Index: head/sys/amd64/linux32/linux32_machdep.c =================================================================== --- head/sys/amd64/linux32/linux32_machdep.c (revision 284214) +++ head/sys/amd64/linux32/linux32_machdep.c (revision 284215) @@ -1,1004 +1,1004 @@ /*- * Copyright (c) 2004 Tim J. Robbins * Copyright (c) 2002 Doug Rabson * Copyright (c) 2000 Marcel Moolenaar * 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 * in this position and unchanged. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static void bsd_to_linux_rusage(struct rusage *ru, struct l_rusage *lru); struct l_old_select_argv { l_int nfds; l_uintptr_t readfds; l_uintptr_t writefds; l_uintptr_t exceptfds; l_uintptr_t timeout; } __packed; static int linux_mmap_common(struct thread *td, l_uintptr_t addr, l_size_t len, l_int prot, l_int flags, l_int fd, l_loff_t pos); static void bsd_to_linux_rusage(struct rusage *ru, struct l_rusage *lru) { lru->ru_utime.tv_sec = ru->ru_utime.tv_sec; lru->ru_utime.tv_usec = ru->ru_utime.tv_usec; lru->ru_stime.tv_sec = ru->ru_stime.tv_sec; lru->ru_stime.tv_usec = ru->ru_stime.tv_usec; lru->ru_maxrss = ru->ru_maxrss; lru->ru_ixrss = ru->ru_ixrss; lru->ru_idrss = ru->ru_idrss; lru->ru_isrss = ru->ru_isrss; lru->ru_minflt = ru->ru_minflt; lru->ru_majflt = ru->ru_majflt; lru->ru_nswap = ru->ru_nswap; lru->ru_inblock = ru->ru_inblock; lru->ru_oublock = ru->ru_oublock; lru->ru_msgsnd = ru->ru_msgsnd; lru->ru_msgrcv = ru->ru_msgrcv; lru->ru_nsignals = ru->ru_nsignals; lru->ru_nvcsw = ru->ru_nvcsw; lru->ru_nivcsw = ru->ru_nivcsw; } int linux_copyout_rusage(struct rusage *ru, void *uaddr) { struct l_rusage lru; bsd_to_linux_rusage(ru, &lru); return (copyout(&lru, uaddr, sizeof(struct l_rusage))); } int linux_execve(struct thread *td, struct linux_execve_args *args) { struct image_args eargs; char *path; int error; LCONVPATHEXIST(td, args->path, &path); #ifdef DEBUG if (ldebug(execve)) printf(ARGS(execve, "%s"), path); #endif error = freebsd32_exec_copyin_args(&eargs, path, UIO_SYSSPACE, args->argp, args->envp); free(path, M_TEMP); if (error == 0) error = linux_common_execve(td, &eargs); return (error); } CTASSERT(sizeof(struct l_iovec32) == 8); static int linux32_copyinuio(struct l_iovec32 *iovp, l_ulong iovcnt, struct uio **uiop) { struct l_iovec32 iov32; struct iovec *iov; struct uio *uio; uint32_t iovlen; int error, i; *uiop = NULL; if (iovcnt > UIO_MAXIOV) return (EINVAL); iovlen = iovcnt * sizeof(struct iovec); uio = malloc(iovlen + sizeof(*uio), M_IOV, M_WAITOK); iov = (struct iovec *)(uio + 1); for (i = 0; i < iovcnt; i++) { error = copyin(&iovp[i], &iov32, sizeof(struct l_iovec32)); if (error) { free(uio, M_IOV); return (error); } iov[i].iov_base = PTRIN(iov32.iov_base); iov[i].iov_len = iov32.iov_len; } uio->uio_iov = iov; uio->uio_iovcnt = iovcnt; uio->uio_segflg = UIO_USERSPACE; uio->uio_offset = -1; uio->uio_resid = 0; for (i = 0; i < iovcnt; i++) { if (iov->iov_len > INT_MAX - uio->uio_resid) { free(uio, M_IOV); return (EINVAL); } uio->uio_resid += iov->iov_len; iov++; } *uiop = uio; return (0); } int linux32_copyiniov(struct l_iovec32 *iovp32, l_ulong iovcnt, struct iovec **iovp, int error) { struct l_iovec32 iov32; struct iovec *iov; uint32_t iovlen; int i; *iovp = NULL; if (iovcnt > UIO_MAXIOV) return (error); iovlen = iovcnt * sizeof(struct iovec); iov = malloc(iovlen, M_IOV, M_WAITOK); for (i = 0; i < iovcnt; i++) { error = copyin(&iovp32[i], &iov32, sizeof(struct l_iovec32)); if (error) { free(iov, M_IOV); return (error); } iov[i].iov_base = PTRIN(iov32.iov_base); iov[i].iov_len = iov32.iov_len; } *iovp = iov; return(0); } int linux_readv(struct thread *td, struct linux_readv_args *uap) { struct uio *auio; int error; error = linux32_copyinuio(uap->iovp, uap->iovcnt, &auio); if (error) return (error); error = kern_readv(td, uap->fd, auio); free(auio, M_IOV); return (error); } int linux_writev(struct thread *td, struct linux_writev_args *uap) { struct uio *auio; int error; error = linux32_copyinuio(uap->iovp, uap->iovcnt, &auio); if (error) return (error); error = kern_writev(td, uap->fd, auio); free(auio, M_IOV); return (error); } struct l_ipc_kludge { l_uintptr_t msgp; l_long msgtyp; } __packed; int linux_ipc(struct thread *td, struct linux_ipc_args *args) { switch (args->what & 0xFFFF) { case LINUX_SEMOP: { struct linux_semop_args a; a.semid = args->arg1; a.tsops = args->ptr; a.nsops = args->arg2; return (linux_semop(td, &a)); } case LINUX_SEMGET: { struct linux_semget_args a; a.key = args->arg1; a.nsems = args->arg2; a.semflg = args->arg3; return (linux_semget(td, &a)); } case LINUX_SEMCTL: { struct linux_semctl_args a; int error; a.semid = args->arg1; a.semnum = args->arg2; a.cmd = args->arg3; error = copyin(args->ptr, &a.arg, sizeof(a.arg)); if (error) return (error); return (linux_semctl(td, &a)); } case LINUX_MSGSND: { struct linux_msgsnd_args a; a.msqid = args->arg1; a.msgp = args->ptr; a.msgsz = args->arg2; a.msgflg = args->arg3; return (linux_msgsnd(td, &a)); } case LINUX_MSGRCV: { struct linux_msgrcv_args a; a.msqid = args->arg1; a.msgsz = args->arg2; a.msgflg = args->arg3; if ((args->what >> 16) == 0) { struct l_ipc_kludge tmp; int error; if (args->ptr == 0) return (EINVAL); error = copyin(args->ptr, &tmp, sizeof(tmp)); if (error) return (error); a.msgp = PTRIN(tmp.msgp); a.msgtyp = tmp.msgtyp; } else { a.msgp = args->ptr; a.msgtyp = args->arg5; } return (linux_msgrcv(td, &a)); } case LINUX_MSGGET: { struct linux_msgget_args a; a.key = args->arg1; a.msgflg = args->arg2; return (linux_msgget(td, &a)); } case LINUX_MSGCTL: { struct linux_msgctl_args a; a.msqid = args->arg1; a.cmd = args->arg2; a.buf = args->ptr; return (linux_msgctl(td, &a)); } case LINUX_SHMAT: { struct linux_shmat_args a; a.shmid = args->arg1; a.shmaddr = args->ptr; a.shmflg = args->arg2; a.raddr = PTRIN((l_uint)args->arg3); return (linux_shmat(td, &a)); } case LINUX_SHMDT: { struct linux_shmdt_args a; a.shmaddr = args->ptr; return (linux_shmdt(td, &a)); } case LINUX_SHMGET: { struct linux_shmget_args a; a.key = args->arg1; a.size = args->arg2; a.shmflg = args->arg3; return (linux_shmget(td, &a)); } case LINUX_SHMCTL: { struct linux_shmctl_args a; a.shmid = args->arg1; a.cmd = args->arg2; a.buf = args->ptr; return (linux_shmctl(td, &a)); } default: break; } return (EINVAL); } int linux_old_select(struct thread *td, struct linux_old_select_args *args) { struct l_old_select_argv linux_args; struct linux_select_args newsel; int error; #ifdef DEBUG if (ldebug(old_select)) printf(ARGS(old_select, "%p"), args->ptr); #endif error = copyin(args->ptr, &linux_args, sizeof(linux_args)); if (error) return (error); newsel.nfds = linux_args.nfds; newsel.readfds = PTRIN(linux_args.readfds); newsel.writefds = PTRIN(linux_args.writefds); newsel.exceptfds = PTRIN(linux_args.exceptfds); newsel.timeout = PTRIN(linux_args.timeout); return (linux_select(td, &newsel)); } int linux_set_cloned_tls(struct thread *td, void *desc) { struct user_segment_descriptor sd; struct l_user_desc info; struct pcb *pcb; int error; int a[2]; error = copyin(desc, &info, sizeof(struct l_user_desc)); if (error) { printf(LMSG("copyin failed!")); } else { /* We might copy out the entry_number as GUGS32_SEL. */ info.entry_number = GUGS32_SEL; error = copyout(&info, desc, sizeof(struct l_user_desc)); if (error) printf(LMSG("copyout failed!")); a[0] = LINUX_LDT_entry_a(&info); a[1] = LINUX_LDT_entry_b(&info); memcpy(&sd, &a, sizeof(a)); #ifdef DEBUG if (ldebug(clone)) printf("Segment created in clone with " "CLONE_SETTLS: lobase: %x, hibase: %x, " "lolimit: %x, hilimit: %x, type: %i, " "dpl: %i, p: %i, xx: %i, long: %i, " "def32: %i, gran: %i\n", sd.sd_lobase, sd.sd_hibase, sd.sd_lolimit, sd.sd_hilimit, sd.sd_type, sd.sd_dpl, sd.sd_p, sd.sd_xx, sd.sd_long, sd.sd_def32, sd.sd_gran); #endif pcb = td->td_pcb; pcb->pcb_gsbase = (register_t)info.base_addr; /* XXXKIB pcb->pcb_gs32sd = sd; */ td->td_frame->tf_gs = GSEL(GUGS32_SEL, SEL_UPL); set_pcb_flags(pcb, PCB_32BIT); } return (error); } int linux_set_upcall_kse(struct thread *td, register_t stack) { if (stack) td->td_frame->tf_rsp = stack; /* * The newly created Linux thread returns * to the user space by the same path that a parent do. */ td->td_frame->tf_rax = 0; return (0); } #define STACK_SIZE (2 * 1024 * 1024) #define GUARD_SIZE (4 * PAGE_SIZE) int linux_mmap2(struct thread *td, struct linux_mmap2_args *args) { #ifdef DEBUG if (ldebug(mmap2)) printf(ARGS(mmap2, "0x%08x, %d, %d, 0x%08x, %d, %d"), args->addr, args->len, args->prot, args->flags, args->fd, args->pgoff); #endif return (linux_mmap_common(td, PTROUT(args->addr), args->len, args->prot, args->flags, args->fd, (uint64_t)(uint32_t)args->pgoff * PAGE_SIZE)); } int linux_mmap(struct thread *td, struct linux_mmap_args *args) { int error; struct l_mmap_argv linux_args; error = copyin(args->ptr, &linux_args, sizeof(linux_args)); if (error) return (error); #ifdef DEBUG if (ldebug(mmap)) printf(ARGS(mmap, "0x%08x, %d, %d, 0x%08x, %d, %d"), linux_args.addr, linux_args.len, linux_args.prot, linux_args.flags, linux_args.fd, linux_args.pgoff); #endif return (linux_mmap_common(td, linux_args.addr, linux_args.len, linux_args.prot, linux_args.flags, linux_args.fd, (uint32_t)linux_args.pgoff)); } static int linux_mmap_common(struct thread *td, l_uintptr_t addr, l_size_t len, l_int prot, l_int flags, l_int fd, l_loff_t pos) { struct proc *p = td->td_proc; struct mmap_args /* { caddr_t addr; size_t len; int prot; int flags; int fd; long pad; off_t pos; } */ bsd_args; int error; struct file *fp; cap_rights_t rights; error = 0; bsd_args.flags = 0; fp = NULL; /* * Linux mmap(2): * You must specify exactly one of MAP_SHARED and MAP_PRIVATE */ if (!((flags & LINUX_MAP_SHARED) ^ (flags & LINUX_MAP_PRIVATE))) return (EINVAL); if (flags & LINUX_MAP_SHARED) bsd_args.flags |= MAP_SHARED; if (flags & LINUX_MAP_PRIVATE) bsd_args.flags |= MAP_PRIVATE; if (flags & LINUX_MAP_FIXED) bsd_args.flags |= MAP_FIXED; if (flags & LINUX_MAP_ANON) { /* Enforce pos to be on page boundary, then ignore. */ if ((pos & PAGE_MASK) != 0) return (EINVAL); pos = 0; bsd_args.flags |= MAP_ANON; } else bsd_args.flags |= MAP_NOSYNC; if (flags & LINUX_MAP_GROWSDOWN) bsd_args.flags |= MAP_STACK; /* * PROT_READ, PROT_WRITE, or PROT_EXEC implies PROT_READ and PROT_EXEC * on Linux/i386. We do this to ensure maximum compatibility. * Linux/ia64 does the same in i386 emulation mode. */ bsd_args.prot = prot; if (bsd_args.prot & (PROT_READ | PROT_WRITE | PROT_EXEC)) bsd_args.prot |= PROT_READ | PROT_EXEC; /* Linux does not check file descriptor when MAP_ANONYMOUS is set. */ bsd_args.fd = (bsd_args.flags & MAP_ANON) ? -1 : fd; if (bsd_args.fd != -1) { /* * Linux follows Solaris mmap(2) description: * The file descriptor fildes is opened with * read permission, regardless of the * protection options specified. */ error = fget(td, bsd_args.fd, cap_rights_init(&rights, CAP_MMAP), &fp); if (error != 0) return (error); if (fp->f_type != DTYPE_VNODE) { fdrop(fp, td); return (EINVAL); } /* Linux mmap() just fails for O_WRONLY files */ if (!(fp->f_flag & FREAD)) { fdrop(fp, td); return (EACCES); } fdrop(fp, td); } if (flags & LINUX_MAP_GROWSDOWN) { /* * The Linux MAP_GROWSDOWN option does not limit auto * growth of the region. Linux mmap with this option * takes as addr the inital BOS, and as len, the initial * region size. It can then grow down from addr without * limit. However, Linux threads has an implicit internal * limit to stack size of STACK_SIZE. Its just not * enforced explicitly in Linux. But, here we impose * a limit of (STACK_SIZE - GUARD_SIZE) on the stack * region, since we can do this with our mmap. * * Our mmap with MAP_STACK takes addr as the maximum * downsize limit on BOS, and as len the max size of * the region. It then maps the top SGROWSIZ bytes, * and auto grows the region down, up to the limit * in addr. * * If we don't use the MAP_STACK option, the effect * of this code is to allocate a stack region of a * fixed size of (STACK_SIZE - GUARD_SIZE). */ if ((caddr_t)PTRIN(addr) + len > p->p_vmspace->vm_maxsaddr) { /* * Some Linux apps will attempt to mmap * thread stacks near the top of their * address space. If their TOS is greater * than vm_maxsaddr, vm_map_growstack() * will confuse the thread stack with the * process stack and deliver a SEGV if they * attempt to grow the thread stack past their * current stacksize rlimit. To avoid this, * adjust vm_maxsaddr upwards to reflect * the current stacksize rlimit rather * than the maximum possible stacksize. * It would be better to adjust the * mmap'ed region, but some apps do not check * mmap's return value. */ PROC_LOCK(p); p->p_vmspace->vm_maxsaddr = (char *)LINUX32_USRSTACK - - lim_cur(p, RLIMIT_STACK); + lim_cur_proc(p, RLIMIT_STACK); PROC_UNLOCK(p); } /* * This gives us our maximum stack size and a new BOS. * If we're using VM_STACK, then mmap will just map * the top SGROWSIZ bytes, and let the stack grow down * to the limit at BOS. If we're not using VM_STACK * we map the full stack, since we don't have a way * to autogrow it. */ if (len > STACK_SIZE - GUARD_SIZE) { bsd_args.addr = (caddr_t)PTRIN(addr); bsd_args.len = len; } else { bsd_args.addr = (caddr_t)PTRIN(addr) - (STACK_SIZE - GUARD_SIZE - len); bsd_args.len = STACK_SIZE - GUARD_SIZE; } } else { bsd_args.addr = (caddr_t)PTRIN(addr); bsd_args.len = len; } bsd_args.pos = pos; #ifdef DEBUG if (ldebug(mmap)) printf("-> %s(%p, %d, %d, 0x%08x, %d, 0x%x)\n", __func__, (void *)bsd_args.addr, (int)bsd_args.len, bsd_args.prot, bsd_args.flags, bsd_args.fd, (int)bsd_args.pos); #endif error = sys_mmap(td, &bsd_args); #ifdef DEBUG if (ldebug(mmap)) printf("-> %s() return: 0x%x (0x%08x)\n", __func__, error, (u_int)td->td_retval[0]); #endif return (error); } int linux_mprotect(struct thread *td, struct linux_mprotect_args *uap) { struct mprotect_args bsd_args; bsd_args.addr = uap->addr; bsd_args.len = uap->len; bsd_args.prot = uap->prot; if (bsd_args.prot & (PROT_READ | PROT_WRITE | PROT_EXEC)) bsd_args.prot |= PROT_READ | PROT_EXEC; return (sys_mprotect(td, &bsd_args)); } int linux_iopl(struct thread *td, struct linux_iopl_args *args) { int error; if (args->level < 0 || args->level > 3) return (EINVAL); if ((error = priv_check(td, PRIV_IO)) != 0) return (error); if ((error = securelevel_gt(td->td_ucred, 0)) != 0) return (error); td->td_frame->tf_rflags = (td->td_frame->tf_rflags & ~PSL_IOPL) | (args->level * (PSL_IOPL / 3)); return (0); } int linux_sigaction(struct thread *td, struct linux_sigaction_args *args) { l_osigaction_t osa; l_sigaction_t act, oact; int error; #ifdef DEBUG if (ldebug(sigaction)) printf(ARGS(sigaction, "%d, %p, %p"), args->sig, (void *)args->nsa, (void *)args->osa); #endif if (args->nsa != NULL) { error = copyin(args->nsa, &osa, sizeof(l_osigaction_t)); if (error) return (error); act.lsa_handler = osa.lsa_handler; act.lsa_flags = osa.lsa_flags; act.lsa_restorer = osa.lsa_restorer; LINUX_SIGEMPTYSET(act.lsa_mask); act.lsa_mask.__mask = osa.lsa_mask; } error = linux_do_sigaction(td, args->sig, args->nsa ? &act : NULL, args->osa ? &oact : NULL); if (args->osa != NULL && !error) { osa.lsa_handler = oact.lsa_handler; osa.lsa_flags = oact.lsa_flags; osa.lsa_restorer = oact.lsa_restorer; osa.lsa_mask = oact.lsa_mask.__mask; error = copyout(&osa, args->osa, sizeof(l_osigaction_t)); } return (error); } /* * Linux has two extra args, restart and oldmask. We don't use these, * but it seems that "restart" is actually a context pointer that * enables the signal to happen with a different register set. */ int linux_sigsuspend(struct thread *td, struct linux_sigsuspend_args *args) { sigset_t sigmask; l_sigset_t mask; #ifdef DEBUG if (ldebug(sigsuspend)) printf(ARGS(sigsuspend, "%08lx"), (unsigned long)args->mask); #endif LINUX_SIGEMPTYSET(mask); mask.__mask = args->mask; linux_to_bsd_sigset(&mask, &sigmask); return (kern_sigsuspend(td, sigmask)); } int linux_rt_sigsuspend(struct thread *td, struct linux_rt_sigsuspend_args *uap) { l_sigset_t lmask; sigset_t sigmask; int error; #ifdef DEBUG if (ldebug(rt_sigsuspend)) printf(ARGS(rt_sigsuspend, "%p, %d"), (void *)uap->newset, uap->sigsetsize); #endif if (uap->sigsetsize != sizeof(l_sigset_t)) return (EINVAL); error = copyin(uap->newset, &lmask, sizeof(l_sigset_t)); if (error) return (error); linux_to_bsd_sigset(&lmask, &sigmask); return (kern_sigsuspend(td, sigmask)); } int linux_pause(struct thread *td, struct linux_pause_args *args) { struct proc *p = td->td_proc; sigset_t sigmask; #ifdef DEBUG if (ldebug(pause)) printf(ARGS(pause, "")); #endif PROC_LOCK(p); sigmask = td->td_sigmask; PROC_UNLOCK(p); return (kern_sigsuspend(td, sigmask)); } int linux_sigaltstack(struct thread *td, struct linux_sigaltstack_args *uap) { stack_t ss, oss; l_stack_t lss; int error; #ifdef DEBUG if (ldebug(sigaltstack)) printf(ARGS(sigaltstack, "%p, %p"), uap->uss, uap->uoss); #endif if (uap->uss != NULL) { error = copyin(uap->uss, &lss, sizeof(l_stack_t)); if (error) return (error); ss.ss_sp = PTRIN(lss.ss_sp); ss.ss_size = lss.ss_size; ss.ss_flags = linux_to_bsd_sigaltstack(lss.ss_flags); } error = kern_sigaltstack(td, (uap->uss != NULL) ? &ss : NULL, (uap->uoss != NULL) ? &oss : NULL); if (!error && uap->uoss != NULL) { lss.ss_sp = PTROUT(oss.ss_sp); lss.ss_size = oss.ss_size; lss.ss_flags = bsd_to_linux_sigaltstack(oss.ss_flags); error = copyout(&lss, uap->uoss, sizeof(l_stack_t)); } return (error); } int linux_ftruncate64(struct thread *td, struct linux_ftruncate64_args *args) { struct ftruncate_args sa; #ifdef DEBUG if (ldebug(ftruncate64)) printf(ARGS(ftruncate64, "%u, %jd"), args->fd, (intmax_t)args->length); #endif sa.fd = args->fd; sa.length = args->length; return sys_ftruncate(td, &sa); } int linux_gettimeofday(struct thread *td, struct linux_gettimeofday_args *uap) { struct timeval atv; l_timeval atv32; struct timezone rtz; int error = 0; if (uap->tp) { microtime(&atv); atv32.tv_sec = atv.tv_sec; atv32.tv_usec = atv.tv_usec; error = copyout(&atv32, uap->tp, sizeof(atv32)); } if (error == 0 && uap->tzp != NULL) { rtz.tz_minuteswest = tz_minuteswest; rtz.tz_dsttime = tz_dsttime; error = copyout(&rtz, uap->tzp, sizeof(rtz)); } return (error); } int linux_settimeofday(struct thread *td, struct linux_settimeofday_args *uap) { l_timeval atv32; struct timeval atv, *tvp; struct timezone atz, *tzp; int error; if (uap->tp) { error = copyin(uap->tp, &atv32, sizeof(atv32)); if (error) return (error); atv.tv_sec = atv32.tv_sec; atv.tv_usec = atv32.tv_usec; tvp = &atv; } else tvp = NULL; if (uap->tzp) { error = copyin(uap->tzp, &atz, sizeof(atz)); if (error) return (error); tzp = &atz; } else tzp = NULL; return (kern_settimeofday(td, tvp, tzp)); } int linux_getrusage(struct thread *td, struct linux_getrusage_args *uap) { struct rusage s; int error; error = kern_getrusage(td, uap->who, &s); if (error != 0) return (error); if (uap->rusage != NULL) error = linux_copyout_rusage(&s, uap->rusage); return (error); } int linux_set_thread_area(struct thread *td, struct linux_set_thread_area_args *args) { struct l_user_desc info; struct user_segment_descriptor sd; struct pcb *pcb; int a[2]; int error; error = copyin(args->desc, &info, sizeof(struct l_user_desc)); if (error) return (error); #ifdef DEBUG if (ldebug(set_thread_area)) printf(ARGS(set_thread_area, "%i, %x, %x, %i, %i, %i, " "%i, %i, %i"), info.entry_number, info.base_addr, info.limit, info.seg_32bit, info.contents, info.read_exec_only, info.limit_in_pages, info.seg_not_present, info.useable); #endif /* * Semantics of Linux version: every thread in the system has array * of three TLS descriptors. 1st is GLIBC TLS, 2nd is WINE, 3rd unknown. * This syscall loads one of the selected TLS decriptors with a value * and also loads GDT descriptors 6, 7 and 8 with the content of * the per-thread descriptors. * * Semantics of FreeBSD version: I think we can ignore that Linux has * three per-thread descriptors and use just the first one. * The tls_array[] is used only in [gs]et_thread_area() syscalls and * for loading the GDT descriptors. We use just one GDT descriptor * for TLS, so we will load just one. * * XXX: This doesn't work when a user space process tries to use more * than one TLS segment. Comment in the Linux source says wine might * do this. */ /* * GLIBC reads current %gs and call set_thread_area() with it. * We should let GUDATA_SEL and GUGS32_SEL proceed as well because * we use these segments. */ switch (info.entry_number) { case GUGS32_SEL: case GUDATA_SEL: case 6: case -1: info.entry_number = GUGS32_SEL; break; default: return (EINVAL); } /* * We have to copy out the GDT entry we use. * * XXX: What if a user space program does not check the return value * and tries to use 6, 7 or 8? */ error = copyout(&info, args->desc, sizeof(struct l_user_desc)); if (error) return (error); if (LINUX_LDT_empty(&info)) { a[0] = 0; a[1] = 0; } else { a[0] = LINUX_LDT_entry_a(&info); a[1] = LINUX_LDT_entry_b(&info); } memcpy(&sd, &a, sizeof(a)); #ifdef DEBUG if (ldebug(set_thread_area)) printf("Segment created in set_thread_area: " "lobase: %x, hibase: %x, lolimit: %x, hilimit: %x, " "type: %i, dpl: %i, p: %i, xx: %i, long: %i, " "def32: %i, gran: %i\n", sd.sd_lobase, sd.sd_hibase, sd.sd_lolimit, sd.sd_hilimit, sd.sd_type, sd.sd_dpl, sd.sd_p, sd.sd_xx, sd.sd_long, sd.sd_def32, sd.sd_gran); #endif pcb = td->td_pcb; pcb->pcb_gsbase = (register_t)info.base_addr; set_pcb_flags(pcb, PCB_32BIT); update_gdt_gsbase(td, info.base_addr); return (0); } Index: head/sys/compat/linux/linux_misc.c =================================================================== --- head/sys/compat/linux/linux_misc.c (revision 284214) +++ head/sys/compat/linux/linux_misc.c (revision 284215) @@ -1,2508 +1,2502 @@ /*- * Copyright (c) 2002 Doug Rabson * Copyright (c) 1994-1995 Søren Schmidt * 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 * in this position and unchanged. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #include #include #include #if defined(__i386__) #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef COMPAT_LINUX32 #include #include #else #include #include #endif #include #include #include #include #include #include #include #include #include /** * Special DTrace provider for the linuxulator. * * In this file we define the provider for the entire linuxulator. All * modules (= files of the linuxulator) use it. * * We define a different name depending on the emulated bitsize, see * ../..//linux{,32}/linux.h, e.g.: * native bitsize = linuxulator * amd64, 32bit emulation = linuxulator32 */ LIN_SDT_PROVIDER_DEFINE(LINUX_DTRACE); int stclohz; /* Statistics clock frequency */ static unsigned int linux_to_bsd_resource[LINUX_RLIM_NLIMITS] = { RLIMIT_CPU, RLIMIT_FSIZE, RLIMIT_DATA, RLIMIT_STACK, RLIMIT_CORE, RLIMIT_RSS, RLIMIT_NPROC, RLIMIT_NOFILE, RLIMIT_MEMLOCK, RLIMIT_AS }; struct l_sysinfo { l_long uptime; /* Seconds since boot */ l_ulong loads[3]; /* 1, 5, and 15 minute load averages */ #define LINUX_SYSINFO_LOADS_SCALE 65536 l_ulong totalram; /* Total usable main memory size */ l_ulong freeram; /* Available memory size */ l_ulong sharedram; /* Amount of shared memory */ l_ulong bufferram; /* Memory used by buffers */ l_ulong totalswap; /* Total swap space size */ l_ulong freeswap; /* swap space still available */ l_ushort procs; /* Number of current processes */ l_ushort pads; l_ulong totalbig; l_ulong freebig; l_uint mem_unit; char _f[20-2*sizeof(l_long)-sizeof(l_int)]; /* padding */ }; struct l_pselect6arg { l_uintptr_t ss; l_size_t ss_len; }; static int linux_utimensat_nsec_valid(l_long); int linux_sysinfo(struct thread *td, struct linux_sysinfo_args *args) { struct l_sysinfo sysinfo; vm_object_t object; int i, j; struct timespec ts; getnanouptime(&ts); if (ts.tv_nsec != 0) ts.tv_sec++; sysinfo.uptime = ts.tv_sec; /* Use the information from the mib to get our load averages */ for (i = 0; i < 3; i++) sysinfo.loads[i] = averunnable.ldavg[i] * LINUX_SYSINFO_LOADS_SCALE / averunnable.fscale; sysinfo.totalram = physmem * PAGE_SIZE; sysinfo.freeram = sysinfo.totalram - vm_cnt.v_wire_count * PAGE_SIZE; sysinfo.sharedram = 0; mtx_lock(&vm_object_list_mtx); TAILQ_FOREACH(object, &vm_object_list, object_list) if (object->shadow_count > 1) sysinfo.sharedram += object->resident_page_count; mtx_unlock(&vm_object_list_mtx); sysinfo.sharedram *= PAGE_SIZE; sysinfo.bufferram = 0; swap_pager_status(&i, &j); sysinfo.totalswap = i * PAGE_SIZE; sysinfo.freeswap = (i - j) * PAGE_SIZE; sysinfo.procs = nprocs; /* The following are only present in newer Linux kernels. */ sysinfo.totalbig = 0; sysinfo.freebig = 0; sysinfo.mem_unit = 1; return (copyout(&sysinfo, args->info, sizeof(sysinfo))); } int linux_alarm(struct thread *td, struct linux_alarm_args *args) { struct itimerval it, old_it; u_int secs; int error; #ifdef DEBUG if (ldebug(alarm)) printf(ARGS(alarm, "%u"), args->secs); #endif secs = args->secs; if (secs > INT_MAX) secs = INT_MAX; it.it_value.tv_sec = (long) secs; it.it_value.tv_usec = 0; it.it_interval.tv_sec = 0; it.it_interval.tv_usec = 0; error = kern_setitimer(td, ITIMER_REAL, &it, &old_it); if (error) return (error); if (timevalisset(&old_it.it_value)) { if (old_it.it_value.tv_usec != 0) old_it.it_value.tv_sec++; td->td_retval[0] = old_it.it_value.tv_sec; } return (0); } int linux_brk(struct thread *td, struct linux_brk_args *args) { struct vmspace *vm = td->td_proc->p_vmspace; vm_offset_t new, old; struct obreak_args /* { char * nsize; } */ tmp; #ifdef DEBUG if (ldebug(brk)) printf(ARGS(brk, "%p"), (void *)(uintptr_t)args->dsend); #endif old = (vm_offset_t)vm->vm_daddr + ctob(vm->vm_dsize); new = (vm_offset_t)args->dsend; tmp.nsize = (char *)new; if (((caddr_t)new > vm->vm_daddr) && !sys_obreak(td, &tmp)) td->td_retval[0] = (long)new; else td->td_retval[0] = (long)old; return (0); } #if defined(__i386__) /* XXX: what about amd64/linux32? */ int linux_uselib(struct thread *td, struct linux_uselib_args *args) { struct nameidata ni; struct vnode *vp; struct exec *a_out; struct vattr attr; vm_offset_t vmaddr; unsigned long file_offset; unsigned long bss_size; char *library; ssize_t aresid; int error, locked, writecount; LCONVPATHEXIST(td, args->library, &library); #ifdef DEBUG if (ldebug(uselib)) printf(ARGS(uselib, "%s"), library); #endif a_out = NULL; locked = 0; vp = NULL; NDINIT(&ni, LOOKUP, ISOPEN | FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE, library, td); error = namei(&ni); LFREEPATH(library); if (error) goto cleanup; vp = ni.ni_vp; NDFREE(&ni, NDF_ONLY_PNBUF); /* * From here on down, we have a locked vnode that must be unlocked. * XXX: The code below largely duplicates exec_check_permissions(). */ locked = 1; /* Writable? */ error = VOP_GET_WRITECOUNT(vp, &writecount); if (error != 0) goto cleanup; if (writecount != 0) { error = ETXTBSY; goto cleanup; } /* Executable? */ error = VOP_GETATTR(vp, &attr, td->td_ucred); if (error) goto cleanup; if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || ((attr.va_mode & 0111) == 0) || (attr.va_type != VREG)) { /* EACCESS is what exec(2) returns. */ error = ENOEXEC; goto cleanup; } /* Sensible size? */ if (attr.va_size == 0) { error = ENOEXEC; goto cleanup; } /* Can we access it? */ error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); if (error) goto cleanup; /* * XXX: This should use vn_open() so that it is properly authorized, * and to reduce code redundancy all over the place here. * XXX: Not really, it duplicates far more of exec_check_permissions() * than vn_open(). */ #ifdef MAC error = mac_vnode_check_open(td->td_ucred, vp, VREAD); if (error) goto cleanup; #endif error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); if (error) goto cleanup; /* Pull in executable header into exec_map */ error = vm_mmap(exec_map, (vm_offset_t *)&a_out, PAGE_SIZE, VM_PROT_READ, VM_PROT_READ, 0, OBJT_VNODE, vp, 0); if (error) goto cleanup; /* Is it a Linux binary ? */ if (((a_out->a_magic >> 16) & 0xff) != 0x64) { error = ENOEXEC; goto cleanup; } /* * While we are here, we should REALLY do some more checks */ /* Set file/virtual offset based on a.out variant. */ switch ((int)(a_out->a_magic & 0xffff)) { case 0413: /* ZMAGIC */ file_offset = 1024; break; case 0314: /* QMAGIC */ file_offset = 0; break; default: error = ENOEXEC; goto cleanup; } bss_size = round_page(a_out->a_bss); /* Check various fields in header for validity/bounds. */ if (a_out->a_text & PAGE_MASK || a_out->a_data & PAGE_MASK) { error = ENOEXEC; goto cleanup; } /* text + data can't exceed file size */ if (a_out->a_data + a_out->a_text > attr.va_size) { error = EFAULT; goto cleanup; } /* * text/data/bss must not exceed limits * XXX - this is not complete. it should check current usage PLUS * the resources needed by this library. */ PROC_LOCK(td->td_proc); if (a_out->a_text > maxtsiz || - a_out->a_data + bss_size > lim_cur(td->td_proc, RLIMIT_DATA) || + a_out->a_data + bss_size > lim_cur_proc(td->td_proc, RLIMIT_DATA) || racct_set(td->td_proc, RACCT_DATA, a_out->a_data + bss_size) != 0) { PROC_UNLOCK(td->td_proc); error = ENOMEM; goto cleanup; } PROC_UNLOCK(td->td_proc); /* * Prevent more writers. * XXX: Note that if any of the VM operations fail below we don't * clear this flag. */ VOP_SET_TEXT(vp); /* * Lock no longer needed */ locked = 0; VOP_UNLOCK(vp, 0); /* * Check if file_offset page aligned. Currently we cannot handle * misalinged file offsets, and so we read in the entire image * (what a waste). */ if (file_offset & PAGE_MASK) { #ifdef DEBUG printf("uselib: Non page aligned binary %lu\n", file_offset); #endif /* Map text+data read/write/execute */ /* a_entry is the load address and is page aligned */ vmaddr = trunc_page(a_out->a_entry); /* get anon user mapping, read+write+execute */ error = vm_map_find(&td->td_proc->p_vmspace->vm_map, NULL, 0, &vmaddr, a_out->a_text + a_out->a_data, 0, VMFS_NO_SPACE, VM_PROT_ALL, VM_PROT_ALL, 0); if (error) goto cleanup; error = vn_rdwr(UIO_READ, vp, (void *)vmaddr, file_offset, a_out->a_text + a_out->a_data, UIO_USERSPACE, 0, td->td_ucred, NOCRED, &aresid, td); if (error != 0) goto cleanup; if (aresid != 0) { error = ENOEXEC; goto cleanup; } } else { #ifdef DEBUG printf("uselib: Page aligned binary %lu\n", file_offset); #endif /* * for QMAGIC, a_entry is 20 bytes beyond the load address * to skip the executable header */ vmaddr = trunc_page(a_out->a_entry); /* * Map it all into the process's space as a single * copy-on-write "data" segment. */ error = vm_mmap(&td->td_proc->p_vmspace->vm_map, &vmaddr, a_out->a_text + a_out->a_data, VM_PROT_ALL, VM_PROT_ALL, MAP_PRIVATE | MAP_FIXED, OBJT_VNODE, vp, file_offset); if (error) goto cleanup; } #ifdef DEBUG printf("mem=%08lx = %08lx %08lx\n", (long)vmaddr, ((long *)vmaddr)[0], ((long *)vmaddr)[1]); #endif if (bss_size != 0) { /* Calculate BSS start address */ vmaddr = trunc_page(a_out->a_entry) + a_out->a_text + a_out->a_data; /* allocate some 'anon' space */ error = vm_map_find(&td->td_proc->p_vmspace->vm_map, NULL, 0, &vmaddr, bss_size, 0, VMFS_NO_SPACE, VM_PROT_ALL, VM_PROT_ALL, 0); if (error) goto cleanup; } cleanup: /* Unlock vnode if needed */ if (locked) VOP_UNLOCK(vp, 0); /* Release the temporary mapping. */ if (a_out) kmap_free_wakeup(exec_map, (vm_offset_t)a_out, PAGE_SIZE); return (error); } #endif /* __i386__ */ int linux_select(struct thread *td, struct linux_select_args *args) { l_timeval ltv; struct timeval tv0, tv1, utv, *tvp; int error; #ifdef DEBUG if (ldebug(select)) printf(ARGS(select, "%d, %p, %p, %p, %p"), args->nfds, (void *)args->readfds, (void *)args->writefds, (void *)args->exceptfds, (void *)args->timeout); #endif /* * Store current time for computation of the amount of * time left. */ if (args->timeout) { if ((error = copyin(args->timeout, <v, sizeof(ltv)))) goto select_out; utv.tv_sec = ltv.tv_sec; utv.tv_usec = ltv.tv_usec; #ifdef DEBUG if (ldebug(select)) printf(LMSG("incoming timeout (%jd/%ld)"), (intmax_t)utv.tv_sec, utv.tv_usec); #endif if (itimerfix(&utv)) { /* * The timeval was invalid. Convert it to something * valid that will act as it does under Linux. */ utv.tv_sec += utv.tv_usec / 1000000; utv.tv_usec %= 1000000; if (utv.tv_usec < 0) { utv.tv_sec -= 1; utv.tv_usec += 1000000; } if (utv.tv_sec < 0) timevalclear(&utv); } microtime(&tv0); tvp = &utv; } else tvp = NULL; error = kern_select(td, args->nfds, args->readfds, args->writefds, args->exceptfds, tvp, LINUX_NFDBITS); #ifdef DEBUG if (ldebug(select)) printf(LMSG("real select returns %d"), error); #endif if (error) goto select_out; if (args->timeout) { if (td->td_retval[0]) { /* * Compute how much time was left of the timeout, * by subtracting the current time and the time * before we started the call, and subtracting * that result from the user-supplied value. */ microtime(&tv1); timevalsub(&tv1, &tv0); timevalsub(&utv, &tv1); if (utv.tv_sec < 0) timevalclear(&utv); } else timevalclear(&utv); #ifdef DEBUG if (ldebug(select)) printf(LMSG("outgoing timeout (%jd/%ld)"), (intmax_t)utv.tv_sec, utv.tv_usec); #endif ltv.tv_sec = utv.tv_sec; ltv.tv_usec = utv.tv_usec; if ((error = copyout(<v, args->timeout, sizeof(ltv)))) goto select_out; } select_out: #ifdef DEBUG if (ldebug(select)) printf(LMSG("select_out -> %d"), error); #endif return (error); } int linux_mremap(struct thread *td, struct linux_mremap_args *args) { struct munmap_args /* { void *addr; size_t len; } */ bsd_args; int error = 0; #ifdef DEBUG if (ldebug(mremap)) printf(ARGS(mremap, "%p, %08lx, %08lx, %08lx"), (void *)(uintptr_t)args->addr, (unsigned long)args->old_len, (unsigned long)args->new_len, (unsigned long)args->flags); #endif if (args->flags & ~(LINUX_MREMAP_FIXED | LINUX_MREMAP_MAYMOVE)) { td->td_retval[0] = 0; return (EINVAL); } /* * Check for the page alignment. * Linux defines PAGE_MASK to be FreeBSD ~PAGE_MASK. */ if (args->addr & PAGE_MASK) { td->td_retval[0] = 0; return (EINVAL); } args->new_len = round_page(args->new_len); args->old_len = round_page(args->old_len); if (args->new_len > args->old_len) { td->td_retval[0] = 0; return (ENOMEM); } if (args->new_len < args->old_len) { bsd_args.addr = (caddr_t)((uintptr_t)args->addr + args->new_len); bsd_args.len = args->old_len - args->new_len; error = sys_munmap(td, &bsd_args); } td->td_retval[0] = error ? 0 : (uintptr_t)args->addr; return (error); } #define LINUX_MS_ASYNC 0x0001 #define LINUX_MS_INVALIDATE 0x0002 #define LINUX_MS_SYNC 0x0004 int linux_msync(struct thread *td, struct linux_msync_args *args) { struct msync_args bsd_args; bsd_args.addr = (caddr_t)(uintptr_t)args->addr; bsd_args.len = (uintptr_t)args->len; bsd_args.flags = args->fl & ~LINUX_MS_SYNC; return (sys_msync(td, &bsd_args)); } int linux_time(struct thread *td, struct linux_time_args *args) { struct timeval tv; l_time_t tm; int error; #ifdef DEBUG if (ldebug(time)) printf(ARGS(time, "*")); #endif microtime(&tv); tm = tv.tv_sec; if (args->tm && (error = copyout(&tm, args->tm, sizeof(tm)))) return (error); td->td_retval[0] = tm; return (0); } struct l_times_argv { l_clock_t tms_utime; l_clock_t tms_stime; l_clock_t tms_cutime; l_clock_t tms_cstime; }; /* * Glibc versions prior to 2.2.1 always use hard-coded CLK_TCK value. * Since 2.2.1 Glibc uses value exported from kernel via AT_CLKTCK * auxiliary vector entry. */ #define CLK_TCK 100 #define CONVOTCK(r) (r.tv_sec * CLK_TCK + r.tv_usec / (1000000 / CLK_TCK)) #define CONVNTCK(r) (r.tv_sec * stclohz + r.tv_usec / (1000000 / stclohz)) #define CONVTCK(r) (linux_kernver(td) >= LINUX_KERNVER_2004000 ? \ CONVNTCK(r) : CONVOTCK(r)) int linux_times(struct thread *td, struct linux_times_args *args) { struct timeval tv, utime, stime, cutime, cstime; struct l_times_argv tms; struct proc *p; int error; #ifdef DEBUG if (ldebug(times)) printf(ARGS(times, "*")); #endif if (args->buf != NULL) { p = td->td_proc; PROC_LOCK(p); PROC_STATLOCK(p); calcru(p, &utime, &stime); PROC_STATUNLOCK(p); calccru(p, &cutime, &cstime); PROC_UNLOCK(p); tms.tms_utime = CONVTCK(utime); tms.tms_stime = CONVTCK(stime); tms.tms_cutime = CONVTCK(cutime); tms.tms_cstime = CONVTCK(cstime); if ((error = copyout(&tms, args->buf, sizeof(tms)))) return (error); } microuptime(&tv); td->td_retval[0] = (int)CONVTCK(tv); return (0); } int linux_newuname(struct thread *td, struct linux_newuname_args *args) { struct l_new_utsname utsname; char osname[LINUX_MAX_UTSNAME]; char osrelease[LINUX_MAX_UTSNAME]; char *p; #ifdef DEBUG if (ldebug(newuname)) printf(ARGS(newuname, "*")); #endif linux_get_osname(td, osname); linux_get_osrelease(td, osrelease); bzero(&utsname, sizeof(utsname)); strlcpy(utsname.sysname, osname, LINUX_MAX_UTSNAME); getcredhostname(td->td_ucred, utsname.nodename, LINUX_MAX_UTSNAME); getcreddomainname(td->td_ucred, utsname.domainname, LINUX_MAX_UTSNAME); strlcpy(utsname.release, osrelease, LINUX_MAX_UTSNAME); strlcpy(utsname.version, version, LINUX_MAX_UTSNAME); for (p = utsname.version; *p != '\0'; ++p) if (*p == '\n') { *p = '\0'; break; } strlcpy(utsname.machine, linux_kplatform, LINUX_MAX_UTSNAME); return (copyout(&utsname, args->buf, sizeof(utsname))); } struct l_utimbuf { l_time_t l_actime; l_time_t l_modtime; }; int linux_utime(struct thread *td, struct linux_utime_args *args) { struct timeval tv[2], *tvp; struct l_utimbuf lut; char *fname; int error; LCONVPATHEXIST(td, args->fname, &fname); #ifdef DEBUG if (ldebug(utime)) printf(ARGS(utime, "%s, *"), fname); #endif if (args->times) { if ((error = copyin(args->times, &lut, sizeof lut))) { LFREEPATH(fname); return (error); } tv[0].tv_sec = lut.l_actime; tv[0].tv_usec = 0; tv[1].tv_sec = lut.l_modtime; tv[1].tv_usec = 0; tvp = tv; } else tvp = NULL; error = kern_utimesat(td, AT_FDCWD, fname, UIO_SYSSPACE, tvp, UIO_SYSSPACE); LFREEPATH(fname); return (error); } int linux_utimes(struct thread *td, struct linux_utimes_args *args) { l_timeval ltv[2]; struct timeval tv[2], *tvp = NULL; char *fname; int error; LCONVPATHEXIST(td, args->fname, &fname); #ifdef DEBUG if (ldebug(utimes)) printf(ARGS(utimes, "%s, *"), fname); #endif if (args->tptr != NULL) { if ((error = copyin(args->tptr, ltv, sizeof ltv))) { LFREEPATH(fname); return (error); } tv[0].tv_sec = ltv[0].tv_sec; tv[0].tv_usec = ltv[0].tv_usec; tv[1].tv_sec = ltv[1].tv_sec; tv[1].tv_usec = ltv[1].tv_usec; tvp = tv; } error = kern_utimesat(td, AT_FDCWD, fname, UIO_SYSSPACE, tvp, UIO_SYSSPACE); LFREEPATH(fname); return (error); } static int linux_utimensat_nsec_valid(l_long nsec) { if (nsec == LINUX_UTIME_OMIT || nsec == LINUX_UTIME_NOW) return (0); if (nsec >= 0 && nsec <= 999999999) return (0); return (1); } int linux_utimensat(struct thread *td, struct linux_utimensat_args *args) { struct l_timespec l_times[2]; struct timespec times[2], *timesp = NULL; char *path = NULL; int error, dfd, flags = 0; dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->dfd; #ifdef DEBUG if (ldebug(utimensat)) printf(ARGS(utimensat, "%d, *"), dfd); #endif if (args->flags & ~LINUX_AT_SYMLINK_NOFOLLOW) return (EINVAL); if (args->times != NULL) { error = copyin(args->times, l_times, sizeof(l_times)); if (error != 0) return (error); if (linux_utimensat_nsec_valid(l_times[0].tv_nsec) != 0 || linux_utimensat_nsec_valid(l_times[1].tv_nsec) != 0) return (EINVAL); times[0].tv_sec = l_times[0].tv_sec; switch (l_times[0].tv_nsec) { case LINUX_UTIME_OMIT: times[0].tv_nsec = UTIME_OMIT; break; case LINUX_UTIME_NOW: times[0].tv_nsec = UTIME_NOW; break; default: times[0].tv_nsec = l_times[0].tv_nsec; } times[1].tv_sec = l_times[1].tv_sec; switch (l_times[1].tv_nsec) { case LINUX_UTIME_OMIT: times[1].tv_nsec = UTIME_OMIT; break; case LINUX_UTIME_NOW: times[1].tv_nsec = UTIME_NOW; break; default: times[1].tv_nsec = l_times[1].tv_nsec; break; } timesp = times; } if (times[0].tv_nsec == UTIME_OMIT && times[1].tv_nsec == UTIME_OMIT) /* This breaks POSIX, but is what the Linux kernel does * _on purpose_ (documented in the man page for utimensat(2)), * so we must follow that behaviour. */ return (0); if (args->pathname != NULL) LCONVPATHEXIST_AT(td, args->pathname, &path, dfd); else if (args->flags != 0) return (EINVAL); if (args->flags & LINUX_AT_SYMLINK_NOFOLLOW) flags |= AT_SYMLINK_NOFOLLOW; if (path == NULL) error = kern_futimens(td, dfd, timesp, UIO_SYSSPACE); else { error = kern_utimensat(td, dfd, path, UIO_SYSSPACE, timesp, UIO_SYSSPACE, flags); LFREEPATH(path); } return (error); } int linux_futimesat(struct thread *td, struct linux_futimesat_args *args) { l_timeval ltv[2]; struct timeval tv[2], *tvp = NULL; char *fname; int error, dfd; dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->dfd; LCONVPATHEXIST_AT(td, args->filename, &fname, dfd); #ifdef DEBUG if (ldebug(futimesat)) printf(ARGS(futimesat, "%s, *"), fname); #endif if (args->utimes != NULL) { if ((error = copyin(args->utimes, ltv, sizeof ltv))) { LFREEPATH(fname); return (error); } tv[0].tv_sec = ltv[0].tv_sec; tv[0].tv_usec = ltv[0].tv_usec; tv[1].tv_sec = ltv[1].tv_sec; tv[1].tv_usec = ltv[1].tv_usec; tvp = tv; } error = kern_utimesat(td, dfd, fname, UIO_SYSSPACE, tvp, UIO_SYSSPACE); LFREEPATH(fname); return (error); } int linux_common_wait(struct thread *td, int pid, int *status, int options, struct rusage *ru) { int error, tmpstat; error = kern_wait(td, pid, &tmpstat, options, ru); if (error) return (error); if (status) { tmpstat &= 0xffff; if (WIFSIGNALED(tmpstat)) tmpstat = (tmpstat & 0xffffff80) | bsd_to_linux_signal(WTERMSIG(tmpstat)); else if (WIFSTOPPED(tmpstat)) tmpstat = (tmpstat & 0xffff00ff) | (bsd_to_linux_signal(WSTOPSIG(tmpstat)) << 8); else if (WIFCONTINUED(tmpstat)) tmpstat = 0xffff; error = copyout(&tmpstat, status, sizeof(int)); } return (error); } #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32)) int linux_waitpid(struct thread *td, struct linux_waitpid_args *args) { struct linux_wait4_args wait4_args; #ifdef DEBUG if (ldebug(waitpid)) printf(ARGS(waitpid, "%d, %p, %d"), args->pid, (void *)args->status, args->options); #endif wait4_args.pid = args->pid; wait4_args.status = args->status; wait4_args.options = args->options; wait4_args.rusage = NULL; return (linux_wait4(td, &wait4_args)); } #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */ int linux_wait4(struct thread *td, struct linux_wait4_args *args) { int error, options; struct rusage ru, *rup; #ifdef DEBUG if (ldebug(wait4)) printf(ARGS(wait4, "%d, %p, %d, %p"), args->pid, (void *)args->status, args->options, (void *)args->rusage); #endif if (args->options & ~(LINUX_WUNTRACED | LINUX_WNOHANG | LINUX_WCONTINUED | __WCLONE | __WNOTHREAD | __WALL)) return (EINVAL); options = WEXITED; linux_to_bsd_waitopts(args->options, &options); if (args->rusage != NULL) rup = &ru; else rup = NULL; error = linux_common_wait(td, args->pid, args->status, options, rup); if (error != 0) return (error); if (args->rusage != NULL) error = linux_copyout_rusage(&ru, args->rusage); return (error); } int linux_waitid(struct thread *td, struct linux_waitid_args *args) { int status, options, sig; struct __wrusage wru; siginfo_t siginfo; l_siginfo_t lsi; idtype_t idtype; struct proc *p; int error; options = 0; linux_to_bsd_waitopts(args->options, &options); if (options & ~(WNOHANG | WNOWAIT | WEXITED | WUNTRACED | WCONTINUED)) return (EINVAL); if (!(options & (WEXITED | WUNTRACED | WCONTINUED))) return (EINVAL); switch (args->idtype) { case LINUX_P_ALL: idtype = P_ALL; break; case LINUX_P_PID: if (args->id <= 0) return (EINVAL); idtype = P_PID; break; case LINUX_P_PGID: if (args->id <= 0) return (EINVAL); idtype = P_PGID; break; default: return (EINVAL); } error = kern_wait6(td, idtype, args->id, &status, options, &wru, &siginfo); if (error != 0) return (error); if (args->rusage != NULL) { error = linux_copyout_rusage(&wru.wru_children, args->rusage); if (error != 0) return (error); } if (args->info != NULL) { p = td->td_proc; if (td->td_retval[0] == 0) bzero(&lsi, sizeof(lsi)); else { sig = bsd_to_linux_signal(siginfo.si_signo); siginfo_to_lsiginfo(&siginfo, &lsi, sig); } error = copyout(&lsi, args->info, sizeof(lsi)); } td->td_retval[0] = 0; return (error); } int linux_mknod(struct thread *td, struct linux_mknod_args *args) { char *path; int error; LCONVPATHCREAT(td, args->path, &path); #ifdef DEBUG if (ldebug(mknod)) printf(ARGS(mknod, "%s, %d, %ju"), path, args->mode, (uintmax_t)args->dev); #endif switch (args->mode & S_IFMT) { case S_IFIFO: case S_IFSOCK: error = kern_mkfifoat(td, AT_FDCWD, path, UIO_SYSSPACE, args->mode); break; case S_IFCHR: case S_IFBLK: error = kern_mknodat(td, AT_FDCWD, path, UIO_SYSSPACE, args->mode, args->dev); break; case S_IFDIR: error = EPERM; break; case 0: args->mode |= S_IFREG; /* FALLTHROUGH */ case S_IFREG: error = kern_openat(td, AT_FDCWD, path, UIO_SYSSPACE, O_WRONLY | O_CREAT | O_TRUNC, args->mode); if (error == 0) kern_close(td, td->td_retval[0]); break; default: error = EINVAL; break; } LFREEPATH(path); return (error); } int linux_mknodat(struct thread *td, struct linux_mknodat_args *args) { char *path; int error, dfd; dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->dfd; LCONVPATHCREAT_AT(td, args->filename, &path, dfd); #ifdef DEBUG if (ldebug(mknodat)) printf(ARGS(mknodat, "%s, %d, %d"), path, args->mode, args->dev); #endif switch (args->mode & S_IFMT) { case S_IFIFO: case S_IFSOCK: error = kern_mkfifoat(td, dfd, path, UIO_SYSSPACE, args->mode); break; case S_IFCHR: case S_IFBLK: error = kern_mknodat(td, dfd, path, UIO_SYSSPACE, args->mode, args->dev); break; case S_IFDIR: error = EPERM; break; case 0: args->mode |= S_IFREG; /* FALLTHROUGH */ case S_IFREG: error = kern_openat(td, dfd, path, UIO_SYSSPACE, O_WRONLY | O_CREAT | O_TRUNC, args->mode); if (error == 0) kern_close(td, td->td_retval[0]); break; default: error = EINVAL; break; } LFREEPATH(path); return (error); } /* * UGH! This is just about the dumbest idea I've ever heard!! */ int linux_personality(struct thread *td, struct linux_personality_args *args) { #ifdef DEBUG if (ldebug(personality)) printf(ARGS(personality, "%lu"), (unsigned long)args->per); #endif if (args->per != 0) return (EINVAL); /* Yes Jim, it's still a Linux... */ td->td_retval[0] = 0; return (0); } struct l_itimerval { l_timeval it_interval; l_timeval it_value; }; #define B2L_ITIMERVAL(bip, lip) \ (bip)->it_interval.tv_sec = (lip)->it_interval.tv_sec; \ (bip)->it_interval.tv_usec = (lip)->it_interval.tv_usec; \ (bip)->it_value.tv_sec = (lip)->it_value.tv_sec; \ (bip)->it_value.tv_usec = (lip)->it_value.tv_usec; int linux_setitimer(struct thread *td, struct linux_setitimer_args *uap) { int error; struct l_itimerval ls; struct itimerval aitv, oitv; #ifdef DEBUG if (ldebug(setitimer)) printf(ARGS(setitimer, "%p, %p"), (void *)uap->itv, (void *)uap->oitv); #endif if (uap->itv == NULL) { uap->itv = uap->oitv; return (linux_getitimer(td, (struct linux_getitimer_args *)uap)); } error = copyin(uap->itv, &ls, sizeof(ls)); if (error != 0) return (error); B2L_ITIMERVAL(&aitv, &ls); #ifdef DEBUG if (ldebug(setitimer)) { printf("setitimer: value: sec: %jd, usec: %ld\n", (intmax_t)aitv.it_value.tv_sec, aitv.it_value.tv_usec); printf("setitimer: interval: sec: %jd, usec: %ld\n", (intmax_t)aitv.it_interval.tv_sec, aitv.it_interval.tv_usec); } #endif error = kern_setitimer(td, uap->which, &aitv, &oitv); if (error != 0 || uap->oitv == NULL) return (error); B2L_ITIMERVAL(&ls, &oitv); return (copyout(&ls, uap->oitv, sizeof(ls))); } int linux_getitimer(struct thread *td, struct linux_getitimer_args *uap) { int error; struct l_itimerval ls; struct itimerval aitv; #ifdef DEBUG if (ldebug(getitimer)) printf(ARGS(getitimer, "%p"), (void *)uap->itv); #endif error = kern_getitimer(td, uap->which, &aitv); if (error != 0) return (error); B2L_ITIMERVAL(&ls, &aitv); return (copyout(&ls, uap->itv, sizeof(ls))); } #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32)) int linux_nice(struct thread *td, struct linux_nice_args *args) { struct setpriority_args bsd_args; bsd_args.which = PRIO_PROCESS; bsd_args.who = 0; /* current process */ bsd_args.prio = args->inc; return (sys_setpriority(td, &bsd_args)); } #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */ int linux_setgroups(struct thread *td, struct linux_setgroups_args *args) { struct ucred *newcred, *oldcred; l_gid_t *linux_gidset; gid_t *bsd_gidset; int ngrp, error; struct proc *p; ngrp = args->gidsetsize; if (ngrp < 0 || ngrp >= ngroups_max + 1) return (EINVAL); linux_gidset = malloc(ngrp * sizeof(*linux_gidset), M_LINUX, M_WAITOK); error = copyin(args->grouplist, linux_gidset, ngrp * sizeof(l_gid_t)); if (error) goto out; newcred = crget(); p = td->td_proc; PROC_LOCK(p); oldcred = crcopysafe(p, newcred); /* * cr_groups[0] holds egid. Setting the whole set from * the supplied set will cause egid to be changed too. * Keep cr_groups[0] unchanged to prevent that. */ if ((error = priv_check_cred(oldcred, PRIV_CRED_SETGROUPS, 0)) != 0) { PROC_UNLOCK(p); crfree(newcred); goto out; } if (ngrp > 0) { newcred->cr_ngroups = ngrp + 1; bsd_gidset = newcred->cr_groups; ngrp--; while (ngrp >= 0) { bsd_gidset[ngrp + 1] = linux_gidset[ngrp]; ngrp--; } } else newcred->cr_ngroups = 1; setsugid(p); proc_set_cred(p, newcred); PROC_UNLOCK(p); crfree(oldcred); error = 0; out: free(linux_gidset, M_LINUX); return (error); } int linux_getgroups(struct thread *td, struct linux_getgroups_args *args) { struct ucred *cred; l_gid_t *linux_gidset; gid_t *bsd_gidset; int bsd_gidsetsz, ngrp, error; cred = td->td_ucred; bsd_gidset = cred->cr_groups; bsd_gidsetsz = cred->cr_ngroups - 1; /* * cr_groups[0] holds egid. Returning the whole set * here will cause a duplicate. Exclude cr_groups[0] * to prevent that. */ if ((ngrp = args->gidsetsize) == 0) { td->td_retval[0] = bsd_gidsetsz; return (0); } if (ngrp < bsd_gidsetsz) return (EINVAL); ngrp = 0; linux_gidset = malloc(bsd_gidsetsz * sizeof(*linux_gidset), M_LINUX, M_WAITOK); while (ngrp < bsd_gidsetsz) { linux_gidset[ngrp] = bsd_gidset[ngrp + 1]; ngrp++; } error = copyout(linux_gidset, args->grouplist, ngrp * sizeof(l_gid_t)); free(linux_gidset, M_LINUX); if (error) return (error); td->td_retval[0] = ngrp; return (0); } int linux_setrlimit(struct thread *td, struct linux_setrlimit_args *args) { struct rlimit bsd_rlim; struct l_rlimit rlim; u_int which; int error; #ifdef DEBUG if (ldebug(setrlimit)) printf(ARGS(setrlimit, "%d, %p"), args->resource, (void *)args->rlim); #endif if (args->resource >= LINUX_RLIM_NLIMITS) return (EINVAL); which = linux_to_bsd_resource[args->resource]; if (which == -1) return (EINVAL); error = copyin(args->rlim, &rlim, sizeof(rlim)); if (error) return (error); bsd_rlim.rlim_cur = (rlim_t)rlim.rlim_cur; bsd_rlim.rlim_max = (rlim_t)rlim.rlim_max; return (kern_setrlimit(td, which, &bsd_rlim)); } #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32)) int linux_old_getrlimit(struct thread *td, struct linux_old_getrlimit_args *args) { struct l_rlimit rlim; - struct proc *p = td->td_proc; struct rlimit bsd_rlim; u_int which; #ifdef DEBUG if (ldebug(old_getrlimit)) printf(ARGS(old_getrlimit, "%d, %p"), args->resource, (void *)args->rlim); #endif if (args->resource >= LINUX_RLIM_NLIMITS) return (EINVAL); which = linux_to_bsd_resource[args->resource]; if (which == -1) return (EINVAL); - PROC_LOCK(p); - lim_rlimit(p, which, &bsd_rlim); - PROC_UNLOCK(p); + lim_rlimit(td, which, &bsd_rlim); #ifdef COMPAT_LINUX32 rlim.rlim_cur = (unsigned int)bsd_rlim.rlim_cur; if (rlim.rlim_cur == UINT_MAX) rlim.rlim_cur = INT_MAX; rlim.rlim_max = (unsigned int)bsd_rlim.rlim_max; if (rlim.rlim_max == UINT_MAX) rlim.rlim_max = INT_MAX; #else rlim.rlim_cur = (unsigned long)bsd_rlim.rlim_cur; if (rlim.rlim_cur == ULONG_MAX) rlim.rlim_cur = LONG_MAX; rlim.rlim_max = (unsigned long)bsd_rlim.rlim_max; if (rlim.rlim_max == ULONG_MAX) rlim.rlim_max = LONG_MAX; #endif return (copyout(&rlim, args->rlim, sizeof(rlim))); } #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */ int linux_getrlimit(struct thread *td, struct linux_getrlimit_args *args) { struct l_rlimit rlim; - struct proc *p = td->td_proc; struct rlimit bsd_rlim; u_int which; #ifdef DEBUG if (ldebug(getrlimit)) printf(ARGS(getrlimit, "%d, %p"), args->resource, (void *)args->rlim); #endif if (args->resource >= LINUX_RLIM_NLIMITS) return (EINVAL); which = linux_to_bsd_resource[args->resource]; if (which == -1) return (EINVAL); - PROC_LOCK(p); - lim_rlimit(p, which, &bsd_rlim); - PROC_UNLOCK(p); + lim_rlimit(td, which, &bsd_rlim); rlim.rlim_cur = (l_ulong)bsd_rlim.rlim_cur; rlim.rlim_max = (l_ulong)bsd_rlim.rlim_max; return (copyout(&rlim, args->rlim, sizeof(rlim))); } int linux_sched_setscheduler(struct thread *td, struct linux_sched_setscheduler_args *args) { struct sched_param sched_param; struct thread *tdt; int error, policy; #ifdef DEBUG if (ldebug(sched_setscheduler)) printf(ARGS(sched_setscheduler, "%d, %d, %p"), args->pid, args->policy, (const void *)args->param); #endif switch (args->policy) { case LINUX_SCHED_OTHER: policy = SCHED_OTHER; break; case LINUX_SCHED_FIFO: policy = SCHED_FIFO; break; case LINUX_SCHED_RR: policy = SCHED_RR; break; default: return (EINVAL); } error = copyin(args->param, &sched_param, sizeof(sched_param)); if (error) return (error); tdt = linux_tdfind(td, args->pid, -1); if (tdt == NULL) return (ESRCH); error = kern_sched_setscheduler(td, tdt, policy, &sched_param); PROC_UNLOCK(tdt->td_proc); return (error); } int linux_sched_getscheduler(struct thread *td, struct linux_sched_getscheduler_args *args) { struct thread *tdt; int error, policy; #ifdef DEBUG if (ldebug(sched_getscheduler)) printf(ARGS(sched_getscheduler, "%d"), args->pid); #endif tdt = linux_tdfind(td, args->pid, -1); if (tdt == NULL) return (ESRCH); error = kern_sched_getscheduler(td, tdt, &policy); PROC_UNLOCK(tdt->td_proc); switch (policy) { case SCHED_OTHER: td->td_retval[0] = LINUX_SCHED_OTHER; break; case SCHED_FIFO: td->td_retval[0] = LINUX_SCHED_FIFO; break; case SCHED_RR: td->td_retval[0] = LINUX_SCHED_RR; break; } return (error); } int linux_sched_get_priority_max(struct thread *td, struct linux_sched_get_priority_max_args *args) { struct sched_get_priority_max_args bsd; #ifdef DEBUG if (ldebug(sched_get_priority_max)) printf(ARGS(sched_get_priority_max, "%d"), args->policy); #endif switch (args->policy) { case LINUX_SCHED_OTHER: bsd.policy = SCHED_OTHER; break; case LINUX_SCHED_FIFO: bsd.policy = SCHED_FIFO; break; case LINUX_SCHED_RR: bsd.policy = SCHED_RR; break; default: return (EINVAL); } return (sys_sched_get_priority_max(td, &bsd)); } int linux_sched_get_priority_min(struct thread *td, struct linux_sched_get_priority_min_args *args) { struct sched_get_priority_min_args bsd; #ifdef DEBUG if (ldebug(sched_get_priority_min)) printf(ARGS(sched_get_priority_min, "%d"), args->policy); #endif switch (args->policy) { case LINUX_SCHED_OTHER: bsd.policy = SCHED_OTHER; break; case LINUX_SCHED_FIFO: bsd.policy = SCHED_FIFO; break; case LINUX_SCHED_RR: bsd.policy = SCHED_RR; break; default: return (EINVAL); } return (sys_sched_get_priority_min(td, &bsd)); } #define REBOOT_CAD_ON 0x89abcdef #define REBOOT_CAD_OFF 0 #define REBOOT_HALT 0xcdef0123 #define REBOOT_RESTART 0x01234567 #define REBOOT_RESTART2 0xA1B2C3D4 #define REBOOT_POWEROFF 0x4321FEDC #define REBOOT_MAGIC1 0xfee1dead #define REBOOT_MAGIC2 0x28121969 #define REBOOT_MAGIC2A 0x05121996 #define REBOOT_MAGIC2B 0x16041998 int linux_reboot(struct thread *td, struct linux_reboot_args *args) { struct reboot_args bsd_args; #ifdef DEBUG if (ldebug(reboot)) printf(ARGS(reboot, "0x%x"), args->cmd); #endif if (args->magic1 != REBOOT_MAGIC1) return (EINVAL); switch (args->magic2) { case REBOOT_MAGIC2: case REBOOT_MAGIC2A: case REBOOT_MAGIC2B: break; default: return (EINVAL); } switch (args->cmd) { case REBOOT_CAD_ON: case REBOOT_CAD_OFF: return (priv_check(td, PRIV_REBOOT)); case REBOOT_HALT: bsd_args.opt = RB_HALT; break; case REBOOT_RESTART: case REBOOT_RESTART2: bsd_args.opt = 0; break; case REBOOT_POWEROFF: bsd_args.opt = RB_POWEROFF; break; default: return (EINVAL); } return (sys_reboot(td, &bsd_args)); } /* * The FreeBSD native getpid(2), getgid(2) and getuid(2) also modify * td->td_retval[1] when COMPAT_43 is defined. This clobbers registers that * are assumed to be preserved. The following lightweight syscalls fixes * this. See also linux_getgid16() and linux_getuid16() in linux_uid16.c * * linux_getpid() - MP SAFE * linux_getgid() - MP SAFE * linux_getuid() - MP SAFE */ int linux_getpid(struct thread *td, struct linux_getpid_args *args) { #ifdef DEBUG if (ldebug(getpid)) printf(ARGS(getpid, "")); #endif td->td_retval[0] = td->td_proc->p_pid; return (0); } int linux_gettid(struct thread *td, struct linux_gettid_args *args) { struct linux_emuldata *em; #ifdef DEBUG if (ldebug(gettid)) printf(ARGS(gettid, "")); #endif em = em_find(td); KASSERT(em != NULL, ("gettid: emuldata not found.\n")); td->td_retval[0] = em->em_tid; return (0); } int linux_getppid(struct thread *td, struct linux_getppid_args *args) { #ifdef DEBUG if (ldebug(getppid)) printf(ARGS(getppid, "")); #endif PROC_LOCK(td->td_proc); td->td_retval[0] = td->td_proc->p_pptr->p_pid; PROC_UNLOCK(td->td_proc); return (0); } int linux_getgid(struct thread *td, struct linux_getgid_args *args) { #ifdef DEBUG if (ldebug(getgid)) printf(ARGS(getgid, "")); #endif td->td_retval[0] = td->td_ucred->cr_rgid; return (0); } int linux_getuid(struct thread *td, struct linux_getuid_args *args) { #ifdef DEBUG if (ldebug(getuid)) printf(ARGS(getuid, "")); #endif td->td_retval[0] = td->td_ucred->cr_ruid; return (0); } int linux_getsid(struct thread *td, struct linux_getsid_args *args) { struct getsid_args bsd; #ifdef DEBUG if (ldebug(getsid)) printf(ARGS(getsid, "%i"), args->pid); #endif bsd.pid = args->pid; return (sys_getsid(td, &bsd)); } int linux_nosys(struct thread *td, struct nosys_args *ignore) { return (ENOSYS); } int linux_getpriority(struct thread *td, struct linux_getpriority_args *args) { struct getpriority_args bsd_args; int error; #ifdef DEBUG if (ldebug(getpriority)) printf(ARGS(getpriority, "%i, %i"), args->which, args->who); #endif bsd_args.which = args->which; bsd_args.who = args->who; error = sys_getpriority(td, &bsd_args); td->td_retval[0] = 20 - td->td_retval[0]; return (error); } int linux_sethostname(struct thread *td, struct linux_sethostname_args *args) { int name[2]; #ifdef DEBUG if (ldebug(sethostname)) printf(ARGS(sethostname, "*, %i"), args->len); #endif name[0] = CTL_KERN; name[1] = KERN_HOSTNAME; return (userland_sysctl(td, name, 2, 0, 0, 0, args->hostname, args->len, 0, 0)); } int linux_setdomainname(struct thread *td, struct linux_setdomainname_args *args) { int name[2]; #ifdef DEBUG if (ldebug(setdomainname)) printf(ARGS(setdomainname, "*, %i"), args->len); #endif name[0] = CTL_KERN; name[1] = KERN_NISDOMAINNAME; return (userland_sysctl(td, name, 2, 0, 0, 0, args->name, args->len, 0, 0)); } int linux_exit_group(struct thread *td, struct linux_exit_group_args *args) { #ifdef DEBUG if (ldebug(exit_group)) printf(ARGS(exit_group, "%i"), args->error_code); #endif LINUX_CTR2(exit_group, "thread(%d) (%d)", td->td_tid, args->error_code); /* * XXX: we should send a signal to the parent if * SIGNAL_EXIT_GROUP is set. We ignore that (temporarily?) * as it doesnt occur often. */ exit1(td, W_EXITCODE(args->error_code, 0)); /* NOTREACHED */ } #define _LINUX_CAPABILITY_VERSION 0x19980330 struct l_user_cap_header { l_int version; l_int pid; }; struct l_user_cap_data { l_int effective; l_int permitted; l_int inheritable; }; int linux_capget(struct thread *td, struct linux_capget_args *args) { struct l_user_cap_header luch; struct l_user_cap_data lucd; int error; if (args->hdrp == NULL) return (EFAULT); error = copyin(args->hdrp, &luch, sizeof(luch)); if (error != 0) return (error); if (luch.version != _LINUX_CAPABILITY_VERSION) { luch.version = _LINUX_CAPABILITY_VERSION; error = copyout(&luch, args->hdrp, sizeof(luch)); if (error) return (error); return (EINVAL); } if (luch.pid) return (EPERM); if (args->datap) { /* * The current implementation doesn't support setting * a capability (it's essentially a stub) so indicate * that no capabilities are currently set or available * to request. */ bzero (&lucd, sizeof(lucd)); error = copyout(&lucd, args->datap, sizeof(lucd)); } return (error); } int linux_capset(struct thread *td, struct linux_capset_args *args) { struct l_user_cap_header luch; struct l_user_cap_data lucd; int error; if (args->hdrp == NULL || args->datap == NULL) return (EFAULT); error = copyin(args->hdrp, &luch, sizeof(luch)); if (error != 0) return (error); if (luch.version != _LINUX_CAPABILITY_VERSION) { luch.version = _LINUX_CAPABILITY_VERSION; error = copyout(&luch, args->hdrp, sizeof(luch)); if (error) return (error); return (EINVAL); } if (luch.pid) return (EPERM); error = copyin(args->datap, &lucd, sizeof(lucd)); if (error != 0) return (error); /* We currently don't support setting any capabilities. */ if (lucd.effective || lucd.permitted || lucd.inheritable) { linux_msg(td, "capset effective=0x%x, permitted=0x%x, " "inheritable=0x%x is not implemented", (int)lucd.effective, (int)lucd.permitted, (int)lucd.inheritable); return (EPERM); } return (0); } int linux_prctl(struct thread *td, struct linux_prctl_args *args) { int error = 0, max_size; struct proc *p = td->td_proc; char comm[LINUX_MAX_COMM_LEN]; struct linux_emuldata *em; int pdeath_signal; #ifdef DEBUG if (ldebug(prctl)) printf(ARGS(prctl, "%d, %ju, %ju, %ju, %ju"), args->option, (uintmax_t)args->arg2, (uintmax_t)args->arg3, (uintmax_t)args->arg4, (uintmax_t)args->arg5); #endif switch (args->option) { case LINUX_PR_SET_PDEATHSIG: if (!LINUX_SIG_VALID(args->arg2)) return (EINVAL); em = em_find(td); KASSERT(em != NULL, ("prctl: emuldata not found.\n")); em->pdeath_signal = args->arg2; break; case LINUX_PR_GET_PDEATHSIG: em = em_find(td); KASSERT(em != NULL, ("prctl: emuldata not found.\n")); pdeath_signal = em->pdeath_signal; error = copyout(&pdeath_signal, (void *)(register_t)args->arg2, sizeof(pdeath_signal)); break; case LINUX_PR_GET_KEEPCAPS: /* * Indicate that we always clear the effective and * permitted capability sets when the user id becomes * non-zero (actually the capability sets are simply * always zero in the current implementation). */ td->td_retval[0] = 0; break; case LINUX_PR_SET_KEEPCAPS: /* * Ignore requests to keep the effective and permitted * capability sets when the user id becomes non-zero. */ break; case LINUX_PR_SET_NAME: /* * To be on the safe side we need to make sure to not * overflow the size a linux program expects. We already * do this here in the copyin, so that we don't need to * check on copyout. */ max_size = MIN(sizeof(comm), sizeof(p->p_comm)); error = copyinstr((void *)(register_t)args->arg2, comm, max_size, NULL); /* Linux silently truncates the name if it is too long. */ if (error == ENAMETOOLONG) { /* * XXX: copyinstr() isn't documented to populate the * array completely, so do a copyin() to be on the * safe side. This should be changed in case * copyinstr() is changed to guarantee this. */ error = copyin((void *)(register_t)args->arg2, comm, max_size - 1); comm[max_size - 1] = '\0'; } if (error) return (error); PROC_LOCK(p); strlcpy(p->p_comm, comm, sizeof(p->p_comm)); PROC_UNLOCK(p); break; case LINUX_PR_GET_NAME: PROC_LOCK(p); strlcpy(comm, p->p_comm, sizeof(comm)); PROC_UNLOCK(p); error = copyout(comm, (void *)(register_t)args->arg2, strlen(comm) + 1); break; default: error = EINVAL; break; } return (error); } int linux_sched_setparam(struct thread *td, struct linux_sched_setparam_args *uap) { struct sched_param sched_param; struct thread *tdt; int error; #ifdef DEBUG if (ldebug(sched_setparam)) printf(ARGS(sched_setparam, "%d, *"), uap->pid); #endif error = copyin(uap->param, &sched_param, sizeof(sched_param)); if (error) return (error); tdt = linux_tdfind(td, uap->pid, -1); if (tdt == NULL) return (ESRCH); error = kern_sched_setparam(td, tdt, &sched_param); PROC_UNLOCK(tdt->td_proc); return (error); } int linux_sched_getparam(struct thread *td, struct linux_sched_getparam_args *uap) { struct sched_param sched_param; struct thread *tdt; int error; #ifdef DEBUG if (ldebug(sched_getparam)) printf(ARGS(sched_getparam, "%d, *"), uap->pid); #endif tdt = linux_tdfind(td, uap->pid, -1); if (tdt == NULL) return (ESRCH); error = kern_sched_getparam(td, tdt, &sched_param); PROC_UNLOCK(tdt->td_proc); if (error == 0) error = copyout(&sched_param, uap->param, sizeof(sched_param)); return (error); } /* * Get affinity of a process. */ int linux_sched_getaffinity(struct thread *td, struct linux_sched_getaffinity_args *args) { int error; struct thread *tdt; struct cpuset_getaffinity_args cga; #ifdef DEBUG if (ldebug(sched_getaffinity)) printf(ARGS(sched_getaffinity, "%d, %d, *"), args->pid, args->len); #endif if (args->len < sizeof(cpuset_t)) return (EINVAL); tdt = linux_tdfind(td, args->pid, -1); if (tdt == NULL) return (ESRCH); PROC_UNLOCK(tdt->td_proc); cga.level = CPU_LEVEL_WHICH; cga.which = CPU_WHICH_TID; cga.id = tdt->td_tid; cga.cpusetsize = sizeof(cpuset_t); cga.mask = (cpuset_t *) args->user_mask_ptr; if ((error = sys_cpuset_getaffinity(td, &cga)) == 0) td->td_retval[0] = sizeof(cpuset_t); return (error); } /* * Set affinity of a process. */ int linux_sched_setaffinity(struct thread *td, struct linux_sched_setaffinity_args *args) { struct cpuset_setaffinity_args csa; struct thread *tdt; #ifdef DEBUG if (ldebug(sched_setaffinity)) printf(ARGS(sched_setaffinity, "%d, %d, *"), args->pid, args->len); #endif if (args->len < sizeof(cpuset_t)) return (EINVAL); tdt = linux_tdfind(td, args->pid, -1); if (tdt == NULL) return (ESRCH); PROC_UNLOCK(tdt->td_proc); csa.level = CPU_LEVEL_WHICH; csa.which = CPU_WHICH_TID; csa.id = tdt->td_tid; csa.cpusetsize = sizeof(cpuset_t); csa.mask = (cpuset_t *) args->user_mask_ptr; return (sys_cpuset_setaffinity(td, &csa)); } struct linux_rlimit64 { uint64_t rlim_cur; uint64_t rlim_max; }; int linux_prlimit64(struct thread *td, struct linux_prlimit64_args *args) { struct rlimit rlim, nrlim; struct linux_rlimit64 lrlim; struct proc *p; u_int which; int flags; int error; #ifdef DEBUG if (ldebug(prlimit64)) printf(ARGS(prlimit64, "%d, %d, %p, %p"), args->pid, args->resource, (void *)args->new, (void *)args->old); #endif if (args->resource >= LINUX_RLIM_NLIMITS) return (EINVAL); which = linux_to_bsd_resource[args->resource]; if (which == -1) return (EINVAL); if (args->new != NULL) { /* * Note. Unlike FreeBSD where rlim is signed 64-bit Linux * rlim is unsigned 64-bit. FreeBSD treats negative limits * as INFINITY so we do not need a conversion even. */ error = copyin(args->new, &nrlim, sizeof(nrlim)); if (error != 0) return (error); } flags = PGET_HOLD | PGET_NOTWEXIT; if (args->new != NULL) flags |= PGET_CANDEBUG; else flags |= PGET_CANSEE; error = pget(args->pid, flags, &p); if (error != 0) return (error); if (args->old != NULL) { PROC_LOCK(p); - lim_rlimit(p, which, &rlim); + lim_rlimit_proc(p, which, &rlim); PROC_UNLOCK(p); if (rlim.rlim_cur == RLIM_INFINITY) lrlim.rlim_cur = LINUX_RLIM_INFINITY; else lrlim.rlim_cur = rlim.rlim_cur; if (rlim.rlim_max == RLIM_INFINITY) lrlim.rlim_max = LINUX_RLIM_INFINITY; else lrlim.rlim_max = rlim.rlim_max; error = copyout(&lrlim, args->old, sizeof(lrlim)); if (error != 0) goto out; } if (args->new != NULL) error = kern_proc_setrlimit(td, p, which, &nrlim); out: PRELE(p); return (error); } int linux_pselect6(struct thread *td, struct linux_pselect6_args *args) { struct timeval utv, tv0, tv1, *tvp; struct l_pselect6arg lpse6; struct l_timespec lts; struct timespec uts; l_sigset_t l_ss; sigset_t *ssp; sigset_t ss; int error; ssp = NULL; if (args->sig != NULL) { error = copyin(args->sig, &lpse6, sizeof(lpse6)); if (error != 0) return (error); if (lpse6.ss_len != sizeof(l_ss)) return (EINVAL); if (lpse6.ss != 0) { error = copyin(PTRIN(lpse6.ss), &l_ss, sizeof(l_ss)); if (error != 0) return (error); linux_to_bsd_sigset(&l_ss, &ss); ssp = &ss; } } /* * Currently glibc changes nanosecond number to microsecond. * This mean losing precision but for now it is hardly seen. */ if (args->tsp != NULL) { error = copyin(args->tsp, <s, sizeof(lts)); if (error != 0) return (error); error = linux_to_native_timespec(&uts, <s); if (error != 0) return (error); TIMESPEC_TO_TIMEVAL(&utv, &uts); if (itimerfix(&utv)) return (EINVAL); microtime(&tv0); tvp = &utv; } else tvp = NULL; error = kern_pselect(td, args->nfds, args->readfds, args->writefds, args->exceptfds, tvp, ssp, LINUX_NFDBITS); if (error == 0 && args->tsp != NULL) { if (td->td_retval[0] != 0) { /* * Compute how much time was left of the timeout, * by subtracting the current time and the time * before we started the call, and subtracting * that result from the user-supplied value. */ microtime(&tv1); timevalsub(&tv1, &tv0); timevalsub(&utv, &tv1); if (utv.tv_sec < 0) timevalclear(&utv); } else timevalclear(&utv); TIMEVAL_TO_TIMESPEC(&utv, &uts); native_to_linux_timespec(<s, &uts); error = copyout(<s, args->tsp, sizeof(lts)); } return (error); } int linux_ppoll(struct thread *td, struct linux_ppoll_args *args) { struct timespec ts0, ts1; struct l_timespec lts; struct timespec uts, *tsp; l_sigset_t l_ss; sigset_t *ssp; sigset_t ss; int error; if (args->sset != NULL) { if (args->ssize != sizeof(l_ss)) return (EINVAL); error = copyin(args->sset, &l_ss, sizeof(l_ss)); if (error) return (error); linux_to_bsd_sigset(&l_ss, &ss); ssp = &ss; } else ssp = NULL; if (args->tsp != NULL) { error = copyin(args->tsp, <s, sizeof(lts)); if (error) return (error); error = linux_to_native_timespec(&uts, <s); if (error != 0) return (error); nanotime(&ts0); tsp = &uts; } else tsp = NULL; error = kern_poll(td, args->fds, args->nfds, tsp, ssp); if (error == 0 && args->tsp != NULL) { if (td->td_retval[0]) { nanotime(&ts1); timespecsub(&ts1, &ts0); timespecsub(&uts, &ts1); if (uts.tv_sec < 0) timespecclear(&uts); } else timespecclear(&uts); native_to_linux_timespec(<s, &uts); error = copyout(<s, args->tsp, sizeof(lts)); } return (error); } #if defined(DEBUG) || defined(KTR) /* XXX: can be removed when every ldebug(...) and KTR stuff are removed. */ u_char linux_debug_map[howmany(LINUX_SYS_MAXSYSCALL, sizeof(u_char))]; static int linux_debug(int syscall, int toggle, int global) { if (global) { char c = toggle ? 0 : 0xff; memset(linux_debug_map, c, sizeof(linux_debug_map)); return (0); } if (syscall < 0 || syscall >= LINUX_SYS_MAXSYSCALL) return (EINVAL); if (toggle) clrbit(linux_debug_map, syscall); else setbit(linux_debug_map, syscall); return (0); } /* * Usage: sysctl linux.debug=.<0/1> * * E.g.: sysctl linux.debug=21.0 * * As a special case, syscall "all" will apply to all syscalls globally. */ #define LINUX_MAX_DEBUGSTR 16 int linux_sysctl_debug(SYSCTL_HANDLER_ARGS) { char value[LINUX_MAX_DEBUGSTR], *p; int error, sysc, toggle; int global = 0; value[0] = '\0'; error = sysctl_handle_string(oidp, value, LINUX_MAX_DEBUGSTR, req); if (error || req->newptr == NULL) return (error); for (p = value; *p != '\0' && *p != '.'; p++); if (*p == '\0') return (EINVAL); *p++ = '\0'; sysc = strtol(value, NULL, 0); toggle = strtol(p, NULL, 0); if (strcmp(value, "all") == 0) global = 1; error = linux_debug(sysc, toggle, global); return (error); } #endif /* DEBUG || KTR */ int linux_sched_rr_get_interval(struct thread *td, struct linux_sched_rr_get_interval_args *uap) { struct timespec ts; struct l_timespec lts; struct thread *tdt; int error; /* * According to man in case the invalid pid specified * EINVAL should be returned. */ if (uap->pid < 0) return (EINVAL); tdt = linux_tdfind(td, uap->pid, -1); if (tdt == NULL) return (ESRCH); error = kern_sched_rr_get_interval_td(td, tdt, &ts); PROC_UNLOCK(tdt->td_proc); if (error != 0) return (error); native_to_linux_timespec(<s, &ts); return (copyout(<s, uap->interval, sizeof(lts))); } /* * In case when the Linux thread is the initial thread in * the thread group thread id is equal to the process id. * Glibc depends on this magic (assert in pthread_getattr_np.c). */ struct thread * linux_tdfind(struct thread *td, lwpid_t tid, pid_t pid) { struct linux_emuldata *em; struct thread *tdt; struct proc *p; tdt = NULL; if (tid == 0 || tid == td->td_tid) { tdt = td; PROC_LOCK(tdt->td_proc); } else if (tid > PID_MAX) tdt = tdfind(tid, pid); else { /* * Initial thread where the tid equal to the pid. */ p = pfind(tid); if (p != NULL) { if (SV_PROC_ABI(p) != SV_ABI_LINUX) { /* * p is not a Linuxulator process. */ PROC_UNLOCK(p); return (NULL); } FOREACH_THREAD_IN_PROC(p, tdt) { em = em_find(tdt); if (tid == em->em_tid) return (tdt); } PROC_UNLOCK(p); } return (NULL); } return (tdt); } void linux_to_bsd_waitopts(int options, int *bsdopts) { if (options & LINUX_WNOHANG) *bsdopts |= WNOHANG; if (options & LINUX_WUNTRACED) *bsdopts |= WUNTRACED; if (options & LINUX_WEXITED) *bsdopts |= WEXITED; if (options & LINUX_WCONTINUED) *bsdopts |= WCONTINUED; if (options & LINUX_WNOWAIT) *bsdopts |= WNOWAIT; if (options & __WCLONE) *bsdopts |= WLINUXCLONE; } Index: head/sys/compat/svr4/imgact_svr4.c =================================================================== --- head/sys/compat/svr4/imgact_svr4.c (revision 284214) +++ head/sys/compat/svr4/imgact_svr4.c (revision 284215) @@ -1,238 +1,238 @@ /*- * Copyright (c) 1998 Mark Newton * Copyright (c) 1994-1996 Søren Schmidt * All rights reserved. * * Based heavily on /sys/kern/imgact_aout.c which is: * Copyright (c) 1993, David Greenman * * 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 * in this position and unchanged. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static int exec_svr4_imgact(struct image_params *iparams); static int exec_svr4_imgact(imgp) struct image_params *imgp; { const struct exec *a_out = (const struct exec *) imgp->image_header; struct vmspace *vmspace; vm_offset_t vmaddr; unsigned long virtual_offset, file_offset; unsigned long bss_size; ssize_t aresid; int error; if (((a_out->a_magic >> 16) & 0xff) != 0x64) return -1; /* * Set file/virtual offset based on a.out variant. */ switch ((int)(a_out->a_magic & 0xffff)) { case 0413: virtual_offset = 0; file_offset = 1024; break; case 0314: virtual_offset = 4096; file_offset = 0; break; default: return (-1); } bss_size = round_page(a_out->a_bss); #ifdef DEBUG printf("imgact: text: %08lx, data: %08lx, bss: %08lx\n", (u_long)a_out->a_text, (u_long)a_out->a_data, bss_size); #endif /* * Check various fields in header for validity/bounds. */ if (a_out->a_entry < virtual_offset || a_out->a_entry >= virtual_offset + a_out->a_text || a_out->a_text & PAGE_MASK || a_out->a_data & PAGE_MASK) return (-1); /* text + data can't exceed file size */ if (a_out->a_data + a_out->a_text > imgp->attr->va_size) return (EFAULT); /* * text/data/bss must not exceed limits */ PROC_LOCK(imgp->proc); if (a_out->a_text > maxtsiz || - a_out->a_data + bss_size > lim_cur(imgp->proc, RLIMIT_DATA) || + a_out->a_data + bss_size > lim_cur_proc(imgp->proc, RLIMIT_DATA) || racct_set(imgp->proc, RACCT_DATA, a_out->a_data + bss_size) != 0) { PROC_UNLOCK(imgp->proc); return (ENOMEM); } PROC_UNLOCK(imgp->proc); VOP_UNLOCK(imgp->vp, 0); /* * Destroy old process VM and create a new one (with a new stack) */ error = exec_new_vmspace(imgp, &svr4_sysvec); if (error) goto fail; vmspace = imgp->proc->p_vmspace; /* * Check if file_offset page aligned,. * Currently we cannot handle misalinged file offsets, * and so we read in the entire image (what a waste). */ if (file_offset & PAGE_MASK) { #ifdef DEBUG printf("imgact: Non page aligned binary %lu\n", file_offset); #endif /* * Map text+data+bss read/write/execute */ vmaddr = virtual_offset; error = vm_map_find(&vmspace->vm_map, NULL, 0, &vmaddr, a_out->a_text + a_out->a_data + bss_size, 0, VMFS_NO_SPACE, VM_PROT_ALL, VM_PROT_ALL, 0); if (error) goto fail; error = vn_rdwr(UIO_READ, imgp->vp, (void *)vmaddr, file_offset, a_out->a_text + a_out->a_data, UIO_USERSPACE, 0, curthread->td_ucred, NOCRED, &aresid, curthread); if (error != 0) goto fail; if (aresid != 0) { error = ENOEXEC; goto fail; } /* * remove write enable on the 'text' part */ error = vm_map_protect(&vmspace->vm_map, vmaddr, vmaddr + a_out->a_text, VM_PROT_EXECUTE|VM_PROT_READ, TRUE); if (error) goto fail; } else { #ifdef DEBUG printf("imgact: Page aligned binary %lu\n", file_offset); #endif /* * Map text+data read/execute */ vmaddr = virtual_offset; error = vm_mmap(&vmspace->vm_map, &vmaddr, a_out->a_text + a_out->a_data, VM_PROT_READ | VM_PROT_EXECUTE, VM_PROT_ALL, MAP_PRIVATE | MAP_FIXED, OBJT_VNODE, imgp->vp, file_offset); if (error) goto fail; #ifdef DEBUG printf("imgact: startaddr=%08lx, length=%08lx\n", (u_long)vmaddr, (u_long)a_out->a_text + a_out->a_data); #endif /* * allow read/write of data */ error = vm_map_protect(&vmspace->vm_map, vmaddr + a_out->a_text, vmaddr + a_out->a_text + a_out->a_data, VM_PROT_ALL, FALSE); if (error) goto fail; /* * Allocate anon demand-zeroed area for uninitialized data */ if (bss_size != 0) { vmaddr = virtual_offset + a_out->a_text + a_out->a_data; error = vm_map_find(&vmspace->vm_map, NULL, 0, &vmaddr, bss_size, 0, VMFS_NO_SPACE, VM_PROT_ALL, VM_PROT_ALL, 0); if (error) goto fail; #ifdef DEBUG printf("imgact: bssaddr=%08lx, length=%08lx\n", (u_long)vmaddr, bss_size); #endif } } /* Fill in process VM information */ vmspace->vm_tsize = round_page(a_out->a_text) >> PAGE_SHIFT; vmspace->vm_dsize = round_page(a_out->a_data + bss_size) >> PAGE_SHIFT; vmspace->vm_taddr = (caddr_t)virtual_offset; vmspace->vm_daddr = (caddr_t)virtual_offset + a_out->a_text; /* Fill in image_params */ imgp->interpreted = 0; imgp->entry_addr = a_out->a_entry; imgp->proc->p_sysent = &svr4_sysvec; fail: vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); return (error); } /* * Tell kern_execve.c about it, with a little help from the linker. */ struct execsw svr4_execsw = { exec_svr4_imgact, "svr4 ELF" }; EXEC_SET(execsw_set, svr4_execsw); Index: head/sys/compat/svr4/svr4_misc.c =================================================================== --- head/sys/compat/svr4/svr4_misc.c (revision 284214) +++ head/sys/compat/svr4/svr4_misc.c (revision 284215) @@ -1,1681 +1,1671 @@ /*- * Copyright (c) 1998 Mark Newton * Copyright (c) 1994 Christos Zoulas * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ /* * SVR4 compatibility module. * * SVR4 system calls that are implemented differently in BSD are * handled here. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include /* Must come after sys/malloc.h */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #if defined(__FreeBSD__) #include #include #endif #if defined(NetBSD) # if defined(UVM) # include # endif #endif #define BSD_DIRENT(cp) ((struct dirent *)(cp)) static int svr4_mknod(struct thread *, register_t *, char *, svr4_mode_t, svr4_dev_t); static __inline clock_t timeval_to_clock_t(struct timeval *); static int svr4_setinfo (pid_t , struct rusage *, int, svr4_siginfo_t *); struct svr4_hrtcntl_args; static int svr4_hrtcntl (struct thread *, struct svr4_hrtcntl_args *, register_t *); static void bsd_statfs_to_svr4_statvfs(const struct statfs *, struct svr4_statvfs *); static void bsd_statfs_to_svr4_statvfs64(const struct statfs *, struct svr4_statvfs64 *); static struct proc *svr4_pfind(pid_t pid); /* BOGUS noop */ #if defined(BOGUS) int svr4_sys_setitimer(td, uap) struct thread *td; struct svr4_sys_setitimer_args *uap; { td->td_retval[0] = 0; return 0; } #endif int svr4_sys_wait(td, uap) struct thread *td; struct svr4_sys_wait_args *uap; { int error, st, sig; error = kern_wait(td, WAIT_ANY, &st, 0, NULL); if (error) return (error); if (WIFSIGNALED(st)) { sig = WTERMSIG(st); if (sig >= 0 && sig < NSIG) st = (st & ~0177) | SVR4_BSD2SVR4_SIG(sig); } else if (WIFSTOPPED(st)) { sig = WSTOPSIG(st); if (sig >= 0 && sig < NSIG) st = (st & ~0xff00) | (SVR4_BSD2SVR4_SIG(sig) << 8); } /* * It looks like wait(2) on svr4/solaris/2.4 returns * the status in retval[1], and the pid on retval[0]. */ td->td_retval[1] = st; if (uap->status) error = copyout(&st, uap->status, sizeof(st)); return (error); } int svr4_sys_execv(td, uap) struct thread *td; struct svr4_sys_execv_args *uap; { struct image_args eargs; struct vmspace *oldvmspace; char *path; int error; CHECKALTEXIST(td, uap->path, &path); error = pre_execve(td, &oldvmspace); if (error != 0) { free(path, M_TEMP); return (error); } error = exec_copyin_args(&eargs, path, UIO_SYSSPACE, uap->argp, NULL); free(path, M_TEMP); if (error == 0) error = kern_execve(td, &eargs, NULL); post_execve(td, error, oldvmspace); return (error); } int svr4_sys_execve(td, uap) struct thread *td; struct svr4_sys_execve_args *uap; { struct image_args eargs; struct vmspace *oldvmspace; char *path; int error; CHECKALTEXIST(td, uap->path, &path); error = pre_execve(td, &oldvmspace); if (error != 0) { free(path, M_TEMP); return (error); } error = exec_copyin_args(&eargs, path, UIO_SYSSPACE, uap->argp, uap->envp); free(path, M_TEMP); if (error == 0) error = kern_execve(td, &eargs, NULL); post_execve(td, error, oldvmspace); return (error); } int svr4_sys_time(td, v) struct thread *td; struct svr4_sys_time_args *v; { struct svr4_sys_time_args *uap = v; int error = 0; struct timeval tv; microtime(&tv); if (uap->t) error = copyout(&tv.tv_sec, uap->t, sizeof(*(uap->t))); td->td_retval[0] = (int) tv.tv_sec; return error; } /* * Read SVR4-style directory entries. We suck them into kernel space so * that they can be massaged before being copied out to user code. * * This code is ported from the Linux emulator: Changes to the VFS interface * between FreeBSD and NetBSD have made it simpler to port it from there than * to adapt the NetBSD version. */ int svr4_sys_getdents64(td, uap) struct thread *td; struct svr4_sys_getdents64_args *uap; { struct dirent *bdp; struct vnode *vp; caddr_t inp, buf; /* BSD-format */ int len, reclen; /* BSD-format */ caddr_t outp; /* SVR4-format */ int resid, svr4reclen=0; /* SVR4-format */ cap_rights_t rights; struct file *fp; struct uio auio; struct iovec aiov; off_t off; struct svr4_dirent64 svr4_dirent; int buflen, error, eofflag, nbytes, justone; u_long *cookies = NULL, *cookiep; int ncookies; DPRINTF(("svr4_sys_getdents64(%d, *, %d)\n", uap->fd, uap->nbytes)); error = getvnode(td->td_proc->p_fd, uap->fd, cap_rights_init(&rights, CAP_READ), &fp); if (error != 0) return (error); if ((fp->f_flag & FREAD) == 0) { fdrop(fp, td); return (EBADF); } vp = fp->f_vnode; if (vp->v_type != VDIR) { fdrop(fp, td); return (EINVAL); } nbytes = uap->nbytes; if (nbytes == 1) { nbytes = sizeof (struct svr4_dirent64); justone = 1; } else justone = 0; off = fp->f_offset; #define DIRBLKSIZ 512 /* XXX we used to use ufs's DIRBLKSIZ */ buflen = max(DIRBLKSIZ, nbytes); buflen = min(buflen, MAXBSIZE); buf = malloc(buflen, M_TEMP, M_WAITOK); vn_lock(vp, LK_SHARED | LK_RETRY); again: aiov.iov_base = buf; aiov.iov_len = buflen; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_READ; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = td; auio.uio_resid = buflen; auio.uio_offset = off; if (cookies) { free(cookies, M_TEMP); cookies = NULL; } #ifdef MAC error = mac_vnode_check_readdir(td->td_ucred, vp); if (error) goto out; #endif error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, &ncookies, &cookies); if (error) { goto out; } inp = buf; outp = (caddr_t) uap->dp; resid = nbytes; if ((len = buflen - auio.uio_resid) <= 0) { goto eof; } cookiep = cookies; if (cookies) { /* * When using cookies, the vfs has the option of reading from * a different offset than that supplied (UFS truncates the * offset to a block boundary to make sure that it never reads * partway through a directory entry, even if the directory * has been compacted). */ while (len > 0 && ncookies > 0 && *cookiep <= off) { bdp = (struct dirent *) inp; len -= bdp->d_reclen; inp += bdp->d_reclen; cookiep++; ncookies--; } } while (len > 0) { if (cookiep && ncookies == 0) break; bdp = (struct dirent *) inp; reclen = bdp->d_reclen; if (reclen & 3) { DPRINTF(("svr4_readdir: reclen=%d\n", reclen)); error = EFAULT; goto out; } if (bdp->d_fileno == 0) { inp += reclen; if (cookiep) { off = *cookiep++; ncookies--; } else off += reclen; len -= reclen; continue; } svr4reclen = SVR4_RECLEN(&svr4_dirent, bdp->d_namlen); if (reclen > len || resid < svr4reclen) { outp++; break; } svr4_dirent.d_ino = (long) bdp->d_fileno; if (justone) { /* * old svr4-style readdir usage. */ svr4_dirent.d_off = (svr4_off_t) svr4reclen; svr4_dirent.d_reclen = (u_short) bdp->d_namlen; } else { svr4_dirent.d_off = (svr4_off_t)(off + reclen); svr4_dirent.d_reclen = (u_short) svr4reclen; } strlcpy(svr4_dirent.d_name, bdp->d_name, sizeof(svr4_dirent.d_name)); if ((error = copyout((caddr_t)&svr4_dirent, outp, svr4reclen))) goto out; inp += reclen; if (cookiep) { off = *cookiep++; ncookies--; } else off += reclen; outp += svr4reclen; resid -= svr4reclen; len -= reclen; if (justone) break; } if (outp == (caddr_t) uap->dp) goto again; fp->f_offset = off; if (justone) nbytes = resid + svr4reclen; eof: td->td_retval[0] = nbytes - resid; out: VOP_UNLOCK(vp, 0); fdrop(fp, td); if (cookies) free(cookies, M_TEMP); free(buf, M_TEMP); return error; } int svr4_sys_getdents(td, uap) struct thread *td; struct svr4_sys_getdents_args *uap; { struct dirent *bdp; struct vnode *vp; caddr_t inp, buf; /* BSD-format */ int len, reclen; /* BSD-format */ caddr_t outp; /* SVR4-format */ int resid, svr4_reclen; /* SVR4-format */ cap_rights_t rights; struct file *fp; struct uio auio; struct iovec aiov; struct svr4_dirent idb; off_t off; /* true file offset */ int buflen, error, eofflag; u_long *cookiebuf = NULL, *cookie; int ncookies = 0, *retval = td->td_retval; if (uap->nbytes < 0) return (EINVAL); error = getvnode(td->td_proc->p_fd, uap->fd, cap_rights_init(&rights, CAP_READ), &fp); if (error != 0) return (error); if ((fp->f_flag & FREAD) == 0) { fdrop(fp, td); return (EBADF); } vp = fp->f_vnode; if (vp->v_type != VDIR) { fdrop(fp, td); return (EINVAL); } buflen = min(MAXBSIZE, uap->nbytes); buf = malloc(buflen, M_TEMP, M_WAITOK); vn_lock(vp, LK_SHARED | LK_RETRY); off = fp->f_offset; again: aiov.iov_base = buf; aiov.iov_len = buflen; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_READ; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = td; auio.uio_resid = buflen; auio.uio_offset = off; #ifdef MAC error = mac_vnode_check_readdir(td->td_ucred, vp); if (error) goto out; #endif /* * First we read into the malloc'ed buffer, then * we massage it into user space, one record at a time. */ error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, &ncookies, &cookiebuf); if (error) { goto out; } inp = buf; outp = uap->buf; resid = uap->nbytes; if ((len = buflen - auio.uio_resid) == 0) goto eof; for (cookie = cookiebuf; len > 0; len -= reclen) { bdp = (struct dirent *)inp; reclen = bdp->d_reclen; if (reclen & 3) panic("svr4_sys_getdents64: bad reclen"); if (cookie) off = *cookie++; /* each entry points to the next */ else off += reclen; if ((off >> 32) != 0) { uprintf("svr4_sys_getdents64: dir offset too large for emulated program"); error = EINVAL; goto out; } if (bdp->d_fileno == 0) { inp += reclen; /* it is a hole; squish it out */ continue; } svr4_reclen = SVR4_RECLEN(&idb, bdp->d_namlen); if (reclen > len || resid < svr4_reclen) { /* entry too big for buffer, so just stop */ outp++; break; } /* * Massage in place to make a SVR4-shaped dirent (otherwise * we have to worry about touching user memory outside of * the copyout() call). */ idb.d_ino = (svr4_ino_t)bdp->d_fileno; idb.d_off = (svr4_off_t)off; idb.d_reclen = (u_short)svr4_reclen; strlcpy(idb.d_name, bdp->d_name, sizeof(idb.d_name)); if ((error = copyout((caddr_t)&idb, outp, svr4_reclen))) goto out; /* advance past this real entry */ inp += reclen; /* advance output past SVR4-shaped entry */ outp += svr4_reclen; resid -= svr4_reclen; } /* if we squished out the whole block, try again */ if (outp == uap->buf) goto again; fp->f_offset = off; /* update the vnode offset */ eof: *retval = uap->nbytes - resid; out: VOP_UNLOCK(vp, 0); fdrop(fp, td); if (cookiebuf) free(cookiebuf, M_TEMP); free(buf, M_TEMP); return error; } int svr4_sys_mmap(td, uap) struct thread *td; struct svr4_sys_mmap_args *uap; { struct mmap_args mm; int *retval; retval = td->td_retval; #define _MAP_NEW 0x80000000 /* * Verify the arguments. */ if (uap->prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) return EINVAL; /* XXX still needed? */ if (uap->len == 0) return EINVAL; mm.prot = uap->prot; mm.len = uap->len; mm.flags = uap->flags & ~_MAP_NEW; mm.fd = uap->fd; mm.addr = uap->addr; mm.pos = uap->pos; return sys_mmap(td, &mm); } int svr4_sys_mmap64(td, uap) struct thread *td; struct svr4_sys_mmap64_args *uap; { struct mmap_args mm; void *rp; #define _MAP_NEW 0x80000000 /* * Verify the arguments. */ if (uap->prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) return EINVAL; /* XXX still needed? */ if (uap->len == 0) return EINVAL; mm.prot = uap->prot; mm.len = uap->len; mm.flags = uap->flags & ~_MAP_NEW; mm.fd = uap->fd; mm.addr = uap->addr; mm.pos = uap->pos; rp = (void *) round_page((vm_offset_t)(td->td_proc->p_vmspace->vm_daddr + maxdsiz)); if ((mm.flags & MAP_FIXED) == 0 && mm.addr != 0 && (void *)mm.addr < rp) mm.addr = rp; return sys_mmap(td, &mm); } int svr4_sys_fchroot(td, uap) struct thread *td; struct svr4_sys_fchroot_args *uap; { struct filedesc *fdp = td->td_proc->p_fd; struct vnode *vp; struct file *fp; int error; if ((error = priv_check(td, PRIV_VFS_FCHROOT)) != 0) return error; /* XXX: we have the chroot priv... what cap might we need? all? */ if ((error = getvnode(fdp, uap->fd, 0, &fp)) != 0) return error; vp = fp->f_vnode; VREF(vp); fdrop(fp, td); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); error = change_dir(vp, td); if (error) goto fail; #ifdef MAC error = mac_vnode_check_chroot(td->td_ucred, vp); if (error) goto fail; #endif VOP_UNLOCK(vp, 0); error = change_root(vp, td); vrele(vp); return (error); fail: vput(vp); return (error); } static int svr4_mknod(td, retval, path, mode, dev) struct thread *td; register_t *retval; char *path; svr4_mode_t mode; svr4_dev_t dev; { char *newpath; int error; CHECKALTEXIST(td, path, &newpath); if (S_ISFIFO(mode)) { error = kern_mkfifoat(td, AT_FDCWD, newpath, UIO_SYSSPACE, mode); } else { error = kern_mknodat(td, AT_FDCWD, newpath, UIO_SYSSPACE, mode, dev); } free(newpath, M_TEMP); return (error); } int svr4_sys_mknod(td, uap) struct thread *td; struct svr4_sys_mknod_args *uap; { int *retval = td->td_retval; return svr4_mknod(td, retval, uap->path, uap->mode, (svr4_dev_t)svr4_to_bsd_odev_t(uap->dev)); } int svr4_sys_xmknod(td, uap) struct thread *td; struct svr4_sys_xmknod_args *uap; { int *retval = td->td_retval; return svr4_mknod(td, retval, uap->path, uap->mode, (svr4_dev_t)svr4_to_bsd_dev_t(uap->dev)); } int svr4_sys_vhangup(td, uap) struct thread *td; struct svr4_sys_vhangup_args *uap; { return 0; } int svr4_sys_sysconfig(td, uap) struct thread *td; struct svr4_sys_sysconfig_args *uap; { int *retval; retval = &(td->td_retval[0]); switch (uap->name) { case SVR4_CONFIG_NGROUPS: *retval = ngroups_max; break; case SVR4_CONFIG_CHILD_MAX: *retval = maxproc; break; case SVR4_CONFIG_OPEN_FILES: *retval = maxfiles; break; case SVR4_CONFIG_POSIX_VER: *retval = 198808; break; case SVR4_CONFIG_PAGESIZE: *retval = PAGE_SIZE; break; case SVR4_CONFIG_CLK_TCK: *retval = 60; /* should this be `hz', ie. 100? */ break; case SVR4_CONFIG_XOPEN_VER: *retval = 2; /* XXX: What should that be? */ break; case SVR4_CONFIG_PROF_TCK: *retval = 60; /* XXX: What should that be? */ break; case SVR4_CONFIG_NPROC_CONF: *retval = 1; /* Only one processor for now */ break; case SVR4_CONFIG_NPROC_ONLN: *retval = 1; /* And it better be online */ break; case SVR4_CONFIG_AIO_LISTIO_MAX: case SVR4_CONFIG_AIO_MAX: case SVR4_CONFIG_AIO_PRIO_DELTA_MAX: *retval = 0; /* No aio support */ break; case SVR4_CONFIG_DELAYTIMER_MAX: *retval = 0; /* No delaytimer support */ break; case SVR4_CONFIG_MQ_OPEN_MAX: *retval = msginfo.msgmni; break; case SVR4_CONFIG_MQ_PRIO_MAX: *retval = 0; /* XXX: Don't know */ break; case SVR4_CONFIG_RTSIG_MAX: *retval = 0; break; case SVR4_CONFIG_SEM_NSEMS_MAX: *retval = seminfo.semmni; break; case SVR4_CONFIG_SEM_VALUE_MAX: *retval = seminfo.semvmx; break; case SVR4_CONFIG_SIGQUEUE_MAX: *retval = 0; /* XXX: Don't know */ break; case SVR4_CONFIG_SIGRT_MIN: case SVR4_CONFIG_SIGRT_MAX: *retval = 0; /* No real time signals */ break; case SVR4_CONFIG_TIMER_MAX: *retval = 3; /* XXX: real, virtual, profiling */ break; #if defined(NOTYET) case SVR4_CONFIG_PHYS_PAGES: #if defined(UVM) *retval = uvmexp.free; /* XXX: free instead of total */ #else *retval = vm_cnt.v_free_count; /* XXX: free instead of total */ #endif break; case SVR4_CONFIG_AVPHYS_PAGES: #if defined(UVM) *retval = uvmexp.active; /* XXX: active instead of avg */ #else *retval = vm_cnt.v_active_count;/* XXX: active instead of avg */ #endif break; #endif /* NOTYET */ case SVR4_CONFIG_COHERENCY: *retval = 0; /* XXX */ break; case SVR4_CONFIG_SPLIT_CACHE: *retval = 0; /* XXX */ break; case SVR4_CONFIG_ICACHESZ: *retval = 256; /* XXX */ break; case SVR4_CONFIG_DCACHESZ: *retval = 256; /* XXX */ break; case SVR4_CONFIG_ICACHELINESZ: *retval = 64; /* XXX */ break; case SVR4_CONFIG_DCACHELINESZ: *retval = 64; /* XXX */ break; case SVR4_CONFIG_ICACHEBLKSZ: *retval = 64; /* XXX */ break; case SVR4_CONFIG_DCACHEBLKSZ: *retval = 64; /* XXX */ break; case SVR4_CONFIG_DCACHETBLKSZ: *retval = 64; /* XXX */ break; case SVR4_CONFIG_ICACHE_ASSOC: *retval = 1; /* XXX */ break; case SVR4_CONFIG_DCACHE_ASSOC: *retval = 1; /* XXX */ break; case SVR4_CONFIG_MAXPID: *retval = PID_MAX; break; case SVR4_CONFIG_STACK_PROT: *retval = PROT_READ|PROT_WRITE|PROT_EXEC; break; default: return EINVAL; } return 0; } /* ARGSUSED */ int svr4_sys_break(td, uap) struct thread *td; struct svr4_sys_break_args *uap; { struct obreak_args ap; ap.nsize = uap->nsize; return (sys_obreak(td, &ap)); } static __inline clock_t timeval_to_clock_t(tv) struct timeval *tv; { return tv->tv_sec * hz + tv->tv_usec / (1000000 / hz); } int svr4_sys_times(td, uap) struct thread *td; struct svr4_sys_times_args *uap; { struct timeval tv, utime, stime, cutime, cstime; struct tms tms; struct proc *p; int error; p = td->td_proc; PROC_LOCK(p); PROC_STATLOCK(p); calcru(p, &utime, &stime); PROC_STATUNLOCK(p); calccru(p, &cutime, &cstime); PROC_UNLOCK(p); tms.tms_utime = timeval_to_clock_t(&utime); tms.tms_stime = timeval_to_clock_t(&stime); tms.tms_cutime = timeval_to_clock_t(&cutime); tms.tms_cstime = timeval_to_clock_t(&cstime); error = copyout(&tms, uap->tp, sizeof(tms)); if (error) return (error); microtime(&tv); td->td_retval[0] = (int)timeval_to_clock_t(&tv); return (0); } int svr4_sys_ulimit(td, uap) struct thread *td; struct svr4_sys_ulimit_args *uap; { int *retval = td->td_retval; int error; switch (uap->cmd) { case SVR4_GFILLIM: - PROC_LOCK(td->td_proc); - *retval = lim_cur(td->td_proc, RLIMIT_FSIZE) / 512; - PROC_UNLOCK(td->td_proc); + *retval = lim_cur(td, RLIMIT_FSIZE) / 512; if (*retval == -1) *retval = 0x7fffffff; return 0; case SVR4_SFILLIM: { struct rlimit krl; krl.rlim_cur = uap->newlimit * 512; - PROC_LOCK(td->td_proc); - krl.rlim_max = lim_max(td->td_proc, RLIMIT_FSIZE); - PROC_UNLOCK(td->td_proc); + krl.rlim_max = lim_max(td, RLIMIT_FSIZE); error = kern_setrlimit(td, RLIMIT_FSIZE, &krl); if (error) return error; - PROC_LOCK(td->td_proc); - *retval = lim_cur(td->td_proc, RLIMIT_FSIZE); - PROC_UNLOCK(td->td_proc); + *retval = lim_cur(td, RLIMIT_FSIZE); if (*retval == -1) *retval = 0x7fffffff; return 0; } case SVR4_GMEMLIM: { struct vmspace *vm = td->td_proc->p_vmspace; register_t r; - PROC_LOCK(td->td_proc); - r = lim_cur(td->td_proc, RLIMIT_DATA); - PROC_UNLOCK(td->td_proc); + r = lim_cur(td, RLIMIT_DATA); if (r == -1) r = 0x7fffffff; r += (long) vm->vm_daddr; if (r < 0) r = 0x7fffffff; *retval = r; return 0; } case SVR4_GDESLIM: - PROC_LOCK(td->td_proc); - *retval = lim_cur(td->td_proc, RLIMIT_NOFILE); - PROC_UNLOCK(td->td_proc); + *retval = lim_cur(td, RLIMIT_NOFILE); if (*retval == -1) *retval = 0x7fffffff; return 0; default: return EINVAL; } } static struct proc * svr4_pfind(pid) pid_t pid; { struct proc *p; /* look in the live processes */ if ((p = pfind(pid)) == NULL) /* look in the zombies */ p = zpfind(pid); return p; } int svr4_sys_pgrpsys(td, uap) struct thread *td; struct svr4_sys_pgrpsys_args *uap; { int *retval = td->td_retval; struct proc *p = td->td_proc; switch (uap->cmd) { case 1: /* setpgrp() */ /* * SVR4 setpgrp() (which takes no arguments) has the * semantics that the session ID is also created anew, so * in almost every sense, setpgrp() is identical to * setsid() for SVR4. (Under BSD, the difference is that * a setpgid(0,0) will not create a new session.) */ sys_setsid(td, NULL); /*FALLTHROUGH*/ case 0: /* getpgrp() */ PROC_LOCK(p); *retval = p->p_pgrp->pg_id; PROC_UNLOCK(p); return 0; case 2: /* getsid(pid) */ if (uap->pid == 0) PROC_LOCK(p); else if ((p = svr4_pfind(uap->pid)) == NULL) return ESRCH; /* * This has already been initialized to the pid of * the session leader. */ *retval = (register_t) p->p_session->s_sid; PROC_UNLOCK(p); return 0; case 3: /* setsid() */ return sys_setsid(td, NULL); case 4: /* getpgid(pid) */ if (uap->pid == 0) PROC_LOCK(p); else if ((p = svr4_pfind(uap->pid)) == NULL) return ESRCH; *retval = (int) p->p_pgrp->pg_id; PROC_UNLOCK(p); return 0; case 5: /* setpgid(pid, pgid); */ { struct setpgid_args sa; sa.pid = uap->pid; sa.pgid = uap->pgid; return sys_setpgid(td, &sa); } default: return EINVAL; } } struct svr4_hrtcntl_args { int cmd; int fun; int clk; svr4_hrt_interval_t * iv; svr4_hrt_time_t * ti; }; static int svr4_hrtcntl(td, uap, retval) struct thread *td; struct svr4_hrtcntl_args *uap; register_t *retval; { switch (uap->fun) { case SVR4_HRT_CNTL_RES: DPRINTF(("htrcntl(RES)\n")); *retval = SVR4_HRT_USEC; return 0; case SVR4_HRT_CNTL_TOFD: DPRINTF(("htrcntl(TOFD)\n")); { struct timeval tv; svr4_hrt_time_t t; if (uap->clk != SVR4_HRT_CLK_STD) { DPRINTF(("clk == %d\n", uap->clk)); return EINVAL; } if (uap->ti == NULL) { DPRINTF(("ti NULL\n")); return EINVAL; } microtime(&tv); t.h_sec = tv.tv_sec; t.h_rem = tv.tv_usec; t.h_res = SVR4_HRT_USEC; return copyout(&t, uap->ti, sizeof(t)); } case SVR4_HRT_CNTL_START: DPRINTF(("htrcntl(START)\n")); return ENOSYS; case SVR4_HRT_CNTL_GET: DPRINTF(("htrcntl(GET)\n")); return ENOSYS; default: DPRINTF(("Bad htrcntl command %d\n", uap->fun)); return ENOSYS; } } int svr4_sys_hrtsys(td, uap) struct thread *td; struct svr4_sys_hrtsys_args *uap; { int *retval = td->td_retval; switch (uap->cmd) { case SVR4_HRT_CNTL: return svr4_hrtcntl(td, (struct svr4_hrtcntl_args *) uap, retval); case SVR4_HRT_ALRM: DPRINTF(("hrtalarm\n")); return ENOSYS; case SVR4_HRT_SLP: DPRINTF(("hrtsleep\n")); return ENOSYS; case SVR4_HRT_CAN: DPRINTF(("hrtcancel\n")); return ENOSYS; default: DPRINTF(("Bad hrtsys command %d\n", uap->cmd)); return EINVAL; } } static int svr4_setinfo(pid, ru, st, s) pid_t pid; struct rusage *ru; int st; svr4_siginfo_t *s; { svr4_siginfo_t i; int sig; memset(&i, 0, sizeof(i)); i.svr4_si_signo = SVR4_SIGCHLD; i.svr4_si_errno = 0; /* XXX? */ i.svr4_si_pid = pid; if (ru) { i.svr4_si_stime = ru->ru_stime.tv_sec; i.svr4_si_utime = ru->ru_utime.tv_sec; } if (WIFEXITED(st)) { i.svr4_si_status = WEXITSTATUS(st); i.svr4_si_code = SVR4_CLD_EXITED; } else if (WIFSTOPPED(st)) { sig = WSTOPSIG(st); if (sig >= 0 && sig < NSIG) i.svr4_si_status = SVR4_BSD2SVR4_SIG(sig); if (i.svr4_si_status == SVR4_SIGCONT) i.svr4_si_code = SVR4_CLD_CONTINUED; else i.svr4_si_code = SVR4_CLD_STOPPED; } else { sig = WTERMSIG(st); if (sig >= 0 && sig < NSIG) i.svr4_si_status = SVR4_BSD2SVR4_SIG(sig); if (WCOREDUMP(st)) i.svr4_si_code = SVR4_CLD_DUMPED; else i.svr4_si_code = SVR4_CLD_KILLED; } DPRINTF(("siginfo [pid %ld signo %d code %d errno %d status %d]\n", i.svr4_si_pid, i.svr4_si_signo, i.svr4_si_code, i.svr4_si_errno, i.svr4_si_status)); return copyout(&i, s, sizeof(i)); } int svr4_sys_waitsys(td, uap) struct thread *td; struct svr4_sys_waitsys_args *uap; { struct rusage ru; pid_t pid; int nfound, status; int error, *retval = td->td_retval; struct proc *p, *q; DPRINTF(("waitsys(%d, %d, %p, %x)\n", uap->grp, uap->id, uap->info, uap->options)); q = td->td_proc; switch (uap->grp) { case SVR4_P_PID: pid = uap->id; break; case SVR4_P_PGID: PROC_LOCK(q); pid = -q->p_pgid; PROC_UNLOCK(q); break; case SVR4_P_ALL: pid = WAIT_ANY; break; default: return EINVAL; } /* Hand off the easy cases to kern_wait(). */ if (!(uap->options & (SVR4_WNOWAIT)) && (uap->options & (SVR4_WEXITED | SVR4_WTRAPPED))) { int options; options = 0; if (uap->options & SVR4_WSTOPPED) options |= WUNTRACED; if (uap->options & SVR4_WCONTINUED) options |= WCONTINUED; if (uap->options & SVR4_WNOHANG) options |= WNOHANG; error = kern_wait(td, pid, &status, options, &ru); if (error) return (error); if (uap->options & SVR4_WNOHANG && *retval == 0) error = svr4_setinfo(*retval, NULL, 0, uap->info); else error = svr4_setinfo(*retval, &ru, status, uap->info); *retval = 0; return (error); } /* * Ok, handle the weird cases. Either WNOWAIT is set (meaning we * just want to see if there is a process to harvest, we don't * want to actually harvest it), or WEXIT and WTRAPPED are clear * meaning we want to ignore zombies. Either way, we don't have * to handle harvesting zombies here. We do have to duplicate the * other portions of kern_wait() though, especially for WCONTINUED * and WSTOPPED. */ loop: nfound = 0; sx_slock(&proctree_lock); LIST_FOREACH(p, &q->p_children, p_sibling) { PROC_LOCK(p); if (pid != WAIT_ANY && p->p_pid != pid && p->p_pgid != -pid) { PROC_UNLOCK(p); DPRINTF(("pid %d pgid %d != %d\n", p->p_pid, p->p_pgid, pid)); continue; } if (p_canwait(td, p)) { PROC_UNLOCK(p); continue; } nfound++; PROC_SLOCK(p); /* * See if we have a zombie. If so, WNOWAIT should be set, * as otherwise we should have called kern_wait() up above. */ if ((p->p_state == PRS_ZOMBIE) && ((uap->options & (SVR4_WEXITED|SVR4_WTRAPPED)))) { PROC_SUNLOCK(p); KASSERT(uap->options & SVR4_WNOWAIT, ("WNOWAIT is clear")); /* Found a zombie, so cache info in local variables. */ pid = p->p_pid; status = p->p_xstat; ru = p->p_ru; PROC_STATLOCK(p); calcru(p, &ru.ru_utime, &ru.ru_stime); PROC_STATUNLOCK(p); PROC_UNLOCK(p); sx_sunlock(&proctree_lock); /* Copy the info out to userland. */ *retval = 0; DPRINTF(("found %d\n", pid)); return (svr4_setinfo(pid, &ru, status, uap->info)); } /* * See if we have a stopped or continued process. * XXX: This duplicates the same code in kern_wait(). */ if ((p->p_flag & P_STOPPED_SIG) && (p->p_suspcount == p->p_numthreads) && (p->p_flag & P_WAITED) == 0 && (p->p_flag & P_TRACED || uap->options & SVR4_WSTOPPED)) { PROC_SUNLOCK(p); if (((uap->options & SVR4_WNOWAIT)) == 0) p->p_flag |= P_WAITED; sx_sunlock(&proctree_lock); pid = p->p_pid; status = W_STOPCODE(p->p_xstat); ru = p->p_ru; PROC_STATLOCK(p); calcru(p, &ru.ru_utime, &ru.ru_stime); PROC_STATUNLOCK(p); PROC_UNLOCK(p); if (((uap->options & SVR4_WNOWAIT)) == 0) { PROC_LOCK(q); sigqueue_take(p->p_ksi); PROC_UNLOCK(q); } *retval = 0; DPRINTF(("jobcontrol %d\n", pid)); return (svr4_setinfo(pid, &ru, status, uap->info)); } PROC_SUNLOCK(p); if (uap->options & SVR4_WCONTINUED && (p->p_flag & P_CONTINUED)) { sx_sunlock(&proctree_lock); if (((uap->options & SVR4_WNOWAIT)) == 0) p->p_flag &= ~P_CONTINUED; pid = p->p_pid; ru = p->p_ru; status = SIGCONT; PROC_STATLOCK(p); calcru(p, &ru.ru_utime, &ru.ru_stime); PROC_STATUNLOCK(p); PROC_UNLOCK(p); if (((uap->options & SVR4_WNOWAIT)) == 0) { PROC_LOCK(q); sigqueue_take(p->p_ksi); PROC_UNLOCK(q); } *retval = 0; DPRINTF(("jobcontrol %d\n", pid)); return (svr4_setinfo(pid, &ru, status, uap->info)); } PROC_UNLOCK(p); } if (nfound == 0) { sx_sunlock(&proctree_lock); return (ECHILD); } if (uap->options & SVR4_WNOHANG) { sx_sunlock(&proctree_lock); *retval = 0; return (svr4_setinfo(0, NULL, 0, uap->info)); } PROC_LOCK(q); sx_sunlock(&proctree_lock); if (q->p_flag & P_STATCHILD) { q->p_flag &= ~P_STATCHILD; error = 0; } else error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "svr4_wait", 0); PROC_UNLOCK(q); if (error) return error; goto loop; } static void bsd_statfs_to_svr4_statvfs(bfs, sfs) const struct statfs *bfs; struct svr4_statvfs *sfs; { sfs->f_bsize = bfs->f_iosize; /* XXX */ sfs->f_frsize = bfs->f_bsize; sfs->f_blocks = bfs->f_blocks; sfs->f_bfree = bfs->f_bfree; sfs->f_bavail = bfs->f_bavail; sfs->f_files = bfs->f_files; sfs->f_ffree = bfs->f_ffree; sfs->f_favail = bfs->f_ffree; sfs->f_fsid = bfs->f_fsid.val[0]; memcpy(sfs->f_basetype, bfs->f_fstypename, sizeof(sfs->f_basetype)); sfs->f_flag = 0; if (bfs->f_flags & MNT_RDONLY) sfs->f_flag |= SVR4_ST_RDONLY; if (bfs->f_flags & MNT_NOSUID) sfs->f_flag |= SVR4_ST_NOSUID; sfs->f_namemax = MAXNAMLEN; memcpy(sfs->f_fstr, bfs->f_fstypename, sizeof(sfs->f_fstr)); /* XXX */ memset(sfs->f_filler, 0, sizeof(sfs->f_filler)); } static void bsd_statfs_to_svr4_statvfs64(bfs, sfs) const struct statfs *bfs; struct svr4_statvfs64 *sfs; { sfs->f_bsize = bfs->f_iosize; /* XXX */ sfs->f_frsize = bfs->f_bsize; sfs->f_blocks = bfs->f_blocks; sfs->f_bfree = bfs->f_bfree; sfs->f_bavail = bfs->f_bavail; sfs->f_files = bfs->f_files; sfs->f_ffree = bfs->f_ffree; sfs->f_favail = bfs->f_ffree; sfs->f_fsid = bfs->f_fsid.val[0]; memcpy(sfs->f_basetype, bfs->f_fstypename, sizeof(sfs->f_basetype)); sfs->f_flag = 0; if (bfs->f_flags & MNT_RDONLY) sfs->f_flag |= SVR4_ST_RDONLY; if (bfs->f_flags & MNT_NOSUID) sfs->f_flag |= SVR4_ST_NOSUID; sfs->f_namemax = MAXNAMLEN; memcpy(sfs->f_fstr, bfs->f_fstypename, sizeof(sfs->f_fstr)); /* XXX */ memset(sfs->f_filler, 0, sizeof(sfs->f_filler)); } int svr4_sys_statvfs(td, uap) struct thread *td; struct svr4_sys_statvfs_args *uap; { struct svr4_statvfs sfs; struct statfs bfs; char *path; int error; CHECKALTEXIST(td, uap->path, &path); error = kern_statfs(td, path, UIO_SYSSPACE, &bfs); free(path, M_TEMP); if (error) return (error); bsd_statfs_to_svr4_statvfs(&bfs, &sfs); return copyout(&sfs, uap->fs, sizeof(sfs)); } int svr4_sys_fstatvfs(td, uap) struct thread *td; struct svr4_sys_fstatvfs_args *uap; { struct svr4_statvfs sfs; struct statfs bfs; int error; error = kern_fstatfs(td, uap->fd, &bfs); if (error) return (error); bsd_statfs_to_svr4_statvfs(&bfs, &sfs); return copyout(&sfs, uap->fs, sizeof(sfs)); } int svr4_sys_statvfs64(td, uap) struct thread *td; struct svr4_sys_statvfs64_args *uap; { struct svr4_statvfs64 sfs; struct statfs bfs; char *path; int error; CHECKALTEXIST(td, uap->path, &path); error = kern_statfs(td, path, UIO_SYSSPACE, &bfs); free(path, M_TEMP); if (error) return (error); bsd_statfs_to_svr4_statvfs64(&bfs, &sfs); return copyout(&sfs, uap->fs, sizeof(sfs)); } int svr4_sys_fstatvfs64(td, uap) struct thread *td; struct svr4_sys_fstatvfs64_args *uap; { struct svr4_statvfs64 sfs; struct statfs bfs; int error; error = kern_fstatfs(td, uap->fd, &bfs); if (error) return (error); bsd_statfs_to_svr4_statvfs64(&bfs, &sfs); return copyout(&sfs, uap->fs, sizeof(sfs)); } int svr4_sys_alarm(td, uap) struct thread *td; struct svr4_sys_alarm_args *uap; { struct itimerval itv, oitv; int error; timevalclear(&itv.it_interval); itv.it_value.tv_sec = uap->sec; itv.it_value.tv_usec = 0; error = kern_setitimer(td, ITIMER_REAL, &itv, &oitv); if (error) return (error); if (oitv.it_value.tv_usec != 0) oitv.it_value.tv_sec++; td->td_retval[0] = oitv.it_value.tv_sec; return (0); } int svr4_sys_gettimeofday(td, uap) struct thread *td; struct svr4_sys_gettimeofday_args *uap; { if (uap->tp) { struct timeval atv; microtime(&atv); return copyout(&atv, uap->tp, sizeof (atv)); } return 0; } int svr4_sys_facl(td, uap) struct thread *td; struct svr4_sys_facl_args *uap; { int *retval; retval = td->td_retval; *retval = 0; switch (uap->cmd) { case SVR4_SYS_SETACL: /* We don't support acls on any filesystem */ return ENOSYS; case SVR4_SYS_GETACL: return copyout(retval, &uap->num, sizeof(uap->num)); case SVR4_SYS_GETACLCNT: return 0; default: return EINVAL; } } int svr4_sys_acl(td, uap) struct thread *td; struct svr4_sys_acl_args *uap; { /* XXX: for now the same */ return svr4_sys_facl(td, (struct svr4_sys_facl_args *)uap); } int svr4_sys_auditsys(td, uap) struct thread *td; struct svr4_sys_auditsys_args *uap; { /* * XXX: Big brother is *not* watching. */ return 0; } int svr4_sys_memcntl(td, uap) struct thread *td; struct svr4_sys_memcntl_args *uap; { switch (uap->cmd) { case SVR4_MC_SYNC: { struct msync_args msa; msa.addr = uap->addr; msa.len = uap->len; msa.flags = (int)uap->arg; return sys_msync(td, &msa); } case SVR4_MC_ADVISE: { struct madvise_args maa; maa.addr = uap->addr; maa.len = uap->len; maa.behav = (int)uap->arg; return sys_madvise(td, &maa); } case SVR4_MC_LOCK: case SVR4_MC_UNLOCK: case SVR4_MC_LOCKAS: case SVR4_MC_UNLOCKAS: return EOPNOTSUPP; default: return ENOSYS; } } int svr4_sys_nice(td, uap) struct thread *td; struct svr4_sys_nice_args *uap; { struct setpriority_args ap; int error; ap.which = PRIO_PROCESS; ap.who = 0; ap.prio = uap->prio; if ((error = sys_setpriority(td, &ap)) != 0) return error; /* the cast is stupid, but the structures are the same */ if ((error = sys_getpriority(td, (struct getpriority_args *)&ap)) != 0) return error; return 0; } int svr4_sys_resolvepath(td, uap) struct thread *td; struct svr4_sys_resolvepath_args *uap; { struct nameidata nd; int error, *retval = td->td_retval; unsigned int ncopy; NDINIT(&nd, LOOKUP, NOFOLLOW | SAVENAME, UIO_USERSPACE, uap->path, td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_NO_FREE_PNBUF); ncopy = min(uap->bufsiz, strlen(nd.ni_cnd.cn_pnbuf) + 1); if ((error = copyout(nd.ni_cnd.cn_pnbuf, uap->buf, ncopy)) != 0) goto bad; *retval = ncopy; bad: NDFREE(&nd, NDF_ONLY_PNBUF); return error; } Index: head/sys/compat/svr4/svr4_resource.c =================================================================== --- head/sys/compat/svr4/svr4_resource.c (revision 284214) +++ head/sys/compat/svr4/svr4_resource.c (revision 284215) @@ -1,314 +1,306 @@ /*- * Copyright (c) 1998 The NetBSD Foundation, Inc. * All rights reserved. * * This code is derived from software contributed to The NetBSD Foundation * by Christos Zoulas. * * 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 NETBSD FOUNDATION, INC. 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 FOUNDATION 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. */ /*- * Portions of this software have been derived from software contributed * to the FreeBSD Project by Mark Newton. * * Copyright (c) 1999 Mark Newton * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * Derived from: $NetBSD: svr4_resource.c,v 1.3 1998/12/13 18:00:52 christos Exp $ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static __inline int svr4_to_native_rl(int); static __inline int svr4_to_native_rl(rl) int rl; { switch (rl) { case SVR4_RLIMIT_CPU: return RLIMIT_CPU; case SVR4_RLIMIT_FSIZE: return RLIMIT_FSIZE; case SVR4_RLIMIT_DATA: return RLIMIT_DATA; case SVR4_RLIMIT_STACK: return RLIMIT_STACK; case SVR4_RLIMIT_CORE: return RLIMIT_CORE; case SVR4_RLIMIT_NOFILE: return RLIMIT_NOFILE; case SVR4_RLIMIT_VMEM: return RLIMIT_VMEM; default: return -1; } } /* * Check if the resource limit fits within the BSD range and it is not * one of the magic SVR4 limit values */ #define OKLIMIT(l) (((int32_t)(l)) >= 0 && ((int32_t)(l)) < 0x7fffffff && \ ((svr4_rlim_t)(l)) != SVR4_RLIM_INFINITY && \ ((svr4_rlim_t)(l)) != SVR4_RLIM_SAVED_CUR && \ ((svr4_rlim_t)(l)) != SVR4_RLIM_SAVED_MAX) #define OKLIMIT64(l) (((rlim_t)(l)) >= 0 && ((rlim_t)(l)) < RLIM_INFINITY && \ ((svr4_rlim64_t)(l)) != SVR4_RLIM64_INFINITY && \ ((svr4_rlim64_t)(l)) != SVR4_RLIM64_SAVED_CUR && \ ((svr4_rlim64_t)(l)) != SVR4_RLIM64_SAVED_MAX) int svr4_sys_getrlimit(td, uap) struct thread *td; struct svr4_sys_getrlimit_args *uap; { int rl = svr4_to_native_rl(uap->which); struct rlimit blim; struct svr4_rlimit slim; if (rl == -1) return EINVAL; - PROC_LOCK(td->td_proc); - lim_rlimit(td->td_proc, rl, &blim); - PROC_UNLOCK(td->td_proc); + lim_rlimit(td, rl, &blim); /* * Our infinity, is their maxfiles. */ if (rl == RLIMIT_NOFILE && blim.rlim_max == RLIM_INFINITY) blim.rlim_max = maxfiles; /* * If the limit can be be represented, it is returned. * Otherwise, if rlim_cur == rlim_max, return RLIM_SAVED_MAX * else return RLIM_SAVED_CUR */ if (blim.rlim_max == RLIM_INFINITY) slim.rlim_max = SVR4_RLIM_INFINITY; else if (OKLIMIT(blim.rlim_max)) slim.rlim_max = (svr4_rlim_t) blim.rlim_max; else slim.rlim_max = SVR4_RLIM_SAVED_MAX; if (blim.rlim_cur == RLIM_INFINITY) slim.rlim_cur = SVR4_RLIM_INFINITY; else if (OKLIMIT(blim.rlim_cur)) slim.rlim_cur = (svr4_rlim_t) blim.rlim_cur; else if (blim.rlim_max == blim.rlim_cur) slim.rlim_cur = SVR4_RLIM_SAVED_MAX; else slim.rlim_cur = SVR4_RLIM_SAVED_CUR; return copyout(&slim, uap->rlp, sizeof(*uap->rlp)); } int svr4_sys_setrlimit(td, uap) struct thread *td; struct svr4_sys_setrlimit_args *uap; { int rl = svr4_to_native_rl(uap->which); struct rlimit blim, curlim; struct svr4_rlimit slim; int error; if (rl == -1) return EINVAL; if ((error = copyin(uap->rlp, &slim, sizeof(slim))) != 0) return error; - PROC_LOCK(td->td_proc); - lim_rlimit(td->td_proc, rl, &curlim); - PROC_UNLOCK(td->td_proc); + lim_rlimit(td, rl, &curlim); /* * if the limit is SVR4_RLIM_INFINITY, then we set it to our * unlimited. * We should also: If it is SVR4_RLIM_SAVED_MAX, we should set the * new limit to the corresponding saved hard limit, and if * it is equal to SVR4_RLIM_SAVED_CUR, we should set it to the * corresponding saved soft limit. * */ if (slim.rlim_max == SVR4_RLIM_INFINITY) blim.rlim_max = RLIM_INFINITY; else if (OKLIMIT(slim.rlim_max)) blim.rlim_max = (rlim_t) slim.rlim_max; else if (slim.rlim_max == SVR4_RLIM_SAVED_MAX) blim.rlim_max = curlim.rlim_max; else if (slim.rlim_max == SVR4_RLIM_SAVED_CUR) blim.rlim_max = curlim.rlim_cur; if (slim.rlim_cur == SVR4_RLIM_INFINITY) blim.rlim_cur = RLIM_INFINITY; else if (OKLIMIT(slim.rlim_cur)) blim.rlim_cur = (rlim_t) slim.rlim_cur; else if (slim.rlim_cur == SVR4_RLIM_SAVED_MAX) blim.rlim_cur = curlim.rlim_max; else if (slim.rlim_cur == SVR4_RLIM_SAVED_CUR) blim.rlim_cur = curlim.rlim_cur; return (kern_setrlimit(td, rl, &blim)); } int svr4_sys_getrlimit64(td, uap) struct thread *td; struct svr4_sys_getrlimit64_args *uap; { int rl = svr4_to_native_rl(uap->which); struct rlimit blim; struct svr4_rlimit64 slim; if (rl == -1) return EINVAL; - PROC_LOCK(td->td_proc); - lim_rlimit(td->td_proc, rl, &blim); - PROC_UNLOCK(td->td_proc); + lim_rlimit(td, rl, &blim); /* * Our infinity, is their maxfiles. */ if (rl == RLIMIT_NOFILE && blim.rlim_max == RLIM_INFINITY) blim.rlim_max = maxfiles; /* * If the limit can be be represented, it is returned. * Otherwise, if rlim_cur == rlim_max, return SVR4_RLIM_SAVED_MAX * else return SVR4_RLIM_SAVED_CUR */ if (blim.rlim_max == RLIM_INFINITY) slim.rlim_max = SVR4_RLIM64_INFINITY; else if (OKLIMIT64(blim.rlim_max)) slim.rlim_max = (svr4_rlim64_t) blim.rlim_max; else slim.rlim_max = SVR4_RLIM64_SAVED_MAX; if (blim.rlim_cur == RLIM_INFINITY) slim.rlim_cur = SVR4_RLIM64_INFINITY; else if (OKLIMIT64(blim.rlim_cur)) slim.rlim_cur = (svr4_rlim64_t) blim.rlim_cur; else if (blim.rlim_max == blim.rlim_cur) slim.rlim_cur = SVR4_RLIM64_SAVED_MAX; else slim.rlim_cur = SVR4_RLIM64_SAVED_CUR; return copyout(&slim, uap->rlp, sizeof(*uap->rlp)); } int svr4_sys_setrlimit64(td, uap) struct thread *td; struct svr4_sys_setrlimit64_args *uap; { int rl = svr4_to_native_rl(uap->which); struct rlimit blim, curlim; struct svr4_rlimit64 slim; int error; if (rl == -1) return EINVAL; if ((error = copyin(uap->rlp, &slim, sizeof(slim))) != 0) return error; - PROC_LOCK(td->td_proc); - lim_rlimit(td->td_proc, rl, &curlim); - PROC_UNLOCK(td->td_proc); + lim_rlimit(td, rl, &curlim); /* * if the limit is SVR4_RLIM64_INFINITY, then we set it to our * unlimited. * We should also: If it is SVR4_RLIM64_SAVED_MAX, we should set the * new limit to the corresponding saved hard limit, and if * it is equal to SVR4_RLIM64_SAVED_CUR, we should set it to the * corresponding saved soft limit. * */ if (slim.rlim_max == SVR4_RLIM64_INFINITY) blim.rlim_max = RLIM_INFINITY; else if (OKLIMIT64(slim.rlim_max)) blim.rlim_max = (rlim_t) slim.rlim_max; else if (slim.rlim_max == SVR4_RLIM64_SAVED_MAX) blim.rlim_max = curlim.rlim_max; else if (slim.rlim_max == SVR4_RLIM64_SAVED_CUR) blim.rlim_max = curlim.rlim_cur; if (slim.rlim_cur == SVR4_RLIM64_INFINITY) blim.rlim_cur = RLIM_INFINITY; else if (OKLIMIT64(slim.rlim_cur)) blim.rlim_cur = (rlim_t) slim.rlim_cur; else if (slim.rlim_cur == SVR4_RLIM64_SAVED_MAX) blim.rlim_cur = curlim.rlim_max; else if (slim.rlim_cur == SVR4_RLIM64_SAVED_CUR) blim.rlim_cur = curlim.rlim_cur; return (kern_setrlimit(td, rl, &blim)); } Index: head/sys/dev/drm2/i915/i915_gem.c =================================================================== --- head/sys/dev/drm2/i915/i915_gem.c (revision 284214) +++ head/sys/dev/drm2/i915/i915_gem.c (revision 284215) @@ -1,4281 +1,4281 @@ /*- * Copyright © 2008 Intel Corporation * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), * to deal in the Software without restriction, including without limitation * the rights to use, copy, modify, merge, publish, distribute, sublicense, * and/or sell copies of the Software, and to permit persons to whom the * Software is furnished to do so, subject to the following conditions: * * The above copyright notice and this permission notice (including the next * paragraph) shall be included in all copies or substantial portions of the * Software. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS * IN THE SOFTWARE. * * Authors: * Eric Anholt * * Copyright (c) 2011 The FreeBSD Foundation * All rights reserved. * * This software was 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. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include static void i915_gem_object_flush_cpu_write_domain( struct drm_i915_gem_object *obj); static uint32_t i915_gem_get_gtt_size(struct drm_device *dev, uint32_t size, int tiling_mode); static uint32_t i915_gem_get_gtt_alignment(struct drm_device *dev, uint32_t size, int tiling_mode); static int i915_gem_object_bind_to_gtt(struct drm_i915_gem_object *obj, unsigned alignment, bool map_and_fenceable); static int i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj, int flags); static void i915_gem_object_put_pages_gtt(struct drm_i915_gem_object *obj); static void i915_gem_object_put_pages_range(struct drm_i915_gem_object *obj, off_t start, off_t end); static int i915_gem_object_get_pages_range(struct drm_i915_gem_object *obj, off_t start, off_t end); static void i915_gem_object_finish_gtt(struct drm_i915_gem_object *obj); static void i915_gem_object_truncate(struct drm_i915_gem_object *obj); static int i915_gem_object_is_purgeable(struct drm_i915_gem_object *obj); static bool i915_gem_object_is_inactive(struct drm_i915_gem_object *obj); static int i915_gem_object_needs_bit17_swizzle(struct drm_i915_gem_object *obj); static vm_page_t i915_gem_wire_page(vm_object_t object, vm_pindex_t pindex, bool *fresh); static void i915_gem_process_flushing_list(struct intel_ring_buffer *ring, uint32_t flush_domains); static void i915_gem_reset_fences(struct drm_device *dev); static void i915_gem_retire_task_handler(void *arg, int pending); static void i915_gem_lowmem(void *arg); static void i915_gem_write_fence(struct drm_device *dev, int reg, struct drm_i915_gem_object *obj); static int __wait_seqno(struct intel_ring_buffer *ring, u32 seqno, bool interruptible); static int i915_gem_check_olr(struct intel_ring_buffer *ring, u32 seqno); MALLOC_DEFINE(DRM_I915_GEM, "i915gem", "Allocations from i915 gem"); long i915_gem_wired_pages_cnt; static bool cpu_cache_is_coherent(struct drm_device *dev, enum i915_cache_level level) { return HAS_LLC(dev) || level != I915_CACHE_NONE; } static bool cpu_write_needs_clflush(struct drm_i915_gem_object *obj) { if (!cpu_cache_is_coherent(obj->base.dev, obj->cache_level)) return true; return obj->pin_display; } static inline void i915_gem_object_fence_lost(struct drm_i915_gem_object *obj) { if (obj->tiling_mode) i915_gem_release_mmap(obj); /* As we do not have an associated fence register, we will force * a tiling change if we ever need to acquire one. */ obj->fence_dirty = false; obj->fence_reg = I915_FENCE_REG_NONE; } static void i915_gem_info_add_obj(struct drm_i915_private *dev_priv, size_t size) { dev_priv->mm.object_count++; dev_priv->mm.object_memory += size; } static void i915_gem_info_remove_obj(struct drm_i915_private *dev_priv, size_t size) { dev_priv->mm.object_count--; dev_priv->mm.object_memory -= size; } static int i915_gem_wait_for_error(struct drm_device *dev) { struct drm_i915_private *dev_priv; int ret; dev_priv = dev->dev_private; if (!atomic_load_acq_int(&dev_priv->mm.wedged)) return (0); mtx_lock(&dev_priv->error_completion_lock); while (dev_priv->error_completion == 0) { ret = -msleep(&dev_priv->error_completion, &dev_priv->error_completion_lock, PCATCH, "915wco", 0); if (ret == -ERESTART) ret = -ERESTARTSYS; if (ret != 0) { mtx_unlock(&dev_priv->error_completion_lock); return (ret); } } mtx_unlock(&dev_priv->error_completion_lock); if (atomic_load_acq_int(&dev_priv->mm.wedged)) { mtx_lock(&dev_priv->error_completion_lock); dev_priv->error_completion++; mtx_unlock(&dev_priv->error_completion_lock); } return (0); } int i915_mutex_lock_interruptible(struct drm_device *dev) { struct drm_i915_private *dev_priv; int ret; dev_priv = dev->dev_private; ret = i915_gem_wait_for_error(dev); if (ret != 0) return (ret); /* * interruptible shall it be. might indeed be if dev_lock is * changed to sx */ ret = sx_xlock_sig(&dev->dev_struct_lock); if (ret != 0) return (-ret); return (0); } void i915_gem_free_object(struct drm_gem_object *gem_obj) { struct drm_i915_gem_object *obj = to_intel_bo(gem_obj); struct drm_device *dev; drm_i915_private_t *dev_priv; dev = obj->base.dev; dev_priv = dev->dev_private; CTR1(KTR_DRM, "object_destroy_tail %p", obj); if (obj->phys_obj) i915_gem_detach_phys_object(dev, obj); obj->pin_count = 0; if (i915_gem_object_unbind(obj) == -ERESTARTSYS) { bool was_interruptible; was_interruptible = dev_priv->mm.interruptible; dev_priv->mm.interruptible = false; if (i915_gem_object_unbind(obj)) printf("i915_gem_free_object: unbind\n"); dev_priv->mm.interruptible = was_interruptible; } drm_gem_free_mmap_offset(&obj->base); drm_gem_object_release(&obj->base); i915_gem_info_remove_obj(dev_priv, obj->base.size); free(obj->bit_17, DRM_I915_GEM); free(obj, DRM_I915_GEM); } static void init_ring_lists(struct intel_ring_buffer *ring) { INIT_LIST_HEAD(&ring->active_list); INIT_LIST_HEAD(&ring->request_list); INIT_LIST_HEAD(&ring->gpu_write_list); } void i915_gem_load(struct drm_device *dev) { drm_i915_private_t *dev_priv; int i; dev_priv = dev->dev_private; INIT_LIST_HEAD(&dev_priv->mm.active_list); INIT_LIST_HEAD(&dev_priv->mm.flushing_list); INIT_LIST_HEAD(&dev_priv->mm.inactive_list); INIT_LIST_HEAD(&dev_priv->mm.fence_list); INIT_LIST_HEAD(&dev_priv->mm.gtt_list); for (i = 0; i < I915_NUM_RINGS; i++) init_ring_lists(&dev_priv->rings[i]); for (i = 0; i < I915_MAX_NUM_FENCES; i++) INIT_LIST_HEAD(&dev_priv->fence_regs[i].lru_list); TIMEOUT_TASK_INIT(dev_priv->tq, &dev_priv->mm.retire_task, 0, i915_gem_retire_task_handler, dev_priv); dev_priv->error_completion = 0; /* On GEN3 we really need to make sure the ARB C3 LP bit is set */ if (IS_GEN3(dev)) { I915_WRITE(MI_ARB_STATE, _MASKED_BIT_ENABLE(MI_ARB_C3_LP_WRITE_ENABLE)); } dev_priv->relative_constants_mode = I915_EXEC_CONSTANTS_REL_GENERAL; /* Old X drivers will take 0-2 for front, back, depth buffers */ if (!drm_core_check_feature(dev, DRIVER_MODESET)) dev_priv->fence_reg_start = 3; if (INTEL_INFO(dev)->gen >= 4 || IS_I945G(dev) || IS_I945GM(dev) || IS_G33(dev)) dev_priv->num_fence_regs = 16; else dev_priv->num_fence_regs = 8; /* Initialize fence registers to zero */ i915_gem_reset_fences(dev); i915_gem_detect_bit_6_swizzle(dev); dev_priv->mm.interruptible = true; dev_priv->mm.i915_lowmem = EVENTHANDLER_REGISTER(vm_lowmem, i915_gem_lowmem, dev, EVENTHANDLER_PRI_ANY); } int i915_gem_init_ioctl(struct drm_device *dev, void *data, struct drm_file *file) { struct drm_i915_gem_init *args; drm_i915_private_t *dev_priv; int error; if (drm_core_check_feature(dev, DRIVER_MODESET)) return -ENODEV; dev_priv = dev->dev_private; args = data; if (args->gtt_start >= args->gtt_end || (args->gtt_end | args->gtt_start) & (PAGE_SIZE - 1)) return (-EINVAL); if (mtx_initialized(&dev_priv->mm.gtt_space.unused_lock)) return (-EBUSY); /* GEM with user mode setting was never supported on ilk and later. */ if (INTEL_INFO(dev)->gen >= 5) return -ENODEV; /* * XXXKIB. The second-time initialization should be guarded * against. */ DRM_LOCK(dev); error = i915_gem_init_global_gtt(dev, args->gtt_start, args->gtt_end, args->gtt_end); DRM_UNLOCK(dev); return (error); } int i915_gem_idle(struct drm_device *dev) { drm_i915_private_t *dev_priv; int ret; DRM_LOCK(dev); dev_priv = dev->dev_private; if (dev_priv->mm.suspended) { DRM_UNLOCK(dev); return (0); } ret = i915_gpu_idle(dev); if (ret != 0) { DRM_UNLOCK(dev); return (ret); } i915_gem_retire_requests(dev); /* Under UMS, be paranoid and evict. */ if (!drm_core_check_feature(dev, DRIVER_MODESET)) { ret = i915_gem_evict_everything(dev, false); if (ret != 0) { DRM_UNLOCK(dev); return ret; } } i915_gem_reset_fences(dev); /* Hack! Don't let anybody do execbuf while we don't control the chip. * We need to replace this with a semaphore, or something. * And not confound mm.suspended! */ dev_priv->mm.suspended = 1; callout_stop(&dev_priv->hangcheck_timer); i915_kernel_lost_context(dev); i915_gem_cleanup_ringbuffer(dev); DRM_UNLOCK(dev); /* Cancel the retire work handler, which should be idle now. */ taskqueue_cancel_timeout(dev_priv->tq, &dev_priv->mm.retire_task, NULL); return (ret); } void i915_gem_init_swizzling(struct drm_device *dev) { drm_i915_private_t *dev_priv; dev_priv = dev->dev_private; if (INTEL_INFO(dev)->gen < 5 || dev_priv->mm.bit_6_swizzle_x == I915_BIT_6_SWIZZLE_NONE) return; I915_WRITE(DISP_ARB_CTL, I915_READ(DISP_ARB_CTL) | DISP_TILE_SURFACE_SWIZZLING); if (IS_GEN5(dev)) return; I915_WRITE(TILECTL, I915_READ(TILECTL) | TILECTL_SWZCTL); if (IS_GEN6(dev)) I915_WRITE(ARB_MODE, _MASKED_BIT_ENABLE(ARB_MODE_SWIZZLE_SNB)); else I915_WRITE(ARB_MODE, _MASKED_BIT_ENABLE(ARB_MODE_SWIZZLE_IVB)); } void i915_gem_init_ppgtt(struct drm_device *dev) { drm_i915_private_t *dev_priv; struct i915_hw_ppgtt *ppgtt; uint32_t pd_offset, pd_entry; vm_paddr_t pt_addr; struct intel_ring_buffer *ring; u_int first_pd_entry_in_global_pt, i; dev_priv = dev->dev_private; ppgtt = dev_priv->mm.aliasing_ppgtt; if (ppgtt == NULL) return; first_pd_entry_in_global_pt = 512 * 1024 - I915_PPGTT_PD_ENTRIES; for (i = 0; i < ppgtt->num_pd_entries; i++) { pt_addr = VM_PAGE_TO_PHYS(ppgtt->pt_pages[i]); pd_entry = GEN6_PDE_ADDR_ENCODE(pt_addr); pd_entry |= GEN6_PDE_VALID; intel_gtt_write(first_pd_entry_in_global_pt + i, pd_entry); } intel_gtt_read_pte(first_pd_entry_in_global_pt); pd_offset = ppgtt->pd_offset; pd_offset /= 64; /* in cachelines, */ pd_offset <<= 16; if (INTEL_INFO(dev)->gen == 6) { uint32_t ecochk, gab_ctl, ecobits; ecobits = I915_READ(GAC_ECO_BITS); I915_WRITE(GAC_ECO_BITS, ecobits | ECOBITS_PPGTT_CACHE64B); gab_ctl = I915_READ(GAB_CTL); I915_WRITE(GAB_CTL, gab_ctl | GAB_CTL_CONT_AFTER_PAGEFAULT); ecochk = I915_READ(GAM_ECOCHK); I915_WRITE(GAM_ECOCHK, ecochk | ECOCHK_SNB_BIT | ECOCHK_PPGTT_CACHE64B); I915_WRITE(GFX_MODE, _MASKED_BIT_ENABLE(GFX_PPGTT_ENABLE)); } else if (INTEL_INFO(dev)->gen >= 7) { I915_WRITE(GAM_ECOCHK, ECOCHK_PPGTT_CACHE64B); /* GFX_MODE is per-ring on gen7+ */ } for_each_ring(ring, dev_priv, i) { if (INTEL_INFO(dev)->gen >= 7) I915_WRITE(RING_MODE_GEN7(ring), _MASKED_BIT_ENABLE(GFX_PPGTT_ENABLE)); I915_WRITE(RING_PP_DIR_DCLV(ring), PP_DIR_DCLV_2G); I915_WRITE(RING_PP_DIR_BASE(ring), pd_offset); } } int i915_gem_init_hw(struct drm_device *dev) { drm_i915_private_t *dev_priv; int ret; dev_priv = dev->dev_private; i915_gem_init_swizzling(dev); ret = intel_init_render_ring_buffer(dev); if (ret != 0) return (ret); if (HAS_BSD(dev)) { ret = intel_init_bsd_ring_buffer(dev); if (ret != 0) goto cleanup_render_ring; } if (HAS_BLT(dev)) { ret = intel_init_blt_ring_buffer(dev); if (ret != 0) goto cleanup_bsd_ring; } dev_priv->next_seqno = 1; i915_gem_context_init(dev); i915_gem_init_ppgtt(dev); return (0); cleanup_bsd_ring: intel_cleanup_ring_buffer(&dev_priv->rings[VCS]); cleanup_render_ring: intel_cleanup_ring_buffer(&dev_priv->rings[RCS]); return (ret); } static bool intel_enable_ppgtt(struct drm_device *dev) { if (i915_enable_ppgtt >= 0) return i915_enable_ppgtt; /* Disable ppgtt on SNB if VT-d is on. */ if (INTEL_INFO(dev)->gen == 6 && intel_iommu_enabled) return false; return true; } int i915_gem_init(struct drm_device *dev) { struct drm_i915_private *dev_priv = dev->dev_private; unsigned long gtt_size, mappable_size; int ret; gtt_size = dev_priv->mm.gtt.gtt_total_entries << PAGE_SHIFT; mappable_size = dev_priv->mm.gtt.gtt_mappable_entries << PAGE_SHIFT; DRM_LOCK(dev); if (intel_enable_ppgtt(dev) && HAS_ALIASING_PPGTT(dev)) { /* PPGTT pdes are stolen from global gtt ptes, so shrink the * aperture accordingly when using aliasing ppgtt. */ gtt_size -= I915_PPGTT_PD_ENTRIES*PAGE_SIZE; i915_gem_init_global_gtt(dev, 0, mappable_size, gtt_size); ret = i915_gem_init_aliasing_ppgtt(dev); if (ret) { DRM_UNLOCK(dev); return ret; } } else { /* Let GEM Manage all of the aperture. * * However, leave one page at the end still bound to the scratch * page. There are a number of places where the hardware * apparently prefetches past the end of the object, and we've * seen multiple hangs with the GPU head pointer stuck in a * batchbuffer bound at the last page of the aperture. One page * should be enough to keep any prefetching inside of the * aperture. */ i915_gem_init_global_gtt(dev, 0, mappable_size, gtt_size); } ret = i915_gem_init_hw(dev); DRM_UNLOCK(dev); if (ret != 0) { i915_gem_cleanup_aliasing_ppgtt(dev); return (ret); } /* Allow hardware batchbuffers unless told otherwise, but not for KMS. */ if (!drm_core_check_feature(dev, DRIVER_MODESET)) dev_priv->dri1.allow_batchbuffer = 1; return 0; } int i915_gem_get_aperture_ioctl(struct drm_device *dev, void *data, struct drm_file *file) { struct drm_i915_private *dev_priv; struct drm_i915_gem_get_aperture *args; struct drm_i915_gem_object *obj; size_t pinned; dev_priv = dev->dev_private; args = data; pinned = 0; DRM_LOCK(dev); list_for_each_entry(obj, &dev_priv->mm.gtt_list, gtt_list) if (obj->pin_count) pinned += obj->gtt_space->size; DRM_UNLOCK(dev); args->aper_size = dev_priv->mm.gtt_total; args->aper_available_size = args->aper_size - pinned; return (0); } int i915_gem_object_pin(struct drm_i915_gem_object *obj, uint32_t alignment, bool map_and_fenceable) { int ret; if (obj->pin_count == DRM_I915_GEM_OBJECT_MAX_PIN_COUNT) return (-EBUSY); if (obj->gtt_space != NULL) { if ((alignment && obj->gtt_offset & (alignment - 1)) || (map_and_fenceable && !obj->map_and_fenceable)) { DRM_DEBUG("bo is already pinned with incorrect alignment:" " offset=%x, req.alignment=%x, req.map_and_fenceable=%d," " obj->map_and_fenceable=%d\n", obj->gtt_offset, alignment, map_and_fenceable, obj->map_and_fenceable); ret = i915_gem_object_unbind(obj); if (ret != 0) return (ret); } } if (obj->gtt_space == NULL) { ret = i915_gem_object_bind_to_gtt(obj, alignment, map_and_fenceable); if (ret) return (ret); } if (!obj->has_global_gtt_mapping && map_and_fenceable) i915_gem_gtt_bind_object(obj, obj->cache_level); obj->pin_count++; obj->pin_mappable |= map_and_fenceable; return 0; } void i915_gem_object_unpin(struct drm_i915_gem_object *obj) { KASSERT(obj->pin_count != 0, ("zero pin count")); KASSERT(obj->gtt_space != NULL, ("No gtt mapping")); if (--obj->pin_count == 0) obj->pin_mappable = false; } int i915_gem_pin_ioctl(struct drm_device *dev, void *data, struct drm_file *file) { struct drm_i915_gem_pin *args; struct drm_i915_gem_object *obj; struct drm_gem_object *gobj; int ret; args = data; ret = i915_mutex_lock_interruptible(dev); if (ret != 0) return ret; gobj = drm_gem_object_lookup(dev, file, args->handle); if (gobj == NULL) { ret = -ENOENT; goto unlock; } obj = to_intel_bo(gobj); if (obj->madv != I915_MADV_WILLNEED) { DRM_ERROR("Attempting to pin a purgeable buffer\n"); ret = -EINVAL; goto out; } if (obj->pin_filp != NULL && obj->pin_filp != file) { DRM_ERROR("Already pinned in i915_gem_pin_ioctl(): %d\n", args->handle); ret = -EINVAL; goto out; } obj->user_pin_count++; obj->pin_filp = file; if (obj->user_pin_count == 1) { ret = i915_gem_object_pin(obj, args->alignment, true); if (ret != 0) goto out; } /* XXX - flush the CPU caches for pinned objects * as the X server doesn't manage domains yet */ i915_gem_object_flush_cpu_write_domain(obj); args->offset = obj->gtt_offset; out: drm_gem_object_unreference(&obj->base); unlock: DRM_UNLOCK(dev); return (ret); } int i915_gem_unpin_ioctl(struct drm_device *dev, void *data, struct drm_file *file) { struct drm_i915_gem_pin *args; struct drm_i915_gem_object *obj; int ret; args = data; ret = i915_mutex_lock_interruptible(dev); if (ret != 0) return (ret); obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->handle)); if (&obj->base == NULL) { ret = -ENOENT; goto unlock; } if (obj->pin_filp != file) { DRM_ERROR("Not pinned by caller in i915_gem_pin_ioctl(): %d\n", args->handle); ret = -EINVAL; goto out; } obj->user_pin_count--; if (obj->user_pin_count == 0) { obj->pin_filp = NULL; i915_gem_object_unpin(obj); } out: drm_gem_object_unreference(&obj->base); unlock: DRM_UNLOCK(dev); return (ret); } int i915_gem_busy_ioctl(struct drm_device *dev, void *data, struct drm_file *file) { struct drm_i915_gem_busy *args; struct drm_i915_gem_object *obj; int ret; args = data; ret = i915_mutex_lock_interruptible(dev); if (ret != 0) return ret; obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->handle)); if (&obj->base == NULL) { ret = -ENOENT; goto unlock; } args->busy = obj->active; if (args->busy) { if (obj->base.write_domain & I915_GEM_GPU_DOMAINS) { ret = i915_gem_flush_ring(obj->ring, 0, obj->base.write_domain); } else { ret = i915_gem_check_olr(obj->ring, obj->last_rendering_seqno); } i915_gem_retire_requests_ring(obj->ring); args->busy = obj->active; } drm_gem_object_unreference(&obj->base); unlock: DRM_UNLOCK(dev); return (ret); } static int i915_gem_ring_throttle(struct drm_device *dev, struct drm_file *file) { struct drm_i915_private *dev_priv; struct drm_i915_file_private *file_priv; unsigned long recent_enough; struct drm_i915_gem_request *request; struct intel_ring_buffer *ring; u32 seqno; int ret; dev_priv = dev->dev_private; if (atomic_load_acq_int(&dev_priv->mm.wedged)) return (-EIO); file_priv = file->driver_priv; recent_enough = ticks - (20 * hz / 1000); ring = NULL; seqno = 0; mtx_lock(&file_priv->mm.lck); list_for_each_entry(request, &file_priv->mm.request_list, client_list) { if (time_after_eq(request->emitted_jiffies, recent_enough)) break; ring = request->ring; seqno = request->seqno; } mtx_unlock(&file_priv->mm.lck); if (seqno == 0) return (0); ret = __wait_seqno(ring, seqno, true); if (ret == 0) taskqueue_enqueue_timeout(dev_priv->tq, &dev_priv->mm.retire_task, 0); return (ret); } int i915_gem_throttle_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv) { return (i915_gem_ring_throttle(dev, file_priv)); } int i915_gem_madvise_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv) { struct drm_i915_gem_madvise *args; struct drm_i915_gem_object *obj; int ret; args = data; switch (args->madv) { case I915_MADV_DONTNEED: case I915_MADV_WILLNEED: break; default: return (-EINVAL); } ret = i915_mutex_lock_interruptible(dev); if (ret != 0) return (ret); obj = to_intel_bo(drm_gem_object_lookup(dev, file_priv, args->handle)); if (&obj->base == NULL) { ret = -ENOENT; goto unlock; } if (obj->pin_count != 0) { ret = -EINVAL; goto out; } if (obj->madv != I915_MADV_PURGED_INTERNAL) obj->madv = args->madv; if (i915_gem_object_is_purgeable(obj) && obj->gtt_space == NULL) i915_gem_object_truncate(obj); args->retained = obj->madv != I915_MADV_PURGED_INTERNAL; out: drm_gem_object_unreference(&obj->base); unlock: DRM_UNLOCK(dev); return (ret); } void i915_gem_cleanup_ringbuffer(struct drm_device *dev) { drm_i915_private_t *dev_priv; struct intel_ring_buffer *ring; int i; dev_priv = dev->dev_private; for_each_ring(ring, dev_priv, i) intel_cleanup_ring_buffer(ring); } int i915_gem_entervt_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv) { drm_i915_private_t *dev_priv; int ret; if (drm_core_check_feature(dev, DRIVER_MODESET)) return (0); dev_priv = dev->dev_private; if (atomic_load_acq_int(&dev_priv->mm.wedged) != 0) { DRM_ERROR("Reenabling wedged hardware, good luck\n"); atomic_store_rel_int(&dev_priv->mm.wedged, 0); } DRM_LOCK(dev); dev_priv->mm.suspended = 0; ret = i915_gem_init_hw(dev); if (ret != 0) { DRM_UNLOCK(dev); return (ret); } KASSERT(list_empty(&dev_priv->mm.active_list), ("active list")); KASSERT(list_empty(&dev_priv->mm.flushing_list), ("flushing list")); KASSERT(list_empty(&dev_priv->mm.inactive_list), ("inactive list")); DRM_UNLOCK(dev); ret = drm_irq_install(dev); if (ret) goto cleanup_ringbuffer; return (0); cleanup_ringbuffer: DRM_LOCK(dev); i915_gem_cleanup_ringbuffer(dev); dev_priv->mm.suspended = 1; DRM_UNLOCK(dev); return (ret); } int i915_gem_leavevt_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv) { if (drm_core_check_feature(dev, DRIVER_MODESET)) return 0; drm_irq_uninstall(dev); return (i915_gem_idle(dev)); } int i915_gem_create(struct drm_file *file, struct drm_device *dev, uint64_t size, uint32_t *handle_p) { struct drm_i915_gem_object *obj; uint32_t handle; int ret; size = roundup(size, PAGE_SIZE); if (size == 0) return (-EINVAL); obj = i915_gem_alloc_object(dev, size); if (obj == NULL) return (-ENOMEM); ret = drm_gem_handle_create(file, &obj->base, &handle); if (ret != 0) { drm_gem_object_release(&obj->base); i915_gem_info_remove_obj(dev->dev_private, obj->base.size); free(obj, DRM_I915_GEM); return (ret); } /* drop reference from allocate - handle holds it now */ drm_gem_object_unreference(&obj->base); CTR2(KTR_DRM, "object_create %p %x", obj, size); *handle_p = handle; return (0); } int i915_gem_dumb_create(struct drm_file *file, struct drm_device *dev, struct drm_mode_create_dumb *args) { /* have to work out size/pitch and return them */ args->pitch = roundup2(args->width * ((args->bpp + 7) / 8), 64); args->size = args->pitch * args->height; return (i915_gem_create(file, dev, args->size, &args->handle)); } int i915_gem_dumb_destroy(struct drm_file *file, struct drm_device *dev, uint32_t handle) { return (drm_gem_handle_delete(file, handle)); } int i915_gem_create_ioctl(struct drm_device *dev, void *data, struct drm_file *file) { struct drm_i915_gem_create *args = data; return (i915_gem_create(file, dev, args->size, &args->handle)); } #define __user #define __force #define __iomem #define to_user_ptr(x) ((void *)(uintptr_t)(x)) #define offset_in_page(x) ((x) & PAGE_MASK) #define page_to_phys(x) VM_PAGE_TO_PHYS(x) static inline int __copy_to_user_inatomic(void __user *to, const void *from, unsigned n) { return (copyout_nofault(from, to, n) != 0 ? n : 0); } static inline unsigned long __copy_from_user_inatomic_nocache(void *to, const void __user *from, unsigned long n) { /* * XXXKIB. Equivalent Linux function is implemented using * MOVNTI for aligned moves. For unaligned head and tail, * normal move is performed. As such, it is not incorrect, if * only somewhat slower, to use normal copyin. All uses * except shmem_pwrite_fast() have the destination mapped WC. */ return ((copyin_nofault(__DECONST(void *, from), to, n) != 0 ? n : 0)); } static inline int fault_in_multipages_readable(const char __user *uaddr, int size) { char c; int ret = 0; const char __user *end = uaddr + size - 1; if (unlikely(size == 0)) return ret; while (uaddr <= end) { ret = -copyin(uaddr, &c, 1); if (ret != 0) return -EFAULT; uaddr += PAGE_SIZE; } /* Check whether the range spilled into the next page. */ if (((unsigned long)uaddr & ~PAGE_MASK) == ((unsigned long)end & ~PAGE_MASK)) { ret = -copyin(end, &c, 1); } return ret; } static inline int fault_in_multipages_writeable(char __user *uaddr, int size) { int ret = 0; char __user *end = uaddr + size - 1; if (unlikely(size == 0)) return ret; /* * Writing zeroes into userspace here is OK, because we know that if * the zero gets there, we'll be overwriting it. */ while (uaddr <= end) { ret = subyte(uaddr, 0); if (ret != 0) return -EFAULT; uaddr += PAGE_SIZE; } /* Check whether the range spilled into the next page. */ if (((unsigned long)uaddr & ~PAGE_MASK) == ((unsigned long)end & ~PAGE_MASK)) ret = subyte(end, 0); return ret; } static inline int __copy_to_user_swizzled(char __user *cpu_vaddr, const char *gpu_vaddr, int gpu_offset, int length) { int ret, cpu_offset = 0; while (length > 0) { int cacheline_end = roundup2(gpu_offset + 1, 64); int this_length = min(cacheline_end - gpu_offset, length); int swizzled_gpu_offset = gpu_offset ^ 64; ret = __copy_to_user(cpu_vaddr + cpu_offset, gpu_vaddr + swizzled_gpu_offset, this_length); if (ret) return ret + length; cpu_offset += this_length; gpu_offset += this_length; length -= this_length; } return 0; } static inline int __copy_from_user_swizzled(char *gpu_vaddr, int gpu_offset, const char __user *cpu_vaddr, int length) { int ret, cpu_offset = 0; while (length > 0) { int cacheline_end = roundup2(gpu_offset + 1, 64); int this_length = min(cacheline_end - gpu_offset, length); int swizzled_gpu_offset = gpu_offset ^ 64; ret = __copy_from_user(gpu_vaddr + swizzled_gpu_offset, cpu_vaddr + cpu_offset, this_length); if (ret) return ret + length; cpu_offset += this_length; gpu_offset += this_length; length -= this_length; } return 0; } static int i915_gem_phys_pwrite(struct drm_device *dev, struct drm_i915_gem_object *obj, struct drm_i915_gem_pwrite *args, struct drm_file *file_priv) { void *vaddr = (char *)obj->phys_obj->handle->vaddr + args->offset; char __user *user_data = to_user_ptr(args->data_ptr); if (__copy_from_user_inatomic_nocache(vaddr, user_data, args->size)) { unsigned long unwritten; /* The physical object once assigned is fixed for the lifetime * of the obj, so we can safely drop the lock and continue * to access vaddr. */ DRM_UNLOCK(dev); unwritten = copy_from_user(vaddr, user_data, args->size); DRM_LOCK(dev); if (unwritten) return -EFAULT; } i915_gem_chipset_flush(dev); return 0; } /* Per-page copy function for the shmem pread fastpath. * Flushes invalid cachelines before reading the target if * needs_clflush is set. */ static int shmem_pread_fast(vm_page_t page, int shmem_page_offset, int page_length, char __user *user_data, bool page_do_bit17_swizzling, bool needs_clflush) { char *vaddr; struct sf_buf *sf; int ret; if (unlikely(page_do_bit17_swizzling)) return -EINVAL; sched_pin(); sf = sf_buf_alloc(page, SFB_NOWAIT | SFB_CPUPRIVATE); if (sf == NULL) { sched_unpin(); return (-EFAULT); } vaddr = (char *)sf_buf_kva(sf); if (needs_clflush) drm_clflush_virt_range(vaddr + shmem_page_offset, page_length); ret = __copy_to_user_inatomic(user_data, vaddr + shmem_page_offset, page_length); sf_buf_free(sf); sched_unpin(); return ret ? -EFAULT : 0; } static void shmem_clflush_swizzled_range(char *addr, unsigned long length, bool swizzled) { if (unlikely(swizzled)) { unsigned long start = (unsigned long) addr; unsigned long end = (unsigned long) addr + length; /* For swizzling simply ensure that we always flush both * channels. Lame, but simple and it works. Swizzled * pwrite/pread is far from a hotpath - current userspace * doesn't use it at all. */ start = rounddown2(start, 128); end = roundup2(end, 128); drm_clflush_virt_range((void *)start, end - start); } else { drm_clflush_virt_range(addr, length); } } /* Only difference to the fast-path function is that this can handle bit17 * and uses non-atomic copy and kmap functions. */ static int shmem_pread_slow(vm_page_t page, int shmem_page_offset, int page_length, char __user *user_data, bool page_do_bit17_swizzling, bool needs_clflush) { char *vaddr; struct sf_buf *sf; int ret; sf = sf_buf_alloc(page, 0); vaddr = (char *)sf_buf_kva(sf); if (needs_clflush) shmem_clflush_swizzled_range(vaddr + shmem_page_offset, page_length, page_do_bit17_swizzling); if (page_do_bit17_swizzling) ret = __copy_to_user_swizzled(user_data, vaddr, shmem_page_offset, page_length); else ret = __copy_to_user(user_data, vaddr + shmem_page_offset, page_length); sf_buf_free(sf); return ret ? - EFAULT : 0; } static int i915_gem_shmem_pread(struct drm_device *dev, struct drm_i915_gem_object *obj, struct drm_i915_gem_pread *args, struct drm_file *file) { char __user *user_data; ssize_t remain, sremain; off_t offset, soffset; int shmem_page_offset, page_length, ret = 0; int obj_do_bit17_swizzling, page_do_bit17_swizzling; int prefaulted = 0; int needs_clflush = 0; user_data = to_user_ptr(args->data_ptr); sremain = remain = args->size; obj_do_bit17_swizzling = i915_gem_object_needs_bit17_swizzle(obj); if (!(obj->base.read_domains & I915_GEM_DOMAIN_CPU)) { /* If we're not in the cpu read domain, set ourself into the gtt * read domain and manually flush cachelines (if required). This * optimizes for the case when the gpu will dirty the data * anyway again before the next pread happens. */ needs_clflush = !cpu_cache_is_coherent(dev, obj->cache_level); ret = i915_gem_object_set_to_gtt_domain(obj, false); if (ret) return ret; } soffset = offset = args->offset; ret = i915_gem_object_get_pages_range(obj, soffset, soffset + sremain); if (ret) return ret; i915_gem_object_pin_pages(obj); VM_OBJECT_WLOCK(obj->base.vm_obj); for (vm_page_t page = vm_page_find_least(obj->base.vm_obj, OFF_TO_IDX(offset));; page = vm_page_next(page)) { VM_OBJECT_WUNLOCK(obj->base.vm_obj); if (remain <= 0) break; /* Operation in this page * * shmem_page_offset = offset within page in shmem file * page_length = bytes to copy for this page */ shmem_page_offset = offset_in_page(offset); page_length = remain; if ((shmem_page_offset + page_length) > PAGE_SIZE) page_length = PAGE_SIZE - shmem_page_offset; page_do_bit17_swizzling = obj_do_bit17_swizzling && (page_to_phys(page) & (1 << 17)) != 0; ret = shmem_pread_fast(page, shmem_page_offset, page_length, user_data, page_do_bit17_swizzling, needs_clflush); if (ret == 0) goto next_page; DRM_UNLOCK(dev); if (likely(!i915_prefault_disable) && !prefaulted) { ret = fault_in_multipages_writeable(user_data, remain); /* Userspace is tricking us, but we've already clobbered * its pages with the prefault and promised to write the * data up to the first fault. Hence ignore any errors * and just continue. */ (void)ret; prefaulted = 1; } ret = shmem_pread_slow(page, shmem_page_offset, page_length, user_data, page_do_bit17_swizzling, needs_clflush); DRM_LOCK(dev); next_page: vm_page_reference(page); if (ret) goto out; remain -= page_length; user_data += page_length; offset += page_length; VM_OBJECT_WLOCK(obj->base.vm_obj); } out: i915_gem_object_unpin_pages(obj); i915_gem_object_put_pages_range(obj, soffset, soffset + sremain); return ret; } /** * Reads data from the object referenced by handle. * * On error, the contents of *data are undefined. */ int i915_gem_pread_ioctl(struct drm_device *dev, void *data, struct drm_file *file) { struct drm_i915_gem_pread *args = data; struct drm_i915_gem_object *obj; int ret = 0; if (args->size == 0) return 0; if (!useracc(to_user_ptr(args->data_ptr), args->size, VM_PROT_WRITE)) return -EFAULT; ret = i915_mutex_lock_interruptible(dev); if (ret) return ret; obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->handle)); if (&obj->base == NULL) { ret = -ENOENT; goto unlock; } /* Bounds check source. */ if (args->offset > obj->base.size || args->size > obj->base.size - args->offset) { ret = -EINVAL; goto out; } #if 1 KIB_NOTYET(); #else /* prime objects have no backing filp to GEM pread/pwrite * pages from. */ if (!obj->base.filp) { ret = -EINVAL; goto out; } #endif CTR3(KTR_DRM, "pread %p %jx %jx", obj, args->offset, args->size); ret = i915_gem_shmem_pread(dev, obj, args, file); out: drm_gem_object_unreference(&obj->base); unlock: DRM_UNLOCK(dev); return ret; } /* This is the fast write path which cannot handle * page faults in the source data */ static inline int fast_user_write(struct drm_device *dev, off_t page_base, int page_offset, char __user *user_data, int length) { void __iomem *vaddr_atomic; void *vaddr; unsigned long unwritten; vaddr_atomic = pmap_mapdev_attr(dev->agp->base + page_base, length, PAT_WRITE_COMBINING); /* We can use the cpu mem copy function because this is X86. */ vaddr = (char *)vaddr_atomic + page_offset; unwritten = __copy_from_user_inatomic_nocache(vaddr, user_data, length); pmap_unmapdev((vm_offset_t)vaddr_atomic, length); return unwritten; } /** * This is the fast pwrite path, where we copy the data directly from the * user into the GTT, uncached. */ static int i915_gem_gtt_pwrite_fast(struct drm_device *dev, struct drm_i915_gem_object *obj, struct drm_i915_gem_pwrite *args, struct drm_file *file) { ssize_t remain; off_t offset, page_base; char __user *user_data; int page_offset, page_length, ret; ret = i915_gem_object_pin(obj, 0, true); /* XXXKIB ret = i915_gem_obj_ggtt_pin(obj, 0, true, true); */ if (ret != 0) goto out; ret = i915_gem_object_set_to_gtt_domain(obj, true); if (ret) goto out_unpin; ret = i915_gem_object_put_fence(obj); if (ret) goto out_unpin; user_data = to_user_ptr(args->data_ptr); remain = args->size; offset = obj->gtt_offset + args->offset; while (remain > 0) { /* Operation in this page * * page_base = page offset within aperture * page_offset = offset within page * page_length = bytes to copy for this page */ page_base = offset & ~PAGE_MASK; page_offset = offset_in_page(offset); page_length = remain; if ((page_offset + remain) > PAGE_SIZE) page_length = PAGE_SIZE - page_offset; /* If we get a fault while copying data, then (presumably) our * source page isn't available. Return the error and we'll * retry in the slow path. */ if (fast_user_write(dev, page_base, page_offset, user_data, page_length)) { ret = -EFAULT; goto out_unpin; } remain -= page_length; user_data += page_length; offset += page_length; } out_unpin: i915_gem_object_unpin(obj); out: return ret; } /* Per-page copy function for the shmem pwrite fastpath. * Flushes invalid cachelines before writing to the target if * needs_clflush_before is set and flushes out any written cachelines after * writing if needs_clflush is set. */ static int shmem_pwrite_fast(vm_page_t page, int shmem_page_offset, int page_length, char __user *user_data, bool page_do_bit17_swizzling, bool needs_clflush_before, bool needs_clflush_after) { char *vaddr; struct sf_buf *sf; int ret; if (unlikely(page_do_bit17_swizzling)) return -EINVAL; sched_pin(); sf = sf_buf_alloc(page, SFB_NOWAIT | SFB_CPUPRIVATE); if (sf == NULL) { sched_unpin(); return (-EFAULT); } vaddr = (char *)sf_buf_kva(sf); if (needs_clflush_before) drm_clflush_virt_range(vaddr + shmem_page_offset, page_length); ret = __copy_from_user_inatomic_nocache(vaddr + shmem_page_offset, user_data, page_length); if (needs_clflush_after) drm_clflush_virt_range(vaddr + shmem_page_offset, page_length); sf_buf_free(sf); sched_unpin(); return ret ? -EFAULT : 0; } /* Only difference to the fast-path function is that this can handle bit17 * and uses non-atomic copy and kmap functions. */ static int shmem_pwrite_slow(vm_page_t page, int shmem_page_offset, int page_length, char __user *user_data, bool page_do_bit17_swizzling, bool needs_clflush_before, bool needs_clflush_after) { char *vaddr; struct sf_buf *sf; int ret; sf = sf_buf_alloc(page, 0); vaddr = (char *)sf_buf_kva(sf); if (unlikely(needs_clflush_before || page_do_bit17_swizzling)) shmem_clflush_swizzled_range(vaddr + shmem_page_offset, page_length, page_do_bit17_swizzling); if (page_do_bit17_swizzling) ret = __copy_from_user_swizzled(vaddr, shmem_page_offset, user_data, page_length); else ret = __copy_from_user(vaddr + shmem_page_offset, user_data, page_length); if (needs_clflush_after) shmem_clflush_swizzled_range(vaddr + shmem_page_offset, page_length, page_do_bit17_swizzling); sf_buf_free(sf); return ret ? -EFAULT : 0; } static int i915_gem_shmem_pwrite(struct drm_device *dev, struct drm_i915_gem_object *obj, struct drm_i915_gem_pwrite *args, struct drm_file *file) { ssize_t remain, sremain; off_t offset, soffset; char __user *user_data; int shmem_page_offset, page_length, ret = 0; int obj_do_bit17_swizzling, page_do_bit17_swizzling; int hit_slowpath = 0; int needs_clflush_after = 0; int needs_clflush_before = 0; user_data = to_user_ptr(args->data_ptr); sremain = remain = args->size; obj_do_bit17_swizzling = i915_gem_object_needs_bit17_swizzle(obj); if (obj->base.write_domain != I915_GEM_DOMAIN_CPU) { /* If we're not in the cpu write domain, set ourself into the gtt * write domain and manually flush cachelines (if required). This * optimizes for the case when the gpu will use the data * right away and we therefore have to clflush anyway. */ needs_clflush_after = cpu_write_needs_clflush(obj); ret = i915_gem_object_set_to_gtt_domain(obj, true); if (ret) return ret; } /* Same trick applies to invalidate partially written cachelines read * before writing. */ if ((obj->base.read_domains & I915_GEM_DOMAIN_CPU) == 0) needs_clflush_before = !cpu_cache_is_coherent(dev, obj->cache_level); soffset = offset = args->offset; ret = i915_gem_object_get_pages_range(obj, soffset, soffset + sremain); if (ret) return ret; i915_gem_object_pin_pages(obj); obj->dirty = 1; VM_OBJECT_WLOCK(obj->base.vm_obj); for (vm_page_t page = vm_page_find_least(obj->base.vm_obj, OFF_TO_IDX(offset));; page = vm_page_next(page)) { VM_OBJECT_WUNLOCK(obj->base.vm_obj); int partial_cacheline_write; if (remain <= 0) break; /* Operation in this page * * shmem_page_offset = offset within page in shmem file * page_length = bytes to copy for this page */ shmem_page_offset = offset_in_page(offset); page_length = remain; if ((shmem_page_offset + page_length) > PAGE_SIZE) page_length = PAGE_SIZE - shmem_page_offset; /* If we don't overwrite a cacheline completely we need to be * careful to have up-to-date data by first clflushing. Don't * overcomplicate things and flush the entire patch. */ partial_cacheline_write = needs_clflush_before && ((shmem_page_offset | page_length) & (cpu_clflush_line_size - 1)); page_do_bit17_swizzling = obj_do_bit17_swizzling && (page_to_phys(page) & (1 << 17)) != 0; ret = shmem_pwrite_fast(page, shmem_page_offset, page_length, user_data, page_do_bit17_swizzling, partial_cacheline_write, needs_clflush_after); if (ret == 0) goto next_page; hit_slowpath = 1; DRM_UNLOCK(dev); ret = shmem_pwrite_slow(page, shmem_page_offset, page_length, user_data, page_do_bit17_swizzling, partial_cacheline_write, needs_clflush_after); DRM_LOCK(dev); next_page: vm_page_dirty(page); vm_page_reference(page); if (ret) goto out; remain -= page_length; user_data += page_length; offset += page_length; VM_OBJECT_WLOCK(obj->base.vm_obj); } out: i915_gem_object_unpin_pages(obj); i915_gem_object_put_pages_range(obj, soffset, soffset + sremain); if (hit_slowpath) { /* * Fixup: Flush cpu caches in case we didn't flush the dirty * cachelines in-line while writing and the object moved * out of the cpu write domain while we've dropped the lock. */ if (!needs_clflush_after && obj->base.write_domain != I915_GEM_DOMAIN_CPU) { i915_gem_clflush_object(obj); i915_gem_chipset_flush(dev); } } if (needs_clflush_after) i915_gem_chipset_flush(dev); return ret; } /** * Writes data to the object referenced by handle. * * On error, the contents of the buffer that were to be modified are undefined. */ int i915_gem_pwrite_ioctl(struct drm_device *dev, void *data, struct drm_file *file) { struct drm_i915_gem_pwrite *args = data; struct drm_i915_gem_object *obj; int ret; if (args->size == 0) return 0; if (!useracc(to_user_ptr(args->data_ptr), args->size, VM_PROT_READ)) return -EFAULT; if (likely(!i915_prefault_disable)) { ret = fault_in_multipages_readable(to_user_ptr(args->data_ptr), args->size); if (ret) return -EFAULT; } ret = i915_mutex_lock_interruptible(dev); if (ret) return ret; obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->handle)); if (&obj->base == NULL) { ret = -ENOENT; goto unlock; } /* Bounds check destination. */ if (args->offset > obj->base.size || args->size > obj->base.size - args->offset) { ret = -EINVAL; goto out; } #if 1 KIB_NOTYET(); #else /* prime objects have no backing filp to GEM pread/pwrite * pages from. */ if (!obj->base.filp) { ret = -EINVAL; goto out; } #endif CTR3(KTR_DRM, "pwrite %p %jx %jx", obj, args->offset, args->size); ret = -EFAULT; /* We can only do the GTT pwrite on untiled buffers, as otherwise * it would end up going through the fenced access, and we'll get * different detiling behavior between reading and writing. * pread/pwrite currently are reading and writing from the CPU * perspective, requiring manual detiling by the client. */ if (obj->phys_obj) { ret = i915_gem_phys_pwrite(dev, obj, args, file); goto out; } if (obj->tiling_mode == I915_TILING_NONE && obj->base.write_domain != I915_GEM_DOMAIN_CPU && cpu_write_needs_clflush(obj)) { ret = i915_gem_gtt_pwrite_fast(dev, obj, args, file); /* Note that the gtt paths might fail with non-page-backed user * pointers (e.g. gtt mappings when moving data between * textures). Fallback to the shmem path in that case. */ } if (ret == -EFAULT || ret == -ENOSPC) ret = i915_gem_shmem_pwrite(dev, obj, args, file); out: drm_gem_object_unreference(&obj->base); unlock: DRM_UNLOCK(dev); return ret; } #undef __user #undef __force #undef __iomem #undef to_user_ptr #undef offset_in_page #undef page_to_phys int i915_gem_set_domain_ioctl(struct drm_device *dev, void *data, struct drm_file *file) { struct drm_i915_gem_set_domain *args; struct drm_i915_gem_object *obj; uint32_t read_domains; uint32_t write_domain; int ret; args = data; read_domains = args->read_domains; write_domain = args->write_domain; if ((write_domain & I915_GEM_GPU_DOMAINS) != 0 || (read_domains & I915_GEM_GPU_DOMAINS) != 0 || (write_domain != 0 && read_domains != write_domain)) return (-EINVAL); ret = i915_mutex_lock_interruptible(dev); if (ret != 0) return (ret); obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->handle)); if (&obj->base == NULL) { ret = -ENOENT; goto unlock; } if ((read_domains & I915_GEM_DOMAIN_GTT) != 0) { ret = i915_gem_object_set_to_gtt_domain(obj, write_domain != 0); if (ret == -EINVAL) ret = 0; } else ret = i915_gem_object_set_to_cpu_domain(obj, write_domain != 0); drm_gem_object_unreference(&obj->base); unlock: DRM_UNLOCK(dev); return (ret); } int i915_gem_sw_finish_ioctl(struct drm_device *dev, void *data, struct drm_file *file) { struct drm_i915_gem_sw_finish *args; struct drm_i915_gem_object *obj; int ret; args = data; ret = i915_mutex_lock_interruptible(dev); if (ret != 0) return (ret); obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->handle)); if (&obj->base == NULL) { ret = -ENOENT; goto unlock; } if (obj->pin_count != 0) i915_gem_object_flush_cpu_write_domain(obj); drm_gem_object_unreference(&obj->base); unlock: DRM_UNLOCK(dev); return (ret); } int i915_gem_mmap_ioctl(struct drm_device *dev, void *data, struct drm_file *file) { struct drm_i915_gem_mmap *args; struct drm_gem_object *obj; struct proc *p; vm_map_t map; vm_offset_t addr; vm_size_t size; int error, rv; args = data; obj = drm_gem_object_lookup(dev, file, args->handle); if (obj == NULL) return (-ENOENT); error = 0; if (args->size == 0) goto out; p = curproc; map = &p->p_vmspace->vm_map; size = round_page(args->size); PROC_LOCK(p); - if (map->size + size > lim_cur(p, RLIMIT_VMEM)) { + if (map->size + size > lim_cur_proc(p, RLIMIT_VMEM)) { PROC_UNLOCK(p); error = -ENOMEM; goto out; } PROC_UNLOCK(p); addr = 0; vm_object_reference(obj->vm_obj); rv = vm_map_find(map, obj->vm_obj, args->offset, &addr, args->size, 0, VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE, VM_PROT_READ | VM_PROT_WRITE, MAP_INHERIT_SHARE); if (rv != KERN_SUCCESS) { vm_object_deallocate(obj->vm_obj); error = -vm_mmap_to_errno(rv); } else { args->addr_ptr = (uint64_t)addr; } out: drm_gem_object_unreference(obj); return (error); } static int i915_gem_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t foff, struct ucred *cred, u_short *color) { *color = 0; /* XXXKIB */ return (0); } int i915_intr_pf; static int i915_gem_pager_fault(vm_object_t vm_obj, vm_ooffset_t offset, int prot, vm_page_t *mres) { struct drm_gem_object *gem_obj; struct drm_i915_gem_object *obj; struct drm_device *dev; drm_i915_private_t *dev_priv; vm_page_t m, oldm; int cause, ret; bool write; gem_obj = vm_obj->handle; obj = to_intel_bo(gem_obj); dev = obj->base.dev; dev_priv = dev->dev_private; #if 0 write = (prot & VM_PROT_WRITE) != 0; #else write = true; #endif vm_object_pip_add(vm_obj, 1); /* * Remove the placeholder page inserted by vm_fault() from the * object before dropping the object lock. If * i915_gem_release_mmap() is active in parallel on this gem * object, then it owns the drm device sx and might find the * placeholder already. Then, since the page is busy, * i915_gem_release_mmap() sleeps waiting for the busy state * of the page cleared. We will be not able to acquire drm * device lock until i915_gem_release_mmap() is able to make a * progress. */ if (*mres != NULL) { oldm = *mres; vm_page_lock(oldm); vm_page_remove(oldm); vm_page_unlock(oldm); *mres = NULL; } else oldm = NULL; VM_OBJECT_WUNLOCK(vm_obj); retry: cause = ret = 0; m = NULL; if (i915_intr_pf) { ret = i915_mutex_lock_interruptible(dev); if (ret != 0) { cause = 10; goto out; } } else DRM_LOCK(dev); /* * Since the object lock was dropped, other thread might have * faulted on the same GTT address and instantiated the * mapping for the page. Recheck. */ VM_OBJECT_WLOCK(vm_obj); m = vm_page_lookup(vm_obj, OFF_TO_IDX(offset)); if (m != NULL) { if (vm_page_busied(m)) { DRM_UNLOCK(dev); vm_page_lock(m); VM_OBJECT_WUNLOCK(vm_obj); vm_page_busy_sleep(m, "915pee"); goto retry; } goto have_page; } else VM_OBJECT_WUNLOCK(vm_obj); /* Now bind it into the GTT if needed */ if (!obj->map_and_fenceable) { ret = i915_gem_object_unbind(obj); if (ret != 0) { cause = 20; goto unlock; } } if (!obj->gtt_space) { ret = i915_gem_object_bind_to_gtt(obj, 0, true); if (ret != 0) { cause = 30; goto unlock; } ret = i915_gem_object_set_to_gtt_domain(obj, write); if (ret != 0) { cause = 40; goto unlock; } } if (!obj->has_global_gtt_mapping) i915_gem_gtt_bind_object(obj, obj->cache_level); ret = i915_gem_object_get_fence(obj); if (ret != 0) { cause = 50; goto unlock; } if (i915_gem_object_is_inactive(obj)) list_move_tail(&obj->mm_list, &dev_priv->mm.inactive_list); obj->fault_mappable = true; VM_OBJECT_WLOCK(vm_obj); m = PHYS_TO_VM_PAGE(dev->agp->base + obj->gtt_offset + offset); KASSERT((m->flags & PG_FICTITIOUS) != 0, ("physical address %#jx not fictitious", (uintmax_t)(dev->agp->base + obj->gtt_offset + offset))); if (m == NULL) { VM_OBJECT_WUNLOCK(vm_obj); cause = 60; ret = -EFAULT; goto unlock; } KASSERT((m->flags & PG_FICTITIOUS) != 0, ("not fictitious %p", m)); KASSERT(m->wire_count == 1, ("wire_count not 1 %p", m)); if (vm_page_busied(m)) { DRM_UNLOCK(dev); vm_page_lock(m); VM_OBJECT_WUNLOCK(vm_obj); vm_page_busy_sleep(m, "915pbs"); goto retry; } if (vm_page_insert(m, vm_obj, OFF_TO_IDX(offset))) { DRM_UNLOCK(dev); VM_OBJECT_WUNLOCK(vm_obj); VM_WAIT; goto retry; } m->valid = VM_PAGE_BITS_ALL; have_page: *mres = m; vm_page_xbusy(m); CTR4(KTR_DRM, "fault %p %jx %x phys %x", gem_obj, offset, prot, m->phys_addr); DRM_UNLOCK(dev); if (oldm != NULL) { vm_page_lock(oldm); vm_page_free(oldm); vm_page_unlock(oldm); } vm_object_pip_wakeup(vm_obj); return (VM_PAGER_OK); unlock: DRM_UNLOCK(dev); out: KASSERT(ret != 0, ("i915_gem_pager_fault: wrong return")); CTR5(KTR_DRM, "fault_fail %p %jx %x err %d %d", gem_obj, offset, prot, -ret, cause); if (ret == -EAGAIN || ret == -EIO || ret == -EINTR) { kern_yield(PRI_USER); goto retry; } VM_OBJECT_WLOCK(vm_obj); vm_object_pip_wakeup(vm_obj); return (VM_PAGER_ERROR); } static void i915_gem_pager_dtor(void *handle) { struct drm_gem_object *obj; struct drm_device *dev; obj = handle; dev = obj->dev; DRM_LOCK(dev); drm_gem_free_mmap_offset(obj); i915_gem_release_mmap(to_intel_bo(obj)); drm_gem_object_unreference(obj); DRM_UNLOCK(dev); } struct cdev_pager_ops i915_gem_pager_ops = { .cdev_pg_fault = i915_gem_pager_fault, .cdev_pg_ctor = i915_gem_pager_ctor, .cdev_pg_dtor = i915_gem_pager_dtor }; int i915_gem_mmap_gtt(struct drm_file *file, struct drm_device *dev, uint32_t handle, uint64_t *offset) { struct drm_i915_private *dev_priv; struct drm_i915_gem_object *obj; int ret; dev_priv = dev->dev_private; ret = i915_mutex_lock_interruptible(dev); if (ret != 0) return (ret); obj = to_intel_bo(drm_gem_object_lookup(dev, file, handle)); if (&obj->base == NULL) { ret = -ENOENT; goto unlock; } if (obj->base.size > dev_priv->mm.gtt_mappable_end) { ret = -E2BIG; goto out; } if (obj->madv != I915_MADV_WILLNEED) { DRM_ERROR("Attempting to mmap a purgeable buffer\n"); ret = -EINVAL; goto out; } ret = drm_gem_create_mmap_offset(&obj->base); if (ret != 0) goto out; *offset = DRM_GEM_MAPPING_OFF(obj->base.map_list.key) | DRM_GEM_MAPPING_KEY; out: drm_gem_object_unreference(&obj->base); unlock: DRM_UNLOCK(dev); return (ret); } int i915_gem_mmap_gtt_ioctl(struct drm_device *dev, void *data, struct drm_file *file) { struct drm_i915_private *dev_priv; struct drm_i915_gem_mmap_gtt *args; dev_priv = dev->dev_private; args = data; return (i915_gem_mmap_gtt(file, dev, args->handle, &args->offset)); } struct drm_i915_gem_object * i915_gem_alloc_object(struct drm_device *dev, size_t size) { struct drm_i915_private *dev_priv; struct drm_i915_gem_object *obj; dev_priv = dev->dev_private; obj = malloc(sizeof(*obj), DRM_I915_GEM, M_WAITOK | M_ZERO); if (drm_gem_object_init(dev, &obj->base, size) != 0) { free(obj, DRM_I915_GEM); return (NULL); } obj->base.write_domain = I915_GEM_DOMAIN_CPU; obj->base.read_domains = I915_GEM_DOMAIN_CPU; if (HAS_LLC(dev)) obj->cache_level = I915_CACHE_LLC; else obj->cache_level = I915_CACHE_NONE; obj->base.driver_private = NULL; obj->fence_reg = I915_FENCE_REG_NONE; INIT_LIST_HEAD(&obj->mm_list); INIT_LIST_HEAD(&obj->gtt_list); INIT_LIST_HEAD(&obj->ring_list); INIT_LIST_HEAD(&obj->exec_list); INIT_LIST_HEAD(&obj->gpu_write_list); obj->madv = I915_MADV_WILLNEED; /* Avoid an unnecessary call to unbind on the first bind. */ obj->map_and_fenceable = true; i915_gem_info_add_obj(dev_priv, size); return (obj); } void i915_gem_clflush_object(struct drm_i915_gem_object *obj) { /* If we don't have a page list set up, then we're not pinned * to GPU, and we can ignore the cache flush because it'll happen * again at bind time. */ if (obj->pages == NULL) return; /* If the GPU is snooping the contents of the CPU cache, * we do not need to manually clear the CPU cache lines. However, * the caches are only snooped when the render cache is * flushed/invalidated. As we always have to emit invalidations * and flushes when moving into and out of the RENDER domain, correct * snooping behaviour occurs naturally as the result of our domain * tracking. */ if (obj->cache_level != I915_CACHE_NONE) return; CTR1(KTR_DRM, "object_clflush %p", obj); drm_clflush_pages(obj->pages, obj->base.size / PAGE_SIZE); } static void i915_gem_object_flush_cpu_write_domain(struct drm_i915_gem_object *obj) { uint32_t old_write_domain; if (obj->base.write_domain != I915_GEM_DOMAIN_CPU) return; i915_gem_clflush_object(obj); intel_gtt_chipset_flush(); old_write_domain = obj->base.write_domain; obj->base.write_domain = 0; CTR3(KTR_DRM, "object_change_domain flush_cpu_write %p %x %x", obj, obj->base.read_domains, old_write_domain); } static int i915_gem_object_flush_gpu_write_domain(struct drm_i915_gem_object *obj) { if ((obj->base.write_domain & I915_GEM_GPU_DOMAINS) == 0) return (0); return (i915_gem_flush_ring(obj->ring, 0, obj->base.write_domain)); } static void i915_gem_object_flush_gtt_write_domain(struct drm_i915_gem_object *obj) { uint32_t old_write_domain; if (obj->base.write_domain != I915_GEM_DOMAIN_GTT) return; wmb(); old_write_domain = obj->base.write_domain; obj->base.write_domain = 0; CTR3(KTR_DRM, "object_change_domain flush gtt_write %p %x %x", obj, obj->base.read_domains, old_write_domain); } int i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool write) { drm_i915_private_t *dev_priv = obj->base.dev->dev_private; uint32_t old_write_domain, old_read_domains; int ret; if (obj->gtt_space == NULL) return (-EINVAL); if (obj->base.write_domain == I915_GEM_DOMAIN_GTT) return 0; ret = i915_gem_object_flush_gpu_write_domain(obj); if (ret != 0) return (ret); if (obj->pending_gpu_write || write) { ret = i915_gem_object_wait_rendering(obj); if (ret != 0) return (ret); } i915_gem_object_flush_cpu_write_domain(obj); old_write_domain = obj->base.write_domain; old_read_domains = obj->base.read_domains; KASSERT((obj->base.write_domain & ~I915_GEM_DOMAIN_GTT) == 0, ("In GTT write domain")); obj->base.read_domains |= I915_GEM_DOMAIN_GTT; if (write) { obj->base.read_domains = I915_GEM_DOMAIN_GTT; obj->base.write_domain = I915_GEM_DOMAIN_GTT; obj->dirty = 1; } /* And bump the LRU for this access */ if (i915_gem_object_is_inactive(obj)) list_move_tail(&obj->mm_list, &dev_priv->mm.inactive_list); CTR3(KTR_DRM, "object_change_domain set_to_gtt %p %x %x", obj, old_read_domains, old_write_domain); return (0); } int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, enum i915_cache_level cache_level) { struct drm_device *dev; drm_i915_private_t *dev_priv; int ret; if (obj->cache_level == cache_level) return 0; if (obj->pin_count) { DRM_DEBUG("can not change the cache level of pinned objects\n"); return (-EBUSY); } dev = obj->base.dev; dev_priv = dev->dev_private; if (obj->gtt_space) { ret = i915_gem_object_finish_gpu(obj); if (ret != 0) return (ret); i915_gem_object_finish_gtt(obj); /* Before SandyBridge, you could not use tiling or fence * registers with snooped memory, so relinquish any fences * currently pointing to our region in the aperture. */ if (INTEL_INFO(obj->base.dev)->gen < 6) { ret = i915_gem_object_put_fence(obj); if (ret != 0) return (ret); } if (obj->has_global_gtt_mapping) i915_gem_gtt_bind_object(obj, cache_level); if (obj->has_aliasing_ppgtt_mapping) i915_ppgtt_bind_object(dev_priv->mm.aliasing_ppgtt, obj, cache_level); } if (cache_level == I915_CACHE_NONE) { u32 old_read_domains, old_write_domain; /* If we're coming from LLC cached, then we haven't * actually been tracking whether the data is in the * CPU cache or not, since we only allow one bit set * in obj->write_domain and have been skipping the clflushes. * Just set it to the CPU cache for now. */ KASSERT((obj->base.write_domain & ~I915_GEM_DOMAIN_CPU) == 0, ("obj %p in CPU write domain", obj)); KASSERT((obj->base.read_domains & ~I915_GEM_DOMAIN_CPU) == 0, ("obj %p in CPU read domain", obj)); old_read_domains = obj->base.read_domains; old_write_domain = obj->base.write_domain; obj->base.read_domains = I915_GEM_DOMAIN_CPU; obj->base.write_domain = I915_GEM_DOMAIN_CPU; CTR3(KTR_DRM, "object_change_domain set_cache_level %p %x %x", obj, old_read_domains, old_write_domain); } obj->cache_level = cache_level; return (0); } static bool is_pin_display(struct drm_i915_gem_object *obj) { /* There are 3 sources that pin objects: * 1. The display engine (scanouts, sprites, cursors); * 2. Reservations for execbuffer; * 3. The user. * * We can ignore reservations as we hold the struct_mutex and * are only called outside of the reservation path. The user * can only increment pin_count once, and so if after * subtracting the potential reference by the user, any pin_count * remains, it must be due to another use by the display engine. */ return obj->pin_count - !!obj->user_pin_count; } int i915_gem_object_pin_to_display_plane(struct drm_i915_gem_object *obj, u32 alignment, struct intel_ring_buffer *pipelined) { u32 old_read_domains, old_write_domain; int ret; ret = i915_gem_object_flush_gpu_write_domain(obj); if (ret != 0) return (ret); if (pipelined != obj->ring) { ret = i915_gem_object_sync(obj, pipelined); if (ret) return (ret); } obj->pin_display = true; ret = i915_gem_object_set_cache_level(obj, I915_CACHE_NONE); if (ret != 0) goto err_unpin_display; ret = i915_gem_object_pin(obj, alignment, true); if (ret != 0) goto err_unpin_display; i915_gem_object_flush_cpu_write_domain(obj); old_write_domain = obj->base.write_domain; old_read_domains = obj->base.read_domains; KASSERT((obj->base.write_domain & ~I915_GEM_DOMAIN_GTT) == 0, ("obj %p in GTT write domain", obj)); obj->base.read_domains |= I915_GEM_DOMAIN_GTT; CTR3(KTR_DRM, "object_change_domain pin_to_display_plan %p %x %x", obj, old_read_domains, obj->base.write_domain); return (0); err_unpin_display: obj->pin_display = is_pin_display(obj); return ret; } void i915_gem_object_unpin_from_display_plane(struct drm_i915_gem_object *obj) { i915_gem_object_unpin(obj); obj->pin_display = is_pin_display(obj); } int i915_gem_object_finish_gpu(struct drm_i915_gem_object *obj) { int ret; if ((obj->base.read_domains & I915_GEM_GPU_DOMAINS) == 0) return (0); if (obj->base.write_domain & I915_GEM_GPU_DOMAINS) { ret = i915_gem_flush_ring(obj->ring, 0, obj->base.write_domain); if (ret != 0) return (ret); } ret = i915_gem_object_wait_rendering(obj); if (ret != 0) return (ret); obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS; return (0); } int i915_gem_object_set_to_cpu_domain(struct drm_i915_gem_object *obj, bool write) { uint32_t old_write_domain, old_read_domains; int ret; if (obj->base.write_domain == I915_GEM_DOMAIN_CPU) return 0; ret = i915_gem_object_flush_gpu_write_domain(obj); if (ret != 0) return (ret); if (write || obj->pending_gpu_write) { ret = i915_gem_object_wait_rendering(obj); if (ret != 0) return (ret); } i915_gem_object_flush_gtt_write_domain(obj); old_write_domain = obj->base.write_domain; old_read_domains = obj->base.read_domains; if ((obj->base.read_domains & I915_GEM_DOMAIN_CPU) == 0) { i915_gem_clflush_object(obj); obj->base.read_domains |= I915_GEM_DOMAIN_CPU; } KASSERT((obj->base.write_domain & ~I915_GEM_DOMAIN_CPU) == 0, ("In cpu write domain")); if (write) { obj->base.read_domains = I915_GEM_DOMAIN_CPU; obj->base.write_domain = I915_GEM_DOMAIN_CPU; } CTR3(KTR_DRM, "object_change_domain set_to_cpu %p %x %x", obj, old_read_domains, old_write_domain); return (0); } static uint32_t i915_gem_get_gtt_size(struct drm_device *dev, uint32_t size, int tiling_mode) { uint32_t gtt_size; if (INTEL_INFO(dev)->gen >= 4 || tiling_mode == I915_TILING_NONE) return (size); /* Previous chips need a power-of-two fence region when tiling */ if (INTEL_INFO(dev)->gen == 3) gtt_size = 1024*1024; else gtt_size = 512*1024; while (gtt_size < size) gtt_size <<= 1; return (gtt_size); } /** * i915_gem_get_gtt_alignment - return required GTT alignment for an object * @obj: object to check * * Return the required GTT alignment for an object, taking into account * potential fence register mapping. */ static uint32_t i915_gem_get_gtt_alignment(struct drm_device *dev, uint32_t size, int tiling_mode) { /* * Minimum alignment is 4k (GTT page size), but might be greater * if a fence register is needed for the object. */ if (INTEL_INFO(dev)->gen >= 4 || tiling_mode == I915_TILING_NONE) return (4096); /* * Previous chips need to be aligned to the size of the smallest * fence register that can contain the object. */ return (i915_gem_get_gtt_size(dev, size, tiling_mode)); } uint32_t i915_gem_get_unfenced_gtt_alignment(struct drm_device *dev, uint32_t size, int tiling_mode) { if (tiling_mode == I915_TILING_NONE) return (4096); /* * Minimum alignment is 4k (GTT page size) for sane hw. */ if (INTEL_INFO(dev)->gen >= 4 || IS_G33(dev)) return (4096); /* * Previous hardware however needs to be aligned to a power-of-two * tile height. The simplest method for determining this is to reuse * the power-of-tile object size. */ return (i915_gem_get_gtt_size(dev, size, tiling_mode)); } static int i915_gem_object_bind_to_gtt(struct drm_i915_gem_object *obj, unsigned alignment, bool map_and_fenceable) { struct drm_device *dev; struct drm_i915_private *dev_priv; struct drm_mm_node *free_space; uint32_t size, fence_size, fence_alignment, unfenced_alignment; bool mappable, fenceable; int ret; dev = obj->base.dev; dev_priv = dev->dev_private; if (obj->madv != I915_MADV_WILLNEED) { DRM_ERROR("Attempting to bind a purgeable object\n"); return (-EINVAL); } fence_size = i915_gem_get_gtt_size(dev, obj->base.size, obj->tiling_mode); fence_alignment = i915_gem_get_gtt_alignment(dev, obj->base.size, obj->tiling_mode); unfenced_alignment = i915_gem_get_unfenced_gtt_alignment(dev, obj->base.size, obj->tiling_mode); if (alignment == 0) alignment = map_and_fenceable ? fence_alignment : unfenced_alignment; if (map_and_fenceable && (alignment & (fence_alignment - 1)) != 0) { DRM_ERROR("Invalid object alignment requested %u\n", alignment); return (-EINVAL); } size = map_and_fenceable ? fence_size : obj->base.size; /* If the object is bigger than the entire aperture, reject it early * before evicting everything in a vain attempt to find space. */ if (obj->base.size > (map_and_fenceable ? dev_priv->mm.gtt_mappable_end : dev_priv->mm.gtt_total)) { DRM_ERROR( "Attempting to bind an object larger than the aperture\n"); return (-E2BIG); } search_free: if (map_and_fenceable) free_space = drm_mm_search_free_in_range( &dev_priv->mm.gtt_space, size, alignment, 0, dev_priv->mm.gtt_mappable_end, 0); else free_space = drm_mm_search_free(&dev_priv->mm.gtt_space, size, alignment, 0); if (free_space != NULL) { if (map_and_fenceable) obj->gtt_space = drm_mm_get_block_range_generic( free_space, size, alignment, 0, 0, dev_priv->mm.gtt_mappable_end, 1); else obj->gtt_space = drm_mm_get_block_generic(free_space, size, alignment, 0, 1); } if (obj->gtt_space == NULL) { ret = i915_gem_evict_something(dev, size, alignment, map_and_fenceable); if (ret != 0) return (ret); goto search_free; } ret = i915_gem_object_get_pages_gtt(obj, 0); if (ret != 0) { drm_mm_put_block(obj->gtt_space); obj->gtt_space = NULL; /* * i915_gem_object_get_pages_gtt() cannot return * ENOMEM, since we use vm_page_grab(). */ return (ret); } ret = i915_gem_gtt_prepare_object(obj); if (ret != 0) { i915_gem_object_put_pages_gtt(obj); drm_mm_put_block(obj->gtt_space); obj->gtt_space = NULL; if (i915_gem_evict_everything(dev, false)) return (ret); goto search_free; } if (!dev_priv->mm.aliasing_ppgtt) i915_gem_gtt_bind_object(obj, obj->cache_level); list_add_tail(&obj->gtt_list, &dev_priv->mm.gtt_list); list_add_tail(&obj->mm_list, &dev_priv->mm.inactive_list); KASSERT((obj->base.read_domains & I915_GEM_GPU_DOMAINS) == 0, ("Object in gpu read domain")); KASSERT((obj->base.write_domain & I915_GEM_GPU_DOMAINS) == 0, ("Object in gpu write domain")); obj->gtt_offset = obj->gtt_space->start; fenceable = obj->gtt_space->size == fence_size && (obj->gtt_space->start & (fence_alignment - 1)) == 0; mappable = obj->gtt_offset + obj->base.size <= dev_priv->mm.gtt_mappable_end; obj->map_and_fenceable = mappable && fenceable; CTR4(KTR_DRM, "object_bind %p %x %x %d", obj, obj->gtt_offset, obj->base.size, map_and_fenceable); return (0); } int i915_gem_object_sync(struct drm_i915_gem_object *obj, struct intel_ring_buffer *to) { struct intel_ring_buffer *from = obj->ring; u32 seqno; int ret, idx; if (from == NULL || to == from) return 0; if (to == NULL || !i915_semaphore_is_enabled(obj->base.dev)) return i915_gem_object_wait_rendering(obj); idx = intel_ring_sync_index(from, to); seqno = obj->last_rendering_seqno; if (seqno <= from->sync_seqno[idx]) return 0; if (seqno == from->outstanding_lazy_request) { struct drm_i915_gem_request *request; request = malloc(sizeof(*request), DRM_I915_GEM, M_WAITOK | M_ZERO); ret = i915_add_request(from, NULL, request); if (ret) { free(request, DRM_I915_GEM); return ret; } seqno = request->seqno; } ret = to->sync_to(to, from, seqno); if (!ret) from->sync_seqno[idx] = seqno; return ret; } static void i915_gem_object_finish_gtt(struct drm_i915_gem_object *obj) { u32 old_write_domain, old_read_domains; /* Act a barrier for all accesses through the GTT */ mb(); /* Force a pagefault for domain tracking on next user access */ i915_gem_release_mmap(obj); if ((obj->base.read_domains & I915_GEM_DOMAIN_GTT) == 0) return; old_read_domains = obj->base.read_domains; old_write_domain = obj->base.write_domain; obj->base.read_domains &= ~I915_GEM_DOMAIN_GTT; obj->base.write_domain &= ~I915_GEM_DOMAIN_GTT; CTR3(KTR_DRM, "object_change_domain finish gtt %p %x %x", obj, old_read_domains, old_write_domain); } int i915_gem_object_unbind(struct drm_i915_gem_object *obj) { drm_i915_private_t *dev_priv; int ret; dev_priv = obj->base.dev->dev_private; ret = 0; if (obj->gtt_space == NULL) return (0); if (obj->pin_count != 0) { DRM_ERROR("Attempting to unbind pinned buffer\n"); return (-EINVAL); } ret = i915_gem_object_finish_gpu(obj); if (ret == -ERESTARTSYS || ret == -EINTR) return (ret); i915_gem_object_finish_gtt(obj); if (ret == 0) ret = i915_gem_object_set_to_cpu_domain(obj, 1); if (ret == -ERESTARTSYS || ret == -EINTR) return (ret); if (ret != 0) { i915_gem_clflush_object(obj); obj->base.read_domains = obj->base.write_domain = I915_GEM_DOMAIN_CPU; } ret = i915_gem_object_put_fence(obj); if (ret) return (ret); if (obj->has_global_gtt_mapping) i915_gem_gtt_unbind_object(obj); if (obj->has_aliasing_ppgtt_mapping) { i915_ppgtt_unbind_object(dev_priv->mm.aliasing_ppgtt, obj); obj->has_aliasing_ppgtt_mapping = 0; } i915_gem_gtt_finish_object(obj); i915_gem_object_put_pages_gtt(obj); list_del_init(&obj->gtt_list); list_del_init(&obj->mm_list); obj->map_and_fenceable = true; drm_mm_put_block(obj->gtt_space); obj->gtt_space = NULL; obj->gtt_offset = 0; if (i915_gem_object_is_purgeable(obj)) i915_gem_object_truncate(obj); CTR1(KTR_DRM, "object_unbind %p", obj); return (ret); } static void i915_gem_object_put_pages_range_locked(struct drm_i915_gem_object *obj, vm_pindex_t si, vm_pindex_t ei) { vm_object_t vm_obj; vm_page_t m; vm_pindex_t i; vm_obj = obj->base.vm_obj; VM_OBJECT_ASSERT_LOCKED(vm_obj); for (i = si, m = vm_page_lookup(vm_obj, i); i < ei; m = vm_page_next(m), i++) { KASSERT(m->pindex == i, ("pindex %jx %jx", (uintmax_t)m->pindex, (uintmax_t)i)); vm_page_lock(m); vm_page_unwire(m, PQ_INACTIVE); if (m->wire_count == 0) atomic_add_long(&i915_gem_wired_pages_cnt, -1); vm_page_unlock(m); } } static void i915_gem_object_put_pages_range(struct drm_i915_gem_object *obj, off_t start, off_t end) { vm_object_t vm_obj; vm_obj = obj->base.vm_obj; VM_OBJECT_WLOCK(vm_obj); i915_gem_object_put_pages_range_locked(obj, OFF_TO_IDX(trunc_page(start)), OFF_TO_IDX(round_page(end))); VM_OBJECT_WUNLOCK(vm_obj); } static int i915_gem_object_get_pages_range(struct drm_i915_gem_object *obj, off_t start, off_t end) { vm_object_t vm_obj; vm_page_t m; vm_pindex_t si, ei, i; bool need_swizzle, fresh; need_swizzle = i915_gem_object_needs_bit17_swizzle(obj) != 0; vm_obj = obj->base.vm_obj; si = OFF_TO_IDX(trunc_page(start)); ei = OFF_TO_IDX(round_page(end)); VM_OBJECT_WLOCK(vm_obj); for (i = si; i < ei; i++) { m = i915_gem_wire_page(vm_obj, i, &fresh); if (m == NULL) goto failed; if (need_swizzle && fresh) i915_gem_object_do_bit_17_swizzle_page(obj, m); } VM_OBJECT_WUNLOCK(vm_obj); return (0); failed: i915_gem_object_put_pages_range_locked(obj, si, i); VM_OBJECT_WUNLOCK(vm_obj); return (-EIO); } static int i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj, int flags) { struct drm_device *dev; vm_object_t vm_obj; vm_page_t m; vm_pindex_t i, page_count; int res; dev = obj->base.dev; KASSERT(obj->pages == NULL, ("Obj already has pages")); page_count = OFF_TO_IDX(obj->base.size); obj->pages = malloc(page_count * sizeof(vm_page_t), DRM_I915_GEM, M_WAITOK); res = i915_gem_object_get_pages_range(obj, 0, obj->base.size); if (res != 0) { free(obj->pages, DRM_I915_GEM); obj->pages = NULL; return (res); } vm_obj = obj->base.vm_obj; VM_OBJECT_WLOCK(vm_obj); for (i = 0, m = vm_page_lookup(vm_obj, 0); i < page_count; i++, m = vm_page_next(m)) { KASSERT(m->pindex == i, ("pindex %jx %jx", (uintmax_t)m->pindex, (uintmax_t)i)); obj->pages[i] = m; } VM_OBJECT_WUNLOCK(vm_obj); return (0); } #define GEM_PARANOID_CHECK_GTT 0 #if GEM_PARANOID_CHECK_GTT static void i915_gem_assert_pages_not_mapped(struct drm_device *dev, vm_page_t *ma, int page_count) { struct drm_i915_private *dev_priv; vm_paddr_t pa; unsigned long start, end; u_int i; int j; dev_priv = dev->dev_private; start = OFF_TO_IDX(dev_priv->mm.gtt_start); end = OFF_TO_IDX(dev_priv->mm.gtt_end); for (i = start; i < end; i++) { pa = intel_gtt_read_pte_paddr(i); for (j = 0; j < page_count; j++) { if (pa == VM_PAGE_TO_PHYS(ma[j])) { panic("Page %p in GTT pte index %d pte %x", ma[i], i, intel_gtt_read_pte(i)); } } } } #endif static void i915_gem_object_put_pages_gtt(struct drm_i915_gem_object *obj) { vm_page_t m; int page_count, i; KASSERT(obj->madv != I915_MADV_PURGED_INTERNAL, ("Purged object")); if (obj->tiling_mode != I915_TILING_NONE) i915_gem_object_save_bit_17_swizzle(obj); if (obj->madv == I915_MADV_DONTNEED) obj->dirty = 0; page_count = obj->base.size / PAGE_SIZE; VM_OBJECT_WLOCK(obj->base.vm_obj); #if GEM_PARANOID_CHECK_GTT i915_gem_assert_pages_not_mapped(obj->base.dev, obj->pages, page_count); #endif for (i = 0; i < page_count; i++) { m = obj->pages[i]; if (obj->dirty) vm_page_dirty(m); if (obj->madv == I915_MADV_WILLNEED) vm_page_reference(m); vm_page_lock(m); vm_page_unwire(obj->pages[i], PQ_ACTIVE); vm_page_unlock(m); atomic_add_long(&i915_gem_wired_pages_cnt, -1); } VM_OBJECT_WUNLOCK(obj->base.vm_obj); obj->dirty = 0; free(obj->pages, DRM_I915_GEM); obj->pages = NULL; } void i915_gem_release_mmap(struct drm_i915_gem_object *obj) { vm_object_t devobj; vm_page_t m; int i, page_count; if (!obj->fault_mappable) return; CTR3(KTR_DRM, "release_mmap %p %x %x", obj, obj->gtt_offset, OFF_TO_IDX(obj->base.size)); devobj = cdev_pager_lookup(obj); if (devobj != NULL) { page_count = OFF_TO_IDX(obj->base.size); VM_OBJECT_WLOCK(devobj); retry: for (i = 0; i < page_count; i++) { m = vm_page_lookup(devobj, i); if (m == NULL) continue; if (vm_page_sleep_if_busy(m, "915unm")) goto retry; cdev_pager_free_page(devobj, m); } VM_OBJECT_WUNLOCK(devobj); vm_object_deallocate(devobj); } obj->fault_mappable = false; } int i915_gem_object_wait_rendering(struct drm_i915_gem_object *obj) { int ret; KASSERT((obj->base.write_domain & I915_GEM_GPU_DOMAINS) == 0, ("In GPU write domain")); CTR5(KTR_DRM, "object_wait_rendering %p %s %x %d %d", obj, obj->ring != NULL ? obj->ring->name : "none", obj->gtt_offset, obj->active, obj->last_rendering_seqno); if (obj->active) { ret = i915_wait_request(obj->ring, obj->last_rendering_seqno); if (ret != 0) return (ret); i915_gem_retire_requests_ring(obj->ring); } return (0); } void i915_gem_object_move_to_active(struct drm_i915_gem_object *obj, struct intel_ring_buffer *ring, uint32_t seqno) { struct drm_device *dev = obj->base.dev; struct drm_i915_private *dev_priv = dev->dev_private; struct drm_i915_fence_reg *reg; obj->ring = ring; KASSERT(ring != NULL, ("NULL ring")); /* Add a reference if we're newly entering the active list. */ if (!obj->active) { drm_gem_object_reference(&obj->base); obj->active = 1; } /* Move from whatever list we were on to the tail of execution. */ list_move_tail(&obj->mm_list, &dev_priv->mm.active_list); list_move_tail(&obj->ring_list, &ring->active_list); obj->last_rendering_seqno = seqno; if (obj->fenced_gpu_access) { obj->last_fenced_seqno = seqno; /* Bump MRU to take account of the delayed flush */ if (obj->fence_reg != I915_FENCE_REG_NONE) { reg = &dev_priv->fence_regs[obj->fence_reg]; list_move_tail(®->lru_list, &dev_priv->mm.fence_list); } } } static void i915_gem_object_move_off_active(struct drm_i915_gem_object *obj) { list_del_init(&obj->ring_list); obj->last_rendering_seqno = 0; obj->last_fenced_seqno = 0; } static void i915_gem_object_move_to_flushing(struct drm_i915_gem_object *obj) { struct drm_device *dev = obj->base.dev; drm_i915_private_t *dev_priv = dev->dev_private; KASSERT(obj->active, ("Object not active")); list_move_tail(&obj->mm_list, &dev_priv->mm.flushing_list); i915_gem_object_move_off_active(obj); } static void i915_gem_object_move_to_inactive(struct drm_i915_gem_object *obj) { struct drm_device *dev = obj->base.dev; struct drm_i915_private *dev_priv = dev->dev_private; list_move_tail(&obj->mm_list, &dev_priv->mm.inactive_list); KASSERT(list_empty(&obj->gpu_write_list), ("On gpu_write_list")); KASSERT(obj->active, ("Object not active")); obj->ring = NULL; i915_gem_object_move_off_active(obj); obj->fenced_gpu_access = false; obj->active = 0; obj->pending_gpu_write = false; drm_gem_object_unreference(&obj->base); #if 1 KIB_NOTYET(); #else WARN_ON(i915_verify_lists(dev)); #endif } static void i915_gem_object_truncate(struct drm_i915_gem_object *obj) { vm_object_t vm_obj; vm_obj = obj->base.vm_obj; VM_OBJECT_WLOCK(vm_obj); vm_object_page_remove(vm_obj, 0, 0, false); VM_OBJECT_WUNLOCK(vm_obj); drm_gem_free_mmap_offset(&obj->base); obj->madv = I915_MADV_PURGED_INTERNAL; } static inline int i915_gem_object_is_purgeable(struct drm_i915_gem_object *obj) { return (obj->madv == I915_MADV_DONTNEED); } static void i915_gem_process_flushing_list(struct intel_ring_buffer *ring, uint32_t flush_domains) { struct drm_i915_gem_object *obj, *next; uint32_t old_write_domain; list_for_each_entry_safe(obj, next, &ring->gpu_write_list, gpu_write_list) { if (obj->base.write_domain & flush_domains) { old_write_domain = obj->base.write_domain; obj->base.write_domain = 0; list_del_init(&obj->gpu_write_list); i915_gem_object_move_to_active(obj, ring, i915_gem_next_request_seqno(ring)); CTR3(KTR_DRM, "object_change_domain process_flush %p %x %x", obj, obj->base.read_domains, old_write_domain); } } } static int i915_gem_object_needs_bit17_swizzle(struct drm_i915_gem_object *obj) { drm_i915_private_t *dev_priv; dev_priv = obj->base.dev->dev_private; return (dev_priv->mm.bit_6_swizzle_x == I915_BIT_6_SWIZZLE_9_10_17 && obj->tiling_mode != I915_TILING_NONE); } static vm_page_t i915_gem_wire_page(vm_object_t object, vm_pindex_t pindex, bool *fresh) { vm_page_t m; int rv; VM_OBJECT_ASSERT_WLOCKED(object); m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL); if (m->valid != VM_PAGE_BITS_ALL) { if (vm_pager_has_page(object, pindex, NULL, NULL)) { rv = vm_pager_get_pages(object, &m, 1, 0); m = vm_page_lookup(object, pindex); if (m == NULL) return (NULL); if (rv != VM_PAGER_OK) { vm_page_lock(m); vm_page_free(m); vm_page_unlock(m); return (NULL); } if (fresh != NULL) *fresh = true; } else { pmap_zero_page(m); m->valid = VM_PAGE_BITS_ALL; m->dirty = 0; if (fresh != NULL) *fresh = false; } } else if (fresh != NULL) { *fresh = false; } vm_page_lock(m); vm_page_wire(m); vm_page_unlock(m); vm_page_xunbusy(m); atomic_add_long(&i915_gem_wired_pages_cnt, 1); return (m); } int i915_gem_flush_ring(struct intel_ring_buffer *ring, uint32_t invalidate_domains, uint32_t flush_domains) { int ret; if (((invalidate_domains | flush_domains) & I915_GEM_GPU_DOMAINS) == 0) return 0; CTR3(KTR_DRM, "ring_flush %s %x %x", ring->name, invalidate_domains, flush_domains); ret = ring->flush(ring, invalidate_domains, flush_domains); if (ret) return ret; if (flush_domains & I915_GEM_GPU_DOMAINS) i915_gem_process_flushing_list(ring, flush_domains); return 0; } static int i915_ring_idle(struct intel_ring_buffer *ring) { int ret; if (list_empty(&ring->gpu_write_list) && list_empty(&ring->active_list)) return 0; if (!list_empty(&ring->gpu_write_list)) { ret = i915_gem_flush_ring(ring, I915_GEM_GPU_DOMAINS, I915_GEM_GPU_DOMAINS); if (ret != 0) return ret; } return (i915_wait_request(ring, i915_gem_next_request_seqno(ring))); } int i915_gpu_idle(struct drm_device *dev) { drm_i915_private_t *dev_priv = dev->dev_private; struct intel_ring_buffer *ring; int ret, i; /* Flush everything onto the inactive list. */ for_each_ring(ring, dev_priv, i) { ret = i915_switch_context(ring, NULL, DEFAULT_CONTEXT_ID); if (ret) return ret; ret = i915_ring_idle(ring); if (ret) return ret; /* Is the device fubar? */ if (!list_empty(&ring->gpu_write_list)) return -EBUSY; } return 0; } static int i915_gem_check_wedge(struct drm_i915_private *dev_priv) { DRM_LOCK_ASSERT(dev_priv->dev); if (atomic_load_acq_int(&dev_priv->mm.wedged) != 0) { bool recovery_complete; /* Give the error handler a chance to run. */ mtx_lock(&dev_priv->error_completion_lock); recovery_complete = (&dev_priv->error_completion) > 0; mtx_unlock(&dev_priv->error_completion_lock); return (recovery_complete ? -EIO : -EAGAIN); } return 0; } /* * Compare seqno against outstanding lazy request. Emit a request if they are * equal. */ static int i915_gem_check_olr(struct intel_ring_buffer *ring, u32 seqno) { int ret = 0; DRM_LOCK_ASSERT(ring->dev); if (seqno == ring->outstanding_lazy_request) { struct drm_i915_gem_request *request; request = malloc(sizeof(*request), DRM_I915_GEM, M_WAITOK | M_ZERO); ret = i915_add_request(ring, NULL, request); if (ret != 0) { free(request, DRM_I915_GEM); return (ret); } MPASS(seqno == request->seqno); } return ret; } static int __wait_seqno(struct intel_ring_buffer *ring, u32 seqno, bool interruptible) { drm_i915_private_t *dev_priv = ring->dev->dev_private; int ret = 0, flags; if (i915_seqno_passed(ring->get_seqno(ring), seqno)) return 0; CTR2(KTR_DRM, "request_wait_begin %s %d", ring->name, seqno); mtx_lock(&dev_priv->irq_lock); if (!ring->irq_get(ring)) { mtx_unlock(&dev_priv->irq_lock); return (-ENODEV); } flags = interruptible ? PCATCH : 0; while (!i915_seqno_passed(ring->get_seqno(ring), seqno) && !atomic_load_acq_int(&dev_priv->mm.wedged) && ret == 0) { ret = -msleep(ring, &dev_priv->irq_lock, flags, "915gwr", 0); if (ret == -ERESTART) ret = -ERESTARTSYS; } ring->irq_put(ring); mtx_unlock(&dev_priv->irq_lock); CTR3(KTR_DRM, "request_wait_end %s %d %d", ring->name, seqno, ret); return ret; } int i915_wait_request(struct intel_ring_buffer *ring, uint32_t seqno) { drm_i915_private_t *dev_priv; int ret; KASSERT(seqno != 0, ("Zero seqno")); dev_priv = ring->dev->dev_private; ret = 0; ret = i915_gem_check_wedge(dev_priv); if (ret) return ret; ret = i915_gem_check_olr(ring, seqno); if (ret) return ret; ret = __wait_seqno(ring, seqno, dev_priv->mm.interruptible); if (atomic_load_acq_int(&dev_priv->mm.wedged)) ret = -EAGAIN; return (ret); } static u32 i915_gem_get_seqno(struct drm_device *dev) { drm_i915_private_t *dev_priv = dev->dev_private; u32 seqno = dev_priv->next_seqno; /* reserve 0 for non-seqno */ if (++dev_priv->next_seqno == 0) dev_priv->next_seqno = 1; return seqno; } u32 i915_gem_next_request_seqno(struct intel_ring_buffer *ring) { if (ring->outstanding_lazy_request == 0) ring->outstanding_lazy_request = i915_gem_get_seqno(ring->dev); return ring->outstanding_lazy_request; } int i915_add_request(struct intel_ring_buffer *ring, struct drm_file *file, struct drm_i915_gem_request *request) { drm_i915_private_t *dev_priv; struct drm_i915_file_private *file_priv; uint32_t seqno; u32 request_ring_position; int was_empty; int ret; KASSERT(request != NULL, ("NULL request in add")); DRM_LOCK_ASSERT(ring->dev); dev_priv = ring->dev->dev_private; seqno = i915_gem_next_request_seqno(ring); request_ring_position = intel_ring_get_tail(ring); ret = ring->add_request(ring, &seqno); if (ret != 0) return ret; CTR2(KTR_DRM, "request_add %s %d", ring->name, seqno); request->seqno = seqno; request->ring = ring; request->tail = request_ring_position; request->emitted_jiffies = ticks; was_empty = list_empty(&ring->request_list); list_add_tail(&request->list, &ring->request_list); if (file != NULL) { file_priv = file->driver_priv; mtx_lock(&file_priv->mm.lck); request->file_priv = file_priv; list_add_tail(&request->client_list, &file_priv->mm.request_list); mtx_unlock(&file_priv->mm.lck); } ring->outstanding_lazy_request = 0; if (!dev_priv->mm.suspended) { if (i915_enable_hangcheck) { callout_schedule(&dev_priv->hangcheck_timer, DRM_I915_HANGCHECK_PERIOD); } if (was_empty) taskqueue_enqueue_timeout(dev_priv->tq, &dev_priv->mm.retire_task, hz); } return (0); } static inline void i915_gem_request_remove_from_client(struct drm_i915_gem_request *request) { struct drm_i915_file_private *file_priv = request->file_priv; if (!file_priv) return; DRM_LOCK_ASSERT(request->ring->dev); mtx_lock(&file_priv->mm.lck); if (request->file_priv != NULL) { list_del(&request->client_list); request->file_priv = NULL; } mtx_unlock(&file_priv->mm.lck); } void i915_gem_release(struct drm_device *dev, struct drm_file *file) { struct drm_i915_file_private *file_priv; struct drm_i915_gem_request *request; file_priv = file->driver_priv; /* Clean up our request list when the client is going away, so that * later retire_requests won't dereference our soon-to-be-gone * file_priv. */ mtx_lock(&file_priv->mm.lck); while (!list_empty(&file_priv->mm.request_list)) { request = list_first_entry(&file_priv->mm.request_list, struct drm_i915_gem_request, client_list); list_del(&request->client_list); request->file_priv = NULL; } mtx_unlock(&file_priv->mm.lck); } static void i915_gem_reset_ring_lists(struct drm_i915_private *dev_priv, struct intel_ring_buffer *ring) { if (ring->dev != NULL) DRM_LOCK_ASSERT(ring->dev); while (!list_empty(&ring->request_list)) { struct drm_i915_gem_request *request; request = list_first_entry(&ring->request_list, struct drm_i915_gem_request, list); list_del(&request->list); i915_gem_request_remove_from_client(request); free(request, DRM_I915_GEM); } while (!list_empty(&ring->active_list)) { struct drm_i915_gem_object *obj; obj = list_first_entry(&ring->active_list, struct drm_i915_gem_object, ring_list); obj->base.write_domain = 0; list_del_init(&obj->gpu_write_list); i915_gem_object_move_to_inactive(obj); } } static void i915_gem_reset_fences(struct drm_device *dev) { struct drm_i915_private *dev_priv = dev->dev_private; int i; for (i = 0; i < dev_priv->num_fence_regs; i++) { struct drm_i915_fence_reg *reg = &dev_priv->fence_regs[i]; i915_gem_write_fence(dev, i, NULL); if (reg->obj) i915_gem_object_fence_lost(reg->obj); reg->pin_count = 0; reg->obj = NULL; INIT_LIST_HEAD(®->lru_list); } INIT_LIST_HEAD(&dev_priv->mm.fence_list); } void i915_gem_reset(struct drm_device *dev) { struct drm_i915_private *dev_priv = dev->dev_private; struct drm_i915_gem_object *obj; struct intel_ring_buffer *ring; int i; for_each_ring(ring, dev_priv, i) i915_gem_reset_ring_lists(dev_priv, ring); /* Remove anything from the flushing lists. The GPU cache is likely * to be lost on reset along with the data, so simply move the * lost bo to the inactive list. */ while (!list_empty(&dev_priv->mm.flushing_list)) { obj = list_first_entry(&dev_priv->mm.flushing_list, struct drm_i915_gem_object, mm_list); obj->base.write_domain = 0; list_del_init(&obj->gpu_write_list); i915_gem_object_move_to_inactive(obj); } /* Move everything out of the GPU domains to ensure we do any * necessary invalidation upon reuse. */ list_for_each_entry(obj, &dev_priv->mm.inactive_list, mm_list) { obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS; } /* The fence registers are invalidated so clear them out */ i915_gem_reset_fences(dev); } /** * This function clears the request list as sequence numbers are passed. */ void i915_gem_retire_requests_ring(struct intel_ring_buffer *ring) { uint32_t seqno; int i; if (list_empty(&ring->request_list)) return; seqno = ring->get_seqno(ring); CTR2(KTR_DRM, "retire_request_ring %s %d", ring->name, seqno); for (i = 0; i < ARRAY_SIZE(ring->sync_seqno); i++) if (seqno >= ring->sync_seqno[i]) ring->sync_seqno[i] = 0; while (!list_empty(&ring->request_list)) { struct drm_i915_gem_request *request; request = list_first_entry(&ring->request_list, struct drm_i915_gem_request, list); if (!i915_seqno_passed(seqno, request->seqno)) break; CTR2(KTR_DRM, "retire_request_seqno_passed %s %d", ring->name, seqno); ring->last_retired_head = request->tail; list_del(&request->list); i915_gem_request_remove_from_client(request); free(request, DRM_I915_GEM); } /* Move any buffers on the active list that are no longer referenced * by the ringbuffer to the flushing/inactive lists as appropriate. */ while (!list_empty(&ring->active_list)) { struct drm_i915_gem_object *obj; obj = list_first_entry(&ring->active_list, struct drm_i915_gem_object, ring_list); if (!i915_seqno_passed(seqno, obj->last_rendering_seqno)) break; if (obj->base.write_domain != 0) i915_gem_object_move_to_flushing(obj); else i915_gem_object_move_to_inactive(obj); } if (ring->trace_irq_seqno && i915_seqno_passed(seqno, ring->trace_irq_seqno)) { struct drm_i915_private *dev_priv = ring->dev->dev_private; mtx_lock(&dev_priv->irq_lock); ring->irq_put(ring); mtx_unlock(&dev_priv->irq_lock); ring->trace_irq_seqno = 0; } } void i915_gem_retire_requests(struct drm_device *dev) { drm_i915_private_t *dev_priv = dev->dev_private; struct intel_ring_buffer *ring; int i; for_each_ring(ring, dev_priv, i) i915_gem_retire_requests_ring(ring); } static void sandybridge_write_fence_reg(struct drm_device *dev, int reg, struct drm_i915_gem_object *obj) { drm_i915_private_t *dev_priv = dev->dev_private; uint64_t val; if (obj) { u32 size = obj->gtt_space->size; val = (uint64_t)((obj->gtt_offset + size - 4096) & 0xfffff000) << 32; val |= obj->gtt_offset & 0xfffff000; val |= (uint64_t)((obj->stride / 128) - 1) << SANDYBRIDGE_FENCE_PITCH_SHIFT; if (obj->tiling_mode == I915_TILING_Y) val |= 1 << I965_FENCE_TILING_Y_SHIFT; val |= I965_FENCE_REG_VALID; } else val = 0; I915_WRITE64(FENCE_REG_SANDYBRIDGE_0 + reg * 8, val); POSTING_READ(FENCE_REG_SANDYBRIDGE_0 + reg * 8); } static void i965_write_fence_reg(struct drm_device *dev, int reg, struct drm_i915_gem_object *obj) { drm_i915_private_t *dev_priv = dev->dev_private; uint64_t val; if (obj) { u32 size = obj->gtt_space->size; val = (uint64_t)((obj->gtt_offset + size - 4096) & 0xfffff000) << 32; val |= obj->gtt_offset & 0xfffff000; val |= ((obj->stride / 128) - 1) << I965_FENCE_PITCH_SHIFT; if (obj->tiling_mode == I915_TILING_Y) val |= 1 << I965_FENCE_TILING_Y_SHIFT; val |= I965_FENCE_REG_VALID; } else val = 0; I915_WRITE64(FENCE_REG_965_0 + reg * 8, val); POSTING_READ(FENCE_REG_965_0 + reg * 8); } static void i915_write_fence_reg(struct drm_device *dev, int reg, struct drm_i915_gem_object *obj) { drm_i915_private_t *dev_priv = dev->dev_private; u32 val; if (obj) { u32 size = obj->gtt_space->size; int pitch_val; int tile_width; if ((obj->gtt_offset & ~I915_FENCE_START_MASK) || (size & -size) != size || (obj->gtt_offset & (size - 1))) printf( "object 0x%08x [fenceable? %d] not 1M or pot-size (0x%08x) aligned\n", obj->gtt_offset, obj->map_and_fenceable, size); if (obj->tiling_mode == I915_TILING_Y && HAS_128_BYTE_Y_TILING(dev)) tile_width = 128; else tile_width = 512; /* Note: pitch better be a power of two tile widths */ pitch_val = obj->stride / tile_width; pitch_val = ffs(pitch_val) - 1; val = obj->gtt_offset; if (obj->tiling_mode == I915_TILING_Y) val |= 1 << I830_FENCE_TILING_Y_SHIFT; val |= I915_FENCE_SIZE_BITS(size); val |= pitch_val << I830_FENCE_PITCH_SHIFT; val |= I830_FENCE_REG_VALID; } else val = 0; if (reg < 8) reg = FENCE_REG_830_0 + reg * 4; else reg = FENCE_REG_945_8 + (reg - 8) * 4; I915_WRITE(reg, val); POSTING_READ(reg); } static void i830_write_fence_reg(struct drm_device *dev, int reg, struct drm_i915_gem_object *obj) { drm_i915_private_t *dev_priv = dev->dev_private; uint32_t val; if (obj) { u32 size = obj->gtt_space->size; uint32_t pitch_val; if ((obj->gtt_offset & ~I830_FENCE_START_MASK) || (size & -size) != size || (obj->gtt_offset & (size - 1))) printf( "object 0x%08x not 512K or pot-size 0x%08x aligned\n", obj->gtt_offset, size); pitch_val = obj->stride / 128; pitch_val = ffs(pitch_val) - 1; val = obj->gtt_offset; if (obj->tiling_mode == I915_TILING_Y) val |= 1 << I830_FENCE_TILING_Y_SHIFT; val |= I830_FENCE_SIZE_BITS(size); val |= pitch_val << I830_FENCE_PITCH_SHIFT; val |= I830_FENCE_REG_VALID; } else val = 0; I915_WRITE(FENCE_REG_830_0 + reg * 4, val); POSTING_READ(FENCE_REG_830_0 + reg * 4); } static void i915_gem_write_fence(struct drm_device *dev, int reg, struct drm_i915_gem_object *obj) { switch (INTEL_INFO(dev)->gen) { case 7: case 6: sandybridge_write_fence_reg(dev, reg, obj); break; case 5: case 4: i965_write_fence_reg(dev, reg, obj); break; case 3: i915_write_fence_reg(dev, reg, obj); break; case 2: i830_write_fence_reg(dev, reg, obj); break; default: break; } } static inline int fence_number(struct drm_i915_private *dev_priv, struct drm_i915_fence_reg *fence) { return fence - dev_priv->fence_regs; } static void i915_gem_object_update_fence(struct drm_i915_gem_object *obj, struct drm_i915_fence_reg *fence, bool enable) { struct drm_i915_private *dev_priv = obj->base.dev->dev_private; int reg = fence_number(dev_priv, fence); i915_gem_write_fence(obj->base.dev, reg, enable ? obj : NULL); if (enable) { obj->fence_reg = reg; fence->obj = obj; list_move_tail(&fence->lru_list, &dev_priv->mm.fence_list); } else { obj->fence_reg = I915_FENCE_REG_NONE; fence->obj = NULL; list_del_init(&fence->lru_list); } } static int i915_gem_object_flush_fence(struct drm_i915_gem_object *obj) { int ret; if (obj->fenced_gpu_access) { if (obj->base.write_domain & I915_GEM_GPU_DOMAINS) { ret = i915_gem_flush_ring(obj->ring, 0, obj->base.write_domain); if (ret) return ret; } obj->fenced_gpu_access = false; } if (obj->last_fenced_seqno) { ret = i915_wait_request(obj->ring, obj->last_fenced_seqno); if (ret) return ret; obj->last_fenced_seqno = 0; } /* Ensure that all CPU reads are completed before installing a fence * and all writes before removing the fence. */ if (obj->base.read_domains & I915_GEM_DOMAIN_GTT) mb(); return 0; } int i915_gem_object_put_fence(struct drm_i915_gem_object *obj) { struct drm_i915_private *dev_priv = obj->base.dev->dev_private; int ret; ret = i915_gem_object_flush_fence(obj); if (ret) return ret; if (obj->fence_reg == I915_FENCE_REG_NONE) return 0; i915_gem_object_update_fence(obj, &dev_priv->fence_regs[obj->fence_reg], false); i915_gem_object_fence_lost(obj); return 0; } static struct drm_i915_fence_reg * i915_find_fence_reg(struct drm_device *dev) { struct drm_i915_private *dev_priv = dev->dev_private; struct drm_i915_fence_reg *reg, *avail; int i; /* First try to find a free reg */ avail = NULL; for (i = dev_priv->fence_reg_start; i < dev_priv->num_fence_regs; i++) { reg = &dev_priv->fence_regs[i]; if (!reg->obj) return reg; if (!reg->pin_count) avail = reg; } if (avail == NULL) return NULL; /* None available, try to steal one or wait for a user to finish */ list_for_each_entry(reg, &dev_priv->mm.fence_list, lru_list) { if (reg->pin_count) continue; return reg; } return NULL; } int i915_gem_object_get_fence(struct drm_i915_gem_object *obj) { struct drm_device *dev = obj->base.dev; struct drm_i915_private *dev_priv = dev->dev_private; bool enable = obj->tiling_mode != I915_TILING_NONE; struct drm_i915_fence_reg *reg; int ret; /* Have we updated the tiling parameters upon the object and so * will need to serialise the write to the associated fence register? */ if (obj->fence_dirty) { ret = i915_gem_object_flush_fence(obj); if (ret) return ret; } ret = 0; if (obj->fence_reg != I915_FENCE_REG_NONE) { reg = &dev_priv->fence_regs[obj->fence_reg]; if (!obj->fence_dirty) { list_move_tail(®->lru_list, &dev_priv->mm.fence_list); return 0; } } else if (enable) { reg = i915_find_fence_reg(dev); if (reg == NULL) return -EDEADLK; if (reg->obj) { struct drm_i915_gem_object *old = reg->obj; ret = i915_gem_object_flush_fence(old); if (ret) return ret; i915_gem_object_fence_lost(old); } } else return 0; i915_gem_object_update_fence(obj, reg, enable); obj->fence_dirty = false; return 0; } int i915_gem_init_object(struct drm_gem_object *obj) { printf("i915_gem_init_object called\n"); return (0); } static bool i915_gem_object_is_inactive(struct drm_i915_gem_object *obj) { return !obj->active; } static void i915_gem_retire_task_handler(void *arg, int pending) { drm_i915_private_t *dev_priv; struct drm_device *dev; struct intel_ring_buffer *ring; bool idle; int i; dev_priv = arg; dev = dev_priv->dev; /* Come back later if the device is busy... */ if (!sx_try_xlock(&dev->dev_struct_lock)) { taskqueue_enqueue_timeout(dev_priv->tq, &dev_priv->mm.retire_task, hz); return; } CTR0(KTR_DRM, "retire_task"); i915_gem_retire_requests(dev); /* Send a periodic flush down the ring so we don't hold onto GEM * objects indefinitely. */ idle = true; for_each_ring(ring, dev_priv, i) { struct intel_ring_buffer *ring = &dev_priv->rings[i]; if (!list_empty(&ring->gpu_write_list)) { struct drm_i915_gem_request *request; int ret; ret = i915_gem_flush_ring(ring, 0, I915_GEM_GPU_DOMAINS); request = malloc(sizeof(*request), DRM_I915_GEM, M_WAITOK | M_ZERO); if (ret || request == NULL || i915_add_request(ring, NULL, request)) free(request, DRM_I915_GEM); } idle &= list_empty(&ring->request_list); } if (!dev_priv->mm.suspended && !idle) taskqueue_enqueue_timeout(dev_priv->tq, &dev_priv->mm.retire_task, hz); DRM_UNLOCK(dev); } void i915_gem_lastclose(struct drm_device *dev) { int ret; if (drm_core_check_feature(dev, DRIVER_MODESET)) return; ret = i915_gem_idle(dev); if (ret != 0) DRM_ERROR("failed to idle hardware: %d\n", ret); } static int i915_gem_init_phys_object(struct drm_device *dev, int id, int size, int align) { drm_i915_private_t *dev_priv; struct drm_i915_gem_phys_object *phys_obj; int ret; dev_priv = dev->dev_private; if (dev_priv->mm.phys_objs[id - 1] != NULL || size == 0) return (0); phys_obj = malloc(sizeof(struct drm_i915_gem_phys_object), DRM_I915_GEM, M_WAITOK | M_ZERO); phys_obj->id = id; phys_obj->handle = drm_pci_alloc(dev, size, align, BUS_SPACE_MAXADDR); if (phys_obj->handle == NULL) { ret = -ENOMEM; goto free_obj; } pmap_change_attr((vm_offset_t)phys_obj->handle->vaddr, size / PAGE_SIZE, PAT_WRITE_COMBINING); dev_priv->mm.phys_objs[id - 1] = phys_obj; return (0); free_obj: free(phys_obj, DRM_I915_GEM); return (ret); } static void i915_gem_free_phys_object(struct drm_device *dev, int id) { drm_i915_private_t *dev_priv; struct drm_i915_gem_phys_object *phys_obj; dev_priv = dev->dev_private; if (dev_priv->mm.phys_objs[id - 1] == NULL) return; phys_obj = dev_priv->mm.phys_objs[id - 1]; if (phys_obj->cur_obj != NULL) i915_gem_detach_phys_object(dev, phys_obj->cur_obj); drm_pci_free(dev, phys_obj->handle); free(phys_obj, DRM_I915_GEM); dev_priv->mm.phys_objs[id - 1] = NULL; } void i915_gem_free_all_phys_object(struct drm_device *dev) { int i; for (i = I915_GEM_PHYS_CURSOR_0; i <= I915_MAX_PHYS_OBJECT; i++) i915_gem_free_phys_object(dev, i); } void i915_gem_detach_phys_object(struct drm_device *dev, struct drm_i915_gem_object *obj) { vm_page_t m; struct sf_buf *sf; char *vaddr, *dst; int i, page_count; if (obj->phys_obj == NULL) return; vaddr = obj->phys_obj->handle->vaddr; page_count = obj->base.size / PAGE_SIZE; VM_OBJECT_WLOCK(obj->base.vm_obj); for (i = 0; i < page_count; i++) { m = i915_gem_wire_page(obj->base.vm_obj, i, NULL); if (m == NULL) continue; /* XXX */ VM_OBJECT_WUNLOCK(obj->base.vm_obj); sf = sf_buf_alloc(m, 0); if (sf != NULL) { dst = (char *)sf_buf_kva(sf); memcpy(dst, vaddr + IDX_TO_OFF(i), PAGE_SIZE); sf_buf_free(sf); } drm_clflush_pages(&m, 1); VM_OBJECT_WLOCK(obj->base.vm_obj); vm_page_reference(m); vm_page_lock(m); vm_page_dirty(m); vm_page_unwire(m, PQ_INACTIVE); vm_page_unlock(m); atomic_add_long(&i915_gem_wired_pages_cnt, -1); } VM_OBJECT_WUNLOCK(obj->base.vm_obj); intel_gtt_chipset_flush(); obj->phys_obj->cur_obj = NULL; obj->phys_obj = NULL; } int i915_gem_attach_phys_object(struct drm_device *dev, struct drm_i915_gem_object *obj, int id, int align) { drm_i915_private_t *dev_priv; vm_page_t m; struct sf_buf *sf; char *dst, *src; int i, page_count, ret; if (id > I915_MAX_PHYS_OBJECT) return (-EINVAL); if (obj->phys_obj != NULL) { if (obj->phys_obj->id == id) return (0); i915_gem_detach_phys_object(dev, obj); } dev_priv = dev->dev_private; if (dev_priv->mm.phys_objs[id - 1] == NULL) { ret = i915_gem_init_phys_object(dev, id, obj->base.size, align); if (ret != 0) { DRM_ERROR("failed to init phys object %d size: %zu\n", id, obj->base.size); return (ret); } } /* bind to the object */ obj->phys_obj = dev_priv->mm.phys_objs[id - 1]; obj->phys_obj->cur_obj = obj; page_count = obj->base.size / PAGE_SIZE; VM_OBJECT_WLOCK(obj->base.vm_obj); ret = 0; for (i = 0; i < page_count; i++) { m = i915_gem_wire_page(obj->base.vm_obj, i, NULL); if (m == NULL) { ret = -EIO; break; } VM_OBJECT_WUNLOCK(obj->base.vm_obj); sf = sf_buf_alloc(m, 0); src = (char *)sf_buf_kva(sf); dst = (char *)obj->phys_obj->handle->vaddr + IDX_TO_OFF(i); memcpy(dst, src, PAGE_SIZE); sf_buf_free(sf); VM_OBJECT_WLOCK(obj->base.vm_obj); vm_page_reference(m); vm_page_lock(m); vm_page_unwire(m, PQ_INACTIVE); vm_page_unlock(m); atomic_add_long(&i915_gem_wired_pages_cnt, -1); } VM_OBJECT_WUNLOCK(obj->base.vm_obj); return (ret); } static int i915_gpu_is_active(struct drm_device *dev) { drm_i915_private_t *dev_priv; dev_priv = dev->dev_private; return (!list_empty(&dev_priv->mm.flushing_list) || !list_empty(&dev_priv->mm.active_list)); } static void i915_gem_lowmem(void *arg) { struct drm_device *dev; struct drm_i915_private *dev_priv; struct drm_i915_gem_object *obj, *next; int cnt, cnt_fail, cnt_total; dev = arg; dev_priv = dev->dev_private; if (!sx_try_xlock(&dev->dev_struct_lock)) return; CTR0(KTR_DRM, "gem_lowmem"); rescan: /* first scan for clean buffers */ i915_gem_retire_requests(dev); cnt_total = cnt_fail = cnt = 0; list_for_each_entry_safe(obj, next, &dev_priv->mm.inactive_list, mm_list) { if (i915_gem_object_is_purgeable(obj)) { if (i915_gem_object_unbind(obj) != 0) cnt_total++; } else cnt_total++; } /* second pass, evict/count anything still on the inactive list */ list_for_each_entry_safe(obj, next, &dev_priv->mm.inactive_list, mm_list) { if (i915_gem_object_unbind(obj) == 0) cnt++; else cnt_fail++; } if (cnt_fail > cnt_total / 100 && i915_gpu_is_active(dev)) { /* * We are desperate for pages, so as a last resort, wait * for the GPU to finish and discard whatever we can. * This has a dramatic impact to reduce the number of * OOM-killer events whilst running the GPU aggressively. */ if (i915_gpu_idle(dev) == 0) goto rescan; } DRM_UNLOCK(dev); } void i915_gem_unload(struct drm_device *dev) { struct drm_i915_private *dev_priv; dev_priv = dev->dev_private; EVENTHANDLER_DEREGISTER(vm_lowmem, dev_priv->mm.i915_lowmem); } Index: head/sys/fs/fdescfs/fdesc_vfsops.c =================================================================== --- head/sys/fs/fdescfs/fdesc_vfsops.c (revision 284214) +++ head/sys/fs/fdescfs/fdesc_vfsops.c (revision 284215) @@ -1,245 +1,243 @@ /*- * Copyright (c) 1992, 1993, 1995 * The Regents of the University of California. All rights reserved. * * This code is derived from software donated to Berkeley by * Jan-Simon Pendry. * * 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. * * @(#)fdesc_vfsops.c 8.4 (Berkeley) 1/21/94 * * $FreeBSD$ */ /* * /dev/fd Filesystem */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include static MALLOC_DEFINE(M_FDESCMNT, "fdesc_mount", "FDESC mount structure"); static vfs_cmount_t fdesc_cmount; static vfs_mount_t fdesc_mount; static vfs_unmount_t fdesc_unmount; static vfs_statfs_t fdesc_statfs; static vfs_root_t fdesc_root; /* * Compatibility shim for old mount(2) system call. */ int fdesc_cmount(struct mntarg *ma, void *data, uint64_t flags) { return kernel_mount(ma, flags); } /* * Mount the per-process file descriptors (/dev/fd) */ static int fdesc_mount(struct mount *mp) { int error = 0; struct fdescmount *fmp; struct thread *td = curthread; struct vnode *rvp; if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_FDESCFS)) return (EPERM); /* * Update is a no-op */ if (mp->mnt_flag & (MNT_UPDATE | MNT_ROOTFS)) return (EOPNOTSUPP); fmp = malloc(sizeof(struct fdescmount), M_FDESCMNT, M_WAITOK); /* XXX */ /* * We need to initialize a few bits of our local mount point struct to * avoid confusion in allocvp. */ mp->mnt_data = (qaddr_t) fmp; fmp->flags = 0; error = fdesc_allocvp(Froot, -1, FD_ROOT, mp, &rvp); if (error) { free(fmp, M_FDESCMNT); mp->mnt_data = NULL; return (error); } rvp->v_type = VDIR; rvp->v_vflag |= VV_ROOT; fmp->f_root = rvp; VOP_UNLOCK(rvp, 0); /* XXX -- don't mark as local to work around fts() problems */ /*mp->mnt_flag |= MNT_LOCAL;*/ vfs_getnewfsid(mp); vfs_mountedfrom(mp, "fdescfs"); return (0); } static int fdesc_unmount(mp, mntflags) struct mount *mp; int mntflags; { struct fdescmount *fmp; caddr_t data; int error; int flags = 0; fmp = (struct fdescmount *)mp->mnt_data; if (mntflags & MNT_FORCE) { /* The hash mutex protects the private mount flags. */ mtx_lock(&fdesc_hashmtx); fmp->flags |= FMNT_UNMOUNTF; mtx_unlock(&fdesc_hashmtx); flags |= FORCECLOSE; } /* * Clear out buffer cache. I don't think we * ever get anything cached at this level at the * moment, but who knows... * * There is 1 extra root vnode reference corresponding * to f_root. */ if ((error = vflush(mp, 1, flags, curthread)) != 0) return (error); /* * Finally, throw away the fdescmount structure. Hold the hashmtx to * protect the fdescmount structure. */ mtx_lock(&fdesc_hashmtx); data = mp->mnt_data; mp->mnt_data = NULL; mtx_unlock(&fdesc_hashmtx); free(data, M_FDESCMNT); /* XXX */ return (0); } static int fdesc_root(mp, flags, vpp) struct mount *mp; int flags; struct vnode **vpp; { struct vnode *vp; /* * Return locked reference to root. */ vp = VFSTOFDESC(mp)->f_root; vget(vp, LK_EXCLUSIVE | LK_RETRY, curthread); *vpp = vp; return (0); } static int fdesc_statfs(mp, sbp) struct mount *mp; struct statfs *sbp; { struct thread *td; struct filedesc *fdp; int lim; int i; int last; int freefd; uint64_t limit; td = curthread; /* * Compute number of free file descriptors. * [ Strange results will ensue if the open file * limit is ever reduced below the current number * of open files... ] */ - PROC_LOCK(td->td_proc); - lim = lim_cur(td->td_proc, RLIMIT_NOFILE); - PROC_UNLOCK(td->td_proc); + lim = lim_cur(td, RLIMIT_NOFILE); fdp = td->td_proc->p_fd; FILEDESC_SLOCK(fdp); limit = racct_get_limit(td->td_proc, RACCT_NOFILE); if (lim > limit) lim = limit; last = min(fdp->fd_nfiles, lim); freefd = 0; for (i = fdp->fd_freefile; i < last; i++) if (fdp->fd_ofiles[i].fde_file == NULL) freefd++; /* * Adjust for the fact that the fdesc array may not * have been fully allocated yet. */ if (fdp->fd_nfiles < lim) freefd += (lim - fdp->fd_nfiles); FILEDESC_SUNLOCK(fdp); sbp->f_flags = 0; sbp->f_bsize = DEV_BSIZE; sbp->f_iosize = DEV_BSIZE; sbp->f_blocks = 2; /* 1K to keep df happy */ sbp->f_bfree = 0; sbp->f_bavail = 0; sbp->f_files = lim + 1; /* Allow for "." */ sbp->f_ffree = freefd; /* See comments above */ return (0); } static struct vfsops fdesc_vfsops = { .vfs_cmount = fdesc_cmount, .vfs_init = fdesc_init, .vfs_mount = fdesc_mount, .vfs_root = fdesc_root, .vfs_statfs = fdesc_statfs, .vfs_uninit = fdesc_uninit, .vfs_unmount = fdesc_unmount, }; VFS_SET(fdesc_vfsops, fdescfs, VFCF_SYNTHETIC | VFCF_JAIL); Index: head/sys/i386/ibcs2/ibcs2_misc.c =================================================================== --- head/sys/i386/ibcs2/ibcs2_misc.c (revision 284214) +++ head/sys/i386/ibcs2/ibcs2_misc.c (revision 284215) @@ -1,1279 +1,1263 @@ /*- * Copyright (c) 1995 Steven Wallace * Copyright (c) 1994, 1995 Scott Bartram * Copyright (c) 1992, 1993 * The Regents of the University of California. All rights reserved. * * This software was developed by the Computer Systems Engineering group * at Lawrence Berkeley Laboratory under DARPA contract BG 91-66 and * contributed to Berkeley. * * All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Lawrence Berkeley Laboratory. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Berkeley and its contributors. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: Header: sun_misc.c,v 1.16 93/04/07 02:46:27 torek Exp * * @(#)sun_misc.c 8.1 (Berkeley) 6/18/93 */ #include __FBSDID("$FreeBSD$"); /* * IBCS2 compatibility module. * * IBCS2 system calls that are implemented differently in BSD are * handled here. */ #include #include #include #include #include #include #include #include #include #include #include /* Must come after sys/malloc.h */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include int ibcs2_ulimit(td, uap) struct thread *td; struct ibcs2_ulimit_args *uap; { struct rlimit rl; - struct proc *p; int error; #define IBCS2_GETFSIZE 1 #define IBCS2_SETFSIZE 2 #define IBCS2_GETPSIZE 3 #define IBCS2_GETDTABLESIZE 4 - p = td->td_proc; switch (uap->cmd) { case IBCS2_GETFSIZE: - PROC_LOCK(p); - td->td_retval[0] = lim_cur(p, RLIMIT_FSIZE); - PROC_UNLOCK(p); + td->td_retval[0] = lim_cur(td, RLIMIT_FSIZE); if (td->td_retval[0] == -1) td->td_retval[0] = 0x7fffffff; return 0; case IBCS2_SETFSIZE: - PROC_LOCK(p); - rl.rlim_max = lim_max(p, RLIMIT_FSIZE); - PROC_UNLOCK(p); + rl.rlim_max = lim_max(td, RLIMIT_FSIZE); rl.rlim_cur = uap->newlimit; error = kern_setrlimit(td, RLIMIT_FSIZE, &rl); if (!error) { - PROC_LOCK(p); - td->td_retval[0] = lim_cur(p, RLIMIT_FSIZE); - PROC_UNLOCK(p); + td->td_retval[0] = lim_cur(td, RLIMIT_FSIZE); } else { DPRINTF(("failed ")); } return error; case IBCS2_GETPSIZE: - PROC_LOCK(p); - td->td_retval[0] = lim_cur(p, RLIMIT_RSS); /* XXX */ - PROC_UNLOCK(p); + td->td_retval[0] = lim_cur(td, RLIMIT_RSS); /* XXX */ return 0; case IBCS2_GETDTABLESIZE: uap->cmd = IBCS2_SC_OPEN_MAX; return ibcs2_sysconf(td, (struct ibcs2_sysconf_args *)uap); default: return ENOSYS; } } #define IBCS2_WSTOPPED 0177 #define IBCS2_STOPCODE(sig) ((sig) << 8 | IBCS2_WSTOPPED) int ibcs2_wait(td, uap) struct thread *td; struct ibcs2_wait_args *uap; { int error, options, status; int *statusp; pid_t pid; struct trapframe *tf = td->td_frame; if ((tf->tf_eflags & (PSL_Z|PSL_PF|PSL_N|PSL_V)) == (PSL_Z|PSL_PF|PSL_N|PSL_V)) { /* waitpid */ pid = uap->a1; statusp = (int *)uap->a2; options = uap->a3; } else { /* wait */ pid = WAIT_ANY; statusp = (int *)uap->a1; options = 0; } error = kern_wait(td, pid, &status, options, NULL); if (error) return error; if (statusp) { /* * Convert status/signal result. */ if (WIFSTOPPED(status)) { if (WSTOPSIG(status) <= 0 || WSTOPSIG(status) > IBCS2_SIGTBLSZ) return (EINVAL); status = IBCS2_STOPCODE(bsd_to_ibcs2_sig[_SIG_IDX(WSTOPSIG(status))]); } else if (WIFSIGNALED(status)) { if (WTERMSIG(status) <= 0 || WTERMSIG(status) > IBCS2_SIGTBLSZ) return (EINVAL); status = bsd_to_ibcs2_sig[_SIG_IDX(WTERMSIG(status))]; } /* else exit status -- identical */ /* record result/status */ td->td_retval[1] = status; return copyout(&status, statusp, sizeof(status)); } return 0; } int ibcs2_execv(td, uap) struct thread *td; struct ibcs2_execv_args *uap; { struct image_args eargs; struct vmspace *oldvmspace; char *path; int error; CHECKALTEXIST(td, uap->path, &path); error = pre_execve(td, &oldvmspace); if (error != 0) { free(path, M_TEMP); return (error); } error = exec_copyin_args(&eargs, path, UIO_SYSSPACE, uap->argp, NULL); free(path, M_TEMP); if (error == 0) error = kern_execve(td, &eargs, NULL); post_execve(td, error, oldvmspace); return (error); } int ibcs2_execve(td, uap) struct thread *td; struct ibcs2_execve_args *uap; { struct image_args eargs; struct vmspace *oldvmspace; char *path; int error; CHECKALTEXIST(td, uap->path, &path); error = pre_execve(td, &oldvmspace); if (error != 0) { free(path, M_TEMP); return (error); } error = exec_copyin_args(&eargs, path, UIO_SYSSPACE, uap->argp, uap->envp); free(path, M_TEMP); if (error == 0) error = kern_execve(td, &eargs, NULL); post_execve(td, error, oldvmspace); return (error); } int ibcs2_umount(td, uap) struct thread *td; struct ibcs2_umount_args *uap; { struct unmount_args um; um.path = uap->name; um.flags = 0; return sys_unmount(td, &um); } int ibcs2_mount(td, uap) struct thread *td; struct ibcs2_mount_args *uap; { #ifdef notyet int oflags = uap->flags, nflags, error; char fsname[MFSNAMELEN]; if (oflags & (IBCS2_MS_NOSUB | IBCS2_MS_SYS5)) return (EINVAL); if ((oflags & IBCS2_MS_NEWTYPE) == 0) return (EINVAL); nflags = 0; if (oflags & IBCS2_MS_RDONLY) nflags |= MNT_RDONLY; if (oflags & IBCS2_MS_NOSUID) nflags |= MNT_NOSUID; if (oflags & IBCS2_MS_REMOUNT) nflags |= MNT_UPDATE; uap->flags = nflags; if (error = copyinstr((caddr_t)uap->type, fsname, sizeof fsname, (u_int *)0)) return (error); if (strcmp(fsname, "4.2") == 0) { uap->type = (caddr_t)STACK_ALLOC(); if (error = copyout("ufs", uap->type, sizeof("ufs"))) return (error); } else if (strcmp(fsname, "nfs") == 0) { struct ibcs2_nfs_args sna; struct sockaddr_in sain; struct nfs_args na; struct sockaddr sa; if (error = copyin(uap->data, &sna, sizeof sna)) return (error); if (error = copyin(sna.addr, &sain, sizeof sain)) return (error); bcopy(&sain, &sa, sizeof sa); sa.sa_len = sizeof(sain); uap->data = (caddr_t)STACK_ALLOC(); na.addr = (struct sockaddr *)((int)uap->data + sizeof na); na.sotype = SOCK_DGRAM; na.proto = IPPROTO_UDP; na.fh = (nfsv2fh_t *)sna.fh; na.flags = sna.flags; na.wsize = sna.wsize; na.rsize = sna.rsize; na.timeo = sna.timeo; na.retrans = sna.retrans; na.hostname = sna.hostname; if (error = copyout(&sa, na.addr, sizeof sa)) return (error); if (error = copyout(&na, uap->data, sizeof na)) return (error); } return (mount(td, uap)); #else return EINVAL; #endif } /* * Read iBCS2-style directory entries. We suck them into kernel space so * that they can be massaged before being copied out to user code. Like * SunOS, we squish out `empty' entries. * * This is quite ugly, but what do you expect from compatibility code? */ int ibcs2_getdents(td, uap) struct thread *td; register struct ibcs2_getdents_args *uap; { register struct vnode *vp; register caddr_t inp, buf; /* BSD-format */ register int len, reclen; /* BSD-format */ register caddr_t outp; /* iBCS2-format */ register int resid; /* iBCS2-format */ cap_rights_t rights; struct file *fp; struct uio auio; struct iovec aiov; struct ibcs2_dirent idb; off_t off; /* true file offset */ int buflen, error, eofflag; u_long *cookies = NULL, *cookiep; int ncookies; #define BSD_DIRENT(cp) ((struct dirent *)(cp)) #define IBCS2_RECLEN(reclen) (reclen + sizeof(u_short)) error = getvnode(td->td_proc->p_fd, uap->fd, cap_rights_init(&rights, CAP_READ), &fp); if (error != 0) return (error); if ((fp->f_flag & FREAD) == 0) { fdrop(fp, td); return (EBADF); } vp = fp->f_vnode; if (vp->v_type != VDIR) { /* XXX vnode readdir op should do this */ fdrop(fp, td); return (EINVAL); } off = fp->f_offset; #define DIRBLKSIZ 512 /* XXX we used to use ufs's DIRBLKSIZ */ buflen = max(DIRBLKSIZ, uap->nbytes); buflen = min(buflen, MAXBSIZE); buf = malloc(buflen, M_TEMP, M_WAITOK); vn_lock(vp, LK_SHARED | LK_RETRY); again: aiov.iov_base = buf; aiov.iov_len = buflen; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_READ; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = td; auio.uio_resid = buflen; auio.uio_offset = off; if (cookies) { free(cookies, M_TEMP); cookies = NULL; } #ifdef MAC error = mac_vnode_check_readdir(td->td_ucred, vp); if (error) goto out; #endif /* * First we read into the malloc'ed buffer, then * we massage it into user space, one record at a time. */ if ((error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, &ncookies, &cookies)) != 0) goto out; inp = buf; outp = uap->buf; resid = uap->nbytes; if ((len = buflen - auio.uio_resid) <= 0) goto eof; cookiep = cookies; if (cookies) { /* * When using cookies, the vfs has the option of reading from * a different offset than that supplied (UFS truncates the * offset to a block boundary to make sure that it never reads * partway through a directory entry, even if the directory * has been compacted). */ while (len > 0 && ncookies > 0 && *cookiep <= off) { len -= BSD_DIRENT(inp)->d_reclen; inp += BSD_DIRENT(inp)->d_reclen; cookiep++; ncookies--; } } for (; len > 0; len -= reclen) { if (cookiep && ncookies == 0) break; reclen = BSD_DIRENT(inp)->d_reclen; if (reclen & 3) { printf("ibcs2_getdents: reclen=%d\n", reclen); error = EFAULT; goto out; } if (BSD_DIRENT(inp)->d_fileno == 0) { inp += reclen; /* it is a hole; squish it out */ if (cookiep) { off = *cookiep++; ncookies--; } else off += reclen; continue; } if (reclen > len || resid < IBCS2_RECLEN(reclen)) { /* entry too big for buffer, so just stop */ outp++; break; } /* * Massage in place to make an iBCS2-shaped dirent (otherwise * we have to worry about touching user memory outside of * the copyout() call). */ idb.d_ino = (ibcs2_ino_t)BSD_DIRENT(inp)->d_fileno; idb.d_off = (ibcs2_off_t)off; idb.d_reclen = (u_short)IBCS2_RECLEN(reclen); if ((error = copyout((caddr_t)&idb, outp, 10)) != 0 || (error = copyout(BSD_DIRENT(inp)->d_name, outp + 10, BSD_DIRENT(inp)->d_namlen + 1)) != 0) goto out; /* advance past this real entry */ if (cookiep) { off = *cookiep++; ncookies--; } else off += reclen; inp += reclen; /* advance output past iBCS2-shaped entry */ outp += IBCS2_RECLEN(reclen); resid -= IBCS2_RECLEN(reclen); } /* if we squished out the whole block, try again */ if (outp == uap->buf) goto again; fp->f_offset = off; /* update the vnode offset */ eof: td->td_retval[0] = uap->nbytes - resid; out: VOP_UNLOCK(vp, 0); fdrop(fp, td); if (cookies) free(cookies, M_TEMP); free(buf, M_TEMP); return (error); } int ibcs2_read(td, uap) struct thread *td; struct ibcs2_read_args *uap; { register struct vnode *vp; register caddr_t inp, buf; /* BSD-format */ register int len, reclen; /* BSD-format */ register caddr_t outp; /* iBCS2-format */ register int resid; /* iBCS2-format */ cap_rights_t rights; struct file *fp; struct uio auio; struct iovec aiov; struct ibcs2_direct { ibcs2_ino_t ino; char name[14]; } idb; off_t off; /* true file offset */ int buflen, error, eofflag, size; u_long *cookies = NULL, *cookiep; int ncookies; error = getvnode(td->td_proc->p_fd, uap->fd, cap_rights_init(&rights, CAP_READ), &fp); if (error != 0) { if (error == EINVAL) return sys_read(td, (struct read_args *)uap); else return error; } if ((fp->f_flag & FREAD) == 0) { fdrop(fp, td); return (EBADF); } vp = fp->f_vnode; if (vp->v_type != VDIR) { fdrop(fp, td); return sys_read(td, (struct read_args *)uap); } off = fp->f_offset; DPRINTF(("ibcs2_read: read directory\n")); buflen = max(DIRBLKSIZ, uap->nbytes); buflen = min(buflen, MAXBSIZE); buf = malloc(buflen, M_TEMP, M_WAITOK); vn_lock(vp, LK_SHARED | LK_RETRY); again: aiov.iov_base = buf; aiov.iov_len = buflen; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_READ; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = td; auio.uio_resid = buflen; auio.uio_offset = off; if (cookies) { free(cookies, M_TEMP); cookies = NULL; } #ifdef MAC error = mac_vnode_check_readdir(td->td_ucred, vp); if (error) goto out; #endif /* * First we read into the malloc'ed buffer, then * we massage it into user space, one record at a time. */ if ((error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, &ncookies, &cookies)) != 0) { DPRINTF(("VOP_READDIR failed: %d\n", error)); goto out; } inp = buf; outp = uap->buf; resid = uap->nbytes; if ((len = buflen - auio.uio_resid) <= 0) goto eof; cookiep = cookies; if (cookies) { /* * When using cookies, the vfs has the option of reading from * a different offset than that supplied (UFS truncates the * offset to a block boundary to make sure that it never reads * partway through a directory entry, even if the directory * has been compacted). */ while (len > 0 && ncookies > 0 && *cookiep <= off) { len -= BSD_DIRENT(inp)->d_reclen; inp += BSD_DIRENT(inp)->d_reclen; cookiep++; ncookies--; } } for (; len > 0 && resid > 0; len -= reclen) { if (cookiep && ncookies == 0) break; reclen = BSD_DIRENT(inp)->d_reclen; if (reclen & 3) { printf("ibcs2_read: reclen=%d\n", reclen); error = EFAULT; goto out; } if (BSD_DIRENT(inp)->d_fileno == 0) { inp += reclen; /* it is a hole; squish it out */ if (cookiep) { off = *cookiep++; ncookies--; } else off += reclen; continue; } if (reclen > len || resid < sizeof(struct ibcs2_direct)) { /* entry too big for buffer, so just stop */ outp++; break; } /* * Massage in place to make an iBCS2-shaped dirent (otherwise * we have to worry about touching user memory outside of * the copyout() call). * * TODO: if length(filename) > 14, then break filename into * multiple entries and set inode = 0xffff except last */ idb.ino = (BSD_DIRENT(inp)->d_fileno > 0xfffe) ? 0xfffe : BSD_DIRENT(inp)->d_fileno; (void)copystr(BSD_DIRENT(inp)->d_name, idb.name, 14, &size); bzero(idb.name + size, 14 - size); if ((error = copyout(&idb, outp, sizeof(struct ibcs2_direct))) != 0) goto out; /* advance past this real entry */ if (cookiep) { off = *cookiep++; ncookies--; } else off += reclen; inp += reclen; /* advance output past iBCS2-shaped entry */ outp += sizeof(struct ibcs2_direct); resid -= sizeof(struct ibcs2_direct); } /* if we squished out the whole block, try again */ if (outp == uap->buf) goto again; fp->f_offset = off; /* update the vnode offset */ eof: td->td_retval[0] = uap->nbytes - resid; out: VOP_UNLOCK(vp, 0); fdrop(fp, td); if (cookies) free(cookies, M_TEMP); free(buf, M_TEMP); return (error); } int ibcs2_mknod(td, uap) struct thread *td; struct ibcs2_mknod_args *uap; { char *path; int error; CHECKALTCREAT(td, uap->path, &path); if (S_ISFIFO(uap->mode)) { error = kern_mkfifoat(td, AT_FDCWD, path, UIO_SYSSPACE, uap->mode); } else { error = kern_mknodat(td, AT_FDCWD, path, UIO_SYSSPACE, uap->mode, uap->dev); } free(path, M_TEMP); return (error); } int ibcs2_getgroups(td, uap) struct thread *td; struct ibcs2_getgroups_args *uap; { struct ucred *cred; ibcs2_gid_t *iset; u_int i, ngrp; int error; cred = td->td_ucred; ngrp = cred->cr_ngroups; if (uap->gidsetsize == 0) { error = 0; goto out; } if (uap->gidsetsize < ngrp) return (EINVAL); iset = malloc(ngrp * sizeof(*iset), M_TEMP, M_WAITOK); for (i = 0; i < ngrp; i++) iset[i] = (ibcs2_gid_t)cred->cr_groups[i]; error = copyout(iset, uap->gidset, ngrp * sizeof(ibcs2_gid_t)); free(iset, M_TEMP); out: td->td_retval[0] = ngrp; return (error); } int ibcs2_setgroups(td, uap) struct thread *td; struct ibcs2_setgroups_args *uap; { ibcs2_gid_t *iset; gid_t *gp; int error, i; if (uap->gidsetsize < 0 || uap->gidsetsize > ngroups_max + 1) return (EINVAL); if (uap->gidsetsize && uap->gidset == NULL) return (EINVAL); gp = malloc(uap->gidsetsize * sizeof(*gp), M_TEMP, M_WAITOK); if (uap->gidsetsize) { iset = malloc(uap->gidsetsize * sizeof(*iset), M_TEMP, M_WAITOK); error = copyin(uap->gidset, iset, sizeof(ibcs2_gid_t) * uap->gidsetsize); if (error) { free(iset, M_TEMP); goto out; } for (i = 0; i < uap->gidsetsize; i++) gp[i] = (gid_t)iset[i]; } error = kern_setgroups(td, uap->gidsetsize, gp); out: free(gp, M_TEMP); return (error); } int ibcs2_setuid(td, uap) struct thread *td; struct ibcs2_setuid_args *uap; { struct setuid_args sa; sa.uid = (uid_t)uap->uid; return sys_setuid(td, &sa); } int ibcs2_setgid(td, uap) struct thread *td; struct ibcs2_setgid_args *uap; { struct setgid_args sa; sa.gid = (gid_t)uap->gid; return sys_setgid(td, &sa); } int ibcs2_time(td, uap) struct thread *td; struct ibcs2_time_args *uap; { struct timeval tv; microtime(&tv); td->td_retval[0] = tv.tv_sec; if (uap->tp) return copyout((caddr_t)&tv.tv_sec, (caddr_t)uap->tp, sizeof(ibcs2_time_t)); else return 0; } int ibcs2_pathconf(td, uap) struct thread *td; struct ibcs2_pathconf_args *uap; { char *path; int error; CHECKALTEXIST(td, uap->path, &path); uap->name++; /* iBCS2 _PC_* defines are offset by one */ error = kern_pathconf(td, path, UIO_SYSSPACE, uap->name, FOLLOW); free(path, M_TEMP); return (error); } int ibcs2_fpathconf(td, uap) struct thread *td; struct ibcs2_fpathconf_args *uap; { uap->name++; /* iBCS2 _PC_* defines are offset by one */ return sys_fpathconf(td, (struct fpathconf_args *)uap); } int ibcs2_sysconf(td, uap) struct thread *td; struct ibcs2_sysconf_args *uap; { int mib[2], value, len, error; - struct proc *p; - p = td->td_proc; switch(uap->name) { case IBCS2_SC_ARG_MAX: mib[1] = KERN_ARGMAX; break; case IBCS2_SC_CHILD_MAX: - PROC_LOCK(p); - td->td_retval[0] = lim_cur(td->td_proc, RLIMIT_NPROC); - PROC_UNLOCK(p); + td->td_retval[0] = lim_cur(td, RLIMIT_NPROC); return 0; case IBCS2_SC_CLK_TCK: td->td_retval[0] = hz; return 0; case IBCS2_SC_NGROUPS_MAX: mib[1] = KERN_NGROUPS; break; case IBCS2_SC_OPEN_MAX: - PROC_LOCK(p); - td->td_retval[0] = lim_cur(td->td_proc, RLIMIT_NOFILE); - PROC_UNLOCK(p); + td->td_retval[0] = lim_cur(td, RLIMIT_NOFILE); return 0; case IBCS2_SC_JOB_CONTROL: mib[1] = KERN_JOB_CONTROL; break; case IBCS2_SC_SAVED_IDS: mib[1] = KERN_SAVED_IDS; break; case IBCS2_SC_VERSION: mib[1] = KERN_POSIX1; break; case IBCS2_SC_PASS_MAX: td->td_retval[0] = 128; /* XXX - should we create PASS_MAX ? */ return 0; case IBCS2_SC_XOPEN_VERSION: td->td_retval[0] = 2; /* XXX: What should that be? */ return 0; default: return EINVAL; } mib[0] = CTL_KERN; len = sizeof(value); error = kernel_sysctl(td, mib, 2, &value, &len, NULL, 0, NULL, 0); if (error) return error; td->td_retval[0] = value; return 0; } int ibcs2_alarm(td, uap) struct thread *td; struct ibcs2_alarm_args *uap; { struct itimerval itv, oitv; int error; timevalclear(&itv.it_interval); itv.it_value.tv_sec = uap->sec; itv.it_value.tv_usec = 0; error = kern_setitimer(td, ITIMER_REAL, &itv, &oitv); if (error) return (error); if (oitv.it_value.tv_usec != 0) oitv.it_value.tv_sec++; td->td_retval[0] = oitv.it_value.tv_sec; return (0); } int ibcs2_times(td, uap) struct thread *td; struct ibcs2_times_args *uap; { struct rusage ru; struct timeval t; struct tms tms; int error; #define CONVTCK(r) (r.tv_sec * hz + r.tv_usec / (1000000 / hz)) error = kern_getrusage(td, RUSAGE_SELF, &ru); if (error) return (error); tms.tms_utime = CONVTCK(ru.ru_utime); tms.tms_stime = CONVTCK(ru.ru_stime); error = kern_getrusage(td, RUSAGE_CHILDREN, &ru); if (error) return (error); tms.tms_cutime = CONVTCK(ru.ru_utime); tms.tms_cstime = CONVTCK(ru.ru_stime); microtime(&t); td->td_retval[0] = CONVTCK(t); return (copyout(&tms, uap->tp, sizeof(struct tms))); } int ibcs2_stime(td, uap) struct thread *td; struct ibcs2_stime_args *uap; { struct timeval tv; long secs; int error; error = copyin(uap->timep, &secs, sizeof(long)); if (error) return (error); tv.tv_sec = secs; tv.tv_usec = 0; error = kern_settimeofday(td, &tv, NULL); if (error) error = EPERM; return (error); } int ibcs2_utime(td, uap) struct thread *td; struct ibcs2_utime_args *uap; { struct ibcs2_utimbuf ubuf; struct timeval tbuf[2], *tp; char *path; int error; if (uap->buf) { error = copyin(uap->buf, &ubuf, sizeof(ubuf)); if (error) return (error); tbuf[0].tv_sec = ubuf.actime; tbuf[0].tv_usec = 0; tbuf[1].tv_sec = ubuf.modtime; tbuf[1].tv_usec = 0; tp = tbuf; } else tp = NULL; CHECKALTEXIST(td, uap->path, &path); error = kern_utimesat(td, AT_FDCWD, path, UIO_SYSSPACE, tp, UIO_SYSSPACE); free(path, M_TEMP); return (error); } int ibcs2_nice(td, uap) struct thread *td; struct ibcs2_nice_args *uap; { int error; struct setpriority_args sa; sa.which = PRIO_PROCESS; sa.who = 0; sa.prio = td->td_proc->p_nice + uap->incr; if ((error = sys_setpriority(td, &sa)) != 0) return EPERM; td->td_retval[0] = td->td_proc->p_nice; return 0; } /* * iBCS2 getpgrp, setpgrp, setsid, and setpgid */ int ibcs2_pgrpsys(td, uap) struct thread *td; struct ibcs2_pgrpsys_args *uap; { struct proc *p = td->td_proc; switch (uap->type) { case 0: /* getpgrp */ PROC_LOCK(p); td->td_retval[0] = p->p_pgrp->pg_id; PROC_UNLOCK(p); return 0; case 1: /* setpgrp */ { struct setpgid_args sa; sa.pid = 0; sa.pgid = 0; sys_setpgid(td, &sa); PROC_LOCK(p); td->td_retval[0] = p->p_pgrp->pg_id; PROC_UNLOCK(p); return 0; } case 2: /* setpgid */ { struct setpgid_args sa; sa.pid = uap->pid; sa.pgid = uap->pgid; return sys_setpgid(td, &sa); } case 3: /* setsid */ return sys_setsid(td, NULL); default: return EINVAL; } } /* * XXX - need to check for nested calls */ int ibcs2_plock(td, uap) struct thread *td; struct ibcs2_plock_args *uap; { int error; #define IBCS2_UNLOCK 0 #define IBCS2_PROCLOCK 1 #define IBCS2_TEXTLOCK 2 #define IBCS2_DATALOCK 4 switch(uap->cmd) { case IBCS2_UNLOCK: error = priv_check(td, PRIV_VM_MUNLOCK); if (error) return (error); /* XXX - TODO */ return (0); case IBCS2_PROCLOCK: case IBCS2_TEXTLOCK: case IBCS2_DATALOCK: error = priv_check(td, PRIV_VM_MLOCK); if (error) return (error); /* XXX - TODO */ return 0; } return EINVAL; } int ibcs2_uadmin(td, uap) struct thread *td; struct ibcs2_uadmin_args *uap; { #define SCO_A_REBOOT 1 #define SCO_A_SHUTDOWN 2 #define SCO_A_REMOUNT 4 #define SCO_A_CLOCK 8 #define SCO_A_SETCONFIG 128 #define SCO_A_GETDEV 130 #define SCO_AD_HALT 0 #define SCO_AD_BOOT 1 #define SCO_AD_IBOOT 2 #define SCO_AD_PWRDOWN 3 #define SCO_AD_PWRNAP 4 #define SCO_AD_PANICBOOT 1 #define SCO_AD_GETBMAJ 0 #define SCO_AD_GETCMAJ 1 switch(uap->cmd) { case SCO_A_REBOOT: case SCO_A_SHUTDOWN: switch(uap->func) { struct reboot_args r; case SCO_AD_HALT: case SCO_AD_PWRDOWN: case SCO_AD_PWRNAP: r.opt = RB_HALT; return (sys_reboot(td, &r)); case SCO_AD_BOOT: case SCO_AD_IBOOT: r.opt = RB_AUTOBOOT; return (sys_reboot(td, &r)); } return EINVAL; case SCO_A_REMOUNT: case SCO_A_CLOCK: case SCO_A_SETCONFIG: return 0; case SCO_A_GETDEV: return EINVAL; /* XXX - TODO */ } return EINVAL; } int ibcs2_sysfs(td, uap) struct thread *td; struct ibcs2_sysfs_args *uap; { #define IBCS2_GETFSIND 1 #define IBCS2_GETFSTYP 2 #define IBCS2_GETNFSTYP 3 switch(uap->cmd) { case IBCS2_GETFSIND: case IBCS2_GETFSTYP: case IBCS2_GETNFSTYP: break; } return EINVAL; /* XXX - TODO */ } int ibcs2_unlink(td, uap) struct thread *td; struct ibcs2_unlink_args *uap; { char *path; int error; CHECKALTEXIST(td, uap->path, &path); error = kern_unlinkat(td, AT_FDCWD, path, UIO_SYSSPACE, 0); free(path, M_TEMP); return (error); } int ibcs2_chdir(td, uap) struct thread *td; struct ibcs2_chdir_args *uap; { char *path; int error; CHECKALTEXIST(td, uap->path, &path); error = kern_chdir(td, path, UIO_SYSSPACE); free(path, M_TEMP); return (error); } int ibcs2_chmod(td, uap) struct thread *td; struct ibcs2_chmod_args *uap; { char *path; int error; CHECKALTEXIST(td, uap->path, &path); error = kern_fchmodat(td, AT_FDCWD, path, UIO_SYSSPACE, uap->mode, 0); free(path, M_TEMP); return (error); } int ibcs2_chown(td, uap) struct thread *td; struct ibcs2_chown_args *uap; { char *path; int error; CHECKALTEXIST(td, uap->path, &path); error = kern_fchownat(td, AT_FDCWD, path, UIO_SYSSPACE, uap->uid, uap->gid, 0); free(path, M_TEMP); return (error); } int ibcs2_rmdir(td, uap) struct thread *td; struct ibcs2_rmdir_args *uap; { char *path; int error; CHECKALTEXIST(td, uap->path, &path); error = kern_rmdirat(td, AT_FDCWD, path, UIO_SYSSPACE); free(path, M_TEMP); return (error); } int ibcs2_mkdir(td, uap) struct thread *td; struct ibcs2_mkdir_args *uap; { char *path; int error; CHECKALTEXIST(td, uap->path, &path); error = kern_mkdirat(td, AT_FDCWD, path, UIO_SYSSPACE, uap->mode); free(path, M_TEMP); return (error); } int ibcs2_symlink(td, uap) struct thread *td; struct ibcs2_symlink_args *uap; { char *path, *link; int error; CHECKALTEXIST(td, uap->path, &path); /* * Have to expand CHECKALTCREAT() so that 'path' can be freed on * errors. */ error = ibcs2_emul_find(td, uap->link, UIO_USERSPACE, &link, 1); if (link == NULL) { free(path, M_TEMP); return (error); } error = kern_symlinkat(td, path, AT_FDCWD, link, UIO_SYSSPACE); free(path, M_TEMP); free(link, M_TEMP); return (error); } int ibcs2_rename(td, uap) struct thread *td; struct ibcs2_rename_args *uap; { char *from, *to; int error; CHECKALTEXIST(td, uap->from, &from); /* * Have to expand CHECKALTCREAT() so that 'from' can be freed on * errors. */ error = ibcs2_emul_find(td, uap->to, UIO_USERSPACE, &to, 1); if (to == NULL) { free(from, M_TEMP); return (error); } error = kern_renameat(td, AT_FDCWD, from, AT_FDCWD, to, UIO_SYSSPACE); free(from, M_TEMP); free(to, M_TEMP); return (error); } int ibcs2_readlink(td, uap) struct thread *td; struct ibcs2_readlink_args *uap; { char *path; int error; CHECKALTEXIST(td, uap->path, &path); error = kern_readlinkat(td, AT_FDCWD, path, UIO_SYSSPACE, uap->buf, UIO_USERSPACE, uap->count); free(path, M_TEMP); return (error); } Index: head/sys/i386/linux/imgact_linux.c =================================================================== --- head/sys/i386/linux/imgact_linux.c (revision 284214) +++ head/sys/i386/linux/imgact_linux.c (revision 284215) @@ -1,239 +1,239 @@ /*- * Copyright (c) 1994-1996 Søren Schmidt * All rights reserved. * * Based heavily on /sys/kern/imgact_aout.c which is: * Copyright (c) 1993, David Greenman * * 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 * in this position and unchanged. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static int exec_linux_imgact(struct image_params *iparams); static int exec_linux_imgact(struct image_params *imgp) { const struct exec *a_out = (const struct exec *) imgp->image_header; struct vmspace *vmspace; vm_offset_t vmaddr; unsigned long virtual_offset, file_offset; unsigned long bss_size; ssize_t aresid; int error; if (((a_out->a_magic >> 16) & 0xff) != 0x64) return -1; /* * Set file/virtual offset based on a.out variant. */ switch ((int)(a_out->a_magic & 0xffff)) { case 0413: virtual_offset = 0; file_offset = 1024; break; case 0314: virtual_offset = 4096; file_offset = 0; break; default: return (-1); } bss_size = round_page(a_out->a_bss); #ifdef DEBUG printf("imgact: text: %08lx, data: %08lx, bss: %08lx\n", (u_long)a_out->a_text, (u_long)a_out->a_data, bss_size); #endif /* * Check various fields in header for validity/bounds. */ if (a_out->a_entry < virtual_offset || a_out->a_entry >= virtual_offset + a_out->a_text || a_out->a_text & PAGE_MASK || a_out->a_data & PAGE_MASK) return (-1); /* text + data can't exceed file size */ if (a_out->a_data + a_out->a_text > imgp->attr->va_size) return (EFAULT); /* * text/data/bss must not exceed limits */ PROC_LOCK(imgp->proc); if (a_out->a_text > maxtsiz || - a_out->a_data + bss_size > lim_cur(imgp->proc, RLIMIT_DATA) || + a_out->a_data + bss_size > lim_cur_proc(imgp->proc, RLIMIT_DATA) || racct_set(imgp->proc, RACCT_DATA, a_out->a_data + bss_size) != 0) { PROC_UNLOCK(imgp->proc); return (ENOMEM); } PROC_UNLOCK(imgp->proc); VOP_UNLOCK(imgp->vp, 0); /* * Destroy old process VM and create a new one (with a new stack) */ error = exec_new_vmspace(imgp, &linux_sysvec); if (error) goto fail; vmspace = imgp->proc->p_vmspace; /* * Check if file_offset page aligned,. * Currently we cannot handle misaligned file offsets, * and so we read in the entire image (what a waste). */ if (file_offset & PAGE_MASK) { #ifdef DEBUG printf("imgact: Non page aligned binary %lu\n", file_offset); #endif /* * Map text+data+bss read/write/execute */ vmaddr = virtual_offset; error = vm_map_find(&vmspace->vm_map, NULL, 0, &vmaddr, a_out->a_text + a_out->a_data + bss_size, 0, VMFS_NO_SPACE, VM_PROT_ALL, VM_PROT_ALL, 0); if (error) goto fail; error = vn_rdwr(UIO_READ, imgp->vp, (void *)vmaddr, file_offset, a_out->a_text + a_out->a_data, UIO_USERSPACE, 0, curthread->td_ucred, NOCRED, &aresid, curthread); if (error != 0) goto fail; if (aresid != 0) { error = ENOEXEC; goto fail; } /* * remove write enable on the 'text' part */ error = vm_map_protect(&vmspace->vm_map, vmaddr, vmaddr + a_out->a_text, VM_PROT_EXECUTE|VM_PROT_READ, TRUE); if (error) goto fail; } else { #ifdef DEBUG printf("imgact: Page aligned binary %lu\n", file_offset); #endif /* * Map text+data read/execute */ vmaddr = virtual_offset; error = vm_mmap(&vmspace->vm_map, &vmaddr, a_out->a_text + a_out->a_data, VM_PROT_READ | VM_PROT_EXECUTE, VM_PROT_ALL, MAP_PRIVATE | MAP_FIXED, OBJT_VNODE, imgp->vp, file_offset); if (error) goto fail; #ifdef DEBUG printf("imgact: startaddr=%08lx, length=%08lx\n", (u_long)vmaddr, (u_long)a_out->a_text + (u_long)a_out->a_data); #endif /* * allow read/write of data */ error = vm_map_protect(&vmspace->vm_map, vmaddr + a_out->a_text, vmaddr + a_out->a_text + a_out->a_data, VM_PROT_ALL, FALSE); if (error) goto fail; /* * Allocate anon demand-zeroed area for uninitialized data */ if (bss_size != 0) { vmaddr = virtual_offset + a_out->a_text + a_out->a_data; error = vm_map_find(&vmspace->vm_map, NULL, 0, &vmaddr, bss_size, 0, VMFS_NO_SPACE, VM_PROT_ALL, VM_PROT_ALL, 0); if (error) goto fail; #ifdef DEBUG printf("imgact: bssaddr=%08lx, length=%08lx\n", (u_long)vmaddr, bss_size); #endif } } /* Fill in process VM information */ vmspace->vm_tsize = round_page(a_out->a_text) >> PAGE_SHIFT; vmspace->vm_dsize = round_page(a_out->a_data + bss_size) >> PAGE_SHIFT; vmspace->vm_taddr = (caddr_t)(void *)(uintptr_t)virtual_offset; vmspace->vm_daddr = (caddr_t)(void *)(uintptr_t) (virtual_offset + a_out->a_text); /* Fill in image_params */ imgp->interpreted = 0; imgp->entry_addr = a_out->a_entry; imgp->proc->p_sysent = &linux_sysvec; fail: vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); return (error); } /* * Tell kern_execve.c about it, with a little help from the linker. */ static struct execsw linux_execsw = { exec_linux_imgact, "linux a.out" }; EXEC_SET(linuxaout, linux_execsw); Index: head/sys/i386/linux/linux_machdep.c =================================================================== --- head/sys/i386/linux/linux_machdep.c (revision 284214) +++ head/sys/i386/linux/linux_machdep.c (revision 284215) @@ -1,1017 +1,1017 @@ /*- * Copyright (c) 2000 Marcel Moolenaar * 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 * in this position and unchanged. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #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 /* needed for pcb definition in linux_set_thread_area */ #include "opt_posix.h" extern struct sysentvec elf32_freebsd_sysvec; /* defined in i386/i386/elf_machdep.c */ struct l_descriptor { l_uint entry_number; l_ulong base_addr; l_uint limit; l_uint seg_32bit:1; l_uint contents:2; l_uint read_exec_only:1; l_uint limit_in_pages:1; l_uint seg_not_present:1; l_uint useable:1; }; struct l_old_select_argv { l_int nfds; l_fd_set *readfds; l_fd_set *writefds; l_fd_set *exceptfds; struct l_timeval *timeout; }; static int linux_mmap_common(struct thread *td, l_uintptr_t addr, l_size_t len, l_int prot, l_int flags, l_int fd, l_loff_t pos); int linux_execve(struct thread *td, struct linux_execve_args *args) { struct image_args eargs; char *newpath; int error; LCONVPATHEXIST(td, args->path, &newpath); #ifdef DEBUG if (ldebug(execve)) printf(ARGS(execve, "%s"), newpath); #endif error = exec_copyin_args(&eargs, newpath, UIO_SYSSPACE, args->argp, args->envp); free(newpath, M_TEMP); if (error == 0) error = linux_common_execve(td, &eargs); return (error); } struct l_ipc_kludge { struct l_msgbuf *msgp; l_long msgtyp; }; int linux_ipc(struct thread *td, struct linux_ipc_args *args) { switch (args->what & 0xFFFF) { case LINUX_SEMOP: { struct linux_semop_args a; a.semid = args->arg1; a.tsops = args->ptr; a.nsops = args->arg2; return (linux_semop(td, &a)); } case LINUX_SEMGET: { struct linux_semget_args a; a.key = args->arg1; a.nsems = args->arg2; a.semflg = args->arg3; return (linux_semget(td, &a)); } case LINUX_SEMCTL: { struct linux_semctl_args a; int error; a.semid = args->arg1; a.semnum = args->arg2; a.cmd = args->arg3; error = copyin(args->ptr, &a.arg, sizeof(a.arg)); if (error) return (error); return (linux_semctl(td, &a)); } case LINUX_MSGSND: { struct linux_msgsnd_args a; a.msqid = args->arg1; a.msgp = args->ptr; a.msgsz = args->arg2; a.msgflg = args->arg3; return (linux_msgsnd(td, &a)); } case LINUX_MSGRCV: { struct linux_msgrcv_args a; a.msqid = args->arg1; a.msgsz = args->arg2; a.msgflg = args->arg3; if ((args->what >> 16) == 0) { struct l_ipc_kludge tmp; int error; if (args->ptr == NULL) return (EINVAL); error = copyin(args->ptr, &tmp, sizeof(tmp)); if (error) return (error); a.msgp = tmp.msgp; a.msgtyp = tmp.msgtyp; } else { a.msgp = args->ptr; a.msgtyp = args->arg5; } return (linux_msgrcv(td, &a)); } case LINUX_MSGGET: { struct linux_msgget_args a; a.key = args->arg1; a.msgflg = args->arg2; return (linux_msgget(td, &a)); } case LINUX_MSGCTL: { struct linux_msgctl_args a; a.msqid = args->arg1; a.cmd = args->arg2; a.buf = args->ptr; return (linux_msgctl(td, &a)); } case LINUX_SHMAT: { struct linux_shmat_args a; a.shmid = args->arg1; a.shmaddr = args->ptr; a.shmflg = args->arg2; a.raddr = (l_ulong *)args->arg3; return (linux_shmat(td, &a)); } case LINUX_SHMDT: { struct linux_shmdt_args a; a.shmaddr = args->ptr; return (linux_shmdt(td, &a)); } case LINUX_SHMGET: { struct linux_shmget_args a; a.key = args->arg1; a.size = args->arg2; a.shmflg = args->arg3; return (linux_shmget(td, &a)); } case LINUX_SHMCTL: { struct linux_shmctl_args a; a.shmid = args->arg1; a.cmd = args->arg2; a.buf = args->ptr; return (linux_shmctl(td, &a)); } default: break; } return (EINVAL); } int linux_old_select(struct thread *td, struct linux_old_select_args *args) { struct l_old_select_argv linux_args; struct linux_select_args newsel; int error; #ifdef DEBUG if (ldebug(old_select)) printf(ARGS(old_select, "%p"), args->ptr); #endif error = copyin(args->ptr, &linux_args, sizeof(linux_args)); if (error) return (error); newsel.nfds = linux_args.nfds; newsel.readfds = linux_args.readfds; newsel.writefds = linux_args.writefds; newsel.exceptfds = linux_args.exceptfds; newsel.timeout = linux_args.timeout; return (linux_select(td, &newsel)); } int linux_set_cloned_tls(struct thread *td, void *desc) { struct segment_descriptor sd; struct l_user_desc info; int idx, error; int a[2]; error = copyin(desc, &info, sizeof(struct l_user_desc)); if (error) { printf(LMSG("copyin failed!")); } else { idx = info.entry_number; /* * looks like we're getting the idx we returned * in the set_thread_area() syscall */ if (idx != 6 && idx != 3) { printf(LMSG("resetting idx!")); idx = 3; } /* this doesnt happen in practice */ if (idx == 6) { /* we might copy out the entry_number as 3 */ info.entry_number = 3; error = copyout(&info, desc, sizeof(struct l_user_desc)); if (error) printf(LMSG("copyout failed!")); } a[0] = LINUX_LDT_entry_a(&info); a[1] = LINUX_LDT_entry_b(&info); memcpy(&sd, &a, sizeof(a)); #ifdef DEBUG if (ldebug(clone)) printf("Segment created in clone with " "CLONE_SETTLS: lobase: %x, hibase: %x, " "lolimit: %x, hilimit: %x, type: %i, " "dpl: %i, p: %i, xx: %i, def32: %i, " "gran: %i\n", sd.sd_lobase, sd.sd_hibase, sd.sd_lolimit, sd.sd_hilimit, sd.sd_type, sd.sd_dpl, sd.sd_p, sd.sd_xx, sd.sd_def32, sd.sd_gran); #endif /* set %gs */ td->td_pcb->pcb_gsd = sd; td->td_pcb->pcb_gs = GSEL(GUGS_SEL, SEL_UPL); } return (error); } int linux_set_upcall_kse(struct thread *td, register_t stack) { if (stack) td->td_frame->tf_esp = stack; /* * The newly created Linux thread returns * to the user space by the same path that a parent do. */ td->td_frame->tf_eax = 0; return (0); } #define STACK_SIZE (2 * 1024 * 1024) #define GUARD_SIZE (4 * PAGE_SIZE) int linux_mmap2(struct thread *td, struct linux_mmap2_args *args) { #ifdef DEBUG if (ldebug(mmap2)) printf(ARGS(mmap2, "%p, %d, %d, 0x%08x, %d, %d"), (void *)args->addr, args->len, args->prot, args->flags, args->fd, args->pgoff); #endif return (linux_mmap_common(td, args->addr, args->len, args->prot, args->flags, args->fd, (uint64_t)(uint32_t)args->pgoff * PAGE_SIZE)); } int linux_mmap(struct thread *td, struct linux_mmap_args *args) { int error; struct l_mmap_argv linux_args; error = copyin(args->ptr, &linux_args, sizeof(linux_args)); if (error) return (error); #ifdef DEBUG if (ldebug(mmap)) printf(ARGS(mmap, "%p, %d, %d, 0x%08x, %d, %d"), (void *)linux_args.addr, linux_args.len, linux_args.prot, linux_args.flags, linux_args.fd, linux_args.pgoff); #endif return (linux_mmap_common(td, linux_args.addr, linux_args.len, linux_args.prot, linux_args.flags, linux_args.fd, (uint32_t)linux_args.pgoff)); } static int linux_mmap_common(struct thread *td, l_uintptr_t addr, l_size_t len, l_int prot, l_int flags, l_int fd, l_loff_t pos) { struct proc *p = td->td_proc; struct mmap_args /* { caddr_t addr; size_t len; int prot; int flags; int fd; long pad; off_t pos; } */ bsd_args; int error; struct file *fp; cap_rights_t rights; error = 0; bsd_args.flags = 0; fp = NULL; /* * Linux mmap(2): * You must specify exactly one of MAP_SHARED and MAP_PRIVATE */ if (!((flags & LINUX_MAP_SHARED) ^ (flags & LINUX_MAP_PRIVATE))) return (EINVAL); if (flags & LINUX_MAP_SHARED) bsd_args.flags |= MAP_SHARED; if (flags & LINUX_MAP_PRIVATE) bsd_args.flags |= MAP_PRIVATE; if (flags & LINUX_MAP_FIXED) bsd_args.flags |= MAP_FIXED; if (flags & LINUX_MAP_ANON) { /* Enforce pos to be on page boundary, then ignore. */ if ((pos & PAGE_MASK) != 0) return (EINVAL); pos = 0; bsd_args.flags |= MAP_ANON; } else bsd_args.flags |= MAP_NOSYNC; if (flags & LINUX_MAP_GROWSDOWN) bsd_args.flags |= MAP_STACK; /* * PROT_READ, PROT_WRITE, or PROT_EXEC implies PROT_READ and PROT_EXEC * on Linux/i386. We do this to ensure maximum compatibility. * Linux/ia64 does the same in i386 emulation mode. */ bsd_args.prot = prot; if (bsd_args.prot & (PROT_READ | PROT_WRITE | PROT_EXEC)) bsd_args.prot |= PROT_READ | PROT_EXEC; /* Linux does not check file descriptor when MAP_ANONYMOUS is set. */ bsd_args.fd = (bsd_args.flags & MAP_ANON) ? -1 : fd; if (bsd_args.fd != -1) { /* * Linux follows Solaris mmap(2) description: * The file descriptor fildes is opened with * read permission, regardless of the * protection options specified. * * Checking just CAP_MMAP is fine here, since the real work * is done in the FreeBSD mmap(). */ error = fget(td, bsd_args.fd, cap_rights_init(&rights, CAP_MMAP), &fp); if (error != 0) return (error); if (fp->f_type != DTYPE_VNODE) { fdrop(fp, td); return (EINVAL); } /* Linux mmap() just fails for O_WRONLY files */ if (!(fp->f_flag & FREAD)) { fdrop(fp, td); return (EACCES); } fdrop(fp, td); } if (flags & LINUX_MAP_GROWSDOWN) { /* * The Linux MAP_GROWSDOWN option does not limit auto * growth of the region. Linux mmap with this option * takes as addr the inital BOS, and as len, the initial * region size. It can then grow down from addr without * limit. However, linux threads has an implicit internal * limit to stack size of STACK_SIZE. Its just not * enforced explicitly in linux. But, here we impose * a limit of (STACK_SIZE - GUARD_SIZE) on the stack * region, since we can do this with our mmap. * * Our mmap with MAP_STACK takes addr as the maximum * downsize limit on BOS, and as len the max size of * the region. It them maps the top SGROWSIZ bytes, * and auto grows the region down, up to the limit * in addr. * * If we don't use the MAP_STACK option, the effect * of this code is to allocate a stack region of a * fixed size of (STACK_SIZE - GUARD_SIZE). */ if ((caddr_t)PTRIN(addr) + len > p->p_vmspace->vm_maxsaddr) { /* * Some linux apps will attempt to mmap * thread stacks near the top of their * address space. If their TOS is greater * than vm_maxsaddr, vm_map_growstack() * will confuse the thread stack with the * process stack and deliver a SEGV if they * attempt to grow the thread stack past their * current stacksize rlimit. To avoid this, * adjust vm_maxsaddr upwards to reflect * the current stacksize rlimit rather * than the maximum possible stacksize. * It would be better to adjust the * mmap'ed region, but some apps do not check * mmap's return value. */ PROC_LOCK(p); p->p_vmspace->vm_maxsaddr = (char *)USRSTACK - - lim_cur(p, RLIMIT_STACK); + lim_cur_proc(p, RLIMIT_STACK); PROC_UNLOCK(p); } /* * This gives us our maximum stack size and a new BOS. * If we're using VM_STACK, then mmap will just map * the top SGROWSIZ bytes, and let the stack grow down * to the limit at BOS. If we're not using VM_STACK * we map the full stack, since we don't have a way * to autogrow it. */ if (len > STACK_SIZE - GUARD_SIZE) { bsd_args.addr = (caddr_t)PTRIN(addr); bsd_args.len = len; } else { bsd_args.addr = (caddr_t)PTRIN(addr) - (STACK_SIZE - GUARD_SIZE - len); bsd_args.len = STACK_SIZE - GUARD_SIZE; } } else { bsd_args.addr = (caddr_t)PTRIN(addr); bsd_args.len = len; } bsd_args.pos = pos; #ifdef DEBUG if (ldebug(mmap)) printf("-> %s(%p, %d, %d, 0x%08x, %d, 0x%x)\n", __func__, (void *)bsd_args.addr, bsd_args.len, bsd_args.prot, bsd_args.flags, bsd_args.fd, (int)bsd_args.pos); #endif error = sys_mmap(td, &bsd_args); #ifdef DEBUG if (ldebug(mmap)) printf("-> %s() return: 0x%x (0x%08x)\n", __func__, error, (u_int)td->td_retval[0]); #endif return (error); } int linux_mprotect(struct thread *td, struct linux_mprotect_args *uap) { struct mprotect_args bsd_args; bsd_args.addr = uap->addr; bsd_args.len = uap->len; bsd_args.prot = uap->prot; if (bsd_args.prot & (PROT_READ | PROT_WRITE | PROT_EXEC)) bsd_args.prot |= PROT_READ | PROT_EXEC; return (sys_mprotect(td, &bsd_args)); } int linux_ioperm(struct thread *td, struct linux_ioperm_args *args) { int error; struct i386_ioperm_args iia; iia.start = args->start; iia.length = args->length; iia.enable = args->enable; error = i386_set_ioperm(td, &iia); return (error); } int linux_iopl(struct thread *td, struct linux_iopl_args *args) { int error; if (args->level < 0 || args->level > 3) return (EINVAL); if ((error = priv_check(td, PRIV_IO)) != 0) return (error); if ((error = securelevel_gt(td->td_ucred, 0)) != 0) return (error); td->td_frame->tf_eflags = (td->td_frame->tf_eflags & ~PSL_IOPL) | (args->level * (PSL_IOPL / 3)); return (0); } int linux_modify_ldt(struct thread *td, struct linux_modify_ldt_args *uap) { int error; struct i386_ldt_args ldt; struct l_descriptor ld; union descriptor desc; int size, written; switch (uap->func) { case 0x00: /* read_ldt */ ldt.start = 0; ldt.descs = uap->ptr; ldt.num = uap->bytecount / sizeof(union descriptor); error = i386_get_ldt(td, &ldt); td->td_retval[0] *= sizeof(union descriptor); break; case 0x02: /* read_default_ldt = 0 */ size = 5*sizeof(struct l_desc_struct); if (size > uap->bytecount) size = uap->bytecount; for (written = error = 0; written < size && error == 0; written++) error = subyte((char *)uap->ptr + written, 0); td->td_retval[0] = written; break; case 0x01: /* write_ldt */ case 0x11: /* write_ldt */ if (uap->bytecount != sizeof(ld)) return (EINVAL); error = copyin(uap->ptr, &ld, sizeof(ld)); if (error) return (error); ldt.start = ld.entry_number; ldt.descs = &desc; ldt.num = 1; desc.sd.sd_lolimit = (ld.limit & 0x0000ffff); desc.sd.sd_hilimit = (ld.limit & 0x000f0000) >> 16; desc.sd.sd_lobase = (ld.base_addr & 0x00ffffff); desc.sd.sd_hibase = (ld.base_addr & 0xff000000) >> 24; desc.sd.sd_type = SDT_MEMRO | ((ld.read_exec_only ^ 1) << 1) | (ld.contents << 2); desc.sd.sd_dpl = 3; desc.sd.sd_p = (ld.seg_not_present ^ 1); desc.sd.sd_xx = 0; desc.sd.sd_def32 = ld.seg_32bit; desc.sd.sd_gran = ld.limit_in_pages; error = i386_set_ldt(td, &ldt, &desc); break; default: error = ENOSYS; break; } if (error == EOPNOTSUPP) { printf("linux: modify_ldt needs kernel option USER_LDT\n"); error = ENOSYS; } return (error); } int linux_sigaction(struct thread *td, struct linux_sigaction_args *args) { l_osigaction_t osa; l_sigaction_t act, oact; int error; #ifdef DEBUG if (ldebug(sigaction)) printf(ARGS(sigaction, "%d, %p, %p"), args->sig, (void *)args->nsa, (void *)args->osa); #endif if (args->nsa != NULL) { error = copyin(args->nsa, &osa, sizeof(l_osigaction_t)); if (error) return (error); act.lsa_handler = osa.lsa_handler; act.lsa_flags = osa.lsa_flags; act.lsa_restorer = osa.lsa_restorer; LINUX_SIGEMPTYSET(act.lsa_mask); act.lsa_mask.__mask = osa.lsa_mask; } error = linux_do_sigaction(td, args->sig, args->nsa ? &act : NULL, args->osa ? &oact : NULL); if (args->osa != NULL && !error) { osa.lsa_handler = oact.lsa_handler; osa.lsa_flags = oact.lsa_flags; osa.lsa_restorer = oact.lsa_restorer; osa.lsa_mask = oact.lsa_mask.__mask; error = copyout(&osa, args->osa, sizeof(l_osigaction_t)); } return (error); } /* * Linux has two extra args, restart and oldmask. We dont use these, * but it seems that "restart" is actually a context pointer that * enables the signal to happen with a different register set. */ int linux_sigsuspend(struct thread *td, struct linux_sigsuspend_args *args) { sigset_t sigmask; l_sigset_t mask; #ifdef DEBUG if (ldebug(sigsuspend)) printf(ARGS(sigsuspend, "%08lx"), (unsigned long)args->mask); #endif LINUX_SIGEMPTYSET(mask); mask.__mask = args->mask; linux_to_bsd_sigset(&mask, &sigmask); return (kern_sigsuspend(td, sigmask)); } int linux_rt_sigsuspend(struct thread *td, struct linux_rt_sigsuspend_args *uap) { l_sigset_t lmask; sigset_t sigmask; int error; #ifdef DEBUG if (ldebug(rt_sigsuspend)) printf(ARGS(rt_sigsuspend, "%p, %d"), (void *)uap->newset, uap->sigsetsize); #endif if (uap->sigsetsize != sizeof(l_sigset_t)) return (EINVAL); error = copyin(uap->newset, &lmask, sizeof(l_sigset_t)); if (error) return (error); linux_to_bsd_sigset(&lmask, &sigmask); return (kern_sigsuspend(td, sigmask)); } int linux_pause(struct thread *td, struct linux_pause_args *args) { struct proc *p = td->td_proc; sigset_t sigmask; #ifdef DEBUG if (ldebug(pause)) printf(ARGS(pause, "")); #endif PROC_LOCK(p); sigmask = td->td_sigmask; PROC_UNLOCK(p); return (kern_sigsuspend(td, sigmask)); } int linux_sigaltstack(struct thread *td, struct linux_sigaltstack_args *uap) { stack_t ss, oss; l_stack_t lss; int error; #ifdef DEBUG if (ldebug(sigaltstack)) printf(ARGS(sigaltstack, "%p, %p"), uap->uss, uap->uoss); #endif if (uap->uss != NULL) { error = copyin(uap->uss, &lss, sizeof(l_stack_t)); if (error) return (error); ss.ss_sp = lss.ss_sp; ss.ss_size = lss.ss_size; ss.ss_flags = linux_to_bsd_sigaltstack(lss.ss_flags); } error = kern_sigaltstack(td, (uap->uss != NULL) ? &ss : NULL, (uap->uoss != NULL) ? &oss : NULL); if (!error && uap->uoss != NULL) { lss.ss_sp = oss.ss_sp; lss.ss_size = oss.ss_size; lss.ss_flags = bsd_to_linux_sigaltstack(oss.ss_flags); error = copyout(&lss, uap->uoss, sizeof(l_stack_t)); } return (error); } int linux_ftruncate64(struct thread *td, struct linux_ftruncate64_args *args) { struct ftruncate_args sa; #ifdef DEBUG if (ldebug(ftruncate64)) printf(ARGS(ftruncate64, "%u, %jd"), args->fd, (intmax_t)args->length); #endif sa.fd = args->fd; sa.length = args->length; return sys_ftruncate(td, &sa); } int linux_set_thread_area(struct thread *td, struct linux_set_thread_area_args *args) { struct l_user_desc info; int error; int idx; int a[2]; struct segment_descriptor sd; error = copyin(args->desc, &info, sizeof(struct l_user_desc)); if (error) return (error); #ifdef DEBUG if (ldebug(set_thread_area)) printf(ARGS(set_thread_area, "%i, %x, %x, %i, %i, %i, %i, %i, %i\n"), info.entry_number, info.base_addr, info.limit, info.seg_32bit, info.contents, info.read_exec_only, info.limit_in_pages, info.seg_not_present, info.useable); #endif idx = info.entry_number; /* * Semantics of linux version: every thread in the system has array of * 3 tls descriptors. 1st is GLIBC TLS, 2nd is WINE, 3rd unknown. This * syscall loads one of the selected tls decriptors with a value and * also loads GDT descriptors 6, 7 and 8 with the content of the * per-thread descriptors. * * Semantics of fbsd version: I think we can ignore that linux has 3 * per-thread descriptors and use just the 1st one. The tls_array[] * is used only in set/get-thread_area() syscalls and for loading the * GDT descriptors. In fbsd we use just one GDT descriptor for TLS so * we will load just one. * * XXX: this doesn't work when a user space process tries to use more * than 1 TLS segment. Comment in the linux sources says wine might do * this. */ /* * we support just GLIBC TLS now * we should let 3 proceed as well because we use this segment so * if code does two subsequent calls it should succeed */ if (idx != 6 && idx != -1 && idx != 3) return (EINVAL); /* * we have to copy out the GDT entry we use * FreeBSD uses GDT entry #3 for storing %gs so load that * * XXX: what if a user space program doesn't check this value and tries * to use 6, 7 or 8? */ idx = info.entry_number = 3; error = copyout(&info, args->desc, sizeof(struct l_user_desc)); if (error) return (error); if (LINUX_LDT_empty(&info)) { a[0] = 0; a[1] = 0; } else { a[0] = LINUX_LDT_entry_a(&info); a[1] = LINUX_LDT_entry_b(&info); } memcpy(&sd, &a, sizeof(a)); #ifdef DEBUG if (ldebug(set_thread_area)) printf("Segment created in set_thread_area: lobase: %x, hibase: %x, lolimit: %x, hilimit: %x, type: %i, dpl: %i, p: %i, xx: %i, def32: %i, gran: %i\n", sd.sd_lobase, sd.sd_hibase, sd.sd_lolimit, sd.sd_hilimit, sd.sd_type, sd.sd_dpl, sd.sd_p, sd.sd_xx, sd.sd_def32, sd.sd_gran); #endif /* this is taken from i386 version of cpu_set_user_tls() */ critical_enter(); /* set %gs */ td->td_pcb->pcb_gsd = sd; PCPU_GET(fsgs_gdt)[1] = sd; load_gs(GSEL(GUGS_SEL, SEL_UPL)); critical_exit(); return (0); } int linux_get_thread_area(struct thread *td, struct linux_get_thread_area_args *args) { struct l_user_desc info; int error; int idx; struct l_desc_struct desc; struct segment_descriptor sd; #ifdef DEBUG if (ldebug(get_thread_area)) printf(ARGS(get_thread_area, "%p"), args->desc); #endif error = copyin(args->desc, &info, sizeof(struct l_user_desc)); if (error) return (error); idx = info.entry_number; /* XXX: I am not sure if we want 3 to be allowed too. */ if (idx != 6 && idx != 3) return (EINVAL); idx = 3; memset(&info, 0, sizeof(info)); sd = PCPU_GET(fsgs_gdt)[1]; memcpy(&desc, &sd, sizeof(desc)); info.entry_number = idx; info.base_addr = LINUX_GET_BASE(&desc); info.limit = LINUX_GET_LIMIT(&desc); info.seg_32bit = LINUX_GET_32BIT(&desc); info.contents = LINUX_GET_CONTENTS(&desc); info.read_exec_only = !LINUX_GET_WRITABLE(&desc); info.limit_in_pages = LINUX_GET_LIMIT_PAGES(&desc); info.seg_not_present = !LINUX_GET_PRESENT(&desc); info.useable = LINUX_GET_USEABLE(&desc); error = copyout(&info, args->desc, sizeof(struct l_user_desc)); if (error) return (EFAULT); return (0); } /* XXX: this wont work with module - convert it */ int linux_mq_open(struct thread *td, struct linux_mq_open_args *args) { #ifdef P1003_1B_MQUEUE return sys_kmq_open(td, (struct kmq_open_args *) args); #else return (ENOSYS); #endif } int linux_mq_unlink(struct thread *td, struct linux_mq_unlink_args *args) { #ifdef P1003_1B_MQUEUE return sys_kmq_unlink(td, (struct kmq_unlink_args *) args); #else return (ENOSYS); #endif } int linux_mq_timedsend(struct thread *td, struct linux_mq_timedsend_args *args) { #ifdef P1003_1B_MQUEUE return sys_kmq_timedsend(td, (struct kmq_timedsend_args *) args); #else return (ENOSYS); #endif } int linux_mq_timedreceive(struct thread *td, struct linux_mq_timedreceive_args *args) { #ifdef P1003_1B_MQUEUE return sys_kmq_timedreceive(td, (struct kmq_timedreceive_args *) args); #else return (ENOSYS); #endif } int linux_mq_notify(struct thread *td, struct linux_mq_notify_args *args) { #ifdef P1003_1B_MQUEUE return sys_kmq_notify(td, (struct kmq_notify_args *) args); #else return (ENOSYS); #endif } int linux_mq_getsetattr(struct thread *td, struct linux_mq_getsetattr_args *args) { #ifdef P1003_1B_MQUEUE return sys_kmq_setattr(td, (struct kmq_setattr_args *) args); #else return (ENOSYS); #endif } Index: head/sys/kern/imgact_aout.c =================================================================== --- head/sys/kern/imgact_aout.c (revision 284214) +++ head/sys/kern/imgact_aout.c (revision 284215) @@ -1,344 +1,344 @@ /*- * Copyright (c) 1993, David Greenman * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef __amd64__ #include #include #include #include #include #endif static int exec_aout_imgact(struct image_params *imgp); static int aout_fixup(register_t **stack_base, struct image_params *imgp); #if defined(__i386__) struct sysentvec aout_sysvec = { .sv_size = SYS_MAXSYSCALL, .sv_table = sysent, .sv_mask = 0, .sv_sigsize = 0, .sv_sigtbl = NULL, .sv_errsize = 0, .sv_errtbl = NULL, .sv_transtrap = NULL, .sv_fixup = aout_fixup, .sv_sendsig = sendsig, .sv_sigcode = sigcode, .sv_szsigcode = &szsigcode, .sv_prepsyscall = NULL, .sv_name = "FreeBSD a.out", .sv_coredump = NULL, .sv_imgact_try = NULL, .sv_minsigstksz = MINSIGSTKSZ, .sv_pagesize = PAGE_SIZE, .sv_minuser = VM_MIN_ADDRESS, .sv_maxuser = VM_MAXUSER_ADDRESS, .sv_usrstack = USRSTACK, .sv_psstrings = PS_STRINGS, .sv_stackprot = VM_PROT_ALL, .sv_copyout_strings = exec_copyout_strings, .sv_setregs = exec_setregs, .sv_fixlimit = NULL, .sv_maxssiz = NULL, .sv_flags = SV_ABI_FREEBSD | SV_AOUT | SV_IA32 | SV_ILP32, .sv_set_syscall_retval = cpu_set_syscall_retval, .sv_fetch_syscall_args = cpu_fetch_syscall_args, .sv_syscallnames = syscallnames, .sv_schedtail = NULL, .sv_thread_detach = NULL, }; #elif defined(__amd64__) #define AOUT32_USRSTACK 0xbfc00000 #define AOUT32_PS_STRINGS \ (AOUT32_USRSTACK - sizeof(struct freebsd32_ps_strings)) #define AOUT32_MINUSER FREEBSD32_MINUSER extern const char *freebsd32_syscallnames[]; extern u_long ia32_maxssiz; struct sysentvec aout_sysvec = { .sv_size = FREEBSD32_SYS_MAXSYSCALL, .sv_table = freebsd32_sysent, .sv_mask = 0, .sv_sigsize = 0, .sv_sigtbl = NULL, .sv_errsize = 0, .sv_errtbl = NULL, .sv_transtrap = NULL, .sv_fixup = aout_fixup, .sv_sendsig = ia32_sendsig, .sv_sigcode = ia32_sigcode, .sv_szsigcode = &sz_ia32_sigcode, .sv_prepsyscall = NULL, .sv_name = "FreeBSD a.out", .sv_coredump = NULL, .sv_imgact_try = NULL, .sv_minsigstksz = MINSIGSTKSZ, .sv_pagesize = IA32_PAGE_SIZE, .sv_minuser = AOUT32_MINUSER, .sv_maxuser = AOUT32_USRSTACK, .sv_usrstack = AOUT32_USRSTACK, .sv_psstrings = AOUT32_PS_STRINGS, .sv_stackprot = VM_PROT_ALL, .sv_copyout_strings = freebsd32_copyout_strings, .sv_setregs = ia32_setregs, .sv_fixlimit = ia32_fixlimit, .sv_maxssiz = &ia32_maxssiz, .sv_flags = SV_ABI_FREEBSD | SV_AOUT | SV_IA32 | SV_ILP32, .sv_set_syscall_retval = ia32_set_syscall_retval, .sv_fetch_syscall_args = ia32_fetch_syscall_args, .sv_syscallnames = freebsd32_syscallnames, }; #else #error "Port me" #endif static int aout_fixup(register_t **stack_base, struct image_params *imgp) { *(char **)stack_base -= sizeof(uint32_t); return (suword32(*stack_base, imgp->args->argc)); } static int exec_aout_imgact(struct image_params *imgp) { const struct exec *a_out = (const struct exec *) imgp->image_header; struct vmspace *vmspace; vm_map_t map; vm_object_t object; vm_offset_t text_end, data_end; unsigned long virtual_offset; unsigned long file_offset; unsigned long bss_size; int error; /* * Linux and *BSD binaries look very much alike, * only the machine id is different: * 0x64 for Linux, 0x86 for *BSD, 0x00 for BSDI. * NetBSD is in network byte order.. ugh. */ if (((a_out->a_midmag >> 16) & 0xff) != 0x86 && ((a_out->a_midmag >> 16) & 0xff) != 0 && ((((int)ntohl(a_out->a_midmag)) >> 16) & 0xff) != 0x86) return -1; /* * Set file/virtual offset based on a.out variant. * We do two cases: host byte order and network byte order * (for NetBSD compatibility) */ switch ((int)(a_out->a_midmag & 0xffff)) { case ZMAGIC: virtual_offset = 0; if (a_out->a_text) { file_offset = PAGE_SIZE; } else { /* Bill's "screwball mode" */ file_offset = 0; } break; case QMAGIC: virtual_offset = PAGE_SIZE; file_offset = 0; /* Pass PS_STRINGS for BSD/OS binaries only. */ if (N_GETMID(*a_out) == MID_ZERO) imgp->ps_strings = aout_sysvec.sv_psstrings; break; default: /* NetBSD compatibility */ switch ((int)(ntohl(a_out->a_midmag) & 0xffff)) { case ZMAGIC: case QMAGIC: virtual_offset = PAGE_SIZE; file_offset = 0; break; default: return (-1); } } bss_size = roundup(a_out->a_bss, PAGE_SIZE); /* * Check various fields in header for validity/bounds. */ if (/* entry point must lay with text region */ a_out->a_entry < virtual_offset || a_out->a_entry >= virtual_offset + a_out->a_text || /* text and data size must each be page rounded */ a_out->a_text & PAGE_MASK || a_out->a_data & PAGE_MASK #ifdef __amd64__ || /* overflows */ virtual_offset + a_out->a_text + a_out->a_data + bss_size > UINT_MAX #endif ) return (-1); /* text + data can't exceed file size */ if (a_out->a_data + a_out->a_text > imgp->attr->va_size) return (EFAULT); /* * text/data/bss must not exceed limits */ PROC_LOCK(imgp->proc); if (/* text can't exceed maximum text size */ a_out->a_text > maxtsiz || /* data + bss can't exceed rlimit */ - a_out->a_data + bss_size > lim_cur(imgp->proc, RLIMIT_DATA) || + a_out->a_data + bss_size > lim_cur_proc(imgp->proc, RLIMIT_DATA) || racct_set(imgp->proc, RACCT_DATA, a_out->a_data + bss_size) != 0) { PROC_UNLOCK(imgp->proc); return (ENOMEM); } PROC_UNLOCK(imgp->proc); /* * Avoid a possible deadlock if the current address space is destroyed * and that address space maps the locked vnode. In the common case, * the locked vnode's v_usecount is decremented but remains greater * than zero. Consequently, the vnode lock is not needed by vrele(). * However, in cases where the vnode lock is external, such as nullfs, * v_usecount may become zero. */ VOP_UNLOCK(imgp->vp, 0); /* * Destroy old process VM and create a new one (with a new stack) */ error = exec_new_vmspace(imgp, &aout_sysvec); vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); if (error) return (error); /* * The vm space can be changed by exec_new_vmspace */ vmspace = imgp->proc->p_vmspace; object = imgp->object; map = &vmspace->vm_map; vm_map_lock(map); vm_object_reference(object); text_end = virtual_offset + a_out->a_text; error = vm_map_insert(map, object, file_offset, virtual_offset, text_end, VM_PROT_READ | VM_PROT_EXECUTE, VM_PROT_ALL, MAP_COPY_ON_WRITE | MAP_PREFAULT); if (error) { vm_map_unlock(map); vm_object_deallocate(object); return (error); } data_end = text_end + a_out->a_data; if (a_out->a_data) { vm_object_reference(object); error = vm_map_insert(map, object, file_offset + a_out->a_text, text_end, data_end, VM_PROT_ALL, VM_PROT_ALL, MAP_COPY_ON_WRITE | MAP_PREFAULT); if (error) { vm_map_unlock(map); vm_object_deallocate(object); return (error); } } if (bss_size) { error = vm_map_insert(map, NULL, 0, data_end, data_end + bss_size, VM_PROT_ALL, VM_PROT_ALL, 0); if (error) { vm_map_unlock(map); return (error); } } vm_map_unlock(map); /* Fill in process VM information */ vmspace->vm_tsize = a_out->a_text >> PAGE_SHIFT; vmspace->vm_dsize = (a_out->a_data + bss_size) >> PAGE_SHIFT; vmspace->vm_taddr = (caddr_t) (uintptr_t) virtual_offset; vmspace->vm_daddr = (caddr_t) (uintptr_t) (virtual_offset + a_out->a_text); /* Fill in image_params */ imgp->interpreted = 0; imgp->entry_addr = a_out->a_entry; imgp->proc->p_sysent = &aout_sysvec; return (0); } /* * Tell kern_execve.c about it, with a little help from the linker. */ static struct execsw aout_execsw = { exec_aout_imgact, "a.out" }; EXEC_SET(aout, aout_execsw); Index: head/sys/kern/imgact_elf.c =================================================================== --- head/sys/kern/imgact_elf.c (revision 284214) +++ head/sys/kern/imgact_elf.c (revision 284215) @@ -1,2178 +1,2178 @@ /*- * Copyright (c) 2000 David O'Brien * Copyright (c) 1995-1996 Søren Schmidt * Copyright (c) 1996 Peter Wemm * 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 * in this position and unchanged. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_capsicum.h" #include "opt_compat.h" #include "opt_gzio.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define ELF_NOTE_ROUNDSIZE 4 #define OLD_EI_BRAND 8 static int __elfN(check_header)(const Elf_Ehdr *hdr); static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp, const char *interp, int interp_name_len, int32_t *osrel); static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr, u_long *entry, size_t pagesize); static int __elfN(load_section)(struct image_params *imgp, vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot, size_t pagesize); static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp); static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel); static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel); static boolean_t __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote, int32_t *osrel); static vm_prot_t __elfN(trans_prot)(Elf_Word); static Elf_Word __elfN(untrans_prot)(vm_prot_t); SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0, ""); #define CORE_BUF_SIZE (16 * 1024) int __elfN(fallback_brand) = -1; SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0, __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort"); static int elf_legacy_coredump = 0; SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, &elf_legacy_coredump, 0, ""); int __elfN(nxstack) = #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ 1; #else 0; #endif SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, nxstack, CTLFLAG_RW, &__elfN(nxstack), 0, __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack"); #if __ELF_WORD_SIZE == 32 #if defined(__amd64__) int i386_read_exec = 0; SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0, "enable execution from readable segments"); #endif #endif static Elf_Brandinfo *elf_brand_list[MAX_BRANDS]; #define trunc_page_ps(va, ps) ((va) & ~(ps - 1)) #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1)) #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a)) static const char FREEBSD_ABI_VENDOR[] = "FreeBSD"; Elf_Brandnote __elfN(freebsd_brandnote) = { .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR), .hdr.n_descsz = sizeof(int32_t), .hdr.n_type = 1, .vendor = FREEBSD_ABI_VENDOR, .flags = BN_TRANSLATE_OSREL, .trans_osrel = __elfN(freebsd_trans_osrel) }; static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel) { uintptr_t p; p = (uintptr_t)(note + 1); p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); *osrel = *(const int32_t *)(p); return (TRUE); } static const char GNU_ABI_VENDOR[] = "GNU"; static int GNU_KFREEBSD_ABI_DESC = 3; Elf_Brandnote __elfN(kfreebsd_brandnote) = { .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), .hdr.n_descsz = 16, /* XXX at least 16 */ .hdr.n_type = 1, .vendor = GNU_ABI_VENDOR, .flags = BN_TRANSLATE_OSREL, .trans_osrel = kfreebsd_trans_osrel }; static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel) { const Elf32_Word *desc; uintptr_t p; p = (uintptr_t)(note + 1); p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE); desc = (const Elf32_Word *)p; if (desc[0] != GNU_KFREEBSD_ABI_DESC) return (FALSE); /* * Debian GNU/kFreeBSD embed the earliest compatible kernel version * (__FreeBSD_version: Rxx) in the LSB way. */ *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3]; return (TRUE); } int __elfN(insert_brand_entry)(Elf_Brandinfo *entry) { int i; for (i = 0; i < MAX_BRANDS; i++) { if (elf_brand_list[i] == NULL) { elf_brand_list[i] = entry; break; } } if (i == MAX_BRANDS) { printf("WARNING: %s: could not insert brandinfo entry: %p\n", __func__, entry); return (-1); } return (0); } int __elfN(remove_brand_entry)(Elf_Brandinfo *entry) { int i; for (i = 0; i < MAX_BRANDS; i++) { if (elf_brand_list[i] == entry) { elf_brand_list[i] = NULL; break; } } if (i == MAX_BRANDS) return (-1); return (0); } int __elfN(brand_inuse)(Elf_Brandinfo *entry) { struct proc *p; int rval = FALSE; sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { if (p->p_sysent == entry->sysvec) { rval = TRUE; break; } } sx_sunlock(&allproc_lock); return (rval); } static Elf_Brandinfo * __elfN(get_brandinfo)(struct image_params *imgp, const char *interp, int interp_name_len, int32_t *osrel) { const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; Elf_Brandinfo *bi; boolean_t ret; int i; /* * We support four types of branding -- (1) the ELF EI_OSABI field * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string * branding w/in the ELF header, (3) path of the `interp_path' * field, and (4) the ".note.ABI-tag" ELF section. */ /* Look for an ".note.ABI-tag" ELF section */ for (i = 0; i < MAX_BRANDS; i++) { bi = elf_brand_list[i]; if (bi == NULL) continue; if (hdr->e_machine == bi->machine && (bi->flags & (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) { ret = __elfN(check_note)(imgp, bi->brand_note, osrel); if (ret) return (bi); } } /* If the executable has a brand, search for it in the brand list. */ for (i = 0; i < MAX_BRANDS; i++) { bi = elf_brand_list[i]; if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) continue; if (hdr->e_machine == bi->machine && (hdr->e_ident[EI_OSABI] == bi->brand || strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND], bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) return (bi); } /* No known brand, see if the header is recognized by any brand */ for (i = 0; i < MAX_BRANDS; i++) { bi = elf_brand_list[i]; if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY || bi->header_supported == NULL) continue; if (hdr->e_machine == bi->machine) { ret = bi->header_supported(imgp); if (ret) return (bi); } } /* Lacking a known brand, search for a recognized interpreter. */ if (interp != NULL) { for (i = 0; i < MAX_BRANDS; i++) { bi = elf_brand_list[i]; if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) continue; if (hdr->e_machine == bi->machine && /* ELF image p_filesz includes terminating zero */ strlen(bi->interp_path) + 1 == interp_name_len && strncmp(interp, bi->interp_path, interp_name_len) == 0) return (bi); } } /* Lacking a recognized interpreter, try the default brand */ for (i = 0; i < MAX_BRANDS; i++) { bi = elf_brand_list[i]; if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY) continue; if (hdr->e_machine == bi->machine && __elfN(fallback_brand) == bi->brand) return (bi); } return (NULL); } static int __elfN(check_header)(const Elf_Ehdr *hdr) { Elf_Brandinfo *bi; int i; if (!IS_ELF(*hdr) || hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || hdr->e_ident[EI_DATA] != ELF_TARG_DATA || hdr->e_ident[EI_VERSION] != EV_CURRENT || hdr->e_phentsize != sizeof(Elf_Phdr) || hdr->e_version != ELF_TARG_VER) return (ENOEXEC); /* * Make sure we have at least one brand for this machine. */ for (i = 0; i < MAX_BRANDS; i++) { bi = elf_brand_list[i]; if (bi != NULL && bi->machine == hdr->e_machine) break; } if (i == MAX_BRANDS) return (ENOEXEC); return (0); } static int __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot) { struct sf_buf *sf; int error; vm_offset_t off; /* * Create the page if it doesn't exist yet. Ignore errors. */ vm_map_lock(map); vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end), VM_PROT_ALL, VM_PROT_ALL, 0); vm_map_unlock(map); /* * Find the page from the underlying object. */ if (object) { sf = vm_imgact_map_page(object, offset); if (sf == NULL) return (KERN_FAILURE); off = offset - trunc_page(offset); error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start, end - start); vm_imgact_unmap_page(sf); if (error) { return (KERN_FAILURE); } } return (KERN_SUCCESS); } static int __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow) { struct sf_buf *sf; vm_offset_t off; vm_size_t sz; int error, rv; if (start != trunc_page(start)) { rv = __elfN(map_partial)(map, object, offset, start, round_page(start), prot); if (rv) return (rv); offset += round_page(start) - start; start = round_page(start); } if (end != round_page(end)) { rv = __elfN(map_partial)(map, object, offset + trunc_page(end) - start, trunc_page(end), end, prot); if (rv) return (rv); end = trunc_page(end); } if (end > start) { if (offset & PAGE_MASK) { /* * The mapping is not page aligned. This means we have * to copy the data. Sigh. */ rv = vm_map_find(map, NULL, 0, &start, end - start, 0, VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0); if (rv) return (rv); if (object == NULL) return (KERN_SUCCESS); for (; start < end; start += sz) { sf = vm_imgact_map_page(object, offset); if (sf == NULL) return (KERN_FAILURE); off = offset - trunc_page(offset); sz = end - start; if (sz > PAGE_SIZE - off) sz = PAGE_SIZE - off; error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start, sz); vm_imgact_unmap_page(sf); if (error) { return (KERN_FAILURE); } offset += sz; } rv = KERN_SUCCESS; } else { vm_object_reference(object); vm_map_lock(map); rv = vm_map_insert(map, object, offset, start, end, prot, VM_PROT_ALL, cow); vm_map_unlock(map); if (rv != KERN_SUCCESS) vm_object_deallocate(object); } return (rv); } else { return (KERN_SUCCESS); } } static int __elfN(load_section)(struct image_params *imgp, vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot, size_t pagesize) { struct sf_buf *sf; size_t map_len; vm_map_t map; vm_object_t object; vm_offset_t map_addr; int error, rv, cow; size_t copy_len; vm_offset_t file_addr; /* * It's necessary to fail if the filsz + offset taken from the * header is greater than the actual file pager object's size. * If we were to allow this, then the vm_map_find() below would * walk right off the end of the file object and into the ether. * * While I'm here, might as well check for something else that * is invalid: filsz cannot be greater than memsz. */ if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) { uprintf("elf_load_section: truncated ELF file\n"); return (ENOEXEC); } object = imgp->object; map = &imgp->proc->p_vmspace->vm_map; map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize); file_addr = trunc_page_ps(offset, pagesize); /* * We have two choices. We can either clear the data in the last page * of an oversized mapping, or we can start the anon mapping a page * early and copy the initialized data into that first page. We * choose the second.. */ if (memsz > filsz) map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr; else map_len = round_page_ps(offset + filsz, pagesize) - file_addr; if (map_len != 0) { /* cow flags: don't dump readonly sections in core */ cow = MAP_COPY_ON_WRITE | MAP_PREFAULT | (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP); rv = __elfN(map_insert)(map, object, file_addr, /* file offset */ map_addr, /* virtual start */ map_addr + map_len,/* virtual end */ prot, cow); if (rv != KERN_SUCCESS) return (EINVAL); /* we can stop now if we've covered it all */ if (memsz == filsz) { return (0); } } /* * We have to get the remaining bit of the file into the first part * of the oversized map segment. This is normally because the .data * segment in the file is extended to provide bss. It's a neat idea * to try and save a page, but it's a pain in the behind to implement. */ copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize); map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize); map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) - map_addr; /* This had damn well better be true! */ if (map_len != 0) { rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr + map_len, VM_PROT_ALL, 0); if (rv != KERN_SUCCESS) { return (EINVAL); } } if (copy_len != 0) { vm_offset_t off; sf = vm_imgact_map_page(object, offset + filsz); if (sf == NULL) return (EIO); /* send the page fragment to user space */ off = trunc_page_ps(offset + filsz, pagesize) - trunc_page(offset + filsz); error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)map_addr, copy_len); vm_imgact_unmap_page(sf); if (error) { return (error); } } /* * set it to the specified protection. * XXX had better undo the damage from pasting over the cracks here! */ vm_map_protect(map, trunc_page(map_addr), round_page(map_addr + map_len), prot, FALSE); return (0); } /* * Load the file "file" into memory. It may be either a shared object * or an executable. * * The "addr" reference parameter is in/out. On entry, it specifies * the address where a shared object should be loaded. If the file is * an executable, this value is ignored. On exit, "addr" specifies * where the file was actually loaded. * * The "entry" reference parameter is out only. On exit, it specifies * the entry point for the loaded file. */ static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr, u_long *entry, size_t pagesize) { struct { struct nameidata nd; struct vattr attr; struct image_params image_params; } *tempdata; const Elf_Ehdr *hdr = NULL; const Elf_Phdr *phdr = NULL; struct nameidata *nd; struct vattr *attr; struct image_params *imgp; vm_prot_t prot; u_long rbase; u_long base_addr = 0; int error, i, numsegs; #ifdef CAPABILITY_MODE /* * XXXJA: This check can go away once we are sufficiently confident * that the checks in namei() are correct. */ if (IN_CAPABILITY_MODE(curthread)) return (ECAPMODE); #endif tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK); nd = &tempdata->nd; attr = &tempdata->attr; imgp = &tempdata->image_params; /* * Initialize part of the common data */ imgp->proc = p; imgp->attr = attr; imgp->firstpage = NULL; imgp->image_header = NULL; imgp->object = NULL; imgp->execlabel = NULL; NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread); if ((error = namei(nd)) != 0) { nd->ni_vp = NULL; goto fail; } NDFREE(nd, NDF_ONLY_PNBUF); imgp->vp = nd->ni_vp; /* * Check permissions, modes, uid, etc on the file, and "open" it. */ error = exec_check_permissions(imgp); if (error) goto fail; error = exec_map_first_page(imgp); if (error) goto fail; /* * Also make certain that the interpreter stays the same, so set * its VV_TEXT flag, too. */ VOP_SET_TEXT(nd->ni_vp); imgp->object = nd->ni_vp->v_object; hdr = (const Elf_Ehdr *)imgp->image_header; if ((error = __elfN(check_header)(hdr)) != 0) goto fail; if (hdr->e_type == ET_DYN) rbase = *addr; else if (hdr->e_type == ET_EXEC) rbase = 0; else { error = ENOEXEC; goto fail; } /* Only support headers that fit within first page for now */ if ((hdr->e_phoff > PAGE_SIZE) || (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) { error = ENOEXEC; goto fail; } phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); if (!aligned(phdr, Elf_Addr)) { error = ENOEXEC; goto fail; } for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) { if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) { /* Loadable segment */ prot = __elfN(trans_prot)(phdr[i].p_flags); error = __elfN(load_section)(imgp, phdr[i].p_offset, (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase, phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize); if (error != 0) goto fail; /* * Establish the base address if this is the * first segment. */ if (numsegs == 0) base_addr = trunc_page(phdr[i].p_vaddr + rbase); numsegs++; } } *addr = base_addr; *entry = (unsigned long)hdr->e_entry + rbase; fail: if (imgp->firstpage) exec_unmap_first_page(imgp); if (nd->ni_vp) vput(nd->ni_vp); free(tempdata, M_TEMP); return (error); } static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp) { const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header; const Elf_Phdr *phdr; Elf_Auxargs *elf_auxargs; struct vmspace *vmspace; vm_prot_t prot; u_long text_size = 0, data_size = 0, total_size = 0; u_long text_addr = 0, data_addr = 0; u_long seg_size, seg_addr; u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0; int32_t osrel = 0; int error = 0, i, n, interp_name_len = 0; const char *err_str = NULL, *interp = NULL, *newinterp = NULL; Elf_Brandinfo *brand_info; char *path; struct sysentvec *sv; /* * Do we have a valid ELF header ? * * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later * if particular brand doesn't support it. */ if (__elfN(check_header)(hdr) != 0 || (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)) return (-1); /* * From here on down, we return an errno, not -1, as we've * detected an ELF file. */ if ((hdr->e_phoff > PAGE_SIZE) || (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) { /* Only support headers in first page for now */ uprintf("Program headers not in the first page\n"); return (ENOEXEC); } phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); if (!aligned(phdr, Elf_Addr)) { uprintf("Unaligned program headers\n"); return (ENOEXEC); } n = 0; baddr = 0; for (i = 0; i < hdr->e_phnum; i++) { switch (phdr[i].p_type) { case PT_LOAD: if (n == 0) baddr = phdr[i].p_vaddr; n++; break; case PT_INTERP: /* Path to interpreter */ if (phdr[i].p_filesz > MAXPATHLEN || phdr[i].p_offset > PAGE_SIZE || phdr[i].p_filesz > PAGE_SIZE - phdr[i].p_offset) { uprintf("Invalid PT_INTERP\n"); return (ENOEXEC); } interp = imgp->image_header + phdr[i].p_offset; interp_name_len = phdr[i].p_filesz; break; case PT_GNU_STACK: if (__elfN(nxstack)) imgp->stack_prot = __elfN(trans_prot)(phdr[i].p_flags); imgp->stack_sz = phdr[i].p_memsz; break; } } brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len, &osrel); if (brand_info == NULL) { uprintf("ELF binary type \"%u\" not known.\n", hdr->e_ident[EI_OSABI]); return (ENOEXEC); } if (hdr->e_type == ET_DYN) { if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) { uprintf("Cannot execute shared object\n"); return (ENOEXEC); } /* * Honour the base load address from the dso if it is * non-zero for some reason. */ if (baddr == 0) et_dyn_addr = ET_DYN_LOAD_ADDR; else et_dyn_addr = 0; } else et_dyn_addr = 0; sv = brand_info->sysvec; if (interp != NULL && brand_info->interp_newpath != NULL) newinterp = brand_info->interp_newpath; /* * Avoid a possible deadlock if the current address space is destroyed * and that address space maps the locked vnode. In the common case, * the locked vnode's v_usecount is decremented but remains greater * than zero. Consequently, the vnode lock is not needed by vrele(). * However, in cases where the vnode lock is external, such as nullfs, * v_usecount may become zero. * * The VV_TEXT flag prevents modifications to the executable while * the vnode is unlocked. */ VOP_UNLOCK(imgp->vp, 0); error = exec_new_vmspace(imgp, sv); imgp->proc->p_sysent = sv; vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); if (error) return (error); for (i = 0; i < hdr->e_phnum; i++) { switch (phdr[i].p_type) { case PT_LOAD: /* Loadable segment */ if (phdr[i].p_memsz == 0) break; prot = __elfN(trans_prot)(phdr[i].p_flags); error = __elfN(load_section)(imgp, phdr[i].p_offset, (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr, phdr[i].p_memsz, phdr[i].p_filesz, prot, sv->sv_pagesize); if (error != 0) return (error); /* * If this segment contains the program headers, * remember their virtual address for the AT_PHDR * aux entry. Static binaries don't usually include * a PT_PHDR entry. */ if (phdr[i].p_offset == 0 && hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize <= phdr[i].p_filesz) proghdr = phdr[i].p_vaddr + hdr->e_phoff + et_dyn_addr; seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr); seg_size = round_page(phdr[i].p_memsz + phdr[i].p_vaddr + et_dyn_addr - seg_addr); /* * Make the largest executable segment the official * text segment and all others data. * * Note that obreak() assumes that data_addr + * data_size == end of data load area, and the ELF * file format expects segments to be sorted by * address. If multiple data segments exist, the * last one will be used. */ if (phdr[i].p_flags & PF_X && text_size < seg_size) { text_size = seg_size; text_addr = seg_addr; } else { data_size = seg_size; data_addr = seg_addr; } total_size += seg_size; break; case PT_PHDR: /* Program header table info */ proghdr = phdr[i].p_vaddr + et_dyn_addr; break; default: break; } } if (data_addr == 0 && data_size == 0) { data_addr = text_addr; data_size = text_size; } entry = (u_long)hdr->e_entry + et_dyn_addr; /* * Check limits. It should be safe to check the * limits after loading the segments since we do * not actually fault in all the segments pages. */ PROC_LOCK(imgp->proc); - if (data_size > lim_cur(imgp->proc, RLIMIT_DATA)) + if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA)) err_str = "Data segment size exceeds process limit"; else if (text_size > maxtsiz) err_str = "Text segment size exceeds system limit"; - else if (total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) + else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM)) err_str = "Total segment size exceeds process limit"; else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0) err_str = "Data segment size exceeds resource limit"; else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) err_str = "Total segment size exceeds resource limit"; if (err_str != NULL) { PROC_UNLOCK(imgp->proc); uprintf("%s\n", err_str); return (ENOMEM); } vmspace = imgp->proc->p_vmspace; vmspace->vm_tsize = text_size >> PAGE_SHIFT; vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr; vmspace->vm_dsize = data_size >> PAGE_SHIFT; vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr; /* * We load the dynamic linker where a userland call * to mmap(0, ...) would put it. The rationale behind this * calculation is that it leaves room for the heap to grow to * its maximum allowed size. */ - addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(imgp->proc, + addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(curthread, RLIMIT_DATA)); PROC_UNLOCK(imgp->proc); imgp->entry_addr = entry; if (interp != NULL) { int have_interp = FALSE; VOP_UNLOCK(imgp->vp, 0); if (brand_info->emul_path != NULL && brand_info->emul_path[0] != '\0') { path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); snprintf(path, MAXPATHLEN, "%s%s", brand_info->emul_path, interp); error = __elfN(load_file)(imgp->proc, path, &addr, &imgp->entry_addr, sv->sv_pagesize); free(path, M_TEMP); if (error == 0) have_interp = TRUE; } if (!have_interp && newinterp != NULL) { error = __elfN(load_file)(imgp->proc, newinterp, &addr, &imgp->entry_addr, sv->sv_pagesize); if (error == 0) have_interp = TRUE; } if (!have_interp) { error = __elfN(load_file)(imgp->proc, interp, &addr, &imgp->entry_addr, sv->sv_pagesize); } vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); if (error != 0) { uprintf("ELF interpreter %s not found\n", interp); return (error); } } else addr = et_dyn_addr; /* * Construct auxargs table (used by the fixup routine) */ elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK); elf_auxargs->execfd = -1; elf_auxargs->phdr = proghdr; elf_auxargs->phent = hdr->e_phentsize; elf_auxargs->phnum = hdr->e_phnum; elf_auxargs->pagesz = PAGE_SIZE; elf_auxargs->base = addr; elf_auxargs->flags = 0; elf_auxargs->entry = entry; elf_auxargs->hdr_eflags = hdr->e_flags; imgp->auxargs = elf_auxargs; imgp->interpreted = 0; imgp->reloc_base = addr; imgp->proc->p_osrel = osrel; return (error); } #define suword __CONCAT(suword, __ELF_WORD_SIZE) int __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp) { Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs; Elf_Addr *base; Elf_Addr *pos; base = (Elf_Addr *)*stack_base; pos = base + (imgp->args->argc + imgp->args->envc + 2); if (args->execfd != -1) AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); AUXARGS_ENTRY(pos, AT_PHENT, args->phent); AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); AUXARGS_ENTRY(pos, AT_BASE, args->base); #ifdef AT_EHDRFLAGS AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags); #endif if (imgp->execpathp != 0) AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp); AUXARGS_ENTRY(pos, AT_OSRELDATE, imgp->proc->p_ucred->cr_prison->pr_osreldate); if (imgp->canary != 0) { AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary); AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen); } AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus); if (imgp->pagesizes != 0) { AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes); AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen); } if (imgp->sysent->sv_timekeep_base != 0) { AUXARGS_ENTRY(pos, AT_TIMEKEEP, imgp->sysent->sv_timekeep_base); } AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : imgp->sysent->sv_stackprot); AUXARGS_ENTRY(pos, AT_NULL, 0); free(imgp->auxargs, M_TEMP); imgp->auxargs = NULL; base--; suword(base, (long)imgp->args->argc); *stack_base = (register_t *)base; return (0); } /* * Code for generating ELF core dumps. */ typedef void (*segment_callback)(vm_map_entry_t, void *); /* Closure for cb_put_phdr(). */ struct phdr_closure { Elf_Phdr *phdr; /* Program header to fill in */ Elf_Off offset; /* Offset of segment in core file */ }; /* Closure for cb_size_segment(). */ struct sseg_closure { int count; /* Count of writable segments. */ size_t size; /* Total size of all writable segments. */ }; typedef void (*outfunc_t)(void *, struct sbuf *, size_t *); struct note_info { int type; /* Note type. */ outfunc_t outfunc; /* Output function. */ void *outarg; /* Argument for the output function. */ size_t outsize; /* Output size. */ TAILQ_ENTRY(note_info) link; /* Link to the next note info. */ }; TAILQ_HEAD(note_info_list, note_info); /* Coredump output parameters. */ struct coredump_params { off_t offset; struct ucred *active_cred; struct ucred *file_cred; struct thread *td; struct vnode *vp; struct gzio_stream *gzs; }; static void cb_put_phdr(vm_map_entry_t, void *); static void cb_size_segment(vm_map_entry_t, void *); static int core_write(struct coredump_params *, void *, size_t, off_t, enum uio_seg); static void each_writable_segment(struct thread *, segment_callback, void *); static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t, struct note_info_list *, size_t); static void __elfN(prepare_notes)(struct thread *, struct note_info_list *, size_t *); static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t); static void __elfN(putnote)(struct note_info *, struct sbuf *); static size_t register_note(struct note_info_list *, int, outfunc_t, void *); static int sbuf_drain_core_output(void *, const char *, int); static int sbuf_drain_count(void *arg, const char *data, int len); static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *); static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *); static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *); static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *); static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *); static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *); static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *); static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *); static void note_procstat_files(void *, struct sbuf *, size_t *); static void note_procstat_groups(void *, struct sbuf *, size_t *); static void note_procstat_osrel(void *, struct sbuf *, size_t *); static void note_procstat_rlimit(void *, struct sbuf *, size_t *); static void note_procstat_umask(void *, struct sbuf *, size_t *); static void note_procstat_vmmap(void *, struct sbuf *, size_t *); #ifdef GZIO extern int compress_user_cores_gzlevel; /* * Write out a core segment to the compression stream. */ static int compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len) { u_int chunk_len; int error; while (len > 0) { chunk_len = MIN(len, CORE_BUF_SIZE); copyin(base, buf, chunk_len); error = gzio_write(p->gzs, buf, chunk_len); if (error != 0) break; base += chunk_len; len -= chunk_len; } return (error); } static int core_gz_write(void *base, size_t len, off_t offset, void *arg) { return (core_write((struct coredump_params *)arg, base, len, offset, UIO_SYSSPACE)); } #endif /* GZIO */ static int core_write(struct coredump_params *p, void *base, size_t len, off_t offset, enum uio_seg seg) { return (vn_rdwr_inchunks(UIO_WRITE, p->vp, base, len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED, p->active_cred, p->file_cred, NULL, p->td)); } static int core_output(void *base, size_t len, off_t offset, struct coredump_params *p, void *tmpbuf) { #ifdef GZIO if (p->gzs != NULL) return (compress_chunk(p, base, tmpbuf, len)); #endif return (core_write(p, base, len, offset, UIO_USERSPACE)); } /* * Drain into a core file. */ static int sbuf_drain_core_output(void *arg, const char *data, int len) { struct coredump_params *p; int error, locked; p = (struct coredump_params *)arg; /* * Some kern_proc out routines that print to this sbuf may * call us with the process lock held. Draining with the * non-sleepable lock held is unsafe. The lock is needed for * those routines when dumping a live process. In our case we * can safely release the lock before draining and acquire * again after. */ locked = PROC_LOCKED(p->td->td_proc); if (locked) PROC_UNLOCK(p->td->td_proc); #ifdef GZIO if (p->gzs != NULL) error = gzio_write(p->gzs, __DECONST(char *, data), len); else #endif error = core_write(p, __DECONST(void *, data), len, p->offset, UIO_SYSSPACE); if (locked) PROC_LOCK(p->td->td_proc); if (error != 0) return (-error); p->offset += len; return (len); } /* * Drain into a counter. */ static int sbuf_drain_count(void *arg, const char *data __unused, int len) { size_t *sizep; sizep = (size_t *)arg; *sizep += len; return (len); } int __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags) { struct ucred *cred = td->td_ucred; int error = 0; struct sseg_closure seginfo; struct note_info_list notelst; struct coredump_params params; struct note_info *ninfo; void *hdr, *tmpbuf; size_t hdrsize, notesz, coresize; boolean_t compress; compress = (flags & IMGACT_CORE_COMPRESS) != 0; hdr = NULL; TAILQ_INIT(¬elst); /* Size the program segments. */ seginfo.count = 0; seginfo.size = 0; each_writable_segment(td, cb_size_segment, &seginfo); /* * Collect info about the core file header area. */ hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count); __elfN(prepare_notes)(td, ¬elst, ¬esz); coresize = round_page(hdrsize + notesz) + seginfo.size; #ifdef RACCT if (racct_enable) { PROC_LOCK(td->td_proc); error = racct_add(td->td_proc, RACCT_CORE, coresize); PROC_UNLOCK(td->td_proc); if (error != 0) { error = EFAULT; goto done; } } #endif if (coresize >= limit) { error = EFAULT; goto done; } /* Set up core dump parameters. */ params.offset = 0; params.active_cred = cred; params.file_cred = NOCRED; params.td = td; params.vp = vp; params.gzs = NULL; tmpbuf = NULL; #ifdef GZIO /* Create a compression stream if necessary. */ if (compress) { params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE, CORE_BUF_SIZE, compress_user_cores_gzlevel, ¶ms); if (params.gzs == NULL) { error = EFAULT; goto done; } tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO); } #endif /* * Allocate memory for building the header, fill it up, * and write it out following the notes. */ hdr = malloc(hdrsize, M_TEMP, M_WAITOK); if (hdr == NULL) { error = EINVAL; goto done; } error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst, notesz); /* Write the contents of all of the writable segments. */ if (error == 0) { Elf_Phdr *php; off_t offset; int i; php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; offset = round_page(hdrsize + notesz); for (i = 0; i < seginfo.count; i++) { error = core_output((caddr_t)(uintptr_t)php->p_vaddr, php->p_filesz, offset, ¶ms, tmpbuf); if (error != 0) break; offset += php->p_filesz; php++; } #ifdef GZIO if (error == 0 && compress) error = gzio_flush(params.gzs); #endif } if (error) { log(LOG_WARNING, "Failed to write core file for process %s (error %d)\n", curproc->p_comm, error); } done: #ifdef GZIO if (compress) { free(tmpbuf, M_TEMP); gzio_fini(params.gzs); } #endif while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) { TAILQ_REMOVE(¬elst, ninfo, link); free(ninfo, M_TEMP); } if (hdr != NULL) free(hdr, M_TEMP); return (error); } /* * A callback for each_writable_segment() to write out the segment's * program header entry. */ static void cb_put_phdr(entry, closure) vm_map_entry_t entry; void *closure; { struct phdr_closure *phc = (struct phdr_closure *)closure; Elf_Phdr *phdr = phc->phdr; phc->offset = round_page(phc->offset); phdr->p_type = PT_LOAD; phdr->p_offset = phc->offset; phdr->p_vaddr = entry->start; phdr->p_paddr = 0; phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; phdr->p_align = PAGE_SIZE; phdr->p_flags = __elfN(untrans_prot)(entry->protection); phc->offset += phdr->p_filesz; phc->phdr++; } /* * A callback for each_writable_segment() to gather information about * the number of segments and their total size. */ static void cb_size_segment(entry, closure) vm_map_entry_t entry; void *closure; { struct sseg_closure *ssc = (struct sseg_closure *)closure; ssc->count++; ssc->size += entry->end - entry->start; } /* * For each writable segment in the process's memory map, call the given * function with a pointer to the map entry and some arbitrary * caller-supplied data. */ static void each_writable_segment(td, func, closure) struct thread *td; segment_callback func; void *closure; { struct proc *p = td->td_proc; vm_map_t map = &p->p_vmspace->vm_map; vm_map_entry_t entry; vm_object_t backing_object, object; boolean_t ignore_entry; vm_map_lock_read(map); for (entry = map->header.next; entry != &map->header; entry = entry->next) { /* * Don't dump inaccessible mappings, deal with legacy * coredump mode. * * Note that read-only segments related to the elf binary * are marked MAP_ENTRY_NOCOREDUMP now so we no longer * need to arbitrarily ignore such segments. */ if (elf_legacy_coredump) { if ((entry->protection & VM_PROT_RW) != VM_PROT_RW) continue; } else { if ((entry->protection & VM_PROT_ALL) == 0) continue; } /* * Dont include memory segment in the coredump if * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in * madvise(2). Do not dump submaps (i.e. parts of the * kernel map). */ if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP)) continue; if ((object = entry->object.vm_object) == NULL) continue; /* Ignore memory-mapped devices and such things. */ VM_OBJECT_RLOCK(object); while ((backing_object = object->backing_object) != NULL) { VM_OBJECT_RLOCK(backing_object); VM_OBJECT_RUNLOCK(object); object = backing_object; } ignore_entry = object->type != OBJT_DEFAULT && object->type != OBJT_SWAP && object->type != OBJT_VNODE && object->type != OBJT_PHYS; VM_OBJECT_RUNLOCK(object); if (ignore_entry) continue; (*func)(entry, closure); } vm_map_unlock_read(map); } /* * Write the core file header to the file, including padding up to * the page boundary. */ static int __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr, size_t hdrsize, struct note_info_list *notelst, size_t notesz) { struct note_info *ninfo; struct sbuf *sb; int error; /* Fill in the header. */ bzero(hdr, hdrsize); __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz); sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN); sbuf_set_drain(sb, sbuf_drain_core_output, p); sbuf_start_section(sb, NULL); sbuf_bcat(sb, hdr, hdrsize); TAILQ_FOREACH(ninfo, notelst, link) __elfN(putnote)(ninfo, sb); /* Align up to a page boundary for the program segments. */ sbuf_end_section(sb, -1, PAGE_SIZE, 0); error = sbuf_finish(sb); sbuf_delete(sb); return (error); } static void __elfN(prepare_notes)(struct thread *td, struct note_info_list *list, size_t *sizep) { struct proc *p; struct thread *thr; size_t size; p = td->td_proc; size = 0; size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p); /* * To have the debugger select the right thread (LWP) as the initial * thread, we dump the state of the thread passed to us in td first. * This is the thread that causes the core dump and thus likely to * be the right thread one wants to have selected in the debugger. */ thr = td; while (thr != NULL) { size += register_note(list, NT_PRSTATUS, __elfN(note_prstatus), thr); size += register_note(list, NT_FPREGSET, __elfN(note_fpregset), thr); size += register_note(list, NT_THRMISC, __elfN(note_thrmisc), thr); size += register_note(list, -1, __elfN(note_threadmd), thr); thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) : TAILQ_NEXT(thr, td_plist); if (thr == td) thr = TAILQ_NEXT(thr, td_plist); } size += register_note(list, NT_PROCSTAT_PROC, __elfN(note_procstat_proc), p); size += register_note(list, NT_PROCSTAT_FILES, note_procstat_files, p); size += register_note(list, NT_PROCSTAT_VMMAP, note_procstat_vmmap, p); size += register_note(list, NT_PROCSTAT_GROUPS, note_procstat_groups, p); size += register_note(list, NT_PROCSTAT_UMASK, note_procstat_umask, p); size += register_note(list, NT_PROCSTAT_RLIMIT, note_procstat_rlimit, p); size += register_note(list, NT_PROCSTAT_OSREL, note_procstat_osrel, p); size += register_note(list, NT_PROCSTAT_PSSTRINGS, __elfN(note_procstat_psstrings), p); size += register_note(list, NT_PROCSTAT_AUXV, __elfN(note_procstat_auxv), p); *sizep = size; } static void __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs, size_t notesz) { Elf_Ehdr *ehdr; Elf_Phdr *phdr; struct phdr_closure phc; ehdr = (Elf_Ehdr *)hdr; phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)); ehdr->e_ident[EI_MAG0] = ELFMAG0; ehdr->e_ident[EI_MAG1] = ELFMAG1; ehdr->e_ident[EI_MAG2] = ELFMAG2; ehdr->e_ident[EI_MAG3] = ELFMAG3; ehdr->e_ident[EI_CLASS] = ELF_CLASS; ehdr->e_ident[EI_DATA] = ELF_DATA; ehdr->e_ident[EI_VERSION] = EV_CURRENT; ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; ehdr->e_ident[EI_ABIVERSION] = 0; ehdr->e_ident[EI_PAD] = 0; ehdr->e_type = ET_CORE; #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 ehdr->e_machine = ELF_ARCH32; #else ehdr->e_machine = ELF_ARCH; #endif ehdr->e_version = EV_CURRENT; ehdr->e_entry = 0; ehdr->e_phoff = sizeof(Elf_Ehdr); ehdr->e_flags = 0; ehdr->e_ehsize = sizeof(Elf_Ehdr); ehdr->e_phentsize = sizeof(Elf_Phdr); ehdr->e_phnum = numsegs + 1; ehdr->e_shentsize = sizeof(Elf_Shdr); ehdr->e_shnum = 0; ehdr->e_shstrndx = SHN_UNDEF; /* * Fill in the program header entries. */ /* The note segement. */ phdr->p_type = PT_NOTE; phdr->p_offset = hdrsize; phdr->p_vaddr = 0; phdr->p_paddr = 0; phdr->p_filesz = notesz; phdr->p_memsz = 0; phdr->p_flags = PF_R; phdr->p_align = ELF_NOTE_ROUNDSIZE; phdr++; /* All the writable segments from the program. */ phc.phdr = phdr; phc.offset = round_page(hdrsize + notesz); each_writable_segment(td, cb_put_phdr, &phc); } static size_t register_note(struct note_info_list *list, int type, outfunc_t out, void *arg) { struct note_info *ninfo; size_t size, notesize; size = 0; out(arg, NULL, &size); ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK); ninfo->type = type; ninfo->outfunc = out; ninfo->outarg = arg; ninfo->outsize = size; TAILQ_INSERT_TAIL(list, ninfo, link); if (type == -1) return (size); notesize = sizeof(Elf_Note) + /* note header */ roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) + /* note name */ roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ return (notesize); } static size_t append_note_data(const void *src, void *dst, size_t len) { size_t padded_len; padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE); if (dst != NULL) { bcopy(src, dst, len); bzero((char *)dst + len, padded_len - len); } return (padded_len); } size_t __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp) { Elf_Note *note; char *buf; size_t notesize; buf = dst; if (buf != NULL) { note = (Elf_Note *)buf; note->n_namesz = sizeof(FREEBSD_ABI_VENDOR); note->n_descsz = size; note->n_type = type; buf += sizeof(*note); buf += append_note_data(FREEBSD_ABI_VENDOR, buf, sizeof(FREEBSD_ABI_VENDOR)); append_note_data(src, buf, size); if (descp != NULL) *descp = buf; } notesize = sizeof(Elf_Note) + /* note header */ roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) + /* note name */ roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */ return (notesize); } static void __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb) { Elf_Note note; ssize_t old_len; if (ninfo->type == -1) { ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); return; } note.n_namesz = sizeof(FREEBSD_ABI_VENDOR); note.n_descsz = ninfo->outsize; note.n_type = ninfo->type; sbuf_bcat(sb, ¬e, sizeof(note)); sbuf_start_section(sb, &old_len); sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR)); sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); if (note.n_descsz == 0) return; sbuf_start_section(sb, &old_len); ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize); sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0); } /* * Miscellaneous note out functions. */ #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 #include typedef struct prstatus32 elf_prstatus_t; typedef struct prpsinfo32 elf_prpsinfo_t; typedef struct fpreg32 elf_prfpregset_t; typedef struct fpreg32 elf_fpregset_t; typedef struct reg32 elf_gregset_t; typedef struct thrmisc32 elf_thrmisc_t; #define ELF_KERN_PROC_MASK KERN_PROC_MASK32 typedef struct kinfo_proc32 elf_kinfo_proc_t; typedef uint32_t elf_ps_strings_t; #else typedef prstatus_t elf_prstatus_t; typedef prpsinfo_t elf_prpsinfo_t; typedef prfpregset_t elf_prfpregset_t; typedef prfpregset_t elf_fpregset_t; typedef gregset_t elf_gregset_t; typedef thrmisc_t elf_thrmisc_t; #define ELF_KERN_PROC_MASK 0 typedef struct kinfo_proc elf_kinfo_proc_t; typedef vm_offset_t elf_ps_strings_t; #endif static void __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep) { struct proc *p; elf_prpsinfo_t *psinfo; p = (struct proc *)arg; if (sb != NULL) { KASSERT(*sizep == sizeof(*psinfo), ("invalid size")); psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK); psinfo->pr_version = PRPSINFO_VERSION; psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t); strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname)); /* * XXX - We don't fill in the command line arguments properly * yet. */ strlcpy(psinfo->pr_psargs, p->p_comm, sizeof(psinfo->pr_psargs)); sbuf_bcat(sb, psinfo, sizeof(*psinfo)); free(psinfo, M_TEMP); } *sizep = sizeof(*psinfo); } static void __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep) { struct thread *td; elf_prstatus_t *status; td = (struct thread *)arg; if (sb != NULL) { KASSERT(*sizep == sizeof(*status), ("invalid size")); status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK); status->pr_version = PRSTATUS_VERSION; status->pr_statussz = sizeof(elf_prstatus_t); status->pr_gregsetsz = sizeof(elf_gregset_t); status->pr_fpregsetsz = sizeof(elf_fpregset_t); status->pr_osreldate = osreldate; status->pr_cursig = td->td_proc->p_sig; status->pr_pid = td->td_tid; #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 fill_regs32(td, &status->pr_reg); #else fill_regs(td, &status->pr_reg); #endif sbuf_bcat(sb, status, sizeof(*status)); free(status, M_TEMP); } *sizep = sizeof(*status); } static void __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep) { struct thread *td; elf_prfpregset_t *fpregset; td = (struct thread *)arg; if (sb != NULL) { KASSERT(*sizep == sizeof(*fpregset), ("invalid size")); fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK); #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 fill_fpregs32(td, fpregset); #else fill_fpregs(td, fpregset); #endif sbuf_bcat(sb, fpregset, sizeof(*fpregset)); free(fpregset, M_TEMP); } *sizep = sizeof(*fpregset); } static void __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep) { struct thread *td; elf_thrmisc_t thrmisc; td = (struct thread *)arg; if (sb != NULL) { KASSERT(*sizep == sizeof(thrmisc), ("invalid size")); bzero(&thrmisc._pad, sizeof(thrmisc._pad)); strcpy(thrmisc.pr_tname, td->td_name); sbuf_bcat(sb, &thrmisc, sizeof(thrmisc)); } *sizep = sizeof(thrmisc); } /* * Allow for MD specific notes, as well as any MD * specific preparations for writing MI notes. */ static void __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep) { struct thread *td; void *buf; size_t size; td = (struct thread *)arg; size = *sizep; if (size != 0 && sb != NULL) buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK); else buf = NULL; size = 0; __elfN(dump_thread)(td, buf, &size); KASSERT(sb == NULL || *sizep == size, ("invalid size")); if (size != 0 && sb != NULL) sbuf_bcat(sb, buf, size); free(buf, M_TEMP); *sizep = size; } #ifdef KINFO_PROC_SIZE CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE); #endif static void __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep) { struct proc *p; size_t size; int structsize; p = (struct proc *)arg; size = sizeof(structsize) + p->p_numthreads * sizeof(elf_kinfo_proc_t); if (sb != NULL) { KASSERT(*sizep == size, ("invalid size")); structsize = sizeof(elf_kinfo_proc_t); sbuf_bcat(sb, &structsize, sizeof(structsize)); sx_slock(&proctree_lock); PROC_LOCK(p); kern_proc_out(p, sb, ELF_KERN_PROC_MASK); sx_sunlock(&proctree_lock); } *sizep = size; } #ifdef KINFO_FILE_SIZE CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); #endif static void note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep) { struct proc *p; size_t size; int structsize; p = (struct proc *)arg; if (sb == NULL) { size = 0; sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); sbuf_set_drain(sb, sbuf_drain_count, &size); sbuf_bcat(sb, &structsize, sizeof(structsize)); PROC_LOCK(p); kern_proc_filedesc_out(p, sb, -1); sbuf_finish(sb); sbuf_delete(sb); *sizep = size; } else { structsize = sizeof(struct kinfo_file); sbuf_bcat(sb, &structsize, sizeof(structsize)); PROC_LOCK(p); kern_proc_filedesc_out(p, sb, -1); } } #ifdef KINFO_VMENTRY_SIZE CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE); #endif static void note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep) { struct proc *p; size_t size; int structsize; p = (struct proc *)arg; if (sb == NULL) { size = 0; sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); sbuf_set_drain(sb, sbuf_drain_count, &size); sbuf_bcat(sb, &structsize, sizeof(structsize)); PROC_LOCK(p); kern_proc_vmmap_out(p, sb); sbuf_finish(sb); sbuf_delete(sb); *sizep = size; } else { structsize = sizeof(struct kinfo_vmentry); sbuf_bcat(sb, &structsize, sizeof(structsize)); PROC_LOCK(p); kern_proc_vmmap_out(p, sb); } } static void note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep) { struct proc *p; size_t size; int structsize; p = (struct proc *)arg; size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t); if (sb != NULL) { KASSERT(*sizep == size, ("invalid size")); structsize = sizeof(gid_t); sbuf_bcat(sb, &structsize, sizeof(structsize)); sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups * sizeof(gid_t)); } *sizep = size; } static void note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep) { struct proc *p; size_t size; int structsize; p = (struct proc *)arg; size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask); if (sb != NULL) { KASSERT(*sizep == size, ("invalid size")); structsize = sizeof(p->p_fd->fd_cmask); sbuf_bcat(sb, &structsize, sizeof(structsize)); sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask)); } *sizep = size; } static void note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep) { struct proc *p; struct rlimit rlim[RLIM_NLIMITS]; size_t size; int structsize, i; p = (struct proc *)arg; size = sizeof(structsize) + sizeof(rlim); if (sb != NULL) { KASSERT(*sizep == size, ("invalid size")); structsize = sizeof(rlim); sbuf_bcat(sb, &structsize, sizeof(structsize)); PROC_LOCK(p); for (i = 0; i < RLIM_NLIMITS; i++) - lim_rlimit(p, i, &rlim[i]); + lim_rlimit_proc(p, i, &rlim[i]); PROC_UNLOCK(p); sbuf_bcat(sb, rlim, sizeof(rlim)); } *sizep = size; } static void note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep) { struct proc *p; size_t size; int structsize; p = (struct proc *)arg; size = sizeof(structsize) + sizeof(p->p_osrel); if (sb != NULL) { KASSERT(*sizep == size, ("invalid size")); structsize = sizeof(p->p_osrel); sbuf_bcat(sb, &structsize, sizeof(structsize)); sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel)); } *sizep = size; } static void __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep) { struct proc *p; elf_ps_strings_t ps_strings; size_t size; int structsize; p = (struct proc *)arg; size = sizeof(structsize) + sizeof(ps_strings); if (sb != NULL) { KASSERT(*sizep == size, ("invalid size")); structsize = sizeof(ps_strings); #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32 ps_strings = PTROUT(p->p_sysent->sv_psstrings); #else ps_strings = p->p_sysent->sv_psstrings; #endif sbuf_bcat(sb, &structsize, sizeof(structsize)); sbuf_bcat(sb, &ps_strings, sizeof(ps_strings)); } *sizep = size; } static void __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep) { struct proc *p; size_t size; int structsize; p = (struct proc *)arg; if (sb == NULL) { size = 0; sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN); sbuf_set_drain(sb, sbuf_drain_count, &size); sbuf_bcat(sb, &structsize, sizeof(structsize)); PHOLD(p); proc_getauxv(curthread, p, sb); PRELE(p); sbuf_finish(sb); sbuf_delete(sb); *sizep = size; } else { structsize = sizeof(Elf_Auxinfo); sbuf_bcat(sb, &structsize, sizeof(structsize)); PHOLD(p); proc_getauxv(curthread, p, sb); PRELE(p); } } static boolean_t __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote, int32_t *osrel, const Elf_Phdr *pnote) { const Elf_Note *note, *note0, *note_end; const char *note_name; int i; if (pnote == NULL || pnote->p_offset > PAGE_SIZE || pnote->p_filesz > PAGE_SIZE - pnote->p_offset) return (FALSE); note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset); note_end = (const Elf_Note *)(imgp->image_header + pnote->p_offset + pnote->p_filesz); for (i = 0; i < 100 && note >= note0 && note < note_end; i++) { if (!aligned(note, Elf32_Addr) || (const char *)note_end - (const char *)note < sizeof(Elf_Note)) return (FALSE); if (note->n_namesz != checknote->hdr.n_namesz || note->n_descsz != checknote->hdr.n_descsz || note->n_type != checknote->hdr.n_type) goto nextnote; note_name = (const char *)(note + 1); if (note_name + checknote->hdr.n_namesz >= (const char *)note_end || strncmp(checknote->vendor, note_name, checknote->hdr.n_namesz) != 0) goto nextnote; /* * Fetch the osreldate for binary * from the ELF OSABI-note if necessary. */ if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 && checknote->trans_osrel != NULL) return (checknote->trans_osrel(note, osrel)); return (TRUE); nextnote: note = (const Elf_Note *)((const char *)(note + 1) + roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) + roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE)); } return (FALSE); } /* * Try to find the appropriate ABI-note section for checknote, * fetch the osreldate for binary from the ELF OSABI-note. Only the * first page of the image is searched, the same as for headers. */ static boolean_t __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote, int32_t *osrel) { const Elf_Phdr *phdr; const Elf_Ehdr *hdr; int i; hdr = (const Elf_Ehdr *)imgp->image_header; phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff); for (i = 0; i < hdr->e_phnum; i++) { if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i])) return (TRUE); } return (FALSE); } /* * Tell kern_execve.c about it, with a little help from the linker. */ static struct execsw __elfN(execsw) = { __CONCAT(exec_, __elfN(imgact)), __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) }; EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw)); static vm_prot_t __elfN(trans_prot)(Elf_Word flags) { vm_prot_t prot; prot = 0; if (flags & PF_X) prot |= VM_PROT_EXECUTE; if (flags & PF_W) prot |= VM_PROT_WRITE; if (flags & PF_R) prot |= VM_PROT_READ; #if __ELF_WORD_SIZE == 32 #if defined(__amd64__) if (i386_read_exec && (flags & PF_R)) prot |= VM_PROT_EXECUTE; #endif #endif return (prot); } static Elf_Word __elfN(untrans_prot)(vm_prot_t prot) { Elf_Word flags; flags = 0; if (prot & VM_PROT_EXECUTE) flags |= PF_X; if (prot & VM_PROT_READ) flags |= PF_R; if (prot & VM_PROT_WRITE) flags |= PF_W; return (flags); } Index: head/sys/kern/imgact_gzip.c =================================================================== --- head/sys/kern/imgact_gzip.c (revision 284214) +++ head/sys/kern/imgact_gzip.c (revision 284215) @@ -1,390 +1,390 @@ /*- * ---------------------------------------------------------------------------- * "THE BEER-WARE LICENSE" (Revision 42): * wrote this file. As long as you retain this notice you * can do whatever you want with this stuff. If we meet some day, and you think * this stuff is worth it, you can buy me a beer in return. Poul-Henning Kamp * ---------------------------------------------------------------------------- */ /* * This module handles execution of a.out files which have been run through * "gzip". This saves diskspace, but wastes cpu-cycles and VM. * * TODO: * text-segments should be made R/O after being filled * is the vm-stuff safe ? * should handle the entire header of gzip'ed stuff. * inflate isn't quite reentrant yet... * error-handling is a mess... * so is the rest... * tidy up unnecessary includes */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include struct imgact_gzip { struct image_params *ip; struct exec a_out; int error; int gotheader; int where; u_char *inbuf; u_long offset; u_long output; u_long len; int idx; u_long virtual_offset, file_offset, file_end, bss_size; }; static int exec_gzip_imgact(struct image_params *imgp); static int NextByte(void *vp); static int do_aout_hdr(struct imgact_gzip *); static int Flush(void *vp, u_char *, u_long siz); static int exec_gzip_imgact(imgp) struct image_params *imgp; { int error; const u_char *p = (const u_char *) imgp->image_header; struct imgact_gzip igz; struct inflate infl; struct vmspace *vmspace; /* If these four are not OK, it isn't a gzip file */ if (p[0] != 0x1f) return -1; /* 0 Simply magic */ if (p[1] != 0x8b) return -1; /* 1 Simply magic */ if (p[2] != 0x08) return -1; /* 2 Compression method */ if (p[9] != 0x03) return -1; /* 9 OS compressed on */ /* * If this one contains anything but a comment or a filename marker, * we don't want to chew on it */ if (p[3] & ~(0x18)) return ENOEXEC; /* 3 Flags */ /* These are of no use to us */ /* 4-7 Timestamp */ /* 8 Extra flags */ bzero(&igz, sizeof igz); bzero(&infl, sizeof infl); infl.gz_private = (void *) &igz; infl.gz_input = NextByte; infl.gz_output = Flush; igz.ip = imgp; igz.idx = 10; if (p[3] & 0x08) { /* skip a filename */ while (p[igz.idx++]) if (igz.idx >= PAGE_SIZE) return ENOEXEC; } if (p[3] & 0x10) { /* skip a comment */ while (p[igz.idx++]) if (igz.idx >= PAGE_SIZE) return ENOEXEC; } igz.len = imgp->attr->va_size; error = inflate(&infl); /* * The unzipped file may not even have been long enough to contain * a header giving Flush() a chance to return error. Check for this. */ if ( !igz.gotheader ) return ENOEXEC; if ( !error ) { vmspace = imgp->proc->p_vmspace; error = vm_map_protect(&vmspace->vm_map, (vm_offset_t) vmspace->vm_taddr, (vm_offset_t) (vmspace->vm_taddr + (vmspace->vm_tsize << PAGE_SHIFT)) , VM_PROT_READ|VM_PROT_EXECUTE,0); } if (igz.inbuf) kmap_free_wakeup(exec_map, (vm_offset_t)igz.inbuf, PAGE_SIZE); if (igz.error || error) { printf("Output=%lu ", igz.output); printf("Inflate_error=%d igz.error=%d where=%d\n", error, igz.error, igz.where); } if (igz.error) return igz.error; if (error) return ENOEXEC; return 0; } static int do_aout_hdr(struct imgact_gzip * gz) { int error; struct vmspace *vmspace; vm_offset_t vmaddr; /* * Set file/virtual offset based on a.out variant. We do two cases: * host byte order and network byte order (for NetBSD compatibility) */ switch ((int) (gz->a_out.a_midmag & 0xffff)) { case ZMAGIC: gz->virtual_offset = 0; if (gz->a_out.a_text) { gz->file_offset = PAGE_SIZE; } else { /* Bill's "screwball mode" */ gz->file_offset = 0; } break; case QMAGIC: gz->virtual_offset = PAGE_SIZE; gz->file_offset = 0; break; default: /* NetBSD compatibility */ switch ((int) (ntohl(gz->a_out.a_midmag) & 0xffff)) { case ZMAGIC: case QMAGIC: gz->virtual_offset = PAGE_SIZE; gz->file_offset = 0; break; default: gz->where = __LINE__; return (-1); } } gz->bss_size = roundup(gz->a_out.a_bss, PAGE_SIZE); /* * Check various fields in header for validity/bounds. */ if ( /* entry point must lay with text region */ gz->a_out.a_entry < gz->virtual_offset || gz->a_out.a_entry >= gz->virtual_offset + gz->a_out.a_text || /* text and data size must each be page rounded */ gz->a_out.a_text & PAGE_MASK || gz->a_out.a_data & PAGE_MASK) { gz->where = __LINE__; return (-1); } /* * text/data/bss must not exceed limits */ PROC_LOCK(gz->ip->proc); if ( /* text can't exceed maximum text size */ gz->a_out.a_text > maxtsiz || /* data + bss can't exceed rlimit */ gz->a_out.a_data + gz->bss_size > - lim_cur(gz->ip->proc, RLIMIT_DATA) || + lim_cur_proc(gz->ip->proc, RLIMIT_DATA) || racct_set(gz->ip->proc, RACCT_DATA, gz->a_out.a_data + gz->bss_size) != 0) { PROC_UNLOCK(gz->ip->proc); gz->where = __LINE__; return (ENOMEM); } PROC_UNLOCK(gz->ip->proc); /* Find out how far we should go */ gz->file_end = gz->file_offset + gz->a_out.a_text + gz->a_out.a_data; /* * Avoid a possible deadlock if the current address space is destroyed * and that address space maps the locked vnode. In the common case, * the locked vnode's v_usecount is decremented but remains greater * than zero. Consequently, the vnode lock is not needed by vrele(). * However, in cases where the vnode lock is external, such as nullfs, * v_usecount may become zero. */ VOP_UNLOCK(gz->ip->vp, 0); /* * Destroy old process VM and create a new one (with a new stack) */ error = exec_new_vmspace(gz->ip, &aout_sysvec); vn_lock(gz->ip->vp, LK_EXCLUSIVE | LK_RETRY); if (error) { gz->where = __LINE__; return (error); } vmspace = gz->ip->proc->p_vmspace; vmaddr = gz->virtual_offset; error = vm_mmap(&vmspace->vm_map, &vmaddr, gz->a_out.a_text + gz->a_out.a_data, VM_PROT_ALL, VM_PROT_ALL, MAP_ANON | MAP_FIXED, OBJT_DEFAULT, NULL, 0); if (error) { gz->where = __LINE__; return (error); } if (gz->bss_size != 0) { /* * Allocate demand-zeroed area for uninitialized data. * "bss" = 'block started by symbol' - named after the * IBM 7090 instruction of the same name. */ vmaddr = gz->virtual_offset + gz->a_out.a_text + gz->a_out.a_data; error = vm_map_find(&vmspace->vm_map, NULL, 0, &vmaddr, gz->bss_size, 0, VMFS_NO_SPACE, VM_PROT_ALL, VM_PROT_ALL, 0); if (error) { gz->where = __LINE__; return (error); } } /* Fill in process VM information */ vmspace->vm_tsize = gz->a_out.a_text >> PAGE_SHIFT; vmspace->vm_dsize = (gz->a_out.a_data + gz->bss_size) >> PAGE_SHIFT; vmspace->vm_taddr = (caddr_t) (uintptr_t) gz->virtual_offset; vmspace->vm_daddr = (caddr_t) (uintptr_t) (gz->virtual_offset + gz->a_out.a_text); /* Fill in image_params */ gz->ip->interpreted = 0; gz->ip->entry_addr = gz->a_out.a_entry; gz->ip->proc->p_sysent = &aout_sysvec; return 0; } static int NextByte(void *vp) { int error; struct imgact_gzip *igz = (struct imgact_gzip *) vp; if (igz->idx >= igz->len) { igz->where = __LINE__; return GZ_EOF; } if (igz->inbuf && igz->idx < (igz->offset + PAGE_SIZE)) { return igz->inbuf[(igz->idx++) - igz->offset]; } if (igz->inbuf) kmap_free_wakeup(exec_map, (vm_offset_t)igz->inbuf, PAGE_SIZE); igz->offset = igz->idx & ~PAGE_MASK; error = vm_mmap(exec_map, /* map */ (vm_offset_t *) & igz->inbuf, /* address */ PAGE_SIZE, /* size */ VM_PROT_READ, /* protection */ VM_PROT_READ, /* max protection */ 0, /* flags */ OBJT_VNODE, /* handle type */ igz->ip->vp, /* vnode */ igz->offset); /* offset */ if (error) { igz->where = __LINE__; igz->error = error; return GZ_EOF; } return igz->inbuf[(igz->idx++) - igz->offset]; } static int Flush(void *vp, u_char * ptr, u_long siz) { struct imgact_gzip *gz = (struct imgact_gzip *) vp; u_char *p = ptr, *q; int i; /* First, find an a.out-header. */ if (gz->output < sizeof gz->a_out) { q = (u_char *) & gz->a_out; i = min(siz, sizeof gz->a_out - gz->output); bcopy(p, q + gz->output, i); gz->output += i; p += i; siz -= i; if (gz->output == sizeof gz->a_out) { gz->gotheader = 1; i = do_aout_hdr(gz); if (i == -1) { if (!gz->where) gz->where = __LINE__; gz->error = ENOEXEC; return ENOEXEC; } else if (i) { gz->where = __LINE__; gz->error = i; return ENOEXEC; } if (gz->file_offset == 0) { q = (u_char *) (uintptr_t) gz->virtual_offset; copyout(&gz->a_out, q, sizeof gz->a_out); } } } /* Skip over zero-padded first PAGE if needed */ if (gz->output < gz->file_offset && gz->output + siz > gz->file_offset) { i = min(siz, gz->file_offset - gz->output); gz->output += i; p += i; siz -= i; } if (gz->output >= gz->file_offset && gz->output < gz->file_end) { i = min(siz, gz->file_end - gz->output); q = (u_char *) (uintptr_t) (gz->virtual_offset + gz->output - gz->file_offset); copyout(p, q, i); gz->output += i; p += i; siz -= i; } gz->output += siz; return 0; } /* * Tell kern_execve.c about it, with a little help from the linker. */ static struct execsw gzip_execsw = {exec_gzip_imgact, "gzip"}; EXEC_SET(execgzip, gzip_execsw); Index: head/sys/kern/kern_descrip.c =================================================================== --- head/sys/kern/kern_descrip.c (revision 284214) +++ head/sys/kern/kern_descrip.c (revision 284215) @@ -1,3864 +1,3862 @@ /*- * Copyright (c) 1982, 1986, 1989, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_capsicum.h" #include "opt_compat.h" #include "opt_ddb.h" #include "opt_ktrace.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KTRACE #include #endif #include #include #include #include #include static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table"); static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader", "file desc to leader structures"); static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities"); MALLOC_DECLARE(M_FADVISE); static uma_zone_t file_zone; static uma_zone_t filedesc0_zone; static int closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, int holdleaders); static int do_dup(struct thread *td, int flags, int old, int new); static int fd_first_free(struct filedesc *fdp, int low, int size); static int fd_last_used(struct filedesc *fdp, int size); static void fdgrowtable(struct filedesc *fdp, int nfd); static void fdgrowtable_exp(struct filedesc *fdp, int nfd); static void fdunused(struct filedesc *fdp, int fd); static void fdused(struct filedesc *fdp, int fd); -static int getmaxfd(struct proc *p); +static int getmaxfd(struct thread *td); /* Flags for do_dup() */ #define DUP_FIXED 0x1 /* Force fixed allocation. */ #define DUP_FCNTL 0x2 /* fcntl()-style errors. */ #define DUP_CLOEXEC 0x4 /* Atomically set FD_CLOEXEC. */ /* * Each process has: * * - An array of open file descriptors (fd_ofiles) * - An array of file flags (fd_ofileflags) * - A bitmap recording which descriptors are in use (fd_map) * * A process starts out with NDFILE descriptors. The value of NDFILE has * been selected based the historical limit of 20 open files, and an * assumption that the majority of processes, especially short-lived * processes like shells, will never need more. * * If this initial allocation is exhausted, a larger descriptor table and * map are allocated dynamically, and the pointers in the process's struct * filedesc are updated to point to those. This is repeated every time * the process runs out of file descriptors (provided it hasn't hit its * resource limit). * * Since threads may hold references to individual descriptor table * entries, the tables are never freed. Instead, they are placed on a * linked list and freed only when the struct filedesc is released. */ #define NDFILE 20 #define NDSLOTSIZE sizeof(NDSLOTTYPE) #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) #define NDSLOT(x) ((x) / NDENTRIES) #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) /* * SLIST entry used to keep track of ofiles which must be reclaimed when * the process exits. */ struct freetable { struct fdescenttbl *ft_table; SLIST_ENTRY(freetable) ft_next; }; /* * Initial allocation: a filedesc structure + the head of SLIST used to * keep track of old ofiles + enough space for NDFILE descriptors. */ struct fdescenttbl0 { int fdt_nfiles; struct filedescent fdt_ofiles[NDFILE]; }; struct filedesc0 { struct filedesc fd_fd; SLIST_HEAD(, freetable) fd_free; struct fdescenttbl0 fd_dfiles; NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; }; /* * Descriptor management. */ volatile int openfiles; /* actual number of open files */ struct mtx sigio_lock; /* mtx to protect pointers to sigio */ void (*mq_fdclose)(struct thread *td, int fd, struct file *fp); /* * If low >= size, just return low. Otherwise find the first zero bit in the * given bitmap, starting at low and not exceeding size - 1. Return size if * not found. */ static int fd_first_free(struct filedesc *fdp, int low, int size) { NDSLOTTYPE *map = fdp->fd_map; NDSLOTTYPE mask; int off, maxoff; if (low >= size) return (low); off = NDSLOT(low); if (low % NDENTRIES) { mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); if ((mask &= ~map[off]) != 0UL) return (off * NDENTRIES + ffsl(mask) - 1); ++off; } for (maxoff = NDSLOTS(size); off < maxoff; ++off) if (map[off] != ~0UL) return (off * NDENTRIES + ffsl(~map[off]) - 1); return (size); } /* * Find the highest non-zero bit in the given bitmap, starting at 0 and * not exceeding size - 1. Return -1 if not found. */ static int fd_last_used(struct filedesc *fdp, int size) { NDSLOTTYPE *map = fdp->fd_map; NDSLOTTYPE mask; int off, minoff; off = NDSLOT(size); if (size % NDENTRIES) { mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES)); if ((mask &= map[off]) != 0) return (off * NDENTRIES + flsl(mask) - 1); --off; } for (minoff = NDSLOT(0); off >= minoff; --off) if (map[off] != 0) return (off * NDENTRIES + flsl(map[off]) - 1); return (-1); } #ifdef INVARIANTS static int fdisused(struct filedesc *fdp, int fd) { KASSERT(fd >= 0 && fd < fdp->fd_nfiles, ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); } #endif /* * Mark a file descriptor as used. */ static void fdused_init(struct filedesc *fdp, int fd) { KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd)); fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); } static void fdused(struct filedesc *fdp, int fd) { FILEDESC_XLOCK_ASSERT(fdp); fdused_init(fdp, fd); if (fd > fdp->fd_lastfile) fdp->fd_lastfile = fd; if (fd == fdp->fd_freefile) fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles); } /* * Mark a file descriptor as unused. */ static void fdunused(struct filedesc *fdp, int fd) { FILEDESC_XLOCK_ASSERT(fdp); KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd)); KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, ("fd=%d is still in use", fd)); fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); if (fd < fdp->fd_freefile) fdp->fd_freefile = fd; if (fd == fdp->fd_lastfile) fdp->fd_lastfile = fd_last_used(fdp, fd); } /* * Free a file descriptor. * * Avoid some work if fdp is about to be destroyed. */ static inline void fdefree_last(struct filedescent *fde) { filecaps_free(&fde->fde_caps); } static inline void fdfree(struct filedesc *fdp, int fd) { struct filedescent *fde; fde = &fdp->fd_ofiles[fd]; #ifdef CAPABILITIES seq_write_begin(&fde->fde_seq); #endif fdefree_last(fde); bzero(fde, fde_change_size); fdunused(fdp, fd); #ifdef CAPABILITIES seq_write_end(&fde->fde_seq); #endif } /* * System calls on descriptors. */ #ifndef _SYS_SYSPROTO_H_ struct getdtablesize_args { int dummy; }; #endif /* ARGSUSED */ int sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) { - struct proc *p = td->td_proc; +#ifdef RACCT uint64_t lim; +#endif - PROC_LOCK(p); td->td_retval[0] = - min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); + min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc); +#ifdef RACCT + PROC_LOCK(p); lim = racct_get_limit(td->td_proc, RACCT_NOFILE); PROC_UNLOCK(p); if (lim < td->td_retval[0]) td->td_retval[0] = lim; +#endif return (0); } /* * Duplicate a file descriptor to a particular value. * * Note: keep in mind that a potential race condition exists when closing * descriptors from a shared descriptor table (via rfork). */ #ifndef _SYS_SYSPROTO_H_ struct dup2_args { u_int from; u_int to; }; #endif /* ARGSUSED */ int sys_dup2(struct thread *td, struct dup2_args *uap) { return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to)); } /* * Duplicate a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct dup_args { u_int fd; }; #endif /* ARGSUSED */ int sys_dup(struct thread *td, struct dup_args *uap) { return (do_dup(td, 0, (int)uap->fd, 0)); } /* * The file control system call. */ #ifndef _SYS_SYSPROTO_H_ struct fcntl_args { int fd; int cmd; long arg; }; #endif /* ARGSUSED */ int sys_fcntl(struct thread *td, struct fcntl_args *uap) { return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg)); } int kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg) { struct flock fl; struct __oflock ofl; intptr_t arg1; int error; error = 0; switch (cmd) { case F_OGETLK: case F_OSETLK: case F_OSETLKW: /* * Convert old flock structure to new. */ error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl)); fl.l_start = ofl.l_start; fl.l_len = ofl.l_len; fl.l_pid = ofl.l_pid; fl.l_type = ofl.l_type; fl.l_whence = ofl.l_whence; fl.l_sysid = 0; switch (cmd) { case F_OGETLK: cmd = F_GETLK; break; case F_OSETLK: cmd = F_SETLK; break; case F_OSETLKW: cmd = F_SETLKW; break; } arg1 = (intptr_t)&fl; break; case F_GETLK: case F_SETLK: case F_SETLKW: case F_SETLK_REMOTE: error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl)); arg1 = (intptr_t)&fl; break; default: arg1 = arg; break; } if (error) return (error); error = kern_fcntl(td, fd, cmd, arg1); if (error) return (error); if (cmd == F_OGETLK) { ofl.l_start = fl.l_start; ofl.l_len = fl.l_len; ofl.l_pid = fl.l_pid; ofl.l_type = fl.l_type; ofl.l_whence = fl.l_whence; error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl)); } else if (cmd == F_GETLK) { error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl)); } return (error); } int kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) { struct filedesc *fdp; struct flock *flp; struct file *fp, *fp2; struct filedescent *fde; struct proc *p; struct vnode *vp; cap_rights_t rights; int error, flg, tmp; uint64_t bsize; off_t foffset; error = 0; flg = F_POSIX; p = td->td_proc; fdp = p->p_fd; switch (cmd) { case F_DUPFD: tmp = arg; error = do_dup(td, DUP_FCNTL, fd, tmp); break; case F_DUPFD_CLOEXEC: tmp = arg; error = do_dup(td, DUP_FCNTL | DUP_CLOEXEC, fd, tmp); break; case F_DUP2FD: tmp = arg; error = do_dup(td, DUP_FIXED, fd, tmp); break; case F_DUP2FD_CLOEXEC: tmp = arg; error = do_dup(td, DUP_FIXED | DUP_CLOEXEC, fd, tmp); break; case F_GETFD: FILEDESC_SLOCK(fdp); if (fget_locked(fdp, fd) == NULL) { FILEDESC_SUNLOCK(fdp); error = EBADF; break; } fde = &fdp->fd_ofiles[fd]; td->td_retval[0] = (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0; FILEDESC_SUNLOCK(fdp); break; case F_SETFD: FILEDESC_XLOCK(fdp); if (fget_locked(fdp, fd) == NULL) { FILEDESC_XUNLOCK(fdp); error = EBADF; break; } fde = &fdp->fd_ofiles[fd]; fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) | (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); FILEDESC_XUNLOCK(fdp); break; case F_GETFL: error = fget_fcntl(td, fd, cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp); if (error != 0) break; td->td_retval[0] = OFLAGS(fp->f_flag); fdrop(fp, td); break; case F_SETFL: error = fget_fcntl(td, fd, cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp); if (error != 0) break; do { tmp = flg = fp->f_flag; tmp &= ~FCNTLFLAGS; tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); tmp = fp->f_flag & FNONBLOCK; error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); if (error != 0) { fdrop(fp, td); break; } tmp = fp->f_flag & FASYNC; error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); if (error == 0) { fdrop(fp, td); break; } atomic_clear_int(&fp->f_flag, FNONBLOCK); tmp = 0; (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); fdrop(fp, td); break; case F_GETOWN: error = fget_fcntl(td, fd, cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp); if (error != 0) break; error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); if (error == 0) td->td_retval[0] = tmp; fdrop(fp, td); break; case F_SETOWN: error = fget_fcntl(td, fd, cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp); if (error != 0) break; tmp = arg; error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); fdrop(fp, td); break; case F_SETLK_REMOTE: error = priv_check(td, PRIV_NFS_LOCKD); if (error) return (error); flg = F_REMOTE; goto do_setlk; case F_SETLKW: flg |= F_WAIT; /* FALLTHROUGH F_SETLK */ case F_SETLK: do_setlk: cap_rights_init(&rights, CAP_FLOCK); error = fget_unlocked(fdp, fd, &rights, &fp, NULL); if (error != 0) break; if (fp->f_type != DTYPE_VNODE) { error = EBADF; fdrop(fp, td); break; } flp = (struct flock *)arg; if (flp->l_whence == SEEK_CUR) { foffset = foffset_get(fp); if (foffset < 0 || (flp->l_start > 0 && foffset > OFF_MAX - flp->l_start)) { error = EOVERFLOW; fdrop(fp, td); break; } flp->l_start += foffset; } vp = fp->f_vnode; switch (flp->l_type) { case F_RDLCK: if ((fp->f_flag & FREAD) == 0) { error = EBADF; break; } PROC_LOCK(p->p_leader); p->p_leader->p_flag |= P_ADVLOCK; PROC_UNLOCK(p->p_leader); error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, flp, flg); break; case F_WRLCK: if ((fp->f_flag & FWRITE) == 0) { error = EBADF; break; } PROC_LOCK(p->p_leader); p->p_leader->p_flag |= P_ADVLOCK; PROC_UNLOCK(p->p_leader); error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, flp, flg); break; case F_UNLCK: error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, flp, flg); break; case F_UNLCKSYS: /* * Temporary api for testing remote lock * infrastructure. */ if (flg != F_REMOTE) { error = EINVAL; break; } error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCKSYS, flp, flg); break; default: error = EINVAL; break; } if (error != 0 || flp->l_type == F_UNLCK || flp->l_type == F_UNLCKSYS) { fdrop(fp, td); break; } /* * Check for a race with close. * * The vnode is now advisory locked (or unlocked, but this case * is not really important) as the caller requested. * We had to drop the filedesc lock, so we need to recheck if * the descriptor is still valid, because if it was closed * in the meantime we need to remove advisory lock from the * vnode - close on any descriptor leading to an advisory * locked vnode, removes that lock. * We will return 0 on purpose in that case, as the result of * successful advisory lock might have been externally visible * already. This is fine - effectively we pretend to the caller * that the closing thread was a bit slower and that the * advisory lock succeeded before the close. */ error = fget_unlocked(fdp, fd, &rights, &fp2, NULL); if (error != 0) { fdrop(fp, td); break; } if (fp != fp2) { flp->l_whence = SEEK_SET; flp->l_start = 0; flp->l_len = 0; flp->l_type = F_UNLCK; (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, flp, F_POSIX); } fdrop(fp, td); fdrop(fp2, td); break; case F_GETLK: error = fget_unlocked(fdp, fd, cap_rights_init(&rights, CAP_FLOCK), &fp, NULL); if (error != 0) break; if (fp->f_type != DTYPE_VNODE) { error = EBADF; fdrop(fp, td); break; } flp = (struct flock *)arg; if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && flp->l_type != F_UNLCK) { error = EINVAL; fdrop(fp, td); break; } if (flp->l_whence == SEEK_CUR) { foffset = foffset_get(fp); if ((flp->l_start > 0 && foffset > OFF_MAX - flp->l_start) || (flp->l_start < 0 && foffset < OFF_MIN - flp->l_start)) { error = EOVERFLOW; fdrop(fp, td); break; } flp->l_start += foffset; } vp = fp->f_vnode; error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, F_POSIX); fdrop(fp, td); break; case F_RDAHEAD: arg = arg ? 128 * 1024: 0; /* FALLTHROUGH */ case F_READAHEAD: error = fget_unlocked(fdp, fd, NULL, &fp, NULL); if (error != 0) break; if (fp->f_type != DTYPE_VNODE) { fdrop(fp, td); error = EBADF; break; } vp = fp->f_vnode; /* * Exclusive lock synchronizes against f_seqcount reads and * writes in sequential_heuristic(). */ error = vn_lock(vp, LK_EXCLUSIVE); if (error != 0) { fdrop(fp, td); break; } if (arg >= 0) { bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; fp->f_seqcount = (arg + bsize - 1) / bsize; atomic_set_int(&fp->f_flag, FRDAHEAD); } else { atomic_clear_int(&fp->f_flag, FRDAHEAD); } VOP_UNLOCK(vp, 0); fdrop(fp, td); break; default: error = EINVAL; break; } return (error); } static int -getmaxfd(struct proc *p) +getmaxfd(struct thread *td) { - int maxfd; - PROC_LOCK(p); - maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); - PROC_UNLOCK(p); - - return (maxfd); + return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc)); } /* * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). */ static int do_dup(struct thread *td, int flags, int old, int new) { struct filedesc *fdp; struct filedescent *oldfde, *newfde; struct proc *p; struct file *fp; struct file *delfp; int error, maxfd; p = td->td_proc; fdp = p->p_fd; /* * Verify we have a valid descriptor to dup from and possibly to * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should * return EINVAL when the new descriptor is out of bounds. */ if (old < 0) return (EBADF); if (new < 0) return (flags & DUP_FCNTL ? EINVAL : EBADF); - maxfd = getmaxfd(p); + maxfd = getmaxfd(td); if (new >= maxfd) return (flags & DUP_FCNTL ? EINVAL : EBADF); FILEDESC_XLOCK(fdp); if (fget_locked(fdp, old) == NULL) { FILEDESC_XUNLOCK(fdp); return (EBADF); } oldfde = &fdp->fd_ofiles[old]; if (flags & DUP_FIXED && old == new) { td->td_retval[0] = new; if (flags & DUP_CLOEXEC) fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE; FILEDESC_XUNLOCK(fdp); return (0); } fp = oldfde->fde_file; fhold(fp); /* * If the caller specified a file descriptor, make sure the file * table is large enough to hold it, and grab it. Otherwise, just * allocate a new descriptor the usual way. */ if (flags & DUP_FIXED) { if (new >= fdp->fd_nfiles) { /* * The resource limits are here instead of e.g. * fdalloc(), because the file descriptor table may be * shared between processes, so we can't really use * racct_add()/racct_sub(). Instead of counting the * number of actually allocated descriptors, just put * the limit on the size of the file descriptor table. */ #ifdef RACCT if (racct_enable) { PROC_LOCK(p); error = racct_set(p, RACCT_NOFILE, new + 1); PROC_UNLOCK(p); if (error != 0) { FILEDESC_XUNLOCK(fdp); fdrop(fp, td); return (EMFILE); } } #endif fdgrowtable_exp(fdp, new + 1); oldfde = &fdp->fd_ofiles[old]; } newfde = &fdp->fd_ofiles[new]; if (newfde->fde_file == NULL) fdused(fdp, new); } else { if ((error = fdalloc(td, new, &new)) != 0) { FILEDESC_XUNLOCK(fdp); fdrop(fp, td); return (error); } newfde = &fdp->fd_ofiles[new]; } KASSERT(fp == oldfde->fde_file, ("old fd has been modified")); KASSERT(old != new, ("new fd is same as old")); delfp = newfde->fde_file; /* * Duplicate the source descriptor. */ #ifdef CAPABILITIES seq_write_begin(&newfde->fde_seq); #endif filecaps_free(&newfde->fde_caps); memcpy(newfde, oldfde, fde_change_size); filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps); if ((flags & DUP_CLOEXEC) != 0) newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE; else newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE; #ifdef CAPABILITIES seq_write_end(&newfde->fde_seq); #endif td->td_retval[0] = new; if (delfp != NULL) { (void) closefp(fdp, new, delfp, td, 1); /* closefp() drops the FILEDESC lock for us. */ } else { FILEDESC_XUNLOCK(fdp); } return (0); } /* * If sigio is on the list associated with a process or process group, * disable signalling from the device, remove sigio from the list and * free sigio. */ void funsetown(struct sigio **sigiop) { struct sigio *sigio; SIGIO_LOCK(); sigio = *sigiop; if (sigio == NULL) { SIGIO_UNLOCK(); return; } *(sigio->sio_myref) = NULL; if ((sigio)->sio_pgid < 0) { struct pgrp *pg = (sigio)->sio_pgrp; PGRP_LOCK(pg); SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, sigio, sio_pgsigio); PGRP_UNLOCK(pg); } else { struct proc *p = (sigio)->sio_proc; PROC_LOCK(p); SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, sigio, sio_pgsigio); PROC_UNLOCK(p); } SIGIO_UNLOCK(); crfree(sigio->sio_ucred); free(sigio, M_SIGIO); } /* * Free a list of sigio structures. * We only need to lock the SIGIO_LOCK because we have made ourselves * inaccessible to callers of fsetown and therefore do not need to lock * the proc or pgrp struct for the list manipulation. */ void funsetownlst(struct sigiolst *sigiolst) { struct proc *p; struct pgrp *pg; struct sigio *sigio; sigio = SLIST_FIRST(sigiolst); if (sigio == NULL) return; p = NULL; pg = NULL; /* * Every entry of the list should belong * to a single proc or pgrp. */ if (sigio->sio_pgid < 0) { pg = sigio->sio_pgrp; PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); } else /* if (sigio->sio_pgid > 0) */ { p = sigio->sio_proc; PROC_LOCK_ASSERT(p, MA_NOTOWNED); } SIGIO_LOCK(); while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { *(sigio->sio_myref) = NULL; if (pg != NULL) { KASSERT(sigio->sio_pgid < 0, ("Proc sigio in pgrp sigio list")); KASSERT(sigio->sio_pgrp == pg, ("Bogus pgrp in sigio list")); PGRP_LOCK(pg); SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio); PGRP_UNLOCK(pg); } else /* if (p != NULL) */ { KASSERT(sigio->sio_pgid > 0, ("Pgrp sigio in proc sigio list")); KASSERT(sigio->sio_proc == p, ("Bogus proc in sigio list")); PROC_LOCK(p); SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio); PROC_UNLOCK(p); } SIGIO_UNLOCK(); crfree(sigio->sio_ucred); free(sigio, M_SIGIO); SIGIO_LOCK(); } SIGIO_UNLOCK(); } /* * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). * * After permission checking, add a sigio structure to the sigio list for * the process or process group. */ int fsetown(pid_t pgid, struct sigio **sigiop) { struct proc *proc; struct pgrp *pgrp; struct sigio *sigio; int ret; if (pgid == 0) { funsetown(sigiop); return (0); } ret = 0; /* Allocate and fill in the new sigio out of locks. */ sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); sigio->sio_pgid = pgid; sigio->sio_ucred = crhold(curthread->td_ucred); sigio->sio_myref = sigiop; sx_slock(&proctree_lock); if (pgid > 0) { proc = pfind(pgid); if (proc == NULL) { ret = ESRCH; goto fail; } /* * Policy - Don't allow a process to FSETOWN a process * in another session. * * Remove this test to allow maximum flexibility or * restrict FSETOWN to the current process or process * group for maximum safety. */ PROC_UNLOCK(proc); if (proc->p_session != curthread->td_proc->p_session) { ret = EPERM; goto fail; } pgrp = NULL; } else /* if (pgid < 0) */ { pgrp = pgfind(-pgid); if (pgrp == NULL) { ret = ESRCH; goto fail; } PGRP_UNLOCK(pgrp); /* * Policy - Don't allow a process to FSETOWN a process * in another session. * * Remove this test to allow maximum flexibility or * restrict FSETOWN to the current process or process * group for maximum safety. */ if (pgrp->pg_session != curthread->td_proc->p_session) { ret = EPERM; goto fail; } proc = NULL; } funsetown(sigiop); if (pgid > 0) { PROC_LOCK(proc); /* * Since funsetownlst() is called without the proctree * locked, we need to check for P_WEXIT. * XXX: is ESRCH correct? */ if ((proc->p_flag & P_WEXIT) != 0) { PROC_UNLOCK(proc); ret = ESRCH; goto fail; } SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); sigio->sio_proc = proc; PROC_UNLOCK(proc); } else { PGRP_LOCK(pgrp); SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); sigio->sio_pgrp = pgrp; PGRP_UNLOCK(pgrp); } sx_sunlock(&proctree_lock); SIGIO_LOCK(); *sigiop = sigio; SIGIO_UNLOCK(); return (0); fail: sx_sunlock(&proctree_lock); crfree(sigio->sio_ucred); free(sigio, M_SIGIO); return (ret); } /* * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). */ pid_t fgetown(sigiop) struct sigio **sigiop; { pid_t pgid; SIGIO_LOCK(); pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; SIGIO_UNLOCK(); return (pgid); } /* * Function drops the filedesc lock on return. */ static int closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, int holdleaders) { int error; FILEDESC_XLOCK_ASSERT(fdp); if (holdleaders) { if (td->td_proc->p_fdtol != NULL) { /* * Ask fdfree() to sleep to ensure that all relevant * process leaders can be traversed in closef(). */ fdp->fd_holdleaderscount++; } else { holdleaders = 0; } } /* * We now hold the fp reference that used to be owned by the * descriptor array. We have to unlock the FILEDESC *AFTER* * knote_fdclose to prevent a race of the fd getting opened, a knote * added, and deleteing a knote for the new fd. */ knote_fdclose(td, fd); /* * We need to notify mqueue if the object is of type mqueue. */ if (fp->f_type == DTYPE_MQUEUE) mq_fdclose(td, fd, fp); FILEDESC_XUNLOCK(fdp); error = closef(fp, td); if (holdleaders) { FILEDESC_XLOCK(fdp); fdp->fd_holdleaderscount--; if (fdp->fd_holdleaderscount == 0 && fdp->fd_holdleaderswakeup != 0) { fdp->fd_holdleaderswakeup = 0; wakeup(&fdp->fd_holdleaderscount); } FILEDESC_XUNLOCK(fdp); } return (error); } /* * Close a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct close_args { int fd; }; #endif /* ARGSUSED */ int sys_close(td, uap) struct thread *td; struct close_args *uap; { return (kern_close(td, uap->fd)); } int kern_close(td, fd) struct thread *td; int fd; { struct filedesc *fdp; struct file *fp; fdp = td->td_proc->p_fd; AUDIT_SYSCLOSE(td, fd); FILEDESC_XLOCK(fdp); if ((fp = fget_locked(fdp, fd)) == NULL) { FILEDESC_XUNLOCK(fdp); return (EBADF); } fdfree(fdp, fd); /* closefp() drops the FILEDESC lock for us. */ return (closefp(fdp, fd, fp, td, 1)); } /* * Close open file descriptors. */ #ifndef _SYS_SYSPROTO_H_ struct closefrom_args { int lowfd; }; #endif /* ARGSUSED */ int sys_closefrom(struct thread *td, struct closefrom_args *uap) { struct filedesc *fdp; int fd; fdp = td->td_proc->p_fd; AUDIT_ARG_FD(uap->lowfd); /* * Treat negative starting file descriptor values identical to * closefrom(0) which closes all files. */ if (uap->lowfd < 0) uap->lowfd = 0; FILEDESC_SLOCK(fdp); for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) { if (fdp->fd_ofiles[fd].fde_file != NULL) { FILEDESC_SUNLOCK(fdp); (void)kern_close(td, fd); FILEDESC_SLOCK(fdp); } } FILEDESC_SUNLOCK(fdp); return (0); } #if defined(COMPAT_43) /* * Return status information about a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct ofstat_args { int fd; struct ostat *sb; }; #endif /* ARGSUSED */ int ofstat(struct thread *td, struct ofstat_args *uap) { struct ostat oub; struct stat ub; int error; error = kern_fstat(td, uap->fd, &ub); if (error == 0) { cvtstat(&ub, &oub); error = copyout(&oub, uap->sb, sizeof(oub)); } return (error); } #endif /* COMPAT_43 */ /* * Return status information about a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct fstat_args { int fd; struct stat *sb; }; #endif /* ARGSUSED */ int sys_fstat(struct thread *td, struct fstat_args *uap) { struct stat ub; int error; error = kern_fstat(td, uap->fd, &ub); if (error == 0) error = copyout(&ub, uap->sb, sizeof(ub)); return (error); } int kern_fstat(struct thread *td, int fd, struct stat *sbp) { struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(fd); error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp); if (error != 0) return (error); AUDIT_ARG_FILE(td->td_proc, fp); error = fo_stat(fp, sbp, td->td_ucred, td); fdrop(fp, td); #ifdef KTRACE if (error == 0 && KTRPOINT(td, KTR_STRUCT)) ktrstat(sbp); #endif return (error); } /* * Return status information about a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct nfstat_args { int fd; struct nstat *sb; }; #endif /* ARGSUSED */ int sys_nfstat(struct thread *td, struct nfstat_args *uap) { struct nstat nub; struct stat ub; int error; error = kern_fstat(td, uap->fd, &ub); if (error == 0) { cvtnstat(&ub, &nub); error = copyout(&nub, uap->sb, sizeof(nub)); } return (error); } /* * Return pathconf information about a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct fpathconf_args { int fd; int name; }; #endif /* ARGSUSED */ int sys_fpathconf(struct thread *td, struct fpathconf_args *uap) { struct file *fp; struct vnode *vp; cap_rights_t rights; int error; error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp); if (error != 0) return (error); /* If asynchronous I/O is available, it works for all descriptors. */ if (uap->name == _PC_ASYNC_IO) { td->td_retval[0] = async_io_version; goto out; } vp = fp->f_vnode; if (vp != NULL) { vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_PATHCONF(vp, uap->name, td->td_retval); VOP_UNLOCK(vp, 0); } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { if (uap->name != _PC_PIPE_BUF) { error = EINVAL; } else { td->td_retval[0] = PIPE_BUF; error = 0; } } else { error = EOPNOTSUPP; } out: fdrop(fp, td); return (error); } /* * Initialize filecaps structure. */ void filecaps_init(struct filecaps *fcaps) { bzero(fcaps, sizeof(*fcaps)); fcaps->fc_nioctls = -1; } /* * Copy filecaps structure allocating memory for ioctls array if needed. */ void filecaps_copy(const struct filecaps *src, struct filecaps *dst) { size_t size; *dst = *src; if (src->fc_ioctls != NULL) { KASSERT(src->fc_nioctls > 0, ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); bcopy(src->fc_ioctls, dst->fc_ioctls, size); } } /* * Move filecaps structure to the new place and clear the old place. */ void filecaps_move(struct filecaps *src, struct filecaps *dst) { *dst = *src; bzero(src, sizeof(*src)); } /* * Fill the given filecaps structure with full rights. */ static void filecaps_fill(struct filecaps *fcaps) { CAP_ALL(&fcaps->fc_rights); fcaps->fc_ioctls = NULL; fcaps->fc_nioctls = -1; fcaps->fc_fcntls = CAP_FCNTL_ALL; } /* * Free memory allocated within filecaps structure. */ void filecaps_free(struct filecaps *fcaps) { free(fcaps->fc_ioctls, M_FILECAPS); bzero(fcaps, sizeof(*fcaps)); } /* * Validate the given filecaps structure. */ static void filecaps_validate(const struct filecaps *fcaps, const char *func) { KASSERT(cap_rights_is_valid(&fcaps->fc_rights), ("%s: invalid rights", func)); KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, ("%s: invalid fcntls", func)); KASSERT(fcaps->fc_fcntls == 0 || cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), ("%s: fcntls without CAP_FCNTL", func)); KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), ("%s: invalid ioctls", func)); KASSERT(fcaps->fc_nioctls == 0 || cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), ("%s: ioctls without CAP_IOCTL", func)); } static void fdgrowtable_exp(struct filedesc *fdp, int nfd) { int nfd1; FILEDESC_XLOCK_ASSERT(fdp); nfd1 = fdp->fd_nfiles * 2; if (nfd1 < nfd) nfd1 = nfd; fdgrowtable(fdp, nfd1); } /* * Grow the file table to accomodate (at least) nfd descriptors. */ static void fdgrowtable(struct filedesc *fdp, int nfd) { struct filedesc0 *fdp0; struct freetable *ft; struct fdescenttbl *ntable; struct fdescenttbl *otable; int nnfiles, onfiles; NDSLOTTYPE *nmap, *omap; /* * If lastfile is -1 this struct filedesc was just allocated and we are * growing it to accomodate for the one we are going to copy from. There * is no need to have a lock on this one as it's not visible to anyone. */ if (fdp->fd_lastfile != -1) FILEDESC_XLOCK_ASSERT(fdp); KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); /* save old values */ onfiles = fdp->fd_nfiles; otable = fdp->fd_files; omap = fdp->fd_map; /* compute the size of the new table */ nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ if (nnfiles <= onfiles) /* the table is already large enough */ return; /* * Allocate a new table. We need enough space for the number of * entries, file entries themselves and the struct freetable we will use * when we decommission the table and place it on the freelist. * We place the struct freetable in the middle so we don't have * to worry about padding. */ ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) + nnfiles * sizeof(ntable->fdt_ofiles[0]) + sizeof(struct freetable), M_FILEDESC, M_ZERO | M_WAITOK); /* copy the old data */ ntable->fdt_nfiles = nnfiles; memcpy(ntable->fdt_ofiles, otable->fdt_ofiles, onfiles * sizeof(ntable->fdt_ofiles[0])); /* * Allocate a new map only if the old is not large enough. It will * grow at a slower rate than the table as it can map more * entries than the table can hold. */ if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, M_ZERO | M_WAITOK); /* copy over the old data and update the pointer */ memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); fdp->fd_map = nmap; } /* * Make sure that ntable is correctly initialized before we replace * fd_files poiner. Otherwise fget_unlocked() may see inconsistent * data. */ atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable); /* * Do not free the old file table, as some threads may still * reference entries within it. Instead, place it on a freelist * which will be processed when the struct filedesc is released. * * Note that if onfiles == NDFILE, we're dealing with the original * static allocation contained within (struct filedesc0 *)fdp, * which must not be freed. */ if (onfiles > NDFILE) { ft = (struct freetable *)&otable->fdt_ofiles[onfiles]; fdp0 = (struct filedesc0 *)fdp; ft->ft_table = otable; SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); } /* * The map does not have the same possibility of threads still * holding references to it. So always free it as long as it * does not reference the original static allocation. */ if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) free(omap, M_FILEDESC); } /* * Allocate a file descriptor for the process. */ int fdalloc(struct thread *td, int minfd, int *result) { struct proc *p = td->td_proc; struct filedesc *fdp = p->p_fd; int fd, maxfd, allocfd; #ifdef RACCT int error; #endif FILEDESC_XLOCK_ASSERT(fdp); if (fdp->fd_freefile > minfd) minfd = fdp->fd_freefile; - maxfd = getmaxfd(p); + maxfd = getmaxfd(td); /* * Search the bitmap for a free descriptor starting at minfd. * If none is found, grow the file table. */ fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); if (fd >= maxfd) return (EMFILE); if (fd >= fdp->fd_nfiles) { allocfd = min(fd * 2, maxfd); #ifdef RACCT if (racct_enable) { PROC_LOCK(p); error = racct_set(p, RACCT_NOFILE, allocfd); PROC_UNLOCK(p); if (error != 0) return (EMFILE); } #endif /* * fd is already equal to first free descriptor >= minfd, so * we only need to grow the table and we are done. */ fdgrowtable_exp(fdp, allocfd); } /* * Perform some sanity checks, then mark the file descriptor as * used and return it to the caller. */ KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), ("invalid descriptor %d", fd)); KASSERT(!fdisused(fdp, fd), ("fd_first_free() returned non-free descriptor")); KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, ("file descriptor isn't free")); KASSERT(fdp->fd_ofiles[fd].fde_flags == 0, ("file flags are set")); fdused(fdp, fd); *result = fd; return (0); } /* * Allocate n file descriptors for the process. */ int fdallocn(struct thread *td, int minfd, int *fds, int n) { struct proc *p = td->td_proc; struct filedesc *fdp = p->p_fd; int i; FILEDESC_XLOCK_ASSERT(fdp); for (i = 0; i < n; i++) if (fdalloc(td, 0, &fds[i]) != 0) break; if (i < n) { for (i--; i >= 0; i--) fdunused(fdp, fds[i]); return (EMFILE); } return (0); } /* * Create a new open file structure and allocate a file decriptor for the * process that refers to it. We add one reference to the file for the * descriptor table and one reference for resultfp. This is to prevent us * being preempted and the entry in the descriptor table closed after we * release the FILEDESC lock. */ int falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags) { struct file *fp; int error, fd; error = falloc_noinstall(td, &fp); if (error) return (error); /* no reference held on error */ error = finstall(td, fp, &fd, flags, NULL); if (error) { fdrop(fp, td); /* one reference (fp only) */ return (error); } if (resultfp != NULL) *resultfp = fp; /* copy out result */ else fdrop(fp, td); /* release local reference */ if (resultfd != NULL) *resultfd = fd; return (0); } /* * Create a new open file structure without allocating a file descriptor. */ int falloc_noinstall(struct thread *td, struct file **resultfp) { struct file *fp; int maxuserfiles = maxfiles - (maxfiles / 20); static struct timeval lastfail; static int curfail; KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); if ((openfiles >= maxuserfiles && priv_check(td, PRIV_MAXFILES) != 0) || openfiles >= maxfiles) { if (ppsratecheck(&lastfail, &curfail, 1)) { printf("kern.maxfiles limit exceeded by uid %i, " "please see tuning(7).\n", td->td_ucred->cr_ruid); } return (ENFILE); } atomic_add_int(&openfiles, 1); fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); refcount_init(&fp->f_count, 1); fp->f_cred = crhold(td->td_ucred); fp->f_ops = &badfileops; *resultfp = fp; return (0); } /* * Install a file in a file descriptor table. */ int finstall(struct thread *td, struct file *fp, int *fd, int flags, struct filecaps *fcaps) { struct filedesc *fdp = td->td_proc->p_fd; struct filedescent *fde; int error; KASSERT(fd != NULL, ("%s: fd == NULL", __func__)); KASSERT(fp != NULL, ("%s: fp == NULL", __func__)); if (fcaps != NULL) filecaps_validate(fcaps, __func__); FILEDESC_XLOCK(fdp); if ((error = fdalloc(td, 0, fd))) { FILEDESC_XUNLOCK(fdp); return (error); } fhold(fp); fde = &fdp->fd_ofiles[*fd]; #ifdef CAPABILITIES seq_write_begin(&fde->fde_seq); #endif fde->fde_file = fp; if ((flags & O_CLOEXEC) != 0) fde->fde_flags |= UF_EXCLOSE; if (fcaps != NULL) filecaps_move(fcaps, &fde->fde_caps); else filecaps_fill(&fde->fde_caps); #ifdef CAPABILITIES seq_write_end(&fde->fde_seq); #endif FILEDESC_XUNLOCK(fdp); return (0); } /* * Build a new filedesc structure from another. * Copy the current, root, and jail root vnode references. * * If fdp is not NULL, return with it shared locked. */ struct filedesc * fdinit(struct filedesc *fdp, bool prepfiles) { struct filedesc0 *newfdp0; struct filedesc *newfdp; newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO); newfdp = &newfdp0->fd_fd; /* Create the file descriptor table. */ FILEDESC_LOCK_INIT(newfdp); refcount_init(&newfdp->fd_refcnt, 1); refcount_init(&newfdp->fd_holdcnt, 1); newfdp->fd_cmask = CMASK; newfdp->fd_map = newfdp0->fd_dmap; newfdp->fd_lastfile = -1; newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles; newfdp->fd_files->fdt_nfiles = NDFILE; if (fdp == NULL) return (newfdp); if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles) fdgrowtable(newfdp, fdp->fd_lastfile + 1); FILEDESC_SLOCK(fdp); newfdp->fd_cdir = fdp->fd_cdir; if (newfdp->fd_cdir) VREF(newfdp->fd_cdir); newfdp->fd_rdir = fdp->fd_rdir; if (newfdp->fd_rdir) VREF(newfdp->fd_rdir); newfdp->fd_jdir = fdp->fd_jdir; if (newfdp->fd_jdir) VREF(newfdp->fd_jdir); if (!prepfiles) { FILEDESC_SUNLOCK(fdp); } else { while (fdp->fd_lastfile >= newfdp->fd_nfiles) { FILEDESC_SUNLOCK(fdp); fdgrowtable(newfdp, fdp->fd_lastfile + 1); FILEDESC_SLOCK(fdp); } } return (newfdp); } static struct filedesc * fdhold(struct proc *p) { struct filedesc *fdp; PROC_LOCK_ASSERT(p, MA_OWNED); fdp = p->p_fd; if (fdp != NULL) refcount_acquire(&fdp->fd_holdcnt); return (fdp); } static void fddrop(struct filedesc *fdp) { if (fdp->fd_holdcnt > 1) { if (refcount_release(&fdp->fd_holdcnt) == 0) return; } FILEDESC_LOCK_DESTROY(fdp); uma_zfree(filedesc0_zone, fdp); } /* * Share a filedesc structure. */ struct filedesc * fdshare(struct filedesc *fdp) { refcount_acquire(&fdp->fd_refcnt); return (fdp); } /* * Unshare a filedesc structure, if necessary by making a copy */ void fdunshare(struct thread *td) { struct filedesc *tmp; struct proc *p = td->td_proc; if (p->p_fd->fd_refcnt == 1) return; tmp = fdcopy(p->p_fd); fdescfree(td); p->p_fd = tmp; } /* * Copy a filedesc structure. A NULL pointer in returns a NULL reference, * this is to ease callers, not catch errors. */ struct filedesc * fdcopy(struct filedesc *fdp) { struct filedesc *newfdp; struct filedescent *nfde, *ofde; int i; MPASS(fdp != NULL); newfdp = fdinit(fdp, true); /* copy all passable descriptors (i.e. not kqueue) */ newfdp->fd_freefile = -1; for (i = 0; i <= fdp->fd_lastfile; ++i) { ofde = &fdp->fd_ofiles[i]; if (ofde->fde_file == NULL || (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) { if (newfdp->fd_freefile == -1) newfdp->fd_freefile = i; continue; } nfde = &newfdp->fd_ofiles[i]; *nfde = *ofde; filecaps_copy(&ofde->fde_caps, &nfde->fde_caps); fhold(nfde->fde_file); fdused_init(newfdp, i); newfdp->fd_lastfile = i; } if (newfdp->fd_freefile == -1) newfdp->fd_freefile = i; newfdp->fd_cmask = fdp->fd_cmask; FILEDESC_SUNLOCK(fdp); return (newfdp); } /* * Clear POSIX style locks. This is only used when fdp looses a reference (i.e. * one of processes using it exits) and the table used to be shared. */ static void fdclearlocks(struct thread *td) { struct filedesc *fdp; struct filedesc_to_leader *fdtol; struct flock lf; struct file *fp; struct proc *p; struct vnode *vp; int i; p = td->td_proc; fdp = p->p_fd; fdtol = p->p_fdtol; MPASS(fdtol != NULL); FILEDESC_XLOCK(fdp); KASSERT(fdtol->fdl_refcount > 0, ("filedesc_to_refcount botch: fdl_refcount=%d", fdtol->fdl_refcount)); if (fdtol->fdl_refcount == 1 && (p->p_leader->p_flag & P_ADVLOCK) != 0) { for (i = 0; i <= fdp->fd_lastfile; i++) { fp = fdp->fd_ofiles[i].fde_file; if (fp == NULL || fp->f_type != DTYPE_VNODE) continue; fhold(fp); FILEDESC_XUNLOCK(fdp); lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; lf.l_type = F_UNLCK; vp = fp->f_vnode; (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, &lf, F_POSIX); FILEDESC_XLOCK(fdp); fdrop(fp, td); } } retry: if (fdtol->fdl_refcount == 1) { if (fdp->fd_holdleaderscount > 0 && (p->p_leader->p_flag & P_ADVLOCK) != 0) { /* * close() or do_dup() has cleared a reference * in a shared file descriptor table. */ fdp->fd_holdleaderswakeup = 1; sx_sleep(&fdp->fd_holdleaderscount, FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); goto retry; } if (fdtol->fdl_holdcount > 0) { /* * Ensure that fdtol->fdl_leader remains * valid in closef(). */ fdtol->fdl_wakeup = 1; sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); goto retry; } } fdtol->fdl_refcount--; if (fdtol->fdl_refcount == 0 && fdtol->fdl_holdcount == 0) { fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; fdtol->fdl_prev->fdl_next = fdtol->fdl_next; } else fdtol = NULL; p->p_fdtol = NULL; FILEDESC_XUNLOCK(fdp); if (fdtol != NULL) free(fdtol, M_FILEDESC_TO_LEADER); } /* * Release a filedesc structure. */ void fdescfree(struct thread *td) { struct proc *p; struct filedesc0 *fdp0; struct filedesc *fdp; struct freetable *ft, *tft; struct filedescent *fde; struct file *fp; struct vnode *cdir, *jdir, *rdir; int i; p = td->td_proc; fdp = p->p_fd; MPASS(fdp != NULL); #ifdef RACCT if (racct_enable) { PROC_LOCK(p); racct_set(p, RACCT_NOFILE, 0); PROC_UNLOCK(p); } #endif if (td->td_proc->p_fdtol != NULL) fdclearlocks(td); PROC_LOCK(p); p->p_fd = NULL; PROC_UNLOCK(p); if (refcount_release(&fdp->fd_refcnt) == 0) return; FILEDESC_XLOCK(fdp); cdir = fdp->fd_cdir; fdp->fd_cdir = NULL; rdir = fdp->fd_rdir; fdp->fd_rdir = NULL; jdir = fdp->fd_jdir; fdp->fd_jdir = NULL; FILEDESC_XUNLOCK(fdp); for (i = 0; i <= fdp->fd_lastfile; i++) { fde = &fdp->fd_ofiles[i]; fp = fde->fde_file; if (fp != NULL) { fdefree_last(fde); (void) closef(fp, td); } } if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) free(fdp->fd_map, M_FILEDESC); if (fdp->fd_nfiles > NDFILE) free(fdp->fd_files, M_FILEDESC); fdp0 = (struct filedesc0 *)fdp; SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft) free(ft->ft_table, M_FILEDESC); if (cdir != NULL) vrele(cdir); if (rdir != NULL) vrele(rdir); if (jdir != NULL) vrele(jdir); fddrop(fdp); } /* * For setugid programs, we don't want to people to use that setugidness * to generate error messages which write to a file which otherwise would * otherwise be off-limits to the process. We check for filesystems where * the vnode can change out from under us after execve (like [lin]procfs). * * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is * sufficient. We also don't check for setugidness since we know we are. */ static bool is_unsafe(struct file *fp) { struct vnode *vp; if (fp->f_type != DTYPE_VNODE) return (false); vp = fp->f_vnode; return ((vp->v_vflag & VV_PROCDEP) != 0); } /* * Make this setguid thing safe, if at all possible. */ void fdsetugidsafety(struct thread *td) { struct filedesc *fdp; struct file *fp; int i; fdp = td->td_proc->p_fd; KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); MPASS(fdp->fd_nfiles >= 3); for (i = 0; i <= 2; i++) { fp = fdp->fd_ofiles[i].fde_file; if (fp != NULL && is_unsafe(fp)) { FILEDESC_XLOCK(fdp); knote_fdclose(td, i); /* * NULL-out descriptor prior to close to avoid * a race while close blocks. */ fdfree(fdp, i); FILEDESC_XUNLOCK(fdp); (void) closef(fp, td); } } } /* * If a specific file object occupies a specific file descriptor, close the * file descriptor entry and drop a reference on the file object. This is a * convenience function to handle a subsequent error in a function that calls * falloc() that handles the race that another thread might have closed the * file descriptor out from under the thread creating the file object. */ void fdclose(struct thread *td, struct file *fp, int idx) { struct filedesc *fdp = td->td_proc->p_fd; FILEDESC_XLOCK(fdp); if (fdp->fd_ofiles[idx].fde_file == fp) { fdfree(fdp, idx); FILEDESC_XUNLOCK(fdp); fdrop(fp, td); } else FILEDESC_XUNLOCK(fdp); } /* * Close any files on exec? */ void fdcloseexec(struct thread *td) { struct filedesc *fdp; struct filedescent *fde; struct file *fp; int i; fdp = td->td_proc->p_fd; KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); for (i = 0; i <= fdp->fd_lastfile; i++) { fde = &fdp->fd_ofiles[i]; fp = fde->fde_file; if (fp != NULL && (fp->f_type == DTYPE_MQUEUE || (fde->fde_flags & UF_EXCLOSE))) { FILEDESC_XLOCK(fdp); fdfree(fdp, i); (void) closefp(fdp, i, fp, td, 0); /* closefp() drops the FILEDESC lock. */ } } } /* * It is unsafe for set[ug]id processes to be started with file * descriptors 0..2 closed, as these descriptors are given implicit * significance in the Standard C library. fdcheckstd() will create a * descriptor referencing /dev/null for each of stdin, stdout, and * stderr that is not already open. */ int fdcheckstd(struct thread *td) { struct filedesc *fdp; register_t save; int i, error, devnull; fdp = td->td_proc->p_fd; KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); MPASS(fdp->fd_nfiles >= 3); devnull = -1; for (i = 0; i <= 2; i++) { if (fdp->fd_ofiles[i].fde_file != NULL) continue; save = td->td_retval[0]; if (devnull != -1) { error = do_dup(td, DUP_FIXED, devnull, i); } else { error = kern_openat(td, AT_FDCWD, "/dev/null", UIO_SYSSPACE, O_RDWR, 0); if (error == 0) { devnull = td->td_retval[0]; KASSERT(devnull == i, ("we didn't get our fd")); } } td->td_retval[0] = save; if (error != 0) return (error); } return (0); } /* * Internal form of close. Decrement reference count on file structure. * Note: td may be NULL when closing a file that was being passed in a * message. * * XXXRW: Giant is not required for the caller, but often will be held; this * makes it moderately likely the Giant will be recursed in the VFS case. */ int closef(struct file *fp, struct thread *td) { struct vnode *vp; struct flock lf; struct filedesc_to_leader *fdtol; struct filedesc *fdp; /* * POSIX record locking dictates that any close releases ALL * locks owned by this process. This is handled by setting * a flag in the unlock to free ONLY locks obeying POSIX * semantics, and not to free BSD-style file locks. * If the descriptor was in a message, POSIX-style locks * aren't passed with the descriptor, and the thread pointer * will be NULL. Callers should be careful only to pass a * NULL thread pointer when there really is no owning * context that might have locks, or the locks will be * leaked. */ if (fp->f_type == DTYPE_VNODE && td != NULL) { vp = fp->f_vnode; if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; lf.l_type = F_UNLCK; (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, F_UNLCK, &lf, F_POSIX); } fdtol = td->td_proc->p_fdtol; if (fdtol != NULL) { /* * Handle special case where file descriptor table is * shared between multiple process leaders. */ fdp = td->td_proc->p_fd; FILEDESC_XLOCK(fdp); for (fdtol = fdtol->fdl_next; fdtol != td->td_proc->p_fdtol; fdtol = fdtol->fdl_next) { if ((fdtol->fdl_leader->p_flag & P_ADVLOCK) == 0) continue; fdtol->fdl_holdcount++; FILEDESC_XUNLOCK(fdp); lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; lf.l_type = F_UNLCK; vp = fp->f_vnode; (void) VOP_ADVLOCK(vp, (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, F_POSIX); FILEDESC_XLOCK(fdp); fdtol->fdl_holdcount--; if (fdtol->fdl_holdcount == 0 && fdtol->fdl_wakeup != 0) { fdtol->fdl_wakeup = 0; wakeup(fdtol); } } FILEDESC_XUNLOCK(fdp); } } return (fdrop(fp, td)); } /* * Initialize the file pointer with the specified properties. * * The ops are set with release semantics to be certain that the flags, type, * and data are visible when ops is. This is to prevent ops methods from being * called with bad data. */ void finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) { fp->f_data = data; fp->f_flag = flag; fp->f_type = type; atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); } int fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, struct file **fpp, seq_t *seqp) { #ifdef CAPABILITIES struct filedescent *fde; #endif struct fdescenttbl *fdt; struct file *fp; u_int count; #ifdef CAPABILITIES seq_t seq; cap_rights_t haverights; int error; #endif fdt = fdp->fd_files; if ((u_int)fd >= fdt->fdt_nfiles) return (EBADF); /* * Fetch the descriptor locklessly. We avoid fdrop() races by * never raising a refcount above 0. To accomplish this we have * to use a cmpset loop rather than an atomic_add. The descriptor * must be re-verified once we acquire a reference to be certain * that the identity is still correct and we did not lose a race * due to preemption. */ for (;;) { #ifdef CAPABILITIES seq = seq_read(fd_seq(fdt, fd)); fde = &fdt->fdt_ofiles[fd]; haverights = *cap_rights_fde(fde); fp = fde->fde_file; if (!seq_consistent(fd_seq(fdt, fd), seq)) { cpu_spinwait(); continue; } #else fp = fdt->fdt_ofiles[fd].fde_file; #endif if (fp == NULL) return (EBADF); #ifdef CAPABILITIES if (needrightsp != NULL) { error = cap_check(&haverights, needrightsp); if (error != 0) return (error); } #endif retry: count = fp->f_count; if (count == 0) { /* * Force a reload. Other thread could reallocate the * table before this fd was closed, so it possible that * there is a stale fp pointer in cached version. */ fdt = *(struct fdescenttbl * volatile *)&(fdp->fd_files); continue; } /* * Use an acquire barrier to force re-reading of fdt so it is * refreshed for verification. */ if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) == 0) goto retry; fdt = fdp->fd_files; #ifdef CAPABILITIES if (seq_consistent_nomb(fd_seq(fdt, fd), seq)) #else if (fp == fdt->fdt_ofiles[fd].fde_file) #endif break; fdrop(fp, curthread); } *fpp = fp; if (seqp != NULL) { #ifdef CAPABILITIES *seqp = seq; #endif } return (0); } /* * Extract the file pointer associated with the specified descriptor for the * current user process. * * If the descriptor doesn't exist or doesn't match 'flags', EBADF is * returned. * * File's rights will be checked against the capability rights mask. * * If an error occured the non-zero error is returned and *fpp is set to * NULL. Otherwise *fpp is held and set and zero is returned. Caller is * responsible for fdrop(). */ static __inline int _fget(struct thread *td, int fd, struct file **fpp, int flags, cap_rights_t *needrightsp, seq_t *seqp) { struct filedesc *fdp; struct file *fp; cap_rights_t needrights; int error; *fpp = NULL; fdp = td->td_proc->p_fd; if (needrightsp != NULL) needrights = *needrightsp; else cap_rights_init(&needrights); error = fget_unlocked(fdp, fd, &needrights, &fp, seqp); if (error != 0) return (error); if (fp->f_ops == &badfileops) { fdrop(fp, td); return (EBADF); } /* * FREAD and FWRITE failure return EBADF as per POSIX. */ error = 0; switch (flags) { case FREAD: case FWRITE: if ((fp->f_flag & flags) == 0) error = EBADF; break; case FEXEC: if ((fp->f_flag & (FREAD | FEXEC)) == 0 || ((fp->f_flag & FWRITE) != 0)) error = EBADF; break; case 0: break; default: KASSERT(0, ("wrong flags")); } if (error != 0) { fdrop(fp, td); return (error); } *fpp = fp; return (0); } int fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) { return (_fget(td, fd, fpp, 0, rightsp, NULL)); } int fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp, struct file **fpp) { int error; #ifndef CAPABILITIES error = _fget(td, fd, fpp, 0, rightsp, NULL); if (maxprotp != NULL) *maxprotp = VM_PROT_ALL; #else struct filedesc *fdp = td->td_proc->p_fd; seq_t seq; MPASS(cap_rights_is_set(rightsp, CAP_MMAP)); for (;;) { error = _fget(td, fd, fpp, 0, rightsp, &seq); if (error != 0) return (error); /* * If requested, convert capability rights to access flags. */ if (maxprotp != NULL) *maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd)); if (!fd_modified(fdp, fd, seq)) break; fdrop(*fpp, td); } #endif return (error); } int fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) { return (_fget(td, fd, fpp, FREAD, rightsp, NULL)); } int fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) { return (_fget(td, fd, fpp, FWRITE, rightsp, NULL)); } int fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl, struct file **fpp) { struct filedesc *fdp = td->td_proc->p_fd; #ifndef CAPABILITIES return (fget_unlocked(fdp, fd, rightsp, fpp, NULL)); #else int error; seq_t seq; MPASS(cap_rights_is_set(rightsp, CAP_FCNTL)); for (;;) { error = fget_unlocked(fdp, fd, rightsp, fpp, &seq); if (error != 0) return (error); error = cap_fcntl_check(fdp, fd, needfcntl); if (!fd_modified(fdp, fd, seq)) break; fdrop(*fpp, td); } if (error != 0) { fdrop(*fpp, td); *fpp = NULL; } return (error); #endif } /* * Like fget() but loads the underlying vnode, or returns an error if the * descriptor does not represent a vnode. Note that pipes use vnodes but * never have VM objects. The returned vnode will be vref()'d. * * XXX: what about the unused flags ? */ static __inline int _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp, struct vnode **vpp) { struct file *fp; int error; *vpp = NULL; error = _fget(td, fd, &fp, flags, needrightsp, NULL); if (error != 0) return (error); if (fp->f_vnode == NULL) { error = EINVAL; } else { *vpp = fp->f_vnode; vref(*vpp); } fdrop(fp, td); return (error); } int fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) { return (_fgetvp(td, fd, 0, rightsp, vpp)); } int fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, struct filecaps *havecaps, struct vnode **vpp) { struct filedesc *fdp; struct file *fp; #ifdef CAPABILITIES int error; #endif fdp = td->td_proc->p_fd; fp = fget_locked(fdp, fd); if (fp == NULL || fp->f_ops == &badfileops) return (EBADF); #ifdef CAPABILITIES if (needrightsp != NULL) { error = cap_check(cap_rights(fdp, fd), needrightsp); if (error != 0) return (error); } #endif if (fp->f_vnode == NULL) return (EINVAL); *vpp = fp->f_vnode; vref(*vpp); filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps); return (0); } int fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) { return (_fgetvp(td, fd, FREAD, rightsp, vpp)); } int fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) { return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); } #ifdef notyet int fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) { return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); } #endif /* * Like fget() but loads the underlying socket, or returns an error if the * descriptor does not represent a socket. * * We bump the ref count on the returned socket. XXX Also obtain the SX lock * in the future. * * Note: fgetsock() and fputsock() are deprecated, as consumers should rely * on their file descriptor reference to prevent the socket from being free'd * during use. */ int fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp, u_int *fflagp) { struct file *fp; int error; *spp = NULL; if (fflagp != NULL) *fflagp = 0; if ((error = _fget(td, fd, &fp, 0, rightsp, NULL)) != 0) return (error); if (fp->f_type != DTYPE_SOCKET) { error = ENOTSOCK; } else { *spp = fp->f_data; if (fflagp) *fflagp = fp->f_flag; SOCK_LOCK(*spp); soref(*spp); SOCK_UNLOCK(*spp); } fdrop(fp, td); return (error); } /* * Drop the reference count on the socket and XXX release the SX lock in the * future. The last reference closes the socket. * * Note: fputsock() is deprecated, see comment for fgetsock(). */ void fputsock(struct socket *so) { ACCEPT_LOCK(); SOCK_LOCK(so); CURVNET_SET(so->so_vnet); sorele(so); CURVNET_RESTORE(); } /* * Handle the last reference to a file being closed. */ int _fdrop(struct file *fp, struct thread *td) { int error; if (fp->f_count != 0) panic("fdrop: count %d", fp->f_count); error = fo_close(fp, td); atomic_subtract_int(&openfiles, 1); crfree(fp->f_cred); free(fp->f_advice, M_FADVISE); uma_zfree(file_zone, fp); return (error); } /* * Apply an advisory lock on a file descriptor. * * Just attempt to get a record lock of the requested type on the entire file * (l_whence = SEEK_SET, l_start = 0, l_len = 0). */ #ifndef _SYS_SYSPROTO_H_ struct flock_args { int fd; int how; }; #endif /* ARGSUSED */ int sys_flock(struct thread *td, struct flock_args *uap) { struct file *fp; struct vnode *vp; struct flock lf; cap_rights_t rights; int error; error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp); if (error != 0) return (error); if (fp->f_type != DTYPE_VNODE) { fdrop(fp, td); return (EOPNOTSUPP); } vp = fp->f_vnode; lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; if (uap->how & LOCK_UN) { lf.l_type = F_UNLCK; atomic_clear_int(&fp->f_flag, FHASLOCK); error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); goto done2; } if (uap->how & LOCK_EX) lf.l_type = F_WRLCK; else if (uap->how & LOCK_SH) lf.l_type = F_RDLCK; else { error = EBADF; goto done2; } atomic_set_int(&fp->f_flag, FHASLOCK); error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); done2: fdrop(fp, td); return (error); } /* * Duplicate the specified descriptor to a free descriptor. */ int dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, int openerror, int *indxp) { struct filedescent *newfde, *oldfde; struct file *fp; int error, indx; KASSERT(openerror == ENODEV || openerror == ENXIO, ("unexpected error %d in %s", openerror, __func__)); /* * If the to-be-dup'd fd number is greater than the allowed number * of file descriptors, or the fd to be dup'd has already been * closed, then reject. */ FILEDESC_XLOCK(fdp); if ((fp = fget_locked(fdp, dfd)) == NULL) { FILEDESC_XUNLOCK(fdp); return (EBADF); } error = fdalloc(td, 0, &indx); if (error != 0) { FILEDESC_XUNLOCK(fdp); return (error); } /* * There are two cases of interest here. * * For ENODEV simply dup (dfd) to file descriptor (indx) and return. * * For ENXIO steal away the file structure from (dfd) and store it in * (indx). (dfd) is effectively closed by this operation. */ switch (openerror) { case ENODEV: /* * Check that the mode the file is being opened for is a * subset of the mode of the existing descriptor. */ if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { fdunused(fdp, indx); FILEDESC_XUNLOCK(fdp); return (EACCES); } fhold(fp); newfde = &fdp->fd_ofiles[indx]; oldfde = &fdp->fd_ofiles[dfd]; #ifdef CAPABILITIES seq_write_begin(&newfde->fde_seq); #endif memcpy(newfde, oldfde, fde_change_size); filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps); #ifdef CAPABILITIES seq_write_end(&newfde->fde_seq); #endif break; case ENXIO: /* * Steal away the file pointer from dfd and stuff it into indx. */ newfde = &fdp->fd_ofiles[indx]; oldfde = &fdp->fd_ofiles[dfd]; #ifdef CAPABILITIES seq_write_begin(&newfde->fde_seq); #endif memcpy(newfde, oldfde, fde_change_size); bzero(oldfde, fde_change_size); fdunused(fdp, dfd); #ifdef CAPABILITIES seq_write_end(&newfde->fde_seq); #endif break; } FILEDESC_XUNLOCK(fdp); *indxp = indx; return (0); } /* * Scan all active processes and prisons to see if any of them have a current * or root directory of `olddp'. If so, replace them with the new mount point. */ void mountcheckdirs(struct vnode *olddp, struct vnode *newdp) { struct filedesc *fdp; struct prison *pr; struct proc *p; int nrele; if (vrefcnt(olddp) == 1) return; nrele = 0; sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); fdp = fdhold(p); PROC_UNLOCK(p); if (fdp == NULL) continue; FILEDESC_XLOCK(fdp); if (fdp->fd_cdir == olddp) { vref(newdp); fdp->fd_cdir = newdp; nrele++; } if (fdp->fd_rdir == olddp) { vref(newdp); fdp->fd_rdir = newdp; nrele++; } if (fdp->fd_jdir == olddp) { vref(newdp); fdp->fd_jdir = newdp; nrele++; } FILEDESC_XUNLOCK(fdp); fddrop(fdp); } sx_sunlock(&allproc_lock); if (rootvnode == olddp) { vref(newdp); rootvnode = newdp; nrele++; } mtx_lock(&prison0.pr_mtx); if (prison0.pr_root == olddp) { vref(newdp); prison0.pr_root = newdp; nrele++; } mtx_unlock(&prison0.pr_mtx); sx_slock(&allprison_lock); TAILQ_FOREACH(pr, &allprison, pr_list) { mtx_lock(&pr->pr_mtx); if (pr->pr_root == olddp) { vref(newdp); pr->pr_root = newdp; nrele++; } mtx_unlock(&pr->pr_mtx); } sx_sunlock(&allprison_lock); while (nrele--) vrele(olddp); } struct filedesc_to_leader * filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) { struct filedesc_to_leader *fdtol; fdtol = malloc(sizeof(struct filedesc_to_leader), M_FILEDESC_TO_LEADER, M_WAITOK); fdtol->fdl_refcount = 1; fdtol->fdl_holdcount = 0; fdtol->fdl_wakeup = 0; fdtol->fdl_leader = leader; if (old != NULL) { FILEDESC_XLOCK(fdp); fdtol->fdl_next = old->fdl_next; fdtol->fdl_prev = old; old->fdl_next = fdtol; fdtol->fdl_next->fdl_prev = fdtol; FILEDESC_XUNLOCK(fdp); } else { fdtol->fdl_next = fdtol; fdtol->fdl_prev = fdtol; } return (fdtol); } /* * Get file structures globally. */ static int sysctl_kern_file(SYSCTL_HANDLER_ARGS) { struct xfile xf; struct filedesc *fdp; struct file *fp; struct proc *p; int error, n; error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); if (req->oldptr == NULL) { n = 0; sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); if (p->p_state == PRS_NEW) { PROC_UNLOCK(p); continue; } fdp = fdhold(p); PROC_UNLOCK(p); if (fdp == NULL) continue; /* overestimates sparse tables. */ if (fdp->fd_lastfile > 0) n += fdp->fd_lastfile; fddrop(fdp); } sx_sunlock(&allproc_lock); return (SYSCTL_OUT(req, 0, n * sizeof(xf))); } error = 0; bzero(&xf, sizeof(xf)); xf.xf_size = sizeof(xf); sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); if (p->p_state == PRS_NEW) { PROC_UNLOCK(p); continue; } if (p_cansee(req->td, p) != 0) { PROC_UNLOCK(p); continue; } xf.xf_pid = p->p_pid; xf.xf_uid = p->p_ucred->cr_uid; fdp = fdhold(p); PROC_UNLOCK(p); if (fdp == NULL) continue; FILEDESC_SLOCK(fdp); for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) { if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) continue; xf.xf_fd = n; xf.xf_file = fp; xf.xf_data = fp->f_data; xf.xf_vnode = fp->f_vnode; xf.xf_type = fp->f_type; xf.xf_count = fp->f_count; xf.xf_msgcount = 0; xf.xf_offset = foffset_get(fp); xf.xf_flag = fp->f_flag; error = SYSCTL_OUT(req, &xf, sizeof(xf)); if (error) break; } FILEDESC_SUNLOCK(fdp); fddrop(fdp); if (error) break; } sx_sunlock(&allproc_lock); return (error); } SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); #ifdef KINFO_FILE_SIZE CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); #endif static int xlate_fflags(int fflags) { static const struct { int fflag; int kf_fflag; } fflags_table[] = { { FAPPEND, KF_FLAG_APPEND }, { FASYNC, KF_FLAG_ASYNC }, { FFSYNC, KF_FLAG_FSYNC }, { FHASLOCK, KF_FLAG_HASLOCK }, { FNONBLOCK, KF_FLAG_NONBLOCK }, { FREAD, KF_FLAG_READ }, { FWRITE, KF_FLAG_WRITE }, { O_CREAT, KF_FLAG_CREAT }, { O_DIRECT, KF_FLAG_DIRECT }, { O_EXCL, KF_FLAG_EXCL }, { O_EXEC, KF_FLAG_EXEC }, { O_EXLOCK, KF_FLAG_EXLOCK }, { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, { O_SHLOCK, KF_FLAG_SHLOCK }, { O_TRUNC, KF_FLAG_TRUNC } }; unsigned int i; int kflags; kflags = 0; for (i = 0; i < nitems(fflags_table); i++) if (fflags & fflags_table[i].fflag) kflags |= fflags_table[i].kf_fflag; return (kflags); } /* Trim unused data from kf_path by truncating the structure size. */ static void pack_kinfo(struct kinfo_file *kif) { kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + strlen(kif->kf_path) + 1; kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); } static void export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp, struct kinfo_file *kif, struct filedesc *fdp) { int error; bzero(kif, sizeof(*kif)); /* Set a default type to allow for empty fill_kinfo() methods. */ kif->kf_type = KF_TYPE_UNKNOWN; kif->kf_flags = xlate_fflags(fp->f_flag); if (rightsp != NULL) kif->kf_cap_rights = *rightsp; else cap_rights_init(&kif->kf_cap_rights); kif->kf_fd = fd; kif->kf_ref_count = fp->f_count; kif->kf_offset = foffset_get(fp); /* * This may drop the filedesc lock, so the 'fp' cannot be * accessed after this call. */ error = fo_fill_kinfo(fp, kif, fdp); if (error == 0) kif->kf_status |= KF_ATTR_VALID; pack_kinfo(kif); } static void export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags, struct kinfo_file *kif) { int error; bzero(kif, sizeof(*kif)); kif->kf_type = KF_TYPE_VNODE; error = vn_fill_kinfo_vnode(vp, kif); if (error == 0) kif->kf_status |= KF_ATTR_VALID; kif->kf_flags = xlate_fflags(fflags); cap_rights_init(&kif->kf_cap_rights); kif->kf_fd = fd; kif->kf_ref_count = -1; kif->kf_offset = -1; pack_kinfo(kif); vrele(vp); } struct export_fd_buf { struct filedesc *fdp; struct sbuf *sb; ssize_t remainder; struct kinfo_file kif; }; static int export_kinfo_to_sb(struct export_fd_buf *efbuf) { struct kinfo_file *kif; kif = &efbuf->kif; if (efbuf->remainder != -1) { if (efbuf->remainder < kif->kf_structsize) { /* Terminate export. */ efbuf->remainder = 0; return (0); } efbuf->remainder -= kif->kf_structsize; } return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM); } static int export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp, struct export_fd_buf *efbuf) { int error; if (efbuf->remainder == 0) return (0); export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp); FILEDESC_SUNLOCK(efbuf->fdp); error = export_kinfo_to_sb(efbuf); FILEDESC_SLOCK(efbuf->fdp); return (error); } static int export_vnode_to_sb(struct vnode *vp, int fd, int fflags, struct export_fd_buf *efbuf) { int error; if (efbuf->remainder == 0) return (0); if (efbuf->fdp != NULL) FILEDESC_SUNLOCK(efbuf->fdp); export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif); error = export_kinfo_to_sb(efbuf); if (efbuf->fdp != NULL) FILEDESC_SLOCK(efbuf->fdp); return (error); } /* * Store a process file descriptor information to sbuf. * * Takes a locked proc as argument, and returns with the proc unlocked. */ int kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) { struct file *fp; struct filedesc *fdp; struct export_fd_buf *efbuf; struct vnode *cttyvp, *textvp, *tracevp; int error, i; cap_rights_t rights; PROC_LOCK_ASSERT(p, MA_OWNED); /* ktrace vnode */ tracevp = p->p_tracevp; if (tracevp != NULL) vref(tracevp); /* text vnode */ textvp = p->p_textvp; if (textvp != NULL) vref(textvp); /* Controlling tty. */ cttyvp = NULL; if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { cttyvp = p->p_pgrp->pg_session->s_ttyvp; if (cttyvp != NULL) vref(cttyvp); } fdp = fdhold(p); PROC_UNLOCK(p); efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); efbuf->fdp = NULL; efbuf->sb = sb; efbuf->remainder = maxlen; if (tracevp != NULL) export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE, efbuf); if (textvp != NULL) export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf); if (cttyvp != NULL) export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE, efbuf); error = 0; if (fdp == NULL) goto fail; efbuf->fdp = fdp; FILEDESC_SLOCK(fdp); /* working directory */ if (fdp->fd_cdir != NULL) { vref(fdp->fd_cdir); export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf); } /* root directory */ if (fdp->fd_rdir != NULL) { vref(fdp->fd_rdir); export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf); } /* jail directory */ if (fdp->fd_jdir != NULL) { vref(fdp->fd_jdir); export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf); } for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) continue; #ifdef CAPABILITIES rights = *cap_rights(fdp, i); #else /* !CAPABILITIES */ cap_rights_init(&rights); #endif /* * Create sysctl entry. It is OK to drop the filedesc * lock inside of export_file_to_sb() as we will * re-validate and re-evaluate its properties when the * loop continues. */ error = export_file_to_sb(fp, i, &rights, efbuf); if (error != 0 || efbuf->remainder == 0) break; } FILEDESC_SUNLOCK(fdp); fddrop(fdp); fail: free(efbuf, M_TEMP); return (error); } #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) /* * Get per-process file descriptors for use by procstat(1), et al. */ static int sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) { struct sbuf sb; struct proc *p; ssize_t maxlen; int error, error2, *name; name = (int *)arg1; sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); sbuf_clear_flags(&sb, SBUF_INCLUDENUL); error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); if (error != 0) { sbuf_delete(&sb); return (error); } maxlen = req->oldptr != NULL ? req->oldlen : -1; error = kern_proc_filedesc_out(p, &sb, maxlen); error2 = sbuf_finish(&sb); sbuf_delete(&sb); return (error != 0 ? error : error2); } #ifdef KINFO_OFILE_SIZE CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); #endif #ifdef COMPAT_FREEBSD7 static void kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif) { okif->kf_structsize = sizeof(*okif); okif->kf_type = kif->kf_type; okif->kf_fd = kif->kf_fd; okif->kf_ref_count = kif->kf_ref_count; okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE | KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK | KF_FLAG_DIRECT | KF_FLAG_HASLOCK); okif->kf_offset = kif->kf_offset; okif->kf_vnode_type = kif->kf_vnode_type; okif->kf_sock_domain = kif->kf_sock_domain; okif->kf_sock_type = kif->kf_sock_type; okif->kf_sock_protocol = kif->kf_sock_protocol; strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path)); okif->kf_sa_local = kif->kf_sa_local; okif->kf_sa_peer = kif->kf_sa_peer; } static int export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif, struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req) { int error; vref(vp); FILEDESC_SUNLOCK(fdp); export_vnode_to_kinfo(vp, type, 0, kif); kinfo_to_okinfo(kif, okif); error = SYSCTL_OUT(req, okif, sizeof(*okif)); FILEDESC_SLOCK(fdp); return (error); } /* * Get per-process file descriptors for use by procstat(1), et al. */ static int sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) { struct kinfo_ofile *okif; struct kinfo_file *kif; struct filedesc *fdp; int error, i, *name; struct file *fp; struct proc *p; name = (int *)arg1; error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); if (error != 0) return (error); fdp = fdhold(p); PROC_UNLOCK(p); if (fdp == NULL) return (ENOENT); kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK); FILEDESC_SLOCK(fdp); if (fdp->fd_cdir != NULL) export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, okif, fdp, req); if (fdp->fd_rdir != NULL) export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, okif, fdp, req); if (fdp->fd_jdir != NULL) export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, okif, fdp, req); for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) { if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) continue; export_file_to_kinfo(fp, i, NULL, kif, fdp); FILEDESC_SUNLOCK(fdp); kinfo_to_okinfo(kif, okif); error = SYSCTL_OUT(req, okif, sizeof(*okif)); FILEDESC_SLOCK(fdp); if (error) break; } FILEDESC_SUNLOCK(fdp); fddrop(fdp); free(kif, M_TEMP); free(okif, M_TEMP); return (0); } static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, "Process ofiledesc entries"); #endif /* COMPAT_FREEBSD7 */ int vntype_to_kinfo(int vtype) { struct { int vtype; int kf_vtype; } vtypes_table[] = { { VBAD, KF_VTYPE_VBAD }, { VBLK, KF_VTYPE_VBLK }, { VCHR, KF_VTYPE_VCHR }, { VDIR, KF_VTYPE_VDIR }, { VFIFO, KF_VTYPE_VFIFO }, { VLNK, KF_VTYPE_VLNK }, { VNON, KF_VTYPE_VNON }, { VREG, KF_VTYPE_VREG }, { VSOCK, KF_VTYPE_VSOCK } }; unsigned int i; /* * Perform vtype translation. */ for (i = 0; i < nitems(vtypes_table); i++) if (vtypes_table[i].vtype == vtype) return (vtypes_table[i].kf_vtype); return (KF_VTYPE_UNKNOWN); } static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, "Process filedesc entries"); /* * Store a process current working directory information to sbuf. * * Takes a locked proc as argument, and returns with the proc unlocked. */ int kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) { struct filedesc *fdp; struct export_fd_buf *efbuf; int error; PROC_LOCK_ASSERT(p, MA_OWNED); fdp = fdhold(p); PROC_UNLOCK(p); if (fdp == NULL) return (EINVAL); efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); efbuf->fdp = fdp; efbuf->sb = sb; efbuf->remainder = maxlen; FILEDESC_SLOCK(fdp); if (fdp->fd_cdir == NULL) error = EINVAL; else { vref(fdp->fd_cdir); error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf); } FILEDESC_SUNLOCK(fdp); fddrop(fdp); free(efbuf, M_TEMP); return (error); } /* * Get per-process current working directory. */ static int sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS) { struct sbuf sb; struct proc *p; ssize_t maxlen; int error, error2, *name; name = (int *)arg1; sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req); sbuf_clear_flags(&sb, SBUF_INCLUDENUL); error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); if (error != 0) { sbuf_delete(&sb); return (error); } maxlen = req->oldptr != NULL ? req->oldlen : -1; error = kern_proc_cwd_out(p, &sb, maxlen); error2 = sbuf_finish(&sb); sbuf_delete(&sb); return (error != 0 ? error : error2); } static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_cwd, "Process current working directory"); #ifdef DDB /* * For the purposes of debugging, generate a human-readable string for the * file type. */ static const char * file_type_to_name(short type) { switch (type) { case 0: return ("zero"); case DTYPE_VNODE: return ("vnod"); case DTYPE_SOCKET: return ("sock"); case DTYPE_PIPE: return ("pipe"); case DTYPE_FIFO: return ("fifo"); case DTYPE_KQUEUE: return ("kque"); case DTYPE_CRYPTO: return ("crpt"); case DTYPE_MQUEUE: return ("mque"); case DTYPE_SHM: return ("shm"); case DTYPE_SEM: return ("ksem"); default: return ("unkn"); } } /* * For the purposes of debugging, identify a process (if any, perhaps one of * many) that references the passed file in its file descriptor array. Return * NULL if none. */ static struct proc * file_to_first_proc(struct file *fp) { struct filedesc *fdp; struct proc *p; int n; FOREACH_PROC_IN_SYSTEM(p) { if (p->p_state == PRS_NEW) continue; fdp = p->p_fd; if (fdp == NULL) continue; for (n = 0; n <= fdp->fd_lastfile; n++) { if (fp == fdp->fd_ofiles[n].fde_file) return (p); } } return (NULL); } static void db_print_file(struct file *fp, int header) { struct proc *p; if (header) db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", "File", "Type", "Data", "Flag", "GCFl", "Count", "MCount", "Vnode", "FPID", "FCmd"); p = file_to_first_proc(fp); db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp, file_type_to_name(fp->f_type), fp->f_data, fp->f_flag, 0, fp->f_count, 0, fp->f_vnode, p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); } DB_SHOW_COMMAND(file, db_show_file) { struct file *fp; if (!have_addr) { db_printf("usage: show file \n"); return; } fp = (struct file *)addr; db_print_file(fp, 1); } DB_SHOW_COMMAND(files, db_show_files) { struct filedesc *fdp; struct file *fp; struct proc *p; int header; int n; header = 1; FOREACH_PROC_IN_SYSTEM(p) { if (p->p_state == PRS_NEW) continue; if ((fdp = p->p_fd) == NULL) continue; for (n = 0; n <= fdp->fd_lastfile; ++n) { if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) continue; db_print_file(fp, header); header = 0; } } } #endif SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, &maxfilesperproc, 0, "Maximum files allowed open per process"); SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, &maxfiles, 0, "Maximum number of files"); SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); /* ARGSUSED*/ static void filelistinit(void *dummy) { file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); } SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); /*-------------------------------------------------------------------*/ static int badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { return (EBADF); } static int badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td) { return (EINVAL); } static int badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td) { return (EBADF); } static int badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td) { return (0); } static int badfo_kqfilter(struct file *fp, struct knote *kn) { return (EBADF); } static int badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td) { return (EBADF); } static int badfo_close(struct file *fp, struct thread *td) { return (0); } static int badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td) { return (EBADF); } static int badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, struct thread *td) { return (EBADF); } static int badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, int kflags, struct thread *td) { return (EBADF); } static int badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) { return (0); } struct fileops badfileops = { .fo_read = badfo_readwrite, .fo_write = badfo_readwrite, .fo_truncate = badfo_truncate, .fo_ioctl = badfo_ioctl, .fo_poll = badfo_poll, .fo_kqfilter = badfo_kqfilter, .fo_stat = badfo_stat, .fo_close = badfo_close, .fo_chmod = badfo_chmod, .fo_chown = badfo_chown, .fo_sendfile = badfo_sendfile, .fo_fill_kinfo = badfo_fill_kinfo, }; int invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { return (EOPNOTSUPP); } int invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td) { return (EINVAL); } int invfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td) { return (ENOTTY); } int invfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td) { return (poll_no_poll(events)); } int invfo_kqfilter(struct file *fp, struct knote *kn) { return (EINVAL); } int invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td) { return (EINVAL); } int invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, struct thread *td) { return (EINVAL); } int invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, int kflags, struct thread *td) { return (EINVAL); } /*-------------------------------------------------------------------*/ /* * File Descriptor pseudo-device driver (/dev/fd/). * * Opening minor device N dup()s the file (if any) connected to file * descriptor N belonging to the calling process. Note that this driver * consists of only the ``open()'' routine, because all subsequent * references to this file will be direct to the other driver. * * XXX: we could give this one a cloning event handler if necessary. */ /* ARGSUSED */ static int fdopen(struct cdev *dev, int mode, int type, struct thread *td) { /* * XXX Kludge: set curthread->td_dupfd to contain the value of the * the file descriptor being sought for duplication. The error * return ensures that the vnode for this device will be released * by vn_open. Open will detect this special error and take the * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN * will simply report the error. */ td->td_dupfd = dev2unit(dev); return (ENODEV); } static struct cdevsw fildesc_cdevsw = { .d_version = D_VERSION, .d_open = fdopen, .d_name = "FD", }; static void fildesc_drvinit(void *unused) { struct cdev *dev; dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, UID_ROOT, GID_WHEEL, 0666, "fd/0"); make_dev_alias(dev, "stdin"); dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, UID_ROOT, GID_WHEEL, 0666, "fd/1"); make_dev_alias(dev, "stdout"); dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, UID_ROOT, GID_WHEEL, 0666, "fd/2"); make_dev_alias(dev, "stderr"); } SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); Index: head/sys/kern/kern_event.c =================================================================== --- head/sys/kern/kern_event.c (revision 284214) +++ head/sys/kern/kern_event.c (revision 284215) @@ -1,2384 +1,2380 @@ /*- * Copyright (c) 1999,2000,2001 Jonathan Lemon * Copyright 2004 John-Mark Gurney * Copyright (c) 2009 Apple, Inc. * 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. */ #include __FBSDID("$FreeBSD$"); #include "opt_ktrace.h" #include "opt_kqueue.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KTRACE #include #endif #include static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system"); /* * This lock is used if multiple kq locks are required. This possibly * should be made into a per proc lock. */ static struct mtx kq_global; MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF); #define KQ_GLOBAL_LOCK(lck, haslck) do { \ if (!haslck) \ mtx_lock(lck); \ haslck = 1; \ } while (0) #define KQ_GLOBAL_UNLOCK(lck, haslck) do { \ if (haslck) \ mtx_unlock(lck); \ haslck = 0; \ } while (0) TASKQUEUE_DEFINE_THREAD(kqueue); static int kevent_copyout(void *arg, struct kevent *kevp, int count); static int kevent_copyin(void *arg, struct kevent *kevp, int count); static int kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int waitok); static int kqueue_acquire(struct file *fp, struct kqueue **kqp); static void kqueue_release(struct kqueue *kq, int locked); static int kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident, int waitok); static void kqueue_task(void *arg, int pending); static int kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops, const struct timespec *timeout, struct kevent *keva, struct thread *td); static void kqueue_wakeup(struct kqueue *kq); static struct filterops *kqueue_fo_find(int filt); static void kqueue_fo_release(int filt); static fo_ioctl_t kqueue_ioctl; static fo_poll_t kqueue_poll; static fo_kqfilter_t kqueue_kqfilter; static fo_stat_t kqueue_stat; static fo_close_t kqueue_close; static fo_fill_kinfo_t kqueue_fill_kinfo; static struct fileops kqueueops = { .fo_read = invfo_rdwr, .fo_write = invfo_rdwr, .fo_truncate = invfo_truncate, .fo_ioctl = kqueue_ioctl, .fo_poll = kqueue_poll, .fo_kqfilter = kqueue_kqfilter, .fo_stat = kqueue_stat, .fo_close = kqueue_close, .fo_chmod = invfo_chmod, .fo_chown = invfo_chown, .fo_sendfile = invfo_sendfile, .fo_fill_kinfo = kqueue_fill_kinfo, }; static int knote_attach(struct knote *kn, struct kqueue *kq); static void knote_drop(struct knote *kn, struct thread *td); static void knote_enqueue(struct knote *kn); static void knote_dequeue(struct knote *kn); static void knote_init(void); static struct knote *knote_alloc(int waitok); static void knote_free(struct knote *kn); static void filt_kqdetach(struct knote *kn); static int filt_kqueue(struct knote *kn, long hint); static int filt_procattach(struct knote *kn); static void filt_procdetach(struct knote *kn); static int filt_proc(struct knote *kn, long hint); static int filt_fileattach(struct knote *kn); static void filt_timerexpire(void *knx); static int filt_timerattach(struct knote *kn); static void filt_timerdetach(struct knote *kn); static int filt_timer(struct knote *kn, long hint); static int filt_userattach(struct knote *kn); static void filt_userdetach(struct knote *kn); static int filt_user(struct knote *kn, long hint); static void filt_usertouch(struct knote *kn, struct kevent *kev, u_long type); static struct filterops file_filtops = { .f_isfd = 1, .f_attach = filt_fileattach, }; static struct filterops kqread_filtops = { .f_isfd = 1, .f_detach = filt_kqdetach, .f_event = filt_kqueue, }; /* XXX - move to kern_proc.c? */ static struct filterops proc_filtops = { .f_isfd = 0, .f_attach = filt_procattach, .f_detach = filt_procdetach, .f_event = filt_proc, }; static struct filterops timer_filtops = { .f_isfd = 0, .f_attach = filt_timerattach, .f_detach = filt_timerdetach, .f_event = filt_timer, }; static struct filterops user_filtops = { .f_attach = filt_userattach, .f_detach = filt_userdetach, .f_event = filt_user, .f_touch = filt_usertouch, }; static uma_zone_t knote_zone; static atomic_uint kq_ncallouts = ATOMIC_VAR_INIT(0); static unsigned int kq_calloutmax = 4 * 1024; SYSCTL_UINT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW, &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue"); /* XXX - ensure not KN_INFLUX?? */ #define KNOTE_ACTIVATE(kn, islock) do { \ if ((islock)) \ mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED); \ else \ KQ_LOCK((kn)->kn_kq); \ (kn)->kn_status |= KN_ACTIVE; \ if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \ knote_enqueue((kn)); \ if (!(islock)) \ KQ_UNLOCK((kn)->kn_kq); \ } while(0) #define KQ_LOCK(kq) do { \ mtx_lock(&(kq)->kq_lock); \ } while (0) #define KQ_FLUX_WAKEUP(kq) do { \ if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) { \ (kq)->kq_state &= ~KQ_FLUXWAIT; \ wakeup((kq)); \ } \ } while (0) #define KQ_UNLOCK_FLUX(kq) do { \ KQ_FLUX_WAKEUP(kq); \ mtx_unlock(&(kq)->kq_lock); \ } while (0) #define KQ_UNLOCK(kq) do { \ mtx_unlock(&(kq)->kq_lock); \ } while (0) #define KQ_OWNED(kq) do { \ mtx_assert(&(kq)->kq_lock, MA_OWNED); \ } while (0) #define KQ_NOTOWNED(kq) do { \ mtx_assert(&(kq)->kq_lock, MA_NOTOWNED); \ } while (0) #define KN_LIST_LOCK(kn) do { \ if (kn->kn_knlist != NULL) \ kn->kn_knlist->kl_lock(kn->kn_knlist->kl_lockarg); \ } while (0) #define KN_LIST_UNLOCK(kn) do { \ if (kn->kn_knlist != NULL) \ kn->kn_knlist->kl_unlock(kn->kn_knlist->kl_lockarg); \ } while (0) #define KNL_ASSERT_LOCK(knl, islocked) do { \ if (islocked) \ KNL_ASSERT_LOCKED(knl); \ else \ KNL_ASSERT_UNLOCKED(knl); \ } while (0) #ifdef INVARIANTS #define KNL_ASSERT_LOCKED(knl) do { \ knl->kl_assert_locked((knl)->kl_lockarg); \ } while (0) #define KNL_ASSERT_UNLOCKED(knl) do { \ knl->kl_assert_unlocked((knl)->kl_lockarg); \ } while (0) #else /* !INVARIANTS */ #define KNL_ASSERT_LOCKED(knl) do {} while(0) #define KNL_ASSERT_UNLOCKED(knl) do {} while (0) #endif /* INVARIANTS */ #ifndef KN_HASHSIZE #define KN_HASHSIZE 64 /* XXX should be tunable */ #endif #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) static int filt_nullattach(struct knote *kn) { return (ENXIO); }; struct filterops null_filtops = { .f_isfd = 0, .f_attach = filt_nullattach, }; /* XXX - make SYSINIT to add these, and move into respective modules. */ extern struct filterops sig_filtops; extern struct filterops fs_filtops; /* * Table for for all system-defined filters. */ static struct mtx filterops_lock; MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops", MTX_DEF); static struct { struct filterops *for_fop; int for_nolock; int for_refcnt; } sysfilt_ops[EVFILT_SYSCOUNT] = { { &file_filtops, 1 }, /* EVFILT_READ */ { &file_filtops, 1 }, /* EVFILT_WRITE */ { &null_filtops }, /* EVFILT_AIO */ { &file_filtops, 1 }, /* EVFILT_VNODE */ { &proc_filtops, 1 }, /* EVFILT_PROC */ { &sig_filtops, 1 }, /* EVFILT_SIGNAL */ { &timer_filtops, 1 }, /* EVFILT_TIMER */ { &file_filtops, 1 }, /* EVFILT_PROCDESC */ { &fs_filtops, 1 }, /* EVFILT_FS */ { &null_filtops }, /* EVFILT_LIO */ { &user_filtops, 1 }, /* EVFILT_USER */ { &null_filtops }, /* EVFILT_SENDFILE */ }; /* * Simple redirection for all cdevsw style objects to call their fo_kqfilter * method. */ static int filt_fileattach(struct knote *kn) { return (fo_kqfilter(kn->kn_fp, kn)); } /*ARGSUSED*/ static int kqueue_kqfilter(struct file *fp, struct knote *kn) { struct kqueue *kq = kn->kn_fp->f_data; if (kn->kn_filter != EVFILT_READ) return (EINVAL); kn->kn_status |= KN_KQUEUE; kn->kn_fop = &kqread_filtops; knlist_add(&kq->kq_sel.si_note, kn, 0); return (0); } static void filt_kqdetach(struct knote *kn) { struct kqueue *kq = kn->kn_fp->f_data; knlist_remove(&kq->kq_sel.si_note, kn, 0); } /*ARGSUSED*/ static int filt_kqueue(struct knote *kn, long hint) { struct kqueue *kq = kn->kn_fp->f_data; kn->kn_data = kq->kq_count; return (kn->kn_data > 0); } /* XXX - move to kern_proc.c? */ static int filt_procattach(struct knote *kn) { struct proc *p; int immediate; int error; immediate = 0; p = pfind(kn->kn_id); if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) { p = zpfind(kn->kn_id); immediate = 1; } else if (p != NULL && (p->p_flag & P_WEXIT)) { immediate = 1; } if (p == NULL) return (ESRCH); if ((error = p_cansee(curthread, p))) { PROC_UNLOCK(p); return (error); } kn->kn_ptr.p_proc = p; kn->kn_flags |= EV_CLEAR; /* automatically set */ /* * internal flag indicating registration done by kernel */ if (kn->kn_flags & EV_FLAG1) { kn->kn_data = kn->kn_sdata; /* ppid */ kn->kn_fflags = NOTE_CHILD; kn->kn_flags &= ~EV_FLAG1; } if (immediate == 0) knlist_add(&p->p_klist, kn, 1); /* * Immediately activate any exit notes if the target process is a * zombie. This is necessary to handle the case where the target * process, e.g. a child, dies before the kevent is registered. */ if (immediate && filt_proc(kn, NOTE_EXIT)) KNOTE_ACTIVATE(kn, 0); PROC_UNLOCK(p); return (0); } /* * The knote may be attached to a different process, which may exit, * leaving nothing for the knote to be attached to. So when the process * exits, the knote is marked as DETACHED and also flagged as ONESHOT so * it will be deleted when read out. However, as part of the knote deletion, * this routine is called, so a check is needed to avoid actually performing * a detach, because the original process does not exist any more. */ /* XXX - move to kern_proc.c? */ static void filt_procdetach(struct knote *kn) { struct proc *p; p = kn->kn_ptr.p_proc; knlist_remove(&p->p_klist, kn, 0); kn->kn_ptr.p_proc = NULL; } /* XXX - move to kern_proc.c? */ static int filt_proc(struct knote *kn, long hint) { struct proc *p; u_int event; p = kn->kn_ptr.p_proc; /* Mask off extra data. */ event = (u_int)hint & NOTE_PCTRLMASK; /* If the user is interested in this event, record it. */ if (kn->kn_sfflags & event) kn->kn_fflags |= event; /* Process is gone, so flag the event as finished. */ if (event == NOTE_EXIT) { if (!(kn->kn_status & KN_DETACHED)) knlist_remove_inevent(&p->p_klist, kn); kn->kn_flags |= EV_EOF | EV_ONESHOT; kn->kn_ptr.p_proc = NULL; if (kn->kn_fflags & NOTE_EXIT) kn->kn_data = p->p_xstat; if (kn->kn_fflags == 0) kn->kn_flags |= EV_DROP; return (1); } return (kn->kn_fflags != 0); } /* * Called when the process forked. It mostly does the same as the * knote(), activating all knotes registered to be activated when the * process forked. Additionally, for each knote attached to the * parent, check whether user wants to track the new process. If so * attach a new knote to it, and immediately report an event with the * child's pid. */ void knote_fork(struct knlist *list, int pid) { struct kqueue *kq; struct knote *kn; struct kevent kev; int error; if (list == NULL) return; list->kl_lock(list->kl_lockarg); SLIST_FOREACH(kn, &list->kl_list, kn_selnext) { /* * XXX - Why do we skip the kn if it is _INFLUX? Does this * mean we will not properly wake up some notes? */ if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) continue; kq = kn->kn_kq; KQ_LOCK(kq); if ((kn->kn_status & (KN_INFLUX | KN_SCAN)) == KN_INFLUX) { KQ_UNLOCK(kq); continue; } /* * The same as knote(), activate the event. */ if ((kn->kn_sfflags & NOTE_TRACK) == 0) { kn->kn_status |= KN_HASKQLOCK; if (kn->kn_fop->f_event(kn, NOTE_FORK)) KNOTE_ACTIVATE(kn, 1); kn->kn_status &= ~KN_HASKQLOCK; KQ_UNLOCK(kq); continue; } /* * The NOTE_TRACK case. In addition to the activation * of the event, we need to register new event to * track the child. Drop the locks in preparation for * the call to kqueue_register(). */ kn->kn_status |= KN_INFLUX; KQ_UNLOCK(kq); list->kl_unlock(list->kl_lockarg); /* * Activate existing knote and register a knote with * new process. */ kev.ident = pid; kev.filter = kn->kn_filter; kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; kev.fflags = kn->kn_sfflags; kev.data = kn->kn_id; /* parent */ kev.udata = kn->kn_kevent.udata;/* preserve udata */ error = kqueue_register(kq, &kev, NULL, 0); if (error) kn->kn_fflags |= NOTE_TRACKERR; if (kn->kn_fop->f_event(kn, NOTE_FORK)) KNOTE_ACTIVATE(kn, 0); KQ_LOCK(kq); kn->kn_status &= ~KN_INFLUX; KQ_UNLOCK_FLUX(kq); list->kl_lock(list->kl_lockarg); } list->kl_unlock(list->kl_lockarg); } /* * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the * interval timer support code. */ #define NOTE_TIMER_PRECMASK (NOTE_SECONDS|NOTE_MSECONDS|NOTE_USECONDS| \ NOTE_NSECONDS) static __inline sbintime_t timer2sbintime(intptr_t data, int flags) { sbintime_t modifier; switch (flags & NOTE_TIMER_PRECMASK) { case NOTE_SECONDS: modifier = SBT_1S; break; case NOTE_MSECONDS: /* FALLTHROUGH */ case 0: modifier = SBT_1MS; break; case NOTE_USECONDS: modifier = SBT_1US; break; case NOTE_NSECONDS: modifier = SBT_1NS; break; default: return (-1); } #ifdef __LP64__ if (data > SBT_MAX / modifier) return (SBT_MAX); #endif return (modifier * data); } static void filt_timerexpire(void *knx) { struct callout *calloutp; struct knote *kn; kn = knx; kn->kn_data++; KNOTE_ACTIVATE(kn, 0); /* XXX - handle locking */ if ((kn->kn_flags & EV_ONESHOT) != EV_ONESHOT) { calloutp = (struct callout *)kn->kn_hook; *kn->kn_ptr.p_nexttime += timer2sbintime(kn->kn_sdata, kn->kn_sfflags); callout_reset_sbt_on(calloutp, *kn->kn_ptr.p_nexttime, 0, filt_timerexpire, kn, PCPU_GET(cpuid), C_ABSOLUTE); } } /* * data contains amount of time to sleep */ static int filt_timerattach(struct knote *kn) { struct callout *calloutp; sbintime_t to; unsigned int ncallouts; if ((intptr_t)kn->kn_sdata < 0) return (EINVAL); if ((intptr_t)kn->kn_sdata == 0 && (kn->kn_flags & EV_ONESHOT) == 0) kn->kn_sdata = 1; /* Only precision unit are supported in flags so far */ if (kn->kn_sfflags & ~NOTE_TIMER_PRECMASK) return (EINVAL); to = timer2sbintime(kn->kn_sdata, kn->kn_sfflags); if (to < 0) return (EINVAL); ncallouts = atomic_load_explicit(&kq_ncallouts, memory_order_relaxed); do { if (ncallouts >= kq_calloutmax) return (ENOMEM); } while (!atomic_compare_exchange_weak_explicit(&kq_ncallouts, &ncallouts, ncallouts + 1, memory_order_relaxed, memory_order_relaxed)); kn->kn_flags |= EV_CLEAR; /* automatically set */ kn->kn_status &= ~KN_DETACHED; /* knlist_add clears it */ kn->kn_ptr.p_nexttime = malloc(sizeof(sbintime_t), M_KQUEUE, M_WAITOK); calloutp = malloc(sizeof(*calloutp), M_KQUEUE, M_WAITOK); callout_init(calloutp, 1); kn->kn_hook = calloutp; *kn->kn_ptr.p_nexttime = to + sbinuptime(); callout_reset_sbt_on(calloutp, *kn->kn_ptr.p_nexttime, 0, filt_timerexpire, kn, PCPU_GET(cpuid), C_ABSOLUTE); return (0); } static void filt_timerdetach(struct knote *kn) { struct callout *calloutp; unsigned int old; calloutp = (struct callout *)kn->kn_hook; callout_drain(calloutp); free(calloutp, M_KQUEUE); free(kn->kn_ptr.p_nexttime, M_KQUEUE); old = atomic_fetch_sub_explicit(&kq_ncallouts, 1, memory_order_relaxed); KASSERT(old > 0, ("Number of callouts cannot become negative")); kn->kn_status |= KN_DETACHED; /* knlist_remove sets it */ } static int filt_timer(struct knote *kn, long hint) { return (kn->kn_data != 0); } static int filt_userattach(struct knote *kn) { /* * EVFILT_USER knotes are not attached to anything in the kernel. */ kn->kn_hook = NULL; if (kn->kn_fflags & NOTE_TRIGGER) kn->kn_hookid = 1; else kn->kn_hookid = 0; return (0); } static void filt_userdetach(__unused struct knote *kn) { /* * EVFILT_USER knotes are not attached to anything in the kernel. */ } static int filt_user(struct knote *kn, __unused long hint) { return (kn->kn_hookid); } static void filt_usertouch(struct knote *kn, struct kevent *kev, u_long type) { u_int ffctrl; switch (type) { case EVENT_REGISTER: if (kev->fflags & NOTE_TRIGGER) kn->kn_hookid = 1; ffctrl = kev->fflags & NOTE_FFCTRLMASK; kev->fflags &= NOTE_FFLAGSMASK; switch (ffctrl) { case NOTE_FFNOP: break; case NOTE_FFAND: kn->kn_sfflags &= kev->fflags; break; case NOTE_FFOR: kn->kn_sfflags |= kev->fflags; break; case NOTE_FFCOPY: kn->kn_sfflags = kev->fflags; break; default: /* XXX Return error? */ break; } kn->kn_sdata = kev->data; if (kev->flags & EV_CLEAR) { kn->kn_hookid = 0; kn->kn_data = 0; kn->kn_fflags = 0; } break; case EVENT_PROCESS: *kev = kn->kn_kevent; kev->fflags = kn->kn_sfflags; kev->data = kn->kn_sdata; if (kn->kn_flags & EV_CLEAR) { kn->kn_hookid = 0; kn->kn_data = 0; kn->kn_fflags = 0; } break; default: panic("filt_usertouch() - invalid type (%ld)", type); break; } } int sys_kqueue(struct thread *td, struct kqueue_args *uap) { return (kern_kqueue(td, 0)); } int kern_kqueue(struct thread *td, int flags) { struct filedesc *fdp; struct kqueue *kq; struct file *fp; struct proc *p; struct ucred *cred; int fd, error; p = td->td_proc; cred = td->td_ucred; crhold(cred); - PROC_LOCK(p); - if (!chgkqcnt(cred->cr_ruidinfo, 1, lim_cur(td->td_proc, - RLIMIT_KQUEUES))) { - PROC_UNLOCK(p); + if (!chgkqcnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_KQUEUES))) { crfree(cred); return (ENOMEM); } - PROC_UNLOCK(p); fdp = p->p_fd; error = falloc(td, &fp, &fd, flags); if (error) goto done2; /* An extra reference on `fp' has been held for us by falloc(). */ kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO); mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF|MTX_DUPOK); TAILQ_INIT(&kq->kq_head); kq->kq_fdp = fdp; kq->kq_cred = cred; knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock); TASK_INIT(&kq->kq_task, 0, kqueue_task, kq); FILEDESC_XLOCK(fdp); TAILQ_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list); FILEDESC_XUNLOCK(fdp); finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops); fdrop(fp, td); td->td_retval[0] = fd; done2: if (error != 0) { chgkqcnt(cred->cr_ruidinfo, -1, 0); crfree(cred); } return (error); } #ifndef _SYS_SYSPROTO_H_ struct kevent_args { int fd; const struct kevent *changelist; int nchanges; struct kevent *eventlist; int nevents; const struct timespec *timeout; }; #endif int sys_kevent(struct thread *td, struct kevent_args *uap) { struct timespec ts, *tsp; struct kevent_copyops k_ops = { uap, kevent_copyout, kevent_copyin}; int error; #ifdef KTRACE struct uio ktruio; struct iovec ktriov; struct uio *ktruioin = NULL; struct uio *ktruioout = NULL; #endif if (uap->timeout != NULL) { error = copyin(uap->timeout, &ts, sizeof(ts)); if (error) return (error); tsp = &ts; } else tsp = NULL; #ifdef KTRACE if (KTRPOINT(td, KTR_GENIO)) { ktriov.iov_base = uap->changelist; ktriov.iov_len = uap->nchanges * sizeof(struct kevent); ktruio = (struct uio){ .uio_iov = &ktriov, .uio_iovcnt = 1, .uio_segflg = UIO_USERSPACE, .uio_rw = UIO_READ, .uio_td = td }; ktruioin = cloneuio(&ktruio); ktriov.iov_base = uap->eventlist; ktriov.iov_len = uap->nevents * sizeof(struct kevent); ktruioout = cloneuio(&ktruio); } #endif error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents, &k_ops, tsp); #ifdef KTRACE if (ktruioin != NULL) { ktruioin->uio_resid = uap->nchanges * sizeof(struct kevent); ktrgenio(uap->fd, UIO_WRITE, ktruioin, 0); ktruioout->uio_resid = td->td_retval[0] * sizeof(struct kevent); ktrgenio(uap->fd, UIO_READ, ktruioout, error); } #endif return (error); } /* * Copy 'count' items into the destination list pointed to by uap->eventlist. */ static int kevent_copyout(void *arg, struct kevent *kevp, int count) { struct kevent_args *uap; int error; KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); uap = (struct kevent_args *)arg; error = copyout(kevp, uap->eventlist, count * sizeof *kevp); if (error == 0) uap->eventlist += count; return (error); } /* * Copy 'count' items from the list pointed to by uap->changelist. */ static int kevent_copyin(void *arg, struct kevent *kevp, int count) { struct kevent_args *uap; int error; KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); uap = (struct kevent_args *)arg; error = copyin(uap->changelist, kevp, count * sizeof *kevp); if (error == 0) uap->changelist += count; return (error); } int kern_kevent(struct thread *td, int fd, int nchanges, int nevents, struct kevent_copyops *k_ops, const struct timespec *timeout) { cap_rights_t rights; struct file *fp; int error; cap_rights_init(&rights); if (nchanges > 0) cap_rights_set(&rights, CAP_KQUEUE_CHANGE); if (nevents > 0) cap_rights_set(&rights, CAP_KQUEUE_EVENT); error = fget(td, fd, &rights, &fp); if (error != 0) return (error); error = kern_kevent_fp(td, fp, nchanges, nevents, k_ops, timeout); fdrop(fp, td); return (error); } int kern_kevent_fp(struct thread *td, struct file *fp, int nchanges, int nevents, struct kevent_copyops *k_ops, const struct timespec *timeout) { struct kevent keva[KQ_NEVENTS]; struct kevent *kevp, *changes; struct kqueue *kq; int i, n, nerrors, error; error = kqueue_acquire(fp, &kq); if (error != 0) return (error); nerrors = 0; while (nchanges > 0) { n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges; error = k_ops->k_copyin(k_ops->arg, keva, n); if (error) goto done; changes = keva; for (i = 0; i < n; i++) { kevp = &changes[i]; if (!kevp->filter) continue; kevp->flags &= ~EV_SYSFLAGS; error = kqueue_register(kq, kevp, td, 1); if (error || (kevp->flags & EV_RECEIPT)) { if (nevents != 0) { kevp->flags = EV_ERROR; kevp->data = error; (void) k_ops->k_copyout(k_ops->arg, kevp, 1); nevents--; nerrors++; } else { goto done; } } } nchanges -= n; } if (nerrors) { td->td_retval[0] = nerrors; error = 0; goto done; } error = kqueue_scan(kq, nevents, k_ops, timeout, keva, td); done: kqueue_release(kq, 0); return (error); } int kqueue_add_filteropts(int filt, struct filterops *filtops) { int error; error = 0; if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) { printf( "trying to add a filterop that is out of range: %d is beyond %d\n", ~filt, EVFILT_SYSCOUNT); return EINVAL; } mtx_lock(&filterops_lock); if (sysfilt_ops[~filt].for_fop != &null_filtops && sysfilt_ops[~filt].for_fop != NULL) error = EEXIST; else { sysfilt_ops[~filt].for_fop = filtops; sysfilt_ops[~filt].for_refcnt = 0; } mtx_unlock(&filterops_lock); return (error); } int kqueue_del_filteropts(int filt) { int error; error = 0; if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) return EINVAL; mtx_lock(&filterops_lock); if (sysfilt_ops[~filt].for_fop == &null_filtops || sysfilt_ops[~filt].for_fop == NULL) error = EINVAL; else if (sysfilt_ops[~filt].for_refcnt != 0) error = EBUSY; else { sysfilt_ops[~filt].for_fop = &null_filtops; sysfilt_ops[~filt].for_refcnt = 0; } mtx_unlock(&filterops_lock); return error; } static struct filterops * kqueue_fo_find(int filt) { if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) return NULL; if (sysfilt_ops[~filt].for_nolock) return sysfilt_ops[~filt].for_fop; mtx_lock(&filterops_lock); sysfilt_ops[~filt].for_refcnt++; if (sysfilt_ops[~filt].for_fop == NULL) sysfilt_ops[~filt].for_fop = &null_filtops; mtx_unlock(&filterops_lock); return sysfilt_ops[~filt].for_fop; } static void kqueue_fo_release(int filt) { if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) return; if (sysfilt_ops[~filt].for_nolock) return; mtx_lock(&filterops_lock); KASSERT(sysfilt_ops[~filt].for_refcnt > 0, ("filter object refcount not valid on release")); sysfilt_ops[~filt].for_refcnt--; mtx_unlock(&filterops_lock); } /* * A ref to kq (obtained via kqueue_acquire) must be held. waitok will * influence if memory allocation should wait. Make sure it is 0 if you * hold any mutexes. */ static int kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int waitok) { struct filterops *fops; struct file *fp; struct knote *kn, *tkn; cap_rights_t rights; int error, filt, event; int haskqglobal, filedesc_unlock; fp = NULL; kn = NULL; error = 0; haskqglobal = 0; filedesc_unlock = 0; filt = kev->filter; fops = kqueue_fo_find(filt); if (fops == NULL) return EINVAL; if (kev->flags & EV_ADD) tkn = knote_alloc(waitok); /* prevent waiting with locks */ else tkn = NULL; findkn: if (fops->f_isfd) { KASSERT(td != NULL, ("td is NULL")); error = fget(td, kev->ident, cap_rights_init(&rights, CAP_EVENT), &fp); if (error) goto done; if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops, kev->ident, 0) != 0) { /* try again */ fdrop(fp, td); fp = NULL; error = kqueue_expand(kq, fops, kev->ident, waitok); if (error) goto done; goto findkn; } if (fp->f_type == DTYPE_KQUEUE) { /* * if we add some inteligence about what we are doing, * we should be able to support events on ourselves. * We need to know when we are doing this to prevent * getting both the knlist lock and the kq lock since * they are the same thing. */ if (fp->f_data == kq) { error = EINVAL; goto done; } /* * Pre-lock the filedesc before the global * lock mutex, see the comment in * kqueue_close(). */ FILEDESC_XLOCK(td->td_proc->p_fd); filedesc_unlock = 1; KQ_GLOBAL_LOCK(&kq_global, haskqglobal); } KQ_LOCK(kq); if (kev->ident < kq->kq_knlistsize) { SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link) if (kev->filter == kn->kn_filter) break; } } else { if ((kev->flags & EV_ADD) == EV_ADD) kqueue_expand(kq, fops, kev->ident, waitok); KQ_LOCK(kq); if (kq->kq_knhashmask != 0) { struct klist *list; list = &kq->kq_knhash[ KN_HASH((u_long)kev->ident, kq->kq_knhashmask)]; SLIST_FOREACH(kn, list, kn_link) if (kev->ident == kn->kn_id && kev->filter == kn->kn_filter) break; } } /* knote is in the process of changing, wait for it to stablize. */ if (kn != NULL && (kn->kn_status & KN_INFLUX) == KN_INFLUX) { KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); if (filedesc_unlock) { FILEDESC_XUNLOCK(td->td_proc->p_fd); filedesc_unlock = 0; } kq->kq_state |= KQ_FLUXWAIT; msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0); if (fp != NULL) { fdrop(fp, td); fp = NULL; } goto findkn; } /* * kn now contains the matching knote, or NULL if no match */ if (kn == NULL) { if (kev->flags & EV_ADD) { kn = tkn; tkn = NULL; if (kn == NULL) { KQ_UNLOCK(kq); error = ENOMEM; goto done; } kn->kn_fp = fp; kn->kn_kq = kq; kn->kn_fop = fops; /* * apply reference counts to knote structure, and * do not release it at the end of this routine. */ fops = NULL; fp = NULL; kn->kn_sfflags = kev->fflags; kn->kn_sdata = kev->data; kev->fflags = 0; kev->data = 0; kn->kn_kevent = *kev; kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE | EV_ENABLE | EV_DISABLE | EV_FORCEONESHOT); kn->kn_status = KN_INFLUX|KN_DETACHED; error = knote_attach(kn, kq); KQ_UNLOCK(kq); if (error != 0) { tkn = kn; goto done; } if ((error = kn->kn_fop->f_attach(kn)) != 0) { knote_drop(kn, td); goto done; } KN_LIST_LOCK(kn); goto done_ev_add; } else { /* No matching knote and the EV_ADD flag is not set. */ KQ_UNLOCK(kq); error = ENOENT; goto done; } } if (kev->flags & EV_DELETE) { kn->kn_status |= KN_INFLUX; KQ_UNLOCK(kq); if (!(kn->kn_status & KN_DETACHED)) kn->kn_fop->f_detach(kn); knote_drop(kn, td); goto done; } if (kev->flags & EV_FORCEONESHOT) { kn->kn_flags |= EV_ONESHOT; KNOTE_ACTIVATE(kn, 1); } /* * The user may change some filter values after the initial EV_ADD, * but doing so will not reset any filter which has already been * triggered. */ kn->kn_status |= KN_INFLUX | KN_SCAN; KQ_UNLOCK(kq); KN_LIST_LOCK(kn); kn->kn_kevent.udata = kev->udata; if (!fops->f_isfd && fops->f_touch != NULL) { fops->f_touch(kn, kev, EVENT_REGISTER); } else { kn->kn_sfflags = kev->fflags; kn->kn_sdata = kev->data; } /* * We can get here with kn->kn_knlist == NULL. This can happen when * the initial attach event decides that the event is "completed" * already. i.e. filt_procattach is called on a zombie process. It * will call filt_proc which will remove it from the list, and NULL * kn_knlist. */ done_ev_add: if ((kev->flags & EV_DISABLE) && ((kn->kn_status & KN_DISABLED) == 0)) { kn->kn_status |= KN_DISABLED; } if ((kn->kn_status & KN_DISABLED) == 0) event = kn->kn_fop->f_event(kn, 0); else event = 0; KQ_LOCK(kq); if (event) KNOTE_ACTIVATE(kn, 1); kn->kn_status &= ~(KN_INFLUX | KN_SCAN); KN_LIST_UNLOCK(kn); if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) { kn->kn_status &= ~KN_DISABLED; if ((kn->kn_status & KN_ACTIVE) && ((kn->kn_status & KN_QUEUED) == 0)) knote_enqueue(kn); } KQ_UNLOCK_FLUX(kq); done: KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); if (filedesc_unlock) FILEDESC_XUNLOCK(td->td_proc->p_fd); if (fp != NULL) fdrop(fp, td); if (tkn != NULL) knote_free(tkn); if (fops != NULL) kqueue_fo_release(filt); return (error); } static int kqueue_acquire(struct file *fp, struct kqueue **kqp) { int error; struct kqueue *kq; error = 0; kq = fp->f_data; if (fp->f_type != DTYPE_KQUEUE || kq == NULL) return (EBADF); *kqp = kq; KQ_LOCK(kq); if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) { KQ_UNLOCK(kq); return (EBADF); } kq->kq_refcnt++; KQ_UNLOCK(kq); return error; } static void kqueue_release(struct kqueue *kq, int locked) { if (locked) KQ_OWNED(kq); else KQ_LOCK(kq); kq->kq_refcnt--; if (kq->kq_refcnt == 1) wakeup(&kq->kq_refcnt); if (!locked) KQ_UNLOCK(kq); } static void kqueue_schedtask(struct kqueue *kq) { KQ_OWNED(kq); KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN), ("scheduling kqueue task while draining")); if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) { taskqueue_enqueue(taskqueue_kqueue, &kq->kq_task); kq->kq_state |= KQ_TASKSCHED; } } /* * Expand the kq to make sure we have storage for fops/ident pair. * * Return 0 on success (or no work necessary), return errno on failure. * * Not calling hashinit w/ waitok (proper malloc flag) should be safe. * If kqueue_register is called from a non-fd context, there usually/should * be no locks held. */ static int kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident, int waitok) { struct klist *list, *tmp_knhash, *to_free; u_long tmp_knhashmask; int size; int fd; int mflag = waitok ? M_WAITOK : M_NOWAIT; KQ_NOTOWNED(kq); to_free = NULL; if (fops->f_isfd) { fd = ident; if (kq->kq_knlistsize <= fd) { size = kq->kq_knlistsize; while (size <= fd) size += KQEXTENT; list = malloc(size * sizeof(*list), M_KQUEUE, mflag); if (list == NULL) return ENOMEM; KQ_LOCK(kq); if (kq->kq_knlistsize > fd) { to_free = list; list = NULL; } else { if (kq->kq_knlist != NULL) { bcopy(kq->kq_knlist, list, kq->kq_knlistsize * sizeof(*list)); to_free = kq->kq_knlist; kq->kq_knlist = NULL; } bzero((caddr_t)list + kq->kq_knlistsize * sizeof(*list), (size - kq->kq_knlistsize) * sizeof(*list)); kq->kq_knlistsize = size; kq->kq_knlist = list; } KQ_UNLOCK(kq); } } else { if (kq->kq_knhashmask == 0) { tmp_knhash = hashinit(KN_HASHSIZE, M_KQUEUE, &tmp_knhashmask); if (tmp_knhash == NULL) return ENOMEM; KQ_LOCK(kq); if (kq->kq_knhashmask == 0) { kq->kq_knhash = tmp_knhash; kq->kq_knhashmask = tmp_knhashmask; } else { to_free = tmp_knhash; } KQ_UNLOCK(kq); } } free(to_free, M_KQUEUE); KQ_NOTOWNED(kq); return 0; } static void kqueue_task(void *arg, int pending) { struct kqueue *kq; int haskqglobal; haskqglobal = 0; kq = arg; KQ_GLOBAL_LOCK(&kq_global, haskqglobal); KQ_LOCK(kq); KNOTE_LOCKED(&kq->kq_sel.si_note, 0); kq->kq_state &= ~KQ_TASKSCHED; if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) { wakeup(&kq->kq_state); } KQ_UNLOCK(kq); KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); } /* * Scan, update kn_data (if not ONESHOT), and copyout triggered events. * We treat KN_MARKER knotes as if they are INFLUX. */ static int kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops, const struct timespec *tsp, struct kevent *keva, struct thread *td) { struct kevent *kevp; struct knote *kn, *marker; sbintime_t asbt, rsbt; int count, error, haskqglobal, influx, nkev, touch; count = maxevents; nkev = 0; error = 0; haskqglobal = 0; if (maxevents == 0) goto done_nl; rsbt = 0; if (tsp != NULL) { if (tsp->tv_sec < 0 || tsp->tv_nsec < 0 || tsp->tv_nsec >= 1000000000) { error = EINVAL; goto done_nl; } if (timespecisset(tsp)) { if (tsp->tv_sec <= INT32_MAX) { rsbt = tstosbt(*tsp); if (TIMESEL(&asbt, rsbt)) asbt += tc_tick_sbt; if (asbt <= SBT_MAX - rsbt) asbt += rsbt; else asbt = 0; rsbt >>= tc_precexp; } else asbt = 0; } else asbt = -1; } else asbt = 0; marker = knote_alloc(1); if (marker == NULL) { error = ENOMEM; goto done_nl; } marker->kn_status = KN_MARKER; KQ_LOCK(kq); retry: kevp = keva; if (kq->kq_count == 0) { if (asbt == -1) { error = EWOULDBLOCK; } else { kq->kq_state |= KQ_SLEEP; error = msleep_sbt(kq, &kq->kq_lock, PSOCK | PCATCH, "kqread", asbt, rsbt, C_ABSOLUTE); } if (error == 0) goto retry; /* don't restart after signals... */ if (error == ERESTART) error = EINTR; else if (error == EWOULDBLOCK) error = 0; goto done; } TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe); influx = 0; while (count) { KQ_OWNED(kq); kn = TAILQ_FIRST(&kq->kq_head); if ((kn->kn_status == KN_MARKER && kn != marker) || (kn->kn_status & KN_INFLUX) == KN_INFLUX) { if (influx) { influx = 0; KQ_FLUX_WAKEUP(kq); } kq->kq_state |= KQ_FLUXWAIT; error = msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0); continue; } TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) { kn->kn_status &= ~KN_QUEUED; kq->kq_count--; continue; } if (kn == marker) { KQ_FLUX_WAKEUP(kq); if (count == maxevents) goto retry; goto done; } KASSERT((kn->kn_status & KN_INFLUX) == 0, ("KN_INFLUX set when not suppose to be")); if ((kn->kn_flags & EV_DROP) == EV_DROP) { kn->kn_status &= ~KN_QUEUED; kn->kn_status |= KN_INFLUX; kq->kq_count--; KQ_UNLOCK(kq); /* * We don't need to lock the list since we've marked * it _INFLUX. */ if (!(kn->kn_status & KN_DETACHED)) kn->kn_fop->f_detach(kn); knote_drop(kn, td); KQ_LOCK(kq); continue; } else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) { kn->kn_status &= ~KN_QUEUED; kn->kn_status |= KN_INFLUX; kq->kq_count--; KQ_UNLOCK(kq); /* * We don't need to lock the list since we've marked * it _INFLUX. */ *kevp = kn->kn_kevent; if (!(kn->kn_status & KN_DETACHED)) kn->kn_fop->f_detach(kn); knote_drop(kn, td); KQ_LOCK(kq); kn = NULL; } else { kn->kn_status |= KN_INFLUX | KN_SCAN; KQ_UNLOCK(kq); if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE) KQ_GLOBAL_LOCK(&kq_global, haskqglobal); KN_LIST_LOCK(kn); if (kn->kn_fop->f_event(kn, 0) == 0) { KQ_LOCK(kq); KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE | KN_INFLUX | KN_SCAN); kq->kq_count--; KN_LIST_UNLOCK(kn); influx = 1; continue; } touch = (!kn->kn_fop->f_isfd && kn->kn_fop->f_touch != NULL); if (touch) kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS); else *kevp = kn->kn_kevent; KQ_LOCK(kq); KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) { /* * Manually clear knotes who weren't * 'touch'ed. */ if (touch == 0 && kn->kn_flags & EV_CLEAR) { kn->kn_data = 0; kn->kn_fflags = 0; } if (kn->kn_flags & EV_DISPATCH) kn->kn_status |= KN_DISABLED; kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); kq->kq_count--; } else TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); kn->kn_status &= ~(KN_INFLUX | KN_SCAN); KN_LIST_UNLOCK(kn); influx = 1; } /* we are returning a copy to the user */ kevp++; nkev++; count--; if (nkev == KQ_NEVENTS) { influx = 0; KQ_UNLOCK_FLUX(kq); error = k_ops->k_copyout(k_ops->arg, keva, nkev); nkev = 0; kevp = keva; KQ_LOCK(kq); if (error) break; } } TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe); done: KQ_OWNED(kq); KQ_UNLOCK_FLUX(kq); knote_free(marker); done_nl: KQ_NOTOWNED(kq); if (nkev != 0) error = k_ops->k_copyout(k_ops->arg, keva, nkev); td->td_retval[0] = maxevents - count; return (error); } /*ARGSUSED*/ static int kqueue_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *active_cred, struct thread *td) { /* * Enabling sigio causes two major problems: * 1) infinite recursion: * Synopsys: kevent is being used to track signals and have FIOASYNC * set. On receipt of a signal this will cause a kqueue to recurse * into itself over and over. Sending the sigio causes the kqueue * to become ready, which in turn posts sigio again, forever. * Solution: this can be solved by setting a flag in the kqueue that * we have a SIGIO in progress. * 2) locking problems: * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts * us above the proc and pgrp locks. * Solution: Post a signal using an async mechanism, being sure to * record a generation count in the delivery so that we do not deliver * a signal to the wrong process. * * Note, these two mechanisms are somewhat mutually exclusive! */ #if 0 struct kqueue *kq; kq = fp->f_data; switch (cmd) { case FIOASYNC: if (*(int *)data) { kq->kq_state |= KQ_ASYNC; } else { kq->kq_state &= ~KQ_ASYNC; } return (0); case FIOSETOWN: return (fsetown(*(int *)data, &kq->kq_sigio)); case FIOGETOWN: *(int *)data = fgetown(&kq->kq_sigio); return (0); } #endif return (ENOTTY); } /*ARGSUSED*/ static int kqueue_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td) { struct kqueue *kq; int revents = 0; int error; if ((error = kqueue_acquire(fp, &kq))) return POLLERR; KQ_LOCK(kq); if (events & (POLLIN | POLLRDNORM)) { if (kq->kq_count) { revents |= events & (POLLIN | POLLRDNORM); } else { selrecord(td, &kq->kq_sel); if (SEL_WAITING(&kq->kq_sel)) kq->kq_state |= KQ_SEL; } } kqueue_release(kq, 1); KQ_UNLOCK(kq); return (revents); } /*ARGSUSED*/ static int kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred, struct thread *td) { bzero((void *)st, sizeof *st); /* * We no longer return kq_count because the unlocked value is useless. * If you spent all this time getting the count, why not spend your * syscall better by calling kevent? * * XXX - This is needed for libc_r. */ st->st_mode = S_IFIFO; return (0); } /*ARGSUSED*/ static int kqueue_close(struct file *fp, struct thread *td) { struct kqueue *kq = fp->f_data; struct filedesc *fdp; struct knote *kn; int i; int error; int filedesc_unlock; if ((error = kqueue_acquire(fp, &kq))) return error; filedesc_unlock = 0; KQ_LOCK(kq); KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING, ("kqueue already closing")); kq->kq_state |= KQ_CLOSING; if (kq->kq_refcnt > 1) msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0); KASSERT(kq->kq_refcnt == 1, ("other refs are out there!")); fdp = kq->kq_fdp; KASSERT(knlist_empty(&kq->kq_sel.si_note), ("kqueue's knlist not empty")); for (i = 0; i < kq->kq_knlistsize; i++) { while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) { if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) { kq->kq_state |= KQ_FLUXWAIT; msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0); continue; } kn->kn_status |= KN_INFLUX; KQ_UNLOCK(kq); if (!(kn->kn_status & KN_DETACHED)) kn->kn_fop->f_detach(kn); knote_drop(kn, td); KQ_LOCK(kq); } } if (kq->kq_knhashmask != 0) { for (i = 0; i <= kq->kq_knhashmask; i++) { while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) { if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) { kq->kq_state |= KQ_FLUXWAIT; msleep(kq, &kq->kq_lock, PSOCK, "kqclo2", 0); continue; } kn->kn_status |= KN_INFLUX; KQ_UNLOCK(kq); if (!(kn->kn_status & KN_DETACHED)) kn->kn_fop->f_detach(kn); knote_drop(kn, td); KQ_LOCK(kq); } } } if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) { kq->kq_state |= KQ_TASKDRAIN; msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0); } if ((kq->kq_state & KQ_SEL) == KQ_SEL) { selwakeuppri(&kq->kq_sel, PSOCK); if (!SEL_WAITING(&kq->kq_sel)) kq->kq_state &= ~KQ_SEL; } KQ_UNLOCK(kq); /* * We could be called due to the knote_drop() doing fdrop(), * called from kqueue_register(). In this case the global * lock is owned, and filedesc sx is locked before, to not * take the sleepable lock after non-sleepable. */ if (!sx_xlocked(FILEDESC_LOCK(fdp))) { FILEDESC_XLOCK(fdp); filedesc_unlock = 1; } else filedesc_unlock = 0; TAILQ_REMOVE(&fdp->fd_kqlist, kq, kq_list); if (filedesc_unlock) FILEDESC_XUNLOCK(fdp); seldrain(&kq->kq_sel); knlist_destroy(&kq->kq_sel.si_note); mtx_destroy(&kq->kq_lock); kq->kq_fdp = NULL; if (kq->kq_knhash != NULL) free(kq->kq_knhash, M_KQUEUE); if (kq->kq_knlist != NULL) free(kq->kq_knlist, M_KQUEUE); funsetown(&kq->kq_sigio); chgkqcnt(kq->kq_cred->cr_ruidinfo, -1, 0); crfree(kq->kq_cred); free(kq, M_KQUEUE); fp->f_data = NULL; return (0); } static int kqueue_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) { kif->kf_type = KF_TYPE_KQUEUE; return (0); } static void kqueue_wakeup(struct kqueue *kq) { KQ_OWNED(kq); if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) { kq->kq_state &= ~KQ_SLEEP; wakeup(kq); } if ((kq->kq_state & KQ_SEL) == KQ_SEL) { selwakeuppri(&kq->kq_sel, PSOCK); if (!SEL_WAITING(&kq->kq_sel)) kq->kq_state &= ~KQ_SEL; } if (!knlist_empty(&kq->kq_sel.si_note)) kqueue_schedtask(kq); if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) { pgsigio(&kq->kq_sigio, SIGIO, 0); } } /* * Walk down a list of knotes, activating them if their event has triggered. * * There is a possibility to optimize in the case of one kq watching another. * Instead of scheduling a task to wake it up, you could pass enough state * down the chain to make up the parent kqueue. Make this code functional * first. */ void knote(struct knlist *list, long hint, int lockflags) { struct kqueue *kq; struct knote *kn; int error; if (list == NULL) return; KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED); if ((lockflags & KNF_LISTLOCKED) == 0) list->kl_lock(list->kl_lockarg); /* * If we unlock the list lock (and set KN_INFLUX), we can eliminate * the kqueue scheduling, but this will introduce four * lock/unlock's for each knote to test. If we do, continue to use * SLIST_FOREACH, SLIST_FOREACH_SAFE is not safe in our case, it is * only safe if you want to remove the current item, which we are * not doing. */ SLIST_FOREACH(kn, &list->kl_list, kn_selnext) { kq = kn->kn_kq; KQ_LOCK(kq); if ((kn->kn_status & (KN_INFLUX | KN_SCAN)) == KN_INFLUX) { /* * Do not process the influx notes, except for * the influx coming from the kq unlock in the * kqueue_scan(). In the later case, we do * not interfere with the scan, since the code * fragment in kqueue_scan() locks the knlist, * and cannot proceed until we finished. */ KQ_UNLOCK(kq); } else if ((lockflags & KNF_NOKQLOCK) != 0) { kn->kn_status |= KN_INFLUX; KQ_UNLOCK(kq); error = kn->kn_fop->f_event(kn, hint); KQ_LOCK(kq); kn->kn_status &= ~KN_INFLUX; if (error) KNOTE_ACTIVATE(kn, 1); KQ_UNLOCK_FLUX(kq); } else { kn->kn_status |= KN_HASKQLOCK; if (kn->kn_fop->f_event(kn, hint)) KNOTE_ACTIVATE(kn, 1); kn->kn_status &= ~KN_HASKQLOCK; KQ_UNLOCK(kq); } } if ((lockflags & KNF_LISTLOCKED) == 0) list->kl_unlock(list->kl_lockarg); } /* * add a knote to a knlist */ void knlist_add(struct knlist *knl, struct knote *kn, int islocked) { KNL_ASSERT_LOCK(knl, islocked); KQ_NOTOWNED(kn->kn_kq); KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == (KN_INFLUX|KN_DETACHED), ("knote not KN_INFLUX and KN_DETACHED")); if (!islocked) knl->kl_lock(knl->kl_lockarg); SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext); if (!islocked) knl->kl_unlock(knl->kl_lockarg); KQ_LOCK(kn->kn_kq); kn->kn_knlist = knl; kn->kn_status &= ~KN_DETACHED; KQ_UNLOCK(kn->kn_kq); } static void knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked, int kqislocked) { KASSERT(!(!!kqislocked && !knlislocked), ("kq locked w/o knl locked")); KNL_ASSERT_LOCK(knl, knlislocked); mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED); if (!kqislocked) KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == KN_INFLUX, ("knlist_remove called w/o knote being KN_INFLUX or already removed")); if (!knlislocked) knl->kl_lock(knl->kl_lockarg); SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext); kn->kn_knlist = NULL; if (!knlislocked) knl->kl_unlock(knl->kl_lockarg); if (!kqislocked) KQ_LOCK(kn->kn_kq); kn->kn_status |= KN_DETACHED; if (!kqislocked) KQ_UNLOCK(kn->kn_kq); } /* * remove knote from the specified knlist */ void knlist_remove(struct knlist *knl, struct knote *kn, int islocked) { knlist_remove_kq(knl, kn, islocked, 0); } /* * remove knote from the specified knlist while in f_event handler. */ void knlist_remove_inevent(struct knlist *knl, struct knote *kn) { knlist_remove_kq(knl, kn, 1, (kn->kn_status & KN_HASKQLOCK) == KN_HASKQLOCK); } int knlist_empty(struct knlist *knl) { KNL_ASSERT_LOCKED(knl); return SLIST_EMPTY(&knl->kl_list); } static struct mtx knlist_lock; MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects", MTX_DEF); static void knlist_mtx_lock(void *arg); static void knlist_mtx_unlock(void *arg); static void knlist_mtx_lock(void *arg) { mtx_lock((struct mtx *)arg); } static void knlist_mtx_unlock(void *arg) { mtx_unlock((struct mtx *)arg); } static void knlist_mtx_assert_locked(void *arg) { mtx_assert((struct mtx *)arg, MA_OWNED); } static void knlist_mtx_assert_unlocked(void *arg) { mtx_assert((struct mtx *)arg, MA_NOTOWNED); } static void knlist_rw_rlock(void *arg) { rw_rlock((struct rwlock *)arg); } static void knlist_rw_runlock(void *arg) { rw_runlock((struct rwlock *)arg); } static void knlist_rw_assert_locked(void *arg) { rw_assert((struct rwlock *)arg, RA_LOCKED); } static void knlist_rw_assert_unlocked(void *arg) { rw_assert((struct rwlock *)arg, RA_UNLOCKED); } void knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *), void (*kl_unlock)(void *), void (*kl_assert_locked)(void *), void (*kl_assert_unlocked)(void *)) { if (lock == NULL) knl->kl_lockarg = &knlist_lock; else knl->kl_lockarg = lock; if (kl_lock == NULL) knl->kl_lock = knlist_mtx_lock; else knl->kl_lock = kl_lock; if (kl_unlock == NULL) knl->kl_unlock = knlist_mtx_unlock; else knl->kl_unlock = kl_unlock; if (kl_assert_locked == NULL) knl->kl_assert_locked = knlist_mtx_assert_locked; else knl->kl_assert_locked = kl_assert_locked; if (kl_assert_unlocked == NULL) knl->kl_assert_unlocked = knlist_mtx_assert_unlocked; else knl->kl_assert_unlocked = kl_assert_unlocked; SLIST_INIT(&knl->kl_list); } void knlist_init_mtx(struct knlist *knl, struct mtx *lock) { knlist_init(knl, lock, NULL, NULL, NULL, NULL); } void knlist_init_rw_reader(struct knlist *knl, struct rwlock *lock) { knlist_init(knl, lock, knlist_rw_rlock, knlist_rw_runlock, knlist_rw_assert_locked, knlist_rw_assert_unlocked); } void knlist_destroy(struct knlist *knl) { #ifdef INVARIANTS /* * if we run across this error, we need to find the offending * driver and have it call knlist_clear or knlist_delete. */ if (!SLIST_EMPTY(&knl->kl_list)) printf("WARNING: destroying knlist w/ knotes on it!\n"); #endif knl->kl_lockarg = knl->kl_lock = knl->kl_unlock = NULL; SLIST_INIT(&knl->kl_list); } /* * Even if we are locked, we may need to drop the lock to allow any influx * knotes time to "settle". */ void knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn) { struct knote *kn, *kn2; struct kqueue *kq; if (islocked) KNL_ASSERT_LOCKED(knl); else { KNL_ASSERT_UNLOCKED(knl); again: /* need to reacquire lock since we have dropped it */ knl->kl_lock(knl->kl_lockarg); } SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) { kq = kn->kn_kq; KQ_LOCK(kq); if ((kn->kn_status & KN_INFLUX)) { KQ_UNLOCK(kq); continue; } knlist_remove_kq(knl, kn, 1, 1); if (killkn) { kn->kn_status |= KN_INFLUX | KN_DETACHED; KQ_UNLOCK(kq); knote_drop(kn, td); } else { /* Make sure cleared knotes disappear soon */ kn->kn_flags |= (EV_EOF | EV_ONESHOT); KQ_UNLOCK(kq); } kq = NULL; } if (!SLIST_EMPTY(&knl->kl_list)) { /* there are still KN_INFLUX remaining */ kn = SLIST_FIRST(&knl->kl_list); kq = kn->kn_kq; KQ_LOCK(kq); KASSERT(kn->kn_status & KN_INFLUX, ("knote removed w/o list lock")); knl->kl_unlock(knl->kl_lockarg); kq->kq_state |= KQ_FLUXWAIT; msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0); kq = NULL; goto again; } if (islocked) KNL_ASSERT_LOCKED(knl); else { knl->kl_unlock(knl->kl_lockarg); KNL_ASSERT_UNLOCKED(knl); } } /* * Remove all knotes referencing a specified fd must be called with FILEDESC * lock. This prevents a race where a new fd comes along and occupies the * entry and we attach a knote to the fd. */ void knote_fdclose(struct thread *td, int fd) { struct filedesc *fdp = td->td_proc->p_fd; struct kqueue *kq; struct knote *kn; int influx; FILEDESC_XLOCK_ASSERT(fdp); /* * We shouldn't have to worry about new kevents appearing on fd * since filedesc is locked. */ TAILQ_FOREACH(kq, &fdp->fd_kqlist, kq_list) { KQ_LOCK(kq); again: influx = 0; while (kq->kq_knlistsize > fd && (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) { if (kn->kn_status & KN_INFLUX) { /* someone else might be waiting on our knote */ if (influx) wakeup(kq); kq->kq_state |= KQ_FLUXWAIT; msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0); goto again; } kn->kn_status |= KN_INFLUX; KQ_UNLOCK(kq); if (!(kn->kn_status & KN_DETACHED)) kn->kn_fop->f_detach(kn); knote_drop(kn, td); influx = 1; KQ_LOCK(kq); } KQ_UNLOCK_FLUX(kq); } } static int knote_attach(struct knote *kn, struct kqueue *kq) { struct klist *list; KASSERT(kn->kn_status & KN_INFLUX, ("knote not marked INFLUX")); KQ_OWNED(kq); if (kn->kn_fop->f_isfd) { if (kn->kn_id >= kq->kq_knlistsize) return ENOMEM; list = &kq->kq_knlist[kn->kn_id]; } else { if (kq->kq_knhash == NULL) return ENOMEM; list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; } SLIST_INSERT_HEAD(list, kn, kn_link); return 0; } /* * knote must already have been detached using the f_detach method. * no lock need to be held, it is assumed that the KN_INFLUX flag is set * to prevent other removal. */ static void knote_drop(struct knote *kn, struct thread *td) { struct kqueue *kq; struct klist *list; kq = kn->kn_kq; KQ_NOTOWNED(kq); KASSERT((kn->kn_status & KN_INFLUX) == KN_INFLUX, ("knote_drop called without KN_INFLUX set in kn_status")); KQ_LOCK(kq); if (kn->kn_fop->f_isfd) list = &kq->kq_knlist[kn->kn_id]; else list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; if (!SLIST_EMPTY(list)) SLIST_REMOVE(list, kn, knote, kn_link); if (kn->kn_status & KN_QUEUED) knote_dequeue(kn); KQ_UNLOCK_FLUX(kq); if (kn->kn_fop->f_isfd) { fdrop(kn->kn_fp, td); kn->kn_fp = NULL; } kqueue_fo_release(kn->kn_kevent.filter); kn->kn_fop = NULL; knote_free(kn); } static void knote_enqueue(struct knote *kn) { struct kqueue *kq = kn->kn_kq; KQ_OWNED(kn->kn_kq); KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); kn->kn_status |= KN_QUEUED; kq->kq_count++; kqueue_wakeup(kq); } static void knote_dequeue(struct knote *kn) { struct kqueue *kq = kn->kn_kq; KQ_OWNED(kn->kn_kq); KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); kn->kn_status &= ~KN_QUEUED; kq->kq_count--; } static void knote_init(void) { knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); } SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL); static struct knote * knote_alloc(int waitok) { return ((struct knote *)uma_zalloc(knote_zone, (waitok ? M_WAITOK : M_NOWAIT)|M_ZERO)); } static void knote_free(struct knote *kn) { if (kn != NULL) uma_zfree(knote_zone, kn); } /* * Register the kev w/ the kq specified by fd. */ int kqfd_register(int fd, struct kevent *kev, struct thread *td, int waitok) { struct kqueue *kq; struct file *fp; cap_rights_t rights; int error; error = fget(td, fd, cap_rights_init(&rights, CAP_KQUEUE_CHANGE), &fp); if (error != 0) return (error); if ((error = kqueue_acquire(fp, &kq)) != 0) goto noacquire; error = kqueue_register(kq, kev, td, waitok); kqueue_release(kq, 0); noacquire: fdrop(fp, td); return error; } Index: head/sys/kern/kern_exec.c =================================================================== --- head/sys/kern/kern_exec.c (revision 284214) +++ head/sys/kern/kern_exec.c (revision 284215) @@ -1,1522 +1,1522 @@ /*- * Copyright (c) 1993, David Greenman * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_capsicum.h" #include "opt_hwpmc_hooks.h" #include "opt_ktrace.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 #include #ifdef KTRACE #include #endif #include #include #include #include #include #include #include #include #include #ifdef HWPMC_HOOKS #include #endif #include #include #include #ifdef KDTRACE_HOOKS #include dtrace_execexit_func_t dtrace_fasttrap_exec; #endif SDT_PROVIDER_DECLARE(proc); SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *"); SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int"); SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "char *"); MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS); static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS); static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS); static int do_execve(struct thread *td, struct image_args *args, struct mac *mac_p); /* XXX This should be vm_size_t. */ SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD, NULL, 0, sysctl_kern_ps_strings, "LU", ""); /* XXX This should be vm_size_t. */ SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD| CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", ""); SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD, NULL, 0, sysctl_kern_stackprot, "I", ""); u_long ps_arg_cache_limit = PAGE_SIZE / 16; SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, &ps_arg_cache_limit, 0, ""); static int disallow_high_osrel; SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW, &disallow_high_osrel, 0, "Disallow execution of binaries built for higher version of the world"); static int map_at_zero = 0; SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0, "Permit processes to map an object at virtual address 0."); static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS) { struct proc *p; int error; p = curproc; #ifdef SCTL_MASK32 if (req->flags & SCTL_MASK32) { unsigned int val; val = (unsigned int)p->p_sysent->sv_psstrings; error = SYSCTL_OUT(req, &val, sizeof(val)); } else #endif error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings, sizeof(p->p_sysent->sv_psstrings)); return error; } static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS) { struct proc *p; int error; p = curproc; #ifdef SCTL_MASK32 if (req->flags & SCTL_MASK32) { unsigned int val; val = (unsigned int)p->p_sysent->sv_usrstack; error = SYSCTL_OUT(req, &val, sizeof(val)); } else #endif error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack, sizeof(p->p_sysent->sv_usrstack)); return error; } static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS) { struct proc *p; p = curproc; return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot, sizeof(p->p_sysent->sv_stackprot))); } /* * Each of the items is a pointer to a `const struct execsw', hence the * double pointer here. */ static const struct execsw **execsw; #ifndef _SYS_SYSPROTO_H_ struct execve_args { char *fname; char **argv; char **envv; }; #endif int sys_execve(struct thread *td, struct execve_args *uap) { struct image_args args; struct vmspace *oldvmspace; int error; error = pre_execve(td, &oldvmspace); if (error != 0) return (error); error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, uap->argv, uap->envv); if (error == 0) error = kern_execve(td, &args, NULL); post_execve(td, error, oldvmspace); return (error); } #ifndef _SYS_SYSPROTO_H_ struct fexecve_args { int fd; char **argv; char **envv; } #endif int sys_fexecve(struct thread *td, struct fexecve_args *uap) { struct image_args args; struct vmspace *oldvmspace; int error; error = pre_execve(td, &oldvmspace); if (error != 0) return (error); error = exec_copyin_args(&args, NULL, UIO_SYSSPACE, uap->argv, uap->envv); if (error == 0) { args.fd = uap->fd; error = kern_execve(td, &args, NULL); } post_execve(td, error, oldvmspace); return (error); } #ifndef _SYS_SYSPROTO_H_ struct __mac_execve_args { char *fname; char **argv; char **envv; struct mac *mac_p; }; #endif int sys___mac_execve(struct thread *td, struct __mac_execve_args *uap) { #ifdef MAC struct image_args args; struct vmspace *oldvmspace; int error; error = pre_execve(td, &oldvmspace); if (error != 0) return (error); error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, uap->argv, uap->envv); if (error == 0) error = kern_execve(td, &args, uap->mac_p); post_execve(td, error, oldvmspace); return (error); #else return (ENOSYS); #endif } int pre_execve(struct thread *td, struct vmspace **oldvmspace) { struct proc *p; int error; KASSERT(td == curthread, ("non-current thread %p", td)); error = 0; p = td->td_proc; if ((p->p_flag & P_HADTHREADS) != 0) { PROC_LOCK(p); if (thread_single(p, SINGLE_BOUNDARY) != 0) error = ERESTART; PROC_UNLOCK(p); } KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0, ("nested execve")); *oldvmspace = p->p_vmspace; return (error); } void post_execve(struct thread *td, int error, struct vmspace *oldvmspace) { struct proc *p; KASSERT(td == curthread, ("non-current thread %p", td)); p = td->td_proc; if ((p->p_flag & P_HADTHREADS) != 0) { PROC_LOCK(p); /* * If success, we upgrade to SINGLE_EXIT state to * force other threads to suicide. */ if (error == 0) thread_single(p, SINGLE_EXIT); else thread_single_end(p, SINGLE_BOUNDARY); PROC_UNLOCK(p); } if ((td->td_pflags & TDP_EXECVMSPC) != 0) { KASSERT(p->p_vmspace != oldvmspace, ("oldvmspace still used")); vmspace_free(oldvmspace); td->td_pflags &= ~TDP_EXECVMSPC; } } /* * XXX: kern_execve has the astonishing property of not always returning to * the caller. If sufficiently bad things happen during the call to * do_execve(), it can end up calling exit1(); as a result, callers must * avoid doing anything which they might need to undo (e.g., allocating * memory). */ int kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p) { AUDIT_ARG_ARGV(args->begin_argv, args->argc, args->begin_envv - args->begin_argv); AUDIT_ARG_ENVV(args->begin_envv, args->envc, args->endp - args->begin_envv); return (do_execve(td, args, mac_p)); } /* * In-kernel implementation of execve(). All arguments are assumed to be * userspace pointers from the passed thread. */ static int do_execve(td, args, mac_p) struct thread *td; struct image_args *args; struct mac *mac_p; { struct proc *p = td->td_proc; struct nameidata nd; struct ucred *newcred = NULL, *oldcred; struct uidinfo *euip = NULL; register_t *stack_base; int error, i; struct image_params image_params, *imgp; struct vattr attr; int (*img_first)(struct image_params *); struct pargs *oldargs = NULL, *newargs = NULL; struct sigacts *oldsigacts, *newsigacts; #ifdef KTRACE struct vnode *tracevp = NULL; struct ucred *tracecred = NULL; #endif struct vnode *textvp = NULL, *binvp; cap_rights_t rights; int credential_changing; int textset; #ifdef MAC struct label *interpvplabel = NULL; int will_transition; #endif #ifdef HWPMC_HOOKS struct pmckern_procexec pe; #endif static const char fexecv_proc_title[] = "(fexecv)"; imgp = &image_params; /* * Lock the process and set the P_INEXEC flag to indicate that * it should be left alone until we're done here. This is * necessary to avoid race conditions - e.g. in ptrace() - * that might allow a local user to illicitly obtain elevated * privileges. */ PROC_LOCK(p); KASSERT((p->p_flag & P_INEXEC) == 0, ("%s(): process already has P_INEXEC flag", __func__)); p->p_flag |= P_INEXEC; PROC_UNLOCK(p); /* * Initialize part of the common data */ bzero(imgp, sizeof(*imgp)); imgp->proc = p; imgp->attr = &attr; imgp->args = args; #ifdef MAC error = mac_execve_enter(imgp, mac_p); if (error) goto exec_fail; #endif /* * Translate the file name. namei() returns a vnode pointer * in ni_vp amoung other things. * * XXXAUDIT: It would be desirable to also audit the name of the * interpreter if this is an interpreted binary. */ if (args->fname != NULL) { NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td); } SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 ); interpret: if (args->fname != NULL) { #ifdef CAPABILITY_MODE /* * While capability mode can't reach this point via direct * path arguments to execve(), we also don't allow * interpreters to be used in capability mode (for now). * Catch indirect lookups and return a permissions error. */ if (IN_CAPABILITY_MODE(td)) { error = ECAPMODE; goto exec_fail; } #endif error = namei(&nd); if (error) goto exec_fail; binvp = nd.ni_vp; imgp->vp = binvp; } else { AUDIT_ARG_FD(args->fd); /* * Descriptors opened only with O_EXEC or O_RDONLY are allowed. */ error = fgetvp_exec(td, args->fd, cap_rights_init(&rights, CAP_FEXECVE), &binvp); if (error) goto exec_fail; vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY); AUDIT_ARG_VNODE1(binvp); imgp->vp = binvp; } /* * Check file permissions (also 'opens' file) */ error = exec_check_permissions(imgp); if (error) goto exec_fail_dealloc; imgp->object = imgp->vp->v_object; if (imgp->object != NULL) vm_object_reference(imgp->object); /* * Set VV_TEXT now so no one can write to the executable while we're * activating it. * * Remember if this was set before and unset it in case this is not * actually an executable image. */ textset = VOP_IS_TEXT(imgp->vp); VOP_SET_TEXT(imgp->vp); error = exec_map_first_page(imgp); if (error) goto exec_fail_dealloc; imgp->proc->p_osrel = 0; /* * If the current process has a special image activator it * wants to try first, call it. For example, emulating shell * scripts differently. */ error = -1; if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) error = img_first(imgp); /* * Loop through the list of image activators, calling each one. * An activator returns -1 if there is no match, 0 on success, * and an error otherwise. */ for (i = 0; error == -1 && execsw[i]; ++i) { if (execsw[i]->ex_imgact == NULL || execsw[i]->ex_imgact == img_first) { continue; } error = (*execsw[i]->ex_imgact)(imgp); } if (error) { if (error == -1) { if (textset == 0) VOP_UNSET_TEXT(imgp->vp); error = ENOEXEC; } goto exec_fail_dealloc; } /* * Special interpreter operation, cleanup and loop up to try to * activate the interpreter. */ if (imgp->interpreted) { exec_unmap_first_page(imgp); /* * VV_TEXT needs to be unset for scripts. There is a short * period before we determine that something is a script where * VV_TEXT will be set. The vnode lock is held over this * entire period so nothing should illegitimately be blocked. */ VOP_UNSET_TEXT(imgp->vp); /* free name buffer and old vnode */ if (args->fname != NULL) NDFREE(&nd, NDF_ONLY_PNBUF); #ifdef MAC mac_execve_interpreter_enter(binvp, &interpvplabel); #endif if (imgp->opened) { VOP_CLOSE(binvp, FREAD, td->td_ucred, td); imgp->opened = 0; } vput(binvp); vm_object_deallocate(imgp->object); imgp->object = NULL; /* set new name to that of the interpreter */ NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME, UIO_SYSSPACE, imgp->interpreter_name, td); args->fname = imgp->interpreter_name; goto interpret; } /* * NB: We unlock the vnode here because it is believed that none * of the sv_copyout_strings/sv_fixup operations require the vnode. */ VOP_UNLOCK(imgp->vp, 0); /* * Do the best to calculate the full path to the image file. */ if (imgp->auxargs != NULL && ((args->fname != NULL && args->fname[0] == '/') || vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0)) imgp->execpath = args->fname; if (disallow_high_osrel && P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) { error = ENOEXEC; uprintf("Osrel %d for image %s too high\n", p->p_osrel, imgp->execpath != NULL ? imgp->execpath : ""); vn_lock(imgp->vp, LK_SHARED | LK_RETRY); goto exec_fail_dealloc; } /* * Copy out strings (args and env) and initialize stack base */ if (p->p_sysent->sv_copyout_strings) stack_base = (*p->p_sysent->sv_copyout_strings)(imgp); else stack_base = exec_copyout_strings(imgp); /* * If custom stack fixup routine present for this process * let it do the stack setup. * Else stuff argument count as first item on stack */ if (p->p_sysent->sv_fixup != NULL) (*p->p_sysent->sv_fixup)(&stack_base, imgp); else suword(--stack_base, imgp->args->argc); /* * For security and other reasons, the file descriptor table cannot * be shared after an exec. */ fdunshare(td); /* close files on exec */ fdcloseexec(td); /* * Malloc things before we need locks. */ i = imgp->args->begin_envv - imgp->args->begin_argv; /* Cache arguments if they fit inside our allowance */ if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { newargs = pargs_alloc(i); bcopy(imgp->args->begin_argv, newargs->ar_args, i); } vn_lock(imgp->vp, LK_SHARED | LK_RETRY); /* Get a reference to the vnode prior to locking the proc */ VREF(binvp); /* * For security and other reasons, signal handlers cannot * be shared after an exec. The new process gets a copy of the old * handlers. In execsigs(), the new process will have its signals * reset. */ if (sigacts_shared(p->p_sigacts)) { oldsigacts = p->p_sigacts; newsigacts = sigacts_alloc(); sigacts_copy(newsigacts, oldsigacts); } else { oldsigacts = NULL; newsigacts = NULL; /* satisfy gcc */ } PROC_LOCK(p); if (oldsigacts) p->p_sigacts = newsigacts; oldcred = p->p_ucred; /* Stop profiling */ stopprofclock(p); /* reset caught signals */ execsigs(p); /* name this process - nameiexec(p, ndp) */ bzero(p->p_comm, sizeof(p->p_comm)); if (args->fname) bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, min(nd.ni_cnd.cn_namelen, MAXCOMLEN)); else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0) bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title)); bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); #ifdef KTR sched_clear_tdname(td); #endif /* * mark as execed, wakeup the process that vforked (if any) and tell * it that it now has its own resources back */ p->p_flag |= P_EXEC; if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0) p->p_flag2 &= ~P2_NOTRACE; if (p->p_flag & P_PPWAIT) { p->p_flag &= ~(P_PPWAIT | P_PPTRACE); cv_broadcast(&p->p_pwait); } /* * Implement image setuid/setgid. * * Don't honor setuid/setgid if the filesystem prohibits it or if * the process is being traced. * * We disable setuid/setgid/etc in compatibility mode on the basis * that most setugid applications are not written with that * environment in mind, and will therefore almost certainly operate * incorrectly. In principle there's no reason that setugid * applications might not be useful in capability mode, so we may want * to reconsider this conservative design choice in the future. * * XXXMAC: For the time being, use NOSUID to also prohibit * transitions on the file system. */ credential_changing = 0; credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid != attr.va_uid; credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid != attr.va_gid; #ifdef MAC will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, interpvplabel, imgp); credential_changing |= will_transition; #endif if (credential_changing && #ifdef CAPABILITY_MODE ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && #endif (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && (p->p_flag & P_TRACED) == 0) { /* * Turn off syscall tracing for set-id programs, except for * root. Record any set-id flags first to make sure that * we do not regain any tracing during a possible block. */ setsugid(p); #ifdef KTRACE if (p->p_tracecred != NULL && priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0)) ktrprocexec(p, &tracecred, &tracevp); #endif /* * Close any file descriptors 0..2 that reference procfs, * then make sure file descriptors 0..2 are in use. * * Both fdsetugidsafety() and fdcheckstd() may call functions * taking sleepable locks, so temporarily drop our locks. */ PROC_UNLOCK(p); VOP_UNLOCK(imgp->vp, 0); fdsetugidsafety(td); error = fdcheckstd(td); if (error != 0) goto done1; newcred = crdup(oldcred); euip = uifind(attr.va_uid); vn_lock(imgp->vp, LK_SHARED | LK_RETRY); PROC_LOCK(p); /* * Set the new credentials. */ if (attr.va_mode & S_ISUID) change_euid(newcred, euip); if (attr.va_mode & S_ISGID) change_egid(newcred, attr.va_gid); #ifdef MAC if (will_transition) { mac_vnode_execve_transition(oldcred, newcred, imgp->vp, interpvplabel, imgp); } #endif /* * Implement correct POSIX saved-id behavior. * * XXXMAC: Note that the current logic will save the * uid and gid if a MAC domain transition occurs, even * though maybe it shouldn't. */ change_svuid(newcred, newcred->cr_uid); change_svgid(newcred, newcred->cr_gid); proc_set_cred(p, newcred); } else { if (oldcred->cr_uid == oldcred->cr_ruid && oldcred->cr_gid == oldcred->cr_rgid) p->p_flag &= ~P_SUGID; /* * Implement correct POSIX saved-id behavior. * * XXX: It's not clear that the existing behavior is * POSIX-compliant. A number of sources indicate that the * saved uid/gid should only be updated if the new ruid is * not equal to the old ruid, or the new euid is not equal * to the old euid and the new euid is not equal to the old * ruid. The FreeBSD code always updates the saved uid/gid. * Also, this code uses the new (replaced) euid and egid as * the source, which may or may not be the right ones to use. */ if (oldcred->cr_svuid != oldcred->cr_uid || oldcred->cr_svgid != oldcred->cr_gid) { PROC_UNLOCK(p); VOP_UNLOCK(imgp->vp, 0); newcred = crdup(oldcred); vn_lock(imgp->vp, LK_SHARED | LK_RETRY); PROC_LOCK(p); change_svuid(newcred, newcred->cr_uid); change_svgid(newcred, newcred->cr_gid); proc_set_cred(p, newcred); } } /* * Store the vp for use in procfs. This vnode was referenced prior * to locking the proc lock. */ textvp = p->p_textvp; p->p_textvp = binvp; #ifdef KDTRACE_HOOKS /* * Tell the DTrace fasttrap provider about the exec if it * has declared an interest. */ if (dtrace_fasttrap_exec) dtrace_fasttrap_exec(p); #endif /* * Notify others that we exec'd, and clear the P_INEXEC flag * as we're now a bona fide freshly-execed process. */ KNOTE_LOCKED(&p->p_klist, NOTE_EXEC); p->p_flag &= ~P_INEXEC; /* clear "fork but no exec" flag, as we _are_ execing */ p->p_acflag &= ~AFORK; /* * Free any previous argument cache and replace it with * the new argument cache, if any. */ oldargs = p->p_args; p->p_args = newargs; newargs = NULL; #ifdef HWPMC_HOOKS /* * Check if system-wide sampling is in effect or if the * current process is using PMCs. If so, do exec() time * processing. This processing needs to happen AFTER the * P_INEXEC flag is cleared. * * The proc lock needs to be released before taking the PMC * SX. */ if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { PROC_UNLOCK(p); VOP_UNLOCK(imgp->vp, 0); pe.pm_credentialschanged = credential_changing; pe.pm_entryaddr = imgp->entry_addr; PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); vn_lock(imgp->vp, LK_SHARED | LK_RETRY); } else PROC_UNLOCK(p); #else /* !HWPMC_HOOKS */ PROC_UNLOCK(p); #endif /* Set values passed into the program in registers. */ if (p->p_sysent->sv_setregs) (*p->p_sysent->sv_setregs)(td, imgp, (u_long)(uintptr_t)stack_base); else exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base); vfs_mark_atime(imgp->vp, td->td_ucred); SDT_PROBE(proc, kernel, , exec__success, args->fname, 0, 0, 0, 0); VOP_UNLOCK(imgp->vp, 0); done1: /* * Free any resources malloc'd earlier that we didn't use. */ if (euip != NULL) uifree(euip); if (newcred != NULL) crfree(oldcred); /* * Handle deferred decrement of ref counts. */ if (textvp != NULL) vrele(textvp); if (error != 0) vrele(binvp); #ifdef KTRACE if (tracevp != NULL) vrele(tracevp); if (tracecred != NULL) crfree(tracecred); #endif vn_lock(imgp->vp, LK_SHARED | LK_RETRY); pargs_drop(oldargs); pargs_drop(newargs); if (oldsigacts != NULL) sigacts_free(oldsigacts); exec_fail_dealloc: /* * free various allocated resources */ if (imgp->firstpage != NULL) exec_unmap_first_page(imgp); if (imgp->vp != NULL) { if (args->fname) NDFREE(&nd, NDF_ONLY_PNBUF); if (imgp->opened) VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); vput(imgp->vp); } if (imgp->object != NULL) vm_object_deallocate(imgp->object); free(imgp->freepath, M_TEMP); if (error == 0) { PROC_LOCK(p); td->td_dbgflags |= TDB_EXEC; PROC_UNLOCK(p); /* * Stop the process here if its stop event mask has * the S_EXEC bit set. */ STOPEVENT(p, S_EXEC, 0); goto done2; } exec_fail: /* we're done here, clear P_INEXEC */ PROC_LOCK(p); p->p_flag &= ~P_INEXEC; PROC_UNLOCK(p); SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0); done2: #ifdef MAC mac_execve_exit(imgp); mac_execve_interpreter_exit(interpvplabel); #endif exec_free_args(args); if (error && imgp->vmspace_destroyed) { /* sorry, no more process anymore. exit gracefully */ exit1(td, W_EXITCODE(0, SIGABRT)); /* NOT REACHED */ } #ifdef KTRACE if (error == 0) ktrprocctor(p); #endif return (error); } int exec_map_first_page(imgp) struct image_params *imgp; { int rv, i; int initial_pagein; vm_page_t ma[VM_INITIAL_PAGEIN]; vm_object_t object; if (imgp->firstpage != NULL) exec_unmap_first_page(imgp); object = imgp->vp->v_object; if (object == NULL) return (EACCES); VM_OBJECT_WLOCK(object); #if VM_NRESERVLEVEL > 0 vm_object_color(object, 0); #endif ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL); if (ma[0]->valid != VM_PAGE_BITS_ALL) { initial_pagein = VM_INITIAL_PAGEIN; if (initial_pagein > object->size) initial_pagein = object->size; for (i = 1; i < initial_pagein; i++) { if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) { if (ma[i]->valid) break; if (vm_page_tryxbusy(ma[i])) break; } else { ma[i] = vm_page_alloc(object, i, VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED); if (ma[i] == NULL) break; } } initial_pagein = i; rv = vm_pager_get_pages(object, ma, initial_pagein, 0); ma[0] = vm_page_lookup(object, 0); if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) { if (ma[0] != NULL) { vm_page_lock(ma[0]); vm_page_free(ma[0]); vm_page_unlock(ma[0]); } VM_OBJECT_WUNLOCK(object); return (EIO); } } vm_page_xunbusy(ma[0]); vm_page_lock(ma[0]); vm_page_hold(ma[0]); vm_page_activate(ma[0]); vm_page_unlock(ma[0]); VM_OBJECT_WUNLOCK(object); imgp->firstpage = sf_buf_alloc(ma[0], 0); imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); return (0); } void exec_unmap_first_page(imgp) struct image_params *imgp; { vm_page_t m; if (imgp->firstpage != NULL) { m = sf_buf_page(imgp->firstpage); sf_buf_free(imgp->firstpage); imgp->firstpage = NULL; vm_page_lock(m); vm_page_unhold(m); vm_page_unlock(m); } } /* * Destroy old address space, and allocate a new stack * The new stack is only SGROWSIZ large because it is grown * automatically in trap.c. */ int exec_new_vmspace(imgp, sv) struct image_params *imgp; struct sysentvec *sv; { int error; struct proc *p = imgp->proc; struct vmspace *vmspace = p->p_vmspace; vm_object_t obj; struct rlimit rlim_stack; vm_offset_t sv_minuser, stack_addr; vm_map_t map; u_long ssiz; imgp->vmspace_destroyed = 1; imgp->sysent = sv; /* May be called with Giant held */ EVENTHANDLER_INVOKE(process_exec, p, imgp); /* * Blow away entire process VM, if address space not shared, * otherwise, create a new VM space so that other threads are * not disrupted */ map = &vmspace->vm_map; if (map_at_zero) sv_minuser = sv->sv_minuser; else sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser && vm_map_max(map) == sv->sv_maxuser) { shmexit(vmspace); pmap_remove_pages(vmspace_pmap(vmspace)); vm_map_remove(map, vm_map_min(map), vm_map_max(map)); } else { error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); if (error) return (error); vmspace = p->p_vmspace; map = &vmspace->vm_map; } /* Map a shared page */ obj = sv->sv_shared_page_obj; if (obj != NULL) { vm_object_reference(obj); error = vm_map_fixed(map, obj, 0, sv->sv_shared_page_base, sv->sv_shared_page_len, VM_PROT_READ | VM_PROT_EXECUTE, VM_PROT_READ | VM_PROT_EXECUTE, MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); if (error) { vm_object_deallocate(obj); return (error); } } /* Allocate a new stack */ if (imgp->stack_sz != 0) { ssiz = trunc_page(imgp->stack_sz); PROC_LOCK(p); - lim_rlimit(p, RLIMIT_STACK, &rlim_stack); + lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack); PROC_UNLOCK(p); if (ssiz > rlim_stack.rlim_max) ssiz = rlim_stack.rlim_max; if (ssiz > rlim_stack.rlim_cur) { rlim_stack.rlim_cur = ssiz; kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack); } } else if (sv->sv_maxssiz != NULL) { ssiz = *sv->sv_maxssiz; } else { ssiz = maxssiz; } stack_addr = sv->sv_usrstack - ssiz; error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN); if (error) return (error); /* * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they * are still used to enforce the stack rlimit on the process stack. */ vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz; return (0); } /* * Copy out argument and environment strings from the old process address * space into the temporary string buffer. */ int exec_copyin_args(struct image_args *args, char *fname, enum uio_seg segflg, char **argv, char **envv) { u_long argp, envp; int error; size_t length; bzero(args, sizeof(*args)); if (argv == NULL) return (EFAULT); /* * Allocate demand-paged memory for the file name, argument, and * environment strings. */ error = exec_alloc_args(args); if (error != 0) return (error); /* * Copy the file name. */ if (fname != NULL) { args->fname = args->buf; error = (segflg == UIO_SYSSPACE) ? copystr(fname, args->fname, PATH_MAX, &length) : copyinstr(fname, args->fname, PATH_MAX, &length); if (error != 0) goto err_exit; } else length = 0; args->begin_argv = args->buf + length; args->endp = args->begin_argv; args->stringspace = ARG_MAX; /* * extract arguments first */ for (;;) { error = fueword(argv++, &argp); if (error == -1) { error = EFAULT; goto err_exit; } if (argp == 0) break; error = copyinstr((void *)(uintptr_t)argp, args->endp, args->stringspace, &length); if (error != 0) { if (error == ENAMETOOLONG) error = E2BIG; goto err_exit; } args->stringspace -= length; args->endp += length; args->argc++; } args->begin_envv = args->endp; /* * extract environment strings */ if (envv) { for (;;) { error = fueword(envv++, &envp); if (error == -1) { error = EFAULT; goto err_exit; } if (envp == 0) break; error = copyinstr((void *)(uintptr_t)envp, args->endp, args->stringspace, &length); if (error != 0) { if (error == ENAMETOOLONG) error = E2BIG; goto err_exit; } args->stringspace -= length; args->endp += length; args->envc++; } } return (0); err_exit: exec_free_args(args); return (error); } /* * Allocate temporary demand-paged, zero-filled memory for the file name, * argument, and environment strings. Returns zero if the allocation succeeds * and ENOMEM otherwise. */ int exec_alloc_args(struct image_args *args) { args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX); return (args->buf != NULL ? 0 : ENOMEM); } void exec_free_args(struct image_args *args) { if (args->buf != NULL) { kmap_free_wakeup(exec_map, (vm_offset_t)args->buf, PATH_MAX + ARG_MAX); args->buf = NULL; } if (args->fname_buf != NULL) { free(args->fname_buf, M_TEMP); args->fname_buf = NULL; } } /* * Copy strings out to the new process address space, constructing new arg * and env vector tables. Return a pointer to the base so that it can be used * as the initial stack pointer. */ register_t * exec_copyout_strings(imgp) struct image_params *imgp; { int argc, envc; char **vectp; char *stringp; uintptr_t destp; register_t *stack_base; struct ps_strings *arginfo; struct proc *p; size_t execpath_len; int szsigcode, szps; char canary[sizeof(long) * 8]; szps = sizeof(pagesizes[0]) * MAXPAGESIZES; /* * Calculate string base and vector table pointers. * Also deal with signal trampoline code for this exec type. */ if (imgp->execpath != NULL && imgp->auxargs != NULL) execpath_len = strlen(imgp->execpath) + 1; else execpath_len = 0; p = imgp->proc; szsigcode = 0; arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; if (p->p_sysent->sv_sigcode_base == 0) { if (p->p_sysent->sv_szsigcode != NULL) szsigcode = *(p->p_sysent->sv_szsigcode); } destp = (uintptr_t)arginfo; /* * install sigcode */ if (szsigcode != 0) { destp -= szsigcode; destp = rounddown2(destp, sizeof(void *)); copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode); } /* * Copy the image path for the rtld. */ if (execpath_len != 0) { destp -= execpath_len; imgp->execpathp = destp; copyout(imgp->execpath, (void *)destp, execpath_len); } /* * Prepare the canary for SSP. */ arc4rand(canary, sizeof(canary), 0); destp -= sizeof(canary); imgp->canary = destp; copyout(canary, (void *)destp, sizeof(canary)); imgp->canarylen = sizeof(canary); /* * Prepare the pagesizes array. */ destp -= szps; destp = rounddown2(destp, sizeof(void *)); imgp->pagesizes = destp; copyout(pagesizes, (void *)destp, szps); imgp->pagesizeslen = szps; destp -= ARG_MAX - imgp->args->stringspace; destp = rounddown2(destp, sizeof(void *)); /* * If we have a valid auxargs ptr, prepare some room * on the stack. */ if (imgp->auxargs) { /* * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for * lower compatibility. */ imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size : (AT_COUNT * 2); /* * The '+ 2' is for the null pointers at the end of each of * the arg and env vector sets,and imgp->auxarg_size is room * for argument of Runtime loader. */ vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2 + imgp->auxarg_size) * sizeof(char *)); } else { /* * The '+ 2' is for the null pointers at the end of each of * the arg and env vector sets */ vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) * sizeof(char *)); } /* * vectp also becomes our initial stack base */ stack_base = (register_t *)vectp; stringp = imgp->args->begin_argv; argc = imgp->args->argc; envc = imgp->args->envc; /* * Copy out strings - arguments and environment. */ copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace); /* * Fill in "ps_strings" struct for ps, w, etc. */ suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); suword32(&arginfo->ps_nargvstr, argc); /* * Fill in argument portion of vector table. */ for (; argc > 0; --argc) { suword(vectp++, (long)(intptr_t)destp); while (*stringp++ != 0) destp++; destp++; } /* a null vector table pointer separates the argp's from the envp's */ suword(vectp++, 0); suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); suword32(&arginfo->ps_nenvstr, envc); /* * Fill in environment portion of vector table. */ for (; envc > 0; --envc) { suword(vectp++, (long)(intptr_t)destp); while (*stringp++ != 0) destp++; destp++; } /* end of vector table is a null pointer */ suword(vectp, 0); return (stack_base); } /* * Check permissions of file to execute. * Called with imgp->vp locked. * Return 0 for success or error code on failure. */ int exec_check_permissions(imgp) struct image_params *imgp; { struct vnode *vp = imgp->vp; struct vattr *attr = imgp->attr; struct thread *td; int error, writecount; td = curthread; /* Get file attributes */ error = VOP_GETATTR(vp, attr, td->td_ucred); if (error) return (error); #ifdef MAC error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); if (error) return (error); #endif /* * 1) Check if file execution is disabled for the filesystem that * this file resides on. * 2) Ensure that at least one execute bit is on. Otherwise, a * privileged user will always succeed, and we don't want this * to happen unless the file really is executable. * 3) Ensure that the file is a regular file. */ if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || (attr->va_type != VREG)) return (EACCES); /* * Zero length files can't be exec'd */ if (attr->va_size == 0) return (ENOEXEC); /* * Check for execute permission to file based on current credentials. */ error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); if (error) return (error); /* * Check number of open-for-writes on the file and deny execution * if there are any. */ error = VOP_GET_WRITECOUNT(vp, &writecount); if (error != 0) return (error); if (writecount != 0) return (ETXTBSY); /* * Call filesystem specific open routine (which does nothing in the * general case). */ error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); if (error == 0) imgp->opened = 1; return (error); } /* * Exec handler registration */ int exec_register(execsw_arg) const struct execsw *execsw_arg; { const struct execsw **es, **xs, **newexecsw; int count = 2; /* New slot and trailing NULL */ if (execsw) for (es = execsw; *es; es++) count++; newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); if (newexecsw == NULL) return (ENOMEM); xs = newexecsw; if (execsw) for (es = execsw; *es; es++) *xs++ = *es; *xs++ = execsw_arg; *xs = NULL; if (execsw) free(execsw, M_TEMP); execsw = newexecsw; return (0); } int exec_unregister(execsw_arg) const struct execsw *execsw_arg; { const struct execsw **es, **xs, **newexecsw; int count = 1; if (execsw == NULL) panic("unregister with no handlers left?\n"); for (es = execsw; *es; es++) { if (*es == execsw_arg) break; } if (*es == NULL) return (ENOENT); for (es = execsw; *es; es++) if (*es != execsw_arg) count++; newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); if (newexecsw == NULL) return (ENOMEM); xs = newexecsw; for (es = execsw; *es; es++) if (*es != execsw_arg) *xs++ = *es; *xs = NULL; if (execsw) free(execsw, M_TEMP); execsw = newexecsw; return (0); } Index: head/sys/kern/kern_fork.c =================================================================== --- head/sys/kern/kern_fork.c (revision 284214) +++ head/sys/kern/kern_fork.c (revision 284215) @@ -1,1064 +1,1062 @@ /*- * Copyright (c) 1982, 1986, 1989, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_ktrace.h" #include "opt_kstack_pages.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KDTRACE_HOOKS #include dtrace_fork_func_t dtrace_fasttrap_fork; #endif SDT_PROVIDER_DECLARE(proc); SDT_PROBE_DEFINE3(proc, kernel, , create, "struct proc *", "struct proc *", "int"); #ifndef _SYS_SYSPROTO_H_ struct fork_args { int dummy; }; #endif /* ARGSUSED */ int sys_fork(struct thread *td, struct fork_args *uap) { int error; struct proc *p2; error = fork1(td, RFFDG | RFPROC, 0, &p2, NULL, 0); if (error == 0) { td->td_retval[0] = p2->p_pid; td->td_retval[1] = 0; } return (error); } /* ARGUSED */ int sys_pdfork(td, uap) struct thread *td; struct pdfork_args *uap; { int error, fd; struct proc *p2; /* * It is necessary to return fd by reference because 0 is a valid file * descriptor number, and the child needs to be able to distinguish * itself from the parent using the return value. */ error = fork1(td, RFFDG | RFPROC | RFPROCDESC, 0, &p2, &fd, uap->flags); if (error == 0) { td->td_retval[0] = p2->p_pid; td->td_retval[1] = 0; error = copyout(&fd, uap->fdp, sizeof(fd)); } return (error); } /* ARGSUSED */ int sys_vfork(struct thread *td, struct vfork_args *uap) { int error, flags; struct proc *p2; flags = RFFDG | RFPROC | RFPPWAIT | RFMEM; error = fork1(td, flags, 0, &p2, NULL, 0); if (error == 0) { td->td_retval[0] = p2->p_pid; td->td_retval[1] = 0; } return (error); } int sys_rfork(struct thread *td, struct rfork_args *uap) { struct proc *p2; int error; /* Don't allow kernel-only flags. */ if ((uap->flags & RFKERNELONLY) != 0) return (EINVAL); AUDIT_ARG_FFLAGS(uap->flags); error = fork1(td, uap->flags, 0, &p2, NULL, 0); if (error == 0) { td->td_retval[0] = p2 ? p2->p_pid : 0; td->td_retval[1] = 0; } return (error); } int nprocs = 1; /* process 0 */ int lastpid = 0; SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, "Last used PID"); /* * Random component to lastpid generation. We mix in a random factor to make * it a little harder to predict. We sanity check the modulus value to avoid * doing it in critical paths. Don't let it be too small or we pointlessly * waste randomness entropy, and don't let it be impossibly large. Using a * modulus that is too big causes a LOT more process table scans and slows * down fork processing as the pidchecked caching is defeated. */ static int randompid = 0; static int sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) { int error, pid; error = sysctl_wire_old_buffer(req, sizeof(int)); if (error != 0) return(error); sx_xlock(&allproc_lock); pid = randompid; error = sysctl_handle_int(oidp, &pid, 0, req); if (error == 0 && req->newptr != NULL) { if (pid < 0 || pid > pid_max - 100) /* out of range */ pid = pid_max - 100; else if (pid < 2) /* NOP */ pid = 0; else if (pid < 100) /* Make it reasonable */ pid = 100; randompid = pid; } sx_xunlock(&allproc_lock); return (error); } SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); static int fork_findpid(int flags) { struct proc *p; int trypid; static int pidchecked = 0; /* * Requires allproc_lock in order to iterate over the list * of processes, and proctree_lock to access p_pgrp. */ sx_assert(&allproc_lock, SX_LOCKED); sx_assert(&proctree_lock, SX_LOCKED); /* * Find an unused process ID. We remember a range of unused IDs * ready to use (from lastpid+1 through pidchecked-1). * * If RFHIGHPID is set (used during system boot), do not allocate * low-numbered pids. */ trypid = lastpid + 1; if (flags & RFHIGHPID) { if (trypid < 10) trypid = 10; } else { if (randompid) trypid += arc4random() % randompid; } retry: /* * If the process ID prototype has wrapped around, * restart somewhat above 0, as the low-numbered procs * tend to include daemons that don't exit. */ if (trypid >= pid_max) { trypid = trypid % pid_max; if (trypid < 100) trypid += 100; pidchecked = 0; } if (trypid >= pidchecked) { int doingzomb = 0; pidchecked = PID_MAX; /* * Scan the active and zombie procs to check whether this pid * is in use. Remember the lowest pid that's greater * than trypid, so we can avoid checking for a while. * * Avoid reuse of the process group id, session id or * the reaper subtree id. Note that for process group * and sessions, the amount of reserved pids is * limited by process limit. For the subtree ids, the * id is kept reserved only while there is a * non-reaped process in the subtree, so amount of * reserved pids is limited by process limit times * two. */ p = LIST_FIRST(&allproc); again: for (; p != NULL; p = LIST_NEXT(p, p_list)) { while (p->p_pid == trypid || p->p_reapsubtree == trypid || (p->p_pgrp != NULL && (p->p_pgrp->pg_id == trypid || (p->p_session != NULL && p->p_session->s_sid == trypid)))) { trypid++; if (trypid >= pidchecked) goto retry; } if (p->p_pid > trypid && pidchecked > p->p_pid) pidchecked = p->p_pid; if (p->p_pgrp != NULL) { if (p->p_pgrp->pg_id > trypid && pidchecked > p->p_pgrp->pg_id) pidchecked = p->p_pgrp->pg_id; if (p->p_session != NULL && p->p_session->s_sid > trypid && pidchecked > p->p_session->s_sid) pidchecked = p->p_session->s_sid; } } if (!doingzomb) { doingzomb = 1; p = LIST_FIRST(&zombproc); goto again; } } /* * RFHIGHPID does not mess with the lastpid counter during boot. */ if (flags & RFHIGHPID) pidchecked = 0; else lastpid = trypid; return (trypid); } static int fork_norfproc(struct thread *td, int flags) { int error; struct proc *p1; KASSERT((flags & RFPROC) == 0, ("fork_norfproc called with RFPROC set")); p1 = td->td_proc; if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && (flags & (RFCFDG | RFFDG))) { PROC_LOCK(p1); if (thread_single(p1, SINGLE_BOUNDARY)) { PROC_UNLOCK(p1); return (ERESTART); } PROC_UNLOCK(p1); } error = vm_forkproc(td, NULL, NULL, NULL, flags); if (error) goto fail; /* * Close all file descriptors. */ if (flags & RFCFDG) { struct filedesc *fdtmp; fdtmp = fdinit(td->td_proc->p_fd, false); fdescfree(td); p1->p_fd = fdtmp; } /* * Unshare file descriptors (from parent). */ if (flags & RFFDG) fdunshare(td); fail: if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && (flags & (RFCFDG | RFFDG))) { PROC_LOCK(p1); thread_single_end(p1, SINGLE_BOUNDARY); PROC_UNLOCK(p1); } return (error); } static void do_fork(struct thread *td, int flags, struct proc *p2, struct thread *td2, struct vmspace *vm2, int pdflags) { struct proc *p1, *pptr; int p2_held, trypid; struct filedesc *fd; struct filedesc_to_leader *fdtol; struct sigacts *newsigacts; sx_assert(&proctree_lock, SX_SLOCKED); sx_assert(&allproc_lock, SX_XLOCKED); p2_held = 0; p1 = td->td_proc; /* * Increment the nprocs resource before blocking can occur. There * are hard-limits as to the number of processes that can run. */ nprocs++; trypid = fork_findpid(flags); sx_sunlock(&proctree_lock); p2->p_state = PRS_NEW; /* protect against others */ p2->p_pid = trypid; AUDIT_ARG_PID(p2->p_pid); LIST_INSERT_HEAD(&allproc, p2, p_list); allproc_gen++; LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); tidhash_add(td2); PROC_LOCK(p2); PROC_LOCK(p1); sx_xunlock(&allproc_lock); bcopy(&p1->p_startcopy, &p2->p_startcopy, __rangeof(struct proc, p_startcopy, p_endcopy)); pargs_hold(p2->p_args); PROC_UNLOCK(p1); bzero(&p2->p_startzero, __rangeof(struct proc, p_startzero, p_endzero)); /* Tell the prison that we exist. */ prison_proc_hold(p2->p_ucred->cr_prison); PROC_UNLOCK(p2); /* * Malloc things while we don't hold any locks. */ if (flags & RFSIGSHARE) newsigacts = NULL; else newsigacts = sigacts_alloc(); /* * Copy filedesc. */ if (flags & RFCFDG) { fd = fdinit(p1->p_fd, false); fdtol = NULL; } else if (flags & RFFDG) { fd = fdcopy(p1->p_fd); fdtol = NULL; } else { fd = fdshare(p1->p_fd); if (p1->p_fdtol == NULL) p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL, p1->p_leader); if ((flags & RFTHREAD) != 0) { /* * Shared file descriptor table, and shared * process leaders. */ fdtol = p1->p_fdtol; FILEDESC_XLOCK(p1->p_fd); fdtol->fdl_refcount++; FILEDESC_XUNLOCK(p1->p_fd); } else { /* * Shared file descriptor table, and different * process leaders. */ fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p1->p_fd, p2); } } /* * Make a proc table entry for the new process. * Start by zeroing the section of proc that is zero-initialized, * then copy the section that is copied directly from the parent. */ PROC_LOCK(p2); PROC_LOCK(p1); bzero(&td2->td_startzero, __rangeof(struct thread, td_startzero, td_endzero)); bcopy(&td->td_startcopy, &td2->td_startcopy, __rangeof(struct thread, td_startcopy, td_endcopy)); bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name)); td2->td_sigstk = td->td_sigstk; td2->td_flags = TDF_INMEM; td2->td_lend_user_pri = PRI_MAX; #ifdef VIMAGE td2->td_vnet = NULL; td2->td_vnet_lpush = NULL; #endif /* * Allow the scheduler to initialize the child. */ thread_lock(td); sched_fork(td, td2); thread_unlock(td); /* * Duplicate sub-structures as needed. * Increase reference counts on shared objects. */ p2->p_flag = P_INMEM; p2->p_flag2 = p1->p_flag2 & (P2_NOTRACE | P2_NOTRACE_EXEC); p2->p_swtick = ticks; if (p1->p_flag & P_PROFIL) startprofclock(p2); if (flags & RFSIGSHARE) { p2->p_sigacts = sigacts_hold(p1->p_sigacts); } else { sigacts_copy(newsigacts, p1->p_sigacts); p2->p_sigacts = newsigacts; } if (flags & RFTSIGZMB) p2->p_sigparent = RFTSIGNUM(flags); else if (flags & RFLINUXTHPN) p2->p_sigparent = SIGUSR1; else p2->p_sigparent = SIGCHLD; p2->p_textvp = p1->p_textvp; p2->p_fd = fd; p2->p_fdtol = fdtol; if (p1->p_flag2 & P2_INHERIT_PROTECTED) { p2->p_flag |= P_PROTECTED; p2->p_flag2 |= P2_INHERIT_PROTECTED; } /* * p_limit is copy-on-write. Bump its refcount. */ lim_fork(p1, p2); thread_cow_get_proc(td2, p2); pstats_fork(p1->p_stats, p2->p_stats); PROC_UNLOCK(p1); PROC_UNLOCK(p2); /* Bump references to the text vnode (for procfs). */ if (p2->p_textvp) vref(p2->p_textvp); /* * Set up linkage for kernel based threading. */ if ((flags & RFTHREAD) != 0) { mtx_lock(&ppeers_lock); p2->p_peers = p1->p_peers; p1->p_peers = p2; p2->p_leader = p1->p_leader; mtx_unlock(&ppeers_lock); PROC_LOCK(p1->p_leader); if ((p1->p_leader->p_flag & P_WEXIT) != 0) { PROC_UNLOCK(p1->p_leader); /* * The task leader is exiting, so process p1 is * going to be killed shortly. Since p1 obviously * isn't dead yet, we know that the leader is either * sending SIGKILL's to all the processes in this * task or is sleeping waiting for all the peers to * exit. We let p1 complete the fork, but we need * to go ahead and kill the new process p2 since * the task leader may not get a chance to send * SIGKILL to it. We leave it on the list so that * the task leader will wait for this new process * to commit suicide. */ PROC_LOCK(p2); kern_psignal(p2, SIGKILL); PROC_UNLOCK(p2); } else PROC_UNLOCK(p1->p_leader); } else { p2->p_peers = NULL; p2->p_leader = p2; } sx_xlock(&proctree_lock); PGRP_LOCK(p1->p_pgrp); PROC_LOCK(p2); PROC_LOCK(p1); /* * Preserve some more flags in subprocess. P_PROFIL has already * been preserved. */ p2->p_flag |= p1->p_flag & P_SUGID; td2->td_pflags |= td->td_pflags & TDP_ALTSTACK; SESS_LOCK(p1->p_session); if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) p2->p_flag |= P_CONTROLT; SESS_UNLOCK(p1->p_session); if (flags & RFPPWAIT) p2->p_flag |= P_PPWAIT; p2->p_pgrp = p1->p_pgrp; LIST_INSERT_AFTER(p1, p2, p_pglist); PGRP_UNLOCK(p1->p_pgrp); LIST_INIT(&p2->p_children); LIST_INIT(&p2->p_orphans); callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0); /* * If PF_FORK is set, the child process inherits the * procfs ioctl flags from its parent. */ if (p1->p_pfsflags & PF_FORK) { p2->p_stops = p1->p_stops; p2->p_pfsflags = p1->p_pfsflags; } /* * This begins the section where we must prevent the parent * from being swapped. */ _PHOLD(p1); PROC_UNLOCK(p1); /* * Attach the new process to its parent. * * If RFNOWAIT is set, the newly created process becomes a child * of init. This effectively disassociates the child from the * parent. */ if ((flags & RFNOWAIT) != 0) { pptr = p1->p_reaper; p2->p_reaper = pptr; } else { p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ? p1 : p1->p_reaper; pptr = p1; } p2->p_pptr = pptr; LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); LIST_INIT(&p2->p_reaplist); LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling); if (p2->p_reaper == p1) p2->p_reapsubtree = p2->p_pid; sx_xunlock(&proctree_lock); /* Inform accounting that we have forked. */ p2->p_acflag = AFORK; PROC_UNLOCK(p2); #ifdef KTRACE ktrprocfork(p1, p2); #endif /* * Finish creating the child process. It will return via a different * execution path later. (ie: directly into user mode) */ vm_forkproc(td, p2, td2, vm2, flags); if (flags == (RFFDG | RFPROC)) { PCPU_INC(cnt.v_forks); PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize); } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { PCPU_INC(cnt.v_vforks); PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize); } else if (p1 == &proc0) { PCPU_INC(cnt.v_kthreads); PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize); } else { PCPU_INC(cnt.v_rforks); PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize); } /* * Associate the process descriptor with the process before anything * can happen that might cause that process to need the descriptor. * However, don't do this until after fork(2) can no longer fail. */ if (flags & RFPROCDESC) procdesc_new(p2, pdflags); /* * Both processes are set up, now check if any loadable modules want * to adjust anything. */ EVENTHANDLER_INVOKE(process_fork, p1, p2, flags); /* * Set the child start time and mark the process as being complete. */ PROC_LOCK(p2); PROC_LOCK(p1); microuptime(&p2->p_stats->p_start); PROC_SLOCK(p2); p2->p_state = PRS_NORMAL; PROC_SUNLOCK(p2); #ifdef KDTRACE_HOOKS /* * Tell the DTrace fasttrap provider about the new process so that any * tracepoints inherited from the parent can be removed. We have to do * this only after p_state is PRS_NORMAL since the fasttrap module will * use pfind() later on. */ if ((flags & RFMEM) == 0 && dtrace_fasttrap_fork) dtrace_fasttrap_fork(p1, p2); #endif if ((p1->p_flag & (P_TRACED | P_FOLLOWFORK)) == (P_TRACED | P_FOLLOWFORK)) { /* * Arrange for debugger to receive the fork event. * * We can report PL_FLAG_FORKED regardless of * P_FOLLOWFORK settings, but it does not make a sense * for runaway child. */ td->td_dbgflags |= TDB_FORK; td->td_dbg_forked = p2->p_pid; td2->td_dbgflags |= TDB_STOPATFORK; _PHOLD(p2); p2_held = 1; } if (flags & RFPPWAIT) { td->td_pflags |= TDP_RFPPWAIT; td->td_rfppwait_p = p2; } PROC_UNLOCK(p2); if ((flags & RFSTOPPED) == 0) { /* * If RFSTOPPED not requested, make child runnable and * add to run queue. */ thread_lock(td2); TD_SET_CAN_RUN(td2); sched_add(td2, SRQ_BORING); thread_unlock(td2); } /* * Now can be swapped. */ _PRELE(p1); PROC_UNLOCK(p1); /* * Tell any interested parties about the new process. */ knote_fork(&p1->p_klist, p2->p_pid); SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0); /* * Wait until debugger is attached to child. */ PROC_LOCK(p2); while ((td2->td_dbgflags & TDB_STOPATFORK) != 0) cv_wait(&p2->p_dbgwait, &p2->p_mtx); if (p2_held) _PRELE(p2); PROC_UNLOCK(p2); } int fork1(struct thread *td, int flags, int pages, struct proc **procp, int *procdescp, int pdflags) { struct proc *p1; struct proc *newproc; int ok; struct thread *td2; struct vmspace *vm2; vm_ooffset_t mem_charged; int error; static int curfail; static struct timeval lastfail; struct file *fp_procdesc = NULL; /* Check for the undefined or unimplemented flags. */ if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0) return (EINVAL); /* Signal value requires RFTSIGZMB. */ if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0) return (EINVAL); /* Can't copy and clear. */ if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) return (EINVAL); /* Check the validity of the signal number. */ if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG) return (EINVAL); if ((flags & RFPROCDESC) != 0) { /* Can't not create a process yet get a process descriptor. */ if ((flags & RFPROC) == 0) return (EINVAL); /* Must provide a place to put a procdesc if creating one. */ if (procdescp == NULL) return (EINVAL); } p1 = td->td_proc; /* * Here we don't create a new process, but we divorce * certain parts of a process from itself. */ if ((flags & RFPROC) == 0) { *procp = NULL; return (fork_norfproc(td, flags)); } /* * If required, create a process descriptor in the parent first; we * will abandon it if something goes wrong. We don't finit() until * later. */ if (flags & RFPROCDESC) { error = falloc(td, &fp_procdesc, procdescp, 0); if (error != 0) return (error); } mem_charged = 0; vm2 = NULL; if (pages == 0) pages = KSTACK_PAGES; /* Allocate new proc. */ newproc = uma_zalloc(proc_zone, M_WAITOK); td2 = FIRST_THREAD_IN_PROC(newproc); if (td2 == NULL) { td2 = thread_alloc(pages); if (td2 == NULL) { error = ENOMEM; goto fail2; } proc_linkup(newproc, td2); } else { if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) { if (td2->td_kstack != 0) vm_thread_dispose(td2); if (!thread_alloc_stack(td2, pages)) { error = ENOMEM; goto fail2; } } } if ((flags & RFMEM) == 0) { vm2 = vmspace_fork(p1->p_vmspace, &mem_charged); if (vm2 == NULL) { error = ENOMEM; goto fail2; } if (!swap_reserve(mem_charged)) { /* * The swap reservation failed. The accounting * from the entries of the copied vm2 will be * substracted in vmspace_free(), so force the * reservation there. */ swap_reserve_force(mem_charged); error = ENOMEM; goto fail2; } } else vm2 = NULL; /* * XXX: This is ugly; when we copy resource usage, we need to bump * per-cred resource counters. */ proc_set_cred_init(newproc, crhold(td->td_ucred)); /* * Initialize resource accounting for the child process. */ error = racct_proc_fork(p1, newproc); if (error != 0) { error = EAGAIN; goto fail1; } #ifdef MAC mac_proc_init(newproc); #endif knlist_init_mtx(&newproc->p_klist, &newproc->p_mtx); STAILQ_INIT(&newproc->p_ktr); /* We have to lock the process tree while we look for a pid. */ sx_slock(&proctree_lock); /* * Although process entries are dynamically created, we still keep * a global limit on the maximum number we will create. Don't allow * a nonprivileged user to use the last ten processes; don't let root * exceed the limit. The variable nprocs is the current number of * processes, maxproc is the limit. */ sx_xlock(&allproc_lock); if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred, PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) { error = EAGAIN; goto fail; } /* * Increment the count of procs running with this uid. Don't allow * a nonprivileged user to exceed their current limit. * * XXXRW: Can we avoid privilege here if it's not needed? */ error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0); if (error == 0) ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0); else { - PROC_LOCK(p1); ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, - lim_cur(p1, RLIMIT_NPROC)); - PROC_UNLOCK(p1); + lim_cur(td, RLIMIT_NPROC)); } if (ok) { do_fork(td, flags, newproc, td2, vm2, pdflags); /* * Return child proc pointer to parent. */ *procp = newproc; if (flags & RFPROCDESC) { procdesc_finit(newproc->p_procdesc, fp_procdesc); fdrop(fp_procdesc, td); } racct_proc_fork_done(newproc); return (0); } error = EAGAIN; fail: sx_sunlock(&proctree_lock); if (ppsratecheck(&lastfail, &curfail, 1)) printf("maxproc limit exceeded by uid %u (pid %d); see tuning(7) and login.conf(5)\n", td->td_ucred->cr_ruid, p1->p_pid); sx_xunlock(&allproc_lock); #ifdef MAC mac_proc_destroy(newproc); #endif racct_proc_exit(newproc); fail1: crfree(newproc->p_ucred); newproc->p_ucred = NULL; fail2: if (vm2 != NULL) vmspace_free(vm2); uma_zfree(proc_zone, newproc); if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) { fdclose(td, fp_procdesc, *procdescp); fdrop(fp_procdesc, td); } pause("fork", hz / 2); return (error); } /* * Handle the return of a child process from fork1(). This function * is called from the MD fork_trampoline() entry point. */ void fork_exit(void (*callout)(void *, struct trapframe *), void *arg, struct trapframe *frame) { struct proc *p; struct thread *td; struct thread *dtd; td = curthread; p = td->td_proc; KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)", td, td->td_sched, p->p_pid, td->td_name); sched_fork_exit(td); /* * Processes normally resume in mi_switch() after being * cpu_switch()'ed to, but when children start up they arrive here * instead, so we must do much the same things as mi_switch() would. */ if ((dtd = PCPU_GET(deadthread))) { PCPU_SET(deadthread, NULL); thread_stash(dtd); } thread_unlock(td); /* * cpu_set_fork_handler intercepts this function call to * have this call a non-return function to stay in kernel mode. * initproc has its own fork handler, but it does return. */ KASSERT(callout != NULL, ("NULL callout in fork_exit")); callout(arg, frame); /* * Check if a kernel thread misbehaved and returned from its main * function. */ if (p->p_flag & P_KTHREAD) { printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", td->td_name, p->p_pid); kproc_exit(0); } mtx_assert(&Giant, MA_NOTOWNED); if (p->p_sysent->sv_schedtail != NULL) (p->p_sysent->sv_schedtail)(td); } /* * Simplified back end of syscall(), used when returning from fork() * directly into user mode. Giant is not held on entry, and must not * be held on return. This function is passed in to fork_exit() as the * first parameter and is called when returning to a new userland process. */ void fork_return(struct thread *td, struct trapframe *frame) { struct proc *p, *dbg; if (td->td_dbgflags & TDB_STOPATFORK) { p = td->td_proc; sx_xlock(&proctree_lock); PROC_LOCK(p); if ((p->p_pptr->p_flag & (P_TRACED | P_FOLLOWFORK)) == (P_TRACED | P_FOLLOWFORK)) { /* * If debugger still wants auto-attach for the * parent's children, do it now. */ dbg = p->p_pptr->p_pptr; p->p_flag |= P_TRACED; p->p_oppid = p->p_pptr->p_pid; CTR2(KTR_PTRACE, "fork_return: attaching to new child pid %d: oppid %d", p->p_pid, p->p_oppid); proc_reparent(p, dbg); sx_xunlock(&proctree_lock); td->td_dbgflags |= TDB_CHILD; ptracestop(td, SIGSTOP); td->td_dbgflags &= ~TDB_CHILD; } else { /* * ... otherwise clear the request. */ sx_xunlock(&proctree_lock); td->td_dbgflags &= ~TDB_STOPATFORK; cv_broadcast(&p->p_dbgwait); } PROC_UNLOCK(p); } userret(td, frame); #ifdef KTRACE if (KTRPOINT(td, KTR_SYSRET)) ktrsysret(SYS_fork, 0, 0); #endif } Index: head/sys/kern/kern_proc.c =================================================================== --- head/sys/kern/kern_proc.c (revision 284214) +++ head/sys/kern/kern_proc.c (revision 284215) @@ -1,3057 +1,3057 @@ /*- * 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 #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, kernel, ctor, entry, "struct proc *", "int", "void *", "int"); SDT_PROBE_DEFINE4(proc, kernel, ctor, return, "struct proc *", "int", "void *", "int"); SDT_PROBE_DEFINE4(proc, kernel, dtor, entry, "struct proc *", "int", "void *", "struct thread *"); SDT_PROBE_DEFINE3(proc, kernel, dtor, return, "struct proc *", "int", "void *"); SDT_PROBE_DEFINE3(proc, kernel, init, entry, "struct proc *", "int", "int"); SDT_PROBE_DEFINE3(proc, kernel, init, return, "struct proc *", "int", "int"); MALLOC_DEFINE(M_PGRP, "pgrp", "process group header"); MALLOC_DEFINE(M_SESSION, "session", "session header"); static MALLOC_DEFINE(M_PROC, "proc", "Proc structures"); MALLOC_DEFINE(M_SUBPROC, "subproc", "Proc sub-structures"); static void doenterpgrp(struct proc *, struct pgrp *); static void orphanpg(struct pgrp *pg); static void fill_kinfo_aggregate(struct proc *p, struct kinfo_proc *kp); static void fill_kinfo_proc_only(struct proc *p, struct kinfo_proc *kp); static void fill_kinfo_thread(struct thread *td, struct kinfo_proc *kp, int preferthread); static void pgadjustjobc(struct pgrp *pgrp, int entering); static void pgdelete(struct pgrp *); static int proc_ctor(void *mem, int size, void *arg, int flags); static void proc_dtor(void *mem, int size, void *arg); static int proc_init(void *mem, int size, int flags); static void proc_fini(void *mem, int size); static void pargs_free(struct pargs *pa); static struct proc *zpfind_locked(pid_t pid); /* * Other process lists */ struct pidhashhead *pidhashtbl; u_long pidhash; struct pgrphashhead *pgrphashtbl; u_long pgrphash; struct proclist allproc; struct proclist zombproc; struct sx allproc_lock; struct sx proctree_lock; struct mtx ppeers_lock; uma_zone_t proc_zone; 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() { sx_init(&allproc_lock, "allproc"); sx_init(&proctree_lock, "proctree"); mtx_init(&ppeers_lock, "p_peers", NULL, MTX_DEF); LIST_INIT(&allproc); LIST_INIT(&zombproc); pidhashtbl = hashinit(maxproc / 4, M_PROC, &pidhash); pgrphashtbl = hashinit(maxproc / 4, M_PROC, &pgrphash); proc_zone = uma_zcreate("PROC", sched_sizeof_proc(), proc_ctor, proc_dtor, proc_init, proc_fini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); uihashinit(); } /* * Prepare a proc for use. */ static int proc_ctor(void *mem, int size, void *arg, int flags) { struct proc *p; p = (struct proc *)mem; SDT_PROBE(proc, kernel, ctor , entry, p, size, arg, flags, 0); EVENTHANDLER_INVOKE(process_ctor, p); SDT_PROBE(proc, kernel, ctor , return, p, size, arg, flags, 0); 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_PROBE(proc, kernel, dtor, entry, p, size, arg, td, 0); if (td != NULL) { #ifdef INVARIANTS KASSERT((p->p_numthreads == 1), ("bad number of threads in exiting process")); KASSERT(STAILQ_EMPTY(&p->p_ktr), ("proc_dtor: non-empty p_ktr")); #endif /* Free all OSD associated to this thread. */ osd_thread_exit(td); } EVENTHANDLER_INVOKE(process_dtor, p); if (p->p_ksi != NULL) KASSERT(! KSI_ONQ(p->p_ksi), ("SIGCHLD queue")); SDT_PROBE(proc, kernel, dtor, return, p, size, arg, 0, 0); } /* * 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_PROBE(proc, kernel, init, entry, p, size, flags, 0, 0); p->p_sched = (struct p_sched *)&p[1]; 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(); SDT_PROBE(proc, kernel, init, return, p, size, flags, 0, 0); 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(p, pgrp, entering) register struct proc *p; register struct pgrp *pgrp; int entering; { register struct pgrp *hispgrp; register struct session *mysession; 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(p, &p->p_children, p_sibling) { hispgrp = p->p_pgrp; if (hispgrp == pgrp || hispgrp->pg_session != mysession) continue; PROC_LOCK(p); if (p->p_state == PRS_ZOMBIE) { PROC_UNLOCK(p); continue; } PROC_UNLOCK(p); pgadjustjobc(hispgrp, entering); } } /* * 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)) { 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 += td->td_estcpu; 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; /* 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; 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 = p->p_xstat; kp->ki_acflag = p->p_acflag; kp->ki_lock = p->p_lock; if (p->p_pptr) { kp->ki_ppid = proc_realparent(p)->p_pid; if (p->p_flag & P_TRACED) kp->ki_tracer = p->p_pptr->p_pid; } } /* * Fill in information that is thread specific. Must be called with * target process locked. If 'preferthread' is set, overwrite certain * process-related fields that are maintained for both threads and * processes. */ static void fill_kinfo_thread(struct thread *td, struct kinfo_proc *kp, int preferthread) { struct proc *p; p = td->td_proc; kp->ki_tdaddr = td; PROC_LOCK_ASSERT(p, MA_OWNED); if (preferthread) PROC_STATLOCK(p); thread_lock(td); if (td->td_wmesg != NULL) strlcpy(kp->ki_wmesg, td->td_wmesg, sizeof(kp->ki_wmesg)); else bzero(kp->ki_wmesg, sizeof(kp->ki_wmesg)); strlcpy(kp->ki_tdname, td->td_name, sizeof(kp->ki_tdname)); if (TD_ON_LOCK(td)) { kp->ki_kiflag |= KI_LOCKBLOCK; strlcpy(kp->ki_lockname, td->td_lockname, sizeof(kp->ki_lockname)); } else { kp->ki_kiflag &= ~KI_LOCKBLOCK; bzero(kp->ki_lockname, sizeof(kp->ki_lockname)); } if (p->p_state == PRS_NORMAL) { /* approximate. */ if (TD_ON_RUNQ(td) || TD_CAN_RUN(td) || TD_IS_RUNNING(td)) { kp->ki_stat = SRUN; } else if (P_SHOULDSTOP(p)) { kp->ki_stat = SSTOP; } else if (TD_IS_SLEEPING(td)) { kp->ki_stat = SSLEEP; } else if (TD_ON_LOCK(td)) { kp->ki_stat = SLOCK; } else { kp->ki_stat = SWAIT; } } else if (p->p_state == PRS_ZOMBIE) { kp->ki_stat = SZOMB; } else { kp->ki_stat = SIDL; } /* Things in the thread */ kp->ki_wchan = td->td_wchan; kp->ki_pri.pri_level = td->td_priority; kp->ki_pri.pri_native = td->td_base_pri; /* * Note: legacy fields; clamp at the old NOCPU value and/or * the maximum u_char CPU value. */ if (td->td_lastcpu == NOCPU) kp->ki_lastcpu_old = NOCPU_OLD; else if (td->td_lastcpu > MAXCPU_OLD) kp->ki_lastcpu_old = MAXCPU_OLD; else kp->ki_lastcpu_old = td->td_lastcpu; if (td->td_oncpu == NOCPU) kp->ki_oncpu_old = NOCPU_OLD; else if (td->td_oncpu > MAXCPU_OLD) kp->ki_oncpu_old = MAXCPU_OLD; else kp->ki_oncpu_old = td->td_oncpu; kp->ki_lastcpu = td->td_lastcpu; kp->ki_oncpu = td->td_oncpu; kp->ki_tdflags = td->td_flags; kp->ki_tid = td->td_tid; kp->ki_numthreads = p->p_numthreads; kp->ki_pcb = td->td_pcb; kp->ki_kstack = (void *)td->td_kstack; kp->ki_slptime = (ticks - td->td_slptick) / hz; kp->ki_pri.pri_class = td->td_pri_class; kp->ki_pri.pri_user = td->td_user_pri; if (preferthread) { rufetchtd(td, &kp->ki_rusage); kp->ki_runtime = cputick2usec(td->td_rux.rux_runtime); kp->ki_pctcpu = sched_pctcpu(td); kp->ki_estcpu = td->td_estcpu; kp->ki_cow = td->td_cow; } /* We can't get this anymore but ps etc never used it anyway. */ kp->ki_rqindex = 0; if (preferthread) kp->ki_siglist = td->td_siglist; kp->ki_sigmask = td->td_sigmask; thread_unlock(td); if (preferthread) PROC_STATUNLOCK(p); } /* * Fill in a kinfo_proc structure for the specified process. * Must be called with the target process locked. */ void fill_kinfo_proc(struct proc *p, struct kinfo_proc *kp) { MPASS(FIRST_THREAD_IN_PROC(p) != NULL); fill_kinfo_proc_only(p, kp); fill_kinfo_thread(FIRST_THREAD_IN_PROC(p), kp, 0); fill_kinfo_aggregate(p, kp); } struct pstats * pstats_alloc(void) { return (malloc(sizeof(struct pstats), M_SUBPROC, M_ZERO|M_WAITOK)); } /* * Copy parts of p_stats; zero the rest of p_stats (statistics). */ void pstats_fork(struct pstats *src, struct pstats *dst) { bzero(&dst->pstat_startzero, __rangeof(struct pstats, pstat_startzero, pstat_endzero)); bcopy(&src->pstat_startcopy, &dst->pstat_startcopy, __rangeof(struct pstats, pstat_startcopy, pstat_endcopy)); } void pstats_free(struct pstats *ps) { free(ps, M_SUBPROC); } static struct proc * zpfind_locked(pid_t pid) { struct proc *p; sx_assert(&allproc_lock, SX_LOCKED); LIST_FOREACH(p, &zombproc, p_list) { if (p->p_pid == pid) { PROC_LOCK(p); break; } } return (p); } /* * Locate a zombie process by number */ struct proc * zpfind(pid_t pid) { struct proc *p; sx_slock(&allproc_lock); p = zpfind_locked(pid); sx_sunlock(&allproc_lock); return (p); } #ifdef COMPAT_FREEBSD32 /* * This function is typically used to copy out the kernel address, so * it can be replaced by assignment of zero. */ static inline uint32_t ptr32_trim(void *ptr) { uintptr_t uptr; uptr = (uintptr_t)ptr; return ((uptr > UINT_MAX) ? 0 : uptr); } #define PTRTRIM_CP(src,dst,fld) \ do { (dst).fld = ptr32_trim((src).fld); } while (0) static void freebsd32_kinfo_proc_out(const struct kinfo_proc *ki, struct kinfo_proc32 *ki32) { int i; bzero(ki32, sizeof(struct kinfo_proc32)); ki32->ki_structsize = sizeof(struct kinfo_proc32); CP(*ki, *ki32, ki_layout); PTRTRIM_CP(*ki, *ki32, ki_args); PTRTRIM_CP(*ki, *ki32, ki_paddr); PTRTRIM_CP(*ki, *ki32, ki_addr); PTRTRIM_CP(*ki, *ki32, ki_tracep); PTRTRIM_CP(*ki, *ki32, ki_textvp); PTRTRIM_CP(*ki, *ki32, ki_fd); PTRTRIM_CP(*ki, *ki32, ki_vmspace); PTRTRIM_CP(*ki, *ki32, ki_wchan); CP(*ki, *ki32, ki_pid); CP(*ki, *ki32, ki_ppid); CP(*ki, *ki32, ki_pgid); CP(*ki, *ki32, ki_tpgid); CP(*ki, *ki32, ki_sid); CP(*ki, *ki32, ki_tsid); CP(*ki, *ki32, ki_jobc); CP(*ki, *ki32, ki_tdev); CP(*ki, *ki32, ki_siglist); CP(*ki, *ki32, ki_sigmask); CP(*ki, *ki32, ki_sigignore); CP(*ki, *ki32, ki_sigcatch); CP(*ki, *ki32, ki_uid); CP(*ki, *ki32, ki_ruid); CP(*ki, *ki32, ki_svuid); CP(*ki, *ki32, ki_rgid); CP(*ki, *ki32, ki_svgid); CP(*ki, *ki32, ki_ngroups); for (i = 0; i < KI_NGROUPS; i++) CP(*ki, *ki32, ki_groups[i]); CP(*ki, *ki32, ki_size); CP(*ki, *ki32, ki_rssize); CP(*ki, *ki32, ki_swrss); CP(*ki, *ki32, ki_tsize); CP(*ki, *ki32, ki_dsize); CP(*ki, *ki32, ki_ssize); CP(*ki, *ki32, ki_xstat); CP(*ki, *ki32, ki_acflag); CP(*ki, *ki32, ki_pctcpu); CP(*ki, *ki32, ki_estcpu); CP(*ki, *ki32, ki_slptime); CP(*ki, *ki32, ki_swtime); CP(*ki, *ki32, ki_cow); CP(*ki, *ki32, ki_runtime); TV_CP(*ki, *ki32, ki_start); TV_CP(*ki, *ki32, ki_childtime); CP(*ki, *ki32, ki_flag); CP(*ki, *ki32, ki_kiflag); CP(*ki, *ki32, ki_traceflag); CP(*ki, *ki32, ki_stat); CP(*ki, *ki32, ki_nice); CP(*ki, *ki32, ki_lock); CP(*ki, *ki32, ki_rqindex); CP(*ki, *ki32, ki_oncpu); CP(*ki, *ki32, ki_lastcpu); /* XXX TODO: wrap cpu value as appropriate */ CP(*ki, *ki32, ki_oncpu_old); CP(*ki, *ki32, ki_lastcpu_old); bcopy(ki->ki_tdname, ki32->ki_tdname, TDNAMLEN + 1); bcopy(ki->ki_wmesg, ki32->ki_wmesg, WMESGLEN + 1); bcopy(ki->ki_login, ki32->ki_login, LOGNAMELEN + 1); bcopy(ki->ki_lockname, ki32->ki_lockname, LOCKNAMELEN + 1); bcopy(ki->ki_comm, ki32->ki_comm, COMMLEN + 1); bcopy(ki->ki_emul, ki32->ki_emul, KI_EMULNAMELEN + 1); bcopy(ki->ki_loginclass, ki32->ki_loginclass, LOGINCLASSLEN + 1); CP(*ki, *ki32, ki_tracer); CP(*ki, *ki32, ki_flag2); CP(*ki, *ki32, ki_fibnum); CP(*ki, *ki32, ki_cr_flags); CP(*ki, *ki32, ki_jid); CP(*ki, *ki32, ki_numthreads); CP(*ki, *ki32, ki_tid); CP(*ki, *ki32, ki_pri); freebsd32_rusage_out(&ki->ki_rusage, &ki32->ki_rusage); freebsd32_rusage_out(&ki->ki_rusage_ch, &ki32->ki_rusage_ch); PTRTRIM_CP(*ki, *ki32, ki_pcb); PTRTRIM_CP(*ki, *ki32, ki_kstack); PTRTRIM_CP(*ki, *ki32, ki_udata); CP(*ki, *ki32, ki_sflag); CP(*ki, *ki32, ki_tdflags); } #endif int kern_proc_out(struct proc *p, struct sbuf *sb, int flags) { struct thread *td; struct kinfo_proc ki; #ifdef COMPAT_FREEBSD32 struct kinfo_proc32 ki32; #endif int error; PROC_LOCK_ASSERT(p, MA_OWNED); MPASS(FIRST_THREAD_IN_PROC(p) != NULL); error = 0; fill_kinfo_proc(p, &ki); if ((flags & KERN_PROC_NOTHREADS) != 0) { #ifdef COMPAT_FREEBSD32 if ((flags & KERN_PROC_MASK32) != 0) { freebsd32_kinfo_proc_out(&ki, &ki32); if (sbuf_bcat(sb, &ki32, sizeof(ki32)) != 0) error = ENOMEM; } else #endif if (sbuf_bcat(sb, &ki, sizeof(ki)) != 0) error = ENOMEM; } else { FOREACH_THREAD_IN_PROC(p, td) { fill_kinfo_thread(td, &ki, 1); #ifdef COMPAT_FREEBSD32 if ((flags & KERN_PROC_MASK32) != 0) { freebsd32_kinfo_proc_out(&ki, &ki32); if (sbuf_bcat(sb, &ki32, sizeof(ki32)) != 0) error = ENOMEM; } else #endif if (sbuf_bcat(sb, &ki, sizeof(ki)) != 0) error = ENOMEM; if (error != 0) break; } } PROC_UNLOCK(p); return (error); } static int sysctl_out_proc(struct proc *p, struct sysctl_req *req, int flags, int doingzomb) { struct sbuf sb; struct kinfo_proc ki; struct proc *np; int error, error2; pid_t pid; pid = p->p_pid; sbuf_new_for_sysctl(&sb, (char *)&ki, sizeof(ki), req); sbuf_clear_flags(&sb, SBUF_INCLUDENUL); error = kern_proc_out(p, &sb, flags); error2 = sbuf_finish(&sb); sbuf_delete(&sb); if (error != 0) return (error); else if (error2 != 0) return (error2); if (doingzomb) np = zpfind(pid); else { if (pid == 0) return (0); np = pfind(pid); } if (np == NULL) return (ESRCH); if (np != p) { PROC_UNLOCK(np); return (ESRCH); } PROC_UNLOCK(np); return (0); } static int sysctl_kern_proc(SYSCTL_HANDLER_ARGS) { int *name = (int *)arg1; u_int namelen = arg2; struct proc *p; int flags, doingzomb, oid_number; int error = 0; oid_number = oidp->oid_number; if (oid_number != KERN_PROC_ALL && (oid_number & KERN_PROC_INC_THREAD) == 0) flags = KERN_PROC_NOTHREADS; else { flags = 0; oid_number &= ~KERN_PROC_INC_THREAD; } #ifdef COMPAT_FREEBSD32 if (req->flags & SCTL_MASK32) flags |= KERN_PROC_MASK32; #endif if (oid_number == KERN_PROC_PID) { if (namelen != 1) return (EINVAL); error = sysctl_wire_old_buffer(req, 0); if (error) return (error); sx_slock(&proctree_lock); error = pget((pid_t)name[0], PGET_CANSEE, &p); if (error == 0) error = sysctl_out_proc(p, req, flags, 0); sx_sunlock(&proctree_lock); return (error); } switch (oid_number) { case KERN_PROC_ALL: if (namelen != 0) return (EINVAL); break; case KERN_PROC_PROC: if (namelen != 0 && namelen != 1) return (EINVAL); break; default: if (namelen != 1) return (EINVAL); break; } if (!req->oldptr) { /* overestimate by 5 procs */ error = SYSCTL_OUT(req, 0, sizeof (struct kinfo_proc) * 5); if (error) return (error); } error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); sx_slock(&proctree_lock); sx_slock(&allproc_lock); for (doingzomb=0 ; doingzomb < 2 ; doingzomb++) { if (!doingzomb) p = LIST_FIRST(&allproc); else p = LIST_FIRST(&zombproc); for (; p != 0; 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_mem(struct thread *td, struct proc *p, vm_offset_t offset, void* buf, size_t len) { struct iovec iov; struct uio uio; iov.iov_base = (caddr_t)buf; iov.iov_len = len; uio.uio_iov = &iov; uio.uio_iovcnt = 1; uio.uio_offset = offset; uio.uio_resid = (ssize_t)len; uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_READ; uio.uio_td = td; return (proc_rwmem(p, &uio)); } static int proc_read_string(struct thread *td, struct proc *p, const char *sptr, char *buf, size_t len) { size_t i; int error; error = proc_read_mem(td, p, (vm_offset_t)sptr, buf, len); /* * Reading the chunk may validly return EFAULT if the string is shorter * than the chunk and is aligned at the end of the page, assuming the * next page is not mapped. So if EFAULT is returned do a fallback to * one byte read loop. */ if (error == EFAULT) { for (i = 0; i < len; i++, buf++, sptr++) { error = proc_read_mem(td, p, (vm_offset_t)sptr, buf, 1); if (error != 0) return (error); if (*buf == '\0') break; } error = 0; } return (error); } #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 = proc_read_mem(td, p, (vm_offset_t)(p->p_sysent->sv_psstrings), &pss, sizeof(pss)); if (error != 0) return (error); 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++) { error = proc_read_mem(td, p, ptr, &aux, sizeof(aux)); if (error != 0) return (error); 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); error = proc_read_mem(td, p, vptr, proc_vector32, size); if (error != 0) 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 error, i; #ifdef COMPAT_FREEBSD32 if (SV_PROC_FLAG(p, SV_ILP32) != 0) return (get_proc_vector32(td, p, proc_vectorp, vsizep, type)); #endif error = proc_read_mem(td, p, (vm_offset_t)(p->p_sysent->sv_psstrings), &pss, sizeof(pss)); if (error != 0) return (error); 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++) { error = proc_read_mem(td, p, ptr, &aux, sizeof(aux)); if (error != 0) return (error); 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_vector == NULL) return (ENOMEM); error = proc_read_mem(td, p, vptr, proc_vector, size); if (error != 0) { free(proc_vector, M_TEMP); return (error); } *proc_vectorp = proc_vector; *vsizep = vsize; return (0); } #define GET_PS_STRINGS_CHUNK_SZ 256 /* Chunk size (bytes) for ps_strings operations. */ static int get_ps_strings(struct thread *td, struct proc *p, struct sbuf *sb, enum proc_vector_type type) { size_t done, len, nchr, vsize; int error, i; char **proc_vector, *sptr; char pss_string[GET_PS_STRINGS_CHUNK_SZ]; PROC_ASSERT_HELD(p); /* * We are not going to read more than 2 * (PATH_MAX + ARG_MAX) bytes. */ nchr = 2 * (PATH_MAX + ARG_MAX); error = get_proc_vector(td, p, &proc_vector, &vsize, type); if (error != 0) return (error); for (done = 0, i = 0; i < (int)vsize && done < nchr; i++) { /* * The program may have scribbled into its argv array, e.g. to * remove some arguments. If that has happened, break out * before trying to read from NULL. */ if (proc_vector[i] == NULL) break; for (sptr = proc_vector[i]; ; sptr += GET_PS_STRINGS_CHUNK_SZ) { error = proc_read_string(td, p, sptr, pss_string, sizeof(pss_string)); if (error != 0) goto done; len = strnlen(pss_string, GET_PS_STRINGS_CHUNK_SZ); if (done + len >= nchr) len = nchr - done - 1; sbuf_bcat(sb, pss_string, len); if (len != GET_PS_STRINGS_CHUNK_SZ) break; done += GET_PS_STRINGS_CHUNK_SZ; } sbuf_bcat(sb, "", 1); done += len + 1; } done: free(proc_vector, M_TEMP); return (error); } int proc_getargv(struct thread *td, struct proc *p, struct sbuf *sb) { return (get_ps_strings(curthread, p, sb, PROC_ARG)); } int proc_getenvv(struct thread *td, struct proc *p, struct sbuf *sb) { return (get_ps_strings(curthread, p, sb, PROC_ENV)); } int proc_getauxv(struct thread *td, struct proc *p, struct sbuf *sb) { size_t vsize, size; char **auxv; int error; error = get_proc_vector(td, p, &auxv, &vsize, PROC_AUX); if (error == 0) { #ifdef COMPAT_FREEBSD32 if (SV_PROC_FLAG(p, SV_ILP32) != 0) size = vsize * sizeof(Elf32_Auxinfo); else #endif size = vsize * sizeof(Elf_Auxinfo); if (sbuf_bcat(sb, auxv, size) != 0) error = ENOMEM; free(auxv, M_TEMP); } return (error); } /* * This sysctl allows a process to retrieve the argument list or process * title for another process without groping around in the address space * of the other process. It also allow a process to set its own "process * title to a string of its own choice. */ static int sysctl_kern_proc_args(SYSCTL_HANDLER_ARGS) { int *name = (int *)arg1; u_int namelen = arg2; struct pargs *newpa, *pa; struct proc *p; struct sbuf sb; int flags, error = 0, error2; if (namelen != 1) return (EINVAL); flags = PGET_CANSEE; if (req->newptr != NULL) flags |= PGET_ISCURRENT; error = pget((pid_t)name[0], flags, &p); if (error) return (error); pa = p->p_args; if (pa != NULL) { pargs_hold(pa); PROC_UNLOCK(p); error = SYSCTL_OUT(req, pa->ar_args, pa->ar_length); pargs_drop(pa); } else if ((p->p_flag & (P_WEXIT | P_SYSTEM)) == 0) { _PHOLD(p); PROC_UNLOCK(p); sbuf_new_for_sysctl(&sb, NULL, GET_PS_STRINGS_CHUNK_SZ, req); sbuf_clear_flags(&sb, SBUF_INCLUDENUL); error = proc_getargv(curthread, p, &sb); error2 = sbuf_finish(&sb); PRELE(p); sbuf_delete(&sb); if (error == 0 && error2 != 0) error = error2; } else { PROC_UNLOCK(p); } if (error != 0 || req->newptr == NULL) return (error); if (req->newlen + sizeof(struct pargs) > ps_arg_cache_limit) return (ENOMEM); newpa = pargs_alloc(req->newlen); error = SYSCTL_IN(req, newpa->ar_args, req->newlen); if (error != 0) { pargs_free(newpa); return (error); } PROC_LOCK(p); pa = p->p_args; p->p_args = newpa; PROC_UNLOCK(p); pargs_drop(pa); return (0); } /* * This sysctl allows a process to retrieve environment of another process. */ static int sysctl_kern_proc_env(SYSCTL_HANDLER_ARGS) { int *name = (int *)arg1; u_int namelen = arg2; struct proc *p; struct sbuf sb; int error, error2; if (namelen != 1) return (EINVAL); error = pget((pid_t)name[0], PGET_WANTREAD, &p); if (error != 0) return (error); if ((p->p_flag & P_SYSTEM) != 0) { PRELE(p); return (0); } sbuf_new_for_sysctl(&sb, NULL, GET_PS_STRINGS_CHUNK_SZ, req); sbuf_clear_flags(&sb, SBUF_INCLUDENUL); error = proc_getenvv(curthread, p, &sb); error2 = sbuf_finish(&sb); PRELE(p); sbuf_delete(&sb); return (error != 0 ? error : error2); } /* * This sysctl allows a process to retrieve ELF auxiliary vector of * another process. */ static int sysctl_kern_proc_auxv(SYSCTL_HANDLER_ARGS) { int *name = (int *)arg1; u_int namelen = arg2; struct proc *p; struct sbuf sb; int error, error2; if (namelen != 1) return (EINVAL); error = pget((pid_t)name[0], PGET_WANTREAD, &p); if (error != 0) return (error); if ((p->p_flag & P_SYSTEM) != 0) { PRELE(p); return (0); } sbuf_new_for_sysctl(&sb, NULL, GET_PS_STRINGS_CHUNK_SZ, req); sbuf_clear_flags(&sb, SBUF_INCLUDENUL); error = proc_getauxv(curthread, p, &sb); error2 = sbuf_finish(&sb); PRELE(p); sbuf_delete(&sb); return (error != 0 ? error : error2); } /* * This sysctl allows a process to retrieve the path of the executable for * itself or another process. */ static int sysctl_kern_proc_pathname(SYSCTL_HANDLER_ARGS) { pid_t *pidp = (pid_t *)arg1; unsigned int arglen = arg2; struct proc *p; struct vnode *vp; char *retbuf, *freebuf; int error; if (arglen != 1) return (EINVAL); if (*pidp == -1) { /* -1 means this process */ p = req->td->td_proc; } else { error = pget(*pidp, PGET_CANSEE, &p); if (error != 0) return (error); } vp = p->p_textvp; if (vp == NULL) { if (*pidp != -1) PROC_UNLOCK(p); return (0); } vref(vp); if (*pidp != -1) PROC_UNLOCK(p); error = vn_fullpath(req->td, vp, &retbuf, &freebuf); vrele(vp); if (error) return (error); error = SYSCTL_OUT(req, retbuf, strlen(retbuf) + 1); free(freebuf, M_TEMP); return (error); } static int sysctl_kern_proc_sv_name(SYSCTL_HANDLER_ARGS) { struct proc *p; char *sv_name; int *name; int namelen; int error; namelen = arg2; if (namelen != 1) return (EINVAL); name = (int *)arg1; error = pget((pid_t)name[0], PGET_CANSEE, &p); if (error != 0) return (error); sv_name = p->p_sysent->sv_name; PROC_UNLOCK(p); return (sysctl_handle_string(oidp, sv_name, 0, req)); } #ifdef KINFO_OVMENTRY_SIZE CTASSERT(sizeof(struct kinfo_ovmentry) == KINFO_OVMENTRY_SIZE); #endif #ifdef COMPAT_FREEBSD7 static int sysctl_kern_proc_ovmmap(SYSCTL_HANDLER_ARGS) { vm_map_entry_t entry, tmp_entry; unsigned int last_timestamp; char *fullpath, *freepath; struct kinfo_ovmentry *kve; struct vattr va; struct ucred *cred; int error, *name; struct vnode *vp; struct proc *p; vm_map_t map; struct vmspace *vm; name = (int *)arg1; error = pget((pid_t)name[0], PGET_WANTREAD, &p); if (error != 0) return (error); vm = vmspace_acquire_ref(p); if (vm == NULL) { PRELE(p); return (ESRCH); } kve = malloc(sizeof(*kve), M_TEMP, M_WAITOK); map = &vm->vm_map; vm_map_lock_read(map); for (entry = map->header.next; entry != &map->header; entry = entry->next) { vm_object_t obj, tobj, lobj; vm_offset_t addr; if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) continue; bzero(kve, sizeof(*kve)); kve->kve_structsize = sizeof(*kve); kve->kve_private_resident = 0; obj = entry->object.vm_object; if (obj != NULL) { VM_OBJECT_RLOCK(obj); if (obj->shadow_count == 1) kve->kve_private_resident = obj->resident_page_count; } kve->kve_resident = 0; addr = entry->start; while (addr < entry->end) { if (pmap_extract(map->pmap, addr)) kve->kve_resident++; addr += PAGE_SIZE; } for (lobj = tobj = obj; tobj; tobj = tobj->backing_object) { if (tobj != obj) VM_OBJECT_RLOCK(tobj); if (lobj != obj) VM_OBJECT_RUNLOCK(lobj); lobj = tobj; } kve->kve_start = (void*)entry->start; kve->kve_end = (void*)entry->end; kve->kve_offset = (off_t)entry->offset; if (entry->protection & VM_PROT_READ) kve->kve_protection |= KVME_PROT_READ; if (entry->protection & VM_PROT_WRITE) kve->kve_protection |= KVME_PROT_WRITE; if (entry->protection & VM_PROT_EXECUTE) kve->kve_protection |= KVME_PROT_EXEC; if (entry->eflags & MAP_ENTRY_COW) kve->kve_flags |= KVME_FLAG_COW; if (entry->eflags & MAP_ENTRY_NEEDS_COPY) kve->kve_flags |= KVME_FLAG_NEEDS_COPY; if (entry->eflags & MAP_ENTRY_NOCOREDUMP) kve->kve_flags |= KVME_FLAG_NOCOREDUMP; last_timestamp = map->timestamp; vm_map_unlock_read(map); kve->kve_fileid = 0; kve->kve_fsid = 0; freepath = NULL; fullpath = ""; if (lobj) { vp = NULL; switch (lobj->type) { case OBJT_DEFAULT: kve->kve_type = KVME_TYPE_DEFAULT; break; case OBJT_VNODE: kve->kve_type = KVME_TYPE_VNODE; vp = lobj->handle; vref(vp); break; case OBJT_SWAP: if ((lobj->flags & OBJ_TMPFS_NODE) != 0) { kve->kve_type = KVME_TYPE_VNODE; if ((lobj->flags & OBJ_TMPFS) != 0) { vp = lobj->un_pager.swp.swp_tmpfs; vref(vp); } } else { kve->kve_type = KVME_TYPE_SWAP; } break; case OBJT_DEVICE: kve->kve_type = KVME_TYPE_DEVICE; break; case OBJT_PHYS: kve->kve_type = KVME_TYPE_PHYS; break; case OBJT_DEAD: kve->kve_type = KVME_TYPE_DEAD; break; case OBJT_SG: kve->kve_type = KVME_TYPE_SG; break; default: kve->kve_type = KVME_TYPE_UNKNOWN; break; } if (lobj != obj) VM_OBJECT_RUNLOCK(lobj); kve->kve_ref_count = obj->ref_count; kve->kve_shadow_count = obj->shadow_count; VM_OBJECT_RUNLOCK(obj); if (vp != NULL) { vn_fullpath(curthread, vp, &fullpath, &freepath); cred = curthread->td_ucred; vn_lock(vp, LK_SHARED | LK_RETRY); if (VOP_GETATTR(vp, &va, cred) == 0) { kve->kve_fileid = va.va_fileid; kve->kve_fsid = va.va_fsid; } vput(vp); } } else { kve->kve_type = KVME_TYPE_NONE; kve->kve_ref_count = 0; kve->kve_shadow_count = 0; } strlcpy(kve->kve_path, fullpath, sizeof(kve->kve_path)); if (freepath != NULL) free(freepath, M_TEMP); error = SYSCTL_OUT(req, kve, sizeof(*kve)); vm_map_lock_read(map); if (error) break; if (last_timestamp != map->timestamp) { vm_map_lookup_entry(map, addr - 1, &tmp_entry); entry = tmp_entry; } } vm_map_unlock_read(map); vmspace_free(vm); PRELE(p); free(kve, M_TEMP); return (error); } #endif /* COMPAT_FREEBSD7 */ #ifdef KINFO_VMENTRY_SIZE CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE); #endif static void kern_proc_vmmap_resident(vm_map_t map, vm_map_entry_t entry, struct kinfo_vmentry *kve) { vm_object_t obj, tobj; vm_page_t m, m_adv; vm_offset_t addr; vm_paddr_t locked_pa; vm_pindex_t pi, pi_adv, pindex; locked_pa = 0; obj = entry->object.vm_object; addr = entry->start; m_adv = NULL; pi = OFF_TO_IDX(entry->offset); for (; addr < entry->end; addr += IDX_TO_OFF(pi_adv), pi += pi_adv) { if (m_adv != NULL) { m = m_adv; } else { pi_adv = OFF_TO_IDX(entry->end - addr); pindex = pi; for (tobj = obj;; tobj = tobj->backing_object) { m = vm_page_find_least(tobj, pindex); if (m != NULL) { if (m->pindex == pindex) break; if (pi_adv > m->pindex - pindex) { pi_adv = m->pindex - pindex; m_adv = m; } } if (tobj->backing_object == NULL) goto next; pindex += OFF_TO_IDX(tobj-> backing_object_offset); } } m_adv = NULL; if (m->psind != 0 && addr + pagesizes[1] <= entry->end && (addr & (pagesizes[1] - 1)) == 0 && (pmap_mincore(map->pmap, addr, &locked_pa) & MINCORE_SUPER) != 0) { kve->kve_flags |= KVME_FLAG_SUPER; pi_adv = OFF_TO_IDX(pagesizes[1]); } else { /* * We do not test the found page on validity. * Either the page is busy and being paged in, * or it was invalidated. The first case * should be counted as resident, the second * is not so clear; we do account both. */ pi_adv = 1; } kve->kve_resident += pi_adv; next:; } PA_UNLOCK_COND(locked_pa); } /* * Must be called with the process locked and will return unlocked. */ int kern_proc_vmmap_out(struct proc *p, struct sbuf *sb) { 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); 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 */ kve->kve_structsize = offsetof(struct kinfo_vmentry, kve_path) + strlen(kve->kve_path) + 1; kve->kve_structsize = roundup(kve->kve_structsize, sizeof(uint64_t)); 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); 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; numthreads = 0; PROC_LOCK(p); repeat: if (numthreads < p->p_numthreads) { 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); goto repeat; } i = 0; /* * 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. */ 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)) 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(p, which, &rlim); + 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; void stop_all_proc(void) { struct proc *cp, *p; int r, gen; bool restart, seen_stopped, seen_exiting, stopped_some; cp = curproc; /* * stop_all_proc() assumes that all process which have * usermode must be stopped, except current process, for * obvious reasons. Since other threads in the process * establishing global stop could unstop something, disable * calls from multithreaded processes as precaution. The * service must not be user-callable anyway. */ KASSERT((cp->p_flag & P_HADTHREADS) == 0 || (cp->p_flag & P_KTHREAD) != 0, ("mt stop_all_proc")); 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_KTHREAD | P_SYSTEM | P_TOTAL_STOP)) != 0) { PROC_UNLOCK(p); continue; } if ((p->p_flag & P_WEXIT) != 0) { seen_exiting = true; PROC_UNLOCK(p); continue; } if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { /* * Stopped processes are tolerated when there * are no other processes which might continue * them. P_STOPPED_SINGLE but not * P_TOTAL_STOP process still has at least one * thread running. */ seen_stopped = true; PROC_UNLOCK(p); continue; } _PHOLD(p); sx_xunlock(&allproc_lock); r = thread_single(p, SINGLE_ALLPROC); if (r != 0) restart = true; else stopped_some = true; _PRELE(p); PROC_UNLOCK(p); sx_xlock(&allproc_lock); } /* Catch forked children we did not see in iteration. */ if (gen != allproc_gen) restart = true; sx_xunlock(&allproc_lock); if (restart || stopped_some || seen_exiting || seen_stopped) { kern_yield(PRI_USER); goto allproc_loop; } } void resume_all_proc(void) { struct proc *cp, *p; cp = curproc; sx_xlock(&allproc_lock); LIST_REMOVE(cp, p_list); LIST_INSERT_HEAD(&allproc, cp, p_list); for (;;) { p = LIST_NEXT(cp, p_list); if (p == NULL) break; LIST_REMOVE(cp, p_list); LIST_INSERT_AFTER(p, cp, p_list); PROC_LOCK(p); if ((p->p_flag & P_TOTAL_STOP) != 0) { sx_xunlock(&allproc_lock); _PHOLD(p); thread_single_end(p, SINGLE_ALLPROC); _PRELE(p); PROC_UNLOCK(p); sx_xlock(&allproc_lock); } else { PROC_UNLOCK(p); } } sx_xunlock(&allproc_lock); } #define TOTAL_STOP_DEBUG 1 #ifdef TOTAL_STOP_DEBUG volatile static int ap_resume; #include static int sysctl_debug_stop_all_proc(SYSCTL_HANDLER_ARGS) { int error, val; val = 0; ap_resume = 0; error = sysctl_handle_int(oidp, &val, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (val != 0) { stop_all_proc(); syncer_suspend(); while (ap_resume == 0) ; syncer_resume(); resume_all_proc(); } return (0); } SYSCTL_PROC(_debug, OID_AUTO, stop_all_proc, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, __DEVOLATILE(int *, &ap_resume), 0, sysctl_debug_stop_all_proc, "I", ""); #endif Index: head/sys/kern/kern_resource.c =================================================================== --- head/sys/kern/kern_resource.c (revision 284214) +++ head/sys/kern/kern_resource.c (revision 284215) @@ -1,1443 +1,1471 @@ /*- * Copyright (c) 1982, 1986, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_resource.c 8.5 (Berkeley) 1/21/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static MALLOC_DEFINE(M_PLIMIT, "plimit", "plimit structures"); static MALLOC_DEFINE(M_UIDINFO, "uidinfo", "uidinfo structures"); #define UIHASH(uid) (&uihashtbl[(uid) & uihash]) static struct rwlock uihashtbl_lock; static LIST_HEAD(uihashhead, uidinfo) *uihashtbl; static u_long uihash; /* size of hash table - 1 */ static void calcru1(struct proc *p, struct rusage_ext *ruxp, struct timeval *up, struct timeval *sp); static int donice(struct thread *td, struct proc *chgp, int n); static struct uidinfo *uilookup(uid_t uid); static void ruxagg_locked(struct rusage_ext *rux, struct thread *td); static __inline int lim_shared(struct plimit *limp); /* * Resource controls and accounting. */ #ifndef _SYS_SYSPROTO_H_ struct getpriority_args { int which; int who; }; #endif int sys_getpriority(struct thread *td, register struct getpriority_args *uap) { struct proc *p; struct pgrp *pg; int error, low; error = 0; low = PRIO_MAX + 1; switch (uap->which) { case PRIO_PROCESS: if (uap->who == 0) low = td->td_proc->p_nice; else { p = pfind(uap->who); if (p == NULL) break; if (p_cansee(td, p) == 0) low = p->p_nice; PROC_UNLOCK(p); } break; case PRIO_PGRP: sx_slock(&proctree_lock); if (uap->who == 0) { pg = td->td_proc->p_pgrp; PGRP_LOCK(pg); } else { pg = pgfind(uap->who); if (pg == NULL) { sx_sunlock(&proctree_lock); break; } } sx_sunlock(&proctree_lock); LIST_FOREACH(p, &pg->pg_members, p_pglist) { PROC_LOCK(p); if (p->p_state == PRS_NORMAL && p_cansee(td, p) == 0) { if (p->p_nice < low) low = p->p_nice; } PROC_UNLOCK(p); } PGRP_UNLOCK(pg); break; case PRIO_USER: if (uap->who == 0) uap->who = td->td_ucred->cr_uid; sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); if (p->p_state == PRS_NORMAL && p_cansee(td, p) == 0 && p->p_ucred->cr_uid == uap->who) { if (p->p_nice < low) low = p->p_nice; } PROC_UNLOCK(p); } sx_sunlock(&allproc_lock); break; default: error = EINVAL; break; } if (low == PRIO_MAX + 1 && error == 0) error = ESRCH; td->td_retval[0] = low; return (error); } #ifndef _SYS_SYSPROTO_H_ struct setpriority_args { int which; int who; int prio; }; #endif int sys_setpriority(struct thread *td, struct setpriority_args *uap) { struct proc *curp, *p; struct pgrp *pg; int found = 0, error = 0; curp = td->td_proc; switch (uap->which) { case PRIO_PROCESS: if (uap->who == 0) { PROC_LOCK(curp); error = donice(td, curp, uap->prio); PROC_UNLOCK(curp); } else { p = pfind(uap->who); if (p == NULL) break; error = p_cansee(td, p); if (error == 0) error = donice(td, p, uap->prio); PROC_UNLOCK(p); } found++; break; case PRIO_PGRP: sx_slock(&proctree_lock); if (uap->who == 0) { pg = curp->p_pgrp; PGRP_LOCK(pg); } else { pg = pgfind(uap->who); if (pg == NULL) { sx_sunlock(&proctree_lock); break; } } sx_sunlock(&proctree_lock); LIST_FOREACH(p, &pg->pg_members, p_pglist) { PROC_LOCK(p); if (p->p_state == PRS_NORMAL && p_cansee(td, p) == 0) { error = donice(td, p, uap->prio); found++; } PROC_UNLOCK(p); } PGRP_UNLOCK(pg); break; case PRIO_USER: if (uap->who == 0) uap->who = td->td_ucred->cr_uid; sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); if (p->p_state == PRS_NORMAL && p->p_ucred->cr_uid == uap->who && p_cansee(td, p) == 0) { error = donice(td, p, uap->prio); found++; } PROC_UNLOCK(p); } sx_sunlock(&allproc_lock); break; default: error = EINVAL; break; } if (found == 0 && error == 0) error = ESRCH; return (error); } /* * Set "nice" for a (whole) process. */ static int donice(struct thread *td, struct proc *p, int n) { int error; PROC_LOCK_ASSERT(p, MA_OWNED); if ((error = p_cansched(td, p))) return (error); if (n > PRIO_MAX) n = PRIO_MAX; if (n < PRIO_MIN) n = PRIO_MIN; if (n < p->p_nice && priv_check(td, PRIV_SCHED_SETPRIORITY) != 0) return (EACCES); sched_nice(p, n); return (0); } static int unprivileged_idprio; SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_idprio, CTLFLAG_RW, &unprivileged_idprio, 0, "Allow non-root users to set an idle priority"); /* * Set realtime priority for LWP. */ #ifndef _SYS_SYSPROTO_H_ struct rtprio_thread_args { int function; lwpid_t lwpid; struct rtprio *rtp; }; #endif int sys_rtprio_thread(struct thread *td, struct rtprio_thread_args *uap) { struct proc *p; struct rtprio rtp; struct thread *td1; int cierror, error; /* Perform copyin before acquiring locks if needed. */ if (uap->function == RTP_SET) cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio)); else cierror = 0; if (uap->lwpid == 0 || uap->lwpid == td->td_tid) { p = td->td_proc; td1 = td; PROC_LOCK(p); } else { /* Only look up thread in current process */ td1 = tdfind(uap->lwpid, curproc->p_pid); if (td1 == NULL) return (ESRCH); p = td1->td_proc; } switch (uap->function) { case RTP_LOOKUP: if ((error = p_cansee(td, p))) break; pri_to_rtp(td1, &rtp); PROC_UNLOCK(p); return (copyout(&rtp, uap->rtp, sizeof(struct rtprio))); case RTP_SET: if ((error = p_cansched(td, p)) || (error = cierror)) break; /* Disallow setting rtprio in most cases if not superuser. */ /* * Realtime priority has to be restricted for reasons which * should be obvious. However, for idleprio processes, there is * a potential for system deadlock if an idleprio process gains * a lock on a resource that other processes need (and the * idleprio process can't run due to a CPU-bound normal * process). Fix me! XXX * * This problem is not only related to idleprio process. * A user level program can obtain a file lock and hold it * indefinitely. Additionally, without idleprio processes it is * still conceivable that a program with low priority will never * get to run. In short, allowing this feature might make it * easier to lock a resource indefinitely, but it is not the * only thing that makes it possible. */ if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_REALTIME || (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_IDLE && unprivileged_idprio == 0)) { error = priv_check(td, PRIV_SCHED_RTPRIO); if (error) break; } error = rtp_to_pri(&rtp, td1); break; default: error = EINVAL; break; } PROC_UNLOCK(p); return (error); } /* * Set realtime priority. */ #ifndef _SYS_SYSPROTO_H_ struct rtprio_args { int function; pid_t pid; struct rtprio *rtp; }; #endif int sys_rtprio(struct thread *td, register struct rtprio_args *uap) { struct proc *p; struct thread *tdp; struct rtprio rtp; int cierror, error; /* Perform copyin before acquiring locks if needed. */ if (uap->function == RTP_SET) cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio)); else cierror = 0; if (uap->pid == 0) { p = td->td_proc; PROC_LOCK(p); } else { p = pfind(uap->pid); if (p == NULL) return (ESRCH); } switch (uap->function) { case RTP_LOOKUP: if ((error = p_cansee(td, p))) break; /* * Return OUR priority if no pid specified, * or if one is, report the highest priority * in the process. There isn't much more you can do as * there is only room to return a single priority. * Note: specifying our own pid is not the same * as leaving it zero. */ if (uap->pid == 0) { pri_to_rtp(td, &rtp); } else { struct rtprio rtp2; rtp.type = RTP_PRIO_IDLE; rtp.prio = RTP_PRIO_MAX; FOREACH_THREAD_IN_PROC(p, tdp) { pri_to_rtp(tdp, &rtp2); if (rtp2.type < rtp.type || (rtp2.type == rtp.type && rtp2.prio < rtp.prio)) { rtp.type = rtp2.type; rtp.prio = rtp2.prio; } } } PROC_UNLOCK(p); return (copyout(&rtp, uap->rtp, sizeof(struct rtprio))); case RTP_SET: if ((error = p_cansched(td, p)) || (error = cierror)) break; /* * Disallow setting rtprio in most cases if not superuser. * See the comment in sys_rtprio_thread about idprio * threads holding a lock. */ if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_REALTIME || (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_IDLE && !unprivileged_idprio)) { error = priv_check(td, PRIV_SCHED_RTPRIO); if (error) break; } /* * If we are setting our own priority, set just our * thread but if we are doing another process, * do all the threads on that process. If we * specify our own pid we do the latter. */ if (uap->pid == 0) { error = rtp_to_pri(&rtp, td); } else { FOREACH_THREAD_IN_PROC(p, td) { if ((error = rtp_to_pri(&rtp, td)) != 0) break; } } break; default: error = EINVAL; break; } PROC_UNLOCK(p); return (error); } int rtp_to_pri(struct rtprio *rtp, struct thread *td) { u_char newpri, oldclass, oldpri; switch (RTP_PRIO_BASE(rtp->type)) { case RTP_PRIO_REALTIME: if (rtp->prio > RTP_PRIO_MAX) return (EINVAL); newpri = PRI_MIN_REALTIME + rtp->prio; break; case RTP_PRIO_NORMAL: if (rtp->prio > (PRI_MAX_TIMESHARE - PRI_MIN_TIMESHARE)) return (EINVAL); newpri = PRI_MIN_TIMESHARE + rtp->prio; break; case RTP_PRIO_IDLE: if (rtp->prio > RTP_PRIO_MAX) return (EINVAL); newpri = PRI_MIN_IDLE + rtp->prio; break; default: return (EINVAL); } thread_lock(td); oldclass = td->td_pri_class; sched_class(td, rtp->type); /* XXX fix */ oldpri = td->td_user_pri; sched_user_prio(td, newpri); if (td->td_user_pri != oldpri && (oldclass != RTP_PRIO_NORMAL || td->td_pri_class != RTP_PRIO_NORMAL)) sched_prio(td, td->td_user_pri); if (TD_ON_UPILOCK(td) && oldpri != newpri) { critical_enter(); thread_unlock(td); umtx_pi_adjust(td, oldpri); critical_exit(); } else thread_unlock(td); return (0); } void pri_to_rtp(struct thread *td, struct rtprio *rtp) { thread_lock(td); switch (PRI_BASE(td->td_pri_class)) { case PRI_REALTIME: rtp->prio = td->td_base_user_pri - PRI_MIN_REALTIME; break; case PRI_TIMESHARE: rtp->prio = td->td_base_user_pri - PRI_MIN_TIMESHARE; break; case PRI_IDLE: rtp->prio = td->td_base_user_pri - PRI_MIN_IDLE; break; default: break; } rtp->type = td->td_pri_class; thread_unlock(td); } #if defined(COMPAT_43) #ifndef _SYS_SYSPROTO_H_ struct osetrlimit_args { u_int which; struct orlimit *rlp; }; #endif int osetrlimit(struct thread *td, register struct osetrlimit_args *uap) { struct orlimit olim; struct rlimit lim; int error; if ((error = copyin(uap->rlp, &olim, sizeof(struct orlimit)))) return (error); lim.rlim_cur = olim.rlim_cur; lim.rlim_max = olim.rlim_max; error = kern_setrlimit(td, uap->which, &lim); return (error); } #ifndef _SYS_SYSPROTO_H_ struct ogetrlimit_args { u_int which; struct orlimit *rlp; }; #endif int ogetrlimit(struct thread *td, register struct ogetrlimit_args *uap) { struct orlimit olim; struct rlimit rl; - struct proc *p; int error; if (uap->which >= RLIM_NLIMITS) return (EINVAL); - p = td->td_proc; - PROC_LOCK(p); - lim_rlimit(p, uap->which, &rl); - PROC_UNLOCK(p); + lim_rlimit(td, uap->which, &rl); /* * XXX would be more correct to convert only RLIM_INFINITY to the * old RLIM_INFINITY and fail with EOVERFLOW for other larger * values. Most 64->32 and 32->16 conversions, including not * unimportant ones of uids are even more broken than what we * do here (they blindly truncate). We don't do this correctly * here since we have little experience with EOVERFLOW yet. * Elsewhere, getuid() can't fail... */ olim.rlim_cur = rl.rlim_cur > 0x7fffffff ? 0x7fffffff : rl.rlim_cur; olim.rlim_max = rl.rlim_max > 0x7fffffff ? 0x7fffffff : rl.rlim_max; error = copyout(&olim, uap->rlp, sizeof(olim)); return (error); } #endif /* COMPAT_43 */ #ifndef _SYS_SYSPROTO_H_ struct __setrlimit_args { u_int which; struct rlimit *rlp; }; #endif int sys_setrlimit(struct thread *td, register struct __setrlimit_args *uap) { struct rlimit alim; int error; if ((error = copyin(uap->rlp, &alim, sizeof(struct rlimit)))) return (error); error = kern_setrlimit(td, uap->which, &alim); return (error); } static void lim_cb(void *arg) { struct rlimit rlim; struct thread *td; struct proc *p; p = arg; PROC_LOCK_ASSERT(p, MA_OWNED); /* * Check if the process exceeds its cpu resource allocation. If * it reaches the max, arrange to kill the process in ast(). */ if (p->p_cpulimit == RLIM_INFINITY) return; PROC_STATLOCK(p); FOREACH_THREAD_IN_PROC(p, td) { ruxagg(p, td); } PROC_STATUNLOCK(p); if (p->p_rux.rux_runtime > p->p_cpulimit * cpu_tickrate()) { - lim_rlimit(p, RLIMIT_CPU, &rlim); + lim_rlimit_proc(p, RLIMIT_CPU, &rlim); if (p->p_rux.rux_runtime >= rlim.rlim_max * cpu_tickrate()) { killproc(p, "exceeded maximum CPU limit"); } else { if (p->p_cpulimit < rlim.rlim_max) p->p_cpulimit += 5; kern_psignal(p, SIGXCPU); } } if ((p->p_flag & P_WEXIT) == 0) callout_reset_sbt(&p->p_limco, SBT_1S, 0, lim_cb, p, C_PREL(1)); } int kern_setrlimit(struct thread *td, u_int which, struct rlimit *limp) { return (kern_proc_setrlimit(td, td->td_proc, which, limp)); } int kern_proc_setrlimit(struct thread *td, struct proc *p, u_int which, struct rlimit *limp) { struct plimit *newlim, *oldlim; register struct rlimit *alimp; struct rlimit oldssiz; int error; if (which >= RLIM_NLIMITS) return (EINVAL); /* * Preserve historical bugs by treating negative limits as unsigned. */ if (limp->rlim_cur < 0) limp->rlim_cur = RLIM_INFINITY; if (limp->rlim_max < 0) limp->rlim_max = RLIM_INFINITY; oldssiz.rlim_cur = 0; - newlim = NULL; + newlim = lim_alloc(); PROC_LOCK(p); - if (lim_shared(p->p_limit)) { - PROC_UNLOCK(p); - newlim = lim_alloc(); - PROC_LOCK(p); - } oldlim = p->p_limit; alimp = &oldlim->pl_rlimit[which]; if (limp->rlim_cur > alimp->rlim_max || limp->rlim_max > alimp->rlim_max) if ((error = priv_check(td, PRIV_PROC_SETRLIMIT))) { PROC_UNLOCK(p); - if (newlim != NULL) - lim_free(newlim); + lim_free(newlim); return (error); } if (limp->rlim_cur > limp->rlim_max) limp->rlim_cur = limp->rlim_max; - if (newlim != NULL) { - lim_copy(newlim, oldlim); - alimp = &newlim->pl_rlimit[which]; - } + lim_copy(newlim, oldlim); + alimp = &newlim->pl_rlimit[which]; switch (which) { case RLIMIT_CPU: if (limp->rlim_cur != RLIM_INFINITY && p->p_cpulimit == RLIM_INFINITY) callout_reset_sbt(&p->p_limco, SBT_1S, 0, lim_cb, p, C_PREL(1)); p->p_cpulimit = limp->rlim_cur; break; case RLIMIT_DATA: if (limp->rlim_cur > maxdsiz) limp->rlim_cur = maxdsiz; if (limp->rlim_max > maxdsiz) limp->rlim_max = maxdsiz; break; case RLIMIT_STACK: if (limp->rlim_cur > maxssiz) limp->rlim_cur = maxssiz; if (limp->rlim_max > maxssiz) limp->rlim_max = maxssiz; oldssiz = *alimp; if (p->p_sysent->sv_fixlimit != NULL) p->p_sysent->sv_fixlimit(&oldssiz, RLIMIT_STACK); break; case RLIMIT_NOFILE: if (limp->rlim_cur > maxfilesperproc) limp->rlim_cur = maxfilesperproc; if (limp->rlim_max > maxfilesperproc) limp->rlim_max = maxfilesperproc; break; case RLIMIT_NPROC: if (limp->rlim_cur > maxprocperuid) limp->rlim_cur = maxprocperuid; if (limp->rlim_max > maxprocperuid) limp->rlim_max = maxprocperuid; if (limp->rlim_cur < 1) limp->rlim_cur = 1; if (limp->rlim_max < 1) limp->rlim_max = 1; break; } if (p->p_sysent->sv_fixlimit != NULL) p->p_sysent->sv_fixlimit(limp, which); *alimp = *limp; - if (newlim != NULL) - p->p_limit = newlim; + p->p_limit = newlim; + PROC_UPDATE_COW(p); PROC_UNLOCK(p); - if (newlim != NULL) - lim_free(oldlim); + lim_free(oldlim); if (which == RLIMIT_STACK && /* * Skip calls from exec_new_vmspace(), done when stack is * not mapped yet. */ (td != curthread || (p->p_flag & P_INEXEC) == 0)) { /* * Stack is allocated to the max at exec time with only * "rlim_cur" bytes accessible. If stack limit is going * up make more accessible, if going down make inaccessible. */ if (limp->rlim_cur != oldssiz.rlim_cur) { vm_offset_t addr; vm_size_t size; vm_prot_t prot; if (limp->rlim_cur > oldssiz.rlim_cur) { prot = p->p_sysent->sv_stackprot; size = limp->rlim_cur - oldssiz.rlim_cur; addr = p->p_sysent->sv_usrstack - limp->rlim_cur; } else { prot = VM_PROT_NONE; size = oldssiz.rlim_cur - limp->rlim_cur; addr = p->p_sysent->sv_usrstack - oldssiz.rlim_cur; } addr = trunc_page(addr); size = round_page(size); (void)vm_map_protect(&p->p_vmspace->vm_map, addr, addr + size, prot, FALSE); } } return (0); } #ifndef _SYS_SYSPROTO_H_ struct __getrlimit_args { u_int which; struct rlimit *rlp; }; #endif /* ARGSUSED */ int sys_getrlimit(struct thread *td, register struct __getrlimit_args *uap) { struct rlimit rlim; - struct proc *p; int error; if (uap->which >= RLIM_NLIMITS) return (EINVAL); - p = td->td_proc; - PROC_LOCK(p); - lim_rlimit(p, uap->which, &rlim); - PROC_UNLOCK(p); + lim_rlimit(td, uap->which, &rlim); error = copyout(&rlim, uap->rlp, sizeof(struct rlimit)); return (error); } /* * Transform the running time and tick information for children of proc p * into user and system time usage. */ void calccru(struct proc *p, struct timeval *up, struct timeval *sp) { PROC_LOCK_ASSERT(p, MA_OWNED); calcru1(p, &p->p_crux, up, sp); } /* * Transform the running time and tick information in proc p into user * and system time usage. If appropriate, include the current time slice * on this CPU. */ void calcru(struct proc *p, struct timeval *up, struct timeval *sp) { struct thread *td; uint64_t runtime, u; PROC_LOCK_ASSERT(p, MA_OWNED); PROC_STATLOCK_ASSERT(p, MA_OWNED); /* * If we are getting stats for the current process, then add in the * stats that this thread has accumulated in its current time slice. * We reset the thread and CPU state as if we had performed a context * switch right here. */ td = curthread; if (td->td_proc == p) { u = cpu_ticks(); runtime = u - PCPU_GET(switchtime); td->td_runtime += runtime; td->td_incruntime += runtime; PCPU_SET(switchtime, u); } /* Make sure the per-thread stats are current. */ FOREACH_THREAD_IN_PROC(p, td) { if (td->td_incruntime == 0) continue; ruxagg(p, td); } calcru1(p, &p->p_rux, up, sp); } /* Collect resource usage for a single thread. */ void rufetchtd(struct thread *td, struct rusage *ru) { struct proc *p; uint64_t runtime, u; p = td->td_proc; PROC_STATLOCK_ASSERT(p, MA_OWNED); THREAD_LOCK_ASSERT(td, MA_OWNED); /* * If we are getting stats for the current thread, then add in the * stats that this thread has accumulated in its current time slice. * We reset the thread and CPU state as if we had performed a context * switch right here. */ if (td == curthread) { u = cpu_ticks(); runtime = u - PCPU_GET(switchtime); td->td_runtime += runtime; td->td_incruntime += runtime; PCPU_SET(switchtime, u); } ruxagg(p, td); *ru = td->td_ru; calcru1(p, &td->td_rux, &ru->ru_utime, &ru->ru_stime); } static void calcru1(struct proc *p, struct rusage_ext *ruxp, struct timeval *up, struct timeval *sp) { /* {user, system, interrupt, total} {ticks, usec}: */ uint64_t ut, uu, st, su, it, tt, tu; ut = ruxp->rux_uticks; st = ruxp->rux_sticks; it = ruxp->rux_iticks; tt = ut + st + it; if (tt == 0) { /* Avoid divide by zero */ st = 1; tt = 1; } tu = cputick2usec(ruxp->rux_runtime); if ((int64_t)tu < 0) { /* XXX: this should be an assert /phk */ printf("calcru: negative runtime of %jd usec for pid %d (%s)\n", (intmax_t)tu, p->p_pid, p->p_comm); tu = ruxp->rux_tu; } if (tu >= ruxp->rux_tu) { /* * The normal case, time increased. * Enforce monotonicity of bucketed numbers. */ uu = (tu * ut) / tt; if (uu < ruxp->rux_uu) uu = ruxp->rux_uu; su = (tu * st) / tt; if (su < ruxp->rux_su) su = ruxp->rux_su; } else if (tu + 3 > ruxp->rux_tu || 101 * tu > 100 * ruxp->rux_tu) { /* * When we calibrate the cputicker, it is not uncommon to * see the presumably fixed frequency increase slightly over * time as a result of thermal stabilization and NTP * discipline (of the reference clock). We therefore ignore * a bit of backwards slop because we expect to catch up * shortly. We use a 3 microsecond limit to catch low * counts and a 1% limit for high counts. */ uu = ruxp->rux_uu; su = ruxp->rux_su; tu = ruxp->rux_tu; } else { /* tu < ruxp->rux_tu */ /* * What happened here was likely that a laptop, which ran at * a reduced clock frequency at boot, kicked into high gear. * The wisdom of spamming this message in that case is * dubious, but it might also be indicative of something * serious, so lets keep it and hope laptops can be made * more truthful about their CPU speed via ACPI. */ printf("calcru: runtime went backwards from %ju usec " "to %ju usec for pid %d (%s)\n", (uintmax_t)ruxp->rux_tu, (uintmax_t)tu, p->p_pid, p->p_comm); uu = (tu * ut) / tt; su = (tu * st) / tt; } ruxp->rux_uu = uu; ruxp->rux_su = su; ruxp->rux_tu = tu; up->tv_sec = uu / 1000000; up->tv_usec = uu % 1000000; sp->tv_sec = su / 1000000; sp->tv_usec = su % 1000000; } #ifndef _SYS_SYSPROTO_H_ struct getrusage_args { int who; struct rusage *rusage; }; #endif int sys_getrusage(register struct thread *td, register struct getrusage_args *uap) { struct rusage ru; int error; error = kern_getrusage(td, uap->who, &ru); if (error == 0) error = copyout(&ru, uap->rusage, sizeof(struct rusage)); return (error); } int kern_getrusage(struct thread *td, int who, struct rusage *rup) { struct proc *p; int error; error = 0; p = td->td_proc; PROC_LOCK(p); switch (who) { case RUSAGE_SELF: rufetchcalc(p, rup, &rup->ru_utime, &rup->ru_stime); break; case RUSAGE_CHILDREN: *rup = p->p_stats->p_cru; calccru(p, &rup->ru_utime, &rup->ru_stime); break; case RUSAGE_THREAD: PROC_STATLOCK(p); thread_lock(td); rufetchtd(td, rup); thread_unlock(td); PROC_STATUNLOCK(p); break; default: error = EINVAL; } PROC_UNLOCK(p); return (error); } void rucollect(struct rusage *ru, struct rusage *ru2) { long *ip, *ip2; int i; if (ru->ru_maxrss < ru2->ru_maxrss) ru->ru_maxrss = ru2->ru_maxrss; ip = &ru->ru_first; ip2 = &ru2->ru_first; for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--) *ip++ += *ip2++; } void ruadd(struct rusage *ru, struct rusage_ext *rux, struct rusage *ru2, struct rusage_ext *rux2) { rux->rux_runtime += rux2->rux_runtime; rux->rux_uticks += rux2->rux_uticks; rux->rux_sticks += rux2->rux_sticks; rux->rux_iticks += rux2->rux_iticks; rux->rux_uu += rux2->rux_uu; rux->rux_su += rux2->rux_su; rux->rux_tu += rux2->rux_tu; rucollect(ru, ru2); } /* * Aggregate tick counts into the proc's rusage_ext. */ static void ruxagg_locked(struct rusage_ext *rux, struct thread *td) { THREAD_LOCK_ASSERT(td, MA_OWNED); PROC_STATLOCK_ASSERT(td->td_proc, MA_OWNED); rux->rux_runtime += td->td_incruntime; rux->rux_uticks += td->td_uticks; rux->rux_sticks += td->td_sticks; rux->rux_iticks += td->td_iticks; } void ruxagg(struct proc *p, struct thread *td) { thread_lock(td); ruxagg_locked(&p->p_rux, td); ruxagg_locked(&td->td_rux, td); td->td_incruntime = 0; td->td_uticks = 0; td->td_iticks = 0; td->td_sticks = 0; thread_unlock(td); } /* * Update the rusage_ext structure and fetch a valid aggregate rusage * for proc p if storage for one is supplied. */ void rufetch(struct proc *p, struct rusage *ru) { struct thread *td; PROC_STATLOCK_ASSERT(p, MA_OWNED); *ru = p->p_ru; if (p->p_numthreads > 0) { FOREACH_THREAD_IN_PROC(p, td) { ruxagg(p, td); rucollect(ru, &td->td_ru); } } } /* * Atomically perform a rufetch and a calcru together. * Consumers, can safely assume the calcru is executed only once * rufetch is completed. */ void rufetchcalc(struct proc *p, struct rusage *ru, struct timeval *up, struct timeval *sp) { PROC_STATLOCK(p); rufetch(p, ru); calcru(p, up, sp); PROC_STATUNLOCK(p); } /* * Allocate a new resource limits structure and initialize its * reference count and mutex pointer. */ struct plimit * lim_alloc() { struct plimit *limp; limp = malloc(sizeof(struct plimit), M_PLIMIT, M_WAITOK); refcount_init(&limp->pl_refcnt, 1); return (limp); } struct plimit * lim_hold(struct plimit *limp) { refcount_acquire(&limp->pl_refcnt); return (limp); } static __inline int lim_shared(struct plimit *limp) { return (limp->pl_refcnt > 1); } void lim_fork(struct proc *p1, struct proc *p2) { PROC_LOCK_ASSERT(p1, MA_OWNED); PROC_LOCK_ASSERT(p2, MA_OWNED); p2->p_limit = lim_hold(p1->p_limit); callout_init_mtx(&p2->p_limco, &p2->p_mtx, 0); if (p1->p_cpulimit != RLIM_INFINITY) callout_reset_sbt(&p2->p_limco, SBT_1S, 0, lim_cb, p2, C_PREL(1)); } void lim_free(struct plimit *limp) { if (refcount_release(&limp->pl_refcnt)) free((void *)limp, M_PLIMIT); } /* * Make a copy of the plimit structure. * We share these structures copy-on-write after fork. */ void lim_copy(struct plimit *dst, struct plimit *src) { KASSERT(!lim_shared(dst), ("lim_copy to shared limit")); bcopy(src->pl_rlimit, dst->pl_rlimit, sizeof(src->pl_rlimit)); } /* * Return the hard limit for a particular system resource. The * which parameter specifies the index into the rlimit array. */ rlim_t -lim_max(struct proc *p, int which) +lim_max(struct thread *td, int which) { struct rlimit rl; - lim_rlimit(p, which, &rl); + lim_rlimit(td, which, &rl); return (rl.rlim_max); } +rlim_t +lim_max_proc(struct proc *p, int which) +{ + struct rlimit rl; + + lim_rlimit_proc(p, which, &rl); + return (rl.rlim_max); +} + /* * Return the current (soft) limit for a particular system resource. * The which parameter which specifies the index into the rlimit array */ rlim_t -lim_cur(struct proc *p, int which) +lim_cur(struct thread *td, int which) { struct rlimit rl; - lim_rlimit(p, which, &rl); + lim_rlimit(td, which, &rl); return (rl.rlim_cur); } +rlim_t +lim_cur_proc(struct proc *p, int which) +{ + struct rlimit rl; + + lim_rlimit_proc(p, which, &rl); + return (rl.rlim_cur); +} + /* * Return a copy of the entire rlimit structure for the system limit * specified by 'which' in the rlimit structure pointed to by 'rlp'. */ void -lim_rlimit(struct proc *p, int which, struct rlimit *rlp) +lim_rlimit(struct thread *td, int which, struct rlimit *rlp) { + struct proc *p = td->td_proc; + MPASS(td == curthread); + KASSERT(which >= 0 && which < RLIM_NLIMITS, + ("request for invalid resource limit")); + *rlp = td->td_limit->pl_rlimit[which]; + if (p->p_sysent->sv_fixlimit != NULL) + p->p_sysent->sv_fixlimit(rlp, which); +} + +void +lim_rlimit_proc(struct proc *p, int which, struct rlimit *rlp) +{ + PROC_LOCK_ASSERT(p, MA_OWNED); KASSERT(which >= 0 && which < RLIM_NLIMITS, ("request for invalid resource limit")); *rlp = p->p_limit->pl_rlimit[which]; if (p->p_sysent->sv_fixlimit != NULL) p->p_sysent->sv_fixlimit(rlp, which); } void uihashinit() { uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash); rw_init(&uihashtbl_lock, "uidinfo hash"); } /* * Look up a uidinfo struct for the parameter uid. * uihashtbl_lock must be locked. * Increase refcount on uidinfo struct returned. */ static struct uidinfo * uilookup(uid_t uid) { struct uihashhead *uipp; struct uidinfo *uip; rw_assert(&uihashtbl_lock, RA_LOCKED); uipp = UIHASH(uid); LIST_FOREACH(uip, uipp, ui_hash) if (uip->ui_uid == uid) { uihold(uip); break; } return (uip); } /* * Find or allocate a struct uidinfo for a particular uid. * Returns with uidinfo struct referenced. * uifree() should be called on a struct uidinfo when released. */ struct uidinfo * uifind(uid_t uid) { struct uidinfo *new_uip, *uip; rw_rlock(&uihashtbl_lock); uip = uilookup(uid); rw_runlock(&uihashtbl_lock); if (uip != NULL) return (uip); new_uip = malloc(sizeof(*new_uip), M_UIDINFO, M_WAITOK | M_ZERO); racct_create(&new_uip->ui_racct); refcount_init(&new_uip->ui_ref, 1); new_uip->ui_uid = uid; mtx_init(&new_uip->ui_vmsize_mtx, "ui_vmsize", NULL, MTX_DEF); rw_wlock(&uihashtbl_lock); /* * There's a chance someone created our uidinfo while we * were in malloc and not holding the lock, so we have to * make sure we don't insert a duplicate uidinfo. */ if ((uip = uilookup(uid)) == NULL) { LIST_INSERT_HEAD(UIHASH(uid), new_uip, ui_hash); rw_wunlock(&uihashtbl_lock); uip = new_uip; } else { rw_wunlock(&uihashtbl_lock); racct_destroy(&new_uip->ui_racct); mtx_destroy(&new_uip->ui_vmsize_mtx); free(new_uip, M_UIDINFO); } return (uip); } /* * Place another refcount on a uidinfo struct. */ void uihold(struct uidinfo *uip) { refcount_acquire(&uip->ui_ref); } /*- * Since uidinfo structs have a long lifetime, we use an * opportunistic refcounting scheme to avoid locking the lookup hash * for each release. * * If the refcount hits 0, we need to free the structure, * which means we need to lock the hash. * Optimal case: * After locking the struct and lowering the refcount, if we find * that we don't need to free, simply unlock and return. * Suboptimal case: * If refcount lowering results in need to free, bump the count * back up, lose the lock and acquire the locks in the proper * order to try again. */ void uifree(struct uidinfo *uip) { int old; /* Prepare for optimal case. */ old = uip->ui_ref; if (old > 1 && atomic_cmpset_int(&uip->ui_ref, old, old - 1)) return; /* Prepare for suboptimal case. */ rw_wlock(&uihashtbl_lock); if (refcount_release(&uip->ui_ref) == 0) { rw_wunlock(&uihashtbl_lock); return; } racct_destroy(&uip->ui_racct); LIST_REMOVE(uip, ui_hash); rw_wunlock(&uihashtbl_lock); if (uip->ui_sbsize != 0) printf("freeing uidinfo: uid = %d, sbsize = %ld\n", uip->ui_uid, uip->ui_sbsize); if (uip->ui_proccnt != 0) printf("freeing uidinfo: uid = %d, proccnt = %ld\n", uip->ui_uid, uip->ui_proccnt); if (uip->ui_vmsize != 0) printf("freeing uidinfo: uid = %d, swapuse = %lld\n", uip->ui_uid, (unsigned long long)uip->ui_vmsize); mtx_destroy(&uip->ui_vmsize_mtx); free(uip, M_UIDINFO); } #ifdef RACCT void ui_racct_foreach(void (*callback)(struct racct *racct, void *arg2, void *arg3), void *arg2, void *arg3) { struct uidinfo *uip; struct uihashhead *uih; rw_rlock(&uihashtbl_lock); for (uih = &uihashtbl[uihash]; uih >= uihashtbl; uih--) { LIST_FOREACH(uip, uih, ui_hash) { (callback)(uip->ui_racct, arg2, arg3); } } rw_runlock(&uihashtbl_lock); } #endif /* * Change the count associated with number of processes * a given user is using. When 'max' is 0, don't enforce a limit */ int chgproccnt(struct uidinfo *uip, int diff, rlim_t max) { /* Don't allow them to exceed max, but allow subtraction. */ if (diff > 0 && max != 0) { if (atomic_fetchadd_long(&uip->ui_proccnt, (long)diff) + diff > max) { atomic_subtract_long(&uip->ui_proccnt, (long)diff); return (0); } } else { atomic_add_long(&uip->ui_proccnt, (long)diff); if (uip->ui_proccnt < 0) printf("negative proccnt for uid = %d\n", uip->ui_uid); } return (1); } /* * Change the total socket buffer size a user has used. */ int chgsbsize(struct uidinfo *uip, u_int *hiwat, u_int to, rlim_t max) { int diff; diff = to - *hiwat; if (diff > 0) { if (atomic_fetchadd_long(&uip->ui_sbsize, (long)diff) + diff > max) { atomic_subtract_long(&uip->ui_sbsize, (long)diff); return (0); } } else { atomic_add_long(&uip->ui_sbsize, (long)diff); if (uip->ui_sbsize < 0) printf("negative sbsize for uid = %d\n", uip->ui_uid); } *hiwat = to; return (1); } /* * Change the count associated with number of pseudo-terminals * a given user is using. When 'max' is 0, don't enforce a limit */ int chgptscnt(struct uidinfo *uip, int diff, rlim_t max) { /* Don't allow them to exceed max, but allow subtraction. */ if (diff > 0 && max != 0) { if (atomic_fetchadd_long(&uip->ui_ptscnt, (long)diff) + diff > max) { atomic_subtract_long(&uip->ui_ptscnt, (long)diff); return (0); } } else { atomic_add_long(&uip->ui_ptscnt, (long)diff); if (uip->ui_ptscnt < 0) printf("negative ptscnt for uid = %d\n", uip->ui_uid); } return (1); } int chgkqcnt(struct uidinfo *uip, int diff, rlim_t max) { if (diff > 0 && max != 0) { if (atomic_fetchadd_long(&uip->ui_kqcnt, (long)diff) + diff > max) { atomic_subtract_long(&uip->ui_kqcnt, (long)diff); return (0); } } else { atomic_add_long(&uip->ui_kqcnt, (long)diff); if (uip->ui_kqcnt < 0) printf("negative kqcnt for uid = %d\n", uip->ui_uid); } return (1); +} + +void +lim_update_thread(struct thread *td) +{ + struct proc *p; + struct plimit *lim; + + p = td->td_proc; + lim = td->td_limit; + PROC_LOCK_ASSERT(p, MA_OWNED); + td->td_limit = lim_hold(p->p_limit); + if (lim != NULL) + lim_free(lim); } Index: head/sys/kern/kern_sig.c =================================================================== --- head/sys/kern/kern_sig.c (revision 284214) +++ head/sys/kern/kern_sig.c (revision 284215) @@ -1,3557 +1,3557 @@ /*- * Copyright (c) 1982, 1986, 1989, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_sig.c 8.7 (Berkeley) 4/18/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #include "opt_gzio.h" #include "opt_ktrace.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define ONSIG 32 /* NSIG for osig* syscalls. XXX. */ SDT_PROVIDER_DECLARE(proc); SDT_PROBE_DEFINE3(proc, kernel, , signal__send, "struct thread *", "struct proc *", "int"); SDT_PROBE_DEFINE2(proc, kernel, , signal__clear, "int", "ksiginfo_t *"); SDT_PROBE_DEFINE3(proc, kernel, , signal__discard, "struct thread *", "struct proc *", "int"); static int coredump(struct thread *); static int killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi); static int issignal(struct thread *td); static int sigprop(int sig); static void tdsigwakeup(struct thread *, int, sig_t, int); static void sig_suspend_threads(struct thread *, struct proc *, int); static int filt_sigattach(struct knote *kn); static void filt_sigdetach(struct knote *kn); static int filt_signal(struct knote *kn, long hint); static struct thread *sigtd(struct proc *p, int sig, int prop); static void sigqueue_start(void); static uma_zone_t ksiginfo_zone = NULL; struct filterops sig_filtops = { .f_isfd = 0, .f_attach = filt_sigattach, .f_detach = filt_sigdetach, .f_event = filt_signal, }; static int kern_logsigexit = 1; SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW, &kern_logsigexit, 0, "Log processes quitting on abnormal signals to syslog(3)"); static int kern_forcesigexit = 1; SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW, &kern_forcesigexit, 0, "Force trap signal to be handled"); static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW, 0, "POSIX real time signal"); static int max_pending_per_proc = 128; SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW, &max_pending_per_proc, 0, "Max pending signals per proc"); static int preallocate_siginfo = 1024; SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RDTUN, &preallocate_siginfo, 0, "Preallocated signal memory size"); static int signal_overflow = 0; SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD, &signal_overflow, 0, "Number of signals overflew"); static int signal_alloc_fail = 0; SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD, &signal_alloc_fail, 0, "signals failed to be allocated"); SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL); /* * Policy -- Can ucred cr1 send SIGIO to process cr2? * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG * in the right situations. */ #define CANSIGIO(cr1, cr2) \ ((cr1)->cr_uid == 0 || \ (cr1)->cr_ruid == (cr2)->cr_ruid || \ (cr1)->cr_uid == (cr2)->cr_ruid || \ (cr1)->cr_ruid == (cr2)->cr_uid || \ (cr1)->cr_uid == (cr2)->cr_uid) static int sugid_coredump; SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RWTUN, &sugid_coredump, 0, "Allow setuid and setgid processes to dump core"); static int capmode_coredump; SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RWTUN, &capmode_coredump, 0, "Allow processes in capability mode to dump core"); static int do_coredump = 1; SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW, &do_coredump, 0, "Enable/Disable coredumps"); static int set_core_nodump_flag = 0; SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag, 0, "Enable setting the NODUMP flag on coredump files"); static int coredump_devctl = 0; SYSCTL_INT(_kern, OID_AUTO, coredump_devctl, CTLFLAG_RW, &coredump_devctl, 0, "Generate a devctl notification when processes coredump"); /* * Signal properties and actions. * The array below categorizes the signals and their default actions * according to the following properties: */ #define SA_KILL 0x01 /* terminates process by default */ #define SA_CORE 0x02 /* ditto and coredumps */ #define SA_STOP 0x04 /* suspend process */ #define SA_TTYSTOP 0x08 /* ditto, from tty */ #define SA_IGNORE 0x10 /* ignore by default */ #define SA_CONT 0x20 /* continue if suspended */ #define SA_CANTMASK 0x40 /* non-maskable, catchable */ static int sigproptbl[NSIG] = { SA_KILL, /* SIGHUP */ SA_KILL, /* SIGINT */ SA_KILL|SA_CORE, /* SIGQUIT */ SA_KILL|SA_CORE, /* SIGILL */ SA_KILL|SA_CORE, /* SIGTRAP */ SA_KILL|SA_CORE, /* SIGABRT */ SA_KILL|SA_CORE, /* SIGEMT */ SA_KILL|SA_CORE, /* SIGFPE */ SA_KILL, /* SIGKILL */ SA_KILL|SA_CORE, /* SIGBUS */ SA_KILL|SA_CORE, /* SIGSEGV */ SA_KILL|SA_CORE, /* SIGSYS */ SA_KILL, /* SIGPIPE */ SA_KILL, /* SIGALRM */ SA_KILL, /* SIGTERM */ SA_IGNORE, /* SIGURG */ SA_STOP, /* SIGSTOP */ SA_STOP|SA_TTYSTOP, /* SIGTSTP */ SA_IGNORE|SA_CONT, /* SIGCONT */ SA_IGNORE, /* SIGCHLD */ SA_STOP|SA_TTYSTOP, /* SIGTTIN */ SA_STOP|SA_TTYSTOP, /* SIGTTOU */ SA_IGNORE, /* SIGIO */ SA_KILL, /* SIGXCPU */ SA_KILL, /* SIGXFSZ */ SA_KILL, /* SIGVTALRM */ SA_KILL, /* SIGPROF */ SA_IGNORE, /* SIGWINCH */ SA_IGNORE, /* SIGINFO */ SA_KILL, /* SIGUSR1 */ SA_KILL, /* SIGUSR2 */ }; static void reschedule_signals(struct proc *p, sigset_t block, int flags); static void sigqueue_start(void) { ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); uma_prealloc(ksiginfo_zone, preallocate_siginfo); p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS); p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1); p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc); } ksiginfo_t * ksiginfo_alloc(int wait) { int flags; flags = M_ZERO; if (! wait) flags |= M_NOWAIT; if (ksiginfo_zone != NULL) return ((ksiginfo_t *)uma_zalloc(ksiginfo_zone, flags)); return (NULL); } void ksiginfo_free(ksiginfo_t *ksi) { uma_zfree(ksiginfo_zone, ksi); } static __inline int ksiginfo_tryfree(ksiginfo_t *ksi) { if (!(ksi->ksi_flags & KSI_EXT)) { uma_zfree(ksiginfo_zone, ksi); return (1); } return (0); } void sigqueue_init(sigqueue_t *list, struct proc *p) { SIGEMPTYSET(list->sq_signals); SIGEMPTYSET(list->sq_kill); TAILQ_INIT(&list->sq_list); list->sq_proc = p; list->sq_flags = SQ_INIT; } /* * Get a signal's ksiginfo. * Return: * 0 - signal not found * others - signal number */ static int sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si) { struct proc *p = sq->sq_proc; struct ksiginfo *ksi, *next; int count = 0; KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited")); if (!SIGISMEMBER(sq->sq_signals, signo)) return (0); if (SIGISMEMBER(sq->sq_kill, signo)) { count++; SIGDELSET(sq->sq_kill, signo); } TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) { if (ksi->ksi_signo == signo) { if (count == 0) { TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link); ksi->ksi_sigq = NULL; ksiginfo_copy(ksi, si); if (ksiginfo_tryfree(ksi) && p != NULL) p->p_pendingcnt--; } if (++count > 1) break; } } if (count <= 1) SIGDELSET(sq->sq_signals, signo); si->ksi_signo = signo; return (signo); } void sigqueue_take(ksiginfo_t *ksi) { struct ksiginfo *kp; struct proc *p; sigqueue_t *sq; if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL) return; p = sq->sq_proc; TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link); ksi->ksi_sigq = NULL; if (!(ksi->ksi_flags & KSI_EXT) && p != NULL) p->p_pendingcnt--; for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL; kp = TAILQ_NEXT(kp, ksi_link)) { if (kp->ksi_signo == ksi->ksi_signo) break; } if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo)) SIGDELSET(sq->sq_signals, ksi->ksi_signo); } static int sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si) { struct proc *p = sq->sq_proc; struct ksiginfo *ksi; int ret = 0; KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited")); if (signo == SIGKILL || signo == SIGSTOP || si == NULL) { SIGADDSET(sq->sq_kill, signo); goto out_set_bit; } /* directly insert the ksi, don't copy it */ if (si->ksi_flags & KSI_INS) { if (si->ksi_flags & KSI_HEAD) TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link); else TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link); si->ksi_sigq = sq; goto out_set_bit; } if (__predict_false(ksiginfo_zone == NULL)) { SIGADDSET(sq->sq_kill, signo); goto out_set_bit; } if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) { signal_overflow++; ret = EAGAIN; } else if ((ksi = ksiginfo_alloc(0)) == NULL) { signal_alloc_fail++; ret = EAGAIN; } else { if (p != NULL) p->p_pendingcnt++; ksiginfo_copy(si, ksi); ksi->ksi_signo = signo; if (si->ksi_flags & KSI_HEAD) TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link); else TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link); ksi->ksi_sigq = sq; } if ((si->ksi_flags & KSI_TRAP) != 0 || (si->ksi_flags & KSI_SIGQ) == 0) { if (ret != 0) SIGADDSET(sq->sq_kill, signo); ret = 0; goto out_set_bit; } if (ret != 0) return (ret); out_set_bit: SIGADDSET(sq->sq_signals, signo); return (ret); } void sigqueue_flush(sigqueue_t *sq) { struct proc *p = sq->sq_proc; ksiginfo_t *ksi; KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited")); if (p != NULL) PROC_LOCK_ASSERT(p, MA_OWNED); while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) { TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link); ksi->ksi_sigq = NULL; if (ksiginfo_tryfree(ksi) && p != NULL) p->p_pendingcnt--; } SIGEMPTYSET(sq->sq_signals); SIGEMPTYSET(sq->sq_kill); } static void sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set) { sigset_t tmp; struct proc *p1, *p2; ksiginfo_t *ksi, *next; KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited")); KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited")); p1 = src->sq_proc; p2 = dst->sq_proc; /* Move siginfo to target list */ TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) { if (SIGISMEMBER(*set, ksi->ksi_signo)) { TAILQ_REMOVE(&src->sq_list, ksi, ksi_link); if (p1 != NULL) p1->p_pendingcnt--; TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link); ksi->ksi_sigq = dst; if (p2 != NULL) p2->p_pendingcnt++; } } /* Move pending bits to target list */ tmp = src->sq_kill; SIGSETAND(tmp, *set); SIGSETOR(dst->sq_kill, tmp); SIGSETNAND(src->sq_kill, tmp); tmp = src->sq_signals; SIGSETAND(tmp, *set); SIGSETOR(dst->sq_signals, tmp); SIGSETNAND(src->sq_signals, tmp); } #if 0 static void sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo) { sigset_t set; SIGEMPTYSET(set); SIGADDSET(set, signo); sigqueue_move_set(src, dst, &set); } #endif static void sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set) { struct proc *p = sq->sq_proc; ksiginfo_t *ksi, *next; KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited")); /* Remove siginfo queue */ TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) { if (SIGISMEMBER(*set, ksi->ksi_signo)) { TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link); ksi->ksi_sigq = NULL; if (ksiginfo_tryfree(ksi) && p != NULL) p->p_pendingcnt--; } } SIGSETNAND(sq->sq_kill, *set); SIGSETNAND(sq->sq_signals, *set); } void sigqueue_delete(sigqueue_t *sq, int signo) { sigset_t set; SIGEMPTYSET(set); SIGADDSET(set, signo); sigqueue_delete_set(sq, &set); } /* Remove a set of signals for a process */ static void sigqueue_delete_set_proc(struct proc *p, const sigset_t *set) { sigqueue_t worklist; struct thread *td0; PROC_LOCK_ASSERT(p, MA_OWNED); sigqueue_init(&worklist, NULL); sigqueue_move_set(&p->p_sigqueue, &worklist, set); FOREACH_THREAD_IN_PROC(p, td0) sigqueue_move_set(&td0->td_sigqueue, &worklist, set); sigqueue_flush(&worklist); } void sigqueue_delete_proc(struct proc *p, int signo) { sigset_t set; SIGEMPTYSET(set); SIGADDSET(set, signo); sigqueue_delete_set_proc(p, &set); } static void sigqueue_delete_stopmask_proc(struct proc *p) { sigset_t set; SIGEMPTYSET(set); SIGADDSET(set, SIGSTOP); SIGADDSET(set, SIGTSTP); SIGADDSET(set, SIGTTIN); SIGADDSET(set, SIGTTOU); sigqueue_delete_set_proc(p, &set); } /* * Determine signal that should be delivered to thread td, the current * thread, 0 if none. If there is a pending stop signal with default * action, the process stops in issignal(). */ int cursig(struct thread *td) { PROC_LOCK_ASSERT(td->td_proc, MA_OWNED); mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED); THREAD_LOCK_ASSERT(td, MA_NOTOWNED); return (SIGPENDING(td) ? issignal(td) : 0); } /* * Arrange for ast() to handle unmasked pending signals on return to user * mode. This must be called whenever a signal is added to td_sigqueue or * unmasked in td_sigmask. */ void signotify(struct thread *td) { struct proc *p; p = td->td_proc; PROC_LOCK_ASSERT(p, MA_OWNED); if (SIGPENDING(td)) { thread_lock(td); td->td_flags |= TDF_NEEDSIGCHK | TDF_ASTPENDING; thread_unlock(td); } } int sigonstack(size_t sp) { struct thread *td = curthread; return ((td->td_pflags & TDP_ALTSTACK) ? #if defined(COMPAT_43) ((td->td_sigstk.ss_size == 0) ? (td->td_sigstk.ss_flags & SS_ONSTACK) : ((sp - (size_t)td->td_sigstk.ss_sp) < td->td_sigstk.ss_size)) #else ((sp - (size_t)td->td_sigstk.ss_sp) < td->td_sigstk.ss_size) #endif : 0); } static __inline int sigprop(int sig) { if (sig > 0 && sig < NSIG) return (sigproptbl[_SIG_IDX(sig)]); return (0); } int sig_ffs(sigset_t *set) { int i; for (i = 0; i < _SIG_WORDS; i++) if (set->__bits[i]) return (ffs(set->__bits[i]) + (i * 32)); return (0); } static bool sigact_flag_test(struct sigaction *act, int flag) { /* * SA_SIGINFO is reset when signal disposition is set to * ignore or default. Other flags are kept according to user * settings. */ return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO || ((__sighandler_t *)act->sa_sigaction != SIG_IGN && (__sighandler_t *)act->sa_sigaction != SIG_DFL))); } /* * kern_sigaction * sigaction * freebsd4_sigaction * osigaction */ int kern_sigaction(td, sig, act, oact, flags) struct thread *td; register int sig; struct sigaction *act, *oact; int flags; { struct sigacts *ps; struct proc *p = td->td_proc; if (!_SIG_VALID(sig)) return (EINVAL); if (act != NULL && act->sa_handler != SIG_DFL && act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK | SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER | SA_NOCLDWAIT | SA_SIGINFO)) != 0) return (EINVAL); PROC_LOCK(p); ps = p->p_sigacts; mtx_lock(&ps->ps_mtx); if (oact) { oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)]; oact->sa_flags = 0; if (SIGISMEMBER(ps->ps_sigonstack, sig)) oact->sa_flags |= SA_ONSTACK; if (!SIGISMEMBER(ps->ps_sigintr, sig)) oact->sa_flags |= SA_RESTART; if (SIGISMEMBER(ps->ps_sigreset, sig)) oact->sa_flags |= SA_RESETHAND; if (SIGISMEMBER(ps->ps_signodefer, sig)) oact->sa_flags |= SA_NODEFER; if (SIGISMEMBER(ps->ps_siginfo, sig)) { oact->sa_flags |= SA_SIGINFO; oact->sa_sigaction = (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)]; } else oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)]; if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP) oact->sa_flags |= SA_NOCLDSTOP; if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT) oact->sa_flags |= SA_NOCLDWAIT; } if (act) { if ((sig == SIGKILL || sig == SIGSTOP) && act->sa_handler != SIG_DFL) { mtx_unlock(&ps->ps_mtx); PROC_UNLOCK(p); return (EINVAL); } /* * Change setting atomically. */ ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask; SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]); if (sigact_flag_test(act, SA_SIGINFO)) { ps->ps_sigact[_SIG_IDX(sig)] = (__sighandler_t *)act->sa_sigaction; SIGADDSET(ps->ps_siginfo, sig); } else { ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler; SIGDELSET(ps->ps_siginfo, sig); } if (!sigact_flag_test(act, SA_RESTART)) SIGADDSET(ps->ps_sigintr, sig); else SIGDELSET(ps->ps_sigintr, sig); if (sigact_flag_test(act, SA_ONSTACK)) SIGADDSET(ps->ps_sigonstack, sig); else SIGDELSET(ps->ps_sigonstack, sig); if (sigact_flag_test(act, SA_RESETHAND)) SIGADDSET(ps->ps_sigreset, sig); else SIGDELSET(ps->ps_sigreset, sig); if (sigact_flag_test(act, SA_NODEFER)) SIGADDSET(ps->ps_signodefer, sig); else SIGDELSET(ps->ps_signodefer, sig); if (sig == SIGCHLD) { if (act->sa_flags & SA_NOCLDSTOP) ps->ps_flag |= PS_NOCLDSTOP; else ps->ps_flag &= ~PS_NOCLDSTOP; if (act->sa_flags & SA_NOCLDWAIT) { /* * Paranoia: since SA_NOCLDWAIT is implemented * by reparenting the dying child to PID 1 (and * trust it to reap the zombie), PID 1 itself * is forbidden to set SA_NOCLDWAIT. */ if (p->p_pid == 1) ps->ps_flag &= ~PS_NOCLDWAIT; else ps->ps_flag |= PS_NOCLDWAIT; } else ps->ps_flag &= ~PS_NOCLDWAIT; if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN) ps->ps_flag |= PS_CLDSIGIGN; else ps->ps_flag &= ~PS_CLDSIGIGN; } /* * Set bit in ps_sigignore for signals that are set to SIG_IGN, * and for signals set to SIG_DFL where the default is to * ignore. However, don't put SIGCONT in ps_sigignore, as we * have to restart the process. */ if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN || (sigprop(sig) & SA_IGNORE && ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) { /* never to be seen again */ sigqueue_delete_proc(p, sig); if (sig != SIGCONT) /* easier in psignal */ SIGADDSET(ps->ps_sigignore, sig); SIGDELSET(ps->ps_sigcatch, sig); } else { SIGDELSET(ps->ps_sigignore, sig); if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL) SIGDELSET(ps->ps_sigcatch, sig); else SIGADDSET(ps->ps_sigcatch, sig); } #ifdef COMPAT_FREEBSD4 if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN || ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL || (flags & KSA_FREEBSD4) == 0) SIGDELSET(ps->ps_freebsd4, sig); else SIGADDSET(ps->ps_freebsd4, sig); #endif #ifdef COMPAT_43 if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN || ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL || (flags & KSA_OSIGSET) == 0) SIGDELSET(ps->ps_osigset, sig); else SIGADDSET(ps->ps_osigset, sig); #endif } mtx_unlock(&ps->ps_mtx); PROC_UNLOCK(p); return (0); } #ifndef _SYS_SYSPROTO_H_ struct sigaction_args { int sig; struct sigaction *act; struct sigaction *oact; }; #endif int sys_sigaction(td, uap) struct thread *td; register struct sigaction_args *uap; { struct sigaction act, oact; register struct sigaction *actp, *oactp; int error; actp = (uap->act != NULL) ? &act : NULL; oactp = (uap->oact != NULL) ? &oact : NULL; if (actp) { error = copyin(uap->act, actp, sizeof(act)); if (error) return (error); } error = kern_sigaction(td, uap->sig, actp, oactp, 0); if (oactp && !error) error = copyout(oactp, uap->oact, sizeof(oact)); return (error); } #ifdef COMPAT_FREEBSD4 #ifndef _SYS_SYSPROTO_H_ struct freebsd4_sigaction_args { int sig; struct sigaction *act; struct sigaction *oact; }; #endif int freebsd4_sigaction(td, uap) struct thread *td; register struct freebsd4_sigaction_args *uap; { struct sigaction act, oact; register struct sigaction *actp, *oactp; int error; actp = (uap->act != NULL) ? &act : NULL; oactp = (uap->oact != NULL) ? &oact : NULL; if (actp) { error = copyin(uap->act, actp, sizeof(act)); if (error) return (error); } error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4); if (oactp && !error) error = copyout(oactp, uap->oact, sizeof(oact)); return (error); } #endif /* COMAPT_FREEBSD4 */ #ifdef COMPAT_43 /* XXX - COMPAT_FBSD3 */ #ifndef _SYS_SYSPROTO_H_ struct osigaction_args { int signum; struct osigaction *nsa; struct osigaction *osa; }; #endif int osigaction(td, uap) struct thread *td; register struct osigaction_args *uap; { struct osigaction sa; struct sigaction nsa, osa; register struct sigaction *nsap, *osap; int error; if (uap->signum <= 0 || uap->signum >= ONSIG) return (EINVAL); nsap = (uap->nsa != NULL) ? &nsa : NULL; osap = (uap->osa != NULL) ? &osa : NULL; if (nsap) { error = copyin(uap->nsa, &sa, sizeof(sa)); if (error) return (error); nsap->sa_handler = sa.sa_handler; nsap->sa_flags = sa.sa_flags; OSIG2SIG(sa.sa_mask, nsap->sa_mask); } error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET); if (osap && !error) { sa.sa_handler = osap->sa_handler; sa.sa_flags = osap->sa_flags; SIG2OSIG(osap->sa_mask, sa.sa_mask); error = copyout(&sa, uap->osa, sizeof(sa)); } return (error); } #if !defined(__i386__) /* Avoid replicating the same stub everywhere */ int osigreturn(td, uap) struct thread *td; struct osigreturn_args *uap; { return (nosys(td, (struct nosys_args *)uap)); } #endif #endif /* COMPAT_43 */ /* * Initialize signal state for process 0; * set to ignore signals that are ignored by default. */ void siginit(p) struct proc *p; { register int i; struct sigacts *ps; PROC_LOCK(p); ps = p->p_sigacts; mtx_lock(&ps->ps_mtx); for (i = 1; i <= NSIG; i++) { if (sigprop(i) & SA_IGNORE && i != SIGCONT) { SIGADDSET(ps->ps_sigignore, i); } } mtx_unlock(&ps->ps_mtx); PROC_UNLOCK(p); } /* * Reset specified signal to the default disposition. */ static void sigdflt(struct sigacts *ps, int sig) { mtx_assert(&ps->ps_mtx, MA_OWNED); SIGDELSET(ps->ps_sigcatch, sig); if ((sigprop(sig) & SA_IGNORE) != 0 && sig != SIGCONT) SIGADDSET(ps->ps_sigignore, sig); ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL; SIGDELSET(ps->ps_siginfo, sig); } /* * Reset signals for an exec of the specified process. */ void execsigs(struct proc *p) { struct sigacts *ps; int sig; struct thread *td; /* * Reset caught signals. Held signals remain held * through td_sigmask (unless they were caught, * and are now ignored by default). */ PROC_LOCK_ASSERT(p, MA_OWNED); td = FIRST_THREAD_IN_PROC(p); ps = p->p_sigacts; mtx_lock(&ps->ps_mtx); while (SIGNOTEMPTY(ps->ps_sigcatch)) { sig = sig_ffs(&ps->ps_sigcatch); sigdflt(ps, sig); if ((sigprop(sig) & SA_IGNORE) != 0) sigqueue_delete_proc(p, sig); } /* * Reset stack state to the user stack. * Clear set of signals caught on the signal stack. */ td->td_sigstk.ss_flags = SS_DISABLE; td->td_sigstk.ss_size = 0; td->td_sigstk.ss_sp = 0; td->td_pflags &= ~TDP_ALTSTACK; /* * Reset no zombies if child dies flag as Solaris does. */ ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN); if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN) ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL; mtx_unlock(&ps->ps_mtx); } /* * kern_sigprocmask() * * Manipulate signal mask. */ int kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset, int flags) { sigset_t new_block, oset1; struct proc *p; int error; p = td->td_proc; if ((flags & SIGPROCMASK_PROC_LOCKED) != 0) PROC_LOCK_ASSERT(p, MA_OWNED); else PROC_LOCK(p); mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0 ? MA_OWNED : MA_NOTOWNED); if (oset != NULL) *oset = td->td_sigmask; error = 0; if (set != NULL) { switch (how) { case SIG_BLOCK: SIG_CANTMASK(*set); oset1 = td->td_sigmask; SIGSETOR(td->td_sigmask, *set); new_block = td->td_sigmask; SIGSETNAND(new_block, oset1); break; case SIG_UNBLOCK: SIGSETNAND(td->td_sigmask, *set); signotify(td); goto out; case SIG_SETMASK: SIG_CANTMASK(*set); oset1 = td->td_sigmask; if (flags & SIGPROCMASK_OLD) SIGSETLO(td->td_sigmask, *set); else td->td_sigmask = *set; new_block = td->td_sigmask; SIGSETNAND(new_block, oset1); signotify(td); break; default: error = EINVAL; goto out; } /* * The new_block set contains signals that were not previously * blocked, but are blocked now. * * In case we block any signal that was not previously blocked * for td, and process has the signal pending, try to schedule * signal delivery to some thread that does not block the * signal, possibly waking it up. */ if (p->p_numthreads != 1) reschedule_signals(p, new_block, flags); } out: if (!(flags & SIGPROCMASK_PROC_LOCKED)) PROC_UNLOCK(p); return (error); } #ifndef _SYS_SYSPROTO_H_ struct sigprocmask_args { int how; const sigset_t *set; sigset_t *oset; }; #endif int sys_sigprocmask(td, uap) register struct thread *td; struct sigprocmask_args *uap; { sigset_t set, oset; sigset_t *setp, *osetp; int error; setp = (uap->set != NULL) ? &set : NULL; osetp = (uap->oset != NULL) ? &oset : NULL; if (setp) { error = copyin(uap->set, setp, sizeof(set)); if (error) return (error); } error = kern_sigprocmask(td, uap->how, setp, osetp, 0); if (osetp && !error) { error = copyout(osetp, uap->oset, sizeof(oset)); } return (error); } #ifdef COMPAT_43 /* XXX - COMPAT_FBSD3 */ #ifndef _SYS_SYSPROTO_H_ struct osigprocmask_args { int how; osigset_t mask; }; #endif int osigprocmask(td, uap) register struct thread *td; struct osigprocmask_args *uap; { sigset_t set, oset; int error; OSIG2SIG(uap->mask, set); error = kern_sigprocmask(td, uap->how, &set, &oset, 1); SIG2OSIG(oset, td->td_retval[0]); return (error); } #endif /* COMPAT_43 */ int sys_sigwait(struct thread *td, struct sigwait_args *uap) { ksiginfo_t ksi; sigset_t set; int error; error = copyin(uap->set, &set, sizeof(set)); if (error) { td->td_retval[0] = error; return (0); } error = kern_sigtimedwait(td, set, &ksi, NULL); if (error) { if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT) error = ERESTART; if (error == ERESTART) return (error); td->td_retval[0] = error; return (0); } error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo)); td->td_retval[0] = error; return (0); } int sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap) { struct timespec ts; struct timespec *timeout; sigset_t set; ksiginfo_t ksi; int error; if (uap->timeout) { error = copyin(uap->timeout, &ts, sizeof(ts)); if (error) return (error); timeout = &ts; } else timeout = NULL; error = copyin(uap->set, &set, sizeof(set)); if (error) return (error); error = kern_sigtimedwait(td, set, &ksi, timeout); if (error) return (error); if (uap->info) error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t)); if (error == 0) td->td_retval[0] = ksi.ksi_signo; return (error); } int sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap) { ksiginfo_t ksi; sigset_t set; int error; error = copyin(uap->set, &set, sizeof(set)); if (error) return (error); error = kern_sigtimedwait(td, set, &ksi, NULL); if (error) return (error); if (uap->info) error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t)); if (error == 0) td->td_retval[0] = ksi.ksi_signo; return (error); } int kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi, struct timespec *timeout) { struct sigacts *ps; sigset_t saved_mask, new_block; struct proc *p; int error, sig, timo, timevalid = 0; struct timespec rts, ets, ts; struct timeval tv; p = td->td_proc; error = 0; ets.tv_sec = 0; ets.tv_nsec = 0; if (timeout != NULL) { if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) { timevalid = 1; getnanouptime(&rts); ets = rts; timespecadd(&ets, timeout); } } ksiginfo_init(ksi); /* Some signals can not be waited for. */ SIG_CANTMASK(waitset); ps = p->p_sigacts; PROC_LOCK(p); saved_mask = td->td_sigmask; SIGSETNAND(td->td_sigmask, waitset); for (;;) { mtx_lock(&ps->ps_mtx); sig = cursig(td); mtx_unlock(&ps->ps_mtx); if (sig != 0 && SIGISMEMBER(waitset, sig)) { if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 || sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) { error = 0; break; } } if (error != 0) break; /* * POSIX says this must be checked after looking for pending * signals. */ if (timeout != NULL) { if (!timevalid) { error = EINVAL; break; } getnanouptime(&rts); if (timespeccmp(&rts, &ets, >=)) { error = EAGAIN; break; } ts = ets; timespecsub(&ts, &rts); TIMESPEC_TO_TIMEVAL(&tv, &ts); timo = tvtohz(&tv); } else { timo = 0; } error = msleep(ps, &p->p_mtx, PPAUSE|PCATCH, "sigwait", timo); if (timeout != NULL) { if (error == ERESTART) { /* Timeout can not be restarted. */ error = EINTR; } else if (error == EAGAIN) { /* We will calculate timeout by ourself. */ error = 0; } } } new_block = saved_mask; SIGSETNAND(new_block, td->td_sigmask); td->td_sigmask = saved_mask; /* * Fewer signals can be delivered to us, reschedule signal * notification. */ if (p->p_numthreads != 1) reschedule_signals(p, new_block, 0); if (error == 0) { SDT_PROBE(proc, kernel, , signal__clear, sig, ksi, 0, 0, 0); if (ksi->ksi_code == SI_TIMER) itimer_accept(p, ksi->ksi_timerid, ksi); #ifdef KTRACE if (KTRPOINT(td, KTR_PSIG)) { sig_t action; mtx_lock(&ps->ps_mtx); action = ps->ps_sigact[_SIG_IDX(sig)]; mtx_unlock(&ps->ps_mtx); ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code); } #endif if (sig == SIGKILL) sigexit(td, sig); } PROC_UNLOCK(p); return (error); } #ifndef _SYS_SYSPROTO_H_ struct sigpending_args { sigset_t *set; }; #endif int sys_sigpending(td, uap) struct thread *td; struct sigpending_args *uap; { struct proc *p = td->td_proc; sigset_t pending; PROC_LOCK(p); pending = p->p_sigqueue.sq_signals; SIGSETOR(pending, td->td_sigqueue.sq_signals); PROC_UNLOCK(p); return (copyout(&pending, uap->set, sizeof(sigset_t))); } #ifdef COMPAT_43 /* XXX - COMPAT_FBSD3 */ #ifndef _SYS_SYSPROTO_H_ struct osigpending_args { int dummy; }; #endif int osigpending(td, uap) struct thread *td; struct osigpending_args *uap; { struct proc *p = td->td_proc; sigset_t pending; PROC_LOCK(p); pending = p->p_sigqueue.sq_signals; SIGSETOR(pending, td->td_sigqueue.sq_signals); PROC_UNLOCK(p); SIG2OSIG(pending, td->td_retval[0]); return (0); } #endif /* COMPAT_43 */ #if defined(COMPAT_43) /* * Generalized interface signal handler, 4.3-compatible. */ #ifndef _SYS_SYSPROTO_H_ struct osigvec_args { int signum; struct sigvec *nsv; struct sigvec *osv; }; #endif /* ARGSUSED */ int osigvec(td, uap) struct thread *td; register struct osigvec_args *uap; { struct sigvec vec; struct sigaction nsa, osa; register struct sigaction *nsap, *osap; int error; if (uap->signum <= 0 || uap->signum >= ONSIG) return (EINVAL); nsap = (uap->nsv != NULL) ? &nsa : NULL; osap = (uap->osv != NULL) ? &osa : NULL; if (nsap) { error = copyin(uap->nsv, &vec, sizeof(vec)); if (error) return (error); nsap->sa_handler = vec.sv_handler; OSIG2SIG(vec.sv_mask, nsap->sa_mask); nsap->sa_flags = vec.sv_flags; nsap->sa_flags ^= SA_RESTART; /* opposite of SV_INTERRUPT */ } error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET); if (osap && !error) { vec.sv_handler = osap->sa_handler; SIG2OSIG(osap->sa_mask, vec.sv_mask); vec.sv_flags = osap->sa_flags; vec.sv_flags &= ~SA_NOCLDWAIT; vec.sv_flags ^= SA_RESTART; error = copyout(&vec, uap->osv, sizeof(vec)); } return (error); } #ifndef _SYS_SYSPROTO_H_ struct osigblock_args { int mask; }; #endif int osigblock(td, uap) register struct thread *td; struct osigblock_args *uap; { sigset_t set, oset; OSIG2SIG(uap->mask, set); kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0); SIG2OSIG(oset, td->td_retval[0]); return (0); } #ifndef _SYS_SYSPROTO_H_ struct osigsetmask_args { int mask; }; #endif int osigsetmask(td, uap) struct thread *td; struct osigsetmask_args *uap; { sigset_t set, oset; OSIG2SIG(uap->mask, set); kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0); SIG2OSIG(oset, td->td_retval[0]); return (0); } #endif /* COMPAT_43 */ /* * Suspend calling thread until signal, providing mask to be set in the * meantime. */ #ifndef _SYS_SYSPROTO_H_ struct sigsuspend_args { const sigset_t *sigmask; }; #endif /* ARGSUSED */ int sys_sigsuspend(td, uap) struct thread *td; struct sigsuspend_args *uap; { sigset_t mask; int error; error = copyin(uap->sigmask, &mask, sizeof(mask)); if (error) return (error); return (kern_sigsuspend(td, mask)); } int kern_sigsuspend(struct thread *td, sigset_t mask) { struct proc *p = td->td_proc; int has_sig, sig; /* * When returning from sigsuspend, we want * the old mask to be restored after the * signal handler has finished. Thus, we * save it here and mark the sigacts structure * to indicate this. */ PROC_LOCK(p); kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask, SIGPROCMASK_PROC_LOCKED); td->td_pflags |= TDP_OLDMASK; /* * Process signals now. Otherwise, we can get spurious wakeup * due to signal entered process queue, but delivered to other * thread. But sigsuspend should return only on signal * delivery. */ (p->p_sysent->sv_set_syscall_retval)(td, EINTR); for (has_sig = 0; !has_sig;) { while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause", 0) == 0) /* void */; thread_suspend_check(0); mtx_lock(&p->p_sigacts->ps_mtx); while ((sig = cursig(td)) != 0) has_sig += postsig(sig); mtx_unlock(&p->p_sigacts->ps_mtx); } PROC_UNLOCK(p); td->td_errno = EINTR; td->td_pflags |= TDP_NERRNO; return (EJUSTRETURN); } #ifdef COMPAT_43 /* XXX - COMPAT_FBSD3 */ /* * Compatibility sigsuspend call for old binaries. Note nonstandard calling * convention: libc stub passes mask, not pointer, to save a copyin. */ #ifndef _SYS_SYSPROTO_H_ struct osigsuspend_args { osigset_t mask; }; #endif /* ARGSUSED */ int osigsuspend(td, uap) struct thread *td; struct osigsuspend_args *uap; { sigset_t mask; OSIG2SIG(uap->mask, mask); return (kern_sigsuspend(td, mask)); } #endif /* COMPAT_43 */ #if defined(COMPAT_43) #ifndef _SYS_SYSPROTO_H_ struct osigstack_args { struct sigstack *nss; struct sigstack *oss; }; #endif /* ARGSUSED */ int osigstack(td, uap) struct thread *td; register struct osigstack_args *uap; { struct sigstack nss, oss; int error = 0; if (uap->nss != NULL) { error = copyin(uap->nss, &nss, sizeof(nss)); if (error) return (error); } oss.ss_sp = td->td_sigstk.ss_sp; oss.ss_onstack = sigonstack(cpu_getstack(td)); if (uap->nss != NULL) { td->td_sigstk.ss_sp = nss.ss_sp; td->td_sigstk.ss_size = 0; td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK; td->td_pflags |= TDP_ALTSTACK; } if (uap->oss != NULL) error = copyout(&oss, uap->oss, sizeof(oss)); return (error); } #endif /* COMPAT_43 */ #ifndef _SYS_SYSPROTO_H_ struct sigaltstack_args { stack_t *ss; stack_t *oss; }; #endif /* ARGSUSED */ int sys_sigaltstack(td, uap) struct thread *td; register struct sigaltstack_args *uap; { stack_t ss, oss; int error; if (uap->ss != NULL) { error = copyin(uap->ss, &ss, sizeof(ss)); if (error) return (error); } error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL, (uap->oss != NULL) ? &oss : NULL); if (error) return (error); if (uap->oss != NULL) error = copyout(&oss, uap->oss, sizeof(stack_t)); return (error); } int kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss) { struct proc *p = td->td_proc; int oonstack; oonstack = sigonstack(cpu_getstack(td)); if (oss != NULL) { *oss = td->td_sigstk; oss->ss_flags = (td->td_pflags & TDP_ALTSTACK) ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE; } if (ss != NULL) { if (oonstack) return (EPERM); if ((ss->ss_flags & ~SS_DISABLE) != 0) return (EINVAL); if (!(ss->ss_flags & SS_DISABLE)) { if (ss->ss_size < p->p_sysent->sv_minsigstksz) return (ENOMEM); td->td_sigstk = *ss; td->td_pflags |= TDP_ALTSTACK; } else { td->td_pflags &= ~TDP_ALTSTACK; } } return (0); } /* * Common code for kill process group/broadcast kill. * cp is calling process. */ static int killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi) { struct proc *p; struct pgrp *pgrp; int err; int ret; ret = ESRCH; if (all) { /* * broadcast */ sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); if (p->p_pid <= 1 || p->p_flag & P_SYSTEM || p == td->td_proc || p->p_state == PRS_NEW) { PROC_UNLOCK(p); continue; } err = p_cansignal(td, p, sig); if (err == 0) { if (sig) pksignal(p, sig, ksi); ret = err; } else if (ret == ESRCH) ret = err; PROC_UNLOCK(p); } sx_sunlock(&allproc_lock); } else { sx_slock(&proctree_lock); if (pgid == 0) { /* * zero pgid means send to my process group. */ pgrp = td->td_proc->p_pgrp; PGRP_LOCK(pgrp); } else { pgrp = pgfind(pgid); if (pgrp == NULL) { sx_sunlock(&proctree_lock); return (ESRCH); } } sx_sunlock(&proctree_lock); LIST_FOREACH(p, &pgrp->pg_members, p_pglist) { PROC_LOCK(p); if (p->p_pid <= 1 || p->p_flag & P_SYSTEM || p->p_state == PRS_NEW) { PROC_UNLOCK(p); continue; } err = p_cansignal(td, p, sig); if (err == 0) { if (sig) pksignal(p, sig, ksi); ret = err; } else if (ret == ESRCH) ret = err; PROC_UNLOCK(p); } PGRP_UNLOCK(pgrp); } return (ret); } #ifndef _SYS_SYSPROTO_H_ struct kill_args { int pid; int signum; }; #endif /* ARGSUSED */ int sys_kill(struct thread *td, struct kill_args *uap) { ksiginfo_t ksi; struct proc *p; int error; /* * A process in capability mode can send signals only to himself. * The main rationale behind this is that abort(3) is implemented as * kill(getpid(), SIGABRT). */ if (IN_CAPABILITY_MODE(td) && uap->pid != td->td_proc->p_pid) return (ECAPMODE); AUDIT_ARG_SIGNUM(uap->signum); AUDIT_ARG_PID(uap->pid); if ((u_int)uap->signum > _SIG_MAXSIG) return (EINVAL); ksiginfo_init(&ksi); ksi.ksi_signo = uap->signum; ksi.ksi_code = SI_USER; ksi.ksi_pid = td->td_proc->p_pid; ksi.ksi_uid = td->td_ucred->cr_ruid; if (uap->pid > 0) { /* kill single process */ if ((p = pfind(uap->pid)) == NULL) { if ((p = zpfind(uap->pid)) == NULL) return (ESRCH); } AUDIT_ARG_PROCESS(p); error = p_cansignal(td, p, uap->signum); if (error == 0 && uap->signum) pksignal(p, uap->signum, &ksi); PROC_UNLOCK(p); return (error); } switch (uap->pid) { case -1: /* broadcast signal */ return (killpg1(td, uap->signum, 0, 1, &ksi)); case 0: /* signal own process group */ return (killpg1(td, uap->signum, 0, 0, &ksi)); default: /* negative explicit process group */ return (killpg1(td, uap->signum, -uap->pid, 0, &ksi)); } /* NOTREACHED */ } int sys_pdkill(td, uap) struct thread *td; struct pdkill_args *uap; { struct proc *p; cap_rights_t rights; int error; AUDIT_ARG_SIGNUM(uap->signum); AUDIT_ARG_FD(uap->fd); if ((u_int)uap->signum > _SIG_MAXSIG) return (EINVAL); error = procdesc_find(td, uap->fd, cap_rights_init(&rights, CAP_PDKILL), &p); if (error) return (error); AUDIT_ARG_PROCESS(p); error = p_cansignal(td, p, uap->signum); if (error == 0 && uap->signum) kern_psignal(p, uap->signum); PROC_UNLOCK(p); return (error); } #if defined(COMPAT_43) #ifndef _SYS_SYSPROTO_H_ struct okillpg_args { int pgid; int signum; }; #endif /* ARGSUSED */ int okillpg(struct thread *td, struct okillpg_args *uap) { ksiginfo_t ksi; AUDIT_ARG_SIGNUM(uap->signum); AUDIT_ARG_PID(uap->pgid); if ((u_int)uap->signum > _SIG_MAXSIG) return (EINVAL); ksiginfo_init(&ksi); ksi.ksi_signo = uap->signum; ksi.ksi_code = SI_USER; ksi.ksi_pid = td->td_proc->p_pid; ksi.ksi_uid = td->td_ucred->cr_ruid; return (killpg1(td, uap->signum, uap->pgid, 0, &ksi)); } #endif /* COMPAT_43 */ #ifndef _SYS_SYSPROTO_H_ struct sigqueue_args { pid_t pid; int signum; /* union sigval */ void *value; }; #endif int sys_sigqueue(struct thread *td, struct sigqueue_args *uap) { ksiginfo_t ksi; struct proc *p; int error; if ((u_int)uap->signum > _SIG_MAXSIG) return (EINVAL); /* * Specification says sigqueue can only send signal to * single process. */ if (uap->pid <= 0) return (EINVAL); if ((p = pfind(uap->pid)) == NULL) { if ((p = zpfind(uap->pid)) == NULL) return (ESRCH); } error = p_cansignal(td, p, uap->signum); if (error == 0 && uap->signum != 0) { ksiginfo_init(&ksi); ksi.ksi_flags = KSI_SIGQ; ksi.ksi_signo = uap->signum; ksi.ksi_code = SI_QUEUE; ksi.ksi_pid = td->td_proc->p_pid; ksi.ksi_uid = td->td_ucred->cr_ruid; ksi.ksi_value.sival_ptr = uap->value; error = pksignal(p, ksi.ksi_signo, &ksi); } PROC_UNLOCK(p); return (error); } /* * Send a signal to a process group. */ void gsignal(int pgid, int sig, ksiginfo_t *ksi) { struct pgrp *pgrp; if (pgid != 0) { sx_slock(&proctree_lock); pgrp = pgfind(pgid); sx_sunlock(&proctree_lock); if (pgrp != NULL) { pgsignal(pgrp, sig, 0, ksi); PGRP_UNLOCK(pgrp); } } } /* * Send a signal to a process group. If checktty is 1, * limit to members which have a controlling terminal. */ void pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi) { struct proc *p; if (pgrp) { PGRP_LOCK_ASSERT(pgrp, MA_OWNED); LIST_FOREACH(p, &pgrp->pg_members, p_pglist) { PROC_LOCK(p); if (p->p_state == PRS_NORMAL && (checkctty == 0 || p->p_flag & P_CONTROLT)) pksignal(p, sig, ksi); PROC_UNLOCK(p); } } } /* * Recalculate the signal mask and reset the signal disposition after * usermode frame for delivery is formed. Should be called after * mach-specific routine, because sysent->sv_sendsig() needs correct * ps_siginfo and signal mask. */ static void postsig_done(int sig, struct thread *td, struct sigacts *ps) { sigset_t mask; mtx_assert(&ps->ps_mtx, MA_OWNED); td->td_ru.ru_nsignals++; mask = ps->ps_catchmask[_SIG_IDX(sig)]; if (!SIGISMEMBER(ps->ps_signodefer, sig)) SIGADDSET(mask, sig); kern_sigprocmask(td, SIG_BLOCK, &mask, NULL, SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED); if (SIGISMEMBER(ps->ps_sigreset, sig)) sigdflt(ps, sig); } /* * Send a signal caused by a trap to the current thread. If it will be * caught immediately, deliver it with correct code. Otherwise, post it * normally. */ void trapsignal(struct thread *td, ksiginfo_t *ksi) { struct sigacts *ps; struct proc *p; int sig; int code; p = td->td_proc; sig = ksi->ksi_signo; code = ksi->ksi_code; KASSERT(_SIG_VALID(sig), ("invalid signal")); PROC_LOCK(p); ps = p->p_sigacts; mtx_lock(&ps->ps_mtx); if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) && !SIGISMEMBER(td->td_sigmask, sig)) { #ifdef KTRACE if (KTRPOINT(curthread, KTR_PSIG)) ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)], &td->td_sigmask, code); #endif (*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)], ksi, &td->td_sigmask); postsig_done(sig, td, ps); mtx_unlock(&ps->ps_mtx); } else { /* * Avoid a possible infinite loop if the thread * masking the signal or process is ignoring the * signal. */ if (kern_forcesigexit && (SIGISMEMBER(td->td_sigmask, sig) || ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) { SIGDELSET(td->td_sigmask, sig); SIGDELSET(ps->ps_sigcatch, sig); SIGDELSET(ps->ps_sigignore, sig); ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL; } mtx_unlock(&ps->ps_mtx); p->p_code = code; /* XXX for core dump/debugger */ p->p_sig = sig; /* XXX to verify code */ tdsendsignal(p, td, sig, ksi); } PROC_UNLOCK(p); } static struct thread * sigtd(struct proc *p, int sig, int prop) { struct thread *td, *signal_td; PROC_LOCK_ASSERT(p, MA_OWNED); /* * Check if current thread can handle the signal without * switching context to another thread. */ if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig)) return (curthread); signal_td = NULL; FOREACH_THREAD_IN_PROC(p, td) { if (!SIGISMEMBER(td->td_sigmask, sig)) { signal_td = td; break; } } if (signal_td == NULL) signal_td = FIRST_THREAD_IN_PROC(p); return (signal_td); } /* * Send the signal to the process. If the signal has an action, the action * is usually performed by the target process rather than the caller; we add * the signal to the set of pending signals for the process. * * Exceptions: * o When a stop signal is sent to a sleeping process that takes the * default action, the process is stopped without awakening it. * o SIGCONT restarts stopped processes (or puts them back to sleep) * regardless of the signal action (eg, blocked or ignored). * * Other ignored signals are discarded immediately. * * NB: This function may be entered from the debugger via the "kill" DDB * command. There is little that can be done to mitigate the possibly messy * side effects of this unwise possibility. */ void kern_psignal(struct proc *p, int sig) { ksiginfo_t ksi; ksiginfo_init(&ksi); ksi.ksi_signo = sig; ksi.ksi_code = SI_KERNEL; (void) tdsendsignal(p, NULL, sig, &ksi); } int pksignal(struct proc *p, int sig, ksiginfo_t *ksi) { return (tdsendsignal(p, NULL, sig, ksi)); } /* Utility function for finding a thread to send signal event to. */ int sigev_findtd(struct proc *p ,struct sigevent *sigev, struct thread **ttd) { struct thread *td; if (sigev->sigev_notify == SIGEV_THREAD_ID) { td = tdfind(sigev->sigev_notify_thread_id, p->p_pid); if (td == NULL) return (ESRCH); *ttd = td; } else { *ttd = NULL; PROC_LOCK(p); } return (0); } void tdsignal(struct thread *td, int sig) { ksiginfo_t ksi; ksiginfo_init(&ksi); ksi.ksi_signo = sig; ksi.ksi_code = SI_KERNEL; (void) tdsendsignal(td->td_proc, td, sig, &ksi); } void tdksignal(struct thread *td, int sig, ksiginfo_t *ksi) { (void) tdsendsignal(td->td_proc, td, sig, ksi); } int tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi) { sig_t action; sigqueue_t *sigqueue; int prop; struct sigacts *ps; int intrval; int ret = 0; int wakeup_swapper; MPASS(td == NULL || p == td->td_proc); PROC_LOCK_ASSERT(p, MA_OWNED); if (!_SIG_VALID(sig)) panic("%s(): invalid signal %d", __func__, sig); KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__)); /* * IEEE Std 1003.1-2001: return success when killing a zombie. */ if (p->p_state == PRS_ZOMBIE) { if (ksi && (ksi->ksi_flags & KSI_INS)) ksiginfo_tryfree(ksi); return (ret); } ps = p->p_sigacts; KNOTE_LOCKED(&p->p_klist, NOTE_SIGNAL | sig); prop = sigprop(sig); if (td == NULL) { td = sigtd(p, sig, prop); sigqueue = &p->p_sigqueue; } else sigqueue = &td->td_sigqueue; SDT_PROBE(proc, kernel, , signal__send, td, p, sig, 0, 0 ); /* * If the signal is being ignored, * then we forget about it immediately. * (Note: we don't set SIGCONT in ps_sigignore, * and if it is set to SIG_IGN, * action will be SIG_DFL here.) */ mtx_lock(&ps->ps_mtx); if (SIGISMEMBER(ps->ps_sigignore, sig)) { SDT_PROBE(proc, kernel, , signal__discard, td, p, sig, 0, 0 ); mtx_unlock(&ps->ps_mtx); if (ksi && (ksi->ksi_flags & KSI_INS)) ksiginfo_tryfree(ksi); return (ret); } if (SIGISMEMBER(td->td_sigmask, sig)) action = SIG_HOLD; else if (SIGISMEMBER(ps->ps_sigcatch, sig)) action = SIG_CATCH; else action = SIG_DFL; if (SIGISMEMBER(ps->ps_sigintr, sig)) intrval = EINTR; else intrval = ERESTART; mtx_unlock(&ps->ps_mtx); if (prop & SA_CONT) sigqueue_delete_stopmask_proc(p); else if (prop & SA_STOP) { /* * If sending a tty stop signal to a member of an orphaned * process group, discard the signal here if the action * is default; don't stop the process below if sleeping, * and don't clear any pending SIGCONT. */ if ((prop & SA_TTYSTOP) && (p->p_pgrp->pg_jobc == 0) && (action == SIG_DFL)) { if (ksi && (ksi->ksi_flags & KSI_INS)) ksiginfo_tryfree(ksi); return (ret); } sigqueue_delete_proc(p, SIGCONT); if (p->p_flag & P_CONTINUED) { p->p_flag &= ~P_CONTINUED; PROC_LOCK(p->p_pptr); sigqueue_take(p->p_ksi); PROC_UNLOCK(p->p_pptr); } } ret = sigqueue_add(sigqueue, sig, ksi); if (ret != 0) return (ret); signotify(td); /* * Defer further processing for signals which are held, * except that stopped processes must be continued by SIGCONT. */ if (action == SIG_HOLD && !((prop & SA_CONT) && (p->p_flag & P_STOPPED_SIG))) return (ret); /* * SIGKILL: Remove procfs STOPEVENTs. */ if (sig == SIGKILL) { /* from procfs_ioctl.c: PIOCBIC */ p->p_stops = 0; /* from procfs_ioctl.c: PIOCCONT */ p->p_step = 0; wakeup(&p->p_step); } /* * Some signals have a process-wide effect and a per-thread * component. Most processing occurs when the process next * tries to cross the user boundary, however there are some * times when processing needs to be done immediately, such as * waking up threads so that they can cross the user boundary. * We try to do the per-process part here. */ if (P_SHOULDSTOP(p)) { KASSERT(!(p->p_flag & P_WEXIT), ("signal to stopped but exiting process")); if (sig == SIGKILL) { /* * If traced process is already stopped, * then no further action is necessary. */ if (p->p_flag & P_TRACED) goto out; /* * SIGKILL sets process running. * It will die elsewhere. * All threads must be restarted. */ p->p_flag &= ~P_STOPPED_SIG; goto runfast; } if (prop & SA_CONT) { /* * If traced process is already stopped, * then no further action is necessary. */ if (p->p_flag & P_TRACED) goto out; /* * If SIGCONT is default (or ignored), we continue the * process but don't leave the signal in sigqueue as * it has no further action. If SIGCONT is held, we * continue the process and leave the signal in * sigqueue. If the process catches SIGCONT, let it * handle the signal itself. If it isn't waiting on * an event, it goes back to run state. * Otherwise, process goes back to sleep state. */ p->p_flag &= ~P_STOPPED_SIG; PROC_SLOCK(p); if (p->p_numthreads == p->p_suspcount) { PROC_SUNLOCK(p); p->p_flag |= P_CONTINUED; p->p_xstat = SIGCONT; PROC_LOCK(p->p_pptr); childproc_continued(p); PROC_UNLOCK(p->p_pptr); PROC_SLOCK(p); } if (action == SIG_DFL) { thread_unsuspend(p); PROC_SUNLOCK(p); sigqueue_delete(sigqueue, sig); goto out; } if (action == SIG_CATCH) { /* * The process wants to catch it so it needs * to run at least one thread, but which one? */ PROC_SUNLOCK(p); goto runfast; } /* * The signal is not ignored or caught. */ thread_unsuspend(p); PROC_SUNLOCK(p); goto out; } if (prop & SA_STOP) { /* * If traced process is already stopped, * then no further action is necessary. */ if (p->p_flag & P_TRACED) goto out; /* * Already stopped, don't need to stop again * (If we did the shell could get confused). * Just make sure the signal STOP bit set. */ p->p_flag |= P_STOPPED_SIG; sigqueue_delete(sigqueue, sig); goto out; } /* * All other kinds of signals: * If a thread is sleeping interruptibly, simulate a * wakeup so that when it is continued it will be made * runnable and can look at the signal. However, don't make * the PROCESS runnable, leave it stopped. * It may run a bit until it hits a thread_suspend_check(). */ wakeup_swapper = 0; PROC_SLOCK(p); thread_lock(td); if (TD_ON_SLEEPQ(td) && (td->td_flags & TDF_SINTR)) wakeup_swapper = sleepq_abort(td, intrval); thread_unlock(td); PROC_SUNLOCK(p); if (wakeup_swapper) kick_proc0(); goto out; /* * Mutexes are short lived. Threads waiting on them will * hit thread_suspend_check() soon. */ } else if (p->p_state == PRS_NORMAL) { if (p->p_flag & P_TRACED || action == SIG_CATCH) { tdsigwakeup(td, sig, action, intrval); goto out; } MPASS(action == SIG_DFL); if (prop & SA_STOP) { if (p->p_flag & (P_PPWAIT|P_WEXIT)) goto out; p->p_flag |= P_STOPPED_SIG; p->p_xstat = sig; PROC_SLOCK(p); sig_suspend_threads(td, p, 1); if (p->p_numthreads == p->p_suspcount) { /* * only thread sending signal to another * process can reach here, if thread is sending * signal to its process, because thread does * not suspend itself here, p_numthreads * should never be equal to p_suspcount. */ thread_stopped(p); PROC_SUNLOCK(p); sigqueue_delete_proc(p, p->p_xstat); } else PROC_SUNLOCK(p); goto out; } } else { /* Not in "NORMAL" state. discard the signal. */ sigqueue_delete(sigqueue, sig); goto out; } /* * The process is not stopped so we need to apply the signal to all the * running threads. */ runfast: tdsigwakeup(td, sig, action, intrval); PROC_SLOCK(p); thread_unsuspend(p); PROC_SUNLOCK(p); out: /* If we jump here, proc slock should not be owned. */ PROC_SLOCK_ASSERT(p, MA_NOTOWNED); return (ret); } /* * The force of a signal has been directed against a single * thread. We need to see what we can do about knocking it * out of any sleep it may be in etc. */ static void tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval) { struct proc *p = td->td_proc; register int prop; int wakeup_swapper; wakeup_swapper = 0; PROC_LOCK_ASSERT(p, MA_OWNED); prop = sigprop(sig); PROC_SLOCK(p); thread_lock(td); /* * Bring the priority of a thread up if we want it to get * killed in this lifetime. Be careful to avoid bumping the * priority of the idle thread, since we still allow to signal * kernel processes. */ if (action == SIG_DFL && (prop & SA_KILL) != 0 && td->td_priority > PUSER && !TD_IS_IDLETHREAD(td)) sched_prio(td, PUSER); if (TD_ON_SLEEPQ(td)) { /* * If thread is sleeping uninterruptibly * we can't interrupt the sleep... the signal will * be noticed when the process returns through * trap() or syscall(). */ if ((td->td_flags & TDF_SINTR) == 0) goto out; /* * If SIGCONT is default (or ignored) and process is * asleep, we are finished; the process should not * be awakened. */ if ((prop & SA_CONT) && action == SIG_DFL) { thread_unlock(td); PROC_SUNLOCK(p); sigqueue_delete(&p->p_sigqueue, sig); /* * It may be on either list in this state. * Remove from both for now. */ sigqueue_delete(&td->td_sigqueue, sig); return; } /* * Don't awaken a sleeping thread for SIGSTOP if the * STOP signal is deferred. */ if ((prop & SA_STOP) && (td->td_flags & TDF_SBDRY)) goto out; /* * Give low priority threads a better chance to run. */ if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td)) sched_prio(td, PUSER); wakeup_swapper = sleepq_abort(td, intrval); } else { /* * Other states do nothing with the signal immediately, * other than kicking ourselves if we are running. * It will either never be noticed, or noticed very soon. */ #ifdef SMP if (TD_IS_RUNNING(td) && td != curthread) forward_signal(td); #endif } out: PROC_SUNLOCK(p); thread_unlock(td); if (wakeup_swapper) kick_proc0(); } static void sig_suspend_threads(struct thread *td, struct proc *p, int sending) { struct thread *td2; PROC_LOCK_ASSERT(p, MA_OWNED); PROC_SLOCK_ASSERT(p, MA_OWNED); FOREACH_THREAD_IN_PROC(p, td2) { thread_lock(td2); td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK; if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) && (td2->td_flags & TDF_SINTR)) { if (td2->td_flags & TDF_SBDRY) { /* * Once a thread is asleep with * TDF_SBDRY set, it should never * become suspended due to this check. */ KASSERT(!TD_IS_SUSPENDED(td2), ("thread with deferred stops suspended")); } else if (!TD_IS_SUSPENDED(td2)) { thread_suspend_one(td2); } } else if (!TD_IS_SUSPENDED(td2)) { if (sending || td != td2) td2->td_flags |= TDF_ASTPENDING; #ifdef SMP if (TD_IS_RUNNING(td2) && td2 != td) forward_signal(td2); #endif } thread_unlock(td2); } } int ptracestop(struct thread *td, int sig) { struct proc *p = td->td_proc; PROC_LOCK_ASSERT(p, MA_OWNED); KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process")); WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, &p->p_mtx.lock_object, "Stopping for traced signal"); td->td_dbgflags |= TDB_XSIG; td->td_xsig = sig; CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d", td->td_tid, p->p_pid, td->td_dbgflags, sig); PROC_SLOCK(p); while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) { if (p->p_flag & P_SINGLE_EXIT) { td->td_dbgflags &= ~TDB_XSIG; PROC_SUNLOCK(p); return (sig); } /* * Just make wait() to work, the last stopped thread * will win. */ p->p_xstat = sig; p->p_xthread = td; p->p_flag |= (P_STOPPED_SIG|P_STOPPED_TRACE); sig_suspend_threads(td, p, 0); if ((td->td_dbgflags & TDB_STOPATFORK) != 0) { td->td_dbgflags &= ~TDB_STOPATFORK; cv_broadcast(&p->p_dbgwait); } stopme: thread_suspend_switch(td, p); if (p->p_xthread == td) p->p_xthread = NULL; if (!(p->p_flag & P_TRACED)) break; if (td->td_dbgflags & TDB_SUSPEND) { if (p->p_flag & P_SINGLE_EXIT) break; goto stopme; } } PROC_SUNLOCK(p); return (td->td_xsig); } static void reschedule_signals(struct proc *p, sigset_t block, int flags) { struct sigacts *ps; struct thread *td; int sig; PROC_LOCK_ASSERT(p, MA_OWNED); ps = p->p_sigacts; mtx_assert(&ps->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0 ? MA_OWNED : MA_NOTOWNED); if (SIGISEMPTY(p->p_siglist)) return; SIGSETAND(block, p->p_siglist); while ((sig = sig_ffs(&block)) != 0) { SIGDELSET(block, sig); td = sigtd(p, sig, 0); signotify(td); if (!(flags & SIGPROCMASK_PS_LOCKED)) mtx_lock(&ps->ps_mtx); if (p->p_flag & P_TRACED || SIGISMEMBER(ps->ps_sigcatch, sig)) tdsigwakeup(td, sig, SIG_CATCH, (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR : ERESTART)); if (!(flags & SIGPROCMASK_PS_LOCKED)) mtx_unlock(&ps->ps_mtx); } } void tdsigcleanup(struct thread *td) { struct proc *p; sigset_t unblocked; p = td->td_proc; PROC_LOCK_ASSERT(p, MA_OWNED); sigqueue_flush(&td->td_sigqueue); if (p->p_numthreads == 1) return; /* * Since we cannot handle signals, notify signal post code * about this by filling the sigmask. * * Also, if needed, wake up thread(s) that do not block the * same signals as the exiting thread, since the thread might * have been selected for delivery and woken up. */ SIGFILLSET(unblocked); SIGSETNAND(unblocked, td->td_sigmask); SIGFILLSET(td->td_sigmask); reschedule_signals(p, unblocked, 0); } /* * Defer the delivery of SIGSTOP for the current thread. Returns true * if stops were deferred and false if they were already deferred. */ int sigdeferstop(void) { struct thread *td; td = curthread; if (td->td_flags & TDF_SBDRY) return (0); thread_lock(td); td->td_flags |= TDF_SBDRY; thread_unlock(td); return (1); } /* * Permit the delivery of SIGSTOP for the current thread. This does * not immediately suspend if a stop was posted. Instead, the thread * will suspend either via ast() or a subsequent interruptible sleep. */ int sigallowstop(void) { struct thread *td; int prev; td = curthread; thread_lock(td); prev = (td->td_flags & TDF_SBDRY) != 0; td->td_flags &= ~TDF_SBDRY; thread_unlock(td); return (prev); } /* * If the current process has received a signal (should be caught or cause * termination, should interrupt current syscall), return the signal number. * Stop signals with default action are processed immediately, then cleared; * they aren't returned. This is checked after each entry to the system for * a syscall or trap (though this can usually be done without calling issignal * by checking the pending signal masks in cursig.) The normal call * sequence is * * while (sig = cursig(curthread)) * postsig(sig); */ static int issignal(struct thread *td) { struct proc *p; struct sigacts *ps; struct sigqueue *queue; sigset_t sigpending; int sig, prop, newsig; p = td->td_proc; ps = p->p_sigacts; mtx_assert(&ps->ps_mtx, MA_OWNED); PROC_LOCK_ASSERT(p, MA_OWNED); for (;;) { int traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG); sigpending = td->td_sigqueue.sq_signals; SIGSETOR(sigpending, p->p_sigqueue.sq_signals); SIGSETNAND(sigpending, td->td_sigmask); if (p->p_flag & P_PPWAIT || td->td_flags & TDF_SBDRY) SIG_STOPSIGMASK(sigpending); if (SIGISEMPTY(sigpending)) /* no signal to send */ return (0); sig = sig_ffs(&sigpending); if (p->p_stops & S_SIG) { mtx_unlock(&ps->ps_mtx); stopevent(p, S_SIG, sig); mtx_lock(&ps->ps_mtx); } /* * We should see pending but ignored signals * only if P_TRACED was on when they were posted. */ if (SIGISMEMBER(ps->ps_sigignore, sig) && (traced == 0)) { sigqueue_delete(&td->td_sigqueue, sig); sigqueue_delete(&p->p_sigqueue, sig); continue; } if (p->p_flag & P_TRACED && (p->p_flag & P_PPTRACE) == 0) { /* * If traced, always stop. * Remove old signal from queue before the stop. * XXX shrug off debugger, it causes siginfo to * be thrown away. */ queue = &td->td_sigqueue; td->td_dbgksi.ksi_signo = 0; if (sigqueue_get(queue, sig, &td->td_dbgksi) == 0) { queue = &p->p_sigqueue; sigqueue_get(queue, sig, &td->td_dbgksi); } mtx_unlock(&ps->ps_mtx); newsig = ptracestop(td, sig); mtx_lock(&ps->ps_mtx); if (sig != newsig) { /* * If parent wants us to take the signal, * then it will leave it in p->p_xstat; * otherwise we just look for signals again. */ if (newsig == 0) continue; sig = newsig; /* * Put the new signal into td_sigqueue. If the * signal is being masked, look for other * signals. */ sigqueue_add(queue, sig, NULL); if (SIGISMEMBER(td->td_sigmask, sig)) continue; signotify(td); } else { if (td->td_dbgksi.ksi_signo != 0) { td->td_dbgksi.ksi_flags |= KSI_HEAD; if (sigqueue_add(&td->td_sigqueue, sig, &td->td_dbgksi) != 0) td->td_dbgksi.ksi_signo = 0; } if (td->td_dbgksi.ksi_signo == 0) sigqueue_add(&td->td_sigqueue, sig, NULL); } /* * If the traced bit got turned off, go back up * to the top to rescan signals. This ensures * that p_sig* and p_sigact are consistent. */ if ((p->p_flag & P_TRACED) == 0) continue; } prop = sigprop(sig); /* * Decide whether the signal should be returned. * Return the signal's number, or fall through * to clear it from the pending mask. */ switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) { case (intptr_t)SIG_DFL: /* * Don't take default actions on system processes. */ if (p->p_pid <= 1) { #ifdef DIAGNOSTIC /* * Are you sure you want to ignore SIGSEGV * in init? XXX */ printf("Process (pid %lu) got signal %d\n", (u_long)p->p_pid, sig); #endif break; /* == ignore */ } /* * If there is a pending stop signal to process * with default action, stop here, * then clear the signal. However, * if process is member of an orphaned * process group, ignore tty stop signals. */ if (prop & SA_STOP) { if (p->p_flag & (P_TRACED|P_WEXIT) || (p->p_pgrp->pg_jobc == 0 && prop & SA_TTYSTOP)) break; /* == ignore */ mtx_unlock(&ps->ps_mtx); WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, &p->p_mtx.lock_object, "Catching SIGSTOP"); p->p_flag |= P_STOPPED_SIG; p->p_xstat = sig; PROC_SLOCK(p); sig_suspend_threads(td, p, 0); thread_suspend_switch(td, p); PROC_SUNLOCK(p); mtx_lock(&ps->ps_mtx); break; } else if (prop & SA_IGNORE) { /* * Except for SIGCONT, shouldn't get here. * Default action is to ignore; drop it. */ break; /* == ignore */ } else return (sig); /*NOTREACHED*/ case (intptr_t)SIG_IGN: /* * Masking above should prevent us ever trying * to take action on an ignored signal other * than SIGCONT, unless process is traced. */ if ((prop & SA_CONT) == 0 && (p->p_flag & P_TRACED) == 0) printf("issignal\n"); break; /* == ignore */ default: /* * This signal has an action, let * postsig() process it. */ return (sig); } sigqueue_delete(&td->td_sigqueue, sig); /* take the signal! */ sigqueue_delete(&p->p_sigqueue, sig); } /* NOTREACHED */ } void thread_stopped(struct proc *p) { int n; PROC_LOCK_ASSERT(p, MA_OWNED); PROC_SLOCK_ASSERT(p, MA_OWNED); n = p->p_suspcount; if (p == curproc) n++; if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) { PROC_SUNLOCK(p); p->p_flag &= ~P_WAITED; PROC_LOCK(p->p_pptr); childproc_stopped(p, (p->p_flag & P_TRACED) ? CLD_TRAPPED : CLD_STOPPED); PROC_UNLOCK(p->p_pptr); PROC_SLOCK(p); } } /* * Take the action for the specified signal * from the current set of pending signals. */ int postsig(sig) register int sig; { struct thread *td = curthread; register struct proc *p = td->td_proc; struct sigacts *ps; sig_t action; ksiginfo_t ksi; sigset_t returnmask; KASSERT(sig != 0, ("postsig")); PROC_LOCK_ASSERT(p, MA_OWNED); ps = p->p_sigacts; mtx_assert(&ps->ps_mtx, MA_OWNED); ksiginfo_init(&ksi); if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 && sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0) return (0); ksi.ksi_signo = sig; if (ksi.ksi_code == SI_TIMER) itimer_accept(p, ksi.ksi_timerid, &ksi); action = ps->ps_sigact[_SIG_IDX(sig)]; #ifdef KTRACE if (KTRPOINT(td, KTR_PSIG)) ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ? &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code); #endif if (p->p_stops & S_SIG) { mtx_unlock(&ps->ps_mtx); stopevent(p, S_SIG, sig); mtx_lock(&ps->ps_mtx); } if (action == SIG_DFL) { /* * Default action, where the default is to kill * the process. (Other cases were ignored above.) */ mtx_unlock(&ps->ps_mtx); sigexit(td, sig); /* NOTREACHED */ } else { /* * If we get here, the signal must be caught. */ KASSERT(action != SIG_IGN && !SIGISMEMBER(td->td_sigmask, sig), ("postsig action")); /* * Set the new mask value and also defer further * occurrences of this signal. * * Special case: user has done a sigsuspend. Here the * current mask is not of interest, but rather the * mask from before the sigsuspend is what we want * restored after the signal processing is completed. */ if (td->td_pflags & TDP_OLDMASK) { returnmask = td->td_oldsigmask; td->td_pflags &= ~TDP_OLDMASK; } else returnmask = td->td_sigmask; if (p->p_sig == sig) { p->p_code = 0; p->p_sig = 0; } (*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask); postsig_done(sig, td, ps); } return (1); } /* * Kill the current process for stated reason. */ void killproc(p, why) struct proc *p; char *why; { PROC_LOCK_ASSERT(p, MA_OWNED); CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid, p->p_comm); log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n", p->p_pid, p->p_comm, p->p_ucred ? p->p_ucred->cr_uid : -1, why); p->p_flag |= P_WKILLED; kern_psignal(p, SIGKILL); } /* * Force the current process to exit with the specified signal, dumping core * if appropriate. We bypass the normal tests for masked and caught signals, * allowing unrecoverable failures to terminate the process without changing * signal state. Mark the accounting record with the signal termination. * If dumping core, save the signal number for the debugger. Calls exit and * does not return. */ void sigexit(td, sig) struct thread *td; int sig; { struct proc *p = td->td_proc; PROC_LOCK_ASSERT(p, MA_OWNED); p->p_acflag |= AXSIG; /* * We must be single-threading to generate a core dump. This * ensures that the registers in the core file are up-to-date. * Also, the ELF dump handler assumes that the thread list doesn't * change out from under it. * * XXX If another thread attempts to single-thread before us * (e.g. via fork()), we won't get a dump at all. */ if ((sigprop(sig) & SA_CORE) && thread_single(p, SINGLE_NO_EXIT) == 0) { p->p_sig = sig; /* * Log signals which would cause core dumps * (Log as LOG_INFO to appease those who don't want * these messages.) * XXX : Todo, as well as euid, write out ruid too * Note that coredump() drops proc lock. */ if (coredump(td) == 0) sig |= WCOREFLAG; if (kern_logsigexit) log(LOG_INFO, "pid %d (%s), uid %d: exited on signal %d%s\n", p->p_pid, p->p_comm, td->td_ucred ? td->td_ucred->cr_uid : -1, sig &~ WCOREFLAG, sig & WCOREFLAG ? " (core dumped)" : ""); } else PROC_UNLOCK(p); exit1(td, W_EXITCODE(0, sig)); /* NOTREACHED */ } /* * Send queued SIGCHLD to parent when child process's state * is changed. */ static void sigparent(struct proc *p, int reason, int status) { PROC_LOCK_ASSERT(p, MA_OWNED); PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED); if (p->p_ksi != NULL) { p->p_ksi->ksi_signo = SIGCHLD; p->p_ksi->ksi_code = reason; p->p_ksi->ksi_status = status; p->p_ksi->ksi_pid = p->p_pid; p->p_ksi->ksi_uid = p->p_ucred->cr_ruid; if (KSI_ONQ(p->p_ksi)) return; } pksignal(p->p_pptr, SIGCHLD, p->p_ksi); } static void childproc_jobstate(struct proc *p, int reason, int sig) { struct sigacts *ps; PROC_LOCK_ASSERT(p, MA_OWNED); PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED); /* * Wake up parent sleeping in kern_wait(), also send * SIGCHLD to parent, but SIGCHLD does not guarantee * that parent will awake, because parent may masked * the signal. */ p->p_pptr->p_flag |= P_STATCHILD; wakeup(p->p_pptr); ps = p->p_pptr->p_sigacts; mtx_lock(&ps->ps_mtx); if ((ps->ps_flag & PS_NOCLDSTOP) == 0) { mtx_unlock(&ps->ps_mtx); sigparent(p, reason, sig); } else mtx_unlock(&ps->ps_mtx); } void childproc_stopped(struct proc *p, int reason) { /* p_xstat is a plain signal number, not a full wait() status here. */ childproc_jobstate(p, reason, p->p_xstat); } void childproc_continued(struct proc *p) { childproc_jobstate(p, CLD_CONTINUED, SIGCONT); } void childproc_exited(struct proc *p) { int reason; int xstat = p->p_xstat; /* convert to int */ int status; if (WCOREDUMP(xstat)) reason = CLD_DUMPED, status = WTERMSIG(xstat); else if (WIFSIGNALED(xstat)) reason = CLD_KILLED, status = WTERMSIG(xstat); else reason = CLD_EXITED, status = WEXITSTATUS(xstat); /* * XXX avoid calling wakeup(p->p_pptr), the work is * done in exit1(). */ sigparent(p, reason, status); } /* * We only have 1 character for the core count in the format * string, so the range will be 0-9 */ #define MAX_NUM_CORES 10 static int num_cores = 5; static int sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS) { int error; int new_val; new_val = num_cores; error = sysctl_handle_int(oidp, &new_val, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (new_val > MAX_NUM_CORES) new_val = MAX_NUM_CORES; if (new_val < 0) new_val = 0; num_cores = new_val; return (0); } SYSCTL_PROC(_debug, OID_AUTO, ncores, CTLTYPE_INT|CTLFLAG_RW, 0, sizeof(int), sysctl_debug_num_cores_check, "I", ""); #define GZ_SUFFIX ".gz" #ifdef GZIO static int compress_user_cores = 1; SYSCTL_INT(_kern, OID_AUTO, compress_user_cores, CTLFLAG_RWTUN, &compress_user_cores, 0, "Compression of user corefiles"); int compress_user_cores_gzlevel = 6; SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_gzlevel, CTLFLAG_RWTUN, &compress_user_cores_gzlevel, 0, "Corefile gzip compression level"); #else static int compress_user_cores = 0; #endif /* * Protect the access to corefilename[] by allproc_lock. */ #define corefilename_lock allproc_lock static char corefilename[MAXPATHLEN] = {"%N.core"}; static int sysctl_kern_corefile(SYSCTL_HANDLER_ARGS) { int error; sx_xlock(&corefilename_lock); error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename), req); sx_xunlock(&corefilename_lock); return (error); } SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A", "Process corefile name format string"); /* * corefile_open(comm, uid, pid, td, compress, vpp, namep) * Expand the name described in corefilename, using name, uid, and pid * and open/create core file. * corefilename is a printf-like string, with three format specifiers: * %N name of process ("name") * %P process id (pid) * %U user id (uid) * For example, "%N.core" is the default; they can be disabled completely * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P". * This is controlled by the sysctl variable kern.corefile (see above). */ static int corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td, int compress, struct vnode **vpp, char **namep) { struct nameidata nd; struct sbuf sb; const char *format; char *hostname, *name; int indexpos, i, error, cmode, flags, oflags; hostname = NULL; format = corefilename; name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO); indexpos = -1; (void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN); sx_slock(&corefilename_lock); for (i = 0; format[i] != '\0'; i++) { switch (format[i]) { case '%': /* Format character */ i++; switch (format[i]) { case '%': sbuf_putc(&sb, '%'); break; case 'H': /* hostname */ if (hostname == NULL) { hostname = malloc(MAXHOSTNAMELEN, M_TEMP, M_WAITOK); } getcredhostname(td->td_ucred, hostname, MAXHOSTNAMELEN); sbuf_printf(&sb, "%s", hostname); break; case 'I': /* autoincrementing index */ sbuf_printf(&sb, "0"); indexpos = sbuf_len(&sb) - 1; break; case 'N': /* process name */ sbuf_printf(&sb, "%s", comm); break; case 'P': /* process id */ sbuf_printf(&sb, "%u", pid); break; case 'U': /* user id */ sbuf_printf(&sb, "%u", uid); break; default: log(LOG_ERR, "Unknown format character %c in " "corename `%s'\n", format[i], format); break; } break; default: sbuf_putc(&sb, format[i]); break; } } sx_sunlock(&corefilename_lock); free(hostname, M_TEMP); if (compress) sbuf_printf(&sb, GZ_SUFFIX); if (sbuf_error(&sb) != 0) { log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too " "long\n", (long)pid, comm, (u_long)uid); sbuf_delete(&sb); free(name, M_TEMP); return (ENOMEM); } sbuf_finish(&sb); sbuf_delete(&sb); cmode = S_IRUSR | S_IWUSR; oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE | (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0); /* * If the core format has a %I in it, then we need to check * for existing corefiles before returning a name. * To do this we iterate over 0..num_cores to find a * non-existing core file name to use. */ if (indexpos != -1) { for (i = 0; i < num_cores; i++) { flags = O_CREAT | O_EXCL | FWRITE | O_NOFOLLOW; name[indexpos] = '0' + i; NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td); error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred, NULL); if (error) { if (error == EEXIST) continue; log(LOG_ERR, "pid %d (%s), uid (%u): Path `%s' failed " "on initial open test, error = %d\n", pid, comm, uid, name, error); } goto out; } } flags = O_CREAT | FWRITE | O_NOFOLLOW; NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td); error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred, NULL); out: if (error) { #ifdef AUDIT audit_proc_coredump(td, name, error); #endif free(name, M_TEMP); return (error); } NDFREE(&nd, NDF_ONLY_PNBUF); *vpp = nd.ni_vp; *namep = name; return (0); } static int coredump_sanitise_path(const char *path) { size_t i; /* * Only send a subset of ASCII to devd(8) because it * might pass these strings to sh -c. */ for (i = 0; path[i]; i++) if (!(isalpha(path[i]) || isdigit(path[i])) && path[i] != '/' && path[i] != '.' && path[i] != '-') return (0); return (1); } /* * Dump a process' core. The main routine does some * policy checking, and creates the name of the coredump; * then it passes on a vnode and a size limit to the process-specific * coredump routine if there is one; if there _is not_ one, it returns * ENOSYS; otherwise it returns the error from the process-specific routine. */ static int coredump(struct thread *td) { struct proc *p = td->td_proc; struct ucred *cred = td->td_ucred; struct vnode *vp; struct flock lf; struct vattr vattr; int error, error1, locked; char *name; /* name of corefile */ void *rl_cookie; off_t limit; char *data = NULL; char *fullpath, *freepath = NULL; size_t len; static const char comm_name[] = "comm="; static const char core_name[] = "core="; PROC_LOCK_ASSERT(p, MA_OWNED); MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td); _STOPEVENT(p, S_CORE, 0); if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) || (p->p_flag2 & P2_NOTRACE) != 0) { PROC_UNLOCK(p); return (EFAULT); } /* * Note that the bulk of limit checking is done after * the corefile is created. The exception is if the limit * for corefiles is 0, in which case we don't bother * creating the corefile at all. This layout means that * a corefile is truncated instead of not being created, * if it is larger than the limit. */ - limit = (off_t)lim_cur(p, RLIMIT_CORE); + limit = (off_t)lim_cur(td, RLIMIT_CORE); if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) { PROC_UNLOCK(p); return (EFBIG); } PROC_UNLOCK(p); error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td, compress_user_cores, &vp, &name); if (error != 0) return (error); /* * Don't dump to non-regular files or files with links. * Do not dump into system files. */ if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 || vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0) { VOP_UNLOCK(vp, 0); error = EFAULT; goto out; } VOP_UNLOCK(vp, 0); /* Postpone other writers, including core dumps of other processes. */ rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; lf.l_type = F_WRLCK; locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0); VATTR_NULL(&vattr); vattr.va_size = 0; if (set_core_nodump_flag) vattr.va_flags = UF_NODUMP; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); VOP_SETATTR(vp, &vattr, cred); VOP_UNLOCK(vp, 0); PROC_LOCK(p); p->p_acflag |= ACORE; PROC_UNLOCK(p); if (p->p_sysent->sv_coredump != NULL) { error = p->p_sysent->sv_coredump(td, vp, limit, compress_user_cores ? IMGACT_CORE_COMPRESS : 0); } else { error = ENOSYS; } if (locked) { lf.l_type = F_UNLCK; VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK); } vn_rangelock_unlock(vp, rl_cookie); /* * Notify the userland helper that a process triggered a core dump. * This allows the helper to run an automated debugging session. */ if (error != 0 || coredump_devctl == 0) goto out; len = MAXPATHLEN * 2 + sizeof(comm_name) - 1 + sizeof(' ') + sizeof(core_name) - 1; data = malloc(len, M_TEMP, M_WAITOK); if (vn_fullpath_global(td, p->p_textvp, &fullpath, &freepath) != 0) goto out; if (!coredump_sanitise_path(fullpath)) goto out; snprintf(data, len, "%s%s ", comm_name, fullpath); free(freepath, M_TEMP); freepath = NULL; if (vn_fullpath_global(td, vp, &fullpath, &freepath) != 0) goto out; if (!coredump_sanitise_path(fullpath)) goto out; strlcat(data, core_name, len); strlcat(data, fullpath, len); devctl_notify("kernel", "signal", "coredump", data); out: error1 = vn_close(vp, FWRITE, cred, td); if (error == 0) error = error1; #ifdef AUDIT audit_proc_coredump(td, name, error); #endif free(freepath, M_TEMP); free(data, M_TEMP); free(name, M_TEMP); return (error); } /* * Nonexistent system call-- signal process (may want to handle it). Flag * error in case process won't see signal immediately (blocked or ignored). */ #ifndef _SYS_SYSPROTO_H_ struct nosys_args { int dummy; }; #endif /* ARGSUSED */ int nosys(td, args) struct thread *td; struct nosys_args *args; { struct proc *p = td->td_proc; PROC_LOCK(p); tdsignal(td, SIGSYS); PROC_UNLOCK(p); return (ENOSYS); } /* * Send a SIGIO or SIGURG signal to a process or process group using stored * credentials rather than those of the current process. */ void pgsigio(sigiop, sig, checkctty) struct sigio **sigiop; int sig, checkctty; { ksiginfo_t ksi; struct sigio *sigio; ksiginfo_init(&ksi); ksi.ksi_signo = sig; ksi.ksi_code = SI_KERNEL; SIGIO_LOCK(); sigio = *sigiop; if (sigio == NULL) { SIGIO_UNLOCK(); return; } if (sigio->sio_pgid > 0) { PROC_LOCK(sigio->sio_proc); if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred)) kern_psignal(sigio->sio_proc, sig); PROC_UNLOCK(sigio->sio_proc); } else if (sigio->sio_pgid < 0) { struct proc *p; PGRP_LOCK(sigio->sio_pgrp); LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) { PROC_LOCK(p); if (p->p_state == PRS_NORMAL && CANSIGIO(sigio->sio_ucred, p->p_ucred) && (checkctty == 0 || (p->p_flag & P_CONTROLT))) kern_psignal(p, sig); PROC_UNLOCK(p); } PGRP_UNLOCK(sigio->sio_pgrp); } SIGIO_UNLOCK(); } static int filt_sigattach(struct knote *kn) { struct proc *p = curproc; kn->kn_ptr.p_proc = p; kn->kn_flags |= EV_CLEAR; /* automatically set */ knlist_add(&p->p_klist, kn, 0); return (0); } static void filt_sigdetach(struct knote *kn) { struct proc *p = kn->kn_ptr.p_proc; knlist_remove(&p->p_klist, kn, 0); } /* * signal knotes are shared with proc knotes, so we apply a mask to * the hint in order to differentiate them from process hints. This * could be avoided by using a signal-specific knote list, but probably * isn't worth the trouble. */ static int filt_signal(struct knote *kn, long hint) { if (hint & NOTE_SIGNAL) { hint &= ~NOTE_SIGNAL; if (kn->kn_id == hint) kn->kn_data++; } return (kn->kn_data != 0); } struct sigacts * sigacts_alloc(void) { struct sigacts *ps; ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO); refcount_init(&ps->ps_refcnt, 1); mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF); return (ps); } void sigacts_free(struct sigacts *ps) { if (refcount_release(&ps->ps_refcnt) == 0) return; mtx_destroy(&ps->ps_mtx); free(ps, M_SUBPROC); } struct sigacts * sigacts_hold(struct sigacts *ps) { refcount_acquire(&ps->ps_refcnt); return (ps); } void sigacts_copy(struct sigacts *dest, struct sigacts *src) { KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest")); mtx_lock(&src->ps_mtx); bcopy(src, dest, offsetof(struct sigacts, ps_refcnt)); mtx_unlock(&src->ps_mtx); } int sigacts_shared(struct sigacts *ps) { return (ps->ps_refcnt > 1); } Index: head/sys/kern/kern_syscalls.c =================================================================== --- head/sys/kern/kern_syscalls.c (revision 284214) +++ head/sys/kern/kern_syscalls.c (revision 284215) @@ -1,230 +1,231 @@ /*- * Copyright (c) 1999 Assar Westerlund * 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. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include +#include #include #include #include #include #include #include /* * Acts like "nosys" but can be identified in sysent for dynamic call * number assignment for a limited number of calls. * * Place holder for system call slots reserved for loadable modules. */ int lkmnosys(struct thread *td, struct nosys_args *args) { return (nosys(td, args)); } int lkmressys(struct thread *td, struct nosys_args *args) { return (nosys(td, args)); } static void syscall_thread_drain(struct sysent *se) { u_int32_t cnt, oldcnt; do { oldcnt = se->sy_thrcnt; KASSERT((oldcnt & SY_THR_STATIC) == 0, ("drain on static syscall")); cnt = oldcnt | SY_THR_DRAINING; } while (atomic_cmpset_acq_32(&se->sy_thrcnt, oldcnt, cnt) == 0); while (atomic_cmpset_32(&se->sy_thrcnt, SY_THR_DRAINING, SY_THR_ABSENT) == 0) pause("scdrn", hz/2); } int syscall_thread_enter(struct thread *td, struct sysent *se) { u_int32_t cnt, oldcnt; do { oldcnt = se->sy_thrcnt; if ((oldcnt & SY_THR_STATIC) != 0) return (0); if ((oldcnt & (SY_THR_DRAINING | SY_THR_ABSENT)) != 0) return (ENOSYS); cnt = oldcnt + SY_THR_INCR; } while (atomic_cmpset_acq_32(&se->sy_thrcnt, oldcnt, cnt) == 0); return (0); } void syscall_thread_exit(struct thread *td, struct sysent *se) { u_int32_t cnt, oldcnt; do { oldcnt = se->sy_thrcnt; if ((oldcnt & SY_THR_STATIC) != 0) return; cnt = oldcnt - SY_THR_INCR; } while (atomic_cmpset_rel_32(&se->sy_thrcnt, oldcnt, cnt) == 0); } int syscall_register(int *offset, struct sysent *new_sysent, struct sysent *old_sysent, int flags) { int i; if ((flags & ~SY_THR_STATIC) != 0) return (EINVAL); if (*offset == NO_SYSCALL) { for (i = 1; i < SYS_MAXSYSCALL; ++i) if (sysent[i].sy_call == (sy_call_t *)lkmnosys) break; if (i == SYS_MAXSYSCALL) return (ENFILE); *offset = i; } else if (*offset < 0 || *offset >= SYS_MAXSYSCALL) return (EINVAL); else if (sysent[*offset].sy_call != (sy_call_t *)lkmnosys && sysent[*offset].sy_call != (sy_call_t *)lkmressys) return (EEXIST); KASSERT(sysent[*offset].sy_thrcnt == SY_THR_ABSENT, ("dynamic syscall is not protected")); *old_sysent = sysent[*offset]; new_sysent->sy_thrcnt = SY_THR_ABSENT; sysent[*offset] = *new_sysent; atomic_store_rel_32(&sysent[*offset].sy_thrcnt, flags); return (0); } int syscall_deregister(int *offset, struct sysent *old_sysent) { struct sysent *se; if (*offset == 0) return (0); /* XXX? */ se = &sysent[*offset]; if ((se->sy_thrcnt & SY_THR_STATIC) != 0) return (EINVAL); syscall_thread_drain(se); sysent[*offset] = *old_sysent; return (0); } int syscall_module_handler(struct module *mod, int what, void *arg) { struct syscall_module_data *data = arg; modspecific_t ms; int error; switch (what) { case MOD_LOAD: error = syscall_register(data->offset, data->new_sysent, &data->old_sysent, data->flags); if (error) { /* Leave a mark so we know to safely unload below. */ data->offset = NULL; return (error); } ms.intval = *data->offset; MOD_XLOCK; module_setspecific(mod, &ms); MOD_XUNLOCK; if (data->chainevh) error = data->chainevh(mod, what, data->chainarg); return (error); case MOD_UNLOAD: /* * MOD_LOAD failed, so just return without calling the * chained handler since we didn't pass along the MOD_LOAD * event. */ if (data->offset == NULL) return (0); if (data->chainevh) { error = data->chainevh(mod, what, data->chainarg); if (error) return error; } error = syscall_deregister(data->offset, &data->old_sysent); return (error); default: if (data->chainevh) return (data->chainevh(mod, what, data->chainarg)); return (EOPNOTSUPP); } /* NOTREACHED */ } int syscall_helper_register(struct syscall_helper_data *sd, int flags) { struct syscall_helper_data *sd1; int error; for (sd1 = sd; sd1->syscall_no != NO_SYSCALL; sd1++) { error = syscall_register(&sd1->syscall_no, &sd1->new_sysent, &sd1->old_sysent, flags); if (error != 0) { syscall_helper_unregister(sd); return (error); } sd1->registered = 1; } return (0); } int syscall_helper_unregister(struct syscall_helper_data *sd) { struct syscall_helper_data *sd1; for (sd1 = sd; sd1->registered != 0; sd1++) { syscall_deregister(&sd1->syscall_no, &sd1->old_sysent); sd1->registered = 0; } return (0); } Index: head/sys/kern/kern_thread.c =================================================================== --- head/sys/kern/kern_thread.c (revision 284214) +++ head/sys/kern/kern_thread.c (revision 284215) @@ -1,1189 +1,1195 @@ /*- * Copyright (C) 2001 Julian Elischer . * 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(s), this list of conditions and the following disclaimer as * the first lines of this file unmodified other than the possible * addition of one or more copyright notices. * 2. Redistributions in binary form must reproduce the above copyright * notice(s), 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 COPYRIGHT HOLDER(S) ``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 COPYRIGHT HOLDER(S) BE LIABLE FOR ANY * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH * DAMAGE. */ #include "opt_witness.h" #include "opt_hwpmc_hooks.h" #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef HWPMC_HOOKS #include #endif #include #include #include #include #include SDT_PROVIDER_DECLARE(proc); SDT_PROBE_DEFINE(proc, , , lwp__exit); /* * thread related storage. */ static uma_zone_t thread_zone; TAILQ_HEAD(, thread) zombie_threads = TAILQ_HEAD_INITIALIZER(zombie_threads); static struct mtx zombie_lock; MTX_SYSINIT(zombie_lock, &zombie_lock, "zombie lock", MTX_SPIN); static void thread_zombie(struct thread *); static int thread_unsuspend_one(struct thread *td, struct proc *p, bool boundary); #define TID_BUFFER_SIZE 1024 struct mtx tid_lock; static struct unrhdr *tid_unrhdr; static lwpid_t tid_buffer[TID_BUFFER_SIZE]; static int tid_head, tid_tail; static MALLOC_DEFINE(M_TIDHASH, "tidhash", "thread hash"); struct tidhashhead *tidhashtbl; u_long tidhash; struct rwlock tidhash_lock; static lwpid_t tid_alloc(void) { lwpid_t tid; tid = alloc_unr(tid_unrhdr); if (tid != -1) return (tid); mtx_lock(&tid_lock); if (tid_head == tid_tail) { mtx_unlock(&tid_lock); return (-1); } tid = tid_buffer[tid_head]; tid_head = (tid_head + 1) % TID_BUFFER_SIZE; mtx_unlock(&tid_lock); return (tid); } static void tid_free(lwpid_t tid) { lwpid_t tmp_tid = -1; mtx_lock(&tid_lock); if ((tid_tail + 1) % TID_BUFFER_SIZE == tid_head) { tmp_tid = tid_buffer[tid_head]; tid_head = (tid_head + 1) % TID_BUFFER_SIZE; } tid_buffer[tid_tail] = tid; tid_tail = (tid_tail + 1) % TID_BUFFER_SIZE; mtx_unlock(&tid_lock); if (tmp_tid != -1) free_unr(tid_unrhdr, tmp_tid); } /* * Prepare a thread for use. */ static int thread_ctor(void *mem, int size, void *arg, int flags) { struct thread *td; td = (struct thread *)mem; td->td_state = TDS_INACTIVE; td->td_oncpu = NOCPU; td->td_tid = tid_alloc(); /* * Note that td_critnest begins life as 1 because the thread is not * running and is thereby implicitly waiting to be on the receiving * end of a context switch. */ td->td_critnest = 1; td->td_lend_user_pri = PRI_MAX; EVENTHANDLER_INVOKE(thread_ctor, td); #ifdef AUDIT audit_thread_alloc(td); #endif umtx_thread_alloc(td); return (0); } /* * Reclaim a thread after use. */ static void thread_dtor(void *mem, int size, void *arg) { struct thread *td; td = (struct thread *)mem; #ifdef INVARIANTS /* Verify that this thread is in a safe state to free. */ switch (td->td_state) { case TDS_INHIBITED: case TDS_RUNNING: case TDS_CAN_RUN: case TDS_RUNQ: /* * We must never unlink a thread that is in one of * these states, because it is currently active. */ panic("bad state for thread unlinking"); /* NOTREACHED */ case TDS_INACTIVE: break; default: panic("bad thread state"); /* NOTREACHED */ } #endif #ifdef AUDIT audit_thread_free(td); #endif /* Free all OSD associated to this thread. */ osd_thread_exit(td); EVENTHANDLER_INVOKE(thread_dtor, td); tid_free(td->td_tid); } /* * Initialize type-stable parts of a thread (when newly created). */ static int thread_init(void *mem, int size, int flags) { struct thread *td; td = (struct thread *)mem; td->td_sleepqueue = sleepq_alloc(); td->td_turnstile = turnstile_alloc(); td->td_rlqe = NULL; EVENTHANDLER_INVOKE(thread_init, td); td->td_sched = (struct td_sched *)&td[1]; umtx_thread_init(td); td->td_kstack = 0; td->td_sel = NULL; return (0); } /* * Tear down type-stable parts of a thread (just before being discarded). */ static void thread_fini(void *mem, int size) { struct thread *td; td = (struct thread *)mem; EVENTHANDLER_INVOKE(thread_fini, td); rlqentry_free(td->td_rlqe); turnstile_free(td->td_turnstile); sleepq_free(td->td_sleepqueue); umtx_thread_fini(td); seltdfini(td); } /* * For a newly created process, * link up all the structures and its initial threads etc. * called from: * {arch}/{arch}/machdep.c {arch}_init(), init386() etc. * proc_dtor() (should go away) * proc_init() */ void proc_linkup0(struct proc *p, struct thread *td) { TAILQ_INIT(&p->p_threads); /* all threads in proc */ proc_linkup(p, td); } void proc_linkup(struct proc *p, struct thread *td) { sigqueue_init(&p->p_sigqueue, p); p->p_ksi = ksiginfo_alloc(1); if (p->p_ksi != NULL) { /* XXX p_ksi may be null if ksiginfo zone is not ready */ p->p_ksi->ksi_flags = KSI_EXT | KSI_INS; } LIST_INIT(&p->p_mqnotifier); p->p_numthreads = 0; thread_link(td, p); } /* * Initialize global thread allocation resources. */ void threadinit(void) { mtx_init(&tid_lock, "TID lock", NULL, MTX_DEF); /* * pid_max cannot be greater than PID_MAX. * leave one number for thread0. */ tid_unrhdr = new_unrhdr(PID_MAX + 2, INT_MAX, &tid_lock); thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(), thread_ctor, thread_dtor, thread_init, thread_fini, 16 - 1, 0); tidhashtbl = hashinit(maxproc / 2, M_TIDHASH, &tidhash); rw_init(&tidhash_lock, "tidhash"); } /* * Place an unused thread on the zombie list. * Use the slpq as that must be unused by now. */ void thread_zombie(struct thread *td) { mtx_lock_spin(&zombie_lock); TAILQ_INSERT_HEAD(&zombie_threads, td, td_slpq); mtx_unlock_spin(&zombie_lock); } /* * Release a thread that has exited after cpu_throw(). */ void thread_stash(struct thread *td) { atomic_subtract_rel_int(&td->td_proc->p_exitthreads, 1); thread_zombie(td); } /* * Reap zombie resources. */ void thread_reap(void) { struct thread *td_first, *td_next; /* * Don't even bother to lock if none at this instant, * we really don't care about the next instant.. */ if (!TAILQ_EMPTY(&zombie_threads)) { mtx_lock_spin(&zombie_lock); td_first = TAILQ_FIRST(&zombie_threads); if (td_first) TAILQ_INIT(&zombie_threads); mtx_unlock_spin(&zombie_lock); while (td_first) { td_next = TAILQ_NEXT(td_first, td_slpq); thread_cow_free(td_first); thread_free(td_first); td_first = td_next; } } } /* * Allocate a thread. */ struct thread * thread_alloc(int pages) { struct thread *td; thread_reap(); /* check if any zombies to get */ td = (struct thread *)uma_zalloc(thread_zone, M_WAITOK); KASSERT(td->td_kstack == 0, ("thread_alloc got thread with kstack")); if (!vm_thread_new(td, pages)) { uma_zfree(thread_zone, td); return (NULL); } cpu_thread_alloc(td); return (td); } int thread_alloc_stack(struct thread *td, int pages) { KASSERT(td->td_kstack == 0, ("thread_alloc_stack called on a thread with kstack")); if (!vm_thread_new(td, pages)) return (0); cpu_thread_alloc(td); return (1); } /* * Deallocate a thread. */ void thread_free(struct thread *td) { lock_profile_thread_exit(td); if (td->td_cpuset) cpuset_rel(td->td_cpuset); td->td_cpuset = NULL; cpu_thread_free(td); if (td->td_kstack != 0) vm_thread_dispose(td); uma_zfree(thread_zone, td); } void thread_cow_get_proc(struct thread *newtd, struct proc *p) { PROC_LOCK_ASSERT(p, MA_OWNED); newtd->td_ucred = crhold(p->p_ucred); + newtd->td_limit = lim_hold(p->p_limit); newtd->td_cowgen = p->p_cowgen; } void thread_cow_get(struct thread *newtd, struct thread *td) { newtd->td_ucred = crhold(td->td_ucred); + newtd->td_limit = lim_hold(td->td_limit); newtd->td_cowgen = td->td_cowgen; } void thread_cow_free(struct thread *td) { if (td->td_ucred) crfree(td->td_ucred); + if (td->td_limit) + lim_free(td->td_limit); } void thread_cow_update(struct thread *td) { struct proc *p; p = td->td_proc; PROC_LOCK(p); if (td->td_ucred != p->p_ucred) cred_update_thread(td); + if (td->td_limit != p->p_limit) + lim_update_thread(td); td->td_cowgen = p->p_cowgen; PROC_UNLOCK(p); } /* * Discard the current thread and exit from its context. * Always called with scheduler locked. * * Because we can't free a thread while we're operating under its context, * push the current thread into our CPU's deadthread holder. This means * we needn't worry about someone else grabbing our context before we * do a cpu_throw(). */ void thread_exit(void) { uint64_t runtime, new_switchtime; struct thread *td; struct thread *td2; struct proc *p; int wakeup_swapper; td = curthread; p = td->td_proc; PROC_SLOCK_ASSERT(p, MA_OWNED); mtx_assert(&Giant, MA_NOTOWNED); PROC_LOCK_ASSERT(p, MA_OWNED); KASSERT(p != NULL, ("thread exiting without a process")); CTR3(KTR_PROC, "thread_exit: thread %p (pid %ld, %s)", td, (long)p->p_pid, td->td_name); KASSERT(TAILQ_EMPTY(&td->td_sigqueue.sq_list), ("signal pending")); #ifdef AUDIT AUDIT_SYSCALL_EXIT(0, td); #endif /* * drop FPU & debug register state storage, or any other * architecture specific resources that * would not be on a new untouched process. */ cpu_thread_exit(td); /* XXXSMP */ /* * The last thread is left attached to the process * So that the whole bundle gets recycled. Skip * all this stuff if we never had threads. * EXIT clears all sign of other threads when * it goes to single threading, so the last thread always * takes the short path. */ if (p->p_flag & P_HADTHREADS) { if (p->p_numthreads > 1) { atomic_add_int(&td->td_proc->p_exitthreads, 1); thread_unlink(td); td2 = FIRST_THREAD_IN_PROC(p); sched_exit_thread(td2, td); /* * The test below is NOT true if we are the * sole exiting thread. P_STOPPED_SINGLE is unset * in exit1() after it is the only survivor. */ if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { if (p->p_numthreads == p->p_suspcount) { thread_lock(p->p_singlethread); wakeup_swapper = thread_unsuspend_one( p->p_singlethread, p, false); thread_unlock(p->p_singlethread); if (wakeup_swapper) kick_proc0(); } } PCPU_SET(deadthread, td); } else { /* * The last thread is exiting.. but not through exit() */ panic ("thread_exit: Last thread exiting on its own"); } } #ifdef HWPMC_HOOKS /* * If this thread is part of a process that is being tracked by hwpmc(4), * inform the module of the thread's impending exit. */ if (PMC_PROC_IS_USING_PMCS(td->td_proc)) PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT); #endif PROC_UNLOCK(p); PROC_STATLOCK(p); thread_lock(td); PROC_SUNLOCK(p); /* Do the same timestamp bookkeeping that mi_switch() would do. */ new_switchtime = cpu_ticks(); runtime = new_switchtime - PCPU_GET(switchtime); td->td_runtime += runtime; td->td_incruntime += runtime; PCPU_SET(switchtime, new_switchtime); PCPU_SET(switchticks, ticks); PCPU_INC(cnt.v_swtch); /* Save our resource usage in our process. */ td->td_ru.ru_nvcsw++; ruxagg(p, td); rucollect(&p->p_ru, &td->td_ru); PROC_STATUNLOCK(p); td->td_state = TDS_INACTIVE; #ifdef WITNESS witness_thread_exit(td); #endif CTR1(KTR_PROC, "thread_exit: cpu_throw() thread %p", td); sched_throw(td); panic("I'm a teapot!"); /* NOTREACHED */ } /* * Do any thread specific cleanups that may be needed in wait() * called with Giant, proc and schedlock not held. */ void thread_wait(struct proc *p) { struct thread *td; mtx_assert(&Giant, MA_NOTOWNED); KASSERT(p->p_numthreads == 1, ("multiple threads in thread_wait()")); KASSERT(p->p_exitthreads == 0, ("p_exitthreads leaking")); td = FIRST_THREAD_IN_PROC(p); /* Lock the last thread so we spin until it exits cpu_throw(). */ thread_lock(td); thread_unlock(td); lock_profile_thread_exit(td); cpuset_rel(td->td_cpuset); td->td_cpuset = NULL; cpu_thread_clean(td); thread_cow_free(td); thread_reap(); /* check for zombie threads etc. */ } /* * Link a thread to a process. * set up anything that needs to be initialized for it to * be used by the process. */ void thread_link(struct thread *td, struct proc *p) { /* * XXX This can't be enabled because it's called for proc0 before * its lock has been created. * PROC_LOCK_ASSERT(p, MA_OWNED); */ td->td_state = TDS_INACTIVE; td->td_proc = p; td->td_flags = TDF_INMEM; LIST_INIT(&td->td_contested); LIST_INIT(&td->td_lprof[0]); LIST_INIT(&td->td_lprof[1]); sigqueue_init(&td->td_sigqueue, p); callout_init(&td->td_slpcallout, 1); TAILQ_INSERT_TAIL(&p->p_threads, td, td_plist); p->p_numthreads++; } /* * Called from: * thread_exit() */ void thread_unlink(struct thread *td) { struct proc *p = td->td_proc; PROC_LOCK_ASSERT(p, MA_OWNED); TAILQ_REMOVE(&p->p_threads, td, td_plist); p->p_numthreads--; /* could clear a few other things here */ /* Must NOT clear links to proc! */ } static int calc_remaining(struct proc *p, int mode) { int remaining; PROC_LOCK_ASSERT(p, MA_OWNED); PROC_SLOCK_ASSERT(p, MA_OWNED); if (mode == SINGLE_EXIT) remaining = p->p_numthreads; else if (mode == SINGLE_BOUNDARY) remaining = p->p_numthreads - p->p_boundary_count; else if (mode == SINGLE_NO_EXIT || mode == SINGLE_ALLPROC) remaining = p->p_numthreads - p->p_suspcount; else panic("calc_remaining: wrong mode %d", mode); return (remaining); } static int remain_for_mode(int mode) { return (mode == SINGLE_ALLPROC ? 0 : 1); } static int weed_inhib(int mode, struct thread *td2, struct proc *p) { int wakeup_swapper; PROC_LOCK_ASSERT(p, MA_OWNED); PROC_SLOCK_ASSERT(p, MA_OWNED); THREAD_LOCK_ASSERT(td2, MA_OWNED); wakeup_swapper = 0; switch (mode) { case SINGLE_EXIT: if (TD_IS_SUSPENDED(td2)) wakeup_swapper |= thread_unsuspend_one(td2, p, true); if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) wakeup_swapper |= sleepq_abort(td2, EINTR); break; case SINGLE_BOUNDARY: if (TD_IS_SUSPENDED(td2) && (td2->td_flags & TDF_BOUNDARY) == 0) wakeup_swapper |= thread_unsuspend_one(td2, p, false); if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) wakeup_swapper |= sleepq_abort(td2, ERESTART); break; case SINGLE_NO_EXIT: if (TD_IS_SUSPENDED(td2) && (td2->td_flags & TDF_BOUNDARY) == 0) wakeup_swapper |= thread_unsuspend_one(td2, p, false); if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) wakeup_swapper |= sleepq_abort(td2, ERESTART); break; case SINGLE_ALLPROC: /* * ALLPROC suspend tries to avoid spurious EINTR for * threads sleeping interruptable, by suspending the * thread directly, similarly to sig_suspend_threads(). * Since such sleep is not performed at the user * boundary, TDF_BOUNDARY flag is not set, and TDF_ALLPROCSUSP * is used to avoid immediate un-suspend. */ if (TD_IS_SUSPENDED(td2) && (td2->td_flags & (TDF_BOUNDARY | TDF_ALLPROCSUSP)) == 0) wakeup_swapper |= thread_unsuspend_one(td2, p, false); if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) { if ((td2->td_flags & TDF_SBDRY) == 0) { thread_suspend_one(td2); td2->td_flags |= TDF_ALLPROCSUSP; } else { wakeup_swapper |= sleepq_abort(td2, ERESTART); } } break; } return (wakeup_swapper); } /* * Enforce single-threading. * * Returns 1 if the caller must abort (another thread is waiting to * exit the process or similar). Process is locked! * Returns 0 when you are successfully the only thread running. * A process has successfully single threaded in the suspend mode when * There are no threads in user mode. Threads in the kernel must be * allowed to continue until they get to the user boundary. They may even * copy out their return values and data before suspending. They may however be * accelerated in reaching the user boundary as we will wake up * any sleeping threads that are interruptable. (PCATCH). */ int thread_single(struct proc *p, int mode) { struct thread *td; struct thread *td2; int remaining, wakeup_swapper; td = curthread; KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY || mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT, ("invalid mode %d", mode)); /* * If allowing non-ALLPROC singlethreading for non-curproc * callers, calc_remaining() and remain_for_mode() should be * adjusted to also account for td->td_proc != p. For now * this is not implemented because it is not used. */ KASSERT((mode == SINGLE_ALLPROC && td->td_proc != p) || (mode != SINGLE_ALLPROC && td->td_proc == p), ("mode %d proc %p curproc %p", mode, p, td->td_proc)); mtx_assert(&Giant, MA_NOTOWNED); PROC_LOCK_ASSERT(p, MA_OWNED); if ((p->p_flag & P_HADTHREADS) == 0 && mode != SINGLE_ALLPROC) return (0); /* Is someone already single threading? */ if (p->p_singlethread != NULL && p->p_singlethread != td) return (1); if (mode == SINGLE_EXIT) { p->p_flag |= P_SINGLE_EXIT; p->p_flag &= ~P_SINGLE_BOUNDARY; } else { p->p_flag &= ~P_SINGLE_EXIT; if (mode == SINGLE_BOUNDARY) p->p_flag |= P_SINGLE_BOUNDARY; else p->p_flag &= ~P_SINGLE_BOUNDARY; } if (mode == SINGLE_ALLPROC) p->p_flag |= P_TOTAL_STOP; p->p_flag |= P_STOPPED_SINGLE; PROC_SLOCK(p); p->p_singlethread = td; remaining = calc_remaining(p, mode); while (remaining != remain_for_mode(mode)) { if (P_SHOULDSTOP(p) != P_STOPPED_SINGLE) goto stopme; wakeup_swapper = 0; FOREACH_THREAD_IN_PROC(p, td2) { if (td2 == td) continue; thread_lock(td2); td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK; if (TD_IS_INHIBITED(td2)) { wakeup_swapper |= weed_inhib(mode, td2, p); #ifdef SMP } else if (TD_IS_RUNNING(td2) && td != td2) { forward_signal(td2); #endif } thread_unlock(td2); } if (wakeup_swapper) kick_proc0(); remaining = calc_remaining(p, mode); /* * Maybe we suspended some threads.. was it enough? */ if (remaining == remain_for_mode(mode)) break; stopme: /* * Wake us up when everyone else has suspended. * In the mean time we suspend as well. */ thread_suspend_switch(td, p); remaining = calc_remaining(p, mode); } if (mode == SINGLE_EXIT) { /* * Convert the process to an unthreaded process. The * SINGLE_EXIT is called by exit1() or execve(), in * both cases other threads must be retired. */ KASSERT(p->p_numthreads == 1, ("Unthreading with >1 threads")); p->p_singlethread = NULL; p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_HADTHREADS); /* * Wait for any remaining threads to exit cpu_throw(). */ while (p->p_exitthreads != 0) { PROC_SUNLOCK(p); PROC_UNLOCK(p); sched_relinquish(td); PROC_LOCK(p); PROC_SLOCK(p); } } else if (mode == SINGLE_BOUNDARY) { /* * Wait until all suspended threads are removed from * the processors. The thread_suspend_check() * increments p_boundary_count while it is still * running, which makes it possible for the execve() * to destroy vmspace while our other threads are * still using the address space. * * We lock the thread, which is only allowed to * succeed after context switch code finished using * the address space. */ FOREACH_THREAD_IN_PROC(p, td2) { if (td2 == td) continue; thread_lock(td2); KASSERT((td2->td_flags & TDF_BOUNDARY) != 0, ("td %p not on boundary", td2)); KASSERT(TD_IS_SUSPENDED(td2), ("td %p is not suspended", td2)); thread_unlock(td2); } } PROC_SUNLOCK(p); return (0); } bool thread_suspend_check_needed(void) { struct proc *p; struct thread *td; td = curthread; p = td->td_proc; PROC_LOCK_ASSERT(p, MA_OWNED); return (P_SHOULDSTOP(p) || ((p->p_flag & P_TRACED) != 0 && (td->td_dbgflags & TDB_SUSPEND) != 0)); } /* * Called in from locations that can safely check to see * whether we have to suspend or at least throttle for a * single-thread event (e.g. fork). * * Such locations include userret(). * If the "return_instead" argument is non zero, the thread must be able to * accept 0 (caller may continue), or 1 (caller must abort) as a result. * * The 'return_instead' argument tells the function if it may do a * thread_exit() or suspend, or whether the caller must abort and back * out instead. * * If the thread that set the single_threading request has set the * P_SINGLE_EXIT bit in the process flags then this call will never return * if 'return_instead' is false, but will exit. * * P_SINGLE_EXIT | return_instead == 0| return_instead != 0 *---------------+--------------------+--------------------- * 0 | returns 0 | returns 0 or 1 * | when ST ends | immediately *---------------+--------------------+--------------------- * 1 | thread exits | returns 1 * | | immediately * 0 = thread_exit() or suspension ok, * other = return error instead of stopping the thread. * * While a full suspension is under effect, even a single threading * thread would be suspended if it made this call (but it shouldn't). * This call should only be made from places where * thread_exit() would be safe as that may be the outcome unless * return_instead is set. */ int thread_suspend_check(int return_instead) { struct thread *td; struct proc *p; int wakeup_swapper; td = curthread; p = td->td_proc; mtx_assert(&Giant, MA_NOTOWNED); PROC_LOCK_ASSERT(p, MA_OWNED); while (thread_suspend_check_needed()) { if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { KASSERT(p->p_singlethread != NULL, ("singlethread not set")); /* * The only suspension in action is a * single-threading. Single threader need not stop. * XXX Should be safe to access unlocked * as it can only be set to be true by us. */ if (p->p_singlethread == td) return (0); /* Exempt from stopping. */ } if ((p->p_flag & P_SINGLE_EXIT) && return_instead) return (EINTR); /* Should we goto user boundary if we didn't come from there? */ if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE && (p->p_flag & P_SINGLE_BOUNDARY) && return_instead) return (ERESTART); /* * Ignore suspend requests if they are deferred. */ if ((td->td_flags & TDF_SBDRY) != 0) { KASSERT(return_instead, ("TDF_SBDRY set for unsafe thread_suspend_check")); return (0); } /* * If the process is waiting for us to exit, * this thread should just suicide. * Assumes that P_SINGLE_EXIT implies P_STOPPED_SINGLE. */ if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td)) { PROC_UNLOCK(p); tidhash_remove(td); /* * Allow Linux emulation layer to do some work * before thread suicide. */ if (__predict_false(p->p_sysent->sv_thread_detach != NULL)) (p->p_sysent->sv_thread_detach)(td); PROC_LOCK(p); tdsigcleanup(td); umtx_thread_exit(td); PROC_SLOCK(p); thread_stopped(p); thread_exit(); } PROC_SLOCK(p); thread_stopped(p); if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { if (p->p_numthreads == p->p_suspcount + 1) { thread_lock(p->p_singlethread); wakeup_swapper = thread_unsuspend_one( p->p_singlethread, p, false); thread_unlock(p->p_singlethread); if (wakeup_swapper) kick_proc0(); } } PROC_UNLOCK(p); thread_lock(td); /* * When a thread suspends, it just * gets taken off all queues. */ thread_suspend_one(td); if (return_instead == 0) { p->p_boundary_count++; td->td_flags |= TDF_BOUNDARY; } PROC_SUNLOCK(p); mi_switch(SW_INVOL | SWT_SUSPEND, NULL); thread_unlock(td); PROC_LOCK(p); } return (0); } void thread_suspend_switch(struct thread *td, struct proc *p) { KASSERT(!TD_IS_SUSPENDED(td), ("already suspended")); PROC_LOCK_ASSERT(p, MA_OWNED); PROC_SLOCK_ASSERT(p, MA_OWNED); /* * We implement thread_suspend_one in stages here to avoid * dropping the proc lock while the thread lock is owned. */ if (p == td->td_proc) { thread_stopped(p); p->p_suspcount++; } PROC_UNLOCK(p); thread_lock(td); td->td_flags &= ~TDF_NEEDSUSPCHK; TD_SET_SUSPENDED(td); sched_sleep(td, 0); PROC_SUNLOCK(p); DROP_GIANT(); mi_switch(SW_VOL | SWT_SUSPEND, NULL); thread_unlock(td); PICKUP_GIANT(); PROC_LOCK(p); PROC_SLOCK(p); } void thread_suspend_one(struct thread *td) { struct proc *p; p = td->td_proc; PROC_SLOCK_ASSERT(p, MA_OWNED); THREAD_LOCK_ASSERT(td, MA_OWNED); KASSERT(!TD_IS_SUSPENDED(td), ("already suspended")); p->p_suspcount++; td->td_flags &= ~TDF_NEEDSUSPCHK; TD_SET_SUSPENDED(td); sched_sleep(td, 0); } static int thread_unsuspend_one(struct thread *td, struct proc *p, bool boundary) { THREAD_LOCK_ASSERT(td, MA_OWNED); KASSERT(TD_IS_SUSPENDED(td), ("Thread not suspended")); TD_CLR_SUSPENDED(td); td->td_flags &= ~TDF_ALLPROCSUSP; if (td->td_proc == p) { PROC_SLOCK_ASSERT(p, MA_OWNED); p->p_suspcount--; if (boundary && (td->td_flags & TDF_BOUNDARY) != 0) { td->td_flags &= ~TDF_BOUNDARY; p->p_boundary_count--; } } return (setrunnable(td)); } /* * Allow all threads blocked by single threading to continue running. */ void thread_unsuspend(struct proc *p) { struct thread *td; int wakeup_swapper; PROC_LOCK_ASSERT(p, MA_OWNED); PROC_SLOCK_ASSERT(p, MA_OWNED); wakeup_swapper = 0; if (!P_SHOULDSTOP(p)) { FOREACH_THREAD_IN_PROC(p, td) { thread_lock(td); if (TD_IS_SUSPENDED(td)) { wakeup_swapper |= thread_unsuspend_one(td, p, true); } thread_unlock(td); } } else if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE && p->p_numthreads == p->p_suspcount) { /* * Stopping everything also did the job for the single * threading request. Now we've downgraded to single-threaded, * let it continue. */ if (p->p_singlethread->td_proc == p) { thread_lock(p->p_singlethread); wakeup_swapper = thread_unsuspend_one( p->p_singlethread, p, false); thread_unlock(p->p_singlethread); } } if (wakeup_swapper) kick_proc0(); } /* * End the single threading mode.. */ void thread_single_end(struct proc *p, int mode) { struct thread *td; int wakeup_swapper; KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY || mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT, ("invalid mode %d", mode)); PROC_LOCK_ASSERT(p, MA_OWNED); KASSERT((mode == SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) != 0) || (mode != SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) == 0), ("mode %d does not match P_TOTAL_STOP", mode)); KASSERT(mode == SINGLE_ALLPROC || p->p_singlethread == curthread, ("thread_single_end from other thread %p %p", curthread, p->p_singlethread)); KASSERT(mode != SINGLE_BOUNDARY || (p->p_flag & P_SINGLE_BOUNDARY) != 0, ("mis-matched SINGLE_BOUNDARY flags %x", p->p_flag)); p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_SINGLE_BOUNDARY | P_TOTAL_STOP); PROC_SLOCK(p); p->p_singlethread = NULL; wakeup_swapper = 0; /* * If there are other threads they may now run, * unless of course there is a blanket 'stop order' * on the process. The single threader must be allowed * to continue however as this is a bad place to stop. */ if (p->p_numthreads != remain_for_mode(mode) && !P_SHOULDSTOP(p)) { FOREACH_THREAD_IN_PROC(p, td) { thread_lock(td); if (TD_IS_SUSPENDED(td)) { wakeup_swapper |= thread_unsuspend_one(td, p, mode == SINGLE_BOUNDARY); } thread_unlock(td); } } KASSERT(mode != SINGLE_BOUNDARY || p->p_boundary_count == 0, ("inconsistent boundary count %d", p->p_boundary_count)); PROC_SUNLOCK(p); if (wakeup_swapper) kick_proc0(); } struct thread * thread_find(struct proc *p, lwpid_t tid) { struct thread *td; PROC_LOCK_ASSERT(p, MA_OWNED); FOREACH_THREAD_IN_PROC(p, td) { if (td->td_tid == tid) break; } return (td); } /* Locate a thread by number; return with proc lock held. */ struct thread * tdfind(lwpid_t tid, pid_t pid) { #define RUN_THRESH 16 struct thread *td; int run = 0; rw_rlock(&tidhash_lock); LIST_FOREACH(td, TIDHASH(tid), td_hash) { if (td->td_tid == tid) { if (pid != -1 && td->td_proc->p_pid != pid) { td = NULL; break; } PROC_LOCK(td->td_proc); if (td->td_proc->p_state == PRS_NEW) { PROC_UNLOCK(td->td_proc); td = NULL; break; } if (run > RUN_THRESH) { if (rw_try_upgrade(&tidhash_lock)) { LIST_REMOVE(td, td_hash); LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash); rw_wunlock(&tidhash_lock); return (td); } } break; } run++; } rw_runlock(&tidhash_lock); return (td); } void tidhash_add(struct thread *td) { rw_wlock(&tidhash_lock); LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash); rw_wunlock(&tidhash_lock); } void tidhash_remove(struct thread *td) { rw_wlock(&tidhash_lock); LIST_REMOVE(td, td_hash); rw_wunlock(&tidhash_lock); } Index: head/sys/kern/subr_uio.c =================================================================== --- head/sys/kern/subr_uio.c (revision 284214) +++ head/sys/kern/subr_uio.c (revision 284215) @@ -1,570 +1,568 @@ /*- * Copyright (c) 1982, 1986, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Copyright (c) 2014 The FreeBSD Foundation * * Portions of this software were developed by Konstantin Belousov * under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kern_subr.c 8.3 (Berkeley) 1/21/94 */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include SYSCTL_INT(_kern, KERN_IOV_MAX, iov_max, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, UIO_MAXIOV, "Maximum number of elements in an I/O vector; sysconf(_SC_IOV_MAX)"); static int uiomove_faultflag(void *cp, int n, struct uio *uio, int nofault); int copyin_nofault(const void *udaddr, void *kaddr, size_t len) { int error, save; save = vm_fault_disable_pagefaults(); error = copyin(udaddr, kaddr, len); vm_fault_enable_pagefaults(save); return (error); } int copyout_nofault(const void *kaddr, void *udaddr, size_t len) { int error, save; save = vm_fault_disable_pagefaults(); error = copyout(kaddr, udaddr, len); vm_fault_enable_pagefaults(save); return (error); } #define PHYS_PAGE_COUNT(len) (howmany(len, PAGE_SIZE) + 1) int physcopyin(void *src, vm_paddr_t dst, size_t len) { vm_page_t m[PHYS_PAGE_COUNT(len)]; struct iovec iov[1]; struct uio uio; int i; iov[0].iov_base = src; iov[0].iov_len = len; uio.uio_iov = iov; uio.uio_iovcnt = 1; uio.uio_offset = 0; uio.uio_resid = len; uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_WRITE; for (i = 0; i < PHYS_PAGE_COUNT(len); i++, dst += PAGE_SIZE) m[i] = PHYS_TO_VM_PAGE(dst); return (uiomove_fromphys(m, dst & PAGE_MASK, len, &uio)); } int physcopyout(vm_paddr_t src, void *dst, size_t len) { vm_page_t m[PHYS_PAGE_COUNT(len)]; struct iovec iov[1]; struct uio uio; int i; iov[0].iov_base = dst; iov[0].iov_len = len; uio.uio_iov = iov; uio.uio_iovcnt = 1; uio.uio_offset = 0; uio.uio_resid = len; uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_READ; for (i = 0; i < PHYS_PAGE_COUNT(len); i++, src += PAGE_SIZE) m[i] = PHYS_TO_VM_PAGE(src); return (uiomove_fromphys(m, src & PAGE_MASK, len, &uio)); } #undef PHYS_PAGE_COUNT int uiomove(void *cp, int n, struct uio *uio) { return (uiomove_faultflag(cp, n, uio, 0)); } int uiomove_nofault(void *cp, int n, struct uio *uio) { return (uiomove_faultflag(cp, n, uio, 1)); } static int uiomove_faultflag(void *cp, int n, struct uio *uio, int nofault) { struct thread *td; struct iovec *iov; size_t cnt; int error, newflags, save; td = curthread; error = 0; KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE, ("uiomove: mode")); KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == td, ("uiomove proc")); if (!nofault) WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Calling uiomove()"); /* XXX does it make a sense to set TDP_DEADLKTREAT for UIO_SYSSPACE ? */ newflags = TDP_DEADLKTREAT; if (uio->uio_segflg == UIO_USERSPACE && nofault) { /* * Fail if a non-spurious page fault occurs. */ newflags |= TDP_NOFAULTING | TDP_RESETSPUR; } save = curthread_pflags_set(newflags); while (n > 0 && uio->uio_resid) { iov = uio->uio_iov; cnt = iov->iov_len; if (cnt == 0) { uio->uio_iov++; uio->uio_iovcnt--; continue; } if (cnt > n) cnt = n; switch (uio->uio_segflg) { case UIO_USERSPACE: maybe_yield(); if (uio->uio_rw == UIO_READ) error = copyout(cp, iov->iov_base, cnt); else error = copyin(iov->iov_base, cp, cnt); if (error) goto out; break; case UIO_SYSSPACE: if (uio->uio_rw == UIO_READ) bcopy(cp, iov->iov_base, cnt); else bcopy(iov->iov_base, cp, cnt); break; case UIO_NOCOPY: break; } iov->iov_base = (char *)iov->iov_base + cnt; iov->iov_len -= cnt; uio->uio_resid -= cnt; uio->uio_offset += cnt; cp = (char *)cp + cnt; n -= cnt; } out: curthread_pflags_restore(save); return (error); } /* * Wrapper for uiomove() that validates the arguments against a known-good * kernel buffer. Currently, uiomove accepts a signed (n) argument, which * is almost definitely a bad thing, so we catch that here as well. We * return a runtime failure, but it might be desirable to generate a runtime * assertion failure instead. */ int uiomove_frombuf(void *buf, int buflen, struct uio *uio) { size_t offset, n; if (uio->uio_offset < 0 || uio->uio_resid < 0 || (offset = uio->uio_offset) != uio->uio_offset) return (EINVAL); if (buflen <= 0 || offset >= buflen) return (0); if ((n = buflen - offset) > IOSIZE_MAX) return (EINVAL); return (uiomove((char *)buf + offset, n, uio)); } /* * Give next character to user as result of read. */ int ureadc(int c, struct uio *uio) { struct iovec *iov; char *iov_base; WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Calling ureadc()"); again: if (uio->uio_iovcnt == 0 || uio->uio_resid == 0) panic("ureadc"); iov = uio->uio_iov; if (iov->iov_len == 0) { uio->uio_iovcnt--; uio->uio_iov++; goto again; } switch (uio->uio_segflg) { case UIO_USERSPACE: if (subyte(iov->iov_base, c) < 0) return (EFAULT); break; case UIO_SYSSPACE: iov_base = iov->iov_base; *iov_base = c; break; case UIO_NOCOPY: break; } iov->iov_base = (char *)iov->iov_base + 1; iov->iov_len--; uio->uio_resid--; uio->uio_offset++; return (0); } int copyinfrom(const void * __restrict src, void * __restrict dst, size_t len, int seg) { int error = 0; switch (seg) { case UIO_USERSPACE: error = copyin(src, dst, len); break; case UIO_SYSSPACE: bcopy(src, dst, len); break; default: panic("copyinfrom: bad seg %d\n", seg); } return (error); } int copyinstrfrom(const void * __restrict src, void * __restrict dst, size_t len, size_t * __restrict copied, int seg) { int error = 0; switch (seg) { case UIO_USERSPACE: error = copyinstr(src, dst, len, copied); break; case UIO_SYSSPACE: error = copystr(src, dst, len, copied); break; default: panic("copyinstrfrom: bad seg %d\n", seg); } return (error); } int copyiniov(const struct iovec *iovp, u_int iovcnt, struct iovec **iov, int error) { u_int iovlen; *iov = NULL; if (iovcnt > UIO_MAXIOV) return (error); iovlen = iovcnt * sizeof (struct iovec); *iov = malloc(iovlen, M_IOV, M_WAITOK); error = copyin(iovp, *iov, iovlen); if (error) { free(*iov, M_IOV); *iov = NULL; } return (error); } int copyinuio(const struct iovec *iovp, u_int iovcnt, struct uio **uiop) { struct iovec *iov; struct uio *uio; u_int iovlen; int error, i; *uiop = NULL; if (iovcnt > UIO_MAXIOV) return (EINVAL); iovlen = iovcnt * sizeof (struct iovec); uio = malloc(iovlen + sizeof *uio, M_IOV, M_WAITOK); iov = (struct iovec *)(uio + 1); error = copyin(iovp, iov, iovlen); if (error) { free(uio, M_IOV); return (error); } uio->uio_iov = iov; uio->uio_iovcnt = iovcnt; uio->uio_segflg = UIO_USERSPACE; uio->uio_offset = -1; uio->uio_resid = 0; for (i = 0; i < iovcnt; i++) { if (iov->iov_len > IOSIZE_MAX - uio->uio_resid) { free(uio, M_IOV); return (EINVAL); } uio->uio_resid += iov->iov_len; iov++; } *uiop = uio; return (0); } struct uio * cloneuio(struct uio *uiop) { struct uio *uio; int iovlen; iovlen = uiop->uio_iovcnt * sizeof (struct iovec); uio = malloc(iovlen + sizeof *uio, M_IOV, M_WAITOK); *uio = *uiop; uio->uio_iov = (struct iovec *)(uio + 1); bcopy(uiop->uio_iov, uio->uio_iov, iovlen); return (uio); } /* * Map some anonymous memory in user space of size sz, rounded up to the page * boundary. */ int copyout_map(struct thread *td, vm_offset_t *addr, size_t sz) { struct vmspace *vms; int error; vm_size_t size; vms = td->td_proc->p_vmspace; /* * Map somewhere after heap in process memory. */ - PROC_LOCK(td->td_proc); *addr = round_page((vm_offset_t)vms->vm_daddr + - lim_max(td->td_proc, RLIMIT_DATA)); - PROC_UNLOCK(td->td_proc); + lim_max(td, RLIMIT_DATA)); /* round size up to page boundry */ size = (vm_size_t)round_page(sz); error = vm_mmap(&vms->vm_map, addr, size, VM_PROT_READ | VM_PROT_WRITE, VM_PROT_ALL, MAP_PRIVATE | MAP_ANON, OBJT_DEFAULT, NULL, 0); return (error); } /* * Unmap memory in user space. */ int copyout_unmap(struct thread *td, vm_offset_t addr, size_t sz) { vm_map_t map; vm_size_t size; if (sz == 0) return (0); map = &td->td_proc->p_vmspace->vm_map; size = (vm_size_t)round_page(sz); if (vm_map_remove(map, addr, addr + size) != KERN_SUCCESS) return (EINVAL); return (0); } #ifdef NO_FUEWORD /* * XXXKIB The temporal implementation of fue*() functions which do not * handle usermode -1 properly, mixing it with the fault code. Keep * this until MD code is written. Currently sparc64, mips and arm do * not have proper implementation. */ int fueword(volatile const void *base, long *val) { long res; res = fuword(base); if (res == -1) return (-1); *val = res; return (0); } int fueword32(volatile const void *base, int32_t *val) { int32_t res; res = fuword32(base); if (res == -1) return (-1); *val = res; return (0); } #ifdef _LP64 int fueword64(volatile const void *base, int64_t *val) { int32_t res; res = fuword64(base); if (res == -1) return (-1); *val = res; return (0); } #endif int casueword32(volatile uint32_t *base, uint32_t oldval, uint32_t *oldvalp, uint32_t newval) { int32_t ov; ov = casuword32(base, oldval, newval); if (ov == -1) return (-1); *oldvalp = ov; return (0); } int casueword(volatile u_long *p, u_long oldval, u_long *oldvalp, u_long newval) { u_long ov; ov = casuword(p, oldval, newval); if (ov == -1) return (-1); *oldvalp = ov; return (0); } #else /* NO_FUEWORD */ int32_t fuword32(volatile const void *addr) { int rv; int32_t val; rv = fueword32(addr, &val); return (rv == -1 ? -1 : val); } #ifdef _LP64 int64_t fuword64(volatile const void *addr) { int rv; int64_t val; rv = fueword64(addr, &val); return (rv == -1 ? -1 : val); } #endif /* _LP64 */ long fuword(volatile const void *addr) { long val; int rv; rv = fueword(addr, &val); return (rv == -1 ? -1 : val); } uint32_t casuword32(volatile uint32_t *addr, uint32_t old, uint32_t new) { int rv; uint32_t val; rv = casueword32(addr, old, &val, new); return (rv == -1 ? -1 : val); } u_long casuword(volatile u_long *addr, u_long old, u_long new) { int rv; u_long val; rv = casueword(addr, old, &val, new); return (rv == -1 ? -1 : val); } #endif /* NO_FUEWORD */ Index: head/sys/kern/sysv_shm.c =================================================================== --- head/sys/kern/sysv_shm.c (revision 284214) +++ head/sys/kern/sysv_shm.c (revision 284215) @@ -1,1370 +1,1370 @@ /* $NetBSD: sysv_shm.c,v 1.23 1994/07/04 23:25:12 glass Exp $ */ /*- * Copyright (c) 1994 Adam Glass and Charles Hannum. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by Adam Glass and Charles * Hannum. * 4. The names of the authors may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``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 AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ /*- * Copyright (c) 2003-2005 McAfee, Inc. * All rights reserved. * * This software was developed for the FreeBSD Project in part by McAfee * Research, the Security Research Division of McAfee, Inc under DARPA/SPAWAR * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS research * program. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #include "opt_sysvipc.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 FEATURE(sysv_shm, "System V shared memory segments support"); static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments"); static int shmget_allocate_segment(struct thread *td, struct shmget_args *uap, int mode); static int shmget_existing(struct thread *td, struct shmget_args *uap, int mode, int segnum); #define SHMSEG_FREE 0x0200 #define SHMSEG_REMOVED 0x0400 #define SHMSEG_ALLOCATED 0x0800 static int shm_last_free, shm_nused, shmalloced; vm_size_t shm_committed; static struct shmid_kernel *shmsegs; struct shmmap_state { vm_offset_t va; int shmid; }; static void shm_deallocate_segment(struct shmid_kernel *); static int shm_find_segment_by_key(key_t); static struct shmid_kernel *shm_find_segment(int, bool); static int shm_delete_mapping(struct vmspace *vm, struct shmmap_state *); static void shmrealloc(void); static int shminit(void); static int sysvshm_modload(struct module *, int, void *); static int shmunload(void); static void shmexit_myhook(struct vmspace *vm); static void shmfork_myhook(struct proc *p1, struct proc *p2); static int sysctl_shmsegs(SYSCTL_HANDLER_ARGS); /* * Tuneable values. */ #ifndef SHMMAXPGS #define SHMMAXPGS 131072 /* Note: sysv shared memory is swap backed. */ #endif #ifndef SHMMAX #define SHMMAX (SHMMAXPGS*PAGE_SIZE) #endif #ifndef SHMMIN #define SHMMIN 1 #endif #ifndef SHMMNI #define SHMMNI 192 #endif #ifndef SHMSEG #define SHMSEG 128 #endif #ifndef SHMALL #define SHMALL (SHMMAXPGS) #endif struct shminfo shminfo = { .shmmax = SHMMAX, .shmmin = SHMMIN, .shmmni = SHMMNI, .shmseg = SHMSEG, .shmall = SHMALL }; static int shm_use_phys; static int shm_allow_removed; SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RWTUN, &shminfo.shmmax, 0, "Maximum shared memory segment size"); SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RWTUN, &shminfo.shmmin, 0, "Minimum shared memory segment size"); SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RDTUN, &shminfo.shmmni, 0, "Number of shared memory identifiers"); SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RDTUN, &shminfo.shmseg, 0, "Number of segments per process"); SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RWTUN, &shminfo.shmall, 0, "Maximum number of pages available for shared memory"); SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RWTUN, &shm_use_phys, 0, "Enable/Disable locking of shared memory pages in core"); SYSCTL_INT(_kern_ipc, OID_AUTO, shm_allow_removed, CTLFLAG_RWTUN, &shm_allow_removed, 0, "Enable/Disable attachment to attached segments marked for removal"); SYSCTL_PROC(_kern_ipc, OID_AUTO, shmsegs, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_shmsegs, "", "Current number of shared memory segments allocated"); static struct sx sysvshmsx; #define SYSVSHM_LOCK() sx_xlock(&sysvshmsx) #define SYSVSHM_UNLOCK() sx_xunlock(&sysvshmsx) #define SYSVSHM_ASSERT_LOCKED() sx_assert(&sysvshmsx, SA_XLOCKED) static int shm_find_segment_by_key(key_t key) { int i; for (i = 0; i < shmalloced; i++) if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) && shmsegs[i].u.shm_perm.key == key) return (i); return (-1); } /* * Finds segment either by shmid if is_shmid is true, or by segnum if * is_shmid is false. */ static struct shmid_kernel * shm_find_segment(int arg, bool is_shmid) { struct shmid_kernel *shmseg; int segnum; segnum = is_shmid ? IPCID_TO_IX(arg) : arg; if (segnum < 0 || segnum >= shmalloced) return (NULL); shmseg = &shmsegs[segnum]; if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 || (!shm_allow_removed && (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) != 0) || (is_shmid && shmseg->u.shm_perm.seq != IPCID_TO_SEQ(arg))) return (NULL); return (shmseg); } static void shm_deallocate_segment(struct shmid_kernel *shmseg) { vm_size_t size; SYSVSHM_ASSERT_LOCKED(); vm_object_deallocate(shmseg->object); shmseg->object = NULL; size = round_page(shmseg->u.shm_segsz); shm_committed -= btoc(size); shm_nused--; shmseg->u.shm_perm.mode = SHMSEG_FREE; #ifdef MAC mac_sysvshm_cleanup(shmseg); #endif racct_sub_cred(shmseg->cred, RACCT_NSHM, 1); racct_sub_cred(shmseg->cred, RACCT_SHMSIZE, size); crfree(shmseg->cred); shmseg->cred = NULL; } static int shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s) { struct shmid_kernel *shmseg; int segnum, result; vm_size_t size; SYSVSHM_ASSERT_LOCKED(); segnum = IPCID_TO_IX(shmmap_s->shmid); KASSERT(segnum >= 0 && segnum < shmalloced, ("segnum %d shmalloced %d", segnum, shmalloced)); shmseg = &shmsegs[segnum]; size = round_page(shmseg->u.shm_segsz); result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size); if (result != KERN_SUCCESS) return (EINVAL); shmmap_s->shmid = -1; shmseg->u.shm_dtime = time_second; if ((--shmseg->u.shm_nattch <= 0) && (shmseg->u.shm_perm.mode & SHMSEG_REMOVED)) { shm_deallocate_segment(shmseg); shm_last_free = segnum; } return (0); } static int kern_shmdt_locked(struct thread *td, const void *shmaddr) { struct proc *p = td->td_proc; struct shmmap_state *shmmap_s; #ifdef MAC struct shmid_kernel *shmsegptr; #endif int error, i; SYSVSHM_ASSERT_LOCKED(); if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC)) return (ENOSYS); shmmap_s = p->p_vmspace->vm_shm; if (shmmap_s == NULL) return (EINVAL); for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) { if (shmmap_s->shmid != -1 && shmmap_s->va == (vm_offset_t)shmaddr) { break; } } if (i == shminfo.shmseg) return (EINVAL); #ifdef MAC shmsegptr = &shmsegs[IPCID_TO_IX(shmmap_s->shmid)]; error = mac_sysvshm_check_shmdt(td->td_ucred, shmsegptr); if (error != 0) return (error); #endif error = shm_delete_mapping(p->p_vmspace, shmmap_s); return (error); } #ifndef _SYS_SYSPROTO_H_ struct shmdt_args { const void *shmaddr; }; #endif int sys_shmdt(struct thread *td, struct shmdt_args *uap) { int error; SYSVSHM_LOCK(); error = kern_shmdt_locked(td, uap->shmaddr); SYSVSHM_UNLOCK(); return (error); } static int kern_shmat_locked(struct thread *td, int shmid, const void *shmaddr, int shmflg) { struct proc *p = td->td_proc; struct shmid_kernel *shmseg; struct shmmap_state *shmmap_s; vm_offset_t attach_va; vm_prot_t prot; vm_size_t size; int error, i, rv; SYSVSHM_ASSERT_LOCKED(); if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC)) return (ENOSYS); shmmap_s = p->p_vmspace->vm_shm; if (shmmap_s == NULL) { shmmap_s = malloc(shminfo.shmseg * sizeof(struct shmmap_state), M_SHM, M_WAITOK); for (i = 0; i < shminfo.shmseg; i++) shmmap_s[i].shmid = -1; KASSERT(p->p_vmspace->vm_shm == NULL, ("raced")); p->p_vmspace->vm_shm = shmmap_s; } shmseg = shm_find_segment(shmid, true); if (shmseg == NULL) return (EINVAL); error = ipcperm(td, &shmseg->u.shm_perm, (shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W); if (error != 0) return (error); #ifdef MAC error = mac_sysvshm_check_shmat(td->td_ucred, shmseg, shmflg); if (error != 0) return (error); #endif for (i = 0; i < shminfo.shmseg; i++) { if (shmmap_s->shmid == -1) break; shmmap_s++; } if (i >= shminfo.shmseg) return (EMFILE); size = round_page(shmseg->u.shm_segsz); prot = VM_PROT_READ; if ((shmflg & SHM_RDONLY) == 0) prot |= VM_PROT_WRITE; if (shmaddr != NULL) { if ((shmflg & SHM_RND) != 0) attach_va = (vm_offset_t)shmaddr & ~(SHMLBA-1); else if (((vm_offset_t)shmaddr & (SHMLBA-1)) == 0) attach_va = (vm_offset_t)shmaddr; else return (EINVAL); } else { /* * This is just a hint to vm_map_find() about where to * put it. */ PROC_LOCK(p); attach_va = round_page((vm_offset_t)p->p_vmspace->vm_daddr + - lim_max(p, RLIMIT_DATA)); + lim_max_proc(p, RLIMIT_DATA)); PROC_UNLOCK(p); } vm_object_reference(shmseg->object); rv = vm_map_find(&p->p_vmspace->vm_map, shmseg->object, 0, &attach_va, size, 0, shmaddr != NULL ? VMFS_NO_SPACE : VMFS_OPTIMAL_SPACE, prot, prot, MAP_INHERIT_SHARE | MAP_PREFAULT_PARTIAL); if (rv != KERN_SUCCESS) { vm_object_deallocate(shmseg->object); return (ENOMEM); } shmmap_s->va = attach_va; shmmap_s->shmid = shmid; shmseg->u.shm_lpid = p->p_pid; shmseg->u.shm_atime = time_second; shmseg->u.shm_nattch++; td->td_retval[0] = attach_va; return (error); } int kern_shmat(struct thread *td, int shmid, const void *shmaddr, int shmflg) { int error; SYSVSHM_LOCK(); error = kern_shmat_locked(td, shmid, shmaddr, shmflg); SYSVSHM_UNLOCK(); return (error); } #ifndef _SYS_SYSPROTO_H_ struct shmat_args { int shmid; const void *shmaddr; int shmflg; }; #endif int sys_shmat(struct thread *td, struct shmat_args *uap) { return (kern_shmat(td, uap->shmid, uap->shmaddr, uap->shmflg)); } static int kern_shmctl_locked(struct thread *td, int shmid, int cmd, void *buf, size_t *bufsz) { struct shmid_kernel *shmseg; struct shmid_ds *shmidp; struct shm_info shm_info; int error; SYSVSHM_ASSERT_LOCKED(); if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC)) return (ENOSYS); error = 0; switch (cmd) { /* * It is possible that kern_shmctl is being called from the Linux ABI * layer, in which case, we will need to implement IPC_INFO. It should * be noted that other shmctl calls will be funneled through here for * Linix binaries as well. * * NB: The Linux ABI layer will convert this data to structure(s) more * consistent with the Linux ABI. */ case IPC_INFO: memcpy(buf, &shminfo, sizeof(shminfo)); if (bufsz) *bufsz = sizeof(shminfo); td->td_retval[0] = shmalloced; return (0); case SHM_INFO: { shm_info.used_ids = shm_nused; shm_info.shm_rss = 0; /*XXX where to get from ? */ shm_info.shm_tot = 0; /*XXX where to get from ? */ shm_info.shm_swp = 0; /*XXX where to get from ? */ shm_info.swap_attempts = 0; /*XXX where to get from ? */ shm_info.swap_successes = 0; /*XXX where to get from ? */ memcpy(buf, &shm_info, sizeof(shm_info)); if (bufsz != NULL) *bufsz = sizeof(shm_info); td->td_retval[0] = shmalloced; return (0); } } shmseg = shm_find_segment(shmid, cmd != SHM_STAT); if (shmseg == NULL) return (EINVAL); #ifdef MAC error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, cmd); if (error != 0) return (error); #endif switch (cmd) { case SHM_STAT: case IPC_STAT: error = ipcperm(td, &shmseg->u.shm_perm, IPC_R); if (error != 0) return (error); memcpy(buf, &shmseg->u, sizeof(struct shmid_ds)); if (bufsz != NULL) *bufsz = sizeof(struct shmid_ds); if (cmd == SHM_STAT) { td->td_retval[0] = IXSEQ_TO_IPCID(shmid, shmseg->u.shm_perm); } break; case IPC_SET: shmidp = (struct shmid_ds *)buf; error = ipcperm(td, &shmseg->u.shm_perm, IPC_M); if (error != 0) return (error); shmseg->u.shm_perm.uid = shmidp->shm_perm.uid; shmseg->u.shm_perm.gid = shmidp->shm_perm.gid; shmseg->u.shm_perm.mode = (shmseg->u.shm_perm.mode & ~ACCESSPERMS) | (shmidp->shm_perm.mode & ACCESSPERMS); shmseg->u.shm_ctime = time_second; break; case IPC_RMID: error = ipcperm(td, &shmseg->u.shm_perm, IPC_M); if (error != 0) return (error); shmseg->u.shm_perm.key = IPC_PRIVATE; shmseg->u.shm_perm.mode |= SHMSEG_REMOVED; if (shmseg->u.shm_nattch <= 0) { shm_deallocate_segment(shmseg); shm_last_free = IPCID_TO_IX(shmid); } break; #if 0 case SHM_LOCK: case SHM_UNLOCK: #endif default: error = EINVAL; break; } return (error); } int kern_shmctl(struct thread *td, int shmid, int cmd, void *buf, size_t *bufsz) { int error; SYSVSHM_LOCK(); error = kern_shmctl_locked(td, shmid, cmd, buf, bufsz); SYSVSHM_UNLOCK(); return (error); } #ifndef _SYS_SYSPROTO_H_ struct shmctl_args { int shmid; int cmd; struct shmid_ds *buf; }; #endif int sys_shmctl(struct thread *td, struct shmctl_args *uap) { int error = 0; struct shmid_ds buf; size_t bufsz; /* * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support * Linux binaries. If we see the call come through the FreeBSD ABI, * return an error back to the user since we do not to support this. */ if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO || uap->cmd == SHM_STAT) return (EINVAL); /* IPC_SET needs to copyin the buffer before calling kern_shmctl */ if (uap->cmd == IPC_SET) { if ((error = copyin(uap->buf, &buf, sizeof(struct shmid_ds)))) goto done; } error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz); if (error) goto done; /* Cases in which we need to copyout */ switch (uap->cmd) { case IPC_STAT: error = copyout(&buf, uap->buf, bufsz); break; } done: if (error) { /* Invalidate the return value */ td->td_retval[0] = -1; } return (error); } static int shmget_existing(struct thread *td, struct shmget_args *uap, int mode, int segnum) { struct shmid_kernel *shmseg; #ifdef MAC int error; #endif SYSVSHM_ASSERT_LOCKED(); KASSERT(segnum >= 0 && segnum < shmalloced, ("segnum %d shmalloced %d", segnum, shmalloced)); shmseg = &shmsegs[segnum]; if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL)) return (EEXIST); #ifdef MAC error = mac_sysvshm_check_shmget(td->td_ucred, shmseg, uap->shmflg); if (error != 0) return (error); #endif if (uap->size != 0 && uap->size > shmseg->u.shm_segsz) return (EINVAL); td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm); return (0); } static int shmget_allocate_segment(struct thread *td, struct shmget_args *uap, int mode) { struct ucred *cred = td->td_ucred; struct shmid_kernel *shmseg; vm_object_t shm_object; int i, segnum; size_t size; SYSVSHM_ASSERT_LOCKED(); if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax) return (EINVAL); if (shm_nused >= shminfo.shmmni) /* Any shmids left? */ return (ENOSPC); size = round_page(uap->size); if (shm_committed + btoc(size) > shminfo.shmall) return (ENOMEM); if (shm_last_free < 0) { shmrealloc(); /* Maybe expand the shmsegs[] array. */ for (i = 0; i < shmalloced; i++) if (shmsegs[i].u.shm_perm.mode & SHMSEG_FREE) break; if (i == shmalloced) return (ENOSPC); segnum = i; } else { segnum = shm_last_free; shm_last_free = -1; } KASSERT(segnum >= 0 && segnum < shmalloced, ("segnum %d shmalloced %d", segnum, shmalloced)); shmseg = &shmsegs[segnum]; #ifdef RACCT if (racct_enable) { PROC_LOCK(td->td_proc); if (racct_add(td->td_proc, RACCT_NSHM, 1)) { PROC_UNLOCK(td->td_proc); return (ENOSPC); } if (racct_add(td->td_proc, RACCT_SHMSIZE, size)) { racct_sub(td->td_proc, RACCT_NSHM, 1); PROC_UNLOCK(td->td_proc); return (ENOMEM); } PROC_UNLOCK(td->td_proc); } #endif /* * We make sure that we have allocated a pager before we need * to. */ shm_object = vm_pager_allocate(shm_use_phys ? OBJT_PHYS : OBJT_SWAP, 0, size, VM_PROT_DEFAULT, 0, cred); if (shm_object == NULL) { #ifdef RACCT if (racct_enable) { PROC_LOCK(td->td_proc); racct_sub(td->td_proc, RACCT_NSHM, 1); racct_sub(td->td_proc, RACCT_SHMSIZE, size); PROC_UNLOCK(td->td_proc); } #endif return (ENOMEM); } shm_object->pg_color = 0; VM_OBJECT_WLOCK(shm_object); vm_object_clear_flag(shm_object, OBJ_ONEMAPPING); vm_object_set_flag(shm_object, OBJ_COLORED | OBJ_NOSPLIT); VM_OBJECT_WUNLOCK(shm_object); shmseg->object = shm_object; shmseg->u.shm_perm.cuid = shmseg->u.shm_perm.uid = cred->cr_uid; shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = cred->cr_gid; shmseg->u.shm_perm.mode = (mode & ACCESSPERMS) | SHMSEG_ALLOCATED; shmseg->u.shm_perm.key = uap->key; shmseg->u.shm_perm.seq = (shmseg->u.shm_perm.seq + 1) & 0x7fff; shmseg->cred = crhold(cred); shmseg->u.shm_segsz = uap->size; shmseg->u.shm_cpid = td->td_proc->p_pid; shmseg->u.shm_lpid = shmseg->u.shm_nattch = 0; shmseg->u.shm_atime = shmseg->u.shm_dtime = 0; #ifdef MAC mac_sysvshm_create(cred, shmseg); #endif shmseg->u.shm_ctime = time_second; shm_committed += btoc(size); shm_nused++; td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm); return (0); } #ifndef _SYS_SYSPROTO_H_ struct shmget_args { key_t key; size_t size; int shmflg; }; #endif int sys_shmget(struct thread *td, struct shmget_args *uap) { int segnum, mode; int error; if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC)) return (ENOSYS); mode = uap->shmflg & ACCESSPERMS; SYSVSHM_LOCK(); if (uap->key == IPC_PRIVATE) { error = shmget_allocate_segment(td, uap, mode); } else { segnum = shm_find_segment_by_key(uap->key); if (segnum >= 0) error = shmget_existing(td, uap, mode, segnum); else if ((uap->shmflg & IPC_CREAT) == 0) error = ENOENT; else error = shmget_allocate_segment(td, uap, mode); } SYSVSHM_UNLOCK(); return (error); } static void shmfork_myhook(struct proc *p1, struct proc *p2) { struct shmmap_state *shmmap_s; size_t size; int i; SYSVSHM_LOCK(); size = shminfo.shmseg * sizeof(struct shmmap_state); shmmap_s = malloc(size, M_SHM, M_WAITOK); bcopy(p1->p_vmspace->vm_shm, shmmap_s, size); p2->p_vmspace->vm_shm = shmmap_s; for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) { if (shmmap_s->shmid != -1) { KASSERT(IPCID_TO_IX(shmmap_s->shmid) >= 0 && IPCID_TO_IX(shmmap_s->shmid) < shmalloced, ("segnum %d shmalloced %d", IPCID_TO_IX(shmmap_s->shmid), shmalloced)); shmsegs[IPCID_TO_IX(shmmap_s->shmid)].u.shm_nattch++; } } SYSVSHM_UNLOCK(); } static void shmexit_myhook(struct vmspace *vm) { struct shmmap_state *base, *shm; int i; base = vm->vm_shm; if (base != NULL) { vm->vm_shm = NULL; SYSVSHM_LOCK(); for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) { if (shm->shmid != -1) shm_delete_mapping(vm, shm); } SYSVSHM_UNLOCK(); free(base, M_SHM); } } static void shmrealloc(void) { struct shmid_kernel *newsegs; int i; SYSVSHM_ASSERT_LOCKED(); if (shmalloced >= shminfo.shmmni) return; newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK); for (i = 0; i < shmalloced; i++) bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0])); for (; i < shminfo.shmmni; i++) { shmsegs[i].u.shm_perm.mode = SHMSEG_FREE; shmsegs[i].u.shm_perm.seq = 0; #ifdef MAC mac_sysvshm_init(&shmsegs[i]); #endif } free(shmsegs, M_SHM); shmsegs = newsegs; shmalloced = shminfo.shmmni; } static struct syscall_helper_data shm_syscalls[] = { SYSCALL_INIT_HELPER(shmat), SYSCALL_INIT_HELPER(shmctl), SYSCALL_INIT_HELPER(shmdt), SYSCALL_INIT_HELPER(shmget), #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) SYSCALL_INIT_HELPER_COMPAT(freebsd7_shmctl), #endif #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43)) SYSCALL_INIT_HELPER(shmsys), #endif SYSCALL_INIT_LAST }; #ifdef COMPAT_FREEBSD32 #include #include #include #include #include #include static struct syscall_helper_data shm32_syscalls[] = { SYSCALL32_INIT_HELPER_COMPAT(shmat), SYSCALL32_INIT_HELPER_COMPAT(shmdt), SYSCALL32_INIT_HELPER_COMPAT(shmget), SYSCALL32_INIT_HELPER(freebsd32_shmsys), SYSCALL32_INIT_HELPER(freebsd32_shmctl), #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) SYSCALL32_INIT_HELPER(freebsd7_freebsd32_shmctl), #endif SYSCALL_INIT_LAST }; #endif static int shminit(void) { int i, error; #ifndef BURN_BRIDGES if (TUNABLE_ULONG_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall) != 0) printf("kern.ipc.shmmaxpgs is now called kern.ipc.shmall!\n"); #endif if (shminfo.shmmax == SHMMAX) { /* Initialize shmmax dealing with possible overflow. */ for (i = PAGE_SIZE; i != 0; i--) { shminfo.shmmax = shminfo.shmall * i; if ((shminfo.shmmax / shminfo.shmall) == (u_long)i) break; } } shmalloced = shminfo.shmmni; shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK); for (i = 0; i < shmalloced; i++) { shmsegs[i].u.shm_perm.mode = SHMSEG_FREE; shmsegs[i].u.shm_perm.seq = 0; #ifdef MAC mac_sysvshm_init(&shmsegs[i]); #endif } shm_last_free = 0; shm_nused = 0; shm_committed = 0; sx_init(&sysvshmsx, "sysvshmsx"); shmexit_hook = &shmexit_myhook; shmfork_hook = &shmfork_myhook; error = syscall_helper_register(shm_syscalls, SY_THR_STATIC_KLD); if (error != 0) return (error); #ifdef COMPAT_FREEBSD32 error = syscall32_helper_register(shm32_syscalls, SY_THR_STATIC_KLD); if (error != 0) return (error); #endif return (0); } static int shmunload(void) { int i; if (shm_nused > 0) return (EBUSY); #ifdef COMPAT_FREEBSD32 syscall32_helper_unregister(shm32_syscalls); #endif syscall_helper_unregister(shm_syscalls); for (i = 0; i < shmalloced; i++) { #ifdef MAC mac_sysvshm_destroy(&shmsegs[i]); #endif /* * Objects might be still mapped into the processes * address spaces. Actual free would happen on the * last mapping destruction. */ if (shmsegs[i].u.shm_perm.mode != SHMSEG_FREE) vm_object_deallocate(shmsegs[i].object); } free(shmsegs, M_SHM); shmexit_hook = NULL; shmfork_hook = NULL; sx_destroy(&sysvshmsx); return (0); } static int sysctl_shmsegs(SYSCTL_HANDLER_ARGS) { int error; SYSVSHM_LOCK(); error = SYSCTL_OUT(req, shmsegs, shmalloced * sizeof(shmsegs[0])); SYSVSHM_UNLOCK(); return (error); } #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43)) struct oshmid_ds { struct ipc_perm_old shm_perm; /* operation perms */ int shm_segsz; /* size of segment (bytes) */ u_short shm_cpid; /* pid, creator */ u_short shm_lpid; /* pid, last operation */ short shm_nattch; /* no. of current attaches */ time_t shm_atime; /* last attach time */ time_t shm_dtime; /* last detach time */ time_t shm_ctime; /* last change time */ void *shm_handle; /* internal handle for shm segment */ }; struct oshmctl_args { int shmid; int cmd; struct oshmid_ds *ubuf; }; static int oshmctl(struct thread *td, struct oshmctl_args *uap) { #ifdef COMPAT_43 int error = 0; struct shmid_kernel *shmseg; struct oshmid_ds outbuf; if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC)) return (ENOSYS); if (uap->cmd != IPC_STAT) { return (freebsd7_shmctl(td, (struct freebsd7_shmctl_args *)uap)); } SYSVSHM_LOCK(); shmseg = shm_find_segment(uap->shmid, true); if (shmseg == NULL) { SYSVSHM_UNLOCK(); return (EINVAL); } error = ipcperm(td, &shmseg->u.shm_perm, IPC_R); if (error != 0) { SYSVSHM_UNLOCK(); return (error); } #ifdef MAC error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, uap->cmd); if (error != 0) { SYSVSHM_UNLOCK(); return (error); } #endif ipcperm_new2old(&shmseg->u.shm_perm, &outbuf.shm_perm); outbuf.shm_segsz = shmseg->u.shm_segsz; outbuf.shm_cpid = shmseg->u.shm_cpid; outbuf.shm_lpid = shmseg->u.shm_lpid; outbuf.shm_nattch = shmseg->u.shm_nattch; outbuf.shm_atime = shmseg->u.shm_atime; outbuf.shm_dtime = shmseg->u.shm_dtime; outbuf.shm_ctime = shmseg->u.shm_ctime; outbuf.shm_handle = shmseg->object; SYSVSHM_UNLOCK(); error = copyout(&outbuf, uap->ubuf, sizeof(outbuf)); return (error); #else return (EINVAL); #endif } /* XXX casting to (sy_call_t *) is bogus, as usual. */ static sy_call_t *shmcalls[] = { (sy_call_t *)sys_shmat, (sy_call_t *)oshmctl, (sy_call_t *)sys_shmdt, (sy_call_t *)sys_shmget, (sy_call_t *)freebsd7_shmctl }; #ifndef _SYS_SYSPROTO_H_ /* XXX actually varargs. */ struct shmsys_args { int which; int a2; int a3; int a4; }; #endif int sys_shmsys(struct thread *td, struct shmsys_args *uap) { int error; if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC)) return (ENOSYS); if (uap->which < 0 || uap->which >= nitems(shmcalls)) return (EINVAL); error = (*shmcalls[uap->which])(td, &uap->a2); return (error); } #endif /* i386 && (COMPAT_FREEBSD4 || COMPAT_43) */ #ifdef COMPAT_FREEBSD32 int freebsd32_shmsys(struct thread *td, struct freebsd32_shmsys_args *uap) { #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) switch (uap->which) { case 0: { /* shmat */ struct shmat_args ap; ap.shmid = uap->a2; ap.shmaddr = PTRIN(uap->a3); ap.shmflg = uap->a4; return (sysent[SYS_shmat].sy_call(td, &ap)); } case 2: { /* shmdt */ struct shmdt_args ap; ap.shmaddr = PTRIN(uap->a2); return (sysent[SYS_shmdt].sy_call(td, &ap)); } case 3: { /* shmget */ struct shmget_args ap; ap.key = uap->a2; ap.size = uap->a3; ap.shmflg = uap->a4; return (sysent[SYS_shmget].sy_call(td, &ap)); } case 4: { /* shmctl */ struct freebsd7_freebsd32_shmctl_args ap; ap.shmid = uap->a2; ap.cmd = uap->a3; ap.buf = PTRIN(uap->a4); return (freebsd7_freebsd32_shmctl(td, &ap)); } case 1: /* oshmctl */ default: return (EINVAL); } #else return (nosys(td, NULL)); #endif } #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) int freebsd7_freebsd32_shmctl(struct thread *td, struct freebsd7_freebsd32_shmctl_args *uap) { int error = 0; union { struct shmid_ds shmid_ds; struct shm_info shm_info; struct shminfo shminfo; } u; union { struct shmid_ds32_old shmid_ds32; struct shm_info32 shm_info32; struct shminfo32 shminfo32; } u32; size_t sz; if (uap->cmd == IPC_SET) { if ((error = copyin(uap->buf, &u32.shmid_ds32, sizeof(u32.shmid_ds32)))) goto done; freebsd32_ipcperm_old_in(&u32.shmid_ds32.shm_perm, &u.shmid_ds.shm_perm); CP(u32.shmid_ds32, u.shmid_ds, shm_segsz); CP(u32.shmid_ds32, u.shmid_ds, shm_lpid); CP(u32.shmid_ds32, u.shmid_ds, shm_cpid); CP(u32.shmid_ds32, u.shmid_ds, shm_nattch); CP(u32.shmid_ds32, u.shmid_ds, shm_atime); CP(u32.shmid_ds32, u.shmid_ds, shm_dtime); CP(u32.shmid_ds32, u.shmid_ds, shm_ctime); } error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz); if (error) goto done; /* Cases in which we need to copyout */ switch (uap->cmd) { case IPC_INFO: CP(u.shminfo, u32.shminfo32, shmmax); CP(u.shminfo, u32.shminfo32, shmmin); CP(u.shminfo, u32.shminfo32, shmmni); CP(u.shminfo, u32.shminfo32, shmseg); CP(u.shminfo, u32.shminfo32, shmall); error = copyout(&u32.shminfo32, uap->buf, sizeof(u32.shminfo32)); break; case SHM_INFO: CP(u.shm_info, u32.shm_info32, used_ids); CP(u.shm_info, u32.shm_info32, shm_rss); CP(u.shm_info, u32.shm_info32, shm_tot); CP(u.shm_info, u32.shm_info32, shm_swp); CP(u.shm_info, u32.shm_info32, swap_attempts); CP(u.shm_info, u32.shm_info32, swap_successes); error = copyout(&u32.shm_info32, uap->buf, sizeof(u32.shm_info32)); break; case SHM_STAT: case IPC_STAT: freebsd32_ipcperm_old_out(&u.shmid_ds.shm_perm, &u32.shmid_ds32.shm_perm); if (u.shmid_ds.shm_segsz > INT32_MAX) u32.shmid_ds32.shm_segsz = INT32_MAX; else CP(u.shmid_ds, u32.shmid_ds32, shm_segsz); CP(u.shmid_ds, u32.shmid_ds32, shm_lpid); CP(u.shmid_ds, u32.shmid_ds32, shm_cpid); CP(u.shmid_ds, u32.shmid_ds32, shm_nattch); CP(u.shmid_ds, u32.shmid_ds32, shm_atime); CP(u.shmid_ds, u32.shmid_ds32, shm_dtime); CP(u.shmid_ds, u32.shmid_ds32, shm_ctime); u32.shmid_ds32.shm_internal = 0; error = copyout(&u32.shmid_ds32, uap->buf, sizeof(u32.shmid_ds32)); break; } done: if (error) { /* Invalidate the return value */ td->td_retval[0] = -1; } return (error); } #endif int freebsd32_shmctl(struct thread *td, struct freebsd32_shmctl_args *uap) { int error = 0; union { struct shmid_ds shmid_ds; struct shm_info shm_info; struct shminfo shminfo; } u; union { struct shmid_ds32 shmid_ds32; struct shm_info32 shm_info32; struct shminfo32 shminfo32; } u32; size_t sz; if (uap->cmd == IPC_SET) { if ((error = copyin(uap->buf, &u32.shmid_ds32, sizeof(u32.shmid_ds32)))) goto done; freebsd32_ipcperm_in(&u32.shmid_ds32.shm_perm, &u.shmid_ds.shm_perm); CP(u32.shmid_ds32, u.shmid_ds, shm_segsz); CP(u32.shmid_ds32, u.shmid_ds, shm_lpid); CP(u32.shmid_ds32, u.shmid_ds, shm_cpid); CP(u32.shmid_ds32, u.shmid_ds, shm_nattch); CP(u32.shmid_ds32, u.shmid_ds, shm_atime); CP(u32.shmid_ds32, u.shmid_ds, shm_dtime); CP(u32.shmid_ds32, u.shmid_ds, shm_ctime); } error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz); if (error) goto done; /* Cases in which we need to copyout */ switch (uap->cmd) { case IPC_INFO: CP(u.shminfo, u32.shminfo32, shmmax); CP(u.shminfo, u32.shminfo32, shmmin); CP(u.shminfo, u32.shminfo32, shmmni); CP(u.shminfo, u32.shminfo32, shmseg); CP(u.shminfo, u32.shminfo32, shmall); error = copyout(&u32.shminfo32, uap->buf, sizeof(u32.shminfo32)); break; case SHM_INFO: CP(u.shm_info, u32.shm_info32, used_ids); CP(u.shm_info, u32.shm_info32, shm_rss); CP(u.shm_info, u32.shm_info32, shm_tot); CP(u.shm_info, u32.shm_info32, shm_swp); CP(u.shm_info, u32.shm_info32, swap_attempts); CP(u.shm_info, u32.shm_info32, swap_successes); error = copyout(&u32.shm_info32, uap->buf, sizeof(u32.shm_info32)); break; case SHM_STAT: case IPC_STAT: freebsd32_ipcperm_out(&u.shmid_ds.shm_perm, &u32.shmid_ds32.shm_perm); if (u.shmid_ds.shm_segsz > INT32_MAX) u32.shmid_ds32.shm_segsz = INT32_MAX; else CP(u.shmid_ds, u32.shmid_ds32, shm_segsz); CP(u.shmid_ds, u32.shmid_ds32, shm_lpid); CP(u.shmid_ds, u32.shmid_ds32, shm_cpid); CP(u.shmid_ds, u32.shmid_ds32, shm_nattch); CP(u.shmid_ds, u32.shmid_ds32, shm_atime); CP(u.shmid_ds, u32.shmid_ds32, shm_dtime); CP(u.shmid_ds, u32.shmid_ds32, shm_ctime); error = copyout(&u32.shmid_ds32, uap->buf, sizeof(u32.shmid_ds32)); break; } done: if (error) { /* Invalidate the return value */ td->td_retval[0] = -1; } return (error); } #endif #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) #ifndef CP #define CP(src, dst, fld) do { (dst).fld = (src).fld; } while (0) #endif #ifndef _SYS_SYSPROTO_H_ struct freebsd7_shmctl_args { int shmid; int cmd; struct shmid_ds_old *buf; }; #endif int freebsd7_shmctl(struct thread *td, struct freebsd7_shmctl_args *uap) { int error = 0; struct shmid_ds_old old; struct shmid_ds buf; size_t bufsz; /* * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support * Linux binaries. If we see the call come through the FreeBSD ABI, * return an error back to the user since we do not to support this. */ if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO || uap->cmd == SHM_STAT) return (EINVAL); /* IPC_SET needs to copyin the buffer before calling kern_shmctl */ if (uap->cmd == IPC_SET) { if ((error = copyin(uap->buf, &old, sizeof(old)))) goto done; ipcperm_old2new(&old.shm_perm, &buf.shm_perm); CP(old, buf, shm_segsz); CP(old, buf, shm_lpid); CP(old, buf, shm_cpid); CP(old, buf, shm_nattch); CP(old, buf, shm_atime); CP(old, buf, shm_dtime); CP(old, buf, shm_ctime); } error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz); if (error) goto done; /* Cases in which we need to copyout */ switch (uap->cmd) { case IPC_STAT: ipcperm_new2old(&buf.shm_perm, &old.shm_perm); if (buf.shm_segsz > INT_MAX) old.shm_segsz = INT_MAX; else CP(buf, old, shm_segsz); CP(buf, old, shm_lpid); CP(buf, old, shm_cpid); if (buf.shm_nattch > SHRT_MAX) old.shm_nattch = SHRT_MAX; else CP(buf, old, shm_nattch); CP(buf, old, shm_atime); CP(buf, old, shm_dtime); CP(buf, old, shm_ctime); old.shm_internal = NULL; error = copyout(&old, uap->buf, sizeof(old)); break; } done: if (error) { /* Invalidate the return value */ td->td_retval[0] = -1; } return (error); } #endif /* COMPAT_FREEBSD4 || COMPAT_FREEBSD5 || COMPAT_FREEBSD6 || COMPAT_FREEBSD7 */ static int sysvshm_modload(struct module *module, int cmd, void *arg) { int error = 0; switch (cmd) { case MOD_LOAD: error = shminit(); if (error != 0) shmunload(); break; case MOD_UNLOAD: error = shmunload(); break; case MOD_SHUTDOWN: break; default: error = EINVAL; break; } return (error); } static moduledata_t sysvshm_mod = { "sysvshm", &sysvshm_modload, NULL }; DECLARE_MODULE(sysvshm, sysvshm_mod, SI_SUB_SYSV_SHM, SI_ORDER_FIRST); MODULE_VERSION(sysvshm, 1); Index: head/sys/kern/tty_pts.c =================================================================== --- head/sys/kern/tty_pts.c (revision 284214) +++ head/sys/kern/tty_pts.c (revision 284215) @@ -1,867 +1,867 @@ /*- * Copyright (c) 2008 Ed Schouten * All rights reserved. * * Portions of this software were developed under sponsorship from Snow * B.V., the Netherlands. * * 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. */ #include __FBSDID("$FreeBSD$"); /* Add compatibility bits for FreeBSD. */ #define PTS_COMPAT /* Add pty(4) compat bits. */ #define PTS_EXTERNAL /* Add bits to make Linux binaries work. */ #define PTS_LINUX #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 /* * Our utmp(5) format is limited to 8-byte TTY line names. This means * we can at most allocate 1000 pseudo-terminals ("pts/999"). Allow * users to increase this number, assuming they have manually increased * UT_LINESIZE. */ static struct unrhdr *pts_pool; static MALLOC_DEFINE(M_PTS, "pts", "pseudo tty device"); /* * Per-PTS structure. * * List of locks * (t) locked by tty_lock() * (c) const until freeing */ struct pts_softc { int pts_unit; /* (c) Device unit number. */ unsigned int pts_flags; /* (t) Device flags. */ #define PTS_PKT 0x1 /* Packet mode. */ #define PTS_FINISHED 0x2 /* Return errors on read()/write(). */ char pts_pkt; /* (t) Unread packet mode data. */ struct cv pts_inwait; /* (t) Blocking write() on master. */ struct selinfo pts_inpoll; /* (t) Select queue for write(). */ struct cv pts_outwait; /* (t) Blocking read() on master. */ struct selinfo pts_outpoll; /* (t) Select queue for read(). */ #ifdef PTS_EXTERNAL struct cdev *pts_cdev; /* (c) Master device node. */ #endif /* PTS_EXTERNAL */ struct ucred *pts_cred; /* (c) Resource limit. */ }; /* * Controller-side file operations. */ static int ptsdev_read(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { struct tty *tp = fp->f_data; struct pts_softc *psc = tty_softc(tp); int error = 0; char pkt; if (uio->uio_resid == 0) return (0); tty_lock(tp); for (;;) { /* * Implement packet mode. When packet mode is turned on, * the first byte contains a bitmask of events that * occured (start, stop, flush, window size, etc). */ if (psc->pts_flags & PTS_PKT && psc->pts_pkt) { pkt = psc->pts_pkt; psc->pts_pkt = 0; tty_unlock(tp); error = ureadc(pkt, uio); return (error); } /* * Transmit regular data. * * XXX: We shouldn't use ttydisc_getc_poll()! Even * though in this implementation, there is likely going * to be data, we should just call ttydisc_getc_uio() * and use its return value to sleep. */ if (ttydisc_getc_poll(tp)) { if (psc->pts_flags & PTS_PKT) { /* * XXX: Small race. Fortunately PTY * consumers aren't multithreaded. */ tty_unlock(tp); error = ureadc(TIOCPKT_DATA, uio); if (error) return (error); tty_lock(tp); } error = ttydisc_getc_uio(tp, uio); break; } /* Maybe the device isn't used anyway. */ if (psc->pts_flags & PTS_FINISHED) break; /* Wait for more data. */ if (fp->f_flag & O_NONBLOCK) { error = EWOULDBLOCK; break; } error = cv_wait_sig(&psc->pts_outwait, tp->t_mtx); if (error != 0) break; } tty_unlock(tp); return (error); } static int ptsdev_write(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { struct tty *tp = fp->f_data; struct pts_softc *psc = tty_softc(tp); char ib[256], *ibstart; size_t iblen, rintlen; int error = 0; if (uio->uio_resid == 0) return (0); for (;;) { ibstart = ib; iblen = MIN(uio->uio_resid, sizeof ib); error = uiomove(ib, iblen, uio); tty_lock(tp); if (error != 0) { iblen = 0; goto done; } /* * When possible, avoid the slow path. rint_bypass() * copies all input to the input queue at once. */ MPASS(iblen > 0); do { rintlen = ttydisc_rint_simple(tp, ibstart, iblen); ibstart += rintlen; iblen -= rintlen; if (iblen == 0) { /* All data written. */ break; } /* Maybe the device isn't used anyway. */ if (psc->pts_flags & PTS_FINISHED) { error = EIO; goto done; } /* Wait for more data. */ if (fp->f_flag & O_NONBLOCK) { error = EWOULDBLOCK; goto done; } /* Wake up users on the slave side. */ ttydisc_rint_done(tp); error = cv_wait_sig(&psc->pts_inwait, tp->t_mtx); if (error != 0) goto done; } while (iblen > 0); if (uio->uio_resid == 0) break; tty_unlock(tp); } done: ttydisc_rint_done(tp); tty_unlock(tp); /* * Don't account for the part of the buffer that we couldn't * pass to the TTY. */ uio->uio_resid += iblen; return (error); } static int ptsdev_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *active_cred, struct thread *td) { struct tty *tp = fp->f_data; struct pts_softc *psc = tty_softc(tp); int error = 0, sig; switch (cmd) { case FIODTYPE: *(int *)data = D_TTY; return (0); case FIONBIO: /* This device supports non-blocking operation. */ return (0); case FIONREAD: tty_lock(tp); if (psc->pts_flags & PTS_FINISHED) { /* Force read() to be called. */ *(int *)data = 1; } else { *(int *)data = ttydisc_getc_poll(tp); } tty_unlock(tp); return (0); case FIODGNAME: { struct fiodgname_arg *fgn; const char *p; int i; /* Reverse device name lookups, for ptsname() and ttyname(). */ fgn = data; p = tty_devname(tp); i = strlen(p) + 1; if (i > fgn->len) return (EINVAL); return copyout(p, fgn->buf, i); } /* * We need to implement TIOCGPGRP and TIOCGSID here again. When * called on the pseudo-terminal master, it should not check if * the terminal is the foreground terminal of the calling * process. * * TIOCGETA is also implemented here. Various Linux PTY routines * often call isatty(), which is implemented by tcgetattr(). */ #ifdef PTS_LINUX case TIOCGETA: /* Obtain terminal flags through tcgetattr(). */ tty_lock(tp); *(struct termios*)data = tp->t_termios; tty_unlock(tp); return (0); #endif /* PTS_LINUX */ case TIOCSETAF: case TIOCSETAW: /* * We must make sure we turn tcsetattr() calls of TCSAFLUSH and * TCSADRAIN into something different. If an application would * call TCSAFLUSH or TCSADRAIN on the master descriptor, it may * deadlock waiting for all data to be read. */ cmd = TIOCSETA; break; #if defined(PTS_COMPAT) || defined(PTS_LINUX) case TIOCGPTN: /* * Get the device unit number. */ if (psc->pts_unit < 0) return (ENOTTY); *(unsigned int *)data = psc->pts_unit; return (0); #endif /* PTS_COMPAT || PTS_LINUX */ case TIOCGPGRP: /* Get the foreground process group ID. */ tty_lock(tp); if (tp->t_pgrp != NULL) *(int *)data = tp->t_pgrp->pg_id; else *(int *)data = NO_PID; tty_unlock(tp); return (0); case TIOCGSID: /* Get the session leader process ID. */ tty_lock(tp); if (tp->t_session == NULL) error = ENOTTY; else *(int *)data = tp->t_session->s_sid; tty_unlock(tp); return (error); case TIOCPTMASTER: /* Yes, we are a pseudo-terminal master. */ return (0); case TIOCSIG: /* Signal the foreground process group. */ sig = *(int *)data; if (sig < 1 || sig >= NSIG) return (EINVAL); tty_lock(tp); tty_signal_pgrp(tp, sig); tty_unlock(tp); return (0); case TIOCPKT: /* Enable/disable packet mode. */ tty_lock(tp); if (*(int *)data) psc->pts_flags |= PTS_PKT; else psc->pts_flags &= ~PTS_PKT; tty_unlock(tp); return (0); } /* Just redirect this ioctl to the slave device. */ tty_lock(tp); error = tty_ioctl(tp, cmd, data, fp->f_flag, td); tty_unlock(tp); if (error == ENOIOCTL) error = ENOTTY; return (error); } static int ptsdev_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td) { struct tty *tp = fp->f_data; struct pts_softc *psc = tty_softc(tp); int revents = 0; tty_lock(tp); if (psc->pts_flags & PTS_FINISHED) { /* Slave device is not opened. */ tty_unlock(tp); return ((events & (POLLIN|POLLRDNORM)) | POLLHUP); } if (events & (POLLIN|POLLRDNORM)) { /* See if we can getc something. */ if (ttydisc_getc_poll(tp) || (psc->pts_flags & PTS_PKT && psc->pts_pkt)) revents |= events & (POLLIN|POLLRDNORM); } if (events & (POLLOUT|POLLWRNORM)) { /* See if we can rint something. */ if (ttydisc_rint_poll(tp)) revents |= events & (POLLOUT|POLLWRNORM); } /* * No need to check for POLLHUP here. This device cannot be used * as a callout device, which means we always have a carrier, * because the master is. */ if (revents == 0) { /* * This code might look misleading, but the naming of * poll events on this side is the opposite of the slave * device. */ if (events & (POLLIN|POLLRDNORM)) selrecord(td, &psc->pts_outpoll); if (events & (POLLOUT|POLLWRNORM)) selrecord(td, &psc->pts_inpoll); } tty_unlock(tp); return (revents); } /* * kqueue support. */ static void pts_kqops_read_detach(struct knote *kn) { struct file *fp = kn->kn_fp; struct tty *tp = fp->f_data; struct pts_softc *psc = tty_softc(tp); knlist_remove(&psc->pts_outpoll.si_note, kn, 0); } static int pts_kqops_read_event(struct knote *kn, long hint) { struct file *fp = kn->kn_fp; struct tty *tp = fp->f_data; struct pts_softc *psc = tty_softc(tp); if (psc->pts_flags & PTS_FINISHED) { kn->kn_flags |= EV_EOF; return (1); } else { kn->kn_data = ttydisc_getc_poll(tp); return (kn->kn_data > 0); } } static void pts_kqops_write_detach(struct knote *kn) { struct file *fp = kn->kn_fp; struct tty *tp = fp->f_data; struct pts_softc *psc = tty_softc(tp); knlist_remove(&psc->pts_inpoll.si_note, kn, 0); } static int pts_kqops_write_event(struct knote *kn, long hint) { struct file *fp = kn->kn_fp; struct tty *tp = fp->f_data; struct pts_softc *psc = tty_softc(tp); if (psc->pts_flags & PTS_FINISHED) { kn->kn_flags |= EV_EOF; return (1); } else { kn->kn_data = ttydisc_rint_poll(tp); return (kn->kn_data > 0); } } static struct filterops pts_kqops_read = { .f_isfd = 1, .f_detach = pts_kqops_read_detach, .f_event = pts_kqops_read_event, }; static struct filterops pts_kqops_write = { .f_isfd = 1, .f_detach = pts_kqops_write_detach, .f_event = pts_kqops_write_event, }; static int ptsdev_kqfilter(struct file *fp, struct knote *kn) { struct tty *tp = fp->f_data; struct pts_softc *psc = tty_softc(tp); int error = 0; tty_lock(tp); switch (kn->kn_filter) { case EVFILT_READ: kn->kn_fop = &pts_kqops_read; knlist_add(&psc->pts_outpoll.si_note, kn, 1); break; case EVFILT_WRITE: kn->kn_fop = &pts_kqops_write; knlist_add(&psc->pts_inpoll.si_note, kn, 1); break; default: error = EINVAL; break; } tty_unlock(tp); return (error); } static int ptsdev_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td) { struct tty *tp = fp->f_data; #ifdef PTS_EXTERNAL struct pts_softc *psc = tty_softc(tp); #endif /* PTS_EXTERNAL */ struct cdev *dev = tp->t_dev; /* * According to POSIX, we must implement an fstat(). This also * makes this implementation compatible with Linux binaries, * because Linux calls fstat() on the pseudo-terminal master to * obtain st_rdev. * * XXX: POSIX also mentions we must fill in st_dev, but how? */ bzero(sb, sizeof *sb); #ifdef PTS_EXTERNAL if (psc->pts_cdev != NULL) sb->st_ino = sb->st_rdev = dev2udev(psc->pts_cdev); else #endif /* PTS_EXTERNAL */ sb->st_ino = sb->st_rdev = tty_udev(tp); sb->st_atim = dev->si_atime; sb->st_ctim = dev->si_ctime; sb->st_mtim = dev->si_mtime; sb->st_uid = dev->si_uid; sb->st_gid = dev->si_gid; sb->st_mode = dev->si_mode | S_IFCHR; return (0); } static int ptsdev_close(struct file *fp, struct thread *td) { struct tty *tp = fp->f_data; /* Deallocate TTY device. */ tty_lock(tp); tty_rel_gone(tp); /* * Open of /dev/ptmx or /dev/ptyXX changes the type of file * from DTYPE_VNODE to DTYPE_PTS. vn_open() increases vnode * use count, we need to decrement it, and possibly do other * required cleanup. */ if (fp->f_vnode != NULL) return (vnops.fo_close(fp, td)); return (0); } static int ptsdev_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) { struct tty *tp; kif->kf_type = KF_TYPE_PTS; tp = fp->f_data; kif->kf_un.kf_pts.kf_pts_dev = tty_udev(tp); strlcpy(kif->kf_path, tty_devname(tp), sizeof(kif->kf_path)); return (0); } static struct fileops ptsdev_ops = { .fo_read = ptsdev_read, .fo_write = ptsdev_write, .fo_truncate = invfo_truncate, .fo_ioctl = ptsdev_ioctl, .fo_poll = ptsdev_poll, .fo_kqfilter = ptsdev_kqfilter, .fo_stat = ptsdev_stat, .fo_close = ptsdev_close, .fo_chmod = invfo_chmod, .fo_chown = invfo_chown, .fo_sendfile = invfo_sendfile, .fo_fill_kinfo = ptsdev_fill_kinfo, .fo_flags = DFLAG_PASSABLE, }; /* * Driver-side hooks. */ static void ptsdrv_outwakeup(struct tty *tp) { struct pts_softc *psc = tty_softc(tp); cv_broadcast(&psc->pts_outwait); selwakeup(&psc->pts_outpoll); KNOTE_LOCKED(&psc->pts_outpoll.si_note, 0); } static void ptsdrv_inwakeup(struct tty *tp) { struct pts_softc *psc = tty_softc(tp); cv_broadcast(&psc->pts_inwait); selwakeup(&psc->pts_inpoll); KNOTE_LOCKED(&psc->pts_inpoll.si_note, 0); } static int ptsdrv_open(struct tty *tp) { struct pts_softc *psc = tty_softc(tp); psc->pts_flags &= ~PTS_FINISHED; return (0); } static void ptsdrv_close(struct tty *tp) { struct pts_softc *psc = tty_softc(tp); /* Wake up any blocked readers/writers. */ psc->pts_flags |= PTS_FINISHED; ptsdrv_outwakeup(tp); ptsdrv_inwakeup(tp); } static void ptsdrv_pktnotify(struct tty *tp, char event) { struct pts_softc *psc = tty_softc(tp); /* * Clear conflicting flags. */ switch (event) { case TIOCPKT_STOP: psc->pts_pkt &= ~TIOCPKT_START; break; case TIOCPKT_START: psc->pts_pkt &= ~TIOCPKT_STOP; break; case TIOCPKT_NOSTOP: psc->pts_pkt &= ~TIOCPKT_DOSTOP; break; case TIOCPKT_DOSTOP: psc->pts_pkt &= ~TIOCPKT_NOSTOP; break; } psc->pts_pkt |= event; ptsdrv_outwakeup(tp); } static void ptsdrv_free(void *softc) { struct pts_softc *psc = softc; /* Make device number available again. */ if (psc->pts_unit >= 0) free_unr(pts_pool, psc->pts_unit); chgptscnt(psc->pts_cred->cr_ruidinfo, -1, 0); racct_sub_cred(psc->pts_cred, RACCT_NPTS, 1); crfree(psc->pts_cred); seldrain(&psc->pts_inpoll); seldrain(&psc->pts_outpoll); knlist_destroy(&psc->pts_inpoll.si_note); knlist_destroy(&psc->pts_outpoll.si_note); #ifdef PTS_EXTERNAL /* Destroy master device as well. */ if (psc->pts_cdev != NULL) destroy_dev_sched(psc->pts_cdev); #endif /* PTS_EXTERNAL */ free(psc, M_PTS); } static struct ttydevsw pts_class = { .tsw_flags = TF_NOPREFIX, .tsw_outwakeup = ptsdrv_outwakeup, .tsw_inwakeup = ptsdrv_inwakeup, .tsw_open = ptsdrv_open, .tsw_close = ptsdrv_close, .tsw_pktnotify = ptsdrv_pktnotify, .tsw_free = ptsdrv_free, }; #ifndef PTS_EXTERNAL static #endif /* !PTS_EXTERNAL */ int pts_alloc(int fflags, struct thread *td, struct file *fp) { int unit, ok, error; struct tty *tp; struct pts_softc *psc; struct proc *p = td->td_proc; struct ucred *cred = td->td_ucred; /* Resource limiting. */ PROC_LOCK(p); error = racct_add(p, RACCT_NPTS, 1); if (error != 0) { PROC_UNLOCK(p); return (EAGAIN); } - ok = chgptscnt(cred->cr_ruidinfo, 1, lim_cur(p, RLIMIT_NPTS)); + ok = chgptscnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_NPTS)); if (!ok) { racct_sub(p, RACCT_NPTS, 1); PROC_UNLOCK(p); return (EAGAIN); } PROC_UNLOCK(p); /* Try to allocate a new pts unit number. */ unit = alloc_unr(pts_pool); if (unit < 0) { racct_sub(p, RACCT_NPTS, 1); chgptscnt(cred->cr_ruidinfo, -1, 0); return (EAGAIN); } /* Allocate TTY and softc. */ psc = malloc(sizeof(struct pts_softc), M_PTS, M_WAITOK|M_ZERO); cv_init(&psc->pts_inwait, "ptsin"); cv_init(&psc->pts_outwait, "ptsout"); psc->pts_unit = unit; psc->pts_cred = crhold(cred); tp = tty_alloc(&pts_class, psc); knlist_init_mtx(&psc->pts_inpoll.si_note, tp->t_mtx); knlist_init_mtx(&psc->pts_outpoll.si_note, tp->t_mtx); /* Expose the slave device as well. */ tty_makedev(tp, td->td_ucred, "pts/%u", psc->pts_unit); finit(fp, fflags, DTYPE_PTS, tp, &ptsdev_ops); return (0); } #ifdef PTS_EXTERNAL int pts_alloc_external(int fflags, struct thread *td, struct file *fp, struct cdev *dev, const char *name) { int ok, error; struct tty *tp; struct pts_softc *psc; struct proc *p = td->td_proc; struct ucred *cred = td->td_ucred; /* Resource limiting. */ PROC_LOCK(p); error = racct_add(p, RACCT_NPTS, 1); if (error != 0) { PROC_UNLOCK(p); return (EAGAIN); } - ok = chgptscnt(cred->cr_ruidinfo, 1, lim_cur(p, RLIMIT_NPTS)); + ok = chgptscnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_NPTS)); if (!ok) { racct_sub(p, RACCT_NPTS, 1); PROC_UNLOCK(p); return (EAGAIN); } PROC_UNLOCK(p); /* Allocate TTY and softc. */ psc = malloc(sizeof(struct pts_softc), M_PTS, M_WAITOK|M_ZERO); cv_init(&psc->pts_inwait, "ptsin"); cv_init(&psc->pts_outwait, "ptsout"); psc->pts_unit = -1; psc->pts_cdev = dev; psc->pts_cred = crhold(cred); tp = tty_alloc(&pts_class, psc); knlist_init_mtx(&psc->pts_inpoll.si_note, tp->t_mtx); knlist_init_mtx(&psc->pts_outpoll.si_note, tp->t_mtx); /* Expose the slave device as well. */ tty_makedev(tp, td->td_ucred, "%s", name); finit(fp, fflags, DTYPE_PTS, tp, &ptsdev_ops); return (0); } #endif /* PTS_EXTERNAL */ int sys_posix_openpt(struct thread *td, struct posix_openpt_args *uap) { int error, fd; struct file *fp; /* * POSIX states it's unspecified when other flags are passed. We * don't allow this. */ if (uap->flags & ~(O_RDWR|O_NOCTTY|O_CLOEXEC)) return (EINVAL); error = falloc(td, &fp, &fd, uap->flags); if (error) return (error); /* Allocate the actual pseudo-TTY. */ error = pts_alloc(FFLAGS(uap->flags & O_ACCMODE), td, fp); if (error != 0) { fdclose(td, fp, fd); fdrop(fp, td); return (error); } /* Pass it back to userspace. */ td->td_retval[0] = fd; fdrop(fp, td); return (0); } static void pts_init(void *unused) { pts_pool = new_unrhdr(0, INT_MAX, NULL); } SYSINIT(pts, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, pts_init, NULL); Index: head/sys/kern/uipc_sockbuf.c =================================================================== --- head/sys/kern/uipc_sockbuf.c (revision 284214) +++ head/sys/kern/uipc_sockbuf.c (revision 284215) @@ -1,1316 +1,1314 @@ /*- * Copyright (c) 1982, 1986, 1988, 1990, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)uipc_socket2.c 8.1 (Berkeley) 6/10/93 */ #include __FBSDID("$FreeBSD$"); #include "opt_param.h" #include #include /* for aio_swake proto */ #include #include #include #include #include #include #include #include #include #include #include #include /* * Function pointer set by the AIO routines so that the socket buffer code * can call back into the AIO module if it is loaded. */ void (*aio_swake)(struct socket *, struct sockbuf *); /* * Primitive routines for operating on socket buffers */ u_long sb_max = SB_MAX; u_long sb_max_adj = (quad_t)SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */ static u_long sb_efficiency = 8; /* parameter for sbreserve() */ static struct mbuf *sbcut_internal(struct sockbuf *sb, int len); static void sbflush_internal(struct sockbuf *sb); /* * Mark ready "count" mbufs starting with "m". */ int sbready(struct sockbuf *sb, struct mbuf *m, int count) { u_int blocker; SOCKBUF_LOCK_ASSERT(sb); KASSERT(sb->sb_fnrdy != NULL, ("%s: sb %p NULL fnrdy", __func__, sb)); blocker = (sb->sb_fnrdy == m) ? M_BLOCKED : 0; for (int i = 0; i < count; i++, m = m->m_next) { KASSERT(m->m_flags & M_NOTREADY, ("%s: m %p !M_NOTREADY", __func__, m)); m->m_flags &= ~(M_NOTREADY | blocker); if (blocker) sb->sb_acc += m->m_len; } if (!blocker) return (EINPROGRESS); /* This one was blocking all the queue. */ for (; m && (m->m_flags & M_NOTREADY) == 0; m = m->m_next) { KASSERT(m->m_flags & M_BLOCKED, ("%s: m %p !M_BLOCKED", __func__, m)); m->m_flags &= ~M_BLOCKED; sb->sb_acc += m->m_len; } sb->sb_fnrdy = m; return (0); } /* * Adjust sockbuf state reflecting allocation of m. */ void sballoc(struct sockbuf *sb, struct mbuf *m) { SOCKBUF_LOCK_ASSERT(sb); sb->sb_ccc += m->m_len; if (sb->sb_fnrdy == NULL) { if (m->m_flags & M_NOTREADY) sb->sb_fnrdy = m; else sb->sb_acc += m->m_len; } else m->m_flags |= M_BLOCKED; if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) sb->sb_ctl += m->m_len; sb->sb_mbcnt += MSIZE; sb->sb_mcnt += 1; if (m->m_flags & M_EXT) { sb->sb_mbcnt += m->m_ext.ext_size; sb->sb_ccnt += 1; } } /* * Adjust sockbuf state reflecting freeing of m. */ void sbfree(struct sockbuf *sb, struct mbuf *m) { #if 0 /* XXX: not yet: soclose() call path comes here w/o lock. */ SOCKBUF_LOCK_ASSERT(sb); #endif sb->sb_ccc -= m->m_len; if (!(m->m_flags & M_NOTAVAIL)) sb->sb_acc -= m->m_len; if (m == sb->sb_fnrdy) { struct mbuf *n; KASSERT(m->m_flags & M_NOTREADY, ("%s: m %p !M_NOTREADY", __func__, m)); n = m->m_next; while (n != NULL && !(n->m_flags & M_NOTREADY)) { n->m_flags &= ~M_BLOCKED; sb->sb_acc += n->m_len; n = n->m_next; } sb->sb_fnrdy = n; } if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) sb->sb_ctl -= m->m_len; sb->sb_mbcnt -= MSIZE; sb->sb_mcnt -= 1; if (m->m_flags & M_EXT) { sb->sb_mbcnt -= m->m_ext.ext_size; sb->sb_ccnt -= 1; } if (sb->sb_sndptr == m) { sb->sb_sndptr = NULL; sb->sb_sndptroff = 0; } if (sb->sb_sndptroff != 0) sb->sb_sndptroff -= m->m_len; } /* * Socantsendmore indicates that no more data will be sent on the socket; it * would normally be applied to a socket when the user informs the system * that no more data is to be sent, by the protocol code (in case * PRU_SHUTDOWN). Socantrcvmore indicates that no more data will be * received, and will normally be applied to the socket by a protocol when it * detects that the peer will send no more data. Data queued for reading in * the socket may yet be read. */ void socantsendmore_locked(struct socket *so) { SOCKBUF_LOCK_ASSERT(&so->so_snd); so->so_snd.sb_state |= SBS_CANTSENDMORE; sowwakeup_locked(so); mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED); } void socantsendmore(struct socket *so) { SOCKBUF_LOCK(&so->so_snd); socantsendmore_locked(so); mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED); } void socantrcvmore_locked(struct socket *so) { SOCKBUF_LOCK_ASSERT(&so->so_rcv); so->so_rcv.sb_state |= SBS_CANTRCVMORE; sorwakeup_locked(so); mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED); } void socantrcvmore(struct socket *so) { SOCKBUF_LOCK(&so->so_rcv); socantrcvmore_locked(so); mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED); } /* * Wait for data to arrive at/drain from a socket buffer. */ int sbwait(struct sockbuf *sb) { SOCKBUF_LOCK_ASSERT(sb); sb->sb_flags |= SB_WAIT; return (msleep_sbt(&sb->sb_acc, &sb->sb_mtx, (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait", sb->sb_timeo, 0, 0)); } int sblock(struct sockbuf *sb, int flags) { KASSERT((flags & SBL_VALID) == flags, ("sblock: flags invalid (0x%x)", flags)); if (flags & SBL_WAIT) { if ((sb->sb_flags & SB_NOINTR) || (flags & SBL_NOINTR)) { sx_xlock(&sb->sb_sx); return (0); } return (sx_xlock_sig(&sb->sb_sx)); } else { if (sx_try_xlock(&sb->sb_sx) == 0) return (EWOULDBLOCK); return (0); } } void sbunlock(struct sockbuf *sb) { sx_xunlock(&sb->sb_sx); } /* * Wakeup processes waiting on a socket buffer. Do asynchronous notification * via SIGIO if the socket has the SS_ASYNC flag set. * * Called with the socket buffer lock held; will release the lock by the end * of the function. This allows the caller to acquire the socket buffer lock * while testing for the need for various sorts of wakeup and hold it through * to the point where it's no longer required. We currently hold the lock * through calls out to other subsystems (with the exception of kqueue), and * then release it to avoid lock order issues. It's not clear that's * correct. */ void sowakeup(struct socket *so, struct sockbuf *sb) { int ret; SOCKBUF_LOCK_ASSERT(sb); selwakeuppri(&sb->sb_sel, PSOCK); if (!SEL_WAITING(&sb->sb_sel)) sb->sb_flags &= ~SB_SEL; if (sb->sb_flags & SB_WAIT) { sb->sb_flags &= ~SB_WAIT; wakeup(&sb->sb_acc); } KNOTE_LOCKED(&sb->sb_sel.si_note, 0); if (sb->sb_upcall != NULL) { ret = sb->sb_upcall(so, sb->sb_upcallarg, M_NOWAIT); if (ret == SU_ISCONNECTED) { KASSERT(sb == &so->so_rcv, ("SO_SND upcall returned SU_ISCONNECTED")); soupcall_clear(so, SO_RCV); } } else ret = SU_OK; if (sb->sb_flags & SB_AIO) aio_swake(so, sb); SOCKBUF_UNLOCK(sb); if (ret == SU_ISCONNECTED) soisconnected(so); if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL) pgsigio(&so->so_sigio, SIGIO, 0); mtx_assert(SOCKBUF_MTX(sb), MA_NOTOWNED); } /* * Socket buffer (struct sockbuf) utility routines. * * Each socket contains two socket buffers: one for sending data and one for * receiving data. Each buffer contains a queue of mbufs, information about * the number of mbufs and amount of data in the queue, and other fields * allowing select() statements and notification on data availability to be * implemented. * * Data stored in a socket buffer is maintained as a list of records. Each * record is a list of mbufs chained together with the m_next field. Records * are chained together with the m_nextpkt field. The upper level routine * soreceive() expects the following conventions to be observed when placing * information in the receive buffer: * * 1. If the protocol requires each message be preceded by the sender's name, * then a record containing that name must be present before any * associated data (mbuf's must be of type MT_SONAME). * 2. If the protocol supports the exchange of ``access rights'' (really just * additional data associated with the message), and there are ``rights'' * to be received, then a record containing this data should be present * (mbuf's must be of type MT_RIGHTS). * 3. If a name or rights record exists, then it must be followed by a data * record, perhaps of zero length. * * Before using a new socket structure it is first necessary to reserve * buffer space to the socket, by calling sbreserve(). This should commit * some of the available buffer space in the system buffer pool for the * socket (currently, it does nothing but enforce limits). The space should * be released by calling sbrelease() when the socket is destroyed. */ int soreserve(struct socket *so, u_long sndcc, u_long rcvcc) { struct thread *td = curthread; SOCKBUF_LOCK(&so->so_snd); SOCKBUF_LOCK(&so->so_rcv); if (sbreserve_locked(&so->so_snd, sndcc, so, td) == 0) goto bad; if (sbreserve_locked(&so->so_rcv, rcvcc, so, td) == 0) goto bad2; if (so->so_rcv.sb_lowat == 0) so->so_rcv.sb_lowat = 1; if (so->so_snd.sb_lowat == 0) so->so_snd.sb_lowat = MCLBYTES; if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat) so->so_snd.sb_lowat = so->so_snd.sb_hiwat; SOCKBUF_UNLOCK(&so->so_rcv); SOCKBUF_UNLOCK(&so->so_snd); return (0); bad2: sbrelease_locked(&so->so_snd, so); bad: SOCKBUF_UNLOCK(&so->so_rcv); SOCKBUF_UNLOCK(&so->so_snd); return (ENOBUFS); } static int sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS) { int error = 0; u_long tmp_sb_max = sb_max; error = sysctl_handle_long(oidp, &tmp_sb_max, arg2, req); if (error || !req->newptr) return (error); if (tmp_sb_max < MSIZE + MCLBYTES) return (EINVAL); sb_max = tmp_sb_max; sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES); return (0); } /* * Allot mbufs to a sockbuf. Attempt to scale mbmax so that mbcnt doesn't * become limiting if buffering efficiency is near the normal case. */ int sbreserve_locked(struct sockbuf *sb, u_long cc, struct socket *so, struct thread *td) { rlim_t sbsize_limit; SOCKBUF_LOCK_ASSERT(sb); /* * When a thread is passed, we take into account the thread's socket * buffer size limit. The caller will generally pass curthread, but * in the TCP input path, NULL will be passed to indicate that no * appropriate thread resource limits are available. In that case, * we don't apply a process limit. */ if (cc > sb_max_adj) return (0); if (td != NULL) { - PROC_LOCK(td->td_proc); - sbsize_limit = lim_cur(td->td_proc, RLIMIT_SBSIZE); - PROC_UNLOCK(td->td_proc); + sbsize_limit = lim_cur(td, RLIMIT_SBSIZE); } else sbsize_limit = RLIM_INFINITY; if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc, sbsize_limit)) return (0); sb->sb_mbmax = min(cc * sb_efficiency, sb_max); if (sb->sb_lowat > sb->sb_hiwat) sb->sb_lowat = sb->sb_hiwat; return (1); } int sbreserve(struct sockbuf *sb, u_long cc, struct socket *so, struct thread *td) { int error; SOCKBUF_LOCK(sb); error = sbreserve_locked(sb, cc, so, td); SOCKBUF_UNLOCK(sb); return (error); } /* * Free mbufs held by a socket, and reserved mbuf space. */ void sbrelease_internal(struct sockbuf *sb, struct socket *so) { sbflush_internal(sb); (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, RLIM_INFINITY); sb->sb_mbmax = 0; } void sbrelease_locked(struct sockbuf *sb, struct socket *so) { SOCKBUF_LOCK_ASSERT(sb); sbrelease_internal(sb, so); } void sbrelease(struct sockbuf *sb, struct socket *so) { SOCKBUF_LOCK(sb); sbrelease_locked(sb, so); SOCKBUF_UNLOCK(sb); } void sbdestroy(struct sockbuf *sb, struct socket *so) { sbrelease_internal(sb, so); } /* * Routines to add and remove data from an mbuf queue. * * The routines sbappend() or sbappendrecord() are normally called to append * new mbufs to a socket buffer, after checking that adequate space is * available, comparing the function sbspace() with the amount of data to be * added. sbappendrecord() differs from sbappend() in that data supplied is * treated as the beginning of a new record. To place a sender's address, * optional access rights, and data in a socket receive buffer, * sbappendaddr() should be used. To place access rights and data in a * socket receive buffer, sbappendrights() should be used. In either case, * the new data begins a new record. Note that unlike sbappend() and * sbappendrecord(), these routines check for the caller that there will be * enough space to store the data. Each fails if there is not enough space, * or if it cannot find mbufs to store additional information in. * * Reliable protocols may use the socket send buffer to hold data awaiting * acknowledgement. Data is normally copied from a socket send buffer in a * protocol with m_copy for output to a peer, and then removing the data from * the socket buffer with sbdrop() or sbdroprecord() when the data is * acknowledged by the peer. */ #ifdef SOCKBUF_DEBUG void sblastrecordchk(struct sockbuf *sb, const char *file, int line) { struct mbuf *m = sb->sb_mb; SOCKBUF_LOCK_ASSERT(sb); while (m && m->m_nextpkt) m = m->m_nextpkt; if (m != sb->sb_lastrecord) { printf("%s: sb_mb %p sb_lastrecord %p last %p\n", __func__, sb->sb_mb, sb->sb_lastrecord, m); printf("packet chain:\n"); for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) printf("\t%p\n", m); panic("%s from %s:%u", __func__, file, line); } } void sblastmbufchk(struct sockbuf *sb, const char *file, int line) { struct mbuf *m = sb->sb_mb; struct mbuf *n; SOCKBUF_LOCK_ASSERT(sb); while (m && m->m_nextpkt) m = m->m_nextpkt; while (m && m->m_next) m = m->m_next; if (m != sb->sb_mbtail) { printf("%s: sb_mb %p sb_mbtail %p last %p\n", __func__, sb->sb_mb, sb->sb_mbtail, m); printf("packet tree:\n"); for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) { printf("\t"); for (n = m; n != NULL; n = n->m_next) printf("%p ", n); printf("\n"); } panic("%s from %s:%u", __func__, file, line); } } #endif /* SOCKBUF_DEBUG */ #define SBLINKRECORD(sb, m0) do { \ SOCKBUF_LOCK_ASSERT(sb); \ if ((sb)->sb_lastrecord != NULL) \ (sb)->sb_lastrecord->m_nextpkt = (m0); \ else \ (sb)->sb_mb = (m0); \ (sb)->sb_lastrecord = (m0); \ } while (/*CONSTCOND*/0) /* * Append mbuf chain m to the last record in the socket buffer sb. The * additional space associated the mbuf chain is recorded in sb. Empty mbufs * are discarded and mbufs are compacted where possible. */ void sbappend_locked(struct sockbuf *sb, struct mbuf *m) { struct mbuf *n; SOCKBUF_LOCK_ASSERT(sb); if (m == 0) return; m_clrprotoflags(m); SBLASTRECORDCHK(sb); n = sb->sb_mb; if (n) { while (n->m_nextpkt) n = n->m_nextpkt; do { if (n->m_flags & M_EOR) { sbappendrecord_locked(sb, m); /* XXXXXX!!!! */ return; } } while (n->m_next && (n = n->m_next)); } else { /* * XXX Would like to simply use sb_mbtail here, but * XXX I need to verify that I won't miss an EOR that * XXX way. */ if ((n = sb->sb_lastrecord) != NULL) { do { if (n->m_flags & M_EOR) { sbappendrecord_locked(sb, m); /* XXXXXX!!!! */ return; } } while (n->m_next && (n = n->m_next)); } else { /* * If this is the first record in the socket buffer, * it's also the last record. */ sb->sb_lastrecord = m; } } sbcompress(sb, m, n); SBLASTRECORDCHK(sb); } /* * Append mbuf chain m to the last record in the socket buffer sb. The * additional space associated the mbuf chain is recorded in sb. Empty mbufs * are discarded and mbufs are compacted where possible. */ void sbappend(struct sockbuf *sb, struct mbuf *m) { SOCKBUF_LOCK(sb); sbappend_locked(sb, m); SOCKBUF_UNLOCK(sb); } /* * This version of sbappend() should only be used when the caller absolutely * knows that there will never be more than one record in the socket buffer, * that is, a stream protocol (such as TCP). */ void sbappendstream_locked(struct sockbuf *sb, struct mbuf *m, int flags) { SOCKBUF_LOCK_ASSERT(sb); KASSERT(m->m_nextpkt == NULL,("sbappendstream 0")); KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1")); SBLASTMBUFCHK(sb); /* Remove all packet headers and mbuf tags to get a pure data chain. */ m_demote(m, 1, flags & PRUS_NOTREADY ? M_NOTREADY : 0); sbcompress(sb, m, sb->sb_mbtail); sb->sb_lastrecord = sb->sb_mb; SBLASTRECORDCHK(sb); } /* * This version of sbappend() should only be used when the caller absolutely * knows that there will never be more than one record in the socket buffer, * that is, a stream protocol (such as TCP). */ void sbappendstream(struct sockbuf *sb, struct mbuf *m, int flags) { SOCKBUF_LOCK(sb); sbappendstream_locked(sb, m, flags); SOCKBUF_UNLOCK(sb); } #ifdef SOCKBUF_DEBUG void sbcheck(struct sockbuf *sb, const char *file, int line) { struct mbuf *m, *n, *fnrdy; u_long acc, ccc, mbcnt; SOCKBUF_LOCK_ASSERT(sb); acc = ccc = mbcnt = 0; fnrdy = NULL; for (m = sb->sb_mb; m; m = n) { n = m->m_nextpkt; for (; m; m = m->m_next) { if (m->m_len == 0) { printf("sb %p empty mbuf %p\n", sb, m); goto fail; } if ((m->m_flags & M_NOTREADY) && fnrdy == NULL) { if (m != sb->sb_fnrdy) { printf("sb %p: fnrdy %p != m %p\n", sb, sb->sb_fnrdy, m); goto fail; } fnrdy = m; } if (fnrdy) { if (!(m->m_flags & M_NOTAVAIL)) { printf("sb %p: fnrdy %p, m %p is avail\n", sb, sb->sb_fnrdy, m); goto fail; } } else acc += m->m_len; ccc += m->m_len; mbcnt += MSIZE; if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */ mbcnt += m->m_ext.ext_size; } } if (acc != sb->sb_acc || ccc != sb->sb_ccc || mbcnt != sb->sb_mbcnt) { printf("acc %ld/%u ccc %ld/%u mbcnt %ld/%u\n", acc, sb->sb_acc, ccc, sb->sb_ccc, mbcnt, sb->sb_mbcnt); goto fail; } return; fail: panic("%s from %s:%u", __func__, file, line); } #endif /* * As above, except the mbuf chain begins a new record. */ void sbappendrecord_locked(struct sockbuf *sb, struct mbuf *m0) { struct mbuf *m; SOCKBUF_LOCK_ASSERT(sb); if (m0 == 0) return; m_clrprotoflags(m0); /* * Put the first mbuf on the queue. Note this permits zero length * records. */ sballoc(sb, m0); SBLASTRECORDCHK(sb); SBLINKRECORD(sb, m0); sb->sb_mbtail = m0; m = m0->m_next; m0->m_next = 0; if (m && (m0->m_flags & M_EOR)) { m0->m_flags &= ~M_EOR; m->m_flags |= M_EOR; } /* always call sbcompress() so it can do SBLASTMBUFCHK() */ sbcompress(sb, m, m0); } /* * As above, except the mbuf chain begins a new record. */ void sbappendrecord(struct sockbuf *sb, struct mbuf *m0) { SOCKBUF_LOCK(sb); sbappendrecord_locked(sb, m0); SOCKBUF_UNLOCK(sb); } /* Helper routine that appends data, control, and address to a sockbuf. */ static int sbappendaddr_locked_internal(struct sockbuf *sb, const struct sockaddr *asa, struct mbuf *m0, struct mbuf *control, struct mbuf *ctrl_last) { struct mbuf *m, *n, *nlast; #if MSIZE <= 256 if (asa->sa_len > MLEN) return (0); #endif m = m_get(M_NOWAIT, MT_SONAME); if (m == NULL) return (0); m->m_len = asa->sa_len; bcopy(asa, mtod(m, caddr_t), asa->sa_len); if (m0) m_clrprotoflags(m0); if (ctrl_last) ctrl_last->m_next = m0; /* concatenate data to control */ else control = m0; m->m_next = control; for (n = m; n->m_next != NULL; n = n->m_next) sballoc(sb, n); sballoc(sb, n); nlast = n; SBLINKRECORD(sb, m); sb->sb_mbtail = nlast; SBLASTMBUFCHK(sb); SBLASTRECORDCHK(sb); return (1); } /* * Append address and data, and optionally, control (ancillary) data to the * receive queue of a socket. If present, m0 must include a packet header * with total length. Returns 0 if no space in sockbuf or insufficient * mbufs. */ int sbappendaddr_locked(struct sockbuf *sb, const struct sockaddr *asa, struct mbuf *m0, struct mbuf *control) { struct mbuf *ctrl_last; int space = asa->sa_len; SOCKBUF_LOCK_ASSERT(sb); if (m0 && (m0->m_flags & M_PKTHDR) == 0) panic("sbappendaddr_locked"); if (m0) space += m0->m_pkthdr.len; space += m_length(control, &ctrl_last); if (space > sbspace(sb)) return (0); return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last)); } /* * Append address and data, and optionally, control (ancillary) data to the * receive queue of a socket. If present, m0 must include a packet header * with total length. Returns 0 if insufficient mbufs. Does not validate space * on the receiving sockbuf. */ int sbappendaddr_nospacecheck_locked(struct sockbuf *sb, const struct sockaddr *asa, struct mbuf *m0, struct mbuf *control) { struct mbuf *ctrl_last; SOCKBUF_LOCK_ASSERT(sb); ctrl_last = (control == NULL) ? NULL : m_last(control); return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last)); } /* * Append address and data, and optionally, control (ancillary) data to the * receive queue of a socket. If present, m0 must include a packet header * with total length. Returns 0 if no space in sockbuf or insufficient * mbufs. */ int sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa, struct mbuf *m0, struct mbuf *control) { int retval; SOCKBUF_LOCK(sb); retval = sbappendaddr_locked(sb, asa, m0, control); SOCKBUF_UNLOCK(sb); return (retval); } int sbappendcontrol_locked(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control) { struct mbuf *m, *n, *mlast; int space; SOCKBUF_LOCK_ASSERT(sb); if (control == 0) panic("sbappendcontrol_locked"); space = m_length(control, &n) + m_length(m0, NULL); if (space > sbspace(sb)) return (0); m_clrprotoflags(m0); n->m_next = m0; /* concatenate data to control */ SBLASTRECORDCHK(sb); for (m = control; m->m_next; m = m->m_next) sballoc(sb, m); sballoc(sb, m); mlast = m; SBLINKRECORD(sb, control); sb->sb_mbtail = mlast; SBLASTMBUFCHK(sb); SBLASTRECORDCHK(sb); return (1); } int sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control) { int retval; SOCKBUF_LOCK(sb); retval = sbappendcontrol_locked(sb, m0, control); SOCKBUF_UNLOCK(sb); return (retval); } /* * Append the data in mbuf chain (m) into the socket buffer sb following mbuf * (n). If (n) is NULL, the buffer is presumed empty. * * When the data is compressed, mbufs in the chain may be handled in one of * three ways: * * (1) The mbuf may simply be dropped, if it contributes nothing (no data, no * record boundary, and no change in data type). * * (2) The mbuf may be coalesced -- i.e., data in the mbuf may be copied into * an mbuf already in the socket buffer. This can occur if an * appropriate mbuf exists, there is room, both mbufs are not marked as * not ready, and no merging of data types will occur. * * (3) The mbuf may be appended to the end of the existing mbuf chain. * * If any of the new mbufs is marked as M_EOR, mark the last mbuf appended as * end-of-record. */ void sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n) { int eor = 0; struct mbuf *o; SOCKBUF_LOCK_ASSERT(sb); while (m) { eor |= m->m_flags & M_EOR; if (m->m_len == 0 && (eor == 0 || (((o = m->m_next) || (o = n)) && o->m_type == m->m_type))) { if (sb->sb_lastrecord == m) sb->sb_lastrecord = m->m_next; m = m_free(m); continue; } if (n && (n->m_flags & M_EOR) == 0 && M_WRITABLE(n) && ((sb->sb_flags & SB_NOCOALESCE) == 0) && !(m->m_flags & M_NOTREADY) && !(n->m_flags & M_NOTREADY) && m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */ m->m_len <= M_TRAILINGSPACE(n) && n->m_type == m->m_type) { bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len, (unsigned)m->m_len); n->m_len += m->m_len; sb->sb_ccc += m->m_len; if (sb->sb_fnrdy == NULL) sb->sb_acc += m->m_len; if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) /* XXX: Probably don't need.*/ sb->sb_ctl += m->m_len; m = m_free(m); continue; } if (n) n->m_next = m; else sb->sb_mb = m; sb->sb_mbtail = m; sballoc(sb, m); n = m; m->m_flags &= ~M_EOR; m = m->m_next; n->m_next = 0; } if (eor) { KASSERT(n != NULL, ("sbcompress: eor && n == NULL")); n->m_flags |= eor; } SBLASTMBUFCHK(sb); } /* * Free all mbufs in a sockbuf. Check that all resources are reclaimed. */ static void sbflush_internal(struct sockbuf *sb) { while (sb->sb_mbcnt) { /* * Don't call sbcut(sb, 0) if the leading mbuf is non-empty: * we would loop forever. Panic instead. */ if (sb->sb_ccc == 0 && (sb->sb_mb == NULL || sb->sb_mb->m_len)) break; m_freem(sbcut_internal(sb, (int)sb->sb_ccc)); } KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0, ("%s: ccc %u mb %p mbcnt %u", __func__, sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt)); } void sbflush_locked(struct sockbuf *sb) { SOCKBUF_LOCK_ASSERT(sb); sbflush_internal(sb); } void sbflush(struct sockbuf *sb) { SOCKBUF_LOCK(sb); sbflush_locked(sb); SOCKBUF_UNLOCK(sb); } /* * Cut data from (the front of) a sockbuf. */ static struct mbuf * sbcut_internal(struct sockbuf *sb, int len) { struct mbuf *m, *next, *mfree; next = (m = sb->sb_mb) ? m->m_nextpkt : 0; mfree = NULL; while (len > 0) { if (m == NULL) { KASSERT(next, ("%s: no next, len %d", __func__, len)); m = next; next = m->m_nextpkt; } if (m->m_len > len) { KASSERT(!(m->m_flags & M_NOTAVAIL), ("%s: m %p M_NOTAVAIL", __func__, m)); m->m_len -= len; m->m_data += len; sb->sb_ccc -= len; sb->sb_acc -= len; if (sb->sb_sndptroff != 0) sb->sb_sndptroff -= len; if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) sb->sb_ctl -= len; break; } len -= m->m_len; sbfree(sb, m); /* * Do not put M_NOTREADY buffers to the free list, they * are referenced from outside. */ if (m->m_flags & M_NOTREADY) m = m->m_next; else { struct mbuf *n; n = m->m_next; m->m_next = mfree; mfree = m; m = n; } } /* * Free any zero-length mbufs from the buffer. * For SOCK_DGRAM sockets such mbufs represent empty records. * XXX: For SOCK_STREAM sockets such mbufs can appear in the buffer, * when sosend_generic() needs to send only control data. */ while (m && m->m_len == 0) { struct mbuf *n; sbfree(sb, m); n = m->m_next; m->m_next = mfree; mfree = m; m = n; } if (m) { sb->sb_mb = m; m->m_nextpkt = next; } else sb->sb_mb = next; /* * First part is an inline SB_EMPTY_FIXUP(). Second part makes sure * sb_lastrecord is up-to-date if we dropped part of the last record. */ m = sb->sb_mb; if (m == NULL) { sb->sb_mbtail = NULL; sb->sb_lastrecord = NULL; } else if (m->m_nextpkt == NULL) { sb->sb_lastrecord = m; } return (mfree); } /* * Drop data from (the front of) a sockbuf. */ void sbdrop_locked(struct sockbuf *sb, int len) { SOCKBUF_LOCK_ASSERT(sb); m_freem(sbcut_internal(sb, len)); } /* * Drop data from (the front of) a sockbuf, * and return it to caller. */ struct mbuf * sbcut_locked(struct sockbuf *sb, int len) { SOCKBUF_LOCK_ASSERT(sb); return (sbcut_internal(sb, len)); } void sbdrop(struct sockbuf *sb, int len) { struct mbuf *mfree; SOCKBUF_LOCK(sb); mfree = sbcut_internal(sb, len); SOCKBUF_UNLOCK(sb); m_freem(mfree); } /* * Maintain a pointer and offset pair into the socket buffer mbuf chain to * avoid traversal of the entire socket buffer for larger offsets. */ struct mbuf * sbsndptr(struct sockbuf *sb, u_int off, u_int len, u_int *moff) { struct mbuf *m, *ret; KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__)); KASSERT(off + len <= sb->sb_acc, ("%s: beyond sb", __func__)); KASSERT(sb->sb_sndptroff <= sb->sb_acc, ("%s: sndptroff broken", __func__)); /* * Is off below stored offset? Happens on retransmits. * Just return, we can't help here. */ if (sb->sb_sndptroff > off) { *moff = off; return (sb->sb_mb); } /* Return closest mbuf in chain for current offset. */ *moff = off - sb->sb_sndptroff; m = ret = sb->sb_sndptr ? sb->sb_sndptr : sb->sb_mb; if (*moff == m->m_len) { *moff = 0; sb->sb_sndptroff += m->m_len; m = ret = m->m_next; KASSERT(ret->m_len > 0, ("mbuf %p in sockbuf %p chain has no valid data", ret, sb)); } /* Advance by len to be as close as possible for the next transmit. */ for (off = off - sb->sb_sndptroff + len - 1; off > 0 && m != NULL && off >= m->m_len; m = m->m_next) { sb->sb_sndptroff += m->m_len; off -= m->m_len; } if (off > 0 && m == NULL) panic("%s: sockbuf %p and mbuf %p clashing", __func__, sb, ret); sb->sb_sndptr = m; return (ret); } /* * Return the first mbuf and the mbuf data offset for the provided * send offset without changing the "sb_sndptroff" field. */ struct mbuf * sbsndmbuf(struct sockbuf *sb, u_int off, u_int *moff) { struct mbuf *m; KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__)); /* * If the "off" is below the stored offset, which happens on * retransmits, just use "sb_mb": */ if (sb->sb_sndptr == NULL || sb->sb_sndptroff > off) { m = sb->sb_mb; } else { m = sb->sb_sndptr; off -= sb->sb_sndptroff; } while (off > 0 && m != NULL) { if (off < m->m_len) break; off -= m->m_len; m = m->m_next; } *moff = off; return (m); } /* * Drop a record off the front of a sockbuf and move the next record to the * front. */ void sbdroprecord_locked(struct sockbuf *sb) { struct mbuf *m; SOCKBUF_LOCK_ASSERT(sb); m = sb->sb_mb; if (m) { sb->sb_mb = m->m_nextpkt; do { sbfree(sb, m); m = m_free(m); } while (m); } SB_EMPTY_FIXUP(sb); } /* * Drop a record off the front of a sockbuf and move the next record to the * front. */ void sbdroprecord(struct sockbuf *sb) { SOCKBUF_LOCK(sb); sbdroprecord_locked(sb); SOCKBUF_UNLOCK(sb); } /* * Create a "control" mbuf containing the specified data with the specified * type for presentation on a socket buffer. */ struct mbuf * sbcreatecontrol(caddr_t p, int size, int type, int level) { struct cmsghdr *cp; struct mbuf *m; if (CMSG_SPACE((u_int)size) > MCLBYTES) return ((struct mbuf *) NULL); if (CMSG_SPACE((u_int)size) > MLEN) m = m_getcl(M_NOWAIT, MT_CONTROL, 0); else m = m_get(M_NOWAIT, MT_CONTROL); if (m == NULL) return ((struct mbuf *) NULL); cp = mtod(m, struct cmsghdr *); m->m_len = 0; KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m), ("sbcreatecontrol: short mbuf")); /* * Don't leave the padding between the msg header and the * cmsg data and the padding after the cmsg data un-initialized. */ bzero(cp, CMSG_SPACE((u_int)size)); if (p != NULL) (void)memcpy(CMSG_DATA(cp), p, size); m->m_len = CMSG_SPACE(size); cp->cmsg_len = CMSG_LEN(size); cp->cmsg_level = level; cp->cmsg_type = type; return (m); } /* * This does the same for socket buffers that sotoxsocket does for sockets: * generate an user-format data structure describing the socket buffer. Note * that the xsockbuf structure, since it is always embedded in a socket, does * not include a self pointer nor a length. We make this entry point public * in case some other mechanism needs it. */ void sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb) { xsb->sb_cc = sb->sb_ccc; xsb->sb_hiwat = sb->sb_hiwat; xsb->sb_mbcnt = sb->sb_mbcnt; xsb->sb_mcnt = sb->sb_mcnt; xsb->sb_ccnt = sb->sb_ccnt; xsb->sb_mbmax = sb->sb_mbmax; xsb->sb_lowat = sb->sb_lowat; xsb->sb_flags = sb->sb_flags; xsb->sb_timeo = sb->sb_timeo; } /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */ static int dummy; SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, ""); SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_ULONG|CTLFLAG_RW, &sb_max, 0, sysctl_handle_sb_max, "LU", "Maximum socket buffer size"); SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW, &sb_efficiency, 0, "Socket buffer size waste factor"); Index: head/sys/kern/vfs_vnops.c =================================================================== --- head/sys/kern/vfs_vnops.c (revision 284214) +++ head/sys/kern/vfs_vnops.c (revision 284215) @@ -1,2460 +1,2459 @@ /*- * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Copyright (c) 2012 Konstantin Belousov * Copyright (c) 2013, 2014 The FreeBSD Foundation * * Portions of this software were developed by Konstantin Belousov * under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94 */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static fo_rdwr_t vn_read; static fo_rdwr_t vn_write; static fo_rdwr_t vn_io_fault; static fo_truncate_t vn_truncate; static fo_ioctl_t vn_ioctl; static fo_poll_t vn_poll; static fo_kqfilter_t vn_kqfilter; static fo_stat_t vn_statfile; static fo_close_t vn_closefile; static fo_mmap_t vn_mmap; struct fileops vnops = { .fo_read = vn_io_fault, .fo_write = vn_io_fault, .fo_truncate = vn_truncate, .fo_ioctl = vn_ioctl, .fo_poll = vn_poll, .fo_kqfilter = vn_kqfilter, .fo_stat = vn_statfile, .fo_close = vn_closefile, .fo_chmod = vn_chmod, .fo_chown = vn_chown, .fo_sendfile = vn_sendfile, .fo_seek = vn_seek, .fo_fill_kinfo = vn_fill_kinfo, .fo_mmap = vn_mmap, .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE }; static const int io_hold_cnt = 16; static int vn_io_fault_enable = 1; SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW, &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance"); static u_long vn_io_faults_cnt; SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD, &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers"); /* * Returns true if vn_io_fault mode of handling the i/o request should * be used. */ static bool do_vn_io_fault(struct vnode *vp, struct uio *uio) { struct mount *mp; return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG && (mp = vp->v_mount) != NULL && (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable); } /* * Structure used to pass arguments to vn_io_fault1(), to do either * file- or vnode-based I/O calls. */ struct vn_io_fault_args { enum { VN_IO_FAULT_FOP, VN_IO_FAULT_VOP } kind; struct ucred *cred; int flags; union { struct fop_args_tag { struct file *fp; fo_rdwr_t *doio; } fop_args; struct vop_args_tag { struct vnode *vp; } vop_args; } args; }; static int vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args, struct thread *td); int vn_open(ndp, flagp, cmode, fp) struct nameidata *ndp; int *flagp, cmode; struct file *fp; { struct thread *td = ndp->ni_cnd.cn_thread; return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp)); } /* * Common code for vnode open operations via a name lookup. * Lookup the vnode and invoke VOP_CREATE if needed. * Check permissions, and call the VOP_OPEN or VOP_CREATE routine. * * Note that this does NOT free nameidata for the successful case, * due to the NDINIT being done elsewhere. */ int vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags, struct ucred *cred, struct file *fp) { struct vnode *vp; struct mount *mp; struct thread *td = ndp->ni_cnd.cn_thread; struct vattr vat; struct vattr *vap = &vat; int fmode, error; restart: fmode = *flagp; if (fmode & O_CREAT) { ndp->ni_cnd.cn_nameiop = CREATE; /* * Set NOCACHE to avoid flushing the cache when * rolling in many files at once. */ ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | NOCACHE; if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0) ndp->ni_cnd.cn_flags |= FOLLOW; if (!(vn_open_flags & VN_OPEN_NOAUDIT)) ndp->ni_cnd.cn_flags |= AUDITVNODE1; if (vn_open_flags & VN_OPEN_NOCAPCHECK) ndp->ni_cnd.cn_flags |= NOCAPCHECK; bwillwrite(); if ((error = namei(ndp)) != 0) return (error); if (ndp->ni_vp == NULL) { VATTR_NULL(vap); vap->va_type = VREG; vap->va_mode = cmode; if (fmode & O_EXCL) vap->va_vaflags |= VA_EXCLUSIVE; if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(ndp, NDF_ONLY_PNBUF); vput(ndp->ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) return (error); goto restart; } if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0) ndp->ni_cnd.cn_flags |= MAKEENTRY; #ifdef MAC error = mac_vnode_check_create(cred, ndp->ni_dvp, &ndp->ni_cnd, vap); if (error == 0) #endif error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, vap); vput(ndp->ni_dvp); vn_finished_write(mp); if (error) { NDFREE(ndp, NDF_ONLY_PNBUF); return (error); } fmode &= ~O_TRUNC; vp = ndp->ni_vp; } else { if (ndp->ni_dvp == ndp->ni_vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); ndp->ni_dvp = NULL; vp = ndp->ni_vp; if (fmode & O_EXCL) { error = EEXIST; goto bad; } fmode &= ~O_CREAT; } } else { ndp->ni_cnd.cn_nameiop = LOOKUP; ndp->ni_cnd.cn_flags = ISOPEN | ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF; if (!(fmode & FWRITE)) ndp->ni_cnd.cn_flags |= LOCKSHARED; if (!(vn_open_flags & VN_OPEN_NOAUDIT)) ndp->ni_cnd.cn_flags |= AUDITVNODE1; if (vn_open_flags & VN_OPEN_NOCAPCHECK) ndp->ni_cnd.cn_flags |= NOCAPCHECK; if ((error = namei(ndp)) != 0) return (error); vp = ndp->ni_vp; } error = vn_open_vnode(vp, fmode, cred, td, fp); if (error) goto bad; *flagp = fmode; return (0); bad: NDFREE(ndp, NDF_ONLY_PNBUF); vput(vp); *flagp = fmode; ndp->ni_vp = NULL; return (error); } /* * Common code for vnode open operations once a vnode is located. * Check permissions, and call the VOP_OPEN routine. */ int vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred, struct thread *td, struct file *fp) { struct mount *mp; accmode_t accmode; struct flock lf; int error, have_flock, lock_flags, type; if (vp->v_type == VLNK) return (EMLINK); if (vp->v_type == VSOCK) return (EOPNOTSUPP); if (vp->v_type != VDIR && fmode & O_DIRECTORY) return (ENOTDIR); accmode = 0; if (fmode & (FWRITE | O_TRUNC)) { if (vp->v_type == VDIR) return (EISDIR); accmode |= VWRITE; } if (fmode & FREAD) accmode |= VREAD; if (fmode & FEXEC) accmode |= VEXEC; if ((fmode & O_APPEND) && (fmode & FWRITE)) accmode |= VAPPEND; #ifdef MAC if (fmode & O_CREAT) accmode |= VCREAT; if (fmode & O_VERIFY) accmode |= VVERIFY; error = mac_vnode_check_open(cred, vp, accmode); if (error) return (error); accmode &= ~(VCREAT | VVERIFY); #endif if ((fmode & O_CREAT) == 0) { if (accmode & VWRITE) { error = vn_writechk(vp); if (error) return (error); } if (accmode) { error = VOP_ACCESS(vp, accmode, cred, td); if (error) return (error); } } if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) vn_lock(vp, LK_UPGRADE | LK_RETRY); if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0) return (error); if (fmode & (O_EXLOCK | O_SHLOCK)) { KASSERT(fp != NULL, ("open with flock requires fp")); lock_flags = VOP_ISLOCKED(vp); VOP_UNLOCK(vp, 0); lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; if (fmode & O_EXLOCK) lf.l_type = F_WRLCK; else lf.l_type = F_RDLCK; type = F_FLOCK; if ((fmode & FNONBLOCK) == 0) type |= F_WAIT; error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type); have_flock = (error == 0); vn_lock(vp, lock_flags | LK_RETRY); if (error == 0 && vp->v_iflag & VI_DOOMED) error = ENOENT; /* * Another thread might have used this vnode as an * executable while the vnode lock was dropped. * Ensure the vnode is still able to be opened for * writing after the lock has been obtained. */ if (error == 0 && accmode & VWRITE) error = vn_writechk(vp); if (error) { VOP_UNLOCK(vp, 0); if (have_flock) { lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; lf.l_type = F_UNLCK; (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK); } vn_start_write(vp, &mp, V_WAIT); vn_lock(vp, lock_flags | LK_RETRY); (void)VOP_CLOSE(vp, fmode, cred, td); vn_finished_write(mp); /* Prevent second close from fdrop()->vn_close(). */ if (fp != NULL) fp->f_ops= &badfileops; return (error); } fp->f_flag |= FHASLOCK; } if (fmode & FWRITE) { VOP_ADD_WRITECOUNT(vp, 1); CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d", __func__, vp, vp->v_writecount); } ASSERT_VOP_LOCKED(vp, "vn_open_vnode"); return (0); } /* * Check for write permissions on the specified vnode. * Prototype text segments cannot be written. */ int vn_writechk(vp) register struct vnode *vp; { ASSERT_VOP_LOCKED(vp, "vn_writechk"); /* * If there's shared text associated with * the vnode, try to free it up once. If * we fail, we can't allow writing. */ if (VOP_IS_TEXT(vp)) return (ETXTBSY); return (0); } /* * Vnode close call */ int vn_close(vp, flags, file_cred, td) register struct vnode *vp; int flags; struct ucred *file_cred; struct thread *td; { struct mount *mp; int error, lock_flags; if (vp->v_type != VFIFO && (flags & FWRITE) == 0 && MNT_EXTENDED_SHARED(vp->v_mount)) lock_flags = LK_SHARED; else lock_flags = LK_EXCLUSIVE; vn_start_write(vp, &mp, V_WAIT); vn_lock(vp, lock_flags | LK_RETRY); if (flags & FWRITE) { VNASSERT(vp->v_writecount > 0, vp, ("vn_close: negative writecount")); VOP_ADD_WRITECOUNT(vp, -1); CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d", __func__, vp, vp->v_writecount); } error = VOP_CLOSE(vp, flags, file_cred, td); vput(vp); vn_finished_write(mp); return (error); } /* * Heuristic to detect sequential operation. */ static int sequential_heuristic(struct uio *uio, struct file *fp) { ASSERT_VOP_LOCKED(fp->f_vnode, __func__); if (fp->f_flag & FRDAHEAD) return (fp->f_seqcount << IO_SEQSHIFT); /* * Offset 0 is handled specially. open() sets f_seqcount to 1 so * that the first I/O is normally considered to be slightly * sequential. Seeking to offset 0 doesn't change sequentiality * unless previous seeks have reduced f_seqcount to 0, in which * case offset 0 is not special. */ if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || uio->uio_offset == fp->f_nextoff) { /* * f_seqcount is in units of fixed-size blocks so that it * depends mainly on the amount of sequential I/O and not * much on the number of sequential I/O's. The fixed size * of 16384 is hard-coded here since it is (not quite) just * a magic size that works well here. This size is more * closely related to the best I/O size for real disks than * to any block size used by software. */ fp->f_seqcount += howmany(uio->uio_resid, 16384); if (fp->f_seqcount > IO_SEQMAX) fp->f_seqcount = IO_SEQMAX; return (fp->f_seqcount << IO_SEQSHIFT); } /* Not sequential. Quickly draw-down sequentiality. */ if (fp->f_seqcount > 1) fp->f_seqcount = 1; else fp->f_seqcount = 0; return (0); } /* * Package up an I/O request on a vnode into a uio and do it. */ int vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset, enum uio_seg segflg, int ioflg, struct ucred *active_cred, struct ucred *file_cred, ssize_t *aresid, struct thread *td) { struct uio auio; struct iovec aiov; struct mount *mp; struct ucred *cred; void *rl_cookie; struct vn_io_fault_args args; int error, lock_flags; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; aiov.iov_base = base; aiov.iov_len = len; auio.uio_resid = len; auio.uio_offset = offset; auio.uio_segflg = segflg; auio.uio_rw = rw; auio.uio_td = td; error = 0; if ((ioflg & IO_NODELOCKED) == 0) { if ((ioflg & IO_RANGELOCKED) == 0) { if (rw == UIO_READ) { rl_cookie = vn_rangelock_rlock(vp, offset, offset + len); } else { rl_cookie = vn_rangelock_wlock(vp, offset, offset + len); } } else rl_cookie = NULL; mp = NULL; if (rw == UIO_WRITE) { if (vp->v_type != VCHR && (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) goto out; if (MNT_SHARED_WRITES(mp) || ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) lock_flags = LK_SHARED; else lock_flags = LK_EXCLUSIVE; } else lock_flags = LK_SHARED; vn_lock(vp, lock_flags | LK_RETRY); } else rl_cookie = NULL; ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); #ifdef MAC if ((ioflg & IO_NOMACCHECK) == 0) { if (rw == UIO_READ) error = mac_vnode_check_read(active_cred, file_cred, vp); else error = mac_vnode_check_write(active_cred, file_cred, vp); } #endif if (error == 0) { if (file_cred != NULL) cred = file_cred; else cred = active_cred; if (do_vn_io_fault(vp, &auio)) { args.kind = VN_IO_FAULT_VOP; args.cred = cred; args.flags = ioflg; args.args.vop_args.vp = vp; error = vn_io_fault1(vp, &auio, &args, td); } else if (rw == UIO_READ) { error = VOP_READ(vp, &auio, ioflg, cred); } else /* if (rw == UIO_WRITE) */ { error = VOP_WRITE(vp, &auio, ioflg, cred); } } if (aresid) *aresid = auio.uio_resid; else if (auio.uio_resid && error == 0) error = EIO; if ((ioflg & IO_NODELOCKED) == 0) { VOP_UNLOCK(vp, 0); if (mp != NULL) vn_finished_write(mp); } out: if (rl_cookie != NULL) vn_rangelock_unlock(vp, rl_cookie); return (error); } /* * Package up an I/O request on a vnode into a uio and do it. The I/O * request is split up into smaller chunks and we try to avoid saturating * the buffer cache while potentially holding a vnode locked, so we * check bwillwrite() before calling vn_rdwr(). We also call kern_yield() * to give other processes a chance to lock the vnode (either other processes * core'ing the same binary, or unrelated processes scanning the directory). */ int vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred, file_cred, aresid, td) enum uio_rw rw; struct vnode *vp; void *base; size_t len; off_t offset; enum uio_seg segflg; int ioflg; struct ucred *active_cred; struct ucred *file_cred; size_t *aresid; struct thread *td; { int error = 0; ssize_t iaresid; do { int chunk; /* * Force `offset' to a multiple of MAXBSIZE except possibly * for the first chunk, so that filesystems only need to * write full blocks except possibly for the first and last * chunks. */ chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; if (chunk > len) chunk = len; if (rw != UIO_READ && vp->v_type == VREG) bwillwrite(); iaresid = 0; error = vn_rdwr(rw, vp, base, chunk, offset, segflg, ioflg, active_cred, file_cred, &iaresid, td); len -= chunk; /* aresid calc already includes length */ if (error) break; offset += chunk; base = (char *)base + chunk; kern_yield(PRI_USER); } while (len); if (aresid) *aresid = len + iaresid; return (error); } off_t foffset_lock(struct file *fp, int flags) { struct mtx *mtxp; off_t res; KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); #if OFF_MAX <= LONG_MAX /* * Caller only wants the current f_offset value. Assume that * the long and shorter integer types reads are atomic. */ if ((flags & FOF_NOLOCK) != 0) return (fp->f_offset); #endif /* * According to McKusick the vn lock was protecting f_offset here. * It is now protected by the FOFFSET_LOCKED flag. */ mtxp = mtx_pool_find(mtxpool_sleep, fp); mtx_lock(mtxp); if ((flags & FOF_NOLOCK) == 0) { while (fp->f_vnread_flags & FOFFSET_LOCKED) { fp->f_vnread_flags |= FOFFSET_LOCK_WAITING; msleep(&fp->f_vnread_flags, mtxp, PUSER -1, "vofflock", 0); } fp->f_vnread_flags |= FOFFSET_LOCKED; } res = fp->f_offset; mtx_unlock(mtxp); return (res); } void foffset_unlock(struct file *fp, off_t val, int flags) { struct mtx *mtxp; KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); #if OFF_MAX <= LONG_MAX if ((flags & FOF_NOLOCK) != 0) { if ((flags & FOF_NOUPDATE) == 0) fp->f_offset = val; if ((flags & FOF_NEXTOFF) != 0) fp->f_nextoff = val; return; } #endif mtxp = mtx_pool_find(mtxpool_sleep, fp); mtx_lock(mtxp); if ((flags & FOF_NOUPDATE) == 0) fp->f_offset = val; if ((flags & FOF_NEXTOFF) != 0) fp->f_nextoff = val; if ((flags & FOF_NOLOCK) == 0) { KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0, ("Lost FOFFSET_LOCKED")); if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING) wakeup(&fp->f_vnread_flags); fp->f_vnread_flags = 0; } mtx_unlock(mtxp); } void foffset_lock_uio(struct file *fp, struct uio *uio, int flags) { if ((flags & FOF_OFFSET) == 0) uio->uio_offset = foffset_lock(fp, flags); } void foffset_unlock_uio(struct file *fp, struct uio *uio, int flags) { if ((flags & FOF_OFFSET) == 0) foffset_unlock(fp, uio->uio_offset, flags); } static int get_advice(struct file *fp, struct uio *uio) { struct mtx *mtxp; int ret; ret = POSIX_FADV_NORMAL; if (fp->f_advice == NULL) return (ret); mtxp = mtx_pool_find(mtxpool_sleep, fp); mtx_lock(mtxp); if (uio->uio_offset >= fp->f_advice->fa_start && uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end) ret = fp->f_advice->fa_advice; mtx_unlock(mtxp); return (ret); } /* * File table vnode read routine. */ static int vn_read(fp, uio, active_cred, flags, td) struct file *fp; struct uio *uio; struct ucred *active_cred; int flags; struct thread *td; { struct vnode *vp; struct mtx *mtxp; int error, ioflag; int advice; off_t offset, start, end; KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); vp = fp->f_vnode; ioflag = 0; if (fp->f_flag & FNONBLOCK) ioflag |= IO_NDELAY; if (fp->f_flag & O_DIRECT) ioflag |= IO_DIRECT; advice = get_advice(fp, uio); vn_lock(vp, LK_SHARED | LK_RETRY); switch (advice) { case POSIX_FADV_NORMAL: case POSIX_FADV_SEQUENTIAL: case POSIX_FADV_NOREUSE: ioflag |= sequential_heuristic(uio, fp); break; case POSIX_FADV_RANDOM: /* Disable read-ahead for random I/O. */ break; } offset = uio->uio_offset; #ifdef MAC error = mac_vnode_check_read(active_cred, fp->f_cred, vp); if (error == 0) #endif error = VOP_READ(vp, uio, ioflag, fp->f_cred); fp->f_nextoff = uio->uio_offset; VOP_UNLOCK(vp, 0); if (error == 0 && advice == POSIX_FADV_NOREUSE && offset != uio->uio_offset) { /* * Use POSIX_FADV_DONTNEED to flush clean pages and * buffers for the backing file after a * POSIX_FADV_NOREUSE read(2). To optimize the common * case of using POSIX_FADV_NOREUSE with sequential * access, track the previous implicit DONTNEED * request and grow this request to include the * current read(2) in addition to the previous * DONTNEED. With purely sequential access this will * cause the DONTNEED requests to continously grow to * cover all of the previously read regions of the * file. This allows filesystem blocks that are * accessed by multiple calls to read(2) to be flushed * once the last read(2) finishes. */ start = offset; end = uio->uio_offset - 1; mtxp = mtx_pool_find(mtxpool_sleep, fp); mtx_lock(mtxp); if (fp->f_advice != NULL && fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) { if (start != 0 && fp->f_advice->fa_prevend + 1 == start) start = fp->f_advice->fa_prevstart; else if (fp->f_advice->fa_prevstart != 0 && fp->f_advice->fa_prevstart == end + 1) end = fp->f_advice->fa_prevend; fp->f_advice->fa_prevstart = start; fp->f_advice->fa_prevend = end; } mtx_unlock(mtxp); error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED); } return (error); } /* * File table vnode write routine. */ static int vn_write(fp, uio, active_cred, flags, td) struct file *fp; struct uio *uio; struct ucred *active_cred; int flags; struct thread *td; { struct vnode *vp; struct mount *mp; struct mtx *mtxp; int error, ioflag, lock_flags; int advice; off_t offset, start, end; KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); vp = fp->f_vnode; if (vp->v_type == VREG) bwillwrite(); ioflag = IO_UNIT; if (vp->v_type == VREG && (fp->f_flag & O_APPEND)) ioflag |= IO_APPEND; if (fp->f_flag & FNONBLOCK) ioflag |= IO_NDELAY; if (fp->f_flag & O_DIRECT) ioflag |= IO_DIRECT; if ((fp->f_flag & O_FSYNC) || (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))) ioflag |= IO_SYNC; mp = NULL; if (vp->v_type != VCHR && (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) goto unlock; advice = get_advice(fp, uio); if (MNT_SHARED_WRITES(mp) || (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) { lock_flags = LK_SHARED; } else { lock_flags = LK_EXCLUSIVE; } vn_lock(vp, lock_flags | LK_RETRY); switch (advice) { case POSIX_FADV_NORMAL: case POSIX_FADV_SEQUENTIAL: case POSIX_FADV_NOREUSE: ioflag |= sequential_heuristic(uio, fp); break; case POSIX_FADV_RANDOM: /* XXX: Is this correct? */ break; } offset = uio->uio_offset; #ifdef MAC error = mac_vnode_check_write(active_cred, fp->f_cred, vp); if (error == 0) #endif error = VOP_WRITE(vp, uio, ioflag, fp->f_cred); fp->f_nextoff = uio->uio_offset; VOP_UNLOCK(vp, 0); if (vp->v_type != VCHR) vn_finished_write(mp); if (error == 0 && advice == POSIX_FADV_NOREUSE && offset != uio->uio_offset) { /* * Use POSIX_FADV_DONTNEED to flush clean pages and * buffers for the backing file after a * POSIX_FADV_NOREUSE write(2). To optimize the * common case of using POSIX_FADV_NOREUSE with * sequential access, track the previous implicit * DONTNEED request and grow this request to include * the current write(2) in addition to the previous * DONTNEED. With purely sequential access this will * cause the DONTNEED requests to continously grow to * cover all of the previously written regions of the * file. * * Note that the blocks just written are almost * certainly still dirty, so this only works when * VOP_ADVISE() calls from subsequent writes push out * the data written by this write(2) once the backing * buffers are clean. However, as compared to forcing * IO_DIRECT, this gives much saner behavior. Write * clustering is still allowed, and clean pages are * merely moved to the cache page queue rather than * outright thrown away. This means a subsequent * read(2) can still avoid hitting the disk if the * pages have not been reclaimed. * * This does make POSIX_FADV_NOREUSE largely useless * with non-sequential access. However, sequential * access is the more common use case and the flag is * merely advisory. */ start = offset; end = uio->uio_offset - 1; mtxp = mtx_pool_find(mtxpool_sleep, fp); mtx_lock(mtxp); if (fp->f_advice != NULL && fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) { if (start != 0 && fp->f_advice->fa_prevend + 1 == start) start = fp->f_advice->fa_prevstart; else if (fp->f_advice->fa_prevstart != 0 && fp->f_advice->fa_prevstart == end + 1) end = fp->f_advice->fa_prevend; fp->f_advice->fa_prevstart = start; fp->f_advice->fa_prevend = end; } mtx_unlock(mtxp); error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED); } unlock: return (error); } /* * The vn_io_fault() is a wrapper around vn_read() and vn_write() to * prevent the following deadlock: * * Assume that the thread A reads from the vnode vp1 into userspace * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is * currently not resident, then system ends up with the call chain * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] -> * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2) * which establishes lock order vp1->vn_lock, then vp2->vn_lock. * If, at the same time, thread B reads from vnode vp2 into buffer buf2 * backed by the pages of vnode vp1, and some page in buf2 is not * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock. * * To prevent the lock order reversal and deadlock, vn_io_fault() does * not allow page faults to happen during VOP_READ() or VOP_WRITE(). * Instead, it first tries to do the whole range i/o with pagefaults * disabled. If all pages in the i/o buffer are resident and mapped, * VOP will succeed (ignoring the genuine filesystem errors). * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do * i/o in chunks, with all pages in the chunk prefaulted and held * using vm_fault_quick_hold_pages(). * * Filesystems using this deadlock avoidance scheme should use the * array of the held pages from uio, saved in the curthread->td_ma, * instead of doing uiomove(). A helper function * vn_io_fault_uiomove() converts uiomove request into * uiomove_fromphys() over td_ma array. * * Since vnode locks do not cover the whole i/o anymore, rangelocks * make the current i/o request atomic with respect to other i/os and * truncations. */ /* * Decode vn_io_fault_args and perform the corresponding i/o. */ static int vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio, struct thread *td) { switch (args->kind) { case VN_IO_FAULT_FOP: return ((args->args.fop_args.doio)(args->args.fop_args.fp, uio, args->cred, args->flags, td)); case VN_IO_FAULT_VOP: if (uio->uio_rw == UIO_READ) { return (VOP_READ(args->args.vop_args.vp, uio, args->flags, args->cred)); } else if (uio->uio_rw == UIO_WRITE) { return (VOP_WRITE(args->args.vop_args.vp, uio, args->flags, args->cred)); } break; } panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind, uio->uio_rw); } /* * Common code for vn_io_fault(), agnostic to the kind of i/o request. * Uses vn_io_fault_doio() to make the call to an actual i/o function. * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request * into args and call vn_io_fault1() to handle faults during the user * mode buffer accesses. */ static int vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args, struct thread *td) { vm_page_t ma[io_hold_cnt + 2]; struct uio *uio_clone, short_uio; struct iovec short_iovec[1]; vm_page_t *prev_td_ma; vm_prot_t prot; vm_offset_t addr, end; size_t len, resid; ssize_t adv; int error, cnt, save, saveheld, prev_td_ma_cnt; prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ; /* * The UFS follows IO_UNIT directive and replays back both * uio_offset and uio_resid if an error is encountered during the * operation. But, since the iovec may be already advanced, * uio is still in an inconsistent state. * * Cache a copy of the original uio, which is advanced to the redo * point using UIO_NOCOPY below. */ uio_clone = cloneuio(uio); resid = uio->uio_resid; short_uio.uio_segflg = UIO_USERSPACE; short_uio.uio_rw = uio->uio_rw; short_uio.uio_td = uio->uio_td; save = vm_fault_disable_pagefaults(); error = vn_io_fault_doio(args, uio, td); if (error != EFAULT) goto out; atomic_add_long(&vn_io_faults_cnt, 1); uio_clone->uio_segflg = UIO_NOCOPY; uiomove(NULL, resid - uio->uio_resid, uio_clone); uio_clone->uio_segflg = uio->uio_segflg; saveheld = curthread_pflags_set(TDP_UIOHELD); prev_td_ma = td->td_ma; prev_td_ma_cnt = td->td_ma_cnt; while (uio_clone->uio_resid != 0) { len = uio_clone->uio_iov->iov_len; if (len == 0) { KASSERT(uio_clone->uio_iovcnt >= 1, ("iovcnt underflow")); uio_clone->uio_iov++; uio_clone->uio_iovcnt--; continue; } if (len > io_hold_cnt * PAGE_SIZE) len = io_hold_cnt * PAGE_SIZE; addr = (uintptr_t)uio_clone->uio_iov->iov_base; end = round_page(addr + len); if (end < addr) { error = EFAULT; break; } cnt = atop(end - trunc_page(addr)); /* * A perfectly misaligned address and length could cause * both the start and the end of the chunk to use partial * page. +2 accounts for such a situation. */ cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map, addr, len, prot, ma, io_hold_cnt + 2); if (cnt == -1) { error = EFAULT; break; } short_uio.uio_iov = &short_iovec[0]; short_iovec[0].iov_base = (void *)addr; short_uio.uio_iovcnt = 1; short_uio.uio_resid = short_iovec[0].iov_len = len; short_uio.uio_offset = uio_clone->uio_offset; td->td_ma = ma; td->td_ma_cnt = cnt; error = vn_io_fault_doio(args, &short_uio, td); vm_page_unhold_pages(ma, cnt); adv = len - short_uio.uio_resid; uio_clone->uio_iov->iov_base = (char *)uio_clone->uio_iov->iov_base + adv; uio_clone->uio_iov->iov_len -= adv; uio_clone->uio_resid -= adv; uio_clone->uio_offset += adv; uio->uio_resid -= adv; uio->uio_offset += adv; if (error != 0 || adv == 0) break; } td->td_ma = prev_td_ma; td->td_ma_cnt = prev_td_ma_cnt; curthread_pflags_restore(saveheld); out: vm_fault_enable_pagefaults(save); free(uio_clone, M_IOV); return (error); } static int vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { fo_rdwr_t *doio; struct vnode *vp; void *rl_cookie; struct vn_io_fault_args args; int error; doio = uio->uio_rw == UIO_READ ? vn_read : vn_write; vp = fp->f_vnode; foffset_lock_uio(fp, uio, flags); if (do_vn_io_fault(vp, uio)) { args.kind = VN_IO_FAULT_FOP; args.args.fop_args.fp = fp; args.args.fop_args.doio = doio; args.cred = active_cred; args.flags = flags | FOF_OFFSET; if (uio->uio_rw == UIO_READ) { rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset, uio->uio_offset + uio->uio_resid); } else if ((fp->f_flag & O_APPEND) != 0 || (flags & FOF_OFFSET) == 0) { /* For appenders, punt and lock the whole range. */ rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); } else { rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset, uio->uio_offset + uio->uio_resid); } error = vn_io_fault1(vp, uio, &args, td); vn_rangelock_unlock(vp, rl_cookie); } else { error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td); } foffset_unlock_uio(fp, uio, flags); return (error); } /* * Helper function to perform the requested uiomove operation using * the held pages for io->uio_iov[0].iov_base buffer instead of * copyin/copyout. Access to the pages with uiomove_fromphys() * instead of iov_base prevents page faults that could occur due to * pmap_collect() invalidating the mapping created by * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or * object cleanup revoking the write access from page mappings. * * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove() * instead of plain uiomove(). */ int vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio) { struct uio transp_uio; struct iovec transp_iov[1]; struct thread *td; size_t adv; int error, pgadv; td = curthread; if ((td->td_pflags & TDP_UIOHELD) == 0 || uio->uio_segflg != UIO_USERSPACE) return (uiomove(data, xfersize, uio)); KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); transp_iov[0].iov_base = data; transp_uio.uio_iov = &transp_iov[0]; transp_uio.uio_iovcnt = 1; if (xfersize > uio->uio_resid) xfersize = uio->uio_resid; transp_uio.uio_resid = transp_iov[0].iov_len = xfersize; transp_uio.uio_offset = 0; transp_uio.uio_segflg = UIO_SYSSPACE; /* * Since transp_iov points to data, and td_ma page array * corresponds to original uio->uio_iov, we need to invert the * direction of the i/o operation as passed to * uiomove_fromphys(). */ switch (uio->uio_rw) { case UIO_WRITE: transp_uio.uio_rw = UIO_READ; break; case UIO_READ: transp_uio.uio_rw = UIO_WRITE; break; } transp_uio.uio_td = uio->uio_td; error = uiomove_fromphys(td->td_ma, ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK, xfersize, &transp_uio); adv = xfersize - transp_uio.uio_resid; pgadv = (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) - (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT); td->td_ma += pgadv; KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, pgadv)); td->td_ma_cnt -= pgadv; uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv; uio->uio_iov->iov_len -= adv; uio->uio_resid -= adv; uio->uio_offset += adv; return (error); } int vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize, struct uio *uio) { struct thread *td; vm_offset_t iov_base; int cnt, pgadv; td = curthread; if ((td->td_pflags & TDP_UIOHELD) == 0 || uio->uio_segflg != UIO_USERSPACE) return (uiomove_fromphys(ma, offset, xfersize, uio)); KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize; iov_base = (vm_offset_t)uio->uio_iov->iov_base; switch (uio->uio_rw) { case UIO_WRITE: pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma, offset, cnt); break; case UIO_READ: pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK, cnt); break; } pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT); td->td_ma += pgadv; KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, pgadv)); td->td_ma_cnt -= pgadv; uio->uio_iov->iov_base = (char *)(iov_base + cnt); uio->uio_iov->iov_len -= cnt; uio->uio_resid -= cnt; uio->uio_offset += cnt; return (0); } /* * File table truncate routine. */ static int vn_truncate(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td) { struct vattr vattr; struct mount *mp; struct vnode *vp; void *rl_cookie; int error; vp = fp->f_vnode; /* * Lock the whole range for truncation. Otherwise split i/o * might happen partly before and partly after the truncation. */ rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); error = vn_start_write(vp, &mp, V_WAIT | PCATCH); if (error) goto out1; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); if (vp->v_type == VDIR) { error = EISDIR; goto out; } #ifdef MAC error = mac_vnode_check_write(active_cred, fp->f_cred, vp); if (error) goto out; #endif error = vn_writechk(vp); if (error == 0) { VATTR_NULL(&vattr); vattr.va_size = length; error = VOP_SETATTR(vp, &vattr, fp->f_cred); } out: VOP_UNLOCK(vp, 0); vn_finished_write(mp); out1: vn_rangelock_unlock(vp, rl_cookie); return (error); } /* * File table vnode stat routine. */ static int vn_statfile(fp, sb, active_cred, td) struct file *fp; struct stat *sb; struct ucred *active_cred; struct thread *td; { struct vnode *vp = fp->f_vnode; int error; vn_lock(vp, LK_SHARED | LK_RETRY); error = vn_stat(vp, sb, active_cred, fp->f_cred, td); VOP_UNLOCK(vp, 0); return (error); } /* * Stat a vnode; implementation for the stat syscall */ int vn_stat(vp, sb, active_cred, file_cred, td) struct vnode *vp; register struct stat *sb; struct ucred *active_cred; struct ucred *file_cred; struct thread *td; { struct vattr vattr; register struct vattr *vap; int error; u_short mode; #ifdef MAC error = mac_vnode_check_stat(active_cred, file_cred, vp); if (error) return (error); #endif vap = &vattr; /* * Initialize defaults for new and unusual fields, so that file * systems which don't support these fields don't need to know * about them. */ vap->va_birthtime.tv_sec = -1; vap->va_birthtime.tv_nsec = 0; vap->va_fsid = VNOVAL; vap->va_rdev = NODEV; error = VOP_GETATTR(vp, vap, active_cred); if (error) return (error); /* * Zero the spare stat fields */ bzero(sb, sizeof *sb); /* * Copy from vattr table */ if (vap->va_fsid != VNOVAL) sb->st_dev = vap->va_fsid; else sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; sb->st_ino = vap->va_fileid; mode = vap->va_mode; switch (vap->va_type) { case VREG: mode |= S_IFREG; break; case VDIR: mode |= S_IFDIR; break; case VBLK: mode |= S_IFBLK; break; case VCHR: mode |= S_IFCHR; break; case VLNK: mode |= S_IFLNK; break; case VSOCK: mode |= S_IFSOCK; break; case VFIFO: mode |= S_IFIFO; break; default: return (EBADF); }; sb->st_mode = mode; sb->st_nlink = vap->va_nlink; sb->st_uid = vap->va_uid; sb->st_gid = vap->va_gid; sb->st_rdev = vap->va_rdev; if (vap->va_size > OFF_MAX) return (EOVERFLOW); sb->st_size = vap->va_size; sb->st_atim = vap->va_atime; sb->st_mtim = vap->va_mtime; sb->st_ctim = vap->va_ctime; sb->st_birthtim = vap->va_birthtime; /* * According to www.opengroup.org, the meaning of st_blksize is * "a filesystem-specific preferred I/O block size for this * object. In some filesystem types, this may vary from file * to file" * Use miminum/default of PAGE_SIZE (e.g. for VCHR). */ sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize); sb->st_flags = vap->va_flags; if (priv_check(td, PRIV_VFS_GENERATION)) sb->st_gen = 0; else sb->st_gen = vap->va_gen; sb->st_blocks = vap->va_bytes / S_BLKSIZE; return (0); } /* * File table vnode ioctl routine. */ static int vn_ioctl(fp, com, data, active_cred, td) struct file *fp; u_long com; void *data; struct ucred *active_cred; struct thread *td; { struct vattr vattr; struct vnode *vp; int error; vp = fp->f_vnode; switch (vp->v_type) { case VDIR: case VREG: switch (com) { case FIONREAD: vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_GETATTR(vp, &vattr, active_cred); VOP_UNLOCK(vp, 0); if (error == 0) *(int *)data = vattr.va_size - fp->f_offset; return (error); case FIONBIO: case FIOASYNC: return (0); default: return (VOP_IOCTL(vp, com, data, fp->f_flag, active_cred, td)); } default: return (ENOTTY); } } /* * File table vnode poll routine. */ static int vn_poll(fp, events, active_cred, td) struct file *fp; int events; struct ucred *active_cred; struct thread *td; { struct vnode *vp; int error; vp = fp->f_vnode; #ifdef MAC vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); error = mac_vnode_check_poll(active_cred, fp->f_cred, vp); VOP_UNLOCK(vp, 0); if (!error) #endif error = VOP_POLL(vp, events, fp->f_cred, td); return (error); } /* * Acquire the requested lock and then check for validity. LK_RETRY * permits vn_lock to return doomed vnodes. */ int _vn_lock(struct vnode *vp, int flags, char *file, int line) { int error; VNASSERT((flags & LK_TYPE_MASK) != 0, vp, ("vn_lock called with no locktype.")); do { #ifdef DEBUG_VFS_LOCKS KASSERT(vp->v_holdcnt != 0, ("vn_lock %p: zero hold count", vp)); #endif error = VOP_LOCK1(vp, flags, file, line); flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */ KASSERT((flags & LK_RETRY) == 0 || error == 0, ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)", flags, error)); /* * Callers specify LK_RETRY if they wish to get dead vnodes. * If RETRY is not set, we return ENOENT instead. */ if (error == 0 && vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0) { VOP_UNLOCK(vp, 0); error = ENOENT; break; } } while (flags & LK_RETRY && error != 0); return (error); } /* * File table vnode close routine. */ static int vn_closefile(fp, td) struct file *fp; struct thread *td; { struct vnode *vp; struct flock lf; int error; vp = fp->f_vnode; fp->f_ops = &badfileops; if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) vref(vp); error = vn_close(vp, fp->f_flag, fp->f_cred, td); if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) { lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; lf.l_type = F_UNLCK; (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK); vrele(vp); } return (error); } static bool vn_suspendable_mp(struct mount *mp) { return ((mp->mnt_kern_flag & MNTK_SUSPENDABLE) != 0); } static bool vn_suspendable(struct vnode *vp, struct mount **mpp) { if (vp != NULL) *mpp = vp->v_mount; if (*mpp == NULL) return (false); return (vn_suspendable_mp(*mpp)); } /* * Preparing to start a filesystem write operation. If the operation is * permitted, then we bump the count of operations in progress and * proceed. If a suspend request is in progress, we wait until the * suspension is over, and then proceed. */ static int vn_start_write_locked(struct mount *mp, int flags) { int error, mflags; mtx_assert(MNT_MTX(mp), MA_OWNED); error = 0; /* * Check on status of suspension. */ if ((curthread->td_pflags & TDP_IGNSUSP) == 0 || mp->mnt_susp_owner != curthread) { mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0) | (PUSER - 1); while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { if (flags & V_NOWAIT) { error = EWOULDBLOCK; goto unlock; } error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags, "suspfs", 0); if (error) goto unlock; } } if (flags & V_XSLEEP) goto unlock; mp->mnt_writeopcount++; unlock: if (error != 0 || (flags & V_XSLEEP) != 0) MNT_REL(mp); MNT_IUNLOCK(mp); return (error); } int vn_start_write(struct vnode *vp, struct mount **mpp, int flags) { struct mount *mp; int error; KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL), ("V_MNTREF requires mp")); if (!vn_suspendable(vp, mpp)) { if ((flags & V_MNTREF) != 0) vfs_rel(*mpp); return (0); } error = 0; /* * If a vnode is provided, get and return the mount point that * to which it will write. */ if (vp != NULL) { if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { *mpp = NULL; if (error != EOPNOTSUPP) return (error); return (0); } } if ((mp = *mpp) == NULL) return (0); /* * VOP_GETWRITEMOUNT() returns with the mp refcount held through * a vfs_ref(). * As long as a vnode is not provided we need to acquire a * refcount for the provided mountpoint too, in order to * emulate a vfs_ref(). */ MNT_ILOCK(mp); if (vp == NULL && (flags & V_MNTREF) == 0) MNT_REF(mp); return (vn_start_write_locked(mp, flags)); } /* * Secondary suspension. Used by operations such as vop_inactive * routines that are needed by the higher level functions. These * are allowed to proceed until all the higher level functions have * completed (indicated by mnt_writeopcount dropping to zero). At that * time, these operations are halted until the suspension is over. */ int vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags) { struct mount *mp; int error; KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL), ("V_MNTREF requires mp")); if (!vn_suspendable(vp, mpp)) { if ((flags & V_MNTREF) != 0) vfs_rel(*mpp); return (0); } retry: if (vp != NULL) { if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { *mpp = NULL; if (error != EOPNOTSUPP) return (error); return (0); } } /* * If we are not suspended or have not yet reached suspended * mode, then let the operation proceed. */ if ((mp = *mpp) == NULL) return (0); /* * VOP_GETWRITEMOUNT() returns with the mp refcount held through * a vfs_ref(). * As long as a vnode is not provided we need to acquire a * refcount for the provided mountpoint too, in order to * emulate a vfs_ref(). */ MNT_ILOCK(mp); if (vp == NULL && (flags & V_MNTREF) == 0) MNT_REF(mp); if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) { mp->mnt_secondary_writes++; mp->mnt_secondary_accwrites++; MNT_IUNLOCK(mp); return (0); } if (flags & V_NOWAIT) { MNT_REL(mp); MNT_IUNLOCK(mp); return (EWOULDBLOCK); } /* * Wait for the suspension to finish. */ error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP | ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0), "suspfs", 0); vfs_rel(mp); if (error == 0) goto retry; return (error); } /* * Filesystem write operation has completed. If we are suspending and this * operation is the last one, notify the suspender that the suspension is * now in effect. */ void vn_finished_write(mp) struct mount *mp; { if (mp == NULL || !vn_suspendable_mp(mp)) return; MNT_ILOCK(mp); MNT_REL(mp); mp->mnt_writeopcount--; if (mp->mnt_writeopcount < 0) panic("vn_finished_write: neg cnt"); if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && mp->mnt_writeopcount <= 0) wakeup(&mp->mnt_writeopcount); MNT_IUNLOCK(mp); } /* * Filesystem secondary write operation has completed. If we are * suspending and this operation is the last one, notify the suspender * that the suspension is now in effect. */ void vn_finished_secondary_write(mp) struct mount *mp; { if (mp == NULL || !vn_suspendable_mp(mp)) return; MNT_ILOCK(mp); MNT_REL(mp); mp->mnt_secondary_writes--; if (mp->mnt_secondary_writes < 0) panic("vn_finished_secondary_write: neg cnt"); if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && mp->mnt_secondary_writes <= 0) wakeup(&mp->mnt_secondary_writes); MNT_IUNLOCK(mp); } /* * Request a filesystem to suspend write operations. */ int vfs_write_suspend(struct mount *mp, int flags) { int error; MPASS(vn_suspendable_mp(mp)); MNT_ILOCK(mp); if (mp->mnt_susp_owner == curthread) { MNT_IUNLOCK(mp); return (EALREADY); } while (mp->mnt_kern_flag & MNTK_SUSPEND) msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0); /* * Unmount holds a write reference on the mount point. If we * own busy reference and drain for writers, we deadlock with * the reference draining in the unmount path. Callers of * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if * vfs_busy() reference is owned and caller is not in the * unmount context. */ if ((flags & VS_SKIP_UNMOUNT) != 0 && (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) { MNT_IUNLOCK(mp); return (EBUSY); } mp->mnt_kern_flag |= MNTK_SUSPEND; mp->mnt_susp_owner = curthread; if (mp->mnt_writeopcount > 0) (void) msleep(&mp->mnt_writeopcount, MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0); else MNT_IUNLOCK(mp); if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0) vfs_write_resume(mp, 0); return (error); } /* * Request a filesystem to resume write operations. */ void vfs_write_resume(struct mount *mp, int flags) { MPASS(vn_suspendable_mp(mp)); MNT_ILOCK(mp); if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner")); mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 | MNTK_SUSPENDED); mp->mnt_susp_owner = NULL; wakeup(&mp->mnt_writeopcount); wakeup(&mp->mnt_flag); curthread->td_pflags &= ~TDP_IGNSUSP; if ((flags & VR_START_WRITE) != 0) { MNT_REF(mp); mp->mnt_writeopcount++; } MNT_IUNLOCK(mp); if ((flags & VR_NO_SUSPCLR) == 0) VFS_SUSP_CLEAN(mp); } else if ((flags & VR_START_WRITE) != 0) { MNT_REF(mp); vn_start_write_locked(mp, 0); } else { MNT_IUNLOCK(mp); } } /* * Helper loop around vfs_write_suspend() for filesystem unmount VFS * methods. */ int vfs_write_suspend_umnt(struct mount *mp) { int error; MPASS(vn_suspendable_mp(mp)); KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0, ("vfs_write_suspend_umnt: recursed")); /* dounmount() already called vn_start_write(). */ for (;;) { vn_finished_write(mp); error = vfs_write_suspend(mp, 0); if (error != 0) { vn_start_write(NULL, &mp, V_WAIT); return (error); } MNT_ILOCK(mp); if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0) break; MNT_IUNLOCK(mp); vn_start_write(NULL, &mp, V_WAIT); } mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2); wakeup(&mp->mnt_flag); MNT_IUNLOCK(mp); curthread->td_pflags |= TDP_IGNSUSP; return (0); } /* * Implement kqueues for files by translating it to vnode operation. */ static int vn_kqfilter(struct file *fp, struct knote *kn) { return (VOP_KQFILTER(fp->f_vnode, kn)); } /* * Simplified in-kernel wrapper calls for extended attribute access. * Both calls pass in a NULL credential, authorizing as "kernel" access. * Set IO_NODELOCKED in ioflg if the vnode is already locked. */ int vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, int *buflen, char *buf, struct thread *td) { struct uio auio; struct iovec iov; int error; iov.iov_len = *buflen; iov.iov_base = buf; auio.uio_iov = &iov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_READ; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = td; auio.uio_offset = 0; auio.uio_resid = *buflen; if ((ioflg & IO_NODELOCKED) == 0) vn_lock(vp, LK_SHARED | LK_RETRY); ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); /* authorize attribute retrieval as kernel */ error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL, td); if ((ioflg & IO_NODELOCKED) == 0) VOP_UNLOCK(vp, 0); if (error == 0) { *buflen = *buflen - auio.uio_resid; } return (error); } /* * XXX failure mode if partially written? */ int vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, int buflen, char *buf, struct thread *td) { struct uio auio; struct iovec iov; struct mount *mp; int error; iov.iov_len = buflen; iov.iov_base = buf; auio.uio_iov = &iov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_WRITE; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = td; auio.uio_offset = 0; auio.uio_resid = buflen; if ((ioflg & IO_NODELOCKED) == 0) { if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); /* authorize attribute setting as kernel */ error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td); if ((ioflg & IO_NODELOCKED) == 0) { vn_finished_write(mp); VOP_UNLOCK(vp, 0); } return (error); } int vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, struct thread *td) { struct mount *mp; int error; if ((ioflg & IO_NODELOCKED) == 0) { if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); /* authorize attribute removal as kernel */ error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td); if (error == EOPNOTSUPP) error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL, NULL, td); if ((ioflg & IO_NODELOCKED) == 0) { vn_finished_write(mp); VOP_UNLOCK(vp, 0); } return (error); } static int vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags, struct vnode **rvp) { return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp)); } int vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp) { return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino, lkflags, rvp)); } int vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg, int lkflags, struct vnode **rvp) { struct mount *mp; int ltype, error; ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get"); mp = vp->v_mount; ltype = VOP_ISLOCKED(vp); KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED, ("vn_vget_ino: vp not locked")); error = vfs_busy(mp, MBF_NOWAIT); if (error != 0) { vfs_ref(mp); VOP_UNLOCK(vp, 0); error = vfs_busy(mp, 0); vn_lock(vp, ltype | LK_RETRY); vfs_rel(mp); if (error != 0) return (ENOENT); if (vp->v_iflag & VI_DOOMED) { vfs_unbusy(mp); return (ENOENT); } } VOP_UNLOCK(vp, 0); error = alloc(mp, alloc_arg, lkflags, rvp); vfs_unbusy(mp); if (*rvp != vp) vn_lock(vp, ltype | LK_RETRY); if (vp->v_iflag & VI_DOOMED) { if (error == 0) { if (*rvp == vp) vunref(vp); else vput(*rvp); } error = ENOENT; } return (error); } int vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio, - const struct thread *td) + struct thread *td) { if (vp->v_type != VREG || td == NULL) return (0); - PROC_LOCK(td->td_proc); if ((uoff_t)uio->uio_offset + uio->uio_resid > - lim_cur(td->td_proc, RLIMIT_FSIZE)) { + lim_cur(td, RLIMIT_FSIZE)) { + PROC_LOCK(td->td_proc); kern_psignal(td->td_proc, SIGXFSZ); PROC_UNLOCK(td->td_proc); return (EFBIG); } - PROC_UNLOCK(td->td_proc); return (0); } int vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td) { struct vnode *vp; vp = fp->f_vnode; #ifdef AUDIT vn_lock(vp, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(vp); VOP_UNLOCK(vp, 0); #endif return (setfmode(td, active_cred, vp, mode)); } int vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, struct thread *td) { struct vnode *vp; vp = fp->f_vnode; #ifdef AUDIT vn_lock(vp, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(vp); VOP_UNLOCK(vp, 0); #endif return (setfown(td, active_cred, vp, uid, gid)); } void vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end) { vm_object_t object; if ((object = vp->v_object) == NULL) return; VM_OBJECT_WLOCK(object); vm_object_page_remove(object, start, end, 0); VM_OBJECT_WUNLOCK(object); } int vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred) { struct vattr va; daddr_t bn, bnp; uint64_t bsize; off_t noff; int error; KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA, ("Wrong command %lu", cmd)); if (vn_lock(vp, LK_SHARED) != 0) return (EBADF); if (vp->v_type != VREG) { error = ENOTTY; goto unlock; } error = VOP_GETATTR(vp, &va, cred); if (error != 0) goto unlock; noff = *off; if (noff >= va.va_size) { error = ENXIO; goto unlock; } bsize = vp->v_mount->mnt_stat.f_iosize; for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) { error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL); if (error == EOPNOTSUPP) { error = ENOTTY; goto unlock; } if ((bnp == -1 && cmd == FIOSEEKHOLE) || (bnp != -1 && cmd == FIOSEEKDATA)) { noff = bn * bsize; if (noff < *off) noff = *off; goto unlock; } } if (noff > va.va_size) noff = va.va_size; /* noff == va.va_size. There is an implicit hole at the end of file. */ if (cmd == FIOSEEKDATA) error = ENXIO; unlock: VOP_UNLOCK(vp, 0); if (error == 0) *off = noff; return (error); } int vn_seek(struct file *fp, off_t offset, int whence, struct thread *td) { struct ucred *cred; struct vnode *vp; struct vattr vattr; off_t foffset, size; int error, noneg; cred = td->td_ucred; vp = fp->f_vnode; foffset = foffset_lock(fp, 0); noneg = (vp->v_type != VCHR); error = 0; switch (whence) { case L_INCR: if (noneg && (foffset < 0 || (offset > 0 && foffset > OFF_MAX - offset))) { error = EOVERFLOW; break; } offset += foffset; break; case L_XTND: vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_GETATTR(vp, &vattr, cred); VOP_UNLOCK(vp, 0); if (error) break; /* * If the file references a disk device, then fetch * the media size and use that to determine the ending * offset. */ if (vattr.va_size == 0 && vp->v_type == VCHR && fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0) vattr.va_size = size; if (noneg && (vattr.va_size > OFF_MAX || (offset > 0 && vattr.va_size > OFF_MAX - offset))) { error = EOVERFLOW; break; } offset += vattr.va_size; break; case L_SET: break; case SEEK_DATA: error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td); break; case SEEK_HOLE: error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td); break; default: error = EINVAL; } if (error == 0 && noneg && offset < 0) error = EINVAL; if (error != 0) goto drop; VFS_KNOTE_UNLOCKED(vp, 0); td->td_uretoff.tdu_off = offset; drop: foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0); return (error); } int vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred, struct thread *td) { int error; /* * Grant permission if the caller is the owner of the file, or * the super-user, or has ACL_WRITE_ATTRIBUTES permission on * on the file. If the time pointer is null, then write * permission on the file is also sufficient. * * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes: * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES * will be allowed to set the times [..] to the current * server time. */ error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td); if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0) error = VOP_ACCESS(vp, VWRITE, cred, td); return (error); } int vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) { struct vnode *vp; int error; if (fp->f_type == DTYPE_FIFO) kif->kf_type = KF_TYPE_FIFO; else kif->kf_type = KF_TYPE_VNODE; vp = fp->f_vnode; vref(vp); FILEDESC_SUNLOCK(fdp); error = vn_fill_kinfo_vnode(vp, kif); vrele(vp); FILEDESC_SLOCK(fdp); return (error); } int vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif) { struct vattr va; char *fullpath, *freepath; int error; kif->kf_vnode_type = vntype_to_kinfo(vp->v_type); freepath = NULL; fullpath = "-"; error = vn_fullpath(curthread, vp, &fullpath, &freepath); if (error == 0) { strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); } if (freepath != NULL) free(freepath, M_TEMP); /* * Retrieve vnode attributes. */ va.va_fsid = VNOVAL; va.va_rdev = NODEV; vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_GETATTR(vp, &va, curthread->td_ucred); VOP_UNLOCK(vp, 0); if (error != 0) return (error); if (va.va_fsid != VNOVAL) kif->kf_un.kf_file.kf_file_fsid = va.va_fsid; else kif->kf_un.kf_file.kf_file_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; kif->kf_un.kf_file.kf_file_fileid = va.va_fileid; kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode); kif->kf_un.kf_file.kf_file_size = va.va_size; kif->kf_un.kf_file.kf_file_rdev = va.va_rdev; return (0); } int vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, struct thread *td) { #ifdef HWPMC_HOOKS struct pmckern_map_in pkm; #endif struct mount *mp; struct vnode *vp; vm_object_t object; vm_prot_t maxprot; boolean_t writecounted; int error; #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \ defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) /* * POSIX shared-memory objects are defined to have * kernel persistence, and are not defined to support * read(2)/write(2) -- or even open(2). Thus, we can * use MAP_ASYNC to trade on-disk coherence for speed. * The shm_open(3) library routine turns on the FPOSIXSHM * flag to request this behavior. */ if ((fp->f_flag & FPOSIXSHM) != 0) flags |= MAP_NOSYNC; #endif vp = fp->f_vnode; /* * Ensure that file and memory protections are * compatible. Note that we only worry about * writability if mapping is shared; in this case, * current and max prot are dictated by the open file. * XXX use the vnode instead? Problem is: what * credentials do we use for determination? What if * proc does a setuid? */ mp = vp->v_mount; if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) maxprot = VM_PROT_NONE; else maxprot = VM_PROT_EXECUTE; if ((fp->f_flag & FREAD) != 0) maxprot |= VM_PROT_READ; else if ((prot & VM_PROT_READ) != 0) return (EACCES); /* * If we are sharing potential changes via MAP_SHARED and we * are trying to get write permission although we opened it * without asking for it, bail out. */ if ((flags & MAP_SHARED) != 0) { if ((fp->f_flag & FWRITE) != 0) maxprot |= VM_PROT_WRITE; else if ((prot & VM_PROT_WRITE) != 0) return (EACCES); } else { maxprot |= VM_PROT_WRITE; cap_maxprot |= VM_PROT_WRITE; } maxprot &= cap_maxprot; writecounted = FALSE; error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, vp, &foff, &object, &writecounted); if (error != 0) return (error); error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, foff, writecounted, td); if (error != 0) { /* * If this mapping was accounted for in the vnode's * writecount, then undo that now. */ if (writecounted) vnode_pager_release_writecount(object, 0, size); vm_object_deallocate(object); } #ifdef HWPMC_HOOKS /* Inform hwpmc(4) if an executable is being mapped. */ if (error == 0 && (prot & VM_PROT_EXECUTE) != 0) { pkm.pm_file = vp; pkm.pm_address = (uintptr_t) addr; PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm); } #endif return (error); } Index: head/sys/ofed/drivers/infiniband/core/umem.c =================================================================== --- head/sys/ofed/drivers/infiniband/core/umem.c (revision 284214) +++ head/sys/ofed/drivers/infiniband/core/umem.c (revision 284215) @@ -1,442 +1,442 @@ /* * Copyright (c) 2005 Topspin Communications. All rights reserved. * Copyright (c) 2005 Cisco Systems. All rights reserved. * Copyright (c) 2005 Mellanox Technologies. All rights reserved. * * This software is available to you under a choice of one of two * licenses. You may choose to be licensed under the terms of the GNU * General Public License (GPL) Version 2, available from the file * COPYING in the main directory of this source tree, or the * OpenIB.org BSD license below: * * Redistribution and use in source and binary forms, with or * without modification, are permitted provided that the following * conditions are met: * * - Redistributions of source code must retain the above * copyright notice, this list of conditions and the following * disclaimer. * * - 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. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. */ #include #include #include #include #include #include #include #include #include #include #include "uverbs.h" #define IB_UMEM_MAX_PAGE_CHUNK (PAGE_SIZE / sizeof (struct page *)) static int allow_weak_ordering; module_param_named(weak_ordering, allow_weak_ordering, int, 0444); MODULE_PARM_DESC(weak_ordering, "Allow weak ordering for data registered memory"); static struct ib_umem *peer_umem_get(struct ib_peer_memory_client *ib_peer_mem, struct ib_umem *umem, unsigned long addr, int dmasync, int invalidation_supported) { int ret; const struct peer_memory_client *peer_mem = ib_peer_mem->peer_mem; struct invalidation_ctx *invalidation_ctx = NULL; umem->ib_peer_mem = ib_peer_mem; if (invalidation_supported) { invalidation_ctx = kzalloc(sizeof(*invalidation_ctx), GFP_KERNEL); if (!invalidation_ctx) { ret = -ENOMEM; goto out; } umem->invalidation_ctx = invalidation_ctx; invalidation_ctx->umem = umem; mutex_lock(&ib_peer_mem->lock); invalidation_ctx->context_ticket = ib_peer_insert_context(ib_peer_mem, invalidation_ctx); /* unlock before calling get pages to prevent a dead-lock from the callback */ mutex_unlock(&ib_peer_mem->lock); } ret = peer_mem->get_pages(addr, umem->length, umem->writable, 1, &umem->sg_head, umem->peer_mem_client_context, invalidation_ctx ? (void *)invalidation_ctx->context_ticket : NULL); if (invalidation_ctx) { /* taking the lock back, checking that wasn't invalidated at that time */ mutex_lock(&ib_peer_mem->lock); if (invalidation_ctx->peer_invalidated) { printk(KERN_ERR "peer_umem_get: pages were invalidated by peer\n"); ret = -EINVAL; } } if (ret) goto out; umem->page_size = peer_mem->get_page_size (umem->peer_mem_client_context); if (umem->page_size <= 0) goto put_pages; umem->offset = addr & ((unsigned long)umem->page_size - 1); ret = peer_mem->dma_map(&umem->sg_head, umem->peer_mem_client_context, umem->context->device->dma_device, dmasync, &umem->nmap); if (ret) goto put_pages; ib_peer_mem->stats.num_reg_pages += umem->nmap * (umem->page_size >> PAGE_SHIFT); ib_peer_mem->stats.num_alloc_mrs += 1; return umem; put_pages: peer_mem->put_pages(umem->peer_mem_client_context, &umem->sg_head); out: if (invalidation_ctx) { ib_peer_remove_context(ib_peer_mem, invalidation_ctx->context_ticket); mutex_unlock(&umem->ib_peer_mem->lock); kfree(invalidation_ctx); } ib_put_peer_client(ib_peer_mem, umem->peer_mem_client_context, umem->peer_mem_srcu_key); kfree(umem); return ERR_PTR(ret); } static void peer_umem_release(struct ib_umem *umem) { struct ib_peer_memory_client *ib_peer_mem = umem->ib_peer_mem; const struct peer_memory_client *peer_mem = ib_peer_mem->peer_mem; struct invalidation_ctx *invalidation_ctx = umem->invalidation_ctx; if (invalidation_ctx) { int peer_callback; int inflight_invalidation; /* If we are not under peer callback we must take the lock before removing * core ticket from the tree and releasing its umem. * It will let any inflight callbacks to be ended safely. * If we are under peer callback or under error flow of reg_mr so that context * wasn't activated yet lock was already taken. */ if (invalidation_ctx->func && !invalidation_ctx->peer_callback) mutex_lock(&ib_peer_mem->lock); ib_peer_remove_context(ib_peer_mem, invalidation_ctx->context_ticket); /* make sure to check inflight flag after took the lock and remove from tree. * in addition, from that point using local variables for peer_callback and * inflight_invalidation as after the complete invalidation_ctx can't be accessed * any more as it may be freed by the callback. */ peer_callback = invalidation_ctx->peer_callback; inflight_invalidation = invalidation_ctx->inflight_invalidation; if (inflight_invalidation) complete(&invalidation_ctx->comp); /* On peer callback lock is handled externally */ if (!peer_callback) /* unlocking before put_pages */ mutex_unlock(&ib_peer_mem->lock); /* in case under callback context or callback is pending let it free the invalidation context */ if (!peer_callback && !inflight_invalidation) kfree(invalidation_ctx); } peer_mem->dma_unmap(&umem->sg_head, umem->peer_mem_client_context, umem->context->device->dma_device); peer_mem->put_pages(&umem->sg_head, umem->peer_mem_client_context); ib_peer_mem->stats.num_dereg_pages += umem->nmap * (umem->page_size >> PAGE_SHIFT); ib_peer_mem->stats.num_dealloc_mrs += 1; ib_put_peer_client(ib_peer_mem, umem->peer_mem_client_context, umem->peer_mem_srcu_key); kfree(umem); return; } static void __ib_umem_release(struct ib_device *dev, struct ib_umem *umem, int dirty) { vm_object_t object; struct scatterlist *sg; struct page *page; int i; object = NULL; if (umem->nmap > 0) ib_dma_unmap_sg(dev, umem->sg_head.sgl, umem->nmap, DMA_BIDIRECTIONAL); for_each_sg(umem->sg_head.sgl, sg, umem->npages, i) { page = sg_page(sg); if (umem->writable && dirty) { if (object && object != page->object) VM_OBJECT_WUNLOCK(object); if (object != page->object) { object = page->object; VM_OBJECT_WLOCK(object); } vm_page_dirty(page); } } sg_free_table(&umem->sg_head); if (object) VM_OBJECT_WUNLOCK(object); } void ib_umem_activate_invalidation_notifier(struct ib_umem *umem, umem_invalidate_func_t func, void *cookie) { struct invalidation_ctx *invalidation_ctx = umem->invalidation_ctx; invalidation_ctx->func = func; invalidation_ctx->cookie = cookie; /* from that point any pending invalidations can be called */ mutex_unlock(&umem->ib_peer_mem->lock); return; } EXPORT_SYMBOL(ib_umem_activate_invalidation_notifier); /** * ib_umem_get - Pin and DMA map userspace memory. * @context: userspace context to pin memory for * @addr: userspace virtual address to start at * @size: length of region to pin * @access: IB_ACCESS_xxx flags for memory being pinned * @dmasync: flush in-flight DMA when the memory region is written */ struct ib_umem *ib_umem_get_ex(struct ib_ucontext *context, unsigned long addr, size_t size, int access, int dmasync, int invalidation_supported) { struct ib_umem *umem; struct proc *proc; pmap_t pmap; vm_offset_t end, last, start; vm_size_t npages; int error; int ret; int ents; int i; DEFINE_DMA_ATTRS(attrs); struct scatterlist *sg, *sg_list_start; int need_release = 0; error = priv_check(curthread, PRIV_VM_MLOCK); if (error) return ERR_PTR(-error); last = addr + size; start = addr & PAGE_MASK; /* Use the linux PAGE_MASK definition. */ end = roundup2(last, PAGE_SIZE); /* Use PAGE_MASK safe operation. */ if (last < addr || end < addr) return ERR_PTR(-EINVAL); npages = atop(end - start); if (npages > vm_page_max_wired) return ERR_PTR(-ENOMEM); umem = kzalloc(sizeof *umem, GFP_KERNEL); if (!umem) return ERR_PTR(-ENOMEM); proc = curthread->td_proc; PROC_LOCK(proc); if (ptoa(npages + pmap_wired_count(vm_map_pmap(&proc->p_vmspace->vm_map))) > - lim_cur(proc, RLIMIT_MEMLOCK)) { + lim_cur_proc(proc, RLIMIT_MEMLOCK)) { PROC_UNLOCK(proc); kfree(umem); return ERR_PTR(-ENOMEM); } PROC_UNLOCK(proc); if (npages + vm_cnt.v_wire_count > vm_page_max_wired) { kfree(umem); return ERR_PTR(-EAGAIN); } error = vm_map_wire(&proc->p_vmspace->vm_map, start, end, VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES | (umem->writable ? VM_MAP_WIRE_WRITE : 0)); if (error != KERN_SUCCESS) { kfree(umem); return ERR_PTR(-ENOMEM); } umem->context = context; umem->length = size; umem->offset = addr & ~PAGE_MASK; umem->page_size = PAGE_SIZE; umem->start = addr; /* * We ask for writable memory if any access flags other than * "remote read" are set. "Local write" and "remote write" * obviously require write access. "Remote atomic" can do * things like fetch and add, which will modify memory, and * "MW bind" can change permissions by binding a window. */ umem->writable = !!(access & ~IB_ACCESS_REMOTE_READ); if (invalidation_supported || context->peer_mem_private_data) { struct ib_peer_memory_client *peer_mem_client; peer_mem_client = ib_get_peer_client(context, addr, size, &umem->peer_mem_client_context, &umem->peer_mem_srcu_key); if (peer_mem_client) return peer_umem_get(peer_mem_client, umem, addr, dmasync, invalidation_supported); } umem->hugetlb = 0; pmap = vm_map_pmap(&proc->p_vmspace->vm_map); if (npages == 0) { ret = -EINVAL; goto out; } ret = sg_alloc_table(&umem->sg_head, npages, GFP_KERNEL); if (ret) goto out; need_release = 1; sg_list_start = umem->sg_head.sgl; while (npages) { ents = min_t(int, npages, IB_UMEM_MAX_PAGE_CHUNK); umem->npages += ents; for_each_sg(sg_list_start, sg, ents, i) { vm_paddr_t pa; pa = pmap_extract(pmap, start); if (pa == 0) { ret = -ENOMEM; goto out; } sg_set_page(sg, PHYS_TO_VM_PAGE(pa), PAGE_SIZE, 0); npages--; start += PAGE_SIZE; } /* preparing for next loop */ sg_list_start = sg; } umem->nmap = ib_dma_map_sg_attrs(context->device, umem->sg_head.sgl, umem->npages, DMA_BIDIRECTIONAL, &attrs); if (umem->nmap != umem->npages) { ret = -ENOMEM; goto out; } out: if (ret < 0) { if (need_release) __ib_umem_release(context->device, umem, 0); kfree(umem); } return ret < 0 ? ERR_PTR(ret) : umem; } EXPORT_SYMBOL(ib_umem_get_ex); struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr, size_t size, int access, int dmasync) { return ib_umem_get_ex(context, addr, size, access, dmasync, 0); } EXPORT_SYMBOL(ib_umem_get); /** * ib_umem_release - release memory pinned with ib_umem_get * @umem: umem struct to release */ void ib_umem_release(struct ib_umem *umem) { vm_offset_t addr, end, last, start; vm_size_t size; int error; if (umem->ib_peer_mem) { peer_umem_release(umem); return; } __ib_umem_release(umem->context->device, umem, 1); if (umem->context->closing) { kfree(umem); return; } error = priv_check(curthread, PRIV_VM_MUNLOCK); if (error) return; addr = umem->start; size = umem->length; last = addr + size; start = addr & PAGE_MASK; /* Use the linux PAGE_MASK definition. */ end = roundup2(last, PAGE_SIZE); /* Use PAGE_MASK safe operation. */ vm_map_unwire(&curthread->td_proc->p_vmspace->vm_map, start, end, VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); kfree(umem); } EXPORT_SYMBOL(ib_umem_release); int ib_umem_page_count(struct ib_umem *umem) { int shift; int i; int n; struct scatterlist *sg; shift = ilog2(umem->page_size); n = 0; for_each_sg(umem->sg_head.sgl, sg, umem->nmap, i) n += sg_dma_len(sg) >> shift; return n; } EXPORT_SYMBOL(ib_umem_page_count); Index: head/sys/ofed/drivers/infiniband/hw/mthca/mthca_memfree.c =================================================================== --- head/sys/ofed/drivers/infiniband/hw/mthca/mthca_memfree.c (revision 284214) +++ head/sys/ofed/drivers/infiniband/hw/mthca/mthca_memfree.c (revision 284215) @@ -1,880 +1,881 @@ /* * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved. * Copyright (c) 2005 Cisco Systems. All rights reserved. * Copyright (c) 2005 Mellanox Technologies. All rights reserved. * * This software is available to you under a choice of one of two * licenses. You may choose to be licensed under the terms of the GNU * General Public License (GPL) Version 2, available from the file * COPYING in the main directory of this source tree, or the * OpenIB.org BSD license below: * * Redistribution and use in source and binary forms, with or * without modification, are permitted provided that the following * conditions are met: * * - Redistributions of source code must retain the above * copyright notice, this list of conditions and the following * disclaimer. * * - 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. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. */ #include #include #include #include #include "mthca_memfree.h" #include "mthca_dev.h" #include "mthca_cmd.h" /* * We allocate in as big chunks as we can, up to a maximum of 256 KB * per chunk. */ enum { MTHCA_ICM_ALLOC_SIZE = 1 << 18, MTHCA_TABLE_CHUNK_SIZE = 1 << 18 }; struct mthca_user_db_table { struct mutex mutex; struct { u64 uvirt; struct scatterlist mem; int refcount; } page[0]; }; static void mthca_free_icm_pages(struct mthca_dev *dev, struct mthca_icm_chunk *chunk) { int i; if (chunk->nsg > 0) pci_unmap_sg(dev->pdev, chunk->mem, chunk->npages, PCI_DMA_BIDIRECTIONAL); for (i = 0; i < chunk->npages; ++i) __free_pages(sg_page(&chunk->mem[i]), get_order(chunk->mem[i].length)); } static void mthca_free_icm_coherent(struct mthca_dev *dev, struct mthca_icm_chunk *chunk) { int i; for (i = 0; i < chunk->npages; ++i) { dma_free_coherent(&dev->pdev->dev, chunk->mem[i].length, lowmem_page_address(sg_page(&chunk->mem[i])), sg_dma_address(&chunk->mem[i])); } } void mthca_free_icm(struct mthca_dev *dev, struct mthca_icm *icm, int coherent) { struct mthca_icm_chunk *chunk, *tmp; if (!icm) return; list_for_each_entry_safe(chunk, tmp, &icm->chunk_list, list) { if (coherent) mthca_free_icm_coherent(dev, chunk); else mthca_free_icm_pages(dev, chunk); kfree(chunk); } kfree(icm); } static int mthca_alloc_icm_pages(struct scatterlist *mem, int order, gfp_t gfp_mask) { struct page *page; /* * Use __GFP_ZERO because buggy firmware assumes ICM pages are * cleared, and subtle failures are seen if they aren't. */ page = alloc_pages(gfp_mask | __GFP_ZERO, order); if (!page) return -ENOMEM; sg_set_page(mem, page, PAGE_SIZE << order, 0); return 0; } static int mthca_alloc_icm_coherent(struct device *dev, struct scatterlist *mem, int order, gfp_t gfp_mask) { void *buf = dma_alloc_coherent(dev, PAGE_SIZE << order, &sg_dma_address(mem), gfp_mask); if (!buf) return -ENOMEM; sg_set_buf(mem, buf, PAGE_SIZE << order); BUG_ON(mem->offset); sg_dma_len(mem) = PAGE_SIZE << order; return 0; } struct mthca_icm *mthca_alloc_icm(struct mthca_dev *dev, int npages, gfp_t gfp_mask, int coherent) { struct mthca_icm *icm; struct mthca_icm_chunk *chunk = NULL; int cur_order; int ret; /* We use sg_set_buf for coherent allocs, which assumes low memory */ BUG_ON(coherent && (gfp_mask & __GFP_HIGHMEM)); icm = kmalloc(sizeof *icm, gfp_mask & ~(__GFP_HIGHMEM | __GFP_NOWARN)); if (!icm) return icm; icm->refcount = 0; INIT_LIST_HEAD(&icm->chunk_list); cur_order = get_order(MTHCA_ICM_ALLOC_SIZE); while (npages > 0) { if (!chunk) { chunk = kmalloc(sizeof *chunk, gfp_mask & ~(__GFP_HIGHMEM | __GFP_NOWARN)); if (!chunk) goto fail; sg_init_table(chunk->mem, MTHCA_ICM_CHUNK_LEN); chunk->npages = 0; chunk->nsg = 0; list_add_tail(&chunk->list, &icm->chunk_list); } while (1 << cur_order > npages) --cur_order; if (coherent) ret = mthca_alloc_icm_coherent(&dev->pdev->dev, &chunk->mem[chunk->npages], cur_order, gfp_mask); else ret = mthca_alloc_icm_pages(&chunk->mem[chunk->npages], cur_order, gfp_mask); if (!ret) { ++chunk->npages; if (coherent) ++chunk->nsg; else if (chunk->npages == MTHCA_ICM_CHUNK_LEN) { chunk->nsg = pci_map_sg(dev->pdev, chunk->mem, chunk->npages, PCI_DMA_BIDIRECTIONAL); if (chunk->nsg <= 0) goto fail; } if (chunk->npages == MTHCA_ICM_CHUNK_LEN) chunk = NULL; npages -= 1 << cur_order; } else { --cur_order; if (cur_order < 0) goto fail; } } if (!coherent && chunk) { chunk->nsg = pci_map_sg(dev->pdev, chunk->mem, chunk->npages, PCI_DMA_BIDIRECTIONAL); if (chunk->nsg <= 0) goto fail; } return icm; fail: mthca_free_icm(dev, icm, coherent); return NULL; } int mthca_table_get(struct mthca_dev *dev, struct mthca_icm_table *table, int obj) { int i = (obj & (table->num_obj - 1)) * table->obj_size / MTHCA_TABLE_CHUNK_SIZE; int ret = 0; u8 status; mutex_lock(&table->mutex); if (table->icm[i]) { ++table->icm[i]->refcount; goto out; } table->icm[i] = mthca_alloc_icm(dev, MTHCA_TABLE_CHUNK_SIZE >> PAGE_SHIFT, (table->lowmem ? GFP_KERNEL : GFP_HIGHUSER) | __GFP_NOWARN, table->coherent); if (!table->icm[i]) { ret = -ENOMEM; goto out; } if (mthca_MAP_ICM(dev, table->icm[i], table->virt + i * MTHCA_TABLE_CHUNK_SIZE, &status) || status) { mthca_free_icm(dev, table->icm[i], table->coherent); table->icm[i] = NULL; ret = -ENOMEM; goto out; } ++table->icm[i]->refcount; out: mutex_unlock(&table->mutex); return ret; } void mthca_table_put(struct mthca_dev *dev, struct mthca_icm_table *table, int obj) { int i; u8 status; if (!mthca_is_memfree(dev)) return; i = (obj & (table->num_obj - 1)) * table->obj_size / MTHCA_TABLE_CHUNK_SIZE; mutex_lock(&table->mutex); if (--table->icm[i]->refcount == 0) { mthca_UNMAP_ICM(dev, table->virt + i * MTHCA_TABLE_CHUNK_SIZE, MTHCA_TABLE_CHUNK_SIZE / MTHCA_ICM_PAGE_SIZE, &status); mthca_free_icm(dev, table->icm[i], table->coherent); table->icm[i] = NULL; } mutex_unlock(&table->mutex); } void *mthca_table_find(struct mthca_icm_table *table, int obj, dma_addr_t *dma_handle) { int idx, offset, dma_offset, i; struct mthca_icm_chunk *chunk; struct mthca_icm *icm; struct page *page = NULL; if (!table->lowmem) return NULL; mutex_lock(&table->mutex); idx = (obj & (table->num_obj - 1)) * table->obj_size; icm = table->icm[idx / MTHCA_TABLE_CHUNK_SIZE]; dma_offset = offset = idx % MTHCA_TABLE_CHUNK_SIZE; if (!icm) goto out; list_for_each_entry(chunk, &icm->chunk_list, list) { for (i = 0; i < chunk->npages; ++i) { if (dma_handle && dma_offset >= 0) { if (sg_dma_len(&chunk->mem[i]) > dma_offset) *dma_handle = sg_dma_address(&chunk->mem[i]) + dma_offset; dma_offset -= sg_dma_len(&chunk->mem[i]); } /* DMA mapping can merge pages but not split them, * so if we found the page, dma_handle has already * been assigned to. */ if (chunk->mem[i].length > offset) { page = sg_page(&chunk->mem[i]); goto out; } offset -= chunk->mem[i].length; } } out: mutex_unlock(&table->mutex); return page ? lowmem_page_address(page) + offset : NULL; } int mthca_table_get_range(struct mthca_dev *dev, struct mthca_icm_table *table, int start, int end) { int inc = MTHCA_TABLE_CHUNK_SIZE / table->obj_size; int i, err; for (i = start; i <= end; i += inc) { err = mthca_table_get(dev, table, i); if (err) goto fail; } return 0; fail: while (i > start) { i -= inc; mthca_table_put(dev, table, i); } return err; } void mthca_table_put_range(struct mthca_dev *dev, struct mthca_icm_table *table, int start, int end) { int i; if (!mthca_is_memfree(dev)) return; for (i = start; i <= end; i += MTHCA_TABLE_CHUNK_SIZE / table->obj_size) mthca_table_put(dev, table, i); } struct mthca_icm_table *mthca_alloc_icm_table(struct mthca_dev *dev, u64 virt, int obj_size, int nobj, int reserved, int use_lowmem, int use_coherent) { struct mthca_icm_table *table; int obj_per_chunk; int num_icm; unsigned chunk_size; int i; u8 status; obj_per_chunk = MTHCA_TABLE_CHUNK_SIZE / obj_size; num_icm = DIV_ROUND_UP(nobj, obj_per_chunk); table = kmalloc(sizeof *table + num_icm * sizeof *table->icm, GFP_KERNEL); if (!table) return NULL; table->virt = virt; table->num_icm = num_icm; table->num_obj = nobj; table->obj_size = obj_size; table->lowmem = use_lowmem; table->coherent = use_coherent; mutex_init(&table->mutex); for (i = 0; i < num_icm; ++i) table->icm[i] = NULL; for (i = 0; i * MTHCA_TABLE_CHUNK_SIZE < reserved * obj_size; ++i) { chunk_size = MTHCA_TABLE_CHUNK_SIZE; if ((i + 1) * MTHCA_TABLE_CHUNK_SIZE > nobj * obj_size) chunk_size = nobj * obj_size - i * MTHCA_TABLE_CHUNK_SIZE; table->icm[i] = mthca_alloc_icm(dev, chunk_size >> PAGE_SHIFT, (use_lowmem ? GFP_KERNEL : GFP_HIGHUSER) | __GFP_NOWARN, use_coherent); if (!table->icm[i]) goto err; if (mthca_MAP_ICM(dev, table->icm[i], virt + i * MTHCA_TABLE_CHUNK_SIZE, &status) || status) { mthca_free_icm(dev, table->icm[i], table->coherent); table->icm[i] = NULL; goto err; } /* * Add a reference to this ICM chunk so that it never * gets freed (since it contains reserved firmware objects). */ ++table->icm[i]->refcount; } return table; err: for (i = 0; i < num_icm; ++i) if (table->icm[i]) { mthca_UNMAP_ICM(dev, virt + i * MTHCA_TABLE_CHUNK_SIZE, MTHCA_TABLE_CHUNK_SIZE / MTHCA_ICM_PAGE_SIZE, &status); mthca_free_icm(dev, table->icm[i], table->coherent); } kfree(table); return NULL; } void mthca_free_icm_table(struct mthca_dev *dev, struct mthca_icm_table *table) { int i; u8 status; for (i = 0; i < table->num_icm; ++i) if (table->icm[i]) { mthca_UNMAP_ICM(dev, table->virt + i * MTHCA_TABLE_CHUNK_SIZE, MTHCA_TABLE_CHUNK_SIZE / MTHCA_ICM_PAGE_SIZE, &status); mthca_free_icm(dev, table->icm[i], table->coherent); } kfree(table); } static u64 mthca_uarc_virt(struct mthca_dev *dev, struct mthca_uar *uar, int page) { return dev->uar_table.uarc_base + uar->index * dev->uar_table.uarc_size + page * MTHCA_ICM_PAGE_SIZE; } #include #include #include #include #include int mthca_map_user_db(struct mthca_dev *dev, struct mthca_uar *uar, struct mthca_user_db_table *db_tab, int index, u64 uaddr) { #ifdef __linux__ struct page *pages[1]; int ret = 0; u8 status; int i; if (!mthca_is_memfree(dev)) return 0; if (index < 0 || index > dev->uar_table.uarc_size / 8) return -EINVAL; mutex_lock(&db_tab->mutex); i = index / MTHCA_DB_REC_PER_PAGE; if ((db_tab->page[i].refcount >= MTHCA_DB_REC_PER_PAGE) || (db_tab->page[i].uvirt && db_tab->page[i].uvirt != uaddr) || (uaddr & 4095)) { ret = -EINVAL; goto out; } if (db_tab->page[i].refcount) { ++db_tab->page[i].refcount; goto out; } ret = get_user_pages(current, current->mm, uaddr & PAGE_MASK, 1, 1, 0, pages, NULL); if (ret < 0) goto out; sg_set_page(&db_tab->page[i].mem, pages[0], MTHCA_ICM_PAGE_SIZE, uaddr & ~PAGE_MASK); ret = pci_map_sg(dev->pdev, &db_tab->page[i].mem, 1, PCI_DMA_TODEVICE); if (ret < 0) { put_page(pages[0]); goto out; } ret = mthca_MAP_ICM_page(dev, sg_dma_address(&db_tab->page[i].mem), mthca_uarc_virt(dev, uar, i), &status); if (!ret && status) ret = -EINVAL; if (ret) { pci_unmap_sg(dev->pdev, &db_tab->page[i].mem, 1, PCI_DMA_TODEVICE); put_page(sg_page(&db_tab->page[i].mem)); goto out; } db_tab->page[i].uvirt = uaddr; db_tab->page[i].refcount = 1; out: mutex_unlock(&db_tab->mutex); return ret; #else struct proc *proc; vm_offset_t start; vm_paddr_t paddr; pmap_t pmap; vm_page_t m; int ret = 0; u8 status; int i; if (!mthca_is_memfree(dev)) return 0; if (index < 0 || index > dev->uar_table.uarc_size / 8) return -EINVAL; mutex_lock(&db_tab->mutex); i = index / MTHCA_DB_REC_PER_PAGE; start = 0; if ((db_tab->page[i].refcount >= MTHCA_DB_REC_PER_PAGE) || (db_tab->page[i].uvirt && db_tab->page[i].uvirt != uaddr) || (uaddr & 4095)) { ret = -EINVAL; goto out; } if (db_tab->page[i].refcount) { ++db_tab->page[i].refcount; goto out; } proc = curproc; pmap = vm_map_pmap(&proc->p_vmspace->vm_map); PROC_LOCK(proc); - if (ptoa(pmap_wired_count(pmap) + 1) > lim_cur(proc, RLIMIT_MEMLOCK)) { + if (ptoa(pmap_wired_count(pmap) + 1) > + lim_cur_proc(proc, RLIMIT_MEMLOCK)) { PROC_UNLOCK(proc); ret = -ENOMEM; goto out; } PROC_UNLOCK(proc); if (vm_cnt.v_wire_count + 1 > vm_page_max_wired) { ret = -EAGAIN; goto out; } start = uaddr & PAGE_MASK; ret = vm_map_wire(&proc->p_vmspace->vm_map, start, start + PAGE_SIZE, VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES | VM_MAP_WIRE_WRITE); if (ret != KERN_SUCCESS) { start = 0; ret = -ENOMEM; goto out; } paddr = pmap_extract(pmap, uaddr); if (paddr == 0) { ret = -EFAULT; goto out; } m = PHYS_TO_VM_PAGE(paddr); sg_set_page(&db_tab->page[i].mem, m, MTHCA_ICM_PAGE_SIZE, uaddr & ~PAGE_MASK); ret = pci_map_sg(dev->pdev, &db_tab->page[i].mem, 1, PCI_DMA_TODEVICE); if (ret < 0) goto out; ret = mthca_MAP_ICM_page(dev, sg_dma_address(&db_tab->page[i].mem), mthca_uarc_virt(dev, uar, i), &status); if (!ret && status) ret = -EINVAL; if (ret) { pci_unmap_sg(dev->pdev, &db_tab->page[i].mem, 1, PCI_DMA_TODEVICE); goto out; } db_tab->page[i].uvirt = uaddr; db_tab->page[i].refcount = 1; out: if (ret < 0 && start) vm_map_unwire(&curthread->td_proc->p_vmspace->vm_map, start, start + PAGE_SIZE, VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); mutex_unlock(&db_tab->mutex); return ret; #endif } void mthca_unmap_user_db(struct mthca_dev *dev, struct mthca_uar *uar, struct mthca_user_db_table *db_tab, int index) { if (!mthca_is_memfree(dev)) return; /* * To make our bookkeeping simpler, we don't unmap DB * pages until we clean up the whole db table. */ mutex_lock(&db_tab->mutex); --db_tab->page[index / MTHCA_DB_REC_PER_PAGE].refcount; mutex_unlock(&db_tab->mutex); } struct mthca_user_db_table *mthca_init_user_db_tab(struct mthca_dev *dev) { struct mthca_user_db_table *db_tab; int npages; int i; if (!mthca_is_memfree(dev)) return NULL; npages = dev->uar_table.uarc_size / MTHCA_ICM_PAGE_SIZE; db_tab = kmalloc(sizeof *db_tab + npages * sizeof *db_tab->page, GFP_KERNEL); if (!db_tab) return ERR_PTR(-ENOMEM); mutex_init(&db_tab->mutex); for (i = 0; i < npages; ++i) { db_tab->page[i].refcount = 0; db_tab->page[i].uvirt = 0; sg_init_table(&db_tab->page[i].mem, 1); } return db_tab; } void mthca_cleanup_user_db_tab(struct mthca_dev *dev, struct mthca_uar *uar, struct mthca_user_db_table *db_tab) { int i; u8 status; if (!mthca_is_memfree(dev)) return; for (i = 0; i < dev->uar_table.uarc_size / MTHCA_ICM_PAGE_SIZE; ++i) { if (db_tab->page[i].uvirt) { mthca_UNMAP_ICM(dev, mthca_uarc_virt(dev, uar, i), 1, &status); pci_unmap_sg(dev->pdev, &db_tab->page[i].mem, 1, PCI_DMA_TODEVICE); #ifdef __linux__ put_page(sg_page(&db_tab->page[i].mem)); #else vm_offset_t start; start = db_tab->page[i].uvirt & PAGE_MASK; vm_map_unwire(&curthread->td_proc->p_vmspace->vm_map, start, start + PAGE_SIZE, VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); #endif } } kfree(db_tab); } int mthca_alloc_db(struct mthca_dev *dev, enum mthca_db_type type, u32 qn, __be32 **db) { int group; int start, end, dir; int i, j; struct mthca_db_page *page; int ret = 0; u8 status; mutex_lock(&dev->db_tab->mutex); switch (type) { case MTHCA_DB_TYPE_CQ_ARM: case MTHCA_DB_TYPE_SQ: group = 0; start = 0; end = dev->db_tab->max_group1; dir = 1; break; case MTHCA_DB_TYPE_CQ_SET_CI: case MTHCA_DB_TYPE_RQ: case MTHCA_DB_TYPE_SRQ: group = 1; start = dev->db_tab->npages - 1; end = dev->db_tab->min_group2; dir = -1; break; default: ret = -EINVAL; goto out; } for (i = start; i != end; i += dir) if (dev->db_tab->page[i].db_rec && !bitmap_full(dev->db_tab->page[i].used, MTHCA_DB_REC_PER_PAGE)) { page = dev->db_tab->page + i; goto found; } for (i = start; i != end; i += dir) if (!dev->db_tab->page[i].db_rec) { page = dev->db_tab->page + i; goto alloc; } if (dev->db_tab->max_group1 >= dev->db_tab->min_group2 - 1) { ret = -ENOMEM; goto out; } if (group == 0) ++dev->db_tab->max_group1; else --dev->db_tab->min_group2; page = dev->db_tab->page + end; alloc: page->db_rec = dma_alloc_coherent(&dev->pdev->dev, MTHCA_ICM_PAGE_SIZE, &page->mapping, GFP_KERNEL); if (!page->db_rec) { ret = -ENOMEM; goto out; } memset(page->db_rec, 0, MTHCA_ICM_PAGE_SIZE); ret = mthca_MAP_ICM_page(dev, page->mapping, mthca_uarc_virt(dev, &dev->driver_uar, i), &status); if (!ret && status) ret = -EINVAL; if (ret) { dma_free_coherent(&dev->pdev->dev, MTHCA_ICM_PAGE_SIZE, page->db_rec, page->mapping); goto out; } bitmap_zero(page->used, MTHCA_DB_REC_PER_PAGE); found: j = find_first_zero_bit(page->used, MTHCA_DB_REC_PER_PAGE); set_bit(j, page->used); if (group == 1) j = MTHCA_DB_REC_PER_PAGE - 1 - j; ret = i * MTHCA_DB_REC_PER_PAGE + j; page->db_rec[j] = cpu_to_be64((qn << 8) | (type << 5)); *db = (__be32 *) &page->db_rec[j]; out: mutex_unlock(&dev->db_tab->mutex); return ret; } void mthca_free_db(struct mthca_dev *dev, int type, int db_index) { int i, j; struct mthca_db_page *page; u8 status; i = db_index / MTHCA_DB_REC_PER_PAGE; j = db_index % MTHCA_DB_REC_PER_PAGE; page = dev->db_tab->page + i; mutex_lock(&dev->db_tab->mutex); page->db_rec[j] = 0; if (i >= dev->db_tab->min_group2) j = MTHCA_DB_REC_PER_PAGE - 1 - j; clear_bit(j, page->used); if (bitmap_empty(page->used, MTHCA_DB_REC_PER_PAGE) && i >= dev->db_tab->max_group1 - 1) { mthca_UNMAP_ICM(dev, mthca_uarc_virt(dev, &dev->driver_uar, i), 1, &status); dma_free_coherent(&dev->pdev->dev, MTHCA_ICM_PAGE_SIZE, page->db_rec, page->mapping); page->db_rec = NULL; if (i == dev->db_tab->max_group1) { --dev->db_tab->max_group1; /* XXX may be able to unmap more pages now */ } if (i == dev->db_tab->min_group2) ++dev->db_tab->min_group2; } mutex_unlock(&dev->db_tab->mutex); } int mthca_init_db_tab(struct mthca_dev *dev) { int i; if (!mthca_is_memfree(dev)) return 0; dev->db_tab = kmalloc(sizeof *dev->db_tab, GFP_KERNEL); if (!dev->db_tab) return -ENOMEM; mutex_init(&dev->db_tab->mutex); dev->db_tab->npages = dev->uar_table.uarc_size / MTHCA_ICM_PAGE_SIZE; dev->db_tab->max_group1 = 0; dev->db_tab->min_group2 = dev->db_tab->npages - 1; dev->db_tab->page = kmalloc(dev->db_tab->npages * sizeof *dev->db_tab->page, GFP_KERNEL); if (!dev->db_tab->page) { kfree(dev->db_tab); return -ENOMEM; } for (i = 0; i < dev->db_tab->npages; ++i) dev->db_tab->page[i].db_rec = NULL; return 0; } void mthca_cleanup_db_tab(struct mthca_dev *dev) { int i; u8 status; if (!mthca_is_memfree(dev)) return; /* * Because we don't always free our UARC pages when they * become empty to make mthca_free_db() simpler we need to * make a sweep through the doorbell pages and free any * leftover pages now. */ for (i = 0; i < dev->db_tab->npages; ++i) { if (!dev->db_tab->page[i].db_rec) continue; if (!bitmap_empty(dev->db_tab->page[i].used, MTHCA_DB_REC_PER_PAGE)) mthca_warn(dev, "Kernel UARC page %d not empty\n", i); mthca_UNMAP_ICM(dev, mthca_uarc_virt(dev, &dev->driver_uar, i), 1, &status); dma_free_coherent(&dev->pdev->dev, MTHCA_ICM_PAGE_SIZE, dev->db_tab->page[i].db_rec, dev->db_tab->page[i].mapping); } kfree(dev->db_tab->page); kfree(dev->db_tab); } Index: head/sys/sys/proc.h =================================================================== --- head/sys/sys/proc.h (revision 284214) +++ head/sys/sys/proc.h (revision 284215) @@ -1,1035 +1,1036 @@ /*- * Copyright (c) 1986, 1989, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * 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. * * @(#)proc.h 8.15 (Berkeley) 5/19/95 * $FreeBSD$ */ #ifndef _SYS_PROC_H_ #define _SYS_PROC_H_ #include /* For struct callout. */ #include /* For struct klist. */ #include #ifndef _KERNEL #include #endif #include #include #include #include #include #include #include /* XXX. */ #include #include #include #include #include #ifndef _KERNEL #include /* For structs itimerval, timeval. */ #else #include #endif #include #include #include /* Machine-dependent proc substruct. */ /* * One structure allocated per session. * * List of locks * (m) locked by s_mtx mtx * (e) locked by proctree_lock sx * (c) const until freeing */ struct session { u_int s_count; /* Ref cnt; pgrps in session - atomic. */ struct proc *s_leader; /* (m + e) Session leader. */ struct vnode *s_ttyvp; /* (m) Vnode of controlling tty. */ struct cdev_priv *s_ttydp; /* (m) Device of controlling tty. */ struct tty *s_ttyp; /* (e) Controlling tty. */ pid_t s_sid; /* (c) Session ID. */ /* (m) Setlogin() name: */ char s_login[roundup(MAXLOGNAME, sizeof(long))]; struct mtx s_mtx; /* Mutex to protect members. */ }; /* * One structure allocated per process group. * * List of locks * (m) locked by pg_mtx mtx * (e) locked by proctree_lock sx * (c) const until freeing */ struct pgrp { LIST_ENTRY(pgrp) pg_hash; /* (e) Hash chain. */ LIST_HEAD(, proc) pg_members; /* (m + e) Pointer to pgrp members. */ struct session *pg_session; /* (c) Pointer to session. */ struct sigiolst pg_sigiolst; /* (m) List of sigio sources. */ pid_t pg_id; /* (c) Process group id. */ int pg_jobc; /* (m) Job control process count. */ struct mtx pg_mtx; /* Mutex to protect members */ }; /* * pargs, used to hold a copy of the command line, if it had a sane length. */ struct pargs { u_int ar_ref; /* Reference count. */ u_int ar_length; /* Length. */ u_char ar_args[1]; /* Arguments. */ }; /*- * Description of a process. * * This structure contains the information needed to manage a thread of * control, known in UN*X as a process; it has references to substructures * containing descriptions of things that the process uses, but may share * with related processes. The process structure and the substructures * are always addressable except for those marked "(CPU)" below, * which might be addressable only on a processor on which the process * is running. * * Below is a key of locks used to protect each member of struct proc. The * lock is indicated by a reference to a specific character in parens in the * associated comment. * * - not yet protected * a - only touched by curproc or parent during fork/wait * b - created at fork, never changes * (exception aiods switch vmspaces, but they are also * marked 'P_SYSTEM' so hopefully it will be left alone) * c - locked by proc mtx * d - locked by allproc_lock lock * e - locked by proctree_lock lock * f - session mtx * g - process group mtx * h - callout_lock mtx * i - by curproc or the master session mtx * j - locked by proc slock * k - only accessed by curthread * k*- only accessed by curthread and from an interrupt * l - the attaching proc or attaching proc parent * m - Giant * n - not locked, lazy * o - ktrace lock * q - td_contested lock * r - p_peers lock * t - thread lock * u - process stat lock * w - process timer lock * x - created at fork, only changes during single threading in exec * y - created at first aio, doesn't change until exit or exec at which * point we are single-threaded and only curthread changes it * z - zombie threads lock * * If the locking key specifies two identifiers (for example, p_pptr) then * either lock is sufficient for read access, but both locks must be held * for write access. */ struct cpuset; struct kaioinfo; struct kaudit_record; struct kdtrace_proc; struct kdtrace_thread; struct mqueue_notifier; struct nlminfo; struct p_sched; struct proc; struct procdesc; struct racct; struct sbuf; struct sleepqueue; struct td_sched; struct thread; struct trapframe; struct turnstile; /* * XXX: Does this belong in resource.h or resourcevar.h instead? * Resource usage extension. The times in rusage structs in the kernel are * never up to date. The actual times are kept as runtimes and tick counts * (with control info in the "previous" times), and are converted when * userland asks for rusage info. Backwards compatibility prevents putting * this directly in the user-visible rusage struct. * * Locking for p_rux: (cu) means (u) for p_rux and (c) for p_crux. * Locking for td_rux: (t) for all fields. */ struct rusage_ext { uint64_t rux_runtime; /* (cu) Real time. */ uint64_t rux_uticks; /* (cu) Statclock hits in user mode. */ uint64_t rux_sticks; /* (cu) Statclock hits in sys mode. */ uint64_t rux_iticks; /* (cu) Statclock hits in intr mode. */ uint64_t rux_uu; /* (c) Previous user time in usec. */ uint64_t rux_su; /* (c) Previous sys time in usec. */ uint64_t rux_tu; /* (c) Previous total time in usec. */ }; /* * Kernel runnable context (thread). * This is what is put to sleep and reactivated. * Thread context. Processes may have multiple threads. */ struct thread { struct mtx *volatile td_lock; /* replaces sched lock */ struct proc *td_proc; /* (*) Associated process. */ TAILQ_ENTRY(thread) td_plist; /* (*) All threads in this proc. */ TAILQ_ENTRY(thread) td_runq; /* (t) Run queue. */ TAILQ_ENTRY(thread) td_slpq; /* (t) Sleep queue. */ TAILQ_ENTRY(thread) td_lockq; /* (t) Lock queue. */ LIST_ENTRY(thread) td_hash; /* (d) Hash chain. */ struct cpuset *td_cpuset; /* (t) CPU affinity mask. */ struct seltd *td_sel; /* Select queue/channel. */ struct sleepqueue *td_sleepqueue; /* (k) Associated sleep queue. */ struct turnstile *td_turnstile; /* (k) Associated turnstile. */ struct rl_q_entry *td_rlqe; /* (k) Associated range lock entry. */ struct umtx_q *td_umtxq; /* (c?) Link for when we're blocked. */ lwpid_t td_tid; /* (b) Thread ID. */ sigqueue_t td_sigqueue; /* (c) Sigs arrived, not delivered. */ #define td_siglist td_sigqueue.sq_signals u_char td_lend_user_pri; /* (t) Lend user pri. */ /* Cleared during fork1() */ #define td_startzero td_flags int td_flags; /* (t) TDF_* flags. */ int td_inhibitors; /* (t) Why can not run. */ int td_pflags; /* (k) Private thread (TDP_*) flags. */ int td_dupfd; /* (k) Ret value from fdopen. XXX */ int td_sqqueue; /* (t) Sleepqueue queue blocked on. */ void *td_wchan; /* (t) Sleep address. */ const char *td_wmesg; /* (t) Reason for sleep. */ int td_lastcpu; /* (t) Last cpu we were on. */ int td_oncpu; /* (t) Which cpu we are on. */ volatile u_char td_owepreempt; /* (k*) Preempt on last critical_exit */ u_char td_tsqueue; /* (t) Turnstile queue blocked on. */ short td_locks; /* (k) Count of non-spin locks. */ short td_rw_rlocks; /* (k) Count of rwlock read locks. */ short td_lk_slocks; /* (k) Count of lockmgr shared locks. */ short td_stopsched; /* (k) Scheduler stopped. */ struct turnstile *td_blocked; /* (t) Lock thread is blocked on. */ const char *td_lockname; /* (t) Name of lock blocked on. */ LIST_HEAD(, turnstile) td_contested; /* (q) Contested locks. */ struct lock_list_entry *td_sleeplocks; /* (k) Held sleep locks. */ int td_intr_nesting_level; /* (k) Interrupt recursion. */ int td_pinned; /* (k) Temporary cpu pin count. */ struct ucred *td_ucred; /* (k) Reference to credentials. */ + struct plimit *td_limit; /* (k) Resource limits. */ u_int td_estcpu; /* (t) estimated cpu utilization */ int td_slptick; /* (t) Time at sleep. */ int td_blktick; /* (t) Time spent blocked. */ int td_swvoltick; /* (t) Time at last SW_VOL switch. */ u_int td_cow; /* (*) Number of copy-on-write faults */ struct rusage td_ru; /* (t) rusage information. */ struct rusage_ext td_rux; /* (t) Internal rusage information. */ uint64_t td_incruntime; /* (t) Cpu ticks to transfer to proc. */ uint64_t td_runtime; /* (t) How many cpu ticks we've run. */ u_int td_pticks; /* (t) Statclock hits for profiling */ u_int td_sticks; /* (t) Statclock hits in system mode. */ u_int td_iticks; /* (t) Statclock hits in intr mode. */ u_int td_uticks; /* (t) Statclock hits in user mode. */ int td_intrval; /* (t) Return value for sleepq. */ sigset_t td_oldsigmask; /* (k) Saved mask from pre sigpause. */ volatile u_int td_generation; /* (k) For detection of preemption */ stack_t td_sigstk; /* (k) Stack ptr and on-stack flag. */ int td_xsig; /* (c) Signal for ptrace */ u_long td_profil_addr; /* (k) Temporary addr until AST. */ u_int td_profil_ticks; /* (k) Temporary ticks until AST. */ char td_name[MAXCOMLEN + 1]; /* (*) Thread name. */ struct file *td_fpop; /* (k) file referencing cdev under op */ int td_dbgflags; /* (c) Userland debugger flags */ struct ksiginfo td_dbgksi; /* (c) ksi reflected to debugger. */ int td_ng_outbound; /* (k) Thread entered ng from above. */ struct osd td_osd; /* (k) Object specific data. */ struct vm_map_entry *td_map_def_user; /* (k) Deferred entries. */ pid_t td_dbg_forked; /* (c) Child pid for debugger. */ u_int td_vp_reserv; /* (k) Count of reserved vnodes. */ int td_no_sleeping; /* (k) Sleeping disabled count. */ int td_dom_rr_idx; /* (k) RR Numa domain selection. */ void *td_su; /* (k) FFS SU private */ #define td_endzero td_sigmask /* Copied during fork1() or create_thread(). */ #define td_startcopy td_endzero sigset_t td_sigmask; /* (c) Current signal mask. */ u_char td_rqindex; /* (t) Run queue index. */ u_char td_base_pri; /* (t) Thread base kernel priority. */ u_char td_priority; /* (t) Thread active priority. */ u_char td_pri_class; /* (t) Scheduling class. */ u_char td_user_pri; /* (t) User pri from estcpu and nice. */ u_char td_base_user_pri; /* (t) Base user pri */ #define td_endcopy td_pcb /* * Fields that must be manually set in fork1() or create_thread() * or already have been set in the allocator, constructor, etc. */ struct pcb *td_pcb; /* (k) Kernel VA of pcb and kstack. */ enum { TDS_INACTIVE = 0x0, TDS_INHIBITED, TDS_CAN_RUN, TDS_RUNQ, TDS_RUNNING } td_state; /* (t) thread state */ union { register_t tdu_retval[2]; off_t tdu_off; } td_uretoff; /* (k) Syscall aux returns. */ #define td_retval td_uretoff.tdu_retval u_int td_cowgen; /* (k) Generation of COW pointers. */ struct callout td_slpcallout; /* (h) Callout for sleep. */ struct trapframe *td_frame; /* (k) */ struct vm_object *td_kstack_obj;/* (a) Kstack object. */ vm_offset_t td_kstack; /* (a) Kernel VA of kstack. */ int td_kstack_pages; /* (a) Size of the kstack. */ volatile u_int td_critnest; /* (k*) Critical section nest level. */ struct mdthread td_md; /* (k) Any machine-dependent fields. */ struct td_sched *td_sched; /* (*) Scheduler-specific data. */ struct kaudit_record *td_ar; /* (k) Active audit record, if any. */ struct lpohead td_lprof[2]; /* (a) lock profiling objects. */ struct kdtrace_thread *td_dtrace; /* (*) DTrace-specific data. */ int td_errno; /* Error returned by last syscall. */ struct vnet *td_vnet; /* (k) Effective vnet. */ const char *td_vnet_lpush; /* (k) Debugging vnet push / pop. */ struct trapframe *td_intr_frame;/* (k) Frame of the current irq */ struct proc *td_rfppwait_p; /* (k) The vforked child */ struct vm_page **td_ma; /* (k) uio pages held */ int td_ma_cnt; /* (k) size of *td_ma */ void *td_emuldata; /* Emulator state data */ }; struct mtx *thread_lock_block(struct thread *); void thread_lock_unblock(struct thread *, struct mtx *); void thread_lock_set(struct thread *, struct mtx *); #define THREAD_LOCK_ASSERT(td, type) \ do { \ struct mtx *__m = (td)->td_lock; \ if (__m != &blocked_lock) \ mtx_assert(__m, (type)); \ } while (0) #ifdef INVARIANTS #define THREAD_LOCKPTR_ASSERT(td, lock) \ do { \ struct mtx *__m = (td)->td_lock; \ KASSERT((__m == &blocked_lock || __m == (lock)), \ ("Thread %p lock %p does not match %p", td, __m, (lock))); \ } while (0) #else #define THREAD_LOCKPTR_ASSERT(td, lock) #endif /* * Flags kept in td_flags: * To change these you MUST have the scheduler lock. */ #define TDF_BORROWING 0x00000001 /* Thread is borrowing pri from another. */ #define TDF_INPANIC 0x00000002 /* Caused a panic, let it drive crashdump. */ #define TDF_INMEM 0x00000004 /* Thread's stack is in memory. */ #define TDF_SINTR 0x00000008 /* Sleep is interruptible. */ #define TDF_TIMEOUT 0x00000010 /* Timing out during sleep. */ #define TDF_IDLETD 0x00000020 /* This is a per-CPU idle thread. */ #define TDF_CANSWAP 0x00000040 /* Thread can be swapped. */ #define TDF_SLEEPABORT 0x00000080 /* sleepq_abort was called. */ #define TDF_KTH_SUSP 0x00000100 /* kthread is suspended */ #define TDF_ALLPROCSUSP 0x00000200 /* suspended by SINGLE_ALLPROC */ #define TDF_BOUNDARY 0x00000400 /* Thread suspended at user boundary */ #define TDF_ASTPENDING 0x00000800 /* Thread has some asynchronous events. */ #define TDF_TIMOFAIL 0x00001000 /* Timeout from sleep after we were awake. */ #define TDF_SBDRY 0x00002000 /* Stop only on usermode boundary. */ #define TDF_UPIBLOCKED 0x00004000 /* Thread blocked on user PI mutex. */ #define TDF_NEEDSUSPCHK 0x00008000 /* Thread may need to suspend. */ #define TDF_NEEDRESCHED 0x00010000 /* Thread needs to yield. */ #define TDF_NEEDSIGCHK 0x00020000 /* Thread may need signal delivery. */ #define TDF_NOLOAD 0x00040000 /* Ignore during load avg calculations. */ #define TDF_UNUSED19 0x00080000 /* --available-- */ #define TDF_THRWAKEUP 0x00100000 /* Libthr thread must not suspend itself. */ #define TDF_UNUSED21 0x00200000 /* --available-- */ #define TDF_SWAPINREQ 0x00400000 /* Swapin request due to wakeup. */ #define TDF_UNUSED23 0x00800000 /* --available-- */ #define TDF_SCHED0 0x01000000 /* Reserved for scheduler private use */ #define TDF_SCHED1 0x02000000 /* Reserved for scheduler private use */ #define TDF_SCHED2 0x04000000 /* Reserved for scheduler private use */ #define TDF_SCHED3 0x08000000 /* Reserved for scheduler private use */ #define TDF_ALRMPEND 0x10000000 /* Pending SIGVTALRM needs to be posted. */ #define TDF_PROFPEND 0x20000000 /* Pending SIGPROF needs to be posted. */ #define TDF_MACPEND 0x40000000 /* AST-based MAC event pending. */ /* Userland debug flags */ #define TDB_SUSPEND 0x00000001 /* Thread is suspended by debugger */ #define TDB_XSIG 0x00000002 /* Thread is exchanging signal under trace */ #define TDB_USERWR 0x00000004 /* Debugger modified memory or registers */ #define TDB_SCE 0x00000008 /* Thread performs syscall enter */ #define TDB_SCX 0x00000010 /* Thread performs syscall exit */ #define TDB_EXEC 0x00000020 /* TDB_SCX from exec(2) family */ #define TDB_FORK 0x00000040 /* TDB_SCX from fork(2) that created new process */ #define TDB_STOPATFORK 0x00000080 /* Stop at the return from fork (child only) */ #define TDB_CHILD 0x00000100 /* New child indicator for ptrace() */ /* * "Private" flags kept in td_pflags: * These are only written by curthread and thus need no locking. */ #define TDP_OLDMASK 0x00000001 /* Need to restore mask after suspend. */ #define TDP_INKTR 0x00000002 /* Thread is currently in KTR code. */ #define TDP_INKTRACE 0x00000004 /* Thread is currently in KTRACE code. */ #define TDP_BUFNEED 0x00000008 /* Do not recurse into the buf flush */ #define TDP_COWINPROGRESS 0x00000010 /* Snapshot copy-on-write in progress. */ #define TDP_ALTSTACK 0x00000020 /* Have alternate signal stack. */ #define TDP_DEADLKTREAT 0x00000040 /* Lock aquisition - deadlock treatment. */ #define TDP_NOFAULTING 0x00000080 /* Do not handle page faults. */ #define TDP_UNUSED9 0x00000100 /* --available-- */ #define TDP_OWEUPC 0x00000200 /* Call addupc() at next AST. */ #define TDP_ITHREAD 0x00000400 /* Thread is an interrupt thread. */ #define TDP_SYNCIO 0x00000800 /* Local override, disable async i/o. */ #define TDP_SCHED1 0x00001000 /* Reserved for scheduler private use */ #define TDP_SCHED2 0x00002000 /* Reserved for scheduler private use */ #define TDP_SCHED3 0x00004000 /* Reserved for scheduler private use */ #define TDP_SCHED4 0x00008000 /* Reserved for scheduler private use */ #define TDP_GEOM 0x00010000 /* Settle GEOM before finishing syscall */ #define TDP_SOFTDEP 0x00020000 /* Stuck processing softdep worklist */ #define TDP_NORUNNINGBUF 0x00040000 /* Ignore runningbufspace check */ #define TDP_WAKEUP 0x00080000 /* Don't sleep in umtx cond_wait */ #define TDP_INBDFLUSH 0x00100000 /* Already in BO_BDFLUSH, do not recurse */ #define TDP_KTHREAD 0x00200000 /* This is an official kernel thread */ #define TDP_CALLCHAIN 0x00400000 /* Capture thread's callchain */ #define TDP_IGNSUSP 0x00800000 /* Permission to ignore the MNTK_SUSPEND* */ #define TDP_AUDITREC 0x01000000 /* Audit record pending on thread */ #define TDP_RFPPWAIT 0x02000000 /* Handle RFPPWAIT on syscall exit */ #define TDP_RESETSPUR 0x04000000 /* Reset spurious page fault history. */ #define TDP_NERRNO 0x08000000 /* Last errno is already in td_errno */ #define TDP_UIOHELD 0x10000000 /* Current uio has pages held in td_ma */ #define TDP_UNUSED29 0x20000000 /* --available-- */ #define TDP_EXECVMSPC 0x40000000 /* Execve destroyed old vmspace */ /* * Reasons that the current thread can not be run yet. * More than one may apply. */ #define TDI_SUSPENDED 0x0001 /* On suspension queue. */ #define TDI_SLEEPING 0x0002 /* Actually asleep! (tricky). */ #define TDI_SWAPPED 0x0004 /* Stack not in mem. Bad juju if run. */ #define TDI_LOCK 0x0008 /* Stopped on a lock. */ #define TDI_IWAIT 0x0010 /* Awaiting interrupt. */ #define TD_IS_SLEEPING(td) ((td)->td_inhibitors & TDI_SLEEPING) #define TD_ON_SLEEPQ(td) ((td)->td_wchan != NULL) #define TD_IS_SUSPENDED(td) ((td)->td_inhibitors & TDI_SUSPENDED) #define TD_IS_SWAPPED(td) ((td)->td_inhibitors & TDI_SWAPPED) #define TD_ON_LOCK(td) ((td)->td_inhibitors & TDI_LOCK) #define TD_AWAITING_INTR(td) ((td)->td_inhibitors & TDI_IWAIT) #define TD_IS_RUNNING(td) ((td)->td_state == TDS_RUNNING) #define TD_ON_RUNQ(td) ((td)->td_state == TDS_RUNQ) #define TD_CAN_RUN(td) ((td)->td_state == TDS_CAN_RUN) #define TD_IS_INHIBITED(td) ((td)->td_state == TDS_INHIBITED) #define TD_ON_UPILOCK(td) ((td)->td_flags & TDF_UPIBLOCKED) #define TD_IS_IDLETHREAD(td) ((td)->td_flags & TDF_IDLETD) #define TD_SET_INHIB(td, inhib) do { \ (td)->td_state = TDS_INHIBITED; \ (td)->td_inhibitors |= (inhib); \ } while (0) #define TD_CLR_INHIB(td, inhib) do { \ if (((td)->td_inhibitors & (inhib)) && \ (((td)->td_inhibitors &= ~(inhib)) == 0)) \ (td)->td_state = TDS_CAN_RUN; \ } while (0) #define TD_SET_SLEEPING(td) TD_SET_INHIB((td), TDI_SLEEPING) #define TD_SET_SWAPPED(td) TD_SET_INHIB((td), TDI_SWAPPED) #define TD_SET_LOCK(td) TD_SET_INHIB((td), TDI_LOCK) #define TD_SET_SUSPENDED(td) TD_SET_INHIB((td), TDI_SUSPENDED) #define TD_SET_IWAIT(td) TD_SET_INHIB((td), TDI_IWAIT) #define TD_SET_EXITING(td) TD_SET_INHIB((td), TDI_EXITING) #define TD_CLR_SLEEPING(td) TD_CLR_INHIB((td), TDI_SLEEPING) #define TD_CLR_SWAPPED(td) TD_CLR_INHIB((td), TDI_SWAPPED) #define TD_CLR_LOCK(td) TD_CLR_INHIB((td), TDI_LOCK) #define TD_CLR_SUSPENDED(td) TD_CLR_INHIB((td), TDI_SUSPENDED) #define TD_CLR_IWAIT(td) TD_CLR_INHIB((td), TDI_IWAIT) #define TD_SET_RUNNING(td) (td)->td_state = TDS_RUNNING #define TD_SET_RUNQ(td) (td)->td_state = TDS_RUNQ #define TD_SET_CAN_RUN(td) (td)->td_state = TDS_CAN_RUN /* * Process structure. */ struct proc { LIST_ENTRY(proc) p_list; /* (d) List of all processes. */ TAILQ_HEAD(, thread) p_threads; /* (c) all threads. */ struct mtx p_slock; /* process spin lock */ struct ucred *p_ucred; /* (c) Process owner's identity. */ struct filedesc *p_fd; /* (b) Open files. */ struct filedesc_to_leader *p_fdtol; /* (b) Tracking node */ struct pstats *p_stats; /* (b) Accounting/statistics (CPU). */ - struct plimit *p_limit; /* (c) Process limits. */ + struct plimit *p_limit; /* (c) Resource limits. */ struct callout p_limco; /* (c) Limit callout handle */ struct sigacts *p_sigacts; /* (x) Signal actions, state (CPU). */ int p_flag; /* (c) P_* flags. */ int p_flag2; /* (c) P2_* flags. */ enum { PRS_NEW = 0, /* In creation */ PRS_NORMAL, /* threads can be run. */ PRS_ZOMBIE } p_state; /* (j/c) Process status. */ pid_t p_pid; /* (b) Process identifier. */ LIST_ENTRY(proc) p_hash; /* (d) Hash chain. */ LIST_ENTRY(proc) p_pglist; /* (g + e) List of processes in pgrp. */ struct proc *p_pptr; /* (c + e) Pointer to parent process. */ LIST_ENTRY(proc) p_sibling; /* (e) List of sibling processes. */ LIST_HEAD(, proc) p_children; /* (e) Pointer to list of children. */ struct proc *p_reaper; /* (e) My reaper. */ LIST_HEAD(, proc) p_reaplist; /* (e) List of my descendants (if I am reaper). */ LIST_ENTRY(proc) p_reapsibling; /* (e) List of siblings - descendants of the same reaper. */ struct mtx p_mtx; /* (n) Lock for this struct. */ struct mtx p_statmtx; /* Lock for the stats */ struct mtx p_itimmtx; /* Lock for the virt/prof timers */ struct mtx p_profmtx; /* Lock for the profiling */ struct ksiginfo *p_ksi; /* Locked by parent proc lock */ sigqueue_t p_sigqueue; /* (c) Sigs not delivered to a td. */ #define p_siglist p_sigqueue.sq_signals /* The following fields are all zeroed upon creation in fork. */ #define p_startzero p_oppid pid_t p_oppid; /* (c + e) Save ppid in ptrace. XXX */ struct vmspace *p_vmspace; /* (b) Address space. */ u_int p_swtick; /* (c) Tick when swapped in or out. */ u_int p_cowgen; /* (c) Generation of COW pointers. */ struct itimerval p_realtimer; /* (c) Alarm timer. */ struct rusage p_ru; /* (a) Exit information. */ struct rusage_ext p_rux; /* (cu) Internal resource usage. */ struct rusage_ext p_crux; /* (c) Internal child resource usage. */ int p_profthreads; /* (c) Num threads in addupc_task. */ volatile int p_exitthreads; /* (j) Number of threads exiting */ int p_traceflag; /* (o) Kernel trace points. */ struct vnode *p_tracevp; /* (c + o) Trace to vnode. */ struct ucred *p_tracecred; /* (o) Credentials to trace with. */ struct vnode *p_textvp; /* (b) Vnode of executable. */ u_int p_lock; /* (c) Proclock (prevent swap) count. */ struct sigiolst p_sigiolst; /* (c) List of sigio sources. */ int p_sigparent; /* (c) Signal to parent on exit. */ int p_sig; /* (n) For core dump/debugger XXX. */ u_long p_code; /* (n) For core dump/debugger XXX. */ u_int p_stops; /* (c) Stop event bitmask. */ u_int p_stype; /* (c) Stop event type. */ char p_step; /* (c) Process is stopped. */ u_char p_pfsflags; /* (c) Procfs flags. */ struct nlminfo *p_nlminfo; /* (?) Only used by/for lockd. */ struct kaioinfo *p_aioinfo; /* (y) ASYNC I/O info. */ struct thread *p_singlethread;/* (c + j) If single threading this is it */ int p_suspcount; /* (j) Num threads in suspended mode. */ struct thread *p_xthread; /* (c) Trap thread */ int p_boundary_count;/* (j) Num threads at user boundary */ int p_pendingcnt; /* how many signals are pending */ struct itimers *p_itimers; /* (c) POSIX interval timers. */ struct procdesc *p_procdesc; /* (e) Process descriptor, if any. */ u_int p_treeflag; /* (e) P_TREE flags */ /* End area that is zeroed on creation. */ #define p_endzero p_magic /* The following fields are all copied upon creation in fork. */ #define p_startcopy p_endzero u_int p_magic; /* (b) Magic number. */ int p_osrel; /* (x) osreldate for the binary (from ELF note, if any) */ char p_comm[MAXCOMLEN + 1]; /* (b) Process name. */ struct pgrp *p_pgrp; /* (c + e) Pointer to process group. */ struct sysentvec *p_sysent; /* (b) Syscall dispatch info. */ struct pargs *p_args; /* (c) Process arguments. */ rlim_t p_cpulimit; /* (c) Current CPU limit in seconds. */ signed char p_nice; /* (c) Process "nice" value. */ int p_fibnum; /* in this routing domain XXX MRT */ pid_t p_reapsubtree; /* (e) Pid of the direct child of the reaper which spawned our subtree. */ /* End area that is copied on creation. */ #define p_endcopy p_xstat u_short p_xstat; /* (c) Exit status; also stop sig. */ struct knlist p_klist; /* (c) Knotes attached to this proc. */ int p_numthreads; /* (c) Number of threads. */ struct mdproc p_md; /* Any machine-dependent fields. */ struct callout p_itcallout; /* (h + c) Interval timer callout. */ u_short p_acflag; /* (c) Accounting flags. */ struct proc *p_peers; /* (r) */ struct proc *p_leader; /* (b) */ void *p_emuldata; /* (c) Emulator state data. */ struct label *p_label; /* (*) Proc (not subject) MAC label. */ struct p_sched *p_sched; /* (*) Scheduler-specific data. */ STAILQ_HEAD(, ktr_request) p_ktr; /* (o) KTR event queue. */ LIST_HEAD(, mqueue_notifier) p_mqnotifier; /* (c) mqueue notifiers.*/ struct kdtrace_proc *p_dtrace; /* (*) DTrace-specific data. */ struct cv p_pwait; /* (*) wait cv for exit/exec. */ struct cv p_dbgwait; /* (*) wait cv for debugger attach after fork. */ uint64_t p_prev_runtime; /* (c) Resource usage accounting. */ struct racct *p_racct; /* (b) Resource accounting. */ u_char p_throttled; /* (c) Flag for racct pcpu throttling */ /* * An orphan is the child that has beed re-parented to the * debugger as a result of attaching to it. Need to keep * track of them for parent to be able to collect the exit * status of what used to be children. */ LIST_ENTRY(proc) p_orphan; /* (e) List of orphan processes. */ LIST_HEAD(, proc) p_orphans; /* (e) Pointer to list of orphans. */ }; #define p_session p_pgrp->pg_session #define p_pgid p_pgrp->pg_id #define NOCPU (-1) /* For when we aren't on a CPU. */ #define NOCPU_OLD (255) #define MAXCPU_OLD (254) #define PROC_SLOCK(p) mtx_lock_spin(&(p)->p_slock) #define PROC_SUNLOCK(p) mtx_unlock_spin(&(p)->p_slock) #define PROC_SLOCK_ASSERT(p, type) mtx_assert(&(p)->p_slock, (type)) #define PROC_STATLOCK(p) mtx_lock_spin(&(p)->p_statmtx) #define PROC_STATUNLOCK(p) mtx_unlock_spin(&(p)->p_statmtx) #define PROC_STATLOCK_ASSERT(p, type) mtx_assert(&(p)->p_statmtx, (type)) #define PROC_ITIMLOCK(p) mtx_lock_spin(&(p)->p_itimmtx) #define PROC_ITIMUNLOCK(p) mtx_unlock_spin(&(p)->p_itimmtx) #define PROC_ITIMLOCK_ASSERT(p, type) mtx_assert(&(p)->p_itimmtx, (type)) #define PROC_PROFLOCK(p) mtx_lock_spin(&(p)->p_profmtx) #define PROC_PROFUNLOCK(p) mtx_unlock_spin(&(p)->p_profmtx) #define PROC_PROFLOCK_ASSERT(p, type) mtx_assert(&(p)->p_profmtx, (type)) /* These flags are kept in p_flag. */ #define P_ADVLOCK 0x00001 /* Process may hold a POSIX advisory lock. */ #define P_CONTROLT 0x00002 /* Has a controlling terminal. */ #define P_KTHREAD 0x00004 /* Kernel thread (*). */ #define P_FOLLOWFORK 0x00008 /* Attach parent debugger to children. */ #define P_PPWAIT 0x00010 /* Parent is waiting for child to exec/exit. */ #define P_PROFIL 0x00020 /* Has started profiling. */ #define P_STOPPROF 0x00040 /* Has thread requesting to stop profiling. */ #define P_HADTHREADS 0x00080 /* Has had threads (no cleanup shortcuts) */ #define P_SUGID 0x00100 /* Had set id privileges since last exec. */ #define P_SYSTEM 0x00200 /* System proc: no sigs, stats or swapping. */ #define P_SINGLE_EXIT 0x00400 /* Threads suspending should exit, not wait. */ #define P_TRACED 0x00800 /* Debugged process being traced. */ #define P_WAITED 0x01000 /* Someone is waiting for us. */ #define P_WEXIT 0x02000 /* Working on exiting. */ #define P_EXEC 0x04000 /* Process called exec. */ #define P_WKILLED 0x08000 /* Killed, go to kernel/user boundary ASAP. */ #define P_CONTINUED 0x10000 /* Proc has continued from a stopped state. */ #define P_STOPPED_SIG 0x20000 /* Stopped due to SIGSTOP/SIGTSTP. */ #define P_STOPPED_TRACE 0x40000 /* Stopped because of tracing. */ #define P_STOPPED_SINGLE 0x80000 /* Only 1 thread can continue (not to user). */ #define P_PROTECTED 0x100000 /* Do not kill on memory overcommit. */ #define P_SIGEVENT 0x200000 /* Process pending signals changed. */ #define P_SINGLE_BOUNDARY 0x400000 /* Threads should suspend at user boundary. */ #define P_HWPMC 0x800000 /* Process is using HWPMCs */ #define P_JAILED 0x1000000 /* Process is in jail. */ #define P_TOTAL_STOP 0x2000000 /* Stopped in stop_all_proc. */ #define P_INEXEC 0x4000000 /* Process is in execve(). */ #define P_STATCHILD 0x8000000 /* Child process stopped or exited. */ #define P_INMEM 0x10000000 /* Loaded into memory. */ #define P_SWAPPINGOUT 0x20000000 /* Process is being swapped out. */ #define P_SWAPPINGIN 0x40000000 /* Process is being swapped in. */ #define P_PPTRACE 0x80000000 /* PT_TRACEME by vforked child. */ #define P_STOPPED (P_STOPPED_SIG|P_STOPPED_SINGLE|P_STOPPED_TRACE) #define P_SHOULDSTOP(p) ((p)->p_flag & P_STOPPED) #define P_KILLED(p) ((p)->p_flag & P_WKILLED) /* These flags are kept in p_flag2. */ #define P2_INHERIT_PROTECTED 0x00000001 /* New children get P_PROTECTED. */ #define P2_NOTRACE 0x00000002 /* No ptrace(2) attach or coredumps. */ #define P2_NOTRACE_EXEC 0x00000004 /* Keep P2_NOPTRACE on exec(2). */ #define P2_AST_SU 0x00000008 /* Handles SU ast for kthreads. */ /* Flags protected by proctree_lock, kept in p_treeflags. */ #define P_TREE_ORPHANED 0x00000001 /* Reparented, on orphan list */ #define P_TREE_FIRST_ORPHAN 0x00000002 /* First element of orphan list */ #define P_TREE_REAPER 0x00000004 /* Reaper of subtree */ /* * These were process status values (p_stat), now they are only used in * legacy conversion code. */ #define SIDL 1 /* Process being created by fork. */ #define SRUN 2 /* Currently runnable. */ #define SSLEEP 3 /* Sleeping on an address. */ #define SSTOP 4 /* Process debugging or suspension. */ #define SZOMB 5 /* Awaiting collection by parent. */ #define SWAIT 6 /* Waiting for interrupt. */ #define SLOCK 7 /* Blocked on a lock. */ #define P_MAGIC 0xbeefface #ifdef _KERNEL /* Types and flags for mi_switch(). */ #define SW_TYPE_MASK 0xff /* First 8 bits are switch type */ #define SWT_NONE 0 /* Unspecified switch. */ #define SWT_PREEMPT 1 /* Switching due to preemption. */ #define SWT_OWEPREEMPT 2 /* Switching due to opepreempt. */ #define SWT_TURNSTILE 3 /* Turnstile contention. */ #define SWT_SLEEPQ 4 /* Sleepq wait. */ #define SWT_SLEEPQTIMO 5 /* Sleepq timeout wait. */ #define SWT_RELINQUISH 6 /* yield call. */ #define SWT_NEEDRESCHED 7 /* NEEDRESCHED was set. */ #define SWT_IDLE 8 /* Switching from the idle thread. */ #define SWT_IWAIT 9 /* Waiting for interrupts. */ #define SWT_SUSPEND 10 /* Thread suspended. */ #define SWT_REMOTEPREEMPT 11 /* Remote processor preempted. */ #define SWT_REMOTEWAKEIDLE 12 /* Remote processor preempted idle. */ #define SWT_COUNT 13 /* Number of switch types. */ /* Flags */ #define SW_VOL 0x0100 /* Voluntary switch. */ #define SW_INVOL 0x0200 /* Involuntary switch. */ #define SW_PREEMPT 0x0400 /* The invol switch is a preemption */ /* How values for thread_single(). */ #define SINGLE_NO_EXIT 0 #define SINGLE_EXIT 1 #define SINGLE_BOUNDARY 2 #define SINGLE_ALLPROC 3 #ifdef MALLOC_DECLARE MALLOC_DECLARE(M_PARGS); MALLOC_DECLARE(M_PGRP); MALLOC_DECLARE(M_SESSION); MALLOC_DECLARE(M_SUBPROC); #endif #define FOREACH_PROC_IN_SYSTEM(p) \ LIST_FOREACH((p), &allproc, p_list) #define FOREACH_THREAD_IN_PROC(p, td) \ TAILQ_FOREACH((td), &(p)->p_threads, td_plist) #define FIRST_THREAD_IN_PROC(p) TAILQ_FIRST(&(p)->p_threads) /* * We use process IDs <= pid_max <= PID_MAX; PID_MAX + 1 must also fit * in a pid_t, as it is used to represent "no process group". */ #define PID_MAX 99999 #define NO_PID 100000 extern pid_t pid_max; #define SESS_LEADER(p) ((p)->p_session->s_leader == (p)) #define STOPEVENT(p, e, v) do { \ WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, \ "checking stopevent %d", (e)); \ if ((p)->p_stops & (e)) { \ PROC_LOCK(p); \ stopevent((p), (e), (v)); \ PROC_UNLOCK(p); \ } \ } while (0) #define _STOPEVENT(p, e, v) do { \ PROC_LOCK_ASSERT(p, MA_OWNED); \ WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, &p->p_mtx.lock_object, \ "checking stopevent %d", (e)); \ if ((p)->p_stops & (e)) \ stopevent((p), (e), (v)); \ } while (0) /* Lock and unlock a process. */ #define PROC_LOCK(p) mtx_lock(&(p)->p_mtx) #define PROC_TRYLOCK(p) mtx_trylock(&(p)->p_mtx) #define PROC_UNLOCK(p) mtx_unlock(&(p)->p_mtx) #define PROC_LOCKED(p) mtx_owned(&(p)->p_mtx) #define PROC_LOCK_ASSERT(p, type) mtx_assert(&(p)->p_mtx, (type)) /* Lock and unlock a process group. */ #define PGRP_LOCK(pg) mtx_lock(&(pg)->pg_mtx) #define PGRP_UNLOCK(pg) mtx_unlock(&(pg)->pg_mtx) #define PGRP_LOCKED(pg) mtx_owned(&(pg)->pg_mtx) #define PGRP_LOCK_ASSERT(pg, type) mtx_assert(&(pg)->pg_mtx, (type)) #define PGRP_LOCK_PGSIGNAL(pg) do { \ if ((pg) != NULL) \ PGRP_LOCK(pg); \ } while (0) #define PGRP_UNLOCK_PGSIGNAL(pg) do { \ if ((pg) != NULL) \ PGRP_UNLOCK(pg); \ } while (0) /* Lock and unlock a session. */ #define SESS_LOCK(s) mtx_lock(&(s)->s_mtx) #define SESS_UNLOCK(s) mtx_unlock(&(s)->s_mtx) #define SESS_LOCKED(s) mtx_owned(&(s)->s_mtx) #define SESS_LOCK_ASSERT(s, type) mtx_assert(&(s)->s_mtx, (type)) /* Hold process U-area in memory, normally for ptrace/procfs work. */ #define PHOLD(p) do { \ PROC_LOCK(p); \ _PHOLD(p); \ PROC_UNLOCK(p); \ } while (0) #define _PHOLD(p) do { \ PROC_LOCK_ASSERT((p), MA_OWNED); \ KASSERT(!((p)->p_flag & P_WEXIT) || (p) == curproc, \ ("PHOLD of exiting process")); \ (p)->p_lock++; \ if (((p)->p_flag & P_INMEM) == 0) \ faultin((p)); \ } while (0) #define PROC_ASSERT_HELD(p) do { \ KASSERT((p)->p_lock > 0, ("process not held")); \ } while (0) #define PRELE(p) do { \ PROC_LOCK((p)); \ _PRELE((p)); \ PROC_UNLOCK((p)); \ } while (0) #define _PRELE(p) do { \ PROC_LOCK_ASSERT((p), MA_OWNED); \ PROC_ASSERT_HELD(p); \ (--(p)->p_lock); \ if (((p)->p_flag & P_WEXIT) && (p)->p_lock == 0) \ wakeup(&(p)->p_lock); \ } while (0) #define PROC_ASSERT_NOT_HELD(p) do { \ KASSERT((p)->p_lock == 0, ("process held")); \ } while (0) #define PROC_UPDATE_COW(p) do { \ PROC_LOCK_ASSERT((p), MA_OWNED); \ (p)->p_cowgen++; \ } while (0) /* Check whether a thread is safe to be swapped out. */ #define thread_safetoswapout(td) ((td)->td_flags & TDF_CANSWAP) /* Control whether or not it is safe for curthread to sleep. */ #define THREAD_NO_SLEEPING() ((curthread)->td_no_sleeping++) #define THREAD_SLEEPING_OK() ((curthread)->td_no_sleeping--) #define THREAD_CAN_SLEEP() ((curthread)->td_no_sleeping == 0) #define PIDHASH(pid) (&pidhashtbl[(pid) & pidhash]) extern LIST_HEAD(pidhashhead, proc) *pidhashtbl; extern u_long pidhash; #define TIDHASH(tid) (&tidhashtbl[(tid) & tidhash]) extern LIST_HEAD(tidhashhead, thread) *tidhashtbl; extern u_long tidhash; extern struct rwlock tidhash_lock; #define PGRPHASH(pgid) (&pgrphashtbl[(pgid) & pgrphash]) extern LIST_HEAD(pgrphashhead, pgrp) *pgrphashtbl; extern u_long pgrphash; extern struct sx allproc_lock; extern int allproc_gen; extern struct sx proctree_lock; extern struct mtx ppeers_lock; extern struct proc proc0; /* Process slot for swapper. */ extern struct thread thread0; /* Primary thread in proc0. */ extern struct vmspace vmspace0; /* VM space for proc0. */ extern int hogticks; /* Limit on kernel cpu hogs. */ extern int lastpid; extern int nprocs, maxproc; /* Current and max number of procs. */ extern int maxprocperuid; /* Max procs per uid. */ extern u_long ps_arg_cache_limit; LIST_HEAD(proclist, proc); TAILQ_HEAD(procqueue, proc); TAILQ_HEAD(threadqueue, thread); extern struct proclist allproc; /* List of all processes. */ extern struct proclist zombproc; /* List of zombie processes. */ extern struct proc *initproc, *pageproc; /* Process slots for init, pager. */ extern struct uma_zone *proc_zone; struct proc *pfind(pid_t); /* Find process by id. */ struct proc *pfind_locked(pid_t pid); struct pgrp *pgfind(pid_t); /* Find process group by id. */ struct proc *zpfind(pid_t); /* Find zombie process by id. */ /* * pget() flags. */ #define PGET_HOLD 0x00001 /* Hold the process. */ #define PGET_CANSEE 0x00002 /* Check against p_cansee(). */ #define PGET_CANDEBUG 0x00004 /* Check against p_candebug(). */ #define PGET_ISCURRENT 0x00008 /* Check that the found process is current. */ #define PGET_NOTWEXIT 0x00010 /* Check that the process is not in P_WEXIT. */ #define PGET_NOTINEXEC 0x00020 /* Check that the process is not in P_INEXEC. */ #define PGET_NOTID 0x00040 /* Do not assume tid if pid > PID_MAX. */ #define PGET_WANTREAD (PGET_HOLD | PGET_CANDEBUG | PGET_NOTWEXIT) int pget(pid_t pid, int flags, struct proc **pp); void ast(struct trapframe *framep); struct thread *choosethread(void); int cr_cansignal(struct ucred *cred, struct proc *proc, int signum); int enterpgrp(struct proc *p, pid_t pgid, struct pgrp *pgrp, struct session *sess); int enterthispgrp(struct proc *p, struct pgrp *pgrp); void faultin(struct proc *p); void fixjobc(struct proc *p, struct pgrp *pgrp, int entering); int fork1(struct thread *, int, int, struct proc **, int *, int); void fork_exit(void (*)(void *, struct trapframe *), void *, struct trapframe *); void fork_return(struct thread *, struct trapframe *); int inferior(struct proc *p); void kern_yield(int); void kick_proc0(void); int leavepgrp(struct proc *p); int maybe_preempt(struct thread *td); void maybe_yield(void); void mi_switch(int flags, struct thread *newtd); int p_candebug(struct thread *td, struct proc *p); int p_cansee(struct thread *td, struct proc *p); int p_cansched(struct thread *td, struct proc *p); int p_cansignal(struct thread *td, struct proc *p, int signum); int p_canwait(struct thread *td, struct proc *p); struct pargs *pargs_alloc(int len); void pargs_drop(struct pargs *pa); void pargs_hold(struct pargs *pa); int proc_getargv(struct thread *td, struct proc *p, struct sbuf *sb); int proc_getauxv(struct thread *td, struct proc *p, struct sbuf *sb); int proc_getenvv(struct thread *td, struct proc *p, struct sbuf *sb); void procinit(void); void proc_linkup0(struct proc *p, struct thread *td); void proc_linkup(struct proc *p, struct thread *td); struct proc *proc_realparent(struct proc *child); void proc_reap(struct thread *td, struct proc *p, int *status, int options); void proc_reparent(struct proc *child, struct proc *newparent); struct pstats *pstats_alloc(void); void pstats_fork(struct pstats *src, struct pstats *dst); void pstats_free(struct pstats *ps); void reaper_abandon_children(struct proc *p, bool exiting); int securelevel_ge(struct ucred *cr, int level); int securelevel_gt(struct ucred *cr, int level); void sess_hold(struct session *); void sess_release(struct session *); int setrunnable(struct thread *); void setsugid(struct proc *p); int should_yield(void); int sigonstack(size_t sp); void stopevent(struct proc *, u_int, u_int); struct thread *tdfind(lwpid_t, pid_t); void threadinit(void); void tidhash_add(struct thread *); void tidhash_remove(struct thread *); void cpu_idle(int); int cpu_idle_wakeup(int); extern void (*cpu_idle_hook)(sbintime_t); /* Hook to machdep CPU idler. */ void cpu_switch(struct thread *, struct thread *, struct mtx *); void cpu_throw(struct thread *, struct thread *) __dead2; void unsleep(struct thread *); void userret(struct thread *, struct trapframe *); void cpu_exit(struct thread *); void exit1(struct thread *, int) __dead2; struct syscall_args; int cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa); void cpu_fork(struct thread *, struct proc *, struct thread *, int); void cpu_set_fork_handler(struct thread *, void (*)(void *), void *); void cpu_set_syscall_retval(struct thread *, int); void cpu_set_upcall(struct thread *td, struct thread *td0); void cpu_set_upcall_kse(struct thread *, void (*)(void *), void *, stack_t *); int cpu_set_user_tls(struct thread *, void *tls_base); void cpu_thread_alloc(struct thread *); void cpu_thread_clean(struct thread *); void cpu_thread_exit(struct thread *); void cpu_thread_free(struct thread *); void cpu_thread_swapin(struct thread *); void cpu_thread_swapout(struct thread *); struct thread *thread_alloc(int pages); int thread_alloc_stack(struct thread *, int pages); void thread_cow_get_proc(struct thread *newtd, struct proc *p); void thread_cow_get(struct thread *newtd, struct thread *td); void thread_cow_free(struct thread *td); void thread_cow_update(struct thread *td); void thread_exit(void) __dead2; void thread_free(struct thread *td); void thread_link(struct thread *td, struct proc *p); void thread_reap(void); int thread_single(struct proc *p, int how); void thread_single_end(struct proc *p, int how); void thread_stash(struct thread *td); void thread_stopped(struct proc *p); void childproc_stopped(struct proc *child, int reason); void childproc_continued(struct proc *child); void childproc_exited(struct proc *child); int thread_suspend_check(int how); bool thread_suspend_check_needed(void); void thread_suspend_switch(struct thread *, struct proc *p); void thread_suspend_one(struct thread *td); void thread_unlink(struct thread *td); void thread_unsuspend(struct proc *p); void thread_wait(struct proc *p); struct thread *thread_find(struct proc *p, lwpid_t tid); void stop_all_proc(void); void resume_all_proc(void); static __inline int curthread_pflags_set(int flags) { struct thread *td; int save; td = curthread; save = ~flags | (td->td_pflags & flags); td->td_pflags |= flags; return (save); } static __inline void curthread_pflags_restore(int save) { curthread->td_pflags &= save; } #endif /* _KERNEL */ #endif /* !_SYS_PROC_H_ */ Index: head/sys/sys/resourcevar.h =================================================================== --- head/sys/sys/resourcevar.h (revision 284214) +++ head/sys/sys/resourcevar.h (revision 284215) @@ -1,160 +1,165 @@ /*- * Copyright (c) 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. * * @(#)resourcevar.h 8.4 (Berkeley) 1/9/95 * $FreeBSD$ */ #ifndef _SYS_RESOURCEVAR_H_ #define _SYS_RESOURCEVAR_H_ #include #include #ifdef _KERNEL #include #include #endif /* * Kernel per-process accounting / statistics * (not necessarily resident except when running). * * Locking key: * b - created at fork, never changes * c - locked by proc mtx * k - only accessed by curthread * w - locked by proc itim lock * w2 - locked by proc prof lock */ struct pstats { #define pstat_startzero p_cru struct rusage p_cru; /* Stats for reaped children. */ struct itimerval p_timer[3]; /* (w) Virtual-time timers. */ #define pstat_endzero pstat_startcopy #define pstat_startcopy p_prof struct uprof { /* Profile arguments. */ caddr_t pr_base; /* (c + w2) Buffer base. */ u_long pr_size; /* (c + w2) Buffer size. */ u_long pr_off; /* (c + w2) PC offset. */ u_long pr_scale; /* (c + w2) PC scaling. */ } p_prof; #define pstat_endcopy p_start struct timeval p_start; /* (b) Starting time. */ }; #ifdef _KERNEL /* * Kernel shareable process resource limits. Because this structure * is moderately large but changes infrequently, it is normally * shared copy-on-write after forks. */ struct plimit { struct rlimit pl_rlimit[RLIM_NLIMITS]; int pl_refcnt; /* number of references */ }; struct racct; /*- * Per uid resource consumption. This structure is used to track * the total resource consumption (process count, socket buffer size, * etc) for the uid and impose limits. * * Locking guide: * (a) Constant from inception * (b) Lockless, updated using atomics * (c) Locked by global uihashtbl_mtx * (d) Locked by the ui_vmsize_mtx */ struct uidinfo { LIST_ENTRY(uidinfo) ui_hash; /* (c) hash chain of uidinfos */ struct mtx ui_vmsize_mtx; vm_ooffset_t ui_vmsize; /* (d) swap reservation by uid */ long ui_sbsize; /* (b) socket buffer space consumed */ long ui_proccnt; /* (b) number of processes */ long ui_ptscnt; /* (b) number of pseudo-terminals */ long ui_kqcnt; /* (b) number of kqueues */ uid_t ui_uid; /* (a) uid */ u_int ui_ref; /* (b) reference count */ #ifdef RACCT struct racct *ui_racct; /* (a) resource accounting */ #endif }; #define UIDINFO_VMSIZE_LOCK(ui) mtx_lock(&((ui)->ui_vmsize_mtx)) #define UIDINFO_VMSIZE_UNLOCK(ui) mtx_unlock(&((ui)->ui_vmsize_mtx)) struct proc; struct rusage_ext; struct thread; void addupc_intr(struct thread *td, uintfptr_t pc, u_int ticks); void addupc_task(struct thread *td, uintfptr_t pc, u_int ticks); void calccru(struct proc *p, struct timeval *up, struct timeval *sp); void calcru(struct proc *p, struct timeval *up, struct timeval *sp); int chgkqcnt(struct uidinfo *uip, int diff, rlim_t max); int chgproccnt(struct uidinfo *uip, int diff, rlim_t maxval); int chgsbsize(struct uidinfo *uip, u_int *hiwat, u_int to, rlim_t maxval); int chgptscnt(struct uidinfo *uip, int diff, rlim_t maxval); int fuswintr(void *base); int kern_proc_setrlimit(struct thread *td, struct proc *p, u_int which, struct rlimit *limp); struct plimit *lim_alloc(void); void lim_copy(struct plimit *dst, struct plimit *src); -rlim_t lim_cur(struct proc *p, int which); +rlim_t lim_cur(struct thread *td, int which); +rlim_t lim_cur_proc(struct proc *p, int which); void lim_fork(struct proc *p1, struct proc *p2); void lim_free(struct plimit *limp); struct plimit *lim_hold(struct plimit *limp); -rlim_t lim_max(struct proc *p, int which); -void lim_rlimit(struct proc *p, int which, struct rlimit *rlp); +rlim_t lim_max(struct thread *td, int which); +rlim_t lim_max_proc(struct proc *p, int which); +void lim_rlimit(struct thread *td, int which, struct rlimit *rlp); +void lim_rlimit_proc(struct proc *p, int which, struct rlimit *rlp); void ruadd(struct rusage *ru, struct rusage_ext *rux, struct rusage *ru2, struct rusage_ext *rux2); void rucollect(struct rusage *ru, struct rusage *ru2); void rufetch(struct proc *p, struct rusage *ru); void rufetchcalc(struct proc *p, struct rusage *ru, struct timeval *up, struct timeval *sp); void rufetchtd(struct thread *td, struct rusage *ru); void ruxagg(struct proc *p, struct thread *td); int suswintr(void *base, int word); struct uidinfo *uifind(uid_t uid); void uifree(struct uidinfo *uip); void uihashinit(void); void uihold(struct uidinfo *uip); #ifdef RACCT void ui_racct_foreach(void (*callback)(struct racct *racct, void *arg2, void *arg3), void *arg2, void *arg3); #endif + +void lim_update_thread(struct thread *td); #endif /* _KERNEL */ #endif /* !_SYS_RESOURCEVAR_H_ */ Index: head/sys/sys/vnode.h =================================================================== --- head/sys/sys/vnode.h (revision 284214) +++ head/sys/sys/vnode.h (revision 284215) @@ -1,869 +1,869 @@ /*- * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vnode.h 8.7 (Berkeley) 2/4/94 * $FreeBSD$ */ #ifndef _SYS_VNODE_H_ #define _SYS_VNODE_H_ #include #include #include #include #include #include #include #include #include #include /* * The vnode is the focus of all file activity in UNIX. There is a * unique vnode allocated for each active file, each current directory, * each mounted-on file, text file, and the root. */ /* * Vnode types. VNON means no type. */ enum vtype { VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO, VBAD, VMARKER }; /* * Each underlying filesystem allocates its own private area and hangs * it from v_data. If non-null, this area is freed in getnewvnode(). */ struct namecache; struct vpollinfo { struct mtx vpi_lock; /* lock to protect below */ struct selinfo vpi_selinfo; /* identity of poller(s) */ short vpi_events; /* what they are looking for */ short vpi_revents; /* what has happened */ }; /* * Reading or writing any of these items requires holding the appropriate lock. * * Lock reference: * c - namecache mutex * f - freelist mutex * i - interlock * m - mount point interlock * p - pollinfo lock * u - Only a reference to the vnode is needed to read. * v - vnode lock * * Vnodes may be found on many lists. The general way to deal with operating * on a vnode that is on a list is: * 1) Lock the list and find the vnode. * 2) Lock interlock so that the vnode does not go away. * 3) Unlock the list to avoid lock order reversals. * 4) vget with LK_INTERLOCK and check for ENOENT, or * 5) Check for DOOMED if the vnode lock is not required. * 6) Perform your operation, then vput(). */ #if defined(_KERNEL) || defined(_KVM_VNODE) struct vnode { /* * Fields which define the identity of the vnode. These fields are * owned by the filesystem (XXX: and vgone() ?) */ const char *v_tag; /* u type of underlying data */ struct vop_vector *v_op; /* u vnode operations vector */ void *v_data; /* u private data for fs */ /* * Filesystem instance stuff */ struct mount *v_mount; /* u ptr to vfs we are in */ TAILQ_ENTRY(vnode) v_nmntvnodes; /* m vnodes for mount point */ /* * Type specific fields, only one applies to any given vnode. * See #defines below for renaming to v_* namespace. */ union { struct mount *vu_mount; /* v ptr to mountpoint (VDIR) */ struct socket *vu_socket; /* v unix domain net (VSOCK) */ struct cdev *vu_cdev; /* v device (VCHR, VBLK) */ struct fifoinfo *vu_fifoinfo; /* v fifo (VFIFO) */ } v_un; /* * vfs_hash: (mount + inode) -> vnode hash. The hash value * itself is grouped with other int fields, to avoid padding. */ LIST_ENTRY(vnode) v_hashlist; /* * VFS_namecache stuff */ LIST_HEAD(, namecache) v_cache_src; /* c Cache entries from us */ TAILQ_HEAD(, namecache) v_cache_dst; /* c Cache entries to us */ struct namecache *v_cache_dd; /* c Cache entry for .. vnode */ /* * Locking */ struct lock v_lock; /* u (if fs don't have one) */ struct mtx v_interlock; /* lock for "i" things */ struct lock *v_vnlock; /* u pointer to vnode lock */ /* * The machinery of being a vnode */ TAILQ_ENTRY(vnode) v_actfreelist; /* f vnode active/free lists */ struct bufobj v_bufobj; /* * Buffer cache object */ /* * Hooks for various subsystems and features. */ struct vpollinfo *v_pollinfo; /* i Poll events, p for *v_pi */ struct label *v_label; /* MAC label for vnode */ struct lockf *v_lockf; /* Byte-level advisory lock list */ struct rangelock v_rl; /* Byte-range lock */ /* * clustering stuff */ daddr_t v_cstart; /* v start block of cluster */ daddr_t v_lasta; /* v last allocation */ daddr_t v_lastw; /* v last write */ int v_clen; /* v length of cur. cluster */ int v_holdcnt; /* i prevents recycling. */ int v_usecount; /* i ref count of users */ u_int v_iflag; /* i vnode flags (see below) */ u_int v_vflag; /* v vnode flags */ int v_writecount; /* v ref count of writers */ u_int v_hash; enum vtype v_type; /* u vnode type */ }; #endif /* defined(_KERNEL) || defined(_KVM_VNODE) */ #define v_mountedhere v_un.vu_mount #define v_socket v_un.vu_socket #define v_rdev v_un.vu_cdev #define v_fifoinfo v_un.vu_fifoinfo /* XXX: These are temporary to avoid a source sweep at this time */ #define v_object v_bufobj.bo_object /* * Userland version of struct vnode, for sysctl. */ struct xvnode { size_t xv_size; /* sizeof(struct xvnode) */ void *xv_vnode; /* address of real vnode */ u_long xv_flag; /* vnode vflags */ int xv_usecount; /* reference count of users */ int xv_writecount; /* reference count of writers */ int xv_holdcnt; /* page & buffer references */ u_long xv_id; /* capability identifier */ void *xv_mount; /* address of parent mount */ long xv_numoutput; /* num of writes in progress */ enum vtype xv_type; /* vnode type */ union { void *xvu_socket; /* socket, if VSOCK */ void *xvu_fifo; /* fifo, if VFIFO */ dev_t xvu_rdev; /* maj/min, if VBLK/VCHR */ struct { dev_t xvu_dev; /* device, if VDIR/VREG/VLNK */ ino_t xvu_ino; /* id, if VDIR/VREG/VLNK */ } xv_uns; } xv_un; }; #define xv_socket xv_un.xvu_socket #define xv_fifo xv_un.xvu_fifo #define xv_rdev xv_un.xvu_rdev #define xv_dev xv_un.xv_uns.xvu_dev #define xv_ino xv_un.xv_uns.xvu_ino /* We don't need to lock the knlist */ #define VN_KNLIST_EMPTY(vp) ((vp)->v_pollinfo == NULL || \ KNLIST_EMPTY(&(vp)->v_pollinfo->vpi_selinfo.si_note)) #define VN_KNOTE(vp, b, a) \ do { \ if (!VN_KNLIST_EMPTY(vp)) \ KNOTE(&vp->v_pollinfo->vpi_selinfo.si_note, (b), \ (a) | KNF_NOKQLOCK); \ } while (0) #define VN_KNOTE_LOCKED(vp, b) VN_KNOTE(vp, b, KNF_LISTLOCKED) #define VN_KNOTE_UNLOCKED(vp, b) VN_KNOTE(vp, b, 0) /* * Vnode flags. * VI flags are protected by interlock and live in v_iflag * VV flags are protected by the vnode lock and live in v_vflag * * VI_DOOMED is doubly protected by the interlock and vnode lock. Both * are required for writing but the status may be checked with either. */ #define VI_MOUNT 0x0020 /* Mount in progress */ #define VI_AGE 0x0040 /* Insert vnode at head of free list */ #define VI_DOOMED 0x0080 /* This vnode is being recycled */ #define VI_FREE 0x0100 /* This vnode is on the freelist */ #define VI_ACTIVE 0x0200 /* This vnode is on the active list */ #define VI_DOINGINACT 0x0800 /* VOP_INACTIVE is in progress */ #define VI_OWEINACT 0x1000 /* Need to call inactive */ #define VV_ROOT 0x0001 /* root of its filesystem */ #define VV_ISTTY 0x0002 /* vnode represents a tty */ #define VV_NOSYNC 0x0004 /* unlinked, stop syncing */ #define VV_ETERNALDEV 0x0008 /* device that is never destroyed */ #define VV_CACHEDLABEL 0x0010 /* Vnode has valid cached MAC label */ #define VV_TEXT 0x0020 /* vnode is a pure text prototype */ #define VV_COPYONWRITE 0x0040 /* vnode is doing copy-on-write */ #define VV_SYSTEM 0x0080 /* vnode being used by kernel */ #define VV_PROCDEP 0x0100 /* vnode is process dependent */ #define VV_NOKNOTE 0x0200 /* don't activate knotes on this vnode */ #define VV_DELETED 0x0400 /* should be removed */ #define VV_MD 0x0800 /* vnode backs the md device */ #define VV_FORCEINSMQ 0x1000 /* force the insmntque to succeed */ /* * Vnode attributes. A field value of VNOVAL represents a field whose value * is unavailable (getattr) or which is not to be changed (setattr). */ struct vattr { enum vtype va_type; /* vnode type (for create) */ u_short va_mode; /* files access mode and type */ short va_nlink; /* number of references to file */ uid_t va_uid; /* owner user id */ gid_t va_gid; /* owner group id */ dev_t va_fsid; /* filesystem id */ long va_fileid; /* file id */ u_quad_t va_size; /* file size in bytes */ long va_blocksize; /* blocksize preferred for i/o */ struct timespec va_atime; /* time of last access */ struct timespec va_mtime; /* time of last modification */ struct timespec va_ctime; /* time file changed */ struct timespec va_birthtime; /* time file created */ u_long va_gen; /* generation number of file */ u_long va_flags; /* flags defined for file */ dev_t va_rdev; /* device the special file represents */ u_quad_t va_bytes; /* bytes of disk space held by file */ u_quad_t va_filerev; /* file modification number */ u_int va_vaflags; /* operations flags, see below */ long va_spare; /* remain quad aligned */ }; /* * Flags for va_vaflags. */ #define VA_UTIMES_NULL 0x01 /* utimes argument was NULL */ #define VA_EXCLUSIVE 0x02 /* exclusive create request */ /* * Flags for ioflag. (high 16 bits used to ask for read-ahead and * help with write clustering) * NB: IO_NDELAY and IO_DIRECT are linked to fcntl.h */ #define IO_UNIT 0x0001 /* do I/O as atomic unit */ #define IO_APPEND 0x0002 /* append write to end */ #define IO_NDELAY 0x0004 /* FNDELAY flag set in file table */ #define IO_NODELOCKED 0x0008 /* underlying node already locked */ #define IO_ASYNC 0x0010 /* bawrite rather then bdwrite */ #define IO_VMIO 0x0020 /* data already in VMIO space */ #define IO_INVAL 0x0040 /* invalidate after I/O */ #define IO_SYNC 0x0080 /* do I/O synchronously */ #define IO_DIRECT 0x0100 /* attempt to bypass buffer cache */ #define IO_EXT 0x0400 /* operate on external attributes */ #define IO_NORMAL 0x0800 /* operate on regular data */ #define IO_NOMACCHECK 0x1000 /* MAC checks unnecessary */ #define IO_BUFLOCKED 0x2000 /* ffs flag; indir buf is locked */ #define IO_RANGELOCKED 0x4000 /* range locked */ #define IO_SEQMAX 0x7F /* seq heuristic max value */ #define IO_SEQSHIFT 16 /* seq heuristic in upper 16 bits */ /* * Flags for accmode_t. */ #define VEXEC 000000000100 /* execute/search permission */ #define VWRITE 000000000200 /* write permission */ #define VREAD 000000000400 /* read permission */ #define VADMIN 000000010000 /* being the file owner */ #define VAPPEND 000000040000 /* permission to write/append */ /* * VEXPLICIT_DENY makes VOP_ACCESSX(9) return EPERM or EACCES only * if permission was denied explicitly, by a "deny" rule in NFSv4 ACL, * and 0 otherwise. This never happens with ordinary unix access rights * or POSIX.1e ACLs. Obviously, VEXPLICIT_DENY must be OR-ed with * some other V* constant. */ #define VEXPLICIT_DENY 000000100000 #define VREAD_NAMED_ATTRS 000000200000 /* not used */ #define VWRITE_NAMED_ATTRS 000000400000 /* not used */ #define VDELETE_CHILD 000001000000 #define VREAD_ATTRIBUTES 000002000000 /* permission to stat(2) */ #define VWRITE_ATTRIBUTES 000004000000 /* change {m,c,a}time */ #define VDELETE 000010000000 #define VREAD_ACL 000020000000 /* read ACL and file mode */ #define VWRITE_ACL 000040000000 /* change ACL and/or file mode */ #define VWRITE_OWNER 000100000000 /* change file owner */ #define VSYNCHRONIZE 000200000000 /* not used */ #define VCREAT 000400000000 /* creating new file */ #define VVERIFY 001000000000 /* verification required */ /* * Permissions that were traditionally granted only to the file owner. */ #define VADMIN_PERMS (VADMIN | VWRITE_ATTRIBUTES | VWRITE_ACL | \ VWRITE_OWNER) /* * Permissions that were traditionally granted to everyone. */ #define VSTAT_PERMS (VREAD_ATTRIBUTES | VREAD_ACL) /* * Permissions that allow to change the state of the file in any way. */ #define VMODIFY_PERMS (VWRITE | VAPPEND | VADMIN_PERMS | VDELETE_CHILD | \ VDELETE) /* * Token indicating no attribute value yet assigned. */ #define VNOVAL (-1) /* * LK_TIMELOCK timeout for vnode locks (used mainly by the pageout daemon) */ #define VLKTIMEOUT (hz / 20 + 1) #ifdef _KERNEL #ifdef MALLOC_DECLARE MALLOC_DECLARE(M_VNODE); #endif /* * Convert between vnode types and inode formats (since POSIX.1 * defines mode word of stat structure in terms of inode formats). */ extern enum vtype iftovt_tab[]; extern int vttoif_tab[]; #define IFTOVT(mode) (iftovt_tab[((mode) & S_IFMT) >> 12]) #define VTTOIF(indx) (vttoif_tab[(int)(indx)]) #define MAKEIMODE(indx, mode) (int)(VTTOIF(indx) | (mode)) /* * Flags to various vnode functions. */ #define SKIPSYSTEM 0x0001 /* vflush: skip vnodes marked VSYSTEM */ #define FORCECLOSE 0x0002 /* vflush: force file closure */ #define WRITECLOSE 0x0004 /* vflush: only close writable files */ #define EARLYFLUSH 0x0008 /* vflush: early call for ffs_flushfiles */ #define V_SAVE 0x0001 /* vinvalbuf: sync file first */ #define V_ALT 0x0002 /* vinvalbuf: invalidate only alternate bufs */ #define V_NORMAL 0x0004 /* vinvalbuf: invalidate only regular bufs */ #define V_CLEANONLY 0x0008 /* vinvalbuf: invalidate only clean bufs */ #define REVOKEALL 0x0001 /* vop_revoke: revoke all aliases */ #define V_WAIT 0x0001 /* vn_start_write: sleep for suspend */ #define V_NOWAIT 0x0002 /* vn_start_write: don't sleep for suspend */ #define V_XSLEEP 0x0004 /* vn_start_write: just return after sleep */ #define V_MNTREF 0x0010 /* vn_start_write: mp is already ref-ed */ #define VR_START_WRITE 0x0001 /* vfs_write_resume: start write atomically */ #define VR_NO_SUSPCLR 0x0002 /* vfs_write_resume: do not clear suspension */ #define VS_SKIP_UNMOUNT 0x0001 /* vfs_write_suspend: fail if the filesystem is being unmounted */ #define VREF(vp) vref(vp) #ifdef DIAGNOSTIC #define VATTR_NULL(vap) vattr_null(vap) #else #define VATTR_NULL(vap) (*(vap) = va_null) /* initialize a vattr */ #endif /* DIAGNOSTIC */ #define NULLVP ((struct vnode *)NULL) /* * Global vnode data. */ extern struct vnode *rootvnode; /* root (i.e. "/") vnode */ extern int async_io_version; /* 0 or POSIX version of AIO i'face */ extern int desiredvnodes; /* number of vnodes desired */ extern struct uma_zone *namei_zone; extern struct vattr va_null; /* predefined null vattr structure */ #define VI_LOCK(vp) mtx_lock(&(vp)->v_interlock) #define VI_LOCK_FLAGS(vp, flags) mtx_lock_flags(&(vp)->v_interlock, (flags)) #define VI_TRYLOCK(vp) mtx_trylock(&(vp)->v_interlock) #define VI_UNLOCK(vp) mtx_unlock(&(vp)->v_interlock) #define VI_MTX(vp) (&(vp)->v_interlock) #define VN_LOCK_AREC(vp) lockallowrecurse((vp)->v_vnlock) #define VN_LOCK_ASHARE(vp) lockallowshare((vp)->v_vnlock) #define VN_LOCK_DSHARE(vp) lockdisableshare((vp)->v_vnlock) #endif /* _KERNEL */ /* * Mods for extensibility. */ /* * Flags for vdesc_flags: */ #define VDESC_MAX_VPS 16 /* Low order 16 flag bits are reserved for willrele flags for vp arguments. */ #define VDESC_VP0_WILLRELE 0x0001 #define VDESC_VP1_WILLRELE 0x0002 #define VDESC_VP2_WILLRELE 0x0004 #define VDESC_VP3_WILLRELE 0x0008 #define VDESC_NOMAP_VPP 0x0100 #define VDESC_VPP_WILLRELE 0x0200 /* * A generic structure. * This can be used by bypass routines to identify generic arguments. */ struct vop_generic_args { struct vnodeop_desc *a_desc; /* other random data follows, presumably */ }; typedef int vop_bypass_t(struct vop_generic_args *); /* * VDESC_NO_OFFSET is used to identify the end of the offset list * and in places where no such field exists. */ #define VDESC_NO_OFFSET -1 /* * This structure describes the vnode operation taking place. */ struct vnodeop_desc { char *vdesc_name; /* a readable name for debugging */ int vdesc_flags; /* VDESC_* flags */ vop_bypass_t *vdesc_call; /* Function to call */ /* * These ops are used by bypass routines to map and locate arguments. * Creds and procs are not needed in bypass routines, but sometimes * they are useful to (for example) transport layers. * Nameidata is useful because it has a cred in it. */ int *vdesc_vp_offsets; /* list ended by VDESC_NO_OFFSET */ int vdesc_vpp_offset; /* return vpp location */ int vdesc_cred_offset; /* cred location, if any */ int vdesc_thread_offset; /* thread location, if any */ int vdesc_componentname_offset; /* if any */ }; #ifdef _KERNEL /* * A list of all the operation descs. */ extern struct vnodeop_desc *vnodeop_descs[]; #define VOPARG_OFFSETOF(s_type, field) __offsetof(s_type, field) #define VOPARG_OFFSETTO(s_type, s_offset, struct_p) \ ((s_type)(((char*)(struct_p)) + (s_offset))) #ifdef DEBUG_VFS_LOCKS /* * Support code to aid in debugging VFS locking problems. Not totally * reliable since if the thread sleeps between changing the lock * state and checking it with the assert, some other thread could * change the state. They are good enough for debugging a single * filesystem using a single-threaded test. Note that the unreliability is * limited to false negatives; efforts were made to ensure that false * positives cannot occur. */ void assert_vi_locked(struct vnode *vp, const char *str); void assert_vi_unlocked(struct vnode *vp, const char *str); void assert_vop_elocked(struct vnode *vp, const char *str); #if 0 void assert_vop_elocked_other(struct vnode *vp, const char *str); #endif void assert_vop_locked(struct vnode *vp, const char *str); #if 0 voi0 assert_vop_slocked(struct vnode *vp, const char *str); #endif void assert_vop_unlocked(struct vnode *vp, const char *str); #define ASSERT_VI_LOCKED(vp, str) assert_vi_locked((vp), (str)) #define ASSERT_VI_UNLOCKED(vp, str) assert_vi_unlocked((vp), (str)) #define ASSERT_VOP_ELOCKED(vp, str) assert_vop_elocked((vp), (str)) #if 0 #define ASSERT_VOP_ELOCKED_OTHER(vp, str) assert_vop_locked_other((vp), (str)) #endif #define ASSERT_VOP_LOCKED(vp, str) assert_vop_locked((vp), (str)) #if 0 #define ASSERT_VOP_SLOCKED(vp, str) assert_vop_slocked((vp), (str)) #endif #define ASSERT_VOP_UNLOCKED(vp, str) assert_vop_unlocked((vp), (str)) #else /* !DEBUG_VFS_LOCKS */ #define ASSERT_VI_LOCKED(vp, str) ((void)0) #define ASSERT_VI_UNLOCKED(vp, str) ((void)0) #define ASSERT_VOP_ELOCKED(vp, str) ((void)0) #if 0 #define ASSERT_VOP_ELOCKED_OTHER(vp, str) #endif #define ASSERT_VOP_LOCKED(vp, str) ((void)0) #if 0 #define ASSERT_VOP_SLOCKED(vp, str) #endif #define ASSERT_VOP_UNLOCKED(vp, str) ((void)0) #endif /* DEBUG_VFS_LOCKS */ /* * This call works for vnodes in the kernel. */ #define VCALL(c) ((c)->a_desc->vdesc_call(c)) #define DOINGASYNC(vp) \ (((vp)->v_mount->mnt_kern_flag & MNTK_ASYNC) != 0 && \ ((curthread->td_pflags & TDP_SYNCIO) == 0)) /* * VMIO support inline */ extern int vmiodirenable; static __inline int vn_canvmio(struct vnode *vp) { if (vp && (vp->v_type == VREG || (vmiodirenable && vp->v_type == VDIR))) return(TRUE); return(FALSE); } /* * Finally, include the default set of vnode operations. */ typedef void vop_getpages_iodone_t(void *, vm_page_t *, int, int); #include "vnode_if.h" /* vn_open_flags */ #define VN_OPEN_NOAUDIT 0x00000001 #define VN_OPEN_NOCAPCHECK 0x00000002 #define VN_OPEN_NAMECACHE 0x00000004 /* * Public vnode manipulation functions. */ struct componentname; struct file; struct mount; struct nameidata; struct ostat; struct thread; struct proc; struct stat; struct nstat; struct ucred; struct uio; struct vattr; struct vnode; typedef int (*vn_get_ino_t)(struct mount *, void *, int, struct vnode **); /* cache_* may belong in namei.h. */ #define cache_enter(dvp, vp, cnp) \ cache_enter_time(dvp, vp, cnp, NULL, NULL) void cache_enter_time(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, struct timespec *tsp, struct timespec *dtsp); int cache_lookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct timespec *tsp, int *ticksp); void cache_purge(struct vnode *vp); void cache_purge_negative(struct vnode *vp); void cache_purgevfs(struct mount *mp); int change_dir(struct vnode *vp, struct thread *td); int change_root(struct vnode *vp, struct thread *td); void cvtstat(struct stat *st, struct ostat *ost); void cvtnstat(struct stat *sb, struct nstat *nsb); int getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops, struct vnode **vpp); void getnewvnode_reserve(u_int count); void getnewvnode_drop_reserve(void); int insmntque1(struct vnode *vp, struct mount *mp, void (*dtr)(struct vnode *, void *), void *dtr_arg); int insmntque(struct vnode *vp, struct mount *mp); u_quad_t init_va_filerev(void); int speedup_syncer(void); int vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen); #define textvp_fullpath(p, rb, rfb) \ vn_fullpath(FIRST_THREAD_IN_PROC(p), (p)->p_textvp, rb, rfb) int vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf); int vn_fullpath_global(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf); struct vnode * vn_dir_dd_ino(struct vnode *vp); int vn_commname(struct vnode *vn, char *buf, u_int buflen); int vn_path_to_global_path(struct thread *td, struct vnode *vp, char *path, u_int pathlen); int vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid, accmode_t accmode, struct ucred *cred, int *privused); int vaccess_acl_nfs4(enum vtype type, uid_t file_uid, gid_t file_gid, struct acl *aclp, accmode_t accmode, struct ucred *cred, int *privused); int vaccess_acl_posix1e(enum vtype type, uid_t file_uid, gid_t file_gid, struct acl *acl, accmode_t accmode, struct ucred *cred, int *privused); void vattr_null(struct vattr *vap); int vcount(struct vnode *vp); void vdrop(struct vnode *); void vdropl(struct vnode *); int vflush(struct mount *mp, int rootrefs, int flags, struct thread *td); int vget(struct vnode *vp, int lockflag, struct thread *td); void vgone(struct vnode *vp); void vhold(struct vnode *); void vholdl(struct vnode *); void vinactive(struct vnode *, struct thread *); int vinvalbuf(struct vnode *vp, int save, int slpflag, int slptimeo); int vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize); void vunref(struct vnode *); void vn_printf(struct vnode *vp, const char *fmt, ...) __printflike(2,3); #define vprint(label, vp) vn_printf((vp), "%s\n", (label)) int vrecycle(struct vnode *vp); int vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred); int vn_close(struct vnode *vp, int flags, struct ucred *file_cred, struct thread *td); void vn_finished_write(struct mount *mp); void vn_finished_secondary_write(struct mount *mp); int vn_isdisk(struct vnode *vp, int *errp); int _vn_lock(struct vnode *vp, int flags, char *file, int line); #define vn_lock(vp, flags) _vn_lock(vp, flags, __FILE__, __LINE__) int vn_open(struct nameidata *ndp, int *flagp, int cmode, struct file *fp); int vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags, struct ucred *cred, struct file *fp); int vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred, struct thread *td, struct file *fp); void vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end); int vn_pollrecord(struct vnode *vp, struct thread *p, int events); int vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset, enum uio_seg segflg, int ioflg, struct ucred *active_cred, struct ucred *file_cred, ssize_t *aresid, struct thread *td); int vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, void *base, size_t len, off_t offset, enum uio_seg segflg, int ioflg, struct ucred *active_cred, struct ucred *file_cred, size_t *aresid, struct thread *td); int vn_rlimit_fsize(const struct vnode *vn, const struct uio *uio, - const struct thread *td); + struct thread *td); int vn_stat(struct vnode *vp, struct stat *sb, struct ucred *active_cred, struct ucred *file_cred, struct thread *td); int vn_start_write(struct vnode *vp, struct mount **mpp, int flags); int vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags); int vn_writechk(struct vnode *vp); int vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, int *buflen, char *buf, struct thread *td); int vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, int buflen, char *buf, struct thread *td); int vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, struct thread *td); int vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp); int vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg, int lkflags, struct vnode **rvp); int vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred, struct thread *td); int vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio); int vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize, struct uio *uio); #define vn_rangelock_unlock(vp, cookie) \ rangelock_unlock(&(vp)->v_rl, (cookie), VI_MTX(vp)) #define vn_rangelock_unlock_range(vp, cookie, start, end) \ rangelock_unlock_range(&(vp)->v_rl, (cookie), (start), (end), \ VI_MTX(vp)) #define vn_rangelock_rlock(vp, start, end) \ rangelock_rlock(&(vp)->v_rl, (start), (end), VI_MTX(vp)) #define vn_rangelock_wlock(vp, start, end) \ rangelock_wlock(&(vp)->v_rl, (start), (end), VI_MTX(vp)) int vfs_cache_lookup(struct vop_lookup_args *ap); void vfs_timestamp(struct timespec *); void vfs_write_resume(struct mount *mp, int flags); int vfs_write_suspend(struct mount *mp, int flags); int vfs_write_suspend_umnt(struct mount *mp); int vop_stdbmap(struct vop_bmap_args *); int vop_stdfsync(struct vop_fsync_args *); int vop_stdgetwritemount(struct vop_getwritemount_args *); int vop_stdgetpages(struct vop_getpages_args *); int vop_stdinactive(struct vop_inactive_args *); int vop_stdislocked(struct vop_islocked_args *); int vop_stdkqfilter(struct vop_kqfilter_args *); int vop_stdlock(struct vop_lock1_args *); int vop_stdputpages(struct vop_putpages_args *); int vop_stdunlock(struct vop_unlock_args *); int vop_nopoll(struct vop_poll_args *); int vop_stdaccess(struct vop_access_args *ap); int vop_stdaccessx(struct vop_accessx_args *ap); int vop_stdadvise(struct vop_advise_args *ap); int vop_stdadvlock(struct vop_advlock_args *ap); int vop_stdadvlockasync(struct vop_advlockasync_args *ap); int vop_stdadvlockpurge(struct vop_advlockpurge_args *ap); int vop_stdallocate(struct vop_allocate_args *ap); int vop_stdpathconf(struct vop_pathconf_args *); int vop_stdpoll(struct vop_poll_args *); int vop_stdvptocnp(struct vop_vptocnp_args *ap); int vop_stdvptofh(struct vop_vptofh_args *ap); int vop_stdunp_bind(struct vop_unp_bind_args *ap); int vop_stdunp_connect(struct vop_unp_connect_args *ap); int vop_stdunp_detach(struct vop_unp_detach_args *ap); int vop_eopnotsupp(struct vop_generic_args *ap); int vop_ebadf(struct vop_generic_args *ap); int vop_einval(struct vop_generic_args *ap); int vop_enoent(struct vop_generic_args *ap); int vop_enotty(struct vop_generic_args *ap); int vop_null(struct vop_generic_args *ap); int vop_panic(struct vop_generic_args *ap); int dead_poll(struct vop_poll_args *ap); int dead_read(struct vop_read_args *ap); int dead_write(struct vop_write_args *ap); /* These are called from within the actual VOPS. */ void vop_create_post(void *a, int rc); void vop_deleteextattr_post(void *a, int rc); void vop_link_post(void *a, int rc); void vop_lock_pre(void *a); void vop_lock_post(void *a, int rc); void vop_lookup_post(void *a, int rc); void vop_lookup_pre(void *a); void vop_mkdir_post(void *a, int rc); void vop_mknod_post(void *a, int rc); void vop_remove_post(void *a, int rc); void vop_rename_post(void *a, int rc); void vop_rename_pre(void *a); void vop_rmdir_post(void *a, int rc); void vop_setattr_post(void *a, int rc); void vop_setextattr_post(void *a, int rc); void vop_strategy_pre(void *a); void vop_symlink_post(void *a, int rc); void vop_unlock_post(void *a, int rc); void vop_unlock_pre(void *a); void vop_rename_fail(struct vop_rename_args *ap); #define VOP_WRITE_PRE(ap) \ struct vattr va; \ int error, osize, ooffset, noffset; \ \ osize = ooffset = noffset = 0; \ if (!VN_KNLIST_EMPTY((ap)->a_vp)) { \ error = VOP_GETATTR((ap)->a_vp, &va, (ap)->a_cred); \ if (error) \ return (error); \ ooffset = (ap)->a_uio->uio_offset; \ osize = va.va_size; \ } #define VOP_WRITE_POST(ap, ret) \ noffset = (ap)->a_uio->uio_offset; \ if (noffset > ooffset && !VN_KNLIST_EMPTY((ap)->a_vp)) { \ VFS_KNOTE_LOCKED((ap)->a_vp, NOTE_WRITE \ | (noffset > osize ? NOTE_EXTEND : 0)); \ } #define VOP_LOCK(vp, flags) VOP_LOCK1(vp, flags, __FILE__, __LINE__) void vput(struct vnode *vp); void vrele(struct vnode *vp); void vref(struct vnode *vp); int vrefcnt(struct vnode *vp); void v_addpollinfo(struct vnode *vp); int vnode_create_vobject(struct vnode *vp, off_t size, struct thread *td); void vnode_destroy_vobject(struct vnode *vp); extern struct vop_vector fifo_specops; extern struct vop_vector dead_vnodeops; extern struct vop_vector default_vnodeops; extern struct vop_vector devfs_specops; #define VOP_PANIC ((void*)(uintptr_t)vop_panic) #define VOP_NULL ((void*)(uintptr_t)vop_null) #define VOP_EBADF ((void*)(uintptr_t)vop_ebadf) #define VOP_ENOTTY ((void*)(uintptr_t)vop_enotty) #define VOP_EINVAL ((void*)(uintptr_t)vop_einval) #define VOP_ENOENT ((void*)(uintptr_t)vop_enoent) #define VOP_EOPNOTSUPP ((void*)(uintptr_t)vop_eopnotsupp) /* fifo_vnops.c */ int fifo_printinfo(struct vnode *); /* vfs_hash.c */ typedef int vfs_hash_cmp_t(struct vnode *vp, void *arg); int vfs_hash_get(const struct mount *mp, u_int hash, int flags, struct thread *td, struct vnode **vpp, vfs_hash_cmp_t *fn, void *arg); u_int vfs_hash_index(struct vnode *vp); int vfs_hash_insert(struct vnode *vp, u_int hash, int flags, struct thread *td, struct vnode **vpp, vfs_hash_cmp_t *fn, void *arg); void vfs_hash_rehash(struct vnode *vp, u_int hash); void vfs_hash_remove(struct vnode *vp); int vfs_kqfilter(struct vop_kqfilter_args *); void vfs_mark_atime(struct vnode *vp, struct ucred *cred); struct dirent; int vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off); int vfs_unixify_accmode(accmode_t *accmode); void vfs_unp_reclaim(struct vnode *vp); int setfmode(struct thread *td, struct ucred *cred, struct vnode *vp, int mode); int setfown(struct thread *td, struct ucred *cred, struct vnode *vp, uid_t uid, gid_t gid); int vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td); int vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, struct thread *td); #endif /* _KERNEL */ #endif /* !_SYS_VNODE_H_ */ Index: head/sys/vm/swap_pager.c =================================================================== --- head/sys/vm/swap_pager.c (revision 284214) +++ head/sys/vm/swap_pager.c (revision 284215) @@ -1,2842 +1,2840 @@ /*- * Copyright (c) 1998 Matthew Dillon, * Copyright (c) 1994 John S. Dyson * Copyright (c) 1990 University of Utah. * 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 * the Systems Programming Group of the University of Utah Computer * Science Department. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Berkeley and its contributors. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * New Swap System * Matthew Dillon * * Radix Bitmap 'blists'. * * - The new swapper uses the new radix bitmap code. This should scale * to arbitrarily small or arbitrarily large swap spaces and an almost * arbitrary degree of fragmentation. * * Features: * * - on the fly reallocation of swap during putpages. The new system * does not try to keep previously allocated swap blocks for dirty * pages. * * - on the fly deallocation of swap * * - No more garbage collection required. Unnecessarily allocated swap * blocks only exist for dirty vm_page_t's now and these are already * cycled (in a high-load system) by the pager. We also do on-the-fly * removal of invalidated swap blocks when a page is destroyed * or renamed. * * from: Utah $Hdr: swap_pager.c 1.4 91/04/30$ * * @(#)swap_pager.c 8.9 (Berkeley) 3/21/94 * @(#)vm_swap.c 8.5 (Berkeley) 2/17/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_swap.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 #include #include #include #include #include #include /* * SWB_NPAGES must be a power of 2. It may be set to 1, 2, 4, 8, 16 * or 32 pages per allocation. * The 32-page limit is due to the radix code (kern/subr_blist.c). */ #ifndef MAX_PAGEOUT_CLUSTER #define MAX_PAGEOUT_CLUSTER 16 #endif #if !defined(SWB_NPAGES) #define SWB_NPAGES MAX_PAGEOUT_CLUSTER #endif /* * The swblock structure maps an object and a small, fixed-size range * of page indices to disk addresses within a swap area. * The collection of these mappings is implemented as a hash table. * Unused disk addresses within a swap area are allocated and managed * using a blist. */ #define SWCORRECT(n) (sizeof(void *) * (n) / sizeof(daddr_t)) #define SWAP_META_PAGES (SWB_NPAGES * 2) #define SWAP_META_MASK (SWAP_META_PAGES - 1) struct swblock { struct swblock *swb_hnext; vm_object_t swb_object; vm_pindex_t swb_index; int swb_count; daddr_t swb_pages[SWAP_META_PAGES]; }; static MALLOC_DEFINE(M_VMPGDATA, "vm_pgdata", "swap pager private data"); static struct mtx sw_dev_mtx; static TAILQ_HEAD(, swdevt) swtailq = TAILQ_HEAD_INITIALIZER(swtailq); static struct swdevt *swdevhd; /* Allocate from here next */ static int nswapdev; /* Number of swap devices */ int swap_pager_avail; static int swdev_syscall_active = 0; /* serialize swap(on|off) */ static vm_ooffset_t swap_total; SYSCTL_QUAD(_vm, OID_AUTO, swap_total, CTLFLAG_RD, &swap_total, 0, "Total amount of available swap storage."); static vm_ooffset_t swap_reserved; SYSCTL_QUAD(_vm, OID_AUTO, swap_reserved, CTLFLAG_RD, &swap_reserved, 0, "Amount of swap storage needed to back all allocated anonymous memory."); static int overcommit = 0; SYSCTL_INT(_vm, OID_AUTO, overcommit, CTLFLAG_RW, &overcommit, 0, "Configure virtual memory overcommit behavior. See tuning(7) " "for details."); static unsigned long swzone; SYSCTL_ULONG(_vm, OID_AUTO, swzone, CTLFLAG_RD, &swzone, 0, "Actual size of swap metadata zone"); static unsigned long swap_maxpages; SYSCTL_ULONG(_vm, OID_AUTO, swap_maxpages, CTLFLAG_RD, &swap_maxpages, 0, "Maximum amount of swap supported"); /* bits from overcommit */ #define SWAP_RESERVE_FORCE_ON (1 << 0) #define SWAP_RESERVE_RLIMIT_ON (1 << 1) #define SWAP_RESERVE_ALLOW_NONWIRED (1 << 2) int swap_reserve(vm_ooffset_t incr) { return (swap_reserve_by_cred(incr, curthread->td_ucred)); } int swap_reserve_by_cred(vm_ooffset_t incr, struct ucred *cred) { vm_ooffset_t r, s; int res, error; static int curfail; static struct timeval lastfail; struct uidinfo *uip; uip = cred->cr_ruidinfo; if (incr & PAGE_MASK) panic("swap_reserve: & PAGE_MASK"); #ifdef RACCT if (racct_enable) { PROC_LOCK(curproc); error = racct_add(curproc, RACCT_SWAP, incr); PROC_UNLOCK(curproc); if (error != 0) return (0); } #endif res = 0; mtx_lock(&sw_dev_mtx); r = swap_reserved + incr; if (overcommit & SWAP_RESERVE_ALLOW_NONWIRED) { s = vm_cnt.v_page_count - vm_cnt.v_free_reserved - vm_cnt.v_wire_count; s *= PAGE_SIZE; } else s = 0; s += swap_total; if ((overcommit & SWAP_RESERVE_FORCE_ON) == 0 || r <= s || (error = priv_check(curthread, PRIV_VM_SWAP_NOQUOTA)) == 0) { res = 1; swap_reserved = r; } mtx_unlock(&sw_dev_mtx); if (res) { - PROC_LOCK(curproc); UIDINFO_VMSIZE_LOCK(uip); if ((overcommit & SWAP_RESERVE_RLIMIT_ON) != 0 && - uip->ui_vmsize + incr > lim_cur(curproc, RLIMIT_SWAP) && + uip->ui_vmsize + incr > lim_cur(curthread, RLIMIT_SWAP) && priv_check(curthread, PRIV_VM_SWAP_NORLIMIT)) res = 0; else uip->ui_vmsize += incr; UIDINFO_VMSIZE_UNLOCK(uip); - PROC_UNLOCK(curproc); if (!res) { mtx_lock(&sw_dev_mtx); swap_reserved -= incr; mtx_unlock(&sw_dev_mtx); } } if (!res && ppsratecheck(&lastfail, &curfail, 1)) { printf("uid %d, pid %d: swap reservation for %jd bytes failed\n", uip->ui_uid, curproc->p_pid, incr); } #ifdef RACCT if (!res) { PROC_LOCK(curproc); racct_sub(curproc, RACCT_SWAP, incr); PROC_UNLOCK(curproc); } #endif return (res); } void swap_reserve_force(vm_ooffset_t incr) { struct uidinfo *uip; mtx_lock(&sw_dev_mtx); swap_reserved += incr; mtx_unlock(&sw_dev_mtx); #ifdef RACCT PROC_LOCK(curproc); racct_add_force(curproc, RACCT_SWAP, incr); PROC_UNLOCK(curproc); #endif uip = curthread->td_ucred->cr_ruidinfo; PROC_LOCK(curproc); UIDINFO_VMSIZE_LOCK(uip); uip->ui_vmsize += incr; UIDINFO_VMSIZE_UNLOCK(uip); PROC_UNLOCK(curproc); } void swap_release(vm_ooffset_t decr) { struct ucred *cred; PROC_LOCK(curproc); cred = curthread->td_ucred; swap_release_by_cred(decr, cred); PROC_UNLOCK(curproc); } void swap_release_by_cred(vm_ooffset_t decr, struct ucred *cred) { struct uidinfo *uip; uip = cred->cr_ruidinfo; if (decr & PAGE_MASK) panic("swap_release: & PAGE_MASK"); mtx_lock(&sw_dev_mtx); if (swap_reserved < decr) panic("swap_reserved < decr"); swap_reserved -= decr; mtx_unlock(&sw_dev_mtx); UIDINFO_VMSIZE_LOCK(uip); if (uip->ui_vmsize < decr) printf("negative vmsize for uid = %d\n", uip->ui_uid); uip->ui_vmsize -= decr; UIDINFO_VMSIZE_UNLOCK(uip); racct_sub_cred(cred, RACCT_SWAP, decr); } static void swapdev_strategy(struct buf *, struct swdevt *sw); #define SWM_FREE 0x02 /* free, period */ #define SWM_POP 0x04 /* pop out */ int swap_pager_full = 2; /* swap space exhaustion (task killing) */ static int swap_pager_almost_full = 1; /* swap space exhaustion (w/hysteresis)*/ static int nsw_rcount; /* free read buffers */ static int nsw_wcount_sync; /* limit write buffers / synchronous */ static int nsw_wcount_async; /* limit write buffers / asynchronous */ static int nsw_wcount_async_max;/* assigned maximum */ static int nsw_cluster_max; /* maximum VOP I/O allowed */ static int sysctl_swap_async_max(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_vm, OID_AUTO, swap_async_max, CTLTYPE_INT | CTLFLAG_RW, NULL, 0, sysctl_swap_async_max, "I", "Maximum running async swap ops"); static struct swblock **swhash; static int swhash_mask; static struct mtx swhash_mtx; static struct sx sw_alloc_sx; /* * "named" and "unnamed" anon region objects. Try to reduce the overhead * of searching a named list by hashing it just a little. */ #define NOBJLISTS 8 #define NOBJLIST(handle) \ (&swap_pager_object_list[((int)(intptr_t)handle >> 4) & (NOBJLISTS-1)]) static struct mtx sw_alloc_mtx; /* protect list manipulation */ static struct pagerlst swap_pager_object_list[NOBJLISTS]; static uma_zone_t swap_zone; /* * pagerops for OBJT_SWAP - "swap pager". Some ops are also global procedure * calls hooked from other parts of the VM system and do not appear here. * (see vm/swap_pager.h). */ static vm_object_t swap_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t offset, struct ucred *); static void swap_pager_dealloc(vm_object_t object); static int swap_pager_getpages(vm_object_t, vm_page_t *, int, int); static int swap_pager_getpages_async(vm_object_t, vm_page_t *, int, int, pgo_getpages_iodone_t, void *); static void swap_pager_putpages(vm_object_t, vm_page_t *, int, boolean_t, int *); static boolean_t swap_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *before, int *after); static void swap_pager_init(void); static void swap_pager_unswapped(vm_page_t); static void swap_pager_swapoff(struct swdevt *sp); struct pagerops swappagerops = { .pgo_init = swap_pager_init, /* early system initialization of pager */ .pgo_alloc = swap_pager_alloc, /* allocate an OBJT_SWAP object */ .pgo_dealloc = swap_pager_dealloc, /* deallocate an OBJT_SWAP object */ .pgo_getpages = swap_pager_getpages, /* pagein */ .pgo_getpages_async = swap_pager_getpages_async, /* pagein (async) */ .pgo_putpages = swap_pager_putpages, /* pageout */ .pgo_haspage = swap_pager_haspage, /* get backing store status for page */ .pgo_pageunswapped = swap_pager_unswapped, /* remove swap related to page */ }; /* * dmmax is in page-sized chunks with the new swap system. It was * dev-bsized chunks in the old. dmmax is always a power of 2. * * swap_*() routines are externally accessible. swp_*() routines are * internal. */ static int dmmax; static int nswap_lowat = 128; /* in pages, swap_pager_almost_full warn */ static int nswap_hiwat = 512; /* in pages, swap_pager_almost_full warn */ SYSCTL_INT(_vm, OID_AUTO, dmmax, CTLFLAG_RD, &dmmax, 0, "Maximum size of a swap block"); static void swp_sizecheck(void); static void swp_pager_async_iodone(struct buf *bp); static int swapongeom(struct thread *, struct vnode *); static int swaponvp(struct thread *, struct vnode *, u_long); static int swapoff_one(struct swdevt *sp, struct ucred *cred); /* * Swap bitmap functions */ static void swp_pager_freeswapspace(daddr_t blk, int npages); static daddr_t swp_pager_getswapspace(int npages); /* * Metadata functions */ static struct swblock **swp_pager_hash(vm_object_t object, vm_pindex_t index); static void swp_pager_meta_build(vm_object_t, vm_pindex_t, daddr_t); static void swp_pager_meta_free(vm_object_t, vm_pindex_t, daddr_t); static void swp_pager_meta_free_all(vm_object_t); static daddr_t swp_pager_meta_ctl(vm_object_t, vm_pindex_t, int); static void swp_pager_free_nrpage(vm_page_t m) { vm_page_lock(m); if (m->wire_count == 0) vm_page_free(m); vm_page_unlock(m); } /* * SWP_SIZECHECK() - update swap_pager_full indication * * update the swap_pager_almost_full indication and warn when we are * about to run out of swap space, using lowat/hiwat hysteresis. * * Clear swap_pager_full ( task killing ) indication when lowat is met. * * No restrictions on call * This routine may not block. */ static void swp_sizecheck(void) { if (swap_pager_avail < nswap_lowat) { if (swap_pager_almost_full == 0) { printf("swap_pager: out of swap space\n"); swap_pager_almost_full = 1; } } else { swap_pager_full = 0; if (swap_pager_avail > nswap_hiwat) swap_pager_almost_full = 0; } } /* * SWP_PAGER_HASH() - hash swap meta data * * This is an helper function which hashes the swapblk given * the object and page index. It returns a pointer to a pointer * to the object, or a pointer to a NULL pointer if it could not * find a swapblk. */ static struct swblock ** swp_pager_hash(vm_object_t object, vm_pindex_t index) { struct swblock **pswap; struct swblock *swap; index &= ~(vm_pindex_t)SWAP_META_MASK; pswap = &swhash[(index ^ (int)(intptr_t)object) & swhash_mask]; while ((swap = *pswap) != NULL) { if (swap->swb_object == object && swap->swb_index == index ) { break; } pswap = &swap->swb_hnext; } return (pswap); } /* * SWAP_PAGER_INIT() - initialize the swap pager! * * Expected to be started from system init. NOTE: This code is run * before much else so be careful what you depend on. Most of the VM * system has yet to be initialized at this point. */ static void swap_pager_init(void) { /* * Initialize object lists */ int i; for (i = 0; i < NOBJLISTS; ++i) TAILQ_INIT(&swap_pager_object_list[i]); mtx_init(&sw_alloc_mtx, "swap_pager list", NULL, MTX_DEF); mtx_init(&sw_dev_mtx, "swapdev", NULL, MTX_DEF); /* * Device Stripe, in PAGE_SIZE'd blocks */ dmmax = SWB_NPAGES * 2; } /* * SWAP_PAGER_SWAP_INIT() - swap pager initialization from pageout process * * Expected to be started from pageout process once, prior to entering * its main loop. */ void swap_pager_swap_init(void) { unsigned long n, n2; /* * Number of in-transit swap bp operations. Don't * exhaust the pbufs completely. Make sure we * initialize workable values (0 will work for hysteresis * but it isn't very efficient). * * The nsw_cluster_max is constrained by the bp->b_pages[] * array (MAXPHYS/PAGE_SIZE) and our locally defined * MAX_PAGEOUT_CLUSTER. Also be aware that swap ops are * constrained by the swap device interleave stripe size. * * Currently we hardwire nsw_wcount_async to 4. This limit is * designed to prevent other I/O from having high latencies due to * our pageout I/O. The value 4 works well for one or two active swap * devices but is probably a little low if you have more. Even so, * a higher value would probably generate only a limited improvement * with three or four active swap devices since the system does not * typically have to pageout at extreme bandwidths. We will want * at least 2 per swap devices, and 4 is a pretty good value if you * have one NFS swap device due to the command/ack latency over NFS. * So it all works out pretty well. */ nsw_cluster_max = min((MAXPHYS/PAGE_SIZE), MAX_PAGEOUT_CLUSTER); mtx_lock(&pbuf_mtx); nsw_rcount = (nswbuf + 1) / 2; nsw_wcount_sync = (nswbuf + 3) / 4; nsw_wcount_async = 4; nsw_wcount_async_max = nsw_wcount_async; mtx_unlock(&pbuf_mtx); /* * Initialize our zone. Right now I'm just guessing on the number * we need based on the number of pages in the system. Each swblock * can hold 32 pages, so this is probably overkill. This reservation * is typically limited to around 32MB by default. */ n = vm_cnt.v_page_count / 2; if (maxswzone && n > maxswzone / sizeof(struct swblock)) n = maxswzone / sizeof(struct swblock); n2 = n; swap_zone = uma_zcreate("SWAPMETA", sizeof(struct swblock), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE | UMA_ZONE_VM); if (swap_zone == NULL) panic("failed to create swap_zone."); do { if (uma_zone_reserve_kva(swap_zone, n)) break; /* * if the allocation failed, try a zone two thirds the * size of the previous attempt. */ n -= ((n + 2) / 3); } while (n > 0); if (n2 != n) printf("Swap zone entries reduced from %lu to %lu.\n", n2, n); swap_maxpages = n * SWAP_META_PAGES; swzone = n * sizeof(struct swblock); n2 = n; /* * Initialize our meta-data hash table. The swapper does not need to * be quite as efficient as the VM system, so we do not use an * oversized hash table. * * n: size of hash table, must be power of 2 * swhash_mask: hash table index mask */ for (n = 1; n < n2 / 8; n *= 2) ; swhash = malloc(sizeof(struct swblock *) * n, M_VMPGDATA, M_WAITOK | M_ZERO); swhash_mask = n - 1; mtx_init(&swhash_mtx, "swap_pager swhash", NULL, MTX_DEF); } /* * SWAP_PAGER_ALLOC() - allocate a new OBJT_SWAP VM object and instantiate * its metadata structures. * * This routine is called from the mmap and fork code to create a new * OBJT_SWAP object. We do this by creating an OBJT_DEFAULT object * and then converting it with swp_pager_meta_build(). * * This routine may block in vm_object_allocate() and create a named * object lookup race, so we must interlock. * * MPSAFE */ static vm_object_t swap_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t offset, struct ucred *cred) { vm_object_t object; vm_pindex_t pindex; pindex = OFF_TO_IDX(offset + PAGE_MASK + size); if (handle) { mtx_lock(&Giant); /* * Reference existing named region or allocate new one. There * should not be a race here against swp_pager_meta_build() * as called from vm_page_remove() in regards to the lookup * of the handle. */ sx_xlock(&sw_alloc_sx); object = vm_pager_object_lookup(NOBJLIST(handle), handle); if (object == NULL) { if (cred != NULL) { if (!swap_reserve_by_cred(size, cred)) { sx_xunlock(&sw_alloc_sx); mtx_unlock(&Giant); return (NULL); } crhold(cred); } object = vm_object_allocate(OBJT_DEFAULT, pindex); VM_OBJECT_WLOCK(object); object->handle = handle; if (cred != NULL) { object->cred = cred; object->charge = size; } swp_pager_meta_build(object, 0, SWAPBLK_NONE); VM_OBJECT_WUNLOCK(object); } sx_xunlock(&sw_alloc_sx); mtx_unlock(&Giant); } else { if (cred != NULL) { if (!swap_reserve_by_cred(size, cred)) return (NULL); crhold(cred); } object = vm_object_allocate(OBJT_DEFAULT, pindex); VM_OBJECT_WLOCK(object); if (cred != NULL) { object->cred = cred; object->charge = size; } swp_pager_meta_build(object, 0, SWAPBLK_NONE); VM_OBJECT_WUNLOCK(object); } return (object); } /* * SWAP_PAGER_DEALLOC() - remove swap metadata from object * * The swap backing for the object is destroyed. The code is * designed such that we can reinstantiate it later, but this * routine is typically called only when the entire object is * about to be destroyed. * * The object must be locked. */ static void swap_pager_dealloc(vm_object_t object) { /* * Remove from list right away so lookups will fail if we block for * pageout completion. */ if (object->handle != NULL) { mtx_lock(&sw_alloc_mtx); TAILQ_REMOVE(NOBJLIST(object->handle), object, pager_object_list); mtx_unlock(&sw_alloc_mtx); } VM_OBJECT_ASSERT_WLOCKED(object); vm_object_pip_wait(object, "swpdea"); /* * Free all remaining metadata. We only bother to free it from * the swap meta data. We do not attempt to free swapblk's still * associated with vm_page_t's for this object. We do not care * if paging is still in progress on some objects. */ swp_pager_meta_free_all(object); object->handle = NULL; object->type = OBJT_DEAD; } /************************************************************************ * SWAP PAGER BITMAP ROUTINES * ************************************************************************/ /* * SWP_PAGER_GETSWAPSPACE() - allocate raw swap space * * Allocate swap for the requested number of pages. The starting * swap block number (a page index) is returned or SWAPBLK_NONE * if the allocation failed. * * Also has the side effect of advising that somebody made a mistake * when they configured swap and didn't configure enough. * * This routine may not sleep. * * We allocate in round-robin fashion from the configured devices. */ static daddr_t swp_pager_getswapspace(int npages) { daddr_t blk; struct swdevt *sp; int i; blk = SWAPBLK_NONE; mtx_lock(&sw_dev_mtx); sp = swdevhd; for (i = 0; i < nswapdev; i++) { if (sp == NULL) sp = TAILQ_FIRST(&swtailq); if (!(sp->sw_flags & SW_CLOSING)) { blk = blist_alloc(sp->sw_blist, npages); if (blk != SWAPBLK_NONE) { blk += sp->sw_first; sp->sw_used += npages; swap_pager_avail -= npages; swp_sizecheck(); swdevhd = TAILQ_NEXT(sp, sw_list); goto done; } } sp = TAILQ_NEXT(sp, sw_list); } if (swap_pager_full != 2) { printf("swap_pager_getswapspace(%d): failed\n", npages); swap_pager_full = 2; swap_pager_almost_full = 1; } swdevhd = NULL; done: mtx_unlock(&sw_dev_mtx); return (blk); } static int swp_pager_isondev(daddr_t blk, struct swdevt *sp) { return (blk >= sp->sw_first && blk < sp->sw_end); } static void swp_pager_strategy(struct buf *bp) { struct swdevt *sp; mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { if (bp->b_blkno >= sp->sw_first && bp->b_blkno < sp->sw_end) { mtx_unlock(&sw_dev_mtx); if ((sp->sw_flags & SW_UNMAPPED) != 0 && unmapped_buf_allowed) { bp->b_kvaalloc = bp->b_data; bp->b_data = unmapped_buf; bp->b_kvabase = unmapped_buf; bp->b_offset = 0; bp->b_flags |= B_UNMAPPED; } else { pmap_qenter((vm_offset_t)bp->b_data, &bp->b_pages[0], bp->b_bcount / PAGE_SIZE); } sp->sw_strategy(bp, sp); return; } } panic("Swapdev not found"); } /* * SWP_PAGER_FREESWAPSPACE() - free raw swap space * * This routine returns the specified swap blocks back to the bitmap. * * This routine may not sleep. */ static void swp_pager_freeswapspace(daddr_t blk, int npages) { struct swdevt *sp; mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { if (blk >= sp->sw_first && blk < sp->sw_end) { sp->sw_used -= npages; /* * If we are attempting to stop swapping on * this device, we don't want to mark any * blocks free lest they be reused. */ if ((sp->sw_flags & SW_CLOSING) == 0) { blist_free(sp->sw_blist, blk - sp->sw_first, npages); swap_pager_avail += npages; swp_sizecheck(); } mtx_unlock(&sw_dev_mtx); return; } } panic("Swapdev not found"); } /* * SWAP_PAGER_FREESPACE() - frees swap blocks associated with a page * range within an object. * * This is a globally accessible routine. * * This routine removes swapblk assignments from swap metadata. * * The external callers of this routine typically have already destroyed * or renamed vm_page_t's associated with this range in the object so * we should be ok. * * The object must be locked. */ void swap_pager_freespace(vm_object_t object, vm_pindex_t start, vm_size_t size) { swp_pager_meta_free(object, start, size); } /* * SWAP_PAGER_RESERVE() - reserve swap blocks in object * * Assigns swap blocks to the specified range within the object. The * swap blocks are not zeroed. Any previous swap assignment is destroyed. * * Returns 0 on success, -1 on failure. */ int swap_pager_reserve(vm_object_t object, vm_pindex_t start, vm_size_t size) { int n = 0; daddr_t blk = SWAPBLK_NONE; vm_pindex_t beg = start; /* save start index */ VM_OBJECT_WLOCK(object); while (size) { if (n == 0) { n = BLIST_MAX_ALLOC; while ((blk = swp_pager_getswapspace(n)) == SWAPBLK_NONE) { n >>= 1; if (n == 0) { swp_pager_meta_free(object, beg, start - beg); VM_OBJECT_WUNLOCK(object); return (-1); } } } swp_pager_meta_build(object, start, blk); --size; ++start; ++blk; --n; } swp_pager_meta_free(object, start, n); VM_OBJECT_WUNLOCK(object); return (0); } /* * SWAP_PAGER_COPY() - copy blocks from source pager to destination pager * and destroy the source. * * Copy any valid swapblks from the source to the destination. In * cases where both the source and destination have a valid swapblk, * we keep the destination's. * * This routine is allowed to sleep. It may sleep allocating metadata * indirectly through swp_pager_meta_build() or if paging is still in * progress on the source. * * The source object contains no vm_page_t's (which is just as well) * * The source object is of type OBJT_SWAP. * * The source and destination objects must be locked. * Both object locks may temporarily be released. */ void swap_pager_copy(vm_object_t srcobject, vm_object_t dstobject, vm_pindex_t offset, int destroysource) { vm_pindex_t i; VM_OBJECT_ASSERT_WLOCKED(srcobject); VM_OBJECT_ASSERT_WLOCKED(dstobject); /* * If destroysource is set, we remove the source object from the * swap_pager internal queue now. */ if (destroysource) { if (srcobject->handle != NULL) { mtx_lock(&sw_alloc_mtx); TAILQ_REMOVE( NOBJLIST(srcobject->handle), srcobject, pager_object_list ); mtx_unlock(&sw_alloc_mtx); } } /* * transfer source to destination. */ for (i = 0; i < dstobject->size; ++i) { daddr_t dstaddr; /* * Locate (without changing) the swapblk on the destination, * unless it is invalid in which case free it silently, or * if the destination is a resident page, in which case the * source is thrown away. */ dstaddr = swp_pager_meta_ctl(dstobject, i, 0); if (dstaddr == SWAPBLK_NONE) { /* * Destination has no swapblk and is not resident, * copy source. */ daddr_t srcaddr; srcaddr = swp_pager_meta_ctl( srcobject, i + offset, SWM_POP ); if (srcaddr != SWAPBLK_NONE) { /* * swp_pager_meta_build() can sleep. */ vm_object_pip_add(srcobject, 1); VM_OBJECT_WUNLOCK(srcobject); vm_object_pip_add(dstobject, 1); swp_pager_meta_build(dstobject, i, srcaddr); vm_object_pip_wakeup(dstobject); VM_OBJECT_WLOCK(srcobject); vm_object_pip_wakeup(srcobject); } } else { /* * Destination has valid swapblk or it is represented * by a resident page. We destroy the sourceblock. */ swp_pager_meta_ctl(srcobject, i + offset, SWM_FREE); } } /* * Free left over swap blocks in source. * * We have to revert the type to OBJT_DEFAULT so we do not accidently * double-remove the object from the swap queues. */ if (destroysource) { swp_pager_meta_free_all(srcobject); /* * Reverting the type is not necessary, the caller is going * to destroy srcobject directly, but I'm doing it here * for consistency since we've removed the object from its * queues. */ srcobject->type = OBJT_DEFAULT; } } /* * SWAP_PAGER_HASPAGE() - determine if we have good backing store for * the requested page. * * We determine whether good backing store exists for the requested * page and return TRUE if it does, FALSE if it doesn't. * * If TRUE, we also try to determine how much valid, contiguous backing * store exists before and after the requested page within a reasonable * distance. We do not try to restrict it to the swap device stripe * (that is handled in getpages/putpages). It probably isn't worth * doing here. */ static boolean_t swap_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *before, int *after) { daddr_t blk0; VM_OBJECT_ASSERT_LOCKED(object); /* * do we have good backing store at the requested index ? */ blk0 = swp_pager_meta_ctl(object, pindex, 0); if (blk0 == SWAPBLK_NONE) { if (before) *before = 0; if (after) *after = 0; return (FALSE); } /* * find backwards-looking contiguous good backing store */ if (before != NULL) { int i; for (i = 1; i < (SWB_NPAGES/2); ++i) { daddr_t blk; if (i > pindex) break; blk = swp_pager_meta_ctl(object, pindex - i, 0); if (blk != blk0 - i) break; } *before = (i - 1); } /* * find forward-looking contiguous good backing store */ if (after != NULL) { int i; for (i = 1; i < (SWB_NPAGES/2); ++i) { daddr_t blk; blk = swp_pager_meta_ctl(object, pindex + i, 0); if (blk != blk0 + i) break; } *after = (i - 1); } return (TRUE); } /* * SWAP_PAGER_PAGE_UNSWAPPED() - remove swap backing store related to page * * This removes any associated swap backing store, whether valid or * not, from the page. * * This routine is typically called when a page is made dirty, at * which point any associated swap can be freed. MADV_FREE also * calls us in a special-case situation * * NOTE!!! If the page is clean and the swap was valid, the caller * should make the page dirty before calling this routine. This routine * does NOT change the m->dirty status of the page. Also: MADV_FREE * depends on it. * * This routine may not sleep. * * The object containing the page must be locked. */ static void swap_pager_unswapped(vm_page_t m) { swp_pager_meta_ctl(m->object, m->pindex, SWM_FREE); } /* * SWAP_PAGER_GETPAGES() - bring pages in from swap * * Attempt to retrieve (m, count) pages from backing store, but make * sure we retrieve at least m[reqpage]. We try to load in as large * a chunk surrounding m[reqpage] as is contiguous in swap and which * belongs to the same object. * * The code is designed for asynchronous operation and * immediate-notification of 'reqpage' but tends not to be * used that way. Please do not optimize-out this algorithmic * feature, I intend to improve on it in the future. * * The parent has a single vm_object_pip_add() reference prior to * calling us and we should return with the same. * * The parent has BUSY'd the pages. We should return with 'm' * left busy, but the others adjusted. */ static int swap_pager_getpages(vm_object_t object, vm_page_t *m, int count, int reqpage) { struct buf *bp; vm_page_t mreq; int i; int j; daddr_t blk; mreq = m[reqpage]; KASSERT(mreq->object == object, ("swap_pager_getpages: object mismatch %p/%p", object, mreq->object)); /* * Calculate range to retrieve. The pages have already been assigned * their swapblks. We require a *contiguous* range but we know it to * not span devices. If we do not supply it, bad things * happen. Note that blk, iblk & jblk can be SWAPBLK_NONE, but the * loops are set up such that the case(s) are handled implicitly. * * The swp_*() calls must be made with the object locked. */ blk = swp_pager_meta_ctl(mreq->object, mreq->pindex, 0); for (i = reqpage - 1; i >= 0; --i) { daddr_t iblk; iblk = swp_pager_meta_ctl(m[i]->object, m[i]->pindex, 0); if (blk != iblk + (reqpage - i)) break; } ++i; for (j = reqpage + 1; j < count; ++j) { daddr_t jblk; jblk = swp_pager_meta_ctl(m[j]->object, m[j]->pindex, 0); if (blk != jblk - (j - reqpage)) break; } /* * free pages outside our collection range. Note: we never free * mreq, it must remain busy throughout. */ if (0 < i || j < count) { int k; for (k = 0; k < i; ++k) swp_pager_free_nrpage(m[k]); for (k = j; k < count; ++k) swp_pager_free_nrpage(m[k]); } /* * Return VM_PAGER_FAIL if we have nothing to do. Return mreq * still busy, but the others unbusied. */ if (blk == SWAPBLK_NONE) return (VM_PAGER_FAIL); /* * Getpbuf() can sleep. */ VM_OBJECT_WUNLOCK(object); /* * Get a swap buffer header to perform the IO */ bp = getpbuf(&nsw_rcount); bp->b_flags |= B_PAGING; bp->b_iocmd = BIO_READ; bp->b_iodone = swp_pager_async_iodone; bp->b_rcred = crhold(thread0.td_ucred); bp->b_wcred = crhold(thread0.td_ucred); bp->b_blkno = blk - (reqpage - i); bp->b_bcount = PAGE_SIZE * (j - i); bp->b_bufsize = PAGE_SIZE * (j - i); bp->b_pager.pg_reqpage = reqpage - i; VM_OBJECT_WLOCK(object); { int k; for (k = i; k < j; ++k) { bp->b_pages[k - i] = m[k]; m[k]->oflags |= VPO_SWAPINPROG; } } bp->b_npages = j - i; PCPU_INC(cnt.v_swapin); PCPU_ADD(cnt.v_swappgsin, bp->b_npages); /* * We still hold the lock on mreq, and our automatic completion routine * does not remove it. */ vm_object_pip_add(object, bp->b_npages); VM_OBJECT_WUNLOCK(object); /* * perform the I/O. NOTE!!! bp cannot be considered valid after * this point because we automatically release it on completion. * Instead, we look at the one page we are interested in which we * still hold a lock on even through the I/O completion. * * The other pages in our m[] array are also released on completion, * so we cannot assume they are valid anymore either. * * NOTE: b_blkno is destroyed by the call to swapdev_strategy */ BUF_KERNPROC(bp); swp_pager_strategy(bp); /* * wait for the page we want to complete. VPO_SWAPINPROG is always * cleared on completion. If an I/O error occurs, SWAPBLK_NONE * is set in the meta-data. */ VM_OBJECT_WLOCK(object); while ((mreq->oflags & VPO_SWAPINPROG) != 0) { mreq->oflags |= VPO_SWAPSLEEP; PCPU_INC(cnt.v_intrans); if (VM_OBJECT_SLEEP(object, &object->paging_in_progress, PSWP, "swread", hz * 20)) { printf( "swap_pager: indefinite wait buffer: bufobj: %p, blkno: %jd, size: %ld\n", bp->b_bufobj, (intmax_t)bp->b_blkno, bp->b_bcount); } } /* * mreq is left busied after completion, but all the other pages * are freed. If we had an unrecoverable read error the page will * not be valid. */ if (mreq->valid != VM_PAGE_BITS_ALL) { return (VM_PAGER_ERROR); } else { return (VM_PAGER_OK); } /* * A final note: in a low swap situation, we cannot deallocate swap * and mark a page dirty here because the caller is likely to mark * the page clean when we return, causing the page to possibly revert * to all-zero's later. */ } /* * swap_pager_getpages_async(): * * Right now this is emulation of asynchronous operation on top of * swap_pager_getpages(). */ static int swap_pager_getpages_async(vm_object_t object, vm_page_t *m, int count, int reqpage, pgo_getpages_iodone_t iodone, void *arg) { int r, error; r = swap_pager_getpages(object, m, count, reqpage); VM_OBJECT_WUNLOCK(object); switch (r) { case VM_PAGER_OK: error = 0; break; case VM_PAGER_ERROR: error = EIO; break; case VM_PAGER_FAIL: error = EINVAL; break; default: panic("unhandled swap_pager_getpages() error %d", r); } (iodone)(arg, m, count, error); VM_OBJECT_WLOCK(object); return (r); } /* * swap_pager_putpages: * * Assign swap (if necessary) and initiate I/O on the specified pages. * * We support both OBJT_DEFAULT and OBJT_SWAP objects. DEFAULT objects * are automatically converted to SWAP objects. * * In a low memory situation we may block in VOP_STRATEGY(), but the new * vm_page reservation system coupled with properly written VFS devices * should ensure that no low-memory deadlock occurs. This is an area * which needs work. * * The parent has N vm_object_pip_add() references prior to * calling us and will remove references for rtvals[] that are * not set to VM_PAGER_PEND. We need to remove the rest on I/O * completion. * * The parent has soft-busy'd the pages it passes us and will unbusy * those whos rtvals[] entry is not set to VM_PAGER_PEND on return. * We need to unbusy the rest on I/O completion. */ void swap_pager_putpages(vm_object_t object, vm_page_t *m, int count, int flags, int *rtvals) { int i, n; boolean_t sync; if (count && m[0]->object != object) { panic("swap_pager_putpages: object mismatch %p/%p", object, m[0]->object ); } /* * Step 1 * * Turn object into OBJT_SWAP * check for bogus sysops * force sync if not pageout process */ if (object->type != OBJT_SWAP) swp_pager_meta_build(object, 0, SWAPBLK_NONE); VM_OBJECT_WUNLOCK(object); n = 0; if (curproc != pageproc) sync = TRUE; else sync = (flags & VM_PAGER_PUT_SYNC) != 0; /* * Step 2 * * Assign swap blocks and issue I/O. We reallocate swap on the fly. * The page is left dirty until the pageout operation completes * successfully. */ for (i = 0; i < count; i += n) { int j; struct buf *bp; daddr_t blk; /* * Maximum I/O size is limited by a number of factors. */ n = min(BLIST_MAX_ALLOC, count - i); n = min(n, nsw_cluster_max); /* * Get biggest block of swap we can. If we fail, fall * back and try to allocate a smaller block. Don't go * overboard trying to allocate space if it would overly * fragment swap. */ while ( (blk = swp_pager_getswapspace(n)) == SWAPBLK_NONE && n > 4 ) { n >>= 1; } if (blk == SWAPBLK_NONE) { for (j = 0; j < n; ++j) rtvals[i+j] = VM_PAGER_FAIL; continue; } /* * All I/O parameters have been satisfied, build the I/O * request and assign the swap space. */ if (sync == TRUE) { bp = getpbuf(&nsw_wcount_sync); } else { bp = getpbuf(&nsw_wcount_async); bp->b_flags = B_ASYNC; } bp->b_flags |= B_PAGING; bp->b_iocmd = BIO_WRITE; bp->b_rcred = crhold(thread0.td_ucred); bp->b_wcred = crhold(thread0.td_ucred); bp->b_bcount = PAGE_SIZE * n; bp->b_bufsize = PAGE_SIZE * n; bp->b_blkno = blk; VM_OBJECT_WLOCK(object); for (j = 0; j < n; ++j) { vm_page_t mreq = m[i+j]; swp_pager_meta_build( mreq->object, mreq->pindex, blk + j ); vm_page_dirty(mreq); rtvals[i+j] = VM_PAGER_OK; mreq->oflags |= VPO_SWAPINPROG; bp->b_pages[j] = mreq; } VM_OBJECT_WUNLOCK(object); bp->b_npages = n; /* * Must set dirty range for NFS to work. */ bp->b_dirtyoff = 0; bp->b_dirtyend = bp->b_bcount; PCPU_INC(cnt.v_swapout); PCPU_ADD(cnt.v_swappgsout, bp->b_npages); /* * asynchronous * * NOTE: b_blkno is destroyed by the call to swapdev_strategy */ if (sync == FALSE) { bp->b_iodone = swp_pager_async_iodone; BUF_KERNPROC(bp); swp_pager_strategy(bp); for (j = 0; j < n; ++j) rtvals[i+j] = VM_PAGER_PEND; /* restart outter loop */ continue; } /* * synchronous * * NOTE: b_blkno is destroyed by the call to swapdev_strategy */ bp->b_iodone = bdone; swp_pager_strategy(bp); /* * Wait for the sync I/O to complete, then update rtvals. * We just set the rtvals[] to VM_PAGER_PEND so we can call * our async completion routine at the end, thus avoiding a * double-free. */ bwait(bp, PVM, "swwrt"); for (j = 0; j < n; ++j) rtvals[i+j] = VM_PAGER_PEND; /* * Now that we are through with the bp, we can call the * normal async completion, which frees everything up. */ swp_pager_async_iodone(bp); } VM_OBJECT_WLOCK(object); } /* * swp_pager_async_iodone: * * Completion routine for asynchronous reads and writes from/to swap. * Also called manually by synchronous code to finish up a bp. * * This routine may not sleep. */ static void swp_pager_async_iodone(struct buf *bp) { int i; vm_object_t object = NULL; /* * report error */ if (bp->b_ioflags & BIO_ERROR) { printf( "swap_pager: I/O error - %s failed; blkno %ld," "size %ld, error %d\n", ((bp->b_iocmd == BIO_READ) ? "pagein" : "pageout"), (long)bp->b_blkno, (long)bp->b_bcount, bp->b_error ); } /* * remove the mapping for kernel virtual */ if ((bp->b_flags & B_UNMAPPED) != 0) { bp->b_data = bp->b_kvaalloc; bp->b_kvabase = bp->b_kvaalloc; bp->b_flags &= ~B_UNMAPPED; } else pmap_qremove((vm_offset_t)bp->b_data, bp->b_npages); if (bp->b_npages) { object = bp->b_pages[0]->object; VM_OBJECT_WLOCK(object); } /* * cleanup pages. If an error occurs writing to swap, we are in * very serious trouble. If it happens to be a disk error, though, * we may be able to recover by reassigning the swap later on. So * in this case we remove the m->swapblk assignment for the page * but do not free it in the rlist. The errornous block(s) are thus * never reallocated as swap. Redirty the page and continue. */ for (i = 0; i < bp->b_npages; ++i) { vm_page_t m = bp->b_pages[i]; m->oflags &= ~VPO_SWAPINPROG; if (m->oflags & VPO_SWAPSLEEP) { m->oflags &= ~VPO_SWAPSLEEP; wakeup(&object->paging_in_progress); } if (bp->b_ioflags & BIO_ERROR) { /* * If an error occurs I'd love to throw the swapblk * away without freeing it back to swapspace, so it * can never be used again. But I can't from an * interrupt. */ if (bp->b_iocmd == BIO_READ) { /* * When reading, reqpage needs to stay * locked for the parent, but all other * pages can be freed. We still want to * wakeup the parent waiting on the page, * though. ( also: pg_reqpage can be -1 and * not match anything ). * * We have to wake specifically requested pages * up too because we cleared VPO_SWAPINPROG and * someone may be waiting for that. * * NOTE: for reads, m->dirty will probably * be overridden by the original caller of * getpages so don't play cute tricks here. */ m->valid = 0; if (i != bp->b_pager.pg_reqpage) swp_pager_free_nrpage(m); else { vm_page_lock(m); vm_page_flash(m); vm_page_unlock(m); } /* * If i == bp->b_pager.pg_reqpage, do not wake * the page up. The caller needs to. */ } else { /* * If a write error occurs, reactivate page * so it doesn't clog the inactive list, * then finish the I/O. */ vm_page_dirty(m); vm_page_lock(m); vm_page_activate(m); vm_page_unlock(m); vm_page_sunbusy(m); } } else if (bp->b_iocmd == BIO_READ) { /* * NOTE: for reads, m->dirty will probably be * overridden by the original caller of getpages so * we cannot set them in order to free the underlying * swap in a low-swap situation. I don't think we'd * want to do that anyway, but it was an optimization * that existed in the old swapper for a time before * it got ripped out due to precisely this problem. * * If not the requested page then deactivate it. * * Note that the requested page, reqpage, is left * busied, but we still have to wake it up. The * other pages are released (unbusied) by * vm_page_xunbusy(). */ KASSERT(!pmap_page_is_mapped(m), ("swp_pager_async_iodone: page %p is mapped", m)); m->valid = VM_PAGE_BITS_ALL; KASSERT(m->dirty == 0, ("swp_pager_async_iodone: page %p is dirty", m)); /* * We have to wake specifically requested pages * up too because we cleared VPO_SWAPINPROG and * could be waiting for it in getpages. However, * be sure to not unbusy getpages specifically * requested page - getpages expects it to be * left busy. */ if (i != bp->b_pager.pg_reqpage) { vm_page_lock(m); vm_page_deactivate(m); vm_page_unlock(m); vm_page_xunbusy(m); } else { vm_page_lock(m); vm_page_flash(m); vm_page_unlock(m); } } else { /* * For write success, clear the dirty * status, then finish the I/O ( which decrements the * busy count and possibly wakes waiter's up ). */ KASSERT(!pmap_page_is_write_mapped(m), ("swp_pager_async_iodone: page %p is not write" " protected", m)); vm_page_undirty(m); vm_page_sunbusy(m); if (vm_page_count_severe()) { vm_page_lock(m); vm_page_try_to_cache(m); vm_page_unlock(m); } } } /* * adjust pip. NOTE: the original parent may still have its own * pip refs on the object. */ if (object != NULL) { vm_object_pip_wakeupn(object, bp->b_npages); VM_OBJECT_WUNLOCK(object); } /* * swapdev_strategy() manually sets b_vp and b_bufobj before calling * bstrategy(). Set them back to NULL now we're done with it, or we'll * trigger a KASSERT in relpbuf(). */ if (bp->b_vp) { bp->b_vp = NULL; bp->b_bufobj = NULL; } /* * release the physical I/O buffer */ relpbuf( bp, ((bp->b_iocmd == BIO_READ) ? &nsw_rcount : ((bp->b_flags & B_ASYNC) ? &nsw_wcount_async : &nsw_wcount_sync ) ) ); } /* * swap_pager_isswapped: * * Return 1 if at least one page in the given object is paged * out to the given swap device. * * This routine may not sleep. */ int swap_pager_isswapped(vm_object_t object, struct swdevt *sp) { daddr_t index = 0; int bcount; int i; VM_OBJECT_ASSERT_WLOCKED(object); if (object->type != OBJT_SWAP) return (0); mtx_lock(&swhash_mtx); for (bcount = 0; bcount < object->un_pager.swp.swp_bcount; bcount++) { struct swblock *swap; if ((swap = *swp_pager_hash(object, index)) != NULL) { for (i = 0; i < SWAP_META_PAGES; ++i) { if (swp_pager_isondev(swap->swb_pages[i], sp)) { mtx_unlock(&swhash_mtx); return (1); } } } index += SWAP_META_PAGES; } mtx_unlock(&swhash_mtx); return (0); } /* * SWP_PAGER_FORCE_PAGEIN() - force a swap block to be paged in * * This routine dissociates the page at the given index within a * swap block from its backing store, paging it in if necessary. * If the page is paged in, it is placed in the inactive queue, * since it had its backing store ripped out from under it. * We also attempt to swap in all other pages in the swap block, * we only guarantee that the one at the specified index is * paged in. * * XXX - The code to page the whole block in doesn't work, so we * revert to the one-by-one behavior for now. Sigh. */ static inline void swp_pager_force_pagein(vm_object_t object, vm_pindex_t pindex) { vm_page_t m; vm_object_pip_add(object, 1); m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL); if (m->valid == VM_PAGE_BITS_ALL) { vm_object_pip_wakeup(object); vm_page_dirty(m); vm_page_lock(m); vm_page_activate(m); vm_page_unlock(m); vm_page_xunbusy(m); vm_pager_page_unswapped(m); return; } if (swap_pager_getpages(object, &m, 1, 0) != VM_PAGER_OK) panic("swap_pager_force_pagein: read from swap failed");/*XXX*/ vm_object_pip_wakeup(object); vm_page_dirty(m); vm_page_lock(m); vm_page_deactivate(m); vm_page_unlock(m); vm_page_xunbusy(m); vm_pager_page_unswapped(m); } /* * swap_pager_swapoff: * * Page in all of the pages that have been paged out to the * given device. The corresponding blocks in the bitmap must be * marked as allocated and the device must be flagged SW_CLOSING. * There may be no processes swapped out to the device. * * This routine may block. */ static void swap_pager_swapoff(struct swdevt *sp) { struct swblock *swap; int i, j, retries; GIANT_REQUIRED; retries = 0; full_rescan: mtx_lock(&swhash_mtx); for (i = 0; i <= swhash_mask; i++) { /* '<=' is correct here */ restart: for (swap = swhash[i]; swap != NULL; swap = swap->swb_hnext) { vm_object_t object = swap->swb_object; vm_pindex_t pindex = swap->swb_index; for (j = 0; j < SWAP_META_PAGES; ++j) { if (swp_pager_isondev(swap->swb_pages[j], sp)) { /* avoid deadlock */ if (!VM_OBJECT_TRYWLOCK(object)) { break; } else { mtx_unlock(&swhash_mtx); swp_pager_force_pagein(object, pindex + j); VM_OBJECT_WUNLOCK(object); mtx_lock(&swhash_mtx); goto restart; } } } } } mtx_unlock(&swhash_mtx); if (sp->sw_used) { /* * Objects may be locked or paging to the device being * removed, so we will miss their pages and need to * make another pass. We have marked this device as * SW_CLOSING, so the activity should finish soon. */ retries++; if (retries > 100) { panic("swapoff: failed to locate %d swap blocks", sp->sw_used); } pause("swpoff", hz / 20); goto full_rescan; } } /************************************************************************ * SWAP META DATA * ************************************************************************ * * These routines manipulate the swap metadata stored in the * OBJT_SWAP object. * * Swap metadata is implemented with a global hash and not directly * linked into the object. Instead the object simply contains * appropriate tracking counters. */ /* * SWP_PAGER_META_BUILD() - add swap block to swap meta data for object * * We first convert the object to a swap object if it is a default * object. * * The specified swapblk is added to the object's swap metadata. If * the swapblk is not valid, it is freed instead. Any previously * assigned swapblk is freed. */ static void swp_pager_meta_build(vm_object_t object, vm_pindex_t pindex, daddr_t swapblk) { static volatile int exhausted; struct swblock *swap; struct swblock **pswap; int idx; VM_OBJECT_ASSERT_WLOCKED(object); /* * Convert default object to swap object if necessary */ if (object->type != OBJT_SWAP) { object->type = OBJT_SWAP; object->un_pager.swp.swp_bcount = 0; if (object->handle != NULL) { mtx_lock(&sw_alloc_mtx); TAILQ_INSERT_TAIL( NOBJLIST(object->handle), object, pager_object_list ); mtx_unlock(&sw_alloc_mtx); } } /* * Locate hash entry. If not found create, but if we aren't adding * anything just return. If we run out of space in the map we wait * and, since the hash table may have changed, retry. */ retry: mtx_lock(&swhash_mtx); pswap = swp_pager_hash(object, pindex); if ((swap = *pswap) == NULL) { int i; if (swapblk == SWAPBLK_NONE) goto done; swap = *pswap = uma_zalloc(swap_zone, M_NOWAIT | (curproc == pageproc ? M_USE_RESERVE : 0)); if (swap == NULL) { mtx_unlock(&swhash_mtx); VM_OBJECT_WUNLOCK(object); if (uma_zone_exhausted(swap_zone)) { if (atomic_cmpset_int(&exhausted, 0, 1)) printf("swap zone exhausted, " "increase kern.maxswzone\n"); vm_pageout_oom(VM_OOM_SWAPZ); pause("swzonex", 10); } else VM_WAIT; VM_OBJECT_WLOCK(object); goto retry; } if (atomic_cmpset_int(&exhausted, 1, 0)) printf("swap zone ok\n"); swap->swb_hnext = NULL; swap->swb_object = object; swap->swb_index = pindex & ~(vm_pindex_t)SWAP_META_MASK; swap->swb_count = 0; ++object->un_pager.swp.swp_bcount; for (i = 0; i < SWAP_META_PAGES; ++i) swap->swb_pages[i] = SWAPBLK_NONE; } /* * Delete prior contents of metadata */ idx = pindex & SWAP_META_MASK; if (swap->swb_pages[idx] != SWAPBLK_NONE) { swp_pager_freeswapspace(swap->swb_pages[idx], 1); --swap->swb_count; } /* * Enter block into metadata */ swap->swb_pages[idx] = swapblk; if (swapblk != SWAPBLK_NONE) ++swap->swb_count; done: mtx_unlock(&swhash_mtx); } /* * SWP_PAGER_META_FREE() - free a range of blocks in the object's swap metadata * * The requested range of blocks is freed, with any associated swap * returned to the swap bitmap. * * This routine will free swap metadata structures as they are cleaned * out. This routine does *NOT* operate on swap metadata associated * with resident pages. */ static void swp_pager_meta_free(vm_object_t object, vm_pindex_t index, daddr_t count) { VM_OBJECT_ASSERT_LOCKED(object); if (object->type != OBJT_SWAP) return; while (count > 0) { struct swblock **pswap; struct swblock *swap; mtx_lock(&swhash_mtx); pswap = swp_pager_hash(object, index); if ((swap = *pswap) != NULL) { daddr_t v = swap->swb_pages[index & SWAP_META_MASK]; if (v != SWAPBLK_NONE) { swp_pager_freeswapspace(v, 1); swap->swb_pages[index & SWAP_META_MASK] = SWAPBLK_NONE; if (--swap->swb_count == 0) { *pswap = swap->swb_hnext; uma_zfree(swap_zone, swap); --object->un_pager.swp.swp_bcount; } } --count; ++index; } else { int n = SWAP_META_PAGES - (index & SWAP_META_MASK); count -= n; index += n; } mtx_unlock(&swhash_mtx); } } /* * SWP_PAGER_META_FREE_ALL() - destroy all swap metadata associated with object * * This routine locates and destroys all swap metadata associated with * an object. */ static void swp_pager_meta_free_all(vm_object_t object) { daddr_t index = 0; VM_OBJECT_ASSERT_WLOCKED(object); if (object->type != OBJT_SWAP) return; while (object->un_pager.swp.swp_bcount) { struct swblock **pswap; struct swblock *swap; mtx_lock(&swhash_mtx); pswap = swp_pager_hash(object, index); if ((swap = *pswap) != NULL) { int i; for (i = 0; i < SWAP_META_PAGES; ++i) { daddr_t v = swap->swb_pages[i]; if (v != SWAPBLK_NONE) { --swap->swb_count; swp_pager_freeswapspace(v, 1); } } if (swap->swb_count != 0) panic("swap_pager_meta_free_all: swb_count != 0"); *pswap = swap->swb_hnext; uma_zfree(swap_zone, swap); --object->un_pager.swp.swp_bcount; } mtx_unlock(&swhash_mtx); index += SWAP_META_PAGES; } } /* * SWP_PAGER_METACTL() - misc control of swap and vm_page_t meta data. * * This routine is capable of looking up, popping, or freeing * swapblk assignments in the swap meta data or in the vm_page_t. * The routine typically returns the swapblk being looked-up, or popped, * or SWAPBLK_NONE if the block was freed, or SWAPBLK_NONE if the block * was invalid. This routine will automatically free any invalid * meta-data swapblks. * * It is not possible to store invalid swapblks in the swap meta data * (other then a literal 'SWAPBLK_NONE'), so we don't bother checking. * * When acting on a busy resident page and paging is in progress, we * have to wait until paging is complete but otherwise can act on the * busy page. * * SWM_FREE remove and free swap block from metadata * SWM_POP remove from meta data but do not free.. pop it out */ static daddr_t swp_pager_meta_ctl(vm_object_t object, vm_pindex_t pindex, int flags) { struct swblock **pswap; struct swblock *swap; daddr_t r1; int idx; VM_OBJECT_ASSERT_LOCKED(object); /* * The meta data only exists of the object is OBJT_SWAP * and even then might not be allocated yet. */ if (object->type != OBJT_SWAP) return (SWAPBLK_NONE); r1 = SWAPBLK_NONE; mtx_lock(&swhash_mtx); pswap = swp_pager_hash(object, pindex); if ((swap = *pswap) != NULL) { idx = pindex & SWAP_META_MASK; r1 = swap->swb_pages[idx]; if (r1 != SWAPBLK_NONE) { if (flags & SWM_FREE) { swp_pager_freeswapspace(r1, 1); r1 = SWAPBLK_NONE; } if (flags & (SWM_FREE|SWM_POP)) { swap->swb_pages[idx] = SWAPBLK_NONE; if (--swap->swb_count == 0) { *pswap = swap->swb_hnext; uma_zfree(swap_zone, swap); --object->un_pager.swp.swp_bcount; } } } } mtx_unlock(&swhash_mtx); return (r1); } /* * System call swapon(name) enables swapping on device name, * which must be in the swdevsw. Return EBUSY * if already swapping on this device. */ #ifndef _SYS_SYSPROTO_H_ struct swapon_args { char *name; }; #endif /* * MPSAFE */ /* ARGSUSED */ int sys_swapon(struct thread *td, struct swapon_args *uap) { struct vattr attr; struct vnode *vp; struct nameidata nd; int error; error = priv_check(td, PRIV_SWAPON); if (error) return (error); mtx_lock(&Giant); while (swdev_syscall_active) tsleep(&swdev_syscall_active, PUSER - 1, "swpon", 0); swdev_syscall_active = 1; /* * Swap metadata may not fit in the KVM if we have physical * memory of >1GB. */ if (swap_zone == NULL) { error = ENOMEM; goto done; } NDINIT(&nd, LOOKUP, ISOPEN | FOLLOW | AUDITVNODE1, UIO_USERSPACE, uap->name, td); error = namei(&nd); if (error) goto done; NDFREE(&nd, NDF_ONLY_PNBUF); vp = nd.ni_vp; if (vn_isdisk(vp, &error)) { error = swapongeom(td, vp); } else if (vp->v_type == VREG && (vp->v_mount->mnt_vfc->vfc_flags & VFCF_NETWORK) != 0 && (error = VOP_GETATTR(vp, &attr, td->td_ucred)) == 0) { /* * Allow direct swapping to NFS regular files in the same * way that nfs_mountroot() sets up diskless swapping. */ error = swaponvp(td, vp, attr.va_size / DEV_BSIZE); } if (error) vrele(vp); done: swdev_syscall_active = 0; wakeup_one(&swdev_syscall_active); mtx_unlock(&Giant); return (error); } /* * Check that the total amount of swap currently configured does not * exceed half the theoretical maximum. If it does, print a warning * message and return -1; otherwise, return 0. */ static int swapon_check_swzone(unsigned long npages) { unsigned long maxpages; /* absolute maximum we can handle assuming 100% efficiency */ maxpages = uma_zone_get_max(swap_zone) * SWAP_META_PAGES; /* recommend using no more than half that amount */ if (npages > maxpages / 2) { printf("warning: total configured swap (%lu pages) " "exceeds maximum recommended amount (%lu pages).\n", npages, maxpages / 2); printf("warning: increase kern.maxswzone " "or reduce amount of swap.\n"); return (-1); } return (0); } static void swaponsomething(struct vnode *vp, void *id, u_long nblks, sw_strategy_t *strategy, sw_close_t *close, dev_t dev, int flags) { struct swdevt *sp, *tsp; swblk_t dvbase; u_long mblocks; /* * nblks is in DEV_BSIZE'd chunks, convert to PAGE_SIZE'd chunks. * First chop nblks off to page-align it, then convert. * * sw->sw_nblks is in page-sized chunks now too. */ nblks &= ~(ctodb(1) - 1); nblks = dbtoc(nblks); /* * If we go beyond this, we get overflows in the radix * tree bitmap code. */ mblocks = 0x40000000 / BLIST_META_RADIX; if (nblks > mblocks) { printf( "WARNING: reducing swap size to maximum of %luMB per unit\n", mblocks / 1024 / 1024 * PAGE_SIZE); nblks = mblocks; } sp = malloc(sizeof *sp, M_VMPGDATA, M_WAITOK | M_ZERO); sp->sw_vp = vp; sp->sw_id = id; sp->sw_dev = dev; sp->sw_flags = 0; sp->sw_nblks = nblks; sp->sw_used = 0; sp->sw_strategy = strategy; sp->sw_close = close; sp->sw_flags = flags; sp->sw_blist = blist_create(nblks, M_WAITOK); /* * Do not free the first two block in order to avoid overwriting * any bsd label at the front of the partition */ blist_free(sp->sw_blist, 2, nblks - 2); dvbase = 0; mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(tsp, &swtailq, sw_list) { if (tsp->sw_end >= dvbase) { /* * We put one uncovered page between the devices * in order to definitively prevent any cross-device * I/O requests */ dvbase = tsp->sw_end + 1; } } sp->sw_first = dvbase; sp->sw_end = dvbase + nblks; TAILQ_INSERT_TAIL(&swtailq, sp, sw_list); nswapdev++; swap_pager_avail += nblks; swap_total += (vm_ooffset_t)nblks * PAGE_SIZE; swapon_check_swzone(swap_total / PAGE_SIZE); swp_sizecheck(); mtx_unlock(&sw_dev_mtx); } /* * SYSCALL: swapoff(devname) * * Disable swapping on the given device. * * XXX: Badly designed system call: it should use a device index * rather than filename as specification. We keep sw_vp around * only to make this work. */ #ifndef _SYS_SYSPROTO_H_ struct swapoff_args { char *name; }; #endif /* * MPSAFE */ /* ARGSUSED */ int sys_swapoff(struct thread *td, struct swapoff_args *uap) { struct vnode *vp; struct nameidata nd; struct swdevt *sp; int error; error = priv_check(td, PRIV_SWAPOFF); if (error) return (error); mtx_lock(&Giant); while (swdev_syscall_active) tsleep(&swdev_syscall_active, PUSER - 1, "swpoff", 0); swdev_syscall_active = 1; NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNODE1, UIO_USERSPACE, uap->name, td); error = namei(&nd); if (error) goto done; NDFREE(&nd, NDF_ONLY_PNBUF); vp = nd.ni_vp; mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { if (sp->sw_vp == vp) break; } mtx_unlock(&sw_dev_mtx); if (sp == NULL) { error = EINVAL; goto done; } error = swapoff_one(sp, td->td_ucred); done: swdev_syscall_active = 0; wakeup_one(&swdev_syscall_active); mtx_unlock(&Giant); return (error); } static int swapoff_one(struct swdevt *sp, struct ucred *cred) { u_long nblks, dvbase; #ifdef MAC int error; #endif mtx_assert(&Giant, MA_OWNED); #ifdef MAC (void) vn_lock(sp->sw_vp, LK_EXCLUSIVE | LK_RETRY); error = mac_system_check_swapoff(cred, sp->sw_vp); (void) VOP_UNLOCK(sp->sw_vp, 0); if (error != 0) return (error); #endif nblks = sp->sw_nblks; /* * We can turn off this swap device safely only if the * available virtual memory in the system will fit the amount * of data we will have to page back in, plus an epsilon so * the system doesn't become critically low on swap space. */ if (vm_cnt.v_free_count + vm_cnt.v_cache_count + swap_pager_avail < nblks + nswap_lowat) { return (ENOMEM); } /* * Prevent further allocations on this device. */ mtx_lock(&sw_dev_mtx); sp->sw_flags |= SW_CLOSING; for (dvbase = 0; dvbase < sp->sw_end; dvbase += dmmax) { swap_pager_avail -= blist_fill(sp->sw_blist, dvbase, dmmax); } swap_total -= (vm_ooffset_t)nblks * PAGE_SIZE; mtx_unlock(&sw_dev_mtx); /* * Page in the contents of the device and close it. */ swap_pager_swapoff(sp); sp->sw_close(curthread, sp); sp->sw_id = NULL; mtx_lock(&sw_dev_mtx); TAILQ_REMOVE(&swtailq, sp, sw_list); nswapdev--; if (nswapdev == 0) { swap_pager_full = 2; swap_pager_almost_full = 1; } if (swdevhd == sp) swdevhd = NULL; mtx_unlock(&sw_dev_mtx); blist_destroy(sp->sw_blist); free(sp, M_VMPGDATA); return (0); } void swapoff_all(void) { struct swdevt *sp, *spt; const char *devname; int error; mtx_lock(&Giant); while (swdev_syscall_active) tsleep(&swdev_syscall_active, PUSER - 1, "swpoff", 0); swdev_syscall_active = 1; mtx_lock(&sw_dev_mtx); TAILQ_FOREACH_SAFE(sp, &swtailq, sw_list, spt) { mtx_unlock(&sw_dev_mtx); if (vn_isdisk(sp->sw_vp, NULL)) devname = devtoname(sp->sw_vp->v_rdev); else devname = "[file]"; error = swapoff_one(sp, thread0.td_ucred); if (error != 0) { printf("Cannot remove swap device %s (error=%d), " "skipping.\n", devname, error); } else if (bootverbose) { printf("Swap device %s removed.\n", devname); } mtx_lock(&sw_dev_mtx); } mtx_unlock(&sw_dev_mtx); swdev_syscall_active = 0; wakeup_one(&swdev_syscall_active); mtx_unlock(&Giant); } void swap_pager_status(int *total, int *used) { struct swdevt *sp; *total = 0; *used = 0; mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { *total += sp->sw_nblks; *used += sp->sw_used; } mtx_unlock(&sw_dev_mtx); } int swap_dev_info(int name, struct xswdev *xs, char *devname, size_t len) { struct swdevt *sp; const char *tmp_devname; int error, n; n = 0; error = ENOENT; mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { if (n != name) { n++; continue; } xs->xsw_version = XSWDEV_VERSION; xs->xsw_dev = sp->sw_dev; xs->xsw_flags = sp->sw_flags; xs->xsw_nblks = sp->sw_nblks; xs->xsw_used = sp->sw_used; if (devname != NULL) { if (vn_isdisk(sp->sw_vp, NULL)) tmp_devname = devtoname(sp->sw_vp->v_rdev); else tmp_devname = "[file]"; strncpy(devname, tmp_devname, len); } error = 0; break; } mtx_unlock(&sw_dev_mtx); return (error); } static int sysctl_vm_swap_info(SYSCTL_HANDLER_ARGS) { struct xswdev xs; int error; if (arg2 != 1) /* name length */ return (EINVAL); error = swap_dev_info(*(int *)arg1, &xs, NULL, 0); if (error != 0) return (error); error = SYSCTL_OUT(req, &xs, sizeof(xs)); return (error); } SYSCTL_INT(_vm, OID_AUTO, nswapdev, CTLFLAG_RD, &nswapdev, 0, "Number of swap devices"); SYSCTL_NODE(_vm, OID_AUTO, swap_info, CTLFLAG_RD, sysctl_vm_swap_info, "Swap statistics by device"); /* * vmspace_swap_count() - count the approximate swap usage in pages for a * vmspace. * * The map must be locked. * * Swap usage is determined by taking the proportional swap used by * VM objects backing the VM map. To make up for fractional losses, * if the VM object has any swap use at all the associated map entries * count for at least 1 swap page. */ long vmspace_swap_count(struct vmspace *vmspace) { vm_map_t map; vm_map_entry_t cur; vm_object_t object; long count, n; map = &vmspace->vm_map; count = 0; for (cur = map->header.next; cur != &map->header; cur = cur->next) { if ((cur->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 && (object = cur->object.vm_object) != NULL) { VM_OBJECT_WLOCK(object); if (object->type == OBJT_SWAP && object->un_pager.swp.swp_bcount != 0) { n = (cur->end - cur->start) / PAGE_SIZE; count += object->un_pager.swp.swp_bcount * SWAP_META_PAGES * n / object->size + 1; } VM_OBJECT_WUNLOCK(object); } } return (count); } /* * GEOM backend * * Swapping onto disk devices. * */ static g_orphan_t swapgeom_orphan; static struct g_class g_swap_class = { .name = "SWAP", .version = G_VERSION, .orphan = swapgeom_orphan, }; DECLARE_GEOM_CLASS(g_swap_class, g_class); static void swapgeom_close_ev(void *arg, int flags) { struct g_consumer *cp; cp = arg; g_access(cp, -1, -1, 0); g_detach(cp); g_destroy_consumer(cp); } static void swapgeom_done(struct bio *bp2) { struct swdevt *sp; struct buf *bp; struct g_consumer *cp; bp = bp2->bio_caller2; cp = bp2->bio_from; bp->b_ioflags = bp2->bio_flags; if (bp2->bio_error) bp->b_ioflags |= BIO_ERROR; bp->b_resid = bp->b_bcount - bp2->bio_completed; bp->b_error = bp2->bio_error; bufdone(bp); mtx_lock(&sw_dev_mtx); if ((--cp->index) == 0 && cp->private) { if (g_post_event(swapgeom_close_ev, cp, M_NOWAIT, NULL) == 0) { sp = bp2->bio_caller1; sp->sw_id = NULL; } } mtx_unlock(&sw_dev_mtx); g_destroy_bio(bp2); } static void swapgeom_strategy(struct buf *bp, struct swdevt *sp) { struct bio *bio; struct g_consumer *cp; mtx_lock(&sw_dev_mtx); cp = sp->sw_id; if (cp == NULL) { mtx_unlock(&sw_dev_mtx); bp->b_error = ENXIO; bp->b_ioflags |= BIO_ERROR; bufdone(bp); return; } cp->index++; mtx_unlock(&sw_dev_mtx); if (bp->b_iocmd == BIO_WRITE) bio = g_new_bio(); else bio = g_alloc_bio(); if (bio == NULL) { bp->b_error = ENOMEM; bp->b_ioflags |= BIO_ERROR; bufdone(bp); return; } bio->bio_caller1 = sp; bio->bio_caller2 = bp; bio->bio_cmd = bp->b_iocmd; bio->bio_offset = (bp->b_blkno - sp->sw_first) * PAGE_SIZE; bio->bio_length = bp->b_bcount; bio->bio_done = swapgeom_done; if ((bp->b_flags & B_UNMAPPED) != 0) { bio->bio_ma = bp->b_pages; bio->bio_data = unmapped_buf; bio->bio_ma_offset = (vm_offset_t)bp->b_offset & PAGE_MASK; bio->bio_ma_n = bp->b_npages; bio->bio_flags |= BIO_UNMAPPED; } else { bio->bio_data = bp->b_data; bio->bio_ma = NULL; } g_io_request(bio, cp); return; } static void swapgeom_orphan(struct g_consumer *cp) { struct swdevt *sp; int destroy; mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { if (sp->sw_id == cp) { sp->sw_flags |= SW_CLOSING; break; } } cp->private = (void *)(uintptr_t)1; destroy = ((sp != NULL) && (cp->index == 0)); if (destroy) sp->sw_id = NULL; mtx_unlock(&sw_dev_mtx); if (destroy) swapgeom_close_ev(cp, 0); } static void swapgeom_close(struct thread *td, struct swdevt *sw) { struct g_consumer *cp; mtx_lock(&sw_dev_mtx); cp = sw->sw_id; sw->sw_id = NULL; mtx_unlock(&sw_dev_mtx); /* XXX: direct call when Giant untangled */ if (cp != NULL) g_waitfor_event(swapgeom_close_ev, cp, M_WAITOK, NULL); } struct swh0h0 { struct cdev *dev; struct vnode *vp; int error; }; static void swapongeom_ev(void *arg, int flags) { struct swh0h0 *swh; struct g_provider *pp; struct g_consumer *cp; static struct g_geom *gp; struct swdevt *sp; u_long nblks; int error; swh = arg; swh->error = 0; pp = g_dev_getprovider(swh->dev); if (pp == NULL) { swh->error = ENODEV; return; } mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { cp = sp->sw_id; if (cp != NULL && cp->provider == pp) { mtx_unlock(&sw_dev_mtx); swh->error = EBUSY; return; } } mtx_unlock(&sw_dev_mtx); if (gp == NULL) gp = g_new_geomf(&g_swap_class, "swap"); cp = g_new_consumer(gp); cp->index = 0; /* Number of active I/Os. */ cp->private = NULL; /* Orphanization flag */ g_attach(cp, pp); /* * XXX: Everytime you think you can improve the margin for * footshooting, somebody depends on the ability to do so: * savecore(8) wants to write to our swapdev so we cannot * set an exclusive count :-( */ error = g_access(cp, 1, 1, 0); if (error) { g_detach(cp); g_destroy_consumer(cp); swh->error = error; return; } nblks = pp->mediasize / DEV_BSIZE; swaponsomething(swh->vp, cp, nblks, swapgeom_strategy, swapgeom_close, dev2udev(swh->dev), (pp->flags & G_PF_ACCEPT_UNMAPPED) != 0 ? SW_UNMAPPED : 0); swh->error = 0; } static int swapongeom(struct thread *td, struct vnode *vp) { int error; struct swh0h0 swh; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); swh.dev = vp->v_rdev; swh.vp = vp; swh.error = 0; /* XXX: direct call when Giant untangled */ error = g_waitfor_event(swapongeom_ev, &swh, M_WAITOK, NULL); if (!error) error = swh.error; VOP_UNLOCK(vp, 0); return (error); } /* * VNODE backend * * This is used mainly for network filesystem (read: probably only tested * with NFS) swapfiles. * */ static void swapdev_strategy(struct buf *bp, struct swdevt *sp) { struct vnode *vp2; bp->b_blkno = ctodb(bp->b_blkno - sp->sw_first); vp2 = sp->sw_id; vhold(vp2); if (bp->b_iocmd == BIO_WRITE) { if (bp->b_bufobj) bufobj_wdrop(bp->b_bufobj); bufobj_wref(&vp2->v_bufobj); } if (bp->b_bufobj != &vp2->v_bufobj) bp->b_bufobj = &vp2->v_bufobj; bp->b_vp = vp2; bp->b_iooffset = dbtob(bp->b_blkno); bstrategy(bp); return; } static void swapdev_close(struct thread *td, struct swdevt *sp) { VOP_CLOSE(sp->sw_vp, FREAD | FWRITE, td->td_ucred, td); vrele(sp->sw_vp); } static int swaponvp(struct thread *td, struct vnode *vp, u_long nblks) { struct swdevt *sp; int error; if (nblks == 0) return (ENXIO); mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { if (sp->sw_id == vp) { mtx_unlock(&sw_dev_mtx); return (EBUSY); } } mtx_unlock(&sw_dev_mtx); (void) vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); #ifdef MAC error = mac_system_check_swapon(td->td_ucred, vp); if (error == 0) #endif error = VOP_OPEN(vp, FREAD | FWRITE, td->td_ucred, td, NULL); (void) VOP_UNLOCK(vp, 0); if (error) return (error); swaponsomething(vp, vp, nblks, swapdev_strategy, swapdev_close, NODEV, 0); return (0); } static int sysctl_swap_async_max(SYSCTL_HANDLER_ARGS) { int error, new, n; new = nsw_wcount_async_max; error = sysctl_handle_int(oidp, &new, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (new > nswbuf / 2 || new < 1) return (EINVAL); mtx_lock(&pbuf_mtx); while (nsw_wcount_async_max != new) { /* * Adjust difference. If the current async count is too low, * we will need to sqeeze our update slowly in. Sleep with a * higher priority than getpbuf() to finish faster. */ n = new - nsw_wcount_async_max; if (nsw_wcount_async + n >= 0) { nsw_wcount_async += n; nsw_wcount_async_max += n; wakeup(&nsw_wcount_async); } else { nsw_wcount_async_max -= nsw_wcount_async; nsw_wcount_async = 0; msleep(&nsw_wcount_async, &pbuf_mtx, PSWP, "swpsysctl", 0); } } mtx_unlock(&pbuf_mtx); return (0); } Index: head/sys/vm/vm_map.c =================================================================== --- head/sys/vm/vm_map.c (revision 284214) +++ head/sys/vm/vm_map.c (revision 284215) @@ -1,4290 +1,4286 @@ /*- * 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_map.c 8.3 (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. */ /* * Virtual memory mapping module. */ #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 /* * Virtual memory maps provide for the mapping, protection, * and sharing of virtual memory objects. In addition, * this module provides for an efficient virtual copy of * memory from one map to another. * * Synchronization is required prior to most operations. * * Maps consist of an ordered doubly-linked list of simple * entries; a self-adjusting binary search tree of these * entries is used to speed up lookups. * * Since portions of maps are specified by start/end addresses, * which may not align with existing map entries, all * routines merely "clip" entries to these start/end values. * [That is, an entry is split into two, bordering at a * start or end value.] Note that these clippings may not * always be necessary (as the two resulting entries are then * not changed); however, the clipping is done for convenience. * * As mentioned above, virtual copy operations are performed * by copying VM object references from one map to * another, and then marking both regions as copy-on-write. */ static struct mtx map_sleep_mtx; static uma_zone_t mapentzone; static uma_zone_t kmapentzone; static uma_zone_t mapzone; static uma_zone_t vmspace_zone; static int vmspace_zinit(void *mem, int size, int flags); static int vm_map_zinit(void *mem, int ize, int flags); static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max); static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map); static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry); static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry); static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags); #ifdef INVARIANTS static void vm_map_zdtor(void *mem, int size, void *arg); static void vmspace_zdtor(void *mem, int size, void *arg); #endif static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow); static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry, vm_offset_t failed_addr); #define ENTRY_CHARGED(e) ((e)->cred != NULL || \ ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \ !((e)->eflags & MAP_ENTRY_NEEDS_COPY))) /* * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type * stable. */ #define PROC_VMSPACE_LOCK(p) do { } while (0) #define PROC_VMSPACE_UNLOCK(p) do { } while (0) /* * VM_MAP_RANGE_CHECK: [ internal use only ] * * Asserts that the starting and ending region * addresses fall within the valid range of the map. */ #define VM_MAP_RANGE_CHECK(map, start, end) \ { \ if (start < vm_map_min(map)) \ start = vm_map_min(map); \ if (end > vm_map_max(map)) \ end = vm_map_max(map); \ if (start > end) \ start = end; \ } /* * vm_map_startup: * * Initialize the vm_map module. Must be called before * any other vm_map routines. * * Map and entry structures are allocated from the general * purpose memory pool with some exceptions: * * - The kernel map and kmem submap are allocated statically. * - Kernel map entries are allocated out of a static pool. * * These restrictions are necessary since malloc() uses the * maps and requires map entries. */ void vm_map_startup(void) { mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF); mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL, #ifdef INVARIANTS vm_map_zdtor, #else NULL, #endif vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); uma_prealloc(mapzone, MAX_KMAP); kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_MTXCLASS | UMA_ZONE_VM); mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL, #ifdef INVARIANTS vmspace_zdtor, #else NULL, #endif vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); } static int vmspace_zinit(void *mem, int size, int flags) { struct vmspace *vm; vm = (struct vmspace *)mem; vm->vm_map.pmap = NULL; (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags); PMAP_LOCK_INIT(vmspace_pmap(vm)); return (0); } static int vm_map_zinit(void *mem, int size, int flags) { vm_map_t map; map = (vm_map_t)mem; memset(map, 0, sizeof(*map)); mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK); sx_init(&map->lock, "vm map (user)"); return (0); } #ifdef INVARIANTS static void vmspace_zdtor(void *mem, int size, void *arg) { struct vmspace *vm; vm = (struct vmspace *)mem; vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg); } static void vm_map_zdtor(void *mem, int size, void *arg) { vm_map_t map; map = (vm_map_t)mem; KASSERT(map->nentries == 0, ("map %p nentries == %d on free.", map, map->nentries)); KASSERT(map->size == 0, ("map %p size == %lu on free.", map, (unsigned long)map->size)); } #endif /* INVARIANTS */ /* * Allocate a vmspace structure, including a vm_map and pmap, * and initialize those structures. The refcnt is set to 1. * * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit(). */ struct vmspace * vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit) { struct vmspace *vm; vm = uma_zalloc(vmspace_zone, M_WAITOK); KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL")); if (pinit == NULL) pinit = &pmap_pinit; if (!pinit(vmspace_pmap(vm))) { uma_zfree(vmspace_zone, vm); return (NULL); } CTR1(KTR_VM, "vmspace_alloc: %p", vm); _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max); vm->vm_refcnt = 1; vm->vm_shm = NULL; vm->vm_swrss = 0; vm->vm_tsize = 0; vm->vm_dsize = 0; vm->vm_ssize = 0; vm->vm_taddr = 0; vm->vm_daddr = 0; vm->vm_maxsaddr = 0; return (vm); } #ifdef RACCT static void vmspace_container_reset(struct proc *p) { PROC_LOCK(p); racct_set(p, RACCT_DATA, 0); racct_set(p, RACCT_STACK, 0); racct_set(p, RACCT_RSS, 0); racct_set(p, RACCT_MEMLOCK, 0); racct_set(p, RACCT_VMEM, 0); PROC_UNLOCK(p); } #endif static inline void vmspace_dofree(struct vmspace *vm) { CTR1(KTR_VM, "vmspace_free: %p", vm); /* * Make sure any SysV shm is freed, it might not have been in * exit1(). */ shmexit(vm); /* * Lock the map, to wait out all other references to it. * Delete all of the mappings and pages they hold, then call * the pmap module to reclaim anything left. */ (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset, vm->vm_map.max_offset); pmap_release(vmspace_pmap(vm)); vm->vm_map.pmap = NULL; uma_zfree(vmspace_zone, vm); } void vmspace_free(struct vmspace *vm) { WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "vmspace_free() called with non-sleepable lock held"); if (vm->vm_refcnt == 0) panic("vmspace_free: attempt to free already freed vmspace"); if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1) vmspace_dofree(vm); } void vmspace_exitfree(struct proc *p) { struct vmspace *vm; PROC_VMSPACE_LOCK(p); vm = p->p_vmspace; p->p_vmspace = NULL; PROC_VMSPACE_UNLOCK(p); KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace")); vmspace_free(vm); } void vmspace_exit(struct thread *td) { int refcnt; struct vmspace *vm; struct proc *p; /* * Release user portion of address space. * This releases references to vnodes, * which could cause I/O if the file has been unlinked. * Need to do this early enough that we can still sleep. * * The last exiting process to reach this point releases as * much of the environment as it can. vmspace_dofree() is the * slower fallback in case another process had a temporary * reference to the vmspace. */ p = td->td_proc; vm = p->p_vmspace; atomic_add_int(&vmspace0.vm_refcnt, 1); do { refcnt = vm->vm_refcnt; if (refcnt > 1 && p->p_vmspace != &vmspace0) { /* Switch now since other proc might free vmspace */ PROC_VMSPACE_LOCK(p); p->p_vmspace = &vmspace0; PROC_VMSPACE_UNLOCK(p); pmap_activate(td); } } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1)); if (refcnt == 1) { if (p->p_vmspace != vm) { /* vmspace not yet freed, switch back */ PROC_VMSPACE_LOCK(p); p->p_vmspace = vm; PROC_VMSPACE_UNLOCK(p); pmap_activate(td); } pmap_remove_pages(vmspace_pmap(vm)); /* Switch now since this proc will free vmspace */ PROC_VMSPACE_LOCK(p); p->p_vmspace = &vmspace0; PROC_VMSPACE_UNLOCK(p); pmap_activate(td); vmspace_dofree(vm); } #ifdef RACCT if (racct_enable) vmspace_container_reset(p); #endif } /* Acquire reference to vmspace owned by another process. */ struct vmspace * vmspace_acquire_ref(struct proc *p) { struct vmspace *vm; int refcnt; PROC_VMSPACE_LOCK(p); vm = p->p_vmspace; if (vm == NULL) { PROC_VMSPACE_UNLOCK(p); return (NULL); } do { refcnt = vm->vm_refcnt; if (refcnt <= 0) { /* Avoid 0->1 transition */ PROC_VMSPACE_UNLOCK(p); return (NULL); } } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1)); if (vm != p->p_vmspace) { PROC_VMSPACE_UNLOCK(p); vmspace_free(vm); return (NULL); } PROC_VMSPACE_UNLOCK(p); return (vm); } void _vm_map_lock(vm_map_t map, const char *file, int line) { if (map->system_map) mtx_lock_flags_(&map->system_mtx, 0, file, line); else sx_xlock_(&map->lock, file, line); map->timestamp++; } static void vm_map_process_deferred(void) { struct thread *td; vm_map_entry_t entry, next; vm_object_t object; td = curthread; entry = td->td_map_def_user; td->td_map_def_user = NULL; while (entry != NULL) { next = entry->next; if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) { /* * Decrement the object's writemappings and * possibly the vnode's v_writecount. */ KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0, ("Submap with writecount")); object = entry->object.vm_object; KASSERT(object != NULL, ("No object for writecount")); vnode_pager_release_writecount(object, entry->start, entry->end); } vm_map_entry_deallocate(entry, FALSE); entry = next; } } void _vm_map_unlock(vm_map_t map, const char *file, int line) { if (map->system_map) mtx_unlock_flags_(&map->system_mtx, 0, file, line); else { sx_xunlock_(&map->lock, file, line); vm_map_process_deferred(); } } void _vm_map_lock_read(vm_map_t map, const char *file, int line) { if (map->system_map) mtx_lock_flags_(&map->system_mtx, 0, file, line); else sx_slock_(&map->lock, file, line); } void _vm_map_unlock_read(vm_map_t map, const char *file, int line) { if (map->system_map) mtx_unlock_flags_(&map->system_mtx, 0, file, line); else { sx_sunlock_(&map->lock, file, line); vm_map_process_deferred(); } } int _vm_map_trylock(vm_map_t map, const char *file, int line) { int error; error = map->system_map ? !mtx_trylock_flags_(&map->system_mtx, 0, file, line) : !sx_try_xlock_(&map->lock, file, line); if (error == 0) map->timestamp++; return (error == 0); } int _vm_map_trylock_read(vm_map_t map, const char *file, int line) { int error; error = map->system_map ? !mtx_trylock_flags_(&map->system_mtx, 0, file, line) : !sx_try_slock_(&map->lock, file, line); return (error == 0); } /* * _vm_map_lock_upgrade: [ internal use only ] * * Tries to upgrade a read (shared) lock on the specified map to a write * (exclusive) lock. Returns the value "0" if the upgrade succeeds and a * non-zero value if the upgrade fails. If the upgrade fails, the map is * returned without a read or write lock held. * * Requires that the map be read locked. */ int _vm_map_lock_upgrade(vm_map_t map, const char *file, int line) { unsigned int last_timestamp; if (map->system_map) { mtx_assert_(&map->system_mtx, MA_OWNED, file, line); } else { if (!sx_try_upgrade_(&map->lock, file, line)) { last_timestamp = map->timestamp; sx_sunlock_(&map->lock, file, line); vm_map_process_deferred(); /* * If the map's timestamp does not change while the * map is unlocked, then the upgrade succeeds. */ sx_xlock_(&map->lock, file, line); if (last_timestamp != map->timestamp) { sx_xunlock_(&map->lock, file, line); return (1); } } } map->timestamp++; return (0); } void _vm_map_lock_downgrade(vm_map_t map, const char *file, int line) { if (map->system_map) { mtx_assert_(&map->system_mtx, MA_OWNED, file, line); } else sx_downgrade_(&map->lock, file, line); } /* * vm_map_locked: * * Returns a non-zero value if the caller holds a write (exclusive) lock * on the specified map and the value "0" otherwise. */ int vm_map_locked(vm_map_t map) { if (map->system_map) return (mtx_owned(&map->system_mtx)); else return (sx_xlocked(&map->lock)); } #ifdef INVARIANTS static void _vm_map_assert_locked(vm_map_t map, const char *file, int line) { if (map->system_map) mtx_assert_(&map->system_mtx, MA_OWNED, file, line); else sx_assert_(&map->lock, SA_XLOCKED, file, line); } #define VM_MAP_ASSERT_LOCKED(map) \ _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE) #else #define VM_MAP_ASSERT_LOCKED(map) #endif /* * _vm_map_unlock_and_wait: * * Atomically releases the lock on the specified map and puts the calling * thread to sleep. The calling thread will remain asleep until either * vm_map_wakeup() is performed on the map or the specified timeout is * exceeded. * * WARNING! This function does not perform deferred deallocations of * objects and map entries. Therefore, the calling thread is expected to * reacquire the map lock after reawakening and later perform an ordinary * unlock operation, such as vm_map_unlock(), before completing its * operation on the map. */ int _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line) { mtx_lock(&map_sleep_mtx); if (map->system_map) mtx_unlock_flags_(&map->system_mtx, 0, file, line); else sx_xunlock_(&map->lock, file, line); return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps", timo)); } /* * vm_map_wakeup: * * Awaken any threads that have slept on the map using * vm_map_unlock_and_wait(). */ void vm_map_wakeup(vm_map_t map) { /* * Acquire and release map_sleep_mtx to prevent a wakeup() * from being performed (and lost) between the map unlock * and the msleep() in _vm_map_unlock_and_wait(). */ mtx_lock(&map_sleep_mtx); mtx_unlock(&map_sleep_mtx); wakeup(&map->root); } void vm_map_busy(vm_map_t map) { VM_MAP_ASSERT_LOCKED(map); map->busy++; } void vm_map_unbusy(vm_map_t map) { VM_MAP_ASSERT_LOCKED(map); KASSERT(map->busy, ("vm_map_unbusy: not busy")); if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) { vm_map_modflags(map, 0, MAP_BUSY_WAKEUP); wakeup(&map->busy); } } void vm_map_wait_busy(vm_map_t map) { VM_MAP_ASSERT_LOCKED(map); while (map->busy) { vm_map_modflags(map, MAP_BUSY_WAKEUP, 0); if (map->system_map) msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0); else sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0); } map->timestamp++; } long vmspace_resident_count(struct vmspace *vmspace) { return pmap_resident_count(vmspace_pmap(vmspace)); } /* * vm_map_create: * * Creates and returns a new empty VM map with * the given physical map structure, and having * the given lower and upper address bounds. */ vm_map_t vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max) { vm_map_t result; result = uma_zalloc(mapzone, M_WAITOK); CTR1(KTR_VM, "vm_map_create: %p", result); _vm_map_init(result, pmap, min, max); return (result); } /* * Initialize an existing vm_map structure * such as that in the vmspace structure. */ static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max) { map->header.next = map->header.prev = &map->header; map->needs_wakeup = FALSE; map->system_map = 0; map->pmap = pmap; map->min_offset = min; map->max_offset = max; map->flags = 0; map->root = NULL; map->timestamp = 0; map->busy = 0; } void vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max) { _vm_map_init(map, pmap, min, max); mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK); sx_init(&map->lock, "user map"); } /* * vm_map_entry_dispose: [ internal use only ] * * Inverse of vm_map_entry_create. */ static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry) { uma_zfree(map->system_map ? kmapentzone : mapentzone, entry); } /* * vm_map_entry_create: [ internal use only ] * * Allocates a VM map entry for insertion. * No entry fields are filled in. */ static vm_map_entry_t vm_map_entry_create(vm_map_t map) { vm_map_entry_t new_entry; if (map->system_map) new_entry = uma_zalloc(kmapentzone, M_NOWAIT); else new_entry = uma_zalloc(mapentzone, M_WAITOK); if (new_entry == NULL) panic("vm_map_entry_create: kernel resources exhausted"); return (new_entry); } /* * vm_map_entry_set_behavior: * * Set the expected access behavior, either normal, random, or * sequential. */ static inline void vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior) { entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) | (behavior & MAP_ENTRY_BEHAV_MASK); } /* * vm_map_entry_set_max_free: * * Set the max_free field in a vm_map_entry. */ static inline void vm_map_entry_set_max_free(vm_map_entry_t entry) { entry->max_free = entry->adj_free; if (entry->left != NULL && entry->left->max_free > entry->max_free) entry->max_free = entry->left->max_free; if (entry->right != NULL && entry->right->max_free > entry->max_free) entry->max_free = entry->right->max_free; } /* * vm_map_entry_splay: * * The Sleator and Tarjan top-down splay algorithm with the * following variation. Max_free must be computed bottom-up, so * on the downward pass, maintain the left and right spines in * reverse order. Then, make a second pass up each side to fix * the pointers and compute max_free. The time bound is O(log n) * amortized. * * The new root is the vm_map_entry containing "addr", or else an * adjacent entry (lower or higher) if addr is not in the tree. * * The map must be locked, and leaves it so. * * Returns: the new root. */ static vm_map_entry_t vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root) { vm_map_entry_t llist, rlist; vm_map_entry_t ltree, rtree; vm_map_entry_t y; /* Special case of empty tree. */ if (root == NULL) return (root); /* * Pass One: Splay down the tree until we find addr or a NULL * pointer where addr would go. llist and rlist are the two * sides in reverse order (bottom-up), with llist linked by * the right pointer and rlist linked by the left pointer in * the vm_map_entry. Wait until Pass Two to set max_free on * the two spines. */ llist = NULL; rlist = NULL; for (;;) { /* root is never NULL in here. */ if (addr < root->start) { y = root->left; if (y == NULL) break; if (addr < y->start && y->left != NULL) { /* Rotate right and put y on rlist. */ root->left = y->right; y->right = root; vm_map_entry_set_max_free(root); root = y->left; y->left = rlist; rlist = y; } else { /* Put root on rlist. */ root->left = rlist; rlist = root; root = y; } } else if (addr >= root->end) { y = root->right; if (y == NULL) break; if (addr >= y->end && y->right != NULL) { /* Rotate left and put y on llist. */ root->right = y->left; y->left = root; vm_map_entry_set_max_free(root); root = y->right; y->right = llist; llist = y; } else { /* Put root on llist. */ root->right = llist; llist = root; root = y; } } else break; } /* * Pass Two: Walk back up the two spines, flip the pointers * and set max_free. The subtrees of the root go at the * bottom of llist and rlist. */ ltree = root->left; while (llist != NULL) { y = llist->right; llist->right = ltree; vm_map_entry_set_max_free(llist); ltree = llist; llist = y; } rtree = root->right; while (rlist != NULL) { y = rlist->left; rlist->left = rtree; vm_map_entry_set_max_free(rlist); rtree = rlist; rlist = y; } /* * Final assembly: add ltree and rtree as subtrees of root. */ root->left = ltree; root->right = rtree; vm_map_entry_set_max_free(root); return (root); } /* * vm_map_entry_{un,}link: * * Insert/remove entries from maps. */ static void vm_map_entry_link(vm_map_t map, vm_map_entry_t after_where, vm_map_entry_t entry) { CTR4(KTR_VM, "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map, map->nentries, entry, after_where); VM_MAP_ASSERT_LOCKED(map); KASSERT(after_where == &map->header || after_where->end <= entry->start, ("vm_map_entry_link: prev end %jx new start %jx overlap", (uintmax_t)after_where->end, (uintmax_t)entry->start)); KASSERT(after_where->next == &map->header || entry->end <= after_where->next->start, ("vm_map_entry_link: new end %jx next start %jx overlap", (uintmax_t)entry->end, (uintmax_t)after_where->next->start)); map->nentries++; entry->prev = after_where; entry->next = after_where->next; entry->next->prev = entry; after_where->next = entry; if (after_where != &map->header) { if (after_where != map->root) vm_map_entry_splay(after_where->start, map->root); entry->right = after_where->right; entry->left = after_where; after_where->right = NULL; after_where->adj_free = entry->start - after_where->end; vm_map_entry_set_max_free(after_where); } else { entry->right = map->root; entry->left = NULL; } entry->adj_free = (entry->next == &map->header ? map->max_offset : entry->next->start) - entry->end; vm_map_entry_set_max_free(entry); map->root = entry; } static void vm_map_entry_unlink(vm_map_t map, vm_map_entry_t entry) { vm_map_entry_t next, prev, root; VM_MAP_ASSERT_LOCKED(map); if (entry != map->root) vm_map_entry_splay(entry->start, map->root); if (entry->left == NULL) root = entry->right; else { root = vm_map_entry_splay(entry->start, entry->left); root->right = entry->right; root->adj_free = (entry->next == &map->header ? map->max_offset : entry->next->start) - root->end; vm_map_entry_set_max_free(root); } map->root = root; prev = entry->prev; next = entry->next; next->prev = prev; prev->next = next; map->nentries--; CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map, map->nentries, entry); } /* * vm_map_entry_resize_free: * * Recompute the amount of free space following a vm_map_entry * and propagate that value up the tree. Call this function after * resizing a map entry in-place, that is, without a call to * vm_map_entry_link() or _unlink(). * * The map must be locked, and leaves it so. */ static void vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry) { /* * Using splay trees without parent pointers, propagating * max_free up the tree is done by moving the entry to the * root and making the change there. */ if (entry != map->root) map->root = vm_map_entry_splay(entry->start, map->root); entry->adj_free = (entry->next == &map->header ? map->max_offset : entry->next->start) - entry->end; vm_map_entry_set_max_free(entry); } /* * vm_map_lookup_entry: [ internal use only ] * * Finds the map entry containing (or * immediately preceding) the specified address * in the given map; the entry is returned * in the "entry" parameter. The boolean * result indicates whether the address is * actually contained in the map. */ boolean_t vm_map_lookup_entry( vm_map_t map, vm_offset_t address, vm_map_entry_t *entry) /* OUT */ { vm_map_entry_t cur; boolean_t locked; /* * If the map is empty, then the map entry immediately preceding * "address" is the map's header. */ cur = map->root; if (cur == NULL) *entry = &map->header; else if (address >= cur->start && cur->end > address) { *entry = cur; return (TRUE); } else if ((locked = vm_map_locked(map)) || sx_try_upgrade(&map->lock)) { /* * Splay requires a write lock on the map. However, it only * restructures the binary search tree; it does not otherwise * change the map. Thus, the map's timestamp need not change * on a temporary upgrade. */ map->root = cur = vm_map_entry_splay(address, cur); if (!locked) sx_downgrade(&map->lock); /* * If "address" is contained within a map entry, the new root * is that map entry. Otherwise, the new root is a map entry * immediately before or after "address". */ if (address >= cur->start) { *entry = cur; if (cur->end > address) return (TRUE); } else *entry = cur->prev; } else /* * Since the map is only locked for read access, perform a * standard binary search tree lookup for "address". */ for (;;) { if (address < cur->start) { if (cur->left == NULL) { *entry = cur->prev; break; } cur = cur->left; } else if (cur->end > address) { *entry = cur; return (TRUE); } else { if (cur->right == NULL) { *entry = cur; break; } cur = cur->right; } } return (FALSE); } /* * vm_map_insert: * * Inserts the given whole VM object into the target * map at the specified address range. The object's * size should match that of the address range. * * Requires that the map be locked, and leaves it so. * * If object is non-NULL, ref count must be bumped by caller * prior to making call to account for the new entry. */ int vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow) { vm_map_entry_t new_entry, prev_entry, temp_entry; vm_eflags_t protoeflags; struct ucred *cred; vm_inherit_t inheritance; VM_MAP_ASSERT_LOCKED(map); KASSERT((object != kmem_object && object != kernel_object) || (cow & MAP_COPY_ON_WRITE) == 0, ("vm_map_insert: kmem or kernel object and COW")); KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0, ("vm_map_insert: paradoxical MAP_NOFAULT request")); /* * Check that the start and end points are not bogus. */ if ((start < map->min_offset) || (end > map->max_offset) || (start >= end)) return (KERN_INVALID_ADDRESS); /* * Find the entry prior to the proposed starting address; if it's part * of an existing entry, this range is bogus. */ if (vm_map_lookup_entry(map, start, &temp_entry)) return (KERN_NO_SPACE); prev_entry = temp_entry; /* * Assert that the next entry doesn't overlap the end point. */ if ((prev_entry->next != &map->header) && (prev_entry->next->start < end)) return (KERN_NO_SPACE); protoeflags = 0; if (cow & MAP_COPY_ON_WRITE) protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY; if (cow & MAP_NOFAULT) protoeflags |= MAP_ENTRY_NOFAULT; if (cow & MAP_DISABLE_SYNCER) protoeflags |= MAP_ENTRY_NOSYNC; if (cow & MAP_DISABLE_COREDUMP) protoeflags |= MAP_ENTRY_NOCOREDUMP; if (cow & MAP_STACK_GROWS_DOWN) protoeflags |= MAP_ENTRY_GROWS_DOWN; if (cow & MAP_STACK_GROWS_UP) protoeflags |= MAP_ENTRY_GROWS_UP; if (cow & MAP_VN_WRITECOUNT) protoeflags |= MAP_ENTRY_VN_WRITECNT; if (cow & MAP_INHERIT_SHARE) inheritance = VM_INHERIT_SHARE; else inheritance = VM_INHERIT_DEFAULT; cred = NULL; if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT)) goto charged; if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) && ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) { if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start)) return (KERN_RESOURCE_SHORTAGE); KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) || object->cred == NULL, ("OVERCOMMIT: vm_map_insert o %p", object)); cred = curthread->td_ucred; } charged: /* Expand the kernel pmap, if necessary. */ if (map == kernel_map && end > kernel_vm_end) pmap_growkernel(end); if (object != NULL) { /* * OBJ_ONEMAPPING must be cleared unless this mapping * is trivially proven to be the only mapping for any * of the object's pages. (Object granularity * reference counting is insufficient to recognize * aliases with precision.) */ VM_OBJECT_WLOCK(object); if (object->ref_count > 1 || object->shadow_count != 0) vm_object_clear_flag(object, OBJ_ONEMAPPING); VM_OBJECT_WUNLOCK(object); } else if ((prev_entry != &map->header) && (prev_entry->eflags == protoeflags) && (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 && (prev_entry->end == start) && (prev_entry->wired_count == 0) && (prev_entry->cred == cred || (prev_entry->object.vm_object != NULL && (prev_entry->object.vm_object->cred == cred))) && vm_object_coalesce(prev_entry->object.vm_object, prev_entry->offset, (vm_size_t)(prev_entry->end - prev_entry->start), (vm_size_t)(end - prev_entry->end), cred != NULL && (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) { /* * We were able to extend the object. Determine if we * can extend the previous map entry to include the * new range as well. */ if ((prev_entry->inheritance == inheritance) && (prev_entry->protection == prot) && (prev_entry->max_protection == max)) { map->size += (end - prev_entry->end); prev_entry->end = end; vm_map_entry_resize_free(map, prev_entry); vm_map_simplify_entry(map, prev_entry); return (KERN_SUCCESS); } /* * If we can extend the object but cannot extend the * map entry, we have to create a new map entry. We * must bump the ref count on the extended object to * account for it. object may be NULL. */ object = prev_entry->object.vm_object; offset = prev_entry->offset + (prev_entry->end - prev_entry->start); vm_object_reference(object); if (cred != NULL && object != NULL && object->cred != NULL && !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { /* Object already accounts for this uid. */ cred = NULL; } } if (cred != NULL) crhold(cred); /* * Create a new entry */ new_entry = vm_map_entry_create(map); new_entry->start = start; new_entry->end = end; new_entry->cred = NULL; new_entry->eflags = protoeflags; new_entry->object.vm_object = object; new_entry->offset = offset; new_entry->avail_ssize = 0; new_entry->inheritance = inheritance; new_entry->protection = prot; new_entry->max_protection = max; new_entry->wired_count = 0; new_entry->wiring_thread = NULL; new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT; new_entry->next_read = OFF_TO_IDX(offset); KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry), ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry)); new_entry->cred = cred; /* * Insert the new entry into the list */ vm_map_entry_link(map, prev_entry, new_entry); map->size += new_entry->end - new_entry->start; /* * Try to coalesce the new entry with both the previous and next * entries in the list. Previously, we only attempted to coalesce * with the previous entry when object is NULL. Here, we handle the * other cases, which are less common. */ vm_map_simplify_entry(map, new_entry); if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) { vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset), end - start, cow & MAP_PREFAULT_PARTIAL); } return (KERN_SUCCESS); } /* * vm_map_findspace: * * Find the first fit (lowest VM address) for "length" free bytes * beginning at address >= start in the given map. * * In a vm_map_entry, "adj_free" is the amount of free space * adjacent (higher address) to this entry, and "max_free" is the * maximum amount of contiguous free space in its subtree. This * allows finding a free region in one path down the tree, so * O(log n) amortized with splay trees. * * The map must be locked, and leaves it so. * * Returns: 0 on success, and starting address in *addr, * 1 if insufficient space. */ int vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length, vm_offset_t *addr) /* OUT */ { vm_map_entry_t entry; vm_offset_t st; /* * Request must fit within min/max VM address and must avoid * address wrap. */ if (start < map->min_offset) start = map->min_offset; if (start + length > map->max_offset || start + length < start) return (1); /* Empty tree means wide open address space. */ if (map->root == NULL) { *addr = start; return (0); } /* * After splay, if start comes before root node, then there * must be a gap from start to the root. */ map->root = vm_map_entry_splay(start, map->root); if (start + length <= map->root->start) { *addr = start; return (0); } /* * Root is the last node that might begin its gap before * start, and this is the last comparison where address * wrap might be a problem. */ st = (start > map->root->end) ? start : map->root->end; if (length <= map->root->end + map->root->adj_free - st) { *addr = st; return (0); } /* With max_free, can immediately tell if no solution. */ entry = map->root->right; if (entry == NULL || length > entry->max_free) return (1); /* * Search the right subtree in the order: left subtree, root, * right subtree (first fit). The previous splay implies that * all regions in the right subtree have addresses > start. */ while (entry != NULL) { if (entry->left != NULL && entry->left->max_free >= length) entry = entry->left; else if (entry->adj_free >= length) { *addr = entry->end; return (0); } else entry = entry->right; } /* Can't get here, so panic if we do. */ panic("vm_map_findspace: max_free corrupt"); } int vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset, vm_offset_t start, vm_size_t length, vm_prot_t prot, vm_prot_t max, int cow) { vm_offset_t end; int result; end = start + length; KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 || object == NULL, ("vm_map_fixed: non-NULL backing object for stack")); vm_map_lock(map); VM_MAP_RANGE_CHECK(map, start, end); if ((cow & MAP_CHECK_EXCL) == 0) vm_map_delete(map, start, end); if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) { result = vm_map_stack_locked(map, start, length, sgrowsiz, prot, max, cow); } else { result = vm_map_insert(map, object, offset, start, end, prot, max, cow); } vm_map_unlock(map); return (result); } /* * vm_map_find finds an unallocated region in the target address * map with the given length. The search is defined to be * first-fit from the specified address; the region found is * returned in the same parameter. * * If object is non-NULL, ref count must be bumped by caller * prior to making call to account for the new entry. */ int vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset, vm_offset_t *addr, /* IN/OUT */ vm_size_t length, vm_offset_t max_addr, int find_space, vm_prot_t prot, vm_prot_t max, int cow) { vm_offset_t alignment, initial_addr, start; int result; KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 || object == NULL, ("vm_map_find: non-NULL backing object for stack")); if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL || (object->flags & OBJ_COLORED) == 0)) find_space = VMFS_ANY_SPACE; if (find_space >> 8 != 0) { KASSERT((find_space & 0xff) == 0, ("bad VMFS flags")); alignment = (vm_offset_t)1 << (find_space >> 8); } else alignment = 0; initial_addr = *addr; again: start = initial_addr; vm_map_lock(map); do { if (find_space != VMFS_NO_SPACE) { if (vm_map_findspace(map, start, length, addr) || (max_addr != 0 && *addr + length > max_addr)) { vm_map_unlock(map); if (find_space == VMFS_OPTIMAL_SPACE) { find_space = VMFS_ANY_SPACE; goto again; } return (KERN_NO_SPACE); } switch (find_space) { case VMFS_SUPER_SPACE: case VMFS_OPTIMAL_SPACE: pmap_align_superpage(object, offset, addr, length); break; case VMFS_ANY_SPACE: break; default: if ((*addr & (alignment - 1)) != 0) { *addr &= ~(alignment - 1); *addr += alignment; } break; } start = *addr; } if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) { result = vm_map_stack_locked(map, start, length, sgrowsiz, prot, max, cow); } else { result = vm_map_insert(map, object, offset, start, start + length, prot, max, cow); } } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE && find_space != VMFS_ANY_SPACE); vm_map_unlock(map); return (result); } /* * vm_map_simplify_entry: * * Simplify the given map entry by merging with either neighbor. This * routine also has the ability to merge with both neighbors. * * The map must be locked. * * This routine guarentees that the passed entry remains valid (though * possibly extended). When merging, this routine may delete one or * both neighbors. */ void vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry) { vm_map_entry_t next, prev; vm_size_t prevsize, esize; if ((entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP | MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) != 0) return; prev = entry->prev; if (prev != &map->header) { prevsize = prev->end - prev->start; if ( (prev->end == entry->start) && (prev->object.vm_object == entry->object.vm_object) && (!prev->object.vm_object || (prev->offset + prevsize == entry->offset)) && (prev->eflags == entry->eflags) && (prev->protection == entry->protection) && (prev->max_protection == entry->max_protection) && (prev->inheritance == entry->inheritance) && (prev->wired_count == entry->wired_count) && (prev->cred == entry->cred)) { vm_map_entry_unlink(map, prev); entry->start = prev->start; entry->offset = prev->offset; if (entry->prev != &map->header) vm_map_entry_resize_free(map, entry->prev); /* * If the backing object is a vnode object, * vm_object_deallocate() calls vrele(). * However, vrele() does not lock the vnode * because the vnode has additional * references. Thus, the map lock can be kept * without causing a lock-order reversal with * the vnode lock. * * Since we count the number of virtual page * mappings in object->un_pager.vnp.writemappings, * the writemappings value should not be adjusted * when the entry is disposed of. */ if (prev->object.vm_object) vm_object_deallocate(prev->object.vm_object); if (prev->cred != NULL) crfree(prev->cred); vm_map_entry_dispose(map, prev); } } next = entry->next; if (next != &map->header) { esize = entry->end - entry->start; if ((entry->end == next->start) && (next->object.vm_object == entry->object.vm_object) && (!entry->object.vm_object || (entry->offset + esize == next->offset)) && (next->eflags == entry->eflags) && (next->protection == entry->protection) && (next->max_protection == entry->max_protection) && (next->inheritance == entry->inheritance) && (next->wired_count == entry->wired_count) && (next->cred == entry->cred)) { vm_map_entry_unlink(map, next); entry->end = next->end; vm_map_entry_resize_free(map, entry); /* * See comment above. */ if (next->object.vm_object) vm_object_deallocate(next->object.vm_object); if (next->cred != NULL) crfree(next->cred); vm_map_entry_dispose(map, next); } } } /* * vm_map_clip_start: [ internal use only ] * * Asserts that the given entry begins at or after * the specified address; if necessary, * it splits the entry into two. */ #define vm_map_clip_start(map, entry, startaddr) \ { \ if (startaddr > entry->start) \ _vm_map_clip_start(map, entry, startaddr); \ } /* * This routine is called only when it is known that * the entry must be split. */ static void _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start) { vm_map_entry_t new_entry; VM_MAP_ASSERT_LOCKED(map); /* * Split off the front portion -- note that we must insert the new * entry BEFORE this one, so that this entry has the specified * starting address. */ vm_map_simplify_entry(map, entry); /* * If there is no object backing this entry, we might as well create * one now. If we defer it, an object can get created after the map * is clipped, and individual objects will be created for the split-up * map. This is a bit of a hack, but is also about the best place to * put this improvement. */ if (entry->object.vm_object == NULL && !map->system_map) { vm_object_t object; object = vm_object_allocate(OBJT_DEFAULT, atop(entry->end - entry->start)); entry->object.vm_object = object; entry->offset = 0; if (entry->cred != NULL) { object->cred = entry->cred; object->charge = entry->end - entry->start; entry->cred = NULL; } } else if (entry->object.vm_object != NULL && ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && entry->cred != NULL) { VM_OBJECT_WLOCK(entry->object.vm_object); KASSERT(entry->object.vm_object->cred == NULL, ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry)); entry->object.vm_object->cred = entry->cred; entry->object.vm_object->charge = entry->end - entry->start; VM_OBJECT_WUNLOCK(entry->object.vm_object); entry->cred = NULL; } new_entry = vm_map_entry_create(map); *new_entry = *entry; new_entry->end = start; entry->offset += (start - entry->start); entry->start = start; if (new_entry->cred != NULL) crhold(entry->cred); vm_map_entry_link(map, entry->prev, new_entry); if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { vm_object_reference(new_entry->object.vm_object); /* * The object->un_pager.vnp.writemappings for the * object of MAP_ENTRY_VN_WRITECNT type entry shall be * kept as is here. The virtual pages are * re-distributed among the clipped entries, so the sum is * left the same. */ } } /* * vm_map_clip_end: [ internal use only ] * * Asserts that the given entry ends at or before * the specified address; if necessary, * it splits the entry into two. */ #define vm_map_clip_end(map, entry, endaddr) \ { \ if ((endaddr) < (entry->end)) \ _vm_map_clip_end((map), (entry), (endaddr)); \ } /* * This routine is called only when it is known that * the entry must be split. */ static void _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end) { vm_map_entry_t new_entry; VM_MAP_ASSERT_LOCKED(map); /* * If there is no object backing this entry, we might as well create * one now. If we defer it, an object can get created after the map * is clipped, and individual objects will be created for the split-up * map. This is a bit of a hack, but is also about the best place to * put this improvement. */ if (entry->object.vm_object == NULL && !map->system_map) { vm_object_t object; object = vm_object_allocate(OBJT_DEFAULT, atop(entry->end - entry->start)); entry->object.vm_object = object; entry->offset = 0; if (entry->cred != NULL) { object->cred = entry->cred; object->charge = entry->end - entry->start; entry->cred = NULL; } } else if (entry->object.vm_object != NULL && ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) && entry->cred != NULL) { VM_OBJECT_WLOCK(entry->object.vm_object); KASSERT(entry->object.vm_object->cred == NULL, ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry)); entry->object.vm_object->cred = entry->cred; entry->object.vm_object->charge = entry->end - entry->start; VM_OBJECT_WUNLOCK(entry->object.vm_object); entry->cred = NULL; } /* * Create a new entry and insert it AFTER the specified entry */ new_entry = vm_map_entry_create(map); *new_entry = *entry; new_entry->start = entry->end = end; new_entry->offset += (end - entry->start); if (new_entry->cred != NULL) crhold(entry->cred); vm_map_entry_link(map, entry, new_entry); if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { vm_object_reference(new_entry->object.vm_object); } } /* * vm_map_submap: [ kernel use only ] * * Mark the given range as handled by a subordinate map. * * This range must have been created with vm_map_find, * and no other operations may have been performed on this * range prior to calling vm_map_submap. * * Only a limited number of operations can be performed * within this rage after calling vm_map_submap: * vm_fault * [Don't try vm_map_copy!] * * To remove a submapping, one must first remove the * range from the superior map, and then destroy the * submap (if desired). [Better yet, don't try it.] */ int vm_map_submap( vm_map_t map, vm_offset_t start, vm_offset_t end, vm_map_t submap) { vm_map_entry_t entry; int result = KERN_INVALID_ARGUMENT; vm_map_lock(map); VM_MAP_RANGE_CHECK(map, start, end); if (vm_map_lookup_entry(map, start, &entry)) { vm_map_clip_start(map, entry, start); } else entry = entry->next; vm_map_clip_end(map, entry, end); if ((entry->start == start) && (entry->end == end) && ((entry->eflags & MAP_ENTRY_COW) == 0) && (entry->object.vm_object == NULL)) { entry->object.sub_map = submap; entry->eflags |= MAP_ENTRY_IS_SUB_MAP; result = KERN_SUCCESS; } vm_map_unlock(map); return (result); } /* * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified */ #define MAX_INIT_PT 96 /* * vm_map_pmap_enter: * * Preload the specified map's pmap with mappings to the specified * object's memory-resident pages. No further physical pages are * allocated, and no further virtual pages are retrieved from secondary * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a * limited number of page mappings are created at the low-end of the * specified address range. (For this purpose, a superpage mapping * counts as one page mapping.) Otherwise, all resident pages within * the specified address range are mapped. Because these mappings are * being created speculatively, cached pages are not reactivated and * mapped. */ static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot, vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags) { vm_offset_t start; vm_page_t p, p_start; vm_pindex_t mask, psize, threshold, tmpidx; if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL) return; VM_OBJECT_RLOCK(object); if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { VM_OBJECT_RUNLOCK(object); VM_OBJECT_WLOCK(object); if (object->type == OBJT_DEVICE || object->type == OBJT_SG) { pmap_object_init_pt(map->pmap, addr, object, pindex, size); VM_OBJECT_WUNLOCK(object); return; } VM_OBJECT_LOCK_DOWNGRADE(object); } psize = atop(size); if (psize + pindex > object->size) { if (object->size < pindex) { VM_OBJECT_RUNLOCK(object); return; } psize = object->size - pindex; } start = 0; p_start = NULL; threshold = MAX_INIT_PT; p = vm_page_find_least(object, pindex); /* * Assert: the variable p is either (1) the page with the * least pindex greater than or equal to the parameter pindex * or (2) NULL. */ for (; p != NULL && (tmpidx = p->pindex - pindex) < psize; p = TAILQ_NEXT(p, listq)) { /* * don't allow an madvise to blow away our really * free pages allocating pv entries. */ if (((flags & MAP_PREFAULT_MADVISE) != 0 && vm_cnt.v_free_count < vm_cnt.v_free_reserved) || ((flags & MAP_PREFAULT_PARTIAL) != 0 && tmpidx >= threshold)) { psize = tmpidx; break; } if (p->valid == VM_PAGE_BITS_ALL) { if (p_start == NULL) { start = addr + ptoa(tmpidx); p_start = p; } /* Jump ahead if a superpage mapping is possible. */ if (p->psind > 0 && ((addr + ptoa(tmpidx)) & (pagesizes[p->psind] - 1)) == 0) { mask = atop(pagesizes[p->psind]) - 1; if (tmpidx + mask < psize && vm_page_ps_is_valid(p)) { p += mask; threshold += mask; } } } else if (p_start != NULL) { pmap_enter_object(map->pmap, start, addr + ptoa(tmpidx), p_start, prot); p_start = NULL; } } if (p_start != NULL) pmap_enter_object(map->pmap, start, addr + ptoa(psize), p_start, prot); VM_OBJECT_RUNLOCK(object); } /* * vm_map_protect: * * Sets the protection of the specified address * region in the target map. If "set_max" is * specified, the maximum protection is to be set; * otherwise, only the current protection is affected. */ int vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end, vm_prot_t new_prot, boolean_t set_max) { vm_map_entry_t current, entry; vm_object_t obj; struct ucred *cred; vm_prot_t old_prot; if (start == end) return (KERN_SUCCESS); vm_map_lock(map); VM_MAP_RANGE_CHECK(map, start, end); if (vm_map_lookup_entry(map, start, &entry)) { vm_map_clip_start(map, entry, start); } else { entry = entry->next; } /* * Make a first pass to check for protection violations. */ current = entry; while ((current != &map->header) && (current->start < end)) { if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { vm_map_unlock(map); return (KERN_INVALID_ARGUMENT); } if ((new_prot & current->max_protection) != new_prot) { vm_map_unlock(map); return (KERN_PROTECTION_FAILURE); } current = current->next; } /* * Do an accounting pass for private read-only mappings that * now will do cow due to allowed write (e.g. debugger sets * breakpoint on text segment) */ for (current = entry; (current != &map->header) && (current->start < end); current = current->next) { vm_map_clip_end(map, current, end); if (set_max || ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 || ENTRY_CHARGED(current)) { continue; } cred = curthread->td_ucred; obj = current->object.vm_object; if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) { if (!swap_reserve(current->end - current->start)) { vm_map_unlock(map); return (KERN_RESOURCE_SHORTAGE); } crhold(cred); current->cred = cred; continue; } VM_OBJECT_WLOCK(obj); if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) { VM_OBJECT_WUNLOCK(obj); continue; } /* * Charge for the whole object allocation now, since * we cannot distinguish between non-charged and * charged clipped mapping of the same object later. */ KASSERT(obj->charge == 0, ("vm_map_protect: object %p overcharged (entry %p)", obj, current)); if (!swap_reserve(ptoa(obj->size))) { VM_OBJECT_WUNLOCK(obj); vm_map_unlock(map); return (KERN_RESOURCE_SHORTAGE); } crhold(cred); obj->cred = cred; obj->charge = ptoa(obj->size); VM_OBJECT_WUNLOCK(obj); } /* * Go back and fix up protections. [Note that clipping is not * necessary the second time.] */ current = entry; while ((current != &map->header) && (current->start < end)) { old_prot = current->protection; if (set_max) current->protection = (current->max_protection = new_prot) & old_prot; else current->protection = new_prot; /* * For user wired map entries, the normal lazy evaluation of * write access upgrades through soft page faults is * undesirable. Instead, immediately copy any pages that are * copy-on-write and enable write access in the physical map. */ if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 && (current->protection & VM_PROT_WRITE) != 0 && (old_prot & VM_PROT_WRITE) == 0) vm_fault_copy_entry(map, map, current, current, NULL); /* * When restricting access, update the physical map. Worry * about copy-on-write here. */ if ((old_prot & ~current->protection) != 0) { #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \ VM_PROT_ALL) pmap_protect(map->pmap, current->start, current->end, current->protection & MASK(current)); #undef MASK } vm_map_simplify_entry(map, current); current = current->next; } vm_map_unlock(map); return (KERN_SUCCESS); } /* * vm_map_madvise: * * This routine traverses a processes map handling the madvise * system call. Advisories are classified as either those effecting * the vm_map_entry structure, or those effecting the underlying * objects. */ int vm_map_madvise( vm_map_t map, vm_offset_t start, vm_offset_t end, int behav) { vm_map_entry_t current, entry; int modify_map = 0; /* * Some madvise calls directly modify the vm_map_entry, in which case * we need to use an exclusive lock on the map and we need to perform * various clipping operations. Otherwise we only need a read-lock * on the map. */ switch(behav) { case MADV_NORMAL: case MADV_SEQUENTIAL: case MADV_RANDOM: case MADV_NOSYNC: case MADV_AUTOSYNC: case MADV_NOCORE: case MADV_CORE: if (start == end) return (KERN_SUCCESS); modify_map = 1; vm_map_lock(map); break; case MADV_WILLNEED: case MADV_DONTNEED: case MADV_FREE: if (start == end) return (KERN_SUCCESS); vm_map_lock_read(map); break; default: return (KERN_INVALID_ARGUMENT); } /* * Locate starting entry and clip if necessary. */ VM_MAP_RANGE_CHECK(map, start, end); if (vm_map_lookup_entry(map, start, &entry)) { if (modify_map) vm_map_clip_start(map, entry, start); } else { entry = entry->next; } if (modify_map) { /* * madvise behaviors that are implemented in the vm_map_entry. * * We clip the vm_map_entry so that behavioral changes are * limited to the specified address range. */ for (current = entry; (current != &map->header) && (current->start < end); current = current->next ) { if (current->eflags & MAP_ENTRY_IS_SUB_MAP) continue; vm_map_clip_end(map, current, end); switch (behav) { case MADV_NORMAL: vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL); break; case MADV_SEQUENTIAL: vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL); break; case MADV_RANDOM: vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM); break; case MADV_NOSYNC: current->eflags |= MAP_ENTRY_NOSYNC; break; case MADV_AUTOSYNC: current->eflags &= ~MAP_ENTRY_NOSYNC; break; case MADV_NOCORE: current->eflags |= MAP_ENTRY_NOCOREDUMP; break; case MADV_CORE: current->eflags &= ~MAP_ENTRY_NOCOREDUMP; break; default: break; } vm_map_simplify_entry(map, current); } vm_map_unlock(map); } else { vm_pindex_t pstart, pend; /* * madvise behaviors that are implemented in the underlying * vm_object. * * Since we don't clip the vm_map_entry, we have to clip * the vm_object pindex and count. */ for (current = entry; (current != &map->header) && (current->start < end); current = current->next ) { vm_offset_t useEnd, useStart; if (current->eflags & MAP_ENTRY_IS_SUB_MAP) continue; pstart = OFF_TO_IDX(current->offset); pend = pstart + atop(current->end - current->start); useStart = current->start; useEnd = current->end; if (current->start < start) { pstart += atop(start - current->start); useStart = start; } if (current->end > end) { pend -= atop(current->end - end); useEnd = end; } if (pstart >= pend) continue; /* * Perform the pmap_advise() before clearing * PGA_REFERENCED in vm_page_advise(). Otherwise, a * concurrent pmap operation, such as pmap_remove(), * could clear a reference in the pmap and set * PGA_REFERENCED on the page before the pmap_advise() * had completed. Consequently, the page would appear * referenced based upon an old reference that * occurred before this pmap_advise() ran. */ if (behav == MADV_DONTNEED || behav == MADV_FREE) pmap_advise(map->pmap, useStart, useEnd, behav); vm_object_madvise(current->object.vm_object, pstart, pend, behav); /* * Pre-populate paging structures in the * WILLNEED case. For wired entries, the * paging structures are already populated. */ if (behav == MADV_WILLNEED && current->wired_count == 0) { vm_map_pmap_enter(map, useStart, current->protection, current->object.vm_object, pstart, ptoa(pend - pstart), MAP_PREFAULT_MADVISE ); } } vm_map_unlock_read(map); } return (0); } /* * vm_map_inherit: * * Sets the inheritance of the specified address * range in the target map. Inheritance * affects how the map will be shared with * child maps at the time of vmspace_fork. */ int vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end, vm_inherit_t new_inheritance) { vm_map_entry_t entry; vm_map_entry_t temp_entry; switch (new_inheritance) { case VM_INHERIT_NONE: case VM_INHERIT_COPY: case VM_INHERIT_SHARE: break; default: return (KERN_INVALID_ARGUMENT); } if (start == end) return (KERN_SUCCESS); vm_map_lock(map); VM_MAP_RANGE_CHECK(map, start, end); if (vm_map_lookup_entry(map, start, &temp_entry)) { entry = temp_entry; vm_map_clip_start(map, entry, start); } else entry = temp_entry->next; while ((entry != &map->header) && (entry->start < end)) { vm_map_clip_end(map, entry, end); entry->inheritance = new_inheritance; vm_map_simplify_entry(map, entry); entry = entry->next; } vm_map_unlock(map); return (KERN_SUCCESS); } /* * vm_map_unwire: * * Implements both kernel and user unwiring. */ int vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags) { vm_map_entry_t entry, first_entry, tmp_entry; vm_offset_t saved_start; unsigned int last_timestamp; int rv; boolean_t need_wakeup, result, user_unwire; if (start == end) return (KERN_SUCCESS); user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; vm_map_lock(map); VM_MAP_RANGE_CHECK(map, start, end); if (!vm_map_lookup_entry(map, start, &first_entry)) { if (flags & VM_MAP_WIRE_HOLESOK) first_entry = first_entry->next; else { vm_map_unlock(map); return (KERN_INVALID_ADDRESS); } } last_timestamp = map->timestamp; entry = first_entry; while (entry != &map->header && entry->start < end) { if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { /* * We have not yet clipped the entry. */ saved_start = (start >= entry->start) ? start : entry->start; entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; if (vm_map_unlock_and_wait(map, 0)) { /* * Allow interruption of user unwiring? */ } vm_map_lock(map); if (last_timestamp+1 != map->timestamp) { /* * Look again for the entry because the map was * modified while it was unlocked. * Specifically, the entry may have been * clipped, merged, or deleted. */ if (!vm_map_lookup_entry(map, saved_start, &tmp_entry)) { if (flags & VM_MAP_WIRE_HOLESOK) tmp_entry = tmp_entry->next; else { if (saved_start == start) { /* * First_entry has been deleted. */ vm_map_unlock(map); return (KERN_INVALID_ADDRESS); } end = saved_start; rv = KERN_INVALID_ADDRESS; goto done; } } if (entry == first_entry) first_entry = tmp_entry; else first_entry = NULL; entry = tmp_entry; } last_timestamp = map->timestamp; continue; } vm_map_clip_start(map, entry, start); vm_map_clip_end(map, entry, end); /* * Mark the entry in case the map lock is released. (See * above.) */ KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 && entry->wiring_thread == NULL, ("owned map entry %p", entry)); entry->eflags |= MAP_ENTRY_IN_TRANSITION; entry->wiring_thread = curthread; /* * Check the map for holes in the specified region. * If VM_MAP_WIRE_HOLESOK was specified, skip this check. */ if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && (entry->end < end && (entry->next == &map->header || entry->next->start > entry->end))) { end = entry->end; rv = KERN_INVALID_ADDRESS; goto done; } /* * If system unwiring, require that the entry is system wired. */ if (!user_unwire && vm_map_entry_system_wired_count(entry) == 0) { end = entry->end; rv = KERN_INVALID_ARGUMENT; goto done; } entry = entry->next; } rv = KERN_SUCCESS; done: need_wakeup = FALSE; if (first_entry == NULL) { result = vm_map_lookup_entry(map, start, &first_entry); if (!result && (flags & VM_MAP_WIRE_HOLESOK)) first_entry = first_entry->next; else KASSERT(result, ("vm_map_unwire: lookup failed")); } for (entry = first_entry; entry != &map->header && entry->start < end; entry = entry->next) { /* * If VM_MAP_WIRE_HOLESOK was specified, an empty * space in the unwired region could have been mapped * while the map lock was dropped for draining * MAP_ENTRY_IN_TRANSITION. Moreover, another thread * could be simultaneously wiring this new mapping * entry. Detect these cases and skip any entries * marked as in transition by us. */ if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 || entry->wiring_thread != curthread) { KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0, ("vm_map_unwire: !HOLESOK and new/changed entry")); continue; } if (rv == KERN_SUCCESS && (!user_unwire || (entry->eflags & MAP_ENTRY_USER_WIRED))) { if (user_unwire) entry->eflags &= ~MAP_ENTRY_USER_WIRED; if (entry->wired_count == 1) vm_map_entry_unwire(map, entry); else entry->wired_count--; } KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0, ("vm_map_unwire: in-transition flag missing %p", entry)); KASSERT(entry->wiring_thread == curthread, ("vm_map_unwire: alien wire %p", entry)); entry->eflags &= ~MAP_ENTRY_IN_TRANSITION; entry->wiring_thread = NULL; if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; need_wakeup = TRUE; } vm_map_simplify_entry(map, entry); } vm_map_unlock(map); if (need_wakeup) vm_map_wakeup(map); return (rv); } /* * vm_map_wire_entry_failure: * * Handle a wiring failure on the given entry. * * The map should be locked. */ static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry, vm_offset_t failed_addr) { VM_MAP_ASSERT_LOCKED(map); KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 && entry->wired_count == 1, ("vm_map_wire_entry_failure: entry %p isn't being wired", entry)); KASSERT(failed_addr < entry->end, ("vm_map_wire_entry_failure: entry %p was fully wired", entry)); /* * If any pages at the start of this entry were successfully wired, * then unwire them. */ if (failed_addr > entry->start) { pmap_unwire(map->pmap, entry->start, failed_addr); vm_object_unwire(entry->object.vm_object, entry->offset, failed_addr - entry->start, PQ_ACTIVE); } /* * Assign an out-of-range value to represent the failure to wire this * entry. */ entry->wired_count = -1; } /* * vm_map_wire: * * Implements both kernel and user wiring. */ int vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags) { vm_map_entry_t entry, first_entry, tmp_entry; vm_offset_t faddr, saved_end, saved_start; unsigned int last_timestamp; int rv; boolean_t need_wakeup, result, user_wire; vm_prot_t prot; if (start == end) return (KERN_SUCCESS); prot = 0; if (flags & VM_MAP_WIRE_WRITE) prot |= VM_PROT_WRITE; user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE; vm_map_lock(map); VM_MAP_RANGE_CHECK(map, start, end); if (!vm_map_lookup_entry(map, start, &first_entry)) { if (flags & VM_MAP_WIRE_HOLESOK) first_entry = first_entry->next; else { vm_map_unlock(map); return (KERN_INVALID_ADDRESS); } } last_timestamp = map->timestamp; entry = first_entry; while (entry != &map->header && entry->start < end) { if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { /* * We have not yet clipped the entry. */ saved_start = (start >= entry->start) ? start : entry->start; entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; if (vm_map_unlock_and_wait(map, 0)) { /* * Allow interruption of user wiring? */ } vm_map_lock(map); if (last_timestamp + 1 != map->timestamp) { /* * Look again for the entry because the map was * modified while it was unlocked. * Specifically, the entry may have been * clipped, merged, or deleted. */ if (!vm_map_lookup_entry(map, saved_start, &tmp_entry)) { if (flags & VM_MAP_WIRE_HOLESOK) tmp_entry = tmp_entry->next; else { if (saved_start == start) { /* * first_entry has been deleted. */ vm_map_unlock(map); return (KERN_INVALID_ADDRESS); } end = saved_start; rv = KERN_INVALID_ADDRESS; goto done; } } if (entry == first_entry) first_entry = tmp_entry; else first_entry = NULL; entry = tmp_entry; } last_timestamp = map->timestamp; continue; } vm_map_clip_start(map, entry, start); vm_map_clip_end(map, entry, end); /* * Mark the entry in case the map lock is released. (See * above.) */ KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 && entry->wiring_thread == NULL, ("owned map entry %p", entry)); entry->eflags |= MAP_ENTRY_IN_TRANSITION; entry->wiring_thread = curthread; if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || (entry->protection & prot) != prot) { entry->eflags |= MAP_ENTRY_WIRE_SKIPPED; if ((flags & VM_MAP_WIRE_HOLESOK) == 0) { end = entry->end; rv = KERN_INVALID_ADDRESS; goto done; } goto next_entry; } if (entry->wired_count == 0) { entry->wired_count++; saved_start = entry->start; saved_end = entry->end; /* * Release the map lock, relying on the in-transition * mark. Mark the map busy for fork. */ vm_map_busy(map); vm_map_unlock(map); faddr = saved_start; do { /* * Simulate a fault to get the page and enter * it into the physical map. */ if ((rv = vm_fault(map, faddr, VM_PROT_NONE, VM_FAULT_CHANGE_WIRING)) != KERN_SUCCESS) break; } while ((faddr += PAGE_SIZE) < saved_end); vm_map_lock(map); vm_map_unbusy(map); if (last_timestamp + 1 != map->timestamp) { /* * Look again for the entry because the map was * modified while it was unlocked. The entry * may have been clipped, but NOT merged or * deleted. */ result = vm_map_lookup_entry(map, saved_start, &tmp_entry); KASSERT(result, ("vm_map_wire: lookup failed")); if (entry == first_entry) first_entry = tmp_entry; else first_entry = NULL; entry = tmp_entry; while (entry->end < saved_end) { /* * In case of failure, handle entries * that were not fully wired here; * fully wired entries are handled * later. */ if (rv != KERN_SUCCESS && faddr < entry->end) vm_map_wire_entry_failure(map, entry, faddr); entry = entry->next; } } last_timestamp = map->timestamp; if (rv != KERN_SUCCESS) { vm_map_wire_entry_failure(map, entry, faddr); end = entry->end; goto done; } } else if (!user_wire || (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { entry->wired_count++; } /* * Check the map for holes in the specified region. * If VM_MAP_WIRE_HOLESOK was specified, skip this check. */ next_entry: if (((flags & VM_MAP_WIRE_HOLESOK) == 0) && (entry->end < end && (entry->next == &map->header || entry->next->start > entry->end))) { end = entry->end; rv = KERN_INVALID_ADDRESS; goto done; } entry = entry->next; } rv = KERN_SUCCESS; done: need_wakeup = FALSE; if (first_entry == NULL) { result = vm_map_lookup_entry(map, start, &first_entry); if (!result && (flags & VM_MAP_WIRE_HOLESOK)) first_entry = first_entry->next; else KASSERT(result, ("vm_map_wire: lookup failed")); } for (entry = first_entry; entry != &map->header && entry->start < end; entry = entry->next) { if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0) goto next_entry_done; /* * If VM_MAP_WIRE_HOLESOK was specified, an empty * space in the unwired region could have been mapped * while the map lock was dropped for faulting in the * pages or draining MAP_ENTRY_IN_TRANSITION. * Moreover, another thread could be simultaneously * wiring this new mapping entry. Detect these cases * and skip any entries marked as in transition by us. */ if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 || entry->wiring_thread != curthread) { KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0, ("vm_map_wire: !HOLESOK and new/changed entry")); continue; } if (rv == KERN_SUCCESS) { if (user_wire) entry->eflags |= MAP_ENTRY_USER_WIRED; } else if (entry->wired_count == -1) { /* * Wiring failed on this entry. Thus, unwiring is * unnecessary. */ entry->wired_count = 0; } else if (!user_wire || (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) { /* * Undo the wiring. Wiring succeeded on this entry * but failed on a later entry. */ if (entry->wired_count == 1) vm_map_entry_unwire(map, entry); else entry->wired_count--; } next_entry_done: KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0, ("vm_map_wire: in-transition flag missing %p", entry)); KASSERT(entry->wiring_thread == curthread, ("vm_map_wire: alien wire %p", entry)); entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_WIRE_SKIPPED); entry->wiring_thread = NULL; if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) { entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP; need_wakeup = TRUE; } vm_map_simplify_entry(map, entry); } vm_map_unlock(map); if (need_wakeup) vm_map_wakeup(map); return (rv); } /* * vm_map_sync * * Push any dirty cached pages in the address range to their pager. * If syncio is TRUE, dirty pages are written synchronously. * If invalidate is TRUE, any cached pages are freed as well. * * If the size of the region from start to end is zero, we are * supposed to flush all modified pages within the region containing * start. Unfortunately, a region can be split or coalesced with * neighboring regions, making it difficult to determine what the * original region was. Therefore, we approximate this requirement by * flushing the current region containing start. * * Returns an error if any part of the specified range is not mapped. */ int vm_map_sync( vm_map_t map, vm_offset_t start, vm_offset_t end, boolean_t syncio, boolean_t invalidate) { vm_map_entry_t current; vm_map_entry_t entry; vm_size_t size; vm_object_t object; vm_ooffset_t offset; unsigned int last_timestamp; boolean_t failed; vm_map_lock_read(map); VM_MAP_RANGE_CHECK(map, start, end); if (!vm_map_lookup_entry(map, start, &entry)) { vm_map_unlock_read(map); return (KERN_INVALID_ADDRESS); } else if (start == end) { start = entry->start; end = entry->end; } /* * Make a first pass to check for user-wired memory and holes. */ for (current = entry; current != &map->header && current->start < end; current = current->next) { if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) { vm_map_unlock_read(map); return (KERN_INVALID_ARGUMENT); } if (end > current->end && (current->next == &map->header || current->end != current->next->start)) { vm_map_unlock_read(map); return (KERN_INVALID_ADDRESS); } } if (invalidate) pmap_remove(map->pmap, start, end); failed = FALSE; /* * Make a second pass, cleaning/uncaching pages from the indicated * objects as we go. */ for (current = entry; current != &map->header && current->start < end;) { offset = current->offset + (start - current->start); size = (end <= current->end ? end : current->end) - start; if (current->eflags & MAP_ENTRY_IS_SUB_MAP) { vm_map_t smap; vm_map_entry_t tentry; vm_size_t tsize; smap = current->object.sub_map; vm_map_lock_read(smap); (void) vm_map_lookup_entry(smap, offset, &tentry); tsize = tentry->end - offset; if (tsize < size) size = tsize; object = tentry->object.vm_object; offset = tentry->offset + (offset - tentry->start); vm_map_unlock_read(smap); } else { object = current->object.vm_object; } vm_object_reference(object); last_timestamp = map->timestamp; vm_map_unlock_read(map); if (!vm_object_sync(object, offset, size, syncio, invalidate)) failed = TRUE; start += size; vm_object_deallocate(object); vm_map_lock_read(map); if (last_timestamp == map->timestamp || !vm_map_lookup_entry(map, start, ¤t)) current = current->next; } vm_map_unlock_read(map); return (failed ? KERN_FAILURE : KERN_SUCCESS); } /* * vm_map_entry_unwire: [ internal use only ] * * Make the region specified by this entry pageable. * * The map in question should be locked. * [This is the reason for this routine's existence.] */ static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry) { VM_MAP_ASSERT_LOCKED(map); KASSERT(entry->wired_count > 0, ("vm_map_entry_unwire: entry %p isn't wired", entry)); pmap_unwire(map->pmap, entry->start, entry->end); vm_object_unwire(entry->object.vm_object, entry->offset, entry->end - entry->start, PQ_ACTIVE); entry->wired_count = 0; } static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map) { if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) vm_object_deallocate(entry->object.vm_object); uma_zfree(system_map ? kmapentzone : mapentzone, entry); } /* * vm_map_entry_delete: [ internal use only ] * * Deallocate the given entry from the target map. */ static void vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry) { vm_object_t object; vm_pindex_t offidxstart, offidxend, count, size1; vm_ooffset_t size; vm_map_entry_unlink(map, entry); object = entry->object.vm_object; size = entry->end - entry->start; map->size -= size; if (entry->cred != NULL) { swap_release_by_cred(size, entry->cred); crfree(entry->cred); } if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 && (object != NULL)) { KASSERT(entry->cred == NULL || object->cred == NULL || (entry->eflags & MAP_ENTRY_NEEDS_COPY), ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry)); count = OFF_TO_IDX(size); offidxstart = OFF_TO_IDX(entry->offset); offidxend = offidxstart + count; VM_OBJECT_WLOCK(object); if (object->ref_count != 1 && ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING || object == kernel_object || object == kmem_object)) { vm_object_collapse(object); /* * The option OBJPR_NOTMAPPED can be passed here * because vm_map_delete() already performed * pmap_remove() on the only mapping to this range * of pages. */ vm_object_page_remove(object, offidxstart, offidxend, OBJPR_NOTMAPPED); if (object->type == OBJT_SWAP) swap_pager_freespace(object, offidxstart, count); if (offidxend >= object->size && offidxstart < object->size) { size1 = object->size; object->size = offidxstart; if (object->cred != NULL) { size1 -= object->size; KASSERT(object->charge >= ptoa(size1), ("vm_map_entry_delete: object->charge < 0")); swap_release_by_cred(ptoa(size1), object->cred); object->charge -= ptoa(size1); } } } VM_OBJECT_WUNLOCK(object); } else entry->object.vm_object = NULL; if (map->system_map) vm_map_entry_deallocate(entry, TRUE); else { entry->next = curthread->td_map_def_user; curthread->td_map_def_user = entry; } } /* * vm_map_delete: [ internal use only ] * * Deallocates the given address range from the target * map. */ int vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end) { vm_map_entry_t entry; vm_map_entry_t first_entry; VM_MAP_ASSERT_LOCKED(map); if (start == end) return (KERN_SUCCESS); /* * Find the start of the region, and clip it */ if (!vm_map_lookup_entry(map, start, &first_entry)) entry = first_entry->next; else { entry = first_entry; vm_map_clip_start(map, entry, start); } /* * Step through all entries in this region */ while ((entry != &map->header) && (entry->start < end)) { vm_map_entry_t next; /* * Wait for wiring or unwiring of an entry to complete. * Also wait for any system wirings to disappear on * user maps. */ if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 || (vm_map_pmap(map) != kernel_pmap && vm_map_entry_system_wired_count(entry) != 0)) { unsigned int last_timestamp; vm_offset_t saved_start; vm_map_entry_t tmp_entry; saved_start = entry->start; entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; last_timestamp = map->timestamp; (void) vm_map_unlock_and_wait(map, 0); vm_map_lock(map); if (last_timestamp + 1 != map->timestamp) { /* * Look again for the entry because the map was * modified while it was unlocked. * Specifically, the entry may have been * clipped, merged, or deleted. */ if (!vm_map_lookup_entry(map, saved_start, &tmp_entry)) entry = tmp_entry->next; else { entry = tmp_entry; vm_map_clip_start(map, entry, saved_start); } } continue; } vm_map_clip_end(map, entry, end); next = entry->next; /* * Unwire before removing addresses from the pmap; otherwise, * unwiring will put the entries back in the pmap. */ if (entry->wired_count != 0) { vm_map_entry_unwire(map, entry); } pmap_remove(map->pmap, entry->start, entry->end); /* * Delete the entry only after removing all pmap * entries pointing to its pages. (Otherwise, its * page frames may be reallocated, and any modify bits * will be set in the wrong object!) */ vm_map_entry_delete(map, entry); entry = next; } return (KERN_SUCCESS); } /* * vm_map_remove: * * Remove the given address range from the target map. * This is the exported form of vm_map_delete. */ int vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end) { int result; vm_map_lock(map); VM_MAP_RANGE_CHECK(map, start, end); result = vm_map_delete(map, start, end); vm_map_unlock(map); return (result); } /* * vm_map_check_protection: * * Assert that the target map allows the specified privilege on the * entire address region given. The entire region must be allocated. * * WARNING! This code does not and should not check whether the * contents of the region is accessible. For example a smaller file * might be mapped into a larger address space. * * NOTE! This code is also called by munmap(). * * The map must be locked. A read lock is sufficient. */ boolean_t vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end, vm_prot_t protection) { vm_map_entry_t entry; vm_map_entry_t tmp_entry; if (!vm_map_lookup_entry(map, start, &tmp_entry)) return (FALSE); entry = tmp_entry; while (start < end) { if (entry == &map->header) return (FALSE); /* * No holes allowed! */ if (start < entry->start) return (FALSE); /* * Check protection associated with entry. */ if ((entry->protection & protection) != protection) return (FALSE); /* go to next entry */ start = entry->end; entry = entry->next; } return (TRUE); } /* * vm_map_copy_entry: * * Copies the contents of the source entry to the destination * entry. The entries *must* be aligned properly. */ static void vm_map_copy_entry( vm_map_t src_map, vm_map_t dst_map, vm_map_entry_t src_entry, vm_map_entry_t dst_entry, vm_ooffset_t *fork_charge) { vm_object_t src_object; vm_map_entry_t fake_entry; vm_offset_t size; struct ucred *cred; int charged; VM_MAP_ASSERT_LOCKED(dst_map); if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP) return; if (src_entry->wired_count == 0 || (src_entry->protection & VM_PROT_WRITE) == 0) { /* * If the source entry is marked needs_copy, it is already * write-protected. */ if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 && (src_entry->protection & VM_PROT_WRITE) != 0) { pmap_protect(src_map->pmap, src_entry->start, src_entry->end, src_entry->protection & ~VM_PROT_WRITE); } /* * Make a copy of the object. */ size = src_entry->end - src_entry->start; if ((src_object = src_entry->object.vm_object) != NULL) { VM_OBJECT_WLOCK(src_object); charged = ENTRY_CHARGED(src_entry); if ((src_object->handle == NULL) && (src_object->type == OBJT_DEFAULT || src_object->type == OBJT_SWAP)) { vm_object_collapse(src_object); if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) { vm_object_split(src_entry); src_object = src_entry->object.vm_object; } } vm_object_reference_locked(src_object); vm_object_clear_flag(src_object, OBJ_ONEMAPPING); if (src_entry->cred != NULL && !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { KASSERT(src_object->cred == NULL, ("OVERCOMMIT: vm_map_copy_entry: cred %p", src_object)); src_object->cred = src_entry->cred; src_object->charge = size; } VM_OBJECT_WUNLOCK(src_object); dst_entry->object.vm_object = src_object; if (charged) { cred = curthread->td_ucred; crhold(cred); dst_entry->cred = cred; *fork_charge += size; if (!(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) { crhold(cred); src_entry->cred = cred; *fork_charge += size; } } src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY); dst_entry->offset = src_entry->offset; if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) { /* * MAP_ENTRY_VN_WRITECNT cannot * indicate write reference from * src_entry, since the entry is * marked as needs copy. Allocate a * fake entry that is used to * decrement object->un_pager.vnp.writecount * at the appropriate time. Attach * fake_entry to the deferred list. */ fake_entry = vm_map_entry_create(dst_map); fake_entry->eflags = MAP_ENTRY_VN_WRITECNT; src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT; vm_object_reference(src_object); fake_entry->object.vm_object = src_object; fake_entry->start = src_entry->start; fake_entry->end = src_entry->end; fake_entry->next = curthread->td_map_def_user; curthread->td_map_def_user = fake_entry; } } else { dst_entry->object.vm_object = NULL; dst_entry->offset = 0; if (src_entry->cred != NULL) { dst_entry->cred = curthread->td_ucred; crhold(dst_entry->cred); *fork_charge += size; } } pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start, dst_entry->end - dst_entry->start, src_entry->start); } else { /* * We don't want to make writeable wired pages copy-on-write. * Immediately copy these pages into the new map by simulating * page faults. The new pages are pageable. */ vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry, fork_charge); } } /* * vmspace_map_entry_forked: * Update the newly-forked vmspace each time a map entry is inherited * or copied. The values for vm_dsize and vm_tsize are approximate * (and mostly-obsolete ideas in the face of mmap(2) et al.) */ static void vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2, vm_map_entry_t entry) { vm_size_t entrysize; vm_offset_t newend; entrysize = entry->end - entry->start; vm2->vm_map.size += entrysize; if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) { vm2->vm_ssize += btoc(entrysize); } else if (entry->start >= (vm_offset_t)vm1->vm_daddr && entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) { newend = MIN(entry->end, (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)); vm2->vm_dsize += btoc(newend - entry->start); } else if (entry->start >= (vm_offset_t)vm1->vm_taddr && entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) { newend = MIN(entry->end, (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)); vm2->vm_tsize += btoc(newend - entry->start); } } /* * vmspace_fork: * Create a new process vmspace structure and vm_map * based on those of an existing process. The new map * is based on the old map, according to the inheritance * values on the regions in that map. * * XXX It might be worth coalescing the entries added to the new vmspace. * * The source map must not be locked. */ struct vmspace * vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge) { struct vmspace *vm2; vm_map_t new_map, old_map; vm_map_entry_t new_entry, old_entry; vm_object_t object; int locked; old_map = &vm1->vm_map; /* Copy immutable fields of vm1 to vm2. */ vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, NULL); if (vm2 == NULL) return (NULL); vm2->vm_taddr = vm1->vm_taddr; vm2->vm_daddr = vm1->vm_daddr; vm2->vm_maxsaddr = vm1->vm_maxsaddr; vm_map_lock(old_map); if (old_map->busy) vm_map_wait_busy(old_map); new_map = &vm2->vm_map; locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */ KASSERT(locked, ("vmspace_fork: lock failed")); old_entry = old_map->header.next; while (old_entry != &old_map->header) { if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP) panic("vm_map_fork: encountered a submap"); switch (old_entry->inheritance) { case VM_INHERIT_NONE: break; case VM_INHERIT_SHARE: /* * Clone the entry, creating the shared object if necessary. */ object = old_entry->object.vm_object; if (object == NULL) { object = vm_object_allocate(OBJT_DEFAULT, atop(old_entry->end - old_entry->start)); old_entry->object.vm_object = object; old_entry->offset = 0; if (old_entry->cred != NULL) { object->cred = old_entry->cred; object->charge = old_entry->end - old_entry->start; old_entry->cred = NULL; } } /* * Add the reference before calling vm_object_shadow * to insure that a shadow object is created. */ vm_object_reference(object); if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) { vm_object_shadow(&old_entry->object.vm_object, &old_entry->offset, old_entry->end - old_entry->start); old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; /* Transfer the second reference too. */ vm_object_reference( old_entry->object.vm_object); /* * As in vm_map_simplify_entry(), the * vnode lock will not be acquired in * this call to vm_object_deallocate(). */ vm_object_deallocate(object); object = old_entry->object.vm_object; } VM_OBJECT_WLOCK(object); vm_object_clear_flag(object, OBJ_ONEMAPPING); if (old_entry->cred != NULL) { KASSERT(object->cred == NULL, ("vmspace_fork both cred")); object->cred = old_entry->cred; object->charge = old_entry->end - old_entry->start; old_entry->cred = NULL; } /* * Assert the correct state of the vnode * v_writecount while the object is locked, to * not relock it later for the assertion * correctness. */ if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT && object->type == OBJT_VNODE) { KASSERT(((struct vnode *)object->handle)-> v_writecount > 0, ("vmspace_fork: v_writecount %p", object)); KASSERT(object->un_pager.vnp.writemappings > 0, ("vmspace_fork: vnp.writecount %p", object)); } VM_OBJECT_WUNLOCK(object); /* * Clone the entry, referencing the shared object. */ new_entry = vm_map_entry_create(new_map); *new_entry = *old_entry; new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION); new_entry->wiring_thread = NULL; new_entry->wired_count = 0; if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) { vnode_pager_update_writecount(object, new_entry->start, new_entry->end); } /* * Insert the entry into the new map -- we know we're * inserting at the end of the new map. */ vm_map_entry_link(new_map, new_map->header.prev, new_entry); vmspace_map_entry_forked(vm1, vm2, new_entry); /* * Update the physical map */ pmap_copy(new_map->pmap, old_map->pmap, new_entry->start, (old_entry->end - old_entry->start), old_entry->start); break; case VM_INHERIT_COPY: /* * Clone the entry and link into the map. */ new_entry = vm_map_entry_create(new_map); *new_entry = *old_entry; /* * Copied entry is COW over the old object. */ new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT); new_entry->wiring_thread = NULL; new_entry->wired_count = 0; new_entry->object.vm_object = NULL; new_entry->cred = NULL; vm_map_entry_link(new_map, new_map->header.prev, new_entry); vmspace_map_entry_forked(vm1, vm2, new_entry); vm_map_copy_entry(old_map, new_map, old_entry, new_entry, fork_charge); break; } old_entry = old_entry->next; } /* * Use inlined vm_map_unlock() to postpone handling the deferred * map entries, which cannot be done until both old_map and * new_map locks are released. */ sx_xunlock(&old_map->lock); sx_xunlock(&new_map->lock); vm_map_process_deferred(); return (vm2); } int vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, vm_prot_t prot, vm_prot_t max, int cow) { vm_size_t growsize, init_ssize; rlim_t lmemlim, vmemlim; int rv; growsize = sgrowsiz; init_ssize = (max_ssize < growsize) ? max_ssize : growsize; vm_map_lock(map); - PROC_LOCK(curproc); - lmemlim = lim_cur(curproc, RLIMIT_MEMLOCK); - vmemlim = lim_cur(curproc, RLIMIT_VMEM); - PROC_UNLOCK(curproc); + lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK); + vmemlim = lim_cur(curthread, RLIMIT_VMEM); if (!old_mlock && map->flags & MAP_WIREFUTURE) { if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) { rv = KERN_NO_SPACE; goto out; } } /* If we would blow our VMEM resource limit, no go */ if (map->size + init_ssize > vmemlim) { rv = KERN_NO_SPACE; goto out; } rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot, max, cow); out: vm_map_unlock(map); return (rv); } static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow) { vm_map_entry_t new_entry, prev_entry; vm_offset_t bot, top; vm_size_t init_ssize; int orient, rv; /* * The stack orientation is piggybacked with the cow argument. * Extract it into orient and mask the cow argument so that we * don't pass it around further. * NOTE: We explicitly allow bi-directional stacks. */ orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP); KASSERT(orient != 0, ("No stack grow direction")); if (addrbos < vm_map_min(map) || addrbos > vm_map_max(map) || addrbos + max_ssize < addrbos) return (KERN_NO_SPACE); init_ssize = (max_ssize < growsize) ? max_ssize : growsize; /* If addr is already mapped, no go */ if (vm_map_lookup_entry(map, addrbos, &prev_entry)) return (KERN_NO_SPACE); /* * If we can't accomodate max_ssize in the current mapping, no go. * However, we need to be aware that subsequent user mappings might * map into the space we have reserved for stack, and currently this * space is not protected. * * Hopefully we will at least detect this condition when we try to * grow the stack. */ if ((prev_entry->next != &map->header) && (prev_entry->next->start < addrbos + max_ssize)) return (KERN_NO_SPACE); /* * We initially map a stack of only init_ssize. We will grow as * needed later. Depending on the orientation of the stack (i.e. * the grow direction) we either map at the top of the range, the * bottom of the range or in the middle. * * Note: we would normally expect prot and max to be VM_PROT_ALL, * and cow to be 0. Possibly we should eliminate these as input * parameters, and just pass these values here in the insert call. */ if (orient == MAP_STACK_GROWS_DOWN) bot = addrbos + max_ssize - init_ssize; else if (orient == MAP_STACK_GROWS_UP) bot = addrbos; else bot = round_page(addrbos + max_ssize/2 - init_ssize/2); top = bot + init_ssize; rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow); /* Now set the avail_ssize amount. */ if (rv == KERN_SUCCESS) { new_entry = prev_entry->next; if (new_entry->end != top || new_entry->start != bot) panic("Bad entry start/end for new stack entry"); new_entry->avail_ssize = max_ssize - init_ssize; KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 || (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0, ("new entry lacks MAP_ENTRY_GROWS_DOWN")); KASSERT((orient & MAP_STACK_GROWS_UP) == 0 || (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0, ("new entry lacks MAP_ENTRY_GROWS_UP")); } return (rv); } static int stack_guard_page = 0; SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN, &stack_guard_page, 0, "Insert stack guard page ahead of the growable segments."); /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the * desired address is already mapped, or if we successfully grow * the stack. Also returns KERN_SUCCESS if addr is outside the * stack range (this is strange, but preserves compatibility with * the grow function in vm_machdep.c). */ int vm_map_growstack(struct proc *p, vm_offset_t addr) { vm_map_entry_t next_entry, prev_entry; vm_map_entry_t new_entry, stack_entry; struct vmspace *vm = p->p_vmspace; vm_map_t map = &vm->vm_map; vm_offset_t end; vm_size_t growsize; size_t grow_amount, max_grow; rlim_t lmemlim, stacklim, vmemlim; int is_procstack, rv; struct ucred *cred; #ifdef notyet uint64_t limit; #endif #ifdef RACCT int error; #endif + lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK); + stacklim = lim_cur(curthread, RLIMIT_STACK); + vmemlim = lim_cur(curthread, RLIMIT_VMEM); Retry: - PROC_LOCK(p); - lmemlim = lim_cur(p, RLIMIT_MEMLOCK); - stacklim = lim_cur(p, RLIMIT_STACK); - vmemlim = lim_cur(p, RLIMIT_VMEM); - PROC_UNLOCK(p); vm_map_lock_read(map); /* If addr is already in the entry range, no need to grow.*/ if (vm_map_lookup_entry(map, addr, &prev_entry)) { vm_map_unlock_read(map); return (KERN_SUCCESS); } next_entry = prev_entry->next; if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) { /* * This entry does not grow upwards. Since the address lies * beyond this entry, the next entry (if one exists) has to * be a downward growable entry. The entry list header is * never a growable entry, so it suffices to check the flags. */ if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) { vm_map_unlock_read(map); return (KERN_SUCCESS); } stack_entry = next_entry; } else { /* * This entry grows upward. If the next entry does not at * least grow downwards, this is the entry we need to grow. * otherwise we have two possible choices and we have to * select one. */ if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) { /* * We have two choices; grow the entry closest to * the address to minimize the amount of growth. */ if (addr - prev_entry->end <= next_entry->start - addr) stack_entry = prev_entry; else stack_entry = next_entry; } else stack_entry = prev_entry; } if (stack_entry == next_entry) { KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo")); KASSERT(addr < stack_entry->start, ("foo")); end = (prev_entry != &map->header) ? prev_entry->end : stack_entry->start - stack_entry->avail_ssize; grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE); max_grow = stack_entry->start - end; } else { KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo")); KASSERT(addr >= stack_entry->end, ("foo")); end = (next_entry != &map->header) ? next_entry->start : stack_entry->end + stack_entry->avail_ssize; grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE); max_grow = end - stack_entry->end; } if (grow_amount > stack_entry->avail_ssize) { vm_map_unlock_read(map); return (KERN_NO_SPACE); } /* * If there is no longer enough space between the entries nogo, and * adjust the available space. Note: this should only happen if the * user has mapped into the stack area after the stack was created, * and is probably an error. * * This also effectively destroys any guard page the user might have * intended by limiting the stack size. */ if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) { if (vm_map_lock_upgrade(map)) goto Retry; stack_entry->avail_ssize = max_grow; vm_map_unlock(map); return (KERN_NO_SPACE); } is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0; /* * If this is the main process stack, see if we're over the stack * limit. */ if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { vm_map_unlock_read(map); return (KERN_NO_SPACE); } #ifdef RACCT if (racct_enable) { PROC_LOCK(p); if (is_procstack && racct_set(p, RACCT_STACK, ctob(vm->vm_ssize) + grow_amount)) { PROC_UNLOCK(p); vm_map_unlock_read(map); return (KERN_NO_SPACE); } PROC_UNLOCK(p); } #endif /* Round up the grow amount modulo sgrowsiz */ growsize = sgrowsiz; grow_amount = roundup(grow_amount, growsize); if (grow_amount > stack_entry->avail_ssize) grow_amount = stack_entry->avail_ssize; if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) { grow_amount = trunc_page((vm_size_t)stacklim) - ctob(vm->vm_ssize); } #ifdef notyet PROC_LOCK(p); limit = racct_get_available(p, RACCT_STACK); PROC_UNLOCK(p); if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit)) grow_amount = limit - ctob(vm->vm_ssize); #endif if (!old_mlock && map->flags & MAP_WIREFUTURE) { if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) { vm_map_unlock_read(map); rv = KERN_NO_SPACE; goto out; } #ifdef RACCT if (racct_enable) { PROC_LOCK(p); if (racct_set(p, RACCT_MEMLOCK, ptoa(pmap_wired_count(map->pmap)) + grow_amount)) { PROC_UNLOCK(p); vm_map_unlock_read(map); rv = KERN_NO_SPACE; goto out; } PROC_UNLOCK(p); } #endif } /* If we would blow our VMEM resource limit, no go */ if (map->size + grow_amount > vmemlim) { vm_map_unlock_read(map); rv = KERN_NO_SPACE; goto out; } #ifdef RACCT if (racct_enable) { PROC_LOCK(p); if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) { PROC_UNLOCK(p); vm_map_unlock_read(map); rv = KERN_NO_SPACE; goto out; } PROC_UNLOCK(p); } #endif if (vm_map_lock_upgrade(map)) goto Retry; if (stack_entry == next_entry) { /* * Growing downward. */ /* Get the preliminary new entry start value */ addr = stack_entry->start - grow_amount; /* * If this puts us into the previous entry, cut back our * growth to the available space. Also, see the note above. */ if (addr < end) { stack_entry->avail_ssize = max_grow; addr = end; if (stack_guard_page) addr += PAGE_SIZE; } rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start, next_entry->protection, next_entry->max_protection, MAP_STACK_GROWS_DOWN); /* Adjust the available stack space by the amount we grew. */ if (rv == KERN_SUCCESS) { new_entry = prev_entry->next; KASSERT(new_entry == stack_entry->prev, ("foo")); KASSERT(new_entry->end == stack_entry->start, ("foo")); KASSERT(new_entry->start == addr, ("foo")); KASSERT((new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0, ("new entry lacks MAP_ENTRY_GROWS_DOWN")); grow_amount = new_entry->end - new_entry->start; new_entry->avail_ssize = stack_entry->avail_ssize - grow_amount; stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN; } } else { /* * Growing upward. */ addr = stack_entry->end + grow_amount; /* * If this puts us into the next entry, cut back our growth * to the available space. Also, see the note above. */ if (addr > end) { stack_entry->avail_ssize = end - stack_entry->end; addr = end; if (stack_guard_page) addr -= PAGE_SIZE; } grow_amount = addr - stack_entry->end; cred = stack_entry->cred; if (cred == NULL && stack_entry->object.vm_object != NULL) cred = stack_entry->object.vm_object->cred; if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred)) rv = KERN_NO_SPACE; /* Grow the underlying object if applicable. */ else if (stack_entry->object.vm_object == NULL || vm_object_coalesce(stack_entry->object.vm_object, stack_entry->offset, (vm_size_t)(stack_entry->end - stack_entry->start), (vm_size_t)grow_amount, cred != NULL)) { map->size += (addr - stack_entry->end); /* Update the current entry. */ stack_entry->end = addr; stack_entry->avail_ssize -= grow_amount; vm_map_entry_resize_free(map, stack_entry); rv = KERN_SUCCESS; } else rv = KERN_FAILURE; } if (rv == KERN_SUCCESS && is_procstack) vm->vm_ssize += btoc(grow_amount); vm_map_unlock(map); /* * Heed the MAP_WIREFUTURE flag if it was set for this process. */ if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) { vm_map_wire(map, (stack_entry == next_entry) ? addr : addr - grow_amount, (stack_entry == next_entry) ? stack_entry->start : addr, (p->p_flag & P_SYSTEM) ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES); } out: #ifdef RACCT if (racct_enable && rv != KERN_SUCCESS) { PROC_LOCK(p); error = racct_set(p, RACCT_VMEM, map->size); KASSERT(error == 0, ("decreasing RACCT_VMEM failed")); if (!old_mlock) { error = racct_set(p, RACCT_MEMLOCK, ptoa(pmap_wired_count(map->pmap))); KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed")); } error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize)); KASSERT(error == 0, ("decreasing RACCT_STACK failed")); PROC_UNLOCK(p); } #endif return (rv); } /* * Unshare the specified VM space for exec. If other processes are * mapped to it, then create a new one. The new vmspace is null. */ int vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser) { struct vmspace *oldvmspace = p->p_vmspace; struct vmspace *newvmspace; KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0, ("vmspace_exec recursed")); newvmspace = vmspace_alloc(minuser, maxuser, NULL); if (newvmspace == NULL) return (ENOMEM); newvmspace->vm_swrss = oldvmspace->vm_swrss; /* * This code is written like this for prototype purposes. The * goal is to avoid running down the vmspace here, but let the * other process's that are still using the vmspace to finally * run it down. Even though there is little or no chance of blocking * here, it is a good idea to keep this form for future mods. */ PROC_VMSPACE_LOCK(p); p->p_vmspace = newvmspace; PROC_VMSPACE_UNLOCK(p); if (p == curthread->td_proc) pmap_activate(curthread); curthread->td_pflags |= TDP_EXECVMSPC; return (0); } /* * Unshare the specified VM space for forcing COW. This * is called by rfork, for the (RFMEM|RFPROC) == 0 case. */ int vmspace_unshare(struct proc *p) { struct vmspace *oldvmspace = p->p_vmspace; struct vmspace *newvmspace; vm_ooffset_t fork_charge; if (oldvmspace->vm_refcnt == 1) return (0); fork_charge = 0; newvmspace = vmspace_fork(oldvmspace, &fork_charge); if (newvmspace == NULL) return (ENOMEM); if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) { vmspace_free(newvmspace); return (ENOMEM); } PROC_VMSPACE_LOCK(p); p->p_vmspace = newvmspace; PROC_VMSPACE_UNLOCK(p); if (p == curthread->td_proc) pmap_activate(curthread); vmspace_free(oldvmspace); return (0); } /* * vm_map_lookup: * * Finds the VM object, offset, and * protection for a given virtual address in the * specified map, assuming a page fault of the * type specified. * * Leaves the map in question locked for read; return * values are guaranteed until a vm_map_lookup_done * call is performed. Note that the map argument * is in/out; the returned map must be used in * the call to vm_map_lookup_done. * * A handle (out_entry) is returned for use in * vm_map_lookup_done, to make that fast. * * If a lookup is requested with "write protection" * specified, the map may be changed to perform virtual * copying operations, although the data referenced will * remain the same. */ int vm_map_lookup(vm_map_t *var_map, /* IN/OUT */ vm_offset_t vaddr, vm_prot_t fault_typea, vm_map_entry_t *out_entry, /* OUT */ vm_object_t *object, /* OUT */ vm_pindex_t *pindex, /* OUT */ vm_prot_t *out_prot, /* OUT */ boolean_t *wired) /* OUT */ { vm_map_entry_t entry; vm_map_t map = *var_map; vm_prot_t prot; vm_prot_t fault_type = fault_typea; vm_object_t eobject; vm_size_t size; struct ucred *cred; RetryLookup:; vm_map_lock_read(map); /* * Lookup the faulting address. */ if (!vm_map_lookup_entry(map, vaddr, out_entry)) { vm_map_unlock_read(map); return (KERN_INVALID_ADDRESS); } entry = *out_entry; /* * Handle submaps. */ if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { vm_map_t old_map = map; *var_map = map = entry->object.sub_map; vm_map_unlock_read(old_map); goto RetryLookup; } /* * Check whether this task is allowed to have this page. */ prot = entry->protection; fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE); if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) { vm_map_unlock_read(map); return (KERN_PROTECTION_FAILURE); } if ((entry->eflags & MAP_ENTRY_USER_WIRED) && (entry->eflags & MAP_ENTRY_COW) && (fault_type & VM_PROT_WRITE)) { vm_map_unlock_read(map); return (KERN_PROTECTION_FAILURE); } if ((fault_typea & VM_PROT_COPY) != 0 && (entry->max_protection & VM_PROT_WRITE) == 0 && (entry->eflags & MAP_ENTRY_COW) == 0) { vm_map_unlock_read(map); return (KERN_PROTECTION_FAILURE); } /* * If this page is not pageable, we have to get it for all possible * accesses. */ *wired = (entry->wired_count != 0); if (*wired) fault_type = entry->protection; size = entry->end - entry->start; /* * If the entry was copy-on-write, we either ... */ if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { /* * If we want to write the page, we may as well handle that * now since we've got the map locked. * * If we don't need to write the page, we just demote the * permissions allowed. */ if ((fault_type & VM_PROT_WRITE) != 0 || (fault_typea & VM_PROT_COPY) != 0) { /* * Make a new object, and place it in the object * chain. Note that no new references have appeared * -- one just moved from the map to the new * object. */ if (vm_map_lock_upgrade(map)) goto RetryLookup; if (entry->cred == NULL) { /* * The debugger owner is charged for * the memory. */ cred = curthread->td_ucred; crhold(cred); if (!swap_reserve_by_cred(size, cred)) { crfree(cred); vm_map_unlock(map); return (KERN_RESOURCE_SHORTAGE); } entry->cred = cred; } vm_object_shadow(&entry->object.vm_object, &entry->offset, size); entry->eflags &= ~MAP_ENTRY_NEEDS_COPY; eobject = entry->object.vm_object; if (eobject->cred != NULL) { /* * The object was not shadowed. */ swap_release_by_cred(size, entry->cred); crfree(entry->cred); entry->cred = NULL; } else if (entry->cred != NULL) { VM_OBJECT_WLOCK(eobject); eobject->cred = entry->cred; eobject->charge = size; VM_OBJECT_WUNLOCK(eobject); entry->cred = NULL; } vm_map_lock_downgrade(map); } else { /* * We're attempting to read a copy-on-write page -- * don't allow writes. */ prot &= ~VM_PROT_WRITE; } } /* * Create an object if necessary. */ if (entry->object.vm_object == NULL && !map->system_map) { if (vm_map_lock_upgrade(map)) goto RetryLookup; entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT, atop(size)); entry->offset = 0; if (entry->cred != NULL) { VM_OBJECT_WLOCK(entry->object.vm_object); entry->object.vm_object->cred = entry->cred; entry->object.vm_object->charge = size; VM_OBJECT_WUNLOCK(entry->object.vm_object); entry->cred = NULL; } vm_map_lock_downgrade(map); } /* * Return the object/offset from this entry. If the entry was * copy-on-write or empty, it has been fixed up. */ *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); *object = entry->object.vm_object; *out_prot = prot; return (KERN_SUCCESS); } /* * vm_map_lookup_locked: * * Lookup the faulting address. A version of vm_map_lookup that returns * KERN_FAILURE instead of blocking on map lock or memory allocation. */ int vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */ vm_offset_t vaddr, vm_prot_t fault_typea, vm_map_entry_t *out_entry, /* OUT */ vm_object_t *object, /* OUT */ vm_pindex_t *pindex, /* OUT */ vm_prot_t *out_prot, /* OUT */ boolean_t *wired) /* OUT */ { vm_map_entry_t entry; vm_map_t map = *var_map; vm_prot_t prot; vm_prot_t fault_type = fault_typea; /* * Lookup the faulting address. */ if (!vm_map_lookup_entry(map, vaddr, out_entry)) return (KERN_INVALID_ADDRESS); entry = *out_entry; /* * Fail if the entry refers to a submap. */ if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) return (KERN_FAILURE); /* * Check whether this task is allowed to have this page. */ prot = entry->protection; fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE; if ((fault_type & prot) != fault_type) return (KERN_PROTECTION_FAILURE); if ((entry->eflags & MAP_ENTRY_USER_WIRED) && (entry->eflags & MAP_ENTRY_COW) && (fault_type & VM_PROT_WRITE)) return (KERN_PROTECTION_FAILURE); /* * If this page is not pageable, we have to get it for all possible * accesses. */ *wired = (entry->wired_count != 0); if (*wired) fault_type = entry->protection; if (entry->eflags & MAP_ENTRY_NEEDS_COPY) { /* * Fail if the entry was copy-on-write for a write fault. */ if (fault_type & VM_PROT_WRITE) return (KERN_FAILURE); /* * We're attempting to read a copy-on-write page -- * don't allow writes. */ prot &= ~VM_PROT_WRITE; } /* * Fail if an object should be created. */ if (entry->object.vm_object == NULL && !map->system_map) return (KERN_FAILURE); /* * Return the object/offset from this entry. If the entry was * copy-on-write or empty, it has been fixed up. */ *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset); *object = entry->object.vm_object; *out_prot = prot; return (KERN_SUCCESS); } /* * vm_map_lookup_done: * * Releases locks acquired by a vm_map_lookup * (according to the handle returned by that lookup). */ void vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry) { /* * Unlock the main-level map */ vm_map_unlock_read(map); } #include "opt_ddb.h" #ifdef DDB #include #include static void vm_map_print(vm_map_t map) { vm_map_entry_t entry; db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n", (void *)map, (void *)map->pmap, map->nentries, map->timestamp); db_indent += 2; for (entry = map->header.next; entry != &map->header; entry = entry->next) { db_iprintf("map entry %p: start=%p, end=%p\n", (void *)entry, (void *)entry->start, (void *)entry->end); { static char *inheritance_name[4] = {"share", "copy", "none", "donate_copy"}; db_iprintf(" prot=%x/%x/%s", entry->protection, entry->max_protection, inheritance_name[(int)(unsigned char)entry->inheritance]); if (entry->wired_count != 0) db_printf(", wired"); } if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { db_printf(", share=%p, offset=0x%jx\n", (void *)entry->object.sub_map, (uintmax_t)entry->offset); if ((entry->prev == &map->header) || (entry->prev->object.sub_map != entry->object.sub_map)) { db_indent += 2; vm_map_print((vm_map_t)entry->object.sub_map); db_indent -= 2; } } else { if (entry->cred != NULL) db_printf(", ruid %d", entry->cred->cr_ruid); db_printf(", object=%p, offset=0x%jx", (void *)entry->object.vm_object, (uintmax_t)entry->offset); if (entry->object.vm_object && entry->object.vm_object->cred) db_printf(", obj ruid %d charge %jx", entry->object.vm_object->cred->cr_ruid, (uintmax_t)entry->object.vm_object->charge); if (entry->eflags & MAP_ENTRY_COW) db_printf(", copy (%s)", (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done"); db_printf("\n"); if ((entry->prev == &map->header) || (entry->prev->object.vm_object != entry->object.vm_object)) { db_indent += 2; vm_object_print((db_expr_t)(intptr_t) entry->object.vm_object, 0, 0, (char *)0); db_indent -= 2; } } } db_indent -= 2; } DB_SHOW_COMMAND(map, map) { if (!have_addr) { db_printf("usage: show map \n"); return; } vm_map_print((vm_map_t)addr); } DB_SHOW_COMMAND(procvm, procvm) { struct proc *p; if (have_addr) { p = (struct proc *) addr; } else { p = curproc; } db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n", (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map, (void *)vmspace_pmap(p->p_vmspace)); vm_map_print((vm_map_t)&p->p_vmspace->vm_map); } #endif /* DDB */ Index: head/sys/vm/vm_mmap.c =================================================================== --- head/sys/vm/vm_mmap.c (revision 284214) +++ head/sys/vm/vm_mmap.c (revision 284215) @@ -1,1611 +1,1611 @@ /*- * Copyright (c) 1988 University of Utah. * 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 Systems Programming Group of the University of Utah Computer * Science Department. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$ * * @(#)vm_mmap.c 8.4 (Berkeley) 1/12/94 */ /* * Mapped file (mmap) interface to VM */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #include "opt_hwpmc_hooks.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 #include #include #include #include #include #include #ifdef HWPMC_HOOKS #include #endif int old_mlock = 0; SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0, "Do not apply RLIMIT_MEMLOCK on mlockall"); #ifdef MAP_32BIT #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31) #endif #ifndef _SYS_SYSPROTO_H_ struct sbrk_args { int incr; }; #endif /* * MPSAFE */ /* ARGSUSED */ int sys_sbrk(td, uap) struct thread *td; struct sbrk_args *uap; { /* Not yet implemented */ return (EOPNOTSUPP); } #ifndef _SYS_SYSPROTO_H_ struct sstk_args { int incr; }; #endif /* * MPSAFE */ /* ARGSUSED */ int sys_sstk(td, uap) struct thread *td; struct sstk_args *uap; { /* Not yet implemented */ return (EOPNOTSUPP); } #if defined(COMPAT_43) #ifndef _SYS_SYSPROTO_H_ struct getpagesize_args { int dummy; }; #endif int ogetpagesize(td, uap) struct thread *td; struct getpagesize_args *uap; { /* MP SAFE */ td->td_retval[0] = PAGE_SIZE; return (0); } #endif /* COMPAT_43 */ /* * Memory Map (mmap) system call. Note that the file offset * and address are allowed to be NOT page aligned, though if * the MAP_FIXED flag it set, both must have the same remainder * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not * page-aligned, the actual mapping starts at trunc_page(addr) * and the return value is adjusted up by the page offset. * * Generally speaking, only character devices which are themselves * memory-based, such as a video framebuffer, can be mmap'd. Otherwise * there would be no cache coherency between a descriptor and a VM mapping * both to the same character device. */ #ifndef _SYS_SYSPROTO_H_ struct mmap_args { void *addr; size_t len; int prot; int flags; int fd; long pad; off_t pos; }; #endif /* * MPSAFE */ int sys_mmap(td, uap) struct thread *td; struct mmap_args *uap; { struct file *fp; vm_offset_t addr; vm_size_t size, pageoff; vm_prot_t cap_maxprot; int align, error, flags, prot; off_t pos; struct vmspace *vms = td->td_proc->p_vmspace; cap_rights_t rights; addr = (vm_offset_t) uap->addr; size = uap->len; prot = uap->prot; flags = uap->flags; pos = uap->pos; fp = NULL; /* * Ignore old flags that used to be defined but did not do anything. */ flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040); /* * Enforce the constraints. * Mapping of length 0 is only allowed for old binaries. * Anonymous mapping shall specify -1 as filedescriptor and * zero position for new code. Be nice to ancient a.out * binaries and correct pos for anonymous mapping, since old * ld.so sometimes issues anonymous map requests with non-zero * pos. */ if (!SV_CURPROC_FLAG(SV_AOUT)) { if ((uap->len == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) || ((flags & MAP_ANON) != 0 && (uap->fd != -1 || pos != 0))) return (EINVAL); } else { if ((flags & MAP_ANON) != 0) pos = 0; } if (flags & MAP_STACK) { if ((uap->fd != -1) || ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE))) return (EINVAL); flags |= MAP_ANON; pos = 0; } if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE | MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE | MAP_PREFAULT_READ | #ifdef MAP_32BIT MAP_32BIT | #endif MAP_ALIGNMENT_MASK)) != 0) return (EINVAL); if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL) return (EINVAL); if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE)) return (EINVAL); if (prot != PROT_NONE && (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0) return (EINVAL); /* * Align the file position to a page boundary, * and save its page offset component. */ pageoff = (pos & PAGE_MASK); pos -= pageoff; /* Adjust size for rounding (on both ends). */ size += pageoff; /* low end... */ size = (vm_size_t) round_page(size); /* hi end */ /* Ensure alignment is at least a page and fits in a pointer. */ align = flags & MAP_ALIGNMENT_MASK; if (align != 0 && align != MAP_ALIGNED_SUPER && (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY || align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT)) return (EINVAL); /* * Check for illegal addresses. Watch out for address wrap... Note * that VM_*_ADDRESS are not constants due to casts (argh). */ if (flags & MAP_FIXED) { /* * The specified address must have the same remainder * as the file offset taken modulo PAGE_SIZE, so it * should be aligned after adjustment by pageoff. */ addr -= pageoff; if (addr & PAGE_MASK) return (EINVAL); /* Address range must be all in user VM space. */ if (addr < vm_map_min(&vms->vm_map) || addr + size > vm_map_max(&vms->vm_map)) return (EINVAL); if (addr + size < addr) return (EINVAL); #ifdef MAP_32BIT if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR) return (EINVAL); } else if (flags & MAP_32BIT) { /* * For MAP_32BIT, override the hint if it is too high and * do not bother moving the mapping past the heap (since * the heap is usually above 2GB). */ if (addr + size > MAP_32BIT_MAX_ADDR) addr = 0; #endif } else { /* * XXX for non-fixed mappings where no hint is provided or * the hint would fall in the potential heap space, * place it after the end of the largest possible heap. * * There should really be a pmap call to determine a reasonable * location. */ PROC_LOCK(td->td_proc); if (addr == 0 || (addr >= round_page((vm_offset_t)vms->vm_taddr) && addr < round_page((vm_offset_t)vms->vm_daddr + - lim_max(td->td_proc, RLIMIT_DATA)))) + lim_max_proc(td->td_proc, RLIMIT_DATA)))) addr = round_page((vm_offset_t)vms->vm_daddr + - lim_max(td->td_proc, RLIMIT_DATA)); + lim_max_proc(td->td_proc, RLIMIT_DATA)); PROC_UNLOCK(td->td_proc); } if (size == 0) { /* * Return success without mapping anything for old * binaries that request a page-aligned mapping of * length 0. For modern binaries, this function * returns an error earlier. */ error = 0; } else if (flags & MAP_ANON) { /* * Mapping blank space is trivial. * * This relies on VM_PROT_* matching PROT_*. */ error = vm_mmap_object(&vms->vm_map, &addr, size, prot, VM_PROT_ALL, flags, NULL, pos, FALSE, td); } else { /* * Mapping file, get fp for validation and don't let the * descriptor disappear on us if we block. Check capability * rights, but also return the maximum rights to be combined * with maxprot later. */ cap_rights_init(&rights, CAP_MMAP); if (prot & PROT_READ) cap_rights_set(&rights, CAP_MMAP_R); if ((flags & MAP_SHARED) != 0) { if (prot & PROT_WRITE) cap_rights_set(&rights, CAP_MMAP_W); } if (prot & PROT_EXEC) cap_rights_set(&rights, CAP_MMAP_X); error = fget_mmap(td, uap->fd, &rights, &cap_maxprot, &fp); if (error != 0) goto done; if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 && td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) { error = EINVAL; goto done; } /* This relies on VM_PROT_* matching PROT_*. */ error = fo_mmap(fp, &vms->vm_map, &addr, size, prot, cap_maxprot, flags, pos, td); } if (error == 0) td->td_retval[0] = (register_t) (addr + pageoff); done: if (fp) fdrop(fp, td); return (error); } #if defined(COMPAT_FREEBSD6) int freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap) { struct mmap_args oargs; oargs.addr = uap->addr; oargs.len = uap->len; oargs.prot = uap->prot; oargs.flags = uap->flags; oargs.fd = uap->fd; oargs.pos = uap->pos; return (sys_mmap(td, &oargs)); } #endif #ifdef COMPAT_43 #ifndef _SYS_SYSPROTO_H_ struct ommap_args { caddr_t addr; int len; int prot; int flags; int fd; long pos; }; #endif int ommap(td, uap) struct thread *td; struct ommap_args *uap; { struct mmap_args nargs; static const char cvtbsdprot[8] = { 0, PROT_EXEC, PROT_WRITE, PROT_EXEC | PROT_WRITE, PROT_READ, PROT_EXEC | PROT_READ, PROT_WRITE | PROT_READ, PROT_EXEC | PROT_WRITE | PROT_READ, }; #define OMAP_ANON 0x0002 #define OMAP_COPY 0x0020 #define OMAP_SHARED 0x0010 #define OMAP_FIXED 0x0100 nargs.addr = uap->addr; nargs.len = uap->len; nargs.prot = cvtbsdprot[uap->prot & 0x7]; #ifdef COMPAT_FREEBSD32 #if defined(__amd64__) if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) && nargs.prot != 0) nargs.prot |= PROT_EXEC; #endif #endif nargs.flags = 0; if (uap->flags & OMAP_ANON) nargs.flags |= MAP_ANON; if (uap->flags & OMAP_COPY) nargs.flags |= MAP_COPY; if (uap->flags & OMAP_SHARED) nargs.flags |= MAP_SHARED; else nargs.flags |= MAP_PRIVATE; if (uap->flags & OMAP_FIXED) nargs.flags |= MAP_FIXED; nargs.fd = uap->fd; nargs.pos = uap->pos; return (sys_mmap(td, &nargs)); } #endif /* COMPAT_43 */ #ifndef _SYS_SYSPROTO_H_ struct msync_args { void *addr; size_t len; int flags; }; #endif /* * MPSAFE */ int sys_msync(td, uap) struct thread *td; struct msync_args *uap; { vm_offset_t addr; vm_size_t size, pageoff; int flags; vm_map_t map; int rv; addr = (vm_offset_t) uap->addr; size = uap->len; flags = uap->flags; pageoff = (addr & PAGE_MASK); addr -= pageoff; size += pageoff; size = (vm_size_t) round_page(size); if (addr + size < addr) return (EINVAL); if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE)) return (EINVAL); map = &td->td_proc->p_vmspace->vm_map; /* * Clean the pages and interpret the return value. */ rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0, (flags & MS_INVALIDATE) != 0); switch (rv) { case KERN_SUCCESS: return (0); case KERN_INVALID_ADDRESS: return (ENOMEM); case KERN_INVALID_ARGUMENT: return (EBUSY); case KERN_FAILURE: return (EIO); default: return (EINVAL); } } #ifndef _SYS_SYSPROTO_H_ struct munmap_args { void *addr; size_t len; }; #endif /* * MPSAFE */ int sys_munmap(td, uap) struct thread *td; struct munmap_args *uap; { #ifdef HWPMC_HOOKS struct pmckern_map_out pkm; vm_map_entry_t entry; #endif vm_offset_t addr; vm_size_t size, pageoff; vm_map_t map; addr = (vm_offset_t) uap->addr; size = uap->len; if (size == 0) return (EINVAL); pageoff = (addr & PAGE_MASK); addr -= pageoff; size += pageoff; size = (vm_size_t) round_page(size); if (addr + size < addr) return (EINVAL); /* * Check for illegal addresses. Watch out for address wrap... */ map = &td->td_proc->p_vmspace->vm_map; if (addr < vm_map_min(map) || addr + size > vm_map_max(map)) return (EINVAL); vm_map_lock(map); #ifdef HWPMC_HOOKS /* * Inform hwpmc if the address range being unmapped contains * an executable region. */ pkm.pm_address = (uintptr_t) NULL; if (vm_map_lookup_entry(map, addr, &entry)) { for (; entry != &map->header && entry->start < addr + size; entry = entry->next) { if (vm_map_check_protection(map, entry->start, entry->end, VM_PROT_EXECUTE) == TRUE) { pkm.pm_address = (uintptr_t) addr; pkm.pm_size = (size_t) size; break; } } } #endif vm_map_delete(map, addr, addr + size); #ifdef HWPMC_HOOKS /* downgrade the lock to prevent a LOR with the pmc-sx lock */ vm_map_lock_downgrade(map); if (pkm.pm_address != (uintptr_t) NULL) PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm); vm_map_unlock_read(map); #else vm_map_unlock(map); #endif /* vm_map_delete returns nothing but KERN_SUCCESS anyway */ return (0); } #ifndef _SYS_SYSPROTO_H_ struct mprotect_args { const void *addr; size_t len; int prot; }; #endif /* * MPSAFE */ int sys_mprotect(td, uap) struct thread *td; struct mprotect_args *uap; { vm_offset_t addr; vm_size_t size, pageoff; vm_prot_t prot; addr = (vm_offset_t) uap->addr; size = uap->len; prot = uap->prot & VM_PROT_ALL; pageoff = (addr & PAGE_MASK); addr -= pageoff; size += pageoff; size = (vm_size_t) round_page(size); if (addr + size < addr) return (EINVAL); switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr, addr + size, prot, FALSE)) { case KERN_SUCCESS: return (0); case KERN_PROTECTION_FAILURE: return (EACCES); case KERN_RESOURCE_SHORTAGE: return (ENOMEM); } return (EINVAL); } #ifndef _SYS_SYSPROTO_H_ struct minherit_args { void *addr; size_t len; int inherit; }; #endif /* * MPSAFE */ int sys_minherit(td, uap) struct thread *td; struct minherit_args *uap; { vm_offset_t addr; vm_size_t size, pageoff; vm_inherit_t inherit; addr = (vm_offset_t)uap->addr; size = uap->len; inherit = uap->inherit; pageoff = (addr & PAGE_MASK); addr -= pageoff; size += pageoff; size = (vm_size_t) round_page(size); if (addr + size < addr) return (EINVAL); switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr, addr + size, inherit)) { case KERN_SUCCESS: return (0); case KERN_PROTECTION_FAILURE: return (EACCES); } return (EINVAL); } #ifndef _SYS_SYSPROTO_H_ struct madvise_args { void *addr; size_t len; int behav; }; #endif /* * MPSAFE */ int sys_madvise(td, uap) struct thread *td; struct madvise_args *uap; { vm_offset_t start, end; vm_map_t map; int flags; /* * Check for our special case, advising the swap pager we are * "immortal." */ if (uap->behav == MADV_PROTECT) { flags = PPROT_SET; return (kern_procctl(td, P_PID, td->td_proc->p_pid, PROC_SPROTECT, &flags)); } /* * Check for illegal behavior */ if (uap->behav < 0 || uap->behav > MADV_CORE) return (EINVAL); /* * Check for illegal addresses. Watch out for address wrap... Note * that VM_*_ADDRESS are not constants due to casts (argh). */ map = &td->td_proc->p_vmspace->vm_map; if ((vm_offset_t)uap->addr < vm_map_min(map) || (vm_offset_t)uap->addr + uap->len > vm_map_max(map)) return (EINVAL); if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr) return (EINVAL); /* * Since this routine is only advisory, we default to conservative * behavior. */ start = trunc_page((vm_offset_t) uap->addr); end = round_page((vm_offset_t) uap->addr + uap->len); if (vm_map_madvise(map, start, end, uap->behav)) return (EINVAL); return (0); } #ifndef _SYS_SYSPROTO_H_ struct mincore_args { const void *addr; size_t len; char *vec; }; #endif /* * MPSAFE */ int sys_mincore(td, uap) struct thread *td; struct mincore_args *uap; { vm_offset_t addr, first_addr; vm_offset_t end, cend; pmap_t pmap; vm_map_t map; char *vec; int error = 0; int vecindex, lastvecindex; vm_map_entry_t current; vm_map_entry_t entry; vm_object_t object; vm_paddr_t locked_pa; vm_page_t m; vm_pindex_t pindex; int mincoreinfo; unsigned int timestamp; boolean_t locked; /* * Make sure that the addresses presented are valid for user * mode. */ first_addr = addr = trunc_page((vm_offset_t) uap->addr); end = addr + (vm_size_t)round_page(uap->len); map = &td->td_proc->p_vmspace->vm_map; if (end > vm_map_max(map) || end < addr) return (ENOMEM); /* * Address of byte vector */ vec = uap->vec; pmap = vmspace_pmap(td->td_proc->p_vmspace); vm_map_lock_read(map); RestartScan: timestamp = map->timestamp; if (!vm_map_lookup_entry(map, addr, &entry)) { vm_map_unlock_read(map); return (ENOMEM); } /* * Do this on a map entry basis so that if the pages are not * in the current processes address space, we can easily look * up the pages elsewhere. */ lastvecindex = -1; for (current = entry; (current != &map->header) && (current->start < end); current = current->next) { /* * check for contiguity */ if (current->end < end && (entry->next == &map->header || current->next->start > current->end)) { vm_map_unlock_read(map); return (ENOMEM); } /* * ignore submaps (for now) or null objects */ if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) || current->object.vm_object == NULL) continue; /* * limit this scan to the current map entry and the * limits for the mincore call */ if (addr < current->start) addr = current->start; cend = current->end; if (cend > end) cend = end; /* * scan this entry one page at a time */ while (addr < cend) { /* * Check pmap first, it is likely faster, also * it can provide info as to whether we are the * one referencing or modifying the page. */ object = NULL; locked_pa = 0; retry: m = NULL; mincoreinfo = pmap_mincore(pmap, addr, &locked_pa); if (locked_pa != 0) { /* * The page is mapped by this process but not * both accessed and modified. It is also * managed. Acquire the object lock so that * other mappings might be examined. */ m = PHYS_TO_VM_PAGE(locked_pa); if (m->object != object) { if (object != NULL) VM_OBJECT_WUNLOCK(object); object = m->object; locked = VM_OBJECT_TRYWLOCK(object); vm_page_unlock(m); if (!locked) { VM_OBJECT_WLOCK(object); vm_page_lock(m); goto retry; } } else vm_page_unlock(m); KASSERT(m->valid == VM_PAGE_BITS_ALL, ("mincore: page %p is mapped but invalid", m)); } else if (mincoreinfo == 0) { /* * The page is not mapped by this process. If * the object implements managed pages, then * determine if the page is resident so that * the mappings might be examined. */ if (current->object.vm_object != object) { if (object != NULL) VM_OBJECT_WUNLOCK(object); object = current->object.vm_object; VM_OBJECT_WLOCK(object); } if (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP || object->type == OBJT_VNODE) { pindex = OFF_TO_IDX(current->offset + (addr - current->start)); m = vm_page_lookup(object, pindex); if (m == NULL && vm_page_is_cached(object, pindex)) mincoreinfo = MINCORE_INCORE; if (m != NULL && m->valid == 0) m = NULL; if (m != NULL) mincoreinfo = MINCORE_INCORE; } } if (m != NULL) { /* Examine other mappings to the page. */ if (m->dirty == 0 && pmap_is_modified(m)) vm_page_dirty(m); if (m->dirty != 0) mincoreinfo |= MINCORE_MODIFIED_OTHER; /* * The first test for PGA_REFERENCED is an * optimization. The second test is * required because a concurrent pmap * operation could clear the last reference * and set PGA_REFERENCED before the call to * pmap_is_referenced(). */ if ((m->aflags & PGA_REFERENCED) != 0 || pmap_is_referenced(m) || (m->aflags & PGA_REFERENCED) != 0) mincoreinfo |= MINCORE_REFERENCED_OTHER; } if (object != NULL) VM_OBJECT_WUNLOCK(object); /* * subyte may page fault. In case it needs to modify * the map, we release the lock. */ vm_map_unlock_read(map); /* * calculate index into user supplied byte vector */ vecindex = OFF_TO_IDX(addr - first_addr); /* * If we have skipped map entries, we need to make sure that * the byte vector is zeroed for those skipped entries. */ while ((lastvecindex + 1) < vecindex) { ++lastvecindex; error = subyte(vec + lastvecindex, 0); if (error) { error = EFAULT; goto done2; } } /* * Pass the page information to the user */ error = subyte(vec + vecindex, mincoreinfo); if (error) { error = EFAULT; goto done2; } /* * If the map has changed, due to the subyte, the previous * output may be invalid. */ vm_map_lock_read(map); if (timestamp != map->timestamp) goto RestartScan; lastvecindex = vecindex; addr += PAGE_SIZE; } } /* * subyte may page fault. In case it needs to modify * the map, we release the lock. */ vm_map_unlock_read(map); /* * Zero the last entries in the byte vector. */ vecindex = OFF_TO_IDX(end - first_addr); while ((lastvecindex + 1) < vecindex) { ++lastvecindex; error = subyte(vec + lastvecindex, 0); if (error) { error = EFAULT; goto done2; } } /* * If the map has changed, due to the subyte, the previous * output may be invalid. */ vm_map_lock_read(map); if (timestamp != map->timestamp) goto RestartScan; vm_map_unlock_read(map); done2: return (error); } #ifndef _SYS_SYSPROTO_H_ struct mlock_args { const void *addr; size_t len; }; #endif /* * MPSAFE */ int sys_mlock(td, uap) struct thread *td; struct mlock_args *uap; { return (vm_mlock(td->td_proc, td->td_ucred, uap->addr, uap->len)); } int vm_mlock(struct proc *proc, struct ucred *cred, const void *addr0, size_t len) { vm_offset_t addr, end, last, start; vm_size_t npages, size; vm_map_t map; unsigned long nsize; int error; error = priv_check_cred(cred, PRIV_VM_MLOCK, 0); if (error) return (error); addr = (vm_offset_t)addr0; size = len; last = addr + size; start = trunc_page(addr); end = round_page(last); if (last < addr || end < addr) return (EINVAL); npages = atop(end - start); if (npages > vm_page_max_wired) return (ENOMEM); map = &proc->p_vmspace->vm_map; PROC_LOCK(proc); nsize = ptoa(npages + pmap_wired_count(map->pmap)); - if (nsize > lim_cur(proc, RLIMIT_MEMLOCK)) { + if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) { PROC_UNLOCK(proc); return (ENOMEM); } PROC_UNLOCK(proc); if (npages + vm_cnt.v_wire_count > vm_page_max_wired) return (EAGAIN); #ifdef RACCT if (racct_enable) { PROC_LOCK(proc); error = racct_set(proc, RACCT_MEMLOCK, nsize); PROC_UNLOCK(proc); if (error != 0) return (ENOMEM); } #endif error = vm_map_wire(map, start, end, VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); #ifdef RACCT if (racct_enable && error != KERN_SUCCESS) { PROC_LOCK(proc); racct_set(proc, RACCT_MEMLOCK, ptoa(pmap_wired_count(map->pmap))); PROC_UNLOCK(proc); } #endif return (error == KERN_SUCCESS ? 0 : ENOMEM); } #ifndef _SYS_SYSPROTO_H_ struct mlockall_args { int how; }; #endif /* * MPSAFE */ int sys_mlockall(td, uap) struct thread *td; struct mlockall_args *uap; { vm_map_t map; int error; map = &td->td_proc->p_vmspace->vm_map; error = priv_check(td, PRIV_VM_MLOCK); if (error) return (error); if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0)) return (EINVAL); /* * If wiring all pages in the process would cause it to exceed * a hard resource limit, return ENOMEM. */ if (!old_mlock && uap->how & MCL_CURRENT) { PROC_LOCK(td->td_proc); - if (map->size > lim_cur(td->td_proc, RLIMIT_MEMLOCK)) { + if (map->size > lim_cur(td, RLIMIT_MEMLOCK)) { PROC_UNLOCK(td->td_proc); return (ENOMEM); } PROC_UNLOCK(td->td_proc); } #ifdef RACCT if (racct_enable) { PROC_LOCK(td->td_proc); error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size); PROC_UNLOCK(td->td_proc); if (error != 0) return (ENOMEM); } #endif if (uap->how & MCL_FUTURE) { vm_map_lock(map); vm_map_modflags(map, MAP_WIREFUTURE, 0); vm_map_unlock(map); error = 0; } if (uap->how & MCL_CURRENT) { /* * P1003.1-2001 mandates that all currently mapped pages * will be memory resident and locked (wired) upon return * from mlockall(). vm_map_wire() will wire pages, by * calling vm_fault_wire() for each page in the region. */ error = vm_map_wire(map, vm_map_min(map), vm_map_max(map), VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK); error = (error == KERN_SUCCESS ? 0 : EAGAIN); } #ifdef RACCT if (racct_enable && error != KERN_SUCCESS) { PROC_LOCK(td->td_proc); racct_set(td->td_proc, RACCT_MEMLOCK, ptoa(pmap_wired_count(map->pmap))); PROC_UNLOCK(td->td_proc); } #endif return (error); } #ifndef _SYS_SYSPROTO_H_ struct munlockall_args { register_t dummy; }; #endif /* * MPSAFE */ int sys_munlockall(td, uap) struct thread *td; struct munlockall_args *uap; { vm_map_t map; int error; map = &td->td_proc->p_vmspace->vm_map; error = priv_check(td, PRIV_VM_MUNLOCK); if (error) return (error); /* Clear the MAP_WIREFUTURE flag from this vm_map. */ vm_map_lock(map); vm_map_modflags(map, 0, MAP_WIREFUTURE); vm_map_unlock(map); /* Forcibly unwire all pages. */ error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map), VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK); #ifdef RACCT if (racct_enable && error == KERN_SUCCESS) { PROC_LOCK(td->td_proc); racct_set(td->td_proc, RACCT_MEMLOCK, 0); PROC_UNLOCK(td->td_proc); } #endif return (error); } #ifndef _SYS_SYSPROTO_H_ struct munlock_args { const void *addr; size_t len; }; #endif /* * MPSAFE */ int sys_munlock(td, uap) struct thread *td; struct munlock_args *uap; { vm_offset_t addr, end, last, start; vm_size_t size; #ifdef RACCT vm_map_t map; #endif int error; error = priv_check(td, PRIV_VM_MUNLOCK); if (error) return (error); addr = (vm_offset_t)uap->addr; size = uap->len; last = addr + size; start = trunc_page(addr); end = round_page(last); if (last < addr || end < addr) return (EINVAL); error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end, VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); #ifdef RACCT if (racct_enable && error == KERN_SUCCESS) { PROC_LOCK(td->td_proc); map = &td->td_proc->p_vmspace->vm_map; racct_set(td->td_proc, RACCT_MEMLOCK, ptoa(pmap_wired_count(map->pmap))); PROC_UNLOCK(td->td_proc); } #endif return (error == KERN_SUCCESS ? 0 : ENOMEM); } /* * vm_mmap_vnode() * * Helper function for vm_mmap. Perform sanity check specific for mmap * operations on vnodes. */ int vm_mmap_vnode(struct thread *td, vm_size_t objsize, vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp, struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp, boolean_t *writecounted) { struct vattr va; vm_object_t obj; vm_offset_t foff; struct ucred *cred; int error, flags, locktype; cred = td->td_ucred; if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED)) locktype = LK_EXCLUSIVE; else locktype = LK_SHARED; if ((error = vget(vp, locktype, td)) != 0) return (error); foff = *foffp; flags = *flagsp; obj = vp->v_object; if (vp->v_type == VREG) { /* * Get the proper underlying object */ if (obj == NULL) { error = EINVAL; goto done; } if (obj->type == OBJT_VNODE && obj->handle != vp) { vput(vp); vp = (struct vnode *)obj->handle; /* * Bypass filesystems obey the mpsafety of the * underlying fs. Tmpfs never bypasses. */ error = vget(vp, locktype, td); if (error != 0) return (error); } if (locktype == LK_EXCLUSIVE) { *writecounted = TRUE; vnode_pager_update_writecount(obj, 0, objsize); } } else { error = EINVAL; goto done; } if ((error = VOP_GETATTR(vp, &va, cred))) goto done; #ifdef MAC /* This relies on VM_PROT_* matching PROT_*. */ error = mac_vnode_check_mmap(cred, vp, (int)prot, flags); if (error != 0) goto done; #endif if ((flags & MAP_SHARED) != 0) { if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) { if (prot & VM_PROT_WRITE) { error = EPERM; goto done; } *maxprotp &= ~VM_PROT_WRITE; } } /* * If it is a regular file without any references * we do not need to sync it. * Adjust object size to be the size of actual file. */ objsize = round_page(va.va_size); if (va.va_nlink == 0) flags |= MAP_NOSYNC; if (obj->type == OBJT_VNODE) { obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff, cred); if (obj == NULL) { error = ENOMEM; goto done; } } else { KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP, ("wrong object type")); VM_OBJECT_WLOCK(obj); vm_object_reference_locked(obj); #if VM_NRESERVLEVEL > 0 vm_object_color(obj, 0); #endif VM_OBJECT_WUNLOCK(obj); } *objp = obj; *flagsp = flags; vfs_mark_atime(vp, cred); done: if (error != 0 && *writecounted) { *writecounted = FALSE; vnode_pager_update_writecount(obj, objsize, 0); } vput(vp); return (error); } /* * vm_mmap_cdev() * * MPSAFE * * Helper function for vm_mmap. Perform sanity check specific for mmap * operations on cdevs. */ int vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw, vm_ooffset_t *foff, vm_object_t *objp) { vm_object_t obj; int error, flags; flags = *flagsp; if (dsw->d_flags & D_MMAP_ANON) { *objp = NULL; *foff = 0; *maxprotp = VM_PROT_ALL; *flagsp |= MAP_ANON; return (0); } /* * cdevs do not provide private mappings of any kind. */ if ((*maxprotp & VM_PROT_WRITE) == 0 && (prot & VM_PROT_WRITE) != 0) return (EACCES); if (flags & (MAP_PRIVATE|MAP_COPY)) return (EINVAL); /* * Force device mappings to be shared. */ flags |= MAP_SHARED; #ifdef MAC_XXX error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot); if (error != 0) return (error); #endif /* * First, try d_mmap_single(). If that is not implemented * (returns ENODEV), fall back to using the device pager. * Note that d_mmap_single() must return a reference to the * object (it needs to bump the reference count of the object * it returns somehow). * * XXX assumes VM_PROT_* == PROT_* */ error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot); if (error != ENODEV) return (error); obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff, td->td_ucred); if (obj == NULL) return (EINVAL); *objp = obj; *flagsp = flags; return (0); } /* * vm_mmap() * * Internal version of mmap used by exec, sys5 shared memory, and * various device drivers. Handle is either a vnode pointer, a * character device, or NULL for MAP_ANON. */ int vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, vm_prot_t maxprot, int flags, objtype_t handle_type, void *handle, vm_ooffset_t foff) { vm_object_t object; struct thread *td = curthread; int error; boolean_t writecounted; if (size == 0) return (EINVAL); size = round_page(size); writecounted = FALSE; /* * Lookup/allocate object. */ switch (handle_type) { case OBJT_DEVICE: { struct cdevsw *dsw; struct cdev *cdev; int ref; cdev = handle; dsw = dev_refthread(cdev, &ref); if (dsw == NULL) return (ENXIO); error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev, dsw, &foff, &object); dev_relthread(cdev, ref); break; } case OBJT_VNODE: error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, handle, &foff, &object, &writecounted); break; case OBJT_DEFAULT: if (handle == NULL) { error = 0; break; } /* FALLTHROUGH */ default: error = EINVAL; break; } if (error) return (error); error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, foff, writecounted, td); if (error != 0 && object != NULL) { /* * If this mapping was accounted for in the vnode's * writecount, then undo that now. */ if (writecounted) vnode_pager_release_writecount(object, 0, size); vm_object_deallocate(object); } return (error); } /* * Internal version of mmap that maps a specific VM object into an * map. Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap. */ int vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff, boolean_t writecounted, struct thread *td) { boolean_t fitit; int docow, error, findspace, rv; if (map == &td->td_proc->p_vmspace->vm_map) { PROC_LOCK(td->td_proc); - if (map->size + size > lim_cur(td->td_proc, RLIMIT_VMEM)) { + if (map->size + size > lim_cur_proc(td->td_proc, RLIMIT_VMEM)) { PROC_UNLOCK(td->td_proc); return (ENOMEM); } if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) { PROC_UNLOCK(td->td_proc); return (ENOMEM); } if (!old_mlock && map->flags & MAP_WIREFUTURE) { if (ptoa(pmap_wired_count(map->pmap)) + size > - lim_cur(td->td_proc, RLIMIT_MEMLOCK)) { + lim_cur_proc(td->td_proc, RLIMIT_MEMLOCK)) { racct_set_force(td->td_proc, RACCT_VMEM, map->size); PROC_UNLOCK(td->td_proc); return (ENOMEM); } error = racct_set(td->td_proc, RACCT_MEMLOCK, ptoa(pmap_wired_count(map->pmap)) + size); if (error != 0) { racct_set_force(td->td_proc, RACCT_VMEM, map->size); PROC_UNLOCK(td->td_proc); return (error); } } PROC_UNLOCK(td->td_proc); } /* * We currently can only deal with page aligned file offsets. * The mmap() system call already enforces this by subtracting * the page offset from the file offset, but checking here * catches errors in device drivers (e.g. d_single_mmap() * callbacks) and other internal mapping requests (such as in * exec). */ if (foff & PAGE_MASK) return (EINVAL); if ((flags & MAP_FIXED) == 0) { fitit = TRUE; *addr = round_page(*addr); } else { if (*addr != trunc_page(*addr)) return (EINVAL); fitit = FALSE; } if (flags & MAP_ANON) { if (object != NULL || foff != 0) return (EINVAL); docow = 0; } else if (flags & MAP_PREFAULT_READ) docow = MAP_PREFAULT; else docow = MAP_PREFAULT_PARTIAL; if ((flags & (MAP_ANON|MAP_SHARED)) == 0) docow |= MAP_COPY_ON_WRITE; if (flags & MAP_NOSYNC) docow |= MAP_DISABLE_SYNCER; if (flags & MAP_NOCORE) docow |= MAP_DISABLE_COREDUMP; /* Shared memory is also shared with children. */ if (flags & MAP_SHARED) docow |= MAP_INHERIT_SHARE; if (writecounted) docow |= MAP_VN_WRITECOUNT; if (flags & MAP_STACK) { if (object != NULL) return (EINVAL); docow |= MAP_STACK_GROWS_DOWN; } if ((flags & MAP_EXCL) != 0) docow |= MAP_CHECK_EXCL; if (fitit) { if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER) findspace = VMFS_SUPER_SPACE; else if ((flags & MAP_ALIGNMENT_MASK) != 0) findspace = VMFS_ALIGNED_SPACE(flags >> MAP_ALIGNMENT_SHIFT); else findspace = VMFS_OPTIMAL_SPACE; rv = vm_map_find(map, object, foff, addr, size, #ifdef MAP_32BIT flags & MAP_32BIT ? MAP_32BIT_MAX_ADDR : #endif 0, findspace, prot, maxprot, docow); } else { rv = vm_map_fixed(map, object, foff, *addr, size, prot, maxprot, docow); } if (rv == KERN_SUCCESS) { /* * If the process has requested that all future mappings * be wired, then heed this. */ if (map->flags & MAP_WIREFUTURE) { vm_map_wire(map, *addr, *addr + size, VM_MAP_WIRE_USER | ((flags & MAP_STACK) ? VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES)); } } return (vm_mmap_to_errno(rv)); } /* * Translate a Mach VM return code to zero on success or the appropriate errno * on failure. */ int vm_mmap_to_errno(int rv) { switch (rv) { case KERN_SUCCESS: return (0); case KERN_INVALID_ADDRESS: case KERN_NO_SPACE: return (ENOMEM); case KERN_PROTECTION_FAILURE: return (EACCES); default: return (EINVAL); } } Index: head/sys/vm/vm_pageout.c =================================================================== --- head/sys/vm/vm_pageout.c (revision 284214) +++ head/sys/vm/vm_pageout.c (revision 284215) @@ -1,1919 +1,1919 @@ /*- * Copyright (c) 1991 Regents of the University of California. * All rights reserved. * Copyright (c) 1994 John S. Dyson * All rights reserved. * Copyright (c) 1994 David Greenman * All rights reserved. * Copyright (c) 2005 Yahoo! Technologies Norway AS * 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. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Berkeley and its contributors. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: @(#)vm_pageout.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. */ /* * The proverbial page-out daemon. */ #include __FBSDID("$FreeBSD$"); #include "opt_vm.h" #include "opt_kdtrace.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 /* * System initialization */ /* the kernel process "vm_pageout"*/ static void vm_pageout(void); static void vm_pageout_init(void); static int vm_pageout_clean(vm_page_t m); static int vm_pageout_cluster(vm_page_t m); static void vm_pageout_scan(struct vm_domain *vmd, int pass); static void vm_pageout_mightbe_oom(struct vm_domain *vmd, int pass); SYSINIT(pagedaemon_init, SI_SUB_KTHREAD_PAGE, SI_ORDER_FIRST, vm_pageout_init, NULL); struct proc *pageproc; static struct kproc_desc page_kp = { "pagedaemon", vm_pageout, &pageproc }; SYSINIT(pagedaemon, SI_SUB_KTHREAD_PAGE, SI_ORDER_SECOND, kproc_start, &page_kp); SDT_PROVIDER_DEFINE(vm); SDT_PROBE_DEFINE(vm, , , vm__lowmem_cache); SDT_PROBE_DEFINE(vm, , , vm__lowmem_scan); #if !defined(NO_SWAPPING) /* the kernel process "vm_daemon"*/ static void vm_daemon(void); static struct proc *vmproc; static struct kproc_desc vm_kp = { "vmdaemon", vm_daemon, &vmproc }; SYSINIT(vmdaemon, SI_SUB_KTHREAD_VM, SI_ORDER_FIRST, kproc_start, &vm_kp); #endif int vm_pages_needed; /* Event on which pageout daemon sleeps */ int vm_pageout_deficit; /* Estimated number of pages deficit */ int vm_pageout_pages_needed; /* flag saying that the pageout daemon needs pages */ int vm_pageout_wakeup_thresh; #if !defined(NO_SWAPPING) static int vm_pageout_req_swapout; /* XXX */ static int vm_daemon_needed; static struct mtx vm_daemon_mtx; /* Allow for use by vm_pageout before vm_daemon is initialized. */ MTX_SYSINIT(vm_daemon, &vm_daemon_mtx, "vm daemon", MTX_DEF); #endif static int vm_max_launder = 32; static int vm_pageout_update_period; static int defer_swap_pageouts; static int disable_swap_pageouts; static int lowmem_period = 10; static int lowmem_ticks; #if defined(NO_SWAPPING) static int vm_swap_enabled = 0; static int vm_swap_idle_enabled = 0; #else static int vm_swap_enabled = 1; static int vm_swap_idle_enabled = 0; #endif static int vm_panic_on_oom = 0; SYSCTL_INT(_vm, OID_AUTO, panic_on_oom, CTLFLAG_RWTUN, &vm_panic_on_oom, 0, "panic on out of memory instead of killing the largest process"); SYSCTL_INT(_vm, OID_AUTO, pageout_wakeup_thresh, CTLFLAG_RW, &vm_pageout_wakeup_thresh, 0, "free page threshold for waking up the pageout daemon"); SYSCTL_INT(_vm, OID_AUTO, max_launder, CTLFLAG_RW, &vm_max_launder, 0, "Limit dirty flushes in pageout"); SYSCTL_INT(_vm, OID_AUTO, pageout_update_period, CTLFLAG_RW, &vm_pageout_update_period, 0, "Maximum active LRU update period"); SYSCTL_INT(_vm, OID_AUTO, lowmem_period, CTLFLAG_RW, &lowmem_period, 0, "Low memory callback period"); #if defined(NO_SWAPPING) SYSCTL_INT(_vm, VM_SWAPPING_ENABLED, swap_enabled, CTLFLAG_RD, &vm_swap_enabled, 0, "Enable entire process swapout"); SYSCTL_INT(_vm, OID_AUTO, swap_idle_enabled, CTLFLAG_RD, &vm_swap_idle_enabled, 0, "Allow swapout on idle criteria"); #else SYSCTL_INT(_vm, VM_SWAPPING_ENABLED, swap_enabled, CTLFLAG_RW, &vm_swap_enabled, 0, "Enable entire process swapout"); SYSCTL_INT(_vm, OID_AUTO, swap_idle_enabled, CTLFLAG_RW, &vm_swap_idle_enabled, 0, "Allow swapout on idle criteria"); #endif SYSCTL_INT(_vm, OID_AUTO, defer_swapspace_pageouts, CTLFLAG_RW, &defer_swap_pageouts, 0, "Give preference to dirty pages in mem"); SYSCTL_INT(_vm, OID_AUTO, disable_swapspace_pageouts, CTLFLAG_RW, &disable_swap_pageouts, 0, "Disallow swapout of dirty pages"); static int pageout_lock_miss; SYSCTL_INT(_vm, OID_AUTO, pageout_lock_miss, CTLFLAG_RD, &pageout_lock_miss, 0, "vget() lock misses during pageout"); #define VM_PAGEOUT_PAGE_COUNT 16 int vm_pageout_page_count = VM_PAGEOUT_PAGE_COUNT; int vm_page_max_wired; /* XXX max # of wired pages system-wide */ SYSCTL_INT(_vm, OID_AUTO, max_wired, CTLFLAG_RW, &vm_page_max_wired, 0, "System-wide limit to wired page count"); static boolean_t vm_pageout_fallback_object_lock(vm_page_t, vm_page_t *); static boolean_t vm_pageout_launder(struct vm_pagequeue *pq, int, vm_paddr_t, vm_paddr_t); #if !defined(NO_SWAPPING) static void vm_pageout_map_deactivate_pages(vm_map_t, long); static void vm_pageout_object_deactivate_pages(pmap_t, vm_object_t, long); static void vm_req_vmdaemon(int req); #endif static boolean_t vm_pageout_page_lock(vm_page_t, vm_page_t *); /* * Initialize a dummy page for marking the caller's place in the specified * paging queue. In principle, this function only needs to set the flag * PG_MARKER. Nonetheless, it wirte busies and initializes the hold count * to one as safety precautions. */ static void vm_pageout_init_marker(vm_page_t marker, u_short queue) { bzero(marker, sizeof(*marker)); marker->flags = PG_MARKER; marker->busy_lock = VPB_SINGLE_EXCLUSIVER; marker->queue = queue; marker->hold_count = 1; } /* * vm_pageout_fallback_object_lock: * * Lock vm object currently associated with `m'. VM_OBJECT_TRYWLOCK is * known to have failed and page queue must be either PQ_ACTIVE or * PQ_INACTIVE. To avoid lock order violation, unlock the page queues * while locking the vm object. Use marker page to detect page queue * changes and maintain notion of next page on page queue. Return * TRUE if no changes were detected, FALSE otherwise. vm object is * locked on return. * * This function depends on both the lock portion of struct vm_object * and normal struct vm_page being type stable. */ static boolean_t vm_pageout_fallback_object_lock(vm_page_t m, vm_page_t *next) { struct vm_page marker; struct vm_pagequeue *pq; boolean_t unchanged; u_short queue; vm_object_t object; queue = m->queue; vm_pageout_init_marker(&marker, queue); pq = vm_page_pagequeue(m); object = m->object; TAILQ_INSERT_AFTER(&pq->pq_pl, m, &marker, plinks.q); vm_pagequeue_unlock(pq); vm_page_unlock(m); VM_OBJECT_WLOCK(object); vm_page_lock(m); vm_pagequeue_lock(pq); /* Page queue might have changed. */ *next = TAILQ_NEXT(&marker, plinks.q); unchanged = (m->queue == queue && m->object == object && &marker == TAILQ_NEXT(m, plinks.q)); TAILQ_REMOVE(&pq->pq_pl, &marker, plinks.q); return (unchanged); } /* * Lock the page while holding the page queue lock. Use marker page * to detect page queue changes and maintain notion of next page on * page queue. Return TRUE if no changes were detected, FALSE * otherwise. The page is locked on return. The page queue lock might * be dropped and reacquired. * * This function depends on normal struct vm_page being type stable. */ static boolean_t vm_pageout_page_lock(vm_page_t m, vm_page_t *next) { struct vm_page marker; struct vm_pagequeue *pq; boolean_t unchanged; u_short queue; vm_page_lock_assert(m, MA_NOTOWNED); if (vm_page_trylock(m)) return (TRUE); queue = m->queue; vm_pageout_init_marker(&marker, queue); pq = vm_page_pagequeue(m); TAILQ_INSERT_AFTER(&pq->pq_pl, m, &marker, plinks.q); vm_pagequeue_unlock(pq); vm_page_lock(m); vm_pagequeue_lock(pq); /* Page queue might have changed. */ *next = TAILQ_NEXT(&marker, plinks.q); unchanged = (m->queue == queue && &marker == TAILQ_NEXT(m, plinks.q)); TAILQ_REMOVE(&pq->pq_pl, &marker, plinks.q); return (unchanged); } /* * vm_pageout_clean: * * Clean the page and remove it from the laundry. * * We set the busy bit to cause potential page faults on this page to * block. Note the careful timing, however, the busy bit isn't set till * late and we cannot do anything that will mess with the page. */ static int vm_pageout_cluster(vm_page_t m) { vm_object_t object; vm_page_t mc[2*vm_pageout_page_count], pb, ps; int pageout_count; int ib, is, page_base; vm_pindex_t pindex = m->pindex; vm_page_lock_assert(m, MA_OWNED); object = m->object; VM_OBJECT_ASSERT_WLOCKED(object); /* * It doesn't cost us anything to pageout OBJT_DEFAULT or OBJT_SWAP * with the new swapper, but we could have serious problems paging * out other object types if there is insufficient memory. * * Unfortunately, checking free memory here is far too late, so the * check has been moved up a procedural level. */ /* * Can't clean the page if it's busy or held. */ vm_page_assert_unbusied(m); KASSERT(m->hold_count == 0, ("vm_pageout_clean: page %p is held", m)); vm_page_unlock(m); mc[vm_pageout_page_count] = pb = ps = m; pageout_count = 1; page_base = vm_pageout_page_count; ib = 1; is = 1; /* * Scan object for clusterable pages. * * We can cluster ONLY if: ->> the page is NOT * clean, wired, busy, held, or mapped into a * buffer, and one of the following: * 1) The page is inactive, or a seldom used * active page. * -or- * 2) we force the issue. * * During heavy mmap/modification loads the pageout * daemon can really fragment the underlying file * due to flushing pages out of order and not trying * align the clusters (which leave sporatic out-of-order * holes). To solve this problem we do the reverse scan * first and attempt to align our cluster, then do a * forward scan if room remains. */ more: while (ib && pageout_count < vm_pageout_page_count) { vm_page_t p; if (ib > pindex) { ib = 0; break; } if ((p = vm_page_prev(pb)) == NULL || vm_page_busied(p)) { ib = 0; break; } vm_page_lock(p); vm_page_test_dirty(p); if (p->dirty == 0 || p->queue != PQ_INACTIVE || p->hold_count != 0) { /* may be undergoing I/O */ vm_page_unlock(p); ib = 0; break; } vm_page_unlock(p); mc[--page_base] = pb = p; ++pageout_count; ++ib; /* * alignment boundry, stop here and switch directions. Do * not clear ib. */ if ((pindex - (ib - 1)) % vm_pageout_page_count == 0) break; } while (pageout_count < vm_pageout_page_count && pindex + is < object->size) { vm_page_t p; if ((p = vm_page_next(ps)) == NULL || vm_page_busied(p)) break; vm_page_lock(p); vm_page_test_dirty(p); if (p->dirty == 0 || p->queue != PQ_INACTIVE || p->hold_count != 0) { /* may be undergoing I/O */ vm_page_unlock(p); break; } vm_page_unlock(p); mc[page_base + pageout_count] = ps = p; ++pageout_count; ++is; } /* * If we exhausted our forward scan, continue with the reverse scan * when possible, even past a page boundry. This catches boundry * conditions. */ if (ib && pageout_count < vm_pageout_page_count) goto more; /* * we allow reads during pageouts... */ return (vm_pageout_flush(&mc[page_base], pageout_count, 0, 0, NULL, NULL)); } /* * vm_pageout_flush() - launder the given pages * * The given pages are laundered. Note that we setup for the start of * I/O ( i.e. busy the page ), mark it read-only, and bump the object * reference count all in here rather then in the parent. If we want * the parent to do more sophisticated things we may have to change * the ordering. * * Returned runlen is the count of pages between mreq and first * page after mreq with status VM_PAGER_AGAIN. * *eio is set to TRUE if pager returned VM_PAGER_ERROR or VM_PAGER_FAIL * for any page in runlen set. */ int vm_pageout_flush(vm_page_t *mc, int count, int flags, int mreq, int *prunlen, boolean_t *eio) { vm_object_t object = mc[0]->object; int pageout_status[count]; int numpagedout = 0; int i, runlen; VM_OBJECT_ASSERT_WLOCKED(object); /* * Initiate I/O. Bump the vm_page_t->busy counter and * mark the pages read-only. * * We do not have to fixup the clean/dirty bits here... we can * allow the pager to do it after the I/O completes. * * NOTE! mc[i]->dirty may be partial or fragmented due to an * edge case with file fragments. */ for (i = 0; i < count; i++) { KASSERT(mc[i]->valid == VM_PAGE_BITS_ALL, ("vm_pageout_flush: partially invalid page %p index %d/%d", mc[i], i, count)); vm_page_sbusy(mc[i]); pmap_remove_write(mc[i]); } vm_object_pip_add(object, count); vm_pager_put_pages(object, mc, count, flags, pageout_status); runlen = count - mreq; if (eio != NULL) *eio = FALSE; for (i = 0; i < count; i++) { vm_page_t mt = mc[i]; KASSERT(pageout_status[i] == VM_PAGER_PEND || !pmap_page_is_write_mapped(mt), ("vm_pageout_flush: page %p is not write protected", mt)); switch (pageout_status[i]) { case VM_PAGER_OK: case VM_PAGER_PEND: numpagedout++; break; case VM_PAGER_BAD: /* * Page outside of range of object. Right now we * essentially lose the changes by pretending it * worked. */ vm_page_undirty(mt); break; case VM_PAGER_ERROR: case VM_PAGER_FAIL: /* * If page couldn't be paged out, then reactivate the * page so it doesn't clog the inactive list. (We * will try paging out it again later). */ vm_page_lock(mt); vm_page_activate(mt); vm_page_unlock(mt); if (eio != NULL && i >= mreq && i - mreq < runlen) *eio = TRUE; break; case VM_PAGER_AGAIN: if (i >= mreq && i - mreq < runlen) runlen = i - mreq; break; } /* * If the operation is still going, leave the page busy to * block all other accesses. Also, leave the paging in * progress indicator set so that we don't attempt an object * collapse. */ if (pageout_status[i] != VM_PAGER_PEND) { vm_object_pip_wakeup(object); vm_page_sunbusy(mt); if (vm_page_count_severe()) { vm_page_lock(mt); vm_page_try_to_cache(mt); vm_page_unlock(mt); } } } if (prunlen != NULL) *prunlen = runlen; return (numpagedout); } static boolean_t vm_pageout_launder(struct vm_pagequeue *pq, int tries, vm_paddr_t low, vm_paddr_t high) { struct mount *mp; struct vnode *vp; vm_object_t object; vm_paddr_t pa; vm_page_t m, m_tmp, next; int lockmode; vm_pagequeue_lock(pq); TAILQ_FOREACH_SAFE(m, &pq->pq_pl, plinks.q, next) { if ((m->flags & PG_MARKER) != 0) continue; pa = VM_PAGE_TO_PHYS(m); if (pa < low || pa + PAGE_SIZE > high) continue; if (!vm_pageout_page_lock(m, &next) || m->hold_count != 0) { vm_page_unlock(m); continue; } object = m->object; if ((!VM_OBJECT_TRYWLOCK(object) && (!vm_pageout_fallback_object_lock(m, &next) || m->hold_count != 0)) || vm_page_busied(m)) { vm_page_unlock(m); VM_OBJECT_WUNLOCK(object); continue; } vm_page_test_dirty(m); if (m->dirty == 0 && object->ref_count != 0) pmap_remove_all(m); if (m->dirty != 0) { vm_page_unlock(m); if (tries == 0 || (object->flags & OBJ_DEAD) != 0) { VM_OBJECT_WUNLOCK(object); continue; } if (object->type == OBJT_VNODE) { vm_pagequeue_unlock(pq); vp = object->handle; vm_object_reference_locked(object); VM_OBJECT_WUNLOCK(object); (void)vn_start_write(vp, &mp, V_WAIT); lockmode = MNT_SHARED_WRITES(vp->v_mount) ? LK_SHARED : LK_EXCLUSIVE; vn_lock(vp, lockmode | LK_RETRY); VM_OBJECT_WLOCK(object); vm_object_page_clean(object, 0, 0, OBJPC_SYNC); VM_OBJECT_WUNLOCK(object); VOP_UNLOCK(vp, 0); vm_object_deallocate(object); vn_finished_write(mp); return (TRUE); } else if (object->type == OBJT_SWAP || object->type == OBJT_DEFAULT) { vm_pagequeue_unlock(pq); m_tmp = m; vm_pageout_flush(&m_tmp, 1, VM_PAGER_PUT_SYNC, 0, NULL, NULL); VM_OBJECT_WUNLOCK(object); return (TRUE); } } else { /* * Dequeue here to prevent lock recursion in * vm_page_cache(). */ vm_page_dequeue_locked(m); vm_page_cache(m); vm_page_unlock(m); } VM_OBJECT_WUNLOCK(object); } vm_pagequeue_unlock(pq); return (FALSE); } /* * Increase the number of cached pages. The specified value, "tries", * determines which categories of pages are cached: * * 0: All clean, inactive pages within the specified physical address range * are cached. Will not sleep. * 1: The vm_lowmem handlers are called. All inactive pages within * the specified physical address range are cached. May sleep. * 2: The vm_lowmem handlers are called. All inactive and active pages * within the specified physical address range are cached. May sleep. */ void vm_pageout_grow_cache(int tries, vm_paddr_t low, vm_paddr_t high) { int actl, actmax, inactl, inactmax, dom, initial_dom; static int start_dom = 0; if (tries > 0) { /* * Decrease registered cache sizes. The vm_lowmem handlers * may acquire locks and/or sleep, so they can only be invoked * when "tries" is greater than zero. */ SDT_PROBE0(vm, , , vm__lowmem_cache); EVENTHANDLER_INVOKE(vm_lowmem, 0); /* * We do this explicitly after the caches have been drained * above. */ uma_reclaim(); } /* * Make the next scan start on the next domain. */ initial_dom = atomic_fetchadd_int(&start_dom, 1) % vm_ndomains; inactl = 0; inactmax = vm_cnt.v_inactive_count; actl = 0; actmax = tries < 2 ? 0 : vm_cnt.v_active_count; dom = initial_dom; /* * Scan domains in round-robin order, first inactive queues, * then active. Since domain usually owns large physically * contiguous chunk of memory, it makes sense to completely * exhaust one domain before switching to next, while growing * the pool of contiguous physical pages. * * Do not even start launder a domain which cannot contain * the specified address range, as indicated by segments * constituting the domain. */ again: if (inactl < inactmax) { if (vm_phys_domain_intersects(vm_dom[dom].vmd_segs, low, high) && vm_pageout_launder(&vm_dom[dom].vmd_pagequeues[PQ_INACTIVE], tries, low, high)) { inactl++; goto again; } if (++dom == vm_ndomains) dom = 0; if (dom != initial_dom) goto again; } if (actl < actmax) { if (vm_phys_domain_intersects(vm_dom[dom].vmd_segs, low, high) && vm_pageout_launder(&vm_dom[dom].vmd_pagequeues[PQ_ACTIVE], tries, low, high)) { actl++; goto again; } if (++dom == vm_ndomains) dom = 0; if (dom != initial_dom) goto again; } } #if !defined(NO_SWAPPING) /* * vm_pageout_object_deactivate_pages * * Deactivate enough pages to satisfy the inactive target * requirements. * * The object and map must be locked. */ static void vm_pageout_object_deactivate_pages(pmap_t pmap, vm_object_t first_object, long desired) { vm_object_t backing_object, object; vm_page_t p; int act_delta, remove_mode; VM_OBJECT_ASSERT_LOCKED(first_object); if ((first_object->flags & OBJ_FICTITIOUS) != 0) return; for (object = first_object;; object = backing_object) { if (pmap_resident_count(pmap) <= desired) goto unlock_return; VM_OBJECT_ASSERT_LOCKED(object); if ((object->flags & OBJ_UNMANAGED) != 0 || object->paging_in_progress != 0) goto unlock_return; remove_mode = 0; if (object->shadow_count > 1) remove_mode = 1; /* * Scan the object's entire memory queue. */ TAILQ_FOREACH(p, &object->memq, listq) { if (pmap_resident_count(pmap) <= desired) goto unlock_return; if (vm_page_busied(p)) continue; PCPU_INC(cnt.v_pdpages); vm_page_lock(p); if (p->wire_count != 0 || p->hold_count != 0 || !pmap_page_exists_quick(pmap, p)) { vm_page_unlock(p); continue; } act_delta = pmap_ts_referenced(p); if ((p->aflags & PGA_REFERENCED) != 0) { if (act_delta == 0) act_delta = 1; vm_page_aflag_clear(p, PGA_REFERENCED); } if (p->queue != PQ_ACTIVE && act_delta != 0) { vm_page_activate(p); p->act_count += act_delta; } else if (p->queue == PQ_ACTIVE) { if (act_delta == 0) { p->act_count -= min(p->act_count, ACT_DECLINE); if (!remove_mode && p->act_count == 0) { pmap_remove_all(p); vm_page_deactivate(p); } else vm_page_requeue(p); } else { vm_page_activate(p); if (p->act_count < ACT_MAX - ACT_ADVANCE) p->act_count += ACT_ADVANCE; vm_page_requeue(p); } } else if (p->queue == PQ_INACTIVE) pmap_remove_all(p); vm_page_unlock(p); } if ((backing_object = object->backing_object) == NULL) goto unlock_return; VM_OBJECT_RLOCK(backing_object); if (object != first_object) VM_OBJECT_RUNLOCK(object); } unlock_return: if (object != first_object) VM_OBJECT_RUNLOCK(object); } /* * deactivate some number of pages in a map, try to do it fairly, but * that is really hard to do. */ static void vm_pageout_map_deactivate_pages(map, desired) vm_map_t map; long desired; { vm_map_entry_t tmpe; vm_object_t obj, bigobj; int nothingwired; if (!vm_map_trylock(map)) return; bigobj = NULL; nothingwired = TRUE; /* * first, search out the biggest object, and try to free pages from * that. */ tmpe = map->header.next; while (tmpe != &map->header) { if ((tmpe->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { obj = tmpe->object.vm_object; if (obj != NULL && VM_OBJECT_TRYRLOCK(obj)) { if (obj->shadow_count <= 1 && (bigobj == NULL || bigobj->resident_page_count < obj->resident_page_count)) { if (bigobj != NULL) VM_OBJECT_RUNLOCK(bigobj); bigobj = obj; } else VM_OBJECT_RUNLOCK(obj); } } if (tmpe->wired_count > 0) nothingwired = FALSE; tmpe = tmpe->next; } if (bigobj != NULL) { vm_pageout_object_deactivate_pages(map->pmap, bigobj, desired); VM_OBJECT_RUNLOCK(bigobj); } /* * Next, hunt around for other pages to deactivate. We actually * do this search sort of wrong -- .text first is not the best idea. */ tmpe = map->header.next; while (tmpe != &map->header) { if (pmap_resident_count(vm_map_pmap(map)) <= desired) break; if ((tmpe->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { obj = tmpe->object.vm_object; if (obj != NULL) { VM_OBJECT_RLOCK(obj); vm_pageout_object_deactivate_pages(map->pmap, obj, desired); VM_OBJECT_RUNLOCK(obj); } } tmpe = tmpe->next; } /* * Remove all mappings if a process is swapped out, this will free page * table pages. */ if (desired == 0 && nothingwired) { pmap_remove(vm_map_pmap(map), vm_map_min(map), vm_map_max(map)); } vm_map_unlock(map); } #endif /* !defined(NO_SWAPPING) */ /* * Attempt to acquire all of the necessary locks to launder a page and * then call through the clustering layer to PUTPAGES. Wait a short * time for a vnode lock. * * Requires the page and object lock on entry, releases both before return. * Returns 0 on success and an errno otherwise. */ static int vm_pageout_clean(vm_page_t m) { struct vnode *vp; struct mount *mp; vm_object_t object; vm_pindex_t pindex; int error, lockmode; vm_page_assert_locked(m); object = m->object; VM_OBJECT_ASSERT_WLOCKED(object); error = 0; vp = NULL; mp = NULL; /* * The object is already known NOT to be dead. It * is possible for the vget() to block the whole * pageout daemon, but the new low-memory handling * code should prevent it. * * We can't wait forever for the vnode lock, we might * deadlock due to a vn_read() getting stuck in * vm_wait while holding this vnode. We skip the * vnode if we can't get it in a reasonable amount * of time. */ if (object->type == OBJT_VNODE) { vm_page_unlock(m); vp = object->handle; if (vp->v_type == VREG && vn_start_write(vp, &mp, V_NOWAIT) != 0) { mp = NULL; error = EDEADLK; goto unlock_all; } KASSERT(mp != NULL, ("vp %p with NULL v_mount", vp)); vm_object_reference_locked(object); pindex = m->pindex; VM_OBJECT_WUNLOCK(object); lockmode = MNT_SHARED_WRITES(vp->v_mount) ? LK_SHARED : LK_EXCLUSIVE; if (vget(vp, lockmode | LK_TIMELOCK, curthread)) { vp = NULL; error = EDEADLK; goto unlock_mp; } VM_OBJECT_WLOCK(object); vm_page_lock(m); /* * While the object and page were unlocked, the page * may have been: * (1) moved to a different queue, * (2) reallocated to a different object, * (3) reallocated to a different offset, or * (4) cleaned. */ if (m->queue != PQ_INACTIVE || m->object != object || m->pindex != pindex || m->dirty == 0) { vm_page_unlock(m); error = ENXIO; goto unlock_all; } /* * The page may have been busied or held while the object * and page locks were released. */ if (vm_page_busied(m) || m->hold_count != 0) { vm_page_unlock(m); error = EBUSY; goto unlock_all; } } /* * If a page is dirty, then it is either being washed * (but not yet cleaned) or it is still in the * laundry. If it is still in the laundry, then we * start the cleaning operation. */ if (vm_pageout_cluster(m) == 0) error = EIO; unlock_all: VM_OBJECT_WUNLOCK(object); unlock_mp: vm_page_lock_assert(m, MA_NOTOWNED); if (mp != NULL) { if (vp != NULL) vput(vp); vm_object_deallocate(object); vn_finished_write(mp); } return (error); } /* * vm_pageout_scan does the dirty work for the pageout daemon. * * pass 0 - Update active LRU/deactivate pages * pass 1 - Move inactive to cache or free * pass 2 - Launder dirty pages */ static void vm_pageout_scan(struct vm_domain *vmd, int pass) { vm_page_t m, next; struct vm_pagequeue *pq; vm_object_t object; int act_delta, addl_page_shortage, deficit, maxscan, page_shortage; int vnodes_skipped = 0; int maxlaunder; boolean_t queues_locked; /* * If we need to reclaim memory ask kernel caches to return * some. We rate limit to avoid thrashing. */ if (vmd == &vm_dom[0] && pass > 0 && (ticks - lowmem_ticks) / hz >= lowmem_period) { /* * Decrease registered cache sizes. */ SDT_PROBE0(vm, , , vm__lowmem_scan); EVENTHANDLER_INVOKE(vm_lowmem, 0); /* * We do this explicitly after the caches have been * drained above. */ uma_reclaim(); lowmem_ticks = ticks; } /* * The addl_page_shortage is the number of temporarily * stuck pages in the inactive queue. In other words, the * number of pages from the inactive count that should be * discounted in setting the target for the active queue scan. */ addl_page_shortage = 0; /* * Calculate the number of pages we want to either free or move * to the cache. */ if (pass > 0) { deficit = atomic_readandclear_int(&vm_pageout_deficit); page_shortage = vm_paging_target() + deficit; } else page_shortage = deficit = 0; /* * maxlaunder limits the number of dirty pages we flush per scan. * For most systems a smaller value (16 or 32) is more robust under * extreme memory and disk pressure because any unnecessary writes * to disk can result in extreme performance degredation. However, * systems with excessive dirty pages (especially when MAP_NOSYNC is * used) will die horribly with limited laundering. If the pageout * daemon cannot clean enough pages in the first pass, we let it go * all out in succeeding passes. */ if ((maxlaunder = vm_max_launder) <= 1) maxlaunder = 1; if (pass > 1) maxlaunder = 10000; /* * Start scanning the inactive queue for pages we can move to the * cache or free. The scan will stop when the target is reached or * we have scanned the entire inactive queue. Note that m->act_count * is not used to form decisions for the inactive queue, only for the * active queue. */ pq = &vmd->vmd_pagequeues[PQ_INACTIVE]; maxscan = pq->pq_cnt; vm_pagequeue_lock(pq); queues_locked = TRUE; for (m = TAILQ_FIRST(&pq->pq_pl); m != NULL && maxscan-- > 0 && page_shortage > 0; m = next) { vm_pagequeue_assert_locked(pq); KASSERT(queues_locked, ("unlocked queues")); KASSERT(m->queue == PQ_INACTIVE, ("Inactive queue %p", m)); PCPU_INC(cnt.v_pdpages); next = TAILQ_NEXT(m, plinks.q); /* * skip marker pages */ if (m->flags & PG_MARKER) continue; KASSERT((m->flags & PG_FICTITIOUS) == 0, ("Fictitious page %p cannot be in inactive queue", m)); KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("Unmanaged page %p cannot be in inactive queue", m)); /* * The page or object lock acquisitions fail if the * page was removed from the queue or moved to a * different position within the queue. In either * case, addl_page_shortage should not be incremented. */ if (!vm_pageout_page_lock(m, &next)) { vm_page_unlock(m); continue; } object = m->object; if (!VM_OBJECT_TRYWLOCK(object) && !vm_pageout_fallback_object_lock(m, &next)) { vm_page_unlock(m); VM_OBJECT_WUNLOCK(object); continue; } /* * Don't mess with busy pages, keep them at at the * front of the queue, most likely they are being * paged out. Increment addl_page_shortage for busy * pages, because they may leave the inactive queue * shortly after page scan is finished. */ if (vm_page_busied(m)) { vm_page_unlock(m); VM_OBJECT_WUNLOCK(object); addl_page_shortage++; continue; } /* * We unlock the inactive page queue, invalidating the * 'next' pointer. Use our marker to remember our * place. */ TAILQ_INSERT_AFTER(&pq->pq_pl, m, &vmd->vmd_marker, plinks.q); vm_pagequeue_unlock(pq); queues_locked = FALSE; /* * We bump the activation count if the page has been * referenced while in the inactive queue. This makes * it less likely that the page will be added back to the * inactive queue prematurely again. Here we check the * page tables (or emulated bits, if any), given the upper * level VM system not knowing anything about existing * references. */ if ((m->aflags & PGA_REFERENCED) != 0) { vm_page_aflag_clear(m, PGA_REFERENCED); act_delta = 1; } else act_delta = 0; if (object->ref_count != 0) { act_delta += pmap_ts_referenced(m); } else { KASSERT(!pmap_page_is_mapped(m), ("vm_pageout_scan: page %p is mapped", m)); } /* * If the upper level VM system knows about any page * references, we reactivate the page or requeue it. */ if (act_delta != 0) { if (object->ref_count != 0) { vm_page_activate(m); m->act_count += act_delta + ACT_ADVANCE; } else { vm_pagequeue_lock(pq); queues_locked = TRUE; vm_page_requeue_locked(m); } VM_OBJECT_WUNLOCK(object); vm_page_unlock(m); goto relock_queues; } if (m->hold_count != 0) { vm_page_unlock(m); VM_OBJECT_WUNLOCK(object); /* * Held pages are essentially stuck in the * queue. So, they ought to be discounted * from the inactive count. See the * calculation of the page_shortage for the * loop over the active queue below. */ addl_page_shortage++; goto relock_queues; } /* * If the page appears to be clean at the machine-independent * layer, then remove all of its mappings from the pmap in * anticipation of placing it onto the cache queue. If, * however, any of the page's mappings allow write access, * then the page may still be modified until the last of those * mappings are removed. */ vm_page_test_dirty(m); if (m->dirty == 0 && object->ref_count != 0) pmap_remove_all(m); if (m->valid == 0) { /* * Invalid pages can be easily freed */ vm_page_free(m); PCPU_INC(cnt.v_dfree); --page_shortage; } else if (m->dirty == 0) { /* * Clean pages can be placed onto the cache queue. * This effectively frees them. */ vm_page_cache(m); --page_shortage; } else if ((m->flags & PG_WINATCFLS) == 0 && pass < 2) { /* * Dirty pages need to be paged out, but flushing * a page is extremely expensive versus freeing * a clean page. Rather then artificially limiting * the number of pages we can flush, we instead give * dirty pages extra priority on the inactive queue * by forcing them to be cycled through the queue * twice before being flushed, after which the * (now clean) page will cycle through once more * before being freed. This significantly extends * the thrash point for a heavily loaded machine. */ m->flags |= PG_WINATCFLS; vm_pagequeue_lock(pq); queues_locked = TRUE; vm_page_requeue_locked(m); } else if (maxlaunder > 0) { /* * We always want to try to flush some dirty pages if * we encounter them, to keep the system stable. * Normally this number is small, but under extreme * pressure where there are insufficient clean pages * on the inactive queue, we may have to go all out. */ int swap_pageouts_ok; int error; if ((object->type != OBJT_SWAP) && (object->type != OBJT_DEFAULT)) { swap_pageouts_ok = 1; } else { swap_pageouts_ok = !(defer_swap_pageouts || disable_swap_pageouts); swap_pageouts_ok |= (!disable_swap_pageouts && defer_swap_pageouts && vm_page_count_min()); } /* * We don't bother paging objects that are "dead". * Those objects are in a "rundown" state. */ if (!swap_pageouts_ok || (object->flags & OBJ_DEAD)) { vm_pagequeue_lock(pq); vm_page_unlock(m); VM_OBJECT_WUNLOCK(object); queues_locked = TRUE; vm_page_requeue_locked(m); goto relock_queues; } error = vm_pageout_clean(m); /* * Decrement page_shortage on success to account for * the (future) cleaned page. Otherwise we could wind * up laundering or cleaning too many pages. */ if (error == 0) { page_shortage--; maxlaunder--; } else if (error == EDEADLK) { pageout_lock_miss++; vnodes_skipped++; } else if (error == EBUSY) { addl_page_shortage++; } vm_page_lock_assert(m, MA_NOTOWNED); goto relock_queues; } vm_page_unlock(m); VM_OBJECT_WUNLOCK(object); relock_queues: if (!queues_locked) { vm_pagequeue_lock(pq); queues_locked = TRUE; } next = TAILQ_NEXT(&vmd->vmd_marker, plinks.q); TAILQ_REMOVE(&pq->pq_pl, &vmd->vmd_marker, plinks.q); } vm_pagequeue_unlock(pq); #if !defined(NO_SWAPPING) /* * Wakeup the swapout daemon if we didn't cache or free the targeted * number of pages. */ if (vm_swap_enabled && page_shortage > 0) vm_req_vmdaemon(VM_SWAP_NORMAL); #endif /* * Wakeup the sync daemon if we skipped a vnode in a writeable object * and we didn't cache or free enough pages. */ if (vnodes_skipped > 0 && page_shortage > vm_cnt.v_free_target - vm_cnt.v_free_min) (void)speedup_syncer(); /* * Compute the number of pages we want to try to move from the * active queue to the inactive queue. */ page_shortage = vm_cnt.v_inactive_target - vm_cnt.v_inactive_count + vm_paging_target() + deficit + addl_page_shortage; pq = &vmd->vmd_pagequeues[PQ_ACTIVE]; vm_pagequeue_lock(pq); maxscan = pq->pq_cnt; /* * If we're just idle polling attempt to visit every * active page within 'update_period' seconds. */ if (pass == 0 && vm_pageout_update_period != 0) { maxscan /= vm_pageout_update_period; page_shortage = maxscan; } /* * Scan the active queue for things we can deactivate. We nominally * track the per-page activity counter and use it to locate * deactivation candidates. */ m = TAILQ_FIRST(&pq->pq_pl); while (m != NULL && maxscan-- > 0 && page_shortage > 0) { KASSERT(m->queue == PQ_ACTIVE, ("vm_pageout_scan: page %p isn't active", m)); next = TAILQ_NEXT(m, plinks.q); if ((m->flags & PG_MARKER) != 0) { m = next; continue; } KASSERT((m->flags & PG_FICTITIOUS) == 0, ("Fictitious page %p cannot be in active queue", m)); KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("Unmanaged page %p cannot be in active queue", m)); if (!vm_pageout_page_lock(m, &next)) { vm_page_unlock(m); m = next; continue; } /* * The count for pagedaemon pages is done after checking the * page for eligibility... */ PCPU_INC(cnt.v_pdpages); /* * Check to see "how much" the page has been used. */ if ((m->aflags & PGA_REFERENCED) != 0) { vm_page_aflag_clear(m, PGA_REFERENCED); act_delta = 1; } else act_delta = 0; /* * Unlocked object ref count check. Two races are possible. * 1) The ref was transitioning to zero and we saw non-zero, * the pmap bits will be checked unnecessarily. * 2) The ref was transitioning to one and we saw zero. * The page lock prevents a new reference to this page so * we need not check the reference bits. */ if (m->object->ref_count != 0) act_delta += pmap_ts_referenced(m); /* * Advance or decay the act_count based on recent usage. */ if (act_delta != 0) { m->act_count += ACT_ADVANCE + act_delta; if (m->act_count > ACT_MAX) m->act_count = ACT_MAX; } else m->act_count -= min(m->act_count, ACT_DECLINE); /* * Move this page to the tail of the active or inactive * queue depending on usage. */ if (m->act_count == 0) { /* Dequeue to avoid later lock recursion. */ vm_page_dequeue_locked(m); vm_page_deactivate(m); page_shortage--; } else vm_page_requeue_locked(m); vm_page_unlock(m); m = next; } vm_pagequeue_unlock(pq); #if !defined(NO_SWAPPING) /* * Idle process swapout -- run once per second. */ if (vm_swap_idle_enabled) { static long lsec; if (time_second != lsec) { vm_req_vmdaemon(VM_SWAP_IDLE); lsec = time_second; } } #endif /* * If we are critically low on one of RAM or swap and low on * the other, kill the largest process. However, we avoid * doing this on the first pass in order to give ourselves a * chance to flush out dirty vnode-backed pages and to allow * active pages to be moved to the inactive queue and reclaimed. */ vm_pageout_mightbe_oom(vmd, pass); } static int vm_pageout_oom_vote; /* * The pagedaemon threads randlomly select one to perform the * OOM. Trying to kill processes before all pagedaemons * failed to reach free target is premature. */ static void vm_pageout_mightbe_oom(struct vm_domain *vmd, int pass) { int old_vote; if (pass <= 1 || !((swap_pager_avail < 64 && vm_page_count_min()) || (swap_pager_full && vm_paging_target() > 0))) { if (vmd->vmd_oom) { vmd->vmd_oom = FALSE; atomic_subtract_int(&vm_pageout_oom_vote, 1); } return; } if (vmd->vmd_oom) return; vmd->vmd_oom = TRUE; old_vote = atomic_fetchadd_int(&vm_pageout_oom_vote, 1); if (old_vote != vm_ndomains - 1) return; /* * The current pagedaemon thread is the last in the quorum to * start OOM. Initiate the selection and signaling of the * victim. */ vm_pageout_oom(VM_OOM_MEM); /* * After one round of OOM terror, recall our vote. On the * next pass, current pagedaemon would vote again if the low * memory condition is still there, due to vmd_oom being * false. */ vmd->vmd_oom = FALSE; atomic_subtract_int(&vm_pageout_oom_vote, 1); } void vm_pageout_oom(int shortage) { struct proc *p, *bigproc; vm_offset_t size, bigsize; struct thread *td; struct vmspace *vm; /* * We keep the process bigproc locked once we find it to keep anyone * from messing with it; however, there is a possibility of * deadlock if process B is bigproc and one of it's child processes * attempts to propagate a signal to B while we are waiting for A's * lock while walking this list. To avoid this, we don't block on * the process lock but just skip a process if it is already locked. */ bigproc = NULL; bigsize = 0; sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { int breakout; PROC_LOCK(p); /* * If this is a system, protected or killed process, skip it. */ if (p->p_state != PRS_NORMAL || (p->p_flag & (P_INEXEC | P_PROTECTED | P_SYSTEM | P_WEXIT)) != 0 || p->p_pid == 1 || P_KILLED(p) || (p->p_pid < 48 && swap_pager_avail != 0)) { PROC_UNLOCK(p); continue; } /* * If the process is in a non-running type state, * don't touch it. Check all the threads individually. */ breakout = 0; FOREACH_THREAD_IN_PROC(p, td) { thread_lock(td); if (!TD_ON_RUNQ(td) && !TD_IS_RUNNING(td) && !TD_IS_SLEEPING(td) && !TD_IS_SUSPENDED(td)) { thread_unlock(td); breakout = 1; break; } thread_unlock(td); } if (breakout) { PROC_UNLOCK(p); continue; } /* * get the process size */ vm = vmspace_acquire_ref(p); if (vm == NULL) { PROC_UNLOCK(p); continue; } _PHOLD(p); if (!vm_map_trylock_read(&vm->vm_map)) { _PRELE(p); PROC_UNLOCK(p); vmspace_free(vm); continue; } PROC_UNLOCK(p); size = vmspace_swap_count(vm); vm_map_unlock_read(&vm->vm_map); if (shortage == VM_OOM_MEM) size += vmspace_resident_count(vm); vmspace_free(vm); /* * if the this process is bigger than the biggest one * remember it. */ if (size > bigsize) { if (bigproc != NULL) PRELE(bigproc); bigproc = p; bigsize = size; } else { PRELE(p); } } sx_sunlock(&allproc_lock); if (bigproc != NULL) { if (vm_panic_on_oom != 0) panic("out of swap space"); PROC_LOCK(bigproc); killproc(bigproc, "out of swap space"); sched_nice(bigproc, PRIO_MIN); _PRELE(bigproc); PROC_UNLOCK(bigproc); wakeup(&vm_cnt.v_free_count); } } static void vm_pageout_worker(void *arg) { struct vm_domain *domain; int domidx; domidx = (uintptr_t)arg; domain = &vm_dom[domidx]; /* * XXXKIB It could be useful to bind pageout daemon threads to * the cores belonging to the domain, from which vm_page_array * is allocated. */ KASSERT(domain->vmd_segs != 0, ("domain without segments")); vm_pageout_init_marker(&domain->vmd_marker, PQ_INACTIVE); /* * The pageout daemon worker is never done, so loop forever. */ while (TRUE) { /* * If we have enough free memory, wakeup waiters. Do * not clear vm_pages_needed until we reach our target, * otherwise we may be woken up over and over again and * waste a lot of cpu. */ mtx_lock(&vm_page_queue_free_mtx); if (vm_pages_needed && !vm_page_count_min()) { if (!vm_paging_needed()) vm_pages_needed = 0; wakeup(&vm_cnt.v_free_count); } if (vm_pages_needed) { /* * Still not done, take a second pass without waiting * (unlimited dirty cleaning), otherwise sleep a bit * and try again. */ if (domain->vmd_pass > 1) msleep(&vm_pages_needed, &vm_page_queue_free_mtx, PVM, "psleep", hz / 2); } else { /* * Good enough, sleep until required to refresh * stats. */ domain->vmd_pass = 0; msleep(&vm_pages_needed, &vm_page_queue_free_mtx, PVM, "psleep", hz); } if (vm_pages_needed) { vm_cnt.v_pdwakeups++; domain->vmd_pass++; } mtx_unlock(&vm_page_queue_free_mtx); vm_pageout_scan(domain, domain->vmd_pass); } } /* * vm_pageout_init initialises basic pageout daemon settings. */ static void vm_pageout_init(void) { /* * Initialize some paging parameters. */ vm_cnt.v_interrupt_free_min = 2; if (vm_cnt.v_page_count < 2000) vm_pageout_page_count = 8; /* * v_free_reserved needs to include enough for the largest * swap pager structures plus enough for any pv_entry structs * when paging. */ if (vm_cnt.v_page_count > 1024) vm_cnt.v_free_min = 4 + (vm_cnt.v_page_count - 1024) / 200; else vm_cnt.v_free_min = 4; vm_cnt.v_pageout_free_min = (2*MAXBSIZE)/PAGE_SIZE + vm_cnt.v_interrupt_free_min; vm_cnt.v_free_reserved = vm_pageout_page_count + vm_cnt.v_pageout_free_min + (vm_cnt.v_page_count / 768); vm_cnt.v_free_severe = vm_cnt.v_free_min / 2; vm_cnt.v_free_target = 4 * vm_cnt.v_free_min + vm_cnt.v_free_reserved; vm_cnt.v_free_min += vm_cnt.v_free_reserved; vm_cnt.v_free_severe += vm_cnt.v_free_reserved; vm_cnt.v_inactive_target = (3 * vm_cnt.v_free_target) / 2; if (vm_cnt.v_inactive_target > vm_cnt.v_free_count / 3) vm_cnt.v_inactive_target = vm_cnt.v_free_count / 3; /* * Set the default wakeup threshold to be 10% above the minimum * page limit. This keeps the steady state out of shortfall. */ vm_pageout_wakeup_thresh = (vm_cnt.v_free_min / 10) * 11; /* * Set interval in seconds for active scan. We want to visit each * page at least once every ten minutes. This is to prevent worst * case paging behaviors with stale active LRU. */ if (vm_pageout_update_period == 0) vm_pageout_update_period = 600; /* XXX does not really belong here */ if (vm_page_max_wired == 0) vm_page_max_wired = vm_cnt.v_free_count / 3; } /* * vm_pageout is the high level pageout daemon. */ static void vm_pageout(void) { int error; #if MAXMEMDOM > 1 int i; #endif swap_pager_swap_init(); #if MAXMEMDOM > 1 for (i = 1; i < vm_ndomains; i++) { error = kthread_add(vm_pageout_worker, (void *)(uintptr_t)i, curproc, NULL, 0, 0, "dom%d", i); if (error != 0) { panic("starting pageout for domain %d, error %d\n", i, error); } } #endif error = kthread_add(uma_reclaim_worker, NULL, curproc, NULL, 0, 0, "uma"); if (error != 0) panic("starting uma_reclaim helper, error %d\n", error); vm_pageout_worker((void *)(uintptr_t)0); } /* * Unless the free page queue lock is held by the caller, this function * should be regarded as advisory. Specifically, the caller should * not msleep() on &vm_cnt.v_free_count following this function unless * the free page queue lock is held until the msleep() is performed. */ void pagedaemon_wakeup(void) { if (!vm_pages_needed && curthread->td_proc != pageproc) { vm_pages_needed = 1; wakeup(&vm_pages_needed); } } #if !defined(NO_SWAPPING) static void vm_req_vmdaemon(int req) { static int lastrun = 0; mtx_lock(&vm_daemon_mtx); vm_pageout_req_swapout |= req; if ((ticks > (lastrun + hz)) || (ticks < lastrun)) { wakeup(&vm_daemon_needed); lastrun = ticks; } mtx_unlock(&vm_daemon_mtx); } static void vm_daemon(void) { struct rlimit rsslim; struct proc *p; struct thread *td; struct vmspace *vm; int breakout, swapout_flags, tryagain, attempts; #ifdef RACCT uint64_t rsize, ravailable; #endif while (TRUE) { mtx_lock(&vm_daemon_mtx); msleep(&vm_daemon_needed, &vm_daemon_mtx, PPAUSE, "psleep", #ifdef RACCT racct_enable ? hz : 0 #else 0 #endif ); swapout_flags = vm_pageout_req_swapout; vm_pageout_req_swapout = 0; mtx_unlock(&vm_daemon_mtx); if (swapout_flags) swapout_procs(swapout_flags); /* * scan the processes for exceeding their rlimits or if * process is swapped out -- deactivate pages */ tryagain = 0; attempts = 0; again: attempts++; sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { vm_pindex_t limit, size; /* * if this is a system process or if we have already * looked at this process, skip it. */ PROC_LOCK(p); if (p->p_state != PRS_NORMAL || p->p_flag & (P_INEXEC | P_SYSTEM | P_WEXIT)) { PROC_UNLOCK(p); continue; } /* * if the process is in a non-running type state, * don't touch it. */ breakout = 0; FOREACH_THREAD_IN_PROC(p, td) { thread_lock(td); if (!TD_ON_RUNQ(td) && !TD_IS_RUNNING(td) && !TD_IS_SLEEPING(td) && !TD_IS_SUSPENDED(td)) { thread_unlock(td); breakout = 1; break; } thread_unlock(td); } if (breakout) { PROC_UNLOCK(p); continue; } /* * get a limit */ - lim_rlimit(p, RLIMIT_RSS, &rsslim); + lim_rlimit_proc(p, RLIMIT_RSS, &rsslim); limit = OFF_TO_IDX( qmin(rsslim.rlim_cur, rsslim.rlim_max)); /* * let processes that are swapped out really be * swapped out set the limit to nothing (will force a * swap-out.) */ if ((p->p_flag & P_INMEM) == 0) limit = 0; /* XXX */ vm = vmspace_acquire_ref(p); PROC_UNLOCK(p); if (vm == NULL) continue; size = vmspace_resident_count(vm); if (size >= limit) { vm_pageout_map_deactivate_pages( &vm->vm_map, limit); } #ifdef RACCT if (racct_enable) { rsize = IDX_TO_OFF(size); PROC_LOCK(p); racct_set(p, RACCT_RSS, rsize); ravailable = racct_get_available(p, RACCT_RSS); PROC_UNLOCK(p); if (rsize > ravailable) { /* * Don't be overly aggressive; this * might be an innocent process, * and the limit could've been exceeded * by some memory hog. Don't try * to deactivate more than 1/4th * of process' resident set size. */ if (attempts <= 8) { if (ravailable < rsize - (rsize / 4)) { ravailable = rsize - (rsize / 4); } } vm_pageout_map_deactivate_pages( &vm->vm_map, OFF_TO_IDX(ravailable)); /* Update RSS usage after paging out. */ size = vmspace_resident_count(vm); rsize = IDX_TO_OFF(size); PROC_LOCK(p); racct_set(p, RACCT_RSS, rsize); PROC_UNLOCK(p); if (rsize > ravailable) tryagain = 1; } } #endif vmspace_free(vm); } sx_sunlock(&allproc_lock); if (tryagain != 0 && attempts <= 10) goto again; } } #endif /* !defined(NO_SWAPPING) */ Index: head/sys/vm/vm_unix.c =================================================================== --- head/sys/vm/vm_unix.c (revision 284214) +++ head/sys/vm/vm_unix.c (revision 284215) @@ -1,257 +1,255 @@ /*- * Copyright (c) 1988 University of Utah. * 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 Systems Programming Group of the University of Utah Computer * Science Department. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: Utah $Hdr: vm_unix.c 1.1 89/11/07$ * * @(#)vm_unix.c 8.1 (Berkeley) 6/11/93 */ #include "opt_compat.h" /* * Traditional sbrk/grow interface to VM */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #ifndef _SYS_SYSPROTO_H_ struct obreak_args { char *nsize; }; #endif /* * MPSAFE */ /* ARGSUSED */ int sys_obreak(td, uap) struct thread *td; struct obreak_args *uap; { struct vmspace *vm = td->td_proc->p_vmspace; vm_map_t map = &vm->vm_map; vm_offset_t new, old, base; rlim_t datalim, lmemlim, vmemlim; int prot, rv; int error = 0; boolean_t do_map_wirefuture; - PROC_LOCK(td->td_proc); - datalim = lim_cur(td->td_proc, RLIMIT_DATA); - lmemlim = lim_cur(td->td_proc, RLIMIT_MEMLOCK); - vmemlim = lim_cur(td->td_proc, RLIMIT_VMEM); - PROC_UNLOCK(td->td_proc); + datalim = lim_cur(td, RLIMIT_DATA); + lmemlim = lim_cur(td, RLIMIT_MEMLOCK); + vmemlim = lim_cur(td, RLIMIT_VMEM); do_map_wirefuture = FALSE; new = round_page((vm_offset_t)uap->nsize); vm_map_lock(map); base = round_page((vm_offset_t) vm->vm_daddr); old = base + ctob(vm->vm_dsize); if (new > base) { /* * Check the resource limit, but allow a process to reduce * its usage, even if it remains over the limit. */ if (new - base > datalim && new > old) { error = ENOMEM; goto done; } if (new > vm_map_max(map)) { error = ENOMEM; goto done; } } else if (new < base) { /* * This is simply an invalid value. If someone wants to * do fancy address space manipulations, mmap and munmap * can do most of what the user would want. */ error = EINVAL; goto done; } if (new > old) { if (!old_mlock && map->flags & MAP_WIREFUTURE) { if (ptoa(pmap_wired_count(map->pmap)) + (new - old) > lmemlim) { error = ENOMEM; goto done; } } if (map->size + (new - old) > vmemlim) { error = ENOMEM; goto done; } #ifdef RACCT if (racct_enable) { PROC_LOCK(td->td_proc); error = racct_set(td->td_proc, RACCT_DATA, new - base); if (error != 0) { PROC_UNLOCK(td->td_proc); error = ENOMEM; goto done; } error = racct_set(td->td_proc, RACCT_VMEM, map->size + (new - old)); if (error != 0) { racct_set_force(td->td_proc, RACCT_DATA, old - base); PROC_UNLOCK(td->td_proc); error = ENOMEM; goto done; } if (!old_mlock && map->flags & MAP_WIREFUTURE) { error = racct_set(td->td_proc, RACCT_MEMLOCK, ptoa(pmap_wired_count(map->pmap)) + (new - old)); if (error != 0) { racct_set_force(td->td_proc, RACCT_DATA, old - base); racct_set_force(td->td_proc, RACCT_VMEM, map->size); PROC_UNLOCK(td->td_proc); error = ENOMEM; goto done; } } PROC_UNLOCK(td->td_proc); } #endif prot = VM_PROT_RW; #ifdef COMPAT_FREEBSD32 #if defined(__amd64__) if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32)) prot |= VM_PROT_EXECUTE; #endif #endif rv = vm_map_insert(map, NULL, 0, old, new, prot, VM_PROT_ALL, 0); if (rv != KERN_SUCCESS) { #ifdef RACCT if (racct_enable) { PROC_LOCK(td->td_proc); racct_set_force(td->td_proc, RACCT_DATA, old - base); racct_set_force(td->td_proc, RACCT_VMEM, map->size); if (!old_mlock && map->flags & MAP_WIREFUTURE) { racct_set_force(td->td_proc, RACCT_MEMLOCK, ptoa(pmap_wired_count(map->pmap))); } PROC_UNLOCK(td->td_proc); } #endif error = ENOMEM; goto done; } vm->vm_dsize += btoc(new - old); /* * Handle the MAP_WIREFUTURE case for legacy applications, * by marking the newly mapped range of pages as wired. * We are not required to perform a corresponding * vm_map_unwire() before vm_map_delete() below, as * it will forcibly unwire the pages in the range. * * XXX If the pages cannot be wired, no error is returned. */ if ((map->flags & MAP_WIREFUTURE) == MAP_WIREFUTURE) { if (bootverbose) printf("obreak: MAP_WIREFUTURE set\n"); do_map_wirefuture = TRUE; } } else if (new < old) { rv = vm_map_delete(map, new, old); if (rv != KERN_SUCCESS) { error = ENOMEM; goto done; } vm->vm_dsize -= btoc(old - new); #ifdef RACCT if (racct_enable) { PROC_LOCK(td->td_proc); racct_set_force(td->td_proc, RACCT_DATA, new - base); racct_set_force(td->td_proc, RACCT_VMEM, map->size); if (!old_mlock && map->flags & MAP_WIREFUTURE) { racct_set_force(td->td_proc, RACCT_MEMLOCK, ptoa(pmap_wired_count(map->pmap))); } PROC_UNLOCK(td->td_proc); } #endif } done: vm_map_unlock(map); if (do_map_wirefuture) (void) vm_map_wire(map, old, new, VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES); return (error); } #ifndef _SYS_SYSPROTO_H_ struct ovadvise_args { int anom; }; #endif /* * MPSAFE */ /* ARGSUSED */ int sys_ovadvise(td, uap) struct thread *td; struct ovadvise_args *uap; { /* START_GIANT_OPTIONAL */ /* END_GIANT_OPTIONAL */ return (EINVAL); }