2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1989, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
42 #include "opt_capsicum.h"
44 #include "opt_ktrace.h"
46 #include <sys/param.h>
47 #include <sys/systm.h>
49 #include <sys/capsicum.h>
51 #include <sys/fcntl.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
67 #include <sys/protosw.h>
68 #include <sys/racct.h>
69 #include <sys/resourcevar.h>
71 #include <sys/signalvar.h>
76 #include <sys/syscallsubr.h>
77 #include <sys/sysctl.h>
78 #include <sys/sysproto.h>
79 #include <sys/unistd.h>
81 #include <sys/vnode.h>
82 #include <sys/ktrace.h>
86 #include <security/audit/audit.h>
93 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
94 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
95 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
96 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
97 "file desc to leader structures");
98 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
99 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
101 MALLOC_DECLARE(M_FADVISE);
103 static __read_mostly uma_zone_t file_zone;
104 static __read_mostly uma_zone_t filedesc0_zone;
105 __read_mostly uma_zone_t pwd_zone;
108 static int closefp(struct filedesc *fdp, int fd, struct file *fp,
109 struct thread *td, bool holdleaders, bool audit);
110 static void export_file_to_kinfo(struct file *fp, int fd,
111 cap_rights_t *rightsp, struct kinfo_file *kif,
112 struct filedesc *fdp, int flags);
113 static int fd_first_free(struct filedesc *fdp, int low, int size);
114 static void fdgrowtable(struct filedesc *fdp, int nfd);
115 static void fdgrowtable_exp(struct filedesc *fdp, int nfd);
116 static void fdunused(struct filedesc *fdp, int fd);
117 static void fdused(struct filedesc *fdp, int fd);
118 static int fget_unlocked_seq(struct thread *td, int fd,
119 cap_rights_t *needrightsp, struct file **fpp, seqc_t *seqp);
120 static int getmaxfd(struct thread *td);
121 static u_long *filecaps_copy_prep(const struct filecaps *src);
122 static void filecaps_copy_finish(const struct filecaps *src,
123 struct filecaps *dst, u_long *ioctls);
124 static u_long *filecaps_free_prep(struct filecaps *fcaps);
125 static void filecaps_free_finish(u_long *ioctls);
127 static struct pwd *pwd_alloc(void);
132 * - An array of open file descriptors (fd_ofiles)
133 * - An array of file flags (fd_ofileflags)
134 * - A bitmap recording which descriptors are in use (fd_map)
136 * A process starts out with NDFILE descriptors. The value of NDFILE has
137 * been selected based the historical limit of 20 open files, and an
138 * assumption that the majority of processes, especially short-lived
139 * processes like shells, will never need more.
141 * If this initial allocation is exhausted, a larger descriptor table and
142 * map are allocated dynamically, and the pointers in the process's struct
143 * filedesc are updated to point to those. This is repeated every time
144 * the process runs out of file descriptors (provided it hasn't hit its
147 * Since threads may hold references to individual descriptor table
148 * entries, the tables are never freed. Instead, they are placed on a
149 * linked list and freed only when the struct filedesc is released.
152 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
153 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
154 #define NDSLOT(x) ((x) / NDENTRIES)
155 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
156 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
159 * SLIST entry used to keep track of ofiles which must be reclaimed when
163 struct fdescenttbl *ft_table;
164 SLIST_ENTRY(freetable) ft_next;
168 * Initial allocation: a filedesc structure + the head of SLIST used to
169 * keep track of old ofiles + enough space for NDFILE descriptors.
172 struct fdescenttbl0 {
174 struct filedescent fdt_ofiles[NDFILE];
178 struct filedesc fd_fd;
179 SLIST_HEAD(, freetable) fd_free;
180 struct fdescenttbl0 fd_dfiles;
181 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
185 * Descriptor management.
187 static int __exclusive_cache_line openfiles; /* actual number of open files */
188 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
189 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
192 * If low >= size, just return low. Otherwise find the first zero bit in the
193 * given bitmap, starting at low and not exceeding size - 1. Return size if
197 fd_first_free(struct filedesc *fdp, int low, int size)
199 NDSLOTTYPE *map = fdp->fd_map;
207 if (low % NDENTRIES) {
208 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
209 if ((mask &= ~map[off]) != 0UL)
210 return (off * NDENTRIES + ffsl(mask) - 1);
213 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
214 if (map[off] != ~0UL)
215 return (off * NDENTRIES + ffsl(~map[off]) - 1);
220 * Find the last used fd.
222 * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
223 * Otherwise use fdlastfile.
226 fdlastfile_single(struct filedesc *fdp)
228 NDSLOTTYPE *map = fdp->fd_map;
231 off = NDSLOT(fdp->fd_nfiles - 1);
232 for (minoff = NDSLOT(0); off >= minoff; --off)
234 return (off * NDENTRIES + flsl(map[off]) - 1);
239 fdlastfile(struct filedesc *fdp)
242 FILEDESC_LOCK_ASSERT(fdp);
243 return (fdlastfile_single(fdp));
247 fdisused(struct filedesc *fdp, int fd)
250 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
251 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
253 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
257 * Mark a file descriptor as used.
260 fdused_init(struct filedesc *fdp, int fd)
263 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
265 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
269 fdused(struct filedesc *fdp, int fd)
272 FILEDESC_XLOCK_ASSERT(fdp);
274 fdused_init(fdp, fd);
275 if (fd == fdp->fd_freefile)
280 * Mark a file descriptor as unused.
283 fdunused(struct filedesc *fdp, int fd)
286 FILEDESC_XLOCK_ASSERT(fdp);
288 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
289 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
290 ("fd=%d is still in use", fd));
292 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
293 if (fd < fdp->fd_freefile)
294 fdp->fd_freefile = fd;
298 * Free a file descriptor.
300 * Avoid some work if fdp is about to be destroyed.
303 fdefree_last(struct filedescent *fde)
306 filecaps_free(&fde->fde_caps);
310 fdfree(struct filedesc *fdp, int fd)
312 struct filedescent *fde;
314 FILEDESC_XLOCK_ASSERT(fdp);
315 fde = &fdp->fd_ofiles[fd];
317 seqc_write_begin(&fde->fde_seqc);
319 fde->fde_file = NULL;
321 seqc_write_end(&fde->fde_seqc);
328 * System calls on descriptors.
330 #ifndef _SYS_SYSPROTO_H_
331 struct getdtablesize_args {
337 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
343 td->td_retval[0] = getmaxfd(td);
345 PROC_LOCK(td->td_proc);
346 lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
347 PROC_UNLOCK(td->td_proc);
348 if (lim < td->td_retval[0])
349 td->td_retval[0] = lim;
355 * Duplicate a file descriptor to a particular value.
357 * Note: keep in mind that a potential race condition exists when closing
358 * descriptors from a shared descriptor table (via rfork).
360 #ifndef _SYS_SYSPROTO_H_
368 sys_dup2(struct thread *td, struct dup2_args *uap)
371 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
375 * Duplicate a file descriptor.
377 #ifndef _SYS_SYSPROTO_H_
384 sys_dup(struct thread *td, struct dup_args *uap)
387 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
391 * The file control system call.
393 #ifndef _SYS_SYSPROTO_H_
402 sys_fcntl(struct thread *td, struct fcntl_args *uap)
405 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
409 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
423 * Convert old flock structure to new.
425 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
426 fl.l_start = ofl.l_start;
427 fl.l_len = ofl.l_len;
428 fl.l_pid = ofl.l_pid;
429 fl.l_type = ofl.l_type;
430 fl.l_whence = ofl.l_whence;
444 arg1 = (intptr_t)&fl;
450 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
451 arg1 = (intptr_t)&fl;
459 error = kern_fcntl(td, fd, newcmd, arg1);
462 if (cmd == F_OGETLK) {
463 ofl.l_start = fl.l_start;
464 ofl.l_len = fl.l_len;
465 ofl.l_pid = fl.l_pid;
466 ofl.l_type = fl.l_type;
467 ofl.l_whence = fl.l_whence;
468 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
469 } else if (cmd == F_GETLK) {
470 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
476 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
478 struct filedesc *fdp;
480 struct file *fp, *fp2;
481 struct filedescent *fde;
485 struct kinfo_file *kif;
486 int error, flg, kif_sz, seals, tmp;
500 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
503 case F_DUPFD_CLOEXEC:
505 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
510 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
513 case F_DUP2FD_CLOEXEC:
515 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
521 fde = fdeget_noref(fdp, fd);
524 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
527 FILEDESC_SUNLOCK(fdp);
533 fde = fdeget_noref(fdp, fd);
535 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
536 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
539 FILEDESC_XUNLOCK(fdp);
543 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
546 td->td_retval[0] = OFLAGS(fp->f_flag);
551 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
554 if (fp->f_ops == &path_fileops) {
560 tmp = flg = fp->f_flag;
562 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
563 } while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
564 tmp = fp->f_flag & FNONBLOCK;
565 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
570 tmp = fp->f_flag & FASYNC;
571 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
576 atomic_clear_int(&fp->f_flag, FNONBLOCK);
578 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
583 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
586 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
588 td->td_retval[0] = tmp;
593 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
597 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
602 error = priv_check(td, PRIV_NFS_LOCKD);
610 /* FALLTHROUGH F_SETLK */
614 flp = (struct flock *)arg;
615 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
620 error = fget_unlocked(td, fd, &cap_flock_rights, &fp);
623 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
629 if (flp->l_whence == SEEK_CUR) {
630 foffset = foffset_get(fp);
633 foffset > OFF_MAX - flp->l_start)) {
638 flp->l_start += foffset;
642 switch (flp->l_type) {
644 if ((fp->f_flag & FREAD) == 0) {
648 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
649 PROC_LOCK(p->p_leader);
650 p->p_leader->p_flag |= P_ADVLOCK;
651 PROC_UNLOCK(p->p_leader);
653 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
657 if ((fp->f_flag & FWRITE) == 0) {
661 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
662 PROC_LOCK(p->p_leader);
663 p->p_leader->p_flag |= P_ADVLOCK;
664 PROC_UNLOCK(p->p_leader);
666 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
670 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
674 if (flg != F_REMOTE) {
678 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
679 F_UNLCKSYS, flp, flg);
685 if (error != 0 || flp->l_type == F_UNLCK ||
686 flp->l_type == F_UNLCKSYS) {
692 * Check for a race with close.
694 * The vnode is now advisory locked (or unlocked, but this case
695 * is not really important) as the caller requested.
696 * We had to drop the filedesc lock, so we need to recheck if
697 * the descriptor is still valid, because if it was closed
698 * in the meantime we need to remove advisory lock from the
699 * vnode - close on any descriptor leading to an advisory
700 * locked vnode, removes that lock.
701 * We will return 0 on purpose in that case, as the result of
702 * successful advisory lock might have been externally visible
703 * already. This is fine - effectively we pretend to the caller
704 * that the closing thread was a bit slower and that the
705 * advisory lock succeeded before the close.
707 error = fget_unlocked(td, fd, &cap_no_rights, &fp2);
713 flp->l_whence = SEEK_SET;
716 flp->l_type = F_UNLCK;
717 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
718 F_UNLCK, flp, F_POSIX);
725 error = fget_unlocked(td, fd, &cap_flock_rights, &fp);
728 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
733 flp = (struct flock *)arg;
734 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
735 flp->l_type != F_UNLCK) {
740 if (flp->l_whence == SEEK_CUR) {
741 foffset = foffset_get(fp);
742 if ((flp->l_start > 0 &&
743 foffset > OFF_MAX - flp->l_start) ||
745 foffset < OFF_MIN - flp->l_start)) {
750 flp->l_start += foffset;
753 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
759 error = fget_unlocked(td, fd, &cap_no_rights, &fp);
762 error = fo_add_seals(fp, arg);
767 error = fget_unlocked(td, fd, &cap_no_rights, &fp);
770 if (fo_get_seals(fp, &seals) == 0)
771 td->td_retval[0] = seals;
778 arg = arg ? 128 * 1024: 0;
781 error = fget_unlocked(td, fd, &cap_no_rights, &fp);
784 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
790 if (vp->v_type != VREG) {
797 * Exclusive lock synchronizes against f_seqcount reads and
798 * writes in sequential_heuristic().
800 error = vn_lock(vp, LK_EXCLUSIVE);
806 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
807 arg = MIN(arg, INT_MAX - bsize + 1);
808 fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
809 (arg + bsize - 1) / bsize);
810 atomic_set_int(&fp->f_flag, FRDAHEAD);
812 atomic_clear_int(&fp->f_flag, FRDAHEAD);
820 * Check if the vnode is part of a union stack (either the
821 * "union" flag from mount(2) or unionfs).
823 * Prior to introduction of this op libc's readdir would call
824 * fstatfs(2), in effect unnecessarily copying kilobytes of
825 * data just to check fs name and a mount flag.
827 * Fixing the code to handle everything in the kernel instead
828 * is a non-trivial endeavor and has low priority, thus this
829 * horrible kludge facilitates the current behavior in a much
830 * cheaper manner until someone(tm) sorts this out.
832 error = fget_unlocked(td, fd, &cap_no_rights, &fp);
835 if (fp->f_type != DTYPE_VNODE) {
842 * Since we don't prevent dooming the vnode even non-null mp
843 * found can become immediately stale. This is tolerable since
844 * mount points are type-stable (providing safe memory access)
845 * and any vfs op on this vnode going forward will return an
846 * error (meaning return value in this case is meaningless).
848 mp = atomic_load_ptr(&vp->v_mount);
849 if (__predict_false(mp == NULL)) {
854 td->td_retval[0] = 0;
855 if (mp->mnt_kern_flag & MNTK_UNIONFS ||
856 mp->mnt_flag & MNT_UNION)
857 td->td_retval[0] = 1;
862 #ifdef CAPABILITY_MODE
863 if (IN_CAPABILITY_MODE(td)) {
868 error = copyin((void *)arg, &kif_sz, sizeof(kif_sz));
871 if (kif_sz != sizeof(*kif)) {
875 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK | M_ZERO);
877 error = fget_cap_noref(fdp, fd, &cap_fcntl_rights, &fp, NULL);
878 if (error == 0 && fhold(fp)) {
879 export_file_to_kinfo(fp, fd, NULL, kif, fdp, 0);
880 FILEDESC_SUNLOCK(fdp);
882 if ((kif->kf_status & KF_ATTR_VALID) != 0) {
883 kif->kf_structsize = sizeof(*kif);
884 error = copyout(kif, (void *)arg, sizeof(*kif));
889 FILEDESC_SUNLOCK(fdp);
904 getmaxfd(struct thread *td)
907 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
911 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
914 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
916 struct filedesc *fdp;
917 struct filedescent *oldfde, *newfde;
919 struct file *delfp, *oldfp;
920 u_long *oioctls, *nioctls;
927 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
928 MPASS(mode < FDDUP_LASTMODE);
931 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
934 * Verify we have a valid descriptor to dup from and possibly to
935 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
936 * return EINVAL when the new descriptor is out of bounds.
941 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
942 maxfd = getmaxfd(td);
944 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
948 if (fget_noref(fdp, old) == NULL)
950 if (mode == FDDUP_FIXED && old == new) {
951 td->td_retval[0] = new;
952 if (flags & FDDUP_FLAG_CLOEXEC)
953 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
958 oldfde = &fdp->fd_ofiles[old];
959 oldfp = oldfde->fde_file;
964 * If the caller specified a file descriptor, make sure the file
965 * table is large enough to hold it, and grab it. Otherwise, just
966 * allocate a new descriptor the usual way.
971 if ((error = fdalloc(td, new, &new)) != 0) {
977 if (new >= fdp->fd_nfiles) {
979 * The resource limits are here instead of e.g.
980 * fdalloc(), because the file descriptor table may be
981 * shared between processes, so we can't really use
982 * racct_add()/racct_sub(). Instead of counting the
983 * number of actually allocated descriptors, just put
984 * the limit on the size of the file descriptor table.
987 if (RACCT_ENABLED()) {
988 error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
996 fdgrowtable_exp(fdp, new + 1);
998 if (!fdisused(fdp, new))
1002 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
1005 KASSERT(old != new, ("new fd is same as old"));
1007 /* Refetch oldfde because the table may have grown and old one freed. */
1008 oldfde = &fdp->fd_ofiles[old];
1009 KASSERT(oldfp == oldfde->fde_file,
1010 ("fdt_ofiles shift from growth observed at fd %d",
1013 newfde = &fdp->fd_ofiles[new];
1014 delfp = newfde->fde_file;
1016 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
1019 * Duplicate the source descriptor.
1022 seqc_write_begin(&newfde->fde_seqc);
1024 oioctls = filecaps_free_prep(&newfde->fde_caps);
1025 fde_copy(oldfde, newfde);
1026 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
1028 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
1029 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
1031 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
1033 seqc_write_end(&newfde->fde_seqc);
1035 td->td_retval[0] = new;
1039 if (delfp != NULL) {
1040 (void) closefp(fdp, new, delfp, td, true, false);
1041 FILEDESC_UNLOCK_ASSERT(fdp);
1044 FILEDESC_XUNLOCK(fdp);
1047 filecaps_free_finish(oioctls);
1052 sigiofree(struct sigio *sigio)
1054 crfree(sigio->sio_ucred);
1055 free(sigio, M_SIGIO);
1058 static struct sigio *
1059 funsetown_locked(struct sigio *sigio)
1064 SIGIO_ASSERT_LOCKED();
1068 *sigio->sio_myref = NULL;
1069 if (sigio->sio_pgid < 0) {
1070 pg = sigio->sio_pgrp;
1072 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio);
1075 p = sigio->sio_proc;
1077 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio);
1084 * If sigio is on the list associated with a process or process group,
1085 * disable signalling from the device, remove sigio from the list and
1089 funsetown(struct sigio **sigiop)
1091 struct sigio *sigio;
1093 /* Racy check, consumers must provide synchronization. */
1094 if (*sigiop == NULL)
1098 sigio = funsetown_locked(*sigiop);
1105 * Free a list of sigio structures. The caller must ensure that new sigio
1106 * structures cannot be added after this point. For process groups this is
1107 * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1111 funsetownlst(struct sigiolst *sigiolst)
1115 struct sigio *sigio, *tmp;
1118 sigio = SLIST_FIRST(sigiolst);
1126 sigio = SLIST_FIRST(sigiolst);
1127 if (sigio == NULL) {
1133 * Every entry of the list should belong to a single proc or pgrp.
1135 if (sigio->sio_pgid < 0) {
1136 pg = sigio->sio_pgrp;
1137 sx_assert(&proctree_lock, SX_XLOCKED);
1139 } else /* if (sigio->sio_pgid > 0) */ {
1140 p = sigio->sio_proc;
1142 KASSERT((p->p_flag & P_WEXIT) != 0,
1143 ("%s: process %p is not exiting", __func__, p));
1146 SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1147 *sigio->sio_myref = NULL;
1149 KASSERT(sigio->sio_pgid < 0,
1150 ("Proc sigio in pgrp sigio list"));
1151 KASSERT(sigio->sio_pgrp == pg,
1152 ("Bogus pgrp in sigio list"));
1153 } else /* if (p != NULL) */ {
1154 KASSERT(sigio->sio_pgid > 0,
1155 ("Pgrp sigio in proc sigio list"));
1156 KASSERT(sigio->sio_proc == p,
1157 ("Bogus proc in sigio list"));
1167 SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1172 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1174 * After permission checking, add a sigio structure to the sigio list for
1175 * the process or process group.
1178 fsetown(pid_t pgid, struct sigio **sigiop)
1182 struct sigio *osigio, *sigio;
1190 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1191 sigio->sio_pgid = pgid;
1192 sigio->sio_ucred = crhold(curthread->td_ucred);
1193 sigio->sio_myref = sigiop;
1197 ret = pget(pgid, PGET_NOTWEXIT | PGET_NOTID | PGET_HOLD, &proc);
1199 osigio = funsetown_locked(*sigiop);
1203 if ((proc->p_flag & P_WEXIT) != 0) {
1205 } else if (proc->p_session !=
1206 curthread->td_proc->p_session) {
1208 * Policy - Don't allow a process to FSETOWN a
1209 * process in another session.
1211 * Remove this test to allow maximum flexibility
1212 * or restrict FSETOWN to the current process or
1213 * process group for maximum safety.
1217 sigio->sio_proc = proc;
1218 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio,
1223 } else /* if (pgid < 0) */ {
1224 sx_slock(&proctree_lock);
1226 osigio = funsetown_locked(*sigiop);
1227 pgrp = pgfind(-pgid);
1231 if (pgrp->pg_session != curthread->td_proc->p_session) {
1233 * Policy - Don't allow a process to FSETOWN a
1234 * process in another session.
1236 * Remove this test to allow maximum flexibility
1237 * or restrict FSETOWN to the current process or
1238 * process group for maximum safety.
1242 sigio->sio_pgrp = pgrp;
1243 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio,
1248 sx_sunlock(&proctree_lock);
1259 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1262 fgetown(struct sigio **sigiop)
1267 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1273 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1278 FILEDESC_XLOCK_ASSERT(fdp);
1281 * We now hold the fp reference that used to be owned by the
1282 * descriptor array. We have to unlock the FILEDESC *AFTER*
1283 * knote_fdclose to prevent a race of the fd getting opened, a knote
1284 * added, and deleteing a knote for the new fd.
1286 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1287 knote_fdclose(td, fd);
1290 * We need to notify mqueue if the object is of type mqueue.
1292 if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1293 mq_fdclose(td, fd, fp);
1294 FILEDESC_XUNLOCK(fdp);
1297 if (AUDITING_TD(td) && audit)
1298 audit_sysclose(td, fd, fp);
1300 error = closef(fp, td);
1303 * All paths leading up to closefp() will have already removed or
1304 * replaced the fd in the filedesc table, so a restart would not
1305 * operate on the same file.
1307 if (error == ERESTART)
1314 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1315 bool holdleaders, bool audit)
1319 FILEDESC_XLOCK_ASSERT(fdp);
1322 if (td->td_proc->p_fdtol != NULL) {
1324 * Ask fdfree() to sleep to ensure that all relevant
1325 * process leaders can be traversed in closef().
1327 fdp->fd_holdleaderscount++;
1329 holdleaders = false;
1333 error = closefp_impl(fdp, fd, fp, td, audit);
1335 FILEDESC_XLOCK(fdp);
1336 fdp->fd_holdleaderscount--;
1337 if (fdp->fd_holdleaderscount == 0 &&
1338 fdp->fd_holdleaderswakeup != 0) {
1339 fdp->fd_holdleaderswakeup = 0;
1340 wakeup(&fdp->fd_holdleaderscount);
1342 FILEDESC_XUNLOCK(fdp);
1348 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1349 bool holdleaders, bool audit)
1352 FILEDESC_XLOCK_ASSERT(fdp);
1354 if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1355 return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1357 return (closefp_impl(fdp, fd, fp, td, audit));
1362 * Close a file descriptor.
1364 #ifndef _SYS_SYSPROTO_H_
1371 sys_close(struct thread *td, struct close_args *uap)
1374 return (kern_close(td, uap->fd));
1378 kern_close(struct thread *td, int fd)
1380 struct filedesc *fdp;
1383 fdp = td->td_proc->p_fd;
1385 FILEDESC_XLOCK(fdp);
1386 if ((fp = fget_noref(fdp, fd)) == NULL) {
1387 FILEDESC_XUNLOCK(fdp);
1392 /* closefp() drops the FILEDESC lock for us. */
1393 return (closefp(fdp, fd, fp, td, true, true));
1397 close_range_cloexec(struct thread *td, u_int lowfd, u_int highfd)
1399 struct filedesc *fdp;
1400 struct fdescenttbl *fdt;
1401 struct filedescent *fde;
1404 fdp = td->td_proc->p_fd;
1405 FILEDESC_XLOCK(fdp);
1406 fdt = atomic_load_ptr(&fdp->fd_files);
1407 highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1409 if (__predict_false(fd > highfd)) {
1412 for (; fd <= highfd; fd++) {
1413 fde = &fdt->fdt_ofiles[fd];
1414 if (fde->fde_file != NULL)
1415 fde->fde_flags |= UF_EXCLOSE;
1418 FILEDESC_XUNLOCK(fdp);
1423 close_range_impl(struct thread *td, u_int lowfd, u_int highfd)
1425 struct filedesc *fdp;
1426 const struct fdescenttbl *fdt;
1430 fdp = td->td_proc->p_fd;
1431 FILEDESC_XLOCK(fdp);
1432 fdt = atomic_load_ptr(&fdp->fd_files);
1433 highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1435 if (__predict_false(fd > highfd)) {
1439 fp = fdt->fdt_ofiles[fd].fde_file;
1445 (void) closefp(fdp, fd, fp, td, true, true);
1448 FILEDESC_XLOCK(fdp);
1449 fdt = atomic_load_ptr(&fdp->fd_files);
1454 FILEDESC_XUNLOCK(fdp);
1460 kern_close_range(struct thread *td, int flags, u_int lowfd, u_int highfd)
1464 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1465 * open should not be a usage error. From a close_range() perspective,
1466 * close_range(3, ~0U, 0) in the same scenario should also likely not
1467 * be a usage error as all fd above 3 are in-fact already closed.
1469 if (highfd < lowfd) {
1473 if ((flags & CLOSE_RANGE_CLOEXEC) != 0)
1474 return (close_range_cloexec(td, lowfd, highfd));
1476 return (close_range_impl(td, lowfd, highfd));
1479 #ifndef _SYS_SYSPROTO_H_
1480 struct close_range_args {
1487 sys_close_range(struct thread *td, struct close_range_args *uap)
1490 AUDIT_ARG_FD(uap->lowfd);
1491 AUDIT_ARG_CMD(uap->highfd);
1492 AUDIT_ARG_FFLAGS(uap->flags);
1494 if ((uap->flags & ~(CLOSE_RANGE_CLOEXEC)) != 0)
1496 return (kern_close_range(td, uap->flags, uap->lowfd, uap->highfd));
1499 #ifdef COMPAT_FREEBSD12
1501 * Close open file descriptors.
1503 #ifndef _SYS_SYSPROTO_H_
1504 struct freebsd12_closefrom_args {
1510 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1514 AUDIT_ARG_FD(uap->lowfd);
1517 * Treat negative starting file descriptor values identical to
1518 * closefrom(0) which closes all files.
1520 lowfd = MAX(0, uap->lowfd);
1521 return (kern_close_range(td, 0, lowfd, ~0U));
1523 #endif /* COMPAT_FREEBSD12 */
1525 #if defined(COMPAT_43)
1527 * Return status information about a file descriptor.
1529 #ifndef _SYS_SYSPROTO_H_
1530 struct ofstat_args {
1537 ofstat(struct thread *td, struct ofstat_args *uap)
1543 error = kern_fstat(td, uap->fd, &ub);
1546 error = copyout(&oub, uap->sb, sizeof(oub));
1550 #endif /* COMPAT_43 */
1552 #if defined(COMPAT_FREEBSD11)
1554 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1557 struct freebsd11_stat osb;
1560 error = kern_fstat(td, uap->fd, &sb);
1563 error = freebsd11_cvtstat(&sb, &osb);
1565 error = copyout(&osb, uap->sb, sizeof(osb));
1568 #endif /* COMPAT_FREEBSD11 */
1571 * Return status information about a file descriptor.
1573 #ifndef _SYS_SYSPROTO_H_
1581 sys_fstat(struct thread *td, struct fstat_args *uap)
1586 error = kern_fstat(td, uap->fd, &ub);
1588 error = copyout(&ub, uap->sb, sizeof(ub));
1593 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1600 error = fget(td, fd, &cap_fstat_rights, &fp);
1601 if (__predict_false(error != 0))
1604 AUDIT_ARG_FILE(td->td_proc, fp);
1606 error = fo_stat(fp, sbp, td->td_ucred);
1608 #ifdef __STAT_TIME_T_EXT
1609 sbp->st_atim_ext = 0;
1610 sbp->st_mtim_ext = 0;
1611 sbp->st_ctim_ext = 0;
1612 sbp->st_btim_ext = 0;
1615 if (KTRPOINT(td, KTR_STRUCT))
1616 ktrstat_error(sbp, error);
1621 #if defined(COMPAT_FREEBSD11)
1623 * Return status information about a file descriptor.
1625 #ifndef _SYS_SYSPROTO_H_
1626 struct freebsd11_nfstat_args {
1633 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1639 error = kern_fstat(td, uap->fd, &ub);
1642 error = freebsd11_cvtnstat(&ub, &nub);
1644 error = copyout(&nub, uap->sb, sizeof(nub));
1647 #endif /* COMPAT_FREEBSD11 */
1650 * Return pathconf information about a file descriptor.
1652 #ifndef _SYS_SYSPROTO_H_
1653 struct fpathconf_args {
1660 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1665 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1667 td->td_retval[0] = value;
1672 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1678 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1682 if (name == _PC_ASYNC_IO) {
1683 *valuep = _POSIX_ASYNCHRONOUS_IO;
1688 vn_lock(vp, LK_SHARED | LK_RETRY);
1689 error = VOP_PATHCONF(vp, name, valuep);
1691 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1692 if (name != _PC_PIPE_BUF) {
1707 * Copy filecaps structure allocating memory for ioctls array if needed.
1709 * The last parameter indicates whether the fdtable is locked. If it is not and
1710 * ioctls are encountered, copying fails and the caller must lock the table.
1712 * Note that if the table was not locked, the caller has to check the relevant
1713 * sequence counter to determine whether the operation was successful.
1716 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1720 if (src->fc_ioctls != NULL && !locked)
1722 memcpy(dst, src, sizeof(*src));
1723 if (src->fc_ioctls == NULL)
1726 KASSERT(src->fc_nioctls > 0,
1727 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1729 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1730 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1731 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1736 filecaps_copy_prep(const struct filecaps *src)
1741 if (__predict_true(src->fc_ioctls == NULL))
1744 KASSERT(src->fc_nioctls > 0,
1745 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1747 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1748 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1753 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1759 if (__predict_true(src->fc_ioctls == NULL)) {
1760 MPASS(ioctls == NULL);
1764 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1765 dst->fc_ioctls = ioctls;
1766 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1770 * Move filecaps structure to the new place and clear the old place.
1773 filecaps_move(struct filecaps *src, struct filecaps *dst)
1777 bzero(src, sizeof(*src));
1781 * Fill the given filecaps structure with full rights.
1784 filecaps_fill(struct filecaps *fcaps)
1787 CAP_ALL(&fcaps->fc_rights);
1788 fcaps->fc_ioctls = NULL;
1789 fcaps->fc_nioctls = -1;
1790 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1794 * Free memory allocated within filecaps structure.
1797 filecaps_free(struct filecaps *fcaps)
1800 free(fcaps->fc_ioctls, M_FILECAPS);
1801 bzero(fcaps, sizeof(*fcaps));
1805 filecaps_free_prep(struct filecaps *fcaps)
1809 ioctls = fcaps->fc_ioctls;
1810 bzero(fcaps, sizeof(*fcaps));
1815 filecaps_free_finish(u_long *ioctls)
1818 free(ioctls, M_FILECAPS);
1822 * Validate the given filecaps structure.
1825 filecaps_validate(const struct filecaps *fcaps, const char *func)
1828 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1829 ("%s: invalid rights", func));
1830 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1831 ("%s: invalid fcntls", func));
1832 KASSERT(fcaps->fc_fcntls == 0 ||
1833 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1834 ("%s: fcntls without CAP_FCNTL", func));
1835 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1836 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1837 ("%s: invalid ioctls", func));
1838 KASSERT(fcaps->fc_nioctls == 0 ||
1839 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1840 ("%s: ioctls without CAP_IOCTL", func));
1844 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1848 FILEDESC_XLOCK_ASSERT(fdp);
1850 nfd1 = fdp->fd_nfiles * 2;
1853 fdgrowtable(fdp, nfd1);
1857 * Grow the file table to accommodate (at least) nfd descriptors.
1860 fdgrowtable(struct filedesc *fdp, int nfd)
1862 struct filedesc0 *fdp0;
1863 struct freetable *ft;
1864 struct fdescenttbl *ntable;
1865 struct fdescenttbl *otable;
1866 int nnfiles, onfiles;
1867 NDSLOTTYPE *nmap, *omap;
1869 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1871 /* save old values */
1872 onfiles = fdp->fd_nfiles;
1873 otable = fdp->fd_files;
1876 /* compute the size of the new table */
1877 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1878 if (nnfiles <= onfiles)
1879 /* the table is already large enough */
1883 * Allocate a new table. We need enough space for the number of
1884 * entries, file entries themselves and the struct freetable we will use
1885 * when we decommission the table and place it on the freelist.
1886 * We place the struct freetable in the middle so we don't have
1887 * to worry about padding.
1889 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1890 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1891 sizeof(struct freetable),
1892 M_FILEDESC, M_ZERO | M_WAITOK);
1893 /* copy the old data */
1894 ntable->fdt_nfiles = nnfiles;
1895 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1896 onfiles * sizeof(ntable->fdt_ofiles[0]));
1899 * Allocate a new map only if the old is not large enough. It will
1900 * grow at a slower rate than the table as it can map more
1901 * entries than the table can hold.
1903 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1904 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1906 /* copy over the old data and update the pointer */
1907 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1912 * Make sure that ntable is correctly initialized before we replace
1913 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1916 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1919 * Free the old file table when not shared by other threads or processes.
1920 * The old file table is considered to be shared when either are true:
1921 * - The process has more than one thread.
1922 * - The file descriptor table has been shared via fdshare().
1924 * When shared, the old file table will be placed on a freelist
1925 * which will be processed when the struct filedesc is released.
1927 * Note that if onfiles == NDFILE, we're dealing with the original
1928 * static allocation contained within (struct filedesc0 *)fdp,
1929 * which must not be freed.
1931 if (onfiles > NDFILE) {
1933 * Note we may be called here from fdinit while allocating a
1934 * table for a new process in which case ->p_fd points
1937 if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) {
1938 free(otable, M_FILEDESC);
1940 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1941 fdp0 = (struct filedesc0 *)fdp;
1942 ft->ft_table = otable;
1943 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1947 * The map does not have the same possibility of threads still
1948 * holding references to it. So always free it as long as it
1949 * does not reference the original static allocation.
1951 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1952 free(omap, M_FILEDESC);
1956 * Allocate a file descriptor for the process.
1959 fdalloc(struct thread *td, int minfd, int *result)
1961 struct proc *p = td->td_proc;
1962 struct filedesc *fdp = p->p_fd;
1963 int fd, maxfd, allocfd;
1968 FILEDESC_XLOCK_ASSERT(fdp);
1970 if (fdp->fd_freefile > minfd)
1971 minfd = fdp->fd_freefile;
1973 maxfd = getmaxfd(td);
1976 * Search the bitmap for a free descriptor starting at minfd.
1977 * If none is found, grow the file table.
1979 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1980 if (__predict_false(fd >= maxfd))
1982 if (__predict_false(fd >= fdp->fd_nfiles)) {
1983 allocfd = min(fd * 2, maxfd);
1985 if (RACCT_ENABLED()) {
1986 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1992 * fd is already equal to first free descriptor >= minfd, so
1993 * we only need to grow the table and we are done.
1995 fdgrowtable_exp(fdp, allocfd);
1999 * Perform some sanity checks, then mark the file descriptor as
2000 * used and return it to the caller.
2002 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
2003 ("invalid descriptor %d", fd));
2004 KASSERT(!fdisused(fdp, fd),
2005 ("fd_first_free() returned non-free descriptor"));
2006 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
2007 ("file descriptor isn't free"));
2014 * Allocate n file descriptors for the process.
2017 fdallocn(struct thread *td, int minfd, int *fds, int n)
2019 struct proc *p = td->td_proc;
2020 struct filedesc *fdp = p->p_fd;
2023 FILEDESC_XLOCK_ASSERT(fdp);
2025 for (i = 0; i < n; i++)
2026 if (fdalloc(td, 0, &fds[i]) != 0)
2030 for (i--; i >= 0; i--)
2031 fdunused(fdp, fds[i]);
2039 * Create a new open file structure and allocate a file descriptor for the
2040 * process that refers to it. We add one reference to the file for the
2041 * descriptor table and one reference for resultfp. This is to prevent us
2042 * being preempted and the entry in the descriptor table closed after we
2043 * release the FILEDESC lock.
2046 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
2047 struct filecaps *fcaps)
2052 MPASS(resultfp != NULL);
2053 MPASS(resultfd != NULL);
2055 error = _falloc_noinstall(td, &fp, 2);
2056 if (__predict_false(error != 0)) {
2060 error = finstall_refed(td, fp, &fd, flags, fcaps);
2061 if (__predict_false(error != 0)) {
2062 falloc_abort(td, fp);
2073 * Create a new open file structure without allocating a file descriptor.
2076 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2079 int maxuserfiles = maxfiles - (maxfiles / 20);
2081 static struct timeval lastfail;
2084 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2087 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2088 if ((openfiles_new >= maxuserfiles &&
2089 priv_check(td, PRIV_MAXFILES) != 0) ||
2090 openfiles_new >= maxfiles) {
2091 atomic_subtract_int(&openfiles, 1);
2092 if (ppsratecheck(&lastfail, &curfail, 1)) {
2093 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2094 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2098 fp = uma_zalloc(file_zone, M_WAITOK);
2099 bzero(fp, sizeof(*fp));
2100 refcount_init(&fp->f_count, n);
2101 fp->f_cred = crhold(td->td_ucred);
2102 fp->f_ops = &badfileops;
2108 falloc_abort(struct thread *td, struct file *fp)
2112 * For assertion purposes.
2114 refcount_init(&fp->f_count, 0);
2119 * Install a file in a file descriptor table.
2122 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2123 struct filecaps *fcaps)
2125 struct filedescent *fde;
2129 filecaps_validate(fcaps, __func__);
2130 FILEDESC_XLOCK_ASSERT(fdp);
2132 fde = &fdp->fd_ofiles[fd];
2134 seqc_write_begin(&fde->fde_seqc);
2137 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2139 filecaps_move(fcaps, &fde->fde_caps);
2141 filecaps_fill(&fde->fde_caps);
2143 seqc_write_end(&fde->fde_seqc);
2148 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2149 struct filecaps *fcaps)
2151 struct filedesc *fdp = td->td_proc->p_fd;
2156 FILEDESC_XLOCK(fdp);
2157 error = fdalloc(td, 0, fd);
2158 if (__predict_true(error == 0)) {
2159 _finstall(fdp, fp, *fd, flags, fcaps);
2161 FILEDESC_XUNLOCK(fdp);
2166 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2167 struct filecaps *fcaps)
2175 error = finstall_refed(td, fp, fd, flags, fcaps);
2176 if (__predict_false(error != 0)) {
2183 * Build a new filedesc structure from another.
2185 * If fdp is not NULL, return with it shared locked.
2190 struct filedesc0 *newfdp0;
2191 struct filedesc *newfdp;
2193 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2194 newfdp = &newfdp0->fd_fd;
2196 /* Create the file descriptor table. */
2197 FILEDESC_LOCK_INIT(newfdp);
2198 refcount_init(&newfdp->fd_refcnt, 1);
2199 refcount_init(&newfdp->fd_holdcnt, 1);
2200 newfdp->fd_map = newfdp0->fd_dmap;
2201 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2202 newfdp->fd_files->fdt_nfiles = NDFILE;
2208 * Build a pwddesc structure from another.
2209 * Copy the current, root, and jail root vnode references.
2211 * If pdp is not NULL, return with it shared locked.
2214 pdinit(struct pwddesc *pdp, bool keeplock)
2216 struct pwddesc *newpdp;
2219 newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2221 PWDDESC_LOCK_INIT(newpdp);
2222 refcount_init(&newpdp->pd_refcount, 1);
2223 newpdp->pd_cmask = CMASK;
2226 newpwd = pwd_alloc();
2227 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2232 newpwd = pwd_hold_pwddesc(pdp);
2233 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2235 PWDDESC_XUNLOCK(pdp);
2240 * Hold either filedesc or pwddesc of the passed process.
2242 * The process lock is used to synchronize against the target exiting and
2245 * Clearing can be ilustrated in 3 steps:
2246 * 1. set the pointer to NULL. Either routine can race against it, hence
2248 * 2. observe the process lock as not taken. Until then fdhold/pdhold can
2249 * race to either still see the pointer or find NULL. It is still safe to
2250 * grab a reference as clearing is stalled.
2251 * 3. after the lock is observed as not taken, any fdhold/pdhold calls are
2252 * guaranteed to see NULL, making it safe to finish clearing
2254 static struct filedesc *
2255 fdhold(struct proc *p)
2257 struct filedesc *fdp;
2259 PROC_LOCK_ASSERT(p, MA_OWNED);
2260 fdp = atomic_load_ptr(&p->p_fd);
2262 refcount_acquire(&fdp->fd_holdcnt);
2266 static struct pwddesc *
2267 pdhold(struct proc *p)
2269 struct pwddesc *pdp;
2271 PROC_LOCK_ASSERT(p, MA_OWNED);
2272 pdp = atomic_load_ptr(&p->p_pd);
2274 refcount_acquire(&pdp->pd_refcount);
2279 fddrop(struct filedesc *fdp)
2282 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2283 if (refcount_release(&fdp->fd_holdcnt) == 0)
2287 FILEDESC_LOCK_DESTROY(fdp);
2288 uma_zfree(filedesc0_zone, fdp);
2292 pddrop(struct pwddesc *pdp)
2296 if (refcount_release_if_not_last(&pdp->pd_refcount))
2300 if (refcount_release(&pdp->pd_refcount) == 0) {
2301 PWDDESC_XUNLOCK(pdp);
2304 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2306 PWDDESC_XUNLOCK(pdp);
2309 PWDDESC_LOCK_DESTROY(pdp);
2310 free(pdp, M_PWDDESC);
2314 * Share a filedesc structure.
2317 fdshare(struct filedesc *fdp)
2320 refcount_acquire(&fdp->fd_refcnt);
2325 * Share a pwddesc structure.
2328 pdshare(struct pwddesc *pdp)
2330 refcount_acquire(&pdp->pd_refcount);
2335 * Unshare a filedesc structure, if necessary by making a copy
2338 fdunshare(struct thread *td)
2340 struct filedesc *tmp;
2341 struct proc *p = td->td_proc;
2343 if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2346 tmp = fdcopy(p->p_fd);
2352 * Unshare a pwddesc structure.
2355 pdunshare(struct thread *td)
2357 struct pwddesc *pdp;
2362 if (refcount_load(&p->p_pd->pd_refcount) == 1)
2365 pdp = pdcopy(p->p_pd);
2371 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2372 * this is to ease callers, not catch errors.
2375 fdcopy(struct filedesc *fdp)
2377 struct filedesc *newfdp;
2378 struct filedescent *nfde, *ofde;
2384 FILEDESC_SLOCK(fdp);
2386 lastfile = fdlastfile(fdp);
2387 if (lastfile < newfdp->fd_nfiles)
2389 FILEDESC_SUNLOCK(fdp);
2390 fdgrowtable(newfdp, lastfile + 1);
2391 FILEDESC_SLOCK(fdp);
2393 /* copy all passable descriptors (i.e. not kqueue) */
2394 newfdp->fd_freefile = fdp->fd_freefile;
2395 FILEDESC_FOREACH_FDE(fdp, i, ofde) {
2396 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2397 !fhold(ofde->fde_file)) {
2398 if (newfdp->fd_freefile == fdp->fd_freefile)
2399 newfdp->fd_freefile = i;
2402 nfde = &newfdp->fd_ofiles[i];
2404 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2405 fdused_init(newfdp, i);
2407 MPASS(newfdp->fd_freefile != -1);
2408 FILEDESC_SUNLOCK(fdp);
2413 * Copy a pwddesc structure.
2416 pdcopy(struct pwddesc *pdp)
2418 struct pwddesc *newpdp;
2422 newpdp = pdinit(pdp, true);
2423 newpdp->pd_cmask = pdp->pd_cmask;
2424 PWDDESC_XUNLOCK(pdp);
2429 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2430 * one of processes using it exits) and the table used to be shared.
2433 fdclearlocks(struct thread *td)
2435 struct filedesc *fdp;
2436 struct filedesc_to_leader *fdtol;
2446 MPASS(fdtol != NULL);
2448 FILEDESC_XLOCK(fdp);
2449 KASSERT(fdtol->fdl_refcount > 0,
2450 ("filedesc_to_refcount botch: fdl_refcount=%d",
2451 fdtol->fdl_refcount));
2452 if (fdtol->fdl_refcount == 1 &&
2453 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2454 FILEDESC_FOREACH_FP(fdp, i, fp) {
2455 if (fp->f_type != DTYPE_VNODE ||
2458 FILEDESC_XUNLOCK(fdp);
2459 lf.l_whence = SEEK_SET;
2462 lf.l_type = F_UNLCK;
2464 (void) VOP_ADVLOCK(vp,
2465 (caddr_t)p->p_leader, F_UNLCK,
2467 FILEDESC_XLOCK(fdp);
2472 if (fdtol->fdl_refcount == 1) {
2473 if (fdp->fd_holdleaderscount > 0 &&
2474 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2476 * close() or kern_dup() has cleared a reference
2477 * in a shared file descriptor table.
2479 fdp->fd_holdleaderswakeup = 1;
2480 sx_sleep(&fdp->fd_holdleaderscount,
2481 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2484 if (fdtol->fdl_holdcount > 0) {
2486 * Ensure that fdtol->fdl_leader remains
2487 * valid in closef().
2489 fdtol->fdl_wakeup = 1;
2490 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2495 fdtol->fdl_refcount--;
2496 if (fdtol->fdl_refcount == 0 &&
2497 fdtol->fdl_holdcount == 0) {
2498 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2499 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2503 FILEDESC_XUNLOCK(fdp);
2505 free(fdtol, M_FILEDESC_TO_LEADER);
2509 * Release a filedesc structure.
2512 fdescfree_fds(struct thread *td, struct filedesc *fdp)
2514 struct filedesc0 *fdp0;
2515 struct freetable *ft, *tft;
2516 struct filedescent *fde;
2520 KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2521 ("%s: fd table %p carries references", __func__, fdp));
2524 * Serialize with threads iterating over the table, if any.
2526 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2527 FILEDESC_XLOCK(fdp);
2528 FILEDESC_XUNLOCK(fdp);
2531 FILEDESC_FOREACH_FDE(fdp, i, fde) {
2534 (void) closef(fp, td);
2537 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2538 free(fdp->fd_map, M_FILEDESC);
2539 if (fdp->fd_nfiles > NDFILE)
2540 free(fdp->fd_files, M_FILEDESC);
2542 fdp0 = (struct filedesc0 *)fdp;
2543 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2544 free(ft->ft_table, M_FILEDESC);
2550 fdescfree(struct thread *td)
2553 struct filedesc *fdp;
2560 if (RACCT_ENABLED())
2561 racct_set_unlocked(p, RACCT_NOFILE, 0);
2564 if (p->p_fdtol != NULL)
2568 * Check fdhold for an explanation.
2570 atomic_store_ptr(&p->p_fd, NULL);
2571 atomic_thread_fence_seq_cst();
2572 PROC_WAIT_UNLOCKED(p);
2574 if (refcount_release(&fdp->fd_refcnt) == 0)
2577 fdescfree_fds(td, fdp);
2581 pdescfree(struct thread *td)
2584 struct pwddesc *pdp;
2591 * Check pdhold for an explanation.
2593 atomic_store_ptr(&p->p_pd, NULL);
2594 atomic_thread_fence_seq_cst();
2595 PROC_WAIT_UNLOCKED(p);
2601 * For setugid programs, we don't want to people to use that setugidness
2602 * to generate error messages which write to a file which otherwise would
2603 * otherwise be off-limits to the process. We check for filesystems where
2604 * the vnode can change out from under us after execve (like [lin]procfs).
2606 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2607 * sufficient. We also don't check for setugidness since we know we are.
2610 is_unsafe(struct file *fp)
2614 if (fp->f_type != DTYPE_VNODE)
2618 return ((vp->v_vflag & VV_PROCDEP) != 0);
2622 * Make this setguid thing safe, if at all possible.
2625 fdsetugidsafety(struct thread *td)
2627 struct filedesc *fdp;
2631 fdp = td->td_proc->p_fd;
2632 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2633 ("the fdtable should not be shared"));
2634 MPASS(fdp->fd_nfiles >= 3);
2635 for (i = 0; i <= 2; i++) {
2636 fp = fdp->fd_ofiles[i].fde_file;
2637 if (fp != NULL && is_unsafe(fp)) {
2638 FILEDESC_XLOCK(fdp);
2639 knote_fdclose(td, i);
2641 * NULL-out descriptor prior to close to avoid
2642 * a race while close blocks.
2645 FILEDESC_XUNLOCK(fdp);
2646 (void) closef(fp, td);
2652 * If a specific file object occupies a specific file descriptor, close the
2653 * file descriptor entry and drop a reference on the file object. This is a
2654 * convenience function to handle a subsequent error in a function that calls
2655 * falloc() that handles the race that another thread might have closed the
2656 * file descriptor out from under the thread creating the file object.
2659 fdclose(struct thread *td, struct file *fp, int idx)
2661 struct filedesc *fdp = td->td_proc->p_fd;
2663 FILEDESC_XLOCK(fdp);
2664 if (fdp->fd_ofiles[idx].fde_file == fp) {
2666 FILEDESC_XUNLOCK(fdp);
2669 FILEDESC_XUNLOCK(fdp);
2673 * Close any files on exec?
2676 fdcloseexec(struct thread *td)
2678 struct filedesc *fdp;
2679 struct filedescent *fde;
2683 fdp = td->td_proc->p_fd;
2684 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2685 ("the fdtable should not be shared"));
2686 FILEDESC_FOREACH_FDE(fdp, i, fde) {
2688 if (fp->f_type == DTYPE_MQUEUE ||
2689 (fde->fde_flags & UF_EXCLOSE)) {
2690 FILEDESC_XLOCK(fdp);
2692 (void) closefp(fdp, i, fp, td, false, false);
2693 FILEDESC_UNLOCK_ASSERT(fdp);
2699 * It is unsafe for set[ug]id processes to be started with file
2700 * descriptors 0..2 closed, as these descriptors are given implicit
2701 * significance in the Standard C library. fdcheckstd() will create a
2702 * descriptor referencing /dev/null for each of stdin, stdout, and
2703 * stderr that is not already open.
2706 fdcheckstd(struct thread *td)
2708 struct filedesc *fdp;
2710 int i, error, devnull;
2712 fdp = td->td_proc->p_fd;
2713 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2714 ("the fdtable should not be shared"));
2715 MPASS(fdp->fd_nfiles >= 3);
2717 for (i = 0; i <= 2; i++) {
2718 if (fdp->fd_ofiles[i].fde_file != NULL)
2721 save = td->td_retval[0];
2722 if (devnull != -1) {
2723 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2725 error = kern_openat(td, AT_FDCWD, "/dev/null",
2726 UIO_SYSSPACE, O_RDWR, 0);
2728 devnull = td->td_retval[0];
2729 KASSERT(devnull == i, ("we didn't get our fd"));
2732 td->td_retval[0] = save;
2740 * Internal form of close. Decrement reference count on file structure.
2741 * Note: td may be NULL when closing a file that was being passed in a
2745 closef(struct file *fp, struct thread *td)
2749 struct filedesc_to_leader *fdtol;
2750 struct filedesc *fdp;
2755 * POSIX record locking dictates that any close releases ALL
2756 * locks owned by this process. This is handled by setting
2757 * a flag in the unlock to free ONLY locks obeying POSIX
2758 * semantics, and not to free BSD-style file locks.
2759 * If the descriptor was in a message, POSIX-style locks
2760 * aren't passed with the descriptor, and the thread pointer
2761 * will be NULL. Callers should be careful only to pass a
2762 * NULL thread pointer when there really is no owning
2763 * context that might have locks, or the locks will be
2766 if (fp->f_type == DTYPE_VNODE) {
2768 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2769 lf.l_whence = SEEK_SET;
2772 lf.l_type = F_UNLCK;
2773 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2774 F_UNLCK, &lf, F_POSIX);
2776 fdtol = td->td_proc->p_fdtol;
2777 if (fdtol != NULL) {
2779 * Handle special case where file descriptor table is
2780 * shared between multiple process leaders.
2782 fdp = td->td_proc->p_fd;
2783 FILEDESC_XLOCK(fdp);
2784 for (fdtol = fdtol->fdl_next;
2785 fdtol != td->td_proc->p_fdtol;
2786 fdtol = fdtol->fdl_next) {
2787 if ((fdtol->fdl_leader->p_flag &
2790 fdtol->fdl_holdcount++;
2791 FILEDESC_XUNLOCK(fdp);
2792 lf.l_whence = SEEK_SET;
2795 lf.l_type = F_UNLCK;
2797 (void) VOP_ADVLOCK(vp,
2798 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2800 FILEDESC_XLOCK(fdp);
2801 fdtol->fdl_holdcount--;
2802 if (fdtol->fdl_holdcount == 0 &&
2803 fdtol->fdl_wakeup != 0) {
2804 fdtol->fdl_wakeup = 0;
2808 FILEDESC_XUNLOCK(fdp);
2811 return (fdrop_close(fp, td));
2815 * Hack for file descriptor passing code.
2818 closef_nothread(struct file *fp)
2825 * Initialize the file pointer with the specified properties.
2827 * The ops are set with release semantics to be certain that the flags, type,
2828 * and data are visible when ops is. This is to prevent ops methods from being
2829 * called with bad data.
2832 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2837 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2841 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2843 fp->f_seqcount[UIO_READ] = 1;
2844 fp->f_seqcount[UIO_WRITE] = 1;
2845 finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2850 fget_cap_noref(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2851 struct file **fpp, struct filecaps *havecapsp)
2853 struct filedescent *fde;
2856 FILEDESC_LOCK_ASSERT(fdp);
2859 fde = fdeget_noref(fdp, fd);
2866 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2871 if (havecapsp != NULL)
2872 filecaps_copy(&fde->fde_caps, havecapsp, true);
2874 *fpp = fde->fde_file;
2883 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2884 struct file **fpp, struct filecaps *havecapsp)
2886 struct filedesc *fdp = td->td_proc->p_fd;
2893 error = fget_unlocked_seq(td, fd, needrightsp, &fp, &seq);
2897 if (havecapsp != NULL) {
2898 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2899 havecapsp, false)) {
2905 if (!fd_modified(fdp, fd, seq))
2914 FILEDESC_SLOCK(fdp);
2915 error = fget_cap_noref(fdp, fd, needrightsp, fpp, havecapsp);
2916 if (error == 0 && !fhold(*fpp))
2918 FILEDESC_SUNLOCK(fdp);
2923 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2924 struct file **fpp, struct filecaps *havecapsp)
2927 error = fget_unlocked(td, fd, needrightsp, fpp);
2928 if (havecapsp != NULL && error == 0)
2929 filecaps_fill(havecapsp);
2937 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2939 const struct filedescent *fde;
2940 const struct fdescenttbl *fdt;
2941 struct filedesc *fdp;
2944 const cap_rights_t *haverights;
2945 cap_rights_t rights;
2948 VFS_SMR_ASSERT_ENTERED();
2950 rights = *ndp->ni_rightsneeded;
2951 cap_rights_set_one(&rights, CAP_LOOKUP);
2953 fdp = curproc->p_fd;
2954 fdt = fdp->fd_files;
2955 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2957 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
2958 fde = &fdt->fdt_ofiles[fd];
2959 haverights = cap_rights_fde_inline(fde);
2961 if (__predict_false(fp == NULL))
2963 if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2965 *fsearch = ((fp->f_flag & FSEARCH) != 0);
2967 if (__predict_false(vp == NULL)) {
2970 if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2974 * Use an acquire barrier to force re-reading of fdt so it is
2975 * refreshed for verification.
2977 atomic_thread_fence_acq();
2978 fdt = fdp->fd_files;
2979 if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)))
2982 * If file descriptor doesn't have all rights,
2983 * all lookups relative to it must also be
2984 * strictly relative.
2986 * Not yet supported by fast path.
2989 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
2990 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
2991 ndp->ni_filecaps.fc_nioctls != -1) {
2993 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
3003 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
3005 const struct fdescenttbl *fdt;
3006 struct filedesc *fdp;
3010 VFS_SMR_ASSERT_ENTERED();
3012 fdp = curproc->p_fd;
3013 fdt = fdp->fd_files;
3014 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3016 fp = fdt->fdt_ofiles[fd].fde_file;
3017 if (__predict_false(fp == NULL))
3019 *fsearch = ((fp->f_flag & FSEARCH) != 0);
3021 if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
3025 * Use an acquire barrier to force re-reading of fdt so it is
3026 * refreshed for verification.
3028 atomic_thread_fence_acq();
3029 fdt = fdp->fd_files;
3030 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3032 filecaps_fill(&ndp->ni_filecaps);
3039 * Fetch the descriptor locklessly.
3041 * We avoid fdrop() races by never raising a refcount above 0. To accomplish
3042 * this we have to use a cmpset loop rather than an atomic_add. The descriptor
3043 * must be re-verified once we acquire a reference to be certain that the
3044 * identity is still correct and we did not lose a race due to preemption.
3046 * Force a reload of fdt when looping. Another thread could reallocate
3047 * the table before this fd was closed, so it is possible that there is
3048 * a stale fp pointer in cached version.
3052 fget_unlocked_seq(struct thread *td, int fd, cap_rights_t *needrightsp,
3053 struct file **fpp, seqc_t *seqp)
3055 struct filedesc *fdp;
3056 const struct filedescent *fde;
3057 const struct fdescenttbl *fdt;
3060 cap_rights_t haverights;
3063 fdp = td->td_proc->p_fd;
3064 fdt = fdp->fd_files;
3065 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3069 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3070 fde = &fdt->fdt_ofiles[fd];
3071 haverights = *cap_rights_fde_inline(fde);
3073 if (__predict_false(fp == NULL)) {
3074 if (seqc_consistent(fd_seqc(fdt, fd), seq))
3076 fdt = atomic_load_ptr(&fdp->fd_files);
3079 error = cap_check_inline(&haverights, needrightsp);
3080 if (__predict_false(error != 0)) {
3081 if (seqc_consistent(fd_seqc(fdt, fd), seq))
3083 fdt = atomic_load_ptr(&fdp->fd_files);
3086 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3087 fdt = atomic_load_ptr(&fdp->fd_files);
3091 * Use an acquire barrier to force re-reading of fdt so it is
3092 * refreshed for verification.
3094 atomic_thread_fence_acq();
3095 fdt = fdp->fd_files;
3096 if (seqc_consistent_no_fence(fd_seqc(fdt, fd), seq))
3108 fget_unlocked_seq(struct thread *td, int fd, cap_rights_t *needrightsp,
3109 struct file **fpp, seqc_t *seqp __unused)
3111 struct filedesc *fdp;
3112 const struct fdescenttbl *fdt;
3115 fdp = td->td_proc->p_fd;
3116 fdt = fdp->fd_files;
3117 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3121 fp = fdt->fdt_ofiles[fd].fde_file;
3122 if (__predict_false(fp == NULL))
3124 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3125 fdt = atomic_load_ptr(&fdp->fd_files);
3129 * Use an acquire barrier to force re-reading of fdt so it is
3130 * refreshed for verification.
3132 atomic_thread_fence_acq();
3133 fdt = fdp->fd_files;
3134 if (__predict_true(fp == fdt->fdt_ofiles[fd].fde_file))
3144 * See the comments in fget_unlocked_seq for an explanation of how this works.
3146 * This is a simplified variant which bails out to the aforementioned routine
3147 * if anything goes wrong. In practice this only happens when userspace is
3148 * racing with itself.
3151 fget_unlocked(struct thread *td, int fd, cap_rights_t *needrightsp,
3154 struct filedesc *fdp;
3156 const struct filedescent *fde;
3158 const struct fdescenttbl *fdt;
3162 const cap_rights_t *haverights;
3165 fdp = td->td_proc->p_fd;
3166 fdt = fdp->fd_files;
3167 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) {
3172 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3173 fde = &fdt->fdt_ofiles[fd];
3174 haverights = cap_rights_fde_inline(fde);
3177 fp = fdt->fdt_ofiles[fd].fde_file;
3179 if (__predict_false(fp == NULL))
3182 if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3185 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3189 * Use an acquire barrier to force re-reading of fdt so it is
3190 * refreshed for verification.
3192 atomic_thread_fence_acq();
3193 fdt = fdp->fd_files;
3195 if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)))
3197 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3206 return (fget_unlocked_seq(td, fd, needrightsp, fpp, NULL));
3210 * Translate fd -> file when the caller guarantees the file descriptor table
3211 * can't be changed by others.
3213 * Note this does not mean the file object itself is only visible to the caller,
3214 * merely that it wont disappear without having to be referenced.
3216 * Must be paired with fput_only_user.
3220 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3223 const struct filedescent *fde;
3224 const struct fdescenttbl *fdt;
3225 const cap_rights_t *haverights;
3229 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3232 if (__predict_false(fd >= fdp->fd_nfiles))
3235 fdt = fdp->fd_files;
3236 fde = &fdt->fdt_ofiles[fd];
3238 if (__predict_false(fp == NULL))
3240 MPASS(refcount_load(&fp->f_count) > 0);
3241 haverights = cap_rights_fde_inline(fde);
3242 error = cap_check_inline(haverights, needrightsp);
3243 if (__predict_false(error != 0))
3250 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3255 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3258 if (__predict_false(fd >= fdp->fd_nfiles))
3261 fp = fdp->fd_ofiles[fd].fde_file;
3262 if (__predict_false(fp == NULL))
3265 MPASS(refcount_load(&fp->f_count) > 0);
3272 * Extract the file pointer associated with the specified descriptor for the
3273 * current user process.
3275 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3278 * File's rights will be checked against the capability rights mask.
3280 * If an error occurred the non-zero error is returned and *fpp is set to
3281 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
3282 * responsible for fdrop().
3285 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3286 cap_rights_t *needrightsp)
3292 error = fget_unlocked(td, fd, needrightsp, &fp);
3293 if (__predict_false(error != 0))
3295 if (__predict_false(fp->f_ops == &badfileops)) {
3301 * FREAD and FWRITE failure return EBADF as per POSIX.
3307 if ((fp->f_flag & flags) == 0)
3311 if (fp->f_ops != &path_fileops &&
3312 ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3313 (fp->f_flag & FWRITE) != 0))
3319 KASSERT(0, ("wrong flags"));
3332 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3335 return (_fget(td, fd, fpp, 0, rightsp));
3339 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3343 #ifndef CAPABILITIES
3344 error = _fget(td, fd, fpp, 0, rightsp);
3345 if (maxprotp != NULL)
3346 *maxprotp = VM_PROT_ALL;
3349 cap_rights_t fdrights;
3350 struct filedesc *fdp;
3355 fdp = td->td_proc->p_fd;
3356 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3358 error = fget_unlocked_seq(td, fd, rightsp, &fp, &seq);
3359 if (__predict_false(error != 0))
3361 if (__predict_false(fp->f_ops == &badfileops)) {
3365 if (maxprotp != NULL)
3366 fdrights = *cap_rights(fdp, fd);
3367 if (!fd_modified(fdp, fd, seq))
3373 * If requested, convert capability rights to access flags.
3375 if (maxprotp != NULL)
3376 *maxprotp = cap_rights_to_vmprot(&fdrights);
3383 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3386 return (_fget(td, fd, fpp, FREAD, rightsp));
3390 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3393 return (_fget(td, fd, fpp, FWRITE, rightsp));
3397 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3400 #ifndef CAPABILITIES
3401 return (fget_unlocked(td, fd, rightsp, fpp));
3403 struct filedesc *fdp = td->td_proc->p_fd;
3409 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3411 error = fget_unlocked_seq(td, fd, rightsp, &fp, &seq);
3414 error = cap_fcntl_check(fdp, fd, needfcntl);
3415 if (!fd_modified(fdp, fd, seq))
3429 * Like fget() but loads the underlying vnode, or returns an error if the
3430 * descriptor does not represent a vnode. Note that pipes use vnodes but
3431 * never have VM objects. The returned vnode will be vref()'d.
3433 * XXX: what about the unused flags ?
3436 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3443 error = _fget(td, fd, &fp, flags, needrightsp);
3446 if (fp->f_vnode == NULL) {
3458 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3461 return (_fgetvp(td, fd, 0, rightsp, vpp));
3465 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3466 struct filecaps *havecaps, struct vnode **vpp)
3468 struct filecaps caps;
3472 error = fget_cap(td, fd, needrightsp, &fp, &caps);
3475 if (fp->f_ops == &badfileops) {
3479 if (fp->f_vnode == NULL) {
3491 filecaps_free(&caps);
3497 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3500 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3504 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3507 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3512 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3516 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3521 * Handle the last reference to a file being closed.
3523 * Without the noinline attribute clang keeps inlining the func thorough this
3524 * file when fdrop is used.
3527 _fdrop(struct file *fp, struct thread *td)
3533 count = refcount_load(&fp->f_count);
3535 panic("fdrop: fp %p count %d", fp, count);
3537 error = fo_close(fp, td);
3538 atomic_subtract_int(&openfiles, 1);
3540 free(fp->f_advice, M_FADVISE);
3541 uma_zfree(file_zone, fp);
3547 * Apply an advisory lock on a file descriptor.
3549 * Just attempt to get a record lock of the requested type on the entire file
3550 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3552 #ifndef _SYS_SYSPROTO_H_
3560 sys_flock(struct thread *td, struct flock_args *uap)
3567 error = fget(td, uap->fd, &cap_flock_rights, &fp);
3571 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
3574 if (fp->f_ops == &path_fileops) {
3580 lf.l_whence = SEEK_SET;
3583 if (uap->how & LOCK_UN) {
3584 lf.l_type = F_UNLCK;
3585 atomic_clear_int(&fp->f_flag, FHASLOCK);
3586 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3589 if (uap->how & LOCK_EX)
3590 lf.l_type = F_WRLCK;
3591 else if (uap->how & LOCK_SH)
3592 lf.l_type = F_RDLCK;
3597 atomic_set_int(&fp->f_flag, FHASLOCK);
3598 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3599 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3605 * Duplicate the specified descriptor to a free descriptor.
3608 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3609 int openerror, int *indxp)
3611 struct filedescent *newfde, *oldfde;
3616 KASSERT(openerror == ENODEV || openerror == ENXIO,
3617 ("unexpected error %d in %s", openerror, __func__));
3620 * If the to-be-dup'd fd number is greater than the allowed number
3621 * of file descriptors, or the fd to be dup'd has already been
3622 * closed, then reject.
3624 FILEDESC_XLOCK(fdp);
3625 if ((fp = fget_noref(fdp, dfd)) == NULL) {
3626 FILEDESC_XUNLOCK(fdp);
3630 error = fdalloc(td, 0, &indx);
3632 FILEDESC_XUNLOCK(fdp);
3637 * There are two cases of interest here.
3639 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3641 * For ENXIO steal away the file structure from (dfd) and store it in
3642 * (indx). (dfd) is effectively closed by this operation.
3644 switch (openerror) {
3647 * Check that the mode the file is being opened for is a
3648 * subset of the mode of the existing descriptor.
3650 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3651 fdunused(fdp, indx);
3652 FILEDESC_XUNLOCK(fdp);
3656 fdunused(fdp, indx);
3657 FILEDESC_XUNLOCK(fdp);
3660 newfde = &fdp->fd_ofiles[indx];
3661 oldfde = &fdp->fd_ofiles[dfd];
3662 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3664 seqc_write_begin(&newfde->fde_seqc);
3666 fde_copy(oldfde, newfde);
3667 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3670 seqc_write_end(&newfde->fde_seqc);
3675 * Steal away the file pointer from dfd and stuff it into indx.
3677 newfde = &fdp->fd_ofiles[indx];
3678 oldfde = &fdp->fd_ofiles[dfd];
3680 seqc_write_begin(&oldfde->fde_seqc);
3681 seqc_write_begin(&newfde->fde_seqc);
3683 fde_copy(oldfde, newfde);
3684 oldfde->fde_file = NULL;
3687 seqc_write_end(&newfde->fde_seqc);
3688 seqc_write_end(&oldfde->fde_seqc);
3692 FILEDESC_XUNLOCK(fdp);
3698 * This sysctl determines if we will allow a process to chroot(2) if it
3699 * has a directory open:
3700 * 0: disallowed for all processes.
3701 * 1: allowed for processes that were not already chroot(2)'ed.
3702 * 2: allowed for all processes.
3705 static int chroot_allow_open_directories = 1;
3707 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3708 &chroot_allow_open_directories, 0,
3709 "Allow a process to chroot(2) if it has a directory open");
3712 * Helper function for raised chroot(2) security function: Refuse if
3713 * any filedescriptors are open directories.
3716 chroot_refuse_vdir_fds(struct filedesc *fdp)
3722 FILEDESC_LOCK_ASSERT(fdp);
3724 FILEDESC_FOREACH_FP(fdp, i, fp) {
3725 if (fp->f_type == DTYPE_VNODE) {
3727 if (vp->v_type == VDIR)
3735 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3738 if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3739 vrefact(oldpwd->pwd_cdir);
3740 newpwd->pwd_cdir = oldpwd->pwd_cdir;
3743 if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3744 vrefact(oldpwd->pwd_rdir);
3745 newpwd->pwd_rdir = oldpwd->pwd_rdir;
3748 if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3749 vrefact(oldpwd->pwd_jdir);
3750 newpwd->pwd_jdir = oldpwd->pwd_jdir;
3755 pwd_hold_pwddesc(struct pwddesc *pdp)
3759 PWDDESC_ASSERT_XLOCKED(pdp);
3760 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3762 refcount_acquire(&pwd->pwd_refcount);
3767 pwd_hold_smr(struct pwd *pwd)
3771 if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3778 pwd_hold(struct thread *td)
3780 struct pwddesc *pdp;
3783 pdp = td->td_proc->p_pd;
3786 pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3787 if (pwd_hold_smr(pwd)) {
3793 pwd = pwd_hold_pwddesc(pdp);
3795 PWDDESC_XUNLOCK(pdp);
3800 pwd_hold_proc(struct proc *p)
3802 struct pwddesc *pdp;
3805 PROC_ASSERT_HELD(p);
3812 pwd = pwd_hold_pwddesc(pdp);
3814 PWDDESC_XUNLOCK(pdp);
3824 pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3825 bzero(pwd, sizeof(*pwd));
3826 refcount_init(&pwd->pwd_refcount, 1);
3831 pwd_drop(struct pwd *pwd)
3834 if (!refcount_release(&pwd->pwd_refcount))
3837 if (pwd->pwd_cdir != NULL)
3838 vrele(pwd->pwd_cdir);
3839 if (pwd->pwd_rdir != NULL)
3840 vrele(pwd->pwd_rdir);
3841 if (pwd->pwd_jdir != NULL)
3842 vrele(pwd->pwd_jdir);
3843 uma_zfree_smr(pwd_zone, pwd);
3847 * The caller is responsible for invoking priv_check() and
3848 * mac_vnode_check_chroot() to authorize this operation.
3851 pwd_chroot(struct thread *td, struct vnode *vp)
3853 struct pwddesc *pdp;
3854 struct filedesc *fdp;
3855 struct pwd *newpwd, *oldpwd;
3858 fdp = td->td_proc->p_fd;
3859 pdp = td->td_proc->p_pd;
3860 newpwd = pwd_alloc();
3861 FILEDESC_SLOCK(fdp);
3863 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3864 if (chroot_allow_open_directories == 0 ||
3865 (chroot_allow_open_directories == 1 &&
3866 oldpwd->pwd_rdir != rootvnode)) {
3867 error = chroot_refuse_vdir_fds(fdp);
3868 FILEDESC_SUNLOCK(fdp);
3870 PWDDESC_XUNLOCK(pdp);
3875 FILEDESC_SUNLOCK(fdp);
3879 newpwd->pwd_rdir = vp;
3880 if (oldpwd->pwd_jdir == NULL) {
3882 newpwd->pwd_jdir = vp;
3884 pwd_fill(oldpwd, newpwd);
3885 pwd_set(pdp, newpwd);
3886 PWDDESC_XUNLOCK(pdp);
3892 pwd_chdir(struct thread *td, struct vnode *vp)
3894 struct pwddesc *pdp;
3895 struct pwd *newpwd, *oldpwd;
3897 VNPASS(vp->v_usecount > 0, vp);
3899 newpwd = pwd_alloc();
3900 pdp = td->td_proc->p_pd;
3902 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3903 newpwd->pwd_cdir = vp;
3904 pwd_fill(oldpwd, newpwd);
3905 pwd_set(pdp, newpwd);
3906 PWDDESC_XUNLOCK(pdp);
3911 * jail_attach(2) changes both root and working directories.
3914 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
3916 struct pwddesc *pdp;
3917 struct filedesc *fdp;
3918 struct pwd *newpwd, *oldpwd;
3921 fdp = td->td_proc->p_fd;
3922 pdp = td->td_proc->p_pd;
3923 newpwd = pwd_alloc();
3924 FILEDESC_SLOCK(fdp);
3926 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3927 error = chroot_refuse_vdir_fds(fdp);
3928 FILEDESC_SUNLOCK(fdp);
3930 PWDDESC_XUNLOCK(pdp);
3936 newpwd->pwd_rdir = vp;
3938 newpwd->pwd_cdir = vp;
3939 if (oldpwd->pwd_jdir == NULL) {
3941 newpwd->pwd_jdir = vp;
3943 pwd_fill(oldpwd, newpwd);
3944 pwd_set(pdp, newpwd);
3945 PWDDESC_XUNLOCK(pdp);
3951 pwd_ensure_dirs(void)
3953 struct pwddesc *pdp;
3954 struct pwd *oldpwd, *newpwd;
3956 pdp = curproc->p_pd;
3958 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3959 if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3960 PWDDESC_XUNLOCK(pdp);
3963 PWDDESC_XUNLOCK(pdp);
3965 newpwd = pwd_alloc();
3967 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3968 pwd_fill(oldpwd, newpwd);
3969 if (newpwd->pwd_cdir == NULL) {
3971 newpwd->pwd_cdir = rootvnode;
3973 if (newpwd->pwd_rdir == NULL) {
3975 newpwd->pwd_rdir = rootvnode;
3977 pwd_set(pdp, newpwd);
3978 PWDDESC_XUNLOCK(pdp);
3983 pwd_set_rootvnode(void)
3985 struct pwddesc *pdp;
3986 struct pwd *oldpwd, *newpwd;
3988 pdp = curproc->p_pd;
3990 newpwd = pwd_alloc();
3992 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3994 newpwd->pwd_cdir = rootvnode;
3996 newpwd->pwd_rdir = rootvnode;
3997 pwd_fill(oldpwd, newpwd);
3998 pwd_set(pdp, newpwd);
3999 PWDDESC_XUNLOCK(pdp);
4004 * Scan all active processes and prisons to see if any of them have a current
4005 * or root directory of `olddp'. If so, replace them with the new mount point.
4008 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
4010 struct pwddesc *pdp;
4011 struct pwd *newpwd, *oldpwd;
4016 if (vrefcnt(olddp) == 1)
4019 newpwd = pwd_alloc();
4020 sx_slock(&allproc_lock);
4021 FOREACH_PROC_IN_SYSTEM(p) {
4028 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4029 if (oldpwd == NULL ||
4030 (oldpwd->pwd_cdir != olddp &&
4031 oldpwd->pwd_rdir != olddp &&
4032 oldpwd->pwd_jdir != olddp)) {
4033 PWDDESC_XUNLOCK(pdp);
4037 if (oldpwd->pwd_cdir == olddp) {
4039 newpwd->pwd_cdir = newdp;
4041 if (oldpwd->pwd_rdir == olddp) {
4043 newpwd->pwd_rdir = newdp;
4045 if (oldpwd->pwd_jdir == olddp) {
4047 newpwd->pwd_jdir = newdp;
4049 pwd_fill(oldpwd, newpwd);
4050 pwd_set(pdp, newpwd);
4051 PWDDESC_XUNLOCK(pdp);
4054 newpwd = pwd_alloc();
4056 sx_sunlock(&allproc_lock);
4058 if (rootvnode == olddp) {
4063 mtx_lock(&prison0.pr_mtx);
4064 if (prison0.pr_root == olddp) {
4066 prison0.pr_root = newdp;
4069 mtx_unlock(&prison0.pr_mtx);
4070 sx_slock(&allprison_lock);
4071 TAILQ_FOREACH(pr, &allprison, pr_list) {
4072 mtx_lock(&pr->pr_mtx);
4073 if (pr->pr_root == olddp) {
4075 pr->pr_root = newdp;
4078 mtx_unlock(&pr->pr_mtx);
4080 sx_sunlock(&allprison_lock);
4085 struct filedesc_to_leader *
4086 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
4088 struct filedesc_to_leader *fdtol;
4090 fdtol = malloc(sizeof(struct filedesc_to_leader),
4091 M_FILEDESC_TO_LEADER, M_WAITOK);
4092 fdtol->fdl_refcount = 1;
4093 fdtol->fdl_holdcount = 0;
4094 fdtol->fdl_wakeup = 0;
4095 fdtol->fdl_leader = leader;
4097 FILEDESC_XLOCK(fdp);
4098 fdtol->fdl_next = old->fdl_next;
4099 fdtol->fdl_prev = old;
4100 old->fdl_next = fdtol;
4101 fdtol->fdl_next->fdl_prev = fdtol;
4102 FILEDESC_XUNLOCK(fdp);
4104 fdtol->fdl_next = fdtol;
4105 fdtol->fdl_prev = fdtol;
4111 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
4114 struct filedesc *fdp;
4116 int count, off, minoff;
4122 if (*(int *)arg1 != 0)
4125 fdp = curproc->p_fd;
4127 FILEDESC_SLOCK(fdp);
4129 off = NDSLOT(fdp->fd_nfiles - 1);
4130 for (minoff = NDSLOT(0); off >= minoff; --off)
4131 count += bitcountl(map[off]);
4132 FILEDESC_SUNLOCK(fdp);
4134 return (SYSCTL_OUT(req, &count, sizeof(count)));
4137 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
4138 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
4139 "Number of open file descriptors");
4142 * Get file structures globally.
4145 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
4148 struct filedesc *fdp;
4153 error = sysctl_wire_old_buffer(req, 0);
4156 if (req->oldptr == NULL) {
4158 sx_slock(&allproc_lock);
4159 FOREACH_PROC_IN_SYSTEM(p) {
4161 if (p->p_state == PRS_NEW) {
4169 /* overestimates sparse tables. */
4170 n += fdp->fd_nfiles;
4173 sx_sunlock(&allproc_lock);
4174 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4177 bzero(&xf, sizeof(xf));
4178 xf.xf_size = sizeof(xf);
4179 sx_slock(&allproc_lock);
4180 FOREACH_PROC_IN_SYSTEM(p) {
4182 if (p->p_state == PRS_NEW) {
4186 if (p_cansee(req->td, p) != 0) {
4190 xf.xf_pid = p->p_pid;
4191 xf.xf_uid = p->p_ucred->cr_uid;
4196 FILEDESC_SLOCK(fdp);
4197 FILEDESC_FOREACH_FP(fdp, n, fp) {
4198 if (refcount_load(&fdp->fd_refcnt) == 0)
4201 xf.xf_file = (uintptr_t)fp;
4202 xf.xf_data = (uintptr_t)fp->f_data;
4203 xf.xf_vnode = (uintptr_t)fp->f_vnode;
4204 xf.xf_type = (uintptr_t)fp->f_type;
4205 xf.xf_count = refcount_load(&fp->f_count);
4207 xf.xf_offset = foffset_get(fp);
4208 xf.xf_flag = fp->f_flag;
4209 error = SYSCTL_OUT(req, &xf, sizeof(xf));
4213 FILEDESC_SUNLOCK(fdp);
4218 sx_sunlock(&allproc_lock);
4222 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4223 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4225 #ifdef KINFO_FILE_SIZE
4226 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4230 xlate_fflags(int fflags)
4232 static const struct {
4235 } fflags_table[] = {
4236 { FAPPEND, KF_FLAG_APPEND },
4237 { FASYNC, KF_FLAG_ASYNC },
4238 { FFSYNC, KF_FLAG_FSYNC },
4239 { FHASLOCK, KF_FLAG_HASLOCK },
4240 { FNONBLOCK, KF_FLAG_NONBLOCK },
4241 { FREAD, KF_FLAG_READ },
4242 { FWRITE, KF_FLAG_WRITE },
4243 { O_CREAT, KF_FLAG_CREAT },
4244 { O_DIRECT, KF_FLAG_DIRECT },
4245 { O_EXCL, KF_FLAG_EXCL },
4246 { O_EXEC, KF_FLAG_EXEC },
4247 { O_EXLOCK, KF_FLAG_EXLOCK },
4248 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4249 { O_SHLOCK, KF_FLAG_SHLOCK },
4250 { O_TRUNC, KF_FLAG_TRUNC }
4256 for (i = 0; i < nitems(fflags_table); i++)
4257 if (fflags & fflags_table[i].fflag)
4258 kflags |= fflags_table[i].kf_fflag;
4262 /* Trim unused data from kf_path by truncating the structure size. */
4264 pack_kinfo(struct kinfo_file *kif)
4267 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4268 strlen(kif->kf_path) + 1;
4269 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4273 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4274 struct kinfo_file *kif, struct filedesc *fdp, int flags)
4278 bzero(kif, sizeof(*kif));
4280 /* Set a default type to allow for empty fill_kinfo() methods. */
4281 kif->kf_type = KF_TYPE_UNKNOWN;
4282 kif->kf_flags = xlate_fflags(fp->f_flag);
4283 if (rightsp != NULL)
4284 kif->kf_cap_rights = *rightsp;
4286 cap_rights_init_zero(&kif->kf_cap_rights);
4288 kif->kf_ref_count = refcount_load(&fp->f_count);
4289 kif->kf_offset = foffset_get(fp);
4292 * This may drop the filedesc lock, so the 'fp' cannot be
4293 * accessed after this call.
4295 error = fo_fill_kinfo(fp, kif, fdp);
4297 kif->kf_status |= KF_ATTR_VALID;
4298 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4301 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4305 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4306 struct kinfo_file *kif, int flags)
4310 bzero(kif, sizeof(*kif));
4312 kif->kf_type = KF_TYPE_VNODE;
4313 error = vn_fill_kinfo_vnode(vp, kif);
4315 kif->kf_status |= KF_ATTR_VALID;
4316 kif->kf_flags = xlate_fflags(fflags);
4317 cap_rights_init_zero(&kif->kf_cap_rights);
4319 kif->kf_ref_count = -1;
4320 kif->kf_offset = -1;
4321 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4324 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4328 struct export_fd_buf {
4329 struct filedesc *fdp;
4330 struct pwddesc *pdp;
4333 struct kinfo_file kif;
4338 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4340 struct kinfo_file *kif;
4343 if (efbuf->remainder != -1) {
4344 if (efbuf->remainder < kif->kf_structsize)
4346 efbuf->remainder -= kif->kf_structsize;
4348 if (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) != 0)
4349 return (sbuf_error(efbuf->sb));
4354 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4355 struct export_fd_buf *efbuf)
4359 if (efbuf->remainder == 0)
4361 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4363 FILEDESC_SUNLOCK(efbuf->fdp);
4364 error = export_kinfo_to_sb(efbuf);
4365 FILEDESC_SLOCK(efbuf->fdp);
4370 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4371 struct export_fd_buf *efbuf)
4375 if (efbuf->remainder == 0)
4377 if (efbuf->pdp != NULL)
4378 PWDDESC_XUNLOCK(efbuf->pdp);
4379 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4380 error = export_kinfo_to_sb(efbuf);
4381 if (efbuf->pdp != NULL)
4382 PWDDESC_XLOCK(efbuf->pdp);
4387 * Store a process file descriptor information to sbuf.
4389 * Takes a locked proc as argument, and returns with the proc unlocked.
4392 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
4396 struct filedesc *fdp;
4397 struct pwddesc *pdp;
4398 struct export_fd_buf *efbuf;
4399 struct vnode *cttyvp, *textvp, *tracevp;
4402 cap_rights_t rights;
4404 PROC_LOCK_ASSERT(p, MA_OWNED);
4407 tracevp = ktr_get_tracevp(p, true);
4409 textvp = p->p_textvp;
4412 /* Controlling tty. */
4414 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4415 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4423 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4427 efbuf->remainder = maxlen;
4428 efbuf->flags = flags;
4431 if (tracevp != NULL)
4432 error = export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE,
4433 FREAD | FWRITE, efbuf);
4434 if (error == 0 && textvp != NULL)
4435 error = export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD,
4437 if (error == 0 && cttyvp != NULL)
4438 error = export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY,
4439 FREAD | FWRITE, efbuf);
4440 if (error != 0 || pdp == NULL || fdp == NULL)
4445 pwd = pwd_hold_pwddesc(pdp);
4447 /* working directory */
4448 if (pwd->pwd_cdir != NULL) {
4449 vrefact(pwd->pwd_cdir);
4450 error = export_vnode_to_sb(pwd->pwd_cdir,
4451 KF_FD_TYPE_CWD, FREAD, efbuf);
4453 /* root directory */
4454 if (error == 0 && pwd->pwd_rdir != NULL) {
4455 vrefact(pwd->pwd_rdir);
4456 error = export_vnode_to_sb(pwd->pwd_rdir,
4457 KF_FD_TYPE_ROOT, FREAD, efbuf);
4459 /* jail directory */
4460 if (error == 0 && pwd->pwd_jdir != NULL) {
4461 vrefact(pwd->pwd_jdir);
4462 error = export_vnode_to_sb(pwd->pwd_jdir,
4463 KF_FD_TYPE_JAIL, FREAD, efbuf);
4466 PWDDESC_XUNLOCK(pdp);
4471 FILEDESC_SLOCK(fdp);
4472 FILEDESC_FOREACH_FP(fdp, i, fp) {
4473 if (refcount_load(&fdp->fd_refcnt) == 0)
4476 rights = *cap_rights(fdp, i);
4477 #else /* !CAPABILITIES */
4478 rights = cap_no_rights;
4481 * Create sysctl entry. It is OK to drop the filedesc
4482 * lock inside of export_file_to_sb() as we will
4483 * re-validate and re-evaluate its properties when the
4486 error = export_file_to_sb(fp, i, &rights, efbuf);
4490 FILEDESC_SUNLOCK(fdp);
4496 free(efbuf, M_TEMP);
4500 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
4503 * Get per-process file descriptors for use by procstat(1), et al.
4506 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4512 int error, error2, *name;
4520 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4521 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4522 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4527 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4528 error = kern_proc_filedesc_out(p, &sb, maxlen,
4529 KERN_FILEDESC_PACK_KINFO);
4530 error2 = sbuf_finish(&sb);
4532 return (error != 0 ? error : error2);
4535 #ifdef COMPAT_FREEBSD7
4536 #ifdef KINFO_OFILE_SIZE
4537 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4541 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4544 okif->kf_structsize = sizeof(*okif);
4545 okif->kf_type = kif->kf_type;
4546 okif->kf_fd = kif->kf_fd;
4547 okif->kf_ref_count = kif->kf_ref_count;
4548 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4549 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4550 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4551 okif->kf_offset = kif->kf_offset;
4552 if (kif->kf_type == KF_TYPE_VNODE)
4553 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4555 okif->kf_vnode_type = KF_VTYPE_VNON;
4556 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4557 if (kif->kf_type == KF_TYPE_SOCKET) {
4558 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4559 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4560 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4561 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4562 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4564 okif->kf_sa_local.ss_family = AF_UNSPEC;
4565 okif->kf_sa_peer.ss_family = AF_UNSPEC;
4570 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4571 struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4576 PWDDESC_XUNLOCK(pdp);
4577 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4578 kinfo_to_okinfo(kif, okif);
4579 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4585 * Get per-process file descriptors for use by procstat(1), et al.
4588 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4590 struct kinfo_ofile *okif;
4591 struct kinfo_file *kif;
4592 struct filedesc *fdp;
4593 struct pwddesc *pdp;
4596 int error, i, *name;
4605 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4612 if (fdp == NULL || pdp == NULL) {
4617 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4618 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4620 pwd = pwd_hold_pwddesc(pdp);
4622 if (pwd->pwd_cdir != NULL)
4623 export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4625 if (pwd->pwd_rdir != NULL)
4626 export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4628 if (pwd->pwd_jdir != NULL)
4629 export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4632 PWDDESC_XUNLOCK(pdp);
4635 FILEDESC_SLOCK(fdp);
4636 FILEDESC_FOREACH_FP(fdp, i, fp) {
4637 if (refcount_load(&fdp->fd_refcnt) == 0)
4639 export_file_to_kinfo(fp, i, NULL, kif, fdp,
4640 KERN_FILEDESC_PACK_KINFO);
4641 FILEDESC_SUNLOCK(fdp);
4642 kinfo_to_okinfo(kif, okif);
4643 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4644 FILEDESC_SLOCK(fdp);
4648 FILEDESC_SUNLOCK(fdp);
4656 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4657 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4658 "Process ofiledesc entries");
4659 #endif /* COMPAT_FREEBSD7 */
4662 vntype_to_kinfo(int vtype)
4667 } vtypes_table[] = {
4668 { VBAD, KF_VTYPE_VBAD },
4669 { VBLK, KF_VTYPE_VBLK },
4670 { VCHR, KF_VTYPE_VCHR },
4671 { VDIR, KF_VTYPE_VDIR },
4672 { VFIFO, KF_VTYPE_VFIFO },
4673 { VLNK, KF_VTYPE_VLNK },
4674 { VNON, KF_VTYPE_VNON },
4675 { VREG, KF_VTYPE_VREG },
4676 { VSOCK, KF_VTYPE_VSOCK }
4681 * Perform vtype translation.
4683 for (i = 0; i < nitems(vtypes_table); i++)
4684 if (vtypes_table[i].vtype == vtype)
4685 return (vtypes_table[i].kf_vtype);
4687 return (KF_VTYPE_UNKNOWN);
4690 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4691 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4692 "Process filedesc entries");
4695 * Store a process current working directory information to sbuf.
4697 * Takes a locked proc as argument, and returns with the proc unlocked.
4700 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
4702 struct pwddesc *pdp;
4704 struct export_fd_buf *efbuf;
4708 PROC_LOCK_ASSERT(p, MA_OWNED);
4715 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4719 efbuf->remainder = maxlen;
4723 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4724 cdir = pwd->pwd_cdir;
4729 error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4731 PWDDESC_XUNLOCK(pdp);
4733 free(efbuf, M_TEMP);
4738 * Get per-process current working directory.
4741 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4747 int error, error2, *name;
4755 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4756 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4757 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4762 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4763 error = kern_proc_cwd_out(p, &sb, maxlen);
4764 error2 = sbuf_finish(&sb);
4766 return (error != 0 ? error : error2);
4769 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4770 sysctl_kern_proc_cwd, "Process current working directory");
4774 * For the purposes of debugging, generate a human-readable string for the
4778 file_type_to_name(short type)
4806 case DTYPE_PROCDESC:
4810 case DTYPE_LINUXTFD:
4818 * For the purposes of debugging, identify a process (if any, perhaps one of
4819 * many) that references the passed file in its file descriptor array. Return
4822 static struct proc *
4823 file_to_first_proc(struct file *fp)
4825 struct filedesc *fdp;
4829 FOREACH_PROC_IN_SYSTEM(p) {
4830 if (p->p_state == PRS_NEW)
4835 for (n = 0; n < fdp->fd_nfiles; n++) {
4836 if (fp == fdp->fd_ofiles[n].fde_file)
4844 db_print_file(struct file *fp, int header)
4846 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4850 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4851 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4852 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4854 p = file_to_first_proc(fp);
4855 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4856 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4857 fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4858 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4863 DB_SHOW_COMMAND(file, db_show_file)
4868 db_printf("usage: show file <addr>\n");
4871 fp = (struct file *)addr;
4872 db_print_file(fp, 1);
4875 DB_SHOW_COMMAND(files, db_show_files)
4877 struct filedesc *fdp;
4884 FOREACH_PROC_IN_SYSTEM(p) {
4885 if (p->p_state == PRS_NEW)
4887 if ((fdp = p->p_fd) == NULL)
4889 for (n = 0; n < fdp->fd_nfiles; ++n) {
4890 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4892 db_print_file(fp, header);
4899 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4900 &maxfilesperproc, 0, "Maximum files allowed open per process");
4902 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4903 &maxfiles, 0, "Maximum number of files");
4905 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4906 &openfiles, 0, "System-wide number of open files");
4910 filelistinit(void *dummy)
4913 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4914 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4915 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4916 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4917 pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4918 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4920 * XXXMJG this is a temporary hack due to boot ordering issues against
4923 vfs_smr = uma_zone_get_smr(pwd_zone);
4924 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4926 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4928 /*-------------------------------------------------------------------*/
4931 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4932 int flags, struct thread *td)
4939 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4947 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4955 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4963 badfo_kqfilter(struct file *fp, struct knote *kn)
4970 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
4977 badfo_close(struct file *fp, struct thread *td)
4984 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4992 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5000 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5001 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5009 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
5015 struct fileops badfileops = {
5016 .fo_read = badfo_readwrite,
5017 .fo_write = badfo_readwrite,
5018 .fo_truncate = badfo_truncate,
5019 .fo_ioctl = badfo_ioctl,
5020 .fo_poll = badfo_poll,
5021 .fo_kqfilter = badfo_kqfilter,
5022 .fo_stat = badfo_stat,
5023 .fo_close = badfo_close,
5024 .fo_chmod = badfo_chmod,
5025 .fo_chown = badfo_chown,
5026 .fo_sendfile = badfo_sendfile,
5027 .fo_fill_kinfo = badfo_fill_kinfo,
5031 path_poll(struct file *fp, int events, struct ucred *active_cred,
5038 path_close(struct file *fp, struct thread *td)
5040 MPASS(fp->f_type == DTYPE_VNODE);
5041 fp->f_ops = &badfileops;
5046 struct fileops path_fileops = {
5047 .fo_read = badfo_readwrite,
5048 .fo_write = badfo_readwrite,
5049 .fo_truncate = badfo_truncate,
5050 .fo_ioctl = badfo_ioctl,
5051 .fo_poll = path_poll,
5052 .fo_kqfilter = vn_kqfilter_opath,
5053 .fo_stat = vn_statfile,
5054 .fo_close = path_close,
5055 .fo_chmod = badfo_chmod,
5056 .fo_chown = badfo_chown,
5057 .fo_sendfile = badfo_sendfile,
5058 .fo_fill_kinfo = vn_fill_kinfo,
5059 .fo_flags = DFLAG_PASSABLE,
5063 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
5064 int flags, struct thread *td)
5067 return (EOPNOTSUPP);
5071 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
5079 invfo_ioctl(struct file *fp, u_long com, void *data,
5080 struct ucred *active_cred, struct thread *td)
5087 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
5091 return (poll_no_poll(events));
5095 invfo_kqfilter(struct file *fp, struct knote *kn)
5102 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5110 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5118 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5119 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5126 /*-------------------------------------------------------------------*/
5129 * File Descriptor pseudo-device driver (/dev/fd/).
5131 * Opening minor device N dup()s the file (if any) connected to file
5132 * descriptor N belonging to the calling process. Note that this driver
5133 * consists of only the ``open()'' routine, because all subsequent
5134 * references to this file will be direct to the other driver.
5136 * XXX: we could give this one a cloning event handler if necessary.
5141 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
5145 * XXX Kludge: set curthread->td_dupfd to contain the value of the
5146 * the file descriptor being sought for duplication. The error
5147 * return ensures that the vnode for this device will be released
5148 * by vn_open. Open will detect this special error and take the
5149 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
5150 * will simply report the error.
5152 td->td_dupfd = dev2unit(dev);
5156 static struct cdevsw fildesc_cdevsw = {
5157 .d_version = D_VERSION,
5163 fildesc_drvinit(void *unused)
5167 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
5168 UID_ROOT, GID_WHEEL, 0666, "fd/0");
5169 make_dev_alias(dev, "stdin");
5170 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
5171 UID_ROOT, GID_WHEEL, 0666, "fd/1");
5172 make_dev_alias(dev, "stdout");
5173 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
5174 UID_ROOT, GID_WHEEL, 0666, "fd/2");
5175 make_dev_alias(dev, "stderr");
5178 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);