Index: stable/11/sys/kern/kern_proc.c =================================================================== --- stable/11/sys/kern/kern_proc.c (revision 327403) +++ stable/11/sys/kern/kern_proc.c (revision 327404) @@ -1,3133 +1,3133 @@ /*- * 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 * 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; +struct sx __exclusive_cache_line allproc_lock; +struct sx __exclusive_cache_line proctree_lock; +struct mtx __exclusive_cache_line 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; struct thread *td; 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); td = FIRST_THREAD_IN_PROC(p); if (td != NULL) { /* Make sure all thread constructors are executed */ EVENTHANDLER_INVOKE(thread_ctor, td); } 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); td_softdep_cleanup(td); MPASS(td->td_su == NULL); /* Make sure all thread destructors are executed */ EVENTHANDLER_INVOKE(thread_dtor, 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(pgid) register pid_t pgid; { register 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(p, pgid, pgrp, sess) register 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(p, pgrp) register 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(p, pgrp) 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(p) register 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(pgrp) register 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(pgrp, entering) 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(pg) struct pgrp *pg; { register 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) { register struct pgrp *pgrp; register struct proc *p; register 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)); if (strlcpy(kp->ki_tdname, td->td_name, sizeof(kp->ki_tdname)) >= sizeof(kp->ki_tdname)) { strlcpy(kp->ki_moretdname, td->td_name + sizeof(kp->ki_tdname) - 1, sizeof(kp->ki_moretdname)); } else { bzero(kp->ki_moretdname, sizeof(kp->ki_moretdname)); } 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); bcopy(ki->ki_moretdname, ki32->ki_moretdname, MAXCOMLEN - TDNAMLEN + 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); PTRTRIM_CP(*ki, *ki32, ki_tdaddr); 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 > ps_arg_cache_limit - sizeof(struct pargs)) 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 = atop(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 = atop(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: stable/11/sys/kern/kern_sysctl.c =================================================================== --- stable/11/sys/kern/kern_sysctl.c (revision 327403) +++ stable/11/sys/kern/kern_sysctl.c (revision 327404) @@ -1,2064 +1,2064 @@ /*- * 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 * 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; +static struct sx __exclusive_cache_line 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_register_disabled_oid(struct sysctl_oid *oidp) { /* * Mark the leaf as dormant if it's not to be immediately enabled. * We do not disable nodes as they can be shared between modules * and it is always safe to access a node. */ KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) == 0, ("internal flag is set in oid_kind")); if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE) oidp->oid_kind |= CTLFLAG_DORMANT; sysctl_register_oid(oidp); } void sysctl_enable_oid(struct sysctl_oid *oidp) { SYSCTL_ASSERT_WLOCKED(); if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) { KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) == 0, ("sysctl node is marked as dormant")); return; } KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) != 0, ("enabling already enabled sysctl oid")); oidp->oid_kind &= ~CTLFLAG_DORMANT; } 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("Warning: can't remove non-dynamic nodes (%s)!\n", oidp->oid_name); 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 | CTLFLAG_DORMANT)) != 0) 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) { if ((oid->oid_kind & CTLFLAG_DORMANT) != 0) return (ENOENT); *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: stable/11/sys/kern/subr_vmem.c =================================================================== --- stable/11/sys/kern/subr_vmem.c (revision 327403) +++ stable/11/sys/kern/subr_vmem.c (revision 327404) @@ -1,1586 +1,1586 @@ /*- * Copyright (c)2006,2007,2008,2009 YAMAMOTO Takashi, * Copyright (c) 2013 EMC Corp. * 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. * * 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. */ /* * From: * $NetBSD: vmem_impl.h,v 1.2 2013/01/29 21:26:24 para Exp $ * $NetBSD: subr_vmem.c,v 1.83 2013/03/06 11:20:10 yamt Exp $ */ /* * reference: * - Magazines and Vmem: Extending the Slab Allocator * to Many CPUs and Arbitrary Resources * http://www.usenix.org/event/usenix01/bonwick.html */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "opt_vm.h" #include #include #include #include #include #include #include #include #include #define VMEM_OPTORDER 5 #define VMEM_OPTVALUE (1 << VMEM_OPTORDER) #define VMEM_MAXORDER \ (VMEM_OPTVALUE - 1 + sizeof(vmem_size_t) * NBBY - VMEM_OPTORDER) #define VMEM_HASHSIZE_MIN 16 #define VMEM_HASHSIZE_MAX 131072 #define VMEM_QCACHE_IDX_MAX 16 #define VMEM_FITMASK (M_BESTFIT | M_FIRSTFIT) #define VMEM_FLAGS \ (M_NOWAIT | M_WAITOK | M_USE_RESERVE | M_NOVM | M_BESTFIT | M_FIRSTFIT) #define BT_FLAGS (M_NOWAIT | M_WAITOK | M_USE_RESERVE | M_NOVM) #define QC_NAME_MAX 16 /* * Data structures private to vmem. */ MALLOC_DEFINE(M_VMEM, "vmem", "vmem internal structures"); typedef struct vmem_btag bt_t; TAILQ_HEAD(vmem_seglist, vmem_btag); LIST_HEAD(vmem_freelist, vmem_btag); LIST_HEAD(vmem_hashlist, vmem_btag); struct qcache { uma_zone_t qc_cache; vmem_t *qc_vmem; vmem_size_t qc_size; char qc_name[QC_NAME_MAX]; }; typedef struct qcache qcache_t; #define QC_POOL_TO_QCACHE(pool) ((qcache_t *)(pool->pr_qcache)) #define VMEM_NAME_MAX 16 /* vmem arena */ struct vmem { struct mtx_padalign vm_lock; struct cv vm_cv; char vm_name[VMEM_NAME_MAX+1]; LIST_ENTRY(vmem) vm_alllist; struct vmem_hashlist vm_hash0[VMEM_HASHSIZE_MIN]; struct vmem_freelist vm_freelist[VMEM_MAXORDER]; struct vmem_seglist vm_seglist; struct vmem_hashlist *vm_hashlist; vmem_size_t vm_hashsize; /* Constant after init */ vmem_size_t vm_qcache_max; vmem_size_t vm_quantum_mask; vmem_size_t vm_import_quantum; int vm_quantum_shift; /* Written on alloc/free */ LIST_HEAD(, vmem_btag) vm_freetags; int vm_nfreetags; int vm_nbusytag; vmem_size_t vm_inuse; vmem_size_t vm_size; /* Used on import. */ vmem_import_t *vm_importfn; vmem_release_t *vm_releasefn; void *vm_arg; /* Space exhaustion callback. */ vmem_reclaim_t *vm_reclaimfn; /* quantum cache */ qcache_t vm_qcache[VMEM_QCACHE_IDX_MAX]; }; /* boundary tag */ struct vmem_btag { TAILQ_ENTRY(vmem_btag) bt_seglist; union { LIST_ENTRY(vmem_btag) u_freelist; /* BT_TYPE_FREE */ LIST_ENTRY(vmem_btag) u_hashlist; /* BT_TYPE_BUSY */ } bt_u; #define bt_hashlist bt_u.u_hashlist #define bt_freelist bt_u.u_freelist vmem_addr_t bt_start; vmem_size_t bt_size; int bt_type; }; #define BT_TYPE_SPAN 1 /* Allocated from importfn */ #define BT_TYPE_SPAN_STATIC 2 /* vmem_add() or create. */ #define BT_TYPE_FREE 3 /* Available space. */ #define BT_TYPE_BUSY 4 /* Used space. */ #define BT_ISSPAN_P(bt) ((bt)->bt_type <= BT_TYPE_SPAN_STATIC) #define BT_END(bt) ((bt)->bt_start + (bt)->bt_size - 1) #if defined(DIAGNOSTIC) static int enable_vmem_check = 1; SYSCTL_INT(_debug, OID_AUTO, vmem_check, CTLFLAG_RWTUN, &enable_vmem_check, 0, "Enable vmem check"); static void vmem_check(vmem_t *); #endif static struct callout vmem_periodic_ch; static int vmem_periodic_interval; static struct task vmem_periodic_wk; -static struct mtx_padalign vmem_list_lock; +static struct mtx_padalign __exclusive_cache_line vmem_list_lock; static LIST_HEAD(, vmem) vmem_list = LIST_HEAD_INITIALIZER(vmem_list); /* ---- misc */ #define VMEM_CONDVAR_INIT(vm, wchan) cv_init(&vm->vm_cv, wchan) #define VMEM_CONDVAR_DESTROY(vm) cv_destroy(&vm->vm_cv) #define VMEM_CONDVAR_WAIT(vm) cv_wait(&vm->vm_cv, &vm->vm_lock) #define VMEM_CONDVAR_BROADCAST(vm) cv_broadcast(&vm->vm_cv) #define VMEM_LOCK(vm) mtx_lock(&vm->vm_lock) #define VMEM_TRYLOCK(vm) mtx_trylock(&vm->vm_lock) #define VMEM_UNLOCK(vm) mtx_unlock(&vm->vm_lock) #define VMEM_LOCK_INIT(vm, name) mtx_init(&vm->vm_lock, (name), NULL, MTX_DEF) #define VMEM_LOCK_DESTROY(vm) mtx_destroy(&vm->vm_lock) #define VMEM_ASSERT_LOCKED(vm) mtx_assert(&vm->vm_lock, MA_OWNED); #define VMEM_ALIGNUP(addr, align) (-(-(addr) & -(align))) #define VMEM_CROSS_P(addr1, addr2, boundary) \ ((((addr1) ^ (addr2)) & -(boundary)) != 0) #define ORDER2SIZE(order) ((order) < VMEM_OPTVALUE ? ((order) + 1) : \ (vmem_size_t)1 << ((order) - (VMEM_OPTVALUE - VMEM_OPTORDER - 1))) #define SIZE2ORDER(size) ((size) <= VMEM_OPTVALUE ? ((size) - 1) : \ (flsl(size) + (VMEM_OPTVALUE - VMEM_OPTORDER - 2))) /* * Maximum number of boundary tags that may be required to satisfy an * allocation. Two may be required to import. Another two may be * required to clip edges. */ #define BT_MAXALLOC 4 /* * Max free limits the number of locally cached boundary tags. We * just want to avoid hitting the zone allocator for every call. */ #define BT_MAXFREE (BT_MAXALLOC * 8) /* Allocator for boundary tags. */ static uma_zone_t vmem_bt_zone; /* boot time arena storage. */ static struct vmem kernel_arena_storage; static struct vmem kmem_arena_storage; static struct vmem buffer_arena_storage; static struct vmem transient_arena_storage; vmem_t *kernel_arena = &kernel_arena_storage; vmem_t *kmem_arena = &kmem_arena_storage; vmem_t *buffer_arena = &buffer_arena_storage; vmem_t *transient_arena = &transient_arena_storage; #ifdef DEBUG_MEMGUARD static struct vmem memguard_arena_storage; vmem_t *memguard_arena = &memguard_arena_storage; #endif /* * Fill the vmem's boundary tag cache. We guarantee that boundary tag * allocation will not fail once bt_fill() passes. To do so we cache * at least the maximum possible tag allocations in the arena. */ static int bt_fill(vmem_t *vm, int flags) { bt_t *bt; VMEM_ASSERT_LOCKED(vm); /* * Only allow the kmem arena to dip into reserve tags. It is the * vmem where new tags come from. */ flags &= BT_FLAGS; if (vm != kmem_arena) flags &= ~M_USE_RESERVE; /* * Loop until we meet the reserve. To minimize the lock shuffle * and prevent simultaneous fills we first try a NOWAIT regardless * of the caller's flags. Specify M_NOVM so we don't recurse while * holding a vmem lock. */ while (vm->vm_nfreetags < BT_MAXALLOC) { bt = uma_zalloc(vmem_bt_zone, (flags & M_USE_RESERVE) | M_NOWAIT | M_NOVM); if (bt == NULL) { VMEM_UNLOCK(vm); bt = uma_zalloc(vmem_bt_zone, flags); VMEM_LOCK(vm); if (bt == NULL && (flags & M_NOWAIT) != 0) break; } LIST_INSERT_HEAD(&vm->vm_freetags, bt, bt_freelist); vm->vm_nfreetags++; } if (vm->vm_nfreetags < BT_MAXALLOC) return ENOMEM; return 0; } /* * Pop a tag off of the freetag stack. */ static bt_t * bt_alloc(vmem_t *vm) { bt_t *bt; VMEM_ASSERT_LOCKED(vm); bt = LIST_FIRST(&vm->vm_freetags); MPASS(bt != NULL); LIST_REMOVE(bt, bt_freelist); vm->vm_nfreetags--; return bt; } /* * Trim the per-vmem free list. Returns with the lock released to * avoid allocator recursions. */ static void bt_freetrim(vmem_t *vm, int freelimit) { LIST_HEAD(, vmem_btag) freetags; bt_t *bt; LIST_INIT(&freetags); VMEM_ASSERT_LOCKED(vm); while (vm->vm_nfreetags > freelimit) { bt = LIST_FIRST(&vm->vm_freetags); LIST_REMOVE(bt, bt_freelist); vm->vm_nfreetags--; LIST_INSERT_HEAD(&freetags, bt, bt_freelist); } VMEM_UNLOCK(vm); while ((bt = LIST_FIRST(&freetags)) != NULL) { LIST_REMOVE(bt, bt_freelist); uma_zfree(vmem_bt_zone, bt); } } static inline void bt_free(vmem_t *vm, bt_t *bt) { VMEM_ASSERT_LOCKED(vm); MPASS(LIST_FIRST(&vm->vm_freetags) != bt); LIST_INSERT_HEAD(&vm->vm_freetags, bt, bt_freelist); vm->vm_nfreetags++; } /* * freelist[0] ... [1, 1] * freelist[1] ... [2, 2] * : * freelist[29] ... [30, 30] * freelist[30] ... [31, 31] * freelist[31] ... [32, 63] * freelist[33] ... [64, 127] * : * freelist[n] ... [(1 << (n - 26)), (1 << (n - 25)) - 1] * : */ static struct vmem_freelist * bt_freehead_tofree(vmem_t *vm, vmem_size_t size) { const vmem_size_t qsize = size >> vm->vm_quantum_shift; const int idx = SIZE2ORDER(qsize); MPASS(size != 0 && qsize != 0); MPASS((size & vm->vm_quantum_mask) == 0); MPASS(idx >= 0); MPASS(idx < VMEM_MAXORDER); return &vm->vm_freelist[idx]; } /* * bt_freehead_toalloc: return the freelist for the given size and allocation * strategy. * * For M_FIRSTFIT, return the list in which any blocks are large enough * for the requested size. otherwise, return the list which can have blocks * large enough for the requested size. */ static struct vmem_freelist * bt_freehead_toalloc(vmem_t *vm, vmem_size_t size, int strat) { const vmem_size_t qsize = size >> vm->vm_quantum_shift; int idx = SIZE2ORDER(qsize); MPASS(size != 0 && qsize != 0); MPASS((size & vm->vm_quantum_mask) == 0); if (strat == M_FIRSTFIT && ORDER2SIZE(idx) != qsize) { idx++; /* check too large request? */ } MPASS(idx >= 0); MPASS(idx < VMEM_MAXORDER); return &vm->vm_freelist[idx]; } /* ---- boundary tag hash */ static struct vmem_hashlist * bt_hashhead(vmem_t *vm, vmem_addr_t addr) { struct vmem_hashlist *list; unsigned int hash; hash = hash32_buf(&addr, sizeof(addr), 0); list = &vm->vm_hashlist[hash % vm->vm_hashsize]; return list; } static bt_t * bt_lookupbusy(vmem_t *vm, vmem_addr_t addr) { struct vmem_hashlist *list; bt_t *bt; VMEM_ASSERT_LOCKED(vm); list = bt_hashhead(vm, addr); LIST_FOREACH(bt, list, bt_hashlist) { if (bt->bt_start == addr) { break; } } return bt; } static void bt_rembusy(vmem_t *vm, bt_t *bt) { VMEM_ASSERT_LOCKED(vm); MPASS(vm->vm_nbusytag > 0); vm->vm_inuse -= bt->bt_size; vm->vm_nbusytag--; LIST_REMOVE(bt, bt_hashlist); } static void bt_insbusy(vmem_t *vm, bt_t *bt) { struct vmem_hashlist *list; VMEM_ASSERT_LOCKED(vm); MPASS(bt->bt_type == BT_TYPE_BUSY); list = bt_hashhead(vm, bt->bt_start); LIST_INSERT_HEAD(list, bt, bt_hashlist); vm->vm_nbusytag++; vm->vm_inuse += bt->bt_size; } /* ---- boundary tag list */ static void bt_remseg(vmem_t *vm, bt_t *bt) { TAILQ_REMOVE(&vm->vm_seglist, bt, bt_seglist); bt_free(vm, bt); } static void bt_insseg(vmem_t *vm, bt_t *bt, bt_t *prev) { TAILQ_INSERT_AFTER(&vm->vm_seglist, prev, bt, bt_seglist); } static void bt_insseg_tail(vmem_t *vm, bt_t *bt) { TAILQ_INSERT_TAIL(&vm->vm_seglist, bt, bt_seglist); } static void bt_remfree(vmem_t *vm, bt_t *bt) { MPASS(bt->bt_type == BT_TYPE_FREE); LIST_REMOVE(bt, bt_freelist); } static void bt_insfree(vmem_t *vm, bt_t *bt) { struct vmem_freelist *list; list = bt_freehead_tofree(vm, bt->bt_size); LIST_INSERT_HEAD(list, bt, bt_freelist); } /* ---- vmem internal functions */ /* * Import from the arena into the quantum cache in UMA. */ static int qc_import(void *arg, void **store, int cnt, int flags) { qcache_t *qc; vmem_addr_t addr; int i; qc = arg; if ((flags & VMEM_FITMASK) == 0) flags |= M_BESTFIT; for (i = 0; i < cnt; i++) { if (vmem_xalloc(qc->qc_vmem, qc->qc_size, 0, 0, 0, VMEM_ADDR_MIN, VMEM_ADDR_MAX, flags, &addr) != 0) break; store[i] = (void *)addr; /* Only guarantee one allocation. */ flags &= ~M_WAITOK; flags |= M_NOWAIT; } return i; } /* * Release memory from the UMA cache to the arena. */ static void qc_release(void *arg, void **store, int cnt) { qcache_t *qc; int i; qc = arg; for (i = 0; i < cnt; i++) vmem_xfree(qc->qc_vmem, (vmem_addr_t)store[i], qc->qc_size); } static void qc_init(vmem_t *vm, vmem_size_t qcache_max) { qcache_t *qc; vmem_size_t size; int qcache_idx_max; int i; MPASS((qcache_max & vm->vm_quantum_mask) == 0); qcache_idx_max = MIN(qcache_max >> vm->vm_quantum_shift, VMEM_QCACHE_IDX_MAX); vm->vm_qcache_max = qcache_idx_max << vm->vm_quantum_shift; for (i = 0; i < qcache_idx_max; i++) { qc = &vm->vm_qcache[i]; size = (i + 1) << vm->vm_quantum_shift; snprintf(qc->qc_name, sizeof(qc->qc_name), "%s-%zu", vm->vm_name, size); qc->qc_vmem = vm; qc->qc_size = size; qc->qc_cache = uma_zcache_create(qc->qc_name, size, NULL, NULL, NULL, NULL, qc_import, qc_release, qc, UMA_ZONE_VM); MPASS(qc->qc_cache); } } static void qc_destroy(vmem_t *vm) { int qcache_idx_max; int i; qcache_idx_max = vm->vm_qcache_max >> vm->vm_quantum_shift; for (i = 0; i < qcache_idx_max; i++) uma_zdestroy(vm->vm_qcache[i].qc_cache); } static void qc_drain(vmem_t *vm) { int qcache_idx_max; int i; qcache_idx_max = vm->vm_qcache_max >> vm->vm_quantum_shift; for (i = 0; i < qcache_idx_max; i++) zone_drain(vm->vm_qcache[i].qc_cache); } #ifndef UMA_MD_SMALL_ALLOC -static struct mtx_padalign vmem_bt_lock; +static struct mtx_padalign __exclusive_cache_line vmem_bt_lock; /* * vmem_bt_alloc: Allocate a new page of boundary tags. * * On architectures with uma_small_alloc there is no recursion; no address * space need be allocated to allocate boundary tags. For the others, we * must handle recursion. Boundary tags are necessary to allocate new * boundary tags. * * UMA guarantees that enough tags are held in reserve to allocate a new * page of kva. We dip into this reserve by specifying M_USE_RESERVE only * when allocating the page to hold new boundary tags. In this way the * reserve is automatically filled by the allocation that uses the reserve. * * We still have to guarantee that the new tags are allocated atomically since * many threads may try concurrently. The bt_lock provides this guarantee. * We convert WAITOK allocations to NOWAIT and then handle the blocking here * on failure. It's ok to return NULL for a WAITOK allocation as UMA will * loop again after checking to see if we lost the race to allocate. * * There is a small race between vmem_bt_alloc() returning the page and the * zone lock being acquired to add the page to the zone. For WAITOK * allocations we just pause briefly. NOWAIT may experience a transient * failure. To alleviate this we permit a small number of simultaneous * fills to proceed concurrently so NOWAIT is less likely to fail unless * we are really out of KVA. */ static void * vmem_bt_alloc(uma_zone_t zone, vm_size_t bytes, uint8_t *pflag, int wait) { vmem_addr_t addr; *pflag = UMA_SLAB_KMEM; /* * Single thread boundary tag allocation so that the address space * and memory are added in one atomic operation. */ mtx_lock(&vmem_bt_lock); if (vmem_xalloc(kmem_arena, bytes, 0, 0, 0, VMEM_ADDR_MIN, VMEM_ADDR_MAX, M_NOWAIT | M_NOVM | M_USE_RESERVE | M_BESTFIT, &addr) == 0) { if (kmem_back(kmem_object, addr, bytes, M_NOWAIT | M_USE_RESERVE) == 0) { mtx_unlock(&vmem_bt_lock); return ((void *)addr); } vmem_xfree(kmem_arena, addr, bytes); mtx_unlock(&vmem_bt_lock); /* * Out of memory, not address space. This may not even be * possible due to M_USE_RESERVE page allocation. */ if (wait & M_WAITOK) VM_WAIT; return (NULL); } mtx_unlock(&vmem_bt_lock); /* * We're either out of address space or lost a fill race. */ if (wait & M_WAITOK) pause("btalloc", 1); return (NULL); } #endif void vmem_startup(void) { mtx_init(&vmem_list_lock, "vmem list lock", NULL, MTX_DEF); vmem_bt_zone = uma_zcreate("vmem btag", sizeof(struct vmem_btag), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM); #ifndef UMA_MD_SMALL_ALLOC mtx_init(&vmem_bt_lock, "btag lock", NULL, MTX_DEF); uma_prealloc(vmem_bt_zone, BT_MAXALLOC); /* * Reserve enough tags to allocate new tags. We allow multiple * CPUs to attempt to allocate new tags concurrently to limit * false restarts in UMA. */ uma_zone_reserve(vmem_bt_zone, BT_MAXALLOC * (mp_ncpus + 1) / 2); uma_zone_set_allocf(vmem_bt_zone, vmem_bt_alloc); #endif } /* ---- rehash */ static int vmem_rehash(vmem_t *vm, vmem_size_t newhashsize) { bt_t *bt; int i; struct vmem_hashlist *newhashlist; struct vmem_hashlist *oldhashlist; vmem_size_t oldhashsize; MPASS(newhashsize > 0); newhashlist = malloc(sizeof(struct vmem_hashlist) * newhashsize, M_VMEM, M_NOWAIT); if (newhashlist == NULL) return ENOMEM; for (i = 0; i < newhashsize; i++) { LIST_INIT(&newhashlist[i]); } VMEM_LOCK(vm); oldhashlist = vm->vm_hashlist; oldhashsize = vm->vm_hashsize; vm->vm_hashlist = newhashlist; vm->vm_hashsize = newhashsize; if (oldhashlist == NULL) { VMEM_UNLOCK(vm); return 0; } for (i = 0; i < oldhashsize; i++) { while ((bt = LIST_FIRST(&oldhashlist[i])) != NULL) { bt_rembusy(vm, bt); bt_insbusy(vm, bt); } } VMEM_UNLOCK(vm); if (oldhashlist != vm->vm_hash0) { free(oldhashlist, M_VMEM); } return 0; } static void vmem_periodic_kick(void *dummy) { taskqueue_enqueue(taskqueue_thread, &vmem_periodic_wk); } static void vmem_periodic(void *unused, int pending) { vmem_t *vm; vmem_size_t desired; vmem_size_t current; mtx_lock(&vmem_list_lock); LIST_FOREACH(vm, &vmem_list, vm_alllist) { #ifdef DIAGNOSTIC /* Convenient time to verify vmem state. */ if (enable_vmem_check == 1) { VMEM_LOCK(vm); vmem_check(vm); VMEM_UNLOCK(vm); } #endif desired = 1 << flsl(vm->vm_nbusytag); desired = MIN(MAX(desired, VMEM_HASHSIZE_MIN), VMEM_HASHSIZE_MAX); current = vm->vm_hashsize; /* Grow in powers of two. Shrink less aggressively. */ if (desired >= current * 2 || desired * 4 <= current) vmem_rehash(vm, desired); /* * Periodically wake up threads waiting for resources, * so they could ask for reclamation again. */ VMEM_CONDVAR_BROADCAST(vm); } mtx_unlock(&vmem_list_lock); callout_reset(&vmem_periodic_ch, vmem_periodic_interval, vmem_periodic_kick, NULL); } static void vmem_start_callout(void *unused) { TASK_INIT(&vmem_periodic_wk, 0, vmem_periodic, NULL); vmem_periodic_interval = hz * 10; callout_init(&vmem_periodic_ch, 1); callout_reset(&vmem_periodic_ch, vmem_periodic_interval, vmem_periodic_kick, NULL); } SYSINIT(vfs, SI_SUB_CONFIGURE, SI_ORDER_ANY, vmem_start_callout, NULL); static void vmem_add1(vmem_t *vm, vmem_addr_t addr, vmem_size_t size, int type) { bt_t *btspan; bt_t *btfree; MPASS(type == BT_TYPE_SPAN || type == BT_TYPE_SPAN_STATIC); MPASS((size & vm->vm_quantum_mask) == 0); btspan = bt_alloc(vm); btspan->bt_type = type; btspan->bt_start = addr; btspan->bt_size = size; bt_insseg_tail(vm, btspan); btfree = bt_alloc(vm); btfree->bt_type = BT_TYPE_FREE; btfree->bt_start = addr; btfree->bt_size = size; bt_insseg(vm, btfree, btspan); bt_insfree(vm, btfree); vm->vm_size += size; } static void vmem_destroy1(vmem_t *vm) { bt_t *bt; /* * Drain per-cpu quantum caches. */ qc_destroy(vm); /* * The vmem should now only contain empty segments. */ VMEM_LOCK(vm); MPASS(vm->vm_nbusytag == 0); while ((bt = TAILQ_FIRST(&vm->vm_seglist)) != NULL) bt_remseg(vm, bt); if (vm->vm_hashlist != NULL && vm->vm_hashlist != vm->vm_hash0) free(vm->vm_hashlist, M_VMEM); bt_freetrim(vm, 0); VMEM_CONDVAR_DESTROY(vm); VMEM_LOCK_DESTROY(vm); free(vm, M_VMEM); } static int vmem_import(vmem_t *vm, vmem_size_t size, vmem_size_t align, int flags) { vmem_addr_t addr; int error; if (vm->vm_importfn == NULL) return EINVAL; /* * To make sure we get a span that meets the alignment we double it * and add the size to the tail. This slightly overestimates. */ if (align != vm->vm_quantum_mask + 1) size = (align * 2) + size; size = roundup(size, vm->vm_import_quantum); /* * Hide MAXALLOC tags so we're guaranteed to be able to add this * span and the tag we want to allocate from it. */ MPASS(vm->vm_nfreetags >= BT_MAXALLOC); vm->vm_nfreetags -= BT_MAXALLOC; VMEM_UNLOCK(vm); error = (vm->vm_importfn)(vm->vm_arg, size, flags, &addr); VMEM_LOCK(vm); vm->vm_nfreetags += BT_MAXALLOC; if (error) return ENOMEM; vmem_add1(vm, addr, size, BT_TYPE_SPAN); return 0; } /* * vmem_fit: check if a bt can satisfy the given restrictions. * * it's a caller's responsibility to ensure the region is big enough * before calling us. */ static int vmem_fit(const bt_t *bt, vmem_size_t size, vmem_size_t align, vmem_size_t phase, vmem_size_t nocross, vmem_addr_t minaddr, vmem_addr_t maxaddr, vmem_addr_t *addrp) { vmem_addr_t start; vmem_addr_t end; MPASS(size > 0); MPASS(bt->bt_size >= size); /* caller's responsibility */ /* * XXX assumption: vmem_addr_t and vmem_size_t are * unsigned integer of the same size. */ start = bt->bt_start; if (start < minaddr) { start = minaddr; } end = BT_END(bt); if (end > maxaddr) end = maxaddr; if (start > end) return (ENOMEM); start = VMEM_ALIGNUP(start - phase, align) + phase; if (start < bt->bt_start) start += align; if (VMEM_CROSS_P(start, start + size - 1, nocross)) { MPASS(align < nocross); start = VMEM_ALIGNUP(start - phase, nocross) + phase; } if (start <= end && end - start >= size - 1) { MPASS((start & (align - 1)) == phase); MPASS(!VMEM_CROSS_P(start, start + size - 1, nocross)); MPASS(minaddr <= start); MPASS(maxaddr == 0 || start + size - 1 <= maxaddr); MPASS(bt->bt_start <= start); MPASS(BT_END(bt) - start >= size - 1); *addrp = start; return (0); } return (ENOMEM); } /* * vmem_clip: Trim the boundary tag edges to the requested start and size. */ static void vmem_clip(vmem_t *vm, bt_t *bt, vmem_addr_t start, vmem_size_t size) { bt_t *btnew; bt_t *btprev; VMEM_ASSERT_LOCKED(vm); MPASS(bt->bt_type == BT_TYPE_FREE); MPASS(bt->bt_size >= size); bt_remfree(vm, bt); if (bt->bt_start != start) { btprev = bt_alloc(vm); btprev->bt_type = BT_TYPE_FREE; btprev->bt_start = bt->bt_start; btprev->bt_size = start - bt->bt_start; bt->bt_start = start; bt->bt_size -= btprev->bt_size; bt_insfree(vm, btprev); bt_insseg(vm, btprev, TAILQ_PREV(bt, vmem_seglist, bt_seglist)); } MPASS(bt->bt_start == start); if (bt->bt_size != size && bt->bt_size - size > vm->vm_quantum_mask) { /* split */ btnew = bt_alloc(vm); btnew->bt_type = BT_TYPE_BUSY; btnew->bt_start = bt->bt_start; btnew->bt_size = size; bt->bt_start = bt->bt_start + size; bt->bt_size -= size; bt_insfree(vm, bt); bt_insseg(vm, btnew, TAILQ_PREV(bt, vmem_seglist, bt_seglist)); bt_insbusy(vm, btnew); bt = btnew; } else { bt->bt_type = BT_TYPE_BUSY; bt_insbusy(vm, bt); } MPASS(bt->bt_size >= size); bt->bt_type = BT_TYPE_BUSY; } /* ---- vmem API */ void vmem_set_import(vmem_t *vm, vmem_import_t *importfn, vmem_release_t *releasefn, void *arg, vmem_size_t import_quantum) { VMEM_LOCK(vm); vm->vm_importfn = importfn; vm->vm_releasefn = releasefn; vm->vm_arg = arg; vm->vm_import_quantum = import_quantum; VMEM_UNLOCK(vm); } void vmem_set_reclaim(vmem_t *vm, vmem_reclaim_t *reclaimfn) { VMEM_LOCK(vm); vm->vm_reclaimfn = reclaimfn; VMEM_UNLOCK(vm); } /* * vmem_init: Initializes vmem arena. */ vmem_t * vmem_init(vmem_t *vm, const char *name, vmem_addr_t base, vmem_size_t size, vmem_size_t quantum, vmem_size_t qcache_max, int flags) { int i; MPASS(quantum > 0); MPASS((quantum & (quantum - 1)) == 0); bzero(vm, sizeof(*vm)); VMEM_CONDVAR_INIT(vm, name); VMEM_LOCK_INIT(vm, name); vm->vm_nfreetags = 0; LIST_INIT(&vm->vm_freetags); strlcpy(vm->vm_name, name, sizeof(vm->vm_name)); vm->vm_quantum_mask = quantum - 1; vm->vm_quantum_shift = flsl(quantum) - 1; vm->vm_nbusytag = 0; vm->vm_size = 0; vm->vm_inuse = 0; qc_init(vm, qcache_max); TAILQ_INIT(&vm->vm_seglist); for (i = 0; i < VMEM_MAXORDER; i++) { LIST_INIT(&vm->vm_freelist[i]); } memset(&vm->vm_hash0, 0, sizeof(vm->vm_hash0)); vm->vm_hashsize = VMEM_HASHSIZE_MIN; vm->vm_hashlist = vm->vm_hash0; if (size != 0) { if (vmem_add(vm, base, size, flags) != 0) { vmem_destroy1(vm); return NULL; } } mtx_lock(&vmem_list_lock); LIST_INSERT_HEAD(&vmem_list, vm, vm_alllist); mtx_unlock(&vmem_list_lock); return vm; } /* * vmem_create: create an arena. */ vmem_t * vmem_create(const char *name, vmem_addr_t base, vmem_size_t size, vmem_size_t quantum, vmem_size_t qcache_max, int flags) { vmem_t *vm; vm = malloc(sizeof(*vm), M_VMEM, flags & (M_WAITOK|M_NOWAIT)); if (vm == NULL) return (NULL); if (vmem_init(vm, name, base, size, quantum, qcache_max, flags) == NULL) return (NULL); return (vm); } void vmem_destroy(vmem_t *vm) { mtx_lock(&vmem_list_lock); LIST_REMOVE(vm, vm_alllist); mtx_unlock(&vmem_list_lock); vmem_destroy1(vm); } vmem_size_t vmem_roundup_size(vmem_t *vm, vmem_size_t size) { return (size + vm->vm_quantum_mask) & ~vm->vm_quantum_mask; } /* * vmem_alloc: allocate resource from the arena. */ int vmem_alloc(vmem_t *vm, vmem_size_t size, int flags, vmem_addr_t *addrp) { const int strat __unused = flags & VMEM_FITMASK; qcache_t *qc; flags &= VMEM_FLAGS; MPASS(size > 0); MPASS(strat == M_BESTFIT || strat == M_FIRSTFIT); if ((flags & M_NOWAIT) == 0) WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "vmem_alloc"); if (size <= vm->vm_qcache_max) { qc = &vm->vm_qcache[(size - 1) >> vm->vm_quantum_shift]; *addrp = (vmem_addr_t)uma_zalloc(qc->qc_cache, flags); if (*addrp == 0) return (ENOMEM); return (0); } return vmem_xalloc(vm, size, 0, 0, 0, VMEM_ADDR_MIN, VMEM_ADDR_MAX, flags, addrp); } int vmem_xalloc(vmem_t *vm, const vmem_size_t size0, vmem_size_t align, const vmem_size_t phase, const vmem_size_t nocross, const vmem_addr_t minaddr, const vmem_addr_t maxaddr, int flags, vmem_addr_t *addrp) { const vmem_size_t size = vmem_roundup_size(vm, size0); struct vmem_freelist *list; struct vmem_freelist *first; struct vmem_freelist *end; vmem_size_t avail; bt_t *bt; int error; int strat; flags &= VMEM_FLAGS; strat = flags & VMEM_FITMASK; MPASS(size0 > 0); MPASS(size > 0); MPASS(strat == M_BESTFIT || strat == M_FIRSTFIT); MPASS((flags & (M_NOWAIT|M_WAITOK)) != (M_NOWAIT|M_WAITOK)); if ((flags & M_NOWAIT) == 0) WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "vmem_xalloc"); MPASS((align & vm->vm_quantum_mask) == 0); MPASS((align & (align - 1)) == 0); MPASS((phase & vm->vm_quantum_mask) == 0); MPASS((nocross & vm->vm_quantum_mask) == 0); MPASS((nocross & (nocross - 1)) == 0); MPASS((align == 0 && phase == 0) || phase < align); MPASS(nocross == 0 || nocross >= size); MPASS(minaddr <= maxaddr); MPASS(!VMEM_CROSS_P(phase, phase + size - 1, nocross)); if (align == 0) align = vm->vm_quantum_mask + 1; *addrp = 0; end = &vm->vm_freelist[VMEM_MAXORDER]; /* * choose a free block from which we allocate. */ first = bt_freehead_toalloc(vm, size, strat); VMEM_LOCK(vm); for (;;) { /* * Make sure we have enough tags to complete the * operation. */ if (vm->vm_nfreetags < BT_MAXALLOC && bt_fill(vm, flags) != 0) { error = ENOMEM; break; } /* * Scan freelists looking for a tag that satisfies the * allocation. If we're doing BESTFIT we may encounter * sizes below the request. If we're doing FIRSTFIT we * inspect only the first element from each list. */ for (list = first; list < end; list++) { LIST_FOREACH(bt, list, bt_freelist) { if (bt->bt_size >= size) { error = vmem_fit(bt, size, align, phase, nocross, minaddr, maxaddr, addrp); if (error == 0) { vmem_clip(vm, bt, *addrp, size); goto out; } } /* FIRST skips to the next list. */ if (strat == M_FIRSTFIT) break; } } /* * Retry if the fast algorithm failed. */ if (strat == M_FIRSTFIT) { strat = M_BESTFIT; first = bt_freehead_toalloc(vm, size, strat); continue; } /* * XXX it is possible to fail to meet restrictions with the * imported region. It is up to the user to specify the * import quantum such that it can satisfy any allocation. */ if (vmem_import(vm, size, align, flags) == 0) continue; /* * Try to free some space from the quantum cache or reclaim * functions if available. */ if (vm->vm_qcache_max != 0 || vm->vm_reclaimfn != NULL) { avail = vm->vm_size - vm->vm_inuse; VMEM_UNLOCK(vm); if (vm->vm_qcache_max != 0) qc_drain(vm); if (vm->vm_reclaimfn != NULL) vm->vm_reclaimfn(vm, flags); VMEM_LOCK(vm); /* If we were successful retry even NOWAIT. */ if (vm->vm_size - vm->vm_inuse > avail) continue; } if ((flags & M_NOWAIT) != 0) { error = ENOMEM; break; } VMEM_CONDVAR_WAIT(vm); } out: VMEM_UNLOCK(vm); if (error != 0 && (flags & M_NOWAIT) == 0) panic("failed to allocate waiting allocation\n"); return (error); } /* * vmem_free: free the resource to the arena. */ void vmem_free(vmem_t *vm, vmem_addr_t addr, vmem_size_t size) { qcache_t *qc; MPASS(size > 0); if (size <= vm->vm_qcache_max) { qc = &vm->vm_qcache[(size - 1) >> vm->vm_quantum_shift]; uma_zfree(qc->qc_cache, (void *)addr); } else vmem_xfree(vm, addr, size); } void vmem_xfree(vmem_t *vm, vmem_addr_t addr, vmem_size_t size) { bt_t *bt; bt_t *t; MPASS(size > 0); VMEM_LOCK(vm); bt = bt_lookupbusy(vm, addr); MPASS(bt != NULL); MPASS(bt->bt_start == addr); MPASS(bt->bt_size == vmem_roundup_size(vm, size) || bt->bt_size - vmem_roundup_size(vm, size) <= vm->vm_quantum_mask); MPASS(bt->bt_type == BT_TYPE_BUSY); bt_rembusy(vm, bt); bt->bt_type = BT_TYPE_FREE; /* coalesce */ t = TAILQ_NEXT(bt, bt_seglist); if (t != NULL && t->bt_type == BT_TYPE_FREE) { MPASS(BT_END(bt) < t->bt_start); /* YYY */ bt->bt_size += t->bt_size; bt_remfree(vm, t); bt_remseg(vm, t); } t = TAILQ_PREV(bt, vmem_seglist, bt_seglist); if (t != NULL && t->bt_type == BT_TYPE_FREE) { MPASS(BT_END(t) < bt->bt_start); /* YYY */ bt->bt_size += t->bt_size; bt->bt_start = t->bt_start; bt_remfree(vm, t); bt_remseg(vm, t); } t = TAILQ_PREV(bt, vmem_seglist, bt_seglist); MPASS(t != NULL); MPASS(BT_ISSPAN_P(t) || t->bt_type == BT_TYPE_BUSY); if (vm->vm_releasefn != NULL && t->bt_type == BT_TYPE_SPAN && t->bt_size == bt->bt_size) { vmem_addr_t spanaddr; vmem_size_t spansize; MPASS(t->bt_start == bt->bt_start); spanaddr = bt->bt_start; spansize = bt->bt_size; bt_remseg(vm, bt); bt_remseg(vm, t); vm->vm_size -= spansize; VMEM_CONDVAR_BROADCAST(vm); bt_freetrim(vm, BT_MAXFREE); (*vm->vm_releasefn)(vm->vm_arg, spanaddr, spansize); } else { bt_insfree(vm, bt); VMEM_CONDVAR_BROADCAST(vm); bt_freetrim(vm, BT_MAXFREE); } } /* * vmem_add: * */ int vmem_add(vmem_t *vm, vmem_addr_t addr, vmem_size_t size, int flags) { int error; error = 0; flags &= VMEM_FLAGS; VMEM_LOCK(vm); if (vm->vm_nfreetags >= BT_MAXALLOC || bt_fill(vm, flags) == 0) vmem_add1(vm, addr, size, BT_TYPE_SPAN_STATIC); else error = ENOMEM; VMEM_UNLOCK(vm); return (error); } /* * vmem_size: information about arenas size */ vmem_size_t vmem_size(vmem_t *vm, int typemask) { int i; switch (typemask) { case VMEM_ALLOC: return vm->vm_inuse; case VMEM_FREE: return vm->vm_size - vm->vm_inuse; case VMEM_FREE|VMEM_ALLOC: return vm->vm_size; case VMEM_MAXFREE: VMEM_LOCK(vm); for (i = VMEM_MAXORDER - 1; i >= 0; i--) { if (LIST_EMPTY(&vm->vm_freelist[i])) continue; VMEM_UNLOCK(vm); return ((vmem_size_t)ORDER2SIZE(i) << vm->vm_quantum_shift); } VMEM_UNLOCK(vm); return (0); default: panic("vmem_size"); } } /* ---- debug */ #if defined(DDB) || defined(DIAGNOSTIC) static void bt_dump(const bt_t *, int (*)(const char *, ...) __printflike(1, 2)); static const char * bt_type_string(int type) { switch (type) { case BT_TYPE_BUSY: return "busy"; case BT_TYPE_FREE: return "free"; case BT_TYPE_SPAN: return "span"; case BT_TYPE_SPAN_STATIC: return "static span"; default: break; } return "BOGUS"; } static void bt_dump(const bt_t *bt, int (*pr)(const char *, ...)) { (*pr)("\t%p: %jx %jx, %d(%s)\n", bt, (intmax_t)bt->bt_start, (intmax_t)bt->bt_size, bt->bt_type, bt_type_string(bt->bt_type)); } static void vmem_dump(const vmem_t *vm , int (*pr)(const char *, ...) __printflike(1, 2)) { const bt_t *bt; int i; (*pr)("vmem %p '%s'\n", vm, vm->vm_name); TAILQ_FOREACH(bt, &vm->vm_seglist, bt_seglist) { bt_dump(bt, pr); } for (i = 0; i < VMEM_MAXORDER; i++) { const struct vmem_freelist *fl = &vm->vm_freelist[i]; if (LIST_EMPTY(fl)) { continue; } (*pr)("freelist[%d]\n", i); LIST_FOREACH(bt, fl, bt_freelist) { bt_dump(bt, pr); } } } #endif /* defined(DDB) || defined(DIAGNOSTIC) */ #if defined(DDB) #include static bt_t * vmem_whatis_lookup(vmem_t *vm, vmem_addr_t addr) { bt_t *bt; TAILQ_FOREACH(bt, &vm->vm_seglist, bt_seglist) { if (BT_ISSPAN_P(bt)) { continue; } if (bt->bt_start <= addr && addr <= BT_END(bt)) { return bt; } } return NULL; } void vmem_whatis(vmem_addr_t addr, int (*pr)(const char *, ...)) { vmem_t *vm; LIST_FOREACH(vm, &vmem_list, vm_alllist) { bt_t *bt; bt = vmem_whatis_lookup(vm, addr); if (bt == NULL) { continue; } (*pr)("%p is %p+%zu in VMEM '%s' (%s)\n", (void *)addr, (void *)bt->bt_start, (vmem_size_t)(addr - bt->bt_start), vm->vm_name, (bt->bt_type == BT_TYPE_BUSY) ? "allocated" : "free"); } } void vmem_printall(const char *modif, int (*pr)(const char *, ...)) { const vmem_t *vm; LIST_FOREACH(vm, &vmem_list, vm_alllist) { vmem_dump(vm, pr); } } void vmem_print(vmem_addr_t addr, const char *modif, int (*pr)(const char *, ...)) { const vmem_t *vm = (const void *)addr; vmem_dump(vm, pr); } DB_SHOW_COMMAND(vmemdump, vmemdump) { if (!have_addr) { db_printf("usage: show vmemdump \n"); return; } vmem_dump((const vmem_t *)addr, db_printf); } DB_SHOW_ALL_COMMAND(vmemdump, vmemdumpall) { const vmem_t *vm; LIST_FOREACH(vm, &vmem_list, vm_alllist) vmem_dump(vm, db_printf); } DB_SHOW_COMMAND(vmem, vmem_summ) { const vmem_t *vm = (const void *)addr; const bt_t *bt; size_t ft[VMEM_MAXORDER], ut[VMEM_MAXORDER]; size_t fs[VMEM_MAXORDER], us[VMEM_MAXORDER]; int ord; if (!have_addr) { db_printf("usage: show vmem \n"); return; } db_printf("vmem %p '%s'\n", vm, vm->vm_name); db_printf("\tquantum:\t%zu\n", vm->vm_quantum_mask + 1); db_printf("\tsize:\t%zu\n", vm->vm_size); db_printf("\tinuse:\t%zu\n", vm->vm_inuse); db_printf("\tfree:\t%zu\n", vm->vm_size - vm->vm_inuse); db_printf("\tbusy tags:\t%d\n", vm->vm_nbusytag); db_printf("\tfree tags:\t%d\n", vm->vm_nfreetags); memset(&ft, 0, sizeof(ft)); memset(&ut, 0, sizeof(ut)); memset(&fs, 0, sizeof(fs)); memset(&us, 0, sizeof(us)); TAILQ_FOREACH(bt, &vm->vm_seglist, bt_seglist) { ord = SIZE2ORDER(bt->bt_size >> vm->vm_quantum_shift); if (bt->bt_type == BT_TYPE_BUSY) { ut[ord]++; us[ord] += bt->bt_size; } else if (bt->bt_type == BT_TYPE_FREE) { ft[ord]++; fs[ord] += bt->bt_size; } } db_printf("\t\t\tinuse\tsize\t\tfree\tsize\n"); for (ord = 0; ord < VMEM_MAXORDER; ord++) { if (ut[ord] == 0 && ft[ord] == 0) continue; db_printf("\t%-15zu %zu\t%-15zu %zu\t%-16zu\n", ORDER2SIZE(ord) << vm->vm_quantum_shift, ut[ord], us[ord], ft[ord], fs[ord]); } } DB_SHOW_ALL_COMMAND(vmem, vmem_summall) { const vmem_t *vm; LIST_FOREACH(vm, &vmem_list, vm_alllist) vmem_summ((db_expr_t)vm, TRUE, count, modif); } #endif /* defined(DDB) */ #define vmem_printf printf #if defined(DIAGNOSTIC) static bool vmem_check_sanity(vmem_t *vm) { const bt_t *bt, *bt2; MPASS(vm != NULL); TAILQ_FOREACH(bt, &vm->vm_seglist, bt_seglist) { if (bt->bt_start > BT_END(bt)) { printf("corrupted tag\n"); bt_dump(bt, vmem_printf); return false; } } TAILQ_FOREACH(bt, &vm->vm_seglist, bt_seglist) { TAILQ_FOREACH(bt2, &vm->vm_seglist, bt_seglist) { if (bt == bt2) { continue; } if (BT_ISSPAN_P(bt) != BT_ISSPAN_P(bt2)) { continue; } if (bt->bt_start <= BT_END(bt2) && bt2->bt_start <= BT_END(bt)) { printf("overwrapped tags\n"); bt_dump(bt, vmem_printf); bt_dump(bt2, vmem_printf); return false; } } } return true; } static void vmem_check(vmem_t *vm) { if (!vmem_check_sanity(vm)) { panic("insanity vmem %p", vm); } } #endif /* defined(DIAGNOSTIC) */ Index: stable/11/sys/kern/vfs_bio.c =================================================================== --- stable/11/sys/kern/vfs_bio.c (revision 327403) +++ stable/11/sys/kern/vfs_bio.c (revision 327404) @@ -1,5001 +1,5001 @@ /*- * Copyright (c) 2004 Poul-Henning Kamp * Copyright (c) 1994,1997 John S. Dyson * Copyright (c) 2013 The FreeBSD Foundation * All rights reserved. * * 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. * * 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 file contains a new buffer I/O scheme implementing a coherent * VM object and buffer cache scheme. Pains have been taken to make * sure that the performance degradation associated with schemes such * as this is not realized. * * Author: John S. Dyson * Significant help during the development and debugging phases * had been provided by David Greenman, also of the FreeBSD core team. * * see man buf(9) for more info. */ #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 #include #include #include #include #include #include #include #include #include #include "opt_compat.h" #include "opt_swap.h" static MALLOC_DEFINE(M_BIOBUF, "biobuf", "BIO buffer"); struct bio_ops bioops; /* I/O operation notification */ struct buf_ops buf_ops_bio = { .bop_name = "buf_ops_bio", .bop_write = bufwrite, .bop_strategy = bufstrategy, .bop_sync = bufsync, .bop_bdflush = bufbdflush, }; static struct buf *buf; /* buffer header pool */ extern struct buf *swbuf; /* Swap buffer header pool. */ caddr_t unmapped_buf; /* Used below and for softdep flushing threads in ufs/ffs/ffs_softdep.c */ struct proc *bufdaemonproc; struct proc *bufspacedaemonproc; static int inmem(struct vnode *vp, daddr_t blkno); static void vm_hold_free_pages(struct buf *bp, int newbsize); static void vm_hold_load_pages(struct buf *bp, vm_offset_t from, vm_offset_t to); static void vfs_page_set_valid(struct buf *bp, vm_ooffset_t off, vm_page_t m); static void vfs_page_set_validclean(struct buf *bp, vm_ooffset_t off, vm_page_t m); static void vfs_clean_pages_dirty_buf(struct buf *bp); static void vfs_setdirty_locked_object(struct buf *bp); static void vfs_vmio_invalidate(struct buf *bp); static void vfs_vmio_truncate(struct buf *bp, int npages); static void vfs_vmio_extend(struct buf *bp, int npages, int size); static int vfs_bio_clcheck(struct vnode *vp, int size, daddr_t lblkno, daddr_t blkno); static int buf_flush(struct vnode *vp, int); static int buf_recycle(bool); static int buf_scan(bool); static int flushbufqueues(struct vnode *, int, int); static void buf_daemon(void); static void bremfreel(struct buf *bp); static __inline void bd_wakeup(void); static int sysctl_runningspace(SYSCTL_HANDLER_ARGS); static void bufkva_reclaim(vmem_t *, int); static void bufkva_free(struct buf *); static int buf_import(void *, void **, int, int); static void buf_release(void *, void **, int); static void maxbcachebuf_adjust(void); #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) static int sysctl_bufspace(SYSCTL_HANDLER_ARGS); #endif int vmiodirenable = TRUE; SYSCTL_INT(_vfs, OID_AUTO, vmiodirenable, CTLFLAG_RW, &vmiodirenable, 0, "Use the VM system for directory writes"); long runningbufspace; SYSCTL_LONG(_vfs, OID_AUTO, runningbufspace, CTLFLAG_RD, &runningbufspace, 0, "Amount of presently outstanding async buffer io"); static long bufspace; #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) SYSCTL_PROC(_vfs, OID_AUTO, bufspace, CTLTYPE_LONG|CTLFLAG_MPSAFE|CTLFLAG_RD, &bufspace, 0, sysctl_bufspace, "L", "Virtual memory used for buffers"); #else SYSCTL_LONG(_vfs, OID_AUTO, bufspace, CTLFLAG_RD, &bufspace, 0, "Physical memory used for buffers"); #endif static long bufkvaspace; SYSCTL_LONG(_vfs, OID_AUTO, bufkvaspace, CTLFLAG_RD, &bufkvaspace, 0, "Kernel virtual memory used for buffers"); static long maxbufspace; SYSCTL_LONG(_vfs, OID_AUTO, maxbufspace, CTLFLAG_RW, &maxbufspace, 0, "Maximum allowed value of bufspace (including metadata)"); static long bufmallocspace; SYSCTL_LONG(_vfs, OID_AUTO, bufmallocspace, CTLFLAG_RD, &bufmallocspace, 0, "Amount of malloced memory for buffers"); static long maxbufmallocspace; SYSCTL_LONG(_vfs, OID_AUTO, maxmallocbufspace, CTLFLAG_RW, &maxbufmallocspace, 0, "Maximum amount of malloced memory for buffers"); static long lobufspace; SYSCTL_LONG(_vfs, OID_AUTO, lobufspace, CTLFLAG_RW, &lobufspace, 0, "Minimum amount of buffers we want to have"); long hibufspace; SYSCTL_LONG(_vfs, OID_AUTO, hibufspace, CTLFLAG_RW, &hibufspace, 0, "Maximum allowed value of bufspace (excluding metadata)"); long bufspacethresh; SYSCTL_LONG(_vfs, OID_AUTO, bufspacethresh, CTLFLAG_RW, &bufspacethresh, 0, "Bufspace consumed before waking the daemon to free some"); static int buffreekvacnt; SYSCTL_INT(_vfs, OID_AUTO, buffreekvacnt, CTLFLAG_RW, &buffreekvacnt, 0, "Number of times we have freed the KVA space from some buffer"); static int bufdefragcnt; SYSCTL_INT(_vfs, OID_AUTO, bufdefragcnt, CTLFLAG_RW, &bufdefragcnt, 0, "Number of times we have had to repeat buffer allocation to defragment"); static long lorunningspace; SYSCTL_PROC(_vfs, OID_AUTO, lorunningspace, CTLTYPE_LONG | CTLFLAG_MPSAFE | CTLFLAG_RW, &lorunningspace, 0, sysctl_runningspace, "L", "Minimum preferred space used for in-progress I/O"); static long hirunningspace; SYSCTL_PROC(_vfs, OID_AUTO, hirunningspace, CTLTYPE_LONG | CTLFLAG_MPSAFE | CTLFLAG_RW, &hirunningspace, 0, sysctl_runningspace, "L", "Maximum amount of space to use for in-progress I/O"); int dirtybufferflushes; SYSCTL_INT(_vfs, OID_AUTO, dirtybufferflushes, CTLFLAG_RW, &dirtybufferflushes, 0, "Number of bdwrite to bawrite conversions to limit dirty buffers"); int bdwriteskip; SYSCTL_INT(_vfs, OID_AUTO, bdwriteskip, CTLFLAG_RW, &bdwriteskip, 0, "Number of buffers supplied to bdwrite with snapshot deadlock risk"); int altbufferflushes; SYSCTL_INT(_vfs, OID_AUTO, altbufferflushes, CTLFLAG_RW, &altbufferflushes, 0, "Number of fsync flushes to limit dirty buffers"); static int recursiveflushes; SYSCTL_INT(_vfs, OID_AUTO, recursiveflushes, CTLFLAG_RW, &recursiveflushes, 0, "Number of flushes skipped due to being recursive"); static int numdirtybuffers; SYSCTL_INT(_vfs, OID_AUTO, numdirtybuffers, CTLFLAG_RD, &numdirtybuffers, 0, "Number of buffers that are dirty (has unwritten changes) at the moment"); static int lodirtybuffers; SYSCTL_INT(_vfs, OID_AUTO, lodirtybuffers, CTLFLAG_RW, &lodirtybuffers, 0, "How many buffers we want to have free before bufdaemon can sleep"); static int hidirtybuffers; SYSCTL_INT(_vfs, OID_AUTO, hidirtybuffers, CTLFLAG_RW, &hidirtybuffers, 0, "When the number of dirty buffers is considered severe"); int dirtybufthresh; SYSCTL_INT(_vfs, OID_AUTO, dirtybufthresh, CTLFLAG_RW, &dirtybufthresh, 0, "Number of bdwrite to bawrite conversions to clear dirty buffers"); static int numfreebuffers; SYSCTL_INT(_vfs, OID_AUTO, numfreebuffers, CTLFLAG_RD, &numfreebuffers, 0, "Number of free buffers"); static int lofreebuffers; SYSCTL_INT(_vfs, OID_AUTO, lofreebuffers, CTLFLAG_RW, &lofreebuffers, 0, "Target number of free buffers"); static int hifreebuffers; SYSCTL_INT(_vfs, OID_AUTO, hifreebuffers, CTLFLAG_RW, &hifreebuffers, 0, "Threshold for clean buffer recycling"); static int getnewbufcalls; SYSCTL_INT(_vfs, OID_AUTO, getnewbufcalls, CTLFLAG_RW, &getnewbufcalls, 0, "Number of calls to getnewbuf"); static int getnewbufrestarts; SYSCTL_INT(_vfs, OID_AUTO, getnewbufrestarts, CTLFLAG_RW, &getnewbufrestarts, 0, "Number of times getnewbuf has had to restart a buffer acquisition"); static int mappingrestarts; SYSCTL_INT(_vfs, OID_AUTO, mappingrestarts, CTLFLAG_RW, &mappingrestarts, 0, "Number of times getblk has had to restart a buffer mapping for " "unmapped buffer"); static int numbufallocfails; SYSCTL_INT(_vfs, OID_AUTO, numbufallocfails, CTLFLAG_RW, &numbufallocfails, 0, "Number of times buffer allocations failed"); static int flushbufqtarget = 100; SYSCTL_INT(_vfs, OID_AUTO, flushbufqtarget, CTLFLAG_RW, &flushbufqtarget, 0, "Amount of work to do in flushbufqueues when helping bufdaemon"); static long notbufdflushes; SYSCTL_LONG(_vfs, OID_AUTO, notbufdflushes, CTLFLAG_RD, ¬bufdflushes, 0, "Number of dirty buffer flushes done by the bufdaemon helpers"); static long barrierwrites; SYSCTL_LONG(_vfs, OID_AUTO, barrierwrites, CTLFLAG_RW, &barrierwrites, 0, "Number of barrier writes"); SYSCTL_INT(_vfs, OID_AUTO, unmapped_buf_allowed, CTLFLAG_RD, &unmapped_buf_allowed, 0, "Permit the use of the unmapped i/o"); int maxbcachebuf = MAXBCACHEBUF; SYSCTL_INT(_vfs, OID_AUTO, maxbcachebuf, CTLFLAG_RDTUN, &maxbcachebuf, 0, "Maximum size of a buffer cache block"); /* * This lock synchronizes access to bd_request. */ -static struct mtx_padalign bdlock; +static struct mtx_padalign __exclusive_cache_line bdlock; /* * This lock protects the runningbufreq and synchronizes runningbufwakeup and * waitrunningbufspace(). */ -static struct mtx_padalign rbreqlock; +static struct mtx_padalign __exclusive_cache_line rbreqlock; /* * Lock that protects needsbuffer and the sleeps/wakeups surrounding it. */ -static struct rwlock_padalign nblock; +static struct rwlock_padalign __exclusive_cache_line nblock; /* * Lock that protects bdirtywait. */ -static struct mtx_padalign bdirtylock; +static struct mtx_padalign __exclusive_cache_line bdirtylock; /* * Wakeup point for bufdaemon, as well as indicator of whether it is already * active. Set to 1 when the bufdaemon is already "on" the queue, 0 when it * is idling. */ static int bd_request; /* * Request/wakeup point for the bufspace daemon. */ static int bufspace_request; /* * Request for the buf daemon to write more buffers than is indicated by * lodirtybuf. This may be necessary to push out excess dependencies or * defragment the address space where a simple count of the number of dirty * buffers is insufficient to characterize the demand for flushing them. */ static int bd_speedupreq; /* * bogus page -- for I/O to/from partially complete buffers * this is a temporary solution to the problem, but it is not * really that bad. it would be better to split the buffer * for input in the case of buffers partially already in memory, * but the code is intricate enough already. */ vm_page_t bogus_page; /* * Synchronization (sleep/wakeup) variable for active buffer space requests. * Set when wait starts, cleared prior to wakeup(). * Used in runningbufwakeup() and waitrunningbufspace(). */ static int runningbufreq; /* * Synchronization (sleep/wakeup) variable for buffer requests. * Can contain the VFS_BIO_NEED flags defined below; setting/clearing is done * by and/or. * Used in numdirtywakeup(), bufspace_wakeup(), bwillwrite(), * getnewbuf(), and getblk(). */ static volatile int needsbuffer; /* * Synchronization for bwillwrite() waiters. */ static int bdirtywait; /* * Definitions for the buffer free lists. */ #define QUEUE_NONE 0 /* on no queue */ #define QUEUE_EMPTY 1 /* empty buffer headers */ #define QUEUE_DIRTY 2 /* B_DELWRI buffers */ #define QUEUE_CLEAN 3 /* non-B_DELWRI buffers */ #define QUEUE_SENTINEL 1024 /* not an queue index, but mark for sentinel */ /* Maximum number of clean buffer queues. */ #define CLEAN_QUEUES 16 /* Configured number of clean queues. */ static int clean_queues; /* Maximum number of buffer queues. */ #define BUFFER_QUEUES (QUEUE_CLEAN + CLEAN_QUEUES) /* Queues for free buffers with various properties */ static TAILQ_HEAD(bqueues, buf) bufqueues[BUFFER_QUEUES] = { { 0 } }; #ifdef INVARIANTS static int bq_len[BUFFER_QUEUES]; #endif /* * Lock for each bufqueue */ -static struct mtx_padalign bqlocks[BUFFER_QUEUES]; +static struct mtx_padalign __exclusive_cache_line bqlocks[BUFFER_QUEUES]; /* * per-cpu empty buffer cache. */ uma_zone_t buf_zone; /* * Single global constant for BUF_WMESG, to avoid getting multiple references. * buf_wmesg is referred from macros. */ const char *buf_wmesg = BUF_WMESG; static int sysctl_runningspace(SYSCTL_HANDLER_ARGS) { long value; int error; value = *(long *)arg1; error = sysctl_handle_long(oidp, &value, 0, req); if (error != 0 || req->newptr == NULL) return (error); mtx_lock(&rbreqlock); if (arg1 == &hirunningspace) { if (value < lorunningspace) error = EINVAL; else hirunningspace = value; } else { KASSERT(arg1 == &lorunningspace, ("%s: unknown arg1", __func__)); if (value > hirunningspace) error = EINVAL; else lorunningspace = value; } mtx_unlock(&rbreqlock); return (error); } #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) static int sysctl_bufspace(SYSCTL_HANDLER_ARGS) { long lvalue; int ivalue; if (sizeof(int) == sizeof(long) || req->oldlen >= sizeof(long)) return (sysctl_handle_long(oidp, arg1, arg2, req)); lvalue = *(long *)arg1; if (lvalue > INT_MAX) /* On overflow, still write out a long to trigger ENOMEM. */ return (sysctl_handle_long(oidp, &lvalue, 0, req)); ivalue = lvalue; return (sysctl_handle_int(oidp, &ivalue, 0, req)); } #endif static int bqcleanq(void) { static int nextq; return ((atomic_fetchadd_int(&nextq, 1) % clean_queues) + QUEUE_CLEAN); } static int bqisclean(int qindex) { return (qindex >= QUEUE_CLEAN && qindex < QUEUE_CLEAN + CLEAN_QUEUES); } /* * bqlock: * * Return the appropriate queue lock based on the index. */ static inline struct mtx * bqlock(int qindex) { return (struct mtx *)&bqlocks[qindex]; } /* * bdirtywakeup: * * Wakeup any bwillwrite() waiters. */ static void bdirtywakeup(void) { mtx_lock(&bdirtylock); if (bdirtywait) { bdirtywait = 0; wakeup(&bdirtywait); } mtx_unlock(&bdirtylock); } /* * bdirtysub: * * Decrement the numdirtybuffers count by one and wakeup any * threads blocked in bwillwrite(). */ static void bdirtysub(void) { if (atomic_fetchadd_int(&numdirtybuffers, -1) == (lodirtybuffers + hidirtybuffers) / 2) bdirtywakeup(); } /* * bdirtyadd: * * Increment the numdirtybuffers count by one and wakeup the buf * daemon if needed. */ static void bdirtyadd(void) { /* * Only do the wakeup once as we cross the boundary. The * buf daemon will keep running until the condition clears. */ if (atomic_fetchadd_int(&numdirtybuffers, 1) == (lodirtybuffers + hidirtybuffers) / 2) bd_wakeup(); } /* * bufspace_wakeup: * * Called when buffer space is potentially available for recovery. * getnewbuf() will block on this flag when it is unable to free * sufficient buffer space. Buffer space becomes recoverable when * bp's get placed back in the queues. */ static void bufspace_wakeup(void) { /* * If someone is waiting for bufspace, wake them up. * * Since needsbuffer is set prior to doing an additional queue * scan it is safe to check for the flag prior to acquiring the * lock. The thread that is preparing to scan again before * blocking would discover the buf we released. */ if (needsbuffer) { rw_rlock(&nblock); if (atomic_cmpset_int(&needsbuffer, 1, 0) == 1) wakeup(__DEVOLATILE(void *, &needsbuffer)); rw_runlock(&nblock); } } /* * bufspace_daemonwakeup: * * Wakeup the daemon responsible for freeing clean bufs. */ static void bufspace_daemonwakeup(void) { rw_rlock(&nblock); if (bufspace_request == 0) { bufspace_request = 1; wakeup(&bufspace_request); } rw_runlock(&nblock); } /* * bufspace_adjust: * * Adjust the reported bufspace for a KVA managed buffer, possibly * waking any waiters. */ static void bufspace_adjust(struct buf *bp, int bufsize) { long space; int diff; KASSERT((bp->b_flags & B_MALLOC) == 0, ("bufspace_adjust: malloc buf %p", bp)); diff = bufsize - bp->b_bufsize; if (diff < 0) { atomic_subtract_long(&bufspace, -diff); bufspace_wakeup(); } else { space = atomic_fetchadd_long(&bufspace, diff); /* Wake up the daemon on the transition. */ if (space < bufspacethresh && space + diff >= bufspacethresh) bufspace_daemonwakeup(); } bp->b_bufsize = bufsize; } /* * bufspace_reserve: * * Reserve bufspace before calling allocbuf(). metadata has a * different space limit than data. */ static int bufspace_reserve(int size, bool metadata) { long limit; long space; if (metadata) limit = maxbufspace; else limit = hibufspace; do { space = bufspace; if (space + size > limit) return (ENOSPC); } while (atomic_cmpset_long(&bufspace, space, space + size) == 0); /* Wake up the daemon on the transition. */ if (space < bufspacethresh && space + size >= bufspacethresh) bufspace_daemonwakeup(); return (0); } /* * bufspace_release: * * Release reserved bufspace after bufspace_adjust() has consumed it. */ static void bufspace_release(int size) { atomic_subtract_long(&bufspace, size); bufspace_wakeup(); } /* * bufspace_wait: * * Wait for bufspace, acting as the buf daemon if a locked vnode is * supplied. needsbuffer must be set in a safe fashion prior to * polling for space. The operation must be re-tried on return. */ static void bufspace_wait(struct vnode *vp, int gbflags, int slpflag, int slptimeo) { struct thread *td; int error, fl, norunbuf; if ((gbflags & GB_NOWAIT_BD) != 0) return; td = curthread; rw_wlock(&nblock); while (needsbuffer != 0) { if (vp != NULL && vp->v_type != VCHR && (td->td_pflags & TDP_BUFNEED) == 0) { rw_wunlock(&nblock); /* * getblk() is called with a vnode locked, and * some majority of the dirty buffers may as * well belong to the vnode. Flushing the * buffers there would make a progress that * cannot be achieved by the buf_daemon, that * cannot lock the vnode. */ norunbuf = ~(TDP_BUFNEED | TDP_NORUNNINGBUF) | (td->td_pflags & TDP_NORUNNINGBUF); /* * Play bufdaemon. The getnewbuf() function * may be called while the thread owns lock * for another dirty buffer for the same * vnode, which makes it impossible to use * VOP_FSYNC() there, due to the buffer lock * recursion. */ td->td_pflags |= TDP_BUFNEED | TDP_NORUNNINGBUF; fl = buf_flush(vp, flushbufqtarget); td->td_pflags &= norunbuf; rw_wlock(&nblock); if (fl != 0) continue; if (needsbuffer == 0) break; } error = rw_sleep(__DEVOLATILE(void *, &needsbuffer), &nblock, (PRIBIO + 4) | slpflag, "newbuf", slptimeo); if (error != 0) break; } rw_wunlock(&nblock); } /* * bufspace_daemon: * * buffer space management daemon. Tries to maintain some marginal * amount of free buffer space so that requesting processes neither * block nor work to reclaim buffers. */ static void bufspace_daemon(void) { for (;;) { kproc_suspend_check(bufspacedaemonproc); /* * Free buffers from the clean queue until we meet our * targets. * * Theory of operation: The buffer cache is most efficient * when some free buffer headers and space are always * available to getnewbuf(). This daemon attempts to prevent * the excessive blocking and synchronization associated * with shortfall. It goes through three phases according * demand: * * 1) The daemon wakes up voluntarily once per-second * during idle periods when the counters are below * the wakeup thresholds (bufspacethresh, lofreebuffers). * * 2) The daemon wakes up as we cross the thresholds * ahead of any potential blocking. This may bounce * slightly according to the rate of consumption and * release. * * 3) The daemon and consumers are starved for working * clean buffers. This is the 'bufspace' sleep below * which will inefficiently trade bufs with bqrelse * until we return to condition 2. */ while (bufspace > lobufspace || numfreebuffers < hifreebuffers) { if (buf_recycle(false) != 0) { atomic_set_int(&needsbuffer, 1); if (buf_recycle(false) != 0) { rw_wlock(&nblock); if (needsbuffer) rw_sleep(__DEVOLATILE(void *, &needsbuffer), &nblock, PRIBIO|PDROP, "bufspace", hz/10); else rw_wunlock(&nblock); } } maybe_yield(); } /* * Re-check our limits under the exclusive nblock. */ rw_wlock(&nblock); if (bufspace < bufspacethresh && numfreebuffers > lofreebuffers) { bufspace_request = 0; rw_sleep(&bufspace_request, &nblock, PRIBIO|PDROP, "-", hz); } else rw_wunlock(&nblock); } } static struct kproc_desc bufspace_kp = { "bufspacedaemon", bufspace_daemon, &bufspacedaemonproc }; SYSINIT(bufspacedaemon, SI_SUB_KTHREAD_BUF, SI_ORDER_FIRST, kproc_start, &bufspace_kp); /* * bufmallocadjust: * * Adjust the reported bufspace for a malloc managed buffer, possibly * waking any waiters. */ static void bufmallocadjust(struct buf *bp, int bufsize) { int diff; KASSERT((bp->b_flags & B_MALLOC) != 0, ("bufmallocadjust: non-malloc buf %p", bp)); diff = bufsize - bp->b_bufsize; if (diff < 0) atomic_subtract_long(&bufmallocspace, -diff); else atomic_add_long(&bufmallocspace, diff); bp->b_bufsize = bufsize; } /* * runningwakeup: * * Wake up processes that are waiting on asynchronous writes to fall * below lorunningspace. */ static void runningwakeup(void) { mtx_lock(&rbreqlock); if (runningbufreq) { runningbufreq = 0; wakeup(&runningbufreq); } mtx_unlock(&rbreqlock); } /* * runningbufwakeup: * * Decrement the outstanding write count according. */ void runningbufwakeup(struct buf *bp) { long space, bspace; bspace = bp->b_runningbufspace; if (bspace == 0) return; space = atomic_fetchadd_long(&runningbufspace, -bspace); KASSERT(space >= bspace, ("runningbufspace underflow %ld %ld", space, bspace)); bp->b_runningbufspace = 0; /* * Only acquire the lock and wakeup on the transition from exceeding * the threshold to falling below it. */ if (space < lorunningspace) return; if (space - bspace > lorunningspace) return; runningwakeup(); } /* * waitrunningbufspace() * * runningbufspace is a measure of the amount of I/O currently * running. This routine is used in async-write situations to * prevent creating huge backups of pending writes to a device. * Only asynchronous writes are governed by this function. * * This does NOT turn an async write into a sync write. It waits * for earlier writes to complete and generally returns before the * caller's write has reached the device. */ void waitrunningbufspace(void) { mtx_lock(&rbreqlock); while (runningbufspace > hirunningspace) { runningbufreq = 1; msleep(&runningbufreq, &rbreqlock, PVM, "wdrain", 0); } mtx_unlock(&rbreqlock); } /* * vfs_buf_test_cache: * * Called when a buffer is extended. This function clears the B_CACHE * bit if the newly extended portion of the buffer does not contain * valid data. */ static __inline void vfs_buf_test_cache(struct buf *bp, vm_ooffset_t foff, vm_offset_t off, vm_offset_t size, vm_page_t m) { VM_OBJECT_ASSERT_LOCKED(m->object); if (bp->b_flags & B_CACHE) { int base = (foff + off) & PAGE_MASK; if (vm_page_is_valid(m, base, size) == 0) bp->b_flags &= ~B_CACHE; } } /* Wake up the buffer daemon if necessary */ static __inline void bd_wakeup(void) { mtx_lock(&bdlock); if (bd_request == 0) { bd_request = 1; wakeup(&bd_request); } mtx_unlock(&bdlock); } /* * Adjust the maxbcachbuf tunable. */ static void maxbcachebuf_adjust(void) { int i; /* * maxbcachebuf must be a power of 2 >= MAXBSIZE. */ i = 2; while (i * 2 <= maxbcachebuf) i *= 2; maxbcachebuf = i; if (maxbcachebuf < MAXBSIZE) maxbcachebuf = MAXBSIZE; if (maxbcachebuf > MAXPHYS) maxbcachebuf = MAXPHYS; if (bootverbose != 0 && maxbcachebuf != MAXBCACHEBUF) printf("maxbcachebuf=%d\n", maxbcachebuf); } /* * bd_speedup - speedup the buffer cache flushing code */ void bd_speedup(void) { int needwake; mtx_lock(&bdlock); needwake = 0; if (bd_speedupreq == 0 || bd_request == 0) needwake = 1; bd_speedupreq = 1; bd_request = 1; if (needwake) wakeup(&bd_request); mtx_unlock(&bdlock); } #ifndef NSWBUF_MIN #define NSWBUF_MIN 16 #endif #ifdef __i386__ #define TRANSIENT_DENOM 5 #else #define TRANSIENT_DENOM 10 #endif /* * Calculating buffer cache scaling values and reserve space for buffer * headers. This is called during low level kernel initialization and * may be called more then once. We CANNOT write to the memory area * being reserved at this time. */ caddr_t kern_vfs_bio_buffer_alloc(caddr_t v, long physmem_est) { int tuned_nbuf; long maxbuf, maxbuf_sz, buf_sz, biotmap_sz; /* * physmem_est is in pages. Convert it to kilobytes (assumes * PAGE_SIZE is >= 1K) */ physmem_est = physmem_est * (PAGE_SIZE / 1024); maxbcachebuf_adjust(); /* * The nominal buffer size (and minimum KVA allocation) is BKVASIZE. * For the first 64MB of ram nominally allocate sufficient buffers to * cover 1/4 of our ram. Beyond the first 64MB allocate additional * buffers to cover 1/10 of our ram over 64MB. When auto-sizing * the buffer cache we limit the eventual kva reservation to * maxbcache bytes. * * factor represents the 1/4 x ram conversion. */ if (nbuf == 0) { int factor = 4 * BKVASIZE / 1024; nbuf = 50; if (physmem_est > 4096) nbuf += min((physmem_est - 4096) / factor, 65536 / factor); if (physmem_est > 65536) nbuf += min((physmem_est - 65536) * 2 / (factor * 5), 32 * 1024 * 1024 / (factor * 5)); if (maxbcache && nbuf > maxbcache / BKVASIZE) nbuf = maxbcache / BKVASIZE; tuned_nbuf = 1; } else tuned_nbuf = 0; /* XXX Avoid unsigned long overflows later on with maxbufspace. */ maxbuf = (LONG_MAX / 3) / BKVASIZE; if (nbuf > maxbuf) { if (!tuned_nbuf) printf("Warning: nbufs lowered from %d to %ld\n", nbuf, maxbuf); nbuf = maxbuf; } /* * Ideal allocation size for the transient bio submap is 10% * of the maximal space buffer map. This roughly corresponds * to the amount of the buffer mapped for typical UFS load. * * Clip the buffer map to reserve space for the transient * BIOs, if its extent is bigger than 90% (80% on i386) of the * maximum buffer map extent on the platform. * * The fall-back to the maxbuf in case of maxbcache unset, * allows to not trim the buffer KVA for the architectures * with ample KVA space. */ if (bio_transient_maxcnt == 0 && unmapped_buf_allowed) { maxbuf_sz = maxbcache != 0 ? maxbcache : maxbuf * BKVASIZE; buf_sz = (long)nbuf * BKVASIZE; if (buf_sz < maxbuf_sz / TRANSIENT_DENOM * (TRANSIENT_DENOM - 1)) { /* * There is more KVA than memory. Do not * adjust buffer map size, and assign the rest * of maxbuf to transient map. */ biotmap_sz = maxbuf_sz - buf_sz; } else { /* * Buffer map spans all KVA we could afford on * this platform. Give 10% (20% on i386) of * the buffer map to the transient bio map. */ biotmap_sz = buf_sz / TRANSIENT_DENOM; buf_sz -= biotmap_sz; } if (biotmap_sz / INT_MAX > MAXPHYS) bio_transient_maxcnt = INT_MAX; else bio_transient_maxcnt = biotmap_sz / MAXPHYS; /* * Artificially limit to 1024 simultaneous in-flight I/Os * using the transient mapping. */ if (bio_transient_maxcnt > 1024) bio_transient_maxcnt = 1024; if (tuned_nbuf) nbuf = buf_sz / BKVASIZE; } /* * swbufs are used as temporary holders for I/O, such as paging I/O. * We have no less then 16 and no more then 256. */ nswbuf = min(nbuf / 4, 256); TUNABLE_INT_FETCH("kern.nswbuf", &nswbuf); if (nswbuf < NSWBUF_MIN) nswbuf = NSWBUF_MIN; /* * Reserve space for the buffer cache buffers */ swbuf = (void *)v; v = (caddr_t)(swbuf + nswbuf); buf = (void *)v; v = (caddr_t)(buf + nbuf); return(v); } /* Initialize the buffer subsystem. Called before use of any buffers. */ void bufinit(void) { struct buf *bp; int i; KASSERT(maxbcachebuf >= MAXBSIZE, ("maxbcachebuf (%d) must be >= MAXBSIZE (%d)\n", maxbcachebuf, MAXBSIZE)); mtx_init(&bqlocks[QUEUE_DIRTY], "bufq dirty lock", NULL, MTX_DEF); mtx_init(&bqlocks[QUEUE_EMPTY], "bufq empty lock", NULL, MTX_DEF); for (i = QUEUE_CLEAN; i < QUEUE_CLEAN + CLEAN_QUEUES; i++) mtx_init(&bqlocks[i], "bufq clean lock", NULL, MTX_DEF); mtx_init(&rbreqlock, "runningbufspace lock", NULL, MTX_DEF); rw_init(&nblock, "needsbuffer lock"); mtx_init(&bdlock, "buffer daemon lock", NULL, MTX_DEF); mtx_init(&bdirtylock, "dirty buf lock", NULL, MTX_DEF); /* next, make a null set of free lists */ for (i = 0; i < BUFFER_QUEUES; i++) TAILQ_INIT(&bufqueues[i]); unmapped_buf = (caddr_t)kva_alloc(MAXPHYS); /* finally, initialize each buffer header and stick on empty q */ for (i = 0; i < nbuf; i++) { bp = &buf[i]; bzero(bp, sizeof *bp); bp->b_flags = B_INVAL; bp->b_rcred = NOCRED; bp->b_wcred = NOCRED; bp->b_qindex = QUEUE_EMPTY; bp->b_xflags = 0; bp->b_data = bp->b_kvabase = unmapped_buf; LIST_INIT(&bp->b_dep); BUF_LOCKINIT(bp); TAILQ_INSERT_TAIL(&bufqueues[QUEUE_EMPTY], bp, b_freelist); #ifdef INVARIANTS bq_len[QUEUE_EMPTY]++; #endif } /* * maxbufspace is the absolute maximum amount of buffer space we are * allowed to reserve in KVM and in real terms. The absolute maximum * is nominally used by metadata. hibufspace is the nominal maximum * used by most other requests. The differential is required to * ensure that metadata deadlocks don't occur. * * maxbufspace is based on BKVASIZE. Allocating buffers larger then * this may result in KVM fragmentation which is not handled optimally * by the system. XXX This is less true with vmem. We could use * PAGE_SIZE. */ maxbufspace = (long)nbuf * BKVASIZE; hibufspace = lmax(3 * maxbufspace / 4, maxbufspace - maxbcachebuf * 10); lobufspace = (hibufspace / 20) * 19; /* 95% */ bufspacethresh = lobufspace + (hibufspace - lobufspace) / 2; /* * Note: The 16 MiB upper limit for hirunningspace was chosen * arbitrarily and may need further tuning. It corresponds to * 128 outstanding write IO requests (if IO size is 128 KiB), * which fits with many RAID controllers' tagged queuing limits. * The lower 1 MiB limit is the historical upper limit for * hirunningspace. */ hirunningspace = lmax(lmin(roundup(hibufspace / 64, maxbcachebuf), 16 * 1024 * 1024), 1024 * 1024); lorunningspace = roundup((hirunningspace * 2) / 3, maxbcachebuf); /* * Limit the amount of malloc memory since it is wired permanently into * the kernel space. Even though this is accounted for in the buffer * allocation, we don't want the malloced region to grow uncontrolled. * The malloc scheme improves memory utilization significantly on * average (small) directories. */ maxbufmallocspace = hibufspace / 20; /* * Reduce the chance of a deadlock occurring by limiting the number * of delayed-write dirty buffers we allow to stack up. */ hidirtybuffers = nbuf / 4 + 20; dirtybufthresh = hidirtybuffers * 9 / 10; numdirtybuffers = 0; /* * To support extreme low-memory systems, make sure hidirtybuffers * cannot eat up all available buffer space. This occurs when our * minimum cannot be met. We try to size hidirtybuffers to 3/4 our * buffer space assuming BKVASIZE'd buffers. */ while ((long)hidirtybuffers * BKVASIZE > 3 * hibufspace / 4) { hidirtybuffers >>= 1; } lodirtybuffers = hidirtybuffers / 2; /* * lofreebuffers should be sufficient to avoid stalling waiting on * buf headers under heavy utilization. The bufs in per-cpu caches * are counted as free but will be unavailable to threads executing * on other cpus. * * hifreebuffers is the free target for the bufspace daemon. This * should be set appropriately to limit work per-iteration. */ lofreebuffers = MIN((nbuf / 25) + (20 * mp_ncpus), 128 * mp_ncpus); hifreebuffers = (3 * lofreebuffers) / 2; numfreebuffers = nbuf; bogus_page = vm_page_alloc(NULL, 0, VM_ALLOC_NOOBJ | VM_ALLOC_NORMAL | VM_ALLOC_WIRED); /* Setup the kva and free list allocators. */ vmem_set_reclaim(buffer_arena, bufkva_reclaim); buf_zone = uma_zcache_create("buf free cache", sizeof(struct buf), NULL, NULL, NULL, NULL, buf_import, buf_release, NULL, 0); /* * Size the clean queue according to the amount of buffer space. * One queue per-256mb up to the max. More queues gives better * concurrency but less accurate LRU. */ clean_queues = MIN(howmany(maxbufspace, 256*1024*1024), CLEAN_QUEUES); } #ifdef INVARIANTS static inline void vfs_buf_check_mapped(struct buf *bp) { KASSERT(bp->b_kvabase != unmapped_buf, ("mapped buf: b_kvabase was not updated %p", bp)); KASSERT(bp->b_data != unmapped_buf, ("mapped buf: b_data was not updated %p", bp)); KASSERT(bp->b_data < unmapped_buf || bp->b_data >= unmapped_buf + MAXPHYS, ("b_data + b_offset unmapped %p", bp)); } static inline void vfs_buf_check_unmapped(struct buf *bp) { KASSERT(bp->b_data == unmapped_buf, ("unmapped buf: corrupted b_data %p", bp)); } #define BUF_CHECK_MAPPED(bp) vfs_buf_check_mapped(bp) #define BUF_CHECK_UNMAPPED(bp) vfs_buf_check_unmapped(bp) #else #define BUF_CHECK_MAPPED(bp) do {} while (0) #define BUF_CHECK_UNMAPPED(bp) do {} while (0) #endif static int isbufbusy(struct buf *bp) { if (((bp->b_flags & B_INVAL) == 0 && BUF_ISLOCKED(bp)) || ((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI)) return (1); return (0); } /* * Shutdown the system cleanly to prepare for reboot, halt, or power off. */ void bufshutdown(int show_busybufs) { static int first_buf_printf = 1; struct buf *bp; int iter, nbusy, pbusy; #ifndef PREEMPTION int subiter; #endif /* * Sync filesystems for shutdown */ wdog_kern_pat(WD_LASTVAL); sys_sync(curthread, NULL); /* * With soft updates, some buffers that are * written will be remarked as dirty until other * buffers are written. */ for (iter = pbusy = 0; iter < 20; iter++) { nbusy = 0; for (bp = &buf[nbuf]; --bp >= buf; ) if (isbufbusy(bp)) nbusy++; if (nbusy == 0) { if (first_buf_printf) printf("All buffers synced."); break; } if (first_buf_printf) { printf("Syncing disks, buffers remaining... "); first_buf_printf = 0; } printf("%d ", nbusy); if (nbusy < pbusy) iter = 0; pbusy = nbusy; wdog_kern_pat(WD_LASTVAL); sys_sync(curthread, NULL); #ifdef PREEMPTION /* * Drop Giant and spin for a while to allow * interrupt threads to run. */ DROP_GIANT(); DELAY(50000 * iter); PICKUP_GIANT(); #else /* * Drop Giant and context switch several times to * allow interrupt threads to run. */ DROP_GIANT(); for (subiter = 0; subiter < 50 * iter; subiter++) { thread_lock(curthread); mi_switch(SW_VOL, NULL); thread_unlock(curthread); DELAY(1000); } PICKUP_GIANT(); #endif } printf("\n"); /* * Count only busy local buffers to prevent forcing * a fsck if we're just a client of a wedged NFS server */ nbusy = 0; for (bp = &buf[nbuf]; --bp >= buf; ) { if (isbufbusy(bp)) { #if 0 /* XXX: This is bogus. We should probably have a BO_REMOTE flag instead */ if (bp->b_dev == NULL) { TAILQ_REMOVE(&mountlist, bp->b_vp->v_mount, mnt_list); continue; } #endif nbusy++; if (show_busybufs > 0) { printf( "%d: buf:%p, vnode:%p, flags:%0x, blkno:%jd, lblkno:%jd, buflock:", nbusy, bp, bp->b_vp, bp->b_flags, (intmax_t)bp->b_blkno, (intmax_t)bp->b_lblkno); BUF_LOCKPRINTINFO(bp); if (show_busybufs > 1) vn_printf(bp->b_vp, "vnode content: "); } } } if (nbusy) { /* * Failed to sync all blocks. Indicate this and don't * unmount filesystems (thus forcing an fsck on reboot). */ printf("Giving up on %d buffers\n", nbusy); DELAY(5000000); /* 5 seconds */ } else { if (!first_buf_printf) printf("Final sync complete\n"); /* * Unmount filesystems */ if (panicstr == NULL) vfs_unmountall(); } swapoff_all(); DELAY(100000); /* wait for console output to finish */ } static void bpmap_qenter(struct buf *bp) { BUF_CHECK_MAPPED(bp); /* * bp->b_data is relative to bp->b_offset, but * bp->b_offset may be offset into the first page. */ bp->b_data = (caddr_t)trunc_page((vm_offset_t)bp->b_data); pmap_qenter((vm_offset_t)bp->b_data, bp->b_pages, bp->b_npages); bp->b_data = (caddr_t)((vm_offset_t)bp->b_data | (vm_offset_t)(bp->b_offset & PAGE_MASK)); } /* * binsfree: * * Insert the buffer into the appropriate free list. */ static void binsfree(struct buf *bp, int qindex) { struct mtx *olock, *nlock; if (qindex != QUEUE_EMPTY) { BUF_ASSERT_XLOCKED(bp); } /* * Stick to the same clean queue for the lifetime of the buf to * limit locking below. Otherwise pick ont sequentially. */ if (qindex == QUEUE_CLEAN) { if (bqisclean(bp->b_qindex)) qindex = bp->b_qindex; else qindex = bqcleanq(); } /* * Handle delayed bremfree() processing. */ nlock = bqlock(qindex); if (bp->b_flags & B_REMFREE) { olock = bqlock(bp->b_qindex); mtx_lock(olock); bremfreel(bp); if (olock != nlock) { mtx_unlock(olock); mtx_lock(nlock); } } else mtx_lock(nlock); if (bp->b_qindex != QUEUE_NONE) panic("binsfree: free buffer onto another queue???"); bp->b_qindex = qindex; if (bp->b_flags & B_AGE) TAILQ_INSERT_HEAD(&bufqueues[bp->b_qindex], bp, b_freelist); else TAILQ_INSERT_TAIL(&bufqueues[bp->b_qindex], bp, b_freelist); #ifdef INVARIANTS bq_len[bp->b_qindex]++; #endif mtx_unlock(nlock); } /* * buf_free: * * Free a buffer to the buf zone once it no longer has valid contents. */ static void buf_free(struct buf *bp) { if (bp->b_flags & B_REMFREE) bremfreef(bp); if (bp->b_vflags & BV_BKGRDINPROG) panic("losing buffer 1"); if (bp->b_rcred != NOCRED) { crfree(bp->b_rcred); bp->b_rcred = NOCRED; } if (bp->b_wcred != NOCRED) { crfree(bp->b_wcred); bp->b_wcred = NOCRED; } if (!LIST_EMPTY(&bp->b_dep)) buf_deallocate(bp); bufkva_free(bp); BUF_UNLOCK(bp); uma_zfree(buf_zone, bp); atomic_add_int(&numfreebuffers, 1); bufspace_wakeup(); } /* * buf_import: * * Import bufs into the uma cache from the buf list. The system still * expects a static array of bufs and much of the synchronization * around bufs assumes type stable storage. As a result, UMA is used * only as a per-cpu cache of bufs still maintained on a global list. */ static int buf_import(void *arg, void **store, int cnt, int flags) { struct buf *bp; int i; mtx_lock(&bqlocks[QUEUE_EMPTY]); for (i = 0; i < cnt; i++) { bp = TAILQ_FIRST(&bufqueues[QUEUE_EMPTY]); if (bp == NULL) break; bremfreel(bp); store[i] = bp; } mtx_unlock(&bqlocks[QUEUE_EMPTY]); return (i); } /* * buf_release: * * Release bufs from the uma cache back to the buffer queues. */ static void buf_release(void *arg, void **store, int cnt) { int i; for (i = 0; i < cnt; i++) binsfree(store[i], QUEUE_EMPTY); } /* * buf_alloc: * * Allocate an empty buffer header. */ static struct buf * buf_alloc(void) { struct buf *bp; bp = uma_zalloc(buf_zone, M_NOWAIT); if (bp == NULL) { bufspace_daemonwakeup(); atomic_add_int(&numbufallocfails, 1); return (NULL); } /* * Wake-up the bufspace daemon on transition. */ if (atomic_fetchadd_int(&numfreebuffers, -1) == lofreebuffers) bufspace_daemonwakeup(); if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0) panic("getnewbuf_empty: Locked buf %p on free queue.", bp); KASSERT(bp->b_vp == NULL, ("bp: %p still has vnode %p.", bp, bp->b_vp)); KASSERT((bp->b_flags & (B_DELWRI | B_NOREUSE)) == 0, ("invalid buffer %p flags %#x", bp, bp->b_flags)); KASSERT((bp->b_xflags & (BX_VNCLEAN|BX_VNDIRTY)) == 0, ("bp: %p still on a buffer list. xflags %X", bp, bp->b_xflags)); KASSERT(bp->b_npages == 0, ("bp: %p still has %d vm pages\n", bp, bp->b_npages)); KASSERT(bp->b_kvasize == 0, ("bp: %p still has kva\n", bp)); KASSERT(bp->b_bufsize == 0, ("bp: %p still has bufspace\n", bp)); bp->b_flags = 0; bp->b_ioflags = 0; bp->b_xflags = 0; bp->b_vflags = 0; bp->b_vp = NULL; bp->b_blkno = bp->b_lblkno = 0; bp->b_offset = NOOFFSET; bp->b_iodone = 0; bp->b_error = 0; bp->b_resid = 0; bp->b_bcount = 0; bp->b_npages = 0; bp->b_dirtyoff = bp->b_dirtyend = 0; bp->b_bufobj = NULL; bp->b_pin_count = 0; bp->b_data = bp->b_kvabase = unmapped_buf; bp->b_fsprivate1 = NULL; bp->b_fsprivate2 = NULL; bp->b_fsprivate3 = NULL; LIST_INIT(&bp->b_dep); return (bp); } /* * buf_qrecycle: * * Free a buffer from the given bufqueue. kva controls whether the * freed buf must own some kva resources. This is used for * defragmenting. */ static int buf_qrecycle(int qindex, bool kva) { struct buf *bp, *nbp; if (kva) atomic_add_int(&bufdefragcnt, 1); nbp = NULL; mtx_lock(&bqlocks[qindex]); nbp = TAILQ_FIRST(&bufqueues[qindex]); /* * Run scan, possibly freeing data and/or kva mappings on the fly * depending. */ while ((bp = nbp) != NULL) { /* * Calculate next bp (we can only use it if we do not * release the bqlock). */ nbp = TAILQ_NEXT(bp, b_freelist); /* * If we are defragging then we need a buffer with * some kva to reclaim. */ if (kva && bp->b_kvasize == 0) continue; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0) continue; /* * Skip buffers with background writes in progress. */ if ((bp->b_vflags & BV_BKGRDINPROG) != 0) { BUF_UNLOCK(bp); continue; } KASSERT(bp->b_qindex == qindex, ("getnewbuf: inconsistent queue %d bp %p", qindex, bp)); /* * NOTE: nbp is now entirely invalid. We can only restart * the scan from this point on. */ bremfreel(bp); mtx_unlock(&bqlocks[qindex]); /* * Requeue the background write buffer with error and * restart the scan. */ if ((bp->b_vflags & BV_BKGRDERR) != 0) { bqrelse(bp); mtx_lock(&bqlocks[qindex]); nbp = TAILQ_FIRST(&bufqueues[qindex]); continue; } bp->b_flags |= B_INVAL; brelse(bp); return (0); } mtx_unlock(&bqlocks[qindex]); return (ENOBUFS); } /* * buf_recycle: * * Iterate through all clean queues until we find a buf to recycle or * exhaust the search. */ static int buf_recycle(bool kva) { int qindex, first_qindex; qindex = first_qindex = bqcleanq(); do { if (buf_qrecycle(qindex, kva) == 0) return (0); if (++qindex == QUEUE_CLEAN + clean_queues) qindex = QUEUE_CLEAN; } while (qindex != first_qindex); return (ENOBUFS); } /* * buf_scan: * * Scan the clean queues looking for a buffer to recycle. needsbuffer * is set on failure so that the caller may optionally bufspace_wait() * in a race-free fashion. */ static int buf_scan(bool defrag) { int error; /* * To avoid heavy synchronization and wakeup races we set * needsbuffer and re-poll before failing. This ensures that * no frees can be missed between an unsuccessful poll and * going to sleep in a synchronized fashion. */ if ((error = buf_recycle(defrag)) != 0) { atomic_set_int(&needsbuffer, 1); bufspace_daemonwakeup(); error = buf_recycle(defrag); } if (error == 0) atomic_add_int(&getnewbufrestarts, 1); return (error); } /* * bremfree: * * Mark the buffer for removal from the appropriate free list. * */ void bremfree(struct buf *bp) { CTR3(KTR_BUF, "bremfree(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT((bp->b_flags & B_REMFREE) == 0, ("bremfree: buffer %p already marked for delayed removal.", bp)); KASSERT(bp->b_qindex != QUEUE_NONE, ("bremfree: buffer %p not on a queue.", bp)); BUF_ASSERT_XLOCKED(bp); bp->b_flags |= B_REMFREE; } /* * bremfreef: * * Force an immediate removal from a free list. Used only in nfs when * it abuses the b_freelist pointer. */ void bremfreef(struct buf *bp) { struct mtx *qlock; qlock = bqlock(bp->b_qindex); mtx_lock(qlock); bremfreel(bp); mtx_unlock(qlock); } /* * bremfreel: * * Removes a buffer from the free list, must be called with the * correct qlock held. */ static void bremfreel(struct buf *bp) { CTR3(KTR_BUF, "bremfreel(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_qindex != QUEUE_NONE, ("bremfreel: buffer %p not on a queue.", bp)); if (bp->b_qindex != QUEUE_EMPTY) { BUF_ASSERT_XLOCKED(bp); } mtx_assert(bqlock(bp->b_qindex), MA_OWNED); TAILQ_REMOVE(&bufqueues[bp->b_qindex], bp, b_freelist); #ifdef INVARIANTS KASSERT(bq_len[bp->b_qindex] >= 1, ("queue %d underflow", bp->b_qindex)); bq_len[bp->b_qindex]--; #endif bp->b_qindex = QUEUE_NONE; bp->b_flags &= ~B_REMFREE; } /* * bufkva_free: * * Free the kva allocation for a buffer. * */ static void bufkva_free(struct buf *bp) { #ifdef INVARIANTS if (bp->b_kvasize == 0) { KASSERT(bp->b_kvabase == unmapped_buf && bp->b_data == unmapped_buf, ("Leaked KVA space on %p", bp)); } else if (buf_mapped(bp)) BUF_CHECK_MAPPED(bp); else BUF_CHECK_UNMAPPED(bp); #endif if (bp->b_kvasize == 0) return; vmem_free(buffer_arena, (vm_offset_t)bp->b_kvabase, bp->b_kvasize); atomic_subtract_long(&bufkvaspace, bp->b_kvasize); atomic_add_int(&buffreekvacnt, 1); bp->b_data = bp->b_kvabase = unmapped_buf; bp->b_kvasize = 0; } /* * bufkva_alloc: * * Allocate the buffer KVA and set b_kvasize and b_kvabase. */ static int bufkva_alloc(struct buf *bp, int maxsize, int gbflags) { vm_offset_t addr; int error; KASSERT((gbflags & GB_UNMAPPED) == 0 || (gbflags & GB_KVAALLOC) != 0, ("Invalid gbflags 0x%x in %s", gbflags, __func__)); bufkva_free(bp); addr = 0; error = vmem_alloc(buffer_arena, maxsize, M_BESTFIT | M_NOWAIT, &addr); if (error != 0) { /* * Buffer map is too fragmented. Request the caller * to defragment the map. */ return (error); } bp->b_kvabase = (caddr_t)addr; bp->b_kvasize = maxsize; atomic_add_long(&bufkvaspace, bp->b_kvasize); if ((gbflags & GB_UNMAPPED) != 0) { bp->b_data = unmapped_buf; BUF_CHECK_UNMAPPED(bp); } else { bp->b_data = bp->b_kvabase; BUF_CHECK_MAPPED(bp); } return (0); } /* * bufkva_reclaim: * * Reclaim buffer kva by freeing buffers holding kva. This is a vmem * callback that fires to avoid returning failure. */ static void bufkva_reclaim(vmem_t *vmem, int flags) { int i; for (i = 0; i < 5; i++) if (buf_scan(true) != 0) break; return; } /* * Attempt to initiate asynchronous I/O on read-ahead blocks. We must * clear BIO_ERROR and B_INVAL prior to initiating I/O . If B_CACHE is set, * the buffer is valid and we do not have to do anything. */ void breada(struct vnode * vp, daddr_t * rablkno, int * rabsize, int cnt, struct ucred * cred) { struct buf *rabp; int i; for (i = 0; i < cnt; i++, rablkno++, rabsize++) { if (inmem(vp, *rablkno)) continue; rabp = getblk(vp, *rablkno, *rabsize, 0, 0, 0); if ((rabp->b_flags & B_CACHE) == 0) { if (!TD_IS_IDLETHREAD(curthread)) { #ifdef RACCT if (racct_enable) { PROC_LOCK(curproc); racct_add_buf(curproc, rabp, 0); PROC_UNLOCK(curproc); } #endif /* RACCT */ curthread->td_ru.ru_inblock++; } rabp->b_flags |= B_ASYNC; rabp->b_flags &= ~B_INVAL; rabp->b_ioflags &= ~BIO_ERROR; rabp->b_iocmd = BIO_READ; if (rabp->b_rcred == NOCRED && cred != NOCRED) rabp->b_rcred = crhold(cred); vfs_busy_pages(rabp, 0); BUF_KERNPROC(rabp); rabp->b_iooffset = dbtob(rabp->b_blkno); bstrategy(rabp); } else { brelse(rabp); } } } /* * Entry point for bread() and breadn() via #defines in sys/buf.h. * * Get a buffer with the specified data. Look in the cache first. We * must clear BIO_ERROR and B_INVAL prior to initiating I/O. If B_CACHE * is set, the buffer is valid and we do not have to do anything, see * getblk(). Also starts asynchronous I/O on read-ahead blocks. * * Always return a NULL buffer pointer (in bpp) when returning an error. */ int breadn_flags(struct vnode *vp, daddr_t blkno, int size, daddr_t *rablkno, int *rabsize, int cnt, struct ucred *cred, int flags, struct buf **bpp) { struct buf *bp; int rv = 0, readwait = 0; CTR3(KTR_BUF, "breadn(%p, %jd, %d)", vp, blkno, size); /* * Can only return NULL if GB_LOCK_NOWAIT flag is specified. */ *bpp = bp = getblk(vp, blkno, size, 0, 0, flags); if (bp == NULL) return (EBUSY); /* if not found in cache, do some I/O */ if ((bp->b_flags & B_CACHE) == 0) { if (!TD_IS_IDLETHREAD(curthread)) { #ifdef RACCT if (racct_enable) { PROC_LOCK(curproc); racct_add_buf(curproc, bp, 0); PROC_UNLOCK(curproc); } #endif /* RACCT */ curthread->td_ru.ru_inblock++; } bp->b_iocmd = BIO_READ; bp->b_flags &= ~B_INVAL; bp->b_ioflags &= ~BIO_ERROR; if (bp->b_rcred == NOCRED && cred != NOCRED) bp->b_rcred = crhold(cred); vfs_busy_pages(bp, 0); bp->b_iooffset = dbtob(bp->b_blkno); bstrategy(bp); ++readwait; } breada(vp, rablkno, rabsize, cnt, cred); if (readwait) { rv = bufwait(bp); if (rv != 0) { brelse(bp); *bpp = NULL; } } return (rv); } /* * Write, release buffer on completion. (Done by iodone * if async). Do not bother writing anything if the buffer * is invalid. * * Note that we set B_CACHE here, indicating that buffer is * fully valid and thus cacheable. This is true even of NFS * now so we set it generally. This could be set either here * or in biodone() since the I/O is synchronous. We put it * here. */ int bufwrite(struct buf *bp) { int oldflags; struct vnode *vp; long space; int vp_md; CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); if ((bp->b_bufobj->bo_flag & BO_DEAD) != 0) { bp->b_flags |= B_INVAL | B_RELBUF; bp->b_flags &= ~B_CACHE; brelse(bp); return (ENXIO); } if (bp->b_flags & B_INVAL) { brelse(bp); return (0); } if (bp->b_flags & B_BARRIER) barrierwrites++; oldflags = bp->b_flags; BUF_ASSERT_HELD(bp); if (bp->b_pin_count > 0) bunpin_wait(bp); KASSERT(!(bp->b_vflags & BV_BKGRDINPROG), ("FFS background buffer should not get here %p", bp)); vp = bp->b_vp; if (vp) vp_md = vp->v_vflag & VV_MD; else vp_md = 0; /* * Mark the buffer clean. Increment the bufobj write count * before bundirty() call, to prevent other thread from seeing * empty dirty list and zero counter for writes in progress, * falsely indicating that the bufobj is clean. */ bufobj_wref(bp->b_bufobj); bundirty(bp); bp->b_flags &= ~B_DONE; bp->b_ioflags &= ~BIO_ERROR; bp->b_flags |= B_CACHE; bp->b_iocmd = BIO_WRITE; vfs_busy_pages(bp, 1); /* * Normal bwrites pipeline writes */ bp->b_runningbufspace = bp->b_bufsize; space = atomic_fetchadd_long(&runningbufspace, bp->b_runningbufspace); if (!TD_IS_IDLETHREAD(curthread)) { #ifdef RACCT if (racct_enable) { PROC_LOCK(curproc); racct_add_buf(curproc, bp, 1); PROC_UNLOCK(curproc); } #endif /* RACCT */ curthread->td_ru.ru_oublock++; } if (oldflags & B_ASYNC) BUF_KERNPROC(bp); bp->b_iooffset = dbtob(bp->b_blkno); bstrategy(bp); if ((oldflags & B_ASYNC) == 0) { int rtval = bufwait(bp); brelse(bp); return (rtval); } else if (space > hirunningspace) { /* * don't allow the async write to saturate the I/O * system. We will not deadlock here because * we are blocking waiting for I/O that is already in-progress * to complete. We do not block here if it is the update * or syncer daemon trying to clean up as that can lead * to deadlock. */ if ((curthread->td_pflags & TDP_NORUNNINGBUF) == 0 && !vp_md) waitrunningbufspace(); } return (0); } void bufbdflush(struct bufobj *bo, struct buf *bp) { struct buf *nbp; if (bo->bo_dirty.bv_cnt > dirtybufthresh + 10) { (void) VOP_FSYNC(bp->b_vp, MNT_NOWAIT, curthread); altbufferflushes++; } else if (bo->bo_dirty.bv_cnt > dirtybufthresh) { BO_LOCK(bo); /* * Try to find a buffer to flush. */ TAILQ_FOREACH(nbp, &bo->bo_dirty.bv_hd, b_bobufs) { if ((nbp->b_vflags & BV_BKGRDINPROG) || BUF_LOCK(nbp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) continue; if (bp == nbp) panic("bdwrite: found ourselves"); BO_UNLOCK(bo); /* Don't countdeps with the bo lock held. */ if (buf_countdeps(nbp, 0)) { BO_LOCK(bo); BUF_UNLOCK(nbp); continue; } if (nbp->b_flags & B_CLUSTEROK) { vfs_bio_awrite(nbp); } else { bremfree(nbp); bawrite(nbp); } dirtybufferflushes++; break; } if (nbp == NULL) BO_UNLOCK(bo); } } /* * Delayed write. (Buffer is marked dirty). Do not bother writing * anything if the buffer is marked invalid. * * Note that since the buffer must be completely valid, we can safely * set B_CACHE. In fact, we have to set B_CACHE here rather then in * biodone() in order to prevent getblk from writing the buffer * out synchronously. */ void bdwrite(struct buf *bp) { struct thread *td = curthread; struct vnode *vp; struct bufobj *bo; CTR3(KTR_BUF, "bdwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); KASSERT((bp->b_flags & B_BARRIER) == 0, ("Barrier request in delayed write %p", bp)); BUF_ASSERT_HELD(bp); if (bp->b_flags & B_INVAL) { brelse(bp); return; } /* * If we have too many dirty buffers, don't create any more. * If we are wildly over our limit, then force a complete * cleanup. Otherwise, just keep the situation from getting * out of control. Note that we have to avoid a recursive * disaster and not try to clean up after our own cleanup! */ vp = bp->b_vp; bo = bp->b_bufobj; if ((td->td_pflags & (TDP_COWINPROGRESS|TDP_INBDFLUSH)) == 0) { td->td_pflags |= TDP_INBDFLUSH; BO_BDFLUSH(bo, bp); td->td_pflags &= ~TDP_INBDFLUSH; } else recursiveflushes++; bdirty(bp); /* * Set B_CACHE, indicating that the buffer is fully valid. This is * true even of NFS now. */ bp->b_flags |= B_CACHE; /* * This bmap keeps the system from needing to do the bmap later, * perhaps when the system is attempting to do a sync. Since it * is likely that the indirect block -- or whatever other datastructure * that the filesystem needs is still in memory now, it is a good * thing to do this. Note also, that if the pageout daemon is * requesting a sync -- there might not be enough memory to do * the bmap then... So, this is important to do. */ if (vp->v_type != VCHR && bp->b_lblkno == bp->b_blkno) { VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, NULL, NULL); } /* * Set the *dirty* buffer range based upon the VM system dirty * pages. * * Mark the buffer pages as clean. We need to do this here to * satisfy the vnode_pager and the pageout daemon, so that it * thinks that the pages have been "cleaned". Note that since * the pages are in a delayed write buffer -- the VFS layer * "will" see that the pages get written out on the next sync, * or perhaps the cluster will be completed. */ vfs_clean_pages_dirty_buf(bp); bqrelse(bp); /* * note: we cannot initiate I/O from a bdwrite even if we wanted to, * due to the softdep code. */ } /* * bdirty: * * Turn buffer into delayed write request. We must clear BIO_READ and * B_RELBUF, and we must set B_DELWRI. We reassign the buffer to * itself to properly update it in the dirty/clean lists. We mark it * B_DONE to ensure that any asynchronization of the buffer properly * clears B_DONE ( else a panic will occur later ). * * bdirty() is kinda like bdwrite() - we have to clear B_INVAL which * might have been set pre-getblk(). Unlike bwrite/bdwrite, bdirty() * should only be called if the buffer is known-good. * * Since the buffer is not on a queue, we do not update the numfreebuffers * count. * * The buffer must be on QUEUE_NONE. */ void bdirty(struct buf *bp) { CTR3(KTR_BUF, "bdirty(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); KASSERT(bp->b_flags & B_REMFREE || bp->b_qindex == QUEUE_NONE, ("bdirty: buffer %p still on queue %d", bp, bp->b_qindex)); BUF_ASSERT_HELD(bp); bp->b_flags &= ~(B_RELBUF); bp->b_iocmd = BIO_WRITE; if ((bp->b_flags & B_DELWRI) == 0) { bp->b_flags |= /* XXX B_DONE | */ B_DELWRI; reassignbuf(bp); bdirtyadd(); } } /* * bundirty: * * Clear B_DELWRI for buffer. * * Since the buffer is not on a queue, we do not update the numfreebuffers * count. * * The buffer must be on QUEUE_NONE. */ void bundirty(struct buf *bp) { CTR3(KTR_BUF, "bundirty(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); KASSERT(bp->b_flags & B_REMFREE || bp->b_qindex == QUEUE_NONE, ("bundirty: buffer %p still on queue %d", bp, bp->b_qindex)); BUF_ASSERT_HELD(bp); if (bp->b_flags & B_DELWRI) { bp->b_flags &= ~B_DELWRI; reassignbuf(bp); bdirtysub(); } /* * Since it is now being written, we can clear its deferred write flag. */ bp->b_flags &= ~B_DEFERRED; } /* * bawrite: * * Asynchronous write. Start output on a buffer, but do not wait for * it to complete. The buffer is released when the output completes. * * bwrite() ( or the VOP routine anyway ) is responsible for handling * B_INVAL buffers. Not us. */ void bawrite(struct buf *bp) { bp->b_flags |= B_ASYNC; (void) bwrite(bp); } /* * babarrierwrite: * * Asynchronous barrier write. Start output on a buffer, but do not * wait for it to complete. Place a write barrier after this write so * that this buffer and all buffers written before it are committed to * the disk before any buffers written after this write are committed * to the disk. The buffer is released when the output completes. */ void babarrierwrite(struct buf *bp) { bp->b_flags |= B_ASYNC | B_BARRIER; (void) bwrite(bp); } /* * bbarrierwrite: * * Synchronous barrier write. Start output on a buffer and wait for * it to complete. Place a write barrier after this write so that * this buffer and all buffers written before it are committed to * the disk before any buffers written after this write are committed * to the disk. The buffer is released when the output completes. */ int bbarrierwrite(struct buf *bp) { bp->b_flags |= B_BARRIER; return (bwrite(bp)); } /* * bwillwrite: * * Called prior to the locking of any vnodes when we are expecting to * write. We do not want to starve the buffer cache with too many * dirty buffers so we block here. By blocking prior to the locking * of any vnodes we attempt to avoid the situation where a locked vnode * prevents the various system daemons from flushing related buffers. */ void bwillwrite(void) { if (numdirtybuffers >= hidirtybuffers) { mtx_lock(&bdirtylock); while (numdirtybuffers >= hidirtybuffers) { bdirtywait = 1; msleep(&bdirtywait, &bdirtylock, (PRIBIO + 4), "flswai", 0); } mtx_unlock(&bdirtylock); } } /* * Return true if we have too many dirty buffers. */ int buf_dirty_count_severe(void) { return(numdirtybuffers >= hidirtybuffers); } /* * brelse: * * Release a busy buffer and, if requested, free its resources. The * buffer will be stashed in the appropriate bufqueue[] allowing it * to be accessed later as a cache entity or reused for other purposes. */ void brelse(struct buf *bp) { int qindex; /* * Many functions erroneously call brelse with a NULL bp under rare * error conditions. Simply return when called with a NULL bp. */ if (bp == NULL) return; CTR3(KTR_BUF, "brelse(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(!(bp->b_flags & (B_CLUSTER|B_PAGING)), ("brelse: inappropriate B_PAGING or B_CLUSTER bp %p", bp)); KASSERT((bp->b_flags & B_VMIO) != 0 || (bp->b_flags & B_NOREUSE) == 0, ("brelse: non-VMIO buffer marked NOREUSE")); if (BUF_LOCKRECURSED(bp)) { /* * Do not process, in particular, do not handle the * B_INVAL/B_RELBUF and do not release to free list. */ BUF_UNLOCK(bp); return; } if (bp->b_flags & B_MANAGED) { bqrelse(bp); return; } if ((bp->b_vflags & (BV_BKGRDINPROG | BV_BKGRDERR)) == BV_BKGRDERR) { BO_LOCK(bp->b_bufobj); bp->b_vflags &= ~BV_BKGRDERR; BO_UNLOCK(bp->b_bufobj); bdirty(bp); } if (bp->b_iocmd == BIO_WRITE && (bp->b_ioflags & BIO_ERROR) && (bp->b_error != ENXIO || !LIST_EMPTY(&bp->b_dep)) && !(bp->b_flags & B_INVAL)) { /* * Failed write, redirty. All errors except ENXIO (which * means the device is gone) are expected to be potentially * transient - underlying media might work if tried again * after EIO, and memory might be available after an ENOMEM. * * Do this also for buffers that failed with ENXIO, but have * non-empty dependencies - the soft updates code might need * to access the buffer to untangle them. * * Must clear BIO_ERROR to prevent pages from being scrapped. */ bp->b_ioflags &= ~BIO_ERROR; bdirty(bp); } else if ((bp->b_flags & (B_NOCACHE | B_INVAL)) || (bp->b_ioflags & BIO_ERROR) || (bp->b_bufsize <= 0)) { /* * Either a failed read I/O, or we were asked to free or not * cache the buffer, or we failed to write to a device that's * no longer present. */ bp->b_flags |= B_INVAL; if (!LIST_EMPTY(&bp->b_dep)) buf_deallocate(bp); if (bp->b_flags & B_DELWRI) bdirtysub(); bp->b_flags &= ~(B_DELWRI | B_CACHE); if ((bp->b_flags & B_VMIO) == 0) { allocbuf(bp, 0); if (bp->b_vp) brelvp(bp); } } /* * We must clear B_RELBUF if B_DELWRI is set. If vfs_vmio_truncate() * is called with B_DELWRI set, the underlying pages may wind up * getting freed causing a previous write (bdwrite()) to get 'lost' * because pages associated with a B_DELWRI bp are marked clean. * * We still allow the B_INVAL case to call vfs_vmio_truncate(), even * if B_DELWRI is set. */ if (bp->b_flags & B_DELWRI) bp->b_flags &= ~B_RELBUF; /* * VMIO buffer rundown. It is not very necessary to keep a VMIO buffer * constituted, not even NFS buffers now. Two flags effect this. If * B_INVAL, the struct buf is invalidated but the VM object is kept * around ( i.e. so it is trivial to reconstitute the buffer later ). * * If BIO_ERROR or B_NOCACHE is set, pages in the VM object will be * invalidated. BIO_ERROR cannot be set for a failed write unless the * buffer is also B_INVAL because it hits the re-dirtying code above. * * Normally we can do this whether a buffer is B_DELWRI or not. If * the buffer is an NFS buffer, it is tracking piecemeal writes or * the commit state and we cannot afford to lose the buffer. If the * buffer has a background write in progress, we need to keep it * around to prevent it from being reconstituted and starting a second * background write. */ if ((bp->b_flags & B_VMIO) && (bp->b_flags & B_NOCACHE || (bp->b_ioflags & BIO_ERROR && bp->b_iocmd == BIO_READ)) && !(bp->b_vp->v_mount != NULL && (bp->b_vp->v_mount->mnt_vfc->vfc_flags & VFCF_NETWORK) != 0 && !vn_isdisk(bp->b_vp, NULL) && (bp->b_flags & B_DELWRI))) { vfs_vmio_invalidate(bp); allocbuf(bp, 0); } if ((bp->b_flags & (B_INVAL | B_RELBUF)) != 0 || (bp->b_flags & (B_DELWRI | B_NOREUSE)) == B_NOREUSE) { allocbuf(bp, 0); bp->b_flags &= ~B_NOREUSE; if (bp->b_vp != NULL) brelvp(bp); } /* * If the buffer has junk contents signal it and eventually * clean up B_DELWRI and diassociate the vnode so that gbincore() * doesn't find it. */ if (bp->b_bufsize == 0 || (bp->b_ioflags & BIO_ERROR) != 0 || (bp->b_flags & (B_INVAL | B_NOCACHE | B_RELBUF)) != 0) bp->b_flags |= B_INVAL; if (bp->b_flags & B_INVAL) { if (bp->b_flags & B_DELWRI) bundirty(bp); if (bp->b_vp) brelvp(bp); } /* buffers with no memory */ if (bp->b_bufsize == 0) { buf_free(bp); return; } /* buffers with junk contents */ if (bp->b_flags & (B_INVAL | B_NOCACHE | B_RELBUF) || (bp->b_ioflags & BIO_ERROR)) { bp->b_xflags &= ~(BX_BKGRDWRITE | BX_ALTDATA); if (bp->b_vflags & BV_BKGRDINPROG) panic("losing buffer 2"); qindex = QUEUE_CLEAN; bp->b_flags |= B_AGE; /* remaining buffers */ } else if (bp->b_flags & B_DELWRI) qindex = QUEUE_DIRTY; else qindex = QUEUE_CLEAN; binsfree(bp, qindex); bp->b_flags &= ~(B_ASYNC | B_NOCACHE | B_AGE | B_RELBUF | B_DIRECT); if ((bp->b_flags & B_DELWRI) == 0 && (bp->b_xflags & BX_VNDIRTY)) panic("brelse: not dirty"); /* unlock */ BUF_UNLOCK(bp); if (qindex == QUEUE_CLEAN) bufspace_wakeup(); } /* * Release a buffer back to the appropriate queue but do not try to free * it. The buffer is expected to be used again soon. * * bqrelse() is used by bdwrite() to requeue a delayed write, and used by * biodone() to requeue an async I/O on completion. It is also used when * known good buffers need to be requeued but we think we may need the data * again soon. * * XXX we should be able to leave the B_RELBUF hint set on completion. */ void bqrelse(struct buf *bp) { int qindex; CTR3(KTR_BUF, "bqrelse(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(!(bp->b_flags & (B_CLUSTER|B_PAGING)), ("bqrelse: inappropriate B_PAGING or B_CLUSTER bp %p", bp)); qindex = QUEUE_NONE; if (BUF_LOCKRECURSED(bp)) { /* do not release to free list */ BUF_UNLOCK(bp); return; } bp->b_flags &= ~(B_ASYNC | B_NOCACHE | B_AGE | B_RELBUF); if (bp->b_flags & B_MANAGED) { if (bp->b_flags & B_REMFREE) bremfreef(bp); goto out; } /* buffers with stale but valid contents */ if ((bp->b_flags & B_DELWRI) != 0 || (bp->b_vflags & (BV_BKGRDINPROG | BV_BKGRDERR)) == BV_BKGRDERR) { BO_LOCK(bp->b_bufobj); bp->b_vflags &= ~BV_BKGRDERR; BO_UNLOCK(bp->b_bufobj); qindex = QUEUE_DIRTY; } else { if ((bp->b_flags & B_DELWRI) == 0 && (bp->b_xflags & BX_VNDIRTY)) panic("bqrelse: not dirty"); if ((bp->b_flags & B_NOREUSE) != 0) { brelse(bp); return; } qindex = QUEUE_CLEAN; } binsfree(bp, qindex); out: /* unlock */ BUF_UNLOCK(bp); if (qindex == QUEUE_CLEAN) bufspace_wakeup(); } /* * Complete I/O to a VMIO backed page. Validate the pages as appropriate, * restore bogus pages. */ static void vfs_vmio_iodone(struct buf *bp) { vm_ooffset_t foff; vm_page_t m; vm_object_t obj; struct vnode *vp; int i, iosize, resid; bool bogus; obj = bp->b_bufobj->bo_object; KASSERT(obj->paging_in_progress >= bp->b_npages, ("vfs_vmio_iodone: paging in progress(%d) < b_npages(%d)", obj->paging_in_progress, bp->b_npages)); vp = bp->b_vp; KASSERT(vp->v_holdcnt > 0, ("vfs_vmio_iodone: vnode %p has zero hold count", vp)); KASSERT(vp->v_object != NULL, ("vfs_vmio_iodone: vnode %p has no vm_object", vp)); foff = bp->b_offset; KASSERT(bp->b_offset != NOOFFSET, ("vfs_vmio_iodone: bp %p has no buffer offset", bp)); bogus = false; iosize = bp->b_bcount - bp->b_resid; VM_OBJECT_WLOCK(obj); for (i = 0; i < bp->b_npages; i++) { resid = ((foff + PAGE_SIZE) & ~(off_t)PAGE_MASK) - foff; if (resid > iosize) resid = iosize; /* * cleanup bogus pages, restoring the originals */ m = bp->b_pages[i]; if (m == bogus_page) { bogus = true; m = vm_page_lookup(obj, OFF_TO_IDX(foff)); if (m == NULL) panic("biodone: page disappeared!"); bp->b_pages[i] = m; } else if ((bp->b_iocmd == BIO_READ) && resid > 0) { /* * In the write case, the valid and clean bits are * already changed correctly ( see bdwrite() ), so we * only need to do this here in the read case. */ KASSERT((m->dirty & vm_page_bits(foff & PAGE_MASK, resid)) == 0, ("vfs_vmio_iodone: page %p " "has unexpected dirty bits", m)); vfs_page_set_valid(bp, foff, m); } KASSERT(OFF_TO_IDX(foff) == m->pindex, ("vfs_vmio_iodone: foff(%jd)/pindex(%ju) mismatch", (intmax_t)foff, (uintmax_t)m->pindex)); vm_page_sunbusy(m); foff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK; iosize -= resid; } vm_object_pip_wakeupn(obj, bp->b_npages); VM_OBJECT_WUNLOCK(obj); if (bogus && buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); pmap_qenter(trunc_page((vm_offset_t)bp->b_data), bp->b_pages, bp->b_npages); } } /* * Unwire a page held by a buf and place it on the appropriate vm queue. */ static void vfs_vmio_unwire(struct buf *bp, vm_page_t m) { bool freed; vm_page_lock(m); if (vm_page_unwire(m, PQ_NONE)) { /* * Determine if the page should be freed before adding * it to the inactive queue. */ if (m->valid == 0) { freed = !vm_page_busied(m); if (freed) vm_page_free(m); } else if ((bp->b_flags & B_DIRECT) != 0) freed = vm_page_try_to_free(m); else freed = false; if (!freed) { /* * If the page is unlikely to be reused, let the * VM know. Otherwise, maintain LRU page * ordering and put the page at the tail of the * inactive queue. */ if ((bp->b_flags & B_NOREUSE) != 0) vm_page_deactivate_noreuse(m); else vm_page_deactivate(m); } } vm_page_unlock(m); } /* * Perform page invalidation when a buffer is released. The fully invalid * pages will be reclaimed later in vfs_vmio_truncate(). */ static void vfs_vmio_invalidate(struct buf *bp) { vm_object_t obj; vm_page_t m; int i, resid, poffset, presid; if (buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); pmap_qremove(trunc_page((vm_offset_t)bp->b_data), bp->b_npages); } else BUF_CHECK_UNMAPPED(bp); /* * Get the base offset and length of the buffer. Note that * in the VMIO case if the buffer block size is not * page-aligned then b_data pointer may not be page-aligned. * But our b_pages[] array *IS* page aligned. * * block sizes less then DEV_BSIZE (usually 512) are not * supported due to the page granularity bits (m->valid, * m->dirty, etc...). * * See man buf(9) for more information */ obj = bp->b_bufobj->bo_object; resid = bp->b_bufsize; poffset = bp->b_offset & PAGE_MASK; VM_OBJECT_WLOCK(obj); for (i = 0; i < bp->b_npages; i++) { m = bp->b_pages[i]; if (m == bogus_page) panic("vfs_vmio_invalidate: Unexpected bogus page."); bp->b_pages[i] = NULL; presid = resid > (PAGE_SIZE - poffset) ? (PAGE_SIZE - poffset) : resid; KASSERT(presid >= 0, ("brelse: extra page")); while (vm_page_xbusied(m)) { vm_page_lock(m); VM_OBJECT_WUNLOCK(obj); vm_page_busy_sleep(m, "mbncsh", true); VM_OBJECT_WLOCK(obj); } if (pmap_page_wired_mappings(m) == 0) vm_page_set_invalid(m, poffset, presid); vfs_vmio_unwire(bp, m); resid -= presid; poffset = 0; } VM_OBJECT_WUNLOCK(obj); bp->b_npages = 0; } /* * Page-granular truncation of an existing VMIO buffer. */ static void vfs_vmio_truncate(struct buf *bp, int desiredpages) { vm_object_t obj; vm_page_t m; int i; if (bp->b_npages == desiredpages) return; if (buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); pmap_qremove((vm_offset_t)trunc_page((vm_offset_t)bp->b_data) + (desiredpages << PAGE_SHIFT), bp->b_npages - desiredpages); } else BUF_CHECK_UNMAPPED(bp); obj = bp->b_bufobj->bo_object; if (obj != NULL) VM_OBJECT_WLOCK(obj); for (i = desiredpages; i < bp->b_npages; i++) { m = bp->b_pages[i]; KASSERT(m != bogus_page, ("allocbuf: bogus page found")); bp->b_pages[i] = NULL; vfs_vmio_unwire(bp, m); } if (obj != NULL) VM_OBJECT_WUNLOCK(obj); bp->b_npages = desiredpages; } /* * Byte granular extension of VMIO buffers. */ static void vfs_vmio_extend(struct buf *bp, int desiredpages, int size) { /* * We are growing the buffer, possibly in a * byte-granular fashion. */ vm_object_t obj; vm_offset_t toff; vm_offset_t tinc; vm_page_t m; /* * Step 1, bring in the VM pages from the object, allocating * them if necessary. We must clear B_CACHE if these pages * are not valid for the range covered by the buffer. */ obj = bp->b_bufobj->bo_object; VM_OBJECT_WLOCK(obj); if (bp->b_npages < desiredpages) { /* * We must allocate system pages since blocking * here could interfere with paging I/O, no * matter which process we are. * * Only exclusive busy can be tested here. * Blocking on shared busy might lead to * deadlocks once allocbuf() is called after * pages are vfs_busy_pages(). */ (void)vm_page_grab_pages(obj, OFF_TO_IDX(bp->b_offset) + bp->b_npages, VM_ALLOC_SYSTEM | VM_ALLOC_IGN_SBUSY | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED, &bp->b_pages[bp->b_npages], desiredpages - bp->b_npages); bp->b_npages = desiredpages; } /* * Step 2. We've loaded the pages into the buffer, * we have to figure out if we can still have B_CACHE * set. Note that B_CACHE is set according to the * byte-granular range ( bcount and size ), not the * aligned range ( newbsize ). * * The VM test is against m->valid, which is DEV_BSIZE * aligned. Needless to say, the validity of the data * needs to also be DEV_BSIZE aligned. Note that this * fails with NFS if the server or some other client * extends the file's EOF. If our buffer is resized, * B_CACHE may remain set! XXX */ toff = bp->b_bcount; tinc = PAGE_SIZE - ((bp->b_offset + toff) & PAGE_MASK); while ((bp->b_flags & B_CACHE) && toff < size) { vm_pindex_t pi; if (tinc > (size - toff)) tinc = size - toff; pi = ((bp->b_offset & PAGE_MASK) + toff) >> PAGE_SHIFT; m = bp->b_pages[pi]; vfs_buf_test_cache(bp, bp->b_offset, toff, tinc, m); toff += tinc; tinc = PAGE_SIZE; } VM_OBJECT_WUNLOCK(obj); /* * Step 3, fixup the KVA pmap. */ if (buf_mapped(bp)) bpmap_qenter(bp); else BUF_CHECK_UNMAPPED(bp); } /* * Check to see if a block at a particular lbn is available for a clustered * write. */ static int vfs_bio_clcheck(struct vnode *vp, int size, daddr_t lblkno, daddr_t blkno) { struct buf *bpa; int match; match = 0; /* If the buf isn't in core skip it */ if ((bpa = gbincore(&vp->v_bufobj, lblkno)) == NULL) return (0); /* If the buf is busy we don't want to wait for it */ if (BUF_LOCK(bpa, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0) return (0); /* Only cluster with valid clusterable delayed write buffers */ if ((bpa->b_flags & (B_DELWRI | B_CLUSTEROK | B_INVAL)) != (B_DELWRI | B_CLUSTEROK)) goto done; if (bpa->b_bufsize != size) goto done; /* * Check to see if it is in the expected place on disk and that the * block has been mapped. */ if ((bpa->b_blkno != bpa->b_lblkno) && (bpa->b_blkno == blkno)) match = 1; done: BUF_UNLOCK(bpa); return (match); } /* * vfs_bio_awrite: * * Implement clustered async writes for clearing out B_DELWRI buffers. * This is much better then the old way of writing only one buffer at * a time. Note that we may not be presented with the buffers in the * correct order, so we search for the cluster in both directions. */ int vfs_bio_awrite(struct buf *bp) { struct bufobj *bo; int i; int j; daddr_t lblkno = bp->b_lblkno; struct vnode *vp = bp->b_vp; int ncl; int nwritten; int size; int maxcl; int gbflags; bo = &vp->v_bufobj; gbflags = (bp->b_data == unmapped_buf) ? GB_UNMAPPED : 0; /* * right now we support clustered writing only to regular files. If * we find a clusterable block we could be in the middle of a cluster * rather then at the beginning. */ if ((vp->v_type == VREG) && (vp->v_mount != 0) && /* Only on nodes that have the size info */ (bp->b_flags & (B_CLUSTEROK | B_INVAL)) == B_CLUSTEROK) { size = vp->v_mount->mnt_stat.f_iosize; maxcl = MAXPHYS / size; BO_RLOCK(bo); for (i = 1; i < maxcl; i++) if (vfs_bio_clcheck(vp, size, lblkno + i, bp->b_blkno + ((i * size) >> DEV_BSHIFT)) == 0) break; for (j = 1; i + j <= maxcl && j <= lblkno; j++) if (vfs_bio_clcheck(vp, size, lblkno - j, bp->b_blkno - ((j * size) >> DEV_BSHIFT)) == 0) break; BO_RUNLOCK(bo); --j; ncl = i + j; /* * this is a possible cluster write */ if (ncl != 1) { BUF_UNLOCK(bp); nwritten = cluster_wbuild(vp, size, lblkno - j, ncl, gbflags); return (nwritten); } } bremfree(bp); bp->b_flags |= B_ASYNC; /* * default (old) behavior, writing out only one block * * XXX returns b_bufsize instead of b_bcount for nwritten? */ nwritten = bp->b_bufsize; (void) bwrite(bp); return (nwritten); } /* * getnewbuf_kva: * * Allocate KVA for an empty buf header according to gbflags. */ static int getnewbuf_kva(struct buf *bp, int gbflags, int maxsize) { if ((gbflags & (GB_UNMAPPED | GB_KVAALLOC)) != GB_UNMAPPED) { /* * In order to keep fragmentation sane we only allocate kva * in BKVASIZE chunks. XXX with vmem we can do page size. */ maxsize = (maxsize + BKVAMASK) & ~BKVAMASK; if (maxsize != bp->b_kvasize && bufkva_alloc(bp, maxsize, gbflags)) return (ENOSPC); } return (0); } /* * getnewbuf: * * Find and initialize a new buffer header, freeing up existing buffers * in the bufqueues as necessary. The new buffer is returned locked. * * We block if: * We have insufficient buffer headers * We have insufficient buffer space * buffer_arena is too fragmented ( space reservation fails ) * If we have to flush dirty buffers ( but we try to avoid this ) * * The caller is responsible for releasing the reserved bufspace after * allocbuf() is called. */ static struct buf * getnewbuf(struct vnode *vp, int slpflag, int slptimeo, int maxsize, int gbflags) { struct buf *bp; bool metadata, reserved; bp = NULL; KASSERT((gbflags & (GB_UNMAPPED | GB_KVAALLOC)) != GB_KVAALLOC, ("GB_KVAALLOC only makes sense with GB_UNMAPPED")); if (!unmapped_buf_allowed) gbflags &= ~(GB_UNMAPPED | GB_KVAALLOC); if (vp == NULL || (vp->v_vflag & (VV_MD | VV_SYSTEM)) != 0 || vp->v_type == VCHR) metadata = true; else metadata = false; atomic_add_int(&getnewbufcalls, 1); reserved = false; do { if (reserved == false && bufspace_reserve(maxsize, metadata) != 0) continue; reserved = true; if ((bp = buf_alloc()) == NULL) continue; if (getnewbuf_kva(bp, gbflags, maxsize) == 0) return (bp); break; } while(buf_scan(false) == 0); if (reserved) atomic_subtract_long(&bufspace, maxsize); if (bp != NULL) { bp->b_flags |= B_INVAL; brelse(bp); } bufspace_wait(vp, gbflags, slpflag, slptimeo); return (NULL); } /* * buf_daemon: * * buffer flushing daemon. Buffers are normally flushed by the * update daemon but if it cannot keep up this process starts to * take the load in an attempt to prevent getnewbuf() from blocking. */ static struct kproc_desc buf_kp = { "bufdaemon", buf_daemon, &bufdaemonproc }; SYSINIT(bufdaemon, SI_SUB_KTHREAD_BUF, SI_ORDER_FIRST, kproc_start, &buf_kp); static int buf_flush(struct vnode *vp, int target) { int flushed; flushed = flushbufqueues(vp, target, 0); if (flushed == 0) { /* * Could not find any buffers without rollback * dependencies, so just write the first one * in the hopes of eventually making progress. */ if (vp != NULL && target > 2) target /= 2; flushbufqueues(vp, target, 1); } return (flushed); } static void buf_daemon() { int lodirty; /* * This process needs to be suspended prior to shutdown sync. */ EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, bufdaemonproc, SHUTDOWN_PRI_LAST); /* * This process is allowed to take the buffer cache to the limit */ curthread->td_pflags |= TDP_NORUNNINGBUF | TDP_BUFNEED; mtx_lock(&bdlock); for (;;) { bd_request = 0; mtx_unlock(&bdlock); kproc_suspend_check(bufdaemonproc); lodirty = lodirtybuffers; if (bd_speedupreq) { lodirty = numdirtybuffers / 2; bd_speedupreq = 0; } /* * Do the flush. Limit the amount of in-transit I/O we * allow to build up, otherwise we would completely saturate * the I/O system. */ while (numdirtybuffers > lodirty) { if (buf_flush(NULL, numdirtybuffers - lodirty) == 0) break; kern_yield(PRI_USER); } /* * Only clear bd_request if we have reached our low water * mark. The buf_daemon normally waits 1 second and * then incrementally flushes any dirty buffers that have * built up, within reason. * * If we were unable to hit our low water mark and couldn't * find any flushable buffers, we sleep for a short period * to avoid endless loops on unlockable buffers. */ mtx_lock(&bdlock); if (numdirtybuffers <= lodirtybuffers) { /* * We reached our low water mark, reset the * request and sleep until we are needed again. * The sleep is just so the suspend code works. */ bd_request = 0; /* * Do an extra wakeup in case dirty threshold * changed via sysctl and the explicit transition * out of shortfall was missed. */ bdirtywakeup(); if (runningbufspace <= lorunningspace) runningwakeup(); msleep(&bd_request, &bdlock, PVM, "psleep", hz); } else { /* * We couldn't find any flushable dirty buffers but * still have too many dirty buffers, we * have to sleep and try again. (rare) */ msleep(&bd_request, &bdlock, PVM, "qsleep", hz / 10); } } } /* * flushbufqueues: * * Try to flush a buffer in the dirty queue. We must be careful to * free up B_INVAL buffers instead of write them, which NFS is * particularly sensitive to. */ static int flushwithdeps = 0; SYSCTL_INT(_vfs, OID_AUTO, flushwithdeps, CTLFLAG_RW, &flushwithdeps, 0, "Number of buffers flushed with dependecies that require rollbacks"); static int flushbufqueues(struct vnode *lvp, int target, int flushdeps) { struct buf *sentinel; struct vnode *vp; struct mount *mp; struct buf *bp; int hasdeps; int flushed; int queue; int error; bool unlock; flushed = 0; queue = QUEUE_DIRTY; bp = NULL; sentinel = malloc(sizeof(struct buf), M_TEMP, M_WAITOK | M_ZERO); sentinel->b_qindex = QUEUE_SENTINEL; mtx_lock(&bqlocks[queue]); TAILQ_INSERT_HEAD(&bufqueues[queue], sentinel, b_freelist); mtx_unlock(&bqlocks[queue]); while (flushed != target) { maybe_yield(); mtx_lock(&bqlocks[queue]); bp = TAILQ_NEXT(sentinel, b_freelist); if (bp != NULL) { TAILQ_REMOVE(&bufqueues[queue], sentinel, b_freelist); TAILQ_INSERT_AFTER(&bufqueues[queue], bp, sentinel, b_freelist); } else { mtx_unlock(&bqlocks[queue]); break; } /* * Skip sentinels inserted by other invocations of the * flushbufqueues(), taking care to not reorder them. * * Only flush the buffers that belong to the * vnode locked by the curthread. */ if (bp->b_qindex == QUEUE_SENTINEL || (lvp != NULL && bp->b_vp != lvp)) { mtx_unlock(&bqlocks[queue]); continue; } error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL); mtx_unlock(&bqlocks[queue]); if (error != 0) continue; if (bp->b_pin_count > 0) { BUF_UNLOCK(bp); continue; } /* * BKGRDINPROG can only be set with the buf and bufobj * locks both held. We tolerate a race to clear it here. */ if ((bp->b_vflags & BV_BKGRDINPROG) != 0 || (bp->b_flags & B_DELWRI) == 0) { BUF_UNLOCK(bp); continue; } if (bp->b_flags & B_INVAL) { bremfreef(bp); brelse(bp); flushed++; continue; } if (!LIST_EMPTY(&bp->b_dep) && buf_countdeps(bp, 0)) { if (flushdeps == 0) { BUF_UNLOCK(bp); continue; } hasdeps = 1; } else hasdeps = 0; /* * We must hold the lock on a vnode before writing * one of its buffers. Otherwise we may confuse, or * in the case of a snapshot vnode, deadlock the * system. * * The lock order here is the reverse of the normal * of vnode followed by buf lock. This is ok because * the NOWAIT will prevent deadlock. */ vp = bp->b_vp; if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { BUF_UNLOCK(bp); continue; } if (lvp == NULL) { unlock = true; error = vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT); } else { ASSERT_VOP_LOCKED(vp, "getbuf"); unlock = false; error = VOP_ISLOCKED(vp) == LK_EXCLUSIVE ? 0 : vn_lock(vp, LK_TRYUPGRADE); } if (error == 0) { CTR3(KTR_BUF, "flushbufqueue(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); if (curproc == bufdaemonproc) { vfs_bio_awrite(bp); } else { bremfree(bp); bwrite(bp); notbufdflushes++; } vn_finished_write(mp); if (unlock) VOP_UNLOCK(vp, 0); flushwithdeps += hasdeps; flushed++; /* * Sleeping on runningbufspace while holding * vnode lock leads to deadlock. */ if (curproc == bufdaemonproc && runningbufspace > hirunningspace) waitrunningbufspace(); continue; } vn_finished_write(mp); BUF_UNLOCK(bp); } mtx_lock(&bqlocks[queue]); TAILQ_REMOVE(&bufqueues[queue], sentinel, b_freelist); mtx_unlock(&bqlocks[queue]); free(sentinel, M_TEMP); return (flushed); } /* * Check to see if a block is currently memory resident. */ struct buf * incore(struct bufobj *bo, daddr_t blkno) { struct buf *bp; BO_RLOCK(bo); bp = gbincore(bo, blkno); BO_RUNLOCK(bo); return (bp); } /* * Returns true if no I/O is needed to access the * associated VM object. This is like incore except * it also hunts around in the VM system for the data. */ static int inmem(struct vnode * vp, daddr_t blkno) { vm_object_t obj; vm_offset_t toff, tinc, size; vm_page_t m; vm_ooffset_t off; ASSERT_VOP_LOCKED(vp, "inmem"); if (incore(&vp->v_bufobj, blkno)) return 1; if (vp->v_mount == NULL) return 0; obj = vp->v_object; if (obj == NULL) return (0); size = PAGE_SIZE; if (size > vp->v_mount->mnt_stat.f_iosize) size = vp->v_mount->mnt_stat.f_iosize; off = (vm_ooffset_t)blkno * (vm_ooffset_t)vp->v_mount->mnt_stat.f_iosize; VM_OBJECT_RLOCK(obj); for (toff = 0; toff < vp->v_mount->mnt_stat.f_iosize; toff += tinc) { m = vm_page_lookup(obj, OFF_TO_IDX(off + toff)); if (!m) goto notinmem; tinc = size; if (tinc > PAGE_SIZE - ((toff + off) & PAGE_MASK)) tinc = PAGE_SIZE - ((toff + off) & PAGE_MASK); if (vm_page_is_valid(m, (vm_offset_t) ((toff + off) & PAGE_MASK), tinc) == 0) goto notinmem; } VM_OBJECT_RUNLOCK(obj); return 1; notinmem: VM_OBJECT_RUNLOCK(obj); return (0); } /* * Set the dirty range for a buffer based on the status of the dirty * bits in the pages comprising the buffer. The range is limited * to the size of the buffer. * * Tell the VM system that the pages associated with this buffer * are clean. This is used for delayed writes where the data is * going to go to disk eventually without additional VM intevention. * * Note that while we only really need to clean through to b_bcount, we * just go ahead and clean through to b_bufsize. */ static void vfs_clean_pages_dirty_buf(struct buf *bp) { vm_ooffset_t foff, noff, eoff; vm_page_t m; int i; if ((bp->b_flags & B_VMIO) == 0 || bp->b_bufsize == 0) return; foff = bp->b_offset; KASSERT(bp->b_offset != NOOFFSET, ("vfs_clean_pages_dirty_buf: no buffer offset")); VM_OBJECT_WLOCK(bp->b_bufobj->bo_object); vfs_drain_busy_pages(bp); vfs_setdirty_locked_object(bp); for (i = 0; i < bp->b_npages; i++) { noff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK; eoff = noff; if (eoff > bp->b_offset + bp->b_bufsize) eoff = bp->b_offset + bp->b_bufsize; m = bp->b_pages[i]; vfs_page_set_validclean(bp, foff, m); /* vm_page_clear_dirty(m, foff & PAGE_MASK, eoff - foff); */ foff = noff; } VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object); } static void vfs_setdirty_locked_object(struct buf *bp) { vm_object_t object; int i; object = bp->b_bufobj->bo_object; VM_OBJECT_ASSERT_WLOCKED(object); /* * We qualify the scan for modified pages on whether the * object has been flushed yet. */ if ((object->flags & OBJ_MIGHTBEDIRTY) != 0) { vm_offset_t boffset; vm_offset_t eoffset; /* * test the pages to see if they have been modified directly * by users through the VM system. */ for (i = 0; i < bp->b_npages; i++) vm_page_test_dirty(bp->b_pages[i]); /* * Calculate the encompassing dirty range, boffset and eoffset, * (eoffset - boffset) bytes. */ for (i = 0; i < bp->b_npages; i++) { if (bp->b_pages[i]->dirty) break; } boffset = (i << PAGE_SHIFT) - (bp->b_offset & PAGE_MASK); for (i = bp->b_npages - 1; i >= 0; --i) { if (bp->b_pages[i]->dirty) { break; } } eoffset = ((i + 1) << PAGE_SHIFT) - (bp->b_offset & PAGE_MASK); /* * Fit it to the buffer. */ if (eoffset > bp->b_bcount) eoffset = bp->b_bcount; /* * If we have a good dirty range, merge with the existing * dirty range. */ if (boffset < eoffset) { if (bp->b_dirtyoff > boffset) bp->b_dirtyoff = boffset; if (bp->b_dirtyend < eoffset) bp->b_dirtyend = eoffset; } } } /* * Allocate the KVA mapping for an existing buffer. * If an unmapped buffer is provided but a mapped buffer is requested, take * also care to properly setup mappings between pages and KVA. */ static void bp_unmapped_get_kva(struct buf *bp, daddr_t blkno, int size, int gbflags) { int bsize, maxsize, need_mapping, need_kva; off_t offset; need_mapping = bp->b_data == unmapped_buf && (gbflags & GB_UNMAPPED) == 0; need_kva = bp->b_kvabase == unmapped_buf && bp->b_data == unmapped_buf && (gbflags & GB_KVAALLOC) != 0; if (!need_mapping && !need_kva) return; BUF_CHECK_UNMAPPED(bp); if (need_mapping && bp->b_kvabase != unmapped_buf) { /* * Buffer is not mapped, but the KVA was already * reserved at the time of the instantiation. Use the * allocated space. */ goto has_addr; } /* * Calculate the amount of the address space we would reserve * if the buffer was mapped. */ bsize = vn_isdisk(bp->b_vp, NULL) ? DEV_BSIZE : bp->b_bufobj->bo_bsize; KASSERT(bsize != 0, ("bsize == 0, check bo->bo_bsize")); offset = blkno * bsize; maxsize = size + (offset & PAGE_MASK); maxsize = imax(maxsize, bsize); while (bufkva_alloc(bp, maxsize, gbflags) != 0) { if ((gbflags & GB_NOWAIT_BD) != 0) { /* * XXXKIB: defragmentation cannot * succeed, not sure what else to do. */ panic("GB_NOWAIT_BD and GB_UNMAPPED %p", bp); } atomic_add_int(&mappingrestarts, 1); bufspace_wait(bp->b_vp, gbflags, 0, 0); } has_addr: if (need_mapping) { /* b_offset is handled by bpmap_qenter. */ bp->b_data = bp->b_kvabase; BUF_CHECK_MAPPED(bp); bpmap_qenter(bp); } } /* * getblk: * * Get a block given a specified block and offset into a file/device. * The buffers B_DONE bit will be cleared on return, making it almost * ready for an I/O initiation. B_INVAL may or may not be set on * return. The caller should clear B_INVAL prior to initiating a * READ. * * For a non-VMIO buffer, B_CACHE is set to the opposite of B_INVAL for * an existing buffer. * * For a VMIO buffer, B_CACHE is modified according to the backing VM. * If getblk()ing a previously 0-sized invalid buffer, B_CACHE is set * and then cleared based on the backing VM. If the previous buffer is * non-0-sized but invalid, B_CACHE will be cleared. * * If getblk() must create a new buffer, the new buffer is returned with * both B_INVAL and B_CACHE clear unless it is a VMIO buffer, in which * case it is returned with B_INVAL clear and B_CACHE set based on the * backing VM. * * getblk() also forces a bwrite() for any B_DELWRI buffer whos * B_CACHE bit is clear. * * What this means, basically, is that the caller should use B_CACHE to * determine whether the buffer is fully valid or not and should clear * B_INVAL prior to issuing a read. If the caller intends to validate * the buffer by loading its data area with something, the caller needs * to clear B_INVAL. If the caller does this without issuing an I/O, * the caller should set B_CACHE ( as an optimization ), else the caller * should issue the I/O and biodone() will set B_CACHE if the I/O was * a write attempt or if it was a successful read. If the caller * intends to issue a READ, the caller must clear B_INVAL and BIO_ERROR * prior to issuing the READ. biodone() will *not* clear B_INVAL. */ struct buf * getblk(struct vnode *vp, daddr_t blkno, int size, int slpflag, int slptimeo, int flags) { struct buf *bp; struct bufobj *bo; int bsize, error, maxsize, vmio; off_t offset; CTR3(KTR_BUF, "getblk(%p, %ld, %d)", vp, (long)blkno, size); KASSERT((flags & (GB_UNMAPPED | GB_KVAALLOC)) != GB_KVAALLOC, ("GB_KVAALLOC only makes sense with GB_UNMAPPED")); ASSERT_VOP_LOCKED(vp, "getblk"); if (size > maxbcachebuf) panic("getblk: size(%d) > maxbcachebuf(%d)\n", size, maxbcachebuf); if (!unmapped_buf_allowed) flags &= ~(GB_UNMAPPED | GB_KVAALLOC); bo = &vp->v_bufobj; loop: BO_RLOCK(bo); bp = gbincore(bo, blkno); if (bp != NULL) { int lockflags; /* * Buffer is in-core. If the buffer is not busy nor managed, * it must be on a queue. */ lockflags = LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK; if (flags & GB_LOCK_NOWAIT) lockflags |= LK_NOWAIT; error = BUF_TIMELOCK(bp, lockflags, BO_LOCKPTR(bo), "getblk", slpflag, slptimeo); /* * If we slept and got the lock we have to restart in case * the buffer changed identities. */ if (error == ENOLCK) goto loop; /* We timed out or were interrupted. */ else if (error) return (NULL); /* If recursed, assume caller knows the rules. */ else if (BUF_LOCKRECURSED(bp)) goto end; /* * The buffer is locked. B_CACHE is cleared if the buffer is * invalid. Otherwise, for a non-VMIO buffer, B_CACHE is set * and for a VMIO buffer B_CACHE is adjusted according to the * backing VM cache. */ if (bp->b_flags & B_INVAL) bp->b_flags &= ~B_CACHE; else if ((bp->b_flags & (B_VMIO | B_INVAL)) == 0) bp->b_flags |= B_CACHE; if (bp->b_flags & B_MANAGED) MPASS(bp->b_qindex == QUEUE_NONE); else bremfree(bp); /* * check for size inconsistencies for non-VMIO case. */ if (bp->b_bcount != size) { if ((bp->b_flags & B_VMIO) == 0 || (size > bp->b_kvasize)) { if (bp->b_flags & B_DELWRI) { /* * If buffer is pinned and caller does * not want sleep waiting for it to be * unpinned, bail out * */ if (bp->b_pin_count > 0) { if (flags & GB_LOCK_NOWAIT) { bqrelse(bp); return (NULL); } else { bunpin_wait(bp); } } bp->b_flags |= B_NOCACHE; bwrite(bp); } else { if (LIST_EMPTY(&bp->b_dep)) { bp->b_flags |= B_RELBUF; brelse(bp); } else { bp->b_flags |= B_NOCACHE; bwrite(bp); } } goto loop; } } /* * Handle the case of unmapped buffer which should * become mapped, or the buffer for which KVA * reservation is requested. */ bp_unmapped_get_kva(bp, blkno, size, flags); /* * If the size is inconsistent in the VMIO case, we can resize * the buffer. This might lead to B_CACHE getting set or * cleared. If the size has not changed, B_CACHE remains * unchanged from its previous state. */ allocbuf(bp, size); KASSERT(bp->b_offset != NOOFFSET, ("getblk: no buffer offset")); /* * A buffer with B_DELWRI set and B_CACHE clear must * be committed before we can return the buffer in * order to prevent the caller from issuing a read * ( due to B_CACHE not being set ) and overwriting * it. * * Most callers, including NFS and FFS, need this to * operate properly either because they assume they * can issue a read if B_CACHE is not set, or because * ( for example ) an uncached B_DELWRI might loop due * to softupdates re-dirtying the buffer. In the latter * case, B_CACHE is set after the first write completes, * preventing further loops. * NOTE! b*write() sets B_CACHE. If we cleared B_CACHE * above while extending the buffer, we cannot allow the * buffer to remain with B_CACHE set after the write * completes or it will represent a corrupt state. To * deal with this we set B_NOCACHE to scrap the buffer * after the write. * * We might be able to do something fancy, like setting * B_CACHE in bwrite() except if B_DELWRI is already set, * so the below call doesn't set B_CACHE, but that gets real * confusing. This is much easier. */ if ((bp->b_flags & (B_CACHE|B_DELWRI)) == B_DELWRI) { bp->b_flags |= B_NOCACHE; bwrite(bp); goto loop; } bp->b_flags &= ~B_DONE; } else { /* * Buffer is not in-core, create new buffer. The buffer * returned by getnewbuf() is locked. Note that the returned * buffer is also considered valid (not marked B_INVAL). */ BO_RUNLOCK(bo); /* * If the user does not want us to create the buffer, bail out * here. */ if (flags & GB_NOCREAT) return NULL; if (numfreebuffers == 0 && TD_IS_IDLETHREAD(curthread)) return NULL; bsize = vn_isdisk(vp, NULL) ? DEV_BSIZE : bo->bo_bsize; KASSERT(bsize != 0, ("bsize == 0, check bo->bo_bsize")); offset = blkno * bsize; vmio = vp->v_object != NULL; if (vmio) { maxsize = size + (offset & PAGE_MASK); } else { maxsize = size; /* Do not allow non-VMIO notmapped buffers. */ flags &= ~(GB_UNMAPPED | GB_KVAALLOC); } maxsize = imax(maxsize, bsize); bp = getnewbuf(vp, slpflag, slptimeo, maxsize, flags); if (bp == NULL) { if (slpflag || slptimeo) return NULL; /* * XXX This is here until the sleep path is diagnosed * enough to work under very low memory conditions. * * There's an issue on low memory, 4BSD+non-preempt * systems (eg MIPS routers with 32MB RAM) where buffer * exhaustion occurs without sleeping for buffer * reclaimation. This just sticks in a loop and * constantly attempts to allocate a buffer, which * hits exhaustion and tries to wakeup bufdaemon. * This never happens because we never yield. * * The real solution is to identify and fix these cases * so we aren't effectively busy-waiting in a loop * until the reclaimation path has cycles to run. */ kern_yield(PRI_USER); goto loop; } /* * This code is used to make sure that a buffer is not * created while the getnewbuf routine is blocked. * This can be a problem whether the vnode is locked or not. * If the buffer is created out from under us, we have to * throw away the one we just created. * * Note: this must occur before we associate the buffer * with the vp especially considering limitations in * the splay tree implementation when dealing with duplicate * lblkno's. */ BO_LOCK(bo); if (gbincore(bo, blkno)) { BO_UNLOCK(bo); bp->b_flags |= B_INVAL; brelse(bp); bufspace_release(maxsize); goto loop; } /* * Insert the buffer into the hash, so that it can * be found by incore. */ bp->b_blkno = bp->b_lblkno = blkno; bp->b_offset = offset; bgetvp(vp, bp); BO_UNLOCK(bo); /* * set B_VMIO bit. allocbuf() the buffer bigger. Since the * buffer size starts out as 0, B_CACHE will be set by * allocbuf() for the VMIO case prior to it testing the * backing store for validity. */ if (vmio) { bp->b_flags |= B_VMIO; KASSERT(vp->v_object == bp->b_bufobj->bo_object, ("ARGH! different b_bufobj->bo_object %p %p %p\n", bp, vp->v_object, bp->b_bufobj->bo_object)); } else { bp->b_flags &= ~B_VMIO; KASSERT(bp->b_bufobj->bo_object == NULL, ("ARGH! has b_bufobj->bo_object %p %p\n", bp, bp->b_bufobj->bo_object)); BUF_CHECK_MAPPED(bp); } allocbuf(bp, size); bufspace_release(maxsize); bp->b_flags &= ~B_DONE; } CTR4(KTR_BUF, "getblk(%p, %ld, %d) = %p", vp, (long)blkno, size, bp); BUF_ASSERT_HELD(bp); end: KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); return (bp); } /* * Get an empty, disassociated buffer of given size. The buffer is initially * set to B_INVAL. */ struct buf * geteblk(int size, int flags) { struct buf *bp; int maxsize; maxsize = (size + BKVAMASK) & ~BKVAMASK; while ((bp = getnewbuf(NULL, 0, 0, maxsize, flags)) == NULL) { if ((flags & GB_NOWAIT_BD) && (curthread->td_pflags & TDP_BUFNEED) != 0) return (NULL); } allocbuf(bp, size); bufspace_release(maxsize); bp->b_flags |= B_INVAL; /* b_dep cleared by getnewbuf() */ BUF_ASSERT_HELD(bp); return (bp); } /* * Truncate the backing store for a non-vmio buffer. */ static void vfs_nonvmio_truncate(struct buf *bp, int newbsize) { if (bp->b_flags & B_MALLOC) { /* * malloced buffers are not shrunk */ if (newbsize == 0) { bufmallocadjust(bp, 0); free(bp->b_data, M_BIOBUF); bp->b_data = bp->b_kvabase; bp->b_flags &= ~B_MALLOC; } return; } vm_hold_free_pages(bp, newbsize); bufspace_adjust(bp, newbsize); } /* * Extend the backing for a non-VMIO buffer. */ static void vfs_nonvmio_extend(struct buf *bp, int newbsize) { caddr_t origbuf; int origbufsize; /* * We only use malloced memory on the first allocation. * and revert to page-allocated memory when the buffer * grows. * * There is a potential smp race here that could lead * to bufmallocspace slightly passing the max. It * is probably extremely rare and not worth worrying * over. */ if (bp->b_bufsize == 0 && newbsize <= PAGE_SIZE/2 && bufmallocspace < maxbufmallocspace) { bp->b_data = malloc(newbsize, M_BIOBUF, M_WAITOK); bp->b_flags |= B_MALLOC; bufmallocadjust(bp, newbsize); return; } /* * If the buffer is growing on its other-than-first * allocation then we revert to the page-allocation * scheme. */ origbuf = NULL; origbufsize = 0; if (bp->b_flags & B_MALLOC) { origbuf = bp->b_data; origbufsize = bp->b_bufsize; bp->b_data = bp->b_kvabase; bufmallocadjust(bp, 0); bp->b_flags &= ~B_MALLOC; newbsize = round_page(newbsize); } vm_hold_load_pages(bp, (vm_offset_t) bp->b_data + bp->b_bufsize, (vm_offset_t) bp->b_data + newbsize); if (origbuf != NULL) { bcopy(origbuf, bp->b_data, origbufsize); free(origbuf, M_BIOBUF); } bufspace_adjust(bp, newbsize); } /* * This code constitutes the buffer memory from either anonymous system * memory (in the case of non-VMIO operations) or from an associated * VM object (in the case of VMIO operations). This code is able to * resize a buffer up or down. * * Note that this code is tricky, and has many complications to resolve * deadlock or inconsistent data situations. Tread lightly!!! * There are B_CACHE and B_DELWRI interactions that must be dealt with by * the caller. Calling this code willy nilly can result in the loss of data. * * allocbuf() only adjusts B_CACHE for VMIO buffers. getblk() deals with * B_CACHE for the non-VMIO case. */ int allocbuf(struct buf *bp, int size) { int newbsize; BUF_ASSERT_HELD(bp); if (bp->b_bcount == size) return (1); if (bp->b_kvasize != 0 && bp->b_kvasize < size) panic("allocbuf: buffer too small"); newbsize = roundup2(size, DEV_BSIZE); if ((bp->b_flags & B_VMIO) == 0) { if ((bp->b_flags & B_MALLOC) == 0) newbsize = round_page(newbsize); /* * Just get anonymous memory from the kernel. Don't * mess with B_CACHE. */ if (newbsize < bp->b_bufsize) vfs_nonvmio_truncate(bp, newbsize); else if (newbsize > bp->b_bufsize) vfs_nonvmio_extend(bp, newbsize); } else { int desiredpages; desiredpages = (size == 0) ? 0 : num_pages((bp->b_offset & PAGE_MASK) + newbsize); if (bp->b_flags & B_MALLOC) panic("allocbuf: VMIO buffer can't be malloced"); /* * Set B_CACHE initially if buffer is 0 length or will become * 0-length. */ if (size == 0 || bp->b_bufsize == 0) bp->b_flags |= B_CACHE; if (newbsize < bp->b_bufsize) vfs_vmio_truncate(bp, desiredpages); /* XXX This looks as if it should be newbsize > b_bufsize */ else if (size > bp->b_bcount) vfs_vmio_extend(bp, desiredpages, size); bufspace_adjust(bp, newbsize); } bp->b_bcount = size; /* requested buffer size. */ return (1); } extern int inflight_transient_maps; void biodone(struct bio *bp) { struct mtx *mtxp; void (*done)(struct bio *); vm_offset_t start, end; if ((bp->bio_flags & BIO_TRANSIENT_MAPPING) != 0) { bp->bio_flags &= ~BIO_TRANSIENT_MAPPING; bp->bio_flags |= BIO_UNMAPPED; start = trunc_page((vm_offset_t)bp->bio_data); end = round_page((vm_offset_t)bp->bio_data + bp->bio_length); bp->bio_data = unmapped_buf; pmap_qremove(start, atop(end - start)); vmem_free(transient_arena, start, end - start); atomic_add_int(&inflight_transient_maps, -1); } done = bp->bio_done; if (done == NULL) { mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); bp->bio_flags |= BIO_DONE; wakeup(bp); mtx_unlock(mtxp); } else done(bp); } /* * Wait for a BIO to finish. */ int biowait(struct bio *bp, const char *wchan) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); while ((bp->bio_flags & BIO_DONE) == 0) msleep(bp, mtxp, PRIBIO, wchan, 0); mtx_unlock(mtxp); if (bp->bio_error != 0) return (bp->bio_error); if (!(bp->bio_flags & BIO_ERROR)) return (0); return (EIO); } void biofinish(struct bio *bp, struct devstat *stat, int error) { if (error) { bp->bio_error = error; bp->bio_flags |= BIO_ERROR; } if (stat != NULL) devstat_end_transaction_bio(stat, bp); biodone(bp); } /* * bufwait: * * Wait for buffer I/O completion, returning error status. The buffer * is left locked and B_DONE on return. B_EINTR is converted into an EINTR * error and cleared. */ int bufwait(struct buf *bp) { if (bp->b_iocmd == BIO_READ) bwait(bp, PRIBIO, "biord"); else bwait(bp, PRIBIO, "biowr"); if (bp->b_flags & B_EINTR) { bp->b_flags &= ~B_EINTR; return (EINTR); } if (bp->b_ioflags & BIO_ERROR) { return (bp->b_error ? bp->b_error : EIO); } else { return (0); } } /* * bufdone: * * Finish I/O on a buffer, optionally calling a completion function. * This is usually called from an interrupt so process blocking is * not allowed. * * biodone is also responsible for setting B_CACHE in a B_VMIO bp. * In a non-VMIO bp, B_CACHE will be set on the next getblk() * assuming B_INVAL is clear. * * For the VMIO case, we set B_CACHE if the op was a read and no * read error occurred, or if the op was a write. B_CACHE is never * set if the buffer is invalid or otherwise uncacheable. * * biodone does not mess with B_INVAL, allowing the I/O routine or the * initiator to leave B_INVAL set to brelse the buffer out of existence * in the biodone routine. */ void bufdone(struct buf *bp) { struct bufobj *dropobj; void (*biodone)(struct buf *); CTR3(KTR_BUF, "bufdone(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); dropobj = NULL; KASSERT(!(bp->b_flags & B_DONE), ("biodone: bp %p already done", bp)); BUF_ASSERT_HELD(bp); runningbufwakeup(bp); if (bp->b_iocmd == BIO_WRITE) dropobj = bp->b_bufobj; /* call optional completion function if requested */ if (bp->b_iodone != NULL) { biodone = bp->b_iodone; bp->b_iodone = NULL; (*biodone) (bp); if (dropobj) bufobj_wdrop(dropobj); return; } bufdone_finish(bp); if (dropobj) bufobj_wdrop(dropobj); } void bufdone_finish(struct buf *bp) { BUF_ASSERT_HELD(bp); if (!LIST_EMPTY(&bp->b_dep)) buf_complete(bp); if (bp->b_flags & B_VMIO) { /* * Set B_CACHE if the op was a normal read and no error * occurred. B_CACHE is set for writes in the b*write() * routines. */ if (bp->b_iocmd == BIO_READ && !(bp->b_flags & (B_INVAL|B_NOCACHE)) && !(bp->b_ioflags & BIO_ERROR)) bp->b_flags |= B_CACHE; vfs_vmio_iodone(bp); } /* * For asynchronous completions, release the buffer now. The brelse * will do a wakeup there if necessary - so no need to do a wakeup * here in the async case. The sync case always needs to do a wakeup. */ if (bp->b_flags & B_ASYNC) { if ((bp->b_flags & (B_NOCACHE | B_INVAL | B_RELBUF)) || (bp->b_ioflags & BIO_ERROR)) brelse(bp); else bqrelse(bp); } else bdone(bp); } /* * This routine is called in lieu of iodone in the case of * incomplete I/O. This keeps the busy status for pages * consistent. */ void vfs_unbusy_pages(struct buf *bp) { int i; vm_object_t obj; vm_page_t m; runningbufwakeup(bp); if (!(bp->b_flags & B_VMIO)) return; obj = bp->b_bufobj->bo_object; VM_OBJECT_WLOCK(obj); for (i = 0; i < bp->b_npages; i++) { m = bp->b_pages[i]; if (m == bogus_page) { m = vm_page_lookup(obj, OFF_TO_IDX(bp->b_offset) + i); if (!m) panic("vfs_unbusy_pages: page missing\n"); bp->b_pages[i] = m; if (buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); pmap_qenter(trunc_page((vm_offset_t)bp->b_data), bp->b_pages, bp->b_npages); } else BUF_CHECK_UNMAPPED(bp); } vm_page_sunbusy(m); } vm_object_pip_wakeupn(obj, bp->b_npages); VM_OBJECT_WUNLOCK(obj); } /* * vfs_page_set_valid: * * Set the valid bits in a page based on the supplied offset. The * range is restricted to the buffer's size. * * This routine is typically called after a read completes. */ static void vfs_page_set_valid(struct buf *bp, vm_ooffset_t off, vm_page_t m) { vm_ooffset_t eoff; /* * Compute the end offset, eoff, such that [off, eoff) does not span a * page boundary and eoff is not greater than the end of the buffer. * The end of the buffer, in this case, is our file EOF, not the * allocation size of the buffer. */ eoff = (off + PAGE_SIZE) & ~(vm_ooffset_t)PAGE_MASK; if (eoff > bp->b_offset + bp->b_bcount) eoff = bp->b_offset + bp->b_bcount; /* * Set valid range. This is typically the entire buffer and thus the * entire page. */ if (eoff > off) vm_page_set_valid_range(m, off & PAGE_MASK, eoff - off); } /* * vfs_page_set_validclean: * * Set the valid bits and clear the dirty bits in a page based on the * supplied offset. The range is restricted to the buffer's size. */ static void vfs_page_set_validclean(struct buf *bp, vm_ooffset_t off, vm_page_t m) { vm_ooffset_t soff, eoff; /* * Start and end offsets in buffer. eoff - soff may not cross a * page boundary or cross the end of the buffer. The end of the * buffer, in this case, is our file EOF, not the allocation size * of the buffer. */ soff = off; eoff = (off + PAGE_SIZE) & ~(off_t)PAGE_MASK; if (eoff > bp->b_offset + bp->b_bcount) eoff = bp->b_offset + bp->b_bcount; /* * Set valid range. This is typically the entire buffer and thus the * entire page. */ if (eoff > soff) { vm_page_set_validclean( m, (vm_offset_t) (soff & PAGE_MASK), (vm_offset_t) (eoff - soff) ); } } /* * Ensure that all buffer pages are not exclusive busied. If any page is * exclusive busy, drain it. */ void vfs_drain_busy_pages(struct buf *bp) { vm_page_t m; int i, last_busied; VM_OBJECT_ASSERT_WLOCKED(bp->b_bufobj->bo_object); last_busied = 0; for (i = 0; i < bp->b_npages; i++) { m = bp->b_pages[i]; if (vm_page_xbusied(m)) { for (; last_busied < i; last_busied++) vm_page_sbusy(bp->b_pages[last_busied]); while (vm_page_xbusied(m)) { vm_page_lock(m); VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object); vm_page_busy_sleep(m, "vbpage", true); VM_OBJECT_WLOCK(bp->b_bufobj->bo_object); } } } for (i = 0; i < last_busied; i++) vm_page_sunbusy(bp->b_pages[i]); } /* * This routine is called before a device strategy routine. * It is used to tell the VM system that paging I/O is in * progress, and treat the pages associated with the buffer * almost as being exclusive busy. Also the object paging_in_progress * flag is handled to make sure that the object doesn't become * inconsistent. * * Since I/O has not been initiated yet, certain buffer flags * such as BIO_ERROR or B_INVAL may be in an inconsistent state * and should be ignored. */ void vfs_busy_pages(struct buf *bp, int clear_modify) { vm_object_t obj; vm_ooffset_t foff; vm_page_t m; int i; bool bogus; if (!(bp->b_flags & B_VMIO)) return; obj = bp->b_bufobj->bo_object; foff = bp->b_offset; KASSERT(bp->b_offset != NOOFFSET, ("vfs_busy_pages: no buffer offset")); VM_OBJECT_WLOCK(obj); vfs_drain_busy_pages(bp); if (bp->b_bufsize != 0) vfs_setdirty_locked_object(bp); bogus = false; for (i = 0; i < bp->b_npages; i++) { m = bp->b_pages[i]; if ((bp->b_flags & B_CLUSTER) == 0) { vm_object_pip_add(obj, 1); vm_page_sbusy(m); } /* * When readying a buffer for a read ( i.e * clear_modify == 0 ), it is important to do * bogus_page replacement for valid pages in * partially instantiated buffers. Partially * instantiated buffers can, in turn, occur when * reconstituting a buffer from its VM backing store * base. We only have to do this if B_CACHE is * clear ( which causes the I/O to occur in the * first place ). The replacement prevents the read * I/O from overwriting potentially dirty VM-backed * pages. XXX bogus page replacement is, uh, bogus. * It may not work properly with small-block devices. * We need to find a better way. */ if (clear_modify) { pmap_remove_write(m); vfs_page_set_validclean(bp, foff, m); } else if (m->valid == VM_PAGE_BITS_ALL && (bp->b_flags & B_CACHE) == 0) { bp->b_pages[i] = bogus_page; bogus = true; } foff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK; } VM_OBJECT_WUNLOCK(obj); if (bogus && buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); pmap_qenter(trunc_page((vm_offset_t)bp->b_data), bp->b_pages, bp->b_npages); } } /* * vfs_bio_set_valid: * * Set the range within the buffer to valid. The range is * relative to the beginning of the buffer, b_offset. Note that * b_offset itself may be offset from the beginning of the first * page. */ void vfs_bio_set_valid(struct buf *bp, int base, int size) { int i, n; vm_page_t m; if (!(bp->b_flags & B_VMIO)) return; /* * Fixup base to be relative to beginning of first page. * Set initial n to be the maximum number of bytes in the * first page that can be validated. */ base += (bp->b_offset & PAGE_MASK); n = PAGE_SIZE - (base & PAGE_MASK); VM_OBJECT_WLOCK(bp->b_bufobj->bo_object); for (i = base / PAGE_SIZE; size > 0 && i < bp->b_npages; ++i) { m = bp->b_pages[i]; if (n > size) n = size; vm_page_set_valid_range(m, base & PAGE_MASK, n); base += n; size -= n; n = PAGE_SIZE; } VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object); } /* * vfs_bio_clrbuf: * * If the specified buffer is a non-VMIO buffer, clear the entire * buffer. If the specified buffer is a VMIO buffer, clear and * validate only the previously invalid portions of the buffer. * This routine essentially fakes an I/O, so we need to clear * BIO_ERROR and B_INVAL. * * Note that while we only theoretically need to clear through b_bcount, * we go ahead and clear through b_bufsize. */ void vfs_bio_clrbuf(struct buf *bp) { int i, j, mask, sa, ea, slide; if ((bp->b_flags & (B_VMIO | B_MALLOC)) != B_VMIO) { clrbuf(bp); return; } bp->b_flags &= ~B_INVAL; bp->b_ioflags &= ~BIO_ERROR; VM_OBJECT_WLOCK(bp->b_bufobj->bo_object); if ((bp->b_npages == 1) && (bp->b_bufsize < PAGE_SIZE) && (bp->b_offset & PAGE_MASK) == 0) { if (bp->b_pages[0] == bogus_page) goto unlock; mask = (1 << (bp->b_bufsize / DEV_BSIZE)) - 1; VM_OBJECT_ASSERT_WLOCKED(bp->b_pages[0]->object); if ((bp->b_pages[0]->valid & mask) == mask) goto unlock; if ((bp->b_pages[0]->valid & mask) == 0) { pmap_zero_page_area(bp->b_pages[0], 0, bp->b_bufsize); bp->b_pages[0]->valid |= mask; goto unlock; } } sa = bp->b_offset & PAGE_MASK; slide = 0; for (i = 0; i < bp->b_npages; i++, sa = 0) { slide = imin(slide + PAGE_SIZE, bp->b_offset + bp->b_bufsize); ea = slide & PAGE_MASK; if (ea == 0) ea = PAGE_SIZE; if (bp->b_pages[i] == bogus_page) continue; j = sa / DEV_BSIZE; mask = ((1 << ((ea - sa) / DEV_BSIZE)) - 1) << j; VM_OBJECT_ASSERT_WLOCKED(bp->b_pages[i]->object); if ((bp->b_pages[i]->valid & mask) == mask) continue; if ((bp->b_pages[i]->valid & mask) == 0) pmap_zero_page_area(bp->b_pages[i], sa, ea - sa); else { for (; sa < ea; sa += DEV_BSIZE, j++) { if ((bp->b_pages[i]->valid & (1 << j)) == 0) { pmap_zero_page_area(bp->b_pages[i], sa, DEV_BSIZE); } } } bp->b_pages[i]->valid |= mask; } unlock: VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object); bp->b_resid = 0; } void vfs_bio_bzero_buf(struct buf *bp, int base, int size) { vm_page_t m; int i, n; if (buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); bzero(bp->b_data + base, size); } else { BUF_CHECK_UNMAPPED(bp); n = PAGE_SIZE - (base & PAGE_MASK); for (i = base / PAGE_SIZE; size > 0 && i < bp->b_npages; ++i) { m = bp->b_pages[i]; if (n > size) n = size; pmap_zero_page_area(m, base & PAGE_MASK, n); base += n; size -= n; n = PAGE_SIZE; } } } /* * vm_hold_load_pages and vm_hold_free_pages get pages into * a buffers address space. The pages are anonymous and are * not associated with a file object. */ static void vm_hold_load_pages(struct buf *bp, vm_offset_t from, vm_offset_t to) { vm_offset_t pg; vm_page_t p; int index; BUF_CHECK_MAPPED(bp); to = round_page(to); from = round_page(from); index = (from - trunc_page((vm_offset_t)bp->b_data)) >> PAGE_SHIFT; for (pg = from; pg < to; pg += PAGE_SIZE, index++) { tryagain: /* * note: must allocate system pages since blocking here * could interfere with paging I/O, no matter which * process we are. */ p = vm_page_alloc(NULL, 0, VM_ALLOC_SYSTEM | VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_COUNT((to - pg) >> PAGE_SHIFT)); if (p == NULL) { VM_WAIT; goto tryagain; } pmap_qenter(pg, &p, 1); bp->b_pages[index] = p; } bp->b_npages = index; } /* Return pages associated with this buf to the vm system */ static void vm_hold_free_pages(struct buf *bp, int newbsize) { vm_offset_t from; vm_page_t p; int index, newnpages; BUF_CHECK_MAPPED(bp); from = round_page((vm_offset_t)bp->b_data + newbsize); newnpages = (from - trunc_page((vm_offset_t)bp->b_data)) >> PAGE_SHIFT; if (bp->b_npages > newnpages) pmap_qremove(from, bp->b_npages - newnpages); for (index = newnpages; index < bp->b_npages; index++) { p = bp->b_pages[index]; bp->b_pages[index] = NULL; p->wire_count--; vm_page_free(p); } atomic_subtract_int(&vm_cnt.v_wire_count, bp->b_npages - newnpages); bp->b_npages = newnpages; } /* * Map an IO request into kernel virtual address space. * * All requests are (re)mapped into kernel VA space. * Notice that we use b_bufsize for the size of the buffer * to be mapped. b_bcount might be modified by the driver. * * Note that even if the caller determines that the address space should * be valid, a race or a smaller-file mapped into a larger space may * actually cause vmapbuf() to fail, so all callers of vmapbuf() MUST * check the return value. * * This function only works with pager buffers. */ int vmapbuf(struct buf *bp, int mapbuf) { vm_prot_t prot; int pidx; if (bp->b_bufsize < 0) return (-1); prot = VM_PROT_READ; if (bp->b_iocmd == BIO_READ) prot |= VM_PROT_WRITE; /* Less backwards than it looks */ if ((pidx = vm_fault_quick_hold_pages(&curproc->p_vmspace->vm_map, (vm_offset_t)bp->b_data, bp->b_bufsize, prot, bp->b_pages, btoc(MAXPHYS))) < 0) return (-1); bp->b_npages = pidx; bp->b_offset = ((vm_offset_t)bp->b_data) & PAGE_MASK; if (mapbuf || !unmapped_buf_allowed) { pmap_qenter((vm_offset_t)bp->b_kvabase, bp->b_pages, pidx); bp->b_data = bp->b_kvabase + bp->b_offset; } else bp->b_data = unmapped_buf; return(0); } /* * Free the io map PTEs associated with this IO operation. * We also invalidate the TLB entries and restore the original b_addr. * * This function only works with pager buffers. */ void vunmapbuf(struct buf *bp) { int npages; npages = bp->b_npages; if (buf_mapped(bp)) pmap_qremove(trunc_page((vm_offset_t)bp->b_data), npages); vm_page_unhold_pages(bp->b_pages, npages); bp->b_data = unmapped_buf; } void bdone(struct buf *bp) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); bp->b_flags |= B_DONE; wakeup(bp); mtx_unlock(mtxp); } void bwait(struct buf *bp, u_char pri, const char *wchan) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); while ((bp->b_flags & B_DONE) == 0) msleep(bp, mtxp, pri, wchan, 0); mtx_unlock(mtxp); } int bufsync(struct bufobj *bo, int waitfor) { return (VOP_FSYNC(bo->__bo_vnode, waitfor, curthread)); } void bufstrategy(struct bufobj *bo, struct buf *bp) { int i = 0; struct vnode *vp; vp = bp->b_vp; KASSERT(vp == bo->bo_private, ("Inconsistent vnode bufstrategy")); KASSERT(vp->v_type != VCHR && vp->v_type != VBLK, ("Wrong vnode in bufstrategy(bp=%p, vp=%p)", bp, vp)); i = VOP_STRATEGY(vp, bp); KASSERT(i == 0, ("VOP_STRATEGY failed bp=%p vp=%p", bp, bp->b_vp)); } void bufobj_wrefl(struct bufobj *bo) { KASSERT(bo != NULL, ("NULL bo in bufobj_wref")); ASSERT_BO_WLOCKED(bo); bo->bo_numoutput++; } void bufobj_wref(struct bufobj *bo) { KASSERT(bo != NULL, ("NULL bo in bufobj_wref")); BO_LOCK(bo); bo->bo_numoutput++; BO_UNLOCK(bo); } void bufobj_wdrop(struct bufobj *bo) { KASSERT(bo != NULL, ("NULL bo in bufobj_wdrop")); BO_LOCK(bo); KASSERT(bo->bo_numoutput > 0, ("bufobj_wdrop non-positive count")); if ((--bo->bo_numoutput == 0) && (bo->bo_flag & BO_WWAIT)) { bo->bo_flag &= ~BO_WWAIT; wakeup(&bo->bo_numoutput); } BO_UNLOCK(bo); } int bufobj_wwait(struct bufobj *bo, int slpflag, int timeo) { int error; KASSERT(bo != NULL, ("NULL bo in bufobj_wwait")); ASSERT_BO_WLOCKED(bo); error = 0; while (bo->bo_numoutput) { bo->bo_flag |= BO_WWAIT; error = msleep(&bo->bo_numoutput, BO_LOCKPTR(bo), slpflag | (PRIBIO + 1), "bo_wwait", timeo); if (error) break; } return (error); } void bpin(struct buf *bp) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); bp->b_pin_count++; mtx_unlock(mtxp); } void bunpin(struct buf *bp) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); if (--bp->b_pin_count == 0) wakeup(bp); mtx_unlock(mtxp); } void bunpin_wait(struct buf *bp) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); while (bp->b_pin_count > 0) msleep(bp, mtxp, PRIBIO, "bwunpin", 0); mtx_unlock(mtxp); } /* * Set bio_data or bio_ma for struct bio from the struct buf. */ void bdata2bio(struct buf *bp, struct bio *bip) { if (!buf_mapped(bp)) { KASSERT(unmapped_buf_allowed, ("unmapped")); bip->bio_ma = bp->b_pages; bip->bio_ma_n = bp->b_npages; bip->bio_data = unmapped_buf; bip->bio_ma_offset = (vm_offset_t)bp->b_offset & PAGE_MASK; bip->bio_flags |= BIO_UNMAPPED; KASSERT(round_page(bip->bio_ma_offset + bip->bio_length) / PAGE_SIZE == bp->b_npages, ("Buffer %p too short: %d %lld %d", bp, bip->bio_ma_offset, (long long)bip->bio_length, bip->bio_ma_n)); } else { bip->bio_data = bp->b_data; bip->bio_ma = NULL; } } static int buf_pager_relbuf; SYSCTL_INT(_vfs, OID_AUTO, buf_pager_relbuf, CTLFLAG_RWTUN, &buf_pager_relbuf, 0, "Make buffer pager release buffers after reading"); /* * The buffer pager. It uses buffer reads to validate pages. * * In contrast to the generic local pager from vm/vnode_pager.c, this * pager correctly and easily handles volumes where the underlying * device block size is greater than the machine page size. The * buffer cache transparently extends the requested page run to be * aligned at the block boundary, and does the necessary bogus page * replacements in the addends to avoid obliterating already valid * pages. * * The only non-trivial issue is that the exclusive busy state for * pages, which is assumed by the vm_pager_getpages() interface, is * incompatible with the VMIO buffer cache's desire to share-busy the * pages. This function performs a trivial downgrade of the pages' * state before reading buffers, and a less trivial upgrade from the * shared-busy to excl-busy state after the read. */ int vfs_bio_getpages(struct vnode *vp, vm_page_t *ma, int count, int *rbehind, int *rahead, vbg_get_lblkno_t get_lblkno, vbg_get_blksize_t get_blksize) { vm_page_t m; vm_object_t object; struct buf *bp; struct mount *mp; daddr_t lbn, lbnp; vm_ooffset_t la, lb, poff, poffe; long bsize; int bo_bs, br_flags, error, i, pgsin, pgsin_a, pgsin_b; bool redo, lpart; object = vp->v_object; mp = vp->v_mount; la = IDX_TO_OFF(ma[count - 1]->pindex); if (la >= object->un_pager.vnp.vnp_size) return (VM_PAGER_BAD); lpart = la + PAGE_SIZE > object->un_pager.vnp.vnp_size; bo_bs = get_blksize(vp, get_lblkno(vp, IDX_TO_OFF(ma[0]->pindex))); /* * Calculate read-ahead, behind and total pages. */ pgsin = count; lb = IDX_TO_OFF(ma[0]->pindex); pgsin_b = OFF_TO_IDX(lb - rounddown2(lb, bo_bs)); pgsin += pgsin_b; if (rbehind != NULL) *rbehind = pgsin_b; pgsin_a = OFF_TO_IDX(roundup2(la, bo_bs) - la); if (la + IDX_TO_OFF(pgsin_a) >= object->un_pager.vnp.vnp_size) pgsin_a = OFF_TO_IDX(roundup2(object->un_pager.vnp.vnp_size, PAGE_SIZE) - la); pgsin += pgsin_a; if (rahead != NULL) *rahead = pgsin_a; PCPU_INC(cnt.v_vnodein); PCPU_ADD(cnt.v_vnodepgsin, pgsin); br_flags = (mp != NULL && (mp->mnt_kern_flag & MNTK_UNMAPPED_BUFS) != 0) ? GB_UNMAPPED : 0; VM_OBJECT_WLOCK(object); again: for (i = 0; i < count; i++) vm_page_busy_downgrade(ma[i]); VM_OBJECT_WUNLOCK(object); lbnp = -1; for (i = 0; i < count; i++) { m = ma[i]; /* * Pages are shared busy and the object lock is not * owned, which together allow for the pages' * invalidation. The racy test for validity avoids * useless creation of the buffer for the most typical * case when invalidation is not used in redo or for * parallel read. The shared->excl upgrade loop at * the end of the function catches the race in a * reliable way (protected by the object lock). */ if (m->valid == VM_PAGE_BITS_ALL) continue; poff = IDX_TO_OFF(m->pindex); poffe = MIN(poff + PAGE_SIZE, object->un_pager.vnp.vnp_size); for (; poff < poffe; poff += bsize) { lbn = get_lblkno(vp, poff); if (lbn == lbnp) goto next_page; lbnp = lbn; bsize = get_blksize(vp, lbn); error = bread_gb(vp, lbn, bsize, curthread->td_ucred, br_flags, &bp); if (error != 0) goto end_pages; if (LIST_EMPTY(&bp->b_dep)) { /* * Invalidation clears m->valid, but * may leave B_CACHE flag if the * buffer existed at the invalidation * time. In this case, recycle the * buffer to do real read on next * bread() after redo. * * Otherwise B_RELBUF is not strictly * necessary, enable to reduce buf * cache pressure. */ if (buf_pager_relbuf || m->valid != VM_PAGE_BITS_ALL) bp->b_flags |= B_RELBUF; bp->b_flags &= ~B_NOCACHE; brelse(bp); } else { bqrelse(bp); } } KASSERT(1 /* racy, enable for debugging */ || m->valid == VM_PAGE_BITS_ALL || i == count - 1, ("buf %d %p invalid", i, m)); if (i == count - 1 && lpart) { VM_OBJECT_WLOCK(object); if (m->valid != 0 && m->valid != VM_PAGE_BITS_ALL) vm_page_zero_invalid(m, TRUE); VM_OBJECT_WUNLOCK(object); } next_page:; } end_pages: VM_OBJECT_WLOCK(object); redo = false; for (i = 0; i < count; i++) { vm_page_sunbusy(ma[i]); ma[i] = vm_page_grab(object, ma[i]->pindex, VM_ALLOC_NORMAL); /* * Since the pages were only sbusy while neither the * buffer nor the object lock was held by us, or * reallocated while vm_page_grab() slept for busy * relinguish, they could have been invalidated. * Recheck the valid bits and re-read as needed. * * Note that the last page is made fully valid in the * read loop, and partial validity for the page at * index count - 1 could mean that the page was * invalidated or removed, so we must restart for * safety as well. */ if (ma[i]->valid != VM_PAGE_BITS_ALL) redo = true; } if (redo && error == 0) goto again; VM_OBJECT_WUNLOCK(object); return (error != 0 ? VM_PAGER_ERROR : VM_PAGER_OK); } #include "opt_ddb.h" #ifdef DDB #include /* DDB command to show buffer data */ DB_SHOW_COMMAND(buffer, db_show_buffer) { /* get args */ struct buf *bp = (struct buf *)addr; if (!have_addr) { db_printf("usage: show buffer \n"); return; } db_printf("buf at %p\n", bp); db_printf("b_flags = 0x%b, b_xflags=0x%b, b_vflags=0x%b\n", (u_int)bp->b_flags, PRINT_BUF_FLAGS, (u_int)bp->b_xflags, PRINT_BUF_XFLAGS, (u_int)bp->b_vflags, PRINT_BUF_VFLAGS); db_printf( "b_error = %d, b_bufsize = %ld, b_bcount = %ld, b_resid = %ld\n" "b_bufobj = (%p), b_data = %p, b_blkno = %jd, b_lblkno = %jd, " "b_dep = %p\n", bp->b_error, bp->b_bufsize, bp->b_bcount, bp->b_resid, bp->b_bufobj, bp->b_data, (intmax_t)bp->b_blkno, (intmax_t)bp->b_lblkno, bp->b_dep.lh_first); db_printf("b_kvabase = %p, b_kvasize = %d\n", bp->b_kvabase, bp->b_kvasize); if (bp->b_npages) { int i; db_printf("b_npages = %d, pages(OBJ, IDX, PA): ", bp->b_npages); for (i = 0; i < bp->b_npages; i++) { vm_page_t m; m = bp->b_pages[i]; db_printf("(%p, 0x%lx, 0x%lx)", (void *)m->object, (u_long)m->pindex, (u_long)VM_PAGE_TO_PHYS(m)); if ((i + 1) < bp->b_npages) db_printf(","); } db_printf("\n"); } db_printf(" "); BUF_LOCKPRINTINFO(bp); } DB_SHOW_COMMAND(lockedbufs, lockedbufs) { struct buf *bp; int i; for (i = 0; i < nbuf; i++) { bp = &buf[i]; if (BUF_ISLOCKED(bp)) { db_show_buffer((uintptr_t)bp, 1, 0, NULL); db_printf("\n"); if (db_pager_quit) break; } } } DB_SHOW_COMMAND(vnodebufs, db_show_vnodebufs) { struct vnode *vp; struct buf *bp; if (!have_addr) { db_printf("usage: show vnodebufs \n"); return; } vp = (struct vnode *)addr; db_printf("Clean buffers:\n"); TAILQ_FOREACH(bp, &vp->v_bufobj.bo_clean.bv_hd, b_bobufs) { db_show_buffer((uintptr_t)bp, 1, 0, NULL); db_printf("\n"); } db_printf("Dirty buffers:\n"); TAILQ_FOREACH(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs) { db_show_buffer((uintptr_t)bp, 1, 0, NULL); db_printf("\n"); } } DB_COMMAND(countfreebufs, db_coundfreebufs) { struct buf *bp; int i, used = 0, nfree = 0; if (have_addr) { db_printf("usage: countfreebufs\n"); return; } for (i = 0; i < nbuf; i++) { bp = &buf[i]; if (bp->b_qindex == QUEUE_EMPTY) nfree++; else used++; } db_printf("Counted %d free, %d used (%d tot)\n", nfree, used, nfree + used); db_printf("numfreebuffers is %d\n", numfreebuffers); } #endif /* DDB */ Index: stable/11/sys/vm/uma_core.c =================================================================== --- stable/11/sys/vm/uma_core.c (revision 327403) +++ stable/11/sys/vm/uma_core.c (revision 327404) @@ -1,3689 +1,3689 @@ /*- * Copyright (c) 2002-2005, 2009, 2013 Jeffrey Roberson * Copyright (c) 2004, 2005 Bosko Milekic * Copyright (c) 2004-2006 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 unmodified, 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 ``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. */ /* * uma_core.c Implementation of the Universal Memory allocator * * This allocator is intended to replace the multitude of similar object caches * in the standard FreeBSD kernel. The intent is to be flexible as well as * efficient. A primary design goal is to return unused memory to the rest of * the system. This will make the system as a whole more flexible due to the * ability to move memory to subsystems which most need it instead of leaving * pools of reserved memory unused. * * The basic ideas stem from similar slab/zone based allocators whose algorithms * are well known. * */ /* * TODO: * - Improve memory usage for large allocations * - Investigate cache size adjustments */ #include __FBSDID("$FreeBSD$"); /* I should really use ktr.. */ /* #define UMA_DEBUG 1 #define UMA_DEBUG_ALLOC 1 #define UMA_DEBUG_ALLOC_1 1 */ #include "opt_ddb.h" #include "opt_param.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 #include #include #include #include #include #include #include #include #ifdef DEBUG_MEMGUARD #include #endif /* * This is the zone and keg from which all zones are spawned. The idea is that * even the zone & keg heads are allocated from the allocator, so we use the * bss section to bootstrap us. */ static struct uma_keg masterkeg; static struct uma_zone masterzone_k; static struct uma_zone masterzone_z; static uma_zone_t kegs = &masterzone_k; static uma_zone_t zones = &masterzone_z; /* This is the zone from which all of uma_slab_t's are allocated. */ static uma_zone_t slabzone; /* * The initial hash tables come out of this zone so they can be allocated * prior to malloc coming up. */ static uma_zone_t hashzone; /* The boot-time adjusted value for cache line alignment. */ int uma_align_cache = 64 - 1; static MALLOC_DEFINE(M_UMAHASH, "UMAHash", "UMA Hash Buckets"); /* * Are we allowed to allocate buckets? */ static int bucketdisable = 1; /* Linked list of all kegs in the system */ static LIST_HEAD(,uma_keg) uma_kegs = LIST_HEAD_INITIALIZER(uma_kegs); /* Linked list of all cache-only zones in the system */ static LIST_HEAD(,uma_zone) uma_cachezones = LIST_HEAD_INITIALIZER(uma_cachezones); /* This RW lock protects the keg list */ -static struct rwlock_padalign uma_rwlock; +static struct rwlock_padalign __exclusive_cache_line uma_rwlock; /* Linked list of boot time pages */ static LIST_HEAD(,uma_slab) uma_boot_pages = LIST_HEAD_INITIALIZER(uma_boot_pages); /* This mutex protects the boot time pages list */ static struct mtx_padalign uma_boot_pages_mtx; static struct sx uma_drain_lock; /* Is the VM done starting up? */ static int booted = 0; #define UMA_STARTUP 1 #define UMA_STARTUP2 2 /* * This is the handle used to schedule events that need to happen * outside of the allocation fast path. */ static struct callout uma_callout; #define UMA_TIMEOUT 20 /* Seconds for callout interval. */ /* * This structure is passed as the zone ctor arg so that I don't have to create * a special allocation function just for zones. */ struct uma_zctor_args { const char *name; size_t size; uma_ctor ctor; uma_dtor dtor; uma_init uminit; uma_fini fini; uma_import import; uma_release release; void *arg; uma_keg_t keg; int align; uint32_t flags; }; struct uma_kctor_args { uma_zone_t zone; size_t size; uma_init uminit; uma_fini fini; int align; uint32_t flags; }; struct uma_bucket_zone { uma_zone_t ubz_zone; char *ubz_name; int ubz_entries; /* Number of items it can hold. */ int ubz_maxsize; /* Maximum allocation size per-item. */ }; /* * Compute the actual number of bucket entries to pack them in power * of two sizes for more efficient space utilization. */ #define BUCKET_SIZE(n) \ (((sizeof(void *) * (n)) - sizeof(struct uma_bucket)) / sizeof(void *)) #define BUCKET_MAX BUCKET_SIZE(256) struct uma_bucket_zone bucket_zones[] = { { NULL, "4 Bucket", BUCKET_SIZE(4), 4096 }, { NULL, "6 Bucket", BUCKET_SIZE(6), 3072 }, { NULL, "8 Bucket", BUCKET_SIZE(8), 2048 }, { NULL, "12 Bucket", BUCKET_SIZE(12), 1536 }, { NULL, "16 Bucket", BUCKET_SIZE(16), 1024 }, { NULL, "32 Bucket", BUCKET_SIZE(32), 512 }, { NULL, "64 Bucket", BUCKET_SIZE(64), 256 }, { NULL, "128 Bucket", BUCKET_SIZE(128), 128 }, { NULL, "256 Bucket", BUCKET_SIZE(256), 64 }, { NULL, NULL, 0} }; /* * Flags and enumerations to be passed to internal functions. */ enum zfreeskip { SKIP_NONE = 0, SKIP_DTOR, SKIP_FINI }; /* Prototypes.. */ static void *noobj_alloc(uma_zone_t, vm_size_t, uint8_t *, int); static void *page_alloc(uma_zone_t, vm_size_t, uint8_t *, int); static void *startup_alloc(uma_zone_t, vm_size_t, uint8_t *, int); static void page_free(void *, vm_size_t, uint8_t); static uma_slab_t keg_alloc_slab(uma_keg_t, uma_zone_t, int); static void cache_drain(uma_zone_t); static void bucket_drain(uma_zone_t, uma_bucket_t); static void bucket_cache_drain(uma_zone_t zone); static int keg_ctor(void *, int, void *, int); static void keg_dtor(void *, int, void *); static int zone_ctor(void *, int, void *, int); static void zone_dtor(void *, int, void *); static int zero_init(void *, int, int); static void keg_small_init(uma_keg_t keg); static void keg_large_init(uma_keg_t keg); static void zone_foreach(void (*zfunc)(uma_zone_t)); static void zone_timeout(uma_zone_t zone); static int hash_alloc(struct uma_hash *); static int hash_expand(struct uma_hash *, struct uma_hash *); static void hash_free(struct uma_hash *hash); static void uma_timeout(void *); static void uma_startup3(void); static void *zone_alloc_item(uma_zone_t, void *, int); static void zone_free_item(uma_zone_t, void *, void *, enum zfreeskip); static void bucket_enable(void); static void bucket_init(void); static uma_bucket_t bucket_alloc(uma_zone_t zone, void *, int); static void bucket_free(uma_zone_t zone, uma_bucket_t, void *); static void bucket_zone_drain(void); static uma_bucket_t zone_alloc_bucket(uma_zone_t zone, void *, int flags); static uma_slab_t zone_fetch_slab(uma_zone_t zone, uma_keg_t last, int flags); static uma_slab_t zone_fetch_slab_multi(uma_zone_t zone, uma_keg_t last, int flags); static void *slab_alloc_item(uma_keg_t keg, uma_slab_t slab); static void slab_free_item(uma_keg_t keg, uma_slab_t slab, void *item); static uma_keg_t uma_kcreate(uma_zone_t zone, size_t size, uma_init uminit, uma_fini fini, int align, uint32_t flags); static int zone_import(uma_zone_t zone, void **bucket, int max, int flags); static void zone_release(uma_zone_t zone, void **bucket, int cnt); static void uma_zero_item(void *item, uma_zone_t zone); void uma_print_zone(uma_zone_t); void uma_print_stats(void); static int sysctl_vm_zone_count(SYSCTL_HANDLER_ARGS); static int sysctl_vm_zone_stats(SYSCTL_HANDLER_ARGS); #ifdef INVARIANTS static void uma_dbg_free(uma_zone_t zone, uma_slab_t slab, void *item); static void uma_dbg_alloc(uma_zone_t zone, uma_slab_t slab, void *item); #endif SYSINIT(uma_startup3, SI_SUB_VM_CONF, SI_ORDER_SECOND, uma_startup3, NULL); SYSCTL_PROC(_vm, OID_AUTO, zone_count, CTLFLAG_RD|CTLTYPE_INT, 0, 0, sysctl_vm_zone_count, "I", "Number of UMA zones"); SYSCTL_PROC(_vm, OID_AUTO, zone_stats, CTLFLAG_RD|CTLTYPE_STRUCT, 0, 0, sysctl_vm_zone_stats, "s,struct uma_type_header", "Zone Stats"); static int zone_warnings = 1; SYSCTL_INT(_vm, OID_AUTO, zone_warnings, CTLFLAG_RWTUN, &zone_warnings, 0, "Warn when UMA zones becomes full"); /* * This routine checks to see whether or not it's safe to enable buckets. */ static void bucket_enable(void) { bucketdisable = vm_page_count_min(); } /* * Initialize bucket_zones, the array of zones of buckets of various sizes. * * For each zone, calculate the memory required for each bucket, consisting * of the header and an array of pointers. */ static void bucket_init(void) { struct uma_bucket_zone *ubz; int size; for (ubz = &bucket_zones[0]; ubz->ubz_entries != 0; ubz++) { size = roundup(sizeof(struct uma_bucket), sizeof(void *)); size += sizeof(void *) * ubz->ubz_entries; ubz->ubz_zone = uma_zcreate(ubz->ubz_name, size, NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_MTXCLASS | UMA_ZFLAG_BUCKET); } } /* * Given a desired number of entries for a bucket, return the zone from which * to allocate the bucket. */ static struct uma_bucket_zone * bucket_zone_lookup(int entries) { struct uma_bucket_zone *ubz; for (ubz = &bucket_zones[0]; ubz->ubz_entries != 0; ubz++) if (ubz->ubz_entries >= entries) return (ubz); ubz--; return (ubz); } static int bucket_select(int size) { struct uma_bucket_zone *ubz; ubz = &bucket_zones[0]; if (size > ubz->ubz_maxsize) return MAX((ubz->ubz_maxsize * ubz->ubz_entries) / size, 1); for (; ubz->ubz_entries != 0; ubz++) if (ubz->ubz_maxsize < size) break; ubz--; return (ubz->ubz_entries); } static uma_bucket_t bucket_alloc(uma_zone_t zone, void *udata, int flags) { struct uma_bucket_zone *ubz; uma_bucket_t bucket; /* * This is to stop us from allocating per cpu buckets while we're * running out of vm.boot_pages. Otherwise, we would exhaust the * boot pages. This also prevents us from allocating buckets in * low memory situations. */ if (bucketdisable) return (NULL); /* * To limit bucket recursion we store the original zone flags * in a cookie passed via zalloc_arg/zfree_arg. This allows the * NOVM flag to persist even through deep recursions. We also * store ZFLAG_BUCKET once we have recursed attempting to allocate * a bucket for a bucket zone so we do not allow infinite bucket * recursion. This cookie will even persist to frees of unused * buckets via the allocation path or bucket allocations in the * free path. */ if ((zone->uz_flags & UMA_ZFLAG_BUCKET) == 0) udata = (void *)(uintptr_t)zone->uz_flags; else { if ((uintptr_t)udata & UMA_ZFLAG_BUCKET) return (NULL); udata = (void *)((uintptr_t)udata | UMA_ZFLAG_BUCKET); } if ((uintptr_t)udata & UMA_ZFLAG_CACHEONLY) flags |= M_NOVM; ubz = bucket_zone_lookup(zone->uz_count); if (ubz->ubz_zone == zone && (ubz + 1)->ubz_entries != 0) ubz++; bucket = uma_zalloc_arg(ubz->ubz_zone, udata, flags); if (bucket) { #ifdef INVARIANTS bzero(bucket->ub_bucket, sizeof(void *) * ubz->ubz_entries); #endif bucket->ub_cnt = 0; bucket->ub_entries = ubz->ubz_entries; } return (bucket); } static void bucket_free(uma_zone_t zone, uma_bucket_t bucket, void *udata) { struct uma_bucket_zone *ubz; KASSERT(bucket->ub_cnt == 0, ("bucket_free: Freeing a non free bucket.")); if ((zone->uz_flags & UMA_ZFLAG_BUCKET) == 0) udata = (void *)(uintptr_t)zone->uz_flags; ubz = bucket_zone_lookup(bucket->ub_entries); uma_zfree_arg(ubz->ubz_zone, bucket, udata); } static void bucket_zone_drain(void) { struct uma_bucket_zone *ubz; for (ubz = &bucket_zones[0]; ubz->ubz_entries != 0; ubz++) zone_drain(ubz->ubz_zone); } static void zone_log_warning(uma_zone_t zone) { static const struct timeval warninterval = { 300, 0 }; if (!zone_warnings || zone->uz_warning == NULL) return; if (ratecheck(&zone->uz_ratecheck, &warninterval)) printf("[zone: %s] %s\n", zone->uz_name, zone->uz_warning); } static inline void zone_maxaction(uma_zone_t zone) { if (zone->uz_maxaction.ta_func != NULL) taskqueue_enqueue(taskqueue_thread, &zone->uz_maxaction); } static void zone_foreach_keg(uma_zone_t zone, void (*kegfn)(uma_keg_t)) { uma_klink_t klink; LIST_FOREACH(klink, &zone->uz_kegs, kl_link) kegfn(klink->kl_keg); } /* * Routine called by timeout which is used to fire off some time interval * based calculations. (stats, hash size, etc.) * * Arguments: * arg Unused * * Returns: * Nothing */ static void uma_timeout(void *unused) { bucket_enable(); zone_foreach(zone_timeout); /* Reschedule this event */ callout_reset(&uma_callout, UMA_TIMEOUT * hz, uma_timeout, NULL); } /* * Routine to perform timeout driven calculations. This expands the * hashes and does per cpu statistics aggregation. * * Returns nothing. */ static void keg_timeout(uma_keg_t keg) { KEG_LOCK(keg); /* * Expand the keg hash table. * * This is done if the number of slabs is larger than the hash size. * What I'm trying to do here is completely reduce collisions. This * may be a little aggressive. Should I allow for two collisions max? */ if (keg->uk_flags & UMA_ZONE_HASH && keg->uk_pages / keg->uk_ppera >= keg->uk_hash.uh_hashsize) { struct uma_hash newhash; struct uma_hash oldhash; int ret; /* * This is so involved because allocating and freeing * while the keg lock is held will lead to deadlock. * I have to do everything in stages and check for * races. */ newhash = keg->uk_hash; KEG_UNLOCK(keg); ret = hash_alloc(&newhash); KEG_LOCK(keg); if (ret) { if (hash_expand(&keg->uk_hash, &newhash)) { oldhash = keg->uk_hash; keg->uk_hash = newhash; } else oldhash = newhash; KEG_UNLOCK(keg); hash_free(&oldhash); return; } } KEG_UNLOCK(keg); } static void zone_timeout(uma_zone_t zone) { zone_foreach_keg(zone, &keg_timeout); } /* * Allocate and zero fill the next sized hash table from the appropriate * backing store. * * Arguments: * hash A new hash structure with the old hash size in uh_hashsize * * Returns: * 1 on success and 0 on failure. */ static int hash_alloc(struct uma_hash *hash) { int oldsize; int alloc; oldsize = hash->uh_hashsize; /* We're just going to go to a power of two greater */ if (oldsize) { hash->uh_hashsize = oldsize * 2; alloc = sizeof(hash->uh_slab_hash[0]) * hash->uh_hashsize; hash->uh_slab_hash = (struct slabhead *)malloc(alloc, M_UMAHASH, M_NOWAIT); } else { alloc = sizeof(hash->uh_slab_hash[0]) * UMA_HASH_SIZE_INIT; hash->uh_slab_hash = zone_alloc_item(hashzone, NULL, M_WAITOK); hash->uh_hashsize = UMA_HASH_SIZE_INIT; } if (hash->uh_slab_hash) { bzero(hash->uh_slab_hash, alloc); hash->uh_hashmask = hash->uh_hashsize - 1; return (1); } return (0); } /* * Expands the hash table for HASH zones. This is done from zone_timeout * to reduce collisions. This must not be done in the regular allocation * path, otherwise, we can recurse on the vm while allocating pages. * * Arguments: * oldhash The hash you want to expand * newhash The hash structure for the new table * * Returns: * Nothing * * Discussion: */ static int hash_expand(struct uma_hash *oldhash, struct uma_hash *newhash) { uma_slab_t slab; int hval; int i; if (!newhash->uh_slab_hash) return (0); if (oldhash->uh_hashsize >= newhash->uh_hashsize) return (0); /* * I need to investigate hash algorithms for resizing without a * full rehash. */ for (i = 0; i < oldhash->uh_hashsize; i++) while (!SLIST_EMPTY(&oldhash->uh_slab_hash[i])) { slab = SLIST_FIRST(&oldhash->uh_slab_hash[i]); SLIST_REMOVE_HEAD(&oldhash->uh_slab_hash[i], us_hlink); hval = UMA_HASH(newhash, slab->us_data); SLIST_INSERT_HEAD(&newhash->uh_slab_hash[hval], slab, us_hlink); } return (1); } /* * Free the hash bucket to the appropriate backing store. * * Arguments: * slab_hash The hash bucket we're freeing * hashsize The number of entries in that hash bucket * * Returns: * Nothing */ static void hash_free(struct uma_hash *hash) { if (hash->uh_slab_hash == NULL) return; if (hash->uh_hashsize == UMA_HASH_SIZE_INIT) zone_free_item(hashzone, hash->uh_slab_hash, NULL, SKIP_NONE); else free(hash->uh_slab_hash, M_UMAHASH); } /* * Frees all outstanding items in a bucket * * Arguments: * zone The zone to free to, must be unlocked. * bucket The free/alloc bucket with items, cpu queue must be locked. * * Returns: * Nothing */ static void bucket_drain(uma_zone_t zone, uma_bucket_t bucket) { int i; if (bucket == NULL) return; if (zone->uz_fini) for (i = 0; i < bucket->ub_cnt; i++) zone->uz_fini(bucket->ub_bucket[i], zone->uz_size); zone->uz_release(zone->uz_arg, bucket->ub_bucket, bucket->ub_cnt); bucket->ub_cnt = 0; } /* * Drains the per cpu caches for a zone. * * NOTE: This may only be called while the zone is being turn down, and not * during normal operation. This is necessary in order that we do not have * to migrate CPUs to drain the per-CPU caches. * * Arguments: * zone The zone to drain, must be unlocked. * * Returns: * Nothing */ static void cache_drain(uma_zone_t zone) { uma_cache_t cache; int cpu; /* * XXX: It is safe to not lock the per-CPU caches, because we're * tearing down the zone anyway. I.e., there will be no further use * of the caches at this point. * * XXX: It would good to be able to assert that the zone is being * torn down to prevent improper use of cache_drain(). * * XXX: We lock the zone before passing into bucket_cache_drain() as * it is used elsewhere. Should the tear-down path be made special * there in some form? */ CPU_FOREACH(cpu) { cache = &zone->uz_cpu[cpu]; bucket_drain(zone, cache->uc_allocbucket); bucket_drain(zone, cache->uc_freebucket); if (cache->uc_allocbucket != NULL) bucket_free(zone, cache->uc_allocbucket, NULL); if (cache->uc_freebucket != NULL) bucket_free(zone, cache->uc_freebucket, NULL); cache->uc_allocbucket = cache->uc_freebucket = NULL; } ZONE_LOCK(zone); bucket_cache_drain(zone); ZONE_UNLOCK(zone); } static void cache_shrink(uma_zone_t zone) { if (zone->uz_flags & UMA_ZFLAG_INTERNAL) return; ZONE_LOCK(zone); zone->uz_count = (zone->uz_count_min + zone->uz_count) / 2; ZONE_UNLOCK(zone); } static void cache_drain_safe_cpu(uma_zone_t zone) { uma_cache_t cache; uma_bucket_t b1, b2; if (zone->uz_flags & UMA_ZFLAG_INTERNAL) return; b1 = b2 = NULL; ZONE_LOCK(zone); critical_enter(); cache = &zone->uz_cpu[curcpu]; if (cache->uc_allocbucket) { if (cache->uc_allocbucket->ub_cnt != 0) LIST_INSERT_HEAD(&zone->uz_buckets, cache->uc_allocbucket, ub_link); else b1 = cache->uc_allocbucket; cache->uc_allocbucket = NULL; } if (cache->uc_freebucket) { if (cache->uc_freebucket->ub_cnt != 0) LIST_INSERT_HEAD(&zone->uz_buckets, cache->uc_freebucket, ub_link); else b2 = cache->uc_freebucket; cache->uc_freebucket = NULL; } critical_exit(); ZONE_UNLOCK(zone); if (b1) bucket_free(zone, b1, NULL); if (b2) bucket_free(zone, b2, NULL); } /* * Safely drain per-CPU caches of a zone(s) to alloc bucket. * This is an expensive call because it needs to bind to all CPUs * one by one and enter a critical section on each of them in order * to safely access their cache buckets. * Zone lock must not be held on call this function. */ static void cache_drain_safe(uma_zone_t zone) { int cpu; /* * Polite bucket sizes shrinking was not enouth, shrink aggressively. */ if (zone) cache_shrink(zone); else zone_foreach(cache_shrink); CPU_FOREACH(cpu) { thread_lock(curthread); sched_bind(curthread, cpu); thread_unlock(curthread); if (zone) cache_drain_safe_cpu(zone); else zone_foreach(cache_drain_safe_cpu); } thread_lock(curthread); sched_unbind(curthread); thread_unlock(curthread); } /* * Drain the cached buckets from a zone. Expects a locked zone on entry. */ static void bucket_cache_drain(uma_zone_t zone) { uma_bucket_t bucket; /* * Drain the bucket queues and free the buckets, we just keep two per * cpu (alloc/free). */ while ((bucket = LIST_FIRST(&zone->uz_buckets)) != NULL) { LIST_REMOVE(bucket, ub_link); ZONE_UNLOCK(zone); bucket_drain(zone, bucket); bucket_free(zone, bucket, NULL); ZONE_LOCK(zone); } /* * Shrink further bucket sizes. Price of single zone lock collision * is probably lower then price of global cache drain. */ if (zone->uz_count > zone->uz_count_min) zone->uz_count--; } static void keg_free_slab(uma_keg_t keg, uma_slab_t slab, int start) { uint8_t *mem; int i; uint8_t flags; mem = slab->us_data; flags = slab->us_flags; i = start; if (keg->uk_fini != NULL) { for (i--; i > -1; i--) keg->uk_fini(slab->us_data + (keg->uk_rsize * i), keg->uk_size); } if (keg->uk_flags & UMA_ZONE_OFFPAGE) zone_free_item(keg->uk_slabzone, slab, NULL, SKIP_NONE); #ifdef UMA_DEBUG printf("%s: Returning %d bytes.\n", keg->uk_name, PAGE_SIZE * keg->uk_ppera); #endif keg->uk_freef(mem, PAGE_SIZE * keg->uk_ppera, flags); } /* * Frees pages from a keg back to the system. This is done on demand from * the pageout daemon. * * Returns nothing. */ static void keg_drain(uma_keg_t keg) { struct slabhead freeslabs = { 0 }; uma_slab_t slab, tmp; /* * We don't want to take pages from statically allocated kegs at this * time */ if (keg->uk_flags & UMA_ZONE_NOFREE || keg->uk_freef == NULL) return; #ifdef UMA_DEBUG printf("%s free items: %u\n", keg->uk_name, keg->uk_free); #endif KEG_LOCK(keg); if (keg->uk_free == 0) goto finished; LIST_FOREACH_SAFE(slab, &keg->uk_free_slab, us_link, tmp) { /* We have nowhere to free these to. */ if (slab->us_flags & UMA_SLAB_BOOT) continue; LIST_REMOVE(slab, us_link); keg->uk_pages -= keg->uk_ppera; keg->uk_free -= keg->uk_ipers; if (keg->uk_flags & UMA_ZONE_HASH) UMA_HASH_REMOVE(&keg->uk_hash, slab, slab->us_data); SLIST_INSERT_HEAD(&freeslabs, slab, us_hlink); } finished: KEG_UNLOCK(keg); while ((slab = SLIST_FIRST(&freeslabs)) != NULL) { SLIST_REMOVE(&freeslabs, slab, uma_slab, us_hlink); keg_free_slab(keg, slab, keg->uk_ipers); } } static void zone_drain_wait(uma_zone_t zone, int waitok) { /* * Set draining to interlock with zone_dtor() so we can release our * locks as we go. Only dtor() should do a WAITOK call since it * is the only call that knows the structure will still be available * when it wakes up. */ ZONE_LOCK(zone); while (zone->uz_flags & UMA_ZFLAG_DRAINING) { if (waitok == M_NOWAIT) goto out; msleep(zone, zone->uz_lockptr, PVM, "zonedrain", 1); } zone->uz_flags |= UMA_ZFLAG_DRAINING; bucket_cache_drain(zone); ZONE_UNLOCK(zone); /* * The DRAINING flag protects us from being freed while * we're running. Normally the uma_rwlock would protect us but we * must be able to release and acquire the right lock for each keg. */ zone_foreach_keg(zone, &keg_drain); ZONE_LOCK(zone); zone->uz_flags &= ~UMA_ZFLAG_DRAINING; wakeup(zone); out: ZONE_UNLOCK(zone); } void zone_drain(uma_zone_t zone) { zone_drain_wait(zone, M_NOWAIT); } /* * Allocate a new slab for a keg. This does not insert the slab onto a list. * * Arguments: * wait Shall we wait? * * Returns: * The slab that was allocated or NULL if there is no memory and the * caller specified M_NOWAIT. */ static uma_slab_t keg_alloc_slab(uma_keg_t keg, uma_zone_t zone, int wait) { uma_alloc allocf; uma_slab_t slab; uint8_t *mem; uint8_t flags; int i; mtx_assert(&keg->uk_lock, MA_OWNED); slab = NULL; mem = NULL; #ifdef UMA_DEBUG printf("alloc_slab: Allocating a new slab for %s\n", keg->uk_name); #endif allocf = keg->uk_allocf; KEG_UNLOCK(keg); if (keg->uk_flags & UMA_ZONE_OFFPAGE) { slab = zone_alloc_item(keg->uk_slabzone, NULL, wait); if (slab == NULL) goto out; } /* * This reproduces the old vm_zone behavior of zero filling pages the * first time they are added to a zone. * * Malloced items are zeroed in uma_zalloc. */ if ((keg->uk_flags & UMA_ZONE_MALLOC) == 0) wait |= M_ZERO; else wait &= ~M_ZERO; if (keg->uk_flags & UMA_ZONE_NODUMP) wait |= M_NODUMP; /* zone is passed for legacy reasons. */ mem = allocf(zone, keg->uk_ppera * PAGE_SIZE, &flags, wait); if (mem == NULL) { if (keg->uk_flags & UMA_ZONE_OFFPAGE) zone_free_item(keg->uk_slabzone, slab, NULL, SKIP_NONE); slab = NULL; goto out; } /* Point the slab into the allocated memory */ if (!(keg->uk_flags & UMA_ZONE_OFFPAGE)) slab = (uma_slab_t )(mem + keg->uk_pgoff); if (keg->uk_flags & UMA_ZONE_VTOSLAB) for (i = 0; i < keg->uk_ppera; i++) vsetslab((vm_offset_t)mem + (i * PAGE_SIZE), slab); slab->us_keg = keg; slab->us_data = mem; slab->us_freecount = keg->uk_ipers; slab->us_flags = flags; BIT_FILL(SLAB_SETSIZE, &slab->us_free); #ifdef INVARIANTS BIT_ZERO(SLAB_SETSIZE, &slab->us_debugfree); #endif if (keg->uk_init != NULL) { for (i = 0; i < keg->uk_ipers; i++) if (keg->uk_init(slab->us_data + (keg->uk_rsize * i), keg->uk_size, wait) != 0) break; if (i != keg->uk_ipers) { keg_free_slab(keg, slab, i); slab = NULL; goto out; } } out: KEG_LOCK(keg); if (slab != NULL) { if (keg->uk_flags & UMA_ZONE_HASH) UMA_HASH_INSERT(&keg->uk_hash, slab, mem); keg->uk_pages += keg->uk_ppera; keg->uk_free += keg->uk_ipers; } return (slab); } /* * This function is intended to be used early on in place of page_alloc() so * that we may use the boot time page cache to satisfy allocations before * the VM is ready. */ static void * startup_alloc(uma_zone_t zone, vm_size_t bytes, uint8_t *pflag, int wait) { uma_keg_t keg; uma_slab_t tmps; int pages, check_pages; keg = zone_first_keg(zone); pages = howmany(bytes, PAGE_SIZE); check_pages = pages - 1; KASSERT(pages > 0, ("startup_alloc can't reserve 0 pages\n")); /* * Check our small startup cache to see if it has pages remaining. */ mtx_lock(&uma_boot_pages_mtx); /* First check if we have enough room. */ tmps = LIST_FIRST(&uma_boot_pages); while (tmps != NULL && check_pages-- > 0) tmps = LIST_NEXT(tmps, us_link); if (tmps != NULL) { /* * It's ok to lose tmps references. The last one will * have tmps->us_data pointing to the start address of * "pages" contiguous pages of memory. */ while (pages-- > 0) { tmps = LIST_FIRST(&uma_boot_pages); LIST_REMOVE(tmps, us_link); } mtx_unlock(&uma_boot_pages_mtx); *pflag = tmps->us_flags; return (tmps->us_data); } mtx_unlock(&uma_boot_pages_mtx); if (booted < UMA_STARTUP2) panic("UMA: Increase vm.boot_pages"); /* * Now that we've booted reset these users to their real allocator. */ #ifdef UMA_MD_SMALL_ALLOC keg->uk_allocf = (keg->uk_ppera > 1) ? page_alloc : uma_small_alloc; #else keg->uk_allocf = page_alloc; #endif return keg->uk_allocf(zone, bytes, pflag, wait); } /* * Allocates a number of pages from the system * * Arguments: * bytes The number of bytes requested * wait Shall we wait? * * Returns: * A pointer to the alloced memory or possibly * NULL if M_NOWAIT is set. */ static void * page_alloc(uma_zone_t zone, vm_size_t bytes, uint8_t *pflag, int wait) { void *p; /* Returned page */ *pflag = UMA_SLAB_KMEM; p = (void *) kmem_malloc(kmem_arena, bytes, wait); return (p); } /* * Allocates a number of pages from within an object * * Arguments: * bytes The number of bytes requested * wait Shall we wait? * * Returns: * A pointer to the alloced memory or possibly * NULL if M_NOWAIT is set. */ static void * noobj_alloc(uma_zone_t zone, vm_size_t bytes, uint8_t *flags, int wait) { TAILQ_HEAD(, vm_page) alloctail; u_long npages; vm_offset_t retkva, zkva; vm_page_t p, p_next; uma_keg_t keg; TAILQ_INIT(&alloctail); keg = zone_first_keg(zone); npages = howmany(bytes, PAGE_SIZE); while (npages > 0) { p = vm_page_alloc(NULL, 0, VM_ALLOC_INTERRUPT | VM_ALLOC_WIRED | VM_ALLOC_NOOBJ); if (p != NULL) { /* * Since the page does not belong to an object, its * listq is unused. */ TAILQ_INSERT_TAIL(&alloctail, p, listq); npages--; continue; } if (wait & M_WAITOK) { VM_WAIT; continue; } /* * Page allocation failed, free intermediate pages and * exit. */ TAILQ_FOREACH_SAFE(p, &alloctail, listq, p_next) { vm_page_unwire(p, PQ_NONE); vm_page_free(p); } return (NULL); } *flags = UMA_SLAB_PRIV; zkva = keg->uk_kva + atomic_fetchadd_long(&keg->uk_offset, round_page(bytes)); retkva = zkva; TAILQ_FOREACH(p, &alloctail, listq) { pmap_qenter(zkva, &p, 1); zkva += PAGE_SIZE; } return ((void *)retkva); } /* * Frees a number of pages to the system * * Arguments: * mem A pointer to the memory to be freed * size The size of the memory being freed * flags The original p->us_flags field * * Returns: * Nothing */ static void page_free(void *mem, vm_size_t size, uint8_t flags) { struct vmem *vmem; if (flags & UMA_SLAB_KMEM) vmem = kmem_arena; else if (flags & UMA_SLAB_KERNEL) vmem = kernel_arena; else panic("UMA: page_free used with invalid flags %d", flags); kmem_free(vmem, (vm_offset_t)mem, size); } /* * Zero fill initializer * * Arguments/Returns follow uma_init specifications */ static int zero_init(void *mem, int size, int flags) { bzero(mem, size); return (0); } /* * Finish creating a small uma keg. This calculates ipers, and the keg size. * * Arguments * keg The zone we should initialize * * Returns * Nothing */ static void keg_small_init(uma_keg_t keg) { u_int rsize; u_int memused; u_int wastedspace; u_int shsize; u_int slabsize; if (keg->uk_flags & UMA_ZONE_PCPU) { u_int ncpus = (mp_maxid + 1) ? (mp_maxid + 1) : MAXCPU; slabsize = sizeof(struct pcpu); keg->uk_ppera = howmany(ncpus * sizeof(struct pcpu), PAGE_SIZE); } else { slabsize = UMA_SLAB_SIZE; keg->uk_ppera = 1; } /* * Calculate the size of each allocation (rsize) according to * alignment. If the requested size is smaller than we have * allocation bits for we round it up. */ rsize = keg->uk_size; if (rsize < slabsize / SLAB_SETSIZE) rsize = slabsize / SLAB_SETSIZE; if (rsize & keg->uk_align) rsize = (rsize & ~keg->uk_align) + (keg->uk_align + 1); keg->uk_rsize = rsize; KASSERT((keg->uk_flags & UMA_ZONE_PCPU) == 0 || keg->uk_rsize < sizeof(struct pcpu), ("%s: size %u too large", __func__, keg->uk_rsize)); if (keg->uk_flags & UMA_ZONE_OFFPAGE) shsize = 0; else shsize = sizeof(struct uma_slab); keg->uk_ipers = (slabsize - shsize) / rsize; KASSERT(keg->uk_ipers > 0 && keg->uk_ipers <= SLAB_SETSIZE, ("%s: keg->uk_ipers %u", __func__, keg->uk_ipers)); memused = keg->uk_ipers * rsize + shsize; wastedspace = slabsize - memused; /* * We can't do OFFPAGE if we're internal or if we've been * asked to not go to the VM for buckets. If we do this we * may end up going to the VM for slabs which we do not * want to do if we're UMA_ZFLAG_CACHEONLY as a result * of UMA_ZONE_VM, which clearly forbids it. */ if ((keg->uk_flags & UMA_ZFLAG_INTERNAL) || (keg->uk_flags & UMA_ZFLAG_CACHEONLY)) return; /* * See if using an OFFPAGE slab will limit our waste. Only do * this if it permits more items per-slab. * * XXX We could try growing slabsize to limit max waste as well. * Historically this was not done because the VM could not * efficiently handle contiguous allocations. */ if ((wastedspace >= slabsize / UMA_MAX_WASTE) && (keg->uk_ipers < (slabsize / keg->uk_rsize))) { keg->uk_ipers = slabsize / keg->uk_rsize; KASSERT(keg->uk_ipers > 0 && keg->uk_ipers <= SLAB_SETSIZE, ("%s: keg->uk_ipers %u", __func__, keg->uk_ipers)); #ifdef UMA_DEBUG printf("UMA decided we need offpage slab headers for " "keg: %s, calculated wastedspace = %d, " "maximum wasted space allowed = %d, " "calculated ipers = %d, " "new wasted space = %d\n", keg->uk_name, wastedspace, slabsize / UMA_MAX_WASTE, keg->uk_ipers, slabsize - keg->uk_ipers * keg->uk_rsize); #endif keg->uk_flags |= UMA_ZONE_OFFPAGE; } if ((keg->uk_flags & UMA_ZONE_OFFPAGE) && (keg->uk_flags & UMA_ZONE_VTOSLAB) == 0) keg->uk_flags |= UMA_ZONE_HASH; } /* * Finish creating a large (> UMA_SLAB_SIZE) uma kegs. Just give in and do * OFFPAGE for now. When I can allow for more dynamic slab sizes this will be * more complicated. * * Arguments * keg The keg we should initialize * * Returns * Nothing */ static void keg_large_init(uma_keg_t keg) { u_int shsize; KASSERT(keg != NULL, ("Keg is null in keg_large_init")); KASSERT((keg->uk_flags & UMA_ZFLAG_CACHEONLY) == 0, ("keg_large_init: Cannot large-init a UMA_ZFLAG_CACHEONLY keg")); KASSERT((keg->uk_flags & UMA_ZONE_PCPU) == 0, ("%s: Cannot large-init a UMA_ZONE_PCPU keg", __func__)); keg->uk_ppera = howmany(keg->uk_size, PAGE_SIZE); keg->uk_ipers = 1; keg->uk_rsize = keg->uk_size; /* Check whether we have enough space to not do OFFPAGE. */ if ((keg->uk_flags & UMA_ZONE_OFFPAGE) == 0) { shsize = sizeof(struct uma_slab); if (shsize & UMA_ALIGN_PTR) shsize = (shsize & ~UMA_ALIGN_PTR) + (UMA_ALIGN_PTR + 1); if (PAGE_SIZE * keg->uk_ppera - keg->uk_rsize < shsize) { /* * We can't do OFFPAGE if we're internal, in which case * we need an extra page per allocation to contain the * slab header. */ if ((keg->uk_flags & UMA_ZFLAG_INTERNAL) == 0) keg->uk_flags |= UMA_ZONE_OFFPAGE; else keg->uk_ppera++; } } if ((keg->uk_flags & UMA_ZONE_OFFPAGE) && (keg->uk_flags & UMA_ZONE_VTOSLAB) == 0) keg->uk_flags |= UMA_ZONE_HASH; } static void keg_cachespread_init(uma_keg_t keg) { int alignsize; int trailer; int pages; int rsize; KASSERT((keg->uk_flags & UMA_ZONE_PCPU) == 0, ("%s: Cannot cachespread-init a UMA_ZONE_PCPU keg", __func__)); alignsize = keg->uk_align + 1; rsize = keg->uk_size; /* * We want one item to start on every align boundary in a page. To * do this we will span pages. We will also extend the item by the * size of align if it is an even multiple of align. Otherwise, it * would fall on the same boundary every time. */ if (rsize & keg->uk_align) rsize = (rsize & ~keg->uk_align) + alignsize; if ((rsize & alignsize) == 0) rsize += alignsize; trailer = rsize - keg->uk_size; pages = (rsize * (PAGE_SIZE / alignsize)) / PAGE_SIZE; pages = MIN(pages, (128 * 1024) / PAGE_SIZE); keg->uk_rsize = rsize; keg->uk_ppera = pages; keg->uk_ipers = ((pages * PAGE_SIZE) + trailer) / rsize; keg->uk_flags |= UMA_ZONE_OFFPAGE | UMA_ZONE_VTOSLAB; KASSERT(keg->uk_ipers <= SLAB_SETSIZE, ("%s: keg->uk_ipers too high(%d) increase max_ipers", __func__, keg->uk_ipers)); } /* * Keg header ctor. This initializes all fields, locks, etc. And inserts * the keg onto the global keg list. * * Arguments/Returns follow uma_ctor specifications * udata Actually uma_kctor_args */ static int keg_ctor(void *mem, int size, void *udata, int flags) { struct uma_kctor_args *arg = udata; uma_keg_t keg = mem; uma_zone_t zone; bzero(keg, size); keg->uk_size = arg->size; keg->uk_init = arg->uminit; keg->uk_fini = arg->fini; keg->uk_align = arg->align; keg->uk_free = 0; keg->uk_reserve = 0; keg->uk_pages = 0; keg->uk_flags = arg->flags; keg->uk_allocf = page_alloc; keg->uk_freef = page_free; keg->uk_slabzone = NULL; /* * The master zone is passed to us at keg-creation time. */ zone = arg->zone; keg->uk_name = zone->uz_name; if (arg->flags & UMA_ZONE_VM) keg->uk_flags |= UMA_ZFLAG_CACHEONLY; if (arg->flags & UMA_ZONE_ZINIT) keg->uk_init = zero_init; if (arg->flags & UMA_ZONE_MALLOC) keg->uk_flags |= UMA_ZONE_VTOSLAB; if (arg->flags & UMA_ZONE_PCPU) #ifdef SMP keg->uk_flags |= UMA_ZONE_OFFPAGE; #else keg->uk_flags &= ~UMA_ZONE_PCPU; #endif if (keg->uk_flags & UMA_ZONE_CACHESPREAD) { keg_cachespread_init(keg); } else { if (keg->uk_size > (UMA_SLAB_SIZE - sizeof(struct uma_slab))) keg_large_init(keg); else keg_small_init(keg); } if (keg->uk_flags & UMA_ZONE_OFFPAGE) keg->uk_slabzone = slabzone; /* * If we haven't booted yet we need allocations to go through the * startup cache until the vm is ready. */ if (keg->uk_ppera == 1) { #ifdef UMA_MD_SMALL_ALLOC keg->uk_allocf = uma_small_alloc; keg->uk_freef = uma_small_free; if (booted < UMA_STARTUP) keg->uk_allocf = startup_alloc; #else if (booted < UMA_STARTUP2) keg->uk_allocf = startup_alloc; #endif } else if (booted < UMA_STARTUP2 && (keg->uk_flags & UMA_ZFLAG_INTERNAL)) keg->uk_allocf = startup_alloc; /* * Initialize keg's lock */ KEG_LOCK_INIT(keg, (arg->flags & UMA_ZONE_MTXCLASS)); /* * If we're putting the slab header in the actual page we need to * figure out where in each page it goes. This calculates a right * justified offset into the memory on an ALIGN_PTR boundary. */ if (!(keg->uk_flags & UMA_ZONE_OFFPAGE)) { u_int totsize; /* Size of the slab struct and free list */ totsize = sizeof(struct uma_slab); if (totsize & UMA_ALIGN_PTR) totsize = (totsize & ~UMA_ALIGN_PTR) + (UMA_ALIGN_PTR + 1); keg->uk_pgoff = (PAGE_SIZE * keg->uk_ppera) - totsize; /* * The only way the following is possible is if with our * UMA_ALIGN_PTR adjustments we are now bigger than * UMA_SLAB_SIZE. I haven't checked whether this is * mathematically possible for all cases, so we make * sure here anyway. */ totsize = keg->uk_pgoff + sizeof(struct uma_slab); if (totsize > PAGE_SIZE * keg->uk_ppera) { printf("zone %s ipers %d rsize %d size %d\n", zone->uz_name, keg->uk_ipers, keg->uk_rsize, keg->uk_size); panic("UMA slab won't fit."); } } if (keg->uk_flags & UMA_ZONE_HASH) hash_alloc(&keg->uk_hash); #ifdef UMA_DEBUG printf("UMA: %s(%p) size %d(%d) flags %#x ipers %d ppera %d out %d free %d\n", zone->uz_name, zone, keg->uk_size, keg->uk_rsize, keg->uk_flags, keg->uk_ipers, keg->uk_ppera, (keg->uk_pages / keg->uk_ppera) * keg->uk_ipers - keg->uk_free, keg->uk_free); #endif LIST_INSERT_HEAD(&keg->uk_zones, zone, uz_link); rw_wlock(&uma_rwlock); LIST_INSERT_HEAD(&uma_kegs, keg, uk_link); rw_wunlock(&uma_rwlock); return (0); } /* * Zone header ctor. This initializes all fields, locks, etc. * * Arguments/Returns follow uma_ctor specifications * udata Actually uma_zctor_args */ static int zone_ctor(void *mem, int size, void *udata, int flags) { struct uma_zctor_args *arg = udata; uma_zone_t zone = mem; uma_zone_t z; uma_keg_t keg; bzero(zone, size); zone->uz_name = arg->name; zone->uz_ctor = arg->ctor; zone->uz_dtor = arg->dtor; zone->uz_slab = zone_fetch_slab; zone->uz_init = NULL; zone->uz_fini = NULL; zone->uz_allocs = 0; zone->uz_frees = 0; zone->uz_fails = 0; zone->uz_sleeps = 0; zone->uz_count = 0; zone->uz_count_min = 0; zone->uz_flags = 0; zone->uz_warning = NULL; timevalclear(&zone->uz_ratecheck); keg = arg->keg; ZONE_LOCK_INIT(zone, (arg->flags & UMA_ZONE_MTXCLASS)); /* * This is a pure cache zone, no kegs. */ if (arg->import) { if (arg->flags & UMA_ZONE_VM) arg->flags |= UMA_ZFLAG_CACHEONLY; zone->uz_flags = arg->flags; zone->uz_size = arg->size; zone->uz_import = arg->import; zone->uz_release = arg->release; zone->uz_arg = arg->arg; zone->uz_lockptr = &zone->uz_lock; rw_wlock(&uma_rwlock); LIST_INSERT_HEAD(&uma_cachezones, zone, uz_link); rw_wunlock(&uma_rwlock); goto out; } /* * Use the regular zone/keg/slab allocator. */ zone->uz_import = (uma_import)zone_import; zone->uz_release = (uma_release)zone_release; zone->uz_arg = zone; if (arg->flags & UMA_ZONE_SECONDARY) { KASSERT(arg->keg != NULL, ("Secondary zone on zero'd keg")); zone->uz_init = arg->uminit; zone->uz_fini = arg->fini; zone->uz_lockptr = &keg->uk_lock; zone->uz_flags |= UMA_ZONE_SECONDARY; rw_wlock(&uma_rwlock); ZONE_LOCK(zone); LIST_FOREACH(z, &keg->uk_zones, uz_link) { if (LIST_NEXT(z, uz_link) == NULL) { LIST_INSERT_AFTER(z, zone, uz_link); break; } } ZONE_UNLOCK(zone); rw_wunlock(&uma_rwlock); } else if (keg == NULL) { if ((keg = uma_kcreate(zone, arg->size, arg->uminit, arg->fini, arg->align, arg->flags)) == NULL) return (ENOMEM); } else { struct uma_kctor_args karg; int error; /* We should only be here from uma_startup() */ karg.size = arg->size; karg.uminit = arg->uminit; karg.fini = arg->fini; karg.align = arg->align; karg.flags = arg->flags; karg.zone = zone; error = keg_ctor(arg->keg, sizeof(struct uma_keg), &karg, flags); if (error) return (error); } /* * Link in the first keg. */ zone->uz_klink.kl_keg = keg; LIST_INSERT_HEAD(&zone->uz_kegs, &zone->uz_klink, kl_link); zone->uz_lockptr = &keg->uk_lock; zone->uz_size = keg->uk_size; zone->uz_flags |= (keg->uk_flags & (UMA_ZONE_INHERIT | UMA_ZFLAG_INHERIT)); /* * Some internal zones don't have room allocated for the per cpu * caches. If we're internal, bail out here. */ if (keg->uk_flags & UMA_ZFLAG_INTERNAL) { KASSERT((zone->uz_flags & UMA_ZONE_SECONDARY) == 0, ("Secondary zone requested UMA_ZFLAG_INTERNAL")); return (0); } out: if ((arg->flags & UMA_ZONE_MAXBUCKET) == 0) zone->uz_count = bucket_select(zone->uz_size); else zone->uz_count = BUCKET_MAX; zone->uz_count_min = zone->uz_count; return (0); } /* * Keg header dtor. This frees all data, destroys locks, frees the hash * table and removes the keg from the global list. * * Arguments/Returns follow uma_dtor specifications * udata unused */ static void keg_dtor(void *arg, int size, void *udata) { uma_keg_t keg; keg = (uma_keg_t)arg; KEG_LOCK(keg); if (keg->uk_free != 0) { printf("Freed UMA keg (%s) was not empty (%d items). " " Lost %d pages of memory.\n", keg->uk_name ? keg->uk_name : "", keg->uk_free, keg->uk_pages); } KEG_UNLOCK(keg); hash_free(&keg->uk_hash); KEG_LOCK_FINI(keg); } /* * Zone header dtor. * * Arguments/Returns follow uma_dtor specifications * udata unused */ static void zone_dtor(void *arg, int size, void *udata) { uma_klink_t klink; uma_zone_t zone; uma_keg_t keg; zone = (uma_zone_t)arg; keg = zone_first_keg(zone); if (!(zone->uz_flags & UMA_ZFLAG_INTERNAL)) cache_drain(zone); rw_wlock(&uma_rwlock); LIST_REMOVE(zone, uz_link); rw_wunlock(&uma_rwlock); /* * XXX there are some races here where * the zone can be drained but zone lock * released and then refilled before we * remove it... we dont care for now */ zone_drain_wait(zone, M_WAITOK); /* * Unlink all of our kegs. */ while ((klink = LIST_FIRST(&zone->uz_kegs)) != NULL) { klink->kl_keg = NULL; LIST_REMOVE(klink, kl_link); if (klink == &zone->uz_klink) continue; free(klink, M_TEMP); } /* * We only destroy kegs from non secondary zones. */ if (keg != NULL && (zone->uz_flags & UMA_ZONE_SECONDARY) == 0) { rw_wlock(&uma_rwlock); LIST_REMOVE(keg, uk_link); rw_wunlock(&uma_rwlock); zone_free_item(kegs, keg, NULL, SKIP_NONE); } ZONE_LOCK_FINI(zone); } /* * Traverses every zone in the system and calls a callback * * Arguments: * zfunc A pointer to a function which accepts a zone * as an argument. * * Returns: * Nothing */ static void zone_foreach(void (*zfunc)(uma_zone_t)) { uma_keg_t keg; uma_zone_t zone; rw_rlock(&uma_rwlock); LIST_FOREACH(keg, &uma_kegs, uk_link) { LIST_FOREACH(zone, &keg->uk_zones, uz_link) zfunc(zone); } rw_runlock(&uma_rwlock); } /* Public functions */ /* See uma.h */ void uma_startup(void *bootmem, int boot_pages) { struct uma_zctor_args args; uma_slab_t slab; int i; #ifdef UMA_DEBUG printf("Creating uma keg headers zone and keg.\n"); #endif rw_init(&uma_rwlock, "UMA lock"); /* "manually" create the initial zone */ memset(&args, 0, sizeof(args)); args.name = "UMA Kegs"; args.size = sizeof(struct uma_keg); args.ctor = keg_ctor; args.dtor = keg_dtor; args.uminit = zero_init; args.fini = NULL; args.keg = &masterkeg; args.align = 32 - 1; args.flags = UMA_ZFLAG_INTERNAL; /* The initial zone has no Per cpu queues so it's smaller */ zone_ctor(kegs, sizeof(struct uma_zone), &args, M_WAITOK); #ifdef UMA_DEBUG printf("Filling boot free list.\n"); #endif for (i = 0; i < boot_pages; i++) { slab = (uma_slab_t)((uint8_t *)bootmem + (i * UMA_SLAB_SIZE)); slab->us_data = (uint8_t *)slab; slab->us_flags = UMA_SLAB_BOOT; LIST_INSERT_HEAD(&uma_boot_pages, slab, us_link); } mtx_init(&uma_boot_pages_mtx, "UMA boot pages", NULL, MTX_DEF); #ifdef UMA_DEBUG printf("Creating uma zone headers zone and keg.\n"); #endif args.name = "UMA Zones"; args.size = sizeof(struct uma_zone) + (sizeof(struct uma_cache) * (mp_maxid + 1)); args.ctor = zone_ctor; args.dtor = zone_dtor; args.uminit = zero_init; args.fini = NULL; args.keg = NULL; args.align = 32 - 1; args.flags = UMA_ZFLAG_INTERNAL; /* The initial zone has no Per cpu queues so it's smaller */ zone_ctor(zones, sizeof(struct uma_zone), &args, M_WAITOK); #ifdef UMA_DEBUG printf("Creating slab and hash zones.\n"); #endif /* Now make a zone for slab headers */ slabzone = uma_zcreate("UMA Slabs", sizeof(struct uma_slab), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZFLAG_INTERNAL); hashzone = uma_zcreate("UMA Hash", sizeof(struct slabhead *) * UMA_HASH_SIZE_INIT, NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZFLAG_INTERNAL); bucket_init(); booted = UMA_STARTUP; #ifdef UMA_DEBUG printf("UMA startup complete.\n"); #endif } /* see uma.h */ void uma_startup2(void) { booted = UMA_STARTUP2; bucket_enable(); sx_init(&uma_drain_lock, "umadrain"); #ifdef UMA_DEBUG printf("UMA startup2 complete.\n"); #endif } /* * Initialize our callout handle * */ static void uma_startup3(void) { #ifdef UMA_DEBUG printf("Starting callout.\n"); #endif callout_init(&uma_callout, 1); callout_reset(&uma_callout, UMA_TIMEOUT * hz, uma_timeout, NULL); #ifdef UMA_DEBUG printf("UMA startup3 complete.\n"); #endif } static uma_keg_t uma_kcreate(uma_zone_t zone, size_t size, uma_init uminit, uma_fini fini, int align, uint32_t flags) { struct uma_kctor_args args; args.size = size; args.uminit = uminit; args.fini = fini; args.align = (align == UMA_ALIGN_CACHE) ? uma_align_cache : align; args.flags = flags; args.zone = zone; return (zone_alloc_item(kegs, &args, M_WAITOK)); } /* See uma.h */ void uma_set_align(int align) { if (align != UMA_ALIGN_CACHE) uma_align_cache = align; } /* See uma.h */ uma_zone_t uma_zcreate(const char *name, size_t size, uma_ctor ctor, uma_dtor dtor, uma_init uminit, uma_fini fini, int align, uint32_t flags) { struct uma_zctor_args args; uma_zone_t res; bool locked; KASSERT(powerof2(align + 1), ("invalid zone alignment %d for \"%s\"", align, name)); /* This stuff is essential for the zone ctor */ memset(&args, 0, sizeof(args)); args.name = name; args.size = size; args.ctor = ctor; args.dtor = dtor; args.uminit = uminit; args.fini = fini; #ifdef INVARIANTS /* * If a zone is being created with an empty constructor and * destructor, pass UMA constructor/destructor which checks for * memory use after free. */ if ((!(flags & (UMA_ZONE_ZINIT | UMA_ZONE_NOFREE))) && ctor == NULL && dtor == NULL && uminit == NULL && fini == NULL) { args.ctor = trash_ctor; args.dtor = trash_dtor; args.uminit = trash_init; args.fini = trash_fini; } #endif args.align = align; args.flags = flags; args.keg = NULL; if (booted < UMA_STARTUP2) { locked = false; } else { sx_slock(&uma_drain_lock); locked = true; } res = zone_alloc_item(zones, &args, M_WAITOK); if (locked) sx_sunlock(&uma_drain_lock); return (res); } /* See uma.h */ uma_zone_t uma_zsecond_create(char *name, uma_ctor ctor, uma_dtor dtor, uma_init zinit, uma_fini zfini, uma_zone_t master) { struct uma_zctor_args args; uma_keg_t keg; uma_zone_t res; bool locked; keg = zone_first_keg(master); memset(&args, 0, sizeof(args)); args.name = name; args.size = keg->uk_size; args.ctor = ctor; args.dtor = dtor; args.uminit = zinit; args.fini = zfini; args.align = keg->uk_align; args.flags = keg->uk_flags | UMA_ZONE_SECONDARY; args.keg = keg; if (booted < UMA_STARTUP2) { locked = false; } else { sx_slock(&uma_drain_lock); locked = true; } /* XXX Attaches only one keg of potentially many. */ res = zone_alloc_item(zones, &args, M_WAITOK); if (locked) sx_sunlock(&uma_drain_lock); return (res); } /* See uma.h */ uma_zone_t uma_zcache_create(char *name, int size, uma_ctor ctor, uma_dtor dtor, uma_init zinit, uma_fini zfini, uma_import zimport, uma_release zrelease, void *arg, int flags) { struct uma_zctor_args args; memset(&args, 0, sizeof(args)); args.name = name; args.size = size; args.ctor = ctor; args.dtor = dtor; args.uminit = zinit; args.fini = zfini; args.import = zimport; args.release = zrelease; args.arg = arg; args.align = 0; args.flags = flags; return (zone_alloc_item(zones, &args, M_WAITOK)); } static void zone_lock_pair(uma_zone_t a, uma_zone_t b) { if (a < b) { ZONE_LOCK(a); mtx_lock_flags(b->uz_lockptr, MTX_DUPOK); } else { ZONE_LOCK(b); mtx_lock_flags(a->uz_lockptr, MTX_DUPOK); } } static void zone_unlock_pair(uma_zone_t a, uma_zone_t b) { ZONE_UNLOCK(a); ZONE_UNLOCK(b); } int uma_zsecond_add(uma_zone_t zone, uma_zone_t master) { uma_klink_t klink; uma_klink_t kl; int error; error = 0; klink = malloc(sizeof(*klink), M_TEMP, M_WAITOK | M_ZERO); zone_lock_pair(zone, master); /* * zone must use vtoslab() to resolve objects and must already be * a secondary. */ if ((zone->uz_flags & (UMA_ZONE_VTOSLAB | UMA_ZONE_SECONDARY)) != (UMA_ZONE_VTOSLAB | UMA_ZONE_SECONDARY)) { error = EINVAL; goto out; } /* * The new master must also use vtoslab(). */ if ((zone->uz_flags & UMA_ZONE_VTOSLAB) != UMA_ZONE_VTOSLAB) { error = EINVAL; goto out; } /* * The underlying object must be the same size. rsize * may be different. */ if (master->uz_size != zone->uz_size) { error = E2BIG; goto out; } /* * Put it at the end of the list. */ klink->kl_keg = zone_first_keg(master); LIST_FOREACH(kl, &zone->uz_kegs, kl_link) { if (LIST_NEXT(kl, kl_link) == NULL) { LIST_INSERT_AFTER(kl, klink, kl_link); break; } } klink = NULL; zone->uz_flags |= UMA_ZFLAG_MULTI; zone->uz_slab = zone_fetch_slab_multi; out: zone_unlock_pair(zone, master); if (klink != NULL) free(klink, M_TEMP); return (error); } /* See uma.h */ void uma_zdestroy(uma_zone_t zone) { sx_slock(&uma_drain_lock); zone_free_item(zones, zone, NULL, SKIP_NONE); sx_sunlock(&uma_drain_lock); } /* See uma.h */ void * uma_zalloc_arg(uma_zone_t zone, void *udata, int flags) { void *item; uma_cache_t cache; uma_bucket_t bucket; int lockfail; int cpu; /* Enable entropy collection for RANDOM_ENABLE_UMA kernel option */ random_harvest_fast_uma(&zone, sizeof(zone), 1, RANDOM_UMA); /* This is the fast path allocation */ #ifdef UMA_DEBUG_ALLOC_1 printf("Allocating one item from %s(%p)\n", zone->uz_name, zone); #endif CTR3(KTR_UMA, "uma_zalloc_arg thread %x zone %s flags %d", curthread, zone->uz_name, flags); if (flags & M_WAITOK) { WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "uma_zalloc_arg: zone \"%s\"", zone->uz_name); } KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), ("uma_zalloc_arg: called with spinlock or critical section held")); #ifdef DEBUG_MEMGUARD if (memguard_cmp_zone(zone)) { item = memguard_alloc(zone->uz_size, flags); if (item != NULL) { if (zone->uz_init != NULL && zone->uz_init(item, zone->uz_size, flags) != 0) return (NULL); if (zone->uz_ctor != NULL && zone->uz_ctor(item, zone->uz_size, udata, flags) != 0) { zone->uz_fini(item, zone->uz_size); return (NULL); } return (item); } /* This is unfortunate but should not be fatal. */ } #endif /* * If possible, allocate from the per-CPU cache. There are two * requirements for safe access to the per-CPU cache: (1) the thread * accessing the cache must not be preempted or yield during access, * and (2) the thread must not migrate CPUs without switching which * cache it accesses. We rely on a critical section to prevent * preemption and migration. We release the critical section in * order to acquire the zone mutex if we are unable to allocate from * the current cache; when we re-acquire the critical section, we * must detect and handle migration if it has occurred. */ critical_enter(); cpu = curcpu; cache = &zone->uz_cpu[cpu]; zalloc_start: bucket = cache->uc_allocbucket; if (bucket != NULL && bucket->ub_cnt > 0) { bucket->ub_cnt--; item = bucket->ub_bucket[bucket->ub_cnt]; #ifdef INVARIANTS bucket->ub_bucket[bucket->ub_cnt] = NULL; #endif KASSERT(item != NULL, ("uma_zalloc: Bucket pointer mangled.")); cache->uc_allocs++; critical_exit(); if (zone->uz_ctor != NULL && zone->uz_ctor(item, zone->uz_size, udata, flags) != 0) { atomic_add_long(&zone->uz_fails, 1); zone_free_item(zone, item, udata, SKIP_DTOR); return (NULL); } #ifdef INVARIANTS uma_dbg_alloc(zone, NULL, item); #endif if (flags & M_ZERO) uma_zero_item(item, zone); return (item); } /* * We have run out of items in our alloc bucket. * See if we can switch with our free bucket. */ bucket = cache->uc_freebucket; if (bucket != NULL && bucket->ub_cnt > 0) { #ifdef UMA_DEBUG_ALLOC printf("uma_zalloc: Swapping empty with alloc.\n"); #endif cache->uc_freebucket = cache->uc_allocbucket; cache->uc_allocbucket = bucket; goto zalloc_start; } /* * Discard any empty allocation bucket while we hold no locks. */ bucket = cache->uc_allocbucket; cache->uc_allocbucket = NULL; critical_exit(); if (bucket != NULL) bucket_free(zone, bucket, udata); /* Short-circuit for zones without buckets and low memory. */ if (zone->uz_count == 0 || bucketdisable) goto zalloc_item; /* * Attempt to retrieve the item from the per-CPU cache has failed, so * we must go back to the zone. This requires the zone lock, so we * must drop the critical section, then re-acquire it when we go back * to the cache. Since the critical section is released, we may be * preempted or migrate. As such, make sure not to maintain any * thread-local state specific to the cache from prior to releasing * the critical section. */ lockfail = 0; if (ZONE_TRYLOCK(zone) == 0) { /* Record contention to size the buckets. */ ZONE_LOCK(zone); lockfail = 1; } critical_enter(); cpu = curcpu; cache = &zone->uz_cpu[cpu]; /* * Since we have locked the zone we may as well send back our stats. */ atomic_add_long(&zone->uz_allocs, cache->uc_allocs); atomic_add_long(&zone->uz_frees, cache->uc_frees); cache->uc_allocs = 0; cache->uc_frees = 0; /* See if we lost the race to fill the cache. */ if (cache->uc_allocbucket != NULL) { ZONE_UNLOCK(zone); goto zalloc_start; } /* * Check the zone's cache of buckets. */ if ((bucket = LIST_FIRST(&zone->uz_buckets)) != NULL) { KASSERT(bucket->ub_cnt != 0, ("uma_zalloc_arg: Returning an empty bucket.")); LIST_REMOVE(bucket, ub_link); cache->uc_allocbucket = bucket; ZONE_UNLOCK(zone); goto zalloc_start; } /* We are no longer associated with this CPU. */ critical_exit(); /* * We bump the uz count when the cache size is insufficient to * handle the working set. */ if (lockfail && zone->uz_count < BUCKET_MAX) zone->uz_count++; ZONE_UNLOCK(zone); /* * Now lets just fill a bucket and put it on the free list. If that * works we'll restart the allocation from the beginning and it * will use the just filled bucket. */ bucket = zone_alloc_bucket(zone, udata, flags); if (bucket != NULL) { ZONE_LOCK(zone); critical_enter(); cpu = curcpu; cache = &zone->uz_cpu[cpu]; /* * See if we lost the race or were migrated. Cache the * initialized bucket to make this less likely or claim * the memory directly. */ if (cache->uc_allocbucket == NULL) cache->uc_allocbucket = bucket; else LIST_INSERT_HEAD(&zone->uz_buckets, bucket, ub_link); ZONE_UNLOCK(zone); goto zalloc_start; } /* * We may not be able to get a bucket so return an actual item. */ #ifdef UMA_DEBUG printf("uma_zalloc_arg: Bucketzone returned NULL\n"); #endif zalloc_item: item = zone_alloc_item(zone, udata, flags); return (item); } static uma_slab_t keg_fetch_slab(uma_keg_t keg, uma_zone_t zone, int flags) { uma_slab_t slab; int reserve; mtx_assert(&keg->uk_lock, MA_OWNED); slab = NULL; reserve = 0; if ((flags & M_USE_RESERVE) == 0) reserve = keg->uk_reserve; for (;;) { /* * Find a slab with some space. Prefer slabs that are partially * used over those that are totally full. This helps to reduce * fragmentation. */ if (keg->uk_free > reserve) { if (!LIST_EMPTY(&keg->uk_part_slab)) { slab = LIST_FIRST(&keg->uk_part_slab); } else { slab = LIST_FIRST(&keg->uk_free_slab); LIST_REMOVE(slab, us_link); LIST_INSERT_HEAD(&keg->uk_part_slab, slab, us_link); } MPASS(slab->us_keg == keg); return (slab); } /* * M_NOVM means don't ask at all! */ if (flags & M_NOVM) break; if (keg->uk_maxpages && keg->uk_pages >= keg->uk_maxpages) { keg->uk_flags |= UMA_ZFLAG_FULL; /* * If this is not a multi-zone, set the FULL bit. * Otherwise slab_multi() takes care of it. */ if ((zone->uz_flags & UMA_ZFLAG_MULTI) == 0) { zone->uz_flags |= UMA_ZFLAG_FULL; zone_log_warning(zone); zone_maxaction(zone); } if (flags & M_NOWAIT) break; zone->uz_sleeps++; msleep(keg, &keg->uk_lock, PVM, "keglimit", 0); continue; } slab = keg_alloc_slab(keg, zone, flags); /* * If we got a slab here it's safe to mark it partially used * and return. We assume that the caller is going to remove * at least one item. */ if (slab) { MPASS(slab->us_keg == keg); LIST_INSERT_HEAD(&keg->uk_part_slab, slab, us_link); return (slab); } /* * We might not have been able to get a slab but another cpu * could have while we were unlocked. Check again before we * fail. */ flags |= M_NOVM; } return (slab); } static uma_slab_t zone_fetch_slab(uma_zone_t zone, uma_keg_t keg, int flags) { uma_slab_t slab; if (keg == NULL) { keg = zone_first_keg(zone); KEG_LOCK(keg); } for (;;) { slab = keg_fetch_slab(keg, zone, flags); if (slab) return (slab); if (flags & (M_NOWAIT | M_NOVM)) break; } KEG_UNLOCK(keg); return (NULL); } /* * uma_zone_fetch_slab_multi: Fetches a slab from one available keg. Returns * with the keg locked. On NULL no lock is held. * * The last pointer is used to seed the search. It is not required. */ static uma_slab_t zone_fetch_slab_multi(uma_zone_t zone, uma_keg_t last, int rflags) { uma_klink_t klink; uma_slab_t slab; uma_keg_t keg; int flags; int empty; int full; /* * Don't wait on the first pass. This will skip limit tests * as well. We don't want to block if we can find a provider * without blocking. */ flags = (rflags & ~M_WAITOK) | M_NOWAIT; /* * Use the last slab allocated as a hint for where to start * the search. */ if (last != NULL) { slab = keg_fetch_slab(last, zone, flags); if (slab) return (slab); KEG_UNLOCK(last); } /* * Loop until we have a slab incase of transient failures * while M_WAITOK is specified. I'm not sure this is 100% * required but we've done it for so long now. */ for (;;) { empty = 0; full = 0; /* * Search the available kegs for slabs. Be careful to hold the * correct lock while calling into the keg layer. */ LIST_FOREACH(klink, &zone->uz_kegs, kl_link) { keg = klink->kl_keg; KEG_LOCK(keg); if ((keg->uk_flags & UMA_ZFLAG_FULL) == 0) { slab = keg_fetch_slab(keg, zone, flags); if (slab) return (slab); } if (keg->uk_flags & UMA_ZFLAG_FULL) full++; else empty++; KEG_UNLOCK(keg); } if (rflags & (M_NOWAIT | M_NOVM)) break; flags = rflags; /* * All kegs are full. XXX We can't atomically check all kegs * and sleep so just sleep for a short period and retry. */ if (full && !empty) { ZONE_LOCK(zone); zone->uz_flags |= UMA_ZFLAG_FULL; zone->uz_sleeps++; zone_log_warning(zone); zone_maxaction(zone); msleep(zone, zone->uz_lockptr, PVM, "zonelimit", hz/100); zone->uz_flags &= ~UMA_ZFLAG_FULL; ZONE_UNLOCK(zone); continue; } } return (NULL); } static void * slab_alloc_item(uma_keg_t keg, uma_slab_t slab) { void *item; uint8_t freei; MPASS(keg == slab->us_keg); mtx_assert(&keg->uk_lock, MA_OWNED); freei = BIT_FFS(SLAB_SETSIZE, &slab->us_free) - 1; BIT_CLR(SLAB_SETSIZE, freei, &slab->us_free); item = slab->us_data + (keg->uk_rsize * freei); slab->us_freecount--; keg->uk_free--; /* Move this slab to the full list */ if (slab->us_freecount == 0) { LIST_REMOVE(slab, us_link); LIST_INSERT_HEAD(&keg->uk_full_slab, slab, us_link); } return (item); } static int zone_import(uma_zone_t zone, void **bucket, int max, int flags) { uma_slab_t slab; uma_keg_t keg; int i; slab = NULL; keg = NULL; /* Try to keep the buckets totally full */ for (i = 0; i < max; ) { if ((slab = zone->uz_slab(zone, keg, flags)) == NULL) break; keg = slab->us_keg; while (slab->us_freecount && i < max) { bucket[i++] = slab_alloc_item(keg, slab); if (keg->uk_free <= keg->uk_reserve) break; } /* Don't grab more than one slab at a time. */ flags &= ~M_WAITOK; flags |= M_NOWAIT; } if (slab != NULL) KEG_UNLOCK(keg); return i; } static uma_bucket_t zone_alloc_bucket(uma_zone_t zone, void *udata, int flags) { uma_bucket_t bucket; int max; /* Don't wait for buckets, preserve caller's NOVM setting. */ bucket = bucket_alloc(zone, udata, M_NOWAIT | (flags & M_NOVM)); if (bucket == NULL) return (NULL); max = MIN(bucket->ub_entries, zone->uz_count); bucket->ub_cnt = zone->uz_import(zone->uz_arg, bucket->ub_bucket, max, flags); /* * Initialize the memory if necessary. */ if (bucket->ub_cnt != 0 && zone->uz_init != NULL) { int i; for (i = 0; i < bucket->ub_cnt; i++) if (zone->uz_init(bucket->ub_bucket[i], zone->uz_size, flags) != 0) break; /* * If we couldn't initialize the whole bucket, put the * rest back onto the freelist. */ if (i != bucket->ub_cnt) { zone->uz_release(zone->uz_arg, &bucket->ub_bucket[i], bucket->ub_cnt - i); #ifdef INVARIANTS bzero(&bucket->ub_bucket[i], sizeof(void *) * (bucket->ub_cnt - i)); #endif bucket->ub_cnt = i; } } if (bucket->ub_cnt == 0) { bucket_free(zone, bucket, udata); atomic_add_long(&zone->uz_fails, 1); return (NULL); } return (bucket); } /* * Allocates a single item from a zone. * * Arguments * zone The zone to alloc for. * udata The data to be passed to the constructor. * flags M_WAITOK, M_NOWAIT, M_ZERO. * * Returns * NULL if there is no memory and M_NOWAIT is set * An item if successful */ static void * zone_alloc_item(uma_zone_t zone, void *udata, int flags) { void *item; item = NULL; #ifdef UMA_DEBUG_ALLOC printf("INTERNAL: Allocating one item from %s(%p)\n", zone->uz_name, zone); #endif if (zone->uz_import(zone->uz_arg, &item, 1, flags) != 1) goto fail; atomic_add_long(&zone->uz_allocs, 1); /* * We have to call both the zone's init (not the keg's init) * and the zone's ctor. This is because the item is going from * a keg slab directly to the user, and the user is expecting it * to be both zone-init'd as well as zone-ctor'd. */ if (zone->uz_init != NULL) { if (zone->uz_init(item, zone->uz_size, flags) != 0) { zone_free_item(zone, item, udata, SKIP_FINI); goto fail; } } if (zone->uz_ctor != NULL) { if (zone->uz_ctor(item, zone->uz_size, udata, flags) != 0) { zone_free_item(zone, item, udata, SKIP_DTOR); goto fail; } } #ifdef INVARIANTS uma_dbg_alloc(zone, NULL, item); #endif if (flags & M_ZERO) uma_zero_item(item, zone); return (item); fail: atomic_add_long(&zone->uz_fails, 1); return (NULL); } /* See uma.h */ void uma_zfree_arg(uma_zone_t zone, void *item, void *udata) { uma_cache_t cache; uma_bucket_t bucket; int lockfail; int cpu; /* Enable entropy collection for RANDOM_ENABLE_UMA kernel option */ random_harvest_fast_uma(&zone, sizeof(zone), 1, RANDOM_UMA); #ifdef UMA_DEBUG_ALLOC_1 printf("Freeing item %p to %s(%p)\n", item, zone->uz_name, zone); #endif CTR2(KTR_UMA, "uma_zfree_arg thread %x zone %s", curthread, zone->uz_name); KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), ("uma_zfree_arg: called with spinlock or critical section held")); /* uma_zfree(..., NULL) does nothing, to match free(9). */ if (item == NULL) return; #ifdef DEBUG_MEMGUARD if (is_memguard_addr(item)) { if (zone->uz_dtor != NULL) zone->uz_dtor(item, zone->uz_size, udata); if (zone->uz_fini != NULL) zone->uz_fini(item, zone->uz_size); memguard_free(item); return; } #endif #ifdef INVARIANTS if (zone->uz_flags & UMA_ZONE_MALLOC) uma_dbg_free(zone, udata, item); else uma_dbg_free(zone, NULL, item); #endif if (zone->uz_dtor != NULL) zone->uz_dtor(item, zone->uz_size, udata); /* * The race here is acceptable. If we miss it we'll just have to wait * a little longer for the limits to be reset. */ if (zone->uz_flags & UMA_ZFLAG_FULL) goto zfree_item; /* * If possible, free to the per-CPU cache. There are two * requirements for safe access to the per-CPU cache: (1) the thread * accessing the cache must not be preempted or yield during access, * and (2) the thread must not migrate CPUs without switching which * cache it accesses. We rely on a critical section to prevent * preemption and migration. We release the critical section in * order to acquire the zone mutex if we are unable to free to the * current cache; when we re-acquire the critical section, we must * detect and handle migration if it has occurred. */ zfree_restart: critical_enter(); cpu = curcpu; cache = &zone->uz_cpu[cpu]; zfree_start: /* * Try to free into the allocbucket first to give LIFO ordering * for cache-hot datastructures. Spill over into the freebucket * if necessary. Alloc will swap them if one runs dry. */ bucket = cache->uc_allocbucket; if (bucket == NULL || bucket->ub_cnt >= bucket->ub_entries) bucket = cache->uc_freebucket; if (bucket != NULL && bucket->ub_cnt < bucket->ub_entries) { KASSERT(bucket->ub_bucket[bucket->ub_cnt] == NULL, ("uma_zfree: Freeing to non free bucket index.")); bucket->ub_bucket[bucket->ub_cnt] = item; bucket->ub_cnt++; cache->uc_frees++; critical_exit(); return; } /* * We must go back the zone, which requires acquiring the zone lock, * which in turn means we must release and re-acquire the critical * section. Since the critical section is released, we may be * preempted or migrate. As such, make sure not to maintain any * thread-local state specific to the cache from prior to releasing * the critical section. */ critical_exit(); if (zone->uz_count == 0 || bucketdisable) goto zfree_item; lockfail = 0; if (ZONE_TRYLOCK(zone) == 0) { /* Record contention to size the buckets. */ ZONE_LOCK(zone); lockfail = 1; } critical_enter(); cpu = curcpu; cache = &zone->uz_cpu[cpu]; /* * Since we have locked the zone we may as well send back our stats. */ atomic_add_long(&zone->uz_allocs, cache->uc_allocs); atomic_add_long(&zone->uz_frees, cache->uc_frees); cache->uc_allocs = 0; cache->uc_frees = 0; bucket = cache->uc_freebucket; if (bucket != NULL && bucket->ub_cnt < bucket->ub_entries) { ZONE_UNLOCK(zone); goto zfree_start; } cache->uc_freebucket = NULL; /* We are no longer associated with this CPU. */ critical_exit(); /* Can we throw this on the zone full list? */ if (bucket != NULL) { #ifdef UMA_DEBUG_ALLOC printf("uma_zfree: Putting old bucket on the free list.\n"); #endif /* ub_cnt is pointing to the last free item */ KASSERT(bucket->ub_cnt != 0, ("uma_zfree: Attempting to insert an empty bucket onto the full list.\n")); LIST_INSERT_HEAD(&zone->uz_buckets, bucket, ub_link); } /* * We bump the uz count when the cache size is insufficient to * handle the working set. */ if (lockfail && zone->uz_count < BUCKET_MAX) zone->uz_count++; ZONE_UNLOCK(zone); #ifdef UMA_DEBUG_ALLOC printf("uma_zfree: Allocating new free bucket.\n"); #endif bucket = bucket_alloc(zone, udata, M_NOWAIT); if (bucket) { critical_enter(); cpu = curcpu; cache = &zone->uz_cpu[cpu]; if (cache->uc_freebucket == NULL) { cache->uc_freebucket = bucket; goto zfree_start; } /* * We lost the race, start over. We have to drop our * critical section to free the bucket. */ critical_exit(); bucket_free(zone, bucket, udata); goto zfree_restart; } /* * If nothing else caught this, we'll just do an internal free. */ zfree_item: zone_free_item(zone, item, udata, SKIP_DTOR); return; } static void slab_free_item(uma_keg_t keg, uma_slab_t slab, void *item) { uint8_t freei; mtx_assert(&keg->uk_lock, MA_OWNED); MPASS(keg == slab->us_keg); /* Do we need to remove from any lists? */ if (slab->us_freecount+1 == keg->uk_ipers) { LIST_REMOVE(slab, us_link); LIST_INSERT_HEAD(&keg->uk_free_slab, slab, us_link); } else if (slab->us_freecount == 0) { LIST_REMOVE(slab, us_link); LIST_INSERT_HEAD(&keg->uk_part_slab, slab, us_link); } /* Slab management. */ freei = ((uintptr_t)item - (uintptr_t)slab->us_data) / keg->uk_rsize; BIT_SET(SLAB_SETSIZE, freei, &slab->us_free); slab->us_freecount++; /* Keg statistics. */ keg->uk_free++; } static void zone_release(uma_zone_t zone, void **bucket, int cnt) { void *item; uma_slab_t slab; uma_keg_t keg; uint8_t *mem; int clearfull; int i; clearfull = 0; keg = zone_first_keg(zone); KEG_LOCK(keg); for (i = 0; i < cnt; i++) { item = bucket[i]; if (!(zone->uz_flags & UMA_ZONE_VTOSLAB)) { mem = (uint8_t *)((uintptr_t)item & (~UMA_SLAB_MASK)); if (zone->uz_flags & UMA_ZONE_HASH) { slab = hash_sfind(&keg->uk_hash, mem); } else { mem += keg->uk_pgoff; slab = (uma_slab_t)mem; } } else { slab = vtoslab((vm_offset_t)item); if (slab->us_keg != keg) { KEG_UNLOCK(keg); keg = slab->us_keg; KEG_LOCK(keg); } } slab_free_item(keg, slab, item); if (keg->uk_flags & UMA_ZFLAG_FULL) { if (keg->uk_pages < keg->uk_maxpages) { keg->uk_flags &= ~UMA_ZFLAG_FULL; clearfull = 1; } /* * We can handle one more allocation. Since we're * clearing ZFLAG_FULL, wake up all procs blocked * on pages. This should be uncommon, so keeping this * simple for now (rather than adding count of blocked * threads etc). */ wakeup(keg); } } KEG_UNLOCK(keg); if (clearfull) { ZONE_LOCK(zone); zone->uz_flags &= ~UMA_ZFLAG_FULL; wakeup(zone); ZONE_UNLOCK(zone); } } /* * Frees a single item to any zone. * * Arguments: * zone The zone to free to * item The item we're freeing * udata User supplied data for the dtor * skip Skip dtors and finis */ static void zone_free_item(uma_zone_t zone, void *item, void *udata, enum zfreeskip skip) { #ifdef INVARIANTS if (skip == SKIP_NONE) { if (zone->uz_flags & UMA_ZONE_MALLOC) uma_dbg_free(zone, udata, item); else uma_dbg_free(zone, NULL, item); } #endif if (skip < SKIP_DTOR && zone->uz_dtor) zone->uz_dtor(item, zone->uz_size, udata); if (skip < SKIP_FINI && zone->uz_fini) zone->uz_fini(item, zone->uz_size); atomic_add_long(&zone->uz_frees, 1); zone->uz_release(zone->uz_arg, &item, 1); } /* See uma.h */ int uma_zone_set_max(uma_zone_t zone, int nitems) { uma_keg_t keg; keg = zone_first_keg(zone); if (keg == NULL) return (0); KEG_LOCK(keg); keg->uk_maxpages = (nitems / keg->uk_ipers) * keg->uk_ppera; if (keg->uk_maxpages * keg->uk_ipers < nitems) keg->uk_maxpages += keg->uk_ppera; nitems = (keg->uk_maxpages / keg->uk_ppera) * keg->uk_ipers; KEG_UNLOCK(keg); return (nitems); } /* See uma.h */ int uma_zone_get_max(uma_zone_t zone) { int nitems; uma_keg_t keg; keg = zone_first_keg(zone); if (keg == NULL) return (0); KEG_LOCK(keg); nitems = (keg->uk_maxpages / keg->uk_ppera) * keg->uk_ipers; KEG_UNLOCK(keg); return (nitems); } /* See uma.h */ void uma_zone_set_warning(uma_zone_t zone, const char *warning) { ZONE_LOCK(zone); zone->uz_warning = warning; ZONE_UNLOCK(zone); } /* See uma.h */ void uma_zone_set_maxaction(uma_zone_t zone, uma_maxaction_t maxaction) { ZONE_LOCK(zone); TASK_INIT(&zone->uz_maxaction, 0, (task_fn_t *)maxaction, zone); ZONE_UNLOCK(zone); } /* See uma.h */ int uma_zone_get_cur(uma_zone_t zone) { int64_t nitems; u_int i; ZONE_LOCK(zone); nitems = zone->uz_allocs - zone->uz_frees; CPU_FOREACH(i) { /* * See the comment in sysctl_vm_zone_stats() regarding the * safety of accessing the per-cpu caches. With the zone lock * held, it is safe, but can potentially result in stale data. */ nitems += zone->uz_cpu[i].uc_allocs - zone->uz_cpu[i].uc_frees; } ZONE_UNLOCK(zone); return (nitems < 0 ? 0 : nitems); } /* See uma.h */ void uma_zone_set_init(uma_zone_t zone, uma_init uminit) { uma_keg_t keg; keg = zone_first_keg(zone); KASSERT(keg != NULL, ("uma_zone_set_init: Invalid zone type")); KEG_LOCK(keg); KASSERT(keg->uk_pages == 0, ("uma_zone_set_init on non-empty keg")); keg->uk_init = uminit; KEG_UNLOCK(keg); } /* See uma.h */ void uma_zone_set_fini(uma_zone_t zone, uma_fini fini) { uma_keg_t keg; keg = zone_first_keg(zone); KASSERT(keg != NULL, ("uma_zone_set_fini: Invalid zone type")); KEG_LOCK(keg); KASSERT(keg->uk_pages == 0, ("uma_zone_set_fini on non-empty keg")); keg->uk_fini = fini; KEG_UNLOCK(keg); } /* See uma.h */ void uma_zone_set_zinit(uma_zone_t zone, uma_init zinit) { ZONE_LOCK(zone); KASSERT(zone_first_keg(zone)->uk_pages == 0, ("uma_zone_set_zinit on non-empty keg")); zone->uz_init = zinit; ZONE_UNLOCK(zone); } /* See uma.h */ void uma_zone_set_zfini(uma_zone_t zone, uma_fini zfini) { ZONE_LOCK(zone); KASSERT(zone_first_keg(zone)->uk_pages == 0, ("uma_zone_set_zfini on non-empty keg")); zone->uz_fini = zfini; ZONE_UNLOCK(zone); } /* See uma.h */ /* XXX uk_freef is not actually used with the zone locked */ void uma_zone_set_freef(uma_zone_t zone, uma_free freef) { uma_keg_t keg; keg = zone_first_keg(zone); KASSERT(keg != NULL, ("uma_zone_set_freef: Invalid zone type")); KEG_LOCK(keg); keg->uk_freef = freef; KEG_UNLOCK(keg); } /* See uma.h */ /* XXX uk_allocf is not actually used with the zone locked */ void uma_zone_set_allocf(uma_zone_t zone, uma_alloc allocf) { uma_keg_t keg; keg = zone_first_keg(zone); KEG_LOCK(keg); keg->uk_allocf = allocf; KEG_UNLOCK(keg); } /* See uma.h */ void uma_zone_reserve(uma_zone_t zone, int items) { uma_keg_t keg; keg = zone_first_keg(zone); if (keg == NULL) return; KEG_LOCK(keg); keg->uk_reserve = items; KEG_UNLOCK(keg); return; } /* See uma.h */ int uma_zone_reserve_kva(uma_zone_t zone, int count) { uma_keg_t keg; vm_offset_t kva; u_int pages; keg = zone_first_keg(zone); if (keg == NULL) return (0); pages = count / keg->uk_ipers; if (pages * keg->uk_ipers < count) pages++; pages *= keg->uk_ppera; #ifdef UMA_MD_SMALL_ALLOC if (keg->uk_ppera > 1) { #else if (1) { #endif kva = kva_alloc((vm_size_t)pages * PAGE_SIZE); if (kva == 0) return (0); } else kva = 0; KEG_LOCK(keg); keg->uk_kva = kva; keg->uk_offset = 0; keg->uk_maxpages = pages; #ifdef UMA_MD_SMALL_ALLOC keg->uk_allocf = (keg->uk_ppera > 1) ? noobj_alloc : uma_small_alloc; #else keg->uk_allocf = noobj_alloc; #endif keg->uk_flags |= UMA_ZONE_NOFREE; KEG_UNLOCK(keg); return (1); } /* See uma.h */ void uma_prealloc(uma_zone_t zone, int items) { int slabs; uma_slab_t slab; uma_keg_t keg; keg = zone_first_keg(zone); if (keg == NULL) return; KEG_LOCK(keg); slabs = items / keg->uk_ipers; if (slabs * keg->uk_ipers < items) slabs++; while (slabs > 0) { slab = keg_alloc_slab(keg, zone, M_WAITOK); if (slab == NULL) break; MPASS(slab->us_keg == keg); LIST_INSERT_HEAD(&keg->uk_free_slab, slab, us_link); slabs--; } KEG_UNLOCK(keg); } /* See uma.h */ static void uma_reclaim_locked(bool kmem_danger) { #ifdef UMA_DEBUG printf("UMA: vm asked us to release pages!\n"); #endif sx_assert(&uma_drain_lock, SA_XLOCKED); bucket_enable(); zone_foreach(zone_drain); if (vm_page_count_min() || kmem_danger) { cache_drain_safe(NULL); zone_foreach(zone_drain); } /* * Some slabs may have been freed but this zone will be visited early * we visit again so that we can free pages that are empty once other * zones are drained. We have to do the same for buckets. */ zone_drain(slabzone); bucket_zone_drain(); } void uma_reclaim(void) { sx_xlock(&uma_drain_lock); uma_reclaim_locked(false); sx_xunlock(&uma_drain_lock); } static int uma_reclaim_needed; void uma_reclaim_wakeup(void) { uma_reclaim_needed = 1; wakeup(&uma_reclaim_needed); } void uma_reclaim_worker(void *arg __unused) { sx_xlock(&uma_drain_lock); for (;;) { sx_sleep(&uma_reclaim_needed, &uma_drain_lock, PVM, "umarcl", 0); if (uma_reclaim_needed) { uma_reclaim_needed = 0; sx_xunlock(&uma_drain_lock); EVENTHANDLER_INVOKE(vm_lowmem, VM_LOW_KMEM); sx_xlock(&uma_drain_lock); uma_reclaim_locked(true); } } } /* See uma.h */ int uma_zone_exhausted(uma_zone_t zone) { int full; ZONE_LOCK(zone); full = (zone->uz_flags & UMA_ZFLAG_FULL); ZONE_UNLOCK(zone); return (full); } int uma_zone_exhausted_nolock(uma_zone_t zone) { return (zone->uz_flags & UMA_ZFLAG_FULL); } void * uma_large_malloc(vm_size_t size, int wait) { void *mem; uma_slab_t slab; uint8_t flags; slab = zone_alloc_item(slabzone, NULL, wait); if (slab == NULL) return (NULL); mem = page_alloc(NULL, size, &flags, wait); if (mem) { vsetslab((vm_offset_t)mem, slab); slab->us_data = mem; slab->us_flags = flags | UMA_SLAB_MALLOC; slab->us_size = size; } else { zone_free_item(slabzone, slab, NULL, SKIP_NONE); } return (mem); } void uma_large_free(uma_slab_t slab) { page_free(slab->us_data, slab->us_size, slab->us_flags); zone_free_item(slabzone, slab, NULL, SKIP_NONE); } static void uma_zero_item(void *item, uma_zone_t zone) { int i; if (zone->uz_flags & UMA_ZONE_PCPU) { CPU_FOREACH(i) bzero(zpcpu_get_cpu(item, i), zone->uz_size); } else bzero(item, zone->uz_size); } void uma_print_stats(void) { zone_foreach(uma_print_zone); } static void slab_print(uma_slab_t slab) { printf("slab: keg %p, data %p, freecount %d\n", slab->us_keg, slab->us_data, slab->us_freecount); } static void cache_print(uma_cache_t cache) { printf("alloc: %p(%d), free: %p(%d)\n", cache->uc_allocbucket, cache->uc_allocbucket?cache->uc_allocbucket->ub_cnt:0, cache->uc_freebucket, cache->uc_freebucket?cache->uc_freebucket->ub_cnt:0); } static void uma_print_keg(uma_keg_t keg) { uma_slab_t slab; printf("keg: %s(%p) size %d(%d) flags %#x ipers %d ppera %d " "out %d free %d limit %d\n", keg->uk_name, keg, keg->uk_size, keg->uk_rsize, keg->uk_flags, keg->uk_ipers, keg->uk_ppera, (keg->uk_pages / keg->uk_ppera) * keg->uk_ipers - keg->uk_free, keg->uk_free, (keg->uk_maxpages / keg->uk_ppera) * keg->uk_ipers); printf("Part slabs:\n"); LIST_FOREACH(slab, &keg->uk_part_slab, us_link) slab_print(slab); printf("Free slabs:\n"); LIST_FOREACH(slab, &keg->uk_free_slab, us_link) slab_print(slab); printf("Full slabs:\n"); LIST_FOREACH(slab, &keg->uk_full_slab, us_link) slab_print(slab); } void uma_print_zone(uma_zone_t zone) { uma_cache_t cache; uma_klink_t kl; int i; printf("zone: %s(%p) size %d flags %#x\n", zone->uz_name, zone, zone->uz_size, zone->uz_flags); LIST_FOREACH(kl, &zone->uz_kegs, kl_link) uma_print_keg(kl->kl_keg); CPU_FOREACH(i) { cache = &zone->uz_cpu[i]; printf("CPU %d Cache:\n", i); cache_print(cache); } } #ifdef DDB /* * Generate statistics across both the zone and its per-cpu cache's. Return * desired statistics if the pointer is non-NULL for that statistic. * * Note: does not update the zone statistics, as it can't safely clear the * per-CPU cache statistic. * * XXXRW: Following the uc_allocbucket and uc_freebucket pointers here isn't * safe from off-CPU; we should modify the caches to track this information * directly so that we don't have to. */ static void uma_zone_sumstat(uma_zone_t z, int *cachefreep, uint64_t *allocsp, uint64_t *freesp, uint64_t *sleepsp) { uma_cache_t cache; uint64_t allocs, frees, sleeps; int cachefree, cpu; allocs = frees = sleeps = 0; cachefree = 0; CPU_FOREACH(cpu) { cache = &z->uz_cpu[cpu]; if (cache->uc_allocbucket != NULL) cachefree += cache->uc_allocbucket->ub_cnt; if (cache->uc_freebucket != NULL) cachefree += cache->uc_freebucket->ub_cnt; allocs += cache->uc_allocs; frees += cache->uc_frees; } allocs += z->uz_allocs; frees += z->uz_frees; sleeps += z->uz_sleeps; if (cachefreep != NULL) *cachefreep = cachefree; if (allocsp != NULL) *allocsp = allocs; if (freesp != NULL) *freesp = frees; if (sleepsp != NULL) *sleepsp = sleeps; } #endif /* DDB */ static int sysctl_vm_zone_count(SYSCTL_HANDLER_ARGS) { uma_keg_t kz; uma_zone_t z; int count; count = 0; rw_rlock(&uma_rwlock); LIST_FOREACH(kz, &uma_kegs, uk_link) { LIST_FOREACH(z, &kz->uk_zones, uz_link) count++; } rw_runlock(&uma_rwlock); return (sysctl_handle_int(oidp, &count, 0, req)); } static int sysctl_vm_zone_stats(SYSCTL_HANDLER_ARGS) { struct uma_stream_header ush; struct uma_type_header uth; struct uma_percpu_stat ups; uma_bucket_t bucket; struct sbuf sbuf; uma_cache_t cache; uma_klink_t kl; uma_keg_t kz; uma_zone_t z; uma_keg_t k; int count, error, i; 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); count = 0; rw_rlock(&uma_rwlock); LIST_FOREACH(kz, &uma_kegs, uk_link) { LIST_FOREACH(z, &kz->uk_zones, uz_link) count++; } /* * Insert stream header. */ bzero(&ush, sizeof(ush)); ush.ush_version = UMA_STREAM_VERSION; ush.ush_maxcpus = (mp_maxid + 1); ush.ush_count = count; (void)sbuf_bcat(&sbuf, &ush, sizeof(ush)); LIST_FOREACH(kz, &uma_kegs, uk_link) { LIST_FOREACH(z, &kz->uk_zones, uz_link) { bzero(&uth, sizeof(uth)); ZONE_LOCK(z); strlcpy(uth.uth_name, z->uz_name, UTH_MAX_NAME); uth.uth_align = kz->uk_align; uth.uth_size = kz->uk_size; uth.uth_rsize = kz->uk_rsize; LIST_FOREACH(kl, &z->uz_kegs, kl_link) { k = kl->kl_keg; uth.uth_maxpages += k->uk_maxpages; uth.uth_pages += k->uk_pages; uth.uth_keg_free += k->uk_free; uth.uth_limit = (k->uk_maxpages / k->uk_ppera) * k->uk_ipers; } /* * A zone is secondary is it is not the first entry * on the keg's zone list. */ if ((z->uz_flags & UMA_ZONE_SECONDARY) && (LIST_FIRST(&kz->uk_zones) != z)) uth.uth_zone_flags = UTH_ZONE_SECONDARY; LIST_FOREACH(bucket, &z->uz_buckets, ub_link) uth.uth_zone_free += bucket->ub_cnt; uth.uth_allocs = z->uz_allocs; uth.uth_frees = z->uz_frees; uth.uth_fails = z->uz_fails; uth.uth_sleeps = z->uz_sleeps; (void)sbuf_bcat(&sbuf, &uth, sizeof(uth)); /* * While it is not normally safe to access the cache * bucket pointers while not on the CPU that owns the * cache, we only allow the pointers to be exchanged * without the zone lock held, not invalidated, so * accept the possible race associated with bucket * exchange during monitoring. */ for (i = 0; i < (mp_maxid + 1); i++) { bzero(&ups, sizeof(ups)); if (kz->uk_flags & UMA_ZFLAG_INTERNAL) goto skip; if (CPU_ABSENT(i)) goto skip; cache = &z->uz_cpu[i]; if (cache->uc_allocbucket != NULL) ups.ups_cache_free += cache->uc_allocbucket->ub_cnt; if (cache->uc_freebucket != NULL) ups.ups_cache_free += cache->uc_freebucket->ub_cnt; ups.ups_allocs = cache->uc_allocs; ups.ups_frees = cache->uc_frees; skip: (void)sbuf_bcat(&sbuf, &ups, sizeof(ups)); } ZONE_UNLOCK(z); } } rw_runlock(&uma_rwlock); error = sbuf_finish(&sbuf); sbuf_delete(&sbuf); return (error); } int sysctl_handle_uma_zone_max(SYSCTL_HANDLER_ARGS) { uma_zone_t zone = *(uma_zone_t *)arg1; int error, max; max = uma_zone_get_max(zone); error = sysctl_handle_int(oidp, &max, 0, req); if (error || !req->newptr) return (error); uma_zone_set_max(zone, max); return (0); } int sysctl_handle_uma_zone_cur(SYSCTL_HANDLER_ARGS) { uma_zone_t zone = *(uma_zone_t *)arg1; int cur; cur = uma_zone_get_cur(zone); return (sysctl_handle_int(oidp, &cur, 0, req)); } #ifdef INVARIANTS static uma_slab_t uma_dbg_getslab(uma_zone_t zone, void *item) { uma_slab_t slab; uma_keg_t keg; uint8_t *mem; mem = (uint8_t *)((uintptr_t)item & (~UMA_SLAB_MASK)); if (zone->uz_flags & UMA_ZONE_VTOSLAB) { slab = vtoslab((vm_offset_t)mem); } else { /* * It is safe to return the slab here even though the * zone is unlocked because the item's allocation state * essentially holds a reference. */ ZONE_LOCK(zone); keg = LIST_FIRST(&zone->uz_kegs)->kl_keg; if (keg->uk_flags & UMA_ZONE_HASH) slab = hash_sfind(&keg->uk_hash, mem); else slab = (uma_slab_t)(mem + keg->uk_pgoff); ZONE_UNLOCK(zone); } return (slab); } /* * Set up the slab's freei data such that uma_dbg_free can function. * */ static void uma_dbg_alloc(uma_zone_t zone, uma_slab_t slab, void *item) { uma_keg_t keg; int freei; if (zone_first_keg(zone) == NULL) return; if (slab == NULL) { slab = uma_dbg_getslab(zone, item); if (slab == NULL) panic("uma: item %p did not belong to zone %s\n", item, zone->uz_name); } keg = slab->us_keg; freei = ((uintptr_t)item - (uintptr_t)slab->us_data) / keg->uk_rsize; if (BIT_ISSET(SLAB_SETSIZE, freei, &slab->us_debugfree)) panic("Duplicate alloc of %p from zone %p(%s) slab %p(%d)\n", item, zone, zone->uz_name, slab, freei); BIT_SET_ATOMIC(SLAB_SETSIZE, freei, &slab->us_debugfree); return; } /* * Verifies freed addresses. Checks for alignment, valid slab membership * and duplicate frees. * */ static void uma_dbg_free(uma_zone_t zone, uma_slab_t slab, void *item) { uma_keg_t keg; int freei; if (zone_first_keg(zone) == NULL) return; if (slab == NULL) { slab = uma_dbg_getslab(zone, item); if (slab == NULL) panic("uma: Freed item %p did not belong to zone %s\n", item, zone->uz_name); } keg = slab->us_keg; freei = ((uintptr_t)item - (uintptr_t)slab->us_data) / keg->uk_rsize; if (freei >= keg->uk_ipers) panic("Invalid free of %p from zone %p(%s) slab %p(%d)\n", item, zone, zone->uz_name, slab, freei); if (((freei * keg->uk_rsize) + slab->us_data) != item) panic("Unaligned free of %p from zone %p(%s) slab %p(%d)\n", item, zone, zone->uz_name, slab, freei); if (!BIT_ISSET(SLAB_SETSIZE, freei, &slab->us_debugfree)) panic("Duplicate free of %p from zone %p(%s) slab %p(%d)\n", item, zone, zone->uz_name, slab, freei); BIT_CLR_ATOMIC(SLAB_SETSIZE, freei, &slab->us_debugfree); } #endif /* INVARIANTS */ #ifdef DDB DB_SHOW_COMMAND(uma, db_show_uma) { uint64_t allocs, frees, sleeps; uma_bucket_t bucket; uma_keg_t kz; uma_zone_t z; int cachefree; db_printf("%18s %8s %8s %8s %12s %8s %8s\n", "Zone", "Size", "Used", "Free", "Requests", "Sleeps", "Bucket"); LIST_FOREACH(kz, &uma_kegs, uk_link) { LIST_FOREACH(z, &kz->uk_zones, uz_link) { if (kz->uk_flags & UMA_ZFLAG_INTERNAL) { allocs = z->uz_allocs; frees = z->uz_frees; sleeps = z->uz_sleeps; cachefree = 0; } else uma_zone_sumstat(z, &cachefree, &allocs, &frees, &sleeps); if (!((z->uz_flags & UMA_ZONE_SECONDARY) && (LIST_FIRST(&kz->uk_zones) != z))) cachefree += kz->uk_free; LIST_FOREACH(bucket, &z->uz_buckets, ub_link) cachefree += bucket->ub_cnt; db_printf("%18s %8ju %8jd %8d %12ju %8ju %8u\n", z->uz_name, (uintmax_t)kz->uk_size, (intmax_t)(allocs - frees), cachefree, (uintmax_t)allocs, sleeps, z->uz_count); if (db_pager_quit) return; } } } DB_SHOW_COMMAND(umacache, db_show_umacache) { uint64_t allocs, frees; uma_bucket_t bucket; uma_zone_t z; int cachefree; db_printf("%18s %8s %8s %8s %12s %8s\n", "Zone", "Size", "Used", "Free", "Requests", "Bucket"); LIST_FOREACH(z, &uma_cachezones, uz_link) { uma_zone_sumstat(z, &cachefree, &allocs, &frees, NULL); LIST_FOREACH(bucket, &z->uz_buckets, ub_link) cachefree += bucket->ub_cnt; db_printf("%18s %8ju %8jd %8d %12ju %8u\n", z->uz_name, (uintmax_t)z->uz_size, (intmax_t)(allocs - frees), cachefree, (uintmax_t)allocs, z->uz_count); if (db_pager_quit) return; } } #endif /* DDB */ Index: stable/11/sys/vm/vm_page.c =================================================================== --- stable/11/sys/vm/vm_page.c (revision 327403) +++ stable/11/sys/vm/vm_page.c (revision 327404) @@ -1,3838 +1,3838 @@ /*- * Copyright (c) 1991 Regents of the University of California. * All rights reserved. * Copyright (c) 1998 Matthew Dillon. All Rights Reserved. * * This code is derived from software contributed to Berkeley by * The Mach Operating System project at Carnegie-Mellon University. * * 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 * 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: @(#)vm_page.c 7.4 (Berkeley) 5/7/91 */ /*- * Copyright (c) 1987, 1990 Carnegie-Mellon University. * All rights reserved. * * Authors: Avadis Tevanian, Jr., Michael Wayne Young * * Permission to use, copy, modify and distribute this software and * its documentation is hereby granted, provided that both the copyright * notice and this permission notice appear in all copies of the * software, derivative works or modified versions, and any portions * thereof, and that both notices appear in supporting documentation. * * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. * * Carnegie Mellon requests users of this software to return to * * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU * School of Computer Science * Carnegie Mellon University * Pittsburgh PA 15213-3890 * * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. */ /* * GENERAL RULES ON VM_PAGE MANIPULATION * * - A page queue lock is required when adding or removing a page from a * page queue regardless of other locks or the busy state of a page. * * * In general, no thread besides the page daemon can acquire or * hold more than one page queue lock at a time. * * * The page daemon can acquire and hold any pair of page queue * locks in any order. * * - The object lock is required when inserting or removing * pages from an object (vm_page_insert() or vm_page_remove()). * */ /* * Resident memory management module. */ #include __FBSDID("$FreeBSD$"); #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 #include #include #include #include #include #include #include #include /* * Associated with page of user-allocatable memory is a * page structure. */ struct vm_domain vm_dom[MAXMEMDOM]; -struct mtx_padalign vm_page_queue_free_mtx; +struct mtx_padalign __exclusive_cache_line vm_page_queue_free_mtx; -struct mtx_padalign pa_lock[PA_LOCK_COUNT]; +struct mtx_padalign __exclusive_cache_line pa_lock[PA_LOCK_COUNT]; vm_page_t vm_page_array; long vm_page_array_size; long first_page; int vm_page_zero_count; static int boot_pages = UMA_BOOT_PAGES; SYSCTL_INT(_vm, OID_AUTO, boot_pages, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &boot_pages, 0, "number of pages allocated for bootstrapping the VM system"); static int pa_tryrelock_restart; SYSCTL_INT(_vm, OID_AUTO, tryrelock_restart, CTLFLAG_RD, &pa_tryrelock_restart, 0, "Number of tryrelock restarts"); static TAILQ_HEAD(, vm_page) blacklist_head; static int sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_vm, OID_AUTO, page_blacklist, CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_page_blacklist, "A", "Blacklist pages"); /* Is the page daemon waiting for free pages? */ static int vm_pageout_pages_needed; static uma_zone_t fakepg_zone; static void vm_page_alloc_check(vm_page_t m); static void vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits); static void vm_page_enqueue(uint8_t queue, vm_page_t m); static void vm_page_free_phys(vm_page_t m); static void vm_page_free_wakeup(void); static void vm_page_init_fakepg(void *dummy); static int vm_page_insert_after(vm_page_t m, vm_object_t object, vm_pindex_t pindex, vm_page_t mpred); static void vm_page_insert_radixdone(vm_page_t m, vm_object_t object, vm_page_t mpred); static int vm_page_reclaim_run(int req_class, u_long npages, vm_page_t m_run, vm_paddr_t high); SYSINIT(vm_page, SI_SUB_VM, SI_ORDER_SECOND, vm_page_init_fakepg, NULL); static void vm_page_init_fakepg(void *dummy) { fakepg_zone = uma_zcreate("fakepg", sizeof(struct vm_page), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE | UMA_ZONE_VM); } /* Make sure that u_long is at least 64 bits when PAGE_SIZE is 32K. */ #if PAGE_SIZE == 32768 #ifdef CTASSERT CTASSERT(sizeof(u_long) >= 8); #endif #endif /* * Try to acquire a physical address lock while a pmap is locked. If we * fail to trylock we unlock and lock the pmap directly and cache the * locked pa in *locked. The caller should then restart their loop in case * the virtual to physical mapping has changed. */ int vm_page_pa_tryrelock(pmap_t pmap, vm_paddr_t pa, vm_paddr_t *locked) { vm_paddr_t lockpa; lockpa = *locked; *locked = pa; if (lockpa) { PA_LOCK_ASSERT(lockpa, MA_OWNED); if (PA_LOCKPTR(pa) == PA_LOCKPTR(lockpa)) return (0); PA_UNLOCK(lockpa); } if (PA_TRYLOCK(pa)) return (0); PMAP_UNLOCK(pmap); atomic_add_int(&pa_tryrelock_restart, 1); PA_LOCK(pa); PMAP_LOCK(pmap); return (EAGAIN); } /* * vm_set_page_size: * * Sets the page size, perhaps based upon the memory * size. Must be called before any use of page-size * dependent functions. */ void vm_set_page_size(void) { if (vm_cnt.v_page_size == 0) vm_cnt.v_page_size = PAGE_SIZE; if (((vm_cnt.v_page_size - 1) & vm_cnt.v_page_size) != 0) panic("vm_set_page_size: page size not a power of two"); } /* * vm_page_blacklist_next: * * Find the next entry in the provided string of blacklist * addresses. Entries are separated by space, comma, or newline. * If an invalid integer is encountered then the rest of the * string is skipped. Updates the list pointer to the next * character, or NULL if the string is exhausted or invalid. */ static vm_paddr_t vm_page_blacklist_next(char **list, char *end) { vm_paddr_t bad; char *cp, *pos; if (list == NULL || *list == NULL) return (0); if (**list =='\0') { *list = NULL; return (0); } /* * If there's no end pointer then the buffer is coming from * the kenv and we know it's null-terminated. */ if (end == NULL) end = *list + strlen(*list); /* Ensure that strtoq() won't walk off the end */ if (*end != '\0') { if (*end == '\n' || *end == ' ' || *end == ',') *end = '\0'; else { printf("Blacklist not terminated, skipping\n"); *list = NULL; return (0); } } for (pos = *list; *pos != '\0'; pos = cp) { bad = strtoq(pos, &cp, 0); if (*cp == '\0' || *cp == ' ' || *cp == ',' || *cp == '\n') { if (bad == 0) { if (++cp < end) continue; else break; } } else break; if (*cp == '\0' || ++cp >= end) *list = NULL; else *list = cp; return (trunc_page(bad)); } printf("Garbage in RAM blacklist, skipping\n"); *list = NULL; return (0); } /* * vm_page_blacklist_check: * * Iterate through the provided string of blacklist addresses, pulling * each entry out of the physical allocator free list and putting it * onto a list for reporting via the vm.page_blacklist sysctl. */ static void vm_page_blacklist_check(char *list, char *end) { vm_paddr_t pa; vm_page_t m; char *next; int ret; next = list; while (next != NULL) { if ((pa = vm_page_blacklist_next(&next, end)) == 0) continue; m = vm_phys_paddr_to_vm_page(pa); if (m == NULL) continue; mtx_lock(&vm_page_queue_free_mtx); ret = vm_phys_unfree_page(m); mtx_unlock(&vm_page_queue_free_mtx); if (ret == TRUE) { TAILQ_INSERT_TAIL(&blacklist_head, m, listq); if (bootverbose) printf("Skipping page with pa 0x%jx\n", (uintmax_t)pa); } } } /* * vm_page_blacklist_load: * * Search for a special module named "ram_blacklist". It'll be a * plain text file provided by the user via the loader directive * of the same name. */ static void vm_page_blacklist_load(char **list, char **end) { void *mod; u_char *ptr; u_int len; mod = NULL; ptr = NULL; mod = preload_search_by_type("ram_blacklist"); if (mod != NULL) { ptr = preload_fetch_addr(mod); len = preload_fetch_size(mod); } *list = ptr; if (ptr != NULL) *end = ptr + len; else *end = NULL; return; } static int sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS) { vm_page_t m; struct sbuf sbuf; int error, first; first = 1; error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); sbuf_new_for_sysctl(&sbuf, NULL, 128, req); TAILQ_FOREACH(m, &blacklist_head, listq) { sbuf_printf(&sbuf, "%s%#jx", first ? "" : ",", (uintmax_t)m->phys_addr); first = 0; } error = sbuf_finish(&sbuf); sbuf_delete(&sbuf); return (error); } static void vm_page_domain_init(struct vm_domain *vmd) { struct vm_pagequeue *pq; int i; *__DECONST(char **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_name) = "vm inactive pagequeue"; *__DECONST(u_int **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_vcnt) = &vm_cnt.v_inactive_count; *__DECONST(char **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_name) = "vm active pagequeue"; *__DECONST(u_int **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_vcnt) = &vm_cnt.v_active_count; *__DECONST(char **, &vmd->vmd_pagequeues[PQ_LAUNDRY].pq_name) = "vm laundry pagequeue"; *__DECONST(int **, &vmd->vmd_pagequeues[PQ_LAUNDRY].pq_vcnt) = &vm_cnt.v_laundry_count; vmd->vmd_page_count = 0; vmd->vmd_free_count = 0; vmd->vmd_segs = 0; vmd->vmd_oom = FALSE; for (i = 0; i < PQ_COUNT; i++) { pq = &vmd->vmd_pagequeues[i]; TAILQ_INIT(&pq->pq_pl); mtx_init(&pq->pq_mutex, pq->pq_name, "vm pagequeue", MTX_DEF | MTX_DUPOK); } } /* * Initialize a physical page in preparation for adding it to the free * lists. */ static void vm_page_init_page(vm_page_t m, vm_paddr_t pa, int segind) { m->object = NULL; m->wire_count = 0; m->busy_lock = VPB_UNBUSIED; m->hold_count = 0; m->flags = 0; m->phys_addr = pa; m->queue = PQ_NONE; m->psind = 0; m->segind = segind; m->order = VM_NFREEORDER; m->pool = VM_FREEPOOL_DEFAULT; m->valid = m->dirty = 0; pmap_page_init(m); } /* * vm_page_startup: * * Initializes the resident memory module. Allocates physical memory for * bootstrapping UMA and some data structures that are used to manage * physical pages. Initializes these structures, and populates the free * page queues. */ vm_offset_t vm_page_startup(vm_offset_t vaddr) { struct vm_domain *vmd; struct vm_phys_seg *seg; vm_page_t m; char *list, *listend; vm_offset_t mapped; vm_paddr_t end, high_avail, low_avail, new_end, page_range, size; vm_paddr_t biggestsize, last_pa, pa; u_long pagecount; int biggestone, i, pages_per_zone, segind; biggestsize = 0; biggestone = 0; vaddr = round_page(vaddr); for (i = 0; phys_avail[i + 1]; i += 2) { phys_avail[i] = round_page(phys_avail[i]); phys_avail[i + 1] = trunc_page(phys_avail[i + 1]); } for (i = 0; phys_avail[i + 1]; i += 2) { size = phys_avail[i + 1] - phys_avail[i]; if (size > biggestsize) { biggestone = i; biggestsize = size; } } end = phys_avail[biggestone+1]; /* * Initialize the page and queue locks. */ mtx_init(&vm_page_queue_free_mtx, "vm page free queue", NULL, MTX_DEF); for (i = 0; i < PA_LOCK_COUNT; i++) mtx_init(&pa_lock[i], "vm page", NULL, MTX_DEF); for (i = 0; i < vm_ndomains; i++) vm_page_domain_init(&vm_dom[i]); /* * Almost all of the pages needed for bootstrapping UMA are used * for zone structures, so if the number of CPUs results in those * structures taking more than one page each, we set aside more pages * in proportion to the zone structure size. */ pages_per_zone = howmany(sizeof(struct uma_zone) + sizeof(struct uma_cache) * (mp_maxid + 1) + roundup2(sizeof(struct uma_slab), sizeof(void *)), UMA_SLAB_SIZE); if (pages_per_zone > 1) { /* Reserve more pages so that we don't run out. */ boot_pages = UMA_BOOT_PAGES_ZONES * pages_per_zone; } /* * Allocate memory for use when boot strapping the kernel memory * allocator. * * CTFLAG_RDTUN doesn't work during the early boot process, so we must * manually fetch the value. */ TUNABLE_INT_FETCH("vm.boot_pages", &boot_pages); new_end = end - (boot_pages * UMA_SLAB_SIZE); new_end = trunc_page(new_end); mapped = pmap_map(&vaddr, new_end, end, VM_PROT_READ | VM_PROT_WRITE); bzero((void *)mapped, end - new_end); uma_startup((void *)mapped, boot_pages); #if defined(__aarch64__) || defined(__amd64__) || defined(__arm__) || \ defined(__i386__) || defined(__mips__) /* * Allocate a bitmap to indicate that a random physical page * needs to be included in a minidump. * * The amd64 port needs this to indicate which direct map pages * need to be dumped, via calls to dump_add_page()/dump_drop_page(). * * However, i386 still needs this workspace internally within the * minidump code. In theory, they are not needed on i386, but are * included should the sf_buf code decide to use them. */ last_pa = 0; for (i = 0; dump_avail[i + 1] != 0; i += 2) if (dump_avail[i + 1] > last_pa) last_pa = dump_avail[i + 1]; page_range = last_pa / PAGE_SIZE; vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY); new_end -= vm_page_dump_size; vm_page_dump = (void *)(uintptr_t)pmap_map(&vaddr, new_end, new_end + vm_page_dump_size, VM_PROT_READ | VM_PROT_WRITE); bzero((void *)vm_page_dump, vm_page_dump_size); #else (void)last_pa; #endif #if defined(__aarch64__) || defined(__amd64__) || defined(__mips__) /* * Include the UMA bootstrap pages and vm_page_dump in a crash dump. * When pmap_map() uses the direct map, they are not automatically * included. */ for (pa = new_end; pa < end; pa += PAGE_SIZE) dump_add_page(pa); #endif phys_avail[biggestone + 1] = new_end; #ifdef __amd64__ /* * Request that the physical pages underlying the message buffer be * included in a crash dump. Since the message buffer is accessed * through the direct map, they are not automatically included. */ pa = DMAP_TO_PHYS((vm_offset_t)msgbufp->msg_ptr); last_pa = pa + round_page(msgbufsize); while (pa < last_pa) { dump_add_page(pa); pa += PAGE_SIZE; } #endif /* * Compute the number of pages of memory that will be available for * use, taking into account the overhead of a page structure per page. * In other words, solve * "available physical memory" - round_page(page_range * * sizeof(struct vm_page)) = page_range * PAGE_SIZE * for page_range. */ low_avail = phys_avail[0]; high_avail = phys_avail[1]; for (i = 0; i < vm_phys_nsegs; i++) { if (vm_phys_segs[i].start < low_avail) low_avail = vm_phys_segs[i].start; if (vm_phys_segs[i].end > high_avail) high_avail = vm_phys_segs[i].end; } /* Skip the first chunk. It is already accounted for. */ for (i = 2; phys_avail[i + 1] != 0; i += 2) { if (phys_avail[i] < low_avail) low_avail = phys_avail[i]; if (phys_avail[i + 1] > high_avail) high_avail = phys_avail[i + 1]; } first_page = low_avail / PAGE_SIZE; #ifdef VM_PHYSSEG_SPARSE size = 0; for (i = 0; i < vm_phys_nsegs; i++) size += vm_phys_segs[i].end - vm_phys_segs[i].start; for (i = 0; phys_avail[i + 1] != 0; i += 2) size += phys_avail[i + 1] - phys_avail[i]; #elif defined(VM_PHYSSEG_DENSE) size = high_avail - low_avail; #else #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined." #endif #ifdef VM_PHYSSEG_DENSE /* * In the VM_PHYSSEG_DENSE case, the number of pages can account for * the overhead of a page structure per page only if vm_page_array is * allocated from the last physical memory chunk. Otherwise, we must * allocate page structures representing the physical memory * underlying vm_page_array, even though they will not be used. */ if (new_end != high_avail) page_range = size / PAGE_SIZE; else #endif { page_range = size / (PAGE_SIZE + sizeof(struct vm_page)); /* * If the partial bytes remaining are large enough for * a page (PAGE_SIZE) without a corresponding * 'struct vm_page', then new_end will contain an * extra page after subtracting the length of the VM * page array. Compensate by subtracting an extra * page from new_end. */ if (size % (PAGE_SIZE + sizeof(struct vm_page)) >= PAGE_SIZE) { if (new_end == high_avail) high_avail -= PAGE_SIZE; new_end -= PAGE_SIZE; } } end = new_end; /* * Reserve an unmapped guard page to trap access to vm_page_array[-1]. * However, because this page is allocated from KVM, out-of-bounds * accesses using the direct map will not be trapped. */ vaddr += PAGE_SIZE; /* * Allocate physical memory for the page structures, and map it. */ new_end = trunc_page(end - page_range * sizeof(struct vm_page)); mapped = pmap_map(&vaddr, new_end, end, VM_PROT_READ | VM_PROT_WRITE); vm_page_array = (vm_page_t)mapped; vm_page_array_size = page_range; #if VM_NRESERVLEVEL > 0 /* * Allocate physical memory for the reservation management system's * data structures, and map it. */ if (high_avail == end) high_avail = new_end; new_end = vm_reserv_startup(&vaddr, new_end, high_avail); #endif #if defined(__aarch64__) || defined(__amd64__) || defined(__mips__) /* * Include vm_page_array and vm_reserv_array in a crash dump. */ for (pa = new_end; pa < end; pa += PAGE_SIZE) dump_add_page(pa); #endif phys_avail[biggestone + 1] = new_end; /* * Add physical memory segments corresponding to the available * physical pages. */ for (i = 0; phys_avail[i + 1] != 0; i += 2) vm_phys_add_seg(phys_avail[i], phys_avail[i + 1]); /* * Initialize the physical memory allocator. */ vm_phys_init(); /* * Initialize the page structures and add every available page to the * physical memory allocator's free lists. */ vm_cnt.v_page_count = 0; vm_cnt.v_free_count = 0; for (segind = 0; segind < vm_phys_nsegs; segind++) { seg = &vm_phys_segs[segind]; for (m = seg->first_page, pa = seg->start; pa < seg->end; m++, pa += PAGE_SIZE) vm_page_init_page(m, pa, segind); /* * Add the segment to the free lists only if it is covered by * one of the ranges in phys_avail. Because we've added the * ranges to the vm_phys_segs array, we can assume that each * segment is either entirely contained in one of the ranges, * or doesn't overlap any of them. */ for (i = 0; phys_avail[i + 1] != 0; i += 2) { if (seg->start < phys_avail[i] || seg->end > phys_avail[i + 1]) continue; m = seg->first_page; pagecount = (u_long)atop(seg->end - seg->start); mtx_lock(&vm_page_queue_free_mtx); vm_phys_free_contig(m, pagecount); vm_phys_freecnt_adj(m, (int)pagecount); mtx_unlock(&vm_page_queue_free_mtx); vm_cnt.v_page_count += (u_int)pagecount; vmd = &vm_dom[seg->domain]; vmd->vmd_page_count += (u_int)pagecount; vmd->vmd_segs |= 1UL << m->segind; break; } } /* * Remove blacklisted pages from the physical memory allocator. */ TAILQ_INIT(&blacklist_head); vm_page_blacklist_load(&list, &listend); vm_page_blacklist_check(list, listend); list = kern_getenv("vm.blacklist"); vm_page_blacklist_check(list, NULL); freeenv(list); #if VM_NRESERVLEVEL > 0 /* * Initialize the reservation management system. */ vm_reserv_init(); #endif return (vaddr); } void vm_page_reference(vm_page_t m) { vm_page_aflag_set(m, PGA_REFERENCED); } /* * vm_page_busy_downgrade: * * Downgrade an exclusive busy page into a single shared busy page. */ void vm_page_busy_downgrade(vm_page_t m) { u_int x; bool locked; vm_page_assert_xbusied(m); locked = mtx_owned(vm_page_lockptr(m)); for (;;) { x = m->busy_lock; x &= VPB_BIT_WAITERS; if (x != 0 && !locked) vm_page_lock(m); if (atomic_cmpset_rel_int(&m->busy_lock, VPB_SINGLE_EXCLUSIVER | x, VPB_SHARERS_WORD(1))) break; if (x != 0 && !locked) vm_page_unlock(m); } if (x != 0) { wakeup(m); if (!locked) vm_page_unlock(m); } } /* * vm_page_sbusied: * * Return a positive value if the page is shared busied, 0 otherwise. */ int vm_page_sbusied(vm_page_t m) { u_int x; x = m->busy_lock; return ((x & VPB_BIT_SHARED) != 0 && x != VPB_UNBUSIED); } /* * vm_page_sunbusy: * * Shared unbusy a page. */ void vm_page_sunbusy(vm_page_t m) { u_int x; vm_page_lock_assert(m, MA_NOTOWNED); vm_page_assert_sbusied(m); for (;;) { x = m->busy_lock; if (VPB_SHARERS(x) > 1) { if (atomic_cmpset_int(&m->busy_lock, x, x - VPB_ONE_SHARER)) break; continue; } if ((x & VPB_BIT_WAITERS) == 0) { KASSERT(x == VPB_SHARERS_WORD(1), ("vm_page_sunbusy: invalid lock state")); if (atomic_cmpset_int(&m->busy_lock, VPB_SHARERS_WORD(1), VPB_UNBUSIED)) break; continue; } KASSERT(x == (VPB_SHARERS_WORD(1) | VPB_BIT_WAITERS), ("vm_page_sunbusy: invalid lock state for waiters")); vm_page_lock(m); if (!atomic_cmpset_int(&m->busy_lock, x, VPB_UNBUSIED)) { vm_page_unlock(m); continue; } wakeup(m); vm_page_unlock(m); break; } } /* * vm_page_busy_sleep: * * Sleep and release the page lock, using the page pointer as wchan. * This is used to implement the hard-path of busying mechanism. * * The given page must be locked. * * If nonshared is true, sleep only if the page is xbusy. */ void vm_page_busy_sleep(vm_page_t m, const char *wmesg, bool nonshared) { u_int x; vm_page_assert_locked(m); x = m->busy_lock; if (x == VPB_UNBUSIED || (nonshared && (x & VPB_BIT_SHARED) != 0) || ((x & VPB_BIT_WAITERS) == 0 && !atomic_cmpset_int(&m->busy_lock, x, x | VPB_BIT_WAITERS))) { vm_page_unlock(m); return; } msleep(m, vm_page_lockptr(m), PVM | PDROP, wmesg, 0); } /* * vm_page_trysbusy: * * Try to shared busy a page. * If the operation succeeds 1 is returned otherwise 0. * The operation never sleeps. */ int vm_page_trysbusy(vm_page_t m) { u_int x; for (;;) { x = m->busy_lock; if ((x & VPB_BIT_SHARED) == 0) return (0); if (atomic_cmpset_acq_int(&m->busy_lock, x, x + VPB_ONE_SHARER)) return (1); } } static void vm_page_xunbusy_locked(vm_page_t m) { vm_page_assert_xbusied(m); vm_page_assert_locked(m); atomic_store_rel_int(&m->busy_lock, VPB_UNBUSIED); /* There is a waiter, do wakeup() instead of vm_page_flash(). */ wakeup(m); } void vm_page_xunbusy_maybelocked(vm_page_t m) { bool lockacq; vm_page_assert_xbusied(m); /* * Fast path for unbusy. If it succeeds, we know that there * are no waiters, so we do not need a wakeup. */ if (atomic_cmpset_rel_int(&m->busy_lock, VPB_SINGLE_EXCLUSIVER, VPB_UNBUSIED)) return; lockacq = !mtx_owned(vm_page_lockptr(m)); if (lockacq) vm_page_lock(m); vm_page_xunbusy_locked(m); if (lockacq) vm_page_unlock(m); } /* * vm_page_xunbusy_hard: * * Called after the first try the exclusive unbusy of a page failed. * It is assumed that the waiters bit is on. */ void vm_page_xunbusy_hard(vm_page_t m) { vm_page_assert_xbusied(m); vm_page_lock(m); vm_page_xunbusy_locked(m); vm_page_unlock(m); } /* * vm_page_flash: * * Wakeup anyone waiting for the page. * The ownership bits do not change. * * The given page must be locked. */ void vm_page_flash(vm_page_t m) { u_int x; vm_page_lock_assert(m, MA_OWNED); for (;;) { x = m->busy_lock; if ((x & VPB_BIT_WAITERS) == 0) return; if (atomic_cmpset_int(&m->busy_lock, x, x & (~VPB_BIT_WAITERS))) break; } wakeup(m); } /* * Avoid releasing and reacquiring the same page lock. */ void vm_page_change_lock(vm_page_t m, struct mtx **mtx) { struct mtx *mtx1; mtx1 = vm_page_lockptr(m); if (*mtx == mtx1) return; if (*mtx != NULL) mtx_unlock(*mtx); *mtx = mtx1; mtx_lock(mtx1); } /* * Keep page from being freed by the page daemon * much of the same effect as wiring, except much lower * overhead and should be used only for *very* temporary * holding ("wiring"). */ void vm_page_hold(vm_page_t mem) { vm_page_lock_assert(mem, MA_OWNED); mem->hold_count++; } void vm_page_unhold(vm_page_t mem) { vm_page_lock_assert(mem, MA_OWNED); KASSERT(mem->hold_count >= 1, ("vm_page_unhold: hold count < 0!!!")); --mem->hold_count; if (mem->hold_count == 0 && (mem->flags & PG_UNHOLDFREE) != 0) vm_page_free_toq(mem); } /* * vm_page_unhold_pages: * * Unhold each of the pages that is referenced by the given array. */ void vm_page_unhold_pages(vm_page_t *ma, int count) { struct mtx *mtx; mtx = NULL; for (; count != 0; count--) { vm_page_change_lock(*ma, &mtx); vm_page_unhold(*ma); ma++; } if (mtx != NULL) mtx_unlock(mtx); } vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa) { vm_page_t m; #ifdef VM_PHYSSEG_SPARSE m = vm_phys_paddr_to_vm_page(pa); if (m == NULL) m = vm_phys_fictitious_to_vm_page(pa); return (m); #elif defined(VM_PHYSSEG_DENSE) long pi; pi = atop(pa); if (pi >= first_page && (pi - first_page) < vm_page_array_size) { m = &vm_page_array[pi - first_page]; return (m); } return (vm_phys_fictitious_to_vm_page(pa)); #else #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined." #endif } /* * vm_page_getfake: * * Create a fictitious page with the specified physical address and * memory attribute. The memory attribute is the only the machine- * dependent aspect of a fictitious page that must be initialized. */ vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr) { vm_page_t m; m = uma_zalloc(fakepg_zone, M_WAITOK | M_ZERO); vm_page_initfake(m, paddr, memattr); return (m); } void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr) { if ((m->flags & PG_FICTITIOUS) != 0) { /* * The page's memattr might have changed since the * previous initialization. Update the pmap to the * new memattr. */ goto memattr; } m->phys_addr = paddr; m->queue = PQ_NONE; /* Fictitious pages don't use "segind". */ m->flags = PG_FICTITIOUS; /* Fictitious pages don't use "order" or "pool". */ m->oflags = VPO_UNMANAGED; m->busy_lock = VPB_SINGLE_EXCLUSIVER; m->wire_count = 1; pmap_page_init(m); memattr: pmap_page_set_memattr(m, memattr); } /* * vm_page_putfake: * * Release a fictitious page. */ void vm_page_putfake(vm_page_t m) { KASSERT((m->oflags & VPO_UNMANAGED) != 0, ("managed %p", m)); KASSERT((m->flags & PG_FICTITIOUS) != 0, ("vm_page_putfake: bad page %p", m)); uma_zfree(fakepg_zone, m); } /* * vm_page_updatefake: * * Update the given fictitious page to the specified physical address and * memory attribute. */ void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr) { KASSERT((m->flags & PG_FICTITIOUS) != 0, ("vm_page_updatefake: bad page %p", m)); m->phys_addr = paddr; pmap_page_set_memattr(m, memattr); } /* * vm_page_free: * * Free a page. */ void vm_page_free(vm_page_t m) { m->flags &= ~PG_ZERO; vm_page_free_toq(m); } /* * vm_page_free_zero: * * Free a page to the zerod-pages queue */ void vm_page_free_zero(vm_page_t m) { m->flags |= PG_ZERO; vm_page_free_toq(m); } /* * Unbusy and handle the page queueing for a page from a getpages request that * was optionally read ahead or behind. */ void vm_page_readahead_finish(vm_page_t m) { /* We shouldn't put invalid pages on queues. */ KASSERT(m->valid != 0, ("%s: %p is invalid", __func__, m)); /* * Since the page is not the actually needed one, whether it should * be activated or deactivated is not obvious. Empirical results * have shown that deactivating the page is usually the best choice, * unless the page is wanted by another thread. */ vm_page_lock(m); if ((m->busy_lock & VPB_BIT_WAITERS) != 0) vm_page_activate(m); else vm_page_deactivate(m); vm_page_unlock(m); vm_page_xunbusy(m); } /* * vm_page_sleep_if_busy: * * Sleep and release the page queues lock if the page is busied. * Returns TRUE if the thread slept. * * The given page must be unlocked and object containing it must * be locked. */ int vm_page_sleep_if_busy(vm_page_t m, const char *msg) { vm_object_t obj; vm_page_lock_assert(m, MA_NOTOWNED); VM_OBJECT_ASSERT_WLOCKED(m->object); if (vm_page_busied(m)) { /* * The page-specific object must be cached because page * identity can change during the sleep, causing the * re-lock of a different object. * It is assumed that a reference to the object is already * held by the callers. */ obj = m->object; vm_page_lock(m); VM_OBJECT_WUNLOCK(obj); vm_page_busy_sleep(m, msg, false); VM_OBJECT_WLOCK(obj); return (TRUE); } return (FALSE); } /* * vm_page_dirty_KBI: [ internal use only ] * * Set all bits in the page's dirty field. * * The object containing the specified page must be locked if the * call is made from the machine-independent layer. * * See vm_page_clear_dirty_mask(). * * This function should only be called by vm_page_dirty(). */ void vm_page_dirty_KBI(vm_page_t m) { /* Refer to this operation by its public name. */ KASSERT(m->valid == VM_PAGE_BITS_ALL, ("vm_page_dirty: page is invalid!")); m->dirty = VM_PAGE_BITS_ALL; } /* * vm_page_insert: [ internal use only ] * * Inserts the given mem entry into the object and object list. * * The object must be locked. */ int vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex) { vm_page_t mpred; VM_OBJECT_ASSERT_WLOCKED(object); mpred = vm_radix_lookup_le(&object->rtree, pindex); return (vm_page_insert_after(m, object, pindex, mpred)); } /* * vm_page_insert_after: * * Inserts the page "m" into the specified object at offset "pindex". * * The page "mpred" must immediately precede the offset "pindex" within * the specified object. * * The object must be locked. */ static int vm_page_insert_after(vm_page_t m, vm_object_t object, vm_pindex_t pindex, vm_page_t mpred) { vm_page_t msucc; VM_OBJECT_ASSERT_WLOCKED(object); KASSERT(m->object == NULL, ("vm_page_insert_after: page already inserted")); if (mpred != NULL) { KASSERT(mpred->object == object, ("vm_page_insert_after: object doesn't contain mpred")); KASSERT(mpred->pindex < pindex, ("vm_page_insert_after: mpred doesn't precede pindex")); msucc = TAILQ_NEXT(mpred, listq); } else msucc = TAILQ_FIRST(&object->memq); if (msucc != NULL) KASSERT(msucc->pindex > pindex, ("vm_page_insert_after: msucc doesn't succeed pindex")); /* * Record the object/offset pair in this page */ m->object = object; m->pindex = pindex; /* * Now link into the object's ordered list of backed pages. */ if (vm_radix_insert(&object->rtree, m)) { m->object = NULL; m->pindex = 0; return (1); } vm_page_insert_radixdone(m, object, mpred); return (0); } /* * vm_page_insert_radixdone: * * Complete page "m" insertion into the specified object after the * radix trie hooking. * * The page "mpred" must precede the offset "m->pindex" within the * specified object. * * The object must be locked. */ static void vm_page_insert_radixdone(vm_page_t m, vm_object_t object, vm_page_t mpred) { VM_OBJECT_ASSERT_WLOCKED(object); KASSERT(object != NULL && m->object == object, ("vm_page_insert_radixdone: page %p has inconsistent object", m)); if (mpred != NULL) { KASSERT(mpred->object == object, ("vm_page_insert_after: object doesn't contain mpred")); KASSERT(mpred->pindex < m->pindex, ("vm_page_insert_after: mpred doesn't precede pindex")); } if (mpred != NULL) TAILQ_INSERT_AFTER(&object->memq, mpred, m, listq); else TAILQ_INSERT_HEAD(&object->memq, m, listq); /* * Show that the object has one more resident page. */ object->resident_page_count++; /* * Hold the vnode until the last page is released. */ if (object->resident_page_count == 1 && object->type == OBJT_VNODE) vhold(object->handle); /* * Since we are inserting a new and possibly dirty page, * update the object's OBJ_MIGHTBEDIRTY flag. */ if (pmap_page_is_write_mapped(m)) vm_object_set_writeable_dirty(object); } /* * vm_page_remove: * * Removes the specified page from its containing object, but does not * invalidate any backing storage. * * The object must be locked. The page must be locked if it is managed. */ void vm_page_remove(vm_page_t m) { vm_object_t object; vm_page_t mrem; if ((m->oflags & VPO_UNMANAGED) == 0) vm_page_assert_locked(m); if ((object = m->object) == NULL) return; VM_OBJECT_ASSERT_WLOCKED(object); if (vm_page_xbusied(m)) vm_page_xunbusy_maybelocked(m); mrem = vm_radix_remove(&object->rtree, m->pindex); KASSERT(mrem == m, ("removed page %p, expected page %p", mrem, m)); /* * Now remove from the object's list of backed pages. */ TAILQ_REMOVE(&object->memq, m, listq); /* * And show that the object has one fewer resident page. */ object->resident_page_count--; /* * The vnode may now be recycled. */ if (object->resident_page_count == 0 && object->type == OBJT_VNODE) vdrop(object->handle); m->object = NULL; } /* * vm_page_lookup: * * Returns the page associated with the object/offset * pair specified; if none is found, NULL is returned. * * The object must be locked. */ vm_page_t vm_page_lookup(vm_object_t object, vm_pindex_t pindex) { VM_OBJECT_ASSERT_LOCKED(object); return (vm_radix_lookup(&object->rtree, pindex)); } /* * vm_page_find_least: * * Returns the page associated with the object with least pindex * greater than or equal to the parameter pindex, or NULL. * * The object must be locked. */ vm_page_t vm_page_find_least(vm_object_t object, vm_pindex_t pindex) { vm_page_t m; VM_OBJECT_ASSERT_LOCKED(object); if ((m = TAILQ_FIRST(&object->memq)) != NULL && m->pindex < pindex) m = vm_radix_lookup_ge(&object->rtree, pindex); return (m); } /* * Returns the given page's successor (by pindex) within the object if it is * resident; if none is found, NULL is returned. * * The object must be locked. */ vm_page_t vm_page_next(vm_page_t m) { vm_page_t next; VM_OBJECT_ASSERT_LOCKED(m->object); if ((next = TAILQ_NEXT(m, listq)) != NULL) { MPASS(next->object == m->object); if (next->pindex != m->pindex + 1) next = NULL; } return (next); } /* * Returns the given page's predecessor (by pindex) within the object if it is * resident; if none is found, NULL is returned. * * The object must be locked. */ vm_page_t vm_page_prev(vm_page_t m) { vm_page_t prev; VM_OBJECT_ASSERT_LOCKED(m->object); if ((prev = TAILQ_PREV(m, pglist, listq)) != NULL) { MPASS(prev->object == m->object); if (prev->pindex != m->pindex - 1) prev = NULL; } return (prev); } /* * Uses the page mnew as a replacement for an existing page at index * pindex which must be already present in the object. * * The existing page must not be on a paging queue. */ vm_page_t vm_page_replace(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex) { vm_page_t mold; VM_OBJECT_ASSERT_WLOCKED(object); KASSERT(mnew->object == NULL, ("vm_page_replace: page already in object")); /* * This function mostly follows vm_page_insert() and * vm_page_remove() without the radix, object count and vnode * dance. Double check such functions for more comments. */ mnew->object = object; mnew->pindex = pindex; mold = vm_radix_replace(&object->rtree, mnew); KASSERT(mold->queue == PQ_NONE, ("vm_page_replace: mold is on a paging queue")); /* Keep the resident page list in sorted order. */ TAILQ_INSERT_AFTER(&object->memq, mold, mnew, listq); TAILQ_REMOVE(&object->memq, mold, listq); mold->object = NULL; vm_page_xunbusy_maybelocked(mold); /* * The object's resident_page_count does not change because we have * swapped one page for another, but OBJ_MIGHTBEDIRTY. */ if (pmap_page_is_write_mapped(mnew)) vm_object_set_writeable_dirty(object); return (mold); } /* * vm_page_rename: * * Move the given memory entry from its * current object to the specified target object/offset. * * Note: swap associated with the page must be invalidated by the move. We * have to do this for several reasons: (1) we aren't freeing the * page, (2) we are dirtying the page, (3) the VM system is probably * moving the page from object A to B, and will then later move * the backing store from A to B and we can't have a conflict. * * Note: we *always* dirty the page. It is necessary both for the * fact that we moved it, and because we may be invalidating * swap. * * The objects must be locked. */ int vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex) { vm_page_t mpred; vm_pindex_t opidx; VM_OBJECT_ASSERT_WLOCKED(new_object); mpred = vm_radix_lookup_le(&new_object->rtree, new_pindex); KASSERT(mpred == NULL || mpred->pindex != new_pindex, ("vm_page_rename: pindex already renamed")); /* * Create a custom version of vm_page_insert() which does not depend * by m_prev and can cheat on the implementation aspects of the * function. */ opidx = m->pindex; m->pindex = new_pindex; if (vm_radix_insert(&new_object->rtree, m)) { m->pindex = opidx; return (1); } /* * The operation cannot fail anymore. The removal must happen before * the listq iterator is tainted. */ m->pindex = opidx; vm_page_lock(m); vm_page_remove(m); /* Return back to the new pindex to complete vm_page_insert(). */ m->pindex = new_pindex; m->object = new_object; vm_page_unlock(m); vm_page_insert_radixdone(m, new_object, mpred); vm_page_dirty(m); return (0); } /* * vm_page_alloc: * * Allocate and return a page that is associated with the specified * object and offset pair. By default, this page is exclusive busied. * * The caller must always specify an allocation class. * * allocation classes: * VM_ALLOC_NORMAL normal process request * VM_ALLOC_SYSTEM system *really* needs a page * VM_ALLOC_INTERRUPT interrupt time request * * optional allocation flags: * VM_ALLOC_COUNT(number) the number of additional pages that the caller * intends to allocate * VM_ALLOC_NOBUSY do not exclusive busy the page * VM_ALLOC_NODUMP do not include the page in a kernel core dump * VM_ALLOC_NOOBJ page is not associated with an object and * should not be exclusive busy * VM_ALLOC_SBUSY shared busy the allocated page * VM_ALLOC_WIRED wire the allocated page * VM_ALLOC_ZERO prefer a zeroed page * * This routine may not sleep. */ vm_page_t vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int req) { vm_page_t m, mpred; int flags, req_class; mpred = NULL; /* XXX: pacify gcc */ KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) && (object != NULL || (req & VM_ALLOC_SBUSY) == 0) && ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) != (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)), ("vm_page_alloc: inconsistent object(%p)/req(%x)", object, req)); if (object != NULL) VM_OBJECT_ASSERT_WLOCKED(object); if (__predict_false((req & VM_ALLOC_IFCACHED) != 0)) return (NULL); req_class = req & VM_ALLOC_CLASS_MASK; /* * The page daemon is allowed to dig deeper into the free page list. */ if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT) req_class = VM_ALLOC_SYSTEM; if (object != NULL) { mpred = vm_radix_lookup_le(&object->rtree, pindex); KASSERT(mpred == NULL || mpred->pindex != pindex, ("vm_page_alloc: pindex already allocated")); } /* * Allocate a page if the number of free pages exceeds the minimum * for the request class. */ mtx_lock(&vm_page_queue_free_mtx); if (vm_cnt.v_free_count > vm_cnt.v_free_reserved || (req_class == VM_ALLOC_SYSTEM && vm_cnt.v_free_count > vm_cnt.v_interrupt_free_min) || (req_class == VM_ALLOC_INTERRUPT && vm_cnt.v_free_count > 0)) { /* * Can we allocate the page from a reservation? */ #if VM_NRESERVLEVEL > 0 if (object == NULL || (object->flags & (OBJ_COLORED | OBJ_FICTITIOUS)) != OBJ_COLORED || (m = vm_reserv_alloc_page(object, pindex, mpred)) == NULL) #endif { /* * If not, allocate it from the free page queues. */ m = vm_phys_alloc_pages(object != NULL ? VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT, 0); #if VM_NRESERVLEVEL > 0 if (m == NULL && vm_reserv_reclaim_inactive()) { m = vm_phys_alloc_pages(object != NULL ? VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT, 0); } #endif } } else { /* * Not allocatable, give up. */ mtx_unlock(&vm_page_queue_free_mtx); atomic_add_int(&vm_pageout_deficit, max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1)); pagedaemon_wakeup(); return (NULL); } /* * At this point we had better have found a good page. */ KASSERT(m != NULL, ("vm_page_alloc: missing page")); vm_phys_freecnt_adj(m, -1); if ((m->flags & PG_ZERO) != 0) vm_page_zero_count--; mtx_unlock(&vm_page_queue_free_mtx); vm_page_alloc_check(m); /* * Initialize the page. Only the PG_ZERO flag is inherited. */ flags = 0; if ((req & VM_ALLOC_ZERO) != 0) flags = PG_ZERO; flags &= m->flags; if ((req & VM_ALLOC_NODUMP) != 0) flags |= PG_NODUMP; m->flags = flags; m->aflags = 0; m->oflags = object == NULL || (object->flags & OBJ_UNMANAGED) != 0 ? VPO_UNMANAGED : 0; m->busy_lock = VPB_UNBUSIED; if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ | VM_ALLOC_SBUSY)) == 0) m->busy_lock = VPB_SINGLE_EXCLUSIVER; if ((req & VM_ALLOC_SBUSY) != 0) m->busy_lock = VPB_SHARERS_WORD(1); if (req & VM_ALLOC_WIRED) { /* * The page lock is not required for wiring a page until that * page is inserted into the object. */ atomic_add_int(&vm_cnt.v_wire_count, 1); m->wire_count = 1; } m->act_count = 0; if (object != NULL) { if (vm_page_insert_after(m, object, pindex, mpred)) { pagedaemon_wakeup(); if (req & VM_ALLOC_WIRED) { atomic_subtract_int(&vm_cnt.v_wire_count, 1); m->wire_count = 0; } KASSERT(m->object == NULL, ("page %p has object", m)); m->oflags = VPO_UNMANAGED; m->busy_lock = VPB_UNBUSIED; /* Don't change PG_ZERO. */ vm_page_free_toq(m); return (NULL); } /* Ignore device objects; the pager sets "memattr" for them. */ if (object->memattr != VM_MEMATTR_DEFAULT && (object->flags & OBJ_FICTITIOUS) == 0) pmap_page_set_memattr(m, object->memattr); } else m->pindex = pindex; /* * Don't wakeup too often - wakeup the pageout daemon when * we would be nearly out of memory. */ if (vm_paging_needed()) pagedaemon_wakeup(); return (m); } /* * vm_page_alloc_contig: * * Allocate a contiguous set of physical pages of the given size "npages" * from the free lists. All of the physical pages must be at or above * the given physical address "low" and below the given physical address * "high". The given value "alignment" determines the alignment of the * first physical page in the set. If the given value "boundary" is * non-zero, then the set of physical pages cannot cross any physical * address boundary that is a multiple of that value. Both "alignment" * and "boundary" must be a power of two. * * If the specified memory attribute, "memattr", is VM_MEMATTR_DEFAULT, * then the memory attribute setting for the physical pages is configured * to the object's memory attribute setting. Otherwise, the memory * attribute setting for the physical pages is configured to "memattr", * overriding the object's memory attribute setting. However, if the * object's memory attribute setting is not VM_MEMATTR_DEFAULT, then the * memory attribute setting for the physical pages cannot be configured * to VM_MEMATTR_DEFAULT. * * The specified object may not contain fictitious pages. * * The caller must always specify an allocation class. * * allocation classes: * VM_ALLOC_NORMAL normal process request * VM_ALLOC_SYSTEM system *really* needs a page * VM_ALLOC_INTERRUPT interrupt time request * * optional allocation flags: * VM_ALLOC_NOBUSY do not exclusive busy the page * VM_ALLOC_NODUMP do not include the page in a kernel core dump * VM_ALLOC_NOOBJ page is not associated with an object and * should not be exclusive busy * VM_ALLOC_SBUSY shared busy the allocated page * VM_ALLOC_WIRED wire the allocated page * VM_ALLOC_ZERO prefer a zeroed page * * This routine may not sleep. */ vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req, u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary, vm_memattr_t memattr) { vm_page_t m, m_ret, mpred; u_int busy_lock, flags, oflags; int req_class; mpred = NULL; /* XXX: pacify gcc */ KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) && (object != NULL || (req & VM_ALLOC_SBUSY) == 0) && ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) != (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)), ("vm_page_alloc_contig: inconsistent object(%p)/req(%x)", object, req)); if (object != NULL) { VM_OBJECT_ASSERT_WLOCKED(object); KASSERT((object->flags & OBJ_FICTITIOUS) == 0, ("vm_page_alloc_contig: object %p has fictitious pages", object)); } KASSERT(npages > 0, ("vm_page_alloc_contig: npages is zero")); req_class = req & VM_ALLOC_CLASS_MASK; /* * The page daemon is allowed to dig deeper into the free page list. */ if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT) req_class = VM_ALLOC_SYSTEM; if (object != NULL) { mpred = vm_radix_lookup_le(&object->rtree, pindex); KASSERT(mpred == NULL || mpred->pindex != pindex, ("vm_page_alloc_contig: pindex already allocated")); } /* * Can we allocate the pages without the number of free pages falling * below the lower bound for the allocation class? */ mtx_lock(&vm_page_queue_free_mtx); if (vm_cnt.v_free_count >= npages + vm_cnt.v_free_reserved || (req_class == VM_ALLOC_SYSTEM && vm_cnt.v_free_count >= npages + vm_cnt.v_interrupt_free_min) || (req_class == VM_ALLOC_INTERRUPT && vm_cnt.v_free_count >= npages)) { /* * Can we allocate the pages from a reservation? */ #if VM_NRESERVLEVEL > 0 retry: if (object == NULL || (object->flags & OBJ_COLORED) == 0 || (m_ret = vm_reserv_alloc_contig(object, pindex, npages, low, high, alignment, boundary, mpred)) == NULL) #endif /* * If not, allocate them from the free page queues. */ m_ret = vm_phys_alloc_contig(npages, low, high, alignment, boundary); } else { mtx_unlock(&vm_page_queue_free_mtx); atomic_add_int(&vm_pageout_deficit, npages); pagedaemon_wakeup(); return (NULL); } if (m_ret != NULL) { vm_phys_freecnt_adj(m_ret, -npages); for (m = m_ret; m < &m_ret[npages]; m++) if ((m->flags & PG_ZERO) != 0) vm_page_zero_count--; } else { #if VM_NRESERVLEVEL > 0 if (vm_reserv_reclaim_contig(npages, low, high, alignment, boundary)) goto retry; #endif } mtx_unlock(&vm_page_queue_free_mtx); if (m_ret == NULL) return (NULL); for (m = m_ret; m < &m_ret[npages]; m++) vm_page_alloc_check(m); /* * Initialize the pages. Only the PG_ZERO flag is inherited. */ flags = 0; if ((req & VM_ALLOC_ZERO) != 0) flags = PG_ZERO; if ((req & VM_ALLOC_NODUMP) != 0) flags |= PG_NODUMP; oflags = object == NULL || (object->flags & OBJ_UNMANAGED) != 0 ? VPO_UNMANAGED : 0; busy_lock = VPB_UNBUSIED; if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ | VM_ALLOC_SBUSY)) == 0) busy_lock = VPB_SINGLE_EXCLUSIVER; if ((req & VM_ALLOC_SBUSY) != 0) busy_lock = VPB_SHARERS_WORD(1); if ((req & VM_ALLOC_WIRED) != 0) atomic_add_int(&vm_cnt.v_wire_count, npages); if (object != NULL) { if (object->memattr != VM_MEMATTR_DEFAULT && memattr == VM_MEMATTR_DEFAULT) memattr = object->memattr; } for (m = m_ret; m < &m_ret[npages]; m++) { m->aflags = 0; m->flags = (m->flags | PG_NODUMP) & flags; m->busy_lock = busy_lock; if ((req & VM_ALLOC_WIRED) != 0) m->wire_count = 1; m->act_count = 0; m->oflags = oflags; if (object != NULL) { if (vm_page_insert_after(m, object, pindex, mpred)) { pagedaemon_wakeup(); if ((req & VM_ALLOC_WIRED) != 0) atomic_subtract_int( &vm_cnt.v_wire_count, npages); KASSERT(m->object == NULL, ("page %p has object", m)); mpred = m; for (m = m_ret; m < &m_ret[npages]; m++) { if (m <= mpred && (req & VM_ALLOC_WIRED) != 0) m->wire_count = 0; m->oflags = VPO_UNMANAGED; m->busy_lock = VPB_UNBUSIED; /* Don't change PG_ZERO. */ vm_page_free_toq(m); } return (NULL); } mpred = m; } else m->pindex = pindex; if (memattr != VM_MEMATTR_DEFAULT) pmap_page_set_memattr(m, memattr); pindex++; } if (vm_paging_needed()) pagedaemon_wakeup(); return (m_ret); } /* * Check a page that has been freshly dequeued from a freelist. */ static void vm_page_alloc_check(vm_page_t m) { KASSERT(m->object == NULL, ("page %p has object", m)); KASSERT(m->queue == PQ_NONE, ("page %p has unexpected queue %d", m, m->queue)); KASSERT(m->wire_count == 0, ("page %p is wired", m)); KASSERT(m->hold_count == 0, ("page %p is held", m)); KASSERT(!vm_page_busied(m), ("page %p is busy", m)); KASSERT(m->dirty == 0, ("page %p is dirty", m)); KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT, ("page %p has unexpected memattr %d", m, pmap_page_get_memattr(m))); KASSERT(m->valid == 0, ("free page %p is valid", m)); } /* * vm_page_alloc_freelist: * * Allocate a physical page from the specified free page list. * * The caller must always specify an allocation class. * * allocation classes: * VM_ALLOC_NORMAL normal process request * VM_ALLOC_SYSTEM system *really* needs a page * VM_ALLOC_INTERRUPT interrupt time request * * optional allocation flags: * VM_ALLOC_COUNT(number) the number of additional pages that the caller * intends to allocate * VM_ALLOC_WIRED wire the allocated page * VM_ALLOC_ZERO prefer a zeroed page * * This routine may not sleep. */ vm_page_t vm_page_alloc_freelist(int flind, int req) { vm_page_t m; u_int flags; int req_class; req_class = req & VM_ALLOC_CLASS_MASK; /* * The page daemon is allowed to dig deeper into the free page list. */ if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT) req_class = VM_ALLOC_SYSTEM; /* * Do not allocate reserved pages unless the req has asked for it. */ mtx_lock(&vm_page_queue_free_mtx); if (vm_cnt.v_free_count > vm_cnt.v_free_reserved || (req_class == VM_ALLOC_SYSTEM && vm_cnt.v_free_count > vm_cnt.v_interrupt_free_min) || (req_class == VM_ALLOC_INTERRUPT && vm_cnt.v_free_count > 0)) m = vm_phys_alloc_freelist_pages(flind, VM_FREEPOOL_DIRECT, 0); else { mtx_unlock(&vm_page_queue_free_mtx); atomic_add_int(&vm_pageout_deficit, max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1)); pagedaemon_wakeup(); return (NULL); } if (m == NULL) { mtx_unlock(&vm_page_queue_free_mtx); return (NULL); } vm_phys_freecnt_adj(m, -1); if ((m->flags & PG_ZERO) != 0) vm_page_zero_count--; mtx_unlock(&vm_page_queue_free_mtx); vm_page_alloc_check(m); /* * Initialize the page. Only the PG_ZERO flag is inherited. */ m->aflags = 0; flags = 0; if ((req & VM_ALLOC_ZERO) != 0) flags = PG_ZERO; m->flags &= flags; if ((req & VM_ALLOC_WIRED) != 0) { /* * The page lock is not required for wiring a page that does * not belong to an object. */ atomic_add_int(&vm_cnt.v_wire_count, 1); m->wire_count = 1; } /* Unmanaged pages don't use "act_count". */ m->oflags = VPO_UNMANAGED; if (vm_paging_needed()) pagedaemon_wakeup(); return (m); } #define VPSC_ANY 0 /* No restrictions. */ #define VPSC_NORESERV 1 /* Skip reservations; implies VPSC_NOSUPER. */ #define VPSC_NOSUPER 2 /* Skip superpages. */ /* * vm_page_scan_contig: * * Scan vm_page_array[] between the specified entries "m_start" and * "m_end" for a run of contiguous physical pages that satisfy the * specified conditions, and return the lowest page in the run. The * specified "alignment" determines the alignment of the lowest physical * page in the run. If the specified "boundary" is non-zero, then the * run of physical pages cannot span a physical address that is a * multiple of "boundary". * * "m_end" is never dereferenced, so it need not point to a vm_page * structure within vm_page_array[]. * * "npages" must be greater than zero. "m_start" and "m_end" must not * span a hole (or discontiguity) in the physical address space. Both * "alignment" and "boundary" must be a power of two. */ vm_page_t vm_page_scan_contig(u_long npages, vm_page_t m_start, vm_page_t m_end, u_long alignment, vm_paddr_t boundary, int options) { struct mtx *m_mtx; vm_object_t object; vm_paddr_t pa; vm_page_t m, m_run; #if VM_NRESERVLEVEL > 0 int level; #endif int m_inc, order, run_ext, run_len; KASSERT(npages > 0, ("npages is 0")); KASSERT(powerof2(alignment), ("alignment is not a power of 2")); KASSERT(powerof2(boundary), ("boundary is not a power of 2")); m_run = NULL; run_len = 0; m_mtx = NULL; for (m = m_start; m < m_end && run_len < npages; m += m_inc) { KASSERT((m->flags & PG_MARKER) == 0, ("page %p is PG_MARKER", m)); KASSERT((m->flags & PG_FICTITIOUS) == 0 || m->wire_count == 1, ("fictitious page %p has invalid wire count", m)); /* * If the current page would be the start of a run, check its * physical address against the end, alignment, and boundary * conditions. If it doesn't satisfy these conditions, either * terminate the scan or advance to the next page that * satisfies the failed condition. */ if (run_len == 0) { KASSERT(m_run == NULL, ("m_run != NULL")); if (m + npages > m_end) break; pa = VM_PAGE_TO_PHYS(m); if ((pa & (alignment - 1)) != 0) { m_inc = atop(roundup2(pa, alignment) - pa); continue; } if (rounddown2(pa ^ (pa + ptoa(npages) - 1), boundary) != 0) { m_inc = atop(roundup2(pa, boundary) - pa); continue; } } else KASSERT(m_run != NULL, ("m_run == NULL")); vm_page_change_lock(m, &m_mtx); m_inc = 1; retry: if (m->wire_count != 0 || m->hold_count != 0) run_ext = 0; #if VM_NRESERVLEVEL > 0 else if ((level = vm_reserv_level(m)) >= 0 && (options & VPSC_NORESERV) != 0) { run_ext = 0; /* Advance to the end of the reservation. */ pa = VM_PAGE_TO_PHYS(m); m_inc = atop(roundup2(pa + 1, vm_reserv_size(level)) - pa); } #endif else if ((object = m->object) != NULL) { /* * The page is considered eligible for relocation if * and only if it could be laundered or reclaimed by * the page daemon. */ if (!VM_OBJECT_TRYRLOCK(object)) { mtx_unlock(m_mtx); VM_OBJECT_RLOCK(object); mtx_lock(m_mtx); if (m->object != object) { /* * The page may have been freed. */ VM_OBJECT_RUNLOCK(object); goto retry; } else if (m->wire_count != 0 || m->hold_count != 0) { run_ext = 0; goto unlock; } } KASSERT((m->flags & PG_UNHOLDFREE) == 0, ("page %p is PG_UNHOLDFREE", m)); /* Don't care: PG_NODUMP, PG_ZERO. */ if (object->type != OBJT_DEFAULT && object->type != OBJT_SWAP && object->type != OBJT_VNODE) { run_ext = 0; #if VM_NRESERVLEVEL > 0 } else if ((options & VPSC_NOSUPER) != 0 && (level = vm_reserv_level_iffullpop(m)) >= 0) { run_ext = 0; /* Advance to the end of the superpage. */ pa = VM_PAGE_TO_PHYS(m); m_inc = atop(roundup2(pa + 1, vm_reserv_size(level)) - pa); #endif } else if (object->memattr == VM_MEMATTR_DEFAULT && m->queue != PQ_NONE && !vm_page_busied(m)) { /* * The page is allocated but eligible for * relocation. Extend the current run by one * page. */ KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT, ("page %p has an unexpected memattr", m)); KASSERT((m->oflags & (VPO_SWAPINPROG | VPO_SWAPSLEEP | VPO_UNMANAGED)) == 0, ("page %p has unexpected oflags", m)); /* Don't care: VPO_NOSYNC. */ run_ext = 1; } else run_ext = 0; unlock: VM_OBJECT_RUNLOCK(object); #if VM_NRESERVLEVEL > 0 } else if (level >= 0) { /* * The page is reserved but not yet allocated. In * other words, it is still free. Extend the current * run by one page. */ run_ext = 1; #endif } else if ((order = m->order) < VM_NFREEORDER) { /* * The page is enqueued in the physical memory * allocator's free page queues. Moreover, it is the * first page in a power-of-two-sized run of * contiguous free pages. Add these pages to the end * of the current run, and jump ahead. */ run_ext = 1 << order; m_inc = 1 << order; } else { /* * Skip the page for one of the following reasons: (1) * It is enqueued in the physical memory allocator's * free page queues. However, it is not the first * page in a run of contiguous free pages. (This case * rarely occurs because the scan is performed in * ascending order.) (2) It is not reserved, and it is * transitioning from free to allocated. (Conversely, * the transition from allocated to free for managed * pages is blocked by the page lock.) (3) It is * allocated but not contained by an object and not * wired, e.g., allocated by Xen's balloon driver. */ run_ext = 0; } /* * Extend or reset the current run of pages. */ if (run_ext > 0) { if (run_len == 0) m_run = m; run_len += run_ext; } else { if (run_len > 0) { m_run = NULL; run_len = 0; } } } if (m_mtx != NULL) mtx_unlock(m_mtx); if (run_len >= npages) return (m_run); return (NULL); } /* * vm_page_reclaim_run: * * Try to relocate each of the allocated virtual pages within the * specified run of physical pages to a new physical address. Free the * physical pages underlying the relocated virtual pages. A virtual page * is relocatable if and only if it could be laundered or reclaimed by * the page daemon. Whenever possible, a virtual page is relocated to a * physical address above "high". * * Returns 0 if every physical page within the run was already free or * just freed by a successful relocation. Otherwise, returns a non-zero * value indicating why the last attempt to relocate a virtual page was * unsuccessful. * * "req_class" must be an allocation class. */ static int vm_page_reclaim_run(int req_class, u_long npages, vm_page_t m_run, vm_paddr_t high) { struct mtx *m_mtx; struct spglist free; vm_object_t object; vm_paddr_t pa; vm_page_t m, m_end, m_new; int error, order, req; KASSERT((req_class & VM_ALLOC_CLASS_MASK) == req_class, ("req_class is not an allocation class")); SLIST_INIT(&free); error = 0; m = m_run; m_end = m_run + npages; m_mtx = NULL; for (; error == 0 && m < m_end; m++) { KASSERT((m->flags & (PG_FICTITIOUS | PG_MARKER)) == 0, ("page %p is PG_FICTITIOUS or PG_MARKER", m)); /* * Avoid releasing and reacquiring the same page lock. */ vm_page_change_lock(m, &m_mtx); retry: if (m->wire_count != 0 || m->hold_count != 0) error = EBUSY; else if ((object = m->object) != NULL) { /* * The page is relocated if and only if it could be * laundered or reclaimed by the page daemon. */ if (!VM_OBJECT_TRYWLOCK(object)) { mtx_unlock(m_mtx); VM_OBJECT_WLOCK(object); mtx_lock(m_mtx); if (m->object != object) { /* * The page may have been freed. */ VM_OBJECT_WUNLOCK(object); goto retry; } else if (m->wire_count != 0 || m->hold_count != 0) { error = EBUSY; goto unlock; } } KASSERT((m->flags & PG_UNHOLDFREE) == 0, ("page %p is PG_UNHOLDFREE", m)); /* Don't care: PG_NODUMP, PG_ZERO. */ if (object->type != OBJT_DEFAULT && object->type != OBJT_SWAP && object->type != OBJT_VNODE) error = EINVAL; else if (object->memattr != VM_MEMATTR_DEFAULT) error = EINVAL; else if (m->queue != PQ_NONE && !vm_page_busied(m)) { KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT, ("page %p has an unexpected memattr", m)); KASSERT((m->oflags & (VPO_SWAPINPROG | VPO_SWAPSLEEP | VPO_UNMANAGED)) == 0, ("page %p has unexpected oflags", m)); /* Don't care: VPO_NOSYNC. */ if (m->valid != 0) { /* * First, try to allocate a new page * that is above "high". Failing * that, try to allocate a new page * that is below "m_run". Allocate * the new page between the end of * "m_run" and "high" only as a last * resort. */ req = req_class | VM_ALLOC_NOOBJ; if ((m->flags & PG_NODUMP) != 0) req |= VM_ALLOC_NODUMP; if (trunc_page(high) != ~(vm_paddr_t)PAGE_MASK) { m_new = vm_page_alloc_contig( NULL, 0, req, 1, round_page(high), ~(vm_paddr_t)0, PAGE_SIZE, 0, VM_MEMATTR_DEFAULT); } else m_new = NULL; if (m_new == NULL) { pa = VM_PAGE_TO_PHYS(m_run); m_new = vm_page_alloc_contig( NULL, 0, req, 1, 0, pa - 1, PAGE_SIZE, 0, VM_MEMATTR_DEFAULT); } if (m_new == NULL) { pa += ptoa(npages); m_new = vm_page_alloc_contig( NULL, 0, req, 1, pa, high, PAGE_SIZE, 0, VM_MEMATTR_DEFAULT); } if (m_new == NULL) { error = ENOMEM; goto unlock; } KASSERT(m_new->wire_count == 0, ("page %p is wired", m)); /* * Replace "m" with the new page. For * vm_page_replace(), "m" must be busy * and dequeued. Finally, change "m" * as if vm_page_free() was called. */ if (object->ref_count != 0) pmap_remove_all(m); m_new->aflags = m->aflags; KASSERT(m_new->oflags == VPO_UNMANAGED, ("page %p is managed", m)); m_new->oflags = m->oflags & VPO_NOSYNC; pmap_copy_page(m, m_new); m_new->valid = m->valid; m_new->dirty = m->dirty; m->flags &= ~PG_ZERO; vm_page_xbusy(m); vm_page_remque(m); vm_page_replace_checked(m_new, object, m->pindex, m); m->valid = 0; vm_page_undirty(m); /* * The new page must be deactivated * before the object is unlocked. */ vm_page_change_lock(m_new, &m_mtx); vm_page_deactivate(m_new); } else { m->flags &= ~PG_ZERO; vm_page_remque(m); vm_page_remove(m); KASSERT(m->dirty == 0, ("page %p is dirty", m)); } SLIST_INSERT_HEAD(&free, m, plinks.s.ss); } else error = EBUSY; unlock: VM_OBJECT_WUNLOCK(object); } else { mtx_lock(&vm_page_queue_free_mtx); order = m->order; if (order < VM_NFREEORDER) { /* * The page is enqueued in the physical memory * allocator's free page queues. Moreover, it * is the first page in a power-of-two-sized * run of contiguous free pages. Jump ahead * to the last page within that run, and * continue from there. */ m += (1 << order) - 1; } #if VM_NRESERVLEVEL > 0 else if (vm_reserv_is_page_free(m)) order = 0; #endif mtx_unlock(&vm_page_queue_free_mtx); if (order == VM_NFREEORDER) error = EINVAL; } } if (m_mtx != NULL) mtx_unlock(m_mtx); if ((m = SLIST_FIRST(&free)) != NULL) { mtx_lock(&vm_page_queue_free_mtx); do { SLIST_REMOVE_HEAD(&free, plinks.s.ss); vm_page_free_phys(m); } while ((m = SLIST_FIRST(&free)) != NULL); vm_page_zero_idle_wakeup(); vm_page_free_wakeup(); mtx_unlock(&vm_page_queue_free_mtx); } return (error); } #define NRUNS 16 CTASSERT(powerof2(NRUNS)); #define RUN_INDEX(count) ((count) & (NRUNS - 1)) #define MIN_RECLAIM 8 /* * vm_page_reclaim_contig: * * Reclaim allocated, contiguous physical memory satisfying the specified * conditions by relocating the virtual pages using that physical memory. * Returns true if reclamation is successful and false otherwise. Since * relocation requires the allocation of physical pages, reclamation may * fail due to a shortage of free pages. When reclamation fails, callers * are expected to perform VM_WAIT before retrying a failed allocation * operation, e.g., vm_page_alloc_contig(). * * The caller must always specify an allocation class through "req". * * allocation classes: * VM_ALLOC_NORMAL normal process request * VM_ALLOC_SYSTEM system *really* needs a page * VM_ALLOC_INTERRUPT interrupt time request * * The optional allocation flags are ignored. * * "npages" must be greater than zero. Both "alignment" and "boundary" * must be a power of two. */ bool vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary) { vm_paddr_t curr_low; vm_page_t m_run, m_runs[NRUNS]; u_long count, reclaimed; int error, i, options, req_class; KASSERT(npages > 0, ("npages is 0")); KASSERT(powerof2(alignment), ("alignment is not a power of 2")); KASSERT(powerof2(boundary), ("boundary is not a power of 2")); req_class = req & VM_ALLOC_CLASS_MASK; /* * The page daemon is allowed to dig deeper into the free page list. */ if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT) req_class = VM_ALLOC_SYSTEM; /* * Return if the number of free pages cannot satisfy the requested * allocation. */ count = vm_cnt.v_free_count; if (count < npages + vm_cnt.v_free_reserved || (count < npages + vm_cnt.v_interrupt_free_min && req_class == VM_ALLOC_SYSTEM) || (count < npages && req_class == VM_ALLOC_INTERRUPT)) return (false); /* * Scan up to three times, relaxing the restrictions ("options") on * the reclamation of reservations and superpages each time. */ for (options = VPSC_NORESERV;;) { /* * Find the highest runs that satisfy the given constraints * and restrictions, and record them in "m_runs". */ curr_low = low; count = 0; for (;;) { m_run = vm_phys_scan_contig(npages, curr_low, high, alignment, boundary, options); if (m_run == NULL) break; curr_low = VM_PAGE_TO_PHYS(m_run) + ptoa(npages); m_runs[RUN_INDEX(count)] = m_run; count++; } /* * Reclaim the highest runs in LIFO (descending) order until * the number of reclaimed pages, "reclaimed", is at least * MIN_RECLAIM. Reset "reclaimed" each time because each * reclamation is idempotent, and runs will (likely) recur * from one scan to the next as restrictions are relaxed. */ reclaimed = 0; for (i = 0; count > 0 && i < NRUNS; i++) { count--; m_run = m_runs[RUN_INDEX(count)]; error = vm_page_reclaim_run(req_class, npages, m_run, high); if (error == 0) { reclaimed += npages; if (reclaimed >= MIN_RECLAIM) return (true); } } /* * Either relax the restrictions on the next scan or return if * the last scan had no restrictions. */ if (options == VPSC_NORESERV) options = VPSC_NOSUPER; else if (options == VPSC_NOSUPER) options = VPSC_ANY; else if (options == VPSC_ANY) return (reclaimed != 0); } } /* * vm_wait: (also see VM_WAIT macro) * * Sleep until free pages are available for allocation. * - Called in various places before memory allocations. */ void vm_wait(void) { mtx_lock(&vm_page_queue_free_mtx); if (curproc == pageproc) { vm_pageout_pages_needed = 1; msleep(&vm_pageout_pages_needed, &vm_page_queue_free_mtx, PDROP | PSWP, "VMWait", 0); } else { if (__predict_false(pageproc == NULL)) panic("vm_wait in early boot"); if (!vm_pageout_wanted) { vm_pageout_wanted = true; wakeup(&vm_pageout_wanted); } vm_pages_needed = true; msleep(&vm_cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PVM, "vmwait", 0); } } /* * vm_waitpfault: (also see VM_WAITPFAULT macro) * * Sleep until free pages are available for allocation. * - Called only in vm_fault so that processes page faulting * can be easily tracked. * - Sleeps at a lower priority than vm_wait() so that vm_wait()ing * processes will be able to grab memory first. Do not change * this balance without careful testing first. */ void vm_waitpfault(void) { mtx_lock(&vm_page_queue_free_mtx); if (!vm_pageout_wanted) { vm_pageout_wanted = true; wakeup(&vm_pageout_wanted); } vm_pages_needed = true; msleep(&vm_cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PUSER, "pfault", 0); } struct vm_pagequeue * vm_page_pagequeue(vm_page_t m) { if (vm_page_in_laundry(m)) return (&vm_dom[0].vmd_pagequeues[m->queue]); else return (&vm_phys_domain(m)->vmd_pagequeues[m->queue]); } /* * vm_page_dequeue: * * Remove the given page from its current page queue. * * The page must be locked. */ void vm_page_dequeue(vm_page_t m) { struct vm_pagequeue *pq; vm_page_assert_locked(m); KASSERT(m->queue < PQ_COUNT, ("vm_page_dequeue: page %p is not queued", m)); pq = vm_page_pagequeue(m); vm_pagequeue_lock(pq); m->queue = PQ_NONE; TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); vm_pagequeue_cnt_dec(pq); vm_pagequeue_unlock(pq); } /* * vm_page_dequeue_locked: * * Remove the given page from its current page queue. * * The page and page queue must be locked. */ void vm_page_dequeue_locked(vm_page_t m) { struct vm_pagequeue *pq; vm_page_lock_assert(m, MA_OWNED); pq = vm_page_pagequeue(m); vm_pagequeue_assert_locked(pq); m->queue = PQ_NONE; TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); vm_pagequeue_cnt_dec(pq); } /* * vm_page_enqueue: * * Add the given page to the specified page queue. * * The page must be locked. */ static void vm_page_enqueue(uint8_t queue, vm_page_t m) { struct vm_pagequeue *pq; vm_page_lock_assert(m, MA_OWNED); KASSERT(queue < PQ_COUNT, ("vm_page_enqueue: invalid queue %u request for page %p", queue, m)); if (queue == PQ_LAUNDRY) pq = &vm_dom[0].vmd_pagequeues[queue]; else pq = &vm_phys_domain(m)->vmd_pagequeues[queue]; vm_pagequeue_lock(pq); m->queue = queue; TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); vm_pagequeue_cnt_inc(pq); vm_pagequeue_unlock(pq); } /* * vm_page_requeue: * * Move the given page to the tail of its current page queue. * * The page must be locked. */ void vm_page_requeue(vm_page_t m) { struct vm_pagequeue *pq; vm_page_lock_assert(m, MA_OWNED); KASSERT(m->queue != PQ_NONE, ("vm_page_requeue: page %p is not queued", m)); pq = vm_page_pagequeue(m); vm_pagequeue_lock(pq); TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); vm_pagequeue_unlock(pq); } /* * vm_page_requeue_locked: * * Move the given page to the tail of its current page queue. * * The page queue must be locked. */ void vm_page_requeue_locked(vm_page_t m) { struct vm_pagequeue *pq; KASSERT(m->queue != PQ_NONE, ("vm_page_requeue_locked: page %p is not queued", m)); pq = vm_page_pagequeue(m); vm_pagequeue_assert_locked(pq); TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); } /* * vm_page_activate: * * Put the specified page on the active list (if appropriate). * Ensure that act_count is at least ACT_INIT but do not otherwise * mess with it. * * The page must be locked. */ void vm_page_activate(vm_page_t m) { int queue; vm_page_lock_assert(m, MA_OWNED); if ((queue = m->queue) != PQ_ACTIVE) { if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) { if (m->act_count < ACT_INIT) m->act_count = ACT_INIT; if (queue != PQ_NONE) vm_page_dequeue(m); vm_page_enqueue(PQ_ACTIVE, m); } else KASSERT(queue == PQ_NONE, ("vm_page_activate: wired page %p is queued", m)); } else { if (m->act_count < ACT_INIT) m->act_count = ACT_INIT; } } /* * vm_page_free_wakeup: * * Helper routine for vm_page_free_toq(). This routine is called * when a page is added to the free queues. * * The page queues must be locked. */ static void vm_page_free_wakeup(void) { mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); /* * if pageout daemon needs pages, then tell it that there are * some free. */ if (vm_pageout_pages_needed && vm_cnt.v_free_count >= vm_cnt.v_pageout_free_min) { wakeup(&vm_pageout_pages_needed); vm_pageout_pages_needed = 0; } /* * wakeup processes that are waiting on memory if we hit a * high water mark. And wakeup scheduler process if we have * lots of memory. this process will swapin processes. */ if (vm_pages_needed && !vm_page_count_min()) { vm_pages_needed = false; wakeup(&vm_cnt.v_free_count); } } /* * vm_page_free_prep: * * Prepares the given page to be put on the free list, * disassociating it from any VM object. The caller may return * the page to the free list only if this function returns true. * * The object must be locked. The page must be locked if it is * managed. For a queued managed page, the pagequeue_locked * argument specifies whether the page queue is already locked. */ bool vm_page_free_prep(vm_page_t m, bool pagequeue_locked) { #if defined(DIAGNOSTIC) && defined(PHYS_TO_DMAP) if ((m->flags & PG_ZERO) != 0) { uint64_t *p; int i; p = (uint64_t *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)); for (i = 0; i < PAGE_SIZE / sizeof(uint64_t); i++, p++) KASSERT(*p == 0, ("vm_page_free_prep %p PG_ZERO %d %jx", m, i, (uintmax_t)*p)); } #endif if ((m->oflags & VPO_UNMANAGED) == 0) { vm_page_lock_assert(m, MA_OWNED); KASSERT(!pmap_page_is_mapped(m), ("vm_page_free_toq: freeing mapped page %p", m)); } else KASSERT(m->queue == PQ_NONE, ("vm_page_free_toq: unmanaged page %p is queued", m)); PCPU_INC(cnt.v_tfree); if (vm_page_sbusied(m)) panic("vm_page_free: freeing busy page %p", m); /* * Unqueue, then remove page. Note that we cannot destroy * the page here because we do not want to call the pager's * callback routine until after we've put the page on the * appropriate free queue. */ if (m->queue != PQ_NONE) { if (pagequeue_locked) vm_page_dequeue_locked(m); else vm_page_dequeue(m); } vm_page_remove(m); /* * If fictitious remove object association and * return, otherwise delay object association removal. */ if ((m->flags & PG_FICTITIOUS) != 0) return (false); m->valid = 0; vm_page_undirty(m); if (m->wire_count != 0) panic("vm_page_free: freeing wired page %p", m); if (m->hold_count != 0) { m->flags &= ~PG_ZERO; KASSERT((m->flags & PG_UNHOLDFREE) == 0, ("vm_page_free: freeing PG_UNHOLDFREE page %p", m)); m->flags |= PG_UNHOLDFREE; return (false); } /* * Restore the default memory attribute to the page. */ if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT) pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT); return (true); } /* * Insert the page into the physical memory allocator's free page * queues. This is the last step to free a page. */ static void vm_page_free_phys(vm_page_t m) { mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); vm_phys_freecnt_adj(m, 1); #if VM_NRESERVLEVEL > 0 if (!vm_reserv_free_page(m)) #endif vm_phys_free_pages(m, 0); if ((m->flags & PG_ZERO) != 0) ++vm_page_zero_count; else vm_page_zero_idle_wakeup(); } void vm_page_free_phys_pglist(struct pglist *tq) { vm_page_t m; if (TAILQ_EMPTY(tq)) return; mtx_lock(&vm_page_queue_free_mtx); TAILQ_FOREACH(m, tq, listq) vm_page_free_phys(m); vm_page_free_wakeup(); mtx_unlock(&vm_page_queue_free_mtx); } /* * vm_page_free_toq: * * Returns the given page to the free list, disassociating it * from any VM object. * * The object must be locked. The page must be locked if it is * managed. */ void vm_page_free_toq(vm_page_t m) { if (!vm_page_free_prep(m, false)) return; mtx_lock(&vm_page_queue_free_mtx); vm_page_free_phys(m); vm_page_free_wakeup(); mtx_unlock(&vm_page_queue_free_mtx); } /* * vm_page_wire: * * Mark this page as wired down by yet * another map, removing it from paging queues * as necessary. * * If the page is fictitious, then its wire count must remain one. * * The page must be locked. */ void vm_page_wire(vm_page_t m) { /* * Only bump the wire statistics if the page is not already wired, * and only unqueue the page if it is on some queue (if it is unmanaged * it is already off the queues). */ vm_page_lock_assert(m, MA_OWNED); if ((m->flags & PG_FICTITIOUS) != 0) { KASSERT(m->wire_count == 1, ("vm_page_wire: fictitious page %p's wire count isn't one", m)); return; } if (m->wire_count == 0) { KASSERT((m->oflags & VPO_UNMANAGED) == 0 || m->queue == PQ_NONE, ("vm_page_wire: unmanaged page %p is queued", m)); vm_page_remque(m); atomic_add_int(&vm_cnt.v_wire_count, 1); } m->wire_count++; KASSERT(m->wire_count != 0, ("vm_page_wire: wire_count overflow m=%p", m)); } /* * vm_page_unwire: * * Release one wiring of the specified page, potentially allowing it to be * paged out. Returns TRUE if the number of wirings transitions to zero and * FALSE otherwise. * * Only managed pages belonging to an object can be paged out. If the number * of wirings transitions to zero and the page is eligible for page out, then * the page is added to the specified paging queue (unless PQ_NONE is * specified). * * If a page is fictitious, then its wire count must always be one. * * A managed page must be locked. */ boolean_t vm_page_unwire(vm_page_t m, uint8_t queue) { KASSERT(queue < PQ_COUNT || queue == PQ_NONE, ("vm_page_unwire: invalid queue %u request for page %p", queue, m)); if ((m->oflags & VPO_UNMANAGED) == 0) vm_page_assert_locked(m); if ((m->flags & PG_FICTITIOUS) != 0) { KASSERT(m->wire_count == 1, ("vm_page_unwire: fictitious page %p's wire count isn't one", m)); return (FALSE); } if (m->wire_count > 0) { m->wire_count--; if (m->wire_count == 0) { atomic_subtract_int(&vm_cnt.v_wire_count, 1); if ((m->oflags & VPO_UNMANAGED) == 0 && m->object != NULL && queue != PQ_NONE) vm_page_enqueue(queue, m); return (TRUE); } else return (FALSE); } else panic("vm_page_unwire: page %p's wire count is zero", m); } /* * Move the specified page to the inactive queue. * * Normally, "noreuse" is FALSE, resulting in LRU ordering of the inactive * queue. However, setting "noreuse" to TRUE will accelerate the specified * page's reclamation, but it will not unmap the page from any address space. * This is implemented by inserting the page near the head of the inactive * queue, using a marker page to guide FIFO insertion ordering. * * The page must be locked. */ static inline void _vm_page_deactivate(vm_page_t m, boolean_t noreuse) { struct vm_pagequeue *pq; int queue; vm_page_assert_locked(m); /* * Ignore if the page is already inactive, unless it is unlikely to be * reactivated. */ if ((queue = m->queue) == PQ_INACTIVE && !noreuse) return; if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) { pq = &vm_phys_domain(m)->vmd_pagequeues[PQ_INACTIVE]; /* Avoid multiple acquisitions of the inactive queue lock. */ if (queue == PQ_INACTIVE) { vm_pagequeue_lock(pq); vm_page_dequeue_locked(m); } else { if (queue != PQ_NONE) vm_page_dequeue(m); vm_pagequeue_lock(pq); } m->queue = PQ_INACTIVE; if (noreuse) TAILQ_INSERT_BEFORE(&vm_phys_domain(m)->vmd_inacthead, m, plinks.q); else TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); vm_pagequeue_cnt_inc(pq); vm_pagequeue_unlock(pq); } } /* * Move the specified page to the inactive queue. * * The page must be locked. */ void vm_page_deactivate(vm_page_t m) { _vm_page_deactivate(m, FALSE); } /* * Move the specified page to the inactive queue with the expectation * that it is unlikely to be reused. * * The page must be locked. */ void vm_page_deactivate_noreuse(vm_page_t m) { _vm_page_deactivate(m, TRUE); } /* * vm_page_launder * * Put a page in the laundry. */ void vm_page_launder(vm_page_t m) { int queue; vm_page_assert_locked(m); if ((queue = m->queue) != PQ_LAUNDRY) { if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) { if (queue != PQ_NONE) vm_page_dequeue(m); vm_page_enqueue(PQ_LAUNDRY, m); } else KASSERT(queue == PQ_NONE, ("wired page %p is queued", m)); } } /* * vm_page_try_to_free() * * Attempt to free the page. If we cannot free it, we do nothing. * true is returned on success, false on failure. */ bool vm_page_try_to_free(vm_page_t m) { vm_page_assert_locked(m); if (m->object != NULL) VM_OBJECT_ASSERT_WLOCKED(m->object); if (m->dirty != 0 || m->hold_count != 0 || m->wire_count != 0 || (m->oflags & VPO_UNMANAGED) != 0 || vm_page_busied(m)) return (false); if (m->object != NULL && m->object->ref_count != 0) { pmap_remove_all(m); if (m->dirty != 0) return (false); } vm_page_free(m); return (true); } /* * vm_page_advise * * Apply the specified advice to the given page. * * The object and page must be locked. */ void vm_page_advise(vm_page_t m, int advice) { vm_page_assert_locked(m); VM_OBJECT_ASSERT_WLOCKED(m->object); if (advice == MADV_FREE) /* * Mark the page clean. This will allow the page to be freed * without first paging it out. MADV_FREE pages are often * quickly reused by malloc(3), so we do not do anything that * would result in a page fault on a later access. */ vm_page_undirty(m); else if (advice != MADV_DONTNEED) { if (advice == MADV_WILLNEED) vm_page_activate(m); return; } /* * Clear any references to the page. Otherwise, the page daemon will * immediately reactivate the page. */ vm_page_aflag_clear(m, PGA_REFERENCED); if (advice != MADV_FREE && m->dirty == 0 && pmap_is_modified(m)) vm_page_dirty(m); /* * Place clean pages near the head of the inactive queue rather than * the tail, thus defeating the queue's LRU operation and ensuring that * the page will be reused quickly. Dirty pages not already in the * laundry are moved there. */ if (m->dirty == 0) vm_page_deactivate_noreuse(m); else vm_page_launder(m); } /* * Grab a page, waiting until we are waken up due to the page * changing state. We keep on waiting, if the page continues * to be in the object. If the page doesn't exist, first allocate it * and then conditionally zero it. * * This routine may sleep. * * The object must be locked on entry. The lock will, however, be released * and reacquired if the routine sleeps. */ vm_page_t vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags) { vm_page_t m; int sleep; VM_OBJECT_ASSERT_WLOCKED(object); KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 || (allocflags & VM_ALLOC_IGN_SBUSY) != 0, ("vm_page_grab: VM_ALLOC_SBUSY/VM_ALLOC_IGN_SBUSY mismatch")); retrylookup: if ((m = vm_page_lookup(object, pindex)) != NULL) { sleep = (allocflags & VM_ALLOC_IGN_SBUSY) != 0 ? vm_page_xbusied(m) : vm_page_busied(m); if (sleep) { if ((allocflags & VM_ALLOC_NOWAIT) != 0) return (NULL); /* * Reference the page before unlocking and * sleeping so that the page daemon is less * likely to reclaim it. */ vm_page_aflag_set(m, PGA_REFERENCED); vm_page_lock(m); VM_OBJECT_WUNLOCK(object); vm_page_busy_sleep(m, "pgrbwt", (allocflags & VM_ALLOC_IGN_SBUSY) != 0); VM_OBJECT_WLOCK(object); goto retrylookup; } else { if ((allocflags & VM_ALLOC_WIRED) != 0) { vm_page_lock(m); vm_page_wire(m); vm_page_unlock(m); } if ((allocflags & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0) vm_page_xbusy(m); if ((allocflags & VM_ALLOC_SBUSY) != 0) vm_page_sbusy(m); return (m); } } m = vm_page_alloc(object, pindex, allocflags); if (m == NULL) { if ((allocflags & VM_ALLOC_NOWAIT) != 0) return (NULL); VM_OBJECT_WUNLOCK(object); VM_WAIT; VM_OBJECT_WLOCK(object); goto retrylookup; } if (allocflags & VM_ALLOC_ZERO && (m->flags & PG_ZERO) == 0) pmap_zero_page(m); return (m); } /* * Return the specified range of pages from the given object. For each * page offset within the range, if a page already exists within the object * at that offset and it is busy, then wait for it to change state. If, * instead, the page doesn't exist, then allocate it. * * The caller must always specify an allocation class. * * allocation classes: * VM_ALLOC_NORMAL normal process request * VM_ALLOC_SYSTEM system *really* needs the pages * * The caller must always specify that the pages are to be busied and/or * wired. * * optional allocation flags: * VM_ALLOC_IGN_SBUSY do not sleep on soft busy pages * VM_ALLOC_NOBUSY do not exclusive busy the page * VM_ALLOC_NOWAIT do not sleep * VM_ALLOC_SBUSY set page to sbusy state * VM_ALLOC_WIRED wire the pages * VM_ALLOC_ZERO zero and validate any invalid pages * * If VM_ALLOC_NOWAIT is not specified, this routine may sleep. Otherwise, it * may return a partial prefix of the requested range. */ int vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags, vm_page_t *ma, int count) { vm_page_t m; int i; bool sleep; VM_OBJECT_ASSERT_WLOCKED(object); KASSERT(((u_int)allocflags >> VM_ALLOC_COUNT_SHIFT) == 0, ("vm_page_grap_pages: VM_ALLOC_COUNT() is not allowed")); KASSERT((allocflags & VM_ALLOC_NOBUSY) == 0 || (allocflags & VM_ALLOC_WIRED) != 0, ("vm_page_grab_pages: the pages must be busied or wired")); KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 || (allocflags & VM_ALLOC_IGN_SBUSY) != 0, ("vm_page_grab_pages: VM_ALLOC_SBUSY/IGN_SBUSY mismatch")); if (count == 0) return (0); i = 0; retrylookup: m = vm_page_lookup(object, pindex + i); for (; i < count; i++) { if (m != NULL) { sleep = (allocflags & VM_ALLOC_IGN_SBUSY) != 0 ? vm_page_xbusied(m) : vm_page_busied(m); if (sleep) { if ((allocflags & VM_ALLOC_NOWAIT) != 0) break; /* * Reference the page before unlocking and * sleeping so that the page daemon is less * likely to reclaim it. */ vm_page_aflag_set(m, PGA_REFERENCED); vm_page_lock(m); VM_OBJECT_WUNLOCK(object); vm_page_busy_sleep(m, "grbmaw", (allocflags & VM_ALLOC_IGN_SBUSY) != 0); VM_OBJECT_WLOCK(object); goto retrylookup; } if ((allocflags & VM_ALLOC_WIRED) != 0) { vm_page_lock(m); vm_page_wire(m); vm_page_unlock(m); } if ((allocflags & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0) vm_page_xbusy(m); if ((allocflags & VM_ALLOC_SBUSY) != 0) vm_page_sbusy(m); } else { m = vm_page_alloc(object, pindex + i, (allocflags & ~VM_ALLOC_IGN_SBUSY) | VM_ALLOC_COUNT(count - i)); if (m == NULL) { if ((allocflags & VM_ALLOC_NOWAIT) != 0) break; VM_OBJECT_WUNLOCK(object); VM_WAIT; VM_OBJECT_WLOCK(object); goto retrylookup; } } if (m->valid == 0 && (allocflags & VM_ALLOC_ZERO) != 0) { if ((m->flags & PG_ZERO) == 0) pmap_zero_page(m); m->valid = VM_PAGE_BITS_ALL; } ma[i] = m; m = vm_page_next(m); } return (i); } /* * Mapping function for valid or dirty bits in a page. * * Inputs are required to range within a page. */ vm_page_bits_t vm_page_bits(int base, int size) { int first_bit; int last_bit; KASSERT( base + size <= PAGE_SIZE, ("vm_page_bits: illegal base/size %d/%d", base, size) ); if (size == 0) /* handle degenerate case */ return (0); first_bit = base >> DEV_BSHIFT; last_bit = (base + size - 1) >> DEV_BSHIFT; return (((vm_page_bits_t)2 << last_bit) - ((vm_page_bits_t)1 << first_bit)); } /* * vm_page_set_valid_range: * * Sets portions of a page valid. The arguments are expected * to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive * of any partial chunks touched by the range. The invalid portion of * such chunks will be zeroed. * * (base + size) must be less then or equal to PAGE_SIZE. */ void vm_page_set_valid_range(vm_page_t m, int base, int size) { int endoff, frag; VM_OBJECT_ASSERT_WLOCKED(m->object); if (size == 0) /* handle degenerate case */ return; /* * If the base is not DEV_BSIZE aligned and the valid * bit is clear, we have to zero out a portion of the * first block. */ if ((frag = rounddown2(base, DEV_BSIZE)) != base && (m->valid & (1 << (base >> DEV_BSHIFT))) == 0) pmap_zero_page_area(m, frag, base - frag); /* * If the ending offset is not DEV_BSIZE aligned and the * valid bit is clear, we have to zero out a portion of * the last block. */ endoff = base + size; if ((frag = rounddown2(endoff, DEV_BSIZE)) != endoff && (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0) pmap_zero_page_area(m, endoff, DEV_BSIZE - (endoff & (DEV_BSIZE - 1))); /* * Assert that no previously invalid block that is now being validated * is already dirty. */ KASSERT((~m->valid & vm_page_bits(base, size) & m->dirty) == 0, ("vm_page_set_valid_range: page %p is dirty", m)); /* * Set valid bits inclusive of any overlap. */ m->valid |= vm_page_bits(base, size); } /* * Clear the given bits from the specified page's dirty field. */ static __inline void vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits) { uintptr_t addr; #if PAGE_SIZE < 16384 int shift; #endif /* * If the object is locked and the page is neither exclusive busy nor * write mapped, then the page's dirty field cannot possibly be * set by a concurrent pmap operation. */ VM_OBJECT_ASSERT_WLOCKED(m->object); if (!vm_page_xbusied(m) && !pmap_page_is_write_mapped(m)) m->dirty &= ~pagebits; else { /* * The pmap layer can call vm_page_dirty() without * holding a distinguished lock. The combination of * the object's lock and an atomic operation suffice * to guarantee consistency of the page dirty field. * * For PAGE_SIZE == 32768 case, compiler already * properly aligns the dirty field, so no forcible * alignment is needed. Only require existence of * atomic_clear_64 when page size is 32768. */ addr = (uintptr_t)&m->dirty; #if PAGE_SIZE == 32768 atomic_clear_64((uint64_t *)addr, pagebits); #elif PAGE_SIZE == 16384 atomic_clear_32((uint32_t *)addr, pagebits); #else /* PAGE_SIZE <= 8192 */ /* * Use a trick to perform a 32-bit atomic on the * containing aligned word, to not depend on the existence * of atomic_clear_{8, 16}. */ shift = addr & (sizeof(uint32_t) - 1); #if BYTE_ORDER == BIG_ENDIAN shift = (sizeof(uint32_t) - sizeof(m->dirty) - shift) * NBBY; #else shift *= NBBY; #endif addr &= ~(sizeof(uint32_t) - 1); atomic_clear_32((uint32_t *)addr, pagebits << shift); #endif /* PAGE_SIZE */ } } /* * vm_page_set_validclean: * * Sets portions of a page valid and clean. The arguments are expected * to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive * of any partial chunks touched by the range. The invalid portion of * such chunks will be zero'd. * * (base + size) must be less then or equal to PAGE_SIZE. */ void vm_page_set_validclean(vm_page_t m, int base, int size) { vm_page_bits_t oldvalid, pagebits; int endoff, frag; VM_OBJECT_ASSERT_WLOCKED(m->object); if (size == 0) /* handle degenerate case */ return; /* * If the base is not DEV_BSIZE aligned and the valid * bit is clear, we have to zero out a portion of the * first block. */ if ((frag = rounddown2(base, DEV_BSIZE)) != base && (m->valid & ((vm_page_bits_t)1 << (base >> DEV_BSHIFT))) == 0) pmap_zero_page_area(m, frag, base - frag); /* * If the ending offset is not DEV_BSIZE aligned and the * valid bit is clear, we have to zero out a portion of * the last block. */ endoff = base + size; if ((frag = rounddown2(endoff, DEV_BSIZE)) != endoff && (m->valid & ((vm_page_bits_t)1 << (endoff >> DEV_BSHIFT))) == 0) pmap_zero_page_area(m, endoff, DEV_BSIZE - (endoff & (DEV_BSIZE - 1))); /* * Set valid, clear dirty bits. If validating the entire * page we can safely clear the pmap modify bit. We also * use this opportunity to clear the VPO_NOSYNC flag. If a process * takes a write fault on a MAP_NOSYNC memory area the flag will * be set again. * * We set valid bits inclusive of any overlap, but we can only * clear dirty bits for DEV_BSIZE chunks that are fully within * the range. */ oldvalid = m->valid; pagebits = vm_page_bits(base, size); m->valid |= pagebits; #if 0 /* NOT YET */ if ((frag = base & (DEV_BSIZE - 1)) != 0) { frag = DEV_BSIZE - frag; base += frag; size -= frag; if (size < 0) size = 0; } pagebits = vm_page_bits(base, size & (DEV_BSIZE - 1)); #endif if (base == 0 && size == PAGE_SIZE) { /* * The page can only be modified within the pmap if it is * mapped, and it can only be mapped if it was previously * fully valid. */ if (oldvalid == VM_PAGE_BITS_ALL) /* * Perform the pmap_clear_modify() first. Otherwise, * a concurrent pmap operation, such as * pmap_protect(), could clear a modification in the * pmap and set the dirty field on the page before * pmap_clear_modify() had begun and after the dirty * field was cleared here. */ pmap_clear_modify(m); m->dirty = 0; m->oflags &= ~VPO_NOSYNC; } else if (oldvalid != VM_PAGE_BITS_ALL) m->dirty &= ~pagebits; else vm_page_clear_dirty_mask(m, pagebits); } void vm_page_clear_dirty(vm_page_t m, int base, int size) { vm_page_clear_dirty_mask(m, vm_page_bits(base, size)); } /* * vm_page_set_invalid: * * Invalidates DEV_BSIZE'd chunks within a page. Both the * valid and dirty bits for the effected areas are cleared. */ void vm_page_set_invalid(vm_page_t m, int base, int size) { vm_page_bits_t bits; vm_object_t object; object = m->object; VM_OBJECT_ASSERT_WLOCKED(object); if (object->type == OBJT_VNODE && base == 0 && IDX_TO_OFF(m->pindex) + size >= object->un_pager.vnp.vnp_size) bits = VM_PAGE_BITS_ALL; else bits = vm_page_bits(base, size); if (object->ref_count != 0 && m->valid == VM_PAGE_BITS_ALL && bits != 0) pmap_remove_all(m); KASSERT((bits == 0 && m->valid == VM_PAGE_BITS_ALL) || !pmap_page_is_mapped(m), ("vm_page_set_invalid: page %p is mapped", m)); m->valid &= ~bits; m->dirty &= ~bits; } /* * vm_page_zero_invalid() * * The kernel assumes that the invalid portions of a page contain * garbage, but such pages can be mapped into memory by user code. * When this occurs, we must zero out the non-valid portions of the * page so user code sees what it expects. * * Pages are most often semi-valid when the end of a file is mapped * into memory and the file's size is not page aligned. */ void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid) { int b; int i; VM_OBJECT_ASSERT_WLOCKED(m->object); /* * Scan the valid bits looking for invalid sections that * must be zeroed. Invalid sub-DEV_BSIZE'd areas ( where the * valid bit may be set ) have already been zeroed by * vm_page_set_validclean(). */ for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) { if (i == (PAGE_SIZE / DEV_BSIZE) || (m->valid & ((vm_page_bits_t)1 << i))) { if (i > b) { pmap_zero_page_area(m, b << DEV_BSHIFT, (i - b) << DEV_BSHIFT); } b = i + 1; } } /* * setvalid is TRUE when we can safely set the zero'd areas * as being valid. We can do this if there are no cache consistancy * issues. e.g. it is ok to do with UFS, but not ok to do with NFS. */ if (setvalid) m->valid = VM_PAGE_BITS_ALL; } /* * vm_page_is_valid: * * Is (partial) page valid? Note that the case where size == 0 * will return FALSE in the degenerate case where the page is * entirely invalid, and TRUE otherwise. */ int vm_page_is_valid(vm_page_t m, int base, int size) { vm_page_bits_t bits; VM_OBJECT_ASSERT_LOCKED(m->object); bits = vm_page_bits(base, size); return (m->valid != 0 && (m->valid & bits) == bits); } /* * Returns true if all of the specified predicates are true for the entire * (super)page and false otherwise. */ bool vm_page_ps_test(vm_page_t m, int flags, vm_page_t skip_m) { vm_object_t object; int i, npages; object = m->object; VM_OBJECT_ASSERT_LOCKED(object); npages = atop(pagesizes[m->psind]); /* * The physically contiguous pages that make up a superpage, i.e., a * page with a page size index ("psind") greater than zero, will * occupy adjacent entries in vm_page_array[]. */ for (i = 0; i < npages; i++) { /* Always test object consistency, including "skip_m". */ if (m[i].object != object) return (false); if (&m[i] == skip_m) continue; if ((flags & PS_NONE_BUSY) != 0 && vm_page_busied(&m[i])) return (false); if ((flags & PS_ALL_DIRTY) != 0) { /* * Calling vm_page_test_dirty() or pmap_is_modified() * might stop this case from spuriously returning * "false". However, that would require a write lock * on the object containing "m[i]". */ if (m[i].dirty != VM_PAGE_BITS_ALL) return (false); } if ((flags & PS_ALL_VALID) != 0 && m[i].valid != VM_PAGE_BITS_ALL) return (false); } return (true); } /* * Set the page's dirty bits if the page is modified. */ void vm_page_test_dirty(vm_page_t m) { VM_OBJECT_ASSERT_WLOCKED(m->object); if (m->dirty != VM_PAGE_BITS_ALL && pmap_is_modified(m)) vm_page_dirty(m); } void vm_page_lock_KBI(vm_page_t m, const char *file, int line) { mtx_lock_flags_(vm_page_lockptr(m), 0, file, line); } void vm_page_unlock_KBI(vm_page_t m, const char *file, int line) { mtx_unlock_flags_(vm_page_lockptr(m), 0, file, line); } int vm_page_trylock_KBI(vm_page_t m, const char *file, int line) { return (mtx_trylock_flags_(vm_page_lockptr(m), 0, file, line)); } #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT) void vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line) { vm_page_lock_assert_KBI(m, MA_OWNED, file, line); } void vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line) { mtx_assert_(vm_page_lockptr(m), a, file, line); } #endif #ifdef INVARIANTS void vm_page_object_lock_assert(vm_page_t m) { /* * Certain of the page's fields may only be modified by the * holder of the containing object's lock or the exclusive busy. * holder. Unfortunately, the holder of the write busy is * not recorded, and thus cannot be checked here. */ if (m->object != NULL && !vm_page_xbusied(m)) VM_OBJECT_ASSERT_WLOCKED(m->object); } void vm_page_assert_pga_writeable(vm_page_t m, uint8_t bits) { if ((bits & PGA_WRITEABLE) == 0) return; /* * The PGA_WRITEABLE flag can only be set if the page is * managed, is exclusively busied or the object is locked. * Currently, this flag is only set by pmap_enter(). */ KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("PGA_WRITEABLE on unmanaged page")); if (!vm_page_xbusied(m)) VM_OBJECT_ASSERT_LOCKED(m->object); } #endif #include "opt_ddb.h" #ifdef DDB #include #include DB_SHOW_COMMAND(page, vm_page_print_page_info) { db_printf("vm_cnt.v_free_count: %d\n", vm_cnt.v_free_count); db_printf("vm_cnt.v_inactive_count: %d\n", vm_cnt.v_inactive_count); db_printf("vm_cnt.v_active_count: %d\n", vm_cnt.v_active_count); db_printf("vm_cnt.v_laundry_count: %d\n", vm_cnt.v_laundry_count); db_printf("vm_cnt.v_wire_count: %d\n", vm_cnt.v_wire_count); db_printf("vm_cnt.v_free_reserved: %d\n", vm_cnt.v_free_reserved); db_printf("vm_cnt.v_free_min: %d\n", vm_cnt.v_free_min); db_printf("vm_cnt.v_free_target: %d\n", vm_cnt.v_free_target); db_printf("vm_cnt.v_inactive_target: %d\n", vm_cnt.v_inactive_target); } DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info) { int dom; db_printf("pq_free %d\n", vm_cnt.v_free_count); for (dom = 0; dom < vm_ndomains; dom++) { db_printf( "dom %d page_cnt %d free %d pq_act %d pq_inact %d pq_laund %d\n", dom, vm_dom[dom].vmd_page_count, vm_dom[dom].vmd_free_count, vm_dom[dom].vmd_pagequeues[PQ_ACTIVE].pq_cnt, vm_dom[dom].vmd_pagequeues[PQ_INACTIVE].pq_cnt, vm_dom[dom].vmd_pagequeues[PQ_LAUNDRY].pq_cnt); } } DB_SHOW_COMMAND(pginfo, vm_page_print_pginfo) { vm_page_t m; boolean_t phys; if (!have_addr) { db_printf("show pginfo addr\n"); return; } phys = strchr(modif, 'p') != NULL; if (phys) m = PHYS_TO_VM_PAGE(addr); else m = (vm_page_t)addr; db_printf( "page %p obj %p pidx 0x%jx phys 0x%jx q %d hold %d wire %d\n" " af 0x%x of 0x%x f 0x%x act %d busy %x valid 0x%x dirty 0x%x\n", m, m->object, (uintmax_t)m->pindex, (uintmax_t)m->phys_addr, m->queue, m->hold_count, m->wire_count, m->aflags, m->oflags, m->flags, m->act_count, m->busy_lock, m->valid, m->dirty); } #endif /* DDB */ Index: stable/11/sys/vm/vm_pager.c =================================================================== --- stable/11/sys/vm/vm_pager.c (revision 327403) +++ stable/11/sys/vm/vm_pager.c (revision 327404) @@ -1,546 +1,546 @@ /*- * Copyright (c) 1991, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * The Mach Operating System project at Carnegie-Mellon University. * * 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 * 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: @(#)vm_pager.c 8.6 (Berkeley) 1/12/94 * * * Copyright (c) 1987, 1990 Carnegie-Mellon University. * All rights reserved. * * Authors: Avadis Tevanian, Jr., Michael Wayne Young * * Permission to use, copy, modify and distribute this software and * its documentation is hereby granted, provided that both the copyright * notice and this permission notice appear in all copies of the * software, derivative works or modified versions, and any portions * thereof, and that both notices appear in supporting documentation. * * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. * * Carnegie Mellon requests users of this software to return to * * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU * School of Computer Science * Carnegie Mellon University * Pittsburgh PA 15213-3890 * * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. */ /* * Paging space routine stubs. Emulates a matchmaker-like interface * for builtin pagers. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include int cluster_pbuf_freecnt = -1; /* unlimited to begin with */ struct buf *swbuf; static int dead_pager_getpages(vm_object_t, vm_page_t *, int, int *, int *); static vm_object_t dead_pager_alloc(void *, vm_ooffset_t, vm_prot_t, vm_ooffset_t, struct ucred *); static void dead_pager_putpages(vm_object_t, vm_page_t *, int, int, int *); static boolean_t dead_pager_haspage(vm_object_t, vm_pindex_t, int *, int *); static void dead_pager_dealloc(vm_object_t); static int dead_pager_getpages(vm_object_t obj, vm_page_t *ma, int count, int *rbehind, int *rahead) { return (VM_PAGER_FAIL); } static vm_object_t dead_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t off, struct ucred *cred) { return (NULL); } static void dead_pager_putpages(vm_object_t object, vm_page_t *m, int count, int flags, int *rtvals) { int i; for (i = 0; i < count; i++) rtvals[i] = VM_PAGER_AGAIN; } static int dead_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *prev, int *next) { if (prev != NULL) *prev = 0; if (next != NULL) *next = 0; return (FALSE); } static void dead_pager_dealloc(vm_object_t object) { } static struct pagerops deadpagerops = { .pgo_alloc = dead_pager_alloc, .pgo_dealloc = dead_pager_dealloc, .pgo_getpages = dead_pager_getpages, .pgo_putpages = dead_pager_putpages, .pgo_haspage = dead_pager_haspage, }; struct pagerops *pagertab[] = { &defaultpagerops, /* OBJT_DEFAULT */ &swappagerops, /* OBJT_SWAP */ &vnodepagerops, /* OBJT_VNODE */ &devicepagerops, /* OBJT_DEVICE */ &physpagerops, /* OBJT_PHYS */ &deadpagerops, /* OBJT_DEAD */ &sgpagerops, /* OBJT_SG */ &mgtdevicepagerops, /* OBJT_MGTDEVICE */ }; /* * Kernel address space for mapping pages. * Used by pagers where KVAs are needed for IO. * * XXX needs to be large enough to support the number of pending async * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size * (MAXPHYS == 64k) if you want to get the most efficiency. */ -struct mtx_padalign pbuf_mtx; +struct mtx_padalign __exclusive_cache_line pbuf_mtx; static TAILQ_HEAD(swqueue, buf) bswlist; static int bswneeded; vm_offset_t swapbkva; /* swap buffers kva */ void vm_pager_init(void) { struct pagerops **pgops; TAILQ_INIT(&bswlist); /* * Initialize known pagers */ for (pgops = pagertab; pgops < &pagertab[nitems(pagertab)]; pgops++) if ((*pgops)->pgo_init != NULL) (*(*pgops)->pgo_init)(); } void vm_pager_bufferinit(void) { struct buf *bp; int i; mtx_init(&pbuf_mtx, "pbuf mutex", NULL, MTX_DEF); bp = swbuf; /* * Now set up swap and physical I/O buffer headers. */ for (i = 0; i < nswbuf; i++, bp++) { TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist); BUF_LOCKINIT(bp); LIST_INIT(&bp->b_dep); bp->b_rcred = bp->b_wcred = NOCRED; bp->b_xflags = 0; } cluster_pbuf_freecnt = nswbuf / 2; vnode_pbuf_freecnt = nswbuf / 2 + 1; vnode_async_pbuf_freecnt = nswbuf / 2; } /* * Allocate an instance of a pager of the given type. * Size, protection and offset parameters are passed in for pagers that * need to perform page-level validation (e.g. the device pager). */ vm_object_t vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t off, struct ucred *cred) { vm_object_t ret; struct pagerops *ops; ops = pagertab[type]; if (ops) ret = (*ops->pgo_alloc)(handle, size, prot, off, cred); else ret = NULL; return (ret); } /* * The object must be locked. */ void vm_pager_deallocate(vm_object_t object) { VM_OBJECT_ASSERT_WLOCKED(object); (*pagertab[object->type]->pgo_dealloc) (object); } static void vm_pager_assert_in(vm_object_t object, vm_page_t *m, int count) { #ifdef INVARIANTS VM_OBJECT_ASSERT_WLOCKED(object); KASSERT(count > 0, ("%s: 0 count", __func__)); /* * All pages must be busied, not mapped, not fully valid, * not dirty and belong to the proper object. */ for (int i = 0 ; i < count; i++) { vm_page_assert_xbusied(m[i]); KASSERT(!pmap_page_is_mapped(m[i]), ("%s: page %p is mapped", __func__, m[i])); KASSERT(m[i]->valid != VM_PAGE_BITS_ALL, ("%s: request for a valid page %p", __func__, m[i])); KASSERT(m[i]->dirty == 0, ("%s: page %p is dirty", __func__, m[i])); KASSERT(m[i]->object == object, ("%s: wrong object %p/%p", __func__, object, m[i]->object)); } #endif } /* * Page in the pages for the object using its associated pager. * The requested page must be fully valid on successful return. */ int vm_pager_get_pages(vm_object_t object, vm_page_t *m, int count, int *rbehind, int *rahead) { #ifdef INVARIANTS vm_pindex_t pindex = m[0]->pindex; #endif int r; vm_pager_assert_in(object, m, count); r = (*pagertab[object->type]->pgo_getpages)(object, m, count, rbehind, rahead); if (r != VM_PAGER_OK) return (r); for (int i = 0; i < count; i++) { /* * If pager has replaced a page, assert that it had * updated the array. */ KASSERT(m[i] == vm_page_lookup(object, pindex++), ("%s: mismatch page %p pindex %ju", __func__, m[i], (uintmax_t )pindex - 1)); /* * Zero out partially filled data. */ if (m[i]->valid != VM_PAGE_BITS_ALL) vm_page_zero_invalid(m[i], TRUE); } return (VM_PAGER_OK); } int vm_pager_get_pages_async(vm_object_t object, vm_page_t *m, int count, int *rbehind, int *rahead, pgo_getpages_iodone_t iodone, void *arg) { vm_pager_assert_in(object, m, count); return ((*pagertab[object->type]->pgo_getpages_async)(object, m, count, rbehind, rahead, iodone, arg)); } /* * vm_pager_put_pages() - inline, see vm/vm_pager.h * vm_pager_has_page() - inline, see vm/vm_pager.h */ /* * Search the specified pager object list for an object with the * specified handle. If an object with the specified handle is found, * increase its reference count and return it. Otherwise, return NULL. * * The pager object list must be locked. */ vm_object_t vm_pager_object_lookup(struct pagerlst *pg_list, void *handle) { vm_object_t object; TAILQ_FOREACH(object, pg_list, pager_object_list) { if (object->handle == handle) { VM_OBJECT_WLOCK(object); if ((object->flags & OBJ_DEAD) == 0) { vm_object_reference_locked(object); VM_OBJECT_WUNLOCK(object); break; } VM_OBJECT_WUNLOCK(object); } } return (object); } /* * initialize a physical buffer */ /* * XXX This probably belongs in vfs_bio.c */ static void initpbuf(struct buf *bp) { KASSERT(bp->b_bufobj == NULL, ("initpbuf with bufobj")); KASSERT(bp->b_vp == NULL, ("initpbuf with vp")); bp->b_rcred = NOCRED; bp->b_wcred = NOCRED; bp->b_qindex = 0; /* On no queue (QUEUE_NONE) */ bp->b_kvabase = (caddr_t)(MAXPHYS * (bp - swbuf)) + swapbkva; bp->b_data = bp->b_kvabase; bp->b_kvasize = MAXPHYS; bp->b_flags = 0; bp->b_xflags = 0; bp->b_ioflags = 0; bp->b_iodone = NULL; bp->b_error = 0; BUF_LOCK(bp, LK_EXCLUSIVE, NULL); } /* * allocate a physical buffer * * There are a limited number (nswbuf) of physical buffers. We need * to make sure that no single subsystem is able to hog all of them, * so each subsystem implements a counter which is typically initialized * to 1/2 nswbuf. getpbuf() decrements this counter in allocation and * increments it on release, and blocks if the counter hits zero. A * subsystem may initialize the counter to -1 to disable the feature, * but it must still be sure to match up all uses of getpbuf() with * relpbuf() using the same variable. * * NOTE: pfreecnt can be NULL, but this 'feature' will be removed * relatively soon when the rest of the subsystems get smart about it. XXX */ struct buf * getpbuf(int *pfreecnt) { struct buf *bp; mtx_lock(&pbuf_mtx); for (;;) { if (pfreecnt != NULL) { while (*pfreecnt == 0) { msleep(pfreecnt, &pbuf_mtx, PVM, "wswbuf0", 0); } } /* get a bp from the swap buffer header pool */ if ((bp = TAILQ_FIRST(&bswlist)) != NULL) break; bswneeded = 1; msleep(&bswneeded, &pbuf_mtx, PVM, "wswbuf1", 0); /* loop in case someone else grabbed one */ } TAILQ_REMOVE(&bswlist, bp, b_freelist); if (pfreecnt) --*pfreecnt; mtx_unlock(&pbuf_mtx); initpbuf(bp); return (bp); } /* * allocate a physical buffer, if one is available. * * Note that there is no NULL hack here - all subsystems using this * call understand how to use pfreecnt. */ struct buf * trypbuf(int *pfreecnt) { struct buf *bp; mtx_lock(&pbuf_mtx); if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) { mtx_unlock(&pbuf_mtx); return NULL; } TAILQ_REMOVE(&bswlist, bp, b_freelist); --*pfreecnt; mtx_unlock(&pbuf_mtx); initpbuf(bp); return (bp); } /* * release a physical buffer * * NOTE: pfreecnt can be NULL, but this 'feature' will be removed * relatively soon when the rest of the subsystems get smart about it. XXX */ void relpbuf(struct buf *bp, int *pfreecnt) { if (bp->b_rcred != NOCRED) { crfree(bp->b_rcred); bp->b_rcred = NOCRED; } if (bp->b_wcred != NOCRED) { crfree(bp->b_wcred); bp->b_wcred = NOCRED; } KASSERT(bp->b_vp == NULL, ("relpbuf with vp")); KASSERT(bp->b_bufobj == NULL, ("relpbuf with bufobj")); BUF_UNLOCK(bp); mtx_lock(&pbuf_mtx); TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist); if (bswneeded) { bswneeded = 0; wakeup(&bswneeded); } if (pfreecnt) { if (++*pfreecnt == 1) wakeup(pfreecnt); } mtx_unlock(&pbuf_mtx); } /* * Associate a p-buffer with a vnode. * * Also sets B_PAGING flag to indicate that vnode is not fully associated * with the buffer. i.e. the bp has not been linked into the vnode or * ref-counted. */ void pbgetvp(struct vnode *vp, struct buf *bp) { KASSERT(bp->b_vp == NULL, ("pbgetvp: not free")); KASSERT(bp->b_bufobj == NULL, ("pbgetvp: not free (bufobj)")); bp->b_vp = vp; bp->b_flags |= B_PAGING; bp->b_bufobj = &vp->v_bufobj; } /* * Associate a p-buffer with a vnode. * * Also sets B_PAGING flag to indicate that vnode is not fully associated * with the buffer. i.e. the bp has not been linked into the vnode or * ref-counted. */ void pbgetbo(struct bufobj *bo, struct buf *bp) { KASSERT(bp->b_vp == NULL, ("pbgetbo: not free (vnode)")); KASSERT(bp->b_bufobj == NULL, ("pbgetbo: not free (bufobj)")); bp->b_flags |= B_PAGING; bp->b_bufobj = bo; } /* * Disassociate a p-buffer from a vnode. */ void pbrelvp(struct buf *bp) { KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL")); KASSERT(bp->b_bufobj != NULL, ("pbrelvp: NULL bufobj")); KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0, ("pbrelvp: pager buf on vnode list.")); bp->b_vp = NULL; bp->b_bufobj = NULL; bp->b_flags &= ~B_PAGING; } /* * Disassociate a p-buffer from a bufobj. */ void pbrelbo(struct buf *bp) { KASSERT(bp->b_vp == NULL, ("pbrelbo: vnode")); KASSERT(bp->b_bufobj != NULL, ("pbrelbo: NULL bufobj")); KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0, ("pbrelbo: pager buf on vnode list.")); bp->b_bufobj = NULL; bp->b_flags &= ~B_PAGING; } Index: stable/11 =================================================================== --- stable/11 (revision 327403) +++ stable/11 (revision 327404) Property changes on: stable/11 ___________________________________________________________________ Modified: svn:mergeinfo ## -0,0 +0,1 ## Merged /head:r323234,323305,324044