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);
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);
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);
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);
764 error = fo_add_seals(fp, arg);
769 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
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);
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);
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
1448 sbp->st_atim_ext = 0;
1449 sbp->st_mtim_ext = 0;
1450 sbp->st_ctim_ext = 0;
1451 sbp->st_btim_ext = 0;
1454 if (KTRPOINT(td, KTR_STRUCT))
1455 ktrstat_error(sbp, error);
1460 #if defined(COMPAT_FREEBSD11)
1462 * Return status information about a file descriptor.
1464 #ifndef _SYS_SYSPROTO_H_
1465 struct freebsd11_nfstat_args {
1472 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1478 error = kern_fstat(td, uap->fd, &ub);
1480 freebsd11_cvtnstat(&ub, &nub);
1481 error = copyout(&nub, uap->sb, sizeof(nub));
1485 #endif /* COMPAT_FREEBSD11 */
1488 * Return pathconf information about a file descriptor.
1490 #ifndef _SYS_SYSPROTO_H_
1491 struct fpathconf_args {
1498 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1503 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1505 td->td_retval[0] = value;
1510 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1516 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1520 if (name == _PC_ASYNC_IO) {
1521 *valuep = _POSIX_ASYNCHRONOUS_IO;
1526 vn_lock(vp, LK_SHARED | LK_RETRY);
1527 error = VOP_PATHCONF(vp, name, valuep);
1529 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1530 if (name != _PC_PIPE_BUF) {
1545 * Copy filecaps structure allocating memory for ioctls array if needed.
1547 * The last parameter indicates whether the fdtable is locked. If it is not and
1548 * ioctls are encountered, copying fails and the caller must lock the table.
1550 * Note that if the table was not locked, the caller has to check the relevant
1551 * sequence counter to determine whether the operation was successful.
1554 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1558 if (src->fc_ioctls != NULL && !locked)
1560 memcpy(dst, src, sizeof(*src));
1561 if (src->fc_ioctls == NULL)
1564 KASSERT(src->fc_nioctls > 0,
1565 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1567 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1568 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1569 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1574 filecaps_copy_prep(const struct filecaps *src)
1579 if (__predict_true(src->fc_ioctls == NULL))
1582 KASSERT(src->fc_nioctls > 0,
1583 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1585 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1586 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1591 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1597 if (__predict_true(src->fc_ioctls == NULL)) {
1598 MPASS(ioctls == NULL);
1602 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1603 dst->fc_ioctls = ioctls;
1604 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1608 * Move filecaps structure to the new place and clear the old place.
1611 filecaps_move(struct filecaps *src, struct filecaps *dst)
1615 bzero(src, sizeof(*src));
1619 * Fill the given filecaps structure with full rights.
1622 filecaps_fill(struct filecaps *fcaps)
1625 CAP_ALL(&fcaps->fc_rights);
1626 fcaps->fc_ioctls = NULL;
1627 fcaps->fc_nioctls = -1;
1628 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1632 * Free memory allocated within filecaps structure.
1635 filecaps_free(struct filecaps *fcaps)
1638 free(fcaps->fc_ioctls, M_FILECAPS);
1639 bzero(fcaps, sizeof(*fcaps));
1643 filecaps_free_prep(struct filecaps *fcaps)
1647 ioctls = fcaps->fc_ioctls;
1648 bzero(fcaps, sizeof(*fcaps));
1653 filecaps_free_finish(u_long *ioctls)
1656 free(ioctls, M_FILECAPS);
1660 * Validate the given filecaps structure.
1663 filecaps_validate(const struct filecaps *fcaps, const char *func)
1666 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1667 ("%s: invalid rights", func));
1668 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1669 ("%s: invalid fcntls", func));
1670 KASSERT(fcaps->fc_fcntls == 0 ||
1671 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1672 ("%s: fcntls without CAP_FCNTL", func));
1673 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1674 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1675 ("%s: invalid ioctls", func));
1676 KASSERT(fcaps->fc_nioctls == 0 ||
1677 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1678 ("%s: ioctls without CAP_IOCTL", func));
1682 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1686 FILEDESC_XLOCK_ASSERT(fdp);
1688 nfd1 = fdp->fd_nfiles * 2;
1691 fdgrowtable(fdp, nfd1);
1695 * Grow the file table to accommodate (at least) nfd descriptors.
1698 fdgrowtable(struct filedesc *fdp, int nfd)
1700 struct filedesc0 *fdp0;
1701 struct freetable *ft;
1702 struct fdescenttbl *ntable;
1703 struct fdescenttbl *otable;
1704 int nnfiles, onfiles;
1705 NDSLOTTYPE *nmap, *omap;
1708 * If lastfile is -1 this struct filedesc was just allocated and we are
1709 * growing it to accommodate for the one we are going to copy from. There
1710 * is no need to have a lock on this one as it's not visible to anyone.
1712 if (fdp->fd_lastfile != -1)
1713 FILEDESC_XLOCK_ASSERT(fdp);
1715 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1717 /* save old values */
1718 onfiles = fdp->fd_nfiles;
1719 otable = fdp->fd_files;
1722 /* compute the size of the new table */
1723 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1724 if (nnfiles <= onfiles)
1725 /* the table is already large enough */
1729 * Allocate a new table. We need enough space for the number of
1730 * entries, file entries themselves and the struct freetable we will use
1731 * when we decommission the table and place it on the freelist.
1732 * We place the struct freetable in the middle so we don't have
1733 * to worry about padding.
1735 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1736 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1737 sizeof(struct freetable),
1738 M_FILEDESC, M_ZERO | M_WAITOK);
1739 /* copy the old data */
1740 ntable->fdt_nfiles = nnfiles;
1741 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1742 onfiles * sizeof(ntable->fdt_ofiles[0]));
1745 * Allocate a new map only if the old is not large enough. It will
1746 * grow at a slower rate than the table as it can map more
1747 * entries than the table can hold.
1749 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1750 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1752 /* copy over the old data and update the pointer */
1753 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1758 * Make sure that ntable is correctly initialized before we replace
1759 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1762 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1765 * Do not free the old file table, as some threads may still
1766 * reference entries within it. Instead, place it on a freelist
1767 * which will be processed when the struct filedesc is released.
1769 * Note that if onfiles == NDFILE, we're dealing with the original
1770 * static allocation contained within (struct filedesc0 *)fdp,
1771 * which must not be freed.
1773 if (onfiles > NDFILE) {
1774 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1775 fdp0 = (struct filedesc0 *)fdp;
1776 ft->ft_table = otable;
1777 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1780 * The map does not have the same possibility of threads still
1781 * holding references to it. So always free it as long as it
1782 * does not reference the original static allocation.
1784 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1785 free(omap, M_FILEDESC);
1789 * Allocate a file descriptor for the process.
1792 fdalloc(struct thread *td, int minfd, int *result)
1794 struct proc *p = td->td_proc;
1795 struct filedesc *fdp = p->p_fd;
1796 int fd, maxfd, allocfd;
1801 FILEDESC_XLOCK_ASSERT(fdp);
1803 if (fdp->fd_freefile > minfd)
1804 minfd = fdp->fd_freefile;
1806 maxfd = getmaxfd(td);
1809 * Search the bitmap for a free descriptor starting at minfd.
1810 * If none is found, grow the file table.
1812 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1815 if (fd >= fdp->fd_nfiles) {
1816 allocfd = min(fd * 2, maxfd);
1818 if (RACCT_ENABLED()) {
1819 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1825 * fd is already equal to first free descriptor >= minfd, so
1826 * we only need to grow the table and we are done.
1828 fdgrowtable_exp(fdp, allocfd);
1832 * Perform some sanity checks, then mark the file descriptor as
1833 * used and return it to the caller.
1835 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1836 ("invalid descriptor %d", fd));
1837 KASSERT(!fdisused(fdp, fd),
1838 ("fd_first_free() returned non-free descriptor"));
1839 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1840 ("file descriptor isn't free"));
1847 * Allocate n file descriptors for the process.
1850 fdallocn(struct thread *td, int minfd, int *fds, int n)
1852 struct proc *p = td->td_proc;
1853 struct filedesc *fdp = p->p_fd;
1856 FILEDESC_XLOCK_ASSERT(fdp);
1858 for (i = 0; i < n; i++)
1859 if (fdalloc(td, 0, &fds[i]) != 0)
1863 for (i--; i >= 0; i--)
1864 fdunused(fdp, fds[i]);
1872 * Create a new open file structure and allocate a file descriptor for the
1873 * process that refers to it. We add one reference to the file for the
1874 * descriptor table and one reference for resultfp. This is to prevent us
1875 * being preempted and the entry in the descriptor table closed after we
1876 * release the FILEDESC lock.
1879 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1880 struct filecaps *fcaps)
1885 error = falloc_noinstall(td, &fp);
1887 return (error); /* no reference held on error */
1889 error = finstall(td, fp, &fd, flags, fcaps);
1891 fdrop(fp, td); /* one reference (fp only) */
1895 if (resultfp != NULL)
1896 *resultfp = fp; /* copy out result */
1898 fdrop(fp, td); /* release local reference */
1900 if (resultfd != NULL)
1907 * Create a new open file structure without allocating a file descriptor.
1910 falloc_noinstall(struct thread *td, struct file **resultfp)
1913 int maxuserfiles = maxfiles - (maxfiles / 20);
1915 static struct timeval lastfail;
1918 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1920 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1921 if ((openfiles_new >= maxuserfiles &&
1922 priv_check(td, PRIV_MAXFILES) != 0) ||
1923 openfiles_new >= maxfiles) {
1924 atomic_subtract_int(&openfiles, 1);
1925 if (ppsratecheck(&lastfail, &curfail, 1)) {
1926 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1927 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1931 fp = uma_zalloc(file_zone, M_WAITOK);
1932 bzero(fp, sizeof(*fp));
1933 refcount_init(&fp->f_count, 1);
1934 fp->f_cred = crhold(td->td_ucred);
1935 fp->f_ops = &badfileops;
1941 * Install a file in a file descriptor table.
1944 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1945 struct filecaps *fcaps)
1947 struct filedescent *fde;
1951 filecaps_validate(fcaps, __func__);
1952 FILEDESC_XLOCK_ASSERT(fdp);
1954 fde = &fdp->fd_ofiles[fd];
1956 seqc_write_begin(&fde->fde_seqc);
1959 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1961 filecaps_move(fcaps, &fde->fde_caps);
1963 filecaps_fill(&fde->fde_caps);
1965 seqc_write_end(&fde->fde_seqc);
1970 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1971 struct filecaps *fcaps)
1973 struct filedesc *fdp = td->td_proc->p_fd;
1980 FILEDESC_XLOCK(fdp);
1981 error = fdalloc(td, 0, fd);
1982 if (__predict_false(error != 0)) {
1983 FILEDESC_XUNLOCK(fdp);
1987 _finstall(fdp, fp, *fd, flags, fcaps);
1988 FILEDESC_XUNLOCK(fdp);
1993 * Build a new filedesc structure from another.
1994 * Copy the current, root, and jail root vnode references.
1996 * If fdp is not NULL, return with it shared locked.
1999 fdinit(struct filedesc *fdp, bool prepfiles)
2001 struct filedesc0 *newfdp0;
2002 struct filedesc *newfdp;
2004 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2005 newfdp = &newfdp0->fd_fd;
2007 /* Create the file descriptor table. */
2008 FILEDESC_LOCK_INIT(newfdp);
2009 refcount_init(&newfdp->fd_refcnt, 1);
2010 refcount_init(&newfdp->fd_holdcnt, 1);
2011 newfdp->fd_cmask = CMASK;
2012 newfdp->fd_map = newfdp0->fd_dmap;
2013 newfdp->fd_lastfile = -1;
2014 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2015 newfdp->fd_files->fdt_nfiles = NDFILE;
2020 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
2021 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
2023 FILEDESC_SLOCK(fdp);
2024 newfdp->fd_cdir = fdp->fd_cdir;
2025 if (newfdp->fd_cdir)
2026 vrefact(newfdp->fd_cdir);
2027 newfdp->fd_rdir = fdp->fd_rdir;
2028 if (newfdp->fd_rdir)
2029 vrefact(newfdp->fd_rdir);
2030 newfdp->fd_jdir = fdp->fd_jdir;
2031 if (newfdp->fd_jdir)
2032 vrefact(newfdp->fd_jdir);
2035 FILEDESC_SUNLOCK(fdp);
2037 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
2038 FILEDESC_SUNLOCK(fdp);
2039 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
2040 FILEDESC_SLOCK(fdp);
2047 static struct filedesc *
2048 fdhold(struct proc *p)
2050 struct filedesc *fdp;
2052 PROC_LOCK_ASSERT(p, MA_OWNED);
2055 refcount_acquire(&fdp->fd_holdcnt);
2060 fddrop(struct filedesc *fdp)
2063 if (fdp->fd_holdcnt > 1) {
2064 if (refcount_release(&fdp->fd_holdcnt) == 0)
2068 FILEDESC_LOCK_DESTROY(fdp);
2069 uma_zfree(filedesc0_zone, fdp);
2073 * Share a filedesc structure.
2076 fdshare(struct filedesc *fdp)
2079 refcount_acquire(&fdp->fd_refcnt);
2084 * Unshare a filedesc structure, if necessary by making a copy
2087 fdunshare(struct thread *td)
2089 struct filedesc *tmp;
2090 struct proc *p = td->td_proc;
2092 if (p->p_fd->fd_refcnt == 1)
2095 tmp = fdcopy(p->p_fd);
2101 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2105 td->td_proc->p_fd = fdp;
2109 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2110 * this is to ease callers, not catch errors.
2113 fdcopy(struct filedesc *fdp)
2115 struct filedesc *newfdp;
2116 struct filedescent *nfde, *ofde;
2121 newfdp = fdinit(fdp, true);
2122 /* copy all passable descriptors (i.e. not kqueue) */
2123 newfdp->fd_freefile = -1;
2124 for (i = 0; i <= fdp->fd_lastfile; ++i) {
2125 ofde = &fdp->fd_ofiles[i];
2126 if (ofde->fde_file == NULL ||
2127 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2128 !fhold(ofde->fde_file)) {
2129 if (newfdp->fd_freefile == -1)
2130 newfdp->fd_freefile = i;
2133 nfde = &newfdp->fd_ofiles[i];
2135 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2136 fdused_init(newfdp, i);
2137 newfdp->fd_lastfile = i;
2139 if (newfdp->fd_freefile == -1)
2140 newfdp->fd_freefile = i;
2141 newfdp->fd_cmask = fdp->fd_cmask;
2142 FILEDESC_SUNLOCK(fdp);
2147 * Copies a filedesc structure, while remapping all file descriptors
2148 * stored inside using a translation table.
2150 * File descriptors are copied over to the new file descriptor table,
2151 * regardless of whether the close-on-exec flag is set.
2154 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2155 struct filedesc **ret)
2157 struct filedesc *newfdp;
2158 struct filedescent *nfde, *ofde;
2163 newfdp = fdinit(fdp, true);
2164 if (nfds > fdp->fd_lastfile + 1) {
2165 /* New table cannot be larger than the old one. */
2169 /* Copy all passable descriptors (i.e. not kqueue). */
2170 newfdp->fd_freefile = nfds;
2171 for (i = 0; i < nfds; ++i) {
2172 if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2173 /* File descriptor out of bounds. */
2177 ofde = &fdp->fd_ofiles[fds[i]];
2178 if (ofde->fde_file == NULL) {
2179 /* Unused file descriptor. */
2183 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2184 /* File descriptor cannot be passed. */
2188 if (!fhold(nfde->fde_file)) {
2192 nfde = &newfdp->fd_ofiles[i];
2194 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2195 fdused_init(newfdp, i);
2196 newfdp->fd_lastfile = i;
2198 newfdp->fd_cmask = fdp->fd_cmask;
2199 FILEDESC_SUNLOCK(fdp);
2203 FILEDESC_SUNLOCK(fdp);
2204 fdescfree_remapped(newfdp);
2209 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2210 * one of processes using it exits) and the table used to be shared.
2213 fdclearlocks(struct thread *td)
2215 struct filedesc *fdp;
2216 struct filedesc_to_leader *fdtol;
2226 MPASS(fdtol != NULL);
2228 FILEDESC_XLOCK(fdp);
2229 KASSERT(fdtol->fdl_refcount > 0,
2230 ("filedesc_to_refcount botch: fdl_refcount=%d",
2231 fdtol->fdl_refcount));
2232 if (fdtol->fdl_refcount == 1 &&
2233 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2234 for (i = 0; i <= fdp->fd_lastfile; i++) {
2235 fp = fdp->fd_ofiles[i].fde_file;
2236 if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2239 FILEDESC_XUNLOCK(fdp);
2240 lf.l_whence = SEEK_SET;
2243 lf.l_type = F_UNLCK;
2245 (void) VOP_ADVLOCK(vp,
2246 (caddr_t)p->p_leader, F_UNLCK,
2248 FILEDESC_XLOCK(fdp);
2253 if (fdtol->fdl_refcount == 1) {
2254 if (fdp->fd_holdleaderscount > 0 &&
2255 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2257 * close() or kern_dup() has cleared a reference
2258 * in a shared file descriptor table.
2260 fdp->fd_holdleaderswakeup = 1;
2261 sx_sleep(&fdp->fd_holdleaderscount,
2262 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2265 if (fdtol->fdl_holdcount > 0) {
2267 * Ensure that fdtol->fdl_leader remains
2268 * valid in closef().
2270 fdtol->fdl_wakeup = 1;
2271 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2276 fdtol->fdl_refcount--;
2277 if (fdtol->fdl_refcount == 0 &&
2278 fdtol->fdl_holdcount == 0) {
2279 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2280 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2284 FILEDESC_XUNLOCK(fdp);
2286 free(fdtol, M_FILEDESC_TO_LEADER);
2290 * Release a filedesc structure.
2293 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2295 struct filedesc0 *fdp0;
2296 struct freetable *ft, *tft;
2297 struct filedescent *fde;
2301 for (i = 0; i <= fdp->fd_lastfile; i++) {
2302 fde = &fdp->fd_ofiles[i];
2307 (void) closef(fp, td);
2313 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2314 free(fdp->fd_map, M_FILEDESC);
2315 if (fdp->fd_nfiles > NDFILE)
2316 free(fdp->fd_files, M_FILEDESC);
2318 fdp0 = (struct filedesc0 *)fdp;
2319 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2320 free(ft->ft_table, M_FILEDESC);
2326 fdescfree(struct thread *td)
2329 struct filedesc *fdp;
2330 struct vnode *cdir, *jdir, *rdir;
2337 if (RACCT_ENABLED())
2338 racct_set_unlocked(p, RACCT_NOFILE, 0);
2341 if (p->p_fdtol != NULL)
2348 if (refcount_release(&fdp->fd_refcnt) == 0)
2351 FILEDESC_XLOCK(fdp);
2352 cdir = fdp->fd_cdir;
2353 fdp->fd_cdir = NULL;
2354 rdir = fdp->fd_rdir;
2355 fdp->fd_rdir = NULL;
2356 jdir = fdp->fd_jdir;
2357 fdp->fd_jdir = NULL;
2358 FILEDESC_XUNLOCK(fdp);
2367 fdescfree_fds(td, fdp, 1);
2371 fdescfree_remapped(struct filedesc *fdp)
2374 if (fdp->fd_cdir != NULL)
2375 vrele(fdp->fd_cdir);
2376 if (fdp->fd_rdir != NULL)
2377 vrele(fdp->fd_rdir);
2378 if (fdp->fd_jdir != NULL)
2379 vrele(fdp->fd_jdir);
2381 fdescfree_fds(curthread, fdp, 0);
2385 * For setugid programs, we don't want to people to use that setugidness
2386 * to generate error messages which write to a file which otherwise would
2387 * otherwise be off-limits to the process. We check for filesystems where
2388 * the vnode can change out from under us after execve (like [lin]procfs).
2390 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2391 * sufficient. We also don't check for setugidness since we know we are.
2394 is_unsafe(struct file *fp)
2398 if (fp->f_type != DTYPE_VNODE)
2402 return ((vp->v_vflag & VV_PROCDEP) != 0);
2406 * Make this setguid thing safe, if at all possible.
2409 fdsetugidsafety(struct thread *td)
2411 struct filedesc *fdp;
2415 fdp = td->td_proc->p_fd;
2416 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2417 MPASS(fdp->fd_nfiles >= 3);
2418 for (i = 0; i <= 2; i++) {
2419 fp = fdp->fd_ofiles[i].fde_file;
2420 if (fp != NULL && is_unsafe(fp)) {
2421 FILEDESC_XLOCK(fdp);
2422 knote_fdclose(td, i);
2424 * NULL-out descriptor prior to close to avoid
2425 * a race while close blocks.
2428 FILEDESC_XUNLOCK(fdp);
2429 (void) closef(fp, td);
2435 * If a specific file object occupies a specific file descriptor, close the
2436 * file descriptor entry and drop a reference on the file object. This is a
2437 * convenience function to handle a subsequent error in a function that calls
2438 * falloc() that handles the race that another thread might have closed the
2439 * file descriptor out from under the thread creating the file object.
2442 fdclose(struct thread *td, struct file *fp, int idx)
2444 struct filedesc *fdp = td->td_proc->p_fd;
2446 FILEDESC_XLOCK(fdp);
2447 if (fdp->fd_ofiles[idx].fde_file == fp) {
2449 FILEDESC_XUNLOCK(fdp);
2452 FILEDESC_XUNLOCK(fdp);
2456 * Close any files on exec?
2459 fdcloseexec(struct thread *td)
2461 struct filedesc *fdp;
2462 struct filedescent *fde;
2466 fdp = td->td_proc->p_fd;
2467 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2468 for (i = 0; i <= fdp->fd_lastfile; i++) {
2469 fde = &fdp->fd_ofiles[i];
2471 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2472 (fde->fde_flags & UF_EXCLOSE))) {
2473 FILEDESC_XLOCK(fdp);
2475 (void) closefp(fdp, i, fp, td, 0);
2476 FILEDESC_UNLOCK_ASSERT(fdp);
2482 * It is unsafe for set[ug]id processes to be started with file
2483 * descriptors 0..2 closed, as these descriptors are given implicit
2484 * significance in the Standard C library. fdcheckstd() will create a
2485 * descriptor referencing /dev/null for each of stdin, stdout, and
2486 * stderr that is not already open.
2489 fdcheckstd(struct thread *td)
2491 struct filedesc *fdp;
2493 int i, error, devnull;
2495 fdp = td->td_proc->p_fd;
2496 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2497 MPASS(fdp->fd_nfiles >= 3);
2499 for (i = 0; i <= 2; i++) {
2500 if (fdp->fd_ofiles[i].fde_file != NULL)
2503 save = td->td_retval[0];
2504 if (devnull != -1) {
2505 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2507 error = kern_openat(td, AT_FDCWD, "/dev/null",
2508 UIO_SYSSPACE, O_RDWR, 0);
2510 devnull = td->td_retval[0];
2511 KASSERT(devnull == i, ("we didn't get our fd"));
2514 td->td_retval[0] = save;
2522 * Internal form of close. Decrement reference count on file structure.
2523 * Note: td may be NULL when closing a file that was being passed in a
2527 closef(struct file *fp, struct thread *td)
2531 struct filedesc_to_leader *fdtol;
2532 struct filedesc *fdp;
2535 * POSIX record locking dictates that any close releases ALL
2536 * locks owned by this process. This is handled by setting
2537 * a flag in the unlock to free ONLY locks obeying POSIX
2538 * semantics, and not to free BSD-style file locks.
2539 * If the descriptor was in a message, POSIX-style locks
2540 * aren't passed with the descriptor, and the thread pointer
2541 * will be NULL. Callers should be careful only to pass a
2542 * NULL thread pointer when there really is no owning
2543 * context that might have locks, or the locks will be
2546 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2548 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2549 lf.l_whence = SEEK_SET;
2552 lf.l_type = F_UNLCK;
2553 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2554 F_UNLCK, &lf, F_POSIX);
2556 fdtol = td->td_proc->p_fdtol;
2557 if (fdtol != NULL) {
2559 * Handle special case where file descriptor table is
2560 * shared between multiple process leaders.
2562 fdp = td->td_proc->p_fd;
2563 FILEDESC_XLOCK(fdp);
2564 for (fdtol = fdtol->fdl_next;
2565 fdtol != td->td_proc->p_fdtol;
2566 fdtol = fdtol->fdl_next) {
2567 if ((fdtol->fdl_leader->p_flag &
2570 fdtol->fdl_holdcount++;
2571 FILEDESC_XUNLOCK(fdp);
2572 lf.l_whence = SEEK_SET;
2575 lf.l_type = F_UNLCK;
2577 (void) VOP_ADVLOCK(vp,
2578 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2580 FILEDESC_XLOCK(fdp);
2581 fdtol->fdl_holdcount--;
2582 if (fdtol->fdl_holdcount == 0 &&
2583 fdtol->fdl_wakeup != 0) {
2584 fdtol->fdl_wakeup = 0;
2588 FILEDESC_XUNLOCK(fdp);
2591 return (fdrop(fp, td));
2595 * Initialize the file pointer with the specified properties.
2597 * The ops are set with release semantics to be certain that the flags, type,
2598 * and data are visible when ops is. This is to prevent ops methods from being
2599 * called with bad data.
2602 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2607 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2611 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2612 struct file **fpp, struct filecaps *havecapsp)
2614 struct filedescent *fde;
2617 FILEDESC_LOCK_ASSERT(fdp);
2619 fde = fdeget_locked(fdp, fd);
2626 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2631 if (havecapsp != NULL)
2632 filecaps_copy(&fde->fde_caps, havecapsp, true);
2634 *fpp = fde->fde_file;
2642 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2643 struct file **fpp, struct filecaps *havecapsp)
2645 struct filedesc *fdp = td->td_proc->p_fd;
2647 #ifndef CAPABILITIES
2648 error = fget_unlocked(fdp, fd, needrightsp, fpp);
2649 if (havecapsp != NULL && error == 0)
2650 filecaps_fill(havecapsp);
2657 error = fget_unlocked_seq(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_seq(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;
2698 cap_rights_t haverights;
2702 fdt = fdp->fd_files;
2703 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2706 * Fetch the descriptor locklessly. We avoid fdrop() races by
2707 * never raising a refcount above 0. To accomplish this we have
2708 * to use a cmpset loop rather than an atomic_add. The descriptor
2709 * must be re-verified once we acquire a reference to be certain
2710 * that the identity is still correct and we did not lose a race
2711 * due to preemption.
2715 seq = seqc_read(fd_seqc(fdt, fd));
2716 fde = &fdt->fdt_ofiles[fd];
2717 haverights = *cap_rights_fde_inline(fde);
2719 if (!seqc_consistent(fd_seqc(fdt, fd), seq))
2722 fp = fdt->fdt_ofiles[fd].fde_file;
2727 error = cap_check_inline(&haverights, needrightsp);
2731 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
2733 * The count was found either saturated or zero.
2734 * This re-read is not any more racy than using the
2735 * return value from fcmpset.
2737 if (fp->f_count != 0)
2740 * Force a reload. Other thread could reallocate the
2741 * table before this fd was closed, so it is possible
2742 * that there is a stale fp pointer in cached version.
2744 fdt = (struct fdescenttbl *)atomic_load_ptr(&fdp->fd_files);
2748 * Use an acquire barrier to force re-reading of fdt so it is
2749 * refreshed for verification.
2751 atomic_thread_fence_acq();
2752 fdt = fdp->fd_files;
2754 if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
2756 if (fp == fdt->fdt_ofiles[fd].fde_file)
2759 fdrop(fp, curthread);
2771 * See the comments in fget_unlocked_seq for an explanation of how this works.
2773 * This is a simplified variant which bails out to the aforementioned routine
2774 * if anything goes wrong. In practice this only happens when userspace is
2775 * racing with itself.
2778 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2782 const struct filedescent *fde;
2784 const struct fdescenttbl *fdt;
2788 const cap_rights_t *haverights;
2791 fdt = fdp->fd_files;
2792 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2795 seq = seqc_read_any(fd_seqc(fdt, fd));
2796 if (__predict_false(seqc_in_modify(seq)))
2798 fde = &fdt->fdt_ofiles[fd];
2799 haverights = cap_rights_fde_inline(fde);
2802 fp = fdt->fdt_ofiles[fd].fde_file;
2804 if (__predict_false(fp == NULL))
2807 if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
2810 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
2814 * Use an acquire barrier to force re-reading of fdt so it is
2815 * refreshed for verification.
2817 atomic_thread_fence_acq();
2818 fdt = fdp->fd_files;
2820 if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
2822 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
2828 fdrop(fp, curthread);
2830 return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
2834 * Extract the file pointer associated with the specified descriptor for the
2835 * current user process.
2837 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2840 * File's rights will be checked against the capability rights mask.
2842 * If an error occurred the non-zero error is returned and *fpp is set to
2843 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
2844 * responsible for fdrop().
2847 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2848 cap_rights_t *needrightsp)
2850 struct filedesc *fdp;
2855 fdp = td->td_proc->p_fd;
2856 error = fget_unlocked(fdp, fd, needrightsp, &fp);
2857 if (__predict_false(error != 0))
2859 if (__predict_false(fp->f_ops == &badfileops)) {
2865 * FREAD and FWRITE failure return EBADF as per POSIX.
2871 if ((fp->f_flag & flags) == 0)
2875 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2876 ((fp->f_flag & FWRITE) != 0))
2882 KASSERT(0, ("wrong flags"));
2895 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2898 return (_fget(td, fd, fpp, 0, rightsp));
2902 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
2906 #ifndef CAPABILITIES
2907 error = _fget(td, fd, fpp, 0, rightsp);
2908 if (maxprotp != NULL)
2909 *maxprotp = VM_PROT_ALL;
2912 cap_rights_t fdrights;
2913 struct filedesc *fdp;
2918 fdp = td->td_proc->p_fd;
2919 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2921 error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
2922 if (__predict_false(error != 0))
2924 if (__predict_false(fp->f_ops == &badfileops)) {
2928 if (maxprotp != NULL)
2929 fdrights = *cap_rights(fdp, fd);
2930 if (!fd_modified(fdp, fd, seq))
2936 * If requested, convert capability rights to access flags.
2938 if (maxprotp != NULL)
2939 *maxprotp = cap_rights_to_vmprot(&fdrights);
2946 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2949 return (_fget(td, fd, fpp, FREAD, rightsp));
2953 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2956 return (_fget(td, fd, fpp, FWRITE, rightsp));
2960 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2963 struct filedesc *fdp = td->td_proc->p_fd;
2964 #ifndef CAPABILITIES
2965 return (fget_unlocked(fdp, fd, rightsp, fpp));
2972 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2974 error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
2977 error = cap_fcntl_check(fdp, fd, needfcntl);
2978 if (!fd_modified(fdp, fd, seq))
2992 * Like fget() but loads the underlying vnode, or returns an error if the
2993 * descriptor does not represent a vnode. Note that pipes use vnodes but
2994 * never have VM objects. The returned vnode will be vref()'d.
2996 * XXX: what about the unused flags ?
2999 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3006 error = _fget(td, fd, &fp, flags, needrightsp);
3009 if (fp->f_vnode == NULL) {
3021 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3024 return (_fgetvp(td, fd, 0, rightsp, vpp));
3028 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3029 struct filecaps *havecaps, struct vnode **vpp)
3031 struct filedesc *fdp;
3032 struct filecaps caps;
3036 fdp = td->td_proc->p_fd;
3037 error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
3040 if (fp->f_ops == &badfileops) {
3044 if (fp->f_vnode == NULL) {
3055 filecaps_free(&caps);
3060 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3063 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3067 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3070 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3075 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3079 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3084 * Handle the last reference to a file being closed.
3086 * Without the noinline attribute clang keeps inlining the func thorough this
3087 * file when fdrop is used.
3090 _fdrop(struct file *fp, struct thread *td)
3094 if (fp->f_count != 0)
3095 panic("fdrop: count %d", fp->f_count);
3096 error = fo_close(fp, td);
3097 atomic_subtract_int(&openfiles, 1);
3099 free(fp->f_advice, M_FADVISE);
3100 uma_zfree(file_zone, fp);
3106 * Apply an advisory lock on a file descriptor.
3108 * Just attempt to get a record lock of the requested type on the entire file
3109 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3111 #ifndef _SYS_SYSPROTO_H_
3119 sys_flock(struct thread *td, struct flock_args *uap)
3126 error = fget(td, uap->fd, &cap_flock_rights, &fp);
3129 if (fp->f_type != DTYPE_VNODE) {
3131 return (EOPNOTSUPP);
3135 lf.l_whence = SEEK_SET;
3138 if (uap->how & LOCK_UN) {
3139 lf.l_type = F_UNLCK;
3140 atomic_clear_int(&fp->f_flag, FHASLOCK);
3141 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3144 if (uap->how & LOCK_EX)
3145 lf.l_type = F_WRLCK;
3146 else if (uap->how & LOCK_SH)
3147 lf.l_type = F_RDLCK;
3152 atomic_set_int(&fp->f_flag, FHASLOCK);
3153 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3154 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3160 * Duplicate the specified descriptor to a free descriptor.
3163 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3164 int openerror, int *indxp)
3166 struct filedescent *newfde, *oldfde;
3171 KASSERT(openerror == ENODEV || openerror == ENXIO,
3172 ("unexpected error %d in %s", openerror, __func__));
3175 * If the to-be-dup'd fd number is greater than the allowed number
3176 * of file descriptors, or the fd to be dup'd has already been
3177 * closed, then reject.
3179 FILEDESC_XLOCK(fdp);
3180 if ((fp = fget_locked(fdp, dfd)) == NULL) {
3181 FILEDESC_XUNLOCK(fdp);
3185 error = fdalloc(td, 0, &indx);
3187 FILEDESC_XUNLOCK(fdp);
3192 * There are two cases of interest here.
3194 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3196 * For ENXIO steal away the file structure from (dfd) and store it in
3197 * (indx). (dfd) is effectively closed by this operation.
3199 switch (openerror) {
3202 * Check that the mode the file is being opened for is a
3203 * subset of the mode of the existing descriptor.
3205 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3206 fdunused(fdp, indx);
3207 FILEDESC_XUNLOCK(fdp);
3211 fdunused(fdp, indx);
3212 FILEDESC_XUNLOCK(fdp);
3215 newfde = &fdp->fd_ofiles[indx];
3216 oldfde = &fdp->fd_ofiles[dfd];
3217 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3219 seqc_write_begin(&newfde->fde_seqc);
3221 memcpy(newfde, oldfde, fde_change_size);
3222 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3225 seqc_write_end(&newfde->fde_seqc);
3230 * Steal away the file pointer from dfd and stuff it into indx.
3232 newfde = &fdp->fd_ofiles[indx];
3233 oldfde = &fdp->fd_ofiles[dfd];
3235 seqc_write_begin(&newfde->fde_seqc);
3237 memcpy(newfde, oldfde, fde_change_size);
3238 oldfde->fde_file = NULL;
3241 seqc_write_end(&newfde->fde_seqc);
3245 FILEDESC_XUNLOCK(fdp);
3251 * This sysctl determines if we will allow a process to chroot(2) if it
3252 * has a directory open:
3253 * 0: disallowed for all processes.
3254 * 1: allowed for processes that were not already chroot(2)'ed.
3255 * 2: allowed for all processes.
3258 static int chroot_allow_open_directories = 1;
3260 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3261 &chroot_allow_open_directories, 0,
3262 "Allow a process to chroot(2) if it has a directory open");
3265 * Helper function for raised chroot(2) security function: Refuse if
3266 * any filedescriptors are open directories.
3269 chroot_refuse_vdir_fds(struct filedesc *fdp)
3275 FILEDESC_LOCK_ASSERT(fdp);
3277 for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3278 fp = fget_locked(fdp, fd);
3281 if (fp->f_type == DTYPE_VNODE) {
3283 if (vp->v_type == VDIR)
3291 * Common routine for kern_chroot() and jail_attach(). The caller is
3292 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3293 * authorize this operation.
3296 pwd_chroot(struct thread *td, struct vnode *vp)
3298 struct filedesc *fdp;
3299 struct vnode *oldvp;
3302 fdp = td->td_proc->p_fd;
3303 FILEDESC_XLOCK(fdp);
3304 if (chroot_allow_open_directories == 0 ||
3305 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3306 error = chroot_refuse_vdir_fds(fdp);
3308 FILEDESC_XUNLOCK(fdp);
3312 oldvp = fdp->fd_rdir;
3315 if (fdp->fd_jdir == NULL) {
3319 FILEDESC_XUNLOCK(fdp);
3325 pwd_chdir(struct thread *td, struct vnode *vp)
3327 struct filedesc *fdp;
3328 struct vnode *oldvp;
3330 fdp = td->td_proc->p_fd;
3331 FILEDESC_XLOCK(fdp);
3332 VNASSERT(vp->v_usecount > 0, vp,
3333 ("chdir to a vnode with zero usecount"));
3334 oldvp = fdp->fd_cdir;
3336 FILEDESC_XUNLOCK(fdp);
3341 * Scan all active processes and prisons to see if any of them have a current
3342 * or root directory of `olddp'. If so, replace them with the new mount point.
3345 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3347 struct filedesc *fdp;
3352 if (vrefcnt(olddp) == 1)
3355 sx_slock(&allproc_lock);
3356 FOREACH_PROC_IN_SYSTEM(p) {
3362 FILEDESC_XLOCK(fdp);
3363 if (fdp->fd_cdir == olddp) {
3365 fdp->fd_cdir = newdp;
3368 if (fdp->fd_rdir == olddp) {
3370 fdp->fd_rdir = newdp;
3373 if (fdp->fd_jdir == olddp) {
3375 fdp->fd_jdir = newdp;
3378 FILEDESC_XUNLOCK(fdp);
3381 sx_sunlock(&allproc_lock);
3382 if (rootvnode == olddp) {
3387 mtx_lock(&prison0.pr_mtx);
3388 if (prison0.pr_root == olddp) {
3390 prison0.pr_root = newdp;
3393 mtx_unlock(&prison0.pr_mtx);
3394 sx_slock(&allprison_lock);
3395 TAILQ_FOREACH(pr, &allprison, pr_list) {
3396 mtx_lock(&pr->pr_mtx);
3397 if (pr->pr_root == olddp) {
3399 pr->pr_root = newdp;
3402 mtx_unlock(&pr->pr_mtx);
3404 sx_sunlock(&allprison_lock);
3409 struct filedesc_to_leader *
3410 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3412 struct filedesc_to_leader *fdtol;
3414 fdtol = malloc(sizeof(struct filedesc_to_leader),
3415 M_FILEDESC_TO_LEADER, M_WAITOK);
3416 fdtol->fdl_refcount = 1;
3417 fdtol->fdl_holdcount = 0;
3418 fdtol->fdl_wakeup = 0;
3419 fdtol->fdl_leader = leader;
3421 FILEDESC_XLOCK(fdp);
3422 fdtol->fdl_next = old->fdl_next;
3423 fdtol->fdl_prev = old;
3424 old->fdl_next = fdtol;
3425 fdtol->fdl_next->fdl_prev = fdtol;
3426 FILEDESC_XUNLOCK(fdp);
3428 fdtol->fdl_next = fdtol;
3429 fdtol->fdl_prev = fdtol;
3435 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3437 struct filedesc *fdp;
3438 int i, count, slots;
3440 if (*(int *)arg1 != 0)
3443 fdp = curproc->p_fd;
3445 FILEDESC_SLOCK(fdp);
3446 slots = NDSLOTS(fdp->fd_lastfile + 1);
3447 for (i = 0; i < slots; i++)
3448 count += bitcountl(fdp->fd_map[i]);
3449 FILEDESC_SUNLOCK(fdp);
3451 return (SYSCTL_OUT(req, &count, sizeof(count)));
3454 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3455 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3456 "Number of open file descriptors");
3459 * Get file structures globally.
3462 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3465 struct filedesc *fdp;
3470 error = sysctl_wire_old_buffer(req, 0);
3473 if (req->oldptr == NULL) {
3475 sx_slock(&allproc_lock);
3476 FOREACH_PROC_IN_SYSTEM(p) {
3478 if (p->p_state == PRS_NEW) {
3486 /* overestimates sparse tables. */
3487 if (fdp->fd_lastfile > 0)
3488 n += fdp->fd_lastfile;
3491 sx_sunlock(&allproc_lock);
3492 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3495 bzero(&xf, sizeof(xf));
3496 xf.xf_size = sizeof(xf);
3497 sx_slock(&allproc_lock);
3498 FOREACH_PROC_IN_SYSTEM(p) {
3500 if (p->p_state == PRS_NEW) {
3504 if (p_cansee(req->td, p) != 0) {
3508 xf.xf_pid = p->p_pid;
3509 xf.xf_uid = p->p_ucred->cr_uid;
3514 FILEDESC_SLOCK(fdp);
3515 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3516 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3519 xf.xf_file = (uintptr_t)fp;
3520 xf.xf_data = (uintptr_t)fp->f_data;
3521 xf.xf_vnode = (uintptr_t)fp->f_vnode;
3522 xf.xf_type = (uintptr_t)fp->f_type;
3523 xf.xf_count = fp->f_count;
3525 xf.xf_offset = foffset_get(fp);
3526 xf.xf_flag = fp->f_flag;
3527 error = SYSCTL_OUT(req, &xf, sizeof(xf));
3531 FILEDESC_SUNLOCK(fdp);
3536 sx_sunlock(&allproc_lock);
3540 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3541 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3543 #ifdef KINFO_FILE_SIZE
3544 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3548 xlate_fflags(int fflags)
3550 static const struct {
3553 } fflags_table[] = {
3554 { FAPPEND, KF_FLAG_APPEND },
3555 { FASYNC, KF_FLAG_ASYNC },
3556 { FFSYNC, KF_FLAG_FSYNC },
3557 { FHASLOCK, KF_FLAG_HASLOCK },
3558 { FNONBLOCK, KF_FLAG_NONBLOCK },
3559 { FREAD, KF_FLAG_READ },
3560 { FWRITE, KF_FLAG_WRITE },
3561 { O_CREAT, KF_FLAG_CREAT },
3562 { O_DIRECT, KF_FLAG_DIRECT },
3563 { O_EXCL, KF_FLAG_EXCL },
3564 { O_EXEC, KF_FLAG_EXEC },
3565 { O_EXLOCK, KF_FLAG_EXLOCK },
3566 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3567 { O_SHLOCK, KF_FLAG_SHLOCK },
3568 { O_TRUNC, KF_FLAG_TRUNC }
3574 for (i = 0; i < nitems(fflags_table); i++)
3575 if (fflags & fflags_table[i].fflag)
3576 kflags |= fflags_table[i].kf_fflag;
3580 /* Trim unused data from kf_path by truncating the structure size. */
3582 pack_kinfo(struct kinfo_file *kif)
3585 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3586 strlen(kif->kf_path) + 1;
3587 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3591 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3592 struct kinfo_file *kif, struct filedesc *fdp, int flags)
3596 bzero(kif, sizeof(*kif));
3598 /* Set a default type to allow for empty fill_kinfo() methods. */
3599 kif->kf_type = KF_TYPE_UNKNOWN;
3600 kif->kf_flags = xlate_fflags(fp->f_flag);
3601 if (rightsp != NULL)
3602 kif->kf_cap_rights = *rightsp;
3604 cap_rights_init(&kif->kf_cap_rights);
3606 kif->kf_ref_count = fp->f_count;
3607 kif->kf_offset = foffset_get(fp);
3610 * This may drop the filedesc lock, so the 'fp' cannot be
3611 * accessed after this call.
3613 error = fo_fill_kinfo(fp, kif, fdp);
3615 kif->kf_status |= KF_ATTR_VALID;
3616 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3619 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3623 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3624 struct kinfo_file *kif, int flags)
3628 bzero(kif, sizeof(*kif));
3630 kif->kf_type = KF_TYPE_VNODE;
3631 error = vn_fill_kinfo_vnode(vp, kif);
3633 kif->kf_status |= KF_ATTR_VALID;
3634 kif->kf_flags = xlate_fflags(fflags);
3635 cap_rights_init(&kif->kf_cap_rights);
3637 kif->kf_ref_count = -1;
3638 kif->kf_offset = -1;
3639 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3642 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3646 struct export_fd_buf {
3647 struct filedesc *fdp;
3650 struct kinfo_file kif;
3655 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3657 struct kinfo_file *kif;
3660 if (efbuf->remainder != -1) {
3661 if (efbuf->remainder < kif->kf_structsize) {
3662 /* Terminate export. */
3663 efbuf->remainder = 0;
3666 efbuf->remainder -= kif->kf_structsize;
3668 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3672 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3673 struct export_fd_buf *efbuf)
3677 if (efbuf->remainder == 0)
3679 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3681 FILEDESC_SUNLOCK(efbuf->fdp);
3682 error = export_kinfo_to_sb(efbuf);
3683 FILEDESC_SLOCK(efbuf->fdp);
3688 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3689 struct export_fd_buf *efbuf)
3693 if (efbuf->remainder == 0)
3695 if (efbuf->fdp != NULL)
3696 FILEDESC_SUNLOCK(efbuf->fdp);
3697 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3698 error = export_kinfo_to_sb(efbuf);
3699 if (efbuf->fdp != NULL)
3700 FILEDESC_SLOCK(efbuf->fdp);
3705 * Store a process file descriptor information to sbuf.
3707 * Takes a locked proc as argument, and returns with the proc unlocked.
3710 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
3714 struct filedesc *fdp;
3715 struct export_fd_buf *efbuf;
3716 struct vnode *cttyvp, *textvp, *tracevp;
3718 cap_rights_t rights;
3720 PROC_LOCK_ASSERT(p, MA_OWNED);
3723 tracevp = p->p_tracevp;
3724 if (tracevp != NULL)
3727 textvp = p->p_textvp;
3730 /* Controlling tty. */
3732 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3733 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3739 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3742 efbuf->remainder = maxlen;
3743 efbuf->flags = flags;
3744 if (tracevp != NULL)
3745 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3748 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3750 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3756 FILEDESC_SLOCK(fdp);
3757 /* working directory */
3758 if (fdp->fd_cdir != NULL) {
3759 vrefact(fdp->fd_cdir);
3760 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3762 /* root directory */
3763 if (fdp->fd_rdir != NULL) {
3764 vrefact(fdp->fd_rdir);
3765 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3767 /* jail directory */
3768 if (fdp->fd_jdir != NULL) {
3769 vrefact(fdp->fd_jdir);
3770 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3772 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3773 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3776 rights = *cap_rights(fdp, i);
3777 #else /* !CAPABILITIES */
3778 rights = cap_no_rights;
3781 * Create sysctl entry. It is OK to drop the filedesc
3782 * lock inside of export_file_to_sb() as we will
3783 * re-validate and re-evaluate its properties when the
3786 error = export_file_to_sb(fp, i, &rights, efbuf);
3787 if (error != 0 || efbuf->remainder == 0)
3790 FILEDESC_SUNLOCK(fdp);
3793 free(efbuf, M_TEMP);
3797 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
3800 * Get per-process file descriptors for use by procstat(1), et al.
3803 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3808 int error, error2, *name;
3812 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3813 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3814 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3819 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3820 error = kern_proc_filedesc_out(p, &sb, maxlen,
3821 KERN_FILEDESC_PACK_KINFO);
3822 error2 = sbuf_finish(&sb);
3824 return (error != 0 ? error : error2);
3827 #ifdef COMPAT_FREEBSD7
3828 #ifdef KINFO_OFILE_SIZE
3829 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3833 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3836 okif->kf_structsize = sizeof(*okif);
3837 okif->kf_type = kif->kf_type;
3838 okif->kf_fd = kif->kf_fd;
3839 okif->kf_ref_count = kif->kf_ref_count;
3840 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3841 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3842 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3843 okif->kf_offset = kif->kf_offset;
3844 if (kif->kf_type == KF_TYPE_VNODE)
3845 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
3847 okif->kf_vnode_type = KF_VTYPE_VNON;
3848 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3849 if (kif->kf_type == KF_TYPE_SOCKET) {
3850 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
3851 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
3852 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
3853 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
3854 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
3856 okif->kf_sa_local.ss_family = AF_UNSPEC;
3857 okif->kf_sa_peer.ss_family = AF_UNSPEC;
3862 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3863 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3868 FILEDESC_SUNLOCK(fdp);
3869 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3870 kinfo_to_okinfo(kif, okif);
3871 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3872 FILEDESC_SLOCK(fdp);
3877 * Get per-process file descriptors for use by procstat(1), et al.
3880 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3882 struct kinfo_ofile *okif;
3883 struct kinfo_file *kif;
3884 struct filedesc *fdp;
3885 int error, i, *name;
3890 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3897 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3898 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3899 FILEDESC_SLOCK(fdp);
3900 if (fdp->fd_cdir != NULL)
3901 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3903 if (fdp->fd_rdir != NULL)
3904 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3906 if (fdp->fd_jdir != NULL)
3907 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3909 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3910 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3912 export_file_to_kinfo(fp, i, NULL, kif, fdp,
3913 KERN_FILEDESC_PACK_KINFO);
3914 FILEDESC_SUNLOCK(fdp);
3915 kinfo_to_okinfo(kif, okif);
3916 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3917 FILEDESC_SLOCK(fdp);
3921 FILEDESC_SUNLOCK(fdp);
3928 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3929 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3930 "Process ofiledesc entries");
3931 #endif /* COMPAT_FREEBSD7 */
3934 vntype_to_kinfo(int vtype)
3939 } vtypes_table[] = {
3940 { VBAD, KF_VTYPE_VBAD },
3941 { VBLK, KF_VTYPE_VBLK },
3942 { VCHR, KF_VTYPE_VCHR },
3943 { VDIR, KF_VTYPE_VDIR },
3944 { VFIFO, KF_VTYPE_VFIFO },
3945 { VLNK, KF_VTYPE_VLNK },
3946 { VNON, KF_VTYPE_VNON },
3947 { VREG, KF_VTYPE_VREG },
3948 { VSOCK, KF_VTYPE_VSOCK }
3953 * Perform vtype translation.
3955 for (i = 0; i < nitems(vtypes_table); i++)
3956 if (vtypes_table[i].vtype == vtype)
3957 return (vtypes_table[i].kf_vtype);
3959 return (KF_VTYPE_UNKNOWN);
3962 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3963 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3964 "Process filedesc entries");
3967 * Store a process current working directory information to sbuf.
3969 * Takes a locked proc as argument, and returns with the proc unlocked.
3972 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
3974 struct filedesc *fdp;
3975 struct export_fd_buf *efbuf;
3978 PROC_LOCK_ASSERT(p, MA_OWNED);
3985 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3988 efbuf->remainder = maxlen;
3990 FILEDESC_SLOCK(fdp);
3991 if (fdp->fd_cdir == NULL)
3994 vrefact(fdp->fd_cdir);
3995 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3998 FILEDESC_SUNLOCK(fdp);
4000 free(efbuf, M_TEMP);
4005 * Get per-process current working directory.
4008 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4013 int error, error2, *name;
4017 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4018 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4019 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4024 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4025 error = kern_proc_cwd_out(p, &sb, maxlen);
4026 error2 = sbuf_finish(&sb);
4028 return (error != 0 ? error : error2);
4031 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4032 sysctl_kern_proc_cwd, "Process current working directory");
4036 * For the purposes of debugging, generate a human-readable string for the
4040 file_type_to_name(short type)
4068 case DTYPE_PROCDESC:
4070 case DTYPE_LINUXEFD:
4072 case DTYPE_LINUXTFD:
4080 * For the purposes of debugging, identify a process (if any, perhaps one of
4081 * many) that references the passed file in its file descriptor array. Return
4084 static struct proc *
4085 file_to_first_proc(struct file *fp)
4087 struct filedesc *fdp;
4091 FOREACH_PROC_IN_SYSTEM(p) {
4092 if (p->p_state == PRS_NEW)
4097 for (n = 0; n <= fdp->fd_lastfile; n++) {
4098 if (fp == fdp->fd_ofiles[n].fde_file)
4106 db_print_file(struct file *fp, int header)
4108 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4112 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4113 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4114 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4116 p = file_to_first_proc(fp);
4117 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4118 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4119 fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode,
4120 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4125 DB_SHOW_COMMAND(file, db_show_file)
4130 db_printf("usage: show file <addr>\n");
4133 fp = (struct file *)addr;
4134 db_print_file(fp, 1);
4137 DB_SHOW_COMMAND(files, db_show_files)
4139 struct filedesc *fdp;
4146 FOREACH_PROC_IN_SYSTEM(p) {
4147 if (p->p_state == PRS_NEW)
4149 if ((fdp = p->p_fd) == NULL)
4151 for (n = 0; n <= fdp->fd_lastfile; ++n) {
4152 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4154 db_print_file(fp, header);
4161 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4162 &maxfilesperproc, 0, "Maximum files allowed open per process");
4164 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4165 &maxfiles, 0, "Maximum number of files");
4167 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4168 &openfiles, 0, "System-wide number of open files");
4172 filelistinit(void *dummy)
4175 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4176 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4177 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4178 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4179 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4181 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4183 /*-------------------------------------------------------------------*/
4186 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4187 int flags, struct thread *td)
4194 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4202 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4210 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4218 badfo_kqfilter(struct file *fp, struct knote *kn)
4225 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4233 badfo_close(struct file *fp, struct thread *td)
4240 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4248 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4256 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4257 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4265 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4271 struct fileops badfileops = {
4272 .fo_read = badfo_readwrite,
4273 .fo_write = badfo_readwrite,
4274 .fo_truncate = badfo_truncate,
4275 .fo_ioctl = badfo_ioctl,
4276 .fo_poll = badfo_poll,
4277 .fo_kqfilter = badfo_kqfilter,
4278 .fo_stat = badfo_stat,
4279 .fo_close = badfo_close,
4280 .fo_chmod = badfo_chmod,
4281 .fo_chown = badfo_chown,
4282 .fo_sendfile = badfo_sendfile,
4283 .fo_fill_kinfo = badfo_fill_kinfo,
4287 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4288 int flags, struct thread *td)
4291 return (EOPNOTSUPP);
4295 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4303 invfo_ioctl(struct file *fp, u_long com, void *data,
4304 struct ucred *active_cred, struct thread *td)
4311 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4315 return (poll_no_poll(events));
4319 invfo_kqfilter(struct file *fp, struct knote *kn)
4326 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4334 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4342 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4343 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4350 /*-------------------------------------------------------------------*/
4353 * File Descriptor pseudo-device driver (/dev/fd/).
4355 * Opening minor device N dup()s the file (if any) connected to file
4356 * descriptor N belonging to the calling process. Note that this driver
4357 * consists of only the ``open()'' routine, because all subsequent
4358 * references to this file will be direct to the other driver.
4360 * XXX: we could give this one a cloning event handler if necessary.
4365 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4369 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4370 * the file descriptor being sought for duplication. The error
4371 * return ensures that the vnode for this device will be released
4372 * by vn_open. Open will detect this special error and take the
4373 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4374 * will simply report the error.
4376 td->td_dupfd = dev2unit(dev);
4380 static struct cdevsw fildesc_cdevsw = {
4381 .d_version = D_VERSION,
4387 fildesc_drvinit(void *unused)
4391 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4392 UID_ROOT, GID_WHEEL, 0666, "fd/0");
4393 make_dev_alias(dev, "stdin");
4394 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4395 UID_ROOT, GID_WHEEL, 0666, "fd/1");
4396 make_dev_alias(dev, "stdout");
4397 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4398 UID_ROOT, GID_WHEEL, 0666, "fd/2");
4399 make_dev_alias(dev, "stderr");
4402 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);