Index: head/sys/kern/kern_acct.c =================================================================== --- head/sys/kern/kern_acct.c (revision 305831) +++ head/sys/kern/kern_acct.c (revision 305832) @@ -1,647 +1,647 @@ /*- * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * Copyright (c) 2005 Robert N. M. Watson * All rights reserved. * * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * Copyright (c) 1994 Christopher G. Demetriou * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Berkeley and its contributors. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_acct.c 8.1 (Berkeley) 6/14/93 */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * The routines implemented in this file are described in: * Leffler, et al.: The Design and Implementation of the 4.3BSD * UNIX Operating System (Addison Welley, 1989) * on pages 62-63. * On May 2007 the historic 3 bits base 8 exponent, 13 bit fraction * compt_t representation described in the above reference was replaced * with that of IEEE-754 floats. * * Arguably, to simplify accounting operations, this mechanism should * be replaced by one in which an accounting log file (similar to /dev/klog) * is read by a user process, etc. However, that has its own problems. */ /* Floating point definitions from . */ #define FLT_MANT_DIG 24 /* p */ #define FLT_MAX_EXP 128 /* emax */ /* * Internal accounting functions. * The former's operation is described in Leffler, et al., and the latter * was provided by UCB with the 4.4BSD-Lite release */ static uint32_t encode_timeval(struct timeval); static uint32_t encode_long(long); static void acctwatch(void); static void acct_thread(void *); static int acct_disable(struct thread *, int); /* * Accounting vnode pointer, saved vnode pointer, and flags for each. * acct_sx protects against changes to the active vnode and credentials * while accounting records are being committed to disk. */ static int acct_configured; static int acct_suspended; static struct vnode *acct_vp; static struct ucred *acct_cred; static struct plimit *acct_limit; static int acct_flags; static struct sx acct_sx; SX_SYSINIT(acct, &acct_sx, "acct_sx"); /* * State of the accounting kthread. */ static int acct_state; #define ACCT_RUNNING 1 /* Accounting kthread is running. */ #define ACCT_EXITREQ 2 /* Accounting kthread should exit. */ /* * Values associated with enabling and disabling accounting */ static int acctsuspend = 2; /* stop accounting when < 2% free space left */ SYSCTL_INT(_kern, OID_AUTO, acct_suspend, CTLFLAG_RW, &acctsuspend, 0, "percentage of free disk space below which accounting stops"); static int acctresume = 4; /* resume when free space risen to > 4% */ SYSCTL_INT(_kern, OID_AUTO, acct_resume, CTLFLAG_RW, &acctresume, 0, "percentage of free disk space above which accounting resumes"); static int acctchkfreq = 15; /* frequency (in seconds) to check space */ static int sysctl_acct_chkfreq(SYSCTL_HANDLER_ARGS) { int error, value; /* Write out the old value. */ error = SYSCTL_OUT(req, &acctchkfreq, sizeof(int)); if (error || req->newptr == NULL) return (error); /* Read in and verify the new value. */ error = SYSCTL_IN(req, &value, sizeof(int)); if (error) return (error); if (value <= 0) return (EINVAL); acctchkfreq = value; return (0); } SYSCTL_PROC(_kern, OID_AUTO, acct_chkfreq, CTLTYPE_INT|CTLFLAG_RW, &acctchkfreq, 0, sysctl_acct_chkfreq, "I", "frequency for checking the free space"); SYSCTL_INT(_kern, OID_AUTO, acct_configured, CTLFLAG_RD, &acct_configured, 0, "Accounting configured or not"); SYSCTL_INT(_kern, OID_AUTO, acct_suspended, CTLFLAG_RD, &acct_suspended, 0, "Accounting suspended or not"); /* * Accounting system call. Written based on the specification and previous * implementation done by Mark Tinguely. */ int sys_acct(struct thread *td, struct acct_args *uap) { struct nameidata nd; int error, flags, i, replacing; error = priv_check(td, PRIV_ACCT); if (error) return (error); /* * If accounting is to be started to a file, open that file for * appending and make sure it's a 'normal'. */ if (uap->path != NULL) { NDINIT(&nd, LOOKUP, NOFOLLOW | AUDITVNODE1, UIO_USERSPACE, uap->path, td); flags = FWRITE | O_APPEND; error = vn_open(&nd, &flags, 0, NULL); if (error) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); #ifdef MAC error = mac_system_check_acct(td->td_ucred, nd.ni_vp); if (error) { VOP_UNLOCK(nd.ni_vp, 0); vn_close(nd.ni_vp, flags, td->td_ucred, td); return (error); } #endif VOP_UNLOCK(nd.ni_vp, 0); if (nd.ni_vp->v_type != VREG) { vn_close(nd.ni_vp, flags, td->td_ucred, td); return (EACCES); } #ifdef MAC } else { error = mac_system_check_acct(td->td_ucred, NULL); if (error) return (error); #endif } /* * Disallow concurrent access to the accounting vnode while we swap * it out, in order to prevent access after close. */ sx_xlock(&acct_sx); /* * Don't log spurious disable/enable messages if we are * switching from one accounting file to another due to log * rotation. */ replacing = (acct_vp != NULL && uap->path != NULL); /* * If accounting was previously enabled, kill the old space-watcher, * close the file, and (if no new file was specified, leave). Reset * the suspended state regardless of whether accounting remains * enabled. */ acct_suspended = 0; if (acct_vp != NULL) error = acct_disable(td, !replacing); if (uap->path == NULL) { if (acct_state & ACCT_RUNNING) { acct_state |= ACCT_EXITREQ; wakeup(&acct_state); } sx_xunlock(&acct_sx); return (error); } /* * Create our own plimit object without limits. It will be assigned * to exiting processes. */ acct_limit = lim_alloc(); for (i = 0; i < RLIM_NLIMITS; i++) acct_limit->pl_rlimit[i].rlim_cur = acct_limit->pl_rlimit[i].rlim_max = RLIM_INFINITY; /* * Save the new accounting file vnode, and schedule the new * free space watcher. */ acct_vp = nd.ni_vp; acct_cred = crhold(td->td_ucred); acct_flags = flags; if (acct_state & ACCT_RUNNING) acct_state &= ~ACCT_EXITREQ; else { /* * Try to start up an accounting kthread. We may start more * than one, but if so the extras will commit suicide as * soon as they start up. */ error = kproc_create(acct_thread, NULL, NULL, 0, 0, "accounting"); if (error) { (void) acct_disable(td, 0); sx_xunlock(&acct_sx); log(LOG_NOTICE, "Unable to start accounting thread\n"); return (error); } } acct_configured = 1; sx_xunlock(&acct_sx); if (!replacing) log(LOG_NOTICE, "Accounting enabled\n"); return (error); } /* * Disable currently in-progress accounting by closing the vnode, dropping * our reference to the credential, and clearing the vnode's flags. */ static int acct_disable(struct thread *td, int logging) { int error; sx_assert(&acct_sx, SX_XLOCKED); error = vn_close(acct_vp, acct_flags, acct_cred, td); crfree(acct_cred); lim_free(acct_limit); acct_configured = 0; acct_vp = NULL; acct_cred = NULL; acct_flags = 0; if (logging) log(LOG_NOTICE, "Accounting disabled\n"); return (error); } /* * Write out process accounting information, on process exit. * Data to be written out is specified in Leffler, et al. * and are enumerated below. (They're also noted in the system * "acct.h" header file.) */ int acct_process(struct thread *td) { struct acctv2 acct; struct timeval ut, st, tmp; struct plimit *oldlim; struct proc *p; struct rusage ru; int t, ret; /* * Lockless check of accounting condition before doing the hard * work. */ if (acct_vp == NULL || acct_suspended) return (0); sx_slock(&acct_sx); /* * If accounting isn't enabled, don't bother. Have to check again * once we own the lock in case we raced with disabling of accounting * by another thread. */ if (acct_vp == NULL || acct_suspended) { sx_sunlock(&acct_sx); return (0); } p = td->td_proc; /* * Get process accounting information. */ sx_slock(&proctree_lock); PROC_LOCK(p); /* (1) The terminal from which the process was started */ if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp) acct.ac_tty = tty_udev(p->p_pgrp->pg_session->s_ttyp); else acct.ac_tty = NODEV; sx_sunlock(&proctree_lock); /* (2) The name of the command that ran */ bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm); /* (3) The amount of user and system time that was used */ rufetchcalc(p, &ru, &ut, &st); acct.ac_utime = encode_timeval(ut); acct.ac_stime = encode_timeval(st); /* (4) The elapsed time the command ran (and its starting time) */ getboottime(&tmp); timevaladd(&tmp, &p->p_stats->p_start); acct.ac_btime = tmp.tv_sec; microuptime(&tmp); timevalsub(&tmp, &p->p_stats->p_start); acct.ac_etime = encode_timeval(tmp); /* (5) The average amount of memory used */ tmp = ut; timevaladd(&tmp, &st); /* Convert tmp (i.e. u + s) into hz units to match ru_i*. */ t = tmp.tv_sec * hz + tmp.tv_usec / tick; if (t) acct.ac_mem = encode_long((ru.ru_ixrss + ru.ru_idrss + + ru.ru_isrss) / t); else acct.ac_mem = 0; /* (6) The number of disk I/O operations done */ acct.ac_io = encode_long(ru.ru_inblock + ru.ru_oublock); /* (7) The UID and GID of the process */ acct.ac_uid = p->p_ucred->cr_ruid; acct.ac_gid = p->p_ucred->cr_rgid; /* (8) The boolean flags that tell how the process terminated, etc. */ acct.ac_flagx = p->p_acflag; /* Setup ancillary structure fields. */ acct.ac_flagx |= ANVER; acct.ac_zero = 0; acct.ac_version = 2; acct.ac_len = acct.ac_len2 = sizeof(acct); /* * Eliminate rlimits (file size limit in particular). */ oldlim = p->p_limit; p->p_limit = lim_hold(acct_limit); PROC_UNLOCK(p); lim_free(oldlim); /* * Write the accounting information to the file. */ ret = vn_rdwr(UIO_WRITE, acct_vp, (caddr_t)&acct, sizeof (acct), (off_t)0, UIO_SYSSPACE, IO_APPEND|IO_UNIT, acct_cred, NOCRED, NULL, td); sx_sunlock(&acct_sx); return (ret); } /* FLOAT_CONVERSION_START (Regression testing; don't remove this line.) */ /* Convert timevals and longs into IEEE-754 bit patterns. */ /* Mantissa mask (MSB is implied, so subtract 1). */ #define MANT_MASK ((1 << (FLT_MANT_DIG - 1)) - 1) /* * We calculate integer values to a precision of approximately * 28 bits. * This is high-enough precision to fill the 24 float bits * and low-enough to avoid overflowing the 32 int bits. */ #define CALC_BITS 28 /* log_2(1000000). */ #define LOG2_1M 20 /* * Convert the elements of a timeval into a 32-bit word holding * the bits of a IEEE-754 float. * The float value represents the timeval's value in microsecond units. */ static uint32_t encode_timeval(struct timeval tv) { int log2_s; int val, exp; /* Unnormalized value and exponent */ int norm_exp; /* Normalized exponent */ int shift; /* * First calculate value and exponent to about CALC_BITS precision. * Note that the following conditionals have been ordered so that * the most common cases appear first. */ if (tv.tv_sec == 0) { if (tv.tv_usec == 0) return (0); exp = 0; val = tv.tv_usec; } else { /* * Calculate the value to a precision of approximately * CALC_BITS. */ log2_s = fls(tv.tv_sec) - 1; if (log2_s + LOG2_1M < CALC_BITS) { exp = 0; val = 1000000 * tv.tv_sec + tv.tv_usec; } else { exp = log2_s + LOG2_1M - CALC_BITS; val = (unsigned int)(((uint64_t)1000000 * tv.tv_sec + tv.tv_usec) >> exp); } } /* Now normalize and pack the value into an IEEE-754 float. */ norm_exp = fls(val) - 1; shift = FLT_MANT_DIG - norm_exp - 1; #ifdef ACCT_DEBUG printf("val=%d exp=%d shift=%d log2(val)=%d\n", val, exp, shift, norm_exp); printf("exp=%x mant=%x\n", FLT_MAX_EXP - 1 + exp + norm_exp, ((shift > 0 ? (val << shift) : (val >> -shift)) & MANT_MASK)); #endif return (((FLT_MAX_EXP - 1 + exp + norm_exp) << (FLT_MANT_DIG - 1)) | ((shift > 0 ? val << shift : val >> -shift) & MANT_MASK)); } /* * Convert a non-negative long value into the bit pattern of * an IEEE-754 float value. */ static uint32_t encode_long(long val) { int norm_exp; /* Normalized exponent */ int shift; if (val == 0) return (0); if (val < 0) { log(LOG_NOTICE, "encode_long: negative value %ld in accounting record\n", val); val = LONG_MAX; } norm_exp = fls(val) - 1; shift = FLT_MANT_DIG - norm_exp - 1; #ifdef ACCT_DEBUG printf("val=%d shift=%d log2(val)=%d\n", val, shift, norm_exp); printf("exp=%x mant=%x\n", FLT_MAX_EXP - 1 + exp + norm_exp, ((shift > 0 ? (val << shift) : (val >> -shift)) & MANT_MASK)); #endif return (((FLT_MAX_EXP - 1 + norm_exp) << (FLT_MANT_DIG - 1)) | ((shift > 0 ? val << shift : val >> -shift) & MANT_MASK)); } /* FLOAT_CONVERSION_END (Regression testing; don't remove this line.) */ /* * Periodically check the filesystem to see if accounting * should be turned on or off. Beware the case where the vnode * has been vgone()'d out from underneath us, e.g. when the file * system containing the accounting file has been forcibly unmounted. */ /* ARGSUSED */ static void acctwatch(void) { struct statfs sb; sx_assert(&acct_sx, SX_XLOCKED); /* * If accounting was disabled before our kthread was scheduled, * then acct_vp might be NULL. If so, just ask our kthread to * exit and return. */ if (acct_vp == NULL) { acct_state |= ACCT_EXITREQ; return; } /* * If our vnode is no longer valid, tear it down and signal the * accounting thread to die. */ if (acct_vp->v_type == VBAD) { (void) acct_disable(NULL, 1); acct_state |= ACCT_EXITREQ; return; } /* * Stopping here is better than continuing, maybe it will be VBAD * next time around. */ if (VFS_STATFS(acct_vp->v_mount, &sb) < 0) return; if (acct_suspended) { if (sb.f_bavail > (int64_t)(acctresume * sb.f_blocks / 100)) { acct_suspended = 0; log(LOG_NOTICE, "Accounting resumed\n"); } } else { if (sb.f_bavail <= (int64_t)(acctsuspend * sb.f_blocks / 100)) { acct_suspended = 1; log(LOG_NOTICE, "Accounting suspended\n"); } } } /* * The main loop for the dedicated kernel thread that periodically calls * acctwatch(). */ static void acct_thread(void *dummy) { u_char pri; /* This is a low-priority kernel thread. */ pri = PRI_MAX_KERN; thread_lock(curthread); sched_prio(curthread, pri); thread_unlock(curthread); /* If another accounting kthread is already running, just die. */ sx_xlock(&acct_sx); if (acct_state & ACCT_RUNNING) { sx_xunlock(&acct_sx); kproc_exit(0); } acct_state |= ACCT_RUNNING; /* Loop until we are asked to exit. */ while (!(acct_state & ACCT_EXITREQ)) { /* Perform our periodic checks. */ acctwatch(); /* * We check this flag again before sleeping since the * acctwatch() might have shut down accounting and asked us * to exit. */ if (!(acct_state & ACCT_EXITREQ)) { sx_sleep(&acct_state, &acct_sx, 0, "-", acctchkfreq * hz); } } /* * Acknowledge the exit request and shutdown. We clear both the * exit request and running flags. */ acct_state = 0; sx_xunlock(&acct_sx); kproc_exit(0); } Index: head/sys/kern/kern_clock.c =================================================================== --- head/sys/kern/kern_clock.c (revision 305831) +++ head/sys/kern/kern_clock.c (revision 305832) @@ -1,933 +1,933 @@ /*- * Copyright (c) 1982, 1986, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_clock.c 8.5 (Berkeley) 1/21/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_kdb.h" #include "opt_device_polling.h" #include "opt_hwpmc_hooks.h" #include "opt_ntp.h" #include "opt_watchdog.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef GPROF #include #endif #ifdef HWPMC_HOOKS #include PMC_SOFT_DEFINE( , , clock, hard); PMC_SOFT_DEFINE( , , clock, stat); PMC_SOFT_DEFINE_EX( , , clock, prof, \ cpu_startprofclock, cpu_stopprofclock); #endif #ifdef DEVICE_POLLING extern void hardclock_device_poll(void); #endif /* DEVICE_POLLING */ static void initclocks(void *dummy); SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL); /* Spin-lock protecting profiling statistics. */ static struct mtx time_lock; SDT_PROVIDER_DECLARE(sched); SDT_PROBE_DEFINE2(sched, , , tick, "struct thread *", "struct proc *"); static int sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS) { int error; long cp_time[CPUSTATES]; #ifdef SCTL_MASK32 int i; unsigned int cp_time32[CPUSTATES]; #endif read_cpu_time(cp_time); #ifdef SCTL_MASK32 if (req->flags & SCTL_MASK32) { if (!req->oldptr) return SYSCTL_OUT(req, 0, sizeof(cp_time32)); for (i = 0; i < CPUSTATES; i++) cp_time32[i] = (unsigned int)cp_time[i]; error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32)); } else #endif { if (!req->oldptr) return SYSCTL_OUT(req, 0, sizeof(cp_time)); error = SYSCTL_OUT(req, cp_time, sizeof(cp_time)); } return error; } SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE, 0,0, sysctl_kern_cp_time, "LU", "CPU time statistics"); static long empty[CPUSTATES]; static int sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS) { struct pcpu *pcpu; int error; int c; long *cp_time; #ifdef SCTL_MASK32 unsigned int cp_time32[CPUSTATES]; int i; #endif if (!req->oldptr) { #ifdef SCTL_MASK32 if (req->flags & SCTL_MASK32) return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1)); else #endif return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1)); } for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) { if (!CPU_ABSENT(c)) { pcpu = pcpu_find(c); cp_time = pcpu->pc_cp_time; } else { cp_time = empty; } #ifdef SCTL_MASK32 if (req->flags & SCTL_MASK32) { for (i = 0; i < CPUSTATES; i++) cp_time32[i] = (unsigned int)cp_time[i]; error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32)); } else #endif error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES); } return error; } SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE, 0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics"); #ifdef DEADLKRES static const char *blessed[] = { "getblk", "so_snd_sx", "so_rcv_sx", NULL }; static int slptime_threshold = 1800; static int blktime_threshold = 900; static int sleepfreq = 3; static void deadlkres(void) { struct proc *p; struct thread *td; void *wchan; int blkticks, i, slpticks, slptype, tryl, tticks; tryl = 0; for (;;) { blkticks = blktime_threshold * hz; slpticks = slptime_threshold * hz; /* * Avoid to sleep on the sx_lock in order to avoid a possible * priority inversion problem leading to starvation. * If the lock can't be held after 100 tries, panic. */ if (!sx_try_slock(&allproc_lock)) { if (tryl > 100) panic("%s: possible deadlock detected on allproc_lock\n", __func__); tryl++; pause("allproc", sleepfreq * hz); continue; } tryl = 0; FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); if (p->p_state == PRS_NEW) { PROC_UNLOCK(p); continue; } FOREACH_THREAD_IN_PROC(p, td) { thread_lock(td); if (TD_ON_LOCK(td)) { /* * The thread should be blocked on a * turnstile, simply check if the * turnstile channel is in good state. */ MPASS(td->td_blocked != NULL); tticks = ticks - td->td_blktick; thread_unlock(td); if (tticks > blkticks) { /* * Accordingly with provided * thresholds, this thread is * stuck for too long on a * turnstile. */ PROC_UNLOCK(p); sx_sunlock(&allproc_lock); panic("%s: possible deadlock detected for %p, blocked for %d ticks\n", __func__, td, tticks); } } else if (TD_IS_SLEEPING(td) && TD_ON_SLEEPQ(td)) { /* * Check if the thread is sleeping on a * lock, otherwise skip the check. * Drop the thread lock in order to * avoid a LOR with the sleepqueue * spinlock. */ wchan = td->td_wchan; tticks = ticks - td->td_slptick; thread_unlock(td); slptype = sleepq_type(wchan); if ((slptype == SLEEPQ_SX || slptype == SLEEPQ_LK) && tticks > slpticks) { /* * Accordingly with provided * thresholds, this thread is * stuck for too long on a * sleepqueue. * However, being on a * sleepqueue, we might still * check for the blessed * list. */ tryl = 0; for (i = 0; blessed[i] != NULL; i++) { if (!strcmp(blessed[i], td->td_wmesg)) { tryl = 1; break; } } if (tryl != 0) { tryl = 0; continue; } PROC_UNLOCK(p); sx_sunlock(&allproc_lock); panic("%s: possible deadlock detected for %p, blocked for %d ticks\n", __func__, td, tticks); } } else thread_unlock(td); } PROC_UNLOCK(p); } sx_sunlock(&allproc_lock); /* Sleep for sleepfreq seconds. */ pause("-", sleepfreq * hz); } } static struct kthread_desc deadlkres_kd = { "deadlkres", deadlkres, (struct thread **)NULL }; SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd); static SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW, 0, "Deadlock resolver"); SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW, &slptime_threshold, 0, "Number of seconds within is valid to sleep on a sleepqueue"); SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW, &blktime_threshold, 0, "Number of seconds within is valid to block on a turnstile"); SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0, "Number of seconds between any deadlock resolver thread run"); #endif /* DEADLKRES */ void read_cpu_time(long *cp_time) { struct pcpu *pc; int i, j; /* Sum up global cp_time[]. */ bzero(cp_time, sizeof(long) * CPUSTATES); CPU_FOREACH(i) { pc = pcpu_find(i); for (j = 0; j < CPUSTATES; j++) cp_time[j] += pc->pc_cp_time[j]; } } #ifdef SW_WATCHDOG #include static int watchdog_ticks; static int watchdog_enabled; static void watchdog_fire(void); static void watchdog_config(void *, u_int, int *); #endif /* SW_WATCHDOG */ /* * Clock handling routines. * * This code is written to operate with two timers that run independently of * each other. * * The main timer, running hz times per second, is used to trigger interval * timers, timeouts and rescheduling as needed. * * The second timer handles kernel and user profiling, * and does resource use estimation. If the second timer is programmable, * it is randomized to avoid aliasing between the two clocks. For example, * the randomization prevents an adversary from always giving up the cpu * just before its quantum expires. Otherwise, it would never accumulate * cpu ticks. The mean frequency of the second timer is stathz. * * If no second timer exists, stathz will be zero; in this case we drive * profiling and statistics off the main clock. This WILL NOT be accurate; * do not do it unless absolutely necessary. * * The statistics clock may (or may not) be run at a higher rate while * profiling. This profile clock runs at profhz. We require that profhz * be an integral multiple of stathz. * * If the statistics clock is running fast, it must be divided by the ratio * profhz/stathz for statistics. (For profiling, every tick counts.) * * Time-of-day is maintained using a "timecounter", which may or may * not be related to the hardware generating the above mentioned * interrupts. */ int stathz; int profhz; int profprocs; volatile int ticks; int psratio; static DPCPU_DEFINE(int, pcputicks); /* Per-CPU version of ticks. */ #ifdef DEVICE_POLLING static int devpoll_run = 0; #endif /* * Initialize clock frequencies and start both clocks running. */ /* ARGSUSED*/ static void initclocks(dummy) void *dummy; { #ifdef EARLY_AP_STARTUP struct proc *p; struct thread *td; #endif register int i; /* * Set divisors to 1 (normal case) and let the machine-specific * code do its bit. */ mtx_init(&time_lock, "time lock", NULL, MTX_DEF); cpu_initclocks(); /* * Compute profhz/stathz, and fix profhz if needed. */ i = stathz ? stathz : hz; if (profhz == 0) profhz = i; psratio = profhz / i; #ifdef SW_WATCHDOG EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0); #endif /* * Arrange for ticks to wrap 10 minutes after boot to help catch * sign problems sooner. */ ticks = INT_MAX - (hz * 10 * 60); #ifdef EARLY_AP_STARTUP /* * Fixup the tick counts in any blocked or sleeping threads to * account for the jump above. */ sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); if (p->p_state == PRS_NEW) { PROC_UNLOCK(p); continue; } FOREACH_THREAD_IN_PROC(p, td) { thread_lock(td); if (TD_ON_LOCK(td)) { MPASS(td->td_blktick == 0); td->td_blktick = ticks; } if (TD_ON_SLEEPQ(td)) { MPASS(td->td_slptick == 0); td->td_slptick = ticks; } thread_unlock(td); } PROC_UNLOCK(p); } sx_sunlock(&allproc_lock); #endif } /* * Each time the real-time timer fires, this function is called on all CPUs. * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only * the other CPUs in the system need to call this function. */ void hardclock_cpu(int usermode) { struct pstats *pstats; struct thread *td = curthread; struct proc *p = td->td_proc; int flags; /* * Run current process's virtual and profile time, as needed. */ pstats = p->p_stats; flags = 0; if (usermode && timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) { PROC_ITIMLOCK(p); if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) flags |= TDF_ALRMPEND | TDF_ASTPENDING; PROC_ITIMUNLOCK(p); } if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) { PROC_ITIMLOCK(p); if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) flags |= TDF_PROFPEND | TDF_ASTPENDING; PROC_ITIMUNLOCK(p); } thread_lock(td); td->td_flags |= flags; thread_unlock(td); #ifdef HWPMC_HOOKS if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid))) PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL); if (td->td_intr_frame != NULL) PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame); #endif callout_process(sbinuptime()); } /* * The real-time timer, interrupting hz times per second. */ void hardclock(int usermode, uintfptr_t pc) { atomic_add_int(&ticks, 1); hardclock_cpu(usermode); tc_ticktock(1); cpu_tick_calibration(); /* * If no separate statistics clock is available, run it from here. * * XXX: this only works for UP */ if (stathz == 0) { profclock(usermode, pc); statclock(usermode); } #ifdef DEVICE_POLLING hardclock_device_poll(); /* this is very short and quick */ #endif /* DEVICE_POLLING */ #ifdef SW_WATCHDOG if (watchdog_enabled > 0 && --watchdog_ticks <= 0) watchdog_fire(); #endif /* SW_WATCHDOG */ } void hardclock_cnt(int cnt, int usermode) { struct pstats *pstats; struct thread *td = curthread; struct proc *p = td->td_proc; int *t = DPCPU_PTR(pcputicks); int flags, global, newticks; #ifdef SW_WATCHDOG int i; #endif /* SW_WATCHDOG */ /* * Update per-CPU and possibly global ticks values. */ *t += cnt; do { global = ticks; newticks = *t - global; if (newticks <= 0) { if (newticks < -1) *t = global - 1; newticks = 0; break; } } while (!atomic_cmpset_int(&ticks, global, *t)); /* * Run current process's virtual and profile time, as needed. */ pstats = p->p_stats; flags = 0; if (usermode && timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) { PROC_ITIMLOCK(p); if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick * cnt) == 0) flags |= TDF_ALRMPEND | TDF_ASTPENDING; PROC_ITIMUNLOCK(p); } if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) { PROC_ITIMLOCK(p); if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick * cnt) == 0) flags |= TDF_PROFPEND | TDF_ASTPENDING; PROC_ITIMUNLOCK(p); } if (flags != 0) { thread_lock(td); td->td_flags |= flags; thread_unlock(td); } #ifdef HWPMC_HOOKS if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid))) PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL); if (td->td_intr_frame != NULL) PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame); #endif /* We are in charge to handle this tick duty. */ if (newticks > 0) { tc_ticktock(newticks); #ifdef DEVICE_POLLING /* Dangerous and no need to call these things concurrently. */ if (atomic_cmpset_acq_int(&devpoll_run, 0, 1)) { /* This is very short and quick. */ hardclock_device_poll(); atomic_store_rel_int(&devpoll_run, 0); } #endif /* DEVICE_POLLING */ #ifdef SW_WATCHDOG if (watchdog_enabled > 0) { i = atomic_fetchadd_int(&watchdog_ticks, -newticks); if (i > 0 && i <= newticks) watchdog_fire(); } #endif /* SW_WATCHDOG */ } if (curcpu == CPU_FIRST()) cpu_tick_calibration(); } void hardclock_sync(int cpu) { int *t = DPCPU_ID_PTR(cpu, pcputicks); *t = ticks; } /* * Compute number of ticks in the specified amount of time. */ int tvtohz(tv) struct timeval *tv; { register unsigned long ticks; register long sec, usec; /* * If the number of usecs in the whole seconds part of the time * difference fits in a long, then the total number of usecs will * fit in an unsigned long. Compute the total and convert it to * ticks, rounding up and adding 1 to allow for the current tick * to expire. Rounding also depends on unsigned long arithmetic * to avoid overflow. * * Otherwise, if the number of ticks in the whole seconds part of * the time difference fits in a long, then convert the parts to * ticks separately and add, using similar rounding methods and * overflow avoidance. This method would work in the previous * case but it is slightly slower and assumes that hz is integral. * * Otherwise, round the time difference down to the maximum * representable value. * * If ints have 32 bits, then the maximum value for any timeout in * 10ms ticks is 248 days. */ sec = tv->tv_sec; usec = tv->tv_usec; if (usec < 0) { sec--; usec += 1000000; } if (sec < 0) { #ifdef DIAGNOSTIC if (usec > 0) { sec++; usec -= 1000000; } printf("tvotohz: negative time difference %ld sec %ld usec\n", sec, usec); #endif ticks = 1; } else if (sec <= LONG_MAX / 1000000) ticks = howmany(sec * 1000000 + (unsigned long)usec, tick) + 1; else if (sec <= LONG_MAX / hz) ticks = sec * hz + howmany((unsigned long)usec, tick) + 1; else ticks = LONG_MAX; if (ticks > INT_MAX) ticks = INT_MAX; return ((int)ticks); } /* * Start profiling on a process. * * Kernel profiling passes proc0 which never exits and hence * keeps the profile clock running constantly. */ void startprofclock(p) register struct proc *p; { PROC_LOCK_ASSERT(p, MA_OWNED); if (p->p_flag & P_STOPPROF) return; if ((p->p_flag & P_PROFIL) == 0) { p->p_flag |= P_PROFIL; mtx_lock(&time_lock); if (++profprocs == 1) cpu_startprofclock(); mtx_unlock(&time_lock); } } /* * Stop profiling on a process. */ void stopprofclock(p) register struct proc *p; { PROC_LOCK_ASSERT(p, MA_OWNED); if (p->p_flag & P_PROFIL) { if (p->p_profthreads != 0) { while (p->p_profthreads != 0) { p->p_flag |= P_STOPPROF; msleep(&p->p_profthreads, &p->p_mtx, PPAUSE, "stopprof", 0); } } if ((p->p_flag & P_PROFIL) == 0) return; p->p_flag &= ~P_PROFIL; mtx_lock(&time_lock); if (--profprocs == 0) cpu_stopprofclock(); mtx_unlock(&time_lock); } } /* * Statistics clock. Updates rusage information and calls the scheduler * to adjust priorities of the active thread. * * This should be called by all active processors. */ void statclock(int usermode) { statclock_cnt(1, usermode); } void statclock_cnt(int cnt, int usermode) { struct rusage *ru; struct vmspace *vm; struct thread *td; struct proc *p; long rss; long *cp_time; td = curthread; p = td->td_proc; cp_time = (long *)PCPU_PTR(cp_time); if (usermode) { /* * Charge the time as appropriate. */ td->td_uticks += cnt; if (p->p_nice > NZERO) cp_time[CP_NICE] += cnt; else cp_time[CP_USER] += cnt; } else { /* * Came from kernel mode, so we were: * - handling an interrupt, * - doing syscall or trap work on behalf of the current * user process, or * - spinning in the idle loop. * Whichever it is, charge the time as appropriate. * Note that we charge interrupts to the current process, * regardless of whether they are ``for'' that process, * so that we know how much of its real time was spent * in ``non-process'' (i.e., interrupt) work. */ if ((td->td_pflags & TDP_ITHREAD) || td->td_intr_nesting_level >= 2) { td->td_iticks += cnt; cp_time[CP_INTR] += cnt; } else { td->td_pticks += cnt; td->td_sticks += cnt; if (!TD_IS_IDLETHREAD(td)) cp_time[CP_SYS] += cnt; else cp_time[CP_IDLE] += cnt; } } /* Update resource usage integrals and maximums. */ MPASS(p->p_vmspace != NULL); vm = p->p_vmspace; ru = &td->td_ru; ru->ru_ixrss += pgtok(vm->vm_tsize) * cnt; ru->ru_idrss += pgtok(vm->vm_dsize) * cnt; ru->ru_isrss += pgtok(vm->vm_ssize) * cnt; rss = pgtok(vmspace_resident_count(vm)); if (ru->ru_maxrss < rss) ru->ru_maxrss = rss; KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock", "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz); SDT_PROBE2(sched, , , tick, td, td->td_proc); thread_lock_flags(td, MTX_QUIET); for ( ; cnt > 0; cnt--) sched_clock(td); thread_unlock(td); #ifdef HWPMC_HOOKS if (td->td_intr_frame != NULL) PMC_SOFT_CALL_TF( , , clock, stat, td->td_intr_frame); #endif } void profclock(int usermode, uintfptr_t pc) { profclock_cnt(1, usermode, pc); } void profclock_cnt(int cnt, int usermode, uintfptr_t pc) { struct thread *td; #ifdef GPROF struct gmonparam *g; uintfptr_t i; #endif td = curthread; if (usermode) { /* * Came from user mode; CPU was in user state. * If this process is being profiled, record the tick. * if there is no related user location yet, don't * bother trying to count it. */ if (td->td_proc->p_flag & P_PROFIL) addupc_intr(td, pc, cnt); } #ifdef GPROF else { /* * Kernel statistics are just like addupc_intr, only easier. */ g = &_gmonparam; if (g->state == GMON_PROF_ON && pc >= g->lowpc) { i = PC_TO_I(g, pc); if (i < g->textsize) { KCOUNT(g, i) += cnt; } } } #endif #ifdef HWPMC_HOOKS if (td->td_intr_frame != NULL) PMC_SOFT_CALL_TF( , , clock, prof, td->td_intr_frame); #endif } /* * Return information about system clocks. */ static int sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS) { struct clockinfo clkinfo; /* * Construct clockinfo structure. */ bzero(&clkinfo, sizeof(clkinfo)); clkinfo.hz = hz; clkinfo.tick = tick; clkinfo.profhz = profhz; clkinfo.stathz = stathz ? stathz : hz; return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req)); } SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE, 0, 0, sysctl_kern_clockrate, "S,clockinfo", "Rate and period of various kernel clocks"); #ifdef SW_WATCHDOG static void watchdog_config(void *unused __unused, u_int cmd, int *error) { u_int u; u = cmd & WD_INTERVAL; if (u >= WD_TO_1SEC) { watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz; watchdog_enabled = 1; *error = 0; } else { watchdog_enabled = 0; } } /* * Handle a watchdog timeout by dumping interrupt information and * then either dropping to DDB or panicking. */ static void watchdog_fire(void) { int nintr; uint64_t inttotal; u_long *curintr; char *curname; curintr = intrcnt; curname = intrnames; inttotal = 0; nintr = sintrcnt / sizeof(u_long); printf("interrupt total\n"); while (--nintr >= 0) { if (*curintr) printf("%-12s %20lu\n", curname, *curintr); curname += strlen(curname) + 1; inttotal += *curintr++; } printf("Total %20ju\n", (uintmax_t)inttotal); #if defined(KDB) && !defined(KDB_UNATTENDED) kdb_backtrace(); kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout"); #else panic("watchdog timeout"); #endif } #endif /* SW_WATCHDOG */ Index: head/sys/kern/kern_cons.c =================================================================== --- head/sys/kern/kern_cons.c (revision 305831) +++ head/sys/kern/kern_cons.c (revision 305832) @@ -1,734 +1,734 @@ /*- * Copyright (c) 1988 University of Utah. * Copyright (c) 1991 The Regents of the University of California. * Copyright (c) 1999 Michael Smith * Copyright (c) 2005 Pawel Jakub Dawidek * * All rights reserved. * * This code is derived from software contributed to Berkeley by * the Systems Programming Group of the University of Utah Computer * Science Department. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: @(#)cons.c 7.2 (Berkeley) 5/9/91 */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include "opt_syscons.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static MALLOC_DEFINE(M_TTYCONS, "tty console", "tty console handling"); struct cn_device { STAILQ_ENTRY(cn_device) cnd_next; struct consdev *cnd_cn; }; #define CNDEVPATHMAX 32 #define CNDEVTAB_SIZE 4 static struct cn_device cn_devtab[CNDEVTAB_SIZE]; static STAILQ_HEAD(, cn_device) cn_devlist = STAILQ_HEAD_INITIALIZER(cn_devlist); int cons_avail_mask = 0; /* Bit mask. Each registered low level console * which is currently unavailable for inpit * (i.e., if it is in graphics mode) will have * this bit cleared. */ static int cn_mute; static char *consbuf; /* buffer used by `consmsgbuf' */ static struct callout conscallout; /* callout for outputting to constty */ struct msgbuf consmsgbuf; /* message buffer for console tty */ static u_char console_pausing; /* pause after each line during probe */ static char *console_pausestr= ""; struct tty *constty; /* pointer to console "window" tty */ static struct mtx cnputs_mtx; /* Mutex for cnputs(). */ static int use_cnputs_mtx = 0; /* != 0 if cnputs_mtx locking reqd. */ static void constty_timeout(void *arg); static struct consdev cons_consdev; DATA_SET(cons_set, cons_consdev); SET_DECLARE(cons_set, struct consdev); void cninit(void) { struct consdev *best_cn, *cn, **list; /* * Check if we should mute the console (for security reasons perhaps) * It can be changes dynamically using sysctl kern.consmute * once we are up and going. * */ cn_mute = ((boothowto & (RB_MUTE |RB_SINGLE |RB_VERBOSE |RB_ASKNAME)) == RB_MUTE); /* * Find the first console with the highest priority. */ best_cn = NULL; SET_FOREACH(list, cons_set) { cn = *list; cnremove(cn); /* Skip cons_consdev. */ if (cn->cn_ops == NULL) continue; cn->cn_ops->cn_probe(cn); if (cn->cn_pri == CN_DEAD) continue; if (best_cn == NULL || cn->cn_pri > best_cn->cn_pri) best_cn = cn; if (boothowto & RB_MULTIPLE) { /* * Initialize console, and attach to it. */ cn->cn_ops->cn_init(cn); cnadd(cn); } } if (best_cn == NULL) return; if ((boothowto & RB_MULTIPLE) == 0) { best_cn->cn_ops->cn_init(best_cn); cnadd(best_cn); } if (boothowto & RB_PAUSE) console_pausing = 1; /* * Make the best console the preferred console. */ cnselect(best_cn); #ifdef EARLY_PRINTF /* * Release early console. */ early_putc = NULL; #endif } void cninit_finish() { console_pausing = 0; } /* add a new physical console to back the virtual console */ int cnadd(struct consdev *cn) { struct cn_device *cnd; int i; STAILQ_FOREACH(cnd, &cn_devlist, cnd_next) if (cnd->cnd_cn == cn) return (0); for (i = 0; i < CNDEVTAB_SIZE; i++) { cnd = &cn_devtab[i]; if (cnd->cnd_cn == NULL) break; } if (cnd->cnd_cn != NULL) return (ENOMEM); cnd->cnd_cn = cn; if (cn->cn_name[0] == '\0') { /* XXX: it is unclear if/where this print might output */ printf("WARNING: console at %p has no name\n", cn); } STAILQ_INSERT_TAIL(&cn_devlist, cnd, cnd_next); if (STAILQ_FIRST(&cn_devlist) == cnd) ttyconsdev_select(cnd->cnd_cn->cn_name); /* Add device to the active mask. */ cnavailable(cn, (cn->cn_flags & CN_FLAG_NOAVAIL) == 0); return (0); } void cnremove(struct consdev *cn) { struct cn_device *cnd; int i; STAILQ_FOREACH(cnd, &cn_devlist, cnd_next) { if (cnd->cnd_cn != cn) continue; if (STAILQ_FIRST(&cn_devlist) == cnd) ttyconsdev_select(NULL); STAILQ_REMOVE(&cn_devlist, cnd, cn_device, cnd_next); cnd->cnd_cn = NULL; /* Remove this device from available mask. */ for (i = 0; i < CNDEVTAB_SIZE; i++) if (cnd == &cn_devtab[i]) { cons_avail_mask &= ~(1 << i); break; } #if 0 /* * XXX * syscons gets really confused if console resources are * freed after the system has initialized. */ if (cn->cn_term != NULL) cn->cn_ops->cn_term(cn); #endif return; } } void cnselect(struct consdev *cn) { struct cn_device *cnd; STAILQ_FOREACH(cnd, &cn_devlist, cnd_next) { if (cnd->cnd_cn != cn) continue; if (cnd == STAILQ_FIRST(&cn_devlist)) return; STAILQ_REMOVE(&cn_devlist, cnd, cn_device, cnd_next); STAILQ_INSERT_HEAD(&cn_devlist, cnd, cnd_next); ttyconsdev_select(cnd->cnd_cn->cn_name); return; } } void cnavailable(struct consdev *cn, int available) { int i; for (i = 0; i < CNDEVTAB_SIZE; i++) { if (cn_devtab[i].cnd_cn == cn) break; } if (available) { if (i < CNDEVTAB_SIZE) cons_avail_mask |= (1 << i); cn->cn_flags &= ~CN_FLAG_NOAVAIL; } else { if (i < CNDEVTAB_SIZE) cons_avail_mask &= ~(1 << i); cn->cn_flags |= CN_FLAG_NOAVAIL; } } int cnunavailable(void) { return (cons_avail_mask == 0); } /* * sysctl_kern_console() provides output parseable in conscontrol(1). */ static int sysctl_kern_console(SYSCTL_HANDLER_ARGS) { struct cn_device *cnd; struct consdev *cp, **list; char *p; int delete, error; struct sbuf *sb; sb = sbuf_new(NULL, NULL, CNDEVPATHMAX * 2, SBUF_AUTOEXTEND | SBUF_INCLUDENUL); if (sb == NULL) return (ENOMEM); sbuf_clear(sb); STAILQ_FOREACH(cnd, &cn_devlist, cnd_next) sbuf_printf(sb, "%s,", cnd->cnd_cn->cn_name); sbuf_printf(sb, "/"); SET_FOREACH(list, cons_set) { cp = *list; if (cp->cn_name[0] != '\0') sbuf_printf(sb, "%s,", cp->cn_name); } sbuf_finish(sb); error = sysctl_handle_string(oidp, sbuf_data(sb), sbuf_len(sb), req); if (error == 0 && req->newptr != NULL) { p = sbuf_data(sb); error = ENXIO; delete = 0; if (*p == '-') { delete = 1; p++; } SET_FOREACH(list, cons_set) { cp = *list; if (strcmp(p, cp->cn_name) != 0) continue; if (delete) { cnremove(cp); error = 0; } else { error = cnadd(cp); if (error == 0) cnselect(cp); } break; } } sbuf_delete(sb); return (error); } SYSCTL_PROC(_kern, OID_AUTO, console, CTLTYPE_STRING|CTLFLAG_RW, 0, 0, sysctl_kern_console, "A", "Console device control"); /* * User has changed the state of the console muting. * This may require us to open or close the device in question. */ static int sysctl_kern_consmute(SYSCTL_HANDLER_ARGS) { int error; error = sysctl_handle_int(oidp, &cn_mute, 0, req); if (error != 0 || req->newptr == NULL) return (error); return (error); } SYSCTL_PROC(_kern, OID_AUTO, consmute, CTLTYPE_INT|CTLFLAG_RW, 0, sizeof(cn_mute), sysctl_kern_consmute, "I", "State of the console muting"); void cngrab() { struct cn_device *cnd; struct consdev *cn; STAILQ_FOREACH(cnd, &cn_devlist, cnd_next) { cn = cnd->cnd_cn; if (!kdb_active || !(cn->cn_flags & CN_FLAG_NODEBUG)) cn->cn_ops->cn_grab(cn); } } void cnungrab() { struct cn_device *cnd; struct consdev *cn; STAILQ_FOREACH(cnd, &cn_devlist, cnd_next) { cn = cnd->cnd_cn; if (!kdb_active || !(cn->cn_flags & CN_FLAG_NODEBUG)) cn->cn_ops->cn_ungrab(cn); } } /* * Low level console routines. */ int cngetc(void) { int c; if (cn_mute) return (-1); while ((c = cncheckc()) == -1) cpu_spinwait(); if (c == '\r') c = '\n'; /* console input is always ICRNL */ return (c); } int cncheckc(void) { struct cn_device *cnd; struct consdev *cn; int c; if (cn_mute) return (-1); STAILQ_FOREACH(cnd, &cn_devlist, cnd_next) { cn = cnd->cnd_cn; if (!kdb_active || !(cn->cn_flags & CN_FLAG_NODEBUG)) { c = cn->cn_ops->cn_getc(cn); if (c != -1) return (c); } } return (-1); } void cngets(char *cp, size_t size, int visible) { char *lp, *end; int c; cngrab(); lp = cp; end = cp + size - 1; for (;;) { c = cngetc() & 0177; switch (c) { case '\n': case '\r': cnputc(c); *lp = '\0'; cnungrab(); return; case '\b': case '\177': if (lp > cp) { if (visible) cnputs("\b \b"); lp--; } continue; case '\0': continue; default: if (lp < end) { switch (visible) { case GETS_NOECHO: break; case GETS_ECHOPASS: cnputc('*'); break; default: cnputc(c); break; } *lp++ = c; } } } } void cnputc(int c) { struct cn_device *cnd; struct consdev *cn; char *cp; #ifdef EARLY_PRINTF if (early_putc != NULL) { if (c == '\n') early_putc('\r'); early_putc(c); return; } #endif if (cn_mute || c == '\0') return; STAILQ_FOREACH(cnd, &cn_devlist, cnd_next) { cn = cnd->cnd_cn; if (!kdb_active || !(cn->cn_flags & CN_FLAG_NODEBUG)) { if (c == '\n') cn->cn_ops->cn_putc(cn, '\r'); cn->cn_ops->cn_putc(cn, c); } } if (console_pausing && c == '\n' && !kdb_active) { for (cp = console_pausestr; *cp != '\0'; cp++) cnputc(*cp); cngrab(); if (cngetc() == '.') console_pausing = 0; cnungrab(); cnputc('\r'); for (cp = console_pausestr; *cp != '\0'; cp++) cnputc(' '); cnputc('\r'); } } void cnputs(char *p) { int c; int unlock_reqd = 0; if (use_cnputs_mtx) { /* * NOTE: Debug prints and/or witness printouts in * console driver clients can cause the "cnputs_mtx" * mutex to recurse. Simply return if that happens. */ if (mtx_owned(&cnputs_mtx)) return; mtx_lock_spin(&cnputs_mtx); unlock_reqd = 1; } while ((c = *p++) != '\0') cnputc(c); if (unlock_reqd) mtx_unlock_spin(&cnputs_mtx); } static int consmsgbuf_size = 8192; SYSCTL_INT(_kern, OID_AUTO, consmsgbuf_size, CTLFLAG_RW, &consmsgbuf_size, 0, "Console tty buffer size"); /* * Redirect console output to a tty. */ void constty_set(struct tty *tp) { int size; KASSERT(tp != NULL, ("constty_set: NULL tp")); if (consbuf == NULL) { size = consmsgbuf_size; consbuf = malloc(size, M_TTYCONS, M_WAITOK); msgbuf_init(&consmsgbuf, consbuf, size); callout_init(&conscallout, 0); } constty = tp; constty_timeout(NULL); } /* * Disable console redirection to a tty. */ void constty_clear(void) { int c; constty = NULL; if (consbuf == NULL) return; callout_stop(&conscallout); while ((c = msgbuf_getchar(&consmsgbuf)) != -1) cnputc(c); free(consbuf, M_TTYCONS); consbuf = NULL; } /* Times per second to check for pending console tty messages. */ static int constty_wakeups_per_second = 5; SYSCTL_INT(_kern, OID_AUTO, constty_wakeups_per_second, CTLFLAG_RW, &constty_wakeups_per_second, 0, "Times per second to check for pending console tty messages"); static void constty_timeout(void *arg) { int c; if (constty != NULL) { tty_lock(constty); while ((c = msgbuf_getchar(&consmsgbuf)) != -1) { if (tty_putchar(constty, c) < 0) { tty_unlock(constty); constty = NULL; break; } } if (constty != NULL) tty_unlock(constty); } if (constty != NULL) { callout_reset(&conscallout, hz / constty_wakeups_per_second, constty_timeout, NULL); } else { /* Deallocate the constty buffer memory. */ constty_clear(); } } static void cn_drvinit(void *unused) { mtx_init(&cnputs_mtx, "cnputs_mtx", NULL, MTX_SPIN | MTX_NOWITNESS); use_cnputs_mtx = 1; } SYSINIT(cndev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, cn_drvinit, NULL); /* * Sysbeep(), if we have hardware for it */ #ifdef HAS_TIMER_SPKR static int beeping; static struct callout beeping_timer; static void sysbeepstop(void *chan) { timer_spkr_release(); beeping = 0; } int sysbeep(int pitch, int period) { if (timer_spkr_acquire()) { if (!beeping) { /* Something else owns it. */ return (EBUSY); } } timer_spkr_setfreq(pitch); if (!beeping) { beeping = period; callout_reset(&beeping_timer, period, sysbeepstop, NULL); } return (0); } static void sysbeep_init(void *unused) { callout_init(&beeping_timer, 1); } SYSINIT(sysbeep, SI_SUB_SOFTINTR, SI_ORDER_ANY, sysbeep_init, NULL); #else /* * No hardware, no sound */ int sysbeep(int pitch __unused, int period __unused) { return (ENODEV); } #endif /* * Temporary support for sc(4) to vt(4) transition. */ static unsigned vty_prefer; static char vty_name[16]; SYSCTL_STRING(_kern, OID_AUTO, vty, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, vty_name, 0, "Console vty driver"); int vty_enabled(unsigned vty) { static unsigned vty_selected = 0; if (vty_selected == 0) { TUNABLE_STR_FETCH("kern.vty", vty_name, sizeof(vty_name)); do { #if defined(DEV_SC) if (strcmp(vty_name, "sc") == 0) { vty_selected = VTY_SC; break; } #endif #if defined(DEV_VT) if (strcmp(vty_name, "vt") == 0) { vty_selected = VTY_VT; break; } #endif if (vty_prefer != 0) { vty_selected = vty_prefer; break; } #if defined(DEV_VT) vty_selected = VTY_VT; #elif defined(DEV_SC) vty_selected = VTY_SC; #endif } while (0); if (vty_selected == VTY_VT) strcpy(vty_name, "vt"); else if (vty_selected == VTY_SC) strcpy(vty_name, "sc"); } return ((vty_selected & vty) != 0); } void vty_set_preferred(unsigned vty) { vty_prefer = vty; #if !defined(DEV_SC) vty_prefer &= ~VTY_SC; #endif #if !defined(DEV_VT) vty_prefer &= ~VTY_VT; #endif } Index: head/sys/kern/kern_descrip.c =================================================================== --- head/sys/kern/kern_descrip.c (revision 305831) +++ head/sys/kern/kern_descrip.c (revision 305832) @@ -1,4177 +1,4177 @@ /*- * Copyright (c) 1982, 1986, 1989, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_capsicum.h" #include "opt_compat.h" #include "opt_ddb.h" #include "opt_ktrace.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KTRACE #include #endif #include #include #include #include #include static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table"); static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader", "file desc to leader structures"); static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities"); MALLOC_DECLARE(M_FADVISE); static uma_zone_t file_zone; static uma_zone_t filedesc0_zone; static int closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, int holdleaders); static int fd_first_free(struct filedesc *fdp, int low, int size); static int fd_last_used(struct filedesc *fdp, int size); static void fdgrowtable(struct filedesc *fdp, int nfd); static void fdgrowtable_exp(struct filedesc *fdp, int nfd); static void fdunused(struct filedesc *fdp, int fd); static void fdused(struct filedesc *fdp, int fd); static int getmaxfd(struct thread *td); /* * Each process has: * * - An array of open file descriptors (fd_ofiles) * - An array of file flags (fd_ofileflags) * - A bitmap recording which descriptors are in use (fd_map) * * A process starts out with NDFILE descriptors. The value of NDFILE has * been selected based the historical limit of 20 open files, and an * assumption that the majority of processes, especially short-lived * processes like shells, will never need more. * * If this initial allocation is exhausted, a larger descriptor table and * map are allocated dynamically, and the pointers in the process's struct * filedesc are updated to point to those. This is repeated every time * the process runs out of file descriptors (provided it hasn't hit its * resource limit). * * Since threads may hold references to individual descriptor table * entries, the tables are never freed. Instead, they are placed on a * linked list and freed only when the struct filedesc is released. */ #define NDFILE 20 #define NDSLOTSIZE sizeof(NDSLOTTYPE) #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) #define NDSLOT(x) ((x) / NDENTRIES) #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) /* * SLIST entry used to keep track of ofiles which must be reclaimed when * the process exits. */ struct freetable { struct fdescenttbl *ft_table; SLIST_ENTRY(freetable) ft_next; }; /* * Initial allocation: a filedesc structure + the head of SLIST used to * keep track of old ofiles + enough space for NDFILE descriptors. */ struct fdescenttbl0 { int fdt_nfiles; struct filedescent fdt_ofiles[NDFILE]; }; struct filedesc0 { struct filedesc fd_fd; SLIST_HEAD(, freetable) fd_free; struct fdescenttbl0 fd_dfiles; NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; }; /* * Descriptor management. */ volatile int openfiles; /* actual number of open files */ struct mtx sigio_lock; /* mtx to protect pointers to sigio */ void (*mq_fdclose)(struct thread *td, int fd, struct file *fp); /* * If low >= size, just return low. Otherwise find the first zero bit in the * given bitmap, starting at low and not exceeding size - 1. Return size if * not found. */ static int fd_first_free(struct filedesc *fdp, int low, int size) { NDSLOTTYPE *map = fdp->fd_map; NDSLOTTYPE mask; int off, maxoff; if (low >= size) return (low); off = NDSLOT(low); if (low % NDENTRIES) { mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); if ((mask &= ~map[off]) != 0UL) return (off * NDENTRIES + ffsl(mask) - 1); ++off; } for (maxoff = NDSLOTS(size); off < maxoff; ++off) if (map[off] != ~0UL) return (off * NDENTRIES + ffsl(~map[off]) - 1); return (size); } /* * Find the highest non-zero bit in the given bitmap, starting at 0 and * not exceeding size - 1. Return -1 if not found. */ static int fd_last_used(struct filedesc *fdp, int size) { NDSLOTTYPE *map = fdp->fd_map; NDSLOTTYPE mask; int off, minoff; off = NDSLOT(size); if (size % NDENTRIES) { mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES)); if ((mask &= map[off]) != 0) return (off * NDENTRIES + flsl(mask) - 1); --off; } for (minoff = NDSLOT(0); off >= minoff; --off) if (map[off] != 0) return (off * NDENTRIES + flsl(map[off]) - 1); return (-1); } static int fdisused(struct filedesc *fdp, int fd) { KASSERT(fd >= 0 && fd < fdp->fd_nfiles, ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); } /* * Mark a file descriptor as used. */ static void fdused_init(struct filedesc *fdp, int fd) { KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd)); fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); } static void fdused(struct filedesc *fdp, int fd) { FILEDESC_XLOCK_ASSERT(fdp); fdused_init(fdp, fd); if (fd > fdp->fd_lastfile) fdp->fd_lastfile = fd; if (fd == fdp->fd_freefile) fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles); } /* * Mark a file descriptor as unused. */ static void fdunused(struct filedesc *fdp, int fd) { FILEDESC_XLOCK_ASSERT(fdp); KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd)); KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, ("fd=%d is still in use", fd)); fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); if (fd < fdp->fd_freefile) fdp->fd_freefile = fd; if (fd == fdp->fd_lastfile) fdp->fd_lastfile = fd_last_used(fdp, fd); } /* * Free a file descriptor. * * Avoid some work if fdp is about to be destroyed. */ static inline void fdefree_last(struct filedescent *fde) { filecaps_free(&fde->fde_caps); } static inline void fdfree(struct filedesc *fdp, int fd) { struct filedescent *fde; fde = &fdp->fd_ofiles[fd]; #ifdef CAPABILITIES seq_write_begin(&fde->fde_seq); #endif fdefree_last(fde); fde->fde_file = NULL; fdunused(fdp, fd); #ifdef CAPABILITIES seq_write_end(&fde->fde_seq); #endif } void pwd_ensure_dirs(void) { struct filedesc *fdp; fdp = curproc->p_fd; FILEDESC_XLOCK(fdp); if (fdp->fd_cdir == NULL) { fdp->fd_cdir = rootvnode; VREF(rootvnode); } if (fdp->fd_rdir == NULL) { fdp->fd_rdir = rootvnode; VREF(rootvnode); } FILEDESC_XUNLOCK(fdp); } /* * System calls on descriptors. */ #ifndef _SYS_SYSPROTO_H_ struct getdtablesize_args { int dummy; }; #endif /* ARGSUSED */ int sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) { #ifdef RACCT uint64_t lim; #endif td->td_retval[0] = min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc); #ifdef RACCT PROC_LOCK(td->td_proc); lim = racct_get_limit(td->td_proc, RACCT_NOFILE); PROC_UNLOCK(td->td_proc); if (lim < td->td_retval[0]) td->td_retval[0] = lim; #endif return (0); } /* * Duplicate a file descriptor to a particular value. * * Note: keep in mind that a potential race condition exists when closing * descriptors from a shared descriptor table (via rfork). */ #ifndef _SYS_SYSPROTO_H_ struct dup2_args { u_int from; u_int to; }; #endif /* ARGSUSED */ int sys_dup2(struct thread *td, struct dup2_args *uap) { return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to)); } /* * Duplicate a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct dup_args { u_int fd; }; #endif /* ARGSUSED */ int sys_dup(struct thread *td, struct dup_args *uap) { return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0)); } /* * The file control system call. */ #ifndef _SYS_SYSPROTO_H_ struct fcntl_args { int fd; int cmd; long arg; }; #endif /* ARGSUSED */ int sys_fcntl(struct thread *td, struct fcntl_args *uap) { return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg)); } int kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg) { struct flock fl; struct __oflock ofl; intptr_t arg1; int error, newcmd; error = 0; newcmd = cmd; switch (cmd) { case F_OGETLK: case F_OSETLK: case F_OSETLKW: /* * Convert old flock structure to new. */ error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl)); fl.l_start = ofl.l_start; fl.l_len = ofl.l_len; fl.l_pid = ofl.l_pid; fl.l_type = ofl.l_type; fl.l_whence = ofl.l_whence; fl.l_sysid = 0; switch (cmd) { case F_OGETLK: newcmd = F_GETLK; break; case F_OSETLK: newcmd = F_SETLK; break; case F_OSETLKW: newcmd = F_SETLKW; break; } arg1 = (intptr_t)&fl; break; case F_GETLK: case F_SETLK: case F_SETLKW: case F_SETLK_REMOTE: error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl)); arg1 = (intptr_t)&fl; break; default: arg1 = arg; break; } if (error) return (error); error = kern_fcntl(td, fd, newcmd, arg1); if (error) return (error); if (cmd == F_OGETLK) { ofl.l_start = fl.l_start; ofl.l_len = fl.l_len; ofl.l_pid = fl.l_pid; ofl.l_type = fl.l_type; ofl.l_whence = fl.l_whence; error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl)); } else if (cmd == F_GETLK) { error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl)); } return (error); } int kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) { struct filedesc *fdp; struct flock *flp; struct file *fp, *fp2; struct filedescent *fde; struct proc *p; struct vnode *vp; cap_rights_t rights; int error, flg, tmp; uint64_t bsize; off_t foffset; error = 0; flg = F_POSIX; p = td->td_proc; fdp = p->p_fd; switch (cmd) { case F_DUPFD: tmp = arg; error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp); break; case F_DUPFD_CLOEXEC: tmp = arg; error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp); break; case F_DUP2FD: tmp = arg; error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp); break; case F_DUP2FD_CLOEXEC: tmp = arg; error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp); break; case F_GETFD: error = EBADF; FILEDESC_SLOCK(fdp); fde = fdeget_locked(fdp, fd); if (fde != NULL) { td->td_retval[0] = (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0; error = 0; } FILEDESC_SUNLOCK(fdp); break; case F_SETFD: error = EBADF; FILEDESC_XLOCK(fdp); fde = fdeget_locked(fdp, fd); if (fde != NULL) { fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) | (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); error = 0; } FILEDESC_XUNLOCK(fdp); break; case F_GETFL: error = fget_fcntl(td, fd, cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp); if (error != 0) break; td->td_retval[0] = OFLAGS(fp->f_flag); fdrop(fp, td); break; case F_SETFL: error = fget_fcntl(td, fd, cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp); if (error != 0) break; do { tmp = flg = fp->f_flag; tmp &= ~FCNTLFLAGS; tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); tmp = fp->f_flag & FNONBLOCK; error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); if (error != 0) { fdrop(fp, td); break; } tmp = fp->f_flag & FASYNC; error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); if (error == 0) { fdrop(fp, td); break; } atomic_clear_int(&fp->f_flag, FNONBLOCK); tmp = 0; (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); fdrop(fp, td); break; case F_GETOWN: error = fget_fcntl(td, fd, cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp); if (error != 0) break; error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); if (error == 0) td->td_retval[0] = tmp; fdrop(fp, td); break; case F_SETOWN: error = fget_fcntl(td, fd, cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp); if (error != 0) break; tmp = arg; error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); fdrop(fp, td); break; case F_SETLK_REMOTE: error = priv_check(td, PRIV_NFS_LOCKD); if (error) return (error); flg = F_REMOTE; goto do_setlk; case F_SETLKW: flg |= F_WAIT; /* FALLTHROUGH F_SETLK */ case F_SETLK: do_setlk: cap_rights_init(&rights, CAP_FLOCK); error = fget_unlocked(fdp, fd, &rights, &fp, NULL); if (error != 0) break; if (fp->f_type != DTYPE_VNODE) { error = EBADF; fdrop(fp, td); break; } flp = (struct flock *)arg; if (flp->l_whence == SEEK_CUR) { foffset = foffset_get(fp); if (foffset < 0 || (flp->l_start > 0 && foffset > OFF_MAX - flp->l_start)) { error = EOVERFLOW; fdrop(fp, td); break; } flp->l_start += foffset; } vp = fp->f_vnode; switch (flp->l_type) { case F_RDLCK: if ((fp->f_flag & FREAD) == 0) { error = EBADF; break; } PROC_LOCK(p->p_leader); p->p_leader->p_flag |= P_ADVLOCK; PROC_UNLOCK(p->p_leader); error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, flp, flg); break; case F_WRLCK: if ((fp->f_flag & FWRITE) == 0) { error = EBADF; break; } PROC_LOCK(p->p_leader); p->p_leader->p_flag |= P_ADVLOCK; PROC_UNLOCK(p->p_leader); error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, flp, flg); break; case F_UNLCK: error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, flp, flg); break; case F_UNLCKSYS: /* * Temporary api for testing remote lock * infrastructure. */ if (flg != F_REMOTE) { error = EINVAL; break; } error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCKSYS, flp, flg); break; default: error = EINVAL; break; } if (error != 0 || flp->l_type == F_UNLCK || flp->l_type == F_UNLCKSYS) { fdrop(fp, td); break; } /* * Check for a race with close. * * The vnode is now advisory locked (or unlocked, but this case * is not really important) as the caller requested. * We had to drop the filedesc lock, so we need to recheck if * the descriptor is still valid, because if it was closed * in the meantime we need to remove advisory lock from the * vnode - close on any descriptor leading to an advisory * locked vnode, removes that lock. * We will return 0 on purpose in that case, as the result of * successful advisory lock might have been externally visible * already. This is fine - effectively we pretend to the caller * that the closing thread was a bit slower and that the * advisory lock succeeded before the close. */ error = fget_unlocked(fdp, fd, &rights, &fp2, NULL); if (error != 0) { fdrop(fp, td); break; } if (fp != fp2) { flp->l_whence = SEEK_SET; flp->l_start = 0; flp->l_len = 0; flp->l_type = F_UNLCK; (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, flp, F_POSIX); } fdrop(fp, td); fdrop(fp2, td); break; case F_GETLK: error = fget_unlocked(fdp, fd, cap_rights_init(&rights, CAP_FLOCK), &fp, NULL); if (error != 0) break; if (fp->f_type != DTYPE_VNODE) { error = EBADF; fdrop(fp, td); break; } flp = (struct flock *)arg; if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && flp->l_type != F_UNLCK) { error = EINVAL; fdrop(fp, td); break; } if (flp->l_whence == SEEK_CUR) { foffset = foffset_get(fp); if ((flp->l_start > 0 && foffset > OFF_MAX - flp->l_start) || (flp->l_start < 0 && foffset < OFF_MIN - flp->l_start)) { error = EOVERFLOW; fdrop(fp, td); break; } flp->l_start += foffset; } vp = fp->f_vnode; error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, F_POSIX); fdrop(fp, td); break; case F_RDAHEAD: arg = arg ? 128 * 1024: 0; /* FALLTHROUGH */ case F_READAHEAD: error = fget_unlocked(fdp, fd, cap_rights_init(&rights), &fp, NULL); if (error != 0) break; if (fp->f_type != DTYPE_VNODE) { fdrop(fp, td); error = EBADF; break; } vp = fp->f_vnode; /* * Exclusive lock synchronizes against f_seqcount reads and * writes in sequential_heuristic(). */ error = vn_lock(vp, LK_EXCLUSIVE); if (error != 0) { fdrop(fp, td); break; } if (arg >= 0) { bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; fp->f_seqcount = (arg + bsize - 1) / bsize; atomic_set_int(&fp->f_flag, FRDAHEAD); } else { atomic_clear_int(&fp->f_flag, FRDAHEAD); } VOP_UNLOCK(vp, 0); fdrop(fp, td); break; default: error = EINVAL; break; } return (error); } static int getmaxfd(struct thread *td) { return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc)); } /* * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). */ int kern_dup(struct thread *td, u_int mode, int flags, int old, int new) { struct filedesc *fdp; struct filedescent *oldfde, *newfde; struct proc *p; struct file *delfp; int error, maxfd; p = td->td_proc; fdp = p->p_fd; MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0); MPASS(mode < FDDUP_LASTMODE); AUDIT_ARG_FD(old); /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */ /* * Verify we have a valid descriptor to dup from and possibly to * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should * return EINVAL when the new descriptor is out of bounds. */ if (old < 0) return (EBADF); if (new < 0) return (mode == FDDUP_FCNTL ? EINVAL : EBADF); maxfd = getmaxfd(td); if (new >= maxfd) return (mode == FDDUP_FCNTL ? EINVAL : EBADF); error = EBADF; FILEDESC_XLOCK(fdp); if (fget_locked(fdp, old) == NULL) goto unlock; if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) { td->td_retval[0] = new; if (flags & FDDUP_FLAG_CLOEXEC) fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE; error = 0; goto unlock; } /* * If the caller specified a file descriptor, make sure the file * table is large enough to hold it, and grab it. Otherwise, just * allocate a new descriptor the usual way. */ switch (mode) { case FDDUP_NORMAL: case FDDUP_FCNTL: if ((error = fdalloc(td, new, &new)) != 0) goto unlock; break; case FDDUP_MUSTREPLACE: /* Target file descriptor must exist. */ if (fget_locked(fdp, new) == NULL) goto unlock; break; case FDDUP_FIXED: if (new >= fdp->fd_nfiles) { /* * The resource limits are here instead of e.g. * fdalloc(), because the file descriptor table may be * shared between processes, so we can't really use * racct_add()/racct_sub(). Instead of counting the * number of actually allocated descriptors, just put * the limit on the size of the file descriptor table. */ #ifdef RACCT if (racct_enable) { PROC_LOCK(p); error = racct_set(p, RACCT_NOFILE, new + 1); PROC_UNLOCK(p); if (error != 0) { error = EMFILE; goto unlock; } } #endif fdgrowtable_exp(fdp, new + 1); } if (!fdisused(fdp, new)) fdused(fdp, new); break; default: KASSERT(0, ("%s unsupported mode %d", __func__, mode)); } KASSERT(old != new, ("new fd is same as old")); oldfde = &fdp->fd_ofiles[old]; fhold(oldfde->fde_file); newfde = &fdp->fd_ofiles[new]; delfp = newfde->fde_file; /* * Duplicate the source descriptor. */ #ifdef CAPABILITIES seq_write_begin(&newfde->fde_seq); #endif filecaps_free(&newfde->fde_caps); memcpy(newfde, oldfde, fde_change_size); filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps, true); if ((flags & FDDUP_FLAG_CLOEXEC) != 0) newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE; else newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE; #ifdef CAPABILITIES seq_write_end(&newfde->fde_seq); #endif td->td_retval[0] = new; error = 0; if (delfp != NULL) { (void) closefp(fdp, new, delfp, td, 1); FILEDESC_UNLOCK_ASSERT(fdp); } else { unlock: FILEDESC_XUNLOCK(fdp); } return (error); } /* * If sigio is on the list associated with a process or process group, * disable signalling from the device, remove sigio from the list and * free sigio. */ void funsetown(struct sigio **sigiop) { struct sigio *sigio; if (*sigiop == NULL) return; SIGIO_LOCK(); sigio = *sigiop; if (sigio == NULL) { SIGIO_UNLOCK(); return; } *(sigio->sio_myref) = NULL; if ((sigio)->sio_pgid < 0) { struct pgrp *pg = (sigio)->sio_pgrp; PGRP_LOCK(pg); SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, sigio, sio_pgsigio); PGRP_UNLOCK(pg); } else { struct proc *p = (sigio)->sio_proc; PROC_LOCK(p); SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, sigio, sio_pgsigio); PROC_UNLOCK(p); } SIGIO_UNLOCK(); crfree(sigio->sio_ucred); free(sigio, M_SIGIO); } /* * Free a list of sigio structures. * We only need to lock the SIGIO_LOCK because we have made ourselves * inaccessible to callers of fsetown and therefore do not need to lock * the proc or pgrp struct for the list manipulation. */ void funsetownlst(struct sigiolst *sigiolst) { struct proc *p; struct pgrp *pg; struct sigio *sigio; sigio = SLIST_FIRST(sigiolst); if (sigio == NULL) return; p = NULL; pg = NULL; /* * Every entry of the list should belong * to a single proc or pgrp. */ if (sigio->sio_pgid < 0) { pg = sigio->sio_pgrp; PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); } else /* if (sigio->sio_pgid > 0) */ { p = sigio->sio_proc; PROC_LOCK_ASSERT(p, MA_NOTOWNED); } SIGIO_LOCK(); while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { *(sigio->sio_myref) = NULL; if (pg != NULL) { KASSERT(sigio->sio_pgid < 0, ("Proc sigio in pgrp sigio list")); KASSERT(sigio->sio_pgrp == pg, ("Bogus pgrp in sigio list")); PGRP_LOCK(pg); SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio); PGRP_UNLOCK(pg); } else /* if (p != NULL) */ { KASSERT(sigio->sio_pgid > 0, ("Pgrp sigio in proc sigio list")); KASSERT(sigio->sio_proc == p, ("Bogus proc in sigio list")); PROC_LOCK(p); SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio); PROC_UNLOCK(p); } SIGIO_UNLOCK(); crfree(sigio->sio_ucred); free(sigio, M_SIGIO); SIGIO_LOCK(); } SIGIO_UNLOCK(); } /* * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). * * After permission checking, add a sigio structure to the sigio list for * the process or process group. */ int fsetown(pid_t pgid, struct sigio **sigiop) { struct proc *proc; struct pgrp *pgrp; struct sigio *sigio; int ret; if (pgid == 0) { funsetown(sigiop); return (0); } ret = 0; /* Allocate and fill in the new sigio out of locks. */ sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); sigio->sio_pgid = pgid; sigio->sio_ucred = crhold(curthread->td_ucred); sigio->sio_myref = sigiop; sx_slock(&proctree_lock); if (pgid > 0) { proc = pfind(pgid); if (proc == NULL) { ret = ESRCH; goto fail; } /* * Policy - Don't allow a process to FSETOWN a process * in another session. * * Remove this test to allow maximum flexibility or * restrict FSETOWN to the current process or process * group for maximum safety. */ PROC_UNLOCK(proc); if (proc->p_session != curthread->td_proc->p_session) { ret = EPERM; goto fail; } pgrp = NULL; } else /* if (pgid < 0) */ { pgrp = pgfind(-pgid); if (pgrp == NULL) { ret = ESRCH; goto fail; } PGRP_UNLOCK(pgrp); /* * Policy - Don't allow a process to FSETOWN a process * in another session. * * Remove this test to allow maximum flexibility or * restrict FSETOWN to the current process or process * group for maximum safety. */ if (pgrp->pg_session != curthread->td_proc->p_session) { ret = EPERM; goto fail; } proc = NULL; } funsetown(sigiop); if (pgid > 0) { PROC_LOCK(proc); /* * Since funsetownlst() is called without the proctree * locked, we need to check for P_WEXIT. * XXX: is ESRCH correct? */ if ((proc->p_flag & P_WEXIT) != 0) { PROC_UNLOCK(proc); ret = ESRCH; goto fail; } SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); sigio->sio_proc = proc; PROC_UNLOCK(proc); } else { PGRP_LOCK(pgrp); SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); sigio->sio_pgrp = pgrp; PGRP_UNLOCK(pgrp); } sx_sunlock(&proctree_lock); SIGIO_LOCK(); *sigiop = sigio; SIGIO_UNLOCK(); return (0); fail: sx_sunlock(&proctree_lock); crfree(sigio->sio_ucred); free(sigio, M_SIGIO); return (ret); } /* * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). */ pid_t fgetown(sigiop) struct sigio **sigiop; { pid_t pgid; SIGIO_LOCK(); pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; SIGIO_UNLOCK(); return (pgid); } /* * Function drops the filedesc lock on return. */ static int closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, int holdleaders) { int error; FILEDESC_XLOCK_ASSERT(fdp); if (holdleaders) { if (td->td_proc->p_fdtol != NULL) { /* * Ask fdfree() to sleep to ensure that all relevant * process leaders can be traversed in closef(). */ fdp->fd_holdleaderscount++; } else { holdleaders = 0; } } /* * We now hold the fp reference that used to be owned by the * descriptor array. We have to unlock the FILEDESC *AFTER* * knote_fdclose to prevent a race of the fd getting opened, a knote * added, and deleteing a knote for the new fd. */ knote_fdclose(td, fd); /* * We need to notify mqueue if the object is of type mqueue. */ if (fp->f_type == DTYPE_MQUEUE) mq_fdclose(td, fd, fp); FILEDESC_XUNLOCK(fdp); error = closef(fp, td); if (holdleaders) { FILEDESC_XLOCK(fdp); fdp->fd_holdleaderscount--; if (fdp->fd_holdleaderscount == 0 && fdp->fd_holdleaderswakeup != 0) { fdp->fd_holdleaderswakeup = 0; wakeup(&fdp->fd_holdleaderscount); } FILEDESC_XUNLOCK(fdp); } return (error); } /* * Close a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct close_args { int fd; }; #endif /* ARGSUSED */ int sys_close(struct thread *td, struct close_args *uap) { return (kern_close(td, uap->fd)); } int kern_close(struct thread *td, int fd) { struct filedesc *fdp; struct file *fp; fdp = td->td_proc->p_fd; AUDIT_SYSCLOSE(td, fd); FILEDESC_XLOCK(fdp); if ((fp = fget_locked(fdp, fd)) == NULL) { FILEDESC_XUNLOCK(fdp); return (EBADF); } fdfree(fdp, fd); /* closefp() drops the FILEDESC lock for us. */ return (closefp(fdp, fd, fp, td, 1)); } /* * Close open file descriptors. */ #ifndef _SYS_SYSPROTO_H_ struct closefrom_args { int lowfd; }; #endif /* ARGSUSED */ int sys_closefrom(struct thread *td, struct closefrom_args *uap) { struct filedesc *fdp; int fd; fdp = td->td_proc->p_fd; AUDIT_ARG_FD(uap->lowfd); /* * Treat negative starting file descriptor values identical to * closefrom(0) which closes all files. */ if (uap->lowfd < 0) uap->lowfd = 0; FILEDESC_SLOCK(fdp); for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) { if (fdp->fd_ofiles[fd].fde_file != NULL) { FILEDESC_SUNLOCK(fdp); (void)kern_close(td, fd); FILEDESC_SLOCK(fdp); } } FILEDESC_SUNLOCK(fdp); return (0); } #if defined(COMPAT_43) /* * Return status information about a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct ofstat_args { int fd; struct ostat *sb; }; #endif /* ARGSUSED */ int ofstat(struct thread *td, struct ofstat_args *uap) { struct ostat oub; struct stat ub; int error; error = kern_fstat(td, uap->fd, &ub); if (error == 0) { cvtstat(&ub, &oub); error = copyout(&oub, uap->sb, sizeof(oub)); } return (error); } #endif /* COMPAT_43 */ /* * Return status information about a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct fstat_args { int fd; struct stat *sb; }; #endif /* ARGSUSED */ int sys_fstat(struct thread *td, struct fstat_args *uap) { struct stat ub; int error; error = kern_fstat(td, uap->fd, &ub); if (error == 0) error = copyout(&ub, uap->sb, sizeof(ub)); return (error); } int kern_fstat(struct thread *td, int fd, struct stat *sbp) { struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(fd); error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp); if (error != 0) return (error); AUDIT_ARG_FILE(td->td_proc, fp); error = fo_stat(fp, sbp, td->td_ucred, td); fdrop(fp, td); #ifdef KTRACE if (error == 0 && KTRPOINT(td, KTR_STRUCT)) ktrstat(sbp); #endif return (error); } /* * Return status information about a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct nfstat_args { int fd; struct nstat *sb; }; #endif /* ARGSUSED */ int sys_nfstat(struct thread *td, struct nfstat_args *uap) { struct nstat nub; struct stat ub; int error; error = kern_fstat(td, uap->fd, &ub); if (error == 0) { cvtnstat(&ub, &nub); error = copyout(&nub, uap->sb, sizeof(nub)); } return (error); } /* * Return pathconf information about a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct fpathconf_args { int fd; int name; }; #endif /* ARGSUSED */ int sys_fpathconf(struct thread *td, struct fpathconf_args *uap) { struct file *fp; struct vnode *vp; cap_rights_t rights; int error; error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp); if (error != 0) return (error); if (uap->name == _PC_ASYNC_IO) { td->td_retval[0] = _POSIX_ASYNCHRONOUS_IO; goto out; } vp = fp->f_vnode; if (vp != NULL) { vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_PATHCONF(vp, uap->name, td->td_retval); VOP_UNLOCK(vp, 0); } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { if (uap->name != _PC_PIPE_BUF) { error = EINVAL; } else { td->td_retval[0] = PIPE_BUF; error = 0; } } else { error = EOPNOTSUPP; } out: fdrop(fp, td); return (error); } /* * Initialize filecaps structure. */ void filecaps_init(struct filecaps *fcaps) { bzero(fcaps, sizeof(*fcaps)); fcaps->fc_nioctls = -1; } /* * Copy filecaps structure allocating memory for ioctls array if needed. * * The last parameter indicates whether the fdtable is locked. If it is not and * ioctls are encountered, copying fails and the caller must lock the table. * * Note that if the table was not locked, the caller has to check the relevant * sequence counter to determine whether the operation was successful. */ int filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked) { size_t size; *dst = *src; if (src->fc_ioctls == NULL) return (0); if (!locked) return (1); KASSERT(src->fc_nioctls > 0, ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); bcopy(src->fc_ioctls, dst->fc_ioctls, size); return (0); } /* * Move filecaps structure to the new place and clear the old place. */ void filecaps_move(struct filecaps *src, struct filecaps *dst) { *dst = *src; bzero(src, sizeof(*src)); } /* * Fill the given filecaps structure with full rights. */ static void filecaps_fill(struct filecaps *fcaps) { CAP_ALL(&fcaps->fc_rights); fcaps->fc_ioctls = NULL; fcaps->fc_nioctls = -1; fcaps->fc_fcntls = CAP_FCNTL_ALL; } /* * Free memory allocated within filecaps structure. */ void filecaps_free(struct filecaps *fcaps) { free(fcaps->fc_ioctls, M_FILECAPS); bzero(fcaps, sizeof(*fcaps)); } /* * Validate the given filecaps structure. */ static void filecaps_validate(const struct filecaps *fcaps, const char *func) { KASSERT(cap_rights_is_valid(&fcaps->fc_rights), ("%s: invalid rights", func)); KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, ("%s: invalid fcntls", func)); KASSERT(fcaps->fc_fcntls == 0 || cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), ("%s: fcntls without CAP_FCNTL", func)); KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), ("%s: invalid ioctls", func)); KASSERT(fcaps->fc_nioctls == 0 || cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), ("%s: ioctls without CAP_IOCTL", func)); } static void fdgrowtable_exp(struct filedesc *fdp, int nfd) { int nfd1; FILEDESC_XLOCK_ASSERT(fdp); nfd1 = fdp->fd_nfiles * 2; if (nfd1 < nfd) nfd1 = nfd; fdgrowtable(fdp, nfd1); } /* * Grow the file table to accommodate (at least) nfd descriptors. */ static void fdgrowtable(struct filedesc *fdp, int nfd) { struct filedesc0 *fdp0; struct freetable *ft; struct fdescenttbl *ntable; struct fdescenttbl *otable; int nnfiles, onfiles; NDSLOTTYPE *nmap, *omap; /* * If lastfile is -1 this struct filedesc was just allocated and we are * growing it to accommodate for the one we are going to copy from. There * is no need to have a lock on this one as it's not visible to anyone. */ if (fdp->fd_lastfile != -1) FILEDESC_XLOCK_ASSERT(fdp); KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); /* save old values */ onfiles = fdp->fd_nfiles; otable = fdp->fd_files; omap = fdp->fd_map; /* compute the size of the new table */ nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ if (nnfiles <= onfiles) /* the table is already large enough */ return; /* * Allocate a new table. We need enough space for the number of * entries, file entries themselves and the struct freetable we will use * when we decommission the table and place it on the freelist. * We place the struct freetable in the middle so we don't have * to worry about padding. */ ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) + nnfiles * sizeof(ntable->fdt_ofiles[0]) + sizeof(struct freetable), M_FILEDESC, M_ZERO | M_WAITOK); /* copy the old data */ ntable->fdt_nfiles = nnfiles; memcpy(ntable->fdt_ofiles, otable->fdt_ofiles, onfiles * sizeof(ntable->fdt_ofiles[0])); /* * Allocate a new map only if the old is not large enough. It will * grow at a slower rate than the table as it can map more * entries than the table can hold. */ if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, M_ZERO | M_WAITOK); /* copy over the old data and update the pointer */ memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); fdp->fd_map = nmap; } /* * Make sure that ntable is correctly initialized before we replace * fd_files poiner. Otherwise fget_unlocked() may see inconsistent * data. */ atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable); /* * Do not free the old file table, as some threads may still * reference entries within it. Instead, place it on a freelist * which will be processed when the struct filedesc is released. * * Note that if onfiles == NDFILE, we're dealing with the original * static allocation contained within (struct filedesc0 *)fdp, * which must not be freed. */ if (onfiles > NDFILE) { ft = (struct freetable *)&otable->fdt_ofiles[onfiles]; fdp0 = (struct filedesc0 *)fdp; ft->ft_table = otable; SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); } /* * The map does not have the same possibility of threads still * holding references to it. So always free it as long as it * does not reference the original static allocation. */ if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) free(omap, M_FILEDESC); } /* * Allocate a file descriptor for the process. */ int fdalloc(struct thread *td, int minfd, int *result) { struct proc *p = td->td_proc; struct filedesc *fdp = p->p_fd; int fd, maxfd, allocfd; #ifdef RACCT int error; #endif FILEDESC_XLOCK_ASSERT(fdp); if (fdp->fd_freefile > minfd) minfd = fdp->fd_freefile; maxfd = getmaxfd(td); /* * Search the bitmap for a free descriptor starting at minfd. * If none is found, grow the file table. */ fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); if (fd >= maxfd) return (EMFILE); if (fd >= fdp->fd_nfiles) { allocfd = min(fd * 2, maxfd); #ifdef RACCT if (racct_enable) { PROC_LOCK(p); error = racct_set(p, RACCT_NOFILE, allocfd); PROC_UNLOCK(p); if (error != 0) return (EMFILE); } #endif /* * fd is already equal to first free descriptor >= minfd, so * we only need to grow the table and we are done. */ fdgrowtable_exp(fdp, allocfd); } /* * Perform some sanity checks, then mark the file descriptor as * used and return it to the caller. */ KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), ("invalid descriptor %d", fd)); KASSERT(!fdisused(fdp, fd), ("fd_first_free() returned non-free descriptor")); KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, ("file descriptor isn't free")); fdused(fdp, fd); *result = fd; return (0); } /* * Allocate n file descriptors for the process. */ int fdallocn(struct thread *td, int minfd, int *fds, int n) { struct proc *p = td->td_proc; struct filedesc *fdp = p->p_fd; int i; FILEDESC_XLOCK_ASSERT(fdp); for (i = 0; i < n; i++) if (fdalloc(td, 0, &fds[i]) != 0) break; if (i < n) { for (i--; i >= 0; i--) fdunused(fdp, fds[i]); return (EMFILE); } return (0); } /* * Create a new open file structure and allocate a file descriptor for the * process that refers to it. We add one reference to the file for the * descriptor table and one reference for resultfp. This is to prevent us * being preempted and the entry in the descriptor table closed after we * release the FILEDESC lock. */ int falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags, struct filecaps *fcaps) { struct file *fp; int error, fd; error = falloc_noinstall(td, &fp); if (error) return (error); /* no reference held on error */ error = finstall(td, fp, &fd, flags, fcaps); if (error) { fdrop(fp, td); /* one reference (fp only) */ return (error); } if (resultfp != NULL) *resultfp = fp; /* copy out result */ else fdrop(fp, td); /* release local reference */ if (resultfd != NULL) *resultfd = fd; return (0); } /* * Create a new open file structure without allocating a file descriptor. */ int falloc_noinstall(struct thread *td, struct file **resultfp) { struct file *fp; int maxuserfiles = maxfiles - (maxfiles / 20); static struct timeval lastfail; static int curfail; KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); if ((openfiles >= maxuserfiles && priv_check(td, PRIV_MAXFILES) != 0) || openfiles >= maxfiles) { if (ppsratecheck(&lastfail, &curfail, 1)) { printf("kern.maxfiles limit exceeded by uid %i, " "please see tuning(7).\n", td->td_ucred->cr_ruid); } return (ENFILE); } atomic_add_int(&openfiles, 1); fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); refcount_init(&fp->f_count, 1); fp->f_cred = crhold(td->td_ucred); fp->f_ops = &badfileops; *resultfp = fp; return (0); } /* * Install a file in a file descriptor table. */ void _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags, struct filecaps *fcaps) { struct filedescent *fde; MPASS(fp != NULL); if (fcaps != NULL) filecaps_validate(fcaps, __func__); FILEDESC_XLOCK_ASSERT(fdp); fde = &fdp->fd_ofiles[fd]; #ifdef CAPABILITIES seq_write_begin(&fde->fde_seq); #endif fde->fde_file = fp; fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0; if (fcaps != NULL) filecaps_move(fcaps, &fde->fde_caps); else filecaps_fill(&fde->fde_caps); #ifdef CAPABILITIES seq_write_end(&fde->fde_seq); #endif } int finstall(struct thread *td, struct file *fp, int *fd, int flags, struct filecaps *fcaps) { struct filedesc *fdp = td->td_proc->p_fd; int error; MPASS(fd != NULL); FILEDESC_XLOCK(fdp); if ((error = fdalloc(td, 0, fd))) { FILEDESC_XUNLOCK(fdp); return (error); } fhold(fp); _finstall(fdp, fp, *fd, flags, fcaps); FILEDESC_XUNLOCK(fdp); return (0); } /* * Build a new filedesc structure from another. * Copy the current, root, and jail root vnode references. * * If fdp is not NULL, return with it shared locked. */ struct filedesc * fdinit(struct filedesc *fdp, bool prepfiles) { struct filedesc0 *newfdp0; struct filedesc *newfdp; newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO); newfdp = &newfdp0->fd_fd; /* Create the file descriptor table. */ FILEDESC_LOCK_INIT(newfdp); refcount_init(&newfdp->fd_refcnt, 1); refcount_init(&newfdp->fd_holdcnt, 1); newfdp->fd_cmask = CMASK; newfdp->fd_map = newfdp0->fd_dmap; newfdp->fd_lastfile = -1; newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles; newfdp->fd_files->fdt_nfiles = NDFILE; if (fdp == NULL) return (newfdp); if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles) fdgrowtable(newfdp, fdp->fd_lastfile + 1); FILEDESC_SLOCK(fdp); newfdp->fd_cdir = fdp->fd_cdir; if (newfdp->fd_cdir) VREF(newfdp->fd_cdir); newfdp->fd_rdir = fdp->fd_rdir; if (newfdp->fd_rdir) VREF(newfdp->fd_rdir); newfdp->fd_jdir = fdp->fd_jdir; if (newfdp->fd_jdir) VREF(newfdp->fd_jdir); if (!prepfiles) { FILEDESC_SUNLOCK(fdp); } else { while (fdp->fd_lastfile >= newfdp->fd_nfiles) { FILEDESC_SUNLOCK(fdp); fdgrowtable(newfdp, fdp->fd_lastfile + 1); FILEDESC_SLOCK(fdp); } } return (newfdp); } static struct filedesc * fdhold(struct proc *p) { struct filedesc *fdp; PROC_LOCK_ASSERT(p, MA_OWNED); fdp = p->p_fd; if (fdp != NULL) refcount_acquire(&fdp->fd_holdcnt); return (fdp); } static void fddrop(struct filedesc *fdp) { if (fdp->fd_holdcnt > 1) { if (refcount_release(&fdp->fd_holdcnt) == 0) return; } FILEDESC_LOCK_DESTROY(fdp); uma_zfree(filedesc0_zone, fdp); } /* * Share a filedesc structure. */ struct filedesc * fdshare(struct filedesc *fdp) { refcount_acquire(&fdp->fd_refcnt); return (fdp); } /* * Unshare a filedesc structure, if necessary by making a copy */ void fdunshare(struct thread *td) { struct filedesc *tmp; struct proc *p = td->td_proc; if (p->p_fd->fd_refcnt == 1) return; tmp = fdcopy(p->p_fd); fdescfree(td); p->p_fd = tmp; } void fdinstall_remapped(struct thread *td, struct filedesc *fdp) { fdescfree(td); td->td_proc->p_fd = fdp; } /* * Copy a filedesc structure. A NULL pointer in returns a NULL reference, * this is to ease callers, not catch errors. */ struct filedesc * fdcopy(struct filedesc *fdp) { struct filedesc *newfdp; struct filedescent *nfde, *ofde; int i; MPASS(fdp != NULL); newfdp = fdinit(fdp, true); /* copy all passable descriptors (i.e. not kqueue) */ newfdp->fd_freefile = -1; for (i = 0; i <= fdp->fd_lastfile; ++i) { ofde = &fdp->fd_ofiles[i]; if (ofde->fde_file == NULL || (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) { if (newfdp->fd_freefile == -1) newfdp->fd_freefile = i; continue; } nfde = &newfdp->fd_ofiles[i]; *nfde = *ofde; filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true); fhold(nfde->fde_file); fdused_init(newfdp, i); newfdp->fd_lastfile = i; } if (newfdp->fd_freefile == -1) newfdp->fd_freefile = i; newfdp->fd_cmask = fdp->fd_cmask; FILEDESC_SUNLOCK(fdp); return (newfdp); } /* * Copies a filedesc structure, while remapping all file descriptors * stored inside using a translation table. * * File descriptors are copied over to the new file descriptor table, * regardless of whether the close-on-exec flag is set. */ int fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds, struct filedesc **ret) { struct filedesc *newfdp; struct filedescent *nfde, *ofde; int error, i; MPASS(fdp != NULL); newfdp = fdinit(fdp, true); if (nfds > fdp->fd_lastfile + 1) { /* New table cannot be larger than the old one. */ error = E2BIG; goto bad; } /* Copy all passable descriptors (i.e. not kqueue). */ newfdp->fd_freefile = nfds; for (i = 0; i < nfds; ++i) { if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) { /* File descriptor out of bounds. */ error = EBADF; goto bad; } ofde = &fdp->fd_ofiles[fds[i]]; if (ofde->fde_file == NULL) { /* Unused file descriptor. */ error = EBADF; goto bad; } if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) { /* File descriptor cannot be passed. */ error = EINVAL; goto bad; } nfde = &newfdp->fd_ofiles[i]; *nfde = *ofde; filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true); fhold(nfde->fde_file); fdused_init(newfdp, i); newfdp->fd_lastfile = i; } newfdp->fd_cmask = fdp->fd_cmask; FILEDESC_SUNLOCK(fdp); *ret = newfdp; return (0); bad: FILEDESC_SUNLOCK(fdp); fdescfree_remapped(newfdp); return (error); } /* * Clear POSIX style locks. This is only used when fdp looses a reference (i.e. * one of processes using it exits) and the table used to be shared. */ static void fdclearlocks(struct thread *td) { struct filedesc *fdp; struct filedesc_to_leader *fdtol; struct flock lf; struct file *fp; struct proc *p; struct vnode *vp; int i; p = td->td_proc; fdp = p->p_fd; fdtol = p->p_fdtol; MPASS(fdtol != NULL); FILEDESC_XLOCK(fdp); KASSERT(fdtol->fdl_refcount > 0, ("filedesc_to_refcount botch: fdl_refcount=%d", fdtol->fdl_refcount)); if (fdtol->fdl_refcount == 1 && (p->p_leader->p_flag & P_ADVLOCK) != 0) { for (i = 0; i <= fdp->fd_lastfile; i++) { fp = fdp->fd_ofiles[i].fde_file; if (fp == NULL || fp->f_type != DTYPE_VNODE) continue; fhold(fp); FILEDESC_XUNLOCK(fdp); lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; lf.l_type = F_UNLCK; vp = fp->f_vnode; (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, &lf, F_POSIX); FILEDESC_XLOCK(fdp); fdrop(fp, td); } } retry: if (fdtol->fdl_refcount == 1) { if (fdp->fd_holdleaderscount > 0 && (p->p_leader->p_flag & P_ADVLOCK) != 0) { /* * close() or kern_dup() has cleared a reference * in a shared file descriptor table. */ fdp->fd_holdleaderswakeup = 1; sx_sleep(&fdp->fd_holdleaderscount, FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); goto retry; } if (fdtol->fdl_holdcount > 0) { /* * Ensure that fdtol->fdl_leader remains * valid in closef(). */ fdtol->fdl_wakeup = 1; sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); goto retry; } } fdtol->fdl_refcount--; if (fdtol->fdl_refcount == 0 && fdtol->fdl_holdcount == 0) { fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; fdtol->fdl_prev->fdl_next = fdtol->fdl_next; } else fdtol = NULL; p->p_fdtol = NULL; FILEDESC_XUNLOCK(fdp); if (fdtol != NULL) free(fdtol, M_FILEDESC_TO_LEADER); } /* * Release a filedesc structure. */ static void fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose) { struct filedesc0 *fdp0; struct freetable *ft, *tft; struct filedescent *fde; struct file *fp; int i; for (i = 0; i <= fdp->fd_lastfile; i++) { fde = &fdp->fd_ofiles[i]; fp = fde->fde_file; if (fp != NULL) { fdefree_last(fde); if (needclose) (void) closef(fp, td); else fdrop(fp, td); } } if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) free(fdp->fd_map, M_FILEDESC); if (fdp->fd_nfiles > NDFILE) free(fdp->fd_files, M_FILEDESC); fdp0 = (struct filedesc0 *)fdp; SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft) free(ft->ft_table, M_FILEDESC); fddrop(fdp); } void fdescfree(struct thread *td) { struct proc *p; struct filedesc *fdp; struct vnode *cdir, *jdir, *rdir; p = td->td_proc; fdp = p->p_fd; MPASS(fdp != NULL); #ifdef RACCT if (racct_enable) { PROC_LOCK(p); racct_set(p, RACCT_NOFILE, 0); PROC_UNLOCK(p); } #endif if (p->p_fdtol != NULL) fdclearlocks(td); PROC_LOCK(p); p->p_fd = NULL; PROC_UNLOCK(p); if (refcount_release(&fdp->fd_refcnt) == 0) return; FILEDESC_XLOCK(fdp); cdir = fdp->fd_cdir; fdp->fd_cdir = NULL; rdir = fdp->fd_rdir; fdp->fd_rdir = NULL; jdir = fdp->fd_jdir; fdp->fd_jdir = NULL; FILEDESC_XUNLOCK(fdp); if (cdir != NULL) vrele(cdir); if (rdir != NULL) vrele(rdir); if (jdir != NULL) vrele(jdir); fdescfree_fds(td, fdp, 1); } void fdescfree_remapped(struct filedesc *fdp) { if (fdp->fd_cdir != NULL) vrele(fdp->fd_cdir); if (fdp->fd_rdir != NULL) vrele(fdp->fd_rdir); if (fdp->fd_jdir != NULL) vrele(fdp->fd_jdir); fdescfree_fds(curthread, fdp, 0); } /* * For setugid programs, we don't want to people to use that setugidness * to generate error messages which write to a file which otherwise would * otherwise be off-limits to the process. We check for filesystems where * the vnode can change out from under us after execve (like [lin]procfs). * * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is * sufficient. We also don't check for setugidness since we know we are. */ static bool is_unsafe(struct file *fp) { struct vnode *vp; if (fp->f_type != DTYPE_VNODE) return (false); vp = fp->f_vnode; return ((vp->v_vflag & VV_PROCDEP) != 0); } /* * Make this setguid thing safe, if at all possible. */ void fdsetugidsafety(struct thread *td) { struct filedesc *fdp; struct file *fp; int i; fdp = td->td_proc->p_fd; KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); MPASS(fdp->fd_nfiles >= 3); for (i = 0; i <= 2; i++) { fp = fdp->fd_ofiles[i].fde_file; if (fp != NULL && is_unsafe(fp)) { FILEDESC_XLOCK(fdp); knote_fdclose(td, i); /* * NULL-out descriptor prior to close to avoid * a race while close blocks. */ fdfree(fdp, i); FILEDESC_XUNLOCK(fdp); (void) closef(fp, td); } } } /* * If a specific file object occupies a specific file descriptor, close the * file descriptor entry and drop a reference on the file object. This is a * convenience function to handle a subsequent error in a function that calls * falloc() that handles the race that another thread might have closed the * file descriptor out from under the thread creating the file object. */ void fdclose(struct thread *td, struct file *fp, int idx) { struct filedesc *fdp = td->td_proc->p_fd; FILEDESC_XLOCK(fdp); if (fdp->fd_ofiles[idx].fde_file == fp) { fdfree(fdp, idx); FILEDESC_XUNLOCK(fdp); fdrop(fp, td); } else FILEDESC_XUNLOCK(fdp); } /* * Close any files on exec? */ void fdcloseexec(struct thread *td) { struct filedesc *fdp; struct filedescent *fde; struct file *fp; int i; fdp = td->td_proc->p_fd; KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); for (i = 0; i <= fdp->fd_lastfile; i++) { fde = &fdp->fd_ofiles[i]; fp = fde->fde_file; if (fp != NULL && (fp->f_type == DTYPE_MQUEUE || (fde->fde_flags & UF_EXCLOSE))) { FILEDESC_XLOCK(fdp); fdfree(fdp, i); (void) closefp(fdp, i, fp, td, 0); FILEDESC_UNLOCK_ASSERT(fdp); } } } /* * It is unsafe for set[ug]id processes to be started with file * descriptors 0..2 closed, as these descriptors are given implicit * significance in the Standard C library. fdcheckstd() will create a * descriptor referencing /dev/null for each of stdin, stdout, and * stderr that is not already open. */ int fdcheckstd(struct thread *td) { struct filedesc *fdp; register_t save; int i, error, devnull; fdp = td->td_proc->p_fd; KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); MPASS(fdp->fd_nfiles >= 3); devnull = -1; for (i = 0; i <= 2; i++) { if (fdp->fd_ofiles[i].fde_file != NULL) continue; save = td->td_retval[0]; if (devnull != -1) { error = kern_dup(td, FDDUP_FIXED, 0, devnull, i); } else { error = kern_openat(td, AT_FDCWD, "/dev/null", UIO_SYSSPACE, O_RDWR, 0); if (error == 0) { devnull = td->td_retval[0]; KASSERT(devnull == i, ("we didn't get our fd")); } } td->td_retval[0] = save; if (error != 0) return (error); } return (0); } /* * Internal form of close. Decrement reference count on file structure. * Note: td may be NULL when closing a file that was being passed in a * message. * * XXXRW: Giant is not required for the caller, but often will be held; this * makes it moderately likely the Giant will be recursed in the VFS case. */ int closef(struct file *fp, struct thread *td) { struct vnode *vp; struct flock lf; struct filedesc_to_leader *fdtol; struct filedesc *fdp; /* * POSIX record locking dictates that any close releases ALL * locks owned by this process. This is handled by setting * a flag in the unlock to free ONLY locks obeying POSIX * semantics, and not to free BSD-style file locks. * If the descriptor was in a message, POSIX-style locks * aren't passed with the descriptor, and the thread pointer * will be NULL. Callers should be careful only to pass a * NULL thread pointer when there really is no owning * context that might have locks, or the locks will be * leaked. */ if (fp->f_type == DTYPE_VNODE && td != NULL) { vp = fp->f_vnode; if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; lf.l_type = F_UNLCK; (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, F_UNLCK, &lf, F_POSIX); } fdtol = td->td_proc->p_fdtol; if (fdtol != NULL) { /* * Handle special case where file descriptor table is * shared between multiple process leaders. */ fdp = td->td_proc->p_fd; FILEDESC_XLOCK(fdp); for (fdtol = fdtol->fdl_next; fdtol != td->td_proc->p_fdtol; fdtol = fdtol->fdl_next) { if ((fdtol->fdl_leader->p_flag & P_ADVLOCK) == 0) continue; fdtol->fdl_holdcount++; FILEDESC_XUNLOCK(fdp); lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; lf.l_type = F_UNLCK; vp = fp->f_vnode; (void) VOP_ADVLOCK(vp, (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, F_POSIX); FILEDESC_XLOCK(fdp); fdtol->fdl_holdcount--; if (fdtol->fdl_holdcount == 0 && fdtol->fdl_wakeup != 0) { fdtol->fdl_wakeup = 0; wakeup(fdtol); } } FILEDESC_XUNLOCK(fdp); } } return (fdrop(fp, td)); } /* * Initialize the file pointer with the specified properties. * * The ops are set with release semantics to be certain that the flags, type, * and data are visible when ops is. This is to prevent ops methods from being * called with bad data. */ void finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) { fp->f_data = data; fp->f_flag = flag; fp->f_type = type; atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); } int fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, struct file **fpp, struct filecaps *havecapsp) { struct filedescent *fde; int error; FILEDESC_LOCK_ASSERT(fdp); fde = fdeget_locked(fdp, fd); if (fde == NULL) { error = EBADF; goto out; } #ifdef CAPABILITIES error = cap_check(cap_rights_fde(fde), needrightsp); if (error != 0) goto out; #endif if (havecapsp != NULL) filecaps_copy(&fde->fde_caps, havecapsp, true); fhold(fde->fde_file); *fpp = fde->fde_file; error = 0; out: return (error); } int fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp, struct file **fpp, struct filecaps *havecapsp) { struct filedesc *fdp; struct file *fp; int error; seq_t seq; fdp = td->td_proc->p_fd; for (;;) { error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq); if (error != 0) return (error); if (havecapsp != NULL) { if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecapsp, false)) { fdrop(fp, td); goto get_locked; } } if (!fd_modified(fdp, fd, seq)) break; fdrop(fp, td); } *fpp = fp; return (0); get_locked: FILEDESC_SLOCK(fdp); error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp); FILEDESC_SUNLOCK(fdp); return (error); } int fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, struct file **fpp, seq_t *seqp) { #ifdef CAPABILITIES struct filedescent *fde; #endif struct fdescenttbl *fdt; struct file *fp; u_int count; #ifdef CAPABILITIES seq_t seq; cap_rights_t haverights; int error; #endif fdt = fdp->fd_files; if ((u_int)fd >= fdt->fdt_nfiles) return (EBADF); /* * Fetch the descriptor locklessly. We avoid fdrop() races by * never raising a refcount above 0. To accomplish this we have * to use a cmpset loop rather than an atomic_add. The descriptor * must be re-verified once we acquire a reference to be certain * that the identity is still correct and we did not lose a race * due to preemption. */ for (;;) { #ifdef CAPABILITIES seq = seq_read(fd_seq(fdt, fd)); fde = &fdt->fdt_ofiles[fd]; haverights = *cap_rights_fde(fde); fp = fde->fde_file; if (!seq_consistent(fd_seq(fdt, fd), seq)) { cpu_spinwait(); continue; } #else fp = fdt->fdt_ofiles[fd].fde_file; #endif if (fp == NULL) return (EBADF); #ifdef CAPABILITIES error = cap_check(&haverights, needrightsp); if (error != 0) return (error); #endif retry: count = fp->f_count; if (count == 0) { /* * Force a reload. Other thread could reallocate the * table before this fd was closed, so it possible that * there is a stale fp pointer in cached version. */ fdt = *(struct fdescenttbl * volatile *)&(fdp->fd_files); continue; } /* * Use an acquire barrier to force re-reading of fdt so it is * refreshed for verification. */ if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) == 0) goto retry; fdt = fdp->fd_files; #ifdef CAPABILITIES if (seq_consistent_nomb(fd_seq(fdt, fd), seq)) #else if (fp == fdt->fdt_ofiles[fd].fde_file) #endif break; fdrop(fp, curthread); } *fpp = fp; if (seqp != NULL) { #ifdef CAPABILITIES *seqp = seq; #endif } return (0); } /* * Extract the file pointer associated with the specified descriptor for the * current user process. * * If the descriptor doesn't exist or doesn't match 'flags', EBADF is * returned. * * File's rights will be checked against the capability rights mask. * * If an error occurred the non-zero error is returned and *fpp is set to * NULL. Otherwise *fpp is held and set and zero is returned. Caller is * responsible for fdrop(). */ static __inline int _fget(struct thread *td, int fd, struct file **fpp, int flags, cap_rights_t *needrightsp, seq_t *seqp) { struct filedesc *fdp; struct file *fp; int error; *fpp = NULL; fdp = td->td_proc->p_fd; error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp); if (error != 0) return (error); if (fp->f_ops == &badfileops) { fdrop(fp, td); return (EBADF); } /* * FREAD and FWRITE failure return EBADF as per POSIX. */ error = 0; switch (flags) { case FREAD: case FWRITE: if ((fp->f_flag & flags) == 0) error = EBADF; break; case FEXEC: if ((fp->f_flag & (FREAD | FEXEC)) == 0 || ((fp->f_flag & FWRITE) != 0)) error = EBADF; break; case 0: break; default: KASSERT(0, ("wrong flags")); } if (error != 0) { fdrop(fp, td); return (error); } *fpp = fp; return (0); } int fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) { return (_fget(td, fd, fpp, 0, rightsp, NULL)); } int fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp, struct file **fpp) { int error; #ifndef CAPABILITIES error = _fget(td, fd, fpp, 0, rightsp, NULL); if (maxprotp != NULL) *maxprotp = VM_PROT_ALL; #else struct filedesc *fdp = td->td_proc->p_fd; seq_t seq; MPASS(cap_rights_is_set(rightsp, CAP_MMAP)); for (;;) { error = _fget(td, fd, fpp, 0, rightsp, &seq); if (error != 0) return (error); /* * If requested, convert capability rights to access flags. */ if (maxprotp != NULL) *maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd)); if (!fd_modified(fdp, fd, seq)) break; fdrop(*fpp, td); } #endif return (error); } int fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) { return (_fget(td, fd, fpp, FREAD, rightsp, NULL)); } int fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) { return (_fget(td, fd, fpp, FWRITE, rightsp, NULL)); } int fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl, struct file **fpp) { struct filedesc *fdp = td->td_proc->p_fd; #ifndef CAPABILITIES return (fget_unlocked(fdp, fd, rightsp, fpp, NULL)); #else int error; seq_t seq; MPASS(cap_rights_is_set(rightsp, CAP_FCNTL)); for (;;) { error = fget_unlocked(fdp, fd, rightsp, fpp, &seq); if (error != 0) return (error); error = cap_fcntl_check(fdp, fd, needfcntl); if (!fd_modified(fdp, fd, seq)) break; fdrop(*fpp, td); } if (error != 0) { fdrop(*fpp, td); *fpp = NULL; } return (error); #endif } /* * Like fget() but loads the underlying vnode, or returns an error if the * descriptor does not represent a vnode. Note that pipes use vnodes but * never have VM objects. The returned vnode will be vref()'d. * * XXX: what about the unused flags ? */ static __inline int _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp, struct vnode **vpp) { struct file *fp; int error; *vpp = NULL; error = _fget(td, fd, &fp, flags, needrightsp, NULL); if (error != 0) return (error); if (fp->f_vnode == NULL) { error = EINVAL; } else { *vpp = fp->f_vnode; vref(*vpp); } fdrop(fp, td); return (error); } int fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) { return (_fgetvp(td, fd, 0, rightsp, vpp)); } int fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, struct filecaps *havecaps, struct vnode **vpp) { struct filedesc *fdp; struct file *fp; #ifdef CAPABILITIES int error; #endif fdp = td->td_proc->p_fd; fp = fget_locked(fdp, fd); if (fp == NULL || fp->f_ops == &badfileops) return (EBADF); #ifdef CAPABILITIES error = cap_check(cap_rights(fdp, fd), needrightsp); if (error != 0) return (error); #endif if (fp->f_vnode == NULL) return (EINVAL); *vpp = fp->f_vnode; vref(*vpp); filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps, true); return (0); } int fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) { return (_fgetvp(td, fd, FREAD, rightsp, vpp)); } int fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) { return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); } #ifdef notyet int fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) { return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); } #endif /* * Like fget() but loads the underlying socket, or returns an error if the * descriptor does not represent a socket. * * We bump the ref count on the returned socket. XXX Also obtain the SX lock * in the future. * * Note: fgetsock() and fputsock() are deprecated, as consumers should rely * on their file descriptor reference to prevent the socket from being free'd * during use. */ int fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp, u_int *fflagp) { struct file *fp; int error; *spp = NULL; if (fflagp != NULL) *fflagp = 0; if ((error = _fget(td, fd, &fp, 0, rightsp, NULL)) != 0) return (error); if (fp->f_type != DTYPE_SOCKET) { error = ENOTSOCK; } else { *spp = fp->f_data; if (fflagp) *fflagp = fp->f_flag; SOCK_LOCK(*spp); soref(*spp); SOCK_UNLOCK(*spp); } fdrop(fp, td); return (error); } /* * Drop the reference count on the socket and XXX release the SX lock in the * future. The last reference closes the socket. * * Note: fputsock() is deprecated, see comment for fgetsock(). */ void fputsock(struct socket *so) { ACCEPT_LOCK(); SOCK_LOCK(so); CURVNET_SET(so->so_vnet); sorele(so); CURVNET_RESTORE(); } /* * Handle the last reference to a file being closed. */ int _fdrop(struct file *fp, struct thread *td) { int error; if (fp->f_count != 0) panic("fdrop: count %d", fp->f_count); error = fo_close(fp, td); atomic_subtract_int(&openfiles, 1); crfree(fp->f_cred); free(fp->f_advice, M_FADVISE); uma_zfree(file_zone, fp); return (error); } /* * Apply an advisory lock on a file descriptor. * * Just attempt to get a record lock of the requested type on the entire file * (l_whence = SEEK_SET, l_start = 0, l_len = 0). */ #ifndef _SYS_SYSPROTO_H_ struct flock_args { int fd; int how; }; #endif /* ARGSUSED */ int sys_flock(struct thread *td, struct flock_args *uap) { struct file *fp; struct vnode *vp; struct flock lf; cap_rights_t rights; int error; error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp); if (error != 0) return (error); if (fp->f_type != DTYPE_VNODE) { fdrop(fp, td); return (EOPNOTSUPP); } vp = fp->f_vnode; lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; if (uap->how & LOCK_UN) { lf.l_type = F_UNLCK; atomic_clear_int(&fp->f_flag, FHASLOCK); error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); goto done2; } if (uap->how & LOCK_EX) lf.l_type = F_WRLCK; else if (uap->how & LOCK_SH) lf.l_type = F_RDLCK; else { error = EBADF; goto done2; } atomic_set_int(&fp->f_flag, FHASLOCK); error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); done2: fdrop(fp, td); return (error); } /* * Duplicate the specified descriptor to a free descriptor. */ int dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, int openerror, int *indxp) { struct filedescent *newfde, *oldfde; struct file *fp; int error, indx; KASSERT(openerror == ENODEV || openerror == ENXIO, ("unexpected error %d in %s", openerror, __func__)); /* * If the to-be-dup'd fd number is greater than the allowed number * of file descriptors, or the fd to be dup'd has already been * closed, then reject. */ FILEDESC_XLOCK(fdp); if ((fp = fget_locked(fdp, dfd)) == NULL) { FILEDESC_XUNLOCK(fdp); return (EBADF); } error = fdalloc(td, 0, &indx); if (error != 0) { FILEDESC_XUNLOCK(fdp); return (error); } /* * There are two cases of interest here. * * For ENODEV simply dup (dfd) to file descriptor (indx) and return. * * For ENXIO steal away the file structure from (dfd) and store it in * (indx). (dfd) is effectively closed by this operation. */ switch (openerror) { case ENODEV: /* * Check that the mode the file is being opened for is a * subset of the mode of the existing descriptor. */ if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { fdunused(fdp, indx); FILEDESC_XUNLOCK(fdp); return (EACCES); } fhold(fp); newfde = &fdp->fd_ofiles[indx]; oldfde = &fdp->fd_ofiles[dfd]; #ifdef CAPABILITIES seq_write_begin(&newfde->fde_seq); #endif memcpy(newfde, oldfde, fde_change_size); filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps, true); #ifdef CAPABILITIES seq_write_end(&newfde->fde_seq); #endif break; case ENXIO: /* * Steal away the file pointer from dfd and stuff it into indx. */ newfde = &fdp->fd_ofiles[indx]; oldfde = &fdp->fd_ofiles[dfd]; #ifdef CAPABILITIES seq_write_begin(&newfde->fde_seq); #endif memcpy(newfde, oldfde, fde_change_size); oldfde->fde_file = NULL; fdunused(fdp, dfd); #ifdef CAPABILITIES seq_write_end(&newfde->fde_seq); #endif break; } FILEDESC_XUNLOCK(fdp); *indxp = indx; return (0); } /* * This sysctl determines if we will allow a process to chroot(2) if it * has a directory open: * 0: disallowed for all processes. * 1: allowed for processes that were not already chroot(2)'ed. * 2: allowed for all processes. */ static int chroot_allow_open_directories = 1; SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW, &chroot_allow_open_directories, 0, "Allow a process to chroot(2) if it has a directory open"); /* * Helper function for raised chroot(2) security function: Refuse if * any filedescriptors are open directories. */ static int chroot_refuse_vdir_fds(struct filedesc *fdp) { struct vnode *vp; struct file *fp; int fd; FILEDESC_LOCK_ASSERT(fdp); for (fd = 0; fd <= fdp->fd_lastfile; fd++) { fp = fget_locked(fdp, fd); if (fp == NULL) continue; if (fp->f_type == DTYPE_VNODE) { vp = fp->f_vnode; if (vp->v_type == VDIR) return (EPERM); } } return (0); } /* * Common routine for kern_chroot() and jail_attach(). The caller is * responsible for invoking priv_check() and mac_vnode_check_chroot() to * authorize this operation. */ int pwd_chroot(struct thread *td, struct vnode *vp) { struct filedesc *fdp; struct vnode *oldvp; int error; fdp = td->td_proc->p_fd; FILEDESC_XLOCK(fdp); if (chroot_allow_open_directories == 0 || (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) { error = chroot_refuse_vdir_fds(fdp); if (error != 0) { FILEDESC_XUNLOCK(fdp); return (error); } } oldvp = fdp->fd_rdir; VREF(vp); fdp->fd_rdir = vp; if (fdp->fd_jdir == NULL) { VREF(vp); fdp->fd_jdir = vp; } FILEDESC_XUNLOCK(fdp); vrele(oldvp); return (0); } void pwd_chdir(struct thread *td, struct vnode *vp) { struct filedesc *fdp; struct vnode *oldvp; fdp = td->td_proc->p_fd; FILEDESC_XLOCK(fdp); VNASSERT(vp->v_usecount > 0, vp, ("chdir to a vnode with zero usecount")); oldvp = fdp->fd_cdir; fdp->fd_cdir = vp; FILEDESC_XUNLOCK(fdp); vrele(oldvp); } /* * Scan all active processes and prisons to see if any of them have a current * or root directory of `olddp'. If so, replace them with the new mount point. */ void mountcheckdirs(struct vnode *olddp, struct vnode *newdp) { struct filedesc *fdp; struct prison *pr; struct proc *p; int nrele; if (vrefcnt(olddp) == 1) return; nrele = 0; sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); fdp = fdhold(p); PROC_UNLOCK(p); if (fdp == NULL) continue; FILEDESC_XLOCK(fdp); if (fdp->fd_cdir == olddp) { vref(newdp); fdp->fd_cdir = newdp; nrele++; } if (fdp->fd_rdir == olddp) { vref(newdp); fdp->fd_rdir = newdp; nrele++; } if (fdp->fd_jdir == olddp) { vref(newdp); fdp->fd_jdir = newdp; nrele++; } FILEDESC_XUNLOCK(fdp); fddrop(fdp); } sx_sunlock(&allproc_lock); if (rootvnode == olddp) { vref(newdp); rootvnode = newdp; nrele++; } mtx_lock(&prison0.pr_mtx); if (prison0.pr_root == olddp) { vref(newdp); prison0.pr_root = newdp; nrele++; } mtx_unlock(&prison0.pr_mtx); sx_slock(&allprison_lock); TAILQ_FOREACH(pr, &allprison, pr_list) { mtx_lock(&pr->pr_mtx); if (pr->pr_root == olddp) { vref(newdp); pr->pr_root = newdp; nrele++; } mtx_unlock(&pr->pr_mtx); } sx_sunlock(&allprison_lock); while (nrele--) vrele(olddp); } struct filedesc_to_leader * filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) { struct filedesc_to_leader *fdtol; fdtol = malloc(sizeof(struct filedesc_to_leader), M_FILEDESC_TO_LEADER, M_WAITOK); fdtol->fdl_refcount = 1; fdtol->fdl_holdcount = 0; fdtol->fdl_wakeup = 0; fdtol->fdl_leader = leader; if (old != NULL) { FILEDESC_XLOCK(fdp); fdtol->fdl_next = old->fdl_next; fdtol->fdl_prev = old; old->fdl_next = fdtol; fdtol->fdl_next->fdl_prev = fdtol; FILEDESC_XUNLOCK(fdp); } else { fdtol->fdl_next = fdtol; fdtol->fdl_prev = fdtol; } return (fdtol); } static int sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS) { struct filedesc *fdp; int i, count, slots; if (*(int *)arg1 != 0) return (EINVAL); fdp = curproc->p_fd; count = 0; FILEDESC_SLOCK(fdp); slots = NDSLOTS(fdp->fd_lastfile + 1); for (i = 0; i < slots; i++) count += bitcountl(fdp->fd_map[i]); FILEDESC_SUNLOCK(fdp); return (SYSCTL_OUT(req, &count, sizeof(count))); } static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds, CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds, "Number of open file descriptors"); /* * Get file structures globally. */ static int sysctl_kern_file(SYSCTL_HANDLER_ARGS) { struct xfile xf; struct filedesc *fdp; struct file *fp; struct proc *p; int error, n; error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); if (req->oldptr == NULL) { n = 0; sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); if (p->p_state == PRS_NEW) { PROC_UNLOCK(p); continue; } fdp = fdhold(p); PROC_UNLOCK(p); if (fdp == NULL) continue; /* overestimates sparse tables. */ if (fdp->fd_lastfile > 0) n += fdp->fd_lastfile; fddrop(fdp); } sx_sunlock(&allproc_lock); return (SYSCTL_OUT(req, 0, n * sizeof(xf))); } error = 0; bzero(&xf, sizeof(xf)); xf.xf_size = sizeof(xf); sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); if (p->p_state == PRS_NEW) { PROC_UNLOCK(p); continue; } if (p_cansee(req->td, p) != 0) { PROC_UNLOCK(p); continue; } xf.xf_pid = p->p_pid; xf.xf_uid = p->p_ucred->cr_uid; fdp = fdhold(p); PROC_UNLOCK(p); if (fdp == NULL) continue; FILEDESC_SLOCK(fdp); for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) { if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) continue; xf.xf_fd = n; xf.xf_file = fp; xf.xf_data = fp->f_data; xf.xf_vnode = fp->f_vnode; xf.xf_type = fp->f_type; xf.xf_count = fp->f_count; xf.xf_msgcount = 0; xf.xf_offset = foffset_get(fp); xf.xf_flag = fp->f_flag; error = SYSCTL_OUT(req, &xf, sizeof(xf)); if (error) break; } FILEDESC_SUNLOCK(fdp); fddrop(fdp); if (error) break; } sx_sunlock(&allproc_lock); return (error); } SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); #ifdef KINFO_FILE_SIZE CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); #endif static int xlate_fflags(int fflags) { static const struct { int fflag; int kf_fflag; } fflags_table[] = { { FAPPEND, KF_FLAG_APPEND }, { FASYNC, KF_FLAG_ASYNC }, { FFSYNC, KF_FLAG_FSYNC }, { FHASLOCK, KF_FLAG_HASLOCK }, { FNONBLOCK, KF_FLAG_NONBLOCK }, { FREAD, KF_FLAG_READ }, { FWRITE, KF_FLAG_WRITE }, { O_CREAT, KF_FLAG_CREAT }, { O_DIRECT, KF_FLAG_DIRECT }, { O_EXCL, KF_FLAG_EXCL }, { O_EXEC, KF_FLAG_EXEC }, { O_EXLOCK, KF_FLAG_EXLOCK }, { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, { O_SHLOCK, KF_FLAG_SHLOCK }, { O_TRUNC, KF_FLAG_TRUNC } }; unsigned int i; int kflags; kflags = 0; for (i = 0; i < nitems(fflags_table); i++) if (fflags & fflags_table[i].fflag) kflags |= fflags_table[i].kf_fflag; return (kflags); } /* Trim unused data from kf_path by truncating the structure size. */ static void pack_kinfo(struct kinfo_file *kif) { kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + strlen(kif->kf_path) + 1; kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); } static void export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp, struct kinfo_file *kif, struct filedesc *fdp, int flags) { int error; bzero(kif, sizeof(*kif)); /* Set a default type to allow for empty fill_kinfo() methods. */ kif->kf_type = KF_TYPE_UNKNOWN; kif->kf_flags = xlate_fflags(fp->f_flag); if (rightsp != NULL) kif->kf_cap_rights = *rightsp; else cap_rights_init(&kif->kf_cap_rights); kif->kf_fd = fd; kif->kf_ref_count = fp->f_count; kif->kf_offset = foffset_get(fp); /* * This may drop the filedesc lock, so the 'fp' cannot be * accessed after this call. */ error = fo_fill_kinfo(fp, kif, fdp); if (error == 0) kif->kf_status |= KF_ATTR_VALID; if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) pack_kinfo(kif); else kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); } static void export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags, struct kinfo_file *kif, int flags) { int error; bzero(kif, sizeof(*kif)); kif->kf_type = KF_TYPE_VNODE; error = vn_fill_kinfo_vnode(vp, kif); if (error == 0) kif->kf_status |= KF_ATTR_VALID; kif->kf_flags = xlate_fflags(fflags); cap_rights_init(&kif->kf_cap_rights); kif->kf_fd = fd; kif->kf_ref_count = -1; kif->kf_offset = -1; if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) pack_kinfo(kif); else kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); vrele(vp); } struct export_fd_buf { struct filedesc *fdp; struct sbuf *sb; ssize_t remainder; struct kinfo_file kif; int flags; }; static int export_kinfo_to_sb(struct export_fd_buf *efbuf) { struct kinfo_file *kif; kif = &efbuf->kif; if (efbuf->remainder != -1) { if (efbuf->remainder < kif->kf_structsize) { /* Terminate export. */ efbuf->remainder = 0; return (0); } efbuf->remainder -= kif->kf_structsize; } return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM); } static int export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp, struct export_fd_buf *efbuf) { int error; if (efbuf->remainder == 0) return (0); export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp, efbuf->flags); FILEDESC_SUNLOCK(efbuf->fdp); error = export_kinfo_to_sb(efbuf); FILEDESC_SLOCK(efbuf->fdp); return (error); } static int export_vnode_to_sb(struct vnode *vp, int fd, int fflags, struct export_fd_buf *efbuf) { int error; if (efbuf->remainder == 0) return (0); if (efbuf->fdp != NULL) FILEDESC_SUNLOCK(efbuf->fdp); export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags); error = export_kinfo_to_sb(efbuf); if (efbuf->fdp != NULL) FILEDESC_SLOCK(efbuf->fdp); return (error); } /* * Store a process file descriptor information to sbuf. * * Takes a locked proc as argument, and returns with the proc unlocked. */ int kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen, int flags) { struct file *fp; struct filedesc *fdp; struct export_fd_buf *efbuf; struct vnode *cttyvp, *textvp, *tracevp; int error, i; cap_rights_t rights; PROC_LOCK_ASSERT(p, MA_OWNED); /* ktrace vnode */ tracevp = p->p_tracevp; if (tracevp != NULL) vref(tracevp); /* text vnode */ textvp = p->p_textvp; if (textvp != NULL) vref(textvp); /* Controlling tty. */ cttyvp = NULL; if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { cttyvp = p->p_pgrp->pg_session->s_ttyvp; if (cttyvp != NULL) vref(cttyvp); } fdp = fdhold(p); PROC_UNLOCK(p); efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); efbuf->fdp = NULL; efbuf->sb = sb; efbuf->remainder = maxlen; efbuf->flags = flags; if (tracevp != NULL) export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE, efbuf); if (textvp != NULL) export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf); if (cttyvp != NULL) export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE, efbuf); error = 0; if (fdp == NULL) goto fail; efbuf->fdp = fdp; FILEDESC_SLOCK(fdp); /* working directory */ if (fdp->fd_cdir != NULL) { vref(fdp->fd_cdir); export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf); } /* root directory */ if (fdp->fd_rdir != NULL) { vref(fdp->fd_rdir); export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf); } /* jail directory */ if (fdp->fd_jdir != NULL) { vref(fdp->fd_jdir); export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf); } for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) continue; #ifdef CAPABILITIES rights = *cap_rights(fdp, i); #else /* !CAPABILITIES */ cap_rights_init(&rights); #endif /* * Create sysctl entry. It is OK to drop the filedesc * lock inside of export_file_to_sb() as we will * re-validate and re-evaluate its properties when the * loop continues. */ error = export_file_to_sb(fp, i, &rights, efbuf); if (error != 0 || efbuf->remainder == 0) break; } FILEDESC_SUNLOCK(fdp); fddrop(fdp); fail: free(efbuf, M_TEMP); return (error); } #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) /* * Get per-process file descriptors for use by procstat(1), et al. */ static int sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) { struct sbuf sb; struct proc *p; ssize_t maxlen; int error, error2, *name; name = (int *)arg1; sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); sbuf_clear_flags(&sb, SBUF_INCLUDENUL); error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); if (error != 0) { sbuf_delete(&sb); return (error); } maxlen = req->oldptr != NULL ? req->oldlen : -1; error = kern_proc_filedesc_out(p, &sb, maxlen, KERN_FILEDESC_PACK_KINFO); error2 = sbuf_finish(&sb); sbuf_delete(&sb); return (error != 0 ? error : error2); } #ifdef KINFO_OFILE_SIZE CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); #endif #ifdef COMPAT_FREEBSD7 static void kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif) { okif->kf_structsize = sizeof(*okif); okif->kf_type = kif->kf_type; okif->kf_fd = kif->kf_fd; okif->kf_ref_count = kif->kf_ref_count; okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE | KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK | KF_FLAG_DIRECT | KF_FLAG_HASLOCK); okif->kf_offset = kif->kf_offset; okif->kf_vnode_type = kif->kf_vnode_type; okif->kf_sock_domain = kif->kf_sock_domain; okif->kf_sock_type = kif->kf_sock_type; okif->kf_sock_protocol = kif->kf_sock_protocol; strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path)); okif->kf_sa_local = kif->kf_sa_local; okif->kf_sa_peer = kif->kf_sa_peer; } static int export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif, struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req) { int error; vref(vp); FILEDESC_SUNLOCK(fdp); export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO); kinfo_to_okinfo(kif, okif); error = SYSCTL_OUT(req, okif, sizeof(*okif)); FILEDESC_SLOCK(fdp); return (error); } /* * Get per-process file descriptors for use by procstat(1), et al. */ static int sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) { struct kinfo_ofile *okif; struct kinfo_file *kif; struct filedesc *fdp; int error, i, *name; struct file *fp; struct proc *p; name = (int *)arg1; error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); if (error != 0) return (error); fdp = fdhold(p); PROC_UNLOCK(p); if (fdp == NULL) return (ENOENT); kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK); FILEDESC_SLOCK(fdp); if (fdp->fd_cdir != NULL) export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, okif, fdp, req); if (fdp->fd_rdir != NULL) export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, okif, fdp, req); if (fdp->fd_jdir != NULL) export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, okif, fdp, req); for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) continue; export_file_to_kinfo(fp, i, NULL, kif, fdp, KERN_FILEDESC_PACK_KINFO); FILEDESC_SUNLOCK(fdp); kinfo_to_okinfo(kif, okif); error = SYSCTL_OUT(req, okif, sizeof(*okif)); FILEDESC_SLOCK(fdp); if (error) break; } FILEDESC_SUNLOCK(fdp); fddrop(fdp); free(kif, M_TEMP); free(okif, M_TEMP); return (0); } static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, "Process ofiledesc entries"); #endif /* COMPAT_FREEBSD7 */ int vntype_to_kinfo(int vtype) { struct { int vtype; int kf_vtype; } vtypes_table[] = { { VBAD, KF_VTYPE_VBAD }, { VBLK, KF_VTYPE_VBLK }, { VCHR, KF_VTYPE_VCHR }, { VDIR, KF_VTYPE_VDIR }, { VFIFO, KF_VTYPE_VFIFO }, { VLNK, KF_VTYPE_VLNK }, { VNON, KF_VTYPE_VNON }, { VREG, KF_VTYPE_VREG }, { VSOCK, KF_VTYPE_VSOCK } }; unsigned int i; /* * Perform vtype translation. */ for (i = 0; i < nitems(vtypes_table); i++) if (vtypes_table[i].vtype == vtype) return (vtypes_table[i].kf_vtype); return (KF_VTYPE_UNKNOWN); } static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, "Process filedesc entries"); /* * Store a process current working directory information to sbuf. * * Takes a locked proc as argument, and returns with the proc unlocked. */ int kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) { struct filedesc *fdp; struct export_fd_buf *efbuf; int error; PROC_LOCK_ASSERT(p, MA_OWNED); fdp = fdhold(p); PROC_UNLOCK(p); if (fdp == NULL) return (EINVAL); efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); efbuf->fdp = fdp; efbuf->sb = sb; efbuf->remainder = maxlen; FILEDESC_SLOCK(fdp); if (fdp->fd_cdir == NULL) error = EINVAL; else { vref(fdp->fd_cdir); error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf); } FILEDESC_SUNLOCK(fdp); fddrop(fdp); free(efbuf, M_TEMP); return (error); } /* * Get per-process current working directory. */ static int sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS) { struct sbuf sb; struct proc *p; ssize_t maxlen; int error, error2, *name; name = (int *)arg1; sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req); sbuf_clear_flags(&sb, SBUF_INCLUDENUL); error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); if (error != 0) { sbuf_delete(&sb); return (error); } maxlen = req->oldptr != NULL ? req->oldlen : -1; error = kern_proc_cwd_out(p, &sb, maxlen); error2 = sbuf_finish(&sb); sbuf_delete(&sb); return (error != 0 ? error : error2); } static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_cwd, "Process current working directory"); #ifdef DDB /* * For the purposes of debugging, generate a human-readable string for the * file type. */ static const char * file_type_to_name(short type) { switch (type) { case 0: return ("zero"); case DTYPE_VNODE: return ("vnod"); case DTYPE_SOCKET: return ("sock"); case DTYPE_PIPE: return ("pipe"); case DTYPE_FIFO: return ("fifo"); case DTYPE_KQUEUE: return ("kque"); case DTYPE_CRYPTO: return ("crpt"); case DTYPE_MQUEUE: return ("mque"); case DTYPE_SHM: return ("shm"); case DTYPE_SEM: return ("ksem"); default: return ("unkn"); } } /* * For the purposes of debugging, identify a process (if any, perhaps one of * many) that references the passed file in its file descriptor array. Return * NULL if none. */ static struct proc * file_to_first_proc(struct file *fp) { struct filedesc *fdp; struct proc *p; int n; FOREACH_PROC_IN_SYSTEM(p) { if (p->p_state == PRS_NEW) continue; fdp = p->p_fd; if (fdp == NULL) continue; for (n = 0; n <= fdp->fd_lastfile; n++) { if (fp == fdp->fd_ofiles[n].fde_file) return (p); } } return (NULL); } static void db_print_file(struct file *fp, int header) { struct proc *p; if (header) db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", "File", "Type", "Data", "Flag", "GCFl", "Count", "MCount", "Vnode", "FPID", "FCmd"); p = file_to_first_proc(fp); db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp, file_type_to_name(fp->f_type), fp->f_data, fp->f_flag, 0, fp->f_count, 0, fp->f_vnode, p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); } DB_SHOW_COMMAND(file, db_show_file) { struct file *fp; if (!have_addr) { db_printf("usage: show file \n"); return; } fp = (struct file *)addr; db_print_file(fp, 1); } DB_SHOW_COMMAND(files, db_show_files) { struct filedesc *fdp; struct file *fp; struct proc *p; int header; int n; header = 1; FOREACH_PROC_IN_SYSTEM(p) { if (p->p_state == PRS_NEW) continue; if ((fdp = p->p_fd) == NULL) continue; for (n = 0; n <= fdp->fd_lastfile; ++n) { if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) continue; db_print_file(fp, header); header = 0; } } } #endif SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, &maxfilesperproc, 0, "Maximum files allowed open per process"); SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, &maxfiles, 0, "Maximum number of files"); SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); /* ARGSUSED*/ static void filelistinit(void *dummy) { file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); } SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); /*-------------------------------------------------------------------*/ static int badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { return (EBADF); } static int badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td) { return (EINVAL); } static int badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td) { return (EBADF); } static int badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td) { return (0); } static int badfo_kqfilter(struct file *fp, struct knote *kn) { return (EBADF); } static int badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td) { return (EBADF); } static int badfo_close(struct file *fp, struct thread *td) { return (0); } static int badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td) { return (EBADF); } static int badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, struct thread *td) { return (EBADF); } static int badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, struct thread *td) { return (EBADF); } static int badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) { return (0); } struct fileops badfileops = { .fo_read = badfo_readwrite, .fo_write = badfo_readwrite, .fo_truncate = badfo_truncate, .fo_ioctl = badfo_ioctl, .fo_poll = badfo_poll, .fo_kqfilter = badfo_kqfilter, .fo_stat = badfo_stat, .fo_close = badfo_close, .fo_chmod = badfo_chmod, .fo_chown = badfo_chown, .fo_sendfile = badfo_sendfile, .fo_fill_kinfo = badfo_fill_kinfo, }; int invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { return (EOPNOTSUPP); } int invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td) { return (EINVAL); } int invfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td) { return (ENOTTY); } int invfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td) { return (poll_no_poll(events)); } int invfo_kqfilter(struct file *fp, struct knote *kn) { return (EINVAL); } int invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td) { return (EINVAL); } int invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, struct thread *td) { return (EINVAL); } int invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, struct thread *td) { return (EINVAL); } /*-------------------------------------------------------------------*/ /* * File Descriptor pseudo-device driver (/dev/fd/). * * Opening minor device N dup()s the file (if any) connected to file * descriptor N belonging to the calling process. Note that this driver * consists of only the ``open()'' routine, because all subsequent * references to this file will be direct to the other driver. * * XXX: we could give this one a cloning event handler if necessary. */ /* ARGSUSED */ static int fdopen(struct cdev *dev, int mode, int type, struct thread *td) { /* * XXX Kludge: set curthread->td_dupfd to contain the value of the * the file descriptor being sought for duplication. The error * return ensures that the vnode for this device will be released * by vn_open. Open will detect this special error and take the * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN * will simply report the error. */ td->td_dupfd = dev2unit(dev); return (ENODEV); } static struct cdevsw fildesc_cdevsw = { .d_version = D_VERSION, .d_open = fdopen, .d_name = "FD", }; static void fildesc_drvinit(void *unused) { struct cdev *dev; dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, UID_ROOT, GID_WHEEL, 0666, "fd/0"); make_dev_alias(dev, "stdin"); dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, UID_ROOT, GID_WHEEL, 0666, "fd/1"); make_dev_alias(dev, "stdout"); dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, UID_ROOT, GID_WHEEL, 0666, "fd/2"); make_dev_alias(dev, "stderr"); } SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); Index: head/sys/kern/kern_exit.c =================================================================== --- head/sys/kern/kern_exit.c (revision 305831) +++ head/sys/kern/kern_exit.c (revision 305832) @@ -1,1326 +1,1326 @@ /*- * Copyright (c) 1982, 1986, 1989, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_exit.c 8.7 (Berkeley) 2/12/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #include "opt_ktrace.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* for acct_process() function prototype */ #include #include #include #include #include #ifdef KTRACE #include #endif #include #include #include #include #include #include #include #include #include #include #ifdef KDTRACE_HOOKS #include dtrace_execexit_func_t dtrace_fasttrap_exit; #endif SDT_PROVIDER_DECLARE(proc); SDT_PROBE_DEFINE1(proc, , , exit, "int"); /* Hook for NFS teardown procedure. */ void (*nlminfo_release_p)(struct proc *p); struct proc * proc_realparent(struct proc *child) { struct proc *p, *parent; sx_assert(&proctree_lock, SX_LOCKED); if ((child->p_treeflag & P_TREE_ORPHANED) == 0) { if (child->p_oppid == 0 || child->p_pptr->p_pid == child->p_oppid) parent = child->p_pptr; else parent = initproc; return (parent); } for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) { /* Cannot use LIST_PREV(), since the list head is not known. */ p = __containerof(p->p_orphan.le_prev, struct proc, p_orphan.le_next); KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0, ("missing P_ORPHAN %p", p)); } parent = __containerof(p->p_orphan.le_prev, struct proc, p_orphans.lh_first); return (parent); } void reaper_abandon_children(struct proc *p, bool exiting) { struct proc *p1, *p2, *ptmp; sx_assert(&proctree_lock, SX_LOCKED); KASSERT(p != initproc, ("reaper_abandon_children for initproc")); if ((p->p_treeflag & P_TREE_REAPER) == 0) return; p1 = p->p_reaper; LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) { LIST_REMOVE(p2, p_reapsibling); p2->p_reaper = p1; p2->p_reapsubtree = p->p_reapsubtree; LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling); if (exiting && p2->p_pptr == p) { PROC_LOCK(p2); proc_reparent(p2, p1); PROC_UNLOCK(p2); } } KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty")); p->p_treeflag &= ~P_TREE_REAPER; } static void clear_orphan(struct proc *p) { struct proc *p1; sx_assert(&proctree_lock, SA_XLOCKED); if ((p->p_treeflag & P_TREE_ORPHANED) == 0) return; if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) { p1 = LIST_NEXT(p, p_orphan); if (p1 != NULL) p1->p_treeflag |= P_TREE_FIRST_ORPHAN; p->p_treeflag &= ~P_TREE_FIRST_ORPHAN; } LIST_REMOVE(p, p_orphan); p->p_treeflag &= ~P_TREE_ORPHANED; } /* * exit -- death of process. */ void sys_sys_exit(struct thread *td, struct sys_exit_args *uap) { exit1(td, uap->rval, 0); /* NOTREACHED */ } /* * Exit: deallocate address space and other resources, change proc state to * zombie, and unlink proc from allproc and parent's lists. Save exit status * and rusage for wait(). Check for child processes and orphan them. */ void exit1(struct thread *td, int rval, int signo) { struct proc *p, *nq, *q, *t; struct thread *tdt; mtx_assert(&Giant, MA_NOTOWNED); KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo)); p = td->td_proc; /* * XXX in case we're rebooting we just let init die in order to * work around an unsolved stack overflow seen very late during * shutdown on sparc64 when the gmirror worker process exists. */ if (p == initproc && rebooting == 0) { printf("init died (signal %d, exit %d)\n", signo, rval); panic("Going nowhere without my init!"); } /* * Deref SU mp, since the thread does not return to userspace. */ if (softdep_ast_cleanup != NULL) softdep_ast_cleanup(); /* * MUST abort all other threads before proceeding past here. */ PROC_LOCK(p); /* * First check if some other thread or external request got * here before us. If so, act appropriately: exit or suspend. * We must ensure that stop requests are handled before we set * P_WEXIT. */ thread_suspend_check(0); while (p->p_flag & P_HADTHREADS) { /* * Kill off the other threads. This requires * some co-operation from other parts of the kernel * so it may not be instantaneous. With this state set * any thread entering the kernel from userspace will * thread_exit() in trap(). Any thread attempting to * sleep will return immediately with EINTR or EWOULDBLOCK * which will hopefully force them to back out to userland * freeing resources as they go. Any thread attempting * to return to userland will thread_exit() from userret(). * thread_exit() will unsuspend us when the last of the * other threads exits. * If there is already a thread singler after resumption, * calling thread_single will fail; in that case, we just * re-check all suspension request, the thread should * either be suspended there or exit. */ if (!thread_single(p, SINGLE_EXIT)) /* * All other activity in this process is now * stopped. Threading support has been turned * off. */ break; /* * Recheck for new stop or suspend requests which * might appear while process lock was dropped in * thread_single(). */ thread_suspend_check(0); } KASSERT(p->p_numthreads == 1, ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); racct_sub(p, RACCT_NTHR, 1); /* Let event handler change exit status */ p->p_xexit = rval; p->p_xsig = signo; /* * Wakeup anyone in procfs' PIOCWAIT. They should have a hold * on our vmspace, so we should block below until they have * released their reference to us. Note that if they have * requested S_EXIT stops we will block here until they ack * via PIOCCONT. */ _STOPEVENT(p, S_EXIT, 0); /* * Ignore any pending request to stop due to a stop signal. * Once P_WEXIT is set, future requests will be ignored as * well. */ p->p_flag &= ~P_STOPPED_SIG; KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped")); /* * Note that we are exiting and do another wakeup of anyone in * PIOCWAIT in case they aren't listening for S_EXIT stops or * decided to wait again after we told them we are exiting. */ p->p_flag |= P_WEXIT; wakeup(&p->p_stype); /* * Wait for any processes that have a hold on our vmspace to * release their reference. */ while (p->p_lock > 0) msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); PROC_UNLOCK(p); /* Drain the limit callout while we don't have the proc locked */ callout_drain(&p->p_limco); #ifdef AUDIT /* * The Sun BSM exit token contains two components: an exit status as * passed to exit(), and a return value to indicate what sort of exit * it was. The exit status is WEXITSTATUS(rv), but it's not clear * what the return value is. */ AUDIT_ARG_EXIT(rval, 0); AUDIT_SYSCALL_EXIT(0, td); #endif /* Are we a task leader with peers? */ if (p->p_peers != NULL && p == p->p_leader) { mtx_lock(&ppeers_lock); q = p->p_peers; while (q != NULL) { PROC_LOCK(q); kern_psignal(q, SIGKILL); PROC_UNLOCK(q); q = q->p_peers; } while (p->p_peers != NULL) msleep(p, &ppeers_lock, PWAIT, "exit1", 0); mtx_unlock(&ppeers_lock); } /* * Check if any loadable modules need anything done at process exit. * E.g. SYSV IPC stuff. * Event handler could change exit status. * XXX what if one of these generates an error? */ EVENTHANDLER_INVOKE(process_exit, p); /* * If parent is waiting for us to exit or exec, * P_PPWAIT is set; we will wakeup the parent below. */ PROC_LOCK(p); stopprofclock(p); p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); p->p_ptevents = 0; /* * Stop the real interval timer. If the handler is currently * executing, prevent it from rearming itself and let it finish. */ if (timevalisset(&p->p_realtimer.it_value) && _callout_stop_safe(&p->p_itcallout, CS_EXECUTING, NULL) == 0) { timevalclear(&p->p_realtimer.it_interval); msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0); KASSERT(!timevalisset(&p->p_realtimer.it_value), ("realtime timer is still armed")); } PROC_UNLOCK(p); umtx_thread_exit(td); /* * Reset any sigio structures pointing to us as a result of * F_SETOWN with our pid. */ funsetownlst(&p->p_sigiolst); /* * If this process has an nlminfo data area (for lockd), release it */ if (nlminfo_release_p != NULL && p->p_nlminfo != NULL) (*nlminfo_release_p)(p); /* * Close open files and release open-file table. * This may block! */ fdescfree(td); /* * If this thread tickled GEOM, we need to wait for the giggling to * stop before we return to userland */ if (td->td_pflags & TDP_GEOM) g_waitidle(); /* * Remove ourself from our leader's peer list and wake our leader. */ if (p->p_leader->p_peers != NULL) { mtx_lock(&ppeers_lock); if (p->p_leader->p_peers != NULL) { q = p->p_leader; while (q->p_peers != p) q = q->p_peers; q->p_peers = p->p_peers; wakeup(p->p_leader); } mtx_unlock(&ppeers_lock); } vmspace_exit(td); killjobc(); (void)acct_process(td); #ifdef KTRACE ktrprocexit(td); #endif /* * Release reference to text vnode */ if (p->p_textvp != NULL) { vrele(p->p_textvp); p->p_textvp = NULL; } /* * Release our limits structure. */ lim_free(p->p_limit); p->p_limit = NULL; tidhash_remove(td); /* * Remove proc from allproc queue and pidhash chain. * Place onto zombproc. Unlink from parent's child list. */ sx_xlock(&allproc_lock); LIST_REMOVE(p, p_list); LIST_INSERT_HEAD(&zombproc, p, p_list); LIST_REMOVE(p, p_hash); sx_xunlock(&allproc_lock); /* * Call machine-dependent code to release any * machine-dependent resources other than the address space. * The address space is released by "vmspace_exitfree(p)" in * vm_waitproc(). */ cpu_exit(td); WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); /* * Reparent all children processes: * - traced ones to the original parent (or init if we are that parent) * - the rest to init */ sx_xlock(&proctree_lock); q = LIST_FIRST(&p->p_children); if (q != NULL) /* only need this if any child is S_ZOMB */ wakeup(q->p_reaper); for (; q != NULL; q = nq) { nq = LIST_NEXT(q, p_sibling); PROC_LOCK(q); q->p_sigparent = SIGCHLD; if (!(q->p_flag & P_TRACED)) { proc_reparent(q, q->p_reaper); } else { /* * Traced processes are killed since their existence * means someone is screwing up. */ t = proc_realparent(q); if (t == p) { proc_reparent(q, q->p_reaper); } else { PROC_LOCK(t); proc_reparent(q, t); PROC_UNLOCK(t); } /* * Since q was found on our children list, the * proc_reparent() call moved q to the orphan * list due to present P_TRACED flag. Clear * orphan link for q now while q is locked. */ clear_orphan(q); q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); q->p_flag2 &= ~P2_PTRACE_FSTP; q->p_ptevents = 0; FOREACH_THREAD_IN_PROC(q, tdt) { tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG | TDB_FSTP); } kern_psignal(q, SIGKILL); } PROC_UNLOCK(q); } /* * Also get rid of our orphans. */ while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { PROC_LOCK(q); CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid, q->p_pid); clear_orphan(q); PROC_UNLOCK(q); } /* Save exit status. */ PROC_LOCK(p); p->p_xthread = td; /* Tell the prison that we are gone. */ prison_proc_free(p->p_ucred->cr_prison); #ifdef KDTRACE_HOOKS /* * Tell the DTrace fasttrap provider about the exit if it * has declared an interest. */ if (dtrace_fasttrap_exit) dtrace_fasttrap_exit(p); #endif /* * Notify interested parties of our demise. */ KNOTE_LOCKED(p->p_klist, NOTE_EXIT); #ifdef KDTRACE_HOOKS int reason = CLD_EXITED; if (WCOREDUMP(signo)) reason = CLD_DUMPED; else if (WIFSIGNALED(signo)) reason = CLD_KILLED; SDT_PROBE1(proc, , , exit, reason); #endif /* * If this is a process with a descriptor, we may not need to deliver * a signal to the parent. proctree_lock is held over * procdesc_exit() to serialize concurrent calls to close() and * exit(). */ if (p->p_procdesc == NULL || procdesc_exit(p)) { /* * Notify parent that we're gone. If parent has the * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, * notify process 1 instead (and hope it will handle this * situation). */ PROC_LOCK(p->p_pptr); mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) { struct proc *pp; mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); pp = p->p_pptr; PROC_UNLOCK(pp); proc_reparent(p, p->p_reaper); p->p_sigparent = SIGCHLD; PROC_LOCK(p->p_pptr); /* * Notify parent, so in case he was wait(2)ing or * executing waitpid(2) with our pid, he will * continue. */ wakeup(pp); } else mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) childproc_exited(p); else if (p->p_sigparent != 0) { if (p->p_sigparent == SIGCHLD) childproc_exited(p); else /* LINUX thread */ kern_psignal(p->p_pptr, p->p_sigparent); } } else PROC_LOCK(p->p_pptr); sx_xunlock(&proctree_lock); /* * The state PRS_ZOMBIE prevents other proesses from sending * signal to the process, to avoid memory leak, we free memory * for signal queue at the time when the state is set. */ sigqueue_flush(&p->p_sigqueue); sigqueue_flush(&td->td_sigqueue); /* * We have to wait until after acquiring all locks before * changing p_state. We need to avoid all possible context * switches (including ones from blocking on a mutex) while * marked as a zombie. We also have to set the zombie state * before we release the parent process' proc lock to avoid * a lost wakeup. So, we first call wakeup, then we grab the * sched lock, update the state, and release the parent process' * proc lock. */ wakeup(p->p_pptr); cv_broadcast(&p->p_pwait); sched_exit(p->p_pptr, td); PROC_SLOCK(p); p->p_state = PRS_ZOMBIE; PROC_UNLOCK(p->p_pptr); /* * Save our children's rusage information in our exit rusage. */ PROC_STATLOCK(p); ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); PROC_STATUNLOCK(p); /* * Make sure the scheduler takes this thread out of its tables etc. * This will also release this thread's reference to the ucred. * Other thread parts to release include pcb bits and such. */ thread_exit(); } #ifndef _SYS_SYSPROTO_H_ struct abort2_args { char *why; int nargs; void **args; }; #endif int sys_abort2(struct thread *td, struct abort2_args *uap) { struct proc *p = td->td_proc; struct sbuf *sb; void *uargs[16]; int error, i, sig; /* * Do it right now so we can log either proper call of abort2(), or * note, that invalid argument was passed. 512 is big enough to * handle 16 arguments' descriptions with additional comments. */ sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); sbuf_clear(sb); sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", p->p_comm, p->p_pid, td->td_ucred->cr_uid); /* * Since we can't return from abort2(), send SIGKILL in cases, where * abort2() was called improperly */ sig = SIGKILL; /* Prevent from DoSes from user-space. */ if (uap->nargs < 0 || uap->nargs > 16) goto out; if (uap->nargs > 0) { if (uap->args == NULL) goto out; error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); if (error != 0) goto out; } /* * Limit size of 'reason' string to 128. Will fit even when * maximal number of arguments was chosen to be logged. */ if (uap->why != NULL) { error = sbuf_copyin(sb, uap->why, 128); if (error < 0) goto out; } else { sbuf_printf(sb, "(null)"); } if (uap->nargs > 0) { sbuf_printf(sb, "("); for (i = 0;i < uap->nargs; i++) sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); sbuf_printf(sb, ")"); } /* * Final stage: arguments were proper, string has been * successfully copied from userspace, and copying pointers * from user-space succeed. */ sig = SIGABRT; out: if (sig == SIGKILL) { sbuf_trim(sb); sbuf_printf(sb, " (Reason text inaccessible)"); } sbuf_cat(sb, "\n"); sbuf_finish(sb); log(LOG_INFO, "%s", sbuf_data(sb)); sbuf_delete(sb); exit1(td, 0, sig); return (0); } #ifdef COMPAT_43 /* * The dirty work is handled by kern_wait(). */ int owait(struct thread *td, struct owait_args *uap __unused) { int error, status; error = kern_wait(td, WAIT_ANY, &status, 0, NULL); if (error == 0) td->td_retval[1] = status; return (error); } #endif /* COMPAT_43 */ /* * The dirty work is handled by kern_wait(). */ int sys_wait4(struct thread *td, struct wait4_args *uap) { struct rusage ru, *rup; int error, status; if (uap->rusage != NULL) rup = &ru; else rup = NULL; error = kern_wait(td, uap->pid, &status, uap->options, rup); if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) error = copyout(&status, uap->status, sizeof(status)); if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0) error = copyout(&ru, uap->rusage, sizeof(struct rusage)); return (error); } int sys_wait6(struct thread *td, struct wait6_args *uap) { struct __wrusage wru, *wrup; siginfo_t si, *sip; idtype_t idtype; id_t id; int error, status; idtype = uap->idtype; id = uap->id; if (uap->wrusage != NULL) wrup = &wru; else wrup = NULL; if (uap->info != NULL) { sip = &si; bzero(sip, sizeof(*sip)); } else sip = NULL; /* * We expect all callers of wait6() to know about WEXITED and * WTRAPPED. */ error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) error = copyout(&status, uap->status, sizeof(status)); if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0) error = copyout(&wru, uap->wrusage, sizeof(wru)); if (uap->info != NULL && error == 0) error = copyout(&si, uap->info, sizeof(si)); return (error); } /* * Reap the remains of a zombie process and optionally return status and * rusage. Asserts and will release both the proctree_lock and the process * lock as part of its work. */ void proc_reap(struct thread *td, struct proc *p, int *status, int options) { struct proc *q, *t; sx_assert(&proctree_lock, SA_XLOCKED); PROC_LOCK_ASSERT(p, MA_OWNED); PROC_SLOCK_ASSERT(p, MA_OWNED); KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); q = td->td_proc; PROC_SUNLOCK(p); if (status) *status = KW_EXITCODE(p->p_xexit, p->p_xsig); if (options & WNOWAIT) { /* * Only poll, returning the status. Caller does not wish to * release the proc struct just yet. */ PROC_UNLOCK(p); sx_xunlock(&proctree_lock); return; } PROC_LOCK(q); sigqueue_take(p->p_ksi); PROC_UNLOCK(q); /* * If we got the child via a ptrace 'attach', we need to give it back * to the old parent. */ if (p->p_oppid != 0 && p->p_oppid != p->p_pptr->p_pid) { PROC_UNLOCK(p); t = proc_realparent(p); PROC_LOCK(t); PROC_LOCK(p); CTR2(KTR_PTRACE, "wait: traced child %d moved back to parent %d", p->p_pid, t->p_pid); proc_reparent(p, t); p->p_oppid = 0; PROC_UNLOCK(p); pksignal(t, SIGCHLD, p->p_ksi); wakeup(t); cv_broadcast(&p->p_pwait); PROC_UNLOCK(t); sx_xunlock(&proctree_lock); return; } p->p_oppid = 0; PROC_UNLOCK(p); /* * Remove other references to this process to ensure we have an * exclusive reference. */ sx_xlock(&allproc_lock); LIST_REMOVE(p, p_list); /* off zombproc */ sx_xunlock(&allproc_lock); LIST_REMOVE(p, p_sibling); reaper_abandon_children(p, true); LIST_REMOVE(p, p_reapsibling); PROC_LOCK(p); clear_orphan(p); PROC_UNLOCK(p); leavepgrp(p); if (p->p_procdesc != NULL) procdesc_reap(p); sx_xunlock(&proctree_lock); PROC_LOCK(p); knlist_detach(p->p_klist); p->p_klist = NULL; PROC_UNLOCK(p); /* * Removal from allproc list and process group list paired with * PROC_LOCK which was executed during that time should guarantee * nothing can reach this process anymore. As such further locking * is unnecessary. */ p->p_xexit = p->p_xsig = 0; /* XXX: why? */ PROC_LOCK(q); ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); PROC_UNLOCK(q); /* * Decrement the count of procs running with this uid. */ (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); /* * Destroy resource accounting information associated with the process. */ #ifdef RACCT if (racct_enable) { PROC_LOCK(p); racct_sub(p, RACCT_NPROC, 1); PROC_UNLOCK(p); } #endif racct_proc_exit(p); /* * Free credentials, arguments, and sigacts. */ crfree(p->p_ucred); proc_set_cred(p, NULL); pargs_drop(p->p_args); p->p_args = NULL; sigacts_free(p->p_sigacts); p->p_sigacts = NULL; /* * Do any thread-system specific cleanups. */ thread_wait(p); /* * Give vm and machine-dependent layer a chance to free anything that * cpu_exit couldn't release while still running in process context. */ vm_waitproc(p); #ifdef MAC mac_proc_destroy(p); #endif /* * Free any domain policy that's still hiding around. */ vm_domain_policy_cleanup(&p->p_vm_dom_policy); KASSERT(FIRST_THREAD_IN_PROC(p), ("proc_reap: no residual thread!")); uma_zfree(proc_zone, p); atomic_add_int(&nprocs, -1); } static int proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo, int check_only) { struct rusage *rup; sx_assert(&proctree_lock, SA_XLOCKED); PROC_LOCK(p); switch (idtype) { case P_ALL: if (p->p_procdesc != NULL) { PROC_UNLOCK(p); return (0); } break; case P_PID: if (p->p_pid != (pid_t)id) { PROC_UNLOCK(p); return (0); } break; case P_PGID: if (p->p_pgid != (pid_t)id) { PROC_UNLOCK(p); return (0); } break; case P_SID: if (p->p_session->s_sid != (pid_t)id) { PROC_UNLOCK(p); return (0); } break; case P_UID: if (p->p_ucred->cr_uid != (uid_t)id) { PROC_UNLOCK(p); return (0); } break; case P_GID: if (p->p_ucred->cr_gid != (gid_t)id) { PROC_UNLOCK(p); return (0); } break; case P_JAILID: if (p->p_ucred->cr_prison->pr_id != (int)id) { PROC_UNLOCK(p); return (0); } break; /* * It seems that the thread structures get zeroed out * at process exit. This makes it impossible to * support P_SETID, P_CID or P_CPUID. */ default: PROC_UNLOCK(p); return (0); } if (p_canwait(td, p)) { PROC_UNLOCK(p); return (0); } if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { PROC_UNLOCK(p); return (0); } /* * This special case handles a kthread spawned by linux_clone * (see linux_misc.c). The linux_wait4 and linux_waitpid * functions need to be able to distinguish between waiting * on a process and waiting on a thread. It is a thread if * p_sigparent is not SIGCHLD, and the WLINUXCLONE option * signifies we want to wait for threads and not processes. */ if ((p->p_sigparent != SIGCHLD) ^ ((options & WLINUXCLONE) != 0)) { PROC_UNLOCK(p); return (0); } if (siginfo != NULL) { bzero(siginfo, sizeof(*siginfo)); siginfo->si_errno = 0; /* * SUSv4 requires that the si_signo value is always * SIGCHLD. Obey it despite the rfork(2) interface * allows to request other signal for child exit * notification. */ siginfo->si_signo = SIGCHLD; /* * This is still a rough estimate. We will fix the * cases TRAPPED, STOPPED, and CONTINUED later. */ if (WCOREDUMP(p->p_xsig)) { siginfo->si_code = CLD_DUMPED; siginfo->si_status = WTERMSIG(p->p_xsig); } else if (WIFSIGNALED(p->p_xsig)) { siginfo->si_code = CLD_KILLED; siginfo->si_status = WTERMSIG(p->p_xsig); } else { siginfo->si_code = CLD_EXITED; siginfo->si_status = p->p_xexit; } siginfo->si_pid = p->p_pid; siginfo->si_uid = p->p_ucred->cr_uid; /* * The si_addr field would be useful additional * detail, but apparently the PC value may be lost * when we reach this point. bzero() above sets * siginfo->si_addr to NULL. */ } /* * There should be no reason to limit resources usage info to * exited processes only. A snapshot about any resources used * by a stopped process may be exactly what is needed. */ if (wrusage != NULL) { rup = &wrusage->wru_self; *rup = p->p_ru; PROC_STATLOCK(p); calcru(p, &rup->ru_utime, &rup->ru_stime); PROC_STATUNLOCK(p); rup = &wrusage->wru_children; *rup = p->p_stats->p_cru; calccru(p, &rup->ru_utime, &rup->ru_stime); } if (p->p_state == PRS_ZOMBIE && !check_only) { PROC_SLOCK(p); proc_reap(td, p, status, options); return (-1); } PROC_UNLOCK(p); return (1); } int kern_wait(struct thread *td, pid_t pid, int *status, int options, struct rusage *rusage) { struct __wrusage wru, *wrup; idtype_t idtype; id_t id; int ret; /* * Translate the special pid values into the (idtype, pid) * pair for kern_wait6. The WAIT_MYPGRP case is handled by * kern_wait6() on its own. */ if (pid == WAIT_ANY) { idtype = P_ALL; id = 0; } else if (pid < 0) { idtype = P_PGID; id = (id_t)-pid; } else { idtype = P_PID; id = (id_t)pid; } if (rusage != NULL) wrup = &wru; else wrup = NULL; /* * For backward compatibility we implicitly add flags WEXITED * and WTRAPPED here. */ options |= WEXITED | WTRAPPED; ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); if (rusage != NULL) *rusage = wru.wru_self; return (ret); } int kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo) { struct proc *p, *q; pid_t pid; int error, nfound, ret; AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ AUDIT_ARG_VALUE(options); q = td->td_proc; if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { PROC_LOCK(q); id = (id_t)q->p_pgid; PROC_UNLOCK(q); idtype = P_PGID; } /* If we don't know the option, just return. */ if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | WEXITED | WTRAPPED | WLINUXCLONE)) != 0) return (EINVAL); if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { /* * We will be unable to find any matching processes, * because there are no known events to look for. * Prefer to return error instead of blocking * indefinitely. */ return (EINVAL); } loop: if (q->p_flag & P_STATCHILD) { PROC_LOCK(q); q->p_flag &= ~P_STATCHILD; PROC_UNLOCK(q); } nfound = 0; sx_xlock(&proctree_lock); LIST_FOREACH(p, &q->p_children, p_sibling) { pid = p->p_pid; ret = proc_to_reap(td, p, idtype, id, status, options, wrusage, siginfo, 0); if (ret == 0) continue; else if (ret == 1) nfound++; else { td->td_retval[0] = pid; return (0); } PROC_LOCK(p); PROC_SLOCK(p); if ((options & WTRAPPED) != 0 && (p->p_flag & P_TRACED) != 0 && (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 && (p->p_suspcount == p->p_numthreads) && ((p->p_flag & P_WAITED) == 0)) { PROC_SUNLOCK(p); if ((options & WNOWAIT) == 0) p->p_flag |= P_WAITED; sx_xunlock(&proctree_lock); if (status != NULL) *status = W_STOPCODE(p->p_xsig); if (siginfo != NULL) { siginfo->si_status = p->p_xsig; siginfo->si_code = CLD_TRAPPED; } if ((options & WNOWAIT) == 0) { PROC_LOCK(q); sigqueue_take(p->p_ksi); PROC_UNLOCK(q); } CTR4(KTR_PTRACE, "wait: returning trapped pid %d status %#x (xstat %d) xthread %d", p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig, p->p_xthread != NULL ? p->p_xthread->td_tid : -1); PROC_UNLOCK(p); td->td_retval[0] = pid; return (0); } if ((options & WUNTRACED) != 0 && (p->p_flag & P_STOPPED_SIG) != 0 && (p->p_suspcount == p->p_numthreads) && ((p->p_flag & P_WAITED) == 0)) { PROC_SUNLOCK(p); if ((options & WNOWAIT) == 0) p->p_flag |= P_WAITED; sx_xunlock(&proctree_lock); if (status != NULL) *status = W_STOPCODE(p->p_xsig); if (siginfo != NULL) { siginfo->si_status = p->p_xsig; siginfo->si_code = CLD_STOPPED; } if ((options & WNOWAIT) == 0) { PROC_LOCK(q); sigqueue_take(p->p_ksi); PROC_UNLOCK(q); } PROC_UNLOCK(p); td->td_retval[0] = pid; return (0); } PROC_SUNLOCK(p); if ((options & WCONTINUED) != 0 && (p->p_flag & P_CONTINUED) != 0) { sx_xunlock(&proctree_lock); if ((options & WNOWAIT) == 0) { p->p_flag &= ~P_CONTINUED; PROC_LOCK(q); sigqueue_take(p->p_ksi); PROC_UNLOCK(q); } PROC_UNLOCK(p); if (status != NULL) *status = SIGCONT; if (siginfo != NULL) { siginfo->si_status = SIGCONT; siginfo->si_code = CLD_CONTINUED; } td->td_retval[0] = pid; return (0); } PROC_UNLOCK(p); } /* * Look in the orphans list too, to allow the parent to * collect it's child exit status even if child is being * debugged. * * Debugger detaches from the parent upon successful * switch-over from parent to child. At this point due to * re-parenting the parent loses the child to debugger and a * wait4(2) call would report that it has no children to wait * for. By maintaining a list of orphans we allow the parent * to successfully wait until the child becomes a zombie. */ if (nfound == 0) { LIST_FOREACH(p, &q->p_orphans, p_orphan) { ret = proc_to_reap(td, p, idtype, id, NULL, options, NULL, NULL, 1); if (ret != 0) { KASSERT(ret != -1, ("reaped an orphan (pid %d)", (int)td->td_retval[0])); nfound++; break; } } } if (nfound == 0) { sx_xunlock(&proctree_lock); return (ECHILD); } if (options & WNOHANG) { sx_xunlock(&proctree_lock); td->td_retval[0] = 0; return (0); } PROC_LOCK(q); sx_xunlock(&proctree_lock); if (q->p_flag & P_STATCHILD) { q->p_flag &= ~P_STATCHILD; error = 0; } else error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0); PROC_UNLOCK(q); if (error) return (error); goto loop; } /* * Make process 'parent' the new parent of process 'child'. * Must be called with an exclusive hold of proctree lock. */ void proc_reparent(struct proc *child, struct proc *parent) { sx_assert(&proctree_lock, SX_XLOCKED); PROC_LOCK_ASSERT(child, MA_OWNED); if (child->p_pptr == parent) return; PROC_LOCK(child->p_pptr); sigqueue_take(child->p_ksi); PROC_UNLOCK(child->p_pptr); LIST_REMOVE(child, p_sibling); LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); clear_orphan(child); if (child->p_flag & P_TRACED) { if (LIST_EMPTY(&child->p_pptr->p_orphans)) { child->p_treeflag |= P_TREE_FIRST_ORPHAN; LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, p_orphan); } else { LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans), child, p_orphan); } child->p_treeflag |= P_TREE_ORPHANED; } child->p_pptr = parent; } Index: head/sys/kern/kern_fork.c =================================================================== --- head/sys/kern/kern_fork.c (revision 305831) +++ head/sys/kern/kern_fork.c (revision 305832) @@ -1,1116 +1,1116 @@ /*- * Copyright (c) 1982, 1986, 1989, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_ktrace.h" #include "opt_kstack_pages.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KDTRACE_HOOKS #include dtrace_fork_func_t dtrace_fasttrap_fork; #endif SDT_PROVIDER_DECLARE(proc); SDT_PROBE_DEFINE3(proc, , , create, "struct proc *", "struct proc *", "int"); #ifndef _SYS_SYSPROTO_H_ struct fork_args { int dummy; }; #endif /* ARGSUSED */ int sys_fork(struct thread *td, struct fork_args *uap) { struct fork_req fr; int error, pid; bzero(&fr, sizeof(fr)); fr.fr_flags = RFFDG | RFPROC; fr.fr_pidp = &pid; error = fork1(td, &fr); if (error == 0) { td->td_retval[0] = pid; td->td_retval[1] = 0; } return (error); } /* ARGUSED */ int sys_pdfork(struct thread *td, struct pdfork_args *uap) { struct fork_req fr; int error, fd, pid; bzero(&fr, sizeof(fr)); fr.fr_flags = RFFDG | RFPROC | RFPROCDESC; fr.fr_pidp = &pid; fr.fr_pd_fd = &fd; fr.fr_pd_flags = uap->flags; /* * It is necessary to return fd by reference because 0 is a valid file * descriptor number, and the child needs to be able to distinguish * itself from the parent using the return value. */ error = fork1(td, &fr); if (error == 0) { td->td_retval[0] = pid; td->td_retval[1] = 0; error = copyout(&fd, uap->fdp, sizeof(fd)); } return (error); } /* ARGSUSED */ int sys_vfork(struct thread *td, struct vfork_args *uap) { struct fork_req fr; int error, pid; bzero(&fr, sizeof(fr)); fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM; fr.fr_pidp = &pid; error = fork1(td, &fr); if (error == 0) { td->td_retval[0] = pid; td->td_retval[1] = 0; } return (error); } int sys_rfork(struct thread *td, struct rfork_args *uap) { struct fork_req fr; int error, pid; /* Don't allow kernel-only flags. */ if ((uap->flags & RFKERNELONLY) != 0) return (EINVAL); AUDIT_ARG_FFLAGS(uap->flags); bzero(&fr, sizeof(fr)); fr.fr_flags = uap->flags; fr.fr_pidp = &pid; error = fork1(td, &fr); if (error == 0) { td->td_retval[0] = pid; td->td_retval[1] = 0; } return (error); } int nprocs = 1; /* process 0 */ int lastpid = 0; SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, "Last used PID"); /* * Random component to lastpid generation. We mix in a random factor to make * it a little harder to predict. We sanity check the modulus value to avoid * doing it in critical paths. Don't let it be too small or we pointlessly * waste randomness entropy, and don't let it be impossibly large. Using a * modulus that is too big causes a LOT more process table scans and slows * down fork processing as the pidchecked caching is defeated. */ static int randompid = 0; static int sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) { int error, pid; error = sysctl_wire_old_buffer(req, sizeof(int)); if (error != 0) return(error); sx_xlock(&allproc_lock); pid = randompid; error = sysctl_handle_int(oidp, &pid, 0, req); if (error == 0 && req->newptr != NULL) { if (pid < 0 || pid > pid_max - 100) /* out of range */ pid = pid_max - 100; else if (pid < 2) /* NOP */ pid = 0; else if (pid < 100) /* Make it reasonable */ pid = 100; randompid = pid; } sx_xunlock(&allproc_lock); return (error); } SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); static int fork_findpid(int flags) { struct proc *p; int trypid; static int pidchecked = 0; /* * Requires allproc_lock in order to iterate over the list * of processes, and proctree_lock to access p_pgrp. */ sx_assert(&allproc_lock, SX_LOCKED); sx_assert(&proctree_lock, SX_LOCKED); /* * Find an unused process ID. We remember a range of unused IDs * ready to use (from lastpid+1 through pidchecked-1). * * If RFHIGHPID is set (used during system boot), do not allocate * low-numbered pids. */ trypid = lastpid + 1; if (flags & RFHIGHPID) { if (trypid < 10) trypid = 10; } else { if (randompid) trypid += arc4random() % randompid; } retry: /* * If the process ID prototype has wrapped around, * restart somewhat above 0, as the low-numbered procs * tend to include daemons that don't exit. */ if (trypid >= pid_max) { trypid = trypid % pid_max; if (trypid < 100) trypid += 100; pidchecked = 0; } if (trypid >= pidchecked) { int doingzomb = 0; pidchecked = PID_MAX; /* * Scan the active and zombie procs to check whether this pid * is in use. Remember the lowest pid that's greater * than trypid, so we can avoid checking for a while. * * Avoid reuse of the process group id, session id or * the reaper subtree id. Note that for process group * and sessions, the amount of reserved pids is * limited by process limit. For the subtree ids, the * id is kept reserved only while there is a * non-reaped process in the subtree, so amount of * reserved pids is limited by process limit times * two. */ p = LIST_FIRST(&allproc); again: for (; p != NULL; p = LIST_NEXT(p, p_list)) { while (p->p_pid == trypid || p->p_reapsubtree == trypid || (p->p_pgrp != NULL && (p->p_pgrp->pg_id == trypid || (p->p_session != NULL && p->p_session->s_sid == trypid)))) { trypid++; if (trypid >= pidchecked) goto retry; } if (p->p_pid > trypid && pidchecked > p->p_pid) pidchecked = p->p_pid; if (p->p_pgrp != NULL) { if (p->p_pgrp->pg_id > trypid && pidchecked > p->p_pgrp->pg_id) pidchecked = p->p_pgrp->pg_id; if (p->p_session != NULL && p->p_session->s_sid > trypid && pidchecked > p->p_session->s_sid) pidchecked = p->p_session->s_sid; } } if (!doingzomb) { doingzomb = 1; p = LIST_FIRST(&zombproc); goto again; } } /* * RFHIGHPID does not mess with the lastpid counter during boot. */ if (flags & RFHIGHPID) pidchecked = 0; else lastpid = trypid; return (trypid); } static int fork_norfproc(struct thread *td, int flags) { int error; struct proc *p1; KASSERT((flags & RFPROC) == 0, ("fork_norfproc called with RFPROC set")); p1 = td->td_proc; if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && (flags & (RFCFDG | RFFDG))) { PROC_LOCK(p1); if (thread_single(p1, SINGLE_BOUNDARY)) { PROC_UNLOCK(p1); return (ERESTART); } PROC_UNLOCK(p1); } error = vm_forkproc(td, NULL, NULL, NULL, flags); if (error) goto fail; /* * Close all file descriptors. */ if (flags & RFCFDG) { struct filedesc *fdtmp; fdtmp = fdinit(td->td_proc->p_fd, false); fdescfree(td); p1->p_fd = fdtmp; } /* * Unshare file descriptors (from parent). */ if (flags & RFFDG) fdunshare(td); fail: if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && (flags & (RFCFDG | RFFDG))) { PROC_LOCK(p1); thread_single_end(p1, SINGLE_BOUNDARY); PROC_UNLOCK(p1); } return (error); } static void do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2, struct vmspace *vm2, struct file *fp_procdesc) { struct proc *p1, *pptr; int trypid; struct filedesc *fd; struct filedesc_to_leader *fdtol; struct sigacts *newsigacts; sx_assert(&proctree_lock, SX_SLOCKED); sx_assert(&allproc_lock, SX_XLOCKED); p1 = td->td_proc; trypid = fork_findpid(fr->fr_flags); sx_sunlock(&proctree_lock); p2->p_state = PRS_NEW; /* protect against others */ p2->p_pid = trypid; AUDIT_ARG_PID(p2->p_pid); LIST_INSERT_HEAD(&allproc, p2, p_list); allproc_gen++; LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); tidhash_add(td2); PROC_LOCK(p2); PROC_LOCK(p1); sx_xunlock(&allproc_lock); bcopy(&p1->p_startcopy, &p2->p_startcopy, __rangeof(struct proc, p_startcopy, p_endcopy)); pargs_hold(p2->p_args); PROC_UNLOCK(p1); bzero(&p2->p_startzero, __rangeof(struct proc, p_startzero, p_endzero)); /* Tell the prison that we exist. */ prison_proc_hold(p2->p_ucred->cr_prison); PROC_UNLOCK(p2); /* * Malloc things while we don't hold any locks. */ if (fr->fr_flags & RFSIGSHARE) newsigacts = NULL; else newsigacts = sigacts_alloc(); /* * Copy filedesc. */ if (fr->fr_flags & RFCFDG) { fd = fdinit(p1->p_fd, false); fdtol = NULL; } else if (fr->fr_flags & RFFDG) { fd = fdcopy(p1->p_fd); fdtol = NULL; } else { fd = fdshare(p1->p_fd); if (p1->p_fdtol == NULL) p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL, p1->p_leader); if ((fr->fr_flags & RFTHREAD) != 0) { /* * Shared file descriptor table, and shared * process leaders. */ fdtol = p1->p_fdtol; FILEDESC_XLOCK(p1->p_fd); fdtol->fdl_refcount++; FILEDESC_XUNLOCK(p1->p_fd); } else { /* * Shared file descriptor table, and different * process leaders. */ fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p1->p_fd, p2); } } /* * Make a proc table entry for the new process. * Start by zeroing the section of proc that is zero-initialized, * then copy the section that is copied directly from the parent. */ PROC_LOCK(p2); PROC_LOCK(p1); bzero(&td2->td_startzero, __rangeof(struct thread, td_startzero, td_endzero)); bcopy(&td->td_startcopy, &td2->td_startcopy, __rangeof(struct thread, td_startcopy, td_endcopy)); bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name)); td2->td_sigstk = td->td_sigstk; td2->td_flags = TDF_INMEM; td2->td_lend_user_pri = PRI_MAX; #ifdef VIMAGE td2->td_vnet = NULL; td2->td_vnet_lpush = NULL; #endif /* * Allow the scheduler to initialize the child. */ thread_lock(td); sched_fork(td, td2); thread_unlock(td); /* * Duplicate sub-structures as needed. * Increase reference counts on shared objects. */ p2->p_flag = P_INMEM; p2->p_flag2 = p1->p_flag2 & (P2_NOTRACE | P2_NOTRACE_EXEC); p2->p_swtick = ticks; if (p1->p_flag & P_PROFIL) startprofclock(p2); /* * Whilst the proc lock is held, copy the VM domain data out * using the VM domain method. */ vm_domain_policy_init(&p2->p_vm_dom_policy); vm_domain_policy_localcopy(&p2->p_vm_dom_policy, &p1->p_vm_dom_policy); if (fr->fr_flags & RFSIGSHARE) { p2->p_sigacts = sigacts_hold(p1->p_sigacts); } else { sigacts_copy(newsigacts, p1->p_sigacts); p2->p_sigacts = newsigacts; } if (fr->fr_flags & RFTSIGZMB) p2->p_sigparent = RFTSIGNUM(fr->fr_flags); else if (fr->fr_flags & RFLINUXTHPN) p2->p_sigparent = SIGUSR1; else p2->p_sigparent = SIGCHLD; p2->p_textvp = p1->p_textvp; p2->p_fd = fd; p2->p_fdtol = fdtol; if (p1->p_flag2 & P2_INHERIT_PROTECTED) { p2->p_flag |= P_PROTECTED; p2->p_flag2 |= P2_INHERIT_PROTECTED; } /* * p_limit is copy-on-write. Bump its refcount. */ lim_fork(p1, p2); thread_cow_get_proc(td2, p2); pstats_fork(p1->p_stats, p2->p_stats); PROC_UNLOCK(p1); PROC_UNLOCK(p2); /* Bump references to the text vnode (for procfs). */ if (p2->p_textvp) vref(p2->p_textvp); /* * Set up linkage for kernel based threading. */ if ((fr->fr_flags & RFTHREAD) != 0) { mtx_lock(&ppeers_lock); p2->p_peers = p1->p_peers; p1->p_peers = p2; p2->p_leader = p1->p_leader; mtx_unlock(&ppeers_lock); PROC_LOCK(p1->p_leader); if ((p1->p_leader->p_flag & P_WEXIT) != 0) { PROC_UNLOCK(p1->p_leader); /* * The task leader is exiting, so process p1 is * going to be killed shortly. Since p1 obviously * isn't dead yet, we know that the leader is either * sending SIGKILL's to all the processes in this * task or is sleeping waiting for all the peers to * exit. We let p1 complete the fork, but we need * to go ahead and kill the new process p2 since * the task leader may not get a chance to send * SIGKILL to it. We leave it on the list so that * the task leader will wait for this new process * to commit suicide. */ PROC_LOCK(p2); kern_psignal(p2, SIGKILL); PROC_UNLOCK(p2); } else PROC_UNLOCK(p1->p_leader); } else { p2->p_peers = NULL; p2->p_leader = p2; } sx_xlock(&proctree_lock); PGRP_LOCK(p1->p_pgrp); PROC_LOCK(p2); PROC_LOCK(p1); /* * Preserve some more flags in subprocess. P_PROFIL has already * been preserved. */ p2->p_flag |= p1->p_flag & P_SUGID; td2->td_pflags |= (td->td_pflags & TDP_ALTSTACK) | TDP_FORKING; SESS_LOCK(p1->p_session); if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) p2->p_flag |= P_CONTROLT; SESS_UNLOCK(p1->p_session); if (fr->fr_flags & RFPPWAIT) p2->p_flag |= P_PPWAIT; p2->p_pgrp = p1->p_pgrp; LIST_INSERT_AFTER(p1, p2, p_pglist); PGRP_UNLOCK(p1->p_pgrp); LIST_INIT(&p2->p_children); LIST_INIT(&p2->p_orphans); callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0); /* * If PF_FORK is set, the child process inherits the * procfs ioctl flags from its parent. */ if (p1->p_pfsflags & PF_FORK) { p2->p_stops = p1->p_stops; p2->p_pfsflags = p1->p_pfsflags; } /* * This begins the section where we must prevent the parent * from being swapped. */ _PHOLD(p1); PROC_UNLOCK(p1); /* * Attach the new process to its parent. * * If RFNOWAIT is set, the newly created process becomes a child * of init. This effectively disassociates the child from the * parent. */ if ((fr->fr_flags & RFNOWAIT) != 0) { pptr = p1->p_reaper; p2->p_reaper = pptr; } else { p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ? p1 : p1->p_reaper; pptr = p1; } p2->p_pptr = pptr; LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); LIST_INIT(&p2->p_reaplist); LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling); if (p2->p_reaper == p1) p2->p_reapsubtree = p2->p_pid; sx_xunlock(&proctree_lock); /* Inform accounting that we have forked. */ p2->p_acflag = AFORK; PROC_UNLOCK(p2); #ifdef KTRACE ktrprocfork(p1, p2); #endif /* * Finish creating the child process. It will return via a different * execution path later. (ie: directly into user mode) */ vm_forkproc(td, p2, td2, vm2, fr->fr_flags); if (fr->fr_flags == (RFFDG | RFPROC)) { PCPU_INC(cnt.v_forks); PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize); } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { PCPU_INC(cnt.v_vforks); PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize); } else if (p1 == &proc0) { PCPU_INC(cnt.v_kthreads); PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize); } else { PCPU_INC(cnt.v_rforks); PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize); } /* * Associate the process descriptor with the process before anything * can happen that might cause that process to need the descriptor. * However, don't do this until after fork(2) can no longer fail. */ if (fr->fr_flags & RFPROCDESC) procdesc_new(p2, fr->fr_pd_flags); /* * Both processes are set up, now check if any loadable modules want * to adjust anything. */ EVENTHANDLER_INVOKE(process_fork, p1, p2, fr->fr_flags); /* * Set the child start time and mark the process as being complete. */ PROC_LOCK(p2); PROC_LOCK(p1); microuptime(&p2->p_stats->p_start); PROC_SLOCK(p2); p2->p_state = PRS_NORMAL; PROC_SUNLOCK(p2); #ifdef KDTRACE_HOOKS /* * Tell the DTrace fasttrap provider about the new process so that any * tracepoints inherited from the parent can be removed. We have to do * this only after p_state is PRS_NORMAL since the fasttrap module will * use pfind() later on. */ if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork) dtrace_fasttrap_fork(p1, p2); #endif /* * Hold the process so that it cannot exit after we make it runnable, * but before we wait for the debugger. */ _PHOLD(p2); if (p1->p_ptevents & PTRACE_FORK) { /* * Arrange for debugger to receive the fork event. * * We can report PL_FLAG_FORKED regardless of * P_FOLLOWFORK settings, but it does not make a sense * for runaway child. */ td->td_dbgflags |= TDB_FORK; td->td_dbg_forked = p2->p_pid; td2->td_dbgflags |= TDB_STOPATFORK; } if (fr->fr_flags & RFPPWAIT) { td->td_pflags |= TDP_RFPPWAIT; td->td_rfppwait_p = p2; td->td_dbgflags |= TDB_VFORK; } PROC_UNLOCK(p2); /* * Now can be swapped. */ _PRELE(p1); PROC_UNLOCK(p1); /* * Tell any interested parties about the new process. */ knote_fork(p1->p_klist, p2->p_pid); SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags); if (fr->fr_flags & RFPROCDESC) { procdesc_finit(p2->p_procdesc, fp_procdesc); fdrop(fp_procdesc, td); } if ((fr->fr_flags & RFSTOPPED) == 0) { /* * If RFSTOPPED not requested, make child runnable and * add to run queue. */ thread_lock(td2); TD_SET_CAN_RUN(td2); sched_add(td2, SRQ_BORING); thread_unlock(td2); if (fr->fr_pidp != NULL) *fr->fr_pidp = p2->p_pid; } else { *fr->fr_procp = p2; } PROC_LOCK(p2); /* * Wait until debugger is attached to child. */ while (td2->td_proc == p2 && (td2->td_dbgflags & TDB_STOPATFORK) != 0) cv_wait(&p2->p_dbgwait, &p2->p_mtx); _PRELE(p2); racct_proc_fork_done(p2); PROC_UNLOCK(p2); } int fork1(struct thread *td, struct fork_req *fr) { struct proc *p1, *newproc; struct thread *td2; struct vmspace *vm2; struct file *fp_procdesc; vm_ooffset_t mem_charged; int error, nprocs_new, ok; static int curfail; static struct timeval lastfail; int flags, pages; flags = fr->fr_flags; pages = fr->fr_pages; if ((flags & RFSTOPPED) != 0) MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL); else MPASS(fr->fr_procp == NULL); /* Check for the undefined or unimplemented flags. */ if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0) return (EINVAL); /* Signal value requires RFTSIGZMB. */ if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0) return (EINVAL); /* Can't copy and clear. */ if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) return (EINVAL); /* Check the validity of the signal number. */ if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG) return (EINVAL); if ((flags & RFPROCDESC) != 0) { /* Can't not create a process yet get a process descriptor. */ if ((flags & RFPROC) == 0) return (EINVAL); /* Must provide a place to put a procdesc if creating one. */ if (fr->fr_pd_fd == NULL) return (EINVAL); /* Check if we are using supported flags. */ if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0) return (EINVAL); } p1 = td->td_proc; /* * Here we don't create a new process, but we divorce * certain parts of a process from itself. */ if ((flags & RFPROC) == 0) { if (fr->fr_procp != NULL) *fr->fr_procp = NULL; else if (fr->fr_pidp != NULL) *fr->fr_pidp = 0; return (fork_norfproc(td, flags)); } fp_procdesc = NULL; newproc = NULL; vm2 = NULL; /* * Increment the nprocs resource before allocations occur. * Although process entries are dynamically created, we still * keep a global limit on the maximum number we will * create. There are hard-limits as to the number of processes * that can run, established by the KVA and memory usage for * the process data. * * Don't allow a nonprivileged user to use the last ten * processes; don't let root exceed the limit. */ nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1; if ((nprocs_new >= maxproc - 10 && priv_check_cred(td->td_ucred, PRIV_MAXPROC, 0) != 0) || nprocs_new >= maxproc) { error = EAGAIN; sx_xlock(&allproc_lock); if (ppsratecheck(&lastfail, &curfail, 1)) { printf("maxproc limit exceeded by uid %u (pid %d); " "see tuning(7) and login.conf(5)\n", td->td_ucred->cr_ruid, p1->p_pid); } sx_xunlock(&allproc_lock); goto fail2; } /* * If required, create a process descriptor in the parent first; we * will abandon it if something goes wrong. We don't finit() until * later. */ if (flags & RFPROCDESC) { error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd, fr->fr_pd_flags, fr->fr_pd_fcaps); if (error != 0) goto fail2; } mem_charged = 0; if (pages == 0) pages = kstack_pages; /* Allocate new proc. */ newproc = uma_zalloc(proc_zone, M_WAITOK); td2 = FIRST_THREAD_IN_PROC(newproc); if (td2 == NULL) { td2 = thread_alloc(pages); if (td2 == NULL) { error = ENOMEM; goto fail2; } proc_linkup(newproc, td2); } else { if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) { if (td2->td_kstack != 0) vm_thread_dispose(td2); if (!thread_alloc_stack(td2, pages)) { error = ENOMEM; goto fail2; } } } if ((flags & RFMEM) == 0) { vm2 = vmspace_fork(p1->p_vmspace, &mem_charged); if (vm2 == NULL) { error = ENOMEM; goto fail2; } if (!swap_reserve(mem_charged)) { /* * The swap reservation failed. The accounting * from the entries of the copied vm2 will be * subtracted in vmspace_free(), so force the * reservation there. */ swap_reserve_force(mem_charged); error = ENOMEM; goto fail2; } } else vm2 = NULL; /* * XXX: This is ugly; when we copy resource usage, we need to bump * per-cred resource counters. */ proc_set_cred_init(newproc, crhold(td->td_ucred)); /* * Initialize resource accounting for the child process. */ error = racct_proc_fork(p1, newproc); if (error != 0) { error = EAGAIN; goto fail1; } #ifdef MAC mac_proc_init(newproc); #endif newproc->p_klist = knlist_alloc(&newproc->p_mtx); STAILQ_INIT(&newproc->p_ktr); /* We have to lock the process tree while we look for a pid. */ sx_slock(&proctree_lock); sx_xlock(&allproc_lock); /* * Increment the count of procs running with this uid. Don't allow * a nonprivileged user to exceed their current limit. * * XXXRW: Can we avoid privilege here if it's not needed? */ error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0); if (error == 0) ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0); else { ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_NPROC)); } if (ok) { do_fork(td, fr, newproc, td2, vm2, fp_procdesc); return (0); } error = EAGAIN; sx_sunlock(&proctree_lock); sx_xunlock(&allproc_lock); #ifdef MAC mac_proc_destroy(newproc); #endif racct_proc_exit(newproc); fail1: crfree(newproc->p_ucred); newproc->p_ucred = NULL; fail2: if (vm2 != NULL) vmspace_free(vm2); uma_zfree(proc_zone, newproc); if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) { fdclose(td, fp_procdesc, *fr->fr_pd_fd); fdrop(fp_procdesc, td); } atomic_add_int(&nprocs, -1); pause("fork", hz / 2); return (error); } /* * Handle the return of a child process from fork1(). This function * is called from the MD fork_trampoline() entry point. */ void fork_exit(void (*callout)(void *, struct trapframe *), void *arg, struct trapframe *frame) { struct proc *p; struct thread *td; struct thread *dtd; td = curthread; p = td->td_proc; KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)", td, td_get_sched(td), p->p_pid, td->td_name); sched_fork_exit(td); /* * Processes normally resume in mi_switch() after being * cpu_switch()'ed to, but when children start up they arrive here * instead, so we must do much the same things as mi_switch() would. */ if ((dtd = PCPU_GET(deadthread))) { PCPU_SET(deadthread, NULL); thread_stash(dtd); } thread_unlock(td); /* * cpu_fork_kthread_handler intercepts this function call to * have this call a non-return function to stay in kernel mode. * initproc has its own fork handler, but it does return. */ KASSERT(callout != NULL, ("NULL callout in fork_exit")); callout(arg, frame); /* * Check if a kernel thread misbehaved and returned from its main * function. */ if (p->p_flag & P_KPROC) { printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", td->td_name, p->p_pid); kthread_exit(); } mtx_assert(&Giant, MA_NOTOWNED); if (p->p_sysent->sv_schedtail != NULL) (p->p_sysent->sv_schedtail)(td); td->td_pflags &= ~TDP_FORKING; } /* * Simplified back end of syscall(), used when returning from fork() * directly into user mode. This function is passed in to fork_exit() * as the first parameter and is called when returning to a new * userland process. */ void fork_return(struct thread *td, struct trapframe *frame) { struct proc *p, *dbg; p = td->td_proc; if (td->td_dbgflags & TDB_STOPATFORK) { sx_xlock(&proctree_lock); PROC_LOCK(p); if (p->p_pptr->p_ptevents & PTRACE_FORK) { /* * If debugger still wants auto-attach for the * parent's children, do it now. */ dbg = p->p_pptr->p_pptr; proc_set_traced(p, true); CTR2(KTR_PTRACE, "fork_return: attaching to new child pid %d: oppid %d", p->p_pid, p->p_oppid); proc_reparent(p, dbg); sx_xunlock(&proctree_lock); td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP; ptracestop(td, SIGSTOP); td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX); } else { /* * ... otherwise clear the request. */ sx_xunlock(&proctree_lock); td->td_dbgflags &= ~TDB_STOPATFORK; cv_broadcast(&p->p_dbgwait); } PROC_UNLOCK(p); } else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) { /* * This is the start of a new thread in a traced * process. Report a system call exit event. */ PROC_LOCK(p); td->td_dbgflags |= TDB_SCX; _STOPEVENT(p, S_SCX, td->td_dbg_sc_code); if ((p->p_ptevents & PTRACE_SCX) != 0 || (td->td_dbgflags & TDB_BORN) != 0) ptracestop(td, SIGTRAP); td->td_dbgflags &= ~(TDB_SCX | TDB_BORN); PROC_UNLOCK(p); } userret(td, frame); #ifdef KTRACE if (KTRPOINT(td, KTR_SYSRET)) ktrsysret(SYS_fork, 0, 0); #endif } Index: head/sys/kern/kern_ktrace.c =================================================================== --- head/sys/kern/kern_ktrace.c (revision 305831) +++ head/sys/kern/kern_ktrace.c (revision 305832) @@ -1,1279 +1,1279 @@ /*- * Copyright (c) 1989, 1993 * The Regents of the University of California. * Copyright (c) 2005 Robert N. M. Watson * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_ktrace.c 8.2 (Berkeley) 9/23/93 */ #include __FBSDID("$FreeBSD$"); #include "opt_ktrace.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * The ktrace facility allows the tracing of certain key events in user space * processes, such as system calls, signal delivery, context switches, and * user generated events using utrace(2). It works by streaming event * records and data to a vnode associated with the process using the * ktrace(2) system call. In general, records can be written directly from * the context that generates the event. One important exception to this is * during a context switch, where sleeping is not permitted. To handle this * case, trace events are generated using in-kernel ktr_request records, and * then delivered to disk at a convenient moment -- either immediately, the * next traceable event, at system call return, or at process exit. * * When dealing with multiple threads or processes writing to the same event * log, ordering guarantees are weak: specifically, if an event has multiple * records (i.e., system call enter and return), they may be interlaced with * records from another event. Process and thread ID information is provided * in the record, and user applications can de-interlace events if required. */ static MALLOC_DEFINE(M_KTRACE, "KTRACE", "KTRACE"); #ifdef KTRACE FEATURE(ktrace, "Kernel support for system-call tracing"); #ifndef KTRACE_REQUEST_POOL #define KTRACE_REQUEST_POOL 100 #endif struct ktr_request { struct ktr_header ktr_header; void *ktr_buffer; union { struct ktr_proc_ctor ktr_proc_ctor; struct ktr_cap_fail ktr_cap_fail; struct ktr_syscall ktr_syscall; struct ktr_sysret ktr_sysret; struct ktr_genio ktr_genio; struct ktr_psig ktr_psig; struct ktr_csw ktr_csw; struct ktr_fault ktr_fault; struct ktr_faultend ktr_faultend; } ktr_data; STAILQ_ENTRY(ktr_request) ktr_list; }; static int data_lengths[] = { [KTR_SYSCALL] = offsetof(struct ktr_syscall, ktr_args), [KTR_SYSRET] = sizeof(struct ktr_sysret), [KTR_NAMEI] = 0, [KTR_GENIO] = sizeof(struct ktr_genio), [KTR_PSIG] = sizeof(struct ktr_psig), [KTR_CSW] = sizeof(struct ktr_csw), [KTR_USER] = 0, [KTR_STRUCT] = 0, [KTR_SYSCTL] = 0, [KTR_PROCCTOR] = sizeof(struct ktr_proc_ctor), [KTR_PROCDTOR] = 0, [KTR_CAPFAIL] = sizeof(struct ktr_cap_fail), [KTR_FAULT] = sizeof(struct ktr_fault), [KTR_FAULTEND] = sizeof(struct ktr_faultend), }; static STAILQ_HEAD(, ktr_request) ktr_free; static SYSCTL_NODE(_kern, OID_AUTO, ktrace, CTLFLAG_RD, 0, "KTRACE options"); static u_int ktr_requestpool = KTRACE_REQUEST_POOL; TUNABLE_INT("kern.ktrace.request_pool", &ktr_requestpool); static u_int ktr_geniosize = PAGE_SIZE; SYSCTL_UINT(_kern_ktrace, OID_AUTO, genio_size, CTLFLAG_RWTUN, &ktr_geniosize, 0, "Maximum size of genio event payload"); static int print_message = 1; static struct mtx ktrace_mtx; static struct sx ktrace_sx; static void ktrace_init(void *dummy); static int sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS); static u_int ktrace_resize_pool(u_int oldsize, u_int newsize); static struct ktr_request *ktr_getrequest_entered(struct thread *td, int type); static struct ktr_request *ktr_getrequest(int type); static void ktr_submitrequest(struct thread *td, struct ktr_request *req); static void ktr_freeproc(struct proc *p, struct ucred **uc, struct vnode **vp); static void ktr_freerequest(struct ktr_request *req); static void ktr_freerequest_locked(struct ktr_request *req); static void ktr_writerequest(struct thread *td, struct ktr_request *req); static int ktrcanset(struct thread *,struct proc *); static int ktrsetchildren(struct thread *,struct proc *,int,int,struct vnode *); static int ktrops(struct thread *,struct proc *,int,int,struct vnode *); static void ktrprocctor_entered(struct thread *, struct proc *); /* * ktrace itself generates events, such as context switches, which we do not * wish to trace. Maintain a flag, TDP_INKTRACE, on each thread to determine * whether or not it is in a region where tracing of events should be * suppressed. */ static void ktrace_enter(struct thread *td) { KASSERT(!(td->td_pflags & TDP_INKTRACE), ("ktrace_enter: flag set")); td->td_pflags |= TDP_INKTRACE; } static void ktrace_exit(struct thread *td) { KASSERT(td->td_pflags & TDP_INKTRACE, ("ktrace_exit: flag not set")); td->td_pflags &= ~TDP_INKTRACE; } static void ktrace_assert(struct thread *td) { KASSERT(td->td_pflags & TDP_INKTRACE, ("ktrace_assert: flag not set")); } static void ktrace_init(void *dummy) { struct ktr_request *req; int i; mtx_init(&ktrace_mtx, "ktrace", NULL, MTX_DEF | MTX_QUIET); sx_init(&ktrace_sx, "ktrace_sx"); STAILQ_INIT(&ktr_free); for (i = 0; i < ktr_requestpool; i++) { req = malloc(sizeof(struct ktr_request), M_KTRACE, M_WAITOK); STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list); } } SYSINIT(ktrace_init, SI_SUB_KTRACE, SI_ORDER_ANY, ktrace_init, NULL); static int sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS) { struct thread *td; u_int newsize, oldsize, wantsize; int error; /* Handle easy read-only case first to avoid warnings from GCC. */ if (!req->newptr) { oldsize = ktr_requestpool; return (SYSCTL_OUT(req, &oldsize, sizeof(u_int))); } error = SYSCTL_IN(req, &wantsize, sizeof(u_int)); if (error) return (error); td = curthread; ktrace_enter(td); oldsize = ktr_requestpool; newsize = ktrace_resize_pool(oldsize, wantsize); ktrace_exit(td); error = SYSCTL_OUT(req, &oldsize, sizeof(u_int)); if (error) return (error); if (wantsize > oldsize && newsize < wantsize) return (ENOSPC); return (0); } SYSCTL_PROC(_kern_ktrace, OID_AUTO, request_pool, CTLTYPE_UINT|CTLFLAG_RW, &ktr_requestpool, 0, sysctl_kern_ktrace_request_pool, "IU", "Pool buffer size for ktrace(1)"); static u_int ktrace_resize_pool(u_int oldsize, u_int newsize) { STAILQ_HEAD(, ktr_request) ktr_new; struct ktr_request *req; int bound; print_message = 1; bound = newsize - oldsize; if (bound == 0) return (ktr_requestpool); if (bound < 0) { mtx_lock(&ktrace_mtx); /* Shrink pool down to newsize if possible. */ while (bound++ < 0) { req = STAILQ_FIRST(&ktr_free); if (req == NULL) break; STAILQ_REMOVE_HEAD(&ktr_free, ktr_list); ktr_requestpool--; free(req, M_KTRACE); } } else { /* Grow pool up to newsize. */ STAILQ_INIT(&ktr_new); while (bound-- > 0) { req = malloc(sizeof(struct ktr_request), M_KTRACE, M_WAITOK); STAILQ_INSERT_HEAD(&ktr_new, req, ktr_list); } mtx_lock(&ktrace_mtx); STAILQ_CONCAT(&ktr_free, &ktr_new); ktr_requestpool += (newsize - oldsize); } mtx_unlock(&ktrace_mtx); return (ktr_requestpool); } /* ktr_getrequest() assumes that ktr_comm[] is the same size as td_name[]. */ CTASSERT(sizeof(((struct ktr_header *)NULL)->ktr_comm) == (sizeof((struct thread *)NULL)->td_name)); static struct ktr_request * ktr_getrequest_entered(struct thread *td, int type) { struct ktr_request *req; struct proc *p = td->td_proc; int pm; mtx_lock(&ktrace_mtx); if (!KTRCHECK(td, type)) { mtx_unlock(&ktrace_mtx); return (NULL); } req = STAILQ_FIRST(&ktr_free); if (req != NULL) { STAILQ_REMOVE_HEAD(&ktr_free, ktr_list); req->ktr_header.ktr_type = type; if (p->p_traceflag & KTRFAC_DROP) { req->ktr_header.ktr_type |= KTR_DROP; p->p_traceflag &= ~KTRFAC_DROP; } mtx_unlock(&ktrace_mtx); microtime(&req->ktr_header.ktr_time); req->ktr_header.ktr_pid = p->p_pid; req->ktr_header.ktr_tid = td->td_tid; bcopy(td->td_name, req->ktr_header.ktr_comm, sizeof(req->ktr_header.ktr_comm)); req->ktr_buffer = NULL; req->ktr_header.ktr_len = 0; } else { p->p_traceflag |= KTRFAC_DROP; pm = print_message; print_message = 0; mtx_unlock(&ktrace_mtx); if (pm) printf("Out of ktrace request objects.\n"); } return (req); } static struct ktr_request * ktr_getrequest(int type) { struct thread *td = curthread; struct ktr_request *req; ktrace_enter(td); req = ktr_getrequest_entered(td, type); if (req == NULL) ktrace_exit(td); return (req); } /* * Some trace generation environments don't permit direct access to VFS, * such as during a context switch where sleeping is not allowed. Under these * circumstances, queue a request to the thread to be written asynchronously * later. */ static void ktr_enqueuerequest(struct thread *td, struct ktr_request *req) { mtx_lock(&ktrace_mtx); STAILQ_INSERT_TAIL(&td->td_proc->p_ktr, req, ktr_list); mtx_unlock(&ktrace_mtx); } /* * Drain any pending ktrace records from the per-thread queue to disk. This * is used both internally before committing other records, and also on * system call return. We drain all the ones we can find at the time when * drain is requested, but don't keep draining after that as those events * may be approximately "after" the current event. */ static void ktr_drain(struct thread *td) { struct ktr_request *queued_req; STAILQ_HEAD(, ktr_request) local_queue; ktrace_assert(td); sx_assert(&ktrace_sx, SX_XLOCKED); STAILQ_INIT(&local_queue); if (!STAILQ_EMPTY(&td->td_proc->p_ktr)) { mtx_lock(&ktrace_mtx); STAILQ_CONCAT(&local_queue, &td->td_proc->p_ktr); mtx_unlock(&ktrace_mtx); while ((queued_req = STAILQ_FIRST(&local_queue))) { STAILQ_REMOVE_HEAD(&local_queue, ktr_list); ktr_writerequest(td, queued_req); ktr_freerequest(queued_req); } } } /* * Submit a trace record for immediate commit to disk -- to be used only * where entering VFS is OK. First drain any pending records that may have * been cached in the thread. */ static void ktr_submitrequest(struct thread *td, struct ktr_request *req) { ktrace_assert(td); sx_xlock(&ktrace_sx); ktr_drain(td); ktr_writerequest(td, req); ktr_freerequest(req); sx_xunlock(&ktrace_sx); ktrace_exit(td); } static void ktr_freerequest(struct ktr_request *req) { mtx_lock(&ktrace_mtx); ktr_freerequest_locked(req); mtx_unlock(&ktrace_mtx); } static void ktr_freerequest_locked(struct ktr_request *req) { mtx_assert(&ktrace_mtx, MA_OWNED); if (req->ktr_buffer != NULL) free(req->ktr_buffer, M_KTRACE); STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list); } /* * Disable tracing for a process and release all associated resources. * The caller is responsible for releasing a reference on the returned * vnode and credentials. */ static void ktr_freeproc(struct proc *p, struct ucred **uc, struct vnode **vp) { struct ktr_request *req; PROC_LOCK_ASSERT(p, MA_OWNED); mtx_assert(&ktrace_mtx, MA_OWNED); *uc = p->p_tracecred; p->p_tracecred = NULL; if (vp != NULL) *vp = p->p_tracevp; p->p_tracevp = NULL; p->p_traceflag = 0; while ((req = STAILQ_FIRST(&p->p_ktr)) != NULL) { STAILQ_REMOVE_HEAD(&p->p_ktr, ktr_list); ktr_freerequest_locked(req); } } void ktrsyscall(code, narg, args) int code, narg; register_t args[]; { struct ktr_request *req; struct ktr_syscall *ktp; size_t buflen; char *buf = NULL; buflen = sizeof(register_t) * narg; if (buflen > 0) { buf = malloc(buflen, M_KTRACE, M_WAITOK); bcopy(args, buf, buflen); } req = ktr_getrequest(KTR_SYSCALL); if (req == NULL) { if (buf != NULL) free(buf, M_KTRACE); return; } ktp = &req->ktr_data.ktr_syscall; ktp->ktr_code = code; ktp->ktr_narg = narg; if (buflen > 0) { req->ktr_header.ktr_len = buflen; req->ktr_buffer = buf; } ktr_submitrequest(curthread, req); } void ktrsysret(code, error, retval) int code, error; register_t retval; { struct ktr_request *req; struct ktr_sysret *ktp; req = ktr_getrequest(KTR_SYSRET); if (req == NULL) return; ktp = &req->ktr_data.ktr_sysret; ktp->ktr_code = code; ktp->ktr_error = error; ktp->ktr_retval = ((error == 0) ? retval: 0); /* what about val2 ? */ ktr_submitrequest(curthread, req); } /* * When a setuid process execs, disable tracing. * * XXX: We toss any pending asynchronous records. */ void ktrprocexec(struct proc *p, struct ucred **uc, struct vnode **vp) { PROC_LOCK_ASSERT(p, MA_OWNED); mtx_lock(&ktrace_mtx); ktr_freeproc(p, uc, vp); mtx_unlock(&ktrace_mtx); } /* * When a process exits, drain per-process asynchronous trace records * and disable tracing. */ void ktrprocexit(struct thread *td) { struct ktr_request *req; struct proc *p; struct ucred *cred; struct vnode *vp; p = td->td_proc; if (p->p_traceflag == 0) return; ktrace_enter(td); req = ktr_getrequest_entered(td, KTR_PROCDTOR); if (req != NULL) ktr_enqueuerequest(td, req); sx_xlock(&ktrace_sx); ktr_drain(td); sx_xunlock(&ktrace_sx); PROC_LOCK(p); mtx_lock(&ktrace_mtx); ktr_freeproc(p, &cred, &vp); mtx_unlock(&ktrace_mtx); PROC_UNLOCK(p); if (vp != NULL) vrele(vp); if (cred != NULL) crfree(cred); ktrace_exit(td); } static void ktrprocctor_entered(struct thread *td, struct proc *p) { struct ktr_proc_ctor *ktp; struct ktr_request *req; struct thread *td2; ktrace_assert(td); td2 = FIRST_THREAD_IN_PROC(p); req = ktr_getrequest_entered(td2, KTR_PROCCTOR); if (req == NULL) return; ktp = &req->ktr_data.ktr_proc_ctor; ktp->sv_flags = p->p_sysent->sv_flags; ktr_enqueuerequest(td2, req); } void ktrprocctor(struct proc *p) { struct thread *td = curthread; if ((p->p_traceflag & KTRFAC_MASK) == 0) return; ktrace_enter(td); ktrprocctor_entered(td, p); ktrace_exit(td); } /* * When a process forks, enable tracing in the new process if needed. */ void ktrprocfork(struct proc *p1, struct proc *p2) { MPASS(p2->p_tracevp == NULL); MPASS(p2->p_traceflag == 0); if (p1->p_traceflag == 0) return; PROC_LOCK(p1); mtx_lock(&ktrace_mtx); if (p1->p_traceflag & KTRFAC_INHERIT) { p2->p_traceflag = p1->p_traceflag; if ((p2->p_tracevp = p1->p_tracevp) != NULL) { VREF(p2->p_tracevp); KASSERT(p1->p_tracecred != NULL, ("ktrace vnode with no cred")); p2->p_tracecred = crhold(p1->p_tracecred); } } mtx_unlock(&ktrace_mtx); PROC_UNLOCK(p1); ktrprocctor(p2); } /* * When a thread returns, drain any asynchronous records generated by the * system call. */ void ktruserret(struct thread *td) { ktrace_enter(td); sx_xlock(&ktrace_sx); ktr_drain(td); sx_xunlock(&ktrace_sx); ktrace_exit(td); } void ktrnamei(path) char *path; { struct ktr_request *req; int namelen; char *buf = NULL; namelen = strlen(path); if (namelen > 0) { buf = malloc(namelen, M_KTRACE, M_WAITOK); bcopy(path, buf, namelen); } req = ktr_getrequest(KTR_NAMEI); if (req == NULL) { if (buf != NULL) free(buf, M_KTRACE); return; } if (namelen > 0) { req->ktr_header.ktr_len = namelen; req->ktr_buffer = buf; } ktr_submitrequest(curthread, req); } void ktrsysctl(name, namelen) int *name; u_int namelen; { struct ktr_request *req; u_int mib[CTL_MAXNAME + 2]; char *mibname; size_t mibnamelen; int error; /* Lookup name of mib. */ KASSERT(namelen <= CTL_MAXNAME, ("sysctl MIB too long")); mib[0] = 0; mib[1] = 1; bcopy(name, mib + 2, namelen * sizeof(*name)); mibnamelen = 128; mibname = malloc(mibnamelen, M_KTRACE, M_WAITOK); error = kernel_sysctl(curthread, mib, namelen + 2, mibname, &mibnamelen, NULL, 0, &mibnamelen, 0); if (error) { free(mibname, M_KTRACE); return; } req = ktr_getrequest(KTR_SYSCTL); if (req == NULL) { free(mibname, M_KTRACE); return; } req->ktr_header.ktr_len = mibnamelen; req->ktr_buffer = mibname; ktr_submitrequest(curthread, req); } void ktrgenio(fd, rw, uio, error) int fd; enum uio_rw rw; struct uio *uio; int error; { struct ktr_request *req; struct ktr_genio *ktg; int datalen; char *buf; if (error) { free(uio, M_IOV); return; } uio->uio_offset = 0; uio->uio_rw = UIO_WRITE; datalen = MIN(uio->uio_resid, ktr_geniosize); buf = malloc(datalen, M_KTRACE, M_WAITOK); error = uiomove(buf, datalen, uio); free(uio, M_IOV); if (error) { free(buf, M_KTRACE); return; } req = ktr_getrequest(KTR_GENIO); if (req == NULL) { free(buf, M_KTRACE); return; } ktg = &req->ktr_data.ktr_genio; ktg->ktr_fd = fd; ktg->ktr_rw = rw; req->ktr_header.ktr_len = datalen; req->ktr_buffer = buf; ktr_submitrequest(curthread, req); } void ktrpsig(sig, action, mask, code) int sig; sig_t action; sigset_t *mask; int code; { struct thread *td = curthread; struct ktr_request *req; struct ktr_psig *kp; req = ktr_getrequest(KTR_PSIG); if (req == NULL) return; kp = &req->ktr_data.ktr_psig; kp->signo = (char)sig; kp->action = action; kp->mask = *mask; kp->code = code; ktr_enqueuerequest(td, req); ktrace_exit(td); } void ktrcsw(out, user, wmesg) int out, user; const char *wmesg; { struct thread *td = curthread; struct ktr_request *req; struct ktr_csw *kc; req = ktr_getrequest(KTR_CSW); if (req == NULL) return; kc = &req->ktr_data.ktr_csw; kc->out = out; kc->user = user; if (wmesg != NULL) strlcpy(kc->wmesg, wmesg, sizeof(kc->wmesg)); else bzero(kc->wmesg, sizeof(kc->wmesg)); ktr_enqueuerequest(td, req); ktrace_exit(td); } void ktrstruct(name, data, datalen) const char *name; void *data; size_t datalen; { struct ktr_request *req; char *buf; size_t buflen, namelen; if (data == NULL) datalen = 0; namelen = strlen(name) + 1; buflen = namelen + datalen; buf = malloc(buflen, M_KTRACE, M_WAITOK); strcpy(buf, name); bcopy(data, buf + namelen, datalen); if ((req = ktr_getrequest(KTR_STRUCT)) == NULL) { free(buf, M_KTRACE); return; } req->ktr_buffer = buf; req->ktr_header.ktr_len = buflen; ktr_submitrequest(curthread, req); } void ktrcapfail(type, needed, held) enum ktr_cap_fail_type type; const cap_rights_t *needed; const cap_rights_t *held; { struct thread *td = curthread; struct ktr_request *req; struct ktr_cap_fail *kcf; req = ktr_getrequest(KTR_CAPFAIL); if (req == NULL) return; kcf = &req->ktr_data.ktr_cap_fail; kcf->cap_type = type; if (needed != NULL) kcf->cap_needed = *needed; else cap_rights_init(&kcf->cap_needed); if (held != NULL) kcf->cap_held = *held; else cap_rights_init(&kcf->cap_held); ktr_enqueuerequest(td, req); ktrace_exit(td); } void ktrfault(vaddr, type) vm_offset_t vaddr; int type; { struct thread *td = curthread; struct ktr_request *req; struct ktr_fault *kf; req = ktr_getrequest(KTR_FAULT); if (req == NULL) return; kf = &req->ktr_data.ktr_fault; kf->vaddr = vaddr; kf->type = type; ktr_enqueuerequest(td, req); ktrace_exit(td); } void ktrfaultend(result) int result; { struct thread *td = curthread; struct ktr_request *req; struct ktr_faultend *kf; req = ktr_getrequest(KTR_FAULTEND); if (req == NULL) return; kf = &req->ktr_data.ktr_faultend; kf->result = result; ktr_enqueuerequest(td, req); ktrace_exit(td); } #endif /* KTRACE */ /* Interface and common routines */ #ifndef _SYS_SYSPROTO_H_ struct ktrace_args { char *fname; int ops; int facs; int pid; }; #endif /* ARGSUSED */ int sys_ktrace(td, uap) struct thread *td; register struct ktrace_args *uap; { #ifdef KTRACE register struct vnode *vp = NULL; register struct proc *p; struct pgrp *pg; int facs = uap->facs & ~KTRFAC_ROOT; int ops = KTROP(uap->ops); int descend = uap->ops & KTRFLAG_DESCEND; int nfound, ret = 0; int flags, error = 0; struct nameidata nd; struct ucred *cred; /* * Need something to (un)trace. */ if (ops != KTROP_CLEARFILE && facs == 0) return (EINVAL); ktrace_enter(td); if (ops != KTROP_CLEAR) { /* * an operation which requires a file argument. */ NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, uap->fname, td); flags = FREAD | FWRITE | O_NOFOLLOW; error = vn_open(&nd, &flags, 0, NULL); if (error) { ktrace_exit(td); return (error); } NDFREE(&nd, NDF_ONLY_PNBUF); vp = nd.ni_vp; VOP_UNLOCK(vp, 0); if (vp->v_type != VREG) { (void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td); ktrace_exit(td); return (EACCES); } } /* * Clear all uses of the tracefile. */ if (ops == KTROP_CLEARFILE) { int vrele_count; vrele_count = 0; sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); if (p->p_tracevp == vp) { if (ktrcanset(td, p)) { mtx_lock(&ktrace_mtx); ktr_freeproc(p, &cred, NULL); mtx_unlock(&ktrace_mtx); vrele_count++; crfree(cred); } else error = EPERM; } PROC_UNLOCK(p); } sx_sunlock(&allproc_lock); if (vrele_count > 0) { while (vrele_count-- > 0) vrele(vp); } goto done; } /* * do it */ sx_slock(&proctree_lock); if (uap->pid < 0) { /* * by process group */ pg = pgfind(-uap->pid); if (pg == NULL) { sx_sunlock(&proctree_lock); error = ESRCH; goto done; } /* * ktrops() may call vrele(). Lock pg_members * by the proctree_lock rather than pg_mtx. */ PGRP_UNLOCK(pg); nfound = 0; LIST_FOREACH(p, &pg->pg_members, p_pglist) { PROC_LOCK(p); if (p->p_state == PRS_NEW || p_cansee(td, p) != 0) { PROC_UNLOCK(p); continue; } nfound++; if (descend) ret |= ktrsetchildren(td, p, ops, facs, vp); else ret |= ktrops(td, p, ops, facs, vp); } if (nfound == 0) { sx_sunlock(&proctree_lock); error = ESRCH; goto done; } } else { /* * by pid */ p = pfind(uap->pid); if (p == NULL) error = ESRCH; else error = p_cansee(td, p); if (error) { if (p != NULL) PROC_UNLOCK(p); sx_sunlock(&proctree_lock); goto done; } if (descend) ret |= ktrsetchildren(td, p, ops, facs, vp); else ret |= ktrops(td, p, ops, facs, vp); } sx_sunlock(&proctree_lock); if (!ret) error = EPERM; done: if (vp != NULL) (void) vn_close(vp, FWRITE, td->td_ucred, td); ktrace_exit(td); return (error); #else /* !KTRACE */ return (ENOSYS); #endif /* KTRACE */ } /* ARGSUSED */ int sys_utrace(td, uap) struct thread *td; register struct utrace_args *uap; { #ifdef KTRACE struct ktr_request *req; void *cp; int error; if (!KTRPOINT(td, KTR_USER)) return (0); if (uap->len > KTR_USER_MAXLEN) return (EINVAL); cp = malloc(uap->len, M_KTRACE, M_WAITOK); error = copyin(uap->addr, cp, uap->len); if (error) { free(cp, M_KTRACE); return (error); } req = ktr_getrequest(KTR_USER); if (req == NULL) { free(cp, M_KTRACE); return (ENOMEM); } req->ktr_buffer = cp; req->ktr_header.ktr_len = uap->len; ktr_submitrequest(td, req); return (0); #else /* !KTRACE */ return (ENOSYS); #endif /* KTRACE */ } #ifdef KTRACE static int ktrops(td, p, ops, facs, vp) struct thread *td; struct proc *p; int ops, facs; struct vnode *vp; { struct vnode *tracevp = NULL; struct ucred *tracecred = NULL; PROC_LOCK_ASSERT(p, MA_OWNED); if (!ktrcanset(td, p)) { PROC_UNLOCK(p); return (0); } if (p->p_flag & P_WEXIT) { /* If the process is exiting, just ignore it. */ PROC_UNLOCK(p); return (1); } mtx_lock(&ktrace_mtx); if (ops == KTROP_SET) { if (p->p_tracevp != vp) { /* * if trace file already in use, relinquish below */ tracevp = p->p_tracevp; VREF(vp); p->p_tracevp = vp; } if (p->p_tracecred != td->td_ucred) { tracecred = p->p_tracecred; p->p_tracecred = crhold(td->td_ucred); } p->p_traceflag |= facs; if (priv_check(td, PRIV_KTRACE) == 0) p->p_traceflag |= KTRFAC_ROOT; } else { /* KTROP_CLEAR */ if (((p->p_traceflag &= ~facs) & KTRFAC_MASK) == 0) /* no more tracing */ ktr_freeproc(p, &tracecred, &tracevp); } mtx_unlock(&ktrace_mtx); if ((p->p_traceflag & KTRFAC_MASK) != 0) ktrprocctor_entered(td, p); PROC_UNLOCK(p); if (tracevp != NULL) vrele(tracevp); if (tracecred != NULL) crfree(tracecred); return (1); } static int ktrsetchildren(td, top, ops, facs, vp) struct thread *td; struct proc *top; int ops, facs; struct vnode *vp; { register struct proc *p; register int ret = 0; p = top; PROC_LOCK_ASSERT(p, MA_OWNED); sx_assert(&proctree_lock, SX_LOCKED); for (;;) { ret |= ktrops(td, p, ops, facs, vp); /* * If this process has children, descend to them next, * otherwise do any siblings, and if done with this level, * follow back up the tree (but not past top). */ if (!LIST_EMPTY(&p->p_children)) p = LIST_FIRST(&p->p_children); else for (;;) { if (p == top) return (ret); if (LIST_NEXT(p, p_sibling)) { p = LIST_NEXT(p, p_sibling); break; } p = p->p_pptr; } PROC_LOCK(p); } /*NOTREACHED*/ } static void ktr_writerequest(struct thread *td, struct ktr_request *req) { struct ktr_header *kth; struct vnode *vp; struct proc *p; struct ucred *cred; struct uio auio; struct iovec aiov[3]; struct mount *mp; int datalen, buflen, vrele_count; int error; /* * We hold the vnode and credential for use in I/O in case ktrace is * disabled on the process as we write out the request. * * XXXRW: This is not ideal: we could end up performing a write after * the vnode has been closed. */ mtx_lock(&ktrace_mtx); vp = td->td_proc->p_tracevp; cred = td->td_proc->p_tracecred; /* * If vp is NULL, the vp has been cleared out from under this * request, so just drop it. Make sure the credential and vnode are * in sync: we should have both or neither. */ if (vp == NULL) { KASSERT(cred == NULL, ("ktr_writerequest: cred != NULL")); mtx_unlock(&ktrace_mtx); return; } VREF(vp); KASSERT(cred != NULL, ("ktr_writerequest: cred == NULL")); crhold(cred); mtx_unlock(&ktrace_mtx); kth = &req->ktr_header; KASSERT(((u_short)kth->ktr_type & ~KTR_DROP) < nitems(data_lengths), ("data_lengths array overflow")); datalen = data_lengths[(u_short)kth->ktr_type & ~KTR_DROP]; buflen = kth->ktr_len; auio.uio_iov = &aiov[0]; auio.uio_offset = 0; auio.uio_segflg = UIO_SYSSPACE; auio.uio_rw = UIO_WRITE; aiov[0].iov_base = (caddr_t)kth; aiov[0].iov_len = sizeof(struct ktr_header); auio.uio_resid = sizeof(struct ktr_header); auio.uio_iovcnt = 1; auio.uio_td = td; if (datalen != 0) { aiov[1].iov_base = (caddr_t)&req->ktr_data; aiov[1].iov_len = datalen; auio.uio_resid += datalen; auio.uio_iovcnt++; kth->ktr_len += datalen; } if (buflen != 0) { KASSERT(req->ktr_buffer != NULL, ("ktrace: nothing to write")); aiov[auio.uio_iovcnt].iov_base = req->ktr_buffer; aiov[auio.uio_iovcnt].iov_len = buflen; auio.uio_resid += buflen; auio.uio_iovcnt++; } vn_start_write(vp, &mp, V_WAIT); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); #ifdef MAC error = mac_vnode_check_write(cred, NOCRED, vp); if (error == 0) #endif error = VOP_WRITE(vp, &auio, IO_UNIT | IO_APPEND, cred); VOP_UNLOCK(vp, 0); vn_finished_write(mp); crfree(cred); if (!error) { vrele(vp); return; } /* * If error encountered, give up tracing on this vnode. We defer * all the vrele()'s on the vnode until after we are finished walking * the various lists to avoid needlessly holding locks. * NB: at this point we still hold the vnode reference that must * not go away as we need the valid vnode to compare with. Thus let * vrele_count start at 1 and the reference will be freed * by the loop at the end after our last use of vp. */ log(LOG_NOTICE, "ktrace write failed, errno %d, tracing stopped\n", error); vrele_count = 1; /* * First, clear this vnode from being used by any processes in the * system. * XXX - If one process gets an EPERM writing to the vnode, should * we really do this? Other processes might have suitable * credentials for the operation. */ cred = NULL; sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); if (p->p_tracevp == vp) { mtx_lock(&ktrace_mtx); ktr_freeproc(p, &cred, NULL); mtx_unlock(&ktrace_mtx); vrele_count++; } PROC_UNLOCK(p); if (cred != NULL) { crfree(cred); cred = NULL; } } sx_sunlock(&allproc_lock); while (vrele_count-- > 0) vrele(vp); } /* * Return true if caller has permission to set the ktracing state * of target. Essentially, the target can't possess any * more permissions than the caller. KTRFAC_ROOT signifies that * root previously set the tracing status on the target process, and * so, only root may further change it. */ static int ktrcanset(td, targetp) struct thread *td; struct proc *targetp; { PROC_LOCK_ASSERT(targetp, MA_OWNED); if (targetp->p_traceflag & KTRFAC_ROOT && priv_check(td, PRIV_KTRACE)) return (0); if (p_candebug(td, targetp) != 0) return (0); return (1); } #endif /* KTRACE */ Index: head/sys/kern/kern_lockf.c =================================================================== --- head/sys/kern/kern_lockf.c (revision 305831) +++ head/sys/kern/kern_lockf.c (revision 305832) @@ -1,2556 +1,2556 @@ /*- * Copyright (c) 2008 Isilon Inc http://www.isilon.com/ * Authors: Doug Rabson * Developed with Red Inc: Alfred Perlstein * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /*- * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Scooter Morris at Genentech Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)ufs_lockf.c 8.3 (Berkeley) 1/6/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_debug_lockf.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef LOCKF_DEBUG #include #include #include static int lockf_debug = 0; /* control debug output */ SYSCTL_INT(_debug, OID_AUTO, lockf_debug, CTLFLAG_RW, &lockf_debug, 0, ""); #endif static MALLOC_DEFINE(M_LOCKF, "lockf", "Byte-range locking structures"); struct owner_edge; struct owner_vertex; struct owner_vertex_list; struct owner_graph; #define NOLOCKF (struct lockf_entry *)0 #define SELF 0x1 #define OTHERS 0x2 static void lf_init(void *); static int lf_hash_owner(caddr_t, struct flock *, int); static int lf_owner_matches(struct lock_owner *, caddr_t, struct flock *, int); static struct lockf_entry * lf_alloc_lock(struct lock_owner *); static int lf_free_lock(struct lockf_entry *); static int lf_clearlock(struct lockf *, struct lockf_entry *); static int lf_overlaps(struct lockf_entry *, struct lockf_entry *); static int lf_blocks(struct lockf_entry *, struct lockf_entry *); static void lf_free_edge(struct lockf_edge *); static struct lockf_edge * lf_alloc_edge(void); static void lf_alloc_vertex(struct lockf_entry *); static int lf_add_edge(struct lockf_entry *, struct lockf_entry *); static void lf_remove_edge(struct lockf_edge *); static void lf_remove_outgoing(struct lockf_entry *); static void lf_remove_incoming(struct lockf_entry *); static int lf_add_outgoing(struct lockf *, struct lockf_entry *); static int lf_add_incoming(struct lockf *, struct lockf_entry *); static int lf_findoverlap(struct lockf_entry **, struct lockf_entry *, int); static struct lockf_entry * lf_getblock(struct lockf *, struct lockf_entry *); static int lf_getlock(struct lockf *, struct lockf_entry *, struct flock *); static void lf_insert_lock(struct lockf *, struct lockf_entry *); static void lf_wakeup_lock(struct lockf *, struct lockf_entry *); static void lf_update_dependancies(struct lockf *, struct lockf_entry *, int all, struct lockf_entry_list *); static void lf_set_start(struct lockf *, struct lockf_entry *, off_t, struct lockf_entry_list*); static void lf_set_end(struct lockf *, struct lockf_entry *, off_t, struct lockf_entry_list*); static int lf_setlock(struct lockf *, struct lockf_entry *, struct vnode *, void **cookiep); static int lf_cancel(struct lockf *, struct lockf_entry *, void *); static void lf_split(struct lockf *, struct lockf_entry *, struct lockf_entry *, struct lockf_entry_list *); #ifdef LOCKF_DEBUG static int graph_reaches(struct owner_vertex *x, struct owner_vertex *y, struct owner_vertex_list *path); static void graph_check(struct owner_graph *g, int checkorder); static void graph_print_vertices(struct owner_vertex_list *set); #endif static int graph_delta_forward(struct owner_graph *g, struct owner_vertex *x, struct owner_vertex *y, struct owner_vertex_list *delta); static int graph_delta_backward(struct owner_graph *g, struct owner_vertex *x, struct owner_vertex *y, struct owner_vertex_list *delta); static int graph_add_indices(int *indices, int n, struct owner_vertex_list *set); static int graph_assign_indices(struct owner_graph *g, int *indices, int nextunused, struct owner_vertex_list *set); static int graph_add_edge(struct owner_graph *g, struct owner_vertex *x, struct owner_vertex *y); static void graph_remove_edge(struct owner_graph *g, struct owner_vertex *x, struct owner_vertex *y); static struct owner_vertex *graph_alloc_vertex(struct owner_graph *g, struct lock_owner *lo); static void graph_free_vertex(struct owner_graph *g, struct owner_vertex *v); static struct owner_graph * graph_init(struct owner_graph *g); #ifdef LOCKF_DEBUG static void lf_print(char *, struct lockf_entry *); static void lf_printlist(char *, struct lockf_entry *); static void lf_print_owner(struct lock_owner *); #endif /* * This structure is used to keep track of both local and remote lock * owners. The lf_owner field of the struct lockf_entry points back at * the lock owner structure. Each possible lock owner (local proc for * POSIX fcntl locks, local file for BSD flock locks or * pair for remote locks) is represented by a unique instance of * struct lock_owner. * * If a lock owner has a lock that blocks some other lock or a lock * that is waiting for some other lock, it also has a vertex in the * owner_graph below. * * Locks: * (s) locked by state->ls_lock * (S) locked by lf_lock_states_lock * (l) locked by lf_lock_owners_lock * (g) locked by lf_owner_graph_lock * (c) const until freeing */ #define LOCK_OWNER_HASH_SIZE 256 struct lock_owner { LIST_ENTRY(lock_owner) lo_link; /* (l) hash chain */ int lo_refs; /* (l) Number of locks referring to this */ int lo_flags; /* (c) Flags passwd to lf_advlock */ caddr_t lo_id; /* (c) Id value passed to lf_advlock */ pid_t lo_pid; /* (c) Process Id of the lock owner */ int lo_sysid; /* (c) System Id of the lock owner */ struct owner_vertex *lo_vertex; /* (g) entry in deadlock graph */ }; LIST_HEAD(lock_owner_list, lock_owner); static struct sx lf_lock_states_lock; static struct lockf_list lf_lock_states; /* (S) */ static struct sx lf_lock_owners_lock; static struct lock_owner_list lf_lock_owners[LOCK_OWNER_HASH_SIZE]; /* (l) */ /* * Structures for deadlock detection. * * We have two types of directed graph, the first is the set of locks, * both active and pending on a vnode. Within this graph, active locks * are terminal nodes in the graph (i.e. have no out-going * edges). Pending locks have out-going edges to each blocking active * lock that prevents the lock from being granted and also to each * older pending lock that would block them if it was active. The * graph for each vnode is naturally acyclic; new edges are only ever * added to or from new nodes (either new pending locks which only add * out-going edges or new active locks which only add in-coming edges) * therefore they cannot create loops in the lock graph. * * The second graph is a global graph of lock owners. Each lock owner * is a vertex in that graph and an edge is added to the graph * whenever an edge is added to a vnode graph, with end points * corresponding to owner of the new pending lock and the owner of the * lock upon which it waits. In order to prevent deadlock, we only add * an edge to this graph if the new edge would not create a cycle. * * The lock owner graph is topologically sorted, i.e. if a node has * any outgoing edges, then it has an order strictly less than any * node to which it has an outgoing edge. We preserve this ordering * (and detect cycles) on edge insertion using Algorithm PK from the * paper "A Dynamic Topological Sort Algorithm for Directed Acyclic * Graphs" (ACM Journal of Experimental Algorithms, Vol 11, Article * No. 1.7) */ struct owner_vertex; struct owner_edge { LIST_ENTRY(owner_edge) e_outlink; /* (g) link from's out-edge list */ LIST_ENTRY(owner_edge) e_inlink; /* (g) link to's in-edge list */ int e_refs; /* (g) number of times added */ struct owner_vertex *e_from; /* (c) out-going from here */ struct owner_vertex *e_to; /* (c) in-coming to here */ }; LIST_HEAD(owner_edge_list, owner_edge); struct owner_vertex { TAILQ_ENTRY(owner_vertex) v_link; /* (g) workspace for edge insertion */ uint32_t v_gen; /* (g) workspace for edge insertion */ int v_order; /* (g) order of vertex in graph */ struct owner_edge_list v_outedges;/* (g) list of out-edges */ struct owner_edge_list v_inedges; /* (g) list of in-edges */ struct lock_owner *v_owner; /* (c) corresponding lock owner */ }; TAILQ_HEAD(owner_vertex_list, owner_vertex); struct owner_graph { struct owner_vertex** g_vertices; /* (g) pointers to vertices */ int g_size; /* (g) number of vertices */ int g_space; /* (g) space allocated for vertices */ int *g_indexbuf; /* (g) workspace for loop detection */ uint32_t g_gen; /* (g) increment when re-ordering */ }; static struct sx lf_owner_graph_lock; static struct owner_graph lf_owner_graph; /* * Initialise various structures and locks. */ static void lf_init(void *dummy) { int i; sx_init(&lf_lock_states_lock, "lock states lock"); LIST_INIT(&lf_lock_states); sx_init(&lf_lock_owners_lock, "lock owners lock"); for (i = 0; i < LOCK_OWNER_HASH_SIZE; i++) LIST_INIT(&lf_lock_owners[i]); sx_init(&lf_owner_graph_lock, "owner graph lock"); graph_init(&lf_owner_graph); } SYSINIT(lf_init, SI_SUB_LOCK, SI_ORDER_FIRST, lf_init, NULL); /* * Generate a hash value for a lock owner. */ static int lf_hash_owner(caddr_t id, struct flock *fl, int flags) { uint32_t h; if (flags & F_REMOTE) { h = HASHSTEP(0, fl->l_pid); h = HASHSTEP(h, fl->l_sysid); } else if (flags & F_FLOCK) { h = ((uintptr_t) id) >> 7; } else { struct proc *p = (struct proc *) id; h = HASHSTEP(0, p->p_pid); h = HASHSTEP(h, 0); } return (h % LOCK_OWNER_HASH_SIZE); } /* * Return true if a lock owner matches the details passed to * lf_advlock. */ static int lf_owner_matches(struct lock_owner *lo, caddr_t id, struct flock *fl, int flags) { if (flags & F_REMOTE) { return lo->lo_pid == fl->l_pid && lo->lo_sysid == fl->l_sysid; } else { return lo->lo_id == id; } } static struct lockf_entry * lf_alloc_lock(struct lock_owner *lo) { struct lockf_entry *lf; lf = malloc(sizeof(struct lockf_entry), M_LOCKF, M_WAITOK|M_ZERO); #ifdef LOCKF_DEBUG if (lockf_debug & 4) printf("Allocated lock %p\n", lf); #endif if (lo) { sx_xlock(&lf_lock_owners_lock); lo->lo_refs++; sx_xunlock(&lf_lock_owners_lock); lf->lf_owner = lo; } return (lf); } static int lf_free_lock(struct lockf_entry *lock) { KASSERT(lock->lf_refs > 0, ("lockf_entry negative ref count %p", lock)); if (--lock->lf_refs > 0) return (0); /* * Adjust the lock_owner reference count and * reclaim the entry if this is the last lock * for that owner. */ struct lock_owner *lo = lock->lf_owner; if (lo) { KASSERT(LIST_EMPTY(&lock->lf_outedges), ("freeing lock with dependencies")); KASSERT(LIST_EMPTY(&lock->lf_inedges), ("freeing lock with dependants")); sx_xlock(&lf_lock_owners_lock); KASSERT(lo->lo_refs > 0, ("lock owner refcount")); lo->lo_refs--; if (lo->lo_refs == 0) { #ifdef LOCKF_DEBUG if (lockf_debug & 1) printf("lf_free_lock: freeing lock owner %p\n", lo); #endif if (lo->lo_vertex) { sx_xlock(&lf_owner_graph_lock); graph_free_vertex(&lf_owner_graph, lo->lo_vertex); sx_xunlock(&lf_owner_graph_lock); } LIST_REMOVE(lo, lo_link); free(lo, M_LOCKF); #ifdef LOCKF_DEBUG if (lockf_debug & 4) printf("Freed lock owner %p\n", lo); #endif } sx_unlock(&lf_lock_owners_lock); } if ((lock->lf_flags & F_REMOTE) && lock->lf_vnode) { vrele(lock->lf_vnode); lock->lf_vnode = NULL; } #ifdef LOCKF_DEBUG if (lockf_debug & 4) printf("Freed lock %p\n", lock); #endif free(lock, M_LOCKF); return (1); } /* * Advisory record locking support */ int lf_advlockasync(struct vop_advlockasync_args *ap, struct lockf **statep, u_quad_t size) { struct lockf *state, *freestate = NULL; struct flock *fl = ap->a_fl; struct lockf_entry *lock; struct vnode *vp = ap->a_vp; caddr_t id = ap->a_id; int flags = ap->a_flags; int hash; struct lock_owner *lo; off_t start, end, oadd; int error; /* * Handle the F_UNLKSYS case first - no need to mess about * creating a lock owner for this one. */ if (ap->a_op == F_UNLCKSYS) { lf_clearremotesys(fl->l_sysid); return (0); } /* * Convert the flock structure into a start and end. */ switch (fl->l_whence) { case SEEK_SET: case SEEK_CUR: /* * Caller is responsible for adding any necessary offset * when SEEK_CUR is used. */ start = fl->l_start; break; case SEEK_END: if (size > OFF_MAX || (fl->l_start > 0 && size > OFF_MAX - fl->l_start)) return (EOVERFLOW); start = size + fl->l_start; break; default: return (EINVAL); } if (start < 0) return (EINVAL); if (fl->l_len < 0) { if (start == 0) return (EINVAL); end = start - 1; start += fl->l_len; if (start < 0) return (EINVAL); } else if (fl->l_len == 0) { end = OFF_MAX; } else { oadd = fl->l_len - 1; if (oadd > OFF_MAX - start) return (EOVERFLOW); end = start + oadd; } retry_setlock: /* * Avoid the common case of unlocking when inode has no locks. */ VI_LOCK(vp); if ((*statep) == NULL) { if (ap->a_op != F_SETLK) { fl->l_type = F_UNLCK; VI_UNLOCK(vp); return (0); } } VI_UNLOCK(vp); /* * Map our arguments to an existing lock owner or create one * if this is the first time we have seen this owner. */ hash = lf_hash_owner(id, fl, flags); sx_xlock(&lf_lock_owners_lock); LIST_FOREACH(lo, &lf_lock_owners[hash], lo_link) if (lf_owner_matches(lo, id, fl, flags)) break; if (!lo) { /* * We initialise the lock with a reference * count which matches the new lockf_entry * structure created below. */ lo = malloc(sizeof(struct lock_owner), M_LOCKF, M_WAITOK|M_ZERO); #ifdef LOCKF_DEBUG if (lockf_debug & 4) printf("Allocated lock owner %p\n", lo); #endif lo->lo_refs = 1; lo->lo_flags = flags; lo->lo_id = id; if (flags & F_REMOTE) { lo->lo_pid = fl->l_pid; lo->lo_sysid = fl->l_sysid; } else if (flags & F_FLOCK) { lo->lo_pid = -1; lo->lo_sysid = 0; } else { struct proc *p = (struct proc *) id; lo->lo_pid = p->p_pid; lo->lo_sysid = 0; } lo->lo_vertex = NULL; #ifdef LOCKF_DEBUG if (lockf_debug & 1) { printf("lf_advlockasync: new lock owner %p ", lo); lf_print_owner(lo); printf("\n"); } #endif LIST_INSERT_HEAD(&lf_lock_owners[hash], lo, lo_link); } else { /* * We have seen this lock owner before, increase its * reference count to account for the new lockf_entry * structure we create below. */ lo->lo_refs++; } sx_xunlock(&lf_lock_owners_lock); /* * Create the lockf structure. We initialise the lf_owner * field here instead of in lf_alloc_lock() to avoid paying * the lf_lock_owners_lock tax twice. */ lock = lf_alloc_lock(NULL); lock->lf_refs = 1; lock->lf_start = start; lock->lf_end = end; lock->lf_owner = lo; lock->lf_vnode = vp; if (flags & F_REMOTE) { /* * For remote locks, the caller may release its ref to * the vnode at any time - we have to ref it here to * prevent it from being recycled unexpectedly. */ vref(vp); } /* * XXX The problem is that VTOI is ufs specific, so it will * break LOCKF_DEBUG for all other FS's other than UFS because * it casts the vnode->data ptr to struct inode *. */ /* lock->lf_inode = VTOI(ap->a_vp); */ lock->lf_inode = (struct inode *)0; lock->lf_type = fl->l_type; LIST_INIT(&lock->lf_outedges); LIST_INIT(&lock->lf_inedges); lock->lf_async_task = ap->a_task; lock->lf_flags = ap->a_flags; /* * Do the requested operation. First find our state structure * and create a new one if necessary - the caller's *statep * variable and the state's ls_threads count is protected by * the vnode interlock. */ VI_LOCK(vp); if (vp->v_iflag & VI_DOOMED) { VI_UNLOCK(vp); lf_free_lock(lock); return (ENOENT); } /* * Allocate a state structure if necessary. */ state = *statep; if (state == NULL) { struct lockf *ls; VI_UNLOCK(vp); ls = malloc(sizeof(struct lockf), M_LOCKF, M_WAITOK|M_ZERO); sx_init(&ls->ls_lock, "ls_lock"); LIST_INIT(&ls->ls_active); LIST_INIT(&ls->ls_pending); ls->ls_threads = 1; sx_xlock(&lf_lock_states_lock); LIST_INSERT_HEAD(&lf_lock_states, ls, ls_link); sx_xunlock(&lf_lock_states_lock); /* * Cope if we lost a race with some other thread while * trying to allocate memory. */ VI_LOCK(vp); if (vp->v_iflag & VI_DOOMED) { VI_UNLOCK(vp); sx_xlock(&lf_lock_states_lock); LIST_REMOVE(ls, ls_link); sx_xunlock(&lf_lock_states_lock); sx_destroy(&ls->ls_lock); free(ls, M_LOCKF); lf_free_lock(lock); return (ENOENT); } if ((*statep) == NULL) { state = *statep = ls; VI_UNLOCK(vp); } else { state = *statep; state->ls_threads++; VI_UNLOCK(vp); sx_xlock(&lf_lock_states_lock); LIST_REMOVE(ls, ls_link); sx_xunlock(&lf_lock_states_lock); sx_destroy(&ls->ls_lock); free(ls, M_LOCKF); } } else { state->ls_threads++; VI_UNLOCK(vp); } sx_xlock(&state->ls_lock); /* * Recheck the doomed vnode after state->ls_lock is * locked. lf_purgelocks() requires that no new threads add * pending locks when vnode is marked by VI_DOOMED flag. */ VI_LOCK(vp); if (vp->v_iflag & VI_DOOMED) { state->ls_threads--; wakeup(state); VI_UNLOCK(vp); sx_xunlock(&state->ls_lock); lf_free_lock(lock); return (ENOENT); } VI_UNLOCK(vp); switch (ap->a_op) { case F_SETLK: error = lf_setlock(state, lock, vp, ap->a_cookiep); break; case F_UNLCK: error = lf_clearlock(state, lock); lf_free_lock(lock); break; case F_GETLK: error = lf_getlock(state, lock, fl); lf_free_lock(lock); break; case F_CANCEL: if (ap->a_cookiep) error = lf_cancel(state, lock, *ap->a_cookiep); else error = EINVAL; lf_free_lock(lock); break; default: lf_free_lock(lock); error = EINVAL; break; } #ifdef INVARIANTS /* * Check for some can't happen stuff. In this case, the active * lock list becoming disordered or containing mutually * blocking locks. We also check the pending list for locks * which should be active (i.e. have no out-going edges). */ LIST_FOREACH(lock, &state->ls_active, lf_link) { struct lockf_entry *lf; if (LIST_NEXT(lock, lf_link)) KASSERT((lock->lf_start <= LIST_NEXT(lock, lf_link)->lf_start), ("locks disordered")); LIST_FOREACH(lf, &state->ls_active, lf_link) { if (lock == lf) break; KASSERT(!lf_blocks(lock, lf), ("two conflicting active locks")); if (lock->lf_owner == lf->lf_owner) KASSERT(!lf_overlaps(lock, lf), ("two overlapping locks from same owner")); } } LIST_FOREACH(lock, &state->ls_pending, lf_link) { KASSERT(!LIST_EMPTY(&lock->lf_outedges), ("pending lock which should be active")); } #endif sx_xunlock(&state->ls_lock); /* * If we have removed the last active lock on the vnode and * this is the last thread that was in-progress, we can free * the state structure. We update the caller's pointer inside * the vnode interlock but call free outside. * * XXX alternatively, keep the state structure around until * the filesystem recycles - requires a callback from the * filesystem. */ VI_LOCK(vp); state->ls_threads--; wakeup(state); if (LIST_EMPTY(&state->ls_active) && state->ls_threads == 0) { KASSERT(LIST_EMPTY(&state->ls_pending), ("freeing state with pending locks")); freestate = state; *statep = NULL; } VI_UNLOCK(vp); if (freestate != NULL) { sx_xlock(&lf_lock_states_lock); LIST_REMOVE(freestate, ls_link); sx_xunlock(&lf_lock_states_lock); sx_destroy(&freestate->ls_lock); free(freestate, M_LOCKF); freestate = NULL; } if (error == EDOOFUS) { KASSERT(ap->a_op == F_SETLK, ("EDOOFUS")); goto retry_setlock; } return (error); } int lf_advlock(struct vop_advlock_args *ap, struct lockf **statep, u_quad_t size) { struct vop_advlockasync_args a; a.a_vp = ap->a_vp; a.a_id = ap->a_id; a.a_op = ap->a_op; a.a_fl = ap->a_fl; a.a_flags = ap->a_flags; a.a_task = NULL; a.a_cookiep = NULL; return (lf_advlockasync(&a, statep, size)); } void lf_purgelocks(struct vnode *vp, struct lockf **statep) { struct lockf *state; struct lockf_entry *lock, *nlock; /* * For this to work correctly, the caller must ensure that no * other threads enter the locking system for this vnode, * e.g. by checking VI_DOOMED. We wake up any threads that are * sleeping waiting for locks on this vnode and then free all * the remaining locks. */ VI_LOCK(vp); KASSERT(vp->v_iflag & VI_DOOMED, ("lf_purgelocks: vp %p has not vgone yet", vp)); state = *statep; if (state) { *statep = NULL; state->ls_threads++; VI_UNLOCK(vp); sx_xlock(&state->ls_lock); sx_xlock(&lf_owner_graph_lock); LIST_FOREACH_SAFE(lock, &state->ls_pending, lf_link, nlock) { LIST_REMOVE(lock, lf_link); lf_remove_outgoing(lock); lf_remove_incoming(lock); /* * If its an async lock, we can just free it * here, otherwise we let the sleeping thread * free it. */ if (lock->lf_async_task) { lf_free_lock(lock); } else { lock->lf_flags |= F_INTR; wakeup(lock); } } sx_xunlock(&lf_owner_graph_lock); sx_xunlock(&state->ls_lock); /* * Wait for all other threads, sleeping and otherwise * to leave. */ VI_LOCK(vp); while (state->ls_threads > 1) msleep(state, VI_MTX(vp), 0, "purgelocks", 0); VI_UNLOCK(vp); /* * We can just free all the active locks since they * will have no dependencies (we removed them all * above). We don't need to bother locking since we * are the last thread using this state structure. */ KASSERT(LIST_EMPTY(&state->ls_pending), ("lock pending for %p", state)); LIST_FOREACH_SAFE(lock, &state->ls_active, lf_link, nlock) { LIST_REMOVE(lock, lf_link); lf_free_lock(lock); } sx_xlock(&lf_lock_states_lock); LIST_REMOVE(state, ls_link); sx_xunlock(&lf_lock_states_lock); sx_destroy(&state->ls_lock); free(state, M_LOCKF); } else { VI_UNLOCK(vp); } } /* * Return non-zero if locks 'x' and 'y' overlap. */ static int lf_overlaps(struct lockf_entry *x, struct lockf_entry *y) { return (x->lf_start <= y->lf_end && x->lf_end >= y->lf_start); } /* * Return non-zero if lock 'x' is blocked by lock 'y' (or vice versa). */ static int lf_blocks(struct lockf_entry *x, struct lockf_entry *y) { return x->lf_owner != y->lf_owner && (x->lf_type == F_WRLCK || y->lf_type == F_WRLCK) && lf_overlaps(x, y); } /* * Allocate a lock edge from the free list */ static struct lockf_edge * lf_alloc_edge(void) { return (malloc(sizeof(struct lockf_edge), M_LOCKF, M_WAITOK|M_ZERO)); } /* * Free a lock edge. */ static void lf_free_edge(struct lockf_edge *e) { free(e, M_LOCKF); } /* * Ensure that the lock's owner has a corresponding vertex in the * owner graph. */ static void lf_alloc_vertex(struct lockf_entry *lock) { struct owner_graph *g = &lf_owner_graph; if (!lock->lf_owner->lo_vertex) lock->lf_owner->lo_vertex = graph_alloc_vertex(g, lock->lf_owner); } /* * Attempt to record an edge from lock x to lock y. Return EDEADLK if * the new edge would cause a cycle in the owner graph. */ static int lf_add_edge(struct lockf_entry *x, struct lockf_entry *y) { struct owner_graph *g = &lf_owner_graph; struct lockf_edge *e; int error; #ifdef INVARIANTS LIST_FOREACH(e, &x->lf_outedges, le_outlink) KASSERT(e->le_to != y, ("adding lock edge twice")); #endif /* * Make sure the two owners have entries in the owner graph. */ lf_alloc_vertex(x); lf_alloc_vertex(y); error = graph_add_edge(g, x->lf_owner->lo_vertex, y->lf_owner->lo_vertex); if (error) return (error); e = lf_alloc_edge(); LIST_INSERT_HEAD(&x->lf_outedges, e, le_outlink); LIST_INSERT_HEAD(&y->lf_inedges, e, le_inlink); e->le_from = x; e->le_to = y; return (0); } /* * Remove an edge from the lock graph. */ static void lf_remove_edge(struct lockf_edge *e) { struct owner_graph *g = &lf_owner_graph; struct lockf_entry *x = e->le_from; struct lockf_entry *y = e->le_to; graph_remove_edge(g, x->lf_owner->lo_vertex, y->lf_owner->lo_vertex); LIST_REMOVE(e, le_outlink); LIST_REMOVE(e, le_inlink); e->le_from = NULL; e->le_to = NULL; lf_free_edge(e); } /* * Remove all out-going edges from lock x. */ static void lf_remove_outgoing(struct lockf_entry *x) { struct lockf_edge *e; while ((e = LIST_FIRST(&x->lf_outedges)) != NULL) { lf_remove_edge(e); } } /* * Remove all in-coming edges from lock x. */ static void lf_remove_incoming(struct lockf_entry *x) { struct lockf_edge *e; while ((e = LIST_FIRST(&x->lf_inedges)) != NULL) { lf_remove_edge(e); } } /* * Walk the list of locks for the file and create an out-going edge * from lock to each blocking lock. */ static int lf_add_outgoing(struct lockf *state, struct lockf_entry *lock) { struct lockf_entry *overlap; int error; LIST_FOREACH(overlap, &state->ls_active, lf_link) { /* * We may assume that the active list is sorted by * lf_start. */ if (overlap->lf_start > lock->lf_end) break; if (!lf_blocks(lock, overlap)) continue; /* * We've found a blocking lock. Add the corresponding * edge to the graphs and see if it would cause a * deadlock. */ error = lf_add_edge(lock, overlap); /* * The only error that lf_add_edge returns is EDEADLK. * Remove any edges we added and return the error. */ if (error) { lf_remove_outgoing(lock); return (error); } } /* * We also need to add edges to sleeping locks that block * us. This ensures that lf_wakeup_lock cannot grant two * mutually blocking locks simultaneously and also enforces a * 'first come, first served' fairness model. Note that this * only happens if we are blocked by at least one active lock * due to the call to lf_getblock in lf_setlock below. */ LIST_FOREACH(overlap, &state->ls_pending, lf_link) { if (!lf_blocks(lock, overlap)) continue; /* * We've found a blocking lock. Add the corresponding * edge to the graphs and see if it would cause a * deadlock. */ error = lf_add_edge(lock, overlap); /* * The only error that lf_add_edge returns is EDEADLK. * Remove any edges we added and return the error. */ if (error) { lf_remove_outgoing(lock); return (error); } } return (0); } /* * Walk the list of pending locks for the file and create an in-coming * edge from lock to each blocking lock. */ static int lf_add_incoming(struct lockf *state, struct lockf_entry *lock) { struct lockf_entry *overlap; int error; LIST_FOREACH(overlap, &state->ls_pending, lf_link) { if (!lf_blocks(lock, overlap)) continue; /* * We've found a blocking lock. Add the corresponding * edge to the graphs and see if it would cause a * deadlock. */ error = lf_add_edge(overlap, lock); /* * The only error that lf_add_edge returns is EDEADLK. * Remove any edges we added and return the error. */ if (error) { lf_remove_incoming(lock); return (error); } } return (0); } /* * Insert lock into the active list, keeping list entries ordered by * increasing values of lf_start. */ static void lf_insert_lock(struct lockf *state, struct lockf_entry *lock) { struct lockf_entry *lf, *lfprev; if (LIST_EMPTY(&state->ls_active)) { LIST_INSERT_HEAD(&state->ls_active, lock, lf_link); return; } lfprev = NULL; LIST_FOREACH(lf, &state->ls_active, lf_link) { if (lf->lf_start > lock->lf_start) { LIST_INSERT_BEFORE(lf, lock, lf_link); return; } lfprev = lf; } LIST_INSERT_AFTER(lfprev, lock, lf_link); } /* * Wake up a sleeping lock and remove it from the pending list now * that all its dependencies have been resolved. The caller should * arrange for the lock to be added to the active list, adjusting any * existing locks for the same owner as needed. */ static void lf_wakeup_lock(struct lockf *state, struct lockf_entry *wakelock) { /* * Remove from ls_pending list and wake up the caller * or start the async notification, as appropriate. */ LIST_REMOVE(wakelock, lf_link); #ifdef LOCKF_DEBUG if (lockf_debug & 1) lf_print("lf_wakeup_lock: awakening", wakelock); #endif /* LOCKF_DEBUG */ if (wakelock->lf_async_task) { taskqueue_enqueue(taskqueue_thread, wakelock->lf_async_task); } else { wakeup(wakelock); } } /* * Re-check all dependent locks and remove edges to locks that we no * longer block. If 'all' is non-zero, the lock has been removed and * we must remove all the dependencies, otherwise it has simply been * reduced but remains active. Any pending locks which have been been * unblocked are added to 'granted' */ static void lf_update_dependancies(struct lockf *state, struct lockf_entry *lock, int all, struct lockf_entry_list *granted) { struct lockf_edge *e, *ne; struct lockf_entry *deplock; LIST_FOREACH_SAFE(e, &lock->lf_inedges, le_inlink, ne) { deplock = e->le_from; if (all || !lf_blocks(lock, deplock)) { sx_xlock(&lf_owner_graph_lock); lf_remove_edge(e); sx_xunlock(&lf_owner_graph_lock); if (LIST_EMPTY(&deplock->lf_outedges)) { lf_wakeup_lock(state, deplock); LIST_INSERT_HEAD(granted, deplock, lf_link); } } } } /* * Set the start of an existing active lock, updating dependencies and * adding any newly woken locks to 'granted'. */ static void lf_set_start(struct lockf *state, struct lockf_entry *lock, off_t new_start, struct lockf_entry_list *granted) { KASSERT(new_start >= lock->lf_start, ("can't increase lock")); lock->lf_start = new_start; LIST_REMOVE(lock, lf_link); lf_insert_lock(state, lock); lf_update_dependancies(state, lock, FALSE, granted); } /* * Set the end of an existing active lock, updating dependencies and * adding any newly woken locks to 'granted'. */ static void lf_set_end(struct lockf *state, struct lockf_entry *lock, off_t new_end, struct lockf_entry_list *granted) { KASSERT(new_end <= lock->lf_end, ("can't increase lock")); lock->lf_end = new_end; lf_update_dependancies(state, lock, FALSE, granted); } /* * Add a lock to the active list, updating or removing any current * locks owned by the same owner and processing any pending locks that * become unblocked as a result. This code is also used for unlock * since the logic for updating existing locks is identical. * * As a result of processing the new lock, we may unblock existing * pending locks as a result of downgrading/unlocking. We simply * activate the newly granted locks by looping. * * Since the new lock already has its dependencies set up, we always * add it to the list (unless its an unlock request). This may * fragment the lock list in some pathological cases but its probably * not a real problem. */ static void lf_activate_lock(struct lockf *state, struct lockf_entry *lock) { struct lockf_entry *overlap, *lf; struct lockf_entry_list granted; int ovcase; LIST_INIT(&granted); LIST_INSERT_HEAD(&granted, lock, lf_link); while (!LIST_EMPTY(&granted)) { lock = LIST_FIRST(&granted); LIST_REMOVE(lock, lf_link); /* * Skip over locks owned by other processes. Handle * any locks that overlap and are owned by ourselves. */ overlap = LIST_FIRST(&state->ls_active); for (;;) { ovcase = lf_findoverlap(&overlap, lock, SELF); #ifdef LOCKF_DEBUG if (ovcase && (lockf_debug & 2)) { printf("lf_setlock: overlap %d", ovcase); lf_print("", overlap); } #endif /* * Six cases: * 0) no overlap * 1) overlap == lock * 2) overlap contains lock * 3) lock contains overlap * 4) overlap starts before lock * 5) overlap ends after lock */ switch (ovcase) { case 0: /* no overlap */ break; case 1: /* overlap == lock */ /* * We have already setup the * dependants for the new lock, taking * into account a possible downgrade * or unlock. Remove the old lock. */ LIST_REMOVE(overlap, lf_link); lf_update_dependancies(state, overlap, TRUE, &granted); lf_free_lock(overlap); break; case 2: /* overlap contains lock */ /* * Just split the existing lock. */ lf_split(state, overlap, lock, &granted); break; case 3: /* lock contains overlap */ /* * Delete the overlap and advance to * the next entry in the list. */ lf = LIST_NEXT(overlap, lf_link); LIST_REMOVE(overlap, lf_link); lf_update_dependancies(state, overlap, TRUE, &granted); lf_free_lock(overlap); overlap = lf; continue; case 4: /* overlap starts before lock */ /* * Just update the overlap end and * move on. */ lf_set_end(state, overlap, lock->lf_start - 1, &granted); overlap = LIST_NEXT(overlap, lf_link); continue; case 5: /* overlap ends after lock */ /* * Change the start of overlap and * re-insert. */ lf_set_start(state, overlap, lock->lf_end + 1, &granted); break; } break; } #ifdef LOCKF_DEBUG if (lockf_debug & 1) { if (lock->lf_type != F_UNLCK) lf_print("lf_activate_lock: activated", lock); else lf_print("lf_activate_lock: unlocked", lock); lf_printlist("lf_activate_lock", lock); } #endif /* LOCKF_DEBUG */ if (lock->lf_type != F_UNLCK) lf_insert_lock(state, lock); } } /* * Cancel a pending lock request, either as a result of a signal or a * cancel request for an async lock. */ static void lf_cancel_lock(struct lockf *state, struct lockf_entry *lock) { struct lockf_entry_list granted; /* * Note it is theoretically possible that cancelling this lock * may allow some other pending lock to become * active. Consider this case: * * Owner Action Result Dependencies * * A: lock [0..0] succeeds * B: lock [2..2] succeeds * C: lock [1..2] blocked C->B * D: lock [0..1] blocked C->B,D->A,D->C * A: unlock [0..0] C->B,D->C * C: cancel [1..2] */ LIST_REMOVE(lock, lf_link); /* * Removing out-going edges is simple. */ sx_xlock(&lf_owner_graph_lock); lf_remove_outgoing(lock); sx_xunlock(&lf_owner_graph_lock); /* * Removing in-coming edges may allow some other lock to * become active - we use lf_update_dependancies to figure * this out. */ LIST_INIT(&granted); lf_update_dependancies(state, lock, TRUE, &granted); lf_free_lock(lock); /* * Feed any newly active locks to lf_activate_lock. */ while (!LIST_EMPTY(&granted)) { lock = LIST_FIRST(&granted); LIST_REMOVE(lock, lf_link); lf_activate_lock(state, lock); } } /* * Set a byte-range lock. */ static int lf_setlock(struct lockf *state, struct lockf_entry *lock, struct vnode *vp, void **cookiep) { static char lockstr[] = "lockf"; int error, priority, stops_deferred; #ifdef LOCKF_DEBUG if (lockf_debug & 1) lf_print("lf_setlock", lock); #endif /* LOCKF_DEBUG */ /* * Set the priority */ priority = PLOCK; if (lock->lf_type == F_WRLCK) priority += 4; if (!(lock->lf_flags & F_NOINTR)) priority |= PCATCH; /* * Scan lock list for this file looking for locks that would block us. */ if (lf_getblock(state, lock)) { /* * Free the structure and return if nonblocking. */ if ((lock->lf_flags & F_WAIT) == 0 && lock->lf_async_task == NULL) { lf_free_lock(lock); error = EAGAIN; goto out; } /* * For flock type locks, we must first remove * any shared locks that we hold before we sleep * waiting for an exclusive lock. */ if ((lock->lf_flags & F_FLOCK) && lock->lf_type == F_WRLCK) { lock->lf_type = F_UNLCK; lf_activate_lock(state, lock); lock->lf_type = F_WRLCK; } /* * We are blocked. Create edges to each blocking lock, * checking for deadlock using the owner graph. For * simplicity, we run deadlock detection for all * locks, posix and otherwise. */ sx_xlock(&lf_owner_graph_lock); error = lf_add_outgoing(state, lock); sx_xunlock(&lf_owner_graph_lock); if (error) { #ifdef LOCKF_DEBUG if (lockf_debug & 1) lf_print("lf_setlock: deadlock", lock); #endif lf_free_lock(lock); goto out; } /* * We have added edges to everything that blocks * us. Sleep until they all go away. */ LIST_INSERT_HEAD(&state->ls_pending, lock, lf_link); #ifdef LOCKF_DEBUG if (lockf_debug & 1) { struct lockf_edge *e; LIST_FOREACH(e, &lock->lf_outedges, le_outlink) { lf_print("lf_setlock: blocking on", e->le_to); lf_printlist("lf_setlock", e->le_to); } } #endif /* LOCKF_DEBUG */ if ((lock->lf_flags & F_WAIT) == 0) { /* * The caller requested async notification - * this callback happens when the blocking * lock is released, allowing the caller to * make another attempt to take the lock. */ *cookiep = (void *) lock; error = EINPROGRESS; goto out; } lock->lf_refs++; stops_deferred = sigdeferstop(SIGDEFERSTOP_ERESTART); error = sx_sleep(lock, &state->ls_lock, priority, lockstr, 0); sigallowstop(stops_deferred); if (lf_free_lock(lock)) { error = EDOOFUS; goto out; } /* * We may have been awakened by a signal and/or by a * debugger continuing us (in which cases we must * remove our lock graph edges) and/or by another * process releasing a lock (in which case our edges * have already been removed and we have been moved to * the active list). We may also have been woken by * lf_purgelocks which we report to the caller as * EINTR. In that case, lf_purgelocks will have * removed our lock graph edges. * * Note that it is possible to receive a signal after * we were successfully woken (and moved to the active * list) but before we resumed execution. In this * case, our lf_outedges list will be clear. We * pretend there was no error. * * Note also, if we have been sleeping long enough, we * may now have incoming edges from some newer lock * which is waiting behind us in the queue. */ if (lock->lf_flags & F_INTR) { error = EINTR; lf_free_lock(lock); goto out; } if (LIST_EMPTY(&lock->lf_outedges)) { error = 0; } else { lf_cancel_lock(state, lock); goto out; } #ifdef LOCKF_DEBUG if (lockf_debug & 1) { lf_print("lf_setlock: granted", lock); } #endif goto out; } /* * It looks like we are going to grant the lock. First add * edges from any currently pending lock that the new lock * would block. */ sx_xlock(&lf_owner_graph_lock); error = lf_add_incoming(state, lock); sx_xunlock(&lf_owner_graph_lock); if (error) { #ifdef LOCKF_DEBUG if (lockf_debug & 1) lf_print("lf_setlock: deadlock", lock); #endif lf_free_lock(lock); goto out; } /* * No blocks!! Add the lock. Note that we will * downgrade or upgrade any overlapping locks this * process already owns. */ lf_activate_lock(state, lock); error = 0; out: return (error); } /* * Remove a byte-range lock on an inode. * * Generally, find the lock (or an overlap to that lock) * and remove it (or shrink it), then wakeup anyone we can. */ static int lf_clearlock(struct lockf *state, struct lockf_entry *unlock) { struct lockf_entry *overlap; overlap = LIST_FIRST(&state->ls_active); if (overlap == NOLOCKF) return (0); #ifdef LOCKF_DEBUG if (unlock->lf_type != F_UNLCK) panic("lf_clearlock: bad type"); if (lockf_debug & 1) lf_print("lf_clearlock", unlock); #endif /* LOCKF_DEBUG */ lf_activate_lock(state, unlock); return (0); } /* * Check whether there is a blocking lock, and if so return its * details in '*fl'. */ static int lf_getlock(struct lockf *state, struct lockf_entry *lock, struct flock *fl) { struct lockf_entry *block; #ifdef LOCKF_DEBUG if (lockf_debug & 1) lf_print("lf_getlock", lock); #endif /* LOCKF_DEBUG */ if ((block = lf_getblock(state, lock))) { fl->l_type = block->lf_type; fl->l_whence = SEEK_SET; fl->l_start = block->lf_start; if (block->lf_end == OFF_MAX) fl->l_len = 0; else fl->l_len = block->lf_end - block->lf_start + 1; fl->l_pid = block->lf_owner->lo_pid; fl->l_sysid = block->lf_owner->lo_sysid; } else { fl->l_type = F_UNLCK; } return (0); } /* * Cancel an async lock request. */ static int lf_cancel(struct lockf *state, struct lockf_entry *lock, void *cookie) { struct lockf_entry *reallock; /* * We need to match this request with an existing lock * request. */ LIST_FOREACH(reallock, &state->ls_pending, lf_link) { if ((void *) reallock == cookie) { /* * Double-check that this lock looks right * (maybe use a rolling ID for the cancel * cookie instead?) */ if (!(reallock->lf_vnode == lock->lf_vnode && reallock->lf_start == lock->lf_start && reallock->lf_end == lock->lf_end)) { return (ENOENT); } /* * Make sure this lock was async and then just * remove it from its wait lists. */ if (!reallock->lf_async_task) { return (ENOENT); } /* * Note that since any other thread must take * state->ls_lock before it can possibly * trigger the async callback, we are safe * from a race with lf_wakeup_lock, i.e. we * can free the lock (actually our caller does * this). */ lf_cancel_lock(state, reallock); return (0); } } /* * We didn't find a matching lock - not much we can do here. */ return (ENOENT); } /* * Walk the list of locks for an inode and * return the first blocking lock. */ static struct lockf_entry * lf_getblock(struct lockf *state, struct lockf_entry *lock) { struct lockf_entry *overlap; LIST_FOREACH(overlap, &state->ls_active, lf_link) { /* * We may assume that the active list is sorted by * lf_start. */ if (overlap->lf_start > lock->lf_end) break; if (!lf_blocks(lock, overlap)) continue; return (overlap); } return (NOLOCKF); } /* * Walk the list of locks for an inode to find an overlapping lock (if * any) and return a classification of that overlap. * * Arguments: * *overlap The place in the lock list to start looking * lock The lock which is being tested * type Pass 'SELF' to test only locks with the same * owner as lock, or 'OTHER' to test only locks * with a different owner * * Returns one of six values: * 0) no overlap * 1) overlap == lock * 2) overlap contains lock * 3) lock contains overlap * 4) overlap starts before lock * 5) overlap ends after lock * * If there is an overlapping lock, '*overlap' is set to point at the * overlapping lock. * * NOTE: this returns only the FIRST overlapping lock. There * may be more than one. */ static int lf_findoverlap(struct lockf_entry **overlap, struct lockf_entry *lock, int type) { struct lockf_entry *lf; off_t start, end; int res; if ((*overlap) == NOLOCKF) { return (0); } #ifdef LOCKF_DEBUG if (lockf_debug & 2) lf_print("lf_findoverlap: looking for overlap in", lock); #endif /* LOCKF_DEBUG */ start = lock->lf_start; end = lock->lf_end; res = 0; while (*overlap) { lf = *overlap; if (lf->lf_start > end) break; if (((type & SELF) && lf->lf_owner != lock->lf_owner) || ((type & OTHERS) && lf->lf_owner == lock->lf_owner)) { *overlap = LIST_NEXT(lf, lf_link); continue; } #ifdef LOCKF_DEBUG if (lockf_debug & 2) lf_print("\tchecking", lf); #endif /* LOCKF_DEBUG */ /* * OK, check for overlap * * Six cases: * 0) no overlap * 1) overlap == lock * 2) overlap contains lock * 3) lock contains overlap * 4) overlap starts before lock * 5) overlap ends after lock */ if (start > lf->lf_end) { /* Case 0 */ #ifdef LOCKF_DEBUG if (lockf_debug & 2) printf("no overlap\n"); #endif /* LOCKF_DEBUG */ *overlap = LIST_NEXT(lf, lf_link); continue; } if (lf->lf_start == start && lf->lf_end == end) { /* Case 1 */ #ifdef LOCKF_DEBUG if (lockf_debug & 2) printf("overlap == lock\n"); #endif /* LOCKF_DEBUG */ res = 1; break; } if (lf->lf_start <= start && lf->lf_end >= end) { /* Case 2 */ #ifdef LOCKF_DEBUG if (lockf_debug & 2) printf("overlap contains lock\n"); #endif /* LOCKF_DEBUG */ res = 2; break; } if (start <= lf->lf_start && end >= lf->lf_end) { /* Case 3 */ #ifdef LOCKF_DEBUG if (lockf_debug & 2) printf("lock contains overlap\n"); #endif /* LOCKF_DEBUG */ res = 3; break; } if (lf->lf_start < start && lf->lf_end >= start) { /* Case 4 */ #ifdef LOCKF_DEBUG if (lockf_debug & 2) printf("overlap starts before lock\n"); #endif /* LOCKF_DEBUG */ res = 4; break; } if (lf->lf_start > start && lf->lf_end > end) { /* Case 5 */ #ifdef LOCKF_DEBUG if (lockf_debug & 2) printf("overlap ends after lock\n"); #endif /* LOCKF_DEBUG */ res = 5; break; } panic("lf_findoverlap: default"); } return (res); } /* * Split an the existing 'lock1', based on the extent of the lock * described by 'lock2'. The existing lock should cover 'lock2' * entirely. * * Any pending locks which have been been unblocked are added to * 'granted' */ static void lf_split(struct lockf *state, struct lockf_entry *lock1, struct lockf_entry *lock2, struct lockf_entry_list *granted) { struct lockf_entry *splitlock; #ifdef LOCKF_DEBUG if (lockf_debug & 2) { lf_print("lf_split", lock1); lf_print("splitting from", lock2); } #endif /* LOCKF_DEBUG */ /* * Check to see if we don't need to split at all. */ if (lock1->lf_start == lock2->lf_start) { lf_set_start(state, lock1, lock2->lf_end + 1, granted); return; } if (lock1->lf_end == lock2->lf_end) { lf_set_end(state, lock1, lock2->lf_start - 1, granted); return; } /* * Make a new lock consisting of the last part of * the encompassing lock. */ splitlock = lf_alloc_lock(lock1->lf_owner); memcpy(splitlock, lock1, sizeof *splitlock); splitlock->lf_refs = 1; if (splitlock->lf_flags & F_REMOTE) vref(splitlock->lf_vnode); /* * This cannot cause a deadlock since any edges we would add * to splitlock already exist in lock1. We must be sure to add * necessary dependencies to splitlock before we reduce lock1 * otherwise we may accidentally grant a pending lock that * was blocked by the tail end of lock1. */ splitlock->lf_start = lock2->lf_end + 1; LIST_INIT(&splitlock->lf_outedges); LIST_INIT(&splitlock->lf_inedges); sx_xlock(&lf_owner_graph_lock); lf_add_incoming(state, splitlock); sx_xunlock(&lf_owner_graph_lock); lf_set_end(state, lock1, lock2->lf_start - 1, granted); /* * OK, now link it in */ lf_insert_lock(state, splitlock); } struct lockdesc { STAILQ_ENTRY(lockdesc) link; struct vnode *vp; struct flock fl; }; STAILQ_HEAD(lockdesclist, lockdesc); int lf_iteratelocks_sysid(int sysid, lf_iterator *fn, void *arg) { struct lockf *ls; struct lockf_entry *lf; struct lockdesc *ldesc; struct lockdesclist locks; int error; /* * In order to keep the locking simple, we iterate over the * active lock lists to build a list of locks that need * releasing. We then call the iterator for each one in turn. * * We take an extra reference to the vnode for the duration to * make sure it doesn't go away before we are finished. */ STAILQ_INIT(&locks); sx_xlock(&lf_lock_states_lock); LIST_FOREACH(ls, &lf_lock_states, ls_link) { sx_xlock(&ls->ls_lock); LIST_FOREACH(lf, &ls->ls_active, lf_link) { if (lf->lf_owner->lo_sysid != sysid) continue; ldesc = malloc(sizeof(struct lockdesc), M_LOCKF, M_WAITOK); ldesc->vp = lf->lf_vnode; vref(ldesc->vp); ldesc->fl.l_start = lf->lf_start; if (lf->lf_end == OFF_MAX) ldesc->fl.l_len = 0; else ldesc->fl.l_len = lf->lf_end - lf->lf_start + 1; ldesc->fl.l_whence = SEEK_SET; ldesc->fl.l_type = F_UNLCK; ldesc->fl.l_pid = lf->lf_owner->lo_pid; ldesc->fl.l_sysid = sysid; STAILQ_INSERT_TAIL(&locks, ldesc, link); } sx_xunlock(&ls->ls_lock); } sx_xunlock(&lf_lock_states_lock); /* * Call the iterator function for each lock in turn. If the * iterator returns an error code, just free the rest of the * lockdesc structures. */ error = 0; while ((ldesc = STAILQ_FIRST(&locks)) != NULL) { STAILQ_REMOVE_HEAD(&locks, link); if (!error) error = fn(ldesc->vp, &ldesc->fl, arg); vrele(ldesc->vp); free(ldesc, M_LOCKF); } return (error); } int lf_iteratelocks_vnode(struct vnode *vp, lf_iterator *fn, void *arg) { struct lockf *ls; struct lockf_entry *lf; struct lockdesc *ldesc; struct lockdesclist locks; int error; /* * In order to keep the locking simple, we iterate over the * active lock lists to build a list of locks that need * releasing. We then call the iterator for each one in turn. * * We take an extra reference to the vnode for the duration to * make sure it doesn't go away before we are finished. */ STAILQ_INIT(&locks); VI_LOCK(vp); ls = vp->v_lockf; if (!ls) { VI_UNLOCK(vp); return (0); } ls->ls_threads++; VI_UNLOCK(vp); sx_xlock(&ls->ls_lock); LIST_FOREACH(lf, &ls->ls_active, lf_link) { ldesc = malloc(sizeof(struct lockdesc), M_LOCKF, M_WAITOK); ldesc->vp = lf->lf_vnode; vref(ldesc->vp); ldesc->fl.l_start = lf->lf_start; if (lf->lf_end == OFF_MAX) ldesc->fl.l_len = 0; else ldesc->fl.l_len = lf->lf_end - lf->lf_start + 1; ldesc->fl.l_whence = SEEK_SET; ldesc->fl.l_type = F_UNLCK; ldesc->fl.l_pid = lf->lf_owner->lo_pid; ldesc->fl.l_sysid = lf->lf_owner->lo_sysid; STAILQ_INSERT_TAIL(&locks, ldesc, link); } sx_xunlock(&ls->ls_lock); VI_LOCK(vp); ls->ls_threads--; wakeup(ls); VI_UNLOCK(vp); /* * Call the iterator function for each lock in turn. If the * iterator returns an error code, just free the rest of the * lockdesc structures. */ error = 0; while ((ldesc = STAILQ_FIRST(&locks)) != NULL) { STAILQ_REMOVE_HEAD(&locks, link); if (!error) error = fn(ldesc->vp, &ldesc->fl, arg); vrele(ldesc->vp); free(ldesc, M_LOCKF); } return (error); } static int lf_clearremotesys_iterator(struct vnode *vp, struct flock *fl, void *arg) { VOP_ADVLOCK(vp, 0, F_UNLCK, fl, F_REMOTE); return (0); } void lf_clearremotesys(int sysid) { KASSERT(sysid != 0, ("Can't clear local locks with F_UNLCKSYS")); lf_iteratelocks_sysid(sysid, lf_clearremotesys_iterator, NULL); } int lf_countlocks(int sysid) { int i; struct lock_owner *lo; int count; count = 0; sx_xlock(&lf_lock_owners_lock); for (i = 0; i < LOCK_OWNER_HASH_SIZE; i++) LIST_FOREACH(lo, &lf_lock_owners[i], lo_link) if (lo->lo_sysid == sysid) count += lo->lo_refs; sx_xunlock(&lf_lock_owners_lock); return (count); } #ifdef LOCKF_DEBUG /* * Return non-zero if y is reachable from x using a brute force * search. If reachable and path is non-null, return the route taken * in path. */ static int graph_reaches(struct owner_vertex *x, struct owner_vertex *y, struct owner_vertex_list *path) { struct owner_edge *e; if (x == y) { if (path) TAILQ_INSERT_HEAD(path, x, v_link); return 1; } LIST_FOREACH(e, &x->v_outedges, e_outlink) { if (graph_reaches(e->e_to, y, path)) { if (path) TAILQ_INSERT_HEAD(path, x, v_link); return 1; } } return 0; } /* * Perform consistency checks on the graph. Make sure the values of * v_order are correct. If checkorder is non-zero, check no vertex can * reach any other vertex with a smaller order. */ static void graph_check(struct owner_graph *g, int checkorder) { int i, j; for (i = 0; i < g->g_size; i++) { if (!g->g_vertices[i]->v_owner) continue; KASSERT(g->g_vertices[i]->v_order == i, ("lock graph vertices disordered")); if (checkorder) { for (j = 0; j < i; j++) { if (!g->g_vertices[j]->v_owner) continue; KASSERT(!graph_reaches(g->g_vertices[i], g->g_vertices[j], NULL), ("lock graph vertices disordered")); } } } } static void graph_print_vertices(struct owner_vertex_list *set) { struct owner_vertex *v; printf("{ "); TAILQ_FOREACH(v, set, v_link) { printf("%d:", v->v_order); lf_print_owner(v->v_owner); if (TAILQ_NEXT(v, v_link)) printf(", "); } printf(" }\n"); } #endif /* * Calculate the sub-set of vertices v from the affected region [y..x] * where v is reachable from y. Return -1 if a loop was detected * (i.e. x is reachable from y, otherwise the number of vertices in * this subset. */ static int graph_delta_forward(struct owner_graph *g, struct owner_vertex *x, struct owner_vertex *y, struct owner_vertex_list *delta) { uint32_t gen; struct owner_vertex *v; struct owner_edge *e; int n; /* * We start with a set containing just y. Then for each vertex * v in the set so far unprocessed, we add each vertex that v * has an out-edge to and that is within the affected region * [y..x]. If we see the vertex x on our travels, stop * immediately. */ TAILQ_INIT(delta); TAILQ_INSERT_TAIL(delta, y, v_link); v = y; n = 1; gen = g->g_gen; while (v) { LIST_FOREACH(e, &v->v_outedges, e_outlink) { if (e->e_to == x) return -1; if (e->e_to->v_order < x->v_order && e->e_to->v_gen != gen) { e->e_to->v_gen = gen; TAILQ_INSERT_TAIL(delta, e->e_to, v_link); n++; } } v = TAILQ_NEXT(v, v_link); } return (n); } /* * Calculate the sub-set of vertices v from the affected region [y..x] * where v reaches x. Return the number of vertices in this subset. */ static int graph_delta_backward(struct owner_graph *g, struct owner_vertex *x, struct owner_vertex *y, struct owner_vertex_list *delta) { uint32_t gen; struct owner_vertex *v; struct owner_edge *e; int n; /* * We start with a set containing just x. Then for each vertex * v in the set so far unprocessed, we add each vertex that v * has an in-edge from and that is within the affected region * [y..x]. */ TAILQ_INIT(delta); TAILQ_INSERT_TAIL(delta, x, v_link); v = x; n = 1; gen = g->g_gen; while (v) { LIST_FOREACH(e, &v->v_inedges, e_inlink) { if (e->e_from->v_order > y->v_order && e->e_from->v_gen != gen) { e->e_from->v_gen = gen; TAILQ_INSERT_HEAD(delta, e->e_from, v_link); n++; } } v = TAILQ_PREV(v, owner_vertex_list, v_link); } return (n); } static int graph_add_indices(int *indices, int n, struct owner_vertex_list *set) { struct owner_vertex *v; int i, j; TAILQ_FOREACH(v, set, v_link) { for (i = n; i > 0 && indices[i - 1] > v->v_order; i--) ; for (j = n - 1; j >= i; j--) indices[j + 1] = indices[j]; indices[i] = v->v_order; n++; } return (n); } static int graph_assign_indices(struct owner_graph *g, int *indices, int nextunused, struct owner_vertex_list *set) { struct owner_vertex *v, *vlowest; while (!TAILQ_EMPTY(set)) { vlowest = NULL; TAILQ_FOREACH(v, set, v_link) { if (!vlowest || v->v_order < vlowest->v_order) vlowest = v; } TAILQ_REMOVE(set, vlowest, v_link); vlowest->v_order = indices[nextunused]; g->g_vertices[vlowest->v_order] = vlowest; nextunused++; } return (nextunused); } static int graph_add_edge(struct owner_graph *g, struct owner_vertex *x, struct owner_vertex *y) { struct owner_edge *e; struct owner_vertex_list deltaF, deltaB; int nF, nB, n, vi, i; int *indices; sx_assert(&lf_owner_graph_lock, SX_XLOCKED); LIST_FOREACH(e, &x->v_outedges, e_outlink) { if (e->e_to == y) { e->e_refs++; return (0); } } #ifdef LOCKF_DEBUG if (lockf_debug & 8) { printf("adding edge %d:", x->v_order); lf_print_owner(x->v_owner); printf(" -> %d:", y->v_order); lf_print_owner(y->v_owner); printf("\n"); } #endif if (y->v_order < x->v_order) { /* * The new edge violates the order. First find the set * of affected vertices reachable from y (deltaF) and * the set of affect vertices affected that reach x * (deltaB), using the graph generation number to * detect whether we have visited a given vertex * already. We re-order the graph so that each vertex * in deltaB appears before each vertex in deltaF. * * If x is a member of deltaF, then the new edge would * create a cycle. Otherwise, we may assume that * deltaF and deltaB are disjoint. */ g->g_gen++; if (g->g_gen == 0) { /* * Generation wrap. */ for (vi = 0; vi < g->g_size; vi++) { g->g_vertices[vi]->v_gen = 0; } g->g_gen++; } nF = graph_delta_forward(g, x, y, &deltaF); if (nF < 0) { #ifdef LOCKF_DEBUG if (lockf_debug & 8) { struct owner_vertex_list path; printf("deadlock: "); TAILQ_INIT(&path); graph_reaches(y, x, &path); graph_print_vertices(&path); } #endif return (EDEADLK); } #ifdef LOCKF_DEBUG if (lockf_debug & 8) { printf("re-ordering graph vertices\n"); printf("deltaF = "); graph_print_vertices(&deltaF); } #endif nB = graph_delta_backward(g, x, y, &deltaB); #ifdef LOCKF_DEBUG if (lockf_debug & 8) { printf("deltaB = "); graph_print_vertices(&deltaB); } #endif /* * We first build a set of vertex indices (vertex * order values) that we may use, then we re-assign * orders first to those vertices in deltaB, then to * deltaF. Note that the contents of deltaF and deltaB * may be partially disordered - we perform an * insertion sort while building our index set. */ indices = g->g_indexbuf; n = graph_add_indices(indices, 0, &deltaF); graph_add_indices(indices, n, &deltaB); /* * We must also be sure to maintain the relative * ordering of deltaF and deltaB when re-assigning * vertices. We do this by iteratively removing the * lowest ordered element from the set and assigning * it the next value from our new ordering. */ i = graph_assign_indices(g, indices, 0, &deltaB); graph_assign_indices(g, indices, i, &deltaF); #ifdef LOCKF_DEBUG if (lockf_debug & 8) { struct owner_vertex_list set; TAILQ_INIT(&set); for (i = 0; i < nB + nF; i++) TAILQ_INSERT_TAIL(&set, g->g_vertices[indices[i]], v_link); printf("new ordering = "); graph_print_vertices(&set); } #endif } KASSERT(x->v_order < y->v_order, ("Failed to re-order graph")); #ifdef LOCKF_DEBUG if (lockf_debug & 8) { graph_check(g, TRUE); } #endif e = malloc(sizeof(struct owner_edge), M_LOCKF, M_WAITOK); LIST_INSERT_HEAD(&x->v_outedges, e, e_outlink); LIST_INSERT_HEAD(&y->v_inedges, e, e_inlink); e->e_refs = 1; e->e_from = x; e->e_to = y; return (0); } /* * Remove an edge x->y from the graph. */ static void graph_remove_edge(struct owner_graph *g, struct owner_vertex *x, struct owner_vertex *y) { struct owner_edge *e; sx_assert(&lf_owner_graph_lock, SX_XLOCKED); LIST_FOREACH(e, &x->v_outedges, e_outlink) { if (e->e_to == y) break; } KASSERT(e, ("Removing non-existent edge from deadlock graph")); e->e_refs--; if (e->e_refs == 0) { #ifdef LOCKF_DEBUG if (lockf_debug & 8) { printf("removing edge %d:", x->v_order); lf_print_owner(x->v_owner); printf(" -> %d:", y->v_order); lf_print_owner(y->v_owner); printf("\n"); } #endif LIST_REMOVE(e, e_outlink); LIST_REMOVE(e, e_inlink); free(e, M_LOCKF); } } /* * Allocate a vertex from the free list. Return ENOMEM if there are * none. */ static struct owner_vertex * graph_alloc_vertex(struct owner_graph *g, struct lock_owner *lo) { struct owner_vertex *v; sx_assert(&lf_owner_graph_lock, SX_XLOCKED); v = malloc(sizeof(struct owner_vertex), M_LOCKF, M_WAITOK); if (g->g_size == g->g_space) { g->g_vertices = realloc(g->g_vertices, 2 * g->g_space * sizeof(struct owner_vertex *), M_LOCKF, M_WAITOK); free(g->g_indexbuf, M_LOCKF); g->g_indexbuf = malloc(2 * g->g_space * sizeof(int), M_LOCKF, M_WAITOK); g->g_space = 2 * g->g_space; } v->v_order = g->g_size; v->v_gen = g->g_gen; g->g_vertices[g->g_size] = v; g->g_size++; LIST_INIT(&v->v_outedges); LIST_INIT(&v->v_inedges); v->v_owner = lo; return (v); } static void graph_free_vertex(struct owner_graph *g, struct owner_vertex *v) { struct owner_vertex *w; int i; sx_assert(&lf_owner_graph_lock, SX_XLOCKED); KASSERT(LIST_EMPTY(&v->v_outedges), ("Freeing vertex with edges")); KASSERT(LIST_EMPTY(&v->v_inedges), ("Freeing vertex with edges")); /* * Remove from the graph's array and close up the gap, * renumbering the other vertices. */ for (i = v->v_order + 1; i < g->g_size; i++) { w = g->g_vertices[i]; w->v_order--; g->g_vertices[i - 1] = w; } g->g_size--; free(v, M_LOCKF); } static struct owner_graph * graph_init(struct owner_graph *g) { g->g_vertices = malloc(10 * sizeof(struct owner_vertex *), M_LOCKF, M_WAITOK); g->g_size = 0; g->g_space = 10; g->g_indexbuf = malloc(g->g_space * sizeof(int), M_LOCKF, M_WAITOK); g->g_gen = 0; return (g); } #ifdef LOCKF_DEBUG /* * Print description of a lock owner */ static void lf_print_owner(struct lock_owner *lo) { if (lo->lo_flags & F_REMOTE) { printf("remote pid %d, system %d", lo->lo_pid, lo->lo_sysid); } else if (lo->lo_flags & F_FLOCK) { printf("file %p", lo->lo_id); } else { printf("local pid %d", lo->lo_pid); } } /* * Print out a lock. */ static void lf_print(char *tag, struct lockf_entry *lock) { printf("%s: lock %p for ", tag, (void *)lock); lf_print_owner(lock->lf_owner); if (lock->lf_inode != (struct inode *)0) printf(" in ino %ju on dev <%s>,", (uintmax_t)lock->lf_inode->i_number, devtoname(lock->lf_inode->i_dev)); printf(" %s, start %jd, end ", lock->lf_type == F_RDLCK ? "shared" : lock->lf_type == F_WRLCK ? "exclusive" : lock->lf_type == F_UNLCK ? "unlock" : "unknown", (intmax_t)lock->lf_start); if (lock->lf_end == OFF_MAX) printf("EOF"); else printf("%jd", (intmax_t)lock->lf_end); if (!LIST_EMPTY(&lock->lf_outedges)) printf(" block %p\n", (void *)LIST_FIRST(&lock->lf_outedges)->le_to); else printf("\n"); } static void lf_printlist(char *tag, struct lockf_entry *lock) { struct lockf_entry *lf, *blk; struct lockf_edge *e; if (lock->lf_inode == (struct inode *)0) return; printf("%s: Lock list for ino %ju on dev <%s>:\n", tag, (uintmax_t)lock->lf_inode->i_number, devtoname(lock->lf_inode->i_dev)); LIST_FOREACH(lf, &lock->lf_vnode->v_lockf->ls_active, lf_link) { printf("\tlock %p for ",(void *)lf); lf_print_owner(lock->lf_owner); printf(", %s, start %jd, end %jd", lf->lf_type == F_RDLCK ? "shared" : lf->lf_type == F_WRLCK ? "exclusive" : lf->lf_type == F_UNLCK ? "unlock" : "unknown", (intmax_t)lf->lf_start, (intmax_t)lf->lf_end); LIST_FOREACH(e, &lf->lf_outedges, le_outlink) { blk = e->le_to; printf("\n\t\tlock request %p for ", (void *)blk); lf_print_owner(blk->lf_owner); printf(", %s, start %jd, end %jd", blk->lf_type == F_RDLCK ? "shared" : blk->lf_type == F_WRLCK ? "exclusive" : blk->lf_type == F_UNLCK ? "unlock" : "unknown", (intmax_t)blk->lf_start, (intmax_t)blk->lf_end); if (!LIST_EMPTY(&blk->lf_inedges)) panic("lf_printlist: bad list"); } printf("\n"); } } #endif /* LOCKF_DEBUG */ Index: head/sys/kern/kern_malloc.c =================================================================== --- head/sys/kern/kern_malloc.c (revision 305831) +++ head/sys/kern/kern_malloc.c (revision 305832) @@ -1,1115 +1,1115 @@ /*- * Copyright (c) 1987, 1991, 1993 * The Regents of the University of California. * Copyright (c) 2005-2009 Robert N. M. Watson * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_malloc.c 8.3 (Berkeley) 1/4/94 */ /* * Kernel malloc(9) implementation -- general purpose kernel memory allocator * based on memory types. Back end is implemented using the UMA(9) zone * allocator. A set of fixed-size buckets are used for smaller allocations, * and a special UMA allocation interface is used for larger allocations. * Callers declare memory types, and statistics are maintained independently * for each memory type. Statistics are maintained per-CPU for performance * reasons. See malloc(9) and comments in malloc.h for a detailed * description. */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include "opt_vm.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DEBUG_MEMGUARD #include #endif #ifdef DEBUG_REDZONE #include #endif #if defined(INVARIANTS) && defined(__i386__) #include #endif #include #ifdef KDTRACE_HOOKS #include dtrace_malloc_probe_func_t dtrace_malloc_probe; #endif /* * When realloc() is called, if the new size is sufficiently smaller than * the old size, realloc() will allocate a new, smaller block to avoid * wasting memory. 'Sufficiently smaller' is defined as: newsize <= * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'. */ #ifndef REALLOC_FRACTION #define REALLOC_FRACTION 1 /* new block if <= half the size */ #endif /* * Centrally define some common malloc types. */ MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches"); MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory"); MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers"); static struct malloc_type *kmemstatistics; static int kmemcount; #define KMEM_ZSHIFT 4 #define KMEM_ZBASE 16 #define KMEM_ZMASK (KMEM_ZBASE - 1) #define KMEM_ZMAX 65536 #define KMEM_ZSIZE (KMEM_ZMAX >> KMEM_ZSHIFT) static uint8_t kmemsize[KMEM_ZSIZE + 1]; #ifndef MALLOC_DEBUG_MAXZONES #define MALLOC_DEBUG_MAXZONES 1 #endif static int numzones = MALLOC_DEBUG_MAXZONES; /* * Small malloc(9) memory allocations are allocated from a set of UMA buckets * of various sizes. * * XXX: The comment here used to read "These won't be powers of two for * long." It's possible that a significant amount of wasted memory could be * recovered by tuning the sizes of these buckets. */ struct { int kz_size; char *kz_name; uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES]; } kmemzones[] = { {16, "16", }, {32, "32", }, {64, "64", }, {128, "128", }, {256, "256", }, {512, "512", }, {1024, "1024", }, {2048, "2048", }, {4096, "4096", }, {8192, "8192", }, {16384, "16384", }, {32768, "32768", }, {65536, "65536", }, {0, NULL}, }; /* * Zone to allocate malloc type descriptions from. For ABI reasons, memory * types are described by a data structure passed by the declaring code, but * the malloc(9) implementation has its own data structure describing the * type and statistics. This permits the malloc(9)-internal data structures * to be modified without breaking binary-compiled kernel modules that * declare malloc types. */ static uma_zone_t mt_zone; u_long vm_kmem_size; SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0, "Size of kernel memory"); static u_long kmem_zmax = KMEM_ZMAX; SYSCTL_ULONG(_vm, OID_AUTO, kmem_zmax, CTLFLAG_RDTUN, &kmem_zmax, 0, "Maximum allocation size that malloc(9) would use UMA as backend"); static u_long vm_kmem_size_min; SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0, "Minimum size of kernel memory"); static u_long vm_kmem_size_max; SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0, "Maximum size of kernel memory"); static u_int vm_kmem_size_scale; SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0, "Scale factor for kernel memory size"); static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size, CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, sysctl_kmem_map_size, "LU", "Current kmem allocation size"); static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free, CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, sysctl_kmem_map_free, "LU", "Free space in kmem"); /* * The malloc_mtx protects the kmemstatistics linked list. */ struct mtx malloc_mtx; #ifdef MALLOC_PROFILE uint64_t krequests[KMEM_ZSIZE + 1]; static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS); #endif static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS); /* * time_uptime of the last malloc(9) failure (induced or real). */ static time_t t_malloc_fail; #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1) static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0, "Kernel malloc debugging options"); #endif /* * malloc(9) fault injection -- cause malloc failures every (n) mallocs when * the caller specifies M_NOWAIT. If set to 0, no failures are caused. */ #ifdef MALLOC_MAKE_FAILURES static int malloc_failure_rate; static int malloc_nowait_count; static int malloc_failure_count; SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RWTUN, &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail"); SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD, &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures"); #endif static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS) { u_long size; size = vmem_size(kmem_arena, VMEM_ALLOC); return (sysctl_handle_long(oidp, &size, 0, req)); } static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS) { u_long size; size = vmem_size(kmem_arena, VMEM_FREE); return (sysctl_handle_long(oidp, &size, 0, req)); } /* * malloc(9) uma zone separation -- sub-page buffer overruns in one * malloc type will affect only a subset of other malloc types. */ #if MALLOC_DEBUG_MAXZONES > 1 static void tunable_set_numzones(void) { TUNABLE_INT_FETCH("debug.malloc.numzones", &numzones); /* Sanity check the number of malloc uma zones. */ if (numzones <= 0) numzones = 1; if (numzones > MALLOC_DEBUG_MAXZONES) numzones = MALLOC_DEBUG_MAXZONES; } SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL); SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &numzones, 0, "Number of malloc uma subzones"); /* * Any number that changes regularly is an okay choice for the * offset. Build numbers are pretty good of you have them. */ static u_int zone_offset = __FreeBSD_version; TUNABLE_INT("debug.malloc.zone_offset", &zone_offset); SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN, &zone_offset, 0, "Separate malloc types by examining the " "Nth character in the malloc type short description."); static u_int mtp_get_subzone(const char *desc) { size_t len; u_int val; if (desc == NULL || (len = strlen(desc)) == 0) return (0); val = desc[zone_offset % len]; return (val % numzones); } #elif MALLOC_DEBUG_MAXZONES == 0 #error "MALLOC_DEBUG_MAXZONES must be positive." #else static inline u_int mtp_get_subzone(const char *desc) { return (0); } #endif /* MALLOC_DEBUG_MAXZONES > 1 */ int malloc_last_fail(void) { return (time_uptime - t_malloc_fail); } /* * An allocation has succeeded -- update malloc type statistics for the * amount of bucket size. Occurs within a critical section so that the * thread isn't preempted and doesn't migrate while updating per-PCU * statistics. */ static void malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size, int zindx) { struct malloc_type_internal *mtip; struct malloc_type_stats *mtsp; critical_enter(); mtip = mtp->ks_handle; mtsp = &mtip->mti_stats[curcpu]; if (size > 0) { mtsp->mts_memalloced += size; mtsp->mts_numallocs++; } if (zindx != -1) mtsp->mts_size |= 1 << zindx; #ifdef KDTRACE_HOOKS if (dtrace_malloc_probe != NULL) { uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC]; if (probe_id != 0) (dtrace_malloc_probe)(probe_id, (uintptr_t) mtp, (uintptr_t) mtip, (uintptr_t) mtsp, size, zindx); } #endif critical_exit(); } void malloc_type_allocated(struct malloc_type *mtp, unsigned long size) { if (size > 0) malloc_type_zone_allocated(mtp, size, -1); } /* * A free operation has occurred -- update malloc type statistics for the * amount of the bucket size. Occurs within a critical section so that the * thread isn't preempted and doesn't migrate while updating per-CPU * statistics. */ void malloc_type_freed(struct malloc_type *mtp, unsigned long size) { struct malloc_type_internal *mtip; struct malloc_type_stats *mtsp; critical_enter(); mtip = mtp->ks_handle; mtsp = &mtip->mti_stats[curcpu]; mtsp->mts_memfreed += size; mtsp->mts_numfrees++; #ifdef KDTRACE_HOOKS if (dtrace_malloc_probe != NULL) { uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE]; if (probe_id != 0) (dtrace_malloc_probe)(probe_id, (uintptr_t) mtp, (uintptr_t) mtip, (uintptr_t) mtsp, size, 0); } #endif critical_exit(); } /* * contigmalloc: * * Allocate a block of physically contiguous memory. * * If M_NOWAIT is set, this routine will not block and return NULL if * the allocation fails. */ void * contigmalloc(unsigned long size, struct malloc_type *type, int flags, vm_paddr_t low, vm_paddr_t high, unsigned long alignment, vm_paddr_t boundary) { void *ret; ret = (void *)kmem_alloc_contig(kernel_arena, size, flags, low, high, alignment, boundary, VM_MEMATTR_DEFAULT); if (ret != NULL) malloc_type_allocated(type, round_page(size)); return (ret); } /* * contigfree: * * Free a block of memory allocated by contigmalloc. * * This routine may not block. */ void contigfree(void *addr, unsigned long size, struct malloc_type *type) { kmem_free(kernel_arena, (vm_offset_t)addr, size); malloc_type_freed(type, round_page(size)); } /* * malloc: * * Allocate a block of memory. * * If M_NOWAIT is set, this routine will not block and return NULL if * the allocation fails. */ void * malloc(unsigned long size, struct malloc_type *mtp, int flags) { int indx; struct malloc_type_internal *mtip; caddr_t va; uma_zone_t zone; #if defined(DIAGNOSTIC) || defined(DEBUG_REDZONE) unsigned long osize = size; #endif #ifdef INVARIANTS KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic")); /* * Check that exactly one of M_WAITOK or M_NOWAIT is specified. */ indx = flags & (M_WAITOK | M_NOWAIT); if (indx != M_NOWAIT && indx != M_WAITOK) { static struct timeval lasterr; static int curerr, once; if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) { printf("Bad malloc flags: %x\n", indx); kdb_backtrace(); flags |= M_WAITOK; once++; } } #endif #ifdef MALLOC_MAKE_FAILURES if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) { atomic_add_int(&malloc_nowait_count, 1); if ((malloc_nowait_count % malloc_failure_rate) == 0) { atomic_add_int(&malloc_failure_count, 1); t_malloc_fail = time_uptime; return (NULL); } } #endif if (flags & M_WAITOK) KASSERT(curthread->td_intr_nesting_level == 0, ("malloc(M_WAITOK) in interrupt context")); KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), ("malloc: called with spinlock or critical section held")); #ifdef DEBUG_MEMGUARD if (memguard_cmp_mtp(mtp, size)) { va = memguard_alloc(size, flags); if (va != NULL) return (va); /* This is unfortunate but should not be fatal. */ } #endif #ifdef DEBUG_REDZONE size = redzone_size_ntor(size); #endif if (size <= kmem_zmax) { mtip = mtp->ks_handle; if (size & KMEM_ZMASK) size = (size & ~KMEM_ZMASK) + KMEM_ZBASE; indx = kmemsize[size >> KMEM_ZSHIFT]; KASSERT(mtip->mti_zone < numzones, ("mti_zone %u out of range %d", mtip->mti_zone, numzones)); zone = kmemzones[indx].kz_zone[mtip->mti_zone]; #ifdef MALLOC_PROFILE krequests[size >> KMEM_ZSHIFT]++; #endif va = uma_zalloc(zone, flags); if (va != NULL) size = zone->uz_size; malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx); } else { size = roundup(size, PAGE_SIZE); zone = NULL; va = uma_large_malloc(size, flags); malloc_type_allocated(mtp, va == NULL ? 0 : size); } if (flags & M_WAITOK) KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL")); else if (va == NULL) t_malloc_fail = time_uptime; #ifdef DIAGNOSTIC if (va != NULL && !(flags & M_ZERO)) { memset(va, 0x70, osize); } #endif #ifdef DEBUG_REDZONE if (va != NULL) va = redzone_setup(va, osize); #endif return ((void *) va); } /* * free: * * Free a block of memory allocated by malloc. * * This routine may not block. */ void free(void *addr, struct malloc_type *mtp) { uma_slab_t slab; u_long size; KASSERT(mtp->ks_magic == M_MAGIC, ("free: bad malloc type magic")); KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), ("free: called with spinlock or critical section held")); /* free(NULL, ...) does nothing */ if (addr == NULL) return; #ifdef DEBUG_MEMGUARD if (is_memguard_addr(addr)) { memguard_free(addr); return; } #endif #ifdef DEBUG_REDZONE redzone_check(addr); addr = redzone_addr_ntor(addr); #endif slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK)); if (slab == NULL) panic("free: address %p(%p) has not been allocated.\n", addr, (void *)((u_long)addr & (~UMA_SLAB_MASK))); if (!(slab->us_flags & UMA_SLAB_MALLOC)) { #ifdef INVARIANTS struct malloc_type **mtpp = addr; #endif size = slab->us_keg->uk_size; #ifdef INVARIANTS /* * Cache a pointer to the malloc_type that most recently freed * this memory here. This way we know who is most likely to * have stepped on it later. * * This code assumes that size is a multiple of 8 bytes for * 64 bit machines */ mtpp = (struct malloc_type **) ((unsigned long)mtpp & ~UMA_ALIGN_PTR); mtpp += (size - sizeof(struct malloc_type *)) / sizeof(struct malloc_type *); *mtpp = mtp; #endif uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab); } else { size = slab->us_size; uma_large_free(slab); } malloc_type_freed(mtp, size); } /* * realloc: change the size of a memory block */ void * realloc(void *addr, unsigned long size, struct malloc_type *mtp, int flags) { uma_slab_t slab; unsigned long alloc; void *newaddr; KASSERT(mtp->ks_magic == M_MAGIC, ("realloc: bad malloc type magic")); KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), ("realloc: called with spinlock or critical section held")); /* realloc(NULL, ...) is equivalent to malloc(...) */ if (addr == NULL) return (malloc(size, mtp, flags)); /* * XXX: Should report free of old memory and alloc of new memory to * per-CPU stats. */ #ifdef DEBUG_MEMGUARD if (is_memguard_addr(addr)) return (memguard_realloc(addr, size, mtp, flags)); #endif #ifdef DEBUG_REDZONE slab = NULL; alloc = redzone_get_size(addr); #else slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK)); /* Sanity check */ KASSERT(slab != NULL, ("realloc: address %p out of range", (void *)addr)); /* Get the size of the original block */ if (!(slab->us_flags & UMA_SLAB_MALLOC)) alloc = slab->us_keg->uk_size; else alloc = slab->us_size; /* Reuse the original block if appropriate */ if (size <= alloc && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE)) return (addr); #endif /* !DEBUG_REDZONE */ /* Allocate a new, bigger (or smaller) block */ if ((newaddr = malloc(size, mtp, flags)) == NULL) return (NULL); /* Copy over original contents */ bcopy(addr, newaddr, min(size, alloc)); free(addr, mtp); return (newaddr); } /* * reallocf: same as realloc() but free memory on failure. */ void * reallocf(void *addr, unsigned long size, struct malloc_type *mtp, int flags) { void *mem; if ((mem = realloc(addr, size, mtp, flags)) == NULL) free(addr, mtp); return (mem); } /* * Wake the uma reclamation pagedaemon thread when we exhaust KVA. It * will call the lowmem handler and uma_reclaim() callbacks in a * context that is safe. */ static void kmem_reclaim(vmem_t *vm, int flags) { uma_reclaim_wakeup(); pagedaemon_wakeup(); } #ifndef __sparc64__ CTASSERT(VM_KMEM_SIZE_SCALE >= 1); #endif /* * Initialize the kernel memory (kmem) arena. */ void kmeminit(void) { u_long mem_size; u_long tmp; #ifdef VM_KMEM_SIZE if (vm_kmem_size == 0) vm_kmem_size = VM_KMEM_SIZE; #endif #ifdef VM_KMEM_SIZE_MIN if (vm_kmem_size_min == 0) vm_kmem_size_min = VM_KMEM_SIZE_MIN; #endif #ifdef VM_KMEM_SIZE_MAX if (vm_kmem_size_max == 0) vm_kmem_size_max = VM_KMEM_SIZE_MAX; #endif /* * Calculate the amount of kernel virtual address (KVA) space that is * preallocated to the kmem arena. In order to support a wide range * of machines, it is a function of the physical memory size, * specifically, * * min(max(physical memory size / VM_KMEM_SIZE_SCALE, * VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX) * * Every architecture must define an integral value for * VM_KMEM_SIZE_SCALE. However, the definitions of VM_KMEM_SIZE_MIN * and VM_KMEM_SIZE_MAX, which represent respectively the floor and * ceiling on this preallocation, are optional. Typically, * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on * a given architecture. */ mem_size = vm_cnt.v_page_count; if (mem_size <= 32768) /* delphij XXX 128MB */ kmem_zmax = PAGE_SIZE; if (vm_kmem_size_scale < 1) vm_kmem_size_scale = VM_KMEM_SIZE_SCALE; /* * Check if we should use defaults for the "vm_kmem_size" * variable: */ if (vm_kmem_size == 0) { vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE; if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min) vm_kmem_size = vm_kmem_size_min; if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max) vm_kmem_size = vm_kmem_size_max; } /* * The amount of KVA space that is preallocated to the * kmem arena can be set statically at compile-time or manually * through the kernel environment. However, it is still limited to * twice the physical memory size, which has been sufficient to handle * the most severe cases of external fragmentation in the kmem arena. */ if (vm_kmem_size / 2 / PAGE_SIZE > mem_size) vm_kmem_size = 2 * mem_size * PAGE_SIZE; vm_kmem_size = round_page(vm_kmem_size); #ifdef DEBUG_MEMGUARD tmp = memguard_fudge(vm_kmem_size, kernel_map); #else tmp = vm_kmem_size; #endif vmem_init(kmem_arena, "kmem arena", kva_alloc(tmp), tmp, PAGE_SIZE, 0, 0); vmem_set_reclaim(kmem_arena, kmem_reclaim); #ifdef DEBUG_MEMGUARD /* * Initialize MemGuard if support compiled in. MemGuard is a * replacement allocator used for detecting tamper-after-free * scenarios as they occur. It is only used for debugging. */ memguard_init(kmem_arena); #endif } /* * Initialize the kernel memory allocator */ /* ARGSUSED*/ static void mallocinit(void *dummy) { int i; uint8_t indx; mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF); kmeminit(); uma_startup2(); if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX) kmem_zmax = KMEM_ZMAX; mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal), #ifdef INVARIANTS mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, #else NULL, NULL, NULL, NULL, #endif UMA_ALIGN_PTR, UMA_ZONE_MALLOC); for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) { int size = kmemzones[indx].kz_size; char *name = kmemzones[indx].kz_name; int subzone; for (subzone = 0; subzone < numzones; subzone++) { kmemzones[indx].kz_zone[subzone] = uma_zcreate(name, size, #ifdef INVARIANTS mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, #else NULL, NULL, NULL, NULL, #endif UMA_ALIGN_PTR, UMA_ZONE_MALLOC); } for (;i <= size; i+= KMEM_ZBASE) kmemsize[i >> KMEM_ZSHIFT] = indx; } } SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_SECOND, mallocinit, NULL); void malloc_init(void *data) { struct malloc_type_internal *mtip; struct malloc_type *mtp; KASSERT(vm_cnt.v_page_count != 0, ("malloc_register before vm_init")); mtp = data; if (mtp->ks_magic != M_MAGIC) panic("malloc_init: bad malloc type magic"); mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO); mtp->ks_handle = mtip; mtip->mti_zone = mtp_get_subzone(mtp->ks_shortdesc); mtx_lock(&malloc_mtx); mtp->ks_next = kmemstatistics; kmemstatistics = mtp; kmemcount++; mtx_unlock(&malloc_mtx); } void malloc_uninit(void *data) { struct malloc_type_internal *mtip; struct malloc_type_stats *mtsp; struct malloc_type *mtp, *temp; uma_slab_t slab; long temp_allocs, temp_bytes; int i; mtp = data; KASSERT(mtp->ks_magic == M_MAGIC, ("malloc_uninit: bad malloc type magic")); KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL")); mtx_lock(&malloc_mtx); mtip = mtp->ks_handle; mtp->ks_handle = NULL; if (mtp != kmemstatistics) { for (temp = kmemstatistics; temp != NULL; temp = temp->ks_next) { if (temp->ks_next == mtp) { temp->ks_next = mtp->ks_next; break; } } KASSERT(temp, ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc)); } else kmemstatistics = mtp->ks_next; kmemcount--; mtx_unlock(&malloc_mtx); /* * Look for memory leaks. */ temp_allocs = temp_bytes = 0; for (i = 0; i < MAXCPU; i++) { mtsp = &mtip->mti_stats[i]; temp_allocs += mtsp->mts_numallocs; temp_allocs -= mtsp->mts_numfrees; temp_bytes += mtsp->mts_memalloced; temp_bytes -= mtsp->mts_memfreed; } if (temp_allocs > 0 || temp_bytes > 0) { printf("Warning: memory type %s leaked memory on destroy " "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc, temp_allocs, temp_bytes); } slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK)); uma_zfree_arg(mt_zone, mtip, slab); } struct malloc_type * malloc_desc2type(const char *desc) { struct malloc_type *mtp; mtx_assert(&malloc_mtx, MA_OWNED); for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { if (strcmp(mtp->ks_shortdesc, desc) == 0) return (mtp); } return (NULL); } static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS) { struct malloc_type_stream_header mtsh; struct malloc_type_internal *mtip; struct malloc_type_header mth; struct malloc_type *mtp; int error, i; struct sbuf sbuf; error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); sbuf_new_for_sysctl(&sbuf, NULL, 128, req); sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL); mtx_lock(&malloc_mtx); /* * Insert stream header. */ bzero(&mtsh, sizeof(mtsh)); mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION; mtsh.mtsh_maxcpus = MAXCPU; mtsh.mtsh_count = kmemcount; (void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh)); /* * Insert alternating sequence of type headers and type statistics. */ for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { mtip = (struct malloc_type_internal *)mtp->ks_handle; /* * Insert type header. */ bzero(&mth, sizeof(mth)); strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME); (void)sbuf_bcat(&sbuf, &mth, sizeof(mth)); /* * Insert type statistics for each CPU. */ for (i = 0; i < MAXCPU; i++) { (void)sbuf_bcat(&sbuf, &mtip->mti_stats[i], sizeof(mtip->mti_stats[i])); } } mtx_unlock(&malloc_mtx); error = sbuf_finish(&sbuf); sbuf_delete(&sbuf); return (error); } SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT, 0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats", "Return malloc types"); SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0, "Count of kernel malloc types"); void malloc_type_list(malloc_type_list_func_t *func, void *arg) { struct malloc_type *mtp, **bufmtp; int count, i; size_t buflen; mtx_lock(&malloc_mtx); restart: mtx_assert(&malloc_mtx, MA_OWNED); count = kmemcount; mtx_unlock(&malloc_mtx); buflen = sizeof(struct malloc_type *) * count; bufmtp = malloc(buflen, M_TEMP, M_WAITOK); mtx_lock(&malloc_mtx); if (count < kmemcount) { free(bufmtp, M_TEMP); goto restart; } for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++) bufmtp[i] = mtp; mtx_unlock(&malloc_mtx); for (i = 0; i < count; i++) (func)(bufmtp[i], arg); free(bufmtp, M_TEMP); } #ifdef DDB DB_SHOW_COMMAND(malloc, db_show_malloc) { struct malloc_type_internal *mtip; struct malloc_type *mtp; uint64_t allocs, frees; uint64_t alloced, freed; int i; db_printf("%18s %12s %12s %12s\n", "Type", "InUse", "MemUse", "Requests"); for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { mtip = (struct malloc_type_internal *)mtp->ks_handle; allocs = 0; frees = 0; alloced = 0; freed = 0; for (i = 0; i < MAXCPU; i++) { allocs += mtip->mti_stats[i].mts_numallocs; frees += mtip->mti_stats[i].mts_numfrees; alloced += mtip->mti_stats[i].mts_memalloced; freed += mtip->mti_stats[i].mts_memfreed; } db_printf("%18s %12ju %12juK %12ju\n", mtp->ks_shortdesc, allocs - frees, (alloced - freed + 1023) / 1024, allocs); if (db_pager_quit) break; } } #if MALLOC_DEBUG_MAXZONES > 1 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches) { struct malloc_type_internal *mtip; struct malloc_type *mtp; u_int subzone; if (!have_addr) { db_printf("Usage: show multizone_matches \n"); return; } mtp = (void *)addr; if (mtp->ks_magic != M_MAGIC) { db_printf("Magic %lx does not match expected %x\n", mtp->ks_magic, M_MAGIC); return; } mtip = mtp->ks_handle; subzone = mtip->mti_zone; for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { mtip = mtp->ks_handle; if (mtip->mti_zone != subzone) continue; db_printf("%s\n", mtp->ks_shortdesc); if (db_pager_quit) break; } } #endif /* MALLOC_DEBUG_MAXZONES > 1 */ #endif /* DDB */ #ifdef MALLOC_PROFILE static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS) { struct sbuf sbuf; uint64_t count; uint64_t waste; uint64_t mem; int error; int rsize; int size; int i; waste = 0; mem = 0; error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); sbuf_new_for_sysctl(&sbuf, NULL, 128, req); sbuf_printf(&sbuf, "\n Size Requests Real Size\n"); for (i = 0; i < KMEM_ZSIZE; i++) { size = i << KMEM_ZSHIFT; rsize = kmemzones[kmemsize[i]].kz_size; count = (long long unsigned)krequests[i]; sbuf_printf(&sbuf, "%6d%28llu%11d\n", size, (unsigned long long)count, rsize); if ((rsize * count) > (size * count)) waste += (rsize * count) - (size * count); mem += (rsize * count); } sbuf_printf(&sbuf, "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n", (unsigned long long)mem, (unsigned long long)waste); error = sbuf_finish(&sbuf); sbuf_delete(&sbuf); return (error); } SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD, NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling"); #endif /* MALLOC_PROFILE */ Index: head/sys/kern/kern_mib.c =================================================================== --- head/sys/kern/kern_mib.c (revision 305831) +++ head/sys/kern/kern_mib.c (revision 305832) @@ -1,593 +1,593 @@ /*- * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Mike Karels at Berkeley Software Design, Inc. * * Quite extensively rewritten by Poul-Henning Kamp of the FreeBSD * project, to make these variables more userfriendly. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_sysctl.c 8.4 (Berkeley) 4/14/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #include "opt_posix.h" #include "opt_config.h" #include #include #include #include #include #include #include #include #include #include #include #include #include SYSCTL_ROOT_NODE(0, sysctl, CTLFLAG_RW, 0, "Sysctl internal magic"); SYSCTL_ROOT_NODE(CTL_KERN, kern, CTLFLAG_RW|CTLFLAG_CAPRD, 0, "High kernel, proc, limits &c"); SYSCTL_ROOT_NODE(CTL_VM, vm, CTLFLAG_RW, 0, "Virtual memory"); SYSCTL_ROOT_NODE(CTL_VFS, vfs, CTLFLAG_RW, 0, "File system"); SYSCTL_ROOT_NODE(CTL_NET, net, CTLFLAG_RW, 0, "Network, (see socket.h)"); SYSCTL_ROOT_NODE(CTL_DEBUG, debug, CTLFLAG_RW, 0, "Debugging"); SYSCTL_NODE(_debug, OID_AUTO, sizeof, CTLFLAG_RW, 0, "Sizeof various things"); SYSCTL_ROOT_NODE(CTL_HW, hw, CTLFLAG_RW, 0, "hardware"); SYSCTL_ROOT_NODE(CTL_MACHDEP, machdep, CTLFLAG_RW, 0, "machine dependent"); SYSCTL_ROOT_NODE(CTL_USER, user, CTLFLAG_RW, 0, "user-level"); SYSCTL_ROOT_NODE(CTL_P1003_1B, p1003_1b, CTLFLAG_RW, 0, "p1003_1b, (see p1003_1b.h)"); SYSCTL_ROOT_NODE(OID_AUTO, compat, CTLFLAG_RW, 0, "Compatibility code"); SYSCTL_ROOT_NODE(OID_AUTO, security, CTLFLAG_RW, 0, "Security"); #ifdef REGRESSION SYSCTL_ROOT_NODE(OID_AUTO, regression, CTLFLAG_RW, 0, "Regression test MIB"); #endif SYSCTL_STRING(_kern, OID_AUTO, ident, CTLFLAG_RD|CTLFLAG_MPSAFE, kern_ident, 0, "Kernel identifier"); SYSCTL_INT(_kern, KERN_OSREV, osrevision, CTLFLAG_RD|CTLFLAG_CAPRD, SYSCTL_NULL_INT_PTR, BSD, "Operating system revision"); SYSCTL_STRING(_kern, KERN_VERSION, version, CTLFLAG_RD|CTLFLAG_MPSAFE, version, 0, "Kernel version"); SYSCTL_STRING(_kern, OID_AUTO, compiler_version, CTLFLAG_RD|CTLFLAG_MPSAFE, compiler_version, 0, "Version of compiler used to compile kernel"); SYSCTL_STRING(_kern, KERN_OSTYPE, ostype, CTLFLAG_RD|CTLFLAG_MPSAFE| CTLFLAG_CAPRD, ostype, 0, "Operating system type"); SYSCTL_INT(_kern, KERN_MAXPROC, maxproc, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &maxproc, 0, "Maximum number of processes"); SYSCTL_INT(_kern, KERN_MAXPROCPERUID, maxprocperuid, CTLFLAG_RW, &maxprocperuid, 0, "Maximum processes allowed per userid"); SYSCTL_INT(_kern, OID_AUTO, maxusers, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &maxusers, 0, "Hint for kernel tuning"); SYSCTL_INT(_kern, KERN_ARGMAX, argmax, CTLFLAG_RD|CTLFLAG_CAPRD, SYSCTL_NULL_INT_PTR, ARG_MAX, "Maximum bytes of argument to execve(2)"); SYSCTL_INT(_kern, KERN_POSIX1, posix1version, CTLFLAG_RD|CTLFLAG_CAPRD, SYSCTL_NULL_INT_PTR, _POSIX_VERSION, "Version of POSIX attempting to comply to"); SYSCTL_INT(_kern, KERN_NGROUPS, ngroups, CTLFLAG_RDTUN | CTLFLAG_NOFETCH | CTLFLAG_CAPRD, &ngroups_max, 0, "Maximum number of supplemental groups a user can belong to"); SYSCTL_INT(_kern, KERN_JOB_CONTROL, job_control, CTLFLAG_RD|CTLFLAG_CAPRD, SYSCTL_NULL_INT_PTR, 1, "Whether job control is available"); #ifdef _POSIX_SAVED_IDS SYSCTL_INT(_kern, KERN_SAVED_IDS, saved_ids, CTLFLAG_RD|CTLFLAG_CAPRD, SYSCTL_NULL_INT_PTR, 1, "Whether saved set-group/user ID is available"); #else SYSCTL_INT(_kern, KERN_SAVED_IDS, saved_ids, CTLFLAG_RD|CTLFLAG_CAPRD, SYSCTL_NULL_INT_PTR, 0, "Whether saved set-group/user ID is available"); #endif char kernelname[MAXPATHLEN] = "/kernel"; /* XXX bloat */ SYSCTL_STRING(_kern, KERN_BOOTFILE, bootfile, CTLFLAG_RW, kernelname, sizeof kernelname, "Name of kernel file booted"); SYSCTL_INT(_hw, HW_NCPU, ncpu, CTLFLAG_RD|CTLFLAG_CAPRD, &mp_ncpus, 0, "Number of active CPUs"); SYSCTL_INT(_hw, HW_BYTEORDER, byteorder, CTLFLAG_RD|CTLFLAG_CAPRD, SYSCTL_NULL_INT_PTR, BYTE_ORDER, "System byte order"); SYSCTL_INT(_hw, HW_PAGESIZE, pagesize, CTLFLAG_RD|CTLFLAG_CAPRD, SYSCTL_NULL_INT_PTR, PAGE_SIZE, "System memory page size"); static int sysctl_kern_arnd(SYSCTL_HANDLER_ARGS) { char buf[256]; size_t len; /*- * This is one of the very few legitimate uses of read_random(9). * Use of arc4random(9) is not recommended as that will ignore * an unsafe (i.e. unseeded) random(4). * * If random(4) is not seeded, then this returns 0, so the * sysctl will return a zero-length buffer. */ len = read_random(buf, MIN(req->oldlen, sizeof(buf))); return (SYSCTL_OUT(req, buf, len)); } SYSCTL_PROC(_kern, KERN_ARND, arandom, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD, NULL, 0, sysctl_kern_arnd, "", "arc4rand"); static int sysctl_hw_physmem(SYSCTL_HANDLER_ARGS) { u_long val; val = ctob(physmem); return (sysctl_handle_long(oidp, &val, 0, req)); } SYSCTL_PROC(_hw, HW_PHYSMEM, physmem, CTLTYPE_ULONG | CTLFLAG_RD, 0, 0, sysctl_hw_physmem, "LU", ""); static int sysctl_hw_realmem(SYSCTL_HANDLER_ARGS) { u_long val; val = ctob(realmem); return (sysctl_handle_long(oidp, &val, 0, req)); } SYSCTL_PROC(_hw, HW_REALMEM, realmem, CTLTYPE_ULONG | CTLFLAG_RD, 0, 0, sysctl_hw_realmem, "LU", ""); static int sysctl_hw_usermem(SYSCTL_HANDLER_ARGS) { u_long val; val = ctob(physmem - vm_cnt.v_wire_count); return (sysctl_handle_long(oidp, &val, 0, req)); } SYSCTL_PROC(_hw, HW_USERMEM, usermem, CTLTYPE_ULONG | CTLFLAG_RD, 0, 0, sysctl_hw_usermem, "LU", ""); SYSCTL_LONG(_hw, OID_AUTO, availpages, CTLFLAG_RD, &physmem, 0, ""); u_long pagesizes[MAXPAGESIZES] = { PAGE_SIZE }; static int sysctl_hw_pagesizes(SYSCTL_HANDLER_ARGS) { int error; #ifdef SCTL_MASK32 int i; uint32_t pagesizes32[MAXPAGESIZES]; if (req->flags & SCTL_MASK32) { /* * Recreate the "pagesizes" array with 32-bit elements. Truncate * any page size greater than UINT32_MAX to zero. */ for (i = 0; i < MAXPAGESIZES; i++) pagesizes32[i] = (uint32_t)pagesizes[i]; error = SYSCTL_OUT(req, pagesizes32, sizeof(pagesizes32)); } else #endif error = SYSCTL_OUT(req, pagesizes, sizeof(pagesizes)); return (error); } SYSCTL_PROC(_hw, OID_AUTO, pagesizes, CTLTYPE_ULONG | CTLFLAG_RD, NULL, 0, sysctl_hw_pagesizes, "LU", "Supported page sizes"); #ifdef SCTL_MASK32 int adaptive_machine_arch = 1; SYSCTL_INT(_debug, OID_AUTO, adaptive_machine_arch, CTLFLAG_RW, &adaptive_machine_arch, 1, "Adapt reported machine architecture to the ABI of the binary"); #endif static int sysctl_hw_machine_arch(SYSCTL_HANDLER_ARGS) { int error; static const char machine_arch[] = MACHINE_ARCH; #ifdef SCTL_MASK32 static const char machine_arch32[] = MACHINE_ARCH32; if ((req->flags & SCTL_MASK32) != 0 && adaptive_machine_arch) error = SYSCTL_OUT(req, machine_arch32, sizeof(machine_arch32)); else #endif error = SYSCTL_OUT(req, machine_arch, sizeof(machine_arch)); return (error); } SYSCTL_PROC(_hw, HW_MACHINE_ARCH, machine_arch, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0, sysctl_hw_machine_arch, "A", "System architecture"); SYSCTL_STRING(_kern, OID_AUTO, supported_archs, CTLFLAG_RD | CTLFLAG_MPSAFE, #ifdef COMPAT_FREEBSD32 MACHINE_ARCH " " MACHINE_ARCH32, 0, "Supported architectures for binaries"); #else MACHINE_ARCH, 0, "Supported architectures for binaries"); #endif static int sysctl_hostname(SYSCTL_HANDLER_ARGS) { struct prison *pr, *cpr; size_t pr_offset; char tmpname[MAXHOSTNAMELEN]; int descend, error, len; /* * This function can set: hostname domainname hostuuid. * Keep that in mind when comments say "hostname". */ pr_offset = (size_t)arg1; len = arg2; KASSERT(len <= sizeof(tmpname), ("length %d too long for %s", len, __func__)); pr = req->td->td_ucred->cr_prison; if (!(pr->pr_allow & PR_ALLOW_SET_HOSTNAME) && req->newptr) return (EPERM); /* * Make a local copy of hostname to get/set so we don't have to hold * the jail mutex during the sysctl copyin/copyout activities. */ mtx_lock(&pr->pr_mtx); bcopy((char *)pr + pr_offset, tmpname, len); mtx_unlock(&pr->pr_mtx); error = sysctl_handle_string(oidp, tmpname, len, req); if (req->newptr != NULL && error == 0) { /* * Copy the locally set hostname to all jails that share * this host info. */ sx_slock(&allprison_lock); while (!(pr->pr_flags & PR_HOST)) pr = pr->pr_parent; mtx_lock(&pr->pr_mtx); bcopy(tmpname, (char *)pr + pr_offset, len); FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend) if (cpr->pr_flags & PR_HOST) descend = 0; else bcopy(tmpname, (char *)cpr + pr_offset, len); mtx_unlock(&pr->pr_mtx); sx_sunlock(&allprison_lock); } return (error); } SYSCTL_PROC(_kern, KERN_HOSTNAME, hostname, CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_MPSAFE, (void *)(offsetof(struct prison, pr_hostname)), MAXHOSTNAMELEN, sysctl_hostname, "A", "Hostname"); SYSCTL_PROC(_kern, KERN_NISDOMAINNAME, domainname, CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_MPSAFE, (void *)(offsetof(struct prison, pr_domainname)), MAXHOSTNAMELEN, sysctl_hostname, "A", "Name of the current YP/NIS domain"); SYSCTL_PROC(_kern, KERN_HOSTUUID, hostuuid, CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_MPSAFE, (void *)(offsetof(struct prison, pr_hostuuid)), HOSTUUIDLEN, sysctl_hostname, "A", "Host UUID"); static int regression_securelevel_nonmonotonic = 0; #ifdef REGRESSION SYSCTL_INT(_regression, OID_AUTO, securelevel_nonmonotonic, CTLFLAG_RW, ®ression_securelevel_nonmonotonic, 0, "securelevel may be lowered"); #endif static int sysctl_kern_securelvl(SYSCTL_HANDLER_ARGS) { struct prison *pr, *cpr; int descend, error, level; pr = req->td->td_ucred->cr_prison; /* * Reading the securelevel is easy, since the current jail's level * is known to be at least as secure as any higher levels. Perform * a lockless read since the securelevel is an integer. */ level = pr->pr_securelevel; error = sysctl_handle_int(oidp, &level, 0, req); if (error || !req->newptr) return (error); /* Permit update only if the new securelevel exceeds the old. */ sx_slock(&allprison_lock); mtx_lock(&pr->pr_mtx); if (!regression_securelevel_nonmonotonic && level < pr->pr_securelevel) { mtx_unlock(&pr->pr_mtx); sx_sunlock(&allprison_lock); return (EPERM); } pr->pr_securelevel = level; /* * Set all child jails to be at least this level, but do not lower * them (even if regression_securelevel_nonmonotonic). */ FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend) { if (cpr->pr_securelevel < level) cpr->pr_securelevel = level; } mtx_unlock(&pr->pr_mtx); sx_sunlock(&allprison_lock); return (error); } SYSCTL_PROC(_kern, KERN_SECURELVL, securelevel, CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, sysctl_kern_securelvl, "I", "Current secure level"); #ifdef INCLUDE_CONFIG_FILE /* Actual kernel configuration options. */ extern char kernconfstring[]; SYSCTL_STRING(_kern, OID_AUTO, conftxt, CTLFLAG_RD, kernconfstring, 0, "Kernel configuration file"); #endif static int sysctl_hostid(SYSCTL_HANDLER_ARGS) { struct prison *pr, *cpr; u_long tmpid; int descend, error; /* * Like sysctl_hostname, except it operates on a u_long * instead of a string, and is used only for hostid. */ pr = req->td->td_ucred->cr_prison; if (!(pr->pr_allow & PR_ALLOW_SET_HOSTNAME) && req->newptr) return (EPERM); tmpid = pr->pr_hostid; error = sysctl_handle_long(oidp, &tmpid, 0, req); if (req->newptr != NULL && error == 0) { sx_slock(&allprison_lock); while (!(pr->pr_flags & PR_HOST)) pr = pr->pr_parent; mtx_lock(&pr->pr_mtx); pr->pr_hostid = tmpid; FOREACH_PRISON_DESCENDANT_LOCKED(pr, cpr, descend) if (cpr->pr_flags & PR_HOST) descend = 0; else cpr->pr_hostid = tmpid; mtx_unlock(&pr->pr_mtx); sx_sunlock(&allprison_lock); } return (error); } SYSCTL_PROC(_kern, KERN_HOSTID, hostid, CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_MPSAFE | CTLFLAG_CAPRD, NULL, 0, sysctl_hostid, "LU", "Host ID"); /* * The osrelease string is copied from the global (osrelease in vers.c) into * prison0 by a sysinit and is inherited by child jails if not changed at jail * creation, so we always return the copy from the current prison data. */ static int sysctl_osrelease(SYSCTL_HANDLER_ARGS) { struct prison *pr; pr = req->td->td_ucred->cr_prison; return (SYSCTL_OUT(req, pr->pr_osrelease, strlen(pr->pr_osrelease) + 1)); } SYSCTL_PROC(_kern, KERN_OSRELEASE, osrelease, CTLTYPE_STRING | CTLFLAG_CAPRD | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_osrelease, "A", "Operating system release"); /* * The osreldate number is copied from the global (osreldate in vers.c) into * prison0 by a sysinit and is inherited by child jails if not changed at jail * creation, so we always return the value from the current prison data. */ static int sysctl_osreldate(SYSCTL_HANDLER_ARGS) { struct prison *pr; pr = req->td->td_ucred->cr_prison; return (SYSCTL_OUT(req, &pr->pr_osreldate, sizeof(pr->pr_osreldate))); } /* * NOTICE: The *userland* release date is available in * /usr/include/osreldate.h */ SYSCTL_PROC(_kern, KERN_OSRELDATE, osreldate, CTLTYPE_INT | CTLFLAG_CAPRD | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_osreldate, "I", "Kernel release date"); SYSCTL_NODE(_kern, OID_AUTO, features, CTLFLAG_RD, 0, "Kernel Features"); #ifdef COMPAT_FREEBSD4 FEATURE(compat_freebsd4, "Compatible with FreeBSD 4"); #endif #ifdef COMPAT_FREEBSD5 FEATURE(compat_freebsd5, "Compatible with FreeBSD 5"); #endif #ifdef COMPAT_FREEBSD6 FEATURE(compat_freebsd6, "Compatible with FreeBSD 6"); #endif #ifdef COMPAT_FREEBSD7 FEATURE(compat_freebsd7, "Compatible with FreeBSD 7"); #endif /* * This is really cheating. These actually live in the libc, something * which I'm not quite sure is a good idea anyway, but in order for * getnext and friends to actually work, we define dummies here. * * XXXRW: These probably should be CTLFLAG_CAPRD. */ SYSCTL_STRING(_user, USER_CS_PATH, cs_path, CTLFLAG_RD, "", 0, "PATH that finds all the standard utilities"); SYSCTL_INT(_user, USER_BC_BASE_MAX, bc_base_max, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, 0, "Max ibase/obase values in bc(1)"); SYSCTL_INT(_user, USER_BC_DIM_MAX, bc_dim_max, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, 0, "Max array size in bc(1)"); SYSCTL_INT(_user, USER_BC_SCALE_MAX, bc_scale_max, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, 0, "Max scale value in bc(1)"); SYSCTL_INT(_user, USER_BC_STRING_MAX, bc_string_max, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, 0, "Max string length in bc(1)"); SYSCTL_INT(_user, USER_COLL_WEIGHTS_MAX, coll_weights_max, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, 0, "Maximum number of weights assigned to an LC_COLLATE locale entry"); SYSCTL_INT(_user, USER_EXPR_NEST_MAX, expr_nest_max, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, 0, ""); SYSCTL_INT(_user, USER_LINE_MAX, line_max, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, 0, "Max length (bytes) of a text-processing utility's input line"); SYSCTL_INT(_user, USER_RE_DUP_MAX, re_dup_max, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, 0, "Maximum number of repeats of a regexp permitted"); SYSCTL_INT(_user, USER_POSIX2_VERSION, posix2_version, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, 0, "The version of POSIX 1003.2 with which the system attempts to comply"); SYSCTL_INT(_user, USER_POSIX2_C_BIND, posix2_c_bind, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, 0, "Whether C development supports the C bindings option"); SYSCTL_INT(_user, USER_POSIX2_C_DEV, posix2_c_dev, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, 0, "Whether system supports the C development utilities option"); SYSCTL_INT(_user, USER_POSIX2_CHAR_TERM, posix2_char_term, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, 0, ""); SYSCTL_INT(_user, USER_POSIX2_FORT_DEV, posix2_fort_dev, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, 0, "Whether system supports FORTRAN development utilities"); SYSCTL_INT(_user, USER_POSIX2_FORT_RUN, posix2_fort_run, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, 0, "Whether system supports FORTRAN runtime utilities"); SYSCTL_INT(_user, USER_POSIX2_LOCALEDEF, posix2_localedef, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, 0, "Whether system supports creation of locales"); SYSCTL_INT(_user, USER_POSIX2_SW_DEV, posix2_sw_dev, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, 0, "Whether system supports software development utilities"); SYSCTL_INT(_user, USER_POSIX2_UPE, posix2_upe, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, 0, "Whether system supports the user portability utilities"); SYSCTL_INT(_user, USER_STREAM_MAX, stream_max, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, 0, "Min Maximum number of streams a process may have open at one time"); SYSCTL_INT(_user, USER_TZNAME_MAX, tzname_max, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, 0, "Min Maximum number of types supported for timezone names"); #include SYSCTL_INT(_debug_sizeof, OID_AUTO, vnode, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, sizeof(struct vnode), "sizeof(struct vnode)"); SYSCTL_INT(_debug_sizeof, OID_AUTO, proc, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, sizeof(struct proc), "sizeof(struct proc)"); static int sysctl_kern_pid_max(SYSCTL_HANDLER_ARGS) { int error, pm; pm = pid_max; error = sysctl_handle_int(oidp, &pm, 0, req); if (error || !req->newptr) return (error); sx_xlock(&proctree_lock); sx_xlock(&allproc_lock); /* * Only permit the values less then PID_MAX. * As a safety measure, do not allow to limit the pid_max too much. */ if (pm < 300 || pm > PID_MAX) error = EINVAL; else pid_max = pm; sx_xunlock(&allproc_lock); sx_xunlock(&proctree_lock); return (error); } SYSCTL_PROC(_kern, OID_AUTO, pid_max, CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NOFETCH | CTLFLAG_MPSAFE, 0, 0, sysctl_kern_pid_max, "I", "Maximum allowed pid"); #include #include SYSCTL_INT(_debug_sizeof, OID_AUTO, bio, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, sizeof(struct bio), "sizeof(struct bio)"); SYSCTL_INT(_debug_sizeof, OID_AUTO, buf, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, sizeof(struct buf), "sizeof(struct buf)"); #include SYSCTL_INT(_debug_sizeof, OID_AUTO, kinfo_proc, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, sizeof(struct kinfo_proc), "sizeof(struct kinfo_proc)"); /* Used by kernel debuggers. */ const int pcb_size = sizeof(struct pcb); SYSCTL_INT(_debug_sizeof, OID_AUTO, pcb, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, sizeof(struct pcb), "sizeof(struct pcb)"); /* XXX compatibility, remove for 6.0 */ #include #include SYSCTL_INT(_kern, OID_AUTO, fallback_elf_brand, CTLFLAG_RW, &__elfN(fallback_brand), sizeof(__elfN(fallback_brand)), "compatibility for kern.fallback_elf_brand"); Index: head/sys/kern/kern_proc.c =================================================================== --- head/sys/kern/kern_proc.c (revision 305831) +++ head/sys/kern/kern_proc.c (revision 305832) @@ -1,3099 +1,3099 @@ /*- * Copyright (c) 1982, 1986, 1989, 1991, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_proc.c 8.7 (Berkeley) 2/14/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #include "opt_ddb.h" #include "opt_ktrace.h" #include "opt_kstack_pages.h" #include "opt_stack.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DDB #include #endif #include #include #include #include #include #include #include #include #ifdef COMPAT_FREEBSD32 #include #include #endif SDT_PROVIDER_DEFINE(proc); SDT_PROBE_DEFINE4(proc, , ctor, entry, "struct proc *", "int", "void *", "int"); SDT_PROBE_DEFINE4(proc, , ctor, return, "struct proc *", "int", "void *", "int"); SDT_PROBE_DEFINE4(proc, , dtor, entry, "struct proc *", "int", "void *", "struct thread *"); SDT_PROBE_DEFINE3(proc, , dtor, return, "struct proc *", "int", "void *"); SDT_PROBE_DEFINE3(proc, , init, entry, "struct proc *", "int", "int"); SDT_PROBE_DEFINE3(proc, , init, return, "struct proc *", "int", "int"); MALLOC_DEFINE(M_PGRP, "pgrp", "process group header"); MALLOC_DEFINE(M_SESSION, "session", "session header"); static MALLOC_DEFINE(M_PROC, "proc", "Proc structures"); MALLOC_DEFINE(M_SUBPROC, "subproc", "Proc sub-structures"); static void doenterpgrp(struct proc *, struct pgrp *); static void orphanpg(struct pgrp *pg); static void fill_kinfo_aggregate(struct proc *p, struct kinfo_proc *kp); static void fill_kinfo_proc_only(struct proc *p, struct kinfo_proc *kp); static void fill_kinfo_thread(struct thread *td, struct kinfo_proc *kp, int preferthread); static void pgadjustjobc(struct pgrp *pgrp, int entering); static void pgdelete(struct pgrp *); static int proc_ctor(void *mem, int size, void *arg, int flags); static void proc_dtor(void *mem, int size, void *arg); static int proc_init(void *mem, int size, int flags); static void proc_fini(void *mem, int size); static void pargs_free(struct pargs *pa); static struct proc *zpfind_locked(pid_t pid); /* * Other process lists */ struct pidhashhead *pidhashtbl; u_long pidhash; struct pgrphashhead *pgrphashtbl; u_long pgrphash; struct proclist allproc; struct proclist zombproc; struct sx allproc_lock; struct sx proctree_lock; struct mtx ppeers_lock; uma_zone_t proc_zone; /* * The offset of various fields in struct proc and struct thread. * These are used by kernel debuggers to enumerate kernel threads and * processes. */ const int proc_off_p_pid = offsetof(struct proc, p_pid); const int proc_off_p_comm = offsetof(struct proc, p_comm); const int proc_off_p_list = offsetof(struct proc, p_list); const int proc_off_p_threads = offsetof(struct proc, p_threads); const int thread_off_td_tid = offsetof(struct thread, td_tid); const int thread_off_td_name = offsetof(struct thread, td_name); const int thread_off_td_oncpu = offsetof(struct thread, td_oncpu); const int thread_off_td_pcb = offsetof(struct thread, td_pcb); const int thread_off_td_plist = offsetof(struct thread, td_plist); int kstack_pages = KSTACK_PAGES; SYSCTL_INT(_kern, OID_AUTO, kstack_pages, CTLFLAG_RD, &kstack_pages, 0, "Kernel stack size in pages"); static int vmmap_skip_res_cnt = 0; SYSCTL_INT(_kern, OID_AUTO, proc_vmmap_skip_resident_count, CTLFLAG_RW, &vmmap_skip_res_cnt, 0, "Skip calculation of the pages resident count in kern.proc.vmmap"); CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE); #ifdef COMPAT_FREEBSD32 CTASSERT(sizeof(struct kinfo_proc32) == KINFO_PROC32_SIZE); #endif /* * Initialize global process hashing structures. */ void procinit(void) { sx_init(&allproc_lock, "allproc"); sx_init(&proctree_lock, "proctree"); mtx_init(&ppeers_lock, "p_peers", NULL, MTX_DEF); LIST_INIT(&allproc); LIST_INIT(&zombproc); pidhashtbl = hashinit(maxproc / 4, M_PROC, &pidhash); pgrphashtbl = hashinit(maxproc / 4, M_PROC, &pgrphash); proc_zone = uma_zcreate("PROC", sched_sizeof_proc(), proc_ctor, proc_dtor, proc_init, proc_fini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); uihashinit(); } /* * Prepare a proc for use. */ static int proc_ctor(void *mem, int size, void *arg, int flags) { struct proc *p; p = (struct proc *)mem; SDT_PROBE4(proc, , ctor , entry, p, size, arg, flags); EVENTHANDLER_INVOKE(process_ctor, p); SDT_PROBE4(proc, , ctor , return, p, size, arg, flags); return (0); } /* * Reclaim a proc after use. */ static void proc_dtor(void *mem, int size, void *arg) { struct proc *p; struct thread *td; /* INVARIANTS checks go here */ p = (struct proc *)mem; td = FIRST_THREAD_IN_PROC(p); SDT_PROBE4(proc, , dtor, entry, p, size, arg, td); if (td != NULL) { #ifdef INVARIANTS KASSERT((p->p_numthreads == 1), ("bad number of threads in exiting process")); KASSERT(STAILQ_EMPTY(&p->p_ktr), ("proc_dtor: non-empty p_ktr")); #endif /* Free all OSD associated to this thread. */ osd_thread_exit(td); } EVENTHANDLER_INVOKE(process_dtor, p); if (p->p_ksi != NULL) KASSERT(! KSI_ONQ(p->p_ksi), ("SIGCHLD queue")); SDT_PROBE3(proc, , dtor, return, p, size, arg); } /* * Initialize type-stable parts of a proc (when newly created). */ static int proc_init(void *mem, int size, int flags) { struct proc *p; p = (struct proc *)mem; SDT_PROBE3(proc, , init, entry, p, size, flags); mtx_init(&p->p_mtx, "process lock", NULL, MTX_DEF | MTX_DUPOK | MTX_NEW); mtx_init(&p->p_slock, "process slock", NULL, MTX_SPIN | MTX_NEW); mtx_init(&p->p_statmtx, "pstatl", NULL, MTX_SPIN | MTX_NEW); mtx_init(&p->p_itimmtx, "pitiml", NULL, MTX_SPIN | MTX_NEW); mtx_init(&p->p_profmtx, "pprofl", NULL, MTX_SPIN | MTX_NEW); cv_init(&p->p_pwait, "ppwait"); cv_init(&p->p_dbgwait, "dbgwait"); TAILQ_INIT(&p->p_threads); /* all threads in proc */ EVENTHANDLER_INVOKE(process_init, p); p->p_stats = pstats_alloc(); p->p_pgrp = NULL; SDT_PROBE3(proc, , init, return, p, size, flags); return (0); } /* * UMA should ensure that this function is never called. * Freeing a proc structure would violate type stability. */ static void proc_fini(void *mem, int size) { #ifdef notnow struct proc *p; p = (struct proc *)mem; EVENTHANDLER_INVOKE(process_fini, p); pstats_free(p->p_stats); thread_free(FIRST_THREAD_IN_PROC(p)); mtx_destroy(&p->p_mtx); if (p->p_ksi != NULL) ksiginfo_free(p->p_ksi); #else panic("proc reclaimed"); #endif } /* * Is p an inferior of the current process? */ int inferior(struct proc *p) { sx_assert(&proctree_lock, SX_LOCKED); PROC_LOCK_ASSERT(p, MA_OWNED); for (; p != curproc; p = proc_realparent(p)) { if (p->p_pid == 0) return (0); } return (1); } struct proc * pfind_locked(pid_t pid) { struct proc *p; sx_assert(&allproc_lock, SX_LOCKED); LIST_FOREACH(p, PIDHASH(pid), p_hash) { if (p->p_pid == pid) { PROC_LOCK(p); if (p->p_state == PRS_NEW) { PROC_UNLOCK(p); p = NULL; } break; } } return (p); } /* * Locate a process by number; return only "live" processes -- i.e., neither * zombies nor newly born but incompletely initialized processes. By not * returning processes in the PRS_NEW state, we allow callers to avoid * testing for that condition to avoid dereferencing p_ucred, et al. */ struct proc * pfind(pid_t pid) { struct proc *p; sx_slock(&allproc_lock); p = pfind_locked(pid); sx_sunlock(&allproc_lock); return (p); } static struct proc * pfind_tid_locked(pid_t tid) { struct proc *p; struct thread *td; sx_assert(&allproc_lock, SX_LOCKED); FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); if (p->p_state == PRS_NEW) { PROC_UNLOCK(p); continue; } FOREACH_THREAD_IN_PROC(p, td) { if (td->td_tid == tid) goto found; } PROC_UNLOCK(p); } found: return (p); } /* * Locate a process group by number. * The caller must hold proctree_lock. */ struct pgrp * pgfind(pid_t pgid) { struct pgrp *pgrp; sx_assert(&proctree_lock, SX_LOCKED); LIST_FOREACH(pgrp, PGRPHASH(pgid), pg_hash) { if (pgrp->pg_id == pgid) { PGRP_LOCK(pgrp); return (pgrp); } } return (NULL); } /* * Locate process and do additional manipulations, depending on flags. */ int pget(pid_t pid, int flags, struct proc **pp) { struct proc *p; int error; sx_slock(&allproc_lock); if (pid <= PID_MAX) { p = pfind_locked(pid); if (p == NULL && (flags & PGET_NOTWEXIT) == 0) p = zpfind_locked(pid); } else if ((flags & PGET_NOTID) == 0) { p = pfind_tid_locked(pid); } else { p = NULL; } sx_sunlock(&allproc_lock); if (p == NULL) return (ESRCH); if ((flags & PGET_CANSEE) != 0) { error = p_cansee(curthread, p); if (error != 0) goto errout; } if ((flags & PGET_CANDEBUG) != 0) { error = p_candebug(curthread, p); if (error != 0) goto errout; } if ((flags & PGET_ISCURRENT) != 0 && curproc != p) { error = EPERM; goto errout; } if ((flags & PGET_NOTWEXIT) != 0 && (p->p_flag & P_WEXIT) != 0) { error = ESRCH; goto errout; } if ((flags & PGET_NOTINEXEC) != 0 && (p->p_flag & P_INEXEC) != 0) { /* * XXXRW: Not clear ESRCH is the right error during proc * execve(). */ error = ESRCH; goto errout; } if ((flags & PGET_HOLD) != 0) { _PHOLD(p); PROC_UNLOCK(p); } *pp = p; return (0); errout: PROC_UNLOCK(p); return (error); } /* * Create a new process group. * pgid must be equal to the pid of p. * Begin a new session if required. */ int enterpgrp(struct proc *p, pid_t pgid, struct pgrp *pgrp, struct session *sess) { sx_assert(&proctree_lock, SX_XLOCKED); KASSERT(pgrp != NULL, ("enterpgrp: pgrp == NULL")); KASSERT(p->p_pid == pgid, ("enterpgrp: new pgrp and pid != pgid")); KASSERT(pgfind(pgid) == NULL, ("enterpgrp: pgrp with pgid exists")); KASSERT(!SESS_LEADER(p), ("enterpgrp: session leader attempted setpgrp")); mtx_init(&pgrp->pg_mtx, "process group", NULL, MTX_DEF | MTX_DUPOK); if (sess != NULL) { /* * new session */ mtx_init(&sess->s_mtx, "session", NULL, MTX_DEF); PROC_LOCK(p); p->p_flag &= ~P_CONTROLT; PROC_UNLOCK(p); PGRP_LOCK(pgrp); sess->s_leader = p; sess->s_sid = p->p_pid; refcount_init(&sess->s_count, 1); sess->s_ttyvp = NULL; sess->s_ttydp = NULL; sess->s_ttyp = NULL; bcopy(p->p_session->s_login, sess->s_login, sizeof(sess->s_login)); pgrp->pg_session = sess; KASSERT(p == curproc, ("enterpgrp: mksession and p != curproc")); } else { pgrp->pg_session = p->p_session; sess_hold(pgrp->pg_session); PGRP_LOCK(pgrp); } pgrp->pg_id = pgid; LIST_INIT(&pgrp->pg_members); /* * As we have an exclusive lock of proctree_lock, * this should not deadlock. */ LIST_INSERT_HEAD(PGRPHASH(pgid), pgrp, pg_hash); pgrp->pg_jobc = 0; SLIST_INIT(&pgrp->pg_sigiolst); PGRP_UNLOCK(pgrp); doenterpgrp(p, pgrp); return (0); } /* * Move p to an existing process group */ int enterthispgrp(struct proc *p, struct pgrp *pgrp) { sx_assert(&proctree_lock, SX_XLOCKED); PROC_LOCK_ASSERT(p, MA_NOTOWNED); PGRP_LOCK_ASSERT(pgrp, MA_NOTOWNED); PGRP_LOCK_ASSERT(p->p_pgrp, MA_NOTOWNED); SESS_LOCK_ASSERT(p->p_session, MA_NOTOWNED); KASSERT(pgrp->pg_session == p->p_session, ("%s: pgrp's session %p, p->p_session %p.\n", __func__, pgrp->pg_session, p->p_session)); KASSERT(pgrp != p->p_pgrp, ("%s: p belongs to pgrp.", __func__)); doenterpgrp(p, pgrp); return (0); } /* * Move p to a process group */ static void doenterpgrp(struct proc *p, struct pgrp *pgrp) { struct pgrp *savepgrp; sx_assert(&proctree_lock, SX_XLOCKED); PROC_LOCK_ASSERT(p, MA_NOTOWNED); PGRP_LOCK_ASSERT(pgrp, MA_NOTOWNED); PGRP_LOCK_ASSERT(p->p_pgrp, MA_NOTOWNED); SESS_LOCK_ASSERT(p->p_session, MA_NOTOWNED); savepgrp = p->p_pgrp; /* * Adjust eligibility of affected pgrps to participate in job control. * Increment eligibility counts before decrementing, otherwise we * could reach 0 spuriously during the first call. */ fixjobc(p, pgrp, 1); fixjobc(p, p->p_pgrp, 0); PGRP_LOCK(pgrp); PGRP_LOCK(savepgrp); PROC_LOCK(p); LIST_REMOVE(p, p_pglist); p->p_pgrp = pgrp; PROC_UNLOCK(p); LIST_INSERT_HEAD(&pgrp->pg_members, p, p_pglist); PGRP_UNLOCK(savepgrp); PGRP_UNLOCK(pgrp); if (LIST_EMPTY(&savepgrp->pg_members)) pgdelete(savepgrp); } /* * remove process from process group */ int leavepgrp(struct proc *p) { struct pgrp *savepgrp; sx_assert(&proctree_lock, SX_XLOCKED); savepgrp = p->p_pgrp; PGRP_LOCK(savepgrp); PROC_LOCK(p); LIST_REMOVE(p, p_pglist); p->p_pgrp = NULL; PROC_UNLOCK(p); PGRP_UNLOCK(savepgrp); if (LIST_EMPTY(&savepgrp->pg_members)) pgdelete(savepgrp); return (0); } /* * delete a process group */ static void pgdelete(struct pgrp *pgrp) { struct session *savesess; struct tty *tp; sx_assert(&proctree_lock, SX_XLOCKED); PGRP_LOCK_ASSERT(pgrp, MA_NOTOWNED); SESS_LOCK_ASSERT(pgrp->pg_session, MA_NOTOWNED); /* * Reset any sigio structures pointing to us as a result of * F_SETOWN with our pgid. */ funsetownlst(&pgrp->pg_sigiolst); PGRP_LOCK(pgrp); tp = pgrp->pg_session->s_ttyp; LIST_REMOVE(pgrp, pg_hash); savesess = pgrp->pg_session; PGRP_UNLOCK(pgrp); /* Remove the reference to the pgrp before deallocating it. */ if (tp != NULL) { tty_lock(tp); tty_rel_pgrp(tp, pgrp); } mtx_destroy(&pgrp->pg_mtx); free(pgrp, M_PGRP); sess_release(savesess); } static void pgadjustjobc(struct pgrp *pgrp, int entering) { PGRP_LOCK(pgrp); if (entering) pgrp->pg_jobc++; else { --pgrp->pg_jobc; if (pgrp->pg_jobc == 0) orphanpg(pgrp); } PGRP_UNLOCK(pgrp); } /* * Adjust pgrp jobc counters when specified process changes process group. * We count the number of processes in each process group that "qualify" * the group for terminal job control (those with a parent in a different * process group of the same session). If that count reaches zero, the * process group becomes orphaned. Check both the specified process' * process group and that of its children. * entering == 0 => p is leaving specified group. * entering == 1 => p is entering specified group. */ void fixjobc(struct proc *p, struct pgrp *pgrp, int entering) { struct pgrp *hispgrp; struct session *mysession; struct proc *q; sx_assert(&proctree_lock, SX_LOCKED); PROC_LOCK_ASSERT(p, MA_NOTOWNED); PGRP_LOCK_ASSERT(pgrp, MA_NOTOWNED); SESS_LOCK_ASSERT(pgrp->pg_session, MA_NOTOWNED); /* * Check p's parent to see whether p qualifies its own process * group; if so, adjust count for p's process group. */ mysession = pgrp->pg_session; if ((hispgrp = p->p_pptr->p_pgrp) != pgrp && hispgrp->pg_session == mysession) pgadjustjobc(pgrp, entering); /* * Check this process' children to see whether they qualify * their process groups; if so, adjust counts for children's * process groups. */ LIST_FOREACH(q, &p->p_children, p_sibling) { hispgrp = q->p_pgrp; if (hispgrp == pgrp || hispgrp->pg_session != mysession) continue; if (q->p_state == PRS_ZOMBIE) continue; pgadjustjobc(hispgrp, entering); } } void killjobc(void) { struct session *sp; struct tty *tp; struct proc *p; struct vnode *ttyvp; p = curproc; MPASS(p->p_flag & P_WEXIT); /* * Do a quick check to see if there is anything to do with the * proctree_lock held. pgrp and LIST_EMPTY checks are for fixjobc(). */ PROC_LOCK(p); if (!SESS_LEADER(p) && (p->p_pgrp == p->p_pptr->p_pgrp) && LIST_EMPTY(&p->p_children)) { PROC_UNLOCK(p); return; } PROC_UNLOCK(p); sx_xlock(&proctree_lock); if (SESS_LEADER(p)) { sp = p->p_session; /* * s_ttyp is not zero'd; we use this to indicate that * the session once had a controlling terminal. (for * logging and informational purposes) */ SESS_LOCK(sp); ttyvp = sp->s_ttyvp; tp = sp->s_ttyp; sp->s_ttyvp = NULL; sp->s_ttydp = NULL; sp->s_leader = NULL; SESS_UNLOCK(sp); /* * Signal foreground pgrp and revoke access to * controlling terminal if it has not been revoked * already. * * Because the TTY may have been revoked in the mean * time and could already have a new session associated * with it, make sure we don't send a SIGHUP to a * foreground process group that does not belong to this * session. */ if (tp != NULL) { tty_lock(tp); if (tp->t_session == sp) tty_signal_pgrp(tp, SIGHUP); tty_unlock(tp); } if (ttyvp != NULL) { sx_xunlock(&proctree_lock); if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) { VOP_REVOKE(ttyvp, REVOKEALL); VOP_UNLOCK(ttyvp, 0); } vrele(ttyvp); sx_xlock(&proctree_lock); } } fixjobc(p, p->p_pgrp, 0); sx_xunlock(&proctree_lock); } /* * A process group has become orphaned; * if there are any stopped processes in the group, * hang-up all process in that group. */ static void orphanpg(struct pgrp *pg) { struct proc *p; PGRP_LOCK_ASSERT(pg, MA_OWNED); LIST_FOREACH(p, &pg->pg_members, p_pglist) { PROC_LOCK(p); if (P_SHOULDSTOP(p) == P_STOPPED_SIG) { PROC_UNLOCK(p); LIST_FOREACH(p, &pg->pg_members, p_pglist) { PROC_LOCK(p); kern_psignal(p, SIGHUP); kern_psignal(p, SIGCONT); PROC_UNLOCK(p); } return; } PROC_UNLOCK(p); } } void sess_hold(struct session *s) { refcount_acquire(&s->s_count); } void sess_release(struct session *s) { if (refcount_release(&s->s_count)) { if (s->s_ttyp != NULL) { tty_lock(s->s_ttyp); tty_rel_sess(s->s_ttyp, s); } mtx_destroy(&s->s_mtx); free(s, M_SESSION); } } #ifdef DDB DB_SHOW_COMMAND(pgrpdump, pgrpdump) { struct pgrp *pgrp; struct proc *p; int i; for (i = 0; i <= pgrphash; i++) { if (!LIST_EMPTY(&pgrphashtbl[i])) { printf("\tindx %d\n", i); LIST_FOREACH(pgrp, &pgrphashtbl[i], pg_hash) { printf( "\tpgrp %p, pgid %ld, sess %p, sesscnt %d, mem %p\n", (void *)pgrp, (long)pgrp->pg_id, (void *)pgrp->pg_session, pgrp->pg_session->s_count, (void *)LIST_FIRST(&pgrp->pg_members)); LIST_FOREACH(p, &pgrp->pg_members, p_pglist) { printf("\t\tpid %ld addr %p pgrp %p\n", (long)p->p_pid, (void *)p, (void *)p->p_pgrp); } } } } } #endif /* DDB */ /* * Calculate the kinfo_proc members which contain process-wide * informations. * Must be called with the target process locked. */ static void fill_kinfo_aggregate(struct proc *p, struct kinfo_proc *kp) { struct thread *td; PROC_LOCK_ASSERT(p, MA_OWNED); kp->ki_estcpu = 0; kp->ki_pctcpu = 0; FOREACH_THREAD_IN_PROC(p, td) { thread_lock(td); kp->ki_pctcpu += sched_pctcpu(td); kp->ki_estcpu += sched_estcpu(td); thread_unlock(td); } } /* * Clear kinfo_proc and fill in any information that is common * to all threads in the process. * Must be called with the target process locked. */ static void fill_kinfo_proc_only(struct proc *p, struct kinfo_proc *kp) { struct thread *td0; struct tty *tp; struct session *sp; struct ucred *cred; struct sigacts *ps; struct timeval boottime; /* For proc_realparent. */ sx_assert(&proctree_lock, SX_LOCKED); PROC_LOCK_ASSERT(p, MA_OWNED); bzero(kp, sizeof(*kp)); kp->ki_structsize = sizeof(*kp); kp->ki_paddr = p; kp->ki_addr =/* p->p_addr; */0; /* XXX */ kp->ki_args = p->p_args; kp->ki_textvp = p->p_textvp; #ifdef KTRACE kp->ki_tracep = p->p_tracevp; kp->ki_traceflag = p->p_traceflag; #endif kp->ki_fd = p->p_fd; kp->ki_vmspace = p->p_vmspace; kp->ki_flag = p->p_flag; kp->ki_flag2 = p->p_flag2; cred = p->p_ucred; if (cred) { kp->ki_uid = cred->cr_uid; kp->ki_ruid = cred->cr_ruid; kp->ki_svuid = cred->cr_svuid; kp->ki_cr_flags = 0; if (cred->cr_flags & CRED_FLAG_CAPMODE) kp->ki_cr_flags |= KI_CRF_CAPABILITY_MODE; /* XXX bde doesn't like KI_NGROUPS */ if (cred->cr_ngroups > KI_NGROUPS) { kp->ki_ngroups = KI_NGROUPS; kp->ki_cr_flags |= KI_CRF_GRP_OVERFLOW; } else kp->ki_ngroups = cred->cr_ngroups; bcopy(cred->cr_groups, kp->ki_groups, kp->ki_ngroups * sizeof(gid_t)); kp->ki_rgid = cred->cr_rgid; kp->ki_svgid = cred->cr_svgid; /* If jailed(cred), emulate the old P_JAILED flag. */ if (jailed(cred)) { kp->ki_flag |= P_JAILED; /* If inside the jail, use 0 as a jail ID. */ if (cred->cr_prison != curthread->td_ucred->cr_prison) kp->ki_jid = cred->cr_prison->pr_id; } strlcpy(kp->ki_loginclass, cred->cr_loginclass->lc_name, sizeof(kp->ki_loginclass)); } ps = p->p_sigacts; if (ps) { mtx_lock(&ps->ps_mtx); kp->ki_sigignore = ps->ps_sigignore; kp->ki_sigcatch = ps->ps_sigcatch; mtx_unlock(&ps->ps_mtx); } if (p->p_state != PRS_NEW && p->p_state != PRS_ZOMBIE && p->p_vmspace != NULL) { struct vmspace *vm = p->p_vmspace; kp->ki_size = vm->vm_map.size; kp->ki_rssize = vmspace_resident_count(vm); /*XXX*/ FOREACH_THREAD_IN_PROC(p, td0) { if (!TD_IS_SWAPPED(td0)) kp->ki_rssize += td0->td_kstack_pages; } kp->ki_swrss = vm->vm_swrss; kp->ki_tsize = vm->vm_tsize; kp->ki_dsize = vm->vm_dsize; kp->ki_ssize = vm->vm_ssize; } else if (p->p_state == PRS_ZOMBIE) kp->ki_stat = SZOMB; if (kp->ki_flag & P_INMEM) kp->ki_sflag = PS_INMEM; else kp->ki_sflag = 0; /* Calculate legacy swtime as seconds since 'swtick'. */ kp->ki_swtime = (ticks - p->p_swtick) / hz; kp->ki_pid = p->p_pid; kp->ki_nice = p->p_nice; kp->ki_fibnum = p->p_fibnum; kp->ki_start = p->p_stats->p_start; getboottime(&boottime); timevaladd(&kp->ki_start, &boottime); PROC_STATLOCK(p); rufetch(p, &kp->ki_rusage); kp->ki_runtime = cputick2usec(p->p_rux.rux_runtime); calcru(p, &kp->ki_rusage.ru_utime, &kp->ki_rusage.ru_stime); PROC_STATUNLOCK(p); calccru(p, &kp->ki_childutime, &kp->ki_childstime); /* Some callers want child times in a single value. */ kp->ki_childtime = kp->ki_childstime; timevaladd(&kp->ki_childtime, &kp->ki_childutime); FOREACH_THREAD_IN_PROC(p, td0) kp->ki_cow += td0->td_cow; tp = NULL; if (p->p_pgrp) { kp->ki_pgid = p->p_pgrp->pg_id; kp->ki_jobc = p->p_pgrp->pg_jobc; sp = p->p_pgrp->pg_session; if (sp != NULL) { kp->ki_sid = sp->s_sid; SESS_LOCK(sp); strlcpy(kp->ki_login, sp->s_login, sizeof(kp->ki_login)); if (sp->s_ttyvp) kp->ki_kiflag |= KI_CTTY; if (SESS_LEADER(p)) kp->ki_kiflag |= KI_SLEADER; /* XXX proctree_lock */ tp = sp->s_ttyp; SESS_UNLOCK(sp); } } if ((p->p_flag & P_CONTROLT) && tp != NULL) { kp->ki_tdev = tty_udev(tp); kp->ki_tpgid = tp->t_pgrp ? tp->t_pgrp->pg_id : NO_PID; if (tp->t_session) kp->ki_tsid = tp->t_session->s_sid; } else kp->ki_tdev = NODEV; if (p->p_comm[0] != '\0') strlcpy(kp->ki_comm, p->p_comm, sizeof(kp->ki_comm)); if (p->p_sysent && p->p_sysent->sv_name != NULL && p->p_sysent->sv_name[0] != '\0') strlcpy(kp->ki_emul, p->p_sysent->sv_name, sizeof(kp->ki_emul)); kp->ki_siglist = p->p_siglist; kp->ki_xstat = KW_EXITCODE(p->p_xexit, p->p_xsig); kp->ki_acflag = p->p_acflag; kp->ki_lock = p->p_lock; if (p->p_pptr) { kp->ki_ppid = proc_realparent(p)->p_pid; if (p->p_flag & P_TRACED) kp->ki_tracer = p->p_pptr->p_pid; } } /* * Fill in information that is thread specific. Must be called with * target process locked. If 'preferthread' is set, overwrite certain * process-related fields that are maintained for both threads and * processes. */ static void fill_kinfo_thread(struct thread *td, struct kinfo_proc *kp, int preferthread) { struct proc *p; p = td->td_proc; kp->ki_tdaddr = td; PROC_LOCK_ASSERT(p, MA_OWNED); if (preferthread) PROC_STATLOCK(p); thread_lock(td); if (td->td_wmesg != NULL) strlcpy(kp->ki_wmesg, td->td_wmesg, sizeof(kp->ki_wmesg)); else bzero(kp->ki_wmesg, sizeof(kp->ki_wmesg)); strlcpy(kp->ki_tdname, td->td_name, sizeof(kp->ki_tdname)); if (TD_ON_LOCK(td)) { kp->ki_kiflag |= KI_LOCKBLOCK; strlcpy(kp->ki_lockname, td->td_lockname, sizeof(kp->ki_lockname)); } else { kp->ki_kiflag &= ~KI_LOCKBLOCK; bzero(kp->ki_lockname, sizeof(kp->ki_lockname)); } if (p->p_state == PRS_NORMAL) { /* approximate. */ if (TD_ON_RUNQ(td) || TD_CAN_RUN(td) || TD_IS_RUNNING(td)) { kp->ki_stat = SRUN; } else if (P_SHOULDSTOP(p)) { kp->ki_stat = SSTOP; } else if (TD_IS_SLEEPING(td)) { kp->ki_stat = SSLEEP; } else if (TD_ON_LOCK(td)) { kp->ki_stat = SLOCK; } else { kp->ki_stat = SWAIT; } } else if (p->p_state == PRS_ZOMBIE) { kp->ki_stat = SZOMB; } else { kp->ki_stat = SIDL; } /* Things in the thread */ kp->ki_wchan = td->td_wchan; kp->ki_pri.pri_level = td->td_priority; kp->ki_pri.pri_native = td->td_base_pri; /* * Note: legacy fields; clamp at the old NOCPU value and/or * the maximum u_char CPU value. */ if (td->td_lastcpu == NOCPU) kp->ki_lastcpu_old = NOCPU_OLD; else if (td->td_lastcpu > MAXCPU_OLD) kp->ki_lastcpu_old = MAXCPU_OLD; else kp->ki_lastcpu_old = td->td_lastcpu; if (td->td_oncpu == NOCPU) kp->ki_oncpu_old = NOCPU_OLD; else if (td->td_oncpu > MAXCPU_OLD) kp->ki_oncpu_old = MAXCPU_OLD; else kp->ki_oncpu_old = td->td_oncpu; kp->ki_lastcpu = td->td_lastcpu; kp->ki_oncpu = td->td_oncpu; kp->ki_tdflags = td->td_flags; kp->ki_tid = td->td_tid; kp->ki_numthreads = p->p_numthreads; kp->ki_pcb = td->td_pcb; kp->ki_kstack = (void *)td->td_kstack; kp->ki_slptime = (ticks - td->td_slptick) / hz; kp->ki_pri.pri_class = td->td_pri_class; kp->ki_pri.pri_user = td->td_user_pri; if (preferthread) { rufetchtd(td, &kp->ki_rusage); kp->ki_runtime = cputick2usec(td->td_rux.rux_runtime); kp->ki_pctcpu = sched_pctcpu(td); kp->ki_estcpu = sched_estcpu(td); kp->ki_cow = td->td_cow; } /* We can't get this anymore but ps etc never used it anyway. */ kp->ki_rqindex = 0; if (preferthread) kp->ki_siglist = td->td_siglist; kp->ki_sigmask = td->td_sigmask; thread_unlock(td); if (preferthread) PROC_STATUNLOCK(p); } /* * Fill in a kinfo_proc structure for the specified process. * Must be called with the target process locked. */ void fill_kinfo_proc(struct proc *p, struct kinfo_proc *kp) { MPASS(FIRST_THREAD_IN_PROC(p) != NULL); fill_kinfo_proc_only(p, kp); fill_kinfo_thread(FIRST_THREAD_IN_PROC(p), kp, 0); fill_kinfo_aggregate(p, kp); } struct pstats * pstats_alloc(void) { return (malloc(sizeof(struct pstats), M_SUBPROC, M_ZERO|M_WAITOK)); } /* * Copy parts of p_stats; zero the rest of p_stats (statistics). */ void pstats_fork(struct pstats *src, struct pstats *dst) { bzero(&dst->pstat_startzero, __rangeof(struct pstats, pstat_startzero, pstat_endzero)); bcopy(&src->pstat_startcopy, &dst->pstat_startcopy, __rangeof(struct pstats, pstat_startcopy, pstat_endcopy)); } void pstats_free(struct pstats *ps) { free(ps, M_SUBPROC); } static struct proc * zpfind_locked(pid_t pid) { struct proc *p; sx_assert(&allproc_lock, SX_LOCKED); LIST_FOREACH(p, &zombproc, p_list) { if (p->p_pid == pid) { PROC_LOCK(p); break; } } return (p); } /* * Locate a zombie process by number */ struct proc * zpfind(pid_t pid) { struct proc *p; sx_slock(&allproc_lock); p = zpfind_locked(pid); sx_sunlock(&allproc_lock); return (p); } #ifdef COMPAT_FREEBSD32 /* * This function is typically used to copy out the kernel address, so * it can be replaced by assignment of zero. */ static inline uint32_t ptr32_trim(void *ptr) { uintptr_t uptr; uptr = (uintptr_t)ptr; return ((uptr > UINT_MAX) ? 0 : uptr); } #define PTRTRIM_CP(src,dst,fld) \ do { (dst).fld = ptr32_trim((src).fld); } while (0) static void freebsd32_kinfo_proc_out(const struct kinfo_proc *ki, struct kinfo_proc32 *ki32) { int i; bzero(ki32, sizeof(struct kinfo_proc32)); ki32->ki_structsize = sizeof(struct kinfo_proc32); CP(*ki, *ki32, ki_layout); PTRTRIM_CP(*ki, *ki32, ki_args); PTRTRIM_CP(*ki, *ki32, ki_paddr); PTRTRIM_CP(*ki, *ki32, ki_addr); PTRTRIM_CP(*ki, *ki32, ki_tracep); PTRTRIM_CP(*ki, *ki32, ki_textvp); PTRTRIM_CP(*ki, *ki32, ki_fd); PTRTRIM_CP(*ki, *ki32, ki_vmspace); PTRTRIM_CP(*ki, *ki32, ki_wchan); CP(*ki, *ki32, ki_pid); CP(*ki, *ki32, ki_ppid); CP(*ki, *ki32, ki_pgid); CP(*ki, *ki32, ki_tpgid); CP(*ki, *ki32, ki_sid); CP(*ki, *ki32, ki_tsid); CP(*ki, *ki32, ki_jobc); CP(*ki, *ki32, ki_tdev); CP(*ki, *ki32, ki_siglist); CP(*ki, *ki32, ki_sigmask); CP(*ki, *ki32, ki_sigignore); CP(*ki, *ki32, ki_sigcatch); CP(*ki, *ki32, ki_uid); CP(*ki, *ki32, ki_ruid); CP(*ki, *ki32, ki_svuid); CP(*ki, *ki32, ki_rgid); CP(*ki, *ki32, ki_svgid); CP(*ki, *ki32, ki_ngroups); for (i = 0; i < KI_NGROUPS; i++) CP(*ki, *ki32, ki_groups[i]); CP(*ki, *ki32, ki_size); CP(*ki, *ki32, ki_rssize); CP(*ki, *ki32, ki_swrss); CP(*ki, *ki32, ki_tsize); CP(*ki, *ki32, ki_dsize); CP(*ki, *ki32, ki_ssize); CP(*ki, *ki32, ki_xstat); CP(*ki, *ki32, ki_acflag); CP(*ki, *ki32, ki_pctcpu); CP(*ki, *ki32, ki_estcpu); CP(*ki, *ki32, ki_slptime); CP(*ki, *ki32, ki_swtime); CP(*ki, *ki32, ki_cow); CP(*ki, *ki32, ki_runtime); TV_CP(*ki, *ki32, ki_start); TV_CP(*ki, *ki32, ki_childtime); CP(*ki, *ki32, ki_flag); CP(*ki, *ki32, ki_kiflag); CP(*ki, *ki32, ki_traceflag); CP(*ki, *ki32, ki_stat); CP(*ki, *ki32, ki_nice); CP(*ki, *ki32, ki_lock); CP(*ki, *ki32, ki_rqindex); CP(*ki, *ki32, ki_oncpu); CP(*ki, *ki32, ki_lastcpu); /* XXX TODO: wrap cpu value as appropriate */ CP(*ki, *ki32, ki_oncpu_old); CP(*ki, *ki32, ki_lastcpu_old); bcopy(ki->ki_tdname, ki32->ki_tdname, TDNAMLEN + 1); bcopy(ki->ki_wmesg, ki32->ki_wmesg, WMESGLEN + 1); bcopy(ki->ki_login, ki32->ki_login, LOGNAMELEN + 1); bcopy(ki->ki_lockname, ki32->ki_lockname, LOCKNAMELEN + 1); bcopy(ki->ki_comm, ki32->ki_comm, COMMLEN + 1); bcopy(ki->ki_emul, ki32->ki_emul, KI_EMULNAMELEN + 1); bcopy(ki->ki_loginclass, ki32->ki_loginclass, LOGINCLASSLEN + 1); CP(*ki, *ki32, ki_tracer); CP(*ki, *ki32, ki_flag2); CP(*ki, *ki32, ki_fibnum); CP(*ki, *ki32, ki_cr_flags); CP(*ki, *ki32, ki_jid); CP(*ki, *ki32, ki_numthreads); CP(*ki, *ki32, ki_tid); CP(*ki, *ki32, ki_pri); freebsd32_rusage_out(&ki->ki_rusage, &ki32->ki_rusage); freebsd32_rusage_out(&ki->ki_rusage_ch, &ki32->ki_rusage_ch); PTRTRIM_CP(*ki, *ki32, ki_pcb); PTRTRIM_CP(*ki, *ki32, ki_kstack); PTRTRIM_CP(*ki, *ki32, ki_udata); CP(*ki, *ki32, ki_sflag); CP(*ki, *ki32, ki_tdflags); } #endif int kern_proc_out(struct proc *p, struct sbuf *sb, int flags) { struct thread *td; struct kinfo_proc ki; #ifdef COMPAT_FREEBSD32 struct kinfo_proc32 ki32; #endif int error; PROC_LOCK_ASSERT(p, MA_OWNED); MPASS(FIRST_THREAD_IN_PROC(p) != NULL); error = 0; fill_kinfo_proc(p, &ki); if ((flags & KERN_PROC_NOTHREADS) != 0) { #ifdef COMPAT_FREEBSD32 if ((flags & KERN_PROC_MASK32) != 0) { freebsd32_kinfo_proc_out(&ki, &ki32); if (sbuf_bcat(sb, &ki32, sizeof(ki32)) != 0) error = ENOMEM; } else #endif if (sbuf_bcat(sb, &ki, sizeof(ki)) != 0) error = ENOMEM; } else { FOREACH_THREAD_IN_PROC(p, td) { fill_kinfo_thread(td, &ki, 1); #ifdef COMPAT_FREEBSD32 if ((flags & KERN_PROC_MASK32) != 0) { freebsd32_kinfo_proc_out(&ki, &ki32); if (sbuf_bcat(sb, &ki32, sizeof(ki32)) != 0) error = ENOMEM; } else #endif if (sbuf_bcat(sb, &ki, sizeof(ki)) != 0) error = ENOMEM; if (error != 0) break; } } PROC_UNLOCK(p); return (error); } static int sysctl_out_proc(struct proc *p, struct sysctl_req *req, int flags, int doingzomb) { struct sbuf sb; struct kinfo_proc ki; struct proc *np; int error, error2; pid_t pid; pid = p->p_pid; sbuf_new_for_sysctl(&sb, (char *)&ki, sizeof(ki), req); sbuf_clear_flags(&sb, SBUF_INCLUDENUL); error = kern_proc_out(p, &sb, flags); error2 = sbuf_finish(&sb); sbuf_delete(&sb); if (error != 0) return (error); else if (error2 != 0) return (error2); if (doingzomb) np = zpfind(pid); else { if (pid == 0) return (0); np = pfind(pid); } if (np == NULL) return (ESRCH); if (np != p) { PROC_UNLOCK(np); return (ESRCH); } PROC_UNLOCK(np); return (0); } static int sysctl_kern_proc(SYSCTL_HANDLER_ARGS) { int *name = (int *)arg1; u_int namelen = arg2; struct proc *p; int flags, doingzomb, oid_number; int error = 0; oid_number = oidp->oid_number; if (oid_number != KERN_PROC_ALL && (oid_number & KERN_PROC_INC_THREAD) == 0) flags = KERN_PROC_NOTHREADS; else { flags = 0; oid_number &= ~KERN_PROC_INC_THREAD; } #ifdef COMPAT_FREEBSD32 if (req->flags & SCTL_MASK32) flags |= KERN_PROC_MASK32; #endif if (oid_number == KERN_PROC_PID) { if (namelen != 1) return (EINVAL); error = sysctl_wire_old_buffer(req, 0); if (error) return (error); sx_slock(&proctree_lock); error = pget((pid_t)name[0], PGET_CANSEE, &p); if (error == 0) error = sysctl_out_proc(p, req, flags, 0); sx_sunlock(&proctree_lock); return (error); } switch (oid_number) { case KERN_PROC_ALL: if (namelen != 0) return (EINVAL); break; case KERN_PROC_PROC: if (namelen != 0 && namelen != 1) return (EINVAL); break; default: if (namelen != 1) return (EINVAL); break; } if (!req->oldptr) { /* overestimate by 5 procs */ error = SYSCTL_OUT(req, 0, sizeof (struct kinfo_proc) * 5); if (error) return (error); } error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); sx_slock(&proctree_lock); sx_slock(&allproc_lock); for (doingzomb=0 ; doingzomb < 2 ; doingzomb++) { if (!doingzomb) p = LIST_FIRST(&allproc); else p = LIST_FIRST(&zombproc); for (; p != NULL; p = LIST_NEXT(p, p_list)) { /* * Skip embryonic processes. */ PROC_LOCK(p); if (p->p_state == PRS_NEW) { PROC_UNLOCK(p); continue; } KASSERT(p->p_ucred != NULL, ("process credential is NULL for non-NEW proc")); /* * Show a user only appropriate processes. */ if (p_cansee(curthread, p)) { PROC_UNLOCK(p); continue; } /* * TODO - make more efficient (see notes below). * do by session. */ switch (oid_number) { case KERN_PROC_GID: if (p->p_ucred->cr_gid != (gid_t)name[0]) { PROC_UNLOCK(p); continue; } break; case KERN_PROC_PGRP: /* could do this by traversing pgrp */ if (p->p_pgrp == NULL || p->p_pgrp->pg_id != (pid_t)name[0]) { PROC_UNLOCK(p); continue; } break; case KERN_PROC_RGID: if (p->p_ucred->cr_rgid != (gid_t)name[0]) { PROC_UNLOCK(p); continue; } break; case KERN_PROC_SESSION: if (p->p_session == NULL || p->p_session->s_sid != (pid_t)name[0]) { PROC_UNLOCK(p); continue; } break; case KERN_PROC_TTY: if ((p->p_flag & P_CONTROLT) == 0 || p->p_session == NULL) { PROC_UNLOCK(p); continue; } /* XXX proctree_lock */ SESS_LOCK(p->p_session); if (p->p_session->s_ttyp == NULL || tty_udev(p->p_session->s_ttyp) != (dev_t)name[0]) { SESS_UNLOCK(p->p_session); PROC_UNLOCK(p); continue; } SESS_UNLOCK(p->p_session); break; case KERN_PROC_UID: if (p->p_ucred->cr_uid != (uid_t)name[0]) { PROC_UNLOCK(p); continue; } break; case KERN_PROC_RUID: if (p->p_ucred->cr_ruid != (uid_t)name[0]) { PROC_UNLOCK(p); continue; } break; case KERN_PROC_PROC: break; default: break; } error = sysctl_out_proc(p, req, flags, doingzomb); if (error) { sx_sunlock(&allproc_lock); sx_sunlock(&proctree_lock); return (error); } } } sx_sunlock(&allproc_lock); sx_sunlock(&proctree_lock); return (0); } struct pargs * pargs_alloc(int len) { struct pargs *pa; pa = malloc(sizeof(struct pargs) + len, M_PARGS, M_WAITOK); refcount_init(&pa->ar_ref, 1); pa->ar_length = len; return (pa); } static void pargs_free(struct pargs *pa) { free(pa, M_PARGS); } void pargs_hold(struct pargs *pa) { if (pa == NULL) return; refcount_acquire(&pa->ar_ref); } void pargs_drop(struct pargs *pa) { if (pa == NULL) return; if (refcount_release(&pa->ar_ref)) pargs_free(pa); } static int proc_read_string(struct thread *td, struct proc *p, const char *sptr, char *buf, size_t len) { ssize_t n; /* * This may return a short read if the string is shorter than the chunk * and is aligned at the end of the page, and the following page is not * mapped. */ n = proc_readmem(td, p, (vm_offset_t)sptr, buf, len); if (n <= 0) return (ENOMEM); return (0); } #define PROC_AUXV_MAX 256 /* Safety limit on auxv size. */ enum proc_vector_type { PROC_ARG, PROC_ENV, PROC_AUX, }; #ifdef COMPAT_FREEBSD32 static int get_proc_vector32(struct thread *td, struct proc *p, char ***proc_vectorp, size_t *vsizep, enum proc_vector_type type) { struct freebsd32_ps_strings pss; Elf32_Auxinfo aux; vm_offset_t vptr, ptr; uint32_t *proc_vector32; char **proc_vector; size_t vsize, size; int i, error; error = 0; if (proc_readmem(td, p, (vm_offset_t)p->p_sysent->sv_psstrings, &pss, sizeof(pss)) != sizeof(pss)) return (ENOMEM); switch (type) { case PROC_ARG: vptr = (vm_offset_t)PTRIN(pss.ps_argvstr); vsize = pss.ps_nargvstr; if (vsize > ARG_MAX) return (ENOEXEC); size = vsize * sizeof(int32_t); break; case PROC_ENV: vptr = (vm_offset_t)PTRIN(pss.ps_envstr); vsize = pss.ps_nenvstr; if (vsize > ARG_MAX) return (ENOEXEC); size = vsize * sizeof(int32_t); break; case PROC_AUX: vptr = (vm_offset_t)PTRIN(pss.ps_envstr) + (pss.ps_nenvstr + 1) * sizeof(int32_t); if (vptr % 4 != 0) return (ENOEXEC); for (ptr = vptr, i = 0; i < PROC_AUXV_MAX; i++) { if (proc_readmem(td, p, ptr, &aux, sizeof(aux)) != sizeof(aux)) return (ENOMEM); if (aux.a_type == AT_NULL) break; ptr += sizeof(aux); } if (aux.a_type != AT_NULL) return (ENOEXEC); vsize = i + 1; size = vsize * sizeof(aux); break; default: KASSERT(0, ("Wrong proc vector type: %d", type)); return (EINVAL); } proc_vector32 = malloc(size, M_TEMP, M_WAITOK); if (proc_readmem(td, p, vptr, proc_vector32, size) != size) { error = ENOMEM; goto done; } if (type == PROC_AUX) { *proc_vectorp = (char **)proc_vector32; *vsizep = vsize; return (0); } proc_vector = malloc(vsize * sizeof(char *), M_TEMP, M_WAITOK); for (i = 0; i < (int)vsize; i++) proc_vector[i] = PTRIN(proc_vector32[i]); *proc_vectorp = proc_vector; *vsizep = vsize; done: free(proc_vector32, M_TEMP); return (error); } #endif static int get_proc_vector(struct thread *td, struct proc *p, char ***proc_vectorp, size_t *vsizep, enum proc_vector_type type) { struct ps_strings pss; Elf_Auxinfo aux; vm_offset_t vptr, ptr; char **proc_vector; size_t vsize, size; int i; #ifdef COMPAT_FREEBSD32 if (SV_PROC_FLAG(p, SV_ILP32) != 0) return (get_proc_vector32(td, p, proc_vectorp, vsizep, type)); #endif if (proc_readmem(td, p, (vm_offset_t)p->p_sysent->sv_psstrings, &pss, sizeof(pss)) != sizeof(pss)) return (ENOMEM); switch (type) { case PROC_ARG: vptr = (vm_offset_t)pss.ps_argvstr; vsize = pss.ps_nargvstr; if (vsize > ARG_MAX) return (ENOEXEC); size = vsize * sizeof(char *); break; case PROC_ENV: vptr = (vm_offset_t)pss.ps_envstr; vsize = pss.ps_nenvstr; if (vsize > ARG_MAX) return (ENOEXEC); size = vsize * sizeof(char *); break; case PROC_AUX: /* * The aux array is just above env array on the stack. Check * that the address is naturally aligned. */ vptr = (vm_offset_t)pss.ps_envstr + (pss.ps_nenvstr + 1) * sizeof(char *); #if __ELF_WORD_SIZE == 64 if (vptr % sizeof(uint64_t) != 0) #else if (vptr % sizeof(uint32_t) != 0) #endif return (ENOEXEC); /* * We count the array size reading the aux vectors from the * stack until AT_NULL vector is returned. So (to keep the code * simple) we read the process stack twice: the first time here * to find the size and the second time when copying the vectors * to the allocated proc_vector. */ for (ptr = vptr, i = 0; i < PROC_AUXV_MAX; i++) { if (proc_readmem(td, p, ptr, &aux, sizeof(aux)) != sizeof(aux)) return (ENOMEM); if (aux.a_type == AT_NULL) break; ptr += sizeof(aux); } /* * If the PROC_AUXV_MAX entries are iterated over, and we have * not reached AT_NULL, it is most likely we are reading wrong * data: either the process doesn't have auxv array or data has * been modified. Return the error in this case. */ if (aux.a_type != AT_NULL) return (ENOEXEC); vsize = i + 1; size = vsize * sizeof(aux); break; default: KASSERT(0, ("Wrong proc vector type: %d", type)); return (EINVAL); /* In case we are built without INVARIANTS. */ } proc_vector = malloc(size, M_TEMP, M_WAITOK); if (proc_readmem(td, p, vptr, proc_vector, size) != size) { free(proc_vector, M_TEMP); return (ENOMEM); } *proc_vectorp = proc_vector; *vsizep = vsize; return (0); } #define GET_PS_STRINGS_CHUNK_SZ 256 /* Chunk size (bytes) for ps_strings operations. */ static int get_ps_strings(struct thread *td, struct proc *p, struct sbuf *sb, enum proc_vector_type type) { size_t done, len, nchr, vsize; int error, i; char **proc_vector, *sptr; char pss_string[GET_PS_STRINGS_CHUNK_SZ]; PROC_ASSERT_HELD(p); /* * We are not going to read more than 2 * (PATH_MAX + ARG_MAX) bytes. */ nchr = 2 * (PATH_MAX + ARG_MAX); error = get_proc_vector(td, p, &proc_vector, &vsize, type); if (error != 0) return (error); for (done = 0, i = 0; i < (int)vsize && done < nchr; i++) { /* * The program may have scribbled into its argv array, e.g. to * remove some arguments. If that has happened, break out * before trying to read from NULL. */ if (proc_vector[i] == NULL) break; for (sptr = proc_vector[i]; ; sptr += GET_PS_STRINGS_CHUNK_SZ) { error = proc_read_string(td, p, sptr, pss_string, sizeof(pss_string)); if (error != 0) goto done; len = strnlen(pss_string, GET_PS_STRINGS_CHUNK_SZ); if (done + len >= nchr) len = nchr - done - 1; sbuf_bcat(sb, pss_string, len); if (len != GET_PS_STRINGS_CHUNK_SZ) break; done += GET_PS_STRINGS_CHUNK_SZ; } sbuf_bcat(sb, "", 1); done += len + 1; } done: free(proc_vector, M_TEMP); return (error); } int proc_getargv(struct thread *td, struct proc *p, struct sbuf *sb) { return (get_ps_strings(curthread, p, sb, PROC_ARG)); } int proc_getenvv(struct thread *td, struct proc *p, struct sbuf *sb) { return (get_ps_strings(curthread, p, sb, PROC_ENV)); } int proc_getauxv(struct thread *td, struct proc *p, struct sbuf *sb) { size_t vsize, size; char **auxv; int error; error = get_proc_vector(td, p, &auxv, &vsize, PROC_AUX); if (error == 0) { #ifdef COMPAT_FREEBSD32 if (SV_PROC_FLAG(p, SV_ILP32) != 0) size = vsize * sizeof(Elf32_Auxinfo); else #endif size = vsize * sizeof(Elf_Auxinfo); if (sbuf_bcat(sb, auxv, size) != 0) error = ENOMEM; free(auxv, M_TEMP); } return (error); } /* * This sysctl allows a process to retrieve the argument list or process * title for another process without groping around in the address space * of the other process. It also allow a process to set its own "process * title to a string of its own choice. */ static int sysctl_kern_proc_args(SYSCTL_HANDLER_ARGS) { int *name = (int *)arg1; u_int namelen = arg2; struct pargs *newpa, *pa; struct proc *p; struct sbuf sb; int flags, error = 0, error2; if (namelen != 1) return (EINVAL); flags = PGET_CANSEE; if (req->newptr != NULL) flags |= PGET_ISCURRENT; error = pget((pid_t)name[0], flags, &p); if (error) return (error); pa = p->p_args; if (pa != NULL) { pargs_hold(pa); PROC_UNLOCK(p); error = SYSCTL_OUT(req, pa->ar_args, pa->ar_length); pargs_drop(pa); } else if ((p->p_flag & (P_WEXIT | P_SYSTEM)) == 0) { _PHOLD(p); PROC_UNLOCK(p); sbuf_new_for_sysctl(&sb, NULL, GET_PS_STRINGS_CHUNK_SZ, req); sbuf_clear_flags(&sb, SBUF_INCLUDENUL); error = proc_getargv(curthread, p, &sb); error2 = sbuf_finish(&sb); PRELE(p); sbuf_delete(&sb); if (error == 0 && error2 != 0) error = error2; } else { PROC_UNLOCK(p); } if (error != 0 || req->newptr == NULL) return (error); if (req->newlen + sizeof(struct pargs) > ps_arg_cache_limit) return (ENOMEM); newpa = pargs_alloc(req->newlen); error = SYSCTL_IN(req, newpa->ar_args, req->newlen); if (error != 0) { pargs_free(newpa); return (error); } PROC_LOCK(p); pa = p->p_args; p->p_args = newpa; PROC_UNLOCK(p); pargs_drop(pa); return (0); } /* * This sysctl allows a process to retrieve environment of another process. */ static int sysctl_kern_proc_env(SYSCTL_HANDLER_ARGS) { int *name = (int *)arg1; u_int namelen = arg2; struct proc *p; struct sbuf sb; int error, error2; if (namelen != 1) return (EINVAL); error = pget((pid_t)name[0], PGET_WANTREAD, &p); if (error != 0) return (error); if ((p->p_flag & P_SYSTEM) != 0) { PRELE(p); return (0); } sbuf_new_for_sysctl(&sb, NULL, GET_PS_STRINGS_CHUNK_SZ, req); sbuf_clear_flags(&sb, SBUF_INCLUDENUL); error = proc_getenvv(curthread, p, &sb); error2 = sbuf_finish(&sb); PRELE(p); sbuf_delete(&sb); return (error != 0 ? error : error2); } /* * This sysctl allows a process to retrieve ELF auxiliary vector of * another process. */ static int sysctl_kern_proc_auxv(SYSCTL_HANDLER_ARGS) { int *name = (int *)arg1; u_int namelen = arg2; struct proc *p; struct sbuf sb; int error, error2; if (namelen != 1) return (EINVAL); error = pget((pid_t)name[0], PGET_WANTREAD, &p); if (error != 0) return (error); if ((p->p_flag & P_SYSTEM) != 0) { PRELE(p); return (0); } sbuf_new_for_sysctl(&sb, NULL, GET_PS_STRINGS_CHUNK_SZ, req); sbuf_clear_flags(&sb, SBUF_INCLUDENUL); error = proc_getauxv(curthread, p, &sb); error2 = sbuf_finish(&sb); PRELE(p); sbuf_delete(&sb); return (error != 0 ? error : error2); } /* * This sysctl allows a process to retrieve the path of the executable for * itself or another process. */ static int sysctl_kern_proc_pathname(SYSCTL_HANDLER_ARGS) { pid_t *pidp = (pid_t *)arg1; unsigned int arglen = arg2; struct proc *p; struct vnode *vp; char *retbuf, *freebuf; int error; if (arglen != 1) return (EINVAL); if (*pidp == -1) { /* -1 means this process */ p = req->td->td_proc; } else { error = pget(*pidp, PGET_CANSEE, &p); if (error != 0) return (error); } vp = p->p_textvp; if (vp == NULL) { if (*pidp != -1) PROC_UNLOCK(p); return (0); } vref(vp); if (*pidp != -1) PROC_UNLOCK(p); error = vn_fullpath(req->td, vp, &retbuf, &freebuf); vrele(vp); if (error) return (error); error = SYSCTL_OUT(req, retbuf, strlen(retbuf) + 1); free(freebuf, M_TEMP); return (error); } static int sysctl_kern_proc_sv_name(SYSCTL_HANDLER_ARGS) { struct proc *p; char *sv_name; int *name; int namelen; int error; namelen = arg2; if (namelen != 1) return (EINVAL); name = (int *)arg1; error = pget((pid_t)name[0], PGET_CANSEE, &p); if (error != 0) return (error); sv_name = p->p_sysent->sv_name; PROC_UNLOCK(p); return (sysctl_handle_string(oidp, sv_name, 0, req)); } #ifdef KINFO_OVMENTRY_SIZE CTASSERT(sizeof(struct kinfo_ovmentry) == KINFO_OVMENTRY_SIZE); #endif #ifdef COMPAT_FREEBSD7 static int sysctl_kern_proc_ovmmap(SYSCTL_HANDLER_ARGS) { vm_map_entry_t entry, tmp_entry; unsigned int last_timestamp; char *fullpath, *freepath; struct kinfo_ovmentry *kve; struct vattr va; struct ucred *cred; int error, *name; struct vnode *vp; struct proc *p; vm_map_t map; struct vmspace *vm; name = (int *)arg1; error = pget((pid_t)name[0], PGET_WANTREAD, &p); if (error != 0) return (error); vm = vmspace_acquire_ref(p); if (vm == NULL) { PRELE(p); return (ESRCH); } kve = malloc(sizeof(*kve), M_TEMP, M_WAITOK); map = &vm->vm_map; vm_map_lock_read(map); for (entry = map->header.next; entry != &map->header; entry = entry->next) { vm_object_t obj, tobj, lobj; vm_offset_t addr; if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) continue; bzero(kve, sizeof(*kve)); kve->kve_structsize = sizeof(*kve); kve->kve_private_resident = 0; obj = entry->object.vm_object; if (obj != NULL) { VM_OBJECT_RLOCK(obj); if (obj->shadow_count == 1) kve->kve_private_resident = obj->resident_page_count; } kve->kve_resident = 0; addr = entry->start; while (addr < entry->end) { if (pmap_extract(map->pmap, addr)) kve->kve_resident++; addr += PAGE_SIZE; } for (lobj = tobj = obj; tobj; tobj = tobj->backing_object) { if (tobj != obj) VM_OBJECT_RLOCK(tobj); if (lobj != obj) VM_OBJECT_RUNLOCK(lobj); lobj = tobj; } kve->kve_start = (void*)entry->start; kve->kve_end = (void*)entry->end; kve->kve_offset = (off_t)entry->offset; if (entry->protection & VM_PROT_READ) kve->kve_protection |= KVME_PROT_READ; if (entry->protection & VM_PROT_WRITE) kve->kve_protection |= KVME_PROT_WRITE; if (entry->protection & VM_PROT_EXECUTE) kve->kve_protection |= KVME_PROT_EXEC; if (entry->eflags & MAP_ENTRY_COW) kve->kve_flags |= KVME_FLAG_COW; if (entry->eflags & MAP_ENTRY_NEEDS_COPY) kve->kve_flags |= KVME_FLAG_NEEDS_COPY; if (entry->eflags & MAP_ENTRY_NOCOREDUMP) kve->kve_flags |= KVME_FLAG_NOCOREDUMP; last_timestamp = map->timestamp; vm_map_unlock_read(map); kve->kve_fileid = 0; kve->kve_fsid = 0; freepath = NULL; fullpath = ""; if (lobj) { vp = NULL; switch (lobj->type) { case OBJT_DEFAULT: kve->kve_type = KVME_TYPE_DEFAULT; break; case OBJT_VNODE: kve->kve_type = KVME_TYPE_VNODE; vp = lobj->handle; vref(vp); break; case OBJT_SWAP: if ((lobj->flags & OBJ_TMPFS_NODE) != 0) { kve->kve_type = KVME_TYPE_VNODE; if ((lobj->flags & OBJ_TMPFS) != 0) { vp = lobj->un_pager.swp.swp_tmpfs; vref(vp); } } else { kve->kve_type = KVME_TYPE_SWAP; } break; case OBJT_DEVICE: kve->kve_type = KVME_TYPE_DEVICE; break; case OBJT_PHYS: kve->kve_type = KVME_TYPE_PHYS; break; case OBJT_DEAD: kve->kve_type = KVME_TYPE_DEAD; break; case OBJT_SG: kve->kve_type = KVME_TYPE_SG; break; default: kve->kve_type = KVME_TYPE_UNKNOWN; break; } if (lobj != obj) VM_OBJECT_RUNLOCK(lobj); kve->kve_ref_count = obj->ref_count; kve->kve_shadow_count = obj->shadow_count; VM_OBJECT_RUNLOCK(obj); if (vp != NULL) { vn_fullpath(curthread, vp, &fullpath, &freepath); cred = curthread->td_ucred; vn_lock(vp, LK_SHARED | LK_RETRY); if (VOP_GETATTR(vp, &va, cred) == 0) { kve->kve_fileid = va.va_fileid; kve->kve_fsid = va.va_fsid; } vput(vp); } } else { kve->kve_type = KVME_TYPE_NONE; kve->kve_ref_count = 0; kve->kve_shadow_count = 0; } strlcpy(kve->kve_path, fullpath, sizeof(kve->kve_path)); if (freepath != NULL) free(freepath, M_TEMP); error = SYSCTL_OUT(req, kve, sizeof(*kve)); vm_map_lock_read(map); if (error) break; if (last_timestamp != map->timestamp) { vm_map_lookup_entry(map, addr - 1, &tmp_entry); entry = tmp_entry; } } vm_map_unlock_read(map); vmspace_free(vm); PRELE(p); free(kve, M_TEMP); return (error); } #endif /* COMPAT_FREEBSD7 */ #ifdef KINFO_VMENTRY_SIZE CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE); #endif static void kern_proc_vmmap_resident(vm_map_t map, vm_map_entry_t entry, struct kinfo_vmentry *kve) { vm_object_t obj, tobj; vm_page_t m, m_adv; vm_offset_t addr; vm_paddr_t locked_pa; vm_pindex_t pi, pi_adv, pindex; locked_pa = 0; obj = entry->object.vm_object; addr = entry->start; m_adv = NULL; pi = OFF_TO_IDX(entry->offset); for (; addr < entry->end; addr += IDX_TO_OFF(pi_adv), pi += pi_adv) { if (m_adv != NULL) { m = m_adv; } else { pi_adv = OFF_TO_IDX(entry->end - addr); pindex = pi; for (tobj = obj;; tobj = tobj->backing_object) { m = vm_page_find_least(tobj, pindex); if (m != NULL) { if (m->pindex == pindex) break; if (pi_adv > m->pindex - pindex) { pi_adv = m->pindex - pindex; m_adv = m; } } if (tobj->backing_object == NULL) goto next; pindex += OFF_TO_IDX(tobj-> backing_object_offset); } } m_adv = NULL; if (m->psind != 0 && addr + pagesizes[1] <= entry->end && (addr & (pagesizes[1] - 1)) == 0 && (pmap_mincore(map->pmap, addr, &locked_pa) & MINCORE_SUPER) != 0) { kve->kve_flags |= KVME_FLAG_SUPER; pi_adv = OFF_TO_IDX(pagesizes[1]); } else { /* * We do not test the found page on validity. * Either the page is busy and being paged in, * or it was invalidated. The first case * should be counted as resident, the second * is not so clear; we do account both. */ pi_adv = 1; } kve->kve_resident += pi_adv; next:; } PA_UNLOCK_COND(locked_pa); } /* * Must be called with the process locked and will return unlocked. */ int kern_proc_vmmap_out(struct proc *p, struct sbuf *sb, ssize_t maxlen, int flags) { vm_map_entry_t entry, tmp_entry; struct vattr va; vm_map_t map; vm_object_t obj, tobj, lobj; char *fullpath, *freepath; struct kinfo_vmentry *kve; struct ucred *cred; struct vnode *vp; struct vmspace *vm; vm_offset_t addr; unsigned int last_timestamp; int error; PROC_LOCK_ASSERT(p, MA_OWNED); _PHOLD(p); PROC_UNLOCK(p); vm = vmspace_acquire_ref(p); if (vm == NULL) { PRELE(p); return (ESRCH); } kve = malloc(sizeof(*kve), M_TEMP, M_WAITOK | M_ZERO); error = 0; map = &vm->vm_map; vm_map_lock_read(map); for (entry = map->header.next; entry != &map->header; entry = entry->next) { if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) continue; addr = entry->end; bzero(kve, sizeof(*kve)); obj = entry->object.vm_object; if (obj != NULL) { for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) { VM_OBJECT_RLOCK(tobj); lobj = tobj; } if (obj->backing_object == NULL) kve->kve_private_resident = obj->resident_page_count; if (!vmmap_skip_res_cnt) kern_proc_vmmap_resident(map, entry, kve); for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) { if (tobj != obj && tobj != lobj) VM_OBJECT_RUNLOCK(tobj); } } else { lobj = NULL; } kve->kve_start = entry->start; kve->kve_end = entry->end; kve->kve_offset = entry->offset; if (entry->protection & VM_PROT_READ) kve->kve_protection |= KVME_PROT_READ; if (entry->protection & VM_PROT_WRITE) kve->kve_protection |= KVME_PROT_WRITE; if (entry->protection & VM_PROT_EXECUTE) kve->kve_protection |= KVME_PROT_EXEC; if (entry->eflags & MAP_ENTRY_COW) kve->kve_flags |= KVME_FLAG_COW; if (entry->eflags & MAP_ENTRY_NEEDS_COPY) kve->kve_flags |= KVME_FLAG_NEEDS_COPY; if (entry->eflags & MAP_ENTRY_NOCOREDUMP) kve->kve_flags |= KVME_FLAG_NOCOREDUMP; if (entry->eflags & MAP_ENTRY_GROWS_UP) kve->kve_flags |= KVME_FLAG_GROWS_UP; if (entry->eflags & MAP_ENTRY_GROWS_DOWN) kve->kve_flags |= KVME_FLAG_GROWS_DOWN; last_timestamp = map->timestamp; vm_map_unlock_read(map); freepath = NULL; fullpath = ""; if (lobj != NULL) { vp = NULL; switch (lobj->type) { case OBJT_DEFAULT: kve->kve_type = KVME_TYPE_DEFAULT; break; case OBJT_VNODE: kve->kve_type = KVME_TYPE_VNODE; vp = lobj->handle; vref(vp); break; case OBJT_SWAP: if ((lobj->flags & OBJ_TMPFS_NODE) != 0) { kve->kve_type = KVME_TYPE_VNODE; if ((lobj->flags & OBJ_TMPFS) != 0) { vp = lobj->un_pager.swp.swp_tmpfs; vref(vp); } } else { kve->kve_type = KVME_TYPE_SWAP; } break; case OBJT_DEVICE: kve->kve_type = KVME_TYPE_DEVICE; break; case OBJT_PHYS: kve->kve_type = KVME_TYPE_PHYS; break; case OBJT_DEAD: kve->kve_type = KVME_TYPE_DEAD; break; case OBJT_SG: kve->kve_type = KVME_TYPE_SG; break; case OBJT_MGTDEVICE: kve->kve_type = KVME_TYPE_MGTDEVICE; break; default: kve->kve_type = KVME_TYPE_UNKNOWN; break; } if (lobj != obj) VM_OBJECT_RUNLOCK(lobj); kve->kve_ref_count = obj->ref_count; kve->kve_shadow_count = obj->shadow_count; VM_OBJECT_RUNLOCK(obj); if (vp != NULL) { vn_fullpath(curthread, vp, &fullpath, &freepath); kve->kve_vn_type = vntype_to_kinfo(vp->v_type); cred = curthread->td_ucred; vn_lock(vp, LK_SHARED | LK_RETRY); if (VOP_GETATTR(vp, &va, cred) == 0) { kve->kve_vn_fileid = va.va_fileid; kve->kve_vn_fsid = va.va_fsid; kve->kve_vn_mode = MAKEIMODE(va.va_type, va.va_mode); kve->kve_vn_size = va.va_size; kve->kve_vn_rdev = va.va_rdev; kve->kve_status = KF_ATTR_VALID; } vput(vp); } } else { kve->kve_type = KVME_TYPE_NONE; kve->kve_ref_count = 0; kve->kve_shadow_count = 0; } strlcpy(kve->kve_path, fullpath, sizeof(kve->kve_path)); if (freepath != NULL) free(freepath, M_TEMP); /* Pack record size down */ if ((flags & KERN_VMMAP_PACK_KINFO) != 0) kve->kve_structsize = offsetof(struct kinfo_vmentry, kve_path) + strlen(kve->kve_path) + 1; else kve->kve_structsize = sizeof(*kve); kve->kve_structsize = roundup(kve->kve_structsize, sizeof(uint64_t)); /* Halt filling and truncate rather than exceeding maxlen */ if (maxlen != -1 && maxlen < kve->kve_structsize) { error = 0; vm_map_lock_read(map); break; } else if (maxlen != -1) maxlen -= kve->kve_structsize; if (sbuf_bcat(sb, kve, kve->kve_structsize) != 0) error = ENOMEM; vm_map_lock_read(map); if (error != 0) break; if (last_timestamp != map->timestamp) { vm_map_lookup_entry(map, addr - 1, &tmp_entry); entry = tmp_entry; } } vm_map_unlock_read(map); vmspace_free(vm); PRELE(p); free(kve, M_TEMP); return (error); } static int sysctl_kern_proc_vmmap(SYSCTL_HANDLER_ARGS) { struct proc *p; struct sbuf sb; int error, error2, *name; name = (int *)arg1; sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_vmentry), req); sbuf_clear_flags(&sb, SBUF_INCLUDENUL); error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); if (error != 0) { sbuf_delete(&sb); return (error); } error = kern_proc_vmmap_out(p, &sb, -1, KERN_VMMAP_PACK_KINFO); error2 = sbuf_finish(&sb); sbuf_delete(&sb); return (error != 0 ? error : error2); } #if defined(STACK) || defined(DDB) static int sysctl_kern_proc_kstack(SYSCTL_HANDLER_ARGS) { struct kinfo_kstack *kkstp; int error, i, *name, numthreads; lwpid_t *lwpidarray; struct thread *td; struct stack *st; struct sbuf sb; struct proc *p; name = (int *)arg1; error = pget((pid_t)name[0], PGET_NOTINEXEC | PGET_WANTREAD, &p); if (error != 0) return (error); kkstp = malloc(sizeof(*kkstp), M_TEMP, M_WAITOK); st = stack_create(); lwpidarray = NULL; PROC_LOCK(p); do { if (lwpidarray != NULL) { free(lwpidarray, M_TEMP); lwpidarray = NULL; } numthreads = p->p_numthreads; PROC_UNLOCK(p); lwpidarray = malloc(sizeof(*lwpidarray) * numthreads, M_TEMP, M_WAITOK | M_ZERO); PROC_LOCK(p); } while (numthreads < p->p_numthreads); /* * XXXRW: During the below loop, execve(2) and countless other sorts * of changes could have taken place. Should we check to see if the * vmspace has been replaced, or the like, in order to prevent * giving a snapshot that spans, say, execve(2), with some threads * before and some after? Among other things, the credentials could * have changed, in which case the right to extract debug info might * no longer be assured. */ i = 0; FOREACH_THREAD_IN_PROC(p, td) { KASSERT(i < numthreads, ("sysctl_kern_proc_kstack: numthreads")); lwpidarray[i] = td->td_tid; i++; } numthreads = i; for (i = 0; i < numthreads; i++) { td = thread_find(p, lwpidarray[i]); if (td == NULL) { continue; } bzero(kkstp, sizeof(*kkstp)); (void)sbuf_new(&sb, kkstp->kkst_trace, sizeof(kkstp->kkst_trace), SBUF_FIXEDLEN); thread_lock(td); kkstp->kkst_tid = td->td_tid; if (TD_IS_SWAPPED(td)) { kkstp->kkst_state = KKST_STATE_SWAPPED; } else if (TD_IS_RUNNING(td)) { if (stack_save_td_running(st, td) == 0) kkstp->kkst_state = KKST_STATE_STACKOK; else kkstp->kkst_state = KKST_STATE_RUNNING; } else { kkstp->kkst_state = KKST_STATE_STACKOK; stack_save_td(st, td); } thread_unlock(td); PROC_UNLOCK(p); stack_sbuf_print(&sb, st); sbuf_finish(&sb); sbuf_delete(&sb); error = SYSCTL_OUT(req, kkstp, sizeof(*kkstp)); PROC_LOCK(p); if (error) break; } _PRELE(p); PROC_UNLOCK(p); if (lwpidarray != NULL) free(lwpidarray, M_TEMP); stack_destroy(st); free(kkstp, M_TEMP); return (error); } #endif /* * This sysctl allows a process to retrieve the full list of groups from * itself or another process. */ static int sysctl_kern_proc_groups(SYSCTL_HANDLER_ARGS) { pid_t *pidp = (pid_t *)arg1; unsigned int arglen = arg2; struct proc *p; struct ucred *cred; int error; if (arglen != 1) return (EINVAL); if (*pidp == -1) { /* -1 means this process */ p = req->td->td_proc; PROC_LOCK(p); } else { error = pget(*pidp, PGET_CANSEE, &p); if (error != 0) return (error); } cred = crhold(p->p_ucred); PROC_UNLOCK(p); error = SYSCTL_OUT(req, cred->cr_groups, cred->cr_ngroups * sizeof(gid_t)); crfree(cred); return (error); } /* * This sysctl allows a process to retrieve or/and set the resource limit for * another process. */ static int sysctl_kern_proc_rlimit(SYSCTL_HANDLER_ARGS) { int *name = (int *)arg1; u_int namelen = arg2; struct rlimit rlim; struct proc *p; u_int which; int flags, error; if (namelen != 2) return (EINVAL); which = (u_int)name[1]; if (which >= RLIM_NLIMITS) return (EINVAL); if (req->newptr != NULL && req->newlen != sizeof(rlim)) return (EINVAL); flags = PGET_HOLD | PGET_NOTWEXIT; if (req->newptr != NULL) flags |= PGET_CANDEBUG; else flags |= PGET_CANSEE; error = pget((pid_t)name[0], flags, &p); if (error != 0) return (error); /* * Retrieve limit. */ if (req->oldptr != NULL) { PROC_LOCK(p); lim_rlimit_proc(p, which, &rlim); PROC_UNLOCK(p); } error = SYSCTL_OUT(req, &rlim, sizeof(rlim)); if (error != 0) goto errout; /* * Set limit. */ if (req->newptr != NULL) { error = SYSCTL_IN(req, &rlim, sizeof(rlim)); if (error == 0) error = kern_proc_setrlimit(curthread, p, which, &rlim); } errout: PRELE(p); return (error); } /* * This sysctl allows a process to retrieve ps_strings structure location of * another process. */ static int sysctl_kern_proc_ps_strings(SYSCTL_HANDLER_ARGS) { int *name = (int *)arg1; u_int namelen = arg2; struct proc *p; vm_offset_t ps_strings; int error; #ifdef COMPAT_FREEBSD32 uint32_t ps_strings32; #endif if (namelen != 1) return (EINVAL); error = pget((pid_t)name[0], PGET_CANDEBUG, &p); if (error != 0) return (error); #ifdef COMPAT_FREEBSD32 if ((req->flags & SCTL_MASK32) != 0) { /* * We return 0 if the 32 bit emulation request is for a 64 bit * process. */ ps_strings32 = SV_PROC_FLAG(p, SV_ILP32) != 0 ? PTROUT(p->p_sysent->sv_psstrings) : 0; PROC_UNLOCK(p); error = SYSCTL_OUT(req, &ps_strings32, sizeof(ps_strings32)); return (error); } #endif ps_strings = p->p_sysent->sv_psstrings; PROC_UNLOCK(p); error = SYSCTL_OUT(req, &ps_strings, sizeof(ps_strings)); return (error); } /* * This sysctl allows a process to retrieve umask of another process. */ static int sysctl_kern_proc_umask(SYSCTL_HANDLER_ARGS) { int *name = (int *)arg1; u_int namelen = arg2; struct proc *p; int error; u_short fd_cmask; if (namelen != 1) return (EINVAL); error = pget((pid_t)name[0], PGET_WANTREAD, &p); if (error != 0) return (error); FILEDESC_SLOCK(p->p_fd); fd_cmask = p->p_fd->fd_cmask; FILEDESC_SUNLOCK(p->p_fd); PRELE(p); error = SYSCTL_OUT(req, &fd_cmask, sizeof(fd_cmask)); return (error); } /* * This sysctl allows a process to set and retrieve binary osreldate of * another process. */ static int sysctl_kern_proc_osrel(SYSCTL_HANDLER_ARGS) { int *name = (int *)arg1; u_int namelen = arg2; struct proc *p; int flags, error, osrel; if (namelen != 1) return (EINVAL); if (req->newptr != NULL && req->newlen != sizeof(osrel)) return (EINVAL); flags = PGET_HOLD | PGET_NOTWEXIT; if (req->newptr != NULL) flags |= PGET_CANDEBUG; else flags |= PGET_CANSEE; error = pget((pid_t)name[0], flags, &p); if (error != 0) return (error); error = SYSCTL_OUT(req, &p->p_osrel, sizeof(p->p_osrel)); if (error != 0) goto errout; if (req->newptr != NULL) { error = SYSCTL_IN(req, &osrel, sizeof(osrel)); if (error != 0) goto errout; if (osrel < 0) { error = EINVAL; goto errout; } p->p_osrel = osrel; } errout: PRELE(p); return (error); } static int sysctl_kern_proc_sigtramp(SYSCTL_HANDLER_ARGS) { int *name = (int *)arg1; u_int namelen = arg2; struct proc *p; struct kinfo_sigtramp kst; const struct sysentvec *sv; int error; #ifdef COMPAT_FREEBSD32 struct kinfo_sigtramp32 kst32; #endif if (namelen != 1) return (EINVAL); error = pget((pid_t)name[0], PGET_CANDEBUG, &p); if (error != 0) return (error); sv = p->p_sysent; #ifdef COMPAT_FREEBSD32 if ((req->flags & SCTL_MASK32) != 0) { bzero(&kst32, sizeof(kst32)); if (SV_PROC_FLAG(p, SV_ILP32)) { if (sv->sv_sigcode_base != 0) { kst32.ksigtramp_start = sv->sv_sigcode_base; kst32.ksigtramp_end = sv->sv_sigcode_base + *sv->sv_szsigcode; } else { kst32.ksigtramp_start = sv->sv_psstrings - *sv->sv_szsigcode; kst32.ksigtramp_end = sv->sv_psstrings; } } PROC_UNLOCK(p); error = SYSCTL_OUT(req, &kst32, sizeof(kst32)); return (error); } #endif bzero(&kst, sizeof(kst)); if (sv->sv_sigcode_base != 0) { kst.ksigtramp_start = (char *)sv->sv_sigcode_base; kst.ksigtramp_end = (char *)sv->sv_sigcode_base + *sv->sv_szsigcode; } else { kst.ksigtramp_start = (char *)sv->sv_psstrings - *sv->sv_szsigcode; kst.ksigtramp_end = (char *)sv->sv_psstrings; } PROC_UNLOCK(p); error = SYSCTL_OUT(req, &kst, sizeof(kst)); return (error); } SYSCTL_NODE(_kern, KERN_PROC, proc, CTLFLAG_RD, 0, "Process table"); SYSCTL_PROC(_kern_proc, KERN_PROC_ALL, all, CTLFLAG_RD|CTLTYPE_STRUCT| CTLFLAG_MPSAFE, 0, 0, sysctl_kern_proc, "S,proc", "Return entire process table"); static SYSCTL_NODE(_kern_proc, KERN_PROC_GID, gid, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc, "Process table"); static SYSCTL_NODE(_kern_proc, KERN_PROC_PGRP, pgrp, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc, "Process table"); static SYSCTL_NODE(_kern_proc, KERN_PROC_RGID, rgid, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc, "Process table"); static SYSCTL_NODE(_kern_proc, KERN_PROC_SESSION, sid, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc, "Process table"); static SYSCTL_NODE(_kern_proc, KERN_PROC_TTY, tty, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc, "Process table"); static SYSCTL_NODE(_kern_proc, KERN_PROC_UID, uid, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc, "Process table"); static SYSCTL_NODE(_kern_proc, KERN_PROC_RUID, ruid, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc, "Process table"); static SYSCTL_NODE(_kern_proc, KERN_PROC_PID, pid, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc, "Process table"); static SYSCTL_NODE(_kern_proc, KERN_PROC_PROC, proc, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc, "Return process table, no threads"); static SYSCTL_NODE(_kern_proc, KERN_PROC_ARGS, args, CTLFLAG_RW | CTLFLAG_CAPWR | CTLFLAG_ANYBODY | CTLFLAG_MPSAFE, sysctl_kern_proc_args, "Process argument list"); static SYSCTL_NODE(_kern_proc, KERN_PROC_ENV, env, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc_env, "Process environment"); static SYSCTL_NODE(_kern_proc, KERN_PROC_AUXV, auxv, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc_auxv, "Process ELF auxiliary vector"); static SYSCTL_NODE(_kern_proc, KERN_PROC_PATHNAME, pathname, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc_pathname, "Process executable path"); static SYSCTL_NODE(_kern_proc, KERN_PROC_SV_NAME, sv_name, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc_sv_name, "Process syscall vector name (ABI type)"); static SYSCTL_NODE(_kern_proc, (KERN_PROC_GID | KERN_PROC_INC_THREAD), gid_td, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc, "Process table"); static SYSCTL_NODE(_kern_proc, (KERN_PROC_PGRP | KERN_PROC_INC_THREAD), pgrp_td, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc, "Process table"); static SYSCTL_NODE(_kern_proc, (KERN_PROC_RGID | KERN_PROC_INC_THREAD), rgid_td, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc, "Process table"); static SYSCTL_NODE(_kern_proc, (KERN_PROC_SESSION | KERN_PROC_INC_THREAD), sid_td, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc, "Process table"); static SYSCTL_NODE(_kern_proc, (KERN_PROC_TTY | KERN_PROC_INC_THREAD), tty_td, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc, "Process table"); static SYSCTL_NODE(_kern_proc, (KERN_PROC_UID | KERN_PROC_INC_THREAD), uid_td, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc, "Process table"); static SYSCTL_NODE(_kern_proc, (KERN_PROC_RUID | KERN_PROC_INC_THREAD), ruid_td, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc, "Process table"); static SYSCTL_NODE(_kern_proc, (KERN_PROC_PID | KERN_PROC_INC_THREAD), pid_td, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc, "Process table"); static SYSCTL_NODE(_kern_proc, (KERN_PROC_PROC | KERN_PROC_INC_THREAD), proc_td, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc, "Return process table, no threads"); #ifdef COMPAT_FREEBSD7 static SYSCTL_NODE(_kern_proc, KERN_PROC_OVMMAP, ovmmap, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc_ovmmap, "Old Process vm map entries"); #endif static SYSCTL_NODE(_kern_proc, KERN_PROC_VMMAP, vmmap, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc_vmmap, "Process vm map entries"); #if defined(STACK) || defined(DDB) static SYSCTL_NODE(_kern_proc, KERN_PROC_KSTACK, kstack, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc_kstack, "Process kernel stacks"); #endif static SYSCTL_NODE(_kern_proc, KERN_PROC_GROUPS, groups, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc_groups, "Process groups"); static SYSCTL_NODE(_kern_proc, KERN_PROC_RLIMIT, rlimit, CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MPSAFE, sysctl_kern_proc_rlimit, "Process resource limits"); static SYSCTL_NODE(_kern_proc, KERN_PROC_PS_STRINGS, ps_strings, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc_ps_strings, "Process ps_strings location"); static SYSCTL_NODE(_kern_proc, KERN_PROC_UMASK, umask, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc_umask, "Process umask"); static SYSCTL_NODE(_kern_proc, KERN_PROC_OSREL, osrel, CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MPSAFE, sysctl_kern_proc_osrel, "Process binary osreldate"); static SYSCTL_NODE(_kern_proc, KERN_PROC_SIGTRAMP, sigtramp, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_kern_proc_sigtramp, "Process signal trampoline location"); int allproc_gen; /* * stop_all_proc() purpose is to stop all process which have usermode, * except current process for obvious reasons. This makes it somewhat * unreliable when invoked from multithreaded process. The service * must not be user-callable anyway. */ void stop_all_proc(void) { struct proc *cp, *p; int r, gen; bool restart, seen_stopped, seen_exiting, stopped_some; cp = curproc; allproc_loop: sx_xlock(&allproc_lock); gen = allproc_gen; seen_exiting = seen_stopped = stopped_some = restart = false; LIST_REMOVE(cp, p_list); LIST_INSERT_HEAD(&allproc, cp, p_list); for (;;) { p = LIST_NEXT(cp, p_list); if (p == NULL) break; LIST_REMOVE(cp, p_list); LIST_INSERT_AFTER(p, cp, p_list); PROC_LOCK(p); if ((p->p_flag & (P_KPROC | P_SYSTEM | P_TOTAL_STOP)) != 0) { PROC_UNLOCK(p); continue; } if ((p->p_flag & P_WEXIT) != 0) { seen_exiting = true; PROC_UNLOCK(p); continue; } if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { /* * Stopped processes are tolerated when there * are no other processes which might continue * them. P_STOPPED_SINGLE but not * P_TOTAL_STOP process still has at least one * thread running. */ seen_stopped = true; PROC_UNLOCK(p); continue; } _PHOLD(p); sx_xunlock(&allproc_lock); r = thread_single(p, SINGLE_ALLPROC); if (r != 0) restart = true; else stopped_some = true; _PRELE(p); PROC_UNLOCK(p); sx_xlock(&allproc_lock); } /* Catch forked children we did not see in iteration. */ if (gen != allproc_gen) restart = true; sx_xunlock(&allproc_lock); if (restart || stopped_some || seen_exiting || seen_stopped) { kern_yield(PRI_USER); goto allproc_loop; } } void resume_all_proc(void) { struct proc *cp, *p; cp = curproc; sx_xlock(&allproc_lock); LIST_REMOVE(cp, p_list); LIST_INSERT_HEAD(&allproc, cp, p_list); for (;;) { p = LIST_NEXT(cp, p_list); if (p == NULL) break; LIST_REMOVE(cp, p_list); LIST_INSERT_AFTER(p, cp, p_list); PROC_LOCK(p); if ((p->p_flag & P_TOTAL_STOP) != 0) { sx_xunlock(&allproc_lock); _PHOLD(p); thread_single_end(p, SINGLE_ALLPROC); _PRELE(p); PROC_UNLOCK(p); sx_xlock(&allproc_lock); } else { PROC_UNLOCK(p); } } sx_xunlock(&allproc_lock); } /* #define TOTAL_STOP_DEBUG 1 */ #ifdef TOTAL_STOP_DEBUG volatile static int ap_resume; #include static int sysctl_debug_stop_all_proc(SYSCTL_HANDLER_ARGS) { int error, val; val = 0; ap_resume = 0; error = sysctl_handle_int(oidp, &val, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (val != 0) { stop_all_proc(); syncer_suspend(); while (ap_resume == 0) ; syncer_resume(); resume_all_proc(); } return (0); } SYSCTL_PROC(_debug, OID_AUTO, stop_all_proc, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, __DEVOLATILE(int *, &ap_resume), 0, sysctl_debug_stop_all_proc, "I", ""); #endif Index: head/sys/kern/kern_prot.c =================================================================== --- head/sys/kern/kern_prot.c (revision 305831) +++ head/sys/kern/kern_prot.c (revision 305832) @@ -1,2213 +1,2213 @@ /*- * Copyright (c) 1982, 1986, 1989, 1990, 1991, 1993 * The Regents of the University of California. * (c) UNIX System Laboratories, Inc. * Copyright (c) 2000-2001 Robert N. M. Watson. * All rights reserved. * * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_prot.c 8.6 (Berkeley) 1/21/94 */ /* * System calls related to processes and protection */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef REGRESSION FEATURE(regression, "Kernel support for interfaces necessary for regression testing (SECURITY RISK!)"); #endif #include #include static MALLOC_DEFINE(M_CRED, "cred", "credentials"); SYSCTL_NODE(_security, OID_AUTO, bsd, CTLFLAG_RW, 0, "BSD security policy"); static void crsetgroups_locked(struct ucred *cr, int ngrp, gid_t *groups); #ifndef _SYS_SYSPROTO_H_ struct getpid_args { int dummy; }; #endif /* ARGSUSED */ int sys_getpid(struct thread *td, struct getpid_args *uap) { struct proc *p = td->td_proc; td->td_retval[0] = p->p_pid; #if defined(COMPAT_43) td->td_retval[1] = kern_getppid(td); #endif return (0); } #ifndef _SYS_SYSPROTO_H_ struct getppid_args { int dummy; }; #endif /* ARGSUSED */ int sys_getppid(struct thread *td, struct getppid_args *uap) { td->td_retval[0] = kern_getppid(td); return (0); } int kern_getppid(struct thread *td) { struct proc *p = td->td_proc; struct proc *pp; int ppid; PROC_LOCK(p); if (!(p->p_flag & P_TRACED)) { ppid = p->p_pptr->p_pid; PROC_UNLOCK(p); } else { PROC_UNLOCK(p); sx_slock(&proctree_lock); pp = proc_realparent(p); ppid = pp->p_pid; sx_sunlock(&proctree_lock); } return (ppid); } /* * Get process group ID; note that POSIX getpgrp takes no parameter. */ #ifndef _SYS_SYSPROTO_H_ struct getpgrp_args { int dummy; }; #endif int sys_getpgrp(struct thread *td, struct getpgrp_args *uap) { struct proc *p = td->td_proc; PROC_LOCK(p); td->td_retval[0] = p->p_pgrp->pg_id; PROC_UNLOCK(p); return (0); } /* Get an arbitrary pid's process group id */ #ifndef _SYS_SYSPROTO_H_ struct getpgid_args { pid_t pid; }; #endif int sys_getpgid(struct thread *td, struct getpgid_args *uap) { struct proc *p; int error; if (uap->pid == 0) { p = td->td_proc; PROC_LOCK(p); } else { p = pfind(uap->pid); if (p == NULL) return (ESRCH); error = p_cansee(td, p); if (error) { PROC_UNLOCK(p); return (error); } } td->td_retval[0] = p->p_pgrp->pg_id; PROC_UNLOCK(p); return (0); } /* * Get an arbitrary pid's session id. */ #ifndef _SYS_SYSPROTO_H_ struct getsid_args { pid_t pid; }; #endif int sys_getsid(struct thread *td, struct getsid_args *uap) { struct proc *p; int error; if (uap->pid == 0) { p = td->td_proc; PROC_LOCK(p); } else { p = pfind(uap->pid); if (p == NULL) return (ESRCH); error = p_cansee(td, p); if (error) { PROC_UNLOCK(p); return (error); } } td->td_retval[0] = p->p_session->s_sid; PROC_UNLOCK(p); return (0); } #ifndef _SYS_SYSPROTO_H_ struct getuid_args { int dummy; }; #endif /* ARGSUSED */ int sys_getuid(struct thread *td, struct getuid_args *uap) { td->td_retval[0] = td->td_ucred->cr_ruid; #if defined(COMPAT_43) td->td_retval[1] = td->td_ucred->cr_uid; #endif return (0); } #ifndef _SYS_SYSPROTO_H_ struct geteuid_args { int dummy; }; #endif /* ARGSUSED */ int sys_geteuid(struct thread *td, struct geteuid_args *uap) { td->td_retval[0] = td->td_ucred->cr_uid; return (0); } #ifndef _SYS_SYSPROTO_H_ struct getgid_args { int dummy; }; #endif /* ARGSUSED */ int sys_getgid(struct thread *td, struct getgid_args *uap) { td->td_retval[0] = td->td_ucred->cr_rgid; #if defined(COMPAT_43) td->td_retval[1] = td->td_ucred->cr_groups[0]; #endif return (0); } /* * Get effective group ID. The "egid" is groups[0], and could be obtained * via getgroups. This syscall exists because it is somewhat painful to do * correctly in a library function. */ #ifndef _SYS_SYSPROTO_H_ struct getegid_args { int dummy; }; #endif /* ARGSUSED */ int sys_getegid(struct thread *td, struct getegid_args *uap) { td->td_retval[0] = td->td_ucred->cr_groups[0]; return (0); } #ifndef _SYS_SYSPROTO_H_ struct getgroups_args { u_int gidsetsize; gid_t *gidset; }; #endif int sys_getgroups(struct thread *td, register struct getgroups_args *uap) { struct ucred *cred; u_int ngrp; int error; cred = td->td_ucred; ngrp = cred->cr_ngroups; if (uap->gidsetsize == 0) { error = 0; goto out; } if (uap->gidsetsize < ngrp) return (EINVAL); error = copyout(cred->cr_groups, uap->gidset, ngrp * sizeof(gid_t)); out: td->td_retval[0] = ngrp; return (error); } #ifndef _SYS_SYSPROTO_H_ struct setsid_args { int dummy; }; #endif /* ARGSUSED */ int sys_setsid(register struct thread *td, struct setsid_args *uap) { struct pgrp *pgrp; int error; struct proc *p = td->td_proc; struct pgrp *newpgrp; struct session *newsess; error = 0; pgrp = NULL; newpgrp = malloc(sizeof(struct pgrp), M_PGRP, M_WAITOK | M_ZERO); newsess = malloc(sizeof(struct session), M_SESSION, M_WAITOK | M_ZERO); sx_xlock(&proctree_lock); if (p->p_pgid == p->p_pid || (pgrp = pgfind(p->p_pid)) != NULL) { if (pgrp != NULL) PGRP_UNLOCK(pgrp); error = EPERM; } else { (void)enterpgrp(p, p->p_pid, newpgrp, newsess); td->td_retval[0] = p->p_pid; newpgrp = NULL; newsess = NULL; } sx_xunlock(&proctree_lock); if (newpgrp != NULL) free(newpgrp, M_PGRP); if (newsess != NULL) free(newsess, M_SESSION); return (error); } /* * set process group (setpgid/old setpgrp) * * caller does setpgid(targpid, targpgid) * * pid must be caller or child of caller (ESRCH) * if a child * pid must be in same session (EPERM) * pid can't have done an exec (EACCES) * if pgid != pid * there must exist some pid in same session having pgid (EPERM) * pid must not be session leader (EPERM) */ #ifndef _SYS_SYSPROTO_H_ struct setpgid_args { int pid; /* target process id */ int pgid; /* target pgrp id */ }; #endif /* ARGSUSED */ int sys_setpgid(struct thread *td, register struct setpgid_args *uap) { struct proc *curp = td->td_proc; register struct proc *targp; /* target process */ register struct pgrp *pgrp; /* target pgrp */ int error; struct pgrp *newpgrp; if (uap->pgid < 0) return (EINVAL); error = 0; newpgrp = malloc(sizeof(struct pgrp), M_PGRP, M_WAITOK | M_ZERO); sx_xlock(&proctree_lock); if (uap->pid != 0 && uap->pid != curp->p_pid) { if ((targp = pfind(uap->pid)) == NULL) { error = ESRCH; goto done; } if (!inferior(targp)) { PROC_UNLOCK(targp); error = ESRCH; goto done; } if ((error = p_cansee(td, targp))) { PROC_UNLOCK(targp); goto done; } if (targp->p_pgrp == NULL || targp->p_session != curp->p_session) { PROC_UNLOCK(targp); error = EPERM; goto done; } if (targp->p_flag & P_EXEC) { PROC_UNLOCK(targp); error = EACCES; goto done; } PROC_UNLOCK(targp); } else targp = curp; if (SESS_LEADER(targp)) { error = EPERM; goto done; } if (uap->pgid == 0) uap->pgid = targp->p_pid; if ((pgrp = pgfind(uap->pgid)) == NULL) { if (uap->pgid == targp->p_pid) { error = enterpgrp(targp, uap->pgid, newpgrp, NULL); if (error == 0) newpgrp = NULL; } else error = EPERM; } else { if (pgrp == targp->p_pgrp) { PGRP_UNLOCK(pgrp); goto done; } if (pgrp->pg_id != targp->p_pid && pgrp->pg_session != curp->p_session) { PGRP_UNLOCK(pgrp); error = EPERM; goto done; } PGRP_UNLOCK(pgrp); error = enterthispgrp(targp, pgrp); } done: sx_xunlock(&proctree_lock); KASSERT((error == 0) || (newpgrp != NULL), ("setpgid failed and newpgrp is NULL")); if (newpgrp != NULL) free(newpgrp, M_PGRP); return (error); } /* * Use the clause in B.4.2.2 that allows setuid/setgid to be 4.2/4.3BSD * compatible. It says that setting the uid/gid to euid/egid is a special * case of "appropriate privilege". Once the rules are expanded out, this * basically means that setuid(nnn) sets all three id's, in all permitted * cases unless _POSIX_SAVED_IDS is enabled. In that case, setuid(getuid()) * does not set the saved id - this is dangerous for traditional BSD * programs. For this reason, we *really* do not want to set * _POSIX_SAVED_IDS and do not want to clear POSIX_APPENDIX_B_4_2_2. */ #define POSIX_APPENDIX_B_4_2_2 #ifndef _SYS_SYSPROTO_H_ struct setuid_args { uid_t uid; }; #endif /* ARGSUSED */ int sys_setuid(struct thread *td, struct setuid_args *uap) { struct proc *p = td->td_proc; struct ucred *newcred, *oldcred; uid_t uid; struct uidinfo *uip; int error; uid = uap->uid; AUDIT_ARG_UID(uid); newcred = crget(); uip = uifind(uid); PROC_LOCK(p); /* * Copy credentials so other references do not see our changes. */ oldcred = crcopysafe(p, newcred); #ifdef MAC error = mac_cred_check_setuid(oldcred, uid); if (error) goto fail; #endif /* * See if we have "permission" by POSIX 1003.1 rules. * * Note that setuid(geteuid()) is a special case of * "appropriate privileges" in appendix B.4.2.2. We need * to use this clause to be compatible with traditional BSD * semantics. Basically, it means that "setuid(xx)" sets all * three id's (assuming you have privs). * * Notes on the logic. We do things in three steps. * 1: We determine if the euid is going to change, and do EPERM * right away. We unconditionally change the euid later if this * test is satisfied, simplifying that part of the logic. * 2: We determine if the real and/or saved uids are going to * change. Determined by compile options. * 3: Change euid last. (after tests in #2 for "appropriate privs") */ if (uid != oldcred->cr_ruid && /* allow setuid(getuid()) */ #ifdef _POSIX_SAVED_IDS uid != oldcred->cr_svuid && /* allow setuid(saved gid) */ #endif #ifdef POSIX_APPENDIX_B_4_2_2 /* Use BSD-compat clause from B.4.2.2 */ uid != oldcred->cr_uid && /* allow setuid(geteuid()) */ #endif (error = priv_check_cred(oldcred, PRIV_CRED_SETUID, 0)) != 0) goto fail; #ifdef _POSIX_SAVED_IDS /* * Do we have "appropriate privileges" (are we root or uid == euid) * If so, we are changing the real uid and/or saved uid. */ if ( #ifdef POSIX_APPENDIX_B_4_2_2 /* Use the clause from B.4.2.2 */ uid == oldcred->cr_uid || #endif /* We are using privs. */ priv_check_cred(oldcred, PRIV_CRED_SETUID, 0) == 0) #endif { /* * Set the real uid and transfer proc count to new user. */ if (uid != oldcred->cr_ruid) { change_ruid(newcred, uip); setsugid(p); } /* * Set saved uid * * XXX always set saved uid even if not _POSIX_SAVED_IDS, as * the security of seteuid() depends on it. B.4.2.2 says it * is important that we should do this. */ if (uid != oldcred->cr_svuid) { change_svuid(newcred, uid); setsugid(p); } } /* * In all permitted cases, we are changing the euid. */ if (uid != oldcred->cr_uid) { change_euid(newcred, uip); setsugid(p); } proc_set_cred(p, newcred); PROC_UNLOCK(p); #ifdef RACCT racct_proc_ucred_changed(p, oldcred, newcred); #endif uifree(uip); crfree(oldcred); return (0); fail: PROC_UNLOCK(p); uifree(uip); crfree(newcred); return (error); } #ifndef _SYS_SYSPROTO_H_ struct seteuid_args { uid_t euid; }; #endif /* ARGSUSED */ int sys_seteuid(struct thread *td, struct seteuid_args *uap) { struct proc *p = td->td_proc; struct ucred *newcred, *oldcred; uid_t euid; struct uidinfo *euip; int error; euid = uap->euid; AUDIT_ARG_EUID(euid); newcred = crget(); euip = uifind(euid); PROC_LOCK(p); /* * Copy credentials so other references do not see our changes. */ oldcred = crcopysafe(p, newcred); #ifdef MAC error = mac_cred_check_seteuid(oldcred, euid); if (error) goto fail; #endif if (euid != oldcred->cr_ruid && /* allow seteuid(getuid()) */ euid != oldcred->cr_svuid && /* allow seteuid(saved uid) */ (error = priv_check_cred(oldcred, PRIV_CRED_SETEUID, 0)) != 0) goto fail; /* * Everything's okay, do it. */ if (oldcred->cr_uid != euid) { change_euid(newcred, euip); setsugid(p); } proc_set_cred(p, newcred); PROC_UNLOCK(p); uifree(euip); crfree(oldcred); return (0); fail: PROC_UNLOCK(p); uifree(euip); crfree(newcred); return (error); } #ifndef _SYS_SYSPROTO_H_ struct setgid_args { gid_t gid; }; #endif /* ARGSUSED */ int sys_setgid(struct thread *td, struct setgid_args *uap) { struct proc *p = td->td_proc; struct ucred *newcred, *oldcred; gid_t gid; int error; gid = uap->gid; AUDIT_ARG_GID(gid); newcred = crget(); PROC_LOCK(p); oldcred = crcopysafe(p, newcred); #ifdef MAC error = mac_cred_check_setgid(oldcred, gid); if (error) goto fail; #endif /* * See if we have "permission" by POSIX 1003.1 rules. * * Note that setgid(getegid()) is a special case of * "appropriate privileges" in appendix B.4.2.2. We need * to use this clause to be compatible with traditional BSD * semantics. Basically, it means that "setgid(xx)" sets all * three id's (assuming you have privs). * * For notes on the logic here, see setuid() above. */ if (gid != oldcred->cr_rgid && /* allow setgid(getgid()) */ #ifdef _POSIX_SAVED_IDS gid != oldcred->cr_svgid && /* allow setgid(saved gid) */ #endif #ifdef POSIX_APPENDIX_B_4_2_2 /* Use BSD-compat clause from B.4.2.2 */ gid != oldcred->cr_groups[0] && /* allow setgid(getegid()) */ #endif (error = priv_check_cred(oldcred, PRIV_CRED_SETGID, 0)) != 0) goto fail; #ifdef _POSIX_SAVED_IDS /* * Do we have "appropriate privileges" (are we root or gid == egid) * If so, we are changing the real uid and saved gid. */ if ( #ifdef POSIX_APPENDIX_B_4_2_2 /* use the clause from B.4.2.2 */ gid == oldcred->cr_groups[0] || #endif /* We are using privs. */ priv_check_cred(oldcred, PRIV_CRED_SETGID, 0) == 0) #endif { /* * Set real gid */ if (oldcred->cr_rgid != gid) { change_rgid(newcred, gid); setsugid(p); } /* * Set saved gid * * XXX always set saved gid even if not _POSIX_SAVED_IDS, as * the security of setegid() depends on it. B.4.2.2 says it * is important that we should do this. */ if (oldcred->cr_svgid != gid) { change_svgid(newcred, gid); setsugid(p); } } /* * In all cases permitted cases, we are changing the egid. * Copy credentials so other references do not see our changes. */ if (oldcred->cr_groups[0] != gid) { change_egid(newcred, gid); setsugid(p); } proc_set_cred(p, newcred); PROC_UNLOCK(p); crfree(oldcred); return (0); fail: PROC_UNLOCK(p); crfree(newcred); return (error); } #ifndef _SYS_SYSPROTO_H_ struct setegid_args { gid_t egid; }; #endif /* ARGSUSED */ int sys_setegid(struct thread *td, struct setegid_args *uap) { struct proc *p = td->td_proc; struct ucred *newcred, *oldcred; gid_t egid; int error; egid = uap->egid; AUDIT_ARG_EGID(egid); newcred = crget(); PROC_LOCK(p); oldcred = crcopysafe(p, newcred); #ifdef MAC error = mac_cred_check_setegid(oldcred, egid); if (error) goto fail; #endif if (egid != oldcred->cr_rgid && /* allow setegid(getgid()) */ egid != oldcred->cr_svgid && /* allow setegid(saved gid) */ (error = priv_check_cred(oldcred, PRIV_CRED_SETEGID, 0)) != 0) goto fail; if (oldcred->cr_groups[0] != egid) { change_egid(newcred, egid); setsugid(p); } proc_set_cred(p, newcred); PROC_UNLOCK(p); crfree(oldcred); return (0); fail: PROC_UNLOCK(p); crfree(newcred); return (error); } #ifndef _SYS_SYSPROTO_H_ struct setgroups_args { u_int gidsetsize; gid_t *gidset; }; #endif /* ARGSUSED */ int sys_setgroups(struct thread *td, struct setgroups_args *uap) { gid_t smallgroups[XU_NGROUPS]; gid_t *groups; u_int gidsetsize; int error; gidsetsize = uap->gidsetsize; if (gidsetsize > ngroups_max + 1) return (EINVAL); if (gidsetsize > XU_NGROUPS) groups = malloc(gidsetsize * sizeof(gid_t), M_TEMP, M_WAITOK); else groups = smallgroups; error = copyin(uap->gidset, groups, gidsetsize * sizeof(gid_t)); if (error == 0) error = kern_setgroups(td, gidsetsize, groups); if (gidsetsize > XU_NGROUPS) free(groups, M_TEMP); return (error); } int kern_setgroups(struct thread *td, u_int ngrp, gid_t *groups) { struct proc *p = td->td_proc; struct ucred *newcred, *oldcred; int error; MPASS(ngrp <= ngroups_max + 1); AUDIT_ARG_GROUPSET(groups, ngrp); newcred = crget(); crextend(newcred, ngrp); PROC_LOCK(p); oldcred = crcopysafe(p, newcred); #ifdef MAC error = mac_cred_check_setgroups(oldcred, ngrp, groups); if (error) goto fail; #endif error = priv_check_cred(oldcred, PRIV_CRED_SETGROUPS, 0); if (error) goto fail; if (ngrp == 0) { /* * setgroups(0, NULL) is a legitimate way of clearing the * groups vector on non-BSD systems (which generally do not * have the egid in the groups[0]). We risk security holes * when running non-BSD software if we do not do the same. */ newcred->cr_ngroups = 1; } else { crsetgroups_locked(newcred, ngrp, groups); } setsugid(p); proc_set_cred(p, newcred); PROC_UNLOCK(p); crfree(oldcred); return (0); fail: PROC_UNLOCK(p); crfree(newcred); return (error); } #ifndef _SYS_SYSPROTO_H_ struct setreuid_args { uid_t ruid; uid_t euid; }; #endif /* ARGSUSED */ int sys_setreuid(register struct thread *td, struct setreuid_args *uap) { struct proc *p = td->td_proc; struct ucred *newcred, *oldcred; uid_t euid, ruid; struct uidinfo *euip, *ruip; int error; euid = uap->euid; ruid = uap->ruid; AUDIT_ARG_EUID(euid); AUDIT_ARG_RUID(ruid); newcred = crget(); euip = uifind(euid); ruip = uifind(ruid); PROC_LOCK(p); oldcred = crcopysafe(p, newcred); #ifdef MAC error = mac_cred_check_setreuid(oldcred, ruid, euid); if (error) goto fail; #endif if (((ruid != (uid_t)-1 && ruid != oldcred->cr_ruid && ruid != oldcred->cr_svuid) || (euid != (uid_t)-1 && euid != oldcred->cr_uid && euid != oldcred->cr_ruid && euid != oldcred->cr_svuid)) && (error = priv_check_cred(oldcred, PRIV_CRED_SETREUID, 0)) != 0) goto fail; if (euid != (uid_t)-1 && oldcred->cr_uid != euid) { change_euid(newcred, euip); setsugid(p); } if (ruid != (uid_t)-1 && oldcred->cr_ruid != ruid) { change_ruid(newcred, ruip); setsugid(p); } if ((ruid != (uid_t)-1 || newcred->cr_uid != newcred->cr_ruid) && newcred->cr_svuid != newcred->cr_uid) { change_svuid(newcred, newcred->cr_uid); setsugid(p); } proc_set_cred(p, newcred); PROC_UNLOCK(p); #ifdef RACCT racct_proc_ucred_changed(p, oldcred, newcred); #endif uifree(ruip); uifree(euip); crfree(oldcred); return (0); fail: PROC_UNLOCK(p); uifree(ruip); uifree(euip); crfree(newcred); return (error); } #ifndef _SYS_SYSPROTO_H_ struct setregid_args { gid_t rgid; gid_t egid; }; #endif /* ARGSUSED */ int sys_setregid(register struct thread *td, struct setregid_args *uap) { struct proc *p = td->td_proc; struct ucred *newcred, *oldcred; gid_t egid, rgid; int error; egid = uap->egid; rgid = uap->rgid; AUDIT_ARG_EGID(egid); AUDIT_ARG_RGID(rgid); newcred = crget(); PROC_LOCK(p); oldcred = crcopysafe(p, newcred); #ifdef MAC error = mac_cred_check_setregid(oldcred, rgid, egid); if (error) goto fail; #endif if (((rgid != (gid_t)-1 && rgid != oldcred->cr_rgid && rgid != oldcred->cr_svgid) || (egid != (gid_t)-1 && egid != oldcred->cr_groups[0] && egid != oldcred->cr_rgid && egid != oldcred->cr_svgid)) && (error = priv_check_cred(oldcred, PRIV_CRED_SETREGID, 0)) != 0) goto fail; if (egid != (gid_t)-1 && oldcred->cr_groups[0] != egid) { change_egid(newcred, egid); setsugid(p); } if (rgid != (gid_t)-1 && oldcred->cr_rgid != rgid) { change_rgid(newcred, rgid); setsugid(p); } if ((rgid != (gid_t)-1 || newcred->cr_groups[0] != newcred->cr_rgid) && newcred->cr_svgid != newcred->cr_groups[0]) { change_svgid(newcred, newcred->cr_groups[0]); setsugid(p); } proc_set_cred(p, newcred); PROC_UNLOCK(p); crfree(oldcred); return (0); fail: PROC_UNLOCK(p); crfree(newcred); return (error); } /* * setresuid(ruid, euid, suid) is like setreuid except control over the saved * uid is explicit. */ #ifndef _SYS_SYSPROTO_H_ struct setresuid_args { uid_t ruid; uid_t euid; uid_t suid; }; #endif /* ARGSUSED */ int sys_setresuid(register struct thread *td, struct setresuid_args *uap) { struct proc *p = td->td_proc; struct ucred *newcred, *oldcred; uid_t euid, ruid, suid; struct uidinfo *euip, *ruip; int error; euid = uap->euid; ruid = uap->ruid; suid = uap->suid; AUDIT_ARG_EUID(euid); AUDIT_ARG_RUID(ruid); AUDIT_ARG_SUID(suid); newcred = crget(); euip = uifind(euid); ruip = uifind(ruid); PROC_LOCK(p); oldcred = crcopysafe(p, newcred); #ifdef MAC error = mac_cred_check_setresuid(oldcred, ruid, euid, suid); if (error) goto fail; #endif if (((ruid != (uid_t)-1 && ruid != oldcred->cr_ruid && ruid != oldcred->cr_svuid && ruid != oldcred->cr_uid) || (euid != (uid_t)-1 && euid != oldcred->cr_ruid && euid != oldcred->cr_svuid && euid != oldcred->cr_uid) || (suid != (uid_t)-1 && suid != oldcred->cr_ruid && suid != oldcred->cr_svuid && suid != oldcred->cr_uid)) && (error = priv_check_cred(oldcred, PRIV_CRED_SETRESUID, 0)) != 0) goto fail; if (euid != (uid_t)-1 && oldcred->cr_uid != euid) { change_euid(newcred, euip); setsugid(p); } if (ruid != (uid_t)-1 && oldcred->cr_ruid != ruid) { change_ruid(newcred, ruip); setsugid(p); } if (suid != (uid_t)-1 && oldcred->cr_svuid != suid) { change_svuid(newcred, suid); setsugid(p); } proc_set_cred(p, newcred); PROC_UNLOCK(p); #ifdef RACCT racct_proc_ucred_changed(p, oldcred, newcred); #endif uifree(ruip); uifree(euip); crfree(oldcred); return (0); fail: PROC_UNLOCK(p); uifree(ruip); uifree(euip); crfree(newcred); return (error); } /* * setresgid(rgid, egid, sgid) is like setregid except control over the saved * gid is explicit. */ #ifndef _SYS_SYSPROTO_H_ struct setresgid_args { gid_t rgid; gid_t egid; gid_t sgid; }; #endif /* ARGSUSED */ int sys_setresgid(register struct thread *td, struct setresgid_args *uap) { struct proc *p = td->td_proc; struct ucred *newcred, *oldcred; gid_t egid, rgid, sgid; int error; egid = uap->egid; rgid = uap->rgid; sgid = uap->sgid; AUDIT_ARG_EGID(egid); AUDIT_ARG_RGID(rgid); AUDIT_ARG_SGID(sgid); newcred = crget(); PROC_LOCK(p); oldcred = crcopysafe(p, newcred); #ifdef MAC error = mac_cred_check_setresgid(oldcred, rgid, egid, sgid); if (error) goto fail; #endif if (((rgid != (gid_t)-1 && rgid != oldcred->cr_rgid && rgid != oldcred->cr_svgid && rgid != oldcred->cr_groups[0]) || (egid != (gid_t)-1 && egid != oldcred->cr_rgid && egid != oldcred->cr_svgid && egid != oldcred->cr_groups[0]) || (sgid != (gid_t)-1 && sgid != oldcred->cr_rgid && sgid != oldcred->cr_svgid && sgid != oldcred->cr_groups[0])) && (error = priv_check_cred(oldcred, PRIV_CRED_SETRESGID, 0)) != 0) goto fail; if (egid != (gid_t)-1 && oldcred->cr_groups[0] != egid) { change_egid(newcred, egid); setsugid(p); } if (rgid != (gid_t)-1 && oldcred->cr_rgid != rgid) { change_rgid(newcred, rgid); setsugid(p); } if (sgid != (gid_t)-1 && oldcred->cr_svgid != sgid) { change_svgid(newcred, sgid); setsugid(p); } proc_set_cred(p, newcred); PROC_UNLOCK(p); crfree(oldcred); return (0); fail: PROC_UNLOCK(p); crfree(newcred); return (error); } #ifndef _SYS_SYSPROTO_H_ struct getresuid_args { uid_t *ruid; uid_t *euid; uid_t *suid; }; #endif /* ARGSUSED */ int sys_getresuid(register struct thread *td, struct getresuid_args *uap) { struct ucred *cred; int error1 = 0, error2 = 0, error3 = 0; cred = td->td_ucred; if (uap->ruid) error1 = copyout(&cred->cr_ruid, uap->ruid, sizeof(cred->cr_ruid)); if (uap->euid) error2 = copyout(&cred->cr_uid, uap->euid, sizeof(cred->cr_uid)); if (uap->suid) error3 = copyout(&cred->cr_svuid, uap->suid, sizeof(cred->cr_svuid)); return (error1 ? error1 : error2 ? error2 : error3); } #ifndef _SYS_SYSPROTO_H_ struct getresgid_args { gid_t *rgid; gid_t *egid; gid_t *sgid; }; #endif /* ARGSUSED */ int sys_getresgid(register struct thread *td, struct getresgid_args *uap) { struct ucred *cred; int error1 = 0, error2 = 0, error3 = 0; cred = td->td_ucred; if (uap->rgid) error1 = copyout(&cred->cr_rgid, uap->rgid, sizeof(cred->cr_rgid)); if (uap->egid) error2 = copyout(&cred->cr_groups[0], uap->egid, sizeof(cred->cr_groups[0])); if (uap->sgid) error3 = copyout(&cred->cr_svgid, uap->sgid, sizeof(cred->cr_svgid)); return (error1 ? error1 : error2 ? error2 : error3); } #ifndef _SYS_SYSPROTO_H_ struct issetugid_args { int dummy; }; #endif /* ARGSUSED */ int sys_issetugid(register struct thread *td, struct issetugid_args *uap) { struct proc *p = td->td_proc; /* * Note: OpenBSD sets a P_SUGIDEXEC flag set at execve() time, * we use P_SUGID because we consider changing the owners as * "tainting" as well. * This is significant for procs that start as root and "become" * a user without an exec - programs cannot know *everything* * that libc *might* have put in their data segment. */ PROC_LOCK(p); td->td_retval[0] = (p->p_flag & P_SUGID) ? 1 : 0; PROC_UNLOCK(p); return (0); } int sys___setugid(struct thread *td, struct __setugid_args *uap) { #ifdef REGRESSION struct proc *p; p = td->td_proc; switch (uap->flag) { case 0: PROC_LOCK(p); p->p_flag &= ~P_SUGID; PROC_UNLOCK(p); return (0); case 1: PROC_LOCK(p); p->p_flag |= P_SUGID; PROC_UNLOCK(p); return (0); default: return (EINVAL); } #else /* !REGRESSION */ return (ENOSYS); #endif /* REGRESSION */ } /* * Check if gid is a member of the group set. */ int groupmember(gid_t gid, struct ucred *cred) { int l; int h; int m; if (cred->cr_groups[0] == gid) return(1); /* * If gid was not our primary group, perform a binary search * of the supplemental groups. This is possible because we * sort the groups in crsetgroups(). */ l = 1; h = cred->cr_ngroups; while (l < h) { m = l + ((h - l) / 2); if (cred->cr_groups[m] < gid) l = m + 1; else h = m; } if ((l < cred->cr_ngroups) && (cred->cr_groups[l] == gid)) return (1); return (0); } /* * Test the active securelevel against a given level. securelevel_gt() * implements (securelevel > level). securelevel_ge() implements * (securelevel >= level). Note that the logic is inverted -- these * functions return EPERM on "success" and 0 on "failure". * * Due to care taken when setting the securelevel, we know that no jail will * be less secure that its parent (or the physical system), so it is sufficient * to test the current jail only. * * XXXRW: Possibly since this has to do with privilege, it should move to * kern_priv.c. */ int securelevel_gt(struct ucred *cr, int level) { return (cr->cr_prison->pr_securelevel > level ? EPERM : 0); } int securelevel_ge(struct ucred *cr, int level) { return (cr->cr_prison->pr_securelevel >= level ? EPERM : 0); } /* * 'see_other_uids' determines whether or not visibility of processes * and sockets with credentials holding different real uids is possible * using a variety of system MIBs. * XXX: data declarations should be together near the beginning of the file. */ static int see_other_uids = 1; SYSCTL_INT(_security_bsd, OID_AUTO, see_other_uids, CTLFLAG_RW, &see_other_uids, 0, "Unprivileged processes may see subjects/objects with different real uid"); /*- * Determine if u1 "can see" the subject specified by u2, according to the * 'see_other_uids' policy. * Returns: 0 for permitted, ESRCH otherwise * Locks: none * References: *u1 and *u2 must not change during the call * u1 may equal u2, in which case only one reference is required */ int cr_canseeotheruids(struct ucred *u1, struct ucred *u2) { if (!see_other_uids && u1->cr_ruid != u2->cr_ruid) { if (priv_check_cred(u1, PRIV_SEEOTHERUIDS, 0) != 0) return (ESRCH); } return (0); } /* * 'see_other_gids' determines whether or not visibility of processes * and sockets with credentials holding different real gids is possible * using a variety of system MIBs. * XXX: data declarations should be together near the beginning of the file. */ static int see_other_gids = 1; SYSCTL_INT(_security_bsd, OID_AUTO, see_other_gids, CTLFLAG_RW, &see_other_gids, 0, "Unprivileged processes may see subjects/objects with different real gid"); /* * Determine if u1 can "see" the subject specified by u2, according to the * 'see_other_gids' policy. * Returns: 0 for permitted, ESRCH otherwise * Locks: none * References: *u1 and *u2 must not change during the call * u1 may equal u2, in which case only one reference is required */ int cr_canseeothergids(struct ucred *u1, struct ucred *u2) { int i, match; if (!see_other_gids) { match = 0; for (i = 0; i < u1->cr_ngroups; i++) { if (groupmember(u1->cr_groups[i], u2)) match = 1; if (match) break; } if (!match) { if (priv_check_cred(u1, PRIV_SEEOTHERGIDS, 0) != 0) return (ESRCH); } } return (0); } /*- * Determine if u1 "can see" the subject specified by u2. * Returns: 0 for permitted, an errno value otherwise * Locks: none * References: *u1 and *u2 must not change during the call * u1 may equal u2, in which case only one reference is required */ int cr_cansee(struct ucred *u1, struct ucred *u2) { int error; if ((error = prison_check(u1, u2))) return (error); #ifdef MAC if ((error = mac_cred_check_visible(u1, u2))) return (error); #endif if ((error = cr_canseeotheruids(u1, u2))) return (error); if ((error = cr_canseeothergids(u1, u2))) return (error); return (0); } /*- * Determine if td "can see" the subject specified by p. * Returns: 0 for permitted, an errno value otherwise * Locks: Sufficient locks to protect p->p_ucred must be held. td really * should be curthread. * References: td and p must be valid for the lifetime of the call */ int p_cansee(struct thread *td, struct proc *p) { /* Wrap cr_cansee() for all functionality. */ KASSERT(td == curthread, ("%s: td not curthread", __func__)); PROC_LOCK_ASSERT(p, MA_OWNED); return (cr_cansee(td->td_ucred, p->p_ucred)); } /* * 'conservative_signals' prevents the delivery of a broad class of * signals by unprivileged processes to processes that have changed their * credentials since the last invocation of execve(). This can prevent * the leakage of cached information or retained privileges as a result * of a common class of signal-related vulnerabilities. However, this * may interfere with some applications that expect to be able to * deliver these signals to peer processes after having given up * privilege. */ static int conservative_signals = 1; SYSCTL_INT(_security_bsd, OID_AUTO, conservative_signals, CTLFLAG_RW, &conservative_signals, 0, "Unprivileged processes prevented from " "sending certain signals to processes whose credentials have changed"); /*- * Determine whether cred may deliver the specified signal to proc. * Returns: 0 for permitted, an errno value otherwise. * Locks: A lock must be held for proc. * References: cred and proc must be valid for the lifetime of the call. */ int cr_cansignal(struct ucred *cred, struct proc *proc, int signum) { int error; PROC_LOCK_ASSERT(proc, MA_OWNED); /* * Jail semantics limit the scope of signalling to proc in the * same jail as cred, if cred is in jail. */ error = prison_check(cred, proc->p_ucred); if (error) return (error); #ifdef MAC if ((error = mac_proc_check_signal(cred, proc, signum))) return (error); #endif if ((error = cr_canseeotheruids(cred, proc->p_ucred))) return (error); if ((error = cr_canseeothergids(cred, proc->p_ucred))) return (error); /* * UNIX signal semantics depend on the status of the P_SUGID * bit on the target process. If the bit is set, then additional * restrictions are placed on the set of available signals. */ if (conservative_signals && (proc->p_flag & P_SUGID)) { switch (signum) { case 0: case SIGKILL: case SIGINT: case SIGTERM: case SIGALRM: case SIGSTOP: case SIGTTIN: case SIGTTOU: case SIGTSTP: case SIGHUP: case SIGUSR1: case SIGUSR2: /* * Generally, permit job and terminal control * signals. */ break; default: /* Not permitted without privilege. */ error = priv_check_cred(cred, PRIV_SIGNAL_SUGID, 0); if (error) return (error); } } /* * Generally, the target credential's ruid or svuid must match the * subject credential's ruid or euid. */ if (cred->cr_ruid != proc->p_ucred->cr_ruid && cred->cr_ruid != proc->p_ucred->cr_svuid && cred->cr_uid != proc->p_ucred->cr_ruid && cred->cr_uid != proc->p_ucred->cr_svuid) { error = priv_check_cred(cred, PRIV_SIGNAL_DIFFCRED, 0); if (error) return (error); } return (0); } /*- * Determine whether td may deliver the specified signal to p. * Returns: 0 for permitted, an errno value otherwise * Locks: Sufficient locks to protect various components of td and p * must be held. td must be curthread, and a lock must be * held for p. * References: td and p must be valid for the lifetime of the call */ int p_cansignal(struct thread *td, struct proc *p, int signum) { KASSERT(td == curthread, ("%s: td not curthread", __func__)); PROC_LOCK_ASSERT(p, MA_OWNED); if (td->td_proc == p) return (0); /* * UNIX signalling semantics require that processes in the same * session always be able to deliver SIGCONT to one another, * overriding the remaining protections. */ /* XXX: This will require an additional lock of some sort. */ if (signum == SIGCONT && td->td_proc->p_session == p->p_session) return (0); /* * Some compat layers use SIGTHR and higher signals for * communication between different kernel threads of the same * process, so that they expect that it's always possible to * deliver them, even for suid applications where cr_cansignal() can * deny such ability for security consideration. It should be * pretty safe to do since the only way to create two processes * with the same p_leader is via rfork(2). */ if (td->td_proc->p_leader != NULL && signum >= SIGTHR && signum < SIGTHR + 4 && td->td_proc->p_leader == p->p_leader) return (0); return (cr_cansignal(td->td_ucred, p, signum)); } /*- * Determine whether td may reschedule p. * Returns: 0 for permitted, an errno value otherwise * Locks: Sufficient locks to protect various components of td and p * must be held. td must be curthread, and a lock must * be held for p. * References: td and p must be valid for the lifetime of the call */ int p_cansched(struct thread *td, struct proc *p) { int error; KASSERT(td == curthread, ("%s: td not curthread", __func__)); PROC_LOCK_ASSERT(p, MA_OWNED); if (td->td_proc == p) return (0); if ((error = prison_check(td->td_ucred, p->p_ucred))) return (error); #ifdef MAC if ((error = mac_proc_check_sched(td->td_ucred, p))) return (error); #endif if ((error = cr_canseeotheruids(td->td_ucred, p->p_ucred))) return (error); if ((error = cr_canseeothergids(td->td_ucred, p->p_ucred))) return (error); if (td->td_ucred->cr_ruid != p->p_ucred->cr_ruid && td->td_ucred->cr_uid != p->p_ucred->cr_ruid) { error = priv_check(td, PRIV_SCHED_DIFFCRED); if (error) return (error); } return (0); } /* * The 'unprivileged_proc_debug' flag may be used to disable a variety of * unprivileged inter-process debugging services, including some procfs * functionality, ptrace(), and ktrace(). In the past, inter-process * debugging has been involved in a variety of security problems, and sites * not requiring the service might choose to disable it when hardening * systems. * * XXX: Should modifying and reading this variable require locking? * XXX: data declarations should be together near the beginning of the file. */ static int unprivileged_proc_debug = 1; SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_proc_debug, CTLFLAG_RW, &unprivileged_proc_debug, 0, "Unprivileged processes may use process debugging facilities"); /*- * Determine whether td may debug p. * Returns: 0 for permitted, an errno value otherwise * Locks: Sufficient locks to protect various components of td and p * must be held. td must be curthread, and a lock must * be held for p. * References: td and p must be valid for the lifetime of the call */ int p_candebug(struct thread *td, struct proc *p) { int credentialchanged, error, grpsubset, i, uidsubset; KASSERT(td == curthread, ("%s: td not curthread", __func__)); PROC_LOCK_ASSERT(p, MA_OWNED); if (!unprivileged_proc_debug) { error = priv_check(td, PRIV_DEBUG_UNPRIV); if (error) return (error); } if (td->td_proc == p) return (0); if ((error = prison_check(td->td_ucred, p->p_ucred))) return (error); #ifdef MAC if ((error = mac_proc_check_debug(td->td_ucred, p))) return (error); #endif if ((error = cr_canseeotheruids(td->td_ucred, p->p_ucred))) return (error); if ((error = cr_canseeothergids(td->td_ucred, p->p_ucred))) return (error); /* * Is p's group set a subset of td's effective group set? This * includes p's egid, group access list, rgid, and svgid. */ grpsubset = 1; for (i = 0; i < p->p_ucred->cr_ngroups; i++) { if (!groupmember(p->p_ucred->cr_groups[i], td->td_ucred)) { grpsubset = 0; break; } } grpsubset = grpsubset && groupmember(p->p_ucred->cr_rgid, td->td_ucred) && groupmember(p->p_ucred->cr_svgid, td->td_ucred); /* * Are the uids present in p's credential equal to td's * effective uid? This includes p's euid, svuid, and ruid. */ uidsubset = (td->td_ucred->cr_uid == p->p_ucred->cr_uid && td->td_ucred->cr_uid == p->p_ucred->cr_svuid && td->td_ucred->cr_uid == p->p_ucred->cr_ruid); /* * Has the credential of the process changed since the last exec()? */ credentialchanged = (p->p_flag & P_SUGID); /* * If p's gids aren't a subset, or the uids aren't a subset, * or the credential has changed, require appropriate privilege * for td to debug p. */ if (!grpsubset || !uidsubset) { error = priv_check(td, PRIV_DEBUG_DIFFCRED); if (error) return (error); } if (credentialchanged) { error = priv_check(td, PRIV_DEBUG_SUGID); if (error) return (error); } /* Can't trace init when securelevel > 0. */ if (p == initproc) { error = securelevel_gt(td->td_ucred, 0); if (error) return (error); } /* * Can't trace a process that's currently exec'ing. * * XXX: Note, this is not a security policy decision, it's a * basic correctness/functionality decision. Therefore, this check * should be moved to the caller's of p_candebug(). */ if ((p->p_flag & P_INEXEC) != 0) return (EBUSY); /* Denied explicitely */ if ((p->p_flag2 & P2_NOTRACE) != 0) { error = priv_check(td, PRIV_DEBUG_DENIED); if (error != 0) return (error); } return (0); } /*- * Determine whether the subject represented by cred can "see" a socket. * Returns: 0 for permitted, ENOENT otherwise. */ int cr_canseesocket(struct ucred *cred, struct socket *so) { int error; error = prison_check(cred, so->so_cred); if (error) return (ENOENT); #ifdef MAC error = mac_socket_check_visible(cred, so); if (error) return (error); #endif if (cr_canseeotheruids(cred, so->so_cred)) return (ENOENT); if (cr_canseeothergids(cred, so->so_cred)) return (ENOENT); return (0); } /*- * Determine whether td can wait for the exit of p. * Returns: 0 for permitted, an errno value otherwise * Locks: Sufficient locks to protect various components of td and p * must be held. td must be curthread, and a lock must * be held for p. * References: td and p must be valid for the lifetime of the call */ int p_canwait(struct thread *td, struct proc *p) { int error; KASSERT(td == curthread, ("%s: td not curthread", __func__)); PROC_LOCK_ASSERT(p, MA_OWNED); if ((error = prison_check(td->td_ucred, p->p_ucred))) return (error); #ifdef MAC if ((error = mac_proc_check_wait(td->td_ucred, p))) return (error); #endif #if 0 /* XXXMAC: This could have odd effects on some shells. */ if ((error = cr_canseeotheruids(td->td_ucred, p->p_ucred))) return (error); #endif return (0); } /* * Allocate a zeroed cred structure. */ struct ucred * crget(void) { register struct ucred *cr; cr = malloc(sizeof(*cr), M_CRED, M_WAITOK | M_ZERO); refcount_init(&cr->cr_ref, 1); #ifdef AUDIT audit_cred_init(cr); #endif #ifdef MAC mac_cred_init(cr); #endif cr->cr_groups = cr->cr_smallgroups; cr->cr_agroups = sizeof(cr->cr_smallgroups) / sizeof(cr->cr_smallgroups[0]); return (cr); } /* * Claim another reference to a ucred structure. */ struct ucred * crhold(struct ucred *cr) { refcount_acquire(&cr->cr_ref); return (cr); } /* * Free a cred structure. Throws away space when ref count gets to 0. */ void crfree(struct ucred *cr) { KASSERT(cr->cr_ref > 0, ("bad ucred refcount: %d", cr->cr_ref)); KASSERT(cr->cr_ref != 0xdeadc0de, ("dangling reference to ucred")); if (refcount_release(&cr->cr_ref)) { /* * Some callers of crget(), such as nfs_statfs(), * allocate a temporary credential, but don't * allocate a uidinfo structure. */ if (cr->cr_uidinfo != NULL) uifree(cr->cr_uidinfo); if (cr->cr_ruidinfo != NULL) uifree(cr->cr_ruidinfo); /* * Free a prison, if any. */ if (cr->cr_prison != NULL) prison_free(cr->cr_prison); if (cr->cr_loginclass != NULL) loginclass_free(cr->cr_loginclass); #ifdef AUDIT audit_cred_destroy(cr); #endif #ifdef MAC mac_cred_destroy(cr); #endif if (cr->cr_groups != cr->cr_smallgroups) free(cr->cr_groups, M_CRED); free(cr, M_CRED); } } /* * Copy a ucred's contents from a template. Does not block. */ void crcopy(struct ucred *dest, struct ucred *src) { KASSERT(dest->cr_ref == 1, ("crcopy of shared ucred")); bcopy(&src->cr_startcopy, &dest->cr_startcopy, (unsigned)((caddr_t)&src->cr_endcopy - (caddr_t)&src->cr_startcopy)); crsetgroups(dest, src->cr_ngroups, src->cr_groups); uihold(dest->cr_uidinfo); uihold(dest->cr_ruidinfo); prison_hold(dest->cr_prison); loginclass_hold(dest->cr_loginclass); #ifdef AUDIT audit_cred_copy(src, dest); #endif #ifdef MAC mac_cred_copy(src, dest); #endif } /* * Dup cred struct to a new held one. */ struct ucred * crdup(struct ucred *cr) { struct ucred *newcr; newcr = crget(); crcopy(newcr, cr); return (newcr); } /* * Fill in a struct xucred based on a struct ucred. */ void cru2x(struct ucred *cr, struct xucred *xcr) { int ngroups; bzero(xcr, sizeof(*xcr)); xcr->cr_version = XUCRED_VERSION; xcr->cr_uid = cr->cr_uid; ngroups = MIN(cr->cr_ngroups, XU_NGROUPS); xcr->cr_ngroups = ngroups; bcopy(cr->cr_groups, xcr->cr_groups, ngroups * sizeof(*cr->cr_groups)); } /* * Set initial process credentials. * Callers are responsible for providing the reference for provided credentials. */ void proc_set_cred_init(struct proc *p, struct ucred *newcred) { p->p_ucred = newcred; } /* * Change process credentials. * Callers are responsible for providing the reference for passed credentials * and for freeing old ones. * * Process has to be locked except when it does not have credentials (as it * should not be visible just yet) or when newcred is NULL (as this can be * only used when the process is about to be freed, at which point it should * not be visible anymore). */ struct ucred * proc_set_cred(struct proc *p, struct ucred *newcred) { struct ucred *oldcred; MPASS(p->p_ucred != NULL); if (newcred == NULL) MPASS(p->p_state == PRS_ZOMBIE); else PROC_LOCK_ASSERT(p, MA_OWNED); oldcred = p->p_ucred; p->p_ucred = newcred; if (newcred != NULL) PROC_UPDATE_COW(p); return (oldcred); } struct ucred * crcopysafe(struct proc *p, struct ucred *cr) { struct ucred *oldcred; int groups; PROC_LOCK_ASSERT(p, MA_OWNED); oldcred = p->p_ucred; while (cr->cr_agroups < oldcred->cr_agroups) { groups = oldcred->cr_agroups; PROC_UNLOCK(p); crextend(cr, groups); PROC_LOCK(p); oldcred = p->p_ucred; } crcopy(cr, oldcred); return (oldcred); } /* * Extend the passed in credential to hold n items. */ void crextend(struct ucred *cr, int n) { int cnt; /* Truncate? */ if (n <= cr->cr_agroups) return; /* * We extend by 2 each time since we're using a power of two * allocator until we need enough groups to fill a page. * Once we're allocating multiple pages, only allocate as many * as we actually need. The case of processes needing a * non-power of two number of pages seems more likely than * a real world process that adds thousands of groups one at a * time. */ if ( n < PAGE_SIZE / sizeof(gid_t) ) { if (cr->cr_agroups == 0) cnt = MINALLOCSIZE / sizeof(gid_t); else cnt = cr->cr_agroups * 2; while (cnt < n) cnt *= 2; } else cnt = roundup2(n, PAGE_SIZE / sizeof(gid_t)); /* Free the old array. */ if (cr->cr_groups != cr->cr_smallgroups) free(cr->cr_groups, M_CRED); cr->cr_groups = malloc(cnt * sizeof(gid_t), M_CRED, M_WAITOK | M_ZERO); cr->cr_agroups = cnt; } /* * Copy groups in to a credential, preserving any necessary invariants. * Currently this includes the sorting of all supplemental gids. * crextend() must have been called before hand to ensure sufficient * space is available. */ static void crsetgroups_locked(struct ucred *cr, int ngrp, gid_t *groups) { int i; int j; gid_t g; KASSERT(cr->cr_agroups >= ngrp, ("cr_ngroups is too small")); bcopy(groups, cr->cr_groups, ngrp * sizeof(gid_t)); cr->cr_ngroups = ngrp; /* * Sort all groups except cr_groups[0] to allow groupmember to * perform a binary search. * * XXX: If large numbers of groups become common this should * be replaced with shell sort like linux uses or possibly * heap sort. */ for (i = 2; i < ngrp; i++) { g = cr->cr_groups[i]; for (j = i-1; j >= 1 && g < cr->cr_groups[j]; j--) cr->cr_groups[j + 1] = cr->cr_groups[j]; cr->cr_groups[j + 1] = g; } } /* * Copy groups in to a credential after expanding it if required. * Truncate the list to (ngroups_max + 1) if it is too large. */ void crsetgroups(struct ucred *cr, int ngrp, gid_t *groups) { if (ngrp > ngroups_max + 1) ngrp = ngroups_max + 1; crextend(cr, ngrp); crsetgroups_locked(cr, ngrp, groups); } /* * Get login name, if available. */ #ifndef _SYS_SYSPROTO_H_ struct getlogin_args { char *namebuf; u_int namelen; }; #endif /* ARGSUSED */ int sys_getlogin(struct thread *td, struct getlogin_args *uap) { char login[MAXLOGNAME]; struct proc *p = td->td_proc; size_t len; if (uap->namelen > MAXLOGNAME) uap->namelen = MAXLOGNAME; PROC_LOCK(p); SESS_LOCK(p->p_session); len = strlcpy(login, p->p_session->s_login, uap->namelen) + 1; SESS_UNLOCK(p->p_session); PROC_UNLOCK(p); if (len > uap->namelen) return (ERANGE); return (copyout(login, uap->namebuf, len)); } /* * Set login name. */ #ifndef _SYS_SYSPROTO_H_ struct setlogin_args { char *namebuf; }; #endif /* ARGSUSED */ int sys_setlogin(struct thread *td, struct setlogin_args *uap) { struct proc *p = td->td_proc; int error; char logintmp[MAXLOGNAME]; CTASSERT(sizeof(p->p_session->s_login) >= sizeof(logintmp)); error = priv_check(td, PRIV_PROC_SETLOGIN); if (error) return (error); error = copyinstr(uap->namebuf, logintmp, sizeof(logintmp), NULL); if (error != 0) { if (error == ENAMETOOLONG) error = EINVAL; return (error); } AUDIT_ARG_LOGIN(logintmp); PROC_LOCK(p); SESS_LOCK(p->p_session); strcpy(p->p_session->s_login, logintmp); SESS_UNLOCK(p->p_session); PROC_UNLOCK(p); return (0); } void setsugid(struct proc *p) { PROC_LOCK_ASSERT(p, MA_OWNED); p->p_flag |= P_SUGID; if (!(p->p_pfsflags & PF_ISUGID)) p->p_stops = 0; } /*- * Change a process's effective uid. * Side effects: newcred->cr_uid and newcred->cr_uidinfo will be modified. * References: newcred must be an exclusive credential reference for the * duration of the call. */ void change_euid(struct ucred *newcred, struct uidinfo *euip) { newcred->cr_uid = euip->ui_uid; uihold(euip); uifree(newcred->cr_uidinfo); newcred->cr_uidinfo = euip; } /*- * Change a process's effective gid. * Side effects: newcred->cr_gid will be modified. * References: newcred must be an exclusive credential reference for the * duration of the call. */ void change_egid(struct ucred *newcred, gid_t egid) { newcred->cr_groups[0] = egid; } /*- * Change a process's real uid. * Side effects: newcred->cr_ruid will be updated, newcred->cr_ruidinfo * will be updated, and the old and new cr_ruidinfo proc * counts will be updated. * References: newcred must be an exclusive credential reference for the * duration of the call. */ void change_ruid(struct ucred *newcred, struct uidinfo *ruip) { (void)chgproccnt(newcred->cr_ruidinfo, -1, 0); newcred->cr_ruid = ruip->ui_uid; uihold(ruip); uifree(newcred->cr_ruidinfo); newcred->cr_ruidinfo = ruip; (void)chgproccnt(newcred->cr_ruidinfo, 1, 0); } /*- * Change a process's real gid. * Side effects: newcred->cr_rgid will be updated. * References: newcred must be an exclusive credential reference for the * duration of the call. */ void change_rgid(struct ucred *newcred, gid_t rgid) { newcred->cr_rgid = rgid; } /*- * Change a process's saved uid. * Side effects: newcred->cr_svuid will be updated. * References: newcred must be an exclusive credential reference for the * duration of the call. */ void change_svuid(struct ucred *newcred, uid_t svuid) { newcred->cr_svuid = svuid; } /*- * Change a process's saved gid. * Side effects: newcred->cr_svgid will be updated. * References: newcred must be an exclusive credential reference for the * duration of the call. */ void change_svgid(struct ucred *newcred, gid_t svgid) { newcred->cr_svgid = svgid; } Index: head/sys/kern/kern_resource.c =================================================================== --- head/sys/kern/kern_resource.c (revision 305831) +++ head/sys/kern/kern_resource.c (revision 305832) @@ -1,1441 +1,1441 @@ /*- * Copyright (c) 1982, 1986, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_resource.c 8.5 (Berkeley) 1/21/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static MALLOC_DEFINE(M_PLIMIT, "plimit", "plimit structures"); static MALLOC_DEFINE(M_UIDINFO, "uidinfo", "uidinfo structures"); #define UIHASH(uid) (&uihashtbl[(uid) & uihash]) static struct rwlock uihashtbl_lock; static LIST_HEAD(uihashhead, uidinfo) *uihashtbl; static u_long uihash; /* size of hash table - 1 */ static void calcru1(struct proc *p, struct rusage_ext *ruxp, struct timeval *up, struct timeval *sp); static int donice(struct thread *td, struct proc *chgp, int n); static struct uidinfo *uilookup(uid_t uid); static void ruxagg_locked(struct rusage_ext *rux, struct thread *td); /* * Resource controls and accounting. */ #ifndef _SYS_SYSPROTO_H_ struct getpriority_args { int which; int who; }; #endif int sys_getpriority(struct thread *td, register struct getpriority_args *uap) { struct proc *p; struct pgrp *pg; int error, low; error = 0; low = PRIO_MAX + 1; switch (uap->which) { case PRIO_PROCESS: if (uap->who == 0) low = td->td_proc->p_nice; else { p = pfind(uap->who); if (p == NULL) break; if (p_cansee(td, p) == 0) low = p->p_nice; PROC_UNLOCK(p); } break; case PRIO_PGRP: sx_slock(&proctree_lock); if (uap->who == 0) { pg = td->td_proc->p_pgrp; PGRP_LOCK(pg); } else { pg = pgfind(uap->who); if (pg == NULL) { sx_sunlock(&proctree_lock); break; } } sx_sunlock(&proctree_lock); LIST_FOREACH(p, &pg->pg_members, p_pglist) { PROC_LOCK(p); if (p->p_state == PRS_NORMAL && p_cansee(td, p) == 0) { if (p->p_nice < low) low = p->p_nice; } PROC_UNLOCK(p); } PGRP_UNLOCK(pg); break; case PRIO_USER: if (uap->who == 0) uap->who = td->td_ucred->cr_uid; sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); if (p->p_state == PRS_NORMAL && p_cansee(td, p) == 0 && p->p_ucred->cr_uid == uap->who) { if (p->p_nice < low) low = p->p_nice; } PROC_UNLOCK(p); } sx_sunlock(&allproc_lock); break; default: error = EINVAL; break; } if (low == PRIO_MAX + 1 && error == 0) error = ESRCH; td->td_retval[0] = low; return (error); } #ifndef _SYS_SYSPROTO_H_ struct setpriority_args { int which; int who; int prio; }; #endif int sys_setpriority(struct thread *td, struct setpriority_args *uap) { struct proc *curp, *p; struct pgrp *pg; int found = 0, error = 0; curp = td->td_proc; switch (uap->which) { case PRIO_PROCESS: if (uap->who == 0) { PROC_LOCK(curp); error = donice(td, curp, uap->prio); PROC_UNLOCK(curp); } else { p = pfind(uap->who); if (p == NULL) break; error = p_cansee(td, p); if (error == 0) error = donice(td, p, uap->prio); PROC_UNLOCK(p); } found++; break; case PRIO_PGRP: sx_slock(&proctree_lock); if (uap->who == 0) { pg = curp->p_pgrp; PGRP_LOCK(pg); } else { pg = pgfind(uap->who); if (pg == NULL) { sx_sunlock(&proctree_lock); break; } } sx_sunlock(&proctree_lock); LIST_FOREACH(p, &pg->pg_members, p_pglist) { PROC_LOCK(p); if (p->p_state == PRS_NORMAL && p_cansee(td, p) == 0) { error = donice(td, p, uap->prio); found++; } PROC_UNLOCK(p); } PGRP_UNLOCK(pg); break; case PRIO_USER: if (uap->who == 0) uap->who = td->td_ucred->cr_uid; sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); if (p->p_state == PRS_NORMAL && p->p_ucred->cr_uid == uap->who && p_cansee(td, p) == 0) { error = donice(td, p, uap->prio); found++; } PROC_UNLOCK(p); } sx_sunlock(&allproc_lock); break; default: error = EINVAL; break; } if (found == 0 && error == 0) error = ESRCH; return (error); } /* * Set "nice" for a (whole) process. */ static int donice(struct thread *td, struct proc *p, int n) { int error; PROC_LOCK_ASSERT(p, MA_OWNED); if ((error = p_cansched(td, p))) return (error); if (n > PRIO_MAX) n = PRIO_MAX; if (n < PRIO_MIN) n = PRIO_MIN; if (n < p->p_nice && priv_check(td, PRIV_SCHED_SETPRIORITY) != 0) return (EACCES); sched_nice(p, n); return (0); } static int unprivileged_idprio; SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_idprio, CTLFLAG_RW, &unprivileged_idprio, 0, "Allow non-root users to set an idle priority"); /* * Set realtime priority for LWP. */ #ifndef _SYS_SYSPROTO_H_ struct rtprio_thread_args { int function; lwpid_t lwpid; struct rtprio *rtp; }; #endif int sys_rtprio_thread(struct thread *td, struct rtprio_thread_args *uap) { struct proc *p; struct rtprio rtp; struct thread *td1; int cierror, error; /* Perform copyin before acquiring locks if needed. */ if (uap->function == RTP_SET) cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio)); else cierror = 0; if (uap->lwpid == 0 || uap->lwpid == td->td_tid) { p = td->td_proc; td1 = td; PROC_LOCK(p); } else { /* Only look up thread in current process */ td1 = tdfind(uap->lwpid, curproc->p_pid); if (td1 == NULL) return (ESRCH); p = td1->td_proc; } switch (uap->function) { case RTP_LOOKUP: if ((error = p_cansee(td, p))) break; pri_to_rtp(td1, &rtp); PROC_UNLOCK(p); return (copyout(&rtp, uap->rtp, sizeof(struct rtprio))); case RTP_SET: if ((error = p_cansched(td, p)) || (error = cierror)) break; /* Disallow setting rtprio in most cases if not superuser. */ /* * Realtime priority has to be restricted for reasons which * should be obvious. However, for idleprio processes, there is * a potential for system deadlock if an idleprio process gains * a lock on a resource that other processes need (and the * idleprio process can't run due to a CPU-bound normal * process). Fix me! XXX * * This problem is not only related to idleprio process. * A user level program can obtain a file lock and hold it * indefinitely. Additionally, without idleprio processes it is * still conceivable that a program with low priority will never * get to run. In short, allowing this feature might make it * easier to lock a resource indefinitely, but it is not the * only thing that makes it possible. */ if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_REALTIME || (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_IDLE && unprivileged_idprio == 0)) { error = priv_check(td, PRIV_SCHED_RTPRIO); if (error) break; } error = rtp_to_pri(&rtp, td1); break; default: error = EINVAL; break; } PROC_UNLOCK(p); return (error); } /* * Set realtime priority. */ #ifndef _SYS_SYSPROTO_H_ struct rtprio_args { int function; pid_t pid; struct rtprio *rtp; }; #endif int sys_rtprio(struct thread *td, register struct rtprio_args *uap) { struct proc *p; struct thread *tdp; struct rtprio rtp; int cierror, error; /* Perform copyin before acquiring locks if needed. */ if (uap->function == RTP_SET) cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio)); else cierror = 0; if (uap->pid == 0) { p = td->td_proc; PROC_LOCK(p); } else { p = pfind(uap->pid); if (p == NULL) return (ESRCH); } switch (uap->function) { case RTP_LOOKUP: if ((error = p_cansee(td, p))) break; /* * Return OUR priority if no pid specified, * or if one is, report the highest priority * in the process. There isn't much more you can do as * there is only room to return a single priority. * Note: specifying our own pid is not the same * as leaving it zero. */ if (uap->pid == 0) { pri_to_rtp(td, &rtp); } else { struct rtprio rtp2; rtp.type = RTP_PRIO_IDLE; rtp.prio = RTP_PRIO_MAX; FOREACH_THREAD_IN_PROC(p, tdp) { pri_to_rtp(tdp, &rtp2); if (rtp2.type < rtp.type || (rtp2.type == rtp.type && rtp2.prio < rtp.prio)) { rtp.type = rtp2.type; rtp.prio = rtp2.prio; } } } PROC_UNLOCK(p); return (copyout(&rtp, uap->rtp, sizeof(struct rtprio))); case RTP_SET: if ((error = p_cansched(td, p)) || (error = cierror)) break; /* * Disallow setting rtprio in most cases if not superuser. * See the comment in sys_rtprio_thread about idprio * threads holding a lock. */ if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_REALTIME || (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_IDLE && !unprivileged_idprio)) { error = priv_check(td, PRIV_SCHED_RTPRIO); if (error) break; } /* * If we are setting our own priority, set just our * thread but if we are doing another process, * do all the threads on that process. If we * specify our own pid we do the latter. */ if (uap->pid == 0) { error = rtp_to_pri(&rtp, td); } else { FOREACH_THREAD_IN_PROC(p, td) { if ((error = rtp_to_pri(&rtp, td)) != 0) break; } } break; default: error = EINVAL; break; } PROC_UNLOCK(p); return (error); } int rtp_to_pri(struct rtprio *rtp, struct thread *td) { u_char newpri, oldclass, oldpri; switch (RTP_PRIO_BASE(rtp->type)) { case RTP_PRIO_REALTIME: if (rtp->prio > RTP_PRIO_MAX) return (EINVAL); newpri = PRI_MIN_REALTIME + rtp->prio; break; case RTP_PRIO_NORMAL: if (rtp->prio > (PRI_MAX_TIMESHARE - PRI_MIN_TIMESHARE)) return (EINVAL); newpri = PRI_MIN_TIMESHARE + rtp->prio; break; case RTP_PRIO_IDLE: if (rtp->prio > RTP_PRIO_MAX) return (EINVAL); newpri = PRI_MIN_IDLE + rtp->prio; break; default: return (EINVAL); } thread_lock(td); oldclass = td->td_pri_class; sched_class(td, rtp->type); /* XXX fix */ oldpri = td->td_user_pri; sched_user_prio(td, newpri); if (td->td_user_pri != oldpri && (oldclass != RTP_PRIO_NORMAL || td->td_pri_class != RTP_PRIO_NORMAL)) sched_prio(td, td->td_user_pri); if (TD_ON_UPILOCK(td) && oldpri != newpri) { critical_enter(); thread_unlock(td); umtx_pi_adjust(td, oldpri); critical_exit(); } else thread_unlock(td); return (0); } void pri_to_rtp(struct thread *td, struct rtprio *rtp) { thread_lock(td); switch (PRI_BASE(td->td_pri_class)) { case PRI_REALTIME: rtp->prio = td->td_base_user_pri - PRI_MIN_REALTIME; break; case PRI_TIMESHARE: rtp->prio = td->td_base_user_pri - PRI_MIN_TIMESHARE; break; case PRI_IDLE: rtp->prio = td->td_base_user_pri - PRI_MIN_IDLE; break; default: break; } rtp->type = td->td_pri_class; thread_unlock(td); } #if defined(COMPAT_43) #ifndef _SYS_SYSPROTO_H_ struct osetrlimit_args { u_int which; struct orlimit *rlp; }; #endif int osetrlimit(struct thread *td, register struct osetrlimit_args *uap) { struct orlimit olim; struct rlimit lim; int error; if ((error = copyin(uap->rlp, &olim, sizeof(struct orlimit)))) return (error); lim.rlim_cur = olim.rlim_cur; lim.rlim_max = olim.rlim_max; error = kern_setrlimit(td, uap->which, &lim); return (error); } #ifndef _SYS_SYSPROTO_H_ struct ogetrlimit_args { u_int which; struct orlimit *rlp; }; #endif int ogetrlimit(struct thread *td, register struct ogetrlimit_args *uap) { struct orlimit olim; struct rlimit rl; int error; if (uap->which >= RLIM_NLIMITS) return (EINVAL); lim_rlimit(td, uap->which, &rl); /* * XXX would be more correct to convert only RLIM_INFINITY to the * old RLIM_INFINITY and fail with EOVERFLOW for other larger * values. Most 64->32 and 32->16 conversions, including not * unimportant ones of uids are even more broken than what we * do here (they blindly truncate). We don't do this correctly * here since we have little experience with EOVERFLOW yet. * Elsewhere, getuid() can't fail... */ olim.rlim_cur = rl.rlim_cur > 0x7fffffff ? 0x7fffffff : rl.rlim_cur; olim.rlim_max = rl.rlim_max > 0x7fffffff ? 0x7fffffff : rl.rlim_max; error = copyout(&olim, uap->rlp, sizeof(olim)); return (error); } #endif /* COMPAT_43 */ #ifndef _SYS_SYSPROTO_H_ struct __setrlimit_args { u_int which; struct rlimit *rlp; }; #endif int sys_setrlimit(struct thread *td, register struct __setrlimit_args *uap) { struct rlimit alim; int error; if ((error = copyin(uap->rlp, &alim, sizeof(struct rlimit)))) return (error); error = kern_setrlimit(td, uap->which, &alim); return (error); } static void lim_cb(void *arg) { struct rlimit rlim; struct thread *td; struct proc *p; p = arg; PROC_LOCK_ASSERT(p, MA_OWNED); /* * Check if the process exceeds its cpu resource allocation. If * it reaches the max, arrange to kill the process in ast(). */ if (p->p_cpulimit == RLIM_INFINITY) return; PROC_STATLOCK(p); FOREACH_THREAD_IN_PROC(p, td) { ruxagg(p, td); } PROC_STATUNLOCK(p); if (p->p_rux.rux_runtime > p->p_cpulimit * cpu_tickrate()) { lim_rlimit_proc(p, RLIMIT_CPU, &rlim); if (p->p_rux.rux_runtime >= rlim.rlim_max * cpu_tickrate()) { killproc(p, "exceeded maximum CPU limit"); } else { if (p->p_cpulimit < rlim.rlim_max) p->p_cpulimit += 5; kern_psignal(p, SIGXCPU); } } if ((p->p_flag & P_WEXIT) == 0) callout_reset_sbt(&p->p_limco, SBT_1S, 0, lim_cb, p, C_PREL(1)); } int kern_setrlimit(struct thread *td, u_int which, struct rlimit *limp) { return (kern_proc_setrlimit(td, td->td_proc, which, limp)); } int kern_proc_setrlimit(struct thread *td, struct proc *p, u_int which, struct rlimit *limp) { struct plimit *newlim, *oldlim; register struct rlimit *alimp; struct rlimit oldssiz; int error; if (which >= RLIM_NLIMITS) return (EINVAL); /* * Preserve historical bugs by treating negative limits as unsigned. */ if (limp->rlim_cur < 0) limp->rlim_cur = RLIM_INFINITY; if (limp->rlim_max < 0) limp->rlim_max = RLIM_INFINITY; oldssiz.rlim_cur = 0; newlim = lim_alloc(); PROC_LOCK(p); oldlim = p->p_limit; alimp = &oldlim->pl_rlimit[which]; if (limp->rlim_cur > alimp->rlim_max || limp->rlim_max > alimp->rlim_max) if ((error = priv_check(td, PRIV_PROC_SETRLIMIT))) { PROC_UNLOCK(p); lim_free(newlim); return (error); } if (limp->rlim_cur > limp->rlim_max) limp->rlim_cur = limp->rlim_max; lim_copy(newlim, oldlim); alimp = &newlim->pl_rlimit[which]; switch (which) { case RLIMIT_CPU: if (limp->rlim_cur != RLIM_INFINITY && p->p_cpulimit == RLIM_INFINITY) callout_reset_sbt(&p->p_limco, SBT_1S, 0, lim_cb, p, C_PREL(1)); p->p_cpulimit = limp->rlim_cur; break; case RLIMIT_DATA: if (limp->rlim_cur > maxdsiz) limp->rlim_cur = maxdsiz; if (limp->rlim_max > maxdsiz) limp->rlim_max = maxdsiz; break; case RLIMIT_STACK: if (limp->rlim_cur > maxssiz) limp->rlim_cur = maxssiz; if (limp->rlim_max > maxssiz) limp->rlim_max = maxssiz; oldssiz = *alimp; if (p->p_sysent->sv_fixlimit != NULL) p->p_sysent->sv_fixlimit(&oldssiz, RLIMIT_STACK); break; case RLIMIT_NOFILE: if (limp->rlim_cur > maxfilesperproc) limp->rlim_cur = maxfilesperproc; if (limp->rlim_max > maxfilesperproc) limp->rlim_max = maxfilesperproc; break; case RLIMIT_NPROC: if (limp->rlim_cur > maxprocperuid) limp->rlim_cur = maxprocperuid; if (limp->rlim_max > maxprocperuid) limp->rlim_max = maxprocperuid; if (limp->rlim_cur < 1) limp->rlim_cur = 1; if (limp->rlim_max < 1) limp->rlim_max = 1; break; } if (p->p_sysent->sv_fixlimit != NULL) p->p_sysent->sv_fixlimit(limp, which); *alimp = *limp; p->p_limit = newlim; PROC_UPDATE_COW(p); PROC_UNLOCK(p); lim_free(oldlim); if (which == RLIMIT_STACK && /* * Skip calls from exec_new_vmspace(), done when stack is * not mapped yet. */ (td != curthread || (p->p_flag & P_INEXEC) == 0)) { /* * Stack is allocated to the max at exec time with only * "rlim_cur" bytes accessible. If stack limit is going * up make more accessible, if going down make inaccessible. */ if (limp->rlim_cur != oldssiz.rlim_cur) { vm_offset_t addr; vm_size_t size; vm_prot_t prot; if (limp->rlim_cur > oldssiz.rlim_cur) { prot = p->p_sysent->sv_stackprot; size = limp->rlim_cur - oldssiz.rlim_cur; addr = p->p_sysent->sv_usrstack - limp->rlim_cur; } else { prot = VM_PROT_NONE; size = oldssiz.rlim_cur - limp->rlim_cur; addr = p->p_sysent->sv_usrstack - oldssiz.rlim_cur; } addr = trunc_page(addr); size = round_page(size); (void)vm_map_protect(&p->p_vmspace->vm_map, addr, addr + size, prot, FALSE); } } return (0); } #ifndef _SYS_SYSPROTO_H_ struct __getrlimit_args { u_int which; struct rlimit *rlp; }; #endif /* ARGSUSED */ int sys_getrlimit(struct thread *td, register struct __getrlimit_args *uap) { struct rlimit rlim; int error; if (uap->which >= RLIM_NLIMITS) return (EINVAL); lim_rlimit(td, uap->which, &rlim); error = copyout(&rlim, uap->rlp, sizeof(struct rlimit)); return (error); } /* * Transform the running time and tick information for children of proc p * into user and system time usage. */ void calccru(struct proc *p, struct timeval *up, struct timeval *sp) { PROC_LOCK_ASSERT(p, MA_OWNED); calcru1(p, &p->p_crux, up, sp); } /* * Transform the running time and tick information in proc p into user * and system time usage. If appropriate, include the current time slice * on this CPU. */ void calcru(struct proc *p, struct timeval *up, struct timeval *sp) { struct thread *td; uint64_t runtime, u; PROC_LOCK_ASSERT(p, MA_OWNED); PROC_STATLOCK_ASSERT(p, MA_OWNED); /* * If we are getting stats for the current process, then add in the * stats that this thread has accumulated in its current time slice. * We reset the thread and CPU state as if we had performed a context * switch right here. */ td = curthread; if (td->td_proc == p) { u = cpu_ticks(); runtime = u - PCPU_GET(switchtime); td->td_runtime += runtime; td->td_incruntime += runtime; PCPU_SET(switchtime, u); } /* Make sure the per-thread stats are current. */ FOREACH_THREAD_IN_PROC(p, td) { if (td->td_incruntime == 0) continue; ruxagg(p, td); } calcru1(p, &p->p_rux, up, sp); } /* Collect resource usage for a single thread. */ void rufetchtd(struct thread *td, struct rusage *ru) { struct proc *p; uint64_t runtime, u; p = td->td_proc; PROC_STATLOCK_ASSERT(p, MA_OWNED); THREAD_LOCK_ASSERT(td, MA_OWNED); /* * If we are getting stats for the current thread, then add in the * stats that this thread has accumulated in its current time slice. * We reset the thread and CPU state as if we had performed a context * switch right here. */ if (td == curthread) { u = cpu_ticks(); runtime = u - PCPU_GET(switchtime); td->td_runtime += runtime; td->td_incruntime += runtime; PCPU_SET(switchtime, u); } ruxagg(p, td); *ru = td->td_ru; calcru1(p, &td->td_rux, &ru->ru_utime, &ru->ru_stime); } static void calcru1(struct proc *p, struct rusage_ext *ruxp, struct timeval *up, struct timeval *sp) { /* {user, system, interrupt, total} {ticks, usec}: */ uint64_t ut, uu, st, su, it, tt, tu; ut = ruxp->rux_uticks; st = ruxp->rux_sticks; it = ruxp->rux_iticks; tt = ut + st + it; if (tt == 0) { /* Avoid divide by zero */ st = 1; tt = 1; } tu = cputick2usec(ruxp->rux_runtime); if ((int64_t)tu < 0) { /* XXX: this should be an assert /phk */ printf("calcru: negative runtime of %jd usec for pid %d (%s)\n", (intmax_t)tu, p->p_pid, p->p_comm); tu = ruxp->rux_tu; } if (tu >= ruxp->rux_tu) { /* * The normal case, time increased. * Enforce monotonicity of bucketed numbers. */ uu = (tu * ut) / tt; if (uu < ruxp->rux_uu) uu = ruxp->rux_uu; su = (tu * st) / tt; if (su < ruxp->rux_su) su = ruxp->rux_su; } else if (tu + 3 > ruxp->rux_tu || 101 * tu > 100 * ruxp->rux_tu) { /* * When we calibrate the cputicker, it is not uncommon to * see the presumably fixed frequency increase slightly over * time as a result of thermal stabilization and NTP * discipline (of the reference clock). We therefore ignore * a bit of backwards slop because we expect to catch up * shortly. We use a 3 microsecond limit to catch low * counts and a 1% limit for high counts. */ uu = ruxp->rux_uu; su = ruxp->rux_su; tu = ruxp->rux_tu; } else { /* tu < ruxp->rux_tu */ /* * What happened here was likely that a laptop, which ran at * a reduced clock frequency at boot, kicked into high gear. * The wisdom of spamming this message in that case is * dubious, but it might also be indicative of something * serious, so lets keep it and hope laptops can be made * more truthful about their CPU speed via ACPI. */ printf("calcru: runtime went backwards from %ju usec " "to %ju usec for pid %d (%s)\n", (uintmax_t)ruxp->rux_tu, (uintmax_t)tu, p->p_pid, p->p_comm); uu = (tu * ut) / tt; su = (tu * st) / tt; } ruxp->rux_uu = uu; ruxp->rux_su = su; ruxp->rux_tu = tu; up->tv_sec = uu / 1000000; up->tv_usec = uu % 1000000; sp->tv_sec = su / 1000000; sp->tv_usec = su % 1000000; } #ifndef _SYS_SYSPROTO_H_ struct getrusage_args { int who; struct rusage *rusage; }; #endif int sys_getrusage(register struct thread *td, register struct getrusage_args *uap) { struct rusage ru; int error; error = kern_getrusage(td, uap->who, &ru); if (error == 0) error = copyout(&ru, uap->rusage, sizeof(struct rusage)); return (error); } int kern_getrusage(struct thread *td, int who, struct rusage *rup) { struct proc *p; int error; error = 0; p = td->td_proc; PROC_LOCK(p); switch (who) { case RUSAGE_SELF: rufetchcalc(p, rup, &rup->ru_utime, &rup->ru_stime); break; case RUSAGE_CHILDREN: *rup = p->p_stats->p_cru; calccru(p, &rup->ru_utime, &rup->ru_stime); break; case RUSAGE_THREAD: PROC_STATLOCK(p); thread_lock(td); rufetchtd(td, rup); thread_unlock(td); PROC_STATUNLOCK(p); break; default: error = EINVAL; } PROC_UNLOCK(p); return (error); } void rucollect(struct rusage *ru, struct rusage *ru2) { long *ip, *ip2; int i; if (ru->ru_maxrss < ru2->ru_maxrss) ru->ru_maxrss = ru2->ru_maxrss; ip = &ru->ru_first; ip2 = &ru2->ru_first; for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--) *ip++ += *ip2++; } void ruadd(struct rusage *ru, struct rusage_ext *rux, struct rusage *ru2, struct rusage_ext *rux2) { rux->rux_runtime += rux2->rux_runtime; rux->rux_uticks += rux2->rux_uticks; rux->rux_sticks += rux2->rux_sticks; rux->rux_iticks += rux2->rux_iticks; rux->rux_uu += rux2->rux_uu; rux->rux_su += rux2->rux_su; rux->rux_tu += rux2->rux_tu; rucollect(ru, ru2); } /* * Aggregate tick counts into the proc's rusage_ext. */ static void ruxagg_locked(struct rusage_ext *rux, struct thread *td) { THREAD_LOCK_ASSERT(td, MA_OWNED); PROC_STATLOCK_ASSERT(td->td_proc, MA_OWNED); rux->rux_runtime += td->td_incruntime; rux->rux_uticks += td->td_uticks; rux->rux_sticks += td->td_sticks; rux->rux_iticks += td->td_iticks; } void ruxagg(struct proc *p, struct thread *td) { thread_lock(td); ruxagg_locked(&p->p_rux, td); ruxagg_locked(&td->td_rux, td); td->td_incruntime = 0; td->td_uticks = 0; td->td_iticks = 0; td->td_sticks = 0; thread_unlock(td); } /* * Update the rusage_ext structure and fetch a valid aggregate rusage * for proc p if storage for one is supplied. */ void rufetch(struct proc *p, struct rusage *ru) { struct thread *td; PROC_STATLOCK_ASSERT(p, MA_OWNED); *ru = p->p_ru; if (p->p_numthreads > 0) { FOREACH_THREAD_IN_PROC(p, td) { ruxagg(p, td); rucollect(ru, &td->td_ru); } } } /* * Atomically perform a rufetch and a calcru together. * Consumers, can safely assume the calcru is executed only once * rufetch is completed. */ void rufetchcalc(struct proc *p, struct rusage *ru, struct timeval *up, struct timeval *sp) { PROC_STATLOCK(p); rufetch(p, ru); calcru(p, up, sp); PROC_STATUNLOCK(p); } /* * Allocate a new resource limits structure and initialize its * reference count and mutex pointer. */ struct plimit * lim_alloc() { struct plimit *limp; limp = malloc(sizeof(struct plimit), M_PLIMIT, M_WAITOK); refcount_init(&limp->pl_refcnt, 1); return (limp); } struct plimit * lim_hold(struct plimit *limp) { refcount_acquire(&limp->pl_refcnt); return (limp); } void lim_fork(struct proc *p1, struct proc *p2) { PROC_LOCK_ASSERT(p1, MA_OWNED); PROC_LOCK_ASSERT(p2, MA_OWNED); p2->p_limit = lim_hold(p1->p_limit); callout_init_mtx(&p2->p_limco, &p2->p_mtx, 0); if (p1->p_cpulimit != RLIM_INFINITY) callout_reset_sbt(&p2->p_limco, SBT_1S, 0, lim_cb, p2, C_PREL(1)); } void lim_free(struct plimit *limp) { if (refcount_release(&limp->pl_refcnt)) free((void *)limp, M_PLIMIT); } /* * Make a copy of the plimit structure. * We share these structures copy-on-write after fork. */ void lim_copy(struct plimit *dst, struct plimit *src) { KASSERT(dst->pl_refcnt <= 1, ("lim_copy to shared limit")); bcopy(src->pl_rlimit, dst->pl_rlimit, sizeof(src->pl_rlimit)); } /* * Return the hard limit for a particular system resource. The * which parameter specifies the index into the rlimit array. */ rlim_t lim_max(struct thread *td, int which) { struct rlimit rl; lim_rlimit(td, which, &rl); return (rl.rlim_max); } rlim_t lim_max_proc(struct proc *p, int which) { struct rlimit rl; lim_rlimit_proc(p, which, &rl); return (rl.rlim_max); } /* * Return the current (soft) limit for a particular system resource. * The which parameter which specifies the index into the rlimit array */ rlim_t lim_cur(struct thread *td, int which) { struct rlimit rl; lim_rlimit(td, which, &rl); return (rl.rlim_cur); } rlim_t lim_cur_proc(struct proc *p, int which) { struct rlimit rl; lim_rlimit_proc(p, which, &rl); return (rl.rlim_cur); } /* * Return a copy of the entire rlimit structure for the system limit * specified by 'which' in the rlimit structure pointed to by 'rlp'. */ void lim_rlimit(struct thread *td, int which, struct rlimit *rlp) { struct proc *p = td->td_proc; MPASS(td == curthread); KASSERT(which >= 0 && which < RLIM_NLIMITS, ("request for invalid resource limit")); *rlp = td->td_limit->pl_rlimit[which]; if (p->p_sysent->sv_fixlimit != NULL) p->p_sysent->sv_fixlimit(rlp, which); } void lim_rlimit_proc(struct proc *p, int which, struct rlimit *rlp) { PROC_LOCK_ASSERT(p, MA_OWNED); KASSERT(which >= 0 && which < RLIM_NLIMITS, ("request for invalid resource limit")); *rlp = p->p_limit->pl_rlimit[which]; if (p->p_sysent->sv_fixlimit != NULL) p->p_sysent->sv_fixlimit(rlp, which); } void uihashinit() { uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash); rw_init(&uihashtbl_lock, "uidinfo hash"); } /* * Look up a uidinfo struct for the parameter uid. * uihashtbl_lock must be locked. * Increase refcount on uidinfo struct returned. */ static struct uidinfo * uilookup(uid_t uid) { struct uihashhead *uipp; struct uidinfo *uip; rw_assert(&uihashtbl_lock, RA_LOCKED); uipp = UIHASH(uid); LIST_FOREACH(uip, uipp, ui_hash) if (uip->ui_uid == uid) { uihold(uip); break; } return (uip); } /* * Find or allocate a struct uidinfo for a particular uid. * Returns with uidinfo struct referenced. * uifree() should be called on a struct uidinfo when released. */ struct uidinfo * uifind(uid_t uid) { struct uidinfo *new_uip, *uip; rw_rlock(&uihashtbl_lock); uip = uilookup(uid); rw_runlock(&uihashtbl_lock); if (uip != NULL) return (uip); new_uip = malloc(sizeof(*new_uip), M_UIDINFO, M_WAITOK | M_ZERO); racct_create(&new_uip->ui_racct); refcount_init(&new_uip->ui_ref, 1); new_uip->ui_uid = uid; mtx_init(&new_uip->ui_vmsize_mtx, "ui_vmsize", NULL, MTX_DEF); rw_wlock(&uihashtbl_lock); /* * There's a chance someone created our uidinfo while we * were in malloc and not holding the lock, so we have to * make sure we don't insert a duplicate uidinfo. */ if ((uip = uilookup(uid)) == NULL) { LIST_INSERT_HEAD(UIHASH(uid), new_uip, ui_hash); rw_wunlock(&uihashtbl_lock); uip = new_uip; } else { rw_wunlock(&uihashtbl_lock); racct_destroy(&new_uip->ui_racct); mtx_destroy(&new_uip->ui_vmsize_mtx); free(new_uip, M_UIDINFO); } return (uip); } /* * Place another refcount on a uidinfo struct. */ void uihold(struct uidinfo *uip) { refcount_acquire(&uip->ui_ref); } /*- * Since uidinfo structs have a long lifetime, we use an * opportunistic refcounting scheme to avoid locking the lookup hash * for each release. * * If the refcount hits 0, we need to free the structure, * which means we need to lock the hash. * Optimal case: * After locking the struct and lowering the refcount, if we find * that we don't need to free, simply unlock and return. * Suboptimal case: * If refcount lowering results in need to free, bump the count * back up, lose the lock and acquire the locks in the proper * order to try again. */ void uifree(struct uidinfo *uip) { int old; /* Prepare for optimal case. */ old = uip->ui_ref; if (old > 1 && atomic_cmpset_int(&uip->ui_ref, old, old - 1)) return; /* Prepare for suboptimal case. */ rw_wlock(&uihashtbl_lock); if (refcount_release(&uip->ui_ref) == 0) { rw_wunlock(&uihashtbl_lock); return; } racct_destroy(&uip->ui_racct); LIST_REMOVE(uip, ui_hash); rw_wunlock(&uihashtbl_lock); if (uip->ui_sbsize != 0) printf("freeing uidinfo: uid = %d, sbsize = %ld\n", uip->ui_uid, uip->ui_sbsize); if (uip->ui_proccnt != 0) printf("freeing uidinfo: uid = %d, proccnt = %ld\n", uip->ui_uid, uip->ui_proccnt); if (uip->ui_vmsize != 0) printf("freeing uidinfo: uid = %d, swapuse = %lld\n", uip->ui_uid, (unsigned long long)uip->ui_vmsize); mtx_destroy(&uip->ui_vmsize_mtx); free(uip, M_UIDINFO); } #ifdef RACCT void ui_racct_foreach(void (*callback)(struct racct *racct, void *arg2, void *arg3), void (*pre)(void), void (*post)(void), void *arg2, void *arg3) { struct uidinfo *uip; struct uihashhead *uih; rw_rlock(&uihashtbl_lock); if (pre != NULL) (pre)(); for (uih = &uihashtbl[uihash]; uih >= uihashtbl; uih--) { LIST_FOREACH(uip, uih, ui_hash) { (callback)(uip->ui_racct, arg2, arg3); } } if (post != NULL) (post)(); rw_runlock(&uihashtbl_lock); } #endif static inline int chglimit(struct uidinfo *uip, long *limit, int diff, rlim_t max, const char *name) { /* Don't allow them to exceed max, but allow subtraction. */ if (diff > 0 && max != 0) { if (atomic_fetchadd_long(limit, (long)diff) + diff > max) { atomic_subtract_long(limit, (long)diff); return (0); } } else { atomic_add_long(limit, (long)diff); if (*limit < 0) printf("negative %s for uid = %d\n", name, uip->ui_uid); } return (1); } /* * Change the count associated with number of processes * a given user is using. When 'max' is 0, don't enforce a limit */ int chgproccnt(struct uidinfo *uip, int diff, rlim_t max) { return (chglimit(uip, &uip->ui_proccnt, diff, max, "proccnt")); } /* * Change the total socket buffer size a user has used. */ int chgsbsize(struct uidinfo *uip, u_int *hiwat, u_int to, rlim_t max) { int diff, rv; diff = to - *hiwat; if (diff > 0 && max == 0) { rv = 0; } else { rv = chglimit(uip, &uip->ui_sbsize, diff, max, "sbsize"); if (rv != 0) *hiwat = to; } return (rv); } /* * Change the count associated with number of pseudo-terminals * a given user is using. When 'max' is 0, don't enforce a limit */ int chgptscnt(struct uidinfo *uip, int diff, rlim_t max) { return (chglimit(uip, &uip->ui_ptscnt, diff, max, "ptscnt")); } int chgkqcnt(struct uidinfo *uip, int diff, rlim_t max) { return (chglimit(uip, &uip->ui_kqcnt, diff, max, "kqcnt")); } int chgumtxcnt(struct uidinfo *uip, int diff, rlim_t max) { return (chglimit(uip, &uip->ui_umtxcnt, diff, max, "umtxcnt")); } Index: head/sys/kern/kern_sendfile.c =================================================================== --- head/sys/kern/kern_sendfile.c (revision 305831) +++ head/sys/kern/kern_sendfile.c (revision 305832) @@ -1,1016 +1,1016 @@ /*- * Copyright (c) 2013-2015 Gleb Smirnoff * Copyright (c) 1998, David Greenman. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * Structure describing a single sendfile(2) I/O, which may consist of * several underlying pager I/Os. * * The syscall context allocates the structure and initializes 'nios' * to 1. As sendfile_swapin() runs through pages and starts asynchronous * paging operations, it increments 'nios'. * * Every I/O completion calls sendfile_iodone(), which decrements the 'nios', * and the syscall also calls sendfile_iodone() after allocating all mbufs, * linking them and sending to socket. Whoever reaches zero 'nios' is * responsible to * call pru_ready on the socket, to notify it of readyness * of the data. */ struct sf_io { volatile u_int nios; u_int error; int npages; struct file *sock_fp; struct mbuf *m; vm_page_t pa[]; }; /* * Structure used to track requests with SF_SYNC flag. */ struct sendfile_sync { struct mtx mtx; struct cv cv; unsigned count; }; counter_u64_t sfstat[sizeof(struct sfstat) / sizeof(uint64_t)]; static void sfstat_init(const void *unused) { COUNTER_ARRAY_ALLOC(sfstat, sizeof(struct sfstat) / sizeof(uint64_t), M_WAITOK); } SYSINIT(sfstat, SI_SUB_MBUF, SI_ORDER_FIRST, sfstat_init, NULL); static int sfstat_sysctl(SYSCTL_HANDLER_ARGS) { struct sfstat s; COUNTER_ARRAY_COPY(sfstat, &s, sizeof(s) / sizeof(uint64_t)); if (req->newptr) COUNTER_ARRAY_ZERO(sfstat, sizeof(s) / sizeof(uint64_t)); return (SYSCTL_OUT(req, &s, sizeof(s))); } SYSCTL_PROC(_kern_ipc, OID_AUTO, sfstat, CTLTYPE_OPAQUE | CTLFLAG_RW, NULL, 0, sfstat_sysctl, "I", "sendfile statistics"); /* * Detach mapped page and release resources back to the system. Called * by mbuf(9) code when last reference to a page is freed. */ void sf_ext_free(void *arg1, void *arg2) { struct sf_buf *sf = arg1; struct sendfile_sync *sfs = arg2; vm_page_t pg = sf_buf_page(sf); sf_buf_free(sf); vm_page_lock(pg); /* * Check for the object going away on us. This can * happen since we don't hold a reference to it. * If so, we're responsible for freeing the page. */ if (vm_page_unwire(pg, PQ_INACTIVE) && pg->object == NULL) vm_page_free(pg); vm_page_unlock(pg); if (sfs != NULL) { mtx_lock(&sfs->mtx); KASSERT(sfs->count > 0, ("Sendfile sync botchup count == 0")); if (--sfs->count == 0) cv_signal(&sfs->cv); mtx_unlock(&sfs->mtx); } } /* * Same as above, but forces the page to be detached from the object * and go into free pool. */ void sf_ext_free_nocache(void *arg1, void *arg2) { struct sf_buf *sf = arg1; struct sendfile_sync *sfs = arg2; vm_page_t pg = sf_buf_page(sf); sf_buf_free(sf); vm_page_lock(pg); if (vm_page_unwire(pg, PQ_NONE)) { vm_object_t obj; /* Try to free the page, but only if it is cheap to. */ if ((obj = pg->object) == NULL) vm_page_free(pg); else if (!vm_page_xbusied(pg) && VM_OBJECT_TRYWLOCK(obj)) { vm_page_free(pg); VM_OBJECT_WUNLOCK(obj); } else vm_page_deactivate(pg); } vm_page_unlock(pg); if (sfs != NULL) { mtx_lock(&sfs->mtx); KASSERT(sfs->count > 0, ("Sendfile sync botchup count == 0")); if (--sfs->count == 0) cv_signal(&sfs->cv); mtx_unlock(&sfs->mtx); } } /* * Helper function to calculate how much data to put into page i of n. * Only first and last pages are special. */ static inline off_t xfsize(int i, int n, off_t off, off_t len) { if (i == 0) return (omin(PAGE_SIZE - (off & PAGE_MASK), len)); if (i == n - 1 && ((off + len) & PAGE_MASK) > 0) return ((off + len) & PAGE_MASK); return (PAGE_SIZE); } /* * Helper function to get offset within object for i page. */ static inline vm_offset_t vmoff(int i, off_t off) { if (i == 0) return ((vm_offset_t)off); return (trunc_page(off + i * PAGE_SIZE)); } /* * Helper function used when allocation of a page or sf_buf failed. * Pretend as if we don't have enough space, subtract xfsize() of * all pages that failed. */ static inline void fixspace(int old, int new, off_t off, int *space) { KASSERT(old > new, ("%s: old %d new %d", __func__, old, new)); /* Subtract last one. */ *space -= xfsize(old - 1, old, off, *space); old--; if (new == old) /* There was only one page. */ return; /* Subtract first one. */ if (new == 0) { *space -= xfsize(0, old, off, *space); new++; } /* Rest of pages are full sized. */ *space -= (old - new) * PAGE_SIZE; KASSERT(*space >= 0, ("%s: space went backwards", __func__)); } /* * I/O completion callback. */ static void sendfile_iodone(void *arg, vm_page_t *pg, int count, int error) { struct sf_io *sfio = arg; struct socket *so; for (int i = 0; i < count; i++) vm_page_xunbusy(pg[i]); if (error) sfio->error = error; if (!refcount_release(&sfio->nios)) return; so = sfio->sock_fp->f_data; if (sfio->error) { struct mbuf *m; /* * I/O operation failed. The state of data in the socket * is now inconsistent, and all what we can do is to tear * it down. Protocol abort method would tear down protocol * state, free all ready mbufs and detach not ready ones. * We will free the mbufs corresponding to this I/O manually. * * The socket would be marked with EIO and made available * for read, so that application receives EIO on next * syscall and eventually closes the socket. */ so->so_proto->pr_usrreqs->pru_abort(so); so->so_error = EIO; m = sfio->m; for (int i = 0; i < sfio->npages; i++) m = m_free(m); } else { CURVNET_SET(so->so_vnet); (void )(so->so_proto->pr_usrreqs->pru_ready)(so, sfio->m, sfio->npages); CURVNET_RESTORE(); } /* XXXGL: curthread */ fdrop(sfio->sock_fp, curthread); free(sfio, M_TEMP); } /* * Iterate through pages vector and request paging for non-valid pages. */ static int sendfile_swapin(vm_object_t obj, struct sf_io *sfio, off_t off, off_t len, int npages, int rhpages, int flags) { vm_page_t *pa = sfio->pa; int nios; nios = 0; flags = (flags & SF_NODISKIO) ? VM_ALLOC_NOWAIT : 0; /* * First grab all the pages and wire them. Note that we grab * only required pages. Readahead pages are dealt with later. */ VM_OBJECT_WLOCK(obj); for (int i = 0; i < npages; i++) { pa[i] = vm_page_grab(obj, OFF_TO_IDX(vmoff(i, off)), VM_ALLOC_WIRED | VM_ALLOC_NORMAL | flags); if (pa[i] == NULL) { npages = i; rhpages = 0; break; } } for (int i = 0; i < npages;) { int j, a, count, rv; /* Skip valid pages. */ if (vm_page_is_valid(pa[i], vmoff(i, off) & PAGE_MASK, xfsize(i, npages, off, len))) { vm_page_xunbusy(pa[i]); SFSTAT_INC(sf_pages_valid); i++; continue; } /* * Now 'i' points to first invalid page, iterate further * to make 'j' point at first valid after a bunch of * invalid ones. */ for (j = i + 1; j < npages; j++) if (vm_page_is_valid(pa[j], vmoff(j, off) & PAGE_MASK, xfsize(j, npages, off, len))) { SFSTAT_INC(sf_pages_valid); break; } /* * Now we got region of invalid pages between 'i' and 'j'. * Check that they belong to pager. They may not be there, * which is a regular situation for shmem pager. For vnode * pager this happens only in case of sparse file. * * Important feature of vm_pager_has_page() is the hint * stored in 'a', about how many pages we can pagein after * this page in a single I/O. */ while (!vm_pager_has_page(obj, OFF_TO_IDX(vmoff(i, off)), NULL, &a) && i < j) { pmap_zero_page(pa[i]); pa[i]->valid = VM_PAGE_BITS_ALL; pa[i]->dirty = 0; vm_page_xunbusy(pa[i]); i++; } if (i == j) continue; /* * We want to pagein as many pages as possible, limited only * by the 'a' hint and actual request. * * We should not pagein into already valid page, thus if * 'j' didn't reach last page, trim by that page. * * When the pagein fulfils the request, also specify readahead. */ if (j < npages) a = min(a, j - i - 1); count = min(a + 1, npages - i); refcount_acquire(&sfio->nios); rv = vm_pager_get_pages_async(obj, pa + i, count, NULL, i + count == npages ? &rhpages : NULL, &sendfile_iodone, sfio); KASSERT(rv == VM_PAGER_OK, ("%s: pager fail obj %p page %p", __func__, obj, pa[i])); SFSTAT_INC(sf_iocnt); SFSTAT_ADD(sf_pages_read, count); if (i + count == npages) SFSTAT_ADD(sf_rhpages_read, rhpages); #ifdef INVARIANTS for (j = i; j < i + count && j < npages; j++) KASSERT(pa[j] == vm_page_lookup(obj, OFF_TO_IDX(vmoff(j, off))), ("pa[j] %p lookup %p\n", pa[j], vm_page_lookup(obj, OFF_TO_IDX(vmoff(j, off))))); #endif i += count; nios++; } VM_OBJECT_WUNLOCK(obj); if (nios == 0 && npages != 0) SFSTAT_INC(sf_noiocnt); return (nios); } static int sendfile_getobj(struct thread *td, struct file *fp, vm_object_t *obj_res, struct vnode **vp_res, struct shmfd **shmfd_res, off_t *obj_size, int *bsize) { struct vattr va; vm_object_t obj; struct vnode *vp; struct shmfd *shmfd; int error; vp = *vp_res = NULL; obj = NULL; shmfd = *shmfd_res = NULL; *bsize = 0; /* * The file descriptor must be a regular file and have a * backing VM object. */ if (fp->f_type == DTYPE_VNODE) { vp = fp->f_vnode; vn_lock(vp, LK_SHARED | LK_RETRY); if (vp->v_type != VREG) { error = EINVAL; goto out; } *bsize = vp->v_mount->mnt_stat.f_iosize; error = VOP_GETATTR(vp, &va, td->td_ucred); if (error != 0) goto out; *obj_size = va.va_size; obj = vp->v_object; if (obj == NULL) { error = EINVAL; goto out; } } else if (fp->f_type == DTYPE_SHM) { error = 0; shmfd = fp->f_data; obj = shmfd->shm_object; *obj_size = shmfd->shm_size; } else { error = EINVAL; goto out; } VM_OBJECT_WLOCK(obj); if ((obj->flags & OBJ_DEAD) != 0) { VM_OBJECT_WUNLOCK(obj); error = EBADF; goto out; } /* * Temporarily increase the backing VM object's reference * count so that a forced reclamation of its vnode does not * immediately destroy it. */ vm_object_reference_locked(obj); VM_OBJECT_WUNLOCK(obj); *obj_res = obj; *vp_res = vp; *shmfd_res = shmfd; out: if (vp != NULL) VOP_UNLOCK(vp, 0); return (error); } static int sendfile_getsock(struct thread *td, int s, struct file **sock_fp, struct socket **so) { cap_rights_t rights; int error; *sock_fp = NULL; *so = NULL; /* * The socket must be a stream socket and connected. */ error = getsock_cap(td, s, cap_rights_init(&rights, CAP_SEND), sock_fp, NULL); if (error != 0) return (error); *so = (*sock_fp)->f_data; if ((*so)->so_type != SOCK_STREAM) return (EINVAL); if (((*so)->so_state & SS_ISCONNECTED) == 0) return (ENOTCONN); return (0); } int vn_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, struct thread *td) { struct file *sock_fp; struct vnode *vp; struct vm_object *obj; struct socket *so; struct mbuf *m, *mh, *mhtail; struct sf_buf *sf; struct shmfd *shmfd; struct sendfile_sync *sfs; struct vattr va; off_t off, sbytes, rem, obj_size; int error, softerr, bsize, hdrlen; obj = NULL; so = NULL; m = mh = NULL; sfs = NULL; hdrlen = sbytes = 0; softerr = 0; error = sendfile_getobj(td, fp, &obj, &vp, &shmfd, &obj_size, &bsize); if (error != 0) return (error); error = sendfile_getsock(td, sockfd, &sock_fp, &so); if (error != 0) goto out; #ifdef MAC error = mac_socket_check_send(td->td_ucred, so); if (error != 0) goto out; #endif SFSTAT_INC(sf_syscalls); SFSTAT_ADD(sf_rhpages_requested, SF_READAHEAD(flags)); if (flags & SF_SYNC) { sfs = malloc(sizeof *sfs, M_TEMP, M_WAITOK | M_ZERO); mtx_init(&sfs->mtx, "sendfile", NULL, MTX_DEF); cv_init(&sfs->cv, "sendfile"); } rem = nbytes ? omin(nbytes, obj_size - offset) : obj_size - offset; /* * Protect against multiple writers to the socket. * * XXXRW: Historically this has assumed non-interruptibility, so now * we implement that, but possibly shouldn't. */ (void)sblock(&so->so_snd, SBL_WAIT | SBL_NOINTR); /* * Loop through the pages of the file, starting with the requested * offset. Get a file page (do I/O if necessary), map the file page * into an sf_buf, attach an mbuf header to the sf_buf, and queue * it on the socket. * This is done in two loops. The inner loop turns as many pages * as it can, up to available socket buffer space, without blocking * into mbufs to have it bulk delivered into the socket send buffer. * The outer loop checks the state and available space of the socket * and takes care of the overall progress. */ for (off = offset; rem > 0; ) { struct sf_io *sfio; vm_page_t *pa; struct mbuf *mtail; int nios, space, npages, rhpages; mtail = NULL; /* * Check the socket state for ongoing connection, * no errors and space in socket buffer. * If space is low allow for the remainder of the * file to be processed if it fits the socket buffer. * Otherwise block in waiting for sufficient space * to proceed, or if the socket is nonblocking, return * to userland with EAGAIN while reporting how far * we've come. * We wait until the socket buffer has significant free * space to do bulk sends. This makes good use of file * system read ahead and allows packet segmentation * offloading hardware to take over lots of work. If * we were not careful here we would send off only one * sfbuf at a time. */ SOCKBUF_LOCK(&so->so_snd); if (so->so_snd.sb_lowat < so->so_snd.sb_hiwat / 2) so->so_snd.sb_lowat = so->so_snd.sb_hiwat / 2; retry_space: if (so->so_snd.sb_state & SBS_CANTSENDMORE) { error = EPIPE; SOCKBUF_UNLOCK(&so->so_snd); goto done; } else if (so->so_error) { error = so->so_error; so->so_error = 0; SOCKBUF_UNLOCK(&so->so_snd); goto done; } space = sbspace(&so->so_snd); if (space < rem && (space <= 0 || space < so->so_snd.sb_lowat)) { if (so->so_state & SS_NBIO) { SOCKBUF_UNLOCK(&so->so_snd); error = EAGAIN; goto done; } /* * sbwait drops the lock while sleeping. * When we loop back to retry_space the * state may have changed and we retest * for it. */ error = sbwait(&so->so_snd); /* * An error from sbwait usually indicates that we've * been interrupted by a signal. If we've sent anything * then return bytes sent, otherwise return the error. */ if (error != 0) { SOCKBUF_UNLOCK(&so->so_snd); goto done; } goto retry_space; } SOCKBUF_UNLOCK(&so->so_snd); /* * At the beginning of the first loop check if any headers * are specified and copy them into mbufs. Reduce space in * the socket buffer by the size of the header mbuf chain. * Clear hdr_uio here and hdrlen at the end of the first loop. */ if (hdr_uio != NULL && hdr_uio->uio_resid > 0) { hdr_uio->uio_td = td; hdr_uio->uio_rw = UIO_WRITE; hdr_uio->uio_resid = min(hdr_uio->uio_resid, space); mh = m_uiotombuf(hdr_uio, M_WAITOK, 0, 0, 0); hdrlen = m_length(mh, &mhtail); space -= hdrlen; hdr_uio = NULL; } if (vp != NULL) { error = vn_lock(vp, LK_SHARED); if (error != 0) goto done; error = VOP_GETATTR(vp, &va, td->td_ucred); if (error != 0 || off >= va.va_size) { VOP_UNLOCK(vp, 0); goto done; } if (va.va_size != obj_size) { if (nbytes == 0) rem += va.va_size - obj_size; else if (offset + nbytes > va.va_size) rem -= (offset + nbytes - va.va_size); obj_size = va.va_size; } } if (space > rem) space = rem; npages = howmany(space + (off & PAGE_MASK), PAGE_SIZE); /* * Calculate maximum allowed number of pages for readahead * at this iteration. First, we allow readahead up to "rem". * If application wants more, let it be, but there is no * reason to go above MAXPHYS. Also check against "obj_size", * since vm_pager_has_page() can hint beyond EOF. */ rhpages = howmany(rem + (off & PAGE_MASK), PAGE_SIZE) - npages; rhpages += SF_READAHEAD(flags); rhpages = min(howmany(MAXPHYS, PAGE_SIZE), rhpages); rhpages = min(howmany(obj_size - trunc_page(off), PAGE_SIZE) - npages, rhpages); sfio = malloc(sizeof(struct sf_io) + npages * sizeof(vm_page_t), M_TEMP, M_WAITOK); refcount_init(&sfio->nios, 1); sfio->error = 0; nios = sendfile_swapin(obj, sfio, off, space, npages, rhpages, flags); /* * Loop and construct maximum sized mbuf chain to be bulk * dumped into socket buffer. */ pa = sfio->pa; for (int i = 0; i < npages; i++) { struct mbuf *m0; /* * If a page wasn't grabbed successfully, then * trim the array. Can happen only with SF_NODISKIO. */ if (pa[i] == NULL) { SFSTAT_INC(sf_busy); fixspace(npages, i, off, &space); npages = i; softerr = EBUSY; break; } /* * Get a sendfile buf. When allocating the * first buffer for mbuf chain, we usually * wait as long as necessary, but this wait * can be interrupted. For consequent * buffers, do not sleep, since several * threads might exhaust the buffers and then * deadlock. */ sf = sf_buf_alloc(pa[i], m != NULL ? SFB_NOWAIT : SFB_CATCH); if (sf == NULL) { SFSTAT_INC(sf_allocfail); for (int j = i; j < npages; j++) { vm_page_lock(pa[j]); vm_page_unwire(pa[j], PQ_INACTIVE); vm_page_unlock(pa[j]); } if (m == NULL) softerr = ENOBUFS; fixspace(npages, i, off, &space); npages = i; break; } m0 = m_get(M_WAITOK, MT_DATA); m0->m_ext.ext_buf = (char *)sf_buf_kva(sf); m0->m_ext.ext_size = PAGE_SIZE; m0->m_ext.ext_arg1 = sf; m0->m_ext.ext_arg2 = sfs; /* * SF_NOCACHE sets the page as being freed upon send. * However, we ignore it for the last page in 'space', * if the page is truncated, and we got more data to * send (rem > space), or if we have readahead * configured (rhpages > 0). */ if ((flags & SF_NOCACHE) == 0 || (i == npages - 1 && ((off + space) & PAGE_MASK) && (rem > space || rhpages > 0))) m0->m_ext.ext_type = EXT_SFBUF; else m0->m_ext.ext_type = EXT_SFBUF_NOCACHE; m0->m_ext.ext_flags = EXT_FLAG_EMBREF; m0->m_ext.ext_count = 1; m0->m_flags |= (M_EXT | M_RDONLY); if (nios) m0->m_flags |= M_NOTREADY; m0->m_data = (char *)sf_buf_kva(sf) + (vmoff(i, off) & PAGE_MASK); m0->m_len = xfsize(i, npages, off, space); if (i == 0) sfio->m = m0; /* Append to mbuf chain. */ if (mtail != NULL) mtail->m_next = m0; else m = m0; mtail = m0; if (sfs != NULL) { mtx_lock(&sfs->mtx); sfs->count++; mtx_unlock(&sfs->mtx); } } if (vp != NULL) VOP_UNLOCK(vp, 0); /* Keep track of bytes processed. */ off += space; rem -= space; /* Prepend header, if any. */ if (hdrlen) { mhtail->m_next = m; m = mh; mh = NULL; } if (m == NULL) { KASSERT(softerr, ("%s: m NULL, no error", __func__)); error = softerr; free(sfio, M_TEMP); goto done; } /* Add the buffer chain to the socket buffer. */ KASSERT(m_length(m, NULL) == space + hdrlen, ("%s: mlen %u space %d hdrlen %d", __func__, m_length(m, NULL), space, hdrlen)); CURVNET_SET(so->so_vnet); if (nios == 0) { /* * If sendfile_swapin() didn't initiate any I/Os, * which happens if all data is cached in VM, then * we can send data right now without the * PRUS_NOTREADY flag. */ free(sfio, M_TEMP); error = (*so->so_proto->pr_usrreqs->pru_send) (so, 0, m, NULL, NULL, td); } else { sfio->sock_fp = sock_fp; sfio->npages = npages; fhold(sock_fp); error = (*so->so_proto->pr_usrreqs->pru_send) (so, PRUS_NOTREADY, m, NULL, NULL, td); sendfile_iodone(sfio, NULL, 0, 0); } CURVNET_RESTORE(); m = NULL; /* pru_send always consumes */ if (error) goto done; sbytes += space + hdrlen; if (hdrlen) hdrlen = 0; if (softerr) { error = softerr; goto done; } } /* * Send trailers. Wimp out and use writev(2). */ if (trl_uio != NULL) { sbunlock(&so->so_snd); error = kern_writev(td, sockfd, trl_uio); if (error == 0) sbytes += td->td_retval[0]; goto out; } done: sbunlock(&so->so_snd); out: /* * If there was no error we have to clear td->td_retval[0] * because it may have been set by writev. */ if (error == 0) { td->td_retval[0] = 0; } if (sent != NULL) { (*sent) = sbytes; } if (obj != NULL) vm_object_deallocate(obj); if (so) fdrop(sock_fp, td); if (m) m_freem(m); if (mh) m_freem(mh); if (sfs != NULL) { mtx_lock(&sfs->mtx); if (sfs->count != 0) cv_wait(&sfs->cv, &sfs->mtx); KASSERT(sfs->count == 0, ("sendfile sync still busy")); cv_destroy(&sfs->cv); mtx_destroy(&sfs->mtx); free(sfs, M_TEMP); } if (error == ERESTART) error = EINTR; return (error); } static int sendfile(struct thread *td, struct sendfile_args *uap, int compat) { struct sf_hdtr hdtr; struct uio *hdr_uio, *trl_uio; struct file *fp; cap_rights_t rights; off_t sbytes; int error; /* * File offset must be positive. If it goes beyond EOF * we send only the header/trailer and no payload data. */ if (uap->offset < 0) return (EINVAL); hdr_uio = trl_uio = NULL; if (uap->hdtr != NULL) { error = copyin(uap->hdtr, &hdtr, sizeof(hdtr)); if (error != 0) goto out; if (hdtr.headers != NULL) { error = copyinuio(hdtr.headers, hdtr.hdr_cnt, &hdr_uio); if (error != 0) goto out; #ifdef COMPAT_FREEBSD4 /* * In FreeBSD < 5.0 the nbytes to send also included * the header. If compat is specified subtract the * header size from nbytes. */ if (compat) { if (uap->nbytes > hdr_uio->uio_resid) uap->nbytes -= hdr_uio->uio_resid; else uap->nbytes = 0; } #endif } if (hdtr.trailers != NULL) { error = copyinuio(hdtr.trailers, hdtr.trl_cnt, &trl_uio); if (error != 0) goto out; } } AUDIT_ARG_FD(uap->fd); /* * sendfile(2) can start at any offset within a file so we require * CAP_READ+CAP_SEEK = CAP_PREAD. */ if ((error = fget_read(td, uap->fd, cap_rights_init(&rights, CAP_PREAD), &fp)) != 0) { goto out; } error = fo_sendfile(fp, uap->s, hdr_uio, trl_uio, uap->offset, uap->nbytes, &sbytes, uap->flags, td); fdrop(fp, td); if (uap->sbytes != NULL) copyout(&sbytes, uap->sbytes, sizeof(off_t)); out: free(hdr_uio, M_IOV); free(trl_uio, M_IOV); return (error); } /* * sendfile(2) * * int sendfile(int fd, int s, off_t offset, size_t nbytes, * struct sf_hdtr *hdtr, off_t *sbytes, int flags) * * Send a file specified by 'fd' and starting at 'offset' to a socket * specified by 's'. Send only 'nbytes' of the file or until EOF if nbytes == * 0. Optionally add a header and/or trailer to the socket output. If * specified, write the total number of bytes sent into *sbytes. */ int sys_sendfile(struct thread *td, struct sendfile_args *uap) { return (sendfile(td, uap, 0)); } #ifdef COMPAT_FREEBSD4 int freebsd4_sendfile(struct thread *td, struct freebsd4_sendfile_args *uap) { struct sendfile_args args; args.fd = uap->fd; args.s = uap->s; args.offset = uap->offset; args.nbytes = uap->nbytes; args.hdtr = uap->hdtr; args.sbytes = uap->sbytes; args.flags = uap->flags; return (sendfile(td, &args, 1)); } #endif /* COMPAT_FREEBSD4 */ Index: head/sys/kern/kern_shutdown.c =================================================================== --- head/sys/kern/kern_shutdown.c (revision 305831) +++ head/sys/kern/kern_shutdown.c (revision 305832) @@ -1,942 +1,942 @@ /*- * Copyright (c) 1986, 1988, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_shutdown.c 8.3 (Berkeley) 1/21/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include "opt_kdb.h" #include "opt_panic.h" #include "opt_sched.h" #include "opt_watchdog.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static MALLOC_DEFINE(M_DUMPER, "dumper", "dumper block buffer"); #ifndef PANIC_REBOOT_WAIT_TIME #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */ #endif static int panic_reboot_wait_time = PANIC_REBOOT_WAIT_TIME; SYSCTL_INT(_kern, OID_AUTO, panic_reboot_wait_time, CTLFLAG_RWTUN, &panic_reboot_wait_time, 0, "Seconds to wait before rebooting after a panic"); /* * Note that stdarg.h and the ANSI style va_start macro is used for both * ANSI and traditional C compilers. */ #include #ifdef KDB #ifdef KDB_UNATTENDED int debugger_on_panic = 0; #else int debugger_on_panic = 1; #endif SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic, CTLFLAG_RWTUN | CTLFLAG_SECURE, &debugger_on_panic, 0, "Run debugger on kernel panic"); #ifdef KDB_TRACE static int trace_on_panic = 1; #else static int trace_on_panic = 0; #endif SYSCTL_INT(_debug, OID_AUTO, trace_on_panic, CTLFLAG_RWTUN | CTLFLAG_SECURE, &trace_on_panic, 0, "Print stack trace on kernel panic"); #endif /* KDB */ static int sync_on_panic = 0; SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RWTUN, &sync_on_panic, 0, "Do a sync before rebooting from a panic"); static SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW, 0, "Shutdown environment"); #ifndef DIAGNOSTIC static int show_busybufs; #else static int show_busybufs = 1; #endif SYSCTL_INT(_kern_shutdown, OID_AUTO, show_busybufs, CTLFLAG_RW, &show_busybufs, 0, ""); int suspend_blocked = 0; SYSCTL_INT(_kern, OID_AUTO, suspend_blocked, CTLFLAG_RW, &suspend_blocked, 0, "Block suspend due to a pending shutdown"); /* * Variable panicstr contains argument to first call to panic; used as flag * to indicate that the kernel has already called panic. */ const char *panicstr; int dumping; /* system is dumping */ int rebooting; /* system is rebooting */ static struct dumperinfo dumper; /* our selected dumper */ /* Context information for dump-debuggers. */ static struct pcb dumppcb; /* Registers. */ lwpid_t dumptid; /* Thread ID. */ static struct cdevsw reroot_cdevsw = { .d_version = D_VERSION, .d_name = "reroot", }; static void poweroff_wait(void *, int); static void shutdown_halt(void *junk, int howto); static void shutdown_panic(void *junk, int howto); static void shutdown_reset(void *junk, int howto); static int kern_reroot(void); /* register various local shutdown events */ static void shutdown_conf(void *unused) { EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL, SHUTDOWN_PRI_FIRST); EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL, SHUTDOWN_PRI_LAST + 100); EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL, SHUTDOWN_PRI_LAST + 100); EVENTHANDLER_REGISTER(shutdown_final, shutdown_reset, NULL, SHUTDOWN_PRI_LAST + 200); } SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL); /* * The only reason this exists is to create the /dev/reroot/ directory, * used by reroot code in init(8) as a mountpoint for tmpfs. */ static void reroot_conf(void *unused) { int error; struct cdev *cdev; error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &cdev, &reroot_cdevsw, NULL, UID_ROOT, GID_WHEEL, 0600, "reroot/reroot"); if (error != 0) { printf("%s: failed to create device node, error %d", __func__, error); } } SYSINIT(reroot_conf, SI_SUB_DEVFS, SI_ORDER_ANY, reroot_conf, NULL); /* * The system call that results in a reboot. */ /* ARGSUSED */ int sys_reboot(struct thread *td, struct reboot_args *uap) { int error; error = 0; #ifdef MAC error = mac_system_check_reboot(td->td_ucred, uap->opt); #endif if (error == 0) error = priv_check(td, PRIV_REBOOT); if (error == 0) { if (uap->opt & RB_REROOT) { error = kern_reroot(); } else { mtx_lock(&Giant); kern_reboot(uap->opt); mtx_unlock(&Giant); } } return (error); } /* * Called by events that want to shut down.. e.g on a PC */ void shutdown_nice(int howto) { if (initproc != NULL) { /* Send a signal to init(8) and have it shutdown the world. */ PROC_LOCK(initproc); if (howto & RB_POWEROFF) kern_psignal(initproc, SIGUSR2); else if (howto & RB_HALT) kern_psignal(initproc, SIGUSR1); else kern_psignal(initproc, SIGINT); PROC_UNLOCK(initproc); } else { /* No init(8) running, so simply reboot. */ kern_reboot(howto | RB_NOSYNC); } } static void print_uptime(void) { int f; struct timespec ts; getnanouptime(&ts); printf("Uptime: "); f = 0; if (ts.tv_sec >= 86400) { printf("%ldd", (long)ts.tv_sec / 86400); ts.tv_sec %= 86400; f = 1; } if (f || ts.tv_sec >= 3600) { printf("%ldh", (long)ts.tv_sec / 3600); ts.tv_sec %= 3600; f = 1; } if (f || ts.tv_sec >= 60) { printf("%ldm", (long)ts.tv_sec / 60); ts.tv_sec %= 60; f = 1; } printf("%lds\n", (long)ts.tv_sec); } int doadump(boolean_t textdump) { boolean_t coredump; int error; error = 0; if (dumping) return (EBUSY); if (dumper.dumper == NULL) return (ENXIO); savectx(&dumppcb); dumptid = curthread->td_tid; dumping++; coredump = TRUE; #ifdef DDB if (textdump && textdump_pending) { coredump = FALSE; textdump_dumpsys(&dumper); } #endif if (coredump) error = dumpsys(&dumper); dumping--; return (error); } /* * Shutdown the system cleanly to prepare for reboot, halt, or power off. */ void kern_reboot(int howto) { static int once = 0; #if defined(SMP) /* * Bind us to CPU 0 so that all shutdown code runs there. Some * systems don't shutdown properly (i.e., ACPI power off) if we * run on another processor. */ if (!SCHEDULER_STOPPED()) { thread_lock(curthread); sched_bind(curthread, 0); thread_unlock(curthread); KASSERT(PCPU_GET(cpuid) == 0, ("boot: not running on cpu 0")); } #endif /* We're in the process of rebooting. */ rebooting = 1; /* We are out of the debugger now. */ kdb_active = 0; /* * Do any callouts that should be done BEFORE syncing the filesystems. */ EVENTHANDLER_INVOKE(shutdown_pre_sync, howto); /* * Now sync filesystems */ if (!cold && (howto & RB_NOSYNC) == 0 && once == 0) { once = 1; bufshutdown(show_busybufs); } print_uptime(); cngrab(); /* * Ok, now do things that assume all filesystem activity has * been completed. */ EVENTHANDLER_INVOKE(shutdown_post_sync, howto); if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping) doadump(TRUE); /* Now that we're going to really halt the system... */ EVENTHANDLER_INVOKE(shutdown_final, howto); for(;;) ; /* safety against shutdown_reset not working */ /* NOTREACHED */ } /* * The system call that results in changing the rootfs. */ static int kern_reroot(void) { struct vnode *oldrootvnode, *vp; struct mount *mp, *devmp; int error; if (curproc != initproc) return (EPERM); /* * Mark the filesystem containing currently-running executable * (the temporary copy of init(8)) busy. */ vp = curproc->p_textvp; error = vn_lock(vp, LK_SHARED); if (error != 0) return (error); mp = vp->v_mount; error = vfs_busy(mp, MBF_NOWAIT); if (error != 0) { vfs_ref(mp); VOP_UNLOCK(vp, 0); error = vfs_busy(mp, 0); vn_lock(vp, LK_SHARED | LK_RETRY); vfs_rel(mp); if (error != 0) { VOP_UNLOCK(vp, 0); return (ENOENT); } if (vp->v_iflag & VI_DOOMED) { VOP_UNLOCK(vp, 0); vfs_unbusy(mp); return (ENOENT); } } VOP_UNLOCK(vp, 0); /* * Remove the filesystem containing currently-running executable * from the mount list, to prevent it from being unmounted * by vfs_unmountall(), and to avoid confusing vfs_mountroot(). * * Also preserve /dev - forcibly unmounting it could cause driver * reinitialization. */ vfs_ref(rootdevmp); devmp = rootdevmp; rootdevmp = NULL; mtx_lock(&mountlist_mtx); TAILQ_REMOVE(&mountlist, mp, mnt_list); TAILQ_REMOVE(&mountlist, devmp, mnt_list); mtx_unlock(&mountlist_mtx); oldrootvnode = rootvnode; /* * Unmount everything except for the two filesystems preserved above. */ vfs_unmountall(); /* * Add /dev back; vfs_mountroot() will move it into its new place. */ mtx_lock(&mountlist_mtx); TAILQ_INSERT_HEAD(&mountlist, devmp, mnt_list); mtx_unlock(&mountlist_mtx); rootdevmp = devmp; vfs_rel(rootdevmp); /* * Mount the new rootfs. */ vfs_mountroot(); /* * Update all references to the old rootvnode. */ mountcheckdirs(oldrootvnode, rootvnode); /* * Add the temporary filesystem back and unbusy it. */ mtx_lock(&mountlist_mtx); TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); mtx_unlock(&mountlist_mtx); vfs_unbusy(mp); return (0); } /* * If the shutdown was a clean halt, behave accordingly. */ static void shutdown_halt(void *junk, int howto) { if (howto & RB_HALT) { printf("\n"); printf("The operating system has halted.\n"); printf("Please press any key to reboot.\n\n"); switch (cngetc()) { case -1: /* No console, just die */ cpu_halt(); /* NOTREACHED */ default: howto &= ~RB_HALT; break; } } } /* * Check to see if the system paniced, pause and then reboot * according to the specified delay. */ static void shutdown_panic(void *junk, int howto) { int loop; if (howto & RB_DUMP) { if (panic_reboot_wait_time != 0) { if (panic_reboot_wait_time != -1) { printf("Automatic reboot in %d seconds - " "press a key on the console to abort\n", panic_reboot_wait_time); for (loop = panic_reboot_wait_time * 10; loop > 0; --loop) { DELAY(1000 * 100); /* 1/10th second */ /* Did user type a key? */ if (cncheckc() != -1) break; } if (!loop) return; } } else { /* zero time specified - reboot NOW */ return; } printf("--> Press a key on the console to reboot,\n"); printf("--> or switch off the system now.\n"); cngetc(); } } /* * Everything done, now reset */ static void shutdown_reset(void *junk, int howto) { printf("Rebooting...\n"); DELAY(1000000); /* wait 1 sec for printf's to complete and be read */ /* * Acquiring smp_ipi_mtx here has a double effect: * - it disables interrupts avoiding CPU0 preemption * by fast handlers (thus deadlocking against other CPUs) * - it avoids deadlocks against smp_rendezvous() or, more * generally, threads busy-waiting, with this spinlock held, * and waiting for responses by threads on other CPUs * (ie. smp_tlb_shootdown()). * * For the !SMP case it just needs to handle the former problem. */ #ifdef SMP mtx_lock_spin(&smp_ipi_mtx); #else spinlock_enter(); #endif /* cpu_boot(howto); */ /* doesn't do anything at the moment */ cpu_reset(); /* NOTREACHED */ /* assuming reset worked */ } #if defined(WITNESS) || defined(INVARIANT_SUPPORT) static int kassert_warn_only = 0; #ifdef KDB static int kassert_do_kdb = 0; #endif #ifdef KTR static int kassert_do_ktr = 0; #endif static int kassert_do_log = 1; static int kassert_log_pps_limit = 4; static int kassert_log_mute_at = 0; static int kassert_log_panic_at = 0; static int kassert_warnings = 0; SYSCTL_NODE(_debug, OID_AUTO, kassert, CTLFLAG_RW, NULL, "kassert options"); SYSCTL_INT(_debug_kassert, OID_AUTO, warn_only, CTLFLAG_RWTUN, &kassert_warn_only, 0, "KASSERT triggers a panic (1) or just a warning (0)"); #ifdef KDB SYSCTL_INT(_debug_kassert, OID_AUTO, do_kdb, CTLFLAG_RWTUN, &kassert_do_kdb, 0, "KASSERT will enter the debugger"); #endif #ifdef KTR SYSCTL_UINT(_debug_kassert, OID_AUTO, do_ktr, CTLFLAG_RWTUN, &kassert_do_ktr, 0, "KASSERT does a KTR, set this to the KTRMASK you want"); #endif SYSCTL_INT(_debug_kassert, OID_AUTO, do_log, CTLFLAG_RWTUN, &kassert_do_log, 0, "KASSERT triggers a panic (1) or just a warning (0)"); SYSCTL_INT(_debug_kassert, OID_AUTO, warnings, CTLFLAG_RWTUN, &kassert_warnings, 0, "number of KASSERTs that have been triggered"); SYSCTL_INT(_debug_kassert, OID_AUTO, log_panic_at, CTLFLAG_RWTUN, &kassert_log_panic_at, 0, "max number of KASSERTS before we will panic"); SYSCTL_INT(_debug_kassert, OID_AUTO, log_pps_limit, CTLFLAG_RWTUN, &kassert_log_pps_limit, 0, "limit number of log messages per second"); SYSCTL_INT(_debug_kassert, OID_AUTO, log_mute_at, CTLFLAG_RWTUN, &kassert_log_mute_at, 0, "max number of KASSERTS to log"); static int kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_debug_kassert, OID_AUTO, kassert, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE, NULL, 0, kassert_sysctl_kassert, "I", "set to trigger a test kassert"); static int kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS) { int error, i; error = sysctl_wire_old_buffer(req, sizeof(int)); if (error == 0) { i = 0; error = sysctl_handle_int(oidp, &i, 0, req); } if (error != 0 || req->newptr == NULL) return (error); KASSERT(0, ("kassert_sysctl_kassert triggered kassert %d", i)); return (0); } /* * Called by KASSERT, this decides if we will panic * or if we will log via printf and/or ktr. */ void kassert_panic(const char *fmt, ...) { static char buf[256]; va_list ap; va_start(ap, fmt); (void)vsnprintf(buf, sizeof(buf), fmt, ap); va_end(ap); /* * panic if we're not just warning, or if we've exceeded * kassert_log_panic_at warnings. */ if (!kassert_warn_only || (kassert_log_panic_at > 0 && kassert_warnings >= kassert_log_panic_at)) { va_start(ap, fmt); vpanic(fmt, ap); /* NORETURN */ } #ifdef KTR if (kassert_do_ktr) CTR0(ktr_mask, buf); #endif /* KTR */ /* * log if we've not yet met the mute limit. */ if (kassert_do_log && (kassert_log_mute_at == 0 || kassert_warnings < kassert_log_mute_at)) { static struct timeval lasterr; static int curerr; if (ppsratecheck(&lasterr, &curerr, kassert_log_pps_limit)) { printf("KASSERT failed: %s\n", buf); kdb_backtrace(); } } #ifdef KDB if (kassert_do_kdb) { kdb_enter(KDB_WHY_KASSERT, buf); } #endif atomic_add_int(&kassert_warnings, 1); } #endif /* * Panic is called on unresolvable fatal errors. It prints "panic: mesg", * and then reboots. If we are called twice, then we avoid trying to sync * the disks as this often leads to recursive panics. */ void panic(const char *fmt, ...) { va_list ap; va_start(ap, fmt); vpanic(fmt, ap); } void vpanic(const char *fmt, va_list ap) { #ifdef SMP cpuset_t other_cpus; #endif struct thread *td = curthread; int bootopt, newpanic; static char buf[256]; spinlock_enter(); #ifdef SMP /* * stop_cpus_hard(other_cpus) should prevent multiple CPUs from * concurrently entering panic. Only the winner will proceed * further. */ if (panicstr == NULL && !kdb_active) { other_cpus = all_cpus; CPU_CLR(PCPU_GET(cpuid), &other_cpus); stop_cpus_hard(other_cpus); } /* * Ensure that the scheduler is stopped while panicking, even if panic * has been entered from kdb. */ td->td_stopsched = 1; #endif bootopt = RB_AUTOBOOT; newpanic = 0; if (panicstr) bootopt |= RB_NOSYNC; else { bootopt |= RB_DUMP; panicstr = fmt; newpanic = 1; } if (newpanic) { (void)vsnprintf(buf, sizeof(buf), fmt, ap); panicstr = buf; cngrab(); printf("panic: %s\n", buf); } else { printf("panic: "); vprintf(fmt, ap); printf("\n"); } #ifdef SMP printf("cpuid = %d\n", PCPU_GET(cpuid)); #endif #ifdef KDB if (newpanic && trace_on_panic) kdb_backtrace(); if (debugger_on_panic) kdb_enter(KDB_WHY_PANIC, "panic"); #endif /*thread_lock(td); */ td->td_flags |= TDF_INPANIC; /* thread_unlock(td); */ if (!sync_on_panic) bootopt |= RB_NOSYNC; kern_reboot(bootopt); } /* * Support for poweroff delay. * * Please note that setting this delay too short might power off your machine * before the write cache on your hard disk has been flushed, leading to * soft-updates inconsistencies. */ #ifndef POWEROFF_DELAY # define POWEROFF_DELAY 5000 #endif static int poweroff_delay = POWEROFF_DELAY; SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW, &poweroff_delay, 0, "Delay before poweroff to write disk caches (msec)"); static void poweroff_wait(void *junk, int howto) { if (!(howto & RB_POWEROFF) || poweroff_delay <= 0) return; DELAY(poweroff_delay * 1000); } /* * Some system processes (e.g. syncer) need to be stopped at appropriate * points in their main loops prior to a system shutdown, so that they * won't interfere with the shutdown process (e.g. by holding a disk buf * to cause sync to fail). For each of these system processes, register * shutdown_kproc() as a handler for one of shutdown events. */ static int kproc_shutdown_wait = 60; SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW, &kproc_shutdown_wait, 0, "Max wait time (sec) to stop for each process"); void kproc_shutdown(void *arg, int howto) { struct proc *p; int error; if (panicstr) return; p = (struct proc *)arg; printf("Waiting (max %d seconds) for system process `%s' to stop... ", kproc_shutdown_wait, p->p_comm); error = kproc_suspend(p, kproc_shutdown_wait * hz); if (error == EWOULDBLOCK) printf("timed out\n"); else printf("done\n"); } void kthread_shutdown(void *arg, int howto) { struct thread *td; int error; if (panicstr) return; td = (struct thread *)arg; printf("Waiting (max %d seconds) for system thread `%s' to stop... ", kproc_shutdown_wait, td->td_name); error = kthread_suspend(td, kproc_shutdown_wait * hz); if (error == EWOULDBLOCK) printf("timed out\n"); else printf("done\n"); } static char dumpdevname[sizeof(((struct cdev*)NULL)->si_name)]; SYSCTL_STRING(_kern_shutdown, OID_AUTO, dumpdevname, CTLFLAG_RD, dumpdevname, 0, "Device for kernel dumps"); /* Registration of dumpers */ int set_dumper(struct dumperinfo *di, const char *devname, struct thread *td) { size_t wantcopy; int error; error = priv_check(td, PRIV_SETDUMPER); if (error != 0) return (error); if (di == NULL) { if (dumper.blockbuf != NULL) free(dumper.blockbuf, M_DUMPER); bzero(&dumper, sizeof(dumper)); dumpdevname[0] = '\0'; return (0); } if (dumper.dumper != NULL) return (EBUSY); dumper = *di; wantcopy = strlcpy(dumpdevname, devname, sizeof(dumpdevname)); if (wantcopy >= sizeof(dumpdevname)) { printf("set_dumper: device name truncated from '%s' -> '%s'\n", devname, dumpdevname); } dumper.blockbuf = malloc(di->blocksize, M_DUMPER, M_WAITOK | M_ZERO); return (0); } /* Call dumper with bounds checking. */ int dump_write(struct dumperinfo *di, void *virtual, vm_offset_t physical, off_t offset, size_t length) { if (length != 0 && (offset < di->mediaoffset || offset - di->mediaoffset + length > di->mediasize)) { printf("Attempt to write outside dump device boundaries.\n" "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n", (intmax_t)offset, (intmax_t)di->mediaoffset, (uintmax_t)length, (intmax_t)di->mediasize); return (ENOSPC); } return (di->dumper(di->priv, virtual, physical, offset, length)); } /* Call dumper with bounds checking. */ int dump_write_pad(struct dumperinfo *di, void *virtual, vm_offset_t physical, off_t offset, size_t length, size_t *size) { char *temp; int ret; if (length > di->blocksize) return (ENOMEM); *size = di->blocksize; if (length == di->blocksize) temp = virtual; else { temp = di->blockbuf; memset(temp + length, 0, di->blocksize - length); memcpy(temp, virtual, length); } ret = dump_write(di, temp, physical, offset, *size); return (ret); } void mkdumpheader(struct kerneldumpheader *kdh, char *magic, uint32_t archver, uint64_t dumplen, uint32_t blksz) { bzero(kdh, sizeof(*kdh)); strlcpy(kdh->magic, magic, sizeof(kdh->magic)); strlcpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture)); kdh->version = htod32(KERNELDUMPVERSION); kdh->architectureversion = htod32(archver); kdh->dumplength = htod64(dumplen); kdh->dumptime = htod64(time_second); kdh->blocksize = htod32(blksz); strlcpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname)); strlcpy(kdh->versionstring, version, sizeof(kdh->versionstring)); if (panicstr != NULL) strlcpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring)); kdh->parity = kerneldump_parity(kdh); } #ifdef DDB DB_SHOW_COMMAND(panic, db_show_panic) { if (panicstr == NULL) db_printf("panicstr not set\n"); else db_printf("panic: %s\n", panicstr); } #endif Index: head/sys/kern/kern_sig.c =================================================================== --- head/sys/kern/kern_sig.c (revision 305831) +++ head/sys/kern/kern_sig.c (revision 305832) @@ -1,3676 +1,3676 @@ /*- * Copyright (c) 1982, 1986, 1989, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_sig.c 8.7 (Berkeley) 4/18/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #include "opt_gzio.h" #include "opt_ktrace.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define ONSIG 32 /* NSIG for osig* syscalls. XXX. */ SDT_PROVIDER_DECLARE(proc); SDT_PROBE_DEFINE3(proc, , , signal__send, "struct thread *", "struct proc *", "int"); SDT_PROBE_DEFINE2(proc, , , signal__clear, "int", "ksiginfo_t *"); SDT_PROBE_DEFINE3(proc, , , signal__discard, "struct thread *", "struct proc *", "int"); static int coredump(struct thread *); static int killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi); static int issignal(struct thread *td); static int sigprop(int sig); static void tdsigwakeup(struct thread *, int, sig_t, int); static int sig_suspend_threads(struct thread *, struct proc *, int); static int filt_sigattach(struct knote *kn); static void filt_sigdetach(struct knote *kn); static int filt_signal(struct knote *kn, long hint); static struct thread *sigtd(struct proc *p, int sig, int prop); static void sigqueue_start(void); static uma_zone_t ksiginfo_zone = NULL; struct filterops sig_filtops = { .f_isfd = 0, .f_attach = filt_sigattach, .f_detach = filt_sigdetach, .f_event = filt_signal, }; static int kern_logsigexit = 1; SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW, &kern_logsigexit, 0, "Log processes quitting on abnormal signals to syslog(3)"); static int kern_forcesigexit = 1; SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW, &kern_forcesigexit, 0, "Force trap signal to be handled"); static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW, 0, "POSIX real time signal"); static int max_pending_per_proc = 128; SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW, &max_pending_per_proc, 0, "Max pending signals per proc"); static int preallocate_siginfo = 1024; SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RDTUN, &preallocate_siginfo, 0, "Preallocated signal memory size"); static int signal_overflow = 0; SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD, &signal_overflow, 0, "Number of signals overflew"); static int signal_alloc_fail = 0; SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD, &signal_alloc_fail, 0, "signals failed to be allocated"); SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL); /* * Policy -- Can ucred cr1 send SIGIO to process cr2? * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG * in the right situations. */ #define CANSIGIO(cr1, cr2) \ ((cr1)->cr_uid == 0 || \ (cr1)->cr_ruid == (cr2)->cr_ruid || \ (cr1)->cr_uid == (cr2)->cr_ruid || \ (cr1)->cr_ruid == (cr2)->cr_uid || \ (cr1)->cr_uid == (cr2)->cr_uid) static int sugid_coredump; SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RWTUN, &sugid_coredump, 0, "Allow setuid and setgid processes to dump core"); static int capmode_coredump; SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RWTUN, &capmode_coredump, 0, "Allow processes in capability mode to dump core"); static int do_coredump = 1; SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW, &do_coredump, 0, "Enable/Disable coredumps"); static int set_core_nodump_flag = 0; SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag, 0, "Enable setting the NODUMP flag on coredump files"); static int coredump_devctl = 0; SYSCTL_INT(_kern, OID_AUTO, coredump_devctl, CTLFLAG_RW, &coredump_devctl, 0, "Generate a devctl notification when processes coredump"); /* * Signal properties and actions. * The array below categorizes the signals and their default actions * according to the following properties: */ #define SIGPROP_KILL 0x01 /* terminates process by default */ #define SIGPROP_CORE 0x02 /* ditto and coredumps */ #define SIGPROP_STOP 0x04 /* suspend process */ #define SIGPROP_TTYSTOP 0x08 /* ditto, from tty */ #define SIGPROP_IGNORE 0x10 /* ignore by default */ #define SIGPROP_CONT 0x20 /* continue if suspended */ #define SIGPROP_CANTMASK 0x40 /* non-maskable, catchable */ static int sigproptbl[NSIG] = { [SIGHUP] = SIGPROP_KILL, [SIGINT] = SIGPROP_KILL, [SIGQUIT] = SIGPROP_KILL | SIGPROP_CORE, [SIGILL] = SIGPROP_KILL | SIGPROP_CORE, [SIGTRAP] = SIGPROP_KILL | SIGPROP_CORE, [SIGABRT] = SIGPROP_KILL | SIGPROP_CORE, [SIGEMT] = SIGPROP_KILL | SIGPROP_CORE, [SIGFPE] = SIGPROP_KILL | SIGPROP_CORE, [SIGKILL] = SIGPROP_KILL, [SIGBUS] = SIGPROP_KILL | SIGPROP_CORE, [SIGSEGV] = SIGPROP_KILL | SIGPROP_CORE, [SIGSYS] = SIGPROP_KILL | SIGPROP_CORE, [SIGPIPE] = SIGPROP_KILL, [SIGALRM] = SIGPROP_KILL, [SIGTERM] = SIGPROP_KILL, [SIGURG] = SIGPROP_IGNORE, [SIGSTOP] = SIGPROP_STOP, [SIGTSTP] = SIGPROP_STOP | SIGPROP_TTYSTOP, [SIGCONT] = SIGPROP_IGNORE | SIGPROP_CONT, [SIGCHLD] = SIGPROP_IGNORE, [SIGTTIN] = SIGPROP_STOP | SIGPROP_TTYSTOP, [SIGTTOU] = SIGPROP_STOP | SIGPROP_TTYSTOP, [SIGIO] = SIGPROP_IGNORE, [SIGXCPU] = SIGPROP_KILL, [SIGXFSZ] = SIGPROP_KILL, [SIGVTALRM] = SIGPROP_KILL, [SIGPROF] = SIGPROP_KILL, [SIGWINCH] = SIGPROP_IGNORE, [SIGINFO] = SIGPROP_IGNORE, [SIGUSR1] = SIGPROP_KILL, [SIGUSR2] = SIGPROP_KILL, }; static void reschedule_signals(struct proc *p, sigset_t block, int flags); static void sigqueue_start(void) { ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); uma_prealloc(ksiginfo_zone, preallocate_siginfo); p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS); p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1); p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc); } ksiginfo_t * ksiginfo_alloc(int wait) { int flags; flags = M_ZERO; if (! wait) flags |= M_NOWAIT; if (ksiginfo_zone != NULL) return ((ksiginfo_t *)uma_zalloc(ksiginfo_zone, flags)); return (NULL); } void ksiginfo_free(ksiginfo_t *ksi) { uma_zfree(ksiginfo_zone, ksi); } static __inline int ksiginfo_tryfree(ksiginfo_t *ksi) { if (!(ksi->ksi_flags & KSI_EXT)) { uma_zfree(ksiginfo_zone, ksi); return (1); } return (0); } void sigqueue_init(sigqueue_t *list, struct proc *p) { SIGEMPTYSET(list->sq_signals); SIGEMPTYSET(list->sq_kill); TAILQ_INIT(&list->sq_list); list->sq_proc = p; list->sq_flags = SQ_INIT; } /* * Get a signal's ksiginfo. * Return: * 0 - signal not found * others - signal number */ static int sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si) { struct proc *p = sq->sq_proc; struct ksiginfo *ksi, *next; int count = 0; KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited")); if (!SIGISMEMBER(sq->sq_signals, signo)) return (0); if (SIGISMEMBER(sq->sq_kill, signo)) { count++; SIGDELSET(sq->sq_kill, signo); } TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) { if (ksi->ksi_signo == signo) { if (count == 0) { TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link); ksi->ksi_sigq = NULL; ksiginfo_copy(ksi, si); if (ksiginfo_tryfree(ksi) && p != NULL) p->p_pendingcnt--; } if (++count > 1) break; } } if (count <= 1) SIGDELSET(sq->sq_signals, signo); si->ksi_signo = signo; return (signo); } void sigqueue_take(ksiginfo_t *ksi) { struct ksiginfo *kp; struct proc *p; sigqueue_t *sq; if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL) return; p = sq->sq_proc; TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link); ksi->ksi_sigq = NULL; if (!(ksi->ksi_flags & KSI_EXT) && p != NULL) p->p_pendingcnt--; for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL; kp = TAILQ_NEXT(kp, ksi_link)) { if (kp->ksi_signo == ksi->ksi_signo) break; } if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo)) SIGDELSET(sq->sq_signals, ksi->ksi_signo); } static int sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si) { struct proc *p = sq->sq_proc; struct ksiginfo *ksi; int ret = 0; KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited")); if (signo == SIGKILL || signo == SIGSTOP || si == NULL) { SIGADDSET(sq->sq_kill, signo); goto out_set_bit; } /* directly insert the ksi, don't copy it */ if (si->ksi_flags & KSI_INS) { if (si->ksi_flags & KSI_HEAD) TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link); else TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link); si->ksi_sigq = sq; goto out_set_bit; } if (__predict_false(ksiginfo_zone == NULL)) { SIGADDSET(sq->sq_kill, signo); goto out_set_bit; } if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) { signal_overflow++; ret = EAGAIN; } else if ((ksi = ksiginfo_alloc(0)) == NULL) { signal_alloc_fail++; ret = EAGAIN; } else { if (p != NULL) p->p_pendingcnt++; ksiginfo_copy(si, ksi); ksi->ksi_signo = signo; if (si->ksi_flags & KSI_HEAD) TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link); else TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link); ksi->ksi_sigq = sq; } if ((si->ksi_flags & KSI_TRAP) != 0 || (si->ksi_flags & KSI_SIGQ) == 0) { if (ret != 0) SIGADDSET(sq->sq_kill, signo); ret = 0; goto out_set_bit; } if (ret != 0) return (ret); out_set_bit: SIGADDSET(sq->sq_signals, signo); return (ret); } void sigqueue_flush(sigqueue_t *sq) { struct proc *p = sq->sq_proc; ksiginfo_t *ksi; KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited")); if (p != NULL) PROC_LOCK_ASSERT(p, MA_OWNED); while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) { TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link); ksi->ksi_sigq = NULL; if (ksiginfo_tryfree(ksi) && p != NULL) p->p_pendingcnt--; } SIGEMPTYSET(sq->sq_signals); SIGEMPTYSET(sq->sq_kill); } static void sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set) { sigset_t tmp; struct proc *p1, *p2; ksiginfo_t *ksi, *next; KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited")); KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited")); p1 = src->sq_proc; p2 = dst->sq_proc; /* Move siginfo to target list */ TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) { if (SIGISMEMBER(*set, ksi->ksi_signo)) { TAILQ_REMOVE(&src->sq_list, ksi, ksi_link); if (p1 != NULL) p1->p_pendingcnt--; TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link); ksi->ksi_sigq = dst; if (p2 != NULL) p2->p_pendingcnt++; } } /* Move pending bits to target list */ tmp = src->sq_kill; SIGSETAND(tmp, *set); SIGSETOR(dst->sq_kill, tmp); SIGSETNAND(src->sq_kill, tmp); tmp = src->sq_signals; SIGSETAND(tmp, *set); SIGSETOR(dst->sq_signals, tmp); SIGSETNAND(src->sq_signals, tmp); } #if 0 static void sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo) { sigset_t set; SIGEMPTYSET(set); SIGADDSET(set, signo); sigqueue_move_set(src, dst, &set); } #endif static void sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set) { struct proc *p = sq->sq_proc; ksiginfo_t *ksi, *next; KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited")); /* Remove siginfo queue */ TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) { if (SIGISMEMBER(*set, ksi->ksi_signo)) { TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link); ksi->ksi_sigq = NULL; if (ksiginfo_tryfree(ksi) && p != NULL) p->p_pendingcnt--; } } SIGSETNAND(sq->sq_kill, *set); SIGSETNAND(sq->sq_signals, *set); } void sigqueue_delete(sigqueue_t *sq, int signo) { sigset_t set; SIGEMPTYSET(set); SIGADDSET(set, signo); sigqueue_delete_set(sq, &set); } /* Remove a set of signals for a process */ static void sigqueue_delete_set_proc(struct proc *p, const sigset_t *set) { sigqueue_t worklist; struct thread *td0; PROC_LOCK_ASSERT(p, MA_OWNED); sigqueue_init(&worklist, NULL); sigqueue_move_set(&p->p_sigqueue, &worklist, set); FOREACH_THREAD_IN_PROC(p, td0) sigqueue_move_set(&td0->td_sigqueue, &worklist, set); sigqueue_flush(&worklist); } void sigqueue_delete_proc(struct proc *p, int signo) { sigset_t set; SIGEMPTYSET(set); SIGADDSET(set, signo); sigqueue_delete_set_proc(p, &set); } static void sigqueue_delete_stopmask_proc(struct proc *p) { sigset_t set; SIGEMPTYSET(set); SIGADDSET(set, SIGSTOP); SIGADDSET(set, SIGTSTP); SIGADDSET(set, SIGTTIN); SIGADDSET(set, SIGTTOU); sigqueue_delete_set_proc(p, &set); } /* * Determine signal that should be delivered to thread td, the current * thread, 0 if none. If there is a pending stop signal with default * action, the process stops in issignal(). */ int cursig(struct thread *td) { PROC_LOCK_ASSERT(td->td_proc, MA_OWNED); mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED); THREAD_LOCK_ASSERT(td, MA_NOTOWNED); return (SIGPENDING(td) ? issignal(td) : 0); } /* * Arrange for ast() to handle unmasked pending signals on return to user * mode. This must be called whenever a signal is added to td_sigqueue or * unmasked in td_sigmask. */ void signotify(struct thread *td) { struct proc *p; p = td->td_proc; PROC_LOCK_ASSERT(p, MA_OWNED); if (SIGPENDING(td)) { thread_lock(td); td->td_flags |= TDF_NEEDSIGCHK | TDF_ASTPENDING; thread_unlock(td); } } int sigonstack(size_t sp) { struct thread *td = curthread; return ((td->td_pflags & TDP_ALTSTACK) ? #if defined(COMPAT_43) ((td->td_sigstk.ss_size == 0) ? (td->td_sigstk.ss_flags & SS_ONSTACK) : ((sp - (size_t)td->td_sigstk.ss_sp) < td->td_sigstk.ss_size)) #else ((sp - (size_t)td->td_sigstk.ss_sp) < td->td_sigstk.ss_size) #endif : 0); } static __inline int sigprop(int sig) { if (sig > 0 && sig < nitems(sigproptbl)) return (sigproptbl[sig]); return (0); } int sig_ffs(sigset_t *set) { int i; for (i = 0; i < _SIG_WORDS; i++) if (set->__bits[i]) return (ffs(set->__bits[i]) + (i * 32)); return (0); } static bool sigact_flag_test(const struct sigaction *act, int flag) { /* * SA_SIGINFO is reset when signal disposition is set to * ignore or default. Other flags are kept according to user * settings. */ return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO || ((__sighandler_t *)act->sa_sigaction != SIG_IGN && (__sighandler_t *)act->sa_sigaction != SIG_DFL))); } /* * kern_sigaction * sigaction * freebsd4_sigaction * osigaction */ int kern_sigaction(struct thread *td, int sig, const struct sigaction *act, struct sigaction *oact, int flags) { struct sigacts *ps; struct proc *p = td->td_proc; if (!_SIG_VALID(sig)) return (EINVAL); if (act != NULL && act->sa_handler != SIG_DFL && act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK | SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER | SA_NOCLDWAIT | SA_SIGINFO)) != 0) return (EINVAL); PROC_LOCK(p); ps = p->p_sigacts; mtx_lock(&ps->ps_mtx); if (oact) { oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)]; oact->sa_flags = 0; if (SIGISMEMBER(ps->ps_sigonstack, sig)) oact->sa_flags |= SA_ONSTACK; if (!SIGISMEMBER(ps->ps_sigintr, sig)) oact->sa_flags |= SA_RESTART; if (SIGISMEMBER(ps->ps_sigreset, sig)) oact->sa_flags |= SA_RESETHAND; if (SIGISMEMBER(ps->ps_signodefer, sig)) oact->sa_flags |= SA_NODEFER; if (SIGISMEMBER(ps->ps_siginfo, sig)) { oact->sa_flags |= SA_SIGINFO; oact->sa_sigaction = (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)]; } else oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)]; if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP) oact->sa_flags |= SA_NOCLDSTOP; if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT) oact->sa_flags |= SA_NOCLDWAIT; } if (act) { if ((sig == SIGKILL || sig == SIGSTOP) && act->sa_handler != SIG_DFL) { mtx_unlock(&ps->ps_mtx); PROC_UNLOCK(p); return (EINVAL); } /* * Change setting atomically. */ ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask; SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]); if (sigact_flag_test(act, SA_SIGINFO)) { ps->ps_sigact[_SIG_IDX(sig)] = (__sighandler_t *)act->sa_sigaction; SIGADDSET(ps->ps_siginfo, sig); } else { ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler; SIGDELSET(ps->ps_siginfo, sig); } if (!sigact_flag_test(act, SA_RESTART)) SIGADDSET(ps->ps_sigintr, sig); else SIGDELSET(ps->ps_sigintr, sig); if (sigact_flag_test(act, SA_ONSTACK)) SIGADDSET(ps->ps_sigonstack, sig); else SIGDELSET(ps->ps_sigonstack, sig); if (sigact_flag_test(act, SA_RESETHAND)) SIGADDSET(ps->ps_sigreset, sig); else SIGDELSET(ps->ps_sigreset, sig); if (sigact_flag_test(act, SA_NODEFER)) SIGADDSET(ps->ps_signodefer, sig); else SIGDELSET(ps->ps_signodefer, sig); if (sig == SIGCHLD) { if (act->sa_flags & SA_NOCLDSTOP) ps->ps_flag |= PS_NOCLDSTOP; else ps->ps_flag &= ~PS_NOCLDSTOP; if (act->sa_flags & SA_NOCLDWAIT) { /* * Paranoia: since SA_NOCLDWAIT is implemented * by reparenting the dying child to PID 1 (and * trust it to reap the zombie), PID 1 itself * is forbidden to set SA_NOCLDWAIT. */ if (p->p_pid == 1) ps->ps_flag &= ~PS_NOCLDWAIT; else ps->ps_flag |= PS_NOCLDWAIT; } else ps->ps_flag &= ~PS_NOCLDWAIT; if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN) ps->ps_flag |= PS_CLDSIGIGN; else ps->ps_flag &= ~PS_CLDSIGIGN; } /* * Set bit in ps_sigignore for signals that are set to SIG_IGN, * and for signals set to SIG_DFL where the default is to * ignore. However, don't put SIGCONT in ps_sigignore, as we * have to restart the process. */ if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN || (sigprop(sig) & SIGPROP_IGNORE && ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) { /* never to be seen again */ sigqueue_delete_proc(p, sig); if (sig != SIGCONT) /* easier in psignal */ SIGADDSET(ps->ps_sigignore, sig); SIGDELSET(ps->ps_sigcatch, sig); } else { SIGDELSET(ps->ps_sigignore, sig); if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL) SIGDELSET(ps->ps_sigcatch, sig); else SIGADDSET(ps->ps_sigcatch, sig); } #ifdef COMPAT_FREEBSD4 if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN || ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL || (flags & KSA_FREEBSD4) == 0) SIGDELSET(ps->ps_freebsd4, sig); else SIGADDSET(ps->ps_freebsd4, sig); #endif #ifdef COMPAT_43 if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN || ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL || (flags & KSA_OSIGSET) == 0) SIGDELSET(ps->ps_osigset, sig); else SIGADDSET(ps->ps_osigset, sig); #endif } mtx_unlock(&ps->ps_mtx); PROC_UNLOCK(p); return (0); } #ifndef _SYS_SYSPROTO_H_ struct sigaction_args { int sig; struct sigaction *act; struct sigaction *oact; }; #endif int sys_sigaction(td, uap) struct thread *td; register struct sigaction_args *uap; { struct sigaction act, oact; register struct sigaction *actp, *oactp; int error; actp = (uap->act != NULL) ? &act : NULL; oactp = (uap->oact != NULL) ? &oact : NULL; if (actp) { error = copyin(uap->act, actp, sizeof(act)); if (error) return (error); } error = kern_sigaction(td, uap->sig, actp, oactp, 0); if (oactp && !error) error = copyout(oactp, uap->oact, sizeof(oact)); return (error); } #ifdef COMPAT_FREEBSD4 #ifndef _SYS_SYSPROTO_H_ struct freebsd4_sigaction_args { int sig; struct sigaction *act; struct sigaction *oact; }; #endif int freebsd4_sigaction(td, uap) struct thread *td; register struct freebsd4_sigaction_args *uap; { struct sigaction act, oact; register struct sigaction *actp, *oactp; int error; actp = (uap->act != NULL) ? &act : NULL; oactp = (uap->oact != NULL) ? &oact : NULL; if (actp) { error = copyin(uap->act, actp, sizeof(act)); if (error) return (error); } error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4); if (oactp && !error) error = copyout(oactp, uap->oact, sizeof(oact)); return (error); } #endif /* COMAPT_FREEBSD4 */ #ifdef COMPAT_43 /* XXX - COMPAT_FBSD3 */ #ifndef _SYS_SYSPROTO_H_ struct osigaction_args { int signum; struct osigaction *nsa; struct osigaction *osa; }; #endif int osigaction(td, uap) struct thread *td; register struct osigaction_args *uap; { struct osigaction sa; struct sigaction nsa, osa; register struct sigaction *nsap, *osap; int error; if (uap->signum <= 0 || uap->signum >= ONSIG) return (EINVAL); nsap = (uap->nsa != NULL) ? &nsa : NULL; osap = (uap->osa != NULL) ? &osa : NULL; if (nsap) { error = copyin(uap->nsa, &sa, sizeof(sa)); if (error) return (error); nsap->sa_handler = sa.sa_handler; nsap->sa_flags = sa.sa_flags; OSIG2SIG(sa.sa_mask, nsap->sa_mask); } error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET); if (osap && !error) { sa.sa_handler = osap->sa_handler; sa.sa_flags = osap->sa_flags; SIG2OSIG(osap->sa_mask, sa.sa_mask); error = copyout(&sa, uap->osa, sizeof(sa)); } return (error); } #if !defined(__i386__) /* Avoid replicating the same stub everywhere */ int osigreturn(td, uap) struct thread *td; struct osigreturn_args *uap; { return (nosys(td, (struct nosys_args *)uap)); } #endif #endif /* COMPAT_43 */ /* * Initialize signal state for process 0; * set to ignore signals that are ignored by default. */ void siginit(p) struct proc *p; { register int i; struct sigacts *ps; PROC_LOCK(p); ps = p->p_sigacts; mtx_lock(&ps->ps_mtx); for (i = 1; i <= NSIG; i++) { if (sigprop(i) & SIGPROP_IGNORE && i != SIGCONT) { SIGADDSET(ps->ps_sigignore, i); } } mtx_unlock(&ps->ps_mtx); PROC_UNLOCK(p); } /* * Reset specified signal to the default disposition. */ static void sigdflt(struct sigacts *ps, int sig) { mtx_assert(&ps->ps_mtx, MA_OWNED); SIGDELSET(ps->ps_sigcatch, sig); if ((sigprop(sig) & SIGPROP_IGNORE) != 0 && sig != SIGCONT) SIGADDSET(ps->ps_sigignore, sig); ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL; SIGDELSET(ps->ps_siginfo, sig); } /* * Reset signals for an exec of the specified process. */ void execsigs(struct proc *p) { sigset_t osigignore; struct sigacts *ps; int sig; struct thread *td; /* * Reset caught signals. Held signals remain held * through td_sigmask (unless they were caught, * and are now ignored by default). */ PROC_LOCK_ASSERT(p, MA_OWNED); td = FIRST_THREAD_IN_PROC(p); ps = p->p_sigacts; mtx_lock(&ps->ps_mtx); while (SIGNOTEMPTY(ps->ps_sigcatch)) { sig = sig_ffs(&ps->ps_sigcatch); sigdflt(ps, sig); if ((sigprop(sig) & SIGPROP_IGNORE) != 0) sigqueue_delete_proc(p, sig); } /* * As CloudABI processes cannot modify signal handlers, fully * reset all signals to their default behavior. Do ignore * SIGPIPE, as it would otherwise be impossible to recover from * writes to broken pipes and sockets. */ if (SV_PROC_ABI(p) == SV_ABI_CLOUDABI) { osigignore = ps->ps_sigignore; while (SIGNOTEMPTY(osigignore)) { sig = sig_ffs(&osigignore); SIGDELSET(osigignore, sig); if (sig != SIGPIPE) sigdflt(ps, sig); } SIGADDSET(ps->ps_sigignore, SIGPIPE); } /* * Reset stack state to the user stack. * Clear set of signals caught on the signal stack. */ td->td_sigstk.ss_flags = SS_DISABLE; td->td_sigstk.ss_size = 0; td->td_sigstk.ss_sp = 0; td->td_pflags &= ~TDP_ALTSTACK; /* * Reset no zombies if child dies flag as Solaris does. */ ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN); if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN) ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL; mtx_unlock(&ps->ps_mtx); } /* * kern_sigprocmask() * * Manipulate signal mask. */ int kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset, int flags) { sigset_t new_block, oset1; struct proc *p; int error; p = td->td_proc; if ((flags & SIGPROCMASK_PROC_LOCKED) != 0) PROC_LOCK_ASSERT(p, MA_OWNED); else PROC_LOCK(p); mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0 ? MA_OWNED : MA_NOTOWNED); if (oset != NULL) *oset = td->td_sigmask; error = 0; if (set != NULL) { switch (how) { case SIG_BLOCK: SIG_CANTMASK(*set); oset1 = td->td_sigmask; SIGSETOR(td->td_sigmask, *set); new_block = td->td_sigmask; SIGSETNAND(new_block, oset1); break; case SIG_UNBLOCK: SIGSETNAND(td->td_sigmask, *set); signotify(td); goto out; case SIG_SETMASK: SIG_CANTMASK(*set); oset1 = td->td_sigmask; if (flags & SIGPROCMASK_OLD) SIGSETLO(td->td_sigmask, *set); else td->td_sigmask = *set; new_block = td->td_sigmask; SIGSETNAND(new_block, oset1); signotify(td); break; default: error = EINVAL; goto out; } /* * The new_block set contains signals that were not previously * blocked, but are blocked now. * * In case we block any signal that was not previously blocked * for td, and process has the signal pending, try to schedule * signal delivery to some thread that does not block the * signal, possibly waking it up. */ if (p->p_numthreads != 1) reschedule_signals(p, new_block, flags); } out: if (!(flags & SIGPROCMASK_PROC_LOCKED)) PROC_UNLOCK(p); return (error); } #ifndef _SYS_SYSPROTO_H_ struct sigprocmask_args { int how; const sigset_t *set; sigset_t *oset; }; #endif int sys_sigprocmask(td, uap) register struct thread *td; struct sigprocmask_args *uap; { sigset_t set, oset; sigset_t *setp, *osetp; int error; setp = (uap->set != NULL) ? &set : NULL; osetp = (uap->oset != NULL) ? &oset : NULL; if (setp) { error = copyin(uap->set, setp, sizeof(set)); if (error) return (error); } error = kern_sigprocmask(td, uap->how, setp, osetp, 0); if (osetp && !error) { error = copyout(osetp, uap->oset, sizeof(oset)); } return (error); } #ifdef COMPAT_43 /* XXX - COMPAT_FBSD3 */ #ifndef _SYS_SYSPROTO_H_ struct osigprocmask_args { int how; osigset_t mask; }; #endif int osigprocmask(td, uap) register struct thread *td; struct osigprocmask_args *uap; { sigset_t set, oset; int error; OSIG2SIG(uap->mask, set); error = kern_sigprocmask(td, uap->how, &set, &oset, 1); SIG2OSIG(oset, td->td_retval[0]); return (error); } #endif /* COMPAT_43 */ int sys_sigwait(struct thread *td, struct sigwait_args *uap) { ksiginfo_t ksi; sigset_t set; int error; error = copyin(uap->set, &set, sizeof(set)); if (error) { td->td_retval[0] = error; return (0); } error = kern_sigtimedwait(td, set, &ksi, NULL); if (error) { if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT) error = ERESTART; if (error == ERESTART) return (error); td->td_retval[0] = error; return (0); } error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo)); td->td_retval[0] = error; return (0); } int sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap) { struct timespec ts; struct timespec *timeout; sigset_t set; ksiginfo_t ksi; int error; if (uap->timeout) { error = copyin(uap->timeout, &ts, sizeof(ts)); if (error) return (error); timeout = &ts; } else timeout = NULL; error = copyin(uap->set, &set, sizeof(set)); if (error) return (error); error = kern_sigtimedwait(td, set, &ksi, timeout); if (error) return (error); if (uap->info) error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t)); if (error == 0) td->td_retval[0] = ksi.ksi_signo; return (error); } int sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap) { ksiginfo_t ksi; sigset_t set; int error; error = copyin(uap->set, &set, sizeof(set)); if (error) return (error); error = kern_sigtimedwait(td, set, &ksi, NULL); if (error) return (error); if (uap->info) error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t)); if (error == 0) td->td_retval[0] = ksi.ksi_signo; return (error); } int kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi, struct timespec *timeout) { struct sigacts *ps; sigset_t saved_mask, new_block; struct proc *p; int error, sig, timo, timevalid = 0; struct timespec rts, ets, ts; struct timeval tv; p = td->td_proc; error = 0; ets.tv_sec = 0; ets.tv_nsec = 0; if (timeout != NULL) { if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) { timevalid = 1; getnanouptime(&rts); ets = rts; timespecadd(&ets, timeout); } } ksiginfo_init(ksi); /* Some signals can not be waited for. */ SIG_CANTMASK(waitset); ps = p->p_sigacts; PROC_LOCK(p); saved_mask = td->td_sigmask; SIGSETNAND(td->td_sigmask, waitset); for (;;) { mtx_lock(&ps->ps_mtx); sig = cursig(td); mtx_unlock(&ps->ps_mtx); KASSERT(sig >= 0, ("sig %d", sig)); if (sig != 0 && SIGISMEMBER(waitset, sig)) { if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 || sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) { error = 0; break; } } if (error != 0) break; /* * POSIX says this must be checked after looking for pending * signals. */ if (timeout != NULL) { if (!timevalid) { error = EINVAL; break; } getnanouptime(&rts); if (timespeccmp(&rts, &ets, >=)) { error = EAGAIN; break; } ts = ets; timespecsub(&ts, &rts); TIMESPEC_TO_TIMEVAL(&tv, &ts); timo = tvtohz(&tv); } else { timo = 0; } error = msleep(ps, &p->p_mtx, PPAUSE|PCATCH, "sigwait", timo); if (timeout != NULL) { if (error == ERESTART) { /* Timeout can not be restarted. */ error = EINTR; } else if (error == EAGAIN) { /* We will calculate timeout by ourself. */ error = 0; } } } new_block = saved_mask; SIGSETNAND(new_block, td->td_sigmask); td->td_sigmask = saved_mask; /* * Fewer signals can be delivered to us, reschedule signal * notification. */ if (p->p_numthreads != 1) reschedule_signals(p, new_block, 0); if (error == 0) { SDT_PROBE2(proc, , , signal__clear, sig, ksi); if (ksi->ksi_code == SI_TIMER) itimer_accept(p, ksi->ksi_timerid, ksi); #ifdef KTRACE if (KTRPOINT(td, KTR_PSIG)) { sig_t action; mtx_lock(&ps->ps_mtx); action = ps->ps_sigact[_SIG_IDX(sig)]; mtx_unlock(&ps->ps_mtx); ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code); } #endif if (sig == SIGKILL) sigexit(td, sig); } PROC_UNLOCK(p); return (error); } #ifndef _SYS_SYSPROTO_H_ struct sigpending_args { sigset_t *set; }; #endif int sys_sigpending(td, uap) struct thread *td; struct sigpending_args *uap; { struct proc *p = td->td_proc; sigset_t pending; PROC_LOCK(p); pending = p->p_sigqueue.sq_signals; SIGSETOR(pending, td->td_sigqueue.sq_signals); PROC_UNLOCK(p); return (copyout(&pending, uap->set, sizeof(sigset_t))); } #ifdef COMPAT_43 /* XXX - COMPAT_FBSD3 */ #ifndef _SYS_SYSPROTO_H_ struct osigpending_args { int dummy; }; #endif int osigpending(td, uap) struct thread *td; struct osigpending_args *uap; { struct proc *p = td->td_proc; sigset_t pending; PROC_LOCK(p); pending = p->p_sigqueue.sq_signals; SIGSETOR(pending, td->td_sigqueue.sq_signals); PROC_UNLOCK(p); SIG2OSIG(pending, td->td_retval[0]); return (0); } #endif /* COMPAT_43 */ #if defined(COMPAT_43) /* * Generalized interface signal handler, 4.3-compatible. */ #ifndef _SYS_SYSPROTO_H_ struct osigvec_args { int signum; struct sigvec *nsv; struct sigvec *osv; }; #endif /* ARGSUSED */ int osigvec(td, uap) struct thread *td; register struct osigvec_args *uap; { struct sigvec vec; struct sigaction nsa, osa; register struct sigaction *nsap, *osap; int error; if (uap->signum <= 0 || uap->signum >= ONSIG) return (EINVAL); nsap = (uap->nsv != NULL) ? &nsa : NULL; osap = (uap->osv != NULL) ? &osa : NULL; if (nsap) { error = copyin(uap->nsv, &vec, sizeof(vec)); if (error) return (error); nsap->sa_handler = vec.sv_handler; OSIG2SIG(vec.sv_mask, nsap->sa_mask); nsap->sa_flags = vec.sv_flags; nsap->sa_flags ^= SA_RESTART; /* opposite of SV_INTERRUPT */ } error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET); if (osap && !error) { vec.sv_handler = osap->sa_handler; SIG2OSIG(osap->sa_mask, vec.sv_mask); vec.sv_flags = osap->sa_flags; vec.sv_flags &= ~SA_NOCLDWAIT; vec.sv_flags ^= SA_RESTART; error = copyout(&vec, uap->osv, sizeof(vec)); } return (error); } #ifndef _SYS_SYSPROTO_H_ struct osigblock_args { int mask; }; #endif int osigblock(td, uap) register struct thread *td; struct osigblock_args *uap; { sigset_t set, oset; OSIG2SIG(uap->mask, set); kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0); SIG2OSIG(oset, td->td_retval[0]); return (0); } #ifndef _SYS_SYSPROTO_H_ struct osigsetmask_args { int mask; }; #endif int osigsetmask(td, uap) struct thread *td; struct osigsetmask_args *uap; { sigset_t set, oset; OSIG2SIG(uap->mask, set); kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0); SIG2OSIG(oset, td->td_retval[0]); return (0); } #endif /* COMPAT_43 */ /* * Suspend calling thread until signal, providing mask to be set in the * meantime. */ #ifndef _SYS_SYSPROTO_H_ struct sigsuspend_args { const sigset_t *sigmask; }; #endif /* ARGSUSED */ int sys_sigsuspend(td, uap) struct thread *td; struct sigsuspend_args *uap; { sigset_t mask; int error; error = copyin(uap->sigmask, &mask, sizeof(mask)); if (error) return (error); return (kern_sigsuspend(td, mask)); } int kern_sigsuspend(struct thread *td, sigset_t mask) { struct proc *p = td->td_proc; int has_sig, sig; /* * When returning from sigsuspend, we want * the old mask to be restored after the * signal handler has finished. Thus, we * save it here and mark the sigacts structure * to indicate this. */ PROC_LOCK(p); kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask, SIGPROCMASK_PROC_LOCKED); td->td_pflags |= TDP_OLDMASK; /* * Process signals now. Otherwise, we can get spurious wakeup * due to signal entered process queue, but delivered to other * thread. But sigsuspend should return only on signal * delivery. */ (p->p_sysent->sv_set_syscall_retval)(td, EINTR); for (has_sig = 0; !has_sig;) { while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause", 0) == 0) /* void */; thread_suspend_check(0); mtx_lock(&p->p_sigacts->ps_mtx); while ((sig = cursig(td)) != 0) { KASSERT(sig >= 0, ("sig %d", sig)); has_sig += postsig(sig); } mtx_unlock(&p->p_sigacts->ps_mtx); } PROC_UNLOCK(p); td->td_errno = EINTR; td->td_pflags |= TDP_NERRNO; return (EJUSTRETURN); } #ifdef COMPAT_43 /* XXX - COMPAT_FBSD3 */ /* * Compatibility sigsuspend call for old binaries. Note nonstandard calling * convention: libc stub passes mask, not pointer, to save a copyin. */ #ifndef _SYS_SYSPROTO_H_ struct osigsuspend_args { osigset_t mask; }; #endif /* ARGSUSED */ int osigsuspend(td, uap) struct thread *td; struct osigsuspend_args *uap; { sigset_t mask; OSIG2SIG(uap->mask, mask); return (kern_sigsuspend(td, mask)); } #endif /* COMPAT_43 */ #if defined(COMPAT_43) #ifndef _SYS_SYSPROTO_H_ struct osigstack_args { struct sigstack *nss; struct sigstack *oss; }; #endif /* ARGSUSED */ int osigstack(td, uap) struct thread *td; register struct osigstack_args *uap; { struct sigstack nss, oss; int error = 0; if (uap->nss != NULL) { error = copyin(uap->nss, &nss, sizeof(nss)); if (error) return (error); } oss.ss_sp = td->td_sigstk.ss_sp; oss.ss_onstack = sigonstack(cpu_getstack(td)); if (uap->nss != NULL) { td->td_sigstk.ss_sp = nss.ss_sp; td->td_sigstk.ss_size = 0; td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK; td->td_pflags |= TDP_ALTSTACK; } if (uap->oss != NULL) error = copyout(&oss, uap->oss, sizeof(oss)); return (error); } #endif /* COMPAT_43 */ #ifndef _SYS_SYSPROTO_H_ struct sigaltstack_args { stack_t *ss; stack_t *oss; }; #endif /* ARGSUSED */ int sys_sigaltstack(td, uap) struct thread *td; register struct sigaltstack_args *uap; { stack_t ss, oss; int error; if (uap->ss != NULL) { error = copyin(uap->ss, &ss, sizeof(ss)); if (error) return (error); } error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL, (uap->oss != NULL) ? &oss : NULL); if (error) return (error); if (uap->oss != NULL) error = copyout(&oss, uap->oss, sizeof(stack_t)); return (error); } int kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss) { struct proc *p = td->td_proc; int oonstack; oonstack = sigonstack(cpu_getstack(td)); if (oss != NULL) { *oss = td->td_sigstk; oss->ss_flags = (td->td_pflags & TDP_ALTSTACK) ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE; } if (ss != NULL) { if (oonstack) return (EPERM); if ((ss->ss_flags & ~SS_DISABLE) != 0) return (EINVAL); if (!(ss->ss_flags & SS_DISABLE)) { if (ss->ss_size < p->p_sysent->sv_minsigstksz) return (ENOMEM); td->td_sigstk = *ss; td->td_pflags |= TDP_ALTSTACK; } else { td->td_pflags &= ~TDP_ALTSTACK; } } return (0); } /* * Common code for kill process group/broadcast kill. * cp is calling process. */ static int killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi) { struct proc *p; struct pgrp *pgrp; int err; int ret; ret = ESRCH; if (all) { /* * broadcast */ sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); if (p->p_pid <= 1 || p->p_flag & P_SYSTEM || p == td->td_proc || p->p_state == PRS_NEW) { PROC_UNLOCK(p); continue; } err = p_cansignal(td, p, sig); if (err == 0) { if (sig) pksignal(p, sig, ksi); ret = err; } else if (ret == ESRCH) ret = err; PROC_UNLOCK(p); } sx_sunlock(&allproc_lock); } else { sx_slock(&proctree_lock); if (pgid == 0) { /* * zero pgid means send to my process group. */ pgrp = td->td_proc->p_pgrp; PGRP_LOCK(pgrp); } else { pgrp = pgfind(pgid); if (pgrp == NULL) { sx_sunlock(&proctree_lock); return (ESRCH); } } sx_sunlock(&proctree_lock); LIST_FOREACH(p, &pgrp->pg_members, p_pglist) { PROC_LOCK(p); if (p->p_pid <= 1 || p->p_flag & P_SYSTEM || p->p_state == PRS_NEW) { PROC_UNLOCK(p); continue; } err = p_cansignal(td, p, sig); if (err == 0) { if (sig) pksignal(p, sig, ksi); ret = err; } else if (ret == ESRCH) ret = err; PROC_UNLOCK(p); } PGRP_UNLOCK(pgrp); } return (ret); } #ifndef _SYS_SYSPROTO_H_ struct kill_args { int pid; int signum; }; #endif /* ARGSUSED */ int sys_kill(struct thread *td, struct kill_args *uap) { ksiginfo_t ksi; struct proc *p; int error; /* * A process in capability mode can send signals only to himself. * The main rationale behind this is that abort(3) is implemented as * kill(getpid(), SIGABRT). */ if (IN_CAPABILITY_MODE(td) && uap->pid != td->td_proc->p_pid) return (ECAPMODE); AUDIT_ARG_SIGNUM(uap->signum); AUDIT_ARG_PID(uap->pid); if ((u_int)uap->signum > _SIG_MAXSIG) return (EINVAL); ksiginfo_init(&ksi); ksi.ksi_signo = uap->signum; ksi.ksi_code = SI_USER; ksi.ksi_pid = td->td_proc->p_pid; ksi.ksi_uid = td->td_ucred->cr_ruid; if (uap->pid > 0) { /* kill single process */ if ((p = pfind(uap->pid)) == NULL) { if ((p = zpfind(uap->pid)) == NULL) return (ESRCH); } AUDIT_ARG_PROCESS(p); error = p_cansignal(td, p, uap->signum); if (error == 0 && uap->signum) pksignal(p, uap->signum, &ksi); PROC_UNLOCK(p); return (error); } switch (uap->pid) { case -1: /* broadcast signal */ return (killpg1(td, uap->signum, 0, 1, &ksi)); case 0: /* signal own process group */ return (killpg1(td, uap->signum, 0, 0, &ksi)); default: /* negative explicit process group */ return (killpg1(td, uap->signum, -uap->pid, 0, &ksi)); } /* NOTREACHED */ } int sys_pdkill(td, uap) struct thread *td; struct pdkill_args *uap; { struct proc *p; cap_rights_t rights; int error; AUDIT_ARG_SIGNUM(uap->signum); AUDIT_ARG_FD(uap->fd); if ((u_int)uap->signum > _SIG_MAXSIG) return (EINVAL); error = procdesc_find(td, uap->fd, cap_rights_init(&rights, CAP_PDKILL), &p); if (error) return (error); AUDIT_ARG_PROCESS(p); error = p_cansignal(td, p, uap->signum); if (error == 0 && uap->signum) kern_psignal(p, uap->signum); PROC_UNLOCK(p); return (error); } #if defined(COMPAT_43) #ifndef _SYS_SYSPROTO_H_ struct okillpg_args { int pgid; int signum; }; #endif /* ARGSUSED */ int okillpg(struct thread *td, struct okillpg_args *uap) { ksiginfo_t ksi; AUDIT_ARG_SIGNUM(uap->signum); AUDIT_ARG_PID(uap->pgid); if ((u_int)uap->signum > _SIG_MAXSIG) return (EINVAL); ksiginfo_init(&ksi); ksi.ksi_signo = uap->signum; ksi.ksi_code = SI_USER; ksi.ksi_pid = td->td_proc->p_pid; ksi.ksi_uid = td->td_ucred->cr_ruid; return (killpg1(td, uap->signum, uap->pgid, 0, &ksi)); } #endif /* COMPAT_43 */ #ifndef _SYS_SYSPROTO_H_ struct sigqueue_args { pid_t pid; int signum; /* union sigval */ void *value; }; #endif int sys_sigqueue(struct thread *td, struct sigqueue_args *uap) { ksiginfo_t ksi; struct proc *p; int error; if ((u_int)uap->signum > _SIG_MAXSIG) return (EINVAL); /* * Specification says sigqueue can only send signal to * single process. */ if (uap->pid <= 0) return (EINVAL); if ((p = pfind(uap->pid)) == NULL) { if ((p = zpfind(uap->pid)) == NULL) return (ESRCH); } error = p_cansignal(td, p, uap->signum); if (error == 0 && uap->signum != 0) { ksiginfo_init(&ksi); ksi.ksi_flags = KSI_SIGQ; ksi.ksi_signo = uap->signum; ksi.ksi_code = SI_QUEUE; ksi.ksi_pid = td->td_proc->p_pid; ksi.ksi_uid = td->td_ucred->cr_ruid; ksi.ksi_value.sival_ptr = uap->value; error = pksignal(p, ksi.ksi_signo, &ksi); } PROC_UNLOCK(p); return (error); } /* * Send a signal to a process group. */ void gsignal(int pgid, int sig, ksiginfo_t *ksi) { struct pgrp *pgrp; if (pgid != 0) { sx_slock(&proctree_lock); pgrp = pgfind(pgid); sx_sunlock(&proctree_lock); if (pgrp != NULL) { pgsignal(pgrp, sig, 0, ksi); PGRP_UNLOCK(pgrp); } } } /* * Send a signal to a process group. If checktty is 1, * limit to members which have a controlling terminal. */ void pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi) { struct proc *p; if (pgrp) { PGRP_LOCK_ASSERT(pgrp, MA_OWNED); LIST_FOREACH(p, &pgrp->pg_members, p_pglist) { PROC_LOCK(p); if (p->p_state == PRS_NORMAL && (checkctty == 0 || p->p_flag & P_CONTROLT)) pksignal(p, sig, ksi); PROC_UNLOCK(p); } } } /* * Recalculate the signal mask and reset the signal disposition after * usermode frame for delivery is formed. Should be called after * mach-specific routine, because sysent->sv_sendsig() needs correct * ps_siginfo and signal mask. */ static void postsig_done(int sig, struct thread *td, struct sigacts *ps) { sigset_t mask; mtx_assert(&ps->ps_mtx, MA_OWNED); td->td_ru.ru_nsignals++; mask = ps->ps_catchmask[_SIG_IDX(sig)]; if (!SIGISMEMBER(ps->ps_signodefer, sig)) SIGADDSET(mask, sig); kern_sigprocmask(td, SIG_BLOCK, &mask, NULL, SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED); if (SIGISMEMBER(ps->ps_sigreset, sig)) sigdflt(ps, sig); } /* * Send a signal caused by a trap to the current thread. If it will be * caught immediately, deliver it with correct code. Otherwise, post it * normally. */ void trapsignal(struct thread *td, ksiginfo_t *ksi) { struct sigacts *ps; struct proc *p; int sig; int code; p = td->td_proc; sig = ksi->ksi_signo; code = ksi->ksi_code; KASSERT(_SIG_VALID(sig), ("invalid signal")); PROC_LOCK(p); ps = p->p_sigacts; mtx_lock(&ps->ps_mtx); if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) && !SIGISMEMBER(td->td_sigmask, sig)) { #ifdef KTRACE if (KTRPOINT(curthread, KTR_PSIG)) ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)], &td->td_sigmask, code); #endif (*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)], ksi, &td->td_sigmask); postsig_done(sig, td, ps); mtx_unlock(&ps->ps_mtx); } else { /* * Avoid a possible infinite loop if the thread * masking the signal or process is ignoring the * signal. */ if (kern_forcesigexit && (SIGISMEMBER(td->td_sigmask, sig) || ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) { SIGDELSET(td->td_sigmask, sig); SIGDELSET(ps->ps_sigcatch, sig); SIGDELSET(ps->ps_sigignore, sig); ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL; } mtx_unlock(&ps->ps_mtx); p->p_code = code; /* XXX for core dump/debugger */ p->p_sig = sig; /* XXX to verify code */ tdsendsignal(p, td, sig, ksi); } PROC_UNLOCK(p); } static struct thread * sigtd(struct proc *p, int sig, int prop) { struct thread *td, *signal_td; PROC_LOCK_ASSERT(p, MA_OWNED); /* * Check if current thread can handle the signal without * switching context to another thread. */ if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig)) return (curthread); signal_td = NULL; FOREACH_THREAD_IN_PROC(p, td) { if (!SIGISMEMBER(td->td_sigmask, sig)) { signal_td = td; break; } } if (signal_td == NULL) signal_td = FIRST_THREAD_IN_PROC(p); return (signal_td); } /* * Send the signal to the process. If the signal has an action, the action * is usually performed by the target process rather than the caller; we add * the signal to the set of pending signals for the process. * * Exceptions: * o When a stop signal is sent to a sleeping process that takes the * default action, the process is stopped without awakening it. * o SIGCONT restarts stopped processes (or puts them back to sleep) * regardless of the signal action (eg, blocked or ignored). * * Other ignored signals are discarded immediately. * * NB: This function may be entered from the debugger via the "kill" DDB * command. There is little that can be done to mitigate the possibly messy * side effects of this unwise possibility. */ void kern_psignal(struct proc *p, int sig) { ksiginfo_t ksi; ksiginfo_init(&ksi); ksi.ksi_signo = sig; ksi.ksi_code = SI_KERNEL; (void) tdsendsignal(p, NULL, sig, &ksi); } int pksignal(struct proc *p, int sig, ksiginfo_t *ksi) { return (tdsendsignal(p, NULL, sig, ksi)); } /* Utility function for finding a thread to send signal event to. */ int sigev_findtd(struct proc *p ,struct sigevent *sigev, struct thread **ttd) { struct thread *td; if (sigev->sigev_notify == SIGEV_THREAD_ID) { td = tdfind(sigev->sigev_notify_thread_id, p->p_pid); if (td == NULL) return (ESRCH); *ttd = td; } else { *ttd = NULL; PROC_LOCK(p); } return (0); } void tdsignal(struct thread *td, int sig) { ksiginfo_t ksi; ksiginfo_init(&ksi); ksi.ksi_signo = sig; ksi.ksi_code = SI_KERNEL; (void) tdsendsignal(td->td_proc, td, sig, &ksi); } void tdksignal(struct thread *td, int sig, ksiginfo_t *ksi) { (void) tdsendsignal(td->td_proc, td, sig, ksi); } int tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi) { sig_t action; sigqueue_t *sigqueue; int prop; struct sigacts *ps; int intrval; int ret = 0; int wakeup_swapper; MPASS(td == NULL || p == td->td_proc); PROC_LOCK_ASSERT(p, MA_OWNED); if (!_SIG_VALID(sig)) panic("%s(): invalid signal %d", __func__, sig); KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__)); /* * IEEE Std 1003.1-2001: return success when killing a zombie. */ if (p->p_state == PRS_ZOMBIE) { if (ksi && (ksi->ksi_flags & KSI_INS)) ksiginfo_tryfree(ksi); return (ret); } ps = p->p_sigacts; KNOTE_LOCKED(p->p_klist, NOTE_SIGNAL | sig); prop = sigprop(sig); if (td == NULL) { td = sigtd(p, sig, prop); sigqueue = &p->p_sigqueue; } else sigqueue = &td->td_sigqueue; SDT_PROBE3(proc, , , signal__send, td, p, sig); /* * If the signal is being ignored, * then we forget about it immediately. * (Note: we don't set SIGCONT in ps_sigignore, * and if it is set to SIG_IGN, * action will be SIG_DFL here.) */ mtx_lock(&ps->ps_mtx); if (SIGISMEMBER(ps->ps_sigignore, sig)) { SDT_PROBE3(proc, , , signal__discard, td, p, sig); mtx_unlock(&ps->ps_mtx); if (ksi && (ksi->ksi_flags & KSI_INS)) ksiginfo_tryfree(ksi); return (ret); } if (SIGISMEMBER(td->td_sigmask, sig)) action = SIG_HOLD; else if (SIGISMEMBER(ps->ps_sigcatch, sig)) action = SIG_CATCH; else action = SIG_DFL; if (SIGISMEMBER(ps->ps_sigintr, sig)) intrval = EINTR; else intrval = ERESTART; mtx_unlock(&ps->ps_mtx); if (prop & SIGPROP_CONT) sigqueue_delete_stopmask_proc(p); else if (prop & SIGPROP_STOP) { /* * If sending a tty stop signal to a member of an orphaned * process group, discard the signal here if the action * is default; don't stop the process below if sleeping, * and don't clear any pending SIGCONT. */ if ((prop & SIGPROP_TTYSTOP) && (p->p_pgrp->pg_jobc == 0) && (action == SIG_DFL)) { if (ksi && (ksi->ksi_flags & KSI_INS)) ksiginfo_tryfree(ksi); return (ret); } sigqueue_delete_proc(p, SIGCONT); if (p->p_flag & P_CONTINUED) { p->p_flag &= ~P_CONTINUED; PROC_LOCK(p->p_pptr); sigqueue_take(p->p_ksi); PROC_UNLOCK(p->p_pptr); } } ret = sigqueue_add(sigqueue, sig, ksi); if (ret != 0) return (ret); signotify(td); /* * Defer further processing for signals which are held, * except that stopped processes must be continued by SIGCONT. */ if (action == SIG_HOLD && !((prop & SIGPROP_CONT) && (p->p_flag & P_STOPPED_SIG))) return (ret); /* * SIGKILL: Remove procfs STOPEVENTs and ptrace events. */ if (sig == SIGKILL) { p->p_ptevents = 0; /* from procfs_ioctl.c: PIOCBIC */ p->p_stops = 0; /* from procfs_ioctl.c: PIOCCONT */ p->p_step = 0; wakeup(&p->p_step); } /* * Some signals have a process-wide effect and a per-thread * component. Most processing occurs when the process next * tries to cross the user boundary, however there are some * times when processing needs to be done immediately, such as * waking up threads so that they can cross the user boundary. * We try to do the per-process part here. */ if (P_SHOULDSTOP(p)) { KASSERT(!(p->p_flag & P_WEXIT), ("signal to stopped but exiting process")); if (sig == SIGKILL) { /* * If traced process is already stopped, * then no further action is necessary. */ if (p->p_flag & P_TRACED) goto out; /* * SIGKILL sets process running. * It will die elsewhere. * All threads must be restarted. */ p->p_flag &= ~P_STOPPED_SIG; goto runfast; } if (prop & SIGPROP_CONT) { /* * If traced process is already stopped, * then no further action is necessary. */ if (p->p_flag & P_TRACED) goto out; /* * If SIGCONT is default (or ignored), we continue the * process but don't leave the signal in sigqueue as * it has no further action. If SIGCONT is held, we * continue the process and leave the signal in * sigqueue. If the process catches SIGCONT, let it * handle the signal itself. If it isn't waiting on * an event, it goes back to run state. * Otherwise, process goes back to sleep state. */ p->p_flag &= ~P_STOPPED_SIG; PROC_SLOCK(p); if (p->p_numthreads == p->p_suspcount) { PROC_SUNLOCK(p); p->p_flag |= P_CONTINUED; p->p_xsig = SIGCONT; PROC_LOCK(p->p_pptr); childproc_continued(p); PROC_UNLOCK(p->p_pptr); PROC_SLOCK(p); } if (action == SIG_DFL) { thread_unsuspend(p); PROC_SUNLOCK(p); sigqueue_delete(sigqueue, sig); goto out; } if (action == SIG_CATCH) { /* * The process wants to catch it so it needs * to run at least one thread, but which one? */ PROC_SUNLOCK(p); goto runfast; } /* * The signal is not ignored or caught. */ thread_unsuspend(p); PROC_SUNLOCK(p); goto out; } if (prop & SIGPROP_STOP) { /* * If traced process is already stopped, * then no further action is necessary. */ if (p->p_flag & P_TRACED) goto out; /* * Already stopped, don't need to stop again * (If we did the shell could get confused). * Just make sure the signal STOP bit set. */ p->p_flag |= P_STOPPED_SIG; sigqueue_delete(sigqueue, sig); goto out; } /* * All other kinds of signals: * If a thread is sleeping interruptibly, simulate a * wakeup so that when it is continued it will be made * runnable and can look at the signal. However, don't make * the PROCESS runnable, leave it stopped. * It may run a bit until it hits a thread_suspend_check(). */ wakeup_swapper = 0; PROC_SLOCK(p); thread_lock(td); if (TD_ON_SLEEPQ(td) && (td->td_flags & TDF_SINTR)) wakeup_swapper = sleepq_abort(td, intrval); thread_unlock(td); PROC_SUNLOCK(p); if (wakeup_swapper) kick_proc0(); goto out; /* * Mutexes are short lived. Threads waiting on them will * hit thread_suspend_check() soon. */ } else if (p->p_state == PRS_NORMAL) { if (p->p_flag & P_TRACED || action == SIG_CATCH) { tdsigwakeup(td, sig, action, intrval); goto out; } MPASS(action == SIG_DFL); if (prop & SIGPROP_STOP) { if (p->p_flag & (P_PPWAIT|P_WEXIT)) goto out; p->p_flag |= P_STOPPED_SIG; p->p_xsig = sig; PROC_SLOCK(p); wakeup_swapper = sig_suspend_threads(td, p, 1); if (p->p_numthreads == p->p_suspcount) { /* * only thread sending signal to another * process can reach here, if thread is sending * signal to its process, because thread does * not suspend itself here, p_numthreads * should never be equal to p_suspcount. */ thread_stopped(p); PROC_SUNLOCK(p); sigqueue_delete_proc(p, p->p_xsig); } else PROC_SUNLOCK(p); if (wakeup_swapper) kick_proc0(); goto out; } } else { /* Not in "NORMAL" state. discard the signal. */ sigqueue_delete(sigqueue, sig); goto out; } /* * The process is not stopped so we need to apply the signal to all the * running threads. */ runfast: tdsigwakeup(td, sig, action, intrval); PROC_SLOCK(p); thread_unsuspend(p); PROC_SUNLOCK(p); out: /* If we jump here, proc slock should not be owned. */ PROC_SLOCK_ASSERT(p, MA_NOTOWNED); return (ret); } /* * The force of a signal has been directed against a single * thread. We need to see what we can do about knocking it * out of any sleep it may be in etc. */ static void tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval) { struct proc *p = td->td_proc; register int prop; int wakeup_swapper; wakeup_swapper = 0; PROC_LOCK_ASSERT(p, MA_OWNED); prop = sigprop(sig); PROC_SLOCK(p); thread_lock(td); /* * Bring the priority of a thread up if we want it to get * killed in this lifetime. Be careful to avoid bumping the * priority of the idle thread, since we still allow to signal * kernel processes. */ if (action == SIG_DFL && (prop & SIGPROP_KILL) != 0 && td->td_priority > PUSER && !TD_IS_IDLETHREAD(td)) sched_prio(td, PUSER); if (TD_ON_SLEEPQ(td)) { /* * If thread is sleeping uninterruptibly * we can't interrupt the sleep... the signal will * be noticed when the process returns through * trap() or syscall(). */ if ((td->td_flags & TDF_SINTR) == 0) goto out; /* * If SIGCONT is default (or ignored) and process is * asleep, we are finished; the process should not * be awakened. */ if ((prop & SIGPROP_CONT) && action == SIG_DFL) { thread_unlock(td); PROC_SUNLOCK(p); sigqueue_delete(&p->p_sigqueue, sig); /* * It may be on either list in this state. * Remove from both for now. */ sigqueue_delete(&td->td_sigqueue, sig); return; } /* * Don't awaken a sleeping thread for SIGSTOP if the * STOP signal is deferred. */ if ((prop & SIGPROP_STOP) != 0 && (td->td_flags & (TDF_SBDRY | TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY) goto out; /* * Give low priority threads a better chance to run. */ if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td)) sched_prio(td, PUSER); wakeup_swapper = sleepq_abort(td, intrval); } else { /* * Other states do nothing with the signal immediately, * other than kicking ourselves if we are running. * It will either never be noticed, or noticed very soon. */ #ifdef SMP if (TD_IS_RUNNING(td) && td != curthread) forward_signal(td); #endif } out: PROC_SUNLOCK(p); thread_unlock(td); if (wakeup_swapper) kick_proc0(); } static int sig_suspend_threads(struct thread *td, struct proc *p, int sending) { struct thread *td2; int wakeup_swapper; PROC_LOCK_ASSERT(p, MA_OWNED); PROC_SLOCK_ASSERT(p, MA_OWNED); wakeup_swapper = 0; FOREACH_THREAD_IN_PROC(p, td2) { thread_lock(td2); td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK; if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) && (td2->td_flags & TDF_SINTR)) { if (td2->td_flags & TDF_SBDRY) { /* * Once a thread is asleep with * TDF_SBDRY and without TDF_SERESTART * or TDF_SEINTR set, it should never * become suspended due to this check. */ KASSERT(!TD_IS_SUSPENDED(td2), ("thread with deferred stops suspended")); if (TD_SBDRY_INTR(td2) && sending) { wakeup_swapper |= sleepq_abort(td2, TD_SBDRY_ERRNO(td2)); } } else if (!TD_IS_SUSPENDED(td2)) { thread_suspend_one(td2); } } else if (!TD_IS_SUSPENDED(td2)) { if (sending || td != td2) td2->td_flags |= TDF_ASTPENDING; #ifdef SMP if (TD_IS_RUNNING(td2) && td2 != td) forward_signal(td2); #endif } thread_unlock(td2); } return (wakeup_swapper); } int ptracestop(struct thread *td, int sig) { struct proc *p = td->td_proc; PROC_LOCK_ASSERT(p, MA_OWNED); KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process")); WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, &p->p_mtx.lock_object, "Stopping for traced signal"); td->td_dbgflags |= TDB_XSIG; td->td_xsig = sig; CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d", td->td_tid, p->p_pid, td->td_dbgflags, sig); PROC_SLOCK(p); while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) { if (p->p_flag & P_SINGLE_EXIT && !(td->td_dbgflags & TDB_EXIT)) { /* * Ignore ptrace stops except for thread exit * events when the process exits. */ td->td_dbgflags &= ~TDB_XSIG; PROC_SUNLOCK(p); return (sig); } /* * Make wait(2) work. Ensure that right after the * attach, the thread which was decided to become the * leader of attach gets reported to the waiter. * Otherwise, just avoid overwriting another thread's * assignment to p_xthread. If another thread has * already set p_xthread, the current thread will get * a chance to report itself upon the next iteration. */ if ((td->td_dbgflags & TDB_FSTP) != 0 || ((p->p_flag2 & P2_PTRACE_FSTP) == 0 && p->p_xthread == NULL)) { p->p_xsig = sig; p->p_xthread = td; td->td_dbgflags &= ~TDB_FSTP; p->p_flag2 &= ~P2_PTRACE_FSTP; p->p_flag |= P_STOPPED_SIG | P_STOPPED_TRACE; sig_suspend_threads(td, p, 0); } if ((td->td_dbgflags & TDB_STOPATFORK) != 0) { td->td_dbgflags &= ~TDB_STOPATFORK; cv_broadcast(&p->p_dbgwait); } stopme: thread_suspend_switch(td, p); if (p->p_xthread == td) p->p_xthread = NULL; if (!(p->p_flag & P_TRACED)) break; if (td->td_dbgflags & TDB_SUSPEND) { if (p->p_flag & P_SINGLE_EXIT) break; goto stopme; } } PROC_SUNLOCK(p); return (td->td_xsig); } static void reschedule_signals(struct proc *p, sigset_t block, int flags) { struct sigacts *ps; struct thread *td; int sig; PROC_LOCK_ASSERT(p, MA_OWNED); ps = p->p_sigacts; mtx_assert(&ps->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0 ? MA_OWNED : MA_NOTOWNED); if (SIGISEMPTY(p->p_siglist)) return; SIGSETAND(block, p->p_siglist); while ((sig = sig_ffs(&block)) != 0) { SIGDELSET(block, sig); td = sigtd(p, sig, 0); signotify(td); if (!(flags & SIGPROCMASK_PS_LOCKED)) mtx_lock(&ps->ps_mtx); if (p->p_flag & P_TRACED || SIGISMEMBER(ps->ps_sigcatch, sig)) tdsigwakeup(td, sig, SIG_CATCH, (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR : ERESTART)); if (!(flags & SIGPROCMASK_PS_LOCKED)) mtx_unlock(&ps->ps_mtx); } } void tdsigcleanup(struct thread *td) { struct proc *p; sigset_t unblocked; p = td->td_proc; PROC_LOCK_ASSERT(p, MA_OWNED); sigqueue_flush(&td->td_sigqueue); if (p->p_numthreads == 1) return; /* * Since we cannot handle signals, notify signal post code * about this by filling the sigmask. * * Also, if needed, wake up thread(s) that do not block the * same signals as the exiting thread, since the thread might * have been selected for delivery and woken up. */ SIGFILLSET(unblocked); SIGSETNAND(unblocked, td->td_sigmask); SIGFILLSET(td->td_sigmask); reschedule_signals(p, unblocked, 0); } static int sigdeferstop_curr_flags(int cflags) { MPASS((cflags & (TDF_SEINTR | TDF_SERESTART)) == 0 || (cflags & TDF_SBDRY) != 0); return (cflags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)); } /* * Defer the delivery of SIGSTOP for the current thread, according to * the requested mode. Returns previous flags, which must be restored * by sigallowstop(). * * TDF_SBDRY, TDF_SEINTR, and TDF_SERESTART flags are only set and * cleared by the current thread, which allow the lock-less read-only * accesses below. */ int sigdeferstop_impl(int mode) { struct thread *td; int cflags, nflags; td = curthread; cflags = sigdeferstop_curr_flags(td->td_flags); switch (mode) { case SIGDEFERSTOP_NOP: nflags = cflags; break; case SIGDEFERSTOP_OFF: nflags = 0; break; case SIGDEFERSTOP_SILENT: nflags = (cflags | TDF_SBDRY) & ~(TDF_SEINTR | TDF_SERESTART); break; case SIGDEFERSTOP_EINTR: nflags = (cflags | TDF_SBDRY | TDF_SEINTR) & ~TDF_SERESTART; break; case SIGDEFERSTOP_ERESTART: nflags = (cflags | TDF_SBDRY | TDF_SERESTART) & ~TDF_SEINTR; break; default: panic("sigdeferstop: invalid mode %x", mode); break; } if (cflags == nflags) return (SIGDEFERSTOP_VAL_NCHG); thread_lock(td); td->td_flags = (td->td_flags & ~cflags) | nflags; thread_unlock(td); return (cflags); } /* * Restores the STOP handling mode, typically permitting the delivery * of SIGSTOP for the current thread. This does not immediately * suspend if a stop was posted. Instead, the thread will suspend * either via ast() or a subsequent interruptible sleep. */ void sigallowstop_impl(int prev) { struct thread *td; int cflags; KASSERT(prev != SIGDEFERSTOP_VAL_NCHG, ("failed sigallowstop")); KASSERT((prev & ~(TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0, ("sigallowstop: incorrect previous mode %x", prev)); td = curthread; cflags = sigdeferstop_curr_flags(td->td_flags); if (cflags != prev) { thread_lock(td); td->td_flags = (td->td_flags & ~cflags) | prev; thread_unlock(td); } } /* * If the current process has received a signal (should be caught or cause * termination, should interrupt current syscall), return the signal number. * Stop signals with default action are processed immediately, then cleared; * they aren't returned. This is checked after each entry to the system for * a syscall or trap (though this can usually be done without calling issignal * by checking the pending signal masks in cursig.) The normal call * sequence is * * while (sig = cursig(curthread)) * postsig(sig); */ static int issignal(struct thread *td) { struct proc *p; struct sigacts *ps; struct sigqueue *queue; sigset_t sigpending; int sig, prop, newsig; p = td->td_proc; ps = p->p_sigacts; mtx_assert(&ps->ps_mtx, MA_OWNED); PROC_LOCK_ASSERT(p, MA_OWNED); for (;;) { int traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG); sigpending = td->td_sigqueue.sq_signals; SIGSETOR(sigpending, p->p_sigqueue.sq_signals); SIGSETNAND(sigpending, td->td_sigmask); if ((p->p_flag & P_PPWAIT) != 0 || (td->td_flags & (TDF_SBDRY | TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY) SIG_STOPSIGMASK(sigpending); if (SIGISEMPTY(sigpending)) /* no signal to send */ return (0); if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED && (p->p_flag2 & P2_PTRACE_FSTP) != 0 && SIGISMEMBER(sigpending, SIGSTOP)) { /* * If debugger just attached, always consume * SIGSTOP from ptrace(PT_ATTACH) first, to * execute the debugger attach ritual in * order. */ sig = SIGSTOP; td->td_dbgflags |= TDB_FSTP; } else { sig = sig_ffs(&sigpending); } if (p->p_stops & S_SIG) { mtx_unlock(&ps->ps_mtx); stopevent(p, S_SIG, sig); mtx_lock(&ps->ps_mtx); } /* * We should see pending but ignored signals * only if P_TRACED was on when they were posted. */ if (SIGISMEMBER(ps->ps_sigignore, sig) && (traced == 0)) { sigqueue_delete(&td->td_sigqueue, sig); sigqueue_delete(&p->p_sigqueue, sig); continue; } if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED) { /* * If traced, always stop. * Remove old signal from queue before the stop. * XXX shrug off debugger, it causes siginfo to * be thrown away. */ queue = &td->td_sigqueue; td->td_dbgksi.ksi_signo = 0; if (sigqueue_get(queue, sig, &td->td_dbgksi) == 0) { queue = &p->p_sigqueue; sigqueue_get(queue, sig, &td->td_dbgksi); } mtx_unlock(&ps->ps_mtx); newsig = ptracestop(td, sig); mtx_lock(&ps->ps_mtx); if (sig != newsig) { /* * If parent wants us to take the signal, * then it will leave it in p->p_xsig; * otherwise we just look for signals again. */ if (newsig == 0) continue; sig = newsig; /* * Put the new signal into td_sigqueue. If the * signal is being masked, look for other * signals. */ sigqueue_add(queue, sig, NULL); if (SIGISMEMBER(td->td_sigmask, sig)) continue; signotify(td); } else { if (td->td_dbgksi.ksi_signo != 0) { td->td_dbgksi.ksi_flags |= KSI_HEAD; if (sigqueue_add(&td->td_sigqueue, sig, &td->td_dbgksi) != 0) td->td_dbgksi.ksi_signo = 0; } if (td->td_dbgksi.ksi_signo == 0) sigqueue_add(&td->td_sigqueue, sig, NULL); } /* * If the traced bit got turned off, go back up * to the top to rescan signals. This ensures * that p_sig* and p_sigact are consistent. */ if ((p->p_flag & P_TRACED) == 0) continue; } prop = sigprop(sig); /* * Decide whether the signal should be returned. * Return the signal's number, or fall through * to clear it from the pending mask. */ switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) { case (intptr_t)SIG_DFL: /* * Don't take default actions on system processes. */ if (p->p_pid <= 1) { #ifdef DIAGNOSTIC /* * Are you sure you want to ignore SIGSEGV * in init? XXX */ printf("Process (pid %lu) got signal %d\n", (u_long)p->p_pid, sig); #endif break; /* == ignore */ } /* * If there is a pending stop signal to process * with default action, stop here, * then clear the signal. However, * if process is member of an orphaned * process group, ignore tty stop signals. */ if (prop & SIGPROP_STOP) { if (p->p_flag & (P_TRACED|P_WEXIT) || (p->p_pgrp->pg_jobc == 0 && prop & SIGPROP_TTYSTOP)) break; /* == ignore */ if (TD_SBDRY_INTR(td)) { KASSERT((td->td_flags & TDF_SBDRY) != 0, ("lost TDF_SBDRY")); return (-1); } mtx_unlock(&ps->ps_mtx); WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, &p->p_mtx.lock_object, "Catching SIGSTOP"); sigqueue_delete(&td->td_sigqueue, sig); sigqueue_delete(&p->p_sigqueue, sig); p->p_flag |= P_STOPPED_SIG; p->p_xsig = sig; PROC_SLOCK(p); sig_suspend_threads(td, p, 0); thread_suspend_switch(td, p); PROC_SUNLOCK(p); mtx_lock(&ps->ps_mtx); goto next; } else if (prop & SIGPROP_IGNORE) { /* * Except for SIGCONT, shouldn't get here. * Default action is to ignore; drop it. */ break; /* == ignore */ } else return (sig); /*NOTREACHED*/ case (intptr_t)SIG_IGN: /* * Masking above should prevent us ever trying * to take action on an ignored signal other * than SIGCONT, unless process is traced. */ if ((prop & SIGPROP_CONT) == 0 && (p->p_flag & P_TRACED) == 0) printf("issignal\n"); break; /* == ignore */ default: /* * This signal has an action, let * postsig() process it. */ return (sig); } sigqueue_delete(&td->td_sigqueue, sig); /* take the signal! */ sigqueue_delete(&p->p_sigqueue, sig); next:; } /* NOTREACHED */ } void thread_stopped(struct proc *p) { int n; PROC_LOCK_ASSERT(p, MA_OWNED); PROC_SLOCK_ASSERT(p, MA_OWNED); n = p->p_suspcount; if (p == curproc) n++; if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) { PROC_SUNLOCK(p); p->p_flag &= ~P_WAITED; PROC_LOCK(p->p_pptr); childproc_stopped(p, (p->p_flag & P_TRACED) ? CLD_TRAPPED : CLD_STOPPED); PROC_UNLOCK(p->p_pptr); PROC_SLOCK(p); } } /* * Take the action for the specified signal * from the current set of pending signals. */ int postsig(sig) register int sig; { struct thread *td = curthread; register struct proc *p = td->td_proc; struct sigacts *ps; sig_t action; ksiginfo_t ksi; sigset_t returnmask; KASSERT(sig != 0, ("postsig")); PROC_LOCK_ASSERT(p, MA_OWNED); ps = p->p_sigacts; mtx_assert(&ps->ps_mtx, MA_OWNED); ksiginfo_init(&ksi); if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 && sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0) return (0); ksi.ksi_signo = sig; if (ksi.ksi_code == SI_TIMER) itimer_accept(p, ksi.ksi_timerid, &ksi); action = ps->ps_sigact[_SIG_IDX(sig)]; #ifdef KTRACE if (KTRPOINT(td, KTR_PSIG)) ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ? &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code); #endif if (p->p_stops & S_SIG) { mtx_unlock(&ps->ps_mtx); stopevent(p, S_SIG, sig); mtx_lock(&ps->ps_mtx); } if (action == SIG_DFL) { /* * Default action, where the default is to kill * the process. (Other cases were ignored above.) */ mtx_unlock(&ps->ps_mtx); sigexit(td, sig); /* NOTREACHED */ } else { /* * If we get here, the signal must be caught. */ KASSERT(action != SIG_IGN && !SIGISMEMBER(td->td_sigmask, sig), ("postsig action")); /* * Set the new mask value and also defer further * occurrences of this signal. * * Special case: user has done a sigsuspend. Here the * current mask is not of interest, but rather the * mask from before the sigsuspend is what we want * restored after the signal processing is completed. */ if (td->td_pflags & TDP_OLDMASK) { returnmask = td->td_oldsigmask; td->td_pflags &= ~TDP_OLDMASK; } else returnmask = td->td_sigmask; if (p->p_sig == sig) { p->p_code = 0; p->p_sig = 0; } (*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask); postsig_done(sig, td, ps); } return (1); } /* * Kill the current process for stated reason. */ void killproc(p, why) struct proc *p; char *why; { PROC_LOCK_ASSERT(p, MA_OWNED); CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid, p->p_comm); log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n", p->p_pid, p->p_comm, p->p_ucred ? p->p_ucred->cr_uid : -1, why); p->p_flag |= P_WKILLED; kern_psignal(p, SIGKILL); } /* * Force the current process to exit with the specified signal, dumping core * if appropriate. We bypass the normal tests for masked and caught signals, * allowing unrecoverable failures to terminate the process without changing * signal state. Mark the accounting record with the signal termination. * If dumping core, save the signal number for the debugger. Calls exit and * does not return. */ void sigexit(td, sig) struct thread *td; int sig; { struct proc *p = td->td_proc; PROC_LOCK_ASSERT(p, MA_OWNED); p->p_acflag |= AXSIG; /* * We must be single-threading to generate a core dump. This * ensures that the registers in the core file are up-to-date. * Also, the ELF dump handler assumes that the thread list doesn't * change out from under it. * * XXX If another thread attempts to single-thread before us * (e.g. via fork()), we won't get a dump at all. */ if ((sigprop(sig) & SIGPROP_CORE) && thread_single(p, SINGLE_NO_EXIT) == 0) { p->p_sig = sig; /* * Log signals which would cause core dumps * (Log as LOG_INFO to appease those who don't want * these messages.) * XXX : Todo, as well as euid, write out ruid too * Note that coredump() drops proc lock. */ if (coredump(td) == 0) sig |= WCOREFLAG; if (kern_logsigexit) log(LOG_INFO, "pid %d (%s), uid %d: exited on signal %d%s\n", p->p_pid, p->p_comm, td->td_ucred ? td->td_ucred->cr_uid : -1, sig &~ WCOREFLAG, sig & WCOREFLAG ? " (core dumped)" : ""); } else PROC_UNLOCK(p); exit1(td, 0, sig); /* NOTREACHED */ } /* * Send queued SIGCHLD to parent when child process's state * is changed. */ static void sigparent(struct proc *p, int reason, int status) { PROC_LOCK_ASSERT(p, MA_OWNED); PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED); if (p->p_ksi != NULL) { p->p_ksi->ksi_signo = SIGCHLD; p->p_ksi->ksi_code = reason; p->p_ksi->ksi_status = status; p->p_ksi->ksi_pid = p->p_pid; p->p_ksi->ksi_uid = p->p_ucred->cr_ruid; if (KSI_ONQ(p->p_ksi)) return; } pksignal(p->p_pptr, SIGCHLD, p->p_ksi); } static void childproc_jobstate(struct proc *p, int reason, int sig) { struct sigacts *ps; PROC_LOCK_ASSERT(p, MA_OWNED); PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED); /* * Wake up parent sleeping in kern_wait(), also send * SIGCHLD to parent, but SIGCHLD does not guarantee * that parent will awake, because parent may masked * the signal. */ p->p_pptr->p_flag |= P_STATCHILD; wakeup(p->p_pptr); ps = p->p_pptr->p_sigacts; mtx_lock(&ps->ps_mtx); if ((ps->ps_flag & PS_NOCLDSTOP) == 0) { mtx_unlock(&ps->ps_mtx); sigparent(p, reason, sig); } else mtx_unlock(&ps->ps_mtx); } void childproc_stopped(struct proc *p, int reason) { childproc_jobstate(p, reason, p->p_xsig); } void childproc_continued(struct proc *p) { childproc_jobstate(p, CLD_CONTINUED, SIGCONT); } void childproc_exited(struct proc *p) { int reason, status; if (WCOREDUMP(p->p_xsig)) { reason = CLD_DUMPED; status = WTERMSIG(p->p_xsig); } else if (WIFSIGNALED(p->p_xsig)) { reason = CLD_KILLED; status = WTERMSIG(p->p_xsig); } else { reason = CLD_EXITED; status = p->p_xexit; } /* * XXX avoid calling wakeup(p->p_pptr), the work is * done in exit1(). */ sigparent(p, reason, status); } /* * We only have 1 character for the core count in the format * string, so the range will be 0-9 */ #define MAX_NUM_CORE_FILES 10 #ifndef NUM_CORE_FILES #define NUM_CORE_FILES 5 #endif CTASSERT(NUM_CORE_FILES >= 0 && NUM_CORE_FILES <= MAX_NUM_CORE_FILES); static int num_cores = NUM_CORE_FILES; static int sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS) { int error; int new_val; new_val = num_cores; error = sysctl_handle_int(oidp, &new_val, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (new_val > MAX_NUM_CORE_FILES) new_val = MAX_NUM_CORE_FILES; if (new_val < 0) new_val = 0; num_cores = new_val; return (0); } SYSCTL_PROC(_debug, OID_AUTO, ncores, CTLTYPE_INT|CTLFLAG_RW, 0, sizeof(int), sysctl_debug_num_cores_check, "I", ""); #define GZ_SUFFIX ".gz" #ifdef GZIO static int compress_user_cores = 1; SYSCTL_INT(_kern, OID_AUTO, compress_user_cores, CTLFLAG_RWTUN, &compress_user_cores, 0, "Compression of user corefiles"); int compress_user_cores_gzlevel = 6; SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_gzlevel, CTLFLAG_RWTUN, &compress_user_cores_gzlevel, 0, "Corefile gzip compression level"); #else static int compress_user_cores = 0; #endif /* * Protect the access to corefilename[] by allproc_lock. */ #define corefilename_lock allproc_lock static char corefilename[MAXPATHLEN] = {"%N.core"}; TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename)); static int sysctl_kern_corefile(SYSCTL_HANDLER_ARGS) { int error; sx_xlock(&corefilename_lock); error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename), req); sx_xunlock(&corefilename_lock); return (error); } SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A", "Process corefile name format string"); /* * corefile_open(comm, uid, pid, td, compress, vpp, namep) * Expand the name described in corefilename, using name, uid, and pid * and open/create core file. * corefilename is a printf-like string, with three format specifiers: * %N name of process ("name") * %P process id (pid) * %U user id (uid) * For example, "%N.core" is the default; they can be disabled completely * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P". * This is controlled by the sysctl variable kern.corefile (see above). */ static int corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td, int compress, struct vnode **vpp, char **namep) { struct nameidata nd; struct sbuf sb; const char *format; char *hostname, *name; int indexpos, i, error, cmode, flags, oflags; hostname = NULL; format = corefilename; name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO); indexpos = -1; (void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN); sx_slock(&corefilename_lock); for (i = 0; format[i] != '\0'; i++) { switch (format[i]) { case '%': /* Format character */ i++; switch (format[i]) { case '%': sbuf_putc(&sb, '%'); break; case 'H': /* hostname */ if (hostname == NULL) { hostname = malloc(MAXHOSTNAMELEN, M_TEMP, M_WAITOK); } getcredhostname(td->td_ucred, hostname, MAXHOSTNAMELEN); sbuf_printf(&sb, "%s", hostname); break; case 'I': /* autoincrementing index */ sbuf_printf(&sb, "0"); indexpos = sbuf_len(&sb) - 1; break; case 'N': /* process name */ sbuf_printf(&sb, "%s", comm); break; case 'P': /* process id */ sbuf_printf(&sb, "%u", pid); break; case 'U': /* user id */ sbuf_printf(&sb, "%u", uid); break; default: log(LOG_ERR, "Unknown format character %c in " "corename `%s'\n", format[i], format); break; } break; default: sbuf_putc(&sb, format[i]); break; } } sx_sunlock(&corefilename_lock); free(hostname, M_TEMP); if (compress) sbuf_printf(&sb, GZ_SUFFIX); if (sbuf_error(&sb) != 0) { log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too " "long\n", (long)pid, comm, (u_long)uid); sbuf_delete(&sb); free(name, M_TEMP); return (ENOMEM); } sbuf_finish(&sb); sbuf_delete(&sb); cmode = S_IRUSR | S_IWUSR; oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE | (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0); /* * If the core format has a %I in it, then we need to check * for existing corefiles before returning a name. * To do this we iterate over 0..num_cores to find a * non-existing core file name to use. */ if (indexpos != -1) { for (i = 0; i < num_cores; i++) { flags = O_CREAT | O_EXCL | FWRITE | O_NOFOLLOW; name[indexpos] = '0' + i; NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td); error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred, NULL); if (error) { if (error == EEXIST) continue; log(LOG_ERR, "pid %d (%s), uid (%u): Path `%s' failed " "on initial open test, error = %d\n", pid, comm, uid, name, error); } goto out; } } flags = O_CREAT | FWRITE | O_NOFOLLOW; NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td); error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred, NULL); out: if (error) { #ifdef AUDIT audit_proc_coredump(td, name, error); #endif free(name, M_TEMP); return (error); } NDFREE(&nd, NDF_ONLY_PNBUF); *vpp = nd.ni_vp; *namep = name; return (0); } static int coredump_sanitise_path(const char *path) { size_t i; /* * Only send a subset of ASCII to devd(8) because it * might pass these strings to sh -c. */ for (i = 0; path[i]; i++) if (!(isalpha(path[i]) || isdigit(path[i])) && path[i] != '/' && path[i] != '.' && path[i] != '-') return (0); return (1); } /* * Dump a process' core. The main routine does some * policy checking, and creates the name of the coredump; * then it passes on a vnode and a size limit to the process-specific * coredump routine if there is one; if there _is not_ one, it returns * ENOSYS; otherwise it returns the error from the process-specific routine. */ static int coredump(struct thread *td) { struct proc *p = td->td_proc; struct ucred *cred = td->td_ucred; struct vnode *vp; struct flock lf; struct vattr vattr; int error, error1, locked; char *name; /* name of corefile */ void *rl_cookie; off_t limit; char *data = NULL; char *fullpath, *freepath = NULL; size_t len; static const char comm_name[] = "comm="; static const char core_name[] = "core="; PROC_LOCK_ASSERT(p, MA_OWNED); MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td); _STOPEVENT(p, S_CORE, 0); if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) || (p->p_flag2 & P2_NOTRACE) != 0) { PROC_UNLOCK(p); return (EFAULT); } /* * Note that the bulk of limit checking is done after * the corefile is created. The exception is if the limit * for corefiles is 0, in which case we don't bother * creating the corefile at all. This layout means that * a corefile is truncated instead of not being created, * if it is larger than the limit. */ limit = (off_t)lim_cur(td, RLIMIT_CORE); if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) { PROC_UNLOCK(p); return (EFBIG); } PROC_UNLOCK(p); error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td, compress_user_cores, &vp, &name); if (error != 0) return (error); /* * Don't dump to non-regular files or files with links. * Do not dump into system files. */ if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 || vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0) { VOP_UNLOCK(vp, 0); error = EFAULT; goto out; } VOP_UNLOCK(vp, 0); /* Postpone other writers, including core dumps of other processes. */ rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; lf.l_type = F_WRLCK; locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0); VATTR_NULL(&vattr); vattr.va_size = 0; if (set_core_nodump_flag) vattr.va_flags = UF_NODUMP; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); VOP_SETATTR(vp, &vattr, cred); VOP_UNLOCK(vp, 0); PROC_LOCK(p); p->p_acflag |= ACORE; PROC_UNLOCK(p); if (p->p_sysent->sv_coredump != NULL) { error = p->p_sysent->sv_coredump(td, vp, limit, compress_user_cores ? IMGACT_CORE_COMPRESS : 0); } else { error = ENOSYS; } if (locked) { lf.l_type = F_UNLCK; VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK); } vn_rangelock_unlock(vp, rl_cookie); /* * Notify the userland helper that a process triggered a core dump. * This allows the helper to run an automated debugging session. */ if (error != 0 || coredump_devctl == 0) goto out; len = MAXPATHLEN * 2 + sizeof(comm_name) - 1 + sizeof(' ') + sizeof(core_name) - 1; data = malloc(len, M_TEMP, M_WAITOK); if (vn_fullpath_global(td, p->p_textvp, &fullpath, &freepath) != 0) goto out; if (!coredump_sanitise_path(fullpath)) goto out; snprintf(data, len, "%s%s ", comm_name, fullpath); free(freepath, M_TEMP); freepath = NULL; if (vn_fullpath_global(td, vp, &fullpath, &freepath) != 0) goto out; if (!coredump_sanitise_path(fullpath)) goto out; strlcat(data, core_name, len); strlcat(data, fullpath, len); devctl_notify("kernel", "signal", "coredump", data); out: error1 = vn_close(vp, FWRITE, cred, td); if (error == 0) error = error1; #ifdef AUDIT audit_proc_coredump(td, name, error); #endif free(freepath, M_TEMP); free(data, M_TEMP); free(name, M_TEMP); return (error); } /* * Nonexistent system call-- signal process (may want to handle it). Flag * error in case process won't see signal immediately (blocked or ignored). */ #ifndef _SYS_SYSPROTO_H_ struct nosys_args { int dummy; }; #endif /* ARGSUSED */ int nosys(td, args) struct thread *td; struct nosys_args *args; { struct proc *p = td->td_proc; PROC_LOCK(p); tdsignal(td, SIGSYS); PROC_UNLOCK(p); return (ENOSYS); } /* * Send a SIGIO or SIGURG signal to a process or process group using stored * credentials rather than those of the current process. */ void pgsigio(sigiop, sig, checkctty) struct sigio **sigiop; int sig, checkctty; { ksiginfo_t ksi; struct sigio *sigio; ksiginfo_init(&ksi); ksi.ksi_signo = sig; ksi.ksi_code = SI_KERNEL; SIGIO_LOCK(); sigio = *sigiop; if (sigio == NULL) { SIGIO_UNLOCK(); return; } if (sigio->sio_pgid > 0) { PROC_LOCK(sigio->sio_proc); if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred)) kern_psignal(sigio->sio_proc, sig); PROC_UNLOCK(sigio->sio_proc); } else if (sigio->sio_pgid < 0) { struct proc *p; PGRP_LOCK(sigio->sio_pgrp); LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) { PROC_LOCK(p); if (p->p_state == PRS_NORMAL && CANSIGIO(sigio->sio_ucred, p->p_ucred) && (checkctty == 0 || (p->p_flag & P_CONTROLT))) kern_psignal(p, sig); PROC_UNLOCK(p); } PGRP_UNLOCK(sigio->sio_pgrp); } SIGIO_UNLOCK(); } static int filt_sigattach(struct knote *kn) { struct proc *p = curproc; kn->kn_ptr.p_proc = p; kn->kn_flags |= EV_CLEAR; /* automatically set */ knlist_add(p->p_klist, kn, 0); return (0); } static void filt_sigdetach(struct knote *kn) { struct proc *p = kn->kn_ptr.p_proc; knlist_remove(p->p_klist, kn, 0); } /* * signal knotes are shared with proc knotes, so we apply a mask to * the hint in order to differentiate them from process hints. This * could be avoided by using a signal-specific knote list, but probably * isn't worth the trouble. */ static int filt_signal(struct knote *kn, long hint) { if (hint & NOTE_SIGNAL) { hint &= ~NOTE_SIGNAL; if (kn->kn_id == hint) kn->kn_data++; } return (kn->kn_data != 0); } struct sigacts * sigacts_alloc(void) { struct sigacts *ps; ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO); refcount_init(&ps->ps_refcnt, 1); mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF); return (ps); } void sigacts_free(struct sigacts *ps) { if (refcount_release(&ps->ps_refcnt) == 0) return; mtx_destroy(&ps->ps_mtx); free(ps, M_SUBPROC); } struct sigacts * sigacts_hold(struct sigacts *ps) { refcount_acquire(&ps->ps_refcnt); return (ps); } void sigacts_copy(struct sigacts *dest, struct sigacts *src) { KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest")); mtx_lock(&src->ps_mtx); bcopy(src, dest, offsetof(struct sigacts, ps_refcnt)); mtx_unlock(&src->ps_mtx); } int sigacts_shared(struct sigacts *ps) { return (ps->ps_refcnt > 1); } Index: head/sys/kern/kern_synch.c =================================================================== --- head/sys/kern/kern_synch.c (revision 305831) +++ head/sys/kern/kern_synch.c (revision 305832) @@ -1,594 +1,594 @@ /*- * Copyright (c) 1982, 1986, 1990, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_synch.c 8.9 (Berkeley) 5/19/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_ktrace.h" #include "opt_sched.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KTRACE #include #include #endif #include #define KTDSTATE(td) \ (((td)->td_inhibitors & TDI_SLEEPING) != 0 ? "sleep" : \ ((td)->td_inhibitors & TDI_SUSPENDED) != 0 ? "suspended" : \ ((td)->td_inhibitors & TDI_SWAPPED) != 0 ? "swapped" : \ ((td)->td_inhibitors & TDI_LOCK) != 0 ? "blocked" : \ ((td)->td_inhibitors & TDI_IWAIT) != 0 ? "iwait" : "yielding") static void synch_setup(void *dummy); SYSINIT(synch_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, synch_setup, NULL); int hogticks; static uint8_t pause_wchan[MAXCPU]; static struct callout loadav_callout; struct loadavg averunnable = { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */ /* * Constants for averages over 1, 5, and 15 minutes * when sampling at 5 second intervals. */ static fixpt_t cexp[3] = { 0.9200444146293232 * FSCALE, /* exp(-1/12) */ 0.9834714538216174 * FSCALE, /* exp(-1/60) */ 0.9944598480048967 * FSCALE, /* exp(-1/180) */ }; /* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */ SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, FSCALE, ""); static void loadav(void *arg); SDT_PROVIDER_DECLARE(sched); SDT_PROBE_DEFINE(sched, , , preempt); static void sleepinit(void *unused) { hogticks = (hz / 10) * 2; /* Default only. */ init_sleepqueues(); } /* * vmem tries to lock the sleepq mutexes when free'ing kva, so make sure * it is available. */ SYSINIT(sleepinit, SI_SUB_KMEM, SI_ORDER_ANY, sleepinit, 0); /* * General sleep call. Suspends the current thread until a wakeup is * performed on the specified identifier. The thread will then be made * runnable with the specified priority. Sleeps at most sbt units of time * (0 means no timeout). If pri includes the PCATCH flag, let signals * interrupt the sleep, otherwise ignore them while sleeping. Returns 0 if * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a * signal becomes pending, ERESTART is returned if the current system * call should be restarted if possible, and EINTR is returned if the system * call should be interrupted by the signal (return EINTR). * * The lock argument is unlocked before the caller is suspended, and * re-locked before _sleep() returns. If priority includes the PDROP * flag the lock is not re-locked before returning. */ int _sleep(void *ident, struct lock_object *lock, int priority, const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags) { struct thread *td; struct proc *p; struct lock_class *class; uintptr_t lock_state; int catch, pri, rval, sleepq_flags; WITNESS_SAVE_DECL(lock_witness); td = curthread; p = td->td_proc; #ifdef KTRACE if (KTRPOINT(td, KTR_CSW)) ktrcsw(1, 0, wmesg); #endif WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock, "Sleeping on \"%s\"", wmesg); KASSERT(sbt != 0 || mtx_owned(&Giant) || lock != NULL, ("sleeping without a lock")); KASSERT(p != NULL, ("msleep1")); KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); if (priority & PDROP) KASSERT(lock != NULL && lock != &Giant.lock_object, ("PDROP requires a non-Giant lock")); if (lock != NULL) class = LOCK_CLASS(lock); else class = NULL; if (SCHEDULER_STOPPED()) { if (lock != NULL && priority & PDROP) class->lc_unlock(lock); return (0); } catch = priority & PCATCH; pri = priority & PRIMASK; /* * If we are already on a sleep queue, then remove us from that * sleep queue first. We have to do this to handle recursive * sleeps. */ if (TD_ON_SLEEPQ(td)) sleepq_remove(td, td->td_wchan); if ((uint8_t *)ident >= &pause_wchan[0] && (uint8_t *)ident <= &pause_wchan[MAXCPU - 1]) sleepq_flags = SLEEPQ_PAUSE; else sleepq_flags = SLEEPQ_SLEEP; if (catch) sleepq_flags |= SLEEPQ_INTERRUPTIBLE; sleepq_lock(ident); CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)", td->td_tid, p->p_pid, td->td_name, wmesg, ident); if (lock == &Giant.lock_object) mtx_assert(&Giant, MA_OWNED); DROP_GIANT(); if (lock != NULL && lock != &Giant.lock_object && !(class->lc_flags & LC_SLEEPABLE)) { WITNESS_SAVE(lock, lock_witness); lock_state = class->lc_unlock(lock); } else /* GCC needs to follow the Yellow Brick Road */ lock_state = -1; /* * We put ourselves on the sleep queue and start our timeout * before calling thread_suspend_check, as we could stop there, * and a wakeup or a SIGCONT (or both) could occur while we were * stopped without resuming us. Thus, we must be ready for sleep * when cursig() is called. If the wakeup happens while we're * stopped, then td will no longer be on a sleep queue upon * return from cursig(). */ sleepq_add(ident, lock, wmesg, sleepq_flags, 0); if (sbt != 0) sleepq_set_timeout_sbt(ident, sbt, pr, flags); if (lock != NULL && class->lc_flags & LC_SLEEPABLE) { sleepq_release(ident); WITNESS_SAVE(lock, lock_witness); lock_state = class->lc_unlock(lock); sleepq_lock(ident); } if (sbt != 0 && catch) rval = sleepq_timedwait_sig(ident, pri); else if (sbt != 0) rval = sleepq_timedwait(ident, pri); else if (catch) rval = sleepq_wait_sig(ident, pri); else { sleepq_wait(ident, pri); rval = 0; } #ifdef KTRACE if (KTRPOINT(td, KTR_CSW)) ktrcsw(0, 0, wmesg); #endif PICKUP_GIANT(); if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) { class->lc_lock(lock, lock_state); WITNESS_RESTORE(lock, lock_witness); } return (rval); } int msleep_spin_sbt(void *ident, struct mtx *mtx, const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags) { struct thread *td; struct proc *p; int rval; WITNESS_SAVE_DECL(mtx); td = curthread; p = td->td_proc; KASSERT(mtx != NULL, ("sleeping without a mutex")); KASSERT(p != NULL, ("msleep1")); KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); if (SCHEDULER_STOPPED()) return (0); sleepq_lock(ident); CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)", td->td_tid, p->p_pid, td->td_name, wmesg, ident); DROP_GIANT(); mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); WITNESS_SAVE(&mtx->lock_object, mtx); mtx_unlock_spin(mtx); /* * We put ourselves on the sleep queue and start our timeout. */ sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0); if (sbt != 0) sleepq_set_timeout_sbt(ident, sbt, pr, flags); /* * Can't call ktrace with any spin locks held so it can lock the * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold * any spin lock. Thus, we have to drop the sleepq spin lock while * we handle those requests. This is safe since we have placed our * thread on the sleep queue already. */ #ifdef KTRACE if (KTRPOINT(td, KTR_CSW)) { sleepq_release(ident); ktrcsw(1, 0, wmesg); sleepq_lock(ident); } #endif #ifdef WITNESS sleepq_release(ident); WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"", wmesg); sleepq_lock(ident); #endif if (sbt != 0) rval = sleepq_timedwait(ident, 0); else { sleepq_wait(ident, 0); rval = 0; } #ifdef KTRACE if (KTRPOINT(td, KTR_CSW)) ktrcsw(0, 0, wmesg); #endif PICKUP_GIANT(); mtx_lock_spin(mtx); WITNESS_RESTORE(&mtx->lock_object, mtx); return (rval); } /* * pause() delays the calling thread by the given number of system ticks. * During cold bootup, pause() uses the DELAY() function instead of * the tsleep() function to do the waiting. The "timo" argument must be * greater than or equal to zero. A "timo" value of zero is equivalent * to a "timo" value of one. */ int pause_sbt(const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags) { KASSERT(sbt >= 0, ("pause: timeout must be >= 0")); /* silently convert invalid timeouts */ if (sbt == 0) sbt = tick_sbt; if (cold || kdb_active || SCHEDULER_STOPPED()) { /* * We delay one second at a time to avoid overflowing the * system specific DELAY() function(s): */ while (sbt >= SBT_1S) { DELAY(1000000); sbt -= SBT_1S; } /* Do the delay remainder, if any */ sbt = howmany(sbt, SBT_1US); if (sbt > 0) DELAY(sbt); return (0); } return (_sleep(&pause_wchan[curcpu], NULL, 0, wmesg, sbt, pr, flags)); } /* * Make all threads sleeping on the specified identifier runnable. */ void wakeup(void *ident) { int wakeup_swapper; sleepq_lock(ident); wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0); sleepq_release(ident); if (wakeup_swapper) { KASSERT(ident != &proc0, ("wakeup and wakeup_swapper and proc0")); kick_proc0(); } } /* * Make a thread sleeping on the specified identifier runnable. * May wake more than one thread if a target thread is currently * swapped out. */ void wakeup_one(void *ident) { int wakeup_swapper; sleepq_lock(ident); wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP, 0, 0); sleepq_release(ident); if (wakeup_swapper) kick_proc0(); } static void kdb_switch(void) { thread_unlock(curthread); kdb_backtrace(); kdb_reenter(); panic("%s: did not reenter debugger", __func__); } /* * The machine independent parts of context switching. */ void mi_switch(int flags, struct thread *newtd) { uint64_t runtime, new_switchtime; struct thread *td; td = curthread; /* XXX */ THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED); KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code")); #ifdef INVARIANTS if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td)) mtx_assert(&Giant, MA_NOTOWNED); #endif KASSERT(td->td_critnest == 1 || panicstr, ("mi_switch: switch in a critical section")); KASSERT((flags & (SW_INVOL | SW_VOL)) != 0, ("mi_switch: switch must be voluntary or involuntary")); KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself")); /* * Don't perform context switches from the debugger. */ if (kdb_active) kdb_switch(); if (SCHEDULER_STOPPED()) return; if (flags & SW_VOL) { td->td_ru.ru_nvcsw++; td->td_swvoltick = ticks; } else { td->td_ru.ru_nivcsw++; td->td_swinvoltick = ticks; } #ifdef SCHED_STATS SCHED_STAT_INC(sched_switch_stats[flags & SW_TYPE_MASK]); #endif /* * Compute the amount of time during which the current * thread was running, and add that to its total so far. */ new_switchtime = cpu_ticks(); runtime = new_switchtime - PCPU_GET(switchtime); td->td_runtime += runtime; td->td_incruntime += runtime; PCPU_SET(switchtime, new_switchtime); td->td_generation++; /* bump preempt-detect counter */ PCPU_INC(cnt.v_swtch); PCPU_SET(switchticks, ticks); CTR4(KTR_PROC, "mi_switch: old thread %ld (td_sched %p, pid %ld, %s)", td->td_tid, td_get_sched(td), td->td_proc->p_pid, td->td_name); #if (KTR_COMPILE & KTR_SCHED) != 0 if (TD_IS_IDLETHREAD(td)) KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "idle", "prio:%d", td->td_priority); else KTR_STATE3(KTR_SCHED, "thread", sched_tdname(td), KTDSTATE(td), "prio:%d", td->td_priority, "wmesg:\"%s\"", td->td_wmesg, "lockname:\"%s\"", td->td_lockname); #endif SDT_PROBE0(sched, , , preempt); sched_switch(td, newtd, flags); KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "running", "prio:%d", td->td_priority); CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)", td->td_tid, td_get_sched(td), td->td_proc->p_pid, td->td_name); /* * If the last thread was exiting, finish cleaning it up. */ if ((td = PCPU_GET(deadthread))) { PCPU_SET(deadthread, NULL); thread_stash(td); } } /* * Change thread state to be runnable, placing it on the run queue if * it is in memory. If it is swapped out, return true so our caller * will know to awaken the swapper. */ int setrunnable(struct thread *td) { THREAD_LOCK_ASSERT(td, MA_OWNED); KASSERT(td->td_proc->p_state != PRS_ZOMBIE, ("setrunnable: pid %d is a zombie", td->td_proc->p_pid)); switch (td->td_state) { case TDS_RUNNING: case TDS_RUNQ: return (0); case TDS_INHIBITED: /* * If we are only inhibited because we are swapped out * then arange to swap in this process. Otherwise just return. */ if (td->td_inhibitors != TDI_SWAPPED) return (0); /* FALLTHROUGH */ case TDS_CAN_RUN: break; default: printf("state is 0x%x", td->td_state); panic("setrunnable(2)"); } if ((td->td_flags & TDF_INMEM) == 0) { if ((td->td_flags & TDF_SWAPINREQ) == 0) { td->td_flags |= TDF_SWAPINREQ; return (1); } } else sched_wakeup(td); return (0); } /* * Compute a tenex style load average of a quantity on * 1, 5 and 15 minute intervals. */ static void loadav(void *arg) { int i, nrun; struct loadavg *avg; nrun = sched_load(); avg = &averunnable; for (i = 0; i < 3; i++) avg->ldavg[i] = (cexp[i] * avg->ldavg[i] + nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT; /* * Schedule the next update to occur after 5 seconds, but add a * random variation to avoid synchronisation with processes that * run at regular intervals. */ callout_reset_sbt(&loadav_callout, SBT_1US * (4000000 + (int)(random() % 2000001)), SBT_1US, loadav, NULL, C_DIRECT_EXEC | C_PREL(32)); } /* ARGSUSED */ static void synch_setup(void *dummy) { callout_init(&loadav_callout, 1); /* Kick off timeout driven events by calling first time. */ loadav(NULL); } int should_yield(void) { return ((u_int)ticks - (u_int)curthread->td_swvoltick >= hogticks); } void maybe_yield(void) { if (should_yield()) kern_yield(PRI_USER); } void kern_yield(int prio) { struct thread *td; td = curthread; DROP_GIANT(); thread_lock(td); if (prio == PRI_USER) prio = td->td_user_pri; if (prio >= 0) sched_prio(td, prio); mi_switch(SW_VOL | SWT_RELINQUISH, NULL); thread_unlock(td); PICKUP_GIANT(); } /* * General purpose yield system call. */ int sys_yield(struct thread *td, struct yield_args *uap) { thread_lock(td); if (PRI_BASE(td->td_pri_class) == PRI_TIMESHARE) sched_prio(td, PRI_MAX_TIMESHARE); mi_switch(SW_VOL | SWT_RELINQUISH, NULL); thread_unlock(td); td->td_retval[0] = 0; return (0); } Index: head/sys/kern/kern_sysctl.c =================================================================== --- head/sys/kern/kern_sysctl.c (revision 305831) +++ head/sys/kern/kern_sysctl.c (revision 305832) @@ -1,2030 +1,2030 @@ /*- * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Mike Karels at Berkeley Software Design, Inc. * * Quite extensively rewritten by Poul-Henning Kamp of the FreeBSD * project, to make these variables more userfriendly. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_sysctl.c 8.4 (Berkeley) 4/14/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_capsicum.h" #include "opt_compat.h" #include "opt_ktrace.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KTRACE #include #endif #include #include #include #include static MALLOC_DEFINE(M_SYSCTL, "sysctl", "sysctl internal magic"); static MALLOC_DEFINE(M_SYSCTLOID, "sysctloid", "sysctl dynamic oids"); static MALLOC_DEFINE(M_SYSCTLTMP, "sysctltmp", "sysctl temp output buffer"); /* * The sysctllock protects the MIB tree. It also protects sysctl * contexts used with dynamic sysctls. The sysctl_register_oid() and * sysctl_unregister_oid() routines require the sysctllock to already * be held, so the sysctl_wlock() and sysctl_wunlock() routines are * provided for the few places in the kernel which need to use that * API rather than using the dynamic API. Use of the dynamic API is * strongly encouraged for most code. * * The sysctlmemlock is used to limit the amount of user memory wired for * sysctl requests. This is implemented by serializing any userland * sysctl requests larger than a single page via an exclusive lock. */ static struct rmlock sysctllock; static struct sx sysctlmemlock; #define SYSCTL_WLOCK() rm_wlock(&sysctllock) #define SYSCTL_WUNLOCK() rm_wunlock(&sysctllock) #define SYSCTL_RLOCK(tracker) rm_rlock(&sysctllock, (tracker)) #define SYSCTL_RUNLOCK(tracker) rm_runlock(&sysctllock, (tracker)) #define SYSCTL_WLOCKED() rm_wowned(&sysctllock) #define SYSCTL_ASSERT_LOCKED() rm_assert(&sysctllock, RA_LOCKED) #define SYSCTL_ASSERT_WLOCKED() rm_assert(&sysctllock, RA_WLOCKED) #define SYSCTL_ASSERT_RLOCKED() rm_assert(&sysctllock, RA_RLOCKED) #define SYSCTL_INIT() rm_init_flags(&sysctllock, "sysctl lock", \ RM_SLEEPABLE) #define SYSCTL_SLEEP(ch, wmesg, timo) \ rm_sleep(ch, &sysctllock, 0, wmesg, timo) static int sysctl_root(SYSCTL_HANDLER_ARGS); /* Root list */ struct sysctl_oid_list sysctl__children = SLIST_HEAD_INITIALIZER(&sysctl__children); static int sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse); static int sysctl_old_kernel(struct sysctl_req *, const void *, size_t); static int sysctl_new_kernel(struct sysctl_req *, void *, size_t); static struct sysctl_oid * sysctl_find_oidname(const char *name, struct sysctl_oid_list *list) { struct sysctl_oid *oidp; SYSCTL_ASSERT_LOCKED(); SLIST_FOREACH(oidp, list, oid_link) { if (strcmp(oidp->oid_name, name) == 0) { return (oidp); } } return (NULL); } /* * Initialization of the MIB tree. * * Order by number in each list. */ void sysctl_wlock(void) { SYSCTL_WLOCK(); } void sysctl_wunlock(void) { SYSCTL_WUNLOCK(); } static int sysctl_root_handler_locked(struct sysctl_oid *oid, void *arg1, intmax_t arg2, struct sysctl_req *req, struct rm_priotracker *tracker) { int error; if (oid->oid_kind & CTLFLAG_DYN) atomic_add_int(&oid->oid_running, 1); if (tracker != NULL) SYSCTL_RUNLOCK(tracker); else SYSCTL_WUNLOCK(); if (!(oid->oid_kind & CTLFLAG_MPSAFE)) mtx_lock(&Giant); error = oid->oid_handler(oid, arg1, arg2, req); if (!(oid->oid_kind & CTLFLAG_MPSAFE)) mtx_unlock(&Giant); KFAIL_POINT_ERROR(_debug_fail_point, sysctl_running, error); if (tracker != NULL) SYSCTL_RLOCK(tracker); else SYSCTL_WLOCK(); if (oid->oid_kind & CTLFLAG_DYN) { if (atomic_fetchadd_int(&oid->oid_running, -1) == 1 && (oid->oid_kind & CTLFLAG_DYING) != 0) wakeup(&oid->oid_running); } return (error); } static void sysctl_load_tunable_by_oid_locked(struct sysctl_oid *oidp) { struct sysctl_req req; struct sysctl_oid *curr; char *penv = NULL; char path[64]; ssize_t rem = sizeof(path); ssize_t len; uint8_t val_8; uint16_t val_16; uint32_t val_32; int val_int; long val_long; int64_t val_64; quad_t val_quad; int error; path[--rem] = 0; for (curr = oidp; curr != NULL; curr = SYSCTL_PARENT(curr)) { len = strlen(curr->oid_name); rem -= len; if (curr != oidp) rem -= 1; if (rem < 0) { printf("OID path exceeds %d bytes\n", (int)sizeof(path)); return; } memcpy(path + rem, curr->oid_name, len); if (curr != oidp) path[rem + len] = '.'; } memset(&req, 0, sizeof(req)); req.td = curthread; req.oldfunc = sysctl_old_kernel; req.newfunc = sysctl_new_kernel; req.lock = REQ_UNWIRED; switch (oidp->oid_kind & CTLTYPE) { case CTLTYPE_INT: if (getenv_int(path + rem, &val_int) == 0) return; req.newlen = sizeof(val_int); req.newptr = &val_int; break; case CTLTYPE_UINT: if (getenv_uint(path + rem, (unsigned int *)&val_int) == 0) return; req.newlen = sizeof(val_int); req.newptr = &val_int; break; case CTLTYPE_LONG: if (getenv_long(path + rem, &val_long) == 0) return; req.newlen = sizeof(val_long); req.newptr = &val_long; break; case CTLTYPE_ULONG: if (getenv_ulong(path + rem, (unsigned long *)&val_long) == 0) return; req.newlen = sizeof(val_long); req.newptr = &val_long; break; case CTLTYPE_S8: if (getenv_int(path + rem, &val_int) == 0) return; val_8 = val_int; req.newlen = sizeof(val_8); req.newptr = &val_8; break; case CTLTYPE_S16: if (getenv_int(path + rem, &val_int) == 0) return; val_16 = val_int; req.newlen = sizeof(val_16); req.newptr = &val_16; break; case CTLTYPE_S32: if (getenv_long(path + rem, &val_long) == 0) return; val_32 = val_long; req.newlen = sizeof(val_32); req.newptr = &val_32; break; case CTLTYPE_S64: if (getenv_quad(path + rem, &val_quad) == 0) return; val_64 = val_quad; req.newlen = sizeof(val_64); req.newptr = &val_64; break; case CTLTYPE_U8: if (getenv_uint(path + rem, (unsigned int *)&val_int) == 0) return; val_8 = val_int; req.newlen = sizeof(val_8); req.newptr = &val_8; break; case CTLTYPE_U16: if (getenv_uint(path + rem, (unsigned int *)&val_int) == 0) return; val_16 = val_int; req.newlen = sizeof(val_16); req.newptr = &val_16; break; case CTLTYPE_U32: if (getenv_ulong(path + rem, (unsigned long *)&val_long) == 0) return; val_32 = val_long; req.newlen = sizeof(val_32); req.newptr = &val_32; break; case CTLTYPE_U64: /* XXX there is no getenv_uquad() */ if (getenv_quad(path + rem, &val_quad) == 0) return; val_64 = val_quad; req.newlen = sizeof(val_64); req.newptr = &val_64; break; case CTLTYPE_STRING: penv = kern_getenv(path + rem); if (penv == NULL) return; req.newlen = strlen(penv); req.newptr = penv; break; default: return; } error = sysctl_root_handler_locked(oidp, oidp->oid_arg1, oidp->oid_arg2, &req, NULL); if (error != 0) printf("Setting sysctl %s failed: %d\n", path + rem, error); if (penv != NULL) freeenv(penv); } void sysctl_register_oid(struct sysctl_oid *oidp) { struct sysctl_oid_list *parent = oidp->oid_parent; struct sysctl_oid *p; struct sysctl_oid *q; int oid_number; int timeout = 2; /* * First check if another oid with the same name already * exists in the parent's list. */ SYSCTL_ASSERT_WLOCKED(); p = sysctl_find_oidname(oidp->oid_name, parent); if (p != NULL) { if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE) { p->oid_refcnt++; return; } else { printf("can't re-use a leaf (%s)!\n", p->oid_name); return; } } /* get current OID number */ oid_number = oidp->oid_number; #if (OID_AUTO >= 0) #error "OID_AUTO is expected to be a negative value" #endif /* * Any negative OID number qualifies as OID_AUTO. Valid OID * numbers should always be positive. * * NOTE: DO NOT change the starting value here, change it in * , and make sure it is at least 256 to * accommodate e.g. net.inet.raw as a static sysctl node. */ if (oid_number < 0) { static int newoid; /* * By decrementing the next OID number we spend less * time inserting the OIDs into a sorted list. */ if (--newoid < CTL_AUTO_START) newoid = 0x7fffffff; oid_number = newoid; } /* * Insert the OID into the parent's list sorted by OID number. */ retry: q = NULL; SLIST_FOREACH(p, parent, oid_link) { /* check if the current OID number is in use */ if (oid_number == p->oid_number) { /* get the next valid OID number */ if (oid_number < CTL_AUTO_START || oid_number == 0x7fffffff) { /* wraparound - restart */ oid_number = CTL_AUTO_START; /* don't loop forever */ if (!timeout--) panic("sysctl: Out of OID numbers\n"); goto retry; } else { oid_number++; } } else if (oid_number < p->oid_number) break; q = p; } /* check for non-auto OID number collision */ if (oidp->oid_number >= 0 && oidp->oid_number < CTL_AUTO_START && oid_number >= CTL_AUTO_START) { printf("sysctl: OID number(%d) is already in use for '%s'\n", oidp->oid_number, oidp->oid_name); } /* update the OID number, if any */ oidp->oid_number = oid_number; if (q != NULL) SLIST_INSERT_AFTER(q, oidp, oid_link); else SLIST_INSERT_HEAD(parent, oidp, oid_link); if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE && #ifdef VIMAGE (oidp->oid_kind & CTLFLAG_VNET) == 0 && #endif (oidp->oid_kind & CTLFLAG_TUN) != 0 && (oidp->oid_kind & CTLFLAG_NOFETCH) == 0) { /* only fetch value once */ oidp->oid_kind |= CTLFLAG_NOFETCH; /* try to fetch value from kernel environment */ sysctl_load_tunable_by_oid_locked(oidp); } } void sysctl_unregister_oid(struct sysctl_oid *oidp) { struct sysctl_oid *p; int error; SYSCTL_ASSERT_WLOCKED(); error = ENOENT; if (oidp->oid_number == OID_AUTO) { error = EINVAL; } else { SLIST_FOREACH(p, oidp->oid_parent, oid_link) { if (p == oidp) { SLIST_REMOVE(oidp->oid_parent, oidp, sysctl_oid, oid_link); error = 0; break; } } } /* * This can happen when a module fails to register and is * being unloaded afterwards. It should not be a panic() * for normal use. */ if (error) printf("%s: failed to unregister sysctl\n", __func__); } /* Initialize a new context to keep track of dynamically added sysctls. */ int sysctl_ctx_init(struct sysctl_ctx_list *c) { if (c == NULL) { return (EINVAL); } /* * No locking here, the caller is responsible for not adding * new nodes to a context until after this function has * returned. */ TAILQ_INIT(c); return (0); } /* Free the context, and destroy all dynamic oids registered in this context */ int sysctl_ctx_free(struct sysctl_ctx_list *clist) { struct sysctl_ctx_entry *e, *e1; int error; error = 0; /* * First perform a "dry run" to check if it's ok to remove oids. * XXX FIXME * XXX This algorithm is a hack. But I don't know any * XXX better solution for now... */ SYSCTL_WLOCK(); TAILQ_FOREACH(e, clist, link) { error = sysctl_remove_oid_locked(e->entry, 0, 0); if (error) break; } /* * Restore deregistered entries, either from the end, * or from the place where error occurred. * e contains the entry that was not unregistered */ if (error) e1 = TAILQ_PREV(e, sysctl_ctx_list, link); else e1 = TAILQ_LAST(clist, sysctl_ctx_list); while (e1 != NULL) { sysctl_register_oid(e1->entry); e1 = TAILQ_PREV(e1, sysctl_ctx_list, link); } if (error) { SYSCTL_WUNLOCK(); return(EBUSY); } /* Now really delete the entries */ e = TAILQ_FIRST(clist); while (e != NULL) { e1 = TAILQ_NEXT(e, link); error = sysctl_remove_oid_locked(e->entry, 1, 0); if (error) panic("sysctl_remove_oid: corrupt tree, entry: %s", e->entry->oid_name); free(e, M_SYSCTLOID); e = e1; } SYSCTL_WUNLOCK(); return (error); } /* Add an entry to the context */ struct sysctl_ctx_entry * sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp) { struct sysctl_ctx_entry *e; SYSCTL_ASSERT_WLOCKED(); if (clist == NULL || oidp == NULL) return(NULL); e = malloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK); e->entry = oidp; TAILQ_INSERT_HEAD(clist, e, link); return (e); } /* Find an entry in the context */ struct sysctl_ctx_entry * sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp) { struct sysctl_ctx_entry *e; SYSCTL_ASSERT_WLOCKED(); if (clist == NULL || oidp == NULL) return(NULL); TAILQ_FOREACH(e, clist, link) { if(e->entry == oidp) return(e); } return (e); } /* * Delete an entry from the context. * NOTE: this function doesn't free oidp! You have to remove it * with sysctl_remove_oid(). */ int sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp) { struct sysctl_ctx_entry *e; if (clist == NULL || oidp == NULL) return (EINVAL); SYSCTL_WLOCK(); e = sysctl_ctx_entry_find(clist, oidp); if (e != NULL) { TAILQ_REMOVE(clist, e, link); SYSCTL_WUNLOCK(); free(e, M_SYSCTLOID); return (0); } else { SYSCTL_WUNLOCK(); return (ENOENT); } } /* * Remove dynamically created sysctl trees. * oidp - top of the tree to be removed * del - if 0 - just deregister, otherwise free up entries as well * recurse - if != 0 traverse the subtree to be deleted */ int sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse) { int error; SYSCTL_WLOCK(); error = sysctl_remove_oid_locked(oidp, del, recurse); SYSCTL_WUNLOCK(); return (error); } int sysctl_remove_name(struct sysctl_oid *parent, const char *name, int del, int recurse) { struct sysctl_oid *p, *tmp; int error; error = ENOENT; SYSCTL_WLOCK(); SLIST_FOREACH_SAFE(p, SYSCTL_CHILDREN(parent), oid_link, tmp) { if (strcmp(p->oid_name, name) == 0) { error = sysctl_remove_oid_locked(p, del, recurse); break; } } SYSCTL_WUNLOCK(); return (error); } static int sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse) { struct sysctl_oid *p, *tmp; int error; SYSCTL_ASSERT_WLOCKED(); if (oidp == NULL) return(EINVAL); if ((oidp->oid_kind & CTLFLAG_DYN) == 0) { printf("can't remove non-dynamic nodes!\n"); return (EINVAL); } /* * WARNING: normal method to do this should be through * sysctl_ctx_free(). Use recursing as the last resort * method to purge your sysctl tree of leftovers... * However, if some other code still references these nodes, * it will panic. */ if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) { if (oidp->oid_refcnt == 1) { SLIST_FOREACH_SAFE(p, SYSCTL_CHILDREN(oidp), oid_link, tmp) { if (!recurse) { printf("Warning: failed attempt to " "remove oid %s with child %s\n", oidp->oid_name, p->oid_name); return (ENOTEMPTY); } error = sysctl_remove_oid_locked(p, del, recurse); if (error) return (error); } } } if (oidp->oid_refcnt > 1 ) { oidp->oid_refcnt--; } else { if (oidp->oid_refcnt == 0) { printf("Warning: bad oid_refcnt=%u (%s)!\n", oidp->oid_refcnt, oidp->oid_name); return (EINVAL); } sysctl_unregister_oid(oidp); if (del) { /* * Wait for all threads running the handler to drain. * This preserves the previous behavior when the * sysctl lock was held across a handler invocation, * and is necessary for module unload correctness. */ while (oidp->oid_running > 0) { oidp->oid_kind |= CTLFLAG_DYING; SYSCTL_SLEEP(&oidp->oid_running, "oidrm", 0); } if (oidp->oid_descr) free(__DECONST(char *, oidp->oid_descr), M_SYSCTLOID); free(__DECONST(char *, oidp->oid_name), M_SYSCTLOID); free(oidp, M_SYSCTLOID); } } return (0); } /* * Create new sysctls at run time. * clist may point to a valid context initialized with sysctl_ctx_init(). */ struct sysctl_oid * sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent, int number, const char *name, int kind, void *arg1, intmax_t arg2, int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr) { struct sysctl_oid *oidp; /* You have to hook up somewhere.. */ if (parent == NULL) return(NULL); /* Check if the node already exists, otherwise create it */ SYSCTL_WLOCK(); oidp = sysctl_find_oidname(name, parent); if (oidp != NULL) { if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) { oidp->oid_refcnt++; /* Update the context */ if (clist != NULL) sysctl_ctx_entry_add(clist, oidp); SYSCTL_WUNLOCK(); return (oidp); } else { SYSCTL_WUNLOCK(); printf("can't re-use a leaf (%s)!\n", name); return (NULL); } } oidp = malloc(sizeof(struct sysctl_oid), M_SYSCTLOID, M_WAITOK|M_ZERO); oidp->oid_parent = parent; SLIST_INIT(&oidp->oid_children); oidp->oid_number = number; oidp->oid_refcnt = 1; oidp->oid_name = strdup(name, M_SYSCTLOID); oidp->oid_handler = handler; oidp->oid_kind = CTLFLAG_DYN | kind; oidp->oid_arg1 = arg1; oidp->oid_arg2 = arg2; oidp->oid_fmt = fmt; if (descr != NULL) oidp->oid_descr = strdup(descr, M_SYSCTLOID); /* Update the context, if used */ if (clist != NULL) sysctl_ctx_entry_add(clist, oidp); /* Register this oid */ sysctl_register_oid(oidp); SYSCTL_WUNLOCK(); return (oidp); } /* * Rename an existing oid. */ void sysctl_rename_oid(struct sysctl_oid *oidp, const char *name) { char *newname; char *oldname; newname = strdup(name, M_SYSCTLOID); SYSCTL_WLOCK(); oldname = __DECONST(char *, oidp->oid_name); oidp->oid_name = newname; SYSCTL_WUNLOCK(); free(oldname, M_SYSCTLOID); } /* * Reparent an existing oid. */ int sysctl_move_oid(struct sysctl_oid *oid, struct sysctl_oid_list *parent) { struct sysctl_oid *oidp; SYSCTL_WLOCK(); if (oid->oid_parent == parent) { SYSCTL_WUNLOCK(); return (0); } oidp = sysctl_find_oidname(oid->oid_name, parent); if (oidp != NULL) { SYSCTL_WUNLOCK(); return (EEXIST); } sysctl_unregister_oid(oid); oid->oid_parent = parent; oid->oid_number = OID_AUTO; sysctl_register_oid(oid); SYSCTL_WUNLOCK(); return (0); } /* * Register the kernel's oids on startup. */ SET_DECLARE(sysctl_set, struct sysctl_oid); static void sysctl_register_all(void *arg) { struct sysctl_oid **oidp; sx_init(&sysctlmemlock, "sysctl mem"); SYSCTL_INIT(); SYSCTL_WLOCK(); SET_FOREACH(oidp, sysctl_set) sysctl_register_oid(*oidp); SYSCTL_WUNLOCK(); } SYSINIT(sysctl, SI_SUB_KMEM, SI_ORDER_FIRST, sysctl_register_all, 0); /* * "Staff-functions" * * These functions implement a presently undocumented interface * used by the sysctl program to walk the tree, and get the type * so it can print the value. * This interface is under work and consideration, and should probably * be killed with a big axe by the first person who can find the time. * (be aware though, that the proper interface isn't as obvious as it * may seem, there are various conflicting requirements. * * {0,0} printf the entire MIB-tree. * {0,1,...} return the name of the "..." OID. * {0,2,...} return the next OID. * {0,3} return the OID of the name in "new" * {0,4,...} return the kind & format info for the "..." OID. * {0,5,...} return the description the "..." OID. */ #ifdef SYSCTL_DEBUG static void sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i) { int k; struct sysctl_oid *oidp; SYSCTL_ASSERT_LOCKED(); SLIST_FOREACH(oidp, l, oid_link) { for (k=0; koid_number, oidp->oid_name); printf("%c%c", oidp->oid_kind & CTLFLAG_RD ? 'R':' ', oidp->oid_kind & CTLFLAG_WR ? 'W':' '); if (oidp->oid_handler) printf(" *Handler"); switch (oidp->oid_kind & CTLTYPE) { case CTLTYPE_NODE: printf(" Node\n"); if (!oidp->oid_handler) { sysctl_sysctl_debug_dump_node( SYSCTL_CHILDREN(oidp), i + 2); } break; case CTLTYPE_INT: printf(" Int\n"); break; case CTLTYPE_UINT: printf(" u_int\n"); break; case CTLTYPE_LONG: printf(" Long\n"); break; case CTLTYPE_ULONG: printf(" u_long\n"); break; case CTLTYPE_STRING: printf(" String\n"); break; case CTLTYPE_S8: printf(" int8_t\n"); break; case CTLTYPE_S16: printf(" int16_t\n"); break; case CTLTYPE_S32: printf(" int32_t\n"); break; case CTLTYPE_S64: printf(" int64_t\n"); break; case CTLTYPE_U8: printf(" uint8_t\n"); break; case CTLTYPE_U16: printf(" uint16_t\n"); break; case CTLTYPE_U32: printf(" uint32_t\n"); break; case CTLTYPE_U64: printf(" uint64_t\n"); break; case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break; default: printf("\n"); } } } static int sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS) { struct rm_priotracker tracker; int error; error = priv_check(req->td, PRIV_SYSCTL_DEBUG); if (error) return (error); SYSCTL_RLOCK(&tracker); sysctl_sysctl_debug_dump_node(&sysctl__children, 0); SYSCTL_RUNLOCK(&tracker); return (ENOENT); } SYSCTL_PROC(_sysctl, 0, debug, CTLTYPE_STRING|CTLFLAG_RD|CTLFLAG_MPSAFE, 0, 0, sysctl_sysctl_debug, "-", ""); #endif static int sysctl_sysctl_name(SYSCTL_HANDLER_ARGS) { int *name = (int *) arg1; u_int namelen = arg2; int error = 0; struct sysctl_oid *oid; struct sysctl_oid_list *lsp = &sysctl__children, *lsp2; struct rm_priotracker tracker; char buf[10]; SYSCTL_RLOCK(&tracker); while (namelen) { if (!lsp) { snprintf(buf,sizeof(buf),"%d",*name); if (req->oldidx) error = SYSCTL_OUT(req, ".", 1); if (!error) error = SYSCTL_OUT(req, buf, strlen(buf)); if (error) goto out; namelen--; name++; continue; } lsp2 = NULL; SLIST_FOREACH(oid, lsp, oid_link) { if (oid->oid_number != *name) continue; if (req->oldidx) error = SYSCTL_OUT(req, ".", 1); if (!error) error = SYSCTL_OUT(req, oid->oid_name, strlen(oid->oid_name)); if (error) goto out; namelen--; name++; if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE) break; if (oid->oid_handler) break; lsp2 = SYSCTL_CHILDREN(oid); break; } lsp = lsp2; } error = SYSCTL_OUT(req, "", 1); out: SYSCTL_RUNLOCK(&tracker); return (error); } /* * XXXRW/JA: Shouldn't return name data for nodes that we don't permit in * capability mode. */ static SYSCTL_NODE(_sysctl, 1, name, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_name, ""); static int sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen, int *next, int *len, int level, struct sysctl_oid **oidpp) { struct sysctl_oid *oidp; SYSCTL_ASSERT_LOCKED(); *len = level; SLIST_FOREACH(oidp, lsp, oid_link) { *next = oidp->oid_number; *oidpp = oidp; if (oidp->oid_kind & CTLFLAG_SKIP) continue; if (!namelen) { if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE) return (0); if (oidp->oid_handler) /* We really should call the handler here...*/ return (0); lsp = SYSCTL_CHILDREN(oidp); if (!sysctl_sysctl_next_ls(lsp, 0, 0, next+1, len, level+1, oidpp)) return (0); goto emptynode; } if (oidp->oid_number < *name) continue; if (oidp->oid_number > *name) { if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE) return (0); if (oidp->oid_handler) return (0); lsp = SYSCTL_CHILDREN(oidp); if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1, len, level+1, oidpp)) return (0); goto next; } if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE) continue; if (oidp->oid_handler) continue; lsp = SYSCTL_CHILDREN(oidp); if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1, len, level+1, oidpp)) return (0); next: namelen = 1; emptynode: *len = level; } return (1); } static int sysctl_sysctl_next(SYSCTL_HANDLER_ARGS) { int *name = (int *) arg1; u_int namelen = arg2; int i, j, error; struct sysctl_oid *oid; struct sysctl_oid_list *lsp = &sysctl__children; struct rm_priotracker tracker; int newoid[CTL_MAXNAME]; SYSCTL_RLOCK(&tracker); i = sysctl_sysctl_next_ls(lsp, name, namelen, newoid, &j, 1, &oid); SYSCTL_RUNLOCK(&tracker); if (i) return (ENOENT); error = SYSCTL_OUT(req, newoid, j * sizeof (int)); return (error); } /* * XXXRW/JA: Shouldn't return next data for nodes that we don't permit in * capability mode. */ static SYSCTL_NODE(_sysctl, 2, next, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_next, ""); static int name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp) { struct sysctl_oid *oidp; struct sysctl_oid_list *lsp = &sysctl__children; char *p; SYSCTL_ASSERT_LOCKED(); for (*len = 0; *len < CTL_MAXNAME;) { p = strsep(&name, "."); oidp = SLIST_FIRST(lsp); for (;; oidp = SLIST_NEXT(oidp, oid_link)) { if (oidp == NULL) return (ENOENT); if (strcmp(p, oidp->oid_name) == 0) break; } *oid++ = oidp->oid_number; (*len)++; if (name == NULL || *name == '\0') { if (oidpp) *oidpp = oidp; return (0); } if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE) break; if (oidp->oid_handler) break; lsp = SYSCTL_CHILDREN(oidp); } return (ENOENT); } static int sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS) { char *p; int error, oid[CTL_MAXNAME], len = 0; struct sysctl_oid *op = NULL; struct rm_priotracker tracker; if (!req->newlen) return (ENOENT); if (req->newlen >= MAXPATHLEN) /* XXX arbitrary, undocumented */ return (ENAMETOOLONG); p = malloc(req->newlen+1, M_SYSCTL, M_WAITOK); error = SYSCTL_IN(req, p, req->newlen); if (error) { free(p, M_SYSCTL); return (error); } p [req->newlen] = '\0'; SYSCTL_RLOCK(&tracker); error = name2oid(p, oid, &len, &op); SYSCTL_RUNLOCK(&tracker); free(p, M_SYSCTL); if (error) return (error); error = SYSCTL_OUT(req, oid, len * sizeof *oid); return (error); } /* * XXXRW/JA: Shouldn't return name2oid data for nodes that we don't permit in * capability mode. */ SYSCTL_PROC(_sysctl, 3, name2oid, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MPSAFE | CTLFLAG_CAPRW, 0, 0, sysctl_sysctl_name2oid, "I", ""); static int sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS) { struct sysctl_oid *oid; struct rm_priotracker tracker; int error; SYSCTL_RLOCK(&tracker); error = sysctl_find_oid(arg1, arg2, &oid, NULL, req); if (error) goto out; if (oid->oid_fmt == NULL) { error = ENOENT; goto out; } error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind)); if (error) goto out; error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1); out: SYSCTL_RUNLOCK(&tracker); return (error); } static SYSCTL_NODE(_sysctl, 4, oidfmt, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD, sysctl_sysctl_oidfmt, ""); static int sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS) { struct sysctl_oid *oid; struct rm_priotracker tracker; int error; SYSCTL_RLOCK(&tracker); error = sysctl_find_oid(arg1, arg2, &oid, NULL, req); if (error) goto out; if (oid->oid_descr == NULL) { error = ENOENT; goto out; } error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1); out: SYSCTL_RUNLOCK(&tracker); return (error); } static SYSCTL_NODE(_sysctl, 5, oiddescr, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD, sysctl_sysctl_oiddescr, ""); /* * Default "handler" functions. */ /* * Handle a bool. * Two cases: * a variable: point arg1 at it. * a constant: pass it in arg2. */ int sysctl_handle_bool(SYSCTL_HANDLER_ARGS) { uint8_t temp; int error; /* * Attempt to get a coherent snapshot by making a copy of the data. */ if (arg1) temp = *(bool *)arg1 ? 1 : 0; else temp = arg2 ? 1 : 0; error = SYSCTL_OUT(req, &temp, sizeof(temp)); if (error || !req->newptr) return (error); if (!arg1) error = EPERM; else { error = SYSCTL_IN(req, &temp, sizeof(temp)); if (!error) *(bool *)arg1 = temp ? 1 : 0; } return (error); } /* * Handle an int8_t, signed or unsigned. * Two cases: * a variable: point arg1 at it. * a constant: pass it in arg2. */ int sysctl_handle_8(SYSCTL_HANDLER_ARGS) { int8_t tmpout; int error = 0; /* * Attempt to get a coherent snapshot by making a copy of the data. */ if (arg1) tmpout = *(int8_t *)arg1; else tmpout = arg2; error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout)); if (error || !req->newptr) return (error); if (!arg1) error = EPERM; else error = SYSCTL_IN(req, arg1, sizeof(tmpout)); return (error); } /* * Handle an int16_t, signed or unsigned. * Two cases: * a variable: point arg1 at it. * a constant: pass it in arg2. */ int sysctl_handle_16(SYSCTL_HANDLER_ARGS) { int16_t tmpout; int error = 0; /* * Attempt to get a coherent snapshot by making a copy of the data. */ if (arg1) tmpout = *(int16_t *)arg1; else tmpout = arg2; error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout)); if (error || !req->newptr) return (error); if (!arg1) error = EPERM; else error = SYSCTL_IN(req, arg1, sizeof(tmpout)); return (error); } /* * Handle an int32_t, signed or unsigned. * Two cases: * a variable: point arg1 at it. * a constant: pass it in arg2. */ int sysctl_handle_32(SYSCTL_HANDLER_ARGS) { int32_t tmpout; int error = 0; /* * Attempt to get a coherent snapshot by making a copy of the data. */ if (arg1) tmpout = *(int32_t *)arg1; else tmpout = arg2; error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout)); if (error || !req->newptr) return (error); if (!arg1) error = EPERM; else error = SYSCTL_IN(req, arg1, sizeof(tmpout)); return (error); } /* * Handle an int, signed or unsigned. * Two cases: * a variable: point arg1 at it. * a constant: pass it in arg2. */ int sysctl_handle_int(SYSCTL_HANDLER_ARGS) { int tmpout, error = 0; /* * Attempt to get a coherent snapshot by making a copy of the data. */ if (arg1) tmpout = *(int *)arg1; else tmpout = arg2; error = SYSCTL_OUT(req, &tmpout, sizeof(int)); if (error || !req->newptr) return (error); if (!arg1) error = EPERM; else error = SYSCTL_IN(req, arg1, sizeof(int)); return (error); } /* * Based on on sysctl_handle_int() convert milliseconds into ticks. * Note: this is used by TCP. */ int sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS) { int error, s, tt; tt = *(int *)arg1; s = (int)((int64_t)tt * 1000 / hz); error = sysctl_handle_int(oidp, &s, 0, req); if (error || !req->newptr) return (error); tt = (int)((int64_t)s * hz / 1000); if (tt < 1) return (EINVAL); *(int *)arg1 = tt; return (0); } /* * Handle a long, signed or unsigned. * Two cases: * a variable: point arg1 at it. * a constant: pass it in arg2. */ int sysctl_handle_long(SYSCTL_HANDLER_ARGS) { int error = 0; long tmplong; #ifdef SCTL_MASK32 int tmpint; #endif /* * Attempt to get a coherent snapshot by making a copy of the data. */ if (arg1) tmplong = *(long *)arg1; else tmplong = arg2; #ifdef SCTL_MASK32 if (req->flags & SCTL_MASK32) { tmpint = tmplong; error = SYSCTL_OUT(req, &tmpint, sizeof(int)); } else #endif error = SYSCTL_OUT(req, &tmplong, sizeof(long)); if (error || !req->newptr) return (error); if (!arg1) error = EPERM; #ifdef SCTL_MASK32 else if (req->flags & SCTL_MASK32) { error = SYSCTL_IN(req, &tmpint, sizeof(int)); *(long *)arg1 = (long)tmpint; } #endif else error = SYSCTL_IN(req, arg1, sizeof(long)); return (error); } /* * Handle a 64 bit int, signed or unsigned. * Two cases: * a variable: point arg1 at it. * a constant: pass it in arg2. */ int sysctl_handle_64(SYSCTL_HANDLER_ARGS) { int error = 0; uint64_t tmpout; /* * Attempt to get a coherent snapshot by making a copy of the data. */ if (arg1) tmpout = *(uint64_t *)arg1; else tmpout = arg2; error = SYSCTL_OUT(req, &tmpout, sizeof(uint64_t)); if (error || !req->newptr) return (error); if (!arg1) error = EPERM; else error = SYSCTL_IN(req, arg1, sizeof(uint64_t)); return (error); } /* * Handle our generic '\0' terminated 'C' string. * Two cases: * a variable string: point arg1 at it, arg2 is max length. * a constant string: point arg1 at it, arg2 is zero. */ int sysctl_handle_string(SYSCTL_HANDLER_ARGS) { size_t outlen; int error = 0, ro_string = 0; /* * A zero-length buffer indicates a fixed size read-only * string: */ if (arg2 == 0) { arg2 = strlen((char *)arg1) + 1; ro_string = 1; } if (req->oldptr != NULL) { char *tmparg; if (ro_string) { tmparg = arg1; } else { /* try to make a coherent snapshot of the string */ tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK); memcpy(tmparg, arg1, arg2); } outlen = strnlen(tmparg, arg2 - 1) + 1; error = SYSCTL_OUT(req, tmparg, outlen); if (!ro_string) free(tmparg, M_SYSCTLTMP); } else { outlen = strnlen((char *)arg1, arg2 - 1) + 1; error = SYSCTL_OUT(req, NULL, outlen); } if (error || !req->newptr) return (error); if ((req->newlen - req->newidx) >= arg2) { error = EINVAL; } else { arg2 = (req->newlen - req->newidx); error = SYSCTL_IN(req, arg1, arg2); ((char *)arg1)[arg2] = '\0'; } return (error); } /* * Handle any kind of opaque data. * arg1 points to it, arg2 is the size. */ int sysctl_handle_opaque(SYSCTL_HANDLER_ARGS) { int error, tries; u_int generation; struct sysctl_req req2; /* * Attempt to get a coherent snapshot, by using the thread * pre-emption counter updated from within mi_switch() to * determine if we were pre-empted during a bcopy() or * copyout(). Make 3 attempts at doing this before giving up. * If we encounter an error, stop immediately. */ tries = 0; req2 = *req; retry: generation = curthread->td_generation; error = SYSCTL_OUT(req, arg1, arg2); if (error) return (error); tries++; if (generation != curthread->td_generation && tries < 3) { *req = req2; goto retry; } error = SYSCTL_IN(req, arg1, arg2); return (error); } /* * Transfer functions to/from kernel space. * XXX: rather untested at this point */ static int sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l) { size_t i = 0; if (req->oldptr) { i = l; if (req->oldlen <= req->oldidx) i = 0; else if (i > req->oldlen - req->oldidx) i = req->oldlen - req->oldidx; if (i > 0) bcopy(p, (char *)req->oldptr + req->oldidx, i); } req->oldidx += l; if (req->oldptr && i != l) return (ENOMEM); return (0); } static int sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l) { if (!req->newptr) return (0); if (req->newlen - req->newidx < l) return (EINVAL); bcopy((char *)req->newptr + req->newidx, p, l); req->newidx += l; return (0); } int kernel_sysctl(struct thread *td, int *name, u_int namelen, void *old, size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags) { int error = 0; struct sysctl_req req; bzero(&req, sizeof req); req.td = td; req.flags = flags; if (oldlenp) { req.oldlen = *oldlenp; } req.validlen = req.oldlen; if (old) { req.oldptr= old; } if (new != NULL) { req.newlen = newlen; req.newptr = new; } req.oldfunc = sysctl_old_kernel; req.newfunc = sysctl_new_kernel; req.lock = REQ_UNWIRED; error = sysctl_root(0, name, namelen, &req); if (req.lock == REQ_WIRED && req.validlen > 0) vsunlock(req.oldptr, req.validlen); if (error && error != ENOMEM) return (error); if (retval) { if (req.oldptr && req.oldidx > req.validlen) *retval = req.validlen; else *retval = req.oldidx; } return (error); } int kernel_sysctlbyname(struct thread *td, char *name, void *old, size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags) { int oid[CTL_MAXNAME]; size_t oidlen, plen; int error; oid[0] = 0; /* sysctl internal magic */ oid[1] = 3; /* name2oid */ oidlen = sizeof(oid); error = kernel_sysctl(td, oid, 2, oid, &oidlen, (void *)name, strlen(name), &plen, flags); if (error) return (error); error = kernel_sysctl(td, oid, plen / sizeof(int), old, oldlenp, new, newlen, retval, flags); return (error); } /* * Transfer function to/from user space. */ static int sysctl_old_user(struct sysctl_req *req, const void *p, size_t l) { size_t i, len, origidx; int error; origidx = req->oldidx; req->oldidx += l; if (req->oldptr == NULL) return (0); /* * If we have not wired the user supplied buffer and we are currently * holding locks, drop a witness warning, as it's possible that * write operations to the user page can sleep. */ if (req->lock != REQ_WIRED) WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "sysctl_old_user()"); i = l; len = req->validlen; if (len <= origidx) i = 0; else { if (i > len - origidx) i = len - origidx; if (req->lock == REQ_WIRED) { error = copyout_nofault(p, (char *)req->oldptr + origidx, i); } else error = copyout(p, (char *)req->oldptr + origidx, i); if (error != 0) return (error); } if (i < l) return (ENOMEM); return (0); } static int sysctl_new_user(struct sysctl_req *req, void *p, size_t l) { int error; if (!req->newptr) return (0); if (req->newlen - req->newidx < l) return (EINVAL); WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "sysctl_new_user()"); error = copyin((char *)req->newptr + req->newidx, p, l); req->newidx += l; return (error); } /* * Wire the user space destination buffer. If set to a value greater than * zero, the len parameter limits the maximum amount of wired memory. */ int sysctl_wire_old_buffer(struct sysctl_req *req, size_t len) { int ret; size_t wiredlen; wiredlen = (len > 0 && len < req->oldlen) ? len : req->oldlen; ret = 0; if (req->lock != REQ_WIRED && req->oldptr && req->oldfunc == sysctl_old_user) { if (wiredlen != 0) { ret = vslock(req->oldptr, wiredlen); if (ret != 0) { if (ret != ENOMEM) return (ret); wiredlen = 0; } } req->lock = REQ_WIRED; req->validlen = wiredlen; } return (0); } int sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid, int *nindx, struct sysctl_req *req) { struct sysctl_oid_list *lsp; struct sysctl_oid *oid; int indx; SYSCTL_ASSERT_LOCKED(); lsp = &sysctl__children; indx = 0; while (indx < CTL_MAXNAME) { SLIST_FOREACH(oid, lsp, oid_link) { if (oid->oid_number == name[indx]) break; } if (oid == NULL) return (ENOENT); indx++; if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) { if (oid->oid_handler != NULL || indx == namelen) { *noid = oid; if (nindx != NULL) *nindx = indx; KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0, ("%s found DYING node %p", __func__, oid)); return (0); } lsp = SYSCTL_CHILDREN(oid); } else if (indx == namelen) { *noid = oid; if (nindx != NULL) *nindx = indx; KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0, ("%s found DYING node %p", __func__, oid)); return (0); } else { return (ENOTDIR); } } return (ENOENT); } /* * Traverse our tree, and find the right node, execute whatever it points * to, and return the resulting error code. */ static int sysctl_root(SYSCTL_HANDLER_ARGS) { struct sysctl_oid *oid; struct rm_priotracker tracker; int error, indx, lvl; SYSCTL_RLOCK(&tracker); error = sysctl_find_oid(arg1, arg2, &oid, &indx, req); if (error) goto out; if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) { /* * You can't call a sysctl when it's a node, but has * no handler. Inform the user that it's a node. * The indx may or may not be the same as namelen. */ if (oid->oid_handler == NULL) { error = EISDIR; goto out; } } /* Is this sysctl writable? */ if (req->newptr && !(oid->oid_kind & CTLFLAG_WR)) { error = EPERM; goto out; } KASSERT(req->td != NULL, ("sysctl_root(): req->td == NULL")); #ifdef CAPABILITY_MODE /* * If the process is in capability mode, then don't permit reading or * writing unless specifically granted for the node. */ if (IN_CAPABILITY_MODE(req->td)) { if ((req->oldptr && !(oid->oid_kind & CTLFLAG_CAPRD)) || (req->newptr && !(oid->oid_kind & CTLFLAG_CAPWR))) { error = EPERM; goto out; } } #endif /* Is this sysctl sensitive to securelevels? */ if (req->newptr && (oid->oid_kind & CTLFLAG_SECURE)) { lvl = (oid->oid_kind & CTLMASK_SECURE) >> CTLSHIFT_SECURE; error = securelevel_gt(req->td->td_ucred, lvl); if (error) goto out; } /* Is this sysctl writable by only privileged users? */ if (req->newptr && !(oid->oid_kind & CTLFLAG_ANYBODY)) { int priv; if (oid->oid_kind & CTLFLAG_PRISON) priv = PRIV_SYSCTL_WRITEJAIL; #ifdef VIMAGE else if ((oid->oid_kind & CTLFLAG_VNET) && prison_owns_vnet(req->td->td_ucred)) priv = PRIV_SYSCTL_WRITEJAIL; #endif else priv = PRIV_SYSCTL_WRITE; error = priv_check(req->td, priv); if (error) goto out; } if (!oid->oid_handler) { error = EINVAL; goto out; } if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) { arg1 = (int *)arg1 + indx; arg2 -= indx; } else { arg1 = oid->oid_arg1; arg2 = oid->oid_arg2; } #ifdef MAC error = mac_system_check_sysctl(req->td->td_ucred, oid, arg1, arg2, req); if (error != 0) goto out; #endif #ifdef VIMAGE if ((oid->oid_kind & CTLFLAG_VNET) && arg1 != NULL) arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1); #endif error = sysctl_root_handler_locked(oid, arg1, arg2, req, &tracker); out: SYSCTL_RUNLOCK(&tracker); return (error); } #ifndef _SYS_SYSPROTO_H_ struct sysctl_args { int *name; u_int namelen; void *old; size_t *oldlenp; void *new; size_t newlen; }; #endif int sys___sysctl(struct thread *td, struct sysctl_args *uap) { int error, i, name[CTL_MAXNAME]; size_t j; if (uap->namelen > CTL_MAXNAME || uap->namelen < 2) return (EINVAL); error = copyin(uap->name, &name, uap->namelen * sizeof(int)); if (error) return (error); error = userland_sysctl(td, name, uap->namelen, uap->old, uap->oldlenp, 0, uap->new, uap->newlen, &j, 0); if (error && error != ENOMEM) return (error); if (uap->oldlenp) { i = copyout(&j, uap->oldlenp, sizeof(j)); if (i) return (i); } return (error); } /* * This is used from various compatibility syscalls too. That's why name * must be in kernel space. */ int userland_sysctl(struct thread *td, int *name, u_int namelen, void *old, size_t *oldlenp, int inkernel, void *new, size_t newlen, size_t *retval, int flags) { int error = 0, memlocked; struct sysctl_req req; bzero(&req, sizeof req); req.td = td; req.flags = flags; if (oldlenp) { if (inkernel) { req.oldlen = *oldlenp; } else { error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp)); if (error) return (error); } } req.validlen = req.oldlen; if (old) { if (!useracc(old, req.oldlen, VM_PROT_WRITE)) return (EFAULT); req.oldptr= old; } if (new != NULL) { if (!useracc(new, newlen, VM_PROT_READ)) return (EFAULT); req.newlen = newlen; req.newptr = new; } req.oldfunc = sysctl_old_user; req.newfunc = sysctl_new_user; req.lock = REQ_UNWIRED; #ifdef KTRACE if (KTRPOINT(curthread, KTR_SYSCTL)) ktrsysctl(name, namelen); #endif if (req.oldptr && req.oldlen > PAGE_SIZE) { memlocked = 1; sx_xlock(&sysctlmemlock); } else memlocked = 0; CURVNET_SET(TD_TO_VNET(td)); for (;;) { req.oldidx = 0; req.newidx = 0; error = sysctl_root(0, name, namelen, &req); if (error != EAGAIN) break; kern_yield(PRI_USER); } CURVNET_RESTORE(); if (req.lock == REQ_WIRED && req.validlen > 0) vsunlock(req.oldptr, req.validlen); if (memlocked) sx_xunlock(&sysctlmemlock); if (error && error != ENOMEM) return (error); if (retval) { if (req.oldptr && req.oldidx > req.validlen) *retval = req.validlen; else *retval = req.oldidx; } return (error); } /* * Drain into a sysctl struct. The user buffer should be wired if a page * fault would cause issue. */ static int sbuf_sysctl_drain(void *arg, const char *data, int len) { struct sysctl_req *req = arg; int error; error = SYSCTL_OUT(req, data, len); KASSERT(error >= 0, ("Got unexpected negative value %d", error)); return (error == 0 ? len : -error); } struct sbuf * sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length, struct sysctl_req *req) { /* Supply a default buffer size if none given. */ if (buf == NULL && length == 0) length = 64; s = sbuf_new(s, buf, length, SBUF_FIXEDLEN | SBUF_INCLUDENUL); sbuf_set_drain(s, sbuf_sysctl_drain, req); return (s); } Index: head/sys/kern/kern_time.c =================================================================== --- head/sys/kern/kern_time.c (revision 305831) +++ head/sys/kern/kern_time.c (revision 305832) @@ -1,1673 +1,1673 @@ /*- * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_time.c 8.1 (Berkeley) 6/10/93 */ #include __FBSDID("$FreeBSD$"); #include "opt_ktrace.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KTRACE #include #endif #include #include #define MAX_CLOCKS (CLOCK_MONOTONIC+1) #define CPUCLOCK_BIT 0x80000000 #define CPUCLOCK_PROCESS_BIT 0x40000000 #define CPUCLOCK_ID_MASK (~(CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT)) #define MAKE_THREAD_CPUCLOCK(tid) (CPUCLOCK_BIT|(tid)) #define MAKE_PROCESS_CPUCLOCK(pid) \ (CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT|(pid)) static struct kclock posix_clocks[MAX_CLOCKS]; static uma_zone_t itimer_zone = NULL; /* * Time of day and interval timer support. * * These routines provide the kernel entry points to get and set * the time-of-day and per-process interval timers. Subroutines * here provide support for adding and subtracting timeval structures * and decrementing interval timers, optionally reloading the interval * timers when they expire. */ static int settime(struct thread *, struct timeval *); static void timevalfix(struct timeval *); static void itimer_start(void); static int itimer_init(void *, int, int); static void itimer_fini(void *, int); static void itimer_enter(struct itimer *); static void itimer_leave(struct itimer *); static struct itimer *itimer_find(struct proc *, int); static void itimers_alloc(struct proc *); static void itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp); static void itimers_event_hook_exit(void *arg, struct proc *p); static int realtimer_create(struct itimer *); static int realtimer_gettime(struct itimer *, struct itimerspec *); static int realtimer_settime(struct itimer *, int, struct itimerspec *, struct itimerspec *); static int realtimer_delete(struct itimer *); static void realtimer_clocktime(clockid_t, struct timespec *); static void realtimer_expire(void *); int register_posix_clock(int, struct kclock *); void itimer_fire(struct itimer *it); int itimespecfix(struct timespec *ts); #define CLOCK_CALL(clock, call, arglist) \ ((*posix_clocks[clock].call) arglist) SYSINIT(posix_timer, SI_SUB_P1003_1B, SI_ORDER_FIRST+4, itimer_start, NULL); static int settime(struct thread *td, struct timeval *tv) { struct timeval delta, tv1, tv2; static struct timeval maxtime, laststep; struct timespec ts; microtime(&tv1); delta = *tv; timevalsub(&delta, &tv1); /* * If the system is secure, we do not allow the time to be * set to a value earlier than 1 second less than the highest * time we have yet seen. The worst a miscreant can do in * this circumstance is "freeze" time. He couldn't go * back to the past. * * We similarly do not allow the clock to be stepped more * than one second, nor more than once per second. This allows * a miscreant to make the clock march double-time, but no worse. */ if (securelevel_gt(td->td_ucred, 1) != 0) { if (delta.tv_sec < 0 || delta.tv_usec < 0) { /* * Update maxtime to latest time we've seen. */ if (tv1.tv_sec > maxtime.tv_sec) maxtime = tv1; tv2 = *tv; timevalsub(&tv2, &maxtime); if (tv2.tv_sec < -1) { tv->tv_sec = maxtime.tv_sec - 1; printf("Time adjustment clamped to -1 second\n"); } } else { if (tv1.tv_sec == laststep.tv_sec) return (EPERM); if (delta.tv_sec > 1) { tv->tv_sec = tv1.tv_sec + 1; printf("Time adjustment clamped to +1 second\n"); } laststep = *tv; } } ts.tv_sec = tv->tv_sec; ts.tv_nsec = tv->tv_usec * 1000; tc_setclock(&ts); resettodr(); return (0); } #ifndef _SYS_SYSPROTO_H_ struct clock_getcpuclockid2_args { id_t id; int which, clockid_t *clock_id; }; #endif /* ARGSUSED */ int sys_clock_getcpuclockid2(struct thread *td, struct clock_getcpuclockid2_args *uap) { clockid_t clk_id; int error; error = kern_clock_getcpuclockid2(td, uap->id, uap->which, &clk_id); if (error == 0) error = copyout(&clk_id, uap->clock_id, sizeof(clockid_t)); return (error); } int kern_clock_getcpuclockid2(struct thread *td, id_t id, int which, clockid_t *clk_id) { struct proc *p; pid_t pid; lwpid_t tid; int error; switch (which) { case CPUCLOCK_WHICH_PID: if (id != 0) { error = pget(id, PGET_CANSEE | PGET_NOTID, &p); if (error != 0) return (error); PROC_UNLOCK(p); pid = id; } else { pid = td->td_proc->p_pid; } *clk_id = MAKE_PROCESS_CPUCLOCK(pid); return (0); case CPUCLOCK_WHICH_TID: tid = id == 0 ? td->td_tid : id; *clk_id = MAKE_THREAD_CPUCLOCK(tid); return (0); default: return (EINVAL); } } #ifndef _SYS_SYSPROTO_H_ struct clock_gettime_args { clockid_t clock_id; struct timespec *tp; }; #endif /* ARGSUSED */ int sys_clock_gettime(struct thread *td, struct clock_gettime_args *uap) { struct timespec ats; int error; error = kern_clock_gettime(td, uap->clock_id, &ats); if (error == 0) error = copyout(&ats, uap->tp, sizeof(ats)); return (error); } static inline void cputick2timespec(uint64_t runtime, struct timespec *ats) { runtime = cputick2usec(runtime); ats->tv_sec = runtime / 1000000; ats->tv_nsec = runtime % 1000000 * 1000; } static void get_thread_cputime(struct thread *targettd, struct timespec *ats) { uint64_t runtime, curtime, switchtime; if (targettd == NULL) { /* current thread */ critical_enter(); switchtime = PCPU_GET(switchtime); curtime = cpu_ticks(); runtime = curthread->td_runtime; critical_exit(); runtime += curtime - switchtime; } else { thread_lock(targettd); runtime = targettd->td_runtime; thread_unlock(targettd); } cputick2timespec(runtime, ats); } static void get_process_cputime(struct proc *targetp, struct timespec *ats) { uint64_t runtime; struct rusage ru; PROC_STATLOCK(targetp); rufetch(targetp, &ru); runtime = targetp->p_rux.rux_runtime; PROC_STATUNLOCK(targetp); cputick2timespec(runtime, ats); } static int get_cputime(struct thread *td, clockid_t clock_id, struct timespec *ats) { struct proc *p, *p2; struct thread *td2; lwpid_t tid; pid_t pid; int error; p = td->td_proc; if ((clock_id & CPUCLOCK_PROCESS_BIT) == 0) { tid = clock_id & CPUCLOCK_ID_MASK; td2 = tdfind(tid, p->p_pid); if (td2 == NULL) return (EINVAL); get_thread_cputime(td2, ats); PROC_UNLOCK(td2->td_proc); } else { pid = clock_id & CPUCLOCK_ID_MASK; error = pget(pid, PGET_CANSEE, &p2); if (error != 0) return (EINVAL); get_process_cputime(p2, ats); PROC_UNLOCK(p2); } return (0); } int kern_clock_gettime(struct thread *td, clockid_t clock_id, struct timespec *ats) { struct timeval sys, user; struct proc *p; p = td->td_proc; switch (clock_id) { case CLOCK_REALTIME: /* Default to precise. */ case CLOCK_REALTIME_PRECISE: nanotime(ats); break; case CLOCK_REALTIME_FAST: getnanotime(ats); break; case CLOCK_VIRTUAL: PROC_LOCK(p); PROC_STATLOCK(p); calcru(p, &user, &sys); PROC_STATUNLOCK(p); PROC_UNLOCK(p); TIMEVAL_TO_TIMESPEC(&user, ats); break; case CLOCK_PROF: PROC_LOCK(p); PROC_STATLOCK(p); calcru(p, &user, &sys); PROC_STATUNLOCK(p); PROC_UNLOCK(p); timevaladd(&user, &sys); TIMEVAL_TO_TIMESPEC(&user, ats); break; case CLOCK_MONOTONIC: /* Default to precise. */ case CLOCK_MONOTONIC_PRECISE: case CLOCK_UPTIME: case CLOCK_UPTIME_PRECISE: nanouptime(ats); break; case CLOCK_UPTIME_FAST: case CLOCK_MONOTONIC_FAST: getnanouptime(ats); break; case CLOCK_SECOND: ats->tv_sec = time_second; ats->tv_nsec = 0; break; case CLOCK_THREAD_CPUTIME_ID: get_thread_cputime(NULL, ats); break; case CLOCK_PROCESS_CPUTIME_ID: PROC_LOCK(p); get_process_cputime(p, ats); PROC_UNLOCK(p); break; default: if ((int)clock_id >= 0) return (EINVAL); return (get_cputime(td, clock_id, ats)); } return (0); } #ifndef _SYS_SYSPROTO_H_ struct clock_settime_args { clockid_t clock_id; const struct timespec *tp; }; #endif /* ARGSUSED */ int sys_clock_settime(struct thread *td, struct clock_settime_args *uap) { struct timespec ats; int error; if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0) return (error); return (kern_clock_settime(td, uap->clock_id, &ats)); } int kern_clock_settime(struct thread *td, clockid_t clock_id, struct timespec *ats) { struct timeval atv; int error; if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0) return (error); if (clock_id != CLOCK_REALTIME) return (EINVAL); if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000 || ats->tv_sec < 0) return (EINVAL); /* XXX Don't convert nsec->usec and back */ TIMESPEC_TO_TIMEVAL(&atv, ats); error = settime(td, &atv); return (error); } #ifndef _SYS_SYSPROTO_H_ struct clock_getres_args { clockid_t clock_id; struct timespec *tp; }; #endif int sys_clock_getres(struct thread *td, struct clock_getres_args *uap) { struct timespec ts; int error; if (uap->tp == NULL) return (0); error = kern_clock_getres(td, uap->clock_id, &ts); if (error == 0) error = copyout(&ts, uap->tp, sizeof(ts)); return (error); } int kern_clock_getres(struct thread *td, clockid_t clock_id, struct timespec *ts) { ts->tv_sec = 0; switch (clock_id) { case CLOCK_REALTIME: case CLOCK_REALTIME_FAST: case CLOCK_REALTIME_PRECISE: case CLOCK_MONOTONIC: case CLOCK_MONOTONIC_FAST: case CLOCK_MONOTONIC_PRECISE: case CLOCK_UPTIME: case CLOCK_UPTIME_FAST: case CLOCK_UPTIME_PRECISE: /* * Round up the result of the division cheaply by adding 1. * Rounding up is especially important if rounding down * would give 0. Perfect rounding is unimportant. */ ts->tv_nsec = 1000000000 / tc_getfrequency() + 1; break; case CLOCK_VIRTUAL: case CLOCK_PROF: /* Accurately round up here because we can do so cheaply. */ ts->tv_nsec = howmany(1000000000, hz); break; case CLOCK_SECOND: ts->tv_sec = 1; ts->tv_nsec = 0; break; case CLOCK_THREAD_CPUTIME_ID: case CLOCK_PROCESS_CPUTIME_ID: cputime: /* sync with cputick2usec */ ts->tv_nsec = 1000000 / cpu_tickrate(); if (ts->tv_nsec == 0) ts->tv_nsec = 1000; break; default: if ((int)clock_id < 0) goto cputime; return (EINVAL); } return (0); } static uint8_t nanowait[MAXCPU]; int kern_nanosleep(struct thread *td, struct timespec *rqt, struct timespec *rmt) { struct timespec ts; sbintime_t sbt, sbtt, prec, tmp; time_t over; int error; if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000) return (EINVAL); if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0)) return (0); ts = *rqt; if (ts.tv_sec > INT32_MAX / 2) { over = ts.tv_sec - INT32_MAX / 2; ts.tv_sec -= over; } else over = 0; tmp = tstosbt(ts); prec = tmp; prec >>= tc_precexp; if (TIMESEL(&sbt, tmp)) sbt += tc_tick_sbt; sbt += tmp; error = tsleep_sbt(&nanowait[curcpu], PWAIT | PCATCH, "nanslp", sbt, prec, C_ABSOLUTE); if (error != EWOULDBLOCK) { if (error == ERESTART) error = EINTR; TIMESEL(&sbtt, tmp); if (rmt != NULL) { ts = sbttots(sbt - sbtt); ts.tv_sec += over; if (ts.tv_sec < 0) timespecclear(&ts); *rmt = ts; } if (sbtt >= sbt) return (0); return (error); } return (0); } #ifndef _SYS_SYSPROTO_H_ struct nanosleep_args { struct timespec *rqtp; struct timespec *rmtp; }; #endif /* ARGSUSED */ int sys_nanosleep(struct thread *td, struct nanosleep_args *uap) { struct timespec rmt, rqt; int error; error = copyin(uap->rqtp, &rqt, sizeof(rqt)); if (error) return (error); if (uap->rmtp && !useracc((caddr_t)uap->rmtp, sizeof(rmt), VM_PROT_WRITE)) return (EFAULT); error = kern_nanosleep(td, &rqt, &rmt); if (error && uap->rmtp) { int error2; error2 = copyout(&rmt, uap->rmtp, sizeof(rmt)); if (error2) error = error2; } return (error); } #ifndef _SYS_SYSPROTO_H_ struct gettimeofday_args { struct timeval *tp; struct timezone *tzp; }; #endif /* ARGSUSED */ int sys_gettimeofday(struct thread *td, struct gettimeofday_args *uap) { struct timeval atv; struct timezone rtz; int error = 0; if (uap->tp) { microtime(&atv); error = copyout(&atv, uap->tp, sizeof (atv)); } if (error == 0 && uap->tzp != NULL) { rtz.tz_minuteswest = tz_minuteswest; rtz.tz_dsttime = tz_dsttime; error = copyout(&rtz, uap->tzp, sizeof (rtz)); } return (error); } #ifndef _SYS_SYSPROTO_H_ struct settimeofday_args { struct timeval *tv; struct timezone *tzp; }; #endif /* ARGSUSED */ int sys_settimeofday(struct thread *td, struct settimeofday_args *uap) { struct timeval atv, *tvp; struct timezone atz, *tzp; int error; if (uap->tv) { error = copyin(uap->tv, &atv, sizeof(atv)); if (error) return (error); tvp = &atv; } else tvp = NULL; if (uap->tzp) { error = copyin(uap->tzp, &atz, sizeof(atz)); if (error) return (error); tzp = &atz; } else tzp = NULL; return (kern_settimeofday(td, tvp, tzp)); } int kern_settimeofday(struct thread *td, struct timeval *tv, struct timezone *tzp) { int error; error = priv_check(td, PRIV_SETTIMEOFDAY); if (error) return (error); /* Verify all parameters before changing time. */ if (tv) { if (tv->tv_usec < 0 || tv->tv_usec >= 1000000 || tv->tv_sec < 0) return (EINVAL); error = settime(td, tv); } if (tzp && error == 0) { tz_minuteswest = tzp->tz_minuteswest; tz_dsttime = tzp->tz_dsttime; } return (error); } /* * Get value of an interval timer. The process virtual and profiling virtual * time timers are kept in the p_stats area, since they can be swapped out. * These are kept internally in the way they are specified externally: in * time until they expire. * * The real time interval timer is kept in the process table slot for the * process, and its value (it_value) is kept as an absolute time rather than * as a delta, so that it is easy to keep periodic real-time signals from * drifting. * * Virtual time timers are processed in the hardclock() routine of * kern_clock.c. The real time timer is processed by a timeout routine, * called from the softclock() routine. Since a callout may be delayed in * real time due to interrupt processing in the system, it is possible for * the real time timeout routine (realitexpire, given below), to be delayed * in real time past when it is supposed to occur. It does not suffice, * therefore, to reload the real timer .it_value from the real time timers * .it_interval. Rather, we compute the next time in absolute time the timer * should go off. */ #ifndef _SYS_SYSPROTO_H_ struct getitimer_args { u_int which; struct itimerval *itv; }; #endif int sys_getitimer(struct thread *td, struct getitimer_args *uap) { struct itimerval aitv; int error; error = kern_getitimer(td, uap->which, &aitv); if (error != 0) return (error); return (copyout(&aitv, uap->itv, sizeof (struct itimerval))); } int kern_getitimer(struct thread *td, u_int which, struct itimerval *aitv) { struct proc *p = td->td_proc; struct timeval ctv; if (which > ITIMER_PROF) return (EINVAL); if (which == ITIMER_REAL) { /* * Convert from absolute to relative time in .it_value * part of real time timer. If time for real time timer * has passed return 0, else return difference between * current time and time for the timer to go off. */ PROC_LOCK(p); *aitv = p->p_realtimer; PROC_UNLOCK(p); if (timevalisset(&aitv->it_value)) { microuptime(&ctv); if (timevalcmp(&aitv->it_value, &ctv, <)) timevalclear(&aitv->it_value); else timevalsub(&aitv->it_value, &ctv); } } else { PROC_ITIMLOCK(p); *aitv = p->p_stats->p_timer[which]; PROC_ITIMUNLOCK(p); } #ifdef KTRACE if (KTRPOINT(td, KTR_STRUCT)) ktritimerval(aitv); #endif return (0); } #ifndef _SYS_SYSPROTO_H_ struct setitimer_args { u_int which; struct itimerval *itv, *oitv; }; #endif int sys_setitimer(struct thread *td, struct setitimer_args *uap) { struct itimerval aitv, oitv; int error; if (uap->itv == NULL) { uap->itv = uap->oitv; return (sys_getitimer(td, (struct getitimer_args *)uap)); } if ((error = copyin(uap->itv, &aitv, sizeof(struct itimerval)))) return (error); error = kern_setitimer(td, uap->which, &aitv, &oitv); if (error != 0 || uap->oitv == NULL) return (error); return (copyout(&oitv, uap->oitv, sizeof(struct itimerval))); } int kern_setitimer(struct thread *td, u_int which, struct itimerval *aitv, struct itimerval *oitv) { struct proc *p = td->td_proc; struct timeval ctv; sbintime_t sbt, pr; if (aitv == NULL) return (kern_getitimer(td, which, oitv)); if (which > ITIMER_PROF) return (EINVAL); #ifdef KTRACE if (KTRPOINT(td, KTR_STRUCT)) ktritimerval(aitv); #endif if (itimerfix(&aitv->it_value) || aitv->it_value.tv_sec > INT32_MAX / 2) return (EINVAL); if (!timevalisset(&aitv->it_value)) timevalclear(&aitv->it_interval); else if (itimerfix(&aitv->it_interval) || aitv->it_interval.tv_sec > INT32_MAX / 2) return (EINVAL); if (which == ITIMER_REAL) { PROC_LOCK(p); if (timevalisset(&p->p_realtimer.it_value)) callout_stop(&p->p_itcallout); microuptime(&ctv); if (timevalisset(&aitv->it_value)) { pr = tvtosbt(aitv->it_value) >> tc_precexp; timevaladd(&aitv->it_value, &ctv); sbt = tvtosbt(aitv->it_value); callout_reset_sbt(&p->p_itcallout, sbt, pr, realitexpire, p, C_ABSOLUTE); } *oitv = p->p_realtimer; p->p_realtimer = *aitv; PROC_UNLOCK(p); if (timevalisset(&oitv->it_value)) { if (timevalcmp(&oitv->it_value, &ctv, <)) timevalclear(&oitv->it_value); else timevalsub(&oitv->it_value, &ctv); } } else { if (aitv->it_interval.tv_sec == 0 && aitv->it_interval.tv_usec != 0 && aitv->it_interval.tv_usec < tick) aitv->it_interval.tv_usec = tick; if (aitv->it_value.tv_sec == 0 && aitv->it_value.tv_usec != 0 && aitv->it_value.tv_usec < tick) aitv->it_value.tv_usec = tick; PROC_ITIMLOCK(p); *oitv = p->p_stats->p_timer[which]; p->p_stats->p_timer[which] = *aitv; PROC_ITIMUNLOCK(p); } #ifdef KTRACE if (KTRPOINT(td, KTR_STRUCT)) ktritimerval(oitv); #endif return (0); } /* * Real interval timer expired: * send process whose timer expired an alarm signal. * If time is not set up to reload, then just return. * Else compute next time timer should go off which is > current time. * This is where delay in processing this timeout causes multiple * SIGALRM calls to be compressed into one. * tvtohz() always adds 1 to allow for the time until the next clock * interrupt being strictly less than 1 clock tick, but we don't want * that here since we want to appear to be in sync with the clock * interrupt even when we're delayed. */ void realitexpire(void *arg) { struct proc *p; struct timeval ctv; sbintime_t isbt; p = (struct proc *)arg; kern_psignal(p, SIGALRM); if (!timevalisset(&p->p_realtimer.it_interval)) { timevalclear(&p->p_realtimer.it_value); if (p->p_flag & P_WEXIT) wakeup(&p->p_itcallout); return; } isbt = tvtosbt(p->p_realtimer.it_interval); if (isbt >= sbt_timethreshold) getmicrouptime(&ctv); else microuptime(&ctv); do { timevaladd(&p->p_realtimer.it_value, &p->p_realtimer.it_interval); } while (timevalcmp(&p->p_realtimer.it_value, &ctv, <=)); callout_reset_sbt(&p->p_itcallout, tvtosbt(p->p_realtimer.it_value), isbt >> tc_precexp, realitexpire, p, C_ABSOLUTE); } /* * Check that a proposed value to load into the .it_value or * .it_interval part of an interval timer is acceptable, and * fix it to have at least minimal value (i.e. if it is less * than the resolution of the clock, round it up.) */ int itimerfix(struct timeval *tv) { if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000) return (EINVAL); if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < (u_int)tick / 16) tv->tv_usec = (u_int)tick / 16; return (0); } /* * Decrement an interval timer by a specified number * of microseconds, which must be less than a second, * i.e. < 1000000. If the timer expires, then reload * it. In this case, carry over (usec - old value) to * reduce the value reloaded into the timer so that * the timer does not drift. This routine assumes * that it is called in a context where the timers * on which it is operating cannot change in value. */ int itimerdecr(struct itimerval *itp, int usec) { if (itp->it_value.tv_usec < usec) { if (itp->it_value.tv_sec == 0) { /* expired, and already in next interval */ usec -= itp->it_value.tv_usec; goto expire; } itp->it_value.tv_usec += 1000000; itp->it_value.tv_sec--; } itp->it_value.tv_usec -= usec; usec = 0; if (timevalisset(&itp->it_value)) return (1); /* expired, exactly at end of interval */ expire: if (timevalisset(&itp->it_interval)) { itp->it_value = itp->it_interval; itp->it_value.tv_usec -= usec; if (itp->it_value.tv_usec < 0) { itp->it_value.tv_usec += 1000000; itp->it_value.tv_sec--; } } else itp->it_value.tv_usec = 0; /* sec is already 0 */ return (0); } /* * Add and subtract routines for timevals. * N.B.: subtract routine doesn't deal with * results which are before the beginning, * it just gets very confused in this case. * Caveat emptor. */ void timevaladd(struct timeval *t1, const struct timeval *t2) { t1->tv_sec += t2->tv_sec; t1->tv_usec += t2->tv_usec; timevalfix(t1); } void timevalsub(struct timeval *t1, const struct timeval *t2) { t1->tv_sec -= t2->tv_sec; t1->tv_usec -= t2->tv_usec; timevalfix(t1); } static void timevalfix(struct timeval *t1) { if (t1->tv_usec < 0) { t1->tv_sec--; t1->tv_usec += 1000000; } if (t1->tv_usec >= 1000000) { t1->tv_sec++; t1->tv_usec -= 1000000; } } /* * ratecheck(): simple time-based rate-limit checking. */ int ratecheck(struct timeval *lasttime, const struct timeval *mininterval) { struct timeval tv, delta; int rv = 0; getmicrouptime(&tv); /* NB: 10ms precision */ delta = tv; timevalsub(&delta, lasttime); /* * check for 0,0 is so that the message will be seen at least once, * even if interval is huge. */ if (timevalcmp(&delta, mininterval, >=) || (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) { *lasttime = tv; rv = 1; } return (rv); } /* * ppsratecheck(): packets (or events) per second limitation. * * Return 0 if the limit is to be enforced (e.g. the caller * should drop a packet because of the rate limitation). * * maxpps of 0 always causes zero to be returned. maxpps of -1 * always causes 1 to be returned; this effectively defeats rate * limiting. * * Note that we maintain the struct timeval for compatibility * with other bsd systems. We reuse the storage and just monitor * clock ticks for minimal overhead. */ int ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps) { int now; /* * Reset the last time and counter if this is the first call * or more than a second has passed since the last update of * lasttime. */ now = ticks; if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) { lasttime->tv_sec = now; *curpps = 1; return (maxpps != 0); } else { (*curpps)++; /* NB: ignore potential overflow */ return (maxpps < 0 || *curpps <= maxpps); } } static void itimer_start(void) { struct kclock rt_clock = { .timer_create = realtimer_create, .timer_delete = realtimer_delete, .timer_settime = realtimer_settime, .timer_gettime = realtimer_gettime, .event_hook = NULL }; itimer_zone = uma_zcreate("itimer", sizeof(struct itimer), NULL, NULL, itimer_init, itimer_fini, UMA_ALIGN_PTR, 0); register_posix_clock(CLOCK_REALTIME, &rt_clock); register_posix_clock(CLOCK_MONOTONIC, &rt_clock); p31b_setcfg(CTL_P1003_1B_TIMERS, 200112L); p31b_setcfg(CTL_P1003_1B_DELAYTIMER_MAX, INT_MAX); p31b_setcfg(CTL_P1003_1B_TIMER_MAX, TIMER_MAX); EVENTHANDLER_REGISTER(process_exit, itimers_event_hook_exit, (void *)ITIMER_EV_EXIT, EVENTHANDLER_PRI_ANY); EVENTHANDLER_REGISTER(process_exec, itimers_event_hook_exec, (void *)ITIMER_EV_EXEC, EVENTHANDLER_PRI_ANY); } int register_posix_clock(int clockid, struct kclock *clk) { if ((unsigned)clockid >= MAX_CLOCKS) { printf("%s: invalid clockid\n", __func__); return (0); } posix_clocks[clockid] = *clk; return (1); } static int itimer_init(void *mem, int size, int flags) { struct itimer *it; it = (struct itimer *)mem; mtx_init(&it->it_mtx, "itimer lock", NULL, MTX_DEF); return (0); } static void itimer_fini(void *mem, int size) { struct itimer *it; it = (struct itimer *)mem; mtx_destroy(&it->it_mtx); } static void itimer_enter(struct itimer *it) { mtx_assert(&it->it_mtx, MA_OWNED); it->it_usecount++; } static void itimer_leave(struct itimer *it) { mtx_assert(&it->it_mtx, MA_OWNED); KASSERT(it->it_usecount > 0, ("invalid it_usecount")); if (--it->it_usecount == 0 && (it->it_flags & ITF_WANTED) != 0) wakeup(it); } #ifndef _SYS_SYSPROTO_H_ struct ktimer_create_args { clockid_t clock_id; struct sigevent * evp; int * timerid; }; #endif int sys_ktimer_create(struct thread *td, struct ktimer_create_args *uap) { struct sigevent *evp, ev; int id; int error; if (uap->evp == NULL) { evp = NULL; } else { error = copyin(uap->evp, &ev, sizeof(ev)); if (error != 0) return (error); evp = &ev; } error = kern_ktimer_create(td, uap->clock_id, evp, &id, -1); if (error == 0) { error = copyout(&id, uap->timerid, sizeof(int)); if (error != 0) kern_ktimer_delete(td, id); } return (error); } int kern_ktimer_create(struct thread *td, clockid_t clock_id, struct sigevent *evp, int *timerid, int preset_id) { struct proc *p = td->td_proc; struct itimer *it; int id; int error; if (clock_id < 0 || clock_id >= MAX_CLOCKS) return (EINVAL); if (posix_clocks[clock_id].timer_create == NULL) return (EINVAL); if (evp != NULL) { if (evp->sigev_notify != SIGEV_NONE && evp->sigev_notify != SIGEV_SIGNAL && evp->sigev_notify != SIGEV_THREAD_ID) return (EINVAL); if ((evp->sigev_notify == SIGEV_SIGNAL || evp->sigev_notify == SIGEV_THREAD_ID) && !_SIG_VALID(evp->sigev_signo)) return (EINVAL); } if (p->p_itimers == NULL) itimers_alloc(p); it = uma_zalloc(itimer_zone, M_WAITOK); it->it_flags = 0; it->it_usecount = 0; it->it_active = 0; timespecclear(&it->it_time.it_value); timespecclear(&it->it_time.it_interval); it->it_overrun = 0; it->it_overrun_last = 0; it->it_clockid = clock_id; it->it_timerid = -1; it->it_proc = p; ksiginfo_init(&it->it_ksi); it->it_ksi.ksi_flags |= KSI_INS | KSI_EXT; error = CLOCK_CALL(clock_id, timer_create, (it)); if (error != 0) goto out; PROC_LOCK(p); if (preset_id != -1) { KASSERT(preset_id >= 0 && preset_id < 3, ("invalid preset_id")); id = preset_id; if (p->p_itimers->its_timers[id] != NULL) { PROC_UNLOCK(p); error = 0; goto out; } } else { /* * Find a free timer slot, skipping those reserved * for setitimer(). */ for (id = 3; id < TIMER_MAX; id++) if (p->p_itimers->its_timers[id] == NULL) break; if (id == TIMER_MAX) { PROC_UNLOCK(p); error = EAGAIN; goto out; } } it->it_timerid = id; p->p_itimers->its_timers[id] = it; if (evp != NULL) it->it_sigev = *evp; else { it->it_sigev.sigev_notify = SIGEV_SIGNAL; switch (clock_id) { default: case CLOCK_REALTIME: it->it_sigev.sigev_signo = SIGALRM; break; case CLOCK_VIRTUAL: it->it_sigev.sigev_signo = SIGVTALRM; break; case CLOCK_PROF: it->it_sigev.sigev_signo = SIGPROF; break; } it->it_sigev.sigev_value.sival_int = id; } if (it->it_sigev.sigev_notify == SIGEV_SIGNAL || it->it_sigev.sigev_notify == SIGEV_THREAD_ID) { it->it_ksi.ksi_signo = it->it_sigev.sigev_signo; it->it_ksi.ksi_code = SI_TIMER; it->it_ksi.ksi_value = it->it_sigev.sigev_value; it->it_ksi.ksi_timerid = id; } PROC_UNLOCK(p); *timerid = id; return (0); out: ITIMER_LOCK(it); CLOCK_CALL(it->it_clockid, timer_delete, (it)); ITIMER_UNLOCK(it); uma_zfree(itimer_zone, it); return (error); } #ifndef _SYS_SYSPROTO_H_ struct ktimer_delete_args { int timerid; }; #endif int sys_ktimer_delete(struct thread *td, struct ktimer_delete_args *uap) { return (kern_ktimer_delete(td, uap->timerid)); } static struct itimer * itimer_find(struct proc *p, int timerid) { struct itimer *it; PROC_LOCK_ASSERT(p, MA_OWNED); if ((p->p_itimers == NULL) || (timerid < 0) || (timerid >= TIMER_MAX) || (it = p->p_itimers->its_timers[timerid]) == NULL) { return (NULL); } ITIMER_LOCK(it); if ((it->it_flags & ITF_DELETING) != 0) { ITIMER_UNLOCK(it); it = NULL; } return (it); } int kern_ktimer_delete(struct thread *td, int timerid) { struct proc *p = td->td_proc; struct itimer *it; PROC_LOCK(p); it = itimer_find(p, timerid); if (it == NULL) { PROC_UNLOCK(p); return (EINVAL); } PROC_UNLOCK(p); it->it_flags |= ITF_DELETING; while (it->it_usecount > 0) { it->it_flags |= ITF_WANTED; msleep(it, &it->it_mtx, PPAUSE, "itimer", 0); } it->it_flags &= ~ITF_WANTED; CLOCK_CALL(it->it_clockid, timer_delete, (it)); ITIMER_UNLOCK(it); PROC_LOCK(p); if (KSI_ONQ(&it->it_ksi)) sigqueue_take(&it->it_ksi); p->p_itimers->its_timers[timerid] = NULL; PROC_UNLOCK(p); uma_zfree(itimer_zone, it); return (0); } #ifndef _SYS_SYSPROTO_H_ struct ktimer_settime_args { int timerid; int flags; const struct itimerspec * value; struct itimerspec * ovalue; }; #endif int sys_ktimer_settime(struct thread *td, struct ktimer_settime_args *uap) { struct itimerspec val, oval, *ovalp; int error; error = copyin(uap->value, &val, sizeof(val)); if (error != 0) return (error); ovalp = uap->ovalue != NULL ? &oval : NULL; error = kern_ktimer_settime(td, uap->timerid, uap->flags, &val, ovalp); if (error == 0 && uap->ovalue != NULL) error = copyout(ovalp, uap->ovalue, sizeof(*ovalp)); return (error); } int kern_ktimer_settime(struct thread *td, int timer_id, int flags, struct itimerspec *val, struct itimerspec *oval) { struct proc *p; struct itimer *it; int error; p = td->td_proc; PROC_LOCK(p); if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) { PROC_UNLOCK(p); error = EINVAL; } else { PROC_UNLOCK(p); itimer_enter(it); error = CLOCK_CALL(it->it_clockid, timer_settime, (it, flags, val, oval)); itimer_leave(it); ITIMER_UNLOCK(it); } return (error); } #ifndef _SYS_SYSPROTO_H_ struct ktimer_gettime_args { int timerid; struct itimerspec * value; }; #endif int sys_ktimer_gettime(struct thread *td, struct ktimer_gettime_args *uap) { struct itimerspec val; int error; error = kern_ktimer_gettime(td, uap->timerid, &val); if (error == 0) error = copyout(&val, uap->value, sizeof(val)); return (error); } int kern_ktimer_gettime(struct thread *td, int timer_id, struct itimerspec *val) { struct proc *p; struct itimer *it; int error; p = td->td_proc; PROC_LOCK(p); if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) { PROC_UNLOCK(p); error = EINVAL; } else { PROC_UNLOCK(p); itimer_enter(it); error = CLOCK_CALL(it->it_clockid, timer_gettime, (it, val)); itimer_leave(it); ITIMER_UNLOCK(it); } return (error); } #ifndef _SYS_SYSPROTO_H_ struct timer_getoverrun_args { int timerid; }; #endif int sys_ktimer_getoverrun(struct thread *td, struct ktimer_getoverrun_args *uap) { return (kern_ktimer_getoverrun(td, uap->timerid)); } int kern_ktimer_getoverrun(struct thread *td, int timer_id) { struct proc *p = td->td_proc; struct itimer *it; int error ; PROC_LOCK(p); if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) { PROC_UNLOCK(p); error = EINVAL; } else { td->td_retval[0] = it->it_overrun_last; ITIMER_UNLOCK(it); PROC_UNLOCK(p); error = 0; } return (error); } static int realtimer_create(struct itimer *it) { callout_init_mtx(&it->it_callout, &it->it_mtx, 0); return (0); } static int realtimer_delete(struct itimer *it) { mtx_assert(&it->it_mtx, MA_OWNED); /* * clear timer's value and interval to tell realtimer_expire * to not rearm the timer. */ timespecclear(&it->it_time.it_value); timespecclear(&it->it_time.it_interval); ITIMER_UNLOCK(it); callout_drain(&it->it_callout); ITIMER_LOCK(it); return (0); } static int realtimer_gettime(struct itimer *it, struct itimerspec *ovalue) { struct timespec cts; mtx_assert(&it->it_mtx, MA_OWNED); realtimer_clocktime(it->it_clockid, &cts); *ovalue = it->it_time; if (ovalue->it_value.tv_sec != 0 || ovalue->it_value.tv_nsec != 0) { timespecsub(&ovalue->it_value, &cts); if (ovalue->it_value.tv_sec < 0 || (ovalue->it_value.tv_sec == 0 && ovalue->it_value.tv_nsec == 0)) { ovalue->it_value.tv_sec = 0; ovalue->it_value.tv_nsec = 1; } } return (0); } static int realtimer_settime(struct itimer *it, int flags, struct itimerspec *value, struct itimerspec *ovalue) { struct timespec cts, ts; struct timeval tv; struct itimerspec val; mtx_assert(&it->it_mtx, MA_OWNED); val = *value; if (itimespecfix(&val.it_value)) return (EINVAL); if (timespecisset(&val.it_value)) { if (itimespecfix(&val.it_interval)) return (EINVAL); } else { timespecclear(&val.it_interval); } if (ovalue != NULL) realtimer_gettime(it, ovalue); it->it_time = val; if (timespecisset(&val.it_value)) { realtimer_clocktime(it->it_clockid, &cts); ts = val.it_value; if ((flags & TIMER_ABSTIME) == 0) { /* Convert to absolute time. */ timespecadd(&it->it_time.it_value, &cts); } else { timespecsub(&ts, &cts); /* * We don't care if ts is negative, tztohz will * fix it. */ } TIMESPEC_TO_TIMEVAL(&tv, &ts); callout_reset(&it->it_callout, tvtohz(&tv), realtimer_expire, it); } else { callout_stop(&it->it_callout); } return (0); } static void realtimer_clocktime(clockid_t id, struct timespec *ts) { if (id == CLOCK_REALTIME) getnanotime(ts); else /* CLOCK_MONOTONIC */ getnanouptime(ts); } int itimer_accept(struct proc *p, int timerid, ksiginfo_t *ksi) { struct itimer *it; PROC_LOCK_ASSERT(p, MA_OWNED); it = itimer_find(p, timerid); if (it != NULL) { ksi->ksi_overrun = it->it_overrun; it->it_overrun_last = it->it_overrun; it->it_overrun = 0; ITIMER_UNLOCK(it); return (0); } return (EINVAL); } int itimespecfix(struct timespec *ts) { if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000) return (EINVAL); if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < tick * 1000) ts->tv_nsec = tick * 1000; return (0); } /* Timeout callback for realtime timer */ static void realtimer_expire(void *arg) { struct timespec cts, ts; struct timeval tv; struct itimer *it; it = (struct itimer *)arg; realtimer_clocktime(it->it_clockid, &cts); /* Only fire if time is reached. */ if (timespeccmp(&cts, &it->it_time.it_value, >=)) { if (timespecisset(&it->it_time.it_interval)) { timespecadd(&it->it_time.it_value, &it->it_time.it_interval); while (timespeccmp(&cts, &it->it_time.it_value, >=)) { if (it->it_overrun < INT_MAX) it->it_overrun++; else it->it_ksi.ksi_errno = ERANGE; timespecadd(&it->it_time.it_value, &it->it_time.it_interval); } } else { /* single shot timer ? */ timespecclear(&it->it_time.it_value); } if (timespecisset(&it->it_time.it_value)) { ts = it->it_time.it_value; timespecsub(&ts, &cts); TIMESPEC_TO_TIMEVAL(&tv, &ts); callout_reset(&it->it_callout, tvtohz(&tv), realtimer_expire, it); } itimer_enter(it); ITIMER_UNLOCK(it); itimer_fire(it); ITIMER_LOCK(it); itimer_leave(it); } else if (timespecisset(&it->it_time.it_value)) { ts = it->it_time.it_value; timespecsub(&ts, &cts); TIMESPEC_TO_TIMEVAL(&tv, &ts); callout_reset(&it->it_callout, tvtohz(&tv), realtimer_expire, it); } } void itimer_fire(struct itimer *it) { struct proc *p = it->it_proc; struct thread *td; if (it->it_sigev.sigev_notify == SIGEV_SIGNAL || it->it_sigev.sigev_notify == SIGEV_THREAD_ID) { if (sigev_findtd(p, &it->it_sigev, &td) != 0) { ITIMER_LOCK(it); timespecclear(&it->it_time.it_value); timespecclear(&it->it_time.it_interval); callout_stop(&it->it_callout); ITIMER_UNLOCK(it); return; } if (!KSI_ONQ(&it->it_ksi)) { it->it_ksi.ksi_errno = 0; ksiginfo_set_sigev(&it->it_ksi, &it->it_sigev); tdsendsignal(p, td, it->it_ksi.ksi_signo, &it->it_ksi); } else { if (it->it_overrun < INT_MAX) it->it_overrun++; else it->it_ksi.ksi_errno = ERANGE; } PROC_UNLOCK(p); } } static void itimers_alloc(struct proc *p) { struct itimers *its; int i; its = malloc(sizeof (struct itimers), M_SUBPROC, M_WAITOK | M_ZERO); LIST_INIT(&its->its_virtual); LIST_INIT(&its->its_prof); TAILQ_INIT(&its->its_worklist); for (i = 0; i < TIMER_MAX; i++) its->its_timers[i] = NULL; PROC_LOCK(p); if (p->p_itimers == NULL) { p->p_itimers = its; PROC_UNLOCK(p); } else { PROC_UNLOCK(p); free(its, M_SUBPROC); } } static void itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp __unused) { itimers_event_hook_exit(arg, p); } /* Clean up timers when some process events are being triggered. */ static void itimers_event_hook_exit(void *arg, struct proc *p) { struct itimers *its; struct itimer *it; int event = (int)(intptr_t)arg; int i; if (p->p_itimers != NULL) { its = p->p_itimers; for (i = 0; i < MAX_CLOCKS; ++i) { if (posix_clocks[i].event_hook != NULL) CLOCK_CALL(i, event_hook, (p, i, event)); } /* * According to susv3, XSI interval timers should be inherited * by new image. */ if (event == ITIMER_EV_EXEC) i = 3; else if (event == ITIMER_EV_EXIT) i = 0; else panic("unhandled event"); for (; i < TIMER_MAX; ++i) { if ((it = its->its_timers[i]) != NULL) kern_ktimer_delete(curthread, i); } if (its->its_timers[0] == NULL && its->its_timers[1] == NULL && its->its_timers[2] == NULL) { free(its, M_SUBPROC); p->p_itimers = NULL; } } } Index: head/sys/kern/kern_timeout.c =================================================================== --- head/sys/kern/kern_timeout.c (revision 305831) +++ head/sys/kern/kern_timeout.c (revision 305832) @@ -1,1665 +1,1665 @@ /*- * Copyright (c) 1982, 1986, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * From: @(#)kern_clock.c 8.5 (Berkeley) 1/21/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_callout_profiling.h" #include "opt_ddb.h" #if defined(__arm__) #include "opt_timer.h" #endif #include "opt_rss.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DDB #include #include #endif #ifdef SMP #include #endif #ifndef NO_EVENTTIMERS DPCPU_DECLARE(sbintime_t, hardclocktime); #endif SDT_PROVIDER_DEFINE(callout_execute); SDT_PROBE_DEFINE1(callout_execute, , , callout__start, "struct callout *"); SDT_PROBE_DEFINE1(callout_execute, , , callout__end, "struct callout *"); #ifdef CALLOUT_PROFILING static int avg_depth; SYSCTL_INT(_debug, OID_AUTO, to_avg_depth, CTLFLAG_RD, &avg_depth, 0, "Average number of items examined per softclock call. Units = 1/1000"); static int avg_gcalls; SYSCTL_INT(_debug, OID_AUTO, to_avg_gcalls, CTLFLAG_RD, &avg_gcalls, 0, "Average number of Giant callouts made per softclock call. Units = 1/1000"); static int avg_lockcalls; SYSCTL_INT(_debug, OID_AUTO, to_avg_lockcalls, CTLFLAG_RD, &avg_lockcalls, 0, "Average number of lock callouts made per softclock call. Units = 1/1000"); static int avg_mpcalls; SYSCTL_INT(_debug, OID_AUTO, to_avg_mpcalls, CTLFLAG_RD, &avg_mpcalls, 0, "Average number of MP callouts made per softclock call. Units = 1/1000"); static int avg_depth_dir; SYSCTL_INT(_debug, OID_AUTO, to_avg_depth_dir, CTLFLAG_RD, &avg_depth_dir, 0, "Average number of direct callouts examined per callout_process call. " "Units = 1/1000"); static int avg_lockcalls_dir; SYSCTL_INT(_debug, OID_AUTO, to_avg_lockcalls_dir, CTLFLAG_RD, &avg_lockcalls_dir, 0, "Average number of lock direct callouts made per " "callout_process call. Units = 1/1000"); static int avg_mpcalls_dir; SYSCTL_INT(_debug, OID_AUTO, to_avg_mpcalls_dir, CTLFLAG_RD, &avg_mpcalls_dir, 0, "Average number of MP direct callouts made per callout_process call. " "Units = 1/1000"); #endif static int ncallout; SYSCTL_INT(_kern, OID_AUTO, ncallout, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &ncallout, 0, "Number of entries in callwheel and size of timeout() preallocation"); #ifdef RSS static int pin_default_swi = 1; static int pin_pcpu_swi = 1; #else static int pin_default_swi = 0; static int pin_pcpu_swi = 0; #endif SYSCTL_INT(_kern, OID_AUTO, pin_default_swi, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &pin_default_swi, 0, "Pin the default (non-per-cpu) swi (shared with PCPU 0 swi)"); SYSCTL_INT(_kern, OID_AUTO, pin_pcpu_swi, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &pin_pcpu_swi, 0, "Pin the per-CPU swis (except PCPU 0, which is also default"); /* * TODO: * allocate more timeout table slots when table overflows. */ u_int callwheelsize, callwheelmask; /* * The callout cpu exec entities represent informations necessary for * describing the state of callouts currently running on the CPU and the ones * necessary for migrating callouts to the new callout cpu. In particular, * the first entry of the array cc_exec_entity holds informations for callout * running in SWI thread context, while the second one holds informations * for callout running directly from hardware interrupt context. * The cached informations are very important for deferring migration when * the migrating callout is already running. */ struct cc_exec { struct callout *cc_curr; void (*cc_drain)(void *); #ifdef SMP void (*ce_migration_func)(void *); void *ce_migration_arg; int ce_migration_cpu; sbintime_t ce_migration_time; sbintime_t ce_migration_prec; #endif bool cc_cancel; bool cc_waiting; }; /* * There is one struct callout_cpu per cpu, holding all relevant * state for the callout processing thread on the individual CPU. */ struct callout_cpu { struct mtx_padalign cc_lock; struct cc_exec cc_exec_entity[2]; struct callout *cc_next; struct callout *cc_callout; struct callout_list *cc_callwheel; struct callout_tailq cc_expireq; struct callout_slist cc_callfree; sbintime_t cc_firstevent; sbintime_t cc_lastscan; void *cc_cookie; u_int cc_bucket; u_int cc_inited; char cc_ktr_event_name[20]; }; #define callout_migrating(c) ((c)->c_iflags & CALLOUT_DFRMIGRATION) #define cc_exec_curr(cc, dir) cc->cc_exec_entity[dir].cc_curr #define cc_exec_drain(cc, dir) cc->cc_exec_entity[dir].cc_drain #define cc_exec_next(cc) cc->cc_next #define cc_exec_cancel(cc, dir) cc->cc_exec_entity[dir].cc_cancel #define cc_exec_waiting(cc, dir) cc->cc_exec_entity[dir].cc_waiting #ifdef SMP #define cc_migration_func(cc, dir) cc->cc_exec_entity[dir].ce_migration_func #define cc_migration_arg(cc, dir) cc->cc_exec_entity[dir].ce_migration_arg #define cc_migration_cpu(cc, dir) cc->cc_exec_entity[dir].ce_migration_cpu #define cc_migration_time(cc, dir) cc->cc_exec_entity[dir].ce_migration_time #define cc_migration_prec(cc, dir) cc->cc_exec_entity[dir].ce_migration_prec struct callout_cpu cc_cpu[MAXCPU]; #define CPUBLOCK MAXCPU #define CC_CPU(cpu) (&cc_cpu[(cpu)]) #define CC_SELF() CC_CPU(PCPU_GET(cpuid)) #else struct callout_cpu cc_cpu; #define CC_CPU(cpu) &cc_cpu #define CC_SELF() &cc_cpu #endif #define CC_LOCK(cc) mtx_lock_spin(&(cc)->cc_lock) #define CC_UNLOCK(cc) mtx_unlock_spin(&(cc)->cc_lock) #define CC_LOCK_ASSERT(cc) mtx_assert(&(cc)->cc_lock, MA_OWNED) static int timeout_cpu; static void callout_cpu_init(struct callout_cpu *cc, int cpu); static void softclock_call_cc(struct callout *c, struct callout_cpu *cc, #ifdef CALLOUT_PROFILING int *mpcalls, int *lockcalls, int *gcalls, #endif int direct); static MALLOC_DEFINE(M_CALLOUT, "callout", "Callout datastructures"); /** * Locked by cc_lock: * cc_curr - If a callout is in progress, it is cc_curr. * If cc_curr is non-NULL, threads waiting in * callout_drain() will be woken up as soon as the * relevant callout completes. * cc_cancel - Changing to 1 with both callout_lock and cc_lock held * guarantees that the current callout will not run. * The softclock() function sets this to 0 before it * drops callout_lock to acquire c_lock, and it calls * the handler only if curr_cancelled is still 0 after * cc_lock is successfully acquired. * cc_waiting - If a thread is waiting in callout_drain(), then * callout_wait is nonzero. Set only when * cc_curr is non-NULL. */ /* * Resets the execution entity tied to a specific callout cpu. */ static void cc_cce_cleanup(struct callout_cpu *cc, int direct) { cc_exec_curr(cc, direct) = NULL; cc_exec_cancel(cc, direct) = false; cc_exec_waiting(cc, direct) = false; #ifdef SMP cc_migration_cpu(cc, direct) = CPUBLOCK; cc_migration_time(cc, direct) = 0; cc_migration_prec(cc, direct) = 0; cc_migration_func(cc, direct) = NULL; cc_migration_arg(cc, direct) = NULL; #endif } /* * Checks if migration is requested by a specific callout cpu. */ static int cc_cce_migrating(struct callout_cpu *cc, int direct) { #ifdef SMP return (cc_migration_cpu(cc, direct) != CPUBLOCK); #else return (0); #endif } /* * Kernel low level callwheel initialization * called on cpu0 during kernel startup. */ static void callout_callwheel_init(void *dummy) { struct callout_cpu *cc; /* * Calculate the size of the callout wheel and the preallocated * timeout() structures. * XXX: Clip callout to result of previous function of maxusers * maximum 384. This is still huge, but acceptable. */ memset(CC_CPU(0), 0, sizeof(cc_cpu)); ncallout = imin(16 + maxproc + maxfiles, 18508); TUNABLE_INT_FETCH("kern.ncallout", &ncallout); /* * Calculate callout wheel size, should be next power of two higher * than 'ncallout'. */ callwheelsize = 1 << fls(ncallout); callwheelmask = callwheelsize - 1; /* * Fetch whether we're pinning the swi's or not. */ TUNABLE_INT_FETCH("kern.pin_default_swi", &pin_default_swi); TUNABLE_INT_FETCH("kern.pin_pcpu_swi", &pin_pcpu_swi); /* * Only cpu0 handles timeout(9) and receives a preallocation. * * XXX: Once all timeout(9) consumers are converted this can * be removed. */ timeout_cpu = PCPU_GET(cpuid); cc = CC_CPU(timeout_cpu); cc->cc_callout = malloc(ncallout * sizeof(struct callout), M_CALLOUT, M_WAITOK); callout_cpu_init(cc, timeout_cpu); } SYSINIT(callwheel_init, SI_SUB_CPU, SI_ORDER_ANY, callout_callwheel_init, NULL); /* * Initialize the per-cpu callout structures. */ static void callout_cpu_init(struct callout_cpu *cc, int cpu) { struct callout *c; int i; mtx_init(&cc->cc_lock, "callout", NULL, MTX_SPIN | MTX_RECURSE); SLIST_INIT(&cc->cc_callfree); cc->cc_inited = 1; cc->cc_callwheel = malloc(sizeof(struct callout_list) * callwheelsize, M_CALLOUT, M_WAITOK); for (i = 0; i < callwheelsize; i++) LIST_INIT(&cc->cc_callwheel[i]); TAILQ_INIT(&cc->cc_expireq); cc->cc_firstevent = SBT_MAX; for (i = 0; i < 2; i++) cc_cce_cleanup(cc, i); snprintf(cc->cc_ktr_event_name, sizeof(cc->cc_ktr_event_name), "callwheel cpu %d", cpu); if (cc->cc_callout == NULL) /* Only cpu0 handles timeout(9) */ return; for (i = 0; i < ncallout; i++) { c = &cc->cc_callout[i]; callout_init(c, 0); c->c_iflags = CALLOUT_LOCAL_ALLOC; SLIST_INSERT_HEAD(&cc->cc_callfree, c, c_links.sle); } } #ifdef SMP /* * Switches the cpu tied to a specific callout. * The function expects a locked incoming callout cpu and returns with * locked outcoming callout cpu. */ static struct callout_cpu * callout_cpu_switch(struct callout *c, struct callout_cpu *cc, int new_cpu) { struct callout_cpu *new_cc; MPASS(c != NULL && cc != NULL); CC_LOCK_ASSERT(cc); /* * Avoid interrupts and preemption firing after the callout cpu * is blocked in order to avoid deadlocks as the new thread * may be willing to acquire the callout cpu lock. */ c->c_cpu = CPUBLOCK; spinlock_enter(); CC_UNLOCK(cc); new_cc = CC_CPU(new_cpu); CC_LOCK(new_cc); spinlock_exit(); c->c_cpu = new_cpu; return (new_cc); } #endif /* * Start standard softclock thread. */ static void start_softclock(void *dummy) { struct callout_cpu *cc; char name[MAXCOMLEN]; #ifdef SMP int cpu; struct intr_event *ie; #endif cc = CC_CPU(timeout_cpu); snprintf(name, sizeof(name), "clock (%d)", timeout_cpu); if (swi_add(&clk_intr_event, name, softclock, cc, SWI_CLOCK, INTR_MPSAFE, &cc->cc_cookie)) panic("died while creating standard software ithreads"); if (pin_default_swi && (intr_event_bind(clk_intr_event, timeout_cpu) != 0)) { printf("%s: timeout clock couldn't be pinned to cpu %d\n", __func__, timeout_cpu); } #ifdef SMP CPU_FOREACH(cpu) { if (cpu == timeout_cpu) continue; cc = CC_CPU(cpu); cc->cc_callout = NULL; /* Only cpu0 handles timeout(9). */ callout_cpu_init(cc, cpu); snprintf(name, sizeof(name), "clock (%d)", cpu); ie = NULL; if (swi_add(&ie, name, softclock, cc, SWI_CLOCK, INTR_MPSAFE, &cc->cc_cookie)) panic("died while creating standard software ithreads"); if (pin_pcpu_swi && (intr_event_bind(ie, cpu) != 0)) { printf("%s: per-cpu clock couldn't be pinned to " "cpu %d\n", __func__, cpu); } } #endif } SYSINIT(start_softclock, SI_SUB_SOFTINTR, SI_ORDER_FIRST, start_softclock, NULL); #define CC_HASH_SHIFT 8 static inline u_int callout_hash(sbintime_t sbt) { return (sbt >> (32 - CC_HASH_SHIFT)); } static inline u_int callout_get_bucket(sbintime_t sbt) { return (callout_hash(sbt) & callwheelmask); } void callout_process(sbintime_t now) { struct callout *tmp, *tmpn; struct callout_cpu *cc; struct callout_list *sc; sbintime_t first, last, max, tmp_max; uint32_t lookahead; u_int firstb, lastb, nowb; #ifdef CALLOUT_PROFILING int depth_dir = 0, mpcalls_dir = 0, lockcalls_dir = 0; #endif cc = CC_SELF(); mtx_lock_spin_flags(&cc->cc_lock, MTX_QUIET); /* Compute the buckets of the last scan and present times. */ firstb = callout_hash(cc->cc_lastscan); cc->cc_lastscan = now; nowb = callout_hash(now); /* Compute the last bucket and minimum time of the bucket after it. */ if (nowb == firstb) lookahead = (SBT_1S / 16); else if (nowb - firstb == 1) lookahead = (SBT_1S / 8); else lookahead = (SBT_1S / 2); first = last = now; first += (lookahead / 2); last += lookahead; last &= (0xffffffffffffffffLLU << (32 - CC_HASH_SHIFT)); lastb = callout_hash(last) - 1; max = last; /* * Check if we wrapped around the entire wheel from the last scan. * In case, we need to scan entirely the wheel for pending callouts. */ if (lastb - firstb >= callwheelsize) { lastb = firstb + callwheelsize - 1; if (nowb - firstb >= callwheelsize) nowb = lastb; } /* Iterate callwheel from firstb to nowb and then up to lastb. */ do { sc = &cc->cc_callwheel[firstb & callwheelmask]; tmp = LIST_FIRST(sc); while (tmp != NULL) { /* Run the callout if present time within allowed. */ if (tmp->c_time <= now) { /* * Consumer told us the callout may be run * directly from hardware interrupt context. */ if (tmp->c_iflags & CALLOUT_DIRECT) { #ifdef CALLOUT_PROFILING ++depth_dir; #endif cc_exec_next(cc) = LIST_NEXT(tmp, c_links.le); cc->cc_bucket = firstb & callwheelmask; LIST_REMOVE(tmp, c_links.le); softclock_call_cc(tmp, cc, #ifdef CALLOUT_PROFILING &mpcalls_dir, &lockcalls_dir, NULL, #endif 1); tmp = cc_exec_next(cc); cc_exec_next(cc) = NULL; } else { tmpn = LIST_NEXT(tmp, c_links.le); LIST_REMOVE(tmp, c_links.le); TAILQ_INSERT_TAIL(&cc->cc_expireq, tmp, c_links.tqe); tmp->c_iflags |= CALLOUT_PROCESSED; tmp = tmpn; } continue; } /* Skip events from distant future. */ if (tmp->c_time >= max) goto next; /* * Event minimal time is bigger than present maximal * time, so it cannot be aggregated. */ if (tmp->c_time > last) { lastb = nowb; goto next; } /* Update first and last time, respecting this event. */ if (tmp->c_time < first) first = tmp->c_time; tmp_max = tmp->c_time + tmp->c_precision; if (tmp_max < last) last = tmp_max; next: tmp = LIST_NEXT(tmp, c_links.le); } /* Proceed with the next bucket. */ firstb++; /* * Stop if we looked after present time and found * some event we can't execute at now. * Stop if we looked far enough into the future. */ } while (((int)(firstb - lastb)) <= 0); cc->cc_firstevent = last; #ifndef NO_EVENTTIMERS cpu_new_callout(curcpu, last, first); #endif #ifdef CALLOUT_PROFILING avg_depth_dir += (depth_dir * 1000 - avg_depth_dir) >> 8; avg_mpcalls_dir += (mpcalls_dir * 1000 - avg_mpcalls_dir) >> 8; avg_lockcalls_dir += (lockcalls_dir * 1000 - avg_lockcalls_dir) >> 8; #endif mtx_unlock_spin_flags(&cc->cc_lock, MTX_QUIET); /* * swi_sched acquires the thread lock, so we don't want to call it * with cc_lock held; incorrect locking order. */ if (!TAILQ_EMPTY(&cc->cc_expireq)) swi_sched(cc->cc_cookie, 0); } static struct callout_cpu * callout_lock(struct callout *c) { struct callout_cpu *cc; int cpu; for (;;) { cpu = c->c_cpu; #ifdef SMP if (cpu == CPUBLOCK) { while (c->c_cpu == CPUBLOCK) cpu_spinwait(); continue; } #endif cc = CC_CPU(cpu); CC_LOCK(cc); if (cpu == c->c_cpu) break; CC_UNLOCK(cc); } return (cc); } static void callout_cc_add(struct callout *c, struct callout_cpu *cc, sbintime_t sbt, sbintime_t precision, void (*func)(void *), void *arg, int cpu, int flags) { int bucket; CC_LOCK_ASSERT(cc); if (sbt < cc->cc_lastscan) sbt = cc->cc_lastscan; c->c_arg = arg; c->c_iflags |= CALLOUT_PENDING; c->c_iflags &= ~CALLOUT_PROCESSED; c->c_flags |= CALLOUT_ACTIVE; if (flags & C_DIRECT_EXEC) c->c_iflags |= CALLOUT_DIRECT; c->c_func = func; c->c_time = sbt; c->c_precision = precision; bucket = callout_get_bucket(c->c_time); CTR3(KTR_CALLOUT, "precision set for %p: %d.%08x", c, (int)(c->c_precision >> 32), (u_int)(c->c_precision & 0xffffffff)); LIST_INSERT_HEAD(&cc->cc_callwheel[bucket], c, c_links.le); if (cc->cc_bucket == bucket) cc_exec_next(cc) = c; #ifndef NO_EVENTTIMERS /* * Inform the eventtimers(4) subsystem there's a new callout * that has been inserted, but only if really required. */ if (SBT_MAX - c->c_time < c->c_precision) c->c_precision = SBT_MAX - c->c_time; sbt = c->c_time + c->c_precision; if (sbt < cc->cc_firstevent) { cc->cc_firstevent = sbt; cpu_new_callout(cpu, sbt, c->c_time); } #endif } static void callout_cc_del(struct callout *c, struct callout_cpu *cc) { if ((c->c_iflags & CALLOUT_LOCAL_ALLOC) == 0) return; c->c_func = NULL; SLIST_INSERT_HEAD(&cc->cc_callfree, c, c_links.sle); } static void softclock_call_cc(struct callout *c, struct callout_cpu *cc, #ifdef CALLOUT_PROFILING int *mpcalls, int *lockcalls, int *gcalls, #endif int direct) { struct rm_priotracker tracker; void (*c_func)(void *); void *c_arg; struct lock_class *class; struct lock_object *c_lock; uintptr_t lock_status; int c_iflags; #ifdef SMP struct callout_cpu *new_cc; void (*new_func)(void *); void *new_arg; int flags, new_cpu; sbintime_t new_prec, new_time; #endif #if defined(DIAGNOSTIC) || defined(CALLOUT_PROFILING) sbintime_t sbt1, sbt2; struct timespec ts2; static sbintime_t maxdt = 2 * SBT_1MS; /* 2 msec */ static timeout_t *lastfunc; #endif KASSERT((c->c_iflags & CALLOUT_PENDING) == CALLOUT_PENDING, ("softclock_call_cc: pend %p %x", c, c->c_iflags)); KASSERT((c->c_flags & CALLOUT_ACTIVE) == CALLOUT_ACTIVE, ("softclock_call_cc: act %p %x", c, c->c_flags)); class = (c->c_lock != NULL) ? LOCK_CLASS(c->c_lock) : NULL; lock_status = 0; if (c->c_flags & CALLOUT_SHAREDLOCK) { if (class == &lock_class_rm) lock_status = (uintptr_t)&tracker; else lock_status = 1; } c_lock = c->c_lock; c_func = c->c_func; c_arg = c->c_arg; c_iflags = c->c_iflags; if (c->c_iflags & CALLOUT_LOCAL_ALLOC) c->c_iflags = CALLOUT_LOCAL_ALLOC; else c->c_iflags &= ~CALLOUT_PENDING; cc_exec_curr(cc, direct) = c; cc_exec_cancel(cc, direct) = false; cc_exec_drain(cc, direct) = NULL; CC_UNLOCK(cc); if (c_lock != NULL) { class->lc_lock(c_lock, lock_status); /* * The callout may have been cancelled * while we switched locks. */ if (cc_exec_cancel(cc, direct)) { class->lc_unlock(c_lock); goto skip; } /* The callout cannot be stopped now. */ cc_exec_cancel(cc, direct) = true; if (c_lock == &Giant.lock_object) { #ifdef CALLOUT_PROFILING (*gcalls)++; #endif CTR3(KTR_CALLOUT, "callout giant %p func %p arg %p", c, c_func, c_arg); } else { #ifdef CALLOUT_PROFILING (*lockcalls)++; #endif CTR3(KTR_CALLOUT, "callout lock %p func %p arg %p", c, c_func, c_arg); } } else { #ifdef CALLOUT_PROFILING (*mpcalls)++; #endif CTR3(KTR_CALLOUT, "callout %p func %p arg %p", c, c_func, c_arg); } KTR_STATE3(KTR_SCHED, "callout", cc->cc_ktr_event_name, "running", "func:%p", c_func, "arg:%p", c_arg, "direct:%d", direct); #if defined(DIAGNOSTIC) || defined(CALLOUT_PROFILING) sbt1 = sbinuptime(); #endif THREAD_NO_SLEEPING(); SDT_PROBE1(callout_execute, , , callout__start, c); c_func(c_arg); SDT_PROBE1(callout_execute, , , callout__end, c); THREAD_SLEEPING_OK(); #if defined(DIAGNOSTIC) || defined(CALLOUT_PROFILING) sbt2 = sbinuptime(); sbt2 -= sbt1; if (sbt2 > maxdt) { if (lastfunc != c_func || sbt2 > maxdt * 2) { ts2 = sbttots(sbt2); printf( "Expensive timeout(9) function: %p(%p) %jd.%09ld s\n", c_func, c_arg, (intmax_t)ts2.tv_sec, ts2.tv_nsec); } maxdt = sbt2; lastfunc = c_func; } #endif KTR_STATE0(KTR_SCHED, "callout", cc->cc_ktr_event_name, "idle"); CTR1(KTR_CALLOUT, "callout %p finished", c); if ((c_iflags & CALLOUT_RETURNUNLOCKED) == 0) class->lc_unlock(c_lock); skip: CC_LOCK(cc); KASSERT(cc_exec_curr(cc, direct) == c, ("mishandled cc_curr")); cc_exec_curr(cc, direct) = NULL; if (cc_exec_drain(cc, direct)) { void (*drain)(void *); drain = cc_exec_drain(cc, direct); cc_exec_drain(cc, direct) = NULL; CC_UNLOCK(cc); drain(c_arg); CC_LOCK(cc); } if (cc_exec_waiting(cc, direct)) { /* * There is someone waiting for the * callout to complete. * If the callout was scheduled for * migration just cancel it. */ if (cc_cce_migrating(cc, direct)) { cc_cce_cleanup(cc, direct); /* * It should be assert here that the callout is not * destroyed but that is not easy. */ c->c_iflags &= ~CALLOUT_DFRMIGRATION; } cc_exec_waiting(cc, direct) = false; CC_UNLOCK(cc); wakeup(&cc_exec_waiting(cc, direct)); CC_LOCK(cc); } else if (cc_cce_migrating(cc, direct)) { KASSERT((c_iflags & CALLOUT_LOCAL_ALLOC) == 0, ("Migrating legacy callout %p", c)); #ifdef SMP /* * If the callout was scheduled for * migration just perform it now. */ new_cpu = cc_migration_cpu(cc, direct); new_time = cc_migration_time(cc, direct); new_prec = cc_migration_prec(cc, direct); new_func = cc_migration_func(cc, direct); new_arg = cc_migration_arg(cc, direct); cc_cce_cleanup(cc, direct); /* * It should be assert here that the callout is not destroyed * but that is not easy. * * As first thing, handle deferred callout stops. */ if (!callout_migrating(c)) { CTR3(KTR_CALLOUT, "deferred cancelled %p func %p arg %p", c, new_func, new_arg); callout_cc_del(c, cc); return; } c->c_iflags &= ~CALLOUT_DFRMIGRATION; new_cc = callout_cpu_switch(c, cc, new_cpu); flags = (direct) ? C_DIRECT_EXEC : 0; callout_cc_add(c, new_cc, new_time, new_prec, new_func, new_arg, new_cpu, flags); CC_UNLOCK(new_cc); CC_LOCK(cc); #else panic("migration should not happen"); #endif } /* * If the current callout is locally allocated (from * timeout(9)) then put it on the freelist. * * Note: we need to check the cached copy of c_iflags because * if it was not local, then it's not safe to deref the * callout pointer. */ KASSERT((c_iflags & CALLOUT_LOCAL_ALLOC) == 0 || c->c_iflags == CALLOUT_LOCAL_ALLOC, ("corrupted callout")); if (c_iflags & CALLOUT_LOCAL_ALLOC) callout_cc_del(c, cc); } /* * The callout mechanism is based on the work of Adam M. Costello and * George Varghese, published in a technical report entitled "Redesigning * the BSD Callout and Timer Facilities" and modified slightly for inclusion * in FreeBSD by Justin T. Gibbs. The original work on the data structures * used in this implementation was published by G. Varghese and T. Lauck in * the paper "Hashed and Hierarchical Timing Wheels: Data Structures for * the Efficient Implementation of a Timer Facility" in the Proceedings of * the 11th ACM Annual Symposium on Operating Systems Principles, * Austin, Texas Nov 1987. */ /* * Software (low priority) clock interrupt. * Run periodic events from timeout queue. */ void softclock(void *arg) { struct callout_cpu *cc; struct callout *c; #ifdef CALLOUT_PROFILING int depth = 0, gcalls = 0, lockcalls = 0, mpcalls = 0; #endif cc = (struct callout_cpu *)arg; CC_LOCK(cc); while ((c = TAILQ_FIRST(&cc->cc_expireq)) != NULL) { TAILQ_REMOVE(&cc->cc_expireq, c, c_links.tqe); softclock_call_cc(c, cc, #ifdef CALLOUT_PROFILING &mpcalls, &lockcalls, &gcalls, #endif 0); #ifdef CALLOUT_PROFILING ++depth; #endif } #ifdef CALLOUT_PROFILING avg_depth += (depth * 1000 - avg_depth) >> 8; avg_mpcalls += (mpcalls * 1000 - avg_mpcalls) >> 8; avg_lockcalls += (lockcalls * 1000 - avg_lockcalls) >> 8; avg_gcalls += (gcalls * 1000 - avg_gcalls) >> 8; #endif CC_UNLOCK(cc); } /* * timeout -- * Execute a function after a specified length of time. * * untimeout -- * Cancel previous timeout function call. * * callout_handle_init -- * Initialize a handle so that using it with untimeout is benign. * * See AT&T BCI Driver Reference Manual for specification. This * implementation differs from that one in that although an * identification value is returned from timeout, the original * arguments to timeout as well as the identifier are used to * identify entries for untimeout. */ struct callout_handle timeout(timeout_t *ftn, void *arg, int to_ticks) { struct callout_cpu *cc; struct callout *new; struct callout_handle handle; cc = CC_CPU(timeout_cpu); CC_LOCK(cc); /* Fill in the next free callout structure. */ new = SLIST_FIRST(&cc->cc_callfree); if (new == NULL) /* XXX Attempt to malloc first */ panic("timeout table full"); SLIST_REMOVE_HEAD(&cc->cc_callfree, c_links.sle); callout_reset(new, to_ticks, ftn, arg); handle.callout = new; CC_UNLOCK(cc); return (handle); } void untimeout(timeout_t *ftn, void *arg, struct callout_handle handle) { struct callout_cpu *cc; /* * Check for a handle that was initialized * by callout_handle_init, but never used * for a real timeout. */ if (handle.callout == NULL) return; cc = callout_lock(handle.callout); if (handle.callout->c_func == ftn && handle.callout->c_arg == arg) callout_stop(handle.callout); CC_UNLOCK(cc); } void callout_handle_init(struct callout_handle *handle) { handle->callout = NULL; } void callout_when(sbintime_t sbt, sbintime_t precision, int flags, sbintime_t *res, sbintime_t *prec_res) { sbintime_t to_sbt, to_pr; if ((flags & (C_ABSOLUTE | C_PRECALC)) != 0) { *res = sbt; *prec_res = precision; return; } if ((flags & C_HARDCLOCK) != 0 && sbt < tick_sbt) sbt = tick_sbt; if ((flags & C_HARDCLOCK) != 0 || #ifdef NO_EVENTTIMERS sbt >= sbt_timethreshold) { to_sbt = getsbinuptime(); /* Add safety belt for the case of hz > 1000. */ to_sbt += tc_tick_sbt - tick_sbt; #else sbt >= sbt_tickthreshold) { /* * Obtain the time of the last hardclock() call on * this CPU directly from the kern_clocksource.c. * This value is per-CPU, but it is equal for all * active ones. */ #ifdef __LP64__ to_sbt = DPCPU_GET(hardclocktime); #else spinlock_enter(); to_sbt = DPCPU_GET(hardclocktime); spinlock_exit(); #endif #endif if ((flags & C_HARDCLOCK) == 0) to_sbt += tick_sbt; } else to_sbt = sbinuptime(); if (SBT_MAX - to_sbt < sbt) to_sbt = SBT_MAX; else to_sbt += sbt; *res = to_sbt; to_pr = ((C_PRELGET(flags) < 0) ? sbt >> tc_precexp : sbt >> C_PRELGET(flags)); *prec_res = to_pr > precision ? to_pr : precision; } /* * New interface; clients allocate their own callout structures. * * callout_reset() - establish or change a timeout * callout_stop() - disestablish a timeout * callout_init() - initialize a callout structure so that it can * safely be passed to callout_reset() and callout_stop() * * defines three convenience macros: * * callout_active() - returns truth if callout has not been stopped, * drained, or deactivated since the last time the callout was * reset. * callout_pending() - returns truth if callout is still waiting for timeout * callout_deactivate() - marks the callout as having been serviced */ int callout_reset_sbt_on(struct callout *c, sbintime_t sbt, sbintime_t prec, void (*ftn)(void *), void *arg, int cpu, int flags) { sbintime_t to_sbt, precision; struct callout_cpu *cc; int cancelled, direct; int ignore_cpu=0; cancelled = 0; if (cpu == -1) { ignore_cpu = 1; } else if ((cpu >= MAXCPU) || ((CC_CPU(cpu))->cc_inited == 0)) { /* Invalid CPU spec */ panic("Invalid CPU in callout %d", cpu); } callout_when(sbt, prec, flags, &to_sbt, &precision); /* * This flag used to be added by callout_cc_add, but the * first time you call this we could end up with the * wrong direct flag if we don't do it before we add. */ if (flags & C_DIRECT_EXEC) { direct = 1; } else { direct = 0; } KASSERT(!direct || c->c_lock == NULL, ("%s: direct callout %p has lock", __func__, c)); cc = callout_lock(c); /* * Don't allow migration of pre-allocated callouts lest they * become unbalanced or handle the case where the user does * not care. */ if ((c->c_iflags & CALLOUT_LOCAL_ALLOC) || ignore_cpu) { cpu = c->c_cpu; } if (cc_exec_curr(cc, direct) == c) { /* * We're being asked to reschedule a callout which is * currently in progress. If there is a lock then we * can cancel the callout if it has not really started. */ if (c->c_lock != NULL && !cc_exec_cancel(cc, direct)) cancelled = cc_exec_cancel(cc, direct) = true; if (cc_exec_waiting(cc, direct) || cc_exec_drain(cc, direct)) { /* * Someone has called callout_drain to kill this * callout. Don't reschedule. */ CTR4(KTR_CALLOUT, "%s %p func %p arg %p", cancelled ? "cancelled" : "failed to cancel", c, c->c_func, c->c_arg); CC_UNLOCK(cc); return (cancelled); } #ifdef SMP if (callout_migrating(c)) { /* * This only occurs when a second callout_reset_sbt_on * is made after a previous one moved it into * deferred migration (below). Note we do *not* change * the prev_cpu even though the previous target may * be different. */ cc_migration_cpu(cc, direct) = cpu; cc_migration_time(cc, direct) = to_sbt; cc_migration_prec(cc, direct) = precision; cc_migration_func(cc, direct) = ftn; cc_migration_arg(cc, direct) = arg; cancelled = 1; CC_UNLOCK(cc); return (cancelled); } #endif } if (c->c_iflags & CALLOUT_PENDING) { if ((c->c_iflags & CALLOUT_PROCESSED) == 0) { if (cc_exec_next(cc) == c) cc_exec_next(cc) = LIST_NEXT(c, c_links.le); LIST_REMOVE(c, c_links.le); } else { TAILQ_REMOVE(&cc->cc_expireq, c, c_links.tqe); } cancelled = 1; c->c_iflags &= ~ CALLOUT_PENDING; c->c_flags &= ~ CALLOUT_ACTIVE; } #ifdef SMP /* * If the callout must migrate try to perform it immediately. * If the callout is currently running, just defer the migration * to a more appropriate moment. */ if (c->c_cpu != cpu) { if (cc_exec_curr(cc, direct) == c) { /* * Pending will have been removed since we are * actually executing the callout on another * CPU. That callout should be waiting on the * lock the caller holds. If we set both * active/and/pending after we return and the * lock on the executing callout proceeds, it * will then see pending is true and return. * At the return from the actual callout execution * the migration will occur in softclock_call_cc * and this new callout will be placed on the * new CPU via a call to callout_cpu_switch() which * will get the lock on the right CPU followed * by a call callout_cc_add() which will add it there. * (see above in softclock_call_cc()). */ cc_migration_cpu(cc, direct) = cpu; cc_migration_time(cc, direct) = to_sbt; cc_migration_prec(cc, direct) = precision; cc_migration_func(cc, direct) = ftn; cc_migration_arg(cc, direct) = arg; c->c_iflags |= (CALLOUT_DFRMIGRATION | CALLOUT_PENDING); c->c_flags |= CALLOUT_ACTIVE; CTR6(KTR_CALLOUT, "migration of %p func %p arg %p in %d.%08x to %u deferred", c, c->c_func, c->c_arg, (int)(to_sbt >> 32), (u_int)(to_sbt & 0xffffffff), cpu); CC_UNLOCK(cc); return (cancelled); } cc = callout_cpu_switch(c, cc, cpu); } #endif callout_cc_add(c, cc, to_sbt, precision, ftn, arg, cpu, flags); CTR6(KTR_CALLOUT, "%sscheduled %p func %p arg %p in %d.%08x", cancelled ? "re" : "", c, c->c_func, c->c_arg, (int)(to_sbt >> 32), (u_int)(to_sbt & 0xffffffff)); CC_UNLOCK(cc); return (cancelled); } /* * Common idioms that can be optimized in the future. */ int callout_schedule_on(struct callout *c, int to_ticks, int cpu) { return callout_reset_on(c, to_ticks, c->c_func, c->c_arg, cpu); } int callout_schedule(struct callout *c, int to_ticks) { return callout_reset_on(c, to_ticks, c->c_func, c->c_arg, c->c_cpu); } int _callout_stop_safe(struct callout *c, int flags, void (*drain)(void *)) { struct callout_cpu *cc, *old_cc; struct lock_class *class; int direct, sq_locked, use_lock; int cancelled, not_on_a_list; if ((flags & CS_DRAIN) != 0) WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, c->c_lock, "calling %s", __func__); /* * Some old subsystems don't hold Giant while running a callout_stop(), * so just discard this check for the moment. */ if ((flags & CS_DRAIN) == 0 && c->c_lock != NULL) { if (c->c_lock == &Giant.lock_object) use_lock = mtx_owned(&Giant); else { use_lock = 1; class = LOCK_CLASS(c->c_lock); class->lc_assert(c->c_lock, LA_XLOCKED); } } else use_lock = 0; if (c->c_iflags & CALLOUT_DIRECT) { direct = 1; } else { direct = 0; } sq_locked = 0; old_cc = NULL; again: cc = callout_lock(c); if ((c->c_iflags & (CALLOUT_DFRMIGRATION | CALLOUT_PENDING)) == (CALLOUT_DFRMIGRATION | CALLOUT_PENDING) && ((c->c_flags & CALLOUT_ACTIVE) == CALLOUT_ACTIVE)) { /* * Special case where this slipped in while we * were migrating *as* the callout is about to * execute. The caller probably holds the lock * the callout wants. * * Get rid of the migration first. Then set * the flag that tells this code *not* to * try to remove it from any lists (its not * on one yet). When the callout wheel runs, * it will ignore this callout. */ c->c_iflags &= ~CALLOUT_PENDING; c->c_flags &= ~CALLOUT_ACTIVE; not_on_a_list = 1; } else { not_on_a_list = 0; } /* * If the callout was migrating while the callout cpu lock was * dropped, just drop the sleepqueue lock and check the states * again. */ if (sq_locked != 0 && cc != old_cc) { #ifdef SMP CC_UNLOCK(cc); sleepq_release(&cc_exec_waiting(old_cc, direct)); sq_locked = 0; old_cc = NULL; goto again; #else panic("migration should not happen"); #endif } /* * If the callout is running, try to stop it or drain it. */ if (cc_exec_curr(cc, direct) == c) { /* * Succeed we to stop it or not, we must clear the * active flag - this is what API users expect. */ c->c_flags &= ~CALLOUT_ACTIVE; if ((flags & CS_DRAIN) != 0) { /* * The current callout is running (or just * about to run) and blocking is allowed, so * just wait for the current invocation to * finish. */ while (cc_exec_curr(cc, direct) == c) { /* * Use direct calls to sleepqueue interface * instead of cv/msleep in order to avoid * a LOR between cc_lock and sleepqueue * chain spinlocks. This piece of code * emulates a msleep_spin() call actually. * * If we already have the sleepqueue chain * locked, then we can safely block. If we * don't already have it locked, however, * we have to drop the cc_lock to lock * it. This opens several races, so we * restart at the beginning once we have * both locks. If nothing has changed, then * we will end up back here with sq_locked * set. */ if (!sq_locked) { CC_UNLOCK(cc); sleepq_lock( &cc_exec_waiting(cc, direct)); sq_locked = 1; old_cc = cc; goto again; } /* * Migration could be cancelled here, but * as long as it is still not sure when it * will be packed up, just let softclock() * take care of it. */ cc_exec_waiting(cc, direct) = true; DROP_GIANT(); CC_UNLOCK(cc); sleepq_add( &cc_exec_waiting(cc, direct), &cc->cc_lock.lock_object, "codrain", SLEEPQ_SLEEP, 0); sleepq_wait( &cc_exec_waiting(cc, direct), 0); sq_locked = 0; old_cc = NULL; /* Reacquire locks previously released. */ PICKUP_GIANT(); CC_LOCK(cc); } } else if (use_lock && !cc_exec_cancel(cc, direct) && (drain == NULL)) { /* * The current callout is waiting for its * lock which we hold. Cancel the callout * and return. After our caller drops the * lock, the callout will be skipped in * softclock(). This *only* works with a * callout_stop() *not* callout_drain() or * callout_async_drain(). */ cc_exec_cancel(cc, direct) = true; CTR3(KTR_CALLOUT, "cancelled %p func %p arg %p", c, c->c_func, c->c_arg); KASSERT(!cc_cce_migrating(cc, direct), ("callout wrongly scheduled for migration")); if (callout_migrating(c)) { c->c_iflags &= ~CALLOUT_DFRMIGRATION; #ifdef SMP cc_migration_cpu(cc, direct) = CPUBLOCK; cc_migration_time(cc, direct) = 0; cc_migration_prec(cc, direct) = 0; cc_migration_func(cc, direct) = NULL; cc_migration_arg(cc, direct) = NULL; #endif } CC_UNLOCK(cc); KASSERT(!sq_locked, ("sleepqueue chain locked")); return (1); } else if (callout_migrating(c)) { /* * The callout is currently being serviced * and the "next" callout is scheduled at * its completion with a migration. We remove * the migration flag so it *won't* get rescheduled, * but we can't stop the one thats running so * we return 0. */ c->c_iflags &= ~CALLOUT_DFRMIGRATION; #ifdef SMP /* * We can't call cc_cce_cleanup here since * if we do it will remove .ce_curr and * its still running. This will prevent a * reschedule of the callout when the * execution completes. */ cc_migration_cpu(cc, direct) = CPUBLOCK; cc_migration_time(cc, direct) = 0; cc_migration_prec(cc, direct) = 0; cc_migration_func(cc, direct) = NULL; cc_migration_arg(cc, direct) = NULL; #endif CTR3(KTR_CALLOUT, "postponing stop %p func %p arg %p", c, c->c_func, c->c_arg); if (drain) { cc_exec_drain(cc, direct) = drain; } CC_UNLOCK(cc); return ((flags & CS_EXECUTING) != 0); } CTR3(KTR_CALLOUT, "failed to stop %p func %p arg %p", c, c->c_func, c->c_arg); if (drain) { cc_exec_drain(cc, direct) = drain; } KASSERT(!sq_locked, ("sleepqueue chain still locked")); cancelled = ((flags & CS_EXECUTING) != 0); } else cancelled = 1; if (sq_locked) sleepq_release(&cc_exec_waiting(cc, direct)); if ((c->c_iflags & CALLOUT_PENDING) == 0) { CTR3(KTR_CALLOUT, "failed to stop %p func %p arg %p", c, c->c_func, c->c_arg); /* * For not scheduled and not executing callout return * negative value. */ if (cc_exec_curr(cc, direct) != c) cancelled = -1; CC_UNLOCK(cc); return (cancelled); } c->c_iflags &= ~CALLOUT_PENDING; c->c_flags &= ~CALLOUT_ACTIVE; CTR3(KTR_CALLOUT, "cancelled %p func %p arg %p", c, c->c_func, c->c_arg); if (not_on_a_list == 0) { if ((c->c_iflags & CALLOUT_PROCESSED) == 0) { if (cc_exec_next(cc) == c) cc_exec_next(cc) = LIST_NEXT(c, c_links.le); LIST_REMOVE(c, c_links.le); } else { TAILQ_REMOVE(&cc->cc_expireq, c, c_links.tqe); } } callout_cc_del(c, cc); CC_UNLOCK(cc); return (cancelled); } void callout_init(struct callout *c, int mpsafe) { bzero(c, sizeof *c); if (mpsafe) { c->c_lock = NULL; c->c_iflags = CALLOUT_RETURNUNLOCKED; } else { c->c_lock = &Giant.lock_object; c->c_iflags = 0; } c->c_cpu = timeout_cpu; } void _callout_init_lock(struct callout *c, struct lock_object *lock, int flags) { bzero(c, sizeof *c); c->c_lock = lock; KASSERT((flags & ~(CALLOUT_RETURNUNLOCKED | CALLOUT_SHAREDLOCK)) == 0, ("callout_init_lock: bad flags %d", flags)); KASSERT(lock != NULL || (flags & CALLOUT_RETURNUNLOCKED) == 0, ("callout_init_lock: CALLOUT_RETURNUNLOCKED with no lock")); KASSERT(lock == NULL || !(LOCK_CLASS(lock)->lc_flags & (LC_SPINLOCK | LC_SLEEPABLE)), ("%s: invalid lock class", __func__)); c->c_iflags = flags & (CALLOUT_RETURNUNLOCKED | CALLOUT_SHAREDLOCK); c->c_cpu = timeout_cpu; } #ifdef APM_FIXUP_CALLTODO /* * Adjust the kernel calltodo timeout list. This routine is used after * an APM resume to recalculate the calltodo timer list values with the * number of hz's we have been sleeping. The next hardclock() will detect * that there are fired timers and run softclock() to execute them. * * Please note, I have not done an exhaustive analysis of what code this * might break. I am motivated to have my select()'s and alarm()'s that * have expired during suspend firing upon resume so that the applications * which set the timer can do the maintanence the timer was for as close * as possible to the originally intended time. Testing this code for a * week showed that resuming from a suspend resulted in 22 to 25 timers * firing, which seemed independent on whether the suspend was 2 hours or * 2 days. Your milage may vary. - Ken Key */ void adjust_timeout_calltodo(struct timeval *time_change) { register struct callout *p; unsigned long delta_ticks; /* * How many ticks were we asleep? * (stolen from tvtohz()). */ /* Don't do anything */ if (time_change->tv_sec < 0) return; else if (time_change->tv_sec <= LONG_MAX / 1000000) delta_ticks = howmany(time_change->tv_sec * 1000000 + time_change->tv_usec, tick) + 1; else if (time_change->tv_sec <= LONG_MAX / hz) delta_ticks = time_change->tv_sec * hz + howmany(time_change->tv_usec, tick) + 1; else delta_ticks = LONG_MAX; if (delta_ticks > INT_MAX) delta_ticks = INT_MAX; /* * Now rip through the timer calltodo list looking for timers * to expire. */ /* don't collide with softclock() */ CC_LOCK(cc); for (p = calltodo.c_next; p != NULL; p = p->c_next) { p->c_time -= delta_ticks; /* Break if the timer had more time on it than delta_ticks */ if (p->c_time > 0) break; /* take back the ticks the timer didn't use (p->c_time <= 0) */ delta_ticks = -p->c_time; } CC_UNLOCK(cc); return; } #endif /* APM_FIXUP_CALLTODO */ static int flssbt(sbintime_t sbt) { sbt += (uint64_t)sbt >> 1; if (sizeof(long) >= sizeof(sbintime_t)) return (flsl(sbt)); if (sbt >= SBT_1S) return (flsl(((uint64_t)sbt) >> 32) + 32); return (flsl(sbt)); } /* * Dump immediate statistic snapshot of the scheduled callouts. */ static int sysctl_kern_callout_stat(SYSCTL_HANDLER_ARGS) { struct callout *tmp; struct callout_cpu *cc; struct callout_list *sc; sbintime_t maxpr, maxt, medpr, medt, now, spr, st, t; int ct[64], cpr[64], ccpbk[32]; int error, val, i, count, tcum, pcum, maxc, c, medc; #ifdef SMP int cpu; #endif val = 0; error = sysctl_handle_int(oidp, &val, 0, req); if (error != 0 || req->newptr == NULL) return (error); count = maxc = 0; st = spr = maxt = maxpr = 0; bzero(ccpbk, sizeof(ccpbk)); bzero(ct, sizeof(ct)); bzero(cpr, sizeof(cpr)); now = sbinuptime(); #ifdef SMP CPU_FOREACH(cpu) { cc = CC_CPU(cpu); #else cc = CC_CPU(timeout_cpu); #endif CC_LOCK(cc); for (i = 0; i < callwheelsize; i++) { sc = &cc->cc_callwheel[i]; c = 0; LIST_FOREACH(tmp, sc, c_links.le) { c++; t = tmp->c_time - now; if (t < 0) t = 0; st += t / SBT_1US; spr += tmp->c_precision / SBT_1US; if (t > maxt) maxt = t; if (tmp->c_precision > maxpr) maxpr = tmp->c_precision; ct[flssbt(t)]++; cpr[flssbt(tmp->c_precision)]++; } if (c > maxc) maxc = c; ccpbk[fls(c + c / 2)]++; count += c; } CC_UNLOCK(cc); #ifdef SMP } #endif for (i = 0, tcum = 0; i < 64 && tcum < count / 2; i++) tcum += ct[i]; medt = (i >= 2) ? (((sbintime_t)1) << (i - 2)) : 0; for (i = 0, pcum = 0; i < 64 && pcum < count / 2; i++) pcum += cpr[i]; medpr = (i >= 2) ? (((sbintime_t)1) << (i - 2)) : 0; for (i = 0, c = 0; i < 32 && c < count / 2; i++) c += ccpbk[i]; medc = (i >= 2) ? (1 << (i - 2)) : 0; printf("Scheduled callouts statistic snapshot:\n"); printf(" Callouts: %6d Buckets: %6d*%-3d Bucket size: 0.%06ds\n", count, callwheelsize, mp_ncpus, 1000000 >> CC_HASH_SHIFT); printf(" C/Bk: med %5d avg %6d.%06jd max %6d\n", medc, count / callwheelsize / mp_ncpus, (uint64_t)count * 1000000 / callwheelsize / mp_ncpus % 1000000, maxc); printf(" Time: med %5jd.%06jds avg %6jd.%06jds max %6jd.%06jds\n", medt / SBT_1S, (medt & 0xffffffff) * 1000000 >> 32, (st / count) / 1000000, (st / count) % 1000000, maxt / SBT_1S, (maxt & 0xffffffff) * 1000000 >> 32); printf(" Prec: med %5jd.%06jds avg %6jd.%06jds max %6jd.%06jds\n", medpr / SBT_1S, (medpr & 0xffffffff) * 1000000 >> 32, (spr / count) / 1000000, (spr / count) % 1000000, maxpr / SBT_1S, (maxpr & 0xffffffff) * 1000000 >> 32); printf(" Distribution: \tbuckets\t time\t tcum\t" " prec\t pcum\n"); for (i = 0, tcum = pcum = 0; i < 64; i++) { if (ct[i] == 0 && cpr[i] == 0) continue; t = (i != 0) ? (((sbintime_t)1) << (i - 1)) : 0; tcum += ct[i]; pcum += cpr[i]; printf(" %10jd.%06jds\t 2**%d\t%7d\t%7d\t%7d\t%7d\n", t / SBT_1S, (t & 0xffffffff) * 1000000 >> 32, i - 1 - (32 - CC_HASH_SHIFT), ct[i], tcum, cpr[i], pcum); } return (error); } SYSCTL_PROC(_kern, OID_AUTO, callout_stat, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0, sysctl_kern_callout_stat, "I", "Dump immediate statistic snapshot of the scheduled callouts"); #ifdef DDB static void _show_callout(struct callout *c) { db_printf("callout %p\n", c); #define C_DB_PRINTF(f, e) db_printf(" %s = " f "\n", #e, c->e); db_printf(" &c_links = %p\n", &(c->c_links)); C_DB_PRINTF("%" PRId64, c_time); C_DB_PRINTF("%" PRId64, c_precision); C_DB_PRINTF("%p", c_arg); C_DB_PRINTF("%p", c_func); C_DB_PRINTF("%p", c_lock); C_DB_PRINTF("%#x", c_flags); C_DB_PRINTF("%#x", c_iflags); C_DB_PRINTF("%d", c_cpu); #undef C_DB_PRINTF } DB_SHOW_COMMAND(callout, db_show_callout) { if (!have_addr) { db_printf("usage: show callout \n"); return; } _show_callout((struct callout *)addr); } #endif /* DDB */ Index: head/sys/kern/kern_xxx.c =================================================================== --- head/sys/kern/kern_xxx.c (revision 305831) +++ head/sys/kern/kern_xxx.c (revision 305832) @@ -1,471 +1,471 @@ /*- * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_xxx.c 8.2 (Berkeley) 11/14/93 */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #include #include #include #include #include #include #include #include #include #include #include #include #if defined(COMPAT_43) #ifndef _SYS_SYSPROTO_H_ struct gethostname_args { char *hostname; u_int len; }; #endif /* ARGSUSED */ int ogethostname(td, uap) struct thread *td; struct gethostname_args *uap; { int name[2]; size_t len = uap->len; name[0] = CTL_KERN; name[1] = KERN_HOSTNAME; return (userland_sysctl(td, name, 2, uap->hostname, &len, 1, 0, 0, 0, 0)); } #ifndef _SYS_SYSPROTO_H_ struct sethostname_args { char *hostname; u_int len; }; #endif /* ARGSUSED */ int osethostname(td, uap) struct thread *td; register struct sethostname_args *uap; { int name[2]; name[0] = CTL_KERN; name[1] = KERN_HOSTNAME; return (userland_sysctl(td, name, 2, 0, 0, 0, uap->hostname, uap->len, 0, 0)); } #ifndef _SYS_SYSPROTO_H_ struct ogethostid_args { int dummy; }; #endif /* ARGSUSED */ int ogethostid(td, uap) struct thread *td; struct ogethostid_args *uap; { size_t len = sizeof(long); int name[2]; name[0] = CTL_KERN; name[1] = KERN_HOSTID; return (kernel_sysctl(td, name, 2, (long *)td->td_retval, &len, NULL, 0, NULL, 0)); } #endif /* COMPAT_43 */ #ifdef COMPAT_43 #ifndef _SYS_SYSPROTO_H_ struct osethostid_args { long hostid; }; #endif /* ARGSUSED */ int osethostid(td, uap) struct thread *td; struct osethostid_args *uap; { int name[2]; name[0] = CTL_KERN; name[1] = KERN_HOSTID; return (kernel_sysctl(td, name, 2, NULL, NULL, &uap->hostid, sizeof(uap->hostid), NULL, 0)); } int oquota(td, uap) struct thread *td; struct oquota_args *uap; { return (ENOSYS); } #define KINFO_PROC (0<<8) #define KINFO_RT (1<<8) #define KINFO_VNODE (2<<8) #define KINFO_FILE (3<<8) #define KINFO_METER (4<<8) #define KINFO_LOADAVG (5<<8) #define KINFO_CLOCKRATE (6<<8) /* Non-standard BSDI extension - only present on their 4.3 net-2 releases */ #define KINFO_BSDI_SYSINFO (101<<8) /* * XXX this is bloat, but I hope it's better here than on the potentially * limited kernel stack... -Peter */ static struct { int bsdi_machine; /* "i386" on BSD/386 */ /* ^^^ this is an offset to the string, relative to the struct start */ char *pad0; long pad1; long pad2; long pad3; u_long pad4; u_long pad5; u_long pad6; int bsdi_ostype; /* "BSD/386" on BSD/386 */ int bsdi_osrelease; /* "1.1" on BSD/386 */ long pad7; long pad8; char *pad9; long pad10; long pad11; int pad12; long pad13; quad_t pad14; long pad15; struct timeval pad16; /* we dont set this, because BSDI's uname used gethostname() instead */ int bsdi_hostname; /* hostname on BSD/386 */ /* the actual string data is appended here */ } bsdi_si; /* * this data is appended to the end of the bsdi_si structure during copyout. * The "char *" offsets are relative to the base of the bsdi_si struct. * This contains "FreeBSD\02.0-BUILT-nnnnnn\0i386\0", and these strings * should not exceed the length of the buffer here... (or else!! :-) */ static char bsdi_strings[80]; /* It had better be less than this! */ #ifndef _SYS_SYSPROTO_H_ struct getkerninfo_args { int op; char *where; size_t *size; int arg; }; #endif int ogetkerninfo(struct thread *td, struct getkerninfo_args *uap) { int error, name[6]; size_t size; u_int needed = 0; switch (uap->op & 0xff00) { case KINFO_RT: name[0] = CTL_NET; name[1] = PF_ROUTE; name[2] = 0; name[3] = (uap->op & 0xff0000) >> 16; name[4] = uap->op & 0xff; name[5] = uap->arg; error = userland_sysctl(td, name, 6, uap->where, uap->size, 0, 0, 0, &size, 0); break; case KINFO_VNODE: name[0] = CTL_KERN; name[1] = KERN_VNODE; error = userland_sysctl(td, name, 2, uap->where, uap->size, 0, 0, 0, &size, 0); break; case KINFO_PROC: name[0] = CTL_KERN; name[1] = KERN_PROC; name[2] = uap->op & 0xff; name[3] = uap->arg; error = userland_sysctl(td, name, 4, uap->where, uap->size, 0, 0, 0, &size, 0); break; case KINFO_FILE: name[0] = CTL_KERN; name[1] = KERN_FILE; error = userland_sysctl(td, name, 2, uap->where, uap->size, 0, 0, 0, &size, 0); break; case KINFO_METER: name[0] = CTL_VM; name[1] = VM_TOTAL; error = userland_sysctl(td, name, 2, uap->where, uap->size, 0, 0, 0, &size, 0); break; case KINFO_LOADAVG: name[0] = CTL_VM; name[1] = VM_LOADAVG; error = userland_sysctl(td, name, 2, uap->where, uap->size, 0, 0, 0, &size, 0); break; case KINFO_CLOCKRATE: name[0] = CTL_KERN; name[1] = KERN_CLOCKRATE; error = userland_sysctl(td, name, 2, uap->where, uap->size, 0, 0, 0, &size, 0); break; case KINFO_BSDI_SYSINFO: { /* * this is pretty crude, but it's just enough for uname() * from BSDI's 1.x libc to work. * * *size gives the size of the buffer before the call, and * the amount of data copied after a successful call. * If successful, the return value is the amount of data * available, which can be larger than *size. * * BSDI's 2.x product apparently fails with ENOMEM if *size * is too small. */ u_int left; char *s; bzero((char *)&bsdi_si, sizeof(bsdi_si)); bzero(bsdi_strings, sizeof(bsdi_strings)); s = bsdi_strings; bsdi_si.bsdi_ostype = (s - bsdi_strings) + sizeof(bsdi_si); strcpy(s, ostype); s += strlen(s) + 1; bsdi_si.bsdi_osrelease = (s - bsdi_strings) + sizeof(bsdi_si); strcpy(s, osrelease); s += strlen(s) + 1; bsdi_si.bsdi_machine = (s - bsdi_strings) + sizeof(bsdi_si); strcpy(s, machine); s += strlen(s) + 1; needed = sizeof(bsdi_si) + (s - bsdi_strings); if ((uap->where == NULL) || (uap->size == NULL)) { /* process is asking how much buffer to supply.. */ size = needed; error = 0; break; } if ((error = copyin(uap->size, &size, sizeof(size))) != 0) break; /* if too much buffer supplied, trim it down */ if (size > needed) size = needed; /* how much of the buffer is remaining */ left = size; if ((error = copyout((char *)&bsdi_si, uap->where, left)) != 0) break; /* is there any point in continuing? */ if (left > sizeof(bsdi_si)) { left -= sizeof(bsdi_si); error = copyout(&bsdi_strings, uap->where + sizeof(bsdi_si), left); } break; } default: error = EOPNOTSUPP; break; } if (error == 0) { td->td_retval[0] = needed ? needed : size; if (uap->size) { error = copyout(&size, uap->size, sizeof(size)); } } return (error); } #endif /* COMPAT_43 */ #ifdef COMPAT_FREEBSD4 /* * This is the FreeBSD-1.1 compatible uname(2) interface. These days it is * done in libc as a wrapper around a bunch of sysctl's. This must maintain * the old 1.1 binary ABI. */ #if SYS_NMLN != 32 #error "FreeBSD-1.1 uname syscall has been broken" #endif #ifndef _SYS_SYSPROTO_H_ struct uname_args { struct utsname *name; }; #endif /* ARGSUSED */ int freebsd4_uname(struct thread *td, struct freebsd4_uname_args *uap) { int name[2], error; size_t len; char *s, *us; name[0] = CTL_KERN; name[1] = KERN_OSTYPE; len = sizeof (uap->name->sysname); error = userland_sysctl(td, name, 2, uap->name->sysname, &len, 1, 0, 0, 0, 0); if (error) return (error); subyte( uap->name->sysname + sizeof(uap->name->sysname) - 1, 0); name[1] = KERN_HOSTNAME; len = sizeof uap->name->nodename; error = userland_sysctl(td, name, 2, uap->name->nodename, &len, 1, 0, 0, 0, 0); if (error) return (error); subyte( uap->name->nodename + sizeof(uap->name->nodename) - 1, 0); name[1] = KERN_OSRELEASE; len = sizeof uap->name->release; error = userland_sysctl(td, name, 2, uap->name->release, &len, 1, 0, 0, 0, 0); if (error) return (error); subyte( uap->name->release + sizeof(uap->name->release) - 1, 0); /* name = KERN_VERSION; len = sizeof uap->name->version; error = userland_sysctl(td, name, 2, uap->name->version, &len, 1, 0, 0, 0, 0); if (error) return (error); subyte( uap->name->version + sizeof(uap->name->version) - 1, 0); */ /* * this stupid hackery to make the version field look like FreeBSD 1.1 */ for(s = version; *s && *s != '#'; s++); for(us = uap->name->version; *s && *s != ':'; s++) { error = subyte( us++, *s); if (error) return (error); } error = subyte( us++, 0); if (error) return (error); name[0] = CTL_HW; name[1] = HW_MACHINE; len = sizeof uap->name->machine; error = userland_sysctl(td, name, 2, uap->name->machine, &len, 1, 0, 0, 0, 0); if (error) return (error); subyte( uap->name->machine + sizeof(uap->name->machine) - 1, 0); return (0); } #ifndef _SYS_SYSPROTO_H_ struct getdomainname_args { char *domainname; int len; }; #endif /* ARGSUSED */ int freebsd4_getdomainname(struct thread *td, struct freebsd4_getdomainname_args *uap) { int name[2]; size_t len = uap->len; name[0] = CTL_KERN; name[1] = KERN_NISDOMAINNAME; return (userland_sysctl(td, name, 2, uap->domainname, &len, 1, 0, 0, 0, 0)); } #ifndef _SYS_SYSPROTO_H_ struct setdomainname_args { char *domainname; int len; }; #endif /* ARGSUSED */ int freebsd4_setdomainname(struct thread *td, struct freebsd4_setdomainname_args *uap) { int name[2]; name[0] = CTL_KERN; name[1] = KERN_NISDOMAINNAME; return (userland_sysctl(td, name, 2, 0, 0, 0, uap->domainname, uap->len, 0, 0)); } #endif /* COMPAT_FREEBSD4 */ Index: head/sys/kern/sched_4bsd.c =================================================================== --- head/sys/kern/sched_4bsd.c (revision 305831) +++ head/sys/kern/sched_4bsd.c (revision 305832) @@ -1,1797 +1,1797 @@ /*- * Copyright (c) 1982, 1986, 1990, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_hwpmc_hooks.h" #include "opt_sched.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef HWPMC_HOOKS #include #endif #ifdef KDTRACE_HOOKS #include int dtrace_vtime_active; dtrace_vtime_switch_func_t dtrace_vtime_switch_func; #endif /* * INVERSE_ESTCPU_WEIGHT is only suitable for statclock() frequencies in * the range 100-256 Hz (approximately). */ #define ESTCPULIM(e) \ min((e), INVERSE_ESTCPU_WEIGHT * (NICE_WEIGHT * (PRIO_MAX - PRIO_MIN) - \ RQ_PPQ) + INVERSE_ESTCPU_WEIGHT - 1) #ifdef SMP #define INVERSE_ESTCPU_WEIGHT (8 * smp_cpus) #else #define INVERSE_ESTCPU_WEIGHT 8 /* 1 / (priorities per estcpu level). */ #endif #define NICE_WEIGHT 1 /* Priorities per nice level. */ #define TS_NAME_LEN (MAXCOMLEN + sizeof(" td ") + sizeof(__XSTRING(UINT_MAX))) /* * The schedulable entity that runs a context. * This is an extension to the thread structure and is tailored to * the requirements of this scheduler. * All fields are protected by the scheduler lock. */ struct td_sched { fixpt_t ts_pctcpu; /* %cpu during p_swtime. */ u_int ts_estcpu; /* Estimated cpu utilization. */ int ts_cpticks; /* Ticks of cpu time. */ int ts_slptime; /* Seconds !RUNNING. */ int ts_slice; /* Remaining part of time slice. */ int ts_flags; struct runq *ts_runq; /* runq the thread is currently on */ #ifdef KTR char ts_name[TS_NAME_LEN]; #endif }; /* flags kept in td_flags */ #define TDF_DIDRUN TDF_SCHED0 /* thread actually ran. */ #define TDF_BOUND TDF_SCHED1 /* Bound to one CPU. */ #define TDF_SLICEEND TDF_SCHED2 /* Thread time slice is over. */ /* flags kept in ts_flags */ #define TSF_AFFINITY 0x0001 /* Has a non-"full" CPU set. */ #define SKE_RUNQ_PCPU(ts) \ ((ts)->ts_runq != 0 && (ts)->ts_runq != &runq) #define THREAD_CAN_SCHED(td, cpu) \ CPU_ISSET((cpu), &(td)->td_cpuset->cs_mask) _Static_assert(sizeof(struct thread) + sizeof(struct td_sched) <= sizeof(struct thread0_storage), "increase struct thread0_storage.t0st_sched size"); static struct mtx sched_lock; static int realstathz = 127; /* stathz is sometimes 0 and run off of hz. */ static int sched_tdcnt; /* Total runnable threads in the system. */ static int sched_slice = 12; /* Thread run time before rescheduling. */ static void setup_runqs(void); static void schedcpu(void); static void schedcpu_thread(void); static void sched_priority(struct thread *td, u_char prio); static void sched_setup(void *dummy); static void maybe_resched(struct thread *td); static void updatepri(struct thread *td); static void resetpriority(struct thread *td); static void resetpriority_thread(struct thread *td); #ifdef SMP static int sched_pickcpu(struct thread *td); static int forward_wakeup(int cpunum); static void kick_other_cpu(int pri, int cpuid); #endif static struct kproc_desc sched_kp = { "schedcpu", schedcpu_thread, NULL }; SYSINIT(schedcpu, SI_SUB_LAST, SI_ORDER_FIRST, kproc_start, &sched_kp); SYSINIT(sched_setup, SI_SUB_RUN_QUEUE, SI_ORDER_FIRST, sched_setup, NULL); static void sched_initticks(void *dummy); SYSINIT(sched_initticks, SI_SUB_CLOCKS, SI_ORDER_THIRD, sched_initticks, NULL); /* * Global run queue. */ static struct runq runq; #ifdef SMP /* * Per-CPU run queues */ static struct runq runq_pcpu[MAXCPU]; long runq_length[MAXCPU]; static cpuset_t idle_cpus_mask; #endif struct pcpuidlestat { u_int idlecalls; u_int oldidlecalls; }; static DPCPU_DEFINE(struct pcpuidlestat, idlestat); static void setup_runqs(void) { #ifdef SMP int i; for (i = 0; i < MAXCPU; ++i) runq_init(&runq_pcpu[i]); #endif runq_init(&runq); } static int sysctl_kern_quantum(SYSCTL_HANDLER_ARGS) { int error, new_val, period; period = 1000000 / realstathz; new_val = period * sched_slice; error = sysctl_handle_int(oidp, &new_val, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (new_val <= 0) return (EINVAL); sched_slice = imax(1, (new_val + period / 2) / period); hogticks = imax(1, (2 * hz * sched_slice + realstathz / 2) / realstathz); return (0); } SYSCTL_NODE(_kern, OID_AUTO, sched, CTLFLAG_RD, 0, "Scheduler"); SYSCTL_STRING(_kern_sched, OID_AUTO, name, CTLFLAG_RD, "4BSD", 0, "Scheduler name"); SYSCTL_PROC(_kern_sched, OID_AUTO, quantum, CTLTYPE_INT | CTLFLAG_RW, NULL, 0, sysctl_kern_quantum, "I", "Quantum for timeshare threads in microseconds"); SYSCTL_INT(_kern_sched, OID_AUTO, slice, CTLFLAG_RW, &sched_slice, 0, "Quantum for timeshare threads in stathz ticks"); #ifdef SMP /* Enable forwarding of wakeups to all other cpus */ static SYSCTL_NODE(_kern_sched, OID_AUTO, ipiwakeup, CTLFLAG_RD, NULL, "Kernel SMP"); static int runq_fuzz = 1; SYSCTL_INT(_kern_sched, OID_AUTO, runq_fuzz, CTLFLAG_RW, &runq_fuzz, 0, ""); static int forward_wakeup_enabled = 1; SYSCTL_INT(_kern_sched_ipiwakeup, OID_AUTO, enabled, CTLFLAG_RW, &forward_wakeup_enabled, 0, "Forwarding of wakeup to idle CPUs"); static int forward_wakeups_requested = 0; SYSCTL_INT(_kern_sched_ipiwakeup, OID_AUTO, requested, CTLFLAG_RD, &forward_wakeups_requested, 0, "Requests for Forwarding of wakeup to idle CPUs"); static int forward_wakeups_delivered = 0; SYSCTL_INT(_kern_sched_ipiwakeup, OID_AUTO, delivered, CTLFLAG_RD, &forward_wakeups_delivered, 0, "Completed Forwarding of wakeup to idle CPUs"); static int forward_wakeup_use_mask = 1; SYSCTL_INT(_kern_sched_ipiwakeup, OID_AUTO, usemask, CTLFLAG_RW, &forward_wakeup_use_mask, 0, "Use the mask of idle cpus"); static int forward_wakeup_use_loop = 0; SYSCTL_INT(_kern_sched_ipiwakeup, OID_AUTO, useloop, CTLFLAG_RW, &forward_wakeup_use_loop, 0, "Use a loop to find idle cpus"); #endif #if 0 static int sched_followon = 0; SYSCTL_INT(_kern_sched, OID_AUTO, followon, CTLFLAG_RW, &sched_followon, 0, "allow threads to share a quantum"); #endif SDT_PROVIDER_DEFINE(sched); SDT_PROBE_DEFINE3(sched, , , change__pri, "struct thread *", "struct proc *", "uint8_t"); SDT_PROBE_DEFINE3(sched, , , dequeue, "struct thread *", "struct proc *", "void *"); SDT_PROBE_DEFINE4(sched, , , enqueue, "struct thread *", "struct proc *", "void *", "int"); SDT_PROBE_DEFINE4(sched, , , lend__pri, "struct thread *", "struct proc *", "uint8_t", "struct thread *"); SDT_PROBE_DEFINE2(sched, , , load__change, "int", "int"); SDT_PROBE_DEFINE2(sched, , , off__cpu, "struct thread *", "struct proc *"); SDT_PROBE_DEFINE(sched, , , on__cpu); SDT_PROBE_DEFINE(sched, , , remain__cpu); SDT_PROBE_DEFINE2(sched, , , surrender, "struct thread *", "struct proc *"); static __inline void sched_load_add(void) { sched_tdcnt++; KTR_COUNTER0(KTR_SCHED, "load", "global load", sched_tdcnt); SDT_PROBE2(sched, , , load__change, NOCPU, sched_tdcnt); } static __inline void sched_load_rem(void) { sched_tdcnt--; KTR_COUNTER0(KTR_SCHED, "load", "global load", sched_tdcnt); SDT_PROBE2(sched, , , load__change, NOCPU, sched_tdcnt); } /* * Arrange to reschedule if necessary, taking the priorities and * schedulers into account. */ static void maybe_resched(struct thread *td) { THREAD_LOCK_ASSERT(td, MA_OWNED); if (td->td_priority < curthread->td_priority) curthread->td_flags |= TDF_NEEDRESCHED; } /* * This function is called when a thread is about to be put on run queue * because it has been made runnable or its priority has been adjusted. It * determines if the new thread should be immediately preempted to. If so, * it switches to it and eventually returns true. If not, it returns false * so that the caller may place the thread on an appropriate run queue. */ int maybe_preempt(struct thread *td) { #ifdef PREEMPTION struct thread *ctd; int cpri, pri; /* * The new thread should not preempt the current thread if any of the * following conditions are true: * * - The kernel is in the throes of crashing (panicstr). * - The current thread has a higher (numerically lower) or * equivalent priority. Note that this prevents curthread from * trying to preempt to itself. * - It is too early in the boot for context switches (cold is set). * - The current thread has an inhibitor set or is in the process of * exiting. In this case, the current thread is about to switch * out anyways, so there's no point in preempting. If we did, * the current thread would not be properly resumed as well, so * just avoid that whole landmine. * - If the new thread's priority is not a realtime priority and * the current thread's priority is not an idle priority and * FULL_PREEMPTION is disabled. * * If all of these conditions are false, but the current thread is in * a nested critical section, then we have to defer the preemption * until we exit the critical section. Otherwise, switch immediately * to the new thread. */ ctd = curthread; THREAD_LOCK_ASSERT(td, MA_OWNED); KASSERT((td->td_inhibitors == 0), ("maybe_preempt: trying to run inhibited thread")); pri = td->td_priority; cpri = ctd->td_priority; if (panicstr != NULL || pri >= cpri || cold /* || dumping */ || TD_IS_INHIBITED(ctd)) return (0); #ifndef FULL_PREEMPTION if (pri > PRI_MAX_ITHD && cpri < PRI_MIN_IDLE) return (0); #endif if (ctd->td_critnest > 1) { CTR1(KTR_PROC, "maybe_preempt: in critical section %d", ctd->td_critnest); ctd->td_owepreempt = 1; return (0); } /* * Thread is runnable but not yet put on system run queue. */ MPASS(ctd->td_lock == td->td_lock); MPASS(TD_ON_RUNQ(td)); TD_SET_RUNNING(td); CTR3(KTR_PROC, "preempting to thread %p (pid %d, %s)\n", td, td->td_proc->p_pid, td->td_name); mi_switch(SW_INVOL | SW_PREEMPT | SWT_PREEMPT, td); /* * td's lock pointer may have changed. We have to return with it * locked. */ spinlock_enter(); thread_unlock(ctd); thread_lock(td); spinlock_exit(); return (1); #else return (0); #endif } /* * Constants for digital decay and forget: * 90% of (ts_estcpu) usage in 5 * loadav time * 95% of (ts_pctcpu) usage in 60 seconds (load insensitive) * Note that, as ps(1) mentions, this can let percentages * total over 100% (I've seen 137.9% for 3 processes). * * Note that schedclock() updates ts_estcpu and p_cpticks asynchronously. * * We wish to decay away 90% of ts_estcpu in (5 * loadavg) seconds. * That is, the system wants to compute a value of decay such * that the following for loop: * for (i = 0; i < (5 * loadavg); i++) * ts_estcpu *= decay; * will compute * ts_estcpu *= 0.1; * for all values of loadavg: * * Mathematically this loop can be expressed by saying: * decay ** (5 * loadavg) ~= .1 * * The system computes decay as: * decay = (2 * loadavg) / (2 * loadavg + 1) * * We wish to prove that the system's computation of decay * will always fulfill the equation: * decay ** (5 * loadavg) ~= .1 * * If we compute b as: * b = 2 * loadavg * then * decay = b / (b + 1) * * We now need to prove two things: * 1) Given factor ** (5 * loadavg) ~= .1, prove factor == b/(b+1) * 2) Given b/(b+1) ** power ~= .1, prove power == (5 * loadavg) * * Facts: * For x close to zero, exp(x) =~ 1 + x, since * exp(x) = 0! + x**1/1! + x**2/2! + ... . * therefore exp(-1/b) =~ 1 - (1/b) = (b-1)/b. * For x close to zero, ln(1+x) =~ x, since * ln(1+x) = x - x**2/2 + x**3/3 - ... -1 < x < 1 * therefore ln(b/(b+1)) = ln(1 - 1/(b+1)) =~ -1/(b+1). * ln(.1) =~ -2.30 * * Proof of (1): * Solve (factor)**(power) =~ .1 given power (5*loadav): * solving for factor, * ln(factor) =~ (-2.30/5*loadav), or * factor =~ exp(-1/((5/2.30)*loadav)) =~ exp(-1/(2*loadav)) = * exp(-1/b) =~ (b-1)/b =~ b/(b+1). QED * * Proof of (2): * Solve (factor)**(power) =~ .1 given factor == (b/(b+1)): * solving for power, * power*ln(b/(b+1)) =~ -2.30, or * power =~ 2.3 * (b + 1) = 4.6*loadav + 2.3 =~ 5*loadav. QED * * Actual power values for the implemented algorithm are as follows: * loadav: 1 2 3 4 * power: 5.68 10.32 14.94 19.55 */ /* calculations for digital decay to forget 90% of usage in 5*loadav sec */ #define loadfactor(loadav) (2 * (loadav)) #define decay_cpu(loadfac, cpu) (((loadfac) * (cpu)) / ((loadfac) + FSCALE)) /* decay 95% of `ts_pctcpu' in 60 seconds; see CCPU_SHIFT before changing */ static fixpt_t ccpu = 0.95122942450071400909 * FSCALE; /* exp(-1/20) */ SYSCTL_UINT(_kern, OID_AUTO, ccpu, CTLFLAG_RD, &ccpu, 0, ""); /* * If `ccpu' is not equal to `exp(-1/20)' and you still want to use the * faster/more-accurate formula, you'll have to estimate CCPU_SHIFT below * and possibly adjust FSHIFT in "param.h" so that (FSHIFT >= CCPU_SHIFT). * * To estimate CCPU_SHIFT for exp(-1/20), the following formula was used: * 1 - exp(-1/20) ~= 0.0487 ~= 0.0488 == 1 (fixed pt, *11* bits). * * If you don't want to bother with the faster/more-accurate formula, you * can set CCPU_SHIFT to (FSHIFT + 1) which will use a slower/less-accurate * (more general) method of calculating the %age of CPU used by a process. */ #define CCPU_SHIFT 11 /* * Recompute process priorities, every hz ticks. * MP-safe, called without the Giant mutex. */ /* ARGSUSED */ static void schedcpu(void) { register fixpt_t loadfac = loadfactor(averunnable.ldavg[0]); struct thread *td; struct proc *p; struct td_sched *ts; int awake; sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); if (p->p_state == PRS_NEW) { PROC_UNLOCK(p); continue; } FOREACH_THREAD_IN_PROC(p, td) { awake = 0; ts = td_get_sched(td); thread_lock(td); /* * Increment sleep time (if sleeping). We * ignore overflow, as above. */ /* * The td_sched slptimes are not touched in wakeup * because the thread may not HAVE everything in * memory? XXX I think this is out of date. */ if (TD_ON_RUNQ(td)) { awake = 1; td->td_flags &= ~TDF_DIDRUN; } else if (TD_IS_RUNNING(td)) { awake = 1; /* Do not clear TDF_DIDRUN */ } else if (td->td_flags & TDF_DIDRUN) { awake = 1; td->td_flags &= ~TDF_DIDRUN; } /* * ts_pctcpu is only for ps and ttyinfo(). */ ts->ts_pctcpu = (ts->ts_pctcpu * ccpu) >> FSHIFT; /* * If the td_sched has been idle the entire second, * stop recalculating its priority until * it wakes up. */ if (ts->ts_cpticks != 0) { #if (FSHIFT >= CCPU_SHIFT) ts->ts_pctcpu += (realstathz == 100) ? ((fixpt_t) ts->ts_cpticks) << (FSHIFT - CCPU_SHIFT) : 100 * (((fixpt_t) ts->ts_cpticks) << (FSHIFT - CCPU_SHIFT)) / realstathz; #else ts->ts_pctcpu += ((FSCALE - ccpu) * (ts->ts_cpticks * FSCALE / realstathz)) >> FSHIFT; #endif ts->ts_cpticks = 0; } /* * If there are ANY running threads in this process, * then don't count it as sleeping. * XXX: this is broken. */ if (awake) { if (ts->ts_slptime > 1) { /* * In an ideal world, this should not * happen, because whoever woke us * up from the long sleep should have * unwound the slptime and reset our * priority before we run at the stale * priority. Should KASSERT at some * point when all the cases are fixed. */ updatepri(td); } ts->ts_slptime = 0; } else ts->ts_slptime++; if (ts->ts_slptime > 1) { thread_unlock(td); continue; } ts->ts_estcpu = decay_cpu(loadfac, ts->ts_estcpu); resetpriority(td); resetpriority_thread(td); thread_unlock(td); } PROC_UNLOCK(p); } sx_sunlock(&allproc_lock); } /* * Main loop for a kthread that executes schedcpu once a second. */ static void schedcpu_thread(void) { for (;;) { schedcpu(); pause("-", hz); } } /* * Recalculate the priority of a process after it has slept for a while. * For all load averages >= 1 and max ts_estcpu of 255, sleeping for at * least six times the loadfactor will decay ts_estcpu to zero. */ static void updatepri(struct thread *td) { struct td_sched *ts; fixpt_t loadfac; unsigned int newcpu; ts = td_get_sched(td); loadfac = loadfactor(averunnable.ldavg[0]); if (ts->ts_slptime > 5 * loadfac) ts->ts_estcpu = 0; else { newcpu = ts->ts_estcpu; ts->ts_slptime--; /* was incremented in schedcpu() */ while (newcpu && --ts->ts_slptime) newcpu = decay_cpu(loadfac, newcpu); ts->ts_estcpu = newcpu; } } /* * Compute the priority of a process when running in user mode. * Arrange to reschedule if the resulting priority is better * than that of the current process. */ static void resetpriority(struct thread *td) { u_int newpriority; if (td->td_pri_class != PRI_TIMESHARE) return; newpriority = PUSER + td_get_sched(td)->ts_estcpu / INVERSE_ESTCPU_WEIGHT + NICE_WEIGHT * (td->td_proc->p_nice - PRIO_MIN); newpriority = min(max(newpriority, PRI_MIN_TIMESHARE), PRI_MAX_TIMESHARE); sched_user_prio(td, newpriority); } /* * Update the thread's priority when the associated process's user * priority changes. */ static void resetpriority_thread(struct thread *td) { /* Only change threads with a time sharing user priority. */ if (td->td_priority < PRI_MIN_TIMESHARE || td->td_priority > PRI_MAX_TIMESHARE) return; /* XXX the whole needresched thing is broken, but not silly. */ maybe_resched(td); sched_prio(td, td->td_user_pri); } /* ARGSUSED */ static void sched_setup(void *dummy) { setup_runqs(); /* Account for thread0. */ sched_load_add(); } /* * This routine determines time constants after stathz and hz are setup. */ static void sched_initticks(void *dummy) { realstathz = stathz ? stathz : hz; sched_slice = realstathz / 10; /* ~100ms */ hogticks = imax(1, (2 * hz * sched_slice + realstathz / 2) / realstathz); } /* External interfaces start here */ /* * Very early in the boot some setup of scheduler-specific * parts of proc0 and of some scheduler resources needs to be done. * Called from: * proc0_init() */ void schedinit(void) { /* * Set up the scheduler specific parts of thread0. */ thread0.td_lock = &sched_lock; td_get_sched(&thread0)->ts_slice = sched_slice; mtx_init(&sched_lock, "sched lock", NULL, MTX_SPIN | MTX_RECURSE); } int sched_runnable(void) { #ifdef SMP return runq_check(&runq) + runq_check(&runq_pcpu[PCPU_GET(cpuid)]); #else return runq_check(&runq); #endif } int sched_rr_interval(void) { /* Convert sched_slice from stathz to hz. */ return (imax(1, (sched_slice * hz + realstathz / 2) / realstathz)); } /* * We adjust the priority of the current process. The priority of a * process gets worse as it accumulates CPU time. The cpu usage * estimator (ts_estcpu) is increased here. resetpriority() will * compute a different priority each time ts_estcpu increases by * INVERSE_ESTCPU_WEIGHT (until PRI_MAX_TIMESHARE is reached). The * cpu usage estimator ramps up quite quickly when the process is * running (linearly), and decays away exponentially, at a rate which * is proportionally slower when the system is busy. The basic * principle is that the system will 90% forget that the process used * a lot of CPU time in 5 * loadav seconds. This causes the system to * favor processes which haven't run much recently, and to round-robin * among other processes. */ void sched_clock(struct thread *td) { struct pcpuidlestat *stat; struct td_sched *ts; THREAD_LOCK_ASSERT(td, MA_OWNED); ts = td_get_sched(td); ts->ts_cpticks++; ts->ts_estcpu = ESTCPULIM(ts->ts_estcpu + 1); if ((ts->ts_estcpu % INVERSE_ESTCPU_WEIGHT) == 0) { resetpriority(td); resetpriority_thread(td); } /* * Force a context switch if the current thread has used up a full * time slice (default is 100ms). */ if (!TD_IS_IDLETHREAD(td) && --ts->ts_slice <= 0) { ts->ts_slice = sched_slice; td->td_flags |= TDF_NEEDRESCHED | TDF_SLICEEND; } stat = DPCPU_PTR(idlestat); stat->oldidlecalls = stat->idlecalls; stat->idlecalls = 0; } /* * Charge child's scheduling CPU usage to parent. */ void sched_exit(struct proc *p, struct thread *td) { KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "proc exit", "prio:%d", td->td_priority); PROC_LOCK_ASSERT(p, MA_OWNED); sched_exit_thread(FIRST_THREAD_IN_PROC(p), td); } void sched_exit_thread(struct thread *td, struct thread *child) { KTR_STATE1(KTR_SCHED, "thread", sched_tdname(child), "exit", "prio:%d", child->td_priority); thread_lock(td); td_get_sched(td)->ts_estcpu = ESTCPULIM(td_get_sched(td)->ts_estcpu + td_get_sched(child)->ts_estcpu); thread_unlock(td); thread_lock(child); if ((child->td_flags & TDF_NOLOAD) == 0) sched_load_rem(); thread_unlock(child); } void sched_fork(struct thread *td, struct thread *childtd) { sched_fork_thread(td, childtd); } void sched_fork_thread(struct thread *td, struct thread *childtd) { struct td_sched *ts, *tsc; childtd->td_oncpu = NOCPU; childtd->td_lastcpu = NOCPU; childtd->td_lock = &sched_lock; childtd->td_cpuset = cpuset_ref(td->td_cpuset); childtd->td_priority = childtd->td_base_pri; ts = td_get_sched(childtd); bzero(ts, sizeof(*ts)); tsc = td_get_sched(td); ts->ts_estcpu = tsc->ts_estcpu; ts->ts_flags |= (tsc->ts_flags & TSF_AFFINITY); ts->ts_slice = 1; } void sched_nice(struct proc *p, int nice) { struct thread *td; PROC_LOCK_ASSERT(p, MA_OWNED); p->p_nice = nice; FOREACH_THREAD_IN_PROC(p, td) { thread_lock(td); resetpriority(td); resetpriority_thread(td); thread_unlock(td); } } void sched_class(struct thread *td, int class) { THREAD_LOCK_ASSERT(td, MA_OWNED); td->td_pri_class = class; } /* * Adjust the priority of a thread. */ static void sched_priority(struct thread *td, u_char prio) { KTR_POINT3(KTR_SCHED, "thread", sched_tdname(td), "priority change", "prio:%d", td->td_priority, "new prio:%d", prio, KTR_ATTR_LINKED, sched_tdname(curthread)); SDT_PROBE3(sched, , , change__pri, td, td->td_proc, prio); if (td != curthread && prio > td->td_priority) { KTR_POINT3(KTR_SCHED, "thread", sched_tdname(curthread), "lend prio", "prio:%d", td->td_priority, "new prio:%d", prio, KTR_ATTR_LINKED, sched_tdname(td)); SDT_PROBE4(sched, , , lend__pri, td, td->td_proc, prio, curthread); } THREAD_LOCK_ASSERT(td, MA_OWNED); if (td->td_priority == prio) return; td->td_priority = prio; if (TD_ON_RUNQ(td) && td->td_rqindex != (prio / RQ_PPQ)) { sched_rem(td); sched_add(td, SRQ_BORING); } } /* * Update a thread's priority when it is lent another thread's * priority. */ void sched_lend_prio(struct thread *td, u_char prio) { td->td_flags |= TDF_BORROWING; sched_priority(td, prio); } /* * Restore a thread's priority when priority propagation is * over. The prio argument is the minimum priority the thread * needs to have to satisfy other possible priority lending * requests. If the thread's regulary priority is less * important than prio the thread will keep a priority boost * of prio. */ void sched_unlend_prio(struct thread *td, u_char prio) { u_char base_pri; if (td->td_base_pri >= PRI_MIN_TIMESHARE && td->td_base_pri <= PRI_MAX_TIMESHARE) base_pri = td->td_user_pri; else base_pri = td->td_base_pri; if (prio >= base_pri) { td->td_flags &= ~TDF_BORROWING; sched_prio(td, base_pri); } else sched_lend_prio(td, prio); } void sched_prio(struct thread *td, u_char prio) { u_char oldprio; /* First, update the base priority. */ td->td_base_pri = prio; /* * If the thread is borrowing another thread's priority, don't ever * lower the priority. */ if (td->td_flags & TDF_BORROWING && td->td_priority < prio) return; /* Change the real priority. */ oldprio = td->td_priority; sched_priority(td, prio); /* * If the thread is on a turnstile, then let the turnstile update * its state. */ if (TD_ON_LOCK(td) && oldprio != prio) turnstile_adjust(td, oldprio); } void sched_user_prio(struct thread *td, u_char prio) { THREAD_LOCK_ASSERT(td, MA_OWNED); td->td_base_user_pri = prio; if (td->td_lend_user_pri <= prio) return; td->td_user_pri = prio; } void sched_lend_user_prio(struct thread *td, u_char prio) { THREAD_LOCK_ASSERT(td, MA_OWNED); td->td_lend_user_pri = prio; td->td_user_pri = min(prio, td->td_base_user_pri); if (td->td_priority > td->td_user_pri) sched_prio(td, td->td_user_pri); else if (td->td_priority != td->td_user_pri) td->td_flags |= TDF_NEEDRESCHED; } void sched_sleep(struct thread *td, int pri) { THREAD_LOCK_ASSERT(td, MA_OWNED); td->td_slptick = ticks; td_get_sched(td)->ts_slptime = 0; if (pri != 0 && PRI_BASE(td->td_pri_class) == PRI_TIMESHARE) sched_prio(td, pri); if (TD_IS_SUSPENDED(td) || pri >= PSOCK) td->td_flags |= TDF_CANSWAP; } void sched_switch(struct thread *td, struct thread *newtd, int flags) { struct mtx *tmtx; struct td_sched *ts; struct proc *p; int preempted; tmtx = NULL; ts = td_get_sched(td); p = td->td_proc; THREAD_LOCK_ASSERT(td, MA_OWNED); /* * Switch to the sched lock to fix things up and pick * a new thread. * Block the td_lock in order to avoid breaking the critical path. */ if (td->td_lock != &sched_lock) { mtx_lock_spin(&sched_lock); tmtx = thread_lock_block(td); } if ((td->td_flags & TDF_NOLOAD) == 0) sched_load_rem(); td->td_lastcpu = td->td_oncpu; preempted = !((td->td_flags & TDF_SLICEEND) || (flags & SWT_RELINQUISH)); td->td_flags &= ~(TDF_NEEDRESCHED | TDF_SLICEEND); td->td_owepreempt = 0; td->td_oncpu = NOCPU; /* * At the last moment, if this thread is still marked RUNNING, * then put it back on the run queue as it has not been suspended * or stopped or any thing else similar. We never put the idle * threads on the run queue, however. */ if (td->td_flags & TDF_IDLETD) { TD_SET_CAN_RUN(td); #ifdef SMP CPU_CLR(PCPU_GET(cpuid), &idle_cpus_mask); #endif } else { if (TD_IS_RUNNING(td)) { /* Put us back on the run queue. */ sched_add(td, preempted ? SRQ_OURSELF|SRQ_YIELDING|SRQ_PREEMPTED : SRQ_OURSELF|SRQ_YIELDING); } } if (newtd) { /* * The thread we are about to run needs to be counted * as if it had been added to the run queue and selected. * It came from: * * A preemption * * An upcall * * A followon */ KASSERT((newtd->td_inhibitors == 0), ("trying to run inhibited thread")); newtd->td_flags |= TDF_DIDRUN; TD_SET_RUNNING(newtd); if ((newtd->td_flags & TDF_NOLOAD) == 0) sched_load_add(); } else { newtd = choosethread(); MPASS(newtd->td_lock == &sched_lock); } if (td != newtd) { #ifdef HWPMC_HOOKS if (PMC_PROC_IS_USING_PMCS(td->td_proc)) PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT); #endif SDT_PROBE2(sched, , , off__cpu, newtd, newtd->td_proc); /* I feel sleepy */ lock_profile_release_lock(&sched_lock.lock_object); #ifdef KDTRACE_HOOKS /* * If DTrace has set the active vtime enum to anything * other than INACTIVE (0), then it should have set the * function to call. */ if (dtrace_vtime_active) (*dtrace_vtime_switch_func)(newtd); #endif cpu_switch(td, newtd, tmtx != NULL ? tmtx : td->td_lock); lock_profile_obtain_lock_success(&sched_lock.lock_object, 0, 0, __FILE__, __LINE__); /* * Where am I? What year is it? * We are in the same thread that went to sleep above, * but any amount of time may have passed. All our context * will still be available as will local variables. * PCPU values however may have changed as we may have * changed CPU so don't trust cached values of them. * New threads will go to fork_exit() instead of here * so if you change things here you may need to change * things there too. * * If the thread above was exiting it will never wake * up again here, so either it has saved everything it * needed to, or the thread_wait() or wait() will * need to reap it. */ SDT_PROBE0(sched, , , on__cpu); #ifdef HWPMC_HOOKS if (PMC_PROC_IS_USING_PMCS(td->td_proc)) PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_IN); #endif } else SDT_PROBE0(sched, , , remain__cpu); #ifdef SMP if (td->td_flags & TDF_IDLETD) CPU_SET(PCPU_GET(cpuid), &idle_cpus_mask); #endif sched_lock.mtx_lock = (uintptr_t)td; td->td_oncpu = PCPU_GET(cpuid); MPASS(td->td_lock == &sched_lock); } void sched_wakeup(struct thread *td) { struct td_sched *ts; THREAD_LOCK_ASSERT(td, MA_OWNED); ts = td_get_sched(td); td->td_flags &= ~TDF_CANSWAP; if (ts->ts_slptime > 1) { updatepri(td); resetpriority(td); } td->td_slptick = 0; ts->ts_slptime = 0; ts->ts_slice = sched_slice; sched_add(td, SRQ_BORING); } #ifdef SMP static int forward_wakeup(int cpunum) { struct pcpu *pc; cpuset_t dontuse, map, map2; u_int id, me; int iscpuset; mtx_assert(&sched_lock, MA_OWNED); CTR0(KTR_RUNQ, "forward_wakeup()"); if ((!forward_wakeup_enabled) || (forward_wakeup_use_mask == 0 && forward_wakeup_use_loop == 0)) return (0); if (!smp_started || cold || panicstr) return (0); forward_wakeups_requested++; /* * Check the idle mask we received against what we calculated * before in the old version. */ me = PCPU_GET(cpuid); /* Don't bother if we should be doing it ourself. */ if (CPU_ISSET(me, &idle_cpus_mask) && (cpunum == NOCPU || me == cpunum)) return (0); CPU_SETOF(me, &dontuse); CPU_OR(&dontuse, &stopped_cpus); CPU_OR(&dontuse, &hlt_cpus_mask); CPU_ZERO(&map2); if (forward_wakeup_use_loop) { STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) { id = pc->pc_cpuid; if (!CPU_ISSET(id, &dontuse) && pc->pc_curthread == pc->pc_idlethread) { CPU_SET(id, &map2); } } } if (forward_wakeup_use_mask) { map = idle_cpus_mask; CPU_NAND(&map, &dontuse); /* If they are both on, compare and use loop if different. */ if (forward_wakeup_use_loop) { if (CPU_CMP(&map, &map2)) { printf("map != map2, loop method preferred\n"); map = map2; } } } else { map = map2; } /* If we only allow a specific CPU, then mask off all the others. */ if (cpunum != NOCPU) { KASSERT((cpunum <= mp_maxcpus),("forward_wakeup: bad cpunum.")); iscpuset = CPU_ISSET(cpunum, &map); if (iscpuset == 0) CPU_ZERO(&map); else CPU_SETOF(cpunum, &map); } if (!CPU_EMPTY(&map)) { forward_wakeups_delivered++; STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) { id = pc->pc_cpuid; if (!CPU_ISSET(id, &map)) continue; if (cpu_idle_wakeup(pc->pc_cpuid)) CPU_CLR(id, &map); } if (!CPU_EMPTY(&map)) ipi_selected(map, IPI_AST); return (1); } if (cpunum == NOCPU) printf("forward_wakeup: Idle processor not found\n"); return (0); } static void kick_other_cpu(int pri, int cpuid) { struct pcpu *pcpu; int cpri; pcpu = pcpu_find(cpuid); if (CPU_ISSET(cpuid, &idle_cpus_mask)) { forward_wakeups_delivered++; if (!cpu_idle_wakeup(cpuid)) ipi_cpu(cpuid, IPI_AST); return; } cpri = pcpu->pc_curthread->td_priority; if (pri >= cpri) return; #if defined(IPI_PREEMPTION) && defined(PREEMPTION) #if !defined(FULL_PREEMPTION) if (pri <= PRI_MAX_ITHD) #endif /* ! FULL_PREEMPTION */ { ipi_cpu(cpuid, IPI_PREEMPT); return; } #endif /* defined(IPI_PREEMPTION) && defined(PREEMPTION) */ pcpu->pc_curthread->td_flags |= TDF_NEEDRESCHED; ipi_cpu(cpuid, IPI_AST); return; } #endif /* SMP */ #ifdef SMP static int sched_pickcpu(struct thread *td) { int best, cpu; mtx_assert(&sched_lock, MA_OWNED); if (td->td_lastcpu != NOCPU && THREAD_CAN_SCHED(td, td->td_lastcpu)) best = td->td_lastcpu; else best = NOCPU; CPU_FOREACH(cpu) { if (!THREAD_CAN_SCHED(td, cpu)) continue; if (best == NOCPU) best = cpu; else if (runq_length[cpu] < runq_length[best]) best = cpu; } KASSERT(best != NOCPU, ("no valid CPUs")); return (best); } #endif void sched_add(struct thread *td, int flags) #ifdef SMP { cpuset_t tidlemsk; struct td_sched *ts; u_int cpu, cpuid; int forwarded = 0; int single_cpu = 0; ts = td_get_sched(td); THREAD_LOCK_ASSERT(td, MA_OWNED); KASSERT((td->td_inhibitors == 0), ("sched_add: trying to run inhibited thread")); KASSERT((TD_CAN_RUN(td) || TD_IS_RUNNING(td)), ("sched_add: bad thread state")); KASSERT(td->td_flags & TDF_INMEM, ("sched_add: thread swapped out")); KTR_STATE2(KTR_SCHED, "thread", sched_tdname(td), "runq add", "prio:%d", td->td_priority, KTR_ATTR_LINKED, sched_tdname(curthread)); KTR_POINT1(KTR_SCHED, "thread", sched_tdname(curthread), "wokeup", KTR_ATTR_LINKED, sched_tdname(td)); SDT_PROBE4(sched, , , enqueue, td, td->td_proc, NULL, flags & SRQ_PREEMPTED); /* * Now that the thread is moving to the run-queue, set the lock * to the scheduler's lock. */ if (td->td_lock != &sched_lock) { mtx_lock_spin(&sched_lock); thread_lock_set(td, &sched_lock); } TD_SET_RUNQ(td); /* * If SMP is started and the thread is pinned or otherwise limited to * a specific set of CPUs, queue the thread to a per-CPU run queue. * Otherwise, queue the thread to the global run queue. * * If SMP has not yet been started we must use the global run queue * as per-CPU state may not be initialized yet and we may crash if we * try to access the per-CPU run queues. */ if (smp_started && (td->td_pinned != 0 || td->td_flags & TDF_BOUND || ts->ts_flags & TSF_AFFINITY)) { if (td->td_pinned != 0) cpu = td->td_lastcpu; else if (td->td_flags & TDF_BOUND) { /* Find CPU from bound runq. */ KASSERT(SKE_RUNQ_PCPU(ts), ("sched_add: bound td_sched not on cpu runq")); cpu = ts->ts_runq - &runq_pcpu[0]; } else /* Find a valid CPU for our cpuset */ cpu = sched_pickcpu(td); ts->ts_runq = &runq_pcpu[cpu]; single_cpu = 1; CTR3(KTR_RUNQ, "sched_add: Put td_sched:%p(td:%p) on cpu%d runq", ts, td, cpu); } else { CTR2(KTR_RUNQ, "sched_add: adding td_sched:%p (td:%p) to gbl runq", ts, td); cpu = NOCPU; ts->ts_runq = &runq; } cpuid = PCPU_GET(cpuid); if (single_cpu && cpu != cpuid) { kick_other_cpu(td->td_priority, cpu); } else { if (!single_cpu) { tidlemsk = idle_cpus_mask; CPU_NAND(&tidlemsk, &hlt_cpus_mask); CPU_CLR(cpuid, &tidlemsk); if (!CPU_ISSET(cpuid, &idle_cpus_mask) && ((flags & SRQ_INTR) == 0) && !CPU_EMPTY(&tidlemsk)) forwarded = forward_wakeup(cpu); } if (!forwarded) { if ((flags & SRQ_YIELDING) == 0 && maybe_preempt(td)) return; else maybe_resched(td); } } if ((td->td_flags & TDF_NOLOAD) == 0) sched_load_add(); runq_add(ts->ts_runq, td, flags); if (cpu != NOCPU) runq_length[cpu]++; } #else /* SMP */ { struct td_sched *ts; ts = td_get_sched(td); THREAD_LOCK_ASSERT(td, MA_OWNED); KASSERT((td->td_inhibitors == 0), ("sched_add: trying to run inhibited thread")); KASSERT((TD_CAN_RUN(td) || TD_IS_RUNNING(td)), ("sched_add: bad thread state")); KASSERT(td->td_flags & TDF_INMEM, ("sched_add: thread swapped out")); KTR_STATE2(KTR_SCHED, "thread", sched_tdname(td), "runq add", "prio:%d", td->td_priority, KTR_ATTR_LINKED, sched_tdname(curthread)); KTR_POINT1(KTR_SCHED, "thread", sched_tdname(curthread), "wokeup", KTR_ATTR_LINKED, sched_tdname(td)); SDT_PROBE4(sched, , , enqueue, td, td->td_proc, NULL, flags & SRQ_PREEMPTED); /* * Now that the thread is moving to the run-queue, set the lock * to the scheduler's lock. */ if (td->td_lock != &sched_lock) { mtx_lock_spin(&sched_lock); thread_lock_set(td, &sched_lock); } TD_SET_RUNQ(td); CTR2(KTR_RUNQ, "sched_add: adding td_sched:%p (td:%p) to runq", ts, td); ts->ts_runq = &runq; /* * If we are yielding (on the way out anyhow) or the thread * being saved is US, then don't try be smart about preemption * or kicking off another CPU as it won't help and may hinder. * In the YIEDLING case, we are about to run whoever is being * put in the queue anyhow, and in the OURSELF case, we are * putting ourself on the run queue which also only happens * when we are about to yield. */ if ((flags & SRQ_YIELDING) == 0) { if (maybe_preempt(td)) return; } if ((td->td_flags & TDF_NOLOAD) == 0) sched_load_add(); runq_add(ts->ts_runq, td, flags); maybe_resched(td); } #endif /* SMP */ void sched_rem(struct thread *td) { struct td_sched *ts; ts = td_get_sched(td); KASSERT(td->td_flags & TDF_INMEM, ("sched_rem: thread swapped out")); KASSERT(TD_ON_RUNQ(td), ("sched_rem: thread not on run queue")); mtx_assert(&sched_lock, MA_OWNED); KTR_STATE2(KTR_SCHED, "thread", sched_tdname(td), "runq rem", "prio:%d", td->td_priority, KTR_ATTR_LINKED, sched_tdname(curthread)); SDT_PROBE3(sched, , , dequeue, td, td->td_proc, NULL); if ((td->td_flags & TDF_NOLOAD) == 0) sched_load_rem(); #ifdef SMP if (ts->ts_runq != &runq) runq_length[ts->ts_runq - runq_pcpu]--; #endif runq_remove(ts->ts_runq, td); TD_SET_CAN_RUN(td); } /* * Select threads to run. Note that running threads still consume a * slot. */ struct thread * sched_choose(void) { struct thread *td; struct runq *rq; mtx_assert(&sched_lock, MA_OWNED); #ifdef SMP struct thread *tdcpu; rq = &runq; td = runq_choose_fuzz(&runq, runq_fuzz); tdcpu = runq_choose(&runq_pcpu[PCPU_GET(cpuid)]); if (td == NULL || (tdcpu != NULL && tdcpu->td_priority < td->td_priority)) { CTR2(KTR_RUNQ, "choosing td %p from pcpu runq %d", tdcpu, PCPU_GET(cpuid)); td = tdcpu; rq = &runq_pcpu[PCPU_GET(cpuid)]; } else { CTR1(KTR_RUNQ, "choosing td_sched %p from main runq", td); } #else rq = &runq; td = runq_choose(&runq); #endif if (td) { #ifdef SMP if (td == tdcpu) runq_length[PCPU_GET(cpuid)]--; #endif runq_remove(rq, td); td->td_flags |= TDF_DIDRUN; KASSERT(td->td_flags & TDF_INMEM, ("sched_choose: thread swapped out")); return (td); } return (PCPU_GET(idlethread)); } void sched_preempt(struct thread *td) { SDT_PROBE2(sched, , , surrender, td, td->td_proc); thread_lock(td); if (td->td_critnest > 1) td->td_owepreempt = 1; else mi_switch(SW_INVOL | SW_PREEMPT | SWT_PREEMPT, NULL); thread_unlock(td); } void sched_userret(struct thread *td) { /* * XXX we cheat slightly on the locking here to avoid locking in * the usual case. Setting td_priority here is essentially an * incomplete workaround for not setting it properly elsewhere. * Now that some interrupt handlers are threads, not setting it * properly elsewhere can clobber it in the window between setting * it here and returning to user mode, so don't waste time setting * it perfectly here. */ KASSERT((td->td_flags & TDF_BORROWING) == 0, ("thread with borrowed priority returning to userland")); if (td->td_priority != td->td_user_pri) { thread_lock(td); td->td_priority = td->td_user_pri; td->td_base_pri = td->td_user_pri; thread_unlock(td); } } void sched_bind(struct thread *td, int cpu) { struct td_sched *ts; THREAD_LOCK_ASSERT(td, MA_OWNED|MA_NOTRECURSED); KASSERT(td == curthread, ("sched_bind: can only bind curthread")); ts = td_get_sched(td); td->td_flags |= TDF_BOUND; #ifdef SMP ts->ts_runq = &runq_pcpu[cpu]; if (PCPU_GET(cpuid) == cpu) return; mi_switch(SW_VOL, NULL); #endif } void sched_unbind(struct thread* td) { THREAD_LOCK_ASSERT(td, MA_OWNED); KASSERT(td == curthread, ("sched_unbind: can only bind curthread")); td->td_flags &= ~TDF_BOUND; } int sched_is_bound(struct thread *td) { THREAD_LOCK_ASSERT(td, MA_OWNED); return (td->td_flags & TDF_BOUND); } void sched_relinquish(struct thread *td) { thread_lock(td); mi_switch(SW_VOL | SWT_RELINQUISH, NULL); thread_unlock(td); } int sched_load(void) { return (sched_tdcnt); } int sched_sizeof_proc(void) { return (sizeof(struct proc)); } int sched_sizeof_thread(void) { return (sizeof(struct thread) + sizeof(struct td_sched)); } fixpt_t sched_pctcpu(struct thread *td) { struct td_sched *ts; THREAD_LOCK_ASSERT(td, MA_OWNED); ts = td_get_sched(td); return (ts->ts_pctcpu); } #ifdef RACCT /* * Calculates the contribution to the thread cpu usage for the latest * (unfinished) second. */ fixpt_t sched_pctcpu_delta(struct thread *td) { struct td_sched *ts; fixpt_t delta; int realstathz; THREAD_LOCK_ASSERT(td, MA_OWNED); ts = td_get_sched(td); delta = 0; realstathz = stathz ? stathz : hz; if (ts->ts_cpticks != 0) { #if (FSHIFT >= CCPU_SHIFT) delta = (realstathz == 100) ? ((fixpt_t) ts->ts_cpticks) << (FSHIFT - CCPU_SHIFT) : 100 * (((fixpt_t) ts->ts_cpticks) << (FSHIFT - CCPU_SHIFT)) / realstathz; #else delta = ((FSCALE - ccpu) * (ts->ts_cpticks * FSCALE / realstathz)) >> FSHIFT; #endif } return (delta); } #endif u_int sched_estcpu(struct thread *td) { return (td_get_sched(td)->ts_estcpu); } /* * The actual idle process. */ void sched_idletd(void *dummy) { struct pcpuidlestat *stat; THREAD_NO_SLEEPING(); stat = DPCPU_PTR(idlestat); for (;;) { mtx_assert(&Giant, MA_NOTOWNED); while (sched_runnable() == 0) { cpu_idle(stat->idlecalls + stat->oldidlecalls > 64); stat->idlecalls++; } mtx_lock_spin(&sched_lock); mi_switch(SW_VOL | SWT_IDLE, NULL); mtx_unlock_spin(&sched_lock); } } /* * A CPU is entering for the first time or a thread is exiting. */ void sched_throw(struct thread *td) { /* * Correct spinlock nesting. The idle thread context that we are * borrowing was created so that it would start out with a single * spin lock (sched_lock) held in fork_trampoline(). Since we've * explicitly acquired locks in this function, the nesting count * is now 2 rather than 1. Since we are nested, calling * spinlock_exit() will simply adjust the counts without allowing * spin lock using code to interrupt us. */ if (td == NULL) { mtx_lock_spin(&sched_lock); spinlock_exit(); PCPU_SET(switchtime, cpu_ticks()); PCPU_SET(switchticks, ticks); } else { lock_profile_release_lock(&sched_lock.lock_object); MPASS(td->td_lock == &sched_lock); td->td_lastcpu = td->td_oncpu; td->td_oncpu = NOCPU; } mtx_assert(&sched_lock, MA_OWNED); KASSERT(curthread->td_md.md_spinlock_count == 1, ("invalid count")); cpu_throw(td, choosethread()); /* doesn't return */ } void sched_fork_exit(struct thread *td) { /* * Finish setting up thread glue so that it begins execution in a * non-nested critical section with sched_lock held but not recursed. */ td->td_oncpu = PCPU_GET(cpuid); sched_lock.mtx_lock = (uintptr_t)td; lock_profile_obtain_lock_success(&sched_lock.lock_object, 0, 0, __FILE__, __LINE__); THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED); } char * sched_tdname(struct thread *td) { #ifdef KTR struct td_sched *ts; ts = td_get_sched(td); if (ts->ts_name[0] == '\0') snprintf(ts->ts_name, sizeof(ts->ts_name), "%s tid %d", td->td_name, td->td_tid); return (ts->ts_name); #else return (td->td_name); #endif } #ifdef KTR void sched_clear_tdname(struct thread *td) { struct td_sched *ts; ts = td_get_sched(td); ts->ts_name[0] = '\0'; } #endif void sched_affinity(struct thread *td) { #ifdef SMP struct td_sched *ts; int cpu; THREAD_LOCK_ASSERT(td, MA_OWNED); /* * Set the TSF_AFFINITY flag if there is at least one CPU this * thread can't run on. */ ts = td_get_sched(td); ts->ts_flags &= ~TSF_AFFINITY; CPU_FOREACH(cpu) { if (!THREAD_CAN_SCHED(td, cpu)) { ts->ts_flags |= TSF_AFFINITY; break; } } /* * If this thread can run on all CPUs, nothing else to do. */ if (!(ts->ts_flags & TSF_AFFINITY)) return; /* Pinned threads and bound threads should be left alone. */ if (td->td_pinned != 0 || td->td_flags & TDF_BOUND) return; switch (td->td_state) { case TDS_RUNQ: /* * If we are on a per-CPU runqueue that is in the set, * then nothing needs to be done. */ if (ts->ts_runq != &runq && THREAD_CAN_SCHED(td, ts->ts_runq - runq_pcpu)) return; /* Put this thread on a valid per-CPU runqueue. */ sched_rem(td); sched_add(td, SRQ_BORING); break; case TDS_RUNNING: /* * See if our current CPU is in the set. If not, force a * context switch. */ if (THREAD_CAN_SCHED(td, td->td_oncpu)) return; td->td_flags |= TDF_NEEDRESCHED; if (td != curthread) ipi_cpu(cpu, IPI_AST); break; default: break; } #endif } Index: head/sys/kern/subr_autoconf.c =================================================================== --- head/sys/kern/subr_autoconf.c (revision 305831) +++ head/sys/kern/subr_autoconf.c (revision 305832) @@ -1,230 +1,230 @@ /*- * Copyright (c) 1992, 1993 * The Regents of the University of California. All rights reserved. * * This software was developed by the Computer Systems Engineering group * at Lawrence Berkeley Laboratory under DARPA contract BG 91-66 and * contributed to Berkeley. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)subr_autoconf.c 8.1 (Berkeley) 6/10/93 * */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include #include #include #include #include #include /* * Autoconfiguration subroutines. */ /* * "Interrupt driven config" functions. */ static TAILQ_HEAD(, intr_config_hook) intr_config_hook_list = TAILQ_HEAD_INITIALIZER(intr_config_hook_list); static struct intr_config_hook *next_to_notify; static struct mtx intr_config_hook_lock; MTX_SYSINIT(intr_config_hook, &intr_config_hook_lock, "intr config", MTX_DEF); /* ARGSUSED */ static void run_interrupt_driven_config_hooks(void); /* * If we wait too long for an interrupt-driven config hook to return, print * a diagnostic. */ #define WARNING_INTERVAL_SECS 60 static void run_interrupt_driven_config_hooks_warning(int warned) { struct intr_config_hook *hook_entry; char namebuf[64]; long offset; if (warned < 6) { printf("run_interrupt_driven_hooks: still waiting after %d " "seconds for", warned * WARNING_INTERVAL_SECS); TAILQ_FOREACH(hook_entry, &intr_config_hook_list, ich_links) { if (linker_search_symbol_name( (caddr_t)hook_entry->ich_func, namebuf, sizeof(namebuf), &offset) == 0) printf(" %s", namebuf); else printf(" %p", hook_entry->ich_func); } printf("\n"); } KASSERT(warned < 6, ("run_interrupt_driven_config_hooks: waited too long")); } static void run_interrupt_driven_config_hooks() { static int running; struct intr_config_hook *hook_entry; mtx_lock(&intr_config_hook_lock); /* * If hook processing is already active, any newly * registered hooks will eventually be notified. * Let the currently running session issue these * notifications. */ if (running != 0) { mtx_unlock(&intr_config_hook_lock); return; } running = 1; while (next_to_notify != NULL) { hook_entry = next_to_notify; next_to_notify = TAILQ_NEXT(hook_entry, ich_links); mtx_unlock(&intr_config_hook_lock); (*hook_entry->ich_func)(hook_entry->ich_arg); mtx_lock(&intr_config_hook_lock); } running = 0; mtx_unlock(&intr_config_hook_lock); } static void boot_run_interrupt_driven_config_hooks(void *dummy) { int warned; run_interrupt_driven_config_hooks(); /* Block boot processing until all hooks are disestablished. */ mtx_lock(&intr_config_hook_lock); warned = 0; while (!TAILQ_EMPTY(&intr_config_hook_list)) { if (msleep(&intr_config_hook_list, &intr_config_hook_lock, 0, "conifhk", WARNING_INTERVAL_SECS * hz) == EWOULDBLOCK) { mtx_unlock(&intr_config_hook_lock); warned++; run_interrupt_driven_config_hooks_warning(warned); mtx_lock(&intr_config_hook_lock); } } mtx_unlock(&intr_config_hook_lock); } SYSINIT(intr_config_hooks, SI_SUB_INT_CONFIG_HOOKS, SI_ORDER_FIRST, boot_run_interrupt_driven_config_hooks, NULL); /* * Register a hook that will be called after "cold" * autoconfiguration is complete and interrupts can * be used to complete initialization. */ int config_intrhook_establish(struct intr_config_hook *hook) { struct intr_config_hook *hook_entry; mtx_lock(&intr_config_hook_lock); TAILQ_FOREACH(hook_entry, &intr_config_hook_list, ich_links) if (hook_entry == hook) break; if (hook_entry != NULL) { mtx_unlock(&intr_config_hook_lock); printf("config_intrhook_establish: establishing an " "already established hook.\n"); return (1); } TAILQ_INSERT_TAIL(&intr_config_hook_list, hook, ich_links); if (next_to_notify == NULL) next_to_notify = hook; mtx_unlock(&intr_config_hook_lock); if (cold == 0) /* * XXX Call from a task since not all drivers expect * to be re-entered at the time a hook is established. */ /* XXX Sufficient for modules loaded after initial config??? */ run_interrupt_driven_config_hooks(); return (0); } void config_intrhook_disestablish(struct intr_config_hook *hook) { struct intr_config_hook *hook_entry; mtx_lock(&intr_config_hook_lock); TAILQ_FOREACH(hook_entry, &intr_config_hook_list, ich_links) if (hook_entry == hook) break; if (hook_entry == NULL) panic("config_intrhook_disestablish: disestablishing an " "unestablished hook"); if (next_to_notify == hook) next_to_notify = TAILQ_NEXT(hook, ich_links); TAILQ_REMOVE(&intr_config_hook_list, hook, ich_links); /* Wakeup anyone watching the list */ wakeup(&intr_config_hook_list); mtx_unlock(&intr_config_hook_lock); } #ifdef DDB #include DB_SHOW_COMMAND(conifhk, db_show_conifhk) { struct intr_config_hook *hook_entry; char namebuf[64]; long offset; TAILQ_FOREACH(hook_entry, &intr_config_hook_list, ich_links) { if (linker_ddb_search_symbol_name( (caddr_t)hook_entry->ich_func, namebuf, sizeof(namebuf), &offset) == 0) { db_printf("hook: %p at %s+%#lx arg: %p\n", hook_entry->ich_func, namebuf, offset, hook_entry->ich_arg); } else { db_printf("hook: %p at ??+?? arg %p\n", hook_entry->ich_func, hook_entry->ich_arg); } } } #endif /* DDB */ Index: head/sys/kern/subr_blist.c =================================================================== --- head/sys/kern/subr_blist.c (revision 305831) +++ head/sys/kern/subr_blist.c (revision 305832) @@ -1,1095 +1,1095 @@ /*- * Copyright (c) 1998 Matthew Dillon. All Rights Reserved. * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE * GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ /* * BLIST.C - Bitmap allocator/deallocator, using a radix tree with hinting * * This module implements a general bitmap allocator/deallocator. The * allocator eats around 2 bits per 'block'. The module does not * try to interpret the meaning of a 'block' other than to return * SWAPBLK_NONE on an allocation failure. * * A radix tree is used to maintain the bitmap. Two radix constants are * involved: One for the bitmaps contained in the leaf nodes (typically * 32), and one for the meta nodes (typically 16). Both meta and leaf * nodes have a hint field. This field gives us a hint as to the largest * free contiguous range of blocks under the node. It may contain a * value that is too high, but will never contain a value that is too * low. When the radix tree is searched, allocation failures in subtrees * update the hint. * * The radix tree also implements two collapsed states for meta nodes: * the ALL-ALLOCATED state and the ALL-FREE state. If a meta node is * in either of these two states, all information contained underneath * the node is considered stale. These states are used to optimize * allocation and freeing operations. * * The hinting greatly increases code efficiency for allocations while * the general radix structure optimizes both allocations and frees. The * radix tree should be able to operate well no matter how much * fragmentation there is and no matter how large a bitmap is used. * * The blist code wires all necessary memory at creation time. Neither * allocations nor frees require interaction with the memory subsystem. * The non-blocking features of the blist code are used in the swap code * (vm/swap_pager.c). * * LAYOUT: The radix tree is laid out recursively using a * linear array. Each meta node is immediately followed (laid out * sequentially in memory) by BLIST_META_RADIX lower level nodes. This * is a recursive structure but one that can be easily scanned through * a very simple 'skip' calculation. In order to support large radixes, * portions of the tree may reside outside our memory allocation. We * handle this with an early-termination optimization (when bighint is * set to -1) on the scan. The memory allocation is only large enough * to cover the number of blocks requested at creation time even if it * must be encompassed in larger root-node radix. * * NOTE: the allocator cannot currently allocate more than * BLIST_BMAP_RADIX blocks per call. It will panic with 'allocation too * large' if you try. This is an area that could use improvement. The * radix is large enough that this restriction does not effect the swap * system, though. Currently only the allocation code is effected by * this algorithmic unfeature. The freeing code can handle arbitrary * ranges. * * This code can be compiled stand-alone for debugging. */ #include __FBSDID("$FreeBSD$"); #ifdef _KERNEL #include #include #include #include #include #include #include #include #else #ifndef BLIST_NO_DEBUG #define BLIST_DEBUG #endif #define SWAPBLK_NONE ((daddr_t)-1) #include #include #include #include #include #define malloc(a,b,c) calloc(a, 1) #define free(a,b) free(a) typedef unsigned int u_daddr_t; #include void panic(const char *ctl, ...); #endif /* * static support functions */ static daddr_t blst_leaf_alloc(blmeta_t *scan, daddr_t blk, int count); static daddr_t blst_meta_alloc(blmeta_t *scan, daddr_t blk, daddr_t count, daddr_t radix, int skip); static void blst_leaf_free(blmeta_t *scan, daddr_t relblk, int count); static void blst_meta_free(blmeta_t *scan, daddr_t freeBlk, daddr_t count, daddr_t radix, int skip, daddr_t blk); static void blst_copy(blmeta_t *scan, daddr_t blk, daddr_t radix, daddr_t skip, blist_t dest, daddr_t count); static int blst_leaf_fill(blmeta_t *scan, daddr_t blk, int count); static int blst_meta_fill(blmeta_t *scan, daddr_t allocBlk, daddr_t count, daddr_t radix, int skip, daddr_t blk); static daddr_t blst_radix_init(blmeta_t *scan, daddr_t radix, int skip, daddr_t count); #ifndef _KERNEL static void blst_radix_print(blmeta_t *scan, daddr_t blk, daddr_t radix, int skip, int tab); #endif #ifdef _KERNEL static MALLOC_DEFINE(M_SWAP, "SWAP", "Swap space"); #endif /* * blist_create() - create a blist capable of handling up to the specified * number of blocks * * blocks - must be greater than 0 * flags - malloc flags * * The smallest blist consists of a single leaf node capable of * managing BLIST_BMAP_RADIX blocks. */ blist_t blist_create(daddr_t blocks, int flags) { blist_t bl; int radix; int skip = 0; /* * Calculate radix and skip field used for scanning. */ radix = BLIST_BMAP_RADIX; while (radix < blocks) { radix *= BLIST_META_RADIX; skip = (skip + 1) * BLIST_META_RADIX; } bl = malloc(sizeof(struct blist), M_SWAP, flags | M_ZERO); bl->bl_blocks = blocks; bl->bl_radix = radix; bl->bl_skip = skip; bl->bl_rootblks = 1 + blst_radix_init(NULL, bl->bl_radix, bl->bl_skip, blocks); bl->bl_root = malloc(sizeof(blmeta_t) * bl->bl_rootblks, M_SWAP, flags); #if defined(BLIST_DEBUG) printf( "BLIST representing %lld blocks (%lld MB of swap)" ", requiring %lldK of ram\n", (long long)bl->bl_blocks, (long long)bl->bl_blocks * 4 / 1024, (long long)(bl->bl_rootblks * sizeof(blmeta_t) + 1023) / 1024 ); printf("BLIST raw radix tree contains %lld records\n", (long long)bl->bl_rootblks); #endif blst_radix_init(bl->bl_root, bl->bl_radix, bl->bl_skip, blocks); return(bl); } void blist_destroy(blist_t bl) { free(bl->bl_root, M_SWAP); free(bl, M_SWAP); } /* * blist_alloc() - reserve space in the block bitmap. Return the base * of a contiguous region or SWAPBLK_NONE if space could * not be allocated. */ daddr_t blist_alloc(blist_t bl, daddr_t count) { daddr_t blk = SWAPBLK_NONE; if (bl) { if (bl->bl_radix == BLIST_BMAP_RADIX) blk = blst_leaf_alloc(bl->bl_root, 0, count); else blk = blst_meta_alloc(bl->bl_root, 0, count, bl->bl_radix, bl->bl_skip); if (blk != SWAPBLK_NONE) bl->bl_free -= count; } return(blk); } /* * blist_free() - free up space in the block bitmap. Return the base * of a contiguous region. Panic if an inconsistancy is * found. */ void blist_free(blist_t bl, daddr_t blkno, daddr_t count) { if (bl) { if (bl->bl_radix == BLIST_BMAP_RADIX) blst_leaf_free(bl->bl_root, blkno, count); else blst_meta_free(bl->bl_root, blkno, count, bl->bl_radix, bl->bl_skip, 0); bl->bl_free += count; } } /* * blist_fill() - mark a region in the block bitmap as off-limits * to the allocator (i.e. allocate it), ignoring any * existing allocations. Return the number of blocks * actually filled that were free before the call. */ int blist_fill(blist_t bl, daddr_t blkno, daddr_t count) { int filled; if (bl) { if (bl->bl_radix == BLIST_BMAP_RADIX) filled = blst_leaf_fill(bl->bl_root, blkno, count); else filled = blst_meta_fill(bl->bl_root, blkno, count, bl->bl_radix, bl->bl_skip, 0); bl->bl_free -= filled; return filled; } else return 0; } /* * blist_resize() - resize an existing radix tree to handle the * specified number of blocks. This will reallocate * the tree and transfer the previous bitmap to the new * one. When extending the tree you can specify whether * the new blocks are to left allocated or freed. */ void blist_resize(blist_t *pbl, daddr_t count, int freenew, int flags) { blist_t newbl = blist_create(count, flags); blist_t save = *pbl; *pbl = newbl; if (count > save->bl_blocks) count = save->bl_blocks; blst_copy(save->bl_root, 0, save->bl_radix, save->bl_skip, newbl, count); /* * If resizing upwards, should we free the new space or not? */ if (freenew && count < newbl->bl_blocks) { blist_free(newbl, count, newbl->bl_blocks - count); } blist_destroy(save); } #ifdef BLIST_DEBUG /* * blist_print() - dump radix tree */ void blist_print(blist_t bl) { printf("BLIST {\n"); blst_radix_print(bl->bl_root, 0, bl->bl_radix, bl->bl_skip, 4); printf("}\n"); } #endif /************************************************************************ * ALLOCATION SUPPORT FUNCTIONS * ************************************************************************ * * These support functions do all the actual work. They may seem * rather longish, but that's because I've commented them up. The * actual code is straight forward. * */ /* * blist_leaf_alloc() - allocate at a leaf in the radix tree (a bitmap). * * This is the core of the allocator and is optimized for the 1 block * and the BLIST_BMAP_RADIX block allocation cases. Other cases are * somewhat slower. The 1 block allocation case is log2 and extremely * quick. */ static daddr_t blst_leaf_alloc( blmeta_t *scan, daddr_t blk, int count ) { u_daddr_t orig = scan->u.bmu_bitmap; if (orig == 0) { /* * Optimize bitmap all-allocated case. Also, count = 1 * case assumes at least 1 bit is free in the bitmap, so * we have to take care of this case here. */ scan->bm_bighint = 0; return(SWAPBLK_NONE); } if (count == 1) { /* * Optimized code to allocate one bit out of the bitmap */ u_daddr_t mask; int j = BLIST_BMAP_RADIX/2; int r = 0; mask = (u_daddr_t)-1 >> (BLIST_BMAP_RADIX/2); while (j) { if ((orig & mask) == 0) { r += j; orig >>= j; } j >>= 1; mask >>= j; } scan->u.bmu_bitmap &= ~(1 << r); return(blk + r); } if (count <= BLIST_BMAP_RADIX) { /* * non-optimized code to allocate N bits out of the bitmap. * The more bits, the faster the code runs. It will run * the slowest allocating 2 bits, but since there aren't any * memory ops in the core loop (or shouldn't be, anyway), * you probably won't notice the difference. */ int j; int n = BLIST_BMAP_RADIX - count; u_daddr_t mask; mask = (u_daddr_t)-1 >> n; for (j = 0; j <= n; ++j) { if ((orig & mask) == mask) { scan->u.bmu_bitmap &= ~mask; return(blk + j); } mask = (mask << 1); } } /* * We couldn't allocate count in this subtree, update bighint. */ scan->bm_bighint = count - 1; return(SWAPBLK_NONE); } /* * blist_meta_alloc() - allocate at a meta in the radix tree. * * Attempt to allocate at a meta node. If we can't, we update * bighint and return a failure. Updating bighint optimize future * calls that hit this node. We have to check for our collapse cases * and we have a few optimizations strewn in as well. */ static daddr_t blst_meta_alloc( blmeta_t *scan, daddr_t blk, daddr_t count, daddr_t radix, int skip ) { int i; int next_skip = ((u_int)skip / BLIST_META_RADIX); if (scan->u.bmu_avail == 0) { /* * ALL-ALLOCATED special case */ scan->bm_bighint = count; return(SWAPBLK_NONE); } if (scan->u.bmu_avail == radix) { radix /= BLIST_META_RADIX; /* * ALL-FREE special case, initialize uninitialize * sublevel. */ for (i = 1; i <= skip; i += next_skip) { if (scan[i].bm_bighint == (daddr_t)-1) break; if (next_skip == 1) { scan[i].u.bmu_bitmap = (u_daddr_t)-1; scan[i].bm_bighint = BLIST_BMAP_RADIX; } else { scan[i].bm_bighint = radix; scan[i].u.bmu_avail = radix; } } } else { radix /= BLIST_META_RADIX; } for (i = 1; i <= skip; i += next_skip) { if (count <= scan[i].bm_bighint) { /* * count fits in object */ daddr_t r; if (next_skip == 1) { r = blst_leaf_alloc(&scan[i], blk, count); } else { r = blst_meta_alloc(&scan[i], blk, count, radix, next_skip - 1); } if (r != SWAPBLK_NONE) { scan->u.bmu_avail -= count; if (scan->bm_bighint > scan->u.bmu_avail) scan->bm_bighint = scan->u.bmu_avail; return(r); } } else if (scan[i].bm_bighint == (daddr_t)-1) { /* * Terminator */ break; } else if (count > radix) { /* * count does not fit in object even if it were * complete free. */ panic("blist_meta_alloc: allocation too large"); } blk += radix; } /* * We couldn't allocate count in this subtree, update bighint. */ if (scan->bm_bighint >= count) scan->bm_bighint = count - 1; return(SWAPBLK_NONE); } /* * BLST_LEAF_FREE() - free allocated block from leaf bitmap * */ static void blst_leaf_free( blmeta_t *scan, daddr_t blk, int count ) { /* * free some data in this bitmap * * e.g. * 0000111111111110000 * \_________/\__/ * v n */ int n = blk & (BLIST_BMAP_RADIX - 1); u_daddr_t mask; mask = ((u_daddr_t)-1 << n) & ((u_daddr_t)-1 >> (BLIST_BMAP_RADIX - count - n)); if (scan->u.bmu_bitmap & mask) panic("blst_radix_free: freeing free block"); scan->u.bmu_bitmap |= mask; /* * We could probably do a better job here. We are required to make * bighint at least as large as the biggest contiguous block of * data. If we just shoehorn it, a little extra overhead will * be incured on the next allocation (but only that one typically). */ scan->bm_bighint = BLIST_BMAP_RADIX; } /* * BLST_META_FREE() - free allocated blocks from radix tree meta info * * This support routine frees a range of blocks from the bitmap. * The range must be entirely enclosed by this radix node. If a * meta node, we break the range down recursively to free blocks * in subnodes (which means that this code can free an arbitrary * range whereas the allocation code cannot allocate an arbitrary * range). */ static void blst_meta_free( blmeta_t *scan, daddr_t freeBlk, daddr_t count, daddr_t radix, int skip, daddr_t blk ) { int i; int next_skip = ((u_int)skip / BLIST_META_RADIX); #if 0 printf("free (%llx,%lld) FROM (%llx,%lld)\n", (long long)freeBlk, (long long)count, (long long)blk, (long long)radix ); #endif if (scan->u.bmu_avail == 0) { /* * ALL-ALLOCATED special case, with possible * shortcut to ALL-FREE special case. */ scan->u.bmu_avail = count; scan->bm_bighint = count; if (count != radix) { for (i = 1; i <= skip; i += next_skip) { if (scan[i].bm_bighint == (daddr_t)-1) break; scan[i].bm_bighint = 0; if (next_skip == 1) { scan[i].u.bmu_bitmap = 0; } else { scan[i].u.bmu_avail = 0; } } /* fall through */ } } else { scan->u.bmu_avail += count; /* scan->bm_bighint = radix; */ } /* * ALL-FREE special case. */ if (scan->u.bmu_avail == radix) return; if (scan->u.bmu_avail > radix) panic("blst_meta_free: freeing already free blocks (%lld) %lld/%lld", (long long)count, (long long)scan->u.bmu_avail, (long long)radix); /* * Break the free down into its components */ radix /= BLIST_META_RADIX; i = (freeBlk - blk) / radix; blk += i * radix; i = i * next_skip + 1; while (i <= skip && blk < freeBlk + count) { daddr_t v; v = blk + radix - freeBlk; if (v > count) v = count; if (scan->bm_bighint == (daddr_t)-1) panic("blst_meta_free: freeing unexpected range"); if (next_skip == 1) { blst_leaf_free(&scan[i], freeBlk, v); } else { blst_meta_free(&scan[i], freeBlk, v, radix, next_skip - 1, blk); } if (scan->bm_bighint < scan[i].bm_bighint) scan->bm_bighint = scan[i].bm_bighint; count -= v; freeBlk += v; blk += radix; i += next_skip; } } /* * BLIST_RADIX_COPY() - copy one radix tree to another * * Locates free space in the source tree and frees it in the destination * tree. The space may not already be free in the destination. */ static void blst_copy( blmeta_t *scan, daddr_t blk, daddr_t radix, daddr_t skip, blist_t dest, daddr_t count ) { int next_skip; int i; /* * Leaf node */ if (radix == BLIST_BMAP_RADIX) { u_daddr_t v = scan->u.bmu_bitmap; if (v == (u_daddr_t)-1) { blist_free(dest, blk, count); } else if (v != 0) { int i; for (i = 0; i < BLIST_BMAP_RADIX && i < count; ++i) { if (v & (1 << i)) blist_free(dest, blk + i, 1); } } return; } /* * Meta node */ if (scan->u.bmu_avail == 0) { /* * Source all allocated, leave dest allocated */ return; } if (scan->u.bmu_avail == radix) { /* * Source all free, free entire dest */ if (count < radix) blist_free(dest, blk, count); else blist_free(dest, blk, radix); return; } radix /= BLIST_META_RADIX; next_skip = ((u_int)skip / BLIST_META_RADIX); for (i = 1; count && i <= skip; i += next_skip) { if (scan[i].bm_bighint == (daddr_t)-1) break; if (count >= radix) { blst_copy( &scan[i], blk, radix, next_skip - 1, dest, radix ); count -= radix; } else { if (count) { blst_copy( &scan[i], blk, radix, next_skip - 1, dest, count ); } count = 0; } blk += radix; } } /* * BLST_LEAF_FILL() - allocate specific blocks in leaf bitmap * * This routine allocates all blocks in the specified range * regardless of any existing allocations in that range. Returns * the number of blocks allocated by the call. */ static int blst_leaf_fill(blmeta_t *scan, daddr_t blk, int count) { int n = blk & (BLIST_BMAP_RADIX - 1); int nblks; u_daddr_t mask, bitmap; mask = ((u_daddr_t)-1 << n) & ((u_daddr_t)-1 >> (BLIST_BMAP_RADIX - count - n)); /* Count the number of blocks we're about to allocate */ bitmap = scan->u.bmu_bitmap & mask; for (nblks = 0; bitmap != 0; nblks++) bitmap &= bitmap - 1; scan->u.bmu_bitmap &= ~mask; return nblks; } /* * BLIST_META_FILL() - allocate specific blocks at a meta node * * This routine allocates the specified range of blocks, * regardless of any existing allocations in the range. The * range must be within the extent of this node. Returns the * number of blocks allocated by the call. */ static int blst_meta_fill( blmeta_t *scan, daddr_t allocBlk, daddr_t count, daddr_t radix, int skip, daddr_t blk ) { int i; int next_skip = ((u_int)skip / BLIST_META_RADIX); int nblks = 0; if (count == radix || scan->u.bmu_avail == 0) { /* * ALL-ALLOCATED special case */ nblks = scan->u.bmu_avail; scan->u.bmu_avail = 0; scan->bm_bighint = count; return nblks; } if (scan->u.bmu_avail == radix) { radix /= BLIST_META_RADIX; /* * ALL-FREE special case, initialize sublevel */ for (i = 1; i <= skip; i += next_skip) { if (scan[i].bm_bighint == (daddr_t)-1) break; if (next_skip == 1) { scan[i].u.bmu_bitmap = (u_daddr_t)-1; scan[i].bm_bighint = BLIST_BMAP_RADIX; } else { scan[i].bm_bighint = radix; scan[i].u.bmu_avail = radix; } } } else { radix /= BLIST_META_RADIX; } if (count > radix) panic("blist_meta_fill: allocation too large"); i = (allocBlk - blk) / radix; blk += i * radix; i = i * next_skip + 1; while (i <= skip && blk < allocBlk + count) { daddr_t v; v = blk + radix - allocBlk; if (v > count) v = count; if (scan->bm_bighint == (daddr_t)-1) panic("blst_meta_fill: filling unexpected range"); if (next_skip == 1) { nblks += blst_leaf_fill(&scan[i], allocBlk, v); } else { nblks += blst_meta_fill(&scan[i], allocBlk, v, radix, next_skip - 1, blk); } count -= v; allocBlk += v; blk += radix; i += next_skip; } scan->u.bmu_avail -= nblks; return nblks; } /* * BLST_RADIX_INIT() - initialize radix tree * * Initialize our meta structures and bitmaps and calculate the exact * amount of space required to manage 'count' blocks - this space may * be considerably less than the calculated radix due to the large * RADIX values we use. */ static daddr_t blst_radix_init(blmeta_t *scan, daddr_t radix, int skip, daddr_t count) { int i; int next_skip; daddr_t memindex = 0; /* * Leaf node */ if (radix == BLIST_BMAP_RADIX) { if (scan) { scan->bm_bighint = 0; scan->u.bmu_bitmap = 0; } return(memindex); } /* * Meta node. If allocating the entire object we can special * case it. However, we need to figure out how much memory * is required to manage 'count' blocks, so we continue on anyway. */ if (scan) { scan->bm_bighint = 0; scan->u.bmu_avail = 0; } radix /= BLIST_META_RADIX; next_skip = ((u_int)skip / BLIST_META_RADIX); for (i = 1; i <= skip; i += next_skip) { if (count >= radix) { /* * Allocate the entire object */ memindex = i + blst_radix_init( ((scan) ? &scan[i] : NULL), radix, next_skip - 1, radix ); count -= radix; } else if (count > 0) { /* * Allocate a partial object */ memindex = i + blst_radix_init( ((scan) ? &scan[i] : NULL), radix, next_skip - 1, count ); count = 0; } else { /* * Add terminator and break out */ if (scan) scan[i].bm_bighint = (daddr_t)-1; break; } } if (memindex < i) memindex = i; return(memindex); } #ifdef BLIST_DEBUG static void blst_radix_print(blmeta_t *scan, daddr_t blk, daddr_t radix, int skip, int tab) { int i; int next_skip; int lastState = 0; if (radix == BLIST_BMAP_RADIX) { printf( "%*.*s(%08llx,%lld): bitmap %08llx big=%lld\n", tab, tab, "", (long long)blk, (long long)radix, (long long)scan->u.bmu_bitmap, (long long)scan->bm_bighint ); return; } if (scan->u.bmu_avail == 0) { printf( "%*.*s(%08llx,%lld) ALL ALLOCATED\n", tab, tab, "", (long long)blk, (long long)radix ); return; } if (scan->u.bmu_avail == radix) { printf( "%*.*s(%08llx,%lld) ALL FREE\n", tab, tab, "", (long long)blk, (long long)radix ); return; } printf( "%*.*s(%08llx,%lld): subtree (%lld/%lld) big=%lld {\n", tab, tab, "", (long long)blk, (long long)radix, (long long)scan->u.bmu_avail, (long long)radix, (long long)scan->bm_bighint ); radix /= BLIST_META_RADIX; next_skip = ((u_int)skip / BLIST_META_RADIX); tab += 4; for (i = 1; i <= skip; i += next_skip) { if (scan[i].bm_bighint == (daddr_t)-1) { printf( "%*.*s(%08llx,%lld): Terminator\n", tab, tab, "", (long long)blk, (long long)radix ); lastState = 0; break; } blst_radix_print( &scan[i], blk, radix, next_skip - 1, tab ); blk += radix; } tab -= 4; printf( "%*.*s}\n", tab, tab, "" ); } #endif #ifdef BLIST_DEBUG int main(int ac, char **av) { int size = 1024; int i; blist_t bl; for (i = 1; i < ac; ++i) { const char *ptr = av[i]; if (*ptr != '-') { size = strtol(ptr, NULL, 0); continue; } ptr += 2; fprintf(stderr, "Bad option: %s\n", ptr - 2); exit(1); } bl = blist_create(size, M_WAITOK); blist_free(bl, 0, size); for (;;) { char buf[1024]; daddr_t da = 0; daddr_t count = 0; printf("%lld/%lld/%lld> ", (long long)bl->bl_free, (long long)size, (long long)bl->bl_radix); fflush(stdout); if (fgets(buf, sizeof(buf), stdin) == NULL) break; switch(buf[0]) { case 'r': if (sscanf(buf + 1, "%lld", &count) == 1) { blist_resize(&bl, count, 1); } else { printf("?\n"); } case 'p': blist_print(bl); break; case 'a': if (sscanf(buf + 1, "%lld", &count) == 1) { daddr_t blk = blist_alloc(bl, count); printf(" R=%08llx\n", (long long)blk); } else { printf("?\n"); } break; case 'f': if (sscanf(buf + 1, "%llx %lld", (long long *)&da, (long long *)&count) == 2) { blist_free(bl, da, count); } else { printf("?\n"); } break; case 'l': if (sscanf(buf + 1, "%llx %lld", (long long *)&da, (long long *)&count) == 2) { printf(" n=%d\n", blist_fill(bl, da, count)); } else { printf("?\n"); } break; case '?': case 'h': puts( "p -print\n" "a %d -allocate\n" "f %x %d -free\n" "l %x %d -fill\n" "r %d -resize\n" "h/? -help" ); break; default: printf("?\n"); break; } } return(0); } void panic(const char *ctl, ...) { va_list va; va_start(va, ctl); vfprintf(stderr, ctl, va); fprintf(stderr, "\n"); va_end(va); exit(1); } #endif Index: head/sys/kern/subr_clock.c =================================================================== --- head/sys/kern/subr_clock.c (revision 305831) +++ head/sys/kern/subr_clock.c (revision 305832) @@ -1,224 +1,224 @@ /*- * Copyright (c) 1988 University of Utah. * Copyright (c) 1982, 1990, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * the Systems Programming Group of the University of Utah Computer * Science Department. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: Utah $Hdr: clock.c 1.18 91/01/21$ * from: @(#)clock.c 8.2 (Berkeley) 1/12/94 * from: NetBSD: clock_subr.c,v 1.6 2001/07/07 17:04:02 thorpej Exp * and * from: src/sys/i386/isa/clock.c,v 1.176 2001/09/04 */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include int tz_minuteswest; int tz_dsttime; /* * The adjkerntz and wall_cmos_clock sysctls are in the "machdep" sysctl * namespace because they were misplaced there originally. */ static int adjkerntz; static int sysctl_machdep_adjkerntz(SYSCTL_HANDLER_ARGS) { int error; error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2, req); if (!error && req->newptr) resettodr(); return (error); } SYSCTL_PROC(_machdep, OID_AUTO, adjkerntz, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, &adjkerntz, 0, sysctl_machdep_adjkerntz, "I", "Local offset from UTC in seconds"); static int ct_debug; SYSCTL_INT(_debug, OID_AUTO, clocktime, CTLFLAG_RW, &ct_debug, 0, "Enable printing of clocktime debugging"); static int wall_cmos_clock; SYSCTL_INT(_machdep, OID_AUTO, wall_cmos_clock, CTLFLAG_RW, &wall_cmos_clock, 0, "Enables application of machdep.adjkerntz"); /*--------------------------------------------------------------------* * Generic routines to convert between a POSIX date * (seconds since 1/1/1970) and yr/mo/day/hr/min/sec * Derived from NetBSD arch/hp300/hp300/clock.c */ #define FEBRUARY 2 #define days_in_year(y) (leapyear(y) ? 366 : 365) #define days_in_month(y, m) \ (month_days[(m) - 1] + (m == FEBRUARY ? leapyear(y) : 0)) /* Day of week. Days are counted from 1/1/1970, which was a Thursday */ #define day_of_week(days) (((days) + 4) % 7) static const int month_days[12] = { 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31 }; /* * This inline avoids some unnecessary modulo operations * as compared with the usual macro: * ( ((year % 4) == 0 && * (year % 100) != 0) || * ((year % 400) == 0) ) * It is otherwise equivalent. */ static int leapyear(int year) { int rv = 0; if ((year & 3) == 0) { rv = 1; if ((year % 100) == 0) { rv = 0; if ((year % 400) == 0) rv = 1; } } return (rv); } static void print_ct(struct clocktime *ct) { printf("[%04d-%02d-%02d %02d:%02d:%02d]", ct->year, ct->mon, ct->day, ct->hour, ct->min, ct->sec); } int clock_ct_to_ts(struct clocktime *ct, struct timespec *ts) { int i, year, days; year = ct->year; if (ct_debug) { printf("ct_to_ts("); print_ct(ct); printf(")"); } /* Sanity checks. */ if (ct->mon < 1 || ct->mon > 12 || ct->day < 1 || ct->day > days_in_month(year, ct->mon) || ct->hour > 23 || ct->min > 59 || ct->sec > 59 || (sizeof(time_t) == 4 && year > 2037)) { /* time_t overflow */ if (ct_debug) printf(" = EINVAL\n"); return (EINVAL); } /* * Compute days since start of time * First from years, then from months. */ days = 0; for (i = POSIX_BASE_YEAR; i < year; i++) days += days_in_year(i); /* Months */ for (i = 1; i < ct->mon; i++) days += days_in_month(year, i); days += (ct->day - 1); ts->tv_sec = (((time_t)days * 24 + ct->hour) * 60 + ct->min) * 60 + ct->sec; ts->tv_nsec = ct->nsec; if (ct_debug) printf(" = %ld.%09ld\n", (long)ts->tv_sec, (long)ts->tv_nsec); return (0); } void clock_ts_to_ct(struct timespec *ts, struct clocktime *ct) { int i, year, days; time_t rsec; /* remainder seconds */ time_t secs; secs = ts->tv_sec; days = secs / SECDAY; rsec = secs % SECDAY; ct->dow = day_of_week(days); /* Subtract out whole years, counting them in i. */ for (year = POSIX_BASE_YEAR; days >= days_in_year(year); year++) days -= days_in_year(year); ct->year = year; /* Subtract out whole months, counting them in i. */ for (i = 1; days >= days_in_month(year, i); i++) days -= days_in_month(year, i); ct->mon = i; /* Days are what is left over (+1) from all that. */ ct->day = days + 1; /* Hours, minutes, seconds are easy */ ct->hour = rsec / 3600; rsec = rsec % 3600; ct->min = rsec / 60; rsec = rsec % 60; ct->sec = rsec; ct->nsec = ts->tv_nsec; if (ct_debug) { printf("ts_to_ct(%ld.%09ld) = ", (long)ts->tv_sec, (long)ts->tv_nsec); print_ct(ct); printf("\n"); } } int utc_offset(void) { return (tz_minuteswest * 60 + (wall_cmos_clock ? adjkerntz : 0)); } Index: head/sys/kern/subr_hash.c =================================================================== --- head/sys/kern/subr_hash.c (revision 305831) +++ head/sys/kern/subr_hash.c (revision 305832) @@ -1,152 +1,152 @@ /*- * Copyright (c) 1982, 1986, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_subr.c 8.3 (Berkeley) 1/21/94 */ #include __FBSDID("$FreeBSD$"); #include #include #include static __inline int hash_mflags(int flags) { return ((flags & HASH_NOWAIT) ? M_NOWAIT : M_WAITOK); } /* * General routine to allocate a hash table with control of memory flags. */ void * hashinit_flags(int elements, struct malloc_type *type, u_long *hashmask, int flags) { long hashsize; LIST_HEAD(generic, generic) *hashtbl; int i; KASSERT(elements > 0, ("%s: bad elements", __func__)); /* Exactly one of HASH_WAITOK and HASH_NOWAIT must be set. */ KASSERT((flags & HASH_WAITOK) ^ (flags & HASH_NOWAIT), ("Bad flags (0x%x) passed to hashinit_flags", flags)); for (hashsize = 1; hashsize <= elements; hashsize <<= 1) continue; hashsize >>= 1; hashtbl = malloc((u_long)hashsize * sizeof(*hashtbl), type, hash_mflags(flags)); if (hashtbl != NULL) { for (i = 0; i < hashsize; i++) LIST_INIT(&hashtbl[i]); *hashmask = hashsize - 1; } return (hashtbl); } /* * Allocate and initialize a hash table with default flag: may sleep. */ void * hashinit(int elements, struct malloc_type *type, u_long *hashmask) { return (hashinit_flags(elements, type, hashmask, HASH_WAITOK)); } void hashdestroy(void *vhashtbl, struct malloc_type *type, u_long hashmask) { LIST_HEAD(generic, generic) *hashtbl, *hp; hashtbl = vhashtbl; for (hp = hashtbl; hp <= &hashtbl[hashmask]; hp++) KASSERT(LIST_EMPTY(hp), ("%s: hashtbl %p not empty " "(malloc type %s)", __func__, hashtbl, type->ks_shortdesc)); free(hashtbl, type); } static const int primes[] = { 1, 13, 31, 61, 127, 251, 509, 761, 1021, 1531, 2039, 2557, 3067, 3583, 4093, 4603, 5119, 5623, 6143, 6653, 7159, 7673, 8191, 12281, 16381, 24571, 32749 }; #define NPRIMES nitems(primes) /* * General routine to allocate a prime number sized hash table with control of * memory flags. */ void * phashinit_flags(int elements, struct malloc_type *type, u_long *nentries, int flags) { long hashsize; LIST_HEAD(generic, generic) *hashtbl; int i; KASSERT(elements > 0, ("%s: bad elements", __func__)); /* Exactly one of HASH_WAITOK and HASH_NOWAIT must be set. */ KASSERT((flags & HASH_WAITOK) ^ (flags & HASH_NOWAIT), ("Bad flags (0x%x) passed to phashinit_flags", flags)); for (i = 1, hashsize = primes[1]; hashsize <= elements;) { i++; if (i == NPRIMES) break; hashsize = primes[i]; } hashsize = primes[i - 1]; hashtbl = malloc((u_long)hashsize * sizeof(*hashtbl), type, hash_mflags(flags)); if (hashtbl == NULL) return (NULL); for (i = 0; i < hashsize; i++) LIST_INIT(&hashtbl[i]); *nentries = hashsize; return (hashtbl); } /* * Allocate and initialize a prime number sized hash table with default flag: * may sleep. */ void * phashinit(int elements, struct malloc_type *type, u_long *nentries) { return (phashinit_flags(elements, type, nentries, HASH_WAITOK)); } Index: head/sys/kern/subr_log.c =================================================================== --- head/sys/kern/subr_log.c (revision 305831) +++ head/sys/kern/subr_log.c (revision 305832) @@ -1,310 +1,310 @@ /*- * Copyright (c) 1982, 1986, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)subr_log.c 8.1 (Berkeley) 6/10/93 */ /* * Error log buffer for kernel printf's. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #define LOG_RDPRI (PZERO + 1) #define LOG_ASYNC 0x04 static d_open_t logopen; static d_close_t logclose; static d_read_t logread; static d_ioctl_t logioctl; static d_poll_t logpoll; static d_kqfilter_t logkqfilter; static void logtimeout(void *arg); static struct cdevsw log_cdevsw = { .d_version = D_VERSION, .d_open = logopen, .d_close = logclose, .d_read = logread, .d_ioctl = logioctl, .d_poll = logpoll, .d_kqfilter = logkqfilter, .d_name = "log", }; static int logkqread(struct knote *note, long hint); static void logkqdetach(struct knote *note); static struct filterops log_read_filterops = { .f_isfd = 1, .f_attach = NULL, .f_detach = logkqdetach, .f_event = logkqread, }; static struct logsoftc { int sc_state; /* see above for possibilities */ struct selinfo sc_selp; /* process waiting on select call */ struct sigio *sc_sigio; /* information for async I/O */ struct callout sc_callout; /* callout to wakeup syslog */ } logsoftc; int log_open; /* also used in log() */ static struct cv log_wakeup; struct mtx msgbuf_lock; MTX_SYSINIT(msgbuf_lock, &msgbuf_lock, "msgbuf lock", MTX_DEF); /* Times per second to check for a pending syslog wakeup. */ static int log_wakeups_per_second = 5; SYSCTL_INT(_kern, OID_AUTO, log_wakeups_per_second, CTLFLAG_RW, &log_wakeups_per_second, 0, ""); /*ARGSUSED*/ static int logopen(struct cdev *dev, int flags, int mode, struct thread *td) { if (log_wakeups_per_second < 1) { printf("syslog wakeup is less than one. Adjusting to 1.\n"); log_wakeups_per_second = 1; } mtx_lock(&msgbuf_lock); if (log_open) { mtx_unlock(&msgbuf_lock); return (EBUSY); } log_open = 1; callout_reset_sbt(&logsoftc.sc_callout, SBT_1S / log_wakeups_per_second, 0, logtimeout, NULL, C_PREL(1)); mtx_unlock(&msgbuf_lock); fsetown(td->td_proc->p_pid, &logsoftc.sc_sigio); /* signal process only */ return (0); } /*ARGSUSED*/ static int logclose(struct cdev *dev, int flag, int mode, struct thread *td) { funsetown(&logsoftc.sc_sigio); mtx_lock(&msgbuf_lock); callout_stop(&logsoftc.sc_callout); logsoftc.sc_state = 0; log_open = 0; mtx_unlock(&msgbuf_lock); return (0); } /*ARGSUSED*/ static int logread(struct cdev *dev, struct uio *uio, int flag) { char buf[128]; struct msgbuf *mbp = msgbufp; int error = 0, l; mtx_lock(&msgbuf_lock); while (msgbuf_getcount(mbp) == 0) { if (flag & IO_NDELAY) { mtx_unlock(&msgbuf_lock); return (EWOULDBLOCK); } if ((error = cv_wait_sig(&log_wakeup, &msgbuf_lock)) != 0) { mtx_unlock(&msgbuf_lock); return (error); } } while (uio->uio_resid > 0) { l = imin(sizeof(buf), uio->uio_resid); l = msgbuf_getbytes(mbp, buf, l); if (l == 0) break; mtx_unlock(&msgbuf_lock); error = uiomove(buf, l, uio); if (error || uio->uio_resid == 0) return (error); mtx_lock(&msgbuf_lock); } mtx_unlock(&msgbuf_lock); return (error); } /*ARGSUSED*/ static int logpoll(struct cdev *dev, int events, struct thread *td) { int revents = 0; if (events & (POLLIN | POLLRDNORM)) { mtx_lock(&msgbuf_lock); if (msgbuf_getcount(msgbufp) > 0) revents |= events & (POLLIN | POLLRDNORM); else selrecord(td, &logsoftc.sc_selp); mtx_unlock(&msgbuf_lock); } return (revents); } static int logkqfilter(struct cdev *dev, struct knote *kn) { if (kn->kn_filter != EVFILT_READ) return (EINVAL); kn->kn_fop = &log_read_filterops; kn->kn_hook = NULL; mtx_lock(&msgbuf_lock); knlist_add(&logsoftc.sc_selp.si_note, kn, 1); mtx_unlock(&msgbuf_lock); return (0); } static int logkqread(struct knote *kn, long hint) { mtx_assert(&msgbuf_lock, MA_OWNED); kn->kn_data = msgbuf_getcount(msgbufp); return (kn->kn_data != 0); } static void logkqdetach(struct knote *kn) { mtx_lock(&msgbuf_lock); knlist_remove(&logsoftc.sc_selp.si_note, kn, 1); mtx_unlock(&msgbuf_lock); } static void logtimeout(void *arg) { if (!log_open) return; if (msgbuftrigger == 0) goto done; msgbuftrigger = 0; selwakeuppri(&logsoftc.sc_selp, LOG_RDPRI); KNOTE_LOCKED(&logsoftc.sc_selp.si_note, 0); if ((logsoftc.sc_state & LOG_ASYNC) && logsoftc.sc_sigio != NULL) pgsigio(&logsoftc.sc_sigio, SIGIO, 0); cv_broadcastpri(&log_wakeup, LOG_RDPRI); done: if (log_wakeups_per_second < 1) { printf("syslog wakeup is less than one. Adjusting to 1.\n"); log_wakeups_per_second = 1; } callout_reset_sbt(&logsoftc.sc_callout, SBT_1S / log_wakeups_per_second, 0, logtimeout, NULL, C_PREL(1)); } /*ARGSUSED*/ static int logioctl(struct cdev *dev, u_long com, caddr_t data, int flag, struct thread *td) { switch (com) { /* return number of characters immediately available */ case FIONREAD: *(int *)data = msgbuf_getcount(msgbufp); break; case FIONBIO: break; case FIOASYNC: mtx_lock(&msgbuf_lock); if (*(int *)data) logsoftc.sc_state |= LOG_ASYNC; else logsoftc.sc_state &= ~LOG_ASYNC; mtx_unlock(&msgbuf_lock); break; case FIOSETOWN: return (fsetown(*(int *)data, &logsoftc.sc_sigio)); case FIOGETOWN: *(int *)data = fgetown(&logsoftc.sc_sigio); break; /* This is deprecated, FIOSETOWN should be used instead. */ case TIOCSPGRP: return (fsetown(-(*(int *)data), &logsoftc.sc_sigio)); /* This is deprecated, FIOGETOWN should be used instead */ case TIOCGPGRP: *(int *)data = -fgetown(&logsoftc.sc_sigio); break; default: return (ENOTTY); } return (0); } static void log_drvinit(void *unused) { cv_init(&log_wakeup, "klog"); callout_init_mtx(&logsoftc.sc_callout, &msgbuf_lock, 0); knlist_init_mtx(&logsoftc.sc_selp.si_note, &msgbuf_lock); make_dev_credf(MAKEDEV_ETERNAL, &log_cdevsw, 0, NULL, UID_ROOT, GID_WHEEL, 0600, "klog"); } SYSINIT(logdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,log_drvinit,NULL); Index: head/sys/kern/subr_mchain.c =================================================================== --- head/sys/kern/subr_mchain.c (revision 305831) +++ head/sys/kern/subr_mchain.c (revision 305832) @@ -1,555 +1,555 @@ /*- * Copyright (c) 2000, 2001 Boris Popov * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the author nor the names of any co-contributors + * 3. Neither the name of the author nor the names of any co-contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include FEATURE(libmchain, "mchain library"); MODULE_VERSION(libmchain, 1); #define MBERROR(format, ...) printf("%s(%d): "format, __func__ , \ __LINE__ , ## __VA_ARGS__) #define MBPANIC(format, ...) printf("%s(%d): "format, __func__ , \ __LINE__ , ## __VA_ARGS__) /* * Various helper functions */ int mb_init(struct mbchain *mbp) { struct mbuf *m; m = m_gethdr(M_WAITOK, MT_DATA); m->m_len = 0; mb_initm(mbp, m); return (0); } void mb_initm(struct mbchain *mbp, struct mbuf *m) { bzero(mbp, sizeof(*mbp)); mbp->mb_top = mbp->mb_cur = m; mbp->mb_mleft = M_TRAILINGSPACE(m); } void mb_done(struct mbchain *mbp) { if (mbp->mb_top) { m_freem(mbp->mb_top); mbp->mb_top = NULL; } } struct mbuf * mb_detach(struct mbchain *mbp) { struct mbuf *m; m = mbp->mb_top; mbp->mb_top = NULL; return (m); } int mb_fixhdr(struct mbchain *mbp) { return (mbp->mb_top->m_pkthdr.len = m_fixhdr(mbp->mb_top)); } /* * Check if object of size 'size' fit to the current position and * allocate new mbuf if not. Advance pointers and increase length of mbuf(s). * Return pointer to the object placeholder or NULL if any error occurred. * Note: size should be <= MLEN */ caddr_t mb_reserve(struct mbchain *mbp, int size) { struct mbuf *m, *mn; caddr_t bpos; if (size > MLEN) panic("mb_reserve: size = %d\n", size); m = mbp->mb_cur; if (mbp->mb_mleft < size) { mn = m_get(M_WAITOK, MT_DATA); mbp->mb_cur = m->m_next = mn; m = mn; m->m_len = 0; mbp->mb_mleft = M_TRAILINGSPACE(m); } mbp->mb_mleft -= size; mbp->mb_count += size; bpos = mtod(m, caddr_t) + m->m_len; m->m_len += size; return (bpos); } int mb_put_padbyte(struct mbchain *mbp) { caddr_t dst; uint8_t x = 0; dst = mtod(mbp->mb_cur, caddr_t) + mbp->mb_cur->m_len; /* Only add padding if address is odd */ if ((unsigned long)dst & 1) return (mb_put_mem(mbp, (caddr_t)&x, sizeof(x), MB_MSYSTEM)); else return (0); } int mb_put_uint8(struct mbchain *mbp, uint8_t x) { return (mb_put_mem(mbp, (caddr_t)&x, sizeof(x), MB_MSYSTEM)); } int mb_put_uint16be(struct mbchain *mbp, uint16_t x) { x = htobe16(x); return (mb_put_mem(mbp, (caddr_t)&x, sizeof(x), MB_MSYSTEM)); } int mb_put_uint16le(struct mbchain *mbp, uint16_t x) { x = htole16(x); return (mb_put_mem(mbp, (caddr_t)&x, sizeof(x), MB_MSYSTEM)); } int mb_put_uint32be(struct mbchain *mbp, uint32_t x) { x = htobe32(x); return (mb_put_mem(mbp, (caddr_t)&x, sizeof(x), MB_MSYSTEM)); } int mb_put_uint32le(struct mbchain *mbp, uint32_t x) { x = htole32(x); return (mb_put_mem(mbp, (caddr_t)&x, sizeof(x), MB_MSYSTEM)); } int mb_put_int64be(struct mbchain *mbp, int64_t x) { x = htobe64(x); return (mb_put_mem(mbp, (caddr_t)&x, sizeof(x), MB_MSYSTEM)); } int mb_put_int64le(struct mbchain *mbp, int64_t x) { x = htole64(x); return (mb_put_mem(mbp, (caddr_t)&x, sizeof(x), MB_MSYSTEM)); } int mb_put_mem(struct mbchain *mbp, c_caddr_t source, int size, int type) { struct mbuf *m; caddr_t dst; c_caddr_t src; int cplen, error, mleft, count; size_t srclen, dstlen; m = mbp->mb_cur; mleft = mbp->mb_mleft; while (size > 0) { if (mleft == 0) { if (m->m_next == NULL) m = m_getm(m, size, M_WAITOK, MT_DATA); else m = m->m_next; mleft = M_TRAILINGSPACE(m); continue; } cplen = mleft > size ? size : mleft; srclen = dstlen = cplen; dst = mtod(m, caddr_t) + m->m_len; switch (type) { case MB_MCUSTOM: srclen = size; dstlen = mleft; error = mbp->mb_copy(mbp, source, dst, &srclen, &dstlen); if (error) return (error); break; case MB_MINLINE: for (src = source, count = cplen; count; count--) *dst++ = *src++; break; case MB_MSYSTEM: bcopy(source, dst, cplen); break; case MB_MUSER: error = copyin(source, dst, cplen); if (error) return (error); break; case MB_MZERO: bzero(dst, cplen); break; } size -= srclen; source += srclen; m->m_len += dstlen; mleft -= dstlen; mbp->mb_count += dstlen; } mbp->mb_cur = m; mbp->mb_mleft = mleft; return (0); } int mb_put_mbuf(struct mbchain *mbp, struct mbuf *m) { mbp->mb_cur->m_next = m; while (m) { mbp->mb_count += m->m_len; if (m->m_next == NULL) break; m = m->m_next; } mbp->mb_mleft = M_TRAILINGSPACE(m); mbp->mb_cur = m; return (0); } /* * copies a uio scatter/gather list to an mbuf chain. */ int mb_put_uio(struct mbchain *mbp, struct uio *uiop, int size) { long left; int mtype, error; mtype = (uiop->uio_segflg == UIO_SYSSPACE) ? MB_MSYSTEM : MB_MUSER; while (size > 0 && uiop->uio_resid) { if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL) return (EFBIG); left = uiop->uio_iov->iov_len; if (left == 0) { uiop->uio_iov++; uiop->uio_iovcnt--; continue; } if (left > size) left = size; error = mb_put_mem(mbp, uiop->uio_iov->iov_base, left, mtype); if (error) return (error); uiop->uio_offset += left; uiop->uio_resid -= left; uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left; uiop->uio_iov->iov_len -= left; size -= left; } return (0); } /* * Routines for fetching data from an mbuf chain */ int md_init(struct mdchain *mdp) { struct mbuf *m; m = m_gethdr(M_WAITOK, MT_DATA); m->m_len = 0; md_initm(mdp, m); return (0); } void md_initm(struct mdchain *mdp, struct mbuf *m) { bzero(mdp, sizeof(*mdp)); mdp->md_top = mdp->md_cur = m; mdp->md_pos = mtod(m, u_char*); } void md_done(struct mdchain *mdp) { if (mdp->md_top) { m_freem(mdp->md_top); mdp->md_top = NULL; } } /* * Append a separate mbuf chain. It is caller responsibility to prevent * multiple calls to fetch/record routines. */ void md_append_record(struct mdchain *mdp, struct mbuf *top) { struct mbuf *m; if (mdp->md_top == NULL) { md_initm(mdp, top); return; } m = mdp->md_top; while (m->m_nextpkt) m = m->m_nextpkt; m->m_nextpkt = top; top->m_nextpkt = NULL; return; } /* * Put next record in place of existing */ int md_next_record(struct mdchain *mdp) { struct mbuf *m; if (mdp->md_top == NULL) return (ENOENT); m = mdp->md_top->m_nextpkt; md_done(mdp); if (m == NULL) return (ENOENT); md_initm(mdp, m); return (0); } int md_get_uint8(struct mdchain *mdp, uint8_t *x) { return (md_get_mem(mdp, x, 1, MB_MINLINE)); } int md_get_uint16(struct mdchain *mdp, uint16_t *x) { return (md_get_mem(mdp, (caddr_t)x, 2, MB_MINLINE)); } int md_get_uint16le(struct mdchain *mdp, uint16_t *x) { uint16_t v; int error = md_get_uint16(mdp, &v); if (x != NULL) *x = le16toh(v); return (error); } int md_get_uint16be(struct mdchain *mdp, uint16_t *x) { uint16_t v; int error = md_get_uint16(mdp, &v); if (x != NULL) *x = be16toh(v); return (error); } int md_get_uint32(struct mdchain *mdp, uint32_t *x) { return (md_get_mem(mdp, (caddr_t)x, 4, MB_MINLINE)); } int md_get_uint32be(struct mdchain *mdp, uint32_t *x) { uint32_t v; int error; error = md_get_uint32(mdp, &v); if (x != NULL) *x = be32toh(v); return (error); } int md_get_uint32le(struct mdchain *mdp, uint32_t *x) { uint32_t v; int error; error = md_get_uint32(mdp, &v); if (x != NULL) *x = le32toh(v); return (error); } int md_get_int64(struct mdchain *mdp, int64_t *x) { return (md_get_mem(mdp, (caddr_t)x, 8, MB_MINLINE)); } int md_get_int64be(struct mdchain *mdp, int64_t *x) { int64_t v; int error; error = md_get_int64(mdp, &v); if (x != NULL) *x = be64toh(v); return (error); } int md_get_int64le(struct mdchain *mdp, int64_t *x) { int64_t v; int error; error = md_get_int64(mdp, &v); if (x != NULL) *x = le64toh(v); return (error); } int md_get_mem(struct mdchain *mdp, caddr_t target, int size, int type) { struct mbuf *m = mdp->md_cur; int error; u_int count; u_char *s; while (size > 0) { if (m == NULL) { MBERROR("incomplete copy\n"); return (EBADRPC); } s = mdp->md_pos; count = mtod(m, u_char*) + m->m_len - s; if (count == 0) { mdp->md_cur = m = m->m_next; if (m) s = mdp->md_pos = mtod(m, caddr_t); continue; } if (count > size) count = size; size -= count; mdp->md_pos += count; if (target == NULL) continue; switch (type) { case MB_MUSER: error = copyout(s, target, count); if (error) return error; break; case MB_MSYSTEM: bcopy(s, target, count); break; case MB_MINLINE: while (count--) *target++ = *s++; continue; } target += count; } return (0); } int md_get_mbuf(struct mdchain *mdp, int size, struct mbuf **ret) { struct mbuf *m = mdp->md_cur, *rm; rm = m_copym(m, mdp->md_pos - mtod(m, u_char*), size, M_WAITOK); md_get_mem(mdp, NULL, size, MB_MZERO); *ret = rm; return (0); } int md_get_uio(struct mdchain *mdp, struct uio *uiop, int size) { char *uiocp; long left; int mtype, error; mtype = (uiop->uio_segflg == UIO_SYSSPACE) ? MB_MSYSTEM : MB_MUSER; while (size > 0 && uiop->uio_resid) { if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL) return (EFBIG); left = uiop->uio_iov->iov_len; if (left == 0) { uiop->uio_iov++; uiop->uio_iovcnt--; continue; } uiocp = uiop->uio_iov->iov_base; if (left > size) left = size; error = md_get_mem(mdp, uiocp, left, mtype); if (error) return (error); uiop->uio_offset += left; uiop->uio_resid -= left; uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left; uiop->uio_iov->iov_len -= left; size -= left; } return (0); } Index: head/sys/kern/subr_param.c =================================================================== --- head/sys/kern/subr_param.c (revision 305831) +++ head/sys/kern/subr_param.c (revision 305832) @@ -1,300 +1,300 @@ /*- * Copyright (c) 1980, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)param.c 8.3 (Berkeley) 8/20/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_param.h" #include "opt_msgbuf.h" #include "opt_maxusers.h" #include #include #include #include #include #include #include #include #include #include /* * System parameter formulae. */ #ifndef HZ # if defined(__mips__) || defined(__arm__) # define HZ 100 # else # define HZ 1000 # endif # ifndef HZ_VM # define HZ_VM 100 # endif #else # ifndef HZ_VM # define HZ_VM HZ # endif #endif #define NPROC (20 + 16 * maxusers) #ifndef NBUF #define NBUF 0 #endif #ifndef MAXFILES #define MAXFILES (40 + 32 * maxusers) #endif static int sysctl_kern_vm_guest(SYSCTL_HANDLER_ARGS); int hz; /* system clock's frequency */ int tick; /* usec per tick (1000000 / hz) */ struct bintime tick_bt; /* bintime per tick (1s / hz) */ sbintime_t tick_sbt; int maxusers; /* base tunable */ int maxproc; /* maximum # of processes */ int maxprocperuid; /* max # of procs per user */ int maxfiles; /* sys. wide open files limit */ int maxfilesperproc; /* per-proc open files limit */ int msgbufsize; /* size of kernel message buffer */ int nbuf; int bio_transient_maxcnt; int ngroups_max; /* max # groups per process */ int nswbuf; pid_t pid_max = PID_MAX; long maxswzone; /* max swmeta KVA storage */ long maxbcache; /* max buffer cache KVA storage */ long maxpipekva; /* Limit on pipe KVA */ int vm_guest = VM_GUEST_NO; /* Running as virtual machine guest? */ u_long maxtsiz; /* max text size */ u_long dfldsiz; /* initial data size limit */ u_long maxdsiz; /* max data size */ u_long dflssiz; /* initial stack size limit */ u_long maxssiz; /* max stack size */ u_long sgrowsiz; /* amount to grow stack */ SYSCTL_INT(_kern, OID_AUTO, hz, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &hz, 0, "Number of clock ticks per second"); SYSCTL_INT(_kern, OID_AUTO, nbuf, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &nbuf, 0, "Number of buffers in the buffer cache"); SYSCTL_INT(_kern, OID_AUTO, nswbuf, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &nswbuf, 0, "Number of swap buffers"); SYSCTL_INT(_kern, OID_AUTO, msgbufsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &msgbufsize, 0, "Size of the kernel message buffer"); SYSCTL_LONG(_kern, OID_AUTO, maxswzone, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &maxswzone, 0, "Maximum memory for swap metadata"); SYSCTL_LONG(_kern, OID_AUTO, maxbcache, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &maxbcache, 0, "Maximum value of vfs.maxbufspace"); SYSCTL_INT(_kern, OID_AUTO, bio_transient_maxcnt, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &bio_transient_maxcnt, 0, "Maximum number of transient BIOs mappings"); SYSCTL_ULONG(_kern, OID_AUTO, maxtsiz, CTLFLAG_RWTUN | CTLFLAG_NOFETCH, &maxtsiz, 0, "Maximum text size"); SYSCTL_ULONG(_kern, OID_AUTO, dfldsiz, CTLFLAG_RWTUN | CTLFLAG_NOFETCH, &dfldsiz, 0, "Initial data size limit"); SYSCTL_ULONG(_kern, OID_AUTO, maxdsiz, CTLFLAG_RWTUN | CTLFLAG_NOFETCH, &maxdsiz, 0, "Maximum data size"); SYSCTL_ULONG(_kern, OID_AUTO, dflssiz, CTLFLAG_RWTUN | CTLFLAG_NOFETCH, &dflssiz, 0, "Initial stack size limit"); SYSCTL_ULONG(_kern, OID_AUTO, maxssiz, CTLFLAG_RWTUN | CTLFLAG_NOFETCH, &maxssiz, 0, "Maximum stack size"); SYSCTL_ULONG(_kern, OID_AUTO, sgrowsiz, CTLFLAG_RWTUN | CTLFLAG_NOFETCH, &sgrowsiz, 0, "Amount to grow stack on a stack fault"); SYSCTL_PROC(_kern, OID_AUTO, vm_guest, CTLFLAG_RD | CTLTYPE_STRING, NULL, 0, sysctl_kern_vm_guest, "A", "Virtual machine guest detected?"); /* * The elements of this array are ordered based upon the values of the * corresponding enum VM_GUEST members. */ static const char *const vm_guest_sysctl_names[] = { "none", "generic", "xen", "hv", "vmware", "kvm", NULL }; CTASSERT(nitems(vm_guest_sysctl_names) - 1 == VM_LAST); /* * Boot time overrides that are not scaled against main memory */ void init_param1(void) { #if !defined(__mips__) && !defined(__arm64__) && !defined(__sparc64__) TUNABLE_INT_FETCH("kern.kstack_pages", &kstack_pages); #endif hz = -1; TUNABLE_INT_FETCH("kern.hz", &hz); if (hz == -1) hz = vm_guest > VM_GUEST_NO ? HZ_VM : HZ; tick = 1000000 / hz; tick_sbt = SBT_1S / hz; tick_bt = sbttobt(tick_sbt); #ifdef VM_SWZONE_SIZE_MAX maxswzone = VM_SWZONE_SIZE_MAX; #endif TUNABLE_LONG_FETCH("kern.maxswzone", &maxswzone); #ifdef VM_BCACHE_SIZE_MAX maxbcache = VM_BCACHE_SIZE_MAX; #endif TUNABLE_LONG_FETCH("kern.maxbcache", &maxbcache); msgbufsize = MSGBUF_SIZE; TUNABLE_INT_FETCH("kern.msgbufsize", &msgbufsize); maxtsiz = MAXTSIZ; TUNABLE_ULONG_FETCH("kern.maxtsiz", &maxtsiz); dfldsiz = DFLDSIZ; TUNABLE_ULONG_FETCH("kern.dfldsiz", &dfldsiz); maxdsiz = MAXDSIZ; TUNABLE_ULONG_FETCH("kern.maxdsiz", &maxdsiz); dflssiz = DFLSSIZ; TUNABLE_ULONG_FETCH("kern.dflssiz", &dflssiz); maxssiz = MAXSSIZ; TUNABLE_ULONG_FETCH("kern.maxssiz", &maxssiz); sgrowsiz = SGROWSIZ; TUNABLE_ULONG_FETCH("kern.sgrowsiz", &sgrowsiz); /* * Let the administrator set {NGROUPS_MAX}, but disallow values * less than NGROUPS_MAX which would violate POSIX.1-2008 or * greater than INT_MAX-1 which would result in overflow. */ ngroups_max = NGROUPS_MAX; TUNABLE_INT_FETCH("kern.ngroups", &ngroups_max); if (ngroups_max < NGROUPS_MAX) ngroups_max = NGROUPS_MAX; /* * Only allow to lower the maximal pid. * Prevent setting up a non-bootable system if pid_max is too low. */ TUNABLE_INT_FETCH("kern.pid_max", &pid_max); if (pid_max > PID_MAX) pid_max = PID_MAX; else if (pid_max < 300) pid_max = 300; TUNABLE_INT_FETCH("vfs.unmapped_buf_allowed", &unmapped_buf_allowed); } /* * Boot time overrides that are scaled against main memory */ void init_param2(long physpages) { /* Base parameters */ maxusers = MAXUSERS; TUNABLE_INT_FETCH("kern.maxusers", &maxusers); if (maxusers == 0) { maxusers = physpages / (2 * 1024 * 1024 / PAGE_SIZE); if (maxusers < 32) maxusers = 32; #ifdef VM_MAX_AUTOTUNE_MAXUSERS if (maxusers > VM_MAX_AUTOTUNE_MAXUSERS) maxusers = VM_MAX_AUTOTUNE_MAXUSERS; #endif /* * Scales down the function in which maxusers grows once * we hit 384. */ if (maxusers > 384) maxusers = 384 + ((maxusers - 384) / 8); } /* * The following can be overridden after boot via sysctl. Note: * unless overriden, these macros are ultimately based on maxusers. * Limit maxproc so that kmap entries cannot be exhausted by * processes. */ maxproc = NPROC; TUNABLE_INT_FETCH("kern.maxproc", &maxproc); if (maxproc > (physpages / 12)) maxproc = physpages / 12; if (maxproc > pid_max) maxproc = pid_max; maxprocperuid = (maxproc * 9) / 10; /* * The default limit for maxfiles is 1/12 of the number of * physical page but not less than 16 times maxusers. * At most it can be 1/6 the number of physical pages. */ maxfiles = imax(MAXFILES, physpages / 8); TUNABLE_INT_FETCH("kern.maxfiles", &maxfiles); if (maxfiles > (physpages / 4)) maxfiles = physpages / 4; maxfilesperproc = (maxfiles / 10) * 9; TUNABLE_INT_FETCH("kern.maxfilesperproc", &maxfilesperproc); /* * Cannot be changed after boot. */ nbuf = NBUF; TUNABLE_INT_FETCH("kern.nbuf", &nbuf); TUNABLE_INT_FETCH("kern.bio_transient_maxcnt", &bio_transient_maxcnt); /* * The default for maxpipekva is min(1/64 of the kernel address space, * max(1/64 of main memory, 512KB)). See sys_pipe.c for more details. */ maxpipekva = (physpages / 64) * PAGE_SIZE; TUNABLE_LONG_FETCH("kern.ipc.maxpipekva", &maxpipekva); if (maxpipekva < 512 * 1024) maxpipekva = 512 * 1024; if (maxpipekva > (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / 64) maxpipekva = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / 64; } /* * Sysctl stringifying handler for kern.vm_guest. */ static int sysctl_kern_vm_guest(SYSCTL_HANDLER_ARGS) { return (SYSCTL_OUT_STR(req, vm_guest_sysctl_names[vm_guest])); } Index: head/sys/kern/subr_pcpu.c =================================================================== --- head/sys/kern/subr_pcpu.c (revision 305831) +++ head/sys/kern/subr_pcpu.c (revision 305832) @@ -1,419 +1,419 @@ /*- * Copyright (c) 2001 Wind River Systems, Inc. * All rights reserved. * Written by: John Baldwin * * Copyright (c) 2009 Jeffrey Roberson * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the author nor the names of any co-contributors + * 3. Neither the name of the author nor the names of any co-contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * This module provides MI support for per-cpu data. * * Each architecture determines the mapping of logical CPU IDs to physical * CPUs. The requirements of this mapping are as follows: * - Logical CPU IDs must reside in the range 0 ... MAXCPU - 1. * - The mapping is not required to be dense. That is, there may be * gaps in the mappings. * - The platform sets the value of MAXCPU in . * - It is suggested, but not required, that in the non-SMP case, the * platform define MAXCPU to be 1 and define the logical ID of the * sole CPU as 0. */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include #include #include #include #include #include #include #include #include #include #include static MALLOC_DEFINE(M_PCPU, "Per-cpu", "Per-cpu resource accouting."); struct dpcpu_free { uintptr_t df_start; int df_len; TAILQ_ENTRY(dpcpu_free) df_link; }; static DPCPU_DEFINE(char, modspace[DPCPU_MODMIN]); static TAILQ_HEAD(, dpcpu_free) dpcpu_head = TAILQ_HEAD_INITIALIZER(dpcpu_head); static struct sx dpcpu_lock; uintptr_t dpcpu_off[MAXCPU]; struct pcpu *cpuid_to_pcpu[MAXCPU]; struct cpuhead cpuhead = STAILQ_HEAD_INITIALIZER(cpuhead); /* * Initialize the MI portions of a struct pcpu. */ void pcpu_init(struct pcpu *pcpu, int cpuid, size_t size) { bzero(pcpu, size); KASSERT(cpuid >= 0 && cpuid < MAXCPU, ("pcpu_init: invalid cpuid %d", cpuid)); pcpu->pc_cpuid = cpuid; cpuid_to_pcpu[cpuid] = pcpu; STAILQ_INSERT_TAIL(&cpuhead, pcpu, pc_allcpu); cpu_pcpu_init(pcpu, cpuid, size); pcpu->pc_rm_queue.rmq_next = &pcpu->pc_rm_queue; pcpu->pc_rm_queue.rmq_prev = &pcpu->pc_rm_queue; } void dpcpu_init(void *dpcpu, int cpuid) { struct pcpu *pcpu; pcpu = pcpu_find(cpuid); pcpu->pc_dynamic = (uintptr_t)dpcpu - DPCPU_START; /* * Initialize defaults from our linker section. */ memcpy(dpcpu, (void *)DPCPU_START, DPCPU_BYTES); /* * Place it in the global pcpu offset array. */ dpcpu_off[cpuid] = pcpu->pc_dynamic; } static void dpcpu_startup(void *dummy __unused) { struct dpcpu_free *df; df = malloc(sizeof(*df), M_PCPU, M_WAITOK | M_ZERO); df->df_start = (uintptr_t)&DPCPU_NAME(modspace); df->df_len = DPCPU_MODMIN; TAILQ_INSERT_HEAD(&dpcpu_head, df, df_link); sx_init(&dpcpu_lock, "dpcpu alloc lock"); } SYSINIT(dpcpu, SI_SUB_KLD, SI_ORDER_FIRST, dpcpu_startup, 0); /* * UMA_PCPU_ZONE zones, that are available for all kernel * consumers. Right now 64 bit zone is used for counter(9) * and pointer zone is used by flowtable. */ uma_zone_t pcpu_zone_64; uma_zone_t pcpu_zone_ptr; static void pcpu_zones_startup(void) { pcpu_zone_64 = uma_zcreate("64 pcpu", sizeof(uint64_t), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_PCPU); if (sizeof(uint64_t) == sizeof(void *)) pcpu_zone_ptr = pcpu_zone_64; else pcpu_zone_ptr = uma_zcreate("ptr pcpu", sizeof(void *), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_PCPU); } SYSINIT(pcpu_zones, SI_SUB_KMEM, SI_ORDER_ANY, pcpu_zones_startup, NULL); /* * First-fit extent based allocator for allocating space in the per-cpu * region reserved for modules. This is only intended for use by the * kernel linkers to place module linker sets. */ void * dpcpu_alloc(int size) { struct dpcpu_free *df; void *s; s = NULL; size = roundup2(size, sizeof(void *)); sx_xlock(&dpcpu_lock); TAILQ_FOREACH(df, &dpcpu_head, df_link) { if (df->df_len < size) continue; if (df->df_len == size) { s = (void *)df->df_start; TAILQ_REMOVE(&dpcpu_head, df, df_link); free(df, M_PCPU); break; } s = (void *)df->df_start; df->df_len -= size; df->df_start = df->df_start + size; break; } sx_xunlock(&dpcpu_lock); return (s); } /* * Free dynamic per-cpu space at module unload time. */ void dpcpu_free(void *s, int size) { struct dpcpu_free *df; struct dpcpu_free *dn; uintptr_t start; uintptr_t end; size = roundup2(size, sizeof(void *)); start = (uintptr_t)s; end = start + size; /* * Free a region of space and merge it with as many neighbors as * possible. Keeping the list sorted simplifies this operation. */ sx_xlock(&dpcpu_lock); TAILQ_FOREACH(df, &dpcpu_head, df_link) { if (df->df_start > end) break; /* * If we expand at the end of an entry we may have to * merge it with the one following it as well. */ if (df->df_start + df->df_len == start) { df->df_len += size; dn = TAILQ_NEXT(df, df_link); if (df->df_start + df->df_len == dn->df_start) { df->df_len += dn->df_len; TAILQ_REMOVE(&dpcpu_head, dn, df_link); free(dn, M_PCPU); } sx_xunlock(&dpcpu_lock); return; } if (df->df_start == end) { df->df_start = start; df->df_len += size; sx_xunlock(&dpcpu_lock); return; } } dn = malloc(sizeof(*df), M_PCPU, M_WAITOK | M_ZERO); dn->df_start = start; dn->df_len = size; if (df) TAILQ_INSERT_BEFORE(df, dn, df_link); else TAILQ_INSERT_TAIL(&dpcpu_head, dn, df_link); sx_xunlock(&dpcpu_lock); } /* * Initialize the per-cpu storage from an updated linker-set region. */ void dpcpu_copy(void *s, int size) { #ifdef SMP uintptr_t dpcpu; int i; CPU_FOREACH(i) { dpcpu = dpcpu_off[i]; if (dpcpu == 0) continue; memcpy((void *)(dpcpu + (uintptr_t)s), s, size); } #else memcpy((void *)(dpcpu_off[0] + (uintptr_t)s), s, size); #endif } /* * Destroy a struct pcpu. */ void pcpu_destroy(struct pcpu *pcpu) { STAILQ_REMOVE(&cpuhead, pcpu, pcpu, pc_allcpu); cpuid_to_pcpu[pcpu->pc_cpuid] = NULL; dpcpu_off[pcpu->pc_cpuid] = 0; } /* * Locate a struct pcpu by cpu id. */ struct pcpu * pcpu_find(u_int cpuid) { return (cpuid_to_pcpu[cpuid]); } int sysctl_dpcpu_quad(SYSCTL_HANDLER_ARGS) { uintptr_t dpcpu; int64_t count; int i; count = 0; CPU_FOREACH(i) { dpcpu = dpcpu_off[i]; if (dpcpu == 0) continue; count += *(int64_t *)(dpcpu + (uintptr_t)arg1); } return (SYSCTL_OUT(req, &count, sizeof(count))); } int sysctl_dpcpu_long(SYSCTL_HANDLER_ARGS) { uintptr_t dpcpu; long count; int i; count = 0; CPU_FOREACH(i) { dpcpu = dpcpu_off[i]; if (dpcpu == 0) continue; count += *(long *)(dpcpu + (uintptr_t)arg1); } return (SYSCTL_OUT(req, &count, sizeof(count))); } int sysctl_dpcpu_int(SYSCTL_HANDLER_ARGS) { uintptr_t dpcpu; int count; int i; count = 0; CPU_FOREACH(i) { dpcpu = dpcpu_off[i]; if (dpcpu == 0) continue; count += *(int *)(dpcpu + (uintptr_t)arg1); } return (SYSCTL_OUT(req, &count, sizeof(count))); } #ifdef DDB DB_SHOW_COMMAND(dpcpu_off, db_show_dpcpu_off) { int id; CPU_FOREACH(id) { db_printf("dpcpu_off[%2d] = 0x%jx (+ DPCPU_START = %p)\n", id, (uintmax_t)dpcpu_off[id], (void *)(uintptr_t)(dpcpu_off[id] + DPCPU_START)); } } static void show_pcpu(struct pcpu *pc) { struct thread *td; db_printf("cpuid = %d\n", pc->pc_cpuid); db_printf("dynamic pcpu = %p\n", (void *)pc->pc_dynamic); db_printf("curthread = "); td = pc->pc_curthread; if (td != NULL) db_printf("%p: pid %d tid %d \"%s\"\n", td, td->td_proc->p_pid, td->td_tid, td->td_name); else db_printf("none\n"); db_printf("curpcb = %p\n", pc->pc_curpcb); db_printf("fpcurthread = "); td = pc->pc_fpcurthread; if (td != NULL) db_printf("%p: pid %d \"%s\"\n", td, td->td_proc->p_pid, td->td_name); else db_printf("none\n"); db_printf("idlethread = "); td = pc->pc_idlethread; if (td != NULL) db_printf("%p: tid %d \"%s\"\n", td, td->td_tid, td->td_name); else db_printf("none\n"); db_show_mdpcpu(pc); #ifdef VIMAGE db_printf("curvnet = %p\n", pc->pc_curthread->td_vnet); #endif #ifdef WITNESS db_printf("spin locks held:\n"); witness_list_locks(&pc->pc_spinlocks, db_printf); #endif } DB_SHOW_COMMAND(pcpu, db_show_pcpu) { struct pcpu *pc; int id; if (have_addr) id = ((addr >> 4) % 16) * 10 + (addr % 16); else id = PCPU_GET(cpuid); pc = pcpu_find(id); if (pc == NULL) { db_printf("CPU %d not found\n", id); return; } show_pcpu(pc); } DB_SHOW_ALL_COMMAND(pcpu, db_show_cpu_all) { struct pcpu *pc; int id; db_printf("Current CPU: %d\n\n", PCPU_GET(cpuid)); for (id = 0; id <= mp_maxid; id++) { pc = pcpu_find(id); if (pc != NULL) { show_pcpu(pc); db_printf("\n"); } } } DB_SHOW_ALIAS(allpcpu, db_show_cpu_all); #endif Index: head/sys/kern/subr_prf.c =================================================================== --- head/sys/kern/subr_prf.c (revision 305831) +++ head/sys/kern/subr_prf.c (revision 305832) @@ -1,1219 +1,1219 @@ /*- * Copyright (c) 1986, 1988, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)subr_prf.c 8.3 (Berkeley) 1/21/94 */ #include __FBSDID("$FreeBSD$"); #ifdef _KERNEL #include "opt_ddb.h" #include "opt_printf.h" #endif /* _KERNEL */ #include #ifdef _KERNEL #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #endif #include #include #ifdef DDB #include #endif /* * Note that stdarg.h and the ANSI style va_start macro is used for both * ANSI and traditional C compilers. */ #include #ifdef _KERNEL #define TOCONS 0x01 #define TOTTY 0x02 #define TOLOG 0x04 /* Max number conversion buffer length: a u_quad_t in base 2, plus NUL byte. */ #define MAXNBUF (sizeof(intmax_t) * NBBY + 1) struct putchar_arg { int flags; int pri; struct tty *tty; char *p_bufr; size_t n_bufr; char *p_next; size_t remain; }; struct snprintf_arg { char *str; size_t remain; }; extern int log_open; static void msglogchar(int c, int pri); static void msglogstr(char *str, int pri, int filter_cr); static void putchar(int ch, void *arg); static char *ksprintn(char *nbuf, uintmax_t num, int base, int *len, int upper); static void snprintf_func(int ch, void *arg); static int msgbufmapped; /* Set when safe to use msgbuf */ int msgbuftrigger; static int log_console_output = 1; SYSCTL_INT(_kern, OID_AUTO, log_console_output, CTLFLAG_RWTUN, &log_console_output, 0, "Duplicate console output to the syslog"); /* * See the comment in log_console() below for more explanation of this. */ static int log_console_add_linefeed; SYSCTL_INT(_kern, OID_AUTO, log_console_add_linefeed, CTLFLAG_RWTUN, &log_console_add_linefeed, 0, "log_console() adds extra newlines"); static int always_console_output; SYSCTL_INT(_kern, OID_AUTO, always_console_output, CTLFLAG_RWTUN, &always_console_output, 0, "Always output to console despite TIOCCONS"); /* * Warn that a system table is full. */ void tablefull(const char *tab) { log(LOG_ERR, "%s: table is full\n", tab); } /* * Uprintf prints to the controlling terminal for the current process. */ int uprintf(const char *fmt, ...) { va_list ap; struct putchar_arg pca; struct proc *p; struct thread *td; int retval; td = curthread; if (TD_IS_IDLETHREAD(td)) return (0); sx_slock(&proctree_lock); p = td->td_proc; PROC_LOCK(p); if ((p->p_flag & P_CONTROLT) == 0) { PROC_UNLOCK(p); sx_sunlock(&proctree_lock); return (0); } SESS_LOCK(p->p_session); pca.tty = p->p_session->s_ttyp; SESS_UNLOCK(p->p_session); PROC_UNLOCK(p); if (pca.tty == NULL) { sx_sunlock(&proctree_lock); return (0); } pca.flags = TOTTY; pca.p_bufr = NULL; va_start(ap, fmt); tty_lock(pca.tty); sx_sunlock(&proctree_lock); retval = kvprintf(fmt, putchar, &pca, 10, ap); tty_unlock(pca.tty); va_end(ap); return (retval); } /* * tprintf and vtprintf print on the controlling terminal associated with the * given session, possibly to the log as well. */ void tprintf(struct proc *p, int pri, const char *fmt, ...) { va_list ap; va_start(ap, fmt); vtprintf(p, pri, fmt, ap); va_end(ap); } void vtprintf(struct proc *p, int pri, const char *fmt, va_list ap) { struct tty *tp = NULL; int flags = 0; struct putchar_arg pca; struct session *sess = NULL; sx_slock(&proctree_lock); if (pri != -1) flags |= TOLOG; if (p != NULL) { PROC_LOCK(p); if (p->p_flag & P_CONTROLT && p->p_session->s_ttyvp) { sess = p->p_session; sess_hold(sess); PROC_UNLOCK(p); tp = sess->s_ttyp; if (tp != NULL && tty_checkoutq(tp)) flags |= TOTTY; else tp = NULL; } else PROC_UNLOCK(p); } pca.pri = pri; pca.tty = tp; pca.flags = flags; pca.p_bufr = NULL; if (pca.tty != NULL) tty_lock(pca.tty); sx_sunlock(&proctree_lock); kvprintf(fmt, putchar, &pca, 10, ap); if (pca.tty != NULL) tty_unlock(pca.tty); if (sess != NULL) sess_release(sess); msgbuftrigger = 1; } /* * Ttyprintf displays a message on a tty; it should be used only by * the tty driver, or anything that knows the underlying tty will not * be revoke(2)'d away. Other callers should use tprintf. */ int ttyprintf(struct tty *tp, const char *fmt, ...) { va_list ap; struct putchar_arg pca; int retval; va_start(ap, fmt); pca.tty = tp; pca.flags = TOTTY; pca.p_bufr = NULL; retval = kvprintf(fmt, putchar, &pca, 10, ap); va_end(ap); return (retval); } static int _vprintf(int level, int flags, const char *fmt, va_list ap) { struct putchar_arg pca; int retval; #ifdef PRINTF_BUFR_SIZE char bufr[PRINTF_BUFR_SIZE]; #endif pca.tty = NULL; pca.pri = level; pca.flags = flags; #ifdef PRINTF_BUFR_SIZE pca.p_bufr = bufr; pca.p_next = pca.p_bufr; pca.n_bufr = sizeof(bufr); pca.remain = sizeof(bufr); *pca.p_next = '\0'; #else /* Don't buffer console output. */ pca.p_bufr = NULL; #endif retval = kvprintf(fmt, putchar, &pca, 10, ap); #ifdef PRINTF_BUFR_SIZE /* Write any buffered console/log output: */ if (*pca.p_bufr != '\0') { if (pca.flags & TOLOG) msglogstr(pca.p_bufr, level, /*filter_cr*/1); if (pca.flags & TOCONS) cnputs(pca.p_bufr); } #endif return (retval); } /* * Log writes to the log buffer, and guarantees not to sleep (so can be * called by interrupt routines). If there is no process reading the * log yet, it writes to the console also. */ void log(int level, const char *fmt, ...) { va_list ap; va_start(ap, fmt); vlog(level, fmt, ap); va_end(ap); } void vlog(int level, const char *fmt, va_list ap) { (void)_vprintf(level, log_open ? TOLOG : TOCONS | TOLOG, fmt, ap); msgbuftrigger = 1; } #define CONSCHUNK 128 void log_console(struct uio *uio) { int c, error, nl; char *consbuffer; int pri; if (!log_console_output) return; pri = LOG_INFO | LOG_CONSOLE; uio = cloneuio(uio); consbuffer = malloc(CONSCHUNK, M_TEMP, M_WAITOK); nl = 0; while (uio->uio_resid > 0) { c = imin(uio->uio_resid, CONSCHUNK - 1); error = uiomove(consbuffer, c, uio); if (error != 0) break; /* Make sure we're NUL-terminated */ consbuffer[c] = '\0'; if (consbuffer[c - 1] == '\n') nl = 1; else nl = 0; msglogstr(consbuffer, pri, /*filter_cr*/ 1); } /* * The previous behavior in log_console() is preserved when * log_console_add_linefeed is non-zero. For that behavior, if an * individual console write came in that was not terminated with a * line feed, it would add a line feed. * * This results in different data in the message buffer than * appears on the system console (which doesn't add extra line feed * characters). * * A number of programs and rc scripts write a line feed, or a period * and a line feed when they have completed their operation. On * the console, this looks seamless, but when displayed with * 'dmesg -a', you wind up with output that looks like this: * * Updating motd: * . * * On the console, it looks like this: * Updating motd:. * * We could add logic to detect that situation, or just not insert * the extra newlines. Set the kern.log_console_add_linefeed * sysctl/tunable variable to get the old behavior. */ if (!nl && log_console_add_linefeed) { consbuffer[0] = '\n'; consbuffer[1] = '\0'; msglogstr(consbuffer, pri, /*filter_cr*/ 1); } msgbuftrigger = 1; free(uio, M_IOV); free(consbuffer, M_TEMP); return; } int printf(const char *fmt, ...) { va_list ap; int retval; va_start(ap, fmt); retval = vprintf(fmt, ap); va_end(ap); return (retval); } int vprintf(const char *fmt, va_list ap) { int retval; retval = _vprintf(-1, TOCONS | TOLOG, fmt, ap); if (!panicstr) msgbuftrigger = 1; return (retval); } static void putbuf(int c, struct putchar_arg *ap) { /* Check if no console output buffer was provided. */ if (ap->p_bufr == NULL) { /* Output direct to the console. */ if (ap->flags & TOCONS) cnputc(c); if (ap->flags & TOLOG) msglogchar(c, ap->pri); } else { /* Buffer the character: */ *ap->p_next++ = c; ap->remain--; /* Always leave the buffer zero terminated. */ *ap->p_next = '\0'; /* Check if the buffer needs to be flushed. */ if (ap->remain == 2 || c == '\n') { if (ap->flags & TOLOG) msglogstr(ap->p_bufr, ap->pri, /*filter_cr*/1); if (ap->flags & TOCONS) { if ((panicstr == NULL) && (constty != NULL)) msgbuf_addstr(&consmsgbuf, -1, ap->p_bufr, /*filter_cr*/ 0); if ((constty == NULL) ||(always_console_output)) cnputs(ap->p_bufr); } ap->p_next = ap->p_bufr; ap->remain = ap->n_bufr; *ap->p_next = '\0'; } /* * Since we fill the buffer up one character at a time, * this should not happen. We should always catch it when * ap->remain == 2 (if not sooner due to a newline), flush * the buffer and move on. One way this could happen is * if someone sets PRINTF_BUFR_SIZE to 1 or something * similarly silly. */ KASSERT(ap->remain > 2, ("Bad buffer logic, remain = %zd", ap->remain)); } } /* * Print a character on console or users terminal. If destination is * the console then the last bunch of characters are saved in msgbuf for * inspection later. */ static void putchar(int c, void *arg) { struct putchar_arg *ap = (struct putchar_arg*) arg; struct tty *tp = ap->tty; int flags = ap->flags; /* Don't use the tty code after a panic or while in ddb. */ if (kdb_active) { if (c != '\0') cnputc(c); return; } if ((flags & TOTTY) && tp != NULL && panicstr == NULL) tty_putchar(tp, c); if ((flags & (TOCONS | TOLOG)) && c != '\0') putbuf(c, ap); } /* * Scaled down version of sprintf(3). */ int sprintf(char *buf, const char *cfmt, ...) { int retval; va_list ap; va_start(ap, cfmt); retval = kvprintf(cfmt, NULL, (void *)buf, 10, ap); buf[retval] = '\0'; va_end(ap); return (retval); } /* * Scaled down version of vsprintf(3). */ int vsprintf(char *buf, const char *cfmt, va_list ap) { int retval; retval = kvprintf(cfmt, NULL, (void *)buf, 10, ap); buf[retval] = '\0'; return (retval); } /* * Scaled down version of snprintf(3). */ int snprintf(char *str, size_t size, const char *format, ...) { int retval; va_list ap; va_start(ap, format); retval = vsnprintf(str, size, format, ap); va_end(ap); return(retval); } /* * Scaled down version of vsnprintf(3). */ int vsnprintf(char *str, size_t size, const char *format, va_list ap) { struct snprintf_arg info; int retval; info.str = str; info.remain = size; retval = kvprintf(format, snprintf_func, &info, 10, ap); if (info.remain >= 1) *info.str++ = '\0'; return (retval); } /* * Kernel version which takes radix argument vsnprintf(3). */ int vsnrprintf(char *str, size_t size, int radix, const char *format, va_list ap) { struct snprintf_arg info; int retval; info.str = str; info.remain = size; retval = kvprintf(format, snprintf_func, &info, radix, ap); if (info.remain >= 1) *info.str++ = '\0'; return (retval); } static void snprintf_func(int ch, void *arg) { struct snprintf_arg *const info = arg; if (info->remain >= 2) { *info->str++ = ch; info->remain--; } } /* * Put a NUL-terminated ASCII number (base <= 36) in a buffer in reverse * order; return an optional length and a pointer to the last character * written in the buffer (i.e., the first character of the string). * The buffer pointed to by `nbuf' must have length >= MAXNBUF. */ static char * ksprintn(char *nbuf, uintmax_t num, int base, int *lenp, int upper) { char *p, c; p = nbuf; *p = '\0'; do { c = hex2ascii(num % base); *++p = upper ? toupper(c) : c; } while (num /= base); if (lenp) *lenp = p - nbuf; return (p); } /* * Scaled down version of printf(3). * * Two additional formats: * * The format %b is supported to decode error registers. * Its usage is: * * printf("reg=%b\n", regval, "*"); * * where is the output base expressed as a control character, e.g. * \10 gives octal; \20 gives hex. Each arg is a sequence of characters, * the first of which gives the bit number to be inspected (origin 1), and * the next characters (up to a control character, i.e. a character <= 32), * give the name of the register. Thus: * * kvprintf("reg=%b\n", 3, "\10\2BITTWO\1BITONE"); * * would produce output: * * reg=3 * * XXX: %D -- Hexdump, takes pointer and separator string: * ("%6D", ptr, ":") -> XX:XX:XX:XX:XX:XX * ("%*D", len, ptr, " " -> XX XX XX XX ... */ int kvprintf(char const *fmt, void (*func)(int, void*), void *arg, int radix, va_list ap) { #define PCHAR(c) {int cc=(c); if (func) (*func)(cc,arg); else *d++ = cc; retval++; } char nbuf[MAXNBUF]; char *d; const char *p, *percent, *q; u_char *up; int ch, n; uintmax_t num; int base, lflag, qflag, tmp, width, ladjust, sharpflag, neg, sign, dot; int cflag, hflag, jflag, tflag, zflag; int dwidth, upper; char padc; int stop = 0, retval = 0; num = 0; if (!func) d = (char *) arg; else d = NULL; if (fmt == NULL) fmt = "(fmt null)\n"; if (radix < 2 || radix > 36) radix = 10; for (;;) { padc = ' '; width = 0; while ((ch = (u_char)*fmt++) != '%' || stop) { if (ch == '\0') return (retval); PCHAR(ch); } percent = fmt - 1; qflag = 0; lflag = 0; ladjust = 0; sharpflag = 0; neg = 0; sign = 0; dot = 0; dwidth = 0; upper = 0; cflag = 0; hflag = 0; jflag = 0; tflag = 0; zflag = 0; reswitch: switch (ch = (u_char)*fmt++) { case '.': dot = 1; goto reswitch; case '#': sharpflag = 1; goto reswitch; case '+': sign = 1; goto reswitch; case '-': ladjust = 1; goto reswitch; case '%': PCHAR(ch); break; case '*': if (!dot) { width = va_arg(ap, int); if (width < 0) { ladjust = !ladjust; width = -width; } } else { dwidth = va_arg(ap, int); } goto reswitch; case '0': if (!dot) { padc = '0'; goto reswitch; } case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9': for (n = 0;; ++fmt) { n = n * 10 + ch - '0'; ch = *fmt; if (ch < '0' || ch > '9') break; } if (dot) dwidth = n; else width = n; goto reswitch; case 'b': num = (u_int)va_arg(ap, int); p = va_arg(ap, char *); for (q = ksprintn(nbuf, num, *p++, NULL, 0); *q;) PCHAR(*q--); if (num == 0) break; for (tmp = 0; *p;) { n = *p++; if (num & (1 << (n - 1))) { PCHAR(tmp ? ',' : '<'); for (; (n = *p) > ' '; ++p) PCHAR(n); tmp = 1; } else for (; *p > ' '; ++p) continue; } if (tmp) PCHAR('>'); break; case 'c': width -= 1; if (!ladjust && width > 0) while (width--) PCHAR(padc); PCHAR(va_arg(ap, int)); if (ladjust && width > 0) while (width--) PCHAR(padc); break; case 'D': up = va_arg(ap, u_char *); p = va_arg(ap, char *); if (!width) width = 16; while(width--) { PCHAR(hex2ascii(*up >> 4)); PCHAR(hex2ascii(*up & 0x0f)); up++; if (width) for (q=p;*q;q++) PCHAR(*q); } break; case 'd': case 'i': base = 10; sign = 1; goto handle_sign; case 'h': if (hflag) { hflag = 0; cflag = 1; } else hflag = 1; goto reswitch; case 'j': jflag = 1; goto reswitch; case 'l': if (lflag) { lflag = 0; qflag = 1; } else lflag = 1; goto reswitch; case 'n': if (jflag) *(va_arg(ap, intmax_t *)) = retval; else if (qflag) *(va_arg(ap, quad_t *)) = retval; else if (lflag) *(va_arg(ap, long *)) = retval; else if (zflag) *(va_arg(ap, size_t *)) = retval; else if (hflag) *(va_arg(ap, short *)) = retval; else if (cflag) *(va_arg(ap, char *)) = retval; else *(va_arg(ap, int *)) = retval; break; case 'o': base = 8; goto handle_nosign; case 'p': base = 16; sharpflag = (width == 0); sign = 0; num = (uintptr_t)va_arg(ap, void *); goto number; case 'q': qflag = 1; goto reswitch; case 'r': base = radix; if (sign) goto handle_sign; goto handle_nosign; case 's': p = va_arg(ap, char *); if (p == NULL) p = "(null)"; if (!dot) n = strlen (p); else for (n = 0; n < dwidth && p[n]; n++) continue; width -= n; if (!ladjust && width > 0) while (width--) PCHAR(padc); while (n--) PCHAR(*p++); if (ladjust && width > 0) while (width--) PCHAR(padc); break; case 't': tflag = 1; goto reswitch; case 'u': base = 10; goto handle_nosign; case 'X': upper = 1; case 'x': base = 16; goto handle_nosign; case 'y': base = 16; sign = 1; goto handle_sign; case 'z': zflag = 1; goto reswitch; handle_nosign: sign = 0; if (jflag) num = va_arg(ap, uintmax_t); else if (qflag) num = va_arg(ap, u_quad_t); else if (tflag) num = va_arg(ap, ptrdiff_t); else if (lflag) num = va_arg(ap, u_long); else if (zflag) num = va_arg(ap, size_t); else if (hflag) num = (u_short)va_arg(ap, int); else if (cflag) num = (u_char)va_arg(ap, int); else num = va_arg(ap, u_int); goto number; handle_sign: if (jflag) num = va_arg(ap, intmax_t); else if (qflag) num = va_arg(ap, quad_t); else if (tflag) num = va_arg(ap, ptrdiff_t); else if (lflag) num = va_arg(ap, long); else if (zflag) num = va_arg(ap, ssize_t); else if (hflag) num = (short)va_arg(ap, int); else if (cflag) num = (char)va_arg(ap, int); else num = va_arg(ap, int); number: if (sign && (intmax_t)num < 0) { neg = 1; num = -(intmax_t)num; } p = ksprintn(nbuf, num, base, &n, upper); tmp = 0; if (sharpflag && num != 0) { if (base == 8) tmp++; else if (base == 16) tmp += 2; } if (neg) tmp++; if (!ladjust && padc == '0') dwidth = width - tmp; width -= tmp + imax(dwidth, n); dwidth -= n; if (!ladjust) while (width-- > 0) PCHAR(' '); if (neg) PCHAR('-'); if (sharpflag && num != 0) { if (base == 8) { PCHAR('0'); } else if (base == 16) { PCHAR('0'); PCHAR('x'); } } while (dwidth-- > 0) PCHAR('0'); while (*p) PCHAR(*p--); if (ladjust) while (width-- > 0) PCHAR(' '); break; default: while (percent < fmt) PCHAR(*percent++); /* * Since we ignore a formatting argument it is no * longer safe to obey the remaining formatting * arguments as the arguments will no longer match * the format specs. */ stop = 1; break; } } #undef PCHAR } /* * Put character in log buffer with a particular priority. */ static void msglogchar(int c, int pri) { static int lastpri = -1; static int dangling; char nbuf[MAXNBUF]; char *p; if (!msgbufmapped) return; if (c == '\0' || c == '\r') return; if (pri != -1 && pri != lastpri) { if (dangling) { msgbuf_addchar(msgbufp, '\n'); dangling = 0; } msgbuf_addchar(msgbufp, '<'); for (p = ksprintn(nbuf, (uintmax_t)pri, 10, NULL, 0); *p;) msgbuf_addchar(msgbufp, *p--); msgbuf_addchar(msgbufp, '>'); lastpri = pri; } msgbuf_addchar(msgbufp, c); if (c == '\n') { dangling = 0; lastpri = -1; } else { dangling = 1; } } static void msglogstr(char *str, int pri, int filter_cr) { if (!msgbufmapped) return; msgbuf_addstr(msgbufp, pri, str, filter_cr); } void msgbufinit(void *ptr, int size) { char *cp; static struct msgbuf *oldp = NULL; size -= sizeof(*msgbufp); cp = (char *)ptr; msgbufp = (struct msgbuf *)(cp + size); msgbuf_reinit(msgbufp, cp, size); if (msgbufmapped && oldp != msgbufp) msgbuf_copy(oldp, msgbufp); msgbufmapped = 1; oldp = msgbufp; } static int unprivileged_read_msgbuf = 1; SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_read_msgbuf, CTLFLAG_RW, &unprivileged_read_msgbuf, 0, "Unprivileged processes may read the kernel message buffer"); /* Sysctls for accessing/clearing the msgbuf */ static int sysctl_kern_msgbuf(SYSCTL_HANDLER_ARGS) { char buf[128]; u_int seq; int error, len; if (!unprivileged_read_msgbuf) { error = priv_check(req->td, PRIV_MSGBUF); if (error) return (error); } /* Read the whole buffer, one chunk at a time. */ mtx_lock(&msgbuf_lock); msgbuf_peekbytes(msgbufp, NULL, 0, &seq); for (;;) { len = msgbuf_peekbytes(msgbufp, buf, sizeof(buf), &seq); mtx_unlock(&msgbuf_lock); if (len == 0) return (SYSCTL_OUT(req, "", 1)); /* add nulterm */ error = sysctl_handle_opaque(oidp, buf, len, req); if (error) return (error); mtx_lock(&msgbuf_lock); } } SYSCTL_PROC(_kern, OID_AUTO, msgbuf, CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_msgbuf, "A", "Contents of kernel message buffer"); static int msgbuf_clearflag; static int sysctl_kern_msgbuf_clear(SYSCTL_HANDLER_ARGS) { int error; error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2, req); if (!error && req->newptr) { mtx_lock(&msgbuf_lock); msgbuf_clear(msgbufp); mtx_unlock(&msgbuf_lock); msgbuf_clearflag = 0; } return (error); } SYSCTL_PROC(_kern, OID_AUTO, msgbuf_clear, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE | CTLFLAG_MPSAFE, &msgbuf_clearflag, 0, sysctl_kern_msgbuf_clear, "I", "Clear kernel message buffer"); #ifdef DDB DB_SHOW_COMMAND(msgbuf, db_show_msgbuf) { int i, j; if (!msgbufmapped) { db_printf("msgbuf not mapped yet\n"); return; } db_printf("msgbufp = %p\n", msgbufp); db_printf("magic = %x, size = %d, r= %u, w = %u, ptr = %p, cksum= %u\n", msgbufp->msg_magic, msgbufp->msg_size, msgbufp->msg_rseq, msgbufp->msg_wseq, msgbufp->msg_ptr, msgbufp->msg_cksum); for (i = 0; i < msgbufp->msg_size && !db_pager_quit; i++) { j = MSGBUF_SEQ_TO_POS(msgbufp, i + msgbufp->msg_rseq); db_printf("%c", msgbufp->msg_ptr[j]); } db_printf("\n"); } #endif /* DDB */ void hexdump(const void *ptr, int length, const char *hdr, int flags) { int i, j, k; int cols; const unsigned char *cp; char delim; if ((flags & HD_DELIM_MASK) != 0) delim = (flags & HD_DELIM_MASK) >> 8; else delim = ' '; if ((flags & HD_COLUMN_MASK) != 0) cols = flags & HD_COLUMN_MASK; else cols = 16; cp = ptr; for (i = 0; i < length; i+= cols) { if (hdr != NULL) printf("%s", hdr); if ((flags & HD_OMIT_COUNT) == 0) printf("%04x ", i); if ((flags & HD_OMIT_HEX) == 0) { for (j = 0; j < cols; j++) { k = i + j; if (k < length) printf("%c%02x", delim, cp[k]); else printf(" "); } } if ((flags & HD_OMIT_CHARS) == 0) { printf(" |"); for (j = 0; j < cols; j++) { k = i + j; if (k >= length) printf(" "); else if (cp[k] >= ' ' && cp[k] <= '~') printf("%c", cp[k]); else printf("."); } printf("|"); } printf("\n"); } } #endif /* _KERNEL */ void sbuf_hexdump(struct sbuf *sb, const void *ptr, int length, const char *hdr, int flags) { int i, j, k; int cols; const unsigned char *cp; char delim; if ((flags & HD_DELIM_MASK) != 0) delim = (flags & HD_DELIM_MASK) >> 8; else delim = ' '; if ((flags & HD_COLUMN_MASK) != 0) cols = flags & HD_COLUMN_MASK; else cols = 16; cp = ptr; for (i = 0; i < length; i+= cols) { if (hdr != NULL) sbuf_printf(sb, "%s", hdr); if ((flags & HD_OMIT_COUNT) == 0) sbuf_printf(sb, "%04x ", i); if ((flags & HD_OMIT_HEX) == 0) { for (j = 0; j < cols; j++) { k = i + j; if (k < length) sbuf_printf(sb, "%c%02x", delim, cp[k]); else sbuf_printf(sb, " "); } } if ((flags & HD_OMIT_CHARS) == 0) { sbuf_printf(sb, " |"); for (j = 0; j < cols; j++) { k = i + j; if (k >= length) sbuf_printf(sb, " "); else if (cp[k] >= ' ' && cp[k] <= '~') sbuf_printf(sb, "%c", cp[k]); else sbuf_printf(sb, "."); } sbuf_printf(sb, "|"); } sbuf_printf(sb, "\n"); } } #ifdef _KERNEL void counted_warning(unsigned *counter, const char *msg) { struct thread *td; unsigned c; for (;;) { c = *counter; if (c == 0) break; if (atomic_cmpset_int(counter, c, c - 1)) { td = curthread; log(LOG_INFO, "pid %d (%s) %s%s\n", td->td_proc->p_pid, td->td_name, msg, c > 1 ? "" : " - not logging anymore"); break; } } } #endif Index: head/sys/kern/subr_prof.c =================================================================== --- head/sys/kern/subr_prof.c (revision 305831) +++ head/sys/kern/subr_prof.c (revision 305832) @@ -1,543 +1,543 @@ /*- * Copyright (c) 1982, 1986, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)subr_prof.c 8.3 (Berkeley) 9/23/93 */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #ifdef GPROF #include #include #undef MCOUNT static MALLOC_DEFINE(M_GPROF, "gprof", "kernel profiling buffer"); static void kmstartup(void *); SYSINIT(kmem, SI_SUB_KPROF, SI_ORDER_FIRST, kmstartup, NULL); struct gmonparam _gmonparam = { GMON_PROF_OFF }; #ifdef GUPROF void nullfunc_loop_profiled() { int i; for (i = 0; i < CALIB_SCALE; i++) nullfunc_profiled(); } #define nullfunc_loop_profiled_end nullfunc_profiled /* XXX */ void nullfunc_profiled() { } #endif /* GUPROF */ /* * Update the histograms to support extending the text region arbitrarily. * This is done slightly naively (no sparse regions), so will waste slight * amounts of memory, but will overall work nicely enough to allow profiling * of KLDs. */ void kmupetext(uintfptr_t nhighpc) { struct gmonparam np; /* slightly large */ struct gmonparam *p = &_gmonparam; char *cp; GIANT_REQUIRED; bcopy(p, &np, sizeof(*p)); np.highpc = ROUNDUP(nhighpc, HISTFRACTION * sizeof(HISTCOUNTER)); if (np.highpc <= p->highpc) return; np.textsize = np.highpc - p->lowpc; np.kcountsize = np.textsize / HISTFRACTION; np.hashfraction = HASHFRACTION; np.fromssize = np.textsize / HASHFRACTION; np.tolimit = np.textsize * ARCDENSITY / 100; if (np.tolimit < MINARCS) np.tolimit = MINARCS; else if (np.tolimit > MAXARCS) np.tolimit = MAXARCS; np.tossize = np.tolimit * sizeof(struct tostruct); cp = malloc(np.kcountsize + np.fromssize + np.tossize, M_GPROF, M_WAITOK); /* * Check for something else extending highpc while we slept. */ if (np.highpc <= p->highpc) { free(cp, M_GPROF); return; } np.tos = (struct tostruct *)cp; cp += np.tossize; np.kcount = (HISTCOUNTER *)cp; cp += np.kcountsize; np.froms = (u_short *)cp; #ifdef GUPROF /* Reinitialize pointers to overhead counters. */ np.cputime_count = &KCOUNT(&np, PC_TO_I(&np, cputime)); np.mcount_count = &KCOUNT(&np, PC_TO_I(&np, mcount)); np.mexitcount_count = &KCOUNT(&np, PC_TO_I(&np, mexitcount)); #endif critical_enter(); bcopy(p->tos, np.tos, p->tossize); bzero((char *)np.tos + p->tossize, np.tossize - p->tossize); bcopy(p->kcount, np.kcount, p->kcountsize); bzero((char *)np.kcount + p->kcountsize, np.kcountsize - p->kcountsize); bcopy(p->froms, np.froms, p->fromssize); bzero((char *)np.froms + p->fromssize, np.fromssize - p->fromssize); cp = (char *)p->tos; bcopy(&np, p, sizeof(*p)); critical_exit(); free(cp, M_GPROF); } static void kmstartup(dummy) void *dummy; { char *cp; struct gmonparam *p = &_gmonparam; #ifdef GUPROF int cputime_overhead; int empty_loop_time; int i; int mcount_overhead; int mexitcount_overhead; int nullfunc_loop_overhead; int nullfunc_loop_profiled_time; uintfptr_t tmp_addr; #endif /* * Round lowpc and highpc to multiples of the density we're using * so the rest of the scaling (here and in gprof) stays in ints. */ p->lowpc = ROUNDDOWN((u_long)btext, HISTFRACTION * sizeof(HISTCOUNTER)); p->highpc = ROUNDUP((u_long)etext, HISTFRACTION * sizeof(HISTCOUNTER)); p->textsize = p->highpc - p->lowpc; printf("Profiling kernel, textsize=%lu [%jx..%jx]\n", p->textsize, (uintmax_t)p->lowpc, (uintmax_t)p->highpc); p->kcountsize = p->textsize / HISTFRACTION; p->hashfraction = HASHFRACTION; p->fromssize = p->textsize / HASHFRACTION; p->tolimit = p->textsize * ARCDENSITY / 100; if (p->tolimit < MINARCS) p->tolimit = MINARCS; else if (p->tolimit > MAXARCS) p->tolimit = MAXARCS; p->tossize = p->tolimit * sizeof(struct tostruct); cp = (char *)malloc(p->kcountsize + p->fromssize + p->tossize, M_GPROF, M_WAITOK | M_ZERO); p->tos = (struct tostruct *)cp; cp += p->tossize; p->kcount = (HISTCOUNTER *)cp; cp += p->kcountsize; p->froms = (u_short *)cp; p->histcounter_type = FUNCTION_ALIGNMENT / HISTFRACTION * NBBY; #ifdef GUPROF /* Signed counters. */ p->histcounter_type = -p->histcounter_type; /* Initialize pointers to overhead counters. */ p->cputime_count = &KCOUNT(p, PC_TO_I(p, cputime)); p->mcount_count = &KCOUNT(p, PC_TO_I(p, mcount)); p->mexitcount_count = &KCOUNT(p, PC_TO_I(p, mexitcount)); /* * Disable interrupts to avoid interference while we calibrate * things. */ critical_enter(); /* * Determine overheads. * XXX this needs to be repeated for each useful timer/counter. */ cputime_overhead = 0; startguprof(p); for (i = 0; i < CALIB_SCALE; i++) cputime_overhead += cputime(); empty_loop(); startguprof(p); empty_loop(); empty_loop_time = cputime(); nullfunc_loop_profiled(); /* * Start profiling. There won't be any normal function calls since * interrupts are disabled, but we will call the profiling routines * directly to determine their overheads. */ p->state = GMON_PROF_HIRES; startguprof(p); nullfunc_loop_profiled(); startguprof(p); for (i = 0; i < CALIB_SCALE; i++) MCOUNT_OVERHEAD(sys_profil); mcount_overhead = KCOUNT(p, PC_TO_I(p, sys_profil)); startguprof(p); for (i = 0; i < CALIB_SCALE; i++) MEXITCOUNT_OVERHEAD(); MEXITCOUNT_OVERHEAD_GETLABEL(tmp_addr); mexitcount_overhead = KCOUNT(p, PC_TO_I(p, tmp_addr)); p->state = GMON_PROF_OFF; stopguprof(p); critical_exit(); nullfunc_loop_profiled_time = 0; for (tmp_addr = (uintfptr_t)nullfunc_loop_profiled; tmp_addr < (uintfptr_t)nullfunc_loop_profiled_end; tmp_addr += HISTFRACTION * sizeof(HISTCOUNTER)) nullfunc_loop_profiled_time += KCOUNT(p, PC_TO_I(p, tmp_addr)); #define CALIB_DOSCALE(count) (((count) + CALIB_SCALE / 3) / CALIB_SCALE) #define c2n(count, freq) ((int)((count) * 1000000000LL / freq)) printf("cputime %d, empty_loop %d, nullfunc_loop_profiled %d, mcount %d, mexitcount %d\n", CALIB_DOSCALE(c2n(cputime_overhead, p->profrate)), CALIB_DOSCALE(c2n(empty_loop_time, p->profrate)), CALIB_DOSCALE(c2n(nullfunc_loop_profiled_time, p->profrate)), CALIB_DOSCALE(c2n(mcount_overhead, p->profrate)), CALIB_DOSCALE(c2n(mexitcount_overhead, p->profrate))); cputime_overhead -= empty_loop_time; mcount_overhead -= empty_loop_time; mexitcount_overhead -= empty_loop_time; /*- * Profiling overheads are determined by the times between the * following events: * MC1: mcount() is called * MC2: cputime() (called from mcount()) latches the timer * MC3: mcount() completes * ME1: mexitcount() is called * ME2: cputime() (called from mexitcount()) latches the timer * ME3: mexitcount() completes. * The times between the events vary slightly depending on instruction * combination and cache misses, etc. Attempt to determine the * minimum times. These can be subtracted from the profiling times * without much risk of reducing the profiling times below what they * would be when profiling is not configured. Abbreviate: * ab = minimum time between MC1 and MC3 * a = minimum time between MC1 and MC2 * b = minimum time between MC2 and MC3 * cd = minimum time between ME1 and ME3 * c = minimum time between ME1 and ME2 * d = minimum time between ME2 and ME3. * These satisfy the relations: * ab <= mcount_overhead (just measured) * a + b <= ab * cd <= mexitcount_overhead (just measured) * c + d <= cd * a + d <= nullfunc_loop_profiled_time (just measured) * a >= 0, b >= 0, c >= 0, d >= 0. * Assume that ab and cd are equal to the minimums. */ p->cputime_overhead = CALIB_DOSCALE(cputime_overhead); p->mcount_overhead = CALIB_DOSCALE(mcount_overhead - cputime_overhead); p->mexitcount_overhead = CALIB_DOSCALE(mexitcount_overhead - cputime_overhead); nullfunc_loop_overhead = nullfunc_loop_profiled_time - empty_loop_time; p->mexitcount_post_overhead = CALIB_DOSCALE((mcount_overhead - nullfunc_loop_overhead) / 4); p->mexitcount_pre_overhead = p->mexitcount_overhead + p->cputime_overhead - p->mexitcount_post_overhead; p->mcount_pre_overhead = CALIB_DOSCALE(nullfunc_loop_overhead) - p->mexitcount_post_overhead; p->mcount_post_overhead = p->mcount_overhead + p->cputime_overhead - p->mcount_pre_overhead; printf( "Profiling overheads: mcount: %d+%d, %d+%d; mexitcount: %d+%d, %d+%d nsec\n", c2n(p->cputime_overhead, p->profrate), c2n(p->mcount_overhead, p->profrate), c2n(p->mcount_pre_overhead, p->profrate), c2n(p->mcount_post_overhead, p->profrate), c2n(p->cputime_overhead, p->profrate), c2n(p->mexitcount_overhead, p->profrate), c2n(p->mexitcount_pre_overhead, p->profrate), c2n(p->mexitcount_post_overhead, p->profrate)); printf( "Profiling overheads: mcount: %d+%d, %d+%d; mexitcount: %d+%d, %d+%d cycles\n", p->cputime_overhead, p->mcount_overhead, p->mcount_pre_overhead, p->mcount_post_overhead, p->cputime_overhead, p->mexitcount_overhead, p->mexitcount_pre_overhead, p->mexitcount_post_overhead); #endif /* GUPROF */ } /* * Return kernel profiling information. */ static int sysctl_kern_prof(SYSCTL_HANDLER_ARGS) { int *name = (int *) arg1; u_int namelen = arg2; struct gmonparam *gp = &_gmonparam; int error; int state; /* all sysctl names at this level are terminal */ if (namelen != 1) return (ENOTDIR); /* overloaded */ switch (name[0]) { case GPROF_STATE: state = gp->state; error = sysctl_handle_int(oidp, &state, 0, req); if (error) return (error); if (!req->newptr) return (0); if (state == GMON_PROF_OFF) { gp->state = state; PROC_LOCK(&proc0); stopprofclock(&proc0); PROC_UNLOCK(&proc0); stopguprof(gp); } else if (state == GMON_PROF_ON) { gp->state = GMON_PROF_OFF; stopguprof(gp); gp->profrate = profhz; PROC_LOCK(&proc0); startprofclock(&proc0); PROC_UNLOCK(&proc0); gp->state = state; #ifdef GUPROF } else if (state == GMON_PROF_HIRES) { gp->state = GMON_PROF_OFF; PROC_LOCK(&proc0); stopprofclock(&proc0); PROC_UNLOCK(&proc0); startguprof(gp); gp->state = state; #endif } else if (state != gp->state) return (EINVAL); return (0); case GPROF_COUNT: return (sysctl_handle_opaque(oidp, gp->kcount, gp->kcountsize, req)); case GPROF_FROMS: return (sysctl_handle_opaque(oidp, gp->froms, gp->fromssize, req)); case GPROF_TOS: return (sysctl_handle_opaque(oidp, gp->tos, gp->tossize, req)); case GPROF_GMONPARAM: return (sysctl_handle_opaque(oidp, gp, sizeof *gp, req)); default: return (EOPNOTSUPP); } /* NOTREACHED */ } static SYSCTL_NODE(_kern, KERN_PROF, prof, CTLFLAG_RW, sysctl_kern_prof, ""); #endif /* GPROF */ /* * Profiling system call. * * The scale factor is a fixed point number with 16 bits of fraction, so that * 1.0 is represented as 0x10000. A scale factor of 0 turns off profiling. */ #ifndef _SYS_SYSPROTO_H_ struct profil_args { caddr_t samples; size_t size; size_t offset; u_int scale; }; #endif /* ARGSUSED */ int sys_profil(struct thread *td, struct profil_args *uap) { struct uprof *upp; struct proc *p; if (uap->scale > (1 << 16)) return (EINVAL); p = td->td_proc; if (uap->scale == 0) { PROC_LOCK(p); stopprofclock(p); PROC_UNLOCK(p); return (0); } PROC_LOCK(p); upp = &td->td_proc->p_stats->p_prof; PROC_PROFLOCK(p); upp->pr_off = uap->offset; upp->pr_scale = uap->scale; upp->pr_base = uap->samples; upp->pr_size = uap->size; PROC_PROFUNLOCK(p); startprofclock(p); PROC_UNLOCK(p); return (0); } /* * Scale is a fixed-point number with the binary point 16 bits * into the value, and is <= 1.0. pc is at most 32 bits, so the * intermediate result is at most 48 bits. */ #define PC_TO_INDEX(pc, prof) \ ((int)(((u_quad_t)((pc) - (prof)->pr_off) * \ (u_quad_t)((prof)->pr_scale)) >> 16) & ~1) /* * Collect user-level profiling statistics; called on a profiling tick, * when a process is running in user-mode. This routine may be called * from an interrupt context. We try to update the user profiling buffers * cheaply with fuswintr() and suswintr(). If that fails, we revert to * an AST that will vector us to trap() with a context in which copyin * and copyout will work. Trap will then call addupc_task(). * * Note that we may (rarely) not get around to the AST soon enough, and * lose profile ticks when the next tick overwrites this one, but in this * case the system is overloaded and the profile is probably already * inaccurate. */ void addupc_intr(struct thread *td, uintfptr_t pc, u_int ticks) { struct uprof *prof; caddr_t addr; u_int i; int v; if (ticks == 0) return; prof = &td->td_proc->p_stats->p_prof; PROC_PROFLOCK(td->td_proc); if (pc < prof->pr_off || (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size) { PROC_PROFUNLOCK(td->td_proc); return; /* out of range; ignore */ } addr = prof->pr_base + i; PROC_PROFUNLOCK(td->td_proc); if ((v = fuswintr(addr)) == -1 || suswintr(addr, v + ticks) == -1) { td->td_profil_addr = pc; td->td_profil_ticks = ticks; td->td_pflags |= TDP_OWEUPC; thread_lock(td); td->td_flags |= TDF_ASTPENDING; thread_unlock(td); } } /* * Much like before, but we can afford to take faults here. If the * update fails, we simply turn off profiling. */ void addupc_task(struct thread *td, uintfptr_t pc, u_int ticks) { struct proc *p = td->td_proc; struct uprof *prof; caddr_t addr; u_int i; u_short v; int stop = 0; if (ticks == 0) return; PROC_LOCK(p); if (!(p->p_flag & P_PROFIL)) { PROC_UNLOCK(p); return; } p->p_profthreads++; prof = &p->p_stats->p_prof; PROC_PROFLOCK(p); if (pc < prof->pr_off || (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size) { PROC_PROFUNLOCK(p); goto out; } addr = prof->pr_base + i; PROC_PROFUNLOCK(p); PROC_UNLOCK(p); if (copyin(addr, &v, sizeof(v)) == 0) { v += ticks; if (copyout(&v, addr, sizeof(v)) == 0) { PROC_LOCK(p); goto out; } } stop = 1; PROC_LOCK(p); out: if (--p->p_profthreads == 0) { if (p->p_flag & P_STOPPROF) { wakeup(&p->p_profthreads); p->p_flag &= ~P_STOPPROF; stop = 0; } } if (stop) stopprofclock(p); PROC_UNLOCK(p); } Index: head/sys/kern/subr_rtc.c =================================================================== --- head/sys/kern/subr_rtc.c (revision 305831) +++ head/sys/kern/subr_rtc.c (revision 305832) @@ -1,185 +1,185 @@ /*- * Copyright (c) 1988 University of Utah. * Copyright (c) 1982, 1990, 1993 * The Regents of the University of California. * Copyright (c) 2011 The FreeBSD Foundation * All rights reserved. * * This code is derived from software contributed to Berkeley by * the Systems Programming Group of the University of Utah Computer * Science Department. * * Portions of this software were developed by Julien Ridoux at the University * of Melbourne under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: Utah $Hdr: clock.c 1.18 91/01/21$ * from: @(#)clock.c 8.2 (Berkeley) 1/12/94 * from: NetBSD: clock_subr.c,v 1.6 2001/07/07 17:04:02 thorpej Exp * and * from: src/sys/i386/isa/clock.c,v 1.176 2001/09/04 */ /* * Helpers for time-of-day clocks. This is useful for architectures that need * support multiple models of such clocks, and generally serves to make the * code more machine-independent. * If the clock in question can also be used as a time counter, the driver * needs to initiate this. * This code is not yet used by all architectures. */ #include __FBSDID("$FreeBSD$"); #include "opt_ffclock.h" #include #include #include #include #include #include #include #include #ifdef FFCLOCK #include #endif #include #include "clock_if.h" static device_t clock_dev = NULL; static long clock_res; static struct timespec clock_adj; static struct mtx resettodr_lock; MTX_SYSINIT(resettodr_init, &resettodr_lock, "tod2rl", MTX_DEF); /* XXX: should be kern. now, it's no longer machdep. */ static int disable_rtc_set; SYSCTL_INT(_machdep, OID_AUTO, disable_rtc_set, CTLFLAG_RW, &disable_rtc_set, 0, "Disallow adjusting time-of-day clock"); void clock_register(device_t dev, long res) /* res has units of microseconds */ { if (clock_dev != NULL) { if (clock_res <= res) { if (bootverbose) device_printf(dev, "not installed as " "time-of-day clock: clock %s has higher " "resolution\n", device_get_name(clock_dev)); return; } if (bootverbose) device_printf(clock_dev, "removed as " "time-of-day clock: clock %s has higher " "resolution\n", device_get_name(dev)); } clock_dev = dev; clock_res = res; clock_adj.tv_sec = res / 2 / 1000000; clock_adj.tv_nsec = res / 2 % 1000000 * 1000; if (bootverbose) device_printf(dev, "registered as a time-of-day clock " "(resolution %ldus, adjustment %jd.%09jds)\n", res, (intmax_t)clock_adj.tv_sec, (intmax_t)clock_adj.tv_nsec); } /* * inittodr and settodr derived from the i386 versions written * by Christoph Robitschko , reintroduced and * updated by Chris Stenton 8/10/94 */ /* * Initialize the time of day register, based on the time base which is, e.g. * from a filesystem. */ void inittodr(time_t base) { struct timespec ts; int error; if (clock_dev == NULL) { printf("warning: no time-of-day clock registered, system time " "will not be set accurately\n"); goto wrong_time; } /* XXX: We should poll all registered RTCs in case of failure */ error = CLOCK_GETTIME(clock_dev, &ts); if (error != 0 && error != EINVAL) { printf("warning: clock_gettime failed (%d), the system time " "will not be set accurately\n", error); goto wrong_time; } if (error == EINVAL || ts.tv_sec < 0) { printf("Invalid time in real time clock.\n" "Check and reset the date immediately!\n"); goto wrong_time; } ts.tv_sec += utc_offset(); timespecadd(&ts, &clock_adj); tc_setclock(&ts); #ifdef FFCLOCK ffclock_reset_clock(&ts); #endif return; wrong_time: if (base > 0) { ts.tv_sec = base; ts.tv_nsec = 0; tc_setclock(&ts); } } /* * Write system time back to RTC */ void resettodr(void) { struct timespec ts; int error; if (disable_rtc_set || clock_dev == NULL) return; getnanotime(&ts); timespecadd(&ts, &clock_adj); ts.tv_sec -= utc_offset(); /* XXX: We should really set all registered RTCs */ mtx_lock(&resettodr_lock); error = CLOCK_SETTIME(clock_dev, &ts); mtx_unlock(&resettodr_lock); if (error != 0) printf("warning: clock_settime failed (%d), time-of-day clock " "not adjusted to system time\n", error); } Index: head/sys/kern/subr_scanf.c =================================================================== --- head/sys/kern/subr_scanf.c (revision 305831) +++ head/sys/kern/subr_scanf.c (revision 305832) @@ -1,641 +1,641 @@ /*- * Copyright (c) 1990, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Chris Torek. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * From: Id: vfscanf.c,v 1.13 1998/09/25 12:20:27 obrien Exp * From: static char sccsid[] = "@(#)strtol.c 8.1 (Berkeley) 6/4/93"; * From: static char sccsid[] = "@(#)strtoul.c 8.1 (Berkeley) 6/4/93"; */ #include __FBSDID("$FreeBSD$"); #include #include #include #include /* * Note that stdarg.h and the ANSI style va_start macro is used for both * ANSI and traditional C compilers. */ #include #define BUF 32 /* Maximum length of numeric string. */ /* * Flags used during conversion. */ #define LONG 0x01 /* l: long or double */ #define SHORT 0x04 /* h: short */ #define SUPPRESS 0x08 /* suppress assignment */ #define POINTER 0x10 /* weird %p pointer (`fake hex') */ #define NOSKIP 0x20 /* do not skip blanks */ #define QUAD 0x400 #define SHORTSHORT 0x4000 /** hh: char */ /* * The following are used in numeric conversions only: * SIGNOK, NDIGITS, DPTOK, and EXPOK are for floating point; * SIGNOK, NDIGITS, PFXOK, and NZDIGITS are for integral. */ #define SIGNOK 0x40 /* +/- is (still) legal */ #define NDIGITS 0x80 /* no digits detected */ #define DPTOK 0x100 /* (float) decimal point is still legal */ #define EXPOK 0x200 /* (float) exponent (e+3, etc) still legal */ #define PFXOK 0x100 /* 0x prefix is (still) legal */ #define NZDIGITS 0x200 /* no zero digits detected */ /* * Conversion types. */ #define CT_CHAR 0 /* %c conversion */ #define CT_CCL 1 /* %[...] conversion */ #define CT_STRING 2 /* %s conversion */ #define CT_INT 3 /* integer, i.e., strtoq or strtouq */ typedef u_quad_t (*ccfntype)(const char *, char **, int); static const u_char *__sccl(char *, const u_char *); int sscanf(const char *ibuf, const char *fmt, ...) { va_list ap; int ret; va_start(ap, fmt); ret = vsscanf(ibuf, fmt, ap); va_end(ap); return(ret); } int vsscanf(const char *inp, char const *fmt0, va_list ap) { int inr; const u_char *fmt = (const u_char *)fmt0; int c; /* character from format, or conversion */ size_t width; /* field width, or 0 */ char *p; /* points into all kinds of strings */ int n; /* handy integer */ int flags; /* flags as defined above */ char *p0; /* saves original value of p when necessary */ int nassigned; /* number of fields assigned */ int nconversions; /* number of conversions */ int nread; /* number of characters consumed from fp */ int base; /* base argument to strtoq/strtouq */ ccfntype ccfn; /* conversion function (strtoq/strtouq) */ char ccltab[256]; /* character class table for %[...] */ char buf[BUF]; /* buffer for numeric conversions */ /* `basefix' is used to avoid `if' tests in the integer scanner */ static short basefix[17] = { 10, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16 }; inr = strlen(inp); nassigned = 0; nconversions = 0; nread = 0; base = 0; /* XXX just to keep gcc happy */ ccfn = NULL; /* XXX just to keep gcc happy */ for (;;) { c = *fmt++; if (c == 0) return (nassigned); if (isspace(c)) { while (inr > 0 && isspace(*inp)) nread++, inr--, inp++; continue; } if (c != '%') goto literal; width = 0; flags = 0; /* * switch on the format. continue if done; * break once format type is derived. */ again: c = *fmt++; switch (c) { case '%': literal: if (inr <= 0) goto input_failure; if (*inp != c) goto match_failure; inr--, inp++; nread++; continue; case '*': flags |= SUPPRESS; goto again; case 'l': if (flags & LONG){ flags &= ~LONG; flags |= QUAD; } else { flags |= LONG; } goto again; case 'q': flags |= QUAD; goto again; case 'h': if (flags & SHORT){ flags &= ~SHORT; flags |= SHORTSHORT; } else { flags |= SHORT; } goto again; case '0': case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9': width = width * 10 + c - '0'; goto again; /* * Conversions. * */ case 'd': c = CT_INT; ccfn = (ccfntype)strtoq; base = 10; break; case 'i': c = CT_INT; ccfn = (ccfntype)strtoq; base = 0; break; case 'o': c = CT_INT; ccfn = strtouq; base = 8; break; case 'u': c = CT_INT; ccfn = strtouq; base = 10; break; case 'x': flags |= PFXOK; /* enable 0x prefixing */ c = CT_INT; ccfn = strtouq; base = 16; break; case 's': c = CT_STRING; break; case '[': fmt = __sccl(ccltab, fmt); flags |= NOSKIP; c = CT_CCL; break; case 'c': flags |= NOSKIP; c = CT_CHAR; break; case 'p': /* pointer format is like hex */ flags |= POINTER | PFXOK; c = CT_INT; ccfn = strtouq; base = 16; break; case 'n': nconversions++; if (flags & SUPPRESS) /* ??? */ continue; if (flags & SHORTSHORT) *va_arg(ap, char *) = nread; else if (flags & SHORT) *va_arg(ap, short *) = nread; else if (flags & LONG) *va_arg(ap, long *) = nread; else if (flags & QUAD) *va_arg(ap, quad_t *) = nread; else *va_arg(ap, int *) = nread; continue; } /* * We have a conversion that requires input. */ if (inr <= 0) goto input_failure; /* * Consume leading white space, except for formats * that suppress this. */ if ((flags & NOSKIP) == 0) { while (isspace(*inp)) { nread++; if (--inr > 0) inp++; else goto input_failure; } /* * Note that there is at least one character in * the buffer, so conversions that do not set NOSKIP * can no longer result in an input failure. */ } /* * Do the conversion. */ switch (c) { case CT_CHAR: /* scan arbitrary characters (sets NOSKIP) */ if (width == 0) width = 1; if (flags & SUPPRESS) { size_t sum = 0; for (;;) { if ((n = inr) < width) { sum += n; width -= n; inp += n; if (sum == 0) goto input_failure; break; } else { sum += width; inr -= width; inp += width; break; } } nread += sum; } else { bcopy(inp, va_arg(ap, char *), width); inr -= width; inp += width; nread += width; nassigned++; } nconversions++; break; case CT_CCL: /* scan a (nonempty) character class (sets NOSKIP) */ if (width == 0) width = (size_t)~0; /* `infinity' */ /* take only those things in the class */ if (flags & SUPPRESS) { n = 0; while (ccltab[(unsigned char)*inp]) { n++, inr--, inp++; if (--width == 0) break; if (inr <= 0) { if (n == 0) goto input_failure; break; } } if (n == 0) goto match_failure; } else { p0 = p = va_arg(ap, char *); while (ccltab[(unsigned char)*inp]) { inr--; *p++ = *inp++; if (--width == 0) break; if (inr <= 0) { if (p == p0) goto input_failure; break; } } n = p - p0; if (n == 0) goto match_failure; *p = 0; nassigned++; } nread += n; nconversions++; break; case CT_STRING: /* like CCL, but zero-length string OK, & no NOSKIP */ if (width == 0) width = (size_t)~0; if (flags & SUPPRESS) { n = 0; while (!isspace(*inp)) { n++, inr--, inp++; if (--width == 0) break; if (inr <= 0) break; } nread += n; } else { p0 = p = va_arg(ap, char *); while (!isspace(*inp)) { inr--; *p++ = *inp++; if (--width == 0) break; if (inr <= 0) break; } *p = 0; nread += p - p0; nassigned++; } nconversions++; continue; case CT_INT: /* scan an integer as if by strtoq/strtouq */ #ifdef hardway if (width == 0 || width > sizeof(buf) - 1) width = sizeof(buf) - 1; #else /* size_t is unsigned, hence this optimisation */ if (--width > sizeof(buf) - 2) width = sizeof(buf) - 2; width++; #endif flags |= SIGNOK | NDIGITS | NZDIGITS; for (p = buf; width; width--) { c = *inp; /* * Switch on the character; `goto ok' * if we accept it as a part of number. */ switch (c) { /* * The digit 0 is always legal, but is * special. For %i conversions, if no * digits (zero or nonzero) have been * scanned (only signs), we will have * base==0. In that case, we should set * it to 8 and enable 0x prefixing. * Also, if we have not scanned zero digits * before this, do not turn off prefixing * (someone else will turn it off if we * have scanned any nonzero digits). */ case '0': if (base == 0) { base = 8; flags |= PFXOK; } if (flags & NZDIGITS) flags &= ~(SIGNOK|NZDIGITS|NDIGITS); else flags &= ~(SIGNOK|PFXOK|NDIGITS); goto ok; /* 1 through 7 always legal */ case '1': case '2': case '3': case '4': case '5': case '6': case '7': base = basefix[base]; flags &= ~(SIGNOK | PFXOK | NDIGITS); goto ok; /* digits 8 and 9 ok iff decimal or hex */ case '8': case '9': base = basefix[base]; if (base <= 8) break; /* not legal here */ flags &= ~(SIGNOK | PFXOK | NDIGITS); goto ok; /* letters ok iff hex */ case 'A': case 'B': case 'C': case 'D': case 'E': case 'F': case 'a': case 'b': case 'c': case 'd': case 'e': case 'f': /* no need to fix base here */ if (base <= 10) break; /* not legal here */ flags &= ~(SIGNOK | PFXOK | NDIGITS); goto ok; /* sign ok only as first character */ case '+': case '-': if (flags & SIGNOK) { flags &= ~SIGNOK; goto ok; } break; /* x ok iff flag still set & 2nd char */ case 'x': case 'X': if (flags & PFXOK && p == buf + 1) { base = 16; /* if %i */ flags &= ~PFXOK; goto ok; } break; } /* * If we got here, c is not a legal character * for a number. Stop accumulating digits. */ break; ok: /* * c is legal: store it and look at the next. */ *p++ = c; if (--inr > 0) inp++; else break; /* end of input */ } /* * If we had only a sign, it is no good; push * back the sign. If the number ends in `x', * it was [sign] '0' 'x', so push back the x * and treat it as [sign] '0'. */ if (flags & NDIGITS) { if (p > buf) { inp--; inr++; } goto match_failure; } c = ((u_char *)p)[-1]; if (c == 'x' || c == 'X') { --p; inp--; inr++; } if ((flags & SUPPRESS) == 0) { u_quad_t res; *p = 0; res = (*ccfn)(buf, (char **)NULL, base); if (flags & POINTER) *va_arg(ap, void **) = (void *)(uintptr_t)res; else if (flags & SHORTSHORT) *va_arg(ap, char *) = res; else if (flags & SHORT) *va_arg(ap, short *) = res; else if (flags & LONG) *va_arg(ap, long *) = res; else if (flags & QUAD) *va_arg(ap, quad_t *) = res; else *va_arg(ap, int *) = res; nassigned++; } nread += p - buf; nconversions++; break; } } input_failure: return (nconversions != 0 ? nassigned : -1); match_failure: return (nassigned); } /* * Fill in the given table from the scanset at the given format * (just after `['). Return a pointer to the character past the * closing `]'. The table has a 1 wherever characters should be * considered part of the scanset. */ static const u_char * __sccl(char *tab, const u_char *fmt) { int c, n, v; /* first `clear' the whole table */ c = *fmt++; /* first char hat => negated scanset */ if (c == '^') { v = 1; /* default => accept */ c = *fmt++; /* get new first char */ } else v = 0; /* default => reject */ /* XXX: Will not work if sizeof(tab*) > sizeof(char) */ for (n = 0; n < 256; n++) tab[n] = v; /* memset(tab, v, 256) */ if (c == 0) return (fmt - 1);/* format ended before closing ] */ /* * Now set the entries corresponding to the actual scanset * to the opposite of the above. * * The first character may be ']' (or '-') without being special; * the last character may be '-'. */ v = 1 - v; for (;;) { tab[c] = v; /* take character c */ doswitch: n = *fmt++; /* and examine the next */ switch (n) { case 0: /* format ended too soon */ return (fmt - 1); case '-': /* * A scanset of the form * [01+-] * is defined as `the digit 0, the digit 1, * the character +, the character -', but * the effect of a scanset such as * [a-zA-Z0-9] * is implementation defined. The V7 Unix * scanf treats `a-z' as `the letters a through * z', but treats `a-a' as `the letter a, the * character -, and the letter a'. * * For compatibility, the `-' is not considered * to define a range if the character following * it is either a close bracket (required by ANSI) * or is not numerically greater than the character * we just stored in the table (c). */ n = *fmt; if (n == ']' || n < c) { c = '-'; break; /* resume the for(;;) */ } fmt++; /* fill in the range */ do { tab[++c] = v; } while (c < n); c = n; /* * Alas, the V7 Unix scanf also treats formats * such as [a-c-e] as `the letters a through e'. * This too is permitted by the standard.... */ goto doswitch; break; case ']': /* end of scanset */ return (fmt); default: /* just another character */ c = n; break; } } /* NOTREACHED */ } Index: head/sys/kern/subr_uio.c =================================================================== --- head/sys/kern/subr_uio.c (revision 305831) +++ head/sys/kern/subr_uio.c (revision 305832) @@ -1,622 +1,622 @@ /*- * Copyright (c) 1982, 1986, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Copyright (c) 2014 The FreeBSD Foundation * * Portions of this software were developed by Konstantin Belousov * under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_subr.c 8.3 (Berkeley) 1/21/94 */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include SYSCTL_INT(_kern, KERN_IOV_MAX, iov_max, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, UIO_MAXIOV, "Maximum number of elements in an I/O vector; sysconf(_SC_IOV_MAX)"); static int uiomove_faultflag(void *cp, int n, struct uio *uio, int nofault); int copyin_nofault(const void *udaddr, void *kaddr, size_t len) { int error, save; save = vm_fault_disable_pagefaults(); error = copyin(udaddr, kaddr, len); vm_fault_enable_pagefaults(save); return (error); } int copyout_nofault(const void *kaddr, void *udaddr, size_t len) { int error, save; save = vm_fault_disable_pagefaults(); error = copyout(kaddr, udaddr, len); vm_fault_enable_pagefaults(save); return (error); } #define PHYS_PAGE_COUNT(len) (howmany(len, PAGE_SIZE) + 1) int physcopyin(void *src, vm_paddr_t dst, size_t len) { vm_page_t m[PHYS_PAGE_COUNT(len)]; struct iovec iov[1]; struct uio uio; int i; iov[0].iov_base = src; iov[0].iov_len = len; uio.uio_iov = iov; uio.uio_iovcnt = 1; uio.uio_offset = 0; uio.uio_resid = len; uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_WRITE; for (i = 0; i < PHYS_PAGE_COUNT(len); i++, dst += PAGE_SIZE) m[i] = PHYS_TO_VM_PAGE(dst); return (uiomove_fromphys(m, dst & PAGE_MASK, len, &uio)); } int physcopyout(vm_paddr_t src, void *dst, size_t len) { vm_page_t m[PHYS_PAGE_COUNT(len)]; struct iovec iov[1]; struct uio uio; int i; iov[0].iov_base = dst; iov[0].iov_len = len; uio.uio_iov = iov; uio.uio_iovcnt = 1; uio.uio_offset = 0; uio.uio_resid = len; uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_READ; for (i = 0; i < PHYS_PAGE_COUNT(len); i++, src += PAGE_SIZE) m[i] = PHYS_TO_VM_PAGE(src); return (uiomove_fromphys(m, src & PAGE_MASK, len, &uio)); } #undef PHYS_PAGE_COUNT int physcopyin_vlist(bus_dma_segment_t *src, off_t offset, vm_paddr_t dst, size_t len) { size_t seg_len; int error; error = 0; while (offset >= src->ds_len) { offset -= src->ds_len; src++; } while (len > 0 && error == 0) { seg_len = MIN(src->ds_len - offset, len); error = physcopyin((void *)(uintptr_t)(src->ds_addr + offset), dst, seg_len); offset = 0; src++; len -= seg_len; dst += seg_len; } return (error); } int physcopyout_vlist(vm_paddr_t src, bus_dma_segment_t *dst, off_t offset, size_t len) { size_t seg_len; int error; error = 0; while (offset >= dst->ds_len) { offset -= dst->ds_len; dst++; } while (len > 0 && error == 0) { seg_len = MIN(dst->ds_len - offset, len); error = physcopyout(src, (void *)(uintptr_t)(dst->ds_addr + offset), seg_len); offset = 0; dst++; len -= seg_len; src += seg_len; } return (error); } int uiomove(void *cp, int n, struct uio *uio) { return (uiomove_faultflag(cp, n, uio, 0)); } int uiomove_nofault(void *cp, int n, struct uio *uio) { return (uiomove_faultflag(cp, n, uio, 1)); } static int uiomove_faultflag(void *cp, int n, struct uio *uio, int nofault) { struct thread *td; struct iovec *iov; size_t cnt; int error, newflags, save; td = curthread; error = 0; KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE, ("uiomove: mode")); KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == td, ("uiomove proc")); if (!nofault) WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Calling uiomove()"); /* XXX does it make a sense to set TDP_DEADLKTREAT for UIO_SYSSPACE ? */ newflags = TDP_DEADLKTREAT; if (uio->uio_segflg == UIO_USERSPACE && nofault) { /* * Fail if a non-spurious page fault occurs. */ newflags |= TDP_NOFAULTING | TDP_RESETSPUR; } save = curthread_pflags_set(newflags); while (n > 0 && uio->uio_resid) { iov = uio->uio_iov; cnt = iov->iov_len; if (cnt == 0) { uio->uio_iov++; uio->uio_iovcnt--; continue; } if (cnt > n) cnt = n; switch (uio->uio_segflg) { case UIO_USERSPACE: maybe_yield(); if (uio->uio_rw == UIO_READ) error = copyout(cp, iov->iov_base, cnt); else error = copyin(iov->iov_base, cp, cnt); if (error) goto out; break; case UIO_SYSSPACE: if (uio->uio_rw == UIO_READ) bcopy(cp, iov->iov_base, cnt); else bcopy(iov->iov_base, cp, cnt); break; case UIO_NOCOPY: break; } iov->iov_base = (char *)iov->iov_base + cnt; iov->iov_len -= cnt; uio->uio_resid -= cnt; uio->uio_offset += cnt; cp = (char *)cp + cnt; n -= cnt; } out: curthread_pflags_restore(save); return (error); } /* * Wrapper for uiomove() that validates the arguments against a known-good * kernel buffer. Currently, uiomove accepts a signed (n) argument, which * is almost definitely a bad thing, so we catch that here as well. We * return a runtime failure, but it might be desirable to generate a runtime * assertion failure instead. */ int uiomove_frombuf(void *buf, int buflen, struct uio *uio) { size_t offset, n; if (uio->uio_offset < 0 || uio->uio_resid < 0 || (offset = uio->uio_offset) != uio->uio_offset) return (EINVAL); if (buflen <= 0 || offset >= buflen) return (0); if ((n = buflen - offset) > IOSIZE_MAX) return (EINVAL); return (uiomove((char *)buf + offset, n, uio)); } /* * Give next character to user as result of read. */ int ureadc(int c, struct uio *uio) { struct iovec *iov; char *iov_base; WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Calling ureadc()"); again: if (uio->uio_iovcnt == 0 || uio->uio_resid == 0) panic("ureadc"); iov = uio->uio_iov; if (iov->iov_len == 0) { uio->uio_iovcnt--; uio->uio_iov++; goto again; } switch (uio->uio_segflg) { case UIO_USERSPACE: if (subyte(iov->iov_base, c) < 0) return (EFAULT); break; case UIO_SYSSPACE: iov_base = iov->iov_base; *iov_base = c; break; case UIO_NOCOPY: break; } iov->iov_base = (char *)iov->iov_base + 1; iov->iov_len--; uio->uio_resid--; uio->uio_offset++; return (0); } int copyinfrom(const void * __restrict src, void * __restrict dst, size_t len, int seg) { int error = 0; switch (seg) { case UIO_USERSPACE: error = copyin(src, dst, len); break; case UIO_SYSSPACE: bcopy(src, dst, len); break; default: panic("copyinfrom: bad seg %d\n", seg); } return (error); } int copyinstrfrom(const void * __restrict src, void * __restrict dst, size_t len, size_t * __restrict copied, int seg) { int error = 0; switch (seg) { case UIO_USERSPACE: error = copyinstr(src, dst, len, copied); break; case UIO_SYSSPACE: error = copystr(src, dst, len, copied); break; default: panic("copyinstrfrom: bad seg %d\n", seg); } return (error); } int copyiniov(const struct iovec *iovp, u_int iovcnt, struct iovec **iov, int error) { u_int iovlen; *iov = NULL; if (iovcnt > UIO_MAXIOV) return (error); iovlen = iovcnt * sizeof (struct iovec); *iov = malloc(iovlen, M_IOV, M_WAITOK); error = copyin(iovp, *iov, iovlen); if (error) { free(*iov, M_IOV); *iov = NULL; } return (error); } int copyinuio(const struct iovec *iovp, u_int iovcnt, struct uio **uiop) { struct iovec *iov; struct uio *uio; u_int iovlen; int error, i; *uiop = NULL; if (iovcnt > UIO_MAXIOV) return (EINVAL); iovlen = iovcnt * sizeof (struct iovec); uio = malloc(iovlen + sizeof *uio, M_IOV, M_WAITOK); iov = (struct iovec *)(uio + 1); error = copyin(iovp, iov, iovlen); if (error) { free(uio, M_IOV); return (error); } uio->uio_iov = iov; uio->uio_iovcnt = iovcnt; uio->uio_segflg = UIO_USERSPACE; uio->uio_offset = -1; uio->uio_resid = 0; for (i = 0; i < iovcnt; i++) { if (iov->iov_len > IOSIZE_MAX - uio->uio_resid) { free(uio, M_IOV); return (EINVAL); } uio->uio_resid += iov->iov_len; iov++; } *uiop = uio; return (0); } struct uio * cloneuio(struct uio *uiop) { struct uio *uio; int iovlen; iovlen = uiop->uio_iovcnt * sizeof (struct iovec); uio = malloc(iovlen + sizeof *uio, M_IOV, M_WAITOK); *uio = *uiop; uio->uio_iov = (struct iovec *)(uio + 1); bcopy(uiop->uio_iov, uio->uio_iov, iovlen); return (uio); } /* * Map some anonymous memory in user space of size sz, rounded up to the page * boundary. */ int copyout_map(struct thread *td, vm_offset_t *addr, size_t sz) { struct vmspace *vms; int error; vm_size_t size; vms = td->td_proc->p_vmspace; /* * Map somewhere after heap in process memory. */ *addr = round_page((vm_offset_t)vms->vm_daddr + lim_max(td, RLIMIT_DATA)); /* round size up to page boundary */ size = (vm_size_t)round_page(sz); error = vm_mmap(&vms->vm_map, addr, size, VM_PROT_READ | VM_PROT_WRITE, VM_PROT_ALL, MAP_PRIVATE | MAP_ANON, OBJT_DEFAULT, NULL, 0); return (error); } /* * Unmap memory in user space. */ int copyout_unmap(struct thread *td, vm_offset_t addr, size_t sz) { vm_map_t map; vm_size_t size; if (sz == 0) return (0); map = &td->td_proc->p_vmspace->vm_map; size = (vm_size_t)round_page(sz); if (vm_map_remove(map, addr, addr + size) != KERN_SUCCESS) return (EINVAL); return (0); } #ifdef NO_FUEWORD /* * XXXKIB The temporal implementation of fue*() functions which do not * handle usermode -1 properly, mixing it with the fault code. Keep * this until MD code is written. Currently sparc64 and mips do not * have proper implementation. */ int fueword(volatile const void *base, long *val) { long res; res = fuword(base); if (res == -1) return (-1); *val = res; return (0); } int fueword32(volatile const void *base, int32_t *val) { int32_t res; res = fuword32(base); if (res == -1) return (-1); *val = res; return (0); } #ifdef _LP64 int fueword64(volatile const void *base, int64_t *val) { int32_t res; res = fuword64(base); if (res == -1) return (-1); *val = res; return (0); } #endif int casueword32(volatile uint32_t *base, uint32_t oldval, uint32_t *oldvalp, uint32_t newval) { int32_t ov; ov = casuword32(base, oldval, newval); if (ov == -1) return (-1); *oldvalp = ov; return (0); } int casueword(volatile u_long *p, u_long oldval, u_long *oldvalp, u_long newval) { u_long ov; ov = casuword(p, oldval, newval); if (ov == -1) return (-1); *oldvalp = ov; return (0); } #else /* NO_FUEWORD */ int32_t fuword32(volatile const void *addr) { int rv; int32_t val; rv = fueword32(addr, &val); return (rv == -1 ? -1 : val); } #ifdef _LP64 int64_t fuword64(volatile const void *addr) { int rv; int64_t val; rv = fueword64(addr, &val); return (rv == -1 ? -1 : val); } #endif /* _LP64 */ long fuword(volatile const void *addr) { long val; int rv; rv = fueword(addr, &val); return (rv == -1 ? -1 : val); } uint32_t casuword32(volatile uint32_t *addr, uint32_t old, uint32_t new) { int rv; uint32_t val; rv = casueword32(addr, old, &val, new); return (rv == -1 ? -1 : val); } u_long casuword(volatile u_long *addr, u_long old, u_long new) { int rv; u_long val; rv = casueword(addr, old, &val, new); return (rv == -1 ? -1 : val); } #endif /* NO_FUEWORD */ Index: head/sys/kern/sys_generic.c =================================================================== --- head/sys/kern/sys_generic.c (revision 305831) +++ head/sys/kern/sys_generic.c (revision 305832) @@ -1,1931 +1,1931 @@ /*- * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)sys_generic.c 8.5 (Berkeley) 1/21/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_capsicum.h" #include "opt_compat.h" #include "opt_ktrace.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KTRACE #include #endif #include /* * The following macro defines how many bytes will be allocated from * the stack instead of memory allocated when passing the IOCTL data * structures from userspace and to the kernel. Some IOCTLs having * small data structures are used very frequently and this small * buffer on the stack gives a significant speedup improvement for * those requests. The value of this define should be greater or equal * to 64 bytes and should also be power of two. The data structure is * currently hard-aligned to a 8-byte boundary on the stack. This * should currently be sufficient for all supported platforms. */ #define SYS_IOCTL_SMALL_SIZE 128 /* bytes */ #define SYS_IOCTL_SMALL_ALIGN 8 /* bytes */ #ifdef __LP64__ static int iosize_max_clamp = 0; SYSCTL_INT(_debug, OID_AUTO, iosize_max_clamp, CTLFLAG_RW, &iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX"); static int devfs_iosize_max_clamp = 1; SYSCTL_INT(_debug, OID_AUTO, devfs_iosize_max_clamp, CTLFLAG_RW, &devfs_iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX for devices"); #endif /* * Assert that the return value of read(2) and write(2) syscalls fits * into a register. If not, an architecture will need to provide the * usermode wrappers to reconstruct the result. */ CTASSERT(sizeof(register_t) >= sizeof(size_t)); static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer"); static MALLOC_DEFINE(M_SELECT, "select", "select() buffer"); MALLOC_DEFINE(M_IOV, "iov", "large iov's"); static int pollout(struct thread *, struct pollfd *, struct pollfd *, u_int); static int pollscan(struct thread *, struct pollfd *, u_int); static int pollrescan(struct thread *); static int selscan(struct thread *, fd_mask **, fd_mask **, int); static int selrescan(struct thread *, fd_mask **, fd_mask **); static void selfdalloc(struct thread *, void *); static void selfdfree(struct seltd *, struct selfd *); static int dofileread(struct thread *, int, struct file *, struct uio *, off_t, int); static int dofilewrite(struct thread *, int, struct file *, struct uio *, off_t, int); static void doselwakeup(struct selinfo *, int); static void seltdinit(struct thread *); static int seltdwait(struct thread *, sbintime_t, sbintime_t); static void seltdclear(struct thread *); /* * One seltd per-thread allocated on demand as needed. * * t - protected by st_mtx * k - Only accessed by curthread or read-only */ struct seltd { STAILQ_HEAD(, selfd) st_selq; /* (k) List of selfds. */ struct selfd *st_free1; /* (k) free fd for read set. */ struct selfd *st_free2; /* (k) free fd for write set. */ struct mtx st_mtx; /* Protects struct seltd */ struct cv st_wait; /* (t) Wait channel. */ int st_flags; /* (t) SELTD_ flags. */ }; #define SELTD_PENDING 0x0001 /* We have pending events. */ #define SELTD_RESCAN 0x0002 /* Doing a rescan. */ /* * One selfd allocated per-thread per-file-descriptor. * f - protected by sf_mtx */ struct selfd { STAILQ_ENTRY(selfd) sf_link; /* (k) fds owned by this td. */ TAILQ_ENTRY(selfd) sf_threads; /* (f) fds on this selinfo. */ struct selinfo *sf_si; /* (f) selinfo when linked. */ struct mtx *sf_mtx; /* Pointer to selinfo mtx. */ struct seltd *sf_td; /* (k) owning seltd. */ void *sf_cookie; /* (k) fd or pollfd. */ u_int sf_refs; }; static uma_zone_t selfd_zone; static struct mtx_pool *mtxpool_select; #ifdef __LP64__ size_t devfs_iosize_max(void) { return (devfs_iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ? INT_MAX : SSIZE_MAX); } size_t iosize_max(void) { return (iosize_max_clamp || SV_CURPROC_FLAG(SV_ILP32) ? INT_MAX : SSIZE_MAX); } #endif #ifndef _SYS_SYSPROTO_H_ struct read_args { int fd; void *buf; size_t nbyte; }; #endif int sys_read(td, uap) struct thread *td; struct read_args *uap; { struct uio auio; struct iovec aiov; int error; if (uap->nbyte > IOSIZE_MAX) return (EINVAL); aiov.iov_base = uap->buf; aiov.iov_len = uap->nbyte; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_resid = uap->nbyte; auio.uio_segflg = UIO_USERSPACE; error = kern_readv(td, uap->fd, &auio); return(error); } /* * Positioned read system call */ #ifndef _SYS_SYSPROTO_H_ struct pread_args { int fd; void *buf; size_t nbyte; int pad; off_t offset; }; #endif int sys_pread(td, uap) struct thread *td; struct pread_args *uap; { struct uio auio; struct iovec aiov; int error; if (uap->nbyte > IOSIZE_MAX) return (EINVAL); aiov.iov_base = uap->buf; aiov.iov_len = uap->nbyte; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_resid = uap->nbyte; auio.uio_segflg = UIO_USERSPACE; error = kern_preadv(td, uap->fd, &auio, uap->offset); return(error); } #if defined(COMPAT_FREEBSD6) int freebsd6_pread(td, uap) struct thread *td; struct freebsd6_pread_args *uap; { struct pread_args oargs; oargs.fd = uap->fd; oargs.buf = uap->buf; oargs.nbyte = uap->nbyte; oargs.offset = uap->offset; return (sys_pread(td, &oargs)); } #endif /* * Scatter read system call. */ #ifndef _SYS_SYSPROTO_H_ struct readv_args { int fd; struct iovec *iovp; u_int iovcnt; }; #endif int sys_readv(struct thread *td, struct readv_args *uap) { struct uio *auio; int error; error = copyinuio(uap->iovp, uap->iovcnt, &auio); if (error) return (error); error = kern_readv(td, uap->fd, auio); free(auio, M_IOV); return (error); } int kern_readv(struct thread *td, int fd, struct uio *auio) { struct file *fp; cap_rights_t rights; int error; error = fget_read(td, fd, cap_rights_init(&rights, CAP_READ), &fp); if (error) return (error); error = dofileread(td, fd, fp, auio, (off_t)-1, 0); fdrop(fp, td); return (error); } /* * Scatter positioned read system call. */ #ifndef _SYS_SYSPROTO_H_ struct preadv_args { int fd; struct iovec *iovp; u_int iovcnt; off_t offset; }; #endif int sys_preadv(struct thread *td, struct preadv_args *uap) { struct uio *auio; int error; error = copyinuio(uap->iovp, uap->iovcnt, &auio); if (error) return (error); error = kern_preadv(td, uap->fd, auio, uap->offset); free(auio, M_IOV); return (error); } int kern_preadv(td, fd, auio, offset) struct thread *td; int fd; struct uio *auio; off_t offset; { struct file *fp; cap_rights_t rights; int error; error = fget_read(td, fd, cap_rights_init(&rights, CAP_PREAD), &fp); if (error) return (error); if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE)) error = ESPIPE; else if (offset < 0 && fp->f_vnode->v_type != VCHR) error = EINVAL; else error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET); fdrop(fp, td); return (error); } /* * Common code for readv and preadv that reads data in * from a file using the passed in uio, offset, and flags. */ static int dofileread(td, fd, fp, auio, offset, flags) struct thread *td; int fd; struct file *fp; struct uio *auio; off_t offset; int flags; { ssize_t cnt; int error; #ifdef KTRACE struct uio *ktruio = NULL; #endif AUDIT_ARG_FD(fd); /* Finish zero length reads right here */ if (auio->uio_resid == 0) { td->td_retval[0] = 0; return(0); } auio->uio_rw = UIO_READ; auio->uio_offset = offset; auio->uio_td = td; #ifdef KTRACE if (KTRPOINT(td, KTR_GENIO)) ktruio = cloneuio(auio); #endif cnt = auio->uio_resid; if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) { if (auio->uio_resid != cnt && (error == ERESTART || error == EINTR || error == EWOULDBLOCK)) error = 0; } cnt -= auio->uio_resid; #ifdef KTRACE if (ktruio != NULL) { ktruio->uio_resid = cnt; ktrgenio(fd, UIO_READ, ktruio, error); } #endif td->td_retval[0] = cnt; return (error); } #ifndef _SYS_SYSPROTO_H_ struct write_args { int fd; const void *buf; size_t nbyte; }; #endif int sys_write(td, uap) struct thread *td; struct write_args *uap; { struct uio auio; struct iovec aiov; int error; if (uap->nbyte > IOSIZE_MAX) return (EINVAL); aiov.iov_base = (void *)(uintptr_t)uap->buf; aiov.iov_len = uap->nbyte; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_resid = uap->nbyte; auio.uio_segflg = UIO_USERSPACE; error = kern_writev(td, uap->fd, &auio); return(error); } /* * Positioned write system call. */ #ifndef _SYS_SYSPROTO_H_ struct pwrite_args { int fd; const void *buf; size_t nbyte; int pad; off_t offset; }; #endif int sys_pwrite(td, uap) struct thread *td; struct pwrite_args *uap; { struct uio auio; struct iovec aiov; int error; if (uap->nbyte > IOSIZE_MAX) return (EINVAL); aiov.iov_base = (void *)(uintptr_t)uap->buf; aiov.iov_len = uap->nbyte; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_resid = uap->nbyte; auio.uio_segflg = UIO_USERSPACE; error = kern_pwritev(td, uap->fd, &auio, uap->offset); return(error); } #if defined(COMPAT_FREEBSD6) int freebsd6_pwrite(td, uap) struct thread *td; struct freebsd6_pwrite_args *uap; { struct pwrite_args oargs; oargs.fd = uap->fd; oargs.buf = uap->buf; oargs.nbyte = uap->nbyte; oargs.offset = uap->offset; return (sys_pwrite(td, &oargs)); } #endif /* * Gather write system call. */ #ifndef _SYS_SYSPROTO_H_ struct writev_args { int fd; struct iovec *iovp; u_int iovcnt; }; #endif int sys_writev(struct thread *td, struct writev_args *uap) { struct uio *auio; int error; error = copyinuio(uap->iovp, uap->iovcnt, &auio); if (error) return (error); error = kern_writev(td, uap->fd, auio); free(auio, M_IOV); return (error); } int kern_writev(struct thread *td, int fd, struct uio *auio) { struct file *fp; cap_rights_t rights; int error; error = fget_write(td, fd, cap_rights_init(&rights, CAP_WRITE), &fp); if (error) return (error); error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0); fdrop(fp, td); return (error); } /* * Gather positioned write system call. */ #ifndef _SYS_SYSPROTO_H_ struct pwritev_args { int fd; struct iovec *iovp; u_int iovcnt; off_t offset; }; #endif int sys_pwritev(struct thread *td, struct pwritev_args *uap) { struct uio *auio; int error; error = copyinuio(uap->iovp, uap->iovcnt, &auio); if (error) return (error); error = kern_pwritev(td, uap->fd, auio, uap->offset); free(auio, M_IOV); return (error); } int kern_pwritev(td, fd, auio, offset) struct thread *td; struct uio *auio; int fd; off_t offset; { struct file *fp; cap_rights_t rights; int error; error = fget_write(td, fd, cap_rights_init(&rights, CAP_PWRITE), &fp); if (error) return (error); if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE)) error = ESPIPE; else if (offset < 0 && fp->f_vnode->v_type != VCHR) error = EINVAL; else error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET); fdrop(fp, td); return (error); } /* * Common code for writev and pwritev that writes data to * a file using the passed in uio, offset, and flags. */ static int dofilewrite(td, fd, fp, auio, offset, flags) struct thread *td; int fd; struct file *fp; struct uio *auio; off_t offset; int flags; { ssize_t cnt; int error; #ifdef KTRACE struct uio *ktruio = NULL; #endif AUDIT_ARG_FD(fd); auio->uio_rw = UIO_WRITE; auio->uio_td = td; auio->uio_offset = offset; #ifdef KTRACE if (KTRPOINT(td, KTR_GENIO)) ktruio = cloneuio(auio); #endif cnt = auio->uio_resid; if (fp->f_type == DTYPE_VNODE && (fp->f_vnread_flags & FDEVFS_VNODE) == 0) bwillwrite(); if ((error = fo_write(fp, auio, td->td_ucred, flags, td))) { if (auio->uio_resid != cnt && (error == ERESTART || error == EINTR || error == EWOULDBLOCK)) error = 0; /* Socket layer is responsible for issuing SIGPIPE. */ if (fp->f_type != DTYPE_SOCKET && error == EPIPE) { PROC_LOCK(td->td_proc); tdsignal(td, SIGPIPE); PROC_UNLOCK(td->td_proc); } } cnt -= auio->uio_resid; #ifdef KTRACE if (ktruio != NULL) { ktruio->uio_resid = cnt; ktrgenio(fd, UIO_WRITE, ktruio, error); } #endif td->td_retval[0] = cnt; return (error); } /* * Truncate a file given a file descriptor. * * Can't use fget_write() here, since must return EINVAL and not EBADF if the * descriptor isn't writable. */ int kern_ftruncate(td, fd, length) struct thread *td; int fd; off_t length; { struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(fd); if (length < 0) return (EINVAL); error = fget(td, fd, cap_rights_init(&rights, CAP_FTRUNCATE), &fp); if (error) return (error); AUDIT_ARG_FILE(td->td_proc, fp); if (!(fp->f_flag & FWRITE)) { fdrop(fp, td); return (EINVAL); } error = fo_truncate(fp, length, td->td_ucred, td); fdrop(fp, td); return (error); } #ifndef _SYS_SYSPROTO_H_ struct ftruncate_args { int fd; int pad; off_t length; }; #endif int sys_ftruncate(td, uap) struct thread *td; struct ftruncate_args *uap; { return (kern_ftruncate(td, uap->fd, uap->length)); } #if defined(COMPAT_43) #ifndef _SYS_SYSPROTO_H_ struct oftruncate_args { int fd; long length; }; #endif int oftruncate(td, uap) struct thread *td; struct oftruncate_args *uap; { return (kern_ftruncate(td, uap->fd, uap->length)); } #endif /* COMPAT_43 */ #ifndef _SYS_SYSPROTO_H_ struct ioctl_args { int fd; u_long com; caddr_t data; }; #endif /* ARGSUSED */ int sys_ioctl(struct thread *td, struct ioctl_args *uap) { u_char smalldata[SYS_IOCTL_SMALL_SIZE] __aligned(SYS_IOCTL_SMALL_ALIGN); u_long com; int arg, error; u_int size; caddr_t data; if (uap->com > 0xffffffff) { printf( "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n", td->td_proc->p_pid, td->td_name, uap->com); uap->com &= 0xffffffff; } com = uap->com; /* * Interpret high order word to find amount of data to be * copied to/from the user's address space. */ size = IOCPARM_LEN(com); if ((size > IOCPARM_MAX) || ((com & (IOC_VOID | IOC_IN | IOC_OUT)) == 0) || #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43) ((com & IOC_OUT) && size == 0) || #else ((com & (IOC_IN | IOC_OUT)) && size == 0) || #endif ((com & IOC_VOID) && size > 0 && size != sizeof(int))) return (ENOTTY); if (size > 0) { if (com & IOC_VOID) { /* Integer argument. */ arg = (intptr_t)uap->data; data = (void *)&arg; size = 0; } else { if (size > SYS_IOCTL_SMALL_SIZE) data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK); else data = smalldata; } } else data = (void *)&uap->data; if (com & IOC_IN) { error = copyin(uap->data, data, (u_int)size); if (error != 0) goto out; } else if (com & IOC_OUT) { /* * Zero the buffer so the user always * gets back something deterministic. */ bzero(data, size); } error = kern_ioctl(td, uap->fd, com, data); if (error == 0 && (com & IOC_OUT)) error = copyout(data, uap->data, (u_int)size); out: if (size > SYS_IOCTL_SMALL_SIZE) free(data, M_IOCTLOPS); return (error); } int kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data) { struct file *fp; struct filedesc *fdp; #ifndef CAPABILITIES cap_rights_t rights; #endif int error, tmp, locked; AUDIT_ARG_FD(fd); AUDIT_ARG_CMD(com); fdp = td->td_proc->p_fd; switch (com) { case FIONCLEX: case FIOCLEX: FILEDESC_XLOCK(fdp); locked = LA_XLOCKED; break; default: #ifdef CAPABILITIES FILEDESC_SLOCK(fdp); locked = LA_SLOCKED; #else locked = LA_UNLOCKED; #endif break; } #ifdef CAPABILITIES if ((fp = fget_locked(fdp, fd)) == NULL) { error = EBADF; goto out; } if ((error = cap_ioctl_check(fdp, fd, com)) != 0) { fp = NULL; /* fhold() was not called yet */ goto out; } fhold(fp); if (locked == LA_SLOCKED) { FILEDESC_SUNLOCK(fdp); locked = LA_UNLOCKED; } #else error = fget(td, fd, cap_rights_init(&rights, CAP_IOCTL), &fp); if (error != 0) { fp = NULL; goto out; } #endif if ((fp->f_flag & (FREAD | FWRITE)) == 0) { error = EBADF; goto out; } switch (com) { case FIONCLEX: fdp->fd_ofiles[fd].fde_flags &= ~UF_EXCLOSE; goto out; case FIOCLEX: fdp->fd_ofiles[fd].fde_flags |= UF_EXCLOSE; goto out; case FIONBIO: if ((tmp = *(int *)data)) atomic_set_int(&fp->f_flag, FNONBLOCK); else atomic_clear_int(&fp->f_flag, FNONBLOCK); data = (void *)&tmp; break; case FIOASYNC: if ((tmp = *(int *)data)) atomic_set_int(&fp->f_flag, FASYNC); else atomic_clear_int(&fp->f_flag, FASYNC); data = (void *)&tmp; break; } error = fo_ioctl(fp, com, data, td->td_ucred, td); out: switch (locked) { case LA_XLOCKED: FILEDESC_XUNLOCK(fdp); break; #ifdef CAPABILITIES case LA_SLOCKED: FILEDESC_SUNLOCK(fdp); break; #endif default: FILEDESC_UNLOCK_ASSERT(fdp); break; } if (fp != NULL) fdrop(fp, td); return (error); } int poll_no_poll(int events) { /* * Return true for read/write. If the user asked for something * special, return POLLNVAL, so that clients have a way of * determining reliably whether or not the extended * functionality is present without hard-coding knowledge * of specific filesystem implementations. */ if (events & ~POLLSTANDARD) return (POLLNVAL); return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM)); } int sys_pselect(struct thread *td, struct pselect_args *uap) { struct timespec ts; struct timeval tv, *tvp; sigset_t set, *uset; int error; if (uap->ts != NULL) { error = copyin(uap->ts, &ts, sizeof(ts)); if (error != 0) return (error); TIMESPEC_TO_TIMEVAL(&tv, &ts); tvp = &tv; } else tvp = NULL; if (uap->sm != NULL) { error = copyin(uap->sm, &set, sizeof(set)); if (error != 0) return (error); uset = &set; } else uset = NULL; return (kern_pselect(td, uap->nd, uap->in, uap->ou, uap->ex, tvp, uset, NFDBITS)); } int kern_pselect(struct thread *td, int nd, fd_set *in, fd_set *ou, fd_set *ex, struct timeval *tvp, sigset_t *uset, int abi_nfdbits) { int error; if (uset != NULL) { error = kern_sigprocmask(td, SIG_SETMASK, uset, &td->td_oldsigmask, 0); if (error != 0) return (error); td->td_pflags |= TDP_OLDMASK; /* * Make sure that ast() is called on return to * usermode and TDP_OLDMASK is cleared, restoring old * sigmask. */ thread_lock(td); td->td_flags |= TDF_ASTPENDING; thread_unlock(td); } error = kern_select(td, nd, in, ou, ex, tvp, abi_nfdbits); return (error); } #ifndef _SYS_SYSPROTO_H_ struct select_args { int nd; fd_set *in, *ou, *ex; struct timeval *tv; }; #endif int sys_select(struct thread *td, struct select_args *uap) { struct timeval tv, *tvp; int error; if (uap->tv != NULL) { error = copyin(uap->tv, &tv, sizeof(tv)); if (error) return (error); tvp = &tv; } else tvp = NULL; return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp, NFDBITS)); } /* * In the unlikely case when user specified n greater then the last * open file descriptor, check that no bits are set after the last * valid fd. We must return EBADF if any is set. * * There are applications that rely on the behaviour. * * nd is fd_lastfile + 1. */ static int select_check_badfd(fd_set *fd_in, int nd, int ndu, int abi_nfdbits) { char *addr, *oaddr; int b, i, res; uint8_t bits; if (nd >= ndu || fd_in == NULL) return (0); oaddr = NULL; bits = 0; /* silence gcc */ for (i = nd; i < ndu; i++) { b = i / NBBY; #if BYTE_ORDER == LITTLE_ENDIAN addr = (char *)fd_in + b; #else addr = (char *)fd_in; if (abi_nfdbits == NFDBITS) { addr += rounddown(b, sizeof(fd_mask)) + sizeof(fd_mask) - 1 - b % sizeof(fd_mask); } else { addr += rounddown(b, sizeof(uint32_t)) + sizeof(uint32_t) - 1 - b % sizeof(uint32_t); } #endif if (addr != oaddr) { res = fubyte(addr); if (res == -1) return (EFAULT); oaddr = addr; bits = res; } if ((bits & (1 << (i % NBBY))) != 0) return (EBADF); } return (0); } int kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou, fd_set *fd_ex, struct timeval *tvp, int abi_nfdbits) { struct filedesc *fdp; /* * The magic 2048 here is chosen to be just enough for FD_SETSIZE * infds with the new FD_SETSIZE of 1024, and more than enough for * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE * of 256. */ fd_mask s_selbits[howmany(2048, NFDBITS)]; fd_mask *ibits[3], *obits[3], *selbits, *sbp; struct timeval rtv; sbintime_t asbt, precision, rsbt; u_int nbufbytes, ncpbytes, ncpubytes, nfdbits; int error, lf, ndu; if (nd < 0) return (EINVAL); fdp = td->td_proc->p_fd; ndu = nd; lf = fdp->fd_lastfile; if (nd > lf + 1) nd = lf + 1; error = select_check_badfd(fd_in, nd, ndu, abi_nfdbits); if (error != 0) return (error); error = select_check_badfd(fd_ou, nd, ndu, abi_nfdbits); if (error != 0) return (error); error = select_check_badfd(fd_ex, nd, ndu, abi_nfdbits); if (error != 0) return (error); /* * Allocate just enough bits for the non-null fd_sets. Use the * preallocated auto buffer if possible. */ nfdbits = roundup(nd, NFDBITS); ncpbytes = nfdbits / NBBY; ncpubytes = roundup(nd, abi_nfdbits) / NBBY; nbufbytes = 0; if (fd_in != NULL) nbufbytes += 2 * ncpbytes; if (fd_ou != NULL) nbufbytes += 2 * ncpbytes; if (fd_ex != NULL) nbufbytes += 2 * ncpbytes; if (nbufbytes <= sizeof s_selbits) selbits = &s_selbits[0]; else selbits = malloc(nbufbytes, M_SELECT, M_WAITOK); /* * Assign pointers into the bit buffers and fetch the input bits. * Put the output buffers together so that they can be bzeroed * together. */ sbp = selbits; #define getbits(name, x) \ do { \ if (name == NULL) { \ ibits[x] = NULL; \ obits[x] = NULL; \ } else { \ ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp; \ obits[x] = sbp; \ sbp += ncpbytes / sizeof *sbp; \ error = copyin(name, ibits[x], ncpubytes); \ if (error != 0) \ goto done; \ bzero((char *)ibits[x] + ncpubytes, \ ncpbytes - ncpubytes); \ } \ } while (0) getbits(fd_in, 0); getbits(fd_ou, 1); getbits(fd_ex, 2); #undef getbits #if BYTE_ORDER == BIG_ENDIAN && defined(__LP64__) /* * XXX: swizzle_fdset assumes that if abi_nfdbits != NFDBITS, * we are running under 32-bit emulation. This should be more * generic. */ #define swizzle_fdset(bits) \ if (abi_nfdbits != NFDBITS && bits != NULL) { \ int i; \ for (i = 0; i < ncpbytes / sizeof *sbp; i++) \ bits[i] = (bits[i] >> 32) | (bits[i] << 32); \ } #else #define swizzle_fdset(bits) #endif /* Make sure the bit order makes it through an ABI transition */ swizzle_fdset(ibits[0]); swizzle_fdset(ibits[1]); swizzle_fdset(ibits[2]); if (nbufbytes != 0) bzero(selbits, nbufbytes / 2); precision = 0; if (tvp != NULL) { rtv = *tvp; if (rtv.tv_sec < 0 || rtv.tv_usec < 0 || rtv.tv_usec >= 1000000) { error = EINVAL; goto done; } if (!timevalisset(&rtv)) asbt = 0; else if (rtv.tv_sec <= INT32_MAX) { rsbt = tvtosbt(rtv); precision = rsbt; precision >>= tc_precexp; if (TIMESEL(&asbt, rsbt)) asbt += tc_tick_sbt; if (asbt <= SBT_MAX - rsbt) asbt += rsbt; else asbt = -1; } else asbt = -1; } else asbt = -1; seltdinit(td); /* Iterate until the timeout expires or descriptors become ready. */ for (;;) { error = selscan(td, ibits, obits, nd); if (error || td->td_retval[0] != 0) break; error = seltdwait(td, asbt, precision); if (error) break; error = selrescan(td, ibits, obits); if (error || td->td_retval[0] != 0) break; } seltdclear(td); done: /* select is not restarted after signals... */ if (error == ERESTART) error = EINTR; if (error == EWOULDBLOCK) error = 0; /* swizzle bit order back, if necessary */ swizzle_fdset(obits[0]); swizzle_fdset(obits[1]); swizzle_fdset(obits[2]); #undef swizzle_fdset #define putbits(name, x) \ if (name && (error2 = copyout(obits[x], name, ncpubytes))) \ error = error2; if (error == 0) { int error2; putbits(fd_in, 0); putbits(fd_ou, 1); putbits(fd_ex, 2); #undef putbits } if (selbits != &s_selbits[0]) free(selbits, M_SELECT); return (error); } /* * Convert a select bit set to poll flags. * * The backend always returns POLLHUP/POLLERR if appropriate and we * return this as a set bit in any set. */ static int select_flags[3] = { POLLRDNORM | POLLHUP | POLLERR, POLLWRNORM | POLLHUP | POLLERR, POLLRDBAND | POLLERR }; /* * Compute the fo_poll flags required for a fd given by the index and * bit position in the fd_mask array. */ static __inline int selflags(fd_mask **ibits, int idx, fd_mask bit) { int flags; int msk; flags = 0; for (msk = 0; msk < 3; msk++) { if (ibits[msk] == NULL) continue; if ((ibits[msk][idx] & bit) == 0) continue; flags |= select_flags[msk]; } return (flags); } /* * Set the appropriate output bits given a mask of fired events and the * input bits originally requested. */ static __inline int selsetbits(fd_mask **ibits, fd_mask **obits, int idx, fd_mask bit, int events) { int msk; int n; n = 0; for (msk = 0; msk < 3; msk++) { if ((events & select_flags[msk]) == 0) continue; if (ibits[msk] == NULL) continue; if ((ibits[msk][idx] & bit) == 0) continue; /* * XXX Check for a duplicate set. This can occur because a * socket calls selrecord() twice for each poll() call * resulting in two selfds per real fd. selrescan() will * call selsetbits twice as a result. */ if ((obits[msk][idx] & bit) != 0) continue; obits[msk][idx] |= bit; n++; } return (n); } static __inline int getselfd_cap(struct filedesc *fdp, int fd, struct file **fpp) { cap_rights_t rights; cap_rights_init(&rights, CAP_EVENT); return (fget_unlocked(fdp, fd, &rights, fpp, NULL)); } /* * Traverse the list of fds attached to this thread's seltd and check for * completion. */ static int selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits) { struct filedesc *fdp; struct selinfo *si; struct seltd *stp; struct selfd *sfp; struct selfd *sfn; struct file *fp; fd_mask bit; int fd, ev, n, idx; int error; fdp = td->td_proc->p_fd; stp = td->td_sel; n = 0; STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) { fd = (int)(uintptr_t)sfp->sf_cookie; si = sfp->sf_si; selfdfree(stp, sfp); /* If the selinfo wasn't cleared the event didn't fire. */ if (si != NULL) continue; error = getselfd_cap(fdp, fd, &fp); if (error) return (error); idx = fd / NFDBITS; bit = (fd_mask)1 << (fd % NFDBITS); ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td); fdrop(fp, td); if (ev != 0) n += selsetbits(ibits, obits, idx, bit, ev); } stp->st_flags = 0; td->td_retval[0] = n; return (0); } /* * Perform the initial filedescriptor scan and register ourselves with * each selinfo. */ static int selscan(td, ibits, obits, nfd) struct thread *td; fd_mask **ibits, **obits; int nfd; { struct filedesc *fdp; struct file *fp; fd_mask bit; int ev, flags, end, fd; int n, idx; int error; fdp = td->td_proc->p_fd; n = 0; for (idx = 0, fd = 0; fd < nfd; idx++) { end = imin(fd + NFDBITS, nfd); for (bit = 1; fd < end; bit <<= 1, fd++) { /* Compute the list of events we're interested in. */ flags = selflags(ibits, idx, bit); if (flags == 0) continue; error = getselfd_cap(fdp, fd, &fp); if (error) return (error); selfdalloc(td, (void *)(uintptr_t)fd); ev = fo_poll(fp, flags, td->td_ucred, td); fdrop(fp, td); if (ev != 0) n += selsetbits(ibits, obits, idx, bit, ev); } } td->td_retval[0] = n; return (0); } int sys_poll(struct thread *td, struct poll_args *uap) { struct timespec ts, *tsp; if (uap->timeout != INFTIM) { if (uap->timeout < 0) return (EINVAL); ts.tv_sec = uap->timeout / 1000; ts.tv_nsec = (uap->timeout % 1000) * 1000000; tsp = &ts; } else tsp = NULL; return (kern_poll(td, uap->fds, uap->nfds, tsp, NULL)); } int kern_poll(struct thread *td, struct pollfd *fds, u_int nfds, struct timespec *tsp, sigset_t *uset) { struct pollfd *bits; struct pollfd smallbits[32]; sbintime_t sbt, precision, tmp; time_t over; struct timespec ts; int error; size_t ni; precision = 0; if (tsp != NULL) { if (tsp->tv_sec < 0) return (EINVAL); if (tsp->tv_nsec < 0 || tsp->tv_nsec >= 1000000000) return (EINVAL); if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) sbt = 0; else { ts = *tsp; if (ts.tv_sec > INT32_MAX / 2) { over = ts.tv_sec - INT32_MAX / 2; ts.tv_sec -= over; } else over = 0; tmp = tstosbt(ts); precision = tmp; precision >>= tc_precexp; if (TIMESEL(&sbt, tmp)) sbt += tc_tick_sbt; sbt += tmp; } } else sbt = -1; if (nfds > maxfilesperproc && nfds > FD_SETSIZE) return (EINVAL); ni = nfds * sizeof(struct pollfd); if (ni > sizeof(smallbits)) bits = malloc(ni, M_TEMP, M_WAITOK); else bits = smallbits; error = copyin(fds, bits, ni); if (error) goto done; if (uset != NULL) { error = kern_sigprocmask(td, SIG_SETMASK, uset, &td->td_oldsigmask, 0); if (error) goto done; td->td_pflags |= TDP_OLDMASK; /* * Make sure that ast() is called on return to * usermode and TDP_OLDMASK is cleared, restoring old * sigmask. */ thread_lock(td); td->td_flags |= TDF_ASTPENDING; thread_unlock(td); } seltdinit(td); /* Iterate until the timeout expires or descriptors become ready. */ for (;;) { error = pollscan(td, bits, nfds); if (error || td->td_retval[0] != 0) break; error = seltdwait(td, sbt, precision); if (error) break; error = pollrescan(td); if (error || td->td_retval[0] != 0) break; } seltdclear(td); done: /* poll is not restarted after signals... */ if (error == ERESTART) error = EINTR; if (error == EWOULDBLOCK) error = 0; if (error == 0) { error = pollout(td, bits, fds, nfds); if (error) goto out; } out: if (ni > sizeof(smallbits)) free(bits, M_TEMP); return (error); } int sys_ppoll(struct thread *td, struct ppoll_args *uap) { struct timespec ts, *tsp; sigset_t set, *ssp; int error; if (uap->ts != NULL) { error = copyin(uap->ts, &ts, sizeof(ts)); if (error) return (error); tsp = &ts; } else tsp = NULL; if (uap->set != NULL) { error = copyin(uap->set, &set, sizeof(set)); if (error) return (error); ssp = &set; } else ssp = NULL; /* * fds is still a pointer to user space. kern_poll() will * take care of copyin that array to the kernel space. */ return (kern_poll(td, uap->fds, uap->nfds, tsp, ssp)); } static int pollrescan(struct thread *td) { struct seltd *stp; struct selfd *sfp; struct selfd *sfn; struct selinfo *si; struct filedesc *fdp; struct file *fp; struct pollfd *fd; #ifdef CAPABILITIES cap_rights_t rights; #endif int n; n = 0; fdp = td->td_proc->p_fd; stp = td->td_sel; FILEDESC_SLOCK(fdp); STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) { fd = (struct pollfd *)sfp->sf_cookie; si = sfp->sf_si; selfdfree(stp, sfp); /* If the selinfo wasn't cleared the event didn't fire. */ if (si != NULL) continue; fp = fdp->fd_ofiles[fd->fd].fde_file; #ifdef CAPABILITIES if (fp == NULL || cap_check(cap_rights(fdp, fd->fd), cap_rights_init(&rights, CAP_EVENT)) != 0) #else if (fp == NULL) #endif { fd->revents = POLLNVAL; n++; continue; } /* * Note: backend also returns POLLHUP and * POLLERR if appropriate. */ fd->revents = fo_poll(fp, fd->events, td->td_ucred, td); if (fd->revents != 0) n++; } FILEDESC_SUNLOCK(fdp); stp->st_flags = 0; td->td_retval[0] = n; return (0); } static int pollout(td, fds, ufds, nfd) struct thread *td; struct pollfd *fds; struct pollfd *ufds; u_int nfd; { int error = 0; u_int i = 0; u_int n = 0; for (i = 0; i < nfd; i++) { error = copyout(&fds->revents, &ufds->revents, sizeof(ufds->revents)); if (error) return (error); if (fds->revents != 0) n++; fds++; ufds++; } td->td_retval[0] = n; return (0); } static int pollscan(td, fds, nfd) struct thread *td; struct pollfd *fds; u_int nfd; { struct filedesc *fdp = td->td_proc->p_fd; struct file *fp; #ifdef CAPABILITIES cap_rights_t rights; #endif int i, n = 0; FILEDESC_SLOCK(fdp); for (i = 0; i < nfd; i++, fds++) { if (fds->fd > fdp->fd_lastfile) { fds->revents = POLLNVAL; n++; } else if (fds->fd < 0) { fds->revents = 0; } else { fp = fdp->fd_ofiles[fds->fd].fde_file; #ifdef CAPABILITIES if (fp == NULL || cap_check(cap_rights(fdp, fds->fd), cap_rights_init(&rights, CAP_EVENT)) != 0) #else if (fp == NULL) #endif { fds->revents = POLLNVAL; n++; } else { /* * Note: backend also returns POLLHUP and * POLLERR if appropriate. */ selfdalloc(td, fds); fds->revents = fo_poll(fp, fds->events, td->td_ucred, td); /* * POSIX requires POLLOUT to be never * set simultaneously with POLLHUP. */ if ((fds->revents & POLLHUP) != 0) fds->revents &= ~POLLOUT; if (fds->revents != 0) n++; } } } FILEDESC_SUNLOCK(fdp); td->td_retval[0] = n; return (0); } /* * XXX This was created specifically to support netncp and netsmb. This * allows the caller to specify a socket to wait for events on. It returns * 0 if any events matched and an error otherwise. There is no way to * determine which events fired. */ int selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td) { struct timeval rtv; sbintime_t asbt, precision, rsbt; int error; precision = 0; /* stupid gcc! */ if (tvp != NULL) { rtv = *tvp; if (rtv.tv_sec < 0 || rtv.tv_usec < 0 || rtv.tv_usec >= 1000000) return (EINVAL); if (!timevalisset(&rtv)) asbt = 0; else if (rtv.tv_sec <= INT32_MAX) { rsbt = tvtosbt(rtv); precision = rsbt; precision >>= tc_precexp; if (TIMESEL(&asbt, rsbt)) asbt += tc_tick_sbt; if (asbt <= SBT_MAX - rsbt) asbt += rsbt; else asbt = -1; } else asbt = -1; } else asbt = -1; seltdinit(td); /* * Iterate until the timeout expires or the socket becomes ready. */ for (;;) { selfdalloc(td, NULL); error = sopoll(so, events, NULL, td); /* error here is actually the ready events. */ if (error) return (0); error = seltdwait(td, asbt, precision); if (error) break; } seltdclear(td); /* XXX Duplicates ncp/smb behavior. */ if (error == ERESTART) error = 0; return (error); } /* * Preallocate two selfds associated with 'cookie'. Some fo_poll routines * have two select sets, one for read and another for write. */ static void selfdalloc(struct thread *td, void *cookie) { struct seltd *stp; stp = td->td_sel; if (stp->st_free1 == NULL) stp->st_free1 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO); stp->st_free1->sf_td = stp; stp->st_free1->sf_cookie = cookie; if (stp->st_free2 == NULL) stp->st_free2 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO); stp->st_free2->sf_td = stp; stp->st_free2->sf_cookie = cookie; } static void selfdfree(struct seltd *stp, struct selfd *sfp) { STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link); if (sfp->sf_si != NULL) { mtx_lock(sfp->sf_mtx); if (sfp->sf_si != NULL) { TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads); refcount_release(&sfp->sf_refs); } mtx_unlock(sfp->sf_mtx); } if (refcount_release(&sfp->sf_refs)) uma_zfree(selfd_zone, sfp); } /* Drain the waiters tied to all the selfd belonging the specified selinfo. */ void seldrain(sip) struct selinfo *sip; { /* * This feature is already provided by doselwakeup(), thus it is * enough to go for it. * Eventually, the context, should take care to avoid races * between thread calling select()/poll() and file descriptor * detaching, but, again, the races are just the same as * selwakeup(). */ doselwakeup(sip, -1); } /* * Record a select request. */ void selrecord(selector, sip) struct thread *selector; struct selinfo *sip; { struct selfd *sfp; struct seltd *stp; struct mtx *mtxp; stp = selector->td_sel; /* * Don't record when doing a rescan. */ if (stp->st_flags & SELTD_RESCAN) return; /* * Grab one of the preallocated descriptors. */ sfp = NULL; if ((sfp = stp->st_free1) != NULL) stp->st_free1 = NULL; else if ((sfp = stp->st_free2) != NULL) stp->st_free2 = NULL; else panic("selrecord: No free selfd on selq"); mtxp = sip->si_mtx; if (mtxp == NULL) mtxp = mtx_pool_find(mtxpool_select, sip); /* * Initialize the sfp and queue it in the thread. */ sfp->sf_si = sip; sfp->sf_mtx = mtxp; refcount_init(&sfp->sf_refs, 2); STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link); /* * Now that we've locked the sip, check for initialization. */ mtx_lock(mtxp); if (sip->si_mtx == NULL) { sip->si_mtx = mtxp; TAILQ_INIT(&sip->si_tdlist); } /* * Add this thread to the list of selfds listening on this selinfo. */ TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads); mtx_unlock(sip->si_mtx); } /* Wake up a selecting thread. */ void selwakeup(sip) struct selinfo *sip; { doselwakeup(sip, -1); } /* Wake up a selecting thread, and set its priority. */ void selwakeuppri(sip, pri) struct selinfo *sip; int pri; { doselwakeup(sip, pri); } /* * Do a wakeup when a selectable event occurs. */ static void doselwakeup(sip, pri) struct selinfo *sip; int pri; { struct selfd *sfp; struct selfd *sfn; struct seltd *stp; /* If it's not initialized there can't be any waiters. */ if (sip->si_mtx == NULL) return; /* * Locking the selinfo locks all selfds associated with it. */ mtx_lock(sip->si_mtx); TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) { /* * Once we remove this sfp from the list and clear the * sf_si seltdclear will know to ignore this si. */ TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads); sfp->sf_si = NULL; stp = sfp->sf_td; mtx_lock(&stp->st_mtx); stp->st_flags |= SELTD_PENDING; cv_broadcastpri(&stp->st_wait, pri); mtx_unlock(&stp->st_mtx); if (refcount_release(&sfp->sf_refs)) uma_zfree(selfd_zone, sfp); } mtx_unlock(sip->si_mtx); } static void seltdinit(struct thread *td) { struct seltd *stp; if ((stp = td->td_sel) != NULL) goto out; td->td_sel = stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO); mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF); cv_init(&stp->st_wait, "select"); out: stp->st_flags = 0; STAILQ_INIT(&stp->st_selq); } static int seltdwait(struct thread *td, sbintime_t sbt, sbintime_t precision) { struct seltd *stp; int error; stp = td->td_sel; /* * An event of interest may occur while we do not hold the seltd * locked so check the pending flag before we sleep. */ mtx_lock(&stp->st_mtx); /* * Any further calls to selrecord will be a rescan. */ stp->st_flags |= SELTD_RESCAN; if (stp->st_flags & SELTD_PENDING) { mtx_unlock(&stp->st_mtx); return (0); } if (sbt == 0) error = EWOULDBLOCK; else if (sbt != -1) error = cv_timedwait_sig_sbt(&stp->st_wait, &stp->st_mtx, sbt, precision, C_ABSOLUTE); else error = cv_wait_sig(&stp->st_wait, &stp->st_mtx); mtx_unlock(&stp->st_mtx); return (error); } void seltdfini(struct thread *td) { struct seltd *stp; stp = td->td_sel; if (stp == NULL) return; if (stp->st_free1) uma_zfree(selfd_zone, stp->st_free1); if (stp->st_free2) uma_zfree(selfd_zone, stp->st_free2); td->td_sel = NULL; free(stp, M_SELECT); } /* * Remove the references to the thread from all of the objects we were * polling. */ static void seltdclear(struct thread *td) { struct seltd *stp; struct selfd *sfp; struct selfd *sfn; stp = td->td_sel; STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) selfdfree(stp, sfp); stp->st_flags = 0; } static void selectinit(void *); SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL); static void selectinit(void *dummy __unused) { selfd_zone = uma_zcreate("selfd", sizeof(struct selfd), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF); } /* * Set up a syscall return value that follows the convention specified for * posix_* functions. */ int kern_posix_error(struct thread *td, int error) { if (error <= 0) return (error); td->td_errno = error; td->td_pflags |= TDP_NERRNO; td->td_retval[0] = error; return (0); } Index: head/sys/kern/sys_socket.c =================================================================== --- head/sys/kern/sys_socket.c (revision 305831) +++ head/sys/kern/sys_socket.c (revision 305832) @@ -1,802 +1,802 @@ /*- * Copyright (c) 1982, 1986, 1990, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)sys_socket.c 8.1 (Berkeley) 6/10/93 */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* XXX */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static SYSCTL_NODE(_kern_ipc, OID_AUTO, aio, CTLFLAG_RD, NULL, "socket AIO stats"); static int empty_results; SYSCTL_INT(_kern_ipc_aio, OID_AUTO, empty_results, CTLFLAG_RD, &empty_results, 0, "socket operation returned EAGAIN"); static int empty_retries; SYSCTL_INT(_kern_ipc_aio, OID_AUTO, empty_retries, CTLFLAG_RD, &empty_retries, 0, "socket operation retries"); static fo_rdwr_t soo_read; static fo_rdwr_t soo_write; static fo_ioctl_t soo_ioctl; static fo_poll_t soo_poll; extern fo_kqfilter_t soo_kqfilter; static fo_stat_t soo_stat; static fo_close_t soo_close; static fo_fill_kinfo_t soo_fill_kinfo; static fo_aio_queue_t soo_aio_queue; static void soo_aio_cancel(struct kaiocb *job); struct fileops socketops = { .fo_read = soo_read, .fo_write = soo_write, .fo_truncate = invfo_truncate, .fo_ioctl = soo_ioctl, .fo_poll = soo_poll, .fo_kqfilter = soo_kqfilter, .fo_stat = soo_stat, .fo_close = soo_close, .fo_chmod = invfo_chmod, .fo_chown = invfo_chown, .fo_sendfile = invfo_sendfile, .fo_fill_kinfo = soo_fill_kinfo, .fo_aio_queue = soo_aio_queue, .fo_flags = DFLAG_PASSABLE }; static int soo_read(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { struct socket *so = fp->f_data; int error; #ifdef MAC error = mac_socket_check_receive(active_cred, so); if (error) return (error); #endif error = soreceive(so, 0, uio, 0, 0, 0); return (error); } static int soo_write(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { struct socket *so = fp->f_data; int error; #ifdef MAC error = mac_socket_check_send(active_cred, so); if (error) return (error); #endif error = sosend(so, 0, uio, 0, 0, 0, uio->uio_td); if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0) { PROC_LOCK(uio->uio_td->td_proc); tdsignal(uio->uio_td, SIGPIPE); PROC_UNLOCK(uio->uio_td->td_proc); } return (error); } static int soo_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *active_cred, struct thread *td) { struct socket *so = fp->f_data; int error = 0; switch (cmd) { case FIONBIO: SOCK_LOCK(so); if (*(int *)data) so->so_state |= SS_NBIO; else so->so_state &= ~SS_NBIO; SOCK_UNLOCK(so); break; case FIOASYNC: /* * XXXRW: This code separately acquires SOCK_LOCK(so) and * SOCKBUF_LOCK(&so->so_rcv) even though they are the same * mutex to avoid introducing the assumption that they are * the same. */ if (*(int *)data) { SOCK_LOCK(so); so->so_state |= SS_ASYNC; SOCK_UNLOCK(so); SOCKBUF_LOCK(&so->so_rcv); so->so_rcv.sb_flags |= SB_ASYNC; SOCKBUF_UNLOCK(&so->so_rcv); SOCKBUF_LOCK(&so->so_snd); so->so_snd.sb_flags |= SB_ASYNC; SOCKBUF_UNLOCK(&so->so_snd); } else { SOCK_LOCK(so); so->so_state &= ~SS_ASYNC; SOCK_UNLOCK(so); SOCKBUF_LOCK(&so->so_rcv); so->so_rcv.sb_flags &= ~SB_ASYNC; SOCKBUF_UNLOCK(&so->so_rcv); SOCKBUF_LOCK(&so->so_snd); so->so_snd.sb_flags &= ~SB_ASYNC; SOCKBUF_UNLOCK(&so->so_snd); } break; case FIONREAD: /* Unlocked read. */ *(int *)data = sbavail(&so->so_rcv); break; case FIONWRITE: /* Unlocked read. */ *(int *)data = sbavail(&so->so_snd); break; case FIONSPACE: /* Unlocked read. */ if ((so->so_snd.sb_hiwat < sbused(&so->so_snd)) || (so->so_snd.sb_mbmax < so->so_snd.sb_mbcnt)) *(int *)data = 0; else *(int *)data = sbspace(&so->so_snd); break; case FIOSETOWN: error = fsetown(*(int *)data, &so->so_sigio); break; case FIOGETOWN: *(int *)data = fgetown(&so->so_sigio); break; case SIOCSPGRP: error = fsetown(-(*(int *)data), &so->so_sigio); break; case SIOCGPGRP: *(int *)data = -fgetown(&so->so_sigio); break; case SIOCATMARK: /* Unlocked read. */ *(int *)data = (so->so_rcv.sb_state & SBS_RCVATMARK) != 0; break; default: /* * Interface/routing/protocol specific ioctls: interface and * routing ioctls should have a different entry since a * socket is unnecessary. */ if (IOCGROUP(cmd) == 'i') error = ifioctl(so, cmd, data, td); else if (IOCGROUP(cmd) == 'r') { CURVNET_SET(so->so_vnet); error = rtioctl_fib(cmd, data, so->so_fibnum); CURVNET_RESTORE(); } else { CURVNET_SET(so->so_vnet); error = ((*so->so_proto->pr_usrreqs->pru_control) (so, cmd, data, 0, td)); CURVNET_RESTORE(); } break; } return (error); } static int soo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td) { struct socket *so = fp->f_data; #ifdef MAC int error; error = mac_socket_check_poll(active_cred, so); if (error) return (error); #endif return (sopoll(so, events, fp->f_cred, td)); } static int soo_stat(struct file *fp, struct stat *ub, struct ucred *active_cred, struct thread *td) { struct socket *so = fp->f_data; struct sockbuf *sb; #ifdef MAC int error; #endif bzero((caddr_t)ub, sizeof (*ub)); ub->st_mode = S_IFSOCK; #ifdef MAC error = mac_socket_check_stat(active_cred, so); if (error) return (error); #endif /* * If SBS_CANTRCVMORE is set, but there's still data left in the * receive buffer, the socket is still readable. */ sb = &so->so_rcv; SOCKBUF_LOCK(sb); if ((sb->sb_state & SBS_CANTRCVMORE) == 0 || sbavail(sb)) ub->st_mode |= S_IRUSR | S_IRGRP | S_IROTH; ub->st_size = sbavail(sb) - sb->sb_ctl; SOCKBUF_UNLOCK(sb); sb = &so->so_snd; SOCKBUF_LOCK(sb); if ((sb->sb_state & SBS_CANTSENDMORE) == 0) ub->st_mode |= S_IWUSR | S_IWGRP | S_IWOTH; SOCKBUF_UNLOCK(sb); ub->st_uid = so->so_cred->cr_uid; ub->st_gid = so->so_cred->cr_gid; return (*so->so_proto->pr_usrreqs->pru_sense)(so, ub); } /* * API socket close on file pointer. We call soclose() to close the socket * (including initiating closing protocols). soclose() will sorele() the * file reference but the actual socket will not go away until the socket's * ref count hits 0. */ static int soo_close(struct file *fp, struct thread *td) { int error = 0; struct socket *so; so = fp->f_data; fp->f_ops = &badfileops; fp->f_data = NULL; if (so) error = soclose(so); return (error); } static int soo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) { struct sockaddr *sa; struct inpcb *inpcb; struct unpcb *unpcb; struct socket *so; int error; kif->kf_type = KF_TYPE_SOCKET; so = fp->f_data; kif->kf_sock_domain = so->so_proto->pr_domain->dom_family; kif->kf_sock_type = so->so_type; kif->kf_sock_protocol = so->so_proto->pr_protocol; kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb; switch (kif->kf_sock_domain) { case AF_INET: case AF_INET6: if (kif->kf_sock_protocol == IPPROTO_TCP) { if (so->so_pcb != NULL) { inpcb = (struct inpcb *)(so->so_pcb); kif->kf_un.kf_sock.kf_sock_inpcb = (uintptr_t)inpcb->inp_ppcb; } } break; case AF_UNIX: if (so->so_pcb != NULL) { unpcb = (struct unpcb *)(so->so_pcb); if (unpcb->unp_conn) { kif->kf_un.kf_sock.kf_sock_unpconn = (uintptr_t)unpcb->unp_conn; kif->kf_un.kf_sock.kf_sock_rcv_sb_state = so->so_rcv.sb_state; kif->kf_un.kf_sock.kf_sock_snd_sb_state = so->so_snd.sb_state; } } break; } error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa); if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { bcopy(sa, &kif->kf_sa_local, sa->sa_len); free(sa, M_SONAME); } error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa); if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { bcopy(sa, &kif->kf_sa_peer, sa->sa_len); free(sa, M_SONAME); } strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name, sizeof(kif->kf_path)); return (0); } /* * Use the 'backend3' field in AIO jobs to store the amount of data * completed by the AIO job so far. */ #define aio_done backend3 static STAILQ_HEAD(, task) soaio_jobs; static struct mtx soaio_jobs_lock; static struct task soaio_kproc_task; static int soaio_starting, soaio_idle, soaio_queued; static struct unrhdr *soaio_kproc_unr; static int soaio_max_procs = MAX_AIO_PROCS; SYSCTL_INT(_kern_ipc_aio, OID_AUTO, max_procs, CTLFLAG_RW, &soaio_max_procs, 0, "Maximum number of kernel processes to use for async socket IO"); static int soaio_num_procs; SYSCTL_INT(_kern_ipc_aio, OID_AUTO, num_procs, CTLFLAG_RD, &soaio_num_procs, 0, "Number of active kernel processes for async socket IO"); static int soaio_target_procs = TARGET_AIO_PROCS; SYSCTL_INT(_kern_ipc_aio, OID_AUTO, target_procs, CTLFLAG_RD, &soaio_target_procs, 0, "Preferred number of ready kernel processes for async socket IO"); static int soaio_lifetime; SYSCTL_INT(_kern_ipc_aio, OID_AUTO, lifetime, CTLFLAG_RW, &soaio_lifetime, 0, "Maximum lifetime for idle aiod"); static void soaio_kproc_loop(void *arg) { struct proc *p; struct vmspace *myvm; struct task *task; int error, id, pending; id = (intptr_t)arg; /* * Grab an extra reference on the daemon's vmspace so that it * doesn't get freed by jobs that switch to a different * vmspace. */ p = curproc; myvm = vmspace_acquire_ref(p); mtx_lock(&soaio_jobs_lock); MPASS(soaio_starting > 0); soaio_starting--; for (;;) { while (!STAILQ_EMPTY(&soaio_jobs)) { task = STAILQ_FIRST(&soaio_jobs); STAILQ_REMOVE_HEAD(&soaio_jobs, ta_link); soaio_queued--; pending = task->ta_pending; task->ta_pending = 0; mtx_unlock(&soaio_jobs_lock); task->ta_func(task->ta_context, pending); mtx_lock(&soaio_jobs_lock); } MPASS(soaio_queued == 0); if (p->p_vmspace != myvm) { mtx_unlock(&soaio_jobs_lock); vmspace_switch_aio(myvm); mtx_lock(&soaio_jobs_lock); continue; } soaio_idle++; error = mtx_sleep(&soaio_idle, &soaio_jobs_lock, 0, "-", soaio_lifetime); soaio_idle--; if (error == EWOULDBLOCK && STAILQ_EMPTY(&soaio_jobs) && soaio_num_procs > soaio_target_procs) break; } soaio_num_procs--; mtx_unlock(&soaio_jobs_lock); free_unr(soaio_kproc_unr, id); kproc_exit(0); } static void soaio_kproc_create(void *context, int pending) { struct proc *p; int error, id; mtx_lock(&soaio_jobs_lock); for (;;) { if (soaio_num_procs < soaio_target_procs) { /* Must create */ } else if (soaio_num_procs >= soaio_max_procs) { /* * Hit the limit on kernel processes, don't * create another one. */ break; } else if (soaio_queued <= soaio_idle + soaio_starting) { /* * No more AIO jobs waiting for a process to be * created, so stop. */ break; } soaio_starting++; mtx_unlock(&soaio_jobs_lock); id = alloc_unr(soaio_kproc_unr); error = kproc_create(soaio_kproc_loop, (void *)(intptr_t)id, &p, 0, 0, "soaiod%d", id); if (error != 0) { free_unr(soaio_kproc_unr, id); mtx_lock(&soaio_jobs_lock); soaio_starting--; break; } mtx_lock(&soaio_jobs_lock); soaio_num_procs++; } mtx_unlock(&soaio_jobs_lock); } void soaio_enqueue(struct task *task) { mtx_lock(&soaio_jobs_lock); MPASS(task->ta_pending == 0); task->ta_pending++; STAILQ_INSERT_TAIL(&soaio_jobs, task, ta_link); soaio_queued++; if (soaio_queued <= soaio_idle) wakeup_one(&soaio_idle); else if (soaio_num_procs < soaio_max_procs) taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task); mtx_unlock(&soaio_jobs_lock); } static void soaio_init(void) { soaio_lifetime = AIOD_LIFETIME_DEFAULT; STAILQ_INIT(&soaio_jobs); mtx_init(&soaio_jobs_lock, "soaio jobs", NULL, MTX_DEF); soaio_kproc_unr = new_unrhdr(1, INT_MAX, NULL); TASK_INIT(&soaio_kproc_task, 0, soaio_kproc_create, NULL); if (soaio_target_procs > 0) taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task); } SYSINIT(soaio, SI_SUB_VFS, SI_ORDER_ANY, soaio_init, NULL); static __inline int soaio_ready(struct socket *so, struct sockbuf *sb) { return (sb == &so->so_rcv ? soreadable(so) : sowriteable(so)); } static void soaio_process_job(struct socket *so, struct sockbuf *sb, struct kaiocb *job) { struct ucred *td_savedcred; struct thread *td; struct file *fp; struct uio uio; struct iovec iov; size_t cnt, done; long ru_before; int error, flags; SOCKBUF_UNLOCK(sb); aio_switch_vmspace(job); td = curthread; fp = job->fd_file; retry: td_savedcred = td->td_ucred; td->td_ucred = job->cred; done = job->aio_done; cnt = job->uaiocb.aio_nbytes - done; iov.iov_base = (void *)((uintptr_t)job->uaiocb.aio_buf + done); iov.iov_len = cnt; uio.uio_iov = &iov; uio.uio_iovcnt = 1; uio.uio_offset = 0; uio.uio_resid = cnt; uio.uio_segflg = UIO_USERSPACE; uio.uio_td = td; flags = MSG_NBIO; /* * For resource usage accounting, only count a completed request * as a single message to avoid counting multiple calls to * sosend/soreceive on a blocking socket. */ if (sb == &so->so_rcv) { uio.uio_rw = UIO_READ; ru_before = td->td_ru.ru_msgrcv; #ifdef MAC error = mac_socket_check_receive(fp->f_cred, so); if (error == 0) #endif error = soreceive(so, NULL, &uio, NULL, NULL, &flags); if (td->td_ru.ru_msgrcv != ru_before) job->msgrcv = 1; } else { uio.uio_rw = UIO_WRITE; ru_before = td->td_ru.ru_msgsnd; #ifdef MAC error = mac_socket_check_send(fp->f_cred, so); if (error == 0) #endif error = sosend(so, NULL, &uio, NULL, NULL, flags, td); if (td->td_ru.ru_msgsnd != ru_before) job->msgsnd = 1; if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0) { PROC_LOCK(job->userproc); kern_psignal(job->userproc, SIGPIPE); PROC_UNLOCK(job->userproc); } } done += cnt - uio.uio_resid; job->aio_done = done; td->td_ucred = td_savedcred; if (error == EWOULDBLOCK) { /* * The request was either partially completed or not * completed at all due to racing with a read() or * write() on the socket. If the socket is * non-blocking, return with any partial completion. * If the socket is blocking or if no progress has * been made, requeue this request at the head of the * queue to try again when the socket is ready. */ MPASS(done != job->uaiocb.aio_nbytes); SOCKBUF_LOCK(sb); if (done == 0 || !(so->so_state & SS_NBIO)) { empty_results++; if (soaio_ready(so, sb)) { empty_retries++; SOCKBUF_UNLOCK(sb); goto retry; } if (!aio_set_cancel_function(job, soo_aio_cancel)) { SOCKBUF_UNLOCK(sb); if (done != 0) aio_complete(job, done, 0); else aio_cancel(job); SOCKBUF_LOCK(sb); } else { TAILQ_INSERT_HEAD(&sb->sb_aiojobq, job, list); } return; } SOCKBUF_UNLOCK(sb); } if (done != 0 && (error == ERESTART || error == EINTR || error == EWOULDBLOCK)) error = 0; if (error) aio_complete(job, -1, error); else aio_complete(job, done, 0); SOCKBUF_LOCK(sb); } static void soaio_process_sb(struct socket *so, struct sockbuf *sb) { struct kaiocb *job; SOCKBUF_LOCK(sb); while (!TAILQ_EMPTY(&sb->sb_aiojobq) && soaio_ready(so, sb)) { job = TAILQ_FIRST(&sb->sb_aiojobq); TAILQ_REMOVE(&sb->sb_aiojobq, job, list); if (!aio_clear_cancel_function(job)) continue; soaio_process_job(so, sb, job); } /* * If there are still pending requests, the socket must not be * ready so set SB_AIO to request a wakeup when the socket * becomes ready. */ if (!TAILQ_EMPTY(&sb->sb_aiojobq)) sb->sb_flags |= SB_AIO; sb->sb_flags &= ~SB_AIO_RUNNING; SOCKBUF_UNLOCK(sb); ACCEPT_LOCK(); SOCK_LOCK(so); sorele(so); } void soaio_rcv(void *context, int pending) { struct socket *so; so = context; soaio_process_sb(so, &so->so_rcv); } void soaio_snd(void *context, int pending) { struct socket *so; so = context; soaio_process_sb(so, &so->so_snd); } void sowakeup_aio(struct socket *so, struct sockbuf *sb) { SOCKBUF_LOCK_ASSERT(sb); sb->sb_flags &= ~SB_AIO; if (sb->sb_flags & SB_AIO_RUNNING) return; sb->sb_flags |= SB_AIO_RUNNING; if (sb == &so->so_snd) SOCK_LOCK(so); soref(so); if (sb == &so->so_snd) SOCK_UNLOCK(so); soaio_enqueue(&sb->sb_aiotask); } static void soo_aio_cancel(struct kaiocb *job) { struct socket *so; struct sockbuf *sb; long done; int opcode; so = job->fd_file->f_data; opcode = job->uaiocb.aio_lio_opcode; if (opcode == LIO_READ) sb = &so->so_rcv; else { MPASS(opcode == LIO_WRITE); sb = &so->so_snd; } SOCKBUF_LOCK(sb); if (!aio_cancel_cleared(job)) TAILQ_REMOVE(&sb->sb_aiojobq, job, list); if (TAILQ_EMPTY(&sb->sb_aiojobq)) sb->sb_flags &= ~SB_AIO; SOCKBUF_UNLOCK(sb); done = job->aio_done; if (done != 0) aio_complete(job, done, 0); else aio_cancel(job); } static int soo_aio_queue(struct file *fp, struct kaiocb *job) { struct socket *so; struct sockbuf *sb; int error; so = fp->f_data; error = (*so->so_proto->pr_usrreqs->pru_aio_queue)(so, job); if (error == 0) return (0); switch (job->uaiocb.aio_lio_opcode) { case LIO_READ: sb = &so->so_rcv; break; case LIO_WRITE: sb = &so->so_snd; break; default: return (EINVAL); } SOCKBUF_LOCK(sb); if (!aio_set_cancel_function(job, soo_aio_cancel)) panic("new job was cancelled"); TAILQ_INSERT_TAIL(&sb->sb_aiojobq, job, list); if (!(sb->sb_flags & SB_AIO_RUNNING)) { if (soaio_ready(so, sb)) sowakeup_aio(so, sb); else sb->sb_flags |= SB_AIO; } SOCKBUF_UNLOCK(sb); return (0); } Index: head/sys/kern/tty_compat.c =================================================================== --- head/sys/kern/tty_compat.c (revision 305831) +++ head/sys/kern/tty_compat.c (revision 305832) @@ -1,484 +1,484 @@ /*- * Copyright (c) 1982, 1986, 1991, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)tty_compat.c 8.1 (Berkeley) 6/10/93 */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" /* * mapping routines for old line discipline (yuck) */ #include #include #include #include #include #include struct speedtab { int sp_speed; /* Speed. */ int sp_code; /* Code. */ }; static int ttcompatgetflags(struct tty *tp); static void ttcompatsetflags(struct tty *tp, struct termios *t); static void ttcompatsetlflags(struct tty *tp, struct termios *t); static int ttcompatspeedtab(int speed, struct speedtab *table); static int ttydebug = 0; SYSCTL_INT(_debug, OID_AUTO, ttydebug, CTLFLAG_RW, &ttydebug, 0, ""); static struct speedtab compatspeeds[] = { #define MAX_SPEED 17 { 115200, 17 }, { 57600, 16 }, { 38400, 15 }, { 19200, 14 }, { 9600, 13 }, { 4800, 12 }, { 2400, 11 }, { 1800, 10 }, { 1200, 9 }, { 600, 8 }, { 300, 7 }, { 200, 6 }, { 150, 5 }, { 134, 4 }, { 110, 3 }, { 75, 2 }, { 50, 1 }, { 0, 0 }, { -1, -1 }, }; static int compatspcodes[] = { 0, 50, 75, 110, 134, 150, 200, 300, 600, 1200, 1800, 2400, 4800, 9600, 19200, 38400, 57600, 115200, }; static int ttcompatspeedtab(int speed, struct speedtab *table) { if (speed == 0) return (0); /* hangup */ for ( ; table->sp_speed > 0; table++) if (table->sp_speed <= speed) /* nearest one, rounded down */ return (table->sp_code); return (1); /* 50, min and not hangup */ } static int ttsetcompat(struct tty *tp, u_long *com, caddr_t data, struct termios *term) { switch (*com) { case TIOCSETP: case TIOCSETN: { struct sgttyb *sg = (struct sgttyb *)data; int speed; if ((speed = sg->sg_ispeed) > MAX_SPEED || speed < 0) return(EINVAL); else if (speed != ttcompatspeedtab(tp->t_termios.c_ispeed, compatspeeds)) term->c_ispeed = compatspcodes[speed]; else term->c_ispeed = tp->t_termios.c_ispeed; if ((speed = sg->sg_ospeed) > MAX_SPEED || speed < 0) return(EINVAL); else if (speed != ttcompatspeedtab(tp->t_termios.c_ospeed, compatspeeds)) term->c_ospeed = compatspcodes[speed]; else term->c_ospeed = tp->t_termios.c_ospeed; term->c_cc[VERASE] = sg->sg_erase; term->c_cc[VKILL] = sg->sg_kill; tp->t_compatflags = (tp->t_compatflags&0xffff0000) | (sg->sg_flags&0xffff); ttcompatsetflags(tp, term); *com = (*com == TIOCSETP) ? TIOCSETAF : TIOCSETA; break; } case TIOCSETC: { struct tchars *tc = (struct tchars *)data; cc_t *cc; cc = term->c_cc; cc[VINTR] = tc->t_intrc; cc[VQUIT] = tc->t_quitc; cc[VSTART] = tc->t_startc; cc[VSTOP] = tc->t_stopc; cc[VEOF] = tc->t_eofc; cc[VEOL] = tc->t_brkc; if (tc->t_brkc == (char)_POSIX_VDISABLE) cc[VEOL2] = _POSIX_VDISABLE; *com = TIOCSETA; break; } case TIOCSLTC: { struct ltchars *ltc = (struct ltchars *)data; cc_t *cc; cc = term->c_cc; cc[VSUSP] = ltc->t_suspc; cc[VDSUSP] = ltc->t_dsuspc; cc[VREPRINT] = ltc->t_rprntc; cc[VDISCARD] = ltc->t_flushc; cc[VWERASE] = ltc->t_werasc; cc[VLNEXT] = ltc->t_lnextc; *com = TIOCSETA; break; } case TIOCLBIS: case TIOCLBIC: case TIOCLSET: if (*com == TIOCLSET) tp->t_compatflags = (tp->t_compatflags&0xffff) | *(int *)data<<16; else { tp->t_compatflags = (ttcompatgetflags(tp)&0xffff0000) | (tp->t_compatflags&0xffff); if (*com == TIOCLBIS) tp->t_compatflags |= *(int *)data<<16; else tp->t_compatflags &= ~(*(int *)data<<16); } ttcompatsetlflags(tp, term); *com = TIOCSETA; break; } return 0; } /*ARGSUSED*/ int tty_ioctl_compat(struct tty *tp, u_long com, caddr_t data, int fflag, struct thread *td) { switch (com) { case TIOCSETP: case TIOCSETN: case TIOCSETC: case TIOCSLTC: case TIOCLBIS: case TIOCLBIC: case TIOCLSET: { struct termios term; int error; term = tp->t_termios; if ((error = ttsetcompat(tp, &com, data, &term)) != 0) return error; return tty_ioctl(tp, com, &term, fflag, td); } case TIOCGETP: { struct sgttyb *sg = (struct sgttyb *)data; cc_t *cc = tp->t_termios.c_cc; sg->sg_ospeed = ttcompatspeedtab(tp->t_termios.c_ospeed, compatspeeds); if (tp->t_termios.c_ispeed == 0) sg->sg_ispeed = sg->sg_ospeed; else sg->sg_ispeed = ttcompatspeedtab(tp->t_termios.c_ispeed, compatspeeds); sg->sg_erase = cc[VERASE]; sg->sg_kill = cc[VKILL]; sg->sg_flags = tp->t_compatflags = ttcompatgetflags(tp); break; } case TIOCGETC: { struct tchars *tc = (struct tchars *)data; cc_t *cc = tp->t_termios.c_cc; tc->t_intrc = cc[VINTR]; tc->t_quitc = cc[VQUIT]; tc->t_startc = cc[VSTART]; tc->t_stopc = cc[VSTOP]; tc->t_eofc = cc[VEOF]; tc->t_brkc = cc[VEOL]; break; } case TIOCGLTC: { struct ltchars *ltc = (struct ltchars *)data; cc_t *cc = tp->t_termios.c_cc; ltc->t_suspc = cc[VSUSP]; ltc->t_dsuspc = cc[VDSUSP]; ltc->t_rprntc = cc[VREPRINT]; ltc->t_flushc = cc[VDISCARD]; ltc->t_werasc = cc[VWERASE]; ltc->t_lnextc = cc[VLNEXT]; break; } case TIOCLGET: tp->t_compatflags = (ttcompatgetflags(tp) & 0xffff0000UL) | (tp->t_compatflags & 0xffff); *(int *)data = tp->t_compatflags>>16; if (ttydebug) printf("CLGET: returning %x\n", *(int *)data); break; case OTIOCGETD: *(int *)data = 2; break; case OTIOCSETD: { int ldisczero = 0; return (tty_ioctl(tp, TIOCSETD, *(int *)data == 2 ? (caddr_t)&ldisczero : data, fflag, td)); } case OTIOCCONS: *(int *)data = 1; return (tty_ioctl(tp, TIOCCONS, data, fflag, td)); default: return (ENOIOCTL); } return (0); } static int ttcompatgetflags(struct tty *tp) { tcflag_t iflag = tp->t_termios.c_iflag; tcflag_t lflag = tp->t_termios.c_lflag; tcflag_t oflag = tp->t_termios.c_oflag; tcflag_t cflag = tp->t_termios.c_cflag; int flags = 0; if (iflag&IXOFF) flags |= TANDEM; if (iflag&ICRNL || oflag&ONLCR) flags |= CRMOD; if ((cflag&CSIZE) == CS8) { flags |= PASS8; if (iflag&ISTRIP) flags |= ANYP; } else if (cflag&PARENB) { if (iflag&INPCK) { if (cflag&PARODD) flags |= ODDP; else flags |= EVENP; } else flags |= EVENP | ODDP; } if ((lflag&ICANON) == 0) { /* fudge */ if (iflag&(INPCK|ISTRIP|IXON) || lflag&(IEXTEN|ISIG) || (cflag&(CSIZE|PARENB)) != CS8) flags |= CBREAK; else flags |= RAW; } if (!(flags&RAW) && !(oflag&OPOST) && (cflag&(CSIZE|PARENB)) == CS8) flags |= LITOUT; if (cflag&MDMBUF) flags |= MDMBUF; if ((cflag&HUPCL) == 0) flags |= NOHANG; if (oflag&TAB3) flags |= XTABS; if (lflag&ECHOE) flags |= CRTERA|CRTBS; if (lflag&ECHOKE) flags |= CRTKIL|CRTBS; if (lflag&ECHOPRT) flags |= PRTERA; if (lflag&ECHOCTL) flags |= CTLECH; if ((iflag&IXANY) == 0) flags |= DECCTQ; flags |= lflag&(ECHO|TOSTOP|FLUSHO|PENDIN|NOFLSH); if (ttydebug) printf("getflags: %x\n", flags); return (flags); } static void ttcompatsetflags(struct tty *tp, struct termios *t) { int flags = tp->t_compatflags; tcflag_t iflag = t->c_iflag; tcflag_t oflag = t->c_oflag; tcflag_t lflag = t->c_lflag; tcflag_t cflag = t->c_cflag; if (flags & RAW) { iflag = IGNBRK; lflag &= ~(ECHOCTL|ISIG|ICANON|IEXTEN); } else { iflag &= ~(PARMRK|IGNPAR|IGNCR|INLCR); iflag |= BRKINT|IXON|IMAXBEL; lflag |= ISIG|IEXTEN|ECHOCTL; /* XXX was echoctl on ? */ if (flags & XTABS) oflag |= TAB3; else oflag &= ~TAB3; if (flags & CBREAK) lflag &= ~ICANON; else lflag |= ICANON; if (flags&CRMOD) { iflag |= ICRNL; oflag |= ONLCR; } else { iflag &= ~ICRNL; oflag &= ~ONLCR; } } if (flags&ECHO) lflag |= ECHO; else lflag &= ~ECHO; cflag &= ~(CSIZE|PARENB); if (flags&(RAW|LITOUT|PASS8)) { cflag |= CS8; if (!(flags&(RAW|PASS8)) || (flags&(RAW|PASS8|ANYP)) == (PASS8|ANYP)) iflag |= ISTRIP; else iflag &= ~ISTRIP; if (flags&(RAW|LITOUT)) oflag &= ~OPOST; else oflag |= OPOST; } else { cflag |= CS7|PARENB; iflag |= ISTRIP; oflag |= OPOST; } /* XXX don't set INPCK if RAW or PASS8? */ if ((flags&(EVENP|ODDP)) == EVENP) { iflag |= INPCK; cflag &= ~PARODD; } else if ((flags&(EVENP|ODDP)) == ODDP) { iflag |= INPCK; cflag |= PARODD; } else iflag &= ~INPCK; if (flags&TANDEM) iflag |= IXOFF; else iflag &= ~IXOFF; if ((flags&DECCTQ) == 0) iflag |= IXANY; else iflag &= ~IXANY; t->c_iflag = iflag; t->c_oflag = oflag; t->c_lflag = lflag; t->c_cflag = cflag; } static void ttcompatsetlflags(struct tty *tp, struct termios *t) { int flags = tp->t_compatflags; tcflag_t iflag = t->c_iflag; tcflag_t oflag = t->c_oflag; tcflag_t lflag = t->c_lflag; tcflag_t cflag = t->c_cflag; iflag &= ~(PARMRK|IGNPAR|IGNCR|INLCR); if (flags&CRTERA) lflag |= ECHOE; else lflag &= ~ECHOE; if (flags&CRTKIL) lflag |= ECHOKE; else lflag &= ~ECHOKE; if (flags&PRTERA) lflag |= ECHOPRT; else lflag &= ~ECHOPRT; if (flags&CTLECH) lflag |= ECHOCTL; else lflag &= ~ECHOCTL; if (flags&TANDEM) iflag |= IXOFF; else iflag &= ~IXOFF; if ((flags&DECCTQ) == 0) iflag |= IXANY; else iflag &= ~IXANY; if (flags & MDMBUF) cflag |= MDMBUF; else cflag &= ~MDMBUF; if (flags&NOHANG) cflag &= ~HUPCL; else cflag |= HUPCL; lflag &= ~(TOSTOP|FLUSHO|PENDIN|NOFLSH); lflag |= flags&(TOSTOP|FLUSHO|PENDIN|NOFLSH); /* * The next if-else statement is copied from above so don't bother * checking it separately. We could avoid fiddlling with the * character size if the mode is already RAW or if neither the * LITOUT bit or the PASS8 bit is being changed, but the delta of * the change is not available here and skipping the RAW case would * make the code different from above. */ cflag &= ~(CSIZE|PARENB); if (flags&(RAW|LITOUT|PASS8)) { cflag |= CS8; if (!(flags&(RAW|PASS8)) || (flags&(RAW|PASS8|ANYP)) == (PASS8|ANYP)) iflag |= ISTRIP; else iflag &= ~ISTRIP; if (flags&(RAW|LITOUT)) oflag &= ~OPOST; else oflag |= OPOST; } else { cflag |= CS7|PARENB; iflag |= ISTRIP; oflag |= OPOST; } t->c_iflag = iflag; t->c_oflag = oflag; t->c_lflag = lflag; t->c_cflag = cflag; } Index: head/sys/kern/tty_info.c =================================================================== --- head/sys/kern/tty_info.c (revision 305831) +++ head/sys/kern/tty_info.c (revision 305832) @@ -1,313 +1,313 @@ /*- * Copyright (c) 1982, 1986, 1990, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Copyright (c) 2002 Networks Associates Technologies, Inc. * All rights reserved. * * Portions of this software were developed for the FreeBSD Project by * ThinkSec AS and NAI Labs, the Security Research Division of Network * Associates, Inc. under DARPA/SPAWAR contract N66001-01-C-8035 * ("CBOSS"), as part of the DARPA CHATS research program. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include /* * Returns 1 if p2 is "better" than p1 * * The algorithm for picking the "interesting" process is thus: * * 1) Only foreground processes are eligible - implied. * 2) Runnable processes are favored over anything else. The runner * with the highest cpu utilization is picked (p_estcpu). Ties are * broken by picking the highest pid. * 3) The sleeper with the shortest sleep time is next. With ties, * we pick out just "short-term" sleepers (P_SINTR == 0). * 4) Further ties are broken by picking the highest pid. */ #define TESTAB(a, b) ((a)<<1 | (b)) #define ONLYA 2 #define ONLYB 1 #define BOTH 3 static int proc_sum(struct proc *p, fixpt_t *estcpup) { struct thread *td; int estcpu; int val; val = 0; estcpu = 0; FOREACH_THREAD_IN_PROC(p, td) { thread_lock(td); if (TD_ON_RUNQ(td) || TD_IS_RUNNING(td)) val = 1; estcpu += sched_pctcpu(td); thread_unlock(td); } *estcpup = estcpu; return (val); } static int thread_compare(struct thread *td, struct thread *td2) { int runa, runb; int slpa, slpb; fixpt_t esta, estb; if (td == NULL) return (1); /* * Fetch running stats, pctcpu usage, and interruptable flag. */ thread_lock(td); runa = TD_IS_RUNNING(td) | TD_ON_RUNQ(td); slpa = td->td_flags & TDF_SINTR; esta = sched_pctcpu(td); thread_unlock(td); thread_lock(td2); runb = TD_IS_RUNNING(td2) | TD_ON_RUNQ(td2); estb = sched_pctcpu(td2); slpb = td2->td_flags & TDF_SINTR; thread_unlock(td2); /* * see if at least one of them is runnable */ switch (TESTAB(runa, runb)) { case ONLYA: return (0); case ONLYB: return (1); case BOTH: break; } /* * favor one with highest recent cpu utilization */ if (estb > esta) return (1); if (esta > estb) return (0); /* * favor one sleeping in a non-interruptible sleep */ switch (TESTAB(slpa, slpb)) { case ONLYA: return (0); case ONLYB: return (1); case BOTH: break; } return (td < td2); } static int proc_compare(struct proc *p1, struct proc *p2) { int runa, runb; fixpt_t esta, estb; if (p1 == NULL) return (1); /* * Fetch various stats about these processes. After we drop the * lock the information could be stale but the race is unimportant. */ PROC_LOCK(p1); runa = proc_sum(p1, &esta); PROC_UNLOCK(p1); PROC_LOCK(p2); runb = proc_sum(p2, &estb); PROC_UNLOCK(p2); /* * see if at least one of them is runnable */ switch (TESTAB(runa, runb)) { case ONLYA: return (0); case ONLYB: return (1); case BOTH: break; } /* * favor one with highest recent cpu utilization */ if (estb > esta) return (1); if (esta > estb) return (0); /* * weed out zombies */ switch (TESTAB(p1->p_state == PRS_ZOMBIE, p2->p_state == PRS_ZOMBIE)) { case ONLYA: return (1); case ONLYB: return (0); case BOTH: break; } return (p2->p_pid > p1->p_pid); /* tie - return highest pid */ } /* * Report on state of foreground process group. */ void tty_info(struct tty *tp) { struct timeval rtime, utime, stime; struct proc *p, *ppick; struct thread *td, *tdpick; const char *stateprefix, *state; long rss; int load, pctcpu; pid_t pid; char comm[MAXCOMLEN + 1]; struct rusage ru; tty_lock_assert(tp, MA_OWNED); if (tty_checkoutq(tp) == 0) return; /* Print load average. */ load = (averunnable.ldavg[0] * 100 + FSCALE / 2) >> FSHIFT; ttyprintf(tp, "%sload: %d.%02d ", tp->t_column == 0 ? "" : "\n", load / 100, load % 100); if (tp->t_session == NULL) { ttyprintf(tp, "not a controlling terminal\n"); return; } if (tp->t_pgrp == NULL) { ttyprintf(tp, "no foreground process group\n"); return; } PGRP_LOCK(tp->t_pgrp); if (LIST_EMPTY(&tp->t_pgrp->pg_members)) { PGRP_UNLOCK(tp->t_pgrp); ttyprintf(tp, "empty foreground process group\n"); return; } /* * Pick the most interesting process and copy some of its * state for printing later. This operation could rely on stale * data as we can't hold the proc slock or thread locks over the * whole list. However, we're guaranteed not to reference an exited * thread or proc since we hold the tty locked. */ p = NULL; LIST_FOREACH(ppick, &tp->t_pgrp->pg_members, p_pglist) if (proc_compare(p, ppick)) p = ppick; PROC_LOCK(p); PGRP_UNLOCK(tp->t_pgrp); td = NULL; FOREACH_THREAD_IN_PROC(p, tdpick) if (thread_compare(td, tdpick)) td = tdpick; stateprefix = ""; thread_lock(td); if (TD_IS_RUNNING(td)) state = "running"; else if (TD_ON_RUNQ(td) || TD_CAN_RUN(td)) state = "runnable"; else if (TD_IS_SLEEPING(td)) { /* XXX: If we're sleeping, are we ever not in a queue? */ if (TD_ON_SLEEPQ(td)) state = td->td_wmesg; else state = "sleeping without queue"; } else if (TD_ON_LOCK(td)) { state = td->td_lockname; stateprefix = "*"; } else if (TD_IS_SUSPENDED(td)) state = "suspended"; else if (TD_AWAITING_INTR(td)) state = "intrwait"; else if (p->p_state == PRS_ZOMBIE) state = "zombie"; else state = "unknown"; pctcpu = (sched_pctcpu(td) * 10000 + FSCALE / 2) >> FSHIFT; thread_unlock(td); if (p->p_state == PRS_NEW || p->p_state == PRS_ZOMBIE) rss = 0; else rss = pgtok(vmspace_resident_count(p->p_vmspace)); microuptime(&rtime); timevalsub(&rtime, &p->p_stats->p_start); rufetchcalc(p, &ru, &utime, &stime); pid = p->p_pid; strlcpy(comm, p->p_comm, sizeof comm); PROC_UNLOCK(p); /* Print command, pid, state, rtime, utime, stime, %cpu, and rss. */ ttyprintf(tp, " cmd: %s %d [%s%s] %ld.%02ldr %ld.%02ldu %ld.%02lds %d%% %ldk\n", comm, pid, stateprefix, state, (long)rtime.tv_sec, rtime.tv_usec / 10000, (long)utime.tv_sec, utime.tv_usec / 10000, (long)stime.tv_sec, stime.tv_usec / 10000, pctcpu / 100, rss); } Index: head/sys/kern/uipc_domain.c =================================================================== --- head/sys/kern/uipc_domain.c (revision 305831) +++ head/sys/kern/uipc_domain.c (revision 305832) @@ -1,521 +1,521 @@ /*- * Copyright (c) 1982, 1986, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)uipc_domain.c 8.2 (Berkeley) 10/18/93 */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include /* * System initialization * * Note: domain initialization takes place on a per domain basis * as a result of traversing a SYSINIT linker set. Most likely, * each domain would want to call DOMAIN_SET(9) itself, which * would cause the domain to be added just after domaininit() * is called during startup. * * See DOMAIN_SET(9) for details on its use. */ static void domaininit(void *); SYSINIT(domain, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, domaininit, NULL); static void domainfinalize(void *); SYSINIT(domainfin, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST, domainfinalize, NULL); static struct callout pffast_callout; static struct callout pfslow_callout; static void pffasttimo(void *); static void pfslowtimo(void *); struct domain *domains; /* registered protocol domains */ int domain_init_status = 0; static struct mtx dom_mtx; /* domain list lock */ MTX_SYSINIT(domain, &dom_mtx, "domain list", MTX_DEF); /* * Dummy protocol specific user requests function pointer array. * All functions return EOPNOTSUPP. */ struct pr_usrreqs nousrreqs = { .pru_accept = pru_accept_notsupp, .pru_attach = pru_attach_notsupp, .pru_bind = pru_bind_notsupp, .pru_connect = pru_connect_notsupp, .pru_connect2 = pru_connect2_notsupp, .pru_control = pru_control_notsupp, .pru_disconnect = pru_disconnect_notsupp, .pru_listen = pru_listen_notsupp, .pru_peeraddr = pru_peeraddr_notsupp, .pru_rcvd = pru_rcvd_notsupp, .pru_rcvoob = pru_rcvoob_notsupp, .pru_send = pru_send_notsupp, .pru_sense = pru_sense_null, .pru_shutdown = pru_shutdown_notsupp, .pru_sockaddr = pru_sockaddr_notsupp, .pru_sosend = pru_sosend_notsupp, .pru_soreceive = pru_soreceive_notsupp, .pru_sopoll = pru_sopoll_notsupp, }; static void protosw_init(struct protosw *pr) { struct pr_usrreqs *pu; pu = pr->pr_usrreqs; KASSERT(pu != NULL, ("protosw_init: %ssw[%d] has no usrreqs!", pr->pr_domain->dom_name, (int)(pr - pr->pr_domain->dom_protosw))); /* * Protocol switch methods fall into three categories: mandatory, * mandatory but protosw_init() provides a default, and optional. * * For true protocols (i.e., pru_attach != NULL), KASSERT truly * mandatory methods with no defaults, and initialize defaults for * other mandatory methods if the protocol hasn't defined an * implementation (NULL function pointer). */ #if 0 if (pu->pru_attach != NULL) { KASSERT(pu->pru_abort != NULL, ("protosw_init: %ssw[%d] pru_abort NULL", pr->pr_domain->dom_name, (int)(pr - pr->pr_domain->dom_protosw))); KASSERT(pu->pru_send != NULL, ("protosw_init: %ssw[%d] pru_send NULL", pr->pr_domain->dom_name, (int)(pr - pr->pr_domain->dom_protosw))); } #endif #define DEFAULT(foo, bar) if ((foo) == NULL) (foo) = (bar) DEFAULT(pu->pru_accept, pru_accept_notsupp); DEFAULT(pu->pru_aio_queue, pru_aio_queue_notsupp); DEFAULT(pu->pru_bind, pru_bind_notsupp); DEFAULT(pu->pru_bindat, pru_bindat_notsupp); DEFAULT(pu->pru_connect, pru_connect_notsupp); DEFAULT(pu->pru_connect2, pru_connect2_notsupp); DEFAULT(pu->pru_connectat, pru_connectat_notsupp); DEFAULT(pu->pru_control, pru_control_notsupp); DEFAULT(pu->pru_disconnect, pru_disconnect_notsupp); DEFAULT(pu->pru_listen, pru_listen_notsupp); DEFAULT(pu->pru_peeraddr, pru_peeraddr_notsupp); DEFAULT(pu->pru_rcvd, pru_rcvd_notsupp); DEFAULT(pu->pru_rcvoob, pru_rcvoob_notsupp); DEFAULT(pu->pru_sense, pru_sense_null); DEFAULT(pu->pru_shutdown, pru_shutdown_notsupp); DEFAULT(pu->pru_sockaddr, pru_sockaddr_notsupp); DEFAULT(pu->pru_sosend, sosend_generic); DEFAULT(pu->pru_soreceive, soreceive_generic); DEFAULT(pu->pru_sopoll, sopoll_generic); DEFAULT(pu->pru_ready, pru_ready_notsupp); #undef DEFAULT if (pr->pr_init) (*pr->pr_init)(); } /* * Add a new protocol domain to the list of supported domains * Note: you cant unload it again because a socket may be using it. * XXX can't fail at this time. */ void domain_init(void *arg) { struct domain *dp = arg; struct protosw *pr; if (dp->dom_init) (*dp->dom_init)(); for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) protosw_init(pr); /* * update global information about maximums */ max_hdr = max_linkhdr + max_protohdr; max_datalen = MHLEN - max_hdr; if (max_datalen < 1) panic("%s: max_datalen < 1", __func__); } #ifdef VIMAGE void vnet_domain_init(void *arg) { /* Virtualized case is no different -- call init functions. */ domain_init(arg); } void vnet_domain_uninit(void *arg) { struct domain *dp = arg; if (dp->dom_destroy) (*dp->dom_destroy)(); } #endif /* * Add a new protocol domain to the list of supported domains * Note: you cant unload it again because a socket may be using it. * XXX can't fail at this time. */ void domain_add(void *data) { struct domain *dp; dp = (struct domain *)data; mtx_lock(&dom_mtx); dp->dom_next = domains; domains = dp; KASSERT(domain_init_status >= 1, ("attempt to domain_add(%s) before domaininit()", dp->dom_name)); #ifndef INVARIANTS if (domain_init_status < 1) printf("WARNING: attempt to domain_add(%s) before " "domaininit()\n", dp->dom_name); #endif #ifdef notyet KASSERT(domain_init_status < 2, ("attempt to domain_add(%s) after domainfinalize()", dp->dom_name)); #else if (domain_init_status >= 2) printf("WARNING: attempt to domain_add(%s) after " "domainfinalize()\n", dp->dom_name); #endif mtx_unlock(&dom_mtx); } /* ARGSUSED*/ static void domaininit(void *dummy) { if (max_linkhdr < 16) /* XXX */ max_linkhdr = 16; callout_init(&pffast_callout, 1); callout_init(&pfslow_callout, 1); mtx_lock(&dom_mtx); KASSERT(domain_init_status == 0, ("domaininit called too late!")); domain_init_status = 1; mtx_unlock(&dom_mtx); } /* ARGSUSED*/ static void domainfinalize(void *dummy) { mtx_lock(&dom_mtx); KASSERT(domain_init_status == 1, ("domainfinalize called too late!")); domain_init_status = 2; mtx_unlock(&dom_mtx); callout_reset(&pffast_callout, 1, pffasttimo, NULL); callout_reset(&pfslow_callout, 1, pfslowtimo, NULL); } struct domain * pffinddomain(int family) { struct domain *dp; for (dp = domains; dp != NULL; dp = dp->dom_next) if (dp->dom_family == family) return (dp); return (NULL); } struct protosw * pffindtype(int family, int type) { struct domain *dp; struct protosw *pr; dp = pffinddomain(family); if (dp == NULL) return (NULL); for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) if (pr->pr_type && pr->pr_type == type) return (pr); return (NULL); } struct protosw * pffindproto(int family, int protocol, int type) { struct domain *dp; struct protosw *pr; struct protosw *maybe; maybe = NULL; if (family == 0) return (NULL); dp = pffinddomain(family); if (dp == NULL) return (NULL); for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) { if ((pr->pr_protocol == protocol) && (pr->pr_type == type)) return (pr); if (type == SOCK_RAW && pr->pr_type == SOCK_RAW && pr->pr_protocol == 0 && maybe == NULL) maybe = pr; } return (maybe); } /* * The caller must make sure that the new protocol is fully set up and ready to * accept requests before it is registered. */ int pf_proto_register(int family, struct protosw *npr) { VNET_ITERATOR_DECL(vnet_iter); struct domain *dp; struct protosw *pr, *fpr; /* Sanity checks. */ if (family == 0) return (EPFNOSUPPORT); if (npr->pr_type == 0) return (EPROTOTYPE); if (npr->pr_protocol == 0) return (EPROTONOSUPPORT); if (npr->pr_usrreqs == NULL) return (ENXIO); /* Try to find the specified domain based on the family. */ dp = pffinddomain(family); if (dp == NULL) return (EPFNOSUPPORT); /* Initialize backpointer to struct domain. */ npr->pr_domain = dp; fpr = NULL; /* * Protect us against races when two protocol registrations for * the same protocol happen at the same time. */ mtx_lock(&dom_mtx); /* The new protocol must not yet exist. */ for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) { if ((pr->pr_type == npr->pr_type) && (pr->pr_protocol == npr->pr_protocol)) { mtx_unlock(&dom_mtx); return (EEXIST); /* XXX: Check only protocol? */ } /* While here, remember the first free spacer. */ if ((fpr == NULL) && (pr->pr_protocol == PROTO_SPACER)) fpr = pr; } /* If no free spacer is found we can't add the new protocol. */ if (fpr == NULL) { mtx_unlock(&dom_mtx); return (ENOMEM); } /* Copy the new struct protosw over the spacer. */ bcopy(npr, fpr, sizeof(*fpr)); /* Job is done, no more protection required. */ mtx_unlock(&dom_mtx); /* Initialize and activate the protocol. */ VNET_LIST_RLOCK(); VNET_FOREACH(vnet_iter) { CURVNET_SET_QUIET(vnet_iter); protosw_init(fpr); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK(); return (0); } /* * The caller must make sure the protocol and its functions correctly shut down * all sockets and release all locks and memory references. */ int pf_proto_unregister(int family, int protocol, int type) { struct domain *dp; struct protosw *pr, *dpr; /* Sanity checks. */ if (family == 0) return (EPFNOSUPPORT); if (protocol == 0) return (EPROTONOSUPPORT); if (type == 0) return (EPROTOTYPE); /* Try to find the specified domain based on the family type. */ dp = pffinddomain(family); if (dp == NULL) return (EPFNOSUPPORT); dpr = NULL; /* Lock out everyone else while we are manipulating the protosw. */ mtx_lock(&dom_mtx); /* The protocol must exist and only once. */ for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) { if ((pr->pr_type == type) && (pr->pr_protocol == protocol)) { if (dpr != NULL) { mtx_unlock(&dom_mtx); return (EMLINK); /* Should not happen! */ } else dpr = pr; } } /* Protocol does not exist. */ if (dpr == NULL) { mtx_unlock(&dom_mtx); return (EPROTONOSUPPORT); } /* De-orbit the protocol and make the slot available again. */ dpr->pr_type = 0; dpr->pr_domain = dp; dpr->pr_protocol = PROTO_SPACER; dpr->pr_flags = 0; dpr->pr_input = NULL; dpr->pr_output = NULL; dpr->pr_ctlinput = NULL; dpr->pr_ctloutput = NULL; dpr->pr_init = NULL; dpr->pr_fasttimo = NULL; dpr->pr_slowtimo = NULL; dpr->pr_drain = NULL; dpr->pr_usrreqs = &nousrreqs; /* Job is done, not more protection required. */ mtx_unlock(&dom_mtx); return (0); } void pfctlinput(int cmd, struct sockaddr *sa) { struct domain *dp; struct protosw *pr; for (dp = domains; dp; dp = dp->dom_next) for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) if (pr->pr_ctlinput) (*pr->pr_ctlinput)(cmd, sa, (void *)0); } void pfctlinput2(int cmd, struct sockaddr *sa, void *ctlparam) { struct domain *dp; struct protosw *pr; if (!sa) return; for (dp = domains; dp; dp = dp->dom_next) { /* * the check must be made by xx_ctlinput() anyways, to * make sure we use data item pointed to by ctlparam in * correct way. the following check is made just for safety. */ if (dp->dom_family != sa->sa_family) continue; for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) if (pr->pr_ctlinput) (*pr->pr_ctlinput)(cmd, sa, ctlparam); } } static void pfslowtimo(void *arg) { struct domain *dp; struct protosw *pr; for (dp = domains; dp; dp = dp->dom_next) for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) if (pr->pr_slowtimo) (*pr->pr_slowtimo)(); callout_reset(&pfslow_callout, hz/2, pfslowtimo, NULL); } static void pffasttimo(void *arg) { struct domain *dp; struct protosw *pr; for (dp = domains; dp; dp = dp->dom_next) for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) if (pr->pr_fasttimo) (*pr->pr_fasttimo)(); callout_reset(&pffast_callout, hz/5, pffasttimo, NULL); } Index: head/sys/kern/uipc_mbuf.c =================================================================== --- head/sys/kern/uipc_mbuf.c (revision 305831) +++ head/sys/kern/uipc_mbuf.c (revision 305832) @@ -1,1869 +1,1869 @@ /*- * Copyright (c) 1982, 1986, 1988, 1991, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_param.h" #include "opt_mbuf_stress_test.h" #include "opt_mbuf_profiling.h" #include #include #include #include #include #include #include #include #include #include #include #include SDT_PROBE_DEFINE5_XLATE(sdt, , , m__init, "struct mbuf *", "mbufinfo_t *", "uint32_t", "uint32_t", "uint16_t", "uint16_t", "uint32_t", "uint32_t", "uint32_t", "uint32_t"); SDT_PROBE_DEFINE3_XLATE(sdt, , , m__gethdr, "uint32_t", "uint32_t", "uint16_t", "uint16_t", "struct mbuf *", "mbufinfo_t *"); SDT_PROBE_DEFINE3_XLATE(sdt, , , m__get, "uint32_t", "uint32_t", "uint16_t", "uint16_t", "struct mbuf *", "mbufinfo_t *"); SDT_PROBE_DEFINE4_XLATE(sdt, , , m__getcl, "uint32_t", "uint32_t", "uint16_t", "uint16_t", "uint32_t", "uint32_t", "struct mbuf *", "mbufinfo_t *"); SDT_PROBE_DEFINE3_XLATE(sdt, , , m__clget, "struct mbuf *", "mbufinfo_t *", "uint32_t", "uint32_t", "uint32_t", "uint32_t"); SDT_PROBE_DEFINE4_XLATE(sdt, , , m__cljget, "struct mbuf *", "mbufinfo_t *", "uint32_t", "uint32_t", "uint32_t", "uint32_t", "void*", "void*"); SDT_PROBE_DEFINE(sdt, , , m__cljset); SDT_PROBE_DEFINE1_XLATE(sdt, , , m__free, "struct mbuf *", "mbufinfo_t *"); SDT_PROBE_DEFINE1_XLATE(sdt, , , m__freem, "struct mbuf *", "mbufinfo_t *"); #include int max_linkhdr; int max_protohdr; int max_hdr; int max_datalen; #ifdef MBUF_STRESS_TEST int m_defragpackets; int m_defragbytes; int m_defraguseless; int m_defragfailure; int m_defragrandomfailures; #endif /* * sysctl(8) exported objects */ SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RD, &max_linkhdr, 0, "Size of largest link layer header"); SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RD, &max_protohdr, 0, "Size of largest protocol layer header"); SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RD, &max_hdr, 0, "Size of largest link plus protocol header"); SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RD, &max_datalen, 0, "Minimum space left in mbuf after max_hdr"); #ifdef MBUF_STRESS_TEST SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD, &m_defragpackets, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD, &m_defragbytes, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD, &m_defraguseless, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD, &m_defragfailure, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW, &m_defragrandomfailures, 0, ""); #endif /* * Ensure the correct size of various mbuf parameters. It could be off due * to compiler-induced padding and alignment artifacts. */ CTASSERT(MSIZE - offsetof(struct mbuf, m_dat) == MLEN); CTASSERT(MSIZE - offsetof(struct mbuf, m_pktdat) == MHLEN); /* * mbuf data storage should be 64-bit aligned regardless of architectural * pointer size; check this is the case with and without a packet header. */ CTASSERT(offsetof(struct mbuf, m_dat) % 8 == 0); CTASSERT(offsetof(struct mbuf, m_pktdat) % 8 == 0); /* * While the specific values here don't matter too much (i.e., +/- a few * words), we do want to ensure that changes to these values are carefully * reasoned about and properly documented. This is especially the case as * network-protocol and device-driver modules encode these layouts, and must * be recompiled if the structures change. Check these values at compile time * against the ones documented in comments in mbuf.h. * * NB: Possibly they should be documented there via #define's and not just * comments. */ #if defined(__LP64__) CTASSERT(offsetof(struct mbuf, m_dat) == 32); CTASSERT(sizeof(struct pkthdr) == 56); CTASSERT(sizeof(struct m_ext) == 48); #else CTASSERT(offsetof(struct mbuf, m_dat) == 24); CTASSERT(sizeof(struct pkthdr) == 48); CTASSERT(sizeof(struct m_ext) == 28); #endif /* * Assert that the queue(3) macros produce code of the same size as an old * plain pointer does. */ #ifdef INVARIANTS static struct mbuf m_assertbuf; CTASSERT(sizeof(m_assertbuf.m_slist) == sizeof(m_assertbuf.m_next)); CTASSERT(sizeof(m_assertbuf.m_stailq) == sizeof(m_assertbuf.m_next)); CTASSERT(sizeof(m_assertbuf.m_slistpkt) == sizeof(m_assertbuf.m_nextpkt)); CTASSERT(sizeof(m_assertbuf.m_stailqpkt) == sizeof(m_assertbuf.m_nextpkt)); #endif /* * Attach the cluster from *m to *n, set up m_ext in *n * and bump the refcount of the cluster. */ void mb_dupcl(struct mbuf *n, struct mbuf *m) { volatile u_int *refcnt; KASSERT(m->m_flags & M_EXT, ("%s: M_EXT not set on %p", __func__, m)); KASSERT(!(n->m_flags & M_EXT), ("%s: M_EXT set on %p", __func__, n)); n->m_ext = m->m_ext; n->m_flags |= M_EXT; n->m_flags |= m->m_flags & M_RDONLY; /* See if this is the mbuf that holds the embedded refcount. */ if (m->m_ext.ext_flags & EXT_FLAG_EMBREF) { refcnt = n->m_ext.ext_cnt = &m->m_ext.ext_count; n->m_ext.ext_flags &= ~EXT_FLAG_EMBREF; } else { KASSERT(m->m_ext.ext_cnt != NULL, ("%s: no refcounting pointer on %p", __func__, m)); refcnt = m->m_ext.ext_cnt; } if (*refcnt == 1) *refcnt += 1; else atomic_add_int(refcnt, 1); } void m_demote_pkthdr(struct mbuf *m) { M_ASSERTPKTHDR(m); m_tag_delete_chain(m, NULL); m->m_flags &= ~M_PKTHDR; bzero(&m->m_pkthdr, sizeof(struct pkthdr)); } /* * Clean up mbuf (chain) from any tags and packet headers. * If "all" is set then the first mbuf in the chain will be * cleaned too. */ void m_demote(struct mbuf *m0, int all, int flags) { struct mbuf *m; for (m = all ? m0 : m0->m_next; m != NULL; m = m->m_next) { KASSERT(m->m_nextpkt == NULL, ("%s: m_nextpkt in m %p, m0 %p", __func__, m, m0)); if (m->m_flags & M_PKTHDR) m_demote_pkthdr(m); m->m_flags = m->m_flags & (M_EXT | M_RDONLY | M_NOFREE | flags); } } /* * Sanity checks on mbuf (chain) for use in KASSERT() and general * debugging. * Returns 0 or panics when bad and 1 on all tests passed. * Sanitize, 0 to run M_SANITY_ACTION, 1 to garble things so they * blow up later. */ int m_sanity(struct mbuf *m0, int sanitize) { struct mbuf *m; caddr_t a, b; int pktlen = 0; #ifdef INVARIANTS #define M_SANITY_ACTION(s) panic("mbuf %p: " s, m) #else #define M_SANITY_ACTION(s) printf("mbuf %p: " s, m) #endif for (m = m0; m != NULL; m = m->m_next) { /* * Basic pointer checks. If any of these fails then some * unrelated kernel memory before or after us is trashed. * No way to recover from that. */ a = M_START(m); b = a + M_SIZE(m); if ((caddr_t)m->m_data < a) M_SANITY_ACTION("m_data outside mbuf data range left"); if ((caddr_t)m->m_data > b) M_SANITY_ACTION("m_data outside mbuf data range right"); if ((caddr_t)m->m_data + m->m_len > b) M_SANITY_ACTION("m_data + m_len exeeds mbuf space"); /* m->m_nextpkt may only be set on first mbuf in chain. */ if (m != m0 && m->m_nextpkt != NULL) { if (sanitize) { m_freem(m->m_nextpkt); m->m_nextpkt = (struct mbuf *)0xDEADC0DE; } else M_SANITY_ACTION("m->m_nextpkt on in-chain mbuf"); } /* packet length (not mbuf length!) calculation */ if (m0->m_flags & M_PKTHDR) pktlen += m->m_len; /* m_tags may only be attached to first mbuf in chain. */ if (m != m0 && m->m_flags & M_PKTHDR && !SLIST_EMPTY(&m->m_pkthdr.tags)) { if (sanitize) { m_tag_delete_chain(m, NULL); /* put in 0xDEADC0DE perhaps? */ } else M_SANITY_ACTION("m_tags on in-chain mbuf"); } /* M_PKTHDR may only be set on first mbuf in chain */ if (m != m0 && m->m_flags & M_PKTHDR) { if (sanitize) { bzero(&m->m_pkthdr, sizeof(m->m_pkthdr)); m->m_flags &= ~M_PKTHDR; /* put in 0xDEADCODE and leave hdr flag in */ } else M_SANITY_ACTION("M_PKTHDR on in-chain mbuf"); } } m = m0; if (pktlen && pktlen != m->m_pkthdr.len) { if (sanitize) m->m_pkthdr.len = 0; else M_SANITY_ACTION("m_pkthdr.len != mbuf chain length"); } return 1; #undef M_SANITY_ACTION } /* * Non-inlined part of m_init(). */ int m_pkthdr_init(struct mbuf *m, int how) { #ifdef MAC int error; #endif m->m_data = m->m_pktdat; bzero(&m->m_pkthdr, sizeof(m->m_pkthdr)); #ifdef MAC /* If the label init fails, fail the alloc */ error = mac_mbuf_init(m, how); if (error) return (error); #endif return (0); } /* * "Move" mbuf pkthdr from "from" to "to". * "from" must have M_PKTHDR set, and "to" must be empty. */ void m_move_pkthdr(struct mbuf *to, struct mbuf *from) { #if 0 /* see below for why these are not enabled */ M_ASSERTPKTHDR(to); /* Note: with MAC, this may not be a good assertion. */ KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_move_pkthdr: to has tags")); #endif #ifdef MAC /* * XXXMAC: It could be this should also occur for non-MAC? */ if (to->m_flags & M_PKTHDR) m_tag_delete_chain(to, NULL); #endif to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT); if ((to->m_flags & M_EXT) == 0) to->m_data = to->m_pktdat; to->m_pkthdr = from->m_pkthdr; /* especially tags */ SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */ from->m_flags &= ~M_PKTHDR; } /* * Duplicate "from"'s mbuf pkthdr in "to". * "from" must have M_PKTHDR set, and "to" must be empty. * In particular, this does a deep copy of the packet tags. */ int m_dup_pkthdr(struct mbuf *to, const struct mbuf *from, int how) { #if 0 /* * The mbuf allocator only initializes the pkthdr * when the mbuf is allocated with m_gethdr(). Many users * (e.g. m_copy*, m_prepend) use m_get() and then * smash the pkthdr as needed causing these * assertions to trip. For now just disable them. */ M_ASSERTPKTHDR(to); /* Note: with MAC, this may not be a good assertion. */ KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags")); #endif MBUF_CHECKSLEEP(how); #ifdef MAC if (to->m_flags & M_PKTHDR) m_tag_delete_chain(to, NULL); #endif to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT); if ((to->m_flags & M_EXT) == 0) to->m_data = to->m_pktdat; to->m_pkthdr = from->m_pkthdr; SLIST_INIT(&to->m_pkthdr.tags); return (m_tag_copy_chain(to, from, how)); } /* * Lesser-used path for M_PREPEND: * allocate new mbuf to prepend to chain, * copy junk along. */ struct mbuf * m_prepend(struct mbuf *m, int len, int how) { struct mbuf *mn; if (m->m_flags & M_PKTHDR) mn = m_gethdr(how, m->m_type); else mn = m_get(how, m->m_type); if (mn == NULL) { m_freem(m); return (NULL); } if (m->m_flags & M_PKTHDR) m_move_pkthdr(mn, m); mn->m_next = m; m = mn; if (len < M_SIZE(m)) M_ALIGN(m, len); m->m_len = len; return (m); } /* * Make a copy of an mbuf chain starting "off0" bytes from the beginning, * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf. * The wait parameter is a choice of M_WAITOK/M_NOWAIT from caller. * Note that the copy is read-only, because clusters are not copied, * only their reference counts are incremented. */ struct mbuf * m_copym(struct mbuf *m, int off0, int len, int wait) { struct mbuf *n, **np; int off = off0; struct mbuf *top; int copyhdr = 0; KASSERT(off >= 0, ("m_copym, negative off %d", off)); KASSERT(len >= 0, ("m_copym, negative len %d", len)); MBUF_CHECKSLEEP(wait); if (off == 0 && m->m_flags & M_PKTHDR) copyhdr = 1; while (off > 0) { KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain")); if (off < m->m_len) break; off -= m->m_len; m = m->m_next; } np = ⊤ top = NULL; while (len > 0) { if (m == NULL) { KASSERT(len == M_COPYALL, ("m_copym, length > size of mbuf chain")); break; } if (copyhdr) n = m_gethdr(wait, m->m_type); else n = m_get(wait, m->m_type); *np = n; if (n == NULL) goto nospace; if (copyhdr) { if (!m_dup_pkthdr(n, m, wait)) goto nospace; if (len == M_COPYALL) n->m_pkthdr.len -= off0; else n->m_pkthdr.len = len; copyhdr = 0; } n->m_len = min(len, m->m_len - off); if (m->m_flags & M_EXT) { n->m_data = m->m_data + off; mb_dupcl(n, m); } else bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t), (u_int)n->m_len); if (len != M_COPYALL) len -= n->m_len; off = 0; m = m->m_next; np = &n->m_next; } return (top); nospace: m_freem(top); return (NULL); } /* * Copy an entire packet, including header (which must be present). * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'. * Note that the copy is read-only, because clusters are not copied, * only their reference counts are incremented. * Preserve alignment of the first mbuf so if the creator has left * some room at the beginning (e.g. for inserting protocol headers) * the copies still have the room available. */ struct mbuf * m_copypacket(struct mbuf *m, int how) { struct mbuf *top, *n, *o; MBUF_CHECKSLEEP(how); n = m_get(how, m->m_type); top = n; if (n == NULL) goto nospace; if (!m_dup_pkthdr(n, m, how)) goto nospace; n->m_len = m->m_len; if (m->m_flags & M_EXT) { n->m_data = m->m_data; mb_dupcl(n, m); } else { n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat ); bcopy(mtod(m, char *), mtod(n, char *), n->m_len); } m = m->m_next; while (m) { o = m_get(how, m->m_type); if (o == NULL) goto nospace; n->m_next = o; n = n->m_next; n->m_len = m->m_len; if (m->m_flags & M_EXT) { n->m_data = m->m_data; mb_dupcl(n, m); } else { bcopy(mtod(m, char *), mtod(n, char *), n->m_len); } m = m->m_next; } return top; nospace: m_freem(top); return (NULL); } /* * Copy data from an mbuf chain starting "off" bytes from the beginning, * continuing for "len" bytes, into the indicated buffer. */ void m_copydata(const struct mbuf *m, int off, int len, caddr_t cp) { u_int count; KASSERT(off >= 0, ("m_copydata, negative off %d", off)); KASSERT(len >= 0, ("m_copydata, negative len %d", len)); while (off > 0) { KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain")); if (off < m->m_len) break; off -= m->m_len; m = m->m_next; } while (len > 0) { KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain")); count = min(m->m_len - off, len); bcopy(mtod(m, caddr_t) + off, cp, count); len -= count; cp += count; off = 0; m = m->m_next; } } /* * Copy a packet header mbuf chain into a completely new chain, including * copying any mbuf clusters. Use this instead of m_copypacket() when * you need a writable copy of an mbuf chain. */ struct mbuf * m_dup(const struct mbuf *m, int how) { struct mbuf **p, *top = NULL; int remain, moff, nsize; MBUF_CHECKSLEEP(how); /* Sanity check */ if (m == NULL) return (NULL); M_ASSERTPKTHDR(m); /* While there's more data, get a new mbuf, tack it on, and fill it */ remain = m->m_pkthdr.len; moff = 0; p = ⊤ while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */ struct mbuf *n; /* Get the next new mbuf */ if (remain >= MINCLSIZE) { n = m_getcl(how, m->m_type, 0); nsize = MCLBYTES; } else { n = m_get(how, m->m_type); nsize = MLEN; } if (n == NULL) goto nospace; if (top == NULL) { /* First one, must be PKTHDR */ if (!m_dup_pkthdr(n, m, how)) { m_free(n); goto nospace; } if ((n->m_flags & M_EXT) == 0) nsize = MHLEN; n->m_flags &= ~M_RDONLY; } n->m_len = 0; /* Link it into the new chain */ *p = n; p = &n->m_next; /* Copy data from original mbuf(s) into new mbuf */ while (n->m_len < nsize && m != NULL) { int chunk = min(nsize - n->m_len, m->m_len - moff); bcopy(m->m_data + moff, n->m_data + n->m_len, chunk); moff += chunk; n->m_len += chunk; remain -= chunk; if (moff == m->m_len) { m = m->m_next; moff = 0; } } /* Check correct total mbuf length */ KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL), ("%s: bogus m_pkthdr.len", __func__)); } return (top); nospace: m_freem(top); return (NULL); } /* * Concatenate mbuf chain n to m. * Both chains must be of the same type (e.g. MT_DATA). * Any m_pkthdr is not updated. */ void m_cat(struct mbuf *m, struct mbuf *n) { while (m->m_next) m = m->m_next; while (n) { if (!M_WRITABLE(m) || M_TRAILINGSPACE(m) < n->m_len) { /* just join the two chains */ m->m_next = n; return; } /* splat the data from one into the other */ bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, (u_int)n->m_len); m->m_len += n->m_len; n = m_free(n); } } /* * Concatenate two pkthdr mbuf chains. */ void m_catpkt(struct mbuf *m, struct mbuf *n) { M_ASSERTPKTHDR(m); M_ASSERTPKTHDR(n); m->m_pkthdr.len += n->m_pkthdr.len; m_demote(n, 1, 0); m_cat(m, n); } void m_adj(struct mbuf *mp, int req_len) { int len = req_len; struct mbuf *m; int count; if ((m = mp) == NULL) return; if (len >= 0) { /* * Trim from head. */ while (m != NULL && len > 0) { if (m->m_len <= len) { len -= m->m_len; m->m_len = 0; m = m->m_next; } else { m->m_len -= len; m->m_data += len; len = 0; } } if (mp->m_flags & M_PKTHDR) mp->m_pkthdr.len -= (req_len - len); } else { /* * Trim from tail. Scan the mbuf chain, * calculating its length and finding the last mbuf. * If the adjustment only affects this mbuf, then just * adjust and return. Otherwise, rescan and truncate * after the remaining size. */ len = -len; count = 0; for (;;) { count += m->m_len; if (m->m_next == (struct mbuf *)0) break; m = m->m_next; } if (m->m_len >= len) { m->m_len -= len; if (mp->m_flags & M_PKTHDR) mp->m_pkthdr.len -= len; return; } count -= len; if (count < 0) count = 0; /* * Correct length for chain is "count". * Find the mbuf with last data, adjust its length, * and toss data from remaining mbufs on chain. */ m = mp; if (m->m_flags & M_PKTHDR) m->m_pkthdr.len = count; for (; m; m = m->m_next) { if (m->m_len >= count) { m->m_len = count; if (m->m_next != NULL) { m_freem(m->m_next); m->m_next = NULL; } break; } count -= m->m_len; } } } /* * Rearange an mbuf chain so that len bytes are contiguous * and in the data area of an mbuf (so that mtod will work * for a structure of size len). Returns the resulting * mbuf chain on success, frees it and returns null on failure. * If there is room, it will add up to max_protohdr-len extra bytes to the * contiguous region in an attempt to avoid being called next time. */ struct mbuf * m_pullup(struct mbuf *n, int len) { struct mbuf *m; int count; int space; /* * If first mbuf has no cluster, and has room for len bytes * without shifting current data, pullup into it, * otherwise allocate a new mbuf to prepend to the chain. */ if ((n->m_flags & M_EXT) == 0 && n->m_data + len < &n->m_dat[MLEN] && n->m_next) { if (n->m_len >= len) return (n); m = n; n = n->m_next; len -= m->m_len; } else { if (len > MHLEN) goto bad; m = m_get(M_NOWAIT, n->m_type); if (m == NULL) goto bad; if (n->m_flags & M_PKTHDR) m_move_pkthdr(m, n); } space = &m->m_dat[MLEN] - (m->m_data + m->m_len); do { count = min(min(max(len, max_protohdr), space), n->m_len); bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, (u_int)count); len -= count; m->m_len += count; n->m_len -= count; space -= count; if (n->m_len) n->m_data += count; else n = m_free(n); } while (len > 0 && n); if (len > 0) { (void) m_free(m); goto bad; } m->m_next = n; return (m); bad: m_freem(n); return (NULL); } /* * Like m_pullup(), except a new mbuf is always allocated, and we allow * the amount of empty space before the data in the new mbuf to be specified * (in the event that the caller expects to prepend later). */ struct mbuf * m_copyup(struct mbuf *n, int len, int dstoff) { struct mbuf *m; int count, space; if (len > (MHLEN - dstoff)) goto bad; m = m_get(M_NOWAIT, n->m_type); if (m == NULL) goto bad; if (n->m_flags & M_PKTHDR) m_move_pkthdr(m, n); m->m_data += dstoff; space = &m->m_dat[MLEN] - (m->m_data + m->m_len); do { count = min(min(max(len, max_protohdr), space), n->m_len); memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t), (unsigned)count); len -= count; m->m_len += count; n->m_len -= count; space -= count; if (n->m_len) n->m_data += count; else n = m_free(n); } while (len > 0 && n); if (len > 0) { (void) m_free(m); goto bad; } m->m_next = n; return (m); bad: m_freem(n); return (NULL); } /* * Partition an mbuf chain in two pieces, returning the tail -- * all but the first len0 bytes. In case of failure, it returns NULL and * attempts to restore the chain to its original state. * * Note that the resulting mbufs might be read-only, because the new * mbuf can end up sharing an mbuf cluster with the original mbuf if * the "breaking point" happens to lie within a cluster mbuf. Use the * M_WRITABLE() macro to check for this case. */ struct mbuf * m_split(struct mbuf *m0, int len0, int wait) { struct mbuf *m, *n; u_int len = len0, remain; MBUF_CHECKSLEEP(wait); for (m = m0; m && len > m->m_len; m = m->m_next) len -= m->m_len; if (m == NULL) return (NULL); remain = m->m_len - len; if (m0->m_flags & M_PKTHDR && remain == 0) { n = m_gethdr(wait, m0->m_type); if (n == NULL) return (NULL); n->m_next = m->m_next; m->m_next = NULL; n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; n->m_pkthdr.len = m0->m_pkthdr.len - len0; m0->m_pkthdr.len = len0; return (n); } else if (m0->m_flags & M_PKTHDR) { n = m_gethdr(wait, m0->m_type); if (n == NULL) return (NULL); n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; n->m_pkthdr.len = m0->m_pkthdr.len - len0; m0->m_pkthdr.len = len0; if (m->m_flags & M_EXT) goto extpacket; if (remain > MHLEN) { /* m can't be the lead packet */ M_ALIGN(n, 0); n->m_next = m_split(m, len, wait); if (n->m_next == NULL) { (void) m_free(n); return (NULL); } else { n->m_len = 0; return (n); } } else M_ALIGN(n, remain); } else if (remain == 0) { n = m->m_next; m->m_next = NULL; return (n); } else { n = m_get(wait, m->m_type); if (n == NULL) return (NULL); M_ALIGN(n, remain); } extpacket: if (m->m_flags & M_EXT) { n->m_data = m->m_data + len; mb_dupcl(n, m); } else { bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain); } n->m_len = remain; m->m_len = len; n->m_next = m->m_next; m->m_next = NULL; return (n); } /* * Routine to copy from device local memory into mbufs. * Note that `off' argument is offset into first mbuf of target chain from * which to begin copying the data to. */ struct mbuf * m_devget(char *buf, int totlen, int off, struct ifnet *ifp, void (*copy)(char *from, caddr_t to, u_int len)) { struct mbuf *m; struct mbuf *top = NULL, **mp = ⊤ int len; if (off < 0 || off > MHLEN) return (NULL); while (totlen > 0) { if (top == NULL) { /* First one, must be PKTHDR */ if (totlen + off >= MINCLSIZE) { m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); len = MCLBYTES; } else { m = m_gethdr(M_NOWAIT, MT_DATA); len = MHLEN; /* Place initial small packet/header at end of mbuf */ if (m && totlen + off + max_linkhdr <= MHLEN) { m->m_data += max_linkhdr; len -= max_linkhdr; } } if (m == NULL) return NULL; m->m_pkthdr.rcvif = ifp; m->m_pkthdr.len = totlen; } else { if (totlen + off >= MINCLSIZE) { m = m_getcl(M_NOWAIT, MT_DATA, 0); len = MCLBYTES; } else { m = m_get(M_NOWAIT, MT_DATA); len = MLEN; } if (m == NULL) { m_freem(top); return NULL; } } if (off) { m->m_data += off; len -= off; off = 0; } m->m_len = len = min(totlen, len); if (copy) copy(buf, mtod(m, caddr_t), (u_int)len); else bcopy(buf, mtod(m, caddr_t), (u_int)len); buf += len; *mp = m; mp = &m->m_next; totlen -= len; } return (top); } /* * Copy data from a buffer back into the indicated mbuf chain, * starting "off" bytes from the beginning, extending the mbuf * chain if necessary. */ void m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp) { int mlen; struct mbuf *m = m0, *n; int totlen = 0; if (m0 == NULL) return; while (off > (mlen = m->m_len)) { off -= mlen; totlen += mlen; if (m->m_next == NULL) { n = m_get(M_NOWAIT, m->m_type); if (n == NULL) goto out; bzero(mtod(n, caddr_t), MLEN); n->m_len = min(MLEN, len + off); m->m_next = n; } m = m->m_next; } while (len > 0) { if (m->m_next == NULL && (len > m->m_len - off)) { m->m_len += min(len - (m->m_len - off), M_TRAILINGSPACE(m)); } mlen = min (m->m_len - off, len); bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen); cp += mlen; len -= mlen; mlen += off; off = 0; totlen += mlen; if (len == 0) break; if (m->m_next == NULL) { n = m_get(M_NOWAIT, m->m_type); if (n == NULL) break; n->m_len = min(MLEN, len); m->m_next = n; } m = m->m_next; } out: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) m->m_pkthdr.len = totlen; } /* * Append the specified data to the indicated mbuf chain, * Extend the mbuf chain if the new data does not fit in * existing space. * * Return 1 if able to complete the job; otherwise 0. */ int m_append(struct mbuf *m0, int len, c_caddr_t cp) { struct mbuf *m, *n; int remainder, space; for (m = m0; m->m_next != NULL; m = m->m_next) ; remainder = len; space = M_TRAILINGSPACE(m); if (space > 0) { /* * Copy into available space. */ if (space > remainder) space = remainder; bcopy(cp, mtod(m, caddr_t) + m->m_len, space); m->m_len += space; cp += space, remainder -= space; } while (remainder > 0) { /* * Allocate a new mbuf; could check space * and allocate a cluster instead. */ n = m_get(M_NOWAIT, m->m_type); if (n == NULL) break; n->m_len = min(MLEN, remainder); bcopy(cp, mtod(n, caddr_t), n->m_len); cp += n->m_len, remainder -= n->m_len; m->m_next = n; m = n; } if (m0->m_flags & M_PKTHDR) m0->m_pkthdr.len += len - remainder; return (remainder == 0); } /* * Apply function f to the data in an mbuf chain starting "off" bytes from * the beginning, continuing for "len" bytes. */ int m_apply(struct mbuf *m, int off, int len, int (*f)(void *, void *, u_int), void *arg) { u_int count; int rval; KASSERT(off >= 0, ("m_apply, negative off %d", off)); KASSERT(len >= 0, ("m_apply, negative len %d", len)); while (off > 0) { KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); if (off < m->m_len) break; off -= m->m_len; m = m->m_next; } while (len > 0) { KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); count = min(m->m_len - off, len); rval = (*f)(arg, mtod(m, caddr_t) + off, count); if (rval) return (rval); len -= count; off = 0; m = m->m_next; } return (0); } /* * Return a pointer to mbuf/offset of location in mbuf chain. */ struct mbuf * m_getptr(struct mbuf *m, int loc, int *off) { while (loc >= 0) { /* Normal end of search. */ if (m->m_len > loc) { *off = loc; return (m); } else { loc -= m->m_len; if (m->m_next == NULL) { if (loc == 0) { /* Point at the end of valid data. */ *off = m->m_len; return (m); } return (NULL); } m = m->m_next; } } return (NULL); } void m_print(const struct mbuf *m, int maxlen) { int len; int pdata; const struct mbuf *m2; if (m == NULL) { printf("mbuf: %p\n", m); return; } if (m->m_flags & M_PKTHDR) len = m->m_pkthdr.len; else len = -1; m2 = m; while (m2 != NULL && (len == -1 || len)) { pdata = m2->m_len; if (maxlen != -1 && pdata > maxlen) pdata = maxlen; printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len, m2->m_next, m2->m_flags, "\20\20freelist\17skipfw" "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly" "\3eor\2pkthdr\1ext", pdata ? "" : "\n"); if (pdata) printf(", %*D\n", pdata, (u_char *)m2->m_data, "-"); if (len != -1) len -= m2->m_len; m2 = m2->m_next; } if (len > 0) printf("%d bytes unaccounted for.\n", len); return; } u_int m_fixhdr(struct mbuf *m0) { u_int len; len = m_length(m0, NULL); m0->m_pkthdr.len = len; return (len); } u_int m_length(struct mbuf *m0, struct mbuf **last) { struct mbuf *m; u_int len; len = 0; for (m = m0; m != NULL; m = m->m_next) { len += m->m_len; if (m->m_next == NULL) break; } if (last != NULL) *last = m; return (len); } /* * Defragment a mbuf chain, returning the shortest possible * chain of mbufs and clusters. If allocation fails and * this cannot be completed, NULL will be returned, but * the passed in chain will be unchanged. Upon success, * the original chain will be freed, and the new chain * will be returned. * * If a non-packet header is passed in, the original * mbuf (chain?) will be returned unharmed. */ struct mbuf * m_defrag(struct mbuf *m0, int how) { struct mbuf *m_new = NULL, *m_final = NULL; int progress = 0, length; MBUF_CHECKSLEEP(how); if (!(m0->m_flags & M_PKTHDR)) return (m0); m_fixhdr(m0); /* Needed sanity check */ #ifdef MBUF_STRESS_TEST if (m_defragrandomfailures) { int temp = arc4random() & 0xff; if (temp == 0xba) goto nospace; } #endif if (m0->m_pkthdr.len > MHLEN) m_final = m_getcl(how, MT_DATA, M_PKTHDR); else m_final = m_gethdr(how, MT_DATA); if (m_final == NULL) goto nospace; if (m_dup_pkthdr(m_final, m0, how) == 0) goto nospace; m_new = m_final; while (progress < m0->m_pkthdr.len) { length = m0->m_pkthdr.len - progress; if (length > MCLBYTES) length = MCLBYTES; if (m_new == NULL) { if (length > MLEN) m_new = m_getcl(how, MT_DATA, 0); else m_new = m_get(how, MT_DATA); if (m_new == NULL) goto nospace; } m_copydata(m0, progress, length, mtod(m_new, caddr_t)); progress += length; m_new->m_len = length; if (m_new != m_final) m_cat(m_final, m_new); m_new = NULL; } #ifdef MBUF_STRESS_TEST if (m0->m_next == NULL) m_defraguseless++; #endif m_freem(m0); m0 = m_final; #ifdef MBUF_STRESS_TEST m_defragpackets++; m_defragbytes += m0->m_pkthdr.len; #endif return (m0); nospace: #ifdef MBUF_STRESS_TEST m_defragfailure++; #endif if (m_final) m_freem(m_final); return (NULL); } /* * Defragment an mbuf chain, returning at most maxfrags separate * mbufs+clusters. If this is not possible NULL is returned and * the original mbuf chain is left in it's present (potentially * modified) state. We use two techniques: collapsing consecutive * mbufs and replacing consecutive mbufs by a cluster. * * NB: this should really be named m_defrag but that name is taken */ struct mbuf * m_collapse(struct mbuf *m0, int how, int maxfrags) { struct mbuf *m, *n, *n2, **prev; u_int curfrags; /* * Calculate the current number of frags. */ curfrags = 0; for (m = m0; m != NULL; m = m->m_next) curfrags++; /* * First, try to collapse mbufs. Note that we always collapse * towards the front so we don't need to deal with moving the * pkthdr. This may be suboptimal if the first mbuf has much * less data than the following. */ m = m0; again: for (;;) { n = m->m_next; if (n == NULL) break; if (M_WRITABLE(m) && n->m_len < M_TRAILINGSPACE(m)) { bcopy(mtod(n, void *), mtod(m, char *) + m->m_len, n->m_len); m->m_len += n->m_len; m->m_next = n->m_next; m_free(n); if (--curfrags <= maxfrags) return m0; } else m = n; } KASSERT(maxfrags > 1, ("maxfrags %u, but normal collapse failed", maxfrags)); /* * Collapse consecutive mbufs to a cluster. */ prev = &m0->m_next; /* NB: not the first mbuf */ while ((n = *prev) != NULL) { if ((n2 = n->m_next) != NULL && n->m_len + n2->m_len < MCLBYTES) { m = m_getcl(how, MT_DATA, 0); if (m == NULL) goto bad; bcopy(mtod(n, void *), mtod(m, void *), n->m_len); bcopy(mtod(n2, void *), mtod(m, char *) + n->m_len, n2->m_len); m->m_len = n->m_len + n2->m_len; m->m_next = n2->m_next; *prev = m; m_free(n); m_free(n2); if (--curfrags <= maxfrags) /* +1 cl -2 mbufs */ return m0; /* * Still not there, try the normal collapse * again before we allocate another cluster. */ goto again; } prev = &n->m_next; } /* * No place where we can collapse to a cluster; punt. * This can occur if, for example, you request 2 frags * but the packet requires that both be clusters (we * never reallocate the first mbuf to avoid moving the * packet header). */ bad: return NULL; } #ifdef MBUF_STRESS_TEST /* * Fragment an mbuf chain. There's no reason you'd ever want to do * this in normal usage, but it's great for stress testing various * mbuf consumers. * * If fragmentation is not possible, the original chain will be * returned. * * Possible length values: * 0 no fragmentation will occur * > 0 each fragment will be of the specified length * -1 each fragment will be the same random value in length * -2 each fragment's length will be entirely random * (Random values range from 1 to 256) */ struct mbuf * m_fragment(struct mbuf *m0, int how, int length) { struct mbuf *m_new = NULL, *m_final = NULL; int progress = 0; if (!(m0->m_flags & M_PKTHDR)) return (m0); if ((length == 0) || (length < -2)) return (m0); m_fixhdr(m0); /* Needed sanity check */ m_final = m_getcl(how, MT_DATA, M_PKTHDR); if (m_final == NULL) goto nospace; if (m_dup_pkthdr(m_final, m0, how) == 0) goto nospace; m_new = m_final; if (length == -1) length = 1 + (arc4random() & 255); while (progress < m0->m_pkthdr.len) { int fraglen; if (length > 0) fraglen = length; else fraglen = 1 + (arc4random() & 255); if (fraglen > m0->m_pkthdr.len - progress) fraglen = m0->m_pkthdr.len - progress; if (fraglen > MCLBYTES) fraglen = MCLBYTES; if (m_new == NULL) { m_new = m_getcl(how, MT_DATA, 0); if (m_new == NULL) goto nospace; } m_copydata(m0, progress, fraglen, mtod(m_new, caddr_t)); progress += fraglen; m_new->m_len = fraglen; if (m_new != m_final) m_cat(m_final, m_new); m_new = NULL; } m_freem(m0); m0 = m_final; return (m0); nospace: if (m_final) m_freem(m_final); /* Return the original chain on failure */ return (m0); } #endif /* * Copy the contents of uio into a properly sized mbuf chain. */ struct mbuf * m_uiotombuf(struct uio *uio, int how, int len, int align, int flags) { struct mbuf *m, *mb; int error, length; ssize_t total; int progress = 0; /* * len can be zero or an arbitrary large value bound by * the total data supplied by the uio. */ if (len > 0) total = min(uio->uio_resid, len); else total = uio->uio_resid; /* * The smallest unit returned by m_getm2() is a single mbuf * with pkthdr. We can't align past it. */ if (align >= MHLEN) return (NULL); /* * Give us the full allocation or nothing. * If len is zero return the smallest empty mbuf. */ m = m_getm2(NULL, max(total + align, 1), how, MT_DATA, flags); if (m == NULL) return (NULL); m->m_data += align; /* Fill all mbufs with uio data and update header information. */ for (mb = m; mb != NULL; mb = mb->m_next) { length = min(M_TRAILINGSPACE(mb), total - progress); error = uiomove(mtod(mb, void *), length, uio); if (error) { m_freem(m); return (NULL); } mb->m_len = length; progress += length; if (flags & M_PKTHDR) m->m_pkthdr.len += length; } KASSERT(progress == total, ("%s: progress != total", __func__)); return (m); } /* * Copy an mbuf chain into a uio limited by len if set. */ int m_mbuftouio(struct uio *uio, struct mbuf *m, int len) { int error, length, total; int progress = 0; if (len > 0) total = min(uio->uio_resid, len); else total = uio->uio_resid; /* Fill the uio with data from the mbufs. */ for (; m != NULL; m = m->m_next) { length = min(m->m_len, total - progress); error = uiomove(mtod(m, void *), length, uio); if (error) return (error); progress += length; } return (0); } /* * Create a writable copy of the mbuf chain. While doing this * we compact the chain with a goal of producing a chain with * at most two mbufs. The second mbuf in this chain is likely * to be a cluster. The primary purpose of this work is to create * a writable packet for encryption, compression, etc. The * secondary goal is to linearize the data so the data can be * passed to crypto hardware in the most efficient manner possible. */ struct mbuf * m_unshare(struct mbuf *m0, int how) { struct mbuf *m, *mprev; struct mbuf *n, *mfirst, *mlast; int len, off; mprev = NULL; for (m = m0; m != NULL; m = mprev->m_next) { /* * Regular mbufs are ignored unless there's a cluster * in front of it that we can use to coalesce. We do * the latter mainly so later clusters can be coalesced * also w/o having to handle them specially (i.e. convert * mbuf+cluster -> cluster). This optimization is heavily * influenced by the assumption that we're running over * Ethernet where MCLBYTES is large enough that the max * packet size will permit lots of coalescing into a * single cluster. This in turn permits efficient * crypto operations, especially when using hardware. */ if ((m->m_flags & M_EXT) == 0) { if (mprev && (mprev->m_flags & M_EXT) && m->m_len <= M_TRAILINGSPACE(mprev)) { /* XXX: this ignores mbuf types */ memcpy(mtod(mprev, caddr_t) + mprev->m_len, mtod(m, caddr_t), m->m_len); mprev->m_len += m->m_len; mprev->m_next = m->m_next; /* unlink from chain */ m_free(m); /* reclaim mbuf */ #if 0 newipsecstat.ips_mbcoalesced++; #endif } else { mprev = m; } continue; } /* * Writable mbufs are left alone (for now). */ if (M_WRITABLE(m)) { mprev = m; continue; } /* * Not writable, replace with a copy or coalesce with * the previous mbuf if possible (since we have to copy * it anyway, we try to reduce the number of mbufs and * clusters so that future work is easier). */ KASSERT(m->m_flags & M_EXT, ("m_flags 0x%x", m->m_flags)); /* NB: we only coalesce into a cluster or larger */ if (mprev != NULL && (mprev->m_flags & M_EXT) && m->m_len <= M_TRAILINGSPACE(mprev)) { /* XXX: this ignores mbuf types */ memcpy(mtod(mprev, caddr_t) + mprev->m_len, mtod(m, caddr_t), m->m_len); mprev->m_len += m->m_len; mprev->m_next = m->m_next; /* unlink from chain */ m_free(m); /* reclaim mbuf */ #if 0 newipsecstat.ips_clcoalesced++; #endif continue; } /* * Allocate new space to hold the copy and copy the data. * We deal with jumbo mbufs (i.e. m_len > MCLBYTES) by * splitting them into clusters. We could just malloc a * buffer and make it external but too many device drivers * don't know how to break up the non-contiguous memory when * doing DMA. */ n = m_getcl(how, m->m_type, m->m_flags & M_COPYFLAGS); if (n == NULL) { m_freem(m0); return (NULL); } if (m->m_flags & M_PKTHDR) { KASSERT(mprev == NULL, ("%s: m0 %p, m %p has M_PKTHDR", __func__, m0, m)); m_move_pkthdr(n, m); } len = m->m_len; off = 0; mfirst = n; mlast = NULL; for (;;) { int cc = min(len, MCLBYTES); memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off, cc); n->m_len = cc; if (mlast != NULL) mlast->m_next = n; mlast = n; #if 0 newipsecstat.ips_clcopied++; #endif len -= cc; if (len <= 0) break; off += cc; n = m_getcl(how, m->m_type, m->m_flags & M_COPYFLAGS); if (n == NULL) { m_freem(mfirst); m_freem(m0); return (NULL); } } n->m_next = m->m_next; if (mprev == NULL) m0 = mfirst; /* new head of chain */ else mprev->m_next = mfirst; /* replace old mbuf */ m_free(m); /* release old mbuf */ mprev = mfirst; } return (m0); } #ifdef MBUF_PROFILING #define MP_BUCKETS 32 /* don't just change this as things may overflow.*/ struct mbufprofile { uintmax_t wasted[MP_BUCKETS]; uintmax_t used[MP_BUCKETS]; uintmax_t segments[MP_BUCKETS]; } mbprof; #define MP_MAXDIGITS 21 /* strlen("16,000,000,000,000,000,000") == 21 */ #define MP_NUMLINES 6 #define MP_NUMSPERLINE 16 #define MP_EXTRABYTES 64 /* > strlen("used:\nwasted:\nsegments:\n") */ /* work out max space needed and add a bit of spare space too */ #define MP_MAXLINE ((MP_MAXDIGITS+1) * MP_NUMSPERLINE) #define MP_BUFSIZE ((MP_MAXLINE * MP_NUMLINES) + 1 + MP_EXTRABYTES) char mbprofbuf[MP_BUFSIZE]; void m_profile(struct mbuf *m) { int segments = 0; int used = 0; int wasted = 0; while (m) { segments++; used += m->m_len; if (m->m_flags & M_EXT) { wasted += MHLEN - sizeof(m->m_ext) + m->m_ext.ext_size - m->m_len; } else { if (m->m_flags & M_PKTHDR) wasted += MHLEN - m->m_len; else wasted += MLEN - m->m_len; } m = m->m_next; } /* be paranoid.. it helps */ if (segments > MP_BUCKETS - 1) segments = MP_BUCKETS - 1; if (used > 100000) used = 100000; if (wasted > 100000) wasted = 100000; /* store in the appropriate bucket */ /* don't bother locking. if it's slightly off, so what? */ mbprof.segments[segments]++; mbprof.used[fls(used)]++; mbprof.wasted[fls(wasted)]++; } static void mbprof_textify(void) { int offset; char *c; uint64_t *p; p = &mbprof.wasted[0]; c = mbprofbuf; offset = snprintf(c, MP_MAXLINE + 10, "wasted:\n" "%ju %ju %ju %ju %ju %ju %ju %ju " "%ju %ju %ju %ju %ju %ju %ju %ju\n", p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); #ifdef BIG_ARRAY p = &mbprof.wasted[16]; c += offset; offset = snprintf(c, MP_MAXLINE, "%ju %ju %ju %ju %ju %ju %ju %ju " "%ju %ju %ju %ju %ju %ju %ju %ju\n", p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); #endif p = &mbprof.used[0]; c += offset; offset = snprintf(c, MP_MAXLINE + 10, "used:\n" "%ju %ju %ju %ju %ju %ju %ju %ju " "%ju %ju %ju %ju %ju %ju %ju %ju\n", p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); #ifdef BIG_ARRAY p = &mbprof.used[16]; c += offset; offset = snprintf(c, MP_MAXLINE, "%ju %ju %ju %ju %ju %ju %ju %ju " "%ju %ju %ju %ju %ju %ju %ju %ju\n", p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); #endif p = &mbprof.segments[0]; c += offset; offset = snprintf(c, MP_MAXLINE + 10, "segments:\n" "%ju %ju %ju %ju %ju %ju %ju %ju " "%ju %ju %ju %ju %ju %ju %ju %ju\n", p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); #ifdef BIG_ARRAY p = &mbprof.segments[16]; c += offset; offset = snprintf(c, MP_MAXLINE, "%ju %ju %ju %ju %ju %ju %ju %ju " "%ju %ju %ju %ju %ju %ju %ju %jju", p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7], p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); #endif } static int mbprof_handler(SYSCTL_HANDLER_ARGS) { int error; mbprof_textify(); error = SYSCTL_OUT(req, mbprofbuf, strlen(mbprofbuf) + 1); return (error); } static int mbprof_clr_handler(SYSCTL_HANDLER_ARGS) { int clear, error; clear = 0; error = sysctl_handle_int(oidp, &clear, 0, req); if (error || !req->newptr) return (error); if (clear) { bzero(&mbprof, sizeof(mbprof)); } return (error); } SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofile, CTLTYPE_STRING|CTLFLAG_RD, NULL, 0, mbprof_handler, "A", "mbuf profiling statistics"); SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofileclr, CTLTYPE_INT|CTLFLAG_RW, NULL, 0, mbprof_clr_handler, "I", "clear mbuf profiling statistics"); #endif Index: head/sys/kern/uipc_mbuf2.c =================================================================== --- head/sys/kern/uipc_mbuf2.c (revision 305831) +++ head/sys/kern/uipc_mbuf2.c (revision 305832) @@ -1,451 +1,451 @@ /* $KAME: uipc_mbuf2.c,v 1.31 2001/11/28 11:08:53 itojun Exp $ */ /* $NetBSD: uipc_mbuf.c,v 1.40 1999/04/01 00:23:25 thorpej Exp $ */ /*- * Copyright (C) 1999 WIDE Project. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the project nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /*- * Copyright (c) 1982, 1986, 1988, 1991, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)uipc_mbuf.c 8.4 (Berkeley) 2/14/95 */ #include __FBSDID("$FreeBSD$"); /*#define PULLDOWN_DEBUG*/ #include #include #include #include #include #include #include #include static MALLOC_DEFINE(M_PACKET_TAGS, MBUF_TAG_MEM_NAME, "packet-attached information"); /* can't call it m_dup(), as freebsd[34] uses m_dup() with different arg */ static struct mbuf *m_dup1(struct mbuf *, int, int, int); /* * ensure that [off, off + len) is contiguous on the mbuf chain "m". * packet chain before "off" is kept untouched. * if offp == NULL, the target will start at on resulting chain. * if offp != NULL, the target will start at on resulting chain. * * on error return (NULL return value), original "m" will be freed. * * XXX: M_TRAILINGSPACE/M_LEADINGSPACE only permitted on writable ext_buf. */ struct mbuf * m_pulldown(struct mbuf *m, int off, int len, int *offp) { struct mbuf *n, *o; int hlen, tlen, olen; int writable; /* check invalid arguments. */ if (m == NULL) panic("m == NULL in m_pulldown()"); if (len > MCLBYTES) { m_freem(m); return NULL; /* impossible */ } #ifdef PULLDOWN_DEBUG { struct mbuf *t; printf("before:"); for (t = m; t; t = t->m_next) printf(" %d", t->m_len); printf("\n"); } #endif n = m; while (n != NULL && off > 0) { if (n->m_len > off) break; off -= n->m_len; n = n->m_next; } /* be sure to point non-empty mbuf */ while (n != NULL && n->m_len == 0) n = n->m_next; if (!n) { m_freem(m); return NULL; /* mbuf chain too short */ } /* * The following comment is dated but still partially applies: * * XXX: This code is flawed because it considers a "writable" mbuf * data region to require all of the following: * (i) mbuf _has_ to have M_EXT set; if it is just a regular * mbuf, it is still not considered "writable." * (ii) since mbuf has M_EXT, the ext_type _has_ to be * EXT_CLUSTER. Anything else makes it non-writable. * (iii) M_WRITABLE() must evaluate true. * Ideally, the requirement should only be (iii). * * If we're writable, we're sure we're writable, because the ref. count * cannot increase from 1, as that would require possession of mbuf * n by someone else (which is impossible). However, if we're _not_ * writable, we may eventually become writable )if the ref. count drops * to 1), but we'll fail to notice it unless we re-evaluate * M_WRITABLE(). For now, we only evaluate once at the beginning and * live with this. */ writable = 0; if ((n->m_flags & M_EXT) == 0 || (n->m_ext.ext_type == EXT_CLUSTER && M_WRITABLE(n))) writable = 1; /* * the target data is on . * if we got enough data on the mbuf "n", we're done. */ if ((off == 0 || offp) && len <= n->m_len - off && writable) goto ok; /* * when len <= n->m_len - off and off != 0, it is a special case. * len bytes from sits in single mbuf, but the caller does * not like the starting position (off). * chop the current mbuf into two pieces, set off to 0. */ if (len <= n->m_len - off) { o = m_dup1(n, off, n->m_len - off, M_NOWAIT); if (o == NULL) { m_freem(m); return NULL; /* ENOBUFS */ } n->m_len = off; o->m_next = n->m_next; n->m_next = o; n = n->m_next; off = 0; goto ok; } /* * we need to take hlen from and tlen from m_next, 0>, * and construct contiguous mbuf with m_len == len. * note that hlen + tlen == len, and tlen > 0. */ hlen = n->m_len - off; tlen = len - hlen; /* * ensure that we have enough trailing data on mbuf chain. * if not, we can do nothing about the chain. */ olen = 0; for (o = n->m_next; o != NULL; o = o->m_next) olen += o->m_len; if (hlen + olen < len) { m_freem(m); return NULL; /* mbuf chain too short */ } /* * easy cases first. * we need to use m_copydata() to get data from m_next, 0>. */ if ((off == 0 || offp) && M_TRAILINGSPACE(n) >= tlen && writable) { m_copydata(n->m_next, 0, tlen, mtod(n, caddr_t) + n->m_len); n->m_len += tlen; m_adj(n->m_next, tlen); goto ok; } if ((off == 0 || offp) && M_LEADINGSPACE(n->m_next) >= hlen && writable) { n->m_next->m_data -= hlen; n->m_next->m_len += hlen; bcopy(mtod(n, caddr_t) + off, mtod(n->m_next, caddr_t), hlen); n->m_len -= hlen; n = n->m_next; off = 0; goto ok; } /* * now, we need to do the hard way. don't m_copy as there's no room * on both end. */ if (len > MLEN) o = m_getcl(M_NOWAIT, m->m_type, 0); else o = m_get(M_NOWAIT, m->m_type); if (!o) { m_freem(m); return NULL; /* ENOBUFS */ } /* get hlen from into */ o->m_len = hlen; bcopy(mtod(n, caddr_t) + off, mtod(o, caddr_t), hlen); n->m_len -= hlen; /* get tlen from m_next, 0> into */ m_copydata(n->m_next, 0, tlen, mtod(o, caddr_t) + o->m_len); o->m_len += tlen; m_adj(n->m_next, tlen); o->m_next = n->m_next; n->m_next = o; n = o; off = 0; ok: #ifdef PULLDOWN_DEBUG { struct mbuf *t; printf("after:"); for (t = m; t; t = t->m_next) printf("%c%d", t == n ? '*' : ' ', t->m_len); printf(" (off=%d)\n", off); } #endif if (offp) *offp = off; return n; } static struct mbuf * m_dup1(struct mbuf *m, int off, int len, int wait) { struct mbuf *n; int copyhdr; if (len > MCLBYTES) return NULL; if (off == 0 && (m->m_flags & M_PKTHDR) != 0) copyhdr = 1; else copyhdr = 0; if (len >= MINCLSIZE) { if (copyhdr == 1) n = m_getcl(wait, m->m_type, M_PKTHDR); else n = m_getcl(wait, m->m_type, 0); } else { if (copyhdr == 1) n = m_gethdr(wait, m->m_type); else n = m_get(wait, m->m_type); } if (!n) return NULL; /* ENOBUFS */ if (copyhdr && !m_dup_pkthdr(n, m, wait)) { m_free(n); return NULL; } m_copydata(m, off, len, mtod(n, caddr_t)); n->m_len = len; return n; } /* Free a packet tag. */ void m_tag_free_default(struct m_tag *t) { #ifdef MAC if (t->m_tag_id == PACKET_TAG_MACLABEL) mac_mbuf_tag_destroy(t); #endif free(t, M_PACKET_TAGS); } /* Get a packet tag structure along with specified data following. */ struct m_tag * m_tag_alloc(uint32_t cookie, int type, int len, int wait) { struct m_tag *t; MBUF_CHECKSLEEP(wait); if (len < 0) return NULL; t = malloc(len + sizeof(struct m_tag), M_PACKET_TAGS, wait); if (t == NULL) return NULL; m_tag_setup(t, cookie, type, len); t->m_tag_free = m_tag_free_default; return t; } /* Unlink and free a packet tag. */ void m_tag_delete(struct mbuf *m, struct m_tag *t) { KASSERT(m && t, ("m_tag_delete: null argument, m %p t %p", m, t)); m_tag_unlink(m, t); m_tag_free(t); } /* Unlink and free a packet tag chain, starting from given tag. */ void m_tag_delete_chain(struct mbuf *m, struct m_tag *t) { struct m_tag *p, *q; KASSERT(m, ("m_tag_delete_chain: null mbuf")); if (t != NULL) p = t; else p = SLIST_FIRST(&m->m_pkthdr.tags); if (p == NULL) return; while ((q = SLIST_NEXT(p, m_tag_link)) != NULL) m_tag_delete(m, q); m_tag_delete(m, p); } /* * Strip off all tags that would normally vanish when * passing through a network interface. Only persistent * tags will exist after this; these are expected to remain * so long as the mbuf chain exists, regardless of the * path the mbufs take. */ void m_tag_delete_nonpersistent(struct mbuf *m) { struct m_tag *p, *q; SLIST_FOREACH_SAFE(p, &m->m_pkthdr.tags, m_tag_link, q) if ((p->m_tag_id & MTAG_PERSISTENT) == 0) m_tag_delete(m, p); } /* Find a tag, starting from a given position. */ struct m_tag * m_tag_locate(struct mbuf *m, uint32_t cookie, int type, struct m_tag *t) { struct m_tag *p; KASSERT(m, ("m_tag_locate: null mbuf")); if (t == NULL) p = SLIST_FIRST(&m->m_pkthdr.tags); else p = SLIST_NEXT(t, m_tag_link); while (p != NULL) { if (p->m_tag_cookie == cookie && p->m_tag_id == type) return p; p = SLIST_NEXT(p, m_tag_link); } return NULL; } /* Copy a single tag. */ struct m_tag * m_tag_copy(struct m_tag *t, int how) { struct m_tag *p; MBUF_CHECKSLEEP(how); KASSERT(t, ("m_tag_copy: null tag")); p = m_tag_alloc(t->m_tag_cookie, t->m_tag_id, t->m_tag_len, how); if (p == NULL) return (NULL); #ifdef MAC /* * XXXMAC: we should probably pass off the initialization, and * copying here? can we hide that PACKET_TAG_MACLABEL is * special from the mbuf code? */ if (t->m_tag_id == PACKET_TAG_MACLABEL) { if (mac_mbuf_tag_init(p, how) != 0) { m_tag_free(p); return (NULL); } mac_mbuf_tag_copy(t, p); } else #endif bcopy(t + 1, p + 1, t->m_tag_len); /* Copy the data */ return p; } /* * Copy two tag chains. The destination mbuf (to) loses any attached * tags even if the operation fails. This should not be a problem, as * m_tag_copy_chain() is typically called with a newly-allocated * destination mbuf. */ int m_tag_copy_chain(struct mbuf *to, const struct mbuf *from, int how) { struct m_tag *p, *t, *tprev = NULL; MBUF_CHECKSLEEP(how); KASSERT(to && from, ("m_tag_copy_chain: null argument, to %p from %p", to, from)); m_tag_delete_chain(to, NULL); SLIST_FOREACH(p, &from->m_pkthdr.tags, m_tag_link) { t = m_tag_copy(p, how); if (t == NULL) { m_tag_delete_chain(to, NULL); return 0; } if (tprev == NULL) SLIST_INSERT_HEAD(&to->m_pkthdr.tags, t, m_tag_link); else SLIST_INSERT_AFTER(tprev, t, m_tag_link); tprev = t; } return 1; } Index: head/sys/kern/uipc_sockbuf.c =================================================================== --- head/sys/kern/uipc_sockbuf.c (revision 305831) +++ head/sys/kern/uipc_sockbuf.c (revision 305832) @@ -1,1332 +1,1332 @@ /*- * Copyright (c) 1982, 1986, 1988, 1990, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)uipc_socket2.c 8.1 (Berkeley) 6/10/93 */ #include __FBSDID("$FreeBSD$"); #include "opt_param.h" #include #include /* for aio_swake proto */ #include #include #include #include #include #include #include #include #include #include #include #include #include /* * Function pointer set by the AIO routines so that the socket buffer code * can call back into the AIO module if it is loaded. */ void (*aio_swake)(struct socket *, struct sockbuf *); /* * Primitive routines for operating on socket buffers */ u_long sb_max = SB_MAX; u_long sb_max_adj = (quad_t)SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */ static u_long sb_efficiency = 8; /* parameter for sbreserve() */ static struct mbuf *sbcut_internal(struct sockbuf *sb, int len); static void sbflush_internal(struct sockbuf *sb); /* * Our own version of m_clrprotoflags(), that can preserve M_NOTREADY. */ static void sbm_clrprotoflags(struct mbuf *m, int flags) { int mask; mask = ~M_PROTOFLAGS; if (flags & PRUS_NOTREADY) mask |= M_NOTREADY; while (m) { m->m_flags &= mask; m = m->m_next; } } /* * Mark ready "count" mbufs starting with "m". */ int sbready(struct sockbuf *sb, struct mbuf *m, int count) { u_int blocker; SOCKBUF_LOCK_ASSERT(sb); KASSERT(sb->sb_fnrdy != NULL, ("%s: sb %p NULL fnrdy", __func__, sb)); blocker = (sb->sb_fnrdy == m) ? M_BLOCKED : 0; for (int i = 0; i < count; i++, m = m->m_next) { KASSERT(m->m_flags & M_NOTREADY, ("%s: m %p !M_NOTREADY", __func__, m)); m->m_flags &= ~(M_NOTREADY | blocker); if (blocker) sb->sb_acc += m->m_len; } if (!blocker) return (EINPROGRESS); /* This one was blocking all the queue. */ for (; m && (m->m_flags & M_NOTREADY) == 0; m = m->m_next) { KASSERT(m->m_flags & M_BLOCKED, ("%s: m %p !M_BLOCKED", __func__, m)); m->m_flags &= ~M_BLOCKED; sb->sb_acc += m->m_len; } sb->sb_fnrdy = m; return (0); } /* * Adjust sockbuf state reflecting allocation of m. */ void sballoc(struct sockbuf *sb, struct mbuf *m) { SOCKBUF_LOCK_ASSERT(sb); sb->sb_ccc += m->m_len; if (sb->sb_fnrdy == NULL) { if (m->m_flags & M_NOTREADY) sb->sb_fnrdy = m; else sb->sb_acc += m->m_len; } else m->m_flags |= M_BLOCKED; if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) sb->sb_ctl += m->m_len; sb->sb_mbcnt += MSIZE; sb->sb_mcnt += 1; if (m->m_flags & M_EXT) { sb->sb_mbcnt += m->m_ext.ext_size; sb->sb_ccnt += 1; } } /* * Adjust sockbuf state reflecting freeing of m. */ void sbfree(struct sockbuf *sb, struct mbuf *m) { #if 0 /* XXX: not yet: soclose() call path comes here w/o lock. */ SOCKBUF_LOCK_ASSERT(sb); #endif sb->sb_ccc -= m->m_len; if (!(m->m_flags & M_NOTAVAIL)) sb->sb_acc -= m->m_len; if (m == sb->sb_fnrdy) { struct mbuf *n; KASSERT(m->m_flags & M_NOTREADY, ("%s: m %p !M_NOTREADY", __func__, m)); n = m->m_next; while (n != NULL && !(n->m_flags & M_NOTREADY)) { n->m_flags &= ~M_BLOCKED; sb->sb_acc += n->m_len; n = n->m_next; } sb->sb_fnrdy = n; } if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) sb->sb_ctl -= m->m_len; sb->sb_mbcnt -= MSIZE; sb->sb_mcnt -= 1; if (m->m_flags & M_EXT) { sb->sb_mbcnt -= m->m_ext.ext_size; sb->sb_ccnt -= 1; } if (sb->sb_sndptr == m) { sb->sb_sndptr = NULL; sb->sb_sndptroff = 0; } if (sb->sb_sndptroff != 0) sb->sb_sndptroff -= m->m_len; } /* * Socantsendmore indicates that no more data will be sent on the socket; it * would normally be applied to a socket when the user informs the system * that no more data is to be sent, by the protocol code (in case * PRU_SHUTDOWN). Socantrcvmore indicates that no more data will be * received, and will normally be applied to the socket by a protocol when it * detects that the peer will send no more data. Data queued for reading in * the socket may yet be read. */ void socantsendmore_locked(struct socket *so) { SOCKBUF_LOCK_ASSERT(&so->so_snd); so->so_snd.sb_state |= SBS_CANTSENDMORE; sowwakeup_locked(so); mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED); } void socantsendmore(struct socket *so) { SOCKBUF_LOCK(&so->so_snd); socantsendmore_locked(so); mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED); } void socantrcvmore_locked(struct socket *so) { SOCKBUF_LOCK_ASSERT(&so->so_rcv); so->so_rcv.sb_state |= SBS_CANTRCVMORE; sorwakeup_locked(so); mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED); } void socantrcvmore(struct socket *so) { SOCKBUF_LOCK(&so->so_rcv); socantrcvmore_locked(so); mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED); } /* * Wait for data to arrive at/drain from a socket buffer. */ int sbwait(struct sockbuf *sb) { SOCKBUF_LOCK_ASSERT(sb); sb->sb_flags |= SB_WAIT; return (msleep_sbt(&sb->sb_acc, &sb->sb_mtx, (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait", sb->sb_timeo, 0, 0)); } int sblock(struct sockbuf *sb, int flags) { KASSERT((flags & SBL_VALID) == flags, ("sblock: flags invalid (0x%x)", flags)); if (flags & SBL_WAIT) { if ((sb->sb_flags & SB_NOINTR) || (flags & SBL_NOINTR)) { sx_xlock(&sb->sb_sx); return (0); } return (sx_xlock_sig(&sb->sb_sx)); } else { if (sx_try_xlock(&sb->sb_sx) == 0) return (EWOULDBLOCK); return (0); } } void sbunlock(struct sockbuf *sb) { sx_xunlock(&sb->sb_sx); } /* * Wakeup processes waiting on a socket buffer. Do asynchronous notification * via SIGIO if the socket has the SS_ASYNC flag set. * * Called with the socket buffer lock held; will release the lock by the end * of the function. This allows the caller to acquire the socket buffer lock * while testing for the need for various sorts of wakeup and hold it through * to the point where it's no longer required. We currently hold the lock * through calls out to other subsystems (with the exception of kqueue), and * then release it to avoid lock order issues. It's not clear that's * correct. */ void sowakeup(struct socket *so, struct sockbuf *sb) { int ret; SOCKBUF_LOCK_ASSERT(sb); selwakeuppri(&sb->sb_sel, PSOCK); if (!SEL_WAITING(&sb->sb_sel)) sb->sb_flags &= ~SB_SEL; if (sb->sb_flags & SB_WAIT) { sb->sb_flags &= ~SB_WAIT; wakeup(&sb->sb_acc); } KNOTE_LOCKED(&sb->sb_sel.si_note, 0); if (sb->sb_upcall != NULL) { ret = sb->sb_upcall(so, sb->sb_upcallarg, M_NOWAIT); if (ret == SU_ISCONNECTED) { KASSERT(sb == &so->so_rcv, ("SO_SND upcall returned SU_ISCONNECTED")); soupcall_clear(so, SO_RCV); } } else ret = SU_OK; if (sb->sb_flags & SB_AIO) sowakeup_aio(so, sb); SOCKBUF_UNLOCK(sb); if (ret == SU_ISCONNECTED) soisconnected(so); if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL) pgsigio(&so->so_sigio, SIGIO, 0); mtx_assert(SOCKBUF_MTX(sb), MA_NOTOWNED); } /* * Socket buffer (struct sockbuf) utility routines. * * Each socket contains two socket buffers: one for sending data and one for * receiving data. Each buffer contains a queue of mbufs, information about * the number of mbufs and amount of data in the queue, and other fields * allowing select() statements and notification on data availability to be * implemented. * * Data stored in a socket buffer is maintained as a list of records. Each * record is a list of mbufs chained together with the m_next field. Records * are chained together with the m_nextpkt field. The upper level routine * soreceive() expects the following conventions to be observed when placing * information in the receive buffer: * * 1. If the protocol requires each message be preceded by the sender's name, * then a record containing that name must be present before any * associated data (mbuf's must be of type MT_SONAME). * 2. If the protocol supports the exchange of ``access rights'' (really just * additional data associated with the message), and there are ``rights'' * to be received, then a record containing this data should be present * (mbuf's must be of type MT_RIGHTS). * 3. If a name or rights record exists, then it must be followed by a data * record, perhaps of zero length. * * Before using a new socket structure it is first necessary to reserve * buffer space to the socket, by calling sbreserve(). This should commit * some of the available buffer space in the system buffer pool for the * socket (currently, it does nothing but enforce limits). The space should * be released by calling sbrelease() when the socket is destroyed. */ int soreserve(struct socket *so, u_long sndcc, u_long rcvcc) { struct thread *td = curthread; SOCKBUF_LOCK(&so->so_snd); SOCKBUF_LOCK(&so->so_rcv); if (sbreserve_locked(&so->so_snd, sndcc, so, td) == 0) goto bad; if (sbreserve_locked(&so->so_rcv, rcvcc, so, td) == 0) goto bad2; if (so->so_rcv.sb_lowat == 0) so->so_rcv.sb_lowat = 1; if (so->so_snd.sb_lowat == 0) so->so_snd.sb_lowat = MCLBYTES; if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat) so->so_snd.sb_lowat = so->so_snd.sb_hiwat; SOCKBUF_UNLOCK(&so->so_rcv); SOCKBUF_UNLOCK(&so->so_snd); return (0); bad2: sbrelease_locked(&so->so_snd, so); bad: SOCKBUF_UNLOCK(&so->so_rcv); SOCKBUF_UNLOCK(&so->so_snd); return (ENOBUFS); } static int sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS) { int error = 0; u_long tmp_sb_max = sb_max; error = sysctl_handle_long(oidp, &tmp_sb_max, arg2, req); if (error || !req->newptr) return (error); if (tmp_sb_max < MSIZE + MCLBYTES) return (EINVAL); sb_max = tmp_sb_max; sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES); return (0); } /* * Allot mbufs to a sockbuf. Attempt to scale mbmax so that mbcnt doesn't * become limiting if buffering efficiency is near the normal case. */ int sbreserve_locked(struct sockbuf *sb, u_long cc, struct socket *so, struct thread *td) { rlim_t sbsize_limit; SOCKBUF_LOCK_ASSERT(sb); /* * When a thread is passed, we take into account the thread's socket * buffer size limit. The caller will generally pass curthread, but * in the TCP input path, NULL will be passed to indicate that no * appropriate thread resource limits are available. In that case, * we don't apply a process limit. */ if (cc > sb_max_adj) return (0); if (td != NULL) { sbsize_limit = lim_cur(td, RLIMIT_SBSIZE); } else sbsize_limit = RLIM_INFINITY; if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc, sbsize_limit)) return (0); sb->sb_mbmax = min(cc * sb_efficiency, sb_max); if (sb->sb_lowat > sb->sb_hiwat) sb->sb_lowat = sb->sb_hiwat; return (1); } int sbreserve(struct sockbuf *sb, u_long cc, struct socket *so, struct thread *td) { int error; SOCKBUF_LOCK(sb); error = sbreserve_locked(sb, cc, so, td); SOCKBUF_UNLOCK(sb); return (error); } /* * Free mbufs held by a socket, and reserved mbuf space. */ void sbrelease_internal(struct sockbuf *sb, struct socket *so) { sbflush_internal(sb); (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, RLIM_INFINITY); sb->sb_mbmax = 0; } void sbrelease_locked(struct sockbuf *sb, struct socket *so) { SOCKBUF_LOCK_ASSERT(sb); sbrelease_internal(sb, so); } void sbrelease(struct sockbuf *sb, struct socket *so) { SOCKBUF_LOCK(sb); sbrelease_locked(sb, so); SOCKBUF_UNLOCK(sb); } void sbdestroy(struct sockbuf *sb, struct socket *so) { sbrelease_internal(sb, so); } /* * Routines to add and remove data from an mbuf queue. * * The routines sbappend() or sbappendrecord() are normally called to append * new mbufs to a socket buffer, after checking that adequate space is * available, comparing the function sbspace() with the amount of data to be * added. sbappendrecord() differs from sbappend() in that data supplied is * treated as the beginning of a new record. To place a sender's address, * optional access rights, and data in a socket receive buffer, * sbappendaddr() should be used. To place access rights and data in a * socket receive buffer, sbappendrights() should be used. In either case, * the new data begins a new record. Note that unlike sbappend() and * sbappendrecord(), these routines check for the caller that there will be * enough space to store the data. Each fails if there is not enough space, * or if it cannot find mbufs to store additional information in. * * Reliable protocols may use the socket send buffer to hold data awaiting * acknowledgement. Data is normally copied from a socket send buffer in a * protocol with m_copy for output to a peer, and then removing the data from * the socket buffer with sbdrop() or sbdroprecord() when the data is * acknowledged by the peer. */ #ifdef SOCKBUF_DEBUG void sblastrecordchk(struct sockbuf *sb, const char *file, int line) { struct mbuf *m = sb->sb_mb; SOCKBUF_LOCK_ASSERT(sb); while (m && m->m_nextpkt) m = m->m_nextpkt; if (m != sb->sb_lastrecord) { printf("%s: sb_mb %p sb_lastrecord %p last %p\n", __func__, sb->sb_mb, sb->sb_lastrecord, m); printf("packet chain:\n"); for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) printf("\t%p\n", m); panic("%s from %s:%u", __func__, file, line); } } void sblastmbufchk(struct sockbuf *sb, const char *file, int line) { struct mbuf *m = sb->sb_mb; struct mbuf *n; SOCKBUF_LOCK_ASSERT(sb); while (m && m->m_nextpkt) m = m->m_nextpkt; while (m && m->m_next) m = m->m_next; if (m != sb->sb_mbtail) { printf("%s: sb_mb %p sb_mbtail %p last %p\n", __func__, sb->sb_mb, sb->sb_mbtail, m); printf("packet tree:\n"); for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) { printf("\t"); for (n = m; n != NULL; n = n->m_next) printf("%p ", n); printf("\n"); } panic("%s from %s:%u", __func__, file, line); } } #endif /* SOCKBUF_DEBUG */ #define SBLINKRECORD(sb, m0) do { \ SOCKBUF_LOCK_ASSERT(sb); \ if ((sb)->sb_lastrecord != NULL) \ (sb)->sb_lastrecord->m_nextpkt = (m0); \ else \ (sb)->sb_mb = (m0); \ (sb)->sb_lastrecord = (m0); \ } while (/*CONSTCOND*/0) /* * Append mbuf chain m to the last record in the socket buffer sb. The * additional space associated the mbuf chain is recorded in sb. Empty mbufs * are discarded and mbufs are compacted where possible. */ void sbappend_locked(struct sockbuf *sb, struct mbuf *m, int flags) { struct mbuf *n; SOCKBUF_LOCK_ASSERT(sb); if (m == NULL) return; sbm_clrprotoflags(m, flags); SBLASTRECORDCHK(sb); n = sb->sb_mb; if (n) { while (n->m_nextpkt) n = n->m_nextpkt; do { if (n->m_flags & M_EOR) { sbappendrecord_locked(sb, m); /* XXXXXX!!!! */ return; } } while (n->m_next && (n = n->m_next)); } else { /* * XXX Would like to simply use sb_mbtail here, but * XXX I need to verify that I won't miss an EOR that * XXX way. */ if ((n = sb->sb_lastrecord) != NULL) { do { if (n->m_flags & M_EOR) { sbappendrecord_locked(sb, m); /* XXXXXX!!!! */ return; } } while (n->m_next && (n = n->m_next)); } else { /* * If this is the first record in the socket buffer, * it's also the last record. */ sb->sb_lastrecord = m; } } sbcompress(sb, m, n); SBLASTRECORDCHK(sb); } /* * Append mbuf chain m to the last record in the socket buffer sb. The * additional space associated the mbuf chain is recorded in sb. Empty mbufs * are discarded and mbufs are compacted where possible. */ void sbappend(struct sockbuf *sb, struct mbuf *m, int flags) { SOCKBUF_LOCK(sb); sbappend_locked(sb, m, flags); SOCKBUF_UNLOCK(sb); } /* * This version of sbappend() should only be used when the caller absolutely * knows that there will never be more than one record in the socket buffer, * that is, a stream protocol (such as TCP). */ void sbappendstream_locked(struct sockbuf *sb, struct mbuf *m, int flags) { SOCKBUF_LOCK_ASSERT(sb); KASSERT(m->m_nextpkt == NULL,("sbappendstream 0")); KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1")); SBLASTMBUFCHK(sb); /* Remove all packet headers and mbuf tags to get a pure data chain. */ m_demote(m, 1, flags & PRUS_NOTREADY ? M_NOTREADY : 0); sbcompress(sb, m, sb->sb_mbtail); sb->sb_lastrecord = sb->sb_mb; SBLASTRECORDCHK(sb); } /* * This version of sbappend() should only be used when the caller absolutely * knows that there will never be more than one record in the socket buffer, * that is, a stream protocol (such as TCP). */ void sbappendstream(struct sockbuf *sb, struct mbuf *m, int flags) { SOCKBUF_LOCK(sb); sbappendstream_locked(sb, m, flags); SOCKBUF_UNLOCK(sb); } #ifdef SOCKBUF_DEBUG void sbcheck(struct sockbuf *sb, const char *file, int line) { struct mbuf *m, *n, *fnrdy; u_long acc, ccc, mbcnt; SOCKBUF_LOCK_ASSERT(sb); acc = ccc = mbcnt = 0; fnrdy = NULL; for (m = sb->sb_mb; m; m = n) { n = m->m_nextpkt; for (; m; m = m->m_next) { if (m->m_len == 0) { printf("sb %p empty mbuf %p\n", sb, m); goto fail; } if ((m->m_flags & M_NOTREADY) && fnrdy == NULL) { if (m != sb->sb_fnrdy) { printf("sb %p: fnrdy %p != m %p\n", sb, sb->sb_fnrdy, m); goto fail; } fnrdy = m; } if (fnrdy) { if (!(m->m_flags & M_NOTAVAIL)) { printf("sb %p: fnrdy %p, m %p is avail\n", sb, sb->sb_fnrdy, m); goto fail; } } else acc += m->m_len; ccc += m->m_len; mbcnt += MSIZE; if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */ mbcnt += m->m_ext.ext_size; } } if (acc != sb->sb_acc || ccc != sb->sb_ccc || mbcnt != sb->sb_mbcnt) { printf("acc %ld/%u ccc %ld/%u mbcnt %ld/%u\n", acc, sb->sb_acc, ccc, sb->sb_ccc, mbcnt, sb->sb_mbcnt); goto fail; } return; fail: panic("%s from %s:%u", __func__, file, line); } #endif /* * As above, except the mbuf chain begins a new record. */ void sbappendrecord_locked(struct sockbuf *sb, struct mbuf *m0) { struct mbuf *m; SOCKBUF_LOCK_ASSERT(sb); if (m0 == NULL) return; m_clrprotoflags(m0); /* * Put the first mbuf on the queue. Note this permits zero length * records. */ sballoc(sb, m0); SBLASTRECORDCHK(sb); SBLINKRECORD(sb, m0); sb->sb_mbtail = m0; m = m0->m_next; m0->m_next = 0; if (m && (m0->m_flags & M_EOR)) { m0->m_flags &= ~M_EOR; m->m_flags |= M_EOR; } /* always call sbcompress() so it can do SBLASTMBUFCHK() */ sbcompress(sb, m, m0); } /* * As above, except the mbuf chain begins a new record. */ void sbappendrecord(struct sockbuf *sb, struct mbuf *m0) { SOCKBUF_LOCK(sb); sbappendrecord_locked(sb, m0); SOCKBUF_UNLOCK(sb); } /* Helper routine that appends data, control, and address to a sockbuf. */ static int sbappendaddr_locked_internal(struct sockbuf *sb, const struct sockaddr *asa, struct mbuf *m0, struct mbuf *control, struct mbuf *ctrl_last) { struct mbuf *m, *n, *nlast; #if MSIZE <= 256 if (asa->sa_len > MLEN) return (0); #endif m = m_get(M_NOWAIT, MT_SONAME); if (m == NULL) return (0); m->m_len = asa->sa_len; bcopy(asa, mtod(m, caddr_t), asa->sa_len); if (m0) m_clrprotoflags(m0); if (ctrl_last) ctrl_last->m_next = m0; /* concatenate data to control */ else control = m0; m->m_next = control; for (n = m; n->m_next != NULL; n = n->m_next) sballoc(sb, n); sballoc(sb, n); nlast = n; SBLINKRECORD(sb, m); sb->sb_mbtail = nlast; SBLASTMBUFCHK(sb); SBLASTRECORDCHK(sb); return (1); } /* * Append address and data, and optionally, control (ancillary) data to the * receive queue of a socket. If present, m0 must include a packet header * with total length. Returns 0 if no space in sockbuf or insufficient * mbufs. */ int sbappendaddr_locked(struct sockbuf *sb, const struct sockaddr *asa, struct mbuf *m0, struct mbuf *control) { struct mbuf *ctrl_last; int space = asa->sa_len; SOCKBUF_LOCK_ASSERT(sb); if (m0 && (m0->m_flags & M_PKTHDR) == 0) panic("sbappendaddr_locked"); if (m0) space += m0->m_pkthdr.len; space += m_length(control, &ctrl_last); if (space > sbspace(sb)) return (0); return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last)); } /* * Append address and data, and optionally, control (ancillary) data to the * receive queue of a socket. If present, m0 must include a packet header * with total length. Returns 0 if insufficient mbufs. Does not validate space * on the receiving sockbuf. */ int sbappendaddr_nospacecheck_locked(struct sockbuf *sb, const struct sockaddr *asa, struct mbuf *m0, struct mbuf *control) { struct mbuf *ctrl_last; SOCKBUF_LOCK_ASSERT(sb); ctrl_last = (control == NULL) ? NULL : m_last(control); return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last)); } /* * Append address and data, and optionally, control (ancillary) data to the * receive queue of a socket. If present, m0 must include a packet header * with total length. Returns 0 if no space in sockbuf or insufficient * mbufs. */ int sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa, struct mbuf *m0, struct mbuf *control) { int retval; SOCKBUF_LOCK(sb); retval = sbappendaddr_locked(sb, asa, m0, control); SOCKBUF_UNLOCK(sb); return (retval); } int sbappendcontrol_locked(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control) { struct mbuf *m, *n, *mlast; int space; SOCKBUF_LOCK_ASSERT(sb); if (control == NULL) panic("sbappendcontrol_locked"); space = m_length(control, &n) + m_length(m0, NULL); if (space > sbspace(sb)) return (0); m_clrprotoflags(m0); n->m_next = m0; /* concatenate data to control */ SBLASTRECORDCHK(sb); for (m = control; m->m_next; m = m->m_next) sballoc(sb, m); sballoc(sb, m); mlast = m; SBLINKRECORD(sb, control); sb->sb_mbtail = mlast; SBLASTMBUFCHK(sb); SBLASTRECORDCHK(sb); return (1); } int sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control) { int retval; SOCKBUF_LOCK(sb); retval = sbappendcontrol_locked(sb, m0, control); SOCKBUF_UNLOCK(sb); return (retval); } /* * Append the data in mbuf chain (m) into the socket buffer sb following mbuf * (n). If (n) is NULL, the buffer is presumed empty. * * When the data is compressed, mbufs in the chain may be handled in one of * three ways: * * (1) The mbuf may simply be dropped, if it contributes nothing (no data, no * record boundary, and no change in data type). * * (2) The mbuf may be coalesced -- i.e., data in the mbuf may be copied into * an mbuf already in the socket buffer. This can occur if an * appropriate mbuf exists, there is room, both mbufs are not marked as * not ready, and no merging of data types will occur. * * (3) The mbuf may be appended to the end of the existing mbuf chain. * * If any of the new mbufs is marked as M_EOR, mark the last mbuf appended as * end-of-record. */ void sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n) { int eor = 0; struct mbuf *o; SOCKBUF_LOCK_ASSERT(sb); while (m) { eor |= m->m_flags & M_EOR; if (m->m_len == 0 && (eor == 0 || (((o = m->m_next) || (o = n)) && o->m_type == m->m_type))) { if (sb->sb_lastrecord == m) sb->sb_lastrecord = m->m_next; m = m_free(m); continue; } if (n && (n->m_flags & M_EOR) == 0 && M_WRITABLE(n) && ((sb->sb_flags & SB_NOCOALESCE) == 0) && !(m->m_flags & M_NOTREADY) && !(n->m_flags & M_NOTREADY) && m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */ m->m_len <= M_TRAILINGSPACE(n) && n->m_type == m->m_type) { bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len, (unsigned)m->m_len); n->m_len += m->m_len; sb->sb_ccc += m->m_len; if (sb->sb_fnrdy == NULL) sb->sb_acc += m->m_len; if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) /* XXX: Probably don't need.*/ sb->sb_ctl += m->m_len; m = m_free(m); continue; } if (n) n->m_next = m; else sb->sb_mb = m; sb->sb_mbtail = m; sballoc(sb, m); n = m; m->m_flags &= ~M_EOR; m = m->m_next; n->m_next = 0; } if (eor) { KASSERT(n != NULL, ("sbcompress: eor && n == NULL")); n->m_flags |= eor; } SBLASTMBUFCHK(sb); } /* * Free all mbufs in a sockbuf. Check that all resources are reclaimed. */ static void sbflush_internal(struct sockbuf *sb) { while (sb->sb_mbcnt) { /* * Don't call sbcut(sb, 0) if the leading mbuf is non-empty: * we would loop forever. Panic instead. */ if (sb->sb_ccc == 0 && (sb->sb_mb == NULL || sb->sb_mb->m_len)) break; m_freem(sbcut_internal(sb, (int)sb->sb_ccc)); } KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0, ("%s: ccc %u mb %p mbcnt %u", __func__, sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt)); } void sbflush_locked(struct sockbuf *sb) { SOCKBUF_LOCK_ASSERT(sb); sbflush_internal(sb); } void sbflush(struct sockbuf *sb) { SOCKBUF_LOCK(sb); sbflush_locked(sb); SOCKBUF_UNLOCK(sb); } /* * Cut data from (the front of) a sockbuf. */ static struct mbuf * sbcut_internal(struct sockbuf *sb, int len) { struct mbuf *m, *next, *mfree; next = (m = sb->sb_mb) ? m->m_nextpkt : 0; mfree = NULL; while (len > 0) { if (m == NULL) { KASSERT(next, ("%s: no next, len %d", __func__, len)); m = next; next = m->m_nextpkt; } if (m->m_len > len) { KASSERT(!(m->m_flags & M_NOTAVAIL), ("%s: m %p M_NOTAVAIL", __func__, m)); m->m_len -= len; m->m_data += len; sb->sb_ccc -= len; sb->sb_acc -= len; if (sb->sb_sndptroff != 0) sb->sb_sndptroff -= len; if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) sb->sb_ctl -= len; break; } len -= m->m_len; sbfree(sb, m); /* * Do not put M_NOTREADY buffers to the free list, they * are referenced from outside. */ if (m->m_flags & M_NOTREADY) m = m->m_next; else { struct mbuf *n; n = m->m_next; m->m_next = mfree; mfree = m; m = n; } } /* * Free any zero-length mbufs from the buffer. * For SOCK_DGRAM sockets such mbufs represent empty records. * XXX: For SOCK_STREAM sockets such mbufs can appear in the buffer, * when sosend_generic() needs to send only control data. */ while (m && m->m_len == 0) { struct mbuf *n; sbfree(sb, m); n = m->m_next; m->m_next = mfree; mfree = m; m = n; } if (m) { sb->sb_mb = m; m->m_nextpkt = next; } else sb->sb_mb = next; /* * First part is an inline SB_EMPTY_FIXUP(). Second part makes sure * sb_lastrecord is up-to-date if we dropped part of the last record. */ m = sb->sb_mb; if (m == NULL) { sb->sb_mbtail = NULL; sb->sb_lastrecord = NULL; } else if (m->m_nextpkt == NULL) { sb->sb_lastrecord = m; } return (mfree); } /* * Drop data from (the front of) a sockbuf. */ void sbdrop_locked(struct sockbuf *sb, int len) { SOCKBUF_LOCK_ASSERT(sb); m_freem(sbcut_internal(sb, len)); } /* * Drop data from (the front of) a sockbuf, * and return it to caller. */ struct mbuf * sbcut_locked(struct sockbuf *sb, int len) { SOCKBUF_LOCK_ASSERT(sb); return (sbcut_internal(sb, len)); } void sbdrop(struct sockbuf *sb, int len) { struct mbuf *mfree; SOCKBUF_LOCK(sb); mfree = sbcut_internal(sb, len); SOCKBUF_UNLOCK(sb); m_freem(mfree); } /* * Maintain a pointer and offset pair into the socket buffer mbuf chain to * avoid traversal of the entire socket buffer for larger offsets. */ struct mbuf * sbsndptr(struct sockbuf *sb, u_int off, u_int len, u_int *moff) { struct mbuf *m, *ret; KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__)); KASSERT(off + len <= sb->sb_acc, ("%s: beyond sb", __func__)); KASSERT(sb->sb_sndptroff <= sb->sb_acc, ("%s: sndptroff broken", __func__)); /* * Is off below stored offset? Happens on retransmits. * Just return, we can't help here. */ if (sb->sb_sndptroff > off) { *moff = off; return (sb->sb_mb); } /* Return closest mbuf in chain for current offset. */ *moff = off - sb->sb_sndptroff; m = ret = sb->sb_sndptr ? sb->sb_sndptr : sb->sb_mb; if (*moff == m->m_len) { *moff = 0; sb->sb_sndptroff += m->m_len; m = ret = m->m_next; KASSERT(ret->m_len > 0, ("mbuf %p in sockbuf %p chain has no valid data", ret, sb)); } /* Advance by len to be as close as possible for the next transmit. */ for (off = off - sb->sb_sndptroff + len - 1; off > 0 && m != NULL && off >= m->m_len; m = m->m_next) { sb->sb_sndptroff += m->m_len; off -= m->m_len; } if (off > 0 && m == NULL) panic("%s: sockbuf %p and mbuf %p clashing", __func__, sb, ret); sb->sb_sndptr = m; return (ret); } /* * Return the first mbuf and the mbuf data offset for the provided * send offset without changing the "sb_sndptroff" field. */ struct mbuf * sbsndmbuf(struct sockbuf *sb, u_int off, u_int *moff) { struct mbuf *m; KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__)); /* * If the "off" is below the stored offset, which happens on * retransmits, just use "sb_mb": */ if (sb->sb_sndptr == NULL || sb->sb_sndptroff > off) { m = sb->sb_mb; } else { m = sb->sb_sndptr; off -= sb->sb_sndptroff; } while (off > 0 && m != NULL) { if (off < m->m_len) break; off -= m->m_len; m = m->m_next; } *moff = off; return (m); } /* * Drop a record off the front of a sockbuf and move the next record to the * front. */ void sbdroprecord_locked(struct sockbuf *sb) { struct mbuf *m; SOCKBUF_LOCK_ASSERT(sb); m = sb->sb_mb; if (m) { sb->sb_mb = m->m_nextpkt; do { sbfree(sb, m); m = m_free(m); } while (m); } SB_EMPTY_FIXUP(sb); } /* * Drop a record off the front of a sockbuf and move the next record to the * front. */ void sbdroprecord(struct sockbuf *sb) { SOCKBUF_LOCK(sb); sbdroprecord_locked(sb); SOCKBUF_UNLOCK(sb); } /* * Create a "control" mbuf containing the specified data with the specified * type for presentation on a socket buffer. */ struct mbuf * sbcreatecontrol(caddr_t p, int size, int type, int level) { struct cmsghdr *cp; struct mbuf *m; if (CMSG_SPACE((u_int)size) > MCLBYTES) return ((struct mbuf *) NULL); if (CMSG_SPACE((u_int)size) > MLEN) m = m_getcl(M_NOWAIT, MT_CONTROL, 0); else m = m_get(M_NOWAIT, MT_CONTROL); if (m == NULL) return ((struct mbuf *) NULL); cp = mtod(m, struct cmsghdr *); m->m_len = 0; KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m), ("sbcreatecontrol: short mbuf")); /* * Don't leave the padding between the msg header and the * cmsg data and the padding after the cmsg data un-initialized. */ bzero(cp, CMSG_SPACE((u_int)size)); if (p != NULL) (void)memcpy(CMSG_DATA(cp), p, size); m->m_len = CMSG_SPACE(size); cp->cmsg_len = CMSG_LEN(size); cp->cmsg_level = level; cp->cmsg_type = type; return (m); } /* * This does the same for socket buffers that sotoxsocket does for sockets: * generate an user-format data structure describing the socket buffer. Note * that the xsockbuf structure, since it is always embedded in a socket, does * not include a self pointer nor a length. We make this entry point public * in case some other mechanism needs it. */ void sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb) { xsb->sb_cc = sb->sb_ccc; xsb->sb_hiwat = sb->sb_hiwat; xsb->sb_mbcnt = sb->sb_mbcnt; xsb->sb_mcnt = sb->sb_mcnt; xsb->sb_ccnt = sb->sb_ccnt; xsb->sb_mbmax = sb->sb_mbmax; xsb->sb_lowat = sb->sb_lowat; xsb->sb_flags = sb->sb_flags; xsb->sb_timeo = sb->sb_timeo; } /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */ static int dummy; SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, ""); SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_ULONG|CTLFLAG_RW, &sb_max, 0, sysctl_handle_sb_max, "LU", "Maximum socket buffer size"); SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW, &sb_efficiency, 0, "Socket buffer size waste factor"); Index: head/sys/kern/uipc_socket.c =================================================================== --- head/sys/kern/uipc_socket.c (revision 305831) +++ head/sys/kern/uipc_socket.c (revision 305832) @@ -1,3728 +1,3728 @@ /*- * Copyright (c) 1982, 1986, 1988, 1990, 1993 * The Regents of the University of California. * Copyright (c) 2004 The FreeBSD Foundation * Copyright (c) 2004-2008 Robert N. M. Watson * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94 */ /* * Comments on the socket life cycle: * * soalloc() sets of socket layer state for a socket, called only by * socreate() and sonewconn(). Socket layer private. * * sodealloc() tears down socket layer state for a socket, called only by * sofree() and sonewconn(). Socket layer private. * * pru_attach() associates protocol layer state with an allocated socket; * called only once, may fail, aborting socket allocation. This is called * from socreate() and sonewconn(). Socket layer private. * * pru_detach() disassociates protocol layer state from an attached socket, * and will be called exactly once for sockets in which pru_attach() has * been successfully called. If pru_attach() returned an error, * pru_detach() will not be called. Socket layer private. * * pru_abort() and pru_close() notify the protocol layer that the last * consumer of a socket is starting to tear down the socket, and that the * protocol should terminate the connection. Historically, pru_abort() also * detached protocol state from the socket state, but this is no longer the * case. * * socreate() creates a socket and attaches protocol state. This is a public * interface that may be used by socket layer consumers to create new * sockets. * * sonewconn() creates a socket and attaches protocol state. This is a * public interface that may be used by protocols to create new sockets when * a new connection is received and will be available for accept() on a * listen socket. * * soclose() destroys a socket after possibly waiting for it to disconnect. * This is a public interface that socket consumers should use to close and * release a socket when done with it. * * soabort() destroys a socket without waiting for it to disconnect (used * only for incoming connections that are already partially or fully * connected). This is used internally by the socket layer when clearing * listen socket queues (due to overflow or close on the listen socket), but * is also a public interface protocols may use to abort connections in * their incomplete listen queues should they no longer be required. Sockets * placed in completed connection listen queues should not be aborted for * reasons described in the comment above the soclose() implementation. This * is not a general purpose close routine, and except in the specific * circumstances described here, should not be used. * * sofree() will free a socket and its protocol state if all references on * the socket have been released, and is the public interface to attempt to * free a socket when a reference is removed. This is a socket layer private * interface. * * NOTE: In addition to socreate() and soclose(), which provide a single * socket reference to the consumer to be managed as required, there are two * calls to explicitly manage socket references, soref(), and sorele(). * Currently, these are generally required only when transitioning a socket * from a listen queue to a file descriptor, in order to prevent garbage * collection of the socket at an untimely moment. For a number of reasons, * these interfaces are not preferred, and should be avoided. * * NOTE: With regard to VNETs the general rule is that callers do not set * curvnet. Exceptions to this rule include soabort(), sodisconnect(), * sofree() (and with that sorele(), sotryfree()), as well as sonewconn() * and sorflush(), which are usually called from a pre-set VNET context. * sopoll() currently does not need a VNET context to be set. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_compat.h" #include #include #include #include #include #include #include #include #include #include #include /* for struct knote */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef COMPAT_FREEBSD32 #include #include #include #endif static int soreceive_rcvoob(struct socket *so, struct uio *uio, int flags); static void filt_sordetach(struct knote *kn); static int filt_soread(struct knote *kn, long hint); static void filt_sowdetach(struct knote *kn); static int filt_sowrite(struct knote *kn, long hint); static int filt_solisten(struct knote *kn, long hint); static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id); fo_kqfilter_t soo_kqfilter; static struct filterops solisten_filtops = { .f_isfd = 1, .f_detach = filt_sordetach, .f_event = filt_solisten, }; static struct filterops soread_filtops = { .f_isfd = 1, .f_detach = filt_sordetach, .f_event = filt_soread, }; static struct filterops sowrite_filtops = { .f_isfd = 1, .f_detach = filt_sowdetach, .f_event = filt_sowrite, }; so_gen_t so_gencnt; /* generation count for sockets */ MALLOC_DEFINE(M_SONAME, "soname", "socket name"); MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); #define VNET_SO_ASSERT(so) \ VNET_ASSERT(curvnet != NULL, \ ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so))); VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]); #define V_socket_hhh VNET(socket_hhh) /* * Limit on the number of connections in the listen queue waiting * for accept(2). * NB: The original sysctl somaxconn is still available but hidden * to prevent confusion about the actual purpose of this number. */ static u_int somaxconn = SOMAXCONN; static int sysctl_somaxconn(SYSCTL_HANDLER_ARGS) { int error; int val; val = somaxconn; error = sysctl_handle_int(oidp, &val, 0, req); if (error || !req->newptr ) return (error); /* * The purpose of the UINT_MAX / 3 limit, is so that the formula * 3 * so_qlimit / 2 * below, will not overflow. */ if (val < 1 || val > UINT_MAX / 3) return (EINVAL); somaxconn = val; return (0); } SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, CTLTYPE_UINT | CTLFLAG_RW, 0, sizeof(int), sysctl_somaxconn, "I", "Maximum listen socket pending connection accept queue size"); SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP, 0, sizeof(int), sysctl_somaxconn, "I", "Maximum listen socket pending connection accept queue size (compat)"); static int numopensockets; SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD, &numopensockets, 0, "Number of open sockets"); /* * accept_mtx locks down per-socket fields relating to accept queues. See * socketvar.h for an annotation of the protected fields of struct socket. */ struct mtx accept_mtx; MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF); /* * so_global_mtx protects so_gencnt, numopensockets, and the per-socket * so_gencnt field. */ static struct mtx so_global_mtx; MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF); /* * General IPC sysctl name space, used by sockets and a variety of other IPC * types. */ SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC"); /* * Initialize the socket subsystem and set up the socket * memory allocator. */ static uma_zone_t socket_zone; int maxsockets; static void socket_zone_change(void *tag) { maxsockets = uma_zone_set_max(socket_zone, maxsockets); } static void socket_hhook_register(int subtype) { if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype, &V_socket_hhh[subtype], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) printf("%s: WARNING: unable to register hook\n", __func__); } static void socket_hhook_deregister(int subtype) { if (hhook_head_deregister(V_socket_hhh[subtype]) != 0) printf("%s: WARNING: unable to deregister hook\n", __func__); } static void socket_init(void *tag) { socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); maxsockets = uma_zone_set_max(socket_zone, maxsockets); uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached"); EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL, EVENTHANDLER_PRI_FIRST); } SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL); static void socket_vnet_init(const void *unused __unused) { int i; /* We expect a contiguous range */ for (i = 0; i <= HHOOK_SOCKET_LAST; i++) socket_hhook_register(i); } VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_vnet_init, NULL); static void socket_vnet_uninit(const void *unused __unused) { int i; for (i = 0; i <= HHOOK_SOCKET_LAST; i++) socket_hhook_deregister(i); } VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_vnet_uninit, NULL); /* * Initialise maxsockets. This SYSINIT must be run after * tunable_mbinit(). */ static void init_maxsockets(void *ignored) { TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets); maxsockets = imax(maxsockets, maxfiles); } SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL); /* * Sysctl to get and set the maximum global sockets limit. Notify protocols * of the change so that they can update their dependent limits as required. */ static int sysctl_maxsockets(SYSCTL_HANDLER_ARGS) { int error, newmaxsockets; newmaxsockets = maxsockets; error = sysctl_handle_int(oidp, &newmaxsockets, 0, req); if (error == 0 && req->newptr) { if (newmaxsockets > maxsockets && newmaxsockets <= maxfiles) { maxsockets = newmaxsockets; EVENTHANDLER_INVOKE(maxsockets_change); } else error = EINVAL; } return (error); } SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW, &maxsockets, 0, sysctl_maxsockets, "IU", "Maximum number of sockets available"); /* * Socket operation routines. These routines are called by the routines in * sys_socket.c or from a system process, and implement the semantics of * socket operations by switching out to the protocol specific routines. */ /* * Get a socket structure from our zone, and initialize it. Note that it * would probably be better to allocate socket and PCB at the same time, but * I'm not convinced that all the protocols can be easily modified to do * this. * * soalloc() returns a socket with a ref count of 0. */ static struct socket * soalloc(struct vnet *vnet) { struct socket *so; so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO); if (so == NULL) return (NULL); #ifdef MAC if (mac_socket_init(so, M_NOWAIT) != 0) { uma_zfree(socket_zone, so); return (NULL); } #endif if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) { uma_zfree(socket_zone, so); return (NULL); } SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd"); SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv"); sx_init(&so->so_snd.sb_sx, "so_snd_sx"); sx_init(&so->so_rcv.sb_sx, "so_rcv_sx"); TAILQ_INIT(&so->so_snd.sb_aiojobq); TAILQ_INIT(&so->so_rcv.sb_aiojobq); TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so); TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so); #ifdef VIMAGE VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p", __func__, __LINE__, so)); so->so_vnet = vnet; #endif /* We shouldn't need the so_global_mtx */ if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) { /* Do we need more comprehensive error returns? */ uma_zfree(socket_zone, so); return (NULL); } mtx_lock(&so_global_mtx); so->so_gencnt = ++so_gencnt; ++numopensockets; #ifdef VIMAGE vnet->vnet_sockcnt++; #endif mtx_unlock(&so_global_mtx); return (so); } /* * Free the storage associated with a socket at the socket layer, tear down * locks, labels, etc. All protocol state is assumed already to have been * torn down (and possibly never set up) by the caller. */ static void sodealloc(struct socket *so) { KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count)); KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL")); mtx_lock(&so_global_mtx); so->so_gencnt = ++so_gencnt; --numopensockets; /* Could be below, but faster here. */ #ifdef VIMAGE VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p", __func__, __LINE__, so)); so->so_vnet->vnet_sockcnt--; #endif mtx_unlock(&so_global_mtx); if (so->so_rcv.sb_hiwat) (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY); if (so->so_snd.sb_hiwat) (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 0, RLIM_INFINITY); /* remove accept filter if one is present. */ if (so->so_accf != NULL) do_setopt_accept_filter(so, NULL); #ifdef MAC mac_socket_destroy(so); #endif hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE); crfree(so->so_cred); khelp_destroy_osd(&so->osd); sx_destroy(&so->so_snd.sb_sx); sx_destroy(&so->so_rcv.sb_sx); SOCKBUF_LOCK_DESTROY(&so->so_snd); SOCKBUF_LOCK_DESTROY(&so->so_rcv); uma_zfree(socket_zone, so); } /* * socreate returns a socket with a ref count of 1. The socket should be * closed with soclose(). */ int socreate(int dom, struct socket **aso, int type, int proto, struct ucred *cred, struct thread *td) { struct protosw *prp; struct socket *so; int error; if (proto) prp = pffindproto(dom, proto, type); else prp = pffindtype(dom, type); if (prp == NULL) { /* No support for domain. */ if (pffinddomain(dom) == NULL) return (EAFNOSUPPORT); /* No support for socket type. */ if (proto == 0 && type != 0) return (EPROTOTYPE); return (EPROTONOSUPPORT); } if (prp->pr_usrreqs->pru_attach == NULL || prp->pr_usrreqs->pru_attach == pru_attach_notsupp) return (EPROTONOSUPPORT); if (prison_check_af(cred, prp->pr_domain->dom_family) != 0) return (EPROTONOSUPPORT); if (prp->pr_type != type) return (EPROTOTYPE); so = soalloc(CRED_TO_VNET(cred)); if (so == NULL) return (ENOBUFS); TAILQ_INIT(&so->so_incomp); TAILQ_INIT(&so->so_comp); so->so_type = type; so->so_cred = crhold(cred); if ((prp->pr_domain->dom_family == PF_INET) || (prp->pr_domain->dom_family == PF_INET6) || (prp->pr_domain->dom_family == PF_ROUTE)) so->so_fibnum = td->td_proc->p_fibnum; else so->so_fibnum = 0; so->so_proto = prp; #ifdef MAC mac_socket_create(cred, so); #endif knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv)); knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd)); so->so_count = 1; /* * Auto-sizing of socket buffers is managed by the protocols and * the appropriate flags must be set in the pru_attach function. */ CURVNET_SET(so->so_vnet); error = (*prp->pr_usrreqs->pru_attach)(so, proto, td); CURVNET_RESTORE(); if (error) { KASSERT(so->so_count == 1, ("socreate: so_count %d", so->so_count)); so->so_count = 0; sodealloc(so); return (error); } *aso = so; return (0); } #ifdef REGRESSION static int regression_sonewconn_earlytest = 1; SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW, ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test"); #endif /* * When an attempt at a new connection is noted on a socket which accepts * connections, sonewconn is called. If the connection is possible (subject * to space constraints, etc.) then we allocate a new structure, properly * linked into the data structure of the original socket, and return this. * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED. * * Note: the ref count on the socket is 0 on return. */ struct socket * sonewconn(struct socket *head, int connstatus) { static struct timeval lastover; static struct timeval overinterval = { 60, 0 }; static int overcount; struct socket *so; int over; ACCEPT_LOCK(); over = (head->so_qlen > 3 * head->so_qlimit / 2); ACCEPT_UNLOCK(); #ifdef REGRESSION if (regression_sonewconn_earlytest && over) { #else if (over) { #endif overcount++; if (ratecheck(&lastover, &overinterval)) { log(LOG_DEBUG, "%s: pcb %p: Listen queue overflow: " "%i already in queue awaiting acceptance " "(%d occurrences)\n", __func__, head->so_pcb, head->so_qlen, overcount); overcount = 0; } return (NULL); } VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p", __func__, __LINE__, head)); so = soalloc(head->so_vnet); if (so == NULL) { log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: " "limit reached or out of memory\n", __func__, head->so_pcb); return (NULL); } if ((head->so_options & SO_ACCEPTFILTER) != 0) connstatus = 0; so->so_head = head; so->so_type = head->so_type; so->so_options = head->so_options &~ SO_ACCEPTCONN; so->so_linger = head->so_linger; so->so_state = head->so_state | SS_NOFDREF; so->so_fibnum = head->so_fibnum; so->so_proto = head->so_proto; so->so_cred = crhold(head->so_cred); #ifdef MAC mac_socket_newconn(head, so); #endif knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv)); knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd)); VNET_SO_ASSERT(head); if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) { sodealloc(so); log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n", __func__, head->so_pcb); return (NULL); } if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) { sodealloc(so); log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n", __func__, head->so_pcb); return (NULL); } so->so_rcv.sb_lowat = head->so_rcv.sb_lowat; so->so_snd.sb_lowat = head->so_snd.sb_lowat; so->so_rcv.sb_timeo = head->so_rcv.sb_timeo; so->so_snd.sb_timeo = head->so_snd.sb_timeo; so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE; so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE; so->so_state |= connstatus; ACCEPT_LOCK(); /* * The accept socket may be tearing down but we just * won a race on the ACCEPT_LOCK. * However, if sctp_peeloff() is called on a 1-to-many * style socket, the SO_ACCEPTCONN doesn't need to be set. */ if (!(head->so_options & SO_ACCEPTCONN) && ((head->so_proto->pr_protocol != IPPROTO_SCTP) || (head->so_type != SOCK_SEQPACKET))) { SOCK_LOCK(so); so->so_head = NULL; sofree(so); /* NB: returns ACCEPT_UNLOCK'ed. */ return (NULL); } if (connstatus) { TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); so->so_qstate |= SQ_COMP; head->so_qlen++; } else { /* * Keep removing sockets from the head until there's room for * us to insert on the tail. In pre-locking revisions, this * was a simple if(), but as we could be racing with other * threads and soabort() requires dropping locks, we must * loop waiting for the condition to be true. */ while (head->so_incqlen > head->so_qlimit) { struct socket *sp; sp = TAILQ_FIRST(&head->so_incomp); TAILQ_REMOVE(&head->so_incomp, sp, so_list); head->so_incqlen--; sp->so_qstate &= ~SQ_INCOMP; sp->so_head = NULL; ACCEPT_UNLOCK(); soabort(sp); ACCEPT_LOCK(); } TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list); so->so_qstate |= SQ_INCOMP; head->so_incqlen++; } ACCEPT_UNLOCK(); if (connstatus) { sorwakeup(head); wakeup_one(&head->so_timeo); } return (so); } int sobind(struct socket *so, struct sockaddr *nam, struct thread *td) { int error; CURVNET_SET(so->so_vnet); error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td); CURVNET_RESTORE(); return (error); } int sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) { int error; CURVNET_SET(so->so_vnet); error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td); CURVNET_RESTORE(); return (error); } /* * solisten() transitions a socket from a non-listening state to a listening * state, but can also be used to update the listen queue depth on an * existing listen socket. The protocol will call back into the sockets * layer using solisten_proto_check() and solisten_proto() to check and set * socket-layer listen state. Call backs are used so that the protocol can * acquire both protocol and socket layer locks in whatever order is required * by the protocol. * * Protocol implementors are advised to hold the socket lock across the * socket-layer test and set to avoid races at the socket layer. */ int solisten(struct socket *so, int backlog, struct thread *td) { int error; CURVNET_SET(so->so_vnet); error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td); CURVNET_RESTORE(); return (error); } int solisten_proto_check(struct socket *so) { SOCK_LOCK_ASSERT(so); if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) return (EINVAL); return (0); } void solisten_proto(struct socket *so, int backlog) { SOCK_LOCK_ASSERT(so); if (backlog < 0 || backlog > somaxconn) backlog = somaxconn; so->so_qlimit = backlog; so->so_options |= SO_ACCEPTCONN; } /* * Evaluate the reference count and named references on a socket; if no * references remain, free it. This should be called whenever a reference is * released, such as in sorele(), but also when named reference flags are * cleared in socket or protocol code. * * sofree() will free the socket if: * * - There are no outstanding file descriptor references or related consumers * (so_count == 0). * * - The socket has been closed by user space, if ever open (SS_NOFDREF). * * - The protocol does not have an outstanding strong reference on the socket * (SS_PROTOREF). * * - The socket is not in a completed connection queue, so a process has been * notified that it is present. If it is removed, the user process may * block in accept() despite select() saying the socket was ready. */ void sofree(struct socket *so) { struct protosw *pr = so->so_proto; struct socket *head; ACCEPT_LOCK_ASSERT(); SOCK_LOCK_ASSERT(so); if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 || (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) { SOCK_UNLOCK(so); ACCEPT_UNLOCK(); return; } head = so->so_head; if (head != NULL) { KASSERT((so->so_qstate & SQ_COMP) != 0 || (so->so_qstate & SQ_INCOMP) != 0, ("sofree: so_head != NULL, but neither SQ_COMP nor " "SQ_INCOMP")); KASSERT((so->so_qstate & SQ_COMP) == 0 || (so->so_qstate & SQ_INCOMP) == 0, ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP")); TAILQ_REMOVE(&head->so_incomp, so, so_list); head->so_incqlen--; so->so_qstate &= ~SQ_INCOMP; so->so_head = NULL; } KASSERT((so->so_qstate & SQ_COMP) == 0 && (so->so_qstate & SQ_INCOMP) == 0, ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)", so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP)); if (so->so_options & SO_ACCEPTCONN) { KASSERT((TAILQ_EMPTY(&so->so_comp)), ("sofree: so_comp populated")); KASSERT((TAILQ_EMPTY(&so->so_incomp)), ("sofree: so_incomp populated")); } SOCK_UNLOCK(so); ACCEPT_UNLOCK(); VNET_SO_ASSERT(so); if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) (*pr->pr_domain->dom_dispose)(so); if (pr->pr_usrreqs->pru_detach != NULL) (*pr->pr_usrreqs->pru_detach)(so); /* * From this point on, we assume that no other references to this * socket exist anywhere else in the stack. Therefore, no locks need * to be acquired or held. * * We used to do a lot of socket buffer and socket locking here, as * well as invoke sorflush() and perform wakeups. The direct call to * dom_dispose() and sbrelease_internal() are an inlining of what was * necessary from sorflush(). * * Notice that the socket buffer and kqueue state are torn down * before calling pru_detach. This means that protocols shold not * assume they can perform socket wakeups, etc, in their detach code. */ sbdestroy(&so->so_snd, so); sbdestroy(&so->so_rcv, so); seldrain(&so->so_snd.sb_sel); seldrain(&so->so_rcv.sb_sel); knlist_destroy(&so->so_rcv.sb_sel.si_note); knlist_destroy(&so->so_snd.sb_sel.si_note); sodealloc(so); } /* * Close a socket on last file table reference removal. Initiate disconnect * if connected. Free socket when disconnect complete. * * This function will sorele() the socket. Note that soclose() may be called * prior to the ref count reaching zero. The actual socket structure will * not be freed until the ref count reaches zero. */ int soclose(struct socket *so) { int error = 0; KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter")); CURVNET_SET(so->so_vnet); funsetown(&so->so_sigio); if (so->so_state & SS_ISCONNECTED) { if ((so->so_state & SS_ISDISCONNECTING) == 0) { error = sodisconnect(so); if (error) { if (error == ENOTCONN) error = 0; goto drop; } } if (so->so_options & SO_LINGER) { if ((so->so_state & SS_ISDISCONNECTING) && (so->so_state & SS_NBIO)) goto drop; while (so->so_state & SS_ISCONNECTED) { error = tsleep(&so->so_timeo, PSOCK | PCATCH, "soclos", so->so_linger * hz); if (error) break; } } } drop: if (so->so_proto->pr_usrreqs->pru_close != NULL) (*so->so_proto->pr_usrreqs->pru_close)(so); ACCEPT_LOCK(); if (so->so_options & SO_ACCEPTCONN) { struct socket *sp; /* * Prevent new additions to the accept queues due * to ACCEPT_LOCK races while we are draining them. */ so->so_options &= ~SO_ACCEPTCONN; while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) { TAILQ_REMOVE(&so->so_incomp, sp, so_list); so->so_incqlen--; sp->so_qstate &= ~SQ_INCOMP; sp->so_head = NULL; ACCEPT_UNLOCK(); soabort(sp); ACCEPT_LOCK(); } while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) { TAILQ_REMOVE(&so->so_comp, sp, so_list); so->so_qlen--; sp->so_qstate &= ~SQ_COMP; sp->so_head = NULL; ACCEPT_UNLOCK(); soabort(sp); ACCEPT_LOCK(); } KASSERT((TAILQ_EMPTY(&so->so_comp)), ("%s: so_comp populated", __func__)); KASSERT((TAILQ_EMPTY(&so->so_incomp)), ("%s: so_incomp populated", __func__)); } SOCK_LOCK(so); KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF")); so->so_state |= SS_NOFDREF; sorele(so); /* NB: Returns with ACCEPT_UNLOCK(). */ CURVNET_RESTORE(); return (error); } /* * soabort() is used to abruptly tear down a connection, such as when a * resource limit is reached (listen queue depth exceeded), or if a listen * socket is closed while there are sockets waiting to be accepted. * * This interface is tricky, because it is called on an unreferenced socket, * and must be called only by a thread that has actually removed the socket * from the listen queue it was on, or races with other threads are risked. * * This interface will call into the protocol code, so must not be called * with any socket locks held. Protocols do call it while holding their own * recursible protocol mutexes, but this is something that should be subject * to review in the future. */ void soabort(struct socket *so) { /* * In as much as is possible, assert that no references to this * socket are held. This is not quite the same as asserting that the * current thread is responsible for arranging for no references, but * is as close as we can get for now. */ KASSERT(so->so_count == 0, ("soabort: so_count")); KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF")); KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF")); KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP")); KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP")); VNET_SO_ASSERT(so); if (so->so_proto->pr_usrreqs->pru_abort != NULL) (*so->so_proto->pr_usrreqs->pru_abort)(so); ACCEPT_LOCK(); SOCK_LOCK(so); sofree(so); } int soaccept(struct socket *so, struct sockaddr **nam) { int error; SOCK_LOCK(so); KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF")); so->so_state &= ~SS_NOFDREF; SOCK_UNLOCK(so); CURVNET_SET(so->so_vnet); error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam); CURVNET_RESTORE(); return (error); } int soconnect(struct socket *so, struct sockaddr *nam, struct thread *td) { return (soconnectat(AT_FDCWD, so, nam, td)); } int soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) { int error; if (so->so_options & SO_ACCEPTCONN) return (EOPNOTSUPP); CURVNET_SET(so->so_vnet); /* * If protocol is connection-based, can only connect once. * Otherwise, if connected, try to disconnect first. This allows * user to disconnect by connecting to, e.g., a null address. */ if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && ((so->so_proto->pr_flags & PR_CONNREQUIRED) || (error = sodisconnect(so)))) { error = EISCONN; } else { /* * Prevent accumulated error from previous connection from * biting us. */ so->so_error = 0; if (fd == AT_FDCWD) { error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td); } else { error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd, so, nam, td); } } CURVNET_RESTORE(); return (error); } int soconnect2(struct socket *so1, struct socket *so2) { int error; CURVNET_SET(so1->so_vnet); error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2); CURVNET_RESTORE(); return (error); } int sodisconnect(struct socket *so) { int error; if ((so->so_state & SS_ISCONNECTED) == 0) return (ENOTCONN); if (so->so_state & SS_ISDISCONNECTING) return (EALREADY); VNET_SO_ASSERT(so); error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so); return (error); } #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT) int sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, struct mbuf *top, struct mbuf *control, int flags, struct thread *td) { long space; ssize_t resid; int clen = 0, error, dontroute; KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM")); KASSERT(so->so_proto->pr_flags & PR_ATOMIC, ("sosend_dgram: !PR_ATOMIC")); if (uio != NULL) resid = uio->uio_resid; else resid = top->m_pkthdr.len; /* * In theory resid should be unsigned. However, space must be * signed, as it might be less than 0 if we over-committed, and we * must use a signed comparison of space and resid. On the other * hand, a negative resid causes us to loop sending 0-length * segments to the protocol. */ if (resid < 0) { error = EINVAL; goto out; } dontroute = (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0; if (td != NULL) td->td_ru.ru_msgsnd++; if (control != NULL) clen = control->m_len; SOCKBUF_LOCK(&so->so_snd); if (so->so_snd.sb_state & SBS_CANTSENDMORE) { SOCKBUF_UNLOCK(&so->so_snd); error = EPIPE; goto out; } if (so->so_error) { error = so->so_error; so->so_error = 0; SOCKBUF_UNLOCK(&so->so_snd); goto out; } if ((so->so_state & SS_ISCONNECTED) == 0) { /* * `sendto' and `sendmsg' is allowed on a connection-based * socket if it supports implied connect. Return ENOTCONN if * not connected and no address is supplied. */ if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { if ((so->so_state & SS_ISCONFIRMING) == 0 && !(resid == 0 && clen != 0)) { SOCKBUF_UNLOCK(&so->so_snd); error = ENOTCONN; goto out; } } else if (addr == NULL) { if (so->so_proto->pr_flags & PR_CONNREQUIRED) error = ENOTCONN; else error = EDESTADDRREQ; SOCKBUF_UNLOCK(&so->so_snd); goto out; } } /* * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a * problem and need fixing. */ space = sbspace(&so->so_snd); if (flags & MSG_OOB) space += 1024; space -= clen; SOCKBUF_UNLOCK(&so->so_snd); if (resid > space) { error = EMSGSIZE; goto out; } if (uio == NULL) { resid = 0; if (flags & MSG_EOR) top->m_flags |= M_EOR; } else { /* * Copy the data from userland into a mbuf chain. * If no data is to be copied in, a single empty mbuf * is returned. */ top = m_uiotombuf(uio, M_WAITOK, space, max_hdr, (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0))); if (top == NULL) { error = EFAULT; /* only possible error */ goto out; } space -= resid - uio->uio_resid; resid = uio->uio_resid; } KASSERT(resid == 0, ("sosend_dgram: resid != 0")); /* * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock * than with. */ if (dontroute) { SOCK_LOCK(so); so->so_options |= SO_DONTROUTE; SOCK_UNLOCK(so); } /* * XXX all the SBS_CANTSENDMORE checks previously done could be out * of date. We could have received a reset packet in an interrupt or * maybe we slept while doing page faults in uiomove() etc. We could * probably recheck again inside the locking protection here, but * there are probably other places that this also happens. We must * rethink this. */ VNET_SO_ASSERT(so); error = (*so->so_proto->pr_usrreqs->pru_send)(so, (flags & MSG_OOB) ? PRUS_OOB : /* * If the user set MSG_EOF, the protocol understands this flag and * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND. */ ((flags & MSG_EOF) && (so->so_proto->pr_flags & PR_IMPLOPCL) && (resid <= 0)) ? PRUS_EOF : /* If there is more to send set PRUS_MORETOCOME */ (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, top, addr, control, td); if (dontroute) { SOCK_LOCK(so); so->so_options &= ~SO_DONTROUTE; SOCK_UNLOCK(so); } clen = 0; control = NULL; top = NULL; out: if (top != NULL) m_freem(top); if (control != NULL) m_freem(control); return (error); } /* * Send on a socket. If send must go all at once and message is larger than * send buffering, then hard error. Lock against other senders. If must go * all at once and not enough room now, then inform user that this would * block and do nothing. Otherwise, if nonblocking, send as much as * possible. The data to be sent is described by "uio" if nonzero, otherwise * by the mbuf chain "top" (which must be null if uio is not). Data provided * in mbuf chain must be small enough to send all at once. * * Returns nonzero on error, timeout or signal; callers must check for short * counts if EINTR/ERESTART are returned. Data and control buffers are freed * on return. */ int sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio, struct mbuf *top, struct mbuf *control, int flags, struct thread *td) { long space; ssize_t resid; int clen = 0, error, dontroute; int atomic = sosendallatonce(so) || top; if (uio != NULL) resid = uio->uio_resid; else resid = top->m_pkthdr.len; /* * In theory resid should be unsigned. However, space must be * signed, as it might be less than 0 if we over-committed, and we * must use a signed comparison of space and resid. On the other * hand, a negative resid causes us to loop sending 0-length * segments to the protocol. * * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM * type sockets since that's an error. */ if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { error = EINVAL; goto out; } dontroute = (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && (so->so_proto->pr_flags & PR_ATOMIC); if (td != NULL) td->td_ru.ru_msgsnd++; if (control != NULL) clen = control->m_len; error = sblock(&so->so_snd, SBLOCKWAIT(flags)); if (error) goto out; restart: do { SOCKBUF_LOCK(&so->so_snd); if (so->so_snd.sb_state & SBS_CANTSENDMORE) { SOCKBUF_UNLOCK(&so->so_snd); error = EPIPE; goto release; } if (so->so_error) { error = so->so_error; so->so_error = 0; SOCKBUF_UNLOCK(&so->so_snd); goto release; } if ((so->so_state & SS_ISCONNECTED) == 0) { /* * `sendto' and `sendmsg' is allowed on a connection- * based socket if it supports implied connect. * Return ENOTCONN if not connected and no address is * supplied. */ if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { if ((so->so_state & SS_ISCONFIRMING) == 0 && !(resid == 0 && clen != 0)) { SOCKBUF_UNLOCK(&so->so_snd); error = ENOTCONN; goto release; } } else if (addr == NULL) { SOCKBUF_UNLOCK(&so->so_snd); if (so->so_proto->pr_flags & PR_CONNREQUIRED) error = ENOTCONN; else error = EDESTADDRREQ; goto release; } } space = sbspace(&so->so_snd); if (flags & MSG_OOB) space += 1024; if ((atomic && resid > so->so_snd.sb_hiwat) || clen > so->so_snd.sb_hiwat) { SOCKBUF_UNLOCK(&so->so_snd); error = EMSGSIZE; goto release; } if (space < resid + clen && (atomic || space < so->so_snd.sb_lowat || space < clen)) { if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) { SOCKBUF_UNLOCK(&so->so_snd); error = EWOULDBLOCK; goto release; } error = sbwait(&so->so_snd); SOCKBUF_UNLOCK(&so->so_snd); if (error) goto release; goto restart; } SOCKBUF_UNLOCK(&so->so_snd); space -= clen; do { if (uio == NULL) { resid = 0; if (flags & MSG_EOR) top->m_flags |= M_EOR; } else { /* * Copy the data from userland into a mbuf * chain. If resid is 0, which can happen * only if we have control to send, then * a single empty mbuf is returned. This * is a workaround to prevent protocol send * methods to panic. */ top = m_uiotombuf(uio, M_WAITOK, space, (atomic ? max_hdr : 0), (atomic ? M_PKTHDR : 0) | ((flags & MSG_EOR) ? M_EOR : 0)); if (top == NULL) { error = EFAULT; /* only possible error */ goto release; } space -= resid - uio->uio_resid; resid = uio->uio_resid; } if (dontroute) { SOCK_LOCK(so); so->so_options |= SO_DONTROUTE; SOCK_UNLOCK(so); } /* * XXX all the SBS_CANTSENDMORE checks previously * done could be out of date. We could have received * a reset packet in an interrupt or maybe we slept * while doing page faults in uiomove() etc. We * could probably recheck again inside the locking * protection here, but there are probably other * places that this also happens. We must rethink * this. */ VNET_SO_ASSERT(so); error = (*so->so_proto->pr_usrreqs->pru_send)(so, (flags & MSG_OOB) ? PRUS_OOB : /* * If the user set MSG_EOF, the protocol understands * this flag and nothing left to send then use * PRU_SEND_EOF instead of PRU_SEND. */ ((flags & MSG_EOF) && (so->so_proto->pr_flags & PR_IMPLOPCL) && (resid <= 0)) ? PRUS_EOF : /* If there is more to send set PRUS_MORETOCOME. */ (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, top, addr, control, td); if (dontroute) { SOCK_LOCK(so); so->so_options &= ~SO_DONTROUTE; SOCK_UNLOCK(so); } clen = 0; control = NULL; top = NULL; if (error) goto release; } while (resid && space > 0); } while (resid); release: sbunlock(&so->so_snd); out: if (top != NULL) m_freem(top); if (control != NULL) m_freem(control); return (error); } int sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, struct mbuf *top, struct mbuf *control, int flags, struct thread *td) { int error; CURVNET_SET(so->so_vnet); error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top, control, flags, td); CURVNET_RESTORE(); return (error); } /* * The part of soreceive() that implements reading non-inline out-of-band * data from a socket. For more complete comments, see soreceive(), from * which this code originated. * * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is * unable to return an mbuf chain to the caller. */ static int soreceive_rcvoob(struct socket *so, struct uio *uio, int flags) { struct protosw *pr = so->so_proto; struct mbuf *m; int error; KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0")); VNET_SO_ASSERT(so); m = m_get(M_WAITOK, MT_DATA); error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK); if (error) goto bad; do { error = uiomove(mtod(m, void *), (int) min(uio->uio_resid, m->m_len), uio); m = m_free(m); } while (uio->uio_resid && error == 0 && m); bad: if (m != NULL) m_freem(m); return (error); } /* * Following replacement or removal of the first mbuf on the first mbuf chain * of a socket buffer, push necessary state changes back into the socket * buffer so that other consumers see the values consistently. 'nextrecord' * is the callers locally stored value of the original value of * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes. * NOTE: 'nextrecord' may be NULL. */ static __inline void sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord) { SOCKBUF_LOCK_ASSERT(sb); /* * First, update for the new value of nextrecord. If necessary, make * it the first record. */ if (sb->sb_mb != NULL) sb->sb_mb->m_nextpkt = nextrecord; else sb->sb_mb = nextrecord; /* * Now update any dependent socket buffer fields to reflect the new * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the * addition of a second clause that takes care of the case where * sb_mb has been updated, but remains the last record. */ if (sb->sb_mb == NULL) { sb->sb_mbtail = NULL; sb->sb_lastrecord = NULL; } else if (sb->sb_mb->m_nextpkt == NULL) sb->sb_lastrecord = sb->sb_mb; } /* * Implement receive operations on a socket. We depend on the way that * records are added to the sockbuf by sbappend. In particular, each record * (mbufs linked through m_next) must begin with an address if the protocol * so specifies, followed by an optional mbuf or mbufs containing ancillary * data, and then zero or more mbufs of data. In order to allow parallelism * between network receive and copying to user space, as well as avoid * sleeping with a mutex held, we release the socket buffer mutex during the * user space copy. Although the sockbuf is locked, new data may still be * appended, and thus we must maintain consistency of the sockbuf during that * time. * * The caller may receive the data as a single mbuf chain by supplying an * mbuf **mp0 for use in returning the chain. The uio is then used only for * the count in uio_resid. */ int soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) { struct mbuf *m, **mp; int flags, error, offset; ssize_t len; struct protosw *pr = so->so_proto; struct mbuf *nextrecord; int moff, type = 0; ssize_t orig_resid = uio->uio_resid; mp = mp0; if (psa != NULL) *psa = NULL; if (controlp != NULL) *controlp = NULL; if (flagsp != NULL) flags = *flagsp &~ MSG_EOR; else flags = 0; if (flags & MSG_OOB) return (soreceive_rcvoob(so, uio, flags)); if (mp != NULL) *mp = NULL; if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING) && uio->uio_resid) { VNET_SO_ASSERT(so); (*pr->pr_usrreqs->pru_rcvd)(so, 0); } error = sblock(&so->so_rcv, SBLOCKWAIT(flags)); if (error) return (error); restart: SOCKBUF_LOCK(&so->so_rcv); m = so->so_rcv.sb_mb; /* * If we have less data than requested, block awaiting more (subject * to any timeout) if: * 1. the current count is less than the low water mark, or * 2. MSG_DONTWAIT is not set */ if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && sbavail(&so->so_rcv) < uio->uio_resid) && sbavail(&so->so_rcv) < so->so_rcv.sb_lowat && m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { KASSERT(m != NULL || !sbavail(&so->so_rcv), ("receive: m == %p sbavail == %u", m, sbavail(&so->so_rcv))); if (so->so_error) { if (m != NULL) goto dontblock; error = so->so_error; if ((flags & MSG_PEEK) == 0) so->so_error = 0; SOCKBUF_UNLOCK(&so->so_rcv); goto release; } SOCKBUF_LOCK_ASSERT(&so->so_rcv); if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { if (m == NULL) { SOCKBUF_UNLOCK(&so->so_rcv); goto release; } else goto dontblock; } for (; m != NULL; m = m->m_next) if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { m = so->so_rcv.sb_mb; goto dontblock; } if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && (so->so_proto->pr_flags & PR_CONNREQUIRED)) { SOCKBUF_UNLOCK(&so->so_rcv); error = ENOTCONN; goto release; } if (uio->uio_resid == 0) { SOCKBUF_UNLOCK(&so->so_rcv); goto release; } if ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO))) { SOCKBUF_UNLOCK(&so->so_rcv); error = EWOULDBLOCK; goto release; } SBLASTRECORDCHK(&so->so_rcv); SBLASTMBUFCHK(&so->so_rcv); error = sbwait(&so->so_rcv); SOCKBUF_UNLOCK(&so->so_rcv); if (error) goto release; goto restart; } dontblock: /* * From this point onward, we maintain 'nextrecord' as a cache of the * pointer to the next record in the socket buffer. We must keep the * various socket buffer pointers and local stack versions of the * pointers in sync, pushing out modifications before dropping the * socket buffer mutex, and re-reading them when picking it up. * * Otherwise, we will race with the network stack appending new data * or records onto the socket buffer by using inconsistent/stale * versions of the field, possibly resulting in socket buffer * corruption. * * By holding the high-level sblock(), we prevent simultaneous * readers from pulling off the front of the socket buffer. */ SOCKBUF_LOCK_ASSERT(&so->so_rcv); if (uio->uio_td) uio->uio_td->td_ru.ru_msgrcv++; KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb")); SBLASTRECORDCHK(&so->so_rcv); SBLASTMBUFCHK(&so->so_rcv); nextrecord = m->m_nextpkt; if (pr->pr_flags & PR_ADDR) { KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); orig_resid = 0; if (psa != NULL) *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_NOWAIT); if (flags & MSG_PEEK) { m = m->m_next; } else { sbfree(&so->so_rcv, m); so->so_rcv.sb_mb = m_free(m); m = so->so_rcv.sb_mb; sockbuf_pushsync(&so->so_rcv, nextrecord); } } /* * Process one or more MT_CONTROL mbufs present before any data mbufs * in the first mbuf chain on the socket buffer. If MSG_PEEK, we * just copy the data; if !MSG_PEEK, we call into the protocol to * perform externalization (or freeing if controlp == NULL). */ if (m != NULL && m->m_type == MT_CONTROL) { struct mbuf *cm = NULL, *cmn; struct mbuf **cme = &cm; do { if (flags & MSG_PEEK) { if (controlp != NULL) { *controlp = m_copym(m, 0, m->m_len, M_NOWAIT); controlp = &(*controlp)->m_next; } m = m->m_next; } else { sbfree(&so->so_rcv, m); so->so_rcv.sb_mb = m->m_next; m->m_next = NULL; *cme = m; cme = &(*cme)->m_next; m = so->so_rcv.sb_mb; } } while (m != NULL && m->m_type == MT_CONTROL); if ((flags & MSG_PEEK) == 0) sockbuf_pushsync(&so->so_rcv, nextrecord); while (cm != NULL) { cmn = cm->m_next; cm->m_next = NULL; if (pr->pr_domain->dom_externalize != NULL) { SOCKBUF_UNLOCK(&so->so_rcv); VNET_SO_ASSERT(so); error = (*pr->pr_domain->dom_externalize) (cm, controlp, flags); SOCKBUF_LOCK(&so->so_rcv); } else if (controlp != NULL) *controlp = cm; else m_freem(cm); if (controlp != NULL) { orig_resid = 0; while (*controlp != NULL) controlp = &(*controlp)->m_next; } cm = cmn; } if (m != NULL) nextrecord = so->so_rcv.sb_mb->m_nextpkt; else nextrecord = so->so_rcv.sb_mb; orig_resid = 0; } if (m != NULL) { if ((flags & MSG_PEEK) == 0) { KASSERT(m->m_nextpkt == nextrecord, ("soreceive: post-control, nextrecord !sync")); if (nextrecord == NULL) { KASSERT(so->so_rcv.sb_mb == m, ("soreceive: post-control, sb_mb!=m")); KASSERT(so->so_rcv.sb_lastrecord == m, ("soreceive: post-control, lastrecord!=m")); } } type = m->m_type; if (type == MT_OOBDATA) flags |= MSG_OOB; } else { if ((flags & MSG_PEEK) == 0) { KASSERT(so->so_rcv.sb_mb == nextrecord, ("soreceive: sb_mb != nextrecord")); if (so->so_rcv.sb_mb == NULL) { KASSERT(so->so_rcv.sb_lastrecord == NULL, ("soreceive: sb_lastercord != NULL")); } } } SOCKBUF_LOCK_ASSERT(&so->so_rcv); SBLASTRECORDCHK(&so->so_rcv); SBLASTMBUFCHK(&so->so_rcv); /* * Now continue to read any data mbufs off of the head of the socket * buffer until the read request is satisfied. Note that 'type' is * used to store the type of any mbuf reads that have happened so far * such that soreceive() can stop reading if the type changes, which * causes soreceive() to return only one of regular data and inline * out-of-band data in a single socket receive operation. */ moff = 0; offset = 0; while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0 && error == 0) { /* * If the type of mbuf has changed since the last mbuf * examined ('type'), end the receive operation. */ SOCKBUF_LOCK_ASSERT(&so->so_rcv); if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) { if (type != m->m_type) break; } else if (type == MT_OOBDATA) break; else KASSERT(m->m_type == MT_DATA, ("m->m_type == %d", m->m_type)); so->so_rcv.sb_state &= ~SBS_RCVATMARK; len = uio->uio_resid; if (so->so_oobmark && len > so->so_oobmark - offset) len = so->so_oobmark - offset; if (len > m->m_len - moff) len = m->m_len - moff; /* * If mp is set, just pass back the mbufs. Otherwise copy * them out via the uio, then free. Sockbuf must be * consistent here (points to current mbuf, it points to next * record) when we drop priority; we must note any additions * to the sockbuf when we block interrupts again. */ if (mp == NULL) { SOCKBUF_LOCK_ASSERT(&so->so_rcv); SBLASTRECORDCHK(&so->so_rcv); SBLASTMBUFCHK(&so->so_rcv); SOCKBUF_UNLOCK(&so->so_rcv); error = uiomove(mtod(m, char *) + moff, (int)len, uio); SOCKBUF_LOCK(&so->so_rcv); if (error) { /* * The MT_SONAME mbuf has already been removed * from the record, so it is necessary to * remove the data mbufs, if any, to preserve * the invariant in the case of PR_ADDR that * requires MT_SONAME mbufs at the head of * each record. */ if (m && pr->pr_flags & PR_ATOMIC && ((flags & MSG_PEEK) == 0)) (void)sbdroprecord_locked(&so->so_rcv); SOCKBUF_UNLOCK(&so->so_rcv); goto release; } } else uio->uio_resid -= len; SOCKBUF_LOCK_ASSERT(&so->so_rcv); if (len == m->m_len - moff) { if (m->m_flags & M_EOR) flags |= MSG_EOR; if (flags & MSG_PEEK) { m = m->m_next; moff = 0; } else { nextrecord = m->m_nextpkt; sbfree(&so->so_rcv, m); if (mp != NULL) { m->m_nextpkt = NULL; *mp = m; mp = &m->m_next; so->so_rcv.sb_mb = m = m->m_next; *mp = NULL; } else { so->so_rcv.sb_mb = m_free(m); m = so->so_rcv.sb_mb; } sockbuf_pushsync(&so->so_rcv, nextrecord); SBLASTRECORDCHK(&so->so_rcv); SBLASTMBUFCHK(&so->so_rcv); } } else { if (flags & MSG_PEEK) moff += len; else { if (mp != NULL) { if (flags & MSG_DONTWAIT) { *mp = m_copym(m, 0, len, M_NOWAIT); if (*mp == NULL) { /* * m_copym() couldn't * allocate an mbuf. * Adjust uio_resid back * (it was adjusted * down by len bytes, * which we didn't end * up "copying" over). */ uio->uio_resid += len; break; } } else { SOCKBUF_UNLOCK(&so->so_rcv); *mp = m_copym(m, 0, len, M_WAITOK); SOCKBUF_LOCK(&so->so_rcv); } } sbcut_locked(&so->so_rcv, len); } } SOCKBUF_LOCK_ASSERT(&so->so_rcv); if (so->so_oobmark) { if ((flags & MSG_PEEK) == 0) { so->so_oobmark -= len; if (so->so_oobmark == 0) { so->so_rcv.sb_state |= SBS_RCVATMARK; break; } } else { offset += len; if (offset == so->so_oobmark) break; } } if (flags & MSG_EOR) break; /* * If the MSG_WAITALL flag is set (for non-atomic socket), we * must not quit until "uio->uio_resid == 0" or an error * termination. If a signal/timeout occurs, return with a * short count but without error. Keep sockbuf locked * against other readers. */ while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && !sosendallatonce(so) && nextrecord == NULL) { SOCKBUF_LOCK_ASSERT(&so->so_rcv); if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE) break; /* * Notify the protocol that some data has been * drained before blocking. */ if (pr->pr_flags & PR_WANTRCVD) { SOCKBUF_UNLOCK(&so->so_rcv); VNET_SO_ASSERT(so); (*pr->pr_usrreqs->pru_rcvd)(so, flags); SOCKBUF_LOCK(&so->so_rcv); } SBLASTRECORDCHK(&so->so_rcv); SBLASTMBUFCHK(&so->so_rcv); /* * We could receive some data while was notifying * the protocol. Skip blocking in this case. */ if (so->so_rcv.sb_mb == NULL) { error = sbwait(&so->so_rcv); if (error) { SOCKBUF_UNLOCK(&so->so_rcv); goto release; } } m = so->so_rcv.sb_mb; if (m != NULL) nextrecord = m->m_nextpkt; } } SOCKBUF_LOCK_ASSERT(&so->so_rcv); if (m != NULL && pr->pr_flags & PR_ATOMIC) { flags |= MSG_TRUNC; if ((flags & MSG_PEEK) == 0) (void) sbdroprecord_locked(&so->so_rcv); } if ((flags & MSG_PEEK) == 0) { if (m == NULL) { /* * First part is an inline SB_EMPTY_FIXUP(). Second * part makes sure sb_lastrecord is up-to-date if * there is still data in the socket buffer. */ so->so_rcv.sb_mb = nextrecord; if (so->so_rcv.sb_mb == NULL) { so->so_rcv.sb_mbtail = NULL; so->so_rcv.sb_lastrecord = NULL; } else if (nextrecord->m_nextpkt == NULL) so->so_rcv.sb_lastrecord = nextrecord; } SBLASTRECORDCHK(&so->so_rcv); SBLASTMBUFCHK(&so->so_rcv); /* * If soreceive() is being done from the socket callback, * then don't need to generate ACK to peer to update window, * since ACK will be generated on return to TCP. */ if (!(flags & MSG_SOCALLBCK) && (pr->pr_flags & PR_WANTRCVD)) { SOCKBUF_UNLOCK(&so->so_rcv); VNET_SO_ASSERT(so); (*pr->pr_usrreqs->pru_rcvd)(so, flags); SOCKBUF_LOCK(&so->so_rcv); } } SOCKBUF_LOCK_ASSERT(&so->so_rcv); if (orig_resid == uio->uio_resid && orig_resid && (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) { SOCKBUF_UNLOCK(&so->so_rcv); goto restart; } SOCKBUF_UNLOCK(&so->so_rcv); if (flagsp != NULL) *flagsp |= flags; release: sbunlock(&so->so_rcv); return (error); } /* * Optimized version of soreceive() for stream (TCP) sockets. * XXXAO: (MSG_WAITALL | MSG_PEEK) isn't properly handled. */ int soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) { int len = 0, error = 0, flags, oresid; struct sockbuf *sb; struct mbuf *m, *n = NULL; /* We only do stream sockets. */ if (so->so_type != SOCK_STREAM) return (EINVAL); if (psa != NULL) *psa = NULL; if (controlp != NULL) return (EINVAL); if (flagsp != NULL) flags = *flagsp &~ MSG_EOR; else flags = 0; if (flags & MSG_OOB) return (soreceive_rcvoob(so, uio, flags)); if (mp0 != NULL) *mp0 = NULL; sb = &so->so_rcv; /* Prevent other readers from entering the socket. */ error = sblock(sb, SBLOCKWAIT(flags)); if (error) goto out; SOCKBUF_LOCK(sb); /* Easy one, no space to copyout anything. */ if (uio->uio_resid == 0) { error = EINVAL; goto out; } oresid = uio->uio_resid; /* We will never ever get anything unless we are or were connected. */ if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) { error = ENOTCONN; goto out; } restart: SOCKBUF_LOCK_ASSERT(&so->so_rcv); /* Abort if socket has reported problems. */ if (so->so_error) { if (sbavail(sb) > 0) goto deliver; if (oresid > uio->uio_resid) goto out; error = so->so_error; if (!(flags & MSG_PEEK)) so->so_error = 0; goto out; } /* Door is closed. Deliver what is left, if any. */ if (sb->sb_state & SBS_CANTRCVMORE) { if (sbavail(sb) > 0) goto deliver; else goto out; } /* Socket buffer is empty and we shall not block. */ if (sbavail(sb) == 0 && ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) { error = EAGAIN; goto out; } /* Socket buffer got some data that we shall deliver now. */ if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) && ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)) || sbavail(sb) >= sb->sb_lowat || sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat) ) { goto deliver; } /* On MSG_WAITALL we must wait until all data or error arrives. */ if ((flags & MSG_WAITALL) && (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat)) goto deliver; /* * Wait and block until (more) data comes in. * NB: Drops the sockbuf lock during wait. */ error = sbwait(sb); if (error) goto out; goto restart; deliver: SOCKBUF_LOCK_ASSERT(&so->so_rcv); KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__)); KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__)); /* Statistics. */ if (uio->uio_td) uio->uio_td->td_ru.ru_msgrcv++; /* Fill uio until full or current end of socket buffer is reached. */ len = min(uio->uio_resid, sbavail(sb)); if (mp0 != NULL) { /* Dequeue as many mbufs as possible. */ if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) { if (*mp0 == NULL) *mp0 = sb->sb_mb; else m_cat(*mp0, sb->sb_mb); for (m = sb->sb_mb; m != NULL && m->m_len <= len; m = m->m_next) { KASSERT(!(m->m_flags & M_NOTAVAIL), ("%s: m %p not available", __func__, m)); len -= m->m_len; uio->uio_resid -= m->m_len; sbfree(sb, m); n = m; } n->m_next = NULL; sb->sb_mb = m; sb->sb_lastrecord = sb->sb_mb; if (sb->sb_mb == NULL) SB_EMPTY_FIXUP(sb); } /* Copy the remainder. */ if (len > 0) { KASSERT(sb->sb_mb != NULL, ("%s: len > 0 && sb->sb_mb empty", __func__)); m = m_copym(sb->sb_mb, 0, len, M_NOWAIT); if (m == NULL) len = 0; /* Don't flush data from sockbuf. */ else uio->uio_resid -= len; if (*mp0 != NULL) m_cat(*mp0, m); else *mp0 = m; if (*mp0 == NULL) { error = ENOBUFS; goto out; } } } else { /* NB: Must unlock socket buffer as uiomove may sleep. */ SOCKBUF_UNLOCK(sb); error = m_mbuftouio(uio, sb->sb_mb, len); SOCKBUF_LOCK(sb); if (error) goto out; } SBLASTRECORDCHK(sb); SBLASTMBUFCHK(sb); /* * Remove the delivered data from the socket buffer unless we * were only peeking. */ if (!(flags & MSG_PEEK)) { if (len > 0) sbdrop_locked(sb, len); /* Notify protocol that we drained some data. */ if ((so->so_proto->pr_flags & PR_WANTRCVD) && (((flags & MSG_WAITALL) && uio->uio_resid > 0) || !(flags & MSG_SOCALLBCK))) { SOCKBUF_UNLOCK(sb); VNET_SO_ASSERT(so); (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags); SOCKBUF_LOCK(sb); } } /* * For MSG_WAITALL we may have to loop again and wait for * more data to come in. */ if ((flags & MSG_WAITALL) && uio->uio_resid > 0) goto restart; out: SOCKBUF_LOCK_ASSERT(sb); SBLASTRECORDCHK(sb); SBLASTMBUFCHK(sb); SOCKBUF_UNLOCK(sb); sbunlock(sb); return (error); } /* * Optimized version of soreceive() for simple datagram cases from userspace. * Unlike in the stream case, we're able to drop a datagram if copyout() * fails, and because we handle datagrams atomically, we don't need to use a * sleep lock to prevent I/O interlacing. */ int soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) { struct mbuf *m, *m2; int flags, error; ssize_t len; struct protosw *pr = so->so_proto; struct mbuf *nextrecord; if (psa != NULL) *psa = NULL; if (controlp != NULL) *controlp = NULL; if (flagsp != NULL) flags = *flagsp &~ MSG_EOR; else flags = 0; /* * For any complicated cases, fall back to the full * soreceive_generic(). */ if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB)) return (soreceive_generic(so, psa, uio, mp0, controlp, flagsp)); /* * Enforce restrictions on use. */ KASSERT((pr->pr_flags & PR_WANTRCVD) == 0, ("soreceive_dgram: wantrcvd")); KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic")); KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0, ("soreceive_dgram: SBS_RCVATMARK")); KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0, ("soreceive_dgram: P_CONNREQUIRED")); /* * Loop blocking while waiting for a datagram. */ SOCKBUF_LOCK(&so->so_rcv); while ((m = so->so_rcv.sb_mb) == NULL) { KASSERT(sbavail(&so->so_rcv) == 0, ("soreceive_dgram: sb_mb NULL but sbavail %u", sbavail(&so->so_rcv))); if (so->so_error) { error = so->so_error; so->so_error = 0; SOCKBUF_UNLOCK(&so->so_rcv); return (error); } if (so->so_rcv.sb_state & SBS_CANTRCVMORE || uio->uio_resid == 0) { SOCKBUF_UNLOCK(&so->so_rcv); return (0); } if ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO))) { SOCKBUF_UNLOCK(&so->so_rcv); return (EWOULDBLOCK); } SBLASTRECORDCHK(&so->so_rcv); SBLASTMBUFCHK(&so->so_rcv); error = sbwait(&so->so_rcv); if (error) { SOCKBUF_UNLOCK(&so->so_rcv); return (error); } } SOCKBUF_LOCK_ASSERT(&so->so_rcv); if (uio->uio_td) uio->uio_td->td_ru.ru_msgrcv++; SBLASTRECORDCHK(&so->so_rcv); SBLASTMBUFCHK(&so->so_rcv); nextrecord = m->m_nextpkt; if (nextrecord == NULL) { KASSERT(so->so_rcv.sb_lastrecord == m, ("soreceive_dgram: lastrecord != m")); } KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord, ("soreceive_dgram: m_nextpkt != nextrecord")); /* * Pull 'm' and its chain off the front of the packet queue. */ so->so_rcv.sb_mb = NULL; sockbuf_pushsync(&so->so_rcv, nextrecord); /* * Walk 'm's chain and free that many bytes from the socket buffer. */ for (m2 = m; m2 != NULL; m2 = m2->m_next) sbfree(&so->so_rcv, m2); /* * Do a few last checks before we let go of the lock. */ SBLASTRECORDCHK(&so->so_rcv); SBLASTMBUFCHK(&so->so_rcv); SOCKBUF_UNLOCK(&so->so_rcv); if (pr->pr_flags & PR_ADDR) { KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); if (psa != NULL) *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_NOWAIT); m = m_free(m); } if (m == NULL) { /* XXXRW: Can this happen? */ return (0); } /* * Packet to copyout() is now in 'm' and it is disconnected from the * queue. * * Process one or more MT_CONTROL mbufs present before any data mbufs * in the first mbuf chain on the socket buffer. We call into the * protocol to perform externalization (or freeing if controlp == * NULL). In some cases there can be only MT_CONTROL mbufs without * MT_DATA mbufs. */ if (m->m_type == MT_CONTROL) { struct mbuf *cm = NULL, *cmn; struct mbuf **cme = &cm; do { m2 = m->m_next; m->m_next = NULL; *cme = m; cme = &(*cme)->m_next; m = m2; } while (m != NULL && m->m_type == MT_CONTROL); while (cm != NULL) { cmn = cm->m_next; cm->m_next = NULL; if (pr->pr_domain->dom_externalize != NULL) { error = (*pr->pr_domain->dom_externalize) (cm, controlp, flags); } else if (controlp != NULL) *controlp = cm; else m_freem(cm); if (controlp != NULL) { while (*controlp != NULL) controlp = &(*controlp)->m_next; } cm = cmn; } } KASSERT(m == NULL || m->m_type == MT_DATA, ("soreceive_dgram: !data")); while (m != NULL && uio->uio_resid > 0) { len = uio->uio_resid; if (len > m->m_len) len = m->m_len; error = uiomove(mtod(m, char *), (int)len, uio); if (error) { m_freem(m); return (error); } if (len == m->m_len) m = m_free(m); else { m->m_data += len; m->m_len -= len; } } if (m != NULL) { flags |= MSG_TRUNC; m_freem(m); } if (flagsp != NULL) *flagsp |= flags; return (0); } int soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) { int error; CURVNET_SET(so->so_vnet); error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0, controlp, flagsp)); CURVNET_RESTORE(); return (error); } int soshutdown(struct socket *so, int how) { struct protosw *pr = so->so_proto; int error; if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) return (EINVAL); if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) return (ENOTCONN); CURVNET_SET(so->so_vnet); if (pr->pr_usrreqs->pru_flush != NULL) (*pr->pr_usrreqs->pru_flush)(so, how); if (how != SHUT_WR) sorflush(so); if (how != SHUT_RD) { error = (*pr->pr_usrreqs->pru_shutdown)(so); wakeup(&so->so_timeo); CURVNET_RESTORE(); return (error); } wakeup(&so->so_timeo); CURVNET_RESTORE(); return (0); } void sorflush(struct socket *so) { struct sockbuf *sb = &so->so_rcv; struct protosw *pr = so->so_proto; struct socket aso; VNET_SO_ASSERT(so); /* * In order to avoid calling dom_dispose with the socket buffer mutex * held, and in order to generally avoid holding the lock for a long * time, we make a copy of the socket buffer and clear the original * (except locks, state). The new socket buffer copy won't have * initialized locks so we can only call routines that won't use or * assert those locks. * * Dislodge threads currently blocked in receive and wait to acquire * a lock against other simultaneous readers before clearing the * socket buffer. Don't let our acquire be interrupted by a signal * despite any existing socket disposition on interruptable waiting. */ socantrcvmore(so); (void) sblock(sb, SBL_WAIT | SBL_NOINTR); /* * Invalidate/clear most of the sockbuf structure, but leave selinfo * and mutex data unchanged. */ SOCKBUF_LOCK(sb); bzero(&aso, sizeof(aso)); aso.so_pcb = so->so_pcb; bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero, sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); bzero(&sb->sb_startzero, sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); SOCKBUF_UNLOCK(sb); sbunlock(sb); /* * Dispose of special rights and flush the copied socket. Don't call * any unsafe routines (that rely on locks being initialized) on aso. */ if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) (*pr->pr_domain->dom_dispose)(&aso); sbrelease_internal(&aso.so_rcv, so); } /* * Wrapper for Socket established helper hook. * Parameters: socket, context of the hook point, hook id. */ static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id) { struct socket_hhook_data hhook_data = { .so = so, .hctx = hctx, .m = NULL, .status = 0 }; CURVNET_SET(so->so_vnet); HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd); CURVNET_RESTORE(); /* Ugly but needed, since hhooks return void for now */ return (hhook_data.status); } /* * Perhaps this routine, and sooptcopyout(), below, ought to come in an * additional variant to handle the case where the option value needs to be * some kind of integer, but not a specific size. In addition to their use * here, these functions are also called by the protocol-level pr_ctloutput() * routines. */ int sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) { size_t valsize; /* * If the user gives us more than we wanted, we ignore it, but if we * don't get the minimum length the caller wants, we return EINVAL. * On success, sopt->sopt_valsize is set to however much we actually * retrieved. */ if ((valsize = sopt->sopt_valsize) < minlen) return EINVAL; if (valsize > len) sopt->sopt_valsize = valsize = len; if (sopt->sopt_td != NULL) return (copyin(sopt->sopt_val, buf, valsize)); bcopy(sopt->sopt_val, buf, valsize); return (0); } /* * Kernel version of setsockopt(2). * * XXX: optlen is size_t, not socklen_t */ int so_setsockopt(struct socket *so, int level, int optname, void *optval, size_t optlen) { struct sockopt sopt; sopt.sopt_level = level; sopt.sopt_name = optname; sopt.sopt_dir = SOPT_SET; sopt.sopt_val = optval; sopt.sopt_valsize = optlen; sopt.sopt_td = NULL; return (sosetopt(so, &sopt)); } int sosetopt(struct socket *so, struct sockopt *sopt) { int error, optval; struct linger l; struct timeval tv; sbintime_t val; uint32_t val32; #ifdef MAC struct mac extmac; #endif CURVNET_SET(so->so_vnet); error = 0; if (sopt->sopt_level != SOL_SOCKET) { if (so->so_proto->pr_ctloutput != NULL) { error = (*so->so_proto->pr_ctloutput)(so, sopt); CURVNET_RESTORE(); return (error); } error = ENOPROTOOPT; } else { switch (sopt->sopt_name) { case SO_ACCEPTFILTER: error = do_setopt_accept_filter(so, sopt); if (error) goto bad; break; case SO_LINGER: error = sooptcopyin(sopt, &l, sizeof l, sizeof l); if (error) goto bad; SOCK_LOCK(so); so->so_linger = l.l_linger; if (l.l_onoff) so->so_options |= SO_LINGER; else so->so_options &= ~SO_LINGER; SOCK_UNLOCK(so); break; case SO_DEBUG: case SO_KEEPALIVE: case SO_DONTROUTE: case SO_USELOOPBACK: case SO_BROADCAST: case SO_REUSEADDR: case SO_REUSEPORT: case SO_OOBINLINE: case SO_TIMESTAMP: case SO_BINTIME: case SO_NOSIGPIPE: case SO_NO_DDP: case SO_NO_OFFLOAD: error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); if (error) goto bad; SOCK_LOCK(so); if (optval) so->so_options |= sopt->sopt_name; else so->so_options &= ~sopt->sopt_name; SOCK_UNLOCK(so); break; case SO_SETFIB: error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); if (error) goto bad; if (optval < 0 || optval >= rt_numfibs) { error = EINVAL; goto bad; } if (((so->so_proto->pr_domain->dom_family == PF_INET) || (so->so_proto->pr_domain->dom_family == PF_INET6) || (so->so_proto->pr_domain->dom_family == PF_ROUTE))) so->so_fibnum = optval; else so->so_fibnum = 0; break; case SO_USER_COOKIE: error = sooptcopyin(sopt, &val32, sizeof val32, sizeof val32); if (error) goto bad; so->so_user_cookie = val32; break; case SO_SNDBUF: case SO_RCVBUF: case SO_SNDLOWAT: case SO_RCVLOWAT: error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); if (error) goto bad; /* * Values < 1 make no sense for any of these options, * so disallow them. */ if (optval < 1) { error = EINVAL; goto bad; } switch (sopt->sopt_name) { case SO_SNDBUF: case SO_RCVBUF: if (sbreserve(sopt->sopt_name == SO_SNDBUF ? &so->so_snd : &so->so_rcv, (u_long)optval, so, curthread) == 0) { error = ENOBUFS; goto bad; } (sopt->sopt_name == SO_SNDBUF ? &so->so_snd : &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE; break; /* * Make sure the low-water is never greater than the * high-water. */ case SO_SNDLOWAT: SOCKBUF_LOCK(&so->so_snd); so->so_snd.sb_lowat = (optval > so->so_snd.sb_hiwat) ? so->so_snd.sb_hiwat : optval; SOCKBUF_UNLOCK(&so->so_snd); break; case SO_RCVLOWAT: SOCKBUF_LOCK(&so->so_rcv); so->so_rcv.sb_lowat = (optval > so->so_rcv.sb_hiwat) ? so->so_rcv.sb_hiwat : optval; SOCKBUF_UNLOCK(&so->so_rcv); break; } break; case SO_SNDTIMEO: case SO_RCVTIMEO: #ifdef COMPAT_FREEBSD32 if (SV_CURPROC_FLAG(SV_ILP32)) { struct timeval32 tv32; error = sooptcopyin(sopt, &tv32, sizeof tv32, sizeof tv32); CP(tv32, tv, tv_sec); CP(tv32, tv, tv_usec); } else #endif error = sooptcopyin(sopt, &tv, sizeof tv, sizeof tv); if (error) goto bad; if (tv.tv_sec < 0 || tv.tv_usec < 0 || tv.tv_usec >= 1000000) { error = EDOM; goto bad; } if (tv.tv_sec > INT32_MAX) val = SBT_MAX; else val = tvtosbt(tv); switch (sopt->sopt_name) { case SO_SNDTIMEO: so->so_snd.sb_timeo = val; break; case SO_RCVTIMEO: so->so_rcv.sb_timeo = val; break; } break; case SO_LABEL: #ifdef MAC error = sooptcopyin(sopt, &extmac, sizeof extmac, sizeof extmac); if (error) goto bad; error = mac_setsockopt_label(sopt->sopt_td->td_ucred, so, &extmac); #else error = EOPNOTSUPP; #endif break; default: if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) error = hhook_run_socket(so, sopt, HHOOK_SOCKET_OPT); else error = ENOPROTOOPT; break; } if (error == 0 && so->so_proto->pr_ctloutput != NULL) (void)(*so->so_proto->pr_ctloutput)(so, sopt); } bad: CURVNET_RESTORE(); return (error); } /* * Helper routine for getsockopt. */ int sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) { int error; size_t valsize; error = 0; /* * Documented get behavior is that we always return a value, possibly * truncated to fit in the user's buffer. Traditional behavior is * that we always tell the user precisely how much we copied, rather * than something useful like the total amount we had available for * her. Note that this interface is not idempotent; the entire * answer must be generated ahead of time. */ valsize = min(len, sopt->sopt_valsize); sopt->sopt_valsize = valsize; if (sopt->sopt_val != NULL) { if (sopt->sopt_td != NULL) error = copyout(buf, sopt->sopt_val, valsize); else bcopy(buf, sopt->sopt_val, valsize); } return (error); } int sogetopt(struct socket *so, struct sockopt *sopt) { int error, optval; struct linger l; struct timeval tv; #ifdef MAC struct mac extmac; #endif CURVNET_SET(so->so_vnet); error = 0; if (sopt->sopt_level != SOL_SOCKET) { if (so->so_proto->pr_ctloutput != NULL) error = (*so->so_proto->pr_ctloutput)(so, sopt); else error = ENOPROTOOPT; CURVNET_RESTORE(); return (error); } else { switch (sopt->sopt_name) { case SO_ACCEPTFILTER: error = do_getopt_accept_filter(so, sopt); break; case SO_LINGER: SOCK_LOCK(so); l.l_onoff = so->so_options & SO_LINGER; l.l_linger = so->so_linger; SOCK_UNLOCK(so); error = sooptcopyout(sopt, &l, sizeof l); break; case SO_USELOOPBACK: case SO_DONTROUTE: case SO_DEBUG: case SO_KEEPALIVE: case SO_REUSEADDR: case SO_REUSEPORT: case SO_BROADCAST: case SO_OOBINLINE: case SO_ACCEPTCONN: case SO_TIMESTAMP: case SO_BINTIME: case SO_NOSIGPIPE: optval = so->so_options & sopt->sopt_name; integer: error = sooptcopyout(sopt, &optval, sizeof optval); break; case SO_TYPE: optval = so->so_type; goto integer; case SO_PROTOCOL: optval = so->so_proto->pr_protocol; goto integer; case SO_ERROR: SOCK_LOCK(so); optval = so->so_error; so->so_error = 0; SOCK_UNLOCK(so); goto integer; case SO_SNDBUF: optval = so->so_snd.sb_hiwat; goto integer; case SO_RCVBUF: optval = so->so_rcv.sb_hiwat; goto integer; case SO_SNDLOWAT: optval = so->so_snd.sb_lowat; goto integer; case SO_RCVLOWAT: optval = so->so_rcv.sb_lowat; goto integer; case SO_SNDTIMEO: case SO_RCVTIMEO: tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ? so->so_snd.sb_timeo : so->so_rcv.sb_timeo); #ifdef COMPAT_FREEBSD32 if (SV_CURPROC_FLAG(SV_ILP32)) { struct timeval32 tv32; CP(tv, tv32, tv_sec); CP(tv, tv32, tv_usec); error = sooptcopyout(sopt, &tv32, sizeof tv32); } else #endif error = sooptcopyout(sopt, &tv, sizeof tv); break; case SO_LABEL: #ifdef MAC error = sooptcopyin(sopt, &extmac, sizeof(extmac), sizeof(extmac)); if (error) goto bad; error = mac_getsockopt_label(sopt->sopt_td->td_ucred, so, &extmac); if (error) goto bad; error = sooptcopyout(sopt, &extmac, sizeof extmac); #else error = EOPNOTSUPP; #endif break; case SO_PEERLABEL: #ifdef MAC error = sooptcopyin(sopt, &extmac, sizeof(extmac), sizeof(extmac)); if (error) goto bad; error = mac_getsockopt_peerlabel( sopt->sopt_td->td_ucred, so, &extmac); if (error) goto bad; error = sooptcopyout(sopt, &extmac, sizeof extmac); #else error = EOPNOTSUPP; #endif break; case SO_LISTENQLIMIT: optval = so->so_qlimit; goto integer; case SO_LISTENQLEN: optval = so->so_qlen; goto integer; case SO_LISTENINCQLEN: optval = so->so_incqlen; goto integer; default: if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) error = hhook_run_socket(so, sopt, HHOOK_SOCKET_OPT); else error = ENOPROTOOPT; break; } } #ifdef MAC bad: #endif CURVNET_RESTORE(); return (error); } int soopt_getm(struct sockopt *sopt, struct mbuf **mp) { struct mbuf *m, *m_prev; int sopt_size = sopt->sopt_valsize; MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); if (m == NULL) return ENOBUFS; if (sopt_size > MLEN) { MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT); if ((m->m_flags & M_EXT) == 0) { m_free(m); return ENOBUFS; } m->m_len = min(MCLBYTES, sopt_size); } else { m->m_len = min(MLEN, sopt_size); } sopt_size -= m->m_len; *mp = m; m_prev = m; while (sopt_size) { MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); if (m == NULL) { m_freem(*mp); return ENOBUFS; } if (sopt_size > MLEN) { MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK : M_NOWAIT); if ((m->m_flags & M_EXT) == 0) { m_freem(m); m_freem(*mp); return ENOBUFS; } m->m_len = min(MCLBYTES, sopt_size); } else { m->m_len = min(MLEN, sopt_size); } sopt_size -= m->m_len; m_prev->m_next = m; m_prev = m; } return (0); } int soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) { struct mbuf *m0 = m; if (sopt->sopt_val == NULL) return (0); while (m != NULL && sopt->sopt_valsize >= m->m_len) { if (sopt->sopt_td != NULL) { int error; error = copyin(sopt->sopt_val, mtod(m, char *), m->m_len); if (error != 0) { m_freem(m0); return(error); } } else bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); sopt->sopt_valsize -= m->m_len; sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; m = m->m_next; } if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ panic("ip6_sooptmcopyin"); return (0); } int soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) { struct mbuf *m0 = m; size_t valsize = 0; if (sopt->sopt_val == NULL) return (0); while (m != NULL && sopt->sopt_valsize >= m->m_len) { if (sopt->sopt_td != NULL) { int error; error = copyout(mtod(m, char *), sopt->sopt_val, m->m_len); if (error != 0) { m_freem(m0); return(error); } } else bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); sopt->sopt_valsize -= m->m_len; sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; valsize += m->m_len; m = m->m_next; } if (m != NULL) { /* enough soopt buffer should be given from user-land */ m_freem(m0); return(EINVAL); } sopt->sopt_valsize = valsize; return (0); } /* * sohasoutofband(): protocol notifies socket layer of the arrival of new * out-of-band data, which will then notify socket consumers. */ void sohasoutofband(struct socket *so) { if (so->so_sigio != NULL) pgsigio(&so->so_sigio, SIGURG, 0); selwakeuppri(&so->so_rcv.sb_sel, PSOCK); } int sopoll(struct socket *so, int events, struct ucred *active_cred, struct thread *td) { /* * We do not need to set or assert curvnet as long as everyone uses * sopoll_generic(). */ return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred, td)); } int sopoll_generic(struct socket *so, int events, struct ucred *active_cred, struct thread *td) { int revents = 0; SOCKBUF_LOCK(&so->so_snd); SOCKBUF_LOCK(&so->so_rcv); if (events & (POLLIN | POLLRDNORM)) if (soreadabledata(so)) revents |= events & (POLLIN | POLLRDNORM); if (events & (POLLOUT | POLLWRNORM)) if (sowriteable(so)) revents |= events & (POLLOUT | POLLWRNORM); if (events & (POLLPRI | POLLRDBAND)) if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK)) revents |= events & (POLLPRI | POLLRDBAND); if ((events & POLLINIGNEOF) == 0) { if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { revents |= events & (POLLIN | POLLRDNORM); if (so->so_snd.sb_state & SBS_CANTSENDMORE) revents |= POLLHUP; } } if (revents == 0) { if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) { selrecord(td, &so->so_rcv.sb_sel); so->so_rcv.sb_flags |= SB_SEL; } if (events & (POLLOUT | POLLWRNORM)) { selrecord(td, &so->so_snd.sb_sel); so->so_snd.sb_flags |= SB_SEL; } } SOCKBUF_UNLOCK(&so->so_rcv); SOCKBUF_UNLOCK(&so->so_snd); return (revents); } int soo_kqfilter(struct file *fp, struct knote *kn) { struct socket *so = kn->kn_fp->f_data; struct sockbuf *sb; switch (kn->kn_filter) { case EVFILT_READ: if (so->so_options & SO_ACCEPTCONN) kn->kn_fop = &solisten_filtops; else kn->kn_fop = &soread_filtops; sb = &so->so_rcv; break; case EVFILT_WRITE: kn->kn_fop = &sowrite_filtops; sb = &so->so_snd; break; default: return (EINVAL); } SOCKBUF_LOCK(sb); knlist_add(&sb->sb_sel.si_note, kn, 1); sb->sb_flags |= SB_KNOTE; SOCKBUF_UNLOCK(sb); return (0); } /* * Some routines that return EOPNOTSUPP for entry points that are not * supported by a protocol. Fill in as needed. */ int pru_accept_notsupp(struct socket *so, struct sockaddr **nam) { return EOPNOTSUPP; } int pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job) { return EOPNOTSUPP; } int pru_attach_notsupp(struct socket *so, int proto, struct thread *td) { return EOPNOTSUPP; } int pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) { return EOPNOTSUPP; } int pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) { return EOPNOTSUPP; } int pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) { return EOPNOTSUPP; } int pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) { return EOPNOTSUPP; } int pru_connect2_notsupp(struct socket *so1, struct socket *so2) { return EOPNOTSUPP; } int pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data, struct ifnet *ifp, struct thread *td) { return EOPNOTSUPP; } int pru_disconnect_notsupp(struct socket *so) { return EOPNOTSUPP; } int pru_listen_notsupp(struct socket *so, int backlog, struct thread *td) { return EOPNOTSUPP; } int pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam) { return EOPNOTSUPP; } int pru_rcvd_notsupp(struct socket *so, int flags) { return EOPNOTSUPP; } int pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) { return EOPNOTSUPP; } int pru_send_notsupp(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, struct mbuf *control, struct thread *td) { return EOPNOTSUPP; } int pru_ready_notsupp(struct socket *so, struct mbuf *m, int count) { return (EOPNOTSUPP); } /* * This isn't really a ``null'' operation, but it's the default one and * doesn't do anything destructive. */ int pru_sense_null(struct socket *so, struct stat *sb) { sb->st_blksize = so->so_snd.sb_hiwat; return 0; } int pru_shutdown_notsupp(struct socket *so) { return EOPNOTSUPP; } int pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam) { return EOPNOTSUPP; } int pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio, struct mbuf *top, struct mbuf *control, int flags, struct thread *td) { return EOPNOTSUPP; } int pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr, struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) { return EOPNOTSUPP; } int pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred, struct thread *td) { return EOPNOTSUPP; } static void filt_sordetach(struct knote *kn) { struct socket *so = kn->kn_fp->f_data; SOCKBUF_LOCK(&so->so_rcv); knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1); if (knlist_empty(&so->so_rcv.sb_sel.si_note)) so->so_rcv.sb_flags &= ~SB_KNOTE; SOCKBUF_UNLOCK(&so->so_rcv); } /*ARGSUSED*/ static int filt_soread(struct knote *kn, long hint) { struct socket *so; so = kn->kn_fp->f_data; SOCKBUF_LOCK_ASSERT(&so->so_rcv); kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl; if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { kn->kn_flags |= EV_EOF; kn->kn_fflags = so->so_error; return (1); } else if (so->so_error) /* temporary udp error */ return (1); if (kn->kn_sfflags & NOTE_LOWAT) { if (kn->kn_data >= kn->kn_sdata) return 1; } else { if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat) return 1; } /* This hook returning non-zero indicates an event, not error */ return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD)); } static void filt_sowdetach(struct knote *kn) { struct socket *so = kn->kn_fp->f_data; SOCKBUF_LOCK(&so->so_snd); knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1); if (knlist_empty(&so->so_snd.sb_sel.si_note)) so->so_snd.sb_flags &= ~SB_KNOTE; SOCKBUF_UNLOCK(&so->so_snd); } /*ARGSUSED*/ static int filt_sowrite(struct knote *kn, long hint) { struct socket *so; so = kn->kn_fp->f_data; SOCKBUF_LOCK_ASSERT(&so->so_snd); kn->kn_data = sbspace(&so->so_snd); hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE); if (so->so_snd.sb_state & SBS_CANTSENDMORE) { kn->kn_flags |= EV_EOF; kn->kn_fflags = so->so_error; return (1); } else if (so->so_error) /* temporary udp error */ return (1); else if (((so->so_state & SS_ISCONNECTED) == 0) && (so->so_proto->pr_flags & PR_CONNREQUIRED)) return (0); else if (kn->kn_sfflags & NOTE_LOWAT) return (kn->kn_data >= kn->kn_sdata); else return (kn->kn_data >= so->so_snd.sb_lowat); } /*ARGSUSED*/ static int filt_solisten(struct knote *kn, long hint) { struct socket *so = kn->kn_fp->f_data; kn->kn_data = so->so_qlen; return (!TAILQ_EMPTY(&so->so_comp)); } int socheckuid(struct socket *so, uid_t uid) { if (so == NULL) return (EPERM); if (so->so_cred->cr_uid != uid) return (EPERM); return (0); } /* * These functions are used by protocols to notify the socket layer (and its * consumers) of state changes in the sockets driven by protocol-side events. */ /* * Procedures to manipulate state flags of socket and do appropriate wakeups. * * Normal sequence from the active (originating) side is that * soisconnecting() is called during processing of connect() call, resulting * in an eventual call to soisconnected() if/when the connection is * established. When the connection is torn down soisdisconnecting() is * called during processing of disconnect() call, and soisdisconnected() is * called when the connection to the peer is totally severed. The semantics * of these routines are such that connectionless protocols can call * soisconnected() and soisdisconnected() only, bypassing the in-progress * calls when setting up a ``connection'' takes no time. * * From the passive side, a socket is created with two queues of sockets: * so_incomp for connections in progress and so_comp for connections already * made and awaiting user acceptance. As a protocol is preparing incoming * connections, it creates a socket structure queued on so_incomp by calling * sonewconn(). When the connection is established, soisconnected() is * called, and transfers the socket structure to so_comp, making it available * to accept(). * * If a socket is closed with sockets on either so_incomp or so_comp, these * sockets are dropped. * * If higher-level protocols are implemented in the kernel, the wakeups done * here will sometimes cause software-interrupt process scheduling. */ void soisconnecting(struct socket *so) { SOCK_LOCK(so); so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); so->so_state |= SS_ISCONNECTING; SOCK_UNLOCK(so); } void soisconnected(struct socket *so) { struct socket *head; int ret; restart: ACCEPT_LOCK(); SOCK_LOCK(so); so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); so->so_state |= SS_ISCONNECTED; head = so->so_head; if (head != NULL && (so->so_qstate & SQ_INCOMP)) { if ((so->so_options & SO_ACCEPTFILTER) == 0) { SOCK_UNLOCK(so); TAILQ_REMOVE(&head->so_incomp, so, so_list); head->so_incqlen--; so->so_qstate &= ~SQ_INCOMP; TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); head->so_qlen++; so->so_qstate |= SQ_COMP; ACCEPT_UNLOCK(); sorwakeup(head); wakeup_one(&head->so_timeo); } else { ACCEPT_UNLOCK(); soupcall_set(so, SO_RCV, head->so_accf->so_accept_filter->accf_callback, head->so_accf->so_accept_filter_arg); so->so_options &= ~SO_ACCEPTFILTER; ret = head->so_accf->so_accept_filter->accf_callback(so, head->so_accf->so_accept_filter_arg, M_NOWAIT); if (ret == SU_ISCONNECTED) soupcall_clear(so, SO_RCV); SOCK_UNLOCK(so); if (ret == SU_ISCONNECTED) goto restart; } return; } SOCK_UNLOCK(so); ACCEPT_UNLOCK(); wakeup(&so->so_timeo); sorwakeup(so); sowwakeup(so); } void soisdisconnecting(struct socket *so) { /* * Note: This code assumes that SOCK_LOCK(so) and * SOCKBUF_LOCK(&so->so_rcv) are the same. */ SOCKBUF_LOCK(&so->so_rcv); so->so_state &= ~SS_ISCONNECTING; so->so_state |= SS_ISDISCONNECTING; socantrcvmore_locked(so); SOCKBUF_LOCK(&so->so_snd); socantsendmore_locked(so); wakeup(&so->so_timeo); } void soisdisconnected(struct socket *so) { /* * Note: This code assumes that SOCK_LOCK(so) and * SOCKBUF_LOCK(&so->so_rcv) are the same. */ SOCKBUF_LOCK(&so->so_rcv); so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); so->so_state |= SS_ISDISCONNECTED; socantrcvmore_locked(so); SOCKBUF_LOCK(&so->so_snd); sbdrop_locked(&so->so_snd, sbused(&so->so_snd)); socantsendmore_locked(so); wakeup(&so->so_timeo); } /* * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. */ struct sockaddr * sodupsockaddr(const struct sockaddr *sa, int mflags) { struct sockaddr *sa2; sa2 = malloc(sa->sa_len, M_SONAME, mflags); if (sa2) bcopy(sa, sa2, sa->sa_len); return sa2; } /* * Register per-socket buffer upcalls. */ void soupcall_set(struct socket *so, int which, int (*func)(struct socket *, void *, int), void *arg) { struct sockbuf *sb; switch (which) { case SO_RCV: sb = &so->so_rcv; break; case SO_SND: sb = &so->so_snd; break; default: panic("soupcall_set: bad which"); } SOCKBUF_LOCK_ASSERT(sb); #if 0 /* XXX: accf_http actually wants to do this on purpose. */ KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall")); #endif sb->sb_upcall = func; sb->sb_upcallarg = arg; sb->sb_flags |= SB_UPCALL; } void soupcall_clear(struct socket *so, int which) { struct sockbuf *sb; switch (which) { case SO_RCV: sb = &so->so_rcv; break; case SO_SND: sb = &so->so_snd; break; default: panic("soupcall_clear: bad which"); } SOCKBUF_LOCK_ASSERT(sb); KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear")); sb->sb_upcall = NULL; sb->sb_upcallarg = NULL; sb->sb_flags &= ~SB_UPCALL; } /* * Create an external-format (``xsocket'') structure using the information in * the kernel-format socket structure pointed to by so. This is done to * reduce the spew of irrelevant information over this interface, to isolate * user code from changes in the kernel structure, and potentially to provide * information-hiding if we decide that some of this information should be * hidden from users. */ void sotoxsocket(struct socket *so, struct xsocket *xso) { xso->xso_len = sizeof *xso; xso->xso_so = so; xso->so_type = so->so_type; xso->so_options = so->so_options; xso->so_linger = so->so_linger; xso->so_state = so->so_state; xso->so_pcb = so->so_pcb; xso->xso_protocol = so->so_proto->pr_protocol; xso->xso_family = so->so_proto->pr_domain->dom_family; xso->so_qlen = so->so_qlen; xso->so_incqlen = so->so_incqlen; xso->so_qlimit = so->so_qlimit; xso->so_timeo = so->so_timeo; xso->so_error = so->so_error; xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; xso->so_oobmark = so->so_oobmark; sbtoxsockbuf(&so->so_snd, &xso->so_snd); sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); xso->so_uid = so->so_cred->cr_uid; } /* * Socket accessor functions to provide external consumers with * a safe interface to socket state * */ void so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *), void *arg) { TAILQ_FOREACH(so, &so->so_comp, so_list) func(so, arg); } struct sockbuf * so_sockbuf_rcv(struct socket *so) { return (&so->so_rcv); } struct sockbuf * so_sockbuf_snd(struct socket *so) { return (&so->so_snd); } int so_state_get(const struct socket *so) { return (so->so_state); } void so_state_set(struct socket *so, int val) { so->so_state = val; } int so_options_get(const struct socket *so) { return (so->so_options); } void so_options_set(struct socket *so, int val) { so->so_options = val; } int so_error_get(const struct socket *so) { return (so->so_error); } void so_error_set(struct socket *so, int val) { so->so_error = val; } int so_linger_get(const struct socket *so) { return (so->so_linger); } void so_linger_set(struct socket *so, int val) { so->so_linger = val; } struct protosw * so_protosw_get(const struct socket *so) { return (so->so_proto); } void so_protosw_set(struct socket *so, struct protosw *val) { so->so_proto = val; } void so_sorwakeup(struct socket *so) { sorwakeup(so); } void so_sowwakeup(struct socket *so) { sowwakeup(so); } void so_sorwakeup_locked(struct socket *so) { sorwakeup_locked(so); } void so_sowwakeup_locked(struct socket *so) { sowwakeup_locked(so); } void so_lock(struct socket *so) { SOCK_LOCK(so); } void so_unlock(struct socket *so) { SOCK_UNLOCK(so); } Index: head/sys/kern/uipc_syscalls.c =================================================================== --- head/sys/kern/uipc_syscalls.c (revision 305831) +++ head/sys/kern/uipc_syscalls.c (revision 305832) @@ -1,1575 +1,1575 @@ /*- * Copyright (c) 1982, 1986, 1989, 1990, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)uipc_syscalls.c 8.4 (Berkeley) 2/21/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_capsicum.h" #include "opt_inet.h" #include "opt_inet6.h" #include "opt_compat.h" #include "opt_ktrace.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KTRACE #include #endif #ifdef COMPAT_FREEBSD32 #include #endif #include #include #include /* * Flags for accept1() and kern_accept4(), in addition to SOCK_CLOEXEC * and SOCK_NONBLOCK. */ #define ACCEPT4_INHERIT 0x1 #define ACCEPT4_COMPAT 0x2 static int sendit(struct thread *td, int s, struct msghdr *mp, int flags); static int recvit(struct thread *td, int s, struct msghdr *mp, void *namelenp); static int accept1(struct thread *td, int s, struct sockaddr *uname, socklen_t *anamelen, int flags); static int getsockname1(struct thread *td, struct getsockname_args *uap, int compat); static int getpeername1(struct thread *td, struct getpeername_args *uap, int compat); static int sockargs(struct mbuf **, char *, socklen_t, int); /* * Convert a user file descriptor to a kernel file entry and check if required * capability rights are present. * A reference on the file entry is held upon returning. */ int getsock_cap(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp, u_int *fflagp) { struct file *fp; int error; error = fget_unlocked(td->td_proc->p_fd, fd, rightsp, &fp, NULL); if (error != 0) return (error); if (fp->f_type != DTYPE_SOCKET) { fdrop(fp, td); return (ENOTSOCK); } if (fflagp != NULL) *fflagp = fp->f_flag; *fpp = fp; return (0); } /* * System call interface to the socket abstraction. */ #if defined(COMPAT_43) #define COMPAT_OLDSOCK #endif int sys_socket(struct thread *td, struct socket_args *uap) { struct socket *so; struct file *fp; int fd, error, type, oflag, fflag; AUDIT_ARG_SOCKET(uap->domain, uap->type, uap->protocol); type = uap->type; oflag = 0; fflag = 0; if ((type & SOCK_CLOEXEC) != 0) { type &= ~SOCK_CLOEXEC; oflag |= O_CLOEXEC; } if ((type & SOCK_NONBLOCK) != 0) { type &= ~SOCK_NONBLOCK; fflag |= FNONBLOCK; } #ifdef MAC error = mac_socket_check_create(td->td_ucred, uap->domain, type, uap->protocol); if (error != 0) return (error); #endif error = falloc(td, &fp, &fd, oflag); if (error != 0) return (error); /* An extra reference on `fp' has been held for us by falloc(). */ error = socreate(uap->domain, &so, type, uap->protocol, td->td_ucred, td); if (error != 0) { fdclose(td, fp, fd); } else { finit(fp, FREAD | FWRITE | fflag, DTYPE_SOCKET, so, &socketops); if ((fflag & FNONBLOCK) != 0) (void) fo_ioctl(fp, FIONBIO, &fflag, td->td_ucred, td); td->td_retval[0] = fd; } fdrop(fp, td); return (error); } int sys_bind(struct thread *td, struct bind_args *uap) { struct sockaddr *sa; int error; error = getsockaddr(&sa, uap->name, uap->namelen); if (error == 0) { error = kern_bindat(td, AT_FDCWD, uap->s, sa); free(sa, M_SONAME); } return (error); } int kern_bindat(struct thread *td, int dirfd, int fd, struct sockaddr *sa) { struct socket *so; struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(fd); AUDIT_ARG_SOCKADDR(td, dirfd, sa); error = getsock_cap(td, fd, cap_rights_init(&rights, CAP_BIND), &fp, NULL); if (error != 0) return (error); so = fp->f_data; #ifdef KTRACE if (KTRPOINT(td, KTR_STRUCT)) ktrsockaddr(sa); #endif #ifdef MAC error = mac_socket_check_bind(td->td_ucred, so, sa); if (error == 0) { #endif if (dirfd == AT_FDCWD) error = sobind(so, sa, td); else error = sobindat(dirfd, so, sa, td); #ifdef MAC } #endif fdrop(fp, td); return (error); } int sys_bindat(struct thread *td, struct bindat_args *uap) { struct sockaddr *sa; int error; error = getsockaddr(&sa, uap->name, uap->namelen); if (error == 0) { error = kern_bindat(td, uap->fd, uap->s, sa); free(sa, M_SONAME); } return (error); } int sys_listen(struct thread *td, struct listen_args *uap) { struct socket *so; struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(uap->s); error = getsock_cap(td, uap->s, cap_rights_init(&rights, CAP_LISTEN), &fp, NULL); if (error == 0) { so = fp->f_data; #ifdef MAC error = mac_socket_check_listen(td->td_ucred, so); if (error == 0) #endif error = solisten(so, uap->backlog, td); fdrop(fp, td); } return(error); } /* * accept1() */ static int accept1(td, s, uname, anamelen, flags) struct thread *td; int s; struct sockaddr *uname; socklen_t *anamelen; int flags; { struct sockaddr *name; socklen_t namelen; struct file *fp; int error; if (uname == NULL) return (kern_accept4(td, s, NULL, NULL, flags, NULL)); error = copyin(anamelen, &namelen, sizeof (namelen)); if (error != 0) return (error); error = kern_accept4(td, s, &name, &namelen, flags, &fp); if (error != 0) return (error); if (error == 0 && uname != NULL) { #ifdef COMPAT_OLDSOCK if (flags & ACCEPT4_COMPAT) ((struct osockaddr *)name)->sa_family = name->sa_family; #endif error = copyout(name, uname, namelen); } if (error == 0) error = copyout(&namelen, anamelen, sizeof(namelen)); if (error != 0) fdclose(td, fp, td->td_retval[0]); fdrop(fp, td); free(name, M_SONAME); return (error); } int kern_accept(struct thread *td, int s, struct sockaddr **name, socklen_t *namelen, struct file **fp) { return (kern_accept4(td, s, name, namelen, ACCEPT4_INHERIT, fp)); } int kern_accept4(struct thread *td, int s, struct sockaddr **name, socklen_t *namelen, int flags, struct file **fp) { struct file *headfp, *nfp = NULL; struct sockaddr *sa = NULL; struct socket *head, *so; cap_rights_t rights; u_int fflag; pid_t pgid; int error, fd, tmp; if (name != NULL) *name = NULL; AUDIT_ARG_FD(s); error = getsock_cap(td, s, cap_rights_init(&rights, CAP_ACCEPT), &headfp, &fflag); if (error != 0) return (error); head = headfp->f_data; if ((head->so_options & SO_ACCEPTCONN) == 0) { error = EINVAL; goto done; } #ifdef MAC error = mac_socket_check_accept(td->td_ucred, head); if (error != 0) goto done; #endif error = falloc(td, &nfp, &fd, (flags & SOCK_CLOEXEC) ? O_CLOEXEC : 0); if (error != 0) goto done; ACCEPT_LOCK(); if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->so_comp)) { ACCEPT_UNLOCK(); error = EWOULDBLOCK; goto noconnection; } while (TAILQ_EMPTY(&head->so_comp) && head->so_error == 0) { if (head->so_rcv.sb_state & SBS_CANTRCVMORE) { head->so_error = ECONNABORTED; break; } error = msleep(&head->so_timeo, &accept_mtx, PSOCK | PCATCH, "accept", 0); if (error != 0) { ACCEPT_UNLOCK(); goto noconnection; } } if (head->so_error) { error = head->so_error; head->so_error = 0; ACCEPT_UNLOCK(); goto noconnection; } so = TAILQ_FIRST(&head->so_comp); KASSERT(!(so->so_qstate & SQ_INCOMP), ("accept1: so SQ_INCOMP")); KASSERT(so->so_qstate & SQ_COMP, ("accept1: so not SQ_COMP")); /* * Before changing the flags on the socket, we have to bump the * reference count. Otherwise, if the protocol calls sofree(), * the socket will be released due to a zero refcount. */ SOCK_LOCK(so); /* soref() and so_state update */ soref(so); /* file descriptor reference */ TAILQ_REMOVE(&head->so_comp, so, so_list); head->so_qlen--; if (flags & ACCEPT4_INHERIT) so->so_state |= (head->so_state & SS_NBIO); else so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0; so->so_qstate &= ~SQ_COMP; so->so_head = NULL; SOCK_UNLOCK(so); ACCEPT_UNLOCK(); /* An extra reference on `nfp' has been held for us by falloc(). */ td->td_retval[0] = fd; /* connection has been removed from the listen queue */ KNOTE_UNLOCKED(&head->so_rcv.sb_sel.si_note, 0); if (flags & ACCEPT4_INHERIT) { pgid = fgetown(&head->so_sigio); if (pgid != 0) fsetown(pgid, &so->so_sigio); } else { fflag &= ~(FNONBLOCK | FASYNC); if (flags & SOCK_NONBLOCK) fflag |= FNONBLOCK; } finit(nfp, fflag, DTYPE_SOCKET, so, &socketops); /* Sync socket nonblocking/async state with file flags */ tmp = fflag & FNONBLOCK; (void) fo_ioctl(nfp, FIONBIO, &tmp, td->td_ucred, td); tmp = fflag & FASYNC; (void) fo_ioctl(nfp, FIOASYNC, &tmp, td->td_ucred, td); sa = NULL; error = soaccept(so, &sa); if (error != 0) goto noconnection; if (sa == NULL) { if (name) *namelen = 0; goto done; } AUDIT_ARG_SOCKADDR(td, AT_FDCWD, sa); if (name) { /* check sa_len before it is destroyed */ if (*namelen > sa->sa_len) *namelen = sa->sa_len; #ifdef KTRACE if (KTRPOINT(td, KTR_STRUCT)) ktrsockaddr(sa); #endif *name = sa; sa = NULL; } noconnection: free(sa, M_SONAME); /* * close the new descriptor, assuming someone hasn't ripped it * out from under us. */ if (error != 0) fdclose(td, nfp, fd); /* * Release explicitly held references before returning. We return * a reference on nfp to the caller on success if they request it. */ done: if (fp != NULL) { if (error == 0) { *fp = nfp; nfp = NULL; } else *fp = NULL; } if (nfp != NULL) fdrop(nfp, td); fdrop(headfp, td); return (error); } int sys_accept(td, uap) struct thread *td; struct accept_args *uap; { return (accept1(td, uap->s, uap->name, uap->anamelen, ACCEPT4_INHERIT)); } int sys_accept4(td, uap) struct thread *td; struct accept4_args *uap; { if (uap->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK)) return (EINVAL); return (accept1(td, uap->s, uap->name, uap->anamelen, uap->flags)); } #ifdef COMPAT_OLDSOCK int oaccept(td, uap) struct thread *td; struct accept_args *uap; { return (accept1(td, uap->s, uap->name, uap->anamelen, ACCEPT4_INHERIT | ACCEPT4_COMPAT)); } #endif /* COMPAT_OLDSOCK */ int sys_connect(struct thread *td, struct connect_args *uap) { struct sockaddr *sa; int error; error = getsockaddr(&sa, uap->name, uap->namelen); if (error == 0) { error = kern_connectat(td, AT_FDCWD, uap->s, sa); free(sa, M_SONAME); } return (error); } int kern_connectat(struct thread *td, int dirfd, int fd, struct sockaddr *sa) { struct socket *so; struct file *fp; cap_rights_t rights; int error, interrupted = 0; AUDIT_ARG_FD(fd); AUDIT_ARG_SOCKADDR(td, dirfd, sa); error = getsock_cap(td, fd, cap_rights_init(&rights, CAP_CONNECT), &fp, NULL); if (error != 0) return (error); so = fp->f_data; if (so->so_state & SS_ISCONNECTING) { error = EALREADY; goto done1; } #ifdef KTRACE if (KTRPOINT(td, KTR_STRUCT)) ktrsockaddr(sa); #endif #ifdef MAC error = mac_socket_check_connect(td->td_ucred, so, sa); if (error != 0) goto bad; #endif if (dirfd == AT_FDCWD) error = soconnect(so, sa, td); else error = soconnectat(dirfd, so, sa, td); if (error != 0) goto bad; if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) { error = EINPROGRESS; goto done1; } SOCK_LOCK(so); while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { error = msleep(&so->so_timeo, SOCK_MTX(so), PSOCK | PCATCH, "connec", 0); if (error != 0) { if (error == EINTR || error == ERESTART) interrupted = 1; break; } } if (error == 0) { error = so->so_error; so->so_error = 0; } SOCK_UNLOCK(so); bad: if (!interrupted) so->so_state &= ~SS_ISCONNECTING; if (error == ERESTART) error = EINTR; done1: fdrop(fp, td); return (error); } int sys_connectat(struct thread *td, struct connectat_args *uap) { struct sockaddr *sa; int error; error = getsockaddr(&sa, uap->name, uap->namelen); if (error == 0) { error = kern_connectat(td, uap->fd, uap->s, sa); free(sa, M_SONAME); } return (error); } int kern_socketpair(struct thread *td, int domain, int type, int protocol, int *rsv) { struct file *fp1, *fp2; struct socket *so1, *so2; int fd, error, oflag, fflag; AUDIT_ARG_SOCKET(domain, type, protocol); oflag = 0; fflag = 0; if ((type & SOCK_CLOEXEC) != 0) { type &= ~SOCK_CLOEXEC; oflag |= O_CLOEXEC; } if ((type & SOCK_NONBLOCK) != 0) { type &= ~SOCK_NONBLOCK; fflag |= FNONBLOCK; } #ifdef MAC /* We might want to have a separate check for socket pairs. */ error = mac_socket_check_create(td->td_ucred, domain, type, protocol); if (error != 0) return (error); #endif error = socreate(domain, &so1, type, protocol, td->td_ucred, td); if (error != 0) return (error); error = socreate(domain, &so2, type, protocol, td->td_ucred, td); if (error != 0) goto free1; /* On success extra reference to `fp1' and 'fp2' is set by falloc. */ error = falloc(td, &fp1, &fd, oflag); if (error != 0) goto free2; rsv[0] = fd; fp1->f_data = so1; /* so1 already has ref count */ error = falloc(td, &fp2, &fd, oflag); if (error != 0) goto free3; fp2->f_data = so2; /* so2 already has ref count */ rsv[1] = fd; error = soconnect2(so1, so2); if (error != 0) goto free4; if (type == SOCK_DGRAM) { /* * Datagram socket connection is asymmetric. */ error = soconnect2(so2, so1); if (error != 0) goto free4; } finit(fp1, FREAD | FWRITE | fflag, DTYPE_SOCKET, fp1->f_data, &socketops); finit(fp2, FREAD | FWRITE | fflag, DTYPE_SOCKET, fp2->f_data, &socketops); if ((fflag & FNONBLOCK) != 0) { (void) fo_ioctl(fp1, FIONBIO, &fflag, td->td_ucred, td); (void) fo_ioctl(fp2, FIONBIO, &fflag, td->td_ucred, td); } fdrop(fp1, td); fdrop(fp2, td); return (0); free4: fdclose(td, fp2, rsv[1]); fdrop(fp2, td); free3: fdclose(td, fp1, rsv[0]); fdrop(fp1, td); free2: if (so2 != NULL) (void)soclose(so2); free1: if (so1 != NULL) (void)soclose(so1); return (error); } int sys_socketpair(struct thread *td, struct socketpair_args *uap) { int error, sv[2]; error = kern_socketpair(td, uap->domain, uap->type, uap->protocol, sv); if (error != 0) return (error); error = copyout(sv, uap->rsv, 2 * sizeof(int)); if (error != 0) { (void)kern_close(td, sv[0]); (void)kern_close(td, sv[1]); } return (error); } static int sendit(struct thread *td, int s, struct msghdr *mp, int flags) { struct mbuf *control; struct sockaddr *to; int error; #ifdef CAPABILITY_MODE if (IN_CAPABILITY_MODE(td) && (mp->msg_name != NULL)) return (ECAPMODE); #endif if (mp->msg_name != NULL) { error = getsockaddr(&to, mp->msg_name, mp->msg_namelen); if (error != 0) { to = NULL; goto bad; } mp->msg_name = to; } else { to = NULL; } if (mp->msg_control) { if (mp->msg_controllen < sizeof(struct cmsghdr) #ifdef COMPAT_OLDSOCK && mp->msg_flags != MSG_COMPAT #endif ) { error = EINVAL; goto bad; } error = sockargs(&control, mp->msg_control, mp->msg_controllen, MT_CONTROL); if (error != 0) goto bad; #ifdef COMPAT_OLDSOCK if (mp->msg_flags == MSG_COMPAT) { struct cmsghdr *cm; M_PREPEND(control, sizeof(*cm), M_WAITOK); cm = mtod(control, struct cmsghdr *); cm->cmsg_len = control->m_len; cm->cmsg_level = SOL_SOCKET; cm->cmsg_type = SCM_RIGHTS; } #endif } else { control = NULL; } error = kern_sendit(td, s, mp, flags, control, UIO_USERSPACE); bad: free(to, M_SONAME); return (error); } int kern_sendit(struct thread *td, int s, struct msghdr *mp, int flags, struct mbuf *control, enum uio_seg segflg) { struct file *fp; struct uio auio; struct iovec *iov; struct socket *so; cap_rights_t rights; #ifdef KTRACE struct uio *ktruio = NULL; #endif ssize_t len; int i, error; AUDIT_ARG_FD(s); cap_rights_init(&rights, CAP_SEND); if (mp->msg_name != NULL) { AUDIT_ARG_SOCKADDR(td, AT_FDCWD, mp->msg_name); cap_rights_set(&rights, CAP_CONNECT); } error = getsock_cap(td, s, &rights, &fp, NULL); if (error != 0) return (error); so = (struct socket *)fp->f_data; #ifdef KTRACE if (mp->msg_name != NULL && KTRPOINT(td, KTR_STRUCT)) ktrsockaddr(mp->msg_name); #endif #ifdef MAC if (mp->msg_name != NULL) { error = mac_socket_check_connect(td->td_ucred, so, mp->msg_name); if (error != 0) goto bad; } error = mac_socket_check_send(td->td_ucred, so); if (error != 0) goto bad; #endif auio.uio_iov = mp->msg_iov; auio.uio_iovcnt = mp->msg_iovlen; auio.uio_segflg = segflg; auio.uio_rw = UIO_WRITE; auio.uio_td = td; auio.uio_offset = 0; /* XXX */ auio.uio_resid = 0; iov = mp->msg_iov; for (i = 0; i < mp->msg_iovlen; i++, iov++) { if ((auio.uio_resid += iov->iov_len) < 0) { error = EINVAL; goto bad; } } #ifdef KTRACE if (KTRPOINT(td, KTR_GENIO)) ktruio = cloneuio(&auio); #endif len = auio.uio_resid; error = sosend(so, mp->msg_name, &auio, 0, control, flags, td); if (error != 0) { if (auio.uio_resid != len && (error == ERESTART || error == EINTR || error == EWOULDBLOCK)) error = 0; /* Generation of SIGPIPE can be controlled per socket */ if (error == EPIPE && !(so->so_options & SO_NOSIGPIPE) && !(flags & MSG_NOSIGNAL)) { PROC_LOCK(td->td_proc); tdsignal(td, SIGPIPE); PROC_UNLOCK(td->td_proc); } } if (error == 0) td->td_retval[0] = len - auio.uio_resid; #ifdef KTRACE if (ktruio != NULL) { ktruio->uio_resid = td->td_retval[0]; ktrgenio(s, UIO_WRITE, ktruio, error); } #endif bad: fdrop(fp, td); return (error); } int sys_sendto(struct thread *td, struct sendto_args *uap) { struct msghdr msg; struct iovec aiov; msg.msg_name = uap->to; msg.msg_namelen = uap->tolen; msg.msg_iov = &aiov; msg.msg_iovlen = 1; msg.msg_control = 0; #ifdef COMPAT_OLDSOCK msg.msg_flags = 0; #endif aiov.iov_base = uap->buf; aiov.iov_len = uap->len; return (sendit(td, uap->s, &msg, uap->flags)); } #ifdef COMPAT_OLDSOCK int osend(struct thread *td, struct osend_args *uap) { struct msghdr msg; struct iovec aiov; msg.msg_name = 0; msg.msg_namelen = 0; msg.msg_iov = &aiov; msg.msg_iovlen = 1; aiov.iov_base = uap->buf; aiov.iov_len = uap->len; msg.msg_control = 0; msg.msg_flags = 0; return (sendit(td, uap->s, &msg, uap->flags)); } int osendmsg(struct thread *td, struct osendmsg_args *uap) { struct msghdr msg; struct iovec *iov; int error; error = copyin(uap->msg, &msg, sizeof (struct omsghdr)); if (error != 0) return (error); error = copyiniov(msg.msg_iov, msg.msg_iovlen, &iov, EMSGSIZE); if (error != 0) return (error); msg.msg_iov = iov; msg.msg_flags = MSG_COMPAT; error = sendit(td, uap->s, &msg, uap->flags); free(iov, M_IOV); return (error); } #endif int sys_sendmsg(struct thread *td, struct sendmsg_args *uap) { struct msghdr msg; struct iovec *iov; int error; error = copyin(uap->msg, &msg, sizeof (msg)); if (error != 0) return (error); error = copyiniov(msg.msg_iov, msg.msg_iovlen, &iov, EMSGSIZE); if (error != 0) return (error); msg.msg_iov = iov; #ifdef COMPAT_OLDSOCK msg.msg_flags = 0; #endif error = sendit(td, uap->s, &msg, uap->flags); free(iov, M_IOV); return (error); } int kern_recvit(struct thread *td, int s, struct msghdr *mp, enum uio_seg fromseg, struct mbuf **controlp) { struct uio auio; struct iovec *iov; struct mbuf *m, *control = NULL; caddr_t ctlbuf; struct file *fp; struct socket *so; struct sockaddr *fromsa = NULL; cap_rights_t rights; #ifdef KTRACE struct uio *ktruio = NULL; #endif ssize_t len; int error, i; if (controlp != NULL) *controlp = NULL; AUDIT_ARG_FD(s); error = getsock_cap(td, s, cap_rights_init(&rights, CAP_RECV), &fp, NULL); if (error != 0) return (error); so = fp->f_data; #ifdef MAC error = mac_socket_check_receive(td->td_ucred, so); if (error != 0) { fdrop(fp, td); return (error); } #endif auio.uio_iov = mp->msg_iov; auio.uio_iovcnt = mp->msg_iovlen; auio.uio_segflg = UIO_USERSPACE; auio.uio_rw = UIO_READ; auio.uio_td = td; auio.uio_offset = 0; /* XXX */ auio.uio_resid = 0; iov = mp->msg_iov; for (i = 0; i < mp->msg_iovlen; i++, iov++) { if ((auio.uio_resid += iov->iov_len) < 0) { fdrop(fp, td); return (EINVAL); } } #ifdef KTRACE if (KTRPOINT(td, KTR_GENIO)) ktruio = cloneuio(&auio); #endif len = auio.uio_resid; error = soreceive(so, &fromsa, &auio, NULL, (mp->msg_control || controlp) ? &control : NULL, &mp->msg_flags); if (error != 0) { if (auio.uio_resid != len && (error == ERESTART || error == EINTR || error == EWOULDBLOCK)) error = 0; } if (fromsa != NULL) AUDIT_ARG_SOCKADDR(td, AT_FDCWD, fromsa); #ifdef KTRACE if (ktruio != NULL) { ktruio->uio_resid = len - auio.uio_resid; ktrgenio(s, UIO_READ, ktruio, error); } #endif if (error != 0) goto out; td->td_retval[0] = len - auio.uio_resid; if (mp->msg_name) { len = mp->msg_namelen; if (len <= 0 || fromsa == NULL) len = 0; else { /* save sa_len before it is destroyed by MSG_COMPAT */ len = MIN(len, fromsa->sa_len); #ifdef COMPAT_OLDSOCK if (mp->msg_flags & MSG_COMPAT) ((struct osockaddr *)fromsa)->sa_family = fromsa->sa_family; #endif if (fromseg == UIO_USERSPACE) { error = copyout(fromsa, mp->msg_name, (unsigned)len); if (error != 0) goto out; } else bcopy(fromsa, mp->msg_name, len); } mp->msg_namelen = len; } if (mp->msg_control && controlp == NULL) { #ifdef COMPAT_OLDSOCK /* * We assume that old recvmsg calls won't receive access * rights and other control info, esp. as control info * is always optional and those options didn't exist in 4.3. * If we receive rights, trim the cmsghdr; anything else * is tossed. */ if (control && mp->msg_flags & MSG_COMPAT) { if (mtod(control, struct cmsghdr *)->cmsg_level != SOL_SOCKET || mtod(control, struct cmsghdr *)->cmsg_type != SCM_RIGHTS) { mp->msg_controllen = 0; goto out; } control->m_len -= sizeof (struct cmsghdr); control->m_data += sizeof (struct cmsghdr); } #endif len = mp->msg_controllen; m = control; mp->msg_controllen = 0; ctlbuf = mp->msg_control; while (m && len > 0) { unsigned int tocopy; if (len >= m->m_len) tocopy = m->m_len; else { mp->msg_flags |= MSG_CTRUNC; tocopy = len; } if ((error = copyout(mtod(m, caddr_t), ctlbuf, tocopy)) != 0) goto out; ctlbuf += tocopy; len -= tocopy; m = m->m_next; } mp->msg_controllen = ctlbuf - (caddr_t)mp->msg_control; } out: fdrop(fp, td); #ifdef KTRACE if (fromsa && KTRPOINT(td, KTR_STRUCT)) ktrsockaddr(fromsa); #endif free(fromsa, M_SONAME); if (error == 0 && controlp != NULL) *controlp = control; else if (control) m_freem(control); return (error); } static int recvit(struct thread *td, int s, struct msghdr *mp, void *namelenp) { int error; error = kern_recvit(td, s, mp, UIO_USERSPACE, NULL); if (error != 0) return (error); if (namelenp != NULL) { error = copyout(&mp->msg_namelen, namelenp, sizeof (socklen_t)); #ifdef COMPAT_OLDSOCK if (mp->msg_flags & MSG_COMPAT) error = 0; /* old recvfrom didn't check */ #endif } return (error); } int sys_recvfrom(struct thread *td, struct recvfrom_args *uap) { struct msghdr msg; struct iovec aiov; int error; if (uap->fromlenaddr) { error = copyin(uap->fromlenaddr, &msg.msg_namelen, sizeof (msg.msg_namelen)); if (error != 0) goto done2; } else { msg.msg_namelen = 0; } msg.msg_name = uap->from; msg.msg_iov = &aiov; msg.msg_iovlen = 1; aiov.iov_base = uap->buf; aiov.iov_len = uap->len; msg.msg_control = 0; msg.msg_flags = uap->flags; error = recvit(td, uap->s, &msg, uap->fromlenaddr); done2: return (error); } #ifdef COMPAT_OLDSOCK int orecvfrom(struct thread *td, struct recvfrom_args *uap) { uap->flags |= MSG_COMPAT; return (sys_recvfrom(td, uap)); } #endif #ifdef COMPAT_OLDSOCK int orecv(struct thread *td, struct orecv_args *uap) { struct msghdr msg; struct iovec aiov; msg.msg_name = 0; msg.msg_namelen = 0; msg.msg_iov = &aiov; msg.msg_iovlen = 1; aiov.iov_base = uap->buf; aiov.iov_len = uap->len; msg.msg_control = 0; msg.msg_flags = uap->flags; return (recvit(td, uap->s, &msg, NULL)); } /* * Old recvmsg. This code takes advantage of the fact that the old msghdr * overlays the new one, missing only the flags, and with the (old) access * rights where the control fields are now. */ int orecvmsg(struct thread *td, struct orecvmsg_args *uap) { struct msghdr msg; struct iovec *iov; int error; error = copyin(uap->msg, &msg, sizeof (struct omsghdr)); if (error != 0) return (error); error = copyiniov(msg.msg_iov, msg.msg_iovlen, &iov, EMSGSIZE); if (error != 0) return (error); msg.msg_flags = uap->flags | MSG_COMPAT; msg.msg_iov = iov; error = recvit(td, uap->s, &msg, &uap->msg->msg_namelen); if (msg.msg_controllen && error == 0) error = copyout(&msg.msg_controllen, &uap->msg->msg_accrightslen, sizeof (int)); free(iov, M_IOV); return (error); } #endif int sys_recvmsg(struct thread *td, struct recvmsg_args *uap) { struct msghdr msg; struct iovec *uiov, *iov; int error; error = copyin(uap->msg, &msg, sizeof (msg)); if (error != 0) return (error); error = copyiniov(msg.msg_iov, msg.msg_iovlen, &iov, EMSGSIZE); if (error != 0) return (error); msg.msg_flags = uap->flags; #ifdef COMPAT_OLDSOCK msg.msg_flags &= ~MSG_COMPAT; #endif uiov = msg.msg_iov; msg.msg_iov = iov; error = recvit(td, uap->s, &msg, NULL); if (error == 0) { msg.msg_iov = uiov; error = copyout(&msg, uap->msg, sizeof(msg)); } free(iov, M_IOV); return (error); } int sys_shutdown(struct thread *td, struct shutdown_args *uap) { struct socket *so; struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(uap->s); error = getsock_cap(td, uap->s, cap_rights_init(&rights, CAP_SHUTDOWN), &fp, NULL); if (error == 0) { so = fp->f_data; error = soshutdown(so, uap->how); /* * Previous versions did not return ENOTCONN, but 0 in * case the socket was not connected. Some important * programs like syslogd up to r279016, 2015-02-19, * still depend on this behavior. */ if (error == ENOTCONN && td->td_proc->p_osrel < P_OSREL_SHUTDOWN_ENOTCONN) error = 0; fdrop(fp, td); } return (error); } int sys_setsockopt(struct thread *td, struct setsockopt_args *uap) { return (kern_setsockopt(td, uap->s, uap->level, uap->name, uap->val, UIO_USERSPACE, uap->valsize)); } int kern_setsockopt(struct thread *td, int s, int level, int name, void *val, enum uio_seg valseg, socklen_t valsize) { struct socket *so; struct file *fp; struct sockopt sopt; cap_rights_t rights; int error; if (val == NULL && valsize != 0) return (EFAULT); if ((int)valsize < 0) return (EINVAL); sopt.sopt_dir = SOPT_SET; sopt.sopt_level = level; sopt.sopt_name = name; sopt.sopt_val = val; sopt.sopt_valsize = valsize; switch (valseg) { case UIO_USERSPACE: sopt.sopt_td = td; break; case UIO_SYSSPACE: sopt.sopt_td = NULL; break; default: panic("kern_setsockopt called with bad valseg"); } AUDIT_ARG_FD(s); error = getsock_cap(td, s, cap_rights_init(&rights, CAP_SETSOCKOPT), &fp, NULL); if (error == 0) { so = fp->f_data; error = sosetopt(so, &sopt); fdrop(fp, td); } return(error); } int sys_getsockopt(struct thread *td, struct getsockopt_args *uap) { socklen_t valsize; int error; if (uap->val) { error = copyin(uap->avalsize, &valsize, sizeof (valsize)); if (error != 0) return (error); } error = kern_getsockopt(td, uap->s, uap->level, uap->name, uap->val, UIO_USERSPACE, &valsize); if (error == 0) error = copyout(&valsize, uap->avalsize, sizeof (valsize)); return (error); } /* * Kernel version of getsockopt. * optval can be a userland or userspace. optlen is always a kernel pointer. */ int kern_getsockopt(struct thread *td, int s, int level, int name, void *val, enum uio_seg valseg, socklen_t *valsize) { struct socket *so; struct file *fp; struct sockopt sopt; cap_rights_t rights; int error; if (val == NULL) *valsize = 0; if ((int)*valsize < 0) return (EINVAL); sopt.sopt_dir = SOPT_GET; sopt.sopt_level = level; sopt.sopt_name = name; sopt.sopt_val = val; sopt.sopt_valsize = (size_t)*valsize; /* checked non-negative above */ switch (valseg) { case UIO_USERSPACE: sopt.sopt_td = td; break; case UIO_SYSSPACE: sopt.sopt_td = NULL; break; default: panic("kern_getsockopt called with bad valseg"); } AUDIT_ARG_FD(s); error = getsock_cap(td, s, cap_rights_init(&rights, CAP_GETSOCKOPT), &fp, NULL); if (error == 0) { so = fp->f_data; error = sogetopt(so, &sopt); *valsize = sopt.sopt_valsize; fdrop(fp, td); } return (error); } /* * getsockname1() - Get socket name. */ static int getsockname1(struct thread *td, struct getsockname_args *uap, int compat) { struct sockaddr *sa; socklen_t len; int error; error = copyin(uap->alen, &len, sizeof(len)); if (error != 0) return (error); error = kern_getsockname(td, uap->fdes, &sa, &len); if (error != 0) return (error); if (len != 0) { #ifdef COMPAT_OLDSOCK if (compat) ((struct osockaddr *)sa)->sa_family = sa->sa_family; #endif error = copyout(sa, uap->asa, (u_int)len); } free(sa, M_SONAME); if (error == 0) error = copyout(&len, uap->alen, sizeof(len)); return (error); } int kern_getsockname(struct thread *td, int fd, struct sockaddr **sa, socklen_t *alen) { struct socket *so; struct file *fp; cap_rights_t rights; socklen_t len; int error; AUDIT_ARG_FD(fd); error = getsock_cap(td, fd, cap_rights_init(&rights, CAP_GETSOCKNAME), &fp, NULL); if (error != 0) return (error); so = fp->f_data; *sa = NULL; CURVNET_SET(so->so_vnet); error = (*so->so_proto->pr_usrreqs->pru_sockaddr)(so, sa); CURVNET_RESTORE(); if (error != 0) goto bad; if (*sa == NULL) len = 0; else len = MIN(*alen, (*sa)->sa_len); *alen = len; #ifdef KTRACE if (KTRPOINT(td, KTR_STRUCT)) ktrsockaddr(*sa); #endif bad: fdrop(fp, td); if (error != 0 && *sa != NULL) { free(*sa, M_SONAME); *sa = NULL; } return (error); } int sys_getsockname(struct thread *td, struct getsockname_args *uap) { return (getsockname1(td, uap, 0)); } #ifdef COMPAT_OLDSOCK int ogetsockname(struct thread *td, struct getsockname_args *uap) { return (getsockname1(td, uap, 1)); } #endif /* COMPAT_OLDSOCK */ /* * getpeername1() - Get name of peer for connected socket. */ static int getpeername1(struct thread *td, struct getpeername_args *uap, int compat) { struct sockaddr *sa; socklen_t len; int error; error = copyin(uap->alen, &len, sizeof (len)); if (error != 0) return (error); error = kern_getpeername(td, uap->fdes, &sa, &len); if (error != 0) return (error); if (len != 0) { #ifdef COMPAT_OLDSOCK if (compat) ((struct osockaddr *)sa)->sa_family = sa->sa_family; #endif error = copyout(sa, uap->asa, (u_int)len); } free(sa, M_SONAME); if (error == 0) error = copyout(&len, uap->alen, sizeof(len)); return (error); } int kern_getpeername(struct thread *td, int fd, struct sockaddr **sa, socklen_t *alen) { struct socket *so; struct file *fp; cap_rights_t rights; socklen_t len; int error; AUDIT_ARG_FD(fd); error = getsock_cap(td, fd, cap_rights_init(&rights, CAP_GETPEERNAME), &fp, NULL); if (error != 0) return (error); so = fp->f_data; if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) { error = ENOTCONN; goto done; } *sa = NULL; CURVNET_SET(so->so_vnet); error = (*so->so_proto->pr_usrreqs->pru_peeraddr)(so, sa); CURVNET_RESTORE(); if (error != 0) goto bad; if (*sa == NULL) len = 0; else len = MIN(*alen, (*sa)->sa_len); *alen = len; #ifdef KTRACE if (KTRPOINT(td, KTR_STRUCT)) ktrsockaddr(*sa); #endif bad: if (error != 0 && *sa != NULL) { free(*sa, M_SONAME); *sa = NULL; } done: fdrop(fp, td); return (error); } int sys_getpeername(struct thread *td, struct getpeername_args *uap) { return (getpeername1(td, uap, 0)); } #ifdef COMPAT_OLDSOCK int ogetpeername(struct thread *td, struct ogetpeername_args *uap) { /* XXX uap should have type `getpeername_args *' to begin with. */ return (getpeername1(td, (struct getpeername_args *)uap, 1)); } #endif /* COMPAT_OLDSOCK */ static int sockargs(struct mbuf **mp, char *buf, socklen_t buflen, int type) { struct sockaddr *sa; struct mbuf *m; int error; if (buflen > MLEN) { #ifdef COMPAT_OLDSOCK if (type == MT_SONAME && buflen <= 112) buflen = MLEN; /* unix domain compat. hack */ else #endif if (buflen > MCLBYTES) return (EINVAL); } m = m_get2(buflen, M_WAITOK, type, 0); m->m_len = buflen; error = copyin(buf, mtod(m, void *), buflen); if (error != 0) (void) m_free(m); else { *mp = m; if (type == MT_SONAME) { sa = mtod(m, struct sockaddr *); #if defined(COMPAT_OLDSOCK) && BYTE_ORDER != BIG_ENDIAN if (sa->sa_family == 0 && sa->sa_len < AF_MAX) sa->sa_family = sa->sa_len; #endif sa->sa_len = buflen; } } return (error); } int getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len) { struct sockaddr *sa; int error; if (len > SOCK_MAXADDRLEN) return (ENAMETOOLONG); if (len < offsetof(struct sockaddr, sa_data[0])) return (EINVAL); sa = malloc(len, M_SONAME, M_WAITOK); error = copyin(uaddr, sa, len); if (error != 0) { free(sa, M_SONAME); } else { #if defined(COMPAT_OLDSOCK) && BYTE_ORDER != BIG_ENDIAN if (sa->sa_family == 0 && sa->sa_len < AF_MAX) sa->sa_family = sa->sa_len; #endif sa->sa_len = len; *namp = sa; } return (error); } Index: head/sys/kern/uipc_usrreq.c =================================================================== --- head/sys/kern/uipc_usrreq.c (revision 305831) +++ head/sys/kern/uipc_usrreq.c (revision 305832) @@ -1,2571 +1,2571 @@ /*- * Copyright (c) 1982, 1986, 1989, 1991, 1993 * The Regents of the University of California. * Copyright (c) 2004-2009 Robert N. M. Watson * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94 */ /* * UNIX Domain (Local) Sockets * * This is an implementation of UNIX (local) domain sockets. Each socket has * an associated struct unpcb (UNIX protocol control block). Stream sockets * may be connected to 0 or 1 other socket. Datagram sockets may be * connected to 0, 1, or many other sockets. Sockets may be created and * connected in pairs (socketpair(2)), or bound/connected to using the file * system name space. For most purposes, only the receive socket buffer is * used, as sending on one socket delivers directly to the receive socket * buffer of a second socket. * * The implementation is substantially complicated by the fact that * "ancillary data", such as file descriptors or credentials, may be passed * across UNIX domain sockets. The potential for passing UNIX domain sockets * over other UNIX domain sockets requires the implementation of a simple * garbage collector to find and tear down cycles of disconnected sockets. * * TODO: * RDM * rethink name space problems * need a proper out-of-band */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include #include #include #include #include /* XXX must be before */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DDB #include #endif #include #include MALLOC_DECLARE(M_FILECAPS); /* * Locking key: * (l) Locked using list lock * (g) Locked using linkage lock */ static uma_zone_t unp_zone; static unp_gen_t unp_gencnt; /* (l) */ static u_int unp_count; /* (l) Count of local sockets. */ static ino_t unp_ino; /* Prototype for fake inode numbers. */ static int unp_rights; /* (g) File descriptors in flight. */ static struct unp_head unp_shead; /* (l) List of stream sockets. */ static struct unp_head unp_dhead; /* (l) List of datagram sockets. */ static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */ struct unp_defer { SLIST_ENTRY(unp_defer) ud_link; struct file *ud_fp; }; static SLIST_HEAD(, unp_defer) unp_defers; static int unp_defers_count; static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; /* * Garbage collection of cyclic file descriptor/socket references occurs * asynchronously in a taskqueue context in order to avoid recursion and * reentrance in the UNIX domain socket, file descriptor, and socket layer * code. See unp_gc() for a full description. */ static struct timeout_task unp_gc_task; /* * The close of unix domain sockets attached as SCM_RIGHTS is * postponed to the taskqueue, to avoid arbitrary recursion depth. * The attached sockets might have another sockets attached. */ static struct task unp_defer_task; /* * Both send and receive buffers are allocated PIPSIZ bytes of buffering for * stream sockets, although the total for sender and receiver is actually * only PIPSIZ. * * Datagram sockets really use the sendspace as the maximum datagram size, * and don't really want to reserve the sendspace. Their recvspace should be * large enough for at least one max-size datagram plus address. */ #ifndef PIPSIZ #define PIPSIZ 8192 #endif static u_long unpst_sendspace = PIPSIZ; static u_long unpst_recvspace = PIPSIZ; static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ static u_long unpdg_recvspace = 4*1024; static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */ static u_long unpsp_recvspace = PIPSIZ; static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain"); static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0, "SOCK_STREAM"); static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM"); static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0, "SOCK_SEQPACKET"); SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, &unpst_sendspace, 0, "Default stream send space."); SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, &unpst_recvspace, 0, "Default stream receive space."); SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, &unpdg_sendspace, 0, "Default datagram send space."); SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, &unpdg_recvspace, 0, "Default datagram receive space."); SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW, &unpsp_sendspace, 0, "Default seqpacket send space."); SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW, &unpsp_recvspace, 0, "Default seqpacket receive space."); SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "File descriptors in flight."); SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD, &unp_defers_count, 0, "File descriptors deferred to taskqueue for close."); /* * Locking and synchronization: * * Three types of locks exit in the local domain socket implementation: a * global list mutex, a global linkage rwlock, and per-unpcb mutexes. Of the * global locks, the list lock protects the socket count, global generation * number, and stream/datagram global lists. The linkage lock protects the * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be * held exclusively over the acquisition of multiple unpcb locks to prevent * deadlock. * * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer, * allocated in pru_attach() and freed in pru_detach(). The validity of that * pointer is an invariant, so no lock is required to dereference the so_pcb * pointer if a valid socket reference is held by the caller. In practice, * this is always true during operations performed on a socket. Each unpcb * has a back-pointer to its socket, unp_socket, which will be stable under * the same circumstances. * * This pointer may only be safely dereferenced as long as a valid reference * to the unpcb is held. Typically, this reference will be from the socket, * or from another unpcb when the referring unpcb's lock is held (in order * that the reference not be invalidated during use). For example, to follow * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn, * as unp_socket remains valid as long as the reference to unp_conn is valid. * * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx. Individual * atomic reads without the lock may be performed "lockless", but more * complex reads and read-modify-writes require the mutex to be held. No * lock order is defined between unpcb locks -- multiple unpcb locks may be * acquired at the same time only when holding the linkage rwlock * exclusively, which prevents deadlocks. * * Blocking with UNIX domain sockets is a tricky issue: unlike most network * protocols, bind() is a non-atomic operation, and connect() requires * potential sleeping in the protocol, due to potentially waiting on local or * distributed file systems. We try to separate "lookup" operations, which * may sleep, and the IPC operations themselves, which typically can occur * with relative atomicity as locks can be held over the entire operation. * * Another tricky issue is simultaneous multi-threaded or multi-process * access to a single UNIX domain socket. These are handled by the flags * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or * binding, both of which involve dropping UNIX domain socket locks in order * to perform namei() and other file system operations. */ static struct rwlock unp_link_rwlock; static struct mtx unp_list_lock; static struct mtx unp_defers_lock; #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \ "unp_link_rwlock") #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \ RA_LOCKED) #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ RA_UNLOCKED) #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock) #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock) #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock) #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock) #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ RA_WLOCKED) #define UNP_LIST_LOCK_INIT() mtx_init(&unp_list_lock, \ "unp_list_lock", NULL, MTX_DEF) #define UNP_LIST_LOCK() mtx_lock(&unp_list_lock) #define UNP_LIST_UNLOCK() mtx_unlock(&unp_list_lock) #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \ "unp_defer", NULL, MTX_DEF) #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock) #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock) #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \ "unp_mtx", "unp_mtx", \ MTX_DUPOK|MTX_DEF|MTX_RECURSE) #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx) #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx) #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx) #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED) static int uipc_connect2(struct socket *, struct socket *); static int uipc_ctloutput(struct socket *, struct sockopt *); static int unp_connect(struct socket *, struct sockaddr *, struct thread *); static int unp_connectat(int, struct socket *, struct sockaddr *, struct thread *); static int unp_connect2(struct socket *so, struct socket *so2, int); static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2); static void unp_dispose(struct socket *so); static void unp_dispose_mbuf(struct mbuf *); static void unp_shutdown(struct unpcb *); static void unp_drop(struct unpcb *); static void unp_gc(__unused void *, int); static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int)); static void unp_discard(struct file *); static void unp_freerights(struct filedescent **, int); static void unp_init(void); static int unp_internalize(struct mbuf **, struct thread *); static void unp_internalize_fp(struct file *); static int unp_externalize(struct mbuf *, struct mbuf **, int); static int unp_externalize_fp(struct file *); static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *); static void unp_process_defers(void * __unused, int); /* * Definitions of protocols supported in the LOCAL domain. */ static struct domain localdomain; static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream; static struct pr_usrreqs uipc_usrreqs_seqpacket; static struct protosw localsw[] = { { .pr_type = SOCK_STREAM, .pr_domain = &localdomain, .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS, .pr_ctloutput = &uipc_ctloutput, .pr_usrreqs = &uipc_usrreqs_stream }, { .pr_type = SOCK_DGRAM, .pr_domain = &localdomain, .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS, .pr_ctloutput = &uipc_ctloutput, .pr_usrreqs = &uipc_usrreqs_dgram }, { .pr_type = SOCK_SEQPACKET, .pr_domain = &localdomain, /* * XXXRW: For now, PR_ADDR because soreceive will bump into them * due to our use of sbappendaddr. A new sbappend variants is needed * that supports both atomic record writes and control data. */ .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD| PR_RIGHTS, .pr_ctloutput = &uipc_ctloutput, .pr_usrreqs = &uipc_usrreqs_seqpacket, }, }; static struct domain localdomain = { .dom_family = AF_LOCAL, .dom_name = "local", .dom_init = unp_init, .dom_externalize = unp_externalize, .dom_dispose = unp_dispose, .dom_protosw = localsw, .dom_protoswNPROTOSW = &localsw[nitems(localsw)] }; DOMAIN_SET(local); static void uipc_abort(struct socket *so) { struct unpcb *unp, *unp2; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); UNP_LINK_WLOCK(); UNP_PCB_LOCK(unp); unp2 = unp->unp_conn; if (unp2 != NULL) { UNP_PCB_LOCK(unp2); unp_drop(unp2); UNP_PCB_UNLOCK(unp2); } UNP_PCB_UNLOCK(unp); UNP_LINK_WUNLOCK(); } static int uipc_accept(struct socket *so, struct sockaddr **nam) { struct unpcb *unp, *unp2; const struct sockaddr *sa; /* * Pass back name of connected socket, if it was bound and we are * still connected (our peer may have closed already!). */ unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); UNP_LINK_RLOCK(); unp2 = unp->unp_conn; if (unp2 != NULL && unp2->unp_addr != NULL) { UNP_PCB_LOCK(unp2); sa = (struct sockaddr *) unp2->unp_addr; bcopy(sa, *nam, sa->sa_len); UNP_PCB_UNLOCK(unp2); } else { sa = &sun_noname; bcopy(sa, *nam, sa->sa_len); } UNP_LINK_RUNLOCK(); return (0); } static int uipc_attach(struct socket *so, int proto, struct thread *td) { u_long sendspace, recvspace; struct unpcb *unp; int error; KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { switch (so->so_type) { case SOCK_STREAM: sendspace = unpst_sendspace; recvspace = unpst_recvspace; break; case SOCK_DGRAM: sendspace = unpdg_sendspace; recvspace = unpdg_recvspace; break; case SOCK_SEQPACKET: sendspace = unpsp_sendspace; recvspace = unpsp_recvspace; break; default: panic("uipc_attach"); } error = soreserve(so, sendspace, recvspace); if (error) return (error); } unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO); if (unp == NULL) return (ENOBUFS); LIST_INIT(&unp->unp_refs); UNP_PCB_LOCK_INIT(unp); unp->unp_socket = so; so->so_pcb = unp; unp->unp_refcount = 1; if (so->so_head != NULL) unp->unp_flags |= UNP_NASCENT; UNP_LIST_LOCK(); unp->unp_gencnt = ++unp_gencnt; unp_count++; switch (so->so_type) { case SOCK_STREAM: LIST_INSERT_HEAD(&unp_shead, unp, unp_link); break; case SOCK_DGRAM: LIST_INSERT_HEAD(&unp_dhead, unp, unp_link); break; case SOCK_SEQPACKET: LIST_INSERT_HEAD(&unp_sphead, unp, unp_link); break; default: panic("uipc_attach"); } UNP_LIST_UNLOCK(); return (0); } static int uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) { struct sockaddr_un *soun = (struct sockaddr_un *)nam; struct vattr vattr; int error, namelen; struct nameidata nd; struct unpcb *unp; struct vnode *vp; struct mount *mp; cap_rights_t rights; char *buf; if (nam->sa_family != AF_UNIX) return (EAFNOSUPPORT); unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); if (soun->sun_len > sizeof(struct sockaddr_un)) return (EINVAL); namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); if (namelen <= 0) return (EINVAL); /* * We don't allow simultaneous bind() calls on a single UNIX domain * socket, so flag in-progress operations, and return an error if an * operation is already in progress. * * Historically, we have not allowed a socket to be rebound, so this * also returns an error. Not allowing re-binding simplifies the * implementation and avoids a great many possible failure modes. */ UNP_PCB_LOCK(unp); if (unp->unp_vnode != NULL) { UNP_PCB_UNLOCK(unp); return (EINVAL); } if (unp->unp_flags & UNP_BINDING) { UNP_PCB_UNLOCK(unp); return (EALREADY); } unp->unp_flags |= UNP_BINDING; UNP_PCB_UNLOCK(unp); buf = malloc(namelen + 1, M_TEMP, M_WAITOK); bcopy(soun->sun_path, buf, namelen); buf[namelen] = 0; restart: NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE, UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_BINDAT), td); /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ error = namei(&nd); if (error) goto error; vp = nd.ni_vp; if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); if (nd.ni_dvp == vp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); if (vp != NULL) { vrele(vp); error = EADDRINUSE; goto error; } error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); if (error) goto error; goto restart; } VATTR_NULL(&vattr); vattr.va_type = VSOCK; vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); #ifdef MAC error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, &vattr); #endif if (error == 0) error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if (error) { vn_finished_write(mp); goto error; } vp = nd.ni_vp; ASSERT_VOP_ELOCKED(vp, "uipc_bind"); soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); UNP_LINK_WLOCK(); UNP_PCB_LOCK(unp); VOP_UNP_BIND(vp, unp->unp_socket); unp->unp_vnode = vp; unp->unp_addr = soun; unp->unp_flags &= ~UNP_BINDING; UNP_PCB_UNLOCK(unp); UNP_LINK_WUNLOCK(); VOP_UNLOCK(vp, 0); vn_finished_write(mp); free(buf, M_TEMP); return (0); error: UNP_PCB_LOCK(unp); unp->unp_flags &= ~UNP_BINDING; UNP_PCB_UNLOCK(unp); free(buf, M_TEMP); return (error); } static int uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) { return (uipc_bindat(AT_FDCWD, so, nam, td)); } static int uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) { int error; KASSERT(td == curthread, ("uipc_connect: td != curthread")); UNP_LINK_WLOCK(); error = unp_connect(so, nam, td); UNP_LINK_WUNLOCK(); return (error); } static int uipc_connectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) { int error; KASSERT(td == curthread, ("uipc_connectat: td != curthread")); UNP_LINK_WLOCK(); error = unp_connectat(fd, so, nam, td); UNP_LINK_WUNLOCK(); return (error); } static void uipc_close(struct socket *so) { struct unpcb *unp, *unp2; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_close: unp == NULL")); UNP_LINK_WLOCK(); UNP_PCB_LOCK(unp); unp2 = unp->unp_conn; if (unp2 != NULL) { UNP_PCB_LOCK(unp2); unp_disconnect(unp, unp2); UNP_PCB_UNLOCK(unp2); } UNP_PCB_UNLOCK(unp); UNP_LINK_WUNLOCK(); } static int uipc_connect2(struct socket *so1, struct socket *so2) { struct unpcb *unp, *unp2; int error; UNP_LINK_WLOCK(); unp = so1->so_pcb; KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); UNP_PCB_LOCK(unp); unp2 = so2->so_pcb; KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL")); UNP_PCB_LOCK(unp2); error = unp_connect2(so1, so2, PRU_CONNECT2); UNP_PCB_UNLOCK(unp2); UNP_PCB_UNLOCK(unp); UNP_LINK_WUNLOCK(); return (error); } static void uipc_detach(struct socket *so) { struct unpcb *unp, *unp2; struct sockaddr_un *saved_unp_addr; struct vnode *vp; int freeunp, local_unp_rights; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); vp = NULL; local_unp_rights = 0; UNP_LIST_LOCK(); LIST_REMOVE(unp, unp_link); unp->unp_gencnt = ++unp_gencnt; --unp_count; UNP_LIST_UNLOCK(); if ((unp->unp_flags & UNP_NASCENT) != 0) { UNP_PCB_LOCK(unp); goto teardown; } UNP_LINK_WLOCK(); UNP_PCB_LOCK(unp); /* * XXXRW: Should assert vp->v_socket == so. */ if ((vp = unp->unp_vnode) != NULL) { VOP_UNP_DETACH(vp); unp->unp_vnode = NULL; } unp2 = unp->unp_conn; if (unp2 != NULL) { UNP_PCB_LOCK(unp2); unp_disconnect(unp, unp2); UNP_PCB_UNLOCK(unp2); } /* * We hold the linkage lock exclusively, so it's OK to acquire * multiple pcb locks at a time. */ while (!LIST_EMPTY(&unp->unp_refs)) { struct unpcb *ref = LIST_FIRST(&unp->unp_refs); UNP_PCB_LOCK(ref); unp_drop(ref); UNP_PCB_UNLOCK(ref); } local_unp_rights = unp_rights; UNP_LINK_WUNLOCK(); teardown: unp->unp_socket->so_pcb = NULL; saved_unp_addr = unp->unp_addr; unp->unp_addr = NULL; unp->unp_refcount--; freeunp = (unp->unp_refcount == 0); if (saved_unp_addr != NULL) free(saved_unp_addr, M_SONAME); if (freeunp) { UNP_PCB_LOCK_DESTROY(unp); uma_zfree(unp_zone, unp); } else UNP_PCB_UNLOCK(unp); if (vp) vrele(vp); if (local_unp_rights) taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1); } static int uipc_disconnect(struct socket *so) { struct unpcb *unp, *unp2; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); UNP_LINK_WLOCK(); UNP_PCB_LOCK(unp); unp2 = unp->unp_conn; if (unp2 != NULL) { UNP_PCB_LOCK(unp2); unp_disconnect(unp, unp2); UNP_PCB_UNLOCK(unp2); } UNP_PCB_UNLOCK(unp); UNP_LINK_WUNLOCK(); return (0); } static int uipc_listen(struct socket *so, int backlog, struct thread *td) { struct unpcb *unp; int error; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_listen: unp == NULL")); UNP_PCB_LOCK(unp); if (unp->unp_vnode == NULL) { /* Already connected or not bound to an address. */ error = unp->unp_conn != NULL ? EINVAL : EDESTADDRREQ; UNP_PCB_UNLOCK(unp); return (error); } SOCK_LOCK(so); error = solisten_proto_check(so); if (error == 0) { cru2x(td->td_ucred, &unp->unp_peercred); unp->unp_flags |= UNP_HAVEPCCACHED; solisten_proto(so, backlog); } SOCK_UNLOCK(so); UNP_PCB_UNLOCK(unp); return (error); } static int uipc_peeraddr(struct socket *so, struct sockaddr **nam) { struct unpcb *unp, *unp2; const struct sockaddr *sa; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); UNP_LINK_RLOCK(); /* * XXX: It seems that this test always fails even when connection is * established. So, this else clause is added as workaround to * return PF_LOCAL sockaddr. */ unp2 = unp->unp_conn; if (unp2 != NULL) { UNP_PCB_LOCK(unp2); if (unp2->unp_addr != NULL) sa = (struct sockaddr *) unp2->unp_addr; else sa = &sun_noname; bcopy(sa, *nam, sa->sa_len); UNP_PCB_UNLOCK(unp2); } else { sa = &sun_noname; bcopy(sa, *nam, sa->sa_len); } UNP_LINK_RUNLOCK(); return (0); } static int uipc_rcvd(struct socket *so, int flags) { struct unpcb *unp, *unp2; struct socket *so2; u_int mbcnt, sbcc; unp = sotounpcb(so); KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET, ("%s: socktype %d", __func__, so->so_type)); /* * Adjust backpressure on sender and wakeup any waiting to write. * * The unp lock is acquired to maintain the validity of the unp_conn * pointer; no lock on unp2 is required as unp2->unp_socket will be * static as long as we don't permit unp2 to disconnect from unp, * which is prevented by the lock on unp. We cache values from * so_rcv to avoid holding the so_rcv lock over the entire * transaction on the remote so_snd. */ SOCKBUF_LOCK(&so->so_rcv); mbcnt = so->so_rcv.sb_mbcnt; sbcc = sbavail(&so->so_rcv); SOCKBUF_UNLOCK(&so->so_rcv); /* * There is a benign race condition at this point. If we're planning to * clear SB_STOP, but uipc_send is called on the connected socket at * this instant, it might add data to the sockbuf and set SB_STOP. Then * we would erroneously clear SB_STOP below, even though the sockbuf is * full. The race is benign because the only ill effect is to allow the * sockbuf to exceed its size limit, and the size limits are not * strictly guaranteed anyway. */ UNP_PCB_LOCK(unp); unp2 = unp->unp_conn; if (unp2 == NULL) { UNP_PCB_UNLOCK(unp); return (0); } so2 = unp2->unp_socket; SOCKBUF_LOCK(&so2->so_snd); if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax) so2->so_snd.sb_flags &= ~SB_STOP; sowwakeup_locked(so2); UNP_PCB_UNLOCK(unp); return (0); } static int uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, struct mbuf *control, struct thread *td) { struct unpcb *unp, *unp2; struct socket *so2; u_int mbcnt, sbcc; int error = 0; unp = sotounpcb(so); KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM || so->so_type == SOCK_SEQPACKET, ("%s: socktype %d", __func__, so->so_type)); if (flags & PRUS_OOB) { error = EOPNOTSUPP; goto release; } if (control != NULL && (error = unp_internalize(&control, td))) goto release; if ((nam != NULL) || (flags & PRUS_EOF)) UNP_LINK_WLOCK(); else UNP_LINK_RLOCK(); switch (so->so_type) { case SOCK_DGRAM: { const struct sockaddr *from; unp2 = unp->unp_conn; if (nam != NULL) { UNP_LINK_WLOCK_ASSERT(); if (unp2 != NULL) { error = EISCONN; break; } error = unp_connect(so, nam, td); if (error) break; unp2 = unp->unp_conn; } /* * Because connect() and send() are non-atomic in a sendto() * with a target address, it's possible that the socket will * have disconnected before the send() can run. In that case * return the slightly counter-intuitive but otherwise * correct error that the socket is not connected. */ if (unp2 == NULL) { error = ENOTCONN; break; } /* Lockless read. */ if (unp2->unp_flags & UNP_WANTCRED) control = unp_addsockcred(td, control); UNP_PCB_LOCK(unp); if (unp->unp_addr != NULL) from = (struct sockaddr *)unp->unp_addr; else from = &sun_noname; so2 = unp2->unp_socket; SOCKBUF_LOCK(&so2->so_rcv); if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { sorwakeup_locked(so2); m = NULL; control = NULL; } else { SOCKBUF_UNLOCK(&so2->so_rcv); error = ENOBUFS; } if (nam != NULL) { UNP_LINK_WLOCK_ASSERT(); UNP_PCB_LOCK(unp2); unp_disconnect(unp, unp2); UNP_PCB_UNLOCK(unp2); } UNP_PCB_UNLOCK(unp); break; } case SOCK_SEQPACKET: case SOCK_STREAM: if ((so->so_state & SS_ISCONNECTED) == 0) { if (nam != NULL) { UNP_LINK_WLOCK_ASSERT(); error = unp_connect(so, nam, td); if (error) break; /* XXX */ } else { error = ENOTCONN; break; } } /* Lockless read. */ if (so->so_snd.sb_state & SBS_CANTSENDMORE) { error = EPIPE; break; } /* * Because connect() and send() are non-atomic in a sendto() * with a target address, it's possible that the socket will * have disconnected before the send() can run. In that case * return the slightly counter-intuitive but otherwise * correct error that the socket is not connected. * * Locking here must be done carefully: the linkage lock * prevents interconnections between unpcbs from changing, so * we can traverse from unp to unp2 without acquiring unp's * lock. Socket buffer locks follow unpcb locks, so we can * acquire both remote and lock socket buffer locks. */ unp2 = unp->unp_conn; if (unp2 == NULL) { error = ENOTCONN; break; } so2 = unp2->unp_socket; UNP_PCB_LOCK(unp2); SOCKBUF_LOCK(&so2->so_rcv); if (unp2->unp_flags & UNP_WANTCRED) { /* * Credentials are passed only once on SOCK_STREAM * and SOCK_SEQPACKET. */ unp2->unp_flags &= ~UNP_WANTCRED; control = unp_addsockcred(td, control); } /* * Send to paired receive port, and then reduce send buffer * hiwater marks to maintain backpressure. Wake up readers. */ switch (so->so_type) { case SOCK_STREAM: if (control != NULL) { if (sbappendcontrol_locked(&so2->so_rcv, m, control)) control = NULL; } else sbappend_locked(&so2->so_rcv, m, flags); break; case SOCK_SEQPACKET: { const struct sockaddr *from; from = &sun_noname; /* * Don't check for space available in so2->so_rcv. * Unix domain sockets only check for space in the * sending sockbuf, and that check is performed one * level up the stack. */ if (sbappendaddr_nospacecheck_locked(&so2->so_rcv, from, m, control)) control = NULL; break; } } mbcnt = so2->so_rcv.sb_mbcnt; sbcc = sbavail(&so2->so_rcv); if (sbcc) sorwakeup_locked(so2); else SOCKBUF_UNLOCK(&so2->so_rcv); /* * The PCB lock on unp2 protects the SB_STOP flag. Without it, * it would be possible for uipc_rcvd to be called at this * point, drain the receiving sockbuf, clear SB_STOP, and then * we would set SB_STOP below. That could lead to an empty * sockbuf having SB_STOP set */ SOCKBUF_LOCK(&so->so_snd); if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax) so->so_snd.sb_flags |= SB_STOP; SOCKBUF_UNLOCK(&so->so_snd); UNP_PCB_UNLOCK(unp2); m = NULL; break; } /* * PRUS_EOF is equivalent to pru_send followed by pru_shutdown. */ if (flags & PRUS_EOF) { UNP_PCB_LOCK(unp); socantsendmore(so); unp_shutdown(unp); UNP_PCB_UNLOCK(unp); } if ((nam != NULL) || (flags & PRUS_EOF)) UNP_LINK_WUNLOCK(); else UNP_LINK_RUNLOCK(); if (control != NULL && error != 0) unp_dispose_mbuf(control); release: if (control != NULL) m_freem(control); if (m != NULL) m_freem(m); return (error); } static int uipc_ready(struct socket *so, struct mbuf *m, int count) { struct unpcb *unp, *unp2; struct socket *so2; int error; unp = sotounpcb(so); UNP_LINK_RLOCK(); unp2 = unp->unp_conn; UNP_PCB_LOCK(unp2); so2 = unp2->unp_socket; SOCKBUF_LOCK(&so2->so_rcv); if ((error = sbready(&so2->so_rcv, m, count)) == 0) sorwakeup_locked(so2); else SOCKBUF_UNLOCK(&so2->so_rcv); UNP_PCB_UNLOCK(unp2); UNP_LINK_RUNLOCK(); return (error); } static int uipc_sense(struct socket *so, struct stat *sb) { struct unpcb *unp; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); sb->st_blksize = so->so_snd.sb_hiwat; UNP_PCB_LOCK(unp); sb->st_dev = NODEV; if (unp->unp_ino == 0) unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; sb->st_ino = unp->unp_ino; UNP_PCB_UNLOCK(unp); return (0); } static int uipc_shutdown(struct socket *so) { struct unpcb *unp; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); UNP_LINK_WLOCK(); UNP_PCB_LOCK(unp); socantsendmore(so); unp_shutdown(unp); UNP_PCB_UNLOCK(unp); UNP_LINK_WUNLOCK(); return (0); } static int uipc_sockaddr(struct socket *so, struct sockaddr **nam) { struct unpcb *unp; const struct sockaddr *sa; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); UNP_PCB_LOCK(unp); if (unp->unp_addr != NULL) sa = (struct sockaddr *) unp->unp_addr; else sa = &sun_noname; bcopy(sa, *nam, sa->sa_len); UNP_PCB_UNLOCK(unp); return (0); } static struct pr_usrreqs uipc_usrreqs_dgram = { .pru_abort = uipc_abort, .pru_accept = uipc_accept, .pru_attach = uipc_attach, .pru_bind = uipc_bind, .pru_bindat = uipc_bindat, .pru_connect = uipc_connect, .pru_connectat = uipc_connectat, .pru_connect2 = uipc_connect2, .pru_detach = uipc_detach, .pru_disconnect = uipc_disconnect, .pru_listen = uipc_listen, .pru_peeraddr = uipc_peeraddr, .pru_rcvd = uipc_rcvd, .pru_send = uipc_send, .pru_sense = uipc_sense, .pru_shutdown = uipc_shutdown, .pru_sockaddr = uipc_sockaddr, .pru_soreceive = soreceive_dgram, .pru_close = uipc_close, }; static struct pr_usrreqs uipc_usrreqs_seqpacket = { .pru_abort = uipc_abort, .pru_accept = uipc_accept, .pru_attach = uipc_attach, .pru_bind = uipc_bind, .pru_bindat = uipc_bindat, .pru_connect = uipc_connect, .pru_connectat = uipc_connectat, .pru_connect2 = uipc_connect2, .pru_detach = uipc_detach, .pru_disconnect = uipc_disconnect, .pru_listen = uipc_listen, .pru_peeraddr = uipc_peeraddr, .pru_rcvd = uipc_rcvd, .pru_send = uipc_send, .pru_sense = uipc_sense, .pru_shutdown = uipc_shutdown, .pru_sockaddr = uipc_sockaddr, .pru_soreceive = soreceive_generic, /* XXX: or...? */ .pru_close = uipc_close, }; static struct pr_usrreqs uipc_usrreqs_stream = { .pru_abort = uipc_abort, .pru_accept = uipc_accept, .pru_attach = uipc_attach, .pru_bind = uipc_bind, .pru_bindat = uipc_bindat, .pru_connect = uipc_connect, .pru_connectat = uipc_connectat, .pru_connect2 = uipc_connect2, .pru_detach = uipc_detach, .pru_disconnect = uipc_disconnect, .pru_listen = uipc_listen, .pru_peeraddr = uipc_peeraddr, .pru_rcvd = uipc_rcvd, .pru_send = uipc_send, .pru_ready = uipc_ready, .pru_sense = uipc_sense, .pru_shutdown = uipc_shutdown, .pru_sockaddr = uipc_sockaddr, .pru_soreceive = soreceive_generic, .pru_close = uipc_close, }; static int uipc_ctloutput(struct socket *so, struct sockopt *sopt) { struct unpcb *unp; struct xucred xu; int error, optval; if (sopt->sopt_level != 0) return (EINVAL); unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); error = 0; switch (sopt->sopt_dir) { case SOPT_GET: switch (sopt->sopt_name) { case LOCAL_PEERCRED: UNP_PCB_LOCK(unp); if (unp->unp_flags & UNP_HAVEPC) xu = unp->unp_peercred; else { if (so->so_type == SOCK_STREAM) error = ENOTCONN; else error = EINVAL; } UNP_PCB_UNLOCK(unp); if (error == 0) error = sooptcopyout(sopt, &xu, sizeof(xu)); break; case LOCAL_CREDS: /* Unlocked read. */ optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; error = sooptcopyout(sopt, &optval, sizeof(optval)); break; case LOCAL_CONNWAIT: /* Unlocked read. */ optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; error = sooptcopyout(sopt, &optval, sizeof(optval)); break; default: error = EOPNOTSUPP; break; } break; case SOPT_SET: switch (sopt->sopt_name) { case LOCAL_CREDS: case LOCAL_CONNWAIT: error = sooptcopyin(sopt, &optval, sizeof(optval), sizeof(optval)); if (error) break; #define OPTSET(bit) do { \ UNP_PCB_LOCK(unp); \ if (optval) \ unp->unp_flags |= bit; \ else \ unp->unp_flags &= ~bit; \ UNP_PCB_UNLOCK(unp); \ } while (0) switch (sopt->sopt_name) { case LOCAL_CREDS: OPTSET(UNP_WANTCRED); break; case LOCAL_CONNWAIT: OPTSET(UNP_CONNWAIT); break; default: break; } break; #undef OPTSET default: error = ENOPROTOOPT; break; } break; default: error = EOPNOTSUPP; break; } return (error); } static int unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) { return (unp_connectat(AT_FDCWD, so, nam, td)); } static int unp_connectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) { struct sockaddr_un *soun = (struct sockaddr_un *)nam; struct vnode *vp; struct socket *so2, *so3; struct unpcb *unp, *unp2, *unp3; struct nameidata nd; char buf[SOCK_MAXADDRLEN]; struct sockaddr *sa; cap_rights_t rights; int error, len; if (nam->sa_family != AF_UNIX) return (EAFNOSUPPORT); UNP_LINK_WLOCK_ASSERT(); unp = sotounpcb(so); KASSERT(unp != NULL, ("unp_connect: unp == NULL")); if (nam->sa_len > sizeof(struct sockaddr_un)) return (EINVAL); len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); if (len <= 0) return (EINVAL); bcopy(soun->sun_path, buf, len); buf[len] = 0; UNP_PCB_LOCK(unp); if (unp->unp_flags & UNP_CONNECTING) { UNP_PCB_UNLOCK(unp); return (EALREADY); } UNP_LINK_WUNLOCK(); unp->unp_flags |= UNP_CONNECTING; UNP_PCB_UNLOCK(unp); sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td); error = namei(&nd); if (error) vp = NULL; else vp = nd.ni_vp; ASSERT_VOP_LOCKED(vp, "unp_connect"); NDFREE(&nd, NDF_ONLY_PNBUF); if (error) goto bad; if (vp->v_type != VSOCK) { error = ENOTSOCK; goto bad; } #ifdef MAC error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); if (error) goto bad; #endif error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); if (error) goto bad; unp = sotounpcb(so); KASSERT(unp != NULL, ("unp_connect: unp == NULL")); /* * Lock linkage lock for two reasons: make sure v_socket is stable, * and to protect simultaneous locking of multiple pcbs. */ UNP_LINK_WLOCK(); VOP_UNP_CONNECT(vp, &so2); if (so2 == NULL) { error = ECONNREFUSED; goto bad2; } if (so->so_type != so2->so_type) { error = EPROTOTYPE; goto bad2; } if (so->so_proto->pr_flags & PR_CONNREQUIRED) { if (so2->so_options & SO_ACCEPTCONN) { CURVNET_SET(so2->so_vnet); so3 = sonewconn(so2, 0); CURVNET_RESTORE(); } else so3 = NULL; if (so3 == NULL) { error = ECONNREFUSED; goto bad2; } unp = sotounpcb(so); unp2 = sotounpcb(so2); unp3 = sotounpcb(so3); UNP_PCB_LOCK(unp); UNP_PCB_LOCK(unp2); UNP_PCB_LOCK(unp3); if (unp2->unp_addr != NULL) { bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); unp3->unp_addr = (struct sockaddr_un *) sa; sa = NULL; } /* * The connector's (client's) credentials are copied from its * process structure at the time of connect() (which is now). */ cru2x(td->td_ucred, &unp3->unp_peercred); unp3->unp_flags |= UNP_HAVEPC; /* * The receiver's (server's) credentials are copied from the * unp_peercred member of socket on which the former called * listen(); uipc_listen() cached that process's credentials * at that time so we can use them now. */ KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, ("unp_connect: listener without cached peercred")); memcpy(&unp->unp_peercred, &unp2->unp_peercred, sizeof(unp->unp_peercred)); unp->unp_flags |= UNP_HAVEPC; if (unp2->unp_flags & UNP_WANTCRED) unp3->unp_flags |= UNP_WANTCRED; UNP_PCB_UNLOCK(unp3); UNP_PCB_UNLOCK(unp2); UNP_PCB_UNLOCK(unp); #ifdef MAC mac_socketpeer_set_from_socket(so, so3); mac_socketpeer_set_from_socket(so3, so); #endif so2 = so3; } unp = sotounpcb(so); KASSERT(unp != NULL, ("unp_connect: unp == NULL")); unp2 = sotounpcb(so2); KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL")); UNP_PCB_LOCK(unp); UNP_PCB_LOCK(unp2); error = unp_connect2(so, so2, PRU_CONNECT); UNP_PCB_UNLOCK(unp2); UNP_PCB_UNLOCK(unp); bad2: UNP_LINK_WUNLOCK(); bad: if (vp != NULL) vput(vp); free(sa, M_SONAME); UNP_LINK_WLOCK(); UNP_PCB_LOCK(unp); unp->unp_flags &= ~UNP_CONNECTING; UNP_PCB_UNLOCK(unp); return (error); } static int unp_connect2(struct socket *so, struct socket *so2, int req) { struct unpcb *unp; struct unpcb *unp2; unp = sotounpcb(so); KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); unp2 = sotounpcb(so2); KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); UNP_LINK_WLOCK_ASSERT(); UNP_PCB_LOCK_ASSERT(unp); UNP_PCB_LOCK_ASSERT(unp2); if (so2->so_type != so->so_type) return (EPROTOTYPE); unp2->unp_flags &= ~UNP_NASCENT; unp->unp_conn = unp2; switch (so->so_type) { case SOCK_DGRAM: LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); soisconnected(so); break; case SOCK_STREAM: case SOCK_SEQPACKET: unp2->unp_conn = unp; if (req == PRU_CONNECT && ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) soisconnecting(so); else soisconnected(so); soisconnected(so2); break; default: panic("unp_connect2"); } return (0); } static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2) { struct socket *so; KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL")); UNP_LINK_WLOCK_ASSERT(); UNP_PCB_LOCK_ASSERT(unp); UNP_PCB_LOCK_ASSERT(unp2); unp->unp_conn = NULL; switch (unp->unp_socket->so_type) { case SOCK_DGRAM: LIST_REMOVE(unp, unp_reflink); so = unp->unp_socket; SOCK_LOCK(so); so->so_state &= ~SS_ISCONNECTED; SOCK_UNLOCK(so); break; case SOCK_STREAM: case SOCK_SEQPACKET: soisdisconnected(unp->unp_socket); unp2->unp_conn = NULL; soisdisconnected(unp2->unp_socket); break; } } /* * unp_pcblist() walks the global list of struct unpcb's to generate a * pointer list, bumping the refcount on each unpcb. It then copies them out * sequentially, validating the generation number on each to see if it has * been detached. All of this is necessary because copyout() may sleep on * disk I/O. */ static int unp_pcblist(SYSCTL_HANDLER_ARGS) { int error, i, n; int freeunp; struct unpcb *unp, **unp_list; unp_gen_t gencnt; struct xunpgen *xug; struct unp_head *head; struct xunpcb *xu; switch ((intptr_t)arg1) { case SOCK_STREAM: head = &unp_shead; break; case SOCK_DGRAM: head = &unp_dhead; break; case SOCK_SEQPACKET: head = &unp_sphead; break; default: panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1); } /* * The process of preparing the PCB list is too time-consuming and * resource-intensive to repeat twice on every request. */ if (req->oldptr == NULL) { n = unp_count; req->oldidx = 2 * (sizeof *xug) + (n + n/8) * sizeof(struct xunpcb); return (0); } if (req->newptr != NULL) return (EPERM); /* * OK, now we're committed to doing something. */ xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); UNP_LIST_LOCK(); gencnt = unp_gencnt; n = unp_count; UNP_LIST_UNLOCK(); xug->xug_len = sizeof *xug; xug->xug_count = n; xug->xug_gen = gencnt; xug->xug_sogen = so_gencnt; error = SYSCTL_OUT(req, xug, sizeof *xug); if (error) { free(xug, M_TEMP); return (error); } unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); UNP_LIST_LOCK(); for (unp = LIST_FIRST(head), i = 0; unp && i < n; unp = LIST_NEXT(unp, unp_link)) { UNP_PCB_LOCK(unp); if (unp->unp_gencnt <= gencnt) { if (cr_cansee(req->td->td_ucred, unp->unp_socket->so_cred)) { UNP_PCB_UNLOCK(unp); continue; } unp_list[i++] = unp; unp->unp_refcount++; } UNP_PCB_UNLOCK(unp); } UNP_LIST_UNLOCK(); n = i; /* In case we lost some during malloc. */ error = 0; xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); for (i = 0; i < n; i++) { unp = unp_list[i]; UNP_PCB_LOCK(unp); unp->unp_refcount--; if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) { xu->xu_len = sizeof *xu; xu->xu_unpp = unp; /* * XXX - need more locking here to protect against * connect/disconnect races for SMP. */ if (unp->unp_addr != NULL) bcopy(unp->unp_addr, &xu->xu_addr, unp->unp_addr->sun_len); if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL) bcopy(unp->unp_conn->unp_addr, &xu->xu_caddr, unp->unp_conn->unp_addr->sun_len); bcopy(unp, &xu->xu_unp, sizeof *unp); sotoxsocket(unp->unp_socket, &xu->xu_socket); UNP_PCB_UNLOCK(unp); error = SYSCTL_OUT(req, xu, sizeof *xu); } else { freeunp = (unp->unp_refcount == 0); UNP_PCB_UNLOCK(unp); if (freeunp) { UNP_PCB_LOCK_DESTROY(unp); uma_zfree(unp_zone, unp); } } } free(xu, M_TEMP); if (!error) { /* * Give the user an updated idea of our state. If the * generation differs from what we told her before, she knows * that something happened while we were processing this * request, and it might be necessary to retry. */ xug->xug_gen = unp_gencnt; xug->xug_sogen = so_gencnt; xug->xug_count = unp_count; error = SYSCTL_OUT(req, xug, sizeof *xug); } free(unp_list, M_TEMP); free(xug, M_TEMP); return (error); } SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", "List of active local datagram sockets"); SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", "List of active local stream sockets"); SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", "List of active local seqpacket sockets"); static void unp_shutdown(struct unpcb *unp) { struct unpcb *unp2; struct socket *so; UNP_LINK_WLOCK_ASSERT(); UNP_PCB_LOCK_ASSERT(unp); unp2 = unp->unp_conn; if ((unp->unp_socket->so_type == SOCK_STREAM || (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) { so = unp2->unp_socket; if (so != NULL) socantrcvmore(so); } } static void unp_drop(struct unpcb *unp) { struct socket *so = unp->unp_socket; struct unpcb *unp2; UNP_LINK_WLOCK_ASSERT(); UNP_PCB_LOCK_ASSERT(unp); /* * Regardless of whether the socket's peer dropped the connection * with this socket by aborting or disconnecting, POSIX requires * that ECONNRESET is returned. */ so->so_error = ECONNRESET; unp2 = unp->unp_conn; if (unp2 == NULL) return; UNP_PCB_LOCK(unp2); unp_disconnect(unp, unp2); UNP_PCB_UNLOCK(unp2); } static void unp_freerights(struct filedescent **fdep, int fdcount) { struct file *fp; int i; KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount)); for (i = 0; i < fdcount; i++) { fp = fdep[i]->fde_file; filecaps_free(&fdep[i]->fde_caps); unp_discard(fp); } free(fdep[0], M_FILECAPS); } static int unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags) { struct thread *td = curthread; /* XXX */ struct cmsghdr *cm = mtod(control, struct cmsghdr *); int i; int *fdp; struct filedesc *fdesc = td->td_proc->p_fd; struct filedescent **fdep; void *data; socklen_t clen = control->m_len, datalen; int error, newfds; u_int newlen; UNP_LINK_UNLOCK_ASSERT(); error = 0; if (controlp != NULL) /* controlp == NULL => free control messages */ *controlp = NULL; while (cm != NULL) { if (sizeof(*cm) > clen || cm->cmsg_len > clen) { error = EINVAL; break; } data = CMSG_DATA(cm); datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; if (cm->cmsg_level == SOL_SOCKET && cm->cmsg_type == SCM_RIGHTS) { newfds = datalen / sizeof(*fdep); if (newfds == 0) goto next; fdep = data; /* If we're not outputting the descriptors free them. */ if (error || controlp == NULL) { unp_freerights(fdep, newfds); goto next; } FILEDESC_XLOCK(fdesc); /* * Now change each pointer to an fd in the global * table to an integer that is the index to the local * fd table entry that we set up to point to the * global one we are transferring. */ newlen = newfds * sizeof(int); *controlp = sbcreatecontrol(NULL, newlen, SCM_RIGHTS, SOL_SOCKET); if (*controlp == NULL) { FILEDESC_XUNLOCK(fdesc); error = E2BIG; unp_freerights(fdep, newfds); goto next; } fdp = (int *) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); if (fdallocn(td, 0, fdp, newfds) != 0) { FILEDESC_XUNLOCK(fdesc); error = EMSGSIZE; unp_freerights(fdep, newfds); m_freem(*controlp); *controlp = NULL; goto next; } for (i = 0; i < newfds; i++, fdp++) { _finstall(fdesc, fdep[i]->fde_file, *fdp, (flags & MSG_CMSG_CLOEXEC) != 0 ? UF_EXCLOSE : 0, &fdep[i]->fde_caps); unp_externalize_fp(fdep[i]->fde_file); } FILEDESC_XUNLOCK(fdesc); free(fdep[0], M_FILECAPS); } else { /* We can just copy anything else across. */ if (error || controlp == NULL) goto next; *controlp = sbcreatecontrol(NULL, datalen, cm->cmsg_type, cm->cmsg_level); if (*controlp == NULL) { error = ENOBUFS; goto next; } bcopy(data, CMSG_DATA(mtod(*controlp, struct cmsghdr *)), datalen); } controlp = &(*controlp)->m_next; next: if (CMSG_SPACE(datalen) < clen) { clen -= CMSG_SPACE(datalen); cm = (struct cmsghdr *) ((caddr_t)cm + CMSG_SPACE(datalen)); } else { clen = 0; cm = NULL; } } m_freem(control); return (error); } static void unp_zone_change(void *tag) { uma_zone_set_max(unp_zone, maxsockets); } static void unp_init(void) { #ifdef VIMAGE if (!IS_DEFAULT_VNET(curvnet)) return; #endif unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); if (unp_zone == NULL) panic("unp_init"); uma_zone_set_max(unp_zone, maxsockets); uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached"); EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, NULL, EVENTHANDLER_PRI_ANY); LIST_INIT(&unp_dhead); LIST_INIT(&unp_shead); LIST_INIT(&unp_sphead); SLIST_INIT(&unp_defers); TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL); TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL); UNP_LINK_LOCK_INIT(); UNP_LIST_LOCK_INIT(); UNP_DEFERRED_LOCK_INIT(); } static int unp_internalize(struct mbuf **controlp, struct thread *td) { struct mbuf *control = *controlp; struct proc *p = td->td_proc; struct filedesc *fdesc = p->p_fd; struct bintime *bt; struct cmsghdr *cm = mtod(control, struct cmsghdr *); struct cmsgcred *cmcred; struct filedescent *fde, **fdep, *fdev; struct file *fp; struct timeval *tv; int i, *fdp; void *data; socklen_t clen = control->m_len, datalen; int error, oldfds; u_int newlen; UNP_LINK_UNLOCK_ASSERT(); error = 0; *controlp = NULL; while (cm != NULL) { if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) { error = EINVAL; goto out; } data = CMSG_DATA(cm); datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; switch (cm->cmsg_type) { /* * Fill in credential information. */ case SCM_CREDS: *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), SCM_CREDS, SOL_SOCKET); if (*controlp == NULL) { error = ENOBUFS; goto out; } cmcred = (struct cmsgcred *) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); cmcred->cmcred_pid = p->p_pid; cmcred->cmcred_uid = td->td_ucred->cr_ruid; cmcred->cmcred_gid = td->td_ucred->cr_rgid; cmcred->cmcred_euid = td->td_ucred->cr_uid; cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); for (i = 0; i < cmcred->cmcred_ngroups; i++) cmcred->cmcred_groups[i] = td->td_ucred->cr_groups[i]; break; case SCM_RIGHTS: oldfds = datalen / sizeof (int); if (oldfds == 0) break; /* * Check that all the FDs passed in refer to legal * files. If not, reject the entire operation. */ fdp = data; FILEDESC_SLOCK(fdesc); for (i = 0; i < oldfds; i++, fdp++) { fp = fget_locked(fdesc, *fdp); if (fp == NULL) { FILEDESC_SUNLOCK(fdesc); error = EBADF; goto out; } if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { FILEDESC_SUNLOCK(fdesc); error = EOPNOTSUPP; goto out; } } /* * Now replace the integer FDs with pointers to the * file structure and capability rights. */ newlen = oldfds * sizeof(fdep[0]); *controlp = sbcreatecontrol(NULL, newlen, SCM_RIGHTS, SOL_SOCKET); if (*controlp == NULL) { FILEDESC_SUNLOCK(fdesc); error = E2BIG; goto out; } fdp = data; fdep = (struct filedescent **) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS, M_WAITOK); for (i = 0; i < oldfds; i++, fdev++, fdp++) { fde = &fdesc->fd_ofiles[*fdp]; fdep[i] = fdev; fdep[i]->fde_file = fde->fde_file; filecaps_copy(&fde->fde_caps, &fdep[i]->fde_caps, true); unp_internalize_fp(fdep[i]->fde_file); } FILEDESC_SUNLOCK(fdesc); break; case SCM_TIMESTAMP: *controlp = sbcreatecontrol(NULL, sizeof(*tv), SCM_TIMESTAMP, SOL_SOCKET); if (*controlp == NULL) { error = ENOBUFS; goto out; } tv = (struct timeval *) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); microtime(tv); break; case SCM_BINTIME: *controlp = sbcreatecontrol(NULL, sizeof(*bt), SCM_BINTIME, SOL_SOCKET); if (*controlp == NULL) { error = ENOBUFS; goto out; } bt = (struct bintime *) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); bintime(bt); break; default: error = EINVAL; goto out; } controlp = &(*controlp)->m_next; if (CMSG_SPACE(datalen) < clen) { clen -= CMSG_SPACE(datalen); cm = (struct cmsghdr *) ((caddr_t)cm + CMSG_SPACE(datalen)); } else { clen = 0; cm = NULL; } } out: m_freem(control); return (error); } static struct mbuf * unp_addsockcred(struct thread *td, struct mbuf *control) { struct mbuf *m, *n, *n_prev; struct sockcred *sc; const struct cmsghdr *cm; int ngroups; int i; ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); if (m == NULL) return (control); sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); sc->sc_uid = td->td_ucred->cr_ruid; sc->sc_euid = td->td_ucred->cr_uid; sc->sc_gid = td->td_ucred->cr_rgid; sc->sc_egid = td->td_ucred->cr_gid; sc->sc_ngroups = ngroups; for (i = 0; i < sc->sc_ngroups; i++) sc->sc_groups[i] = td->td_ucred->cr_groups[i]; /* * Unlink SCM_CREDS control messages (struct cmsgcred), since just * created SCM_CREDS control message (struct sockcred) has another * format. */ if (control != NULL) for (n = control, n_prev = NULL; n != NULL;) { cm = mtod(n, struct cmsghdr *); if (cm->cmsg_level == SOL_SOCKET && cm->cmsg_type == SCM_CREDS) { if (n_prev == NULL) control = n->m_next; else n_prev->m_next = n->m_next; n = m_free(n); } else { n_prev = n; n = n->m_next; } } /* Prepend it to the head. */ m->m_next = control; return (m); } static struct unpcb * fptounp(struct file *fp) { struct socket *so; if (fp->f_type != DTYPE_SOCKET) return (NULL); if ((so = fp->f_data) == NULL) return (NULL); if (so->so_proto->pr_domain != &localdomain) return (NULL); return sotounpcb(so); } static void unp_discard(struct file *fp) { struct unp_defer *dr; if (unp_externalize_fp(fp)) { dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK); dr->ud_fp = fp; UNP_DEFERRED_LOCK(); SLIST_INSERT_HEAD(&unp_defers, dr, ud_link); UNP_DEFERRED_UNLOCK(); atomic_add_int(&unp_defers_count, 1); taskqueue_enqueue(taskqueue_thread, &unp_defer_task); } else (void) closef(fp, (struct thread *)NULL); } static void unp_process_defers(void *arg __unused, int pending) { struct unp_defer *dr; SLIST_HEAD(, unp_defer) drl; int count; SLIST_INIT(&drl); for (;;) { UNP_DEFERRED_LOCK(); if (SLIST_FIRST(&unp_defers) == NULL) { UNP_DEFERRED_UNLOCK(); break; } SLIST_SWAP(&unp_defers, &drl, unp_defer); UNP_DEFERRED_UNLOCK(); count = 0; while ((dr = SLIST_FIRST(&drl)) != NULL) { SLIST_REMOVE_HEAD(&drl, ud_link); closef(dr->ud_fp, NULL); free(dr, M_TEMP); count++; } atomic_add_int(&unp_defers_count, -count); } } static void unp_internalize_fp(struct file *fp) { struct unpcb *unp; UNP_LINK_WLOCK(); if ((unp = fptounp(fp)) != NULL) { unp->unp_file = fp; unp->unp_msgcount++; } fhold(fp); unp_rights++; UNP_LINK_WUNLOCK(); } static int unp_externalize_fp(struct file *fp) { struct unpcb *unp; int ret; UNP_LINK_WLOCK(); if ((unp = fptounp(fp)) != NULL) { unp->unp_msgcount--; ret = 1; } else ret = 0; unp_rights--; UNP_LINK_WUNLOCK(); return (ret); } /* * unp_defer indicates whether additional work has been defered for a future * pass through unp_gc(). It is thread local and does not require explicit * synchronization. */ static int unp_marked; static int unp_unreachable; static void unp_accessable(struct filedescent **fdep, int fdcount) { struct unpcb *unp; struct file *fp; int i; for (i = 0; i < fdcount; i++) { fp = fdep[i]->fde_file; if ((unp = fptounp(fp)) == NULL) continue; if (unp->unp_gcflag & UNPGC_REF) continue; unp->unp_gcflag &= ~UNPGC_DEAD; unp->unp_gcflag |= UNPGC_REF; unp_marked++; } } static void unp_gc_process(struct unpcb *unp) { struct socket *soa; struct socket *so; struct file *fp; /* Already processed. */ if (unp->unp_gcflag & UNPGC_SCANNED) return; fp = unp->unp_file; /* * Check for a socket potentially in a cycle. It must be in a * queue as indicated by msgcount, and this must equal the file * reference count. Note that when msgcount is 0 the file is NULL. */ if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp && unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) { unp->unp_gcflag |= UNPGC_DEAD; unp_unreachable++; return; } /* * Mark all sockets we reference with RIGHTS. */ so = unp->unp_socket; if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) { SOCKBUF_LOCK(&so->so_rcv); unp_scan(so->so_rcv.sb_mb, unp_accessable); SOCKBUF_UNLOCK(&so->so_rcv); } /* * Mark all sockets in our accept queue. */ ACCEPT_LOCK(); TAILQ_FOREACH(soa, &so->so_comp, so_list) { if ((sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS) != 0) continue; SOCKBUF_LOCK(&soa->so_rcv); unp_scan(soa->so_rcv.sb_mb, unp_accessable); SOCKBUF_UNLOCK(&soa->so_rcv); } ACCEPT_UNLOCK(); unp->unp_gcflag |= UNPGC_SCANNED; } static int unp_recycled; SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, "Number of unreachable sockets claimed by the garbage collector."); static int unp_taskcount; SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, "Number of times the garbage collector has run."); static void unp_gc(__unused void *arg, int pending) { struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead, NULL }; struct unp_head **head; struct file *f, **unref; struct unpcb *unp; int i, total; unp_taskcount++; UNP_LIST_LOCK(); /* * First clear all gc flags from previous runs, apart from * UNPGC_IGNORE_RIGHTS. */ for (head = heads; *head != NULL; head++) LIST_FOREACH(unp, *head, unp_link) unp->unp_gcflag = (unp->unp_gcflag & UNPGC_IGNORE_RIGHTS); /* * Scan marking all reachable sockets with UNPGC_REF. Once a socket * is reachable all of the sockets it references are reachable. * Stop the scan once we do a complete loop without discovering * a new reachable socket. */ do { unp_unreachable = 0; unp_marked = 0; for (head = heads; *head != NULL; head++) LIST_FOREACH(unp, *head, unp_link) unp_gc_process(unp); } while (unp_marked); UNP_LIST_UNLOCK(); if (unp_unreachable == 0) return; /* * Allocate space for a local list of dead unpcbs. */ unref = malloc(unp_unreachable * sizeof(struct file *), M_TEMP, M_WAITOK); /* * Iterate looking for sockets which have been specifically marked * as as unreachable and store them locally. */ UNP_LINK_RLOCK(); UNP_LIST_LOCK(); for (total = 0, head = heads; *head != NULL; head++) LIST_FOREACH(unp, *head, unp_link) if ((unp->unp_gcflag & UNPGC_DEAD) != 0) { f = unp->unp_file; if (unp->unp_msgcount == 0 || f == NULL || f->f_count != unp->unp_msgcount) continue; unref[total++] = f; fhold(f); KASSERT(total <= unp_unreachable, ("unp_gc: incorrect unreachable count.")); } UNP_LIST_UNLOCK(); UNP_LINK_RUNLOCK(); /* * Now flush all sockets, free'ing rights. This will free the * struct files associated with these sockets but leave each socket * with one remaining ref. */ for (i = 0; i < total; i++) { struct socket *so; so = unref[i]->f_data; CURVNET_SET(so->so_vnet); sorflush(so); CURVNET_RESTORE(); } /* * And finally release the sockets so they can be reclaimed. */ for (i = 0; i < total; i++) fdrop(unref[i], NULL); unp_recycled += total; free(unref, M_TEMP); } static void unp_dispose_mbuf(struct mbuf *m) { if (m) unp_scan(m, unp_freerights); } /* * Synchronize against unp_gc, which can trip over data as we are freeing it. */ static void unp_dispose(struct socket *so) { struct unpcb *unp; unp = sotounpcb(so); UNP_LIST_LOCK(); unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS; UNP_LIST_UNLOCK(); unp_dispose_mbuf(so->so_rcv.sb_mb); } static void unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int)) { struct mbuf *m; struct cmsghdr *cm; void *data; socklen_t clen, datalen; while (m0 != NULL) { for (m = m0; m; m = m->m_next) { if (m->m_type != MT_CONTROL) continue; cm = mtod(m, struct cmsghdr *); clen = m->m_len; while (cm != NULL) { if (sizeof(*cm) > clen || cm->cmsg_len > clen) break; data = CMSG_DATA(cm); datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; if (cm->cmsg_level == SOL_SOCKET && cm->cmsg_type == SCM_RIGHTS) { (*op)(data, datalen / sizeof(struct filedescent *)); } if (CMSG_SPACE(datalen) < clen) { clen -= CMSG_SPACE(datalen); cm = (struct cmsghdr *) ((caddr_t)cm + CMSG_SPACE(datalen)); } else { clen = 0; cm = NULL; } } } m0 = m0->m_nextpkt; } } /* * A helper function called by VFS before socket-type vnode reclamation. * For an active vnode it clears unp_vnode pointer and decrements unp_vnode * use count. */ void vfs_unp_reclaim(struct vnode *vp) { struct socket *so; struct unpcb *unp; int active; ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim"); KASSERT(vp->v_type == VSOCK, ("vfs_unp_reclaim: vp->v_type != VSOCK")); active = 0; UNP_LINK_WLOCK(); VOP_UNP_CONNECT(vp, &so); if (so == NULL) goto done; unp = sotounpcb(so); if (unp == NULL) goto done; UNP_PCB_LOCK(unp); if (unp->unp_vnode == vp) { VOP_UNP_DETACH(vp); unp->unp_vnode = NULL; active = 1; } UNP_PCB_UNLOCK(unp); done: UNP_LINK_WUNLOCK(); if (active) vunref(vp); } #ifdef DDB static void db_print_indent(int indent) { int i; for (i = 0; i < indent; i++) db_printf(" "); } static void db_print_unpflags(int unp_flags) { int comma; comma = 0; if (unp_flags & UNP_HAVEPC) { db_printf("%sUNP_HAVEPC", comma ? ", " : ""); comma = 1; } if (unp_flags & UNP_HAVEPCCACHED) { db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : ""); comma = 1; } if (unp_flags & UNP_WANTCRED) { db_printf("%sUNP_WANTCRED", comma ? ", " : ""); comma = 1; } if (unp_flags & UNP_CONNWAIT) { db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); comma = 1; } if (unp_flags & UNP_CONNECTING) { db_printf("%sUNP_CONNECTING", comma ? ", " : ""); comma = 1; } if (unp_flags & UNP_BINDING) { db_printf("%sUNP_BINDING", comma ? ", " : ""); comma = 1; } } static void db_print_xucred(int indent, struct xucred *xu) { int comma, i; db_print_indent(indent); db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n", xu->cr_version, xu->cr_uid, xu->cr_ngroups); db_print_indent(indent); db_printf("cr_groups: "); comma = 0; for (i = 0; i < xu->cr_ngroups; i++) { db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); comma = 1; } db_printf("\n"); } static void db_print_unprefs(int indent, struct unp_head *uh) { struct unpcb *unp; int counter; counter = 0; LIST_FOREACH(unp, uh, unp_reflink) { if (counter % 4 == 0) db_print_indent(indent); db_printf("%p ", unp); if (counter % 4 == 3) db_printf("\n"); counter++; } if (counter != 0 && counter % 4 != 0) db_printf("\n"); } DB_SHOW_COMMAND(unpcb, db_show_unpcb) { struct unpcb *unp; if (!have_addr) { db_printf("usage: show unpcb \n"); return; } unp = (struct unpcb *)addr; db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, unp->unp_vnode); db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino, unp->unp_conn); db_printf("unp_refs:\n"); db_print_unprefs(2, &unp->unp_refs); /* XXXRW: Would be nice to print the full address, if any. */ db_printf("unp_addr: %p\n", unp->unp_addr); db_printf("unp_gencnt: %llu\n", (unsigned long long)unp->unp_gencnt); db_printf("unp_flags: %x (", unp->unp_flags); db_print_unpflags(unp->unp_flags); db_printf(")\n"); db_printf("unp_peercred:\n"); db_print_xucred(2, &unp->unp_peercred); db_printf("unp_refcount: %u\n", unp->unp_refcount); } #endif Index: head/sys/kern/vfs_cache.c =================================================================== --- head/sys/kern/vfs_cache.c (revision 305831) +++ head/sys/kern/vfs_cache.c (revision 305832) @@ -1,1739 +1,1739 @@ /*- * Copyright (c) 1989, 1993, 1995 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Poul-Henning Kamp of the FreeBSD Project. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_cache.c 8.5 (Berkeley) 3/22/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_ktrace.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KTRACE #include #endif #include SDT_PROVIDER_DECLARE(vfs); SDT_PROBE_DEFINE3(vfs, namecache, enter, done, "struct vnode *", "char *", "struct vnode *"); SDT_PROBE_DEFINE2(vfs, namecache, enter_negative, done, "struct vnode *", "char *"); SDT_PROBE_DEFINE1(vfs, namecache, fullpath, entry, "struct vnode *"); SDT_PROBE_DEFINE3(vfs, namecache, fullpath, hit, "struct vnode *", "char *", "struct vnode *"); SDT_PROBE_DEFINE1(vfs, namecache, fullpath, miss, "struct vnode *"); SDT_PROBE_DEFINE3(vfs, namecache, fullpath, return, "int", "struct vnode *", "char *"); SDT_PROBE_DEFINE3(vfs, namecache, lookup, hit, "struct vnode *", "char *", "struct vnode *"); SDT_PROBE_DEFINE2(vfs, namecache, lookup, hit__negative, "struct vnode *", "char *"); SDT_PROBE_DEFINE2(vfs, namecache, lookup, miss, "struct vnode *", "char *"); SDT_PROBE_DEFINE1(vfs, namecache, purge, done, "struct vnode *"); SDT_PROBE_DEFINE1(vfs, namecache, purge_negative, done, "struct vnode *"); SDT_PROBE_DEFINE1(vfs, namecache, purgevfs, done, "struct mount *"); SDT_PROBE_DEFINE3(vfs, namecache, zap, done, "struct vnode *", "char *", "struct vnode *"); SDT_PROBE_DEFINE2(vfs, namecache, zap_negative, done, "struct vnode *", "char *"); /* * This structure describes the elements in the cache of recent * names looked up by namei. */ struct namecache { LIST_ENTRY(namecache) nc_hash; /* hash chain */ LIST_ENTRY(namecache) nc_src; /* source vnode list */ TAILQ_ENTRY(namecache) nc_dst; /* destination vnode list */ struct vnode *nc_dvp; /* vnode of parent of name */ struct vnode *nc_vp; /* vnode the name refers to */ u_char nc_flag; /* flag bits */ u_char nc_nlen; /* length of name */ char nc_name[0]; /* segment name + nul */ }; /* * struct namecache_ts repeats struct namecache layout up to the * nc_nlen member. * struct namecache_ts is used in place of struct namecache when time(s) need * to be stored. The nc_dotdottime field is used when a cache entry is mapping * both a non-dotdot directory name plus dotdot for the directory's * parent. */ struct namecache_ts { LIST_ENTRY(namecache) nc_hash; /* hash chain */ LIST_ENTRY(namecache) nc_src; /* source vnode list */ TAILQ_ENTRY(namecache) nc_dst; /* destination vnode list */ struct vnode *nc_dvp; /* vnode of parent of name */ struct vnode *nc_vp; /* vnode the name refers to */ u_char nc_flag; /* flag bits */ u_char nc_nlen; /* length of name */ struct timespec nc_time; /* timespec provided by fs */ struct timespec nc_dotdottime; /* dotdot timespec provided by fs */ int nc_ticks; /* ticks value when entry was added */ char nc_name[0]; /* segment name + nul */ }; /* * Flags in namecache.nc_flag */ #define NCF_WHITE 0x01 #define NCF_ISDOTDOT 0x02 #define NCF_TS 0x04 #define NCF_DTS 0x08 #define NCF_DVDROP 0x10 /* * Name caching works as follows: * * Names found by directory scans are retained in a cache * for future reference. It is managed LRU, so frequently * used names will hang around. Cache is indexed by hash value * obtained from (vp, name) where vp refers to the directory * containing name. * * If it is a "negative" entry, (i.e. for a name that is known NOT to * exist) the vnode pointer will be NULL. * * Upon reaching the last segment of a path, if the reference * is for DELETE, or NOCACHE is set (rewrite), and the * name is located in the cache, it will be dropped. * * These locks are used (in the order in which they can be taken): * NAME TYPE ROLE * cache_lock rwlock global, needed for all modifications * bucketlock rwlock for access to given hash bucket * ncneg_mtx mtx negative entry LRU management * * A name -> vnode lookup can be safely performed by either locking cache_lock * or the relevant hash bucket. * * ".." and vnode -> name lookups require cache_lock. * * Modifications require both cache_lock and relevant bucketlock taken for * writing. * * Negative entry LRU management requires ncneg_mtx taken on top of either * cache_lock or bucketlock. */ /* * Structures associated with name caching. */ #define NCHHASH(hash) \ (&nchashtbl[(hash) & nchash]) static LIST_HEAD(nchashhead, namecache) *nchashtbl; /* Hash Table */ static TAILQ_HEAD(, namecache) ncneg; /* Hash Table */ static u_long nchash; /* size of hash table */ SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0, "Size of namecache hash table"); static u_long ncnegfactor = 16; /* ratio of negative entries */ SYSCTL_ULONG(_vfs, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0, "Ratio of negative namecache entries"); static u_long numneg; /* number of negative entries allocated */ SYSCTL_ULONG(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0, "Number of negative entries in namecache"); static u_long numcache; /* number of cache entries allocated */ SYSCTL_ULONG(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0, "Number of namecache entries"); static u_long numcachehv; /* number of cache entries with vnodes held */ SYSCTL_ULONG(_debug, OID_AUTO, numcachehv, CTLFLAG_RD, &numcachehv, 0, "Number of namecache entries with vnodes held"); u_int ncsizefactor = 2; SYSCTL_UINT(_vfs, OID_AUTO, ncsizefactor, CTLFLAG_RW, &ncsizefactor, 0, "Size factor for namecache"); struct nchstats nchstats; /* cache effectiveness statistics */ static struct rwlock cache_lock; RW_SYSINIT(vfscache, &cache_lock, "ncglobal"); #define CACHE_TRY_WLOCK() rw_try_wlock(&cache_lock) #define CACHE_UPGRADE_LOCK() rw_try_upgrade(&cache_lock) #define CACHE_RLOCK() rw_rlock(&cache_lock) #define CACHE_RUNLOCK() rw_runlock(&cache_lock) #define CACHE_WLOCK() rw_wlock(&cache_lock) #define CACHE_WUNLOCK() rw_wunlock(&cache_lock) static struct mtx_padalign ncneg_mtx; MTX_SYSINIT(vfscache_neg, &ncneg_mtx, "ncneg", MTX_DEF); static u_int numbucketlocks; static struct rwlock_padalign *bucketlocks; #define HASH2BUCKETLOCK(hash) \ ((struct rwlock *)(&bucketlocks[((hash) % numbucketlocks)])) /* * UMA zones for the VFS cache. * * The small cache is used for entries with short names, which are the * most common. The large cache is used for entries which are too big to * fit in the small cache. */ static uma_zone_t cache_zone_small; static uma_zone_t cache_zone_small_ts; static uma_zone_t cache_zone_large; static uma_zone_t cache_zone_large_ts; #define CACHE_PATH_CUTOFF 35 static struct namecache * cache_alloc(int len, int ts) { if (len > CACHE_PATH_CUTOFF) { if (ts) return (uma_zalloc(cache_zone_large_ts, M_WAITOK)); else return (uma_zalloc(cache_zone_large, M_WAITOK)); } if (ts) return (uma_zalloc(cache_zone_small_ts, M_WAITOK)); else return (uma_zalloc(cache_zone_small, M_WAITOK)); } static void cache_free(struct namecache *ncp) { int ts; if (ncp == NULL) return; ts = ncp->nc_flag & NCF_TS; if ((ncp->nc_flag & NCF_DVDROP) != 0) vdrop(ncp->nc_dvp); if (ncp->nc_nlen <= CACHE_PATH_CUTOFF) { if (ts) uma_zfree(cache_zone_small_ts, ncp); else uma_zfree(cache_zone_small, ncp); } else if (ts) uma_zfree(cache_zone_large_ts, ncp); else uma_zfree(cache_zone_large, ncp); } static char * nc_get_name(struct namecache *ncp) { struct namecache_ts *ncp_ts; if ((ncp->nc_flag & NCF_TS) == 0) return (ncp->nc_name); ncp_ts = (struct namecache_ts *)ncp; return (ncp_ts->nc_name); } static void cache_out_ts(struct namecache *ncp, struct timespec *tsp, int *ticksp) { KASSERT((ncp->nc_flag & NCF_TS) != 0 || (tsp == NULL && ticksp == NULL), ("No NCF_TS")); if (tsp != NULL) *tsp = ((struct namecache_ts *)ncp)->nc_time; if (ticksp != NULL) *ticksp = ((struct namecache_ts *)ncp)->nc_ticks; } static int doingcache = 1; /* 1 => enable the cache */ SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0, "VFS namecache enabled"); /* Export size information to userland */ SYSCTL_INT(_debug_sizeof, OID_AUTO, namecache, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, sizeof(struct namecache), "sizeof(struct namecache)"); /* * The new name cache statistics */ static SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0, "Name cache statistics"); #define STATNODE_ULONG(name, descr) \ SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, descr); #define STATNODE_COUNTER(name, descr) \ static counter_u64_t name; \ SYSCTL_COUNTER_U64(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, descr); STATNODE_ULONG(numneg, "Number of negative cache entries"); STATNODE_ULONG(numcache, "Number of cache entries"); STATNODE_COUNTER(numcalls, "Number of cache lookups"); STATNODE_COUNTER(dothits, "Number of '.' hits"); STATNODE_COUNTER(dotdothits, "Number of '..' hits"); STATNODE_COUNTER(numchecks, "Number of checks in lookup"); STATNODE_COUNTER(nummiss, "Number of cache misses"); STATNODE_COUNTER(nummisszap, "Number of cache misses we do not want to cache"); STATNODE_COUNTER(numposzaps, "Number of cache hits (positive) we do not want to cache"); STATNODE_COUNTER(numposhits, "Number of cache hits (positive)"); STATNODE_COUNTER(numnegzaps, "Number of cache hits (negative) we do not want to cache"); STATNODE_COUNTER(numneghits, "Number of cache hits (negative)"); /* These count for kern___getcwd(), too. */ STATNODE_COUNTER(numfullpathcalls, "Number of fullpath search calls"); STATNODE_COUNTER(numfullpathfail1, "Number of fullpath search errors (ENOTDIR)"); STATNODE_COUNTER(numfullpathfail2, "Number of fullpath search errors (VOP_VPTOCNP failures)"); STATNODE_COUNTER(numfullpathfail4, "Number of fullpath search errors (ENOMEM)"); STATNODE_COUNTER(numfullpathfound, "Number of successful fullpath calls"); static long numupgrades; STATNODE_ULONG(numupgrades, "Number of updates of the cache after lookup (write lock + retry)"); static long zap_and_exit_bucket_fail; STATNODE_ULONG(zap_and_exit_bucket_fail, "Number of times bucketlocked zap_and_exit case failed to writelock"); static void cache_zap(struct namecache *ncp); static int vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen); static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir, char *buf, char **retbuf, u_int buflen); static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries"); static uint32_t cache_get_hash(char *name, u_char len, struct vnode *dvp) { uint32_t hash; hash = fnv_32_buf(name, len, FNV1_32_INIT); hash = fnv_32_buf(&dvp, sizeof(dvp), hash); return (hash); } #ifdef INVARIANTS static void cache_assert_bucket_locked(struct namecache *ncp, int mode) { struct rwlock *bucketlock; uint32_t hash; hash = cache_get_hash(nc_get_name(ncp), ncp->nc_nlen, ncp->nc_dvp); bucketlock = HASH2BUCKETLOCK(hash); rw_assert(bucketlock, mode); } #else #define cache_assert_bucket_locked(x, y) do { } while (0) #endif static void cache_lock_all_buckets(void) { u_int i; for (i = 0; i < numbucketlocks; i++) rw_wlock(&bucketlocks[i]); } static void cache_unlock_all_buckets(void) { u_int i; for (i = 0; i < numbucketlocks; i++) rw_wunlock(&bucketlocks[i]); } static int sysctl_nchstats(SYSCTL_HANDLER_ARGS) { struct nchstats snap; if (req->oldptr == NULL) return (SYSCTL_OUT(req, 0, sizeof(snap))); snap = nchstats; snap.ncs_goodhits = counter_u64_fetch(numposhits); snap.ncs_neghits = counter_u64_fetch(numneghits); snap.ncs_badhits = counter_u64_fetch(numposzaps) + counter_u64_fetch(numnegzaps); snap.ncs_miss = counter_u64_fetch(nummisszap) + counter_u64_fetch(nummiss); return (SYSCTL_OUT(req, &snap, sizeof(snap))); } SYSCTL_PROC(_vfs_cache, OID_AUTO, nchstats, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, sysctl_nchstats, "LU", "VFS cache effectiveness statistics"); #ifdef DIAGNOSTIC /* * Grab an atomic snapshot of the name cache hash chain lengths */ static SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL, "hash table stats"); static int sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS) { struct nchashhead *ncpp; struct namecache *ncp; int i, error, n_nchash, *cntbuf; retry: n_nchash = nchash + 1; /* nchash is max index, not count */ if (req->oldptr == NULL) return SYSCTL_OUT(req, 0, n_nchash * sizeof(int)); cntbuf = malloc(n_nchash * sizeof(int), M_TEMP, M_ZERO | M_WAITOK); CACHE_RLOCK(); if (n_nchash != nchash + 1) { CACHE_RUNLOCK(); free(cntbuf, M_TEMP); goto retry; } /* Scan hash tables counting entries */ for (ncpp = nchashtbl, i = 0; i < n_nchash; ncpp++, i++) LIST_FOREACH(ncp, ncpp, nc_hash) cntbuf[i]++; CACHE_RUNLOCK(); for (error = 0, i = 0; i < n_nchash; i++) if ((error = SYSCTL_OUT(req, &cntbuf[i], sizeof(int))) != 0) break; free(cntbuf, M_TEMP); return (error); } SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD| CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_rawnchash, "S,int", "nchash chain lengths"); static int sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS) { int error; struct nchashhead *ncpp; struct namecache *ncp; int n_nchash; int count, maxlength, used, pct; if (!req->oldptr) return SYSCTL_OUT(req, 0, 4 * sizeof(int)); CACHE_RLOCK(); n_nchash = nchash + 1; /* nchash is max index, not count */ used = 0; maxlength = 0; /* Scan hash tables for applicable entries */ for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) { count = 0; LIST_FOREACH(ncp, ncpp, nc_hash) { count++; } if (count) used++; if (maxlength < count) maxlength = count; } n_nchash = nchash + 1; CACHE_RUNLOCK(); pct = (used * 100) / (n_nchash / 100); error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash)); if (error) return (error); error = SYSCTL_OUT(req, &used, sizeof(used)); if (error) return (error); error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength)); if (error) return (error); error = SYSCTL_OUT(req, &pct, sizeof(pct)); if (error) return (error); return (0); } SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD| CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_nchash, "I", "nchash statistics (number of total/used buckets, maximum chain length, usage percentage)"); #endif /* * Negative entries management */ static void cache_negative_hit(struct namecache *ncp) { mtx_lock(&ncneg_mtx); TAILQ_REMOVE(&ncneg, ncp, nc_dst); TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst); mtx_unlock(&ncneg_mtx); } static void cache_negative_insert(struct namecache *ncp) { rw_assert(&cache_lock, RA_WLOCKED); cache_assert_bucket_locked(ncp, RA_WLOCKED); MPASS(ncp->nc_vp == NULL); mtx_lock(&ncneg_mtx); TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst); numneg++; mtx_unlock(&ncneg_mtx); } static void cache_negative_remove(struct namecache *ncp) { rw_assert(&cache_lock, RA_WLOCKED); cache_assert_bucket_locked(ncp, RA_WLOCKED); MPASS(ncp->nc_vp == NULL); mtx_lock(&ncneg_mtx); TAILQ_REMOVE(&ncneg, ncp, nc_dst); numneg--; mtx_unlock(&ncneg_mtx); } static struct namecache * cache_negative_zap_one(void) { struct namecache *ncp; rw_assert(&cache_lock, RA_WLOCKED); ncp = TAILQ_FIRST(&ncneg); KASSERT(ncp->nc_vp == NULL, ("ncp %p vp %p on ncneg", ncp, ncp->nc_vp)); cache_zap(ncp); return (ncp); } /* * cache_zap(): * * Removes a namecache entry from cache, whether it contains an actual * pointer to a vnode or if it is just a negative cache entry. */ static void cache_zap_locked(struct namecache *ncp) { rw_assert(&cache_lock, RA_WLOCKED); cache_assert_bucket_locked(ncp, RA_WLOCKED); CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp, ncp->nc_vp); if (ncp->nc_vp != NULL) { SDT_PROBE3(vfs, namecache, zap, done, ncp->nc_dvp, nc_get_name(ncp), ncp->nc_vp); } else { SDT_PROBE2(vfs, namecache, zap_negative, done, ncp->nc_dvp, nc_get_name(ncp)); } LIST_REMOVE(ncp, nc_hash); if (ncp->nc_flag & NCF_ISDOTDOT) { if (ncp == ncp->nc_dvp->v_cache_dd) ncp->nc_dvp->v_cache_dd = NULL; } else { LIST_REMOVE(ncp, nc_src); if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) { ncp->nc_flag |= NCF_DVDROP; numcachehv--; } } if (ncp->nc_vp) { TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst); if (ncp == ncp->nc_vp->v_cache_dd) ncp->nc_vp->v_cache_dd = NULL; } else { cache_negative_remove(ncp); } numcache--; } static void cache_zap(struct namecache *ncp) { struct rwlock *bucketlock; uint32_t hash; rw_assert(&cache_lock, RA_WLOCKED); hash = cache_get_hash(nc_get_name(ncp), ncp->nc_nlen, ncp->nc_dvp); bucketlock = HASH2BUCKETLOCK(hash); rw_wlock(bucketlock); cache_zap_locked(ncp); rw_wunlock(bucketlock); } /* * Lookup an entry in the cache * * Lookup is called with dvp pointing to the directory to search, * cnp pointing to the name of the entry being sought. If the lookup * succeeds, the vnode is returned in *vpp, and a status of -1 is * returned. If the lookup determines that the name does not exist * (negative caching), a status of ENOENT is returned. If the lookup * fails, a status of zero is returned. If the directory vnode is * recycled out from under us due to a forced unmount, a status of * ENOENT is returned. * * vpp is locked and ref'd on return. If we're looking up DOTDOT, dvp is * unlocked. If we're looking up . an extra ref is taken, but the lock is * not recursively acquired. */ enum { UNLOCKED, WLOCKED, RLOCKED }; static void cache_unlock(int cache_locked) { switch (cache_locked) { case UNLOCKED: break; case WLOCKED: CACHE_WUNLOCK(); break; case RLOCKED: CACHE_RUNLOCK(); break; } } int cache_lookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct timespec *tsp, int *ticksp) { struct rwlock *bucketlock; struct namecache *ncp; uint32_t hash; int error, ltype, cache_locked; if (!doingcache) { cnp->cn_flags &= ~MAKEENTRY; return (0); } retry: bucketlock = NULL; cache_locked = UNLOCKED; error = 0; counter_u64_add(numcalls, 1); retry_wlocked: if (cnp->cn_nameptr[0] == '.') { if (cnp->cn_namelen == 1) { *vpp = dvp; CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .", dvp, cnp->cn_nameptr); counter_u64_add(dothits, 1); SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ".", *vpp); if (tsp != NULL) timespecclear(tsp); if (ticksp != NULL) *ticksp = ticks; VREF(*vpp); /* * When we lookup "." we still can be asked to lock it * differently... */ ltype = cnp->cn_lkflags & LK_TYPE_MASK; if (ltype != VOP_ISLOCKED(*vpp)) { if (ltype == LK_EXCLUSIVE) { vn_lock(*vpp, LK_UPGRADE | LK_RETRY); if ((*vpp)->v_iflag & VI_DOOMED) { /* forced unmount */ vrele(*vpp); *vpp = NULL; return (ENOENT); } } else vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY); } return (-1); } if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') { counter_u64_add(dotdothits, 1); if (cache_locked == UNLOCKED) { CACHE_RLOCK(); cache_locked = RLOCKED; } if (dvp->v_cache_dd == NULL) { SDT_PROBE3(vfs, namecache, lookup, miss, dvp, "..", NULL); goto unlock; } if ((cnp->cn_flags & MAKEENTRY) == 0) { if (cache_locked != WLOCKED && !CACHE_UPGRADE_LOCK()) goto wlock; ncp = NULL; if (dvp->v_cache_dd->nc_flag & NCF_ISDOTDOT) { ncp = dvp->v_cache_dd; cache_zap(ncp); } dvp->v_cache_dd = NULL; CACHE_WUNLOCK(); cache_free(ncp); return (0); } ncp = dvp->v_cache_dd; if (ncp->nc_flag & NCF_ISDOTDOT) *vpp = ncp->nc_vp; else *vpp = ncp->nc_dvp; /* Return failure if negative entry was found. */ if (*vpp == NULL) goto negative_success; CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..", dvp, cnp->cn_nameptr, *vpp); SDT_PROBE3(vfs, namecache, lookup, hit, dvp, "..", *vpp); cache_out_ts(ncp, tsp, ticksp); if ((ncp->nc_flag & (NCF_ISDOTDOT | NCF_DTS)) == NCF_DTS && tsp != NULL) *tsp = ((struct namecache_ts *)ncp)-> nc_dotdottime; goto success; } } hash = cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp); if (cache_locked == UNLOCKED) { bucketlock = HASH2BUCKETLOCK(hash); rw_rlock(bucketlock); } LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) { counter_u64_add(numchecks, 1); if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && !bcmp(nc_get_name(ncp), cnp->cn_nameptr, ncp->nc_nlen)) break; } /* We failed to find an entry */ if (ncp == NULL) { SDT_PROBE3(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr, NULL); if ((cnp->cn_flags & MAKEENTRY) == 0) { counter_u64_add(nummisszap, 1); } else { counter_u64_add(nummiss, 1); } goto unlock; } /* We don't want to have an entry, so dump it */ if ((cnp->cn_flags & MAKEENTRY) == 0) { counter_u64_add(numposzaps, 1); goto zap_and_exit; } /* We found a "positive" match, return the vnode */ if (ncp->nc_vp) { counter_u64_add(numposhits, 1); *vpp = ncp->nc_vp; CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p", dvp, cnp->cn_nameptr, *vpp, ncp); SDT_PROBE3(vfs, namecache, lookup, hit, dvp, nc_get_name(ncp), *vpp); cache_out_ts(ncp, tsp, ticksp); goto success; } negative_success: /* We found a negative match, and want to create it, so purge */ if (cnp->cn_nameiop == CREATE) { counter_u64_add(numnegzaps, 1); goto zap_and_exit; } counter_u64_add(numneghits, 1); cache_negative_hit(ncp); if (ncp->nc_flag & NCF_WHITE) cnp->cn_flags |= ISWHITEOUT; SDT_PROBE2(vfs, namecache, lookup, hit__negative, dvp, nc_get_name(ncp)); cache_out_ts(ncp, tsp, ticksp); MPASS(bucketlock != NULL || cache_locked != UNLOCKED); if (bucketlock != NULL) rw_runlock(bucketlock); cache_unlock(cache_locked); return (ENOENT); wlock: /* * We need to update the cache after our lookup, so upgrade to * a write lock and retry the operation. */ CACHE_RUNLOCK(); wlock_unlocked: CACHE_WLOCK(); numupgrades++; cache_locked = WLOCKED; goto retry_wlocked; success: /* * On success we return a locked and ref'd vnode as per the lookup * protocol. */ MPASS(dvp != *vpp); ltype = 0; /* silence gcc warning */ if (cnp->cn_flags & ISDOTDOT) { ltype = VOP_ISLOCKED(dvp); VOP_UNLOCK(dvp, 0); } vhold(*vpp); MPASS(bucketlock != NULL || cache_locked != UNLOCKED); if (bucketlock != NULL) rw_runlock(bucketlock); cache_unlock(cache_locked); error = vget(*vpp, cnp->cn_lkflags | LK_VNHELD, cnp->cn_thread); if (cnp->cn_flags & ISDOTDOT) { vn_lock(dvp, ltype | LK_RETRY); if (dvp->v_iflag & VI_DOOMED) { if (error == 0) vput(*vpp); *vpp = NULL; return (ENOENT); } } if (error) { *vpp = NULL; goto retry; } if ((cnp->cn_flags & ISLASTCN) && (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) { ASSERT_VOP_ELOCKED(*vpp, "cache_lookup"); } return (-1); unlock: MPASS(bucketlock != NULL || cache_locked != UNLOCKED); if (bucketlock != NULL) rw_runlock(bucketlock); cache_unlock(cache_locked); return (0); zap_and_exit: if (bucketlock != NULL) { rw_assert(&cache_lock, RA_UNLOCKED); if (!CACHE_TRY_WLOCK()) { rw_runlock(bucketlock); bucketlock = NULL; zap_and_exit_bucket_fail++; goto wlock_unlocked; } cache_locked = WLOCKED; rw_runlock(bucketlock); bucketlock = NULL; } else if (cache_locked != WLOCKED && !CACHE_UPGRADE_LOCK()) goto wlock; cache_zap(ncp); CACHE_WUNLOCK(); cache_free(ncp); return (0); } /* * Add an entry to the cache. */ void cache_enter_time(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, struct timespec *tsp, struct timespec *dtsp) { struct rwlock *bucketlock; struct namecache *ncp, *n2, *ndd, *nneg; struct namecache_ts *n3; struct nchashhead *ncpp; uint32_t hash; int flag; int len; CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr); VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp, ("cache_enter: Adding a doomed vnode")); VNASSERT(dvp == NULL || (dvp->v_iflag & VI_DOOMED) == 0, dvp, ("cache_enter: Doomed vnode used as src")); if (!doingcache) return; /* * Avoid blowout in namecache entries. */ if (numcache >= desiredvnodes * ncsizefactor) return; ndd = nneg = NULL; flag = 0; if (cnp->cn_nameptr[0] == '.') { if (cnp->cn_namelen == 1) return; if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') { CACHE_WLOCK(); /* * If dotdot entry already exists, just retarget it * to new parent vnode, otherwise continue with new * namecache entry allocation. */ if ((ncp = dvp->v_cache_dd) != NULL && ncp->nc_flag & NCF_ISDOTDOT) { KASSERT(ncp->nc_dvp == dvp, ("wrong isdotdot parent")); if (ncp->nc_vp != NULL) { TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst); } else { cache_negative_remove(ncp); } if (vp != NULL) { TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst); } else { cache_negative_insert(ncp); } ncp->nc_vp = vp; CACHE_WUNLOCK(); return; } dvp->v_cache_dd = NULL; SDT_PROBE3(vfs, namecache, enter, done, dvp, "..", vp); CACHE_WUNLOCK(); flag = NCF_ISDOTDOT; } } /* * Calculate the hash key and setup as much of the new * namecache entry as possible before acquiring the lock. */ ncp = cache_alloc(cnp->cn_namelen, tsp != NULL); ncp->nc_vp = vp; ncp->nc_dvp = dvp; ncp->nc_flag = flag; if (tsp != NULL) { n3 = (struct namecache_ts *)ncp; n3->nc_time = *tsp; n3->nc_ticks = ticks; n3->nc_flag |= NCF_TS; if (dtsp != NULL) { n3->nc_dotdottime = *dtsp; n3->nc_flag |= NCF_DTS; } } len = ncp->nc_nlen = cnp->cn_namelen; hash = cache_get_hash(cnp->cn_nameptr, len, dvp); strlcpy(nc_get_name(ncp), cnp->cn_nameptr, len + 1); CACHE_WLOCK(); /* * See if this vnode or negative entry is already in the cache * with this name. This can happen with concurrent lookups of * the same path name. */ ncpp = NCHHASH(hash); LIST_FOREACH(n2, ncpp, nc_hash) { if (n2->nc_dvp == dvp && n2->nc_nlen == cnp->cn_namelen && !bcmp(nc_get_name(n2), cnp->cn_nameptr, n2->nc_nlen)) { if (tsp != NULL) { KASSERT((n2->nc_flag & NCF_TS) != 0, ("no NCF_TS")); n3 = (struct namecache_ts *)n2; n3->nc_time = ((struct namecache_ts *)ncp)->nc_time; n3->nc_ticks = ((struct namecache_ts *)ncp)->nc_ticks; if (dtsp != NULL) { n3->nc_dotdottime = ((struct namecache_ts *)ncp)-> nc_dotdottime; n3->nc_flag |= NCF_DTS; } } CACHE_WUNLOCK(); cache_free(ncp); return; } } if (flag == NCF_ISDOTDOT) { /* * See if we are trying to add .. entry, but some other lookup * has populated v_cache_dd pointer already. */ if (dvp->v_cache_dd != NULL) { CACHE_WUNLOCK(); cache_free(ncp); return; } KASSERT(vp == NULL || vp->v_type == VDIR, ("wrong vnode type %p", vp)); dvp->v_cache_dd = ncp; } numcache++; if (vp != NULL) { if (vp->v_type == VDIR) { if (flag != NCF_ISDOTDOT) { /* * For this case, the cache entry maps both the * directory name in it and the name ".." for the * directory's parent. */ if ((ndd = vp->v_cache_dd) != NULL) { if ((ndd->nc_flag & NCF_ISDOTDOT) != 0) cache_zap(ndd); else ndd = NULL; } vp->v_cache_dd = ncp; } } else { vp->v_cache_dd = NULL; } } if (flag != NCF_ISDOTDOT) { if (LIST_EMPTY(&dvp->v_cache_src)) { vhold(dvp); numcachehv++; } LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src); } bucketlock = HASH2BUCKETLOCK(hash); rw_wlock(bucketlock); /* * Insert the new namecache entry into the appropriate chain * within the cache entries table. */ LIST_INSERT_HEAD(ncpp, ncp, nc_hash); /* * If the entry is "negative", we place it into the * "negative" cache queue, otherwise, we place it into the * destination vnode's cache entries queue. */ if (vp != NULL) { TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst); SDT_PROBE3(vfs, namecache, enter, done, dvp, nc_get_name(ncp), vp); } else { if (cnp->cn_flags & ISWHITEOUT) ncp->nc_flag |= NCF_WHITE; cache_negative_insert(ncp); SDT_PROBE2(vfs, namecache, enter_negative, done, dvp, nc_get_name(ncp)); } rw_wunlock(bucketlock); if (numneg * ncnegfactor > numcache) nneg = cache_negative_zap_one(); CACHE_WUNLOCK(); cache_free(ndd); cache_free(nneg); } static u_int cache_roundup_2(u_int val) { u_int res; for (res = 1; res <= val; res <<= 1) continue; return (res); } /* * Name cache initialization, from vfs_init() when we are booting */ static void nchinit(void *dummy __unused) { u_int i; TAILQ_INIT(&ncneg); cache_zone_small = uma_zcreate("S VFS Cache", sizeof(struct namecache) + CACHE_PATH_CUTOFF + 1, NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT); cache_zone_small_ts = uma_zcreate("STS VFS Cache", sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1, NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT); cache_zone_large = uma_zcreate("L VFS Cache", sizeof(struct namecache) + NAME_MAX + 1, NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT); cache_zone_large_ts = uma_zcreate("LTS VFS Cache", sizeof(struct namecache_ts) + NAME_MAX + 1, NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT); nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash); numbucketlocks = cache_roundup_2(mp_ncpus * 16); if (numbucketlocks > nchash) numbucketlocks = nchash; bucketlocks = malloc(sizeof(*bucketlocks) * numbucketlocks, M_VFSCACHE, M_WAITOK | M_ZERO); for (i = 0; i < numbucketlocks; i++) rw_init_flags(&bucketlocks[i], "ncbuc", RW_DUPOK); numcalls = counter_u64_alloc(M_WAITOK); dothits = counter_u64_alloc(M_WAITOK); dotdothits = counter_u64_alloc(M_WAITOK); numchecks = counter_u64_alloc(M_WAITOK); nummiss = counter_u64_alloc(M_WAITOK); nummisszap = counter_u64_alloc(M_WAITOK); numposzaps = counter_u64_alloc(M_WAITOK); numposhits = counter_u64_alloc(M_WAITOK); numnegzaps = counter_u64_alloc(M_WAITOK); numneghits = counter_u64_alloc(M_WAITOK); numfullpathcalls = counter_u64_alloc(M_WAITOK); numfullpathfail1 = counter_u64_alloc(M_WAITOK); numfullpathfail2 = counter_u64_alloc(M_WAITOK); numfullpathfail4 = counter_u64_alloc(M_WAITOK); numfullpathfound = counter_u64_alloc(M_WAITOK); } SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL); void cache_changesize(int newmaxvnodes) { struct nchashhead *new_nchashtbl, *old_nchashtbl; u_long new_nchash, old_nchash; struct namecache *ncp; uint32_t hash; int i; new_nchashtbl = hashinit(newmaxvnodes * 2, M_VFSCACHE, &new_nchash); /* If same hash table size, nothing to do */ if (nchash == new_nchash) { free(new_nchashtbl, M_VFSCACHE); return; } /* * Move everything from the old hash table to the new table. * None of the namecache entries in the table can be removed * because to do so, they have to be removed from the hash table. */ CACHE_WLOCK(); cache_lock_all_buckets(); old_nchashtbl = nchashtbl; old_nchash = nchash; nchashtbl = new_nchashtbl; nchash = new_nchash; for (i = 0; i <= old_nchash; i++) { while ((ncp = LIST_FIRST(&old_nchashtbl[i])) != NULL) { hash = cache_get_hash(nc_get_name(ncp), ncp->nc_nlen, ncp->nc_dvp); LIST_REMOVE(ncp, nc_hash); LIST_INSERT_HEAD(NCHHASH(hash), ncp, nc_hash); } } cache_unlock_all_buckets(); CACHE_WUNLOCK(); free(old_nchashtbl, M_VFSCACHE); } /* * Invalidate all entries to a particular vnode. */ void cache_purge(struct vnode *vp) { TAILQ_HEAD(, namecache) ncps; struct namecache *ncp, *nnp; CTR1(KTR_VFS, "cache_purge(%p)", vp); SDT_PROBE1(vfs, namecache, purge, done, vp); TAILQ_INIT(&ncps); CACHE_WLOCK(); while (!LIST_EMPTY(&vp->v_cache_src)) { ncp = LIST_FIRST(&vp->v_cache_src); cache_zap(ncp); TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); } while (!TAILQ_EMPTY(&vp->v_cache_dst)) { ncp = TAILQ_FIRST(&vp->v_cache_dst); cache_zap(ncp); TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); } if (vp->v_cache_dd != NULL) { ncp = vp->v_cache_dd; KASSERT(ncp->nc_flag & NCF_ISDOTDOT, ("lost dotdot link")); cache_zap(ncp); TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); } KASSERT(vp->v_cache_dd == NULL, ("incomplete purge")); CACHE_WUNLOCK(); TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) { cache_free(ncp); } } /* * Invalidate all negative entries for a particular directory vnode. */ void cache_purge_negative(struct vnode *vp) { TAILQ_HEAD(, namecache) ncps; struct namecache *ncp, *nnp; CTR1(KTR_VFS, "cache_purge_negative(%p)", vp); SDT_PROBE1(vfs, namecache, purge_negative, done, vp); TAILQ_INIT(&ncps); CACHE_WLOCK(); LIST_FOREACH_SAFE(ncp, &vp->v_cache_src, nc_src, nnp) { if (ncp->nc_vp != NULL) continue; cache_zap(ncp); TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); } CACHE_WUNLOCK(); TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) { cache_free(ncp); } } /* * Flush all entries referencing a particular filesystem. */ void cache_purgevfs(struct mount *mp) { TAILQ_HEAD(, namecache) ncps; struct rwlock *bucketlock; struct nchashhead *bucket; struct namecache *ncp, *nnp; u_long i, j, n_nchash; /* Scan hash tables for applicable entries */ SDT_PROBE1(vfs, namecache, purgevfs, done, mp); TAILQ_INIT(&ncps); CACHE_WLOCK(); n_nchash = nchash + 1; for (i = 0; i < numbucketlocks; i++) { bucketlock = (struct rwlock *)&bucketlocks[i]; rw_wlock(bucketlock); for (j = i; j < n_nchash; j += numbucketlocks) { bucket = &nchashtbl[j]; LIST_FOREACH_SAFE(ncp, bucket, nc_hash, nnp) { cache_assert_bucket_locked(ncp, RA_WLOCKED); if (ncp->nc_dvp->v_mount != mp) continue; cache_zap_locked(ncp); TAILQ_INSERT_HEAD(&ncps, ncp, nc_dst); } } rw_wunlock(bucketlock); } CACHE_WUNLOCK(); TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) { cache_free(ncp); } } /* * Perform canonical checks and cache lookup and pass on to filesystem * through the vop_cachedlookup only if needed. */ int vfs_cache_lookup(struct vop_lookup_args *ap) { struct vnode *dvp; int error; struct vnode **vpp = ap->a_vpp; struct componentname *cnp = ap->a_cnp; struct ucred *cred = cnp->cn_cred; int flags = cnp->cn_flags; struct thread *td = cnp->cn_thread; *vpp = NULL; dvp = ap->a_dvp; if (dvp->v_type != VDIR) return (ENOTDIR); if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) return (EROFS); error = VOP_ACCESS(dvp, VEXEC, cred, td); if (error) return (error); error = cache_lookup(dvp, vpp, cnp, NULL, NULL); if (error == 0) return (VOP_CACHEDLOOKUP(dvp, vpp, cnp)); if (error == -1) return (0); return (error); } /* * XXX All of these sysctls would probably be more productive dead. */ static int disablecwd; SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0, "Disable the getcwd syscall"); /* Implementation of the getcwd syscall. */ int sys___getcwd(struct thread *td, struct __getcwd_args *uap) { return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen, MAXPATHLEN)); } int kern___getcwd(struct thread *td, char *buf, enum uio_seg bufseg, u_int buflen, u_int path_max) { char *bp, *tmpbuf; struct filedesc *fdp; struct vnode *cdir, *rdir; int error; if (disablecwd) return (ENODEV); if (buflen < 2) return (EINVAL); if (buflen > path_max) buflen = path_max; tmpbuf = malloc(buflen, M_TEMP, M_WAITOK); fdp = td->td_proc->p_fd; FILEDESC_SLOCK(fdp); cdir = fdp->fd_cdir; VREF(cdir); rdir = fdp->fd_rdir; VREF(rdir); FILEDESC_SUNLOCK(fdp); error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen); vrele(rdir); vrele(cdir); if (!error) { if (bufseg == UIO_SYSSPACE) bcopy(bp, buf, strlen(bp) + 1); else error = copyout(bp, buf, strlen(bp) + 1); #ifdef KTRACE if (KTRPOINT(curthread, KTR_NAMEI)) ktrnamei(bp); #endif } free(tmpbuf, M_TEMP); return (error); } /* * Thus begins the fullpath magic. */ static int disablefullpath; SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0, "Disable the vn_fullpath function"); /* * Retrieve the full filesystem path that correspond to a vnode from the name * cache (if available) */ int vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf) { char *buf; struct filedesc *fdp; struct vnode *rdir; int error; if (disablefullpath) return (ENODEV); if (vn == NULL) return (EINVAL); buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); fdp = td->td_proc->p_fd; FILEDESC_SLOCK(fdp); rdir = fdp->fd_rdir; VREF(rdir); FILEDESC_SUNLOCK(fdp); error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN); vrele(rdir); if (!error) *freebuf = buf; else free(buf, M_TEMP); return (error); } /* * This function is similar to vn_fullpath, but it attempts to lookup the * pathname relative to the global root mount point. This is required for the * auditing sub-system, as audited pathnames must be absolute, relative to the * global root mount point. */ int vn_fullpath_global(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf) { char *buf; int error; if (disablefullpath) return (ENODEV); if (vn == NULL) return (EINVAL); buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN); if (!error) *freebuf = buf; else free(buf, M_TEMP); return (error); } int vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen) { int error; CACHE_RLOCK(); error = vn_vptocnp_locked(vp, cred, buf, buflen); if (error == 0) CACHE_RUNLOCK(); return (error); } static int vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen) { struct vnode *dvp; struct namecache *ncp; int error; TAILQ_FOREACH(ncp, &((*vp)->v_cache_dst), nc_dst) { if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) break; } if (ncp != NULL) { if (*buflen < ncp->nc_nlen) { CACHE_RUNLOCK(); vrele(*vp); counter_u64_add(numfullpathfail4, 1); error = ENOMEM; SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); return (error); } *buflen -= ncp->nc_nlen; memcpy(buf + *buflen, nc_get_name(ncp), ncp->nc_nlen); SDT_PROBE3(vfs, namecache, fullpath, hit, ncp->nc_dvp, nc_get_name(ncp), vp); dvp = *vp; *vp = ncp->nc_dvp; vref(*vp); CACHE_RUNLOCK(); vrele(dvp); CACHE_RLOCK(); return (0); } SDT_PROBE1(vfs, namecache, fullpath, miss, vp); CACHE_RUNLOCK(); vn_lock(*vp, LK_SHARED | LK_RETRY); error = VOP_VPTOCNP(*vp, &dvp, cred, buf, buflen); vput(*vp); if (error) { counter_u64_add(numfullpathfail2, 1); SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); return (error); } *vp = dvp; CACHE_RLOCK(); if (dvp->v_iflag & VI_DOOMED) { /* forced unmount */ CACHE_RUNLOCK(); vrele(dvp); error = ENOENT; SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); return (error); } /* * *vp has its use count incremented still. */ return (0); } /* * The magic behind kern___getcwd() and vn_fullpath(). */ static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir, char *buf, char **retbuf, u_int buflen) { int error, slash_prefixed; #ifdef KDTRACE_HOOKS struct vnode *startvp = vp; #endif struct vnode *vp1; buflen--; buf[buflen] = '\0'; error = 0; slash_prefixed = 0; SDT_PROBE1(vfs, namecache, fullpath, entry, vp); counter_u64_add(numfullpathcalls, 1); vref(vp); CACHE_RLOCK(); if (vp->v_type != VDIR) { error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen); if (error) return (error); if (buflen == 0) { CACHE_RUNLOCK(); vrele(vp); return (ENOMEM); } buf[--buflen] = '/'; slash_prefixed = 1; } while (vp != rdir && vp != rootvnode) { if (vp->v_vflag & VV_ROOT) { if (vp->v_iflag & VI_DOOMED) { /* forced unmount */ CACHE_RUNLOCK(); vrele(vp); error = ENOENT; SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); break; } vp1 = vp->v_mount->mnt_vnodecovered; vref(vp1); CACHE_RUNLOCK(); vrele(vp); vp = vp1; CACHE_RLOCK(); continue; } if (vp->v_type != VDIR) { CACHE_RUNLOCK(); vrele(vp); counter_u64_add(numfullpathfail1, 1); error = ENOTDIR; SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); break; } error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen); if (error) break; if (buflen == 0) { CACHE_RUNLOCK(); vrele(vp); error = ENOMEM; SDT_PROBE3(vfs, namecache, fullpath, return, error, startvp, NULL); break; } buf[--buflen] = '/'; slash_prefixed = 1; } if (error) return (error); if (!slash_prefixed) { if (buflen == 0) { CACHE_RUNLOCK(); vrele(vp); counter_u64_add(numfullpathfail4, 1); SDT_PROBE3(vfs, namecache, fullpath, return, ENOMEM, startvp, NULL); return (ENOMEM); } buf[--buflen] = '/'; } counter_u64_add(numfullpathfound, 1); CACHE_RUNLOCK(); vrele(vp); SDT_PROBE3(vfs, namecache, fullpath, return, 0, startvp, buf + buflen); *retbuf = buf + buflen; return (0); } struct vnode * vn_dir_dd_ino(struct vnode *vp) { struct namecache *ncp; struct vnode *ddvp; ASSERT_VOP_LOCKED(vp, "vn_dir_dd_ino"); CACHE_RLOCK(); TAILQ_FOREACH(ncp, &(vp->v_cache_dst), nc_dst) { if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) continue; ddvp = ncp->nc_dvp; vhold(ddvp); CACHE_RUNLOCK(); if (vget(ddvp, LK_SHARED | LK_NOWAIT | LK_VNHELD, curthread)) return (NULL); return (ddvp); } CACHE_RUNLOCK(); return (NULL); } int vn_commname(struct vnode *vp, char *buf, u_int buflen) { struct namecache *ncp; int l; CACHE_RLOCK(); TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst) if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) break; if (ncp == NULL) { CACHE_RUNLOCK(); return (ENOENT); } l = min(ncp->nc_nlen, buflen - 1); memcpy(buf, nc_get_name(ncp), l); CACHE_RUNLOCK(); buf[l] = '\0'; return (0); } /* ABI compat shims for old kernel modules. */ #undef cache_enter void cache_enter(struct vnode *dvp, struct vnode *vp, struct componentname *cnp); void cache_enter(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) { cache_enter_time(dvp, vp, cnp, NULL, NULL); } /* * This function updates path string to vnode's full global path * and checks the size of the new path string against the pathlen argument. * * Requires a locked, referenced vnode. * Vnode is re-locked on success or ENODEV, otherwise unlocked. * * If sysctl debug.disablefullpath is set, ENODEV is returned, * vnode is left locked and path remain untouched. * * If vp is a directory, the call to vn_fullpath_global() always succeeds * because it falls back to the ".." lookup if the namecache lookup fails. */ int vn_path_to_global_path(struct thread *td, struct vnode *vp, char *path, u_int pathlen) { struct nameidata nd; struct vnode *vp1; char *rpath, *fbuf; int error; ASSERT_VOP_ELOCKED(vp, __func__); /* Return ENODEV if sysctl debug.disablefullpath==1 */ if (disablefullpath) return (ENODEV); /* Construct global filesystem path from vp. */ VOP_UNLOCK(vp, 0); error = vn_fullpath_global(td, vp, &rpath, &fbuf); if (error != 0) { vrele(vp); return (error); } if (strlen(rpath) >= pathlen) { vrele(vp); error = ENAMETOOLONG; goto out; } /* * Re-lookup the vnode by path to detect a possible rename. * As a side effect, the vnode is relocked. * If vnode was renamed, return ENOENT. */ NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE, path, td); error = namei(&nd); if (error != 0) { vrele(vp); goto out; } NDFREE(&nd, NDF_ONLY_PNBUF); vp1 = nd.ni_vp; vrele(vp); if (vp1 == vp) strcpy(path, rpath); else { vput(vp1); error = ENOENT; } out: free(fbuf, M_TEMP); return (error); } Index: head/sys/kern/vfs_cluster.c =================================================================== --- head/sys/kern/vfs_cluster.c (revision 305831) +++ head/sys/kern/vfs_cluster.c (revision 305832) @@ -1,1063 +1,1063 @@ /*- * Copyright (c) 1993 * The Regents of the University of California. All rights reserved. * Modifications/enhancements: * Copyright (c) 1995 John S. Dyson. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_cluster.c 8.7 (Berkeley) 2/13/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_debug_cluster.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #if defined(CLUSTERDEBUG) static int rcluster= 0; SYSCTL_INT(_debug, OID_AUTO, rcluster, CTLFLAG_RW, &rcluster, 0, "Debug VFS clustering code"); #endif static MALLOC_DEFINE(M_SEGMENT, "cl_savebuf", "cluster_save buffer"); static struct cluster_save *cluster_collectbufs(struct vnode *vp, struct buf *last_bp, int gbflags); static struct buf *cluster_rbuild(struct vnode *vp, u_quad_t filesize, daddr_t lbn, daddr_t blkno, long size, int run, int gbflags, struct buf *fbp); static void cluster_callback(struct buf *); static int write_behind = 1; SYSCTL_INT(_vfs, OID_AUTO, write_behind, CTLFLAG_RW, &write_behind, 0, "Cluster write-behind; 0: disable, 1: enable, 2: backed off"); static int read_max = 64; SYSCTL_INT(_vfs, OID_AUTO, read_max, CTLFLAG_RW, &read_max, 0, "Cluster read-ahead max block count"); static int read_min = 1; SYSCTL_INT(_vfs, OID_AUTO, read_min, CTLFLAG_RW, &read_min, 0, "Cluster read min block count"); /* Page expended to mark partially backed buffers */ extern vm_page_t bogus_page; /* * Read data to a buf, including read-ahead if we find this to be beneficial. * cluster_read replaces bread. */ int cluster_read(struct vnode *vp, u_quad_t filesize, daddr_t lblkno, long size, struct ucred *cred, long totread, int seqcount, int gbflags, struct buf **bpp) { struct buf *bp, *rbp, *reqbp; struct bufobj *bo; daddr_t blkno, origblkno; int maxra, racluster; int error, ncontig; int i; error = 0; bo = &vp->v_bufobj; if (!unmapped_buf_allowed) gbflags &= ~GB_UNMAPPED; /* * Try to limit the amount of read-ahead by a few * ad-hoc parameters. This needs work!!! */ racluster = vp->v_mount->mnt_iosize_max / size; maxra = seqcount; maxra = min(read_max, maxra); maxra = min(nbuf/8, maxra); if (((u_quad_t)(lblkno + maxra + 1) * size) > filesize) maxra = (filesize / size) - lblkno; /* * get the requested block */ *bpp = reqbp = bp = getblk(vp, lblkno, size, 0, 0, gbflags); if (bp == NULL) return (EBUSY); origblkno = lblkno; /* * if it is in the cache, then check to see if the reads have been * sequential. If they have, then try some read-ahead, otherwise * back-off on prospective read-aheads. */ if (bp->b_flags & B_CACHE) { if (!seqcount) { return 0; } else if ((bp->b_flags & B_RAM) == 0) { return 0; } else { bp->b_flags &= ~B_RAM; BO_RLOCK(bo); for (i = 1; i < maxra; i++) { /* * Stop if the buffer does not exist or it * is invalid (about to go away?) */ rbp = gbincore(&vp->v_bufobj, lblkno+i); if (rbp == NULL || (rbp->b_flags & B_INVAL)) break; /* * Set another read-ahead mark so we know * to check again. (If we can lock the * buffer without waiting) */ if ((((i % racluster) == (racluster - 1)) || (i == (maxra - 1))) && (0 == BUF_LOCK(rbp, LK_EXCLUSIVE | LK_NOWAIT, NULL))) { rbp->b_flags |= B_RAM; BUF_UNLOCK(rbp); } } BO_RUNLOCK(bo); if (i >= maxra) { return 0; } lblkno += i; } reqbp = bp = NULL; /* * If it isn't in the cache, then get a chunk from * disk if sequential, otherwise just get the block. */ } else { off_t firstread = bp->b_offset; int nblks; long minread; KASSERT(bp->b_offset != NOOFFSET, ("cluster_read: no buffer offset")); ncontig = 0; /* * Adjust totread if needed */ minread = read_min * size; if (minread > totread) totread = minread; /* * Compute the total number of blocks that we should read * synchronously. */ if (firstread + totread > filesize) totread = filesize - firstread; nblks = howmany(totread, size); if (nblks > racluster) nblks = racluster; /* * Now compute the number of contiguous blocks. */ if (nblks > 1) { error = VOP_BMAP(vp, lblkno, NULL, &blkno, &ncontig, NULL); /* * If this failed to map just do the original block. */ if (error || blkno == -1) ncontig = 0; } /* * If we have contiguous data available do a cluster * otherwise just read the requested block. */ if (ncontig) { /* Account for our first block. */ ncontig = min(ncontig + 1, nblks); if (ncontig < nblks) nblks = ncontig; bp = cluster_rbuild(vp, filesize, lblkno, blkno, size, nblks, gbflags, bp); lblkno += (bp->b_bufsize / size); } else { bp->b_flags |= B_RAM; bp->b_iocmd = BIO_READ; lblkno += 1; } } /* * handle the synchronous read so that it is available ASAP. */ if (bp) { if ((bp->b_flags & B_CLUSTER) == 0) { vfs_busy_pages(bp, 0); } bp->b_flags &= ~B_INVAL; bp->b_ioflags &= ~BIO_ERROR; if ((bp->b_flags & B_ASYNC) || bp->b_iodone != NULL) BUF_KERNPROC(bp); bp->b_iooffset = dbtob(bp->b_blkno); bstrategy(bp); #ifdef RACCT if (racct_enable) { PROC_LOCK(curproc); racct_add_buf(curproc, bp, 0); PROC_UNLOCK(curproc); } #endif /* RACCT */ curthread->td_ru.ru_inblock++; } /* * If we have been doing sequential I/O, then do some read-ahead. */ while (lblkno < (origblkno + maxra)) { error = VOP_BMAP(vp, lblkno, NULL, &blkno, &ncontig, NULL); if (error) break; if (blkno == -1) break; /* * We could throttle ncontig here by maxra but we might as * well read the data if it is contiguous. We're throttled * by racluster anyway. */ if (ncontig) { ncontig = min(ncontig + 1, racluster); rbp = cluster_rbuild(vp, filesize, lblkno, blkno, size, ncontig, gbflags, NULL); lblkno += (rbp->b_bufsize / size); if (rbp->b_flags & B_DELWRI) { bqrelse(rbp); continue; } } else { rbp = getblk(vp, lblkno, size, 0, 0, gbflags); lblkno += 1; if (rbp->b_flags & B_DELWRI) { bqrelse(rbp); continue; } rbp->b_flags |= B_ASYNC | B_RAM; rbp->b_iocmd = BIO_READ; rbp->b_blkno = blkno; } if (rbp->b_flags & B_CACHE) { rbp->b_flags &= ~B_ASYNC; bqrelse(rbp); continue; } if ((rbp->b_flags & B_CLUSTER) == 0) { vfs_busy_pages(rbp, 0); } rbp->b_flags &= ~B_INVAL; rbp->b_ioflags &= ~BIO_ERROR; if ((rbp->b_flags & B_ASYNC) || rbp->b_iodone != NULL) BUF_KERNPROC(rbp); rbp->b_iooffset = dbtob(rbp->b_blkno); bstrategy(rbp); #ifdef RACCT if (racct_enable) { PROC_LOCK(curproc); racct_add_buf(curproc, rbp, 0); PROC_UNLOCK(curproc); } #endif /* RACCT */ curthread->td_ru.ru_inblock++; } if (reqbp) { /* * Like bread, always brelse() the buffer when * returning an error. */ error = bufwait(reqbp); if (error != 0) { brelse(reqbp); *bpp = NULL; } } return (error); } /* * If blocks are contiguous on disk, use this to provide clustered * read ahead. We will read as many blocks as possible sequentially * and then parcel them up into logical blocks in the buffer hash table. */ static struct buf * cluster_rbuild(struct vnode *vp, u_quad_t filesize, daddr_t lbn, daddr_t blkno, long size, int run, int gbflags, struct buf *fbp) { struct buf *bp, *tbp; daddr_t bn; off_t off; long tinc, tsize; int i, inc, j, k, toff; KASSERT(size == vp->v_mount->mnt_stat.f_iosize, ("cluster_rbuild: size %ld != f_iosize %jd\n", size, (intmax_t)vp->v_mount->mnt_stat.f_iosize)); /* * avoid a division */ while ((u_quad_t) size * (lbn + run) > filesize) { --run; } if (fbp) { tbp = fbp; tbp->b_iocmd = BIO_READ; } else { tbp = getblk(vp, lbn, size, 0, 0, gbflags); if (tbp->b_flags & B_CACHE) return tbp; tbp->b_flags |= B_ASYNC | B_RAM; tbp->b_iocmd = BIO_READ; } tbp->b_blkno = blkno; if( (tbp->b_flags & B_MALLOC) || ((tbp->b_flags & B_VMIO) == 0) || (run <= 1) ) return tbp; bp = trypbuf(&cluster_pbuf_freecnt); if (bp == NULL) return tbp; /* * We are synthesizing a buffer out of vm_page_t's, but * if the block size is not page aligned then the starting * address may not be either. Inherit the b_data offset * from the original buffer. */ bp->b_flags = B_ASYNC | B_CLUSTER | B_VMIO; if ((gbflags & GB_UNMAPPED) != 0) { bp->b_data = unmapped_buf; } else { bp->b_data = (char *)((vm_offset_t)bp->b_data | ((vm_offset_t)tbp->b_data & PAGE_MASK)); } bp->b_iocmd = BIO_READ; bp->b_iodone = cluster_callback; bp->b_blkno = blkno; bp->b_lblkno = lbn; bp->b_offset = tbp->b_offset; KASSERT(bp->b_offset != NOOFFSET, ("cluster_rbuild: no buffer offset")); pbgetvp(vp, bp); TAILQ_INIT(&bp->b_cluster.cluster_head); bp->b_bcount = 0; bp->b_bufsize = 0; bp->b_npages = 0; inc = btodb(size); for (bn = blkno, i = 0; i < run; ++i, bn += inc) { if (i == 0) { VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object); vfs_drain_busy_pages(tbp); vm_object_pip_add(tbp->b_bufobj->bo_object, tbp->b_npages); for (k = 0; k < tbp->b_npages; k++) vm_page_sbusy(tbp->b_pages[k]); VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object); } else { if ((bp->b_npages * PAGE_SIZE) + round_page(size) > vp->v_mount->mnt_iosize_max) { break; } tbp = getblk(vp, lbn + i, size, 0, 0, GB_LOCK_NOWAIT | (gbflags & GB_UNMAPPED)); /* Don't wait around for locked bufs. */ if (tbp == NULL) break; /* * Stop scanning if the buffer is fully valid * (marked B_CACHE), or locked (may be doing a * background write), or if the buffer is not * VMIO backed. The clustering code can only deal * with VMIO-backed buffers. The bo lock is not * required for the BKGRDINPROG check since it * can not be set without the buf lock. */ if ((tbp->b_vflags & BV_BKGRDINPROG) || (tbp->b_flags & B_CACHE) || (tbp->b_flags & B_VMIO) == 0) { bqrelse(tbp); break; } /* * The buffer must be completely invalid in order to * take part in the cluster. If it is partially valid * then we stop. */ off = tbp->b_offset; tsize = size; VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object); for (j = 0; tsize > 0; j++) { toff = off & PAGE_MASK; tinc = tsize; if (toff + tinc > PAGE_SIZE) tinc = PAGE_SIZE - toff; VM_OBJECT_ASSERT_WLOCKED(tbp->b_pages[j]->object); if ((tbp->b_pages[j]->valid & vm_page_bits(toff, tinc)) != 0) break; if (vm_page_xbusied(tbp->b_pages[j])) break; vm_object_pip_add(tbp->b_bufobj->bo_object, 1); vm_page_sbusy(tbp->b_pages[j]); off += tinc; tsize -= tinc; } if (tsize > 0) { clean_sbusy: vm_object_pip_add(tbp->b_bufobj->bo_object, -j); for (k = 0; k < j; k++) vm_page_sunbusy(tbp->b_pages[k]); VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object); bqrelse(tbp); break; } VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object); /* * Set a read-ahead mark as appropriate */ if ((fbp && (i == 1)) || (i == (run - 1))) tbp->b_flags |= B_RAM; /* * Set the buffer up for an async read (XXX should * we do this only if we do not wind up brelse()ing?). * Set the block number if it isn't set, otherwise * if it is make sure it matches the block number we * expect. */ tbp->b_flags |= B_ASYNC; tbp->b_iocmd = BIO_READ; if (tbp->b_blkno == tbp->b_lblkno) { tbp->b_blkno = bn; } else if (tbp->b_blkno != bn) { VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object); goto clean_sbusy; } } /* * XXX fbp from caller may not be B_ASYNC, but we are going * to biodone() it in cluster_callback() anyway */ BUF_KERNPROC(tbp); TAILQ_INSERT_TAIL(&bp->b_cluster.cluster_head, tbp, b_cluster.cluster_entry); VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object); for (j = 0; j < tbp->b_npages; j += 1) { vm_page_t m; m = tbp->b_pages[j]; if ((bp->b_npages == 0) || (bp->b_pages[bp->b_npages-1] != m)) { bp->b_pages[bp->b_npages] = m; bp->b_npages++; } if (m->valid == VM_PAGE_BITS_ALL) tbp->b_pages[j] = bogus_page; } VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object); /* * Don't inherit tbp->b_bufsize as it may be larger due to * a non-page-aligned size. Instead just aggregate using * 'size'. */ if (tbp->b_bcount != size) printf("warning: tbp->b_bcount wrong %ld vs %ld\n", tbp->b_bcount, size); if (tbp->b_bufsize != size) printf("warning: tbp->b_bufsize wrong %ld vs %ld\n", tbp->b_bufsize, size); bp->b_bcount += size; bp->b_bufsize += size; } /* * Fully valid pages in the cluster are already good and do not need * to be re-read from disk. Replace the page with bogus_page */ VM_OBJECT_WLOCK(bp->b_bufobj->bo_object); for (j = 0; j < bp->b_npages; j++) { VM_OBJECT_ASSERT_WLOCKED(bp->b_pages[j]->object); if (bp->b_pages[j]->valid == VM_PAGE_BITS_ALL) bp->b_pages[j] = bogus_page; } VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object); if (bp->b_bufsize > bp->b_kvasize) panic("cluster_rbuild: b_bufsize(%ld) > b_kvasize(%d)\n", bp->b_bufsize, bp->b_kvasize); if (buf_mapped(bp)) { pmap_qenter(trunc_page((vm_offset_t) bp->b_data), (vm_page_t *)bp->b_pages, bp->b_npages); } return (bp); } /* * Cleanup after a clustered read or write. * This is complicated by the fact that any of the buffers might have * extra memory (if there were no empty buffer headers at allocbuf time) * that we will need to shift around. */ static void cluster_callback(bp) struct buf *bp; { struct buf *nbp, *tbp; int error = 0; /* * Must propagate errors to all the components. */ if (bp->b_ioflags & BIO_ERROR) error = bp->b_error; if (buf_mapped(bp)) { pmap_qremove(trunc_page((vm_offset_t) bp->b_data), bp->b_npages); } /* * Move memory from the large cluster buffer into the component * buffers and mark IO as done on these. */ for (tbp = TAILQ_FIRST(&bp->b_cluster.cluster_head); tbp; tbp = nbp) { nbp = TAILQ_NEXT(&tbp->b_cluster, cluster_entry); if (error) { tbp->b_ioflags |= BIO_ERROR; tbp->b_error = error; } else { tbp->b_dirtyoff = tbp->b_dirtyend = 0; tbp->b_flags &= ~B_INVAL; tbp->b_ioflags &= ~BIO_ERROR; /* * XXX the bdwrite()/bqrelse() issued during * cluster building clears B_RELBUF (see bqrelse() * comment). If direct I/O was specified, we have * to restore it here to allow the buffer and VM * to be freed. */ if (tbp->b_flags & B_DIRECT) tbp->b_flags |= B_RELBUF; } bufdone(tbp); } pbrelvp(bp); relpbuf(bp, &cluster_pbuf_freecnt); } /* * cluster_wbuild_wb: * * Implement modified write build for cluster. * * write_behind = 0 write behind disabled * write_behind = 1 write behind normal (default) * write_behind = 2 write behind backed-off */ static __inline int cluster_wbuild_wb(struct vnode *vp, long size, daddr_t start_lbn, int len, int gbflags) { int r = 0; switch (write_behind) { case 2: if (start_lbn < len) break; start_lbn -= len; /* FALLTHROUGH */ case 1: r = cluster_wbuild(vp, size, start_lbn, len, gbflags); /* FALLTHROUGH */ default: /* FALLTHROUGH */ break; } return(r); } /* * Do clustered write for FFS. * * Three cases: * 1. Write is not sequential (write asynchronously) * Write is sequential: * 2. beginning of cluster - begin cluster * 3. middle of a cluster - add to cluster * 4. end of a cluster - asynchronously write cluster */ void cluster_write(struct vnode *vp, struct buf *bp, u_quad_t filesize, int seqcount, int gbflags) { daddr_t lbn; int maxclen, cursize; int lblocksize; int async; if (!unmapped_buf_allowed) gbflags &= ~GB_UNMAPPED; if (vp->v_type == VREG) { async = DOINGASYNC(vp); lblocksize = vp->v_mount->mnt_stat.f_iosize; } else { async = 0; lblocksize = bp->b_bufsize; } lbn = bp->b_lblkno; KASSERT(bp->b_offset != NOOFFSET, ("cluster_write: no buffer offset")); /* Initialize vnode to beginning of file. */ if (lbn == 0) vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0; if (vp->v_clen == 0 || lbn != vp->v_lastw + 1 || (bp->b_blkno != vp->v_lasta + btodb(lblocksize))) { maxclen = vp->v_mount->mnt_iosize_max / lblocksize - 1; if (vp->v_clen != 0) { /* * Next block is not sequential. * * If we are not writing at end of file, the process * seeked to another point in the file since its last * write, or we have reached our maximum cluster size, * then push the previous cluster. Otherwise try * reallocating to make it sequential. * * Change to algorithm: only push previous cluster if * it was sequential from the point of view of the * seqcount heuristic, otherwise leave the buffer * intact so we can potentially optimize the I/O * later on in the buf_daemon or update daemon * flush. */ cursize = vp->v_lastw - vp->v_cstart + 1; if (((u_quad_t) bp->b_offset + lblocksize) != filesize || lbn != vp->v_lastw + 1 || vp->v_clen <= cursize) { if (!async && seqcount > 0) { cluster_wbuild_wb(vp, lblocksize, vp->v_cstart, cursize, gbflags); } } else { struct buf **bpp, **endbp; struct cluster_save *buflist; buflist = cluster_collectbufs(vp, bp, gbflags); endbp = &buflist->bs_children [buflist->bs_nchildren - 1]; if (VOP_REALLOCBLKS(vp, buflist)) { /* * Failed, push the previous cluster * if *really* writing sequentially * in the logical file (seqcount > 1), * otherwise delay it in the hopes that * the low level disk driver can * optimize the write ordering. */ for (bpp = buflist->bs_children; bpp < endbp; bpp++) brelse(*bpp); free(buflist, M_SEGMENT); if (seqcount > 1) { cluster_wbuild_wb(vp, lblocksize, vp->v_cstart, cursize, gbflags); } } else { /* * Succeeded, keep building cluster. */ for (bpp = buflist->bs_children; bpp <= endbp; bpp++) bdwrite(*bpp); free(buflist, M_SEGMENT); vp->v_lastw = lbn; vp->v_lasta = bp->b_blkno; return; } } } /* * Consider beginning a cluster. If at end of file, make * cluster as large as possible, otherwise find size of * existing cluster. */ if ((vp->v_type == VREG) && ((u_quad_t) bp->b_offset + lblocksize) != filesize && (bp->b_blkno == bp->b_lblkno) && (VOP_BMAP(vp, lbn, NULL, &bp->b_blkno, &maxclen, NULL) || bp->b_blkno == -1)) { bawrite(bp); vp->v_clen = 0; vp->v_lasta = bp->b_blkno; vp->v_cstart = lbn + 1; vp->v_lastw = lbn; return; } vp->v_clen = maxclen; if (!async && maxclen == 0) { /* I/O not contiguous */ vp->v_cstart = lbn + 1; bawrite(bp); } else { /* Wait for rest of cluster */ vp->v_cstart = lbn; bdwrite(bp); } } else if (lbn == vp->v_cstart + vp->v_clen) { /* * At end of cluster, write it out if seqcount tells us we * are operating sequentially, otherwise let the buf or * update daemon handle it. */ bdwrite(bp); if (seqcount > 1) { cluster_wbuild_wb(vp, lblocksize, vp->v_cstart, vp->v_clen + 1, gbflags); } vp->v_clen = 0; vp->v_cstart = lbn + 1; } else if (vm_page_count_severe()) { /* * We are low on memory, get it going NOW */ bawrite(bp); } else { /* * In the middle of a cluster, so just delay the I/O for now. */ bdwrite(bp); } vp->v_lastw = lbn; vp->v_lasta = bp->b_blkno; } /* * This is an awful lot like cluster_rbuild...wish they could be combined. * The last lbn argument is the current block on which I/O is being * performed. Check to see that it doesn't fall in the middle of * the current block (if last_bp == NULL). */ int cluster_wbuild(struct vnode *vp, long size, daddr_t start_lbn, int len, int gbflags) { struct buf *bp, *tbp; struct bufobj *bo; int i, j; int totalwritten = 0; int dbsize = btodb(size); if (!unmapped_buf_allowed) gbflags &= ~GB_UNMAPPED; bo = &vp->v_bufobj; while (len > 0) { /* * If the buffer is not delayed-write (i.e. dirty), or it * is delayed-write but either locked or inval, it cannot * partake in the clustered write. */ BO_LOCK(bo); if ((tbp = gbincore(&vp->v_bufobj, start_lbn)) == NULL || (tbp->b_vflags & BV_BKGRDINPROG)) { BO_UNLOCK(bo); ++start_lbn; --len; continue; } if (BUF_LOCK(tbp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, BO_LOCKPTR(bo))) { ++start_lbn; --len; continue; } if ((tbp->b_flags & (B_INVAL | B_DELWRI)) != B_DELWRI) { BUF_UNLOCK(tbp); ++start_lbn; --len; continue; } bremfree(tbp); tbp->b_flags &= ~B_DONE; /* * Extra memory in the buffer, punt on this buffer. * XXX we could handle this in most cases, but we would * have to push the extra memory down to after our max * possible cluster size and then potentially pull it back * up if the cluster was terminated prematurely--too much * hassle. */ if (((tbp->b_flags & (B_CLUSTEROK | B_MALLOC | B_VMIO)) != (B_CLUSTEROK | B_VMIO)) || (tbp->b_bcount != tbp->b_bufsize) || (tbp->b_bcount != size) || (len == 1) || ((bp = (vp->v_vflag & VV_MD) != 0 ? trypbuf(&cluster_pbuf_freecnt) : getpbuf(&cluster_pbuf_freecnt)) == NULL)) { totalwritten += tbp->b_bufsize; bawrite(tbp); ++start_lbn; --len; continue; } /* * We got a pbuf to make the cluster in. * so initialise it. */ TAILQ_INIT(&bp->b_cluster.cluster_head); bp->b_bcount = 0; bp->b_bufsize = 0; bp->b_npages = 0; if (tbp->b_wcred != NOCRED) bp->b_wcred = crhold(tbp->b_wcred); bp->b_blkno = tbp->b_blkno; bp->b_lblkno = tbp->b_lblkno; bp->b_offset = tbp->b_offset; /* * We are synthesizing a buffer out of vm_page_t's, but * if the block size is not page aligned then the starting * address may not be either. Inherit the b_data offset * from the original buffer. */ if ((gbflags & GB_UNMAPPED) == 0 || (tbp->b_flags & B_VMIO) == 0) { bp->b_data = (char *)((vm_offset_t)bp->b_data | ((vm_offset_t)tbp->b_data & PAGE_MASK)); } else { bp->b_data = unmapped_buf; } bp->b_flags |= B_CLUSTER | (tbp->b_flags & (B_VMIO | B_NEEDCOMMIT)); bp->b_iodone = cluster_callback; pbgetvp(vp, bp); /* * From this location in the file, scan forward to see * if there are buffers with adjacent data that need to * be written as well. */ for (i = 0; i < len; ++i, ++start_lbn) { if (i != 0) { /* If not the first buffer */ /* * If the adjacent data is not even in core it * can't need to be written. */ BO_LOCK(bo); if ((tbp = gbincore(bo, start_lbn)) == NULL || (tbp->b_vflags & BV_BKGRDINPROG)) { BO_UNLOCK(bo); break; } /* * If it IS in core, but has different * characteristics, or is locked (which * means it could be undergoing a background * I/O or be in a weird state), then don't * cluster with it. */ if (BUF_LOCK(tbp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, BO_LOCKPTR(bo))) break; if ((tbp->b_flags & (B_VMIO | B_CLUSTEROK | B_INVAL | B_DELWRI | B_NEEDCOMMIT)) != (B_DELWRI | B_CLUSTEROK | (bp->b_flags & (B_VMIO | B_NEEDCOMMIT))) || tbp->b_wcred != bp->b_wcred) { BUF_UNLOCK(tbp); break; } /* * Check that the combined cluster * would make sense with regard to pages * and would not be too large */ if ((tbp->b_bcount != size) || ((bp->b_blkno + (dbsize * i)) != tbp->b_blkno) || ((tbp->b_npages + bp->b_npages) > (vp->v_mount->mnt_iosize_max / PAGE_SIZE))) { BUF_UNLOCK(tbp); break; } /* * Ok, it's passed all the tests, * so remove it from the free list * and mark it busy. We will use it. */ bremfree(tbp); tbp->b_flags &= ~B_DONE; } /* end of code for non-first buffers only */ /* * If the IO is via the VM then we do some * special VM hackery (yuck). Since the buffer's * block size may not be page-aligned it is possible * for a page to be shared between two buffers. We * have to get rid of the duplication when building * the cluster. */ if (tbp->b_flags & B_VMIO) { vm_page_t m; VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object); if (i == 0) { vfs_drain_busy_pages(tbp); } else { /* if not first buffer */ for (j = 0; j < tbp->b_npages; j += 1) { m = tbp->b_pages[j]; if (vm_page_xbusied(m)) { VM_OBJECT_WUNLOCK( tbp->b_object); bqrelse(tbp); goto finishcluster; } } } for (j = 0; j < tbp->b_npages; j += 1) { m = tbp->b_pages[j]; vm_page_sbusy(m); vm_object_pip_add(m->object, 1); if ((bp->b_npages == 0) || (bp->b_pages[bp->b_npages - 1] != m)) { bp->b_pages[bp->b_npages] = m; bp->b_npages++; } } VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object); } bp->b_bcount += size; bp->b_bufsize += size; /* * If any of the clustered buffers have their * B_BARRIER flag set, transfer that request to * the cluster. */ bp->b_flags |= (tbp->b_flags & B_BARRIER); tbp->b_flags &= ~(B_DONE | B_BARRIER); tbp->b_flags |= B_ASYNC; tbp->b_ioflags &= ~BIO_ERROR; tbp->b_iocmd = BIO_WRITE; bundirty(tbp); reassignbuf(tbp); /* put on clean list */ bufobj_wref(tbp->b_bufobj); BUF_KERNPROC(tbp); TAILQ_INSERT_TAIL(&bp->b_cluster.cluster_head, tbp, b_cluster.cluster_entry); } finishcluster: if (buf_mapped(bp)) { pmap_qenter(trunc_page((vm_offset_t) bp->b_data), (vm_page_t *)bp->b_pages, bp->b_npages); } if (bp->b_bufsize > bp->b_kvasize) panic( "cluster_wbuild: b_bufsize(%ld) > b_kvasize(%d)\n", bp->b_bufsize, bp->b_kvasize); totalwritten += bp->b_bufsize; bp->b_dirtyoff = 0; bp->b_dirtyend = bp->b_bufsize; bawrite(bp); len -= i; } return totalwritten; } /* * Collect together all the buffers in a cluster. * Plus add one additional buffer. */ static struct cluster_save * cluster_collectbufs(struct vnode *vp, struct buf *last_bp, int gbflags) { struct cluster_save *buflist; struct buf *bp; daddr_t lbn; int i, len; len = vp->v_lastw - vp->v_cstart + 1; buflist = malloc(sizeof(struct buf *) * (len + 1) + sizeof(*buflist), M_SEGMENT, M_WAITOK); buflist->bs_nchildren = 0; buflist->bs_children = (struct buf **) (buflist + 1); for (lbn = vp->v_cstart, i = 0; i < len; lbn++, i++) { (void)bread_gb(vp, lbn, last_bp->b_bcount, NOCRED, gbflags, &bp); buflist->bs_children[i] = bp; if (bp->b_blkno == bp->b_lblkno) VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, NULL, NULL); } buflist->bs_children[i] = bp = last_bp; if (bp->b_blkno == bp->b_lblkno) VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, NULL, NULL); buflist->bs_nchildren = i + 1; return (buflist); } Index: head/sys/kern/vfs_default.c =================================================================== --- head/sys/kern/vfs_default.c (revision 305831) +++ head/sys/kern/vfs_default.c (revision 305832) @@ -1,1323 +1,1323 @@ /*- * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed * to Berkeley by John Heidemann of the UCLA Ficus project. * * Source: * @(#)i405_init.c 2.10 92/04/27 UCLA Ficus project * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static int vop_nolookup(struct vop_lookup_args *); static int vop_norename(struct vop_rename_args *); static int vop_nostrategy(struct vop_strategy_args *); static int get_next_dirent(struct vnode *vp, struct dirent **dpp, char *dirbuf, int dirbuflen, off_t *off, char **cpos, int *len, int *eofflag, struct thread *td); static int dirent_exists(struct vnode *vp, const char *dirname, struct thread *td); #define DIRENT_MINSIZE (sizeof(struct dirent) - (MAXNAMLEN+1) + 4) static int vop_stdis_text(struct vop_is_text_args *ap); static int vop_stdset_text(struct vop_set_text_args *ap); static int vop_stdunset_text(struct vop_unset_text_args *ap); static int vop_stdget_writecount(struct vop_get_writecount_args *ap); static int vop_stdadd_writecount(struct vop_add_writecount_args *ap); static int vop_stdfdatasync(struct vop_fdatasync_args *ap); static int vop_stdgetpages_async(struct vop_getpages_async_args *ap); /* * This vnode table stores what we want to do if the filesystem doesn't * implement a particular VOP. * * If there is no specific entry here, we will return EOPNOTSUPP. * * Note that every filesystem has to implement either vop_access * or vop_accessx; failing to do so will result in immediate crash * due to stack overflow, as vop_stdaccess() calls vop_stdaccessx(), * which calls vop_stdaccess() etc. */ struct vop_vector default_vnodeops = { .vop_default = NULL, .vop_bypass = VOP_EOPNOTSUPP, .vop_access = vop_stdaccess, .vop_accessx = vop_stdaccessx, .vop_advise = vop_stdadvise, .vop_advlock = vop_stdadvlock, .vop_advlockasync = vop_stdadvlockasync, .vop_advlockpurge = vop_stdadvlockpurge, .vop_allocate = vop_stdallocate, .vop_bmap = vop_stdbmap, .vop_close = VOP_NULL, .vop_fsync = VOP_NULL, .vop_fdatasync = vop_stdfdatasync, .vop_getpages = vop_stdgetpages, .vop_getpages_async = vop_stdgetpages_async, .vop_getwritemount = vop_stdgetwritemount, .vop_inactive = VOP_NULL, .vop_ioctl = VOP_ENOTTY, .vop_kqfilter = vop_stdkqfilter, .vop_islocked = vop_stdislocked, .vop_lock1 = vop_stdlock, .vop_lookup = vop_nolookup, .vop_open = VOP_NULL, .vop_pathconf = VOP_EINVAL, .vop_poll = vop_nopoll, .vop_putpages = vop_stdputpages, .vop_readlink = VOP_EINVAL, .vop_rename = vop_norename, .vop_revoke = VOP_PANIC, .vop_strategy = vop_nostrategy, .vop_unlock = vop_stdunlock, .vop_vptocnp = vop_stdvptocnp, .vop_vptofh = vop_stdvptofh, .vop_unp_bind = vop_stdunp_bind, .vop_unp_connect = vop_stdunp_connect, .vop_unp_detach = vop_stdunp_detach, .vop_is_text = vop_stdis_text, .vop_set_text = vop_stdset_text, .vop_unset_text = vop_stdunset_text, .vop_get_writecount = vop_stdget_writecount, .vop_add_writecount = vop_stdadd_writecount, }; /* * Series of placeholder functions for various error returns for * VOPs. */ int vop_eopnotsupp(struct vop_generic_args *ap) { /* printf("vop_notsupp[%s]\n", ap->a_desc->vdesc_name); */ return (EOPNOTSUPP); } int vop_ebadf(struct vop_generic_args *ap) { return (EBADF); } int vop_enotty(struct vop_generic_args *ap) { return (ENOTTY); } int vop_einval(struct vop_generic_args *ap) { return (EINVAL); } int vop_enoent(struct vop_generic_args *ap) { return (ENOENT); } int vop_null(struct vop_generic_args *ap) { return (0); } /* * Helper function to panic on some bad VOPs in some filesystems. */ int vop_panic(struct vop_generic_args *ap) { panic("filesystem goof: vop_panic[%s]", ap->a_desc->vdesc_name); } /* * vop_std and vop_no are default functions for use by * filesystems that need the "default reasonable" implementation for a * particular operation. * * The documentation for the operations they implement exists (if it exists) * in the VOP_(9) manpage (all uppercase). */ /* * Default vop for filesystems that do not support name lookup */ static int vop_nolookup(ap) struct vop_lookup_args /* { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; } */ *ap; { *ap->a_vpp = NULL; return (ENOTDIR); } /* * vop_norename: * * Handle unlock and reference counting for arguments of vop_rename * for filesystems that do not implement rename operation. */ static int vop_norename(struct vop_rename_args *ap) { vop_rename_fail(ap); return (EOPNOTSUPP); } /* * vop_nostrategy: * * Strategy routine for VFS devices that have none. * * BIO_ERROR and B_INVAL must be cleared prior to calling any strategy * routine. Typically this is done for a BIO_READ strategy call. * Typically B_INVAL is assumed to already be clear prior to a write * and should not be cleared manually unless you just made the buffer * invalid. BIO_ERROR should be cleared either way. */ static int vop_nostrategy (struct vop_strategy_args *ap) { printf("No strategy for buffer at %p\n", ap->a_bp); vn_printf(ap->a_vp, "vnode "); ap->a_bp->b_ioflags |= BIO_ERROR; ap->a_bp->b_error = EOPNOTSUPP; bufdone(ap->a_bp); return (EOPNOTSUPP); } static int get_next_dirent(struct vnode *vp, struct dirent **dpp, char *dirbuf, int dirbuflen, off_t *off, char **cpos, int *len, int *eofflag, struct thread *td) { int error, reclen; struct uio uio; struct iovec iov; struct dirent *dp; KASSERT(VOP_ISLOCKED(vp), ("vp %p is not locked", vp)); KASSERT(vp->v_type == VDIR, ("vp %p is not a directory", vp)); if (*len == 0) { iov.iov_base = dirbuf; iov.iov_len = dirbuflen; uio.uio_iov = &iov; uio.uio_iovcnt = 1; uio.uio_offset = *off; uio.uio_resid = dirbuflen; uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_READ; uio.uio_td = td; *eofflag = 0; #ifdef MAC error = mac_vnode_check_readdir(td->td_ucred, vp); if (error == 0) #endif error = VOP_READDIR(vp, &uio, td->td_ucred, eofflag, NULL, NULL); if (error) return (error); *off = uio.uio_offset; *cpos = dirbuf; *len = (dirbuflen - uio.uio_resid); if (*len == 0) return (ENOENT); } dp = (struct dirent *)(*cpos); reclen = dp->d_reclen; *dpp = dp; /* check for malformed directory.. */ if (reclen < DIRENT_MINSIZE) return (EINVAL); *cpos += reclen; *len -= reclen; return (0); } /* * Check if a named file exists in a given directory vnode. */ static int dirent_exists(struct vnode *vp, const char *dirname, struct thread *td) { char *dirbuf, *cpos; int error, eofflag, dirbuflen, len, found; off_t off; struct dirent *dp; struct vattr va; KASSERT(VOP_ISLOCKED(vp), ("vp %p is not locked", vp)); KASSERT(vp->v_type == VDIR, ("vp %p is not a directory", vp)); found = 0; error = VOP_GETATTR(vp, &va, td->td_ucred); if (error) return (found); dirbuflen = DEV_BSIZE; if (dirbuflen < va.va_blocksize) dirbuflen = va.va_blocksize; dirbuf = (char *)malloc(dirbuflen, M_TEMP, M_WAITOK); off = 0; len = 0; do { error = get_next_dirent(vp, &dp, dirbuf, dirbuflen, &off, &cpos, &len, &eofflag, td); if (error) goto out; if (dp->d_type != DT_WHT && dp->d_fileno != 0 && strcmp(dp->d_name, dirname) == 0) { found = 1; goto out; } } while (len > 0 || !eofflag); out: free(dirbuf, M_TEMP); return (found); } int vop_stdaccess(struct vop_access_args *ap) { KASSERT((ap->a_accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0, ("invalid bit in accmode")); return (VOP_ACCESSX(ap->a_vp, ap->a_accmode, ap->a_cred, ap->a_td)); } int vop_stdaccessx(struct vop_accessx_args *ap) { int error; accmode_t accmode = ap->a_accmode; error = vfs_unixify_accmode(&accmode); if (error != 0) return (error); if (accmode == 0) return (0); return (VOP_ACCESS(ap->a_vp, accmode, ap->a_cred, ap->a_td)); } /* * Advisory record locking support */ int vop_stdadvlock(struct vop_advlock_args *ap) { struct vnode *vp; struct vattr vattr; int error; vp = ap->a_vp; if (ap->a_fl->l_whence == SEEK_END) { /* * The NFSv4 server must avoid doing a vn_lock() here, since it * can deadlock the nfsd threads, due to a LOR. Fortunately * the NFSv4 server always uses SEEK_SET and this code is * only required for the SEEK_END case. */ vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_GETATTR(vp, &vattr, curthread->td_ucred); VOP_UNLOCK(vp, 0); if (error) return (error); } else vattr.va_size = 0; return (lf_advlock(ap, &(vp->v_lockf), vattr.va_size)); } int vop_stdadvlockasync(struct vop_advlockasync_args *ap) { struct vnode *vp; struct vattr vattr; int error; vp = ap->a_vp; if (ap->a_fl->l_whence == SEEK_END) { /* The size argument is only needed for SEEK_END. */ vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_GETATTR(vp, &vattr, curthread->td_ucred); VOP_UNLOCK(vp, 0); if (error) return (error); } else vattr.va_size = 0; return (lf_advlockasync(ap, &(vp->v_lockf), vattr.va_size)); } int vop_stdadvlockpurge(struct vop_advlockpurge_args *ap) { struct vnode *vp; vp = ap->a_vp; lf_purgelocks(vp, &vp->v_lockf); return (0); } /* * vop_stdpathconf: * * Standard implementation of POSIX pathconf, to get information about limits * for a filesystem. * Override per filesystem for the case where the filesystem has smaller * limits. */ int vop_stdpathconf(ap) struct vop_pathconf_args /* { struct vnode *a_vp; int a_name; int *a_retval; } */ *ap; { switch (ap->a_name) { case _PC_ASYNC_IO: *ap->a_retval = _POSIX_ASYNCHRONOUS_IO; return (0); case _PC_NAME_MAX: *ap->a_retval = NAME_MAX; return (0); case _PC_PATH_MAX: *ap->a_retval = PATH_MAX; return (0); case _PC_LINK_MAX: *ap->a_retval = LINK_MAX; return (0); case _PC_MAX_CANON: *ap->a_retval = MAX_CANON; return (0); case _PC_MAX_INPUT: *ap->a_retval = MAX_INPUT; return (0); case _PC_PIPE_BUF: *ap->a_retval = PIPE_BUF; return (0); case _PC_CHOWN_RESTRICTED: *ap->a_retval = 1; return (0); case _PC_VDISABLE: *ap->a_retval = _POSIX_VDISABLE; return (0); default: return (EINVAL); } /* NOTREACHED */ } /* * Standard lock, unlock and islocked functions. */ int vop_stdlock(ap) struct vop_lock1_args /* { struct vnode *a_vp; int a_flags; char *file; int line; } */ *ap; { struct vnode *vp = ap->a_vp; return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp), LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file, ap->a_line)); } /* See above. */ int vop_stdunlock(ap) struct vop_unlock_args /* { struct vnode *a_vp; int a_flags; } */ *ap; { struct vnode *vp = ap->a_vp; return (lockmgr(vp->v_vnlock, ap->a_flags | LK_RELEASE, VI_MTX(vp))); } /* See above. */ int vop_stdislocked(ap) struct vop_islocked_args /* { struct vnode *a_vp; } */ *ap; { return (lockstatus(ap->a_vp->v_vnlock)); } /* * Return true for select/poll. */ int vop_nopoll(ap) struct vop_poll_args /* { struct vnode *a_vp; int a_events; struct ucred *a_cred; struct thread *a_td; } */ *ap; { return (poll_no_poll(ap->a_events)); } /* * Implement poll for local filesystems that support it. */ int vop_stdpoll(ap) struct vop_poll_args /* { struct vnode *a_vp; int a_events; struct ucred *a_cred; struct thread *a_td; } */ *ap; { if (ap->a_events & ~POLLSTANDARD) return (vn_pollrecord(ap->a_vp, ap->a_td, ap->a_events)); return (ap->a_events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM)); } /* * Return our mount point, as we will take charge of the writes. */ int vop_stdgetwritemount(ap) struct vop_getwritemount_args /* { struct vnode *a_vp; struct mount **a_mpp; } */ *ap; { struct mount *mp; /* * XXX Since this is called unlocked we may be recycled while * attempting to ref the mount. If this is the case or mountpoint * will be set to NULL. We only have to prevent this call from * returning with a ref to an incorrect mountpoint. It is not * harmful to return with a ref to our previous mountpoint. */ mp = ap->a_vp->v_mount; if (mp != NULL) { vfs_ref(mp); if (mp != ap->a_vp->v_mount) { vfs_rel(mp); mp = NULL; } } *(ap->a_mpp) = mp; return (0); } /* XXX Needs good comment and VOP_BMAP(9) manpage */ int vop_stdbmap(ap) struct vop_bmap_args /* { struct vnode *a_vp; daddr_t a_bn; struct bufobj **a_bop; daddr_t *a_bnp; int *a_runp; int *a_runb; } */ *ap; { if (ap->a_bop != NULL) *ap->a_bop = &ap->a_vp->v_bufobj; if (ap->a_bnp != NULL) *ap->a_bnp = ap->a_bn * btodb(ap->a_vp->v_mount->mnt_stat.f_iosize); if (ap->a_runp != NULL) *ap->a_runp = 0; if (ap->a_runb != NULL) *ap->a_runb = 0; return (0); } int vop_stdfsync(ap) struct vop_fsync_args /* { struct vnode *a_vp; int a_waitfor; struct thread *a_td; } */ *ap; { struct vnode *vp = ap->a_vp; struct buf *bp; struct bufobj *bo; struct buf *nbp; int error = 0; int maxretry = 1000; /* large, arbitrarily chosen */ bo = &vp->v_bufobj; BO_LOCK(bo); loop1: /* * MARK/SCAN initialization to avoid infinite loops. */ TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) { bp->b_vflags &= ~BV_SCANNED; bp->b_error = 0; } /* * Flush all dirty buffers associated with a vnode. */ loop2: TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if ((bp->b_vflags & BV_SCANNED) != 0) continue; bp->b_vflags |= BV_SCANNED; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { if (ap->a_waitfor != MNT_WAIT) continue; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_INTERLOCK | LK_SLEEPFAIL, BO_LOCKPTR(bo)) != 0) { BO_LOCK(bo); goto loop1; } BO_LOCK(bo); } BO_UNLOCK(bo); KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); if ((bp->b_flags & B_DELWRI) == 0) panic("fsync: not dirty"); if ((vp->v_object != NULL) && (bp->b_flags & B_CLUSTEROK)) { vfs_bio_awrite(bp); } else { bremfree(bp); bawrite(bp); } BO_LOCK(bo); goto loop2; } /* * If synchronous the caller expects us to completely resolve all * dirty buffers in the system. Wait for in-progress I/O to * complete (which could include background bitmap writes), then * retry if dirty blocks still exist. */ if (ap->a_waitfor == MNT_WAIT) { bufobj_wwait(bo, 0, 0); if (bo->bo_dirty.bv_cnt > 0) { /* * If we are unable to write any of these buffers * then we fail now rather than trying endlessly * to write them out. */ TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) if ((error = bp->b_error) == 0) continue; if (error == 0 && --maxretry >= 0) goto loop1; error = EAGAIN; } } BO_UNLOCK(bo); if (error == EAGAIN) vn_printf(vp, "fsync: giving up on dirty "); return (error); } static int vop_stdfdatasync(struct vop_fdatasync_args *ap) { return (VOP_FSYNC(ap->a_vp, MNT_WAIT, ap->a_td)); } int vop_stdfdatasync_buf(struct vop_fdatasync_args *ap) { struct vop_fsync_args apf; apf.a_vp = ap->a_vp; apf.a_waitfor = MNT_WAIT; apf.a_td = ap->a_td; return (vop_stdfsync(&apf)); } /* XXX Needs good comment and more info in the manpage (VOP_GETPAGES(9)). */ int vop_stdgetpages(ap) struct vop_getpages_args /* { struct vnode *a_vp; vm_page_t *a_m; int a_count; int *a_rbehind; int *a_rahead; } */ *ap; { return vnode_pager_generic_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_rbehind, ap->a_rahead, NULL, NULL); } static int vop_stdgetpages_async(struct vop_getpages_async_args *ap) { int error; error = VOP_GETPAGES(ap->a_vp, ap->a_m, ap->a_count, ap->a_rbehind, ap->a_rahead); ap->a_iodone(ap->a_arg, ap->a_m, ap->a_count, error); return (error); } int vop_stdkqfilter(struct vop_kqfilter_args *ap) { return vfs_kqfilter(ap); } /* XXX Needs good comment and more info in the manpage (VOP_PUTPAGES(9)). */ int vop_stdputpages(ap) struct vop_putpages_args /* { struct vnode *a_vp; vm_page_t *a_m; int a_count; int a_sync; int *a_rtvals; } */ *ap; { return vnode_pager_generic_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync, ap->a_rtvals); } int vop_stdvptofh(struct vop_vptofh_args *ap) { return (EOPNOTSUPP); } int vop_stdvptocnp(struct vop_vptocnp_args *ap) { struct vnode *vp = ap->a_vp; struct vnode **dvp = ap->a_vpp; struct ucred *cred = ap->a_cred; char *buf = ap->a_buf; int *buflen = ap->a_buflen; char *dirbuf, *cpos; int i, error, eofflag, dirbuflen, flags, locked, len, covered; off_t off; ino_t fileno; struct vattr va; struct nameidata nd; struct thread *td; struct dirent *dp; struct vnode *mvp; i = *buflen; error = 0; covered = 0; td = curthread; if (vp->v_type != VDIR) return (ENOENT); error = VOP_GETATTR(vp, &va, cred); if (error) return (error); VREF(vp); locked = VOP_ISLOCKED(vp); VOP_UNLOCK(vp, 0); NDINIT_ATVP(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, "..", vp, td); flags = FREAD; error = vn_open_cred(&nd, &flags, 0, VN_OPEN_NOAUDIT, cred, NULL); if (error) { vn_lock(vp, locked | LK_RETRY); return (error); } NDFREE(&nd, NDF_ONLY_PNBUF); mvp = *dvp = nd.ni_vp; if (vp->v_mount != (*dvp)->v_mount && ((*dvp)->v_vflag & VV_ROOT) && ((*dvp)->v_mount->mnt_flag & MNT_UNION)) { *dvp = (*dvp)->v_mount->mnt_vnodecovered; VREF(mvp); VOP_UNLOCK(mvp, 0); vn_close(mvp, FREAD, cred, td); VREF(*dvp); vn_lock(*dvp, LK_SHARED | LK_RETRY); covered = 1; } fileno = va.va_fileid; dirbuflen = DEV_BSIZE; if (dirbuflen < va.va_blocksize) dirbuflen = va.va_blocksize; dirbuf = (char *)malloc(dirbuflen, M_TEMP, M_WAITOK); if ((*dvp)->v_type != VDIR) { error = ENOENT; goto out; } off = 0; len = 0; do { /* call VOP_READDIR of parent */ error = get_next_dirent(*dvp, &dp, dirbuf, dirbuflen, &off, &cpos, &len, &eofflag, td); if (error) goto out; if ((dp->d_type != DT_WHT) && (dp->d_fileno == fileno)) { if (covered) { VOP_UNLOCK(*dvp, 0); vn_lock(mvp, LK_SHARED | LK_RETRY); if (dirent_exists(mvp, dp->d_name, td)) { error = ENOENT; VOP_UNLOCK(mvp, 0); vn_lock(*dvp, LK_SHARED | LK_RETRY); goto out; } VOP_UNLOCK(mvp, 0); vn_lock(*dvp, LK_SHARED | LK_RETRY); } i -= dp->d_namlen; if (i < 0) { error = ENOMEM; goto out; } if (dp->d_namlen == 1 && dp->d_name[0] == '.') { error = ENOENT; } else { bcopy(dp->d_name, buf + i, dp->d_namlen); error = 0; } goto out; } } while (len > 0 || !eofflag); error = ENOENT; out: free(dirbuf, M_TEMP); if (!error) { *buflen = i; vref(*dvp); } if (covered) { vput(*dvp); vrele(mvp); } else { VOP_UNLOCK(mvp, 0); vn_close(mvp, FREAD, cred, td); } vn_lock(vp, locked | LK_RETRY); return (error); } int vop_stdallocate(struct vop_allocate_args *ap) { #ifdef __notyet__ struct statfs sfs; #endif struct iovec aiov; struct vattr vattr, *vap; struct uio auio; off_t fsize, len, cur, offset; uint8_t *buf; struct thread *td; struct vnode *vp; size_t iosize; int error; buf = NULL; error = 0; td = curthread; vap = &vattr; vp = ap->a_vp; len = *ap->a_len; offset = *ap->a_offset; error = VOP_GETATTR(vp, vap, td->td_ucred); if (error != 0) goto out; fsize = vap->va_size; iosize = vap->va_blocksize; if (iosize == 0) iosize = BLKDEV_IOSIZE; if (iosize > MAXPHYS) iosize = MAXPHYS; buf = malloc(iosize, M_TEMP, M_WAITOK); #ifdef __notyet__ /* * Check if the filesystem sets f_maxfilesize; if not use * VOP_SETATTR to perform the check. */ error = VFS_STATFS(vp->v_mount, &sfs, td); if (error != 0) goto out; if (sfs.f_maxfilesize) { if (offset > sfs.f_maxfilesize || len > sfs.f_maxfilesize || offset + len > sfs.f_maxfilesize) { error = EFBIG; goto out; } } else #endif if (offset + len > vap->va_size) { /* * Test offset + len against the filesystem's maxfilesize. */ VATTR_NULL(vap); vap->va_size = offset + len; error = VOP_SETATTR(vp, vap, td->td_ucred); if (error != 0) goto out; VATTR_NULL(vap); vap->va_size = fsize; error = VOP_SETATTR(vp, vap, td->td_ucred); if (error != 0) goto out; } for (;;) { /* * Read and write back anything below the nominal file * size. There's currently no way outside the filesystem * to know whether this area is sparse or not. */ cur = iosize; if ((offset % iosize) != 0) cur -= (offset % iosize); if (cur > len) cur = len; if (offset < fsize) { aiov.iov_base = buf; aiov.iov_len = cur; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_offset = offset; auio.uio_resid = cur; auio.uio_segflg = UIO_SYSSPACE; auio.uio_rw = UIO_READ; auio.uio_td = td; error = VOP_READ(vp, &auio, 0, td->td_ucred); if (error != 0) break; if (auio.uio_resid > 0) { bzero(buf + cur - auio.uio_resid, auio.uio_resid); } } else { bzero(buf, cur); } aiov.iov_base = buf; aiov.iov_len = cur; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_offset = offset; auio.uio_resid = cur; auio.uio_segflg = UIO_SYSSPACE; auio.uio_rw = UIO_WRITE; auio.uio_td = td; error = VOP_WRITE(vp, &auio, 0, td->td_ucred); if (error != 0) break; len -= cur; offset += cur; if (len == 0) break; if (should_yield()) break; } out: *ap->a_len = len; *ap->a_offset = offset; free(buf, M_TEMP); return (error); } int vop_stdadvise(struct vop_advise_args *ap) { struct vnode *vp; struct bufobj *bo; daddr_t startn, endn; off_t start, end; int bsize, error; vp = ap->a_vp; switch (ap->a_advice) { case POSIX_FADV_WILLNEED: /* * Do nothing for now. Filesystems should provide a * custom method which starts an asynchronous read of * the requested region. */ error = 0; break; case POSIX_FADV_DONTNEED: error = 0; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); if (vp->v_iflag & VI_DOOMED) { VOP_UNLOCK(vp, 0); break; } /* * Deactivate pages in the specified range from the backing VM * object. Pages that are resident in the buffer cache will * remain wired until their corresponding buffers are released * below. */ if (vp->v_object != NULL) { start = trunc_page(ap->a_start); end = round_page(ap->a_end); VM_OBJECT_WLOCK(vp->v_object); vm_object_page_noreuse(vp->v_object, OFF_TO_IDX(start), OFF_TO_IDX(end)); VM_OBJECT_WUNLOCK(vp->v_object); } bo = &vp->v_bufobj; BO_RLOCK(bo); bsize = vp->v_bufobj.bo_bsize; startn = ap->a_start / bsize; endn = ap->a_end / bsize; error = bnoreuselist(&bo->bo_clean, bo, startn, endn); if (error == 0) error = bnoreuselist(&bo->bo_dirty, bo, startn, endn); BO_RUNLOCK(bo); VOP_UNLOCK(vp, 0); break; default: error = EINVAL; break; } return (error); } int vop_stdunp_bind(struct vop_unp_bind_args *ap) { ap->a_vp->v_socket = ap->a_socket; return (0); } int vop_stdunp_connect(struct vop_unp_connect_args *ap) { *ap->a_socket = ap->a_vp->v_socket; return (0); } int vop_stdunp_detach(struct vop_unp_detach_args *ap) { ap->a_vp->v_socket = NULL; return (0); } static int vop_stdis_text(struct vop_is_text_args *ap) { return ((ap->a_vp->v_vflag & VV_TEXT) != 0); } static int vop_stdset_text(struct vop_set_text_args *ap) { ap->a_vp->v_vflag |= VV_TEXT; return (0); } static int vop_stdunset_text(struct vop_unset_text_args *ap) { ap->a_vp->v_vflag &= ~VV_TEXT; return (0); } static int vop_stdget_writecount(struct vop_get_writecount_args *ap) { *ap->a_writecount = ap->a_vp->v_writecount; return (0); } static int vop_stdadd_writecount(struct vop_add_writecount_args *ap) { ap->a_vp->v_writecount += ap->a_inc; return (0); } /* * vfs default ops * used to fill the vfs function table to get reasonable default return values. */ int vfs_stdroot (mp, flags, vpp) struct mount *mp; int flags; struct vnode **vpp; { return (EOPNOTSUPP); } int vfs_stdstatfs (mp, sbp) struct mount *mp; struct statfs *sbp; { return (EOPNOTSUPP); } int vfs_stdquotactl (mp, cmds, uid, arg) struct mount *mp; int cmds; uid_t uid; void *arg; { return (EOPNOTSUPP); } int vfs_stdsync(mp, waitfor) struct mount *mp; int waitfor; { struct vnode *vp, *mvp; struct thread *td; int error, lockreq, allerror = 0; td = curthread; lockreq = LK_EXCLUSIVE | LK_INTERLOCK; if (waitfor != MNT_WAIT) lockreq |= LK_NOWAIT; /* * Force stale buffer cache information to be flushed. */ loop: MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { if (vp->v_bufobj.bo_dirty.bv_cnt == 0) { VI_UNLOCK(vp); continue; } if ((error = vget(vp, lockreq, td)) != 0) { if (error == ENOENT) { MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); goto loop; } continue; } error = VOP_FSYNC(vp, waitfor, td); if (error) allerror = error; vput(vp); } return (allerror); } int vfs_stdnosync (mp, waitfor) struct mount *mp; int waitfor; { return (0); } int vfs_stdvget (mp, ino, flags, vpp) struct mount *mp; ino_t ino; int flags; struct vnode **vpp; { return (EOPNOTSUPP); } int vfs_stdfhtovp (mp, fhp, flags, vpp) struct mount *mp; struct fid *fhp; int flags; struct vnode **vpp; { return (EOPNOTSUPP); } int vfs_stdinit (vfsp) struct vfsconf *vfsp; { return (0); } int vfs_stduninit (vfsp) struct vfsconf *vfsp; { return(0); } int vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, attrname) struct mount *mp; int cmd; struct vnode *filename_vp; int attrnamespace; const char *attrname; { if (filename_vp != NULL) VOP_UNLOCK(filename_vp, 0); return (EOPNOTSUPP); } int vfs_stdsysctl(mp, op, req) struct mount *mp; fsctlop_t op; struct sysctl_req *req; { return (EOPNOTSUPP); } /* end of vfs default ops */ Index: head/sys/kern/vfs_export.c =================================================================== --- head/sys/kern/vfs_export.c (revision 305831) +++ head/sys/kern/vfs_export.c (revision 305832) @@ -1,520 +1,520 @@ /*- * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static MALLOC_DEFINE(M_NETADDR, "export_host", "Export host address structure"); static struct radix_node_head *vfs_create_addrlist_af( struct radix_node_head **prnh, int off); static void vfs_free_addrlist(struct netexport *nep); static int vfs_free_netcred(struct radix_node *rn, void *w); static void vfs_free_addrlist_af(struct radix_node_head **prnh); static int vfs_hang_addrlist(struct mount *mp, struct netexport *nep, struct export_args *argp); static struct netcred *vfs_export_lookup(struct mount *, struct sockaddr *); /* * Network address lookup element */ struct netcred { struct radix_node netc_rnodes[2]; int netc_exflags; struct ucred *netc_anon; int netc_numsecflavors; int netc_secflavors[MAXSECFLAVORS]; }; /* * Network export information */ struct netexport { struct netcred ne_defexported; /* Default export */ struct radix_node_head *ne4; struct radix_node_head *ne6; }; /* * Build hash lists of net addresses and hang them off the mount point. * Called by vfs_export() to set up the lists of export addresses. */ static int vfs_hang_addrlist(struct mount *mp, struct netexport *nep, struct export_args *argp) { register struct netcred *np; register struct radix_node_head *rnh; register int i; struct radix_node *rn; struct sockaddr *saddr, *smask = NULL; #if defined(INET6) || defined(INET) int off; #endif int error; /* * XXX: This routine converts from a `struct xucred' * (argp->ex_anon) to a `struct ucred' (np->netc_anon). This * operation is questionable; for example, what should be done * with fields like cr_uidinfo and cr_prison? Currently, this * routine does not touch them (leaves them as NULL). */ if (argp->ex_anon.cr_version != XUCRED_VERSION) { vfs_mount_error(mp, "ex_anon.cr_version: %d != %d", argp->ex_anon.cr_version, XUCRED_VERSION); return (EINVAL); } if (argp->ex_addrlen == 0) { if (mp->mnt_flag & MNT_DEFEXPORTED) { vfs_mount_error(mp, "MNT_DEFEXPORTED already set for mount %p", mp); return (EPERM); } np = &nep->ne_defexported; np->netc_exflags = argp->ex_flags; np->netc_anon = crget(); np->netc_anon->cr_uid = argp->ex_anon.cr_uid; crsetgroups(np->netc_anon, argp->ex_anon.cr_ngroups, argp->ex_anon.cr_groups); np->netc_anon->cr_prison = &prison0; prison_hold(np->netc_anon->cr_prison); np->netc_numsecflavors = argp->ex_numsecflavors; bcopy(argp->ex_secflavors, np->netc_secflavors, sizeof(np->netc_secflavors)); MNT_ILOCK(mp); mp->mnt_flag |= MNT_DEFEXPORTED; MNT_IUNLOCK(mp); return (0); } #if MSIZE <= 256 if (argp->ex_addrlen > MLEN) { vfs_mount_error(mp, "ex_addrlen %d is greater than %d", argp->ex_addrlen, MLEN); return (EINVAL); } #endif i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK | M_ZERO); saddr = (struct sockaddr *) (np + 1); if ((error = copyin(argp->ex_addr, saddr, argp->ex_addrlen))) goto out; if (saddr->sa_family == AF_UNSPEC || saddr->sa_family > AF_MAX) { error = EINVAL; vfs_mount_error(mp, "Invalid saddr->sa_family: %d"); goto out; } if (saddr->sa_len > argp->ex_addrlen) saddr->sa_len = argp->ex_addrlen; if (argp->ex_masklen) { smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); error = copyin(argp->ex_mask, smask, argp->ex_masklen); if (error) goto out; if (smask->sa_len > argp->ex_masklen) smask->sa_len = argp->ex_masklen; } rnh = NULL; switch (saddr->sa_family) { #ifdef INET case AF_INET: if ((rnh = nep->ne4) == NULL) { off = offsetof(struct sockaddr_in, sin_addr) << 3; rnh = vfs_create_addrlist_af(&nep->ne4, off); } break; #endif #ifdef INET6 case AF_INET6: if ((rnh = nep->ne6) == NULL) { off = offsetof(struct sockaddr_in6, sin6_addr) << 3; rnh = vfs_create_addrlist_af(&nep->ne6, off); } break; #endif } if (rnh == NULL) { error = ENOBUFS; vfs_mount_error(mp, "%s %s %d", "Unable to initialize radix node head ", "for address family", saddr->sa_family); goto out; } RADIX_NODE_HEAD_LOCK(rnh); rn = (*rnh->rnh_addaddr)(saddr, smask, &rnh->rh, np->netc_rnodes); RADIX_NODE_HEAD_UNLOCK(rnh); if (rn == NULL || np != (struct netcred *)rn) { /* already exists */ error = EPERM; vfs_mount_error(mp, "netcred already exists for given addr/mask"); goto out; } np->netc_exflags = argp->ex_flags; np->netc_anon = crget(); np->netc_anon->cr_uid = argp->ex_anon.cr_uid; crsetgroups(np->netc_anon, argp->ex_anon.cr_ngroups, argp->ex_anon.cr_groups); np->netc_anon->cr_prison = &prison0; prison_hold(np->netc_anon->cr_prison); np->netc_numsecflavors = argp->ex_numsecflavors; bcopy(argp->ex_secflavors, np->netc_secflavors, sizeof(np->netc_secflavors)); return (0); out: free(np, M_NETADDR); return (error); } /* Helper for vfs_free_addrlist. */ /* ARGSUSED */ static int vfs_free_netcred(struct radix_node *rn, void *w) { struct radix_node_head *rnh = (struct radix_node_head *) w; struct ucred *cred; (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, &rnh->rh); cred = ((struct netcred *)rn)->netc_anon; if (cred != NULL) crfree(cred); free(rn, M_NETADDR); return (0); } static struct radix_node_head * vfs_create_addrlist_af(struct radix_node_head **prnh, int off) { if (rn_inithead((void **)prnh, off) == 0) return (NULL); RADIX_NODE_HEAD_LOCK_INIT(*prnh); return (*prnh); } static void vfs_free_addrlist_af(struct radix_node_head **prnh) { struct radix_node_head *rnh; rnh = *prnh; RADIX_NODE_HEAD_LOCK(rnh); (*rnh->rnh_walktree)(&rnh->rh, vfs_free_netcred, rnh); RADIX_NODE_HEAD_UNLOCK(rnh); RADIX_NODE_HEAD_DESTROY(rnh); rn_detachhead((void **)prnh); prnh = NULL; } /* * Free the net address hash lists that are hanging off the mount points. */ static void vfs_free_addrlist(struct netexport *nep) { struct ucred *cred; if (nep->ne4 != NULL) vfs_free_addrlist_af(&nep->ne4); if (nep->ne6 != NULL) vfs_free_addrlist_af(&nep->ne6); cred = nep->ne_defexported.netc_anon; if (cred != NULL) crfree(cred); } /* * High level function to manipulate export options on a mount point * and the passed in netexport. * Struct export_args *argp is the variable used to twiddle options, * the structure is described in sys/mount.h */ int vfs_export(struct mount *mp, struct export_args *argp) { struct netexport *nep; int error; if (argp->ex_numsecflavors < 0 || argp->ex_numsecflavors >= MAXSECFLAVORS) return (EINVAL); error = 0; lockmgr(&mp->mnt_explock, LK_EXCLUSIVE, NULL); nep = mp->mnt_export; if (argp->ex_flags & MNT_DELEXPORT) { if (nep == NULL) { error = ENOENT; goto out; } if (mp->mnt_flag & MNT_EXPUBLIC) { vfs_setpublicfs(NULL, NULL, NULL); MNT_ILOCK(mp); mp->mnt_flag &= ~MNT_EXPUBLIC; MNT_IUNLOCK(mp); } vfs_free_addrlist(nep); mp->mnt_export = NULL; free(nep, M_MOUNT); nep = NULL; MNT_ILOCK(mp); mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); MNT_IUNLOCK(mp); } if (argp->ex_flags & MNT_EXPORTED) { if (nep == NULL) { nep = malloc(sizeof(struct netexport), M_MOUNT, M_WAITOK | M_ZERO); mp->mnt_export = nep; } if (argp->ex_flags & MNT_EXPUBLIC) { if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) goto out; MNT_ILOCK(mp); mp->mnt_flag |= MNT_EXPUBLIC; MNT_IUNLOCK(mp); } if ((error = vfs_hang_addrlist(mp, nep, argp))) goto out; MNT_ILOCK(mp); mp->mnt_flag |= MNT_EXPORTED; MNT_IUNLOCK(mp); } out: lockmgr(&mp->mnt_explock, LK_RELEASE, NULL); /* * Once we have executed the vfs_export() command, we do * not want to keep the "export" option around in the * options list, since that will cause subsequent MNT_UPDATE * calls to fail. The export information is saved in * mp->mnt_export, so we can safely delete the "export" mount option * here. */ vfs_deleteopt(mp->mnt_optnew, "export"); vfs_deleteopt(mp->mnt_opt, "export"); return (error); } /* * Set the publicly exported filesystem (WebNFS). Currently, only * one public filesystem is possible in the spec (RFC 2054 and 2055) */ int vfs_setpublicfs(struct mount *mp, struct netexport *nep, struct export_args *argp) { int error; struct vnode *rvp; char *cp; /* * mp == NULL -> invalidate the current info, the FS is * no longer exported. May be called from either vfs_export * or unmount, so check if it hasn't already been done. */ if (mp == NULL) { if (nfs_pub.np_valid) { nfs_pub.np_valid = 0; if (nfs_pub.np_index != NULL) { free(nfs_pub.np_index, M_TEMP); nfs_pub.np_index = NULL; } } return (0); } /* * Only one allowed at a time. */ if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) return (EBUSY); /* * Get real filehandle for root of exported FS. */ bzero(&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rvp))) return (error); if ((error = VOP_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) return (error); vput(rvp); /* * If an indexfile was specified, pull it in. */ if (argp->ex_indexfile != NULL) { if (nfs_pub.np_index != NULL) nfs_pub.np_index = malloc(MAXNAMLEN + 1, M_TEMP, M_WAITOK); error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, MAXNAMLEN, (size_t *)0); if (!error) { /* * Check for illegal filenames. */ for (cp = nfs_pub.np_index; *cp; cp++) { if (*cp == '/') { error = EINVAL; break; } } } if (error) { free(nfs_pub.np_index, M_TEMP); nfs_pub.np_index = NULL; return (error); } } nfs_pub.np_mount = mp; nfs_pub.np_valid = 1; return (0); } /* * Used by the filesystems to determine if a given network address * (passed in 'nam') is present in their exports list, returns a pointer * to struct netcred so that the filesystem can examine it for * access rights (read/write/etc). */ static struct netcred * vfs_export_lookup(struct mount *mp, struct sockaddr *nam) { struct netexport *nep; register struct netcred *np; register struct radix_node_head *rnh; struct sockaddr *saddr; nep = mp->mnt_export; if (nep == NULL) return (NULL); np = NULL; if (mp->mnt_flag & MNT_EXPORTED) { /* * Lookup in the export list first. */ if (nam != NULL) { saddr = nam; rnh = NULL; switch (saddr->sa_family) { case AF_INET: rnh = nep->ne4; break; case AF_INET6: rnh = nep->ne6; break; } if (rnh != NULL) { RADIX_NODE_HEAD_RLOCK(rnh); np = (struct netcred *) (*rnh->rnh_matchaddr)(saddr, &rnh->rh); RADIX_NODE_HEAD_RUNLOCK(rnh); if (np && np->netc_rnodes->rn_flags & RNF_ROOT) np = NULL; } } /* * If no address match, use the default if it exists. */ if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) np = &nep->ne_defexported; } return (np); } /* * XXX: This comment comes from the deprecated ufs_check_export() * XXX: and may not entirely apply, but lacking something better: * This is the generic part of fhtovp called after the underlying * filesystem has validated the file handle. * * Verify that a host should have access to a filesystem. */ int vfs_stdcheckexp(struct mount *mp, struct sockaddr *nam, int *extflagsp, struct ucred **credanonp, int *numsecflavors, int **secflavors) { struct netcred *np; lockmgr(&mp->mnt_explock, LK_SHARED, NULL); np = vfs_export_lookup(mp, nam); if (np == NULL) { lockmgr(&mp->mnt_explock, LK_RELEASE, NULL); *credanonp = NULL; return (EACCES); } *extflagsp = np->netc_exflags; if ((*credanonp = np->netc_anon) != NULL) crhold(*credanonp); if (numsecflavors) *numsecflavors = np->netc_numsecflavors; if (secflavors) *secflavors = np->netc_secflavors; lockmgr(&mp->mnt_explock, LK_RELEASE, NULL); return (0); } Index: head/sys/kern/vfs_init.c =================================================================== --- head/sys/kern/vfs_init.c (revision 305831) +++ head/sys/kern/vfs_init.c (revision 305832) @@ -1,370 +1,370 @@ /*- * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed * to Berkeley by John Heidemann of the UCLA Ficus project. * * Source: * @(#)i405_init.c 2.10 92/04/27 UCLA Ficus project * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_init.c 8.3 (Berkeley) 1/4/94 */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include static int vfs_register(struct vfsconf *); static int vfs_unregister(struct vfsconf *); MALLOC_DEFINE(M_VNODE, "vnodes", "Dynamically allocated vnodes"); /* * The highest defined VFS number. */ int maxvfsconf = VFS_GENERIC + 1; /* * Single-linked list of configured VFSes. * New entries are added/deleted by vfs_register()/vfs_unregister() */ struct vfsconfhead vfsconf = TAILQ_HEAD_INITIALIZER(vfsconf); struct sx vfsconf_sx; SX_SYSINIT(vfsconf, &vfsconf_sx, "vfsconf"); /* * Loader.conf variable vfs.typenumhash enables setting vfc_typenum using a hash * calculation on vfc_name, so that it doesn't change when file systems are * loaded in a different order. This will avoid the NFS server file handles from * changing for file systems that use vfc_typenum in their fsid. */ static int vfs_typenumhash = 1; SYSCTL_INT(_vfs, OID_AUTO, typenumhash, CTLFLAG_RDTUN, &vfs_typenumhash, 0, "Set vfc_typenum using a hash calculation on vfc_name, so that it does not" "change when file systems are loaded in a different order."); /* * A Zen vnode attribute structure. * * Initialized when the first filesystem registers by vfs_register(). */ struct vattr va_null; /* * vfs_init.c * * Allocate and fill in operations vectors. * * An undocumented feature of this approach to defining operations is that * there can be multiple entries in vfs_opv_descs for the same operations * vector. This allows third parties to extend the set of operations * supported by another layer in a binary compatibile way. For example, * assume that NFS needed to be modified to support Ficus. NFS has an entry * (probably nfs_vnopdeop_decls) declaring all the operations NFS supports by * default. Ficus could add another entry (ficus_nfs_vnodeop_decl_entensions) * listing those new operations Ficus adds to NFS, all without modifying the * NFS code. (Of couse, the OTW NFS protocol still needs to be munged, but * that is a(whole)nother story.) This is a feature. */ /* * Routines having to do with the management of the vnode table. */ static struct vfsconf * vfs_byname_locked(const char *name) { struct vfsconf *vfsp; sx_assert(&vfsconf_sx, SA_LOCKED); if (!strcmp(name, "ffs")) name = "ufs"; TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { if (!strcmp(name, vfsp->vfc_name)) return (vfsp); } return (NULL); } struct vfsconf * vfs_byname(const char *name) { struct vfsconf *vfsp; vfsconf_slock(); vfsp = vfs_byname_locked(name); vfsconf_sunlock(); return (vfsp); } struct vfsconf * vfs_byname_kld(const char *fstype, struct thread *td, int *error) { struct vfsconf *vfsp; int fileid, loaded; vfsp = vfs_byname(fstype); if (vfsp != NULL) return (vfsp); /* Try to load the respective module. */ *error = kern_kldload(td, fstype, &fileid); loaded = (*error == 0); if (*error == EEXIST) *error = 0; if (*error) return (NULL); /* Look up again to see if the VFS was loaded. */ vfsp = vfs_byname(fstype); if (vfsp == NULL) { if (loaded) (void)kern_kldunload(td, fileid, LINKER_UNLOAD_FORCE); *error = ENODEV; return (NULL); } return (vfsp); } /* Register a new filesystem type in the global table */ static int vfs_register(struct vfsconf *vfc) { struct sysctl_oid *oidp; struct vfsops *vfsops; static int once; struct vfsconf *tvfc; uint32_t hashval; int secondpass; if (!once) { vattr_null(&va_null); once = 1; } if (vfc->vfc_version != VFS_VERSION) { printf("ERROR: filesystem %s, unsupported ABI version %x\n", vfc->vfc_name, vfc->vfc_version); return (EINVAL); } vfsconf_lock(); if (vfs_byname_locked(vfc->vfc_name) != NULL) { vfsconf_unlock(); return (EEXIST); } if (vfs_typenumhash != 0) { /* * Calculate a hash on vfc_name to use for vfc_typenum. Unless * all of 1<->255 are assigned, it is limited to 8bits since * that is what ZFS uses from vfc_typenum and is also the * preferred range for vfs_getnewfsid(). */ hashval = fnv_32_str(vfc->vfc_name, FNV1_32_INIT); hashval &= 0xff; secondpass = 0; do { /* Look for and fix any collision. */ TAILQ_FOREACH(tvfc, &vfsconf, vfc_list) { if (hashval == tvfc->vfc_typenum) { if (hashval == 255 && secondpass == 0) { hashval = 1; secondpass = 1; } else hashval++; break; } } } while (tvfc != NULL); vfc->vfc_typenum = hashval; if (vfc->vfc_typenum >= maxvfsconf) maxvfsconf = vfc->vfc_typenum + 1; } else vfc->vfc_typenum = maxvfsconf++; TAILQ_INSERT_TAIL(&vfsconf, vfc, vfc_list); /* * Initialise unused ``struct vfsops'' fields, to use * the vfs_std*() functions. Note, we need the mount * and unmount operations, at the least. The check * for vfsops available is just a debugging aid. */ KASSERT(vfc->vfc_vfsops != NULL, ("Filesystem %s has no vfsops", vfc->vfc_name)); /* * Check the mount and unmount operations. */ vfsops = vfc->vfc_vfsops; KASSERT(vfsops->vfs_mount != NULL, ("Filesystem %s has no mount op", vfc->vfc_name)); KASSERT(vfsops->vfs_unmount != NULL, ("Filesystem %s has no unmount op", vfc->vfc_name)); if (vfsops->vfs_root == NULL) /* return file system's root vnode */ vfsops->vfs_root = vfs_stdroot; if (vfsops->vfs_quotactl == NULL) /* quota control */ vfsops->vfs_quotactl = vfs_stdquotactl; if (vfsops->vfs_statfs == NULL) /* return file system's status */ vfsops->vfs_statfs = vfs_stdstatfs; if (vfsops->vfs_sync == NULL) /* * flush unwritten data (nosync) * file systems can use vfs_stdsync * explicitly by setting it in the * vfsop vector. */ vfsops->vfs_sync = vfs_stdnosync; if (vfsops->vfs_vget == NULL) /* convert an inode number to a vnode */ vfsops->vfs_vget = vfs_stdvget; if (vfsops->vfs_fhtovp == NULL) /* turn an NFS file handle into a vnode */ vfsops->vfs_fhtovp = vfs_stdfhtovp; if (vfsops->vfs_checkexp == NULL) /* check if file system is exported */ vfsops->vfs_checkexp = vfs_stdcheckexp; if (vfsops->vfs_init == NULL) /* file system specific initialisation */ vfsops->vfs_init = vfs_stdinit; if (vfsops->vfs_uninit == NULL) /* file system specific uninitialisation */ vfsops->vfs_uninit = vfs_stduninit; if (vfsops->vfs_extattrctl == NULL) /* extended attribute control */ vfsops->vfs_extattrctl = vfs_stdextattrctl; if (vfsops->vfs_sysctl == NULL) vfsops->vfs_sysctl = vfs_stdsysctl; /* * Call init function for this VFS... */ (*(vfc->vfc_vfsops->vfs_init))(vfc); vfsconf_unlock(); /* * If this filesystem has a sysctl node under vfs * (i.e. vfs.xxfs), then change the oid number of that node to * match the filesystem's type number. This allows user code * which uses the type number to read sysctl variables defined * by the filesystem to continue working. Since the oids are * in a sorted list, we need to make sure the order is * preserved by re-registering the oid after modifying its * number. */ sysctl_wlock(); SLIST_FOREACH(oidp, SYSCTL_CHILDREN(&sysctl___vfs), oid_link) { if (strcmp(oidp->oid_name, vfc->vfc_name) == 0) { sysctl_unregister_oid(oidp); oidp->oid_number = vfc->vfc_typenum; sysctl_register_oid(oidp); break; } } sysctl_wunlock(); return (0); } /* Remove registration of a filesystem type */ static int vfs_unregister(struct vfsconf *vfc) { struct vfsconf *vfsp; int error, maxtypenum; vfsconf_lock(); vfsp = vfs_byname_locked(vfc->vfc_name); if (vfsp == NULL) { vfsconf_unlock(); return (EINVAL); } if (vfsp->vfc_refcount != 0) { vfsconf_unlock(); return (EBUSY); } if (vfc->vfc_vfsops->vfs_uninit != NULL) { error = (*vfc->vfc_vfsops->vfs_uninit)(vfsp); if (error != 0) { vfsconf_unlock(); return (error); } } TAILQ_REMOVE(&vfsconf, vfsp, vfc_list); maxtypenum = VFS_GENERIC; TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) if (maxtypenum < vfsp->vfc_typenum) maxtypenum = vfsp->vfc_typenum; maxvfsconf = maxtypenum + 1; vfsconf_unlock(); return (0); } /* * Standard kernel module handling code for filesystem modules. * Referenced from VFS_SET(). */ int vfs_modevent(module_t mod, int type, void *data) { struct vfsconf *vfc; int error = 0; vfc = (struct vfsconf *)data; switch (type) { case MOD_LOAD: if (vfc) error = vfs_register(vfc); break; case MOD_UNLOAD: if (vfc) error = vfs_unregister(vfc); break; default: error = EOPNOTSUPP; break; } return (error); } Index: head/sys/kern/vfs_lookup.c =================================================================== --- head/sys/kern/vfs_lookup.c (revision 305831) +++ head/sys/kern/vfs_lookup.c (revision 305832) @@ -1,1258 +1,1258 @@ /*- * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_lookup.c 8.4 (Berkeley) 2/16/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_capsicum.h" #include "opt_ktrace.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KTRACE #include #endif #include #include #include #define NAMEI_DIAGNOSTIC 1 #undef NAMEI_DIAGNOSTIC SDT_PROVIDER_DECLARE(vfs); SDT_PROBE_DEFINE3(vfs, namei, lookup, entry, "struct vnode *", "char *", "unsigned long"); SDT_PROBE_DEFINE2(vfs, namei, lookup, return, "int", "struct vnode *"); /* * Allocation zone for namei */ uma_zone_t namei_zone; /* * Placeholder vnode for mp traversal */ static struct vnode *vp_crossmp; static void nameiinit(void *dummy __unused) { namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); getnewvnode("crossmp", NULL, &dead_vnodeops, &vp_crossmp); vn_lock(vp_crossmp, LK_EXCLUSIVE); VN_LOCK_ASHARE(vp_crossmp); VOP_UNLOCK(vp_crossmp, 0); } SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL); static int lookup_shared = 1; SYSCTL_INT(_vfs, OID_AUTO, lookup_shared, CTLFLAG_RWTUN, &lookup_shared, 0, "Enables/Disables shared locks for path name translation"); static void namei_cleanup_cnp(struct componentname *cnp) { uma_zfree(namei_zone, cnp->cn_pnbuf); #ifdef DIAGNOSTIC cnp->cn_pnbuf = NULL; cnp->cn_nameptr = NULL; #endif } static int namei_handle_root(struct nameidata *ndp, struct vnode **dpp) { struct componentname *cnp; cnp = &ndp->ni_cnd; if (ndp->ni_strictrelative != 0) { #ifdef KTRACE if (KTRPOINT(curthread, KTR_CAPFAIL)) ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL); #endif return (ENOTCAPABLE); } while (*(cnp->cn_nameptr) == '/') { cnp->cn_nameptr++; ndp->ni_pathlen--; } *dpp = ndp->ni_rootdir; VREF(*dpp); return (0); } /* * Convert a pathname into a pointer to a locked vnode. * * The FOLLOW flag is set when symbolic links are to be followed * when they occur at the end of the name translation process. * Symbolic links are always followed for all other pathname * components other than the last. * * The segflg defines whether the name is to be copied from user * space or kernel space. * * Overall outline of namei: * * copy in name * get starting directory * while (!done && !error) { * call lookup to search path. * if symbolic link, massage name in buffer and continue * } */ int namei(struct nameidata *ndp) { struct filedesc *fdp; /* pointer to file descriptor state */ char *cp; /* pointer into pathname argument */ struct vnode *dp; /* the directory we are searching */ struct iovec aiov; /* uio for reading symbolic links */ struct uio auio; int error, linklen, startdir_used; struct componentname *cnp = &ndp->ni_cnd; struct thread *td = cnp->cn_thread; struct proc *p = td->td_proc; ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred; KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc")); KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0, ("namei: nameiop contaminated with flags")); KASSERT((cnp->cn_flags & OPMASK) == 0, ("namei: flags contaminated with nameiops")); MPASS(ndp->ni_startdir == NULL || ndp->ni_startdir->v_type == VDIR || ndp->ni_startdir->v_type == VBAD); if (!lookup_shared) cnp->cn_flags &= ~LOCKSHARED; fdp = p->p_fd; /* We will set this ourselves if we need it. */ cnp->cn_flags &= ~TRAILINGSLASH; /* * Get a buffer for the name to be translated, and copy the * name into the buffer. */ if ((cnp->cn_flags & HASBUF) == 0) cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK); if (ndp->ni_segflg == UIO_SYSSPACE) error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN, (size_t *)&ndp->ni_pathlen); else error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN, (size_t *)&ndp->ni_pathlen); /* * Don't allow empty pathnames. */ if (error == 0 && *cnp->cn_pnbuf == '\0') error = ENOENT; #ifdef CAPABILITY_MODE /* * In capability mode, lookups must be "strictly relative" (i.e. * not an absolute path, and not containing '..' components) to * a real file descriptor, not the pseudo-descriptor AT_FDCWD. */ if (error == 0 && IN_CAPABILITY_MODE(td) && (cnp->cn_flags & NOCAPCHECK) == 0) { ndp->ni_strictrelative = 1; if (ndp->ni_dirfd == AT_FDCWD) { #ifdef KTRACE if (KTRPOINT(td, KTR_CAPFAIL)) ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL); #endif error = ECAPMODE; } } #endif if (error != 0) { namei_cleanup_cnp(cnp); ndp->ni_vp = NULL; return (error); } ndp->ni_loopcnt = 0; #ifdef KTRACE if (KTRPOINT(td, KTR_NAMEI)) { KASSERT(cnp->cn_thread == curthread, ("namei not using curthread")); ktrnamei(cnp->cn_pnbuf); } #endif /* * Get starting point for the translation. */ FILEDESC_SLOCK(fdp); ndp->ni_rootdir = fdp->fd_rdir; VREF(ndp->ni_rootdir); ndp->ni_topdir = fdp->fd_jdir; /* * If we are auditing the kernel pathname, save the user pathname. */ if (cnp->cn_flags & AUDITVNODE1) AUDIT_ARG_UPATH1(td, ndp->ni_dirfd, cnp->cn_pnbuf); if (cnp->cn_flags & AUDITVNODE2) AUDIT_ARG_UPATH2(td, ndp->ni_dirfd, cnp->cn_pnbuf); startdir_used = 0; dp = NULL; cnp->cn_nameptr = cnp->cn_pnbuf; if (cnp->cn_pnbuf[0] == '/') { error = namei_handle_root(ndp, &dp); } else { if (ndp->ni_startdir != NULL) { dp = ndp->ni_startdir; startdir_used = 1; } else if (ndp->ni_dirfd == AT_FDCWD) { dp = fdp->fd_cdir; VREF(dp); } else { cap_rights_t rights; rights = ndp->ni_rightsneeded; cap_rights_set(&rights, CAP_LOOKUP); if (cnp->cn_flags & AUDITVNODE1) AUDIT_ARG_ATFD1(ndp->ni_dirfd); if (cnp->cn_flags & AUDITVNODE2) AUDIT_ARG_ATFD2(ndp->ni_dirfd); error = fgetvp_rights(td, ndp->ni_dirfd, &rights, &ndp->ni_filecaps, &dp); if (error == EINVAL) error = ENOTDIR; #ifdef CAPABILITIES /* * If file descriptor doesn't have all rights, * all lookups relative to it must also be * strictly relative. */ CAP_ALL(&rights); if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) || ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL || ndp->ni_filecaps.fc_nioctls != -1) { ndp->ni_strictrelative = 1; } #endif } if (error == 0 && dp->v_type != VDIR) error = ENOTDIR; } FILEDESC_SUNLOCK(fdp); if (ndp->ni_startdir != NULL && !startdir_used) vrele(ndp->ni_startdir); if (error != 0) { if (dp != NULL) vrele(dp); goto out; } SDT_PROBE3(vfs, namei, lookup, entry, dp, cnp->cn_pnbuf, cnp->cn_flags); for (;;) { ndp->ni_startdir = dp; error = lookup(ndp); if (error != 0) goto out; /* * If not a symbolic link, we're done. */ if ((cnp->cn_flags & ISSYMLINK) == 0) { vrele(ndp->ni_rootdir); if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) { namei_cleanup_cnp(cnp); } else cnp->cn_flags |= HASBUF; SDT_PROBE2(vfs, namei, lookup, return, 0, ndp->ni_vp); return (0); } if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { error = ELOOP; break; } #ifdef MAC if ((cnp->cn_flags & NOMACCHECK) == 0) { error = mac_vnode_check_readlink(td->td_ucred, ndp->ni_vp); if (error != 0) break; } #endif if (ndp->ni_pathlen > 1) cp = uma_zalloc(namei_zone, M_WAITOK); else cp = cnp->cn_pnbuf; aiov.iov_base = cp; aiov.iov_len = MAXPATHLEN; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_offset = 0; auio.uio_rw = UIO_READ; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = td; auio.uio_resid = MAXPATHLEN; error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); if (error != 0) { if (ndp->ni_pathlen > 1) uma_zfree(namei_zone, cp); break; } linklen = MAXPATHLEN - auio.uio_resid; if (linklen == 0) { if (ndp->ni_pathlen > 1) uma_zfree(namei_zone, cp); error = ENOENT; break; } if (linklen + ndp->ni_pathlen >= MAXPATHLEN) { if (ndp->ni_pathlen > 1) uma_zfree(namei_zone, cp); error = ENAMETOOLONG; break; } if (ndp->ni_pathlen > 1) { bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen); uma_zfree(namei_zone, cnp->cn_pnbuf); cnp->cn_pnbuf = cp; } else cnp->cn_pnbuf[linklen] = '\0'; ndp->ni_pathlen += linklen; vput(ndp->ni_vp); dp = ndp->ni_dvp; /* * Check if root directory should replace current directory. */ cnp->cn_nameptr = cnp->cn_pnbuf; if (*(cnp->cn_nameptr) == '/') { vrele(dp); error = namei_handle_root(ndp, &dp); if (error != 0) goto out; } } vput(ndp->ni_vp); ndp->ni_vp = NULL; vrele(ndp->ni_dvp); out: vrele(ndp->ni_rootdir); namei_cleanup_cnp(cnp); SDT_PROBE2(vfs, namei, lookup, return, error, NULL); return (error); } static int compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags) { if (mp == NULL || ((lkflags & LK_SHARED) && (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) || ((cnflags & ISDOTDOT) && (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) { lkflags &= ~LK_SHARED; lkflags |= LK_EXCLUSIVE; } lkflags |= LK_NODDLKTREAT; return (lkflags); } static __inline int needs_exclusive_leaf(struct mount *mp, int flags) { /* * Intermediate nodes can use shared locks, we only need to * force an exclusive lock for leaf nodes. */ if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF)) return (0); /* Always use exclusive locks if LOCKSHARED isn't set. */ if (!(flags & LOCKSHARED)) return (1); /* * For lookups during open(), if the mount point supports * extended shared operations, then use a shared lock for the * leaf node, otherwise use an exclusive lock. */ if ((flags & ISOPEN) != 0) return (!MNT_EXTENDED_SHARED(mp)); /* * Lookup requests outside of open() that specify LOCKSHARED * only need a shared lock on the leaf vnode. */ return (0); } /* * Search a pathname. * This is a very central and rather complicated routine. * * The pathname is pointed to by ni_ptr and is of length ni_pathlen. * The starting directory is taken from ni_startdir. The pathname is * descended until done, or a symbolic link is encountered. The variable * ni_more is clear if the path is completed; it is set to one if a * symbolic link needing interpretation is encountered. * * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on * whether the name is to be looked up, created, renamed, or deleted. * When CREATE, RENAME, or DELETE is specified, information usable in * creating, renaming, or deleting a directory entry may be calculated. * If flag has LOCKPARENT or'ed into it, the parent directory is returned * locked. If flag has WANTPARENT or'ed into it, the parent directory is * returned unlocked. Otherwise the parent directory is not returned. If * the target of the pathname exists and LOCKLEAF is or'ed into the flag * the target is returned locked, otherwise it is returned unlocked. * When creating or renaming and LOCKPARENT is specified, the target may not * be ".". When deleting and LOCKPARENT is specified, the target may be ".". * * Overall outline of lookup: * * dirloop: * identify next component of name at ndp->ni_ptr * handle degenerate case where name is null string * if .. and crossing mount points and on mounted filesys, find parent * call VOP_LOOKUP routine for next component name * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set * component vnode returned in ni_vp (if it exists), locked. * if result vnode is mounted on and crossing mount points, * find mounted on vnode * if more components of name, do next level at dirloop * return the answer in ni_vp, locked if LOCKLEAF set * if LOCKPARENT set, return locked parent in ni_dvp * if WANTPARENT set, return unlocked parent in ni_dvp */ int lookup(struct nameidata *ndp) { char *cp; /* pointer into pathname argument */ struct vnode *dp = NULL; /* the directory we are searching */ struct vnode *tdp; /* saved dp */ struct mount *mp; /* mount table entry */ struct prison *pr; int docache; /* == 0 do not cache last component */ int wantparent; /* 1 => wantparent or lockparent flag */ int rdonly; /* lookup read-only flag bit */ int error = 0; int dpunlocked = 0; /* dp has already been unlocked */ int relookup = 0; /* do not consume the path component */ struct componentname *cnp = &ndp->ni_cnd; int lkflags_save; int ni_dvp_unlocked; /* * Setup: break out flag bits into variables. */ ni_dvp_unlocked = 0; wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT); KASSERT(cnp->cn_nameiop == LOOKUP || wantparent, ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT.")); docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE; if (cnp->cn_nameiop == DELETE || (wantparent && cnp->cn_nameiop != CREATE && cnp->cn_nameiop != LOOKUP)) docache = 0; rdonly = cnp->cn_flags & RDONLY; cnp->cn_flags &= ~ISSYMLINK; ndp->ni_dvp = NULL; /* * We use shared locks until we hit the parent of the last cn then * we adjust based on the requesting flags. */ if (lookup_shared) cnp->cn_lkflags = LK_SHARED; else cnp->cn_lkflags = LK_EXCLUSIVE; dp = ndp->ni_startdir; ndp->ni_startdir = NULLVP; vn_lock(dp, compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY, cnp->cn_flags)); dirloop: /* * Search a new directory. * * The last component of the filename is left accessible via * cnp->cn_nameptr for callers that need the name. Callers needing * the name set the SAVENAME flag. When done, they assume * responsibility for freeing the pathname buffer. */ for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++) continue; cnp->cn_namelen = cp - cnp->cn_nameptr; if (cnp->cn_namelen > NAME_MAX) { error = ENAMETOOLONG; goto bad; } #ifdef NAMEI_DIAGNOSTIC { char c = *cp; *cp = '\0'; printf("{%s}: ", cnp->cn_nameptr); *cp = c; } #endif ndp->ni_pathlen -= cnp->cn_namelen; ndp->ni_next = cp; /* * Replace multiple slashes by a single slash and trailing slashes * by a null. This must be done before VOP_LOOKUP() because some * fs's don't know about trailing slashes. Remember if there were * trailing slashes to handle symlinks, existing non-directories * and non-existing files that won't be directories specially later. */ while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) { cp++; ndp->ni_pathlen--; if (*cp == '\0') { *ndp->ni_next = '\0'; cnp->cn_flags |= TRAILINGSLASH; } } ndp->ni_next = cp; cnp->cn_flags |= MAKEENTRY; if (*cp == '\0' && docache == 0) cnp->cn_flags &= ~MAKEENTRY; if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.') cnp->cn_flags |= ISDOTDOT; else cnp->cn_flags &= ~ISDOTDOT; if (*ndp->ni_next == 0) cnp->cn_flags |= ISLASTCN; else cnp->cn_flags &= ~ISLASTCN; if ((cnp->cn_flags & ISLASTCN) != 0 && cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { error = EINVAL; goto bad; } /* * Check for degenerate name (e.g. / or "") * which is a way of talking about a directory, * e.g. like "/." or ".". */ if (cnp->cn_nameptr[0] == '\0') { if (dp->v_type != VDIR) { error = ENOTDIR; goto bad; } if (cnp->cn_nameiop != LOOKUP) { error = EISDIR; goto bad; } if (wantparent) { ndp->ni_dvp = dp; VREF(dp); } ndp->ni_vp = dp; if (cnp->cn_flags & AUDITVNODE1) AUDIT_ARG_VNODE1(dp); else if (cnp->cn_flags & AUDITVNODE2) AUDIT_ARG_VNODE2(dp); if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF))) VOP_UNLOCK(dp, 0); /* XXX This should probably move to the top of function. */ if (cnp->cn_flags & SAVESTART) panic("lookup: SAVESTART"); goto success; } /* * Handle "..": five special cases. * 0. If doing a capability lookup, return ENOTCAPABLE (this is a * fairly conservative design choice, but it's the only one that we * are satisfied guarantees the property we're looking for). * 1. Return an error if this is the last component of * the name and the operation is DELETE or RENAME. * 2. If at root directory (e.g. after chroot) * or at absolute root directory * then ignore it so can't get out. * 3. If this vnode is the root of a mounted * filesystem, then replace it with the * vnode which was mounted on so we take the * .. in the other filesystem. * 4. If the vnode is the top directory of * the jail or chroot, don't let them out. */ if (cnp->cn_flags & ISDOTDOT) { if (ndp->ni_strictrelative != 0) { #ifdef KTRACE if (KTRPOINT(curthread, KTR_CAPFAIL)) ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL); #endif error = ENOTCAPABLE; goto bad; } if ((cnp->cn_flags & ISLASTCN) != 0 && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { error = EINVAL; goto bad; } for (;;) { for (pr = cnp->cn_cred->cr_prison; pr != NULL; pr = pr->pr_parent) if (dp == pr->pr_root) break; if (dp == ndp->ni_rootdir || dp == ndp->ni_topdir || dp == rootvnode || pr != NULL || ((dp->v_vflag & VV_ROOT) != 0 && (cnp->cn_flags & NOCROSSMOUNT) != 0)) { ndp->ni_dvp = dp; ndp->ni_vp = dp; VREF(dp); goto nextname; } if ((dp->v_vflag & VV_ROOT) == 0) break; if (dp->v_iflag & VI_DOOMED) { /* forced unmount */ error = ENOENT; goto bad; } tdp = dp; dp = dp->v_mount->mnt_vnodecovered; VREF(dp); vput(tdp); vn_lock(dp, compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY, ISDOTDOT)); } } /* * We now have a segment name to search for, and a directory to search. */ unionlookup: #ifdef MAC if ((cnp->cn_flags & NOMACCHECK) == 0) { error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp, cnp); if (error) goto bad; } #endif ndp->ni_dvp = dp; ndp->ni_vp = NULL; ASSERT_VOP_LOCKED(dp, "lookup"); /* * If we have a shared lock we may need to upgrade the lock for the * last operation. */ if (dp != vp_crossmp && VOP_ISLOCKED(dp) == LK_SHARED && (cnp->cn_flags & ISLASTCN) && (cnp->cn_flags & LOCKPARENT)) vn_lock(dp, LK_UPGRADE|LK_RETRY); if ((dp->v_iflag & VI_DOOMED) != 0) { error = ENOENT; goto bad; } /* * If we're looking up the last component and we need an exclusive * lock, adjust our lkflags. */ if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags)) cnp->cn_lkflags = LK_EXCLUSIVE; #ifdef NAMEI_DIAGNOSTIC vn_printf(dp, "lookup in "); #endif lkflags_save = cnp->cn_lkflags; cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags, cnp->cn_flags); error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp); cnp->cn_lkflags = lkflags_save; if (error != 0) { KASSERT(ndp->ni_vp == NULL, ("leaf should be empty")); #ifdef NAMEI_DIAGNOSTIC printf("not found\n"); #endif if ((error == ENOENT) && (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) && (dp->v_mount->mnt_flag & MNT_UNION)) { tdp = dp; dp = dp->v_mount->mnt_vnodecovered; VREF(dp); vput(tdp); vn_lock(dp, compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY, cnp->cn_flags)); goto unionlookup; } if (error == ERELOOKUP) { vref(dp); ndp->ni_vp = dp; error = 0; relookup = 1; goto good; } if (error != EJUSTRETURN) goto bad; /* * At this point, we know we're at the end of the * pathname. If creating / renaming, we can consider * allowing the file or directory to be created / renamed, * provided we're not on a read-only filesystem. */ if (rdonly) { error = EROFS; goto bad; } /* trailing slash only allowed for directories */ if ((cnp->cn_flags & TRAILINGSLASH) && !(cnp->cn_flags & WILLBEDIR)) { error = ENOENT; goto bad; } if ((cnp->cn_flags & LOCKPARENT) == 0) VOP_UNLOCK(dp, 0); /* * We return with ni_vp NULL to indicate that the entry * doesn't currently exist, leaving a pointer to the * (possibly locked) directory vnode in ndp->ni_dvp. */ if (cnp->cn_flags & SAVESTART) { ndp->ni_startdir = ndp->ni_dvp; VREF(ndp->ni_startdir); } goto success; } good: #ifdef NAMEI_DIAGNOSTIC printf("found\n"); #endif dp = ndp->ni_vp; /* * Check to see if the vnode has been mounted on; * if so find the root of the mounted filesystem. */ while (dp->v_type == VDIR && (mp = dp->v_mountedhere) && (cnp->cn_flags & NOCROSSMOUNT) == 0) { if (vfs_busy(mp, 0)) continue; vput(dp); if (dp != ndp->ni_dvp) vput(ndp->ni_dvp); else vrele(ndp->ni_dvp); vref(vp_crossmp); ndp->ni_dvp = vp_crossmp; error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags, cnp->cn_flags), &tdp); vfs_unbusy(mp); if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT)) panic("vp_crossmp exclusively locked or reclaimed"); if (error) { dpunlocked = 1; goto bad2; } ndp->ni_vp = dp = tdp; } /* * Check for symbolic link */ if ((dp->v_type == VLNK) && ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) || *ndp->ni_next == '/')) { cnp->cn_flags |= ISSYMLINK; if (dp->v_iflag & VI_DOOMED) { /* * We can't know whether the directory was mounted with * NOSYMFOLLOW, so we can't follow safely. */ error = ENOENT; goto bad2; } if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) { error = EACCES; goto bad2; } /* * Symlink code always expects an unlocked dvp. */ if (ndp->ni_dvp != ndp->ni_vp) { VOP_UNLOCK(ndp->ni_dvp, 0); ni_dvp_unlocked = 1; } goto success; } nextname: /* * Not a symbolic link that we will follow. Continue with the * next component if there is any; otherwise, we're done. */ KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/', ("lookup: invalid path state.")); if (relookup) { relookup = 0; if (ndp->ni_dvp != dp) vput(ndp->ni_dvp); else vrele(ndp->ni_dvp); goto dirloop; } if (*ndp->ni_next == '/') { cnp->cn_nameptr = ndp->ni_next; while (*cnp->cn_nameptr == '/') { cnp->cn_nameptr++; ndp->ni_pathlen--; } if (ndp->ni_dvp != dp) vput(ndp->ni_dvp); else vrele(ndp->ni_dvp); goto dirloop; } /* * If we're processing a path with a trailing slash, * check that the end result is a directory. */ if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) { error = ENOTDIR; goto bad2; } /* * Disallow directory write attempts on read-only filesystems. */ if (rdonly && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { error = EROFS; goto bad2; } if (cnp->cn_flags & SAVESTART) { ndp->ni_startdir = ndp->ni_dvp; VREF(ndp->ni_startdir); } if (!wantparent) { ni_dvp_unlocked = 2; if (ndp->ni_dvp != dp) vput(ndp->ni_dvp); else vrele(ndp->ni_dvp); } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) { VOP_UNLOCK(ndp->ni_dvp, 0); ni_dvp_unlocked = 1; } if (cnp->cn_flags & AUDITVNODE1) AUDIT_ARG_VNODE1(dp); else if (cnp->cn_flags & AUDITVNODE2) AUDIT_ARG_VNODE2(dp); if ((cnp->cn_flags & LOCKLEAF) == 0) VOP_UNLOCK(dp, 0); success: /* * Because of lookup_shared we may have the vnode shared locked, but * the caller may want it to be exclusively locked. */ if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) && VOP_ISLOCKED(dp) != LK_EXCLUSIVE) { vn_lock(dp, LK_UPGRADE | LK_RETRY); if (dp->v_iflag & VI_DOOMED) { error = ENOENT; goto bad2; } } return (0); bad2: if (ni_dvp_unlocked != 2) { if (dp != ndp->ni_dvp && !ni_dvp_unlocked) vput(ndp->ni_dvp); else vrele(ndp->ni_dvp); } bad: if (!dpunlocked) vput(dp); ndp->ni_vp = NULL; return (error); } /* * relookup - lookup a path name component * Used by lookup to re-acquire things. */ int relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp) { struct vnode *dp = NULL; /* the directory we are searching */ int wantparent; /* 1 => wantparent or lockparent flag */ int rdonly; /* lookup read-only flag bit */ int error = 0; KASSERT(cnp->cn_flags & ISLASTCN, ("relookup: Not given last component.")); /* * Setup: break out flag bits into variables. */ wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT); KASSERT(wantparent, ("relookup: parent not wanted.")); rdonly = cnp->cn_flags & RDONLY; cnp->cn_flags &= ~ISSYMLINK; dp = dvp; cnp->cn_lkflags = LK_EXCLUSIVE; vn_lock(dp, LK_EXCLUSIVE | LK_RETRY); /* * Search a new directory. * * The last component of the filename is left accessible via * cnp->cn_nameptr for callers that need the name. Callers needing * the name set the SAVENAME flag. When done, they assume * responsibility for freeing the pathname buffer. */ #ifdef NAMEI_DIAGNOSTIC printf("{%s}: ", cnp->cn_nameptr); #endif /* * Check for "" which represents the root directory after slash * removal. */ if (cnp->cn_nameptr[0] == '\0') { /* * Support only LOOKUP for "/" because lookup() * can't succeed for CREATE, DELETE and RENAME. */ KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP")); KASSERT(dp->v_type == VDIR, ("dp is not a directory")); if (!(cnp->cn_flags & LOCKLEAF)) VOP_UNLOCK(dp, 0); *vpp = dp; /* XXX This should probably move to the top of function. */ if (cnp->cn_flags & SAVESTART) panic("lookup: SAVESTART"); return (0); } if (cnp->cn_flags & ISDOTDOT) panic ("relookup: lookup on dot-dot"); /* * We now have a segment name to search for, and a directory to search. */ #ifdef NAMEI_DIAGNOSTIC vn_printf(dp, "search in "); #endif if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) { KASSERT(*vpp == NULL, ("leaf should be empty")); if (error != EJUSTRETURN) goto bad; /* * If creating and at end of pathname, then can consider * allowing file to be created. */ if (rdonly) { error = EROFS; goto bad; } /* ASSERT(dvp == ndp->ni_startdir) */ if (cnp->cn_flags & SAVESTART) VREF(dvp); if ((cnp->cn_flags & LOCKPARENT) == 0) VOP_UNLOCK(dp, 0); /* * We return with ni_vp NULL to indicate that the entry * doesn't currently exist, leaving a pointer to the * (possibly locked) directory vnode in ndp->ni_dvp. */ return (0); } dp = *vpp; /* * Disallow directory write attempts on read-only filesystems. */ if (rdonly && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { if (dvp == dp) vrele(dvp); else vput(dvp); error = EROFS; goto bad; } /* * Set the parent lock/ref state to the requested state. */ if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) { if (wantparent) VOP_UNLOCK(dvp, 0); else vput(dvp); } else if (!wantparent) vrele(dvp); /* * Check for symbolic link */ KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW), ("relookup: symlink found.\n")); /* ASSERT(dvp == ndp->ni_startdir) */ if (cnp->cn_flags & SAVESTART) VREF(dvp); if ((cnp->cn_flags & LOCKLEAF) == 0) VOP_UNLOCK(dp, 0); return (0); bad: vput(dp); *vpp = NULL; return (error); } void NDINIT_ALL(struct nameidata *ndp, u_long op, u_long flags, enum uio_seg segflg, const char *namep, int dirfd, struct vnode *startdir, cap_rights_t *rightsp, struct thread *td) { ndp->ni_cnd.cn_nameiop = op; ndp->ni_cnd.cn_flags = flags; ndp->ni_segflg = segflg; ndp->ni_dirp = namep; ndp->ni_dirfd = dirfd; ndp->ni_startdir = startdir; ndp->ni_strictrelative = 0; if (rightsp != NULL) ndp->ni_rightsneeded = *rightsp; else cap_rights_init(&ndp->ni_rightsneeded); filecaps_init(&ndp->ni_filecaps); ndp->ni_cnd.cn_thread = td; } /* * Free data allocated by namei(); see namei(9) for details. */ void NDFREE(struct nameidata *ndp, const u_int flags) { int unlock_dvp; int unlock_vp; unlock_dvp = 0; unlock_vp = 0; if (!(flags & NDF_NO_FREE_PNBUF) && (ndp->ni_cnd.cn_flags & HASBUF)) { uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); ndp->ni_cnd.cn_flags &= ~HASBUF; } if (!(flags & NDF_NO_VP_UNLOCK) && (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp) unlock_vp = 1; if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) { if (unlock_vp) { vput(ndp->ni_vp); unlock_vp = 0; } else vrele(ndp->ni_vp); ndp->ni_vp = NULL; } if (unlock_vp) VOP_UNLOCK(ndp->ni_vp, 0); if (!(flags & NDF_NO_DVP_UNLOCK) && (ndp->ni_cnd.cn_flags & LOCKPARENT) && ndp->ni_dvp != ndp->ni_vp) unlock_dvp = 1; if (!(flags & NDF_NO_DVP_RELE) && (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) { if (unlock_dvp) { vput(ndp->ni_dvp); unlock_dvp = 0; } else vrele(ndp->ni_dvp); ndp->ni_dvp = NULL; } if (unlock_dvp) VOP_UNLOCK(ndp->ni_dvp, 0); if (!(flags & NDF_NO_STARTDIR_RELE) && (ndp->ni_cnd.cn_flags & SAVESTART)) { vrele(ndp->ni_startdir); ndp->ni_startdir = NULL; } } /* * Determine if there is a suitable alternate filename under the specified * prefix for the specified path. If the create flag is set, then the * alternate prefix will be used so long as the parent directory exists. * This is used by the various compatibility ABIs so that Linux binaries prefer * files under /compat/linux for example. The chosen path (whether under * the prefix or under /) is returned in a kernel malloc'd buffer pointed * to by pathbuf. The caller is responsible for free'ing the buffer from * the M_TEMP bucket if one is returned. */ int kern_alternate_path(struct thread *td, const char *prefix, const char *path, enum uio_seg pathseg, char **pathbuf, int create, int dirfd) { struct nameidata nd, ndroot; char *ptr, *buf, *cp; size_t len, sz; int error; buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK); *pathbuf = buf; /* Copy the prefix into the new pathname as a starting point. */ len = strlcpy(buf, prefix, MAXPATHLEN); if (len >= MAXPATHLEN) { *pathbuf = NULL; free(buf, M_TEMP); return (EINVAL); } sz = MAXPATHLEN - len; ptr = buf + len; /* Append the filename to the prefix. */ if (pathseg == UIO_SYSSPACE) error = copystr(path, ptr, sz, &len); else error = copyinstr(path, ptr, sz, &len); if (error) { *pathbuf = NULL; free(buf, M_TEMP); return (error); } /* Only use a prefix with absolute pathnames. */ if (*ptr != '/') { error = EINVAL; goto keeporig; } if (dirfd != AT_FDCWD) { /* * We want the original because the "prefix" is * included in the already opened dirfd. */ bcopy(ptr, buf, len); return (0); } /* * We know that there is a / somewhere in this pathname. * Search backwards for it, to find the file's parent dir * to see if it exists in the alternate tree. If it does, * and we want to create a file (cflag is set). We don't * need to worry about the root comparison in this case. */ if (create) { for (cp = &ptr[len] - 1; *cp != '/'; cp--); *cp = '\0'; NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, buf, td); error = namei(&nd); *cp = '/'; if (error != 0) goto keeporig; } else { NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, buf, td); error = namei(&nd); if (error != 0) goto keeporig; /* * We now compare the vnode of the prefix to the one * vnode asked. If they resolve to be the same, then we * ignore the match so that the real root gets used. * This avoids the problem of traversing "../.." to find the * root directory and never finding it, because "/" resolves * to the emulation root directory. This is expensive :-( */ NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix, td); /* We shouldn't ever get an error from this namei(). */ error = namei(&ndroot); if (error == 0) { if (nd.ni_vp == ndroot.ni_vp) error = ENOENT; NDFREE(&ndroot, NDF_ONLY_PNBUF); vrele(ndroot.ni_vp); } } NDFREE(&nd, NDF_ONLY_PNBUF); vrele(nd.ni_vp); keeporig: /* If there was an error, use the original path name. */ if (error) bcopy(ptr, buf, len); return (error); } Index: head/sys/kern/vfs_mount.c =================================================================== --- head/sys/kern/vfs_mount.c (revision 305831) +++ head/sys/kern/vfs_mount.c (revision 305832) @@ -1,1992 +1,1992 @@ /*- * Copyright (c) 1999-2004 Poul-Henning Kamp * Copyright (c) 1999 Michael Smith * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define VFS_MOUNTARG_SIZE_MAX (1024 * 64) static int vfs_domount(struct thread *td, const char *fstype, char *fspath, uint64_t fsflags, struct vfsoptlist **optlist); static void free_mntarg(struct mntarg *ma); static int usermount = 0; SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, "Unprivileged users may mount and unmount file systems"); MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure"); static uma_zone_t mount_zone; /* List of mounted filesystems. */ struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); /* For any iteration/modification of mountlist */ struct mtx mountlist_mtx; MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF); /* * Global opts, taken by all filesystems */ static const char *global_opts[] = { "errmsg", "fstype", "fspath", "ro", "rw", "nosuid", "noexec", NULL }; static int mount_init(void *mem, int size, int flags) { struct mount *mp; mp = (struct mount *)mem; mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0); return (0); } static void mount_fini(void *mem, int size) { struct mount *mp; mp = (struct mount *)mem; lockdestroy(&mp->mnt_explock); mtx_destroy(&mp->mnt_mtx); } static void vfs_mount_init(void *dummy __unused) { mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL, NULL, mount_init, mount_fini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); } SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL); /* * --------------------------------------------------------------------- * Functions for building and sanitizing the mount options */ /* Remove one mount option. */ static void vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt) { TAILQ_REMOVE(opts, opt, link); free(opt->name, M_MOUNT); if (opt->value != NULL) free(opt->value, M_MOUNT); free(opt, M_MOUNT); } /* Release all resources related to the mount options. */ void vfs_freeopts(struct vfsoptlist *opts) { struct vfsopt *opt; while (!TAILQ_EMPTY(opts)) { opt = TAILQ_FIRST(opts); vfs_freeopt(opts, opt); } free(opts, M_MOUNT); } void vfs_deleteopt(struct vfsoptlist *opts, const char *name) { struct vfsopt *opt, *temp; if (opts == NULL) return; TAILQ_FOREACH_SAFE(opt, opts, link, temp) { if (strcmp(opt->name, name) == 0) vfs_freeopt(opts, opt); } } static int vfs_isopt_ro(const char *opt) { if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 || strcmp(opt, "norw") == 0) return (1); return (0); } static int vfs_isopt_rw(const char *opt) { if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0) return (1); return (0); } /* * Check if options are equal (with or without the "no" prefix). */ static int vfs_equalopts(const char *opt1, const char *opt2) { char *p; /* "opt" vs. "opt" or "noopt" vs. "noopt" */ if (strcmp(opt1, opt2) == 0) return (1); /* "noopt" vs. "opt" */ if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) return (1); /* "opt" vs. "noopt" */ if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) return (1); while ((p = strchr(opt1, '.')) != NULL && !strncmp(opt1, opt2, ++p - opt1)) { opt2 += p - opt1; opt1 = p; /* "foo.noopt" vs. "foo.opt" */ if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) return (1); /* "foo.opt" vs. "foo.noopt" */ if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) return (1); } /* "ro" / "rdonly" / "norw" / "rw" / "noro" */ if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) && (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2))) return (1); return (0); } /* * If a mount option is specified several times, * (with or without the "no" prefix) only keep * the last occurrence of it. */ static void vfs_sanitizeopts(struct vfsoptlist *opts) { struct vfsopt *opt, *opt2, *tmp; TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { opt2 = TAILQ_PREV(opt, vfsoptlist, link); while (opt2 != NULL) { if (vfs_equalopts(opt->name, opt2->name)) { tmp = TAILQ_PREV(opt2, vfsoptlist, link); vfs_freeopt(opts, opt2); opt2 = tmp; } else { opt2 = TAILQ_PREV(opt2, vfsoptlist, link); } } } } /* * Build a linked list of mount options from a struct uio. */ int vfs_buildopts(struct uio *auio, struct vfsoptlist **options) { struct vfsoptlist *opts; struct vfsopt *opt; size_t memused, namelen, optlen; unsigned int i, iovcnt; int error; opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); TAILQ_INIT(opts); memused = 0; iovcnt = auio->uio_iovcnt; for (i = 0; i < iovcnt; i += 2) { namelen = auio->uio_iov[i].iov_len; optlen = auio->uio_iov[i + 1].iov_len; memused += sizeof(struct vfsopt) + optlen + namelen; /* * Avoid consuming too much memory, and attempts to overflow * memused. */ if (memused > VFS_MOUNTARG_SIZE_MAX || optlen > VFS_MOUNTARG_SIZE_MAX || namelen > VFS_MOUNTARG_SIZE_MAX) { error = EINVAL; goto bad; } opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); opt->name = malloc(namelen, M_MOUNT, M_WAITOK); opt->value = NULL; opt->len = 0; opt->pos = i / 2; opt->seen = 0; /* * Do this early, so jumps to "bad" will free the current * option. */ TAILQ_INSERT_TAIL(opts, opt, link); if (auio->uio_segflg == UIO_SYSSPACE) { bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); } else { error = copyin(auio->uio_iov[i].iov_base, opt->name, namelen); if (error) goto bad; } /* Ensure names are null-terminated strings. */ if (namelen == 0 || opt->name[namelen - 1] != '\0') { error = EINVAL; goto bad; } if (optlen != 0) { opt->len = optlen; opt->value = malloc(optlen, M_MOUNT, M_WAITOK); if (auio->uio_segflg == UIO_SYSSPACE) { bcopy(auio->uio_iov[i + 1].iov_base, opt->value, optlen); } else { error = copyin(auio->uio_iov[i + 1].iov_base, opt->value, optlen); if (error) goto bad; } } } vfs_sanitizeopts(opts); *options = opts; return (0); bad: vfs_freeopts(opts); return (error); } /* * Merge the old mount options with the new ones passed * in the MNT_UPDATE case. * * XXX: This function will keep a "nofoo" option in the new * options. E.g, if the option's canonical name is "foo", * "nofoo" ends up in the mount point's active options. */ static void vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts) { struct vfsopt *opt, *new; TAILQ_FOREACH(opt, oldopts, link) { new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); new->name = strdup(opt->name, M_MOUNT); if (opt->len != 0) { new->value = malloc(opt->len, M_MOUNT, M_WAITOK); bcopy(opt->value, new->value, opt->len); } else new->value = NULL; new->len = opt->len; new->seen = opt->seen; TAILQ_INSERT_HEAD(toopts, new, link); } vfs_sanitizeopts(toopts); } /* * Mount a filesystem. */ int sys_nmount(td, uap) struct thread *td; struct nmount_args /* { struct iovec *iovp; unsigned int iovcnt; int flags; } */ *uap; { struct uio *auio; int error; u_int iovcnt; uint64_t flags; /* * Mount flags are now 64-bits. On 32-bit archtectures only * 32-bits are passed in, but from here on everything handles * 64-bit flags correctly. */ flags = uap->flags; AUDIT_ARG_FFLAGS(flags); CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__, uap->iovp, uap->iovcnt, flags); /* * Filter out MNT_ROOTFS. We do not want clients of nmount() in * userspace to set this flag, but we must filter it out if we want * MNT_UPDATE on the root file system to work. * MNT_ROOTFS should only be set by the kernel when mounting its * root file system. */ flags &= ~MNT_ROOTFS; iovcnt = uap->iovcnt; /* * Check that we have an even number of iovec's * and that we have at least two options. */ if ((iovcnt & 1) || (iovcnt < 4)) { CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__, uap->iovcnt); return (EINVAL); } error = copyinuio(uap->iovp, iovcnt, &auio); if (error) { CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno", __func__, error); return (error); } error = vfs_donmount(td, flags, auio); free(auio, M_IOV); return (error); } /* * --------------------------------------------------------------------- * Various utility functions */ void vfs_ref(struct mount *mp) { CTR2(KTR_VFS, "%s: mp %p", __func__, mp); MNT_ILOCK(mp); MNT_REF(mp); MNT_IUNLOCK(mp); } void vfs_rel(struct mount *mp) { CTR2(KTR_VFS, "%s: mp %p", __func__, mp); MNT_ILOCK(mp); MNT_REL(mp); MNT_IUNLOCK(mp); } /* * Allocate and initialize the mount point struct. */ struct mount * vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath, struct ucred *cred) { struct mount *mp; mp = uma_zalloc(mount_zone, M_WAITOK); bzero(&mp->mnt_startzero, __rangeof(struct mount, mnt_startzero, mnt_endzero)); TAILQ_INIT(&mp->mnt_nvnodelist); mp->mnt_nvnodelistsize = 0; TAILQ_INIT(&mp->mnt_activevnodelist); mp->mnt_activevnodelistsize = 0; mp->mnt_ref = 0; (void) vfs_busy(mp, MBF_NOWAIT); atomic_add_acq_int(&vfsp->vfc_refcount, 1); mp->mnt_op = vfsp->vfc_vfsops; mp->mnt_vfc = vfsp; mp->mnt_stat.f_type = vfsp->vfc_typenum; mp->mnt_gen++; strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); mp->mnt_vnodecovered = vp; mp->mnt_cred = crdup(cred); mp->mnt_stat.f_owner = cred->cr_uid; strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); mp->mnt_iosize_max = DFLTPHYS; #ifdef MAC mac_mount_init(mp); mac_mount_create(cred, mp); #endif arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0); TAILQ_INIT(&mp->mnt_uppers); return (mp); } /* * Destroy the mount struct previously allocated by vfs_mount_alloc(). */ void vfs_mount_destroy(struct mount *mp) { MNT_ILOCK(mp); mp->mnt_kern_flag |= MNTK_REFEXPIRE; if (mp->mnt_kern_flag & MNTK_MWAIT) { mp->mnt_kern_flag &= ~MNTK_MWAIT; wakeup(mp); } while (mp->mnt_ref) msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0); KASSERT(mp->mnt_ref == 0, ("%s: invalid refcount in the drain path @ %s:%d", __func__, __FILE__, __LINE__)); if (mp->mnt_writeopcount != 0) panic("vfs_mount_destroy: nonzero writeopcount"); if (mp->mnt_secondary_writes != 0) panic("vfs_mount_destroy: nonzero secondary_writes"); atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1); if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) { struct vnode *vp; TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) vn_printf(vp, "dangling vnode "); panic("unmount: dangling vnode"); } KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers")); if (mp->mnt_nvnodelistsize != 0) panic("vfs_mount_destroy: nonzero nvnodelistsize"); if (mp->mnt_activevnodelistsize != 0) panic("vfs_mount_destroy: nonzero activevnodelistsize"); if (mp->mnt_lockref != 0) panic("vfs_mount_destroy: nonzero lock refcount"); MNT_IUNLOCK(mp); #ifdef MAC mac_mount_destroy(mp); #endif if (mp->mnt_opt != NULL) vfs_freeopts(mp->mnt_opt); crfree(mp->mnt_cred); uma_zfree(mount_zone, mp); } int vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions) { struct vfsoptlist *optlist; struct vfsopt *opt, *tmp_opt; char *fstype, *fspath, *errmsg; int error, fstypelen, fspathlen, errmsg_len, errmsg_pos; errmsg = fspath = NULL; errmsg_len = fspathlen = 0; errmsg_pos = -1; error = vfs_buildopts(fsoptions, &optlist); if (error) return (error); if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0) errmsg_pos = vfs_getopt_pos(optlist, "errmsg"); /* * We need these two options before the others, * and they are mandatory for any filesystem. * Ensure they are NUL terminated as well. */ fstypelen = 0; error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); if (error || fstype[fstypelen - 1] != '\0') { error = EINVAL; if (errmsg != NULL) strncpy(errmsg, "Invalid fstype", errmsg_len); goto bail; } fspathlen = 0; error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); if (error || fspath[fspathlen - 1] != '\0') { error = EINVAL; if (errmsg != NULL) strncpy(errmsg, "Invalid fspath", errmsg_len); goto bail; } /* * We need to see if we have the "update" option * before we call vfs_domount(), since vfs_domount() has special * logic based on MNT_UPDATE. This is very important * when we want to update the root filesystem. */ TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) { if (strcmp(opt->name, "update") == 0) { fsflags |= MNT_UPDATE; vfs_freeopt(optlist, opt); } else if (strcmp(opt->name, "async") == 0) fsflags |= MNT_ASYNC; else if (strcmp(opt->name, "force") == 0) { fsflags |= MNT_FORCE; vfs_freeopt(optlist, opt); } else if (strcmp(opt->name, "reload") == 0) { fsflags |= MNT_RELOAD; vfs_freeopt(optlist, opt); } else if (strcmp(opt->name, "multilabel") == 0) fsflags |= MNT_MULTILABEL; else if (strcmp(opt->name, "noasync") == 0) fsflags &= ~MNT_ASYNC; else if (strcmp(opt->name, "noatime") == 0) fsflags |= MNT_NOATIME; else if (strcmp(opt->name, "atime") == 0) { free(opt->name, M_MOUNT); opt->name = strdup("nonoatime", M_MOUNT); } else if (strcmp(opt->name, "noclusterr") == 0) fsflags |= MNT_NOCLUSTERR; else if (strcmp(opt->name, "clusterr") == 0) { free(opt->name, M_MOUNT); opt->name = strdup("nonoclusterr", M_MOUNT); } else if (strcmp(opt->name, "noclusterw") == 0) fsflags |= MNT_NOCLUSTERW; else if (strcmp(opt->name, "clusterw") == 0) { free(opt->name, M_MOUNT); opt->name = strdup("nonoclusterw", M_MOUNT); } else if (strcmp(opt->name, "noexec") == 0) fsflags |= MNT_NOEXEC; else if (strcmp(opt->name, "exec") == 0) { free(opt->name, M_MOUNT); opt->name = strdup("nonoexec", M_MOUNT); } else if (strcmp(opt->name, "nosuid") == 0) fsflags |= MNT_NOSUID; else if (strcmp(opt->name, "suid") == 0) { free(opt->name, M_MOUNT); opt->name = strdup("nonosuid", M_MOUNT); } else if (strcmp(opt->name, "nosymfollow") == 0) fsflags |= MNT_NOSYMFOLLOW; else if (strcmp(opt->name, "symfollow") == 0) { free(opt->name, M_MOUNT); opt->name = strdup("nonosymfollow", M_MOUNT); } else if (strcmp(opt->name, "noro") == 0) fsflags &= ~MNT_RDONLY; else if (strcmp(opt->name, "rw") == 0) fsflags &= ~MNT_RDONLY; else if (strcmp(opt->name, "ro") == 0) fsflags |= MNT_RDONLY; else if (strcmp(opt->name, "rdonly") == 0) { free(opt->name, M_MOUNT); opt->name = strdup("ro", M_MOUNT); fsflags |= MNT_RDONLY; } else if (strcmp(opt->name, "suiddir") == 0) fsflags |= MNT_SUIDDIR; else if (strcmp(opt->name, "sync") == 0) fsflags |= MNT_SYNCHRONOUS; else if (strcmp(opt->name, "union") == 0) fsflags |= MNT_UNION; else if (strcmp(opt->name, "automounted") == 0) { fsflags |= MNT_AUTOMOUNTED; vfs_freeopt(optlist, opt); } } /* * Be ultra-paranoid about making sure the type and fspath * variables will fit in our mp buffers, including the * terminating NUL. */ if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) { error = ENAMETOOLONG; goto bail; } error = vfs_domount(td, fstype, fspath, fsflags, &optlist); bail: /* copyout the errmsg */ if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) && errmsg_len > 0 && errmsg != NULL) { if (fsoptions->uio_segflg == UIO_SYSSPACE) { bcopy(errmsg, fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); } else { copyout(errmsg, fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); } } if (optlist != NULL) vfs_freeopts(optlist); return (error); } /* * Old mount API. */ #ifndef _SYS_SYSPROTO_H_ struct mount_args { char *type; char *path; int flags; caddr_t data; }; #endif /* ARGSUSED */ int sys_mount(td, uap) struct thread *td; struct mount_args /* { char *type; char *path; int flags; caddr_t data; } */ *uap; { char *fstype; struct vfsconf *vfsp = NULL; struct mntarg *ma = NULL; uint64_t flags; int error; /* * Mount flags are now 64-bits. On 32-bit architectures only * 32-bits are passed in, but from here on everything handles * 64-bit flags correctly. */ flags = uap->flags; AUDIT_ARG_FFLAGS(flags); /* * Filter out MNT_ROOTFS. We do not want clients of mount() in * userspace to set this flag, but we must filter it out if we want * MNT_UPDATE on the root file system to work. * MNT_ROOTFS should only be set by the kernel when mounting its * root file system. */ flags &= ~MNT_ROOTFS; fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); if (error) { free(fstype, M_TEMP); return (error); } AUDIT_ARG_TEXT(fstype); vfsp = vfs_byname_kld(fstype, td, &error); free(fstype, M_TEMP); if (vfsp == NULL) return (ENOENT); if (vfsp->vfc_vfsops->vfs_cmount == NULL) return (EOPNOTSUPP); ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN); ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); ma = mount_argb(ma, flags & MNT_RDONLY, "noro"); ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid"); ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec"); error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags); return (error); } /* * vfs_domount_first(): first file system mount (not update) */ static int vfs_domount_first( struct thread *td, /* Calling thread. */ struct vfsconf *vfsp, /* File system type. */ char *fspath, /* Mount path. */ struct vnode *vp, /* Vnode to be covered. */ uint64_t fsflags, /* Flags common to all filesystems. */ struct vfsoptlist **optlist /* Options local to the filesystem. */ ) { struct vattr va; struct mount *mp; struct vnode *newdp; int error; ASSERT_VOP_ELOCKED(vp, __func__); KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here")); /* * If the user is not root, ensure that they own the directory * onto which we are attempting to mount. */ error = VOP_GETATTR(vp, &va, td->td_ucred); if (error == 0 && va.va_uid != td->td_ucred->cr_uid) error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 0); if (error == 0) error = vinvalbuf(vp, V_SAVE, 0, 0); if (error == 0 && vp->v_type != VDIR) error = ENOTDIR; if (error == 0) { VI_LOCK(vp); if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL) vp->v_iflag |= VI_MOUNT; else error = EBUSY; VI_UNLOCK(vp); } if (error != 0) { vput(vp); return (error); } VOP_UNLOCK(vp, 0); /* Allocate and initialize the filesystem. */ mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred); /* XXXMAC: pass to vfs_mount_alloc? */ mp->mnt_optnew = *optlist; /* Set the mount level flags. */ mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY)); /* * Mount the filesystem. * XXX The final recipients of VFS_MOUNT just overwrite the ndp they * get. No freeing of cn_pnbuf. */ error = VFS_MOUNT(mp); if (error != 0) { vfs_unbusy(mp); vfs_mount_destroy(mp); VI_LOCK(vp); vp->v_iflag &= ~VI_MOUNT; VI_UNLOCK(vp); vrele(vp); return (error); } if (mp->mnt_opt != NULL) vfs_freeopts(mp->mnt_opt); mp->mnt_opt = mp->mnt_optnew; *optlist = NULL; (void)VFS_STATFS(mp, &mp->mnt_stat); /* * Prevent external consumers of mount options from reading mnt_optnew. */ mp->mnt_optnew = NULL; MNT_ILOCK(mp); if ((mp->mnt_flag & MNT_ASYNC) != 0 && (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) mp->mnt_kern_flag |= MNTK_ASYNC; else mp->mnt_kern_flag &= ~MNTK_ASYNC; MNT_IUNLOCK(mp); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); cache_purge(vp); VI_LOCK(vp); vp->v_iflag &= ~VI_MOUNT; VI_UNLOCK(vp); vp->v_mountedhere = mp; /* Place the new filesystem at the end of the mount list. */ mtx_lock(&mountlist_mtx); TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); mtx_unlock(&mountlist_mtx); vfs_event_signal(NULL, VQ_MOUNT, 0); if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) panic("mount: lost mount"); VOP_UNLOCK(vp, 0); EVENTHANDLER_INVOKE(vfs_mounted, mp, newdp, td); VOP_UNLOCK(newdp, 0); mountcheckdirs(vp, newdp); vrele(newdp); if ((mp->mnt_flag & MNT_RDONLY) == 0) vfs_allocate_syncvnode(mp); vfs_unbusy(mp); return (0); } /* * vfs_domount_update(): update of mounted file system */ static int vfs_domount_update( struct thread *td, /* Calling thread. */ struct vnode *vp, /* Mount point vnode. */ uint64_t fsflags, /* Flags common to all filesystems. */ struct vfsoptlist **optlist /* Options local to the filesystem. */ ) { struct export_args export; void *bufp; struct mount *mp; int error, export_error, len; uint64_t flag; ASSERT_VOP_ELOCKED(vp, __func__); KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here")); mp = vp->v_mount; if ((vp->v_vflag & VV_ROOT) == 0) { if (vfs_copyopt(*optlist, "export", &export, sizeof(export)) == 0) error = EXDEV; else error = EINVAL; vput(vp); return (error); } /* * We only allow the filesystem to be reloaded if it * is currently mounted read-only. */ flag = mp->mnt_flag; if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) { vput(vp); return (EOPNOTSUPP); /* Needs translation */ } /* * Only privileged root, or (if MNT_USER is set) the user that * did the original mount is permitted to update it. */ error = vfs_suser(mp, td); if (error != 0) { vput(vp); return (error); } if (vfs_busy(mp, MBF_NOWAIT)) { vput(vp); return (EBUSY); } VI_LOCK(vp); if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) { VI_UNLOCK(vp); vfs_unbusy(mp); vput(vp); return (EBUSY); } vp->v_iflag |= VI_MOUNT; VI_UNLOCK(vp); VOP_UNLOCK(vp, 0); MNT_ILOCK(mp); mp->mnt_flag &= ~MNT_UPDATEMASK; mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY); if ((mp->mnt_flag & MNT_ASYNC) == 0) mp->mnt_kern_flag &= ~MNTK_ASYNC; MNT_IUNLOCK(mp); mp->mnt_optnew = *optlist; vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); /* * Mount the filesystem. * XXX The final recipients of VFS_MOUNT just overwrite the ndp they * get. No freeing of cn_pnbuf. */ error = VFS_MOUNT(mp); export_error = 0; /* Process the export option. */ if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp, &len) == 0) { /* Assume that there is only 1 ABI for each length. */ switch (len) { case (sizeof(struct oexport_args)): bzero(&export, sizeof(export)); /* FALLTHROUGH */ case (sizeof(export)): bcopy(bufp, &export, len); export_error = vfs_export(mp, &export); break; default: export_error = EINVAL; break; } } MNT_ILOCK(mp); if (error == 0) { mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | MNT_SNAPSHOT); } else { /* * If we fail, restore old mount flags. MNT_QUOTA is special, * because it is not part of MNT_UPDATEMASK, but it could have * changed in the meantime if quotactl(2) was called. * All in all we want current value of MNT_QUOTA, not the old * one. */ mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA); } if ((mp->mnt_flag & MNT_ASYNC) != 0 && (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) mp->mnt_kern_flag |= MNTK_ASYNC; else mp->mnt_kern_flag &= ~MNTK_ASYNC; MNT_IUNLOCK(mp); if (error != 0) goto end; if (mp->mnt_opt != NULL) vfs_freeopts(mp->mnt_opt); mp->mnt_opt = mp->mnt_optnew; *optlist = NULL; (void)VFS_STATFS(mp, &mp->mnt_stat); /* * Prevent external consumers of mount options from reading * mnt_optnew. */ mp->mnt_optnew = NULL; if ((mp->mnt_flag & MNT_RDONLY) == 0) vfs_allocate_syncvnode(mp); else vfs_deallocate_syncvnode(mp); end: vfs_unbusy(mp); VI_LOCK(vp); vp->v_iflag &= ~VI_MOUNT; VI_UNLOCK(vp); vrele(vp); return (error != 0 ? error : export_error); } /* * vfs_domount(): actually attempt a filesystem mount. */ static int vfs_domount( struct thread *td, /* Calling thread. */ const char *fstype, /* Filesystem type. */ char *fspath, /* Mount path. */ uint64_t fsflags, /* Flags common to all filesystems. */ struct vfsoptlist **optlist /* Options local to the filesystem. */ ) { struct vfsconf *vfsp; struct nameidata nd; struct vnode *vp; char *pathbuf; int error; /* * Be ultra-paranoid about making sure the type and fspath * variables will fit in our mp buffers, including the * terminating NUL. */ if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) return (ENAMETOOLONG); if (jailed(td->td_ucred) || usermount == 0) { if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) return (error); } /* * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. */ if (fsflags & MNT_EXPORTED) { error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); if (error) return (error); } if (fsflags & MNT_SUIDDIR) { error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); if (error) return (error); } /* * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. */ if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) fsflags |= MNT_NOSUID | MNT_USER; } /* Load KLDs before we lock the covered vnode to avoid reversals. */ vfsp = NULL; if ((fsflags & MNT_UPDATE) == 0) { /* Don't try to load KLDs if we're mounting the root. */ if (fsflags & MNT_ROOTFS) vfsp = vfs_byname(fstype); else vfsp = vfs_byname_kld(fstype, td, &error); if (vfsp == NULL) return (ENODEV); if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL)) return (EPERM); } /* * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE. */ NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE, fspath, td); error = namei(&nd); if (error != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); vp = nd.ni_vp; if ((fsflags & MNT_UPDATE) == 0) { pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); strcpy(pathbuf, fspath); error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN); /* debug.disablefullpath == 1 results in ENODEV */ if (error == 0 || error == ENODEV) { error = vfs_domount_first(td, vfsp, pathbuf, vp, fsflags, optlist); } free(pathbuf, M_TEMP); } else error = vfs_domount_update(td, vp, fsflags, optlist); return (error); } /* * Unmount a filesystem. * * Note: unmount takes a path to the vnode mounted on as argument, not * special file (as before). */ #ifndef _SYS_SYSPROTO_H_ struct unmount_args { char *path; int flags; }; #endif /* ARGSUSED */ int sys_unmount(struct thread *td, struct unmount_args *uap) { struct nameidata nd; struct mount *mp; char *pathbuf; int error, id0, id1; AUDIT_ARG_VALUE(uap->flags); if (jailed(td->td_ucred) || usermount == 0) { error = priv_check(td, PRIV_VFS_UNMOUNT); if (error) return (error); } pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); if (error) { free(pathbuf, M_TEMP); return (error); } if (uap->flags & MNT_BYFSID) { AUDIT_ARG_TEXT(pathbuf); /* Decode the filesystem ID. */ if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { free(pathbuf, M_TEMP); return (EINVAL); } mtx_lock(&mountlist_mtx); TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { if (mp->mnt_stat.f_fsid.val[0] == id0 && mp->mnt_stat.f_fsid.val[1] == id1) { vfs_ref(mp); break; } } mtx_unlock(&mountlist_mtx); } else { /* * Try to find global path for path argument. */ NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE, pathbuf, td); if (namei(&nd) == 0) { NDFREE(&nd, NDF_ONLY_PNBUF); error = vn_path_to_global_path(td, nd.ni_vp, pathbuf, MNAMELEN); if (error == 0 || error == ENODEV) vput(nd.ni_vp); } mtx_lock(&mountlist_mtx); TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) { vfs_ref(mp); break; } } mtx_unlock(&mountlist_mtx); } free(pathbuf, M_TEMP); if (mp == NULL) { /* * Previously we returned ENOENT for a nonexistent path and * EINVAL for a non-mountpoint. We cannot tell these apart * now, so in the !MNT_BYFSID case return the more likely * EINVAL for compatibility. */ return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); } /* * Don't allow unmounting the root filesystem. */ if (mp->mnt_flag & MNT_ROOTFS) { vfs_rel(mp); return (EINVAL); } error = dounmount(mp, uap->flags, td); return (error); } /* * Return error if any of the vnodes, ignoring the root vnode * and the syncer vnode, have non-zero usecount. */ static int vfs_check_usecounts(struct mount *mp) { struct vnode *vp, *mvp; MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON && vp->v_usecount != 0) { VI_UNLOCK(vp); MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); return (EBUSY); } VI_UNLOCK(vp); } return (0); } /* * Do the actual filesystem unmount. */ int dounmount(struct mount *mp, int flags, struct thread *td) { struct vnode *coveredvp, *fsrootvp; int error; uint64_t async_flag; int mnt_gen_r; if ((coveredvp = mp->mnt_vnodecovered) != NULL) { mnt_gen_r = mp->mnt_gen; VI_LOCK(coveredvp); vholdl(coveredvp); vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); /* * Check for mp being unmounted while waiting for the * covered vnode lock. */ if (coveredvp->v_mountedhere != mp || coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { VOP_UNLOCK(coveredvp, 0); vdrop(coveredvp); vfs_rel(mp); return (EBUSY); } } /* * Only privileged root, or (if MNT_USER is set) the user that did the * original mount is permitted to unmount this filesystem. */ error = vfs_suser(mp, td); if (error != 0) { if (coveredvp != NULL) { VOP_UNLOCK(coveredvp, 0); vdrop(coveredvp); } vfs_rel(mp); return (error); } vn_start_write(NULL, &mp, V_WAIT | V_MNTREF); MNT_ILOCK(mp); if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 || !TAILQ_EMPTY(&mp->mnt_uppers)) { MNT_IUNLOCK(mp); if (coveredvp != NULL) { VOP_UNLOCK(coveredvp, 0); vdrop(coveredvp); } vn_finished_write(mp); return (EBUSY); } mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ; if (flags & MNT_NONBUSY) { MNT_IUNLOCK(mp); error = vfs_check_usecounts(mp); MNT_ILOCK(mp); if (error != 0) { mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ); MNT_IUNLOCK(mp); if (coveredvp != NULL) { VOP_UNLOCK(coveredvp, 0); vdrop(coveredvp); } vn_finished_write(mp); return (error); } } /* Allow filesystems to detect that a forced unmount is in progress. */ if (flags & MNT_FORCE) { mp->mnt_kern_flag |= MNTK_UNMOUNTF; MNT_IUNLOCK(mp); /* * Must be done after setting MNTK_UNMOUNTF and before * waiting for mnt_lockref to become 0. */ VFS_PURGE(mp); MNT_ILOCK(mp); } error = 0; if (mp->mnt_lockref) { mp->mnt_kern_flag |= MNTK_DRAINING; error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS, "mount drain", 0); } MNT_IUNLOCK(mp); KASSERT(mp->mnt_lockref == 0, ("%s: invalid lock refcount in the drain path @ %s:%d", __func__, __FILE__, __LINE__)); KASSERT(error == 0, ("%s: invalid return value for msleep in the drain path @ %s:%d", __func__, __FILE__, __LINE__)); if (mp->mnt_flag & MNT_EXPUBLIC) vfs_setpublicfs(NULL, NULL, NULL); /* * From now, we can claim that the use reference on the * coveredvp is ours, and the ref can be released only by * successfull unmount by us, or left for later unmount * attempt. The previously acquired hold reference is no * longer needed to protect the vnode from reuse. */ if (coveredvp != NULL) vdrop(coveredvp); vfs_msync(mp, MNT_WAIT); MNT_ILOCK(mp); async_flag = mp->mnt_flag & MNT_ASYNC; mp->mnt_flag &= ~MNT_ASYNC; mp->mnt_kern_flag &= ~MNTK_ASYNC; MNT_IUNLOCK(mp); cache_purgevfs(mp); /* remove cache entries for this file sys */ vfs_deallocate_syncvnode(mp); /* * For forced unmounts, move process cdir/rdir refs on the fs root * vnode to the covered vnode. For non-forced unmounts we want * such references to cause an EBUSY error. */ if ((flags & MNT_FORCE) && VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { if (mp->mnt_vnodecovered != NULL && (mp->mnt_flag & MNT_IGNORE) == 0) mountcheckdirs(fsrootvp, mp->mnt_vnodecovered); if (fsrootvp == rootvnode) { vrele(rootvnode); rootvnode = NULL; } vput(fsrootvp); } if ((mp->mnt_flag & MNT_RDONLY) != 0 || (flags & MNT_FORCE) != 0 || (error = VFS_SYNC(mp, MNT_WAIT)) == 0) error = VFS_UNMOUNT(mp, flags); vn_finished_write(mp); /* * If we failed to flush the dirty blocks for this mount point, * undo all the cdir/rdir and rootvnode changes we made above. * Unless we failed to do so because the device is reporting that * it doesn't exist anymore. */ if (error && error != ENXIO) { if ((flags & MNT_FORCE) && VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { if (mp->mnt_vnodecovered != NULL && (mp->mnt_flag & MNT_IGNORE) == 0) mountcheckdirs(mp->mnt_vnodecovered, fsrootvp); if (rootvnode == NULL) { rootvnode = fsrootvp; vref(rootvnode); } vput(fsrootvp); } MNT_ILOCK(mp); mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ; if ((mp->mnt_flag & MNT_RDONLY) == 0) { MNT_IUNLOCK(mp); vfs_allocate_syncvnode(mp); MNT_ILOCK(mp); } mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); mp->mnt_flag |= async_flag; if ((mp->mnt_flag & MNT_ASYNC) != 0 && (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) mp->mnt_kern_flag |= MNTK_ASYNC; if (mp->mnt_kern_flag & MNTK_MWAIT) { mp->mnt_kern_flag &= ~MNTK_MWAIT; wakeup(mp); } MNT_IUNLOCK(mp); if (coveredvp) VOP_UNLOCK(coveredvp, 0); return (error); } mtx_lock(&mountlist_mtx); TAILQ_REMOVE(&mountlist, mp, mnt_list); mtx_unlock(&mountlist_mtx); EVENTHANDLER_INVOKE(vfs_unmounted, mp, td); if (coveredvp != NULL) { coveredvp->v_mountedhere = NULL; vput(coveredvp); } vfs_event_signal(NULL, VQ_UNMOUNT, 0); if (mp == rootdevmp) rootdevmp = NULL; vfs_mount_destroy(mp); return (0); } /* * Report errors during filesystem mounting. */ void vfs_mount_error(struct mount *mp, const char *fmt, ...) { struct vfsoptlist *moptlist = mp->mnt_optnew; va_list ap; int error, len; char *errmsg; error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); if (error || errmsg == NULL || len <= 0) return; va_start(ap, fmt); vsnprintf(errmsg, (size_t)len, fmt, ap); va_end(ap); } void vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...) { va_list ap; int error, len; char *errmsg; error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len); if (error || errmsg == NULL || len <= 0) return; va_start(ap, fmt); vsnprintf(errmsg, (size_t)len, fmt, ap); va_end(ap); } /* * --------------------------------------------------------------------- * Functions for querying mount options/arguments from filesystems. */ /* * Check that no unknown options are given */ int vfs_filteropt(struct vfsoptlist *opts, const char **legal) { struct vfsopt *opt; char errmsg[255]; const char **t, *p, *q; int ret = 0; TAILQ_FOREACH(opt, opts, link) { p = opt->name; q = NULL; if (p[0] == 'n' && p[1] == 'o') q = p + 2; for(t = global_opts; *t != NULL; t++) { if (strcmp(*t, p) == 0) break; if (q != NULL) { if (strcmp(*t, q) == 0) break; } } if (*t != NULL) continue; for(t = legal; *t != NULL; t++) { if (strcmp(*t, p) == 0) break; if (q != NULL) { if (strcmp(*t, q) == 0) break; } } if (*t != NULL) continue; snprintf(errmsg, sizeof(errmsg), "mount option <%s> is unknown", p); ret = EINVAL; } if (ret != 0) { TAILQ_FOREACH(opt, opts, link) { if (strcmp(opt->name, "errmsg") == 0) { strncpy((char *)opt->value, errmsg, opt->len); break; } } if (opt == NULL) printf("%s\n", errmsg); } return (ret); } /* * Get a mount option by its name. * * Return 0 if the option was found, ENOENT otherwise. * If len is non-NULL it will be filled with the length * of the option. If buf is non-NULL, it will be filled * with the address of the option. */ int vfs_getopt(opts, name, buf, len) struct vfsoptlist *opts; const char *name; void **buf; int *len; { struct vfsopt *opt; KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); TAILQ_FOREACH(opt, opts, link) { if (strcmp(name, opt->name) == 0) { opt->seen = 1; if (len != NULL) *len = opt->len; if (buf != NULL) *buf = opt->value; return (0); } } return (ENOENT); } int vfs_getopt_pos(struct vfsoptlist *opts, const char *name) { struct vfsopt *opt; if (opts == NULL) return (-1); TAILQ_FOREACH(opt, opts, link) { if (strcmp(name, opt->name) == 0) { opt->seen = 1; return (opt->pos); } } return (-1); } int vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value) { char *opt_value, *vtp; quad_t iv; int error, opt_len; error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len); if (error != 0) return (error); if (opt_len == 0 || opt_value == NULL) return (EINVAL); if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0') return (EINVAL); iv = strtoq(opt_value, &vtp, 0); if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0')) return (EINVAL); if (iv < 0) return (EINVAL); switch (vtp[0]) { case 't': case 'T': iv *= 1024; case 'g': case 'G': iv *= 1024; case 'm': case 'M': iv *= 1024; case 'k': case 'K': iv *= 1024; case '\0': break; default: return (EINVAL); } *value = iv; return (0); } char * vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) { struct vfsopt *opt; *error = 0; TAILQ_FOREACH(opt, opts, link) { if (strcmp(name, opt->name) != 0) continue; opt->seen = 1; if (opt->len == 0 || ((char *)opt->value)[opt->len - 1] != '\0') { *error = EINVAL; return (NULL); } return (opt->value); } *error = ENOENT; return (NULL); } int vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w, uint64_t val) { struct vfsopt *opt; TAILQ_FOREACH(opt, opts, link) { if (strcmp(name, opt->name) == 0) { opt->seen = 1; if (w != NULL) *w |= val; return (1); } } if (w != NULL) *w &= ~val; return (0); } int vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) { va_list ap; struct vfsopt *opt; int ret; KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); TAILQ_FOREACH(opt, opts, link) { if (strcmp(name, opt->name) != 0) continue; opt->seen = 1; if (opt->len == 0 || opt->value == NULL) return (0); if (((char *)opt->value)[opt->len - 1] != '\0') return (0); va_start(ap, fmt); ret = vsscanf(opt->value, fmt, ap); va_end(ap); return (ret); } return (0); } int vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len) { struct vfsopt *opt; TAILQ_FOREACH(opt, opts, link) { if (strcmp(name, opt->name) != 0) continue; opt->seen = 1; if (opt->value == NULL) opt->len = len; else { if (opt->len != len) return (EINVAL); bcopy(value, opt->value, len); } return (0); } return (ENOENT); } int vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len) { struct vfsopt *opt; TAILQ_FOREACH(opt, opts, link) { if (strcmp(name, opt->name) != 0) continue; opt->seen = 1; if (opt->value == NULL) opt->len = len; else { if (opt->len < len) return (EINVAL); opt->len = len; bcopy(value, opt->value, len); } return (0); } return (ENOENT); } int vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value) { struct vfsopt *opt; TAILQ_FOREACH(opt, opts, link) { if (strcmp(name, opt->name) != 0) continue; opt->seen = 1; if (opt->value == NULL) opt->len = strlen(value) + 1; else if (strlcpy(opt->value, value, opt->len) >= opt->len) return (EINVAL); return (0); } return (ENOENT); } /* * Find and copy a mount option. * * The size of the buffer has to be specified * in len, if it is not the same length as the * mount option, EINVAL is returned. * Returns ENOENT if the option is not found. */ int vfs_copyopt(opts, name, dest, len) struct vfsoptlist *opts; const char *name; void *dest; int len; { struct vfsopt *opt; KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); TAILQ_FOREACH(opt, opts, link) { if (strcmp(name, opt->name) == 0) { opt->seen = 1; if (len != opt->len) return (EINVAL); bcopy(opt->value, dest, opt->len); return (0); } } return (ENOENT); } int __vfs_statfs(struct mount *mp, struct statfs *sbp) { int error; error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat); if (sbp != &mp->mnt_stat) *sbp = mp->mnt_stat; return (error); } void vfs_mountedfrom(struct mount *mp, const char *from) { bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); strlcpy(mp->mnt_stat.f_mntfromname, from, sizeof mp->mnt_stat.f_mntfromname); } /* * --------------------------------------------------------------------- * This is the api for building mount args and mounting filesystems from * inside the kernel. * * The API works by accumulation of individual args. First error is * latched. * * XXX: should be documented in new manpage kernel_mount(9) */ /* A memory allocation which must be freed when we are done */ struct mntaarg { SLIST_ENTRY(mntaarg) next; }; /* The header for the mount arguments */ struct mntarg { struct iovec *v; int len; int error; SLIST_HEAD(, mntaarg) list; }; /* * Add a boolean argument. * * flag is the boolean value. * name must start with "no". */ struct mntarg * mount_argb(struct mntarg *ma, int flag, const char *name) { KASSERT(name[0] == 'n' && name[1] == 'o', ("mount_argb(...,%s): name must start with 'no'", name)); return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); } /* * Add an argument printf style */ struct mntarg * mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) { va_list ap; struct mntaarg *maa; struct sbuf *sb; int len; if (ma == NULL) { ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); SLIST_INIT(&ma->list); } if (ma->error) return (ma); ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), M_MOUNT, M_WAITOK); ma->v[ma->len].iov_base = (void *)(uintptr_t)name; ma->v[ma->len].iov_len = strlen(name) + 1; ma->len++; sb = sbuf_new_auto(); va_start(ap, fmt); sbuf_vprintf(sb, fmt, ap); va_end(ap); sbuf_finish(sb); len = sbuf_len(sb) + 1; maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); SLIST_INSERT_HEAD(&ma->list, maa, next); bcopy(sbuf_data(sb), maa + 1, len); sbuf_delete(sb); ma->v[ma->len].iov_base = maa + 1; ma->v[ma->len].iov_len = len; ma->len++; return (ma); } /* * Add an argument which is a userland string. */ struct mntarg * mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) { struct mntaarg *maa; char *tbuf; if (val == NULL) return (ma); if (ma == NULL) { ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); SLIST_INIT(&ma->list); } if (ma->error) return (ma); maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); SLIST_INSERT_HEAD(&ma->list, maa, next); tbuf = (void *)(maa + 1); ma->error = copyinstr(val, tbuf, len, NULL); return (mount_arg(ma, name, tbuf, -1)); } /* * Plain argument. * * If length is -1, treat value as a C string. */ struct mntarg * mount_arg(struct mntarg *ma, const char *name, const void *val, int len) { if (ma == NULL) { ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); SLIST_INIT(&ma->list); } if (ma->error) return (ma); ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), M_MOUNT, M_WAITOK); ma->v[ma->len].iov_base = (void *)(uintptr_t)name; ma->v[ma->len].iov_len = strlen(name) + 1; ma->len++; ma->v[ma->len].iov_base = (void *)(uintptr_t)val; if (len < 0) ma->v[ma->len].iov_len = strlen(val) + 1; else ma->v[ma->len].iov_len = len; ma->len++; return (ma); } /* * Free a mntarg structure */ static void free_mntarg(struct mntarg *ma) { struct mntaarg *maa; while (!SLIST_EMPTY(&ma->list)) { maa = SLIST_FIRST(&ma->list); SLIST_REMOVE_HEAD(&ma->list, next); free(maa, M_MOUNT); } free(ma->v, M_MOUNT); free(ma, M_MOUNT); } /* * Mount a filesystem */ int kernel_mount(struct mntarg *ma, uint64_t flags) { struct uio auio; int error; KASSERT(ma != NULL, ("kernel_mount NULL ma")); KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); auio.uio_iov = ma->v; auio.uio_iovcnt = ma->len; auio.uio_segflg = UIO_SYSSPACE; error = ma->error; if (!error) error = vfs_donmount(curthread, flags, &auio); free_mntarg(ma); return (error); } /* * A printflike function to mount a filesystem. */ int kernel_vmount(int flags, ...) { struct mntarg *ma = NULL; va_list ap; const char *cp; const void *vp; int error; va_start(ap, flags); for (;;) { cp = va_arg(ap, const char *); if (cp == NULL) break; vp = va_arg(ap, const void *); ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0)); } va_end(ap); error = kernel_mount(ma, flags); return (error); } void vfs_oexport_conv(const struct oexport_args *oexp, struct export_args *exp) { bcopy(oexp, exp, sizeof(*oexp)); exp->ex_numsecflavors = 0; } Index: head/sys/kern/vfs_mountroot.c =================================================================== --- head/sys/kern/vfs_mountroot.c (revision 305831) +++ head/sys/kern/vfs_mountroot.c (revision 305832) @@ -1,1105 +1,1105 @@ /*- * Copyright (c) 2010 Marcel Moolenaar * Copyright (c) 1999-2004 Poul-Henning Kamp * Copyright (c) 1999 Michael Smith * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include "opt_rootdevname.h" #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * The root filesystem is detailed in the kernel environment variable * vfs.root.mountfrom, which is expected to be in the general format * * :[][ :[] ...] * vfsname := the name of a VFS known to the kernel and capable * of being mounted as root * path := disk device name or other data used by the filesystem * to locate its physical store * * If the environment variable vfs.root.mountfrom is a space separated list, * each list element is tried in turn and the root filesystem will be mounted * from the first one that succeeds. * * The environment variable vfs.root.mountfrom.options is a comma delimited * set of string mount options. These mount options must be parseable * by nmount() in the kernel. */ static int parse_mount(char **); static struct mntarg *parse_mountroot_options(struct mntarg *, const char *); static int sysctl_vfs_root_mount_hold(SYSCTL_HANDLER_ARGS); static void vfs_mountroot_wait(void); static int vfs_mountroot_wait_if_neccessary(const char *fs, const char *dev); /* * The vnode of the system's root (/ in the filesystem, without chroot * active.) */ struct vnode *rootvnode; /* * Mount of the system's /dev. */ struct mount *rootdevmp; char *rootdevnames[2] = {NULL, NULL}; struct mtx root_holds_mtx; MTX_SYSINIT(root_holds, &root_holds_mtx, "root_holds", MTX_DEF); struct root_hold_token { const char *who; LIST_ENTRY(root_hold_token) list; }; static LIST_HEAD(, root_hold_token) root_holds = LIST_HEAD_INITIALIZER(root_holds); enum action { A_CONTINUE, A_PANIC, A_REBOOT, A_RETRY }; static enum action root_mount_onfail = A_CONTINUE; static int root_mount_mddev; static int root_mount_complete; /* By default wait up to 3 seconds for devices to appear. */ static int root_mount_timeout = 3; TUNABLE_INT("vfs.mountroot.timeout", &root_mount_timeout); SYSCTL_PROC(_vfs, OID_AUTO, root_mount_hold, CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_vfs_root_mount_hold, "A", "List of root mount hold tokens"); static int sysctl_vfs_root_mount_hold(SYSCTL_HANDLER_ARGS) { struct sbuf sb; struct root_hold_token *h; int error; sbuf_new(&sb, NULL, 256, SBUF_AUTOEXTEND | SBUF_INCLUDENUL); mtx_lock(&root_holds_mtx); LIST_FOREACH(h, &root_holds, list) { if (h != LIST_FIRST(&root_holds)) sbuf_putc(&sb, ' '); sbuf_printf(&sb, "%s", h->who); } mtx_unlock(&root_holds_mtx); error = sbuf_finish(&sb); if (error == 0) error = SYSCTL_OUT(req, sbuf_data(&sb), sbuf_len(&sb)); sbuf_delete(&sb); return (error); } struct root_hold_token * root_mount_hold(const char *identifier) { struct root_hold_token *h; if (root_mounted()) return (NULL); h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK); h->who = identifier; mtx_lock(&root_holds_mtx); LIST_INSERT_HEAD(&root_holds, h, list); mtx_unlock(&root_holds_mtx); return (h); } void root_mount_rel(struct root_hold_token *h) { if (h == NULL) return; mtx_lock(&root_holds_mtx); LIST_REMOVE(h, list); wakeup(&root_holds); mtx_unlock(&root_holds_mtx); free(h, M_DEVBUF); } int root_mounted(void) { /* No mutex is acquired here because int stores are atomic. */ return (root_mount_complete); } static void set_rootvnode(void) { struct proc *p; if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode)) panic("Cannot find root vnode"); VOP_UNLOCK(rootvnode, 0); p = curthread->td_proc; FILEDESC_XLOCK(p->p_fd); if (p->p_fd->fd_cdir != NULL) vrele(p->p_fd->fd_cdir); p->p_fd->fd_cdir = rootvnode; VREF(rootvnode); if (p->p_fd->fd_rdir != NULL) vrele(p->p_fd->fd_rdir); p->p_fd->fd_rdir = rootvnode; VREF(rootvnode); FILEDESC_XUNLOCK(p->p_fd); } static int vfs_mountroot_devfs(struct thread *td, struct mount **mpp) { struct vfsoptlist *opts; struct vfsconf *vfsp; struct mount *mp; int error; *mpp = NULL; if (rootdevmp != NULL) { /* * Already have /dev; this happens during rerooting. */ error = vfs_busy(rootdevmp, 0); if (error != 0) return (error); *mpp = rootdevmp; } else { vfsp = vfs_byname("devfs"); KASSERT(vfsp != NULL, ("Could not find devfs by name")); if (vfsp == NULL) return (ENOENT); mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred); error = VFS_MOUNT(mp); KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error)); if (error) return (error); opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); TAILQ_INIT(opts); mp->mnt_opt = opts; mtx_lock(&mountlist_mtx); TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); mtx_unlock(&mountlist_mtx); *mpp = mp; rootdevmp = mp; } set_rootvnode(); error = kern_symlinkat(td, "/", AT_FDCWD, "dev", UIO_SYSSPACE); if (error) printf("kern_symlink /dev -> / returns %d\n", error); return (error); } static void vfs_mountroot_shuffle(struct thread *td, struct mount *mpdevfs) { struct nameidata nd; struct mount *mporoot, *mpnroot; struct vnode *vp, *vporoot, *vpdevfs; char *fspath; int error; mpnroot = TAILQ_NEXT(mpdevfs, mnt_list); /* Shuffle the mountlist. */ mtx_lock(&mountlist_mtx); mporoot = TAILQ_FIRST(&mountlist); TAILQ_REMOVE(&mountlist, mpdevfs, mnt_list); if (mporoot != mpdevfs) { TAILQ_REMOVE(&mountlist, mpnroot, mnt_list); TAILQ_INSERT_HEAD(&mountlist, mpnroot, mnt_list); } TAILQ_INSERT_TAIL(&mountlist, mpdevfs, mnt_list); mtx_unlock(&mountlist_mtx); cache_purgevfs(mporoot); if (mporoot != mpdevfs) cache_purgevfs(mpdevfs); VFS_ROOT(mporoot, LK_EXCLUSIVE, &vporoot); VI_LOCK(vporoot); vporoot->v_iflag &= ~VI_MOUNT; VI_UNLOCK(vporoot); vporoot->v_mountedhere = NULL; mporoot->mnt_flag &= ~MNT_ROOTFS; mporoot->mnt_vnodecovered = NULL; vput(vporoot); /* Set up the new rootvnode, and purge the cache */ mpnroot->mnt_vnodecovered = NULL; set_rootvnode(); cache_purgevfs(rootvnode->v_mount); if (mporoot != mpdevfs) { /* Remount old root under /.mount or /mnt */ fspath = "/.mount"; NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspath, td); error = namei(&nd); if (error) { NDFREE(&nd, NDF_ONLY_PNBUF); fspath = "/mnt"; NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspath, td); error = namei(&nd); } if (!error) { vp = nd.ni_vp; error = (vp->v_type == VDIR) ? 0 : ENOTDIR; if (!error) error = vinvalbuf(vp, V_SAVE, 0, 0); if (!error) { cache_purge(vp); mporoot->mnt_vnodecovered = vp; vp->v_mountedhere = mporoot; strlcpy(mporoot->mnt_stat.f_mntonname, fspath, MNAMELEN); VOP_UNLOCK(vp, 0); } else vput(vp); } NDFREE(&nd, NDF_ONLY_PNBUF); if (error) printf("mountroot: unable to remount previous root " "under /.mount or /mnt (error %d)\n", error); } /* Remount devfs under /dev */ NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td); error = namei(&nd); if (!error) { vp = nd.ni_vp; error = (vp->v_type == VDIR) ? 0 : ENOTDIR; if (!error) error = vinvalbuf(vp, V_SAVE, 0, 0); if (!error) { vpdevfs = mpdevfs->mnt_vnodecovered; if (vpdevfs != NULL) { cache_purge(vpdevfs); vpdevfs->v_mountedhere = NULL; vrele(vpdevfs); } mpdevfs->mnt_vnodecovered = vp; vp->v_mountedhere = mpdevfs; VOP_UNLOCK(vp, 0); } else vput(vp); } if (error) printf("mountroot: unable to remount devfs under /dev " "(error %d)\n", error); NDFREE(&nd, NDF_ONLY_PNBUF); if (mporoot == mpdevfs) { vfs_unbusy(mpdevfs); /* Unlink the no longer needed /dev/dev -> / symlink */ error = kern_unlinkat(td, AT_FDCWD, "/dev/dev", UIO_SYSSPACE, 0); if (error) printf("mountroot: unable to unlink /dev/dev " "(error %d)\n", error); } } /* * Configuration parser. */ /* Parser character classes. */ #define CC_WHITESPACE -1 #define CC_NONWHITESPACE -2 /* Parse errors. */ #define PE_EOF -1 #define PE_EOL -2 static __inline int parse_peek(char **conf) { return (**conf); } static __inline void parse_poke(char **conf, int c) { **conf = c; } static __inline void parse_advance(char **conf) { (*conf)++; } static int parse_skipto(char **conf, int mc) { int c, match; while (1) { c = parse_peek(conf); if (c == 0) return (PE_EOF); switch (mc) { case CC_WHITESPACE: match = (c == ' ' || c == '\t' || c == '\n') ? 1 : 0; break; case CC_NONWHITESPACE: if (c == '\n') return (PE_EOL); match = (c != ' ' && c != '\t') ? 1 : 0; break; default: match = (c == mc) ? 1 : 0; break; } if (match) break; parse_advance(conf); } return (0); } static int parse_token(char **conf, char **tok) { char *p; size_t len; int error; *tok = NULL; error = parse_skipto(conf, CC_NONWHITESPACE); if (error) return (error); p = *conf; error = parse_skipto(conf, CC_WHITESPACE); len = *conf - p; *tok = malloc(len + 1, M_TEMP, M_WAITOK | M_ZERO); bcopy(p, *tok, len); return (0); } static void parse_dir_ask_printenv(const char *var) { char *val; val = kern_getenv(var); if (val != NULL) { printf(" %s=%s\n", var, val); freeenv(val); } } static int parse_dir_ask(char **conf) { char name[80]; char *mnt; int error; vfs_mountroot_wait(); printf("\nLoader variables:\n"); parse_dir_ask_printenv("vfs.root.mountfrom"); parse_dir_ask_printenv("vfs.root.mountfrom.options"); printf("\nManual root filesystem specification:\n"); printf(" : [options]\n"); printf(" Mount using filesystem \n"); printf(" and with the specified (optional) option list.\n"); printf("\n"); printf(" eg. ufs:/dev/da0s1a\n"); printf(" zfs:tank\n"); printf(" cd9660:/dev/cd0 ro\n"); printf(" (which is equivalent to: "); printf("mount -t cd9660 -o ro /dev/cd0 /)\n"); printf("\n"); printf(" ? List valid disk boot devices\n"); printf(" . Yield 1 second (for background tasks)\n"); printf(" Abort manual input\n"); do { error = EINVAL; printf("\nmountroot> "); cngets(name, sizeof(name), GETS_ECHO); if (name[0] == '\0') break; if (name[0] == '?' && name[1] == '\0') { printf("\nList of GEOM managed disk devices:\n "); g_dev_print(); continue; } if (name[0] == '.' && name[1] == '\0') { pause("rmask", hz); continue; } mnt = name; error = parse_mount(&mnt); if (error == -1) printf("Invalid file system specification.\n"); } while (error != 0); return (error); } static int parse_dir_md(char **conf) { struct stat sb; struct thread *td; struct md_ioctl *mdio; char *path, *tok; int error, fd, len; td = curthread; error = parse_token(conf, &tok); if (error) return (error); len = strlen(tok); mdio = malloc(sizeof(*mdio) + len + 1, M_TEMP, M_WAITOK | M_ZERO); path = (void *)(mdio + 1); bcopy(tok, path, len); free(tok, M_TEMP); /* Get file status. */ error = kern_statat(td, 0, AT_FDCWD, path, UIO_SYSSPACE, &sb, NULL); if (error) goto out; /* Open /dev/mdctl so that we can attach/detach. */ error = kern_openat(td, AT_FDCWD, "/dev/" MDCTL_NAME, UIO_SYSSPACE, O_RDWR, 0); if (error) goto out; fd = td->td_retval[0]; mdio->md_version = MDIOVERSION; mdio->md_type = MD_VNODE; if (root_mount_mddev != -1) { mdio->md_unit = root_mount_mddev; DROP_GIANT(); error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio); PICKUP_GIANT(); /* Ignore errors. We don't care. */ root_mount_mddev = -1; } mdio->md_file = (void *)(mdio + 1); mdio->md_options = MD_AUTOUNIT | MD_READONLY; mdio->md_mediasize = sb.st_size; mdio->md_unit = 0; DROP_GIANT(); error = kern_ioctl(td, fd, MDIOCATTACH, (void *)mdio); PICKUP_GIANT(); if (error) goto out; if (mdio->md_unit > 9) { printf("rootmount: too many md units\n"); mdio->md_file = NULL; mdio->md_options = 0; mdio->md_mediasize = 0; DROP_GIANT(); error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio); PICKUP_GIANT(); /* Ignore errors. We don't care. */ error = ERANGE; goto out; } root_mount_mddev = mdio->md_unit; printf(MD_NAME "%u attached to %s\n", root_mount_mddev, mdio->md_file); error = kern_close(td, fd); out: free(mdio, M_TEMP); return (error); } static int parse_dir_onfail(char **conf) { char *action; int error; error = parse_token(conf, &action); if (error) return (error); if (!strcmp(action, "continue")) root_mount_onfail = A_CONTINUE; else if (!strcmp(action, "panic")) root_mount_onfail = A_PANIC; else if (!strcmp(action, "reboot")) root_mount_onfail = A_REBOOT; else if (!strcmp(action, "retry")) root_mount_onfail = A_RETRY; else { printf("rootmount: %s: unknown action\n", action); error = EINVAL; } free(action, M_TEMP); return (0); } static int parse_dir_timeout(char **conf) { char *tok, *endtok; long secs; int error; error = parse_token(conf, &tok); if (error) return (error); secs = strtol(tok, &endtok, 0); error = (secs < 0 || *endtok != '\0') ? EINVAL : 0; if (!error) root_mount_timeout = secs; free(tok, M_TEMP); return (error); } static int parse_directive(char **conf) { char *dir; int error; error = parse_token(conf, &dir); if (error) return (error); if (strcmp(dir, ".ask") == 0) error = parse_dir_ask(conf); else if (strcmp(dir, ".md") == 0) error = parse_dir_md(conf); else if (strcmp(dir, ".onfail") == 0) error = parse_dir_onfail(conf); else if (strcmp(dir, ".timeout") == 0) error = parse_dir_timeout(conf); else { printf("mountroot: invalid directive `%s'\n", dir); /* Ignore the rest of the line. */ (void)parse_skipto(conf, '\n'); error = EINVAL; } free(dir, M_TEMP); return (error); } static int parse_mount_dev_present(const char *dev) { struct nameidata nd; int error; NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, dev, curthread); error = namei(&nd); if (!error) vput(nd.ni_vp); NDFREE(&nd, NDF_ONLY_PNBUF); return (error != 0) ? 0 : 1; } #define ERRMSGL 255 static int parse_mount(char **conf) { char *errmsg; struct mntarg *ma; char *dev, *fs, *opts, *tok; int error; error = parse_token(conf, &tok); if (error) return (error); fs = tok; error = parse_skipto(&tok, ':'); if (error) { free(fs, M_TEMP); return (error); } parse_poke(&tok, '\0'); parse_advance(&tok); dev = tok; if (root_mount_mddev != -1) { /* Handle substitution for the md unit number. */ tok = strstr(dev, "md#"); if (tok != NULL) tok[2] = '0' + root_mount_mddev; } /* Parse options. */ error = parse_token(conf, &tok); opts = (error == 0) ? tok : NULL; printf("Trying to mount root from %s:%s [%s]...\n", fs, dev, (opts != NULL) ? opts : ""); errmsg = malloc(ERRMSGL, M_TEMP, M_WAITOK | M_ZERO); if (vfs_byname(fs) == NULL) { strlcpy(errmsg, "unknown file system", ERRMSGL); error = ENOENT; goto out; } error = vfs_mountroot_wait_if_neccessary(fs, dev); if (error != 0) goto out; ma = NULL; ma = mount_arg(ma, "fstype", fs, -1); ma = mount_arg(ma, "fspath", "/", -1); ma = mount_arg(ma, "from", dev, -1); ma = mount_arg(ma, "errmsg", errmsg, ERRMSGL); ma = mount_arg(ma, "ro", NULL, 0); ma = parse_mountroot_options(ma, opts); error = kernel_mount(ma, MNT_ROOTFS); out: if (error) { printf("Mounting from %s:%s failed with error %d", fs, dev, error); if (errmsg[0] != '\0') printf(": %s", errmsg); printf(".\n"); } free(fs, M_TEMP); free(errmsg, M_TEMP); if (opts != NULL) free(opts, M_TEMP); /* kernel_mount can return -1 on error. */ return ((error < 0) ? EDOOFUS : error); } #undef ERRMSGL static int vfs_mountroot_parse(struct sbuf *sb, struct mount *mpdevfs) { struct mount *mp; char *conf; int error; root_mount_mddev = -1; retry: conf = sbuf_data(sb); mp = TAILQ_NEXT(mpdevfs, mnt_list); error = (mp == NULL) ? 0 : EDOOFUS; root_mount_onfail = A_CONTINUE; while (mp == NULL) { error = parse_skipto(&conf, CC_NONWHITESPACE); if (error == PE_EOL) { parse_advance(&conf); continue; } if (error < 0) break; switch (parse_peek(&conf)) { case '#': error = parse_skipto(&conf, '\n'); break; case '.': error = parse_directive(&conf); break; default: error = parse_mount(&conf); if (error == -1) { printf("mountroot: invalid file system " "specification.\n"); error = 0; } break; } if (error < 0) break; /* Ignore any trailing garbage on the line. */ if (parse_peek(&conf) != '\n') { printf("mountroot: advancing to next directive...\n"); (void)parse_skipto(&conf, '\n'); } mp = TAILQ_NEXT(mpdevfs, mnt_list); } if (mp != NULL) return (0); /* * We failed to mount (a new) root. */ switch (root_mount_onfail) { case A_CONTINUE: break; case A_PANIC: panic("mountroot: unable to (re-)mount root."); /* NOTREACHED */ case A_RETRY: goto retry; case A_REBOOT: kern_reboot(RB_NOSYNC); /* NOTREACHED */ } return (error); } static void vfs_mountroot_conf0(struct sbuf *sb) { char *s, *tok, *mnt, *opt; int error; sbuf_printf(sb, ".onfail panic\n"); sbuf_printf(sb, ".timeout %d\n", root_mount_timeout); if (boothowto & RB_ASKNAME) sbuf_printf(sb, ".ask\n"); #ifdef ROOTDEVNAME if (boothowto & RB_DFLTROOT) sbuf_printf(sb, "%s\n", ROOTDEVNAME); #endif if (boothowto & RB_CDROM) { sbuf_printf(sb, "cd9660:/dev/cd0 ro\n"); sbuf_printf(sb, ".timeout 0\n"); sbuf_printf(sb, "cd9660:/dev/cd1 ro\n"); sbuf_printf(sb, ".timeout %d\n", root_mount_timeout); } s = kern_getenv("vfs.root.mountfrom"); if (s != NULL) { opt = kern_getenv("vfs.root.mountfrom.options"); tok = s; error = parse_token(&tok, &mnt); while (!error) { sbuf_printf(sb, "%s %s\n", mnt, (opt != NULL) ? opt : ""); free(mnt, M_TEMP); error = parse_token(&tok, &mnt); } if (opt != NULL) freeenv(opt); freeenv(s); } if (rootdevnames[0] != NULL) sbuf_printf(sb, "%s\n", rootdevnames[0]); if (rootdevnames[1] != NULL) sbuf_printf(sb, "%s\n", rootdevnames[1]); #ifdef ROOTDEVNAME if (!(boothowto & RB_DFLTROOT)) sbuf_printf(sb, "%s\n", ROOTDEVNAME); #endif if (!(boothowto & RB_ASKNAME)) sbuf_printf(sb, ".ask\n"); } static int vfs_mountroot_readconf(struct thread *td, struct sbuf *sb) { static char buf[128]; struct nameidata nd; off_t ofs; ssize_t resid; int error, flags, len; NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/.mount.conf", td); flags = FREAD; error = vn_open(&nd, &flags, 0, NULL); if (error) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); ofs = 0; len = sizeof(buf) - 1; while (1) { error = vn_rdwr(UIO_READ, nd.ni_vp, buf, len, ofs, UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, NOCRED, &resid, td); if (error) break; if (resid == len) break; buf[len - resid] = 0; sbuf_printf(sb, "%s", buf); ofs += len - resid; } VOP_UNLOCK(nd.ni_vp, 0); vn_close(nd.ni_vp, FREAD, td->td_ucred, td); return (error); } static void vfs_mountroot_wait(void) { struct root_hold_token *h; struct timeval lastfail; int curfail; curfail = 0; while (1) { DROP_GIANT(); g_waitidle(); PICKUP_GIANT(); mtx_lock(&root_holds_mtx); if (LIST_EMPTY(&root_holds)) { mtx_unlock(&root_holds_mtx); break; } if (ppsratecheck(&lastfail, &curfail, 1)) { printf("Root mount waiting for:"); LIST_FOREACH(h, &root_holds, list) printf(" %s", h->who); printf("\n"); } msleep(&root_holds, &root_holds_mtx, PZERO | PDROP, "roothold", hz); } } static int vfs_mountroot_wait_if_neccessary(const char *fs, const char *dev) { int delay, timeout; /* * In case of ZFS and NFS we don't have a way to wait for * specific device. */ if (strcmp(fs, "zfs") == 0 || strstr(fs, "nfs") != NULL || dev[0] == '\0') { vfs_mountroot_wait(); return (0); } /* * Otherwise, no point in waiting if the device is already there. * Note that we must wait for GEOM to finish reconfiguring itself, * eg for geom_part(4) to finish tasting. */ DROP_GIANT(); g_waitidle(); PICKUP_GIANT(); if (parse_mount_dev_present(dev)) return (0); /* * No luck. Let's wait. This code looks weird, but it's that way * to behave exactly as it used to work before. */ vfs_mountroot_wait(); printf("mountroot: waiting for device %s...\n", dev); delay = hz / 10; timeout = root_mount_timeout * hz; do { pause("rmdev", delay); timeout -= delay; } while (timeout > 0 && !parse_mount_dev_present(dev)); if (timeout <= 0) return (ENODEV); return (0); } void vfs_mountroot(void) { struct mount *mp; struct sbuf *sb; struct thread *td; time_t timebase; int error; td = curthread; sb = sbuf_new_auto(); vfs_mountroot_conf0(sb); sbuf_finish(sb); error = vfs_mountroot_devfs(td, &mp); while (!error) { error = vfs_mountroot_parse(sb, mp); if (!error) { vfs_mountroot_shuffle(td, mp); sbuf_clear(sb); error = vfs_mountroot_readconf(td, sb); sbuf_finish(sb); } } sbuf_delete(sb); /* * Iterate over all currently mounted file systems and use * the time stamp found to check and/or initialize the RTC. * Call inittodr() only once and pass it the largest of the * timestamps we encounter. */ timebase = 0; mtx_lock(&mountlist_mtx); mp = TAILQ_FIRST(&mountlist); while (mp != NULL) { if (mp->mnt_time > timebase) timebase = mp->mnt_time; mp = TAILQ_NEXT(mp, mnt_list); } mtx_unlock(&mountlist_mtx); inittodr(timebase); /* Keep prison0's root in sync with the global rootvnode. */ mtx_lock(&prison0.pr_mtx); prison0.pr_root = rootvnode; vref(prison0.pr_root); mtx_unlock(&prison0.pr_mtx); mtx_lock(&root_holds_mtx); atomic_store_rel_int(&root_mount_complete, 1); wakeup(&root_mount_complete); mtx_unlock(&root_holds_mtx); EVENTHANDLER_INVOKE(mountroot); } static struct mntarg * parse_mountroot_options(struct mntarg *ma, const char *options) { char *p; char *name, *name_arg; char *val, *val_arg; char *opts; if (options == NULL || options[0] == '\0') return (ma); p = opts = strdup(options, M_MOUNT); if (opts == NULL) { return (ma); } while((name = strsep(&p, ",")) != NULL) { if (name[0] == '\0') break; val = strchr(name, '='); if (val != NULL) { *val = '\0'; ++val; } if( strcmp(name, "rw") == 0 || strcmp(name, "noro") == 0) { /* * The first time we mount the root file system, * we need to mount 'ro', so We need to ignore * 'rw' and 'noro' mount options. */ continue; } name_arg = strdup(name, M_MOUNT); val_arg = NULL; if (val != NULL) val_arg = strdup(val, M_MOUNT); ma = mount_arg(ma, name_arg, val_arg, (val_arg != NULL ? -1 : 0)); } free(opts, M_MOUNT); return (ma); } Index: head/sys/kern/vfs_subr.c =================================================================== --- head/sys/kern/vfs_subr.c (revision 305831) +++ head/sys/kern/vfs_subr.c (revision 305832) @@ -1,5327 +1,5327 @@ /*- * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95 */ /* * External virtual filesystem routines */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #include "opt_ddb.h" #include "opt_watchdog.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DDB #include #endif static void delmntque(struct vnode *vp); static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag, int slptimeo); static void syncer_shutdown(void *arg, int howto); static int vtryrecycle(struct vnode *vp); static void v_init_counters(struct vnode *); static void v_incr_usecount(struct vnode *); static void v_incr_usecount_locked(struct vnode *); static void v_incr_devcount(struct vnode *); static void v_decr_devcount(struct vnode *); static void vgonel(struct vnode *); static void vfs_knllock(void *arg); static void vfs_knlunlock(void *arg); static void vfs_knl_assert_locked(void *arg); static void vfs_knl_assert_unlocked(void *arg); static void destroy_vpollinfo(struct vpollinfo *vi); /* * Number of vnodes in existence. Increased whenever getnewvnode() * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode. */ static unsigned long numvnodes; SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "Number of vnodes in existence"); static u_long vnodes_created; SYSCTL_ULONG(_vfs, OID_AUTO, vnodes_created, CTLFLAG_RD, &vnodes_created, 0, "Number of vnodes created by getnewvnode"); /* * Conversion tables for conversion from vnode types to inode formats * and back. */ enum vtype iftovt_tab[16] = { VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, }; int vttoif_tab[10] = { 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT }; /* * List of vnodes that are ready for recycling. */ static TAILQ_HEAD(freelst, vnode) vnode_free_list; /* * "Free" vnode target. Free vnodes are rarely completely free, but are * just ones that are cheap to recycle. Usually they are for files which * have been stat'd but not read; these usually have inode and namecache * data attached to them. This target is the preferred minimum size of a * sub-cache consisting mostly of such files. The system balances the size * of this sub-cache with its complement to try to prevent either from * thrashing while the other is relatively inactive. The targets express * a preference for the best balance. * * "Above" this target there are 2 further targets (watermarks) related * to recyling of free vnodes. In the best-operating case, the cache is * exactly full, the free list has size between vlowat and vhiwat above the * free target, and recycling from it and normal use maintains this state. * Sometimes the free list is below vlowat or even empty, but this state * is even better for immediate use provided the cache is not full. * Otherwise, vnlru_proc() runs to reclaim enough vnodes (usually non-free * ones) to reach one of these states. The watermarks are currently hard- * coded as 4% and 9% of the available space higher. These and the default * of 25% for wantfreevnodes are too large if the memory size is large. * E.g., 9% of 75% of MAXVNODES is more than 566000 vnodes to reclaim * whenever vnlru_proc() becomes active. */ static u_long wantfreevnodes; SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "Target for minimum number of \"free\" vnodes"); static u_long freevnodes; SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "Number of \"free\" vnodes"); static u_long recycles_count; SYSCTL_ULONG(_vfs, OID_AUTO, recycles, CTLFLAG_RD, &recycles_count, 0, "Number of vnodes recycled to meet vnode cache targets"); /* * Various variables used for debugging the new implementation of * reassignbuf(). * XXX these are probably of (very) limited utility now. */ static int reassignbufcalls; SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "Number of calls to reassignbuf"); static u_long free_owe_inact; SYSCTL_ULONG(_vfs, OID_AUTO, free_owe_inact, CTLFLAG_RD, &free_owe_inact, 0, "Number of times free vnodes kept on active list due to VFS " "owing inactivation"); /* To keep more than one thread at a time from running vfs_getnewfsid */ static struct mtx mntid_mtx; /* * Lock for any access to the following: * vnode_free_list * numvnodes * freevnodes */ static struct mtx vnode_free_list_mtx; /* Publicly exported FS */ struct nfs_public nfs_pub; static uma_zone_t buf_trie_zone; /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */ static uma_zone_t vnode_zone; static uma_zone_t vnodepoll_zone; /* * The workitem queue. * * It is useful to delay writes of file data and filesystem metadata * for tens of seconds so that quickly created and deleted files need * not waste disk bandwidth being created and removed. To realize this, * we append vnodes to a "workitem" queue. When running with a soft * updates implementation, most pending metadata dependencies should * not wait for more than a few seconds. Thus, mounted on block devices * are delayed only about a half the time that file data is delayed. * Similarly, directory updates are more critical, so are only delayed * about a third the time that file data is delayed. Thus, there are * SYNCER_MAXDELAY queues that are processed round-robin at a rate of * one each second (driven off the filesystem syncer process). The * syncer_delayno variable indicates the next queue that is to be processed. * Items that need to be processed soon are placed in this queue: * * syncer_workitem_pending[syncer_delayno] * * A delay of fifteen seconds is done by placing the request fifteen * entries later in the queue: * * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] * */ static int syncer_delayno; static long syncer_mask; LIST_HEAD(synclist, bufobj); static struct synclist *syncer_workitem_pending; /* * The sync_mtx protects: * bo->bo_synclist * sync_vnode_count * syncer_delayno * syncer_state * syncer_workitem_pending * syncer_worklist_len * rushjob */ static struct mtx sync_mtx; static struct cv sync_wakeup; #define SYNCER_MAXDELAY 32 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ static int syncdelay = 30; /* max time to delay syncing data */ static int filedelay = 30; /* time to delay syncing files */ SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "Time to delay syncing files (in seconds)"); static int dirdelay = 29; /* time to delay syncing directories */ SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "Time to delay syncing directories (in seconds)"); static int metadelay = 28; /* time to delay syncing metadata */ SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "Time to delay syncing metadata (in seconds)"); static int rushjob; /* number of slots to run ASAP */ static int stat_rush_requests; /* number of times I/O speeded up */ SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "Number of times I/O speeded up (rush requests)"); /* * When shutting down the syncer, run it at four times normal speed. */ #define SYNCER_SHUTDOWN_SPEEDUP 4 static int sync_vnode_count; static int syncer_worklist_len; static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY } syncer_state; /* Target for maximum number of vnodes. */ int desiredvnodes; static int gapvnodes; /* gap between wanted and desired */ static int vhiwat; /* enough extras after expansion */ static int vlowat; /* minimal extras before expansion */ static int vstir; /* nonzero to stir non-free vnodes */ static volatile int vsmalltrigger = 8; /* pref to keep if > this many pages */ static int sysctl_update_desiredvnodes(SYSCTL_HANDLER_ARGS) { int error, old_desiredvnodes; old_desiredvnodes = desiredvnodes; if ((error = sysctl_handle_int(oidp, arg1, arg2, req)) != 0) return (error); if (old_desiredvnodes != desiredvnodes) { wantfreevnodes = desiredvnodes / 4; /* XXX locking seems to be incomplete. */ vfs_hash_changesize(desiredvnodes); cache_changesize(desiredvnodes); } return (0); } SYSCTL_PROC(_kern, KERN_MAXVNODES, maxvnodes, CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, &desiredvnodes, 0, sysctl_update_desiredvnodes, "I", "Target for maximum number of vnodes"); SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW, &wantfreevnodes, 0, "Old name for vfs.wantfreevnodes (legacy)"); static int vnlru_nowhere; SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0, "Number of times the vnlru process ran without success"); /* Shift count for (uintptr_t)vp to initialize vp->v_hash. */ static int vnsz2log; /* * Support for the bufobj clean & dirty pctrie. */ static void * buf_trie_alloc(struct pctrie *ptree) { return uma_zalloc(buf_trie_zone, M_NOWAIT); } static void buf_trie_free(struct pctrie *ptree, void *node) { uma_zfree(buf_trie_zone, node); } PCTRIE_DEFINE(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free); /* * Initialize the vnode management data structures. * * Reevaluate the following cap on the number of vnodes after the physical * memory size exceeds 512GB. In the limit, as the physical memory size * grows, the ratio of the memory size in KB to to vnodes approaches 64:1. */ #ifndef MAXVNODES_MAX #define MAXVNODES_MAX (512 * 1024 * 1024 / 64) /* 8M */ #endif /* * Initialize a vnode as it first enters the zone. */ static int vnode_init(void *mem, int size, int flags) { struct vnode *vp; struct bufobj *bo; vp = mem; bzero(vp, size); /* * Setup locks. */ vp->v_vnlock = &vp->v_lock; mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF); /* * By default, don't allow shared locks unless filesystems opt-in. */ lockinit(vp->v_vnlock, PVFS, "vnode", VLKTIMEOUT, LK_NOSHARE | LK_IS_VNODE); /* * Initialize bufobj. */ bo = &vp->v_bufobj; bo->__bo_vnode = vp; rw_init(BO_LOCKPTR(bo), "bufobj interlock"); bo->bo_private = vp; TAILQ_INIT(&bo->bo_clean.bv_hd); TAILQ_INIT(&bo->bo_dirty.bv_hd); /* * Initialize namecache. */ LIST_INIT(&vp->v_cache_src); TAILQ_INIT(&vp->v_cache_dst); /* * Initialize rangelocks. */ rangelock_init(&vp->v_rl); return (0); } /* * Free a vnode when it is cleared from the zone. */ static void vnode_fini(void *mem, int size) { struct vnode *vp; struct bufobj *bo; vp = mem; rangelock_destroy(&vp->v_rl); lockdestroy(vp->v_vnlock); mtx_destroy(&vp->v_interlock); bo = &vp->v_bufobj; rw_destroy(BO_LOCKPTR(bo)); } /* * Provide the size of NFS nclnode and NFS fh for calculation of the * vnode memory consumption. The size is specified directly to * eliminate dependency on NFS-private header. * * Other filesystems may use bigger or smaller (like UFS and ZFS) * private inode data, but the NFS-based estimation is ample enough. * Still, we care about differences in the size between 64- and 32-bit * platforms. * * Namecache structure size is heuristically * sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1. */ #ifdef _LP64 #define NFS_NCLNODE_SZ (528 + 64) #define NC_SZ 148 #else #define NFS_NCLNODE_SZ (360 + 32) #define NC_SZ 92 #endif static void vntblinit(void *dummy __unused) { u_int i; int physvnodes, virtvnodes; /* * Desiredvnodes is a function of the physical memory size and the * kernel's heap size. Generally speaking, it scales with the * physical memory size. The ratio of desiredvnodes to the physical * memory size is 1:16 until desiredvnodes exceeds 98,304. * Thereafter, the * marginal ratio of desiredvnodes to the physical memory size is * 1:64. However, desiredvnodes is limited by the kernel's heap * size. The memory required by desiredvnodes vnodes and vm objects * must not exceed 1/10th of the kernel's heap size. */ physvnodes = maxproc + pgtok(vm_cnt.v_page_count) / 64 + 3 * min(98304 * 16, pgtok(vm_cnt.v_page_count)) / 64; virtvnodes = vm_kmem_size / (10 * (sizeof(struct vm_object) + sizeof(struct vnode) + NC_SZ * ncsizefactor + NFS_NCLNODE_SZ)); desiredvnodes = min(physvnodes, virtvnodes); if (desiredvnodes > MAXVNODES_MAX) { if (bootverbose) printf("Reducing kern.maxvnodes %d -> %d\n", desiredvnodes, MAXVNODES_MAX); desiredvnodes = MAXVNODES_MAX; } wantfreevnodes = desiredvnodes / 4; mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF); TAILQ_INIT(&vnode_free_list); mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF); vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL, vnode_init, vnode_fini, UMA_ALIGN_PTR, 0); vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); /* * Preallocate enough nodes to support one-per buf so that * we can not fail an insert. reassignbuf() callers can not * tolerate the insertion failure. */ buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(), NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE | UMA_ZONE_VM); uma_prealloc(buf_trie_zone, nbuf); /* * Initialize the filesystem syncer. */ syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE, &syncer_mask); syncer_maxdelay = syncer_mask + 1; mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF); cv_init(&sync_wakeup, "syncer"); for (i = 1; i <= sizeof(struct vnode); i <<= 1) vnsz2log++; vnsz2log--; } SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL); /* * Mark a mount point as busy. Used to synchronize access and to delay * unmounting. Eventually, mountlist_mtx is not released on failure. * * vfs_busy() is a custom lock, it can block the caller. * vfs_busy() only sleeps if the unmount is active on the mount point. * For a mountpoint mp, vfs_busy-enforced lock is before lock of any * vnode belonging to mp. * * Lookup uses vfs_busy() to traverse mount points. * root fs var fs * / vnode lock A / vnode lock (/var) D * /var vnode lock B /log vnode lock(/var/log) E * vfs_busy lock C vfs_busy lock F * * Within each file system, the lock order is C->A->B and F->D->E. * * When traversing across mounts, the system follows that lock order: * * C->A->B * | * +->F->D->E * * The lookup() process for namei("/var") illustrates the process: * VOP_LOOKUP() obtains B while A is held * vfs_busy() obtains a shared lock on F while A and B are held * vput() releases lock on B * vput() releases lock on A * VFS_ROOT() obtains lock on D while shared lock on F is held * vfs_unbusy() releases shared lock on F * vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A. * Attempt to lock A (instead of vp_crossmp) while D is held would * violate the global order, causing deadlocks. * * dounmount() locks B while F is drained. */ int vfs_busy(struct mount *mp, int flags) { MPASS((flags & ~MBF_MASK) == 0); CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags); MNT_ILOCK(mp); MNT_REF(mp); /* * If mount point is currently being unmounted, sleep until the * mount point fate is decided. If thread doing the unmounting fails, * it will clear MNTK_UNMOUNT flag before waking us up, indicating * that this mount point has survived the unmount attempt and vfs_busy * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE * flag in addition to MNTK_UNMOUNT, indicating that mount point is * about to be really destroyed. vfs_busy needs to release its * reference on the mount point in this case and return with ENOENT, * telling the caller that mount mount it tried to busy is no longer * valid. */ while (mp->mnt_kern_flag & MNTK_UNMOUNT) { if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) { MNT_REL(mp); MNT_IUNLOCK(mp); CTR1(KTR_VFS, "%s: failed busying before sleeping", __func__); return (ENOENT); } if (flags & MBF_MNTLSTLOCK) mtx_unlock(&mountlist_mtx); mp->mnt_kern_flag |= MNTK_MWAIT; msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0); if (flags & MBF_MNTLSTLOCK) mtx_lock(&mountlist_mtx); MNT_ILOCK(mp); } if (flags & MBF_MNTLSTLOCK) mtx_unlock(&mountlist_mtx); mp->mnt_lockref++; MNT_IUNLOCK(mp); return (0); } /* * Free a busy filesystem. */ void vfs_unbusy(struct mount *mp) { CTR2(KTR_VFS, "%s: mp %p", __func__, mp); MNT_ILOCK(mp); MNT_REL(mp); KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref")); mp->mnt_lockref--; if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) { MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT); CTR1(KTR_VFS, "%s: waking up waiters", __func__); mp->mnt_kern_flag &= ~MNTK_DRAINING; wakeup(&mp->mnt_lockref); } MNT_IUNLOCK(mp); } /* * Lookup a mount point by filesystem identifier. */ struct mount * vfs_getvfs(fsid_t *fsid) { struct mount *mp; CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { vfs_ref(mp); mtx_unlock(&mountlist_mtx); return (mp); } } mtx_unlock(&mountlist_mtx); CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); return ((struct mount *) 0); } /* * Lookup a mount point by filesystem identifier, busying it before * returning. * * To avoid congestion on mountlist_mtx, implement simple direct-mapped * cache for popular filesystem identifiers. The cache is lockess, using * the fact that struct mount's are never freed. In worst case we may * get pointer to unmounted or even different filesystem, so we have to * check what we got, and go slow way if so. */ struct mount * vfs_busyfs(fsid_t *fsid) { #define FSID_CACHE_SIZE 256 typedef struct mount * volatile vmp_t; static vmp_t cache[FSID_CACHE_SIZE]; struct mount *mp; int error; uint32_t hash; CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); hash = fsid->val[0] ^ fsid->val[1]; hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1); mp = cache[hash]; if (mp == NULL || mp->mnt_stat.f_fsid.val[0] != fsid->val[0] || mp->mnt_stat.f_fsid.val[1] != fsid->val[1]) goto slow; if (vfs_busy(mp, 0) != 0) { cache[hash] = NULL; goto slow; } if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) return (mp); else vfs_unbusy(mp); slow: mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { error = vfs_busy(mp, MBF_MNTLSTLOCK); if (error) { cache[hash] = NULL; mtx_unlock(&mountlist_mtx); return (NULL); } cache[hash] = mp; return (mp); } } CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); mtx_unlock(&mountlist_mtx); return ((struct mount *) 0); } /* * Check if a user can access privileged mount options. */ int vfs_suser(struct mount *mp, struct thread *td) { int error; /* * If the thread is jailed, but this is not a jail-friendly file * system, deny immediately. */ if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred)) return (EPERM); /* * If the file system was mounted outside the jail of the calling * thread, deny immediately. */ if (prison_check(td->td_ucred, mp->mnt_cred) != 0) return (EPERM); /* * If file system supports delegated administration, we don't check * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified * by the file system itself. * If this is not the user that did original mount, we check for * the PRIV_VFS_MOUNT_OWNER privilege. */ if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) && mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) { if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0) return (error); } return (0); } /* * Get a new unique fsid. Try to make its val[0] unique, since this value * will be used to create fake device numbers for stat(). Also try (but * not so hard) make its val[0] unique mod 2^16, since some emulators only * support 16-bit device numbers. We end up with unique val[0]'s for the * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. * * Keep in mind that several mounts may be running in parallel. Starting * the search one past where the previous search terminated is both a * micro-optimization and a defense against returning the same fsid to * different mounts. */ void vfs_getnewfsid(struct mount *mp) { static uint16_t mntid_base; struct mount *nmp; fsid_t tfsid; int mtype; CTR2(KTR_VFS, "%s: mp %p", __func__, mp); mtx_lock(&mntid_mtx); mtype = mp->mnt_vfc->vfc_typenum; tfsid.val[1] = mtype; mtype = (mtype & 0xFF) << 24; for (;;) { tfsid.val[0] = makedev(255, mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); mntid_base++; if ((nmp = vfs_getvfs(&tfsid)) == NULL) break; vfs_rel(nmp); } mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; mtx_unlock(&mntid_mtx); } /* * Knob to control the precision of file timestamps: * * 0 = seconds only; nanoseconds zeroed. * 1 = seconds and nanoseconds, accurate within 1/HZ. * 2 = seconds and nanoseconds, truncated to microseconds. * >=3 = seconds and nanoseconds, maximum precision. */ enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; static int timestamp_precision = TSP_USEC; SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, ×tamp_precision, 0, "File timestamp precision (0: seconds, " "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, " "3+: sec + ns (max. precision))"); /* * Get a current timestamp. */ void vfs_timestamp(struct timespec *tsp) { struct timeval tv; switch (timestamp_precision) { case TSP_SEC: tsp->tv_sec = time_second; tsp->tv_nsec = 0; break; case TSP_HZ: getnanotime(tsp); break; case TSP_USEC: microtime(&tv); TIMEVAL_TO_TIMESPEC(&tv, tsp); break; case TSP_NSEC: default: nanotime(tsp); break; } } /* * Set vnode attributes to VNOVAL */ void vattr_null(struct vattr *vap) { vap->va_type = VNON; vap->va_size = VNOVAL; vap->va_bytes = VNOVAL; vap->va_mode = VNOVAL; vap->va_nlink = VNOVAL; vap->va_uid = VNOVAL; vap->va_gid = VNOVAL; vap->va_fsid = VNOVAL; vap->va_fileid = VNOVAL; vap->va_blocksize = VNOVAL; vap->va_rdev = VNOVAL; vap->va_atime.tv_sec = VNOVAL; vap->va_atime.tv_nsec = VNOVAL; vap->va_mtime.tv_sec = VNOVAL; vap->va_mtime.tv_nsec = VNOVAL; vap->va_ctime.tv_sec = VNOVAL; vap->va_ctime.tv_nsec = VNOVAL; vap->va_birthtime.tv_sec = VNOVAL; vap->va_birthtime.tv_nsec = VNOVAL; vap->va_flags = VNOVAL; vap->va_gen = VNOVAL; vap->va_vaflags = 0; } /* * This routine is called when we have too many vnodes. It attempts * to free vnodes and will potentially free vnodes that still * have VM backing store (VM backing store is typically the cause * of a vnode blowout so we want to do this). Therefore, this operation * is not considered cheap. * * A number of conditions may prevent a vnode from being reclaimed. * the buffer cache may have references on the vnode, a directory * vnode may still have references due to the namei cache representing * underlying files, or the vnode may be in active use. It is not * desirable to reuse such vnodes. These conditions may cause the * number of vnodes to reach some minimum value regardless of what * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. */ static int vlrureclaim(struct mount *mp, int reclaim_nc_src, int trigger) { struct vnode *vp; int count, done, target; done = 0; vn_start_write(NULL, &mp, V_WAIT); MNT_ILOCK(mp); count = mp->mnt_nvnodelistsize; target = count * (int64_t)gapvnodes / imax(desiredvnodes, 1); target = target / 10 + 1; while (count != 0 && done < target) { vp = TAILQ_FIRST(&mp->mnt_nvnodelist); while (vp != NULL && vp->v_type == VMARKER) vp = TAILQ_NEXT(vp, v_nmntvnodes); if (vp == NULL) break; /* * XXX LRU is completely broken for non-free vnodes. First * by calling here in mountpoint order, then by moving * unselected vnodes to the end here, and most grossly by * removing the vlruvp() function that was supposed to * maintain the order. (This function was born broken * since syncer problems prevented it doing anything.) The * order is closer to LRC (C = Created). * * LRU reclaiming of vnodes seems to have last worked in * FreeBSD-3 where LRU wasn't mentioned under any spelling. * Then there was no hold count, and inactive vnodes were * simply put on the free list in LRU order. The separate * lists also break LRU. We prefer to reclaim from the * free list for technical reasons. This tends to thrash * the free list to keep very unrecently used held vnodes. * The problem is mitigated by keeping the free list large. */ TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); --count; if (!VI_TRYLOCK(vp)) goto next_iter; /* * If it's been deconstructed already, it's still * referenced, or it exceeds the trigger, skip it. * Also skip free vnodes. We are trying to make space * to expand the free list, not reduce it. */ if (vp->v_usecount || (!reclaim_nc_src && !LIST_EMPTY(&vp->v_cache_src)) || ((vp->v_iflag & VI_FREE) != 0) || (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL && vp->v_object->resident_page_count > trigger)) { VI_UNLOCK(vp); goto next_iter; } MNT_IUNLOCK(mp); vholdl(vp); if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) { vdrop(vp); goto next_iter_mntunlocked; } VI_LOCK(vp); /* * v_usecount may have been bumped after VOP_LOCK() dropped * the vnode interlock and before it was locked again. * * It is not necessary to recheck VI_DOOMED because it can * only be set by another thread that holds both the vnode * lock and vnode interlock. If another thread has the * vnode lock before we get to VOP_LOCK() and obtains the * vnode interlock after VOP_LOCK() drops the vnode * interlock, the other thread will be unable to drop the * vnode lock before our VOP_LOCK() call fails. */ if (vp->v_usecount || (!reclaim_nc_src && !LIST_EMPTY(&vp->v_cache_src)) || (vp->v_iflag & VI_FREE) != 0 || (vp->v_object != NULL && vp->v_object->resident_page_count > trigger)) { VOP_UNLOCK(vp, LK_INTERLOCK); vdrop(vp); goto next_iter_mntunlocked; } KASSERT((vp->v_iflag & VI_DOOMED) == 0, ("VI_DOOMED unexpectedly detected in vlrureclaim()")); atomic_add_long(&recycles_count, 1); vgonel(vp); VOP_UNLOCK(vp, 0); vdropl(vp); done++; next_iter_mntunlocked: if (!should_yield()) goto relock_mnt; goto yield; next_iter: if (!should_yield()) continue; MNT_IUNLOCK(mp); yield: kern_yield(PRI_USER); relock_mnt: MNT_ILOCK(mp); } MNT_IUNLOCK(mp); vn_finished_write(mp); return done; } static int max_vnlru_free = 10000; /* limit on vnode free requests per call */ SYSCTL_INT(_debug, OID_AUTO, max_vnlru_free, CTLFLAG_RW, &max_vnlru_free, 0, "limit on vnode free requests per call to the vnlru_free routine"); /* * Attempt to reduce the free list by the requested amount. */ static void vnlru_free_locked(int count, struct vfsops *mnt_op) { struct vnode *vp; struct mount *mp; mtx_assert(&vnode_free_list_mtx, MA_OWNED); if (count > max_vnlru_free) count = max_vnlru_free; for (; count > 0; count--) { vp = TAILQ_FIRST(&vnode_free_list); /* * The list can be modified while the free_list_mtx * has been dropped and vp could be NULL here. */ if (!vp) break; VNASSERT(vp->v_op != NULL, vp, ("vnlru_free: vnode already reclaimed.")); KASSERT((vp->v_iflag & VI_FREE) != 0, ("Removing vnode not on freelist")); KASSERT((vp->v_iflag & VI_ACTIVE) == 0, ("Mangling active vnode")); TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist); /* * Don't recycle if our vnode is from different type * of mount point. Note that mp is type-safe, the * check does not reach unmapped address even if * vnode is reclaimed. * Don't recycle if we can't get the interlock without * blocking. */ if ((mnt_op != NULL && (mp = vp->v_mount) != NULL && mp->mnt_op != mnt_op) || !VI_TRYLOCK(vp)) { TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist); continue; } VNASSERT((vp->v_iflag & VI_FREE) != 0 && vp->v_holdcnt == 0, vp, ("vp inconsistent on freelist")); /* * The clear of VI_FREE prevents activation of the * vnode. There is no sense in putting the vnode on * the mount point active list, only to remove it * later during recycling. Inline the relevant part * of vholdl(), to avoid triggering assertions or * activating. */ freevnodes--; vp->v_iflag &= ~VI_FREE; refcount_acquire(&vp->v_holdcnt); mtx_unlock(&vnode_free_list_mtx); VI_UNLOCK(vp); vtryrecycle(vp); /* * If the recycled succeeded this vdrop will actually free * the vnode. If not it will simply place it back on * the free list. */ vdrop(vp); mtx_lock(&vnode_free_list_mtx); } } void vnlru_free(int count, struct vfsops *mnt_op) { mtx_lock(&vnode_free_list_mtx); vnlru_free_locked(count, mnt_op); mtx_unlock(&vnode_free_list_mtx); } /* XXX some names and initialization are bad for limits and watermarks. */ static int vspace(void) { int space; gapvnodes = imax(desiredvnodes - wantfreevnodes, 100); vhiwat = gapvnodes / 11; /* 9% -- just under the 10% in vlrureclaim() */ vlowat = vhiwat / 2; if (numvnodes > desiredvnodes) return (0); space = desiredvnodes - numvnodes; if (freevnodes > wantfreevnodes) space += freevnodes - wantfreevnodes; return (space); } /* * Attempt to recycle vnodes in a context that is always safe to block. * Calling vlrurecycle() from the bowels of filesystem code has some * interesting deadlock problems. */ static struct proc *vnlruproc; static int vnlruproc_sig; static void vnlru_proc(void) { struct mount *mp, *nmp; unsigned long ofreevnodes, onumvnodes; int done, force, reclaim_nc_src, trigger, usevnodes; EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, vnlruproc, SHUTDOWN_PRI_FIRST); force = 0; for (;;) { kproc_suspend_check(vnlruproc); mtx_lock(&vnode_free_list_mtx); /* * If numvnodes is too large (due to desiredvnodes being * adjusted using its sysctl, or emergency growth), first * try to reduce it by discarding from the free list. */ if (numvnodes > desiredvnodes && freevnodes > 0) vnlru_free_locked(ulmin(numvnodes - desiredvnodes, freevnodes), NULL); /* * Sleep if the vnode cache is in a good state. This is * when it is not over-full and has space for about a 4% * or 9% expansion (by growing its size or inexcessively * reducing its free list). Otherwise, try to reclaim * space for a 10% expansion. */ if (vstir && force == 0) { force = 1; vstir = 0; } if (vspace() >= vlowat && force == 0) { vnlruproc_sig = 0; wakeup(&vnlruproc_sig); msleep(vnlruproc, &vnode_free_list_mtx, PVFS|PDROP, "vlruwt", hz); continue; } mtx_unlock(&vnode_free_list_mtx); done = 0; ofreevnodes = freevnodes; onumvnodes = numvnodes; /* * Calculate parameters for recycling. These are the same * throughout the loop to give some semblance of fairness. * The trigger point is to avoid recycling vnodes with lots * of resident pages. We aren't trying to free memory; we * are trying to recycle or at least free vnodes. */ if (numvnodes <= desiredvnodes) usevnodes = numvnodes - freevnodes; else usevnodes = numvnodes; if (usevnodes <= 0) usevnodes = 1; /* * The trigger value is is chosen to give a conservatively * large value to ensure that it alone doesn't prevent * making progress. The value can easily be so large that * it is effectively infinite in some congested and * misconfigured cases, and this is necessary. Normally * it is about 8 to 100 (pages), which is quite large. */ trigger = vm_cnt.v_page_count * 2 / usevnodes; if (force < 2) trigger = vsmalltrigger; reclaim_nc_src = force >= 3; mtx_lock(&mountlist_mtx); for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } done += vlrureclaim(mp, reclaim_nc_src, trigger); mtx_lock(&mountlist_mtx); nmp = TAILQ_NEXT(mp, mnt_list); vfs_unbusy(mp); } mtx_unlock(&mountlist_mtx); if (onumvnodes > desiredvnodes && numvnodes <= desiredvnodes) uma_reclaim(); if (done == 0) { if (force == 0 || force == 1) { force = 2; continue; } if (force == 2) { force = 3; continue; } force = 0; vnlru_nowhere++; tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3); } else kern_yield(PRI_USER); /* * After becoming active to expand above low water, keep * active until above high water. */ force = vspace() < vhiwat; } } static struct kproc_desc vnlru_kp = { "vnlru", vnlru_proc, &vnlruproc }; SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp); /* * Routines having to do with the management of the vnode table. */ /* * Try to recycle a freed vnode. We abort if anyone picks up a reference * before we actually vgone(). This function must be called with the vnode * held to prevent the vnode from being returned to the free list midway * through vgone(). */ static int vtryrecycle(struct vnode *vp) { struct mount *vnmp; CTR2(KTR_VFS, "%s: vp %p", __func__, vp); VNASSERT(vp->v_holdcnt, vp, ("vtryrecycle: Recycling vp %p without a reference.", vp)); /* * This vnode may found and locked via some other list, if so we * can't recycle it yet. */ if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { CTR2(KTR_VFS, "%s: impossible to recycle, vp %p lock is already held", __func__, vp); return (EWOULDBLOCK); } /* * Don't recycle if its filesystem is being suspended. */ if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) { VOP_UNLOCK(vp, 0); CTR2(KTR_VFS, "%s: impossible to recycle, cannot start the write for %p", __func__, vp); return (EBUSY); } /* * If we got this far, we need to acquire the interlock and see if * anyone picked up this vnode from another list. If not, we will * mark it with DOOMED via vgonel() so that anyone who does find it * will skip over it. */ VI_LOCK(vp); if (vp->v_usecount) { VOP_UNLOCK(vp, LK_INTERLOCK); vn_finished_write(vnmp); CTR2(KTR_VFS, "%s: impossible to recycle, %p is already referenced", __func__, vp); return (EBUSY); } if ((vp->v_iflag & VI_DOOMED) == 0) { atomic_add_long(&recycles_count, 1); vgonel(vp); } VOP_UNLOCK(vp, LK_INTERLOCK); vn_finished_write(vnmp); return (0); } static void vcheckspace(void) { if (vspace() < vlowat && vnlruproc_sig == 0) { vnlruproc_sig = 1; wakeup(vnlruproc); } } /* * Wait if necessary for space for a new vnode. */ static int getnewvnode_wait(int suspended) { mtx_assert(&vnode_free_list_mtx, MA_OWNED); if (numvnodes >= desiredvnodes) { if (suspended) { /* * The file system is being suspended. We cannot * risk a deadlock here, so allow allocation of * another vnode even if this would give too many. */ return (0); } if (vnlruproc_sig == 0) { vnlruproc_sig = 1; /* avoid unnecessary wakeups */ wakeup(vnlruproc); } msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS, "vlruwk", hz); } /* Post-adjust like the pre-adjust in getnewvnode(). */ if (numvnodes + 1 > desiredvnodes && freevnodes > 1) vnlru_free_locked(1, NULL); return (numvnodes >= desiredvnodes ? ENFILE : 0); } /* * This hack is fragile, and probably not needed any more now that the * watermark handling works. */ void getnewvnode_reserve(u_int count) { struct thread *td; /* Pre-adjust like the pre-adjust in getnewvnode(), with any count. */ /* XXX no longer so quick, but this part is not racy. */ mtx_lock(&vnode_free_list_mtx); if (numvnodes + count > desiredvnodes && freevnodes > wantfreevnodes) vnlru_free_locked(ulmin(numvnodes + count - desiredvnodes, freevnodes - wantfreevnodes), NULL); mtx_unlock(&vnode_free_list_mtx); td = curthread; /* First try to be quick and racy. */ if (atomic_fetchadd_long(&numvnodes, count) + count <= desiredvnodes) { td->td_vp_reserv += count; vcheckspace(); /* XXX no longer so quick, but more racy */ return; } else atomic_subtract_long(&numvnodes, count); mtx_lock(&vnode_free_list_mtx); while (count > 0) { if (getnewvnode_wait(0) == 0) { count--; td->td_vp_reserv++; atomic_add_long(&numvnodes, 1); } } vcheckspace(); mtx_unlock(&vnode_free_list_mtx); } /* * This hack is fragile, especially if desiredvnodes or wantvnodes are * misconfgured or changed significantly. Reducing desiredvnodes below * the reserved amount should cause bizarre behaviour like reducing it * below the number of active vnodes -- the system will try to reduce * numvnodes to match, but should fail, so the subtraction below should * not overflow. */ void getnewvnode_drop_reserve(void) { struct thread *td; td = curthread; atomic_subtract_long(&numvnodes, td->td_vp_reserv); td->td_vp_reserv = 0; } /* * Return the next vnode from the free list. */ int getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops, struct vnode **vpp) { struct vnode *vp; struct thread *td; struct lock_object *lo; static int cyclecount; int error; CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag); vp = NULL; td = curthread; if (td->td_vp_reserv > 0) { td->td_vp_reserv -= 1; goto alloc; } mtx_lock(&vnode_free_list_mtx); if (numvnodes < desiredvnodes) cyclecount = 0; else if (cyclecount++ >= freevnodes) { cyclecount = 0; vstir = 1; } /* * Grow the vnode cache if it will not be above its target max * after growing. Otherwise, if the free list is nonempty, try * to reclaim 1 item from it before growing the cache (possibly * above its target max if the reclamation failed or is delayed). * Otherwise, wait for some space. In all cases, schedule * vnlru_proc() if we are getting short of space. The watermarks * should be chosen so that we never wait or even reclaim from * the free list to below its target minimum. */ if (numvnodes + 1 <= desiredvnodes) ; else if (freevnodes > 0) vnlru_free_locked(1, NULL); else { error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)); #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */ if (error != 0) { mtx_unlock(&vnode_free_list_mtx); return (error); } #endif } vcheckspace(); atomic_add_long(&numvnodes, 1); mtx_unlock(&vnode_free_list_mtx); alloc: atomic_add_long(&vnodes_created, 1); vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK); /* * Locks are given the generic name "vnode" when created. * Follow the historic practice of using the filesystem * name when they allocated, e.g., "zfs", "ufs", "nfs, etc. * * Locks live in a witness group keyed on their name. Thus, * when a lock is renamed, it must also move from the witness * group of its old name to the witness group of its new name. * * The change only needs to be made when the vnode moves * from one filesystem type to another. We ensure that each * filesystem use a single static name pointer for its tag so * that we can compare pointers rather than doing a strcmp(). */ lo = &vp->v_vnlock->lock_object; if (lo->lo_name != tag) { lo->lo_name = tag; WITNESS_DESTROY(lo); WITNESS_INIT(lo, tag); } /* * By default, don't allow shared locks unless filesystems opt-in. */ vp->v_vnlock->lock_object.lo_flags |= LK_NOSHARE; /* * Finalize various vnode identity bits. */ KASSERT(vp->v_object == NULL, ("stale v_object %p", vp)); KASSERT(vp->v_lockf == NULL, ("stale v_lockf %p", vp)); KASSERT(vp->v_pollinfo == NULL, ("stale v_pollinfo %p", vp)); vp->v_type = VNON; vp->v_tag = tag; vp->v_op = vops; v_init_counters(vp); vp->v_bufobj.bo_ops = &buf_ops_bio; #ifdef MAC mac_vnode_init(vp); if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0) mac_vnode_associate_singlelabel(mp, vp); else if (mp == NULL && vops != &dead_vnodeops) printf("NULL mp in getnewvnode()\n"); #endif if (mp != NULL) { vp->v_bufobj.bo_bsize = mp->mnt_stat.f_iosize; if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0) vp->v_vflag |= VV_NOKNOTE; } /* * For the filesystems which do not use vfs_hash_insert(), * still initialize v_hash to have vfs_hash_index() useful. * E.g., nullfs uses vfs_hash_index() on the lower vnode for * its own hashing. */ vp->v_hash = (uintptr_t)vp >> vnsz2log; *vpp = vp; return (0); } /* * Delete from old mount point vnode list, if on one. */ static void delmntque(struct vnode *vp) { struct mount *mp; int active; mp = vp->v_mount; if (mp == NULL) return; MNT_ILOCK(mp); VI_LOCK(vp); KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize, ("Active vnode list size %d > Vnode list size %d", mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize)); active = vp->v_iflag & VI_ACTIVE; vp->v_iflag &= ~VI_ACTIVE; if (active) { mtx_lock(&vnode_free_list_mtx); TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist); mp->mnt_activevnodelistsize--; mtx_unlock(&vnode_free_list_mtx); } vp->v_mount = NULL; VI_UNLOCK(vp); VNASSERT(mp->mnt_nvnodelistsize > 0, vp, ("bad mount point vnode list size")); TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); mp->mnt_nvnodelistsize--; MNT_REL(mp); MNT_IUNLOCK(mp); } static void insmntque_stddtr(struct vnode *vp, void *dtr_arg) { vp->v_data = NULL; vp->v_op = &dead_vnodeops; vgone(vp); vput(vp); } /* * Insert into list of vnodes for the new mount point, if available. */ int insmntque1(struct vnode *vp, struct mount *mp, void (*dtr)(struct vnode *, void *), void *dtr_arg) { KASSERT(vp->v_mount == NULL, ("insmntque: vnode already on per mount vnode list")); VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)")); ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp"); /* * We acquire the vnode interlock early to ensure that the * vnode cannot be recycled by another process releasing a * holdcnt on it before we get it on both the vnode list * and the active vnode list. The mount mutex protects only * manipulation of the vnode list and the vnode freelist * mutex protects only manipulation of the active vnode list. * Hence the need to hold the vnode interlock throughout. */ MNT_ILOCK(mp); VI_LOCK(vp); if (((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 && ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 || mp->mnt_nvnodelistsize == 0)) && (vp->v_vflag & VV_FORCEINSMQ) == 0) { VI_UNLOCK(vp); MNT_IUNLOCK(mp); if (dtr != NULL) dtr(vp, dtr_arg); return (EBUSY); } vp->v_mount = mp; MNT_REF(mp); TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); VNASSERT(mp->mnt_nvnodelistsize >= 0, vp, ("neg mount point vnode list size")); mp->mnt_nvnodelistsize++; KASSERT((vp->v_iflag & VI_ACTIVE) == 0, ("Activating already active vnode")); vp->v_iflag |= VI_ACTIVE; mtx_lock(&vnode_free_list_mtx); TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist); mp->mnt_activevnodelistsize++; mtx_unlock(&vnode_free_list_mtx); VI_UNLOCK(vp); MNT_IUNLOCK(mp); return (0); } int insmntque(struct vnode *vp, struct mount *mp) { return (insmntque1(vp, mp, insmntque_stddtr, NULL)); } /* * Flush out and invalidate all buffers associated with a bufobj * Called with the underlying object locked. */ int bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo) { int error; BO_LOCK(bo); if (flags & V_SAVE) { error = bufobj_wwait(bo, slpflag, slptimeo); if (error) { BO_UNLOCK(bo); return (error); } if (bo->bo_dirty.bv_cnt > 0) { BO_UNLOCK(bo); if ((error = BO_SYNC(bo, MNT_WAIT)) != 0) return (error); /* * XXX We could save a lock/unlock if this was only * enabled under INVARIANTS */ BO_LOCK(bo); if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) panic("vinvalbuf: dirty bufs"); } } /* * If you alter this loop please notice that interlock is dropped and * reacquired in flushbuflist. Special care is needed to ensure that * no race conditions occur from this. */ do { error = flushbuflist(&bo->bo_clean, flags, bo, slpflag, slptimeo); if (error == 0 && !(flags & V_CLEANONLY)) error = flushbuflist(&bo->bo_dirty, flags, bo, slpflag, slptimeo); if (error != 0 && error != EAGAIN) { BO_UNLOCK(bo); return (error); } } while (error != 0); /* * Wait for I/O to complete. XXX needs cleaning up. The vnode can * have write I/O in-progress but if there is a VM object then the * VM object can also have read-I/O in-progress. */ do { bufobj_wwait(bo, 0, 0); BO_UNLOCK(bo); if (bo->bo_object != NULL) { VM_OBJECT_WLOCK(bo->bo_object); vm_object_pip_wait(bo->bo_object, "bovlbx"); VM_OBJECT_WUNLOCK(bo->bo_object); } BO_LOCK(bo); } while (bo->bo_numoutput > 0); BO_UNLOCK(bo); /* * Destroy the copy in the VM cache, too. */ if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) { VM_OBJECT_WLOCK(bo->bo_object); vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ? OBJPR_CLEANONLY : 0); VM_OBJECT_WUNLOCK(bo->bo_object); } #ifdef INVARIANTS BO_LOCK(bo); if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 && (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0)) panic("vinvalbuf: flush failed"); BO_UNLOCK(bo); #endif return (0); } /* * Flush out and invalidate all buffers associated with a vnode. * Called with the underlying object locked. */ int vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) { CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags); ASSERT_VOP_LOCKED(vp, "vinvalbuf"); if (vp->v_object != NULL && vp->v_object->handle != vp) return (0); return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo)); } /* * Flush out buffers on the specified list. * */ static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag, int slptimeo) { struct buf *bp, *nbp; int retval, error; daddr_t lblkno; b_xflags_t xflags; ASSERT_BO_WLOCKED(bo); retval = 0; TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) { if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) || ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) { continue; } lblkno = 0; xflags = 0; if (nbp != NULL) { lblkno = nbp->b_lblkno; xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN); } retval = EAGAIN; error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo), "flushbuf", slpflag, slptimeo); if (error) { BO_LOCK(bo); return (error != ENOLCK ? error : EAGAIN); } KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); /* * XXX Since there are no node locks for NFS, I * believe there is a slight chance that a delayed * write will occur while sleeping just above, so * check for it. */ if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && (flags & V_SAVE)) { bremfree(bp); bp->b_flags |= B_ASYNC; bwrite(bp); BO_LOCK(bo); return (EAGAIN); /* XXX: why not loop ? */ } bremfree(bp); bp->b_flags |= (B_INVAL | B_RELBUF); bp->b_flags &= ~B_ASYNC; brelse(bp); BO_LOCK(bo); nbp = gbincore(bo, lblkno); if (nbp == NULL || (nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) != xflags) break; /* nbp invalid */ } return (retval); } int bnoreuselist(struct bufv *bufv, struct bufobj *bo, daddr_t startn, daddr_t endn) { struct buf *bp; int error; daddr_t lblkno; ASSERT_BO_LOCKED(bo); for (lblkno = startn;;) { again: bp = BUF_PCTRIE_LOOKUP_GE(&bufv->bv_root, lblkno); if (bp == NULL || bp->b_lblkno >= endn || bp->b_lblkno < startn) break; error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo), "brlsfl", 0, 0); if (error != 0) { BO_RLOCK(bo); if (error == ENOLCK) goto again; return (error); } KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); lblkno = bp->b_lblkno + 1; if ((bp->b_flags & B_MANAGED) == 0) bremfree(bp); bp->b_flags |= B_RELBUF; /* * In the VMIO case, use the B_NOREUSE flag to hint that the * pages backing each buffer in the range are unlikely to be * reused. Dirty buffers will have the hint applied once * they've been written. */ if (bp->b_vp->v_object != NULL) bp->b_flags |= B_NOREUSE; brelse(bp); BO_RLOCK(bo); } return (0); } /* * Truncate a file's buffer and pages to a specified length. This * is in lieu of the old vinvalbuf mechanism, which performed unneeded * sync activity. */ int vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize) { struct buf *bp, *nbp; int anyfreed; int trunclbn; struct bufobj *bo; CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__, vp, cred, blksize, (uintmax_t)length); /* * Round up to the *next* lbn. */ trunclbn = howmany(length, blksize); ASSERT_VOP_LOCKED(vp, "vtruncbuf"); restart: bo = &vp->v_bufobj; BO_LOCK(bo); anyfreed = 1; for (;anyfreed;) { anyfreed = 0; TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) { if (bp->b_lblkno < trunclbn) continue; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) goto restart; bremfree(bp); bp->b_flags |= (B_INVAL | B_RELBUF); bp->b_flags &= ~B_ASYNC; brelse(bp); anyfreed = 1; BO_LOCK(bo); if (nbp != NULL && (((nbp->b_xflags & BX_VNCLEAN) == 0) || (nbp->b_vp != vp) || (nbp->b_flags & B_DELWRI))) { BO_UNLOCK(bo); goto restart; } } TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if (bp->b_lblkno < trunclbn) continue; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) goto restart; bremfree(bp); bp->b_flags |= (B_INVAL | B_RELBUF); bp->b_flags &= ~B_ASYNC; brelse(bp); anyfreed = 1; BO_LOCK(bo); if (nbp != NULL && (((nbp->b_xflags & BX_VNDIRTY) == 0) || (nbp->b_vp != vp) || (nbp->b_flags & B_DELWRI) == 0)) { BO_UNLOCK(bo); goto restart; } } } if (length > 0) { restartsync: TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if (bp->b_lblkno > 0) continue; /* * Since we hold the vnode lock this should only * fail if we're racing with the buf daemon. */ if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { goto restart; } VNASSERT((bp->b_flags & B_DELWRI), vp, ("buf(%p) on dirty queue without DELWRI", bp)); bremfree(bp); bawrite(bp); BO_LOCK(bo); goto restartsync; } } bufobj_wwait(bo, 0, 0); BO_UNLOCK(bo); vnode_pager_setsize(vp, length); return (0); } static void buf_vlist_remove(struct buf *bp) { struct bufv *bv; KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); ASSERT_BO_WLOCKED(bp->b_bufobj); KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) != (BX_VNDIRTY|BX_VNCLEAN), ("buf_vlist_remove: Buf %p is on two lists", bp)); if (bp->b_xflags & BX_VNDIRTY) bv = &bp->b_bufobj->bo_dirty; else bv = &bp->b_bufobj->bo_clean; BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno); TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs); bv->bv_cnt--; bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); } /* * Add the buffer to the sorted clean or dirty block list. * * NOTE: xflags is passed as a constant, optimizing this inline function! */ static void buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags) { struct bufv *bv; struct buf *n; int error; ASSERT_BO_WLOCKED(bo); KASSERT((xflags & BX_VNDIRTY) == 0 || (bo->bo_flag & BO_DEAD) == 0, ("dead bo %p", bo)); KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags)); bp->b_xflags |= xflags; if (xflags & BX_VNDIRTY) bv = &bo->bo_dirty; else bv = &bo->bo_clean; /* * Keep the list ordered. Optimize empty list insertion. Assume * we tend to grow at the tail so lookup_le should usually be cheaper * than _ge. */ if (bv->bv_cnt == 0 || bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno) TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs); else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL) TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs); else TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs); error = BUF_PCTRIE_INSERT(&bv->bv_root, bp); if (error) panic("buf_vlist_add: Preallocated nodes insufficient."); bv->bv_cnt++; } /* * Look up a buffer using the buffer tries. */ struct buf * gbincore(struct bufobj *bo, daddr_t lblkno) { struct buf *bp; ASSERT_BO_LOCKED(bo); bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno); if (bp != NULL) return (bp); return BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno); } /* * Associate a buffer with a vnode. */ void bgetvp(struct vnode *vp, struct buf *bp) { struct bufobj *bo; bo = &vp->v_bufobj; ASSERT_BO_WLOCKED(bo); VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free")); CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags); VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp, ("bgetvp: bp already attached! %p", bp)); vhold(vp); bp->b_vp = vp; bp->b_bufobj = bo; /* * Insert onto list for new vnode. */ buf_vlist_add(bp, bo, BX_VNCLEAN); } /* * Disassociate a buffer from a vnode. */ void brelvp(struct buf *bp) { struct bufobj *bo; struct vnode *vp; CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); /* * Delete from old vnode list, if on one. */ vp = bp->b_vp; /* XXX */ bo = bp->b_bufobj; BO_LOCK(bo); if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) buf_vlist_remove(bp); else panic("brelvp: Buffer %p not on queue.", bp); if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { bo->bo_flag &= ~BO_ONWORKLST; mtx_lock(&sync_mtx); LIST_REMOVE(bo, bo_synclist); syncer_worklist_len--; mtx_unlock(&sync_mtx); } bp->b_vp = NULL; bp->b_bufobj = NULL; BO_UNLOCK(bo); vdrop(vp); } /* * Add an item to the syncer work queue. */ static void vn_syncer_add_to_worklist(struct bufobj *bo, int delay) { int slot; ASSERT_BO_WLOCKED(bo); mtx_lock(&sync_mtx); if (bo->bo_flag & BO_ONWORKLST) LIST_REMOVE(bo, bo_synclist); else { bo->bo_flag |= BO_ONWORKLST; syncer_worklist_len++; } if (delay > syncer_maxdelay - 2) delay = syncer_maxdelay - 2; slot = (syncer_delayno + delay) & syncer_mask; LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist); mtx_unlock(&sync_mtx); } static int sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS) { int error, len; mtx_lock(&sync_mtx); len = syncer_worklist_len - sync_vnode_count; mtx_unlock(&sync_mtx); error = SYSCTL_OUT(req, &len, sizeof(len)); return (error); } SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0, sysctl_vfs_worklist_len, "I", "Syncer thread worklist length"); static struct proc *updateproc; static void sched_sync(void); static struct kproc_desc up_kp = { "syncer", sched_sync, &updateproc }; SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp); static int sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td) { struct vnode *vp; struct mount *mp; *bo = LIST_FIRST(slp); if (*bo == NULL) return (0); vp = (*bo)->__bo_vnode; /* XXX */ if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0) return (1); /* * We use vhold in case the vnode does not * successfully sync. vhold prevents the vnode from * going away when we unlock the sync_mtx so that * we can acquire the vnode interlock. */ vholdl(vp); mtx_unlock(&sync_mtx); VI_UNLOCK(vp); if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { vdrop(vp); mtx_lock(&sync_mtx); return (*bo == LIST_FIRST(slp)); } vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); (void) VOP_FSYNC(vp, MNT_LAZY, td); VOP_UNLOCK(vp, 0); vn_finished_write(mp); BO_LOCK(*bo); if (((*bo)->bo_flag & BO_ONWORKLST) != 0) { /* * Put us back on the worklist. The worklist * routine will remove us from our current * position and then add us back in at a later * position. */ vn_syncer_add_to_worklist(*bo, syncdelay); } BO_UNLOCK(*bo); vdrop(vp); mtx_lock(&sync_mtx); return (0); } static int first_printf = 1; /* * System filesystem synchronizer daemon. */ static void sched_sync(void) { struct synclist *next, *slp; struct bufobj *bo; long starttime; struct thread *td = curthread; int last_work_seen; int net_worklist_len; int syncer_final_iter; int error; last_work_seen = 0; syncer_final_iter = 0; syncer_state = SYNCER_RUNNING; starttime = time_uptime; td->td_pflags |= TDP_NORUNNINGBUF; EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc, SHUTDOWN_PRI_LAST); mtx_lock(&sync_mtx); for (;;) { if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter == 0) { mtx_unlock(&sync_mtx); kproc_suspend_check(td->td_proc); mtx_lock(&sync_mtx); } net_worklist_len = syncer_worklist_len - sync_vnode_count; if (syncer_state != SYNCER_RUNNING && starttime != time_uptime) { if (first_printf) { printf("\nSyncing disks, vnodes remaining... "); first_printf = 0; } printf("%d ", net_worklist_len); } starttime = time_uptime; /* * Push files whose dirty time has expired. Be careful * of interrupt race on slp queue. * * Skip over empty worklist slots when shutting down. */ do { slp = &syncer_workitem_pending[syncer_delayno]; syncer_delayno += 1; if (syncer_delayno == syncer_maxdelay) syncer_delayno = 0; next = &syncer_workitem_pending[syncer_delayno]; /* * If the worklist has wrapped since the * it was emptied of all but syncer vnodes, * switch to the FINAL_DELAY state and run * for one more second. */ if (syncer_state == SYNCER_SHUTTING_DOWN && net_worklist_len == 0 && last_work_seen == syncer_delayno) { syncer_state = SYNCER_FINAL_DELAY; syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP; } } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) && syncer_worklist_len > 0); /* * Keep track of the last time there was anything * on the worklist other than syncer vnodes. * Return to the SHUTTING_DOWN state if any * new work appears. */ if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING) last_work_seen = syncer_delayno; if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY) syncer_state = SYNCER_SHUTTING_DOWN; while (!LIST_EMPTY(slp)) { error = sync_vnode(slp, &bo, td); if (error == 1) { LIST_REMOVE(bo, bo_synclist); LIST_INSERT_HEAD(next, bo, bo_synclist); continue; } if (first_printf == 0) { /* * Drop the sync mutex, because some watchdog * drivers need to sleep while patting */ mtx_unlock(&sync_mtx); wdog_kern_pat(WD_LASTVAL); mtx_lock(&sync_mtx); } } if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0) syncer_final_iter--; /* * The variable rushjob allows the kernel to speed up the * processing of the filesystem syncer process. A rushjob * value of N tells the filesystem syncer to process the next * N seconds worth of work on its queue ASAP. Currently rushjob * is used by the soft update code to speed up the filesystem * syncer process when the incore state is getting so far * ahead of the disk that the kernel memory pool is being * threatened with exhaustion. */ if (rushjob > 0) { rushjob -= 1; continue; } /* * Just sleep for a short period of time between * iterations when shutting down to allow some I/O * to happen. * * If it has taken us less than a second to process the * current work, then wait. Otherwise start right over * again. We can still lose time if any single round * takes more than two seconds, but it does not really * matter as we are just trying to generally pace the * filesystem activity. */ if (syncer_state != SYNCER_RUNNING || time_uptime == starttime) { thread_lock(td); sched_prio(td, PPAUSE); thread_unlock(td); } if (syncer_state != SYNCER_RUNNING) cv_timedwait(&sync_wakeup, &sync_mtx, hz / SYNCER_SHUTDOWN_SPEEDUP); else if (time_uptime == starttime) cv_timedwait(&sync_wakeup, &sync_mtx, hz); } } /* * Request the syncer daemon to speed up its work. * We never push it to speed up more than half of its * normal turn time, otherwise it could take over the cpu. */ int speedup_syncer(void) { int ret = 0; mtx_lock(&sync_mtx); if (rushjob < syncdelay / 2) { rushjob += 1; stat_rush_requests += 1; ret = 1; } mtx_unlock(&sync_mtx); cv_broadcast(&sync_wakeup); return (ret); } /* * Tell the syncer to speed up its work and run though its work * list several times, then tell it to shut down. */ static void syncer_shutdown(void *arg, int howto) { if (howto & RB_NOSYNC) return; mtx_lock(&sync_mtx); syncer_state = SYNCER_SHUTTING_DOWN; rushjob = 0; mtx_unlock(&sync_mtx); cv_broadcast(&sync_wakeup); kproc_shutdown(arg, howto); } void syncer_suspend(void) { syncer_shutdown(updateproc, 0); } void syncer_resume(void) { mtx_lock(&sync_mtx); first_printf = 1; syncer_state = SYNCER_RUNNING; mtx_unlock(&sync_mtx); cv_broadcast(&sync_wakeup); kproc_resume(updateproc); } /* * Reassign a buffer from one vnode to another. * Used to assign file specific control information * (indirect blocks) to the vnode to which they belong. */ void reassignbuf(struct buf *bp) { struct vnode *vp; struct bufobj *bo; int delay; #ifdef INVARIANTS struct bufv *bv; #endif vp = bp->b_vp; bo = bp->b_bufobj; ++reassignbufcalls; CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); /* * B_PAGING flagged buffers cannot be reassigned because their vp * is not fully linked in. */ if (bp->b_flags & B_PAGING) panic("cannot reassign paging buffer"); /* * Delete from old vnode list, if on one. */ BO_LOCK(bo); if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) buf_vlist_remove(bp); else panic("reassignbuf: Buffer %p not on queue.", bp); /* * If dirty, put on list of dirty buffers; otherwise insert onto list * of clean buffers. */ if (bp->b_flags & B_DELWRI) { if ((bo->bo_flag & BO_ONWORKLST) == 0) { switch (vp->v_type) { case VDIR: delay = dirdelay; break; case VCHR: delay = metadelay; break; default: delay = filedelay; } vn_syncer_add_to_worklist(bo, delay); } buf_vlist_add(bp, bo, BX_VNDIRTY); } else { buf_vlist_add(bp, bo, BX_VNCLEAN); if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { mtx_lock(&sync_mtx); LIST_REMOVE(bo, bo_synclist); syncer_worklist_len--; mtx_unlock(&sync_mtx); bo->bo_flag &= ~BO_ONWORKLST; } } #ifdef INVARIANTS bv = &bo->bo_clean; bp = TAILQ_FIRST(&bv->bv_hd); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); bp = TAILQ_LAST(&bv->bv_hd, buflists); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); bv = &bo->bo_dirty; bp = TAILQ_FIRST(&bv->bv_hd); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); bp = TAILQ_LAST(&bv->bv_hd, buflists); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); #endif BO_UNLOCK(bo); } /* * A temporary hack until refcount_* APIs are sorted out. */ static __inline int vfs_refcount_acquire_if_not_zero(volatile u_int *count) { u_int old; for (;;) { old = *count; if (old == 0) return (0); if (atomic_cmpset_int(count, old, old + 1)) return (1); } } static __inline int vfs_refcount_release_if_not_last(volatile u_int *count) { u_int old; for (;;) { old = *count; if (old == 1) return (0); if (atomic_cmpset_int(count, old, old - 1)) return (1); } } static void v_init_counters(struct vnode *vp) { VNASSERT(vp->v_type == VNON && vp->v_data == NULL && vp->v_iflag == 0, vp, ("%s called for an initialized vnode", __FUNCTION__)); ASSERT_VI_UNLOCKED(vp, __FUNCTION__); refcount_init(&vp->v_holdcnt, 1); refcount_init(&vp->v_usecount, 1); } static void v_incr_usecount_locked(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); if ((vp->v_iflag & VI_OWEINACT) != 0) { VNASSERT(vp->v_usecount == 0, vp, ("vnode with usecount and VI_OWEINACT set")); vp->v_iflag &= ~VI_OWEINACT; } refcount_acquire(&vp->v_usecount); v_incr_devcount(vp); } /* * Increment the use and hold counts on the vnode, taking care to reference * the driver's usecount if this is a chardev. The _vhold() will remove * the vnode from the free list if it is presently free. */ static void v_incr_usecount(struct vnode *vp) { ASSERT_VI_UNLOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if (vp->v_type != VCHR && vfs_refcount_acquire_if_not_zero(&vp->v_usecount)) { VNASSERT((vp->v_iflag & VI_OWEINACT) == 0, vp, ("vnode with usecount and VI_OWEINACT set")); } else { VI_LOCK(vp); v_incr_usecount_locked(vp); VI_UNLOCK(vp); } } /* * Increment si_usecount of the associated device, if any. */ static void v_incr_devcount(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __FUNCTION__); if (vp->v_type == VCHR && vp->v_rdev != NULL) { dev_lock(); vp->v_rdev->si_usecount++; dev_unlock(); } } /* * Decrement si_usecount of the associated device, if any. */ static void v_decr_devcount(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __FUNCTION__); if (vp->v_type == VCHR && vp->v_rdev != NULL) { dev_lock(); vp->v_rdev->si_usecount--; dev_unlock(); } } /* * Grab a particular vnode from the free list, increment its * reference count and lock it. VI_DOOMED is set if the vnode * is being destroyed. Only callers who specify LK_RETRY will * see doomed vnodes. If inactive processing was delayed in * vput try to do it here. * * Notes on lockless counter manipulation: * _vhold, vputx and other routines make various decisions based * on either holdcnt or usecount being 0. As long as either counter * is not transitioning 0->1 nor 1->0, the manipulation can be done * with atomic operations. Otherwise the interlock is taken covering * both the atomic and additional actions. */ int vget(struct vnode *vp, int flags, struct thread *td) { int error, oweinact; VNASSERT((flags & LK_TYPE_MASK) != 0, vp, ("vget: invalid lock operation")); if ((flags & LK_INTERLOCK) != 0) ASSERT_VI_LOCKED(vp, __func__); else ASSERT_VI_UNLOCKED(vp, __func__); if ((flags & LK_VNHELD) != 0) VNASSERT((vp->v_holdcnt > 0), vp, ("vget: LK_VNHELD passed but vnode not held")); CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags); if ((flags & LK_VNHELD) == 0) _vhold(vp, (flags & LK_INTERLOCK) != 0); if ((error = vn_lock(vp, flags)) != 0) { vdrop(vp); CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__, vp); return (error); } if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0) panic("vget: vn_lock failed to return ENOENT\n"); /* * We don't guarantee that any particular close will * trigger inactive processing so just make a best effort * here at preventing a reference to a removed file. If * we don't succeed no harm is done. * * Upgrade our holdcnt to a usecount. */ if (vp->v_type == VCHR || !vfs_refcount_acquire_if_not_zero(&vp->v_usecount)) { VI_LOCK(vp); if ((vp->v_iflag & VI_OWEINACT) == 0) { oweinact = 0; } else { oweinact = 1; vp->v_iflag &= ~VI_OWEINACT; } refcount_acquire(&vp->v_usecount); v_incr_devcount(vp); if (oweinact && VOP_ISLOCKED(vp) == LK_EXCLUSIVE && (flags & LK_NOWAIT) == 0) vinactive(vp, td); VI_UNLOCK(vp); } return (0); } /* * Increase the reference count of a vnode. */ void vref(struct vnode *vp) { CTR2(KTR_VFS, "%s: vp %p", __func__, vp); _vhold(vp, false); v_incr_usecount(vp); } void vrefl(struct vnode *vp) { CTR2(KTR_VFS, "%s: vp %p", __func__, vp); _vhold(vp, true); v_incr_usecount_locked(vp); } /* * Return reference count of a vnode. * * The results of this call are only guaranteed when some mechanism is used to * stop other processes from gaining references to the vnode. This may be the * case if the caller holds the only reference. This is also useful when stale * data is acceptable as race conditions may be accounted for by some other * means. */ int vrefcnt(struct vnode *vp) { return (vp->v_usecount); } #define VPUTX_VRELE 1 #define VPUTX_VPUT 2 #define VPUTX_VUNREF 3 /* * Decrement the use and hold counts for a vnode. * * See an explanation near vget() as to why atomic operation is safe. */ static void vputx(struct vnode *vp, int func) { int error; KASSERT(vp != NULL, ("vputx: null vp")); if (func == VPUTX_VUNREF) ASSERT_VOP_LOCKED(vp, "vunref"); else if (func == VPUTX_VPUT) ASSERT_VOP_LOCKED(vp, "vput"); else KASSERT(func == VPUTX_VRELE, ("vputx: wrong func")); ASSERT_VI_UNLOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if (vp->v_type != VCHR && vfs_refcount_release_if_not_last(&vp->v_usecount)) { if (func == VPUTX_VPUT) VOP_UNLOCK(vp, 0); vdrop(vp); return; } VI_LOCK(vp); /* * We want to hold the vnode until the inactive finishes to * prevent vgone() races. We drop the use count here and the * hold count below when we're done. */ if (!refcount_release(&vp->v_usecount) || (vp->v_iflag & VI_DOINGINACT)) { if (func == VPUTX_VPUT) VOP_UNLOCK(vp, 0); v_decr_devcount(vp); vdropl(vp); return; } v_decr_devcount(vp); error = 0; if (vp->v_usecount != 0) { vn_printf(vp, "vputx: usecount not zero for vnode "); panic("vputx: usecount not zero"); } CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp); /* * We must call VOP_INACTIVE with the node locked. Mark * as VI_DOINGINACT to avoid recursion. */ vp->v_iflag |= VI_OWEINACT; switch (func) { case VPUTX_VRELE: error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK); VI_LOCK(vp); break; case VPUTX_VPUT: if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK | LK_NOWAIT); VI_LOCK(vp); } break; case VPUTX_VUNREF: if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK); VI_LOCK(vp); } break; } VNASSERT(vp->v_usecount == 0 || (vp->v_iflag & VI_OWEINACT) == 0, vp, ("vnode with usecount and VI_OWEINACT set")); if (error == 0) { if (vp->v_iflag & VI_OWEINACT) vinactive(vp, curthread); if (func != VPUTX_VUNREF) VOP_UNLOCK(vp, 0); } vdropl(vp); } /* * Vnode put/release. * If count drops to zero, call inactive routine and return to freelist. */ void vrele(struct vnode *vp) { vputx(vp, VPUTX_VRELE); } /* * Release an already locked vnode. This give the same effects as * unlock+vrele(), but takes less time and avoids releasing and * re-aquiring the lock (as vrele() acquires the lock internally.) */ void vput(struct vnode *vp) { vputx(vp, VPUTX_VPUT); } /* * Release an exclusively locked vnode. Do not unlock the vnode lock. */ void vunref(struct vnode *vp) { vputx(vp, VPUTX_VUNREF); } /* * Increase the hold count and activate if this is the first reference. */ void _vhold(struct vnode *vp, bool locked) { struct mount *mp; if (locked) ASSERT_VI_LOCKED(vp, __func__); else ASSERT_VI_UNLOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if (!locked && vfs_refcount_acquire_if_not_zero(&vp->v_holdcnt)) { VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("_vhold: vnode with holdcnt is free")); return; } if (!locked) VI_LOCK(vp); if ((vp->v_iflag & VI_FREE) == 0) { refcount_acquire(&vp->v_holdcnt); if (!locked) VI_UNLOCK(vp); return; } VNASSERT(vp->v_holdcnt == 0, vp, ("%s: wrong hold count", __func__)); VNASSERT(vp->v_op != NULL, vp, ("%s: vnode already reclaimed.", __func__)); /* * Remove a vnode from the free list, mark it as in use, * and put it on the active list. */ mtx_lock(&vnode_free_list_mtx); TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist); freevnodes--; vp->v_iflag &= ~VI_FREE; KASSERT((vp->v_iflag & VI_ACTIVE) == 0, ("Activating already active vnode")); vp->v_iflag |= VI_ACTIVE; mp = vp->v_mount; TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist); mp->mnt_activevnodelistsize++; mtx_unlock(&vnode_free_list_mtx); refcount_acquire(&vp->v_holdcnt); if (!locked) VI_UNLOCK(vp); } /* * Drop the hold count of the vnode. If this is the last reference to * the vnode we place it on the free list unless it has been vgone'd * (marked VI_DOOMED) in which case we will free it. * * Because the vnode vm object keeps a hold reference on the vnode if * there is at least one resident non-cached page, the vnode cannot * leave the active list without the page cleanup done. */ void _vdrop(struct vnode *vp, bool locked) { struct bufobj *bo; struct mount *mp; int active; if (locked) ASSERT_VI_LOCKED(vp, __func__); else ASSERT_VI_UNLOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if ((int)vp->v_holdcnt <= 0) panic("vdrop: holdcnt %d", vp->v_holdcnt); if (vfs_refcount_release_if_not_last(&vp->v_holdcnt)) { if (locked) VI_UNLOCK(vp); return; } if (!locked) VI_LOCK(vp); if (refcount_release(&vp->v_holdcnt) == 0) { VI_UNLOCK(vp); return; } if ((vp->v_iflag & VI_DOOMED) == 0) { /* * Mark a vnode as free: remove it from its active list * and put it up for recycling on the freelist. */ VNASSERT(vp->v_op != NULL, vp, ("vdropl: vnode already reclaimed.")); VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free")); VNASSERT(vp->v_holdcnt == 0, vp, ("vdropl: freeing when we shouldn't")); active = vp->v_iflag & VI_ACTIVE; if ((vp->v_iflag & VI_OWEINACT) == 0) { vp->v_iflag &= ~VI_ACTIVE; mp = vp->v_mount; mtx_lock(&vnode_free_list_mtx); if (active) { TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist); mp->mnt_activevnodelistsize--; } TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist); freevnodes++; vp->v_iflag |= VI_FREE; mtx_unlock(&vnode_free_list_mtx); } else { atomic_add_long(&free_owe_inact, 1); } VI_UNLOCK(vp); return; } /* * The vnode has been marked for destruction, so free it. * * The vnode will be returned to the zone where it will * normally remain until it is needed for another vnode. We * need to cleanup (or verify that the cleanup has already * been done) any residual data left from its current use * so as not to contaminate the freshly allocated vnode. */ CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp); atomic_subtract_long(&numvnodes, 1); bo = &vp->v_bufobj; VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("cleaned vnode still on the free list.")); VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't")); VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count")); VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count")); VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count")); VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's")); VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0")); VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp, ("clean blk trie not empty")); VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0")); VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp, ("dirty blk trie not empty")); VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst")); VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src")); VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for ..")); VNASSERT(TAILQ_EMPTY(&vp->v_rl.rl_waiters), vp, ("Dangling rangelock waiters")); VI_UNLOCK(vp); #ifdef MAC mac_vnode_destroy(vp); #endif if (vp->v_pollinfo != NULL) { destroy_vpollinfo(vp->v_pollinfo); vp->v_pollinfo = NULL; } #ifdef INVARIANTS /* XXX Elsewhere we detect an already freed vnode via NULL v_op. */ vp->v_op = NULL; #endif bzero(&vp->v_un, sizeof(vp->v_un)); vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0; vp->v_iflag = 0; vp->v_vflag = 0; bo->bo_flag = 0; uma_zfree(vnode_zone, vp); } /* * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT * flags. DOINGINACT prevents us from recursing in calls to vinactive. * OWEINACT tracks whether a vnode missed a call to inactive due to a * failed lock upgrade. */ void vinactive(struct vnode *vp, struct thread *td) { struct vm_object *obj; ASSERT_VOP_ELOCKED(vp, "vinactive"); ASSERT_VI_LOCKED(vp, "vinactive"); VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp, ("vinactive: recursed on VI_DOINGINACT")); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); vp->v_iflag |= VI_DOINGINACT; vp->v_iflag &= ~VI_OWEINACT; VI_UNLOCK(vp); /* * Before moving off the active list, we must be sure that any * modified pages are converted into the vnode's dirty * buffers, since these will no longer be checked once the * vnode is on the inactive list. * * The write-out of the dirty pages is asynchronous. At the * point that VOP_INACTIVE() is called, there could still be * pending I/O and dirty pages in the object. */ obj = vp->v_object; if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) { VM_OBJECT_WLOCK(obj); vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC); VM_OBJECT_WUNLOCK(obj); } VOP_INACTIVE(vp, td); VI_LOCK(vp); VNASSERT(vp->v_iflag & VI_DOINGINACT, vp, ("vinactive: lost VI_DOINGINACT")); vp->v_iflag &= ~VI_DOINGINACT; } /* * Remove any vnodes in the vnode table belonging to mount point mp. * * If FORCECLOSE is not specified, there should not be any active ones, * return error if any are found (nb: this is a user error, not a * system error). If FORCECLOSE is specified, detach any active vnodes * that are found. * * If WRITECLOSE is set, only flush out regular file vnodes open for * writing. * * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped. * * `rootrefs' specifies the base reference count for the root vnode * of this filesystem. The root vnode is considered busy if its * v_usecount exceeds this value. On a successful return, vflush(, td) * will call vrele() on the root vnode exactly rootrefs times. * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must * be zero. */ #ifdef DIAGNOSTIC static int busyprt = 0; /* print out busy vnodes */ SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes"); #endif int vflush(struct mount *mp, int rootrefs, int flags, struct thread *td) { struct vnode *vp, *mvp, *rootvp = NULL; struct vattr vattr; int busy = 0, error; CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp, rootrefs, flags); if (rootrefs > 0) { KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, ("vflush: bad args")); /* * Get the filesystem root vnode. We can vput() it * immediately, since with rootrefs > 0, it won't go away. */ if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) { CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d", __func__, error); return (error); } vput(rootvp); } loop: MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { vholdl(vp); error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE); if (error) { vdrop(vp); MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); goto loop; } /* * Skip over a vnodes marked VV_SYSTEM. */ if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) { VOP_UNLOCK(vp, 0); vdrop(vp); continue; } /* * If WRITECLOSE is set, flush out unlinked but still open * files (even if open only for reading) and regular file * vnodes open for writing. */ if (flags & WRITECLOSE) { if (vp->v_object != NULL) { VM_OBJECT_WLOCK(vp->v_object); vm_object_page_clean(vp->v_object, 0, 0, 0); VM_OBJECT_WUNLOCK(vp->v_object); } error = VOP_FSYNC(vp, MNT_WAIT, td); if (error != 0) { VOP_UNLOCK(vp, 0); vdrop(vp); MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); return (error); } error = VOP_GETATTR(vp, &vattr, td->td_ucred); VI_LOCK(vp); if ((vp->v_type == VNON || (error == 0 && vattr.va_nlink > 0)) && (vp->v_writecount == 0 || vp->v_type != VREG)) { VOP_UNLOCK(vp, 0); vdropl(vp); continue; } } else VI_LOCK(vp); /* * With v_usecount == 0, all we need to do is clear out the * vnode data structures and we are done. * * If FORCECLOSE is set, forcibly close the vnode. */ if (vp->v_usecount == 0 || (flags & FORCECLOSE)) { vgonel(vp); } else { busy++; #ifdef DIAGNOSTIC if (busyprt) vn_printf(vp, "vflush: busy vnode "); #endif } VOP_UNLOCK(vp, 0); vdropl(vp); } if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { /* * If just the root vnode is busy, and if its refcount * is equal to `rootrefs', then go ahead and kill it. */ VI_LOCK(rootvp); KASSERT(busy > 0, ("vflush: not busy")); VNASSERT(rootvp->v_usecount >= rootrefs, rootvp, ("vflush: usecount %d < rootrefs %d", rootvp->v_usecount, rootrefs)); if (busy == 1 && rootvp->v_usecount == rootrefs) { VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK); vgone(rootvp); VOP_UNLOCK(rootvp, 0); busy = 0; } else VI_UNLOCK(rootvp); } if (busy) { CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__, busy); return (EBUSY); } for (; rootrefs > 0; rootrefs--) vrele(rootvp); return (0); } /* * Recycle an unused vnode to the front of the free list. */ int vrecycle(struct vnode *vp) { int recycled; ASSERT_VOP_ELOCKED(vp, "vrecycle"); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); recycled = 0; VI_LOCK(vp); if (vp->v_usecount == 0) { recycled = 1; vgonel(vp); } VI_UNLOCK(vp); return (recycled); } /* * Eliminate all activity associated with a vnode * in preparation for reuse. */ void vgone(struct vnode *vp) { VI_LOCK(vp); vgonel(vp); VI_UNLOCK(vp); } static void notify_lowervp_vfs_dummy(struct mount *mp __unused, struct vnode *lowervp __unused) { } /* * Notify upper mounts about reclaimed or unlinked vnode. */ void vfs_notify_upper(struct vnode *vp, int event) { static struct vfsops vgonel_vfsops = { .vfs_reclaim_lowervp = notify_lowervp_vfs_dummy, .vfs_unlink_lowervp = notify_lowervp_vfs_dummy, }; struct mount *mp, *ump, *mmp; mp = vp->v_mount; if (mp == NULL) return; MNT_ILOCK(mp); if (TAILQ_EMPTY(&mp->mnt_uppers)) goto unlock; MNT_IUNLOCK(mp); mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO); mmp->mnt_op = &vgonel_vfsops; mmp->mnt_kern_flag |= MNTK_MARKER; MNT_ILOCK(mp); mp->mnt_kern_flag |= MNTK_VGONE_UPPER; for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) { if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) { ump = TAILQ_NEXT(ump, mnt_upper_link); continue; } TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link); MNT_IUNLOCK(mp); switch (event) { case VFS_NOTIFY_UPPER_RECLAIM: VFS_RECLAIM_LOWERVP(ump, vp); break; case VFS_NOTIFY_UPPER_UNLINK: VFS_UNLINK_LOWERVP(ump, vp); break; default: KASSERT(0, ("invalid event %d", event)); break; } MNT_ILOCK(mp); ump = TAILQ_NEXT(mmp, mnt_upper_link); TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link); } free(mmp, M_TEMP); mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER; if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) { mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER; wakeup(&mp->mnt_uppers); } unlock: MNT_IUNLOCK(mp); } /* * vgone, with the vp interlock held. */ static void vgonel(struct vnode *vp) { struct thread *td; int oweinact; int active; struct mount *mp; ASSERT_VOP_ELOCKED(vp, "vgonel"); ASSERT_VI_LOCKED(vp, "vgonel"); VNASSERT(vp->v_holdcnt, vp, ("vgonel: vp %p has no reference.", vp)); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); td = curthread; /* * Don't vgonel if we're already doomed. */ if (vp->v_iflag & VI_DOOMED) return; vp->v_iflag |= VI_DOOMED; /* * Check to see if the vnode is in use. If so, we have to call * VOP_CLOSE() and VOP_INACTIVE(). */ active = vp->v_usecount; oweinact = (vp->v_iflag & VI_OWEINACT); VI_UNLOCK(vp); vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM); /* * If purging an active vnode, it must be closed and * deactivated before being reclaimed. */ if (active) VOP_CLOSE(vp, FNONBLOCK, NOCRED, td); if (oweinact || active) { VI_LOCK(vp); if ((vp->v_iflag & VI_DOINGINACT) == 0) vinactive(vp, td); VI_UNLOCK(vp); } if (vp->v_type == VSOCK) vfs_unp_reclaim(vp); /* * Clean out any buffers associated with the vnode. * If the flush fails, just toss the buffers. */ mp = NULL; if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd)) (void) vn_start_secondary_write(vp, &mp, V_WAIT); if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) { while (vinvalbuf(vp, 0, 0, 0) != 0) ; } BO_LOCK(&vp->v_bufobj); KASSERT(TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd) && vp->v_bufobj.bo_dirty.bv_cnt == 0 && TAILQ_EMPTY(&vp->v_bufobj.bo_clean.bv_hd) && vp->v_bufobj.bo_clean.bv_cnt == 0, ("vp %p bufobj not invalidated", vp)); /* * For VMIO bufobj, BO_DEAD is set in vm_object_terminate() * after the object's page queue is flushed. */ if (vp->v_bufobj.bo_object == NULL) vp->v_bufobj.bo_flag |= BO_DEAD; BO_UNLOCK(&vp->v_bufobj); /* * Reclaim the vnode. */ if (VOP_RECLAIM(vp, td)) panic("vgone: cannot reclaim"); if (mp != NULL) vn_finished_secondary_write(mp); VNASSERT(vp->v_object == NULL, vp, ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag)); /* * Clear the advisory locks and wake up waiting threads. */ (void)VOP_ADVLOCKPURGE(vp); vp->v_lockf = NULL; /* * Delete from old mount point vnode list. */ delmntque(vp); cache_purge(vp); /* * Done with purge, reset to the standard lock and invalidate * the vnode. */ VI_LOCK(vp); vp->v_vnlock = &vp->v_lock; vp->v_op = &dead_vnodeops; vp->v_tag = "none"; vp->v_type = VBAD; } /* * Calculate the total number of references to a special device. */ int vcount(struct vnode *vp) { int count; dev_lock(); count = vp->v_rdev->si_usecount; dev_unlock(); return (count); } /* * Same as above, but using the struct cdev *as argument */ int count_dev(struct cdev *dev) { int count; dev_lock(); count = dev->si_usecount; dev_unlock(); return(count); } /* * Print out a description of a vnode. */ static char *typename[] = {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD", "VMARKER"}; void vn_printf(struct vnode *vp, const char *fmt, ...) { va_list ap; char buf[256], buf2[16]; u_long flags; va_start(ap, fmt); vprintf(fmt, ap); va_end(ap); printf("%p: ", (void *)vp); printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]); printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n", vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere); buf[0] = '\0'; buf[1] = '\0'; if (vp->v_vflag & VV_ROOT) strlcat(buf, "|VV_ROOT", sizeof(buf)); if (vp->v_vflag & VV_ISTTY) strlcat(buf, "|VV_ISTTY", sizeof(buf)); if (vp->v_vflag & VV_NOSYNC) strlcat(buf, "|VV_NOSYNC", sizeof(buf)); if (vp->v_vflag & VV_ETERNALDEV) strlcat(buf, "|VV_ETERNALDEV", sizeof(buf)); if (vp->v_vflag & VV_CACHEDLABEL) strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf)); if (vp->v_vflag & VV_TEXT) strlcat(buf, "|VV_TEXT", sizeof(buf)); if (vp->v_vflag & VV_COPYONWRITE) strlcat(buf, "|VV_COPYONWRITE", sizeof(buf)); if (vp->v_vflag & VV_SYSTEM) strlcat(buf, "|VV_SYSTEM", sizeof(buf)); if (vp->v_vflag & VV_PROCDEP) strlcat(buf, "|VV_PROCDEP", sizeof(buf)); if (vp->v_vflag & VV_NOKNOTE) strlcat(buf, "|VV_NOKNOTE", sizeof(buf)); if (vp->v_vflag & VV_DELETED) strlcat(buf, "|VV_DELETED", sizeof(buf)); if (vp->v_vflag & VV_MD) strlcat(buf, "|VV_MD", sizeof(buf)); if (vp->v_vflag & VV_FORCEINSMQ) strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf)); flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV | VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP | VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ); if (flags != 0) { snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags); strlcat(buf, buf2, sizeof(buf)); } if (vp->v_iflag & VI_MOUNT) strlcat(buf, "|VI_MOUNT", sizeof(buf)); if (vp->v_iflag & VI_DOOMED) strlcat(buf, "|VI_DOOMED", sizeof(buf)); if (vp->v_iflag & VI_FREE) strlcat(buf, "|VI_FREE", sizeof(buf)); if (vp->v_iflag & VI_ACTIVE) strlcat(buf, "|VI_ACTIVE", sizeof(buf)); if (vp->v_iflag & VI_DOINGINACT) strlcat(buf, "|VI_DOINGINACT", sizeof(buf)); if (vp->v_iflag & VI_OWEINACT) strlcat(buf, "|VI_OWEINACT", sizeof(buf)); flags = vp->v_iflag & ~(VI_MOUNT | VI_DOOMED | VI_FREE | VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT); if (flags != 0) { snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags); strlcat(buf, buf2, sizeof(buf)); } printf(" flags (%s)\n", buf + 1); if (mtx_owned(VI_MTX(vp))) printf(" VI_LOCKed"); if (vp->v_object != NULL) printf(" v_object %p ref %d pages %d " "cleanbuf %d dirtybuf %d\n", vp->v_object, vp->v_object->ref_count, vp->v_object->resident_page_count, vp->v_bufobj.bo_clean.bv_cnt, vp->v_bufobj.bo_dirty.bv_cnt); printf(" "); lockmgr_printinfo(vp->v_vnlock); if (vp->v_data != NULL) VOP_PRINT(vp); } #ifdef DDB /* * List all of the locked vnodes in the system. * Called when debugging the kernel. */ DB_SHOW_COMMAND(lockedvnods, lockedvnodes) { struct mount *mp; struct vnode *vp; /* * Note: because this is DDB, we can't obey the locking semantics * for these structures, which means we could catch an inconsistent * state and dereference a nasty pointer. Not much to be done * about that. */ db_printf("Locked vnodes\n"); TAILQ_FOREACH(mp, &mountlist, mnt_list) { TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (vp->v_type != VMARKER && VOP_ISLOCKED(vp)) vn_printf(vp, "vnode "); } } } /* * Show details about the given vnode. */ DB_SHOW_COMMAND(vnode, db_show_vnode) { struct vnode *vp; if (!have_addr) return; vp = (struct vnode *)addr; vn_printf(vp, "vnode "); } /* * Show details about the given mount point. */ DB_SHOW_COMMAND(mount, db_show_mount) { struct mount *mp; struct vfsopt *opt; struct statfs *sp; struct vnode *vp; char buf[512]; uint64_t mflags; u_int flags; if (!have_addr) { /* No address given, print short info about all mount points. */ TAILQ_FOREACH(mp, &mountlist, mnt_list) { db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename); if (db_pager_quit) break; } db_printf("\nMore info: show mount \n"); return; } mp = (struct mount *)addr; db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename); buf[0] = '\0'; mflags = mp->mnt_flag; #define MNT_FLAG(flag) do { \ if (mflags & (flag)) { \ if (buf[0] != '\0') \ strlcat(buf, ", ", sizeof(buf)); \ strlcat(buf, (#flag) + 4, sizeof(buf)); \ mflags &= ~(flag); \ } \ } while (0) MNT_FLAG(MNT_RDONLY); MNT_FLAG(MNT_SYNCHRONOUS); MNT_FLAG(MNT_NOEXEC); MNT_FLAG(MNT_NOSUID); MNT_FLAG(MNT_NFS4ACLS); MNT_FLAG(MNT_UNION); MNT_FLAG(MNT_ASYNC); MNT_FLAG(MNT_SUIDDIR); MNT_FLAG(MNT_SOFTDEP); MNT_FLAG(MNT_NOSYMFOLLOW); MNT_FLAG(MNT_GJOURNAL); MNT_FLAG(MNT_MULTILABEL); MNT_FLAG(MNT_ACLS); MNT_FLAG(MNT_NOATIME); MNT_FLAG(MNT_NOCLUSTERR); MNT_FLAG(MNT_NOCLUSTERW); MNT_FLAG(MNT_SUJ); MNT_FLAG(MNT_EXRDONLY); MNT_FLAG(MNT_EXPORTED); MNT_FLAG(MNT_DEFEXPORTED); MNT_FLAG(MNT_EXPORTANON); MNT_FLAG(MNT_EXKERB); MNT_FLAG(MNT_EXPUBLIC); MNT_FLAG(MNT_LOCAL); MNT_FLAG(MNT_QUOTA); MNT_FLAG(MNT_ROOTFS); MNT_FLAG(MNT_USER); MNT_FLAG(MNT_IGNORE); MNT_FLAG(MNT_UPDATE); MNT_FLAG(MNT_DELEXPORT); MNT_FLAG(MNT_RELOAD); MNT_FLAG(MNT_FORCE); MNT_FLAG(MNT_SNAPSHOT); MNT_FLAG(MNT_BYFSID); #undef MNT_FLAG if (mflags != 0) { if (buf[0] != '\0') strlcat(buf, ", ", sizeof(buf)); snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), "0x%016jx", mflags); } db_printf(" mnt_flag = %s\n", buf); buf[0] = '\0'; flags = mp->mnt_kern_flag; #define MNT_KERN_FLAG(flag) do { \ if (flags & (flag)) { \ if (buf[0] != '\0') \ strlcat(buf, ", ", sizeof(buf)); \ strlcat(buf, (#flag) + 5, sizeof(buf)); \ flags &= ~(flag); \ } \ } while (0) MNT_KERN_FLAG(MNTK_UNMOUNTF); MNT_KERN_FLAG(MNTK_ASYNC); MNT_KERN_FLAG(MNTK_SOFTDEP); MNT_KERN_FLAG(MNTK_NOINSMNTQ); MNT_KERN_FLAG(MNTK_DRAINING); MNT_KERN_FLAG(MNTK_REFEXPIRE); MNT_KERN_FLAG(MNTK_EXTENDED_SHARED); MNT_KERN_FLAG(MNTK_SHARED_WRITES); MNT_KERN_FLAG(MNTK_NO_IOPF); MNT_KERN_FLAG(MNTK_VGONE_UPPER); MNT_KERN_FLAG(MNTK_VGONE_WAITER); MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT); MNT_KERN_FLAG(MNTK_MARKER); MNT_KERN_FLAG(MNTK_USES_BCACHE); MNT_KERN_FLAG(MNTK_NOASYNC); MNT_KERN_FLAG(MNTK_UNMOUNT); MNT_KERN_FLAG(MNTK_MWAIT); MNT_KERN_FLAG(MNTK_SUSPEND); MNT_KERN_FLAG(MNTK_SUSPEND2); MNT_KERN_FLAG(MNTK_SUSPENDED); MNT_KERN_FLAG(MNTK_LOOKUP_SHARED); MNT_KERN_FLAG(MNTK_NOKNOTE); #undef MNT_KERN_FLAG if (flags != 0) { if (buf[0] != '\0') strlcat(buf, ", ", sizeof(buf)); snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), "0x%08x", flags); } db_printf(" mnt_kern_flag = %s\n", buf); db_printf(" mnt_opt = "); opt = TAILQ_FIRST(mp->mnt_opt); if (opt != NULL) { db_printf("%s", opt->name); opt = TAILQ_NEXT(opt, link); while (opt != NULL) { db_printf(", %s", opt->name); opt = TAILQ_NEXT(opt, link); } } db_printf("\n"); sp = &mp->mnt_stat; db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx " "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju " "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju " "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n", (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags, (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize, (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree, (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files, (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites, (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads, (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax, (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]); db_printf(" mnt_cred = { uid=%u ruid=%u", (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid); if (jailed(mp->mnt_cred)) db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id); db_printf(" }\n"); db_printf(" mnt_ref = %d\n", mp->mnt_ref); db_printf(" mnt_gen = %d\n", mp->mnt_gen); db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize); db_printf(" mnt_activevnodelistsize = %d\n", mp->mnt_activevnodelistsize); db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount); db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen); db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max); db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed); db_printf(" mnt_lockref = %d\n", mp->mnt_lockref); db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes); db_printf(" mnt_secondary_accwrites = %d\n", mp->mnt_secondary_accwrites); db_printf(" mnt_gjprovider = %s\n", mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL"); db_printf("\n\nList of active vnodes\n"); TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) { if (vp->v_type != VMARKER) { vn_printf(vp, "vnode "); if (db_pager_quit) break; } } db_printf("\n\nList of inactive vnodes\n"); TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) { vn_printf(vp, "vnode "); if (db_pager_quit) break; } } } #endif /* DDB */ /* * Fill in a struct xvfsconf based on a struct vfsconf. */ static int vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp) { struct xvfsconf xvfsp; bzero(&xvfsp, sizeof(xvfsp)); strcpy(xvfsp.vfc_name, vfsp->vfc_name); xvfsp.vfc_typenum = vfsp->vfc_typenum; xvfsp.vfc_refcount = vfsp->vfc_refcount; xvfsp.vfc_flags = vfsp->vfc_flags; /* * These are unused in userland, we keep them * to not break binary compatibility. */ xvfsp.vfc_vfsops = NULL; xvfsp.vfc_next = NULL; return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); } #ifdef COMPAT_FREEBSD32 struct xvfsconf32 { uint32_t vfc_vfsops; char vfc_name[MFSNAMELEN]; int32_t vfc_typenum; int32_t vfc_refcount; int32_t vfc_flags; uint32_t vfc_next; }; static int vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp) { struct xvfsconf32 xvfsp; strcpy(xvfsp.vfc_name, vfsp->vfc_name); xvfsp.vfc_typenum = vfsp->vfc_typenum; xvfsp.vfc_refcount = vfsp->vfc_refcount; xvfsp.vfc_flags = vfsp->vfc_flags; xvfsp.vfc_vfsops = 0; xvfsp.vfc_next = 0; return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); } #endif /* * Top level filesystem related information gathering. */ static int sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS) { struct vfsconf *vfsp; int error; error = 0; vfsconf_slock(); TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { #ifdef COMPAT_FREEBSD32 if (req->flags & SCTL_MASK32) error = vfsconf2x32(req, vfsp); else #endif error = vfsconf2x(req, vfsp); if (error) break; } vfsconf_sunlock(); return (error); } SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_vfs_conflist, "S,xvfsconf", "List of all configured filesystems"); #ifndef BURN_BRIDGES static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS); static int vfs_sysctl(SYSCTL_HANDLER_ARGS) { int *name = (int *)arg1 - 1; /* XXX */ u_int namelen = arg2 + 1; /* XXX */ struct vfsconf *vfsp; log(LOG_WARNING, "userland calling deprecated sysctl, " "please rebuild world\n"); #if 1 || defined(COMPAT_PRELITE2) /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ if (namelen == 1) return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); #endif switch (name[1]) { case VFS_MAXTYPENUM: if (namelen != 2) return (ENOTDIR); return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); case VFS_CONF: if (namelen != 3) return (ENOTDIR); /* overloaded */ vfsconf_slock(); TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { if (vfsp->vfc_typenum == name[2]) break; } vfsconf_sunlock(); if (vfsp == NULL) return (EOPNOTSUPP); #ifdef COMPAT_FREEBSD32 if (req->flags & SCTL_MASK32) return (vfsconf2x32(req, vfsp)); else #endif return (vfsconf2x(req, vfsp)); } return (EOPNOTSUPP); } static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP | CTLFLAG_MPSAFE, vfs_sysctl, "Generic filesystem"); #if 1 || defined(COMPAT_PRELITE2) static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) { int error; struct vfsconf *vfsp; struct ovfsconf ovfs; vfsconf_slock(); TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { bzero(&ovfs, sizeof(ovfs)); ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ strcpy(ovfs.vfc_name, vfsp->vfc_name); ovfs.vfc_index = vfsp->vfc_typenum; ovfs.vfc_refcount = vfsp->vfc_refcount; ovfs.vfc_flags = vfsp->vfc_flags; error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); if (error != 0) { vfsconf_sunlock(); return (error); } } vfsconf_sunlock(); return (0); } #endif /* 1 || COMPAT_PRELITE2 */ #endif /* !BURN_BRIDGES */ #define KINFO_VNODESLOP 10 #ifdef notyet /* * Dump vnode list (via sysctl). */ /* ARGSUSED */ static int sysctl_vnode(SYSCTL_HANDLER_ARGS) { struct xvnode *xvn; struct mount *mp; struct vnode *vp; int error, len, n; /* * Stale numvnodes access is not fatal here. */ req->lock = 0; len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn; if (!req->oldptr) /* Make an estimate */ return (SYSCTL_OUT(req, 0, len)); error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK); n = 0; mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) continue; MNT_ILOCK(mp); TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (n == len) break; vref(vp); xvn[n].xv_size = sizeof *xvn; xvn[n].xv_vnode = vp; xvn[n].xv_id = 0; /* XXX compat */ #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field XV_COPY(usecount); XV_COPY(writecount); XV_COPY(holdcnt); XV_COPY(mount); XV_COPY(numoutput); XV_COPY(type); #undef XV_COPY xvn[n].xv_flag = vp->v_vflag; switch (vp->v_type) { case VREG: case VDIR: case VLNK: break; case VBLK: case VCHR: if (vp->v_rdev == NULL) { vrele(vp); continue; } xvn[n].xv_dev = dev2udev(vp->v_rdev); break; case VSOCK: xvn[n].xv_socket = vp->v_socket; break; case VFIFO: xvn[n].xv_fifo = vp->v_fifoinfo; break; case VNON: case VBAD: default: /* shouldn't happen? */ vrele(vp); continue; } vrele(vp); ++n; } MNT_IUNLOCK(mp); mtx_lock(&mountlist_mtx); vfs_unbusy(mp); if (n == len) break; } mtx_unlock(&mountlist_mtx); error = SYSCTL_OUT(req, xvn, n * sizeof *xvn); free(xvn, M_TEMP); return (error); } SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, sysctl_vnode, "S,xvnode", ""); #endif static void unmount_or_warn(struct mount *mp) { int error; error = dounmount(mp, MNT_FORCE, curthread); if (error != 0) { printf("unmount of %s failed (", mp->mnt_stat.f_mntonname); if (error == EBUSY) printf("BUSY)\n"); else printf("%d)\n", error); } } /* * Unmount all filesystems. The list is traversed in reverse order * of mounting to avoid dependencies. */ void vfs_unmountall(void) { struct mount *mp, *tmp; CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__); /* * Since this only runs when rebooting, it is not interlocked. */ TAILQ_FOREACH_REVERSE_SAFE(mp, &mountlist, mntlist, mnt_list, tmp) { vfs_ref(mp); /* * Forcibly unmounting "/dev" before "/" would prevent clean * unmount of the latter. */ if (mp == rootdevmp) continue; unmount_or_warn(mp); } if (rootdevmp != NULL) unmount_or_warn(rootdevmp); } /* * perform msync on all vnodes under a mount point * the mount point must be locked. */ void vfs_msync(struct mount *mp, int flags) { struct vnode *vp, *mvp; struct vm_object *obj; CTR2(KTR_VFS, "%s: mp %p", __func__, mp); MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) { obj = vp->v_object; if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 && (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) { if (!vget(vp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, curthread)) { if (vp->v_vflag & VV_NOSYNC) { /* unlinked */ vput(vp); continue; } obj = vp->v_object; if (obj != NULL) { VM_OBJECT_WLOCK(obj); vm_object_page_clean(obj, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); VM_OBJECT_WUNLOCK(obj); } vput(vp); } } else VI_UNLOCK(vp); } } static void destroy_vpollinfo_free(struct vpollinfo *vi) { knlist_destroy(&vi->vpi_selinfo.si_note); mtx_destroy(&vi->vpi_lock); uma_zfree(vnodepoll_zone, vi); } static void destroy_vpollinfo(struct vpollinfo *vi) { knlist_clear(&vi->vpi_selinfo.si_note, 1); seldrain(&vi->vpi_selinfo); destroy_vpollinfo_free(vi); } /* * Initialize per-vnode helper structure to hold poll-related state. */ void v_addpollinfo(struct vnode *vp) { struct vpollinfo *vi; if (vp->v_pollinfo != NULL) return; vi = uma_zalloc(vnodepoll_zone, M_WAITOK | M_ZERO); mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF); knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock, vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked); VI_LOCK(vp); if (vp->v_pollinfo != NULL) { VI_UNLOCK(vp); destroy_vpollinfo_free(vi); return; } vp->v_pollinfo = vi; VI_UNLOCK(vp); } /* * Record a process's interest in events which might happen to * a vnode. Because poll uses the historic select-style interface * internally, this routine serves as both the ``check for any * pending events'' and the ``record my interest in future events'' * functions. (These are done together, while the lock is held, * to avoid race conditions.) */ int vn_pollrecord(struct vnode *vp, struct thread *td, int events) { v_addpollinfo(vp); mtx_lock(&vp->v_pollinfo->vpi_lock); if (vp->v_pollinfo->vpi_revents & events) { /* * This leaves events we are not interested * in available for the other process which * which presumably had requested them * (otherwise they would never have been * recorded). */ events &= vp->v_pollinfo->vpi_revents; vp->v_pollinfo->vpi_revents &= ~events; mtx_unlock(&vp->v_pollinfo->vpi_lock); return (events); } vp->v_pollinfo->vpi_events |= events; selrecord(td, &vp->v_pollinfo->vpi_selinfo); mtx_unlock(&vp->v_pollinfo->vpi_lock); return (0); } /* * Routine to create and manage a filesystem syncer vnode. */ #define sync_close ((int (*)(struct vop_close_args *))nullop) static int sync_fsync(struct vop_fsync_args *); static int sync_inactive(struct vop_inactive_args *); static int sync_reclaim(struct vop_reclaim_args *); static struct vop_vector sync_vnodeops = { .vop_bypass = VOP_EOPNOTSUPP, .vop_close = sync_close, /* close */ .vop_fsync = sync_fsync, /* fsync */ .vop_inactive = sync_inactive, /* inactive */ .vop_reclaim = sync_reclaim, /* reclaim */ .vop_lock1 = vop_stdlock, /* lock */ .vop_unlock = vop_stdunlock, /* unlock */ .vop_islocked = vop_stdislocked, /* islocked */ }; /* * Create a new filesystem syncer vnode for the specified mount point. */ void vfs_allocate_syncvnode(struct mount *mp) { struct vnode *vp; struct bufobj *bo; static long start, incr, next; int error; /* Allocate a new vnode */ error = getnewvnode("syncer", mp, &sync_vnodeops, &vp); if (error != 0) panic("vfs_allocate_syncvnode: getnewvnode() failed"); vp->v_type = VNON; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); vp->v_vflag |= VV_FORCEINSMQ; error = insmntque(vp, mp); if (error != 0) panic("vfs_allocate_syncvnode: insmntque() failed"); vp->v_vflag &= ~VV_FORCEINSMQ; VOP_UNLOCK(vp, 0); /* * Place the vnode onto the syncer worklist. We attempt to * scatter them about on the list so that they will go off * at evenly distributed times even if all the filesystems * are mounted at once. */ next += incr; if (next == 0 || next > syncer_maxdelay) { start /= 2; incr /= 2; if (start == 0) { start = syncer_maxdelay / 2; incr = syncer_maxdelay; } next = start; } bo = &vp->v_bufobj; BO_LOCK(bo); vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0); /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */ mtx_lock(&sync_mtx); sync_vnode_count++; if (mp->mnt_syncer == NULL) { mp->mnt_syncer = vp; vp = NULL; } mtx_unlock(&sync_mtx); BO_UNLOCK(bo); if (vp != NULL) { vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); vgone(vp); vput(vp); } } void vfs_deallocate_syncvnode(struct mount *mp) { struct vnode *vp; mtx_lock(&sync_mtx); vp = mp->mnt_syncer; if (vp != NULL) mp->mnt_syncer = NULL; mtx_unlock(&sync_mtx); if (vp != NULL) vrele(vp); } /* * Do a lazy sync of the filesystem. */ static int sync_fsync(struct vop_fsync_args *ap) { struct vnode *syncvp = ap->a_vp; struct mount *mp = syncvp->v_mount; int error, save; struct bufobj *bo; /* * We only need to do something if this is a lazy evaluation. */ if (ap->a_waitfor != MNT_LAZY) return (0); /* * Move ourselves to the back of the sync list. */ bo = &syncvp->v_bufobj; BO_LOCK(bo); vn_syncer_add_to_worklist(bo, syncdelay); BO_UNLOCK(bo); /* * Walk the list of vnodes pushing all that are dirty and * not already on the sync list. */ if (vfs_busy(mp, MBF_NOWAIT) != 0) return (0); if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) { vfs_unbusy(mp); return (0); } save = curthread_pflags_set(TDP_SYNCIO); vfs_msync(mp, MNT_NOWAIT); error = VFS_SYNC(mp, MNT_LAZY); curthread_pflags_restore(save); vn_finished_write(mp); vfs_unbusy(mp); return (error); } /* * The syncer vnode is no referenced. */ static int sync_inactive(struct vop_inactive_args *ap) { vgone(ap->a_vp); return (0); } /* * The syncer vnode is no longer needed and is being decommissioned. * * Modifications to the worklist must be protected by sync_mtx. */ static int sync_reclaim(struct vop_reclaim_args *ap) { struct vnode *vp = ap->a_vp; struct bufobj *bo; bo = &vp->v_bufobj; BO_LOCK(bo); mtx_lock(&sync_mtx); if (vp->v_mount->mnt_syncer == vp) vp->v_mount->mnt_syncer = NULL; if (bo->bo_flag & BO_ONWORKLST) { LIST_REMOVE(bo, bo_synclist); syncer_worklist_len--; sync_vnode_count--; bo->bo_flag &= ~BO_ONWORKLST; } mtx_unlock(&sync_mtx); BO_UNLOCK(bo); return (0); } /* * Check if vnode represents a disk device */ int vn_isdisk(struct vnode *vp, int *errp) { int error; if (vp->v_type != VCHR) { error = ENOTBLK; goto out; } error = 0; dev_lock(); if (vp->v_rdev == NULL) error = ENXIO; else if (vp->v_rdev->si_devsw == NULL) error = ENXIO; else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK)) error = ENOTBLK; dev_unlock(); out: if (errp != NULL) *errp = error; return (error == 0); } /* * Common filesystem object access control check routine. Accepts a * vnode's type, "mode", uid and gid, requested access mode, credentials, * and optional call-by-reference privused argument allowing vaccess() * to indicate to the caller whether privilege was used to satisfy the * request (obsoleted). Returns 0 on success, or an errno on failure. */ int vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid, accmode_t accmode, struct ucred *cred, int *privused) { accmode_t dac_granted; accmode_t priv_granted; KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0, ("invalid bit in accmode")); KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE), ("VAPPEND without VWRITE")); /* * Look for a normal, non-privileged way to access the file/directory * as requested. If it exists, go with that. */ if (privused != NULL) *privused = 0; dac_granted = 0; /* Check the owner. */ if (cred->cr_uid == file_uid) { dac_granted |= VADMIN; if (file_mode & S_IXUSR) dac_granted |= VEXEC; if (file_mode & S_IRUSR) dac_granted |= VREAD; if (file_mode & S_IWUSR) dac_granted |= (VWRITE | VAPPEND); if ((accmode & dac_granted) == accmode) return (0); goto privcheck; } /* Otherwise, check the groups (first match) */ if (groupmember(file_gid, cred)) { if (file_mode & S_IXGRP) dac_granted |= VEXEC; if (file_mode & S_IRGRP) dac_granted |= VREAD; if (file_mode & S_IWGRP) dac_granted |= (VWRITE | VAPPEND); if ((accmode & dac_granted) == accmode) return (0); goto privcheck; } /* Otherwise, check everyone else. */ if (file_mode & S_IXOTH) dac_granted |= VEXEC; if (file_mode & S_IROTH) dac_granted |= VREAD; if (file_mode & S_IWOTH) dac_granted |= (VWRITE | VAPPEND); if ((accmode & dac_granted) == accmode) return (0); privcheck: /* * Build a privilege mask to determine if the set of privileges * satisfies the requirements when combined with the granted mask * from above. For each privilege, if the privilege is required, * bitwise or the request type onto the priv_granted mask. */ priv_granted = 0; if (type == VDIR) { /* * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC * requests, instead of PRIV_VFS_EXEC. */ if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0)) priv_granted |= VEXEC; } else { /* * Ensure that at least one execute bit is on. Otherwise, * a privileged user will always succeed, and we don't want * this to happen unless the file really is executable. */ if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 && !priv_check_cred(cred, PRIV_VFS_EXEC, 0)) priv_granted |= VEXEC; } if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) && !priv_check_cred(cred, PRIV_VFS_READ, 0)) priv_granted |= VREAD; if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) && !priv_check_cred(cred, PRIV_VFS_WRITE, 0)) priv_granted |= (VWRITE | VAPPEND); if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) && !priv_check_cred(cred, PRIV_VFS_ADMIN, 0)) priv_granted |= VADMIN; if ((accmode & (priv_granted | dac_granted)) == accmode) { /* XXX audit: privilege used */ if (privused != NULL) *privused = 1; return (0); } return ((accmode & VADMIN) ? EPERM : EACCES); } /* * Credential check based on process requesting service, and per-attribute * permissions. */ int extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred, struct thread *td, accmode_t accmode) { /* * Kernel-invoked always succeeds. */ if (cred == NOCRED) return (0); /* * Do not allow privileged processes in jail to directly manipulate * system attributes. */ switch (attrnamespace) { case EXTATTR_NAMESPACE_SYSTEM: /* Potentially should be: return (EPERM); */ return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0)); case EXTATTR_NAMESPACE_USER: return (VOP_ACCESS(vp, accmode, cred, td)); default: return (EPERM); } } #ifdef DEBUG_VFS_LOCKS /* * This only exists to suppress warnings from unlocked specfs accesses. It is * no longer ok to have an unlocked VFS. */ #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \ (vp)->v_type == VCHR || (vp)->v_type == VBAD) int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "Drop into debugger on lock violation"); int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "Check for interlock across VOPs"); int vfs_badlock_print = 1; /* Print lock violations. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "Print lock violations"); int vfs_badlock_vnode = 1; /* Print vnode details on lock violations. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_vnode, CTLFLAG_RW, &vfs_badlock_vnode, 0, "Print vnode details on lock violations"); #ifdef KDB int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "Print backtrace at lock violations"); #endif static void vfs_badlock(const char *msg, const char *str, struct vnode *vp) { #ifdef KDB if (vfs_badlock_backtrace) kdb_backtrace(); #endif if (vfs_badlock_vnode) vn_printf(vp, "vnode "); if (vfs_badlock_print) printf("%s: %p %s\n", str, (void *)vp, msg); if (vfs_badlock_ddb) kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); } void assert_vi_locked(struct vnode *vp, const char *str) { if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp))) vfs_badlock("interlock is not locked but should be", str, vp); } void assert_vi_unlocked(struct vnode *vp, const char *str) { if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp))) vfs_badlock("interlock is locked but should not be", str, vp); } void assert_vop_locked(struct vnode *vp, const char *str) { int locked; if (!IGNORE_LOCK(vp)) { locked = VOP_ISLOCKED(vp); if (locked == 0 || locked == LK_EXCLOTHER) vfs_badlock("is not locked but should be", str, vp); } } void assert_vop_unlocked(struct vnode *vp, const char *str) { if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) vfs_badlock("is locked but should not be", str, vp); } void assert_vop_elocked(struct vnode *vp, const char *str) { if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) vfs_badlock("is not exclusive locked but should be", str, vp); } #endif /* DEBUG_VFS_LOCKS */ void vop_rename_fail(struct vop_rename_args *ap) { if (ap->a_tvp != NULL) vput(ap->a_tvp); if (ap->a_tdvp == ap->a_tvp) vrele(ap->a_tdvp); else vput(ap->a_tdvp); vrele(ap->a_fdvp); vrele(ap->a_fvp); } void vop_rename_pre(void *ap) { struct vop_rename_args *a = ap; #ifdef DEBUG_VFS_LOCKS if (a->a_tvp) ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME"); ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME"); ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME"); ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME"); /* Check the source (from). */ if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock && (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock)) ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked"); if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock) ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked"); /* Check the target. */ if (a->a_tvp) ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked"); ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked"); #endif if (a->a_tdvp != a->a_fdvp) vhold(a->a_fdvp); if (a->a_tvp != a->a_fvp) vhold(a->a_fvp); vhold(a->a_tdvp); if (a->a_tvp) vhold(a->a_tvp); } #ifdef DEBUG_VFS_LOCKS void vop_strategy_pre(void *ap) { struct vop_strategy_args *a; struct buf *bp; a = ap; bp = a->a_bp; /* * Cluster ops lock their component buffers but not the IO container. */ if ((bp->b_flags & B_CLUSTER) != 0) return; if (panicstr == NULL && !BUF_ISLOCKED(bp)) { if (vfs_badlock_print) printf( "VOP_STRATEGY: bp is not locked but should be\n"); if (vfs_badlock_ddb) kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); } } void vop_lock_pre(void *ap) { struct vop_lock1_args *a = ap; if ((a->a_flags & LK_INTERLOCK) == 0) ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); else ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK"); } void vop_lock_post(void *ap, int rc) { struct vop_lock1_args *a = ap; ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0) ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK"); } void vop_unlock_pre(void *ap) { struct vop_unlock_args *a = ap; if (a->a_flags & LK_INTERLOCK) ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK"); ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK"); } void vop_unlock_post(void *ap, int rc) { struct vop_unlock_args *a = ap; if (a->a_flags & LK_INTERLOCK) ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK"); } #endif void vop_create_post(void *ap, int rc) { struct vop_create_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); } void vop_deleteextattr_post(void *ap, int rc) { struct vop_deleteextattr_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); } void vop_link_post(void *ap, int rc) { struct vop_link_args *a = ap; if (!rc) { VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK); VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE); } } void vop_mkdir_post(void *ap, int rc) { struct vop_mkdir_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); } void vop_mknod_post(void *ap, int rc) { struct vop_mknod_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); } void vop_reclaim_post(void *ap, int rc) { struct vop_reclaim_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_REVOKE); } void vop_remove_post(void *ap, int rc) { struct vop_remove_args *a = ap; if (!rc) { VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); } } void vop_rename_post(void *ap, int rc) { struct vop_rename_args *a = ap; long hint; if (!rc) { hint = NOTE_WRITE; if (a->a_fdvp == a->a_tdvp) { if (a->a_tvp != NULL && a->a_tvp->v_type == VDIR) hint |= NOTE_LINK; VFS_KNOTE_UNLOCKED(a->a_fdvp, hint); VFS_KNOTE_UNLOCKED(a->a_tdvp, hint); } else { hint |= NOTE_EXTEND; if (a->a_fvp->v_type == VDIR) hint |= NOTE_LINK; VFS_KNOTE_UNLOCKED(a->a_fdvp, hint); if (a->a_fvp->v_type == VDIR && a->a_tvp != NULL && a->a_tvp->v_type == VDIR) hint &= ~NOTE_LINK; VFS_KNOTE_UNLOCKED(a->a_tdvp, hint); } VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME); if (a->a_tvp) VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE); } if (a->a_tdvp != a->a_fdvp) vdrop(a->a_fdvp); if (a->a_tvp != a->a_fvp) vdrop(a->a_fvp); vdrop(a->a_tdvp); if (a->a_tvp) vdrop(a->a_tvp); } void vop_rmdir_post(void *ap, int rc) { struct vop_rmdir_args *a = ap; if (!rc) { VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); } } void vop_setattr_post(void *ap, int rc) { struct vop_setattr_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); } void vop_setextattr_post(void *ap, int rc) { struct vop_setextattr_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); } void vop_symlink_post(void *ap, int rc) { struct vop_symlink_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); } void vop_open_post(void *ap, int rc) { struct vop_open_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_OPEN); } void vop_close_post(void *ap, int rc) { struct vop_close_args *a = ap; if (!rc && (a->a_cred != NOCRED || /* filter out revokes */ (a->a_vp->v_iflag & VI_DOOMED) == 0)) { VFS_KNOTE_LOCKED(a->a_vp, (a->a_fflag & FWRITE) != 0 ? NOTE_CLOSE_WRITE : NOTE_CLOSE); } } void vop_read_post(void *ap, int rc) { struct vop_read_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_READ); } void vop_readdir_post(void *ap, int rc) { struct vop_readdir_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_READ); } static struct knlist fs_knlist; static void vfs_event_init(void *arg) { knlist_init_mtx(&fs_knlist, NULL); } /* XXX - correct order? */ SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL); void vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused) { KNOTE_UNLOCKED(&fs_knlist, event); } static int filt_fsattach(struct knote *kn); static void filt_fsdetach(struct knote *kn); static int filt_fsevent(struct knote *kn, long hint); struct filterops fs_filtops = { .f_isfd = 0, .f_attach = filt_fsattach, .f_detach = filt_fsdetach, .f_event = filt_fsevent }; static int filt_fsattach(struct knote *kn) { kn->kn_flags |= EV_CLEAR; knlist_add(&fs_knlist, kn, 0); return (0); } static void filt_fsdetach(struct knote *kn) { knlist_remove(&fs_knlist, kn, 0); } static int filt_fsevent(struct knote *kn, long hint) { kn->kn_fflags |= hint; return (kn->kn_fflags != 0); } static int sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS) { struct vfsidctl vc; int error; struct mount *mp; error = SYSCTL_IN(req, &vc, sizeof(vc)); if (error) return (error); if (vc.vc_vers != VFS_CTL_VERS1) return (EINVAL); mp = vfs_getvfs(&vc.vc_fsid); if (mp == NULL) return (ENOENT); /* ensure that a specific sysctl goes to the right filesystem. */ if (strcmp(vc.vc_fstypename, "*") != 0 && strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) { vfs_rel(mp); return (EINVAL); } VCTLTOREQ(&vc, req); error = VFS_SYSCTL(mp, vc.vc_op, req); vfs_rel(mp); return (error); } SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "", "Sysctl by fsid"); /* * Function to initialize a va_filerev field sensibly. * XXX: Wouldn't a random number make a lot more sense ?? */ u_quad_t init_va_filerev(void) { struct bintime bt; getbinuptime(&bt); return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL)); } static int filt_vfsread(struct knote *kn, long hint); static int filt_vfswrite(struct knote *kn, long hint); static int filt_vfsvnode(struct knote *kn, long hint); static void filt_vfsdetach(struct knote *kn); static struct filterops vfsread_filtops = { .f_isfd = 1, .f_detach = filt_vfsdetach, .f_event = filt_vfsread }; static struct filterops vfswrite_filtops = { .f_isfd = 1, .f_detach = filt_vfsdetach, .f_event = filt_vfswrite }; static struct filterops vfsvnode_filtops = { .f_isfd = 1, .f_detach = filt_vfsdetach, .f_event = filt_vfsvnode }; static void vfs_knllock(void *arg) { struct vnode *vp = arg; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } static void vfs_knlunlock(void *arg) { struct vnode *vp = arg; VOP_UNLOCK(vp, 0); } static void vfs_knl_assert_locked(void *arg) { #ifdef DEBUG_VFS_LOCKS struct vnode *vp = arg; ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked"); #endif } static void vfs_knl_assert_unlocked(void *arg) { #ifdef DEBUG_VFS_LOCKS struct vnode *vp = arg; ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked"); #endif } int vfs_kqfilter(struct vop_kqfilter_args *ap) { struct vnode *vp = ap->a_vp; struct knote *kn = ap->a_kn; struct knlist *knl; switch (kn->kn_filter) { case EVFILT_READ: kn->kn_fop = &vfsread_filtops; break; case EVFILT_WRITE: kn->kn_fop = &vfswrite_filtops; break; case EVFILT_VNODE: kn->kn_fop = &vfsvnode_filtops; break; default: return (EINVAL); } kn->kn_hook = (caddr_t)vp; v_addpollinfo(vp); if (vp->v_pollinfo == NULL) return (ENOMEM); knl = &vp->v_pollinfo->vpi_selinfo.si_note; vhold(vp); knlist_add(knl, kn, 0); return (0); } /* * Detach knote from vnode */ static void filt_vfsdetach(struct knote *kn) { struct vnode *vp = (struct vnode *)kn->kn_hook; KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo")); knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0); vdrop(vp); } /*ARGSUSED*/ static int filt_vfsread(struct knote *kn, long hint) { struct vnode *vp = (struct vnode *)kn->kn_hook; struct vattr va; int res; /* * filesystem is gone, so set the EOF flag and schedule * the knote for deletion. */ if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) { VI_LOCK(vp); kn->kn_flags |= (EV_EOF | EV_ONESHOT); VI_UNLOCK(vp); return (1); } if (VOP_GETATTR(vp, &va, curthread->td_ucred)) return (0); VI_LOCK(vp); kn->kn_data = va.va_size - kn->kn_fp->f_offset; res = (kn->kn_sfflags & NOTE_FILE_POLL) != 0 || kn->kn_data != 0; VI_UNLOCK(vp); return (res); } /*ARGSUSED*/ static int filt_vfswrite(struct knote *kn, long hint) { struct vnode *vp = (struct vnode *)kn->kn_hook; VI_LOCK(vp); /* * filesystem is gone, so set the EOF flag and schedule * the knote for deletion. */ if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) kn->kn_flags |= (EV_EOF | EV_ONESHOT); kn->kn_data = 0; VI_UNLOCK(vp); return (1); } static int filt_vfsvnode(struct knote *kn, long hint) { struct vnode *vp = (struct vnode *)kn->kn_hook; int res; VI_LOCK(vp); if (kn->kn_sfflags & hint) kn->kn_fflags |= hint; if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) { kn->kn_flags |= EV_EOF; VI_UNLOCK(vp); return (1); } res = (kn->kn_fflags != 0); VI_UNLOCK(vp); return (res); } int vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off) { int error; if (dp->d_reclen > ap->a_uio->uio_resid) return (ENAMETOOLONG); error = uiomove(dp, dp->d_reclen, ap->a_uio); if (error) { if (ap->a_ncookies != NULL) { if (ap->a_cookies != NULL) free(ap->a_cookies, M_TEMP); ap->a_cookies = NULL; *ap->a_ncookies = 0; } return (error); } if (ap->a_ncookies == NULL) return (0); KASSERT(ap->a_cookies, ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!")); *ap->a_cookies = realloc(*ap->a_cookies, (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO); (*ap->a_cookies)[*ap->a_ncookies] = off; *ap->a_ncookies += 1; return (0); } /* * Mark for update the access time of the file if the filesystem * supports VOP_MARKATIME. This functionality is used by execve and * mmap, so we want to avoid the I/O implied by directly setting * va_atime for the sake of efficiency. */ void vfs_mark_atime(struct vnode *vp, struct ucred *cred) { struct mount *mp; mp = vp->v_mount; ASSERT_VOP_LOCKED(vp, "vfs_mark_atime"); if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) (void)VOP_MARKATIME(vp); } /* * The purpose of this routine is to remove granularity from accmode_t, * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE, * VADMIN and VAPPEND. * * If it returns 0, the caller is supposed to continue with the usual * access checks using 'accmode' as modified by this routine. If it * returns nonzero value, the caller is supposed to return that value * as errno. * * Note that after this routine runs, accmode may be zero. */ int vfs_unixify_accmode(accmode_t *accmode) { /* * There is no way to specify explicit "deny" rule using * file mode or POSIX.1e ACLs. */ if (*accmode & VEXPLICIT_DENY) { *accmode = 0; return (0); } /* * None of these can be translated into usual access bits. * Also, the common case for NFSv4 ACLs is to not contain * either of these bits. Caller should check for VWRITE * on the containing directory instead. */ if (*accmode & (VDELETE_CHILD | VDELETE)) return (EPERM); if (*accmode & VADMIN_PERMS) { *accmode &= ~VADMIN_PERMS; *accmode |= VADMIN; } /* * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL * or VSYNCHRONIZE using file mode or POSIX.1e ACL. */ *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE); return (0); } /* * These are helper functions for filesystems to traverse all * their vnodes. See MNT_VNODE_FOREACH_ALL() in sys/mount.h. * * This interface replaces MNT_VNODE_FOREACH. */ MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); struct vnode * __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp) { struct vnode *vp; if (should_yield()) kern_yield(PRI_USER); MNT_ILOCK(mp); KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); vp = TAILQ_NEXT(*mvp, v_nmntvnodes); while (vp != NULL && (vp->v_type == VMARKER || (vp->v_iflag & VI_DOOMED) != 0)) vp = TAILQ_NEXT(vp, v_nmntvnodes); /* Check if we are done */ if (vp == NULL) { __mnt_vnode_markerfree_all(mvp, mp); /* MNT_IUNLOCK(mp); -- done in above function */ mtx_assert(MNT_MTX(mp), MA_NOTOWNED); return (NULL); } TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); VI_LOCK(vp); MNT_IUNLOCK(mp); return (vp); } struct vnode * __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp) { struct vnode *vp; *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO); MNT_ILOCK(mp); MNT_REF(mp); (*mvp)->v_type = VMARKER; vp = TAILQ_FIRST(&mp->mnt_nvnodelist); while (vp != NULL && (vp->v_type == VMARKER || (vp->v_iflag & VI_DOOMED) != 0)) vp = TAILQ_NEXT(vp, v_nmntvnodes); /* Check if we are done */ if (vp == NULL) { MNT_REL(mp); MNT_IUNLOCK(mp); free(*mvp, M_VNODE_MARKER); *mvp = NULL; return (NULL); } (*mvp)->v_mount = mp; TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); VI_LOCK(vp); MNT_IUNLOCK(mp); return (vp); } void __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp) { if (*mvp == NULL) { MNT_IUNLOCK(mp); return; } mtx_assert(MNT_MTX(mp), MA_OWNED); KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); MNT_REL(mp); MNT_IUNLOCK(mp); free(*mvp, M_VNODE_MARKER); *mvp = NULL; } /* * These are helper functions for filesystems to traverse their * active vnodes. See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h */ static void mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp) { KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); MNT_ILOCK(mp); MNT_REL(mp); MNT_IUNLOCK(mp); free(*mvp, M_VNODE_MARKER); *mvp = NULL; } static struct vnode * mnt_vnode_next_active(struct vnode **mvp, struct mount *mp) { struct vnode *vp, *nvp; mtx_assert(&vnode_free_list_mtx, MA_OWNED); KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); restart: vp = TAILQ_NEXT(*mvp, v_actfreelist); TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist); while (vp != NULL) { if (vp->v_type == VMARKER) { vp = TAILQ_NEXT(vp, v_actfreelist); continue; } if (!VI_TRYLOCK(vp)) { if (mp_ncpus == 1 || should_yield()) { TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist); mtx_unlock(&vnode_free_list_mtx); pause("vnacti", 1); mtx_lock(&vnode_free_list_mtx); goto restart; } continue; } KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp)); KASSERT(vp->v_mount == mp || vp->v_mount == NULL, ("alien vnode on the active list %p %p", vp, mp)); if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0) break; nvp = TAILQ_NEXT(vp, v_actfreelist); VI_UNLOCK(vp); vp = nvp; } /* Check if we are done */ if (vp == NULL) { mtx_unlock(&vnode_free_list_mtx); mnt_vnode_markerfree_active(mvp, mp); return (NULL); } TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist); mtx_unlock(&vnode_free_list_mtx); ASSERT_VI_LOCKED(vp, "active iter"); KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp)); return (vp); } struct vnode * __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp) { if (should_yield()) kern_yield(PRI_USER); mtx_lock(&vnode_free_list_mtx); return (mnt_vnode_next_active(mvp, mp)); } struct vnode * __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp) { struct vnode *vp; *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO); MNT_ILOCK(mp); MNT_REF(mp); MNT_IUNLOCK(mp); (*mvp)->v_type = VMARKER; (*mvp)->v_mount = mp; mtx_lock(&vnode_free_list_mtx); vp = TAILQ_FIRST(&mp->mnt_activevnodelist); if (vp == NULL) { mtx_unlock(&vnode_free_list_mtx); mnt_vnode_markerfree_active(mvp, mp); return (NULL); } TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist); return (mnt_vnode_next_active(mvp, mp)); } void __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp) { if (*mvp == NULL) return; mtx_lock(&vnode_free_list_mtx); TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist); mtx_unlock(&vnode_free_list_mtx); mnt_vnode_markerfree_active(mvp, mp); } Index: head/sys/kern/vfs_syscalls.c =================================================================== --- head/sys/kern/vfs_syscalls.c (revision 305831) +++ head/sys/kern/vfs_syscalls.c (revision 305832) @@ -1,4680 +1,4680 @@ /*- * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_syscalls.c 8.13 (Berkeley) 4/15/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_capsicum.h" #include "opt_compat.h" #include "opt_ktrace.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KTRACE #include #endif #include #include #include #include #include #include #include #include MALLOC_DEFINE(M_FADVISE, "fadvise", "posix_fadvise(2) information"); SDT_PROVIDER_DEFINE(vfs); SDT_PROBE_DEFINE2(vfs, , stat, mode, "char *", "int"); SDT_PROBE_DEFINE2(vfs, , stat, reg, "char *", "int"); static int kern_chflagsat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, u_long flags, int atflag); static int setfflags(struct thread *td, struct vnode *, u_long); static int getutimes(const struct timeval *, enum uio_seg, struct timespec *); static int getutimens(const struct timespec *, enum uio_seg, struct timespec *, int *); static int setutimes(struct thread *td, struct vnode *, const struct timespec *, int, int); static int vn_access(struct vnode *vp, int user_flags, struct ucred *cred, struct thread *td); /* * Sync each mounted filesystem. */ #ifndef _SYS_SYSPROTO_H_ struct sync_args { int dummy; }; #endif /* ARGSUSED */ int sys_sync(td, uap) struct thread *td; struct sync_args *uap; { struct mount *mp, *nmp; int save; mtx_lock(&mountlist_mtx); for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } if ((mp->mnt_flag & MNT_RDONLY) == 0 && vn_start_write(NULL, &mp, V_NOWAIT) == 0) { save = curthread_pflags_set(TDP_SYNCIO); vfs_msync(mp, MNT_NOWAIT); VFS_SYNC(mp, MNT_NOWAIT); curthread_pflags_restore(save); vn_finished_write(mp); } mtx_lock(&mountlist_mtx); nmp = TAILQ_NEXT(mp, mnt_list); vfs_unbusy(mp); } mtx_unlock(&mountlist_mtx); return (0); } /* * Change filesystem quotas. */ #ifndef _SYS_SYSPROTO_H_ struct quotactl_args { char *path; int cmd; int uid; caddr_t arg; }; #endif int sys_quotactl(td, uap) struct thread *td; register struct quotactl_args /* { char *path; int cmd; int uid; caddr_t arg; } */ *uap; { struct mount *mp; struct nameidata nd; int error; AUDIT_ARG_CMD(uap->cmd); AUDIT_ARG_UID(uap->uid); if (!prison_allow(td->td_ucred, PR_ALLOW_QUOTAS)) return (EPERM); NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_USERSPACE, uap->path, td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); mp = nd.ni_vp->v_mount; vfs_ref(mp); vput(nd.ni_vp); error = vfs_busy(mp, 0); vfs_rel(mp); if (error != 0) return (error); error = VFS_QUOTACTL(mp, uap->cmd, uap->uid, uap->arg); /* * Since quota on operation typically needs to open quota * file, the Q_QUOTAON handler needs to unbusy the mount point * before calling into namei. Otherwise, unmount might be * started between two vfs_busy() invocations (first is our, * second is from mount point cross-walk code in lookup()), * causing deadlock. * * Require that Q_QUOTAON handles the vfs_busy() reference on * its own, always returning with ubusied mount point. */ if ((uap->cmd >> SUBCMDSHIFT) != Q_QUOTAON) vfs_unbusy(mp); return (error); } /* * Used by statfs conversion routines to scale the block size up if * necessary so that all of the block counts are <= 'max_size'. Note * that 'max_size' should be a bitmask, i.e. 2^n - 1 for some non-zero * value of 'n'. */ void statfs_scale_blocks(struct statfs *sf, long max_size) { uint64_t count; int shift; KASSERT(powerof2(max_size + 1), ("%s: invalid max_size", __func__)); /* * Attempt to scale the block counts to give a more accurate * overview to userland of the ratio of free space to used * space. To do this, find the largest block count and compute * a divisor that lets it fit into a signed integer <= max_size. */ if (sf->f_bavail < 0) count = -sf->f_bavail; else count = sf->f_bavail; count = MAX(sf->f_blocks, MAX(sf->f_bfree, count)); if (count <= max_size) return; count >>= flsl(max_size); shift = 0; while (count > 0) { shift++; count >>=1; } sf->f_bsize <<= shift; sf->f_blocks >>= shift; sf->f_bfree >>= shift; sf->f_bavail >>= shift; } /* * Get filesystem statistics. */ #ifndef _SYS_SYSPROTO_H_ struct statfs_args { char *path; struct statfs *buf; }; #endif int sys_statfs(td, uap) struct thread *td; register struct statfs_args /* { char *path; struct statfs *buf; } */ *uap; { struct statfs sf; int error; error = kern_statfs(td, uap->path, UIO_USERSPACE, &sf); if (error == 0) error = copyout(&sf, uap->buf, sizeof(sf)); return (error); } int kern_statfs(struct thread *td, char *path, enum uio_seg pathseg, struct statfs *buf) { struct mount *mp; struct statfs *sp, sb; struct nameidata nd; int error; NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1, pathseg, path, td); error = namei(&nd); if (error != 0) return (error); mp = nd.ni_vp->v_mount; vfs_ref(mp); NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_vp); error = vfs_busy(mp, 0); vfs_rel(mp); if (error != 0) return (error); #ifdef MAC error = mac_mount_check_stat(td->td_ucred, mp); if (error != 0) goto out; #endif /* * Set these in case the underlying filesystem fails to do so. */ sp = &mp->mnt_stat; sp->f_version = STATFS_VERSION; sp->f_namemax = NAME_MAX; sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; error = VFS_STATFS(mp, sp); if (error != 0) goto out; if (priv_check(td, PRIV_VFS_GENERATION)) { bcopy(sp, &sb, sizeof(sb)); sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0; prison_enforce_statfs(td->td_ucred, mp, &sb); sp = &sb; } *buf = *sp; out: vfs_unbusy(mp); return (error); } /* * Get filesystem statistics. */ #ifndef _SYS_SYSPROTO_H_ struct fstatfs_args { int fd; struct statfs *buf; }; #endif int sys_fstatfs(td, uap) struct thread *td; register struct fstatfs_args /* { int fd; struct statfs *buf; } */ *uap; { struct statfs sf; int error; error = kern_fstatfs(td, uap->fd, &sf); if (error == 0) error = copyout(&sf, uap->buf, sizeof(sf)); return (error); } int kern_fstatfs(struct thread *td, int fd, struct statfs *buf) { struct file *fp; struct mount *mp; struct statfs *sp, sb; struct vnode *vp; cap_rights_t rights; int error; AUDIT_ARG_FD(fd); error = getvnode(td, fd, cap_rights_init(&rights, CAP_FSTATFS), &fp); if (error != 0) return (error); vp = fp->f_vnode; vn_lock(vp, LK_SHARED | LK_RETRY); #ifdef AUDIT AUDIT_ARG_VNODE1(vp); #endif mp = vp->v_mount; if (mp) vfs_ref(mp); VOP_UNLOCK(vp, 0); fdrop(fp, td); if (mp == NULL) { error = EBADF; goto out; } error = vfs_busy(mp, 0); vfs_rel(mp); if (error != 0) return (error); #ifdef MAC error = mac_mount_check_stat(td->td_ucred, mp); if (error != 0) goto out; #endif /* * Set these in case the underlying filesystem fails to do so. */ sp = &mp->mnt_stat; sp->f_version = STATFS_VERSION; sp->f_namemax = NAME_MAX; sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; error = VFS_STATFS(mp, sp); if (error != 0) goto out; if (priv_check(td, PRIV_VFS_GENERATION)) { bcopy(sp, &sb, sizeof(sb)); sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0; prison_enforce_statfs(td->td_ucred, mp, &sb); sp = &sb; } *buf = *sp; out: if (mp) vfs_unbusy(mp); return (error); } /* * Get statistics on all filesystems. */ #ifndef _SYS_SYSPROTO_H_ struct getfsstat_args { struct statfs *buf; long bufsize; int flags; }; #endif int sys_getfsstat(td, uap) struct thread *td; register struct getfsstat_args /* { struct statfs *buf; long bufsize; int flags; } */ *uap; { size_t count; int error; if (uap->bufsize < 0 || uap->bufsize > SIZE_MAX) return (EINVAL); error = kern_getfsstat(td, &uap->buf, uap->bufsize, &count, UIO_USERSPACE, uap->flags); if (error == 0) td->td_retval[0] = count; return (error); } /* * If (bufsize > 0 && bufseg == UIO_SYSSPACE) * The caller is responsible for freeing memory which will be allocated * in '*buf'. */ int kern_getfsstat(struct thread *td, struct statfs **buf, size_t bufsize, size_t *countp, enum uio_seg bufseg, int flags) { struct mount *mp, *nmp; struct statfs *sfsp, *sp, sb; size_t count, maxcount; int error; maxcount = bufsize / sizeof(struct statfs); if (bufsize == 0) sfsp = NULL; else if (bufseg == UIO_USERSPACE) sfsp = *buf; else /* if (bufseg == UIO_SYSSPACE) */ { count = 0; mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { count++; } mtx_unlock(&mountlist_mtx); if (maxcount > count) maxcount = count; sfsp = *buf = malloc(maxcount * sizeof(struct statfs), M_TEMP, M_WAITOK); } count = 0; mtx_lock(&mountlist_mtx); for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { if (prison_canseemount(td->td_ucred, mp) != 0) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } #ifdef MAC if (mac_mount_check_stat(td->td_ucred, mp) != 0) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } #endif if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } if (sfsp && count < maxcount) { sp = &mp->mnt_stat; /* * Set these in case the underlying filesystem * fails to do so. */ sp->f_version = STATFS_VERSION; sp->f_namemax = NAME_MAX; sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; /* * If MNT_NOWAIT or MNT_LAZY is specified, do not * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY * overrides MNT_WAIT. */ if (((flags & (MNT_LAZY|MNT_NOWAIT)) == 0 || (flags & MNT_WAIT)) && (error = VFS_STATFS(mp, sp))) { mtx_lock(&mountlist_mtx); nmp = TAILQ_NEXT(mp, mnt_list); vfs_unbusy(mp); continue; } if (priv_check(td, PRIV_VFS_GENERATION)) { bcopy(sp, &sb, sizeof(sb)); sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0; prison_enforce_statfs(td->td_ucred, mp, &sb); sp = &sb; } if (bufseg == UIO_SYSSPACE) bcopy(sp, sfsp, sizeof(*sp)); else /* if (bufseg == UIO_USERSPACE) */ { error = copyout(sp, sfsp, sizeof(*sp)); if (error != 0) { vfs_unbusy(mp); return (error); } } sfsp++; } count++; mtx_lock(&mountlist_mtx); nmp = TAILQ_NEXT(mp, mnt_list); vfs_unbusy(mp); } mtx_unlock(&mountlist_mtx); if (sfsp && count > maxcount) *countp = maxcount; else *countp = count; return (0); } #ifdef COMPAT_FREEBSD4 /* * Get old format filesystem statistics. */ static void cvtstatfs(struct statfs *, struct ostatfs *); #ifndef _SYS_SYSPROTO_H_ struct freebsd4_statfs_args { char *path; struct ostatfs *buf; }; #endif int freebsd4_statfs(td, uap) struct thread *td; struct freebsd4_statfs_args /* { char *path; struct ostatfs *buf; } */ *uap; { struct ostatfs osb; struct statfs sf; int error; error = kern_statfs(td, uap->path, UIO_USERSPACE, &sf); if (error != 0) return (error); cvtstatfs(&sf, &osb); return (copyout(&osb, uap->buf, sizeof(osb))); } /* * Get filesystem statistics. */ #ifndef _SYS_SYSPROTO_H_ struct freebsd4_fstatfs_args { int fd; struct ostatfs *buf; }; #endif int freebsd4_fstatfs(td, uap) struct thread *td; struct freebsd4_fstatfs_args /* { int fd; struct ostatfs *buf; } */ *uap; { struct ostatfs osb; struct statfs sf; int error; error = kern_fstatfs(td, uap->fd, &sf); if (error != 0) return (error); cvtstatfs(&sf, &osb); return (copyout(&osb, uap->buf, sizeof(osb))); } /* * Get statistics on all filesystems. */ #ifndef _SYS_SYSPROTO_H_ struct freebsd4_getfsstat_args { struct ostatfs *buf; long bufsize; int flags; }; #endif int freebsd4_getfsstat(td, uap) struct thread *td; register struct freebsd4_getfsstat_args /* { struct ostatfs *buf; long bufsize; int flags; } */ *uap; { struct statfs *buf, *sp; struct ostatfs osb; size_t count, size; int error; if (uap->bufsize < 0) return (EINVAL); count = uap->bufsize / sizeof(struct ostatfs); if (count > SIZE_MAX / sizeof(struct statfs)) return (EINVAL); size = count * sizeof(struct statfs); error = kern_getfsstat(td, &buf, size, &count, UIO_SYSSPACE, uap->flags); td->td_retval[0] = count; if (size != 0) { sp = buf; while (count != 0 && error == 0) { cvtstatfs(sp, &osb); error = copyout(&osb, uap->buf, sizeof(osb)); sp++; uap->buf++; count--; } free(buf, M_TEMP); } return (error); } /* * Implement fstatfs() for (NFS) file handles. */ #ifndef _SYS_SYSPROTO_H_ struct freebsd4_fhstatfs_args { struct fhandle *u_fhp; struct ostatfs *buf; }; #endif int freebsd4_fhstatfs(td, uap) struct thread *td; struct freebsd4_fhstatfs_args /* { struct fhandle *u_fhp; struct ostatfs *buf; } */ *uap; { struct ostatfs osb; struct statfs sf; fhandle_t fh; int error; error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t)); if (error != 0) return (error); error = kern_fhstatfs(td, fh, &sf); if (error != 0) return (error); cvtstatfs(&sf, &osb); return (copyout(&osb, uap->buf, sizeof(osb))); } /* * Convert a new format statfs structure to an old format statfs structure. */ static void cvtstatfs(nsp, osp) struct statfs *nsp; struct ostatfs *osp; { statfs_scale_blocks(nsp, LONG_MAX); bzero(osp, sizeof(*osp)); osp->f_bsize = nsp->f_bsize; osp->f_iosize = MIN(nsp->f_iosize, LONG_MAX); osp->f_blocks = nsp->f_blocks; osp->f_bfree = nsp->f_bfree; osp->f_bavail = nsp->f_bavail; osp->f_files = MIN(nsp->f_files, LONG_MAX); osp->f_ffree = MIN(nsp->f_ffree, LONG_MAX); osp->f_owner = nsp->f_owner; osp->f_type = nsp->f_type; osp->f_flags = nsp->f_flags; osp->f_syncwrites = MIN(nsp->f_syncwrites, LONG_MAX); osp->f_asyncwrites = MIN(nsp->f_asyncwrites, LONG_MAX); osp->f_syncreads = MIN(nsp->f_syncreads, LONG_MAX); osp->f_asyncreads = MIN(nsp->f_asyncreads, LONG_MAX); strlcpy(osp->f_fstypename, nsp->f_fstypename, MIN(MFSNAMELEN, OMFSNAMELEN)); strlcpy(osp->f_mntonname, nsp->f_mntonname, MIN(MNAMELEN, OMNAMELEN)); strlcpy(osp->f_mntfromname, nsp->f_mntfromname, MIN(MNAMELEN, OMNAMELEN)); osp->f_fsid = nsp->f_fsid; } #endif /* COMPAT_FREEBSD4 */ /* * Change current working directory to a given file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct fchdir_args { int fd; }; #endif int sys_fchdir(td, uap) struct thread *td; struct fchdir_args /* { int fd; } */ *uap; { struct vnode *vp, *tdp; struct mount *mp; struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(uap->fd); error = getvnode(td, uap->fd, cap_rights_init(&rights, CAP_FCHDIR), &fp); if (error != 0) return (error); vp = fp->f_vnode; VREF(vp); fdrop(fp, td); vn_lock(vp, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(vp); error = change_dir(vp, td); while (!error && (mp = vp->v_mountedhere) != NULL) { if (vfs_busy(mp, 0)) continue; error = VFS_ROOT(mp, LK_SHARED, &tdp); vfs_unbusy(mp); if (error != 0) break; vput(vp); vp = tdp; } if (error != 0) { vput(vp); return (error); } VOP_UNLOCK(vp, 0); pwd_chdir(td, vp); return (0); } /* * Change current working directory (``.''). */ #ifndef _SYS_SYSPROTO_H_ struct chdir_args { char *path; }; #endif int sys_chdir(td, uap) struct thread *td; struct chdir_args /* { char *path; } */ *uap; { return (kern_chdir(td, uap->path, UIO_USERSPACE)); } int kern_chdir(struct thread *td, char *path, enum uio_seg pathseg) { struct nameidata nd; int error; NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1, pathseg, path, td); if ((error = namei(&nd)) != 0) return (error); if ((error = change_dir(nd.ni_vp, td)) != 0) { vput(nd.ni_vp); NDFREE(&nd, NDF_ONLY_PNBUF); return (error); } VOP_UNLOCK(nd.ni_vp, 0); NDFREE(&nd, NDF_ONLY_PNBUF); pwd_chdir(td, nd.ni_vp); return (0); } /* * Change notion of root (``/'') directory. */ #ifndef _SYS_SYSPROTO_H_ struct chroot_args { char *path; }; #endif int sys_chroot(td, uap) struct thread *td; struct chroot_args /* { char *path; } */ *uap; { struct nameidata nd; int error; error = priv_check(td, PRIV_VFS_CHROOT); if (error != 0) return (error); NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1, UIO_USERSPACE, uap->path, td); error = namei(&nd); if (error != 0) goto error; error = change_dir(nd.ni_vp, td); if (error != 0) goto e_vunlock; #ifdef MAC error = mac_vnode_check_chroot(td->td_ucred, nd.ni_vp); if (error != 0) goto e_vunlock; #endif VOP_UNLOCK(nd.ni_vp, 0); error = pwd_chroot(td, nd.ni_vp); vrele(nd.ni_vp); NDFREE(&nd, NDF_ONLY_PNBUF); return (error); e_vunlock: vput(nd.ni_vp); error: NDFREE(&nd, NDF_ONLY_PNBUF); return (error); } /* * Common routine for chroot and chdir. Callers must provide a locked vnode * instance. */ int change_dir(vp, td) struct vnode *vp; struct thread *td; { #ifdef MAC int error; #endif ASSERT_VOP_LOCKED(vp, "change_dir(): vp not locked"); if (vp->v_type != VDIR) return (ENOTDIR); #ifdef MAC error = mac_vnode_check_chdir(td->td_ucred, vp); if (error != 0) return (error); #endif return (VOP_ACCESS(vp, VEXEC, td->td_ucred, td)); } static __inline void flags_to_rights(int flags, cap_rights_t *rightsp) { if (flags & O_EXEC) { cap_rights_set(rightsp, CAP_FEXECVE); } else { switch ((flags & O_ACCMODE)) { case O_RDONLY: cap_rights_set(rightsp, CAP_READ); break; case O_RDWR: cap_rights_set(rightsp, CAP_READ); /* FALLTHROUGH */ case O_WRONLY: cap_rights_set(rightsp, CAP_WRITE); if (!(flags & (O_APPEND | O_TRUNC))) cap_rights_set(rightsp, CAP_SEEK); break; } } if (flags & O_CREAT) cap_rights_set(rightsp, CAP_CREATE); if (flags & O_TRUNC) cap_rights_set(rightsp, CAP_FTRUNCATE); if (flags & (O_SYNC | O_FSYNC)) cap_rights_set(rightsp, CAP_FSYNC); if (flags & (O_EXLOCK | O_SHLOCK)) cap_rights_set(rightsp, CAP_FLOCK); } /* * Check permissions, allocate an open file structure, and call the device * open routine if any. */ #ifndef _SYS_SYSPROTO_H_ struct open_args { char *path; int flags; int mode; }; #endif int sys_open(td, uap) struct thread *td; register struct open_args /* { char *path; int flags; int mode; } */ *uap; { return (kern_openat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->flags, uap->mode)); } #ifndef _SYS_SYSPROTO_H_ struct openat_args { int fd; char *path; int flag; int mode; }; #endif int sys_openat(struct thread *td, struct openat_args *uap) { AUDIT_ARG_FD(uap->fd); return (kern_openat(td, uap->fd, uap->path, UIO_USERSPACE, uap->flag, uap->mode)); } int kern_openat(struct thread *td, int fd, char *path, enum uio_seg pathseg, int flags, int mode) { struct proc *p = td->td_proc; struct filedesc *fdp = p->p_fd; struct file *fp; struct vnode *vp; struct nameidata nd; cap_rights_t rights; int cmode, error, indx; indx = -1; AUDIT_ARG_FFLAGS(flags); AUDIT_ARG_MODE(mode); cap_rights_init(&rights, CAP_LOOKUP); flags_to_rights(flags, &rights); /* * Only one of the O_EXEC, O_RDONLY, O_WRONLY and O_RDWR flags * may be specified. */ if (flags & O_EXEC) { if (flags & O_ACCMODE) return (EINVAL); } else if ((flags & O_ACCMODE) == O_ACCMODE) { return (EINVAL); } else { flags = FFLAGS(flags); } /* * Allocate a file structure. The descriptor to reference it * is allocated and set by finstall() below. */ error = falloc_noinstall(td, &fp); if (error != 0) return (error); /* * An extra reference on `fp' has been held for us by * falloc_noinstall(). */ /* Set the flags early so the finit in devfs can pick them up. */ fp->f_flag = flags & FMASK; cmode = ((mode & ~fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT; NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | AUDITVNODE1, pathseg, path, fd, &rights, td); td->td_dupfd = -1; /* XXX check for fdopen */ error = vn_open(&nd, &flags, cmode, fp); if (error != 0) { /* * If the vn_open replaced the method vector, something * wonderous happened deep below and we just pass it up * pretending we know what we do. */ if (error == ENXIO && fp->f_ops != &badfileops) goto success; /* * Handle special fdopen() case. bleh. * * Don't do this for relative (capability) lookups; we don't * understand exactly what would happen, and we don't think * that it ever should. */ if (nd.ni_strictrelative == 0 && (error == ENODEV || error == ENXIO) && td->td_dupfd >= 0) { error = dupfdopen(td, fdp, td->td_dupfd, flags, error, &indx); if (error == 0) goto success; } goto bad; } td->td_dupfd = 0; NDFREE(&nd, NDF_ONLY_PNBUF); vp = nd.ni_vp; /* * Store the vnode, for any f_type. Typically, the vnode use * count is decremented by direct call to vn_closefile() for * files that switched type in the cdevsw fdopen() method. */ fp->f_vnode = vp; /* * If the file wasn't claimed by devfs bind it to the normal * vnode operations here. */ if (fp->f_ops == &badfileops) { KASSERT(vp->v_type != VFIFO, ("Unexpected fifo.")); fp->f_seqcount = 1; finit(fp, (flags & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE, vp, &vnops); } VOP_UNLOCK(vp, 0); if (flags & O_TRUNC) { error = fo_truncate(fp, 0, td->td_ucred, td); if (error != 0) goto bad; } success: /* * If we haven't already installed the FD (for dupfdopen), do so now. */ if (indx == -1) { struct filecaps *fcaps; #ifdef CAPABILITIES if (nd.ni_strictrelative == 1) fcaps = &nd.ni_filecaps; else #endif fcaps = NULL; error = finstall(td, fp, &indx, flags, fcaps); /* On success finstall() consumes fcaps. */ if (error != 0) { filecaps_free(&nd.ni_filecaps); goto bad; } } else { filecaps_free(&nd.ni_filecaps); } /* * Release our private reference, leaving the one associated with * the descriptor table intact. */ fdrop(fp, td); td->td_retval[0] = indx; return (0); bad: KASSERT(indx == -1, ("indx=%d, should be -1", indx)); fdrop(fp, td); return (error); } #ifdef COMPAT_43 /* * Create a file. */ #ifndef _SYS_SYSPROTO_H_ struct ocreat_args { char *path; int mode; }; #endif int ocreat(td, uap) struct thread *td; register struct ocreat_args /* { char *path; int mode; } */ *uap; { return (kern_openat(td, AT_FDCWD, uap->path, UIO_USERSPACE, O_WRONLY | O_CREAT | O_TRUNC, uap->mode)); } #endif /* COMPAT_43 */ /* * Create a special file. */ #ifndef _SYS_SYSPROTO_H_ struct mknod_args { char *path; int mode; int dev; }; #endif int sys_mknod(td, uap) struct thread *td; register struct mknod_args /* { char *path; int mode; int dev; } */ *uap; { return (kern_mknodat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->mode, uap->dev)); } #ifndef _SYS_SYSPROTO_H_ struct mknodat_args { int fd; char *path; mode_t mode; dev_t dev; }; #endif int sys_mknodat(struct thread *td, struct mknodat_args *uap) { return (kern_mknodat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode, uap->dev)); } int kern_mknodat(struct thread *td, int fd, char *path, enum uio_seg pathseg, int mode, int dev) { struct vnode *vp; struct mount *mp; struct vattr vattr; struct nameidata nd; cap_rights_t rights; int error, whiteout = 0; AUDIT_ARG_MODE(mode); AUDIT_ARG_DEV(dev); switch (mode & S_IFMT) { case S_IFCHR: case S_IFBLK: error = priv_check(td, PRIV_VFS_MKNOD_DEV); if (error == 0 && dev == VNOVAL) error = EINVAL; break; case S_IFMT: error = priv_check(td, PRIV_VFS_MKNOD_BAD); break; case S_IFWHT: error = priv_check(td, PRIV_VFS_MKNOD_WHT); break; case S_IFIFO: if (dev == 0) return (kern_mkfifoat(td, fd, path, pathseg, mode)); /* FALLTHROUGH */ default: error = EINVAL; break; } if (error != 0) return (error); restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE1 | NOCACHE, pathseg, path, fd, cap_rights_init(&rights, CAP_MKNODAT), td); if ((error = namei(&nd)) != 0) return (error); vp = nd.ni_vp; if (vp != NULL) { NDFREE(&nd, NDF_ONLY_PNBUF); if (vp == nd.ni_dvp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); vrele(vp); return (EEXIST); } else { VATTR_NULL(&vattr); vattr.va_mode = (mode & ALLPERMS) & ~td->td_proc->p_fd->fd_cmask; vattr.va_rdev = dev; whiteout = 0; switch (mode & S_IFMT) { case S_IFMT: /* used by badsect to flag bad sectors */ vattr.va_type = VBAD; break; case S_IFCHR: vattr.va_type = VCHR; break; case S_IFBLK: vattr.va_type = VBLK; break; case S_IFWHT: whiteout = 1; break; default: panic("kern_mknod: invalid mode"); } } if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) return (error); goto restart; } #ifdef MAC if (error == 0 && !whiteout) error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, &vattr); #endif if (error == 0) { if (whiteout) error = VOP_WHITEOUT(nd.ni_dvp, &nd.ni_cnd, CREATE); else { error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); if (error == 0) vput(nd.ni_vp); } } NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); vn_finished_write(mp); return (error); } /* * Create a named pipe. */ #ifndef _SYS_SYSPROTO_H_ struct mkfifo_args { char *path; int mode; }; #endif int sys_mkfifo(td, uap) struct thread *td; register struct mkfifo_args /* { char *path; int mode; } */ *uap; { return (kern_mkfifoat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->mode)); } #ifndef _SYS_SYSPROTO_H_ struct mkfifoat_args { int fd; char *path; mode_t mode; }; #endif int sys_mkfifoat(struct thread *td, struct mkfifoat_args *uap) { return (kern_mkfifoat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode)); } int kern_mkfifoat(struct thread *td, int fd, char *path, enum uio_seg pathseg, int mode) { struct mount *mp; struct vattr vattr; struct nameidata nd; cap_rights_t rights; int error; AUDIT_ARG_MODE(mode); restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE1 | NOCACHE, pathseg, path, fd, cap_rights_init(&rights, CAP_MKFIFOAT), td); if ((error = namei(&nd)) != 0) return (error); if (nd.ni_vp != NULL) { NDFREE(&nd, NDF_ONLY_PNBUF); if (nd.ni_vp == nd.ni_dvp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); vrele(nd.ni_vp); return (EEXIST); } if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) return (error); goto restart; } VATTR_NULL(&vattr); vattr.va_type = VFIFO; vattr.va_mode = (mode & ALLPERMS) & ~td->td_proc->p_fd->fd_cmask; #ifdef MAC error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, &vattr); if (error != 0) goto out; #endif error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); if (error == 0) vput(nd.ni_vp); #ifdef MAC out: #endif vput(nd.ni_dvp); vn_finished_write(mp); NDFREE(&nd, NDF_ONLY_PNBUF); return (error); } /* * Make a hard file link. */ #ifndef _SYS_SYSPROTO_H_ struct link_args { char *path; char *link; }; #endif int sys_link(td, uap) struct thread *td; register struct link_args /* { char *path; char *link; } */ *uap; { return (kern_linkat(td, AT_FDCWD, AT_FDCWD, uap->path, uap->link, UIO_USERSPACE, FOLLOW)); } #ifndef _SYS_SYSPROTO_H_ struct linkat_args { int fd1; char *path1; int fd2; char *path2; int flag; }; #endif int sys_linkat(struct thread *td, struct linkat_args *uap) { int flag; flag = uap->flag; if (flag & ~AT_SYMLINK_FOLLOW) return (EINVAL); return (kern_linkat(td, uap->fd1, uap->fd2, uap->path1, uap->path2, UIO_USERSPACE, (flag & AT_SYMLINK_FOLLOW) ? FOLLOW : NOFOLLOW)); } int hardlink_check_uid = 0; SYSCTL_INT(_security_bsd, OID_AUTO, hardlink_check_uid, CTLFLAG_RW, &hardlink_check_uid, 0, "Unprivileged processes cannot create hard links to files owned by other " "users"); static int hardlink_check_gid = 0; SYSCTL_INT(_security_bsd, OID_AUTO, hardlink_check_gid, CTLFLAG_RW, &hardlink_check_gid, 0, "Unprivileged processes cannot create hard links to files owned by other " "groups"); static int can_hardlink(struct vnode *vp, struct ucred *cred) { struct vattr va; int error; if (!hardlink_check_uid && !hardlink_check_gid) return (0); error = VOP_GETATTR(vp, &va, cred); if (error != 0) return (error); if (hardlink_check_uid && cred->cr_uid != va.va_uid) { error = priv_check_cred(cred, PRIV_VFS_LINK, 0); if (error != 0) return (error); } if (hardlink_check_gid && !groupmember(va.va_gid, cred)) { error = priv_check_cred(cred, PRIV_VFS_LINK, 0); if (error != 0) return (error); } return (0); } int kern_linkat(struct thread *td, int fd1, int fd2, char *path1, char *path2, enum uio_seg segflg, int follow) { struct vnode *vp; struct mount *mp; struct nameidata nd; cap_rights_t rights; int error; again: bwillwrite(); NDINIT_ATRIGHTS(&nd, LOOKUP, follow | AUDITVNODE1, segflg, path1, fd1, cap_rights_init(&rights, CAP_LINKAT_SOURCE), td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); vp = nd.ni_vp; if (vp->v_type == VDIR) { vrele(vp); return (EPERM); /* POSIX */ } NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE2 | NOCACHE, segflg, path2, fd2, cap_rights_init(&rights, CAP_LINKAT_TARGET), td); if ((error = namei(&nd)) == 0) { if (nd.ni_vp != NULL) { NDFREE(&nd, NDF_ONLY_PNBUF); if (nd.ni_dvp == nd.ni_vp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); vrele(nd.ni_vp); vrele(vp); return (EEXIST); } else if (nd.ni_dvp->v_mount != vp->v_mount) { /* * Cross-device link. No need to recheck * vp->v_type, since it cannot change, except * to VBAD. */ NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); vrele(vp); return (EXDEV); } else if ((error = vn_lock(vp, LK_EXCLUSIVE)) == 0) { error = can_hardlink(vp, td->td_ucred); #ifdef MAC if (error == 0) error = mac_vnode_check_link(td->td_ucred, nd.ni_dvp, vp, &nd.ni_cnd); #endif if (error != 0) { vput(vp); vput(nd.ni_dvp); NDFREE(&nd, NDF_ONLY_PNBUF); return (error); } error = vn_start_write(vp, &mp, V_NOWAIT); if (error != 0) { vput(vp); vput(nd.ni_dvp); NDFREE(&nd, NDF_ONLY_PNBUF); error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); if (error != 0) return (error); goto again; } error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd); VOP_UNLOCK(vp, 0); vput(nd.ni_dvp); vn_finished_write(mp); NDFREE(&nd, NDF_ONLY_PNBUF); } else { vput(nd.ni_dvp); NDFREE(&nd, NDF_ONLY_PNBUF); vrele(vp); goto again; } } vrele(vp); return (error); } /* * Make a symbolic link. */ #ifndef _SYS_SYSPROTO_H_ struct symlink_args { char *path; char *link; }; #endif int sys_symlink(td, uap) struct thread *td; register struct symlink_args /* { char *path; char *link; } */ *uap; { return (kern_symlinkat(td, uap->path, AT_FDCWD, uap->link, UIO_USERSPACE)); } #ifndef _SYS_SYSPROTO_H_ struct symlinkat_args { char *path; int fd; char *path2; }; #endif int sys_symlinkat(struct thread *td, struct symlinkat_args *uap) { return (kern_symlinkat(td, uap->path1, uap->fd, uap->path2, UIO_USERSPACE)); } int kern_symlinkat(struct thread *td, char *path1, int fd, char *path2, enum uio_seg segflg) { struct mount *mp; struct vattr vattr; char *syspath; struct nameidata nd; int error; cap_rights_t rights; if (segflg == UIO_SYSSPACE) { syspath = path1; } else { syspath = uma_zalloc(namei_zone, M_WAITOK); if ((error = copyinstr(path1, syspath, MAXPATHLEN, NULL)) != 0) goto out; } AUDIT_ARG_TEXT(syspath); restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE1 | NOCACHE, segflg, path2, fd, cap_rights_init(&rights, CAP_SYMLINKAT), td); if ((error = namei(&nd)) != 0) goto out; if (nd.ni_vp) { NDFREE(&nd, NDF_ONLY_PNBUF); if (nd.ni_vp == nd.ni_dvp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); vrele(nd.ni_vp); error = EEXIST; goto out; } if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) goto out; goto restart; } VATTR_NULL(&vattr); vattr.va_mode = ACCESSPERMS &~ td->td_proc->p_fd->fd_cmask; #ifdef MAC vattr.va_type = VLNK; error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, &vattr); if (error != 0) goto out2; #endif error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr, syspath); if (error == 0) vput(nd.ni_vp); #ifdef MAC out2: #endif NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); vn_finished_write(mp); out: if (segflg != UIO_SYSSPACE) uma_zfree(namei_zone, syspath); return (error); } /* * Delete a whiteout from the filesystem. */ int sys_undelete(td, uap) struct thread *td; register struct undelete_args /* { char *path; } */ *uap; { struct mount *mp; struct nameidata nd; int error; restart: bwillwrite(); NDINIT(&nd, DELETE, LOCKPARENT | DOWHITEOUT | AUDITVNODE1, UIO_USERSPACE, uap->path, td); error = namei(&nd); if (error != 0) return (error); if (nd.ni_vp != NULLVP || !(nd.ni_cnd.cn_flags & ISWHITEOUT)) { NDFREE(&nd, NDF_ONLY_PNBUF); if (nd.ni_vp == nd.ni_dvp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); if (nd.ni_vp) vrele(nd.ni_vp); return (EEXIST); } if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) return (error); goto restart; } error = VOP_WHITEOUT(nd.ni_dvp, &nd.ni_cnd, DELETE); NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); vn_finished_write(mp); return (error); } /* * Delete a name from the filesystem. */ #ifndef _SYS_SYSPROTO_H_ struct unlink_args { char *path; }; #endif int sys_unlink(td, uap) struct thread *td; struct unlink_args /* { char *path; } */ *uap; { return (kern_unlinkat(td, AT_FDCWD, uap->path, UIO_USERSPACE, 0)); } #ifndef _SYS_SYSPROTO_H_ struct unlinkat_args { int fd; char *path; int flag; }; #endif int sys_unlinkat(struct thread *td, struct unlinkat_args *uap) { int flag = uap->flag; int fd = uap->fd; char *path = uap->path; if (flag & ~AT_REMOVEDIR) return (EINVAL); if (flag & AT_REMOVEDIR) return (kern_rmdirat(td, fd, path, UIO_USERSPACE)); else return (kern_unlinkat(td, fd, path, UIO_USERSPACE, 0)); } int kern_unlinkat(struct thread *td, int fd, char *path, enum uio_seg pathseg, ino_t oldinum) { struct mount *mp; struct vnode *vp; struct nameidata nd; struct stat sb; cap_rights_t rights; int error; restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, DELETE, LOCKPARENT | LOCKLEAF | AUDITVNODE1, pathseg, path, fd, cap_rights_init(&rights, CAP_UNLINKAT), td); if ((error = namei(&nd)) != 0) return (error == EINVAL ? EPERM : error); vp = nd.ni_vp; if (vp->v_type == VDIR && oldinum == 0) { error = EPERM; /* POSIX */ } else if (oldinum != 0 && ((error = vn_stat(vp, &sb, td->td_ucred, NOCRED, td)) == 0) && sb.st_ino != oldinum) { error = EIDRM; /* Identifier removed */ } else { /* * The root of a mounted filesystem cannot be deleted. * * XXX: can this only be a VDIR case? */ if (vp->v_vflag & VV_ROOT) error = EBUSY; } if (error == 0) { if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if (vp == nd.ni_dvp) vrele(vp); else vput(vp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) return (error); goto restart; } #ifdef MAC error = mac_vnode_check_unlink(td->td_ucred, nd.ni_dvp, vp, &nd.ni_cnd); if (error != 0) goto out; #endif vfs_notify_upper(vp, VFS_NOTIFY_UPPER_UNLINK); error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd); #ifdef MAC out: #endif vn_finished_write(mp); } NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if (vp == nd.ni_dvp) vrele(vp); else vput(vp); return (error); } /* * Reposition read/write file offset. */ #ifndef _SYS_SYSPROTO_H_ struct lseek_args { int fd; int pad; off_t offset; int whence; }; #endif int sys_lseek(td, uap) struct thread *td; register struct lseek_args /* { int fd; int pad; off_t offset; int whence; } */ *uap; { struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(uap->fd); error = fget(td, uap->fd, cap_rights_init(&rights, CAP_SEEK), &fp); if (error != 0) return (error); error = (fp->f_ops->fo_flags & DFLAG_SEEKABLE) != 0 ? fo_seek(fp, uap->offset, uap->whence, td) : ESPIPE; fdrop(fp, td); return (error); } #if defined(COMPAT_43) /* * Reposition read/write file offset. */ #ifndef _SYS_SYSPROTO_H_ struct olseek_args { int fd; long offset; int whence; }; #endif int olseek(td, uap) struct thread *td; register struct olseek_args /* { int fd; long offset; int whence; } */ *uap; { struct lseek_args /* { int fd; int pad; off_t offset; int whence; } */ nuap; nuap.fd = uap->fd; nuap.offset = uap->offset; nuap.whence = uap->whence; return (sys_lseek(td, &nuap)); } #endif /* COMPAT_43 */ #if defined(COMPAT_FREEBSD6) /* Version with the 'pad' argument */ int freebsd6_lseek(td, uap) struct thread *td; register struct freebsd6_lseek_args *uap; { struct lseek_args ouap; ouap.fd = uap->fd; ouap.offset = uap->offset; ouap.whence = uap->whence; return (sys_lseek(td, &ouap)); } #endif /* * Check access permissions using passed credentials. */ static int vn_access(vp, user_flags, cred, td) struct vnode *vp; int user_flags; struct ucred *cred; struct thread *td; { accmode_t accmode; int error; /* Flags == 0 means only check for existence. */ if (user_flags == 0) return (0); accmode = 0; if (user_flags & R_OK) accmode |= VREAD; if (user_flags & W_OK) accmode |= VWRITE; if (user_flags & X_OK) accmode |= VEXEC; #ifdef MAC error = mac_vnode_check_access(cred, vp, accmode); if (error != 0) return (error); #endif if ((accmode & VWRITE) == 0 || (error = vn_writechk(vp)) == 0) error = VOP_ACCESS(vp, accmode, cred, td); return (error); } /* * Check access permissions using "real" credentials. */ #ifndef _SYS_SYSPROTO_H_ struct access_args { char *path; int amode; }; #endif int sys_access(td, uap) struct thread *td; register struct access_args /* { char *path; int amode; } */ *uap; { return (kern_accessat(td, AT_FDCWD, uap->path, UIO_USERSPACE, 0, uap->amode)); } #ifndef _SYS_SYSPROTO_H_ struct faccessat_args { int dirfd; char *path; int amode; int flag; } #endif int sys_faccessat(struct thread *td, struct faccessat_args *uap) { return (kern_accessat(td, uap->fd, uap->path, UIO_USERSPACE, uap->flag, uap->amode)); } int kern_accessat(struct thread *td, int fd, char *path, enum uio_seg pathseg, int flag, int amode) { struct ucred *cred, *usecred; struct vnode *vp; struct nameidata nd; cap_rights_t rights; int error; if (flag & ~AT_EACCESS) return (EINVAL); if (amode != F_OK && (amode & ~(R_OK | W_OK | X_OK)) != 0) return (EINVAL); /* * Create and modify a temporary credential instead of one that * is potentially shared (if we need one). */ cred = td->td_ucred; if ((flag & AT_EACCESS) == 0 && ((cred->cr_uid != cred->cr_ruid || cred->cr_rgid != cred->cr_groups[0]))) { usecred = crdup(cred); usecred->cr_uid = cred->cr_ruid; usecred->cr_groups[0] = cred->cr_rgid; td->td_ucred = usecred; } else usecred = cred; AUDIT_ARG_VALUE(amode); NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1, pathseg, path, fd, cap_rights_init(&rights, CAP_FSTAT), td); if ((error = namei(&nd)) != 0) goto out; vp = nd.ni_vp; error = vn_access(vp, amode, usecred, td); NDFREE(&nd, NDF_ONLY_PNBUF); vput(vp); out: if (usecred != cred) { td->td_ucred = cred; crfree(usecred); } return (error); } /* * Check access permissions using "effective" credentials. */ #ifndef _SYS_SYSPROTO_H_ struct eaccess_args { char *path; int amode; }; #endif int sys_eaccess(td, uap) struct thread *td; register struct eaccess_args /* { char *path; int amode; } */ *uap; { return (kern_accessat(td, AT_FDCWD, uap->path, UIO_USERSPACE, AT_EACCESS, uap->amode)); } #if defined(COMPAT_43) /* * Get file status; this version follows links. */ #ifndef _SYS_SYSPROTO_H_ struct ostat_args { char *path; struct ostat *ub; }; #endif int ostat(td, uap) struct thread *td; register struct ostat_args /* { char *path; struct ostat *ub; } */ *uap; { struct stat sb; struct ostat osb; int error; error = kern_statat(td, 0, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); cvtstat(&sb, &osb); return (copyout(&osb, uap->ub, sizeof (osb))); } /* * Get file status; this version does not follow links. */ #ifndef _SYS_SYSPROTO_H_ struct olstat_args { char *path; struct ostat *ub; }; #endif int olstat(td, uap) struct thread *td; register struct olstat_args /* { char *path; struct ostat *ub; } */ *uap; { struct stat sb; struct ostat osb; int error; error = kern_statat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); cvtstat(&sb, &osb); return (copyout(&osb, uap->ub, sizeof (osb))); } /* * Convert from an old to a new stat structure. */ void cvtstat(st, ost) struct stat *st; struct ostat *ost; { bzero(ost, sizeof(*ost)); ost->st_dev = st->st_dev; ost->st_ino = st->st_ino; ost->st_mode = st->st_mode; ost->st_nlink = st->st_nlink; ost->st_uid = st->st_uid; ost->st_gid = st->st_gid; ost->st_rdev = st->st_rdev; if (st->st_size < (quad_t)1 << 32) ost->st_size = st->st_size; else ost->st_size = -2; ost->st_atim = st->st_atim; ost->st_mtim = st->st_mtim; ost->st_ctim = st->st_ctim; ost->st_blksize = st->st_blksize; ost->st_blocks = st->st_blocks; ost->st_flags = st->st_flags; ost->st_gen = st->st_gen; } #endif /* COMPAT_43 */ /* * Get file status; this version follows links. */ #ifndef _SYS_SYSPROTO_H_ struct stat_args { char *path; struct stat *ub; }; #endif int sys_stat(td, uap) struct thread *td; register struct stat_args /* { char *path; struct stat *ub; } */ *uap; { struct stat sb; int error; error = kern_statat(td, 0, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error == 0) error = copyout(&sb, uap->ub, sizeof (sb)); return (error); } #ifndef _SYS_SYSPROTO_H_ struct fstatat_args { int fd; char *path; struct stat *buf; int flag; } #endif int sys_fstatat(struct thread *td, struct fstatat_args *uap) { struct stat sb; int error; error = kern_statat(td, uap->flag, uap->fd, uap->path, UIO_USERSPACE, &sb, NULL); if (error == 0) error = copyout(&sb, uap->buf, sizeof (sb)); return (error); } int kern_statat(struct thread *td, int flag, int fd, char *path, enum uio_seg pathseg, struct stat *sbp, void (*hook)(struct vnode *vp, struct stat *sbp)) { struct nameidata nd; struct stat sb; cap_rights_t rights; int error; if (flag & ~AT_SYMLINK_NOFOLLOW) return (EINVAL); NDINIT_ATRIGHTS(&nd, LOOKUP, ((flag & AT_SYMLINK_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKSHARED | LOCKLEAF | AUDITVNODE1, pathseg, path, fd, cap_rights_init(&rights, CAP_FSTAT), td); if ((error = namei(&nd)) != 0) return (error); error = vn_stat(nd.ni_vp, &sb, td->td_ucred, NOCRED, td); if (error == 0) { SDT_PROBE2(vfs, , stat, mode, path, sb.st_mode); if (S_ISREG(sb.st_mode)) SDT_PROBE2(vfs, , stat, reg, path, pathseg); if (__predict_false(hook != NULL)) hook(nd.ni_vp, &sb); } NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_vp); if (error != 0) return (error); *sbp = sb; #ifdef KTRACE if (KTRPOINT(td, KTR_STRUCT)) ktrstat(&sb); #endif return (0); } /* * Get file status; this version does not follow links. */ #ifndef _SYS_SYSPROTO_H_ struct lstat_args { char *path; struct stat *ub; }; #endif int sys_lstat(td, uap) struct thread *td; register struct lstat_args /* { char *path; struct stat *ub; } */ *uap; { struct stat sb; int error; error = kern_statat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error == 0) error = copyout(&sb, uap->ub, sizeof (sb)); return (error); } /* * Implementation of the NetBSD [l]stat() functions. */ void cvtnstat(sb, nsb) struct stat *sb; struct nstat *nsb; { bzero(nsb, sizeof *nsb); nsb->st_dev = sb->st_dev; nsb->st_ino = sb->st_ino; nsb->st_mode = sb->st_mode; nsb->st_nlink = sb->st_nlink; nsb->st_uid = sb->st_uid; nsb->st_gid = sb->st_gid; nsb->st_rdev = sb->st_rdev; nsb->st_atim = sb->st_atim; nsb->st_mtim = sb->st_mtim; nsb->st_ctim = sb->st_ctim; nsb->st_size = sb->st_size; nsb->st_blocks = sb->st_blocks; nsb->st_blksize = sb->st_blksize; nsb->st_flags = sb->st_flags; nsb->st_gen = sb->st_gen; nsb->st_birthtim = sb->st_birthtim; } #ifndef _SYS_SYSPROTO_H_ struct nstat_args { char *path; struct nstat *ub; }; #endif int sys_nstat(td, uap) struct thread *td; register struct nstat_args /* { char *path; struct nstat *ub; } */ *uap; { struct stat sb; struct nstat nsb; int error; error = kern_statat(td, 0, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); cvtnstat(&sb, &nsb); return (copyout(&nsb, uap->ub, sizeof (nsb))); } /* * NetBSD lstat. Get file status; this version does not follow links. */ #ifndef _SYS_SYSPROTO_H_ struct lstat_args { char *path; struct stat *ub; }; #endif int sys_nlstat(td, uap) struct thread *td; register struct nlstat_args /* { char *path; struct nstat *ub; } */ *uap; { struct stat sb; struct nstat nsb; int error; error = kern_statat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); cvtnstat(&sb, &nsb); return (copyout(&nsb, uap->ub, sizeof (nsb))); } /* * Get configurable pathname variables. */ #ifndef _SYS_SYSPROTO_H_ struct pathconf_args { char *path; int name; }; #endif int sys_pathconf(td, uap) struct thread *td; register struct pathconf_args /* { char *path; int name; } */ *uap; { return (kern_pathconf(td, uap->path, UIO_USERSPACE, uap->name, FOLLOW)); } #ifndef _SYS_SYSPROTO_H_ struct lpathconf_args { char *path; int name; }; #endif int sys_lpathconf(td, uap) struct thread *td; register struct lpathconf_args /* { char *path; int name; } */ *uap; { return (kern_pathconf(td, uap->path, UIO_USERSPACE, uap->name, NOFOLLOW)); } int kern_pathconf(struct thread *td, char *path, enum uio_seg pathseg, int name, u_long flags) { struct nameidata nd; int error; NDINIT(&nd, LOOKUP, LOCKSHARED | LOCKLEAF | AUDITVNODE1 | flags, pathseg, path, td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); error = VOP_PATHCONF(nd.ni_vp, name, td->td_retval); vput(nd.ni_vp); return (error); } /* * Return target name of a symbolic link. */ #ifndef _SYS_SYSPROTO_H_ struct readlink_args { char *path; char *buf; size_t count; }; #endif int sys_readlink(td, uap) struct thread *td; register struct readlink_args /* { char *path; char *buf; size_t count; } */ *uap; { return (kern_readlinkat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->buf, UIO_USERSPACE, uap->count)); } #ifndef _SYS_SYSPROTO_H_ struct readlinkat_args { int fd; char *path; char *buf; size_t bufsize; }; #endif int sys_readlinkat(struct thread *td, struct readlinkat_args *uap) { return (kern_readlinkat(td, uap->fd, uap->path, UIO_USERSPACE, uap->buf, UIO_USERSPACE, uap->bufsize)); } int kern_readlinkat(struct thread *td, int fd, char *path, enum uio_seg pathseg, char *buf, enum uio_seg bufseg, size_t count) { struct vnode *vp; struct iovec aiov; struct uio auio; struct nameidata nd; int error; if (count > IOSIZE_MAX) return (EINVAL); NDINIT_AT(&nd, LOOKUP, NOFOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1, pathseg, path, fd, td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); vp = nd.ni_vp; #ifdef MAC error = mac_vnode_check_readlink(td->td_ucred, vp); if (error != 0) { vput(vp); return (error); } #endif if (vp->v_type != VLNK) error = EINVAL; else { aiov.iov_base = buf; aiov.iov_len = count; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_offset = 0; auio.uio_rw = UIO_READ; auio.uio_segflg = bufseg; auio.uio_td = td; auio.uio_resid = count; error = VOP_READLINK(vp, &auio, td->td_ucred); td->td_retval[0] = count - auio.uio_resid; } vput(vp); return (error); } /* * Common implementation code for chflags() and fchflags(). */ static int setfflags(td, vp, flags) struct thread *td; struct vnode *vp; u_long flags; { struct mount *mp; struct vattr vattr; int error; /* We can't support the value matching VNOVAL. */ if (flags == VNOVAL) return (EOPNOTSUPP); /* * Prevent non-root users from setting flags on devices. When * a device is reused, users can retain ownership of the device * if they are allowed to set flags and programs assume that * chown can't fail when done as root. */ if (vp->v_type == VCHR || vp->v_type == VBLK) { error = priv_check(td, PRIV_VFS_CHFLAGS_DEV); if (error != 0) return (error); } if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) return (error); VATTR_NULL(&vattr); vattr.va_flags = flags; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); #ifdef MAC error = mac_vnode_check_setflags(td->td_ucred, vp, vattr.va_flags); if (error == 0) #endif error = VOP_SETATTR(vp, &vattr, td->td_ucred); VOP_UNLOCK(vp, 0); vn_finished_write(mp); return (error); } /* * Change flags of a file given a path name. */ #ifndef _SYS_SYSPROTO_H_ struct chflags_args { const char *path; u_long flags; }; #endif int sys_chflags(td, uap) struct thread *td; register struct chflags_args /* { const char *path; u_long flags; } */ *uap; { return (kern_chflagsat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->flags, 0)); } #ifndef _SYS_SYSPROTO_H_ struct chflagsat_args { int fd; const char *path; u_long flags; int atflag; } #endif int sys_chflagsat(struct thread *td, struct chflagsat_args *uap) { int fd = uap->fd; const char *path = uap->path; u_long flags = uap->flags; int atflag = uap->atflag; if (atflag & ~AT_SYMLINK_NOFOLLOW) return (EINVAL); return (kern_chflagsat(td, fd, path, UIO_USERSPACE, flags, atflag)); } /* * Same as chflags() but doesn't follow symlinks. */ int sys_lchflags(td, uap) struct thread *td; register struct lchflags_args /* { const char *path; u_long flags; } */ *uap; { return (kern_chflagsat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->flags, AT_SYMLINK_NOFOLLOW)); } static int kern_chflagsat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, u_long flags, int atflag) { struct nameidata nd; cap_rights_t rights; int error, follow; AUDIT_ARG_FFLAGS(flags); follow = (atflag & AT_SYMLINK_NOFOLLOW) ? NOFOLLOW : FOLLOW; NDINIT_ATRIGHTS(&nd, LOOKUP, follow | AUDITVNODE1, pathseg, path, fd, cap_rights_init(&rights, CAP_FCHFLAGS), td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); error = setfflags(td, nd.ni_vp, flags); vrele(nd.ni_vp); return (error); } /* * Change flags of a file given a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct fchflags_args { int fd; u_long flags; }; #endif int sys_fchflags(td, uap) struct thread *td; register struct fchflags_args /* { int fd; u_long flags; } */ *uap; { struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(uap->fd); AUDIT_ARG_FFLAGS(uap->flags); error = getvnode(td, uap->fd, cap_rights_init(&rights, CAP_FCHFLAGS), &fp); if (error != 0) return (error); #ifdef AUDIT vn_lock(fp->f_vnode, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(fp->f_vnode); VOP_UNLOCK(fp->f_vnode, 0); #endif error = setfflags(td, fp->f_vnode, uap->flags); fdrop(fp, td); return (error); } /* * Common implementation code for chmod(), lchmod() and fchmod(). */ int setfmode(td, cred, vp, mode) struct thread *td; struct ucred *cred; struct vnode *vp; int mode; { struct mount *mp; struct vattr vattr; int error; if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); VATTR_NULL(&vattr); vattr.va_mode = mode & ALLPERMS; #ifdef MAC error = mac_vnode_check_setmode(cred, vp, vattr.va_mode); if (error == 0) #endif error = VOP_SETATTR(vp, &vattr, cred); VOP_UNLOCK(vp, 0); vn_finished_write(mp); return (error); } /* * Change mode of a file given path name. */ #ifndef _SYS_SYSPROTO_H_ struct chmod_args { char *path; int mode; }; #endif int sys_chmod(td, uap) struct thread *td; register struct chmod_args /* { char *path; int mode; } */ *uap; { return (kern_fchmodat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->mode, 0)); } #ifndef _SYS_SYSPROTO_H_ struct fchmodat_args { int dirfd; char *path; mode_t mode; int flag; } #endif int sys_fchmodat(struct thread *td, struct fchmodat_args *uap) { int flag = uap->flag; int fd = uap->fd; char *path = uap->path; mode_t mode = uap->mode; if (flag & ~AT_SYMLINK_NOFOLLOW) return (EINVAL); return (kern_fchmodat(td, fd, path, UIO_USERSPACE, mode, flag)); } /* * Change mode of a file given path name (don't follow links.) */ #ifndef _SYS_SYSPROTO_H_ struct lchmod_args { char *path; int mode; }; #endif int sys_lchmod(td, uap) struct thread *td; register struct lchmod_args /* { char *path; int mode; } */ *uap; { return (kern_fchmodat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->mode, AT_SYMLINK_NOFOLLOW)); } int kern_fchmodat(struct thread *td, int fd, char *path, enum uio_seg pathseg, mode_t mode, int flag) { struct nameidata nd; cap_rights_t rights; int error, follow; AUDIT_ARG_MODE(mode); follow = (flag & AT_SYMLINK_NOFOLLOW) ? NOFOLLOW : FOLLOW; NDINIT_ATRIGHTS(&nd, LOOKUP, follow | AUDITVNODE1, pathseg, path, fd, cap_rights_init(&rights, CAP_FCHMOD), td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); error = setfmode(td, td->td_ucred, nd.ni_vp, mode); vrele(nd.ni_vp); return (error); } /* * Change mode of a file given a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct fchmod_args { int fd; int mode; }; #endif int sys_fchmod(struct thread *td, struct fchmod_args *uap) { struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(uap->fd); AUDIT_ARG_MODE(uap->mode); error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FCHMOD), &fp); if (error != 0) return (error); error = fo_chmod(fp, uap->mode, td->td_ucred, td); fdrop(fp, td); return (error); } /* * Common implementation for chown(), lchown(), and fchown() */ int setfown(td, cred, vp, uid, gid) struct thread *td; struct ucred *cred; struct vnode *vp; uid_t uid; gid_t gid; { struct mount *mp; struct vattr vattr; int error; if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); VATTR_NULL(&vattr); vattr.va_uid = uid; vattr.va_gid = gid; #ifdef MAC error = mac_vnode_check_setowner(cred, vp, vattr.va_uid, vattr.va_gid); if (error == 0) #endif error = VOP_SETATTR(vp, &vattr, cred); VOP_UNLOCK(vp, 0); vn_finished_write(mp); return (error); } /* * Set ownership given a path name. */ #ifndef _SYS_SYSPROTO_H_ struct chown_args { char *path; int uid; int gid; }; #endif int sys_chown(td, uap) struct thread *td; register struct chown_args /* { char *path; int uid; int gid; } */ *uap; { return (kern_fchownat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->uid, uap->gid, 0)); } #ifndef _SYS_SYSPROTO_H_ struct fchownat_args { int fd; const char * path; uid_t uid; gid_t gid; int flag; }; #endif int sys_fchownat(struct thread *td, struct fchownat_args *uap) { int flag; flag = uap->flag; if (flag & ~AT_SYMLINK_NOFOLLOW) return (EINVAL); return (kern_fchownat(td, uap->fd, uap->path, UIO_USERSPACE, uap->uid, uap->gid, uap->flag)); } int kern_fchownat(struct thread *td, int fd, char *path, enum uio_seg pathseg, int uid, int gid, int flag) { struct nameidata nd; cap_rights_t rights; int error, follow; AUDIT_ARG_OWNER(uid, gid); follow = (flag & AT_SYMLINK_NOFOLLOW) ? NOFOLLOW : FOLLOW; NDINIT_ATRIGHTS(&nd, LOOKUP, follow | AUDITVNODE1, pathseg, path, fd, cap_rights_init(&rights, CAP_FCHOWN), td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); error = setfown(td, td->td_ucred, nd.ni_vp, uid, gid); vrele(nd.ni_vp); return (error); } /* * Set ownership given a path name, do not cross symlinks. */ #ifndef _SYS_SYSPROTO_H_ struct lchown_args { char *path; int uid; int gid; }; #endif int sys_lchown(td, uap) struct thread *td; register struct lchown_args /* { char *path; int uid; int gid; } */ *uap; { return (kern_fchownat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->uid, uap->gid, AT_SYMLINK_NOFOLLOW)); } /* * Set ownership given a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct fchown_args { int fd; int uid; int gid; }; #endif int sys_fchown(td, uap) struct thread *td; register struct fchown_args /* { int fd; int uid; int gid; } */ *uap; { struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(uap->fd); AUDIT_ARG_OWNER(uap->uid, uap->gid); error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FCHOWN), &fp); if (error != 0) return (error); error = fo_chown(fp, uap->uid, uap->gid, td->td_ucred, td); fdrop(fp, td); return (error); } /* * Common implementation code for utimes(), lutimes(), and futimes(). */ static int getutimes(usrtvp, tvpseg, tsp) const struct timeval *usrtvp; enum uio_seg tvpseg; struct timespec *tsp; { struct timeval tv[2]; const struct timeval *tvp; int error; if (usrtvp == NULL) { vfs_timestamp(&tsp[0]); tsp[1] = tsp[0]; } else { if (tvpseg == UIO_SYSSPACE) { tvp = usrtvp; } else { if ((error = copyin(usrtvp, tv, sizeof(tv))) != 0) return (error); tvp = tv; } if (tvp[0].tv_usec < 0 || tvp[0].tv_usec >= 1000000 || tvp[1].tv_usec < 0 || tvp[1].tv_usec >= 1000000) return (EINVAL); TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]); TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]); } return (0); } /* * Common implementation code for futimens(), utimensat(). */ #define UTIMENS_NULL 0x1 #define UTIMENS_EXIT 0x2 static int getutimens(const struct timespec *usrtsp, enum uio_seg tspseg, struct timespec *tsp, int *retflags) { struct timespec tsnow; int error; vfs_timestamp(&tsnow); *retflags = 0; if (usrtsp == NULL) { tsp[0] = tsnow; tsp[1] = tsnow; *retflags |= UTIMENS_NULL; return (0); } if (tspseg == UIO_SYSSPACE) { tsp[0] = usrtsp[0]; tsp[1] = usrtsp[1]; } else if ((error = copyin(usrtsp, tsp, sizeof(*tsp) * 2)) != 0) return (error); if (tsp[0].tv_nsec == UTIME_OMIT && tsp[1].tv_nsec == UTIME_OMIT) *retflags |= UTIMENS_EXIT; if (tsp[0].tv_nsec == UTIME_NOW && tsp[1].tv_nsec == UTIME_NOW) *retflags |= UTIMENS_NULL; if (tsp[0].tv_nsec == UTIME_OMIT) tsp[0].tv_sec = VNOVAL; else if (tsp[0].tv_nsec == UTIME_NOW) tsp[0] = tsnow; else if (tsp[0].tv_nsec < 0 || tsp[0].tv_nsec >= 1000000000L) return (EINVAL); if (tsp[1].tv_nsec == UTIME_OMIT) tsp[1].tv_sec = VNOVAL; else if (tsp[1].tv_nsec == UTIME_NOW) tsp[1] = tsnow; else if (tsp[1].tv_nsec < 0 || tsp[1].tv_nsec >= 1000000000L) return (EINVAL); return (0); } /* * Common implementation code for utimes(), lutimes(), futimes(), futimens(), * and utimensat(). */ static int setutimes(td, vp, ts, numtimes, nullflag) struct thread *td; struct vnode *vp; const struct timespec *ts; int numtimes; int nullflag; { struct mount *mp; struct vattr vattr; int error, setbirthtime; if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); setbirthtime = 0; if (numtimes < 3 && !VOP_GETATTR(vp, &vattr, td->td_ucred) && timespeccmp(&ts[1], &vattr.va_birthtime, < )) setbirthtime = 1; VATTR_NULL(&vattr); vattr.va_atime = ts[0]; vattr.va_mtime = ts[1]; if (setbirthtime) vattr.va_birthtime = ts[1]; if (numtimes > 2) vattr.va_birthtime = ts[2]; if (nullflag) vattr.va_vaflags |= VA_UTIMES_NULL; #ifdef MAC error = mac_vnode_check_setutimes(td->td_ucred, vp, vattr.va_atime, vattr.va_mtime); #endif if (error == 0) error = VOP_SETATTR(vp, &vattr, td->td_ucred); VOP_UNLOCK(vp, 0); vn_finished_write(mp); return (error); } /* * Set the access and modification times of a file. */ #ifndef _SYS_SYSPROTO_H_ struct utimes_args { char *path; struct timeval *tptr; }; #endif int sys_utimes(td, uap) struct thread *td; register struct utimes_args /* { char *path; struct timeval *tptr; } */ *uap; { return (kern_utimesat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->tptr, UIO_USERSPACE)); } #ifndef _SYS_SYSPROTO_H_ struct futimesat_args { int fd; const char * path; const struct timeval * times; }; #endif int sys_futimesat(struct thread *td, struct futimesat_args *uap) { return (kern_utimesat(td, uap->fd, uap->path, UIO_USERSPACE, uap->times, UIO_USERSPACE)); } int kern_utimesat(struct thread *td, int fd, char *path, enum uio_seg pathseg, struct timeval *tptr, enum uio_seg tptrseg) { struct nameidata nd; struct timespec ts[2]; cap_rights_t rights; int error; if ((error = getutimes(tptr, tptrseg, ts)) != 0) return (error); NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | AUDITVNODE1, pathseg, path, fd, cap_rights_init(&rights, CAP_FUTIMES), td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); error = setutimes(td, nd.ni_vp, ts, 2, tptr == NULL); vrele(nd.ni_vp); return (error); } /* * Set the access and modification times of a file. */ #ifndef _SYS_SYSPROTO_H_ struct lutimes_args { char *path; struct timeval *tptr; }; #endif int sys_lutimes(td, uap) struct thread *td; register struct lutimes_args /* { char *path; struct timeval *tptr; } */ *uap; { return (kern_lutimes(td, uap->path, UIO_USERSPACE, uap->tptr, UIO_USERSPACE)); } int kern_lutimes(struct thread *td, char *path, enum uio_seg pathseg, struct timeval *tptr, enum uio_seg tptrseg) { struct timespec ts[2]; struct nameidata nd; int error; if ((error = getutimes(tptr, tptrseg, ts)) != 0) return (error); NDINIT(&nd, LOOKUP, NOFOLLOW | AUDITVNODE1, pathseg, path, td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); error = setutimes(td, nd.ni_vp, ts, 2, tptr == NULL); vrele(nd.ni_vp); return (error); } /* * Set the access and modification times of a file. */ #ifndef _SYS_SYSPROTO_H_ struct futimes_args { int fd; struct timeval *tptr; }; #endif int sys_futimes(td, uap) struct thread *td; register struct futimes_args /* { int fd; struct timeval *tptr; } */ *uap; { return (kern_futimes(td, uap->fd, uap->tptr, UIO_USERSPACE)); } int kern_futimes(struct thread *td, int fd, struct timeval *tptr, enum uio_seg tptrseg) { struct timespec ts[2]; struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(fd); error = getutimes(tptr, tptrseg, ts); if (error != 0) return (error); error = getvnode(td, fd, cap_rights_init(&rights, CAP_FUTIMES), &fp); if (error != 0) return (error); #ifdef AUDIT vn_lock(fp->f_vnode, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(fp->f_vnode); VOP_UNLOCK(fp->f_vnode, 0); #endif error = setutimes(td, fp->f_vnode, ts, 2, tptr == NULL); fdrop(fp, td); return (error); } int sys_futimens(struct thread *td, struct futimens_args *uap) { return (kern_futimens(td, uap->fd, uap->times, UIO_USERSPACE)); } int kern_futimens(struct thread *td, int fd, struct timespec *tptr, enum uio_seg tptrseg) { struct timespec ts[2]; struct file *fp; cap_rights_t rights; int error, flags; AUDIT_ARG_FD(fd); error = getutimens(tptr, tptrseg, ts, &flags); if (error != 0) return (error); if (flags & UTIMENS_EXIT) return (0); error = getvnode(td, fd, cap_rights_init(&rights, CAP_FUTIMES), &fp); if (error != 0) return (error); #ifdef AUDIT vn_lock(fp->f_vnode, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(fp->f_vnode); VOP_UNLOCK(fp->f_vnode, 0); #endif error = setutimes(td, fp->f_vnode, ts, 2, flags & UTIMENS_NULL); fdrop(fp, td); return (error); } int sys_utimensat(struct thread *td, struct utimensat_args *uap) { return (kern_utimensat(td, uap->fd, uap->path, UIO_USERSPACE, uap->times, UIO_USERSPACE, uap->flag)); } int kern_utimensat(struct thread *td, int fd, char *path, enum uio_seg pathseg, struct timespec *tptr, enum uio_seg tptrseg, int flag) { struct nameidata nd; struct timespec ts[2]; cap_rights_t rights; int error, flags; if (flag & ~AT_SYMLINK_NOFOLLOW) return (EINVAL); if ((error = getutimens(tptr, tptrseg, ts, &flags)) != 0) return (error); NDINIT_ATRIGHTS(&nd, LOOKUP, ((flag & AT_SYMLINK_NOFOLLOW) ? NOFOLLOW : FOLLOW) | AUDITVNODE1, pathseg, path, fd, cap_rights_init(&rights, CAP_FUTIMES), td); if ((error = namei(&nd)) != 0) return (error); /* * We are allowed to call namei() regardless of 2xUTIME_OMIT. * POSIX states: * "If both tv_nsec fields are UTIME_OMIT... EACCESS may be detected." * "Search permission is denied by a component of the path prefix." */ NDFREE(&nd, NDF_ONLY_PNBUF); if ((flags & UTIMENS_EXIT) == 0) error = setutimes(td, nd.ni_vp, ts, 2, flags & UTIMENS_NULL); vrele(nd.ni_vp); return (error); } /* * Truncate a file given its path name. */ #ifndef _SYS_SYSPROTO_H_ struct truncate_args { char *path; int pad; off_t length; }; #endif int sys_truncate(td, uap) struct thread *td; register struct truncate_args /* { char *path; int pad; off_t length; } */ *uap; { return (kern_truncate(td, uap->path, UIO_USERSPACE, uap->length)); } int kern_truncate(struct thread *td, char *path, enum uio_seg pathseg, off_t length) { struct mount *mp; struct vnode *vp; void *rl_cookie; struct vattr vattr; struct nameidata nd; int error; if (length < 0) return(EINVAL); NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNODE1, pathseg, path, td); if ((error = namei(&nd)) != 0) return (error); vp = nd.ni_vp; rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) { vn_rangelock_unlock(vp, rl_cookie); vrele(vp); return (error); } NDFREE(&nd, NDF_ONLY_PNBUF); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); if (vp->v_type == VDIR) error = EISDIR; #ifdef MAC else if ((error = mac_vnode_check_write(td->td_ucred, NOCRED, vp))) { } #endif else if ((error = vn_writechk(vp)) == 0 && (error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td)) == 0) { VATTR_NULL(&vattr); vattr.va_size = length; error = VOP_SETATTR(vp, &vattr, td->td_ucred); } VOP_UNLOCK(vp, 0); vn_finished_write(mp); vn_rangelock_unlock(vp, rl_cookie); vrele(vp); return (error); } #if defined(COMPAT_43) /* * Truncate a file given its path name. */ #ifndef _SYS_SYSPROTO_H_ struct otruncate_args { char *path; long length; }; #endif int otruncate(td, uap) struct thread *td; register struct otruncate_args /* { char *path; long length; } */ *uap; { struct truncate_args /* { char *path; int pad; off_t length; } */ nuap; nuap.path = uap->path; nuap.length = uap->length; return (sys_truncate(td, &nuap)); } #endif /* COMPAT_43 */ #if defined(COMPAT_FREEBSD6) /* Versions with the pad argument */ int freebsd6_truncate(struct thread *td, struct freebsd6_truncate_args *uap) { struct truncate_args ouap; ouap.path = uap->path; ouap.length = uap->length; return (sys_truncate(td, &ouap)); } int freebsd6_ftruncate(struct thread *td, struct freebsd6_ftruncate_args *uap) { struct ftruncate_args ouap; ouap.fd = uap->fd; ouap.length = uap->length; return (sys_ftruncate(td, &ouap)); } #endif int kern_fsync(struct thread *td, int fd, bool fullsync) { struct vnode *vp; struct mount *mp; struct file *fp; cap_rights_t rights; int error, lock_flags; AUDIT_ARG_FD(fd); error = getvnode(td, fd, cap_rights_init(&rights, CAP_FSYNC), &fp); if (error != 0) return (error); vp = fp->f_vnode; #if 0 if (!fullsync) /* XXXKIB: compete outstanding aio writes */; #endif error = vn_start_write(vp, &mp, V_WAIT | PCATCH); if (error != 0) goto drop; if (MNT_SHARED_WRITES(mp) || ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) { lock_flags = LK_SHARED; } else { lock_flags = LK_EXCLUSIVE; } vn_lock(vp, lock_flags | LK_RETRY); AUDIT_ARG_VNODE1(vp); if (vp->v_object != NULL) { VM_OBJECT_WLOCK(vp->v_object); vm_object_page_clean(vp->v_object, 0, 0, 0); VM_OBJECT_WUNLOCK(vp->v_object); } error = fullsync ? VOP_FSYNC(vp, MNT_WAIT, td) : VOP_FDATASYNC(vp, td); VOP_UNLOCK(vp, 0); vn_finished_write(mp); drop: fdrop(fp, td); return (error); } /* * Sync an open file. */ #ifndef _SYS_SYSPROTO_H_ struct fsync_args { int fd; }; #endif int sys_fsync(struct thread *td, struct fsync_args *uap) { return (kern_fsync(td, uap->fd, true)); } int sys_fdatasync(struct thread *td, struct fdatasync_args *uap) { return (kern_fsync(td, uap->fd, false)); } /* * Rename files. Source and destination must either both be directories, or * both not be directories. If target is a directory, it must be empty. */ #ifndef _SYS_SYSPROTO_H_ struct rename_args { char *from; char *to; }; #endif int sys_rename(td, uap) struct thread *td; register struct rename_args /* { char *from; char *to; } */ *uap; { return (kern_renameat(td, AT_FDCWD, uap->from, AT_FDCWD, uap->to, UIO_USERSPACE)); } #ifndef _SYS_SYSPROTO_H_ struct renameat_args { int oldfd; char *old; int newfd; char *new; }; #endif int sys_renameat(struct thread *td, struct renameat_args *uap) { return (kern_renameat(td, uap->oldfd, uap->old, uap->newfd, uap->new, UIO_USERSPACE)); } int kern_renameat(struct thread *td, int oldfd, char *old, int newfd, char *new, enum uio_seg pathseg) { struct mount *mp = NULL; struct vnode *tvp, *fvp, *tdvp; struct nameidata fromnd, tond; cap_rights_t rights; int error; again: bwillwrite(); #ifdef MAC NDINIT_ATRIGHTS(&fromnd, DELETE, LOCKPARENT | LOCKLEAF | SAVESTART | AUDITVNODE1, pathseg, old, oldfd, cap_rights_init(&rights, CAP_RENAMEAT_SOURCE), td); #else NDINIT_ATRIGHTS(&fromnd, DELETE, WANTPARENT | SAVESTART | AUDITVNODE1, pathseg, old, oldfd, cap_rights_init(&rights, CAP_RENAMEAT_SOURCE), td); #endif if ((error = namei(&fromnd)) != 0) return (error); #ifdef MAC error = mac_vnode_check_rename_from(td->td_ucred, fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd); VOP_UNLOCK(fromnd.ni_dvp, 0); if (fromnd.ni_dvp != fromnd.ni_vp) VOP_UNLOCK(fromnd.ni_vp, 0); #endif fvp = fromnd.ni_vp; NDINIT_ATRIGHTS(&tond, RENAME, LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART | AUDITVNODE2, pathseg, new, newfd, cap_rights_init(&rights, CAP_RENAMEAT_TARGET), td); if (fromnd.ni_vp->v_type == VDIR) tond.ni_cnd.cn_flags |= WILLBEDIR; if ((error = namei(&tond)) != 0) { /* Translate error code for rename("dir1", "dir2/."). */ if (error == EISDIR && fvp->v_type == VDIR) error = EINVAL; NDFREE(&fromnd, NDF_ONLY_PNBUF); vrele(fromnd.ni_dvp); vrele(fvp); goto out1; } tdvp = tond.ni_dvp; tvp = tond.ni_vp; error = vn_start_write(fvp, &mp, V_NOWAIT); if (error != 0) { NDFREE(&fromnd, NDF_ONLY_PNBUF); NDFREE(&tond, NDF_ONLY_PNBUF); if (tvp != NULL) vput(tvp); if (tdvp == tvp) vrele(tdvp); else vput(tdvp); vrele(fromnd.ni_dvp); vrele(fvp); vrele(tond.ni_startdir); if (fromnd.ni_startdir != NULL) vrele(fromnd.ni_startdir); error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); if (error != 0) return (error); goto again; } if (tvp != NULL) { if (fvp->v_type == VDIR && tvp->v_type != VDIR) { error = ENOTDIR; goto out; } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { error = EISDIR; goto out; } #ifdef CAPABILITIES if (newfd != AT_FDCWD) { /* * If the target already exists we require CAP_UNLINKAT * from 'newfd'. */ error = cap_check(&tond.ni_filecaps.fc_rights, cap_rights_init(&rights, CAP_UNLINKAT)); if (error != 0) goto out; } #endif } if (fvp == tdvp) { error = EINVAL; goto out; } /* * If the source is the same as the destination (that is, if they * are links to the same vnode), then there is nothing to do. */ if (fvp == tvp) error = -1; #ifdef MAC else error = mac_vnode_check_rename_to(td->td_ucred, tdvp, tond.ni_vp, fromnd.ni_dvp == tdvp, &tond.ni_cnd); #endif out: if (error == 0) { error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd, tond.ni_dvp, tond.ni_vp, &tond.ni_cnd); NDFREE(&fromnd, NDF_ONLY_PNBUF); NDFREE(&tond, NDF_ONLY_PNBUF); } else { NDFREE(&fromnd, NDF_ONLY_PNBUF); NDFREE(&tond, NDF_ONLY_PNBUF); if (tvp != NULL) vput(tvp); if (tdvp == tvp) vrele(tdvp); else vput(tdvp); vrele(fromnd.ni_dvp); vrele(fvp); } vrele(tond.ni_startdir); vn_finished_write(mp); out1: if (fromnd.ni_startdir) vrele(fromnd.ni_startdir); if (error == -1) return (0); return (error); } /* * Make a directory file. */ #ifndef _SYS_SYSPROTO_H_ struct mkdir_args { char *path; int mode; }; #endif int sys_mkdir(td, uap) struct thread *td; register struct mkdir_args /* { char *path; int mode; } */ *uap; { return (kern_mkdirat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->mode)); } #ifndef _SYS_SYSPROTO_H_ struct mkdirat_args { int fd; char *path; mode_t mode; }; #endif int sys_mkdirat(struct thread *td, struct mkdirat_args *uap) { return (kern_mkdirat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode)); } int kern_mkdirat(struct thread *td, int fd, char *path, enum uio_seg segflg, int mode) { struct mount *mp; struct vnode *vp; struct vattr vattr; struct nameidata nd; cap_rights_t rights; int error; AUDIT_ARG_MODE(mode); restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE1 | NOCACHE, segflg, path, fd, cap_rights_init(&rights, CAP_MKDIRAT), td); nd.ni_cnd.cn_flags |= WILLBEDIR; if ((error = namei(&nd)) != 0) return (error); vp = nd.ni_vp; if (vp != NULL) { NDFREE(&nd, NDF_ONLY_PNBUF); /* * XXX namei called with LOCKPARENT but not LOCKLEAF has * the strange behaviour of leaving the vnode unlocked * if the target is the same vnode as the parent. */ if (vp == nd.ni_dvp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); vrele(vp); return (EEXIST); } if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) return (error); goto restart; } VATTR_NULL(&vattr); vattr.va_type = VDIR; vattr.va_mode = (mode & ACCESSPERMS) &~ td->td_proc->p_fd->fd_cmask; #ifdef MAC error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, &vattr); if (error != 0) goto out; #endif error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); #ifdef MAC out: #endif NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if (error == 0) vput(nd.ni_vp); vn_finished_write(mp); return (error); } /* * Remove a directory file. */ #ifndef _SYS_SYSPROTO_H_ struct rmdir_args { char *path; }; #endif int sys_rmdir(td, uap) struct thread *td; struct rmdir_args /* { char *path; } */ *uap; { return (kern_rmdirat(td, AT_FDCWD, uap->path, UIO_USERSPACE)); } int kern_rmdirat(struct thread *td, int fd, char *path, enum uio_seg pathseg) { struct mount *mp; struct vnode *vp; struct nameidata nd; cap_rights_t rights; int error; restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, DELETE, LOCKPARENT | LOCKLEAF | AUDITVNODE1, pathseg, path, fd, cap_rights_init(&rights, CAP_UNLINKAT), td); if ((error = namei(&nd)) != 0) return (error); vp = nd.ni_vp; if (vp->v_type != VDIR) { error = ENOTDIR; goto out; } /* * No rmdir "." please. */ if (nd.ni_dvp == vp) { error = EINVAL; goto out; } /* * The root of a mounted filesystem cannot be deleted. */ if (vp->v_vflag & VV_ROOT) { error = EBUSY; goto out; } #ifdef MAC error = mac_vnode_check_unlink(td->td_ucred, nd.ni_dvp, vp, &nd.ni_cnd); if (error != 0) goto out; #endif if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(vp); if (nd.ni_dvp == vp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) return (error); goto restart; } vfs_notify_upper(vp, VFS_NOTIFY_UPPER_UNLINK); error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); vn_finished_write(mp); out: NDFREE(&nd, NDF_ONLY_PNBUF); vput(vp); if (nd.ni_dvp == vp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); return (error); } #ifdef COMPAT_43 /* * Read a block of directory entries in a filesystem independent format. */ #ifndef _SYS_SYSPROTO_H_ struct ogetdirentries_args { int fd; char *buf; u_int count; long *basep; }; #endif int ogetdirentries(struct thread *td, struct ogetdirentries_args *uap) { long loff; int error; error = kern_ogetdirentries(td, uap, &loff); if (error == 0) error = copyout(&loff, uap->basep, sizeof(long)); return (error); } int kern_ogetdirentries(struct thread *td, struct ogetdirentries_args *uap, long *ploff) { struct vnode *vp; struct file *fp; struct uio auio, kuio; struct iovec aiov, kiov; struct dirent *dp, *edp; cap_rights_t rights; caddr_t dirbuf; int error, eofflag, readcnt; long loff; off_t foffset; /* XXX arbitrary sanity limit on `count'. */ if (uap->count > 64 * 1024) return (EINVAL); error = getvnode(td, uap->fd, cap_rights_init(&rights, CAP_READ), &fp); if (error != 0) return (error); if ((fp->f_flag & FREAD) == 0) { fdrop(fp, td); return (EBADF); } vp = fp->f_vnode; foffset = foffset_lock(fp, 0); unionread: if (vp->v_type != VDIR) { foffset_unlock(fp, foffset, 0); fdrop(fp, td); return (EINVAL); } aiov.iov_base = uap->buf; aiov.iov_len = uap->count; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_READ; auio.uio_segflg = UIO_USERSPACE; auio.uio_td = td; auio.uio_resid = uap->count; vn_lock(vp, LK_SHARED | LK_RETRY); loff = auio.uio_offset = foffset; #ifdef MAC error = mac_vnode_check_readdir(td->td_ucred, vp); if (error != 0) { VOP_UNLOCK(vp, 0); foffset_unlock(fp, foffset, FOF_NOUPDATE); fdrop(fp, td); return (error); } #endif # if (BYTE_ORDER != LITTLE_ENDIAN) if (vp->v_mount->mnt_maxsymlinklen <= 0) { error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL); foffset = auio.uio_offset; } else # endif { kuio = auio; kuio.uio_iov = &kiov; kuio.uio_segflg = UIO_SYSSPACE; kiov.iov_len = uap->count; dirbuf = malloc(uap->count, M_TEMP, M_WAITOK); kiov.iov_base = dirbuf; error = VOP_READDIR(vp, &kuio, fp->f_cred, &eofflag, NULL, NULL); foffset = kuio.uio_offset; if (error == 0) { readcnt = uap->count - kuio.uio_resid; edp = (struct dirent *)&dirbuf[readcnt]; for (dp = (struct dirent *)dirbuf; dp < edp; ) { # if (BYTE_ORDER == LITTLE_ENDIAN) /* * The expected low byte of * dp->d_namlen is our dp->d_type. * The high MBZ byte of dp->d_namlen * is our dp->d_namlen. */ dp->d_type = dp->d_namlen; dp->d_namlen = 0; # else /* * The dp->d_type is the high byte * of the expected dp->d_namlen, * so must be zero'ed. */ dp->d_type = 0; # endif if (dp->d_reclen > 0) { dp = (struct dirent *) ((char *)dp + dp->d_reclen); } else { error = EIO; break; } } if (dp >= edp) error = uiomove(dirbuf, readcnt, &auio); } free(dirbuf, M_TEMP); } if (error != 0) { VOP_UNLOCK(vp, 0); foffset_unlock(fp, foffset, 0); fdrop(fp, td); return (error); } if (uap->count == auio.uio_resid && (vp->v_vflag & VV_ROOT) && (vp->v_mount->mnt_flag & MNT_UNION)) { struct vnode *tvp = vp; vp = vp->v_mount->mnt_vnodecovered; VREF(vp); fp->f_vnode = vp; fp->f_data = vp; foffset = 0; vput(tvp); goto unionread; } VOP_UNLOCK(vp, 0); foffset_unlock(fp, foffset, 0); fdrop(fp, td); td->td_retval[0] = uap->count - auio.uio_resid; if (error == 0) *ploff = loff; return (error); } #endif /* COMPAT_43 */ /* * Read a block of directory entries in a filesystem independent format. */ #ifndef _SYS_SYSPROTO_H_ struct getdirentries_args { int fd; char *buf; u_int count; long *basep; }; #endif int sys_getdirentries(td, uap) struct thread *td; register struct getdirentries_args /* { int fd; char *buf; u_int count; long *basep; } */ *uap; { long base; int error; error = kern_getdirentries(td, uap->fd, uap->buf, uap->count, &base, NULL, UIO_USERSPACE); if (error != 0) return (error); if (uap->basep != NULL) error = copyout(&base, uap->basep, sizeof(long)); return (error); } int kern_getdirentries(struct thread *td, int fd, char *buf, u_int count, long *basep, ssize_t *residp, enum uio_seg bufseg) { struct vnode *vp; struct file *fp; struct uio auio; struct iovec aiov; cap_rights_t rights; long loff; int error, eofflag; off_t foffset; AUDIT_ARG_FD(fd); if (count > IOSIZE_MAX) return (EINVAL); auio.uio_resid = count; error = getvnode(td, fd, cap_rights_init(&rights, CAP_READ), &fp); if (error != 0) return (error); if ((fp->f_flag & FREAD) == 0) { fdrop(fp, td); return (EBADF); } vp = fp->f_vnode; foffset = foffset_lock(fp, 0); unionread: if (vp->v_type != VDIR) { error = EINVAL; goto fail; } aiov.iov_base = buf; aiov.iov_len = count; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_READ; auio.uio_segflg = bufseg; auio.uio_td = td; vn_lock(vp, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(vp); loff = auio.uio_offset = foffset; #ifdef MAC error = mac_vnode_check_readdir(td->td_ucred, vp); if (error == 0) #endif error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL); foffset = auio.uio_offset; if (error != 0) { VOP_UNLOCK(vp, 0); goto fail; } if (count == auio.uio_resid && (vp->v_vflag & VV_ROOT) && (vp->v_mount->mnt_flag & MNT_UNION)) { struct vnode *tvp = vp; vp = vp->v_mount->mnt_vnodecovered; VREF(vp); fp->f_vnode = vp; fp->f_data = vp; foffset = 0; vput(tvp); goto unionread; } VOP_UNLOCK(vp, 0); *basep = loff; if (residp != NULL) *residp = auio.uio_resid; td->td_retval[0] = count - auio.uio_resid; fail: foffset_unlock(fp, foffset, 0); fdrop(fp, td); return (error); } #ifndef _SYS_SYSPROTO_H_ struct getdents_args { int fd; char *buf; size_t count; }; #endif int sys_getdents(td, uap) struct thread *td; register struct getdents_args /* { int fd; char *buf; u_int count; } */ *uap; { struct getdirentries_args ap; ap.fd = uap->fd; ap.buf = uap->buf; ap.count = uap->count; ap.basep = NULL; return (sys_getdirentries(td, &ap)); } /* * Set the mode mask for creation of filesystem nodes. */ #ifndef _SYS_SYSPROTO_H_ struct umask_args { int newmask; }; #endif int sys_umask(td, uap) struct thread *td; struct umask_args /* { int newmask; } */ *uap; { struct filedesc *fdp; fdp = td->td_proc->p_fd; FILEDESC_XLOCK(fdp); td->td_retval[0] = fdp->fd_cmask; fdp->fd_cmask = uap->newmask & ALLPERMS; FILEDESC_XUNLOCK(fdp); return (0); } /* * Void all references to file by ripping underlying filesystem away from * vnode. */ #ifndef _SYS_SYSPROTO_H_ struct revoke_args { char *path; }; #endif int sys_revoke(td, uap) struct thread *td; register struct revoke_args /* { char *path; } */ *uap; { struct vnode *vp; struct vattr vattr; struct nameidata nd; int error; NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_USERSPACE, uap->path, td); if ((error = namei(&nd)) != 0) return (error); vp = nd.ni_vp; NDFREE(&nd, NDF_ONLY_PNBUF); if (vp->v_type != VCHR || vp->v_rdev == NULL) { error = EINVAL; goto out; } #ifdef MAC error = mac_vnode_check_revoke(td->td_ucred, vp); if (error != 0) goto out; #endif error = VOP_GETATTR(vp, &vattr, td->td_ucred); if (error != 0) goto out; if (td->td_ucred->cr_uid != vattr.va_uid) { error = priv_check(td, PRIV_VFS_ADMIN); if (error != 0) goto out; } if (vcount(vp) > 1) VOP_REVOKE(vp, REVOKEALL); out: vput(vp); return (error); } /* * Convert a user file descriptor to a kernel file entry and check that, if it * is a capability, the correct rights are present. A reference on the file * entry is held upon returning. */ int getvnode(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) { struct file *fp; int error; error = fget_unlocked(td->td_proc->p_fd, fd, rightsp, &fp, NULL); if (error != 0) return (error); /* * The file could be not of the vnode type, or it may be not * yet fully initialized, in which case the f_vnode pointer * may be set, but f_ops is still badfileops. E.g., * devfs_open() transiently create such situation to * facilitate csw d_fdopen(). * * Dupfdopen() handling in kern_openat() installs the * half-baked file into the process descriptor table, allowing * other thread to dereference it. Guard against the race by * checking f_ops. */ if (fp->f_vnode == NULL || fp->f_ops == &badfileops) { fdrop(fp, td); return (EINVAL); } *fpp = fp; return (0); } /* * Get an (NFS) file handle. */ #ifndef _SYS_SYSPROTO_H_ struct lgetfh_args { char *fname; fhandle_t *fhp; }; #endif int sys_lgetfh(td, uap) struct thread *td; register struct lgetfh_args *uap; { struct nameidata nd; fhandle_t fh; register struct vnode *vp; int error; error = priv_check(td, PRIV_VFS_GETFH); if (error != 0) return (error); NDINIT(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | AUDITVNODE1, UIO_USERSPACE, uap->fname, td); error = namei(&nd); if (error != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); vp = nd.ni_vp; bzero(&fh, sizeof(fh)); fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid; error = VOP_VPTOFH(vp, &fh.fh_fid); vput(vp); if (error == 0) error = copyout(&fh, uap->fhp, sizeof (fh)); return (error); } #ifndef _SYS_SYSPROTO_H_ struct getfh_args { char *fname; fhandle_t *fhp; }; #endif int sys_getfh(td, uap) struct thread *td; register struct getfh_args *uap; { struct nameidata nd; fhandle_t fh; register struct vnode *vp; int error; error = priv_check(td, PRIV_VFS_GETFH); if (error != 0) return (error); NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_USERSPACE, uap->fname, td); error = namei(&nd); if (error != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); vp = nd.ni_vp; bzero(&fh, sizeof(fh)); fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid; error = VOP_VPTOFH(vp, &fh.fh_fid); vput(vp); if (error == 0) error = copyout(&fh, uap->fhp, sizeof (fh)); return (error); } /* * syscall for the rpc.lockd to use to translate a NFS file handle into an * open descriptor. * * warning: do not remove the priv_check() call or this becomes one giant * security hole. */ #ifndef _SYS_SYSPROTO_H_ struct fhopen_args { const struct fhandle *u_fhp; int flags; }; #endif int sys_fhopen(td, uap) struct thread *td; struct fhopen_args /* { const struct fhandle *u_fhp; int flags; } */ *uap; { struct mount *mp; struct vnode *vp; struct fhandle fhp; struct file *fp; int fmode, error; int indx; error = priv_check(td, PRIV_VFS_FHOPEN); if (error != 0) return (error); indx = -1; fmode = FFLAGS(uap->flags); /* why not allow a non-read/write open for our lockd? */ if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT)) return (EINVAL); error = copyin(uap->u_fhp, &fhp, sizeof(fhp)); if (error != 0) return(error); /* find the mount point */ mp = vfs_busyfs(&fhp.fh_fsid); if (mp == NULL) return (ESTALE); /* now give me my vnode, it gets returned to me locked */ error = VFS_FHTOVP(mp, &fhp.fh_fid, LK_EXCLUSIVE, &vp); vfs_unbusy(mp); if (error != 0) return (error); error = falloc_noinstall(td, &fp); if (error != 0) { vput(vp); return (error); } /* * An extra reference on `fp' has been held for us by * falloc_noinstall(). */ #ifdef INVARIANTS td->td_dupfd = -1; #endif error = vn_open_vnode(vp, fmode, td->td_ucred, td, fp); if (error != 0) { KASSERT(fp->f_ops == &badfileops, ("VOP_OPEN in fhopen() set f_ops")); KASSERT(td->td_dupfd < 0, ("fhopen() encountered fdopen()")); vput(vp); goto bad; } #ifdef INVARIANTS td->td_dupfd = 0; #endif fp->f_vnode = vp; fp->f_seqcount = 1; finit(fp, (fmode & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE, vp, &vnops); VOP_UNLOCK(vp, 0); if ((fmode & O_TRUNC) != 0) { error = fo_truncate(fp, 0, td->td_ucred, td); if (error != 0) goto bad; } error = finstall(td, fp, &indx, fmode, NULL); bad: fdrop(fp, td); td->td_retval[0] = indx; return (error); } /* * Stat an (NFS) file handle. */ #ifndef _SYS_SYSPROTO_H_ struct fhstat_args { struct fhandle *u_fhp; struct stat *sb; }; #endif int sys_fhstat(td, uap) struct thread *td; register struct fhstat_args /* { struct fhandle *u_fhp; struct stat *sb; } */ *uap; { struct stat sb; struct fhandle fh; int error; error = copyin(uap->u_fhp, &fh, sizeof(fh)); if (error != 0) return (error); error = kern_fhstat(td, fh, &sb); if (error == 0) error = copyout(&sb, uap->sb, sizeof(sb)); return (error); } int kern_fhstat(struct thread *td, struct fhandle fh, struct stat *sb) { struct mount *mp; struct vnode *vp; int error; error = priv_check(td, PRIV_VFS_FHSTAT); if (error != 0) return (error); if ((mp = vfs_busyfs(&fh.fh_fsid)) == NULL) return (ESTALE); error = VFS_FHTOVP(mp, &fh.fh_fid, LK_EXCLUSIVE, &vp); vfs_unbusy(mp); if (error != 0) return (error); error = vn_stat(vp, sb, td->td_ucred, NOCRED, td); vput(vp); return (error); } /* * Implement fstatfs() for (NFS) file handles. */ #ifndef _SYS_SYSPROTO_H_ struct fhstatfs_args { struct fhandle *u_fhp; struct statfs *buf; }; #endif int sys_fhstatfs(td, uap) struct thread *td; struct fhstatfs_args /* { struct fhandle *u_fhp; struct statfs *buf; } */ *uap; { struct statfs sf; fhandle_t fh; int error; error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t)); if (error != 0) return (error); error = kern_fhstatfs(td, fh, &sf); if (error != 0) return (error); return (copyout(&sf, uap->buf, sizeof(sf))); } int kern_fhstatfs(struct thread *td, fhandle_t fh, struct statfs *buf) { struct statfs *sp; struct mount *mp; struct vnode *vp; int error; error = priv_check(td, PRIV_VFS_FHSTATFS); if (error != 0) return (error); if ((mp = vfs_busyfs(&fh.fh_fsid)) == NULL) return (ESTALE); error = VFS_FHTOVP(mp, &fh.fh_fid, LK_EXCLUSIVE, &vp); if (error != 0) { vfs_unbusy(mp); return (error); } vput(vp); error = prison_canseemount(td->td_ucred, mp); if (error != 0) goto out; #ifdef MAC error = mac_mount_check_stat(td->td_ucred, mp); if (error != 0) goto out; #endif /* * Set these in case the underlying filesystem fails to do so. */ sp = &mp->mnt_stat; sp->f_version = STATFS_VERSION; sp->f_namemax = NAME_MAX; sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; error = VFS_STATFS(mp, sp); if (error == 0) *buf = *sp; out: vfs_unbusy(mp); return (error); } int kern_posix_fallocate(struct thread *td, int fd, off_t offset, off_t len) { struct file *fp; struct mount *mp; struct vnode *vp; cap_rights_t rights; off_t olen, ooffset; int error; if (offset < 0 || len <= 0) return (EINVAL); /* Check for wrap. */ if (offset > OFF_MAX - len) return (EFBIG); error = fget(td, fd, cap_rights_init(&rights, CAP_WRITE), &fp); if (error != 0) return (error); if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) { error = ESPIPE; goto out; } if ((fp->f_flag & FWRITE) == 0) { error = EBADF; goto out; } if (fp->f_type != DTYPE_VNODE) { error = ENODEV; goto out; } vp = fp->f_vnode; if (vp->v_type != VREG) { error = ENODEV; goto out; } /* Allocating blocks may take a long time, so iterate. */ for (;;) { olen = len; ooffset = offset; bwillwrite(); mp = NULL; error = vn_start_write(vp, &mp, V_WAIT | PCATCH); if (error != 0) break; error = vn_lock(vp, LK_EXCLUSIVE); if (error != 0) { vn_finished_write(mp); break; } #ifdef MAC error = mac_vnode_check_write(td->td_ucred, fp->f_cred, vp); if (error == 0) #endif error = VOP_ALLOCATE(vp, &offset, &len); VOP_UNLOCK(vp, 0); vn_finished_write(mp); if (olen + ooffset != offset + len) { panic("offset + len changed from %jx/%jx to %jx/%jx", ooffset, olen, offset, len); } if (error != 0 || len == 0) break; KASSERT(olen > len, ("Iteration did not make progress?")); maybe_yield(); } out: fdrop(fp, td); return (error); } int sys_posix_fallocate(struct thread *td, struct posix_fallocate_args *uap) { int error; error = kern_posix_fallocate(td, uap->fd, uap->offset, uap->len); return (kern_posix_error(td, error)); } /* * Unlike madvise(2), we do not make a best effort to remember every * possible caching hint. Instead, we remember the last setting with * the exception that we will allow POSIX_FADV_NORMAL to adjust the * region of any current setting. */ int kern_posix_fadvise(struct thread *td, int fd, off_t offset, off_t len, int advice) { struct fadvise_info *fa, *new; struct file *fp; struct vnode *vp; cap_rights_t rights; off_t end; int error; if (offset < 0 || len < 0 || offset > OFF_MAX - len) return (EINVAL); switch (advice) { case POSIX_FADV_SEQUENTIAL: case POSIX_FADV_RANDOM: case POSIX_FADV_NOREUSE: new = malloc(sizeof(*fa), M_FADVISE, M_WAITOK); break; case POSIX_FADV_NORMAL: case POSIX_FADV_WILLNEED: case POSIX_FADV_DONTNEED: new = NULL; break; default: return (EINVAL); } /* XXX: CAP_POSIX_FADVISE? */ error = fget(td, fd, cap_rights_init(&rights), &fp); if (error != 0) goto out; if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) { error = ESPIPE; goto out; } if (fp->f_type != DTYPE_VNODE) { error = ENODEV; goto out; } vp = fp->f_vnode; if (vp->v_type != VREG) { error = ENODEV; goto out; } if (len == 0) end = OFF_MAX; else end = offset + len - 1; switch (advice) { case POSIX_FADV_SEQUENTIAL: case POSIX_FADV_RANDOM: case POSIX_FADV_NOREUSE: /* * Try to merge any existing non-standard region with * this new region if possible, otherwise create a new * non-standard region for this request. */ mtx_pool_lock(mtxpool_sleep, fp); fa = fp->f_advice; if (fa != NULL && fa->fa_advice == advice && ((fa->fa_start <= end && fa->fa_end >= offset) || (end != OFF_MAX && fa->fa_start == end + 1) || (fa->fa_end != OFF_MAX && fa->fa_end + 1 == offset))) { if (offset < fa->fa_start) fa->fa_start = offset; if (end > fa->fa_end) fa->fa_end = end; } else { new->fa_advice = advice; new->fa_start = offset; new->fa_end = end; fp->f_advice = new; new = fa; } mtx_pool_unlock(mtxpool_sleep, fp); break; case POSIX_FADV_NORMAL: /* * If a the "normal" region overlaps with an existing * non-standard region, trim or remove the * non-standard region. */ mtx_pool_lock(mtxpool_sleep, fp); fa = fp->f_advice; if (fa != NULL) { if (offset <= fa->fa_start && end >= fa->fa_end) { new = fa; fp->f_advice = NULL; } else if (offset <= fa->fa_start && end >= fa->fa_start) fa->fa_start = end + 1; else if (offset <= fa->fa_end && end >= fa->fa_end) fa->fa_end = offset - 1; else if (offset >= fa->fa_start && end <= fa->fa_end) { /* * If the "normal" region is a middle * portion of the existing * non-standard region, just remove * the whole thing rather than picking * one side or the other to * preserve. */ new = fa; fp->f_advice = NULL; } } mtx_pool_unlock(mtxpool_sleep, fp); break; case POSIX_FADV_WILLNEED: case POSIX_FADV_DONTNEED: error = VOP_ADVISE(vp, offset, end, advice); break; } out: if (fp != NULL) fdrop(fp, td); free(new, M_FADVISE); return (error); } int sys_posix_fadvise(struct thread *td, struct posix_fadvise_args *uap) { int error; error = kern_posix_fadvise(td, uap->fd, uap->offset, uap->len, uap->advice); return (kern_posix_error(td, error)); } Index: head/sys/kern/vfs_vnops.c =================================================================== --- head/sys/kern/vfs_vnops.c (revision 305831) +++ head/sys/kern/vfs_vnops.c (revision 305832) @@ -1,2477 +1,2477 @@ /*- * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Copyright (c) 2012 Konstantin Belousov * Copyright (c) 2013, 2014 The FreeBSD Foundation * * Portions of this software were developed by Konstantin Belousov * under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * 4. Neither the name of the University nor the names of its contributors + * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_hwpmc_hooks.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef HWPMC_HOOKS #include #endif static fo_rdwr_t vn_read; static fo_rdwr_t vn_write; static fo_rdwr_t vn_io_fault; static fo_truncate_t vn_truncate; static fo_ioctl_t vn_ioctl; static fo_poll_t vn_poll; static fo_kqfilter_t vn_kqfilter; static fo_stat_t vn_statfile; static fo_close_t vn_closefile; static fo_mmap_t vn_mmap; struct fileops vnops = { .fo_read = vn_io_fault, .fo_write = vn_io_fault, .fo_truncate = vn_truncate, .fo_ioctl = vn_ioctl, .fo_poll = vn_poll, .fo_kqfilter = vn_kqfilter, .fo_stat = vn_statfile, .fo_close = vn_closefile, .fo_chmod = vn_chmod, .fo_chown = vn_chown, .fo_sendfile = vn_sendfile, .fo_seek = vn_seek, .fo_fill_kinfo = vn_fill_kinfo, .fo_mmap = vn_mmap, .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE }; static const int io_hold_cnt = 16; static int vn_io_fault_enable = 1; SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW, &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance"); static int vn_io_fault_prefault = 0; SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_prefault, CTLFLAG_RW, &vn_io_fault_prefault, 0, "Enable vn_io_fault prefaulting"); static u_long vn_io_faults_cnt; SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD, &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers"); /* * Returns true if vn_io_fault mode of handling the i/o request should * be used. */ static bool do_vn_io_fault(struct vnode *vp, struct uio *uio) { struct mount *mp; return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG && (mp = vp->v_mount) != NULL && (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable); } /* * Structure used to pass arguments to vn_io_fault1(), to do either * file- or vnode-based I/O calls. */ struct vn_io_fault_args { enum { VN_IO_FAULT_FOP, VN_IO_FAULT_VOP } kind; struct ucred *cred; int flags; union { struct fop_args_tag { struct file *fp; fo_rdwr_t *doio; } fop_args; struct vop_args_tag { struct vnode *vp; } vop_args; } args; }; static int vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args, struct thread *td); int vn_open(ndp, flagp, cmode, fp) struct nameidata *ndp; int *flagp, cmode; struct file *fp; { struct thread *td = ndp->ni_cnd.cn_thread; return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp)); } /* * Common code for vnode open operations via a name lookup. * Lookup the vnode and invoke VOP_CREATE if needed. * Check permissions, and call the VOP_OPEN or VOP_CREATE routine. * * Note that this does NOT free nameidata for the successful case, * due to the NDINIT being done elsewhere. */ int vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags, struct ucred *cred, struct file *fp) { struct vnode *vp; struct mount *mp; struct thread *td = ndp->ni_cnd.cn_thread; struct vattr vat; struct vattr *vap = &vat; int fmode, error; restart: fmode = *flagp; if ((fmode & (O_CREAT | O_EXCL | O_DIRECTORY)) == (O_CREAT | O_EXCL | O_DIRECTORY)) return (EINVAL); else if ((fmode & (O_CREAT | O_DIRECTORY)) == O_CREAT) { ndp->ni_cnd.cn_nameiop = CREATE; /* * Set NOCACHE to avoid flushing the cache when * rolling in many files at once. */ ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | NOCACHE; if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0) ndp->ni_cnd.cn_flags |= FOLLOW; if (!(vn_open_flags & VN_OPEN_NOAUDIT)) ndp->ni_cnd.cn_flags |= AUDITVNODE1; if (vn_open_flags & VN_OPEN_NOCAPCHECK) ndp->ni_cnd.cn_flags |= NOCAPCHECK; bwillwrite(); if ((error = namei(ndp)) != 0) return (error); if (ndp->ni_vp == NULL) { VATTR_NULL(vap); vap->va_type = VREG; vap->va_mode = cmode; if (fmode & O_EXCL) vap->va_vaflags |= VA_EXCLUSIVE; if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(ndp, NDF_ONLY_PNBUF); vput(ndp->ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) return (error); goto restart; } if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0) ndp->ni_cnd.cn_flags |= MAKEENTRY; #ifdef MAC error = mac_vnode_check_create(cred, ndp->ni_dvp, &ndp->ni_cnd, vap); if (error == 0) #endif error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, vap); vput(ndp->ni_dvp); vn_finished_write(mp); if (error) { NDFREE(ndp, NDF_ONLY_PNBUF); return (error); } fmode &= ~O_TRUNC; vp = ndp->ni_vp; } else { if (ndp->ni_dvp == ndp->ni_vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); ndp->ni_dvp = NULL; vp = ndp->ni_vp; if (fmode & O_EXCL) { error = EEXIST; goto bad; } fmode &= ~O_CREAT; } } else { ndp->ni_cnd.cn_nameiop = LOOKUP; ndp->ni_cnd.cn_flags = ISOPEN | ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF; if (!(fmode & FWRITE)) ndp->ni_cnd.cn_flags |= LOCKSHARED; if (!(vn_open_flags & VN_OPEN_NOAUDIT)) ndp->ni_cnd.cn_flags |= AUDITVNODE1; if (vn_open_flags & VN_OPEN_NOCAPCHECK) ndp->ni_cnd.cn_flags |= NOCAPCHECK; if ((error = namei(ndp)) != 0) return (error); vp = ndp->ni_vp; } error = vn_open_vnode(vp, fmode, cred, td, fp); if (error) goto bad; *flagp = fmode; return (0); bad: NDFREE(ndp, NDF_ONLY_PNBUF); vput(vp); *flagp = fmode; ndp->ni_vp = NULL; return (error); } /* * Common code for vnode open operations once a vnode is located. * Check permissions, and call the VOP_OPEN routine. */ int vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred, struct thread *td, struct file *fp) { accmode_t accmode; struct flock lf; int error, lock_flags, type; if (vp->v_type == VLNK) return (EMLINK); if (vp->v_type == VSOCK) return (EOPNOTSUPP); if (vp->v_type != VDIR && fmode & O_DIRECTORY) return (ENOTDIR); accmode = 0; if (fmode & (FWRITE | O_TRUNC)) { if (vp->v_type == VDIR) return (EISDIR); accmode |= VWRITE; } if (fmode & FREAD) accmode |= VREAD; if (fmode & FEXEC) accmode |= VEXEC; if ((fmode & O_APPEND) && (fmode & FWRITE)) accmode |= VAPPEND; #ifdef MAC if (fmode & O_CREAT) accmode |= VCREAT; if (fmode & O_VERIFY) accmode |= VVERIFY; error = mac_vnode_check_open(cred, vp, accmode); if (error) return (error); accmode &= ~(VCREAT | VVERIFY); #endif if ((fmode & O_CREAT) == 0) { if (accmode & VWRITE) { error = vn_writechk(vp); if (error) return (error); } if (accmode) { error = VOP_ACCESS(vp, accmode, cred, td); if (error) return (error); } } if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) vn_lock(vp, LK_UPGRADE | LK_RETRY); if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0) return (error); if (fmode & (O_EXLOCK | O_SHLOCK)) { KASSERT(fp != NULL, ("open with flock requires fp")); lock_flags = VOP_ISLOCKED(vp); VOP_UNLOCK(vp, 0); lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; if (fmode & O_EXLOCK) lf.l_type = F_WRLCK; else lf.l_type = F_RDLCK; type = F_FLOCK; if ((fmode & FNONBLOCK) == 0) type |= F_WAIT; error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type); if (error == 0) fp->f_flag |= FHASLOCK; vn_lock(vp, lock_flags | LK_RETRY); if (error == 0 && vp->v_iflag & VI_DOOMED) error = ENOENT; /* * Another thread might have used this vnode as an * executable while the vnode lock was dropped. * Ensure the vnode is still able to be opened for * writing after the lock has been obtained. */ if (error == 0 && accmode & VWRITE) error = vn_writechk(vp); if (error != 0) { fp->f_flag |= FOPENFAILED; fp->f_vnode = vp; if (fp->f_ops == &badfileops) { fp->f_type = DTYPE_VNODE; fp->f_ops = &vnops; } vref(vp); } } if (error == 0 && fmode & FWRITE) { VOP_ADD_WRITECOUNT(vp, 1); CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d", __func__, vp, vp->v_writecount); } ASSERT_VOP_LOCKED(vp, "vn_open_vnode"); return (error); } /* * Check for write permissions on the specified vnode. * Prototype text segments cannot be written. */ int vn_writechk(vp) register struct vnode *vp; { ASSERT_VOP_LOCKED(vp, "vn_writechk"); /* * If there's shared text associated with * the vnode, try to free it up once. If * we fail, we can't allow writing. */ if (VOP_IS_TEXT(vp)) return (ETXTBSY); return (0); } /* * Vnode close call */ int vn_close(vp, flags, file_cred, td) register struct vnode *vp; int flags; struct ucred *file_cred; struct thread *td; { struct mount *mp; int error, lock_flags; if (vp->v_type != VFIFO && (flags & FWRITE) == 0 && MNT_EXTENDED_SHARED(vp->v_mount)) lock_flags = LK_SHARED; else lock_flags = LK_EXCLUSIVE; vn_start_write(vp, &mp, V_WAIT); vn_lock(vp, lock_flags | LK_RETRY); AUDIT_ARG_VNODE1(vp); if ((flags & (FWRITE | FOPENFAILED)) == FWRITE) { VNASSERT(vp->v_writecount > 0, vp, ("vn_close: negative writecount")); VOP_ADD_WRITECOUNT(vp, -1); CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d", __func__, vp, vp->v_writecount); } error = VOP_CLOSE(vp, flags, file_cred, td); vput(vp); vn_finished_write(mp); return (error); } /* * Heuristic to detect sequential operation. */ static int sequential_heuristic(struct uio *uio, struct file *fp) { ASSERT_VOP_LOCKED(fp->f_vnode, __func__); if (fp->f_flag & FRDAHEAD) return (fp->f_seqcount << IO_SEQSHIFT); /* * Offset 0 is handled specially. open() sets f_seqcount to 1 so * that the first I/O is normally considered to be slightly * sequential. Seeking to offset 0 doesn't change sequentiality * unless previous seeks have reduced f_seqcount to 0, in which * case offset 0 is not special. */ if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || uio->uio_offset == fp->f_nextoff) { /* * f_seqcount is in units of fixed-size blocks so that it * depends mainly on the amount of sequential I/O and not * much on the number of sequential I/O's. The fixed size * of 16384 is hard-coded here since it is (not quite) just * a magic size that works well here. This size is more * closely related to the best I/O size for real disks than * to any block size used by software. */ fp->f_seqcount += howmany(uio->uio_resid, 16384); if (fp->f_seqcount > IO_SEQMAX) fp->f_seqcount = IO_SEQMAX; return (fp->f_seqcount << IO_SEQSHIFT); } /* Not sequential. Quickly draw-down sequentiality. */ if (fp->f_seqcount > 1) fp->f_seqcount = 1; else fp->f_seqcount = 0; return (0); } /* * Package up an I/O request on a vnode into a uio and do it. */ int vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset, enum uio_seg segflg, int ioflg, struct ucred *active_cred, struct ucred *file_cred, ssize_t *aresid, struct thread *td) { struct uio auio; struct iovec aiov; struct mount *mp; struct ucred *cred; void *rl_cookie; struct vn_io_fault_args args; int error, lock_flags; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; aiov.iov_base = base; aiov.iov_len = len; auio.uio_resid = len; auio.uio_offset = offset; auio.uio_segflg = segflg; auio.uio_rw = rw; auio.uio_td = td; error = 0; if ((ioflg & IO_NODELOCKED) == 0) { if ((ioflg & IO_RANGELOCKED) == 0) { if (rw == UIO_READ) { rl_cookie = vn_rangelock_rlock(vp, offset, offset + len); } else { rl_cookie = vn_rangelock_wlock(vp, offset, offset + len); } } else rl_cookie = NULL; mp = NULL; if (rw == UIO_WRITE) { if (vp->v_type != VCHR && (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) goto out; if (MNT_SHARED_WRITES(mp) || ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) lock_flags = LK_SHARED; else lock_flags = LK_EXCLUSIVE; } else lock_flags = LK_SHARED; vn_lock(vp, lock_flags | LK_RETRY); } else rl_cookie = NULL; ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); #ifdef MAC if ((ioflg & IO_NOMACCHECK) == 0) { if (rw == UIO_READ) error = mac_vnode_check_read(active_cred, file_cred, vp); else error = mac_vnode_check_write(active_cred, file_cred, vp); } #endif if (error == 0) { if (file_cred != NULL) cred = file_cred; else cred = active_cred; if (do_vn_io_fault(vp, &auio)) { args.kind = VN_IO_FAULT_VOP; args.cred = cred; args.flags = ioflg; args.args.vop_args.vp = vp; error = vn_io_fault1(vp, &auio, &args, td); } else if (rw == UIO_READ) { error = VOP_READ(vp, &auio, ioflg, cred); } else /* if (rw == UIO_WRITE) */ { error = VOP_WRITE(vp, &auio, ioflg, cred); } } if (aresid) *aresid = auio.uio_resid; else if (auio.uio_resid && error == 0) error = EIO; if ((ioflg & IO_NODELOCKED) == 0) { VOP_UNLOCK(vp, 0); if (mp != NULL) vn_finished_write(mp); } out: if (rl_cookie != NULL) vn_rangelock_unlock(vp, rl_cookie); return (error); } /* * Package up an I/O request on a vnode into a uio and do it. The I/O * request is split up into smaller chunks and we try to avoid saturating * the buffer cache while potentially holding a vnode locked, so we * check bwillwrite() before calling vn_rdwr(). We also call kern_yield() * to give other processes a chance to lock the vnode (either other processes * core'ing the same binary, or unrelated processes scanning the directory). */ int vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred, file_cred, aresid, td) enum uio_rw rw; struct vnode *vp; void *base; size_t len; off_t offset; enum uio_seg segflg; int ioflg; struct ucred *active_cred; struct ucred *file_cred; size_t *aresid; struct thread *td; { int error = 0; ssize_t iaresid; do { int chunk; /* * Force `offset' to a multiple of MAXBSIZE except possibly * for the first chunk, so that filesystems only need to * write full blocks except possibly for the first and last * chunks. */ chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; if (chunk > len) chunk = len; if (rw != UIO_READ && vp->v_type == VREG) bwillwrite(); iaresid = 0; error = vn_rdwr(rw, vp, base, chunk, offset, segflg, ioflg, active_cred, file_cred, &iaresid, td); len -= chunk; /* aresid calc already includes length */ if (error) break; offset += chunk; base = (char *)base + chunk; kern_yield(PRI_USER); } while (len); if (aresid) *aresid = len + iaresid; return (error); } off_t foffset_lock(struct file *fp, int flags) { struct mtx *mtxp; off_t res; KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); #if OFF_MAX <= LONG_MAX /* * Caller only wants the current f_offset value. Assume that * the long and shorter integer types reads are atomic. */ if ((flags & FOF_NOLOCK) != 0) return (fp->f_offset); #endif /* * According to McKusick the vn lock was protecting f_offset here. * It is now protected by the FOFFSET_LOCKED flag. */ mtxp = mtx_pool_find(mtxpool_sleep, fp); mtx_lock(mtxp); if ((flags & FOF_NOLOCK) == 0) { while (fp->f_vnread_flags & FOFFSET_LOCKED) { fp->f_vnread_flags |= FOFFSET_LOCK_WAITING; msleep(&fp->f_vnread_flags, mtxp, PUSER -1, "vofflock", 0); } fp->f_vnread_flags |= FOFFSET_LOCKED; } res = fp->f_offset; mtx_unlock(mtxp); return (res); } void foffset_unlock(struct file *fp, off_t val, int flags) { struct mtx *mtxp; KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); #if OFF_MAX <= LONG_MAX if ((flags & FOF_NOLOCK) != 0) { if ((flags & FOF_NOUPDATE) == 0) fp->f_offset = val; if ((flags & FOF_NEXTOFF) != 0) fp->f_nextoff = val; return; } #endif mtxp = mtx_pool_find(mtxpool_sleep, fp); mtx_lock(mtxp); if ((flags & FOF_NOUPDATE) == 0) fp->f_offset = val; if ((flags & FOF_NEXTOFF) != 0) fp->f_nextoff = val; if ((flags & FOF_NOLOCK) == 0) { KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0, ("Lost FOFFSET_LOCKED")); if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING) wakeup(&fp->f_vnread_flags); fp->f_vnread_flags = 0; } mtx_unlock(mtxp); } void foffset_lock_uio(struct file *fp, struct uio *uio, int flags) { if ((flags & FOF_OFFSET) == 0) uio->uio_offset = foffset_lock(fp, flags); } void foffset_unlock_uio(struct file *fp, struct uio *uio, int flags) { if ((flags & FOF_OFFSET) == 0) foffset_unlock(fp, uio->uio_offset, flags); } static int get_advice(struct file *fp, struct uio *uio) { struct mtx *mtxp; int ret; ret = POSIX_FADV_NORMAL; if (fp->f_advice == NULL || fp->f_vnode->v_type != VREG) return (ret); mtxp = mtx_pool_find(mtxpool_sleep, fp); mtx_lock(mtxp); if (fp->f_advice != NULL && uio->uio_offset >= fp->f_advice->fa_start && uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end) ret = fp->f_advice->fa_advice; mtx_unlock(mtxp); return (ret); } /* * File table vnode read routine. */ static int vn_read(fp, uio, active_cred, flags, td) struct file *fp; struct uio *uio; struct ucred *active_cred; int flags; struct thread *td; { struct vnode *vp; off_t orig_offset; int error, ioflag; int advice; KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); vp = fp->f_vnode; ioflag = 0; if (fp->f_flag & FNONBLOCK) ioflag |= IO_NDELAY; if (fp->f_flag & O_DIRECT) ioflag |= IO_DIRECT; advice = get_advice(fp, uio); vn_lock(vp, LK_SHARED | LK_RETRY); switch (advice) { case POSIX_FADV_NORMAL: case POSIX_FADV_SEQUENTIAL: case POSIX_FADV_NOREUSE: ioflag |= sequential_heuristic(uio, fp); break; case POSIX_FADV_RANDOM: /* Disable read-ahead for random I/O. */ break; } orig_offset = uio->uio_offset; #ifdef MAC error = mac_vnode_check_read(active_cred, fp->f_cred, vp); if (error == 0) #endif error = VOP_READ(vp, uio, ioflag, fp->f_cred); fp->f_nextoff = uio->uio_offset; VOP_UNLOCK(vp, 0); if (error == 0 && advice == POSIX_FADV_NOREUSE && orig_offset != uio->uio_offset) /* * Use POSIX_FADV_DONTNEED to flush pages and buffers * for the backing file after a POSIX_FADV_NOREUSE * read(2). */ error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1, POSIX_FADV_DONTNEED); return (error); } /* * File table vnode write routine. */ static int vn_write(fp, uio, active_cred, flags, td) struct file *fp; struct uio *uio; struct ucred *active_cred; int flags; struct thread *td; { struct vnode *vp; struct mount *mp; off_t orig_offset; int error, ioflag, lock_flags; int advice; KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); vp = fp->f_vnode; if (vp->v_type == VREG) bwillwrite(); ioflag = IO_UNIT; if (vp->v_type == VREG && (fp->f_flag & O_APPEND)) ioflag |= IO_APPEND; if (fp->f_flag & FNONBLOCK) ioflag |= IO_NDELAY; if (fp->f_flag & O_DIRECT) ioflag |= IO_DIRECT; if ((fp->f_flag & O_FSYNC) || (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))) ioflag |= IO_SYNC; mp = NULL; if (vp->v_type != VCHR && (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) goto unlock; advice = get_advice(fp, uio); if (MNT_SHARED_WRITES(mp) || (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) { lock_flags = LK_SHARED; } else { lock_flags = LK_EXCLUSIVE; } vn_lock(vp, lock_flags | LK_RETRY); switch (advice) { case POSIX_FADV_NORMAL: case POSIX_FADV_SEQUENTIAL: case POSIX_FADV_NOREUSE: ioflag |= sequential_heuristic(uio, fp); break; case POSIX_FADV_RANDOM: /* XXX: Is this correct? */ break; } orig_offset = uio->uio_offset; #ifdef MAC error = mac_vnode_check_write(active_cred, fp->f_cred, vp); if (error == 0) #endif error = VOP_WRITE(vp, uio, ioflag, fp->f_cred); fp->f_nextoff = uio->uio_offset; VOP_UNLOCK(vp, 0); if (vp->v_type != VCHR) vn_finished_write(mp); if (error == 0 && advice == POSIX_FADV_NOREUSE && orig_offset != uio->uio_offset) /* * Use POSIX_FADV_DONTNEED to flush pages and buffers * for the backing file after a POSIX_FADV_NOREUSE * write(2). */ error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1, POSIX_FADV_DONTNEED); unlock: return (error); } /* * The vn_io_fault() is a wrapper around vn_read() and vn_write() to * prevent the following deadlock: * * Assume that the thread A reads from the vnode vp1 into userspace * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is * currently not resident, then system ends up with the call chain * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] -> * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2) * which establishes lock order vp1->vn_lock, then vp2->vn_lock. * If, at the same time, thread B reads from vnode vp2 into buffer buf2 * backed by the pages of vnode vp1, and some page in buf2 is not * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock. * * To prevent the lock order reversal and deadlock, vn_io_fault() does * not allow page faults to happen during VOP_READ() or VOP_WRITE(). * Instead, it first tries to do the whole range i/o with pagefaults * disabled. If all pages in the i/o buffer are resident and mapped, * VOP will succeed (ignoring the genuine filesystem errors). * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do * i/o in chunks, with all pages in the chunk prefaulted and held * using vm_fault_quick_hold_pages(). * * Filesystems using this deadlock avoidance scheme should use the * array of the held pages from uio, saved in the curthread->td_ma, * instead of doing uiomove(). A helper function * vn_io_fault_uiomove() converts uiomove request into * uiomove_fromphys() over td_ma array. * * Since vnode locks do not cover the whole i/o anymore, rangelocks * make the current i/o request atomic with respect to other i/os and * truncations. */ /* * Decode vn_io_fault_args and perform the corresponding i/o. */ static int vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio, struct thread *td) { switch (args->kind) { case VN_IO_FAULT_FOP: return ((args->args.fop_args.doio)(args->args.fop_args.fp, uio, args->cred, args->flags, td)); case VN_IO_FAULT_VOP: if (uio->uio_rw == UIO_READ) { return (VOP_READ(args->args.vop_args.vp, uio, args->flags, args->cred)); } else if (uio->uio_rw == UIO_WRITE) { return (VOP_WRITE(args->args.vop_args.vp, uio, args->flags, args->cred)); } break; } panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind, uio->uio_rw); } static int vn_io_fault_touch(char *base, const struct uio *uio) { int r; r = fubyte(base); if (r == -1 || (uio->uio_rw == UIO_READ && subyte(base, r) == -1)) return (EFAULT); return (0); } static int vn_io_fault_prefault_user(const struct uio *uio) { char *base; const struct iovec *iov; size_t len; ssize_t resid; int error, i; KASSERT(uio->uio_segflg == UIO_USERSPACE, ("vn_io_fault_prefault userspace")); error = i = 0; iov = uio->uio_iov; resid = uio->uio_resid; base = iov->iov_base; len = iov->iov_len; while (resid > 0) { error = vn_io_fault_touch(base, uio); if (error != 0) break; if (len < PAGE_SIZE) { if (len != 0) { error = vn_io_fault_touch(base + len - 1, uio); if (error != 0) break; resid -= len; } if (++i >= uio->uio_iovcnt) break; iov = uio->uio_iov + i; base = iov->iov_base; len = iov->iov_len; } else { len -= PAGE_SIZE; base += PAGE_SIZE; resid -= PAGE_SIZE; } } return (error); } /* * Common code for vn_io_fault(), agnostic to the kind of i/o request. * Uses vn_io_fault_doio() to make the call to an actual i/o function. * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request * into args and call vn_io_fault1() to handle faults during the user * mode buffer accesses. */ static int vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args, struct thread *td) { vm_page_t ma[io_hold_cnt + 2]; struct uio *uio_clone, short_uio; struct iovec short_iovec[1]; vm_page_t *prev_td_ma; vm_prot_t prot; vm_offset_t addr, end; size_t len, resid; ssize_t adv; int error, cnt, save, saveheld, prev_td_ma_cnt; if (vn_io_fault_prefault) { error = vn_io_fault_prefault_user(uio); if (error != 0) return (error); /* Or ignore ? */ } prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ; /* * The UFS follows IO_UNIT directive and replays back both * uio_offset and uio_resid if an error is encountered during the * operation. But, since the iovec may be already advanced, * uio is still in an inconsistent state. * * Cache a copy of the original uio, which is advanced to the redo * point using UIO_NOCOPY below. */ uio_clone = cloneuio(uio); resid = uio->uio_resid; short_uio.uio_segflg = UIO_USERSPACE; short_uio.uio_rw = uio->uio_rw; short_uio.uio_td = uio->uio_td; save = vm_fault_disable_pagefaults(); error = vn_io_fault_doio(args, uio, td); if (error != EFAULT) goto out; atomic_add_long(&vn_io_faults_cnt, 1); uio_clone->uio_segflg = UIO_NOCOPY; uiomove(NULL, resid - uio->uio_resid, uio_clone); uio_clone->uio_segflg = uio->uio_segflg; saveheld = curthread_pflags_set(TDP_UIOHELD); prev_td_ma = td->td_ma; prev_td_ma_cnt = td->td_ma_cnt; while (uio_clone->uio_resid != 0) { len = uio_clone->uio_iov->iov_len; if (len == 0) { KASSERT(uio_clone->uio_iovcnt >= 1, ("iovcnt underflow")); uio_clone->uio_iov++; uio_clone->uio_iovcnt--; continue; } if (len > io_hold_cnt * PAGE_SIZE) len = io_hold_cnt * PAGE_SIZE; addr = (uintptr_t)uio_clone->uio_iov->iov_base; end = round_page(addr + len); if (end < addr) { error = EFAULT; break; } cnt = atop(end - trunc_page(addr)); /* * A perfectly misaligned address and length could cause * both the start and the end of the chunk to use partial * page. +2 accounts for such a situation. */ cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map, addr, len, prot, ma, io_hold_cnt + 2); if (cnt == -1) { error = EFAULT; break; } short_uio.uio_iov = &short_iovec[0]; short_iovec[0].iov_base = (void *)addr; short_uio.uio_iovcnt = 1; short_uio.uio_resid = short_iovec[0].iov_len = len; short_uio.uio_offset = uio_clone->uio_offset; td->td_ma = ma; td->td_ma_cnt = cnt; error = vn_io_fault_doio(args, &short_uio, td); vm_page_unhold_pages(ma, cnt); adv = len - short_uio.uio_resid; uio_clone->uio_iov->iov_base = (char *)uio_clone->uio_iov->iov_base + adv; uio_clone->uio_iov->iov_len -= adv; uio_clone->uio_resid -= adv; uio_clone->uio_offset += adv; uio->uio_resid -= adv; uio->uio_offset += adv; if (error != 0 || adv == 0) break; } td->td_ma = prev_td_ma; td->td_ma_cnt = prev_td_ma_cnt; curthread_pflags_restore(saveheld); out: vm_fault_enable_pagefaults(save); free(uio_clone, M_IOV); return (error); } static int vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { fo_rdwr_t *doio; struct vnode *vp; void *rl_cookie; struct vn_io_fault_args args; int error; doio = uio->uio_rw == UIO_READ ? vn_read : vn_write; vp = fp->f_vnode; foffset_lock_uio(fp, uio, flags); if (do_vn_io_fault(vp, uio)) { args.kind = VN_IO_FAULT_FOP; args.args.fop_args.fp = fp; args.args.fop_args.doio = doio; args.cred = active_cred; args.flags = flags | FOF_OFFSET; if (uio->uio_rw == UIO_READ) { rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset, uio->uio_offset + uio->uio_resid); } else if ((fp->f_flag & O_APPEND) != 0 || (flags & FOF_OFFSET) == 0) { /* For appenders, punt and lock the whole range. */ rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); } else { rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset, uio->uio_offset + uio->uio_resid); } error = vn_io_fault1(vp, uio, &args, td); vn_rangelock_unlock(vp, rl_cookie); } else { error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td); } foffset_unlock_uio(fp, uio, flags); return (error); } /* * Helper function to perform the requested uiomove operation using * the held pages for io->uio_iov[0].iov_base buffer instead of * copyin/copyout. Access to the pages with uiomove_fromphys() * instead of iov_base prevents page faults that could occur due to * pmap_collect() invalidating the mapping created by * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or * object cleanup revoking the write access from page mappings. * * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove() * instead of plain uiomove(). */ int vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio) { struct uio transp_uio; struct iovec transp_iov[1]; struct thread *td; size_t adv; int error, pgadv; td = curthread; if ((td->td_pflags & TDP_UIOHELD) == 0 || uio->uio_segflg != UIO_USERSPACE) return (uiomove(data, xfersize, uio)); KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); transp_iov[0].iov_base = data; transp_uio.uio_iov = &transp_iov[0]; transp_uio.uio_iovcnt = 1; if (xfersize > uio->uio_resid) xfersize = uio->uio_resid; transp_uio.uio_resid = transp_iov[0].iov_len = xfersize; transp_uio.uio_offset = 0; transp_uio.uio_segflg = UIO_SYSSPACE; /* * Since transp_iov points to data, and td_ma page array * corresponds to original uio->uio_iov, we need to invert the * direction of the i/o operation as passed to * uiomove_fromphys(). */ switch (uio->uio_rw) { case UIO_WRITE: transp_uio.uio_rw = UIO_READ; break; case UIO_READ: transp_uio.uio_rw = UIO_WRITE; break; } transp_uio.uio_td = uio->uio_td; error = uiomove_fromphys(td->td_ma, ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK, xfersize, &transp_uio); adv = xfersize - transp_uio.uio_resid; pgadv = (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) - (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT); td->td_ma += pgadv; KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, pgadv)); td->td_ma_cnt -= pgadv; uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv; uio->uio_iov->iov_len -= adv; uio->uio_resid -= adv; uio->uio_offset += adv; return (error); } int vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize, struct uio *uio) { struct thread *td; vm_offset_t iov_base; int cnt, pgadv; td = curthread; if ((td->td_pflags & TDP_UIOHELD) == 0 || uio->uio_segflg != UIO_USERSPACE) return (uiomove_fromphys(ma, offset, xfersize, uio)); KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize; iov_base = (vm_offset_t)uio->uio_iov->iov_base; switch (uio->uio_rw) { case UIO_WRITE: pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma, offset, cnt); break; case UIO_READ: pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK, cnt); break; } pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT); td->td_ma += pgadv; KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, pgadv)); td->td_ma_cnt -= pgadv; uio->uio_iov->iov_base = (char *)(iov_base + cnt); uio->uio_iov->iov_len -= cnt; uio->uio_resid -= cnt; uio->uio_offset += cnt; return (0); } /* * File table truncate routine. */ static int vn_truncate(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td) { struct vattr vattr; struct mount *mp; struct vnode *vp; void *rl_cookie; int error; vp = fp->f_vnode; /* * Lock the whole range for truncation. Otherwise split i/o * might happen partly before and partly after the truncation. */ rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); error = vn_start_write(vp, &mp, V_WAIT | PCATCH); if (error) goto out1; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); AUDIT_ARG_VNODE1(vp); if (vp->v_type == VDIR) { error = EISDIR; goto out; } #ifdef MAC error = mac_vnode_check_write(active_cred, fp->f_cred, vp); if (error) goto out; #endif error = vn_writechk(vp); if (error == 0) { VATTR_NULL(&vattr); vattr.va_size = length; if ((fp->f_flag & O_FSYNC) != 0) vattr.va_vaflags |= VA_SYNC; error = VOP_SETATTR(vp, &vattr, fp->f_cred); } out: VOP_UNLOCK(vp, 0); vn_finished_write(mp); out1: vn_rangelock_unlock(vp, rl_cookie); return (error); } /* * File table vnode stat routine. */ static int vn_statfile(fp, sb, active_cred, td) struct file *fp; struct stat *sb; struct ucred *active_cred; struct thread *td; { struct vnode *vp = fp->f_vnode; int error; vn_lock(vp, LK_SHARED | LK_RETRY); error = vn_stat(vp, sb, active_cred, fp->f_cred, td); VOP_UNLOCK(vp, 0); return (error); } /* * Stat a vnode; implementation for the stat syscall */ int vn_stat(vp, sb, active_cred, file_cred, td) struct vnode *vp; register struct stat *sb; struct ucred *active_cred; struct ucred *file_cred; struct thread *td; { struct vattr vattr; register struct vattr *vap; int error; u_short mode; AUDIT_ARG_VNODE1(vp); #ifdef MAC error = mac_vnode_check_stat(active_cred, file_cred, vp); if (error) return (error); #endif vap = &vattr; /* * Initialize defaults for new and unusual fields, so that file * systems which don't support these fields don't need to know * about them. */ vap->va_birthtime.tv_sec = -1; vap->va_birthtime.tv_nsec = 0; vap->va_fsid = VNOVAL; vap->va_rdev = NODEV; error = VOP_GETATTR(vp, vap, active_cred); if (error) return (error); /* * Zero the spare stat fields */ bzero(sb, sizeof *sb); /* * Copy from vattr table */ if (vap->va_fsid != VNOVAL) sb->st_dev = vap->va_fsid; else sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; sb->st_ino = vap->va_fileid; mode = vap->va_mode; switch (vap->va_type) { case VREG: mode |= S_IFREG; break; case VDIR: mode |= S_IFDIR; break; case VBLK: mode |= S_IFBLK; break; case VCHR: mode |= S_IFCHR; break; case VLNK: mode |= S_IFLNK; break; case VSOCK: mode |= S_IFSOCK; break; case VFIFO: mode |= S_IFIFO; break; default: return (EBADF); } sb->st_mode = mode; sb->st_nlink = vap->va_nlink; sb->st_uid = vap->va_uid; sb->st_gid = vap->va_gid; sb->st_rdev = vap->va_rdev; if (vap->va_size > OFF_MAX) return (EOVERFLOW); sb->st_size = vap->va_size; sb->st_atim = vap->va_atime; sb->st_mtim = vap->va_mtime; sb->st_ctim = vap->va_ctime; sb->st_birthtim = vap->va_birthtime; /* * According to www.opengroup.org, the meaning of st_blksize is * "a filesystem-specific preferred I/O block size for this * object. In some filesystem types, this may vary from file * to file" * Use miminum/default of PAGE_SIZE (e.g. for VCHR). */ sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize); sb->st_flags = vap->va_flags; if (priv_check(td, PRIV_VFS_GENERATION)) sb->st_gen = 0; else sb->st_gen = vap->va_gen; sb->st_blocks = vap->va_bytes / S_BLKSIZE; return (0); } /* * File table vnode ioctl routine. */ static int vn_ioctl(fp, com, data, active_cred, td) struct file *fp; u_long com; void *data; struct ucred *active_cred; struct thread *td; { struct vattr vattr; struct vnode *vp; int error; vp = fp->f_vnode; switch (vp->v_type) { case VDIR: case VREG: switch (com) { case FIONREAD: vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_GETATTR(vp, &vattr, active_cred); VOP_UNLOCK(vp, 0); if (error == 0) *(int *)data = vattr.va_size - fp->f_offset; return (error); case FIONBIO: case FIOASYNC: return (0); default: return (VOP_IOCTL(vp, com, data, fp->f_flag, active_cred, td)); } break; case VCHR: return (VOP_IOCTL(vp, com, data, fp->f_flag, active_cred, td)); default: return (ENOTTY); } } /* * File table vnode poll routine. */ static int vn_poll(fp, events, active_cred, td) struct file *fp; int events; struct ucred *active_cred; struct thread *td; { struct vnode *vp; int error; vp = fp->f_vnode; #ifdef MAC vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); AUDIT_ARG_VNODE1(vp); error = mac_vnode_check_poll(active_cred, fp->f_cred, vp); VOP_UNLOCK(vp, 0); if (!error) #endif error = VOP_POLL(vp, events, fp->f_cred, td); return (error); } /* * Acquire the requested lock and then check for validity. LK_RETRY * permits vn_lock to return doomed vnodes. */ int _vn_lock(struct vnode *vp, int flags, char *file, int line) { int error; VNASSERT((flags & LK_TYPE_MASK) != 0, vp, ("vn_lock called with no locktype.")); do { #ifdef DEBUG_VFS_LOCKS KASSERT(vp->v_holdcnt != 0, ("vn_lock %p: zero hold count", vp)); #endif error = VOP_LOCK1(vp, flags, file, line); flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */ KASSERT((flags & LK_RETRY) == 0 || error == 0, ("LK_RETRY set with incompatible flags (0x%x) or an error occurred (%d)", flags, error)); /* * Callers specify LK_RETRY if they wish to get dead vnodes. * If RETRY is not set, we return ENOENT instead. */ if (error == 0 && vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0) { VOP_UNLOCK(vp, 0); error = ENOENT; break; } } while (flags & LK_RETRY && error != 0); return (error); } /* * File table vnode close routine. */ static int vn_closefile(fp, td) struct file *fp; struct thread *td; { struct vnode *vp; struct flock lf; int error; vp = fp->f_vnode; fp->f_ops = &badfileops; if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) vref(vp); error = vn_close(vp, fp->f_flag, fp->f_cred, td); if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) { lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; lf.l_type = F_UNLCK; (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK); vrele(vp); } return (error); } static bool vn_suspendable(struct mount *mp) { return (mp->mnt_op->vfs_susp_clean != NULL); } /* * Preparing to start a filesystem write operation. If the operation is * permitted, then we bump the count of operations in progress and * proceed. If a suspend request is in progress, we wait until the * suspension is over, and then proceed. */ static int vn_start_write_locked(struct mount *mp, int flags) { int error, mflags; mtx_assert(MNT_MTX(mp), MA_OWNED); error = 0; /* * Check on status of suspension. */ if ((curthread->td_pflags & TDP_IGNSUSP) == 0 || mp->mnt_susp_owner != curthread) { mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0) | (PUSER - 1); while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { if (flags & V_NOWAIT) { error = EWOULDBLOCK; goto unlock; } error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags, "suspfs", 0); if (error) goto unlock; } } if (flags & V_XSLEEP) goto unlock; mp->mnt_writeopcount++; unlock: if (error != 0 || (flags & V_XSLEEP) != 0) MNT_REL(mp); MNT_IUNLOCK(mp); return (error); } int vn_start_write(struct vnode *vp, struct mount **mpp, int flags) { struct mount *mp; int error; KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL), ("V_MNTREF requires mp")); error = 0; /* * If a vnode is provided, get and return the mount point that * to which it will write. */ if (vp != NULL) { if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { *mpp = NULL; if (error != EOPNOTSUPP) return (error); return (0); } } if ((mp = *mpp) == NULL) return (0); if (!vn_suspendable(mp)) { if (vp != NULL || (flags & V_MNTREF) != 0) vfs_rel(mp); return (0); } /* * VOP_GETWRITEMOUNT() returns with the mp refcount held through * a vfs_ref(). * As long as a vnode is not provided we need to acquire a * refcount for the provided mountpoint too, in order to * emulate a vfs_ref(). */ MNT_ILOCK(mp); if (vp == NULL && (flags & V_MNTREF) == 0) MNT_REF(mp); return (vn_start_write_locked(mp, flags)); } /* * Secondary suspension. Used by operations such as vop_inactive * routines that are needed by the higher level functions. These * are allowed to proceed until all the higher level functions have * completed (indicated by mnt_writeopcount dropping to zero). At that * time, these operations are halted until the suspension is over. */ int vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags) { struct mount *mp; int error; KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL), ("V_MNTREF requires mp")); retry: if (vp != NULL) { if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { *mpp = NULL; if (error != EOPNOTSUPP) return (error); return (0); } } /* * If we are not suspended or have not yet reached suspended * mode, then let the operation proceed. */ if ((mp = *mpp) == NULL) return (0); if (!vn_suspendable(mp)) { if (vp != NULL || (flags & V_MNTREF) != 0) vfs_rel(mp); return (0); } /* * VOP_GETWRITEMOUNT() returns with the mp refcount held through * a vfs_ref(). * As long as a vnode is not provided we need to acquire a * refcount for the provided mountpoint too, in order to * emulate a vfs_ref(). */ MNT_ILOCK(mp); if (vp == NULL && (flags & V_MNTREF) == 0) MNT_REF(mp); if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) { mp->mnt_secondary_writes++; mp->mnt_secondary_accwrites++; MNT_IUNLOCK(mp); return (0); } if (flags & V_NOWAIT) { MNT_REL(mp); MNT_IUNLOCK(mp); return (EWOULDBLOCK); } /* * Wait for the suspension to finish. */ error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP | ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0), "suspfs", 0); vfs_rel(mp); if (error == 0) goto retry; return (error); } /* * Filesystem write operation has completed. If we are suspending and this * operation is the last one, notify the suspender that the suspension is * now in effect. */ void vn_finished_write(mp) struct mount *mp; { if (mp == NULL || !vn_suspendable(mp)) return; MNT_ILOCK(mp); MNT_REL(mp); mp->mnt_writeopcount--; if (mp->mnt_writeopcount < 0) panic("vn_finished_write: neg cnt"); if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && mp->mnt_writeopcount <= 0) wakeup(&mp->mnt_writeopcount); MNT_IUNLOCK(mp); } /* * Filesystem secondary write operation has completed. If we are * suspending and this operation is the last one, notify the suspender * that the suspension is now in effect. */ void vn_finished_secondary_write(mp) struct mount *mp; { if (mp == NULL || !vn_suspendable(mp)) return; MNT_ILOCK(mp); MNT_REL(mp); mp->mnt_secondary_writes--; if (mp->mnt_secondary_writes < 0) panic("vn_finished_secondary_write: neg cnt"); if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && mp->mnt_secondary_writes <= 0) wakeup(&mp->mnt_secondary_writes); MNT_IUNLOCK(mp); } /* * Request a filesystem to suspend write operations. */ int vfs_write_suspend(struct mount *mp, int flags) { int error; MPASS(vn_suspendable(mp)); MNT_ILOCK(mp); if (mp->mnt_susp_owner == curthread) { MNT_IUNLOCK(mp); return (EALREADY); } while (mp->mnt_kern_flag & MNTK_SUSPEND) msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0); /* * Unmount holds a write reference on the mount point. If we * own busy reference and drain for writers, we deadlock with * the reference draining in the unmount path. Callers of * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if * vfs_busy() reference is owned and caller is not in the * unmount context. */ if ((flags & VS_SKIP_UNMOUNT) != 0 && (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) { MNT_IUNLOCK(mp); return (EBUSY); } mp->mnt_kern_flag |= MNTK_SUSPEND; mp->mnt_susp_owner = curthread; if (mp->mnt_writeopcount > 0) (void) msleep(&mp->mnt_writeopcount, MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0); else MNT_IUNLOCK(mp); if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0) vfs_write_resume(mp, 0); return (error); } /* * Request a filesystem to resume write operations. */ void vfs_write_resume(struct mount *mp, int flags) { MPASS(vn_suspendable(mp)); MNT_ILOCK(mp); if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner")); mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 | MNTK_SUSPENDED); mp->mnt_susp_owner = NULL; wakeup(&mp->mnt_writeopcount); wakeup(&mp->mnt_flag); curthread->td_pflags &= ~TDP_IGNSUSP; if ((flags & VR_START_WRITE) != 0) { MNT_REF(mp); mp->mnt_writeopcount++; } MNT_IUNLOCK(mp); if ((flags & VR_NO_SUSPCLR) == 0) VFS_SUSP_CLEAN(mp); } else if ((flags & VR_START_WRITE) != 0) { MNT_REF(mp); vn_start_write_locked(mp, 0); } else { MNT_IUNLOCK(mp); } } /* * Helper loop around vfs_write_suspend() for filesystem unmount VFS * methods. */ int vfs_write_suspend_umnt(struct mount *mp) { int error; MPASS(vn_suspendable(mp)); KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0, ("vfs_write_suspend_umnt: recursed")); /* dounmount() already called vn_start_write(). */ for (;;) { vn_finished_write(mp); error = vfs_write_suspend(mp, 0); if (error != 0) { vn_start_write(NULL, &mp, V_WAIT); return (error); } MNT_ILOCK(mp); if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0) break; MNT_IUNLOCK(mp); vn_start_write(NULL, &mp, V_WAIT); } mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2); wakeup(&mp->mnt_flag); MNT_IUNLOCK(mp); curthread->td_pflags |= TDP_IGNSUSP; return (0); } /* * Implement kqueues for files by translating it to vnode operation. */ static int vn_kqfilter(struct file *fp, struct knote *kn) { return (VOP_KQFILTER(fp->f_vnode, kn)); } /* * Simplified in-kernel wrapper calls for extended attribute access. * Both calls pass in a NULL credential, authorizing as "kernel" access. * Set IO_NODELOCKED in ioflg if the vnode is already locked. */ int vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, int *buflen, char *buf, struct thread *td) { struct uio auio; struct iovec iov; int error; iov.iov_len = *buflen; iov.iov_base = buf; auio.uio_iov = &iov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_READ; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = td; auio.uio_offset = 0; auio.uio_resid = *buflen; if ((ioflg & IO_NODELOCKED) == 0) vn_lock(vp, LK_SHARED | LK_RETRY); ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); /* authorize attribute retrieval as kernel */ error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL, td); if ((ioflg & IO_NODELOCKED) == 0) VOP_UNLOCK(vp, 0); if (error == 0) { *buflen = *buflen - auio.uio_resid; } return (error); } /* * XXX failure mode if partially written? */ int vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, int buflen, char *buf, struct thread *td) { struct uio auio; struct iovec iov; struct mount *mp; int error; iov.iov_len = buflen; iov.iov_base = buf; auio.uio_iov = &iov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_WRITE; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = td; auio.uio_offset = 0; auio.uio_resid = buflen; if ((ioflg & IO_NODELOCKED) == 0) { if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); /* authorize attribute setting as kernel */ error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td); if ((ioflg & IO_NODELOCKED) == 0) { vn_finished_write(mp); VOP_UNLOCK(vp, 0); } return (error); } int vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, struct thread *td) { struct mount *mp; int error; if ((ioflg & IO_NODELOCKED) == 0) { if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); /* authorize attribute removal as kernel */ error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td); if (error == EOPNOTSUPP) error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL, NULL, td); if ((ioflg & IO_NODELOCKED) == 0) { vn_finished_write(mp); VOP_UNLOCK(vp, 0); } return (error); } static int vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags, struct vnode **rvp) { return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp)); } int vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp) { return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino, lkflags, rvp)); } int vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg, int lkflags, struct vnode **rvp) { struct mount *mp; int ltype, error; ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get"); mp = vp->v_mount; ltype = VOP_ISLOCKED(vp); KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED, ("vn_vget_ino: vp not locked")); error = vfs_busy(mp, MBF_NOWAIT); if (error != 0) { vfs_ref(mp); VOP_UNLOCK(vp, 0); error = vfs_busy(mp, 0); vn_lock(vp, ltype | LK_RETRY); vfs_rel(mp); if (error != 0) return (ENOENT); if (vp->v_iflag & VI_DOOMED) { vfs_unbusy(mp); return (ENOENT); } } VOP_UNLOCK(vp, 0); error = alloc(mp, alloc_arg, lkflags, rvp); vfs_unbusy(mp); if (*rvp != vp) vn_lock(vp, ltype | LK_RETRY); if (vp->v_iflag & VI_DOOMED) { if (error == 0) { if (*rvp == vp) vunref(vp); else vput(*rvp); } error = ENOENT; } return (error); } int vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio, struct thread *td) { if (vp->v_type != VREG || td == NULL) return (0); if ((uoff_t)uio->uio_offset + uio->uio_resid > lim_cur(td, RLIMIT_FSIZE)) { PROC_LOCK(td->td_proc); kern_psignal(td->td_proc, SIGXFSZ); PROC_UNLOCK(td->td_proc); return (EFBIG); } return (0); } int vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td) { struct vnode *vp; vp = fp->f_vnode; #ifdef AUDIT vn_lock(vp, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(vp); VOP_UNLOCK(vp, 0); #endif return (setfmode(td, active_cred, vp, mode)); } int vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, struct thread *td) { struct vnode *vp; vp = fp->f_vnode; #ifdef AUDIT vn_lock(vp, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(vp); VOP_UNLOCK(vp, 0); #endif return (setfown(td, active_cred, vp, uid, gid)); } void vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end) { vm_object_t object; if ((object = vp->v_object) == NULL) return; VM_OBJECT_WLOCK(object); vm_object_page_remove(object, start, end, 0); VM_OBJECT_WUNLOCK(object); } int vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred) { struct vattr va; daddr_t bn, bnp; uint64_t bsize; off_t noff; int error; KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA, ("Wrong command %lu", cmd)); if (vn_lock(vp, LK_SHARED) != 0) return (EBADF); if (vp->v_type != VREG) { error = ENOTTY; goto unlock; } error = VOP_GETATTR(vp, &va, cred); if (error != 0) goto unlock; noff = *off; if (noff >= va.va_size) { error = ENXIO; goto unlock; } bsize = vp->v_mount->mnt_stat.f_iosize; for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) { error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL); if (error == EOPNOTSUPP) { error = ENOTTY; goto unlock; } if ((bnp == -1 && cmd == FIOSEEKHOLE) || (bnp != -1 && cmd == FIOSEEKDATA)) { noff = bn * bsize; if (noff < *off) noff = *off; goto unlock; } } if (noff > va.va_size) noff = va.va_size; /* noff == va.va_size. There is an implicit hole at the end of file. */ if (cmd == FIOSEEKDATA) error = ENXIO; unlock: VOP_UNLOCK(vp, 0); if (error == 0) *off = noff; return (error); } int vn_seek(struct file *fp, off_t offset, int whence, struct thread *td) { struct ucred *cred; struct vnode *vp; struct vattr vattr; off_t foffset, size; int error, noneg; cred = td->td_ucred; vp = fp->f_vnode; foffset = foffset_lock(fp, 0); noneg = (vp->v_type != VCHR); error = 0; switch (whence) { case L_INCR: if (noneg && (foffset < 0 || (offset > 0 && foffset > OFF_MAX - offset))) { error = EOVERFLOW; break; } offset += foffset; break; case L_XTND: vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_GETATTR(vp, &vattr, cred); VOP_UNLOCK(vp, 0); if (error) break; /* * If the file references a disk device, then fetch * the media size and use that to determine the ending * offset. */ if (vattr.va_size == 0 && vp->v_type == VCHR && fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0) vattr.va_size = size; if (noneg && (vattr.va_size > OFF_MAX || (offset > 0 && vattr.va_size > OFF_MAX - offset))) { error = EOVERFLOW; break; } offset += vattr.va_size; break; case L_SET: break; case SEEK_DATA: error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td); break; case SEEK_HOLE: error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td); break; default: error = EINVAL; } if (error == 0 && noneg && offset < 0) error = EINVAL; if (error != 0) goto drop; VFS_KNOTE_UNLOCKED(vp, 0); td->td_uretoff.tdu_off = offset; drop: foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0); return (error); } int vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred, struct thread *td) { int error; /* * Grant permission if the caller is the owner of the file, or * the super-user, or has ACL_WRITE_ATTRIBUTES permission on * on the file. If the time pointer is null, then write * permission on the file is also sufficient. * * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes: * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES * will be allowed to set the times [..] to the current * server time. */ error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td); if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0) error = VOP_ACCESS(vp, VWRITE, cred, td); return (error); } int vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) { struct vnode *vp; int error; if (fp->f_type == DTYPE_FIFO) kif->kf_type = KF_TYPE_FIFO; else kif->kf_type = KF_TYPE_VNODE; vp = fp->f_vnode; vref(vp); FILEDESC_SUNLOCK(fdp); error = vn_fill_kinfo_vnode(vp, kif); vrele(vp); FILEDESC_SLOCK(fdp); return (error); } static inline void vn_fill_junk(struct kinfo_file *kif) { size_t len, olen; /* * Simulate vn_fullpath returning changing values for a given * vp during e.g. coredump. */ len = (arc4random() % (sizeof(kif->kf_path) - 2)) + 1; olen = strlen(kif->kf_path); if (len < olen) strcpy(&kif->kf_path[len - 1], "$"); else for (; olen < len; olen++) strcpy(&kif->kf_path[olen], "A"); } int vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif) { struct vattr va; char *fullpath, *freepath; int error; kif->kf_vnode_type = vntype_to_kinfo(vp->v_type); freepath = NULL; fullpath = "-"; error = vn_fullpath(curthread, vp, &fullpath, &freepath); if (error == 0) { strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); } if (freepath != NULL) free(freepath, M_TEMP); KFAIL_POINT_CODE(DEBUG_FP, fill_kinfo_vnode__random_path, vn_fill_junk(kif); ); /* * Retrieve vnode attributes. */ va.va_fsid = VNOVAL; va.va_rdev = NODEV; vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_GETATTR(vp, &va, curthread->td_ucred); VOP_UNLOCK(vp, 0); if (error != 0) return (error); if (va.va_fsid != VNOVAL) kif->kf_un.kf_file.kf_file_fsid = va.va_fsid; else kif->kf_un.kf_file.kf_file_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; kif->kf_un.kf_file.kf_file_fileid = va.va_fileid; kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode); kif->kf_un.kf_file.kf_file_size = va.va_size; kif->kf_un.kf_file.kf_file_rdev = va.va_rdev; return (0); } int vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, struct thread *td) { #ifdef HWPMC_HOOKS struct pmckern_map_in pkm; #endif struct mount *mp; struct vnode *vp; vm_object_t object; vm_prot_t maxprot; boolean_t writecounted; int error; #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \ defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) /* * POSIX shared-memory objects are defined to have * kernel persistence, and are not defined to support * read(2)/write(2) -- or even open(2). Thus, we can * use MAP_ASYNC to trade on-disk coherence for speed. * The shm_open(3) library routine turns on the FPOSIXSHM * flag to request this behavior. */ if ((fp->f_flag & FPOSIXSHM) != 0) flags |= MAP_NOSYNC; #endif vp = fp->f_vnode; /* * Ensure that file and memory protections are * compatible. Note that we only worry about * writability if mapping is shared; in this case, * current and max prot are dictated by the open file. * XXX use the vnode instead? Problem is: what * credentials do we use for determination? What if * proc does a setuid? */ mp = vp->v_mount; if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) maxprot = VM_PROT_NONE; else maxprot = VM_PROT_EXECUTE; if ((fp->f_flag & FREAD) != 0) maxprot |= VM_PROT_READ; else if ((prot & VM_PROT_READ) != 0) return (EACCES); /* * If we are sharing potential changes via MAP_SHARED and we * are trying to get write permission although we opened it * without asking for it, bail out. */ if ((flags & MAP_SHARED) != 0) { if ((fp->f_flag & FWRITE) != 0) maxprot |= VM_PROT_WRITE; else if ((prot & VM_PROT_WRITE) != 0) return (EACCES); } else { maxprot |= VM_PROT_WRITE; cap_maxprot |= VM_PROT_WRITE; } maxprot &= cap_maxprot; writecounted = FALSE; error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, vp, &foff, &object, &writecounted); if (error != 0) return (error); error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, foff, writecounted, td); if (error != 0) { /* * If this mapping was accounted for in the vnode's * writecount, then undo that now. */ if (writecounted) vnode_pager_release_writecount(object, 0, size); vm_object_deallocate(object); } #ifdef HWPMC_HOOKS /* Inform hwpmc(4) if an executable is being mapped. */ if (error == 0 && (prot & VM_PROT_EXECUTE) != 0) { pkm.pm_file = vp; pkm.pm_address = (uintptr_t) *addr; PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm); } #endif return (error); } Index: head/sys/kern/vnode_if.src =================================================================== --- head/sys/kern/vnode_if.src (revision 305831) +++ head/sys/kern/vnode_if.src (revision 305832) @@ -1,752 +1,752 @@ #- # Copyright (c) 1992, 1993 # The Regents of the University of California. All rights reserved. # # Redistribution and use in source and binary forms, with or without # modification, are permitted provided that the following conditions # are met: # 1. Redistributions of source code must retain the above copyright # notice, this list of conditions and the following disclaimer. # 2. Redistributions in binary form must reproduce the above copyright # notice, this list of conditions and the following disclaimer in the # documentation and/or other materials provided with the distribution. -# 4. Neither the name of the University nor the names of its contributors +# 3. Neither the name of the University nor the names of its contributors # may be used to endorse or promote products derived from this software # without specific prior written permission. # # THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND # ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE # IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE # ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE # FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL # DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS # OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) # HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT # LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY # OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF # SUCH DAMAGE. # # @(#)vnode_if.src 8.12 (Berkeley) 5/14/95 # $FreeBSD$ # # # Above each of the vop descriptors in lines starting with %% # is a specification of the locking protocol used by each vop call. # The first column is the name of the variable, the remaining three # columns are in, out and error respectively. The "in" column defines # the lock state on input, the "out" column defines the state on successful # return, and the "error" column defines the locking state on error exit. # # The locking value can take the following values: # L: locked; not converted to type of lock. # E: locked with exclusive lock for this process. # U: unlocked. # -: not applicable. vnode does not yet (or no longer) exists. # =: the same on input and output, may be either L or U. # # The paramater named "vpp" is assumed to be always used with double # indirection (**vpp) and that name is hard-coded in vnode_if.awk ! # # Lines starting with %! specify a pre or post-condition function # to call before/after the vop call. # # If other such parameters are introduced, they have to be added to # the AWK script at the head of the definition of "add_debug_code()". # vop_islocked { IN struct vnode *vp; }; %% lookup dvp L L L %% lookup vpp - L - # XXX - the lookup locking protocol defies simple description and depends # on the flags and operation fields in the (cnp) structure. Note # especially that *vpp may equal dvp and both may be locked. vop_lookup { IN struct vnode *dvp; INOUT struct vnode **vpp; IN struct componentname *cnp; }; %% cachedlookup dvp L L L %% cachedlookup vpp - L - # This must be an exact copy of lookup. See kern/vfs_cache.c for details. vop_cachedlookup { IN struct vnode *dvp; INOUT struct vnode **vpp; IN struct componentname *cnp; }; %% create dvp E E E %% create vpp - L - %! create post vop_create_post vop_create { IN struct vnode *dvp; OUT struct vnode **vpp; IN struct componentname *cnp; IN struct vattr *vap; }; %% whiteout dvp E E E vop_whiteout { IN struct vnode *dvp; IN struct componentname *cnp; IN int flags; }; %% mknod dvp E E E %% mknod vpp - L - %! mknod post vop_mknod_post vop_mknod { IN struct vnode *dvp; OUT struct vnode **vpp; IN struct componentname *cnp; IN struct vattr *vap; }; %% open vp L L L %! open post vop_open_post vop_open { IN struct vnode *vp; IN int mode; IN struct ucred *cred; IN struct thread *td; IN struct file *fp; }; %% close vp L L L %! close post vop_close_post vop_close { IN struct vnode *vp; IN int fflag; IN struct ucred *cred; IN struct thread *td; }; %% access vp L L L vop_access { IN struct vnode *vp; IN accmode_t accmode; IN struct ucred *cred; IN struct thread *td; }; %% accessx vp L L L vop_accessx { IN struct vnode *vp; IN accmode_t accmode; IN struct ucred *cred; IN struct thread *td; }; %% getattr vp L L L vop_getattr { IN struct vnode *vp; OUT struct vattr *vap; IN struct ucred *cred; }; %% setattr vp E E E %! setattr post vop_setattr_post vop_setattr { IN struct vnode *vp; IN struct vattr *vap; IN struct ucred *cred; }; %% markatime vp L L L vop_markatime { IN struct vnode *vp; }; %% read vp L L L %! read post vop_read_post vop_read { IN struct vnode *vp; INOUT struct uio *uio; IN int ioflag; IN struct ucred *cred; }; %% write vp L L L %! write pre VOP_WRITE_PRE %! write post VOP_WRITE_POST vop_write { IN struct vnode *vp; INOUT struct uio *uio; IN int ioflag; IN struct ucred *cred; }; %% ioctl vp U U U vop_ioctl { IN struct vnode *vp; IN u_long command; IN void *data; IN int fflag; IN struct ucred *cred; IN struct thread *td; }; %% poll vp U U U vop_poll { IN struct vnode *vp; IN int events; IN struct ucred *cred; IN struct thread *td; }; %% kqfilter vp U U U vop_kqfilter { IN struct vnode *vp; IN struct knote *kn; }; %% revoke vp L L L vop_revoke { IN struct vnode *vp; IN int flags; }; %% fsync vp L L L vop_fsync { IN struct vnode *vp; IN int waitfor; IN struct thread *td; }; %% remove dvp E E E %% remove vp E E E %! remove post vop_remove_post vop_remove { IN struct vnode *dvp; IN struct vnode *vp; IN struct componentname *cnp; }; %% link tdvp E E E %% link vp E E E %! link post vop_link_post vop_link { IN struct vnode *tdvp; IN struct vnode *vp; IN struct componentname *cnp; }; %! rename pre vop_rename_pre %! rename post vop_rename_post vop_rename { IN WILLRELE struct vnode *fdvp; IN WILLRELE struct vnode *fvp; IN struct componentname *fcnp; IN WILLRELE struct vnode *tdvp; IN WILLRELE struct vnode *tvp; IN struct componentname *tcnp; }; %% mkdir dvp E E E %% mkdir vpp - E - %! mkdir post vop_mkdir_post vop_mkdir { IN struct vnode *dvp; OUT struct vnode **vpp; IN struct componentname *cnp; IN struct vattr *vap; }; %% rmdir dvp E E E %% rmdir vp E E E %! rmdir post vop_rmdir_post vop_rmdir { IN struct vnode *dvp; IN struct vnode *vp; IN struct componentname *cnp; }; %% symlink dvp E E E %% symlink vpp - E - %! symlink post vop_symlink_post vop_symlink { IN struct vnode *dvp; OUT struct vnode **vpp; IN struct componentname *cnp; IN struct vattr *vap; IN char *target; }; %% readdir vp L L L %! readdir post vop_readdir_post vop_readdir { IN struct vnode *vp; INOUT struct uio *uio; IN struct ucred *cred; INOUT int *eofflag; OUT int *ncookies; INOUT u_long **cookies; }; %% readlink vp L L L vop_readlink { IN struct vnode *vp; INOUT struct uio *uio; IN struct ucred *cred; }; %% inactive vp E E E vop_inactive { IN struct vnode *vp; IN struct thread *td; }; %% reclaim vp E E E %! reclaim post vop_reclaim_post vop_reclaim { IN struct vnode *vp; IN struct thread *td; }; %! lock1 pre vop_lock_pre %! lock1 post vop_lock_post vop_lock1 { IN struct vnode *vp; IN int flags; IN char *file; IN int line; }; %! unlock pre vop_unlock_pre %! unlock post vop_unlock_post vop_unlock { IN struct vnode *vp; IN int flags; }; %% bmap vp L L L vop_bmap { IN struct vnode *vp; IN daddr_t bn; OUT struct bufobj **bop; IN daddr_t *bnp; OUT int *runp; OUT int *runb; }; %% strategy vp L L L %! strategy pre vop_strategy_pre vop_strategy { IN struct vnode *vp; IN struct buf *bp; }; %% getwritemount vp = = = vop_getwritemount { IN struct vnode *vp; OUT struct mount **mpp; }; %% print vp - - - vop_print { IN struct vnode *vp; }; %% pathconf vp L L L vop_pathconf { IN struct vnode *vp; IN int name; OUT register_t *retval; }; %% advlock vp U U U vop_advlock { IN struct vnode *vp; IN void *id; IN int op; IN struct flock *fl; IN int flags; }; %% advlockasync vp U U U vop_advlockasync { IN struct vnode *vp; IN void *id; IN int op; IN struct flock *fl; IN int flags; IN struct task *task; INOUT void **cookiep; }; %% advlockpurge vp E E E vop_advlockpurge { IN struct vnode *vp; }; %% reallocblks vp E E E vop_reallocblks { IN struct vnode *vp; IN struct cluster_save *buflist; }; %% getpages vp L L L vop_getpages { IN struct vnode *vp; IN vm_page_t *m; IN int count; IN int *rbehind; IN int *rahead; }; %% getpages_async vp L L L vop_getpages_async { IN struct vnode *vp; IN vm_page_t *m; IN int count; IN int *rbehind; IN int *rahead; IN vop_getpages_iodone_t *iodone; IN void *arg; }; %% putpages vp L L L vop_putpages { IN struct vnode *vp; IN vm_page_t *m; IN int count; IN int sync; IN int *rtvals; }; %% getacl vp L L L vop_getacl { IN struct vnode *vp; IN acl_type_t type; OUT struct acl *aclp; IN struct ucred *cred; IN struct thread *td; }; %% setacl vp E E E vop_setacl { IN struct vnode *vp; IN acl_type_t type; IN struct acl *aclp; IN struct ucred *cred; IN struct thread *td; }; %% aclcheck vp = = = vop_aclcheck { IN struct vnode *vp; IN acl_type_t type; IN struct acl *aclp; IN struct ucred *cred; IN struct thread *td; }; %% closeextattr vp L L L vop_closeextattr { IN struct vnode *vp; IN int commit; IN struct ucred *cred; IN struct thread *td; }; %% getextattr vp L L L vop_getextattr { IN struct vnode *vp; IN int attrnamespace; IN const char *name; INOUT struct uio *uio; OUT size_t *size; IN struct ucred *cred; IN struct thread *td; }; %% listextattr vp L L L vop_listextattr { IN struct vnode *vp; IN int attrnamespace; INOUT struct uio *uio; OUT size_t *size; IN struct ucred *cred; IN struct thread *td; }; %% openextattr vp L L L vop_openextattr { IN struct vnode *vp; IN struct ucred *cred; IN struct thread *td; }; %% deleteextattr vp E E E %! deleteextattr post vop_deleteextattr_post vop_deleteextattr { IN struct vnode *vp; IN int attrnamespace; IN const char *name; IN struct ucred *cred; IN struct thread *td; }; %% setextattr vp E E E %! setextattr post vop_setextattr_post vop_setextattr { IN struct vnode *vp; IN int attrnamespace; IN const char *name; INOUT struct uio *uio; IN struct ucred *cred; IN struct thread *td; }; %% setlabel vp E E E vop_setlabel { IN struct vnode *vp; IN struct label *label; IN struct ucred *cred; IN struct thread *td; }; %% vptofh vp = = = vop_vptofh { IN struct vnode *vp; IN struct fid *fhp; }; %% vptocnp vp L L L %% vptocnp vpp - U - vop_vptocnp { IN struct vnode *vp; OUT struct vnode **vpp; IN struct ucred *cred; INOUT char *buf; INOUT int *buflen; }; %% allocate vp E E E vop_allocate { IN struct vnode *vp; INOUT off_t *offset; INOUT off_t *len; }; %% advise vp U U U vop_advise { IN struct vnode *vp; IN off_t start; IN off_t end; IN int advice; }; %% unp_bind vp E E E vop_unp_bind { IN struct vnode *vp; IN struct socket *socket; }; %% unp_connect vp L L L vop_unp_connect { IN struct vnode *vp; OUT struct socket **socket; }; %% unp_detach vp = = = vop_unp_detach { IN struct vnode *vp; }; %% is_text vp L L L vop_is_text { IN struct vnode *vp; }; %% set_text vp E E E vop_set_text { IN struct vnode *vp; }; %% vop_unset_text vp E E E vop_unset_text { IN struct vnode *vp; }; %% get_writecount vp L L L vop_get_writecount { IN struct vnode *vp; OUT int *writecount; }; %% add_writecount vp E E E vop_add_writecount { IN struct vnode *vp; IN int inc; }; %% fdatasync vp L L L vop_fdatasync { IN struct vnode *vp; IN struct thread *td; }; # The VOPs below are spares at the end of the table to allow new VOPs to be # added in stable branches without breaking the KBI. New VOPs in HEAD should # be added above these spares. When merging a new VOP to a stable branch, # the new VOP should replace one of the spares. vop_spare1 { IN struct vnode *vp; }; vop_spare2 { IN struct vnode *vp; }; vop_spare3 { IN struct vnode *vp; }; vop_spare4 { IN struct vnode *vp; }; vop_spare5 { IN struct vnode *vp; };