2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1989, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
42 #include "opt_capsicum.h"
44 #include "opt_ktrace.h"
46 #include <sys/param.h>
47 #include <sys/systm.h>
49 #include <sys/capsicum.h>
51 #include <sys/fcntl.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
66 #include <sys/protosw.h>
67 #include <sys/racct.h>
68 #include <sys/resourcevar.h>
70 #include <sys/signalvar.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysctl.h>
77 #include <sys/sysproto.h>
78 #include <sys/unistd.h>
80 #include <sys/vnode.h>
82 #include <sys/ktrace.h>
87 #include <security/audit/audit.h>
94 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
95 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
96 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
97 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
98 "file desc to leader structures");
99 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
100 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
102 MALLOC_DECLARE(M_FADVISE);
104 static __read_mostly uma_zone_t file_zone;
105 static __read_mostly uma_zone_t filedesc0_zone;
106 __read_mostly uma_zone_t pwd_zone;
109 static int closefp(struct filedesc *fdp, int fd, struct file *fp,
110 struct thread *td, int holdleaders);
111 static int fd_first_free(struct filedesc *fdp, int low, int size);
112 static void fdgrowtable(struct filedesc *fdp, int nfd);
113 static void fdgrowtable_exp(struct filedesc *fdp, int nfd);
114 static void fdunused(struct filedesc *fdp, int fd);
115 static void fdused(struct filedesc *fdp, int fd);
116 static int getmaxfd(struct thread *td);
117 static u_long *filecaps_copy_prep(const struct filecaps *src);
118 static void filecaps_copy_finish(const struct filecaps *src,
119 struct filecaps *dst, u_long *ioctls);
120 static u_long *filecaps_free_prep(struct filecaps *fcaps);
121 static void filecaps_free_finish(u_long *ioctls);
123 static struct pwd *pwd_alloc(void);
128 * - An array of open file descriptors (fd_ofiles)
129 * - An array of file flags (fd_ofileflags)
130 * - A bitmap recording which descriptors are in use (fd_map)
132 * A process starts out with NDFILE descriptors. The value of NDFILE has
133 * been selected based the historical limit of 20 open files, and an
134 * assumption that the majority of processes, especially short-lived
135 * processes like shells, will never need more.
137 * If this initial allocation is exhausted, a larger descriptor table and
138 * map are allocated dynamically, and the pointers in the process's struct
139 * filedesc are updated to point to those. This is repeated every time
140 * the process runs out of file descriptors (provided it hasn't hit its
143 * Since threads may hold references to individual descriptor table
144 * entries, the tables are never freed. Instead, they are placed on a
145 * linked list and freed only when the struct filedesc is released.
148 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
149 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
150 #define NDSLOT(x) ((x) / NDENTRIES)
151 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
152 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
155 * SLIST entry used to keep track of ofiles which must be reclaimed when
159 struct fdescenttbl *ft_table;
160 SLIST_ENTRY(freetable) ft_next;
164 * Initial allocation: a filedesc structure + the head of SLIST used to
165 * keep track of old ofiles + enough space for NDFILE descriptors.
168 struct fdescenttbl0 {
170 struct filedescent fdt_ofiles[NDFILE];
174 struct filedesc fd_fd;
175 SLIST_HEAD(, freetable) fd_free;
176 struct fdescenttbl0 fd_dfiles;
177 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
181 * Descriptor management.
183 static int __exclusive_cache_line openfiles; /* actual number of open files */
184 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
185 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
188 * If low >= size, just return low. Otherwise find the first zero bit in the
189 * given bitmap, starting at low and not exceeding size - 1. Return size if
193 fd_first_free(struct filedesc *fdp, int low, int size)
195 NDSLOTTYPE *map = fdp->fd_map;
203 if (low % NDENTRIES) {
204 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
205 if ((mask &= ~map[off]) != 0UL)
206 return (off * NDENTRIES + ffsl(mask) - 1);
209 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
210 if (map[off] != ~0UL)
211 return (off * NDENTRIES + ffsl(~map[off]) - 1);
216 * Find the last used fd.
218 * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
219 * Otherwise use fdlastfile.
222 fdlastfile_single(struct filedesc *fdp)
224 NDSLOTTYPE *map = fdp->fd_map;
227 off = NDSLOT(fdp->fd_nfiles - 1);
228 for (minoff = NDSLOT(0); off >= minoff; --off)
230 return (off * NDENTRIES + flsl(map[off]) - 1);
235 fdlastfile(struct filedesc *fdp)
238 FILEDESC_LOCK_ASSERT(fdp);
239 return (fdlastfile_single(fdp));
243 fdisused(struct filedesc *fdp, int fd)
246 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
247 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
249 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
253 * Mark a file descriptor as used.
256 fdused_init(struct filedesc *fdp, int fd)
259 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
261 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
265 fdused(struct filedesc *fdp, int fd)
268 FILEDESC_XLOCK_ASSERT(fdp);
270 fdused_init(fdp, fd);
271 if (fd == fdp->fd_freefile)
276 * Mark a file descriptor as unused.
279 fdunused(struct filedesc *fdp, int fd)
282 FILEDESC_XLOCK_ASSERT(fdp);
284 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
285 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
286 ("fd=%d is still in use", fd));
288 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
289 if (fd < fdp->fd_freefile)
290 fdp->fd_freefile = fd;
294 * Free a file descriptor.
296 * Avoid some work if fdp is about to be destroyed.
299 fdefree_last(struct filedescent *fde)
302 filecaps_free(&fde->fde_caps);
306 fdfree(struct filedesc *fdp, int fd)
308 struct filedescent *fde;
310 fde = &fdp->fd_ofiles[fd];
312 seqc_write_begin(&fde->fde_seqc);
314 fde->fde_file = NULL;
316 seqc_write_end(&fde->fde_seqc);
323 * System calls on descriptors.
325 #ifndef _SYS_SYSPROTO_H_
326 struct getdtablesize_args {
332 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
338 td->td_retval[0] = getmaxfd(td);
340 PROC_LOCK(td->td_proc);
341 lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
342 PROC_UNLOCK(td->td_proc);
343 if (lim < td->td_retval[0])
344 td->td_retval[0] = lim;
350 * Duplicate a file descriptor to a particular value.
352 * Note: keep in mind that a potential race condition exists when closing
353 * descriptors from a shared descriptor table (via rfork).
355 #ifndef _SYS_SYSPROTO_H_
363 sys_dup2(struct thread *td, struct dup2_args *uap)
366 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
370 * Duplicate a file descriptor.
372 #ifndef _SYS_SYSPROTO_H_
379 sys_dup(struct thread *td, struct dup_args *uap)
382 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
386 * The file control system call.
388 #ifndef _SYS_SYSPROTO_H_
397 sys_fcntl(struct thread *td, struct fcntl_args *uap)
400 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
404 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
418 * Convert old flock structure to new.
420 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
421 fl.l_start = ofl.l_start;
422 fl.l_len = ofl.l_len;
423 fl.l_pid = ofl.l_pid;
424 fl.l_type = ofl.l_type;
425 fl.l_whence = ofl.l_whence;
439 arg1 = (intptr_t)&fl;
445 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
446 arg1 = (intptr_t)&fl;
454 error = kern_fcntl(td, fd, newcmd, arg1);
457 if (cmd == F_OGETLK) {
458 ofl.l_start = fl.l_start;
459 ofl.l_len = fl.l_len;
460 ofl.l_pid = fl.l_pid;
461 ofl.l_type = fl.l_type;
462 ofl.l_whence = fl.l_whence;
463 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
464 } else if (cmd == F_GETLK) {
465 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
471 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
473 struct filedesc *fdp;
475 struct file *fp, *fp2;
476 struct filedescent *fde;
480 int error, flg, seals, tmp;
494 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
497 case F_DUPFD_CLOEXEC:
499 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
504 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
507 case F_DUP2FD_CLOEXEC:
509 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
515 fde = fdeget_locked(fdp, fd);
518 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
521 FILEDESC_SUNLOCK(fdp);
527 fde = fdeget_locked(fdp, fd);
529 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
530 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
533 FILEDESC_XUNLOCK(fdp);
537 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
540 td->td_retval[0] = OFLAGS(fp->f_flag);
545 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
549 tmp = flg = fp->f_flag;
551 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
552 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
553 tmp = fp->f_flag & FNONBLOCK;
554 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
559 tmp = fp->f_flag & FASYNC;
560 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
565 atomic_clear_int(&fp->f_flag, FNONBLOCK);
567 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
572 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
575 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
577 td->td_retval[0] = tmp;
582 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
586 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
591 error = priv_check(td, PRIV_NFS_LOCKD);
599 /* FALLTHROUGH F_SETLK */
603 flp = (struct flock *)arg;
604 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
609 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
612 if (fp->f_type != DTYPE_VNODE) {
618 if (flp->l_whence == SEEK_CUR) {
619 foffset = foffset_get(fp);
622 foffset > OFF_MAX - flp->l_start)) {
627 flp->l_start += foffset;
631 switch (flp->l_type) {
633 if ((fp->f_flag & FREAD) == 0) {
637 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
638 PROC_LOCK(p->p_leader);
639 p->p_leader->p_flag |= P_ADVLOCK;
640 PROC_UNLOCK(p->p_leader);
642 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
646 if ((fp->f_flag & FWRITE) == 0) {
650 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
651 PROC_LOCK(p->p_leader);
652 p->p_leader->p_flag |= P_ADVLOCK;
653 PROC_UNLOCK(p->p_leader);
655 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
659 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
663 if (flg != F_REMOTE) {
667 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
668 F_UNLCKSYS, flp, flg);
674 if (error != 0 || flp->l_type == F_UNLCK ||
675 flp->l_type == F_UNLCKSYS) {
681 * Check for a race with close.
683 * The vnode is now advisory locked (or unlocked, but this case
684 * is not really important) as the caller requested.
685 * We had to drop the filedesc lock, so we need to recheck if
686 * the descriptor is still valid, because if it was closed
687 * in the meantime we need to remove advisory lock from the
688 * vnode - close on any descriptor leading to an advisory
689 * locked vnode, removes that lock.
690 * We will return 0 on purpose in that case, as the result of
691 * successful advisory lock might have been externally visible
692 * already. This is fine - effectively we pretend to the caller
693 * that the closing thread was a bit slower and that the
694 * advisory lock succeeded before the close.
696 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
702 flp->l_whence = SEEK_SET;
705 flp->l_type = F_UNLCK;
706 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
707 F_UNLCK, flp, F_POSIX);
714 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
717 if (fp->f_type != DTYPE_VNODE) {
722 flp = (struct flock *)arg;
723 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
724 flp->l_type != F_UNLCK) {
729 if (flp->l_whence == SEEK_CUR) {
730 foffset = foffset_get(fp);
731 if ((flp->l_start > 0 &&
732 foffset > OFF_MAX - flp->l_start) ||
734 foffset < OFF_MIN - flp->l_start)) {
739 flp->l_start += foffset;
742 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
748 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
751 error = fo_add_seals(fp, arg);
756 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
759 if (fo_get_seals(fp, &seals) == 0)
760 td->td_retval[0] = seals;
767 arg = arg ? 128 * 1024: 0;
770 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
773 if (fp->f_type != DTYPE_VNODE) {
779 if (vp->v_type != VREG) {
786 * Exclusive lock synchronizes against f_seqcount reads and
787 * writes in sequential_heuristic().
789 error = vn_lock(vp, LK_EXCLUSIVE);
795 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
796 arg = MIN(arg, INT_MAX - bsize + 1);
797 fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
798 (arg + bsize - 1) / bsize);
799 atomic_set_int(&fp->f_flag, FRDAHEAD);
801 atomic_clear_int(&fp->f_flag, FRDAHEAD);
809 * Check if the vnode is part of a union stack (either the
810 * "union" flag from mount(2) or unionfs).
812 * Prior to introduction of this op libc's readdir would call
813 * fstatfs(2), in effect unnecessarily copying kilobytes of
814 * data just to check fs name and a mount flag.
816 * Fixing the code to handle everything in the kernel instead
817 * is a non-trivial endeavor and has low priority, thus this
818 * horrible kludge facilitates the current behavior in a much
819 * cheaper manner until someone(tm) sorts this out.
821 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
824 if (fp->f_type != DTYPE_VNODE) {
831 * Since we don't prevent dooming the vnode even non-null mp
832 * found can become immediately stale. This is tolerable since
833 * mount points are type-stable (providing safe memory access)
834 * and any vfs op on this vnode going forward will return an
835 * error (meaning return value in this case is meaningless).
837 mp = atomic_load_ptr(&vp->v_mount);
838 if (__predict_false(mp == NULL)) {
843 td->td_retval[0] = 0;
844 if (mp->mnt_kern_flag & MNTK_UNIONFS ||
845 mp->mnt_flag & MNT_UNION)
846 td->td_retval[0] = 1;
858 getmaxfd(struct thread *td)
861 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
865 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
868 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
870 struct filedesc *fdp;
871 struct filedescent *oldfde, *newfde;
873 struct file *delfp, *oldfp;
874 u_long *oioctls, *nioctls;
881 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
882 MPASS(mode < FDDUP_LASTMODE);
885 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
888 * Verify we have a valid descriptor to dup from and possibly to
889 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
890 * return EINVAL when the new descriptor is out of bounds.
895 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
896 maxfd = getmaxfd(td);
898 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
902 if (fget_locked(fdp, old) == NULL)
904 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
905 td->td_retval[0] = new;
906 if (flags & FDDUP_FLAG_CLOEXEC)
907 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
912 oldfde = &fdp->fd_ofiles[old];
913 oldfp = oldfde->fde_file;
918 * If the caller specified a file descriptor, make sure the file
919 * table is large enough to hold it, and grab it. Otherwise, just
920 * allocate a new descriptor the usual way.
925 if ((error = fdalloc(td, new, &new)) != 0) {
930 case FDDUP_MUSTREPLACE:
931 /* Target file descriptor must exist. */
932 if (fget_locked(fdp, new) == NULL) {
938 if (new >= fdp->fd_nfiles) {
940 * The resource limits are here instead of e.g.
941 * fdalloc(), because the file descriptor table may be
942 * shared between processes, so we can't really use
943 * racct_add()/racct_sub(). Instead of counting the
944 * number of actually allocated descriptors, just put
945 * the limit on the size of the file descriptor table.
948 if (RACCT_ENABLED()) {
949 error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
957 fdgrowtable_exp(fdp, new + 1);
959 if (!fdisused(fdp, new))
963 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
966 KASSERT(old != new, ("new fd is same as old"));
968 /* Refetch oldfde because the table may have grown and old one freed. */
969 oldfde = &fdp->fd_ofiles[old];
970 KASSERT(oldfp == oldfde->fde_file,
971 ("fdt_ofiles shift from growth observed at fd %d",
974 newfde = &fdp->fd_ofiles[new];
975 delfp = newfde->fde_file;
977 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
980 * Duplicate the source descriptor.
983 seqc_write_begin(&newfde->fde_seqc);
985 oioctls = filecaps_free_prep(&newfde->fde_caps);
986 memcpy(newfde, oldfde, fde_change_size);
987 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
989 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
990 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
992 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
994 seqc_write_end(&newfde->fde_seqc);
996 td->td_retval[0] = new;
1000 if (delfp != NULL) {
1001 (void) closefp(fdp, new, delfp, td, 1);
1002 FILEDESC_UNLOCK_ASSERT(fdp);
1005 FILEDESC_XUNLOCK(fdp);
1008 filecaps_free_finish(oioctls);
1013 sigiofree(struct sigio *sigio)
1015 crfree(sigio->sio_ucred);
1016 free(sigio, M_SIGIO);
1019 static struct sigio *
1020 funsetown_locked(struct sigio *sigio)
1025 SIGIO_ASSERT_LOCKED();
1029 *(sigio->sio_myref) = NULL;
1030 if (sigio->sio_pgid < 0) {
1031 pg = sigio->sio_pgrp;
1033 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
1034 sigio, sio_pgsigio);
1037 p = sigio->sio_proc;
1039 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
1040 sigio, sio_pgsigio);
1047 * If sigio is on the list associated with a process or process group,
1048 * disable signalling from the device, remove sigio from the list and
1052 funsetown(struct sigio **sigiop)
1054 struct sigio *sigio;
1056 /* Racy check, consumers must provide synchronization. */
1057 if (*sigiop == NULL)
1061 sigio = funsetown_locked(*sigiop);
1068 * Free a list of sigio structures. The caller must ensure that new sigio
1069 * structures cannot be added after this point. For process groups this is
1070 * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1074 funsetownlst(struct sigiolst *sigiolst)
1078 struct sigio *sigio, *tmp;
1081 sigio = SLIST_FIRST(sigiolst);
1089 sigio = SLIST_FIRST(sigiolst);
1090 if (sigio == NULL) {
1096 * Every entry of the list should belong to a single proc or pgrp.
1098 if (sigio->sio_pgid < 0) {
1099 pg = sigio->sio_pgrp;
1100 sx_assert(&proctree_lock, SX_XLOCKED);
1102 } else /* if (sigio->sio_pgid > 0) */ {
1103 p = sigio->sio_proc;
1105 KASSERT((p->p_flag & P_WEXIT) != 0,
1106 ("%s: process %p is not exiting", __func__, p));
1109 SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1110 *sigio->sio_myref = NULL;
1112 KASSERT(sigio->sio_pgid < 0,
1113 ("Proc sigio in pgrp sigio list"));
1114 KASSERT(sigio->sio_pgrp == pg,
1115 ("Bogus pgrp in sigio list"));
1116 } else /* if (p != NULL) */ {
1117 KASSERT(sigio->sio_pgid > 0,
1118 ("Pgrp sigio in proc sigio list"));
1119 KASSERT(sigio->sio_proc == p,
1120 ("Bogus proc in sigio list"));
1130 SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1135 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1137 * After permission checking, add a sigio structure to the sigio list for
1138 * the process or process group.
1141 fsetown(pid_t pgid, struct sigio **sigiop)
1145 struct sigio *osigio, *sigio;
1155 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1156 sigio->sio_pgid = pgid;
1157 sigio->sio_ucred = crhold(curthread->td_ucred);
1158 sigio->sio_myref = sigiop;
1160 sx_slock(&proctree_lock);
1162 osigio = funsetown_locked(*sigiop);
1171 * Policy - Don't allow a process to FSETOWN a process
1172 * in another session.
1174 * Remove this test to allow maximum flexibility or
1175 * restrict FSETOWN to the current process or process
1176 * group for maximum safety.
1178 if (proc->p_session != curthread->td_proc->p_session) {
1184 sigio->sio_proc = proc;
1185 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1187 } else /* if (pgid < 0) */ {
1188 pgrp = pgfind(-pgid);
1195 * Policy - Don't allow a process to FSETOWN a process
1196 * in another session.
1198 * Remove this test to allow maximum flexibility or
1199 * restrict FSETOWN to the current process or process
1200 * group for maximum safety.
1202 if (pgrp->pg_session != curthread->td_proc->p_session) {
1208 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1209 sigio->sio_pgrp = pgrp;
1212 sx_sunlock(&proctree_lock);
1221 sx_sunlock(&proctree_lock);
1229 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1232 fgetown(struct sigio **sigiop)
1237 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1243 * Function drops the filedesc lock on return.
1246 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1251 FILEDESC_XLOCK_ASSERT(fdp);
1254 if (td->td_proc->p_fdtol != NULL) {
1256 * Ask fdfree() to sleep to ensure that all relevant
1257 * process leaders can be traversed in closef().
1259 fdp->fd_holdleaderscount++;
1266 * We now hold the fp reference that used to be owned by the
1267 * descriptor array. We have to unlock the FILEDESC *AFTER*
1268 * knote_fdclose to prevent a race of the fd getting opened, a knote
1269 * added, and deleteing a knote for the new fd.
1271 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1272 knote_fdclose(td, fd);
1275 * We need to notify mqueue if the object is of type mqueue.
1277 if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1278 mq_fdclose(td, fd, fp);
1279 FILEDESC_XUNLOCK(fdp);
1281 error = closef(fp, td);
1284 * All paths leading up to closefp() will have already removed or
1285 * replaced the fd in the filedesc table, so a restart would not
1286 * operate on the same file.
1288 if (error == ERESTART)
1292 FILEDESC_XLOCK(fdp);
1293 fdp->fd_holdleaderscount--;
1294 if (fdp->fd_holdleaderscount == 0 &&
1295 fdp->fd_holdleaderswakeup != 0) {
1296 fdp->fd_holdleaderswakeup = 0;
1297 wakeup(&fdp->fd_holdleaderscount);
1299 FILEDESC_XUNLOCK(fdp);
1305 * Close a file descriptor.
1307 #ifndef _SYS_SYSPROTO_H_
1314 sys_close(struct thread *td, struct close_args *uap)
1317 return (kern_close(td, uap->fd));
1321 kern_close(struct thread *td, int fd)
1323 struct filedesc *fdp;
1326 fdp = td->td_proc->p_fd;
1328 AUDIT_SYSCLOSE(td, fd);
1330 FILEDESC_XLOCK(fdp);
1331 if ((fp = fget_locked(fdp, fd)) == NULL) {
1332 FILEDESC_XUNLOCK(fdp);
1337 /* closefp() drops the FILEDESC lock for us. */
1338 return (closefp(fdp, fd, fp, td, 1));
1342 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1344 struct filedesc *fdp;
1345 int fd, ret, lastfile;
1348 fdp = td->td_proc->p_fd;
1349 FILEDESC_SLOCK(fdp);
1352 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1353 * open should not be a usage error. From a close_range() perspective,
1354 * close_range(3, ~0U, 0) in the same scenario should also likely not
1355 * be a usage error as all fd above 3 are in-fact already closed.
1357 if (highfd < lowfd) {
1363 * If lastfile == -1, we're dealing with either a fresh file
1364 * table or one in which every fd has been closed. Just return
1365 * successful; there's nothing left to do.
1367 lastfile = fdlastfile(fdp);
1370 /* Clamped to [lowfd, lastfile] */
1371 highfd = MIN(highfd, lastfile);
1372 for (fd = lowfd; fd <= highfd; fd++) {
1373 if (fdp->fd_ofiles[fd].fde_file != NULL) {
1374 FILEDESC_SUNLOCK(fdp);
1375 (void)kern_close(td, fd);
1376 FILEDESC_SLOCK(fdp);
1380 FILEDESC_SUNLOCK(fdp);
1384 #ifndef _SYS_SYSPROTO_H_
1385 struct close_range_args {
1392 sys_close_range(struct thread *td, struct close_range_args *uap)
1395 /* No flags currently defined */
1396 if (uap->flags != 0)
1398 return (kern_close_range(td, uap->lowfd, uap->highfd));
1401 #ifdef COMPAT_FREEBSD12
1403 * Close open file descriptors.
1405 #ifndef _SYS_SYSPROTO_H_
1406 struct freebsd12_closefrom_args {
1412 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1416 AUDIT_ARG_FD(uap->lowfd);
1419 * Treat negative starting file descriptor values identical to
1420 * closefrom(0) which closes all files.
1422 lowfd = MAX(0, uap->lowfd);
1423 return (kern_close_range(td, lowfd, ~0U));
1425 #endif /* COMPAT_FREEBSD12 */
1427 #if defined(COMPAT_43)
1429 * Return status information about a file descriptor.
1431 #ifndef _SYS_SYSPROTO_H_
1432 struct ofstat_args {
1439 ofstat(struct thread *td, struct ofstat_args *uap)
1445 error = kern_fstat(td, uap->fd, &ub);
1448 error = copyout(&oub, uap->sb, sizeof(oub));
1452 #endif /* COMPAT_43 */
1454 #if defined(COMPAT_FREEBSD11)
1456 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1459 struct freebsd11_stat osb;
1462 error = kern_fstat(td, uap->fd, &sb);
1465 error = freebsd11_cvtstat(&sb, &osb);
1467 error = copyout(&osb, uap->sb, sizeof(osb));
1470 #endif /* COMPAT_FREEBSD11 */
1473 * Return status information about a file descriptor.
1475 #ifndef _SYS_SYSPROTO_H_
1483 sys_fstat(struct thread *td, struct fstat_args *uap)
1488 error = kern_fstat(td, uap->fd, &ub);
1490 error = copyout(&ub, uap->sb, sizeof(ub));
1495 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1502 error = fget(td, fd, &cap_fstat_rights, &fp);
1503 if (__predict_false(error != 0))
1506 AUDIT_ARG_FILE(td->td_proc, fp);
1508 error = fo_stat(fp, sbp, td->td_ucred, td);
1510 #ifdef __STAT_TIME_T_EXT
1511 sbp->st_atim_ext = 0;
1512 sbp->st_mtim_ext = 0;
1513 sbp->st_ctim_ext = 0;
1514 sbp->st_btim_ext = 0;
1517 if (KTRPOINT(td, KTR_STRUCT))
1518 ktrstat_error(sbp, error);
1523 #if defined(COMPAT_FREEBSD11)
1525 * Return status information about a file descriptor.
1527 #ifndef _SYS_SYSPROTO_H_
1528 struct freebsd11_nfstat_args {
1535 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1541 error = kern_fstat(td, uap->fd, &ub);
1543 freebsd11_cvtnstat(&ub, &nub);
1544 error = copyout(&nub, uap->sb, sizeof(nub));
1548 #endif /* COMPAT_FREEBSD11 */
1551 * Return pathconf information about a file descriptor.
1553 #ifndef _SYS_SYSPROTO_H_
1554 struct fpathconf_args {
1561 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1566 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1568 td->td_retval[0] = value;
1573 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1579 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1583 if (name == _PC_ASYNC_IO) {
1584 *valuep = _POSIX_ASYNCHRONOUS_IO;
1589 vn_lock(vp, LK_SHARED | LK_RETRY);
1590 error = VOP_PATHCONF(vp, name, valuep);
1592 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1593 if (name != _PC_PIPE_BUF) {
1608 * Copy filecaps structure allocating memory for ioctls array if needed.
1610 * The last parameter indicates whether the fdtable is locked. If it is not and
1611 * ioctls are encountered, copying fails and the caller must lock the table.
1613 * Note that if the table was not locked, the caller has to check the relevant
1614 * sequence counter to determine whether the operation was successful.
1617 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1621 if (src->fc_ioctls != NULL && !locked)
1623 memcpy(dst, src, sizeof(*src));
1624 if (src->fc_ioctls == NULL)
1627 KASSERT(src->fc_nioctls > 0,
1628 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1630 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1631 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1632 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1637 filecaps_copy_prep(const struct filecaps *src)
1642 if (__predict_true(src->fc_ioctls == NULL))
1645 KASSERT(src->fc_nioctls > 0,
1646 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1648 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1649 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1654 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1660 if (__predict_true(src->fc_ioctls == NULL)) {
1661 MPASS(ioctls == NULL);
1665 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1666 dst->fc_ioctls = ioctls;
1667 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1671 * Move filecaps structure to the new place and clear the old place.
1674 filecaps_move(struct filecaps *src, struct filecaps *dst)
1678 bzero(src, sizeof(*src));
1682 * Fill the given filecaps structure with full rights.
1685 filecaps_fill(struct filecaps *fcaps)
1688 CAP_ALL(&fcaps->fc_rights);
1689 fcaps->fc_ioctls = NULL;
1690 fcaps->fc_nioctls = -1;
1691 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1695 * Free memory allocated within filecaps structure.
1698 filecaps_free(struct filecaps *fcaps)
1701 free(fcaps->fc_ioctls, M_FILECAPS);
1702 bzero(fcaps, sizeof(*fcaps));
1706 filecaps_free_prep(struct filecaps *fcaps)
1710 ioctls = fcaps->fc_ioctls;
1711 bzero(fcaps, sizeof(*fcaps));
1716 filecaps_free_finish(u_long *ioctls)
1719 free(ioctls, M_FILECAPS);
1723 * Validate the given filecaps structure.
1726 filecaps_validate(const struct filecaps *fcaps, const char *func)
1729 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1730 ("%s: invalid rights", func));
1731 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1732 ("%s: invalid fcntls", func));
1733 KASSERT(fcaps->fc_fcntls == 0 ||
1734 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1735 ("%s: fcntls without CAP_FCNTL", func));
1736 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1737 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1738 ("%s: invalid ioctls", func));
1739 KASSERT(fcaps->fc_nioctls == 0 ||
1740 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1741 ("%s: ioctls without CAP_IOCTL", func));
1745 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1749 FILEDESC_XLOCK_ASSERT(fdp);
1751 nfd1 = fdp->fd_nfiles * 2;
1754 fdgrowtable(fdp, nfd1);
1758 * Grow the file table to accommodate (at least) nfd descriptors.
1761 fdgrowtable(struct filedesc *fdp, int nfd)
1763 struct filedesc0 *fdp0;
1764 struct freetable *ft;
1765 struct fdescenttbl *ntable;
1766 struct fdescenttbl *otable;
1767 int nnfiles, onfiles;
1768 NDSLOTTYPE *nmap, *omap;
1770 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1772 /* save old values */
1773 onfiles = fdp->fd_nfiles;
1774 otable = fdp->fd_files;
1777 /* compute the size of the new table */
1778 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1779 if (nnfiles <= onfiles)
1780 /* the table is already large enough */
1784 * Allocate a new table. We need enough space for the number of
1785 * entries, file entries themselves and the struct freetable we will use
1786 * when we decommission the table and place it on the freelist.
1787 * We place the struct freetable in the middle so we don't have
1788 * to worry about padding.
1790 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1791 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1792 sizeof(struct freetable),
1793 M_FILEDESC, M_ZERO | M_WAITOK);
1794 /* copy the old data */
1795 ntable->fdt_nfiles = nnfiles;
1796 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1797 onfiles * sizeof(ntable->fdt_ofiles[0]));
1800 * Allocate a new map only if the old is not large enough. It will
1801 * grow at a slower rate than the table as it can map more
1802 * entries than the table can hold.
1804 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1805 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1807 /* copy over the old data and update the pointer */
1808 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1813 * Make sure that ntable is correctly initialized before we replace
1814 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1817 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1820 * Free the old file table when not shared by other threads or processes.
1821 * The old file table is considered to be shared when either are true:
1822 * - The process has more than one thread.
1823 * - The file descriptor table has been shared via fdshare().
1825 * When shared, the old file table will be placed on a freelist
1826 * which will be processed when the struct filedesc is released.
1828 * Note that if onfiles == NDFILE, we're dealing with the original
1829 * static allocation contained within (struct filedesc0 *)fdp,
1830 * which must not be freed.
1832 if (onfiles > NDFILE) {
1833 if (curproc->p_numthreads == 1 &&
1834 refcount_load(&fdp->fd_refcnt) == 1)
1835 free(otable, M_FILEDESC);
1837 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1838 fdp0 = (struct filedesc0 *)fdp;
1839 ft->ft_table = otable;
1840 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1844 * The map does not have the same possibility of threads still
1845 * holding references to it. So always free it as long as it
1846 * does not reference the original static allocation.
1848 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1849 free(omap, M_FILEDESC);
1853 * Allocate a file descriptor for the process.
1856 fdalloc(struct thread *td, int minfd, int *result)
1858 struct proc *p = td->td_proc;
1859 struct filedesc *fdp = p->p_fd;
1860 int fd, maxfd, allocfd;
1865 FILEDESC_XLOCK_ASSERT(fdp);
1867 if (fdp->fd_freefile > minfd)
1868 minfd = fdp->fd_freefile;
1870 maxfd = getmaxfd(td);
1873 * Search the bitmap for a free descriptor starting at minfd.
1874 * If none is found, grow the file table.
1876 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1877 if (__predict_false(fd >= maxfd))
1879 if (__predict_false(fd >= fdp->fd_nfiles)) {
1880 allocfd = min(fd * 2, maxfd);
1882 if (RACCT_ENABLED()) {
1883 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1889 * fd is already equal to first free descriptor >= minfd, so
1890 * we only need to grow the table and we are done.
1892 fdgrowtable_exp(fdp, allocfd);
1896 * Perform some sanity checks, then mark the file descriptor as
1897 * used and return it to the caller.
1899 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1900 ("invalid descriptor %d", fd));
1901 KASSERT(!fdisused(fdp, fd),
1902 ("fd_first_free() returned non-free descriptor"));
1903 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1904 ("file descriptor isn't free"));
1911 * Allocate n file descriptors for the process.
1914 fdallocn(struct thread *td, int minfd, int *fds, int n)
1916 struct proc *p = td->td_proc;
1917 struct filedesc *fdp = p->p_fd;
1920 FILEDESC_XLOCK_ASSERT(fdp);
1922 for (i = 0; i < n; i++)
1923 if (fdalloc(td, 0, &fds[i]) != 0)
1927 for (i--; i >= 0; i--)
1928 fdunused(fdp, fds[i]);
1936 * Create a new open file structure and allocate a file descriptor for the
1937 * process that refers to it. We add one reference to the file for the
1938 * descriptor table and one reference for resultfp. This is to prevent us
1939 * being preempted and the entry in the descriptor table closed after we
1940 * release the FILEDESC lock.
1943 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1944 struct filecaps *fcaps)
1949 error = falloc_noinstall(td, &fp);
1951 return (error); /* no reference held on error */
1953 error = finstall(td, fp, &fd, flags, fcaps);
1955 fdrop(fp, td); /* one reference (fp only) */
1959 if (resultfp != NULL)
1960 *resultfp = fp; /* copy out result */
1962 fdrop(fp, td); /* release local reference */
1964 if (resultfd != NULL)
1971 * Create a new open file structure without allocating a file descriptor.
1974 falloc_noinstall(struct thread *td, struct file **resultfp)
1977 int maxuserfiles = maxfiles - (maxfiles / 20);
1979 static struct timeval lastfail;
1982 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1984 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1985 if ((openfiles_new >= maxuserfiles &&
1986 priv_check(td, PRIV_MAXFILES) != 0) ||
1987 openfiles_new >= maxfiles) {
1988 atomic_subtract_int(&openfiles, 1);
1989 if (ppsratecheck(&lastfail, &curfail, 1)) {
1990 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1991 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1995 fp = uma_zalloc(file_zone, M_WAITOK);
1996 bzero(fp, sizeof(*fp));
1997 refcount_init(&fp->f_count, 1);
1998 fp->f_cred = crhold(td->td_ucred);
1999 fp->f_ops = &badfileops;
2005 * Install a file in a file descriptor table.
2008 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2009 struct filecaps *fcaps)
2011 struct filedescent *fde;
2015 filecaps_validate(fcaps, __func__);
2016 FILEDESC_XLOCK_ASSERT(fdp);
2018 fde = &fdp->fd_ofiles[fd];
2020 seqc_write_begin(&fde->fde_seqc);
2023 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2025 filecaps_move(fcaps, &fde->fde_caps);
2027 filecaps_fill(&fde->fde_caps);
2029 seqc_write_end(&fde->fde_seqc);
2034 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2035 struct filecaps *fcaps)
2037 struct filedesc *fdp = td->td_proc->p_fd;
2044 FILEDESC_XLOCK(fdp);
2045 error = fdalloc(td, 0, fd);
2046 if (__predict_false(error != 0)) {
2047 FILEDESC_XUNLOCK(fdp);
2051 _finstall(fdp, fp, *fd, flags, fcaps);
2052 FILEDESC_XUNLOCK(fdp);
2057 * Build a new filedesc structure from another.
2059 * If fdp is not NULL, return with it shared locked.
2062 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2064 struct filedesc0 *newfdp0;
2065 struct filedesc *newfdp;
2068 MPASS(lastfile != NULL);
2070 MPASS(lastfile == NULL);
2072 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2073 newfdp = &newfdp0->fd_fd;
2075 /* Create the file descriptor table. */
2076 FILEDESC_LOCK_INIT(newfdp);
2077 refcount_init(&newfdp->fd_refcnt, 1);
2078 refcount_init(&newfdp->fd_holdcnt, 1);
2079 newfdp->fd_map = newfdp0->fd_dmap;
2080 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2081 newfdp->fd_files->fdt_nfiles = NDFILE;
2086 FILEDESC_SLOCK(fdp);
2088 FILEDESC_SUNLOCK(fdp);
2093 *lastfile = fdlastfile(fdp);
2094 if (*lastfile < newfdp->fd_nfiles)
2096 FILEDESC_SUNLOCK(fdp);
2097 fdgrowtable(newfdp, *lastfile + 1);
2098 FILEDESC_SLOCK(fdp);
2105 * Build a pwddesc structure from another.
2106 * Copy the current, root, and jail root vnode references.
2108 * If pdp is not NULL, return with it shared locked.
2111 pdinit(struct pwddesc *pdp, bool keeplock)
2113 struct pwddesc *newpdp;
2116 newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2118 PWDDESC_LOCK_INIT(newpdp);
2119 refcount_init(&newpdp->pd_refcount, 1);
2120 newpdp->pd_cmask = CMASK;
2123 newpwd = pwd_alloc();
2124 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2129 newpwd = pwd_hold_pwddesc(pdp);
2130 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2132 PWDDESC_XUNLOCK(pdp);
2136 static struct filedesc *
2137 fdhold(struct proc *p)
2139 struct filedesc *fdp;
2141 PROC_LOCK_ASSERT(p, MA_OWNED);
2144 refcount_acquire(&fdp->fd_holdcnt);
2148 static struct pwddesc *
2149 pdhold(struct proc *p)
2151 struct pwddesc *pdp;
2153 PROC_LOCK_ASSERT(p, MA_OWNED);
2156 refcount_acquire(&pdp->pd_refcount);
2161 fddrop(struct filedesc *fdp)
2164 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2165 if (refcount_release(&fdp->fd_holdcnt) == 0)
2169 FILEDESC_LOCK_DESTROY(fdp);
2170 uma_zfree(filedesc0_zone, fdp);
2174 pddrop(struct pwddesc *pdp)
2178 if (refcount_release_if_not_last(&pdp->pd_refcount))
2182 if (refcount_release(&pdp->pd_refcount) == 0) {
2183 PWDDESC_XUNLOCK(pdp);
2186 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2188 PWDDESC_XUNLOCK(pdp);
2191 PWDDESC_LOCK_DESTROY(pdp);
2192 free(pdp, M_PWDDESC);
2196 * Share a filedesc structure.
2199 fdshare(struct filedesc *fdp)
2202 refcount_acquire(&fdp->fd_refcnt);
2207 * Share a pwddesc structure.
2210 pdshare(struct pwddesc *pdp)
2212 refcount_acquire(&pdp->pd_refcount);
2217 * Unshare a filedesc structure, if necessary by making a copy
2220 fdunshare(struct thread *td)
2222 struct filedesc *tmp;
2223 struct proc *p = td->td_proc;
2225 if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2228 tmp = fdcopy(p->p_fd);
2234 * Unshare a pwddesc structure.
2237 pdunshare(struct thread *td)
2239 struct pwddesc *pdp;
2244 if (p->p_pd->pd_refcount == 1)
2247 pdp = pdcopy(p->p_pd);
2253 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2257 td->td_proc->p_fd = fdp;
2261 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2262 * this is to ease callers, not catch errors.
2265 fdcopy(struct filedesc *fdp)
2267 struct filedesc *newfdp;
2268 struct filedescent *nfde, *ofde;
2273 newfdp = fdinit(fdp, true, &lastfile);
2274 /* copy all passable descriptors (i.e. not kqueue) */
2275 newfdp->fd_freefile = -1;
2276 for (i = 0; i <= lastfile; ++i) {
2277 ofde = &fdp->fd_ofiles[i];
2278 if (ofde->fde_file == NULL ||
2279 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2280 !fhold(ofde->fde_file)) {
2281 if (newfdp->fd_freefile == -1)
2282 newfdp->fd_freefile = i;
2285 nfde = &newfdp->fd_ofiles[i];
2287 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2288 fdused_init(newfdp, i);
2290 if (newfdp->fd_freefile == -1)
2291 newfdp->fd_freefile = i;
2292 FILEDESC_SUNLOCK(fdp);
2297 * Copy a pwddesc structure.
2300 pdcopy(struct pwddesc *pdp)
2302 struct pwddesc *newpdp;
2306 newpdp = pdinit(pdp, true);
2307 newpdp->pd_cmask = pdp->pd_cmask;
2308 PWDDESC_XUNLOCK(pdp);
2313 * Copies a filedesc structure, while remapping all file descriptors
2314 * stored inside using a translation table.
2316 * File descriptors are copied over to the new file descriptor table,
2317 * regardless of whether the close-on-exec flag is set.
2320 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2321 struct filedesc **ret)
2323 struct filedesc *newfdp;
2324 struct filedescent *nfde, *ofde;
2325 int error, i, lastfile;
2329 newfdp = fdinit(fdp, true, &lastfile);
2330 if (nfds > lastfile + 1) {
2331 /* New table cannot be larger than the old one. */
2335 /* Copy all passable descriptors (i.e. not kqueue). */
2336 newfdp->fd_freefile = nfds;
2337 for (i = 0; i < nfds; ++i) {
2338 if (fds[i] < 0 || fds[i] > lastfile) {
2339 /* File descriptor out of bounds. */
2343 ofde = &fdp->fd_ofiles[fds[i]];
2344 if (ofde->fde_file == NULL) {
2345 /* Unused file descriptor. */
2349 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2350 /* File descriptor cannot be passed. */
2354 if (!fhold(ofde->fde_file)) {
2358 nfde = &newfdp->fd_ofiles[i];
2360 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2361 fdused_init(newfdp, i);
2363 FILEDESC_SUNLOCK(fdp);
2367 FILEDESC_SUNLOCK(fdp);
2368 fdescfree_remapped(newfdp);
2373 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2374 * one of processes using it exits) and the table used to be shared.
2377 fdclearlocks(struct thread *td)
2379 struct filedesc *fdp;
2380 struct filedesc_to_leader *fdtol;
2390 MPASS(fdtol != NULL);
2392 FILEDESC_XLOCK(fdp);
2393 KASSERT(fdtol->fdl_refcount > 0,
2394 ("filedesc_to_refcount botch: fdl_refcount=%d",
2395 fdtol->fdl_refcount));
2396 if (fdtol->fdl_refcount == 1 &&
2397 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2398 lastfile = fdlastfile(fdp);
2399 for (i = 0; i <= lastfile; i++) {
2400 fp = fdp->fd_ofiles[i].fde_file;
2401 if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2404 FILEDESC_XUNLOCK(fdp);
2405 lf.l_whence = SEEK_SET;
2408 lf.l_type = F_UNLCK;
2410 (void) VOP_ADVLOCK(vp,
2411 (caddr_t)p->p_leader, F_UNLCK,
2413 FILEDESC_XLOCK(fdp);
2418 if (fdtol->fdl_refcount == 1) {
2419 if (fdp->fd_holdleaderscount > 0 &&
2420 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2422 * close() or kern_dup() has cleared a reference
2423 * in a shared file descriptor table.
2425 fdp->fd_holdleaderswakeup = 1;
2426 sx_sleep(&fdp->fd_holdleaderscount,
2427 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2430 if (fdtol->fdl_holdcount > 0) {
2432 * Ensure that fdtol->fdl_leader remains
2433 * valid in closef().
2435 fdtol->fdl_wakeup = 1;
2436 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2441 fdtol->fdl_refcount--;
2442 if (fdtol->fdl_refcount == 0 &&
2443 fdtol->fdl_holdcount == 0) {
2444 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2445 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2449 FILEDESC_XUNLOCK(fdp);
2451 free(fdtol, M_FILEDESC_TO_LEADER);
2455 * Release a filedesc structure.
2458 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2460 struct filedesc0 *fdp0;
2461 struct freetable *ft, *tft;
2462 struct filedescent *fde;
2466 KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2467 ("%s: fd table %p carries references", __func__, fdp));
2470 * Serialize with threads iterating over the table, if any.
2472 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2473 FILEDESC_XLOCK(fdp);
2474 FILEDESC_XUNLOCK(fdp);
2477 lastfile = fdlastfile_single(fdp);
2478 for (i = 0; i <= lastfile; i++) {
2479 fde = &fdp->fd_ofiles[i];
2484 (void) closef(fp, td);
2490 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2491 free(fdp->fd_map, M_FILEDESC);
2492 if (fdp->fd_nfiles > NDFILE)
2493 free(fdp->fd_files, M_FILEDESC);
2495 fdp0 = (struct filedesc0 *)fdp;
2496 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2497 free(ft->ft_table, M_FILEDESC);
2503 fdescfree(struct thread *td)
2506 struct filedesc *fdp;
2513 if (RACCT_ENABLED())
2514 racct_set_unlocked(p, RACCT_NOFILE, 0);
2517 if (p->p_fdtol != NULL)
2524 if (refcount_release(&fdp->fd_refcnt) == 0)
2527 fdescfree_fds(td, fdp, 1);
2531 pdescfree(struct thread *td)
2534 struct pwddesc *pdp;
2548 fdescfree_remapped(struct filedesc *fdp)
2551 /* fdescfree_fds() asserts that fd_refcnt == 0. */
2552 if (!refcount_release(&fdp->fd_refcnt))
2553 panic("%s: fd table %p has extra references", __func__, fdp);
2555 fdescfree_fds(curthread, fdp, 0);
2559 * For setugid programs, we don't want to people to use that setugidness
2560 * to generate error messages which write to a file which otherwise would
2561 * otherwise be off-limits to the process. We check for filesystems where
2562 * the vnode can change out from under us after execve (like [lin]procfs).
2564 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2565 * sufficient. We also don't check for setugidness since we know we are.
2568 is_unsafe(struct file *fp)
2572 if (fp->f_type != DTYPE_VNODE)
2576 return ((vp->v_vflag & VV_PROCDEP) != 0);
2580 * Make this setguid thing safe, if at all possible.
2583 fdsetugidsafety(struct thread *td)
2585 struct filedesc *fdp;
2589 fdp = td->td_proc->p_fd;
2590 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2591 ("the fdtable should not be shared"));
2592 MPASS(fdp->fd_nfiles >= 3);
2593 for (i = 0; i <= 2; i++) {
2594 fp = fdp->fd_ofiles[i].fde_file;
2595 if (fp != NULL && is_unsafe(fp)) {
2596 FILEDESC_XLOCK(fdp);
2597 knote_fdclose(td, i);
2599 * NULL-out descriptor prior to close to avoid
2600 * a race while close blocks.
2603 FILEDESC_XUNLOCK(fdp);
2604 (void) closef(fp, td);
2610 * If a specific file object occupies a specific file descriptor, close the
2611 * file descriptor entry and drop a reference on the file object. This is a
2612 * convenience function to handle a subsequent error in a function that calls
2613 * falloc() that handles the race that another thread might have closed the
2614 * file descriptor out from under the thread creating the file object.
2617 fdclose(struct thread *td, struct file *fp, int idx)
2619 struct filedesc *fdp = td->td_proc->p_fd;
2621 FILEDESC_XLOCK(fdp);
2622 if (fdp->fd_ofiles[idx].fde_file == fp) {
2624 FILEDESC_XUNLOCK(fdp);
2627 FILEDESC_XUNLOCK(fdp);
2631 * Close any files on exec?
2634 fdcloseexec(struct thread *td)
2636 struct filedesc *fdp;
2637 struct filedescent *fde;
2641 fdp = td->td_proc->p_fd;
2642 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2643 ("the fdtable should not be shared"));
2644 lastfile = fdlastfile_single(fdp);
2645 for (i = 0; i <= lastfile; i++) {
2646 fde = &fdp->fd_ofiles[i];
2648 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2649 (fde->fde_flags & UF_EXCLOSE))) {
2650 FILEDESC_XLOCK(fdp);
2652 (void) closefp(fdp, i, fp, td, 0);
2653 FILEDESC_UNLOCK_ASSERT(fdp);
2659 * It is unsafe for set[ug]id processes to be started with file
2660 * descriptors 0..2 closed, as these descriptors are given implicit
2661 * significance in the Standard C library. fdcheckstd() will create a
2662 * descriptor referencing /dev/null for each of stdin, stdout, and
2663 * stderr that is not already open.
2666 fdcheckstd(struct thread *td)
2668 struct filedesc *fdp;
2670 int i, error, devnull;
2672 fdp = td->td_proc->p_fd;
2673 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2674 ("the fdtable should not be shared"));
2675 MPASS(fdp->fd_nfiles >= 3);
2677 for (i = 0; i <= 2; i++) {
2678 if (fdp->fd_ofiles[i].fde_file != NULL)
2681 save = td->td_retval[0];
2682 if (devnull != -1) {
2683 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2685 error = kern_openat(td, AT_FDCWD, "/dev/null",
2686 UIO_SYSSPACE, O_RDWR, 0);
2688 devnull = td->td_retval[0];
2689 KASSERT(devnull == i, ("we didn't get our fd"));
2692 td->td_retval[0] = save;
2700 * Internal form of close. Decrement reference count on file structure.
2701 * Note: td may be NULL when closing a file that was being passed in a
2705 closef(struct file *fp, struct thread *td)
2709 struct filedesc_to_leader *fdtol;
2710 struct filedesc *fdp;
2713 * POSIX record locking dictates that any close releases ALL
2714 * locks owned by this process. This is handled by setting
2715 * a flag in the unlock to free ONLY locks obeying POSIX
2716 * semantics, and not to free BSD-style file locks.
2717 * If the descriptor was in a message, POSIX-style locks
2718 * aren't passed with the descriptor, and the thread pointer
2719 * will be NULL. Callers should be careful only to pass a
2720 * NULL thread pointer when there really is no owning
2721 * context that might have locks, or the locks will be
2724 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2726 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2727 lf.l_whence = SEEK_SET;
2730 lf.l_type = F_UNLCK;
2731 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2732 F_UNLCK, &lf, F_POSIX);
2734 fdtol = td->td_proc->p_fdtol;
2735 if (fdtol != NULL) {
2737 * Handle special case where file descriptor table is
2738 * shared between multiple process leaders.
2740 fdp = td->td_proc->p_fd;
2741 FILEDESC_XLOCK(fdp);
2742 for (fdtol = fdtol->fdl_next;
2743 fdtol != td->td_proc->p_fdtol;
2744 fdtol = fdtol->fdl_next) {
2745 if ((fdtol->fdl_leader->p_flag &
2748 fdtol->fdl_holdcount++;
2749 FILEDESC_XUNLOCK(fdp);
2750 lf.l_whence = SEEK_SET;
2753 lf.l_type = F_UNLCK;
2755 (void) VOP_ADVLOCK(vp,
2756 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2758 FILEDESC_XLOCK(fdp);
2759 fdtol->fdl_holdcount--;
2760 if (fdtol->fdl_holdcount == 0 &&
2761 fdtol->fdl_wakeup != 0) {
2762 fdtol->fdl_wakeup = 0;
2766 FILEDESC_XUNLOCK(fdp);
2769 return (fdrop_close(fp, td));
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_any(fd_seqc(fdt, fd));
2897 if (__predict_false(seqc_in_modify(seq)))
2899 fde = &fdt->fdt_ofiles[fd];
2900 haverights = cap_rights_fde_inline(fde);
2902 if (__predict_false(fp == NULL))
2904 if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2906 *fsearch = ((fp->f_flag & FSEARCH) != 0);
2908 if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2911 if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2915 * Use an acquire barrier to force re-reading of fdt so it is
2916 * refreshed for verification.
2918 atomic_thread_fence_acq();
2919 fdt = fdp->fd_files;
2920 if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
2923 * If file descriptor doesn't have all rights,
2924 * all lookups relative to it must also be
2925 * strictly relative.
2927 * Not yet supported by fast path.
2930 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
2931 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
2932 ndp->ni_filecaps.fc_nioctls != -1) {
2934 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
2944 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2946 const struct fdescenttbl *fdt;
2947 struct filedesc *fdp;
2951 VFS_SMR_ASSERT_ENTERED();
2953 fdp = curproc->p_fd;
2954 fdt = fdp->fd_files;
2955 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2957 fp = fdt->fdt_ofiles[fd].fde_file;
2958 if (__predict_false(fp == NULL))
2960 *fsearch = ((fp->f_flag & FSEARCH) != 0);
2962 if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2966 * Use an acquire barrier to force re-reading of fdt so it is
2967 * refreshed for verification.
2969 atomic_thread_fence_acq();
2970 fdt = fdp->fd_files;
2971 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
2973 filecaps_fill(&ndp->ni_filecaps);
2980 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2981 struct file **fpp, seqc_t *seqp)
2984 const struct filedescent *fde;
2986 const struct fdescenttbl *fdt;
2990 cap_rights_t haverights;
2994 fdt = fdp->fd_files;
2995 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2998 * Fetch the descriptor locklessly. We avoid fdrop() races by
2999 * never raising a refcount above 0. To accomplish this we have
3000 * to use a cmpset loop rather than an atomic_add. The descriptor
3001 * must be re-verified once we acquire a reference to be certain
3002 * that the identity is still correct and we did not lose a race
3003 * due to preemption.
3007 seq = seqc_read(fd_seqc(fdt, fd));
3008 fde = &fdt->fdt_ofiles[fd];
3009 haverights = *cap_rights_fde_inline(fde);
3011 if (!seqc_consistent(fd_seqc(fdt, fd), seq))
3014 fp = fdt->fdt_ofiles[fd].fde_file;
3019 error = cap_check_inline(&haverights, needrightsp);
3023 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3025 * Force a reload. Other thread could reallocate the
3026 * table before this fd was closed, so it is possible
3027 * that there is a stale fp pointer in cached version.
3029 fdt = atomic_load_ptr(&fdp->fd_files);
3033 * Use an acquire barrier to force re-reading of fdt so it is
3034 * refreshed for verification.
3036 atomic_thread_fence_acq();
3037 fdt = fdp->fd_files;
3039 if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
3041 if (fp == fdt->fdt_ofiles[fd].fde_file)
3044 fdrop(fp, curthread);
3056 * See the comments in fget_unlocked_seq for an explanation of how this works.
3058 * This is a simplified variant which bails out to the aforementioned routine
3059 * if anything goes wrong. In practice this only happens when userspace is
3060 * racing with itself.
3063 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3067 const struct filedescent *fde;
3069 const struct fdescenttbl *fdt;
3073 const cap_rights_t *haverights;
3076 fdt = fdp->fd_files;
3077 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3080 seq = seqc_read_any(fd_seqc(fdt, fd));
3081 if (__predict_false(seqc_in_modify(seq)))
3083 fde = &fdt->fdt_ofiles[fd];
3084 haverights = cap_rights_fde_inline(fde);
3087 fp = fdt->fdt_ofiles[fd].fde_file;
3089 if (__predict_false(fp == NULL))
3092 if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3095 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3099 * Use an acquire barrier to force re-reading of fdt so it is
3100 * refreshed for verification.
3102 atomic_thread_fence_acq();
3103 fdt = fdp->fd_files;
3105 if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3107 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3113 fdrop(fp, curthread);
3115 return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
3119 * Extract the file pointer associated with the specified descriptor for the
3120 * current user process.
3122 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3125 * File's rights will be checked against the capability rights mask.
3127 * If an error occurred the non-zero error is returned and *fpp is set to
3128 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
3129 * responsible for fdrop().
3132 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3133 cap_rights_t *needrightsp)
3135 struct filedesc *fdp;
3140 fdp = td->td_proc->p_fd;
3141 error = fget_unlocked(fdp, fd, needrightsp, &fp);
3142 if (__predict_false(error != 0))
3144 if (__predict_false(fp->f_ops == &badfileops)) {
3150 * FREAD and FWRITE failure return EBADF as per POSIX.
3156 if ((fp->f_flag & flags) == 0)
3160 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3161 ((fp->f_flag & FWRITE) != 0))
3167 KASSERT(0, ("wrong flags"));
3180 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3183 return (_fget(td, fd, fpp, 0, rightsp));
3187 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3191 #ifndef CAPABILITIES
3192 error = _fget(td, fd, fpp, 0, rightsp);
3193 if (maxprotp != NULL)
3194 *maxprotp = VM_PROT_ALL;
3197 cap_rights_t fdrights;
3198 struct filedesc *fdp;
3203 fdp = td->td_proc->p_fd;
3204 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3206 error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3207 if (__predict_false(error != 0))
3209 if (__predict_false(fp->f_ops == &badfileops)) {
3213 if (maxprotp != NULL)
3214 fdrights = *cap_rights(fdp, fd);
3215 if (!fd_modified(fdp, fd, seq))
3221 * If requested, convert capability rights to access flags.
3223 if (maxprotp != NULL)
3224 *maxprotp = cap_rights_to_vmprot(&fdrights);
3231 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3234 return (_fget(td, fd, fpp, FREAD, rightsp));
3238 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3241 return (_fget(td, fd, fpp, FWRITE, rightsp));
3245 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3248 struct filedesc *fdp = td->td_proc->p_fd;
3249 #ifndef CAPABILITIES
3250 return (fget_unlocked(fdp, fd, rightsp, fpp));
3257 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3259 error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3262 error = cap_fcntl_check(fdp, fd, needfcntl);
3263 if (!fd_modified(fdp, fd, seq))
3277 * Like fget() but loads the underlying vnode, or returns an error if the
3278 * descriptor does not represent a vnode. Note that pipes use vnodes but
3279 * never have VM objects. The returned vnode will be vref()'d.
3281 * XXX: what about the unused flags ?
3284 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3291 error = _fget(td, fd, &fp, flags, needrightsp);
3294 if (fp->f_vnode == NULL) {
3306 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3309 return (_fgetvp(td, fd, 0, rightsp, vpp));
3313 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3314 struct filecaps *havecaps, struct vnode **vpp)
3316 struct filecaps caps;
3320 error = fget_cap(td, fd, needrightsp, &fp, &caps);
3323 if (fp->f_ops == &badfileops) {
3327 if (fp->f_vnode == NULL) {
3339 filecaps_free(&caps);
3345 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3348 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3352 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3355 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3360 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3364 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3369 * Handle the last reference to a file being closed.
3371 * Without the noinline attribute clang keeps inlining the func thorough this
3372 * file when fdrop is used.
3375 _fdrop(struct file *fp, struct thread *td)
3381 count = refcount_load(&fp->f_count);
3383 panic("fdrop: fp %p count %d", fp, count);
3385 error = fo_close(fp, td);
3386 atomic_subtract_int(&openfiles, 1);
3388 free(fp->f_advice, M_FADVISE);
3389 uma_zfree(file_zone, fp);
3395 * Apply an advisory lock on a file descriptor.
3397 * Just attempt to get a record lock of the requested type on the entire file
3398 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3400 #ifndef _SYS_SYSPROTO_H_
3408 sys_flock(struct thread *td, struct flock_args *uap)
3415 error = fget(td, uap->fd, &cap_flock_rights, &fp);
3418 if (fp->f_type != DTYPE_VNODE) {
3420 return (EOPNOTSUPP);
3424 lf.l_whence = SEEK_SET;
3427 if (uap->how & LOCK_UN) {
3428 lf.l_type = F_UNLCK;
3429 atomic_clear_int(&fp->f_flag, FHASLOCK);
3430 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3433 if (uap->how & LOCK_EX)
3434 lf.l_type = F_WRLCK;
3435 else if (uap->how & LOCK_SH)
3436 lf.l_type = F_RDLCK;
3441 atomic_set_int(&fp->f_flag, FHASLOCK);
3442 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3443 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3449 * Duplicate the specified descriptor to a free descriptor.
3452 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3453 int openerror, int *indxp)
3455 struct filedescent *newfde, *oldfde;
3460 KASSERT(openerror == ENODEV || openerror == ENXIO,
3461 ("unexpected error %d in %s", openerror, __func__));
3464 * If the to-be-dup'd fd number is greater than the allowed number
3465 * of file descriptors, or the fd to be dup'd has already been
3466 * closed, then reject.
3468 FILEDESC_XLOCK(fdp);
3469 if ((fp = fget_locked(fdp, dfd)) == NULL) {
3470 FILEDESC_XUNLOCK(fdp);
3474 error = fdalloc(td, 0, &indx);
3476 FILEDESC_XUNLOCK(fdp);
3481 * There are two cases of interest here.
3483 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3485 * For ENXIO steal away the file structure from (dfd) and store it in
3486 * (indx). (dfd) is effectively closed by this operation.
3488 switch (openerror) {
3491 * Check that the mode the file is being opened for is a
3492 * subset of the mode of the existing descriptor.
3494 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3495 fdunused(fdp, indx);
3496 FILEDESC_XUNLOCK(fdp);
3500 fdunused(fdp, indx);
3501 FILEDESC_XUNLOCK(fdp);
3504 newfde = &fdp->fd_ofiles[indx];
3505 oldfde = &fdp->fd_ofiles[dfd];
3506 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3508 seqc_write_begin(&newfde->fde_seqc);
3510 memcpy(newfde, oldfde, fde_change_size);
3511 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3514 seqc_write_end(&newfde->fde_seqc);
3519 * Steal away the file pointer from dfd and stuff it into indx.
3521 newfde = &fdp->fd_ofiles[indx];
3522 oldfde = &fdp->fd_ofiles[dfd];
3524 seqc_write_begin(&newfde->fde_seqc);
3526 memcpy(newfde, oldfde, fde_change_size);
3527 oldfde->fde_file = NULL;
3530 seqc_write_end(&newfde->fde_seqc);
3534 FILEDESC_XUNLOCK(fdp);
3540 * This sysctl determines if we will allow a process to chroot(2) if it
3541 * has a directory open:
3542 * 0: disallowed for all processes.
3543 * 1: allowed for processes that were not already chroot(2)'ed.
3544 * 2: allowed for all processes.
3547 static int chroot_allow_open_directories = 1;
3549 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3550 &chroot_allow_open_directories, 0,
3551 "Allow a process to chroot(2) if it has a directory open");
3554 * Helper function for raised chroot(2) security function: Refuse if
3555 * any filedescriptors are open directories.
3558 chroot_refuse_vdir_fds(struct filedesc *fdp)
3564 FILEDESC_LOCK_ASSERT(fdp);
3566 lastfile = fdlastfile(fdp);
3567 for (fd = 0; fd <= lastfile; fd++) {
3568 fp = fget_locked(fdp, fd);
3571 if (fp->f_type == DTYPE_VNODE) {
3573 if (vp->v_type == VDIR)
3581 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3584 if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3585 vrefact(oldpwd->pwd_cdir);
3586 newpwd->pwd_cdir = oldpwd->pwd_cdir;
3589 if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3590 vrefact(oldpwd->pwd_rdir);
3591 newpwd->pwd_rdir = oldpwd->pwd_rdir;
3594 if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3595 vrefact(oldpwd->pwd_jdir);
3596 newpwd->pwd_jdir = oldpwd->pwd_jdir;
3601 pwd_hold_pwddesc(struct pwddesc *pdp)
3605 PWDDESC_ASSERT_XLOCKED(pdp);
3606 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3608 refcount_acquire(&pwd->pwd_refcount);
3613 pwd_hold_smr(struct pwd *pwd)
3617 if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3624 pwd_hold(struct thread *td)
3626 struct pwddesc *pdp;
3629 pdp = td->td_proc->p_pd;
3632 pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3633 if (pwd_hold_smr(pwd)) {
3639 pwd = pwd_hold_pwddesc(pdp);
3641 PWDDESC_XUNLOCK(pdp);
3650 pwd = vfs_smr_entered_load(&curproc->p_pd->pd_pwd);
3660 pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3661 bzero(pwd, sizeof(*pwd));
3662 refcount_init(&pwd->pwd_refcount, 1);
3667 pwd_drop(struct pwd *pwd)
3670 if (!refcount_release(&pwd->pwd_refcount))
3673 if (pwd->pwd_cdir != NULL)
3674 vrele(pwd->pwd_cdir);
3675 if (pwd->pwd_rdir != NULL)
3676 vrele(pwd->pwd_rdir);
3677 if (pwd->pwd_jdir != NULL)
3678 vrele(pwd->pwd_jdir);
3679 uma_zfree_smr(pwd_zone, pwd);
3683 * Common routine for kern_chroot() and jail_attach(). The caller is
3684 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3685 * authorize this operation.
3688 pwd_chroot(struct thread *td, struct vnode *vp)
3690 struct pwddesc *pdp;
3691 struct filedesc *fdp;
3692 struct pwd *newpwd, *oldpwd;
3695 fdp = td->td_proc->p_fd;
3696 pdp = td->td_proc->p_pd;
3697 newpwd = pwd_alloc();
3698 FILEDESC_SLOCK(fdp);
3700 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3701 if (chroot_allow_open_directories == 0 ||
3702 (chroot_allow_open_directories == 1 &&
3703 oldpwd->pwd_rdir != rootvnode)) {
3704 error = chroot_refuse_vdir_fds(fdp);
3705 FILEDESC_SUNLOCK(fdp);
3707 PWDDESC_XUNLOCK(pdp);
3712 FILEDESC_SUNLOCK(fdp);
3716 newpwd->pwd_rdir = vp;
3717 if (oldpwd->pwd_jdir == NULL) {
3719 newpwd->pwd_jdir = vp;
3721 pwd_fill(oldpwd, newpwd);
3722 pwd_set(pdp, newpwd);
3723 PWDDESC_XUNLOCK(pdp);
3729 pwd_chdir(struct thread *td, struct vnode *vp)
3731 struct pwddesc *pdp;
3732 struct pwd *newpwd, *oldpwd;
3734 VNPASS(vp->v_usecount > 0, vp);
3736 newpwd = pwd_alloc();
3737 pdp = td->td_proc->p_pd;
3739 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3740 newpwd->pwd_cdir = vp;
3741 pwd_fill(oldpwd, newpwd);
3742 pwd_set(pdp, newpwd);
3743 PWDDESC_XUNLOCK(pdp);
3748 pwd_ensure_dirs(void)
3750 struct pwddesc *pdp;
3751 struct pwd *oldpwd, *newpwd;
3753 pdp = curproc->p_pd;
3755 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3756 if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3757 PWDDESC_XUNLOCK(pdp);
3760 PWDDESC_XUNLOCK(pdp);
3762 newpwd = pwd_alloc();
3764 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3765 pwd_fill(oldpwd, newpwd);
3766 if (newpwd->pwd_cdir == NULL) {
3768 newpwd->pwd_cdir = rootvnode;
3770 if (newpwd->pwd_rdir == NULL) {
3772 newpwd->pwd_rdir = rootvnode;
3774 pwd_set(pdp, newpwd);
3775 PWDDESC_XUNLOCK(pdp);
3780 pwd_set_rootvnode(void)
3782 struct pwddesc *pdp;
3783 struct pwd *oldpwd, *newpwd;
3785 pdp = curproc->p_pd;
3787 newpwd = pwd_alloc();
3789 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3791 newpwd->pwd_cdir = rootvnode;
3793 newpwd->pwd_rdir = rootvnode;
3794 pwd_fill(oldpwd, newpwd);
3795 pwd_set(pdp, newpwd);
3796 PWDDESC_XUNLOCK(pdp);
3801 * Scan all active processes and prisons to see if any of them have a current
3802 * or root directory of `olddp'. If so, replace them with the new mount point.
3805 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3807 struct pwddesc *pdp;
3808 struct pwd *newpwd, *oldpwd;
3813 if (vrefcnt(olddp) == 1)
3816 newpwd = pwd_alloc();
3817 sx_slock(&allproc_lock);
3818 FOREACH_PROC_IN_SYSTEM(p) {
3825 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3826 if (oldpwd == NULL ||
3827 (oldpwd->pwd_cdir != olddp &&
3828 oldpwd->pwd_rdir != olddp &&
3829 oldpwd->pwd_jdir != olddp)) {
3830 PWDDESC_XUNLOCK(pdp);
3834 if (oldpwd->pwd_cdir == olddp) {
3836 newpwd->pwd_cdir = newdp;
3838 if (oldpwd->pwd_rdir == olddp) {
3840 newpwd->pwd_rdir = newdp;
3842 if (oldpwd->pwd_jdir == olddp) {
3844 newpwd->pwd_jdir = newdp;
3846 pwd_fill(oldpwd, newpwd);
3847 pwd_set(pdp, newpwd);
3848 PWDDESC_XUNLOCK(pdp);
3851 newpwd = pwd_alloc();
3853 sx_sunlock(&allproc_lock);
3855 if (rootvnode == olddp) {
3860 mtx_lock(&prison0.pr_mtx);
3861 if (prison0.pr_root == olddp) {
3863 prison0.pr_root = newdp;
3866 mtx_unlock(&prison0.pr_mtx);
3867 sx_slock(&allprison_lock);
3868 TAILQ_FOREACH(pr, &allprison, pr_list) {
3869 mtx_lock(&pr->pr_mtx);
3870 if (pr->pr_root == olddp) {
3872 pr->pr_root = newdp;
3875 mtx_unlock(&pr->pr_mtx);
3877 sx_sunlock(&allprison_lock);
3882 struct filedesc_to_leader *
3883 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3885 struct filedesc_to_leader *fdtol;
3887 fdtol = malloc(sizeof(struct filedesc_to_leader),
3888 M_FILEDESC_TO_LEADER, M_WAITOK);
3889 fdtol->fdl_refcount = 1;
3890 fdtol->fdl_holdcount = 0;
3891 fdtol->fdl_wakeup = 0;
3892 fdtol->fdl_leader = leader;
3894 FILEDESC_XLOCK(fdp);
3895 fdtol->fdl_next = old->fdl_next;
3896 fdtol->fdl_prev = old;
3897 old->fdl_next = fdtol;
3898 fdtol->fdl_next->fdl_prev = fdtol;
3899 FILEDESC_XUNLOCK(fdp);
3901 fdtol->fdl_next = fdtol;
3902 fdtol->fdl_prev = fdtol;
3908 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3911 struct filedesc *fdp;
3912 int count, off, minoff;
3914 if (*(int *)arg1 != 0)
3917 fdp = curproc->p_fd;
3919 FILEDESC_SLOCK(fdp);
3921 off = NDSLOT(fdp->fd_nfiles - 1);
3922 for (minoff = NDSLOT(0); off >= minoff; --off)
3923 count += bitcountl(map[off]);
3924 FILEDESC_SUNLOCK(fdp);
3926 return (SYSCTL_OUT(req, &count, sizeof(count)));
3929 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3930 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3931 "Number of open file descriptors");
3934 * Get file structures globally.
3937 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3940 struct filedesc *fdp;
3943 int error, n, lastfile;
3945 error = sysctl_wire_old_buffer(req, 0);
3948 if (req->oldptr == NULL) {
3950 sx_slock(&allproc_lock);
3951 FOREACH_PROC_IN_SYSTEM(p) {
3953 if (p->p_state == PRS_NEW) {
3961 /* overestimates sparse tables. */
3962 n += fdp->fd_nfiles;
3965 sx_sunlock(&allproc_lock);
3966 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3969 bzero(&xf, sizeof(xf));
3970 xf.xf_size = sizeof(xf);
3971 sx_slock(&allproc_lock);
3972 FOREACH_PROC_IN_SYSTEM(p) {
3974 if (p->p_state == PRS_NEW) {
3978 if (p_cansee(req->td, p) != 0) {
3982 xf.xf_pid = p->p_pid;
3983 xf.xf_uid = p->p_ucred->cr_uid;
3988 FILEDESC_SLOCK(fdp);
3989 lastfile = fdlastfile(fdp);
3990 for (n = 0; refcount_load(&fdp->fd_refcnt) > 0 && n <= lastfile;
3992 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3995 xf.xf_file = (uintptr_t)fp;
3996 xf.xf_data = (uintptr_t)fp->f_data;
3997 xf.xf_vnode = (uintptr_t)fp->f_vnode;
3998 xf.xf_type = (uintptr_t)fp->f_type;
3999 xf.xf_count = refcount_load(&fp->f_count);
4001 xf.xf_offset = foffset_get(fp);
4002 xf.xf_flag = fp->f_flag;
4003 error = SYSCTL_OUT(req, &xf, sizeof(xf));
4007 FILEDESC_SUNLOCK(fdp);
4012 sx_sunlock(&allproc_lock);
4016 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4017 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4019 #ifdef KINFO_FILE_SIZE
4020 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4024 xlate_fflags(int fflags)
4026 static const struct {
4029 } fflags_table[] = {
4030 { FAPPEND, KF_FLAG_APPEND },
4031 { FASYNC, KF_FLAG_ASYNC },
4032 { FFSYNC, KF_FLAG_FSYNC },
4033 { FHASLOCK, KF_FLAG_HASLOCK },
4034 { FNONBLOCK, KF_FLAG_NONBLOCK },
4035 { FREAD, KF_FLAG_READ },
4036 { FWRITE, KF_FLAG_WRITE },
4037 { O_CREAT, KF_FLAG_CREAT },
4038 { O_DIRECT, KF_FLAG_DIRECT },
4039 { O_EXCL, KF_FLAG_EXCL },
4040 { O_EXEC, KF_FLAG_EXEC },
4041 { O_EXLOCK, KF_FLAG_EXLOCK },
4042 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4043 { O_SHLOCK, KF_FLAG_SHLOCK },
4044 { O_TRUNC, KF_FLAG_TRUNC }
4050 for (i = 0; i < nitems(fflags_table); i++)
4051 if (fflags & fflags_table[i].fflag)
4052 kflags |= fflags_table[i].kf_fflag;
4056 /* Trim unused data from kf_path by truncating the structure size. */
4058 pack_kinfo(struct kinfo_file *kif)
4061 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4062 strlen(kif->kf_path) + 1;
4063 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4067 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4068 struct kinfo_file *kif, struct filedesc *fdp, int flags)
4072 bzero(kif, sizeof(*kif));
4074 /* Set a default type to allow for empty fill_kinfo() methods. */
4075 kif->kf_type = KF_TYPE_UNKNOWN;
4076 kif->kf_flags = xlate_fflags(fp->f_flag);
4077 if (rightsp != NULL)
4078 kif->kf_cap_rights = *rightsp;
4080 cap_rights_init_zero(&kif->kf_cap_rights);
4082 kif->kf_ref_count = refcount_load(&fp->f_count);
4083 kif->kf_offset = foffset_get(fp);
4086 * This may drop the filedesc lock, so the 'fp' cannot be
4087 * accessed after this call.
4089 error = fo_fill_kinfo(fp, kif, fdp);
4091 kif->kf_status |= KF_ATTR_VALID;
4092 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4095 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4099 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4100 struct kinfo_file *kif, int flags)
4104 bzero(kif, sizeof(*kif));
4106 kif->kf_type = KF_TYPE_VNODE;
4107 error = vn_fill_kinfo_vnode(vp, kif);
4109 kif->kf_status |= KF_ATTR_VALID;
4110 kif->kf_flags = xlate_fflags(fflags);
4111 cap_rights_init_zero(&kif->kf_cap_rights);
4113 kif->kf_ref_count = -1;
4114 kif->kf_offset = -1;
4115 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4118 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4122 struct export_fd_buf {
4123 struct filedesc *fdp;
4124 struct pwddesc *pdp;
4127 struct kinfo_file kif;
4132 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4134 struct kinfo_file *kif;
4137 if (efbuf->remainder != -1) {
4138 if (efbuf->remainder < kif->kf_structsize) {
4139 /* Terminate export. */
4140 efbuf->remainder = 0;
4143 efbuf->remainder -= kif->kf_structsize;
4145 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
4149 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4150 struct export_fd_buf *efbuf)
4154 if (efbuf->remainder == 0)
4156 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4158 FILEDESC_SUNLOCK(efbuf->fdp);
4159 error = export_kinfo_to_sb(efbuf);
4160 FILEDESC_SLOCK(efbuf->fdp);
4165 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4166 struct export_fd_buf *efbuf)
4170 if (efbuf->remainder == 0)
4172 if (efbuf->pdp != NULL)
4173 PWDDESC_XUNLOCK(efbuf->pdp);
4174 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4175 error = export_kinfo_to_sb(efbuf);
4176 if (efbuf->pdp != NULL)
4177 PWDDESC_XLOCK(efbuf->pdp);
4182 * Store a process file descriptor information to sbuf.
4184 * Takes a locked proc as argument, and returns with the proc unlocked.
4187 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
4191 struct filedesc *fdp;
4192 struct pwddesc *pdp;
4193 struct export_fd_buf *efbuf;
4194 struct vnode *cttyvp, *textvp, *tracevp;
4196 int error, i, lastfile;
4197 cap_rights_t rights;
4199 PROC_LOCK_ASSERT(p, MA_OWNED);
4202 tracevp = p->p_tracevp;
4203 if (tracevp != NULL)
4206 textvp = p->p_textvp;
4209 /* Controlling tty. */
4211 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4212 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4219 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4223 efbuf->remainder = maxlen;
4224 efbuf->flags = flags;
4225 if (tracevp != NULL)
4226 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
4229 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
4231 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
4234 if (pdp == NULL || fdp == NULL)
4239 pwd = pwd_hold_pwddesc(pdp);
4241 /* working directory */
4242 if (pwd->pwd_cdir != NULL) {
4243 vrefact(pwd->pwd_cdir);
4244 export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4246 /* root directory */
4247 if (pwd->pwd_rdir != NULL) {
4248 vrefact(pwd->pwd_rdir);
4249 export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
4251 /* jail directory */
4252 if (pwd->pwd_jdir != NULL) {
4253 vrefact(pwd->pwd_jdir);
4254 export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
4257 PWDDESC_XUNLOCK(pdp);
4260 FILEDESC_SLOCK(fdp);
4261 lastfile = fdlastfile(fdp);
4262 for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4263 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4266 rights = *cap_rights(fdp, i);
4267 #else /* !CAPABILITIES */
4268 rights = cap_no_rights;
4271 * Create sysctl entry. It is OK to drop the filedesc
4272 * lock inside of export_file_to_sb() as we will
4273 * re-validate and re-evaluate its properties when the
4276 error = export_file_to_sb(fp, i, &rights, efbuf);
4277 if (error != 0 || efbuf->remainder == 0)
4280 FILEDESC_SUNLOCK(fdp);
4286 free(efbuf, M_TEMP);
4290 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
4293 * Get per-process file descriptors for use by procstat(1), et al.
4296 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4301 int error, error2, *name;
4305 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4306 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4307 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4312 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4313 error = kern_proc_filedesc_out(p, &sb, maxlen,
4314 KERN_FILEDESC_PACK_KINFO);
4315 error2 = sbuf_finish(&sb);
4317 return (error != 0 ? error : error2);
4320 #ifdef COMPAT_FREEBSD7
4321 #ifdef KINFO_OFILE_SIZE
4322 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4326 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4329 okif->kf_structsize = sizeof(*okif);
4330 okif->kf_type = kif->kf_type;
4331 okif->kf_fd = kif->kf_fd;
4332 okif->kf_ref_count = kif->kf_ref_count;
4333 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4334 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4335 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4336 okif->kf_offset = kif->kf_offset;
4337 if (kif->kf_type == KF_TYPE_VNODE)
4338 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4340 okif->kf_vnode_type = KF_VTYPE_VNON;
4341 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4342 if (kif->kf_type == KF_TYPE_SOCKET) {
4343 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4344 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4345 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4346 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4347 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4349 okif->kf_sa_local.ss_family = AF_UNSPEC;
4350 okif->kf_sa_peer.ss_family = AF_UNSPEC;
4355 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4356 struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4361 PWDDESC_XUNLOCK(pdp);
4362 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4363 kinfo_to_okinfo(kif, okif);
4364 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4370 * Get per-process file descriptors for use by procstat(1), et al.
4373 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4375 struct kinfo_ofile *okif;
4376 struct kinfo_file *kif;
4377 struct filedesc *fdp;
4378 struct pwddesc *pdp;
4380 int error, i, lastfile, *name;
4385 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4392 if (fdp == NULL || pdp == NULL) {
4397 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4398 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4400 pwd = pwd_hold_pwddesc(pdp);
4402 if (pwd->pwd_cdir != NULL)
4403 export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4405 if (pwd->pwd_rdir != NULL)
4406 export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4408 if (pwd->pwd_jdir != NULL)
4409 export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4412 PWDDESC_XUNLOCK(pdp);
4415 FILEDESC_SLOCK(fdp);
4416 lastfile = fdlastfile(fdp);
4417 for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4418 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4420 export_file_to_kinfo(fp, i, NULL, kif, fdp,
4421 KERN_FILEDESC_PACK_KINFO);
4422 FILEDESC_SUNLOCK(fdp);
4423 kinfo_to_okinfo(kif, okif);
4424 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4425 FILEDESC_SLOCK(fdp);
4429 FILEDESC_SUNLOCK(fdp);
4437 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4438 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4439 "Process ofiledesc entries");
4440 #endif /* COMPAT_FREEBSD7 */
4443 vntype_to_kinfo(int vtype)
4448 } vtypes_table[] = {
4449 { VBAD, KF_VTYPE_VBAD },
4450 { VBLK, KF_VTYPE_VBLK },
4451 { VCHR, KF_VTYPE_VCHR },
4452 { VDIR, KF_VTYPE_VDIR },
4453 { VFIFO, KF_VTYPE_VFIFO },
4454 { VLNK, KF_VTYPE_VLNK },
4455 { VNON, KF_VTYPE_VNON },
4456 { VREG, KF_VTYPE_VREG },
4457 { VSOCK, KF_VTYPE_VSOCK }
4462 * Perform vtype translation.
4464 for (i = 0; i < nitems(vtypes_table); i++)
4465 if (vtypes_table[i].vtype == vtype)
4466 return (vtypes_table[i].kf_vtype);
4468 return (KF_VTYPE_UNKNOWN);
4471 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4472 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4473 "Process filedesc entries");
4476 * Store a process current working directory information to sbuf.
4478 * Takes a locked proc as argument, and returns with the proc unlocked.
4481 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
4483 struct pwddesc *pdp;
4485 struct export_fd_buf *efbuf;
4489 PROC_LOCK_ASSERT(p, MA_OWNED);
4496 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4499 efbuf->remainder = maxlen;
4502 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4503 cdir = pwd->pwd_cdir;
4508 error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4510 PWDDESC_XUNLOCK(pdp);
4512 free(efbuf, M_TEMP);
4517 * Get per-process current working directory.
4520 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4525 int error, error2, *name;
4529 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4530 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4531 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4536 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4537 error = kern_proc_cwd_out(p, &sb, maxlen);
4538 error2 = sbuf_finish(&sb);
4540 return (error != 0 ? error : error2);
4543 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4544 sysctl_kern_proc_cwd, "Process current working directory");
4548 * For the purposes of debugging, generate a human-readable string for the
4552 file_type_to_name(short type)
4580 case DTYPE_PROCDESC:
4582 case DTYPE_LINUXEFD:
4584 case DTYPE_LINUXTFD:
4592 * For the purposes of debugging, identify a process (if any, perhaps one of
4593 * many) that references the passed file in its file descriptor array. Return
4596 static struct proc *
4597 file_to_first_proc(struct file *fp)
4599 struct filedesc *fdp;
4603 FOREACH_PROC_IN_SYSTEM(p) {
4604 if (p->p_state == PRS_NEW)
4609 for (n = 0; n < fdp->fd_nfiles; n++) {
4610 if (fp == fdp->fd_ofiles[n].fde_file)
4618 db_print_file(struct file *fp, int header)
4620 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4624 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4625 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4626 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4628 p = file_to_first_proc(fp);
4629 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4630 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4631 fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4632 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4637 DB_SHOW_COMMAND(file, db_show_file)
4642 db_printf("usage: show file <addr>\n");
4645 fp = (struct file *)addr;
4646 db_print_file(fp, 1);
4649 DB_SHOW_COMMAND(files, db_show_files)
4651 struct filedesc *fdp;
4658 FOREACH_PROC_IN_SYSTEM(p) {
4659 if (p->p_state == PRS_NEW)
4661 if ((fdp = p->p_fd) == NULL)
4663 for (n = 0; n < fdp->fd_nfiles; ++n) {
4664 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4666 db_print_file(fp, header);
4673 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4674 &maxfilesperproc, 0, "Maximum files allowed open per process");
4676 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4677 &maxfiles, 0, "Maximum number of files");
4679 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4680 &openfiles, 0, "System-wide number of open files");
4684 filelistinit(void *dummy)
4687 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4688 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4689 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4690 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4691 pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4692 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4694 * XXXMJG this is a temporary hack due to boot ordering issues against
4697 vfs_smr = uma_zone_get_smr(pwd_zone);
4698 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4700 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4702 /*-------------------------------------------------------------------*/
4705 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4706 int flags, struct thread *td)
4713 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4721 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4729 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4737 badfo_kqfilter(struct file *fp, struct knote *kn)
4744 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4752 badfo_close(struct file *fp, struct thread *td)
4759 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4767 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4775 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4776 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4784 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4790 struct fileops badfileops = {
4791 .fo_read = badfo_readwrite,
4792 .fo_write = badfo_readwrite,
4793 .fo_truncate = badfo_truncate,
4794 .fo_ioctl = badfo_ioctl,
4795 .fo_poll = badfo_poll,
4796 .fo_kqfilter = badfo_kqfilter,
4797 .fo_stat = badfo_stat,
4798 .fo_close = badfo_close,
4799 .fo_chmod = badfo_chmod,
4800 .fo_chown = badfo_chown,
4801 .fo_sendfile = badfo_sendfile,
4802 .fo_fill_kinfo = badfo_fill_kinfo,
4806 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4807 int flags, struct thread *td)
4810 return (EOPNOTSUPP);
4814 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4822 invfo_ioctl(struct file *fp, u_long com, void *data,
4823 struct ucred *active_cred, struct thread *td)
4830 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4834 return (poll_no_poll(events));
4838 invfo_kqfilter(struct file *fp, struct knote *kn)
4845 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4853 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4861 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4862 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4869 /*-------------------------------------------------------------------*/
4872 * File Descriptor pseudo-device driver (/dev/fd/).
4874 * Opening minor device N dup()s the file (if any) connected to file
4875 * descriptor N belonging to the calling process. Note that this driver
4876 * consists of only the ``open()'' routine, because all subsequent
4877 * references to this file will be direct to the other driver.
4879 * XXX: we could give this one a cloning event handler if necessary.
4884 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4888 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4889 * the file descriptor being sought for duplication. The error
4890 * return ensures that the vnode for this device will be released
4891 * by vn_open. Open will detect this special error and take the
4892 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4893 * will simply report the error.
4895 td->td_dupfd = dev2unit(dev);
4899 static struct cdevsw fildesc_cdevsw = {
4900 .d_version = D_VERSION,
4906 fildesc_drvinit(void *unused)
4910 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4911 UID_ROOT, GID_WHEEL, 0666, "fd/0");
4912 make_dev_alias(dev, "stdin");
4913 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4914 UID_ROOT, GID_WHEEL, 0666, "fd/1");
4915 make_dev_alias(dev, "stdout");
4916 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4917 UID_ROOT, GID_WHEEL, 0666, "fd/2");
4918 make_dev_alias(dev, "stderr");
4921 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);