2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1989, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
42 #include "opt_capsicum.h"
44 #include "opt_ktrace.h"
46 #include <sys/param.h>
47 #include <sys/systm.h>
49 #include <sys/capsicum.h>
51 #include <sys/fcntl.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
66 #include <sys/protosw.h>
67 #include <sys/racct.h>
68 #include <sys/resourcevar.h>
70 #include <sys/signalvar.h>
74 #include <sys/syscallsubr.h>
75 #include <sys/sysctl.h>
76 #include <sys/sysproto.h>
77 #include <sys/unistd.h>
79 #include <sys/vnode.h>
81 #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_FILEDESC_TO_LEADER, "filedesc_to_leader",
95 "file desc to leader structures");
96 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
97 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
99 MALLOC_DECLARE(M_FADVISE);
101 static __read_mostly uma_zone_t file_zone;
102 static __read_mostly uma_zone_t filedesc0_zone;
104 static int closefp(struct filedesc *fdp, int fd, struct file *fp,
105 struct thread *td, int holdleaders);
106 static int fd_first_free(struct filedesc *fdp, int low, int size);
107 static int fd_last_used(struct filedesc *fdp, int size);
108 static void fdgrowtable(struct filedesc *fdp, int nfd);
109 static void fdgrowtable_exp(struct filedesc *fdp, int nfd);
110 static void fdunused(struct filedesc *fdp, int fd);
111 static void fdused(struct filedesc *fdp, int fd);
112 static int getmaxfd(struct thread *td);
113 static u_long *filecaps_copy_prep(const struct filecaps *src);
114 static void filecaps_copy_finish(const struct filecaps *src,
115 struct filecaps *dst, u_long *ioctls);
116 static u_long *filecaps_free_prep(struct filecaps *fcaps);
117 static void filecaps_free_finish(u_long *ioctls);
122 * - An array of open file descriptors (fd_ofiles)
123 * - An array of file flags (fd_ofileflags)
124 * - A bitmap recording which descriptors are in use (fd_map)
126 * A process starts out with NDFILE descriptors. The value of NDFILE has
127 * been selected based the historical limit of 20 open files, and an
128 * assumption that the majority of processes, especially short-lived
129 * processes like shells, will never need more.
131 * If this initial allocation is exhausted, a larger descriptor table and
132 * map are allocated dynamically, and the pointers in the process's struct
133 * filedesc are updated to point to those. This is repeated every time
134 * the process runs out of file descriptors (provided it hasn't hit its
137 * Since threads may hold references to individual descriptor table
138 * entries, the tables are never freed. Instead, they are placed on a
139 * linked list and freed only when the struct filedesc is released.
142 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
143 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
144 #define NDSLOT(x) ((x) / NDENTRIES)
145 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
146 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
149 * SLIST entry used to keep track of ofiles which must be reclaimed when
153 struct fdescenttbl *ft_table;
154 SLIST_ENTRY(freetable) ft_next;
158 * Initial allocation: a filedesc structure + the head of SLIST used to
159 * keep track of old ofiles + enough space for NDFILE descriptors.
162 struct fdescenttbl0 {
164 struct filedescent fdt_ofiles[NDFILE];
168 struct filedesc fd_fd;
169 SLIST_HEAD(, freetable) fd_free;
170 struct fdescenttbl0 fd_dfiles;
171 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
175 * Descriptor management.
177 static int __exclusive_cache_line openfiles; /* actual number of open files */
178 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
179 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
182 * If low >= size, just return low. Otherwise find the first zero bit in the
183 * given bitmap, starting at low and not exceeding size - 1. Return size if
187 fd_first_free(struct filedesc *fdp, int low, int size)
189 NDSLOTTYPE *map = fdp->fd_map;
197 if (low % NDENTRIES) {
198 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
199 if ((mask &= ~map[off]) != 0UL)
200 return (off * NDENTRIES + ffsl(mask) - 1);
203 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
204 if (map[off] != ~0UL)
205 return (off * NDENTRIES + ffsl(~map[off]) - 1);
210 * Find the highest non-zero bit in the given bitmap, starting at 0 and
211 * not exceeding size - 1. Return -1 if not found.
214 fd_last_used(struct filedesc *fdp, int size)
216 NDSLOTTYPE *map = fdp->fd_map;
221 if (size % NDENTRIES) {
222 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
223 if ((mask &= map[off]) != 0)
224 return (off * NDENTRIES + flsl(mask) - 1);
227 for (minoff = NDSLOT(0); off >= minoff; --off)
229 return (off * NDENTRIES + flsl(map[off]) - 1);
234 fdisused(struct filedesc *fdp, int fd)
237 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
238 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
240 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
244 * Mark a file descriptor as used.
247 fdused_init(struct filedesc *fdp, int fd)
250 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
252 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
256 fdused(struct filedesc *fdp, int fd)
259 FILEDESC_XLOCK_ASSERT(fdp);
261 fdused_init(fdp, fd);
262 if (fd > fdp->fd_lastfile)
263 fdp->fd_lastfile = fd;
264 if (fd == fdp->fd_freefile)
269 * Mark a file descriptor as unused.
272 fdunused(struct filedesc *fdp, int fd)
275 FILEDESC_XLOCK_ASSERT(fdp);
277 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
278 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
279 ("fd=%d is still in use", fd));
281 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
282 if (fd < fdp->fd_freefile)
283 fdp->fd_freefile = fd;
284 if (fd == fdp->fd_lastfile)
285 fdp->fd_lastfile = fd_last_used(fdp, fd);
289 * Free a file descriptor.
291 * Avoid some work if fdp is about to be destroyed.
294 fdefree_last(struct filedescent *fde)
297 filecaps_free(&fde->fde_caps);
301 fdfree(struct filedesc *fdp, int fd)
303 struct filedescent *fde;
305 fde = &fdp->fd_ofiles[fd];
307 seqc_write_begin(&fde->fde_seqc);
309 fde->fde_file = NULL;
311 seqc_write_end(&fde->fde_seqc);
318 pwd_ensure_dirs(void)
320 struct filedesc *fdp;
324 if (fdp->fd_cdir == NULL) {
325 fdp->fd_cdir = rootvnode;
328 if (fdp->fd_rdir == NULL) {
329 fdp->fd_rdir = rootvnode;
332 FILEDESC_XUNLOCK(fdp);
336 * System calls on descriptors.
338 #ifndef _SYS_SYSPROTO_H_
339 struct getdtablesize_args {
345 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
351 td->td_retval[0] = getmaxfd(td);
353 PROC_LOCK(td->td_proc);
354 lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
355 PROC_UNLOCK(td->td_proc);
356 if (lim < td->td_retval[0])
357 td->td_retval[0] = lim;
363 * Duplicate a file descriptor to a particular value.
365 * Note: keep in mind that a potential race condition exists when closing
366 * descriptors from a shared descriptor table (via rfork).
368 #ifndef _SYS_SYSPROTO_H_
376 sys_dup2(struct thread *td, struct dup2_args *uap)
379 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
383 * Duplicate a file descriptor.
385 #ifndef _SYS_SYSPROTO_H_
392 sys_dup(struct thread *td, struct dup_args *uap)
395 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
399 * The file control system call.
401 #ifndef _SYS_SYSPROTO_H_
410 sys_fcntl(struct thread *td, struct fcntl_args *uap)
413 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
417 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
431 * Convert old flock structure to new.
433 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
434 fl.l_start = ofl.l_start;
435 fl.l_len = ofl.l_len;
436 fl.l_pid = ofl.l_pid;
437 fl.l_type = ofl.l_type;
438 fl.l_whence = ofl.l_whence;
452 arg1 = (intptr_t)&fl;
458 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
459 arg1 = (intptr_t)&fl;
467 error = kern_fcntl(td, fd, newcmd, arg1);
470 if (cmd == F_OGETLK) {
471 ofl.l_start = fl.l_start;
472 ofl.l_len = fl.l_len;
473 ofl.l_pid = fl.l_pid;
474 ofl.l_type = fl.l_type;
475 ofl.l_whence = fl.l_whence;
476 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
477 } else if (cmd == F_GETLK) {
478 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
484 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
486 struct filedesc *fdp;
488 struct file *fp, *fp2;
489 struct filedescent *fde;
493 int error, flg, seals, tmp;
507 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
510 case F_DUPFD_CLOEXEC:
512 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
517 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
520 case F_DUP2FD_CLOEXEC:
522 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
528 fde = fdeget_locked(fdp, fd);
531 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
534 FILEDESC_SUNLOCK(fdp);
540 fde = fdeget_locked(fdp, fd);
542 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
543 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
546 FILEDESC_XUNLOCK(fdp);
550 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
553 td->td_retval[0] = OFLAGS(fp->f_flag);
558 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
562 tmp = flg = fp->f_flag;
564 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
565 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
566 tmp = fp->f_flag & FNONBLOCK;
567 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
572 tmp = fp->f_flag & FASYNC;
573 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
578 atomic_clear_int(&fp->f_flag, FNONBLOCK);
580 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
585 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
588 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
590 td->td_retval[0] = tmp;
595 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
599 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
604 error = priv_check(td, PRIV_NFS_LOCKD);
612 /* FALLTHROUGH F_SETLK */
616 flp = (struct flock *)arg;
617 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
622 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL);
625 if (fp->f_type != DTYPE_VNODE) {
631 if (flp->l_whence == SEEK_CUR) {
632 foffset = foffset_get(fp);
635 foffset > OFF_MAX - flp->l_start)) {
640 flp->l_start += foffset;
644 switch (flp->l_type) {
646 if ((fp->f_flag & FREAD) == 0) {
650 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
651 PROC_LOCK(p->p_leader);
652 p->p_leader->p_flag |= P_ADVLOCK;
653 PROC_UNLOCK(p->p_leader);
655 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
659 if ((fp->f_flag & FWRITE) == 0) {
663 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
664 PROC_LOCK(p->p_leader);
665 p->p_leader->p_flag |= P_ADVLOCK;
666 PROC_UNLOCK(p->p_leader);
668 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
672 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
676 if (flg != F_REMOTE) {
680 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
681 F_UNLCKSYS, flp, flg);
687 if (error != 0 || flp->l_type == F_UNLCK ||
688 flp->l_type == F_UNLCKSYS) {
694 * Check for a race with close.
696 * The vnode is now advisory locked (or unlocked, but this case
697 * is not really important) as the caller requested.
698 * We had to drop the filedesc lock, so we need to recheck if
699 * the descriptor is still valid, because if it was closed
700 * in the meantime we need to remove advisory lock from the
701 * vnode - close on any descriptor leading to an advisory
702 * locked vnode, removes that lock.
703 * We will return 0 on purpose in that case, as the result of
704 * successful advisory lock might have been externally visible
705 * already. This is fine - effectively we pretend to the caller
706 * that the closing thread was a bit slower and that the
707 * advisory lock succeeded before the close.
709 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2, NULL);
715 flp->l_whence = SEEK_SET;
718 flp->l_type = F_UNLCK;
719 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
720 F_UNLCK, flp, F_POSIX);
727 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL);
730 if (fp->f_type != DTYPE_VNODE) {
735 flp = (struct flock *)arg;
736 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
737 flp->l_type != F_UNLCK) {
742 if (flp->l_whence == SEEK_CUR) {
743 foffset = foffset_get(fp);
744 if ((flp->l_start > 0 &&
745 foffset > OFF_MAX - flp->l_start) ||
747 foffset < OFF_MIN - flp->l_start)) {
752 flp->l_start += foffset;
755 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
761 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp, NULL);
764 error = fo_add_seals(fp, arg);
769 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp, NULL);
772 if (fo_get_seals(fp, &seals) == 0)
773 td->td_retval[0] = seals;
780 arg = arg ? 128 * 1024: 0;
783 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp, NULL);
786 if (fp->f_type != DTYPE_VNODE) {
792 if (vp->v_type != VREG) {
799 * Exclusive lock synchronizes against f_seqcount reads and
800 * writes in sequential_heuristic().
802 error = vn_lock(vp, LK_EXCLUSIVE);
808 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
809 arg = MIN(arg, INT_MAX - bsize + 1);
810 fp->f_seqcount = MIN(IO_SEQMAX,
811 (arg + bsize - 1) / bsize);
812 atomic_set_int(&fp->f_flag, FRDAHEAD);
814 atomic_clear_int(&fp->f_flag, FRDAHEAD);
822 * Check if the vnode is part of a union stack (either the
823 * "union" flag from mount(2) or unionfs).
825 * Prior to introduction of this op libc's readdir would call
826 * fstatfs(2), in effect unnecessarily copying kilobytes of
827 * data just to check fs name and a mount flag.
829 * Fixing the code to handle everything in the kernel instead
830 * is a non-trivial endeavor and has low priority, thus this
831 * horrible kludge facilitates the current behavior in a much
832 * cheaper manner until someone(tm) sorts this out.
834 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp, NULL);
837 if (fp->f_type != DTYPE_VNODE) {
844 * Since we don't prevent dooming the vnode even non-null mp
845 * found can become immediately stale. This is tolerable since
846 * mount points are type-stable (providing safe memory access)
847 * and any vfs op on this vnode going forward will return an
848 * error (meaning return value in this case is meaningless).
850 mp = (struct mount *)atomic_load_ptr(&vp->v_mount);
851 if (__predict_false(mp == NULL)) {
856 td->td_retval[0] = 0;
857 if (mp->mnt_kern_flag & MNTK_UNIONFS ||
858 mp->mnt_flag & MNT_UNION)
859 td->td_retval[0] = 1;
871 getmaxfd(struct thread *td)
874 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
878 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
881 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
883 struct filedesc *fdp;
884 struct filedescent *oldfde, *newfde;
887 u_long *oioctls, *nioctls;
894 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
895 MPASS(mode < FDDUP_LASTMODE);
898 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
901 * Verify we have a valid descriptor to dup from and possibly to
902 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
903 * return EINVAL when the new descriptor is out of bounds.
908 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
909 maxfd = getmaxfd(td);
911 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
915 if (fget_locked(fdp, old) == NULL)
917 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
918 td->td_retval[0] = new;
919 if (flags & FDDUP_FLAG_CLOEXEC)
920 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
925 oldfde = &fdp->fd_ofiles[old];
926 if (!fhold(oldfde->fde_file))
930 * If the caller specified a file descriptor, make sure the file
931 * table is large enough to hold it, and grab it. Otherwise, just
932 * allocate a new descriptor the usual way.
937 if ((error = fdalloc(td, new, &new)) != 0) {
938 fdrop(oldfde->fde_file, td);
942 case FDDUP_MUSTREPLACE:
943 /* Target file descriptor must exist. */
944 if (fget_locked(fdp, new) == NULL) {
945 fdrop(oldfde->fde_file, td);
950 if (new >= fdp->fd_nfiles) {
952 * The resource limits are here instead of e.g.
953 * fdalloc(), because the file descriptor table may be
954 * shared between processes, so we can't really use
955 * racct_add()/racct_sub(). Instead of counting the
956 * number of actually allocated descriptors, just put
957 * the limit on the size of the file descriptor table.
960 if (RACCT_ENABLED()) {
961 error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
964 fdrop(oldfde->fde_file, td);
969 fdgrowtable_exp(fdp, new + 1);
971 if (!fdisused(fdp, new))
975 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
978 KASSERT(old != new, ("new fd is same as old"));
980 newfde = &fdp->fd_ofiles[new];
981 delfp = newfde->fde_file;
983 oioctls = filecaps_free_prep(&newfde->fde_caps);
984 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
987 * Duplicate the source descriptor.
990 seqc_write_begin(&newfde->fde_seqc);
992 memcpy(newfde, oldfde, fde_change_size);
993 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
995 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
996 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
998 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
1000 seqc_write_end(&newfde->fde_seqc);
1002 td->td_retval[0] = new;
1006 if (delfp != NULL) {
1007 (void) closefp(fdp, new, delfp, td, 1);
1008 FILEDESC_UNLOCK_ASSERT(fdp);
1011 FILEDESC_XUNLOCK(fdp);
1014 filecaps_free_finish(oioctls);
1019 * If sigio is on the list associated with a process or process group,
1020 * disable signalling from the device, remove sigio from the list and
1024 funsetown(struct sigio **sigiop)
1026 struct sigio *sigio;
1028 if (*sigiop == NULL)
1032 if (sigio == NULL) {
1036 *(sigio->sio_myref) = NULL;
1037 if ((sigio)->sio_pgid < 0) {
1038 struct pgrp *pg = (sigio)->sio_pgrp;
1040 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
1041 sigio, sio_pgsigio);
1044 struct proc *p = (sigio)->sio_proc;
1046 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
1047 sigio, sio_pgsigio);
1051 crfree(sigio->sio_ucred);
1052 free(sigio, M_SIGIO);
1056 * Free a list of sigio structures.
1057 * We only need to lock the SIGIO_LOCK because we have made ourselves
1058 * inaccessible to callers of fsetown and therefore do not need to lock
1059 * the proc or pgrp struct for the list manipulation.
1062 funsetownlst(struct sigiolst *sigiolst)
1066 struct sigio *sigio;
1068 sigio = SLIST_FIRST(sigiolst);
1075 * Every entry of the list should belong
1076 * to a single proc or pgrp.
1078 if (sigio->sio_pgid < 0) {
1079 pg = sigio->sio_pgrp;
1080 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
1081 } else /* if (sigio->sio_pgid > 0) */ {
1082 p = sigio->sio_proc;
1083 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1087 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1088 *(sigio->sio_myref) = NULL;
1090 KASSERT(sigio->sio_pgid < 0,
1091 ("Proc sigio in pgrp sigio list"));
1092 KASSERT(sigio->sio_pgrp == pg,
1093 ("Bogus pgrp in sigio list"));
1095 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1098 } else /* if (p != NULL) */ {
1099 KASSERT(sigio->sio_pgid > 0,
1100 ("Pgrp sigio in proc sigio list"));
1101 KASSERT(sigio->sio_proc == p,
1102 ("Bogus proc in sigio list"));
1104 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1109 crfree(sigio->sio_ucred);
1110 free(sigio, M_SIGIO);
1117 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1119 * After permission checking, add a sigio structure to the sigio list for
1120 * the process or process group.
1123 fsetown(pid_t pgid, struct sigio **sigiop)
1127 struct sigio *sigio;
1137 /* Allocate and fill in the new sigio out of locks. */
1138 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1139 sigio->sio_pgid = pgid;
1140 sigio->sio_ucred = crhold(curthread->td_ucred);
1141 sigio->sio_myref = sigiop;
1143 sx_slock(&proctree_lock);
1152 * Policy - Don't allow a process to FSETOWN a process
1153 * in another session.
1155 * Remove this test to allow maximum flexibility or
1156 * restrict FSETOWN to the current process or process
1157 * group for maximum safety.
1160 if (proc->p_session != curthread->td_proc->p_session) {
1166 } else /* if (pgid < 0) */ {
1167 pgrp = pgfind(-pgid);
1175 * Policy - Don't allow a process to FSETOWN a process
1176 * in another session.
1178 * Remove this test to allow maximum flexibility or
1179 * restrict FSETOWN to the current process or process
1180 * group for maximum safety.
1182 if (pgrp->pg_session != curthread->td_proc->p_session) {
1193 * Since funsetownlst() is called without the proctree
1194 * locked, we need to check for P_WEXIT.
1195 * XXX: is ESRCH correct?
1197 if ((proc->p_flag & P_WEXIT) != 0) {
1202 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1203 sigio->sio_proc = proc;
1207 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1208 sigio->sio_pgrp = pgrp;
1211 sx_sunlock(&proctree_lock);
1218 sx_sunlock(&proctree_lock);
1219 crfree(sigio->sio_ucred);
1220 free(sigio, M_SIGIO);
1225 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1228 fgetown(struct sigio **sigiop)
1233 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1239 * Function drops the filedesc lock on return.
1242 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1247 FILEDESC_XLOCK_ASSERT(fdp);
1250 if (td->td_proc->p_fdtol != NULL) {
1252 * Ask fdfree() to sleep to ensure that all relevant
1253 * process leaders can be traversed in closef().
1255 fdp->fd_holdleaderscount++;
1262 * We now hold the fp reference that used to be owned by the
1263 * descriptor array. We have to unlock the FILEDESC *AFTER*
1264 * knote_fdclose to prevent a race of the fd getting opened, a knote
1265 * added, and deleteing a knote for the new fd.
1267 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1268 knote_fdclose(td, fd);
1271 * We need to notify mqueue if the object is of type mqueue.
1273 if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1274 mq_fdclose(td, fd, fp);
1275 FILEDESC_XUNLOCK(fdp);
1277 error = closef(fp, td);
1279 FILEDESC_XLOCK(fdp);
1280 fdp->fd_holdleaderscount--;
1281 if (fdp->fd_holdleaderscount == 0 &&
1282 fdp->fd_holdleaderswakeup != 0) {
1283 fdp->fd_holdleaderswakeup = 0;
1284 wakeup(&fdp->fd_holdleaderscount);
1286 FILEDESC_XUNLOCK(fdp);
1292 * Close a file descriptor.
1294 #ifndef _SYS_SYSPROTO_H_
1301 sys_close(struct thread *td, struct close_args *uap)
1304 return (kern_close(td, uap->fd));
1308 kern_close(struct thread *td, int fd)
1310 struct filedesc *fdp;
1313 fdp = td->td_proc->p_fd;
1315 AUDIT_SYSCLOSE(td, fd);
1317 FILEDESC_XLOCK(fdp);
1318 if ((fp = fget_locked(fdp, fd)) == NULL) {
1319 FILEDESC_XUNLOCK(fdp);
1324 /* closefp() drops the FILEDESC lock for us. */
1325 return (closefp(fdp, fd, fp, td, 1));
1329 * Close open file descriptors.
1331 #ifndef _SYS_SYSPROTO_H_
1332 struct closefrom_args {
1338 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1340 struct filedesc *fdp;
1343 fdp = td->td_proc->p_fd;
1344 AUDIT_ARG_FD(uap->lowfd);
1347 * Treat negative starting file descriptor values identical to
1348 * closefrom(0) which closes all files.
1352 FILEDESC_SLOCK(fdp);
1353 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1354 if (fdp->fd_ofiles[fd].fde_file != NULL) {
1355 FILEDESC_SUNLOCK(fdp);
1356 (void)kern_close(td, fd);
1357 FILEDESC_SLOCK(fdp);
1360 FILEDESC_SUNLOCK(fdp);
1364 #if defined(COMPAT_43)
1366 * Return status information about a file descriptor.
1368 #ifndef _SYS_SYSPROTO_H_
1369 struct ofstat_args {
1376 ofstat(struct thread *td, struct ofstat_args *uap)
1382 error = kern_fstat(td, uap->fd, &ub);
1385 error = copyout(&oub, uap->sb, sizeof(oub));
1389 #endif /* COMPAT_43 */
1391 #if defined(COMPAT_FREEBSD11)
1393 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1396 struct freebsd11_stat osb;
1399 error = kern_fstat(td, uap->fd, &sb);
1402 error = freebsd11_cvtstat(&sb, &osb);
1404 error = copyout(&osb, uap->sb, sizeof(osb));
1407 #endif /* COMPAT_FREEBSD11 */
1410 * Return status information about a file descriptor.
1412 #ifndef _SYS_SYSPROTO_H_
1420 sys_fstat(struct thread *td, struct fstat_args *uap)
1425 error = kern_fstat(td, uap->fd, &ub);
1427 error = copyout(&ub, uap->sb, sizeof(ub));
1432 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1439 error = fget(td, fd, &cap_fstat_rights, &fp);
1440 if (__predict_false(error != 0))
1443 AUDIT_ARG_FILE(td->td_proc, fp);
1445 error = fo_stat(fp, sbp, td->td_ucred, td);
1447 #ifdef __STAT_TIME_T_EXT
1449 sbp->st_atim_ext = 0;
1450 sbp->st_mtim_ext = 0;
1451 sbp->st_ctim_ext = 0;
1452 sbp->st_btim_ext = 0;
1456 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1462 #if defined(COMPAT_FREEBSD11)
1464 * Return status information about a file descriptor.
1466 #ifndef _SYS_SYSPROTO_H_
1467 struct freebsd11_nfstat_args {
1474 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1480 error = kern_fstat(td, uap->fd, &ub);
1482 freebsd11_cvtnstat(&ub, &nub);
1483 error = copyout(&nub, uap->sb, sizeof(nub));
1487 #endif /* COMPAT_FREEBSD11 */
1490 * Return pathconf information about a file descriptor.
1492 #ifndef _SYS_SYSPROTO_H_
1493 struct fpathconf_args {
1500 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1505 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1507 td->td_retval[0] = value;
1512 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1518 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1522 if (name == _PC_ASYNC_IO) {
1523 *valuep = _POSIX_ASYNCHRONOUS_IO;
1528 vn_lock(vp, LK_SHARED | LK_RETRY);
1529 error = VOP_PATHCONF(vp, name, valuep);
1531 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1532 if (name != _PC_PIPE_BUF) {
1547 * Copy filecaps structure allocating memory for ioctls array if needed.
1549 * The last parameter indicates whether the fdtable is locked. If it is not and
1550 * ioctls are encountered, copying fails and the caller must lock the table.
1552 * Note that if the table was not locked, the caller has to check the relevant
1553 * sequence counter to determine whether the operation was successful.
1556 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1560 if (src->fc_ioctls != NULL && !locked)
1562 memcpy(dst, src, sizeof(*src));
1563 if (src->fc_ioctls == NULL)
1566 KASSERT(src->fc_nioctls > 0,
1567 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1569 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1570 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1571 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1576 filecaps_copy_prep(const struct filecaps *src)
1581 if (__predict_true(src->fc_ioctls == NULL))
1584 KASSERT(src->fc_nioctls > 0,
1585 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1587 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1588 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1593 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1599 if (__predict_true(src->fc_ioctls == NULL)) {
1600 MPASS(ioctls == NULL);
1604 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1605 dst->fc_ioctls = ioctls;
1606 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1610 * Move filecaps structure to the new place and clear the old place.
1613 filecaps_move(struct filecaps *src, struct filecaps *dst)
1617 bzero(src, sizeof(*src));
1621 * Fill the given filecaps structure with full rights.
1624 filecaps_fill(struct filecaps *fcaps)
1627 CAP_ALL(&fcaps->fc_rights);
1628 fcaps->fc_ioctls = NULL;
1629 fcaps->fc_nioctls = -1;
1630 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1634 * Free memory allocated within filecaps structure.
1637 filecaps_free(struct filecaps *fcaps)
1640 free(fcaps->fc_ioctls, M_FILECAPS);
1641 bzero(fcaps, sizeof(*fcaps));
1645 filecaps_free_prep(struct filecaps *fcaps)
1649 ioctls = fcaps->fc_ioctls;
1650 bzero(fcaps, sizeof(*fcaps));
1655 filecaps_free_finish(u_long *ioctls)
1658 free(ioctls, M_FILECAPS);
1662 * Validate the given filecaps structure.
1665 filecaps_validate(const struct filecaps *fcaps, const char *func)
1668 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1669 ("%s: invalid rights", func));
1670 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1671 ("%s: invalid fcntls", func));
1672 KASSERT(fcaps->fc_fcntls == 0 ||
1673 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1674 ("%s: fcntls without CAP_FCNTL", func));
1675 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1676 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1677 ("%s: invalid ioctls", func));
1678 KASSERT(fcaps->fc_nioctls == 0 ||
1679 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1680 ("%s: ioctls without CAP_IOCTL", func));
1684 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1688 FILEDESC_XLOCK_ASSERT(fdp);
1690 nfd1 = fdp->fd_nfiles * 2;
1693 fdgrowtable(fdp, nfd1);
1697 * Grow the file table to accommodate (at least) nfd descriptors.
1700 fdgrowtable(struct filedesc *fdp, int nfd)
1702 struct filedesc0 *fdp0;
1703 struct freetable *ft;
1704 struct fdescenttbl *ntable;
1705 struct fdescenttbl *otable;
1706 int nnfiles, onfiles;
1707 NDSLOTTYPE *nmap, *omap;
1710 * If lastfile is -1 this struct filedesc was just allocated and we are
1711 * growing it to accommodate for the one we are going to copy from. There
1712 * is no need to have a lock on this one as it's not visible to anyone.
1714 if (fdp->fd_lastfile != -1)
1715 FILEDESC_XLOCK_ASSERT(fdp);
1717 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1719 /* save old values */
1720 onfiles = fdp->fd_nfiles;
1721 otable = fdp->fd_files;
1724 /* compute the size of the new table */
1725 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1726 if (nnfiles <= onfiles)
1727 /* the table is already large enough */
1731 * Allocate a new table. We need enough space for the number of
1732 * entries, file entries themselves and the struct freetable we will use
1733 * when we decommission the table and place it on the freelist.
1734 * We place the struct freetable in the middle so we don't have
1735 * to worry about padding.
1737 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1738 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1739 sizeof(struct freetable),
1740 M_FILEDESC, M_ZERO | M_WAITOK);
1741 /* copy the old data */
1742 ntable->fdt_nfiles = nnfiles;
1743 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1744 onfiles * sizeof(ntable->fdt_ofiles[0]));
1747 * Allocate a new map only if the old is not large enough. It will
1748 * grow at a slower rate than the table as it can map more
1749 * entries than the table can hold.
1751 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1752 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1754 /* copy over the old data and update the pointer */
1755 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1760 * Make sure that ntable is correctly initialized before we replace
1761 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1764 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1767 * Do not free the old file table, as some threads may still
1768 * reference entries within it. Instead, place it on a freelist
1769 * which will be processed when the struct filedesc is released.
1771 * Note that if onfiles == NDFILE, we're dealing with the original
1772 * static allocation contained within (struct filedesc0 *)fdp,
1773 * which must not be freed.
1775 if (onfiles > NDFILE) {
1776 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1777 fdp0 = (struct filedesc0 *)fdp;
1778 ft->ft_table = otable;
1779 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1782 * The map does not have the same possibility of threads still
1783 * holding references to it. So always free it as long as it
1784 * does not reference the original static allocation.
1786 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1787 free(omap, M_FILEDESC);
1791 * Allocate a file descriptor for the process.
1794 fdalloc(struct thread *td, int minfd, int *result)
1796 struct proc *p = td->td_proc;
1797 struct filedesc *fdp = p->p_fd;
1798 int fd, maxfd, allocfd;
1803 FILEDESC_XLOCK_ASSERT(fdp);
1805 if (fdp->fd_freefile > minfd)
1806 minfd = fdp->fd_freefile;
1808 maxfd = getmaxfd(td);
1811 * Search the bitmap for a free descriptor starting at minfd.
1812 * If none is found, grow the file table.
1814 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1817 if (fd >= fdp->fd_nfiles) {
1818 allocfd = min(fd * 2, maxfd);
1820 if (RACCT_ENABLED()) {
1821 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1827 * fd is already equal to first free descriptor >= minfd, so
1828 * we only need to grow the table and we are done.
1830 fdgrowtable_exp(fdp, allocfd);
1834 * Perform some sanity checks, then mark the file descriptor as
1835 * used and return it to the caller.
1837 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1838 ("invalid descriptor %d", fd));
1839 KASSERT(!fdisused(fdp, fd),
1840 ("fd_first_free() returned non-free descriptor"));
1841 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1842 ("file descriptor isn't free"));
1849 * Allocate n file descriptors for the process.
1852 fdallocn(struct thread *td, int minfd, int *fds, int n)
1854 struct proc *p = td->td_proc;
1855 struct filedesc *fdp = p->p_fd;
1858 FILEDESC_XLOCK_ASSERT(fdp);
1860 for (i = 0; i < n; i++)
1861 if (fdalloc(td, 0, &fds[i]) != 0)
1865 for (i--; i >= 0; i--)
1866 fdunused(fdp, fds[i]);
1874 * Create a new open file structure and allocate a file descriptor for the
1875 * process that refers to it. We add one reference to the file for the
1876 * descriptor table and one reference for resultfp. This is to prevent us
1877 * being preempted and the entry in the descriptor table closed after we
1878 * release the FILEDESC lock.
1881 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1882 struct filecaps *fcaps)
1887 error = falloc_noinstall(td, &fp);
1889 return (error); /* no reference held on error */
1891 error = finstall(td, fp, &fd, flags, fcaps);
1893 fdrop(fp, td); /* one reference (fp only) */
1897 if (resultfp != NULL)
1898 *resultfp = fp; /* copy out result */
1900 fdrop(fp, td); /* release local reference */
1902 if (resultfd != NULL)
1909 * Create a new open file structure without allocating a file descriptor.
1912 falloc_noinstall(struct thread *td, struct file **resultfp)
1915 int maxuserfiles = maxfiles - (maxfiles / 20);
1917 static struct timeval lastfail;
1920 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1922 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1923 if ((openfiles_new >= maxuserfiles &&
1924 priv_check(td, PRIV_MAXFILES) != 0) ||
1925 openfiles_new >= maxfiles) {
1926 atomic_subtract_int(&openfiles, 1);
1927 if (ppsratecheck(&lastfail, &curfail, 1)) {
1928 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1929 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1933 fp = uma_zalloc(file_zone, M_WAITOK);
1934 bzero(fp, sizeof(*fp));
1935 refcount_init(&fp->f_count, 1);
1936 fp->f_cred = crhold(td->td_ucred);
1937 fp->f_ops = &badfileops;
1943 * Install a file in a file descriptor table.
1946 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1947 struct filecaps *fcaps)
1949 struct filedescent *fde;
1953 filecaps_validate(fcaps, __func__);
1954 FILEDESC_XLOCK_ASSERT(fdp);
1956 fde = &fdp->fd_ofiles[fd];
1958 seqc_write_begin(&fde->fde_seqc);
1961 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1963 filecaps_move(fcaps, &fde->fde_caps);
1965 filecaps_fill(&fde->fde_caps);
1967 seqc_write_end(&fde->fde_seqc);
1972 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1973 struct filecaps *fcaps)
1975 struct filedesc *fdp = td->td_proc->p_fd;
1982 FILEDESC_XLOCK(fdp);
1983 if ((error = fdalloc(td, 0, fd))) {
1984 FILEDESC_XUNLOCK(fdp);
1988 _finstall(fdp, fp, *fd, flags, fcaps);
1989 FILEDESC_XUNLOCK(fdp);
1994 * Build a new filedesc structure from another.
1995 * Copy the current, root, and jail root vnode references.
1997 * If fdp is not NULL, return with it shared locked.
2000 fdinit(struct filedesc *fdp, bool prepfiles)
2002 struct filedesc0 *newfdp0;
2003 struct filedesc *newfdp;
2005 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2006 newfdp = &newfdp0->fd_fd;
2008 /* Create the file descriptor table. */
2009 FILEDESC_LOCK_INIT(newfdp);
2010 refcount_init(&newfdp->fd_refcnt, 1);
2011 refcount_init(&newfdp->fd_holdcnt, 1);
2012 newfdp->fd_cmask = CMASK;
2013 newfdp->fd_map = newfdp0->fd_dmap;
2014 newfdp->fd_lastfile = -1;
2015 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2016 newfdp->fd_files->fdt_nfiles = NDFILE;
2021 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
2022 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
2024 FILEDESC_SLOCK(fdp);
2025 newfdp->fd_cdir = fdp->fd_cdir;
2026 if (newfdp->fd_cdir)
2027 vrefact(newfdp->fd_cdir);
2028 newfdp->fd_rdir = fdp->fd_rdir;
2029 if (newfdp->fd_rdir)
2030 vrefact(newfdp->fd_rdir);
2031 newfdp->fd_jdir = fdp->fd_jdir;
2032 if (newfdp->fd_jdir)
2033 vrefact(newfdp->fd_jdir);
2036 FILEDESC_SUNLOCK(fdp);
2038 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
2039 FILEDESC_SUNLOCK(fdp);
2040 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
2041 FILEDESC_SLOCK(fdp);
2048 static struct filedesc *
2049 fdhold(struct proc *p)
2051 struct filedesc *fdp;
2053 PROC_LOCK_ASSERT(p, MA_OWNED);
2056 refcount_acquire(&fdp->fd_holdcnt);
2061 fddrop(struct filedesc *fdp)
2064 if (fdp->fd_holdcnt > 1) {
2065 if (refcount_release(&fdp->fd_holdcnt) == 0)
2069 FILEDESC_LOCK_DESTROY(fdp);
2070 uma_zfree(filedesc0_zone, fdp);
2074 * Share a filedesc structure.
2077 fdshare(struct filedesc *fdp)
2080 refcount_acquire(&fdp->fd_refcnt);
2085 * Unshare a filedesc structure, if necessary by making a copy
2088 fdunshare(struct thread *td)
2090 struct filedesc *tmp;
2091 struct proc *p = td->td_proc;
2093 if (p->p_fd->fd_refcnt == 1)
2096 tmp = fdcopy(p->p_fd);
2102 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2106 td->td_proc->p_fd = fdp;
2110 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2111 * this is to ease callers, not catch errors.
2114 fdcopy(struct filedesc *fdp)
2116 struct filedesc *newfdp;
2117 struct filedescent *nfde, *ofde;
2122 newfdp = fdinit(fdp, true);
2123 /* copy all passable descriptors (i.e. not kqueue) */
2124 newfdp->fd_freefile = -1;
2125 for (i = 0; i <= fdp->fd_lastfile; ++i) {
2126 ofde = &fdp->fd_ofiles[i];
2127 if (ofde->fde_file == NULL ||
2128 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2129 !fhold(ofde->fde_file)) {
2130 if (newfdp->fd_freefile == -1)
2131 newfdp->fd_freefile = i;
2134 nfde = &newfdp->fd_ofiles[i];
2136 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2137 fdused_init(newfdp, i);
2138 newfdp->fd_lastfile = i;
2140 if (newfdp->fd_freefile == -1)
2141 newfdp->fd_freefile = i;
2142 newfdp->fd_cmask = fdp->fd_cmask;
2143 FILEDESC_SUNLOCK(fdp);
2148 * Copies a filedesc structure, while remapping all file descriptors
2149 * stored inside using a translation table.
2151 * File descriptors are copied over to the new file descriptor table,
2152 * regardless of whether the close-on-exec flag is set.
2155 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2156 struct filedesc **ret)
2158 struct filedesc *newfdp;
2159 struct filedescent *nfde, *ofde;
2164 newfdp = fdinit(fdp, true);
2165 if (nfds > fdp->fd_lastfile + 1) {
2166 /* New table cannot be larger than the old one. */
2170 /* Copy all passable descriptors (i.e. not kqueue). */
2171 newfdp->fd_freefile = nfds;
2172 for (i = 0; i < nfds; ++i) {
2173 if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2174 /* File descriptor out of bounds. */
2178 ofde = &fdp->fd_ofiles[fds[i]];
2179 if (ofde->fde_file == NULL) {
2180 /* Unused file descriptor. */
2184 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2185 /* File descriptor cannot be passed. */
2189 if (!fhold(nfde->fde_file)) {
2193 nfde = &newfdp->fd_ofiles[i];
2195 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2196 fdused_init(newfdp, i);
2197 newfdp->fd_lastfile = i;
2199 newfdp->fd_cmask = fdp->fd_cmask;
2200 FILEDESC_SUNLOCK(fdp);
2204 FILEDESC_SUNLOCK(fdp);
2205 fdescfree_remapped(newfdp);
2210 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2211 * one of processes using it exits) and the table used to be shared.
2214 fdclearlocks(struct thread *td)
2216 struct filedesc *fdp;
2217 struct filedesc_to_leader *fdtol;
2227 MPASS(fdtol != NULL);
2229 FILEDESC_XLOCK(fdp);
2230 KASSERT(fdtol->fdl_refcount > 0,
2231 ("filedesc_to_refcount botch: fdl_refcount=%d",
2232 fdtol->fdl_refcount));
2233 if (fdtol->fdl_refcount == 1 &&
2234 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2235 for (i = 0; i <= fdp->fd_lastfile; i++) {
2236 fp = fdp->fd_ofiles[i].fde_file;
2237 if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2240 FILEDESC_XUNLOCK(fdp);
2241 lf.l_whence = SEEK_SET;
2244 lf.l_type = F_UNLCK;
2246 (void) VOP_ADVLOCK(vp,
2247 (caddr_t)p->p_leader, F_UNLCK,
2249 FILEDESC_XLOCK(fdp);
2254 if (fdtol->fdl_refcount == 1) {
2255 if (fdp->fd_holdleaderscount > 0 &&
2256 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2258 * close() or kern_dup() has cleared a reference
2259 * in a shared file descriptor table.
2261 fdp->fd_holdleaderswakeup = 1;
2262 sx_sleep(&fdp->fd_holdleaderscount,
2263 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2266 if (fdtol->fdl_holdcount > 0) {
2268 * Ensure that fdtol->fdl_leader remains
2269 * valid in closef().
2271 fdtol->fdl_wakeup = 1;
2272 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2277 fdtol->fdl_refcount--;
2278 if (fdtol->fdl_refcount == 0 &&
2279 fdtol->fdl_holdcount == 0) {
2280 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2281 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2285 FILEDESC_XUNLOCK(fdp);
2287 free(fdtol, M_FILEDESC_TO_LEADER);
2291 * Release a filedesc structure.
2294 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2296 struct filedesc0 *fdp0;
2297 struct freetable *ft, *tft;
2298 struct filedescent *fde;
2302 for (i = 0; i <= fdp->fd_lastfile; i++) {
2303 fde = &fdp->fd_ofiles[i];
2308 (void) closef(fp, td);
2314 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2315 free(fdp->fd_map, M_FILEDESC);
2316 if (fdp->fd_nfiles > NDFILE)
2317 free(fdp->fd_files, M_FILEDESC);
2319 fdp0 = (struct filedesc0 *)fdp;
2320 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2321 free(ft->ft_table, M_FILEDESC);
2327 fdescfree(struct thread *td)
2330 struct filedesc *fdp;
2331 struct vnode *cdir, *jdir, *rdir;
2338 if (RACCT_ENABLED())
2339 racct_set_unlocked(p, RACCT_NOFILE, 0);
2342 if (p->p_fdtol != NULL)
2349 if (refcount_release(&fdp->fd_refcnt) == 0)
2352 FILEDESC_XLOCK(fdp);
2353 cdir = fdp->fd_cdir;
2354 fdp->fd_cdir = NULL;
2355 rdir = fdp->fd_rdir;
2356 fdp->fd_rdir = NULL;
2357 jdir = fdp->fd_jdir;
2358 fdp->fd_jdir = NULL;
2359 FILEDESC_XUNLOCK(fdp);
2368 fdescfree_fds(td, fdp, 1);
2372 fdescfree_remapped(struct filedesc *fdp)
2375 if (fdp->fd_cdir != NULL)
2376 vrele(fdp->fd_cdir);
2377 if (fdp->fd_rdir != NULL)
2378 vrele(fdp->fd_rdir);
2379 if (fdp->fd_jdir != NULL)
2380 vrele(fdp->fd_jdir);
2382 fdescfree_fds(curthread, fdp, 0);
2386 * For setugid programs, we don't want to people to use that setugidness
2387 * to generate error messages which write to a file which otherwise would
2388 * otherwise be off-limits to the process. We check for filesystems where
2389 * the vnode can change out from under us after execve (like [lin]procfs).
2391 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2392 * sufficient. We also don't check for setugidness since we know we are.
2395 is_unsafe(struct file *fp)
2399 if (fp->f_type != DTYPE_VNODE)
2403 return ((vp->v_vflag & VV_PROCDEP) != 0);
2407 * Make this setguid thing safe, if at all possible.
2410 fdsetugidsafety(struct thread *td)
2412 struct filedesc *fdp;
2416 fdp = td->td_proc->p_fd;
2417 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2418 MPASS(fdp->fd_nfiles >= 3);
2419 for (i = 0; i <= 2; i++) {
2420 fp = fdp->fd_ofiles[i].fde_file;
2421 if (fp != NULL && is_unsafe(fp)) {
2422 FILEDESC_XLOCK(fdp);
2423 knote_fdclose(td, i);
2425 * NULL-out descriptor prior to close to avoid
2426 * a race while close blocks.
2429 FILEDESC_XUNLOCK(fdp);
2430 (void) closef(fp, td);
2436 * If a specific file object occupies a specific file descriptor, close the
2437 * file descriptor entry and drop a reference on the file object. This is a
2438 * convenience function to handle a subsequent error in a function that calls
2439 * falloc() that handles the race that another thread might have closed the
2440 * file descriptor out from under the thread creating the file object.
2443 fdclose(struct thread *td, struct file *fp, int idx)
2445 struct filedesc *fdp = td->td_proc->p_fd;
2447 FILEDESC_XLOCK(fdp);
2448 if (fdp->fd_ofiles[idx].fde_file == fp) {
2450 FILEDESC_XUNLOCK(fdp);
2453 FILEDESC_XUNLOCK(fdp);
2457 * Close any files on exec?
2460 fdcloseexec(struct thread *td)
2462 struct filedesc *fdp;
2463 struct filedescent *fde;
2467 fdp = td->td_proc->p_fd;
2468 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2469 for (i = 0; i <= fdp->fd_lastfile; i++) {
2470 fde = &fdp->fd_ofiles[i];
2472 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2473 (fde->fde_flags & UF_EXCLOSE))) {
2474 FILEDESC_XLOCK(fdp);
2476 (void) closefp(fdp, i, fp, td, 0);
2477 FILEDESC_UNLOCK_ASSERT(fdp);
2483 * It is unsafe for set[ug]id processes to be started with file
2484 * descriptors 0..2 closed, as these descriptors are given implicit
2485 * significance in the Standard C library. fdcheckstd() will create a
2486 * descriptor referencing /dev/null for each of stdin, stdout, and
2487 * stderr that is not already open.
2490 fdcheckstd(struct thread *td)
2492 struct filedesc *fdp;
2494 int i, error, devnull;
2496 fdp = td->td_proc->p_fd;
2497 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2498 MPASS(fdp->fd_nfiles >= 3);
2500 for (i = 0; i <= 2; i++) {
2501 if (fdp->fd_ofiles[i].fde_file != NULL)
2504 save = td->td_retval[0];
2505 if (devnull != -1) {
2506 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2508 error = kern_openat(td, AT_FDCWD, "/dev/null",
2509 UIO_SYSSPACE, O_RDWR, 0);
2511 devnull = td->td_retval[0];
2512 KASSERT(devnull == i, ("we didn't get our fd"));
2515 td->td_retval[0] = save;
2523 * Internal form of close. Decrement reference count on file structure.
2524 * Note: td may be NULL when closing a file that was being passed in a
2528 closef(struct file *fp, struct thread *td)
2532 struct filedesc_to_leader *fdtol;
2533 struct filedesc *fdp;
2536 * POSIX record locking dictates that any close releases ALL
2537 * locks owned by this process. This is handled by setting
2538 * a flag in the unlock to free ONLY locks obeying POSIX
2539 * semantics, and not to free BSD-style file locks.
2540 * If the descriptor was in a message, POSIX-style locks
2541 * aren't passed with the descriptor, and the thread pointer
2542 * will be NULL. Callers should be careful only to pass a
2543 * NULL thread pointer when there really is no owning
2544 * context that might have locks, or the locks will be
2547 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2549 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2550 lf.l_whence = SEEK_SET;
2553 lf.l_type = F_UNLCK;
2554 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2555 F_UNLCK, &lf, F_POSIX);
2557 fdtol = td->td_proc->p_fdtol;
2558 if (fdtol != NULL) {
2560 * Handle special case where file descriptor table is
2561 * shared between multiple process leaders.
2563 fdp = td->td_proc->p_fd;
2564 FILEDESC_XLOCK(fdp);
2565 for (fdtol = fdtol->fdl_next;
2566 fdtol != td->td_proc->p_fdtol;
2567 fdtol = fdtol->fdl_next) {
2568 if ((fdtol->fdl_leader->p_flag &
2571 fdtol->fdl_holdcount++;
2572 FILEDESC_XUNLOCK(fdp);
2573 lf.l_whence = SEEK_SET;
2576 lf.l_type = F_UNLCK;
2578 (void) VOP_ADVLOCK(vp,
2579 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2581 FILEDESC_XLOCK(fdp);
2582 fdtol->fdl_holdcount--;
2583 if (fdtol->fdl_holdcount == 0 &&
2584 fdtol->fdl_wakeup != 0) {
2585 fdtol->fdl_wakeup = 0;
2589 FILEDESC_XUNLOCK(fdp);
2592 return (fdrop(fp, td));
2596 * Initialize the file pointer with the specified properties.
2598 * The ops are set with release semantics to be certain that the flags, type,
2599 * and data are visible when ops is. This is to prevent ops methods from being
2600 * called with bad data.
2603 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2608 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2612 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2613 struct file **fpp, struct filecaps *havecapsp)
2615 struct filedescent *fde;
2618 FILEDESC_LOCK_ASSERT(fdp);
2620 fde = fdeget_locked(fdp, fd);
2627 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2632 if (havecapsp != NULL)
2633 filecaps_copy(&fde->fde_caps, havecapsp, true);
2635 *fpp = fde->fde_file;
2643 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2644 struct file **fpp, struct filecaps *havecapsp)
2646 struct filedesc *fdp = td->td_proc->p_fd;
2648 #ifndef CAPABILITIES
2649 error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2650 if (error == 0 && havecapsp != NULL)
2651 filecaps_fill(havecapsp);
2657 error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2661 if (havecapsp != NULL) {
2662 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2663 havecapsp, false)) {
2669 if (!fd_modified(fdp, fd, seq))
2678 FILEDESC_SLOCK(fdp);
2679 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2680 if (error == 0 && !fhold(*fpp))
2682 FILEDESC_SUNLOCK(fdp);
2688 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2689 struct file **fpp, seqc_t *seqp)
2692 const struct filedescent *fde;
2694 const struct fdescenttbl *fdt;
2699 cap_rights_t haverights;
2703 fdt = fdp->fd_files;
2704 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2707 * Fetch the descriptor locklessly. We avoid fdrop() races by
2708 * never raising a refcount above 0. To accomplish this we have
2709 * to use a cmpset loop rather than an atomic_add. The descriptor
2710 * must be re-verified once we acquire a reference to be certain
2711 * that the identity is still correct and we did not lose a race
2712 * due to preemption.
2716 seq = seqc_read(fd_seqc(fdt, fd));
2717 fde = &fdt->fdt_ofiles[fd];
2718 haverights = *cap_rights_fde_inline(fde);
2720 if (!seqc_consistent(fd_seqc(fdt, fd), seq))
2723 fp = fdt->fdt_ofiles[fd].fde_file;
2728 error = cap_check(&haverights, needrightsp);
2732 count = fp->f_count;
2736 * Force a reload. Other thread could reallocate the
2737 * table before this fd was closed, so it possible that
2738 * there is a stale fp pointer in cached version.
2740 fdt = (struct fdescenttbl *)atomic_load_ptr(&fdp->fd_files);
2743 if (__predict_false(count + 1 < count))
2747 * Use an acquire barrier to force re-reading of fdt so it is
2748 * refreshed for verification.
2750 if (__predict_false(atomic_fcmpset_acq_int(&fp->f_count,
2751 &count, count + 1) == 0))
2753 fdt = fdp->fd_files;
2755 if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
2757 if (fp == fdt->fdt_ofiles[fd].fde_file)
2760 fdrop(fp, curthread);
2772 * Extract the file pointer associated with the specified descriptor for the
2773 * current user process.
2775 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2778 * File's rights will be checked against the capability rights mask.
2780 * If an error occurred the non-zero error is returned and *fpp is set to
2781 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
2782 * responsible for fdrop().
2785 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2786 cap_rights_t *needrightsp, seqc_t *seqp)
2788 struct filedesc *fdp;
2793 fdp = td->td_proc->p_fd;
2794 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2795 if (__predict_false(error != 0))
2797 if (__predict_false(fp->f_ops == &badfileops)) {
2803 * FREAD and FWRITE failure return EBADF as per POSIX.
2809 if ((fp->f_flag & flags) == 0)
2813 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2814 ((fp->f_flag & FWRITE) != 0))
2820 KASSERT(0, ("wrong flags"));
2833 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2836 return (_fget(td, fd, fpp, 0, rightsp, NULL));
2840 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2844 #ifndef CAPABILITIES
2845 error = _fget(td, fd, fpp, 0, rightsp, NULL);
2846 if (maxprotp != NULL)
2847 *maxprotp = VM_PROT_ALL;
2849 cap_rights_t fdrights;
2850 struct filedesc *fdp = td->td_proc->p_fd;
2853 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2855 error = _fget(td, fd, fpp, 0, rightsp, &seq);
2858 if (maxprotp != NULL)
2859 fdrights = *cap_rights(fdp, fd);
2860 if (!fd_modified(fdp, fd, seq))
2866 * If requested, convert capability rights to access flags.
2868 if (maxprotp != NULL)
2869 *maxprotp = cap_rights_to_vmprot(&fdrights);
2875 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2878 return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2882 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2885 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2889 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2892 struct filedesc *fdp = td->td_proc->p_fd;
2893 #ifndef CAPABILITIES
2894 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2899 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2901 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2904 error = cap_fcntl_check(fdp, fd, needfcntl);
2905 if (!fd_modified(fdp, fd, seq))
2918 * Like fget() but loads the underlying vnode, or returns an error if the
2919 * descriptor does not represent a vnode. Note that pipes use vnodes but
2920 * never have VM objects. The returned vnode will be vref()'d.
2922 * XXX: what about the unused flags ?
2925 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2932 error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2935 if (fp->f_vnode == NULL) {
2947 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2950 return (_fgetvp(td, fd, 0, rightsp, vpp));
2954 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2955 struct filecaps *havecaps, struct vnode **vpp)
2957 struct filedesc *fdp;
2958 struct filecaps caps;
2962 fdp = td->td_proc->p_fd;
2963 error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2966 if (fp->f_ops == &badfileops) {
2970 if (fp->f_vnode == NULL) {
2981 filecaps_free(&caps);
2986 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2989 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2993 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2996 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3001 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3005 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3010 * Handle the last reference to a file being closed.
3012 * Without the noinline attribute clang keeps inlining the func thorough this
3013 * file when fdrop is used.
3016 _fdrop(struct file *fp, struct thread *td)
3020 if (fp->f_count != 0)
3021 panic("fdrop: count %d", fp->f_count);
3022 error = fo_close(fp, td);
3023 atomic_subtract_int(&openfiles, 1);
3025 free(fp->f_advice, M_FADVISE);
3026 uma_zfree(file_zone, fp);
3032 * Apply an advisory lock on a file descriptor.
3034 * Just attempt to get a record lock of the requested type on the entire file
3035 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3037 #ifndef _SYS_SYSPROTO_H_
3045 sys_flock(struct thread *td, struct flock_args *uap)
3052 error = fget(td, uap->fd, &cap_flock_rights, &fp);
3055 if (fp->f_type != DTYPE_VNODE) {
3057 return (EOPNOTSUPP);
3061 lf.l_whence = SEEK_SET;
3064 if (uap->how & LOCK_UN) {
3065 lf.l_type = F_UNLCK;
3066 atomic_clear_int(&fp->f_flag, FHASLOCK);
3067 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3070 if (uap->how & LOCK_EX)
3071 lf.l_type = F_WRLCK;
3072 else if (uap->how & LOCK_SH)
3073 lf.l_type = F_RDLCK;
3078 atomic_set_int(&fp->f_flag, FHASLOCK);
3079 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3080 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3086 * Duplicate the specified descriptor to a free descriptor.
3089 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3090 int openerror, int *indxp)
3092 struct filedescent *newfde, *oldfde;
3097 KASSERT(openerror == ENODEV || openerror == ENXIO,
3098 ("unexpected error %d in %s", openerror, __func__));
3101 * If the to-be-dup'd fd number is greater than the allowed number
3102 * of file descriptors, or the fd to be dup'd has already been
3103 * closed, then reject.
3105 FILEDESC_XLOCK(fdp);
3106 if ((fp = fget_locked(fdp, dfd)) == NULL) {
3107 FILEDESC_XUNLOCK(fdp);
3111 error = fdalloc(td, 0, &indx);
3113 FILEDESC_XUNLOCK(fdp);
3118 * There are two cases of interest here.
3120 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3122 * For ENXIO steal away the file structure from (dfd) and store it in
3123 * (indx). (dfd) is effectively closed by this operation.
3125 switch (openerror) {
3128 * Check that the mode the file is being opened for is a
3129 * subset of the mode of the existing descriptor.
3131 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3132 fdunused(fdp, indx);
3133 FILEDESC_XUNLOCK(fdp);
3137 fdunused(fdp, indx);
3138 FILEDESC_XUNLOCK(fdp);
3141 newfde = &fdp->fd_ofiles[indx];
3142 oldfde = &fdp->fd_ofiles[dfd];
3143 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3145 seqc_write_begin(&newfde->fde_seqc);
3147 memcpy(newfde, oldfde, fde_change_size);
3148 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3151 seqc_write_end(&newfde->fde_seqc);
3156 * Steal away the file pointer from dfd and stuff it into indx.
3158 newfde = &fdp->fd_ofiles[indx];
3159 oldfde = &fdp->fd_ofiles[dfd];
3161 seqc_write_begin(&newfde->fde_seqc);
3163 memcpy(newfde, oldfde, fde_change_size);
3164 oldfde->fde_file = NULL;
3167 seqc_write_end(&newfde->fde_seqc);
3171 FILEDESC_XUNLOCK(fdp);
3177 * This sysctl determines if we will allow a process to chroot(2) if it
3178 * has a directory open:
3179 * 0: disallowed for all processes.
3180 * 1: allowed for processes that were not already chroot(2)'ed.
3181 * 2: allowed for all processes.
3184 static int chroot_allow_open_directories = 1;
3186 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3187 &chroot_allow_open_directories, 0,
3188 "Allow a process to chroot(2) if it has a directory open");
3191 * Helper function for raised chroot(2) security function: Refuse if
3192 * any filedescriptors are open directories.
3195 chroot_refuse_vdir_fds(struct filedesc *fdp)
3201 FILEDESC_LOCK_ASSERT(fdp);
3203 for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3204 fp = fget_locked(fdp, fd);
3207 if (fp->f_type == DTYPE_VNODE) {
3209 if (vp->v_type == VDIR)
3217 * Common routine for kern_chroot() and jail_attach(). The caller is
3218 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3219 * authorize this operation.
3222 pwd_chroot(struct thread *td, struct vnode *vp)
3224 struct filedesc *fdp;
3225 struct vnode *oldvp;
3228 fdp = td->td_proc->p_fd;
3229 FILEDESC_XLOCK(fdp);
3230 if (chroot_allow_open_directories == 0 ||
3231 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3232 error = chroot_refuse_vdir_fds(fdp);
3234 FILEDESC_XUNLOCK(fdp);
3238 oldvp = fdp->fd_rdir;
3241 if (fdp->fd_jdir == NULL) {
3245 FILEDESC_XUNLOCK(fdp);
3251 pwd_chdir(struct thread *td, struct vnode *vp)
3253 struct filedesc *fdp;
3254 struct vnode *oldvp;
3256 fdp = td->td_proc->p_fd;
3257 FILEDESC_XLOCK(fdp);
3258 VNASSERT(vp->v_usecount > 0, vp,
3259 ("chdir to a vnode with zero usecount"));
3260 oldvp = fdp->fd_cdir;
3262 FILEDESC_XUNLOCK(fdp);
3267 * Scan all active processes and prisons to see if any of them have a current
3268 * or root directory of `olddp'. If so, replace them with the new mount point.
3271 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3273 struct filedesc *fdp;
3278 if (vrefcnt(olddp) == 1)
3281 sx_slock(&allproc_lock);
3282 FOREACH_PROC_IN_SYSTEM(p) {
3288 FILEDESC_XLOCK(fdp);
3289 if (fdp->fd_cdir == olddp) {
3291 fdp->fd_cdir = newdp;
3294 if (fdp->fd_rdir == olddp) {
3296 fdp->fd_rdir = newdp;
3299 if (fdp->fd_jdir == olddp) {
3301 fdp->fd_jdir = newdp;
3304 FILEDESC_XUNLOCK(fdp);
3307 sx_sunlock(&allproc_lock);
3308 if (rootvnode == olddp) {
3313 mtx_lock(&prison0.pr_mtx);
3314 if (prison0.pr_root == olddp) {
3316 prison0.pr_root = newdp;
3319 mtx_unlock(&prison0.pr_mtx);
3320 sx_slock(&allprison_lock);
3321 TAILQ_FOREACH(pr, &allprison, pr_list) {
3322 mtx_lock(&pr->pr_mtx);
3323 if (pr->pr_root == olddp) {
3325 pr->pr_root = newdp;
3328 mtx_unlock(&pr->pr_mtx);
3330 sx_sunlock(&allprison_lock);
3335 struct filedesc_to_leader *
3336 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3338 struct filedesc_to_leader *fdtol;
3340 fdtol = malloc(sizeof(struct filedesc_to_leader),
3341 M_FILEDESC_TO_LEADER, M_WAITOK);
3342 fdtol->fdl_refcount = 1;
3343 fdtol->fdl_holdcount = 0;
3344 fdtol->fdl_wakeup = 0;
3345 fdtol->fdl_leader = leader;
3347 FILEDESC_XLOCK(fdp);
3348 fdtol->fdl_next = old->fdl_next;
3349 fdtol->fdl_prev = old;
3350 old->fdl_next = fdtol;
3351 fdtol->fdl_next->fdl_prev = fdtol;
3352 FILEDESC_XUNLOCK(fdp);
3354 fdtol->fdl_next = fdtol;
3355 fdtol->fdl_prev = fdtol;
3361 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3363 struct filedesc *fdp;
3364 int i, count, slots;
3366 if (*(int *)arg1 != 0)
3369 fdp = curproc->p_fd;
3371 FILEDESC_SLOCK(fdp);
3372 slots = NDSLOTS(fdp->fd_lastfile + 1);
3373 for (i = 0; i < slots; i++)
3374 count += bitcountl(fdp->fd_map[i]);
3375 FILEDESC_SUNLOCK(fdp);
3377 return (SYSCTL_OUT(req, &count, sizeof(count)));
3380 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3381 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3382 "Number of open file descriptors");
3385 * Get file structures globally.
3388 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3391 struct filedesc *fdp;
3396 error = sysctl_wire_old_buffer(req, 0);
3399 if (req->oldptr == NULL) {
3401 sx_slock(&allproc_lock);
3402 FOREACH_PROC_IN_SYSTEM(p) {
3404 if (p->p_state == PRS_NEW) {
3412 /* overestimates sparse tables. */
3413 if (fdp->fd_lastfile > 0)
3414 n += fdp->fd_lastfile;
3417 sx_sunlock(&allproc_lock);
3418 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3421 bzero(&xf, sizeof(xf));
3422 xf.xf_size = sizeof(xf);
3423 sx_slock(&allproc_lock);
3424 FOREACH_PROC_IN_SYSTEM(p) {
3426 if (p->p_state == PRS_NEW) {
3430 if (p_cansee(req->td, p) != 0) {
3434 xf.xf_pid = p->p_pid;
3435 xf.xf_uid = p->p_ucred->cr_uid;
3440 FILEDESC_SLOCK(fdp);
3441 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3442 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3445 xf.xf_file = (uintptr_t)fp;
3446 xf.xf_data = (uintptr_t)fp->f_data;
3447 xf.xf_vnode = (uintptr_t)fp->f_vnode;
3448 xf.xf_type = (uintptr_t)fp->f_type;
3449 xf.xf_count = fp->f_count;
3451 xf.xf_offset = foffset_get(fp);
3452 xf.xf_flag = fp->f_flag;
3453 error = SYSCTL_OUT(req, &xf, sizeof(xf));
3457 FILEDESC_SUNLOCK(fdp);
3462 sx_sunlock(&allproc_lock);
3466 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3467 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3469 #ifdef KINFO_FILE_SIZE
3470 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3474 xlate_fflags(int fflags)
3476 static const struct {
3479 } fflags_table[] = {
3480 { FAPPEND, KF_FLAG_APPEND },
3481 { FASYNC, KF_FLAG_ASYNC },
3482 { FFSYNC, KF_FLAG_FSYNC },
3483 { FHASLOCK, KF_FLAG_HASLOCK },
3484 { FNONBLOCK, KF_FLAG_NONBLOCK },
3485 { FREAD, KF_FLAG_READ },
3486 { FWRITE, KF_FLAG_WRITE },
3487 { O_CREAT, KF_FLAG_CREAT },
3488 { O_DIRECT, KF_FLAG_DIRECT },
3489 { O_EXCL, KF_FLAG_EXCL },
3490 { O_EXEC, KF_FLAG_EXEC },
3491 { O_EXLOCK, KF_FLAG_EXLOCK },
3492 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3493 { O_SHLOCK, KF_FLAG_SHLOCK },
3494 { O_TRUNC, KF_FLAG_TRUNC }
3500 for (i = 0; i < nitems(fflags_table); i++)
3501 if (fflags & fflags_table[i].fflag)
3502 kflags |= fflags_table[i].kf_fflag;
3506 /* Trim unused data from kf_path by truncating the structure size. */
3508 pack_kinfo(struct kinfo_file *kif)
3511 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3512 strlen(kif->kf_path) + 1;
3513 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3517 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3518 struct kinfo_file *kif, struct filedesc *fdp, int flags)
3522 bzero(kif, sizeof(*kif));
3524 /* Set a default type to allow for empty fill_kinfo() methods. */
3525 kif->kf_type = KF_TYPE_UNKNOWN;
3526 kif->kf_flags = xlate_fflags(fp->f_flag);
3527 if (rightsp != NULL)
3528 kif->kf_cap_rights = *rightsp;
3530 cap_rights_init(&kif->kf_cap_rights);
3532 kif->kf_ref_count = fp->f_count;
3533 kif->kf_offset = foffset_get(fp);
3536 * This may drop the filedesc lock, so the 'fp' cannot be
3537 * accessed after this call.
3539 error = fo_fill_kinfo(fp, kif, fdp);
3541 kif->kf_status |= KF_ATTR_VALID;
3542 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3545 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3549 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3550 struct kinfo_file *kif, int flags)
3554 bzero(kif, sizeof(*kif));
3556 kif->kf_type = KF_TYPE_VNODE;
3557 error = vn_fill_kinfo_vnode(vp, kif);
3559 kif->kf_status |= KF_ATTR_VALID;
3560 kif->kf_flags = xlate_fflags(fflags);
3561 cap_rights_init(&kif->kf_cap_rights);
3563 kif->kf_ref_count = -1;
3564 kif->kf_offset = -1;
3565 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3568 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3572 struct export_fd_buf {
3573 struct filedesc *fdp;
3576 struct kinfo_file kif;
3581 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3583 struct kinfo_file *kif;
3586 if (efbuf->remainder != -1) {
3587 if (efbuf->remainder < kif->kf_structsize) {
3588 /* Terminate export. */
3589 efbuf->remainder = 0;
3592 efbuf->remainder -= kif->kf_structsize;
3594 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3598 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3599 struct export_fd_buf *efbuf)
3603 if (efbuf->remainder == 0)
3605 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3607 FILEDESC_SUNLOCK(efbuf->fdp);
3608 error = export_kinfo_to_sb(efbuf);
3609 FILEDESC_SLOCK(efbuf->fdp);
3614 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3615 struct export_fd_buf *efbuf)
3619 if (efbuf->remainder == 0)
3621 if (efbuf->fdp != NULL)
3622 FILEDESC_SUNLOCK(efbuf->fdp);
3623 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3624 error = export_kinfo_to_sb(efbuf);
3625 if (efbuf->fdp != NULL)
3626 FILEDESC_SLOCK(efbuf->fdp);
3631 * Store a process file descriptor information to sbuf.
3633 * Takes a locked proc as argument, and returns with the proc unlocked.
3636 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
3640 struct filedesc *fdp;
3641 struct export_fd_buf *efbuf;
3642 struct vnode *cttyvp, *textvp, *tracevp;
3644 cap_rights_t rights;
3646 PROC_LOCK_ASSERT(p, MA_OWNED);
3649 tracevp = p->p_tracevp;
3650 if (tracevp != NULL)
3653 textvp = p->p_textvp;
3656 /* Controlling tty. */
3658 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3659 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3665 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3668 efbuf->remainder = maxlen;
3669 efbuf->flags = flags;
3670 if (tracevp != NULL)
3671 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3674 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3676 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3682 FILEDESC_SLOCK(fdp);
3683 /* working directory */
3684 if (fdp->fd_cdir != NULL) {
3685 vrefact(fdp->fd_cdir);
3686 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3688 /* root directory */
3689 if (fdp->fd_rdir != NULL) {
3690 vrefact(fdp->fd_rdir);
3691 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3693 /* jail directory */
3694 if (fdp->fd_jdir != NULL) {
3695 vrefact(fdp->fd_jdir);
3696 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3698 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3699 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3702 rights = *cap_rights(fdp, i);
3703 #else /* !CAPABILITIES */
3704 rights = cap_no_rights;
3707 * Create sysctl entry. It is OK to drop the filedesc
3708 * lock inside of export_file_to_sb() as we will
3709 * re-validate and re-evaluate its properties when the
3712 error = export_file_to_sb(fp, i, &rights, efbuf);
3713 if (error != 0 || efbuf->remainder == 0)
3716 FILEDESC_SUNLOCK(fdp);
3719 free(efbuf, M_TEMP);
3723 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
3726 * Get per-process file descriptors for use by procstat(1), et al.
3729 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3734 int error, error2, *name;
3738 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3739 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3740 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3745 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3746 error = kern_proc_filedesc_out(p, &sb, maxlen,
3747 KERN_FILEDESC_PACK_KINFO);
3748 error2 = sbuf_finish(&sb);
3750 return (error != 0 ? error : error2);
3753 #ifdef COMPAT_FREEBSD7
3754 #ifdef KINFO_OFILE_SIZE
3755 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3759 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3762 okif->kf_structsize = sizeof(*okif);
3763 okif->kf_type = kif->kf_type;
3764 okif->kf_fd = kif->kf_fd;
3765 okif->kf_ref_count = kif->kf_ref_count;
3766 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3767 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3768 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3769 okif->kf_offset = kif->kf_offset;
3770 if (kif->kf_type == KF_TYPE_VNODE)
3771 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
3773 okif->kf_vnode_type = KF_VTYPE_VNON;
3774 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3775 if (kif->kf_type == KF_TYPE_SOCKET) {
3776 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
3777 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
3778 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
3779 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
3780 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
3782 okif->kf_sa_local.ss_family = AF_UNSPEC;
3783 okif->kf_sa_peer.ss_family = AF_UNSPEC;
3788 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3789 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3794 FILEDESC_SUNLOCK(fdp);
3795 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3796 kinfo_to_okinfo(kif, okif);
3797 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3798 FILEDESC_SLOCK(fdp);
3803 * Get per-process file descriptors for use by procstat(1), et al.
3806 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3808 struct kinfo_ofile *okif;
3809 struct kinfo_file *kif;
3810 struct filedesc *fdp;
3811 int error, i, *name;
3816 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3823 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3824 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3825 FILEDESC_SLOCK(fdp);
3826 if (fdp->fd_cdir != NULL)
3827 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3829 if (fdp->fd_rdir != NULL)
3830 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3832 if (fdp->fd_jdir != NULL)
3833 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3835 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3836 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3838 export_file_to_kinfo(fp, i, NULL, kif, fdp,
3839 KERN_FILEDESC_PACK_KINFO);
3840 FILEDESC_SUNLOCK(fdp);
3841 kinfo_to_okinfo(kif, okif);
3842 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3843 FILEDESC_SLOCK(fdp);
3847 FILEDESC_SUNLOCK(fdp);
3854 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3855 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3856 "Process ofiledesc entries");
3857 #endif /* COMPAT_FREEBSD7 */
3860 vntype_to_kinfo(int vtype)
3865 } vtypes_table[] = {
3866 { VBAD, KF_VTYPE_VBAD },
3867 { VBLK, KF_VTYPE_VBLK },
3868 { VCHR, KF_VTYPE_VCHR },
3869 { VDIR, KF_VTYPE_VDIR },
3870 { VFIFO, KF_VTYPE_VFIFO },
3871 { VLNK, KF_VTYPE_VLNK },
3872 { VNON, KF_VTYPE_VNON },
3873 { VREG, KF_VTYPE_VREG },
3874 { VSOCK, KF_VTYPE_VSOCK }
3879 * Perform vtype translation.
3881 for (i = 0; i < nitems(vtypes_table); i++)
3882 if (vtypes_table[i].vtype == vtype)
3883 return (vtypes_table[i].kf_vtype);
3885 return (KF_VTYPE_UNKNOWN);
3888 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3889 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3890 "Process filedesc entries");
3893 * Store a process current working directory information to sbuf.
3895 * Takes a locked proc as argument, and returns with the proc unlocked.
3898 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
3900 struct filedesc *fdp;
3901 struct export_fd_buf *efbuf;
3904 PROC_LOCK_ASSERT(p, MA_OWNED);
3911 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3914 efbuf->remainder = maxlen;
3916 FILEDESC_SLOCK(fdp);
3917 if (fdp->fd_cdir == NULL)
3920 vrefact(fdp->fd_cdir);
3921 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3924 FILEDESC_SUNLOCK(fdp);
3926 free(efbuf, M_TEMP);
3931 * Get per-process current working directory.
3934 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3939 int error, error2, *name;
3943 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3944 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3945 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3950 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3951 error = kern_proc_cwd_out(p, &sb, maxlen);
3952 error2 = sbuf_finish(&sb);
3954 return (error != 0 ? error : error2);
3957 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3958 sysctl_kern_proc_cwd, "Process current working directory");
3962 * For the purposes of debugging, generate a human-readable string for the
3966 file_type_to_name(short type)
3994 case DTYPE_PROCDESC:
3996 case DTYPE_LINUXEFD:
3998 case DTYPE_LINUXTFD:
4006 * For the purposes of debugging, identify a process (if any, perhaps one of
4007 * many) that references the passed file in its file descriptor array. Return
4010 static struct proc *
4011 file_to_first_proc(struct file *fp)
4013 struct filedesc *fdp;
4017 FOREACH_PROC_IN_SYSTEM(p) {
4018 if (p->p_state == PRS_NEW)
4023 for (n = 0; n <= fdp->fd_lastfile; n++) {
4024 if (fp == fdp->fd_ofiles[n].fde_file)
4032 db_print_file(struct file *fp, int header)
4034 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4038 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4039 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4040 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4042 p = file_to_first_proc(fp);
4043 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4044 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4045 fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode,
4046 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4051 DB_SHOW_COMMAND(file, db_show_file)
4056 db_printf("usage: show file <addr>\n");
4059 fp = (struct file *)addr;
4060 db_print_file(fp, 1);
4063 DB_SHOW_COMMAND(files, db_show_files)
4065 struct filedesc *fdp;
4072 FOREACH_PROC_IN_SYSTEM(p) {
4073 if (p->p_state == PRS_NEW)
4075 if ((fdp = p->p_fd) == NULL)
4077 for (n = 0; n <= fdp->fd_lastfile; ++n) {
4078 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4080 db_print_file(fp, header);
4087 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4088 &maxfilesperproc, 0, "Maximum files allowed open per process");
4090 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4091 &maxfiles, 0, "Maximum number of files");
4093 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4094 &openfiles, 0, "System-wide number of open files");
4098 filelistinit(void *dummy)
4101 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4102 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4103 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4104 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4105 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4107 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4109 /*-------------------------------------------------------------------*/
4112 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4113 int flags, struct thread *td)
4120 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4128 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4136 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4144 badfo_kqfilter(struct file *fp, struct knote *kn)
4151 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4159 badfo_close(struct file *fp, struct thread *td)
4166 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4174 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4182 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4183 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4191 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4197 struct fileops badfileops = {
4198 .fo_read = badfo_readwrite,
4199 .fo_write = badfo_readwrite,
4200 .fo_truncate = badfo_truncate,
4201 .fo_ioctl = badfo_ioctl,
4202 .fo_poll = badfo_poll,
4203 .fo_kqfilter = badfo_kqfilter,
4204 .fo_stat = badfo_stat,
4205 .fo_close = badfo_close,
4206 .fo_chmod = badfo_chmod,
4207 .fo_chown = badfo_chown,
4208 .fo_sendfile = badfo_sendfile,
4209 .fo_fill_kinfo = badfo_fill_kinfo,
4213 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4214 int flags, struct thread *td)
4217 return (EOPNOTSUPP);
4221 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4229 invfo_ioctl(struct file *fp, u_long com, void *data,
4230 struct ucred *active_cred, struct thread *td)
4237 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4241 return (poll_no_poll(events));
4245 invfo_kqfilter(struct file *fp, struct knote *kn)
4252 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4260 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4268 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4269 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4276 /*-------------------------------------------------------------------*/
4279 * File Descriptor pseudo-device driver (/dev/fd/).
4281 * Opening minor device N dup()s the file (if any) connected to file
4282 * descriptor N belonging to the calling process. Note that this driver
4283 * consists of only the ``open()'' routine, because all subsequent
4284 * references to this file will be direct to the other driver.
4286 * XXX: we could give this one a cloning event handler if necessary.
4291 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4295 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4296 * the file descriptor being sought for duplication. The error
4297 * return ensures that the vnode for this device will be released
4298 * by vn_open. Open will detect this special error and take the
4299 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4300 * will simply report the error.
4302 td->td_dupfd = dev2unit(dev);
4306 static struct cdevsw fildesc_cdevsw = {
4307 .d_version = D_VERSION,
4313 fildesc_drvinit(void *unused)
4317 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4318 UID_ROOT, GID_WHEEL, 0666, "fd/0");
4319 make_dev_alias(dev, "stdin");
4320 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4321 UID_ROOT, GID_WHEEL, 0666, "fd/1");
4322 make_dev_alias(dev, "stdout");
4323 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4324 UID_ROOT, GID_WHEEL, 0666, "fd/2");
4325 make_dev_alias(dev, "stderr");
4328 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);