2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
40 #include "opt_compat.h"
43 #include <sys/param.h>
44 #include <sys/systm.h>
47 #include <sys/fcntl.h>
49 #include <sys/filedesc.h>
50 #include <sys/filio.h>
52 #include <sys/kernel.h>
53 #include <sys/limits.h>
55 #include <sys/malloc.h>
56 #include <sys/mount.h>
57 #include <sys/mqueue.h>
58 #include <sys/mutex.h>
59 #include <sys/namei.h>
61 #include <sys/resourcevar.h>
62 #include <sys/signalvar.h>
63 #include <sys/socketvar.h>
66 #include <sys/syscallsubr.h>
67 #include <sys/sysctl.h>
68 #include <sys/sysproto.h>
69 #include <sys/unistd.h>
70 #include <sys/vnode.h>
72 #include <security/audit/audit.h>
78 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
79 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
80 "file desc to leader structures");
81 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
83 static uma_zone_t file_zone;
86 /* How to treat 'new' parameter when allocating a fd for do_dup(). */
87 enum dup_type { DUP_VARIABLE, DUP_FIXED };
89 static int do_dup(struct thread *td, enum dup_type type, int old, int new,
91 static int fd_first_free(struct filedesc *, int, int);
92 static int fd_last_used(struct filedesc *, int, int);
93 static void fdgrowtable(struct filedesc *, int);
94 static int fdrop_locked(struct file *fp, struct thread *td);
95 static void fdunused(struct filedesc *fdp, int fd);
96 static void fdused(struct filedesc *fdp, int fd);
99 * A process is initially started out with NDFILE descriptors stored within
100 * this structure, selected to be enough for typical applications based on
101 * the historical limit of 20 open files (and the usage of descriptors by
102 * shells). If these descriptors are exhausted, a larger descriptor table
103 * may be allocated, up to a process' resource limit; the internal arrays
107 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
108 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
109 #define NDSLOT(x) ((x) / NDENTRIES)
110 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
111 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
114 * Storage required per open file descriptor.
116 #define OFILESIZE (sizeof(struct file *) + sizeof(char))
119 * Basic allocation of descriptors:
120 * one of the above, plus arrays for NDFILE descriptors.
123 struct filedesc fd_fd;
125 * These arrays are used when the number of open files is
126 * <= NDFILE, and are then pointed to by the pointers above.
128 struct file *fd_dfiles[NDFILE];
129 char fd_dfileflags[NDFILE];
130 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
134 * Descriptor management.
136 struct filelist filehead; /* head of list of open files */
137 int openfiles; /* actual number of open files */
138 struct sx filelist_lock; /* sx to protect filelist */
139 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
140 void (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
142 /* A mutex to protect the association between a proc and filedesc. */
143 static struct mtx fdesc_mtx;
146 * Find the first zero bit in the given bitmap, starting at low and not
147 * exceeding size - 1.
150 fd_first_free(struct filedesc *fdp, int low, int size)
152 NDSLOTTYPE *map = fdp->fd_map;
160 if (low % NDENTRIES) {
161 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
162 if ((mask &= ~map[off]) != 0UL)
163 return (off * NDENTRIES + ffsl(mask) - 1);
166 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
167 if (map[off] != ~0UL)
168 return (off * NDENTRIES + ffsl(~map[off]) - 1);
173 * Find the highest non-zero bit in the given bitmap, starting at low and
174 * not exceeding size - 1.
177 fd_last_used(struct filedesc *fdp, int low, int size)
179 NDSLOTTYPE *map = fdp->fd_map;
187 if (size % NDENTRIES) {
188 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
189 if ((mask &= map[off]) != 0)
190 return (off * NDENTRIES + flsl(mask) - 1);
193 for (minoff = NDSLOT(low); off >= minoff; --off)
195 return (off * NDENTRIES + flsl(map[off]) - 1);
200 fdisused(struct filedesc *fdp, int fd)
202 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
203 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
204 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
208 * Mark a file descriptor as used.
211 fdused(struct filedesc *fdp, int fd)
213 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
214 KASSERT(!fdisused(fdp, fd),
215 ("fd already used"));
216 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
217 if (fd > fdp->fd_lastfile)
218 fdp->fd_lastfile = fd;
219 if (fd == fdp->fd_freefile)
220 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
224 * Mark a file descriptor as unused.
227 fdunused(struct filedesc *fdp, int fd)
229 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
230 KASSERT(fdisused(fdp, fd),
231 ("fd is already unused"));
232 KASSERT(fdp->fd_ofiles[fd] == NULL,
233 ("fd is still in use"));
234 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
235 if (fd < fdp->fd_freefile)
236 fdp->fd_freefile = fd;
237 if (fd == fdp->fd_lastfile)
238 fdp->fd_lastfile = fd_last_used(fdp, 0, fd);
242 * System calls on descriptors.
244 #ifndef _SYS_SYSPROTO_H_
245 struct getdtablesize_args {
254 getdtablesize(struct thread *td, struct getdtablesize_args *uap)
256 struct proc *p = td->td_proc;
260 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
266 * Duplicate a file descriptor to a particular value.
268 * note: keep in mind that a potential race condition exists when closing
269 * descriptors from a shared descriptor table (via rfork).
271 #ifndef _SYS_SYSPROTO_H_
282 dup2(struct thread *td, struct dup2_args *uap)
285 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
290 * Duplicate a file descriptor.
292 #ifndef _SYS_SYSPROTO_H_
302 dup(struct thread *td, struct dup_args *uap)
305 return (do_dup(td, DUP_VARIABLE, (int)uap->fd, 0, td->td_retval));
309 * The file control system call.
311 #ifndef _SYS_SYSPROTO_H_
323 fcntl(struct thread *td, struct fcntl_args *uap)
334 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl));
343 error = kern_fcntl(td, uap->fd, uap->cmd, arg);
346 if (uap->cmd == F_GETLK)
347 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl));
352 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
354 struct filedesc *fdp;
365 * XXXRW: Some fcntl() calls require Giant -- others don't. Try to
366 * avoid grabbing Giant for calls we know don't need it.
386 if ((unsigned)fd >= fdp->fd_nfiles ||
387 (fp = fdp->fd_ofiles[fd]) == NULL) {
388 FILEDESC_UNLOCK(fdp);
392 pop = &fdp->fd_ofileflags[fd];
396 /* mtx_assert(&Giant, MA_NOTOWNED); */
397 FILEDESC_UNLOCK(fdp);
400 if (newmin >= lim_cur(p, RLIMIT_NOFILE) ||
401 newmin >= maxfilesperproc) {
407 error = do_dup(td, DUP_VARIABLE, fd, newmin, td->td_retval);
411 /* mtx_assert(&Giant, MA_NOTOWNED); */
412 td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0;
413 FILEDESC_UNLOCK(fdp);
417 /* mtx_assert(&Giant, MA_NOTOWNED); */
418 *pop = (*pop &~ UF_EXCLOSE) |
419 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
420 FILEDESC_UNLOCK(fdp);
424 /* mtx_assert(&Giant, MA_NOTOWNED); */
426 td->td_retval[0] = OFLAGS(fp->f_flag);
428 FILEDESC_UNLOCK(fdp);
432 mtx_assert(&Giant, MA_OWNED);
435 fp->f_flag &= ~FCNTLFLAGS;
436 fp->f_flag |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
438 FILEDESC_UNLOCK(fdp);
439 tmp = fp->f_flag & FNONBLOCK;
440 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
445 tmp = fp->f_flag & FASYNC;
446 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
452 fp->f_flag &= ~FNONBLOCK;
455 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
460 mtx_assert(&Giant, MA_OWNED);
462 FILEDESC_UNLOCK(fdp);
463 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
465 td->td_retval[0] = tmp;
470 mtx_assert(&Giant, MA_OWNED);
472 FILEDESC_UNLOCK(fdp);
474 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
479 mtx_assert(&Giant, MA_OWNED);
481 /* FALLTHROUGH F_SETLK */
484 mtx_assert(&Giant, MA_OWNED);
485 if (fp->f_type != DTYPE_VNODE) {
486 FILEDESC_UNLOCK(fdp);
491 flp = (struct flock *)arg;
492 if (flp->l_whence == SEEK_CUR) {
493 if (fp->f_offset < 0 ||
495 fp->f_offset > OFF_MAX - flp->l_start)) {
496 FILEDESC_UNLOCK(fdp);
500 flp->l_start += fp->f_offset;
504 * VOP_ADVLOCK() may block.
507 FILEDESC_UNLOCK(fdp);
510 switch (flp->l_type) {
512 if ((fp->f_flag & FREAD) == 0) {
516 PROC_LOCK(p->p_leader);
517 p->p_leader->p_flag |= P_ADVLOCK;
518 PROC_UNLOCK(p->p_leader);
519 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
523 if ((fp->f_flag & FWRITE) == 0) {
527 PROC_LOCK(p->p_leader);
528 p->p_leader->p_flag |= P_ADVLOCK;
529 PROC_UNLOCK(p->p_leader);
530 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
534 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
541 /* Check for race with close */
542 FILEDESC_LOCK_FAST(fdp);
543 if ((unsigned) fd >= fdp->fd_nfiles ||
544 fp != fdp->fd_ofiles[fd]) {
545 FILEDESC_UNLOCK_FAST(fdp);
546 flp->l_whence = SEEK_SET;
549 flp->l_type = F_UNLCK;
550 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
551 F_UNLCK, flp, F_POSIX);
553 FILEDESC_UNLOCK_FAST(fdp);
558 mtx_assert(&Giant, MA_OWNED);
559 if (fp->f_type != DTYPE_VNODE) {
560 FILEDESC_UNLOCK(fdp);
564 flp = (struct flock *)arg;
565 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
566 flp->l_type != F_UNLCK) {
567 FILEDESC_UNLOCK(fdp);
571 if (flp->l_whence == SEEK_CUR) {
572 if ((flp->l_start > 0 &&
573 fp->f_offset > OFF_MAX - flp->l_start) ||
575 fp->f_offset < OFF_MIN - flp->l_start)) {
576 FILEDESC_UNLOCK(fdp);
580 flp->l_start += fp->f_offset;
583 * VOP_ADVLOCK() may block.
586 FILEDESC_UNLOCK(fdp);
588 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
593 FILEDESC_UNLOCK(fdp);
604 * Common code for dup, dup2, and fcntl(F_DUPFD).
607 do_dup(struct thread *td, enum dup_type type, int old, int new, register_t *retval)
609 struct filedesc *fdp;
613 int error, holdleaders, maxfd;
615 KASSERT((type == DUP_VARIABLE || type == DUP_FIXED),
616 ("invalid dup type %d", type));
622 * Verify we have a valid descriptor to dup from and possibly to
625 if (old < 0 || new < 0)
628 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
634 if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) {
635 FILEDESC_UNLOCK(fdp);
638 if (type == DUP_FIXED && old == new) {
640 FILEDESC_UNLOCK(fdp);
643 fp = fdp->fd_ofiles[old];
647 * If the caller specified a file descriptor, make sure the file
648 * table is large enough to hold it, and grab it. Otherwise, just
649 * allocate a new descriptor the usual way. Since the filedesc
650 * lock may be temporarily dropped in the process, we have to look
653 if (type == DUP_FIXED) {
654 if (new >= fdp->fd_nfiles)
655 fdgrowtable(fdp, new + 1);
656 if (fdp->fd_ofiles[new] == NULL)
659 if ((error = fdalloc(td, new, &new)) != 0) {
660 FILEDESC_UNLOCK(fdp);
667 * If the old file changed out from under us then treat it as a
668 * bad file descriptor. Userland should do its own locking to
671 if (fdp->fd_ofiles[old] != fp) {
672 /* we've allocated a descriptor which we won't use */
673 if (fdp->fd_ofiles[new] == NULL)
675 FILEDESC_UNLOCK(fdp);
680 ("new fd is same as old"));
683 * Save info on the descriptor being overwritten. We cannot close
684 * it without introducing an ownership race for the slot, since we
685 * need to drop the filedesc lock to call closef().
687 * XXX this duplicates parts of close().
689 delfp = fdp->fd_ofiles[new];
692 if (td->td_proc->p_fdtol != NULL) {
694 * Ask fdfree() to sleep to ensure that all relevant
695 * process leaders can be traversed in closef().
697 fdp->fd_holdleaderscount++;
703 * Duplicate the source descriptor
705 fdp->fd_ofiles[new] = fp;
706 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
707 if (new > fdp->fd_lastfile)
708 fdp->fd_lastfile = new;
712 * If we dup'd over a valid file, we now own the reference to it
713 * and must dispose of it using closef() semantics (as if a
714 * close() were performed on it).
716 * XXX this duplicates parts of close().
719 knote_fdclose(td, new);
720 if (delfp->f_type == DTYPE_MQUEUE)
721 mq_fdclose(td, new, delfp);
722 FILEDESC_UNLOCK(fdp);
723 (void) closef(delfp, td);
725 FILEDESC_LOCK_FAST(fdp);
726 fdp->fd_holdleaderscount--;
727 if (fdp->fd_holdleaderscount == 0 &&
728 fdp->fd_holdleaderswakeup != 0) {
729 fdp->fd_holdleaderswakeup = 0;
730 wakeup(&fdp->fd_holdleaderscount);
732 FILEDESC_UNLOCK_FAST(fdp);
735 FILEDESC_UNLOCK(fdp);
741 * If sigio is on the list associated with a process or process group,
742 * disable signalling from the device, remove sigio from the list and
746 funsetown(struct sigio **sigiop)
756 *(sigio->sio_myref) = NULL;
757 if ((sigio)->sio_pgid < 0) {
758 struct pgrp *pg = (sigio)->sio_pgrp;
760 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
764 struct proc *p = (sigio)->sio_proc;
766 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
771 crfree(sigio->sio_ucred);
772 FREE(sigio, M_SIGIO);
776 * Free a list of sigio structures.
777 * We only need to lock the SIGIO_LOCK because we have made ourselves
778 * inaccessible to callers of fsetown and therefore do not need to lock
779 * the proc or pgrp struct for the list manipulation.
782 funsetownlst(struct sigiolst *sigiolst)
788 sigio = SLIST_FIRST(sigiolst);
795 * Every entry of the list should belong
796 * to a single proc or pgrp.
798 if (sigio->sio_pgid < 0) {
799 pg = sigio->sio_pgrp;
800 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
801 } else /* if (sigio->sio_pgid > 0) */ {
803 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
807 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
808 *(sigio->sio_myref) = NULL;
810 KASSERT(sigio->sio_pgid < 0,
811 ("Proc sigio in pgrp sigio list"));
812 KASSERT(sigio->sio_pgrp == pg,
813 ("Bogus pgrp in sigio list"));
815 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
818 } else /* if (p != NULL) */ {
819 KASSERT(sigio->sio_pgid > 0,
820 ("Pgrp sigio in proc sigio list"));
821 KASSERT(sigio->sio_proc == p,
822 ("Bogus proc in sigio list"));
824 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
829 crfree(sigio->sio_ucred);
830 FREE(sigio, M_SIGIO);
837 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
839 * After permission checking, add a sigio structure to the sigio list for
840 * the process or process group.
843 fsetown(pid_t pgid, struct sigio **sigiop)
857 /* Allocate and fill in the new sigio out of locks. */
858 MALLOC(sigio, struct sigio *, sizeof(struct sigio), M_SIGIO, M_WAITOK);
859 sigio->sio_pgid = pgid;
860 sigio->sio_ucred = crhold(curthread->td_ucred);
861 sigio->sio_myref = sigiop;
863 sx_slock(&proctree_lock);
872 * Policy - Don't allow a process to FSETOWN a process
873 * in another session.
875 * Remove this test to allow maximum flexibility or
876 * restrict FSETOWN to the current process or process
877 * group for maximum safety.
880 if (proc->p_session != curthread->td_proc->p_session) {
886 } else /* if (pgid < 0) */ {
887 pgrp = pgfind(-pgid);
895 * Policy - Don't allow a process to FSETOWN a process
896 * in another session.
898 * Remove this test to allow maximum flexibility or
899 * restrict FSETOWN to the current process or process
900 * group for maximum safety.
902 if (pgrp->pg_session != curthread->td_proc->p_session) {
913 * Since funsetownlst() is called without the proctree
914 * locked, we need to check for P_WEXIT.
915 * XXX: is ESRCH correct?
917 if ((proc->p_flag & P_WEXIT) != 0) {
922 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
923 sigio->sio_proc = proc;
927 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
928 sigio->sio_pgrp = pgrp;
931 sx_sunlock(&proctree_lock);
938 sx_sunlock(&proctree_lock);
939 crfree(sigio->sio_ucred);
940 FREE(sigio, M_SIGIO);
945 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
949 struct sigio **sigiop;
954 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
960 * Close a file descriptor.
962 #ifndef _SYS_SYSPROTO_H_
974 struct close_args *uap;
977 return (kern_close(td, uap->fd));
985 struct filedesc *fdp;
992 fdp = td->td_proc->p_fd;
994 AUDIT_SYSCLOSE(td, fd);
997 if ((unsigned)fd >= fdp->fd_nfiles ||
998 (fp = fdp->fd_ofiles[fd]) == NULL) {
999 FILEDESC_UNLOCK(fdp);
1002 fdp->fd_ofiles[fd] = NULL;
1003 fdp->fd_ofileflags[fd] = 0;
1005 if (td->td_proc->p_fdtol != NULL) {
1007 * Ask fdfree() to sleep to ensure that all relevant
1008 * process leaders can be traversed in closef().
1010 fdp->fd_holdleaderscount++;
1015 * We now hold the fp reference that used to be owned by the descriptor
1017 * We have to unlock the FILEDESC *AFTER* knote_fdclose to prevent a
1018 * race of the fd getting opened, a knote added, and deleteing a knote
1021 knote_fdclose(td, fd);
1022 if (fp->f_type == DTYPE_MQUEUE)
1023 mq_fdclose(td, fd, fp);
1024 FILEDESC_UNLOCK(fdp);
1026 error = closef(fp, td);
1028 FILEDESC_LOCK_FAST(fdp);
1029 fdp->fd_holdleaderscount--;
1030 if (fdp->fd_holdleaderscount == 0 &&
1031 fdp->fd_holdleaderswakeup != 0) {
1032 fdp->fd_holdleaderswakeup = 0;
1033 wakeup(&fdp->fd_holdleaderscount);
1035 FILEDESC_UNLOCK_FAST(fdp);
1040 #if defined(COMPAT_43)
1042 * Return status information about a file descriptor.
1044 #ifndef _SYS_SYSPROTO_H_
1045 struct ofstat_args {
1055 ofstat(struct thread *td, struct ofstat_args *uap)
1061 error = kern_fstat(td, uap->fd, &ub);
1064 error = copyout(&oub, uap->sb, sizeof(oub));
1068 #endif /* COMPAT_43 */
1071 * Return status information about a file descriptor.
1073 #ifndef _SYS_SYSPROTO_H_
1084 fstat(struct thread *td, struct fstat_args *uap)
1089 error = kern_fstat(td, uap->fd, &ub);
1091 error = copyout(&ub, uap->sb, sizeof(ub));
1096 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1103 if ((error = fget(td, fd, &fp)) != 0)
1106 AUDIT_ARG(file, td->td_proc, fp);
1108 error = fo_stat(fp, sbp, td->td_ucred, td);
1114 * Return status information about a file descriptor.
1116 #ifndef _SYS_SYSPROTO_H_
1117 struct nfstat_args {
1127 nfstat(struct thread *td, struct nfstat_args *uap)
1133 error = kern_fstat(td, uap->fd, &ub);
1135 cvtnstat(&ub, &nub);
1136 error = copyout(&nub, uap->sb, sizeof(nub));
1142 * Return pathconf information about a file descriptor.
1144 #ifndef _SYS_SYSPROTO_H_
1145 struct fpathconf_args {
1155 fpathconf(struct thread *td, struct fpathconf_args *uap)
1161 if ((error = fget(td, uap->fd, &fp)) != 0)
1164 /* If asynchronous I/O is available, it works for all descriptors. */
1165 if (uap->name == _PC_ASYNC_IO) {
1166 td->td_retval[0] = async_io_version;
1172 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1173 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1174 error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1175 VOP_UNLOCK(vp, 0, td);
1176 VFS_UNLOCK_GIANT(vfslocked);
1177 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1178 if (uap->name != _PC_PIPE_BUF) {
1181 td->td_retval[0] = PIPE_BUF;
1193 * Grow the file table to accomodate (at least) nfd descriptors. This may
1194 * block and drop the filedesc lock, but it will reacquire it before
1198 fdgrowtable(struct filedesc *fdp, int nfd)
1200 struct file **ntable;
1202 int nnfiles, onfiles;
1205 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1207 KASSERT(fdp->fd_nfiles > 0,
1208 ("zero-length file table"));
1210 /* compute the size of the new table */
1211 onfiles = fdp->fd_nfiles;
1212 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1213 if (nnfiles <= onfiles)
1214 /* the table is already large enough */
1217 /* allocate a new table and (if required) new bitmaps */
1218 FILEDESC_UNLOCK(fdp);
1219 MALLOC(ntable, struct file **, nnfiles * OFILESIZE,
1220 M_FILEDESC, M_ZERO | M_WAITOK);
1221 nfileflags = (char *)&ntable[nnfiles];
1222 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles))
1223 MALLOC(nmap, NDSLOTTYPE *, NDSLOTS(nnfiles) * NDSLOTSIZE,
1224 M_FILEDESC, M_ZERO | M_WAITOK);
1230 * We now have new tables ready to go. Since we dropped the
1231 * filedesc lock to call malloc(), watch out for a race.
1233 onfiles = fdp->fd_nfiles;
1234 if (onfiles >= nnfiles) {
1235 /* we lost the race, but that's OK */
1236 free(ntable, M_FILEDESC);
1238 free(nmap, M_FILEDESC);
1241 bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable));
1242 bcopy(fdp->fd_ofileflags, nfileflags, onfiles);
1243 if (onfiles > NDFILE)
1244 free(fdp->fd_ofiles, M_FILEDESC);
1245 fdp->fd_ofiles = ntable;
1246 fdp->fd_ofileflags = nfileflags;
1247 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1248 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap));
1249 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1250 free(fdp->fd_map, M_FILEDESC);
1253 fdp->fd_nfiles = nnfiles;
1257 * Allocate a file descriptor for the process.
1260 fdalloc(struct thread *td, int minfd, int *result)
1262 struct proc *p = td->td_proc;
1263 struct filedesc *fdp = p->p_fd;
1266 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1268 if (fdp->fd_freefile > minfd)
1269 minfd = fdp->fd_freefile;
1272 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1276 * Search the bitmap for a free descriptor. If none is found, try
1277 * to grow the file table. Keep at it until we either get a file
1278 * descriptor or run into process or system limits; fdgrowtable()
1279 * may drop the filedesc lock, so we're in a race.
1282 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1285 if (fd < fdp->fd_nfiles)
1287 fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd));
1291 * Perform some sanity checks, then mark the file descriptor as
1292 * used and return it to the caller.
1294 KASSERT(!fdisused(fdp, fd),
1295 ("fd_first_free() returned non-free descriptor"));
1296 KASSERT(fdp->fd_ofiles[fd] == NULL,
1297 ("free descriptor isn't"));
1298 fdp->fd_ofileflags[fd] = 0; /* XXX needed? */
1305 * Check to see whether n user file descriptors
1306 * are available to the process p.
1309 fdavail(struct thread *td, int n)
1311 struct proc *p = td->td_proc;
1312 struct filedesc *fdp = td->td_proc->p_fd;
1316 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1319 lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1321 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1323 last = min(fdp->fd_nfiles, lim);
1324 fpp = &fdp->fd_ofiles[fdp->fd_freefile];
1325 for (i = last - fdp->fd_freefile; --i >= 0; fpp++) {
1326 if (*fpp == NULL && --n <= 0)
1333 * Create a new open file structure and allocate
1334 * a file decriptor for the process that refers to it.
1335 * We add one reference to the file for the descriptor table
1336 * and one reference for resultfp. This is to prevent us being
1337 * preempted and the entry in the descriptor table closed after
1338 * we release the FILEDESC lock.
1341 falloc(struct thread *td, struct file **resultfp, int *resultfd)
1343 struct proc *p = td->td_proc;
1344 struct file *fp, *fq;
1346 int maxuserfiles = maxfiles - (maxfiles / 20);
1347 static struct timeval lastfail;
1350 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1351 sx_xlock(&filelist_lock);
1353 if ((openfiles >= maxuserfiles &&
1354 suser_cred(td->td_ucred, SUSER_RUID) != 0) ||
1355 openfiles >= maxfiles) {
1356 if (ppsratecheck(&lastfail, &curfail, 1)) {
1357 printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n",
1358 td->td_ucred->cr_ruid);
1360 sx_xunlock(&filelist_lock);
1361 uma_zfree(file_zone, fp);
1367 * If the process has file descriptor zero open, add the new file
1368 * descriptor to the list of open files at that point, otherwise
1369 * put it at the front of the list of open files.
1371 fp->f_mtxp = mtx_pool_alloc(mtxpool_sleep);
1375 fp->f_cred = crhold(td->td_ucred);
1376 fp->f_ops = &badfileops;
1379 FILEDESC_LOCK(p->p_fd);
1380 if ((fq = p->p_fd->fd_ofiles[0])) {
1381 LIST_INSERT_AFTER(fq, fp, f_list);
1383 LIST_INSERT_HEAD(&filehead, fp, f_list);
1385 sx_xunlock(&filelist_lock);
1386 if ((error = fdalloc(td, 0, &i))) {
1387 FILEDESC_UNLOCK(p->p_fd);
1393 p->p_fd->fd_ofiles[i] = fp;
1394 FILEDESC_UNLOCK(p->p_fd);
1403 * Build a new filedesc structure from another.
1404 * Copy the current, root, and jail root vnode references.
1407 fdinit(struct filedesc *fdp)
1409 struct filedesc0 *newfdp;
1411 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1412 mtx_init(&newfdp->fd_fd.fd_mtx, FILEDESC_LOCK_DESC, NULL, MTX_DEF);
1415 newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1416 if (newfdp->fd_fd.fd_cdir)
1417 VREF(newfdp->fd_fd.fd_cdir);
1418 newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1419 if (newfdp->fd_fd.fd_rdir)
1420 VREF(newfdp->fd_fd.fd_rdir);
1421 newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1422 if (newfdp->fd_fd.fd_jdir)
1423 VREF(newfdp->fd_fd.fd_jdir);
1424 FILEDESC_UNLOCK(fdp);
1427 /* Create the file descriptor table. */
1428 newfdp->fd_fd.fd_refcnt = 1;
1429 newfdp->fd_fd.fd_holdcnt = 1;
1430 newfdp->fd_fd.fd_cmask = CMASK;
1431 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1432 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
1433 newfdp->fd_fd.fd_nfiles = NDFILE;
1434 newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1435 newfdp->fd_fd.fd_lastfile = -1;
1436 return (&newfdp->fd_fd);
1439 static struct filedesc *
1440 fdhold(struct proc *p)
1442 struct filedesc *fdp;
1444 mtx_lock(&fdesc_mtx);
1448 mtx_unlock(&fdesc_mtx);
1453 fddrop(struct filedesc *fdp)
1457 mtx_lock(&fdesc_mtx);
1458 i = --fdp->fd_holdcnt;
1459 mtx_unlock(&fdesc_mtx);
1463 mtx_destroy(&fdp->fd_mtx);
1464 FREE(fdp, M_FILEDESC);
1468 * Share a filedesc structure.
1471 fdshare(struct filedesc *fdp)
1473 FILEDESC_LOCK_FAST(fdp);
1475 FILEDESC_UNLOCK_FAST(fdp);
1480 * Unshare a filedesc structure, if necessary by making a copy
1483 fdunshare(struct proc *p, struct thread *td)
1486 FILEDESC_LOCK_FAST(p->p_fd);
1487 if (p->p_fd->fd_refcnt > 1) {
1488 struct filedesc *tmp;
1490 FILEDESC_UNLOCK_FAST(p->p_fd);
1491 tmp = fdcopy(p->p_fd);
1495 FILEDESC_UNLOCK_FAST(p->p_fd);
1499 * Copy a filedesc structure.
1500 * A NULL pointer in returns a NULL reference, this is to ease callers,
1504 fdcopy(struct filedesc *fdp)
1506 struct filedesc *newfdp;
1509 /* Certain daemons might not have file descriptors. */
1513 newfdp = fdinit(fdp);
1514 FILEDESC_LOCK_FAST(fdp);
1515 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1516 FILEDESC_UNLOCK_FAST(fdp);
1517 FILEDESC_LOCK(newfdp);
1518 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1519 FILEDESC_UNLOCK(newfdp);
1520 FILEDESC_LOCK_FAST(fdp);
1522 /* copy everything except kqueue descriptors */
1523 newfdp->fd_freefile = -1;
1524 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1525 if (fdisused(fdp, i) &&
1526 fdp->fd_ofiles[i]->f_type != DTYPE_KQUEUE) {
1527 newfdp->fd_ofiles[i] = fdp->fd_ofiles[i];
1528 newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i];
1529 fhold(newfdp->fd_ofiles[i]);
1530 newfdp->fd_lastfile = i;
1532 if (newfdp->fd_freefile == -1)
1533 newfdp->fd_freefile = i;
1536 FILEDESC_UNLOCK_FAST(fdp);
1537 FILEDESC_LOCK(newfdp);
1538 for (i = 0; i <= newfdp->fd_lastfile; ++i)
1539 if (newfdp->fd_ofiles[i] != NULL)
1541 FILEDESC_UNLOCK(newfdp);
1542 FILEDESC_LOCK_FAST(fdp);
1543 if (newfdp->fd_freefile == -1)
1544 newfdp->fd_freefile = i;
1545 newfdp->fd_cmask = fdp->fd_cmask;
1546 FILEDESC_UNLOCK_FAST(fdp);
1551 * Release a filedesc structure.
1554 fdfree(struct thread *td)
1556 struct filedesc *fdp;
1559 struct filedesc_to_leader *fdtol;
1561 struct vnode *cdir, *jdir, *rdir, *vp;
1564 /* Certain daemons might not have file descriptors. */
1565 fdp = td->td_proc->p_fd;
1569 /* Check for special need to clear POSIX style locks */
1570 fdtol = td->td_proc->p_fdtol;
1571 if (fdtol != NULL) {
1573 KASSERT(fdtol->fdl_refcount > 0,
1574 ("filedesc_to_refcount botch: fdl_refcount=%d",
1575 fdtol->fdl_refcount));
1576 if (fdtol->fdl_refcount == 1 &&
1577 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1578 for (i = 0, fpp = fdp->fd_ofiles;
1579 i <= fdp->fd_lastfile;
1582 (*fpp)->f_type != DTYPE_VNODE)
1586 FILEDESC_UNLOCK(fdp);
1587 lf.l_whence = SEEK_SET;
1590 lf.l_type = F_UNLCK;
1592 locked = VFS_LOCK_GIANT(vp->v_mount);
1593 (void) VOP_ADVLOCK(vp,
1594 (caddr_t)td->td_proc->
1599 VFS_UNLOCK_GIANT(locked);
1602 fpp = fdp->fd_ofiles + i;
1606 if (fdtol->fdl_refcount == 1) {
1607 if (fdp->fd_holdleaderscount > 0 &&
1608 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1610 * close() or do_dup() has cleared a reference
1611 * in a shared file descriptor table.
1613 fdp->fd_holdleaderswakeup = 1;
1614 msleep(&fdp->fd_holdleaderscount, &fdp->fd_mtx,
1615 PLOCK, "fdlhold", 0);
1618 if (fdtol->fdl_holdcount > 0) {
1620 * Ensure that fdtol->fdl_leader
1621 * remains valid in closef().
1623 fdtol->fdl_wakeup = 1;
1624 msleep(fdtol, &fdp->fd_mtx,
1625 PLOCK, "fdlhold", 0);
1629 fdtol->fdl_refcount--;
1630 if (fdtol->fdl_refcount == 0 &&
1631 fdtol->fdl_holdcount == 0) {
1632 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1633 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1636 td->td_proc->p_fdtol = NULL;
1637 FILEDESC_UNLOCK(fdp);
1639 FREE(fdtol, M_FILEDESC_TO_LEADER);
1641 FILEDESC_LOCK_FAST(fdp);
1642 i = --fdp->fd_refcnt;
1643 FILEDESC_UNLOCK_FAST(fdp);
1647 * We are the last reference to the structure, so we can
1648 * safely assume it will not change out from under us.
1650 fpp = fdp->fd_ofiles;
1651 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
1653 (void) closef(*fpp, td);
1657 /* XXX This should happen earlier. */
1658 mtx_lock(&fdesc_mtx);
1659 td->td_proc->p_fd = NULL;
1660 mtx_unlock(&fdesc_mtx);
1662 if (fdp->fd_nfiles > NDFILE)
1663 FREE(fdp->fd_ofiles, M_FILEDESC);
1664 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
1665 FREE(fdp->fd_map, M_FILEDESC);
1669 cdir = fdp->fd_cdir;
1670 fdp->fd_cdir = NULL;
1671 rdir = fdp->fd_rdir;
1672 fdp->fd_rdir = NULL;
1673 jdir = fdp->fd_jdir;
1674 fdp->fd_jdir = NULL;
1675 FILEDESC_UNLOCK(fdp);
1678 locked = VFS_LOCK_GIANT(cdir->v_mount);
1680 VFS_UNLOCK_GIANT(locked);
1683 locked = VFS_LOCK_GIANT(rdir->v_mount);
1685 VFS_UNLOCK_GIANT(locked);
1688 locked = VFS_LOCK_GIANT(jdir->v_mount);
1690 VFS_UNLOCK_GIANT(locked);
1697 * For setugid programs, we don't want to people to use that setugidness
1698 * to generate error messages which write to a file which otherwise would
1699 * otherwise be off-limits to the process. We check for filesystems where
1700 * the vnode can change out from under us after execve (like [lin]procfs).
1702 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
1703 * sufficient. We also don't check for setugidness since we know we are.
1706 is_unsafe(struct file *fp)
1708 if (fp->f_type == DTYPE_VNODE) {
1709 struct vnode *vp = fp->f_vnode;
1711 if ((vp->v_vflag & VV_PROCDEP) != 0)
1718 * Make this setguid thing safe, if at all possible.
1721 setugidsafety(struct thread *td)
1723 struct filedesc *fdp;
1726 /* Certain daemons might not have file descriptors. */
1727 fdp = td->td_proc->p_fd;
1732 * Note: fdp->fd_ofiles may be reallocated out from under us while
1733 * we are blocked in a close. Be careful!
1736 for (i = 0; i <= fdp->fd_lastfile; i++) {
1739 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
1742 knote_fdclose(td, i);
1744 * NULL-out descriptor prior to close to avoid
1745 * a race while close blocks.
1747 fp = fdp->fd_ofiles[i];
1748 fdp->fd_ofiles[i] = NULL;
1749 fdp->fd_ofileflags[i] = 0;
1751 FILEDESC_UNLOCK(fdp);
1752 (void) closef(fp, td);
1756 FILEDESC_UNLOCK(fdp);
1760 * If a specific file object occupies a specific file descriptor,
1761 * close the file descriptor entry and drop a reference on the file
1762 * object. This is a convenience function to handle a subsequent
1763 * error in a function that calls falloc() that handles the race that
1764 * another thread might have closed the file descriptor out from under
1765 * the thread creating the file object.
1768 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
1772 if (fdp->fd_ofiles[idx] == fp) {
1773 fdp->fd_ofiles[idx] = NULL;
1775 FILEDESC_UNLOCK(fdp);
1778 FILEDESC_UNLOCK(fdp);
1783 * Close any files on exec?
1786 fdcloseexec(struct thread *td)
1788 struct filedesc *fdp;
1791 /* Certain daemons might not have file descriptors. */
1792 fdp = td->td_proc->p_fd;
1799 * We cannot cache fd_ofiles or fd_ofileflags since operations
1800 * may block and rip them out from under us.
1802 for (i = 0; i <= fdp->fd_lastfile; i++) {
1803 if (fdp->fd_ofiles[i] != NULL &&
1804 (fdp->fd_ofiles[i]->f_type == DTYPE_MQUEUE ||
1805 (fdp->fd_ofileflags[i] & UF_EXCLOSE))) {
1808 knote_fdclose(td, i);
1810 * NULL-out descriptor prior to close to avoid
1811 * a race while close blocks.
1813 fp = fdp->fd_ofiles[i];
1814 fdp->fd_ofiles[i] = NULL;
1815 fdp->fd_ofileflags[i] = 0;
1817 if (fp->f_type == DTYPE_MQUEUE)
1818 mq_fdclose(td, i, fp);
1819 FILEDESC_UNLOCK(fdp);
1820 (void) closef(fp, td);
1824 FILEDESC_UNLOCK(fdp);
1828 * It is unsafe for set[ug]id processes to be started with file
1829 * descriptors 0..2 closed, as these descriptors are given implicit
1830 * significance in the Standard C library. fdcheckstd() will create a
1831 * descriptor referencing /dev/null for each of stdin, stdout, and
1832 * stderr that is not already open.
1835 fdcheckstd(struct thread *td)
1837 struct nameidata nd;
1838 struct filedesc *fdp;
1841 int fd, i, error, flags, devnull;
1843 fdp = td->td_proc->p_fd;
1846 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
1849 for (i = 0; i < 3; i++) {
1850 if (fdp->fd_ofiles[i] != NULL)
1854 error = falloc(td, &fp, &fd);
1857 /* Note extra ref on `fp' held for us by falloc(). */
1858 KASSERT(fd == i, ("oof, we didn't get our fd"));
1859 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE,
1861 flags = FREAD | FWRITE;
1862 error = vn_open(&nd, &flags, 0, fd);
1865 * Someone may have closed the entry in the
1866 * file descriptor table, so check it hasn't
1867 * changed before dropping the reference count.
1870 KASSERT(fdp->fd_ofiles[fd] == fp,
1871 ("table not shared, how did it change?"));
1872 fdp->fd_ofiles[fd] = NULL;
1874 FILEDESC_UNLOCK(fdp);
1879 vfslocked = NDHASGIANT(&nd);
1880 NDFREE(&nd, NDF_ONLY_PNBUF);
1882 fp->f_vnode = nd.ni_vp;
1883 if (fp->f_data == NULL)
1884 fp->f_data = nd.ni_vp;
1885 if (fp->f_ops == &badfileops)
1887 fp->f_type = DTYPE_VNODE;
1888 VOP_UNLOCK(nd.ni_vp, 0, td);
1889 VFS_UNLOCK_GIANT(vfslocked);
1893 error = do_dup(td, DUP_FIXED, devnull, i, &retval);
1902 * Internal form of close.
1903 * Decrement reference count on file structure.
1904 * Note: td may be NULL when closing a file that was being passed in a
1907 * XXXRW: Giant is not required for the caller, but often will be held; this
1908 * makes it moderately likely the Giant will be recursed in the VFS case.
1911 closef(struct file *fp, struct thread *td)
1915 struct filedesc_to_leader *fdtol;
1916 struct filedesc *fdp;
1919 * POSIX record locking dictates that any close releases ALL
1920 * locks owned by this process. This is handled by setting
1921 * a flag in the unlock to free ONLY locks obeying POSIX
1922 * semantics, and not to free BSD-style file locks.
1923 * If the descriptor was in a message, POSIX-style locks
1924 * aren't passed with the descriptor, and the thread pointer
1925 * will be NULL. Callers should be careful only to pass a
1926 * NULL thread pointer when there really is no owning
1927 * context that might have locks, or the locks will be
1930 if (fp->f_type == DTYPE_VNODE && td != NULL) {
1934 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1935 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1936 lf.l_whence = SEEK_SET;
1939 lf.l_type = F_UNLCK;
1940 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
1941 F_UNLCK, &lf, F_POSIX);
1943 fdtol = td->td_proc->p_fdtol;
1944 if (fdtol != NULL) {
1946 * Handle special case where file descriptor table
1947 * is shared between multiple process leaders.
1949 fdp = td->td_proc->p_fd;
1951 for (fdtol = fdtol->fdl_next;
1952 fdtol != td->td_proc->p_fdtol;
1953 fdtol = fdtol->fdl_next) {
1954 if ((fdtol->fdl_leader->p_flag &
1957 fdtol->fdl_holdcount++;
1958 FILEDESC_UNLOCK(fdp);
1959 lf.l_whence = SEEK_SET;
1962 lf.l_type = F_UNLCK;
1964 (void) VOP_ADVLOCK(vp,
1965 (caddr_t)fdtol->fdl_leader,
1966 F_UNLCK, &lf, F_POSIX);
1968 fdtol->fdl_holdcount--;
1969 if (fdtol->fdl_holdcount == 0 &&
1970 fdtol->fdl_wakeup != 0) {
1971 fdtol->fdl_wakeup = 0;
1975 FILEDESC_UNLOCK(fdp);
1977 VFS_UNLOCK_GIANT(vfslocked);
1979 return (fdrop(fp, td));
1983 * Extract the file pointer associated with the specified descriptor for
1984 * the current user process.
1986 * If the descriptor doesn't exist, EBADF is returned.
1988 * If the descriptor exists but doesn't match 'flags' then
1989 * return EBADF for read attempts and EINVAL for write attempts.
1991 * If 'hold' is set (non-zero) the file's refcount will be bumped on return.
1992 * It should be dropped with fdrop().
1993 * If it is not set, then the refcount will not be bumped however the
1994 * thread's filedesc struct will be returned locked (for fgetsock).
1996 * If an error occured the non-zero error is returned and *fpp is set to NULL.
1997 * Otherwise *fpp is set and zero is returned.
2000 _fget(struct thread *td, int fd, struct file **fpp, int flags, int hold)
2002 struct filedesc *fdp;
2006 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
2009 if ((fp = fget_locked(fdp, fd)) == NULL || fp->f_ops == &badfileops) {
2010 FILEDESC_UNLOCK(fdp);
2015 * FREAD and FWRITE failure return EBADF as per POSIX.
2017 * Only one flag, or 0, may be specified.
2019 if (flags == FREAD && (fp->f_flag & FREAD) == 0) {
2020 FILEDESC_UNLOCK(fdp);
2023 if (flags == FWRITE && (fp->f_flag & FWRITE) == 0) {
2024 FILEDESC_UNLOCK(fdp);
2029 FILEDESC_UNLOCK(fdp);
2036 fget(struct thread *td, int fd, struct file **fpp)
2039 return(_fget(td, fd, fpp, 0, 1));
2043 fget_read(struct thread *td, int fd, struct file **fpp)
2046 return(_fget(td, fd, fpp, FREAD, 1));
2050 fget_write(struct thread *td, int fd, struct file **fpp)
2053 return(_fget(td, fd, fpp, FWRITE, 1));
2057 * Like fget() but loads the underlying vnode, or returns an error if
2058 * the descriptor does not represent a vnode. Note that pipes use vnodes
2059 * but never have VM objects. The returned vnode will be vref()d.
2061 * XXX: what about the unused flags ?
2064 _fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags)
2070 if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
2072 if (fp->f_vnode == NULL) {
2078 FILEDESC_UNLOCK(td->td_proc->p_fd);
2083 fgetvp(struct thread *td, int fd, struct vnode **vpp)
2086 return (_fgetvp(td, fd, vpp, 0));
2090 fgetvp_read(struct thread *td, int fd, struct vnode **vpp)
2093 return (_fgetvp(td, fd, vpp, FREAD));
2098 fgetvp_write(struct thread *td, int fd, struct vnode **vpp)
2101 return (_fgetvp(td, fd, vpp, FWRITE));
2106 * Like fget() but loads the underlying socket, or returns an error if
2107 * the descriptor does not represent a socket.
2109 * We bump the ref count on the returned socket. XXX Also obtain the SX
2110 * lock in the future.
2112 * XXXRW: fgetsock() and fputsock() are deprecated, as consumers should rely
2113 * on their file descriptor reference to prevent the socket from being
2117 fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp)
2127 if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
2129 if (fp->f_type != DTYPE_SOCKET) {
2134 *fflagp = fp->f_flag;
2139 FILEDESC_UNLOCK(td->td_proc->p_fd);
2144 * Drop the reference count on the socket and XXX release the SX lock in the
2145 * future. The last reference closes the socket.
2147 * XXXRW: fputsock() is deprecated, see comment for fgetsock().
2150 fputsock(struct socket *so)
2160 fdrop(struct file *fp, struct thread *td)
2164 return (fdrop_locked(fp, td));
2168 * Drop reference on struct file passed in, may call closef if the
2169 * reference hits zero.
2170 * Expects struct file locked, and will unlock it.
2173 fdrop_locked(struct file *fp, struct thread *td)
2177 FILE_LOCK_ASSERT(fp, MA_OWNED);
2179 if (--fp->f_count > 0) {
2183 /* We have the last ref so we can proceed without the file lock. */
2185 if (fp->f_count < 0)
2186 panic("fdrop: count < 0");
2187 if (fp->f_ops != &badfileops)
2188 error = fo_close(fp, td);
2192 sx_xlock(&filelist_lock);
2193 LIST_REMOVE(fp, f_list);
2195 sx_xunlock(&filelist_lock);
2197 uma_zfree(file_zone, fp);
2203 * Apply an advisory lock on a file descriptor.
2205 * Just attempt to get a record lock of the requested type on
2206 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2208 #ifndef _SYS_SYSPROTO_H_
2219 flock(struct thread *td, struct flock_args *uap)
2226 if ((error = fget(td, uap->fd, &fp)) != 0)
2228 if (fp->f_type != DTYPE_VNODE) {
2230 return (EOPNOTSUPP);
2235 lf.l_whence = SEEK_SET;
2238 if (uap->how & LOCK_UN) {
2239 lf.l_type = F_UNLCK;
2241 fp->f_flag &= ~FHASLOCK;
2243 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2246 if (uap->how & LOCK_EX)
2247 lf.l_type = F_WRLCK;
2248 else if (uap->how & LOCK_SH)
2249 lf.l_type = F_RDLCK;
2255 fp->f_flag |= FHASLOCK;
2257 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2258 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2265 * Duplicate the specified descriptor to a free descriptor.
2268 dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error)
2274 * If the to-be-dup'd fd number is greater than the allowed number
2275 * of file descriptors, or the fd to be dup'd has already been
2276 * closed, then reject.
2279 if (dfd < 0 || dfd >= fdp->fd_nfiles ||
2280 (wfp = fdp->fd_ofiles[dfd]) == NULL) {
2281 FILEDESC_UNLOCK(fdp);
2286 * There are two cases of interest here.
2288 * For ENODEV simply dup (dfd) to file descriptor
2289 * (indx) and return.
2291 * For ENXIO steal away the file structure from (dfd) and
2292 * store it in (indx). (dfd) is effectively closed by
2295 * Any other error code is just returned.
2300 * Check that the mode the file is being opened for is a
2301 * subset of the mode of the existing descriptor.
2304 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2306 FILEDESC_UNLOCK(fdp);
2309 fp = fdp->fd_ofiles[indx];
2310 fdp->fd_ofiles[indx] = wfp;
2311 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2316 FILEDESC_UNLOCK(fdp);
2319 * We now own the reference to fp that the ofiles[]
2320 * array used to own. Release it.
2323 fdrop_locked(fp, td);
2329 * Steal away the file pointer from dfd and stuff it into indx.
2331 fp = fdp->fd_ofiles[indx];
2332 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2333 fdp->fd_ofiles[dfd] = NULL;
2334 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2335 fdp->fd_ofileflags[dfd] = 0;
2343 * We now own the reference to fp that the ofiles[] array
2344 * used to own. Release it.
2347 fdrop_locked(fp, td);
2349 FILEDESC_UNLOCK(fdp);
2354 FILEDESC_UNLOCK(fdp);
2361 * Scan all active processes to see if any of them have a current
2362 * or root directory of `olddp'. If so, replace them with the new
2366 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2368 struct filedesc *fdp;
2372 if (vrefcnt(olddp) == 1)
2374 sx_slock(&allproc_lock);
2375 LIST_FOREACH(p, &allproc, p_list) {
2380 FILEDESC_LOCK_FAST(fdp);
2381 if (fdp->fd_cdir == olddp) {
2383 fdp->fd_cdir = newdp;
2386 if (fdp->fd_rdir == olddp) {
2388 fdp->fd_rdir = newdp;
2391 FILEDESC_UNLOCK_FAST(fdp);
2396 sx_sunlock(&allproc_lock);
2397 if (rootvnode == olddp) {
2404 struct filedesc_to_leader *
2405 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2407 struct filedesc_to_leader *fdtol;
2409 MALLOC(fdtol, struct filedesc_to_leader *,
2410 sizeof(struct filedesc_to_leader),
2411 M_FILEDESC_TO_LEADER,
2413 fdtol->fdl_refcount = 1;
2414 fdtol->fdl_holdcount = 0;
2415 fdtol->fdl_wakeup = 0;
2416 fdtol->fdl_leader = leader;
2419 fdtol->fdl_next = old->fdl_next;
2420 fdtol->fdl_prev = old;
2421 old->fdl_next = fdtol;
2422 fdtol->fdl_next->fdl_prev = fdtol;
2423 FILEDESC_UNLOCK(fdp);
2425 fdtol->fdl_next = fdtol;
2426 fdtol->fdl_prev = fdtol;
2432 * Get file structures.
2435 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2438 struct filedesc *fdp;
2444 * Note: because the number of file descriptors is calculated
2445 * in different ways for sizing vs returning the data,
2446 * there is information leakage from the first loop. However,
2447 * it is of a similar order of magnitude to the leakage from
2448 * global system statistics such as kern.openfiles.
2450 error = sysctl_wire_old_buffer(req, 0);
2453 if (req->oldptr == NULL) {
2454 n = 16; /* A slight overestimate. */
2455 sx_slock(&filelist_lock);
2456 LIST_FOREACH(fp, &filehead, f_list) {
2458 * We should grab the lock, but this is an
2459 * estimate, so does it really matter?
2461 /* mtx_lock(fp->f_mtxp); */
2463 /* mtx_unlock(f->f_mtxp); */
2465 sx_sunlock(&filelist_lock);
2466 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2469 bzero(&xf, sizeof(xf));
2470 xf.xf_size = sizeof(xf);
2471 sx_slock(&allproc_lock);
2472 LIST_FOREACH(p, &allproc, p_list) {
2473 if (p->p_state == PRS_NEW)
2476 if (p_cansee(req->td, p) != 0) {
2480 xf.xf_pid = p->p_pid;
2481 xf.xf_uid = p->p_ucred->cr_uid;
2486 FILEDESC_LOCK_FAST(fdp);
2487 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
2488 if ((fp = fdp->fd_ofiles[n]) == NULL)
2492 xf.xf_data = fp->f_data;
2493 xf.xf_vnode = fp->f_vnode;
2494 xf.xf_type = fp->f_type;
2495 xf.xf_count = fp->f_count;
2496 xf.xf_msgcount = fp->f_msgcount;
2497 xf.xf_offset = fp->f_offset;
2498 xf.xf_flag = fp->f_flag;
2499 error = SYSCTL_OUT(req, &xf, sizeof(xf));
2503 FILEDESC_UNLOCK_FAST(fdp);
2508 sx_sunlock(&allproc_lock);
2512 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2513 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2517 * For the purposes of debugging, generate a human-readable string for the
2521 file_type_to_name(short type)
2543 * For the purposes of debugging, identify a process (if any, perhaps one of
2544 * many) that references the passed file in its file descriptor array. Return
2547 static struct proc *
2548 file_to_first_proc(struct file *fp)
2550 struct filedesc *fdp;
2554 LIST_FOREACH(p, &allproc, p_list) {
2555 if (p->p_state == PRS_NEW)
2560 for (n = 0; n < fdp->fd_nfiles; n++) {
2561 if (fp == fdp->fd_ofiles[n])
2568 DB_SHOW_COMMAND(files, db_show_files)
2573 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", "File",
2574 "Type", "Data", "Flag", "GCFl", "Count", "MCount", "Vnode",
2576 LIST_FOREACH(fp, &filehead, f_list) {
2577 p = file_to_first_proc(fp);
2578 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
2579 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
2580 fp->f_gcflag, fp->f_count, fp->f_msgcount, fp->f_vnode,
2581 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
2586 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2587 &maxfilesperproc, 0, "Maximum files allowed open per process");
2589 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2590 &maxfiles, 0, "Maximum number of files");
2592 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2593 &openfiles, 0, "System-wide number of open files");
2597 filelistinit(void *dummy)
2600 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
2601 NULL, NULL, UMA_ALIGN_PTR, 0);
2602 sx_init(&filelist_lock, "filelist lock");
2603 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
2604 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
2606 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL)
2608 /*-------------------------------------------------------------------*/
2611 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td)
2618 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td)
2625 badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td)
2632 badfo_kqfilter(struct file *fp, struct knote *kn)
2639 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td)
2646 badfo_close(struct file *fp, struct thread *td)
2652 struct fileops badfileops = {
2653 .fo_read = badfo_readwrite,
2654 .fo_write = badfo_readwrite,
2655 .fo_ioctl = badfo_ioctl,
2656 .fo_poll = badfo_poll,
2657 .fo_kqfilter = badfo_kqfilter,
2658 .fo_stat = badfo_stat,
2659 .fo_close = badfo_close,
2663 /*-------------------------------------------------------------------*/
2666 * File Descriptor pseudo-device driver (/dev/fd/).
2668 * Opening minor device N dup()s the file (if any) connected to file
2669 * descriptor N belonging to the calling process. Note that this driver
2670 * consists of only the ``open()'' routine, because all subsequent
2671 * references to this file will be direct to the other driver.
2673 * XXX: we could give this one a cloning event handler if necessary.
2678 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
2682 * XXX Kludge: set curthread->td_dupfd to contain the value of the
2683 * the file descriptor being sought for duplication. The error
2684 * return ensures that the vnode for this device will be released
2685 * by vn_open. Open will detect this special error and take the
2686 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2687 * will simply report the error.
2689 td->td_dupfd = dev2unit(dev);
2693 static struct cdevsw fildesc_cdevsw = {
2694 .d_version = D_VERSION,
2695 .d_flags = D_NEEDGIANT,
2701 fildesc_drvinit(void *unused)
2705 dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0");
2706 make_dev_alias(dev, "stdin");
2707 dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1");
2708 make_dev_alias(dev, "stdout");
2709 dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2");
2710 make_dev_alias(dev, "stderr");
2713 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL)