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 volatile 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;
506 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
509 case F_DUPFD_CLOEXEC:
511 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
516 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
519 case F_DUP2FD_CLOEXEC:
521 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
527 fde = fdeget_locked(fdp, fd);
530 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
533 FILEDESC_SUNLOCK(fdp);
539 fde = fdeget_locked(fdp, fd);
541 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
542 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
545 FILEDESC_XUNLOCK(fdp);
549 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
552 td->td_retval[0] = OFLAGS(fp->f_flag);
557 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
561 tmp = flg = fp->f_flag;
563 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
564 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
565 tmp = fp->f_flag & FNONBLOCK;
566 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
571 tmp = fp->f_flag & FASYNC;
572 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
577 atomic_clear_int(&fp->f_flag, FNONBLOCK);
579 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
584 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
587 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
589 td->td_retval[0] = tmp;
594 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
598 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
603 error = priv_check(td, PRIV_NFS_LOCKD);
611 /* FALLTHROUGH F_SETLK */
615 flp = (struct flock *)arg;
616 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
621 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL);
624 if (fp->f_type != DTYPE_VNODE) {
630 if (flp->l_whence == SEEK_CUR) {
631 foffset = foffset_get(fp);
634 foffset > OFF_MAX - flp->l_start)) {
639 flp->l_start += foffset;
643 switch (flp->l_type) {
645 if ((fp->f_flag & FREAD) == 0) {
649 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
650 PROC_LOCK(p->p_leader);
651 p->p_leader->p_flag |= P_ADVLOCK;
652 PROC_UNLOCK(p->p_leader);
654 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
658 if ((fp->f_flag & FWRITE) == 0) {
662 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
663 PROC_LOCK(p->p_leader);
664 p->p_leader->p_flag |= P_ADVLOCK;
665 PROC_UNLOCK(p->p_leader);
667 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
671 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
675 if (flg != F_REMOTE) {
679 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
680 F_UNLCKSYS, flp, flg);
686 if (error != 0 || flp->l_type == F_UNLCK ||
687 flp->l_type == F_UNLCKSYS) {
693 * Check for a race with close.
695 * The vnode is now advisory locked (or unlocked, but this case
696 * is not really important) as the caller requested.
697 * We had to drop the filedesc lock, so we need to recheck if
698 * the descriptor is still valid, because if it was closed
699 * in the meantime we need to remove advisory lock from the
700 * vnode - close on any descriptor leading to an advisory
701 * locked vnode, removes that lock.
702 * We will return 0 on purpose in that case, as the result of
703 * successful advisory lock might have been externally visible
704 * already. This is fine - effectively we pretend to the caller
705 * that the closing thread was a bit slower and that the
706 * advisory lock succeeded before the close.
708 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2, NULL);
714 flp->l_whence = SEEK_SET;
717 flp->l_type = F_UNLCK;
718 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
719 F_UNLCK, flp, F_POSIX);
726 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL);
729 if (fp->f_type != DTYPE_VNODE) {
734 flp = (struct flock *)arg;
735 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
736 flp->l_type != F_UNLCK) {
741 if (flp->l_whence == SEEK_CUR) {
742 foffset = foffset_get(fp);
743 if ((flp->l_start > 0 &&
744 foffset > OFF_MAX - flp->l_start) ||
746 foffset < OFF_MIN - flp->l_start)) {
751 flp->l_start += foffset;
754 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
760 arg = arg ? 128 * 1024: 0;
763 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp, NULL);
766 if (fp->f_type != DTYPE_VNODE) {
773 * Exclusive lock synchronizes against f_seqcount reads and
774 * writes in sequential_heuristic().
776 error = vn_lock(vp, LK_EXCLUSIVE);
782 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
783 fp->f_seqcount = (arg + bsize - 1) / bsize;
784 atomic_set_int(&fp->f_flag, FRDAHEAD);
786 atomic_clear_int(&fp->f_flag, FRDAHEAD);
800 getmaxfd(struct thread *td)
803 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
807 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
810 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
812 struct filedesc *fdp;
813 struct filedescent *oldfde, *newfde;
816 u_long *oioctls, *nioctls;
823 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
824 MPASS(mode < FDDUP_LASTMODE);
827 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
830 * Verify we have a valid descriptor to dup from and possibly to
831 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
832 * return EINVAL when the new descriptor is out of bounds.
837 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
838 maxfd = getmaxfd(td);
840 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
844 if (fget_locked(fdp, old) == NULL)
846 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
847 td->td_retval[0] = new;
848 if (flags & FDDUP_FLAG_CLOEXEC)
849 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
855 * If the caller specified a file descriptor, make sure the file
856 * table is large enough to hold it, and grab it. Otherwise, just
857 * allocate a new descriptor the usual way.
862 if ((error = fdalloc(td, new, &new)) != 0)
865 case FDDUP_MUSTREPLACE:
866 /* Target file descriptor must exist. */
867 if (fget_locked(fdp, new) == NULL)
871 if (new >= fdp->fd_nfiles) {
873 * The resource limits are here instead of e.g.
874 * fdalloc(), because the file descriptor table may be
875 * shared between processes, so we can't really use
876 * racct_add()/racct_sub(). Instead of counting the
877 * number of actually allocated descriptors, just put
878 * the limit on the size of the file descriptor table.
881 if (RACCT_ENABLED()) {
882 error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
889 fdgrowtable_exp(fdp, new + 1);
891 if (!fdisused(fdp, new))
895 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
898 KASSERT(old != new, ("new fd is same as old"));
900 oldfde = &fdp->fd_ofiles[old];
901 fhold(oldfde->fde_file);
902 newfde = &fdp->fd_ofiles[new];
903 delfp = newfde->fde_file;
905 oioctls = filecaps_free_prep(&newfde->fde_caps);
906 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
909 * Duplicate the source descriptor.
912 seqc_write_begin(&newfde->fde_seqc);
914 memcpy(newfde, oldfde, fde_change_size);
915 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
917 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
918 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
920 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
922 seqc_write_end(&newfde->fde_seqc);
924 td->td_retval[0] = new;
929 (void) closefp(fdp, new, delfp, td, 1);
930 FILEDESC_UNLOCK_ASSERT(fdp);
933 FILEDESC_XUNLOCK(fdp);
936 filecaps_free_finish(oioctls);
941 * If sigio is on the list associated with a process or process group,
942 * disable signalling from the device, remove sigio from the list and
946 funsetown(struct sigio **sigiop)
958 *(sigio->sio_myref) = NULL;
959 if ((sigio)->sio_pgid < 0) {
960 struct pgrp *pg = (sigio)->sio_pgrp;
962 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
966 struct proc *p = (sigio)->sio_proc;
968 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
973 crfree(sigio->sio_ucred);
974 free(sigio, M_SIGIO);
978 * Free a list of sigio structures.
979 * We only need to lock the SIGIO_LOCK because we have made ourselves
980 * inaccessible to callers of fsetown and therefore do not need to lock
981 * the proc or pgrp struct for the list manipulation.
984 funsetownlst(struct sigiolst *sigiolst)
990 sigio = SLIST_FIRST(sigiolst);
997 * Every entry of the list should belong
998 * to a single proc or pgrp.
1000 if (sigio->sio_pgid < 0) {
1001 pg = sigio->sio_pgrp;
1002 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
1003 } else /* if (sigio->sio_pgid > 0) */ {
1004 p = sigio->sio_proc;
1005 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1009 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1010 *(sigio->sio_myref) = NULL;
1012 KASSERT(sigio->sio_pgid < 0,
1013 ("Proc sigio in pgrp sigio list"));
1014 KASSERT(sigio->sio_pgrp == pg,
1015 ("Bogus pgrp in sigio list"));
1017 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1020 } else /* if (p != NULL) */ {
1021 KASSERT(sigio->sio_pgid > 0,
1022 ("Pgrp sigio in proc sigio list"));
1023 KASSERT(sigio->sio_proc == p,
1024 ("Bogus proc in sigio list"));
1026 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1031 crfree(sigio->sio_ucred);
1032 free(sigio, M_SIGIO);
1039 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1041 * After permission checking, add a sigio structure to the sigio list for
1042 * the process or process group.
1045 fsetown(pid_t pgid, struct sigio **sigiop)
1049 struct sigio *sigio;
1059 /* Allocate and fill in the new sigio out of locks. */
1060 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1061 sigio->sio_pgid = pgid;
1062 sigio->sio_ucred = crhold(curthread->td_ucred);
1063 sigio->sio_myref = sigiop;
1065 sx_slock(&proctree_lock);
1074 * Policy - Don't allow a process to FSETOWN a process
1075 * in another session.
1077 * Remove this test to allow maximum flexibility or
1078 * restrict FSETOWN to the current process or process
1079 * group for maximum safety.
1082 if (proc->p_session != curthread->td_proc->p_session) {
1088 } else /* if (pgid < 0) */ {
1089 pgrp = pgfind(-pgid);
1097 * Policy - Don't allow a process to FSETOWN a process
1098 * in another session.
1100 * Remove this test to allow maximum flexibility or
1101 * restrict FSETOWN to the current process or process
1102 * group for maximum safety.
1104 if (pgrp->pg_session != curthread->td_proc->p_session) {
1115 * Since funsetownlst() is called without the proctree
1116 * locked, we need to check for P_WEXIT.
1117 * XXX: is ESRCH correct?
1119 if ((proc->p_flag & P_WEXIT) != 0) {
1124 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1125 sigio->sio_proc = proc;
1129 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1130 sigio->sio_pgrp = pgrp;
1133 sx_sunlock(&proctree_lock);
1140 sx_sunlock(&proctree_lock);
1141 crfree(sigio->sio_ucred);
1142 free(sigio, M_SIGIO);
1147 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1150 fgetown(struct sigio **sigiop)
1155 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1161 * Function drops the filedesc lock on return.
1164 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1169 FILEDESC_XLOCK_ASSERT(fdp);
1172 if (td->td_proc->p_fdtol != NULL) {
1174 * Ask fdfree() to sleep to ensure that all relevant
1175 * process leaders can be traversed in closef().
1177 fdp->fd_holdleaderscount++;
1184 * We now hold the fp reference that used to be owned by the
1185 * descriptor array. We have to unlock the FILEDESC *AFTER*
1186 * knote_fdclose to prevent a race of the fd getting opened, a knote
1187 * added, and deleteing a knote for the new fd.
1189 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1190 knote_fdclose(td, fd);
1193 * We need to notify mqueue if the object is of type mqueue.
1195 if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1196 mq_fdclose(td, fd, fp);
1197 FILEDESC_XUNLOCK(fdp);
1199 error = closef(fp, td);
1201 FILEDESC_XLOCK(fdp);
1202 fdp->fd_holdleaderscount--;
1203 if (fdp->fd_holdleaderscount == 0 &&
1204 fdp->fd_holdleaderswakeup != 0) {
1205 fdp->fd_holdleaderswakeup = 0;
1206 wakeup(&fdp->fd_holdleaderscount);
1208 FILEDESC_XUNLOCK(fdp);
1214 * Close a file descriptor.
1216 #ifndef _SYS_SYSPROTO_H_
1223 sys_close(struct thread *td, struct close_args *uap)
1226 return (kern_close(td, uap->fd));
1230 kern_close(struct thread *td, int fd)
1232 struct filedesc *fdp;
1235 fdp = td->td_proc->p_fd;
1237 AUDIT_SYSCLOSE(td, fd);
1239 FILEDESC_XLOCK(fdp);
1240 if ((fp = fget_locked(fdp, fd)) == NULL) {
1241 FILEDESC_XUNLOCK(fdp);
1246 /* closefp() drops the FILEDESC lock for us. */
1247 return (closefp(fdp, fd, fp, td, 1));
1251 * Close open file descriptors.
1253 #ifndef _SYS_SYSPROTO_H_
1254 struct closefrom_args {
1260 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1262 struct filedesc *fdp;
1265 fdp = td->td_proc->p_fd;
1266 AUDIT_ARG_FD(uap->lowfd);
1269 * Treat negative starting file descriptor values identical to
1270 * closefrom(0) which closes all files.
1274 FILEDESC_SLOCK(fdp);
1275 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1276 if (fdp->fd_ofiles[fd].fde_file != NULL) {
1277 FILEDESC_SUNLOCK(fdp);
1278 (void)kern_close(td, fd);
1279 FILEDESC_SLOCK(fdp);
1282 FILEDESC_SUNLOCK(fdp);
1286 #if defined(COMPAT_43)
1288 * Return status information about a file descriptor.
1290 #ifndef _SYS_SYSPROTO_H_
1291 struct ofstat_args {
1298 ofstat(struct thread *td, struct ofstat_args *uap)
1304 error = kern_fstat(td, uap->fd, &ub);
1307 error = copyout(&oub, uap->sb, sizeof(oub));
1311 #endif /* COMPAT_43 */
1313 #if defined(COMPAT_FREEBSD11)
1315 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1318 struct freebsd11_stat osb;
1321 error = kern_fstat(td, uap->fd, &sb);
1324 error = freebsd11_cvtstat(&sb, &osb);
1326 error = copyout(&osb, uap->sb, sizeof(osb));
1329 #endif /* COMPAT_FREEBSD11 */
1332 * Return status information about a file descriptor.
1334 #ifndef _SYS_SYSPROTO_H_
1342 sys_fstat(struct thread *td, struct fstat_args *uap)
1347 error = kern_fstat(td, uap->fd, &ub);
1349 error = copyout(&ub, uap->sb, sizeof(ub));
1354 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1361 error = fget(td, fd, &cap_fstat_rights, &fp);
1365 AUDIT_ARG_FILE(td->td_proc, fp);
1367 error = fo_stat(fp, sbp, td->td_ucred, td);
1369 #ifdef __STAT_TIME_T_EXT
1371 sbp->st_atim_ext = 0;
1372 sbp->st_mtim_ext = 0;
1373 sbp->st_ctim_ext = 0;
1374 sbp->st_btim_ext = 0;
1378 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1384 #if defined(COMPAT_FREEBSD11)
1386 * Return status information about a file descriptor.
1388 #ifndef _SYS_SYSPROTO_H_
1389 struct freebsd11_nfstat_args {
1396 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1402 error = kern_fstat(td, uap->fd, &ub);
1404 freebsd11_cvtnstat(&ub, &nub);
1405 error = copyout(&nub, uap->sb, sizeof(nub));
1409 #endif /* COMPAT_FREEBSD11 */
1412 * Return pathconf information about a file descriptor.
1414 #ifndef _SYS_SYSPROTO_H_
1415 struct fpathconf_args {
1422 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1427 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1429 td->td_retval[0] = value;
1434 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1440 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1444 if (name == _PC_ASYNC_IO) {
1445 *valuep = _POSIX_ASYNCHRONOUS_IO;
1450 vn_lock(vp, LK_SHARED | LK_RETRY);
1451 error = VOP_PATHCONF(vp, name, valuep);
1453 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1454 if (name != _PC_PIPE_BUF) {
1469 * Copy filecaps structure allocating memory for ioctls array if needed.
1471 * The last parameter indicates whether the fdtable is locked. If it is not and
1472 * ioctls are encountered, copying fails and the caller must lock the table.
1474 * Note that if the table was not locked, the caller has to check the relevant
1475 * sequence counter to determine whether the operation was successful.
1478 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1482 if (src->fc_ioctls != NULL && !locked)
1484 memcpy(dst, src, sizeof(*src));
1485 if (src->fc_ioctls == NULL)
1488 KASSERT(src->fc_nioctls > 0,
1489 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1491 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1492 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1493 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1498 filecaps_copy_prep(const struct filecaps *src)
1503 if (__predict_true(src->fc_ioctls == NULL))
1506 KASSERT(src->fc_nioctls > 0,
1507 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1509 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1510 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1515 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1521 if (__predict_true(src->fc_ioctls == NULL)) {
1522 MPASS(ioctls == NULL);
1526 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1527 dst->fc_ioctls = ioctls;
1528 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1532 * Move filecaps structure to the new place and clear the old place.
1535 filecaps_move(struct filecaps *src, struct filecaps *dst)
1539 bzero(src, sizeof(*src));
1543 * Fill the given filecaps structure with full rights.
1546 filecaps_fill(struct filecaps *fcaps)
1549 CAP_ALL(&fcaps->fc_rights);
1550 fcaps->fc_ioctls = NULL;
1551 fcaps->fc_nioctls = -1;
1552 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1556 * Free memory allocated within filecaps structure.
1559 filecaps_free(struct filecaps *fcaps)
1562 free(fcaps->fc_ioctls, M_FILECAPS);
1563 bzero(fcaps, sizeof(*fcaps));
1567 filecaps_free_prep(struct filecaps *fcaps)
1571 ioctls = fcaps->fc_ioctls;
1572 bzero(fcaps, sizeof(*fcaps));
1577 filecaps_free_finish(u_long *ioctls)
1580 free(ioctls, M_FILECAPS);
1584 * Validate the given filecaps structure.
1587 filecaps_validate(const struct filecaps *fcaps, const char *func)
1590 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1591 ("%s: invalid rights", func));
1592 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1593 ("%s: invalid fcntls", func));
1594 KASSERT(fcaps->fc_fcntls == 0 ||
1595 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1596 ("%s: fcntls without CAP_FCNTL", func));
1597 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1598 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1599 ("%s: invalid ioctls", func));
1600 KASSERT(fcaps->fc_nioctls == 0 ||
1601 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1602 ("%s: ioctls without CAP_IOCTL", func));
1606 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1610 FILEDESC_XLOCK_ASSERT(fdp);
1612 nfd1 = fdp->fd_nfiles * 2;
1615 fdgrowtable(fdp, nfd1);
1619 * Grow the file table to accommodate (at least) nfd descriptors.
1622 fdgrowtable(struct filedesc *fdp, int nfd)
1624 struct filedesc0 *fdp0;
1625 struct freetable *ft;
1626 struct fdescenttbl *ntable;
1627 struct fdescenttbl *otable;
1628 int nnfiles, onfiles;
1629 NDSLOTTYPE *nmap, *omap;
1632 * If lastfile is -1 this struct filedesc was just allocated and we are
1633 * growing it to accommodate for the one we are going to copy from. There
1634 * is no need to have a lock on this one as it's not visible to anyone.
1636 if (fdp->fd_lastfile != -1)
1637 FILEDESC_XLOCK_ASSERT(fdp);
1639 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1641 /* save old values */
1642 onfiles = fdp->fd_nfiles;
1643 otable = fdp->fd_files;
1646 /* compute the size of the new table */
1647 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1648 if (nnfiles <= onfiles)
1649 /* the table is already large enough */
1653 * Allocate a new table. We need enough space for the number of
1654 * entries, file entries themselves and the struct freetable we will use
1655 * when we decommission the table and place it on the freelist.
1656 * We place the struct freetable in the middle so we don't have
1657 * to worry about padding.
1659 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1660 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1661 sizeof(struct freetable),
1662 M_FILEDESC, M_ZERO | M_WAITOK);
1663 /* copy the old data */
1664 ntable->fdt_nfiles = nnfiles;
1665 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1666 onfiles * sizeof(ntable->fdt_ofiles[0]));
1669 * Allocate a new map only if the old is not large enough. It will
1670 * grow at a slower rate than the table as it can map more
1671 * entries than the table can hold.
1673 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1674 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1676 /* copy over the old data and update the pointer */
1677 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1682 * Make sure that ntable is correctly initialized before we replace
1683 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1686 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1689 * Do not free the old file table, as some threads may still
1690 * reference entries within it. Instead, place it on a freelist
1691 * which will be processed when the struct filedesc is released.
1693 * Note that if onfiles == NDFILE, we're dealing with the original
1694 * static allocation contained within (struct filedesc0 *)fdp,
1695 * which must not be freed.
1697 if (onfiles > NDFILE) {
1698 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1699 fdp0 = (struct filedesc0 *)fdp;
1700 ft->ft_table = otable;
1701 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1704 * The map does not have the same possibility of threads still
1705 * holding references to it. So always free it as long as it
1706 * does not reference the original static allocation.
1708 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1709 free(omap, M_FILEDESC);
1713 * Allocate a file descriptor for the process.
1716 fdalloc(struct thread *td, int minfd, int *result)
1718 struct proc *p = td->td_proc;
1719 struct filedesc *fdp = p->p_fd;
1720 int fd, maxfd, allocfd;
1725 FILEDESC_XLOCK_ASSERT(fdp);
1727 if (fdp->fd_freefile > minfd)
1728 minfd = fdp->fd_freefile;
1730 maxfd = getmaxfd(td);
1733 * Search the bitmap for a free descriptor starting at minfd.
1734 * If none is found, grow the file table.
1736 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1739 if (fd >= fdp->fd_nfiles) {
1740 allocfd = min(fd * 2, maxfd);
1742 if (RACCT_ENABLED()) {
1743 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1749 * fd is already equal to first free descriptor >= minfd, so
1750 * we only need to grow the table and we are done.
1752 fdgrowtable_exp(fdp, allocfd);
1756 * Perform some sanity checks, then mark the file descriptor as
1757 * used and return it to the caller.
1759 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1760 ("invalid descriptor %d", fd));
1761 KASSERT(!fdisused(fdp, fd),
1762 ("fd_first_free() returned non-free descriptor"));
1763 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1764 ("file descriptor isn't free"));
1771 * Allocate n file descriptors for the process.
1774 fdallocn(struct thread *td, int minfd, int *fds, int n)
1776 struct proc *p = td->td_proc;
1777 struct filedesc *fdp = p->p_fd;
1780 FILEDESC_XLOCK_ASSERT(fdp);
1782 for (i = 0; i < n; i++)
1783 if (fdalloc(td, 0, &fds[i]) != 0)
1787 for (i--; i >= 0; i--)
1788 fdunused(fdp, fds[i]);
1796 * Create a new open file structure and allocate a file descriptor for the
1797 * process that refers to it. We add one reference to the file for the
1798 * descriptor table and one reference for resultfp. This is to prevent us
1799 * being preempted and the entry in the descriptor table closed after we
1800 * release the FILEDESC lock.
1803 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1804 struct filecaps *fcaps)
1809 error = falloc_noinstall(td, &fp);
1811 return (error); /* no reference held on error */
1813 error = finstall(td, fp, &fd, flags, fcaps);
1815 fdrop(fp, td); /* one reference (fp only) */
1819 if (resultfp != NULL)
1820 *resultfp = fp; /* copy out result */
1822 fdrop(fp, td); /* release local reference */
1824 if (resultfd != NULL)
1831 * Create a new open file structure without allocating a file descriptor.
1834 falloc_noinstall(struct thread *td, struct file **resultfp)
1837 int maxuserfiles = maxfiles - (maxfiles / 20);
1839 static struct timeval lastfail;
1842 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1844 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1845 if ((openfiles_new >= maxuserfiles &&
1846 priv_check(td, PRIV_MAXFILES) != 0) ||
1847 openfiles_new >= maxfiles) {
1848 atomic_subtract_int(&openfiles, 1);
1849 if (ppsratecheck(&lastfail, &curfail, 1)) {
1850 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1851 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1855 fp = uma_zalloc(file_zone, M_WAITOK);
1856 bzero(fp, sizeof(*fp));
1857 refcount_init(&fp->f_count, 1);
1858 fp->f_cred = crhold(td->td_ucred);
1859 fp->f_ops = &badfileops;
1865 * Install a file in a file descriptor table.
1868 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1869 struct filecaps *fcaps)
1871 struct filedescent *fde;
1875 filecaps_validate(fcaps, __func__);
1876 FILEDESC_XLOCK_ASSERT(fdp);
1878 fde = &fdp->fd_ofiles[fd];
1880 seqc_write_begin(&fde->fde_seqc);
1883 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1885 filecaps_move(fcaps, &fde->fde_caps);
1887 filecaps_fill(&fde->fde_caps);
1889 seqc_write_end(&fde->fde_seqc);
1894 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1895 struct filecaps *fcaps)
1897 struct filedesc *fdp = td->td_proc->p_fd;
1902 FILEDESC_XLOCK(fdp);
1903 if ((error = fdalloc(td, 0, fd))) {
1904 FILEDESC_XUNLOCK(fdp);
1908 _finstall(fdp, fp, *fd, flags, fcaps);
1909 FILEDESC_XUNLOCK(fdp);
1914 * Build a new filedesc structure from another.
1915 * Copy the current, root, and jail root vnode references.
1917 * If fdp is not NULL, return with it shared locked.
1920 fdinit(struct filedesc *fdp, bool prepfiles)
1922 struct filedesc0 *newfdp0;
1923 struct filedesc *newfdp;
1925 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1926 newfdp = &newfdp0->fd_fd;
1928 /* Create the file descriptor table. */
1929 FILEDESC_LOCK_INIT(newfdp);
1930 refcount_init(&newfdp->fd_refcnt, 1);
1931 refcount_init(&newfdp->fd_holdcnt, 1);
1932 newfdp->fd_cmask = CMASK;
1933 newfdp->fd_map = newfdp0->fd_dmap;
1934 newfdp->fd_lastfile = -1;
1935 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1936 newfdp->fd_files->fdt_nfiles = NDFILE;
1941 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1942 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1944 FILEDESC_SLOCK(fdp);
1945 newfdp->fd_cdir = fdp->fd_cdir;
1946 if (newfdp->fd_cdir)
1947 vrefact(newfdp->fd_cdir);
1948 newfdp->fd_rdir = fdp->fd_rdir;
1949 if (newfdp->fd_rdir)
1950 vrefact(newfdp->fd_rdir);
1951 newfdp->fd_jdir = fdp->fd_jdir;
1952 if (newfdp->fd_jdir)
1953 vrefact(newfdp->fd_jdir);
1956 FILEDESC_SUNLOCK(fdp);
1958 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1959 FILEDESC_SUNLOCK(fdp);
1960 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1961 FILEDESC_SLOCK(fdp);
1968 static struct filedesc *
1969 fdhold(struct proc *p)
1971 struct filedesc *fdp;
1973 PROC_LOCK_ASSERT(p, MA_OWNED);
1976 refcount_acquire(&fdp->fd_holdcnt);
1981 fddrop(struct filedesc *fdp)
1984 if (fdp->fd_holdcnt > 1) {
1985 if (refcount_release(&fdp->fd_holdcnt) == 0)
1989 FILEDESC_LOCK_DESTROY(fdp);
1990 uma_zfree(filedesc0_zone, fdp);
1994 * Share a filedesc structure.
1997 fdshare(struct filedesc *fdp)
2000 refcount_acquire(&fdp->fd_refcnt);
2005 * Unshare a filedesc structure, if necessary by making a copy
2008 fdunshare(struct thread *td)
2010 struct filedesc *tmp;
2011 struct proc *p = td->td_proc;
2013 if (p->p_fd->fd_refcnt == 1)
2016 tmp = fdcopy(p->p_fd);
2022 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2026 td->td_proc->p_fd = fdp;
2030 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2031 * this is to ease callers, not catch errors.
2034 fdcopy(struct filedesc *fdp)
2036 struct filedesc *newfdp;
2037 struct filedescent *nfde, *ofde;
2042 newfdp = fdinit(fdp, true);
2043 /* copy all passable descriptors (i.e. not kqueue) */
2044 newfdp->fd_freefile = -1;
2045 for (i = 0; i <= fdp->fd_lastfile; ++i) {
2046 ofde = &fdp->fd_ofiles[i];
2047 if (ofde->fde_file == NULL ||
2048 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2049 if (newfdp->fd_freefile == -1)
2050 newfdp->fd_freefile = i;
2053 nfde = &newfdp->fd_ofiles[i];
2055 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2056 fhold(nfde->fde_file);
2057 fdused_init(newfdp, i);
2058 newfdp->fd_lastfile = i;
2060 if (newfdp->fd_freefile == -1)
2061 newfdp->fd_freefile = i;
2062 newfdp->fd_cmask = fdp->fd_cmask;
2063 FILEDESC_SUNLOCK(fdp);
2068 * Copies a filedesc structure, while remapping all file descriptors
2069 * stored inside using a translation table.
2071 * File descriptors are copied over to the new file descriptor table,
2072 * regardless of whether the close-on-exec flag is set.
2075 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2076 struct filedesc **ret)
2078 struct filedesc *newfdp;
2079 struct filedescent *nfde, *ofde;
2084 newfdp = fdinit(fdp, true);
2085 if (nfds > fdp->fd_lastfile + 1) {
2086 /* New table cannot be larger than the old one. */
2090 /* Copy all passable descriptors (i.e. not kqueue). */
2091 newfdp->fd_freefile = nfds;
2092 for (i = 0; i < nfds; ++i) {
2093 if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2094 /* File descriptor out of bounds. */
2098 ofde = &fdp->fd_ofiles[fds[i]];
2099 if (ofde->fde_file == NULL) {
2100 /* Unused file descriptor. */
2104 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2105 /* File descriptor cannot be passed. */
2109 nfde = &newfdp->fd_ofiles[i];
2111 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2112 fhold(nfde->fde_file);
2113 fdused_init(newfdp, i);
2114 newfdp->fd_lastfile = i;
2116 newfdp->fd_cmask = fdp->fd_cmask;
2117 FILEDESC_SUNLOCK(fdp);
2121 FILEDESC_SUNLOCK(fdp);
2122 fdescfree_remapped(newfdp);
2127 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2128 * one of processes using it exits) and the table used to be shared.
2131 fdclearlocks(struct thread *td)
2133 struct filedesc *fdp;
2134 struct filedesc_to_leader *fdtol;
2144 MPASS(fdtol != NULL);
2146 FILEDESC_XLOCK(fdp);
2147 KASSERT(fdtol->fdl_refcount > 0,
2148 ("filedesc_to_refcount botch: fdl_refcount=%d",
2149 fdtol->fdl_refcount));
2150 if (fdtol->fdl_refcount == 1 &&
2151 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2152 for (i = 0; i <= fdp->fd_lastfile; i++) {
2153 fp = fdp->fd_ofiles[i].fde_file;
2154 if (fp == NULL || fp->f_type != DTYPE_VNODE)
2157 FILEDESC_XUNLOCK(fdp);
2158 lf.l_whence = SEEK_SET;
2161 lf.l_type = F_UNLCK;
2163 (void) VOP_ADVLOCK(vp,
2164 (caddr_t)p->p_leader, F_UNLCK,
2166 FILEDESC_XLOCK(fdp);
2171 if (fdtol->fdl_refcount == 1) {
2172 if (fdp->fd_holdleaderscount > 0 &&
2173 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2175 * close() or kern_dup() has cleared a reference
2176 * in a shared file descriptor table.
2178 fdp->fd_holdleaderswakeup = 1;
2179 sx_sleep(&fdp->fd_holdleaderscount,
2180 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2183 if (fdtol->fdl_holdcount > 0) {
2185 * Ensure that fdtol->fdl_leader remains
2186 * valid in closef().
2188 fdtol->fdl_wakeup = 1;
2189 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2194 fdtol->fdl_refcount--;
2195 if (fdtol->fdl_refcount == 0 &&
2196 fdtol->fdl_holdcount == 0) {
2197 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2198 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2202 FILEDESC_XUNLOCK(fdp);
2204 free(fdtol, M_FILEDESC_TO_LEADER);
2208 * Release a filedesc structure.
2211 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2213 struct filedesc0 *fdp0;
2214 struct freetable *ft, *tft;
2215 struct filedescent *fde;
2219 for (i = 0; i <= fdp->fd_lastfile; i++) {
2220 fde = &fdp->fd_ofiles[i];
2225 (void) closef(fp, td);
2231 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2232 free(fdp->fd_map, M_FILEDESC);
2233 if (fdp->fd_nfiles > NDFILE)
2234 free(fdp->fd_files, M_FILEDESC);
2236 fdp0 = (struct filedesc0 *)fdp;
2237 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2238 free(ft->ft_table, M_FILEDESC);
2244 fdescfree(struct thread *td)
2247 struct filedesc *fdp;
2248 struct vnode *cdir, *jdir, *rdir;
2255 if (RACCT_ENABLED())
2256 racct_set_unlocked(p, RACCT_NOFILE, 0);
2259 if (p->p_fdtol != NULL)
2266 if (refcount_release(&fdp->fd_refcnt) == 0)
2269 FILEDESC_XLOCK(fdp);
2270 cdir = fdp->fd_cdir;
2271 fdp->fd_cdir = NULL;
2272 rdir = fdp->fd_rdir;
2273 fdp->fd_rdir = NULL;
2274 jdir = fdp->fd_jdir;
2275 fdp->fd_jdir = NULL;
2276 FILEDESC_XUNLOCK(fdp);
2285 fdescfree_fds(td, fdp, 1);
2289 fdescfree_remapped(struct filedesc *fdp)
2292 if (fdp->fd_cdir != NULL)
2293 vrele(fdp->fd_cdir);
2294 if (fdp->fd_rdir != NULL)
2295 vrele(fdp->fd_rdir);
2296 if (fdp->fd_jdir != NULL)
2297 vrele(fdp->fd_jdir);
2299 fdescfree_fds(curthread, fdp, 0);
2303 * For setugid programs, we don't want to people to use that setugidness
2304 * to generate error messages which write to a file which otherwise would
2305 * otherwise be off-limits to the process. We check for filesystems where
2306 * the vnode can change out from under us after execve (like [lin]procfs).
2308 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2309 * sufficient. We also don't check for setugidness since we know we are.
2312 is_unsafe(struct file *fp)
2316 if (fp->f_type != DTYPE_VNODE)
2320 return ((vp->v_vflag & VV_PROCDEP) != 0);
2324 * Make this setguid thing safe, if at all possible.
2327 fdsetugidsafety(struct thread *td)
2329 struct filedesc *fdp;
2333 fdp = td->td_proc->p_fd;
2334 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2335 MPASS(fdp->fd_nfiles >= 3);
2336 for (i = 0; i <= 2; i++) {
2337 fp = fdp->fd_ofiles[i].fde_file;
2338 if (fp != NULL && is_unsafe(fp)) {
2339 FILEDESC_XLOCK(fdp);
2340 knote_fdclose(td, i);
2342 * NULL-out descriptor prior to close to avoid
2343 * a race while close blocks.
2346 FILEDESC_XUNLOCK(fdp);
2347 (void) closef(fp, td);
2353 * If a specific file object occupies a specific file descriptor, close the
2354 * file descriptor entry and drop a reference on the file object. This is a
2355 * convenience function to handle a subsequent error in a function that calls
2356 * falloc() that handles the race that another thread might have closed the
2357 * file descriptor out from under the thread creating the file object.
2360 fdclose(struct thread *td, struct file *fp, int idx)
2362 struct filedesc *fdp = td->td_proc->p_fd;
2364 FILEDESC_XLOCK(fdp);
2365 if (fdp->fd_ofiles[idx].fde_file == fp) {
2367 FILEDESC_XUNLOCK(fdp);
2370 FILEDESC_XUNLOCK(fdp);
2374 * Close any files on exec?
2377 fdcloseexec(struct thread *td)
2379 struct filedesc *fdp;
2380 struct filedescent *fde;
2384 fdp = td->td_proc->p_fd;
2385 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2386 for (i = 0; i <= fdp->fd_lastfile; i++) {
2387 fde = &fdp->fd_ofiles[i];
2389 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2390 (fde->fde_flags & UF_EXCLOSE))) {
2391 FILEDESC_XLOCK(fdp);
2393 (void) closefp(fdp, i, fp, td, 0);
2394 FILEDESC_UNLOCK_ASSERT(fdp);
2400 * It is unsafe for set[ug]id processes to be started with file
2401 * descriptors 0..2 closed, as these descriptors are given implicit
2402 * significance in the Standard C library. fdcheckstd() will create a
2403 * descriptor referencing /dev/null for each of stdin, stdout, and
2404 * stderr that is not already open.
2407 fdcheckstd(struct thread *td)
2409 struct filedesc *fdp;
2411 int i, error, devnull;
2413 fdp = td->td_proc->p_fd;
2414 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2415 MPASS(fdp->fd_nfiles >= 3);
2417 for (i = 0; i <= 2; i++) {
2418 if (fdp->fd_ofiles[i].fde_file != NULL)
2421 save = td->td_retval[0];
2422 if (devnull != -1) {
2423 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2425 error = kern_openat(td, AT_FDCWD, "/dev/null",
2426 UIO_SYSSPACE, O_RDWR, 0);
2428 devnull = td->td_retval[0];
2429 KASSERT(devnull == i, ("we didn't get our fd"));
2432 td->td_retval[0] = save;
2440 * Internal form of close. Decrement reference count on file structure.
2441 * Note: td may be NULL when closing a file that was being passed in a
2445 closef(struct file *fp, struct thread *td)
2449 struct filedesc_to_leader *fdtol;
2450 struct filedesc *fdp;
2453 * POSIX record locking dictates that any close releases ALL
2454 * locks owned by this process. This is handled by setting
2455 * a flag in the unlock to free ONLY locks obeying POSIX
2456 * semantics, and not to free BSD-style file locks.
2457 * If the descriptor was in a message, POSIX-style locks
2458 * aren't passed with the descriptor, and the thread pointer
2459 * will be NULL. Callers should be careful only to pass a
2460 * NULL thread pointer when there really is no owning
2461 * context that might have locks, or the locks will be
2464 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2466 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2467 lf.l_whence = SEEK_SET;
2470 lf.l_type = F_UNLCK;
2471 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2472 F_UNLCK, &lf, F_POSIX);
2474 fdtol = td->td_proc->p_fdtol;
2475 if (fdtol != NULL) {
2477 * Handle special case where file descriptor table is
2478 * shared between multiple process leaders.
2480 fdp = td->td_proc->p_fd;
2481 FILEDESC_XLOCK(fdp);
2482 for (fdtol = fdtol->fdl_next;
2483 fdtol != td->td_proc->p_fdtol;
2484 fdtol = fdtol->fdl_next) {
2485 if ((fdtol->fdl_leader->p_flag &
2488 fdtol->fdl_holdcount++;
2489 FILEDESC_XUNLOCK(fdp);
2490 lf.l_whence = SEEK_SET;
2493 lf.l_type = F_UNLCK;
2495 (void) VOP_ADVLOCK(vp,
2496 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2498 FILEDESC_XLOCK(fdp);
2499 fdtol->fdl_holdcount--;
2500 if (fdtol->fdl_holdcount == 0 &&
2501 fdtol->fdl_wakeup != 0) {
2502 fdtol->fdl_wakeup = 0;
2506 FILEDESC_XUNLOCK(fdp);
2509 return (fdrop(fp, td));
2513 * Initialize the file pointer with the specified properties.
2515 * The ops are set with release semantics to be certain that the flags, type,
2516 * and data are visible when ops is. This is to prevent ops methods from being
2517 * called with bad data.
2520 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2525 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2529 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2530 struct file **fpp, struct filecaps *havecapsp)
2532 struct filedescent *fde;
2535 FILEDESC_LOCK_ASSERT(fdp);
2537 fde = fdeget_locked(fdp, fd);
2544 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2549 if (havecapsp != NULL)
2550 filecaps_copy(&fde->fde_caps, havecapsp, true);
2552 *fpp = fde->fde_file;
2560 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2561 struct file **fpp, struct filecaps *havecapsp)
2563 struct filedesc *fdp = td->td_proc->p_fd;
2565 #ifndef CAPABILITIES
2566 error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2567 if (error == 0 && havecapsp != NULL)
2568 filecaps_fill(havecapsp);
2574 error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2578 if (havecapsp != NULL) {
2579 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2580 havecapsp, false)) {
2586 if (!fd_modified(fdp, fd, seq))
2595 FILEDESC_SLOCK(fdp);
2596 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2599 FILEDESC_SUNLOCK(fdp);
2605 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2606 struct file **fpp, seqc_t *seqp)
2609 const struct filedescent *fde;
2611 const struct fdescenttbl *fdt;
2616 cap_rights_t haverights;
2620 fdt = fdp->fd_files;
2621 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2624 * Fetch the descriptor locklessly. We avoid fdrop() races by
2625 * never raising a refcount above 0. To accomplish this we have
2626 * to use a cmpset loop rather than an atomic_add. The descriptor
2627 * must be re-verified once we acquire a reference to be certain
2628 * that the identity is still correct and we did not lose a race
2629 * due to preemption.
2633 seq = seqc_read(fd_seqc(fdt, fd));
2634 fde = &fdt->fdt_ofiles[fd];
2635 haverights = *cap_rights_fde_inline(fde);
2637 if (!seqc_consistent(fd_seqc(fdt, fd), seq))
2640 fp = fdt->fdt_ofiles[fd].fde_file;
2645 error = cap_check(&haverights, needrightsp);
2649 count = fp->f_count;
2653 * Force a reload. Other thread could reallocate the
2654 * table before this fd was closed, so it possible that
2655 * there is a stale fp pointer in cached version.
2657 fdt = *(const struct fdescenttbl * const volatile *)&(fdp->fd_files);
2661 * Use an acquire barrier to force re-reading of fdt so it is
2662 * refreshed for verification.
2664 if (atomic_fcmpset_acq_int(&fp->f_count, &count, count + 1) == 0)
2666 fdt = fdp->fd_files;
2668 if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
2670 if (fp == fdt->fdt_ofiles[fd].fde_file)
2673 fdrop(fp, curthread);
2685 * Extract the file pointer associated with the specified descriptor for the
2686 * current user process.
2688 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2691 * File's rights will be checked against the capability rights mask.
2693 * If an error occurred the non-zero error is returned and *fpp is set to
2694 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
2695 * responsible for fdrop().
2698 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2699 cap_rights_t *needrightsp, seqc_t *seqp)
2701 struct filedesc *fdp;
2706 fdp = td->td_proc->p_fd;
2707 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2710 if (fp->f_ops == &badfileops) {
2716 * FREAD and FWRITE failure return EBADF as per POSIX.
2722 if ((fp->f_flag & flags) == 0)
2726 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2727 ((fp->f_flag & FWRITE) != 0))
2733 KASSERT(0, ("wrong flags"));
2746 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2749 return (_fget(td, fd, fpp, 0, rightsp, NULL));
2753 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2757 #ifndef CAPABILITIES
2758 error = _fget(td, fd, fpp, 0, rightsp, NULL);
2759 if (maxprotp != NULL)
2760 *maxprotp = VM_PROT_ALL;
2762 struct filedesc *fdp = td->td_proc->p_fd;
2765 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2767 error = _fget(td, fd, fpp, 0, rightsp, &seq);
2771 * If requested, convert capability rights to access flags.
2773 if (maxprotp != NULL)
2774 *maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd));
2775 if (!fd_modified(fdp, fd, seq))
2784 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2787 return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2791 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2794 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2798 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2801 struct filedesc *fdp = td->td_proc->p_fd;
2802 #ifndef CAPABILITIES
2803 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2808 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2810 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2813 error = cap_fcntl_check(fdp, fd, needfcntl);
2814 if (!fd_modified(fdp, fd, seq))
2827 * Like fget() but loads the underlying vnode, or returns an error if the
2828 * descriptor does not represent a vnode. Note that pipes use vnodes but
2829 * never have VM objects. The returned vnode will be vref()'d.
2831 * XXX: what about the unused flags ?
2834 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2841 error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2844 if (fp->f_vnode == NULL) {
2856 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2859 return (_fgetvp(td, fd, 0, rightsp, vpp));
2863 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2864 struct filecaps *havecaps, struct vnode **vpp)
2866 struct filedesc *fdp;
2867 struct filecaps caps;
2871 fdp = td->td_proc->p_fd;
2872 error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2875 if (fp->f_ops == &badfileops) {
2879 if (fp->f_vnode == NULL) {
2890 filecaps_free(&caps);
2895 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2898 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2902 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2905 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2910 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2914 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2919 * Handle the last reference to a file being closed.
2921 * Without the noinline attribute clang keeps inlining the func thorough this
2922 * file when fdrop is used.
2925 _fdrop(struct file *fp, struct thread *td)
2929 if (fp->f_count != 0)
2930 panic("fdrop: count %d", fp->f_count);
2931 error = fo_close(fp, td);
2932 atomic_subtract_int(&openfiles, 1);
2934 free(fp->f_advice, M_FADVISE);
2935 uma_zfree(file_zone, fp);
2941 * Apply an advisory lock on a file descriptor.
2943 * Just attempt to get a record lock of the requested type on the entire file
2944 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2946 #ifndef _SYS_SYSPROTO_H_
2954 sys_flock(struct thread *td, struct flock_args *uap)
2961 error = fget(td, uap->fd, &cap_flock_rights, &fp);
2964 if (fp->f_type != DTYPE_VNODE) {
2966 return (EOPNOTSUPP);
2970 lf.l_whence = SEEK_SET;
2973 if (uap->how & LOCK_UN) {
2974 lf.l_type = F_UNLCK;
2975 atomic_clear_int(&fp->f_flag, FHASLOCK);
2976 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2979 if (uap->how & LOCK_EX)
2980 lf.l_type = F_WRLCK;
2981 else if (uap->how & LOCK_SH)
2982 lf.l_type = F_RDLCK;
2987 atomic_set_int(&fp->f_flag, FHASLOCK);
2988 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2989 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2995 * Duplicate the specified descriptor to a free descriptor.
2998 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
2999 int openerror, int *indxp)
3001 struct filedescent *newfde, *oldfde;
3006 KASSERT(openerror == ENODEV || openerror == ENXIO,
3007 ("unexpected error %d in %s", openerror, __func__));
3010 * If the to-be-dup'd fd number is greater than the allowed number
3011 * of file descriptors, or the fd to be dup'd has already been
3012 * closed, then reject.
3014 FILEDESC_XLOCK(fdp);
3015 if ((fp = fget_locked(fdp, dfd)) == NULL) {
3016 FILEDESC_XUNLOCK(fdp);
3020 error = fdalloc(td, 0, &indx);
3022 FILEDESC_XUNLOCK(fdp);
3027 * There are two cases of interest here.
3029 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3031 * For ENXIO steal away the file structure from (dfd) and store it in
3032 * (indx). (dfd) is effectively closed by this operation.
3034 switch (openerror) {
3037 * Check that the mode the file is being opened for is a
3038 * subset of the mode of the existing descriptor.
3040 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3041 fdunused(fdp, indx);
3042 FILEDESC_XUNLOCK(fdp);
3046 newfde = &fdp->fd_ofiles[indx];
3047 oldfde = &fdp->fd_ofiles[dfd];
3048 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3050 seqc_write_begin(&newfde->fde_seqc);
3052 memcpy(newfde, oldfde, fde_change_size);
3053 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3056 seqc_write_end(&newfde->fde_seqc);
3061 * Steal away the file pointer from dfd and stuff it into indx.
3063 newfde = &fdp->fd_ofiles[indx];
3064 oldfde = &fdp->fd_ofiles[dfd];
3066 seqc_write_begin(&newfde->fde_seqc);
3068 memcpy(newfde, oldfde, fde_change_size);
3069 oldfde->fde_file = NULL;
3072 seqc_write_end(&newfde->fde_seqc);
3076 FILEDESC_XUNLOCK(fdp);
3082 * This sysctl determines if we will allow a process to chroot(2) if it
3083 * has a directory open:
3084 * 0: disallowed for all processes.
3085 * 1: allowed for processes that were not already chroot(2)'ed.
3086 * 2: allowed for all processes.
3089 static int chroot_allow_open_directories = 1;
3091 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3092 &chroot_allow_open_directories, 0,
3093 "Allow a process to chroot(2) if it has a directory open");
3096 * Helper function for raised chroot(2) security function: Refuse if
3097 * any filedescriptors are open directories.
3100 chroot_refuse_vdir_fds(struct filedesc *fdp)
3106 FILEDESC_LOCK_ASSERT(fdp);
3108 for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3109 fp = fget_locked(fdp, fd);
3112 if (fp->f_type == DTYPE_VNODE) {
3114 if (vp->v_type == VDIR)
3122 * Common routine for kern_chroot() and jail_attach(). The caller is
3123 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3124 * authorize this operation.
3127 pwd_chroot(struct thread *td, struct vnode *vp)
3129 struct filedesc *fdp;
3130 struct vnode *oldvp;
3133 fdp = td->td_proc->p_fd;
3134 FILEDESC_XLOCK(fdp);
3135 if (chroot_allow_open_directories == 0 ||
3136 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3137 error = chroot_refuse_vdir_fds(fdp);
3139 FILEDESC_XUNLOCK(fdp);
3143 oldvp = fdp->fd_rdir;
3146 if (fdp->fd_jdir == NULL) {
3150 FILEDESC_XUNLOCK(fdp);
3156 pwd_chdir(struct thread *td, struct vnode *vp)
3158 struct filedesc *fdp;
3159 struct vnode *oldvp;
3161 fdp = td->td_proc->p_fd;
3162 FILEDESC_XLOCK(fdp);
3163 VNASSERT(vp->v_usecount > 0, vp,
3164 ("chdir to a vnode with zero usecount"));
3165 oldvp = fdp->fd_cdir;
3167 FILEDESC_XUNLOCK(fdp);
3172 * Scan all active processes and prisons to see if any of them have a current
3173 * or root directory of `olddp'. If so, replace them with the new mount point.
3176 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3178 struct filedesc *fdp;
3183 if (vrefcnt(olddp) == 1)
3186 sx_slock(&allproc_lock);
3187 FOREACH_PROC_IN_SYSTEM(p) {
3193 FILEDESC_XLOCK(fdp);
3194 if (fdp->fd_cdir == olddp) {
3196 fdp->fd_cdir = newdp;
3199 if (fdp->fd_rdir == olddp) {
3201 fdp->fd_rdir = newdp;
3204 if (fdp->fd_jdir == olddp) {
3206 fdp->fd_jdir = newdp;
3209 FILEDESC_XUNLOCK(fdp);
3212 sx_sunlock(&allproc_lock);
3213 if (rootvnode == olddp) {
3218 mtx_lock(&prison0.pr_mtx);
3219 if (prison0.pr_root == olddp) {
3221 prison0.pr_root = newdp;
3224 mtx_unlock(&prison0.pr_mtx);
3225 sx_slock(&allprison_lock);
3226 TAILQ_FOREACH(pr, &allprison, pr_list) {
3227 mtx_lock(&pr->pr_mtx);
3228 if (pr->pr_root == olddp) {
3230 pr->pr_root = newdp;
3233 mtx_unlock(&pr->pr_mtx);
3235 sx_sunlock(&allprison_lock);
3240 struct filedesc_to_leader *
3241 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3243 struct filedesc_to_leader *fdtol;
3245 fdtol = malloc(sizeof(struct filedesc_to_leader),
3246 M_FILEDESC_TO_LEADER, M_WAITOK);
3247 fdtol->fdl_refcount = 1;
3248 fdtol->fdl_holdcount = 0;
3249 fdtol->fdl_wakeup = 0;
3250 fdtol->fdl_leader = leader;
3252 FILEDESC_XLOCK(fdp);
3253 fdtol->fdl_next = old->fdl_next;
3254 fdtol->fdl_prev = old;
3255 old->fdl_next = fdtol;
3256 fdtol->fdl_next->fdl_prev = fdtol;
3257 FILEDESC_XUNLOCK(fdp);
3259 fdtol->fdl_next = fdtol;
3260 fdtol->fdl_prev = fdtol;
3266 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3268 struct filedesc *fdp;
3269 int i, count, slots;
3271 if (*(int *)arg1 != 0)
3274 fdp = curproc->p_fd;
3276 FILEDESC_SLOCK(fdp);
3277 slots = NDSLOTS(fdp->fd_lastfile + 1);
3278 for (i = 0; i < slots; i++)
3279 count += bitcountl(fdp->fd_map[i]);
3280 FILEDESC_SUNLOCK(fdp);
3282 return (SYSCTL_OUT(req, &count, sizeof(count)));
3285 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3286 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3287 "Number of open file descriptors");
3290 * Get file structures globally.
3293 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3296 struct filedesc *fdp;
3301 error = sysctl_wire_old_buffer(req, 0);
3304 if (req->oldptr == NULL) {
3306 sx_slock(&allproc_lock);
3307 FOREACH_PROC_IN_SYSTEM(p) {
3309 if (p->p_state == PRS_NEW) {
3317 /* overestimates sparse tables. */
3318 if (fdp->fd_lastfile > 0)
3319 n += fdp->fd_lastfile;
3322 sx_sunlock(&allproc_lock);
3323 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3326 bzero(&xf, sizeof(xf));
3327 xf.xf_size = sizeof(xf);
3328 sx_slock(&allproc_lock);
3329 FOREACH_PROC_IN_SYSTEM(p) {
3331 if (p->p_state == PRS_NEW) {
3335 if (p_cansee(req->td, p) != 0) {
3339 xf.xf_pid = p->p_pid;
3340 xf.xf_uid = p->p_ucred->cr_uid;
3345 FILEDESC_SLOCK(fdp);
3346 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3347 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3350 xf.xf_file = (uintptr_t)fp;
3351 xf.xf_data = (uintptr_t)fp->f_data;
3352 xf.xf_vnode = (uintptr_t)fp->f_vnode;
3353 xf.xf_type = (uintptr_t)fp->f_type;
3354 xf.xf_count = fp->f_count;
3356 xf.xf_offset = foffset_get(fp);
3357 xf.xf_flag = fp->f_flag;
3358 error = SYSCTL_OUT(req, &xf, sizeof(xf));
3362 FILEDESC_SUNLOCK(fdp);
3367 sx_sunlock(&allproc_lock);
3371 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3372 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3374 #ifdef KINFO_FILE_SIZE
3375 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3379 xlate_fflags(int fflags)
3381 static const struct {
3384 } fflags_table[] = {
3385 { FAPPEND, KF_FLAG_APPEND },
3386 { FASYNC, KF_FLAG_ASYNC },
3387 { FFSYNC, KF_FLAG_FSYNC },
3388 { FHASLOCK, KF_FLAG_HASLOCK },
3389 { FNONBLOCK, KF_FLAG_NONBLOCK },
3390 { FREAD, KF_FLAG_READ },
3391 { FWRITE, KF_FLAG_WRITE },
3392 { O_CREAT, KF_FLAG_CREAT },
3393 { O_DIRECT, KF_FLAG_DIRECT },
3394 { O_EXCL, KF_FLAG_EXCL },
3395 { O_EXEC, KF_FLAG_EXEC },
3396 { O_EXLOCK, KF_FLAG_EXLOCK },
3397 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3398 { O_SHLOCK, KF_FLAG_SHLOCK },
3399 { O_TRUNC, KF_FLAG_TRUNC }
3405 for (i = 0; i < nitems(fflags_table); i++)
3406 if (fflags & fflags_table[i].fflag)
3407 kflags |= fflags_table[i].kf_fflag;
3411 /* Trim unused data from kf_path by truncating the structure size. */
3413 pack_kinfo(struct kinfo_file *kif)
3416 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3417 strlen(kif->kf_path) + 1;
3418 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3422 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3423 struct kinfo_file *kif, struct filedesc *fdp, int flags)
3427 bzero(kif, sizeof(*kif));
3429 /* Set a default type to allow for empty fill_kinfo() methods. */
3430 kif->kf_type = KF_TYPE_UNKNOWN;
3431 kif->kf_flags = xlate_fflags(fp->f_flag);
3432 if (rightsp != NULL)
3433 kif->kf_cap_rights = *rightsp;
3435 cap_rights_init(&kif->kf_cap_rights);
3437 kif->kf_ref_count = fp->f_count;
3438 kif->kf_offset = foffset_get(fp);
3441 * This may drop the filedesc lock, so the 'fp' cannot be
3442 * accessed after this call.
3444 error = fo_fill_kinfo(fp, kif, fdp);
3446 kif->kf_status |= KF_ATTR_VALID;
3447 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3450 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3454 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3455 struct kinfo_file *kif, int flags)
3459 bzero(kif, sizeof(*kif));
3461 kif->kf_type = KF_TYPE_VNODE;
3462 error = vn_fill_kinfo_vnode(vp, kif);
3464 kif->kf_status |= KF_ATTR_VALID;
3465 kif->kf_flags = xlate_fflags(fflags);
3466 cap_rights_init(&kif->kf_cap_rights);
3468 kif->kf_ref_count = -1;
3469 kif->kf_offset = -1;
3470 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3473 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3477 struct export_fd_buf {
3478 struct filedesc *fdp;
3481 struct kinfo_file kif;
3486 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3488 struct kinfo_file *kif;
3491 if (efbuf->remainder != -1) {
3492 if (efbuf->remainder < kif->kf_structsize) {
3493 /* Terminate export. */
3494 efbuf->remainder = 0;
3497 efbuf->remainder -= kif->kf_structsize;
3499 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3503 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3504 struct export_fd_buf *efbuf)
3508 if (efbuf->remainder == 0)
3510 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3512 FILEDESC_SUNLOCK(efbuf->fdp);
3513 error = export_kinfo_to_sb(efbuf);
3514 FILEDESC_SLOCK(efbuf->fdp);
3519 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3520 struct export_fd_buf *efbuf)
3524 if (efbuf->remainder == 0)
3526 if (efbuf->fdp != NULL)
3527 FILEDESC_SUNLOCK(efbuf->fdp);
3528 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3529 error = export_kinfo_to_sb(efbuf);
3530 if (efbuf->fdp != NULL)
3531 FILEDESC_SLOCK(efbuf->fdp);
3536 * Store a process file descriptor information to sbuf.
3538 * Takes a locked proc as argument, and returns with the proc unlocked.
3541 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
3545 struct filedesc *fdp;
3546 struct export_fd_buf *efbuf;
3547 struct vnode *cttyvp, *textvp, *tracevp;
3549 cap_rights_t rights;
3551 PROC_LOCK_ASSERT(p, MA_OWNED);
3554 tracevp = p->p_tracevp;
3555 if (tracevp != NULL)
3558 textvp = p->p_textvp;
3561 /* Controlling tty. */
3563 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3564 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3570 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3573 efbuf->remainder = maxlen;
3574 efbuf->flags = flags;
3575 if (tracevp != NULL)
3576 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3579 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3581 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3587 FILEDESC_SLOCK(fdp);
3588 /* working directory */
3589 if (fdp->fd_cdir != NULL) {
3590 vrefact(fdp->fd_cdir);
3591 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3593 /* root directory */
3594 if (fdp->fd_rdir != NULL) {
3595 vrefact(fdp->fd_rdir);
3596 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3598 /* jail directory */
3599 if (fdp->fd_jdir != NULL) {
3600 vrefact(fdp->fd_jdir);
3601 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3603 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3604 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3607 rights = *cap_rights(fdp, i);
3608 #else /* !CAPABILITIES */
3609 rights = cap_no_rights;
3612 * Create sysctl entry. It is OK to drop the filedesc
3613 * lock inside of export_file_to_sb() as we will
3614 * re-validate and re-evaluate its properties when the
3617 error = export_file_to_sb(fp, i, &rights, efbuf);
3618 if (error != 0 || efbuf->remainder == 0)
3621 FILEDESC_SUNLOCK(fdp);
3624 free(efbuf, M_TEMP);
3628 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
3631 * Get per-process file descriptors for use by procstat(1), et al.
3634 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3639 int error, error2, *name;
3643 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3644 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3645 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3650 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3651 error = kern_proc_filedesc_out(p, &sb, maxlen,
3652 KERN_FILEDESC_PACK_KINFO);
3653 error2 = sbuf_finish(&sb);
3655 return (error != 0 ? error : error2);
3658 #ifdef COMPAT_FREEBSD7
3659 #ifdef KINFO_OFILE_SIZE
3660 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3664 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3667 okif->kf_structsize = sizeof(*okif);
3668 okif->kf_type = kif->kf_type;
3669 okif->kf_fd = kif->kf_fd;
3670 okif->kf_ref_count = kif->kf_ref_count;
3671 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3672 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3673 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3674 okif->kf_offset = kif->kf_offset;
3675 if (kif->kf_type == KF_TYPE_VNODE)
3676 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
3678 okif->kf_vnode_type = KF_VTYPE_VNON;
3679 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3680 if (kif->kf_type == KF_TYPE_SOCKET) {
3681 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
3682 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
3683 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
3684 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
3685 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
3687 okif->kf_sa_local.ss_family = AF_UNSPEC;
3688 okif->kf_sa_peer.ss_family = AF_UNSPEC;
3693 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3694 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3699 FILEDESC_SUNLOCK(fdp);
3700 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3701 kinfo_to_okinfo(kif, okif);
3702 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3703 FILEDESC_SLOCK(fdp);
3708 * Get per-process file descriptors for use by procstat(1), et al.
3711 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3713 struct kinfo_ofile *okif;
3714 struct kinfo_file *kif;
3715 struct filedesc *fdp;
3716 int error, i, *name;
3721 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3728 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3729 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3730 FILEDESC_SLOCK(fdp);
3731 if (fdp->fd_cdir != NULL)
3732 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3734 if (fdp->fd_rdir != NULL)
3735 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3737 if (fdp->fd_jdir != NULL)
3738 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3740 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3741 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3743 export_file_to_kinfo(fp, i, NULL, kif, fdp,
3744 KERN_FILEDESC_PACK_KINFO);
3745 FILEDESC_SUNLOCK(fdp);
3746 kinfo_to_okinfo(kif, okif);
3747 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3748 FILEDESC_SLOCK(fdp);
3752 FILEDESC_SUNLOCK(fdp);
3759 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3760 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3761 "Process ofiledesc entries");
3762 #endif /* COMPAT_FREEBSD7 */
3765 vntype_to_kinfo(int vtype)
3770 } vtypes_table[] = {
3771 { VBAD, KF_VTYPE_VBAD },
3772 { VBLK, KF_VTYPE_VBLK },
3773 { VCHR, KF_VTYPE_VCHR },
3774 { VDIR, KF_VTYPE_VDIR },
3775 { VFIFO, KF_VTYPE_VFIFO },
3776 { VLNK, KF_VTYPE_VLNK },
3777 { VNON, KF_VTYPE_VNON },
3778 { VREG, KF_VTYPE_VREG },
3779 { VSOCK, KF_VTYPE_VSOCK }
3784 * Perform vtype translation.
3786 for (i = 0; i < nitems(vtypes_table); i++)
3787 if (vtypes_table[i].vtype == vtype)
3788 return (vtypes_table[i].kf_vtype);
3790 return (KF_VTYPE_UNKNOWN);
3793 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3794 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3795 "Process filedesc entries");
3798 * Store a process current working directory information to sbuf.
3800 * Takes a locked proc as argument, and returns with the proc unlocked.
3803 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
3805 struct filedesc *fdp;
3806 struct export_fd_buf *efbuf;
3809 PROC_LOCK_ASSERT(p, MA_OWNED);
3816 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3819 efbuf->remainder = maxlen;
3821 FILEDESC_SLOCK(fdp);
3822 if (fdp->fd_cdir == NULL)
3825 vrefact(fdp->fd_cdir);
3826 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3829 FILEDESC_SUNLOCK(fdp);
3831 free(efbuf, M_TEMP);
3836 * Get per-process current working directory.
3839 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3844 int error, error2, *name;
3848 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3849 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3850 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3855 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3856 error = kern_proc_cwd_out(p, &sb, maxlen);
3857 error2 = sbuf_finish(&sb);
3859 return (error != 0 ? error : error2);
3862 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3863 sysctl_kern_proc_cwd, "Process current working directory");
3867 * For the purposes of debugging, generate a human-readable string for the
3871 file_type_to_name(short type)
3899 case DTYPE_PROCDESC:
3901 case DTYPE_LINUXEFD:
3903 case DTYPE_LINUXTFD:
3911 * For the purposes of debugging, identify a process (if any, perhaps one of
3912 * many) that references the passed file in its file descriptor array. Return
3915 static struct proc *
3916 file_to_first_proc(struct file *fp)
3918 struct filedesc *fdp;
3922 FOREACH_PROC_IN_SYSTEM(p) {
3923 if (p->p_state == PRS_NEW)
3928 for (n = 0; n <= fdp->fd_lastfile; n++) {
3929 if (fp == fdp->fd_ofiles[n].fde_file)
3937 db_print_file(struct file *fp, int header)
3939 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
3943 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
3944 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
3945 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
3947 p = file_to_first_proc(fp);
3948 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
3949 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
3950 fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode,
3951 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3956 DB_SHOW_COMMAND(file, db_show_file)
3961 db_printf("usage: show file <addr>\n");
3964 fp = (struct file *)addr;
3965 db_print_file(fp, 1);
3968 DB_SHOW_COMMAND(files, db_show_files)
3970 struct filedesc *fdp;
3977 FOREACH_PROC_IN_SYSTEM(p) {
3978 if (p->p_state == PRS_NEW)
3980 if ((fdp = p->p_fd) == NULL)
3982 for (n = 0; n <= fdp->fd_lastfile; ++n) {
3983 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3985 db_print_file(fp, header);
3992 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3993 &maxfilesperproc, 0, "Maximum files allowed open per process");
3995 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3996 &maxfiles, 0, "Maximum number of files");
3998 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3999 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
4003 filelistinit(void *dummy)
4006 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4007 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4008 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4009 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4010 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4012 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4014 /*-------------------------------------------------------------------*/
4017 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4018 int flags, struct thread *td)
4025 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4033 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4041 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4049 badfo_kqfilter(struct file *fp, struct knote *kn)
4056 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4064 badfo_close(struct file *fp, struct thread *td)
4071 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4079 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4087 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4088 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4096 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4102 struct fileops badfileops = {
4103 .fo_read = badfo_readwrite,
4104 .fo_write = badfo_readwrite,
4105 .fo_truncate = badfo_truncate,
4106 .fo_ioctl = badfo_ioctl,
4107 .fo_poll = badfo_poll,
4108 .fo_kqfilter = badfo_kqfilter,
4109 .fo_stat = badfo_stat,
4110 .fo_close = badfo_close,
4111 .fo_chmod = badfo_chmod,
4112 .fo_chown = badfo_chown,
4113 .fo_sendfile = badfo_sendfile,
4114 .fo_fill_kinfo = badfo_fill_kinfo,
4118 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4119 int flags, struct thread *td)
4122 return (EOPNOTSUPP);
4126 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4134 invfo_ioctl(struct file *fp, u_long com, void *data,
4135 struct ucred *active_cred, struct thread *td)
4142 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4146 return (poll_no_poll(events));
4150 invfo_kqfilter(struct file *fp, struct knote *kn)
4157 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4165 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4173 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4174 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4181 /*-------------------------------------------------------------------*/
4184 * File Descriptor pseudo-device driver (/dev/fd/).
4186 * Opening minor device N dup()s the file (if any) connected to file
4187 * descriptor N belonging to the calling process. Note that this driver
4188 * consists of only the ``open()'' routine, because all subsequent
4189 * references to this file will be direct to the other driver.
4191 * XXX: we could give this one a cloning event handler if necessary.
4196 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4200 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4201 * the file descriptor being sought for duplication. The error
4202 * return ensures that the vnode for this device will be released
4203 * by vn_open. Open will detect this special error and take the
4204 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4205 * will simply report the error.
4207 td->td_dupfd = dev2unit(dev);
4211 static struct cdevsw fildesc_cdevsw = {
4212 .d_version = D_VERSION,
4218 fildesc_drvinit(void *unused)
4222 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4223 UID_ROOT, GID_WHEEL, 0666, "fd/0");
4224 make_dev_alias(dev, "stdin");
4225 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4226 UID_ROOT, GID_WHEEL, 0666, "fd/1");
4227 make_dev_alias(dev, "stdout");
4228 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4229 UID_ROOT, GID_WHEEL, 0666, "fd/2");
4230 make_dev_alias(dev, "stderr");
4233 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);