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 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1399 struct filedesc *fdp;
1400 const struct fdescenttbl *fdt;
1405 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1406 * open should not be a usage error. From a close_range() perspective,
1407 * close_range(3, ~0U, 0) in the same scenario should also likely not
1408 * be a usage error as all fd above 3 are in-fact already closed.
1410 if (highfd < lowfd) {
1414 fdp = td->td_proc->p_fd;
1415 FILEDESC_XLOCK(fdp);
1416 fdt = atomic_load_ptr(&fdp->fd_files);
1417 highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1419 if (__predict_false(fd > highfd)) {
1423 fp = fdt->fdt_ofiles[fd].fde_file;
1429 (void) closefp(fdp, fd, fp, td, true, true);
1432 FILEDESC_XLOCK(fdp);
1433 fdt = atomic_load_ptr(&fdp->fd_files);
1438 FILEDESC_XUNLOCK(fdp);
1443 #ifndef _SYS_SYSPROTO_H_
1444 struct close_range_args {
1451 sys_close_range(struct thread *td, struct close_range_args *uap)
1454 AUDIT_ARG_FD(uap->lowfd);
1455 AUDIT_ARG_CMD(uap->highfd);
1456 AUDIT_ARG_FFLAGS(uap->flags);
1458 /* No flags currently defined */
1459 if (uap->flags != 0)
1461 return (kern_close_range(td, uap->lowfd, uap->highfd));
1464 #ifdef COMPAT_FREEBSD12
1466 * Close open file descriptors.
1468 #ifndef _SYS_SYSPROTO_H_
1469 struct freebsd12_closefrom_args {
1475 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1479 AUDIT_ARG_FD(uap->lowfd);
1482 * Treat negative starting file descriptor values identical to
1483 * closefrom(0) which closes all files.
1485 lowfd = MAX(0, uap->lowfd);
1486 return (kern_close_range(td, lowfd, ~0U));
1488 #endif /* COMPAT_FREEBSD12 */
1490 #if defined(COMPAT_43)
1492 * Return status information about a file descriptor.
1494 #ifndef _SYS_SYSPROTO_H_
1495 struct ofstat_args {
1502 ofstat(struct thread *td, struct ofstat_args *uap)
1508 error = kern_fstat(td, uap->fd, &ub);
1511 error = copyout(&oub, uap->sb, sizeof(oub));
1515 #endif /* COMPAT_43 */
1517 #if defined(COMPAT_FREEBSD11)
1519 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1522 struct freebsd11_stat osb;
1525 error = kern_fstat(td, uap->fd, &sb);
1528 error = freebsd11_cvtstat(&sb, &osb);
1530 error = copyout(&osb, uap->sb, sizeof(osb));
1533 #endif /* COMPAT_FREEBSD11 */
1536 * Return status information about a file descriptor.
1538 #ifndef _SYS_SYSPROTO_H_
1546 sys_fstat(struct thread *td, struct fstat_args *uap)
1551 error = kern_fstat(td, uap->fd, &ub);
1553 error = copyout(&ub, uap->sb, sizeof(ub));
1558 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1565 error = fget(td, fd, &cap_fstat_rights, &fp);
1566 if (__predict_false(error != 0))
1569 AUDIT_ARG_FILE(td->td_proc, fp);
1571 error = fo_stat(fp, sbp, td->td_ucred);
1573 #ifdef __STAT_TIME_T_EXT
1574 sbp->st_atim_ext = 0;
1575 sbp->st_mtim_ext = 0;
1576 sbp->st_ctim_ext = 0;
1577 sbp->st_btim_ext = 0;
1580 if (KTRPOINT(td, KTR_STRUCT))
1581 ktrstat_error(sbp, error);
1586 #if defined(COMPAT_FREEBSD11)
1588 * Return status information about a file descriptor.
1590 #ifndef _SYS_SYSPROTO_H_
1591 struct freebsd11_nfstat_args {
1598 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1604 error = kern_fstat(td, uap->fd, &ub);
1607 error = freebsd11_cvtnstat(&ub, &nub);
1609 error = copyout(&nub, uap->sb, sizeof(nub));
1612 #endif /* COMPAT_FREEBSD11 */
1615 * Return pathconf information about a file descriptor.
1617 #ifndef _SYS_SYSPROTO_H_
1618 struct fpathconf_args {
1625 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1630 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1632 td->td_retval[0] = value;
1637 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1643 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1647 if (name == _PC_ASYNC_IO) {
1648 *valuep = _POSIX_ASYNCHRONOUS_IO;
1653 vn_lock(vp, LK_SHARED | LK_RETRY);
1654 error = VOP_PATHCONF(vp, name, valuep);
1656 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1657 if (name != _PC_PIPE_BUF) {
1672 * Copy filecaps structure allocating memory for ioctls array if needed.
1674 * The last parameter indicates whether the fdtable is locked. If it is not and
1675 * ioctls are encountered, copying fails and the caller must lock the table.
1677 * Note that if the table was not locked, the caller has to check the relevant
1678 * sequence counter to determine whether the operation was successful.
1681 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1685 if (src->fc_ioctls != NULL && !locked)
1687 memcpy(dst, src, sizeof(*src));
1688 if (src->fc_ioctls == NULL)
1691 KASSERT(src->fc_nioctls > 0,
1692 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1694 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1695 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1696 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1701 filecaps_copy_prep(const struct filecaps *src)
1706 if (__predict_true(src->fc_ioctls == NULL))
1709 KASSERT(src->fc_nioctls > 0,
1710 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1712 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1713 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1718 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1724 if (__predict_true(src->fc_ioctls == NULL)) {
1725 MPASS(ioctls == NULL);
1729 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1730 dst->fc_ioctls = ioctls;
1731 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1735 * Move filecaps structure to the new place and clear the old place.
1738 filecaps_move(struct filecaps *src, struct filecaps *dst)
1742 bzero(src, sizeof(*src));
1746 * Fill the given filecaps structure with full rights.
1749 filecaps_fill(struct filecaps *fcaps)
1752 CAP_ALL(&fcaps->fc_rights);
1753 fcaps->fc_ioctls = NULL;
1754 fcaps->fc_nioctls = -1;
1755 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1759 * Free memory allocated within filecaps structure.
1762 filecaps_free(struct filecaps *fcaps)
1765 free(fcaps->fc_ioctls, M_FILECAPS);
1766 bzero(fcaps, sizeof(*fcaps));
1770 filecaps_free_prep(struct filecaps *fcaps)
1774 ioctls = fcaps->fc_ioctls;
1775 bzero(fcaps, sizeof(*fcaps));
1780 filecaps_free_finish(u_long *ioctls)
1783 free(ioctls, M_FILECAPS);
1787 * Validate the given filecaps structure.
1790 filecaps_validate(const struct filecaps *fcaps, const char *func)
1793 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1794 ("%s: invalid rights", func));
1795 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1796 ("%s: invalid fcntls", func));
1797 KASSERT(fcaps->fc_fcntls == 0 ||
1798 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1799 ("%s: fcntls without CAP_FCNTL", func));
1800 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1801 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1802 ("%s: invalid ioctls", func));
1803 KASSERT(fcaps->fc_nioctls == 0 ||
1804 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1805 ("%s: ioctls without CAP_IOCTL", func));
1809 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1813 FILEDESC_XLOCK_ASSERT(fdp);
1815 nfd1 = fdp->fd_nfiles * 2;
1818 fdgrowtable(fdp, nfd1);
1822 * Grow the file table to accommodate (at least) nfd descriptors.
1825 fdgrowtable(struct filedesc *fdp, int nfd)
1827 struct filedesc0 *fdp0;
1828 struct freetable *ft;
1829 struct fdescenttbl *ntable;
1830 struct fdescenttbl *otable;
1831 int nnfiles, onfiles;
1832 NDSLOTTYPE *nmap, *omap;
1834 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1836 /* save old values */
1837 onfiles = fdp->fd_nfiles;
1838 otable = fdp->fd_files;
1841 /* compute the size of the new table */
1842 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1843 if (nnfiles <= onfiles)
1844 /* the table is already large enough */
1848 * Allocate a new table. We need enough space for the number of
1849 * entries, file entries themselves and the struct freetable we will use
1850 * when we decommission the table and place it on the freelist.
1851 * We place the struct freetable in the middle so we don't have
1852 * to worry about padding.
1854 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1855 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1856 sizeof(struct freetable),
1857 M_FILEDESC, M_ZERO | M_WAITOK);
1858 /* copy the old data */
1859 ntable->fdt_nfiles = nnfiles;
1860 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1861 onfiles * sizeof(ntable->fdt_ofiles[0]));
1864 * Allocate a new map only if the old is not large enough. It will
1865 * grow at a slower rate than the table as it can map more
1866 * entries than the table can hold.
1868 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1869 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1871 /* copy over the old data and update the pointer */
1872 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1877 * Make sure that ntable is correctly initialized before we replace
1878 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1881 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1884 * Free the old file table when not shared by other threads or processes.
1885 * The old file table is considered to be shared when either are true:
1886 * - The process has more than one thread.
1887 * - The file descriptor table has been shared via fdshare().
1889 * When shared, the old file table will be placed on a freelist
1890 * which will be processed when the struct filedesc is released.
1892 * Note that if onfiles == NDFILE, we're dealing with the original
1893 * static allocation contained within (struct filedesc0 *)fdp,
1894 * which must not be freed.
1896 if (onfiles > NDFILE) {
1898 * Note we may be called here from fdinit while allocating a
1899 * table for a new process in which case ->p_fd points
1902 if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) {
1903 free(otable, M_FILEDESC);
1905 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1906 fdp0 = (struct filedesc0 *)fdp;
1907 ft->ft_table = otable;
1908 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1912 * The map does not have the same possibility of threads still
1913 * holding references to it. So always free it as long as it
1914 * does not reference the original static allocation.
1916 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1917 free(omap, M_FILEDESC);
1921 * Allocate a file descriptor for the process.
1924 fdalloc(struct thread *td, int minfd, int *result)
1926 struct proc *p = td->td_proc;
1927 struct filedesc *fdp = p->p_fd;
1928 int fd, maxfd, allocfd;
1933 FILEDESC_XLOCK_ASSERT(fdp);
1935 if (fdp->fd_freefile > minfd)
1936 minfd = fdp->fd_freefile;
1938 maxfd = getmaxfd(td);
1941 * Search the bitmap for a free descriptor starting at minfd.
1942 * If none is found, grow the file table.
1944 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1945 if (__predict_false(fd >= maxfd))
1947 if (__predict_false(fd >= fdp->fd_nfiles)) {
1948 allocfd = min(fd * 2, maxfd);
1950 if (RACCT_ENABLED()) {
1951 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1957 * fd is already equal to first free descriptor >= minfd, so
1958 * we only need to grow the table and we are done.
1960 fdgrowtable_exp(fdp, allocfd);
1964 * Perform some sanity checks, then mark the file descriptor as
1965 * used and return it to the caller.
1967 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1968 ("invalid descriptor %d", fd));
1969 KASSERT(!fdisused(fdp, fd),
1970 ("fd_first_free() returned non-free descriptor"));
1971 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1972 ("file descriptor isn't free"));
1979 * Allocate n file descriptors for the process.
1982 fdallocn(struct thread *td, int minfd, int *fds, int n)
1984 struct proc *p = td->td_proc;
1985 struct filedesc *fdp = p->p_fd;
1988 FILEDESC_XLOCK_ASSERT(fdp);
1990 for (i = 0; i < n; i++)
1991 if (fdalloc(td, 0, &fds[i]) != 0)
1995 for (i--; i >= 0; i--)
1996 fdunused(fdp, fds[i]);
2004 * Create a new open file structure and allocate a file descriptor for the
2005 * process that refers to it. We add one reference to the file for the
2006 * descriptor table and one reference for resultfp. This is to prevent us
2007 * being preempted and the entry in the descriptor table closed after we
2008 * release the FILEDESC lock.
2011 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
2012 struct filecaps *fcaps)
2017 MPASS(resultfp != NULL);
2018 MPASS(resultfd != NULL);
2020 error = _falloc_noinstall(td, &fp, 2);
2021 if (__predict_false(error != 0)) {
2025 error = finstall_refed(td, fp, &fd, flags, fcaps);
2026 if (__predict_false(error != 0)) {
2027 falloc_abort(td, fp);
2038 * Create a new open file structure without allocating a file descriptor.
2041 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2044 int maxuserfiles = maxfiles - (maxfiles / 20);
2046 static struct timeval lastfail;
2049 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2052 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2053 if ((openfiles_new >= maxuserfiles &&
2054 priv_check(td, PRIV_MAXFILES) != 0) ||
2055 openfiles_new >= maxfiles) {
2056 atomic_subtract_int(&openfiles, 1);
2057 if (ppsratecheck(&lastfail, &curfail, 1)) {
2058 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2059 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2063 fp = uma_zalloc(file_zone, M_WAITOK);
2064 bzero(fp, sizeof(*fp));
2065 refcount_init(&fp->f_count, n);
2066 fp->f_cred = crhold(td->td_ucred);
2067 fp->f_ops = &badfileops;
2073 falloc_abort(struct thread *td, struct file *fp)
2077 * For assertion purposes.
2079 refcount_init(&fp->f_count, 0);
2084 * Install a file in a file descriptor table.
2087 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2088 struct filecaps *fcaps)
2090 struct filedescent *fde;
2094 filecaps_validate(fcaps, __func__);
2095 FILEDESC_XLOCK_ASSERT(fdp);
2097 fde = &fdp->fd_ofiles[fd];
2099 seqc_write_begin(&fde->fde_seqc);
2102 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2104 filecaps_move(fcaps, &fde->fde_caps);
2106 filecaps_fill(&fde->fde_caps);
2108 seqc_write_end(&fde->fde_seqc);
2113 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2114 struct filecaps *fcaps)
2116 struct filedesc *fdp = td->td_proc->p_fd;
2121 FILEDESC_XLOCK(fdp);
2122 error = fdalloc(td, 0, fd);
2123 if (__predict_true(error == 0)) {
2124 _finstall(fdp, fp, *fd, flags, fcaps);
2126 FILEDESC_XUNLOCK(fdp);
2131 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2132 struct filecaps *fcaps)
2140 error = finstall_refed(td, fp, fd, flags, fcaps);
2141 if (__predict_false(error != 0)) {
2148 * Build a new filedesc structure from another.
2150 * If fdp is not NULL, return with it shared locked.
2155 struct filedesc0 *newfdp0;
2156 struct filedesc *newfdp;
2158 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2159 newfdp = &newfdp0->fd_fd;
2161 /* Create the file descriptor table. */
2162 FILEDESC_LOCK_INIT(newfdp);
2163 refcount_init(&newfdp->fd_refcnt, 1);
2164 refcount_init(&newfdp->fd_holdcnt, 1);
2165 newfdp->fd_map = newfdp0->fd_dmap;
2166 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2167 newfdp->fd_files->fdt_nfiles = NDFILE;
2173 * Build a pwddesc structure from another.
2174 * Copy the current, root, and jail root vnode references.
2176 * If pdp is not NULL, return with it shared locked.
2179 pdinit(struct pwddesc *pdp, bool keeplock)
2181 struct pwddesc *newpdp;
2184 newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2186 PWDDESC_LOCK_INIT(newpdp);
2187 refcount_init(&newpdp->pd_refcount, 1);
2188 newpdp->pd_cmask = CMASK;
2191 newpwd = pwd_alloc();
2192 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2197 newpwd = pwd_hold_pwddesc(pdp);
2198 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2200 PWDDESC_XUNLOCK(pdp);
2205 * Hold either filedesc or pwddesc of the passed process.
2207 * The process lock is used to synchronize against the target exiting and
2210 * Clearing can be ilustrated in 3 steps:
2211 * 1. set the pointer to NULL. Either routine can race against it, hence
2213 * 2. observe the process lock as not taken. Until then fdhold/pdhold can
2214 * race to either still see the pointer or find NULL. It is still safe to
2215 * grab a reference as clearing is stalled.
2216 * 3. after the lock is observed as not taken, any fdhold/pdhold calls are
2217 * guaranteed to see NULL, making it safe to finish clearing
2219 static struct filedesc *
2220 fdhold(struct proc *p)
2222 struct filedesc *fdp;
2224 PROC_LOCK_ASSERT(p, MA_OWNED);
2225 fdp = atomic_load_ptr(&p->p_fd);
2227 refcount_acquire(&fdp->fd_holdcnt);
2231 static struct pwddesc *
2232 pdhold(struct proc *p)
2234 struct pwddesc *pdp;
2236 PROC_LOCK_ASSERT(p, MA_OWNED);
2237 pdp = atomic_load_ptr(&p->p_pd);
2239 refcount_acquire(&pdp->pd_refcount);
2244 fddrop(struct filedesc *fdp)
2247 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2248 if (refcount_release(&fdp->fd_holdcnt) == 0)
2252 FILEDESC_LOCK_DESTROY(fdp);
2253 uma_zfree(filedesc0_zone, fdp);
2257 pddrop(struct pwddesc *pdp)
2261 if (refcount_release_if_not_last(&pdp->pd_refcount))
2265 if (refcount_release(&pdp->pd_refcount) == 0) {
2266 PWDDESC_XUNLOCK(pdp);
2269 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2271 PWDDESC_XUNLOCK(pdp);
2274 PWDDESC_LOCK_DESTROY(pdp);
2275 free(pdp, M_PWDDESC);
2279 * Share a filedesc structure.
2282 fdshare(struct filedesc *fdp)
2285 refcount_acquire(&fdp->fd_refcnt);
2290 * Share a pwddesc structure.
2293 pdshare(struct pwddesc *pdp)
2295 refcount_acquire(&pdp->pd_refcount);
2300 * Unshare a filedesc structure, if necessary by making a copy
2303 fdunshare(struct thread *td)
2305 struct filedesc *tmp;
2306 struct proc *p = td->td_proc;
2308 if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2311 tmp = fdcopy(p->p_fd);
2317 * Unshare a pwddesc structure.
2320 pdunshare(struct thread *td)
2322 struct pwddesc *pdp;
2327 if (refcount_load(&p->p_pd->pd_refcount) == 1)
2330 pdp = pdcopy(p->p_pd);
2336 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2337 * this is to ease callers, not catch errors.
2340 fdcopy(struct filedesc *fdp)
2342 struct filedesc *newfdp;
2343 struct filedescent *nfde, *ofde;
2349 FILEDESC_SLOCK(fdp);
2351 lastfile = fdlastfile(fdp);
2352 if (lastfile < newfdp->fd_nfiles)
2354 FILEDESC_SUNLOCK(fdp);
2355 fdgrowtable(newfdp, lastfile + 1);
2356 FILEDESC_SLOCK(fdp);
2358 /* copy all passable descriptors (i.e. not kqueue) */
2359 newfdp->fd_freefile = fdp->fd_freefile;
2360 FILEDESC_FOREACH_FDE(fdp, i, ofde) {
2361 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2362 !fhold(ofde->fde_file)) {
2363 if (newfdp->fd_freefile == fdp->fd_freefile)
2364 newfdp->fd_freefile = i;
2367 nfde = &newfdp->fd_ofiles[i];
2369 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2370 fdused_init(newfdp, i);
2372 MPASS(newfdp->fd_freefile != -1);
2373 FILEDESC_SUNLOCK(fdp);
2378 * Copy a pwddesc structure.
2381 pdcopy(struct pwddesc *pdp)
2383 struct pwddesc *newpdp;
2387 newpdp = pdinit(pdp, true);
2388 newpdp->pd_cmask = pdp->pd_cmask;
2389 PWDDESC_XUNLOCK(pdp);
2394 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2395 * one of processes using it exits) and the table used to be shared.
2398 fdclearlocks(struct thread *td)
2400 struct filedesc *fdp;
2401 struct filedesc_to_leader *fdtol;
2411 MPASS(fdtol != NULL);
2413 FILEDESC_XLOCK(fdp);
2414 KASSERT(fdtol->fdl_refcount > 0,
2415 ("filedesc_to_refcount botch: fdl_refcount=%d",
2416 fdtol->fdl_refcount));
2417 if (fdtol->fdl_refcount == 1 &&
2418 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2419 FILEDESC_FOREACH_FP(fdp, i, fp) {
2420 if (fp->f_type != DTYPE_VNODE ||
2423 FILEDESC_XUNLOCK(fdp);
2424 lf.l_whence = SEEK_SET;
2427 lf.l_type = F_UNLCK;
2429 (void) VOP_ADVLOCK(vp,
2430 (caddr_t)p->p_leader, F_UNLCK,
2432 FILEDESC_XLOCK(fdp);
2437 if (fdtol->fdl_refcount == 1) {
2438 if (fdp->fd_holdleaderscount > 0 &&
2439 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2441 * close() or kern_dup() has cleared a reference
2442 * in a shared file descriptor table.
2444 fdp->fd_holdleaderswakeup = 1;
2445 sx_sleep(&fdp->fd_holdleaderscount,
2446 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2449 if (fdtol->fdl_holdcount > 0) {
2451 * Ensure that fdtol->fdl_leader remains
2452 * valid in closef().
2454 fdtol->fdl_wakeup = 1;
2455 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2460 fdtol->fdl_refcount--;
2461 if (fdtol->fdl_refcount == 0 &&
2462 fdtol->fdl_holdcount == 0) {
2463 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2464 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2468 FILEDESC_XUNLOCK(fdp);
2470 free(fdtol, M_FILEDESC_TO_LEADER);
2474 * Release a filedesc structure.
2477 fdescfree_fds(struct thread *td, struct filedesc *fdp)
2479 struct filedesc0 *fdp0;
2480 struct freetable *ft, *tft;
2481 struct filedescent *fde;
2485 KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2486 ("%s: fd table %p carries references", __func__, fdp));
2489 * Serialize with threads iterating over the table, if any.
2491 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2492 FILEDESC_XLOCK(fdp);
2493 FILEDESC_XUNLOCK(fdp);
2496 FILEDESC_FOREACH_FDE(fdp, i, fde) {
2499 (void) closef(fp, td);
2502 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2503 free(fdp->fd_map, M_FILEDESC);
2504 if (fdp->fd_nfiles > NDFILE)
2505 free(fdp->fd_files, M_FILEDESC);
2507 fdp0 = (struct filedesc0 *)fdp;
2508 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2509 free(ft->ft_table, M_FILEDESC);
2515 fdescfree(struct thread *td)
2518 struct filedesc *fdp;
2525 if (RACCT_ENABLED())
2526 racct_set_unlocked(p, RACCT_NOFILE, 0);
2529 if (p->p_fdtol != NULL)
2533 * Check fdhold for an explanation.
2535 atomic_store_ptr(&p->p_fd, NULL);
2536 atomic_thread_fence_seq_cst();
2537 PROC_WAIT_UNLOCKED(p);
2539 if (refcount_release(&fdp->fd_refcnt) == 0)
2542 fdescfree_fds(td, fdp);
2546 pdescfree(struct thread *td)
2549 struct pwddesc *pdp;
2556 * Check pdhold for an explanation.
2558 atomic_store_ptr(&p->p_pd, NULL);
2559 atomic_thread_fence_seq_cst();
2560 PROC_WAIT_UNLOCKED(p);
2566 * For setugid programs, we don't want to people to use that setugidness
2567 * to generate error messages which write to a file which otherwise would
2568 * otherwise be off-limits to the process. We check for filesystems where
2569 * the vnode can change out from under us after execve (like [lin]procfs).
2571 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2572 * sufficient. We also don't check for setugidness since we know we are.
2575 is_unsafe(struct file *fp)
2579 if (fp->f_type != DTYPE_VNODE)
2583 return ((vp->v_vflag & VV_PROCDEP) != 0);
2587 * Make this setguid thing safe, if at all possible.
2590 fdsetugidsafety(struct thread *td)
2592 struct filedesc *fdp;
2596 fdp = td->td_proc->p_fd;
2597 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2598 ("the fdtable should not be shared"));
2599 MPASS(fdp->fd_nfiles >= 3);
2600 for (i = 0; i <= 2; i++) {
2601 fp = fdp->fd_ofiles[i].fde_file;
2602 if (fp != NULL && is_unsafe(fp)) {
2603 FILEDESC_XLOCK(fdp);
2604 knote_fdclose(td, i);
2606 * NULL-out descriptor prior to close to avoid
2607 * a race while close blocks.
2610 FILEDESC_XUNLOCK(fdp);
2611 (void) closef(fp, td);
2617 * If a specific file object occupies a specific file descriptor, close the
2618 * file descriptor entry and drop a reference on the file object. This is a
2619 * convenience function to handle a subsequent error in a function that calls
2620 * falloc() that handles the race that another thread might have closed the
2621 * file descriptor out from under the thread creating the file object.
2624 fdclose(struct thread *td, struct file *fp, int idx)
2626 struct filedesc *fdp = td->td_proc->p_fd;
2628 FILEDESC_XLOCK(fdp);
2629 if (fdp->fd_ofiles[idx].fde_file == fp) {
2631 FILEDESC_XUNLOCK(fdp);
2634 FILEDESC_XUNLOCK(fdp);
2638 * Close any files on exec?
2641 fdcloseexec(struct thread *td)
2643 struct filedesc *fdp;
2644 struct filedescent *fde;
2648 fdp = td->td_proc->p_fd;
2649 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2650 ("the fdtable should not be shared"));
2651 FILEDESC_FOREACH_FDE(fdp, i, fde) {
2653 if (fp->f_type == DTYPE_MQUEUE ||
2654 (fde->fde_flags & UF_EXCLOSE)) {
2655 FILEDESC_XLOCK(fdp);
2657 (void) closefp(fdp, i, fp, td, false, false);
2658 FILEDESC_UNLOCK_ASSERT(fdp);
2664 * It is unsafe for set[ug]id processes to be started with file
2665 * descriptors 0..2 closed, as these descriptors are given implicit
2666 * significance in the Standard C library. fdcheckstd() will create a
2667 * descriptor referencing /dev/null for each of stdin, stdout, and
2668 * stderr that is not already open.
2671 fdcheckstd(struct thread *td)
2673 struct filedesc *fdp;
2675 int i, error, devnull;
2677 fdp = td->td_proc->p_fd;
2678 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2679 ("the fdtable should not be shared"));
2680 MPASS(fdp->fd_nfiles >= 3);
2682 for (i = 0; i <= 2; i++) {
2683 if (fdp->fd_ofiles[i].fde_file != NULL)
2686 save = td->td_retval[0];
2687 if (devnull != -1) {
2688 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2690 error = kern_openat(td, AT_FDCWD, "/dev/null",
2691 UIO_SYSSPACE, O_RDWR, 0);
2693 devnull = td->td_retval[0];
2694 KASSERT(devnull == i, ("we didn't get our fd"));
2697 td->td_retval[0] = save;
2705 * Internal form of close. Decrement reference count on file structure.
2706 * Note: td may be NULL when closing a file that was being passed in a
2710 closef(struct file *fp, struct thread *td)
2714 struct filedesc_to_leader *fdtol;
2715 struct filedesc *fdp;
2720 * POSIX record locking dictates that any close releases ALL
2721 * locks owned by this process. This is handled by setting
2722 * a flag in the unlock to free ONLY locks obeying POSIX
2723 * semantics, and not to free BSD-style file locks.
2724 * If the descriptor was in a message, POSIX-style locks
2725 * aren't passed with the descriptor, and the thread pointer
2726 * will be NULL. Callers should be careful only to pass a
2727 * NULL thread pointer when there really is no owning
2728 * context that might have locks, or the locks will be
2731 if (fp->f_type == DTYPE_VNODE) {
2733 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2734 lf.l_whence = SEEK_SET;
2737 lf.l_type = F_UNLCK;
2738 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2739 F_UNLCK, &lf, F_POSIX);
2741 fdtol = td->td_proc->p_fdtol;
2742 if (fdtol != NULL) {
2744 * Handle special case where file descriptor table is
2745 * shared between multiple process leaders.
2747 fdp = td->td_proc->p_fd;
2748 FILEDESC_XLOCK(fdp);
2749 for (fdtol = fdtol->fdl_next;
2750 fdtol != td->td_proc->p_fdtol;
2751 fdtol = fdtol->fdl_next) {
2752 if ((fdtol->fdl_leader->p_flag &
2755 fdtol->fdl_holdcount++;
2756 FILEDESC_XUNLOCK(fdp);
2757 lf.l_whence = SEEK_SET;
2760 lf.l_type = F_UNLCK;
2762 (void) VOP_ADVLOCK(vp,
2763 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2765 FILEDESC_XLOCK(fdp);
2766 fdtol->fdl_holdcount--;
2767 if (fdtol->fdl_holdcount == 0 &&
2768 fdtol->fdl_wakeup != 0) {
2769 fdtol->fdl_wakeup = 0;
2773 FILEDESC_XUNLOCK(fdp);
2776 return (fdrop_close(fp, td));
2780 * Hack for file descriptor passing code.
2783 closef_nothread(struct file *fp)
2790 * Initialize the file pointer with the specified properties.
2792 * The ops are set with release semantics to be certain that the flags, type,
2793 * and data are visible when ops is. This is to prevent ops methods from being
2794 * called with bad data.
2797 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2802 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2806 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2808 fp->f_seqcount[UIO_READ] = 1;
2809 fp->f_seqcount[UIO_WRITE] = 1;
2810 finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2815 fget_cap_noref(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2816 struct file **fpp, struct filecaps *havecapsp)
2818 struct filedescent *fde;
2821 FILEDESC_LOCK_ASSERT(fdp);
2824 fde = fdeget_noref(fdp, fd);
2831 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2836 if (havecapsp != NULL)
2837 filecaps_copy(&fde->fde_caps, havecapsp, true);
2839 *fpp = fde->fde_file;
2848 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2849 struct file **fpp, struct filecaps *havecapsp)
2851 struct filedesc *fdp = td->td_proc->p_fd;
2858 error = fget_unlocked_seq(td, fd, needrightsp, &fp, &seq);
2862 if (havecapsp != NULL) {
2863 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2864 havecapsp, false)) {
2870 if (!fd_modified(fdp, fd, seq))
2879 FILEDESC_SLOCK(fdp);
2880 error = fget_cap_noref(fdp, fd, needrightsp, fpp, havecapsp);
2881 if (error == 0 && !fhold(*fpp))
2883 FILEDESC_SUNLOCK(fdp);
2888 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2889 struct file **fpp, struct filecaps *havecapsp)
2892 error = fget_unlocked(td, fd, needrightsp, fpp);
2893 if (havecapsp != NULL && error == 0)
2894 filecaps_fill(havecapsp);
2902 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2904 const struct filedescent *fde;
2905 const struct fdescenttbl *fdt;
2906 struct filedesc *fdp;
2909 const cap_rights_t *haverights;
2910 cap_rights_t rights;
2913 VFS_SMR_ASSERT_ENTERED();
2915 rights = *ndp->ni_rightsneeded;
2916 cap_rights_set_one(&rights, CAP_LOOKUP);
2918 fdp = curproc->p_fd;
2919 fdt = fdp->fd_files;
2920 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2922 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
2923 fde = &fdt->fdt_ofiles[fd];
2924 haverights = cap_rights_fde_inline(fde);
2926 if (__predict_false(fp == NULL))
2928 if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2930 *fsearch = ((fp->f_flag & FSEARCH) != 0);
2932 if (__predict_false(vp == NULL)) {
2935 if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2939 * Use an acquire barrier to force re-reading of fdt so it is
2940 * refreshed for verification.
2942 atomic_thread_fence_acq();
2943 fdt = fdp->fd_files;
2944 if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)))
2947 * If file descriptor doesn't have all rights,
2948 * all lookups relative to it must also be
2949 * strictly relative.
2951 * Not yet supported by fast path.
2954 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
2955 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
2956 ndp->ni_filecaps.fc_nioctls != -1) {
2958 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
2968 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2970 const struct fdescenttbl *fdt;
2971 struct filedesc *fdp;
2975 VFS_SMR_ASSERT_ENTERED();
2977 fdp = curproc->p_fd;
2978 fdt = fdp->fd_files;
2979 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2981 fp = fdt->fdt_ofiles[fd].fde_file;
2982 if (__predict_false(fp == NULL))
2984 *fsearch = ((fp->f_flag & FSEARCH) != 0);
2986 if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2990 * Use an acquire barrier to force re-reading of fdt so it is
2991 * refreshed for verification.
2993 atomic_thread_fence_acq();
2994 fdt = fdp->fd_files;
2995 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
2997 filecaps_fill(&ndp->ni_filecaps);
3004 * Fetch the descriptor locklessly.
3006 * We avoid fdrop() races by never raising a refcount above 0. To accomplish
3007 * this we have to use a cmpset loop rather than an atomic_add. The descriptor
3008 * must be re-verified once we acquire a reference to be certain that the
3009 * identity is still correct and we did not lose a race due to preemption.
3011 * Force a reload of fdt when looping. Another thread could reallocate
3012 * the table before this fd was closed, so it is possible that there is
3013 * a stale fp pointer in cached version.
3017 fget_unlocked_seq(struct thread *td, int fd, cap_rights_t *needrightsp,
3018 struct file **fpp, seqc_t *seqp)
3020 struct filedesc *fdp;
3021 const struct filedescent *fde;
3022 const struct fdescenttbl *fdt;
3025 cap_rights_t haverights;
3028 fdp = td->td_proc->p_fd;
3029 fdt = fdp->fd_files;
3030 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3034 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3035 fde = &fdt->fdt_ofiles[fd];
3036 haverights = *cap_rights_fde_inline(fde);
3038 if (__predict_false(fp == NULL)) {
3039 if (seqc_consistent(fd_seqc(fdt, fd), seq))
3041 fdt = atomic_load_ptr(&fdp->fd_files);
3044 error = cap_check_inline(&haverights, needrightsp);
3045 if (__predict_false(error != 0)) {
3046 if (seqc_consistent(fd_seqc(fdt, fd), seq))
3048 fdt = atomic_load_ptr(&fdp->fd_files);
3051 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3052 fdt = atomic_load_ptr(&fdp->fd_files);
3056 * Use an acquire barrier to force re-reading of fdt so it is
3057 * refreshed for verification.
3059 atomic_thread_fence_acq();
3060 fdt = fdp->fd_files;
3061 if (seqc_consistent_no_fence(fd_seqc(fdt, fd), seq))
3073 fget_unlocked_seq(struct thread *td, int fd, cap_rights_t *needrightsp,
3074 struct file **fpp, seqc_t *seqp __unused)
3076 struct filedesc *fdp;
3077 const struct fdescenttbl *fdt;
3080 fdp = td->td_proc->p_fd;
3081 fdt = fdp->fd_files;
3082 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3086 fp = fdt->fdt_ofiles[fd].fde_file;
3087 if (__predict_false(fp == NULL))
3089 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3090 fdt = atomic_load_ptr(&fdp->fd_files);
3094 * Use an acquire barrier to force re-reading of fdt so it is
3095 * refreshed for verification.
3097 atomic_thread_fence_acq();
3098 fdt = fdp->fd_files;
3099 if (__predict_true(fp == fdt->fdt_ofiles[fd].fde_file))
3109 * See the comments in fget_unlocked_seq for an explanation of how this works.
3111 * This is a simplified variant which bails out to the aforementioned routine
3112 * if anything goes wrong. In practice this only happens when userspace is
3113 * racing with itself.
3116 fget_unlocked(struct thread *td, int fd, cap_rights_t *needrightsp,
3119 struct filedesc *fdp;
3121 const struct filedescent *fde;
3123 const struct fdescenttbl *fdt;
3127 const cap_rights_t *haverights;
3130 fdp = td->td_proc->p_fd;
3131 fdt = fdp->fd_files;
3132 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) {
3137 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3138 fde = &fdt->fdt_ofiles[fd];
3139 haverights = cap_rights_fde_inline(fde);
3142 fp = fdt->fdt_ofiles[fd].fde_file;
3144 if (__predict_false(fp == NULL))
3147 if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3150 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3154 * Use an acquire barrier to force re-reading of fdt so it is
3155 * refreshed for verification.
3157 atomic_thread_fence_acq();
3158 fdt = fdp->fd_files;
3160 if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)))
3162 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3171 return (fget_unlocked_seq(td, fd, needrightsp, fpp, NULL));
3175 * Translate fd -> file when the caller guarantees the file descriptor table
3176 * can't be changed by others.
3178 * Note this does not mean the file object itself is only visible to the caller,
3179 * merely that it wont disappear without having to be referenced.
3181 * Must be paired with fput_only_user.
3185 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3188 const struct filedescent *fde;
3189 const struct fdescenttbl *fdt;
3190 const cap_rights_t *haverights;
3194 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3197 if (__predict_false(fd >= fdp->fd_nfiles))
3200 fdt = fdp->fd_files;
3201 fde = &fdt->fdt_ofiles[fd];
3203 if (__predict_false(fp == NULL))
3205 MPASS(refcount_load(&fp->f_count) > 0);
3206 haverights = cap_rights_fde_inline(fde);
3207 error = cap_check_inline(haverights, needrightsp);
3208 if (__predict_false(error != 0))
3215 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3220 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3223 if (__predict_false(fd >= fdp->fd_nfiles))
3226 fp = fdp->fd_ofiles[fd].fde_file;
3227 if (__predict_false(fp == NULL))
3230 MPASS(refcount_load(&fp->f_count) > 0);
3237 * Extract the file pointer associated with the specified descriptor for the
3238 * current user process.
3240 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3243 * File's rights will be checked against the capability rights mask.
3245 * If an error occurred the non-zero error is returned and *fpp is set to
3246 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
3247 * responsible for fdrop().
3250 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3251 cap_rights_t *needrightsp)
3257 error = fget_unlocked(td, fd, needrightsp, &fp);
3258 if (__predict_false(error != 0))
3260 if (__predict_false(fp->f_ops == &badfileops)) {
3266 * FREAD and FWRITE failure return EBADF as per POSIX.
3272 if ((fp->f_flag & flags) == 0)
3276 if (fp->f_ops != &path_fileops &&
3277 ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3278 (fp->f_flag & FWRITE) != 0))
3284 KASSERT(0, ("wrong flags"));
3297 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3300 return (_fget(td, fd, fpp, 0, rightsp));
3304 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3308 #ifndef CAPABILITIES
3309 error = _fget(td, fd, fpp, 0, rightsp);
3310 if (maxprotp != NULL)
3311 *maxprotp = VM_PROT_ALL;
3314 cap_rights_t fdrights;
3315 struct filedesc *fdp;
3320 fdp = td->td_proc->p_fd;
3321 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3323 error = fget_unlocked_seq(td, fd, rightsp, &fp, &seq);
3324 if (__predict_false(error != 0))
3326 if (__predict_false(fp->f_ops == &badfileops)) {
3330 if (maxprotp != NULL)
3331 fdrights = *cap_rights(fdp, fd);
3332 if (!fd_modified(fdp, fd, seq))
3338 * If requested, convert capability rights to access flags.
3340 if (maxprotp != NULL)
3341 *maxprotp = cap_rights_to_vmprot(&fdrights);
3348 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3351 return (_fget(td, fd, fpp, FREAD, rightsp));
3355 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3358 return (_fget(td, fd, fpp, FWRITE, rightsp));
3362 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3365 #ifndef CAPABILITIES
3366 return (fget_unlocked(td, fd, rightsp, fpp));
3368 struct filedesc *fdp = td->td_proc->p_fd;
3374 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3376 error = fget_unlocked_seq(td, fd, rightsp, &fp, &seq);
3379 error = cap_fcntl_check(fdp, fd, needfcntl);
3380 if (!fd_modified(fdp, fd, seq))
3394 * Like fget() but loads the underlying vnode, or returns an error if the
3395 * descriptor does not represent a vnode. Note that pipes use vnodes but
3396 * never have VM objects. The returned vnode will be vref()'d.
3398 * XXX: what about the unused flags ?
3401 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3408 error = _fget(td, fd, &fp, flags, needrightsp);
3411 if (fp->f_vnode == NULL) {
3423 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3426 return (_fgetvp(td, fd, 0, rightsp, vpp));
3430 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3431 struct filecaps *havecaps, struct vnode **vpp)
3433 struct filecaps caps;
3437 error = fget_cap(td, fd, needrightsp, &fp, &caps);
3440 if (fp->f_ops == &badfileops) {
3444 if (fp->f_vnode == NULL) {
3456 filecaps_free(&caps);
3462 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3465 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3469 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3472 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3477 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3481 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3486 * Handle the last reference to a file being closed.
3488 * Without the noinline attribute clang keeps inlining the func thorough this
3489 * file when fdrop is used.
3492 _fdrop(struct file *fp, struct thread *td)
3498 count = refcount_load(&fp->f_count);
3500 panic("fdrop: fp %p count %d", fp, count);
3502 error = fo_close(fp, td);
3503 atomic_subtract_int(&openfiles, 1);
3505 free(fp->f_advice, M_FADVISE);
3506 uma_zfree(file_zone, fp);
3512 * Apply an advisory lock on a file descriptor.
3514 * Just attempt to get a record lock of the requested type on the entire file
3515 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3517 #ifndef _SYS_SYSPROTO_H_
3525 sys_flock(struct thread *td, struct flock_args *uap)
3532 error = fget(td, uap->fd, &cap_flock_rights, &fp);
3536 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
3539 if (fp->f_ops == &path_fileops) {
3545 lf.l_whence = SEEK_SET;
3548 if (uap->how & LOCK_UN) {
3549 lf.l_type = F_UNLCK;
3550 atomic_clear_int(&fp->f_flag, FHASLOCK);
3551 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3554 if (uap->how & LOCK_EX)
3555 lf.l_type = F_WRLCK;
3556 else if (uap->how & LOCK_SH)
3557 lf.l_type = F_RDLCK;
3562 atomic_set_int(&fp->f_flag, FHASLOCK);
3563 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3564 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3570 * Duplicate the specified descriptor to a free descriptor.
3573 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3574 int openerror, int *indxp)
3576 struct filedescent *newfde, *oldfde;
3581 KASSERT(openerror == ENODEV || openerror == ENXIO,
3582 ("unexpected error %d in %s", openerror, __func__));
3585 * If the to-be-dup'd fd number is greater than the allowed number
3586 * of file descriptors, or the fd to be dup'd has already been
3587 * closed, then reject.
3589 FILEDESC_XLOCK(fdp);
3590 if ((fp = fget_noref(fdp, dfd)) == NULL) {
3591 FILEDESC_XUNLOCK(fdp);
3595 error = fdalloc(td, 0, &indx);
3597 FILEDESC_XUNLOCK(fdp);
3602 * There are two cases of interest here.
3604 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3606 * For ENXIO steal away the file structure from (dfd) and store it in
3607 * (indx). (dfd) is effectively closed by this operation.
3609 switch (openerror) {
3612 * Check that the mode the file is being opened for is a
3613 * subset of the mode of the existing descriptor.
3615 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3616 fdunused(fdp, indx);
3617 FILEDESC_XUNLOCK(fdp);
3621 fdunused(fdp, indx);
3622 FILEDESC_XUNLOCK(fdp);
3625 newfde = &fdp->fd_ofiles[indx];
3626 oldfde = &fdp->fd_ofiles[dfd];
3627 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3629 seqc_write_begin(&newfde->fde_seqc);
3631 fde_copy(oldfde, newfde);
3632 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3635 seqc_write_end(&newfde->fde_seqc);
3640 * Steal away the file pointer from dfd and stuff it into indx.
3642 newfde = &fdp->fd_ofiles[indx];
3643 oldfde = &fdp->fd_ofiles[dfd];
3645 seqc_write_begin(&oldfde->fde_seqc);
3646 seqc_write_begin(&newfde->fde_seqc);
3648 fde_copy(oldfde, newfde);
3649 oldfde->fde_file = NULL;
3652 seqc_write_end(&newfde->fde_seqc);
3653 seqc_write_end(&oldfde->fde_seqc);
3657 FILEDESC_XUNLOCK(fdp);
3663 * This sysctl determines if we will allow a process to chroot(2) if it
3664 * has a directory open:
3665 * 0: disallowed for all processes.
3666 * 1: allowed for processes that were not already chroot(2)'ed.
3667 * 2: allowed for all processes.
3670 static int chroot_allow_open_directories = 1;
3672 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3673 &chroot_allow_open_directories, 0,
3674 "Allow a process to chroot(2) if it has a directory open");
3677 * Helper function for raised chroot(2) security function: Refuse if
3678 * any filedescriptors are open directories.
3681 chroot_refuse_vdir_fds(struct filedesc *fdp)
3687 FILEDESC_LOCK_ASSERT(fdp);
3689 FILEDESC_FOREACH_FP(fdp, i, fp) {
3690 if (fp->f_type == DTYPE_VNODE) {
3692 if (vp->v_type == VDIR)
3700 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3703 if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3704 vrefact(oldpwd->pwd_cdir);
3705 newpwd->pwd_cdir = oldpwd->pwd_cdir;
3708 if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3709 vrefact(oldpwd->pwd_rdir);
3710 newpwd->pwd_rdir = oldpwd->pwd_rdir;
3713 if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3714 vrefact(oldpwd->pwd_jdir);
3715 newpwd->pwd_jdir = oldpwd->pwd_jdir;
3720 pwd_hold_pwddesc(struct pwddesc *pdp)
3724 PWDDESC_ASSERT_XLOCKED(pdp);
3725 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3727 refcount_acquire(&pwd->pwd_refcount);
3732 pwd_hold_smr(struct pwd *pwd)
3736 if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3743 pwd_hold(struct thread *td)
3745 struct pwddesc *pdp;
3748 pdp = td->td_proc->p_pd;
3751 pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3752 if (pwd_hold_smr(pwd)) {
3758 pwd = pwd_hold_pwddesc(pdp);
3760 PWDDESC_XUNLOCK(pdp);
3765 pwd_hold_proc(struct proc *p)
3767 struct pwddesc *pdp;
3770 PROC_ASSERT_HELD(p);
3777 pwd = pwd_hold_pwddesc(pdp);
3779 PWDDESC_XUNLOCK(pdp);
3789 pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3790 bzero(pwd, sizeof(*pwd));
3791 refcount_init(&pwd->pwd_refcount, 1);
3796 pwd_drop(struct pwd *pwd)
3799 if (!refcount_release(&pwd->pwd_refcount))
3802 if (pwd->pwd_cdir != NULL)
3803 vrele(pwd->pwd_cdir);
3804 if (pwd->pwd_rdir != NULL)
3805 vrele(pwd->pwd_rdir);
3806 if (pwd->pwd_jdir != NULL)
3807 vrele(pwd->pwd_jdir);
3808 uma_zfree_smr(pwd_zone, pwd);
3812 * The caller is responsible for invoking priv_check() and
3813 * mac_vnode_check_chroot() to authorize this operation.
3816 pwd_chroot(struct thread *td, struct vnode *vp)
3818 struct pwddesc *pdp;
3819 struct filedesc *fdp;
3820 struct pwd *newpwd, *oldpwd;
3823 fdp = td->td_proc->p_fd;
3824 pdp = td->td_proc->p_pd;
3825 newpwd = pwd_alloc();
3826 FILEDESC_SLOCK(fdp);
3828 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3829 if (chroot_allow_open_directories == 0 ||
3830 (chroot_allow_open_directories == 1 &&
3831 oldpwd->pwd_rdir != rootvnode)) {
3832 error = chroot_refuse_vdir_fds(fdp);
3833 FILEDESC_SUNLOCK(fdp);
3835 PWDDESC_XUNLOCK(pdp);
3840 FILEDESC_SUNLOCK(fdp);
3844 newpwd->pwd_rdir = vp;
3845 if (oldpwd->pwd_jdir == NULL) {
3847 newpwd->pwd_jdir = vp;
3849 pwd_fill(oldpwd, newpwd);
3850 pwd_set(pdp, newpwd);
3851 PWDDESC_XUNLOCK(pdp);
3857 pwd_chdir(struct thread *td, struct vnode *vp)
3859 struct pwddesc *pdp;
3860 struct pwd *newpwd, *oldpwd;
3862 VNPASS(vp->v_usecount > 0, vp);
3864 newpwd = pwd_alloc();
3865 pdp = td->td_proc->p_pd;
3867 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3868 newpwd->pwd_cdir = vp;
3869 pwd_fill(oldpwd, newpwd);
3870 pwd_set(pdp, newpwd);
3871 PWDDESC_XUNLOCK(pdp);
3876 * jail_attach(2) changes both root and working directories.
3879 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
3881 struct pwddesc *pdp;
3882 struct filedesc *fdp;
3883 struct pwd *newpwd, *oldpwd;
3886 fdp = td->td_proc->p_fd;
3887 pdp = td->td_proc->p_pd;
3888 newpwd = pwd_alloc();
3889 FILEDESC_SLOCK(fdp);
3891 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3892 error = chroot_refuse_vdir_fds(fdp);
3893 FILEDESC_SUNLOCK(fdp);
3895 PWDDESC_XUNLOCK(pdp);
3901 newpwd->pwd_rdir = vp;
3903 newpwd->pwd_cdir = vp;
3904 if (oldpwd->pwd_jdir == NULL) {
3906 newpwd->pwd_jdir = vp;
3908 pwd_fill(oldpwd, newpwd);
3909 pwd_set(pdp, newpwd);
3910 PWDDESC_XUNLOCK(pdp);
3916 pwd_ensure_dirs(void)
3918 struct pwddesc *pdp;
3919 struct pwd *oldpwd, *newpwd;
3921 pdp = curproc->p_pd;
3923 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3924 if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3925 PWDDESC_XUNLOCK(pdp);
3928 PWDDESC_XUNLOCK(pdp);
3930 newpwd = pwd_alloc();
3932 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3933 pwd_fill(oldpwd, newpwd);
3934 if (newpwd->pwd_cdir == NULL) {
3936 newpwd->pwd_cdir = rootvnode;
3938 if (newpwd->pwd_rdir == NULL) {
3940 newpwd->pwd_rdir = rootvnode;
3942 pwd_set(pdp, newpwd);
3943 PWDDESC_XUNLOCK(pdp);
3948 pwd_set_rootvnode(void)
3950 struct pwddesc *pdp;
3951 struct pwd *oldpwd, *newpwd;
3953 pdp = curproc->p_pd;
3955 newpwd = pwd_alloc();
3957 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3959 newpwd->pwd_cdir = rootvnode;
3961 newpwd->pwd_rdir = rootvnode;
3962 pwd_fill(oldpwd, newpwd);
3963 pwd_set(pdp, newpwd);
3964 PWDDESC_XUNLOCK(pdp);
3969 * Scan all active processes and prisons to see if any of them have a current
3970 * or root directory of `olddp'. If so, replace them with the new mount point.
3973 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3975 struct pwddesc *pdp;
3976 struct pwd *newpwd, *oldpwd;
3981 if (vrefcnt(olddp) == 1)
3984 newpwd = pwd_alloc();
3985 sx_slock(&allproc_lock);
3986 FOREACH_PROC_IN_SYSTEM(p) {
3993 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3994 if (oldpwd == NULL ||
3995 (oldpwd->pwd_cdir != olddp &&
3996 oldpwd->pwd_rdir != olddp &&
3997 oldpwd->pwd_jdir != olddp)) {
3998 PWDDESC_XUNLOCK(pdp);
4002 if (oldpwd->pwd_cdir == olddp) {
4004 newpwd->pwd_cdir = newdp;
4006 if (oldpwd->pwd_rdir == olddp) {
4008 newpwd->pwd_rdir = newdp;
4010 if (oldpwd->pwd_jdir == olddp) {
4012 newpwd->pwd_jdir = newdp;
4014 pwd_fill(oldpwd, newpwd);
4015 pwd_set(pdp, newpwd);
4016 PWDDESC_XUNLOCK(pdp);
4019 newpwd = pwd_alloc();
4021 sx_sunlock(&allproc_lock);
4023 if (rootvnode == olddp) {
4028 mtx_lock(&prison0.pr_mtx);
4029 if (prison0.pr_root == olddp) {
4031 prison0.pr_root = newdp;
4034 mtx_unlock(&prison0.pr_mtx);
4035 sx_slock(&allprison_lock);
4036 TAILQ_FOREACH(pr, &allprison, pr_list) {
4037 mtx_lock(&pr->pr_mtx);
4038 if (pr->pr_root == olddp) {
4040 pr->pr_root = newdp;
4043 mtx_unlock(&pr->pr_mtx);
4045 sx_sunlock(&allprison_lock);
4050 struct filedesc_to_leader *
4051 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
4053 struct filedesc_to_leader *fdtol;
4055 fdtol = malloc(sizeof(struct filedesc_to_leader),
4056 M_FILEDESC_TO_LEADER, M_WAITOK);
4057 fdtol->fdl_refcount = 1;
4058 fdtol->fdl_holdcount = 0;
4059 fdtol->fdl_wakeup = 0;
4060 fdtol->fdl_leader = leader;
4062 FILEDESC_XLOCK(fdp);
4063 fdtol->fdl_next = old->fdl_next;
4064 fdtol->fdl_prev = old;
4065 old->fdl_next = fdtol;
4066 fdtol->fdl_next->fdl_prev = fdtol;
4067 FILEDESC_XUNLOCK(fdp);
4069 fdtol->fdl_next = fdtol;
4070 fdtol->fdl_prev = fdtol;
4076 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
4079 struct filedesc *fdp;
4081 int count, off, minoff;
4087 if (*(int *)arg1 != 0)
4090 fdp = curproc->p_fd;
4092 FILEDESC_SLOCK(fdp);
4094 off = NDSLOT(fdp->fd_nfiles - 1);
4095 for (minoff = NDSLOT(0); off >= minoff; --off)
4096 count += bitcountl(map[off]);
4097 FILEDESC_SUNLOCK(fdp);
4099 return (SYSCTL_OUT(req, &count, sizeof(count)));
4102 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
4103 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
4104 "Number of open file descriptors");
4107 * Get file structures globally.
4110 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
4113 struct filedesc *fdp;
4118 error = sysctl_wire_old_buffer(req, 0);
4121 if (req->oldptr == NULL) {
4123 sx_slock(&allproc_lock);
4124 FOREACH_PROC_IN_SYSTEM(p) {
4126 if (p->p_state == PRS_NEW) {
4134 /* overestimates sparse tables. */
4135 n += fdp->fd_nfiles;
4138 sx_sunlock(&allproc_lock);
4139 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4142 bzero(&xf, sizeof(xf));
4143 xf.xf_size = sizeof(xf);
4144 sx_slock(&allproc_lock);
4145 FOREACH_PROC_IN_SYSTEM(p) {
4147 if (p->p_state == PRS_NEW) {
4151 if (p_cansee(req->td, p) != 0) {
4155 xf.xf_pid = p->p_pid;
4156 xf.xf_uid = p->p_ucred->cr_uid;
4161 FILEDESC_SLOCK(fdp);
4162 FILEDESC_FOREACH_FP(fdp, n, fp) {
4163 if (refcount_load(&fdp->fd_refcnt) == 0)
4166 xf.xf_file = (uintptr_t)fp;
4167 xf.xf_data = (uintptr_t)fp->f_data;
4168 xf.xf_vnode = (uintptr_t)fp->f_vnode;
4169 xf.xf_type = (uintptr_t)fp->f_type;
4170 xf.xf_count = refcount_load(&fp->f_count);
4172 xf.xf_offset = foffset_get(fp);
4173 xf.xf_flag = fp->f_flag;
4174 error = SYSCTL_OUT(req, &xf, sizeof(xf));
4178 FILEDESC_SUNLOCK(fdp);
4183 sx_sunlock(&allproc_lock);
4187 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4188 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4190 #ifdef KINFO_FILE_SIZE
4191 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4195 xlate_fflags(int fflags)
4197 static const struct {
4200 } fflags_table[] = {
4201 { FAPPEND, KF_FLAG_APPEND },
4202 { FASYNC, KF_FLAG_ASYNC },
4203 { FFSYNC, KF_FLAG_FSYNC },
4204 { FHASLOCK, KF_FLAG_HASLOCK },
4205 { FNONBLOCK, KF_FLAG_NONBLOCK },
4206 { FREAD, KF_FLAG_READ },
4207 { FWRITE, KF_FLAG_WRITE },
4208 { O_CREAT, KF_FLAG_CREAT },
4209 { O_DIRECT, KF_FLAG_DIRECT },
4210 { O_EXCL, KF_FLAG_EXCL },
4211 { O_EXEC, KF_FLAG_EXEC },
4212 { O_EXLOCK, KF_FLAG_EXLOCK },
4213 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4214 { O_SHLOCK, KF_FLAG_SHLOCK },
4215 { O_TRUNC, KF_FLAG_TRUNC }
4221 for (i = 0; i < nitems(fflags_table); i++)
4222 if (fflags & fflags_table[i].fflag)
4223 kflags |= fflags_table[i].kf_fflag;
4227 /* Trim unused data from kf_path by truncating the structure size. */
4229 pack_kinfo(struct kinfo_file *kif)
4232 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4233 strlen(kif->kf_path) + 1;
4234 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4238 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4239 struct kinfo_file *kif, struct filedesc *fdp, int flags)
4243 bzero(kif, sizeof(*kif));
4245 /* Set a default type to allow for empty fill_kinfo() methods. */
4246 kif->kf_type = KF_TYPE_UNKNOWN;
4247 kif->kf_flags = xlate_fflags(fp->f_flag);
4248 if (rightsp != NULL)
4249 kif->kf_cap_rights = *rightsp;
4251 cap_rights_init_zero(&kif->kf_cap_rights);
4253 kif->kf_ref_count = refcount_load(&fp->f_count);
4254 kif->kf_offset = foffset_get(fp);
4257 * This may drop the filedesc lock, so the 'fp' cannot be
4258 * accessed after this call.
4260 error = fo_fill_kinfo(fp, kif, fdp);
4262 kif->kf_status |= KF_ATTR_VALID;
4263 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4266 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4270 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4271 struct kinfo_file *kif, int flags)
4275 bzero(kif, sizeof(*kif));
4277 kif->kf_type = KF_TYPE_VNODE;
4278 error = vn_fill_kinfo_vnode(vp, kif);
4280 kif->kf_status |= KF_ATTR_VALID;
4281 kif->kf_flags = xlate_fflags(fflags);
4282 cap_rights_init_zero(&kif->kf_cap_rights);
4284 kif->kf_ref_count = -1;
4285 kif->kf_offset = -1;
4286 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4289 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4293 struct export_fd_buf {
4294 struct filedesc *fdp;
4295 struct pwddesc *pdp;
4298 struct kinfo_file kif;
4303 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4305 struct kinfo_file *kif;
4308 if (efbuf->remainder != -1) {
4309 if (efbuf->remainder < kif->kf_structsize)
4311 efbuf->remainder -= kif->kf_structsize;
4313 if (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) != 0)
4314 return (sbuf_error(efbuf->sb));
4319 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4320 struct export_fd_buf *efbuf)
4324 if (efbuf->remainder == 0)
4326 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4328 FILEDESC_SUNLOCK(efbuf->fdp);
4329 error = export_kinfo_to_sb(efbuf);
4330 FILEDESC_SLOCK(efbuf->fdp);
4335 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4336 struct export_fd_buf *efbuf)
4340 if (efbuf->remainder == 0)
4342 if (efbuf->pdp != NULL)
4343 PWDDESC_XUNLOCK(efbuf->pdp);
4344 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4345 error = export_kinfo_to_sb(efbuf);
4346 if (efbuf->pdp != NULL)
4347 PWDDESC_XLOCK(efbuf->pdp);
4352 * Store a process file descriptor information to sbuf.
4354 * Takes a locked proc as argument, and returns with the proc unlocked.
4357 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
4361 struct filedesc *fdp;
4362 struct pwddesc *pdp;
4363 struct export_fd_buf *efbuf;
4364 struct vnode *cttyvp, *textvp, *tracevp;
4367 cap_rights_t rights;
4369 PROC_LOCK_ASSERT(p, MA_OWNED);
4372 tracevp = ktr_get_tracevp(p, true);
4374 textvp = p->p_textvp;
4377 /* Controlling tty. */
4379 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4380 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4388 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4392 efbuf->remainder = maxlen;
4393 efbuf->flags = flags;
4396 if (tracevp != NULL)
4397 error = export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE,
4398 FREAD | FWRITE, efbuf);
4399 if (error == 0 && textvp != NULL)
4400 error = export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD,
4402 if (error == 0 && cttyvp != NULL)
4403 error = export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY,
4404 FREAD | FWRITE, efbuf);
4405 if (error != 0 || pdp == NULL || fdp == NULL)
4410 pwd = pwd_hold_pwddesc(pdp);
4412 /* working directory */
4413 if (pwd->pwd_cdir != NULL) {
4414 vrefact(pwd->pwd_cdir);
4415 error = export_vnode_to_sb(pwd->pwd_cdir,
4416 KF_FD_TYPE_CWD, FREAD, efbuf);
4418 /* root directory */
4419 if (error == 0 && pwd->pwd_rdir != NULL) {
4420 vrefact(pwd->pwd_rdir);
4421 error = export_vnode_to_sb(pwd->pwd_rdir,
4422 KF_FD_TYPE_ROOT, FREAD, efbuf);
4424 /* jail directory */
4425 if (error == 0 && pwd->pwd_jdir != NULL) {
4426 vrefact(pwd->pwd_jdir);
4427 error = export_vnode_to_sb(pwd->pwd_jdir,
4428 KF_FD_TYPE_JAIL, FREAD, efbuf);
4431 PWDDESC_XUNLOCK(pdp);
4436 FILEDESC_SLOCK(fdp);
4437 FILEDESC_FOREACH_FP(fdp, i, fp) {
4438 if (refcount_load(&fdp->fd_refcnt) == 0)
4441 rights = *cap_rights(fdp, i);
4442 #else /* !CAPABILITIES */
4443 rights = cap_no_rights;
4446 * Create sysctl entry. It is OK to drop the filedesc
4447 * lock inside of export_file_to_sb() as we will
4448 * re-validate and re-evaluate its properties when the
4451 error = export_file_to_sb(fp, i, &rights, efbuf);
4455 FILEDESC_SUNLOCK(fdp);
4461 free(efbuf, M_TEMP);
4465 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
4468 * Get per-process file descriptors for use by procstat(1), et al.
4471 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4477 int error, error2, *name;
4485 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4486 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4487 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4492 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4493 error = kern_proc_filedesc_out(p, &sb, maxlen,
4494 KERN_FILEDESC_PACK_KINFO);
4495 error2 = sbuf_finish(&sb);
4497 return (error != 0 ? error : error2);
4500 #ifdef COMPAT_FREEBSD7
4501 #ifdef KINFO_OFILE_SIZE
4502 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4506 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4509 okif->kf_structsize = sizeof(*okif);
4510 okif->kf_type = kif->kf_type;
4511 okif->kf_fd = kif->kf_fd;
4512 okif->kf_ref_count = kif->kf_ref_count;
4513 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4514 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4515 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4516 okif->kf_offset = kif->kf_offset;
4517 if (kif->kf_type == KF_TYPE_VNODE)
4518 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4520 okif->kf_vnode_type = KF_VTYPE_VNON;
4521 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4522 if (kif->kf_type == KF_TYPE_SOCKET) {
4523 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4524 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4525 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4526 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4527 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4529 okif->kf_sa_local.ss_family = AF_UNSPEC;
4530 okif->kf_sa_peer.ss_family = AF_UNSPEC;
4535 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4536 struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4541 PWDDESC_XUNLOCK(pdp);
4542 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4543 kinfo_to_okinfo(kif, okif);
4544 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4550 * Get per-process file descriptors for use by procstat(1), et al.
4553 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4555 struct kinfo_ofile *okif;
4556 struct kinfo_file *kif;
4557 struct filedesc *fdp;
4558 struct pwddesc *pdp;
4561 int error, i, *name;
4570 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4577 if (fdp == NULL || pdp == NULL) {
4582 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4583 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4585 pwd = pwd_hold_pwddesc(pdp);
4587 if (pwd->pwd_cdir != NULL)
4588 export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4590 if (pwd->pwd_rdir != NULL)
4591 export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4593 if (pwd->pwd_jdir != NULL)
4594 export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4597 PWDDESC_XUNLOCK(pdp);
4600 FILEDESC_SLOCK(fdp);
4601 FILEDESC_FOREACH_FP(fdp, i, fp) {
4602 if (refcount_load(&fdp->fd_refcnt) == 0)
4604 export_file_to_kinfo(fp, i, NULL, kif, fdp,
4605 KERN_FILEDESC_PACK_KINFO);
4606 FILEDESC_SUNLOCK(fdp);
4607 kinfo_to_okinfo(kif, okif);
4608 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4609 FILEDESC_SLOCK(fdp);
4613 FILEDESC_SUNLOCK(fdp);
4621 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4622 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4623 "Process ofiledesc entries");
4624 #endif /* COMPAT_FREEBSD7 */
4627 vntype_to_kinfo(int vtype)
4632 } vtypes_table[] = {
4633 { VBAD, KF_VTYPE_VBAD },
4634 { VBLK, KF_VTYPE_VBLK },
4635 { VCHR, KF_VTYPE_VCHR },
4636 { VDIR, KF_VTYPE_VDIR },
4637 { VFIFO, KF_VTYPE_VFIFO },
4638 { VLNK, KF_VTYPE_VLNK },
4639 { VNON, KF_VTYPE_VNON },
4640 { VREG, KF_VTYPE_VREG },
4641 { VSOCK, KF_VTYPE_VSOCK }
4646 * Perform vtype translation.
4648 for (i = 0; i < nitems(vtypes_table); i++)
4649 if (vtypes_table[i].vtype == vtype)
4650 return (vtypes_table[i].kf_vtype);
4652 return (KF_VTYPE_UNKNOWN);
4655 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4656 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4657 "Process filedesc entries");
4660 * Store a process current working directory information to sbuf.
4662 * Takes a locked proc as argument, and returns with the proc unlocked.
4665 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
4667 struct pwddesc *pdp;
4669 struct export_fd_buf *efbuf;
4673 PROC_LOCK_ASSERT(p, MA_OWNED);
4680 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4684 efbuf->remainder = maxlen;
4688 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4689 cdir = pwd->pwd_cdir;
4694 error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4696 PWDDESC_XUNLOCK(pdp);
4698 free(efbuf, M_TEMP);
4703 * Get per-process current working directory.
4706 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4712 int error, error2, *name;
4720 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4721 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4722 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4727 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4728 error = kern_proc_cwd_out(p, &sb, maxlen);
4729 error2 = sbuf_finish(&sb);
4731 return (error != 0 ? error : error2);
4734 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4735 sysctl_kern_proc_cwd, "Process current working directory");
4739 * For the purposes of debugging, generate a human-readable string for the
4743 file_type_to_name(short type)
4771 case DTYPE_PROCDESC:
4775 case DTYPE_LINUXTFD:
4783 * For the purposes of debugging, identify a process (if any, perhaps one of
4784 * many) that references the passed file in its file descriptor array. Return
4787 static struct proc *
4788 file_to_first_proc(struct file *fp)
4790 struct filedesc *fdp;
4794 FOREACH_PROC_IN_SYSTEM(p) {
4795 if (p->p_state == PRS_NEW)
4800 for (n = 0; n < fdp->fd_nfiles; n++) {
4801 if (fp == fdp->fd_ofiles[n].fde_file)
4809 db_print_file(struct file *fp, int header)
4811 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4815 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4816 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4817 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4819 p = file_to_first_proc(fp);
4820 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4821 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4822 fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4823 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4828 DB_SHOW_COMMAND(file, db_show_file)
4833 db_printf("usage: show file <addr>\n");
4836 fp = (struct file *)addr;
4837 db_print_file(fp, 1);
4840 DB_SHOW_COMMAND(files, db_show_files)
4842 struct filedesc *fdp;
4849 FOREACH_PROC_IN_SYSTEM(p) {
4850 if (p->p_state == PRS_NEW)
4852 if ((fdp = p->p_fd) == NULL)
4854 for (n = 0; n < fdp->fd_nfiles; ++n) {
4855 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4857 db_print_file(fp, header);
4864 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4865 &maxfilesperproc, 0, "Maximum files allowed open per process");
4867 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4868 &maxfiles, 0, "Maximum number of files");
4870 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4871 &openfiles, 0, "System-wide number of open files");
4875 filelistinit(void *dummy)
4878 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4879 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4880 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4881 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4882 pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4883 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4885 * XXXMJG this is a temporary hack due to boot ordering issues against
4888 vfs_smr = uma_zone_get_smr(pwd_zone);
4889 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4891 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4893 /*-------------------------------------------------------------------*/
4896 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4897 int flags, struct thread *td)
4904 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4912 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4920 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4928 badfo_kqfilter(struct file *fp, struct knote *kn)
4935 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
4942 badfo_close(struct file *fp, struct thread *td)
4949 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4957 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4965 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4966 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4974 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4980 struct fileops badfileops = {
4981 .fo_read = badfo_readwrite,
4982 .fo_write = badfo_readwrite,
4983 .fo_truncate = badfo_truncate,
4984 .fo_ioctl = badfo_ioctl,
4985 .fo_poll = badfo_poll,
4986 .fo_kqfilter = badfo_kqfilter,
4987 .fo_stat = badfo_stat,
4988 .fo_close = badfo_close,
4989 .fo_chmod = badfo_chmod,
4990 .fo_chown = badfo_chown,
4991 .fo_sendfile = badfo_sendfile,
4992 .fo_fill_kinfo = badfo_fill_kinfo,
4996 path_poll(struct file *fp, int events, struct ucred *active_cred,
5003 path_close(struct file *fp, struct thread *td)
5005 MPASS(fp->f_type == DTYPE_VNODE);
5006 fp->f_ops = &badfileops;
5011 struct fileops path_fileops = {
5012 .fo_read = badfo_readwrite,
5013 .fo_write = badfo_readwrite,
5014 .fo_truncate = badfo_truncate,
5015 .fo_ioctl = badfo_ioctl,
5016 .fo_poll = path_poll,
5017 .fo_kqfilter = vn_kqfilter_opath,
5018 .fo_stat = vn_statfile,
5019 .fo_close = path_close,
5020 .fo_chmod = badfo_chmod,
5021 .fo_chown = badfo_chown,
5022 .fo_sendfile = badfo_sendfile,
5023 .fo_fill_kinfo = vn_fill_kinfo,
5024 .fo_flags = DFLAG_PASSABLE,
5028 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
5029 int flags, struct thread *td)
5032 return (EOPNOTSUPP);
5036 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
5044 invfo_ioctl(struct file *fp, u_long com, void *data,
5045 struct ucred *active_cred, struct thread *td)
5052 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
5056 return (poll_no_poll(events));
5060 invfo_kqfilter(struct file *fp, struct knote *kn)
5067 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5075 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5083 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5084 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5091 /*-------------------------------------------------------------------*/
5094 * File Descriptor pseudo-device driver (/dev/fd/).
5096 * Opening minor device N dup()s the file (if any) connected to file
5097 * descriptor N belonging to the calling process. Note that this driver
5098 * consists of only the ``open()'' routine, because all subsequent
5099 * references to this file will be direct to the other driver.
5101 * XXX: we could give this one a cloning event handler if necessary.
5106 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
5110 * XXX Kludge: set curthread->td_dupfd to contain the value of the
5111 * the file descriptor being sought for duplication. The error
5112 * return ensures that the vnode for this device will be released
5113 * by vn_open. Open will detect this special error and take the
5114 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
5115 * will simply report the error.
5117 td->td_dupfd = dev2unit(dev);
5121 static struct cdevsw fildesc_cdevsw = {
5122 .d_version = D_VERSION,
5128 fildesc_drvinit(void *unused)
5132 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
5133 UID_ROOT, GID_WHEEL, 0666, "fd/0");
5134 make_dev_alias(dev, "stdin");
5135 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
5136 UID_ROOT, GID_WHEEL, 0666, "fd/1");
5137 make_dev_alias(dev, "stdout");
5138 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
5139 UID_ROOT, GID_WHEEL, 0666, "fd/2");
5140 make_dev_alias(dev, "stderr");
5143 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);