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>
67 #include <sys/protosw.h>
68 #include <sys/racct.h>
69 #include <sys/resourcevar.h>
71 #include <sys/signalvar.h>
76 #include <sys/syscallsubr.h>
77 #include <sys/sysctl.h>
78 #include <sys/sysproto.h>
79 #include <sys/unistd.h>
81 #include <sys/vnode.h>
82 #include <sys/ktrace.h>
86 #include <security/audit/audit.h>
93 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
94 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
95 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
96 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
97 "file desc to leader structures");
98 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
99 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
101 MALLOC_DECLARE(M_FADVISE);
103 static __read_mostly uma_zone_t file_zone;
104 static __read_mostly uma_zone_t filedesc0_zone;
105 __read_mostly uma_zone_t pwd_zone;
108 static int closefp(struct filedesc *fdp, int fd, struct file *fp,
109 struct thread *td, bool holdleaders, bool audit);
110 static int fd_first_free(struct filedesc *fdp, int low, int size);
111 static void fdgrowtable(struct filedesc *fdp, int nfd);
112 static void fdgrowtable_exp(struct filedesc *fdp, int nfd);
113 static void fdunused(struct filedesc *fdp, int fd);
114 static void fdused(struct filedesc *fdp, int fd);
115 static int getmaxfd(struct thread *td);
116 static u_long *filecaps_copy_prep(const struct filecaps *src);
117 static void filecaps_copy_finish(const struct filecaps *src,
118 struct filecaps *dst, u_long *ioctls);
119 static u_long *filecaps_free_prep(struct filecaps *fcaps);
120 static void filecaps_free_finish(u_long *ioctls);
122 static struct pwd *pwd_alloc(void);
127 * - An array of open file descriptors (fd_ofiles)
128 * - An array of file flags (fd_ofileflags)
129 * - A bitmap recording which descriptors are in use (fd_map)
131 * A process starts out with NDFILE descriptors. The value of NDFILE has
132 * been selected based the historical limit of 20 open files, and an
133 * assumption that the majority of processes, especially short-lived
134 * processes like shells, will never need more.
136 * If this initial allocation is exhausted, a larger descriptor table and
137 * map are allocated dynamically, and the pointers in the process's struct
138 * filedesc are updated to point to those. This is repeated every time
139 * the process runs out of file descriptors (provided it hasn't hit its
142 * Since threads may hold references to individual descriptor table
143 * entries, the tables are never freed. Instead, they are placed on a
144 * linked list and freed only when the struct filedesc is released.
147 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
148 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
149 #define NDSLOT(x) ((x) / NDENTRIES)
150 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
151 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
154 * SLIST entry used to keep track of ofiles which must be reclaimed when
158 struct fdescenttbl *ft_table;
159 SLIST_ENTRY(freetable) ft_next;
163 * Initial allocation: a filedesc structure + the head of SLIST used to
164 * keep track of old ofiles + enough space for NDFILE descriptors.
167 struct fdescenttbl0 {
169 struct filedescent fdt_ofiles[NDFILE];
173 struct filedesc fd_fd;
174 SLIST_HEAD(, freetable) fd_free;
175 struct fdescenttbl0 fd_dfiles;
176 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
180 * Descriptor management.
182 static int __exclusive_cache_line openfiles; /* actual number of open files */
183 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
184 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
187 * If low >= size, just return low. Otherwise find the first zero bit in the
188 * given bitmap, starting at low and not exceeding size - 1. Return size if
192 fd_first_free(struct filedesc *fdp, int low, int size)
194 NDSLOTTYPE *map = fdp->fd_map;
202 if (low % NDENTRIES) {
203 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
204 if ((mask &= ~map[off]) != 0UL)
205 return (off * NDENTRIES + ffsl(mask) - 1);
208 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
209 if (map[off] != ~0UL)
210 return (off * NDENTRIES + ffsl(~map[off]) - 1);
215 * Find the last used fd.
217 * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
218 * Otherwise use fdlastfile.
221 fdlastfile_single(struct filedesc *fdp)
223 NDSLOTTYPE *map = fdp->fd_map;
226 off = NDSLOT(fdp->fd_nfiles - 1);
227 for (minoff = NDSLOT(0); off >= minoff; --off)
229 return (off * NDENTRIES + flsl(map[off]) - 1);
234 fdlastfile(struct filedesc *fdp)
237 FILEDESC_LOCK_ASSERT(fdp);
238 return (fdlastfile_single(fdp));
242 fdisused(struct filedesc *fdp, int fd)
245 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
246 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
248 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
252 * Mark a file descriptor as used.
255 fdused_init(struct filedesc *fdp, int fd)
258 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
260 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
264 fdused(struct filedesc *fdp, int fd)
267 FILEDESC_XLOCK_ASSERT(fdp);
269 fdused_init(fdp, fd);
270 if (fd == fdp->fd_freefile)
275 * Mark a file descriptor as unused.
278 fdunused(struct filedesc *fdp, int fd)
281 FILEDESC_XLOCK_ASSERT(fdp);
283 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
284 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
285 ("fd=%d is still in use", fd));
287 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
288 if (fd < fdp->fd_freefile)
289 fdp->fd_freefile = fd;
293 * Free a file descriptor.
295 * Avoid some work if fdp is about to be destroyed.
298 fdefree_last(struct filedescent *fde)
301 filecaps_free(&fde->fde_caps);
305 fdfree(struct filedesc *fdp, int fd)
307 struct filedescent *fde;
309 FILEDESC_XLOCK_ASSERT(fdp);
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);
548 if (fp->f_ops == &path_fileops) {
554 tmp = flg = fp->f_flag;
556 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
557 } while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
558 tmp = fp->f_flag & FNONBLOCK;
559 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
564 tmp = fp->f_flag & FASYNC;
565 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
570 atomic_clear_int(&fp->f_flag, FNONBLOCK);
572 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
577 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
580 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
582 td->td_retval[0] = tmp;
587 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
591 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
596 error = priv_check(td, PRIV_NFS_LOCKD);
604 /* FALLTHROUGH F_SETLK */
608 flp = (struct flock *)arg;
609 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
614 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
617 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
623 if (flp->l_whence == SEEK_CUR) {
624 foffset = foffset_get(fp);
627 foffset > OFF_MAX - flp->l_start)) {
632 flp->l_start += foffset;
636 switch (flp->l_type) {
638 if ((fp->f_flag & FREAD) == 0) {
642 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
643 PROC_LOCK(p->p_leader);
644 p->p_leader->p_flag |= P_ADVLOCK;
645 PROC_UNLOCK(p->p_leader);
647 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
651 if ((fp->f_flag & FWRITE) == 0) {
655 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
656 PROC_LOCK(p->p_leader);
657 p->p_leader->p_flag |= P_ADVLOCK;
658 PROC_UNLOCK(p->p_leader);
660 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
664 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
668 if (flg != F_REMOTE) {
672 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
673 F_UNLCKSYS, flp, flg);
679 if (error != 0 || flp->l_type == F_UNLCK ||
680 flp->l_type == F_UNLCKSYS) {
686 * Check for a race with close.
688 * The vnode is now advisory locked (or unlocked, but this case
689 * is not really important) as the caller requested.
690 * We had to drop the filedesc lock, so we need to recheck if
691 * the descriptor is still valid, because if it was closed
692 * in the meantime we need to remove advisory lock from the
693 * vnode - close on any descriptor leading to an advisory
694 * locked vnode, removes that lock.
695 * We will return 0 on purpose in that case, as the result of
696 * successful advisory lock might have been externally visible
697 * already. This is fine - effectively we pretend to the caller
698 * that the closing thread was a bit slower and that the
699 * advisory lock succeeded before the close.
701 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
707 flp->l_whence = SEEK_SET;
710 flp->l_type = F_UNLCK;
711 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
712 F_UNLCK, flp, F_POSIX);
719 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
722 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
727 flp = (struct flock *)arg;
728 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
729 flp->l_type != F_UNLCK) {
734 if (flp->l_whence == SEEK_CUR) {
735 foffset = foffset_get(fp);
736 if ((flp->l_start > 0 &&
737 foffset > OFF_MAX - flp->l_start) ||
739 foffset < OFF_MIN - flp->l_start)) {
744 flp->l_start += foffset;
747 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
753 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
756 error = fo_add_seals(fp, arg);
761 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
764 if (fo_get_seals(fp, &seals) == 0)
765 td->td_retval[0] = seals;
772 arg = arg ? 128 * 1024: 0;
775 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
778 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
784 if (vp->v_type != VREG) {
791 * Exclusive lock synchronizes against f_seqcount reads and
792 * writes in sequential_heuristic().
794 error = vn_lock(vp, LK_EXCLUSIVE);
800 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
801 arg = MIN(arg, INT_MAX - bsize + 1);
802 fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
803 (arg + bsize - 1) / bsize);
804 atomic_set_int(&fp->f_flag, FRDAHEAD);
806 atomic_clear_int(&fp->f_flag, FRDAHEAD);
814 * Check if the vnode is part of a union stack (either the
815 * "union" flag from mount(2) or unionfs).
817 * Prior to introduction of this op libc's readdir would call
818 * fstatfs(2), in effect unnecessarily copying kilobytes of
819 * data just to check fs name and a mount flag.
821 * Fixing the code to handle everything in the kernel instead
822 * is a non-trivial endeavor and has low priority, thus this
823 * horrible kludge facilitates the current behavior in a much
824 * cheaper manner until someone(tm) sorts this out.
826 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
829 if (fp->f_type != DTYPE_VNODE) {
836 * Since we don't prevent dooming the vnode even non-null mp
837 * found can become immediately stale. This is tolerable since
838 * mount points are type-stable (providing safe memory access)
839 * and any vfs op on this vnode going forward will return an
840 * error (meaning return value in this case is meaningless).
842 mp = atomic_load_ptr(&vp->v_mount);
843 if (__predict_false(mp == NULL)) {
848 td->td_retval[0] = 0;
849 if (mp->mnt_kern_flag & MNTK_UNIONFS ||
850 mp->mnt_flag & MNT_UNION)
851 td->td_retval[0] = 1;
863 getmaxfd(struct thread *td)
866 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
870 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
873 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
875 struct filedesc *fdp;
876 struct filedescent *oldfde, *newfde;
878 struct file *delfp, *oldfp;
879 u_long *oioctls, *nioctls;
886 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
887 MPASS(mode < FDDUP_LASTMODE);
890 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
893 * Verify we have a valid descriptor to dup from and possibly to
894 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
895 * return EINVAL when the new descriptor is out of bounds.
900 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
901 maxfd = getmaxfd(td);
903 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
907 if (fget_locked(fdp, old) == NULL)
909 if (mode == FDDUP_FIXED && old == new) {
910 td->td_retval[0] = new;
911 if (flags & FDDUP_FLAG_CLOEXEC)
912 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
917 oldfde = &fdp->fd_ofiles[old];
918 oldfp = oldfde->fde_file;
923 * If the caller specified a file descriptor, make sure the file
924 * table is large enough to hold it, and grab it. Otherwise, just
925 * allocate a new descriptor the usual way.
930 if ((error = fdalloc(td, new, &new)) != 0) {
936 if (new >= fdp->fd_nfiles) {
938 * The resource limits are here instead of e.g.
939 * fdalloc(), because the file descriptor table may be
940 * shared between processes, so we can't really use
941 * racct_add()/racct_sub(). Instead of counting the
942 * number of actually allocated descriptors, just put
943 * the limit on the size of the file descriptor table.
946 if (RACCT_ENABLED()) {
947 error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
955 fdgrowtable_exp(fdp, new + 1);
957 if (!fdisused(fdp, new))
961 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
964 KASSERT(old != new, ("new fd is same as old"));
966 /* Refetch oldfde because the table may have grown and old one freed. */
967 oldfde = &fdp->fd_ofiles[old];
968 KASSERT(oldfp == oldfde->fde_file,
969 ("fdt_ofiles shift from growth observed at fd %d",
972 newfde = &fdp->fd_ofiles[new];
973 delfp = newfde->fde_file;
975 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
978 * Duplicate the source descriptor.
981 seqc_write_begin(&newfde->fde_seqc);
983 oioctls = filecaps_free_prep(&newfde->fde_caps);
984 memcpy(newfde, oldfde, fde_change_size);
985 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
987 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
988 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
990 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
992 seqc_write_end(&newfde->fde_seqc);
994 td->td_retval[0] = new;
999 (void) closefp(fdp, new, delfp, td, true, false);
1000 FILEDESC_UNLOCK_ASSERT(fdp);
1003 FILEDESC_XUNLOCK(fdp);
1006 filecaps_free_finish(oioctls);
1011 sigiofree(struct sigio *sigio)
1013 crfree(sigio->sio_ucred);
1014 free(sigio, M_SIGIO);
1017 static struct sigio *
1018 funsetown_locked(struct sigio *sigio)
1023 SIGIO_ASSERT_LOCKED();
1027 *sigio->sio_myref = NULL;
1028 if (sigio->sio_pgid < 0) {
1029 pg = sigio->sio_pgrp;
1031 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio);
1034 p = sigio->sio_proc;
1036 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio);
1043 * If sigio is on the list associated with a process or process group,
1044 * disable signalling from the device, remove sigio from the list and
1048 funsetown(struct sigio **sigiop)
1050 struct sigio *sigio;
1052 /* Racy check, consumers must provide synchronization. */
1053 if (*sigiop == NULL)
1057 sigio = funsetown_locked(*sigiop);
1064 * Free a list of sigio structures. The caller must ensure that new sigio
1065 * structures cannot be added after this point. For process groups this is
1066 * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1070 funsetownlst(struct sigiolst *sigiolst)
1074 struct sigio *sigio, *tmp;
1077 sigio = SLIST_FIRST(sigiolst);
1085 sigio = SLIST_FIRST(sigiolst);
1086 if (sigio == NULL) {
1092 * Every entry of the list should belong to a single proc or pgrp.
1094 if (sigio->sio_pgid < 0) {
1095 pg = sigio->sio_pgrp;
1096 sx_assert(&proctree_lock, SX_XLOCKED);
1098 } else /* if (sigio->sio_pgid > 0) */ {
1099 p = sigio->sio_proc;
1101 KASSERT((p->p_flag & P_WEXIT) != 0,
1102 ("%s: process %p is not exiting", __func__, p));
1105 SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1106 *sigio->sio_myref = NULL;
1108 KASSERT(sigio->sio_pgid < 0,
1109 ("Proc sigio in pgrp sigio list"));
1110 KASSERT(sigio->sio_pgrp == pg,
1111 ("Bogus pgrp in sigio list"));
1112 } else /* if (p != NULL) */ {
1113 KASSERT(sigio->sio_pgid > 0,
1114 ("Pgrp sigio in proc sigio list"));
1115 KASSERT(sigio->sio_proc == p,
1116 ("Bogus proc in sigio list"));
1126 SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1131 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1133 * After permission checking, add a sigio structure to the sigio list for
1134 * the process or process group.
1137 fsetown(pid_t pgid, struct sigio **sigiop)
1141 struct sigio *osigio, *sigio;
1149 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1150 sigio->sio_pgid = pgid;
1151 sigio->sio_ucred = crhold(curthread->td_ucred);
1152 sigio->sio_myref = sigiop;
1156 ret = pget(pgid, PGET_NOTWEXIT | PGET_NOTID | PGET_HOLD, &proc);
1158 osigio = funsetown_locked(*sigiop);
1162 if ((proc->p_flag & P_WEXIT) != 0) {
1164 } else if (proc->p_session !=
1165 curthread->td_proc->p_session) {
1167 * Policy - Don't allow a process to FSETOWN a
1168 * process in another session.
1170 * Remove this test to allow maximum flexibility
1171 * or restrict FSETOWN to the current process or
1172 * process group for maximum safety.
1176 sigio->sio_proc = proc;
1177 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio,
1182 } else /* if (pgid < 0) */ {
1183 sx_slock(&proctree_lock);
1185 osigio = funsetown_locked(*sigiop);
1186 pgrp = pgfind(-pgid);
1190 if (pgrp->pg_session != curthread->td_proc->p_session) {
1192 * Policy - Don't allow a process to FSETOWN a
1193 * process in another session.
1195 * Remove this test to allow maximum flexibility
1196 * or restrict FSETOWN to the current process or
1197 * process group for maximum safety.
1201 sigio->sio_pgrp = pgrp;
1202 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio,
1207 sx_sunlock(&proctree_lock);
1218 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1221 fgetown(struct sigio **sigiop)
1226 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1232 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1237 FILEDESC_XLOCK_ASSERT(fdp);
1240 * We now hold the fp reference that used to be owned by the
1241 * descriptor array. We have to unlock the FILEDESC *AFTER*
1242 * knote_fdclose to prevent a race of the fd getting opened, a knote
1243 * added, and deleteing a knote for the new fd.
1245 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1246 knote_fdclose(td, fd);
1249 * We need to notify mqueue if the object is of type mqueue.
1251 if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1252 mq_fdclose(td, fd, fp);
1253 FILEDESC_XUNLOCK(fdp);
1256 if (AUDITING_TD(td) && audit)
1257 audit_sysclose(td, fd, fp);
1259 error = closef(fp, td);
1262 * All paths leading up to closefp() will have already removed or
1263 * replaced the fd in the filedesc table, so a restart would not
1264 * operate on the same file.
1266 if (error == ERESTART)
1273 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1274 bool holdleaders, bool audit)
1278 FILEDESC_XLOCK_ASSERT(fdp);
1281 if (td->td_proc->p_fdtol != NULL) {
1283 * Ask fdfree() to sleep to ensure that all relevant
1284 * process leaders can be traversed in closef().
1286 fdp->fd_holdleaderscount++;
1288 holdleaders = false;
1292 error = closefp_impl(fdp, fd, fp, td, audit);
1294 FILEDESC_XLOCK(fdp);
1295 fdp->fd_holdleaderscount--;
1296 if (fdp->fd_holdleaderscount == 0 &&
1297 fdp->fd_holdleaderswakeup != 0) {
1298 fdp->fd_holdleaderswakeup = 0;
1299 wakeup(&fdp->fd_holdleaderscount);
1301 FILEDESC_XUNLOCK(fdp);
1307 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1308 bool holdleaders, bool audit)
1311 FILEDESC_XLOCK_ASSERT(fdp);
1313 if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1314 return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1316 return (closefp_impl(fdp, fd, fp, td, audit));
1321 * Close a file descriptor.
1323 #ifndef _SYS_SYSPROTO_H_
1330 sys_close(struct thread *td, struct close_args *uap)
1333 return (kern_close(td, uap->fd));
1337 kern_close(struct thread *td, int fd)
1339 struct filedesc *fdp;
1342 fdp = td->td_proc->p_fd;
1344 FILEDESC_XLOCK(fdp);
1345 if ((fp = fget_locked(fdp, fd)) == NULL) {
1346 FILEDESC_XUNLOCK(fdp);
1351 /* closefp() drops the FILEDESC lock for us. */
1352 return (closefp(fdp, fd, fp, td, true, true));
1356 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1358 struct filedesc *fdp;
1359 const struct fdescenttbl *fdt;
1364 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1365 * open should not be a usage error. From a close_range() perspective,
1366 * close_range(3, ~0U, 0) in the same scenario should also likely not
1367 * be a usage error as all fd above 3 are in-fact already closed.
1369 if (highfd < lowfd) {
1373 fdp = td->td_proc->p_fd;
1374 FILEDESC_XLOCK(fdp);
1375 fdt = atomic_load_ptr(&fdp->fd_files);
1376 highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1378 if (__predict_false(fd > highfd)) {
1382 fp = fdt->fdt_ofiles[fd].fde_file;
1388 (void) closefp(fdp, fd, fp, td, true, true);
1391 FILEDESC_XLOCK(fdp);
1392 fdt = atomic_load_ptr(&fdp->fd_files);
1397 FILEDESC_XUNLOCK(fdp);
1402 #ifndef _SYS_SYSPROTO_H_
1403 struct close_range_args {
1410 sys_close_range(struct thread *td, struct close_range_args *uap)
1413 AUDIT_ARG_FD(uap->lowfd);
1414 AUDIT_ARG_CMD(uap->highfd);
1415 AUDIT_ARG_FFLAGS(uap->flags);
1417 /* No flags currently defined */
1418 if (uap->flags != 0)
1420 return (kern_close_range(td, uap->lowfd, uap->highfd));
1423 #ifdef COMPAT_FREEBSD12
1425 * Close open file descriptors.
1427 #ifndef _SYS_SYSPROTO_H_
1428 struct freebsd12_closefrom_args {
1434 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1438 AUDIT_ARG_FD(uap->lowfd);
1441 * Treat negative starting file descriptor values identical to
1442 * closefrom(0) which closes all files.
1444 lowfd = MAX(0, uap->lowfd);
1445 return (kern_close_range(td, lowfd, ~0U));
1447 #endif /* COMPAT_FREEBSD12 */
1449 #if defined(COMPAT_43)
1451 * Return status information about a file descriptor.
1453 #ifndef _SYS_SYSPROTO_H_
1454 struct ofstat_args {
1461 ofstat(struct thread *td, struct ofstat_args *uap)
1467 error = kern_fstat(td, uap->fd, &ub);
1470 error = copyout(&oub, uap->sb, sizeof(oub));
1474 #endif /* COMPAT_43 */
1476 #if defined(COMPAT_FREEBSD11)
1478 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1481 struct freebsd11_stat osb;
1484 error = kern_fstat(td, uap->fd, &sb);
1487 error = freebsd11_cvtstat(&sb, &osb);
1489 error = copyout(&osb, uap->sb, sizeof(osb));
1492 #endif /* COMPAT_FREEBSD11 */
1495 * Return status information about a file descriptor.
1497 #ifndef _SYS_SYSPROTO_H_
1505 sys_fstat(struct thread *td, struct fstat_args *uap)
1510 error = kern_fstat(td, uap->fd, &ub);
1512 error = copyout(&ub, uap->sb, sizeof(ub));
1517 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1524 error = fget(td, fd, &cap_fstat_rights, &fp);
1525 if (__predict_false(error != 0))
1528 AUDIT_ARG_FILE(td->td_proc, fp);
1530 error = fo_stat(fp, sbp, td->td_ucred);
1532 #ifdef __STAT_TIME_T_EXT
1533 sbp->st_atim_ext = 0;
1534 sbp->st_mtim_ext = 0;
1535 sbp->st_ctim_ext = 0;
1536 sbp->st_btim_ext = 0;
1539 if (KTRPOINT(td, KTR_STRUCT))
1540 ktrstat_error(sbp, error);
1545 #if defined(COMPAT_FREEBSD11)
1547 * Return status information about a file descriptor.
1549 #ifndef _SYS_SYSPROTO_H_
1550 struct freebsd11_nfstat_args {
1557 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1563 error = kern_fstat(td, uap->fd, &ub);
1565 freebsd11_cvtnstat(&ub, &nub);
1566 error = copyout(&nub, uap->sb, sizeof(nub));
1570 #endif /* COMPAT_FREEBSD11 */
1573 * Return pathconf information about a file descriptor.
1575 #ifndef _SYS_SYSPROTO_H_
1576 struct fpathconf_args {
1583 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1588 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1590 td->td_retval[0] = value;
1595 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1601 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1605 if (name == _PC_ASYNC_IO) {
1606 *valuep = _POSIX_ASYNCHRONOUS_IO;
1611 vn_lock(vp, LK_SHARED | LK_RETRY);
1612 error = VOP_PATHCONF(vp, name, valuep);
1614 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1615 if (name != _PC_PIPE_BUF) {
1630 * Copy filecaps structure allocating memory for ioctls array if needed.
1632 * The last parameter indicates whether the fdtable is locked. If it is not and
1633 * ioctls are encountered, copying fails and the caller must lock the table.
1635 * Note that if the table was not locked, the caller has to check the relevant
1636 * sequence counter to determine whether the operation was successful.
1639 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1643 if (src->fc_ioctls != NULL && !locked)
1645 memcpy(dst, src, sizeof(*src));
1646 if (src->fc_ioctls == NULL)
1649 KASSERT(src->fc_nioctls > 0,
1650 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1652 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1653 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1654 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1659 filecaps_copy_prep(const struct filecaps *src)
1664 if (__predict_true(src->fc_ioctls == NULL))
1667 KASSERT(src->fc_nioctls > 0,
1668 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1670 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1671 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1676 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1682 if (__predict_true(src->fc_ioctls == NULL)) {
1683 MPASS(ioctls == NULL);
1687 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1688 dst->fc_ioctls = ioctls;
1689 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1693 * Move filecaps structure to the new place and clear the old place.
1696 filecaps_move(struct filecaps *src, struct filecaps *dst)
1700 bzero(src, sizeof(*src));
1704 * Fill the given filecaps structure with full rights.
1707 filecaps_fill(struct filecaps *fcaps)
1710 CAP_ALL(&fcaps->fc_rights);
1711 fcaps->fc_ioctls = NULL;
1712 fcaps->fc_nioctls = -1;
1713 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1717 * Free memory allocated within filecaps structure.
1720 filecaps_free(struct filecaps *fcaps)
1723 free(fcaps->fc_ioctls, M_FILECAPS);
1724 bzero(fcaps, sizeof(*fcaps));
1728 filecaps_free_prep(struct filecaps *fcaps)
1732 ioctls = fcaps->fc_ioctls;
1733 bzero(fcaps, sizeof(*fcaps));
1738 filecaps_free_finish(u_long *ioctls)
1741 free(ioctls, M_FILECAPS);
1745 * Validate the given filecaps structure.
1748 filecaps_validate(const struct filecaps *fcaps, const char *func)
1751 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1752 ("%s: invalid rights", func));
1753 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1754 ("%s: invalid fcntls", func));
1755 KASSERT(fcaps->fc_fcntls == 0 ||
1756 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1757 ("%s: fcntls without CAP_FCNTL", func));
1758 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1759 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1760 ("%s: invalid ioctls", func));
1761 KASSERT(fcaps->fc_nioctls == 0 ||
1762 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1763 ("%s: ioctls without CAP_IOCTL", func));
1767 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1771 FILEDESC_XLOCK_ASSERT(fdp);
1773 nfd1 = fdp->fd_nfiles * 2;
1776 fdgrowtable(fdp, nfd1);
1780 * Grow the file table to accommodate (at least) nfd descriptors.
1783 fdgrowtable(struct filedesc *fdp, int nfd)
1785 struct filedesc0 *fdp0;
1786 struct freetable *ft;
1787 struct fdescenttbl *ntable;
1788 struct fdescenttbl *otable;
1789 int nnfiles, onfiles;
1790 NDSLOTTYPE *nmap, *omap;
1792 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1794 /* save old values */
1795 onfiles = fdp->fd_nfiles;
1796 otable = fdp->fd_files;
1799 /* compute the size of the new table */
1800 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1801 if (nnfiles <= onfiles)
1802 /* the table is already large enough */
1806 * Allocate a new table. We need enough space for the number of
1807 * entries, file entries themselves and the struct freetable we will use
1808 * when we decommission the table and place it on the freelist.
1809 * We place the struct freetable in the middle so we don't have
1810 * to worry about padding.
1812 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1813 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1814 sizeof(struct freetable),
1815 M_FILEDESC, M_ZERO | M_WAITOK);
1816 /* copy the old data */
1817 ntable->fdt_nfiles = nnfiles;
1818 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1819 onfiles * sizeof(ntable->fdt_ofiles[0]));
1822 * Allocate a new map only if the old is not large enough. It will
1823 * grow at a slower rate than the table as it can map more
1824 * entries than the table can hold.
1826 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1827 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1829 /* copy over the old data and update the pointer */
1830 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1835 * Make sure that ntable is correctly initialized before we replace
1836 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1839 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1842 * Free the old file table when not shared by other threads or processes.
1843 * The old file table is considered to be shared when either are true:
1844 * - The process has more than one thread.
1845 * - The file descriptor table has been shared via fdshare().
1847 * When shared, the old file table will be placed on a freelist
1848 * which will be processed when the struct filedesc is released.
1850 * Note that if onfiles == NDFILE, we're dealing with the original
1851 * static allocation contained within (struct filedesc0 *)fdp,
1852 * which must not be freed.
1854 if (onfiles > NDFILE) {
1856 * Note we may be called here from fdinit while allocating a
1857 * table for a new process in which case ->p_fd points
1860 if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) {
1861 free(otable, M_FILEDESC);
1863 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1864 fdp0 = (struct filedesc0 *)fdp;
1865 ft->ft_table = otable;
1866 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1870 * The map does not have the same possibility of threads still
1871 * holding references to it. So always free it as long as it
1872 * does not reference the original static allocation.
1874 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1875 free(omap, M_FILEDESC);
1879 * Allocate a file descriptor for the process.
1882 fdalloc(struct thread *td, int minfd, int *result)
1884 struct proc *p = td->td_proc;
1885 struct filedesc *fdp = p->p_fd;
1886 int fd, maxfd, allocfd;
1891 FILEDESC_XLOCK_ASSERT(fdp);
1893 if (fdp->fd_freefile > minfd)
1894 minfd = fdp->fd_freefile;
1896 maxfd = getmaxfd(td);
1899 * Search the bitmap for a free descriptor starting at minfd.
1900 * If none is found, grow the file table.
1902 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1903 if (__predict_false(fd >= maxfd))
1905 if (__predict_false(fd >= fdp->fd_nfiles)) {
1906 allocfd = min(fd * 2, maxfd);
1908 if (RACCT_ENABLED()) {
1909 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1915 * fd is already equal to first free descriptor >= minfd, so
1916 * we only need to grow the table and we are done.
1918 fdgrowtable_exp(fdp, allocfd);
1922 * Perform some sanity checks, then mark the file descriptor as
1923 * used and return it to the caller.
1925 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1926 ("invalid descriptor %d", fd));
1927 KASSERT(!fdisused(fdp, fd),
1928 ("fd_first_free() returned non-free descriptor"));
1929 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1930 ("file descriptor isn't free"));
1937 * Allocate n file descriptors for the process.
1940 fdallocn(struct thread *td, int minfd, int *fds, int n)
1942 struct proc *p = td->td_proc;
1943 struct filedesc *fdp = p->p_fd;
1946 FILEDESC_XLOCK_ASSERT(fdp);
1948 for (i = 0; i < n; i++)
1949 if (fdalloc(td, 0, &fds[i]) != 0)
1953 for (i--; i >= 0; i--)
1954 fdunused(fdp, fds[i]);
1962 * Create a new open file structure and allocate a file descriptor for the
1963 * process that refers to it. We add one reference to the file for the
1964 * descriptor table and one reference for resultfp. This is to prevent us
1965 * being preempted and the entry in the descriptor table closed after we
1966 * release the FILEDESC lock.
1969 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1970 struct filecaps *fcaps)
1975 MPASS(resultfp != NULL);
1976 MPASS(resultfd != NULL);
1978 error = _falloc_noinstall(td, &fp, 2);
1979 if (__predict_false(error != 0)) {
1983 error = finstall_refed(td, fp, &fd, flags, fcaps);
1984 if (__predict_false(error != 0)) {
1985 falloc_abort(td, fp);
1996 * Create a new open file structure without allocating a file descriptor.
1999 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2002 int maxuserfiles = maxfiles - (maxfiles / 20);
2004 static struct timeval lastfail;
2007 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2010 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2011 if ((openfiles_new >= maxuserfiles &&
2012 priv_check(td, PRIV_MAXFILES) != 0) ||
2013 openfiles_new >= maxfiles) {
2014 atomic_subtract_int(&openfiles, 1);
2015 if (ppsratecheck(&lastfail, &curfail, 1)) {
2016 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2017 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2021 fp = uma_zalloc(file_zone, M_WAITOK);
2022 bzero(fp, sizeof(*fp));
2023 refcount_init(&fp->f_count, n);
2024 fp->f_cred = crhold(td->td_ucred);
2025 fp->f_ops = &badfileops;
2031 falloc_abort(struct thread *td, struct file *fp)
2035 * For assertion purposes.
2037 refcount_init(&fp->f_count, 0);
2042 * Install a file in a file descriptor table.
2045 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2046 struct filecaps *fcaps)
2048 struct filedescent *fde;
2052 filecaps_validate(fcaps, __func__);
2053 FILEDESC_XLOCK_ASSERT(fdp);
2055 fde = &fdp->fd_ofiles[fd];
2057 seqc_write_begin(&fde->fde_seqc);
2060 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2062 filecaps_move(fcaps, &fde->fde_caps);
2064 filecaps_fill(&fde->fde_caps);
2066 seqc_write_end(&fde->fde_seqc);
2071 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2072 struct filecaps *fcaps)
2074 struct filedesc *fdp = td->td_proc->p_fd;
2079 FILEDESC_XLOCK(fdp);
2080 error = fdalloc(td, 0, fd);
2081 if (__predict_true(error == 0)) {
2082 _finstall(fdp, fp, *fd, flags, fcaps);
2084 FILEDESC_XUNLOCK(fdp);
2089 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2090 struct filecaps *fcaps)
2098 error = finstall_refed(td, fp, fd, flags, fcaps);
2099 if (__predict_false(error != 0)) {
2106 * Build a new filedesc structure from another.
2108 * If fdp is not NULL, return with it shared locked.
2111 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2113 struct filedesc0 *newfdp0;
2114 struct filedesc *newfdp;
2117 MPASS(lastfile != NULL);
2119 MPASS(lastfile == NULL);
2121 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2122 newfdp = &newfdp0->fd_fd;
2124 /* Create the file descriptor table. */
2125 FILEDESC_LOCK_INIT(newfdp);
2126 refcount_init(&newfdp->fd_refcnt, 1);
2127 refcount_init(&newfdp->fd_holdcnt, 1);
2128 newfdp->fd_map = newfdp0->fd_dmap;
2129 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2130 newfdp->fd_files->fdt_nfiles = NDFILE;
2135 FILEDESC_SLOCK(fdp);
2137 FILEDESC_SUNLOCK(fdp);
2142 *lastfile = fdlastfile(fdp);
2143 if (*lastfile < newfdp->fd_nfiles)
2145 FILEDESC_SUNLOCK(fdp);
2146 fdgrowtable(newfdp, *lastfile + 1);
2147 FILEDESC_SLOCK(fdp);
2154 * Build a pwddesc structure from another.
2155 * Copy the current, root, and jail root vnode references.
2157 * If pdp is not NULL, return with it shared locked.
2160 pdinit(struct pwddesc *pdp, bool keeplock)
2162 struct pwddesc *newpdp;
2165 newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2167 PWDDESC_LOCK_INIT(newpdp);
2168 refcount_init(&newpdp->pd_refcount, 1);
2169 newpdp->pd_cmask = CMASK;
2172 newpwd = pwd_alloc();
2173 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2178 newpwd = pwd_hold_pwddesc(pdp);
2179 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2181 PWDDESC_XUNLOCK(pdp);
2186 * Hold either filedesc or pwddesc of the passed process.
2188 * The process lock is used to synchronize against the target exiting and
2191 * Clearing can be ilustrated in 3 steps:
2192 * 1. set the pointer to NULL. Either routine can race against it, hence
2194 * 2. observe the process lock as not taken. Until then fdhold/pdhold can
2195 * race to either still see the pointer or find NULL. It is still safe to
2196 * grab a reference as clearing is stalled.
2197 * 3. after the lock is observed as not taken, any fdhold/pdhold calls are
2198 * guaranteed to see NULL, making it safe to finish clearing
2200 static struct filedesc *
2201 fdhold(struct proc *p)
2203 struct filedesc *fdp;
2205 PROC_LOCK_ASSERT(p, MA_OWNED);
2206 fdp = atomic_load_ptr(&p->p_fd);
2208 refcount_acquire(&fdp->fd_holdcnt);
2212 static struct pwddesc *
2213 pdhold(struct proc *p)
2215 struct pwddesc *pdp;
2217 PROC_LOCK_ASSERT(p, MA_OWNED);
2218 pdp = atomic_load_ptr(&p->p_pd);
2220 refcount_acquire(&pdp->pd_refcount);
2225 fddrop(struct filedesc *fdp)
2228 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2229 if (refcount_release(&fdp->fd_holdcnt) == 0)
2233 FILEDESC_LOCK_DESTROY(fdp);
2234 uma_zfree(filedesc0_zone, fdp);
2238 pddrop(struct pwddesc *pdp)
2242 if (refcount_release_if_not_last(&pdp->pd_refcount))
2246 if (refcount_release(&pdp->pd_refcount) == 0) {
2247 PWDDESC_XUNLOCK(pdp);
2250 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2252 PWDDESC_XUNLOCK(pdp);
2255 PWDDESC_LOCK_DESTROY(pdp);
2256 free(pdp, M_PWDDESC);
2260 * Share a filedesc structure.
2263 fdshare(struct filedesc *fdp)
2266 refcount_acquire(&fdp->fd_refcnt);
2271 * Share a pwddesc structure.
2274 pdshare(struct pwddesc *pdp)
2276 refcount_acquire(&pdp->pd_refcount);
2281 * Unshare a filedesc structure, if necessary by making a copy
2284 fdunshare(struct thread *td)
2286 struct filedesc *tmp;
2287 struct proc *p = td->td_proc;
2289 if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2292 tmp = fdcopy(p->p_fd);
2298 * Unshare a pwddesc structure.
2301 pdunshare(struct thread *td)
2303 struct pwddesc *pdp;
2308 if (p->p_pd->pd_refcount == 1)
2311 pdp = pdcopy(p->p_pd);
2317 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2318 * this is to ease callers, not catch errors.
2321 fdcopy(struct filedesc *fdp)
2323 struct filedesc *newfdp;
2324 struct filedescent *nfde, *ofde;
2329 newfdp = fdinit(fdp, true, &lastfile);
2330 /* copy all passable descriptors (i.e. not kqueue) */
2331 newfdp->fd_freefile = -1;
2332 for (i = 0; i <= lastfile; ++i) {
2333 ofde = &fdp->fd_ofiles[i];
2334 if (ofde->fde_file == NULL ||
2335 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2336 !fhold(ofde->fde_file)) {
2337 if (newfdp->fd_freefile == -1)
2338 newfdp->fd_freefile = i;
2341 nfde = &newfdp->fd_ofiles[i];
2343 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2344 fdused_init(newfdp, i);
2346 if (newfdp->fd_freefile == -1)
2347 newfdp->fd_freefile = i;
2348 FILEDESC_SUNLOCK(fdp);
2353 * Copy a pwddesc structure.
2356 pdcopy(struct pwddesc *pdp)
2358 struct pwddesc *newpdp;
2362 newpdp = pdinit(pdp, true);
2363 newpdp->pd_cmask = pdp->pd_cmask;
2364 PWDDESC_XUNLOCK(pdp);
2369 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2370 * one of processes using it exits) and the table used to be shared.
2373 fdclearlocks(struct thread *td)
2375 struct filedesc *fdp;
2376 struct filedesc_to_leader *fdtol;
2386 MPASS(fdtol != NULL);
2388 FILEDESC_XLOCK(fdp);
2389 KASSERT(fdtol->fdl_refcount > 0,
2390 ("filedesc_to_refcount botch: fdl_refcount=%d",
2391 fdtol->fdl_refcount));
2392 if (fdtol->fdl_refcount == 1 &&
2393 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2394 lastfile = fdlastfile(fdp);
2395 for (i = 0; i <= lastfile; i++) {
2396 fp = fdp->fd_ofiles[i].fde_file;
2397 if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2400 FILEDESC_XUNLOCK(fdp);
2401 lf.l_whence = SEEK_SET;
2404 lf.l_type = F_UNLCK;
2406 (void) VOP_ADVLOCK(vp,
2407 (caddr_t)p->p_leader, F_UNLCK,
2409 FILEDESC_XLOCK(fdp);
2414 if (fdtol->fdl_refcount == 1) {
2415 if (fdp->fd_holdleaderscount > 0 &&
2416 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2418 * close() or kern_dup() has cleared a reference
2419 * in a shared file descriptor table.
2421 fdp->fd_holdleaderswakeup = 1;
2422 sx_sleep(&fdp->fd_holdleaderscount,
2423 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2426 if (fdtol->fdl_holdcount > 0) {
2428 * Ensure that fdtol->fdl_leader remains
2429 * valid in closef().
2431 fdtol->fdl_wakeup = 1;
2432 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2437 fdtol->fdl_refcount--;
2438 if (fdtol->fdl_refcount == 0 &&
2439 fdtol->fdl_holdcount == 0) {
2440 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2441 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2445 FILEDESC_XUNLOCK(fdp);
2447 free(fdtol, M_FILEDESC_TO_LEADER);
2451 * Release a filedesc structure.
2454 fdescfree_fds(struct thread *td, struct filedesc *fdp)
2456 struct filedesc0 *fdp0;
2457 struct freetable *ft, *tft;
2458 struct filedescent *fde;
2462 KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2463 ("%s: fd table %p carries references", __func__, fdp));
2466 * Serialize with threads iterating over the table, if any.
2468 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2469 FILEDESC_XLOCK(fdp);
2470 FILEDESC_XUNLOCK(fdp);
2473 lastfile = fdlastfile_single(fdp);
2474 for (i = 0; i <= lastfile; i++) {
2475 fde = &fdp->fd_ofiles[i];
2479 (void) closef(fp, td);
2483 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2484 free(fdp->fd_map, M_FILEDESC);
2485 if (fdp->fd_nfiles > NDFILE)
2486 free(fdp->fd_files, M_FILEDESC);
2488 fdp0 = (struct filedesc0 *)fdp;
2489 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2490 free(ft->ft_table, M_FILEDESC);
2496 fdescfree(struct thread *td)
2499 struct filedesc *fdp;
2506 if (RACCT_ENABLED())
2507 racct_set_unlocked(p, RACCT_NOFILE, 0);
2510 if (p->p_fdtol != NULL)
2514 * Check fdhold for an explanation.
2516 atomic_store_ptr(&p->p_fd, NULL);
2517 atomic_thread_fence_seq_cst();
2518 PROC_WAIT_UNLOCKED(p);
2520 if (refcount_release(&fdp->fd_refcnt) == 0)
2523 fdescfree_fds(td, fdp);
2527 pdescfree(struct thread *td)
2530 struct pwddesc *pdp;
2537 * Check pdhold for an explanation.
2539 atomic_store_ptr(&p->p_pd, NULL);
2540 atomic_thread_fence_seq_cst();
2541 PROC_WAIT_UNLOCKED(p);
2547 * For setugid programs, we don't want to people to use that setugidness
2548 * to generate error messages which write to a file which otherwise would
2549 * otherwise be off-limits to the process. We check for filesystems where
2550 * the vnode can change out from under us after execve (like [lin]procfs).
2552 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2553 * sufficient. We also don't check for setugidness since we know we are.
2556 is_unsafe(struct file *fp)
2560 if (fp->f_type != DTYPE_VNODE)
2564 return ((vp->v_vflag & VV_PROCDEP) != 0);
2568 * Make this setguid thing safe, if at all possible.
2571 fdsetugidsafety(struct thread *td)
2573 struct filedesc *fdp;
2577 fdp = td->td_proc->p_fd;
2578 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2579 ("the fdtable should not be shared"));
2580 MPASS(fdp->fd_nfiles >= 3);
2581 for (i = 0; i <= 2; i++) {
2582 fp = fdp->fd_ofiles[i].fde_file;
2583 if (fp != NULL && is_unsafe(fp)) {
2584 FILEDESC_XLOCK(fdp);
2585 knote_fdclose(td, i);
2587 * NULL-out descriptor prior to close to avoid
2588 * a race while close blocks.
2591 FILEDESC_XUNLOCK(fdp);
2592 (void) closef(fp, td);
2598 * If a specific file object occupies a specific file descriptor, close the
2599 * file descriptor entry and drop a reference on the file object. This is a
2600 * convenience function to handle a subsequent error in a function that calls
2601 * falloc() that handles the race that another thread might have closed the
2602 * file descriptor out from under the thread creating the file object.
2605 fdclose(struct thread *td, struct file *fp, int idx)
2607 struct filedesc *fdp = td->td_proc->p_fd;
2609 FILEDESC_XLOCK(fdp);
2610 if (fdp->fd_ofiles[idx].fde_file == fp) {
2612 FILEDESC_XUNLOCK(fdp);
2615 FILEDESC_XUNLOCK(fdp);
2619 * Close any files on exec?
2622 fdcloseexec(struct thread *td)
2624 struct filedesc *fdp;
2625 struct filedescent *fde;
2629 fdp = td->td_proc->p_fd;
2630 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2631 ("the fdtable should not be shared"));
2632 lastfile = fdlastfile_single(fdp);
2633 for (i = 0; i <= lastfile; i++) {
2634 fde = &fdp->fd_ofiles[i];
2636 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2637 (fde->fde_flags & UF_EXCLOSE))) {
2638 FILEDESC_XLOCK(fdp);
2640 (void) closefp(fdp, i, fp, td, false, false);
2641 FILEDESC_UNLOCK_ASSERT(fdp);
2647 * It is unsafe for set[ug]id processes to be started with file
2648 * descriptors 0..2 closed, as these descriptors are given implicit
2649 * significance in the Standard C library. fdcheckstd() will create a
2650 * descriptor referencing /dev/null for each of stdin, stdout, and
2651 * stderr that is not already open.
2654 fdcheckstd(struct thread *td)
2656 struct filedesc *fdp;
2658 int i, error, devnull;
2660 fdp = td->td_proc->p_fd;
2661 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2662 ("the fdtable should not be shared"));
2663 MPASS(fdp->fd_nfiles >= 3);
2665 for (i = 0; i <= 2; i++) {
2666 if (fdp->fd_ofiles[i].fde_file != NULL)
2669 save = td->td_retval[0];
2670 if (devnull != -1) {
2671 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2673 error = kern_openat(td, AT_FDCWD, "/dev/null",
2674 UIO_SYSSPACE, O_RDWR, 0);
2676 devnull = td->td_retval[0];
2677 KASSERT(devnull == i, ("we didn't get our fd"));
2680 td->td_retval[0] = save;
2688 * Internal form of close. Decrement reference count on file structure.
2689 * Note: td may be NULL when closing a file that was being passed in a
2693 closef(struct file *fp, struct thread *td)
2697 struct filedesc_to_leader *fdtol;
2698 struct filedesc *fdp;
2703 * POSIX record locking dictates that any close releases ALL
2704 * locks owned by this process. This is handled by setting
2705 * a flag in the unlock to free ONLY locks obeying POSIX
2706 * semantics, and not to free BSD-style file locks.
2707 * If the descriptor was in a message, POSIX-style locks
2708 * aren't passed with the descriptor, and the thread pointer
2709 * will be NULL. Callers should be careful only to pass a
2710 * NULL thread pointer when there really is no owning
2711 * context that might have locks, or the locks will be
2714 if (fp->f_type == DTYPE_VNODE) {
2716 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2717 lf.l_whence = SEEK_SET;
2720 lf.l_type = F_UNLCK;
2721 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2722 F_UNLCK, &lf, F_POSIX);
2724 fdtol = td->td_proc->p_fdtol;
2725 if (fdtol != NULL) {
2727 * Handle special case where file descriptor table is
2728 * shared between multiple process leaders.
2730 fdp = td->td_proc->p_fd;
2731 FILEDESC_XLOCK(fdp);
2732 for (fdtol = fdtol->fdl_next;
2733 fdtol != td->td_proc->p_fdtol;
2734 fdtol = fdtol->fdl_next) {
2735 if ((fdtol->fdl_leader->p_flag &
2738 fdtol->fdl_holdcount++;
2739 FILEDESC_XUNLOCK(fdp);
2740 lf.l_whence = SEEK_SET;
2743 lf.l_type = F_UNLCK;
2745 (void) VOP_ADVLOCK(vp,
2746 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2748 FILEDESC_XLOCK(fdp);
2749 fdtol->fdl_holdcount--;
2750 if (fdtol->fdl_holdcount == 0 &&
2751 fdtol->fdl_wakeup != 0) {
2752 fdtol->fdl_wakeup = 0;
2756 FILEDESC_XUNLOCK(fdp);
2759 return (fdrop_close(fp, td));
2763 * Hack for file descriptor passing code.
2766 closef_nothread(struct file *fp)
2773 * Initialize the file pointer with the specified properties.
2775 * The ops are set with release semantics to be certain that the flags, type,
2776 * and data are visible when ops is. This is to prevent ops methods from being
2777 * called with bad data.
2780 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2785 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2789 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2791 fp->f_seqcount[UIO_READ] = 1;
2792 fp->f_seqcount[UIO_WRITE] = 1;
2793 finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2798 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2799 struct file **fpp, struct filecaps *havecapsp)
2801 struct filedescent *fde;
2804 FILEDESC_LOCK_ASSERT(fdp);
2806 fde = fdeget_locked(fdp, fd);
2813 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2818 if (havecapsp != NULL)
2819 filecaps_copy(&fde->fde_caps, havecapsp, true);
2821 *fpp = fde->fde_file;
2829 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2830 struct file **fpp, struct filecaps *havecapsp)
2832 struct filedesc *fdp = td->td_proc->p_fd;
2834 #ifndef CAPABILITIES
2835 error = fget_unlocked(fdp, fd, needrightsp, fpp);
2836 if (havecapsp != NULL && error == 0)
2837 filecaps_fill(havecapsp);
2844 error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2848 if (havecapsp != NULL) {
2849 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2850 havecapsp, false)) {
2856 if (!fd_modified(fdp, fd, seq))
2865 FILEDESC_SLOCK(fdp);
2866 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2867 if (error == 0 && !fhold(*fpp))
2869 FILEDESC_SUNLOCK(fdp);
2876 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2878 const struct filedescent *fde;
2879 const struct fdescenttbl *fdt;
2880 struct filedesc *fdp;
2883 const cap_rights_t *haverights;
2884 cap_rights_t rights;
2887 VFS_SMR_ASSERT_ENTERED();
2889 rights = *ndp->ni_rightsneeded;
2890 cap_rights_set_one(&rights, CAP_LOOKUP);
2892 fdp = curproc->p_fd;
2893 fdt = fdp->fd_files;
2894 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2896 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
2897 fde = &fdt->fdt_ofiles[fd];
2898 haverights = cap_rights_fde_inline(fde);
2900 if (__predict_false(fp == NULL))
2902 if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2904 *fsearch = ((fp->f_flag & FSEARCH) != 0);
2906 if (__predict_false(vp == NULL)) {
2909 if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2913 * Use an acquire barrier to force re-reading of fdt so it is
2914 * refreshed for verification.
2916 atomic_thread_fence_acq();
2917 fdt = fdp->fd_files;
2918 if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
2921 * If file descriptor doesn't have all rights,
2922 * all lookups relative to it must also be
2923 * strictly relative.
2925 * Not yet supported by fast path.
2928 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
2929 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
2930 ndp->ni_filecaps.fc_nioctls != -1) {
2932 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
2942 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2944 const struct fdescenttbl *fdt;
2945 struct filedesc *fdp;
2949 VFS_SMR_ASSERT_ENTERED();
2951 fdp = curproc->p_fd;
2952 fdt = fdp->fd_files;
2953 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2955 fp = fdt->fdt_ofiles[fd].fde_file;
2956 if (__predict_false(fp == NULL))
2958 *fsearch = ((fp->f_flag & FSEARCH) != 0);
2960 if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2964 * Use an acquire barrier to force re-reading of fdt so it is
2965 * refreshed for verification.
2967 atomic_thread_fence_acq();
2968 fdt = fdp->fd_files;
2969 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
2971 filecaps_fill(&ndp->ni_filecaps);
2978 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2979 struct file **fpp, seqc_t *seqp)
2982 const struct filedescent *fde;
2984 const struct fdescenttbl *fdt;
2988 cap_rights_t haverights;
2992 fdt = fdp->fd_files;
2993 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2996 * Fetch the descriptor locklessly. We avoid fdrop() races by
2997 * never raising a refcount above 0. To accomplish this we have
2998 * to use a cmpset loop rather than an atomic_add. The descriptor
2999 * must be re-verified once we acquire a reference to be certain
3000 * that the identity is still correct and we did not lose a race
3001 * due to preemption.
3005 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3006 fde = &fdt->fdt_ofiles[fd];
3007 haverights = *cap_rights_fde_inline(fde);
3009 if (!seqc_consistent(fd_seqc(fdt, fd), seq))
3012 fp = fdt->fdt_ofiles[fd].fde_file;
3017 error = cap_check_inline(&haverights, needrightsp);
3021 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3023 * Force a reload. Other thread could reallocate the
3024 * table before this fd was closed, so it is possible
3025 * that there is a stale fp pointer in cached version.
3027 fdt = atomic_load_ptr(&fdp->fd_files);
3031 * Use an acquire barrier to force re-reading of fdt so it is
3032 * refreshed for verification.
3034 atomic_thread_fence_acq();
3035 fdt = fdp->fd_files;
3037 if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
3039 if (fp == fdt->fdt_ofiles[fd].fde_file)
3042 fdrop(fp, curthread);
3054 * See the comments in fget_unlocked_seq for an explanation of how this works.
3056 * This is a simplified variant which bails out to the aforementioned routine
3057 * if anything goes wrong. In practice this only happens when userspace is
3058 * racing with itself.
3061 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3065 const struct filedescent *fde;
3067 const struct fdescenttbl *fdt;
3071 const cap_rights_t *haverights;
3074 fdt = fdp->fd_files;
3075 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3078 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3079 fde = &fdt->fdt_ofiles[fd];
3080 haverights = cap_rights_fde_inline(fde);
3083 fp = fdt->fdt_ofiles[fd].fde_file;
3085 if (__predict_false(fp == NULL))
3088 if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3091 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3095 * Use an acquire barrier to force re-reading of fdt so it is
3096 * refreshed for verification.
3098 atomic_thread_fence_acq();
3099 fdt = fdp->fd_files;
3101 if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3103 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3109 fdrop(fp, curthread);
3111 return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
3115 * Translate fd -> file when the caller guarantees the file descriptor table
3116 * can't be changed by others.
3118 * Note this does not mean the file object itself is only visible to the caller,
3119 * merely that it wont disappear without having to be referenced.
3121 * Must be paired with fput_only_user.
3125 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3128 const struct filedescent *fde;
3129 const struct fdescenttbl *fdt;
3130 const cap_rights_t *haverights;
3134 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3136 if (__predict_false(fd >= fdp->fd_nfiles))
3139 fdt = fdp->fd_files;
3140 fde = &fdt->fdt_ofiles[fd];
3142 if (__predict_false(fp == NULL))
3144 MPASS(refcount_load(&fp->f_count) > 0);
3145 haverights = cap_rights_fde_inline(fde);
3146 error = cap_check_inline(haverights, needrightsp);
3147 if (__predict_false(error != 0))
3154 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3159 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3161 if (__predict_false(fd >= fdp->fd_nfiles))
3164 fp = fdp->fd_ofiles[fd].fde_file;
3165 if (__predict_false(fp == NULL))
3168 MPASS(refcount_load(&fp->f_count) > 0);
3175 * Extract the file pointer associated with the specified descriptor for the
3176 * current user process.
3178 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3181 * File's rights will be checked against the capability rights mask.
3183 * If an error occurred the non-zero error is returned and *fpp is set to
3184 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
3185 * responsible for fdrop().
3188 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3189 cap_rights_t *needrightsp)
3191 struct filedesc *fdp;
3196 fdp = td->td_proc->p_fd;
3197 error = fget_unlocked(fdp, fd, needrightsp, &fp);
3198 if (__predict_false(error != 0))
3200 if (__predict_false(fp->f_ops == &badfileops)) {
3206 * FREAD and FWRITE failure return EBADF as per POSIX.
3212 if ((fp->f_flag & flags) == 0)
3216 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3217 ((fp->f_flag & FWRITE) != 0))
3223 KASSERT(0, ("wrong flags"));
3236 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3239 return (_fget(td, fd, fpp, 0, rightsp));
3243 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3247 #ifndef CAPABILITIES
3248 error = _fget(td, fd, fpp, 0, rightsp);
3249 if (maxprotp != NULL)
3250 *maxprotp = VM_PROT_ALL;
3253 cap_rights_t fdrights;
3254 struct filedesc *fdp;
3259 fdp = td->td_proc->p_fd;
3260 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3262 error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3263 if (__predict_false(error != 0))
3265 if (__predict_false(fp->f_ops == &badfileops)) {
3269 if (maxprotp != NULL)
3270 fdrights = *cap_rights(fdp, fd);
3271 if (!fd_modified(fdp, fd, seq))
3277 * If requested, convert capability rights to access flags.
3279 if (maxprotp != NULL)
3280 *maxprotp = cap_rights_to_vmprot(&fdrights);
3287 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3290 return (_fget(td, fd, fpp, FREAD, rightsp));
3294 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3297 return (_fget(td, fd, fpp, FWRITE, rightsp));
3301 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3304 struct filedesc *fdp = td->td_proc->p_fd;
3305 #ifndef CAPABILITIES
3306 return (fget_unlocked(fdp, fd, rightsp, fpp));
3313 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3315 error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3318 error = cap_fcntl_check(fdp, fd, needfcntl);
3319 if (!fd_modified(fdp, fd, seq))
3333 * Like fget() but loads the underlying vnode, or returns an error if the
3334 * descriptor does not represent a vnode. Note that pipes use vnodes but
3335 * never have VM objects. The returned vnode will be vref()'d.
3337 * XXX: what about the unused flags ?
3340 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3347 error = _fget(td, fd, &fp, flags, needrightsp);
3350 if (fp->f_vnode == NULL) {
3362 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3365 return (_fgetvp(td, fd, 0, rightsp, vpp));
3369 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3370 struct filecaps *havecaps, struct vnode **vpp)
3372 struct filecaps caps;
3376 error = fget_cap(td, fd, needrightsp, &fp, &caps);
3379 if (fp->f_ops == &badfileops) {
3383 if (fp->f_vnode == NULL) {
3395 filecaps_free(&caps);
3401 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3404 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3408 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3411 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3416 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3420 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3425 * Handle the last reference to a file being closed.
3427 * Without the noinline attribute clang keeps inlining the func thorough this
3428 * file when fdrop is used.
3431 _fdrop(struct file *fp, struct thread *td)
3437 count = refcount_load(&fp->f_count);
3439 panic("fdrop: fp %p count %d", fp, count);
3441 error = fo_close(fp, td);
3442 atomic_subtract_int(&openfiles, 1);
3444 free(fp->f_advice, M_FADVISE);
3445 uma_zfree(file_zone, fp);
3451 * Apply an advisory lock on a file descriptor.
3453 * Just attempt to get a record lock of the requested type on the entire file
3454 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3456 #ifndef _SYS_SYSPROTO_H_
3464 sys_flock(struct thread *td, struct flock_args *uap)
3471 error = fget(td, uap->fd, &cap_flock_rights, &fp);
3475 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
3478 if (fp->f_ops == &path_fileops) {
3484 lf.l_whence = SEEK_SET;
3487 if (uap->how & LOCK_UN) {
3488 lf.l_type = F_UNLCK;
3489 atomic_clear_int(&fp->f_flag, FHASLOCK);
3490 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3493 if (uap->how & LOCK_EX)
3494 lf.l_type = F_WRLCK;
3495 else if (uap->how & LOCK_SH)
3496 lf.l_type = F_RDLCK;
3501 atomic_set_int(&fp->f_flag, FHASLOCK);
3502 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3503 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3509 * Duplicate the specified descriptor to a free descriptor.
3512 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3513 int openerror, int *indxp)
3515 struct filedescent *newfde, *oldfde;
3520 KASSERT(openerror == ENODEV || openerror == ENXIO,
3521 ("unexpected error %d in %s", openerror, __func__));
3524 * If the to-be-dup'd fd number is greater than the allowed number
3525 * of file descriptors, or the fd to be dup'd has already been
3526 * closed, then reject.
3528 FILEDESC_XLOCK(fdp);
3529 if ((fp = fget_locked(fdp, dfd)) == NULL) {
3530 FILEDESC_XUNLOCK(fdp);
3534 error = fdalloc(td, 0, &indx);
3536 FILEDESC_XUNLOCK(fdp);
3541 * There are two cases of interest here.
3543 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3545 * For ENXIO steal away the file structure from (dfd) and store it in
3546 * (indx). (dfd) is effectively closed by this operation.
3548 switch (openerror) {
3551 * Check that the mode the file is being opened for is a
3552 * subset of the mode of the existing descriptor.
3554 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3555 fdunused(fdp, indx);
3556 FILEDESC_XUNLOCK(fdp);
3560 fdunused(fdp, indx);
3561 FILEDESC_XUNLOCK(fdp);
3564 newfde = &fdp->fd_ofiles[indx];
3565 oldfde = &fdp->fd_ofiles[dfd];
3566 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3568 seqc_write_begin(&newfde->fde_seqc);
3570 memcpy(newfde, oldfde, fde_change_size);
3571 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3574 seqc_write_end(&newfde->fde_seqc);
3579 * Steal away the file pointer from dfd and stuff it into indx.
3581 newfde = &fdp->fd_ofiles[indx];
3582 oldfde = &fdp->fd_ofiles[dfd];
3584 seqc_write_begin(&newfde->fde_seqc);
3586 memcpy(newfde, oldfde, fde_change_size);
3587 oldfde->fde_file = NULL;
3590 seqc_write_end(&newfde->fde_seqc);
3594 FILEDESC_XUNLOCK(fdp);
3600 * This sysctl determines if we will allow a process to chroot(2) if it
3601 * has a directory open:
3602 * 0: disallowed for all processes.
3603 * 1: allowed for processes that were not already chroot(2)'ed.
3604 * 2: allowed for all processes.
3607 static int chroot_allow_open_directories = 1;
3609 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3610 &chroot_allow_open_directories, 0,
3611 "Allow a process to chroot(2) if it has a directory open");
3614 * Helper function for raised chroot(2) security function: Refuse if
3615 * any filedescriptors are open directories.
3618 chroot_refuse_vdir_fds(struct filedesc *fdp)
3624 FILEDESC_LOCK_ASSERT(fdp);
3626 lastfile = fdlastfile(fdp);
3627 for (fd = 0; fd <= lastfile; fd++) {
3628 fp = fget_locked(fdp, fd);
3631 if (fp->f_type == DTYPE_VNODE) {
3633 if (vp->v_type == VDIR)
3641 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3644 if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3645 vrefact(oldpwd->pwd_cdir);
3646 newpwd->pwd_cdir = oldpwd->pwd_cdir;
3649 if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3650 vrefact(oldpwd->pwd_rdir);
3651 newpwd->pwd_rdir = oldpwd->pwd_rdir;
3654 if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3655 vrefact(oldpwd->pwd_jdir);
3656 newpwd->pwd_jdir = oldpwd->pwd_jdir;
3661 pwd_hold_pwddesc(struct pwddesc *pdp)
3665 PWDDESC_ASSERT_XLOCKED(pdp);
3666 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3668 refcount_acquire(&pwd->pwd_refcount);
3673 pwd_hold_smr(struct pwd *pwd)
3677 if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3684 pwd_hold(struct thread *td)
3686 struct pwddesc *pdp;
3689 pdp = td->td_proc->p_pd;
3692 pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3693 if (pwd_hold_smr(pwd)) {
3699 pwd = pwd_hold_pwddesc(pdp);
3701 PWDDESC_XUNLOCK(pdp);
3706 pwd_hold_proc(struct proc *p)
3708 struct pwddesc *pdp;
3711 PROC_ASSERT_HELD(p);
3718 pwd = pwd_hold_pwddesc(pdp);
3720 PWDDESC_XUNLOCK(pdp);
3730 pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3731 bzero(pwd, sizeof(*pwd));
3732 refcount_init(&pwd->pwd_refcount, 1);
3737 pwd_drop(struct pwd *pwd)
3740 if (!refcount_release(&pwd->pwd_refcount))
3743 if (pwd->pwd_cdir != NULL)
3744 vrele(pwd->pwd_cdir);
3745 if (pwd->pwd_rdir != NULL)
3746 vrele(pwd->pwd_rdir);
3747 if (pwd->pwd_jdir != NULL)
3748 vrele(pwd->pwd_jdir);
3749 uma_zfree_smr(pwd_zone, pwd);
3753 * The caller is responsible for invoking priv_check() and
3754 * mac_vnode_check_chroot() to authorize this operation.
3757 pwd_chroot(struct thread *td, struct vnode *vp)
3759 struct pwddesc *pdp;
3760 struct filedesc *fdp;
3761 struct pwd *newpwd, *oldpwd;
3764 fdp = td->td_proc->p_fd;
3765 pdp = td->td_proc->p_pd;
3766 newpwd = pwd_alloc();
3767 FILEDESC_SLOCK(fdp);
3769 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3770 if (chroot_allow_open_directories == 0 ||
3771 (chroot_allow_open_directories == 1 &&
3772 oldpwd->pwd_rdir != rootvnode)) {
3773 error = chroot_refuse_vdir_fds(fdp);
3774 FILEDESC_SUNLOCK(fdp);
3776 PWDDESC_XUNLOCK(pdp);
3781 FILEDESC_SUNLOCK(fdp);
3785 newpwd->pwd_rdir = vp;
3786 if (oldpwd->pwd_jdir == NULL) {
3788 newpwd->pwd_jdir = vp;
3790 pwd_fill(oldpwd, newpwd);
3791 pwd_set(pdp, newpwd);
3792 PWDDESC_XUNLOCK(pdp);
3798 pwd_chdir(struct thread *td, struct vnode *vp)
3800 struct pwddesc *pdp;
3801 struct pwd *newpwd, *oldpwd;
3803 VNPASS(vp->v_usecount > 0, vp);
3805 newpwd = pwd_alloc();
3806 pdp = td->td_proc->p_pd;
3808 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3809 newpwd->pwd_cdir = vp;
3810 pwd_fill(oldpwd, newpwd);
3811 pwd_set(pdp, newpwd);
3812 PWDDESC_XUNLOCK(pdp);
3817 * jail_attach(2) changes both root and working directories.
3820 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
3822 struct pwddesc *pdp;
3823 struct filedesc *fdp;
3824 struct pwd *newpwd, *oldpwd;
3827 fdp = td->td_proc->p_fd;
3828 pdp = td->td_proc->p_pd;
3829 newpwd = pwd_alloc();
3830 FILEDESC_SLOCK(fdp);
3832 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3833 error = chroot_refuse_vdir_fds(fdp);
3834 FILEDESC_SUNLOCK(fdp);
3836 PWDDESC_XUNLOCK(pdp);
3842 newpwd->pwd_rdir = vp;
3844 newpwd->pwd_cdir = vp;
3845 if (oldpwd->pwd_jdir == NULL) {
3847 newpwd->pwd_jdir = vp;
3849 pwd_fill(oldpwd, newpwd);
3850 pwd_set(pdp, newpwd);
3851 PWDDESC_XUNLOCK(pdp);
3857 pwd_ensure_dirs(void)
3859 struct pwddesc *pdp;
3860 struct pwd *oldpwd, *newpwd;
3862 pdp = curproc->p_pd;
3864 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3865 if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3866 PWDDESC_XUNLOCK(pdp);
3869 PWDDESC_XUNLOCK(pdp);
3871 newpwd = pwd_alloc();
3873 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3874 pwd_fill(oldpwd, newpwd);
3875 if (newpwd->pwd_cdir == NULL) {
3877 newpwd->pwd_cdir = rootvnode;
3879 if (newpwd->pwd_rdir == NULL) {
3881 newpwd->pwd_rdir = rootvnode;
3883 pwd_set(pdp, newpwd);
3884 PWDDESC_XUNLOCK(pdp);
3889 pwd_set_rootvnode(void)
3891 struct pwddesc *pdp;
3892 struct pwd *oldpwd, *newpwd;
3894 pdp = curproc->p_pd;
3896 newpwd = pwd_alloc();
3898 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3900 newpwd->pwd_cdir = rootvnode;
3902 newpwd->pwd_rdir = rootvnode;
3903 pwd_fill(oldpwd, newpwd);
3904 pwd_set(pdp, newpwd);
3905 PWDDESC_XUNLOCK(pdp);
3910 * Scan all active processes and prisons to see if any of them have a current
3911 * or root directory of `olddp'. If so, replace them with the new mount point.
3914 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3916 struct pwddesc *pdp;
3917 struct pwd *newpwd, *oldpwd;
3922 if (vrefcnt(olddp) == 1)
3925 newpwd = pwd_alloc();
3926 sx_slock(&allproc_lock);
3927 FOREACH_PROC_IN_SYSTEM(p) {
3934 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3935 if (oldpwd == NULL ||
3936 (oldpwd->pwd_cdir != olddp &&
3937 oldpwd->pwd_rdir != olddp &&
3938 oldpwd->pwd_jdir != olddp)) {
3939 PWDDESC_XUNLOCK(pdp);
3943 if (oldpwd->pwd_cdir == olddp) {
3945 newpwd->pwd_cdir = newdp;
3947 if (oldpwd->pwd_rdir == olddp) {
3949 newpwd->pwd_rdir = newdp;
3951 if (oldpwd->pwd_jdir == olddp) {
3953 newpwd->pwd_jdir = newdp;
3955 pwd_fill(oldpwd, newpwd);
3956 pwd_set(pdp, newpwd);
3957 PWDDESC_XUNLOCK(pdp);
3960 newpwd = pwd_alloc();
3962 sx_sunlock(&allproc_lock);
3964 if (rootvnode == olddp) {
3969 mtx_lock(&prison0.pr_mtx);
3970 if (prison0.pr_root == olddp) {
3972 prison0.pr_root = newdp;
3975 mtx_unlock(&prison0.pr_mtx);
3976 sx_slock(&allprison_lock);
3977 TAILQ_FOREACH(pr, &allprison, pr_list) {
3978 mtx_lock(&pr->pr_mtx);
3979 if (pr->pr_root == olddp) {
3981 pr->pr_root = newdp;
3984 mtx_unlock(&pr->pr_mtx);
3986 sx_sunlock(&allprison_lock);
3991 struct filedesc_to_leader *
3992 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3994 struct filedesc_to_leader *fdtol;
3996 fdtol = malloc(sizeof(struct filedesc_to_leader),
3997 M_FILEDESC_TO_LEADER, M_WAITOK);
3998 fdtol->fdl_refcount = 1;
3999 fdtol->fdl_holdcount = 0;
4000 fdtol->fdl_wakeup = 0;
4001 fdtol->fdl_leader = leader;
4003 FILEDESC_XLOCK(fdp);
4004 fdtol->fdl_next = old->fdl_next;
4005 fdtol->fdl_prev = old;
4006 old->fdl_next = fdtol;
4007 fdtol->fdl_next->fdl_prev = fdtol;
4008 FILEDESC_XUNLOCK(fdp);
4010 fdtol->fdl_next = fdtol;
4011 fdtol->fdl_prev = fdtol;
4017 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
4020 struct filedesc *fdp;
4022 int count, off, minoff;
4028 if (*(int *)arg1 != 0)
4031 fdp = curproc->p_fd;
4033 FILEDESC_SLOCK(fdp);
4035 off = NDSLOT(fdp->fd_nfiles - 1);
4036 for (minoff = NDSLOT(0); off >= minoff; --off)
4037 count += bitcountl(map[off]);
4038 FILEDESC_SUNLOCK(fdp);
4040 return (SYSCTL_OUT(req, &count, sizeof(count)));
4043 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
4044 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
4045 "Number of open file descriptors");
4048 * Get file structures globally.
4051 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
4054 struct filedesc *fdp;
4057 int error, n, lastfile;
4059 error = sysctl_wire_old_buffer(req, 0);
4062 if (req->oldptr == NULL) {
4064 sx_slock(&allproc_lock);
4065 FOREACH_PROC_IN_SYSTEM(p) {
4067 if (p->p_state == PRS_NEW) {
4075 /* overestimates sparse tables. */
4076 n += fdp->fd_nfiles;
4079 sx_sunlock(&allproc_lock);
4080 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4083 bzero(&xf, sizeof(xf));
4084 xf.xf_size = sizeof(xf);
4085 sx_slock(&allproc_lock);
4086 FOREACH_PROC_IN_SYSTEM(p) {
4088 if (p->p_state == PRS_NEW) {
4092 if (p_cansee(req->td, p) != 0) {
4096 xf.xf_pid = p->p_pid;
4097 xf.xf_uid = p->p_ucred->cr_uid;
4102 FILEDESC_SLOCK(fdp);
4103 lastfile = fdlastfile(fdp);
4104 for (n = 0; refcount_load(&fdp->fd_refcnt) > 0 && n <= lastfile;
4106 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4109 xf.xf_file = (uintptr_t)fp;
4110 xf.xf_data = (uintptr_t)fp->f_data;
4111 xf.xf_vnode = (uintptr_t)fp->f_vnode;
4112 xf.xf_type = (uintptr_t)fp->f_type;
4113 xf.xf_count = refcount_load(&fp->f_count);
4115 xf.xf_offset = foffset_get(fp);
4116 xf.xf_flag = fp->f_flag;
4117 error = SYSCTL_OUT(req, &xf, sizeof(xf));
4121 FILEDESC_SUNLOCK(fdp);
4126 sx_sunlock(&allproc_lock);
4130 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4131 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4133 #ifdef KINFO_FILE_SIZE
4134 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4138 xlate_fflags(int fflags)
4140 static const struct {
4143 } fflags_table[] = {
4144 { FAPPEND, KF_FLAG_APPEND },
4145 { FASYNC, KF_FLAG_ASYNC },
4146 { FFSYNC, KF_FLAG_FSYNC },
4147 { FHASLOCK, KF_FLAG_HASLOCK },
4148 { FNONBLOCK, KF_FLAG_NONBLOCK },
4149 { FREAD, KF_FLAG_READ },
4150 { FWRITE, KF_FLAG_WRITE },
4151 { O_CREAT, KF_FLAG_CREAT },
4152 { O_DIRECT, KF_FLAG_DIRECT },
4153 { O_EXCL, KF_FLAG_EXCL },
4154 { O_EXEC, KF_FLAG_EXEC },
4155 { O_EXLOCK, KF_FLAG_EXLOCK },
4156 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4157 { O_SHLOCK, KF_FLAG_SHLOCK },
4158 { O_TRUNC, KF_FLAG_TRUNC }
4164 for (i = 0; i < nitems(fflags_table); i++)
4165 if (fflags & fflags_table[i].fflag)
4166 kflags |= fflags_table[i].kf_fflag;
4170 /* Trim unused data from kf_path by truncating the structure size. */
4172 pack_kinfo(struct kinfo_file *kif)
4175 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4176 strlen(kif->kf_path) + 1;
4177 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4181 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4182 struct kinfo_file *kif, struct filedesc *fdp, int flags)
4186 bzero(kif, sizeof(*kif));
4188 /* Set a default type to allow for empty fill_kinfo() methods. */
4189 kif->kf_type = KF_TYPE_UNKNOWN;
4190 kif->kf_flags = xlate_fflags(fp->f_flag);
4191 if (rightsp != NULL)
4192 kif->kf_cap_rights = *rightsp;
4194 cap_rights_init_zero(&kif->kf_cap_rights);
4196 kif->kf_ref_count = refcount_load(&fp->f_count);
4197 kif->kf_offset = foffset_get(fp);
4200 * This may drop the filedesc lock, so the 'fp' cannot be
4201 * accessed after this call.
4203 error = fo_fill_kinfo(fp, kif, fdp);
4205 kif->kf_status |= KF_ATTR_VALID;
4206 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4209 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4213 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4214 struct kinfo_file *kif, int flags)
4218 bzero(kif, sizeof(*kif));
4220 kif->kf_type = KF_TYPE_VNODE;
4221 error = vn_fill_kinfo_vnode(vp, kif);
4223 kif->kf_status |= KF_ATTR_VALID;
4224 kif->kf_flags = xlate_fflags(fflags);
4225 cap_rights_init_zero(&kif->kf_cap_rights);
4227 kif->kf_ref_count = -1;
4228 kif->kf_offset = -1;
4229 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4232 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4236 struct export_fd_buf {
4237 struct filedesc *fdp;
4238 struct pwddesc *pdp;
4241 struct kinfo_file kif;
4246 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4248 struct kinfo_file *kif;
4251 if (efbuf->remainder != -1) {
4252 if (efbuf->remainder < kif->kf_structsize) {
4253 /* Terminate export. */
4254 efbuf->remainder = 0;
4257 efbuf->remainder -= kif->kf_structsize;
4259 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
4263 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4264 struct export_fd_buf *efbuf)
4268 if (efbuf->remainder == 0)
4270 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4272 FILEDESC_SUNLOCK(efbuf->fdp);
4273 error = export_kinfo_to_sb(efbuf);
4274 FILEDESC_SLOCK(efbuf->fdp);
4279 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4280 struct export_fd_buf *efbuf)
4284 if (efbuf->remainder == 0)
4286 if (efbuf->pdp != NULL)
4287 PWDDESC_XUNLOCK(efbuf->pdp);
4288 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4289 error = export_kinfo_to_sb(efbuf);
4290 if (efbuf->pdp != NULL)
4291 PWDDESC_XLOCK(efbuf->pdp);
4296 * Store a process file descriptor information to sbuf.
4298 * Takes a locked proc as argument, and returns with the proc unlocked.
4301 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
4305 struct filedesc *fdp;
4306 struct pwddesc *pdp;
4307 struct export_fd_buf *efbuf;
4308 struct vnode *cttyvp, *textvp, *tracevp;
4310 int error, i, lastfile;
4311 cap_rights_t rights;
4313 PROC_LOCK_ASSERT(p, MA_OWNED);
4316 tracevp = ktr_get_tracevp(p, true);
4318 textvp = p->p_textvp;
4321 /* Controlling tty. */
4323 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4324 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4331 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4335 efbuf->remainder = maxlen;
4336 efbuf->flags = flags;
4337 if (tracevp != NULL)
4338 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
4341 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
4343 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
4346 if (pdp == NULL || fdp == NULL)
4351 pwd = pwd_hold_pwddesc(pdp);
4353 /* working directory */
4354 if (pwd->pwd_cdir != NULL) {
4355 vrefact(pwd->pwd_cdir);
4356 export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD,
4359 /* root directory */
4360 if (pwd->pwd_rdir != NULL) {
4361 vrefact(pwd->pwd_rdir);
4362 export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT,
4365 /* jail directory */
4366 if (pwd->pwd_jdir != NULL) {
4367 vrefact(pwd->pwd_jdir);
4368 export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL,
4372 PWDDESC_XUNLOCK(pdp);
4375 FILEDESC_SLOCK(fdp);
4376 lastfile = fdlastfile(fdp);
4377 for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4378 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4381 rights = *cap_rights(fdp, i);
4382 #else /* !CAPABILITIES */
4383 rights = cap_no_rights;
4386 * Create sysctl entry. It is OK to drop the filedesc
4387 * lock inside of export_file_to_sb() as we will
4388 * re-validate and re-evaluate its properties when the
4391 error = export_file_to_sb(fp, i, &rights, efbuf);
4392 if (error != 0 || efbuf->remainder == 0)
4395 FILEDESC_SUNLOCK(fdp);
4401 free(efbuf, M_TEMP);
4405 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
4408 * Get per-process file descriptors for use by procstat(1), et al.
4411 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4417 int error, error2, *name;
4425 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4426 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4427 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4432 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4433 error = kern_proc_filedesc_out(p, &sb, maxlen,
4434 KERN_FILEDESC_PACK_KINFO);
4435 error2 = sbuf_finish(&sb);
4437 return (error != 0 ? error : error2);
4440 #ifdef COMPAT_FREEBSD7
4441 #ifdef KINFO_OFILE_SIZE
4442 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4446 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4449 okif->kf_structsize = sizeof(*okif);
4450 okif->kf_type = kif->kf_type;
4451 okif->kf_fd = kif->kf_fd;
4452 okif->kf_ref_count = kif->kf_ref_count;
4453 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4454 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4455 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4456 okif->kf_offset = kif->kf_offset;
4457 if (kif->kf_type == KF_TYPE_VNODE)
4458 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4460 okif->kf_vnode_type = KF_VTYPE_VNON;
4461 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4462 if (kif->kf_type == KF_TYPE_SOCKET) {
4463 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4464 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4465 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4466 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4467 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4469 okif->kf_sa_local.ss_family = AF_UNSPEC;
4470 okif->kf_sa_peer.ss_family = AF_UNSPEC;
4475 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4476 struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4481 PWDDESC_XUNLOCK(pdp);
4482 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4483 kinfo_to_okinfo(kif, okif);
4484 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4490 * Get per-process file descriptors for use by procstat(1), et al.
4493 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4495 struct kinfo_ofile *okif;
4496 struct kinfo_file *kif;
4497 struct filedesc *fdp;
4498 struct pwddesc *pdp;
4501 int error, i, lastfile, *name;
4510 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4517 if (fdp == NULL || pdp == NULL) {
4522 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4523 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4525 pwd = pwd_hold_pwddesc(pdp);
4527 if (pwd->pwd_cdir != NULL)
4528 export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4530 if (pwd->pwd_rdir != NULL)
4531 export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4533 if (pwd->pwd_jdir != NULL)
4534 export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4537 PWDDESC_XUNLOCK(pdp);
4540 FILEDESC_SLOCK(fdp);
4541 lastfile = fdlastfile(fdp);
4542 for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4543 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4545 export_file_to_kinfo(fp, i, NULL, kif, fdp,
4546 KERN_FILEDESC_PACK_KINFO);
4547 FILEDESC_SUNLOCK(fdp);
4548 kinfo_to_okinfo(kif, okif);
4549 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4550 FILEDESC_SLOCK(fdp);
4554 FILEDESC_SUNLOCK(fdp);
4562 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4563 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4564 "Process ofiledesc entries");
4565 #endif /* COMPAT_FREEBSD7 */
4568 vntype_to_kinfo(int vtype)
4573 } vtypes_table[] = {
4574 { VBAD, KF_VTYPE_VBAD },
4575 { VBLK, KF_VTYPE_VBLK },
4576 { VCHR, KF_VTYPE_VCHR },
4577 { VDIR, KF_VTYPE_VDIR },
4578 { VFIFO, KF_VTYPE_VFIFO },
4579 { VLNK, KF_VTYPE_VLNK },
4580 { VNON, KF_VTYPE_VNON },
4581 { VREG, KF_VTYPE_VREG },
4582 { VSOCK, KF_VTYPE_VSOCK }
4587 * Perform vtype translation.
4589 for (i = 0; i < nitems(vtypes_table); i++)
4590 if (vtypes_table[i].vtype == vtype)
4591 return (vtypes_table[i].kf_vtype);
4593 return (KF_VTYPE_UNKNOWN);
4596 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4597 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4598 "Process filedesc entries");
4601 * Store a process current working directory information to sbuf.
4603 * Takes a locked proc as argument, and returns with the proc unlocked.
4606 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
4608 struct pwddesc *pdp;
4610 struct export_fd_buf *efbuf;
4614 PROC_LOCK_ASSERT(p, MA_OWNED);
4621 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4624 efbuf->remainder = maxlen;
4627 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4628 cdir = pwd->pwd_cdir;
4633 error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4635 PWDDESC_XUNLOCK(pdp);
4637 free(efbuf, M_TEMP);
4642 * Get per-process current working directory.
4645 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4651 int error, error2, *name;
4659 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4660 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4661 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4666 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4667 error = kern_proc_cwd_out(p, &sb, maxlen);
4668 error2 = sbuf_finish(&sb);
4670 return (error != 0 ? error : error2);
4673 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4674 sysctl_kern_proc_cwd, "Process current working directory");
4678 * For the purposes of debugging, generate a human-readable string for the
4682 file_type_to_name(short type)
4710 case DTYPE_PROCDESC:
4714 case DTYPE_LINUXTFD:
4722 * For the purposes of debugging, identify a process (if any, perhaps one of
4723 * many) that references the passed file in its file descriptor array. Return
4726 static struct proc *
4727 file_to_first_proc(struct file *fp)
4729 struct filedesc *fdp;
4733 FOREACH_PROC_IN_SYSTEM(p) {
4734 if (p->p_state == PRS_NEW)
4739 for (n = 0; n < fdp->fd_nfiles; n++) {
4740 if (fp == fdp->fd_ofiles[n].fde_file)
4748 db_print_file(struct file *fp, int header)
4750 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4754 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4755 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4756 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4758 p = file_to_first_proc(fp);
4759 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4760 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4761 fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4762 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4767 DB_SHOW_COMMAND(file, db_show_file)
4772 db_printf("usage: show file <addr>\n");
4775 fp = (struct file *)addr;
4776 db_print_file(fp, 1);
4779 DB_SHOW_COMMAND(files, db_show_files)
4781 struct filedesc *fdp;
4788 FOREACH_PROC_IN_SYSTEM(p) {
4789 if (p->p_state == PRS_NEW)
4791 if ((fdp = p->p_fd) == NULL)
4793 for (n = 0; n < fdp->fd_nfiles; ++n) {
4794 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4796 db_print_file(fp, header);
4803 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4804 &maxfilesperproc, 0, "Maximum files allowed open per process");
4806 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4807 &maxfiles, 0, "Maximum number of files");
4809 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4810 &openfiles, 0, "System-wide number of open files");
4814 filelistinit(void *dummy)
4817 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4818 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4819 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4820 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4821 pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4822 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4824 * XXXMJG this is a temporary hack due to boot ordering issues against
4827 vfs_smr = uma_zone_get_smr(pwd_zone);
4828 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4830 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4832 /*-------------------------------------------------------------------*/
4835 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4836 int flags, struct thread *td)
4843 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4851 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4859 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4867 badfo_kqfilter(struct file *fp, struct knote *kn)
4874 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
4881 badfo_close(struct file *fp, struct thread *td)
4888 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4896 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4904 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4905 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4913 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4919 struct fileops badfileops = {
4920 .fo_read = badfo_readwrite,
4921 .fo_write = badfo_readwrite,
4922 .fo_truncate = badfo_truncate,
4923 .fo_ioctl = badfo_ioctl,
4924 .fo_poll = badfo_poll,
4925 .fo_kqfilter = badfo_kqfilter,
4926 .fo_stat = badfo_stat,
4927 .fo_close = badfo_close,
4928 .fo_chmod = badfo_chmod,
4929 .fo_chown = badfo_chown,
4930 .fo_sendfile = badfo_sendfile,
4931 .fo_fill_kinfo = badfo_fill_kinfo,
4935 path_poll(struct file *fp, int events, struct ucred *active_cred,
4942 path_close(struct file *fp, struct thread *td)
4944 MPASS(fp->f_type == DTYPE_VNODE);
4945 fp->f_ops = &badfileops;
4950 struct fileops path_fileops = {
4951 .fo_read = badfo_readwrite,
4952 .fo_write = badfo_readwrite,
4953 .fo_truncate = badfo_truncate,
4954 .fo_ioctl = badfo_ioctl,
4955 .fo_poll = path_poll,
4956 .fo_kqfilter = vn_kqfilter_opath,
4957 .fo_stat = vn_statfile,
4958 .fo_close = path_close,
4959 .fo_chmod = badfo_chmod,
4960 .fo_chown = badfo_chown,
4961 .fo_sendfile = badfo_sendfile,
4962 .fo_fill_kinfo = vn_fill_kinfo,
4963 .fo_flags = DFLAG_PASSABLE,
4967 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4968 int flags, struct thread *td)
4971 return (EOPNOTSUPP);
4975 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4983 invfo_ioctl(struct file *fp, u_long com, void *data,
4984 struct ucred *active_cred, struct thread *td)
4991 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4995 return (poll_no_poll(events));
4999 invfo_kqfilter(struct file *fp, struct knote *kn)
5006 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5014 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5022 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5023 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5030 /*-------------------------------------------------------------------*/
5033 * File Descriptor pseudo-device driver (/dev/fd/).
5035 * Opening minor device N dup()s the file (if any) connected to file
5036 * descriptor N belonging to the calling process. Note that this driver
5037 * consists of only the ``open()'' routine, because all subsequent
5038 * references to this file will be direct to the other driver.
5040 * XXX: we could give this one a cloning event handler if necessary.
5045 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
5049 * XXX Kludge: set curthread->td_dupfd to contain the value of the
5050 * the file descriptor being sought for duplication. The error
5051 * return ensures that the vnode for this device will be released
5052 * by vn_open. Open will detect this special error and take the
5053 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
5054 * will simply report the error.
5056 td->td_dupfd = dev2unit(dev);
5060 static struct cdevsw fildesc_cdevsw = {
5061 .d_version = D_VERSION,
5067 fildesc_drvinit(void *unused)
5071 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
5072 UID_ROOT, GID_WHEEL, 0666, "fd/0");
5073 make_dev_alias(dev, "stdin");
5074 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
5075 UID_ROOT, GID_WHEEL, 0666, "fd/1");
5076 make_dev_alias(dev, "stdout");
5077 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
5078 UID_ROOT, GID_WHEEL, 0666, "fd/2");
5079 make_dev_alias(dev, "stderr");
5082 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);