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>
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 __read_mostly uma_zone_t file_zone;
101 static __read_mostly 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 fd_first_free(struct filedesc *fdp, int low, int size);
106 static int fd_last_used(struct filedesc *fdp, int size);
107 static void fdgrowtable(struct filedesc *fdp, int nfd);
108 static void fdgrowtable_exp(struct filedesc *fdp, int nfd);
109 static void fdunused(struct filedesc *fdp, int fd);
110 static void fdused(struct filedesc *fdp, int fd);
111 static int getmaxfd(struct thread *td);
116 * - An array of open file descriptors (fd_ofiles)
117 * - An array of file flags (fd_ofileflags)
118 * - A bitmap recording which descriptors are in use (fd_map)
120 * A process starts out with NDFILE descriptors. The value of NDFILE has
121 * been selected based the historical limit of 20 open files, and an
122 * assumption that the majority of processes, especially short-lived
123 * processes like shells, will never need more.
125 * If this initial allocation is exhausted, a larger descriptor table and
126 * map are allocated dynamically, and the pointers in the process's struct
127 * filedesc are updated to point to those. This is repeated every time
128 * the process runs out of file descriptors (provided it hasn't hit its
131 * Since threads may hold references to individual descriptor table
132 * entries, the tables are never freed. Instead, they are placed on a
133 * linked list and freed only when the struct filedesc is released.
136 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
137 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
138 #define NDSLOT(x) ((x) / NDENTRIES)
139 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
140 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
143 * SLIST entry used to keep track of ofiles which must be reclaimed when
147 struct fdescenttbl *ft_table;
148 SLIST_ENTRY(freetable) ft_next;
152 * Initial allocation: a filedesc structure + the head of SLIST used to
153 * keep track of old ofiles + enough space for NDFILE descriptors.
156 struct fdescenttbl0 {
158 struct filedescent fdt_ofiles[NDFILE];
162 struct filedesc fd_fd;
163 SLIST_HEAD(, freetable) fd_free;
164 struct fdescenttbl0 fd_dfiles;
165 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
169 * Descriptor management.
171 volatile int __exclusive_cache_line openfiles; /* actual number of open files */
172 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
173 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
176 * If low >= size, just return low. Otherwise find the first zero bit in the
177 * given bitmap, starting at low and not exceeding size - 1. Return size if
181 fd_first_free(struct filedesc *fdp, int low, int size)
183 NDSLOTTYPE *map = fdp->fd_map;
191 if (low % NDENTRIES) {
192 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
193 if ((mask &= ~map[off]) != 0UL)
194 return (off * NDENTRIES + ffsl(mask) - 1);
197 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
198 if (map[off] != ~0UL)
199 return (off * NDENTRIES + ffsl(~map[off]) - 1);
204 * Find the highest non-zero bit in the given bitmap, starting at 0 and
205 * not exceeding size - 1. Return -1 if not found.
208 fd_last_used(struct filedesc *fdp, int size)
210 NDSLOTTYPE *map = fdp->fd_map;
215 if (size % NDENTRIES) {
216 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
217 if ((mask &= map[off]) != 0)
218 return (off * NDENTRIES + flsl(mask) - 1);
221 for (minoff = NDSLOT(0); off >= minoff; --off)
223 return (off * NDENTRIES + flsl(map[off]) - 1);
228 fdisused(struct filedesc *fdp, int fd)
231 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
232 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
234 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
238 * Mark a file descriptor as used.
241 fdused_init(struct filedesc *fdp, int fd)
244 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
246 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
250 fdused(struct filedesc *fdp, int fd)
253 FILEDESC_XLOCK_ASSERT(fdp);
255 fdused_init(fdp, fd);
256 if (fd > fdp->fd_lastfile)
257 fdp->fd_lastfile = fd;
258 if (fd == fdp->fd_freefile)
259 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
263 * Mark a file descriptor as unused.
266 fdunused(struct filedesc *fdp, int fd)
269 FILEDESC_XLOCK_ASSERT(fdp);
271 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
272 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
273 ("fd=%d is still in use", fd));
275 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
276 if (fd < fdp->fd_freefile)
277 fdp->fd_freefile = fd;
278 if (fd == fdp->fd_lastfile)
279 fdp->fd_lastfile = fd_last_used(fdp, fd);
283 * Free a file descriptor.
285 * Avoid some work if fdp is about to be destroyed.
288 fdefree_last(struct filedescent *fde)
291 filecaps_free(&fde->fde_caps);
295 fdfree(struct filedesc *fdp, int fd)
297 struct filedescent *fde;
299 fde = &fdp->fd_ofiles[fd];
301 seq_write_begin(&fde->fde_seq);
304 fde->fde_file = NULL;
307 seq_write_end(&fde->fde_seq);
312 pwd_ensure_dirs(void)
314 struct filedesc *fdp;
318 if (fdp->fd_cdir == NULL) {
319 fdp->fd_cdir = rootvnode;
322 if (fdp->fd_rdir == NULL) {
323 fdp->fd_rdir = rootvnode;
326 FILEDESC_XUNLOCK(fdp);
330 * System calls on descriptors.
332 #ifndef _SYS_SYSPROTO_H_
333 struct getdtablesize_args {
339 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
346 min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc);
348 PROC_LOCK(td->td_proc);
349 lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
350 PROC_UNLOCK(td->td_proc);
351 if (lim < td->td_retval[0])
352 td->td_retval[0] = lim;
358 * Duplicate a file descriptor to a particular value.
360 * Note: keep in mind that a potential race condition exists when closing
361 * descriptors from a shared descriptor table (via rfork).
363 #ifndef _SYS_SYSPROTO_H_
371 sys_dup2(struct thread *td, struct dup2_args *uap)
374 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
378 * Duplicate a file descriptor.
380 #ifndef _SYS_SYSPROTO_H_
387 sys_dup(struct thread *td, struct dup_args *uap)
390 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
394 * The file control system call.
396 #ifndef _SYS_SYSPROTO_H_
405 sys_fcntl(struct thread *td, struct fcntl_args *uap)
408 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
412 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
426 * Convert old flock structure to new.
428 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
429 fl.l_start = ofl.l_start;
430 fl.l_len = ofl.l_len;
431 fl.l_pid = ofl.l_pid;
432 fl.l_type = ofl.l_type;
433 fl.l_whence = ofl.l_whence;
447 arg1 = (intptr_t)&fl;
453 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
454 arg1 = (intptr_t)&fl;
462 error = kern_fcntl(td, fd, newcmd, arg1);
465 if (cmd == F_OGETLK) {
466 ofl.l_start = fl.l_start;
467 ofl.l_len = fl.l_len;
468 ofl.l_pid = fl.l_pid;
469 ofl.l_type = fl.l_type;
470 ofl.l_whence = fl.l_whence;
471 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
472 } else if (cmd == F_GETLK) {
473 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
479 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
481 struct filedesc *fdp;
483 struct file *fp, *fp2;
484 struct filedescent *fde;
500 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
503 case F_DUPFD_CLOEXEC:
505 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
510 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
513 case F_DUP2FD_CLOEXEC:
515 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
521 fde = fdeget_locked(fdp, fd);
524 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
527 FILEDESC_SUNLOCK(fdp);
533 fde = fdeget_locked(fdp, fd);
535 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
536 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
539 FILEDESC_XUNLOCK(fdp);
543 error = fget_fcntl(td, fd,
544 cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp);
547 td->td_retval[0] = OFLAGS(fp->f_flag);
552 error = fget_fcntl(td, fd,
553 cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp);
557 tmp = flg = fp->f_flag;
559 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
560 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
561 tmp = fp->f_flag & FNONBLOCK;
562 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
567 tmp = fp->f_flag & FASYNC;
568 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
573 atomic_clear_int(&fp->f_flag, FNONBLOCK);
575 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
580 error = fget_fcntl(td, fd,
581 cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp);
584 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
586 td->td_retval[0] = tmp;
591 error = fget_fcntl(td, fd,
592 cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp);
596 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
601 error = priv_check(td, PRIV_NFS_LOCKD);
609 /* FALLTHROUGH F_SETLK */
613 cap_rights_init(&rights, CAP_FLOCK);
614 error = fget_unlocked(fdp, fd, &rights, &fp, NULL);
617 if (fp->f_type != DTYPE_VNODE) {
623 flp = (struct flock *)arg;
624 if (flp->l_whence == SEEK_CUR) {
625 foffset = foffset_get(fp);
628 foffset > OFF_MAX - flp->l_start)) {
633 flp->l_start += foffset;
637 switch (flp->l_type) {
639 if ((fp->f_flag & FREAD) == 0) {
643 PROC_LOCK(p->p_leader);
644 p->p_leader->p_flag |= P_ADVLOCK;
645 PROC_UNLOCK(p->p_leader);
646 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
650 if ((fp->f_flag & FWRITE) == 0) {
654 PROC_LOCK(p->p_leader);
655 p->p_leader->p_flag |= P_ADVLOCK;
656 PROC_UNLOCK(p->p_leader);
657 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
661 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
666 * Temporary api for testing remote lock
669 if (flg != F_REMOTE) {
673 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
674 F_UNLCKSYS, flp, flg);
680 if (error != 0 || flp->l_type == F_UNLCK ||
681 flp->l_type == F_UNLCKSYS) {
687 * Check for a race with close.
689 * The vnode is now advisory locked (or unlocked, but this case
690 * is not really important) as the caller requested.
691 * We had to drop the filedesc lock, so we need to recheck if
692 * the descriptor is still valid, because if it was closed
693 * in the meantime we need to remove advisory lock from the
694 * vnode - close on any descriptor leading to an advisory
695 * locked vnode, removes that lock.
696 * We will return 0 on purpose in that case, as the result of
697 * successful advisory lock might have been externally visible
698 * already. This is fine - effectively we pretend to the caller
699 * that the closing thread was a bit slower and that the
700 * advisory lock succeeded before the close.
702 error = fget_unlocked(fdp, fd, &rights, &fp2, NULL);
708 flp->l_whence = SEEK_SET;
711 flp->l_type = F_UNLCK;
712 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
713 F_UNLCK, flp, F_POSIX);
720 error = fget_unlocked(fdp, fd,
721 cap_rights_init(&rights, CAP_FLOCK), &fp, NULL);
724 if (fp->f_type != DTYPE_VNODE) {
729 flp = (struct flock *)arg;
730 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
731 flp->l_type != F_UNLCK) {
736 if (flp->l_whence == SEEK_CUR) {
737 foffset = foffset_get(fp);
738 if ((flp->l_start > 0 &&
739 foffset > OFF_MAX - flp->l_start) ||
741 foffset < OFF_MIN - flp->l_start)) {
746 flp->l_start += foffset;
749 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
755 arg = arg ? 128 * 1024: 0;
758 error = fget_unlocked(fdp, fd,
759 cap_rights_init(&rights), &fp, NULL);
762 if (fp->f_type != DTYPE_VNODE) {
769 * Exclusive lock synchronizes against f_seqcount reads and
770 * writes in sequential_heuristic().
772 error = vn_lock(vp, LK_EXCLUSIVE);
778 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
779 fp->f_seqcount = (arg + bsize - 1) / bsize;
780 atomic_set_int(&fp->f_flag, FRDAHEAD);
782 atomic_clear_int(&fp->f_flag, FRDAHEAD);
796 getmaxfd(struct thread *td)
799 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
803 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
806 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
808 struct filedesc *fdp;
809 struct filedescent *oldfde, *newfde;
817 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
818 MPASS(mode < FDDUP_LASTMODE);
821 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
824 * Verify we have a valid descriptor to dup from and possibly to
825 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
826 * return EINVAL when the new descriptor is out of bounds.
831 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
832 maxfd = getmaxfd(td);
834 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
838 if (fget_locked(fdp, old) == NULL)
840 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
841 td->td_retval[0] = new;
842 if (flags & FDDUP_FLAG_CLOEXEC)
843 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
849 * If the caller specified a file descriptor, make sure the file
850 * table is large enough to hold it, and grab it. Otherwise, just
851 * allocate a new descriptor the usual way.
856 if ((error = fdalloc(td, new, &new)) != 0)
859 case FDDUP_MUSTREPLACE:
860 /* Target file descriptor must exist. */
861 if (fget_locked(fdp, new) == NULL)
865 if (new >= fdp->fd_nfiles) {
867 * The resource limits are here instead of e.g.
868 * fdalloc(), because the file descriptor table may be
869 * shared between processes, so we can't really use
870 * racct_add()/racct_sub(). Instead of counting the
871 * number of actually allocated descriptors, just put
872 * the limit on the size of the file descriptor table.
877 error = racct_set(p, RACCT_NOFILE, new + 1);
885 fdgrowtable_exp(fdp, new + 1);
887 if (!fdisused(fdp, new))
891 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
894 KASSERT(old != new, ("new fd is same as old"));
896 oldfde = &fdp->fd_ofiles[old];
897 fhold(oldfde->fde_file);
898 newfde = &fdp->fd_ofiles[new];
899 delfp = newfde->fde_file;
902 * Duplicate the source descriptor.
905 seq_write_begin(&newfde->fde_seq);
907 filecaps_free(&newfde->fde_caps);
908 memcpy(newfde, oldfde, fde_change_size);
909 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps, true);
910 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
911 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
913 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
915 seq_write_end(&newfde->fde_seq);
917 td->td_retval[0] = new;
922 (void) closefp(fdp, new, delfp, td, 1);
923 FILEDESC_UNLOCK_ASSERT(fdp);
926 FILEDESC_XUNLOCK(fdp);
933 * If sigio is on the list associated with a process or process group,
934 * disable signalling from the device, remove sigio from the list and
938 funsetown(struct sigio **sigiop)
950 *(sigio->sio_myref) = NULL;
951 if ((sigio)->sio_pgid < 0) {
952 struct pgrp *pg = (sigio)->sio_pgrp;
954 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
958 struct proc *p = (sigio)->sio_proc;
960 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
965 crfree(sigio->sio_ucred);
966 free(sigio, M_SIGIO);
970 * Free a list of sigio structures.
971 * We only need to lock the SIGIO_LOCK because we have made ourselves
972 * inaccessible to callers of fsetown and therefore do not need to lock
973 * the proc or pgrp struct for the list manipulation.
976 funsetownlst(struct sigiolst *sigiolst)
982 sigio = SLIST_FIRST(sigiolst);
989 * Every entry of the list should belong
990 * to a single proc or pgrp.
992 if (sigio->sio_pgid < 0) {
993 pg = sigio->sio_pgrp;
994 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
995 } else /* if (sigio->sio_pgid > 0) */ {
997 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1001 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1002 *(sigio->sio_myref) = NULL;
1004 KASSERT(sigio->sio_pgid < 0,
1005 ("Proc sigio in pgrp sigio list"));
1006 KASSERT(sigio->sio_pgrp == pg,
1007 ("Bogus pgrp in sigio list"));
1009 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1012 } else /* if (p != NULL) */ {
1013 KASSERT(sigio->sio_pgid > 0,
1014 ("Pgrp sigio in proc sigio list"));
1015 KASSERT(sigio->sio_proc == p,
1016 ("Bogus proc in sigio list"));
1018 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1023 crfree(sigio->sio_ucred);
1024 free(sigio, M_SIGIO);
1031 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1033 * After permission checking, add a sigio structure to the sigio list for
1034 * the process or process group.
1037 fsetown(pid_t pgid, struct sigio **sigiop)
1041 struct sigio *sigio;
1051 /* Allocate and fill in the new sigio out of locks. */
1052 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1053 sigio->sio_pgid = pgid;
1054 sigio->sio_ucred = crhold(curthread->td_ucred);
1055 sigio->sio_myref = sigiop;
1057 sx_slock(&proctree_lock);
1066 * Policy - Don't allow a process to FSETOWN a process
1067 * in another session.
1069 * Remove this test to allow maximum flexibility or
1070 * restrict FSETOWN to the current process or process
1071 * group for maximum safety.
1074 if (proc->p_session != curthread->td_proc->p_session) {
1080 } else /* if (pgid < 0) */ {
1081 pgrp = pgfind(-pgid);
1089 * Policy - Don't allow a process to FSETOWN a process
1090 * in another session.
1092 * Remove this test to allow maximum flexibility or
1093 * restrict FSETOWN to the current process or process
1094 * group for maximum safety.
1096 if (pgrp->pg_session != curthread->td_proc->p_session) {
1107 * Since funsetownlst() is called without the proctree
1108 * locked, we need to check for P_WEXIT.
1109 * XXX: is ESRCH correct?
1111 if ((proc->p_flag & P_WEXIT) != 0) {
1116 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1117 sigio->sio_proc = proc;
1121 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1122 sigio->sio_pgrp = pgrp;
1125 sx_sunlock(&proctree_lock);
1132 sx_sunlock(&proctree_lock);
1133 crfree(sigio->sio_ucred);
1134 free(sigio, M_SIGIO);
1139 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1143 struct sigio **sigiop;
1148 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1154 * Function drops the filedesc lock on return.
1157 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1162 FILEDESC_XLOCK_ASSERT(fdp);
1165 if (td->td_proc->p_fdtol != NULL) {
1167 * Ask fdfree() to sleep to ensure that all relevant
1168 * process leaders can be traversed in closef().
1170 fdp->fd_holdleaderscount++;
1177 * We now hold the fp reference that used to be owned by the
1178 * descriptor array. We have to unlock the FILEDESC *AFTER*
1179 * knote_fdclose to prevent a race of the fd getting opened, a knote
1180 * added, and deleteing a knote for the new fd.
1182 knote_fdclose(td, fd);
1185 * We need to notify mqueue if the object is of type mqueue.
1187 if (fp->f_type == DTYPE_MQUEUE)
1188 mq_fdclose(td, fd, fp);
1189 FILEDESC_XUNLOCK(fdp);
1191 error = closef(fp, td);
1193 FILEDESC_XLOCK(fdp);
1194 fdp->fd_holdleaderscount--;
1195 if (fdp->fd_holdleaderscount == 0 &&
1196 fdp->fd_holdleaderswakeup != 0) {
1197 fdp->fd_holdleaderswakeup = 0;
1198 wakeup(&fdp->fd_holdleaderscount);
1200 FILEDESC_XUNLOCK(fdp);
1206 * Close a file descriptor.
1208 #ifndef _SYS_SYSPROTO_H_
1215 sys_close(struct thread *td, struct close_args *uap)
1218 return (kern_close(td, uap->fd));
1222 kern_close(struct thread *td, int fd)
1224 struct filedesc *fdp;
1227 fdp = td->td_proc->p_fd;
1229 AUDIT_SYSCLOSE(td, fd);
1231 FILEDESC_XLOCK(fdp);
1232 if ((fp = fget_locked(fdp, fd)) == NULL) {
1233 FILEDESC_XUNLOCK(fdp);
1238 /* closefp() drops the FILEDESC lock for us. */
1239 return (closefp(fdp, fd, fp, td, 1));
1243 * Close open file descriptors.
1245 #ifndef _SYS_SYSPROTO_H_
1246 struct closefrom_args {
1252 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1254 struct filedesc *fdp;
1257 fdp = td->td_proc->p_fd;
1258 AUDIT_ARG_FD(uap->lowfd);
1261 * Treat negative starting file descriptor values identical to
1262 * closefrom(0) which closes all files.
1266 FILEDESC_SLOCK(fdp);
1267 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1268 if (fdp->fd_ofiles[fd].fde_file != NULL) {
1269 FILEDESC_SUNLOCK(fdp);
1270 (void)kern_close(td, fd);
1271 FILEDESC_SLOCK(fdp);
1274 FILEDESC_SUNLOCK(fdp);
1278 #if defined(COMPAT_43)
1280 * Return status information about a file descriptor.
1282 #ifndef _SYS_SYSPROTO_H_
1283 struct ofstat_args {
1290 ofstat(struct thread *td, struct ofstat_args *uap)
1296 error = kern_fstat(td, uap->fd, &ub);
1299 error = copyout(&oub, uap->sb, sizeof(oub));
1303 #endif /* COMPAT_43 */
1306 * Return status information about a file descriptor.
1308 #ifndef _SYS_SYSPROTO_H_
1316 sys_fstat(struct thread *td, struct fstat_args *uap)
1321 error = kern_fstat(td, uap->fd, &ub);
1323 error = copyout(&ub, uap->sb, sizeof(ub));
1328 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1331 cap_rights_t rights;
1336 error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp);
1340 AUDIT_ARG_FILE(td->td_proc, fp);
1342 error = fo_stat(fp, sbp, td->td_ucred, td);
1345 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1352 * Return status information about a file descriptor.
1354 #ifndef _SYS_SYSPROTO_H_
1355 struct nfstat_args {
1362 sys_nfstat(struct thread *td, struct nfstat_args *uap)
1368 error = kern_fstat(td, uap->fd, &ub);
1370 cvtnstat(&ub, &nub);
1371 error = copyout(&nub, uap->sb, sizeof(nub));
1377 * Return pathconf information about a file descriptor.
1379 #ifndef _SYS_SYSPROTO_H_
1380 struct fpathconf_args {
1387 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1390 return (kern_fpathconf(td, uap->fd, uap->name));
1394 kern_fpathconf(struct thread *td, int fd, int name)
1398 cap_rights_t rights;
1401 error = fget(td, fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp);
1405 if (name == _PC_ASYNC_IO) {
1406 td->td_retval[0] = _POSIX_ASYNCHRONOUS_IO;
1411 vn_lock(vp, LK_SHARED | LK_RETRY);
1412 error = VOP_PATHCONF(vp, name, td->td_retval);
1414 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1415 if (name != _PC_PIPE_BUF) {
1418 td->td_retval[0] = PIPE_BUF;
1430 * Initialize filecaps structure.
1433 filecaps_init(struct filecaps *fcaps)
1436 bzero(fcaps, sizeof(*fcaps));
1437 fcaps->fc_nioctls = -1;
1441 * Copy filecaps structure allocating memory for ioctls array if needed.
1443 * The last parameter indicates whether the fdtable is locked. If it is not and
1444 * ioctls are encountered, copying fails and the caller must lock the table.
1446 * Note that if the table was not locked, the caller has to check the relevant
1447 * sequence counter to determine whether the operation was successful.
1450 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1454 if (src->fc_ioctls != NULL && !locked)
1457 if (src->fc_ioctls == NULL)
1460 KASSERT(src->fc_nioctls > 0,
1461 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1463 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1464 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1465 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1470 * Move filecaps structure to the new place and clear the old place.
1473 filecaps_move(struct filecaps *src, struct filecaps *dst)
1477 bzero(src, sizeof(*src));
1481 * Fill the given filecaps structure with full rights.
1484 filecaps_fill(struct filecaps *fcaps)
1487 CAP_ALL(&fcaps->fc_rights);
1488 fcaps->fc_ioctls = NULL;
1489 fcaps->fc_nioctls = -1;
1490 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1494 * Free memory allocated within filecaps structure.
1497 filecaps_free(struct filecaps *fcaps)
1500 free(fcaps->fc_ioctls, M_FILECAPS);
1501 bzero(fcaps, sizeof(*fcaps));
1505 * Validate the given filecaps structure.
1508 filecaps_validate(const struct filecaps *fcaps, const char *func)
1511 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1512 ("%s: invalid rights", func));
1513 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1514 ("%s: invalid fcntls", func));
1515 KASSERT(fcaps->fc_fcntls == 0 ||
1516 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1517 ("%s: fcntls without CAP_FCNTL", func));
1518 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1519 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1520 ("%s: invalid ioctls", func));
1521 KASSERT(fcaps->fc_nioctls == 0 ||
1522 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1523 ("%s: ioctls without CAP_IOCTL", func));
1527 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1531 FILEDESC_XLOCK_ASSERT(fdp);
1533 nfd1 = fdp->fd_nfiles * 2;
1536 fdgrowtable(fdp, nfd1);
1540 * Grow the file table to accommodate (at least) nfd descriptors.
1543 fdgrowtable(struct filedesc *fdp, int nfd)
1545 struct filedesc0 *fdp0;
1546 struct freetable *ft;
1547 struct fdescenttbl *ntable;
1548 struct fdescenttbl *otable;
1549 int nnfiles, onfiles;
1550 NDSLOTTYPE *nmap, *omap;
1553 * If lastfile is -1 this struct filedesc was just allocated and we are
1554 * growing it to accommodate for the one we are going to copy from. There
1555 * is no need to have a lock on this one as it's not visible to anyone.
1557 if (fdp->fd_lastfile != -1)
1558 FILEDESC_XLOCK_ASSERT(fdp);
1560 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1562 /* save old values */
1563 onfiles = fdp->fd_nfiles;
1564 otable = fdp->fd_files;
1567 /* compute the size of the new table */
1568 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1569 if (nnfiles <= onfiles)
1570 /* the table is already large enough */
1574 * Allocate a new table. We need enough space for the number of
1575 * entries, file entries themselves and the struct freetable we will use
1576 * when we decommission the table and place it on the freelist.
1577 * We place the struct freetable in the middle so we don't have
1578 * to worry about padding.
1580 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1581 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1582 sizeof(struct freetable),
1583 M_FILEDESC, M_ZERO | M_WAITOK);
1584 /* copy the old data */
1585 ntable->fdt_nfiles = nnfiles;
1586 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1587 onfiles * sizeof(ntable->fdt_ofiles[0]));
1590 * Allocate a new map only if the old is not large enough. It will
1591 * grow at a slower rate than the table as it can map more
1592 * entries than the table can hold.
1594 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1595 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1597 /* copy over the old data and update the pointer */
1598 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1603 * Make sure that ntable is correctly initialized before we replace
1604 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1607 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1610 * Do not free the old file table, as some threads may still
1611 * reference entries within it. Instead, place it on a freelist
1612 * which will be processed when the struct filedesc is released.
1614 * Note that if onfiles == NDFILE, we're dealing with the original
1615 * static allocation contained within (struct filedesc0 *)fdp,
1616 * which must not be freed.
1618 if (onfiles > NDFILE) {
1619 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1620 fdp0 = (struct filedesc0 *)fdp;
1621 ft->ft_table = otable;
1622 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1625 * The map does not have the same possibility of threads still
1626 * holding references to it. So always free it as long as it
1627 * does not reference the original static allocation.
1629 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1630 free(omap, M_FILEDESC);
1634 * Allocate a file descriptor for the process.
1637 fdalloc(struct thread *td, int minfd, int *result)
1639 struct proc *p = td->td_proc;
1640 struct filedesc *fdp = p->p_fd;
1641 int fd, maxfd, allocfd;
1646 FILEDESC_XLOCK_ASSERT(fdp);
1648 if (fdp->fd_freefile > minfd)
1649 minfd = fdp->fd_freefile;
1651 maxfd = getmaxfd(td);
1654 * Search the bitmap for a free descriptor starting at minfd.
1655 * If none is found, grow the file table.
1657 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1660 if (fd >= fdp->fd_nfiles) {
1661 allocfd = min(fd * 2, maxfd);
1665 error = racct_set(p, RACCT_NOFILE, allocfd);
1672 * fd is already equal to first free descriptor >= minfd, so
1673 * we only need to grow the table and we are done.
1675 fdgrowtable_exp(fdp, allocfd);
1679 * Perform some sanity checks, then mark the file descriptor as
1680 * used and return it to the caller.
1682 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1683 ("invalid descriptor %d", fd));
1684 KASSERT(!fdisused(fdp, fd),
1685 ("fd_first_free() returned non-free descriptor"));
1686 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1687 ("file descriptor isn't free"));
1694 * Allocate n file descriptors for the process.
1697 fdallocn(struct thread *td, int minfd, int *fds, int n)
1699 struct proc *p = td->td_proc;
1700 struct filedesc *fdp = p->p_fd;
1703 FILEDESC_XLOCK_ASSERT(fdp);
1705 for (i = 0; i < n; i++)
1706 if (fdalloc(td, 0, &fds[i]) != 0)
1710 for (i--; i >= 0; i--)
1711 fdunused(fdp, fds[i]);
1719 * Create a new open file structure and allocate a file descriptor for the
1720 * process that refers to it. We add one reference to the file for the
1721 * descriptor table and one reference for resultfp. This is to prevent us
1722 * being preempted and the entry in the descriptor table closed after we
1723 * release the FILEDESC lock.
1726 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1727 struct filecaps *fcaps)
1732 error = falloc_noinstall(td, &fp);
1734 return (error); /* no reference held on error */
1736 error = finstall(td, fp, &fd, flags, fcaps);
1738 fdrop(fp, td); /* one reference (fp only) */
1742 if (resultfp != NULL)
1743 *resultfp = fp; /* copy out result */
1745 fdrop(fp, td); /* release local reference */
1747 if (resultfd != NULL)
1754 * Create a new open file structure without allocating a file descriptor.
1757 falloc_noinstall(struct thread *td, struct file **resultfp)
1760 int maxuserfiles = maxfiles - (maxfiles / 20);
1762 static struct timeval lastfail;
1765 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1767 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1768 if ((openfiles_new >= maxuserfiles &&
1769 priv_check(td, PRIV_MAXFILES) != 0) ||
1770 openfiles_new >= maxfiles) {
1771 atomic_subtract_int(&openfiles, 1);
1772 if (ppsratecheck(&lastfail, &curfail, 1)) {
1773 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1774 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1778 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1779 refcount_init(&fp->f_count, 1);
1780 fp->f_cred = crhold(td->td_ucred);
1781 fp->f_ops = &badfileops;
1787 * Install a file in a file descriptor table.
1790 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1791 struct filecaps *fcaps)
1793 struct filedescent *fde;
1797 filecaps_validate(fcaps, __func__);
1798 FILEDESC_XLOCK_ASSERT(fdp);
1800 fde = &fdp->fd_ofiles[fd];
1802 seq_write_begin(&fde->fde_seq);
1805 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1807 filecaps_move(fcaps, &fde->fde_caps);
1809 filecaps_fill(&fde->fde_caps);
1811 seq_write_end(&fde->fde_seq);
1816 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1817 struct filecaps *fcaps)
1819 struct filedesc *fdp = td->td_proc->p_fd;
1824 FILEDESC_XLOCK(fdp);
1825 if ((error = fdalloc(td, 0, fd))) {
1826 FILEDESC_XUNLOCK(fdp);
1830 _finstall(fdp, fp, *fd, flags, fcaps);
1831 FILEDESC_XUNLOCK(fdp);
1836 * Build a new filedesc structure from another.
1837 * Copy the current, root, and jail root vnode references.
1839 * If fdp is not NULL, return with it shared locked.
1842 fdinit(struct filedesc *fdp, bool prepfiles)
1844 struct filedesc0 *newfdp0;
1845 struct filedesc *newfdp;
1847 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1848 newfdp = &newfdp0->fd_fd;
1850 /* Create the file descriptor table. */
1851 FILEDESC_LOCK_INIT(newfdp);
1852 refcount_init(&newfdp->fd_refcnt, 1);
1853 refcount_init(&newfdp->fd_holdcnt, 1);
1854 newfdp->fd_cmask = CMASK;
1855 newfdp->fd_map = newfdp0->fd_dmap;
1856 newfdp->fd_lastfile = -1;
1857 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1858 newfdp->fd_files->fdt_nfiles = NDFILE;
1863 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1864 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1866 FILEDESC_SLOCK(fdp);
1867 newfdp->fd_cdir = fdp->fd_cdir;
1868 if (newfdp->fd_cdir)
1869 vrefact(newfdp->fd_cdir);
1870 newfdp->fd_rdir = fdp->fd_rdir;
1871 if (newfdp->fd_rdir)
1872 vrefact(newfdp->fd_rdir);
1873 newfdp->fd_jdir = fdp->fd_jdir;
1874 if (newfdp->fd_jdir)
1875 vrefact(newfdp->fd_jdir);
1878 FILEDESC_SUNLOCK(fdp);
1880 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1881 FILEDESC_SUNLOCK(fdp);
1882 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1883 FILEDESC_SLOCK(fdp);
1890 static struct filedesc *
1891 fdhold(struct proc *p)
1893 struct filedesc *fdp;
1895 PROC_LOCK_ASSERT(p, MA_OWNED);
1898 refcount_acquire(&fdp->fd_holdcnt);
1903 fddrop(struct filedesc *fdp)
1906 if (fdp->fd_holdcnt > 1) {
1907 if (refcount_release(&fdp->fd_holdcnt) == 0)
1911 FILEDESC_LOCK_DESTROY(fdp);
1912 uma_zfree(filedesc0_zone, fdp);
1916 * Share a filedesc structure.
1919 fdshare(struct filedesc *fdp)
1922 refcount_acquire(&fdp->fd_refcnt);
1927 * Unshare a filedesc structure, if necessary by making a copy
1930 fdunshare(struct thread *td)
1932 struct filedesc *tmp;
1933 struct proc *p = td->td_proc;
1935 if (p->p_fd->fd_refcnt == 1)
1938 tmp = fdcopy(p->p_fd);
1944 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
1948 td->td_proc->p_fd = fdp;
1952 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
1953 * this is to ease callers, not catch errors.
1956 fdcopy(struct filedesc *fdp)
1958 struct filedesc *newfdp;
1959 struct filedescent *nfde, *ofde;
1964 newfdp = fdinit(fdp, true);
1965 /* copy all passable descriptors (i.e. not kqueue) */
1966 newfdp->fd_freefile = -1;
1967 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1968 ofde = &fdp->fd_ofiles[i];
1969 if (ofde->fde_file == NULL ||
1970 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
1971 if (newfdp->fd_freefile == -1)
1972 newfdp->fd_freefile = i;
1975 nfde = &newfdp->fd_ofiles[i];
1977 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
1978 fhold(nfde->fde_file);
1979 fdused_init(newfdp, i);
1980 newfdp->fd_lastfile = i;
1982 if (newfdp->fd_freefile == -1)
1983 newfdp->fd_freefile = i;
1984 newfdp->fd_cmask = fdp->fd_cmask;
1985 FILEDESC_SUNLOCK(fdp);
1990 * Copies a filedesc structure, while remapping all file descriptors
1991 * stored inside using a translation table.
1993 * File descriptors are copied over to the new file descriptor table,
1994 * regardless of whether the close-on-exec flag is set.
1997 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
1998 struct filedesc **ret)
2000 struct filedesc *newfdp;
2001 struct filedescent *nfde, *ofde;
2006 newfdp = fdinit(fdp, true);
2007 if (nfds > fdp->fd_lastfile + 1) {
2008 /* New table cannot be larger than the old one. */
2012 /* Copy all passable descriptors (i.e. not kqueue). */
2013 newfdp->fd_freefile = nfds;
2014 for (i = 0; i < nfds; ++i) {
2015 if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2016 /* File descriptor out of bounds. */
2020 ofde = &fdp->fd_ofiles[fds[i]];
2021 if (ofde->fde_file == NULL) {
2022 /* Unused file descriptor. */
2026 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2027 /* File descriptor cannot be passed. */
2031 nfde = &newfdp->fd_ofiles[i];
2033 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2034 fhold(nfde->fde_file);
2035 fdused_init(newfdp, i);
2036 newfdp->fd_lastfile = i;
2038 newfdp->fd_cmask = fdp->fd_cmask;
2039 FILEDESC_SUNLOCK(fdp);
2043 FILEDESC_SUNLOCK(fdp);
2044 fdescfree_remapped(newfdp);
2049 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2050 * one of processes using it exits) and the table used to be shared.
2053 fdclearlocks(struct thread *td)
2055 struct filedesc *fdp;
2056 struct filedesc_to_leader *fdtol;
2066 MPASS(fdtol != NULL);
2068 FILEDESC_XLOCK(fdp);
2069 KASSERT(fdtol->fdl_refcount > 0,
2070 ("filedesc_to_refcount botch: fdl_refcount=%d",
2071 fdtol->fdl_refcount));
2072 if (fdtol->fdl_refcount == 1 &&
2073 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2074 for (i = 0; i <= fdp->fd_lastfile; i++) {
2075 fp = fdp->fd_ofiles[i].fde_file;
2076 if (fp == NULL || fp->f_type != DTYPE_VNODE)
2079 FILEDESC_XUNLOCK(fdp);
2080 lf.l_whence = SEEK_SET;
2083 lf.l_type = F_UNLCK;
2085 (void) VOP_ADVLOCK(vp,
2086 (caddr_t)p->p_leader, F_UNLCK,
2088 FILEDESC_XLOCK(fdp);
2093 if (fdtol->fdl_refcount == 1) {
2094 if (fdp->fd_holdleaderscount > 0 &&
2095 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2097 * close() or kern_dup() has cleared a reference
2098 * in a shared file descriptor table.
2100 fdp->fd_holdleaderswakeup = 1;
2101 sx_sleep(&fdp->fd_holdleaderscount,
2102 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2105 if (fdtol->fdl_holdcount > 0) {
2107 * Ensure that fdtol->fdl_leader remains
2108 * valid in closef().
2110 fdtol->fdl_wakeup = 1;
2111 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2116 fdtol->fdl_refcount--;
2117 if (fdtol->fdl_refcount == 0 &&
2118 fdtol->fdl_holdcount == 0) {
2119 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2120 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2124 FILEDESC_XUNLOCK(fdp);
2126 free(fdtol, M_FILEDESC_TO_LEADER);
2130 * Release a filedesc structure.
2133 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2135 struct filedesc0 *fdp0;
2136 struct freetable *ft, *tft;
2137 struct filedescent *fde;
2141 for (i = 0; i <= fdp->fd_lastfile; i++) {
2142 fde = &fdp->fd_ofiles[i];
2147 (void) closef(fp, td);
2153 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2154 free(fdp->fd_map, M_FILEDESC);
2155 if (fdp->fd_nfiles > NDFILE)
2156 free(fdp->fd_files, M_FILEDESC);
2158 fdp0 = (struct filedesc0 *)fdp;
2159 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2160 free(ft->ft_table, M_FILEDESC);
2166 fdescfree(struct thread *td)
2169 struct filedesc *fdp;
2170 struct vnode *cdir, *jdir, *rdir;
2179 racct_set(p, RACCT_NOFILE, 0);
2184 if (p->p_fdtol != NULL)
2191 if (refcount_release(&fdp->fd_refcnt) == 0)
2194 FILEDESC_XLOCK(fdp);
2195 cdir = fdp->fd_cdir;
2196 fdp->fd_cdir = NULL;
2197 rdir = fdp->fd_rdir;
2198 fdp->fd_rdir = NULL;
2199 jdir = fdp->fd_jdir;
2200 fdp->fd_jdir = NULL;
2201 FILEDESC_XUNLOCK(fdp);
2210 fdescfree_fds(td, fdp, 1);
2214 fdescfree_remapped(struct filedesc *fdp)
2217 if (fdp->fd_cdir != NULL)
2218 vrele(fdp->fd_cdir);
2219 if (fdp->fd_rdir != NULL)
2220 vrele(fdp->fd_rdir);
2221 if (fdp->fd_jdir != NULL)
2222 vrele(fdp->fd_jdir);
2224 fdescfree_fds(curthread, fdp, 0);
2228 * For setugid programs, we don't want to people to use that setugidness
2229 * to generate error messages which write to a file which otherwise would
2230 * otherwise be off-limits to the process. We check for filesystems where
2231 * the vnode can change out from under us after execve (like [lin]procfs).
2233 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2234 * sufficient. We also don't check for setugidness since we know we are.
2237 is_unsafe(struct file *fp)
2241 if (fp->f_type != DTYPE_VNODE)
2245 return ((vp->v_vflag & VV_PROCDEP) != 0);
2249 * Make this setguid thing safe, if at all possible.
2252 fdsetugidsafety(struct thread *td)
2254 struct filedesc *fdp;
2258 fdp = td->td_proc->p_fd;
2259 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2260 MPASS(fdp->fd_nfiles >= 3);
2261 for (i = 0; i <= 2; i++) {
2262 fp = fdp->fd_ofiles[i].fde_file;
2263 if (fp != NULL && is_unsafe(fp)) {
2264 FILEDESC_XLOCK(fdp);
2265 knote_fdclose(td, i);
2267 * NULL-out descriptor prior to close to avoid
2268 * a race while close blocks.
2271 FILEDESC_XUNLOCK(fdp);
2272 (void) closef(fp, td);
2278 * If a specific file object occupies a specific file descriptor, close the
2279 * file descriptor entry and drop a reference on the file object. This is a
2280 * convenience function to handle a subsequent error in a function that calls
2281 * falloc() that handles the race that another thread might have closed the
2282 * file descriptor out from under the thread creating the file object.
2285 fdclose(struct thread *td, struct file *fp, int idx)
2287 struct filedesc *fdp = td->td_proc->p_fd;
2289 FILEDESC_XLOCK(fdp);
2290 if (fdp->fd_ofiles[idx].fde_file == fp) {
2292 FILEDESC_XUNLOCK(fdp);
2295 FILEDESC_XUNLOCK(fdp);
2299 * Close any files on exec?
2302 fdcloseexec(struct thread *td)
2304 struct filedesc *fdp;
2305 struct filedescent *fde;
2309 fdp = td->td_proc->p_fd;
2310 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2311 for (i = 0; i <= fdp->fd_lastfile; i++) {
2312 fde = &fdp->fd_ofiles[i];
2314 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2315 (fde->fde_flags & UF_EXCLOSE))) {
2316 FILEDESC_XLOCK(fdp);
2318 (void) closefp(fdp, i, fp, td, 0);
2319 FILEDESC_UNLOCK_ASSERT(fdp);
2325 * It is unsafe for set[ug]id processes to be started with file
2326 * descriptors 0..2 closed, as these descriptors are given implicit
2327 * significance in the Standard C library. fdcheckstd() will create a
2328 * descriptor referencing /dev/null for each of stdin, stdout, and
2329 * stderr that is not already open.
2332 fdcheckstd(struct thread *td)
2334 struct filedesc *fdp;
2336 int i, error, devnull;
2338 fdp = td->td_proc->p_fd;
2339 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2340 MPASS(fdp->fd_nfiles >= 3);
2342 for (i = 0; i <= 2; i++) {
2343 if (fdp->fd_ofiles[i].fde_file != NULL)
2346 save = td->td_retval[0];
2347 if (devnull != -1) {
2348 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2350 error = kern_openat(td, AT_FDCWD, "/dev/null",
2351 UIO_SYSSPACE, O_RDWR, 0);
2353 devnull = td->td_retval[0];
2354 KASSERT(devnull == i, ("we didn't get our fd"));
2357 td->td_retval[0] = save;
2365 * Internal form of close. Decrement reference count on file structure.
2366 * Note: td may be NULL when closing a file that was being passed in a
2369 * XXXRW: Giant is not required for the caller, but often will be held; this
2370 * makes it moderately likely the Giant will be recursed in the VFS case.
2373 closef(struct file *fp, struct thread *td)
2377 struct filedesc_to_leader *fdtol;
2378 struct filedesc *fdp;
2381 * POSIX record locking dictates that any close releases ALL
2382 * locks owned by this process. This is handled by setting
2383 * a flag in the unlock to free ONLY locks obeying POSIX
2384 * semantics, and not to free BSD-style file locks.
2385 * If the descriptor was in a message, POSIX-style locks
2386 * aren't passed with the descriptor, and the thread pointer
2387 * will be NULL. Callers should be careful only to pass a
2388 * NULL thread pointer when there really is no owning
2389 * context that might have locks, or the locks will be
2392 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2394 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2395 lf.l_whence = SEEK_SET;
2398 lf.l_type = F_UNLCK;
2399 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2400 F_UNLCK, &lf, F_POSIX);
2402 fdtol = td->td_proc->p_fdtol;
2403 if (fdtol != NULL) {
2405 * Handle special case where file descriptor table is
2406 * shared between multiple process leaders.
2408 fdp = td->td_proc->p_fd;
2409 FILEDESC_XLOCK(fdp);
2410 for (fdtol = fdtol->fdl_next;
2411 fdtol != td->td_proc->p_fdtol;
2412 fdtol = fdtol->fdl_next) {
2413 if ((fdtol->fdl_leader->p_flag &
2416 fdtol->fdl_holdcount++;
2417 FILEDESC_XUNLOCK(fdp);
2418 lf.l_whence = SEEK_SET;
2421 lf.l_type = F_UNLCK;
2423 (void) VOP_ADVLOCK(vp,
2424 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2426 FILEDESC_XLOCK(fdp);
2427 fdtol->fdl_holdcount--;
2428 if (fdtol->fdl_holdcount == 0 &&
2429 fdtol->fdl_wakeup != 0) {
2430 fdtol->fdl_wakeup = 0;
2434 FILEDESC_XUNLOCK(fdp);
2437 return (fdrop(fp, td));
2441 * Initialize the file pointer with the specified properties.
2443 * The ops are set with release semantics to be certain that the flags, type,
2444 * and data are visible when ops is. This is to prevent ops methods from being
2445 * called with bad data.
2448 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2453 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2457 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2458 struct file **fpp, struct filecaps *havecapsp)
2460 struct filedescent *fde;
2463 FILEDESC_LOCK_ASSERT(fdp);
2465 fde = fdeget_locked(fdp, fd);
2472 error = cap_check(cap_rights_fde(fde), needrightsp);
2477 if (havecapsp != NULL)
2478 filecaps_copy(&fde->fde_caps, havecapsp, true);
2480 *fpp = fde->fde_file;
2488 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2489 struct file **fpp, struct filecaps *havecapsp)
2491 struct filedesc *fdp = td->td_proc->p_fd;
2493 #ifndef CAPABILITIES
2494 error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2495 if (error == 0 && havecapsp != NULL)
2496 filecaps_fill(havecapsp);
2502 error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2506 if (havecapsp != NULL) {
2507 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2508 havecapsp, false)) {
2514 if (!fd_modified(fdp, fd, seq))
2523 FILEDESC_SLOCK(fdp);
2524 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2527 FILEDESC_SUNLOCK(fdp);
2533 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2534 struct file **fpp, seq_t *seqp)
2537 struct filedescent *fde;
2539 struct fdescenttbl *fdt;
2544 cap_rights_t haverights;
2548 fdt = fdp->fd_files;
2549 if ((u_int)fd >= fdt->fdt_nfiles)
2552 * Fetch the descriptor locklessly. We avoid fdrop() races by
2553 * never raising a refcount above 0. To accomplish this we have
2554 * to use a cmpset loop rather than an atomic_add. The descriptor
2555 * must be re-verified once we acquire a reference to be certain
2556 * that the identity is still correct and we did not lose a race
2557 * due to preemption.
2561 seq = seq_read(fd_seq(fdt, fd));
2562 fde = &fdt->fdt_ofiles[fd];
2563 haverights = *cap_rights_fde(fde);
2565 if (!seq_consistent(fd_seq(fdt, fd), seq))
2568 fp = fdt->fdt_ofiles[fd].fde_file;
2573 error = cap_check(&haverights, needrightsp);
2577 count = fp->f_count;
2581 * Force a reload. Other thread could reallocate the
2582 * table before this fd was closed, so it possible that
2583 * there is a stale fp pointer in cached version.
2585 fdt = *(struct fdescenttbl * volatile *)&(fdp->fd_files);
2589 * Use an acquire barrier to force re-reading of fdt so it is
2590 * refreshed for verification.
2592 if (atomic_fcmpset_acq_int(&fp->f_count, &count, count + 1) == 0)
2594 fdt = fdp->fd_files;
2596 if (seq_consistent_nomb(fd_seq(fdt, fd), seq))
2598 if (fp == fdt->fdt_ofiles[fd].fde_file)
2601 fdrop(fp, curthread);
2613 * Extract the file pointer associated with the specified descriptor for the
2614 * current user process.
2616 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2619 * File's rights will be checked against the capability rights mask.
2621 * If an error occurred the non-zero error is returned and *fpp is set to
2622 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
2623 * responsible for fdrop().
2626 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2627 cap_rights_t *needrightsp, seq_t *seqp)
2629 struct filedesc *fdp;
2634 fdp = td->td_proc->p_fd;
2635 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2638 if (fp->f_ops == &badfileops) {
2644 * FREAD and FWRITE failure return EBADF as per POSIX.
2650 if ((fp->f_flag & flags) == 0)
2654 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2655 ((fp->f_flag & FWRITE) != 0))
2661 KASSERT(0, ("wrong flags"));
2674 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2677 return (_fget(td, fd, fpp, 0, rightsp, NULL));
2681 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2685 #ifndef CAPABILITIES
2686 error = _fget(td, fd, fpp, 0, rightsp, NULL);
2687 if (maxprotp != NULL)
2688 *maxprotp = VM_PROT_ALL;
2690 struct filedesc *fdp = td->td_proc->p_fd;
2693 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2695 error = _fget(td, fd, fpp, 0, rightsp, &seq);
2699 * If requested, convert capability rights to access flags.
2701 if (maxprotp != NULL)
2702 *maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd));
2703 if (!fd_modified(fdp, fd, seq))
2712 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2715 return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2719 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2722 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2726 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2729 struct filedesc *fdp = td->td_proc->p_fd;
2730 #ifndef CAPABILITIES
2731 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2736 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2738 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2741 error = cap_fcntl_check(fdp, fd, needfcntl);
2742 if (!fd_modified(fdp, fd, seq))
2755 * Like fget() but loads the underlying vnode, or returns an error if the
2756 * descriptor does not represent a vnode. Note that pipes use vnodes but
2757 * never have VM objects. The returned vnode will be vref()'d.
2759 * XXX: what about the unused flags ?
2762 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2769 error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2772 if (fp->f_vnode == NULL) {
2784 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2787 return (_fgetvp(td, fd, 0, rightsp, vpp));
2791 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2792 struct filecaps *havecaps, struct vnode **vpp)
2794 struct filedesc *fdp;
2795 struct filecaps caps;
2799 fdp = td->td_proc->p_fd;
2800 error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2803 if (fp->f_ops == &badfileops) {
2807 if (fp->f_vnode == NULL) {
2818 filecaps_free(&caps);
2823 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2826 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2830 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2833 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2838 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2842 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2847 * Handle the last reference to a file being closed.
2850 _fdrop(struct file *fp, struct thread *td)
2854 if (fp->f_count != 0)
2855 panic("fdrop: count %d", fp->f_count);
2856 error = fo_close(fp, td);
2857 atomic_subtract_int(&openfiles, 1);
2859 free(fp->f_advice, M_FADVISE);
2860 uma_zfree(file_zone, fp);
2866 * Apply an advisory lock on a file descriptor.
2868 * Just attempt to get a record lock of the requested type on the entire file
2869 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2871 #ifndef _SYS_SYSPROTO_H_
2879 sys_flock(struct thread *td, struct flock_args *uap)
2884 cap_rights_t rights;
2887 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp);
2890 if (fp->f_type != DTYPE_VNODE) {
2892 return (EOPNOTSUPP);
2896 lf.l_whence = SEEK_SET;
2899 if (uap->how & LOCK_UN) {
2900 lf.l_type = F_UNLCK;
2901 atomic_clear_int(&fp->f_flag, FHASLOCK);
2902 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2905 if (uap->how & LOCK_EX)
2906 lf.l_type = F_WRLCK;
2907 else if (uap->how & LOCK_SH)
2908 lf.l_type = F_RDLCK;
2913 atomic_set_int(&fp->f_flag, FHASLOCK);
2914 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2915 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2921 * Duplicate the specified descriptor to a free descriptor.
2924 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
2925 int openerror, int *indxp)
2927 struct filedescent *newfde, *oldfde;
2931 KASSERT(openerror == ENODEV || openerror == ENXIO,
2932 ("unexpected error %d in %s", openerror, __func__));
2935 * If the to-be-dup'd fd number is greater than the allowed number
2936 * of file descriptors, or the fd to be dup'd has already been
2937 * closed, then reject.
2939 FILEDESC_XLOCK(fdp);
2940 if ((fp = fget_locked(fdp, dfd)) == NULL) {
2941 FILEDESC_XUNLOCK(fdp);
2945 error = fdalloc(td, 0, &indx);
2947 FILEDESC_XUNLOCK(fdp);
2952 * There are two cases of interest here.
2954 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2956 * For ENXIO steal away the file structure from (dfd) and store it in
2957 * (indx). (dfd) is effectively closed by this operation.
2959 switch (openerror) {
2962 * Check that the mode the file is being opened for is a
2963 * subset of the mode of the existing descriptor.
2965 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
2966 fdunused(fdp, indx);
2967 FILEDESC_XUNLOCK(fdp);
2971 newfde = &fdp->fd_ofiles[indx];
2972 oldfde = &fdp->fd_ofiles[dfd];
2974 seq_write_begin(&newfde->fde_seq);
2976 memcpy(newfde, oldfde, fde_change_size);
2977 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps, true);
2979 seq_write_end(&newfde->fde_seq);
2984 * Steal away the file pointer from dfd and stuff it into indx.
2986 newfde = &fdp->fd_ofiles[indx];
2987 oldfde = &fdp->fd_ofiles[dfd];
2989 seq_write_begin(&newfde->fde_seq);
2991 memcpy(newfde, oldfde, fde_change_size);
2992 oldfde->fde_file = NULL;
2995 seq_write_end(&newfde->fde_seq);
2999 FILEDESC_XUNLOCK(fdp);
3005 * This sysctl determines if we will allow a process to chroot(2) if it
3006 * has a directory open:
3007 * 0: disallowed for all processes.
3008 * 1: allowed for processes that were not already chroot(2)'ed.
3009 * 2: allowed for all processes.
3012 static int chroot_allow_open_directories = 1;
3014 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3015 &chroot_allow_open_directories, 0,
3016 "Allow a process to chroot(2) if it has a directory open");
3019 * Helper function for raised chroot(2) security function: Refuse if
3020 * any filedescriptors are open directories.
3023 chroot_refuse_vdir_fds(struct filedesc *fdp)
3029 FILEDESC_LOCK_ASSERT(fdp);
3031 for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3032 fp = fget_locked(fdp, fd);
3035 if (fp->f_type == DTYPE_VNODE) {
3037 if (vp->v_type == VDIR)
3045 * Common routine for kern_chroot() and jail_attach(). The caller is
3046 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3047 * authorize this operation.
3050 pwd_chroot(struct thread *td, struct vnode *vp)
3052 struct filedesc *fdp;
3053 struct vnode *oldvp;
3056 fdp = td->td_proc->p_fd;
3057 FILEDESC_XLOCK(fdp);
3058 if (chroot_allow_open_directories == 0 ||
3059 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3060 error = chroot_refuse_vdir_fds(fdp);
3062 FILEDESC_XUNLOCK(fdp);
3066 oldvp = fdp->fd_rdir;
3069 if (fdp->fd_jdir == NULL) {
3073 FILEDESC_XUNLOCK(fdp);
3079 pwd_chdir(struct thread *td, struct vnode *vp)
3081 struct filedesc *fdp;
3082 struct vnode *oldvp;
3084 fdp = td->td_proc->p_fd;
3085 FILEDESC_XLOCK(fdp);
3086 VNASSERT(vp->v_usecount > 0, vp,
3087 ("chdir to a vnode with zero usecount"));
3088 oldvp = fdp->fd_cdir;
3090 FILEDESC_XUNLOCK(fdp);
3095 * Scan all active processes and prisons to see if any of them have a current
3096 * or root directory of `olddp'. If so, replace them with the new mount point.
3099 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3101 struct filedesc *fdp;
3106 if (vrefcnt(olddp) == 1)
3109 sx_slock(&allproc_lock);
3110 FOREACH_PROC_IN_SYSTEM(p) {
3116 FILEDESC_XLOCK(fdp);
3117 if (fdp->fd_cdir == olddp) {
3119 fdp->fd_cdir = newdp;
3122 if (fdp->fd_rdir == olddp) {
3124 fdp->fd_rdir = newdp;
3127 if (fdp->fd_jdir == olddp) {
3129 fdp->fd_jdir = newdp;
3132 FILEDESC_XUNLOCK(fdp);
3135 sx_sunlock(&allproc_lock);
3136 if (rootvnode == olddp) {
3141 mtx_lock(&prison0.pr_mtx);
3142 if (prison0.pr_root == olddp) {
3144 prison0.pr_root = newdp;
3147 mtx_unlock(&prison0.pr_mtx);
3148 sx_slock(&allprison_lock);
3149 TAILQ_FOREACH(pr, &allprison, pr_list) {
3150 mtx_lock(&pr->pr_mtx);
3151 if (pr->pr_root == olddp) {
3153 pr->pr_root = newdp;
3156 mtx_unlock(&pr->pr_mtx);
3158 sx_sunlock(&allprison_lock);
3163 struct filedesc_to_leader *
3164 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3166 struct filedesc_to_leader *fdtol;
3168 fdtol = malloc(sizeof(struct filedesc_to_leader),
3169 M_FILEDESC_TO_LEADER, M_WAITOK);
3170 fdtol->fdl_refcount = 1;
3171 fdtol->fdl_holdcount = 0;
3172 fdtol->fdl_wakeup = 0;
3173 fdtol->fdl_leader = leader;
3175 FILEDESC_XLOCK(fdp);
3176 fdtol->fdl_next = old->fdl_next;
3177 fdtol->fdl_prev = old;
3178 old->fdl_next = fdtol;
3179 fdtol->fdl_next->fdl_prev = fdtol;
3180 FILEDESC_XUNLOCK(fdp);
3182 fdtol->fdl_next = fdtol;
3183 fdtol->fdl_prev = fdtol;
3189 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3191 struct filedesc *fdp;
3192 int i, count, slots;
3194 if (*(int *)arg1 != 0)
3197 fdp = curproc->p_fd;
3199 FILEDESC_SLOCK(fdp);
3200 slots = NDSLOTS(fdp->fd_lastfile + 1);
3201 for (i = 0; i < slots; i++)
3202 count += bitcountl(fdp->fd_map[i]);
3203 FILEDESC_SUNLOCK(fdp);
3205 return (SYSCTL_OUT(req, &count, sizeof(count)));
3208 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3209 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3210 "Number of open file descriptors");
3213 * Get file structures globally.
3216 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3219 struct filedesc *fdp;
3224 error = sysctl_wire_old_buffer(req, 0);
3227 if (req->oldptr == NULL) {
3229 sx_slock(&allproc_lock);
3230 FOREACH_PROC_IN_SYSTEM(p) {
3232 if (p->p_state == PRS_NEW) {
3240 /* overestimates sparse tables. */
3241 if (fdp->fd_lastfile > 0)
3242 n += fdp->fd_lastfile;
3245 sx_sunlock(&allproc_lock);
3246 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3249 bzero(&xf, sizeof(xf));
3250 xf.xf_size = sizeof(xf);
3251 sx_slock(&allproc_lock);
3252 FOREACH_PROC_IN_SYSTEM(p) {
3254 if (p->p_state == PRS_NEW) {
3258 if (p_cansee(req->td, p) != 0) {
3262 xf.xf_pid = p->p_pid;
3263 xf.xf_uid = p->p_ucred->cr_uid;
3268 FILEDESC_SLOCK(fdp);
3269 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3270 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3274 xf.xf_data = fp->f_data;
3275 xf.xf_vnode = fp->f_vnode;
3276 xf.xf_type = fp->f_type;
3277 xf.xf_count = fp->f_count;
3279 xf.xf_offset = foffset_get(fp);
3280 xf.xf_flag = fp->f_flag;
3281 error = SYSCTL_OUT(req, &xf, sizeof(xf));
3285 FILEDESC_SUNLOCK(fdp);
3290 sx_sunlock(&allproc_lock);
3294 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3295 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3297 #ifdef KINFO_FILE_SIZE
3298 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3302 xlate_fflags(int fflags)
3304 static const struct {
3307 } fflags_table[] = {
3308 { FAPPEND, KF_FLAG_APPEND },
3309 { FASYNC, KF_FLAG_ASYNC },
3310 { FFSYNC, KF_FLAG_FSYNC },
3311 { FHASLOCK, KF_FLAG_HASLOCK },
3312 { FNONBLOCK, KF_FLAG_NONBLOCK },
3313 { FREAD, KF_FLAG_READ },
3314 { FWRITE, KF_FLAG_WRITE },
3315 { O_CREAT, KF_FLAG_CREAT },
3316 { O_DIRECT, KF_FLAG_DIRECT },
3317 { O_EXCL, KF_FLAG_EXCL },
3318 { O_EXEC, KF_FLAG_EXEC },
3319 { O_EXLOCK, KF_FLAG_EXLOCK },
3320 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3321 { O_SHLOCK, KF_FLAG_SHLOCK },
3322 { O_TRUNC, KF_FLAG_TRUNC }
3328 for (i = 0; i < nitems(fflags_table); i++)
3329 if (fflags & fflags_table[i].fflag)
3330 kflags |= fflags_table[i].kf_fflag;
3334 /* Trim unused data from kf_path by truncating the structure size. */
3336 pack_kinfo(struct kinfo_file *kif)
3339 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3340 strlen(kif->kf_path) + 1;
3341 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3345 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3346 struct kinfo_file *kif, struct filedesc *fdp, int flags)
3350 bzero(kif, sizeof(*kif));
3352 /* Set a default type to allow for empty fill_kinfo() methods. */
3353 kif->kf_type = KF_TYPE_UNKNOWN;
3354 kif->kf_flags = xlate_fflags(fp->f_flag);
3355 if (rightsp != NULL)
3356 kif->kf_cap_rights = *rightsp;
3358 cap_rights_init(&kif->kf_cap_rights);
3360 kif->kf_ref_count = fp->f_count;
3361 kif->kf_offset = foffset_get(fp);
3364 * This may drop the filedesc lock, so the 'fp' cannot be
3365 * accessed after this call.
3367 error = fo_fill_kinfo(fp, kif, fdp);
3369 kif->kf_status |= KF_ATTR_VALID;
3370 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3373 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3377 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3378 struct kinfo_file *kif, int flags)
3382 bzero(kif, sizeof(*kif));
3384 kif->kf_type = KF_TYPE_VNODE;
3385 error = vn_fill_kinfo_vnode(vp, kif);
3387 kif->kf_status |= KF_ATTR_VALID;
3388 kif->kf_flags = xlate_fflags(fflags);
3389 cap_rights_init(&kif->kf_cap_rights);
3391 kif->kf_ref_count = -1;
3392 kif->kf_offset = -1;
3393 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3396 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3400 struct export_fd_buf {
3401 struct filedesc *fdp;
3404 struct kinfo_file kif;
3409 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3411 struct kinfo_file *kif;
3414 if (efbuf->remainder != -1) {
3415 if (efbuf->remainder < kif->kf_structsize) {
3416 /* Terminate export. */
3417 efbuf->remainder = 0;
3420 efbuf->remainder -= kif->kf_structsize;
3422 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3426 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3427 struct export_fd_buf *efbuf)
3431 if (efbuf->remainder == 0)
3433 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3435 FILEDESC_SUNLOCK(efbuf->fdp);
3436 error = export_kinfo_to_sb(efbuf);
3437 FILEDESC_SLOCK(efbuf->fdp);
3442 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3443 struct export_fd_buf *efbuf)
3447 if (efbuf->remainder == 0)
3449 if (efbuf->fdp != NULL)
3450 FILEDESC_SUNLOCK(efbuf->fdp);
3451 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3452 error = export_kinfo_to_sb(efbuf);
3453 if (efbuf->fdp != NULL)
3454 FILEDESC_SLOCK(efbuf->fdp);
3459 * Store a process file descriptor information to sbuf.
3461 * Takes a locked proc as argument, and returns with the proc unlocked.
3464 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
3468 struct filedesc *fdp;
3469 struct export_fd_buf *efbuf;
3470 struct vnode *cttyvp, *textvp, *tracevp;
3472 cap_rights_t rights;
3474 PROC_LOCK_ASSERT(p, MA_OWNED);
3477 tracevp = p->p_tracevp;
3478 if (tracevp != NULL)
3481 textvp = p->p_textvp;
3484 /* Controlling tty. */
3486 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3487 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3493 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3496 efbuf->remainder = maxlen;
3497 efbuf->flags = flags;
3498 if (tracevp != NULL)
3499 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3502 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3504 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3510 FILEDESC_SLOCK(fdp);
3511 /* working directory */
3512 if (fdp->fd_cdir != NULL) {
3513 vrefact(fdp->fd_cdir);
3514 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3516 /* root directory */
3517 if (fdp->fd_rdir != NULL) {
3518 vrefact(fdp->fd_rdir);
3519 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3521 /* jail directory */
3522 if (fdp->fd_jdir != NULL) {
3523 vrefact(fdp->fd_jdir);
3524 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3526 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3527 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3530 rights = *cap_rights(fdp, i);
3531 #else /* !CAPABILITIES */
3532 cap_rights_init(&rights);
3535 * Create sysctl entry. It is OK to drop the filedesc
3536 * lock inside of export_file_to_sb() as we will
3537 * re-validate and re-evaluate its properties when the
3540 error = export_file_to_sb(fp, i, &rights, efbuf);
3541 if (error != 0 || efbuf->remainder == 0)
3544 FILEDESC_SUNLOCK(fdp);
3547 free(efbuf, M_TEMP);
3551 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
3554 * Get per-process file descriptors for use by procstat(1), et al.
3557 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3562 int error, error2, *name;
3566 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3567 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3568 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3573 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3574 error = kern_proc_filedesc_out(p, &sb, maxlen,
3575 KERN_FILEDESC_PACK_KINFO);
3576 error2 = sbuf_finish(&sb);
3578 return (error != 0 ? error : error2);
3581 #ifdef KINFO_OFILE_SIZE
3582 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3585 #ifdef COMPAT_FREEBSD7
3587 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3590 okif->kf_structsize = sizeof(*okif);
3591 okif->kf_type = kif->kf_type;
3592 okif->kf_fd = kif->kf_fd;
3593 okif->kf_ref_count = kif->kf_ref_count;
3594 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3595 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3596 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3597 okif->kf_offset = kif->kf_offset;
3598 okif->kf_vnode_type = kif->kf_vnode_type;
3599 okif->kf_sock_domain = kif->kf_sock_domain;
3600 okif->kf_sock_type = kif->kf_sock_type;
3601 okif->kf_sock_protocol = kif->kf_sock_protocol;
3602 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3603 okif->kf_sa_local = kif->kf_sa_local;
3604 okif->kf_sa_peer = kif->kf_sa_peer;
3608 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3609 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3614 FILEDESC_SUNLOCK(fdp);
3615 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3616 kinfo_to_okinfo(kif, okif);
3617 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3618 FILEDESC_SLOCK(fdp);
3623 * Get per-process file descriptors for use by procstat(1), et al.
3626 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3628 struct kinfo_ofile *okif;
3629 struct kinfo_file *kif;
3630 struct filedesc *fdp;
3631 int error, i, *name;
3636 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3643 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3644 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3645 FILEDESC_SLOCK(fdp);
3646 if (fdp->fd_cdir != NULL)
3647 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3649 if (fdp->fd_rdir != NULL)
3650 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3652 if (fdp->fd_jdir != NULL)
3653 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3655 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3656 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3658 export_file_to_kinfo(fp, i, NULL, kif, fdp,
3659 KERN_FILEDESC_PACK_KINFO);
3660 FILEDESC_SUNLOCK(fdp);
3661 kinfo_to_okinfo(kif, okif);
3662 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3663 FILEDESC_SLOCK(fdp);
3667 FILEDESC_SUNLOCK(fdp);
3674 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3675 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3676 "Process ofiledesc entries");
3677 #endif /* COMPAT_FREEBSD7 */
3680 vntype_to_kinfo(int vtype)
3685 } vtypes_table[] = {
3686 { VBAD, KF_VTYPE_VBAD },
3687 { VBLK, KF_VTYPE_VBLK },
3688 { VCHR, KF_VTYPE_VCHR },
3689 { VDIR, KF_VTYPE_VDIR },
3690 { VFIFO, KF_VTYPE_VFIFO },
3691 { VLNK, KF_VTYPE_VLNK },
3692 { VNON, KF_VTYPE_VNON },
3693 { VREG, KF_VTYPE_VREG },
3694 { VSOCK, KF_VTYPE_VSOCK }
3699 * Perform vtype translation.
3701 for (i = 0; i < nitems(vtypes_table); i++)
3702 if (vtypes_table[i].vtype == vtype)
3703 return (vtypes_table[i].kf_vtype);
3705 return (KF_VTYPE_UNKNOWN);
3708 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3709 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3710 "Process filedesc entries");
3713 * Store a process current working directory information to sbuf.
3715 * Takes a locked proc as argument, and returns with the proc unlocked.
3718 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
3720 struct filedesc *fdp;
3721 struct export_fd_buf *efbuf;
3724 PROC_LOCK_ASSERT(p, MA_OWNED);
3731 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3734 efbuf->remainder = maxlen;
3736 FILEDESC_SLOCK(fdp);
3737 if (fdp->fd_cdir == NULL)
3740 vrefact(fdp->fd_cdir);
3741 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3744 FILEDESC_SUNLOCK(fdp);
3746 free(efbuf, M_TEMP);
3751 * Get per-process current working directory.
3754 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3759 int error, error2, *name;
3763 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3764 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3765 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3770 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3771 error = kern_proc_cwd_out(p, &sb, maxlen);
3772 error2 = sbuf_finish(&sb);
3774 return (error != 0 ? error : error2);
3777 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3778 sysctl_kern_proc_cwd, "Process current working directory");
3782 * For the purposes of debugging, generate a human-readable string for the
3786 file_type_to_name(short type)
3816 * For the purposes of debugging, identify a process (if any, perhaps one of
3817 * many) that references the passed file in its file descriptor array. Return
3820 static struct proc *
3821 file_to_first_proc(struct file *fp)
3823 struct filedesc *fdp;
3827 FOREACH_PROC_IN_SYSTEM(p) {
3828 if (p->p_state == PRS_NEW)
3833 for (n = 0; n <= fdp->fd_lastfile; n++) {
3834 if (fp == fdp->fd_ofiles[n].fde_file)
3842 db_print_file(struct file *fp, int header)
3847 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
3848 "File", "Type", "Data", "Flag", "GCFl", "Count",
3849 "MCount", "Vnode", "FPID", "FCmd");
3850 p = file_to_first_proc(fp);
3851 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
3852 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
3853 0, fp->f_count, 0, fp->f_vnode,
3854 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3857 DB_SHOW_COMMAND(file, db_show_file)
3862 db_printf("usage: show file <addr>\n");
3865 fp = (struct file *)addr;
3866 db_print_file(fp, 1);
3869 DB_SHOW_COMMAND(files, db_show_files)
3871 struct filedesc *fdp;
3878 FOREACH_PROC_IN_SYSTEM(p) {
3879 if (p->p_state == PRS_NEW)
3881 if ((fdp = p->p_fd) == NULL)
3883 for (n = 0; n <= fdp->fd_lastfile; ++n) {
3884 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3886 db_print_file(fp, header);
3893 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3894 &maxfilesperproc, 0, "Maximum files allowed open per process");
3896 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3897 &maxfiles, 0, "Maximum number of files");
3899 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3900 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
3904 filelistinit(void *dummy)
3907 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3908 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
3909 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
3910 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
3911 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3913 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3915 /*-------------------------------------------------------------------*/
3918 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
3919 int flags, struct thread *td)
3926 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
3934 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
3942 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
3950 badfo_kqfilter(struct file *fp, struct knote *kn)
3957 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
3965 badfo_close(struct file *fp, struct thread *td)
3972 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
3980 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
3988 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
3989 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
3997 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4003 struct fileops badfileops = {
4004 .fo_read = badfo_readwrite,
4005 .fo_write = badfo_readwrite,
4006 .fo_truncate = badfo_truncate,
4007 .fo_ioctl = badfo_ioctl,
4008 .fo_poll = badfo_poll,
4009 .fo_kqfilter = badfo_kqfilter,
4010 .fo_stat = badfo_stat,
4011 .fo_close = badfo_close,
4012 .fo_chmod = badfo_chmod,
4013 .fo_chown = badfo_chown,
4014 .fo_sendfile = badfo_sendfile,
4015 .fo_fill_kinfo = badfo_fill_kinfo,
4019 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4020 int flags, struct thread *td)
4023 return (EOPNOTSUPP);
4027 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4035 invfo_ioctl(struct file *fp, u_long com, void *data,
4036 struct ucred *active_cred, struct thread *td)
4043 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4047 return (poll_no_poll(events));
4051 invfo_kqfilter(struct file *fp, struct knote *kn)
4058 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4066 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4074 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4075 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4082 /*-------------------------------------------------------------------*/
4085 * File Descriptor pseudo-device driver (/dev/fd/).
4087 * Opening minor device N dup()s the file (if any) connected to file
4088 * descriptor N belonging to the calling process. Note that this driver
4089 * consists of only the ``open()'' routine, because all subsequent
4090 * references to this file will be direct to the other driver.
4092 * XXX: we could give this one a cloning event handler if necessary.
4097 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4101 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4102 * the file descriptor being sought for duplication. The error
4103 * return ensures that the vnode for this device will be released
4104 * by vn_open. Open will detect this special error and take the
4105 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4106 * will simply report the error.
4108 td->td_dupfd = dev2unit(dev);
4112 static struct cdevsw fildesc_cdevsw = {
4113 .d_version = D_VERSION,
4119 fildesc_drvinit(void *unused)
4123 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4124 UID_ROOT, GID_WHEEL, 0666, "fd/0");
4125 make_dev_alias(dev, "stdin");
4126 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4127 UID_ROOT, GID_WHEEL, 0666, "fd/1");
4128 make_dev_alias(dev, "stdout");
4129 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4130 UID_ROOT, GID_WHEEL, 0666, "fd/2");
4131 make_dev_alias(dev, "stderr");
4134 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);