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_capsicum.h"
41 #include "opt_compat.h"
43 #include "opt_ktrace.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
48 #include <sys/capsicum.h>
50 #include <sys/fcntl.h>
52 #include <sys/filedesc.h>
53 #include <sys/filio.h>
55 #include <sys/kernel.h>
56 #include <sys/limits.h>
58 #include <sys/malloc.h>
59 #include <sys/mount.h>
60 #include <sys/mutex.h>
61 #include <sys/namei.h>
62 #include <sys/selinfo.h>
65 #include <sys/protosw.h>
66 #include <sys/racct.h>
67 #include <sys/resourcevar.h>
69 #include <sys/signalvar.h>
70 #include <sys/socketvar.h>
73 #include <sys/syscallsubr.h>
74 #include <sys/sysctl.h>
75 #include <sys/sysproto.h>
76 #include <sys/unistd.h>
78 #include <sys/vnode.h>
80 #include <sys/ktrace.h>
85 #include <security/audit/audit.h>
92 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
93 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
94 "file desc to leader structures");
95 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
96 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
98 MALLOC_DECLARE(M_FADVISE);
100 static uma_zone_t file_zone;
101 static uma_zone_t filedesc0_zone;
103 static int closefp(struct filedesc *fdp, int fd, struct file *fp,
104 struct thread *td, int holdleaders);
105 static int do_dup(struct thread *td, int flags, int old, int new);
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 proc *p);
114 /* Flags for do_dup() */
115 #define DUP_FIXED 0x1 /* Force fixed allocation. */
116 #define DUP_FCNTL 0x2 /* fcntl()-style errors. */
117 #define DUP_CLOEXEC 0x4 /* Atomically set FD_CLOEXEC. */
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 openfiles; /* actual number of open files */
178 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
179 void (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
181 /* A mutex to protect the association between a proc and filedesc. */
182 static struct mtx fdesc_mtx;
185 * If low >= size, just return low. Otherwise find the first zero bit in the
186 * given bitmap, starting at low and not exceeding size - 1. Return size if
190 fd_first_free(struct filedesc *fdp, int low, int size)
192 NDSLOTTYPE *map = fdp->fd_map;
200 if (low % NDENTRIES) {
201 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
202 if ((mask &= ~map[off]) != 0UL)
203 return (off * NDENTRIES + ffsl(mask) - 1);
206 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
207 if (map[off] != ~0UL)
208 return (off * NDENTRIES + ffsl(~map[off]) - 1);
213 * Find the highest non-zero bit in the given bitmap, starting at 0 and
214 * not exceeding size - 1. Return -1 if not found.
217 fd_last_used(struct filedesc *fdp, int size)
219 NDSLOTTYPE *map = fdp->fd_map;
224 if (size % NDENTRIES) {
225 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
226 if ((mask &= map[off]) != 0)
227 return (off * NDENTRIES + flsl(mask) - 1);
230 for (minoff = NDSLOT(0); off >= minoff; --off)
232 return (off * NDENTRIES + flsl(map[off]) - 1);
238 fdisused(struct filedesc *fdp, int fd)
241 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
242 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
244 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
249 * Mark a file descriptor as used.
252 fdused_init(struct filedesc *fdp, int fd)
255 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
257 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
261 fdused(struct filedesc *fdp, int fd)
264 FILEDESC_XLOCK_ASSERT(fdp);
266 fdused_init(fdp, fd);
267 if (fd > fdp->fd_lastfile)
268 fdp->fd_lastfile = fd;
269 if (fd == fdp->fd_freefile)
270 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
274 * Mark a file descriptor as unused.
277 fdunused(struct filedesc *fdp, int fd)
280 FILEDESC_XLOCK_ASSERT(fdp);
282 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
283 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
284 ("fd=%d is still in use", fd));
286 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
287 if (fd < fdp->fd_freefile)
288 fdp->fd_freefile = fd;
289 if (fd == fdp->fd_lastfile)
290 fdp->fd_lastfile = fd_last_used(fdp, fd);
294 * Free a file descriptor.
296 * Avoid some work if fdp is about to be destroyed.
299 fdefree_last(struct filedescent *fde)
302 filecaps_free(&fde->fde_caps);
306 fdfree(struct filedesc *fdp, int fd)
308 struct filedescent *fde;
310 fde = &fdp->fd_ofiles[fd];
312 seq_write_begin(&fde->fde_seq);
315 bzero(fde, fde_change_size);
318 seq_write_end(&fde->fde_seq);
323 * System calls on descriptors.
325 #ifndef _SYS_SYSPROTO_H_
326 struct getdtablesize_args {
332 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
334 struct proc *p = td->td_proc;
339 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
340 lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
342 if (lim < td->td_retval[0])
343 td->td_retval[0] = lim;
348 * Duplicate a file descriptor to a particular value.
350 * Note: keep in mind that a potential race condition exists when closing
351 * descriptors from a shared descriptor table (via rfork).
353 #ifndef _SYS_SYSPROTO_H_
361 sys_dup2(struct thread *td, struct dup2_args *uap)
364 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to));
368 * Duplicate a file descriptor.
370 #ifndef _SYS_SYSPROTO_H_
377 sys_dup(struct thread *td, struct dup_args *uap)
380 return (do_dup(td, 0, (int)uap->fd, 0));
384 * The file control system call.
386 #ifndef _SYS_SYSPROTO_H_
395 sys_fcntl(struct thread *td, struct fcntl_args *uap)
398 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
402 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
415 * Convert old flock structure to new.
417 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
418 fl.l_start = ofl.l_start;
419 fl.l_len = ofl.l_len;
420 fl.l_pid = ofl.l_pid;
421 fl.l_type = ofl.l_type;
422 fl.l_whence = ofl.l_whence;
436 arg1 = (intptr_t)&fl;
442 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
443 arg1 = (intptr_t)&fl;
451 error = kern_fcntl(td, fd, cmd, arg1);
454 if (cmd == F_OGETLK) {
455 ofl.l_start = fl.l_start;
456 ofl.l_len = fl.l_len;
457 ofl.l_pid = fl.l_pid;
458 ofl.l_type = fl.l_type;
459 ofl.l_whence = fl.l_whence;
460 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
461 } else if (cmd == F_GETLK) {
462 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
468 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
470 struct filedesc *fdp;
472 struct file *fp, *fp2;
473 struct filedescent *fde;
489 error = do_dup(td, DUP_FCNTL, fd, tmp);
492 case F_DUPFD_CLOEXEC:
494 error = do_dup(td, DUP_FCNTL | DUP_CLOEXEC, fd, tmp);
499 error = do_dup(td, DUP_FIXED, fd, tmp);
502 case F_DUP2FD_CLOEXEC:
504 error = do_dup(td, DUP_FIXED | DUP_CLOEXEC, fd, tmp);
509 if (fget_locked(fdp, fd) == NULL) {
510 FILEDESC_SUNLOCK(fdp);
514 fde = &fdp->fd_ofiles[fd];
516 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
517 FILEDESC_SUNLOCK(fdp);
522 if (fget_locked(fdp, fd) == NULL) {
523 FILEDESC_XUNLOCK(fdp);
527 fde = &fdp->fd_ofiles[fd];
528 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
529 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
530 FILEDESC_XUNLOCK(fdp);
534 error = fget_fcntl(td, fd,
535 cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp);
538 td->td_retval[0] = OFLAGS(fp->f_flag);
543 error = fget_fcntl(td, fd,
544 cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp);
548 tmp = flg = fp->f_flag;
550 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
551 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
552 tmp = fp->f_flag & FNONBLOCK;
553 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
558 tmp = fp->f_flag & FASYNC;
559 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
564 atomic_clear_int(&fp->f_flag, FNONBLOCK);
566 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
571 error = fget_fcntl(td, fd,
572 cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp);
575 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
577 td->td_retval[0] = tmp;
582 error = fget_fcntl(td, fd,
583 cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp);
587 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
592 error = priv_check(td, PRIV_NFS_LOCKD);
600 /* FALLTHROUGH F_SETLK */
604 cap_rights_init(&rights, CAP_FLOCK);
605 error = fget_unlocked(fdp, fd, &rights, &fp, NULL);
608 if (fp->f_type != DTYPE_VNODE) {
614 flp = (struct flock *)arg;
615 if (flp->l_whence == SEEK_CUR) {
616 foffset = foffset_get(fp);
619 foffset > OFF_MAX - flp->l_start)) {
620 FILEDESC_SUNLOCK(fdp);
625 flp->l_start += foffset;
629 switch (flp->l_type) {
631 if ((fp->f_flag & FREAD) == 0) {
635 PROC_LOCK(p->p_leader);
636 p->p_leader->p_flag |= P_ADVLOCK;
637 PROC_UNLOCK(p->p_leader);
638 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
642 if ((fp->f_flag & FWRITE) == 0) {
646 PROC_LOCK(p->p_leader);
647 p->p_leader->p_flag |= P_ADVLOCK;
648 PROC_UNLOCK(p->p_leader);
649 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
653 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
658 * Temporary api for testing remote lock
661 if (flg != F_REMOTE) {
665 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
666 F_UNLCKSYS, flp, flg);
672 if (error != 0 || flp->l_type == F_UNLCK ||
673 flp->l_type == F_UNLCKSYS) {
679 * Check for a race with close.
681 * The vnode is now advisory locked (or unlocked, but this case
682 * is not really important) as the caller requested.
683 * We had to drop the filedesc lock, so we need to recheck if
684 * the descriptor is still valid, because if it was closed
685 * in the meantime we need to remove advisory lock from the
686 * vnode - close on any descriptor leading to an advisory
687 * locked vnode, removes that lock.
688 * We will return 0 on purpose in that case, as the result of
689 * successful advisory lock might have been externally visible
690 * already. This is fine - effectively we pretend to the caller
691 * that the closing thread was a bit slower and that the
692 * advisory lock succeeded before the close.
694 error = fget_unlocked(fdp, fd, &rights, &fp2, NULL);
700 flp->l_whence = SEEK_SET;
703 flp->l_type = F_UNLCK;
704 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
705 F_UNLCK, flp, F_POSIX);
712 error = fget_unlocked(fdp, fd,
713 cap_rights_init(&rights, CAP_FLOCK), &fp, NULL);
716 if (fp->f_type != DTYPE_VNODE) {
721 flp = (struct flock *)arg;
722 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
723 flp->l_type != F_UNLCK) {
728 if (flp->l_whence == SEEK_CUR) {
729 foffset = foffset_get(fp);
730 if ((flp->l_start > 0 &&
731 foffset > OFF_MAX - flp->l_start) ||
733 foffset < OFF_MIN - flp->l_start)) {
734 FILEDESC_SUNLOCK(fdp);
739 flp->l_start += foffset;
742 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
748 arg = arg ? 128 * 1024: 0;
751 error = fget_unlocked(fdp, fd, NULL, &fp, NULL);
754 if (fp->f_type != DTYPE_VNODE) {
761 * Exclusive lock synchronizes against f_seqcount reads and
762 * writes in sequential_heuristic().
764 error = vn_lock(vp, LK_EXCLUSIVE);
770 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
771 fp->f_seqcount = (arg + bsize - 1) / bsize;
772 atomic_set_int(&fp->f_flag, FRDAHEAD);
774 atomic_clear_int(&fp->f_flag, FRDAHEAD);
788 getmaxfd(struct proc *p)
793 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
800 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
803 do_dup(struct thread *td, int flags, int old, int new)
805 struct filedesc *fdp;
806 struct filedescent *oldfde, *newfde;
816 * Verify we have a valid descriptor to dup from and possibly to
817 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
818 * return EINVAL when the new descriptor is out of bounds.
823 return (flags & DUP_FCNTL ? EINVAL : EBADF);
826 return (flags & DUP_FCNTL ? EINVAL : EBADF);
829 if (fget_locked(fdp, old) == NULL) {
830 FILEDESC_XUNLOCK(fdp);
833 oldfde = &fdp->fd_ofiles[old];
834 if (flags & DUP_FIXED && old == new) {
835 td->td_retval[0] = new;
836 if (flags & DUP_CLOEXEC)
837 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
838 FILEDESC_XUNLOCK(fdp);
841 fp = oldfde->fde_file;
845 * If the caller specified a file descriptor, make sure the file
846 * table is large enough to hold it, and grab it. Otherwise, just
847 * allocate a new descriptor the usual way.
849 if (flags & DUP_FIXED) {
850 if (new >= fdp->fd_nfiles) {
852 * The resource limits are here instead of e.g.
853 * fdalloc(), because the file descriptor table may be
854 * shared between processes, so we can't really use
855 * racct_add()/racct_sub(). Instead of counting the
856 * number of actually allocated descriptors, just put
857 * the limit on the size of the file descriptor table.
861 error = racct_set(p, RACCT_NOFILE, new + 1);
864 FILEDESC_XUNLOCK(fdp);
869 fdgrowtable_exp(fdp, new + 1);
870 oldfde = &fdp->fd_ofiles[old];
872 newfde = &fdp->fd_ofiles[new];
873 if (newfde->fde_file == NULL)
876 if ((error = fdalloc(td, new, &new)) != 0) {
877 FILEDESC_XUNLOCK(fdp);
881 newfde = &fdp->fd_ofiles[new];
884 KASSERT(fp == oldfde->fde_file, ("old fd has been modified"));
885 KASSERT(old != new, ("new fd is same as old"));
887 delfp = newfde->fde_file;
890 * Duplicate the source descriptor.
893 seq_write_begin(&newfde->fde_seq);
895 filecaps_free(&newfde->fde_caps);
896 memcpy(newfde, oldfde, fde_change_size);
897 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps);
898 if ((flags & DUP_CLOEXEC) != 0)
899 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
901 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
903 seq_write_end(&newfde->fde_seq);
905 td->td_retval[0] = new;
908 (void) closefp(fdp, new, delfp, td, 1);
909 /* closefp() drops the FILEDESC lock for us. */
911 FILEDESC_XUNLOCK(fdp);
918 * If sigio is on the list associated with a process or process group,
919 * disable signalling from the device, remove sigio from the list and
923 funsetown(struct sigio **sigiop)
933 *(sigio->sio_myref) = NULL;
934 if ((sigio)->sio_pgid < 0) {
935 struct pgrp *pg = (sigio)->sio_pgrp;
937 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
941 struct proc *p = (sigio)->sio_proc;
943 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
948 crfree(sigio->sio_ucred);
949 free(sigio, M_SIGIO);
953 * Free a list of sigio structures.
954 * We only need to lock the SIGIO_LOCK because we have made ourselves
955 * inaccessible to callers of fsetown and therefore do not need to lock
956 * the proc or pgrp struct for the list manipulation.
959 funsetownlst(struct sigiolst *sigiolst)
965 sigio = SLIST_FIRST(sigiolst);
972 * Every entry of the list should belong
973 * to a single proc or pgrp.
975 if (sigio->sio_pgid < 0) {
976 pg = sigio->sio_pgrp;
977 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
978 } else /* if (sigio->sio_pgid > 0) */ {
980 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
984 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
985 *(sigio->sio_myref) = NULL;
987 KASSERT(sigio->sio_pgid < 0,
988 ("Proc sigio in pgrp sigio list"));
989 KASSERT(sigio->sio_pgrp == pg,
990 ("Bogus pgrp in sigio list"));
992 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
995 } else /* if (p != NULL) */ {
996 KASSERT(sigio->sio_pgid > 0,
997 ("Pgrp sigio in proc sigio list"));
998 KASSERT(sigio->sio_proc == p,
999 ("Bogus proc in sigio list"));
1001 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1006 crfree(sigio->sio_ucred);
1007 free(sigio, M_SIGIO);
1014 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1016 * After permission checking, add a sigio structure to the sigio list for
1017 * the process or process group.
1020 fsetown(pid_t pgid, struct sigio **sigiop)
1024 struct sigio *sigio;
1034 /* Allocate and fill in the new sigio out of locks. */
1035 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1036 sigio->sio_pgid = pgid;
1037 sigio->sio_ucred = crhold(curthread->td_ucred);
1038 sigio->sio_myref = sigiop;
1040 sx_slock(&proctree_lock);
1049 * Policy - Don't allow a process to FSETOWN a process
1050 * in another session.
1052 * Remove this test to allow maximum flexibility or
1053 * restrict FSETOWN to the current process or process
1054 * group for maximum safety.
1057 if (proc->p_session != curthread->td_proc->p_session) {
1063 } else /* if (pgid < 0) */ {
1064 pgrp = pgfind(-pgid);
1072 * Policy - Don't allow a process to FSETOWN a process
1073 * in another session.
1075 * Remove this test to allow maximum flexibility or
1076 * restrict FSETOWN to the current process or process
1077 * group for maximum safety.
1079 if (pgrp->pg_session != curthread->td_proc->p_session) {
1090 * Since funsetownlst() is called without the proctree
1091 * locked, we need to check for P_WEXIT.
1092 * XXX: is ESRCH correct?
1094 if ((proc->p_flag & P_WEXIT) != 0) {
1099 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1100 sigio->sio_proc = proc;
1104 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1105 sigio->sio_pgrp = pgrp;
1108 sx_sunlock(&proctree_lock);
1115 sx_sunlock(&proctree_lock);
1116 crfree(sigio->sio_ucred);
1117 free(sigio, M_SIGIO);
1122 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1126 struct sigio **sigiop;
1131 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1137 * Function drops the filedesc lock on return.
1140 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1145 FILEDESC_XLOCK_ASSERT(fdp);
1148 if (td->td_proc->p_fdtol != NULL) {
1150 * Ask fdfree() to sleep to ensure that all relevant
1151 * process leaders can be traversed in closef().
1153 fdp->fd_holdleaderscount++;
1160 * We now hold the fp reference that used to be owned by the
1161 * descriptor array. We have to unlock the FILEDESC *AFTER*
1162 * knote_fdclose to prevent a race of the fd getting opened, a knote
1163 * added, and deleteing a knote for the new fd.
1165 knote_fdclose(td, fd);
1168 * We need to notify mqueue if the object is of type mqueue.
1170 if (fp->f_type == DTYPE_MQUEUE)
1171 mq_fdclose(td, fd, fp);
1172 FILEDESC_XUNLOCK(fdp);
1174 error = closef(fp, td);
1176 FILEDESC_XLOCK(fdp);
1177 fdp->fd_holdleaderscount--;
1178 if (fdp->fd_holdleaderscount == 0 &&
1179 fdp->fd_holdleaderswakeup != 0) {
1180 fdp->fd_holdleaderswakeup = 0;
1181 wakeup(&fdp->fd_holdleaderscount);
1183 FILEDESC_XUNLOCK(fdp);
1189 * Close a file descriptor.
1191 #ifndef _SYS_SYSPROTO_H_
1200 struct close_args *uap;
1203 return (kern_close(td, uap->fd));
1211 struct filedesc *fdp;
1214 fdp = td->td_proc->p_fd;
1216 AUDIT_SYSCLOSE(td, fd);
1218 FILEDESC_XLOCK(fdp);
1219 if ((fp = fget_locked(fdp, fd)) == NULL) {
1220 FILEDESC_XUNLOCK(fdp);
1225 /* closefp() drops the FILEDESC lock for us. */
1226 return (closefp(fdp, fd, fp, td, 1));
1230 * Close open file descriptors.
1232 #ifndef _SYS_SYSPROTO_H_
1233 struct closefrom_args {
1239 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1241 struct filedesc *fdp;
1244 fdp = td->td_proc->p_fd;
1245 AUDIT_ARG_FD(uap->lowfd);
1248 * Treat negative starting file descriptor values identical to
1249 * closefrom(0) which closes all files.
1253 FILEDESC_SLOCK(fdp);
1254 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1255 if (fdp->fd_ofiles[fd].fde_file != NULL) {
1256 FILEDESC_SUNLOCK(fdp);
1257 (void)kern_close(td, fd);
1258 FILEDESC_SLOCK(fdp);
1261 FILEDESC_SUNLOCK(fdp);
1265 #if defined(COMPAT_43)
1267 * Return status information about a file descriptor.
1269 #ifndef _SYS_SYSPROTO_H_
1270 struct ofstat_args {
1277 ofstat(struct thread *td, struct ofstat_args *uap)
1283 error = kern_fstat(td, uap->fd, &ub);
1286 error = copyout(&oub, uap->sb, sizeof(oub));
1290 #endif /* COMPAT_43 */
1293 * Return status information about a file descriptor.
1295 #ifndef _SYS_SYSPROTO_H_
1303 sys_fstat(struct thread *td, struct fstat_args *uap)
1308 error = kern_fstat(td, uap->fd, &ub);
1310 error = copyout(&ub, uap->sb, sizeof(ub));
1315 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1318 cap_rights_t rights;
1323 error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp);
1327 AUDIT_ARG_FILE(td->td_proc, fp);
1329 error = fo_stat(fp, sbp, td->td_ucred, td);
1332 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1339 * Return status information about a file descriptor.
1341 #ifndef _SYS_SYSPROTO_H_
1342 struct nfstat_args {
1349 sys_nfstat(struct thread *td, struct nfstat_args *uap)
1355 error = kern_fstat(td, uap->fd, &ub);
1357 cvtnstat(&ub, &nub);
1358 error = copyout(&nub, uap->sb, sizeof(nub));
1364 * Return pathconf information about a file descriptor.
1366 #ifndef _SYS_SYSPROTO_H_
1367 struct fpathconf_args {
1374 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1378 cap_rights_t rights;
1381 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp);
1385 /* If asynchronous I/O is available, it works for all descriptors. */
1386 if (uap->name == _PC_ASYNC_IO) {
1387 td->td_retval[0] = async_io_version;
1392 vn_lock(vp, LK_SHARED | LK_RETRY);
1393 error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1395 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1396 if (uap->name != _PC_PIPE_BUF) {
1399 td->td_retval[0] = PIPE_BUF;
1411 * Initialize filecaps structure.
1414 filecaps_init(struct filecaps *fcaps)
1417 bzero(fcaps, sizeof(*fcaps));
1418 fcaps->fc_nioctls = -1;
1422 * Copy filecaps structure allocating memory for ioctls array if needed.
1425 filecaps_copy(const struct filecaps *src, struct filecaps *dst)
1430 if (src->fc_ioctls != NULL) {
1431 KASSERT(src->fc_nioctls > 0,
1432 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1434 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1435 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1436 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1441 * Move filecaps structure to the new place and clear the old place.
1444 filecaps_move(struct filecaps *src, struct filecaps *dst)
1448 bzero(src, sizeof(*src));
1452 * Fill the given filecaps structure with full rights.
1455 filecaps_fill(struct filecaps *fcaps)
1458 CAP_ALL(&fcaps->fc_rights);
1459 fcaps->fc_ioctls = NULL;
1460 fcaps->fc_nioctls = -1;
1461 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1465 * Free memory allocated within filecaps structure.
1468 filecaps_free(struct filecaps *fcaps)
1471 free(fcaps->fc_ioctls, M_FILECAPS);
1472 bzero(fcaps, sizeof(*fcaps));
1476 * Validate the given filecaps structure.
1479 filecaps_validate(const struct filecaps *fcaps, const char *func)
1482 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1483 ("%s: invalid rights", func));
1484 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1485 ("%s: invalid fcntls", func));
1486 KASSERT(fcaps->fc_fcntls == 0 ||
1487 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1488 ("%s: fcntls without CAP_FCNTL", func));
1489 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1490 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1491 ("%s: invalid ioctls", func));
1492 KASSERT(fcaps->fc_nioctls == 0 ||
1493 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1494 ("%s: ioctls without CAP_IOCTL", func));
1498 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1502 FILEDESC_XLOCK_ASSERT(fdp);
1504 nfd1 = fdp->fd_nfiles * 2;
1507 fdgrowtable(fdp, nfd1);
1511 * Grow the file table to accomodate (at least) nfd descriptors.
1514 fdgrowtable(struct filedesc *fdp, int nfd)
1516 struct filedesc0 *fdp0;
1517 struct freetable *ft;
1518 struct fdescenttbl *ntable;
1519 struct fdescenttbl *otable;
1520 int nnfiles, onfiles;
1521 NDSLOTTYPE *nmap, *omap;
1524 * If lastfile is -1 this struct filedesc was just allocated and we are
1525 * growing it to accomodate for the one we are going to copy from. There
1526 * is no need to have a lock on this one as it's not visible to anyone.
1528 if (fdp->fd_lastfile != -1)
1529 FILEDESC_XLOCK_ASSERT(fdp);
1531 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1533 /* save old values */
1534 onfiles = fdp->fd_nfiles;
1535 otable = fdp->fd_files;
1538 /* compute the size of the new table */
1539 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1540 if (nnfiles <= onfiles)
1541 /* the table is already large enough */
1545 * Allocate a new table. We need enough space for the number of
1546 * entries, file entries themselves and the struct freetable we will use
1547 * when we decommission the table and place it on the freelist.
1548 * We place the struct freetable in the middle so we don't have
1549 * to worry about padding.
1551 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1552 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1553 sizeof(struct freetable),
1554 M_FILEDESC, M_ZERO | M_WAITOK);
1555 /* copy the old data */
1556 ntable->fdt_nfiles = nnfiles;
1557 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1558 onfiles * sizeof(ntable->fdt_ofiles[0]));
1561 * Allocate a new map only if the old is not large enough. It will
1562 * grow at a slower rate than the table as it can map more
1563 * entries than the table can hold.
1565 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1566 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1568 /* copy over the old data and update the pointer */
1569 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1574 * Make sure that ntable is correctly initialized before we replace
1575 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1578 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1581 * Do not free the old file table, as some threads may still
1582 * reference entries within it. Instead, place it on a freelist
1583 * which will be processed when the struct filedesc is released.
1585 * Note that if onfiles == NDFILE, we're dealing with the original
1586 * static allocation contained within (struct filedesc0 *)fdp,
1587 * which must not be freed.
1589 if (onfiles > NDFILE) {
1590 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1591 fdp0 = (struct filedesc0 *)fdp;
1592 ft->ft_table = otable;
1593 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1596 * The map does not have the same possibility of threads still
1597 * holding references to it. So always free it as long as it
1598 * does not reference the original static allocation.
1600 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1601 free(omap, M_FILEDESC);
1605 * Allocate a file descriptor for the process.
1608 fdalloc(struct thread *td, int minfd, int *result)
1610 struct proc *p = td->td_proc;
1611 struct filedesc *fdp = p->p_fd;
1612 int fd = -1, maxfd, allocfd;
1617 FILEDESC_XLOCK_ASSERT(fdp);
1619 if (fdp->fd_freefile > minfd)
1620 minfd = fdp->fd_freefile;
1622 maxfd = getmaxfd(p);
1625 * Search the bitmap for a free descriptor starting at minfd.
1626 * If none is found, grow the file table.
1628 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1631 if (fd >= fdp->fd_nfiles) {
1632 allocfd = min(fd * 2, maxfd);
1635 error = racct_set(p, RACCT_NOFILE, allocfd);
1641 * fd is already equal to first free descriptor >= minfd, so
1642 * we only need to grow the table and we are done.
1644 fdgrowtable_exp(fdp, allocfd);
1648 * Perform some sanity checks, then mark the file descriptor as
1649 * used and return it to the caller.
1651 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1652 ("invalid descriptor %d", fd));
1653 KASSERT(!fdisused(fdp, fd),
1654 ("fd_first_free() returned non-free descriptor"));
1655 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1656 ("file descriptor isn't free"));
1657 KASSERT(fdp->fd_ofiles[fd].fde_flags == 0, ("file flags are set"));
1664 * Allocate n file descriptors for the process.
1667 fdallocn(struct thread *td, int minfd, int *fds, int n)
1669 struct proc *p = td->td_proc;
1670 struct filedesc *fdp = p->p_fd;
1673 FILEDESC_XLOCK_ASSERT(fdp);
1675 for (i = 0; i < n; i++)
1676 if (fdalloc(td, 0, &fds[i]) != 0)
1680 for (i--; i >= 0; i--)
1681 fdunused(fdp, fds[i]);
1689 * Create a new open file structure and allocate a file decriptor for the
1690 * process that refers to it. We add one reference to the file for the
1691 * descriptor table and one reference for resultfp. This is to prevent us
1692 * being preempted and the entry in the descriptor table closed after we
1693 * release the FILEDESC lock.
1696 falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags)
1701 error = falloc_noinstall(td, &fp);
1703 return (error); /* no reference held on error */
1705 error = finstall(td, fp, &fd, flags, NULL);
1707 fdrop(fp, td); /* one reference (fp only) */
1711 if (resultfp != NULL)
1712 *resultfp = fp; /* copy out result */
1714 fdrop(fp, td); /* release local reference */
1716 if (resultfd != NULL)
1723 * Create a new open file structure without allocating a file descriptor.
1726 falloc_noinstall(struct thread *td, struct file **resultfp)
1729 int maxuserfiles = maxfiles - (maxfiles / 20);
1730 static struct timeval lastfail;
1733 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1735 if ((openfiles >= maxuserfiles &&
1736 priv_check(td, PRIV_MAXFILES) != 0) ||
1737 openfiles >= maxfiles) {
1738 if (ppsratecheck(&lastfail, &curfail, 1)) {
1739 printf("kern.maxfiles limit exceeded by uid %i, "
1740 "please see tuning(7).\n", td->td_ucred->cr_ruid);
1744 atomic_add_int(&openfiles, 1);
1745 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1746 refcount_init(&fp->f_count, 1);
1747 fp->f_cred = crhold(td->td_ucred);
1748 fp->f_ops = &badfileops;
1754 * Install a file in a file descriptor table.
1757 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1758 struct filecaps *fcaps)
1760 struct filedesc *fdp = td->td_proc->p_fd;
1761 struct filedescent *fde;
1764 KASSERT(fd != NULL, ("%s: fd == NULL", __func__));
1765 KASSERT(fp != NULL, ("%s: fp == NULL", __func__));
1767 filecaps_validate(fcaps, __func__);
1769 FILEDESC_XLOCK(fdp);
1770 if ((error = fdalloc(td, 0, fd))) {
1771 FILEDESC_XUNLOCK(fdp);
1775 fde = &fdp->fd_ofiles[*fd];
1777 seq_write_begin(&fde->fde_seq);
1780 if ((flags & O_CLOEXEC) != 0)
1781 fde->fde_flags |= UF_EXCLOSE;
1783 filecaps_move(fcaps, &fde->fde_caps);
1785 filecaps_fill(&fde->fde_caps);
1787 seq_write_end(&fde->fde_seq);
1789 FILEDESC_XUNLOCK(fdp);
1794 * Build a new filedesc structure from another.
1795 * Copy the current, root, and jail root vnode references.
1797 * If fdp is not NULL, return with it shared locked.
1800 fdinit(struct filedesc *fdp, bool prepfiles)
1802 struct filedesc0 *newfdp0;
1803 struct filedesc *newfdp;
1805 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1806 newfdp = &newfdp0->fd_fd;
1808 /* Create the file descriptor table. */
1809 FILEDESC_LOCK_INIT(newfdp);
1810 newfdp->fd_refcnt = 1;
1811 newfdp->fd_holdcnt = 1;
1812 newfdp->fd_cmask = CMASK;
1813 newfdp->fd_map = newfdp0->fd_dmap;
1814 newfdp->fd_lastfile = -1;
1815 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1816 newfdp->fd_files->fdt_nfiles = NDFILE;
1821 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1822 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1824 FILEDESC_SLOCK(fdp);
1825 newfdp->fd_cdir = fdp->fd_cdir;
1826 if (newfdp->fd_cdir)
1827 VREF(newfdp->fd_cdir);
1828 newfdp->fd_rdir = fdp->fd_rdir;
1829 if (newfdp->fd_rdir)
1830 VREF(newfdp->fd_rdir);
1831 newfdp->fd_jdir = fdp->fd_jdir;
1832 if (newfdp->fd_jdir)
1833 VREF(newfdp->fd_jdir);
1836 FILEDESC_SUNLOCK(fdp);
1838 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1839 FILEDESC_SUNLOCK(fdp);
1840 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1841 FILEDESC_SLOCK(fdp);
1848 static struct filedesc *
1849 fdhold(struct proc *p)
1851 struct filedesc *fdp;
1853 mtx_lock(&fdesc_mtx);
1857 mtx_unlock(&fdesc_mtx);
1862 fddrop(struct filedesc *fdp)
1866 if (fdp->fd_holdcnt > 1) {
1867 mtx_lock(&fdesc_mtx);
1868 i = --fdp->fd_holdcnt;
1869 mtx_unlock(&fdesc_mtx);
1874 FILEDESC_LOCK_DESTROY(fdp);
1875 uma_zfree(filedesc0_zone, fdp);
1879 * Share a filedesc structure.
1882 fdshare(struct filedesc *fdp)
1885 FILEDESC_XLOCK(fdp);
1887 FILEDESC_XUNLOCK(fdp);
1892 * Unshare a filedesc structure, if necessary by making a copy
1895 fdunshare(struct thread *td)
1897 struct filedesc *tmp;
1898 struct proc *p = td->td_proc;
1900 if (p->p_fd->fd_refcnt == 1)
1903 tmp = fdcopy(p->p_fd);
1909 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
1910 * this is to ease callers, not catch errors.
1913 fdcopy(struct filedesc *fdp)
1915 struct filedesc *newfdp;
1916 struct filedescent *nfde, *ofde;
1921 newfdp = fdinit(fdp, true);
1922 /* copy all passable descriptors (i.e. not kqueue) */
1923 newfdp->fd_freefile = -1;
1924 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1925 ofde = &fdp->fd_ofiles[i];
1926 if (ofde->fde_file == NULL ||
1927 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
1928 if (newfdp->fd_freefile == -1)
1929 newfdp->fd_freefile = i;
1932 nfde = &newfdp->fd_ofiles[i];
1934 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps);
1935 fhold(nfde->fde_file);
1936 fdused_init(newfdp, i);
1937 newfdp->fd_lastfile = i;
1939 if (newfdp->fd_freefile == -1)
1940 newfdp->fd_freefile = i;
1941 newfdp->fd_cmask = fdp->fd_cmask;
1942 FILEDESC_SUNLOCK(fdp);
1947 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
1948 * one of processes using it exits) and the table used to be shared.
1951 fdclearlocks(struct thread *td)
1953 struct filedesc *fdp;
1954 struct filedesc_to_leader *fdtol;
1964 MPASS(fdtol != NULL);
1966 FILEDESC_XLOCK(fdp);
1967 KASSERT(fdtol->fdl_refcount > 0,
1968 ("filedesc_to_refcount botch: fdl_refcount=%d",
1969 fdtol->fdl_refcount));
1970 if (fdtol->fdl_refcount == 1 &&
1971 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1972 for (i = 0; i <= fdp->fd_lastfile; i++) {
1973 fp = fdp->fd_ofiles[i].fde_file;
1974 if (fp == NULL || fp->f_type != DTYPE_VNODE)
1977 FILEDESC_XUNLOCK(fdp);
1978 lf.l_whence = SEEK_SET;
1981 lf.l_type = F_UNLCK;
1983 (void) VOP_ADVLOCK(vp,
1984 (caddr_t)p->p_leader, F_UNLCK,
1986 FILEDESC_XLOCK(fdp);
1991 if (fdtol->fdl_refcount == 1) {
1992 if (fdp->fd_holdleaderscount > 0 &&
1993 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1995 * close() or do_dup() has cleared a reference
1996 * in a shared file descriptor table.
1998 fdp->fd_holdleaderswakeup = 1;
1999 sx_sleep(&fdp->fd_holdleaderscount,
2000 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2003 if (fdtol->fdl_holdcount > 0) {
2005 * Ensure that fdtol->fdl_leader remains
2006 * valid in closef().
2008 fdtol->fdl_wakeup = 1;
2009 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2014 fdtol->fdl_refcount--;
2015 if (fdtol->fdl_refcount == 0 &&
2016 fdtol->fdl_holdcount == 0) {
2017 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2018 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2022 FILEDESC_XUNLOCK(fdp);
2024 free(fdtol, M_FILEDESC_TO_LEADER);
2028 * Release a filedesc structure.
2031 fdescfree(struct thread *td)
2033 struct filedesc0 *fdp0;
2034 struct filedesc *fdp;
2035 struct freetable *ft, *tft;
2036 struct filedescent *fde;
2038 struct vnode *cdir, *jdir, *rdir;
2041 fdp = td->td_proc->p_fd;
2045 PROC_LOCK(td->td_proc);
2046 racct_set(td->td_proc, RACCT_NOFILE, 0);
2047 PROC_UNLOCK(td->td_proc);
2050 if (td->td_proc->p_fdtol != NULL)
2053 mtx_lock(&fdesc_mtx);
2054 td->td_proc->p_fd = NULL;
2055 mtx_unlock(&fdesc_mtx);
2057 FILEDESC_XLOCK(fdp);
2058 i = --fdp->fd_refcnt;
2060 FILEDESC_XUNLOCK(fdp);
2064 cdir = fdp->fd_cdir;
2065 fdp->fd_cdir = NULL;
2066 rdir = fdp->fd_rdir;
2067 fdp->fd_rdir = NULL;
2068 jdir = fdp->fd_jdir;
2069 fdp->fd_jdir = NULL;
2070 FILEDESC_XUNLOCK(fdp);
2072 for (i = 0; i <= fdp->fd_lastfile; i++) {
2073 fde = &fdp->fd_ofiles[i];
2077 (void) closef(fp, td);
2081 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2082 free(fdp->fd_map, M_FILEDESC);
2083 if (fdp->fd_nfiles > NDFILE)
2084 free(fdp->fd_files, M_FILEDESC);
2086 fdp0 = (struct filedesc0 *)fdp;
2087 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2088 free(ft->ft_table, M_FILEDESC);
2101 * For setugid programs, we don't want to people to use that setugidness
2102 * to generate error messages which write to a file which otherwise would
2103 * otherwise be off-limits to the process. We check for filesystems where
2104 * the vnode can change out from under us after execve (like [lin]procfs).
2106 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2107 * sufficient. We also don't check for setugidness since we know we are.
2110 is_unsafe(struct file *fp)
2114 if (fp->f_type != DTYPE_VNODE)
2118 return ((vp->v_vflag & VV_PROCDEP) != 0);
2122 * Make this setguid thing safe, if at all possible.
2125 fdsetugidsafety(struct thread *td)
2127 struct filedesc *fdp;
2131 fdp = td->td_proc->p_fd;
2132 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2133 MPASS(fdp->fd_nfiles >= 3);
2134 for (i = 0; i <= 2; i++) {
2135 fp = fdp->fd_ofiles[i].fde_file;
2136 if (fp != NULL && is_unsafe(fp)) {
2137 FILEDESC_XLOCK(fdp);
2138 knote_fdclose(td, i);
2140 * NULL-out descriptor prior to close to avoid
2141 * a race while close blocks.
2144 FILEDESC_XUNLOCK(fdp);
2145 (void) closef(fp, td);
2151 * If a specific file object occupies a specific file descriptor, close the
2152 * file descriptor entry and drop a reference on the file object. This is a
2153 * convenience function to handle a subsequent error in a function that calls
2154 * falloc() that handles the race that another thread might have closed the
2155 * file descriptor out from under the thread creating the file object.
2158 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
2161 FILEDESC_XLOCK(fdp);
2162 if (fdp->fd_ofiles[idx].fde_file == fp) {
2164 FILEDESC_XUNLOCK(fdp);
2167 FILEDESC_XUNLOCK(fdp);
2171 * Close any files on exec?
2174 fdcloseexec(struct thread *td)
2176 struct filedesc *fdp;
2177 struct filedescent *fde;
2181 fdp = td->td_proc->p_fd;
2182 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2183 for (i = 0; i <= fdp->fd_lastfile; i++) {
2184 fde = &fdp->fd_ofiles[i];
2186 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2187 (fde->fde_flags & UF_EXCLOSE))) {
2188 FILEDESC_XLOCK(fdp);
2190 (void) closefp(fdp, i, fp, td, 0);
2191 /* closefp() drops the FILEDESC lock. */
2197 * It is unsafe for set[ug]id processes to be started with file
2198 * descriptors 0..2 closed, as these descriptors are given implicit
2199 * significance in the Standard C library. fdcheckstd() will create a
2200 * descriptor referencing /dev/null for each of stdin, stdout, and
2201 * stderr that is not already open.
2204 fdcheckstd(struct thread *td)
2206 struct filedesc *fdp;
2208 int i, error, devnull;
2210 fdp = td->td_proc->p_fd;
2211 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2212 MPASS(fdp->fd_nfiles >= 3);
2214 for (i = 0; i <= 2; i++) {
2215 if (fdp->fd_ofiles[i].fde_file != NULL)
2218 save = td->td_retval[0];
2219 if (devnull != -1) {
2220 error = do_dup(td, DUP_FIXED, devnull, i);
2222 error = kern_openat(td, AT_FDCWD, "/dev/null",
2223 UIO_SYSSPACE, O_RDWR, 0);
2225 devnull = td->td_retval[0];
2226 KASSERT(devnull == i, ("we didn't get our fd"));
2229 td->td_retval[0] = save;
2237 * Internal form of close. Decrement reference count on file structure.
2238 * Note: td may be NULL when closing a file that was being passed in a
2241 * XXXRW: Giant is not required for the caller, but often will be held; this
2242 * makes it moderately likely the Giant will be recursed in the VFS case.
2245 closef(struct file *fp, struct thread *td)
2249 struct filedesc_to_leader *fdtol;
2250 struct filedesc *fdp;
2253 * POSIX record locking dictates that any close releases ALL
2254 * locks owned by this process. This is handled by setting
2255 * a flag in the unlock to free ONLY locks obeying POSIX
2256 * semantics, and not to free BSD-style file locks.
2257 * If the descriptor was in a message, POSIX-style locks
2258 * aren't passed with the descriptor, and the thread pointer
2259 * will be NULL. Callers should be careful only to pass a
2260 * NULL thread pointer when there really is no owning
2261 * context that might have locks, or the locks will be
2264 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2266 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2267 lf.l_whence = SEEK_SET;
2270 lf.l_type = F_UNLCK;
2271 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2272 F_UNLCK, &lf, F_POSIX);
2274 fdtol = td->td_proc->p_fdtol;
2275 if (fdtol != NULL) {
2277 * Handle special case where file descriptor table is
2278 * shared between multiple process leaders.
2280 fdp = td->td_proc->p_fd;
2281 FILEDESC_XLOCK(fdp);
2282 for (fdtol = fdtol->fdl_next;
2283 fdtol != td->td_proc->p_fdtol;
2284 fdtol = fdtol->fdl_next) {
2285 if ((fdtol->fdl_leader->p_flag &
2288 fdtol->fdl_holdcount++;
2289 FILEDESC_XUNLOCK(fdp);
2290 lf.l_whence = SEEK_SET;
2293 lf.l_type = F_UNLCK;
2295 (void) VOP_ADVLOCK(vp,
2296 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2298 FILEDESC_XLOCK(fdp);
2299 fdtol->fdl_holdcount--;
2300 if (fdtol->fdl_holdcount == 0 &&
2301 fdtol->fdl_wakeup != 0) {
2302 fdtol->fdl_wakeup = 0;
2306 FILEDESC_XUNLOCK(fdp);
2309 return (fdrop(fp, td));
2313 * Initialize the file pointer with the specified properties.
2315 * The ops are set with release semantics to be certain that the flags, type,
2316 * and data are visible when ops is. This is to prevent ops methods from being
2317 * called with bad data.
2320 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2325 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2329 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2330 struct file **fpp, seq_t *seqp)
2333 struct filedescent *fde;
2335 struct fdescenttbl *fdt;
2340 cap_rights_t *haverights;
2344 fdt = fdp->fd_files;
2345 if (fd < 0 || fd >= fdt->fdt_nfiles)
2348 * Fetch the descriptor locklessly. We avoid fdrop() races by
2349 * never raising a refcount above 0. To accomplish this we have
2350 * to use a cmpset loop rather than an atomic_add. The descriptor
2351 * must be re-verified once we acquire a reference to be certain
2352 * that the identity is still correct and we did not lose a race
2353 * due to preemption.
2357 seq = seq_read(fd_seq(fdt, fd));
2358 fde = &fdt->fdt_ofiles[fd];
2359 haverights = cap_rights_fde(fde);
2361 if (!seq_consistent(fd_seq(fdt, fd), seq)) {
2366 fp = fdt->fdt_ofiles[fd].fde_file;
2371 if (needrightsp != NULL) {
2372 error = cap_check(haverights, needrightsp);
2378 count = fp->f_count;
2381 * Force a reload. Other thread could reallocate the
2382 * table before this fd was closed, so it possible that
2383 * there is a stale fp pointer in cached version.
2385 fdt = *(struct fdescenttbl * volatile *)&(fdp->fd_files);
2389 * Use an acquire barrier to force re-reading of fdt so it is
2390 * refreshed for verification.
2392 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) == 0)
2394 fdt = fdp->fd_files;
2396 if (seq_consistent_nomb(fd_seq(fdt, fd), seq))
2398 if (fp == fdt->fdt_ofiles[fd].fde_file)
2401 fdrop(fp, curthread);
2413 * Extract the file pointer associated with the specified descriptor for the
2414 * current user process.
2416 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2419 * File's rights will be checked against the capability rights mask.
2421 * If an error occured the non-zero error is returned and *fpp is set to
2422 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
2423 * responsible for fdrop().
2426 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2427 cap_rights_t *needrightsp, seq_t *seqp)
2429 struct filedesc *fdp;
2431 cap_rights_t needrights;
2435 fdp = td->td_proc->p_fd;
2436 if (needrightsp != NULL)
2437 needrights = *needrightsp;
2439 cap_rights_init(&needrights);
2440 error = fget_unlocked(fdp, fd, &needrights, &fp, seqp);
2443 if (fp->f_ops == &badfileops) {
2449 * FREAD and FWRITE failure return EBADF as per POSIX.
2455 if ((fp->f_flag & flags) == 0)
2459 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2460 ((fp->f_flag & FWRITE) != 0))
2466 KASSERT(0, ("wrong flags"));
2479 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2482 return (_fget(td, fd, fpp, 0, rightsp, NULL));
2486 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2490 #ifndef CAPABILITIES
2491 error = _fget(td, fd, fpp, 0, rightsp, NULL);
2492 if (maxprotp != NULL)
2493 *maxprotp = VM_PROT_ALL;
2495 struct filedesc *fdp = td->td_proc->p_fd;
2498 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2500 error = _fget(td, fd, fpp, 0, rightsp, &seq);
2504 * If requested, convert capability rights to access flags.
2506 if (maxprotp != NULL)
2507 *maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd));
2508 if (!fd_modified(td->td_proc->p_fd, fd, seq))
2517 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2520 return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2524 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2527 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2531 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2534 struct filedesc *fdp = td->td_proc->p_fd;
2535 #ifndef CAPABILITIES
2536 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2541 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2543 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2546 error = cap_fcntl_check(fdp, fd, needfcntl);
2547 if (!fd_modified(fdp, fd, seq))
2560 * Like fget() but loads the underlying vnode, or returns an error if the
2561 * descriptor does not represent a vnode. Note that pipes use vnodes but
2562 * never have VM objects. The returned vnode will be vref()'d.
2564 * XXX: what about the unused flags ?
2567 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2574 error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2577 if (fp->f_vnode == NULL) {
2589 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2592 return (_fgetvp(td, fd, 0, rightsp, vpp));
2596 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2597 struct filecaps *havecaps, struct vnode **vpp)
2599 struct filedesc *fdp;
2605 fdp = td->td_proc->p_fd;
2606 fp = fget_locked(fdp, fd);
2607 if (fp == NULL || fp->f_ops == &badfileops)
2611 if (needrightsp != NULL) {
2612 error = cap_check(cap_rights(fdp, fd), needrightsp);
2618 if (fp->f_vnode == NULL)
2623 filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps);
2629 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2632 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2636 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2639 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2644 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2648 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2653 * Like fget() but loads the underlying socket, or returns an error if the
2654 * descriptor does not represent a socket.
2656 * We bump the ref count on the returned socket. XXX Also obtain the SX lock
2659 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely
2660 * on their file descriptor reference to prevent the socket from being free'd
2664 fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp,
2673 if ((error = _fget(td, fd, &fp, 0, rightsp, NULL)) != 0)
2675 if (fp->f_type != DTYPE_SOCKET) {
2680 *fflagp = fp->f_flag;
2691 * Drop the reference count on the socket and XXX release the SX lock in the
2692 * future. The last reference closes the socket.
2694 * Note: fputsock() is deprecated, see comment for fgetsock().
2697 fputsock(struct socket *so)
2702 CURVNET_SET(so->so_vnet);
2708 * Handle the last reference to a file being closed.
2711 _fdrop(struct file *fp, struct thread *td)
2715 if (fp->f_count != 0)
2716 panic("fdrop: count %d", fp->f_count);
2717 error = fo_close(fp, td);
2718 atomic_subtract_int(&openfiles, 1);
2720 free(fp->f_advice, M_FADVISE);
2721 uma_zfree(file_zone, fp);
2727 * Apply an advisory lock on a file descriptor.
2729 * Just attempt to get a record lock of the requested type on the entire file
2730 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2732 #ifndef _SYS_SYSPROTO_H_
2740 sys_flock(struct thread *td, struct flock_args *uap)
2745 cap_rights_t rights;
2748 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp);
2751 if (fp->f_type != DTYPE_VNODE) {
2753 return (EOPNOTSUPP);
2757 lf.l_whence = SEEK_SET;
2760 if (uap->how & LOCK_UN) {
2761 lf.l_type = F_UNLCK;
2762 atomic_clear_int(&fp->f_flag, FHASLOCK);
2763 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2766 if (uap->how & LOCK_EX)
2767 lf.l_type = F_WRLCK;
2768 else if (uap->how & LOCK_SH)
2769 lf.l_type = F_RDLCK;
2774 atomic_set_int(&fp->f_flag, FHASLOCK);
2775 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2776 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2782 * Duplicate the specified descriptor to a free descriptor.
2785 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
2786 int openerror, int *indxp)
2788 struct filedescent *newfde, *oldfde;
2792 KASSERT(openerror == ENODEV || openerror == ENXIO,
2793 ("unexpected error %d in %s", openerror, __func__));
2796 * If the to-be-dup'd fd number is greater than the allowed number
2797 * of file descriptors, or the fd to be dup'd has already been
2798 * closed, then reject.
2800 FILEDESC_XLOCK(fdp);
2801 if ((fp = fget_locked(fdp, dfd)) == NULL) {
2802 FILEDESC_XUNLOCK(fdp);
2806 error = fdalloc(td, 0, &indx);
2808 FILEDESC_XUNLOCK(fdp);
2813 * There are two cases of interest here.
2815 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2817 * For ENXIO steal away the file structure from (dfd) and store it in
2818 * (indx). (dfd) is effectively closed by this operation.
2820 switch (openerror) {
2823 * Check that the mode the file is being opened for is a
2824 * subset of the mode of the existing descriptor.
2826 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
2827 fdunused(fdp, indx);
2828 FILEDESC_XUNLOCK(fdp);
2832 newfde = &fdp->fd_ofiles[indx];
2833 oldfde = &fdp->fd_ofiles[dfd];
2835 seq_write_begin(&newfde->fde_seq);
2837 memcpy(newfde, oldfde, fde_change_size);
2838 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps);
2840 seq_write_end(&newfde->fde_seq);
2845 * Steal away the file pointer from dfd and stuff it into indx.
2847 newfde = &fdp->fd_ofiles[indx];
2848 oldfde = &fdp->fd_ofiles[dfd];
2850 seq_write_begin(&newfde->fde_seq);
2852 memcpy(newfde, oldfde, fde_change_size);
2853 bzero(oldfde, fde_change_size);
2856 seq_write_end(&newfde->fde_seq);
2860 FILEDESC_XUNLOCK(fdp);
2866 * Scan all active processes and prisons to see if any of them have a current
2867 * or root directory of `olddp'. If so, replace them with the new mount point.
2870 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2872 struct filedesc *fdp;
2877 if (vrefcnt(olddp) == 1)
2880 sx_slock(&allproc_lock);
2881 FOREACH_PROC_IN_SYSTEM(p) {
2885 FILEDESC_XLOCK(fdp);
2886 if (fdp->fd_cdir == olddp) {
2888 fdp->fd_cdir = newdp;
2891 if (fdp->fd_rdir == olddp) {
2893 fdp->fd_rdir = newdp;
2896 if (fdp->fd_jdir == olddp) {
2898 fdp->fd_jdir = newdp;
2901 FILEDESC_XUNLOCK(fdp);
2904 sx_sunlock(&allproc_lock);
2905 if (rootvnode == olddp) {
2910 mtx_lock(&prison0.pr_mtx);
2911 if (prison0.pr_root == olddp) {
2913 prison0.pr_root = newdp;
2916 mtx_unlock(&prison0.pr_mtx);
2917 sx_slock(&allprison_lock);
2918 TAILQ_FOREACH(pr, &allprison, pr_list) {
2919 mtx_lock(&pr->pr_mtx);
2920 if (pr->pr_root == olddp) {
2922 pr->pr_root = newdp;
2925 mtx_unlock(&pr->pr_mtx);
2927 sx_sunlock(&allprison_lock);
2932 struct filedesc_to_leader *
2933 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2935 struct filedesc_to_leader *fdtol;
2937 fdtol = malloc(sizeof(struct filedesc_to_leader),
2938 M_FILEDESC_TO_LEADER,
2940 fdtol->fdl_refcount = 1;
2941 fdtol->fdl_holdcount = 0;
2942 fdtol->fdl_wakeup = 0;
2943 fdtol->fdl_leader = leader;
2945 FILEDESC_XLOCK(fdp);
2946 fdtol->fdl_next = old->fdl_next;
2947 fdtol->fdl_prev = old;
2948 old->fdl_next = fdtol;
2949 fdtol->fdl_next->fdl_prev = fdtol;
2950 FILEDESC_XUNLOCK(fdp);
2952 fdtol->fdl_next = fdtol;
2953 fdtol->fdl_prev = fdtol;
2959 * Get file structures globally.
2962 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2965 struct filedesc *fdp;
2970 error = sysctl_wire_old_buffer(req, 0);
2973 if (req->oldptr == NULL) {
2975 sx_slock(&allproc_lock);
2976 FOREACH_PROC_IN_SYSTEM(p) {
2977 if (p->p_state == PRS_NEW)
2982 /* overestimates sparse tables. */
2983 if (fdp->fd_lastfile > 0)
2984 n += fdp->fd_lastfile;
2987 sx_sunlock(&allproc_lock);
2988 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2991 bzero(&xf, sizeof(xf));
2992 xf.xf_size = sizeof(xf);
2993 sx_slock(&allproc_lock);
2994 FOREACH_PROC_IN_SYSTEM(p) {
2996 if (p->p_state == PRS_NEW) {
3000 if (p_cansee(req->td, p) != 0) {
3004 xf.xf_pid = p->p_pid;
3005 xf.xf_uid = p->p_ucred->cr_uid;
3010 FILEDESC_SLOCK(fdp);
3011 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3012 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3016 xf.xf_data = fp->f_data;
3017 xf.xf_vnode = fp->f_vnode;
3018 xf.xf_type = fp->f_type;
3019 xf.xf_count = fp->f_count;
3021 xf.xf_offset = foffset_get(fp);
3022 xf.xf_flag = fp->f_flag;
3023 error = SYSCTL_OUT(req, &xf, sizeof(xf));
3027 FILEDESC_SUNLOCK(fdp);
3032 sx_sunlock(&allproc_lock);
3036 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3037 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3039 #ifdef KINFO_FILE_SIZE
3040 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3044 xlate_fflags(int fflags)
3046 static const struct {
3049 } fflags_table[] = {
3050 { FAPPEND, KF_FLAG_APPEND },
3051 { FASYNC, KF_FLAG_ASYNC },
3052 { FFSYNC, KF_FLAG_FSYNC },
3053 { FHASLOCK, KF_FLAG_HASLOCK },
3054 { FNONBLOCK, KF_FLAG_NONBLOCK },
3055 { FREAD, KF_FLAG_READ },
3056 { FWRITE, KF_FLAG_WRITE },
3057 { O_CREAT, KF_FLAG_CREAT },
3058 { O_DIRECT, KF_FLAG_DIRECT },
3059 { O_EXCL, KF_FLAG_EXCL },
3060 { O_EXEC, KF_FLAG_EXEC },
3061 { O_EXLOCK, KF_FLAG_EXLOCK },
3062 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3063 { O_SHLOCK, KF_FLAG_SHLOCK },
3064 { O_TRUNC, KF_FLAG_TRUNC }
3070 for (i = 0; i < nitems(fflags_table); i++)
3071 if (fflags & fflags_table[i].fflag)
3072 kflags |= fflags_table[i].kf_fflag;
3076 /* Trim unused data from kf_path by truncating the structure size. */
3078 pack_kinfo(struct kinfo_file *kif)
3081 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3082 strlen(kif->kf_path) + 1;
3083 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3087 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3088 struct kinfo_file *kif, struct filedesc *fdp)
3092 bzero(kif, sizeof(*kif));
3094 /* Set a default type to allow for empty fill_kinfo() methods. */
3095 kif->kf_type = KF_TYPE_UNKNOWN;
3096 kif->kf_flags = xlate_fflags(fp->f_flag);
3097 if (rightsp != NULL)
3098 kif->kf_cap_rights = *rightsp;
3100 cap_rights_init(&kif->kf_cap_rights);
3102 kif->kf_ref_count = fp->f_count;
3103 kif->kf_offset = foffset_get(fp);
3106 * This may drop the filedesc lock, so the 'fp' cannot be
3107 * accessed after this call.
3109 error = fo_fill_kinfo(fp, kif, fdp);
3111 kif->kf_status |= KF_ATTR_VALID;
3116 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3117 struct kinfo_file *kif)
3121 bzero(kif, sizeof(*kif));
3123 kif->kf_type = KF_TYPE_VNODE;
3124 error = vn_fill_kinfo_vnode(vp, kif);
3126 kif->kf_status |= KF_ATTR_VALID;
3127 kif->kf_flags = xlate_fflags(fflags);
3128 cap_rights_init(&kif->kf_cap_rights);
3130 kif->kf_ref_count = -1;
3131 kif->kf_offset = -1;
3136 struct export_fd_buf {
3137 struct filedesc *fdp;
3140 struct kinfo_file kif;
3144 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3146 struct kinfo_file *kif;
3149 if (efbuf->remainder != -1) {
3150 if (efbuf->remainder < kif->kf_structsize) {
3151 /* Terminate export. */
3152 efbuf->remainder = 0;
3155 efbuf->remainder -= kif->kf_structsize;
3157 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3161 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3162 struct export_fd_buf *efbuf)
3166 if (efbuf->remainder == 0)
3168 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp);
3169 FILEDESC_SUNLOCK(efbuf->fdp);
3170 error = export_kinfo_to_sb(efbuf);
3171 FILEDESC_SLOCK(efbuf->fdp);
3176 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3177 struct export_fd_buf *efbuf)
3181 if (efbuf->remainder == 0)
3183 if (efbuf->fdp != NULL)
3184 FILEDESC_SUNLOCK(efbuf->fdp);
3185 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif);
3186 error = export_kinfo_to_sb(efbuf);
3187 if (efbuf->fdp != NULL)
3188 FILEDESC_SLOCK(efbuf->fdp);
3193 * Store a process file descriptor information to sbuf.
3195 * Takes a locked proc as argument, and returns with the proc unlocked.
3198 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
3201 struct filedesc *fdp;
3202 struct export_fd_buf *efbuf;
3203 struct vnode *cttyvp, *textvp, *tracevp;
3205 cap_rights_t rights;
3207 PROC_LOCK_ASSERT(p, MA_OWNED);
3210 tracevp = p->p_tracevp;
3211 if (tracevp != NULL)
3214 textvp = p->p_textvp;
3217 /* Controlling tty. */
3219 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3220 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3226 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3229 efbuf->remainder = maxlen;
3230 if (tracevp != NULL)
3231 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3234 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3236 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3242 FILEDESC_SLOCK(fdp);
3243 /* working directory */
3244 if (fdp->fd_cdir != NULL) {
3246 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3248 /* root directory */
3249 if (fdp->fd_rdir != NULL) {
3251 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3253 /* jail directory */
3254 if (fdp->fd_jdir != NULL) {
3256 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3258 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3259 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3262 rights = *cap_rights(fdp, i);
3263 #else /* !CAPABILITIES */
3264 cap_rights_init(&rights);
3267 * Create sysctl entry. It is OK to drop the filedesc
3268 * lock inside of export_file_to_sb() as we will
3269 * re-validate and re-evaluate its properties when the
3272 error = export_file_to_sb(fp, i, &rights, efbuf);
3273 if (error != 0 || efbuf->remainder == 0)
3276 FILEDESC_SUNLOCK(fdp);
3279 free(efbuf, M_TEMP);
3283 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
3286 * Get per-process file descriptors for use by procstat(1), et al.
3289 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3294 int error, error2, *name;
3298 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3299 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3304 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3305 error = kern_proc_filedesc_out(p, &sb, maxlen);
3306 error2 = sbuf_finish(&sb);
3308 return (error != 0 ? error : error2);
3311 #ifdef KINFO_OFILE_SIZE
3312 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3315 #ifdef COMPAT_FREEBSD7
3317 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3320 okif->kf_structsize = sizeof(*okif);
3321 okif->kf_type = kif->kf_type;
3322 okif->kf_fd = kif->kf_fd;
3323 okif->kf_ref_count = kif->kf_ref_count;
3324 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3325 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3326 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3327 okif->kf_offset = kif->kf_offset;
3328 okif->kf_vnode_type = kif->kf_vnode_type;
3329 okif->kf_sock_domain = kif->kf_sock_domain;
3330 okif->kf_sock_type = kif->kf_sock_type;
3331 okif->kf_sock_protocol = kif->kf_sock_protocol;
3332 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3333 okif->kf_sa_local = kif->kf_sa_local;
3334 okif->kf_sa_peer = kif->kf_sa_peer;
3338 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3339 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3344 FILEDESC_SUNLOCK(fdp);
3345 export_vnode_to_kinfo(vp, type, 0, kif);
3346 kinfo_to_okinfo(kif, okif);
3347 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3348 FILEDESC_SLOCK(fdp);
3353 * Get per-process file descriptors for use by procstat(1), et al.
3356 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3358 struct kinfo_ofile *okif;
3359 struct kinfo_file *kif;
3360 struct filedesc *fdp;
3361 int error, i, *name;
3366 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3373 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3374 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3375 FILEDESC_SLOCK(fdp);
3376 if (fdp->fd_cdir != NULL)
3377 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3379 if (fdp->fd_rdir != NULL)
3380 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3382 if (fdp->fd_jdir != NULL)
3383 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3385 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3386 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3388 export_file_to_kinfo(fp, i, NULL, kif, fdp);
3389 FILEDESC_SUNLOCK(fdp);
3390 kinfo_to_okinfo(kif, okif);
3391 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3392 FILEDESC_SLOCK(fdp);
3396 FILEDESC_SUNLOCK(fdp);
3403 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3404 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3405 "Process ofiledesc entries");
3406 #endif /* COMPAT_FREEBSD7 */
3409 vntype_to_kinfo(int vtype)
3414 } vtypes_table[] = {
3415 { VBAD, KF_VTYPE_VBAD },
3416 { VBLK, KF_VTYPE_VBLK },
3417 { VCHR, KF_VTYPE_VCHR },
3418 { VDIR, KF_VTYPE_VDIR },
3419 { VFIFO, KF_VTYPE_VFIFO },
3420 { VLNK, KF_VTYPE_VLNK },
3421 { VNON, KF_VTYPE_VNON },
3422 { VREG, KF_VTYPE_VREG },
3423 { VSOCK, KF_VTYPE_VSOCK }
3428 * Perform vtype translation.
3430 for (i = 0; i < nitems(vtypes_table); i++)
3431 if (vtypes_table[i].vtype == vtype)
3432 return (vtypes_table[i].kf_vtype);
3434 return (KF_VTYPE_UNKNOWN);
3437 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3438 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3439 "Process filedesc entries");
3442 * Store a process current working directory information to sbuf.
3444 * Takes a locked proc as argument, and returns with the proc unlocked.
3447 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
3449 struct filedesc *fdp;
3450 struct export_fd_buf *efbuf;
3453 PROC_LOCK_ASSERT(p, MA_OWNED);
3460 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3463 efbuf->remainder = maxlen;
3465 FILEDESC_SLOCK(fdp);
3466 if (fdp->fd_cdir == NULL)
3470 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3473 FILEDESC_SUNLOCK(fdp);
3475 free(efbuf, M_TEMP);
3480 * Get per-process current working directory.
3483 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3488 int error, error2, *name;
3492 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3493 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3498 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3499 error = kern_proc_cwd_out(p, &sb, maxlen);
3500 error2 = sbuf_finish(&sb);
3502 return (error != 0 ? error : error2);
3505 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3506 sysctl_kern_proc_cwd, "Process current working directory");
3510 * For the purposes of debugging, generate a human-readable string for the
3514 file_type_to_name(short type)
3544 * For the purposes of debugging, identify a process (if any, perhaps one of
3545 * many) that references the passed file in its file descriptor array. Return
3548 static struct proc *
3549 file_to_first_proc(struct file *fp)
3551 struct filedesc *fdp;
3555 FOREACH_PROC_IN_SYSTEM(p) {
3556 if (p->p_state == PRS_NEW)
3561 for (n = 0; n <= fdp->fd_lastfile; n++) {
3562 if (fp == fdp->fd_ofiles[n].fde_file)
3570 db_print_file(struct file *fp, int header)
3575 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
3576 "File", "Type", "Data", "Flag", "GCFl", "Count",
3577 "MCount", "Vnode", "FPID", "FCmd");
3578 p = file_to_first_proc(fp);
3579 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
3580 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
3581 0, fp->f_count, 0, fp->f_vnode,
3582 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3585 DB_SHOW_COMMAND(file, db_show_file)
3590 db_printf("usage: show file <addr>\n");
3593 fp = (struct file *)addr;
3594 db_print_file(fp, 1);
3597 DB_SHOW_COMMAND(files, db_show_files)
3599 struct filedesc *fdp;
3606 FOREACH_PROC_IN_SYSTEM(p) {
3607 if (p->p_state == PRS_NEW)
3609 if ((fdp = p->p_fd) == NULL)
3611 for (n = 0; n <= fdp->fd_lastfile; ++n) {
3612 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3614 db_print_file(fp, header);
3621 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3622 &maxfilesperproc, 0, "Maximum files allowed open per process");
3624 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3625 &maxfiles, 0, "Maximum number of files");
3627 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3628 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
3632 filelistinit(void *dummy)
3635 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3636 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
3637 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
3638 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
3639 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3640 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
3642 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3644 /*-------------------------------------------------------------------*/
3647 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
3648 int flags, struct thread *td)
3655 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
3663 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
3671 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
3679 badfo_kqfilter(struct file *fp, struct knote *kn)
3686 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
3694 badfo_close(struct file *fp, struct thread *td)
3701 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
3709 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
3717 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
3718 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
3719 int kflags, struct thread *td)
3726 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
3732 struct fileops badfileops = {
3733 .fo_read = badfo_readwrite,
3734 .fo_write = badfo_readwrite,
3735 .fo_truncate = badfo_truncate,
3736 .fo_ioctl = badfo_ioctl,
3737 .fo_poll = badfo_poll,
3738 .fo_kqfilter = badfo_kqfilter,
3739 .fo_stat = badfo_stat,
3740 .fo_close = badfo_close,
3741 .fo_chmod = badfo_chmod,
3742 .fo_chown = badfo_chown,
3743 .fo_sendfile = badfo_sendfile,
3744 .fo_fill_kinfo = badfo_fill_kinfo,
3748 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
3749 int flags, struct thread *td)
3752 return (EOPNOTSUPP);
3756 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
3764 invfo_ioctl(struct file *fp, u_long com, void *data,
3765 struct ucred *active_cred, struct thread *td)
3772 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
3776 return (poll_no_poll(events));
3780 invfo_kqfilter(struct file *fp, struct knote *kn)
3787 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
3795 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
3803 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
3804 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
3805 int kflags, struct thread *td)
3811 /*-------------------------------------------------------------------*/
3814 * File Descriptor pseudo-device driver (/dev/fd/).
3816 * Opening minor device N dup()s the file (if any) connected to file
3817 * descriptor N belonging to the calling process. Note that this driver
3818 * consists of only the ``open()'' routine, because all subsequent
3819 * references to this file will be direct to the other driver.
3821 * XXX: we could give this one a cloning event handler if necessary.
3826 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
3830 * XXX Kludge: set curthread->td_dupfd to contain the value of the
3831 * the file descriptor being sought for duplication. The error
3832 * return ensures that the vnode for this device will be released
3833 * by vn_open. Open will detect this special error and take the
3834 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
3835 * will simply report the error.
3837 td->td_dupfd = dev2unit(dev);
3841 static struct cdevsw fildesc_cdevsw = {
3842 .d_version = D_VERSION,
3848 fildesc_drvinit(void *unused)
3852 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
3853 UID_ROOT, GID_WHEEL, 0666, "fd/0");
3854 make_dev_alias(dev, "stdin");
3855 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
3856 UID_ROOT, GID_WHEEL, 0666, "fd/1");
3857 make_dev_alias(dev, "stdout");
3858 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
3859 UID_ROOT, GID_WHEEL, 0666, "fd/2");
3860 make_dev_alias(dev, "stderr");
3863 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);