2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1989, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
42 #include "opt_capsicum.h"
44 #include "opt_ktrace.h"
46 #include <sys/param.h>
47 #include <sys/systm.h>
49 #include <sys/capsicum.h>
51 #include <sys/fcntl.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
66 #include <sys/protosw.h>
67 #include <sys/racct.h>
68 #include <sys/resourcevar.h>
70 #include <sys/signalvar.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysctl.h>
77 #include <sys/sysproto.h>
78 #include <sys/unistd.h>
80 #include <sys/vnode.h>
82 #include <sys/ktrace.h>
87 #include <security/audit/audit.h>
94 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
95 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
96 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
97 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
98 "file desc to leader structures");
99 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
100 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
102 MALLOC_DECLARE(M_FADVISE);
104 static __read_mostly uma_zone_t file_zone;
105 static __read_mostly uma_zone_t filedesc0_zone;
106 __read_mostly uma_zone_t pwd_zone;
109 static int closefp(struct filedesc *fdp, int fd, struct file *fp,
110 struct thread *td, int holdleaders);
111 static int fd_first_free(struct filedesc *fdp, int low, int size);
112 static void fdgrowtable(struct filedesc *fdp, int nfd);
113 static void fdgrowtable_exp(struct filedesc *fdp, int nfd);
114 static void fdunused(struct filedesc *fdp, int fd);
115 static void fdused(struct filedesc *fdp, int fd);
116 static int getmaxfd(struct thread *td);
117 static u_long *filecaps_copy_prep(const struct filecaps *src);
118 static void filecaps_copy_finish(const struct filecaps *src,
119 struct filecaps *dst, u_long *ioctls);
120 static u_long *filecaps_free_prep(struct filecaps *fcaps);
121 static void filecaps_free_finish(u_long *ioctls);
123 static struct pwd *pwd_alloc(void);
128 * - An array of open file descriptors (fd_ofiles)
129 * - An array of file flags (fd_ofileflags)
130 * - A bitmap recording which descriptors are in use (fd_map)
132 * A process starts out with NDFILE descriptors. The value of NDFILE has
133 * been selected based the historical limit of 20 open files, and an
134 * assumption that the majority of processes, especially short-lived
135 * processes like shells, will never need more.
137 * If this initial allocation is exhausted, a larger descriptor table and
138 * map are allocated dynamically, and the pointers in the process's struct
139 * filedesc are updated to point to those. This is repeated every time
140 * the process runs out of file descriptors (provided it hasn't hit its
143 * Since threads may hold references to individual descriptor table
144 * entries, the tables are never freed. Instead, they are placed on a
145 * linked list and freed only when the struct filedesc is released.
148 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
149 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
150 #define NDSLOT(x) ((x) / NDENTRIES)
151 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
152 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
155 * SLIST entry used to keep track of ofiles which must be reclaimed when
159 struct fdescenttbl *ft_table;
160 SLIST_ENTRY(freetable) ft_next;
164 * Initial allocation: a filedesc structure + the head of SLIST used to
165 * keep track of old ofiles + enough space for NDFILE descriptors.
168 struct fdescenttbl0 {
170 struct filedescent fdt_ofiles[NDFILE];
174 struct filedesc fd_fd;
175 SLIST_HEAD(, freetable) fd_free;
176 struct fdescenttbl0 fd_dfiles;
177 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
181 * Descriptor management.
183 static int __exclusive_cache_line openfiles; /* actual number of open files */
184 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
185 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
188 * If low >= size, just return low. Otherwise find the first zero bit in the
189 * given bitmap, starting at low and not exceeding size - 1. Return size if
193 fd_first_free(struct filedesc *fdp, int low, int size)
195 NDSLOTTYPE *map = fdp->fd_map;
203 if (low % NDENTRIES) {
204 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
205 if ((mask &= ~map[off]) != 0UL)
206 return (off * NDENTRIES + ffsl(mask) - 1);
209 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
210 if (map[off] != ~0UL)
211 return (off * NDENTRIES + ffsl(~map[off]) - 1);
216 * Find the last used fd.
218 * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
219 * Otherwise use fdlastfile.
222 fdlastfile_single(struct filedesc *fdp)
224 NDSLOTTYPE *map = fdp->fd_map;
227 off = NDSLOT(fdp->fd_nfiles - 1);
228 for (minoff = NDSLOT(0); off >= minoff; --off)
230 return (off * NDENTRIES + flsl(map[off]) - 1);
235 fdlastfile(struct filedesc *fdp)
238 FILEDESC_LOCK_ASSERT(fdp);
239 return (fdlastfile_single(fdp));
243 fdisused(struct filedesc *fdp, int fd)
246 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
247 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
249 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
253 * Mark a file descriptor as used.
256 fdused_init(struct filedesc *fdp, int fd)
259 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
261 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
265 fdused(struct filedesc *fdp, int fd)
268 FILEDESC_XLOCK_ASSERT(fdp);
270 fdused_init(fdp, fd);
271 if (fd == fdp->fd_freefile)
276 * Mark a file descriptor as unused.
279 fdunused(struct filedesc *fdp, int fd)
282 FILEDESC_XLOCK_ASSERT(fdp);
284 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
285 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
286 ("fd=%d is still in use", fd));
288 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
289 if (fd < fdp->fd_freefile)
290 fdp->fd_freefile = fd;
294 * Free a file descriptor.
296 * Avoid some work if fdp is about to be destroyed.
299 fdefree_last(struct filedescent *fde)
302 filecaps_free(&fde->fde_caps);
306 fdfree(struct filedesc *fdp, int fd)
308 struct filedescent *fde;
310 fde = &fdp->fd_ofiles[fd];
312 seqc_write_begin(&fde->fde_seqc);
314 fde->fde_file = NULL;
316 seqc_write_end(&fde->fde_seqc);
323 * System calls on descriptors.
325 #ifndef _SYS_SYSPROTO_H_
326 struct getdtablesize_args {
332 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
338 td->td_retval[0] = getmaxfd(td);
340 PROC_LOCK(td->td_proc);
341 lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
342 PROC_UNLOCK(td->td_proc);
343 if (lim < td->td_retval[0])
344 td->td_retval[0] = lim;
350 * Duplicate a file descriptor to a particular value.
352 * Note: keep in mind that a potential race condition exists when closing
353 * descriptors from a shared descriptor table (via rfork).
355 #ifndef _SYS_SYSPROTO_H_
363 sys_dup2(struct thread *td, struct dup2_args *uap)
366 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
370 * Duplicate a file descriptor.
372 #ifndef _SYS_SYSPROTO_H_
379 sys_dup(struct thread *td, struct dup_args *uap)
382 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
386 * The file control system call.
388 #ifndef _SYS_SYSPROTO_H_
397 sys_fcntl(struct thread *td, struct fcntl_args *uap)
400 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
404 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
418 * Convert old flock structure to new.
420 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
421 fl.l_start = ofl.l_start;
422 fl.l_len = ofl.l_len;
423 fl.l_pid = ofl.l_pid;
424 fl.l_type = ofl.l_type;
425 fl.l_whence = ofl.l_whence;
439 arg1 = (intptr_t)&fl;
445 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
446 arg1 = (intptr_t)&fl;
454 error = kern_fcntl(td, fd, newcmd, arg1);
457 if (cmd == F_OGETLK) {
458 ofl.l_start = fl.l_start;
459 ofl.l_len = fl.l_len;
460 ofl.l_pid = fl.l_pid;
461 ofl.l_type = fl.l_type;
462 ofl.l_whence = fl.l_whence;
463 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
464 } else if (cmd == F_GETLK) {
465 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
471 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
473 struct filedesc *fdp;
475 struct file *fp, *fp2;
476 struct filedescent *fde;
480 int error, flg, seals, tmp;
494 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
497 case F_DUPFD_CLOEXEC:
499 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
504 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
507 case F_DUP2FD_CLOEXEC:
509 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
515 fde = fdeget_locked(fdp, fd);
518 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
521 FILEDESC_SUNLOCK(fdp);
527 fde = fdeget_locked(fdp, fd);
529 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
530 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
533 FILEDESC_XUNLOCK(fdp);
537 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
540 td->td_retval[0] = OFLAGS(fp->f_flag);
545 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
549 tmp = flg = fp->f_flag;
551 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
552 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
553 tmp = fp->f_flag & FNONBLOCK;
554 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
559 tmp = fp->f_flag & FASYNC;
560 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
565 atomic_clear_int(&fp->f_flag, FNONBLOCK);
567 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
572 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
575 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
577 td->td_retval[0] = tmp;
582 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
586 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
591 error = priv_check(td, PRIV_NFS_LOCKD);
599 /* FALLTHROUGH F_SETLK */
603 flp = (struct flock *)arg;
604 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
609 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
612 if (fp->f_type != DTYPE_VNODE) {
618 if (flp->l_whence == SEEK_CUR) {
619 foffset = foffset_get(fp);
622 foffset > OFF_MAX - flp->l_start)) {
627 flp->l_start += foffset;
631 switch (flp->l_type) {
633 if ((fp->f_flag & FREAD) == 0) {
637 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
638 PROC_LOCK(p->p_leader);
639 p->p_leader->p_flag |= P_ADVLOCK;
640 PROC_UNLOCK(p->p_leader);
642 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
646 if ((fp->f_flag & FWRITE) == 0) {
650 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
651 PROC_LOCK(p->p_leader);
652 p->p_leader->p_flag |= P_ADVLOCK;
653 PROC_UNLOCK(p->p_leader);
655 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
659 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
663 if (flg != F_REMOTE) {
667 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
668 F_UNLCKSYS, flp, flg);
674 if (error != 0 || flp->l_type == F_UNLCK ||
675 flp->l_type == F_UNLCKSYS) {
681 * Check for a race with close.
683 * The vnode is now advisory locked (or unlocked, but this case
684 * is not really important) as the caller requested.
685 * We had to drop the filedesc lock, so we need to recheck if
686 * the descriptor is still valid, because if it was closed
687 * in the meantime we need to remove advisory lock from the
688 * vnode - close on any descriptor leading to an advisory
689 * locked vnode, removes that lock.
690 * We will return 0 on purpose in that case, as the result of
691 * successful advisory lock might have been externally visible
692 * already. This is fine - effectively we pretend to the caller
693 * that the closing thread was a bit slower and that the
694 * advisory lock succeeded before the close.
696 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
702 flp->l_whence = SEEK_SET;
705 flp->l_type = F_UNLCK;
706 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
707 F_UNLCK, flp, F_POSIX);
714 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
717 if (fp->f_type != DTYPE_VNODE) {
722 flp = (struct flock *)arg;
723 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
724 flp->l_type != F_UNLCK) {
729 if (flp->l_whence == SEEK_CUR) {
730 foffset = foffset_get(fp);
731 if ((flp->l_start > 0 &&
732 foffset > OFF_MAX - flp->l_start) ||
734 foffset < OFF_MIN - flp->l_start)) {
739 flp->l_start += foffset;
742 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
748 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
751 error = fo_add_seals(fp, arg);
756 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
759 if (fo_get_seals(fp, &seals) == 0)
760 td->td_retval[0] = seals;
767 arg = arg ? 128 * 1024: 0;
770 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
773 if (fp->f_type != DTYPE_VNODE) {
779 if (vp->v_type != VREG) {
786 * Exclusive lock synchronizes against f_seqcount reads and
787 * writes in sequential_heuristic().
789 error = vn_lock(vp, LK_EXCLUSIVE);
795 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
796 arg = MIN(arg, INT_MAX - bsize + 1);
797 fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
798 (arg + bsize - 1) / bsize);
799 atomic_set_int(&fp->f_flag, FRDAHEAD);
801 atomic_clear_int(&fp->f_flag, FRDAHEAD);
809 * Check if the vnode is part of a union stack (either the
810 * "union" flag from mount(2) or unionfs).
812 * Prior to introduction of this op libc's readdir would call
813 * fstatfs(2), in effect unnecessarily copying kilobytes of
814 * data just to check fs name and a mount flag.
816 * Fixing the code to handle everything in the kernel instead
817 * is a non-trivial endeavor and has low priority, thus this
818 * horrible kludge facilitates the current behavior in a much
819 * cheaper manner until someone(tm) sorts this out.
821 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
824 if (fp->f_type != DTYPE_VNODE) {
831 * Since we don't prevent dooming the vnode even non-null mp
832 * found can become immediately stale. This is tolerable since
833 * mount points are type-stable (providing safe memory access)
834 * and any vfs op on this vnode going forward will return an
835 * error (meaning return value in this case is meaningless).
837 mp = atomic_load_ptr(&vp->v_mount);
838 if (__predict_false(mp == NULL)) {
843 td->td_retval[0] = 0;
844 if (mp->mnt_kern_flag & MNTK_UNIONFS ||
845 mp->mnt_flag & MNT_UNION)
846 td->td_retval[0] = 1;
858 getmaxfd(struct thread *td)
861 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
865 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
868 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
870 struct filedesc *fdp;
871 struct filedescent *oldfde, *newfde;
874 u_long *oioctls, *nioctls;
881 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
882 MPASS(mode < FDDUP_LASTMODE);
885 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
888 * Verify we have a valid descriptor to dup from and possibly to
889 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
890 * return EINVAL when the new descriptor is out of bounds.
895 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
896 maxfd = getmaxfd(td);
898 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
902 if (fget_locked(fdp, old) == NULL)
904 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
905 td->td_retval[0] = new;
906 if (flags & FDDUP_FLAG_CLOEXEC)
907 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
912 oldfde = &fdp->fd_ofiles[old];
913 if (!fhold(oldfde->fde_file))
917 * If the caller specified a file descriptor, make sure the file
918 * table is large enough to hold it, and grab it. Otherwise, just
919 * allocate a new descriptor the usual way.
924 if ((error = fdalloc(td, new, &new)) != 0) {
925 fdrop(oldfde->fde_file, td);
929 case FDDUP_MUSTREPLACE:
930 /* Target file descriptor must exist. */
931 if (fget_locked(fdp, new) == NULL) {
932 fdrop(oldfde->fde_file, td);
937 if (new >= fdp->fd_nfiles) {
939 * The resource limits are here instead of e.g.
940 * fdalloc(), because the file descriptor table may be
941 * shared between processes, so we can't really use
942 * racct_add()/racct_sub(). Instead of counting the
943 * number of actually allocated descriptors, just put
944 * the limit on the size of the file descriptor table.
947 if (RACCT_ENABLED()) {
948 error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
951 fdrop(oldfde->fde_file, td);
956 fdgrowtable_exp(fdp, new + 1);
958 if (!fdisused(fdp, new))
962 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
965 KASSERT(old != new, ("new fd is same as old"));
967 newfde = &fdp->fd_ofiles[new];
968 delfp = newfde->fde_file;
970 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
973 * Duplicate the source descriptor.
976 seqc_write_begin(&newfde->fde_seqc);
978 oioctls = filecaps_free_prep(&newfde->fde_caps);
979 memcpy(newfde, oldfde, fde_change_size);
980 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
982 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
983 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
985 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
987 seqc_write_end(&newfde->fde_seqc);
989 td->td_retval[0] = new;
994 (void) closefp(fdp, new, delfp, td, 1);
995 FILEDESC_UNLOCK_ASSERT(fdp);
998 FILEDESC_XUNLOCK(fdp);
1001 filecaps_free_finish(oioctls);
1006 sigiofree(struct sigio *sigio)
1008 crfree(sigio->sio_ucred);
1009 free(sigio, M_SIGIO);
1012 static struct sigio *
1013 funsetown_locked(struct sigio *sigio)
1018 SIGIO_ASSERT_LOCKED();
1022 *(sigio->sio_myref) = NULL;
1023 if (sigio->sio_pgid < 0) {
1024 pg = sigio->sio_pgrp;
1026 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
1027 sigio, sio_pgsigio);
1030 p = sigio->sio_proc;
1032 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
1033 sigio, sio_pgsigio);
1040 * If sigio is on the list associated with a process or process group,
1041 * disable signalling from the device, remove sigio from the list and
1045 funsetown(struct sigio **sigiop)
1047 struct sigio *sigio;
1049 /* Racy check, consumers must provide synchronization. */
1050 if (*sigiop == NULL)
1054 sigio = funsetown_locked(*sigiop);
1061 * Free a list of sigio structures. The caller must ensure that new sigio
1062 * structures cannot be added after this point. For process groups this is
1063 * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1067 funsetownlst(struct sigiolst *sigiolst)
1071 struct sigio *sigio, *tmp;
1074 sigio = SLIST_FIRST(sigiolst);
1082 sigio = SLIST_FIRST(sigiolst);
1083 if (sigio == NULL) {
1089 * Every entry of the list should belong to a single proc or pgrp.
1091 if (sigio->sio_pgid < 0) {
1092 pg = sigio->sio_pgrp;
1093 sx_assert(&proctree_lock, SX_XLOCKED);
1095 } else /* if (sigio->sio_pgid > 0) */ {
1096 p = sigio->sio_proc;
1098 KASSERT((p->p_flag & P_WEXIT) != 0,
1099 ("%s: process %p is not exiting", __func__, p));
1102 SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1103 *sigio->sio_myref = NULL;
1105 KASSERT(sigio->sio_pgid < 0,
1106 ("Proc sigio in pgrp sigio list"));
1107 KASSERT(sigio->sio_pgrp == pg,
1108 ("Bogus pgrp in sigio list"));
1109 } else /* if (p != NULL) */ {
1110 KASSERT(sigio->sio_pgid > 0,
1111 ("Pgrp sigio in proc sigio list"));
1112 KASSERT(sigio->sio_proc == p,
1113 ("Bogus proc in sigio list"));
1123 SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1128 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1130 * After permission checking, add a sigio structure to the sigio list for
1131 * the process or process group.
1134 fsetown(pid_t pgid, struct sigio **sigiop)
1138 struct sigio *osigio, *sigio;
1148 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1149 sigio->sio_pgid = pgid;
1150 sigio->sio_ucred = crhold(curthread->td_ucred);
1151 sigio->sio_myref = sigiop;
1153 sx_slock(&proctree_lock);
1155 osigio = funsetown_locked(*sigiop);
1164 * Policy - Don't allow a process to FSETOWN a process
1165 * in another session.
1167 * Remove this test to allow maximum flexibility or
1168 * restrict FSETOWN to the current process or process
1169 * group for maximum safety.
1171 if (proc->p_session != curthread->td_proc->p_session) {
1177 sigio->sio_proc = proc;
1178 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1180 } else /* if (pgid < 0) */ {
1181 pgrp = pgfind(-pgid);
1188 * Policy - Don't allow a process to FSETOWN a process
1189 * in another session.
1191 * Remove this test to allow maximum flexibility or
1192 * restrict FSETOWN to the current process or process
1193 * group for maximum safety.
1195 if (pgrp->pg_session != curthread->td_proc->p_session) {
1201 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1202 sigio->sio_pgrp = pgrp;
1205 sx_sunlock(&proctree_lock);
1214 sx_sunlock(&proctree_lock);
1222 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1225 fgetown(struct sigio **sigiop)
1230 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1236 * Function drops the filedesc lock on return.
1239 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1244 FILEDESC_XLOCK_ASSERT(fdp);
1247 if (td->td_proc->p_fdtol != NULL) {
1249 * Ask fdfree() to sleep to ensure that all relevant
1250 * process leaders can be traversed in closef().
1252 fdp->fd_holdleaderscount++;
1259 * We now hold the fp reference that used to be owned by the
1260 * descriptor array. We have to unlock the FILEDESC *AFTER*
1261 * knote_fdclose to prevent a race of the fd getting opened, a knote
1262 * added, and deleteing a knote for the new fd.
1264 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1265 knote_fdclose(td, fd);
1268 * We need to notify mqueue if the object is of type mqueue.
1270 if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1271 mq_fdclose(td, fd, fp);
1272 FILEDESC_XUNLOCK(fdp);
1274 error = closef(fp, td);
1276 FILEDESC_XLOCK(fdp);
1277 fdp->fd_holdleaderscount--;
1278 if (fdp->fd_holdleaderscount == 0 &&
1279 fdp->fd_holdleaderswakeup != 0) {
1280 fdp->fd_holdleaderswakeup = 0;
1281 wakeup(&fdp->fd_holdleaderscount);
1283 FILEDESC_XUNLOCK(fdp);
1289 * Close a file descriptor.
1291 #ifndef _SYS_SYSPROTO_H_
1298 sys_close(struct thread *td, struct close_args *uap)
1301 return (kern_close(td, uap->fd));
1305 kern_close(struct thread *td, int fd)
1307 struct filedesc *fdp;
1310 fdp = td->td_proc->p_fd;
1312 AUDIT_SYSCLOSE(td, fd);
1314 FILEDESC_XLOCK(fdp);
1315 if ((fp = fget_locked(fdp, fd)) == NULL) {
1316 FILEDESC_XUNLOCK(fdp);
1321 /* closefp() drops the FILEDESC lock for us. */
1322 return (closefp(fdp, fd, fp, td, 1));
1326 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1328 struct filedesc *fdp;
1329 int fd, ret, lastfile;
1332 fdp = td->td_proc->p_fd;
1333 FILEDESC_SLOCK(fdp);
1336 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1337 * open should not be a usage error. From a close_range() perspective,
1338 * close_range(3, ~0U, 0) in the same scenario should also likely not
1339 * be a usage error as all fd above 3 are in-fact already closed.
1341 if (highfd < lowfd) {
1347 * If lastfile == -1, we're dealing with either a fresh file
1348 * table or one in which every fd has been closed. Just return
1349 * successful; there's nothing left to do.
1351 lastfile = fdlastfile(fdp);
1354 /* Clamped to [lowfd, lastfile] */
1355 highfd = MIN(highfd, lastfile);
1356 for (fd = lowfd; fd <= highfd; fd++) {
1357 if (fdp->fd_ofiles[fd].fde_file != NULL) {
1358 FILEDESC_SUNLOCK(fdp);
1359 (void)kern_close(td, fd);
1360 FILEDESC_SLOCK(fdp);
1364 FILEDESC_SUNLOCK(fdp);
1368 #ifndef _SYS_SYSPROTO_H_
1369 struct close_range_args {
1376 sys_close_range(struct thread *td, struct close_range_args *uap)
1379 /* No flags currently defined */
1380 if (uap->flags != 0)
1382 return (kern_close_range(td, uap->lowfd, uap->highfd));
1385 #ifdef COMPAT_FREEBSD12
1387 * Close open file descriptors.
1389 #ifndef _SYS_SYSPROTO_H_
1390 struct freebsd12_closefrom_args {
1396 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1400 AUDIT_ARG_FD(uap->lowfd);
1403 * Treat negative starting file descriptor values identical to
1404 * closefrom(0) which closes all files.
1406 lowfd = MAX(0, uap->lowfd);
1407 return (kern_close_range(td, lowfd, ~0U));
1409 #endif /* COMPAT_FREEBSD12 */
1411 #if defined(COMPAT_43)
1413 * Return status information about a file descriptor.
1415 #ifndef _SYS_SYSPROTO_H_
1416 struct ofstat_args {
1423 ofstat(struct thread *td, struct ofstat_args *uap)
1429 error = kern_fstat(td, uap->fd, &ub);
1432 error = copyout(&oub, uap->sb, sizeof(oub));
1436 #endif /* COMPAT_43 */
1438 #if defined(COMPAT_FREEBSD11)
1440 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1443 struct freebsd11_stat osb;
1446 error = kern_fstat(td, uap->fd, &sb);
1449 error = freebsd11_cvtstat(&sb, &osb);
1451 error = copyout(&osb, uap->sb, sizeof(osb));
1454 #endif /* COMPAT_FREEBSD11 */
1457 * Return status information about a file descriptor.
1459 #ifndef _SYS_SYSPROTO_H_
1467 sys_fstat(struct thread *td, struct fstat_args *uap)
1472 error = kern_fstat(td, uap->fd, &ub);
1474 error = copyout(&ub, uap->sb, sizeof(ub));
1479 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1486 error = fget(td, fd, &cap_fstat_rights, &fp);
1487 if (__predict_false(error != 0))
1490 AUDIT_ARG_FILE(td->td_proc, fp);
1492 error = fo_stat(fp, sbp, td->td_ucred, td);
1494 #ifdef __STAT_TIME_T_EXT
1495 sbp->st_atim_ext = 0;
1496 sbp->st_mtim_ext = 0;
1497 sbp->st_ctim_ext = 0;
1498 sbp->st_btim_ext = 0;
1501 if (KTRPOINT(td, KTR_STRUCT))
1502 ktrstat_error(sbp, error);
1507 #if defined(COMPAT_FREEBSD11)
1509 * Return status information about a file descriptor.
1511 #ifndef _SYS_SYSPROTO_H_
1512 struct freebsd11_nfstat_args {
1519 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1525 error = kern_fstat(td, uap->fd, &ub);
1527 freebsd11_cvtnstat(&ub, &nub);
1528 error = copyout(&nub, uap->sb, sizeof(nub));
1532 #endif /* COMPAT_FREEBSD11 */
1535 * Return pathconf information about a file descriptor.
1537 #ifndef _SYS_SYSPROTO_H_
1538 struct fpathconf_args {
1545 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1550 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1552 td->td_retval[0] = value;
1557 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1563 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1567 if (name == _PC_ASYNC_IO) {
1568 *valuep = _POSIX_ASYNCHRONOUS_IO;
1573 vn_lock(vp, LK_SHARED | LK_RETRY);
1574 error = VOP_PATHCONF(vp, name, valuep);
1576 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1577 if (name != _PC_PIPE_BUF) {
1592 * Copy filecaps structure allocating memory for ioctls array if needed.
1594 * The last parameter indicates whether the fdtable is locked. If it is not and
1595 * ioctls are encountered, copying fails and the caller must lock the table.
1597 * Note that if the table was not locked, the caller has to check the relevant
1598 * sequence counter to determine whether the operation was successful.
1601 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1605 if (src->fc_ioctls != NULL && !locked)
1607 memcpy(dst, src, sizeof(*src));
1608 if (src->fc_ioctls == NULL)
1611 KASSERT(src->fc_nioctls > 0,
1612 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1614 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1615 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1616 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1621 filecaps_copy_prep(const struct filecaps *src)
1626 if (__predict_true(src->fc_ioctls == NULL))
1629 KASSERT(src->fc_nioctls > 0,
1630 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1632 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1633 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1638 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1644 if (__predict_true(src->fc_ioctls == NULL)) {
1645 MPASS(ioctls == NULL);
1649 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1650 dst->fc_ioctls = ioctls;
1651 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1655 * Move filecaps structure to the new place and clear the old place.
1658 filecaps_move(struct filecaps *src, struct filecaps *dst)
1662 bzero(src, sizeof(*src));
1666 * Fill the given filecaps structure with full rights.
1669 filecaps_fill(struct filecaps *fcaps)
1672 CAP_ALL(&fcaps->fc_rights);
1673 fcaps->fc_ioctls = NULL;
1674 fcaps->fc_nioctls = -1;
1675 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1679 * Free memory allocated within filecaps structure.
1682 filecaps_free(struct filecaps *fcaps)
1685 free(fcaps->fc_ioctls, M_FILECAPS);
1686 bzero(fcaps, sizeof(*fcaps));
1690 filecaps_free_prep(struct filecaps *fcaps)
1694 ioctls = fcaps->fc_ioctls;
1695 bzero(fcaps, sizeof(*fcaps));
1700 filecaps_free_finish(u_long *ioctls)
1703 free(ioctls, M_FILECAPS);
1707 * Validate the given filecaps structure.
1710 filecaps_validate(const struct filecaps *fcaps, const char *func)
1713 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1714 ("%s: invalid rights", func));
1715 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1716 ("%s: invalid fcntls", func));
1717 KASSERT(fcaps->fc_fcntls == 0 ||
1718 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1719 ("%s: fcntls without CAP_FCNTL", func));
1720 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1721 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1722 ("%s: invalid ioctls", func));
1723 KASSERT(fcaps->fc_nioctls == 0 ||
1724 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1725 ("%s: ioctls without CAP_IOCTL", func));
1729 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1733 FILEDESC_XLOCK_ASSERT(fdp);
1735 nfd1 = fdp->fd_nfiles * 2;
1738 fdgrowtable(fdp, nfd1);
1742 * Grow the file table to accommodate (at least) nfd descriptors.
1745 fdgrowtable(struct filedesc *fdp, int nfd)
1747 struct filedesc0 *fdp0;
1748 struct freetable *ft;
1749 struct fdescenttbl *ntable;
1750 struct fdescenttbl *otable;
1751 int nnfiles, onfiles;
1752 NDSLOTTYPE *nmap, *omap;
1754 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1756 /* save old values */
1757 onfiles = fdp->fd_nfiles;
1758 otable = fdp->fd_files;
1761 /* compute the size of the new table */
1762 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1763 if (nnfiles <= onfiles)
1764 /* the table is already large enough */
1768 * Allocate a new table. We need enough space for the number of
1769 * entries, file entries themselves and the struct freetable we will use
1770 * when we decommission the table and place it on the freelist.
1771 * We place the struct freetable in the middle so we don't have
1772 * to worry about padding.
1774 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1775 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1776 sizeof(struct freetable),
1777 M_FILEDESC, M_ZERO | M_WAITOK);
1778 /* copy the old data */
1779 ntable->fdt_nfiles = nnfiles;
1780 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1781 onfiles * sizeof(ntable->fdt_ofiles[0]));
1784 * Allocate a new map only if the old is not large enough. It will
1785 * grow at a slower rate than the table as it can map more
1786 * entries than the table can hold.
1788 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1789 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1791 /* copy over the old data and update the pointer */
1792 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1797 * Make sure that ntable is correctly initialized before we replace
1798 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1801 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1804 * Do not free the old file table, as some threads may still
1805 * reference entries within it. Instead, place it on a freelist
1806 * which will be processed when the struct filedesc is released.
1808 * Note that if onfiles == NDFILE, we're dealing with the original
1809 * static allocation contained within (struct filedesc0 *)fdp,
1810 * which must not be freed.
1812 if (onfiles > NDFILE) {
1813 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1814 fdp0 = (struct filedesc0 *)fdp;
1815 ft->ft_table = otable;
1816 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1819 * The map does not have the same possibility of threads still
1820 * holding references to it. So always free it as long as it
1821 * does not reference the original static allocation.
1823 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1824 free(omap, M_FILEDESC);
1828 * Allocate a file descriptor for the process.
1831 fdalloc(struct thread *td, int minfd, int *result)
1833 struct proc *p = td->td_proc;
1834 struct filedesc *fdp = p->p_fd;
1835 int fd, maxfd, allocfd;
1840 FILEDESC_XLOCK_ASSERT(fdp);
1842 if (fdp->fd_freefile > minfd)
1843 minfd = fdp->fd_freefile;
1845 maxfd = getmaxfd(td);
1848 * Search the bitmap for a free descriptor starting at minfd.
1849 * If none is found, grow the file table.
1851 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1852 if (__predict_false(fd >= maxfd))
1854 if (__predict_false(fd >= fdp->fd_nfiles)) {
1855 allocfd = min(fd * 2, maxfd);
1857 if (RACCT_ENABLED()) {
1858 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1864 * fd is already equal to first free descriptor >= minfd, so
1865 * we only need to grow the table and we are done.
1867 fdgrowtable_exp(fdp, allocfd);
1871 * Perform some sanity checks, then mark the file descriptor as
1872 * used and return it to the caller.
1874 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1875 ("invalid descriptor %d", fd));
1876 KASSERT(!fdisused(fdp, fd),
1877 ("fd_first_free() returned non-free descriptor"));
1878 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1879 ("file descriptor isn't free"));
1886 * Allocate n file descriptors for the process.
1889 fdallocn(struct thread *td, int minfd, int *fds, int n)
1891 struct proc *p = td->td_proc;
1892 struct filedesc *fdp = p->p_fd;
1895 FILEDESC_XLOCK_ASSERT(fdp);
1897 for (i = 0; i < n; i++)
1898 if (fdalloc(td, 0, &fds[i]) != 0)
1902 for (i--; i >= 0; i--)
1903 fdunused(fdp, fds[i]);
1911 * Create a new open file structure and allocate a file descriptor for the
1912 * process that refers to it. We add one reference to the file for the
1913 * descriptor table and one reference for resultfp. This is to prevent us
1914 * being preempted and the entry in the descriptor table closed after we
1915 * release the FILEDESC lock.
1918 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1919 struct filecaps *fcaps)
1924 error = falloc_noinstall(td, &fp);
1926 return (error); /* no reference held on error */
1928 error = finstall(td, fp, &fd, flags, fcaps);
1930 fdrop(fp, td); /* one reference (fp only) */
1934 if (resultfp != NULL)
1935 *resultfp = fp; /* copy out result */
1937 fdrop(fp, td); /* release local reference */
1939 if (resultfd != NULL)
1946 * Create a new open file structure without allocating a file descriptor.
1949 falloc_noinstall(struct thread *td, struct file **resultfp)
1952 int maxuserfiles = maxfiles - (maxfiles / 20);
1954 static struct timeval lastfail;
1957 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1959 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1960 if ((openfiles_new >= maxuserfiles &&
1961 priv_check(td, PRIV_MAXFILES) != 0) ||
1962 openfiles_new >= maxfiles) {
1963 atomic_subtract_int(&openfiles, 1);
1964 if (ppsratecheck(&lastfail, &curfail, 1)) {
1965 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1966 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1970 fp = uma_zalloc(file_zone, M_WAITOK);
1971 bzero(fp, sizeof(*fp));
1972 refcount_init(&fp->f_count, 1);
1973 fp->f_cred = crhold(td->td_ucred);
1974 fp->f_ops = &badfileops;
1980 * Install a file in a file descriptor table.
1983 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1984 struct filecaps *fcaps)
1986 struct filedescent *fde;
1990 filecaps_validate(fcaps, __func__);
1991 FILEDESC_XLOCK_ASSERT(fdp);
1993 fde = &fdp->fd_ofiles[fd];
1995 seqc_write_begin(&fde->fde_seqc);
1998 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2000 filecaps_move(fcaps, &fde->fde_caps);
2002 filecaps_fill(&fde->fde_caps);
2004 seqc_write_end(&fde->fde_seqc);
2009 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2010 struct filecaps *fcaps)
2012 struct filedesc *fdp = td->td_proc->p_fd;
2019 FILEDESC_XLOCK(fdp);
2020 error = fdalloc(td, 0, fd);
2021 if (__predict_false(error != 0)) {
2022 FILEDESC_XUNLOCK(fdp);
2026 _finstall(fdp, fp, *fd, flags, fcaps);
2027 FILEDESC_XUNLOCK(fdp);
2032 * Build a new filedesc structure from another.
2034 * If fdp is not NULL, return with it shared locked.
2037 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2039 struct filedesc0 *newfdp0;
2040 struct filedesc *newfdp;
2043 MPASS(lastfile != NULL);
2045 MPASS(lastfile == NULL);
2047 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2048 newfdp = &newfdp0->fd_fd;
2050 /* Create the file descriptor table. */
2051 FILEDESC_LOCK_INIT(newfdp);
2052 refcount_init(&newfdp->fd_refcnt, 1);
2053 refcount_init(&newfdp->fd_holdcnt, 1);
2054 newfdp->fd_map = newfdp0->fd_dmap;
2055 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2056 newfdp->fd_files->fdt_nfiles = NDFILE;
2061 FILEDESC_SLOCK(fdp);
2063 FILEDESC_SUNLOCK(fdp);
2068 *lastfile = fdlastfile(fdp);
2069 if (*lastfile < newfdp->fd_nfiles)
2071 FILEDESC_SUNLOCK(fdp);
2072 fdgrowtable(newfdp, *lastfile + 1);
2073 FILEDESC_SLOCK(fdp);
2080 * Build a pwddesc structure from another.
2081 * Copy the current, root, and jail root vnode references.
2083 * If pdp is not NULL, return with it shared locked.
2086 pdinit(struct pwddesc *pdp, bool keeplock)
2088 struct pwddesc *newpdp;
2091 newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2093 PWDDESC_LOCK_INIT(newpdp);
2094 refcount_init(&newpdp->pd_refcount, 1);
2095 newpdp->pd_cmask = CMASK;
2098 newpwd = pwd_alloc();
2099 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2104 newpwd = pwd_hold_pwddesc(pdp);
2105 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2107 PWDDESC_XUNLOCK(pdp);
2111 static struct filedesc *
2112 fdhold(struct proc *p)
2114 struct filedesc *fdp;
2116 PROC_LOCK_ASSERT(p, MA_OWNED);
2119 refcount_acquire(&fdp->fd_holdcnt);
2123 static struct pwddesc *
2124 pdhold(struct proc *p)
2126 struct pwddesc *pdp;
2128 PROC_LOCK_ASSERT(p, MA_OWNED);
2131 refcount_acquire(&pdp->pd_refcount);
2136 fddrop(struct filedesc *fdp)
2139 if (fdp->fd_holdcnt > 1) {
2140 if (refcount_release(&fdp->fd_holdcnt) == 0)
2144 FILEDESC_LOCK_DESTROY(fdp);
2145 uma_zfree(filedesc0_zone, fdp);
2149 pddrop(struct pwddesc *pdp)
2153 if (refcount_release_if_not_last(&pdp->pd_refcount))
2157 if (refcount_release(&pdp->pd_refcount) == 0) {
2158 PWDDESC_XUNLOCK(pdp);
2161 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2163 PWDDESC_XUNLOCK(pdp);
2166 PWDDESC_LOCK_DESTROY(pdp);
2167 free(pdp, M_PWDDESC);
2171 * Share a filedesc structure.
2174 fdshare(struct filedesc *fdp)
2177 refcount_acquire(&fdp->fd_refcnt);
2182 * Share a pwddesc structure.
2185 pdshare(struct pwddesc *pdp)
2187 refcount_acquire(&pdp->pd_refcount);
2192 * Unshare a filedesc structure, if necessary by making a copy
2195 fdunshare(struct thread *td)
2197 struct filedesc *tmp;
2198 struct proc *p = td->td_proc;
2200 if (p->p_fd->fd_refcnt == 1)
2203 tmp = fdcopy(p->p_fd);
2209 * Unshare a pwddesc structure.
2212 pdunshare(struct thread *td)
2214 struct pwddesc *pdp;
2219 if (p->p_pd->pd_refcount == 1)
2222 pdp = pdcopy(p->p_pd);
2228 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2232 td->td_proc->p_fd = fdp;
2236 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2237 * this is to ease callers, not catch errors.
2240 fdcopy(struct filedesc *fdp)
2242 struct filedesc *newfdp;
2243 struct filedescent *nfde, *ofde;
2248 newfdp = fdinit(fdp, true, &lastfile);
2249 /* copy all passable descriptors (i.e. not kqueue) */
2250 newfdp->fd_freefile = -1;
2251 for (i = 0; i <= lastfile; ++i) {
2252 ofde = &fdp->fd_ofiles[i];
2253 if (ofde->fde_file == NULL ||
2254 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2255 !fhold(ofde->fde_file)) {
2256 if (newfdp->fd_freefile == -1)
2257 newfdp->fd_freefile = i;
2260 nfde = &newfdp->fd_ofiles[i];
2262 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2263 fdused_init(newfdp, i);
2265 if (newfdp->fd_freefile == -1)
2266 newfdp->fd_freefile = i;
2267 FILEDESC_SUNLOCK(fdp);
2272 * Copy a pwddesc structure.
2275 pdcopy(struct pwddesc *pdp)
2277 struct pwddesc *newpdp;
2281 newpdp = pdinit(pdp, true);
2282 newpdp->pd_cmask = pdp->pd_cmask;
2283 PWDDESC_XUNLOCK(pdp);
2288 * Copies a filedesc structure, while remapping all file descriptors
2289 * stored inside using a translation table.
2291 * File descriptors are copied over to the new file descriptor table,
2292 * regardless of whether the close-on-exec flag is set.
2295 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2296 struct filedesc **ret)
2298 struct filedesc *newfdp;
2299 struct filedescent *nfde, *ofde;
2300 int error, i, lastfile;
2304 newfdp = fdinit(fdp, true, &lastfile);
2305 if (nfds > lastfile + 1) {
2306 /* New table cannot be larger than the old one. */
2310 /* Copy all passable descriptors (i.e. not kqueue). */
2311 newfdp->fd_freefile = nfds;
2312 for (i = 0; i < nfds; ++i) {
2313 if (fds[i] < 0 || fds[i] > lastfile) {
2314 /* File descriptor out of bounds. */
2318 ofde = &fdp->fd_ofiles[fds[i]];
2319 if (ofde->fde_file == NULL) {
2320 /* Unused file descriptor. */
2324 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2325 /* File descriptor cannot be passed. */
2329 if (!fhold(ofde->fde_file)) {
2333 nfde = &newfdp->fd_ofiles[i];
2335 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2336 fdused_init(newfdp, i);
2338 FILEDESC_SUNLOCK(fdp);
2342 FILEDESC_SUNLOCK(fdp);
2343 fdescfree_remapped(newfdp);
2348 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2349 * one of processes using it exits) and the table used to be shared.
2352 fdclearlocks(struct thread *td)
2354 struct filedesc *fdp;
2355 struct filedesc_to_leader *fdtol;
2365 MPASS(fdtol != NULL);
2367 FILEDESC_XLOCK(fdp);
2368 KASSERT(fdtol->fdl_refcount > 0,
2369 ("filedesc_to_refcount botch: fdl_refcount=%d",
2370 fdtol->fdl_refcount));
2371 if (fdtol->fdl_refcount == 1 &&
2372 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2373 lastfile = fdlastfile(fdp);
2374 for (i = 0; i <= lastfile; i++) {
2375 fp = fdp->fd_ofiles[i].fde_file;
2376 if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2379 FILEDESC_XUNLOCK(fdp);
2380 lf.l_whence = SEEK_SET;
2383 lf.l_type = F_UNLCK;
2385 (void) VOP_ADVLOCK(vp,
2386 (caddr_t)p->p_leader, F_UNLCK,
2388 FILEDESC_XLOCK(fdp);
2393 if (fdtol->fdl_refcount == 1) {
2394 if (fdp->fd_holdleaderscount > 0 &&
2395 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2397 * close() or kern_dup() has cleared a reference
2398 * in a shared file descriptor table.
2400 fdp->fd_holdleaderswakeup = 1;
2401 sx_sleep(&fdp->fd_holdleaderscount,
2402 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2405 if (fdtol->fdl_holdcount > 0) {
2407 * Ensure that fdtol->fdl_leader remains
2408 * valid in closef().
2410 fdtol->fdl_wakeup = 1;
2411 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2416 fdtol->fdl_refcount--;
2417 if (fdtol->fdl_refcount == 0 &&
2418 fdtol->fdl_holdcount == 0) {
2419 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2420 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2424 FILEDESC_XUNLOCK(fdp);
2426 free(fdtol, M_FILEDESC_TO_LEADER);
2430 * Release a filedesc structure.
2433 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2435 struct filedesc0 *fdp0;
2436 struct freetable *ft, *tft;
2437 struct filedescent *fde;
2441 lastfile = fdlastfile_single(fdp);
2442 for (i = 0; i <= lastfile; i++) {
2443 fde = &fdp->fd_ofiles[i];
2448 (void) closef(fp, td);
2454 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2455 free(fdp->fd_map, M_FILEDESC);
2456 if (fdp->fd_nfiles > NDFILE)
2457 free(fdp->fd_files, M_FILEDESC);
2459 fdp0 = (struct filedesc0 *)fdp;
2460 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2461 free(ft->ft_table, M_FILEDESC);
2467 fdescfree(struct thread *td)
2470 struct filedesc *fdp;
2477 if (RACCT_ENABLED())
2478 racct_set_unlocked(p, RACCT_NOFILE, 0);
2481 if (p->p_fdtol != NULL)
2488 if (refcount_release(&fdp->fd_refcnt) == 0)
2491 fdescfree_fds(td, fdp, 1);
2495 pdescfree(struct thread *td)
2498 struct pwddesc *pdp;
2512 fdescfree_remapped(struct filedesc *fdp)
2514 fdescfree_fds(curthread, fdp, 0);
2518 * For setugid programs, we don't want to people to use that setugidness
2519 * to generate error messages which write to a file which otherwise would
2520 * otherwise be off-limits to the process. We check for filesystems where
2521 * the vnode can change out from under us after execve (like [lin]procfs).
2523 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2524 * sufficient. We also don't check for setugidness since we know we are.
2527 is_unsafe(struct file *fp)
2531 if (fp->f_type != DTYPE_VNODE)
2535 return ((vp->v_vflag & VV_PROCDEP) != 0);
2539 * Make this setguid thing safe, if at all possible.
2542 fdsetugidsafety(struct thread *td)
2544 struct filedesc *fdp;
2548 fdp = td->td_proc->p_fd;
2549 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2550 MPASS(fdp->fd_nfiles >= 3);
2551 for (i = 0; i <= 2; i++) {
2552 fp = fdp->fd_ofiles[i].fde_file;
2553 if (fp != NULL && is_unsafe(fp)) {
2554 FILEDESC_XLOCK(fdp);
2555 knote_fdclose(td, i);
2557 * NULL-out descriptor prior to close to avoid
2558 * a race while close blocks.
2561 FILEDESC_XUNLOCK(fdp);
2562 (void) closef(fp, td);
2568 * If a specific file object occupies a specific file descriptor, close the
2569 * file descriptor entry and drop a reference on the file object. This is a
2570 * convenience function to handle a subsequent error in a function that calls
2571 * falloc() that handles the race that another thread might have closed the
2572 * file descriptor out from under the thread creating the file object.
2575 fdclose(struct thread *td, struct file *fp, int idx)
2577 struct filedesc *fdp = td->td_proc->p_fd;
2579 FILEDESC_XLOCK(fdp);
2580 if (fdp->fd_ofiles[idx].fde_file == fp) {
2582 FILEDESC_XUNLOCK(fdp);
2585 FILEDESC_XUNLOCK(fdp);
2589 * Close any files on exec?
2592 fdcloseexec(struct thread *td)
2594 struct filedesc *fdp;
2595 struct filedescent *fde;
2599 fdp = td->td_proc->p_fd;
2600 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2601 lastfile = fdlastfile_single(fdp);
2602 for (i = 0; i <= lastfile; i++) {
2603 fde = &fdp->fd_ofiles[i];
2605 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2606 (fde->fde_flags & UF_EXCLOSE))) {
2607 FILEDESC_XLOCK(fdp);
2609 (void) closefp(fdp, i, fp, td, 0);
2610 FILEDESC_UNLOCK_ASSERT(fdp);
2616 * It is unsafe for set[ug]id processes to be started with file
2617 * descriptors 0..2 closed, as these descriptors are given implicit
2618 * significance in the Standard C library. fdcheckstd() will create a
2619 * descriptor referencing /dev/null for each of stdin, stdout, and
2620 * stderr that is not already open.
2623 fdcheckstd(struct thread *td)
2625 struct filedesc *fdp;
2627 int i, error, devnull;
2629 fdp = td->td_proc->p_fd;
2630 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2631 MPASS(fdp->fd_nfiles >= 3);
2633 for (i = 0; i <= 2; i++) {
2634 if (fdp->fd_ofiles[i].fde_file != NULL)
2637 save = td->td_retval[0];
2638 if (devnull != -1) {
2639 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2641 error = kern_openat(td, AT_FDCWD, "/dev/null",
2642 UIO_SYSSPACE, O_RDWR, 0);
2644 devnull = td->td_retval[0];
2645 KASSERT(devnull == i, ("we didn't get our fd"));
2648 td->td_retval[0] = save;
2656 * Internal form of close. Decrement reference count on file structure.
2657 * Note: td may be NULL when closing a file that was being passed in a
2661 closef(struct file *fp, struct thread *td)
2665 struct filedesc_to_leader *fdtol;
2666 struct filedesc *fdp;
2669 * POSIX record locking dictates that any close releases ALL
2670 * locks owned by this process. This is handled by setting
2671 * a flag in the unlock to free ONLY locks obeying POSIX
2672 * semantics, and not to free BSD-style file locks.
2673 * If the descriptor was in a message, POSIX-style locks
2674 * aren't passed with the descriptor, and the thread pointer
2675 * will be NULL. Callers should be careful only to pass a
2676 * NULL thread pointer when there really is no owning
2677 * context that might have locks, or the locks will be
2680 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2682 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2683 lf.l_whence = SEEK_SET;
2686 lf.l_type = F_UNLCK;
2687 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2688 F_UNLCK, &lf, F_POSIX);
2690 fdtol = td->td_proc->p_fdtol;
2691 if (fdtol != NULL) {
2693 * Handle special case where file descriptor table is
2694 * shared between multiple process leaders.
2696 fdp = td->td_proc->p_fd;
2697 FILEDESC_XLOCK(fdp);
2698 for (fdtol = fdtol->fdl_next;
2699 fdtol != td->td_proc->p_fdtol;
2700 fdtol = fdtol->fdl_next) {
2701 if ((fdtol->fdl_leader->p_flag &
2704 fdtol->fdl_holdcount++;
2705 FILEDESC_XUNLOCK(fdp);
2706 lf.l_whence = SEEK_SET;
2709 lf.l_type = F_UNLCK;
2711 (void) VOP_ADVLOCK(vp,
2712 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2714 FILEDESC_XLOCK(fdp);
2715 fdtol->fdl_holdcount--;
2716 if (fdtol->fdl_holdcount == 0 &&
2717 fdtol->fdl_wakeup != 0) {
2718 fdtol->fdl_wakeup = 0;
2722 FILEDESC_XUNLOCK(fdp);
2725 return (fdrop(fp, td));
2729 * Initialize the file pointer with the specified properties.
2731 * The ops are set with release semantics to be certain that the flags, type,
2732 * and data are visible when ops is. This is to prevent ops methods from being
2733 * called with bad data.
2736 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2741 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2745 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2747 fp->f_seqcount[UIO_READ] = 1;
2748 fp->f_seqcount[UIO_WRITE] = 1;
2749 finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2754 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2755 struct file **fpp, struct filecaps *havecapsp)
2757 struct filedescent *fde;
2760 FILEDESC_LOCK_ASSERT(fdp);
2762 fde = fdeget_locked(fdp, fd);
2769 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2774 if (havecapsp != NULL)
2775 filecaps_copy(&fde->fde_caps, havecapsp, true);
2777 *fpp = fde->fde_file;
2785 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2786 struct file **fpp, struct filecaps *havecapsp)
2788 struct filedesc *fdp = td->td_proc->p_fd;
2790 #ifndef CAPABILITIES
2791 error = fget_unlocked(fdp, fd, needrightsp, fpp);
2792 if (havecapsp != NULL && error == 0)
2793 filecaps_fill(havecapsp);
2800 error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2804 if (havecapsp != NULL) {
2805 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2806 havecapsp, false)) {
2812 if (!fd_modified(fdp, fd, seq))
2821 FILEDESC_SLOCK(fdp);
2822 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2823 if (error == 0 && !fhold(*fpp))
2825 FILEDESC_SUNLOCK(fdp);
2832 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2834 const struct filedescent *fde;
2835 const struct fdescenttbl *fdt;
2836 struct filedesc *fdp;
2839 const cap_rights_t *haverights;
2840 cap_rights_t rights;
2843 VFS_SMR_ASSERT_ENTERED();
2845 rights = *ndp->ni_rightsneeded;
2846 cap_rights_set_one(&rights, CAP_LOOKUP);
2848 fdp = curproc->p_fd;
2849 fdt = fdp->fd_files;
2850 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2852 seq = seqc_read_any(fd_seqc(fdt, fd));
2853 if (__predict_false(seqc_in_modify(seq)))
2855 fde = &fdt->fdt_ofiles[fd];
2856 haverights = cap_rights_fde_inline(fde);
2858 if (__predict_false(fp == NULL))
2860 if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2862 *fsearch = ((fp->f_flag & FSEARCH) != 0);
2864 if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2867 if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2871 * Use an acquire barrier to force re-reading of fdt so it is
2872 * refreshed for verification.
2874 atomic_thread_fence_acq();
2875 fdt = fdp->fd_files;
2876 if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
2879 * If file descriptor doesn't have all rights,
2880 * all lookups relative to it must also be
2881 * strictly relative.
2883 * Not yet supported by fast path.
2886 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
2887 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
2888 ndp->ni_filecaps.fc_nioctls != -1) {
2890 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
2900 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2902 const struct fdescenttbl *fdt;
2903 struct filedesc *fdp;
2907 VFS_SMR_ASSERT_ENTERED();
2909 fdp = curproc->p_fd;
2910 fdt = fdp->fd_files;
2911 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2913 fp = fdt->fdt_ofiles[fd].fde_file;
2914 if (__predict_false(fp == NULL))
2916 *fsearch = ((fp->f_flag & FSEARCH) != 0);
2918 if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2922 * Use an acquire barrier to force re-reading of fdt so it is
2923 * refreshed for verification.
2925 atomic_thread_fence_acq();
2926 fdt = fdp->fd_files;
2927 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
2929 filecaps_fill(&ndp->ni_filecaps);
2936 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2937 struct file **fpp, seqc_t *seqp)
2940 const struct filedescent *fde;
2942 const struct fdescenttbl *fdt;
2946 cap_rights_t haverights;
2950 fdt = fdp->fd_files;
2951 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2954 * Fetch the descriptor locklessly. We avoid fdrop() races by
2955 * never raising a refcount above 0. To accomplish this we have
2956 * to use a cmpset loop rather than an atomic_add. The descriptor
2957 * must be re-verified once we acquire a reference to be certain
2958 * that the identity is still correct and we did not lose a race
2959 * due to preemption.
2963 seq = seqc_read(fd_seqc(fdt, fd));
2964 fde = &fdt->fdt_ofiles[fd];
2965 haverights = *cap_rights_fde_inline(fde);
2967 if (!seqc_consistent(fd_seqc(fdt, fd), seq))
2970 fp = fdt->fdt_ofiles[fd].fde_file;
2975 error = cap_check_inline(&haverights, needrightsp);
2979 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
2981 * The count was found either saturated or zero.
2982 * This re-read is not any more racy than using the
2983 * return value from fcmpset.
2985 if (refcount_load(&fp->f_count) != 0)
2988 * Force a reload. Other thread could reallocate the
2989 * table before this fd was closed, so it is possible
2990 * that there is a stale fp pointer in cached version.
2992 fdt = atomic_load_ptr(&fdp->fd_files);
2996 * Use an acquire barrier to force re-reading of fdt so it is
2997 * refreshed for verification.
2999 atomic_thread_fence_acq();
3000 fdt = fdp->fd_files;
3002 if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
3004 if (fp == fdt->fdt_ofiles[fd].fde_file)
3007 fdrop(fp, curthread);
3019 * See the comments in fget_unlocked_seq for an explanation of how this works.
3021 * This is a simplified variant which bails out to the aforementioned routine
3022 * if anything goes wrong. In practice this only happens when userspace is
3023 * racing with itself.
3026 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3030 const struct filedescent *fde;
3032 const struct fdescenttbl *fdt;
3036 const cap_rights_t *haverights;
3039 fdt = fdp->fd_files;
3040 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3043 seq = seqc_read_any(fd_seqc(fdt, fd));
3044 if (__predict_false(seqc_in_modify(seq)))
3046 fde = &fdt->fdt_ofiles[fd];
3047 haverights = cap_rights_fde_inline(fde);
3050 fp = fdt->fdt_ofiles[fd].fde_file;
3052 if (__predict_false(fp == NULL))
3055 if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3058 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3062 * Use an acquire barrier to force re-reading of fdt so it is
3063 * refreshed for verification.
3065 atomic_thread_fence_acq();
3066 fdt = fdp->fd_files;
3068 if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3070 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3076 fdrop(fp, curthread);
3078 return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
3082 * Extract the file pointer associated with the specified descriptor for the
3083 * current user process.
3085 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3088 * File's rights will be checked against the capability rights mask.
3090 * If an error occurred the non-zero error is returned and *fpp is set to
3091 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
3092 * responsible for fdrop().
3095 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3096 cap_rights_t *needrightsp)
3098 struct filedesc *fdp;
3103 fdp = td->td_proc->p_fd;
3104 error = fget_unlocked(fdp, fd, needrightsp, &fp);
3105 if (__predict_false(error != 0))
3107 if (__predict_false(fp->f_ops == &badfileops)) {
3113 * FREAD and FWRITE failure return EBADF as per POSIX.
3119 if ((fp->f_flag & flags) == 0)
3123 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3124 ((fp->f_flag & FWRITE) != 0))
3130 KASSERT(0, ("wrong flags"));
3143 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3146 return (_fget(td, fd, fpp, 0, rightsp));
3150 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3154 #ifndef CAPABILITIES
3155 error = _fget(td, fd, fpp, 0, rightsp);
3156 if (maxprotp != NULL)
3157 *maxprotp = VM_PROT_ALL;
3160 cap_rights_t fdrights;
3161 struct filedesc *fdp;
3166 fdp = td->td_proc->p_fd;
3167 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3169 error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3170 if (__predict_false(error != 0))
3172 if (__predict_false(fp->f_ops == &badfileops)) {
3176 if (maxprotp != NULL)
3177 fdrights = *cap_rights(fdp, fd);
3178 if (!fd_modified(fdp, fd, seq))
3184 * If requested, convert capability rights to access flags.
3186 if (maxprotp != NULL)
3187 *maxprotp = cap_rights_to_vmprot(&fdrights);
3194 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3197 return (_fget(td, fd, fpp, FREAD, rightsp));
3201 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3204 return (_fget(td, fd, fpp, FWRITE, rightsp));
3208 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3211 struct filedesc *fdp = td->td_proc->p_fd;
3212 #ifndef CAPABILITIES
3213 return (fget_unlocked(fdp, fd, rightsp, fpp));
3220 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3222 error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3225 error = cap_fcntl_check(fdp, fd, needfcntl);
3226 if (!fd_modified(fdp, fd, seq))
3240 * Like fget() but loads the underlying vnode, or returns an error if the
3241 * descriptor does not represent a vnode. Note that pipes use vnodes but
3242 * never have VM objects. The returned vnode will be vref()'d.
3244 * XXX: what about the unused flags ?
3247 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3254 error = _fget(td, fd, &fp, flags, needrightsp);
3257 if (fp->f_vnode == NULL) {
3269 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3272 return (_fgetvp(td, fd, 0, rightsp, vpp));
3276 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3277 struct filecaps *havecaps, struct vnode **vpp)
3279 struct filecaps caps;
3283 error = fget_cap(td, fd, needrightsp, &fp, &caps);
3286 if (fp->f_ops == &badfileops) {
3290 if (fp->f_vnode == NULL) {
3302 filecaps_free(&caps);
3308 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3311 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3315 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3318 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3323 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3327 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3332 * Handle the last reference to a file being closed.
3334 * Without the noinline attribute clang keeps inlining the func thorough this
3335 * file when fdrop is used.
3338 _fdrop(struct file *fp, struct thread *td)
3344 count = refcount_load(&fp->f_count);
3346 panic("fdrop: fp %p count %d", fp, count);
3348 error = fo_close(fp, td);
3349 atomic_subtract_int(&openfiles, 1);
3351 free(fp->f_advice, M_FADVISE);
3352 uma_zfree(file_zone, fp);
3358 * Apply an advisory lock on a file descriptor.
3360 * Just attempt to get a record lock of the requested type on the entire file
3361 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3363 #ifndef _SYS_SYSPROTO_H_
3371 sys_flock(struct thread *td, struct flock_args *uap)
3378 error = fget(td, uap->fd, &cap_flock_rights, &fp);
3381 if (fp->f_type != DTYPE_VNODE) {
3383 return (EOPNOTSUPP);
3387 lf.l_whence = SEEK_SET;
3390 if (uap->how & LOCK_UN) {
3391 lf.l_type = F_UNLCK;
3392 atomic_clear_int(&fp->f_flag, FHASLOCK);
3393 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3396 if (uap->how & LOCK_EX)
3397 lf.l_type = F_WRLCK;
3398 else if (uap->how & LOCK_SH)
3399 lf.l_type = F_RDLCK;
3404 atomic_set_int(&fp->f_flag, FHASLOCK);
3405 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3406 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3412 * Duplicate the specified descriptor to a free descriptor.
3415 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3416 int openerror, int *indxp)
3418 struct filedescent *newfde, *oldfde;
3423 KASSERT(openerror == ENODEV || openerror == ENXIO,
3424 ("unexpected error %d in %s", openerror, __func__));
3427 * If the to-be-dup'd fd number is greater than the allowed number
3428 * of file descriptors, or the fd to be dup'd has already been
3429 * closed, then reject.
3431 FILEDESC_XLOCK(fdp);
3432 if ((fp = fget_locked(fdp, dfd)) == NULL) {
3433 FILEDESC_XUNLOCK(fdp);
3437 error = fdalloc(td, 0, &indx);
3439 FILEDESC_XUNLOCK(fdp);
3444 * There are two cases of interest here.
3446 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3448 * For ENXIO steal away the file structure from (dfd) and store it in
3449 * (indx). (dfd) is effectively closed by this operation.
3451 switch (openerror) {
3454 * Check that the mode the file is being opened for is a
3455 * subset of the mode of the existing descriptor.
3457 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3458 fdunused(fdp, indx);
3459 FILEDESC_XUNLOCK(fdp);
3463 fdunused(fdp, indx);
3464 FILEDESC_XUNLOCK(fdp);
3467 newfde = &fdp->fd_ofiles[indx];
3468 oldfde = &fdp->fd_ofiles[dfd];
3469 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3471 seqc_write_begin(&newfde->fde_seqc);
3473 memcpy(newfde, oldfde, fde_change_size);
3474 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3477 seqc_write_end(&newfde->fde_seqc);
3482 * Steal away the file pointer from dfd and stuff it into indx.
3484 newfde = &fdp->fd_ofiles[indx];
3485 oldfde = &fdp->fd_ofiles[dfd];
3487 seqc_write_begin(&newfde->fde_seqc);
3489 memcpy(newfde, oldfde, fde_change_size);
3490 oldfde->fde_file = NULL;
3493 seqc_write_end(&newfde->fde_seqc);
3497 FILEDESC_XUNLOCK(fdp);
3503 * This sysctl determines if we will allow a process to chroot(2) if it
3504 * has a directory open:
3505 * 0: disallowed for all processes.
3506 * 1: allowed for processes that were not already chroot(2)'ed.
3507 * 2: allowed for all processes.
3510 static int chroot_allow_open_directories = 1;
3512 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3513 &chroot_allow_open_directories, 0,
3514 "Allow a process to chroot(2) if it has a directory open");
3517 * Helper function for raised chroot(2) security function: Refuse if
3518 * any filedescriptors are open directories.
3521 chroot_refuse_vdir_fds(struct filedesc *fdp)
3527 FILEDESC_LOCK_ASSERT(fdp);
3529 lastfile = fdlastfile(fdp);
3530 for (fd = 0; fd <= lastfile; fd++) {
3531 fp = fget_locked(fdp, fd);
3534 if (fp->f_type == DTYPE_VNODE) {
3536 if (vp->v_type == VDIR)
3544 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3547 if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3548 vrefact(oldpwd->pwd_cdir);
3549 newpwd->pwd_cdir = oldpwd->pwd_cdir;
3552 if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3553 vrefact(oldpwd->pwd_rdir);
3554 newpwd->pwd_rdir = oldpwd->pwd_rdir;
3557 if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3558 vrefact(oldpwd->pwd_jdir);
3559 newpwd->pwd_jdir = oldpwd->pwd_jdir;
3564 pwd_hold_pwddesc(struct pwddesc *pdp)
3568 PWDDESC_ASSERT_XLOCKED(pdp);
3569 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3571 refcount_acquire(&pwd->pwd_refcount);
3576 pwd_hold_smr(struct pwd *pwd)
3580 if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3587 pwd_hold(struct thread *td)
3589 struct pwddesc *pdp;
3592 pdp = td->td_proc->p_pd;
3595 pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3596 if (pwd_hold_smr(pwd)) {
3602 pwd = pwd_hold_pwddesc(pdp);
3604 PWDDESC_XUNLOCK(pdp);
3613 pwd = vfs_smr_entered_load(&curproc->p_pd->pd_pwd);
3623 pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3624 bzero(pwd, sizeof(*pwd));
3625 refcount_init(&pwd->pwd_refcount, 1);
3630 pwd_drop(struct pwd *pwd)
3633 if (!refcount_release(&pwd->pwd_refcount))
3636 if (pwd->pwd_cdir != NULL)
3637 vrele(pwd->pwd_cdir);
3638 if (pwd->pwd_rdir != NULL)
3639 vrele(pwd->pwd_rdir);
3640 if (pwd->pwd_jdir != NULL)
3641 vrele(pwd->pwd_jdir);
3642 uma_zfree_smr(pwd_zone, pwd);
3646 * Common routine for kern_chroot() and jail_attach(). The caller is
3647 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3648 * authorize this operation.
3651 pwd_chroot(struct thread *td, struct vnode *vp)
3653 struct pwddesc *pdp;
3654 struct filedesc *fdp;
3655 struct pwd *newpwd, *oldpwd;
3658 fdp = td->td_proc->p_fd;
3659 pdp = td->td_proc->p_pd;
3660 newpwd = pwd_alloc();
3661 FILEDESC_SLOCK(fdp);
3663 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3664 if (chroot_allow_open_directories == 0 ||
3665 (chroot_allow_open_directories == 1 &&
3666 oldpwd->pwd_rdir != rootvnode)) {
3667 error = chroot_refuse_vdir_fds(fdp);
3668 FILEDESC_SUNLOCK(fdp);
3670 PWDDESC_XUNLOCK(pdp);
3675 FILEDESC_SUNLOCK(fdp);
3679 newpwd->pwd_rdir = vp;
3680 if (oldpwd->pwd_jdir == NULL) {
3682 newpwd->pwd_jdir = vp;
3684 pwd_fill(oldpwd, newpwd);
3685 pwd_set(pdp, newpwd);
3686 PWDDESC_XUNLOCK(pdp);
3692 pwd_chdir(struct thread *td, struct vnode *vp)
3694 struct pwddesc *pdp;
3695 struct pwd *newpwd, *oldpwd;
3697 VNPASS(vp->v_usecount > 0, vp);
3699 newpwd = pwd_alloc();
3700 pdp = td->td_proc->p_pd;
3702 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3703 newpwd->pwd_cdir = vp;
3704 pwd_fill(oldpwd, newpwd);
3705 pwd_set(pdp, newpwd);
3706 PWDDESC_XUNLOCK(pdp);
3711 pwd_ensure_dirs(void)
3713 struct pwddesc *pdp;
3714 struct pwd *oldpwd, *newpwd;
3716 pdp = curproc->p_pd;
3718 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3719 if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3720 PWDDESC_XUNLOCK(pdp);
3723 PWDDESC_XUNLOCK(pdp);
3725 newpwd = pwd_alloc();
3727 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3728 pwd_fill(oldpwd, newpwd);
3729 if (newpwd->pwd_cdir == NULL) {
3731 newpwd->pwd_cdir = rootvnode;
3733 if (newpwd->pwd_rdir == NULL) {
3735 newpwd->pwd_rdir = rootvnode;
3737 pwd_set(pdp, newpwd);
3738 PWDDESC_XUNLOCK(pdp);
3743 pwd_set_rootvnode(void)
3745 struct pwddesc *pdp;
3746 struct pwd *oldpwd, *newpwd;
3748 pdp = curproc->p_pd;
3750 newpwd = pwd_alloc();
3752 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3754 newpwd->pwd_cdir = rootvnode;
3756 newpwd->pwd_rdir = rootvnode;
3757 pwd_fill(oldpwd, newpwd);
3758 pwd_set(pdp, newpwd);
3759 PWDDESC_XUNLOCK(pdp);
3764 * Scan all active processes and prisons to see if any of them have a current
3765 * or root directory of `olddp'. If so, replace them with the new mount point.
3768 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3770 struct pwddesc *pdp;
3771 struct pwd *newpwd, *oldpwd;
3776 if (vrefcnt(olddp) == 1)
3779 newpwd = pwd_alloc();
3780 sx_slock(&allproc_lock);
3781 FOREACH_PROC_IN_SYSTEM(p) {
3788 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3789 if (oldpwd == NULL ||
3790 (oldpwd->pwd_cdir != olddp &&
3791 oldpwd->pwd_rdir != olddp &&
3792 oldpwd->pwd_jdir != olddp)) {
3793 PWDDESC_XUNLOCK(pdp);
3797 if (oldpwd->pwd_cdir == olddp) {
3799 newpwd->pwd_cdir = newdp;
3801 if (oldpwd->pwd_rdir == olddp) {
3803 newpwd->pwd_rdir = newdp;
3805 if (oldpwd->pwd_jdir == olddp) {
3807 newpwd->pwd_jdir = newdp;
3809 pwd_fill(oldpwd, newpwd);
3810 pwd_set(pdp, newpwd);
3811 PWDDESC_XUNLOCK(pdp);
3814 newpwd = pwd_alloc();
3816 sx_sunlock(&allproc_lock);
3818 if (rootvnode == olddp) {
3823 mtx_lock(&prison0.pr_mtx);
3824 if (prison0.pr_root == olddp) {
3826 prison0.pr_root = newdp;
3829 mtx_unlock(&prison0.pr_mtx);
3830 sx_slock(&allprison_lock);
3831 TAILQ_FOREACH(pr, &allprison, pr_list) {
3832 mtx_lock(&pr->pr_mtx);
3833 if (pr->pr_root == olddp) {
3835 pr->pr_root = newdp;
3838 mtx_unlock(&pr->pr_mtx);
3840 sx_sunlock(&allprison_lock);
3845 struct filedesc_to_leader *
3846 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3848 struct filedesc_to_leader *fdtol;
3850 fdtol = malloc(sizeof(struct filedesc_to_leader),
3851 M_FILEDESC_TO_LEADER, M_WAITOK);
3852 fdtol->fdl_refcount = 1;
3853 fdtol->fdl_holdcount = 0;
3854 fdtol->fdl_wakeup = 0;
3855 fdtol->fdl_leader = leader;
3857 FILEDESC_XLOCK(fdp);
3858 fdtol->fdl_next = old->fdl_next;
3859 fdtol->fdl_prev = old;
3860 old->fdl_next = fdtol;
3861 fdtol->fdl_next->fdl_prev = fdtol;
3862 FILEDESC_XUNLOCK(fdp);
3864 fdtol->fdl_next = fdtol;
3865 fdtol->fdl_prev = fdtol;
3871 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3874 struct filedesc *fdp;
3875 int count, off, minoff;
3877 if (*(int *)arg1 != 0)
3880 fdp = curproc->p_fd;
3882 FILEDESC_SLOCK(fdp);
3884 off = NDSLOT(fdp->fd_nfiles - 1);
3885 for (minoff = NDSLOT(0); off >= minoff; --off)
3886 count += bitcountl(map[off]);
3887 FILEDESC_SUNLOCK(fdp);
3889 return (SYSCTL_OUT(req, &count, sizeof(count)));
3892 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3893 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3894 "Number of open file descriptors");
3897 * Get file structures globally.
3900 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3903 struct filedesc *fdp;
3906 int error, n, lastfile;
3908 error = sysctl_wire_old_buffer(req, 0);
3911 if (req->oldptr == NULL) {
3913 sx_slock(&allproc_lock);
3914 FOREACH_PROC_IN_SYSTEM(p) {
3916 if (p->p_state == PRS_NEW) {
3924 /* overestimates sparse tables. */
3925 n += fdp->fd_nfiles;
3928 sx_sunlock(&allproc_lock);
3929 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3932 bzero(&xf, sizeof(xf));
3933 xf.xf_size = sizeof(xf);
3934 sx_slock(&allproc_lock);
3935 FOREACH_PROC_IN_SYSTEM(p) {
3937 if (p->p_state == PRS_NEW) {
3941 if (p_cansee(req->td, p) != 0) {
3945 xf.xf_pid = p->p_pid;
3946 xf.xf_uid = p->p_ucred->cr_uid;
3951 FILEDESC_SLOCK(fdp);
3952 lastfile = fdlastfile(fdp);
3953 for (n = 0; fdp->fd_refcnt > 0 && n <= lastfile; ++n) {
3954 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3957 xf.xf_file = (uintptr_t)fp;
3958 xf.xf_data = (uintptr_t)fp->f_data;
3959 xf.xf_vnode = (uintptr_t)fp->f_vnode;
3960 xf.xf_type = (uintptr_t)fp->f_type;
3961 xf.xf_count = refcount_load(&fp->f_count);
3963 xf.xf_offset = foffset_get(fp);
3964 xf.xf_flag = fp->f_flag;
3965 error = SYSCTL_OUT(req, &xf, sizeof(xf));
3969 FILEDESC_SUNLOCK(fdp);
3974 sx_sunlock(&allproc_lock);
3978 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3979 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3981 #ifdef KINFO_FILE_SIZE
3982 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3986 xlate_fflags(int fflags)
3988 static const struct {
3991 } fflags_table[] = {
3992 { FAPPEND, KF_FLAG_APPEND },
3993 { FASYNC, KF_FLAG_ASYNC },
3994 { FFSYNC, KF_FLAG_FSYNC },
3995 { FHASLOCK, KF_FLAG_HASLOCK },
3996 { FNONBLOCK, KF_FLAG_NONBLOCK },
3997 { FREAD, KF_FLAG_READ },
3998 { FWRITE, KF_FLAG_WRITE },
3999 { O_CREAT, KF_FLAG_CREAT },
4000 { O_DIRECT, KF_FLAG_DIRECT },
4001 { O_EXCL, KF_FLAG_EXCL },
4002 { O_EXEC, KF_FLAG_EXEC },
4003 { O_EXLOCK, KF_FLAG_EXLOCK },
4004 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4005 { O_SHLOCK, KF_FLAG_SHLOCK },
4006 { O_TRUNC, KF_FLAG_TRUNC }
4012 for (i = 0; i < nitems(fflags_table); i++)
4013 if (fflags & fflags_table[i].fflag)
4014 kflags |= fflags_table[i].kf_fflag;
4018 /* Trim unused data from kf_path by truncating the structure size. */
4020 pack_kinfo(struct kinfo_file *kif)
4023 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4024 strlen(kif->kf_path) + 1;
4025 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4029 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4030 struct kinfo_file *kif, struct filedesc *fdp, int flags)
4034 bzero(kif, sizeof(*kif));
4036 /* Set a default type to allow for empty fill_kinfo() methods. */
4037 kif->kf_type = KF_TYPE_UNKNOWN;
4038 kif->kf_flags = xlate_fflags(fp->f_flag);
4039 if (rightsp != NULL)
4040 kif->kf_cap_rights = *rightsp;
4042 cap_rights_init_zero(&kif->kf_cap_rights);
4044 kif->kf_ref_count = refcount_load(&fp->f_count);
4045 kif->kf_offset = foffset_get(fp);
4048 * This may drop the filedesc lock, so the 'fp' cannot be
4049 * accessed after this call.
4051 error = fo_fill_kinfo(fp, kif, fdp);
4053 kif->kf_status |= KF_ATTR_VALID;
4054 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4057 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4061 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4062 struct kinfo_file *kif, int flags)
4066 bzero(kif, sizeof(*kif));
4068 kif->kf_type = KF_TYPE_VNODE;
4069 error = vn_fill_kinfo_vnode(vp, kif);
4071 kif->kf_status |= KF_ATTR_VALID;
4072 kif->kf_flags = xlate_fflags(fflags);
4073 cap_rights_init_zero(&kif->kf_cap_rights);
4075 kif->kf_ref_count = -1;
4076 kif->kf_offset = -1;
4077 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4080 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4084 struct export_fd_buf {
4085 struct filedesc *fdp;
4086 struct pwddesc *pdp;
4089 struct kinfo_file kif;
4094 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4096 struct kinfo_file *kif;
4099 if (efbuf->remainder != -1) {
4100 if (efbuf->remainder < kif->kf_structsize) {
4101 /* Terminate export. */
4102 efbuf->remainder = 0;
4105 efbuf->remainder -= kif->kf_structsize;
4107 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
4111 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4112 struct export_fd_buf *efbuf)
4116 if (efbuf->remainder == 0)
4118 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4120 FILEDESC_SUNLOCK(efbuf->fdp);
4121 error = export_kinfo_to_sb(efbuf);
4122 FILEDESC_SLOCK(efbuf->fdp);
4127 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4128 struct export_fd_buf *efbuf)
4132 if (efbuf->remainder == 0)
4134 if (efbuf->pdp != NULL)
4135 PWDDESC_XUNLOCK(efbuf->pdp);
4136 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4137 error = export_kinfo_to_sb(efbuf);
4138 if (efbuf->pdp != NULL)
4139 PWDDESC_XLOCK(efbuf->pdp);
4144 * Store a process file descriptor information to sbuf.
4146 * Takes a locked proc as argument, and returns with the proc unlocked.
4149 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
4153 struct filedesc *fdp;
4154 struct pwddesc *pdp;
4155 struct export_fd_buf *efbuf;
4156 struct vnode *cttyvp, *textvp, *tracevp;
4158 int error, i, lastfile;
4159 cap_rights_t rights;
4161 PROC_LOCK_ASSERT(p, MA_OWNED);
4164 tracevp = p->p_tracevp;
4165 if (tracevp != NULL)
4168 textvp = p->p_textvp;
4171 /* Controlling tty. */
4173 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4174 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4181 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4185 efbuf->remainder = maxlen;
4186 efbuf->flags = flags;
4187 if (tracevp != NULL)
4188 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
4191 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
4193 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
4196 if (pdp == NULL || fdp == NULL)
4201 pwd = pwd_hold_pwddesc(pdp);
4203 /* working directory */
4204 if (pwd->pwd_cdir != NULL) {
4205 vrefact(pwd->pwd_cdir);
4206 export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4208 /* root directory */
4209 if (pwd->pwd_rdir != NULL) {
4210 vrefact(pwd->pwd_rdir);
4211 export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
4213 /* jail directory */
4214 if (pwd->pwd_jdir != NULL) {
4215 vrefact(pwd->pwd_jdir);
4216 export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
4219 PWDDESC_XUNLOCK(pdp);
4222 FILEDESC_SLOCK(fdp);
4223 lastfile = fdlastfile(fdp);
4224 for (i = 0; fdp->fd_refcnt > 0 && i <= lastfile; i++) {
4225 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4228 rights = *cap_rights(fdp, i);
4229 #else /* !CAPABILITIES */
4230 rights = cap_no_rights;
4233 * Create sysctl entry. It is OK to drop the filedesc
4234 * lock inside of export_file_to_sb() as we will
4235 * re-validate and re-evaluate its properties when the
4238 error = export_file_to_sb(fp, i, &rights, efbuf);
4239 if (error != 0 || efbuf->remainder == 0)
4242 FILEDESC_SUNLOCK(fdp);
4248 free(efbuf, M_TEMP);
4252 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
4255 * Get per-process file descriptors for use by procstat(1), et al.
4258 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4263 int error, error2, *name;
4267 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4268 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4269 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4274 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4275 error = kern_proc_filedesc_out(p, &sb, maxlen,
4276 KERN_FILEDESC_PACK_KINFO);
4277 error2 = sbuf_finish(&sb);
4279 return (error != 0 ? error : error2);
4282 #ifdef COMPAT_FREEBSD7
4283 #ifdef KINFO_OFILE_SIZE
4284 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4288 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4291 okif->kf_structsize = sizeof(*okif);
4292 okif->kf_type = kif->kf_type;
4293 okif->kf_fd = kif->kf_fd;
4294 okif->kf_ref_count = kif->kf_ref_count;
4295 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4296 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4297 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4298 okif->kf_offset = kif->kf_offset;
4299 if (kif->kf_type == KF_TYPE_VNODE)
4300 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4302 okif->kf_vnode_type = KF_VTYPE_VNON;
4303 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4304 if (kif->kf_type == KF_TYPE_SOCKET) {
4305 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4306 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4307 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4308 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4309 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4311 okif->kf_sa_local.ss_family = AF_UNSPEC;
4312 okif->kf_sa_peer.ss_family = AF_UNSPEC;
4317 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4318 struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4323 PWDDESC_XUNLOCK(pdp);
4324 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4325 kinfo_to_okinfo(kif, okif);
4326 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4332 * Get per-process file descriptors for use by procstat(1), et al.
4335 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4337 struct kinfo_ofile *okif;
4338 struct kinfo_file *kif;
4339 struct filedesc *fdp;
4340 struct pwddesc *pdp;
4342 int error, i, lastfile, *name;
4347 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4354 if (fdp == NULL || pdp == NULL) {
4359 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4360 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4362 pwd = pwd_hold_pwddesc(pdp);
4364 if (pwd->pwd_cdir != NULL)
4365 export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4367 if (pwd->pwd_rdir != NULL)
4368 export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4370 if (pwd->pwd_jdir != NULL)
4371 export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4374 PWDDESC_XUNLOCK(pdp);
4377 FILEDESC_SLOCK(fdp);
4378 lastfile = fdlastfile(fdp);
4379 for (i = 0; fdp->fd_refcnt > 0 && i <= lastfile; i++) {
4380 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4382 export_file_to_kinfo(fp, i, NULL, kif, fdp,
4383 KERN_FILEDESC_PACK_KINFO);
4384 FILEDESC_SUNLOCK(fdp);
4385 kinfo_to_okinfo(kif, okif);
4386 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4387 FILEDESC_SLOCK(fdp);
4391 FILEDESC_SUNLOCK(fdp);
4399 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4400 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4401 "Process ofiledesc entries");
4402 #endif /* COMPAT_FREEBSD7 */
4405 vntype_to_kinfo(int vtype)
4410 } vtypes_table[] = {
4411 { VBAD, KF_VTYPE_VBAD },
4412 { VBLK, KF_VTYPE_VBLK },
4413 { VCHR, KF_VTYPE_VCHR },
4414 { VDIR, KF_VTYPE_VDIR },
4415 { VFIFO, KF_VTYPE_VFIFO },
4416 { VLNK, KF_VTYPE_VLNK },
4417 { VNON, KF_VTYPE_VNON },
4418 { VREG, KF_VTYPE_VREG },
4419 { VSOCK, KF_VTYPE_VSOCK }
4424 * Perform vtype translation.
4426 for (i = 0; i < nitems(vtypes_table); i++)
4427 if (vtypes_table[i].vtype == vtype)
4428 return (vtypes_table[i].kf_vtype);
4430 return (KF_VTYPE_UNKNOWN);
4433 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4434 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4435 "Process filedesc entries");
4438 * Store a process current working directory information to sbuf.
4440 * Takes a locked proc as argument, and returns with the proc unlocked.
4443 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
4445 struct pwddesc *pdp;
4447 struct export_fd_buf *efbuf;
4451 PROC_LOCK_ASSERT(p, MA_OWNED);
4458 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4461 efbuf->remainder = maxlen;
4464 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4465 cdir = pwd->pwd_cdir;
4470 error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4472 PWDDESC_XUNLOCK(pdp);
4474 free(efbuf, M_TEMP);
4479 * Get per-process current working directory.
4482 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4487 int error, error2, *name;
4491 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4492 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4493 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4498 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4499 error = kern_proc_cwd_out(p, &sb, maxlen);
4500 error2 = sbuf_finish(&sb);
4502 return (error != 0 ? error : error2);
4505 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4506 sysctl_kern_proc_cwd, "Process current working directory");
4510 * For the purposes of debugging, generate a human-readable string for the
4514 file_type_to_name(short type)
4542 case DTYPE_PROCDESC:
4544 case DTYPE_LINUXEFD:
4546 case DTYPE_LINUXTFD:
4554 * For the purposes of debugging, identify a process (if any, perhaps one of
4555 * many) that references the passed file in its file descriptor array. Return
4558 static struct proc *
4559 file_to_first_proc(struct file *fp)
4561 struct filedesc *fdp;
4565 FOREACH_PROC_IN_SYSTEM(p) {
4566 if (p->p_state == PRS_NEW)
4571 for (n = 0; n < fdp->fd_nfiles; n++) {
4572 if (fp == fdp->fd_ofiles[n].fde_file)
4580 db_print_file(struct file *fp, int header)
4582 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4586 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4587 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4588 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4590 p = file_to_first_proc(fp);
4591 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4592 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4593 fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4594 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4599 DB_SHOW_COMMAND(file, db_show_file)
4604 db_printf("usage: show file <addr>\n");
4607 fp = (struct file *)addr;
4608 db_print_file(fp, 1);
4611 DB_SHOW_COMMAND(files, db_show_files)
4613 struct filedesc *fdp;
4620 FOREACH_PROC_IN_SYSTEM(p) {
4621 if (p->p_state == PRS_NEW)
4623 if ((fdp = p->p_fd) == NULL)
4625 for (n = 0; n < fdp->fd_nfiles; ++n) {
4626 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4628 db_print_file(fp, header);
4635 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4636 &maxfilesperproc, 0, "Maximum files allowed open per process");
4638 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4639 &maxfiles, 0, "Maximum number of files");
4641 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4642 &openfiles, 0, "System-wide number of open files");
4646 filelistinit(void *dummy)
4649 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4650 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4651 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4652 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4653 pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4654 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4656 * XXXMJG this is a temporary hack due to boot ordering issues against
4659 vfs_smr = uma_zone_get_smr(pwd_zone);
4660 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4662 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4664 /*-------------------------------------------------------------------*/
4667 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4668 int flags, struct thread *td)
4675 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4683 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4691 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4699 badfo_kqfilter(struct file *fp, struct knote *kn)
4706 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4714 badfo_close(struct file *fp, struct thread *td)
4721 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4729 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4737 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4738 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4746 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4752 struct fileops badfileops = {
4753 .fo_read = badfo_readwrite,
4754 .fo_write = badfo_readwrite,
4755 .fo_truncate = badfo_truncate,
4756 .fo_ioctl = badfo_ioctl,
4757 .fo_poll = badfo_poll,
4758 .fo_kqfilter = badfo_kqfilter,
4759 .fo_stat = badfo_stat,
4760 .fo_close = badfo_close,
4761 .fo_chmod = badfo_chmod,
4762 .fo_chown = badfo_chown,
4763 .fo_sendfile = badfo_sendfile,
4764 .fo_fill_kinfo = badfo_fill_kinfo,
4768 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4769 int flags, struct thread *td)
4772 return (EOPNOTSUPP);
4776 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4784 invfo_ioctl(struct file *fp, u_long com, void *data,
4785 struct ucred *active_cred, struct thread *td)
4792 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4796 return (poll_no_poll(events));
4800 invfo_kqfilter(struct file *fp, struct knote *kn)
4807 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4815 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4823 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4824 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4831 /*-------------------------------------------------------------------*/
4834 * File Descriptor pseudo-device driver (/dev/fd/).
4836 * Opening minor device N dup()s the file (if any) connected to file
4837 * descriptor N belonging to the calling process. Note that this driver
4838 * consists of only the ``open()'' routine, because all subsequent
4839 * references to this file will be direct to the other driver.
4841 * XXX: we could give this one a cloning event handler if necessary.
4846 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4850 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4851 * the file descriptor being sought for duplication. The error
4852 * return ensures that the vnode for this device will be released
4853 * by vn_open. Open will detect this special error and take the
4854 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4855 * will simply report the error.
4857 td->td_dupfd = dev2unit(dev);
4861 static struct cdevsw fildesc_cdevsw = {
4862 .d_version = D_VERSION,
4868 fildesc_drvinit(void *unused)
4872 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4873 UID_ROOT, GID_WHEEL, 0666, "fd/0");
4874 make_dev_alias(dev, "stdin");
4875 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4876 UID_ROOT, GID_WHEEL, 0666, "fd/1");
4877 make_dev_alias(dev, "stdout");
4878 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4879 UID_ROOT, GID_WHEEL, 0666, "fd/2");
4880 make_dev_alias(dev, "stderr");
4883 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);