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)
158 #define FILEDESC_FOREACH_FDE(fdp, _iterator, _fde) \
159 struct filedesc *_fdp = (fdp); \
160 int _lastfile = fdlastfile_single(_fdp); \
161 for (_iterator = 0; _iterator <= _lastfile; _iterator++) \
162 if ((_fde = &_fdp->fd_ofiles[_iterator])->fde_file != NULL)
164 #define FILEDESC_FOREACH_FP(fdp, _iterator, _fp) \
165 struct filedesc *_fdp = (fdp); \
166 int _lastfile = fdlastfile_single(_fdp); \
167 for (_iterator = 0; _iterator <= _lastfile; _iterator++) \
168 if ((_fp = _fdp->fd_ofiles[_iterator].fde_file) != NULL)
171 * SLIST entry used to keep track of ofiles which must be reclaimed when
175 struct fdescenttbl *ft_table;
176 SLIST_ENTRY(freetable) ft_next;
180 * Initial allocation: a filedesc structure + the head of SLIST used to
181 * keep track of old ofiles + enough space for NDFILE descriptors.
184 struct fdescenttbl0 {
186 struct filedescent fdt_ofiles[NDFILE];
190 struct filedesc fd_fd;
191 SLIST_HEAD(, freetable) fd_free;
192 struct fdescenttbl0 fd_dfiles;
193 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
197 * Descriptor management.
199 static int __exclusive_cache_line openfiles; /* actual number of open files */
200 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
201 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
204 * If low >= size, just return low. Otherwise find the first zero bit in the
205 * given bitmap, starting at low and not exceeding size - 1. Return size if
209 fd_first_free(struct filedesc *fdp, int low, int size)
211 NDSLOTTYPE *map = fdp->fd_map;
219 if (low % NDENTRIES) {
220 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
221 if ((mask &= ~map[off]) != 0UL)
222 return (off * NDENTRIES + ffsl(mask) - 1);
225 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
226 if (map[off] != ~0UL)
227 return (off * NDENTRIES + ffsl(~map[off]) - 1);
232 * Find the last used fd.
234 * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
235 * Otherwise use fdlastfile.
238 fdlastfile_single(struct filedesc *fdp)
240 NDSLOTTYPE *map = fdp->fd_map;
243 off = NDSLOT(fdp->fd_nfiles - 1);
244 for (minoff = NDSLOT(0); off >= minoff; --off)
246 return (off * NDENTRIES + flsl(map[off]) - 1);
251 fdlastfile(struct filedesc *fdp)
254 FILEDESC_LOCK_ASSERT(fdp);
255 return (fdlastfile_single(fdp));
259 fdisused(struct filedesc *fdp, int fd)
262 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
263 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
265 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
269 * Mark a file descriptor as used.
272 fdused_init(struct filedesc *fdp, int fd)
275 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
277 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
281 fdused(struct filedesc *fdp, int fd)
284 FILEDESC_XLOCK_ASSERT(fdp);
286 fdused_init(fdp, fd);
287 if (fd == fdp->fd_freefile)
292 * Mark a file descriptor as unused.
295 fdunused(struct filedesc *fdp, int fd)
298 FILEDESC_XLOCK_ASSERT(fdp);
300 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
301 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
302 ("fd=%d is still in use", fd));
304 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
305 if (fd < fdp->fd_freefile)
306 fdp->fd_freefile = fd;
310 * Free a file descriptor.
312 * Avoid some work if fdp is about to be destroyed.
315 fdefree_last(struct filedescent *fde)
318 filecaps_free(&fde->fde_caps);
322 fdfree(struct filedesc *fdp, int fd)
324 struct filedescent *fde;
326 FILEDESC_XLOCK_ASSERT(fdp);
327 fde = &fdp->fd_ofiles[fd];
329 seqc_write_begin(&fde->fde_seqc);
331 fde->fde_file = NULL;
333 seqc_write_end(&fde->fde_seqc);
340 * System calls on descriptors.
342 #ifndef _SYS_SYSPROTO_H_
343 struct getdtablesize_args {
349 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
355 td->td_retval[0] = getmaxfd(td);
357 PROC_LOCK(td->td_proc);
358 lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
359 PROC_UNLOCK(td->td_proc);
360 if (lim < td->td_retval[0])
361 td->td_retval[0] = lim;
367 * Duplicate a file descriptor to a particular value.
369 * Note: keep in mind that a potential race condition exists when closing
370 * descriptors from a shared descriptor table (via rfork).
372 #ifndef _SYS_SYSPROTO_H_
380 sys_dup2(struct thread *td, struct dup2_args *uap)
383 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
387 * Duplicate a file descriptor.
389 #ifndef _SYS_SYSPROTO_H_
396 sys_dup(struct thread *td, struct dup_args *uap)
399 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
403 * The file control system call.
405 #ifndef _SYS_SYSPROTO_H_
414 sys_fcntl(struct thread *td, struct fcntl_args *uap)
417 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
421 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
435 * Convert old flock structure to new.
437 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
438 fl.l_start = ofl.l_start;
439 fl.l_len = ofl.l_len;
440 fl.l_pid = ofl.l_pid;
441 fl.l_type = ofl.l_type;
442 fl.l_whence = ofl.l_whence;
456 arg1 = (intptr_t)&fl;
462 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
463 arg1 = (intptr_t)&fl;
471 error = kern_fcntl(td, fd, newcmd, arg1);
474 if (cmd == F_OGETLK) {
475 ofl.l_start = fl.l_start;
476 ofl.l_len = fl.l_len;
477 ofl.l_pid = fl.l_pid;
478 ofl.l_type = fl.l_type;
479 ofl.l_whence = fl.l_whence;
480 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
481 } else if (cmd == F_GETLK) {
482 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
488 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
490 struct filedesc *fdp;
492 struct file *fp, *fp2;
493 struct filedescent *fde;
497 struct kinfo_file *kif;
498 int error, flg, kif_sz, seals, tmp, got_set, got_cleared;
512 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
515 case F_DUPFD_CLOEXEC:
517 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
522 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
525 case F_DUP2FD_CLOEXEC:
527 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
533 fde = fdeget_noref(fdp, fd);
536 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
539 FILEDESC_SUNLOCK(fdp);
545 fde = fdeget_noref(fdp, fd);
547 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
548 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
551 FILEDESC_XUNLOCK(fdp);
555 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
558 td->td_retval[0] = OFLAGS(fp->f_flag);
563 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
566 if (fp->f_ops == &path_fileops) {
572 tmp = flg = fp->f_flag;
574 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
575 } while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
576 got_set = tmp & ~flg;
577 got_cleared = flg & ~tmp;
578 tmp = fp->f_flag & FNONBLOCK;
579 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
582 tmp = fp->f_flag & FASYNC;
583 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
588 atomic_clear_int(&fp->f_flag, FNONBLOCK);
590 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
593 tmp = flg = fp->f_flag;
597 } while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
602 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
605 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
607 td->td_retval[0] = tmp;
612 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
616 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
621 error = priv_check(td, PRIV_NFS_LOCKD);
629 /* FALLTHROUGH F_SETLK */
633 flp = (struct flock *)arg;
634 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
639 error = fget_unlocked(td, fd, &cap_flock_rights, &fp);
642 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
648 if (flp->l_whence == SEEK_CUR) {
649 foffset = foffset_get(fp);
652 foffset > OFF_MAX - flp->l_start)) {
657 flp->l_start += foffset;
661 switch (flp->l_type) {
663 if ((fp->f_flag & FREAD) == 0) {
667 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
668 PROC_LOCK(p->p_leader);
669 p->p_leader->p_flag |= P_ADVLOCK;
670 PROC_UNLOCK(p->p_leader);
672 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
676 if ((fp->f_flag & FWRITE) == 0) {
680 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
681 PROC_LOCK(p->p_leader);
682 p->p_leader->p_flag |= P_ADVLOCK;
683 PROC_UNLOCK(p->p_leader);
685 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
689 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
693 if (flg != F_REMOTE) {
697 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
698 F_UNLCKSYS, flp, flg);
704 if (error != 0 || flp->l_type == F_UNLCK ||
705 flp->l_type == F_UNLCKSYS) {
711 * Check for a race with close.
713 * The vnode is now advisory locked (or unlocked, but this case
714 * is not really important) as the caller requested.
715 * We had to drop the filedesc lock, so we need to recheck if
716 * the descriptor is still valid, because if it was closed
717 * in the meantime we need to remove advisory lock from the
718 * vnode - close on any descriptor leading to an advisory
719 * locked vnode, removes that lock.
720 * We will return 0 on purpose in that case, as the result of
721 * successful advisory lock might have been externally visible
722 * already. This is fine - effectively we pretend to the caller
723 * that the closing thread was a bit slower and that the
724 * advisory lock succeeded before the close.
726 error = fget_unlocked(td, fd, &cap_no_rights, &fp2);
732 flp->l_whence = SEEK_SET;
735 flp->l_type = F_UNLCK;
736 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
737 F_UNLCK, flp, F_POSIX);
744 error = fget_unlocked(td, fd, &cap_flock_rights, &fp);
747 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
752 flp = (struct flock *)arg;
753 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
754 flp->l_type != F_UNLCK) {
759 if (flp->l_whence == SEEK_CUR) {
760 foffset = foffset_get(fp);
761 if ((flp->l_start > 0 &&
762 foffset > OFF_MAX - flp->l_start) ||
764 foffset < OFF_MIN - flp->l_start)) {
769 flp->l_start += foffset;
772 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
778 error = fget_unlocked(td, fd, &cap_no_rights, &fp);
781 error = fo_add_seals(fp, arg);
786 error = fget_unlocked(td, fd, &cap_no_rights, &fp);
789 if (fo_get_seals(fp, &seals) == 0)
790 td->td_retval[0] = seals;
797 arg = arg ? 128 * 1024: 0;
800 error = fget_unlocked(td, fd, &cap_no_rights, &fp);
803 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
809 if (vp->v_type != VREG) {
816 * Exclusive lock synchronizes against f_seqcount reads and
817 * writes in sequential_heuristic().
819 error = vn_lock(vp, LK_EXCLUSIVE);
825 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
826 arg = MIN(arg, INT_MAX - bsize + 1);
827 fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
828 (arg + bsize - 1) / bsize);
829 atomic_set_int(&fp->f_flag, FRDAHEAD);
831 atomic_clear_int(&fp->f_flag, FRDAHEAD);
839 * Check if the vnode is part of a union stack (either the
840 * "union" flag from mount(2) or unionfs).
842 * Prior to introduction of this op libc's readdir would call
843 * fstatfs(2), in effect unnecessarily copying kilobytes of
844 * data just to check fs name and a mount flag.
846 * Fixing the code to handle everything in the kernel instead
847 * is a non-trivial endeavor and has low priority, thus this
848 * horrible kludge facilitates the current behavior in a much
849 * cheaper manner until someone(tm) sorts this out.
851 error = fget_unlocked(td, fd, &cap_no_rights, &fp);
854 if (fp->f_type != DTYPE_VNODE) {
861 * Since we don't prevent dooming the vnode even non-null mp
862 * found can become immediately stale. This is tolerable since
863 * mount points are type-stable (providing safe memory access)
864 * and any vfs op on this vnode going forward will return an
865 * error (meaning return value in this case is meaningless).
867 mp = atomic_load_ptr(&vp->v_mount);
868 if (__predict_false(mp == NULL)) {
873 td->td_retval[0] = 0;
874 if (mp->mnt_kern_flag & MNTK_UNIONFS ||
875 mp->mnt_flag & MNT_UNION)
876 td->td_retval[0] = 1;
881 #ifdef CAPABILITY_MODE
882 if (IN_CAPABILITY_MODE(td)) {
887 error = copyin((void *)arg, &kif_sz, sizeof(kif_sz));
890 if (kif_sz != sizeof(*kif)) {
894 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK | M_ZERO);
896 error = fget_cap_noref(fdp, fd, &cap_fcntl_rights, &fp, NULL);
897 if (error == 0 && fhold(fp)) {
898 export_file_to_kinfo(fp, fd, NULL, kif, fdp, 0);
899 FILEDESC_SUNLOCK(fdp);
901 if ((kif->kf_status & KF_ATTR_VALID) != 0) {
902 kif->kf_structsize = sizeof(*kif);
903 error = copyout(kif, (void *)arg, sizeof(*kif));
908 FILEDESC_SUNLOCK(fdp);
923 getmaxfd(struct thread *td)
926 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
930 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
933 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
935 struct filedesc *fdp;
936 struct filedescent *oldfde, *newfde;
938 struct file *delfp, *oldfp;
939 u_long *oioctls, *nioctls;
946 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
947 MPASS(mode < FDDUP_LASTMODE);
950 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
953 * Verify we have a valid descriptor to dup from and possibly to
954 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
955 * return EINVAL when the new descriptor is out of bounds.
960 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
961 maxfd = getmaxfd(td);
963 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
967 if (fget_noref(fdp, old) == NULL)
969 if (mode == FDDUP_FIXED && old == new) {
970 td->td_retval[0] = new;
971 if (flags & FDDUP_FLAG_CLOEXEC)
972 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
977 oldfde = &fdp->fd_ofiles[old];
978 oldfp = oldfde->fde_file;
983 * If the caller specified a file descriptor, make sure the file
984 * table is large enough to hold it, and grab it. Otherwise, just
985 * allocate a new descriptor the usual way.
990 if ((error = fdalloc(td, new, &new)) != 0) {
996 if (new >= fdp->fd_nfiles) {
998 * The resource limits are here instead of e.g.
999 * fdalloc(), because the file descriptor table may be
1000 * shared between processes, so we can't really use
1001 * racct_add()/racct_sub(). Instead of counting the
1002 * number of actually allocated descriptors, just put
1003 * the limit on the size of the file descriptor table.
1006 if (RACCT_ENABLED()) {
1007 error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
1015 fdgrowtable_exp(fdp, new + 1);
1017 if (!fdisused(fdp, new))
1021 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
1024 KASSERT(old != new, ("new fd is same as old"));
1026 /* Refetch oldfde because the table may have grown and old one freed. */
1027 oldfde = &fdp->fd_ofiles[old];
1028 KASSERT(oldfp == oldfde->fde_file,
1029 ("fdt_ofiles shift from growth observed at fd %d",
1032 newfde = &fdp->fd_ofiles[new];
1033 delfp = newfde->fde_file;
1035 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
1038 * Duplicate the source descriptor.
1041 seqc_write_begin(&newfde->fde_seqc);
1043 oioctls = filecaps_free_prep(&newfde->fde_caps);
1044 fde_copy(oldfde, newfde);
1045 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
1047 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
1048 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
1050 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
1052 seqc_write_end(&newfde->fde_seqc);
1054 td->td_retval[0] = new;
1058 if (delfp != NULL) {
1059 (void) closefp(fdp, new, delfp, td, true, false);
1060 FILEDESC_UNLOCK_ASSERT(fdp);
1063 FILEDESC_XUNLOCK(fdp);
1066 filecaps_free_finish(oioctls);
1071 sigiofree(struct sigio *sigio)
1073 crfree(sigio->sio_ucred);
1074 free(sigio, M_SIGIO);
1077 static struct sigio *
1078 funsetown_locked(struct sigio *sigio)
1083 SIGIO_ASSERT_LOCKED();
1087 *sigio->sio_myref = NULL;
1088 if (sigio->sio_pgid < 0) {
1089 pg = sigio->sio_pgrp;
1091 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio);
1094 p = sigio->sio_proc;
1096 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio);
1103 * If sigio is on the list associated with a process or process group,
1104 * disable signalling from the device, remove sigio from the list and
1108 funsetown(struct sigio **sigiop)
1110 struct sigio *sigio;
1112 /* Racy check, consumers must provide synchronization. */
1113 if (*sigiop == NULL)
1117 sigio = funsetown_locked(*sigiop);
1124 * Free a list of sigio structures. The caller must ensure that new sigio
1125 * structures cannot be added after this point. For process groups this is
1126 * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1130 funsetownlst(struct sigiolst *sigiolst)
1134 struct sigio *sigio, *tmp;
1137 sigio = SLIST_FIRST(sigiolst);
1145 sigio = SLIST_FIRST(sigiolst);
1146 if (sigio == NULL) {
1152 * Every entry of the list should belong to a single proc or pgrp.
1154 if (sigio->sio_pgid < 0) {
1155 pg = sigio->sio_pgrp;
1156 sx_assert(&proctree_lock, SX_XLOCKED);
1158 } else /* if (sigio->sio_pgid > 0) */ {
1159 p = sigio->sio_proc;
1161 KASSERT((p->p_flag & P_WEXIT) != 0,
1162 ("%s: process %p is not exiting", __func__, p));
1165 SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1166 *sigio->sio_myref = NULL;
1168 KASSERT(sigio->sio_pgid < 0,
1169 ("Proc sigio in pgrp sigio list"));
1170 KASSERT(sigio->sio_pgrp == pg,
1171 ("Bogus pgrp in sigio list"));
1172 } else /* if (p != NULL) */ {
1173 KASSERT(sigio->sio_pgid > 0,
1174 ("Pgrp sigio in proc sigio list"));
1175 KASSERT(sigio->sio_proc == p,
1176 ("Bogus proc in sigio list"));
1186 SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1191 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1193 * After permission checking, add a sigio structure to the sigio list for
1194 * the process or process group.
1197 fsetown(pid_t pgid, struct sigio **sigiop)
1201 struct sigio *osigio, *sigio;
1209 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1210 sigio->sio_pgid = pgid;
1211 sigio->sio_ucred = crhold(curthread->td_ucred);
1212 sigio->sio_myref = sigiop;
1216 ret = pget(pgid, PGET_NOTWEXIT | PGET_NOTID | PGET_HOLD, &proc);
1218 osigio = funsetown_locked(*sigiop);
1222 if ((proc->p_flag & P_WEXIT) != 0) {
1224 } else if (proc->p_session !=
1225 curthread->td_proc->p_session) {
1227 * Policy - Don't allow a process to FSETOWN a
1228 * process in another session.
1230 * Remove this test to allow maximum flexibility
1231 * or restrict FSETOWN to the current process or
1232 * process group for maximum safety.
1236 sigio->sio_proc = proc;
1237 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio,
1242 } else /* if (pgid < 0) */ {
1243 sx_slock(&proctree_lock);
1245 osigio = funsetown_locked(*sigiop);
1246 pgrp = pgfind(-pgid);
1250 if (pgrp->pg_session != curthread->td_proc->p_session) {
1252 * Policy - Don't allow a process to FSETOWN a
1253 * process in another session.
1255 * Remove this test to allow maximum flexibility
1256 * or restrict FSETOWN to the current process or
1257 * process group for maximum safety.
1261 sigio->sio_pgrp = pgrp;
1262 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio,
1267 sx_sunlock(&proctree_lock);
1278 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1281 fgetown(struct sigio **sigiop)
1286 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1292 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1297 FILEDESC_XLOCK_ASSERT(fdp);
1300 * We now hold the fp reference that used to be owned by the
1301 * descriptor array. We have to unlock the FILEDESC *AFTER*
1302 * knote_fdclose to prevent a race of the fd getting opened, a knote
1303 * added, and deleteing a knote for the new fd.
1305 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1306 knote_fdclose(td, fd);
1309 * We need to notify mqueue if the object is of type mqueue.
1311 if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1312 mq_fdclose(td, fd, fp);
1313 FILEDESC_XUNLOCK(fdp);
1316 if (AUDITING_TD(td) && audit)
1317 audit_sysclose(td, fd, fp);
1319 error = closef(fp, td);
1322 * All paths leading up to closefp() will have already removed or
1323 * replaced the fd in the filedesc table, so a restart would not
1324 * operate on the same file.
1326 if (error == ERESTART)
1333 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1334 bool holdleaders, bool audit)
1338 FILEDESC_XLOCK_ASSERT(fdp);
1341 if (td->td_proc->p_fdtol != NULL) {
1343 * Ask fdfree() to sleep to ensure that all relevant
1344 * process leaders can be traversed in closef().
1346 fdp->fd_holdleaderscount++;
1348 holdleaders = false;
1352 error = closefp_impl(fdp, fd, fp, td, audit);
1354 FILEDESC_XLOCK(fdp);
1355 fdp->fd_holdleaderscount--;
1356 if (fdp->fd_holdleaderscount == 0 &&
1357 fdp->fd_holdleaderswakeup != 0) {
1358 fdp->fd_holdleaderswakeup = 0;
1359 wakeup(&fdp->fd_holdleaderscount);
1361 FILEDESC_XUNLOCK(fdp);
1367 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1368 bool holdleaders, bool audit)
1371 FILEDESC_XLOCK_ASSERT(fdp);
1373 if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1374 return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1376 return (closefp_impl(fdp, fd, fp, td, audit));
1381 * Close a file descriptor.
1383 #ifndef _SYS_SYSPROTO_H_
1390 sys_close(struct thread *td, struct close_args *uap)
1393 return (kern_close(td, uap->fd));
1397 kern_close(struct thread *td, int fd)
1399 struct filedesc *fdp;
1402 fdp = td->td_proc->p_fd;
1404 FILEDESC_XLOCK(fdp);
1405 if ((fp = fget_noref(fdp, fd)) == NULL) {
1406 FILEDESC_XUNLOCK(fdp);
1411 /* closefp() drops the FILEDESC lock for us. */
1412 return (closefp(fdp, fd, fp, td, true, true));
1416 close_range_cloexec(struct thread *td, u_int lowfd, u_int highfd)
1418 struct filedesc *fdp;
1419 struct fdescenttbl *fdt;
1420 struct filedescent *fde;
1423 fdp = td->td_proc->p_fd;
1424 FILEDESC_XLOCK(fdp);
1425 fdt = atomic_load_ptr(&fdp->fd_files);
1426 highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1428 if (__predict_false(fd > highfd)) {
1431 for (; fd <= highfd; fd++) {
1432 fde = &fdt->fdt_ofiles[fd];
1433 if (fde->fde_file != NULL)
1434 fde->fde_flags |= UF_EXCLOSE;
1437 FILEDESC_XUNLOCK(fdp);
1442 close_range_impl(struct thread *td, u_int lowfd, u_int highfd)
1444 struct filedesc *fdp;
1445 const struct fdescenttbl *fdt;
1449 fdp = td->td_proc->p_fd;
1450 FILEDESC_XLOCK(fdp);
1451 fdt = atomic_load_ptr(&fdp->fd_files);
1452 highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1454 if (__predict_false(fd > highfd)) {
1458 fp = fdt->fdt_ofiles[fd].fde_file;
1464 (void) closefp(fdp, fd, fp, td, true, true);
1467 FILEDESC_XLOCK(fdp);
1468 fdt = atomic_load_ptr(&fdp->fd_files);
1473 FILEDESC_XUNLOCK(fdp);
1479 kern_close_range(struct thread *td, int flags, u_int lowfd, u_int highfd)
1483 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1484 * open should not be a usage error. From a close_range() perspective,
1485 * close_range(3, ~0U, 0) in the same scenario should also likely not
1486 * be a usage error as all fd above 3 are in-fact already closed.
1488 if (highfd < lowfd) {
1492 if ((flags & CLOSE_RANGE_CLOEXEC) != 0)
1493 return (close_range_cloexec(td, lowfd, highfd));
1495 return (close_range_impl(td, lowfd, highfd));
1498 #ifndef _SYS_SYSPROTO_H_
1499 struct close_range_args {
1506 sys_close_range(struct thread *td, struct close_range_args *uap)
1509 AUDIT_ARG_FD(uap->lowfd);
1510 AUDIT_ARG_CMD(uap->highfd);
1511 AUDIT_ARG_FFLAGS(uap->flags);
1513 if ((uap->flags & ~(CLOSE_RANGE_CLOEXEC)) != 0)
1515 return (kern_close_range(td, uap->flags, uap->lowfd, uap->highfd));
1518 #ifdef COMPAT_FREEBSD12
1520 * Close open file descriptors.
1522 #ifndef _SYS_SYSPROTO_H_
1523 struct freebsd12_closefrom_args {
1529 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1533 AUDIT_ARG_FD(uap->lowfd);
1536 * Treat negative starting file descriptor values identical to
1537 * closefrom(0) which closes all files.
1539 lowfd = MAX(0, uap->lowfd);
1540 return (kern_close_range(td, 0, lowfd, ~0U));
1542 #endif /* COMPAT_FREEBSD12 */
1544 #if defined(COMPAT_43)
1546 * Return status information about a file descriptor.
1548 #ifndef _SYS_SYSPROTO_H_
1549 struct ofstat_args {
1556 ofstat(struct thread *td, struct ofstat_args *uap)
1562 error = kern_fstat(td, uap->fd, &ub);
1565 error = copyout(&oub, uap->sb, sizeof(oub));
1569 #endif /* COMPAT_43 */
1571 #if defined(COMPAT_FREEBSD11)
1573 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1576 struct freebsd11_stat osb;
1579 error = kern_fstat(td, uap->fd, &sb);
1582 error = freebsd11_cvtstat(&sb, &osb);
1584 error = copyout(&osb, uap->sb, sizeof(osb));
1587 #endif /* COMPAT_FREEBSD11 */
1590 * Return status information about a file descriptor.
1592 #ifndef _SYS_SYSPROTO_H_
1600 sys_fstat(struct thread *td, struct fstat_args *uap)
1605 error = kern_fstat(td, uap->fd, &ub);
1607 error = copyout(&ub, uap->sb, sizeof(ub));
1612 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1619 error = fget(td, fd, &cap_fstat_rights, &fp);
1620 if (__predict_false(error != 0))
1623 AUDIT_ARG_FILE(td->td_proc, fp);
1625 error = fo_stat(fp, sbp, td->td_ucred);
1627 #ifdef __STAT_TIME_T_EXT
1628 sbp->st_atim_ext = 0;
1629 sbp->st_mtim_ext = 0;
1630 sbp->st_ctim_ext = 0;
1631 sbp->st_btim_ext = 0;
1634 if (KTRPOINT(td, KTR_STRUCT))
1635 ktrstat_error(sbp, error);
1640 #if defined(COMPAT_FREEBSD11)
1642 * Return status information about a file descriptor.
1644 #ifndef _SYS_SYSPROTO_H_
1645 struct freebsd11_nfstat_args {
1652 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1658 error = kern_fstat(td, uap->fd, &ub);
1661 error = freebsd11_cvtnstat(&ub, &nub);
1663 error = copyout(&nub, uap->sb, sizeof(nub));
1666 #endif /* COMPAT_FREEBSD11 */
1669 * Return pathconf information about a file descriptor.
1671 #ifndef _SYS_SYSPROTO_H_
1672 struct fpathconf_args {
1679 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1684 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1686 td->td_retval[0] = value;
1691 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1697 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1701 if (name == _PC_ASYNC_IO) {
1702 *valuep = _POSIX_ASYNCHRONOUS_IO;
1707 vn_lock(vp, LK_SHARED | LK_RETRY);
1708 error = VOP_PATHCONF(vp, name, valuep);
1710 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1711 if (name != _PC_PIPE_BUF) {
1726 * Copy filecaps structure allocating memory for ioctls array if needed.
1728 * The last parameter indicates whether the fdtable is locked. If it is not and
1729 * ioctls are encountered, copying fails and the caller must lock the table.
1731 * Note that if the table was not locked, the caller has to check the relevant
1732 * sequence counter to determine whether the operation was successful.
1735 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1739 if (src->fc_ioctls != NULL && !locked)
1741 memcpy(dst, src, sizeof(*src));
1742 if (src->fc_ioctls == NULL)
1745 KASSERT(src->fc_nioctls > 0,
1746 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1748 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1749 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1750 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1755 filecaps_copy_prep(const struct filecaps *src)
1760 if (__predict_true(src->fc_ioctls == NULL))
1763 KASSERT(src->fc_nioctls > 0,
1764 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1766 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1767 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1772 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1778 if (__predict_true(src->fc_ioctls == NULL)) {
1779 MPASS(ioctls == NULL);
1783 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1784 dst->fc_ioctls = ioctls;
1785 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1789 * Move filecaps structure to the new place and clear the old place.
1792 filecaps_move(struct filecaps *src, struct filecaps *dst)
1796 bzero(src, sizeof(*src));
1800 * Fill the given filecaps structure with full rights.
1803 filecaps_fill(struct filecaps *fcaps)
1806 CAP_ALL(&fcaps->fc_rights);
1807 fcaps->fc_ioctls = NULL;
1808 fcaps->fc_nioctls = -1;
1809 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1813 * Free memory allocated within filecaps structure.
1816 filecaps_free_ioctl(struct filecaps *fcaps)
1819 free(fcaps->fc_ioctls, M_FILECAPS);
1820 fcaps->fc_ioctls = NULL;
1824 filecaps_free(struct filecaps *fcaps)
1827 filecaps_free_ioctl(fcaps);
1828 bzero(fcaps, sizeof(*fcaps));
1832 filecaps_free_prep(struct filecaps *fcaps)
1836 ioctls = fcaps->fc_ioctls;
1837 bzero(fcaps, sizeof(*fcaps));
1842 filecaps_free_finish(u_long *ioctls)
1845 free(ioctls, M_FILECAPS);
1849 * Validate the given filecaps structure.
1852 filecaps_validate(const struct filecaps *fcaps, const char *func)
1855 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1856 ("%s: invalid rights", func));
1857 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1858 ("%s: invalid fcntls", func));
1859 KASSERT(fcaps->fc_fcntls == 0 ||
1860 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1861 ("%s: fcntls without CAP_FCNTL", func));
1863 * open calls without WANTIOCTLCAPS free caps but leave the counter
1866 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1867 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1868 ("%s: invalid ioctls", func));
1870 KASSERT(fcaps->fc_nioctls == 0 ||
1871 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1872 ("%s: ioctls without CAP_IOCTL", func));
1876 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1880 FILEDESC_XLOCK_ASSERT(fdp);
1882 nfd1 = fdp->fd_nfiles * 2;
1885 fdgrowtable(fdp, nfd1);
1889 * Grow the file table to accommodate (at least) nfd descriptors.
1892 fdgrowtable(struct filedesc *fdp, int nfd)
1894 struct filedesc0 *fdp0;
1895 struct freetable *ft;
1896 struct fdescenttbl *ntable;
1897 struct fdescenttbl *otable;
1898 int nnfiles, onfiles;
1899 NDSLOTTYPE *nmap, *omap;
1901 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1903 /* save old values */
1904 onfiles = fdp->fd_nfiles;
1905 otable = fdp->fd_files;
1908 /* compute the size of the new table */
1909 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1910 if (nnfiles <= onfiles)
1911 /* the table is already large enough */
1915 * Allocate a new table. We need enough space for the number of
1916 * entries, file entries themselves and the struct freetable we will use
1917 * when we decommission the table and place it on the freelist.
1918 * We place the struct freetable in the middle so we don't have
1919 * to worry about padding.
1921 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1922 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1923 sizeof(struct freetable),
1924 M_FILEDESC, M_ZERO | M_WAITOK);
1925 /* copy the old data */
1926 ntable->fdt_nfiles = nnfiles;
1927 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1928 onfiles * sizeof(ntable->fdt_ofiles[0]));
1931 * Allocate a new map only if the old is not large enough. It will
1932 * grow at a slower rate than the table as it can map more
1933 * entries than the table can hold.
1935 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1936 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1938 /* copy over the old data and update the pointer */
1939 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1944 * Make sure that ntable is correctly initialized before we replace
1945 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1948 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1951 * Free the old file table when not shared by other threads or processes.
1952 * The old file table is considered to be shared when either are true:
1953 * - The process has more than one thread.
1954 * - The file descriptor table has been shared via fdshare().
1956 * When shared, the old file table will be placed on a freelist
1957 * which will be processed when the struct filedesc is released.
1959 * Note that if onfiles == NDFILE, we're dealing with the original
1960 * static allocation contained within (struct filedesc0 *)fdp,
1961 * which must not be freed.
1963 if (onfiles > NDFILE) {
1965 * Note we may be called here from fdinit while allocating a
1966 * table for a new process in which case ->p_fd points
1969 if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) {
1970 free(otable, M_FILEDESC);
1972 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1973 fdp0 = (struct filedesc0 *)fdp;
1974 ft->ft_table = otable;
1975 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1979 * The map does not have the same possibility of threads still
1980 * holding references to it. So always free it as long as it
1981 * does not reference the original static allocation.
1983 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1984 free(omap, M_FILEDESC);
1988 * Allocate a file descriptor for the process.
1991 fdalloc(struct thread *td, int minfd, int *result)
1993 struct proc *p = td->td_proc;
1994 struct filedesc *fdp = p->p_fd;
1995 int fd, maxfd, allocfd;
2000 FILEDESC_XLOCK_ASSERT(fdp);
2002 if (fdp->fd_freefile > minfd)
2003 minfd = fdp->fd_freefile;
2005 maxfd = getmaxfd(td);
2008 * Search the bitmap for a free descriptor starting at minfd.
2009 * If none is found, grow the file table.
2011 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
2012 if (__predict_false(fd >= maxfd))
2014 if (__predict_false(fd >= fdp->fd_nfiles)) {
2015 allocfd = min(fd * 2, maxfd);
2017 if (RACCT_ENABLED()) {
2018 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
2024 * fd is already equal to first free descriptor >= minfd, so
2025 * we only need to grow the table and we are done.
2027 fdgrowtable_exp(fdp, allocfd);
2031 * Perform some sanity checks, then mark the file descriptor as
2032 * used and return it to the caller.
2034 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
2035 ("invalid descriptor %d", fd));
2036 KASSERT(!fdisused(fdp, fd),
2037 ("fd_first_free() returned non-free descriptor"));
2038 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
2039 ("file descriptor isn't free"));
2046 * Allocate n file descriptors for the process.
2049 fdallocn(struct thread *td, int minfd, int *fds, int n)
2051 struct proc *p = td->td_proc;
2052 struct filedesc *fdp = p->p_fd;
2055 FILEDESC_XLOCK_ASSERT(fdp);
2057 for (i = 0; i < n; i++)
2058 if (fdalloc(td, 0, &fds[i]) != 0)
2062 for (i--; i >= 0; i--)
2063 fdunused(fdp, fds[i]);
2071 * Create a new open file structure and allocate a file descriptor for the
2072 * process that refers to it. We add one reference to the file for the
2073 * descriptor table and one reference for resultfp. This is to prevent us
2074 * being preempted and the entry in the descriptor table closed after we
2075 * release the FILEDESC lock.
2078 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
2079 struct filecaps *fcaps)
2084 MPASS(resultfp != NULL);
2085 MPASS(resultfd != NULL);
2087 error = _falloc_noinstall(td, &fp, 2);
2088 if (__predict_false(error != 0)) {
2092 error = finstall_refed(td, fp, &fd, flags, fcaps);
2093 if (__predict_false(error != 0)) {
2094 falloc_abort(td, fp);
2105 * Create a new open file structure without allocating a file descriptor.
2108 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2111 int maxuserfiles = maxfiles - (maxfiles / 20);
2113 static struct timeval lastfail;
2116 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2119 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2120 if ((openfiles_new >= maxuserfiles &&
2121 priv_check(td, PRIV_MAXFILES) != 0) ||
2122 openfiles_new >= maxfiles) {
2123 atomic_subtract_int(&openfiles, 1);
2124 if (ppsratecheck(&lastfail, &curfail, 1)) {
2125 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2126 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2130 fp = uma_zalloc(file_zone, M_WAITOK);
2131 bzero(fp, sizeof(*fp));
2132 refcount_init(&fp->f_count, n);
2133 fp->f_cred = crhold(td->td_ucred);
2134 fp->f_ops = &badfileops;
2140 falloc_abort(struct thread *td, struct file *fp)
2144 * For assertion purposes.
2146 refcount_init(&fp->f_count, 0);
2151 * Install a file in a file descriptor table.
2154 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2155 struct filecaps *fcaps)
2157 struct filedescent *fde;
2161 filecaps_validate(fcaps, __func__);
2162 FILEDESC_XLOCK_ASSERT(fdp);
2164 fde = &fdp->fd_ofiles[fd];
2166 seqc_write_begin(&fde->fde_seqc);
2169 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2171 filecaps_move(fcaps, &fde->fde_caps);
2173 filecaps_fill(&fde->fde_caps);
2175 seqc_write_end(&fde->fde_seqc);
2180 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2181 struct filecaps *fcaps)
2183 struct filedesc *fdp = td->td_proc->p_fd;
2188 FILEDESC_XLOCK(fdp);
2189 error = fdalloc(td, 0, fd);
2190 if (__predict_true(error == 0)) {
2191 _finstall(fdp, fp, *fd, flags, fcaps);
2193 FILEDESC_XUNLOCK(fdp);
2198 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2199 struct filecaps *fcaps)
2207 error = finstall_refed(td, fp, fd, flags, fcaps);
2208 if (__predict_false(error != 0)) {
2215 * Build a new filedesc structure from another.
2217 * If fdp is not NULL, return with it shared locked.
2222 struct filedesc0 *newfdp0;
2223 struct filedesc *newfdp;
2225 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2226 newfdp = &newfdp0->fd_fd;
2228 /* Create the file descriptor table. */
2229 FILEDESC_LOCK_INIT(newfdp);
2230 refcount_init(&newfdp->fd_refcnt, 1);
2231 refcount_init(&newfdp->fd_holdcnt, 1);
2232 newfdp->fd_map = newfdp0->fd_dmap;
2233 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2234 newfdp->fd_files->fdt_nfiles = NDFILE;
2240 * Build a pwddesc structure from another.
2241 * Copy the current, root, and jail root vnode references.
2243 * If pdp is not NULL, return with it shared locked.
2246 pdinit(struct pwddesc *pdp, bool keeplock)
2248 struct pwddesc *newpdp;
2251 newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2253 PWDDESC_LOCK_INIT(newpdp);
2254 refcount_init(&newpdp->pd_refcount, 1);
2255 newpdp->pd_cmask = CMASK;
2258 newpwd = pwd_alloc();
2259 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2264 newpwd = pwd_hold_pwddesc(pdp);
2265 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2267 PWDDESC_XUNLOCK(pdp);
2272 * Hold either filedesc or pwddesc of the passed process.
2274 * The process lock is used to synchronize against the target exiting and
2277 * Clearing can be ilustrated in 3 steps:
2278 * 1. set the pointer to NULL. Either routine can race against it, hence
2280 * 2. observe the process lock as not taken. Until then fdhold/pdhold can
2281 * race to either still see the pointer or find NULL. It is still safe to
2282 * grab a reference as clearing is stalled.
2283 * 3. after the lock is observed as not taken, any fdhold/pdhold calls are
2284 * guaranteed to see NULL, making it safe to finish clearing
2286 static struct filedesc *
2287 fdhold(struct proc *p)
2289 struct filedesc *fdp;
2291 PROC_LOCK_ASSERT(p, MA_OWNED);
2292 fdp = atomic_load_ptr(&p->p_fd);
2294 refcount_acquire(&fdp->fd_holdcnt);
2298 static struct pwddesc *
2299 pdhold(struct proc *p)
2301 struct pwddesc *pdp;
2303 PROC_LOCK_ASSERT(p, MA_OWNED);
2304 pdp = atomic_load_ptr(&p->p_pd);
2306 refcount_acquire(&pdp->pd_refcount);
2311 fddrop(struct filedesc *fdp)
2314 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2315 if (refcount_release(&fdp->fd_holdcnt) == 0)
2319 FILEDESC_LOCK_DESTROY(fdp);
2320 uma_zfree(filedesc0_zone, fdp);
2324 pddrop(struct pwddesc *pdp)
2328 if (refcount_release_if_not_last(&pdp->pd_refcount))
2332 if (refcount_release(&pdp->pd_refcount) == 0) {
2333 PWDDESC_XUNLOCK(pdp);
2336 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2338 PWDDESC_XUNLOCK(pdp);
2341 PWDDESC_LOCK_DESTROY(pdp);
2342 free(pdp, M_PWDDESC);
2346 * Share a filedesc structure.
2349 fdshare(struct filedesc *fdp)
2352 refcount_acquire(&fdp->fd_refcnt);
2357 * Share a pwddesc structure.
2360 pdshare(struct pwddesc *pdp)
2362 refcount_acquire(&pdp->pd_refcount);
2367 * Unshare a filedesc structure, if necessary by making a copy
2370 fdunshare(struct thread *td)
2372 struct filedesc *tmp;
2373 struct proc *p = td->td_proc;
2375 if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2378 tmp = fdcopy(p->p_fd);
2384 * Unshare a pwddesc structure.
2387 pdunshare(struct thread *td)
2389 struct pwddesc *pdp;
2394 if (refcount_load(&p->p_pd->pd_refcount) == 1)
2397 pdp = pdcopy(p->p_pd);
2403 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2404 * this is to ease callers, not catch errors.
2407 fdcopy(struct filedesc *fdp)
2409 struct filedesc *newfdp;
2410 struct filedescent *nfde, *ofde;
2416 FILEDESC_SLOCK(fdp);
2418 lastfile = fdlastfile(fdp);
2419 if (lastfile < newfdp->fd_nfiles)
2421 FILEDESC_SUNLOCK(fdp);
2422 fdgrowtable(newfdp, lastfile + 1);
2423 FILEDESC_SLOCK(fdp);
2425 /* copy all passable descriptors (i.e. not kqueue) */
2426 newfdp->fd_freefile = fdp->fd_freefile;
2427 FILEDESC_FOREACH_FDE(fdp, i, ofde) {
2428 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2429 !fhold(ofde->fde_file)) {
2430 if (newfdp->fd_freefile == fdp->fd_freefile)
2431 newfdp->fd_freefile = i;
2434 nfde = &newfdp->fd_ofiles[i];
2436 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2437 fdused_init(newfdp, i);
2439 MPASS(newfdp->fd_freefile != -1);
2440 FILEDESC_SUNLOCK(fdp);
2445 * Copy a pwddesc structure.
2448 pdcopy(struct pwddesc *pdp)
2450 struct pwddesc *newpdp;
2454 newpdp = pdinit(pdp, true);
2455 newpdp->pd_cmask = pdp->pd_cmask;
2456 PWDDESC_XUNLOCK(pdp);
2461 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2462 * one of processes using it exits) and the table used to be shared.
2465 fdclearlocks(struct thread *td)
2467 struct filedesc *fdp;
2468 struct filedesc_to_leader *fdtol;
2478 MPASS(fdtol != NULL);
2480 FILEDESC_XLOCK(fdp);
2481 KASSERT(fdtol->fdl_refcount > 0,
2482 ("filedesc_to_refcount botch: fdl_refcount=%d",
2483 fdtol->fdl_refcount));
2484 if (fdtol->fdl_refcount == 1 &&
2485 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2486 FILEDESC_FOREACH_FP(fdp, i, fp) {
2487 if (fp->f_type != DTYPE_VNODE ||
2490 FILEDESC_XUNLOCK(fdp);
2491 lf.l_whence = SEEK_SET;
2494 lf.l_type = F_UNLCK;
2496 (void) VOP_ADVLOCK(vp,
2497 (caddr_t)p->p_leader, F_UNLCK,
2499 FILEDESC_XLOCK(fdp);
2504 if (fdtol->fdl_refcount == 1) {
2505 if (fdp->fd_holdleaderscount > 0 &&
2506 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2508 * close() or kern_dup() has cleared a reference
2509 * in a shared file descriptor table.
2511 fdp->fd_holdleaderswakeup = 1;
2512 sx_sleep(&fdp->fd_holdleaderscount,
2513 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2516 if (fdtol->fdl_holdcount > 0) {
2518 * Ensure that fdtol->fdl_leader remains
2519 * valid in closef().
2521 fdtol->fdl_wakeup = 1;
2522 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2527 fdtol->fdl_refcount--;
2528 if (fdtol->fdl_refcount == 0 &&
2529 fdtol->fdl_holdcount == 0) {
2530 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2531 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2535 FILEDESC_XUNLOCK(fdp);
2537 free(fdtol, M_FILEDESC_TO_LEADER);
2541 * Release a filedesc structure.
2544 fdescfree_fds(struct thread *td, struct filedesc *fdp)
2546 struct filedesc0 *fdp0;
2547 struct freetable *ft, *tft;
2548 struct filedescent *fde;
2552 KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2553 ("%s: fd table %p carries references", __func__, fdp));
2556 * Serialize with threads iterating over the table, if any.
2558 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2559 FILEDESC_XLOCK(fdp);
2560 FILEDESC_XUNLOCK(fdp);
2563 FILEDESC_FOREACH_FDE(fdp, i, fde) {
2566 (void) closef(fp, td);
2569 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2570 free(fdp->fd_map, M_FILEDESC);
2571 if (fdp->fd_nfiles > NDFILE)
2572 free(fdp->fd_files, M_FILEDESC);
2574 fdp0 = (struct filedesc0 *)fdp;
2575 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2576 free(ft->ft_table, M_FILEDESC);
2582 fdescfree(struct thread *td)
2585 struct filedesc *fdp;
2592 if (RACCT_ENABLED())
2593 racct_set_unlocked(p, RACCT_NOFILE, 0);
2596 if (p->p_fdtol != NULL)
2600 * Check fdhold for an explanation.
2602 atomic_store_ptr(&p->p_fd, NULL);
2603 atomic_thread_fence_seq_cst();
2604 PROC_WAIT_UNLOCKED(p);
2606 if (refcount_release(&fdp->fd_refcnt) == 0)
2609 fdescfree_fds(td, fdp);
2613 pdescfree(struct thread *td)
2616 struct pwddesc *pdp;
2623 * Check pdhold for an explanation.
2625 atomic_store_ptr(&p->p_pd, NULL);
2626 atomic_thread_fence_seq_cst();
2627 PROC_WAIT_UNLOCKED(p);
2633 * For setugid programs, we don't want to people to use that setugidness
2634 * to generate error messages which write to a file which otherwise would
2635 * otherwise be off-limits to the process. We check for filesystems where
2636 * the vnode can change out from under us after execve (like [lin]procfs).
2638 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2639 * sufficient. We also don't check for setugidness since we know we are.
2642 is_unsafe(struct file *fp)
2646 if (fp->f_type != DTYPE_VNODE)
2650 return ((vp->v_vflag & VV_PROCDEP) != 0);
2654 * Make this setguid thing safe, if at all possible.
2657 fdsetugidsafety(struct thread *td)
2659 struct filedesc *fdp;
2663 fdp = td->td_proc->p_fd;
2664 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2665 ("the fdtable should not be shared"));
2666 MPASS(fdp->fd_nfiles >= 3);
2667 for (i = 0; i <= 2; i++) {
2668 fp = fdp->fd_ofiles[i].fde_file;
2669 if (fp != NULL && is_unsafe(fp)) {
2670 FILEDESC_XLOCK(fdp);
2671 knote_fdclose(td, i);
2673 * NULL-out descriptor prior to close to avoid
2674 * a race while close blocks.
2677 FILEDESC_XUNLOCK(fdp);
2678 (void) closef(fp, td);
2684 * If a specific file object occupies a specific file descriptor, close the
2685 * file descriptor entry and drop a reference on the file object. This is a
2686 * convenience function to handle a subsequent error in a function that calls
2687 * falloc() that handles the race that another thread might have closed the
2688 * file descriptor out from under the thread creating the file object.
2691 fdclose(struct thread *td, struct file *fp, int idx)
2693 struct filedesc *fdp = td->td_proc->p_fd;
2695 FILEDESC_XLOCK(fdp);
2696 if (fdp->fd_ofiles[idx].fde_file == fp) {
2698 FILEDESC_XUNLOCK(fdp);
2701 FILEDESC_XUNLOCK(fdp);
2705 * Close any files on exec?
2708 fdcloseexec(struct thread *td)
2710 struct filedesc *fdp;
2711 struct filedescent *fde;
2715 fdp = td->td_proc->p_fd;
2716 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2717 ("the fdtable should not be shared"));
2718 FILEDESC_FOREACH_FDE(fdp, i, fde) {
2720 if (fp->f_type == DTYPE_MQUEUE ||
2721 (fde->fde_flags & UF_EXCLOSE)) {
2722 FILEDESC_XLOCK(fdp);
2724 (void) closefp(fdp, i, fp, td, false, false);
2725 FILEDESC_UNLOCK_ASSERT(fdp);
2731 * It is unsafe for set[ug]id processes to be started with file
2732 * descriptors 0..2 closed, as these descriptors are given implicit
2733 * significance in the Standard C library. fdcheckstd() will create a
2734 * descriptor referencing /dev/null for each of stdin, stdout, and
2735 * stderr that is not already open.
2738 fdcheckstd(struct thread *td)
2740 struct filedesc *fdp;
2742 int i, error, devnull;
2744 fdp = td->td_proc->p_fd;
2745 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2746 ("the fdtable should not be shared"));
2747 MPASS(fdp->fd_nfiles >= 3);
2749 for (i = 0; i <= 2; i++) {
2750 if (fdp->fd_ofiles[i].fde_file != NULL)
2753 save = td->td_retval[0];
2754 if (devnull != -1) {
2755 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2757 error = kern_openat(td, AT_FDCWD, "/dev/null",
2758 UIO_SYSSPACE, O_RDWR, 0);
2760 devnull = td->td_retval[0];
2761 KASSERT(devnull == i, ("we didn't get our fd"));
2764 td->td_retval[0] = save;
2772 * Internal form of close. Decrement reference count on file structure.
2773 * Note: td may be NULL when closing a file that was being passed in a
2777 closef(struct file *fp, struct thread *td)
2781 struct filedesc_to_leader *fdtol;
2782 struct filedesc *fdp;
2787 * POSIX record locking dictates that any close releases ALL
2788 * locks owned by this process. This is handled by setting
2789 * a flag in the unlock to free ONLY locks obeying POSIX
2790 * semantics, and not to free BSD-style file locks.
2791 * If the descriptor was in a message, POSIX-style locks
2792 * aren't passed with the descriptor, and the thread pointer
2793 * will be NULL. Callers should be careful only to pass a
2794 * NULL thread pointer when there really is no owning
2795 * context that might have locks, or the locks will be
2798 if (fp->f_type == DTYPE_VNODE) {
2800 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2801 lf.l_whence = SEEK_SET;
2804 lf.l_type = F_UNLCK;
2805 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2806 F_UNLCK, &lf, F_POSIX);
2808 fdtol = td->td_proc->p_fdtol;
2809 if (fdtol != NULL) {
2811 * Handle special case where file descriptor table is
2812 * shared between multiple process leaders.
2814 fdp = td->td_proc->p_fd;
2815 FILEDESC_XLOCK(fdp);
2816 for (fdtol = fdtol->fdl_next;
2817 fdtol != td->td_proc->p_fdtol;
2818 fdtol = fdtol->fdl_next) {
2819 if ((fdtol->fdl_leader->p_flag &
2822 fdtol->fdl_holdcount++;
2823 FILEDESC_XUNLOCK(fdp);
2824 lf.l_whence = SEEK_SET;
2827 lf.l_type = F_UNLCK;
2829 (void) VOP_ADVLOCK(vp,
2830 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2832 FILEDESC_XLOCK(fdp);
2833 fdtol->fdl_holdcount--;
2834 if (fdtol->fdl_holdcount == 0 &&
2835 fdtol->fdl_wakeup != 0) {
2836 fdtol->fdl_wakeup = 0;
2840 FILEDESC_XUNLOCK(fdp);
2843 return (fdrop_close(fp, td));
2847 * Hack for file descriptor passing code.
2850 closef_nothread(struct file *fp)
2857 * Initialize the file pointer with the specified properties.
2859 * The ops are set with release semantics to be certain that the flags, type,
2860 * and data are visible when ops is. This is to prevent ops methods from being
2861 * called with bad data.
2864 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2869 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2873 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2875 fp->f_seqcount[UIO_READ] = 1;
2876 fp->f_seqcount[UIO_WRITE] = 1;
2877 finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2882 fget_cap_noref(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2883 struct file **fpp, struct filecaps *havecapsp)
2885 struct filedescent *fde;
2888 FILEDESC_LOCK_ASSERT(fdp);
2891 fde = fdeget_noref(fdp, fd);
2898 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2903 if (havecapsp != NULL)
2904 filecaps_copy(&fde->fde_caps, havecapsp, true);
2906 *fpp = fde->fde_file;
2915 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2916 struct file **fpp, struct filecaps *havecapsp)
2918 struct filedesc *fdp = td->td_proc->p_fd;
2925 error = fget_unlocked_seq(td, fd, needrightsp, &fp, &seq);
2929 if (havecapsp != NULL) {
2930 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2931 havecapsp, false)) {
2937 if (!fd_modified(fdp, fd, seq))
2946 FILEDESC_SLOCK(fdp);
2947 error = fget_cap_noref(fdp, fd, needrightsp, fpp, havecapsp);
2948 if (error == 0 && !fhold(*fpp))
2950 FILEDESC_SUNLOCK(fdp);
2955 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2956 struct file **fpp, struct filecaps *havecapsp)
2959 error = fget_unlocked(td, fd, needrightsp, fpp);
2960 if (havecapsp != NULL && error == 0)
2961 filecaps_fill(havecapsp);
2969 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2971 const struct filedescent *fde;
2972 const struct fdescenttbl *fdt;
2973 struct filedesc *fdp;
2976 const cap_rights_t *haverights;
2977 cap_rights_t rights;
2980 VFS_SMR_ASSERT_ENTERED();
2982 rights = *ndp->ni_rightsneeded;
2983 cap_rights_set_one(&rights, CAP_LOOKUP);
2985 fdp = curproc->p_fd;
2986 fdt = fdp->fd_files;
2987 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2989 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
2990 fde = &fdt->fdt_ofiles[fd];
2991 haverights = cap_rights_fde_inline(fde);
2993 if (__predict_false(fp == NULL))
2995 if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2997 *fsearch = ((fp->f_flag & FSEARCH) != 0);
2999 if (__predict_false(vp == NULL)) {
3002 if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
3006 * Use an acquire barrier to force re-reading of fdt so it is
3007 * refreshed for verification.
3009 atomic_thread_fence_acq();
3010 fdt = fdp->fd_files;
3011 if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)))
3014 * If file descriptor doesn't have all rights,
3015 * all lookups relative to it must also be
3016 * strictly relative.
3018 * Not yet supported by fast path.
3021 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
3022 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
3023 ndp->ni_filecaps.fc_nioctls != -1) {
3025 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
3035 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
3037 const struct fdescenttbl *fdt;
3038 struct filedesc *fdp;
3042 VFS_SMR_ASSERT_ENTERED();
3044 fdp = curproc->p_fd;
3045 fdt = fdp->fd_files;
3046 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3048 fp = fdt->fdt_ofiles[fd].fde_file;
3049 if (__predict_false(fp == NULL))
3051 *fsearch = ((fp->f_flag & FSEARCH) != 0);
3053 if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
3057 * Use an acquire barrier to force re-reading of fdt so it is
3058 * refreshed for verification.
3060 atomic_thread_fence_acq();
3061 fdt = fdp->fd_files;
3062 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3064 filecaps_fill(&ndp->ni_filecaps);
3071 fgetvp_lookup(int fd, struct nameidata *ndp, struct vnode **vpp)
3076 struct componentname *cnp;
3077 cap_rights_t rights;
3081 rights = *ndp->ni_rightsneeded;
3082 cap_rights_set_one(&rights, CAP_LOOKUP);
3085 error = fget_cap(td, ndp->ni_dirfd, &rights, &fp, &ndp->ni_filecaps);
3086 if (__predict_false(error != 0))
3088 if (__predict_false(fp->f_ops == &badfileops)) {
3093 if (__predict_false(vp == NULL)) {
3099 * XXX does not check for VDIR, handled by namei_setup
3101 if ((fp->f_flag & FSEARCH) != 0)
3102 cnp->cn_flags |= NOEXECCHECK;
3107 * If file descriptor doesn't have all rights,
3108 * all lookups relative to it must also be
3109 * strictly relative.
3112 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
3113 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
3114 ndp->ni_filecaps.fc_nioctls != -1) {
3115 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
3116 ndp->ni_resflags |= NIRES_STRICTREL;
3121 * TODO: avoid copying ioctl caps if it can be helped to begin with
3123 if ((cnp->cn_flags & WANTIOCTLCAPS) == 0)
3124 filecaps_free_ioctl(&ndp->ni_filecaps);
3130 filecaps_free(&ndp->ni_filecaps);
3136 * Fetch the descriptor locklessly.
3138 * We avoid fdrop() races by never raising a refcount above 0. To accomplish
3139 * this we have to use a cmpset loop rather than an atomic_add. The descriptor
3140 * must be re-verified once we acquire a reference to be certain that the
3141 * identity is still correct and we did not lose a race due to preemption.
3143 * Force a reload of fdt when looping. Another thread could reallocate
3144 * the table before this fd was closed, so it is possible that there is
3145 * a stale fp pointer in cached version.
3149 fget_unlocked_seq(struct thread *td, int fd, cap_rights_t *needrightsp,
3150 struct file **fpp, seqc_t *seqp)
3152 struct filedesc *fdp;
3153 const struct filedescent *fde;
3154 const struct fdescenttbl *fdt;
3157 cap_rights_t haverights;
3160 fdp = td->td_proc->p_fd;
3161 fdt = fdp->fd_files;
3162 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3166 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3167 fde = &fdt->fdt_ofiles[fd];
3168 haverights = *cap_rights_fde_inline(fde);
3170 if (__predict_false(fp == NULL)) {
3171 if (seqc_consistent(fd_seqc(fdt, fd), seq))
3173 fdt = atomic_load_ptr(&fdp->fd_files);
3176 error = cap_check_inline(&haverights, needrightsp);
3177 if (__predict_false(error != 0)) {
3178 if (seqc_consistent(fd_seqc(fdt, fd), seq))
3180 fdt = atomic_load_ptr(&fdp->fd_files);
3183 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3184 fdt = atomic_load_ptr(&fdp->fd_files);
3188 * Use an acquire barrier to force re-reading of fdt so it is
3189 * refreshed for verification.
3191 atomic_thread_fence_acq();
3192 fdt = fdp->fd_files;
3193 if (seqc_consistent_no_fence(fd_seqc(fdt, fd), seq))
3205 fget_unlocked_seq(struct thread *td, int fd, cap_rights_t *needrightsp,
3206 struct file **fpp, seqc_t *seqp __unused)
3208 struct filedesc *fdp;
3209 const struct fdescenttbl *fdt;
3212 fdp = td->td_proc->p_fd;
3213 fdt = fdp->fd_files;
3214 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3218 fp = fdt->fdt_ofiles[fd].fde_file;
3219 if (__predict_false(fp == NULL))
3221 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3222 fdt = atomic_load_ptr(&fdp->fd_files);
3226 * Use an acquire barrier to force re-reading of fdt so it is
3227 * refreshed for verification.
3229 atomic_thread_fence_acq();
3230 fdt = fdp->fd_files;
3231 if (__predict_true(fp == fdt->fdt_ofiles[fd].fde_file))
3241 * See the comments in fget_unlocked_seq for an explanation of how this works.
3243 * This is a simplified variant which bails out to the aforementioned routine
3244 * if anything goes wrong. In practice this only happens when userspace is
3245 * racing with itself.
3248 fget_unlocked(struct thread *td, int fd, cap_rights_t *needrightsp,
3251 struct filedesc *fdp;
3253 const struct filedescent *fde;
3255 const struct fdescenttbl *fdt;
3259 const cap_rights_t *haverights;
3262 fdp = td->td_proc->p_fd;
3263 fdt = fdp->fd_files;
3264 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) {
3269 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3270 fde = &fdt->fdt_ofiles[fd];
3271 haverights = cap_rights_fde_inline(fde);
3274 fp = fdt->fdt_ofiles[fd].fde_file;
3276 if (__predict_false(fp == NULL))
3279 if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3282 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3286 * Use an acquire barrier to force re-reading of fdt so it is
3287 * refreshed for verification.
3289 atomic_thread_fence_acq();
3290 fdt = fdp->fd_files;
3292 if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)))
3294 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3303 return (fget_unlocked_seq(td, fd, needrightsp, fpp, NULL));
3307 * Translate fd -> file when the caller guarantees the file descriptor table
3308 * can't be changed by others.
3310 * Note this does not mean the file object itself is only visible to the caller,
3311 * merely that it wont disappear without having to be referenced.
3313 * Must be paired with fput_only_user.
3317 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3320 const struct filedescent *fde;
3321 const struct fdescenttbl *fdt;
3322 const cap_rights_t *haverights;
3326 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3329 if (__predict_false(fd >= fdp->fd_nfiles))
3332 fdt = fdp->fd_files;
3333 fde = &fdt->fdt_ofiles[fd];
3335 if (__predict_false(fp == NULL))
3337 MPASS(refcount_load(&fp->f_count) > 0);
3338 haverights = cap_rights_fde_inline(fde);
3339 error = cap_check_inline(haverights, needrightsp);
3340 if (__predict_false(error != 0))
3347 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3352 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3355 if (__predict_false(fd >= fdp->fd_nfiles))
3358 fp = fdp->fd_ofiles[fd].fde_file;
3359 if (__predict_false(fp == NULL))
3362 MPASS(refcount_load(&fp->f_count) > 0);
3369 * Extract the file pointer associated with the specified descriptor for the
3370 * current user process.
3372 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3375 * File's rights will be checked against the capability rights mask.
3377 * If an error occurred the non-zero error is returned and *fpp is set to
3378 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
3379 * responsible for fdrop().
3382 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3383 cap_rights_t *needrightsp)
3389 error = fget_unlocked(td, fd, needrightsp, &fp);
3390 if (__predict_false(error != 0))
3392 if (__predict_false(fp->f_ops == &badfileops)) {
3398 * FREAD and FWRITE failure return EBADF as per POSIX.
3404 if ((fp->f_flag & flags) == 0)
3408 if (fp->f_ops != &path_fileops &&
3409 ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3410 (fp->f_flag & FWRITE) != 0))
3416 KASSERT(0, ("wrong flags"));
3429 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3432 return (_fget(td, fd, fpp, 0, rightsp));
3436 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3440 #ifndef CAPABILITIES
3441 error = _fget(td, fd, fpp, 0, rightsp);
3442 if (maxprotp != NULL)
3443 *maxprotp = VM_PROT_ALL;
3446 cap_rights_t fdrights;
3447 struct filedesc *fdp;
3452 fdp = td->td_proc->p_fd;
3453 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3455 error = fget_unlocked_seq(td, fd, rightsp, &fp, &seq);
3456 if (__predict_false(error != 0))
3458 if (__predict_false(fp->f_ops == &badfileops)) {
3462 if (maxprotp != NULL)
3463 fdrights = *cap_rights(fdp, fd);
3464 if (!fd_modified(fdp, fd, seq))
3470 * If requested, convert capability rights to access flags.
3472 if (maxprotp != NULL)
3473 *maxprotp = cap_rights_to_vmprot(&fdrights);
3480 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3483 return (_fget(td, fd, fpp, FREAD, rightsp));
3487 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3490 return (_fget(td, fd, fpp, FWRITE, rightsp));
3494 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3497 #ifndef CAPABILITIES
3498 return (fget_unlocked(td, fd, rightsp, fpp));
3500 struct filedesc *fdp = td->td_proc->p_fd;
3506 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3508 error = fget_unlocked_seq(td, fd, rightsp, &fp, &seq);
3511 error = cap_fcntl_check(fdp, fd, needfcntl);
3512 if (!fd_modified(fdp, fd, seq))
3526 * Like fget() but loads the underlying vnode, or returns an error if the
3527 * descriptor does not represent a vnode. Note that pipes use vnodes but
3528 * never have VM objects. The returned vnode will be vref()'d.
3530 * XXX: what about the unused flags ?
3533 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3540 error = _fget(td, fd, &fp, flags, needrightsp);
3543 if (fp->f_vnode == NULL) {
3555 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3558 return (_fgetvp(td, fd, 0, rightsp, vpp));
3562 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3563 struct filecaps *havecaps, struct vnode **vpp)
3565 struct filecaps caps;
3569 error = fget_cap(td, fd, needrightsp, &fp, &caps);
3572 if (fp->f_ops == &badfileops) {
3576 if (fp->f_vnode == NULL) {
3588 filecaps_free(&caps);
3594 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3597 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3601 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3604 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3609 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3613 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3618 * Handle the last reference to a file being closed.
3620 * Without the noinline attribute clang keeps inlining the func thorough this
3621 * file when fdrop is used.
3624 _fdrop(struct file *fp, struct thread *td)
3630 count = refcount_load(&fp->f_count);
3632 panic("fdrop: fp %p count %d", fp, count);
3634 error = fo_close(fp, td);
3635 atomic_subtract_int(&openfiles, 1);
3637 free(fp->f_advice, M_FADVISE);
3638 uma_zfree(file_zone, fp);
3644 * Apply an advisory lock on a file descriptor.
3646 * Just attempt to get a record lock of the requested type on the entire file
3647 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3649 #ifndef _SYS_SYSPROTO_H_
3657 sys_flock(struct thread *td, struct flock_args *uap)
3664 error = fget(td, uap->fd, &cap_flock_rights, &fp);
3668 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
3671 if (fp->f_ops == &path_fileops) {
3677 lf.l_whence = SEEK_SET;
3680 if (uap->how & LOCK_UN) {
3681 lf.l_type = F_UNLCK;
3682 atomic_clear_int(&fp->f_flag, FHASLOCK);
3683 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3686 if (uap->how & LOCK_EX)
3687 lf.l_type = F_WRLCK;
3688 else if (uap->how & LOCK_SH)
3689 lf.l_type = F_RDLCK;
3694 atomic_set_int(&fp->f_flag, FHASLOCK);
3695 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3696 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3702 * Duplicate the specified descriptor to a free descriptor.
3705 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3706 int openerror, int *indxp)
3708 struct filedescent *newfde, *oldfde;
3713 KASSERT(openerror == ENODEV || openerror == ENXIO,
3714 ("unexpected error %d in %s", openerror, __func__));
3717 * If the to-be-dup'd fd number is greater than the allowed number
3718 * of file descriptors, or the fd to be dup'd has already been
3719 * closed, then reject.
3721 FILEDESC_XLOCK(fdp);
3722 if ((fp = fget_noref(fdp, dfd)) == NULL) {
3723 FILEDESC_XUNLOCK(fdp);
3727 error = fdalloc(td, 0, &indx);
3729 FILEDESC_XUNLOCK(fdp);
3734 * There are two cases of interest here.
3736 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3738 * For ENXIO steal away the file structure from (dfd) and store it in
3739 * (indx). (dfd) is effectively closed by this operation.
3741 switch (openerror) {
3744 * Check that the mode the file is being opened for is a
3745 * subset of the mode of the existing descriptor.
3747 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3748 fdunused(fdp, indx);
3749 FILEDESC_XUNLOCK(fdp);
3753 fdunused(fdp, indx);
3754 FILEDESC_XUNLOCK(fdp);
3757 newfde = &fdp->fd_ofiles[indx];
3758 oldfde = &fdp->fd_ofiles[dfd];
3759 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3761 seqc_write_begin(&newfde->fde_seqc);
3763 fde_copy(oldfde, newfde);
3764 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3767 seqc_write_end(&newfde->fde_seqc);
3772 * Steal away the file pointer from dfd and stuff it into indx.
3774 newfde = &fdp->fd_ofiles[indx];
3775 oldfde = &fdp->fd_ofiles[dfd];
3777 seqc_write_begin(&oldfde->fde_seqc);
3778 seqc_write_begin(&newfde->fde_seqc);
3780 fde_copy(oldfde, newfde);
3781 oldfde->fde_file = NULL;
3784 seqc_write_end(&newfde->fde_seqc);
3785 seqc_write_end(&oldfde->fde_seqc);
3789 FILEDESC_XUNLOCK(fdp);
3795 * This sysctl determines if we will allow a process to chroot(2) if it
3796 * has a directory open:
3797 * 0: disallowed for all processes.
3798 * 1: allowed for processes that were not already chroot(2)'ed.
3799 * 2: allowed for all processes.
3802 static int chroot_allow_open_directories = 1;
3804 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3805 &chroot_allow_open_directories, 0,
3806 "Allow a process to chroot(2) if it has a directory open");
3809 * Helper function for raised chroot(2) security function: Refuse if
3810 * any filedescriptors are open directories.
3813 chroot_refuse_vdir_fds(struct filedesc *fdp)
3819 FILEDESC_LOCK_ASSERT(fdp);
3821 FILEDESC_FOREACH_FP(fdp, i, fp) {
3822 if (fp->f_type == DTYPE_VNODE) {
3824 if (vp->v_type == VDIR)
3832 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3835 if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3836 vrefact(oldpwd->pwd_cdir);
3837 newpwd->pwd_cdir = oldpwd->pwd_cdir;
3840 if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3841 vrefact(oldpwd->pwd_rdir);
3842 newpwd->pwd_rdir = oldpwd->pwd_rdir;
3845 if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3846 vrefact(oldpwd->pwd_jdir);
3847 newpwd->pwd_jdir = oldpwd->pwd_jdir;
3850 if (newpwd->pwd_adir == NULL && oldpwd->pwd_adir != NULL) {
3851 vrefact(oldpwd->pwd_adir);
3852 newpwd->pwd_adir = oldpwd->pwd_adir;
3857 pwd_hold_pwddesc(struct pwddesc *pdp)
3861 PWDDESC_ASSERT_XLOCKED(pdp);
3862 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3864 refcount_acquire(&pwd->pwd_refcount);
3869 pwd_hold_smr(struct pwd *pwd)
3873 if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3880 pwd_hold(struct thread *td)
3882 struct pwddesc *pdp;
3885 pdp = td->td_proc->p_pd;
3888 pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3889 if (pwd_hold_smr(pwd)) {
3895 pwd = pwd_hold_pwddesc(pdp);
3897 PWDDESC_XUNLOCK(pdp);
3902 pwd_hold_proc(struct proc *p)
3904 struct pwddesc *pdp;
3907 PROC_ASSERT_HELD(p);
3914 pwd = pwd_hold_pwddesc(pdp);
3916 PWDDESC_XUNLOCK(pdp);
3926 pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3927 bzero(pwd, sizeof(*pwd));
3928 refcount_init(&pwd->pwd_refcount, 1);
3933 pwd_drop(struct pwd *pwd)
3936 if (!refcount_release(&pwd->pwd_refcount))
3939 if (pwd->pwd_cdir != NULL)
3940 vrele(pwd->pwd_cdir);
3941 if (pwd->pwd_rdir != NULL)
3942 vrele(pwd->pwd_rdir);
3943 if (pwd->pwd_jdir != NULL)
3944 vrele(pwd->pwd_jdir);
3945 if (pwd->pwd_adir != NULL)
3946 vrele(pwd->pwd_adir);
3947 uma_zfree_smr(pwd_zone, pwd);
3951 * The caller is responsible for invoking priv_check() and
3952 * mac_vnode_check_chroot() to authorize this operation.
3955 pwd_chroot(struct thread *td, struct vnode *vp)
3957 struct pwddesc *pdp;
3958 struct filedesc *fdp;
3959 struct pwd *newpwd, *oldpwd;
3962 fdp = td->td_proc->p_fd;
3963 pdp = td->td_proc->p_pd;
3964 newpwd = pwd_alloc();
3965 FILEDESC_SLOCK(fdp);
3967 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3968 if (chroot_allow_open_directories == 0 ||
3969 (chroot_allow_open_directories == 1 &&
3970 oldpwd->pwd_rdir != rootvnode)) {
3971 error = chroot_refuse_vdir_fds(fdp);
3972 FILEDESC_SUNLOCK(fdp);
3974 PWDDESC_XUNLOCK(pdp);
3979 FILEDESC_SUNLOCK(fdp);
3983 newpwd->pwd_rdir = vp;
3985 newpwd->pwd_adir = vp;
3986 if (oldpwd->pwd_jdir == NULL) {
3988 newpwd->pwd_jdir = vp;
3990 pwd_fill(oldpwd, newpwd);
3991 pwd_set(pdp, newpwd);
3992 PWDDESC_XUNLOCK(pdp);
3998 pwd_chdir(struct thread *td, struct vnode *vp)
4000 struct pwddesc *pdp;
4001 struct pwd *newpwd, *oldpwd;
4003 VNPASS(vp->v_usecount > 0, vp);
4005 newpwd = pwd_alloc();
4006 pdp = td->td_proc->p_pd;
4008 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4009 newpwd->pwd_cdir = vp;
4010 pwd_fill(oldpwd, newpwd);
4011 pwd_set(pdp, newpwd);
4012 PWDDESC_XUNLOCK(pdp);
4017 * Process is transitioning to/from a non-native ABI.
4020 pwd_altroot(struct thread *td, struct vnode *altroot_vp)
4022 struct pwddesc *pdp;
4023 struct pwd *newpwd, *oldpwd;
4025 newpwd = pwd_alloc();
4026 pdp = td->td_proc->p_pd;
4028 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4029 if (altroot_vp != NULL) {
4031 * Native process to a non-native ABI.
4034 vrefact(altroot_vp);
4035 newpwd->pwd_adir = altroot_vp;
4038 * Non-native process to the native ABI.
4041 vrefact(oldpwd->pwd_rdir);
4042 newpwd->pwd_adir = oldpwd->pwd_rdir;
4044 pwd_fill(oldpwd, newpwd);
4045 pwd_set(pdp, newpwd);
4046 PWDDESC_XUNLOCK(pdp);
4051 * jail_attach(2) changes both root and working directories.
4054 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
4056 struct pwddesc *pdp;
4057 struct filedesc *fdp;
4058 struct pwd *newpwd, *oldpwd;
4061 fdp = td->td_proc->p_fd;
4062 pdp = td->td_proc->p_pd;
4063 newpwd = pwd_alloc();
4064 FILEDESC_SLOCK(fdp);
4066 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4067 error = chroot_refuse_vdir_fds(fdp);
4068 FILEDESC_SUNLOCK(fdp);
4070 PWDDESC_XUNLOCK(pdp);
4076 newpwd->pwd_rdir = vp;
4078 newpwd->pwd_cdir = vp;
4079 if (oldpwd->pwd_jdir == NULL) {
4081 newpwd->pwd_jdir = vp;
4084 newpwd->pwd_adir = vp;
4085 pwd_fill(oldpwd, newpwd);
4086 pwd_set(pdp, newpwd);
4087 PWDDESC_XUNLOCK(pdp);
4093 pwd_ensure_dirs(void)
4095 struct pwddesc *pdp;
4096 struct pwd *oldpwd, *newpwd;
4098 pdp = curproc->p_pd;
4100 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4101 if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL &&
4102 oldpwd->pwd_adir != NULL) {
4103 PWDDESC_XUNLOCK(pdp);
4106 PWDDESC_XUNLOCK(pdp);
4108 newpwd = pwd_alloc();
4110 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4111 pwd_fill(oldpwd, newpwd);
4112 if (newpwd->pwd_cdir == NULL) {
4114 newpwd->pwd_cdir = rootvnode;
4116 if (newpwd->pwd_rdir == NULL) {
4118 newpwd->pwd_rdir = rootvnode;
4120 if (newpwd->pwd_adir == NULL) {
4122 newpwd->pwd_adir = rootvnode;
4124 pwd_set(pdp, newpwd);
4125 PWDDESC_XUNLOCK(pdp);
4130 pwd_set_rootvnode(void)
4132 struct pwddesc *pdp;
4133 struct pwd *oldpwd, *newpwd;
4135 pdp = curproc->p_pd;
4137 newpwd = pwd_alloc();
4139 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4141 newpwd->pwd_cdir = rootvnode;
4143 newpwd->pwd_rdir = rootvnode;
4145 newpwd->pwd_adir = rootvnode;
4146 pwd_fill(oldpwd, newpwd);
4147 pwd_set(pdp, newpwd);
4148 PWDDESC_XUNLOCK(pdp);
4153 * Scan all active processes and prisons to see if any of them have a current
4154 * or root directory of `olddp'. If so, replace them with the new mount point.
4157 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
4159 struct pwddesc *pdp;
4160 struct pwd *newpwd, *oldpwd;
4165 if (vrefcnt(olddp) == 1)
4168 newpwd = pwd_alloc();
4169 sx_slock(&allproc_lock);
4170 FOREACH_PROC_IN_SYSTEM(p) {
4177 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4178 if (oldpwd == NULL ||
4179 (oldpwd->pwd_cdir != olddp &&
4180 oldpwd->pwd_rdir != olddp &&
4181 oldpwd->pwd_jdir != olddp &&
4182 oldpwd->pwd_adir != olddp)) {
4183 PWDDESC_XUNLOCK(pdp);
4187 if (oldpwd->pwd_cdir == olddp) {
4189 newpwd->pwd_cdir = newdp;
4191 if (oldpwd->pwd_rdir == olddp) {
4193 newpwd->pwd_rdir = newdp;
4195 if (oldpwd->pwd_jdir == olddp) {
4197 newpwd->pwd_jdir = newdp;
4199 if (oldpwd->pwd_adir == olddp) {
4201 newpwd->pwd_adir = newdp;
4203 pwd_fill(oldpwd, newpwd);
4204 pwd_set(pdp, newpwd);
4205 PWDDESC_XUNLOCK(pdp);
4208 newpwd = pwd_alloc();
4210 sx_sunlock(&allproc_lock);
4212 if (rootvnode == olddp) {
4217 mtx_lock(&prison0.pr_mtx);
4218 if (prison0.pr_root == olddp) {
4220 prison0.pr_root = newdp;
4223 mtx_unlock(&prison0.pr_mtx);
4224 sx_slock(&allprison_lock);
4225 TAILQ_FOREACH(pr, &allprison, pr_list) {
4226 mtx_lock(&pr->pr_mtx);
4227 if (pr->pr_root == olddp) {
4229 pr->pr_root = newdp;
4232 mtx_unlock(&pr->pr_mtx);
4234 sx_sunlock(&allprison_lock);
4240 descrip_check_write_mp(struct filedesc *fdp, struct mount *mp)
4247 FILEDESC_SLOCK(fdp);
4248 FILEDESC_FOREACH_FP(fdp, i, fp) {
4249 if (fp->f_type != DTYPE_VNODE ||
4250 (atomic_load_int(&fp->f_flag) & FWRITE) == 0)
4253 if (vp->v_mount == mp) {
4258 FILEDESC_SUNLOCK(fdp);
4262 struct filedesc_to_leader *
4263 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp,
4264 struct proc *leader)
4266 struct filedesc_to_leader *fdtol;
4268 fdtol = malloc(sizeof(struct filedesc_to_leader),
4269 M_FILEDESC_TO_LEADER, M_WAITOK);
4270 fdtol->fdl_refcount = 1;
4271 fdtol->fdl_holdcount = 0;
4272 fdtol->fdl_wakeup = 0;
4273 fdtol->fdl_leader = leader;
4275 FILEDESC_XLOCK(fdp);
4276 fdtol->fdl_next = old->fdl_next;
4277 fdtol->fdl_prev = old;
4278 old->fdl_next = fdtol;
4279 fdtol->fdl_next->fdl_prev = fdtol;
4280 FILEDESC_XUNLOCK(fdp);
4282 fdtol->fdl_next = fdtol;
4283 fdtol->fdl_prev = fdtol;
4288 struct filedesc_to_leader *
4289 filedesc_to_leader_share(struct filedesc_to_leader *fdtol, struct filedesc *fdp)
4291 FILEDESC_XLOCK(fdp);
4292 fdtol->fdl_refcount++;
4293 FILEDESC_XUNLOCK(fdp);
4298 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
4301 struct filedesc *fdp;
4303 int count, off, minoff;
4309 if (*(int *)arg1 != 0)
4312 fdp = curproc->p_fd;
4314 FILEDESC_SLOCK(fdp);
4316 off = NDSLOT(fdp->fd_nfiles - 1);
4317 for (minoff = NDSLOT(0); off >= minoff; --off)
4318 count += bitcountl(map[off]);
4319 FILEDESC_SUNLOCK(fdp);
4321 return (SYSCTL_OUT(req, &count, sizeof(count)));
4324 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
4325 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
4326 "Number of open file descriptors");
4329 * Get file structures globally.
4332 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
4335 struct filedesc *fdp;
4340 error = sysctl_wire_old_buffer(req, 0);
4343 if (req->oldptr == NULL) {
4345 sx_slock(&allproc_lock);
4346 FOREACH_PROC_IN_SYSTEM(p) {
4348 if (p->p_state == PRS_NEW) {
4356 /* overestimates sparse tables. */
4357 n += fdp->fd_nfiles;
4360 sx_sunlock(&allproc_lock);
4361 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4364 bzero(&xf, sizeof(xf));
4365 xf.xf_size = sizeof(xf);
4366 sx_slock(&allproc_lock);
4367 FOREACH_PROC_IN_SYSTEM(p) {
4369 if (p->p_state == PRS_NEW) {
4373 if (p_cansee(req->td, p) != 0) {
4377 xf.xf_pid = p->p_pid;
4378 xf.xf_uid = p->p_ucred->cr_uid;
4383 FILEDESC_SLOCK(fdp);
4384 if (refcount_load(&fdp->fd_refcnt) == 0)
4386 FILEDESC_FOREACH_FP(fdp, n, fp) {
4388 xf.xf_file = (uintptr_t)fp;
4389 xf.xf_data = (uintptr_t)fp->f_data;
4390 xf.xf_vnode = (uintptr_t)fp->f_vnode;
4391 xf.xf_type = (uintptr_t)fp->f_type;
4392 xf.xf_count = refcount_load(&fp->f_count);
4394 xf.xf_offset = foffset_get(fp);
4395 xf.xf_flag = fp->f_flag;
4396 error = SYSCTL_OUT(req, &xf, sizeof(xf));
4399 * There is no need to re-check the fdtable refcount
4400 * here since the filedesc lock is not dropped in the
4407 FILEDESC_SUNLOCK(fdp);
4412 sx_sunlock(&allproc_lock);
4416 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4417 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4419 #ifdef KINFO_FILE_SIZE
4420 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4424 xlate_fflags(int fflags)
4426 static const struct {
4429 } fflags_table[] = {
4430 { FAPPEND, KF_FLAG_APPEND },
4431 { FASYNC, KF_FLAG_ASYNC },
4432 { FFSYNC, KF_FLAG_FSYNC },
4433 { FHASLOCK, KF_FLAG_HASLOCK },
4434 { FNONBLOCK, KF_FLAG_NONBLOCK },
4435 { FREAD, KF_FLAG_READ },
4436 { FWRITE, KF_FLAG_WRITE },
4437 { O_CREAT, KF_FLAG_CREAT },
4438 { O_DIRECT, KF_FLAG_DIRECT },
4439 { O_EXCL, KF_FLAG_EXCL },
4440 { O_EXEC, KF_FLAG_EXEC },
4441 { O_EXLOCK, KF_FLAG_EXLOCK },
4442 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4443 { O_SHLOCK, KF_FLAG_SHLOCK },
4444 { O_TRUNC, KF_FLAG_TRUNC }
4450 for (i = 0; i < nitems(fflags_table); i++)
4451 if (fflags & fflags_table[i].fflag)
4452 kflags |= fflags_table[i].kf_fflag;
4456 /* Trim unused data from kf_path by truncating the structure size. */
4458 pack_kinfo(struct kinfo_file *kif)
4461 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4462 strlen(kif->kf_path) + 1;
4463 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4467 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4468 struct kinfo_file *kif, struct filedesc *fdp, int flags)
4472 bzero(kif, sizeof(*kif));
4474 /* Set a default type to allow for empty fill_kinfo() methods. */
4475 kif->kf_type = KF_TYPE_UNKNOWN;
4476 kif->kf_flags = xlate_fflags(fp->f_flag);
4477 if (rightsp != NULL)
4478 kif->kf_cap_rights = *rightsp;
4480 cap_rights_init_zero(&kif->kf_cap_rights);
4482 kif->kf_ref_count = refcount_load(&fp->f_count);
4483 kif->kf_offset = foffset_get(fp);
4486 * This may drop the filedesc lock, so the 'fp' cannot be
4487 * accessed after this call.
4489 error = fo_fill_kinfo(fp, kif, fdp);
4491 kif->kf_status |= KF_ATTR_VALID;
4492 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4495 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4499 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4500 struct kinfo_file *kif, int flags)
4504 bzero(kif, sizeof(*kif));
4506 kif->kf_type = KF_TYPE_VNODE;
4507 error = vn_fill_kinfo_vnode(vp, kif);
4509 kif->kf_status |= KF_ATTR_VALID;
4510 kif->kf_flags = xlate_fflags(fflags);
4511 cap_rights_init_zero(&kif->kf_cap_rights);
4513 kif->kf_ref_count = -1;
4514 kif->kf_offset = -1;
4515 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4518 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4522 struct export_fd_buf {
4523 struct filedesc *fdp;
4524 struct pwddesc *pdp;
4527 struct kinfo_file kif;
4532 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4534 struct kinfo_file *kif;
4537 if (efbuf->remainder != -1) {
4538 if (efbuf->remainder < kif->kf_structsize)
4540 efbuf->remainder -= kif->kf_structsize;
4542 if (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) != 0)
4543 return (sbuf_error(efbuf->sb));
4548 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4549 struct export_fd_buf *efbuf)
4553 if (efbuf->remainder == 0)
4555 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4557 FILEDESC_SUNLOCK(efbuf->fdp);
4558 error = export_kinfo_to_sb(efbuf);
4559 FILEDESC_SLOCK(efbuf->fdp);
4564 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4565 struct export_fd_buf *efbuf)
4569 if (efbuf->remainder == 0)
4571 if (efbuf->pdp != NULL)
4572 PWDDESC_XUNLOCK(efbuf->pdp);
4573 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4574 error = export_kinfo_to_sb(efbuf);
4575 if (efbuf->pdp != NULL)
4576 PWDDESC_XLOCK(efbuf->pdp);
4581 * Store a process file descriptor information to sbuf.
4583 * Takes a locked proc as argument, and returns with the proc unlocked.
4586 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
4590 struct filedesc *fdp;
4591 struct pwddesc *pdp;
4592 struct export_fd_buf *efbuf;
4593 struct vnode *cttyvp, *textvp, *tracevp;
4596 cap_rights_t rights;
4598 PROC_LOCK_ASSERT(p, MA_OWNED);
4601 tracevp = ktr_get_tracevp(p, true);
4603 textvp = p->p_textvp;
4606 /* Controlling tty. */
4608 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4609 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4617 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4621 efbuf->remainder = maxlen;
4622 efbuf->flags = flags;
4625 if (tracevp != NULL)
4626 error = export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE,
4627 FREAD | FWRITE, efbuf);
4628 if (error == 0 && textvp != NULL)
4629 error = export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD,
4631 if (error == 0 && cttyvp != NULL)
4632 error = export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY,
4633 FREAD | FWRITE, efbuf);
4634 if (error != 0 || pdp == NULL || fdp == NULL)
4639 pwd = pwd_hold_pwddesc(pdp);
4641 /* working directory */
4642 if (pwd->pwd_cdir != NULL) {
4643 vrefact(pwd->pwd_cdir);
4644 error = export_vnode_to_sb(pwd->pwd_cdir,
4645 KF_FD_TYPE_CWD, FREAD, efbuf);
4647 /* root directory */
4648 if (error == 0 && pwd->pwd_rdir != NULL) {
4649 vrefact(pwd->pwd_rdir);
4650 error = export_vnode_to_sb(pwd->pwd_rdir,
4651 KF_FD_TYPE_ROOT, FREAD, efbuf);
4653 /* jail directory */
4654 if (error == 0 && pwd->pwd_jdir != NULL) {
4655 vrefact(pwd->pwd_jdir);
4656 error = export_vnode_to_sb(pwd->pwd_jdir,
4657 KF_FD_TYPE_JAIL, FREAD, efbuf);
4660 PWDDESC_XUNLOCK(pdp);
4665 FILEDESC_SLOCK(fdp);
4666 if (refcount_load(&fdp->fd_refcnt) == 0)
4668 FILEDESC_FOREACH_FP(fdp, i, fp) {
4670 rights = *cap_rights(fdp, i);
4671 #else /* !CAPABILITIES */
4672 rights = cap_no_rights;
4675 * Create sysctl entry. It is OK to drop the filedesc
4676 * lock inside of export_file_to_sb() as we will
4677 * re-validate and re-evaluate its properties when the
4680 error = export_file_to_sb(fp, i, &rights, efbuf);
4681 if (error != 0 || refcount_load(&fdp->fd_refcnt) == 0)
4685 FILEDESC_SUNLOCK(fdp);
4691 free(efbuf, M_TEMP);
4695 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
4698 * Get per-process file descriptors for use by procstat(1), et al.
4701 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4707 int error, error2, *name;
4715 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4716 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4717 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4722 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4723 error = kern_proc_filedesc_out(p, &sb, maxlen,
4724 KERN_FILEDESC_PACK_KINFO);
4725 error2 = sbuf_finish(&sb);
4727 return (error != 0 ? error : error2);
4730 #ifdef COMPAT_FREEBSD7
4731 #ifdef KINFO_OFILE_SIZE
4732 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4736 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4739 okif->kf_structsize = sizeof(*okif);
4740 okif->kf_type = kif->kf_type;
4741 okif->kf_fd = kif->kf_fd;
4742 okif->kf_ref_count = kif->kf_ref_count;
4743 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4744 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4745 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4746 okif->kf_offset = kif->kf_offset;
4747 if (kif->kf_type == KF_TYPE_VNODE)
4748 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4750 okif->kf_vnode_type = KF_VTYPE_VNON;
4751 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4752 if (kif->kf_type == KF_TYPE_SOCKET) {
4753 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4754 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4755 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4756 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4757 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4759 okif->kf_sa_local.ss_family = AF_UNSPEC;
4760 okif->kf_sa_peer.ss_family = AF_UNSPEC;
4765 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4766 struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4771 PWDDESC_XUNLOCK(pdp);
4772 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4773 kinfo_to_okinfo(kif, okif);
4774 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4780 * Get per-process file descriptors for use by procstat(1), et al.
4783 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4785 struct kinfo_ofile *okif;
4786 struct kinfo_file *kif;
4787 struct filedesc *fdp;
4788 struct pwddesc *pdp;
4791 int error, i, *name;
4800 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4807 if (fdp == NULL || pdp == NULL) {
4812 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4813 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4815 pwd = pwd_hold_pwddesc(pdp);
4817 if (pwd->pwd_cdir != NULL)
4818 export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4820 if (pwd->pwd_rdir != NULL)
4821 export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4823 if (pwd->pwd_jdir != NULL)
4824 export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4827 PWDDESC_XUNLOCK(pdp);
4830 FILEDESC_SLOCK(fdp);
4831 if (refcount_load(&fdp->fd_refcnt) == 0)
4833 FILEDESC_FOREACH_FP(fdp, i, fp) {
4834 export_file_to_kinfo(fp, i, NULL, kif, fdp,
4835 KERN_FILEDESC_PACK_KINFO);
4836 FILEDESC_SUNLOCK(fdp);
4837 kinfo_to_okinfo(kif, okif);
4838 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4839 FILEDESC_SLOCK(fdp);
4840 if (error != 0 || refcount_load(&fdp->fd_refcnt) == 0)
4844 FILEDESC_SUNLOCK(fdp);
4852 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4853 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4854 "Process ofiledesc entries");
4855 #endif /* COMPAT_FREEBSD7 */
4858 vntype_to_kinfo(int vtype)
4863 } vtypes_table[] = {
4864 { VBAD, KF_VTYPE_VBAD },
4865 { VBLK, KF_VTYPE_VBLK },
4866 { VCHR, KF_VTYPE_VCHR },
4867 { VDIR, KF_VTYPE_VDIR },
4868 { VFIFO, KF_VTYPE_VFIFO },
4869 { VLNK, KF_VTYPE_VLNK },
4870 { VNON, KF_VTYPE_VNON },
4871 { VREG, KF_VTYPE_VREG },
4872 { VSOCK, KF_VTYPE_VSOCK }
4877 * Perform vtype translation.
4879 for (i = 0; i < nitems(vtypes_table); i++)
4880 if (vtypes_table[i].vtype == vtype)
4881 return (vtypes_table[i].kf_vtype);
4883 return (KF_VTYPE_UNKNOWN);
4886 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4887 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4888 "Process filedesc entries");
4891 * Store a process current working directory information to sbuf.
4893 * Takes a locked proc as argument, and returns with the proc unlocked.
4896 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
4898 struct pwddesc *pdp;
4900 struct export_fd_buf *efbuf;
4904 PROC_LOCK_ASSERT(p, MA_OWNED);
4911 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4915 efbuf->remainder = maxlen;
4919 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4920 cdir = pwd->pwd_cdir;
4925 error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4927 PWDDESC_XUNLOCK(pdp);
4929 free(efbuf, M_TEMP);
4934 * Get per-process current working directory.
4937 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4943 int error, error2, *name;
4951 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4952 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4953 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4958 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4959 error = kern_proc_cwd_out(p, &sb, maxlen);
4960 error2 = sbuf_finish(&sb);
4962 return (error != 0 ? error : error2);
4965 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4966 sysctl_kern_proc_cwd, "Process current working directory");
4970 * For the purposes of debugging, generate a human-readable string for the
4974 file_type_to_name(short type)
5002 case DTYPE_PROCDESC:
5006 case DTYPE_LINUXTFD:
5014 * For the purposes of debugging, identify a process (if any, perhaps one of
5015 * many) that references the passed file in its file descriptor array. Return
5018 static struct proc *
5019 file_to_first_proc(struct file *fp)
5021 struct filedesc *fdp;
5025 FOREACH_PROC_IN_SYSTEM(p) {
5026 if (p->p_state == PRS_NEW)
5031 for (n = 0; n < fdp->fd_nfiles; n++) {
5032 if (fp == fdp->fd_ofiles[n].fde_file)
5040 db_print_file(struct file *fp, int header)
5042 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
5046 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
5047 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
5048 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
5050 p = file_to_first_proc(fp);
5051 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
5052 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
5053 fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
5054 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
5059 DB_SHOW_COMMAND(file, db_show_file)
5064 db_printf("usage: show file <addr>\n");
5067 fp = (struct file *)addr;
5068 db_print_file(fp, 1);
5071 DB_SHOW_COMMAND_FLAGS(files, db_show_files, DB_CMD_MEMSAFE)
5073 struct filedesc *fdp;
5080 FOREACH_PROC_IN_SYSTEM(p) {
5081 if (p->p_state == PRS_NEW)
5083 if ((fdp = p->p_fd) == NULL)
5085 for (n = 0; n < fdp->fd_nfiles; ++n) {
5086 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
5088 db_print_file(fp, header);
5095 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
5096 &maxfilesperproc, 0, "Maximum files allowed open per process");
5098 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
5099 &maxfiles, 0, "Maximum number of files");
5101 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
5102 &openfiles, 0, "System-wide number of open files");
5106 filelistinit(void *dummy)
5109 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
5110 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
5111 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
5112 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
5113 pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
5114 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
5116 * XXXMJG this is a temporary hack due to boot ordering issues against
5119 vfs_smr = uma_zone_get_smr(pwd_zone);
5120 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
5122 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
5124 /*-------------------------------------------------------------------*/
5127 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
5128 int flags, struct thread *td)
5135 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
5143 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
5151 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
5159 badfo_kqfilter(struct file *fp, struct knote *kn)
5166 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
5173 badfo_close(struct file *fp, struct thread *td)
5180 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5188 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5196 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5197 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5205 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
5211 struct fileops badfileops = {
5212 .fo_read = badfo_readwrite,
5213 .fo_write = badfo_readwrite,
5214 .fo_truncate = badfo_truncate,
5215 .fo_ioctl = badfo_ioctl,
5216 .fo_poll = badfo_poll,
5217 .fo_kqfilter = badfo_kqfilter,
5218 .fo_stat = badfo_stat,
5219 .fo_close = badfo_close,
5220 .fo_chmod = badfo_chmod,
5221 .fo_chown = badfo_chown,
5222 .fo_sendfile = badfo_sendfile,
5223 .fo_fill_kinfo = badfo_fill_kinfo,
5227 path_poll(struct file *fp, int events, struct ucred *active_cred,
5234 path_close(struct file *fp, struct thread *td)
5236 MPASS(fp->f_type == DTYPE_VNODE);
5237 fp->f_ops = &badfileops;
5242 struct fileops path_fileops = {
5243 .fo_read = badfo_readwrite,
5244 .fo_write = badfo_readwrite,
5245 .fo_truncate = badfo_truncate,
5246 .fo_ioctl = badfo_ioctl,
5247 .fo_poll = path_poll,
5248 .fo_kqfilter = vn_kqfilter_opath,
5249 .fo_stat = vn_statfile,
5250 .fo_close = path_close,
5251 .fo_chmod = badfo_chmod,
5252 .fo_chown = badfo_chown,
5253 .fo_sendfile = badfo_sendfile,
5254 .fo_fill_kinfo = vn_fill_kinfo,
5255 .fo_flags = DFLAG_PASSABLE,
5259 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
5260 int flags, struct thread *td)
5263 return (EOPNOTSUPP);
5267 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
5275 invfo_ioctl(struct file *fp, u_long com, void *data,
5276 struct ucred *active_cred, struct thread *td)
5283 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
5287 return (poll_no_poll(events));
5291 invfo_kqfilter(struct file *fp, struct knote *kn)
5298 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5306 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5314 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5315 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5322 /*-------------------------------------------------------------------*/
5325 * File Descriptor pseudo-device driver (/dev/fd/).
5327 * Opening minor device N dup()s the file (if any) connected to file
5328 * descriptor N belonging to the calling process. Note that this driver
5329 * consists of only the ``open()'' routine, because all subsequent
5330 * references to this file will be direct to the other driver.
5332 * XXX: we could give this one a cloning event handler if necessary.
5337 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
5341 * XXX Kludge: set curthread->td_dupfd to contain the value of the
5342 * the file descriptor being sought for duplication. The error
5343 * return ensures that the vnode for this device will be released
5344 * by vn_open. Open will detect this special error and take the
5345 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
5346 * will simply report the error.
5348 td->td_dupfd = dev2unit(dev);
5352 static struct cdevsw fildesc_cdevsw = {
5353 .d_version = D_VERSION,
5359 fildesc_drvinit(void *unused)
5363 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
5364 UID_ROOT, GID_WHEEL, 0666, "fd/0");
5365 make_dev_alias(dev, "stdin");
5366 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
5367 UID_ROOT, GID_WHEEL, 0666, "fd/1");
5368 make_dev_alias(dev, "stdout");
5369 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
5370 UID_ROOT, GID_WHEEL, 0666, "fd/2");
5371 make_dev_alias(dev, "stderr");
5374 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);