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;
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 tmp = fp->f_flag & FNONBLOCK;
577 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);
595 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
598 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
600 td->td_retval[0] = tmp;
605 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
609 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
614 error = priv_check(td, PRIV_NFS_LOCKD);
622 /* FALLTHROUGH F_SETLK */
626 flp = (struct flock *)arg;
627 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
632 error = fget_unlocked(td, fd, &cap_flock_rights, &fp);
635 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
641 if (flp->l_whence == SEEK_CUR) {
642 foffset = foffset_get(fp);
645 foffset > OFF_MAX - flp->l_start)) {
650 flp->l_start += foffset;
654 switch (flp->l_type) {
656 if ((fp->f_flag & FREAD) == 0) {
660 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
661 PROC_LOCK(p->p_leader);
662 p->p_leader->p_flag |= P_ADVLOCK;
663 PROC_UNLOCK(p->p_leader);
665 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
669 if ((fp->f_flag & FWRITE) == 0) {
673 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
674 PROC_LOCK(p->p_leader);
675 p->p_leader->p_flag |= P_ADVLOCK;
676 PROC_UNLOCK(p->p_leader);
678 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
682 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
686 if (flg != F_REMOTE) {
690 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
691 F_UNLCKSYS, flp, flg);
697 if (error != 0 || flp->l_type == F_UNLCK ||
698 flp->l_type == F_UNLCKSYS) {
704 * Check for a race with close.
706 * The vnode is now advisory locked (or unlocked, but this case
707 * is not really important) as the caller requested.
708 * We had to drop the filedesc lock, so we need to recheck if
709 * the descriptor is still valid, because if it was closed
710 * in the meantime we need to remove advisory lock from the
711 * vnode - close on any descriptor leading to an advisory
712 * locked vnode, removes that lock.
713 * We will return 0 on purpose in that case, as the result of
714 * successful advisory lock might have been externally visible
715 * already. This is fine - effectively we pretend to the caller
716 * that the closing thread was a bit slower and that the
717 * advisory lock succeeded before the close.
719 error = fget_unlocked(td, fd, &cap_no_rights, &fp2);
725 flp->l_whence = SEEK_SET;
728 flp->l_type = F_UNLCK;
729 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
730 F_UNLCK, flp, F_POSIX);
737 error = fget_unlocked(td, fd, &cap_flock_rights, &fp);
740 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
745 flp = (struct flock *)arg;
746 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
747 flp->l_type != F_UNLCK) {
752 if (flp->l_whence == SEEK_CUR) {
753 foffset = foffset_get(fp);
754 if ((flp->l_start > 0 &&
755 foffset > OFF_MAX - flp->l_start) ||
757 foffset < OFF_MIN - flp->l_start)) {
762 flp->l_start += foffset;
765 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
771 error = fget_unlocked(td, fd, &cap_no_rights, &fp);
774 error = fo_add_seals(fp, arg);
779 error = fget_unlocked(td, fd, &cap_no_rights, &fp);
782 if (fo_get_seals(fp, &seals) == 0)
783 td->td_retval[0] = seals;
790 arg = arg ? 128 * 1024: 0;
793 error = fget_unlocked(td, fd, &cap_no_rights, &fp);
796 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
802 if (vp->v_type != VREG) {
809 * Exclusive lock synchronizes against f_seqcount reads and
810 * writes in sequential_heuristic().
812 error = vn_lock(vp, LK_EXCLUSIVE);
818 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
819 arg = MIN(arg, INT_MAX - bsize + 1);
820 fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
821 (arg + bsize - 1) / bsize);
822 atomic_set_int(&fp->f_flag, FRDAHEAD);
824 atomic_clear_int(&fp->f_flag, FRDAHEAD);
832 * Check if the vnode is part of a union stack (either the
833 * "union" flag from mount(2) or unionfs).
835 * Prior to introduction of this op libc's readdir would call
836 * fstatfs(2), in effect unnecessarily copying kilobytes of
837 * data just to check fs name and a mount flag.
839 * Fixing the code to handle everything in the kernel instead
840 * is a non-trivial endeavor and has low priority, thus this
841 * horrible kludge facilitates the current behavior in a much
842 * cheaper manner until someone(tm) sorts this out.
844 error = fget_unlocked(td, fd, &cap_no_rights, &fp);
847 if (fp->f_type != DTYPE_VNODE) {
854 * Since we don't prevent dooming the vnode even non-null mp
855 * found can become immediately stale. This is tolerable since
856 * mount points are type-stable (providing safe memory access)
857 * and any vfs op on this vnode going forward will return an
858 * error (meaning return value in this case is meaningless).
860 mp = atomic_load_ptr(&vp->v_mount);
861 if (__predict_false(mp == NULL)) {
866 td->td_retval[0] = 0;
867 if (mp->mnt_kern_flag & MNTK_UNIONFS ||
868 mp->mnt_flag & MNT_UNION)
869 td->td_retval[0] = 1;
874 #ifdef CAPABILITY_MODE
875 if (IN_CAPABILITY_MODE(td)) {
880 error = copyin((void *)arg, &kif_sz, sizeof(kif_sz));
883 if (kif_sz != sizeof(*kif)) {
887 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK | M_ZERO);
889 error = fget_cap_noref(fdp, fd, &cap_fcntl_rights, &fp, NULL);
890 if (error == 0 && fhold(fp)) {
891 export_file_to_kinfo(fp, fd, NULL, kif, fdp, 0);
892 FILEDESC_SUNLOCK(fdp);
894 if ((kif->kf_status & KF_ATTR_VALID) != 0) {
895 kif->kf_structsize = sizeof(*kif);
896 error = copyout(kif, (void *)arg, sizeof(*kif));
901 FILEDESC_SUNLOCK(fdp);
916 getmaxfd(struct thread *td)
919 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
923 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
926 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
928 struct filedesc *fdp;
929 struct filedescent *oldfde, *newfde;
931 struct file *delfp, *oldfp;
932 u_long *oioctls, *nioctls;
939 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
940 MPASS(mode < FDDUP_LASTMODE);
943 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
946 * Verify we have a valid descriptor to dup from and possibly to
947 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
948 * return EINVAL when the new descriptor is out of bounds.
953 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
954 maxfd = getmaxfd(td);
956 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
960 if (fget_noref(fdp, old) == NULL)
962 if (mode == FDDUP_FIXED && old == new) {
963 td->td_retval[0] = new;
964 if (flags & FDDUP_FLAG_CLOEXEC)
965 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
970 oldfde = &fdp->fd_ofiles[old];
971 oldfp = oldfde->fde_file;
976 * If the caller specified a file descriptor, make sure the file
977 * table is large enough to hold it, and grab it. Otherwise, just
978 * allocate a new descriptor the usual way.
983 if ((error = fdalloc(td, new, &new)) != 0) {
989 if (new >= fdp->fd_nfiles) {
991 * The resource limits are here instead of e.g.
992 * fdalloc(), because the file descriptor table may be
993 * shared between processes, so we can't really use
994 * racct_add()/racct_sub(). Instead of counting the
995 * number of actually allocated descriptors, just put
996 * the limit on the size of the file descriptor table.
999 if (RACCT_ENABLED()) {
1000 error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
1008 fdgrowtable_exp(fdp, new + 1);
1010 if (!fdisused(fdp, new))
1014 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
1017 KASSERT(old != new, ("new fd is same as old"));
1019 /* Refetch oldfde because the table may have grown and old one freed. */
1020 oldfde = &fdp->fd_ofiles[old];
1021 KASSERT(oldfp == oldfde->fde_file,
1022 ("fdt_ofiles shift from growth observed at fd %d",
1025 newfde = &fdp->fd_ofiles[new];
1026 delfp = newfde->fde_file;
1028 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
1031 * Duplicate the source descriptor.
1034 seqc_write_begin(&newfde->fde_seqc);
1036 oioctls = filecaps_free_prep(&newfde->fde_caps);
1037 fde_copy(oldfde, newfde);
1038 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
1040 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
1041 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
1043 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
1045 seqc_write_end(&newfde->fde_seqc);
1047 td->td_retval[0] = new;
1051 if (delfp != NULL) {
1052 (void) closefp(fdp, new, delfp, td, true, false);
1053 FILEDESC_UNLOCK_ASSERT(fdp);
1056 FILEDESC_XUNLOCK(fdp);
1059 filecaps_free_finish(oioctls);
1064 sigiofree(struct sigio *sigio)
1066 crfree(sigio->sio_ucred);
1067 free(sigio, M_SIGIO);
1070 static struct sigio *
1071 funsetown_locked(struct sigio *sigio)
1076 SIGIO_ASSERT_LOCKED();
1080 *sigio->sio_myref = NULL;
1081 if (sigio->sio_pgid < 0) {
1082 pg = sigio->sio_pgrp;
1084 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio);
1087 p = sigio->sio_proc;
1089 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio);
1096 * If sigio is on the list associated with a process or process group,
1097 * disable signalling from the device, remove sigio from the list and
1101 funsetown(struct sigio **sigiop)
1103 struct sigio *sigio;
1105 /* Racy check, consumers must provide synchronization. */
1106 if (*sigiop == NULL)
1110 sigio = funsetown_locked(*sigiop);
1117 * Free a list of sigio structures. The caller must ensure that new sigio
1118 * structures cannot be added after this point. For process groups this is
1119 * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1123 funsetownlst(struct sigiolst *sigiolst)
1127 struct sigio *sigio, *tmp;
1130 sigio = SLIST_FIRST(sigiolst);
1138 sigio = SLIST_FIRST(sigiolst);
1139 if (sigio == NULL) {
1145 * Every entry of the list should belong to a single proc or pgrp.
1147 if (sigio->sio_pgid < 0) {
1148 pg = sigio->sio_pgrp;
1149 sx_assert(&proctree_lock, SX_XLOCKED);
1151 } else /* if (sigio->sio_pgid > 0) */ {
1152 p = sigio->sio_proc;
1154 KASSERT((p->p_flag & P_WEXIT) != 0,
1155 ("%s: process %p is not exiting", __func__, p));
1158 SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1159 *sigio->sio_myref = NULL;
1161 KASSERT(sigio->sio_pgid < 0,
1162 ("Proc sigio in pgrp sigio list"));
1163 KASSERT(sigio->sio_pgrp == pg,
1164 ("Bogus pgrp in sigio list"));
1165 } else /* if (p != NULL) */ {
1166 KASSERT(sigio->sio_pgid > 0,
1167 ("Pgrp sigio in proc sigio list"));
1168 KASSERT(sigio->sio_proc == p,
1169 ("Bogus proc in sigio list"));
1179 SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1184 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1186 * After permission checking, add a sigio structure to the sigio list for
1187 * the process or process group.
1190 fsetown(pid_t pgid, struct sigio **sigiop)
1194 struct sigio *osigio, *sigio;
1202 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1203 sigio->sio_pgid = pgid;
1204 sigio->sio_ucred = crhold(curthread->td_ucred);
1205 sigio->sio_myref = sigiop;
1209 ret = pget(pgid, PGET_NOTWEXIT | PGET_NOTID | PGET_HOLD, &proc);
1211 osigio = funsetown_locked(*sigiop);
1215 if ((proc->p_flag & P_WEXIT) != 0) {
1217 } else if (proc->p_session !=
1218 curthread->td_proc->p_session) {
1220 * Policy - Don't allow a process to FSETOWN a
1221 * process in another session.
1223 * Remove this test to allow maximum flexibility
1224 * or restrict FSETOWN to the current process or
1225 * process group for maximum safety.
1229 sigio->sio_proc = proc;
1230 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio,
1235 } else /* if (pgid < 0) */ {
1236 sx_slock(&proctree_lock);
1238 osigio = funsetown_locked(*sigiop);
1239 pgrp = pgfind(-pgid);
1243 if (pgrp->pg_session != curthread->td_proc->p_session) {
1245 * Policy - Don't allow a process to FSETOWN a
1246 * process in another session.
1248 * Remove this test to allow maximum flexibility
1249 * or restrict FSETOWN to the current process or
1250 * process group for maximum safety.
1254 sigio->sio_pgrp = pgrp;
1255 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio,
1260 sx_sunlock(&proctree_lock);
1271 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1274 fgetown(struct sigio **sigiop)
1279 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1285 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1290 FILEDESC_XLOCK_ASSERT(fdp);
1293 * We now hold the fp reference that used to be owned by the
1294 * descriptor array. We have to unlock the FILEDESC *AFTER*
1295 * knote_fdclose to prevent a race of the fd getting opened, a knote
1296 * added, and deleteing a knote for the new fd.
1298 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1299 knote_fdclose(td, fd);
1302 * We need to notify mqueue if the object is of type mqueue.
1304 if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1305 mq_fdclose(td, fd, fp);
1306 FILEDESC_XUNLOCK(fdp);
1309 if (AUDITING_TD(td) && audit)
1310 audit_sysclose(td, fd, fp);
1312 error = closef(fp, td);
1315 * All paths leading up to closefp() will have already removed or
1316 * replaced the fd in the filedesc table, so a restart would not
1317 * operate on the same file.
1319 if (error == ERESTART)
1326 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1327 bool holdleaders, bool audit)
1331 FILEDESC_XLOCK_ASSERT(fdp);
1334 if (td->td_proc->p_fdtol != NULL) {
1336 * Ask fdfree() to sleep to ensure that all relevant
1337 * process leaders can be traversed in closef().
1339 fdp->fd_holdleaderscount++;
1341 holdleaders = false;
1345 error = closefp_impl(fdp, fd, fp, td, audit);
1347 FILEDESC_XLOCK(fdp);
1348 fdp->fd_holdleaderscount--;
1349 if (fdp->fd_holdleaderscount == 0 &&
1350 fdp->fd_holdleaderswakeup != 0) {
1351 fdp->fd_holdleaderswakeup = 0;
1352 wakeup(&fdp->fd_holdleaderscount);
1354 FILEDESC_XUNLOCK(fdp);
1360 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1361 bool holdleaders, bool audit)
1364 FILEDESC_XLOCK_ASSERT(fdp);
1366 if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1367 return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1369 return (closefp_impl(fdp, fd, fp, td, audit));
1374 * Close a file descriptor.
1376 #ifndef _SYS_SYSPROTO_H_
1383 sys_close(struct thread *td, struct close_args *uap)
1386 return (kern_close(td, uap->fd));
1390 kern_close(struct thread *td, int fd)
1392 struct filedesc *fdp;
1395 fdp = td->td_proc->p_fd;
1397 FILEDESC_XLOCK(fdp);
1398 if ((fp = fget_noref(fdp, fd)) == NULL) {
1399 FILEDESC_XUNLOCK(fdp);
1404 /* closefp() drops the FILEDESC lock for us. */
1405 return (closefp(fdp, fd, fp, td, true, true));
1409 close_range_cloexec(struct thread *td, u_int lowfd, u_int highfd)
1411 struct filedesc *fdp;
1412 struct fdescenttbl *fdt;
1413 struct filedescent *fde;
1416 fdp = td->td_proc->p_fd;
1417 FILEDESC_XLOCK(fdp);
1418 fdt = atomic_load_ptr(&fdp->fd_files);
1419 highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1421 if (__predict_false(fd > highfd)) {
1424 for (; fd <= highfd; fd++) {
1425 fde = &fdt->fdt_ofiles[fd];
1426 if (fde->fde_file != NULL)
1427 fde->fde_flags |= UF_EXCLOSE;
1430 FILEDESC_XUNLOCK(fdp);
1435 close_range_impl(struct thread *td, u_int lowfd, u_int highfd)
1437 struct filedesc *fdp;
1438 const struct fdescenttbl *fdt;
1442 fdp = td->td_proc->p_fd;
1443 FILEDESC_XLOCK(fdp);
1444 fdt = atomic_load_ptr(&fdp->fd_files);
1445 highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1447 if (__predict_false(fd > highfd)) {
1451 fp = fdt->fdt_ofiles[fd].fde_file;
1457 (void) closefp(fdp, fd, fp, td, true, true);
1460 FILEDESC_XLOCK(fdp);
1461 fdt = atomic_load_ptr(&fdp->fd_files);
1466 FILEDESC_XUNLOCK(fdp);
1472 kern_close_range(struct thread *td, int flags, u_int lowfd, u_int highfd)
1476 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1477 * open should not be a usage error. From a close_range() perspective,
1478 * close_range(3, ~0U, 0) in the same scenario should also likely not
1479 * be a usage error as all fd above 3 are in-fact already closed.
1481 if (highfd < lowfd) {
1485 if ((flags & CLOSE_RANGE_CLOEXEC) != 0)
1486 return (close_range_cloexec(td, lowfd, highfd));
1488 return (close_range_impl(td, lowfd, highfd));
1491 #ifndef _SYS_SYSPROTO_H_
1492 struct close_range_args {
1499 sys_close_range(struct thread *td, struct close_range_args *uap)
1502 AUDIT_ARG_FD(uap->lowfd);
1503 AUDIT_ARG_CMD(uap->highfd);
1504 AUDIT_ARG_FFLAGS(uap->flags);
1506 if ((uap->flags & ~(CLOSE_RANGE_CLOEXEC)) != 0)
1508 return (kern_close_range(td, uap->flags, uap->lowfd, uap->highfd));
1511 #ifdef COMPAT_FREEBSD12
1513 * Close open file descriptors.
1515 #ifndef _SYS_SYSPROTO_H_
1516 struct freebsd12_closefrom_args {
1522 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1526 AUDIT_ARG_FD(uap->lowfd);
1529 * Treat negative starting file descriptor values identical to
1530 * closefrom(0) which closes all files.
1532 lowfd = MAX(0, uap->lowfd);
1533 return (kern_close_range(td, 0, lowfd, ~0U));
1535 #endif /* COMPAT_FREEBSD12 */
1537 #if defined(COMPAT_43)
1539 * Return status information about a file descriptor.
1541 #ifndef _SYS_SYSPROTO_H_
1542 struct ofstat_args {
1549 ofstat(struct thread *td, struct ofstat_args *uap)
1555 error = kern_fstat(td, uap->fd, &ub);
1558 error = copyout(&oub, uap->sb, sizeof(oub));
1562 #endif /* COMPAT_43 */
1564 #if defined(COMPAT_FREEBSD11)
1566 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1569 struct freebsd11_stat osb;
1572 error = kern_fstat(td, uap->fd, &sb);
1575 error = freebsd11_cvtstat(&sb, &osb);
1577 error = copyout(&osb, uap->sb, sizeof(osb));
1580 #endif /* COMPAT_FREEBSD11 */
1583 * Return status information about a file descriptor.
1585 #ifndef _SYS_SYSPROTO_H_
1593 sys_fstat(struct thread *td, struct fstat_args *uap)
1598 error = kern_fstat(td, uap->fd, &ub);
1600 error = copyout(&ub, uap->sb, sizeof(ub));
1605 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1612 error = fget(td, fd, &cap_fstat_rights, &fp);
1613 if (__predict_false(error != 0))
1616 AUDIT_ARG_FILE(td->td_proc, fp);
1618 error = fo_stat(fp, sbp, td->td_ucred);
1620 #ifdef __STAT_TIME_T_EXT
1621 sbp->st_atim_ext = 0;
1622 sbp->st_mtim_ext = 0;
1623 sbp->st_ctim_ext = 0;
1624 sbp->st_btim_ext = 0;
1627 if (KTRPOINT(td, KTR_STRUCT))
1628 ktrstat_error(sbp, error);
1633 #if defined(COMPAT_FREEBSD11)
1635 * Return status information about a file descriptor.
1637 #ifndef _SYS_SYSPROTO_H_
1638 struct freebsd11_nfstat_args {
1645 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1651 error = kern_fstat(td, uap->fd, &ub);
1654 error = freebsd11_cvtnstat(&ub, &nub);
1656 error = copyout(&nub, uap->sb, sizeof(nub));
1659 #endif /* COMPAT_FREEBSD11 */
1662 * Return pathconf information about a file descriptor.
1664 #ifndef _SYS_SYSPROTO_H_
1665 struct fpathconf_args {
1672 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1677 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1679 td->td_retval[0] = value;
1684 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1690 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1694 if (name == _PC_ASYNC_IO) {
1695 *valuep = _POSIX_ASYNCHRONOUS_IO;
1700 vn_lock(vp, LK_SHARED | LK_RETRY);
1701 error = VOP_PATHCONF(vp, name, valuep);
1703 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1704 if (name != _PC_PIPE_BUF) {
1719 * Copy filecaps structure allocating memory for ioctls array if needed.
1721 * The last parameter indicates whether the fdtable is locked. If it is not and
1722 * ioctls are encountered, copying fails and the caller must lock the table.
1724 * Note that if the table was not locked, the caller has to check the relevant
1725 * sequence counter to determine whether the operation was successful.
1728 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1732 if (src->fc_ioctls != NULL && !locked)
1734 memcpy(dst, src, sizeof(*src));
1735 if (src->fc_ioctls == NULL)
1738 KASSERT(src->fc_nioctls > 0,
1739 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1741 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1742 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1743 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1748 filecaps_copy_prep(const struct filecaps *src)
1753 if (__predict_true(src->fc_ioctls == NULL))
1756 KASSERT(src->fc_nioctls > 0,
1757 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1759 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1760 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1765 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1771 if (__predict_true(src->fc_ioctls == NULL)) {
1772 MPASS(ioctls == NULL);
1776 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1777 dst->fc_ioctls = ioctls;
1778 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1782 * Move filecaps structure to the new place and clear the old place.
1785 filecaps_move(struct filecaps *src, struct filecaps *dst)
1789 bzero(src, sizeof(*src));
1793 * Fill the given filecaps structure with full rights.
1796 filecaps_fill(struct filecaps *fcaps)
1799 CAP_ALL(&fcaps->fc_rights);
1800 fcaps->fc_ioctls = NULL;
1801 fcaps->fc_nioctls = -1;
1802 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1806 * Free memory allocated within filecaps structure.
1809 filecaps_free(struct filecaps *fcaps)
1812 free(fcaps->fc_ioctls, M_FILECAPS);
1813 bzero(fcaps, sizeof(*fcaps));
1817 filecaps_free_prep(struct filecaps *fcaps)
1821 ioctls = fcaps->fc_ioctls;
1822 bzero(fcaps, sizeof(*fcaps));
1827 filecaps_free_finish(u_long *ioctls)
1830 free(ioctls, M_FILECAPS);
1834 * Validate the given filecaps structure.
1837 filecaps_validate(const struct filecaps *fcaps, const char *func)
1840 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1841 ("%s: invalid rights", func));
1842 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1843 ("%s: invalid fcntls", func));
1844 KASSERT(fcaps->fc_fcntls == 0 ||
1845 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1846 ("%s: fcntls without CAP_FCNTL", func));
1847 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1848 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1849 ("%s: invalid ioctls", func));
1850 KASSERT(fcaps->fc_nioctls == 0 ||
1851 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1852 ("%s: ioctls without CAP_IOCTL", func));
1856 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1860 FILEDESC_XLOCK_ASSERT(fdp);
1862 nfd1 = fdp->fd_nfiles * 2;
1865 fdgrowtable(fdp, nfd1);
1869 * Grow the file table to accommodate (at least) nfd descriptors.
1872 fdgrowtable(struct filedesc *fdp, int nfd)
1874 struct filedesc0 *fdp0;
1875 struct freetable *ft;
1876 struct fdescenttbl *ntable;
1877 struct fdescenttbl *otable;
1878 int nnfiles, onfiles;
1879 NDSLOTTYPE *nmap, *omap;
1881 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1883 /* save old values */
1884 onfiles = fdp->fd_nfiles;
1885 otable = fdp->fd_files;
1888 /* compute the size of the new table */
1889 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1890 if (nnfiles <= onfiles)
1891 /* the table is already large enough */
1895 * Allocate a new table. We need enough space for the number of
1896 * entries, file entries themselves and the struct freetable we will use
1897 * when we decommission the table and place it on the freelist.
1898 * We place the struct freetable in the middle so we don't have
1899 * to worry about padding.
1901 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1902 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1903 sizeof(struct freetable),
1904 M_FILEDESC, M_ZERO | M_WAITOK);
1905 /* copy the old data */
1906 ntable->fdt_nfiles = nnfiles;
1907 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1908 onfiles * sizeof(ntable->fdt_ofiles[0]));
1911 * Allocate a new map only if the old is not large enough. It will
1912 * grow at a slower rate than the table as it can map more
1913 * entries than the table can hold.
1915 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1916 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1918 /* copy over the old data and update the pointer */
1919 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1924 * Make sure that ntable is correctly initialized before we replace
1925 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1928 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1931 * Free the old file table when not shared by other threads or processes.
1932 * The old file table is considered to be shared when either are true:
1933 * - The process has more than one thread.
1934 * - The file descriptor table has been shared via fdshare().
1936 * When shared, the old file table will be placed on a freelist
1937 * which will be processed when the struct filedesc is released.
1939 * Note that if onfiles == NDFILE, we're dealing with the original
1940 * static allocation contained within (struct filedesc0 *)fdp,
1941 * which must not be freed.
1943 if (onfiles > NDFILE) {
1945 * Note we may be called here from fdinit while allocating a
1946 * table for a new process in which case ->p_fd points
1949 if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) {
1950 free(otable, M_FILEDESC);
1952 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1953 fdp0 = (struct filedesc0 *)fdp;
1954 ft->ft_table = otable;
1955 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1959 * The map does not have the same possibility of threads still
1960 * holding references to it. So always free it as long as it
1961 * does not reference the original static allocation.
1963 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1964 free(omap, M_FILEDESC);
1968 * Allocate a file descriptor for the process.
1971 fdalloc(struct thread *td, int minfd, int *result)
1973 struct proc *p = td->td_proc;
1974 struct filedesc *fdp = p->p_fd;
1975 int fd, maxfd, allocfd;
1980 FILEDESC_XLOCK_ASSERT(fdp);
1982 if (fdp->fd_freefile > minfd)
1983 minfd = fdp->fd_freefile;
1985 maxfd = getmaxfd(td);
1988 * Search the bitmap for a free descriptor starting at minfd.
1989 * If none is found, grow the file table.
1991 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1992 if (__predict_false(fd >= maxfd))
1994 if (__predict_false(fd >= fdp->fd_nfiles)) {
1995 allocfd = min(fd * 2, maxfd);
1997 if (RACCT_ENABLED()) {
1998 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
2004 * fd is already equal to first free descriptor >= minfd, so
2005 * we only need to grow the table and we are done.
2007 fdgrowtable_exp(fdp, allocfd);
2011 * Perform some sanity checks, then mark the file descriptor as
2012 * used and return it to the caller.
2014 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
2015 ("invalid descriptor %d", fd));
2016 KASSERT(!fdisused(fdp, fd),
2017 ("fd_first_free() returned non-free descriptor"));
2018 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
2019 ("file descriptor isn't free"));
2026 * Allocate n file descriptors for the process.
2029 fdallocn(struct thread *td, int minfd, int *fds, int n)
2031 struct proc *p = td->td_proc;
2032 struct filedesc *fdp = p->p_fd;
2035 FILEDESC_XLOCK_ASSERT(fdp);
2037 for (i = 0; i < n; i++)
2038 if (fdalloc(td, 0, &fds[i]) != 0)
2042 for (i--; i >= 0; i--)
2043 fdunused(fdp, fds[i]);
2051 * Create a new open file structure and allocate a file descriptor for the
2052 * process that refers to it. We add one reference to the file for the
2053 * descriptor table and one reference for resultfp. This is to prevent us
2054 * being preempted and the entry in the descriptor table closed after we
2055 * release the FILEDESC lock.
2058 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
2059 struct filecaps *fcaps)
2064 MPASS(resultfp != NULL);
2065 MPASS(resultfd != NULL);
2067 error = _falloc_noinstall(td, &fp, 2);
2068 if (__predict_false(error != 0)) {
2072 error = finstall_refed(td, fp, &fd, flags, fcaps);
2073 if (__predict_false(error != 0)) {
2074 falloc_abort(td, fp);
2085 * Create a new open file structure without allocating a file descriptor.
2088 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2091 int maxuserfiles = maxfiles - (maxfiles / 20);
2093 static struct timeval lastfail;
2096 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2099 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2100 if ((openfiles_new >= maxuserfiles &&
2101 priv_check(td, PRIV_MAXFILES) != 0) ||
2102 openfiles_new >= maxfiles) {
2103 atomic_subtract_int(&openfiles, 1);
2104 if (ppsratecheck(&lastfail, &curfail, 1)) {
2105 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2106 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2110 fp = uma_zalloc(file_zone, M_WAITOK);
2111 bzero(fp, sizeof(*fp));
2112 refcount_init(&fp->f_count, n);
2113 fp->f_cred = crhold(td->td_ucred);
2114 fp->f_ops = &badfileops;
2120 falloc_abort(struct thread *td, struct file *fp)
2124 * For assertion purposes.
2126 refcount_init(&fp->f_count, 0);
2131 * Install a file in a file descriptor table.
2134 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2135 struct filecaps *fcaps)
2137 struct filedescent *fde;
2141 filecaps_validate(fcaps, __func__);
2142 FILEDESC_XLOCK_ASSERT(fdp);
2144 fde = &fdp->fd_ofiles[fd];
2146 seqc_write_begin(&fde->fde_seqc);
2149 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2151 filecaps_move(fcaps, &fde->fde_caps);
2153 filecaps_fill(&fde->fde_caps);
2155 seqc_write_end(&fde->fde_seqc);
2160 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2161 struct filecaps *fcaps)
2163 struct filedesc *fdp = td->td_proc->p_fd;
2168 FILEDESC_XLOCK(fdp);
2169 error = fdalloc(td, 0, fd);
2170 if (__predict_true(error == 0)) {
2171 _finstall(fdp, fp, *fd, flags, fcaps);
2173 FILEDESC_XUNLOCK(fdp);
2178 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2179 struct filecaps *fcaps)
2187 error = finstall_refed(td, fp, fd, flags, fcaps);
2188 if (__predict_false(error != 0)) {
2195 * Build a new filedesc structure from another.
2197 * If fdp is not NULL, return with it shared locked.
2202 struct filedesc0 *newfdp0;
2203 struct filedesc *newfdp;
2205 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2206 newfdp = &newfdp0->fd_fd;
2208 /* Create the file descriptor table. */
2209 FILEDESC_LOCK_INIT(newfdp);
2210 refcount_init(&newfdp->fd_refcnt, 1);
2211 refcount_init(&newfdp->fd_holdcnt, 1);
2212 newfdp->fd_map = newfdp0->fd_dmap;
2213 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2214 newfdp->fd_files->fdt_nfiles = NDFILE;
2220 * Build a pwddesc structure from another.
2221 * Copy the current, root, and jail root vnode references.
2223 * If pdp is not NULL, return with it shared locked.
2226 pdinit(struct pwddesc *pdp, bool keeplock)
2228 struct pwddesc *newpdp;
2231 newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2233 PWDDESC_LOCK_INIT(newpdp);
2234 refcount_init(&newpdp->pd_refcount, 1);
2235 newpdp->pd_cmask = CMASK;
2238 newpwd = pwd_alloc();
2239 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2244 newpwd = pwd_hold_pwddesc(pdp);
2245 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2247 PWDDESC_XUNLOCK(pdp);
2252 * Hold either filedesc or pwddesc of the passed process.
2254 * The process lock is used to synchronize against the target exiting and
2257 * Clearing can be ilustrated in 3 steps:
2258 * 1. set the pointer to NULL. Either routine can race against it, hence
2260 * 2. observe the process lock as not taken. Until then fdhold/pdhold can
2261 * race to either still see the pointer or find NULL. It is still safe to
2262 * grab a reference as clearing is stalled.
2263 * 3. after the lock is observed as not taken, any fdhold/pdhold calls are
2264 * guaranteed to see NULL, making it safe to finish clearing
2266 static struct filedesc *
2267 fdhold(struct proc *p)
2269 struct filedesc *fdp;
2271 PROC_LOCK_ASSERT(p, MA_OWNED);
2272 fdp = atomic_load_ptr(&p->p_fd);
2274 refcount_acquire(&fdp->fd_holdcnt);
2278 static struct pwddesc *
2279 pdhold(struct proc *p)
2281 struct pwddesc *pdp;
2283 PROC_LOCK_ASSERT(p, MA_OWNED);
2284 pdp = atomic_load_ptr(&p->p_pd);
2286 refcount_acquire(&pdp->pd_refcount);
2291 fddrop(struct filedesc *fdp)
2294 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2295 if (refcount_release(&fdp->fd_holdcnt) == 0)
2299 FILEDESC_LOCK_DESTROY(fdp);
2300 uma_zfree(filedesc0_zone, fdp);
2304 pddrop(struct pwddesc *pdp)
2308 if (refcount_release_if_not_last(&pdp->pd_refcount))
2312 if (refcount_release(&pdp->pd_refcount) == 0) {
2313 PWDDESC_XUNLOCK(pdp);
2316 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2318 PWDDESC_XUNLOCK(pdp);
2321 PWDDESC_LOCK_DESTROY(pdp);
2322 free(pdp, M_PWDDESC);
2326 * Share a filedesc structure.
2329 fdshare(struct filedesc *fdp)
2332 refcount_acquire(&fdp->fd_refcnt);
2337 * Share a pwddesc structure.
2340 pdshare(struct pwddesc *pdp)
2342 refcount_acquire(&pdp->pd_refcount);
2347 * Unshare a filedesc structure, if necessary by making a copy
2350 fdunshare(struct thread *td)
2352 struct filedesc *tmp;
2353 struct proc *p = td->td_proc;
2355 if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2358 tmp = fdcopy(p->p_fd);
2364 * Unshare a pwddesc structure.
2367 pdunshare(struct thread *td)
2369 struct pwddesc *pdp;
2374 if (refcount_load(&p->p_pd->pd_refcount) == 1)
2377 pdp = pdcopy(p->p_pd);
2383 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2384 * this is to ease callers, not catch errors.
2387 fdcopy(struct filedesc *fdp)
2389 struct filedesc *newfdp;
2390 struct filedescent *nfde, *ofde;
2396 FILEDESC_SLOCK(fdp);
2398 lastfile = fdlastfile(fdp);
2399 if (lastfile < newfdp->fd_nfiles)
2401 FILEDESC_SUNLOCK(fdp);
2402 fdgrowtable(newfdp, lastfile + 1);
2403 FILEDESC_SLOCK(fdp);
2405 /* copy all passable descriptors (i.e. not kqueue) */
2406 newfdp->fd_freefile = fdp->fd_freefile;
2407 FILEDESC_FOREACH_FDE(fdp, i, ofde) {
2408 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2409 !fhold(ofde->fde_file)) {
2410 if (newfdp->fd_freefile == fdp->fd_freefile)
2411 newfdp->fd_freefile = i;
2414 nfde = &newfdp->fd_ofiles[i];
2416 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2417 fdused_init(newfdp, i);
2419 MPASS(newfdp->fd_freefile != -1);
2420 FILEDESC_SUNLOCK(fdp);
2425 * Copy a pwddesc structure.
2428 pdcopy(struct pwddesc *pdp)
2430 struct pwddesc *newpdp;
2434 newpdp = pdinit(pdp, true);
2435 newpdp->pd_cmask = pdp->pd_cmask;
2436 PWDDESC_XUNLOCK(pdp);
2441 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2442 * one of processes using it exits) and the table used to be shared.
2445 fdclearlocks(struct thread *td)
2447 struct filedesc *fdp;
2448 struct filedesc_to_leader *fdtol;
2458 MPASS(fdtol != NULL);
2460 FILEDESC_XLOCK(fdp);
2461 KASSERT(fdtol->fdl_refcount > 0,
2462 ("filedesc_to_refcount botch: fdl_refcount=%d",
2463 fdtol->fdl_refcount));
2464 if (fdtol->fdl_refcount == 1 &&
2465 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2466 FILEDESC_FOREACH_FP(fdp, i, fp) {
2467 if (fp->f_type != DTYPE_VNODE ||
2470 FILEDESC_XUNLOCK(fdp);
2471 lf.l_whence = SEEK_SET;
2474 lf.l_type = F_UNLCK;
2476 (void) VOP_ADVLOCK(vp,
2477 (caddr_t)p->p_leader, F_UNLCK,
2479 FILEDESC_XLOCK(fdp);
2484 if (fdtol->fdl_refcount == 1) {
2485 if (fdp->fd_holdleaderscount > 0 &&
2486 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2488 * close() or kern_dup() has cleared a reference
2489 * in a shared file descriptor table.
2491 fdp->fd_holdleaderswakeup = 1;
2492 sx_sleep(&fdp->fd_holdleaderscount,
2493 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2496 if (fdtol->fdl_holdcount > 0) {
2498 * Ensure that fdtol->fdl_leader remains
2499 * valid in closef().
2501 fdtol->fdl_wakeup = 1;
2502 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2507 fdtol->fdl_refcount--;
2508 if (fdtol->fdl_refcount == 0 &&
2509 fdtol->fdl_holdcount == 0) {
2510 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2511 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2515 FILEDESC_XUNLOCK(fdp);
2517 free(fdtol, M_FILEDESC_TO_LEADER);
2521 * Release a filedesc structure.
2524 fdescfree_fds(struct thread *td, struct filedesc *fdp)
2526 struct filedesc0 *fdp0;
2527 struct freetable *ft, *tft;
2528 struct filedescent *fde;
2532 KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2533 ("%s: fd table %p carries references", __func__, fdp));
2536 * Serialize with threads iterating over the table, if any.
2538 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2539 FILEDESC_XLOCK(fdp);
2540 FILEDESC_XUNLOCK(fdp);
2543 FILEDESC_FOREACH_FDE(fdp, i, fde) {
2546 (void) closef(fp, td);
2549 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2550 free(fdp->fd_map, M_FILEDESC);
2551 if (fdp->fd_nfiles > NDFILE)
2552 free(fdp->fd_files, M_FILEDESC);
2554 fdp0 = (struct filedesc0 *)fdp;
2555 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2556 free(ft->ft_table, M_FILEDESC);
2562 fdescfree(struct thread *td)
2565 struct filedesc *fdp;
2572 if (RACCT_ENABLED())
2573 racct_set_unlocked(p, RACCT_NOFILE, 0);
2576 if (p->p_fdtol != NULL)
2580 * Check fdhold for an explanation.
2582 atomic_store_ptr(&p->p_fd, NULL);
2583 atomic_thread_fence_seq_cst();
2584 PROC_WAIT_UNLOCKED(p);
2586 if (refcount_release(&fdp->fd_refcnt) == 0)
2589 fdescfree_fds(td, fdp);
2593 pdescfree(struct thread *td)
2596 struct pwddesc *pdp;
2603 * Check pdhold for an explanation.
2605 atomic_store_ptr(&p->p_pd, NULL);
2606 atomic_thread_fence_seq_cst();
2607 PROC_WAIT_UNLOCKED(p);
2613 * For setugid programs, we don't want to people to use that setugidness
2614 * to generate error messages which write to a file which otherwise would
2615 * otherwise be off-limits to the process. We check for filesystems where
2616 * the vnode can change out from under us after execve (like [lin]procfs).
2618 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2619 * sufficient. We also don't check for setugidness since we know we are.
2622 is_unsafe(struct file *fp)
2626 if (fp->f_type != DTYPE_VNODE)
2630 return ((vp->v_vflag & VV_PROCDEP) != 0);
2634 * Make this setguid thing safe, if at all possible.
2637 fdsetugidsafety(struct thread *td)
2639 struct filedesc *fdp;
2643 fdp = td->td_proc->p_fd;
2644 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2645 ("the fdtable should not be shared"));
2646 MPASS(fdp->fd_nfiles >= 3);
2647 for (i = 0; i <= 2; i++) {
2648 fp = fdp->fd_ofiles[i].fde_file;
2649 if (fp != NULL && is_unsafe(fp)) {
2650 FILEDESC_XLOCK(fdp);
2651 knote_fdclose(td, i);
2653 * NULL-out descriptor prior to close to avoid
2654 * a race while close blocks.
2657 FILEDESC_XUNLOCK(fdp);
2658 (void) closef(fp, td);
2664 * If a specific file object occupies a specific file descriptor, close the
2665 * file descriptor entry and drop a reference on the file object. This is a
2666 * convenience function to handle a subsequent error in a function that calls
2667 * falloc() that handles the race that another thread might have closed the
2668 * file descriptor out from under the thread creating the file object.
2671 fdclose(struct thread *td, struct file *fp, int idx)
2673 struct filedesc *fdp = td->td_proc->p_fd;
2675 FILEDESC_XLOCK(fdp);
2676 if (fdp->fd_ofiles[idx].fde_file == fp) {
2678 FILEDESC_XUNLOCK(fdp);
2681 FILEDESC_XUNLOCK(fdp);
2685 * Close any files on exec?
2688 fdcloseexec(struct thread *td)
2690 struct filedesc *fdp;
2691 struct filedescent *fde;
2695 fdp = td->td_proc->p_fd;
2696 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2697 ("the fdtable should not be shared"));
2698 FILEDESC_FOREACH_FDE(fdp, i, fde) {
2700 if (fp->f_type == DTYPE_MQUEUE ||
2701 (fde->fde_flags & UF_EXCLOSE)) {
2702 FILEDESC_XLOCK(fdp);
2704 (void) closefp(fdp, i, fp, td, false, false);
2705 FILEDESC_UNLOCK_ASSERT(fdp);
2711 * It is unsafe for set[ug]id processes to be started with file
2712 * descriptors 0..2 closed, as these descriptors are given implicit
2713 * significance in the Standard C library. fdcheckstd() will create a
2714 * descriptor referencing /dev/null for each of stdin, stdout, and
2715 * stderr that is not already open.
2718 fdcheckstd(struct thread *td)
2720 struct filedesc *fdp;
2722 int i, error, devnull;
2724 fdp = td->td_proc->p_fd;
2725 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2726 ("the fdtable should not be shared"));
2727 MPASS(fdp->fd_nfiles >= 3);
2729 for (i = 0; i <= 2; i++) {
2730 if (fdp->fd_ofiles[i].fde_file != NULL)
2733 save = td->td_retval[0];
2734 if (devnull != -1) {
2735 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2737 error = kern_openat(td, AT_FDCWD, "/dev/null",
2738 UIO_SYSSPACE, O_RDWR, 0);
2740 devnull = td->td_retval[0];
2741 KASSERT(devnull == i, ("we didn't get our fd"));
2744 td->td_retval[0] = save;
2752 * Internal form of close. Decrement reference count on file structure.
2753 * Note: td may be NULL when closing a file that was being passed in a
2757 closef(struct file *fp, struct thread *td)
2761 struct filedesc_to_leader *fdtol;
2762 struct filedesc *fdp;
2767 * POSIX record locking dictates that any close releases ALL
2768 * locks owned by this process. This is handled by setting
2769 * a flag in the unlock to free ONLY locks obeying POSIX
2770 * semantics, and not to free BSD-style file locks.
2771 * If the descriptor was in a message, POSIX-style locks
2772 * aren't passed with the descriptor, and the thread pointer
2773 * will be NULL. Callers should be careful only to pass a
2774 * NULL thread pointer when there really is no owning
2775 * context that might have locks, or the locks will be
2778 if (fp->f_type == DTYPE_VNODE) {
2780 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2781 lf.l_whence = SEEK_SET;
2784 lf.l_type = F_UNLCK;
2785 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2786 F_UNLCK, &lf, F_POSIX);
2788 fdtol = td->td_proc->p_fdtol;
2789 if (fdtol != NULL) {
2791 * Handle special case where file descriptor table is
2792 * shared between multiple process leaders.
2794 fdp = td->td_proc->p_fd;
2795 FILEDESC_XLOCK(fdp);
2796 for (fdtol = fdtol->fdl_next;
2797 fdtol != td->td_proc->p_fdtol;
2798 fdtol = fdtol->fdl_next) {
2799 if ((fdtol->fdl_leader->p_flag &
2802 fdtol->fdl_holdcount++;
2803 FILEDESC_XUNLOCK(fdp);
2804 lf.l_whence = SEEK_SET;
2807 lf.l_type = F_UNLCK;
2809 (void) VOP_ADVLOCK(vp,
2810 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2812 FILEDESC_XLOCK(fdp);
2813 fdtol->fdl_holdcount--;
2814 if (fdtol->fdl_holdcount == 0 &&
2815 fdtol->fdl_wakeup != 0) {
2816 fdtol->fdl_wakeup = 0;
2820 FILEDESC_XUNLOCK(fdp);
2823 return (fdrop_close(fp, td));
2827 * Hack for file descriptor passing code.
2830 closef_nothread(struct file *fp)
2837 * Initialize the file pointer with the specified properties.
2839 * The ops are set with release semantics to be certain that the flags, type,
2840 * and data are visible when ops is. This is to prevent ops methods from being
2841 * called with bad data.
2844 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2849 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2853 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2855 fp->f_seqcount[UIO_READ] = 1;
2856 fp->f_seqcount[UIO_WRITE] = 1;
2857 finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2862 fget_cap_noref(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2863 struct file **fpp, struct filecaps *havecapsp)
2865 struct filedescent *fde;
2868 FILEDESC_LOCK_ASSERT(fdp);
2871 fde = fdeget_noref(fdp, fd);
2878 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2883 if (havecapsp != NULL)
2884 filecaps_copy(&fde->fde_caps, havecapsp, true);
2886 *fpp = fde->fde_file;
2895 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2896 struct file **fpp, struct filecaps *havecapsp)
2898 struct filedesc *fdp = td->td_proc->p_fd;
2905 error = fget_unlocked_seq(td, fd, needrightsp, &fp, &seq);
2909 if (havecapsp != NULL) {
2910 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2911 havecapsp, false)) {
2917 if (!fd_modified(fdp, fd, seq))
2926 FILEDESC_SLOCK(fdp);
2927 error = fget_cap_noref(fdp, fd, needrightsp, fpp, havecapsp);
2928 if (error == 0 && !fhold(*fpp))
2930 FILEDESC_SUNLOCK(fdp);
2935 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2936 struct file **fpp, struct filecaps *havecapsp)
2939 error = fget_unlocked(td, fd, needrightsp, fpp);
2940 if (havecapsp != NULL && error == 0)
2941 filecaps_fill(havecapsp);
2949 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2951 const struct filedescent *fde;
2952 const struct fdescenttbl *fdt;
2953 struct filedesc *fdp;
2956 const cap_rights_t *haverights;
2957 cap_rights_t rights;
2960 VFS_SMR_ASSERT_ENTERED();
2962 rights = *ndp->ni_rightsneeded;
2963 cap_rights_set_one(&rights, CAP_LOOKUP);
2965 fdp = curproc->p_fd;
2966 fdt = fdp->fd_files;
2967 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2969 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
2970 fde = &fdt->fdt_ofiles[fd];
2971 haverights = cap_rights_fde_inline(fde);
2973 if (__predict_false(fp == NULL))
2975 if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2977 *fsearch = ((fp->f_flag & FSEARCH) != 0);
2979 if (__predict_false(vp == NULL)) {
2982 if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2986 * Use an acquire barrier to force re-reading of fdt so it is
2987 * refreshed for verification.
2989 atomic_thread_fence_acq();
2990 fdt = fdp->fd_files;
2991 if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)))
2994 * If file descriptor doesn't have all rights,
2995 * all lookups relative to it must also be
2996 * strictly relative.
2998 * Not yet supported by fast path.
3001 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
3002 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
3003 ndp->ni_filecaps.fc_nioctls != -1) {
3005 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
3015 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
3017 const struct fdescenttbl *fdt;
3018 struct filedesc *fdp;
3022 VFS_SMR_ASSERT_ENTERED();
3024 fdp = curproc->p_fd;
3025 fdt = fdp->fd_files;
3026 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3028 fp = fdt->fdt_ofiles[fd].fde_file;
3029 if (__predict_false(fp == NULL))
3031 *fsearch = ((fp->f_flag & FSEARCH) != 0);
3033 if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
3037 * Use an acquire barrier to force re-reading of fdt so it is
3038 * refreshed for verification.
3040 atomic_thread_fence_acq();
3041 fdt = fdp->fd_files;
3042 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3044 filecaps_fill(&ndp->ni_filecaps);
3051 * Fetch the descriptor locklessly.
3053 * We avoid fdrop() races by never raising a refcount above 0. To accomplish
3054 * this we have to use a cmpset loop rather than an atomic_add. The descriptor
3055 * must be re-verified once we acquire a reference to be certain that the
3056 * identity is still correct and we did not lose a race due to preemption.
3058 * Force a reload of fdt when looping. Another thread could reallocate
3059 * the table before this fd was closed, so it is possible that there is
3060 * a stale fp pointer in cached version.
3064 fget_unlocked_seq(struct thread *td, int fd, cap_rights_t *needrightsp,
3065 struct file **fpp, seqc_t *seqp)
3067 struct filedesc *fdp;
3068 const struct filedescent *fde;
3069 const struct fdescenttbl *fdt;
3072 cap_rights_t haverights;
3075 fdp = td->td_proc->p_fd;
3076 fdt = fdp->fd_files;
3077 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3081 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3082 fde = &fdt->fdt_ofiles[fd];
3083 haverights = *cap_rights_fde_inline(fde);
3085 if (__predict_false(fp == NULL)) {
3086 if (seqc_consistent(fd_seqc(fdt, fd), seq))
3088 fdt = atomic_load_ptr(&fdp->fd_files);
3091 error = cap_check_inline(&haverights, needrightsp);
3092 if (__predict_false(error != 0)) {
3093 if (seqc_consistent(fd_seqc(fdt, fd), seq))
3095 fdt = atomic_load_ptr(&fdp->fd_files);
3098 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3099 fdt = atomic_load_ptr(&fdp->fd_files);
3103 * Use an acquire barrier to force re-reading of fdt so it is
3104 * refreshed for verification.
3106 atomic_thread_fence_acq();
3107 fdt = fdp->fd_files;
3108 if (seqc_consistent_no_fence(fd_seqc(fdt, fd), seq))
3120 fget_unlocked_seq(struct thread *td, int fd, cap_rights_t *needrightsp,
3121 struct file **fpp, seqc_t *seqp __unused)
3123 struct filedesc *fdp;
3124 const struct fdescenttbl *fdt;
3127 fdp = td->td_proc->p_fd;
3128 fdt = fdp->fd_files;
3129 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3133 fp = fdt->fdt_ofiles[fd].fde_file;
3134 if (__predict_false(fp == NULL))
3136 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3137 fdt = atomic_load_ptr(&fdp->fd_files);
3141 * Use an acquire barrier to force re-reading of fdt so it is
3142 * refreshed for verification.
3144 atomic_thread_fence_acq();
3145 fdt = fdp->fd_files;
3146 if (__predict_true(fp == fdt->fdt_ofiles[fd].fde_file))
3156 * See the comments in fget_unlocked_seq for an explanation of how this works.
3158 * This is a simplified variant which bails out to the aforementioned routine
3159 * if anything goes wrong. In practice this only happens when userspace is
3160 * racing with itself.
3163 fget_unlocked(struct thread *td, int fd, cap_rights_t *needrightsp,
3166 struct filedesc *fdp;
3168 const struct filedescent *fde;
3170 const struct fdescenttbl *fdt;
3174 const cap_rights_t *haverights;
3177 fdp = td->td_proc->p_fd;
3178 fdt = fdp->fd_files;
3179 if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) {
3184 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3185 fde = &fdt->fdt_ofiles[fd];
3186 haverights = cap_rights_fde_inline(fde);
3189 fp = fdt->fdt_ofiles[fd].fde_file;
3191 if (__predict_false(fp == NULL))
3194 if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3197 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3201 * Use an acquire barrier to force re-reading of fdt so it is
3202 * refreshed for verification.
3204 atomic_thread_fence_acq();
3205 fdt = fdp->fd_files;
3207 if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)))
3209 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3218 return (fget_unlocked_seq(td, fd, needrightsp, fpp, NULL));
3222 * Translate fd -> file when the caller guarantees the file descriptor table
3223 * can't be changed by others.
3225 * Note this does not mean the file object itself is only visible to the caller,
3226 * merely that it wont disappear without having to be referenced.
3228 * Must be paired with fput_only_user.
3232 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3235 const struct filedescent *fde;
3236 const struct fdescenttbl *fdt;
3237 const cap_rights_t *haverights;
3241 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3244 if (__predict_false(fd >= fdp->fd_nfiles))
3247 fdt = fdp->fd_files;
3248 fde = &fdt->fdt_ofiles[fd];
3250 if (__predict_false(fp == NULL))
3252 MPASS(refcount_load(&fp->f_count) > 0);
3253 haverights = cap_rights_fde_inline(fde);
3254 error = cap_check_inline(haverights, needrightsp);
3255 if (__predict_false(error != 0))
3262 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3267 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3270 if (__predict_false(fd >= fdp->fd_nfiles))
3273 fp = fdp->fd_ofiles[fd].fde_file;
3274 if (__predict_false(fp == NULL))
3277 MPASS(refcount_load(&fp->f_count) > 0);
3284 * Extract the file pointer associated with the specified descriptor for the
3285 * current user process.
3287 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3290 * File's rights will be checked against the capability rights mask.
3292 * If an error occurred the non-zero error is returned and *fpp is set to
3293 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
3294 * responsible for fdrop().
3297 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3298 cap_rights_t *needrightsp)
3304 error = fget_unlocked(td, fd, needrightsp, &fp);
3305 if (__predict_false(error != 0))
3307 if (__predict_false(fp->f_ops == &badfileops)) {
3313 * FREAD and FWRITE failure return EBADF as per POSIX.
3319 if ((fp->f_flag & flags) == 0)
3323 if (fp->f_ops != &path_fileops &&
3324 ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3325 (fp->f_flag & FWRITE) != 0))
3331 KASSERT(0, ("wrong flags"));
3344 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3347 return (_fget(td, fd, fpp, 0, rightsp));
3351 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3355 #ifndef CAPABILITIES
3356 error = _fget(td, fd, fpp, 0, rightsp);
3357 if (maxprotp != NULL)
3358 *maxprotp = VM_PROT_ALL;
3361 cap_rights_t fdrights;
3362 struct filedesc *fdp;
3367 fdp = td->td_proc->p_fd;
3368 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3370 error = fget_unlocked_seq(td, fd, rightsp, &fp, &seq);
3371 if (__predict_false(error != 0))
3373 if (__predict_false(fp->f_ops == &badfileops)) {
3377 if (maxprotp != NULL)
3378 fdrights = *cap_rights(fdp, fd);
3379 if (!fd_modified(fdp, fd, seq))
3385 * If requested, convert capability rights to access flags.
3387 if (maxprotp != NULL)
3388 *maxprotp = cap_rights_to_vmprot(&fdrights);
3395 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3398 return (_fget(td, fd, fpp, FREAD, rightsp));
3402 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3405 return (_fget(td, fd, fpp, FWRITE, rightsp));
3409 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3412 #ifndef CAPABILITIES
3413 return (fget_unlocked(td, fd, rightsp, fpp));
3415 struct filedesc *fdp = td->td_proc->p_fd;
3421 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3423 error = fget_unlocked_seq(td, fd, rightsp, &fp, &seq);
3426 error = cap_fcntl_check(fdp, fd, needfcntl);
3427 if (!fd_modified(fdp, fd, seq))
3441 * Like fget() but loads the underlying vnode, or returns an error if the
3442 * descriptor does not represent a vnode. Note that pipes use vnodes but
3443 * never have VM objects. The returned vnode will be vref()'d.
3445 * XXX: what about the unused flags ?
3448 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3455 error = _fget(td, fd, &fp, flags, needrightsp);
3458 if (fp->f_vnode == NULL) {
3470 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3473 return (_fgetvp(td, fd, 0, rightsp, vpp));
3477 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3478 struct filecaps *havecaps, struct vnode **vpp)
3480 struct filecaps caps;
3484 error = fget_cap(td, fd, needrightsp, &fp, &caps);
3487 if (fp->f_ops == &badfileops) {
3491 if (fp->f_vnode == NULL) {
3503 filecaps_free(&caps);
3509 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3512 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3516 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3519 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3524 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3528 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3533 * Handle the last reference to a file being closed.
3535 * Without the noinline attribute clang keeps inlining the func thorough this
3536 * file when fdrop is used.
3539 _fdrop(struct file *fp, struct thread *td)
3545 count = refcount_load(&fp->f_count);
3547 panic("fdrop: fp %p count %d", fp, count);
3549 error = fo_close(fp, td);
3550 atomic_subtract_int(&openfiles, 1);
3552 free(fp->f_advice, M_FADVISE);
3553 uma_zfree(file_zone, fp);
3559 * Apply an advisory lock on a file descriptor.
3561 * Just attempt to get a record lock of the requested type on the entire file
3562 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3564 #ifndef _SYS_SYSPROTO_H_
3572 sys_flock(struct thread *td, struct flock_args *uap)
3579 error = fget(td, uap->fd, &cap_flock_rights, &fp);
3583 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
3586 if (fp->f_ops == &path_fileops) {
3592 lf.l_whence = SEEK_SET;
3595 if (uap->how & LOCK_UN) {
3596 lf.l_type = F_UNLCK;
3597 atomic_clear_int(&fp->f_flag, FHASLOCK);
3598 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3601 if (uap->how & LOCK_EX)
3602 lf.l_type = F_WRLCK;
3603 else if (uap->how & LOCK_SH)
3604 lf.l_type = F_RDLCK;
3609 atomic_set_int(&fp->f_flag, FHASLOCK);
3610 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3611 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3617 * Duplicate the specified descriptor to a free descriptor.
3620 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3621 int openerror, int *indxp)
3623 struct filedescent *newfde, *oldfde;
3628 KASSERT(openerror == ENODEV || openerror == ENXIO,
3629 ("unexpected error %d in %s", openerror, __func__));
3632 * If the to-be-dup'd fd number is greater than the allowed number
3633 * of file descriptors, or the fd to be dup'd has already been
3634 * closed, then reject.
3636 FILEDESC_XLOCK(fdp);
3637 if ((fp = fget_noref(fdp, dfd)) == NULL) {
3638 FILEDESC_XUNLOCK(fdp);
3642 error = fdalloc(td, 0, &indx);
3644 FILEDESC_XUNLOCK(fdp);
3649 * There are two cases of interest here.
3651 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3653 * For ENXIO steal away the file structure from (dfd) and store it in
3654 * (indx). (dfd) is effectively closed by this operation.
3656 switch (openerror) {
3659 * Check that the mode the file is being opened for is a
3660 * subset of the mode of the existing descriptor.
3662 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3663 fdunused(fdp, indx);
3664 FILEDESC_XUNLOCK(fdp);
3668 fdunused(fdp, indx);
3669 FILEDESC_XUNLOCK(fdp);
3672 newfde = &fdp->fd_ofiles[indx];
3673 oldfde = &fdp->fd_ofiles[dfd];
3674 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3676 seqc_write_begin(&newfde->fde_seqc);
3678 fde_copy(oldfde, newfde);
3679 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3682 seqc_write_end(&newfde->fde_seqc);
3687 * Steal away the file pointer from dfd and stuff it into indx.
3689 newfde = &fdp->fd_ofiles[indx];
3690 oldfde = &fdp->fd_ofiles[dfd];
3692 seqc_write_begin(&oldfde->fde_seqc);
3693 seqc_write_begin(&newfde->fde_seqc);
3695 fde_copy(oldfde, newfde);
3696 oldfde->fde_file = NULL;
3699 seqc_write_end(&newfde->fde_seqc);
3700 seqc_write_end(&oldfde->fde_seqc);
3704 FILEDESC_XUNLOCK(fdp);
3710 * This sysctl determines if we will allow a process to chroot(2) if it
3711 * has a directory open:
3712 * 0: disallowed for all processes.
3713 * 1: allowed for processes that were not already chroot(2)'ed.
3714 * 2: allowed for all processes.
3717 static int chroot_allow_open_directories = 1;
3719 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3720 &chroot_allow_open_directories, 0,
3721 "Allow a process to chroot(2) if it has a directory open");
3724 * Helper function for raised chroot(2) security function: Refuse if
3725 * any filedescriptors are open directories.
3728 chroot_refuse_vdir_fds(struct filedesc *fdp)
3734 FILEDESC_LOCK_ASSERT(fdp);
3736 FILEDESC_FOREACH_FP(fdp, i, fp) {
3737 if (fp->f_type == DTYPE_VNODE) {
3739 if (vp->v_type == VDIR)
3747 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3750 if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3751 vrefact(oldpwd->pwd_cdir);
3752 newpwd->pwd_cdir = oldpwd->pwd_cdir;
3755 if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3756 vrefact(oldpwd->pwd_rdir);
3757 newpwd->pwd_rdir = oldpwd->pwd_rdir;
3760 if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3761 vrefact(oldpwd->pwd_jdir);
3762 newpwd->pwd_jdir = oldpwd->pwd_jdir;
3767 pwd_hold_pwddesc(struct pwddesc *pdp)
3771 PWDDESC_ASSERT_XLOCKED(pdp);
3772 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3774 refcount_acquire(&pwd->pwd_refcount);
3779 pwd_hold_smr(struct pwd *pwd)
3783 if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3790 pwd_hold(struct thread *td)
3792 struct pwddesc *pdp;
3795 pdp = td->td_proc->p_pd;
3798 pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3799 if (pwd_hold_smr(pwd)) {
3805 pwd = pwd_hold_pwddesc(pdp);
3807 PWDDESC_XUNLOCK(pdp);
3812 pwd_hold_proc(struct proc *p)
3814 struct pwddesc *pdp;
3817 PROC_ASSERT_HELD(p);
3824 pwd = pwd_hold_pwddesc(pdp);
3826 PWDDESC_XUNLOCK(pdp);
3836 pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3837 bzero(pwd, sizeof(*pwd));
3838 refcount_init(&pwd->pwd_refcount, 1);
3843 pwd_drop(struct pwd *pwd)
3846 if (!refcount_release(&pwd->pwd_refcount))
3849 if (pwd->pwd_cdir != NULL)
3850 vrele(pwd->pwd_cdir);
3851 if (pwd->pwd_rdir != NULL)
3852 vrele(pwd->pwd_rdir);
3853 if (pwd->pwd_jdir != NULL)
3854 vrele(pwd->pwd_jdir);
3855 uma_zfree_smr(pwd_zone, pwd);
3859 * The caller is responsible for invoking priv_check() and
3860 * mac_vnode_check_chroot() to authorize this operation.
3863 pwd_chroot(struct thread *td, struct vnode *vp)
3865 struct pwddesc *pdp;
3866 struct filedesc *fdp;
3867 struct pwd *newpwd, *oldpwd;
3870 fdp = td->td_proc->p_fd;
3871 pdp = td->td_proc->p_pd;
3872 newpwd = pwd_alloc();
3873 FILEDESC_SLOCK(fdp);
3875 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3876 if (chroot_allow_open_directories == 0 ||
3877 (chroot_allow_open_directories == 1 &&
3878 oldpwd->pwd_rdir != rootvnode)) {
3879 error = chroot_refuse_vdir_fds(fdp);
3880 FILEDESC_SUNLOCK(fdp);
3882 PWDDESC_XUNLOCK(pdp);
3887 FILEDESC_SUNLOCK(fdp);
3891 newpwd->pwd_rdir = vp;
3892 if (oldpwd->pwd_jdir == NULL) {
3894 newpwd->pwd_jdir = vp;
3896 pwd_fill(oldpwd, newpwd);
3897 pwd_set(pdp, newpwd);
3898 PWDDESC_XUNLOCK(pdp);
3904 pwd_chdir(struct thread *td, struct vnode *vp)
3906 struct pwddesc *pdp;
3907 struct pwd *newpwd, *oldpwd;
3909 VNPASS(vp->v_usecount > 0, vp);
3911 newpwd = pwd_alloc();
3912 pdp = td->td_proc->p_pd;
3914 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3915 newpwd->pwd_cdir = vp;
3916 pwd_fill(oldpwd, newpwd);
3917 pwd_set(pdp, newpwd);
3918 PWDDESC_XUNLOCK(pdp);
3923 * jail_attach(2) changes both root and working directories.
3926 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
3928 struct pwddesc *pdp;
3929 struct filedesc *fdp;
3930 struct pwd *newpwd, *oldpwd;
3933 fdp = td->td_proc->p_fd;
3934 pdp = td->td_proc->p_pd;
3935 newpwd = pwd_alloc();
3936 FILEDESC_SLOCK(fdp);
3938 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3939 error = chroot_refuse_vdir_fds(fdp);
3940 FILEDESC_SUNLOCK(fdp);
3942 PWDDESC_XUNLOCK(pdp);
3948 newpwd->pwd_rdir = vp;
3950 newpwd->pwd_cdir = vp;
3951 if (oldpwd->pwd_jdir == NULL) {
3953 newpwd->pwd_jdir = vp;
3955 pwd_fill(oldpwd, newpwd);
3956 pwd_set(pdp, newpwd);
3957 PWDDESC_XUNLOCK(pdp);
3963 pwd_ensure_dirs(void)
3965 struct pwddesc *pdp;
3966 struct pwd *oldpwd, *newpwd;
3968 pdp = curproc->p_pd;
3970 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3971 if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3972 PWDDESC_XUNLOCK(pdp);
3975 PWDDESC_XUNLOCK(pdp);
3977 newpwd = pwd_alloc();
3979 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3980 pwd_fill(oldpwd, newpwd);
3981 if (newpwd->pwd_cdir == NULL) {
3983 newpwd->pwd_cdir = rootvnode;
3985 if (newpwd->pwd_rdir == NULL) {
3987 newpwd->pwd_rdir = rootvnode;
3989 pwd_set(pdp, newpwd);
3990 PWDDESC_XUNLOCK(pdp);
3995 pwd_set_rootvnode(void)
3997 struct pwddesc *pdp;
3998 struct pwd *oldpwd, *newpwd;
4000 pdp = curproc->p_pd;
4002 newpwd = pwd_alloc();
4004 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4006 newpwd->pwd_cdir = rootvnode;
4008 newpwd->pwd_rdir = rootvnode;
4009 pwd_fill(oldpwd, newpwd);
4010 pwd_set(pdp, newpwd);
4011 PWDDESC_XUNLOCK(pdp);
4016 * Scan all active processes and prisons to see if any of them have a current
4017 * or root directory of `olddp'. If so, replace them with the new mount point.
4020 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
4022 struct pwddesc *pdp;
4023 struct pwd *newpwd, *oldpwd;
4028 if (vrefcnt(olddp) == 1)
4031 newpwd = pwd_alloc();
4032 sx_slock(&allproc_lock);
4033 FOREACH_PROC_IN_SYSTEM(p) {
4040 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4041 if (oldpwd == NULL ||
4042 (oldpwd->pwd_cdir != olddp &&
4043 oldpwd->pwd_rdir != olddp &&
4044 oldpwd->pwd_jdir != olddp)) {
4045 PWDDESC_XUNLOCK(pdp);
4049 if (oldpwd->pwd_cdir == olddp) {
4051 newpwd->pwd_cdir = newdp;
4053 if (oldpwd->pwd_rdir == olddp) {
4055 newpwd->pwd_rdir = newdp;
4057 if (oldpwd->pwd_jdir == olddp) {
4059 newpwd->pwd_jdir = newdp;
4061 pwd_fill(oldpwd, newpwd);
4062 pwd_set(pdp, newpwd);
4063 PWDDESC_XUNLOCK(pdp);
4066 newpwd = pwd_alloc();
4068 sx_sunlock(&allproc_lock);
4070 if (rootvnode == olddp) {
4075 mtx_lock(&prison0.pr_mtx);
4076 if (prison0.pr_root == olddp) {
4078 prison0.pr_root = newdp;
4081 mtx_unlock(&prison0.pr_mtx);
4082 sx_slock(&allprison_lock);
4083 TAILQ_FOREACH(pr, &allprison, pr_list) {
4084 mtx_lock(&pr->pr_mtx);
4085 if (pr->pr_root == olddp) {
4087 pr->pr_root = newdp;
4090 mtx_unlock(&pr->pr_mtx);
4092 sx_sunlock(&allprison_lock);
4097 struct filedesc_to_leader *
4098 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
4100 struct filedesc_to_leader *fdtol;
4102 fdtol = malloc(sizeof(struct filedesc_to_leader),
4103 M_FILEDESC_TO_LEADER, M_WAITOK);
4104 fdtol->fdl_refcount = 1;
4105 fdtol->fdl_holdcount = 0;
4106 fdtol->fdl_wakeup = 0;
4107 fdtol->fdl_leader = leader;
4109 FILEDESC_XLOCK(fdp);
4110 fdtol->fdl_next = old->fdl_next;
4111 fdtol->fdl_prev = old;
4112 old->fdl_next = fdtol;
4113 fdtol->fdl_next->fdl_prev = fdtol;
4114 FILEDESC_XUNLOCK(fdp);
4116 fdtol->fdl_next = fdtol;
4117 fdtol->fdl_prev = fdtol;
4123 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
4126 struct filedesc *fdp;
4128 int count, off, minoff;
4134 if (*(int *)arg1 != 0)
4137 fdp = curproc->p_fd;
4139 FILEDESC_SLOCK(fdp);
4141 off = NDSLOT(fdp->fd_nfiles - 1);
4142 for (minoff = NDSLOT(0); off >= minoff; --off)
4143 count += bitcountl(map[off]);
4144 FILEDESC_SUNLOCK(fdp);
4146 return (SYSCTL_OUT(req, &count, sizeof(count)));
4149 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
4150 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
4151 "Number of open file descriptors");
4154 * Get file structures globally.
4157 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
4160 struct filedesc *fdp;
4165 error = sysctl_wire_old_buffer(req, 0);
4168 if (req->oldptr == NULL) {
4170 sx_slock(&allproc_lock);
4171 FOREACH_PROC_IN_SYSTEM(p) {
4173 if (p->p_state == PRS_NEW) {
4181 /* overestimates sparse tables. */
4182 n += fdp->fd_nfiles;
4185 sx_sunlock(&allproc_lock);
4186 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4189 bzero(&xf, sizeof(xf));
4190 xf.xf_size = sizeof(xf);
4191 sx_slock(&allproc_lock);
4192 FOREACH_PROC_IN_SYSTEM(p) {
4194 if (p->p_state == PRS_NEW) {
4198 if (p_cansee(req->td, p) != 0) {
4202 xf.xf_pid = p->p_pid;
4203 xf.xf_uid = p->p_ucred->cr_uid;
4208 FILEDESC_SLOCK(fdp);
4209 if (refcount_load(&fdp->fd_refcnt) == 0)
4211 FILEDESC_FOREACH_FP(fdp, n, fp) {
4213 xf.xf_file = (uintptr_t)fp;
4214 xf.xf_data = (uintptr_t)fp->f_data;
4215 xf.xf_vnode = (uintptr_t)fp->f_vnode;
4216 xf.xf_type = (uintptr_t)fp->f_type;
4217 xf.xf_count = refcount_load(&fp->f_count);
4219 xf.xf_offset = foffset_get(fp);
4220 xf.xf_flag = fp->f_flag;
4221 error = SYSCTL_OUT(req, &xf, sizeof(xf));
4224 * There is no need to re-check the fdtable refcount
4225 * here since the filedesc lock is not dropped in the
4232 FILEDESC_SUNLOCK(fdp);
4237 sx_sunlock(&allproc_lock);
4241 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4242 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4244 #ifdef KINFO_FILE_SIZE
4245 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4249 xlate_fflags(int fflags)
4251 static const struct {
4254 } fflags_table[] = {
4255 { FAPPEND, KF_FLAG_APPEND },
4256 { FASYNC, KF_FLAG_ASYNC },
4257 { FFSYNC, KF_FLAG_FSYNC },
4258 { FHASLOCK, KF_FLAG_HASLOCK },
4259 { FNONBLOCK, KF_FLAG_NONBLOCK },
4260 { FREAD, KF_FLAG_READ },
4261 { FWRITE, KF_FLAG_WRITE },
4262 { O_CREAT, KF_FLAG_CREAT },
4263 { O_DIRECT, KF_FLAG_DIRECT },
4264 { O_EXCL, KF_FLAG_EXCL },
4265 { O_EXEC, KF_FLAG_EXEC },
4266 { O_EXLOCK, KF_FLAG_EXLOCK },
4267 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4268 { O_SHLOCK, KF_FLAG_SHLOCK },
4269 { O_TRUNC, KF_FLAG_TRUNC }
4275 for (i = 0; i < nitems(fflags_table); i++)
4276 if (fflags & fflags_table[i].fflag)
4277 kflags |= fflags_table[i].kf_fflag;
4281 /* Trim unused data from kf_path by truncating the structure size. */
4283 pack_kinfo(struct kinfo_file *kif)
4286 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4287 strlen(kif->kf_path) + 1;
4288 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4292 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4293 struct kinfo_file *kif, struct filedesc *fdp, int flags)
4297 bzero(kif, sizeof(*kif));
4299 /* Set a default type to allow for empty fill_kinfo() methods. */
4300 kif->kf_type = KF_TYPE_UNKNOWN;
4301 kif->kf_flags = xlate_fflags(fp->f_flag);
4302 if (rightsp != NULL)
4303 kif->kf_cap_rights = *rightsp;
4305 cap_rights_init_zero(&kif->kf_cap_rights);
4307 kif->kf_ref_count = refcount_load(&fp->f_count);
4308 kif->kf_offset = foffset_get(fp);
4311 * This may drop the filedesc lock, so the 'fp' cannot be
4312 * accessed after this call.
4314 error = fo_fill_kinfo(fp, kif, fdp);
4316 kif->kf_status |= KF_ATTR_VALID;
4317 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4320 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4324 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4325 struct kinfo_file *kif, int flags)
4329 bzero(kif, sizeof(*kif));
4331 kif->kf_type = KF_TYPE_VNODE;
4332 error = vn_fill_kinfo_vnode(vp, kif);
4334 kif->kf_status |= KF_ATTR_VALID;
4335 kif->kf_flags = xlate_fflags(fflags);
4336 cap_rights_init_zero(&kif->kf_cap_rights);
4338 kif->kf_ref_count = -1;
4339 kif->kf_offset = -1;
4340 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4343 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4347 struct export_fd_buf {
4348 struct filedesc *fdp;
4349 struct pwddesc *pdp;
4352 struct kinfo_file kif;
4357 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4359 struct kinfo_file *kif;
4362 if (efbuf->remainder != -1) {
4363 if (efbuf->remainder < kif->kf_structsize)
4365 efbuf->remainder -= kif->kf_structsize;
4367 if (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) != 0)
4368 return (sbuf_error(efbuf->sb));
4373 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4374 struct export_fd_buf *efbuf)
4378 if (efbuf->remainder == 0)
4380 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4382 FILEDESC_SUNLOCK(efbuf->fdp);
4383 error = export_kinfo_to_sb(efbuf);
4384 FILEDESC_SLOCK(efbuf->fdp);
4389 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4390 struct export_fd_buf *efbuf)
4394 if (efbuf->remainder == 0)
4396 if (efbuf->pdp != NULL)
4397 PWDDESC_XUNLOCK(efbuf->pdp);
4398 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4399 error = export_kinfo_to_sb(efbuf);
4400 if (efbuf->pdp != NULL)
4401 PWDDESC_XLOCK(efbuf->pdp);
4406 * Store a process file descriptor information to sbuf.
4408 * Takes a locked proc as argument, and returns with the proc unlocked.
4411 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
4415 struct filedesc *fdp;
4416 struct pwddesc *pdp;
4417 struct export_fd_buf *efbuf;
4418 struct vnode *cttyvp, *textvp, *tracevp;
4421 cap_rights_t rights;
4423 PROC_LOCK_ASSERT(p, MA_OWNED);
4426 tracevp = ktr_get_tracevp(p, true);
4428 textvp = p->p_textvp;
4431 /* Controlling tty. */
4433 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4434 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4442 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4446 efbuf->remainder = maxlen;
4447 efbuf->flags = flags;
4450 if (tracevp != NULL)
4451 error = export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE,
4452 FREAD | FWRITE, efbuf);
4453 if (error == 0 && textvp != NULL)
4454 error = export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD,
4456 if (error == 0 && cttyvp != NULL)
4457 error = export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY,
4458 FREAD | FWRITE, efbuf);
4459 if (error != 0 || pdp == NULL || fdp == NULL)
4464 pwd = pwd_hold_pwddesc(pdp);
4466 /* working directory */
4467 if (pwd->pwd_cdir != NULL) {
4468 vrefact(pwd->pwd_cdir);
4469 error = export_vnode_to_sb(pwd->pwd_cdir,
4470 KF_FD_TYPE_CWD, FREAD, efbuf);
4472 /* root directory */
4473 if (error == 0 && pwd->pwd_rdir != NULL) {
4474 vrefact(pwd->pwd_rdir);
4475 error = export_vnode_to_sb(pwd->pwd_rdir,
4476 KF_FD_TYPE_ROOT, FREAD, efbuf);
4478 /* jail directory */
4479 if (error == 0 && pwd->pwd_jdir != NULL) {
4480 vrefact(pwd->pwd_jdir);
4481 error = export_vnode_to_sb(pwd->pwd_jdir,
4482 KF_FD_TYPE_JAIL, FREAD, efbuf);
4485 PWDDESC_XUNLOCK(pdp);
4490 FILEDESC_SLOCK(fdp);
4491 if (refcount_load(&fdp->fd_refcnt) == 0)
4493 FILEDESC_FOREACH_FP(fdp, i, fp) {
4495 rights = *cap_rights(fdp, i);
4496 #else /* !CAPABILITIES */
4497 rights = cap_no_rights;
4500 * Create sysctl entry. It is OK to drop the filedesc
4501 * lock inside of export_file_to_sb() as we will
4502 * re-validate and re-evaluate its properties when the
4505 error = export_file_to_sb(fp, i, &rights, efbuf);
4506 if (error != 0 || refcount_load(&fdp->fd_refcnt) == 0)
4510 FILEDESC_SUNLOCK(fdp);
4516 free(efbuf, M_TEMP);
4520 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
4523 * Get per-process file descriptors for use by procstat(1), et al.
4526 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4532 int error, error2, *name;
4540 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4541 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4542 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4547 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4548 error = kern_proc_filedesc_out(p, &sb, maxlen,
4549 KERN_FILEDESC_PACK_KINFO);
4550 error2 = sbuf_finish(&sb);
4552 return (error != 0 ? error : error2);
4555 #ifdef COMPAT_FREEBSD7
4556 #ifdef KINFO_OFILE_SIZE
4557 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4561 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4564 okif->kf_structsize = sizeof(*okif);
4565 okif->kf_type = kif->kf_type;
4566 okif->kf_fd = kif->kf_fd;
4567 okif->kf_ref_count = kif->kf_ref_count;
4568 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4569 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4570 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4571 okif->kf_offset = kif->kf_offset;
4572 if (kif->kf_type == KF_TYPE_VNODE)
4573 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4575 okif->kf_vnode_type = KF_VTYPE_VNON;
4576 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4577 if (kif->kf_type == KF_TYPE_SOCKET) {
4578 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4579 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4580 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4581 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4582 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4584 okif->kf_sa_local.ss_family = AF_UNSPEC;
4585 okif->kf_sa_peer.ss_family = AF_UNSPEC;
4590 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4591 struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4596 PWDDESC_XUNLOCK(pdp);
4597 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4598 kinfo_to_okinfo(kif, okif);
4599 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4605 * Get per-process file descriptors for use by procstat(1), et al.
4608 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4610 struct kinfo_ofile *okif;
4611 struct kinfo_file *kif;
4612 struct filedesc *fdp;
4613 struct pwddesc *pdp;
4616 int error, i, *name;
4625 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4632 if (fdp == NULL || pdp == NULL) {
4637 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4638 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4640 pwd = pwd_hold_pwddesc(pdp);
4642 if (pwd->pwd_cdir != NULL)
4643 export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4645 if (pwd->pwd_rdir != NULL)
4646 export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4648 if (pwd->pwd_jdir != NULL)
4649 export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4652 PWDDESC_XUNLOCK(pdp);
4655 FILEDESC_SLOCK(fdp);
4656 if (refcount_load(&fdp->fd_refcnt) == 0)
4658 FILEDESC_FOREACH_FP(fdp, i, fp) {
4659 export_file_to_kinfo(fp, i, NULL, kif, fdp,
4660 KERN_FILEDESC_PACK_KINFO);
4661 FILEDESC_SUNLOCK(fdp);
4662 kinfo_to_okinfo(kif, okif);
4663 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4664 FILEDESC_SLOCK(fdp);
4665 if (error != 0 || refcount_load(&fdp->fd_refcnt) == 0)
4669 FILEDESC_SUNLOCK(fdp);
4677 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4678 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4679 "Process ofiledesc entries");
4680 #endif /* COMPAT_FREEBSD7 */
4683 vntype_to_kinfo(int vtype)
4688 } vtypes_table[] = {
4689 { VBAD, KF_VTYPE_VBAD },
4690 { VBLK, KF_VTYPE_VBLK },
4691 { VCHR, KF_VTYPE_VCHR },
4692 { VDIR, KF_VTYPE_VDIR },
4693 { VFIFO, KF_VTYPE_VFIFO },
4694 { VLNK, KF_VTYPE_VLNK },
4695 { VNON, KF_VTYPE_VNON },
4696 { VREG, KF_VTYPE_VREG },
4697 { VSOCK, KF_VTYPE_VSOCK }
4702 * Perform vtype translation.
4704 for (i = 0; i < nitems(vtypes_table); i++)
4705 if (vtypes_table[i].vtype == vtype)
4706 return (vtypes_table[i].kf_vtype);
4708 return (KF_VTYPE_UNKNOWN);
4711 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4712 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4713 "Process filedesc entries");
4716 * Store a process current working directory information to sbuf.
4718 * Takes a locked proc as argument, and returns with the proc unlocked.
4721 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
4723 struct pwddesc *pdp;
4725 struct export_fd_buf *efbuf;
4729 PROC_LOCK_ASSERT(p, MA_OWNED);
4736 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4740 efbuf->remainder = maxlen;
4744 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4745 cdir = pwd->pwd_cdir;
4750 error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4752 PWDDESC_XUNLOCK(pdp);
4754 free(efbuf, M_TEMP);
4759 * Get per-process current working directory.
4762 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4768 int error, error2, *name;
4776 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4777 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4778 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4783 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4784 error = kern_proc_cwd_out(p, &sb, maxlen);
4785 error2 = sbuf_finish(&sb);
4787 return (error != 0 ? error : error2);
4790 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4791 sysctl_kern_proc_cwd, "Process current working directory");
4795 * For the purposes of debugging, generate a human-readable string for the
4799 file_type_to_name(short type)
4827 case DTYPE_PROCDESC:
4831 case DTYPE_LINUXTFD:
4839 * For the purposes of debugging, identify a process (if any, perhaps one of
4840 * many) that references the passed file in its file descriptor array. Return
4843 static struct proc *
4844 file_to_first_proc(struct file *fp)
4846 struct filedesc *fdp;
4850 FOREACH_PROC_IN_SYSTEM(p) {
4851 if (p->p_state == PRS_NEW)
4856 for (n = 0; n < fdp->fd_nfiles; n++) {
4857 if (fp == fdp->fd_ofiles[n].fde_file)
4865 db_print_file(struct file *fp, int header)
4867 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4871 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4872 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4873 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4875 p = file_to_first_proc(fp);
4876 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4877 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4878 fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4879 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4884 DB_SHOW_COMMAND(file, db_show_file)
4889 db_printf("usage: show file <addr>\n");
4892 fp = (struct file *)addr;
4893 db_print_file(fp, 1);
4896 DB_SHOW_COMMAND(files, db_show_files)
4898 struct filedesc *fdp;
4905 FOREACH_PROC_IN_SYSTEM(p) {
4906 if (p->p_state == PRS_NEW)
4908 if ((fdp = p->p_fd) == NULL)
4910 for (n = 0; n < fdp->fd_nfiles; ++n) {
4911 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4913 db_print_file(fp, header);
4920 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4921 &maxfilesperproc, 0, "Maximum files allowed open per process");
4923 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4924 &maxfiles, 0, "Maximum number of files");
4926 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4927 &openfiles, 0, "System-wide number of open files");
4931 filelistinit(void *dummy)
4934 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4935 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4936 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4937 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4938 pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4939 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4941 * XXXMJG this is a temporary hack due to boot ordering issues against
4944 vfs_smr = uma_zone_get_smr(pwd_zone);
4945 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4947 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4949 /*-------------------------------------------------------------------*/
4952 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4953 int flags, struct thread *td)
4960 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4968 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4976 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4984 badfo_kqfilter(struct file *fp, struct knote *kn)
4991 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
4998 badfo_close(struct file *fp, struct thread *td)
5005 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5013 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5021 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5022 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5030 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
5036 struct fileops badfileops = {
5037 .fo_read = badfo_readwrite,
5038 .fo_write = badfo_readwrite,
5039 .fo_truncate = badfo_truncate,
5040 .fo_ioctl = badfo_ioctl,
5041 .fo_poll = badfo_poll,
5042 .fo_kqfilter = badfo_kqfilter,
5043 .fo_stat = badfo_stat,
5044 .fo_close = badfo_close,
5045 .fo_chmod = badfo_chmod,
5046 .fo_chown = badfo_chown,
5047 .fo_sendfile = badfo_sendfile,
5048 .fo_fill_kinfo = badfo_fill_kinfo,
5052 path_poll(struct file *fp, int events, struct ucred *active_cred,
5059 path_close(struct file *fp, struct thread *td)
5061 MPASS(fp->f_type == DTYPE_VNODE);
5062 fp->f_ops = &badfileops;
5067 struct fileops path_fileops = {
5068 .fo_read = badfo_readwrite,
5069 .fo_write = badfo_readwrite,
5070 .fo_truncate = badfo_truncate,
5071 .fo_ioctl = badfo_ioctl,
5072 .fo_poll = path_poll,
5073 .fo_kqfilter = vn_kqfilter_opath,
5074 .fo_stat = vn_statfile,
5075 .fo_close = path_close,
5076 .fo_chmod = badfo_chmod,
5077 .fo_chown = badfo_chown,
5078 .fo_sendfile = badfo_sendfile,
5079 .fo_fill_kinfo = vn_fill_kinfo,
5080 .fo_flags = DFLAG_PASSABLE,
5084 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
5085 int flags, struct thread *td)
5088 return (EOPNOTSUPP);
5092 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
5100 invfo_ioctl(struct file *fp, u_long com, void *data,
5101 struct ucred *active_cred, struct thread *td)
5108 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
5112 return (poll_no_poll(events));
5116 invfo_kqfilter(struct file *fp, struct knote *kn)
5123 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5131 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5139 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5140 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5147 /*-------------------------------------------------------------------*/
5150 * File Descriptor pseudo-device driver (/dev/fd/).
5152 * Opening minor device N dup()s the file (if any) connected to file
5153 * descriptor N belonging to the calling process. Note that this driver
5154 * consists of only the ``open()'' routine, because all subsequent
5155 * references to this file will be direct to the other driver.
5157 * XXX: we could give this one a cloning event handler if necessary.
5162 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
5166 * XXX Kludge: set curthread->td_dupfd to contain the value of the
5167 * the file descriptor being sought for duplication. The error
5168 * return ensures that the vnode for this device will be released
5169 * by vn_open. Open will detect this special error and take the
5170 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
5171 * will simply report the error.
5173 td->td_dupfd = dev2unit(dev);
5177 static struct cdevsw fildesc_cdevsw = {
5178 .d_version = D_VERSION,
5184 fildesc_drvinit(void *unused)
5188 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
5189 UID_ROOT, GID_WHEEL, 0666, "fd/0");
5190 make_dev_alias(dev, "stdin");
5191 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
5192 UID_ROOT, GID_WHEEL, 0666, "fd/1");
5193 make_dev_alias(dev, "stdout");
5194 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
5195 UID_ROOT, GID_WHEEL, 0666, "fd/2");
5196 make_dev_alias(dev, "stderr");
5199 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);