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 * 3. 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 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 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 openfiles; /* actual number of open files */
172 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
173 void (*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;
502 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
505 case F_DUPFD_CLOEXEC:
507 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
512 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
515 case F_DUP2FD_CLOEXEC:
517 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
523 fde = fdeget_locked(fdp, fd);
526 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
529 FILEDESC_SUNLOCK(fdp);
535 fde = fdeget_locked(fdp, fd);
537 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
538 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
541 FILEDESC_XUNLOCK(fdp);
545 error = fget_fcntl(td, fd,
546 cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp);
549 td->td_retval[0] = OFLAGS(fp->f_flag);
554 error = fget_fcntl(td, fd,
555 cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp);
559 tmp = flg = fp->f_flag;
561 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
562 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
563 tmp = fp->f_flag & FNONBLOCK;
564 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
569 tmp = fp->f_flag & FASYNC;
570 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
575 atomic_clear_int(&fp->f_flag, FNONBLOCK);
577 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
582 error = fget_fcntl(td, fd,
583 cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp);
586 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
588 td->td_retval[0] = tmp;
593 error = fget_fcntl(td, fd,
594 cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp);
598 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
603 error = priv_check(td, PRIV_NFS_LOCKD);
611 /* FALLTHROUGH F_SETLK */
615 cap_rights_init(&rights, CAP_FLOCK);
616 error = fget_unlocked(fdp, fd, &rights, &fp, NULL);
619 if (fp->f_type != DTYPE_VNODE) {
625 flp = (struct flock *)arg;
626 if (flp->l_whence == SEEK_CUR) {
627 foffset = foffset_get(fp);
630 foffset > OFF_MAX - flp->l_start)) {
635 flp->l_start += foffset;
639 switch (flp->l_type) {
641 if ((fp->f_flag & FREAD) == 0) {
645 PROC_LOCK(p->p_leader);
646 p->p_leader->p_flag |= P_ADVLOCK;
647 PROC_UNLOCK(p->p_leader);
648 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
652 if ((fp->f_flag & FWRITE) == 0) {
656 PROC_LOCK(p->p_leader);
657 p->p_leader->p_flag |= P_ADVLOCK;
658 PROC_UNLOCK(p->p_leader);
659 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
663 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
668 * Temporary api for testing remote lock
671 if (flg != F_REMOTE) {
675 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
676 F_UNLCKSYS, flp, flg);
682 if (error != 0 || flp->l_type == F_UNLCK ||
683 flp->l_type == F_UNLCKSYS) {
689 * Check for a race with close.
691 * The vnode is now advisory locked (or unlocked, but this case
692 * is not really important) as the caller requested.
693 * We had to drop the filedesc lock, so we need to recheck if
694 * the descriptor is still valid, because if it was closed
695 * in the meantime we need to remove advisory lock from the
696 * vnode - close on any descriptor leading to an advisory
697 * locked vnode, removes that lock.
698 * We will return 0 on purpose in that case, as the result of
699 * successful advisory lock might have been externally visible
700 * already. This is fine - effectively we pretend to the caller
701 * that the closing thread was a bit slower and that the
702 * advisory lock succeeded before the close.
704 error = fget_unlocked(fdp, fd, &rights, &fp2, NULL);
710 flp->l_whence = SEEK_SET;
713 flp->l_type = F_UNLCK;
714 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
715 F_UNLCK, flp, F_POSIX);
722 error = fget_unlocked(fdp, fd,
723 cap_rights_init(&rights, CAP_FLOCK), &fp, NULL);
726 if (fp->f_type != DTYPE_VNODE) {
731 flp = (struct flock *)arg;
732 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
733 flp->l_type != F_UNLCK) {
738 if (flp->l_whence == SEEK_CUR) {
739 foffset = foffset_get(fp);
740 if ((flp->l_start > 0 &&
741 foffset > OFF_MAX - flp->l_start) ||
743 foffset < OFF_MIN - flp->l_start)) {
748 flp->l_start += foffset;
751 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
757 arg = arg ? 128 * 1024: 0;
760 error = fget_unlocked(fdp, fd,
761 cap_rights_init(&rights), &fp, NULL);
764 if (fp->f_type != DTYPE_VNODE) {
771 * Exclusive lock synchronizes against f_seqcount reads and
772 * writes in sequential_heuristic().
774 error = vn_lock(vp, LK_EXCLUSIVE);
780 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
781 fp->f_seqcount = (arg + bsize - 1) / bsize;
782 atomic_set_int(&fp->f_flag, FRDAHEAD);
784 atomic_clear_int(&fp->f_flag, FRDAHEAD);
798 getmaxfd(struct thread *td)
801 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
805 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
808 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
810 struct filedesc *fdp;
811 struct filedescent *oldfde, *newfde;
819 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
820 MPASS(mode < FDDUP_LASTMODE);
823 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
826 * Verify we have a valid descriptor to dup from and possibly to
827 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
828 * return EINVAL when the new descriptor is out of bounds.
833 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
834 maxfd = getmaxfd(td);
836 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
840 if (fget_locked(fdp, old) == NULL)
842 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
843 td->td_retval[0] = new;
844 if (flags & FDDUP_FLAG_CLOEXEC)
845 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
851 * If the caller specified a file descriptor, make sure the file
852 * table is large enough to hold it, and grab it. Otherwise, just
853 * allocate a new descriptor the usual way.
858 if ((error = fdalloc(td, new, &new)) != 0)
861 case FDDUP_MUSTREPLACE:
862 /* Target file descriptor must exist. */
863 if (fget_locked(fdp, new) == NULL)
867 if (new >= fdp->fd_nfiles) {
869 * The resource limits are here instead of e.g.
870 * fdalloc(), because the file descriptor table may be
871 * shared between processes, so we can't really use
872 * racct_add()/racct_sub(). Instead of counting the
873 * number of actually allocated descriptors, just put
874 * the limit on the size of the file descriptor table.
879 error = racct_set(p, RACCT_NOFILE, new + 1);
887 fdgrowtable_exp(fdp, new + 1);
889 if (!fdisused(fdp, new))
893 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
896 KASSERT(old != new, ("new fd is same as old"));
898 oldfde = &fdp->fd_ofiles[old];
899 fhold(oldfde->fde_file);
900 newfde = &fdp->fd_ofiles[new];
901 delfp = newfde->fde_file;
904 * Duplicate the source descriptor.
907 seq_write_begin(&newfde->fde_seq);
909 filecaps_free(&newfde->fde_caps);
910 memcpy(newfde, oldfde, fde_change_size);
911 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps, true);
912 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
913 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
915 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
917 seq_write_end(&newfde->fde_seq);
919 td->td_retval[0] = new;
924 (void) closefp(fdp, new, delfp, td, 1);
925 FILEDESC_UNLOCK_ASSERT(fdp);
928 FILEDESC_XUNLOCK(fdp);
935 * If sigio is on the list associated with a process or process group,
936 * disable signalling from the device, remove sigio from the list and
940 funsetown(struct sigio **sigiop)
952 *(sigio->sio_myref) = NULL;
953 if ((sigio)->sio_pgid < 0) {
954 struct pgrp *pg = (sigio)->sio_pgrp;
956 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
960 struct proc *p = (sigio)->sio_proc;
962 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
967 crfree(sigio->sio_ucred);
968 free(sigio, M_SIGIO);
972 * Free a list of sigio structures.
973 * We only need to lock the SIGIO_LOCK because we have made ourselves
974 * inaccessible to callers of fsetown and therefore do not need to lock
975 * the proc or pgrp struct for the list manipulation.
978 funsetownlst(struct sigiolst *sigiolst)
984 sigio = SLIST_FIRST(sigiolst);
991 * Every entry of the list should belong
992 * to a single proc or pgrp.
994 if (sigio->sio_pgid < 0) {
995 pg = sigio->sio_pgrp;
996 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
997 } else /* if (sigio->sio_pgid > 0) */ {
999 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1003 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1004 *(sigio->sio_myref) = NULL;
1006 KASSERT(sigio->sio_pgid < 0,
1007 ("Proc sigio in pgrp sigio list"));
1008 KASSERT(sigio->sio_pgrp == pg,
1009 ("Bogus pgrp in sigio list"));
1011 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1014 } else /* if (p != NULL) */ {
1015 KASSERT(sigio->sio_pgid > 0,
1016 ("Pgrp sigio in proc sigio list"));
1017 KASSERT(sigio->sio_proc == p,
1018 ("Bogus proc in sigio list"));
1020 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1025 crfree(sigio->sio_ucred);
1026 free(sigio, M_SIGIO);
1033 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1035 * After permission checking, add a sigio structure to the sigio list for
1036 * the process or process group.
1039 fsetown(pid_t pgid, struct sigio **sigiop)
1043 struct sigio *sigio;
1053 /* Allocate and fill in the new sigio out of locks. */
1054 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1055 sigio->sio_pgid = pgid;
1056 sigio->sio_ucred = crhold(curthread->td_ucred);
1057 sigio->sio_myref = sigiop;
1059 sx_slock(&proctree_lock);
1068 * Policy - Don't allow a process to FSETOWN a process
1069 * in another session.
1071 * Remove this test to allow maximum flexibility or
1072 * restrict FSETOWN to the current process or process
1073 * group for maximum safety.
1076 if (proc->p_session != curthread->td_proc->p_session) {
1082 } else /* if (pgid < 0) */ {
1083 pgrp = pgfind(-pgid);
1091 * Policy - Don't allow a process to FSETOWN a process
1092 * in another session.
1094 * Remove this test to allow maximum flexibility or
1095 * restrict FSETOWN to the current process or process
1096 * group for maximum safety.
1098 if (pgrp->pg_session != curthread->td_proc->p_session) {
1109 * Since funsetownlst() is called without the proctree
1110 * locked, we need to check for P_WEXIT.
1111 * XXX: is ESRCH correct?
1113 if ((proc->p_flag & P_WEXIT) != 0) {
1118 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1119 sigio->sio_proc = proc;
1123 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1124 sigio->sio_pgrp = pgrp;
1127 sx_sunlock(&proctree_lock);
1134 sx_sunlock(&proctree_lock);
1135 crfree(sigio->sio_ucred);
1136 free(sigio, M_SIGIO);
1141 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1145 struct sigio **sigiop;
1150 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1156 * Function drops the filedesc lock on return.
1159 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1164 FILEDESC_XLOCK_ASSERT(fdp);
1167 if (td->td_proc->p_fdtol != NULL) {
1169 * Ask fdfree() to sleep to ensure that all relevant
1170 * process leaders can be traversed in closef().
1172 fdp->fd_holdleaderscount++;
1179 * We now hold the fp reference that used to be owned by the
1180 * descriptor array. We have to unlock the FILEDESC *AFTER*
1181 * knote_fdclose to prevent a race of the fd getting opened, a knote
1182 * added, and deleteing a knote for the new fd.
1184 knote_fdclose(td, fd);
1187 * We need to notify mqueue if the object is of type mqueue.
1189 if (fp->f_type == DTYPE_MQUEUE)
1190 mq_fdclose(td, fd, fp);
1191 FILEDESC_XUNLOCK(fdp);
1193 error = closef(fp, td);
1195 FILEDESC_XLOCK(fdp);
1196 fdp->fd_holdleaderscount--;
1197 if (fdp->fd_holdleaderscount == 0 &&
1198 fdp->fd_holdleaderswakeup != 0) {
1199 fdp->fd_holdleaderswakeup = 0;
1200 wakeup(&fdp->fd_holdleaderscount);
1202 FILEDESC_XUNLOCK(fdp);
1208 * Close a file descriptor.
1210 #ifndef _SYS_SYSPROTO_H_
1217 sys_close(struct thread *td, struct close_args *uap)
1220 return (kern_close(td, uap->fd));
1224 kern_close(struct thread *td, int fd)
1226 struct filedesc *fdp;
1229 fdp = td->td_proc->p_fd;
1231 AUDIT_SYSCLOSE(td, fd);
1233 FILEDESC_XLOCK(fdp);
1234 if ((fp = fget_locked(fdp, fd)) == NULL) {
1235 FILEDESC_XUNLOCK(fdp);
1240 /* closefp() drops the FILEDESC lock for us. */
1241 return (closefp(fdp, fd, fp, td, 1));
1245 * Close open file descriptors.
1247 #ifndef _SYS_SYSPROTO_H_
1248 struct closefrom_args {
1254 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1256 struct filedesc *fdp;
1259 fdp = td->td_proc->p_fd;
1260 AUDIT_ARG_FD(uap->lowfd);
1263 * Treat negative starting file descriptor values identical to
1264 * closefrom(0) which closes all files.
1268 FILEDESC_SLOCK(fdp);
1269 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1270 if (fdp->fd_ofiles[fd].fde_file != NULL) {
1271 FILEDESC_SUNLOCK(fdp);
1272 (void)kern_close(td, fd);
1273 FILEDESC_SLOCK(fdp);
1276 FILEDESC_SUNLOCK(fdp);
1280 #if defined(COMPAT_43)
1282 * Return status information about a file descriptor.
1284 #ifndef _SYS_SYSPROTO_H_
1285 struct ofstat_args {
1292 ofstat(struct thread *td, struct ofstat_args *uap)
1298 error = kern_fstat(td, uap->fd, &ub);
1301 error = copyout(&oub, uap->sb, sizeof(oub));
1305 #endif /* COMPAT_43 */
1308 * Return status information about a file descriptor.
1310 #ifndef _SYS_SYSPROTO_H_
1318 sys_fstat(struct thread *td, struct fstat_args *uap)
1323 error = kern_fstat(td, uap->fd, &ub);
1325 error = copyout(&ub, uap->sb, sizeof(ub));
1330 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1333 cap_rights_t rights;
1338 error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp);
1342 AUDIT_ARG_FILE(td->td_proc, fp);
1344 error = fo_stat(fp, sbp, td->td_ucred, td);
1347 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1354 * Return status information about a file descriptor.
1356 #ifndef _SYS_SYSPROTO_H_
1357 struct nfstat_args {
1364 sys_nfstat(struct thread *td, struct nfstat_args *uap)
1370 error = kern_fstat(td, uap->fd, &ub);
1372 cvtnstat(&ub, &nub);
1373 error = copyout(&nub, uap->sb, sizeof(nub));
1379 * Return pathconf information about a file descriptor.
1381 #ifndef _SYS_SYSPROTO_H_
1382 struct fpathconf_args {
1389 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1393 cap_rights_t rights;
1396 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp);
1400 if (uap->name == _PC_ASYNC_IO) {
1401 td->td_retval[0] = _POSIX_ASYNCHRONOUS_IO;
1406 vn_lock(vp, LK_SHARED | LK_RETRY);
1407 error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1409 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1410 if (uap->name != _PC_PIPE_BUF) {
1413 td->td_retval[0] = PIPE_BUF;
1425 * Initialize filecaps structure.
1428 filecaps_init(struct filecaps *fcaps)
1431 bzero(fcaps, sizeof(*fcaps));
1432 fcaps->fc_nioctls = -1;
1436 * Copy filecaps structure allocating memory for ioctls array if needed.
1438 * The last parameter indicates whether the fdtable is locked. If it is not and
1439 * ioctls are encountered, copying fails and the caller must lock the table.
1441 * Note that if the table was not locked, the caller has to check the relevant
1442 * sequence counter to determine whether the operation was successful.
1445 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1450 if (src->fc_ioctls == NULL)
1455 KASSERT(src->fc_nioctls > 0,
1456 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1458 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1459 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1460 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1465 * Move filecaps structure to the new place and clear the old place.
1468 filecaps_move(struct filecaps *src, struct filecaps *dst)
1472 bzero(src, sizeof(*src));
1476 * Fill the given filecaps structure with full rights.
1479 filecaps_fill(struct filecaps *fcaps)
1482 CAP_ALL(&fcaps->fc_rights);
1483 fcaps->fc_ioctls = NULL;
1484 fcaps->fc_nioctls = -1;
1485 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1489 * Free memory allocated within filecaps structure.
1492 filecaps_free(struct filecaps *fcaps)
1495 free(fcaps->fc_ioctls, M_FILECAPS);
1496 bzero(fcaps, sizeof(*fcaps));
1500 * Validate the given filecaps structure.
1503 filecaps_validate(const struct filecaps *fcaps, const char *func)
1506 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1507 ("%s: invalid rights", func));
1508 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1509 ("%s: invalid fcntls", func));
1510 KASSERT(fcaps->fc_fcntls == 0 ||
1511 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1512 ("%s: fcntls without CAP_FCNTL", func));
1513 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1514 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1515 ("%s: invalid ioctls", func));
1516 KASSERT(fcaps->fc_nioctls == 0 ||
1517 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1518 ("%s: ioctls without CAP_IOCTL", func));
1522 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1526 FILEDESC_XLOCK_ASSERT(fdp);
1528 nfd1 = fdp->fd_nfiles * 2;
1531 fdgrowtable(fdp, nfd1);
1535 * Grow the file table to accommodate (at least) nfd descriptors.
1538 fdgrowtable(struct filedesc *fdp, int nfd)
1540 struct filedesc0 *fdp0;
1541 struct freetable *ft;
1542 struct fdescenttbl *ntable;
1543 struct fdescenttbl *otable;
1544 int nnfiles, onfiles;
1545 NDSLOTTYPE *nmap, *omap;
1548 * If lastfile is -1 this struct filedesc was just allocated and we are
1549 * growing it to accommodate for the one we are going to copy from. There
1550 * is no need to have a lock on this one as it's not visible to anyone.
1552 if (fdp->fd_lastfile != -1)
1553 FILEDESC_XLOCK_ASSERT(fdp);
1555 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1557 /* save old values */
1558 onfiles = fdp->fd_nfiles;
1559 otable = fdp->fd_files;
1562 /* compute the size of the new table */
1563 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1564 if (nnfiles <= onfiles)
1565 /* the table is already large enough */
1569 * Allocate a new table. We need enough space for the number of
1570 * entries, file entries themselves and the struct freetable we will use
1571 * when we decommission the table and place it on the freelist.
1572 * We place the struct freetable in the middle so we don't have
1573 * to worry about padding.
1575 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1576 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1577 sizeof(struct freetable),
1578 M_FILEDESC, M_ZERO | M_WAITOK);
1579 /* copy the old data */
1580 ntable->fdt_nfiles = nnfiles;
1581 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1582 onfiles * sizeof(ntable->fdt_ofiles[0]));
1585 * Allocate a new map only if the old is not large enough. It will
1586 * grow at a slower rate than the table as it can map more
1587 * entries than the table can hold.
1589 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1590 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1592 /* copy over the old data and update the pointer */
1593 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1598 * Make sure that ntable is correctly initialized before we replace
1599 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1602 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1605 * Do not free the old file table, as some threads may still
1606 * reference entries within it. Instead, place it on a freelist
1607 * which will be processed when the struct filedesc is released.
1609 * Note that if onfiles == NDFILE, we're dealing with the original
1610 * static allocation contained within (struct filedesc0 *)fdp,
1611 * which must not be freed.
1613 if (onfiles > NDFILE) {
1614 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1615 fdp0 = (struct filedesc0 *)fdp;
1616 ft->ft_table = otable;
1617 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1620 * The map does not have the same possibility of threads still
1621 * holding references to it. So always free it as long as it
1622 * does not reference the original static allocation.
1624 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1625 free(omap, M_FILEDESC);
1629 * Allocate a file descriptor for the process.
1632 fdalloc(struct thread *td, int minfd, int *result)
1634 struct proc *p = td->td_proc;
1635 struct filedesc *fdp = p->p_fd;
1636 int fd, maxfd, allocfd;
1641 FILEDESC_XLOCK_ASSERT(fdp);
1643 if (fdp->fd_freefile > minfd)
1644 minfd = fdp->fd_freefile;
1646 maxfd = getmaxfd(td);
1649 * Search the bitmap for a free descriptor starting at minfd.
1650 * If none is found, grow the file table.
1652 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1655 if (fd >= fdp->fd_nfiles) {
1656 allocfd = min(fd * 2, maxfd);
1660 error = racct_set(p, RACCT_NOFILE, allocfd);
1667 * fd is already equal to first free descriptor >= minfd, so
1668 * we only need to grow the table and we are done.
1670 fdgrowtable_exp(fdp, allocfd);
1674 * Perform some sanity checks, then mark the file descriptor as
1675 * used and return it to the caller.
1677 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1678 ("invalid descriptor %d", fd));
1679 KASSERT(!fdisused(fdp, fd),
1680 ("fd_first_free() returned non-free descriptor"));
1681 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1682 ("file descriptor isn't free"));
1689 * Allocate n file descriptors for the process.
1692 fdallocn(struct thread *td, int minfd, int *fds, int n)
1694 struct proc *p = td->td_proc;
1695 struct filedesc *fdp = p->p_fd;
1698 FILEDESC_XLOCK_ASSERT(fdp);
1700 for (i = 0; i < n; i++)
1701 if (fdalloc(td, 0, &fds[i]) != 0)
1705 for (i--; i >= 0; i--)
1706 fdunused(fdp, fds[i]);
1714 * Create a new open file structure and allocate a file descriptor for the
1715 * process that refers to it. We add one reference to the file for the
1716 * descriptor table and one reference for resultfp. This is to prevent us
1717 * being preempted and the entry in the descriptor table closed after we
1718 * release the FILEDESC lock.
1721 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1722 struct filecaps *fcaps)
1727 error = falloc_noinstall(td, &fp);
1729 return (error); /* no reference held on error */
1731 error = finstall(td, fp, &fd, flags, fcaps);
1733 fdrop(fp, td); /* one reference (fp only) */
1737 if (resultfp != NULL)
1738 *resultfp = fp; /* copy out result */
1740 fdrop(fp, td); /* release local reference */
1742 if (resultfd != NULL)
1749 * Create a new open file structure without allocating a file descriptor.
1752 falloc_noinstall(struct thread *td, struct file **resultfp)
1755 int maxuserfiles = maxfiles - (maxfiles / 20);
1757 static struct timeval lastfail;
1760 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1762 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1763 if ((openfiles_new >= maxuserfiles &&
1764 priv_check(td, PRIV_MAXFILES) != 0) ||
1765 openfiles_new >= maxfiles) {
1766 atomic_subtract_int(&openfiles, 1);
1767 if (ppsratecheck(&lastfail, &curfail, 1)) {
1768 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1769 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1773 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1774 refcount_init(&fp->f_count, 1);
1775 fp->f_cred = crhold(td->td_ucred);
1776 fp->f_ops = &badfileops;
1782 * Install a file in a file descriptor table.
1785 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1786 struct filecaps *fcaps)
1788 struct filedescent *fde;
1792 filecaps_validate(fcaps, __func__);
1793 FILEDESC_XLOCK_ASSERT(fdp);
1795 fde = &fdp->fd_ofiles[fd];
1797 seq_write_begin(&fde->fde_seq);
1800 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1802 filecaps_move(fcaps, &fde->fde_caps);
1804 filecaps_fill(&fde->fde_caps);
1806 seq_write_end(&fde->fde_seq);
1811 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1812 struct filecaps *fcaps)
1814 struct filedesc *fdp = td->td_proc->p_fd;
1819 FILEDESC_XLOCK(fdp);
1820 if ((error = fdalloc(td, 0, fd))) {
1821 FILEDESC_XUNLOCK(fdp);
1825 _finstall(fdp, fp, *fd, flags, fcaps);
1826 FILEDESC_XUNLOCK(fdp);
1831 * Build a new filedesc structure from another.
1832 * Copy the current, root, and jail root vnode references.
1834 * If fdp is not NULL, return with it shared locked.
1837 fdinit(struct filedesc *fdp, bool prepfiles)
1839 struct filedesc0 *newfdp0;
1840 struct filedesc *newfdp;
1842 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1843 newfdp = &newfdp0->fd_fd;
1845 /* Create the file descriptor table. */
1846 FILEDESC_LOCK_INIT(newfdp);
1847 refcount_init(&newfdp->fd_refcnt, 1);
1848 refcount_init(&newfdp->fd_holdcnt, 1);
1849 newfdp->fd_cmask = CMASK;
1850 newfdp->fd_map = newfdp0->fd_dmap;
1851 newfdp->fd_lastfile = -1;
1852 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1853 newfdp->fd_files->fdt_nfiles = NDFILE;
1858 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1859 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1861 FILEDESC_SLOCK(fdp);
1862 newfdp->fd_cdir = fdp->fd_cdir;
1863 if (newfdp->fd_cdir)
1864 vrefact(newfdp->fd_cdir);
1865 newfdp->fd_rdir = fdp->fd_rdir;
1866 if (newfdp->fd_rdir)
1867 vrefact(newfdp->fd_rdir);
1868 newfdp->fd_jdir = fdp->fd_jdir;
1869 if (newfdp->fd_jdir)
1870 vrefact(newfdp->fd_jdir);
1873 FILEDESC_SUNLOCK(fdp);
1875 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1876 FILEDESC_SUNLOCK(fdp);
1877 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1878 FILEDESC_SLOCK(fdp);
1885 static struct filedesc *
1886 fdhold(struct proc *p)
1888 struct filedesc *fdp;
1890 PROC_LOCK_ASSERT(p, MA_OWNED);
1893 refcount_acquire(&fdp->fd_holdcnt);
1898 fddrop(struct filedesc *fdp)
1901 if (fdp->fd_holdcnt > 1) {
1902 if (refcount_release(&fdp->fd_holdcnt) == 0)
1906 FILEDESC_LOCK_DESTROY(fdp);
1907 uma_zfree(filedesc0_zone, fdp);
1911 * Share a filedesc structure.
1914 fdshare(struct filedesc *fdp)
1917 refcount_acquire(&fdp->fd_refcnt);
1922 * Unshare a filedesc structure, if necessary by making a copy
1925 fdunshare(struct thread *td)
1927 struct filedesc *tmp;
1928 struct proc *p = td->td_proc;
1930 if (p->p_fd->fd_refcnt == 1)
1933 tmp = fdcopy(p->p_fd);
1939 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
1943 td->td_proc->p_fd = fdp;
1947 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
1948 * this is to ease callers, not catch errors.
1951 fdcopy(struct filedesc *fdp)
1953 struct filedesc *newfdp;
1954 struct filedescent *nfde, *ofde;
1959 newfdp = fdinit(fdp, true);
1960 /* copy all passable descriptors (i.e. not kqueue) */
1961 newfdp->fd_freefile = -1;
1962 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1963 ofde = &fdp->fd_ofiles[i];
1964 if (ofde->fde_file == NULL ||
1965 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
1966 if (newfdp->fd_freefile == -1)
1967 newfdp->fd_freefile = i;
1970 nfde = &newfdp->fd_ofiles[i];
1972 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
1973 fhold(nfde->fde_file);
1974 fdused_init(newfdp, i);
1975 newfdp->fd_lastfile = i;
1977 if (newfdp->fd_freefile == -1)
1978 newfdp->fd_freefile = i;
1979 newfdp->fd_cmask = fdp->fd_cmask;
1980 FILEDESC_SUNLOCK(fdp);
1985 * Copies a filedesc structure, while remapping all file descriptors
1986 * stored inside using a translation table.
1988 * File descriptors are copied over to the new file descriptor table,
1989 * regardless of whether the close-on-exec flag is set.
1992 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
1993 struct filedesc **ret)
1995 struct filedesc *newfdp;
1996 struct filedescent *nfde, *ofde;
2001 newfdp = fdinit(fdp, true);
2002 if (nfds > fdp->fd_lastfile + 1) {
2003 /* New table cannot be larger than the old one. */
2007 /* Copy all passable descriptors (i.e. not kqueue). */
2008 newfdp->fd_freefile = nfds;
2009 for (i = 0; i < nfds; ++i) {
2010 if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2011 /* File descriptor out of bounds. */
2015 ofde = &fdp->fd_ofiles[fds[i]];
2016 if (ofde->fde_file == NULL) {
2017 /* Unused file descriptor. */
2021 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2022 /* File descriptor cannot be passed. */
2026 nfde = &newfdp->fd_ofiles[i];
2028 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2029 fhold(nfde->fde_file);
2030 fdused_init(newfdp, i);
2031 newfdp->fd_lastfile = i;
2033 newfdp->fd_cmask = fdp->fd_cmask;
2034 FILEDESC_SUNLOCK(fdp);
2038 FILEDESC_SUNLOCK(fdp);
2039 fdescfree_remapped(newfdp);
2044 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2045 * one of processes using it exits) and the table used to be shared.
2048 fdclearlocks(struct thread *td)
2050 struct filedesc *fdp;
2051 struct filedesc_to_leader *fdtol;
2061 MPASS(fdtol != NULL);
2063 FILEDESC_XLOCK(fdp);
2064 KASSERT(fdtol->fdl_refcount > 0,
2065 ("filedesc_to_refcount botch: fdl_refcount=%d",
2066 fdtol->fdl_refcount));
2067 if (fdtol->fdl_refcount == 1 &&
2068 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2069 for (i = 0; i <= fdp->fd_lastfile; i++) {
2070 fp = fdp->fd_ofiles[i].fde_file;
2071 if (fp == NULL || fp->f_type != DTYPE_VNODE)
2074 FILEDESC_XUNLOCK(fdp);
2075 lf.l_whence = SEEK_SET;
2078 lf.l_type = F_UNLCK;
2080 (void) VOP_ADVLOCK(vp,
2081 (caddr_t)p->p_leader, F_UNLCK,
2083 FILEDESC_XLOCK(fdp);
2088 if (fdtol->fdl_refcount == 1) {
2089 if (fdp->fd_holdleaderscount > 0 &&
2090 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2092 * close() or kern_dup() has cleared a reference
2093 * in a shared file descriptor table.
2095 fdp->fd_holdleaderswakeup = 1;
2096 sx_sleep(&fdp->fd_holdleaderscount,
2097 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2100 if (fdtol->fdl_holdcount > 0) {
2102 * Ensure that fdtol->fdl_leader remains
2103 * valid in closef().
2105 fdtol->fdl_wakeup = 1;
2106 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2111 fdtol->fdl_refcount--;
2112 if (fdtol->fdl_refcount == 0 &&
2113 fdtol->fdl_holdcount == 0) {
2114 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2115 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2119 FILEDESC_XUNLOCK(fdp);
2121 free(fdtol, M_FILEDESC_TO_LEADER);
2125 * Release a filedesc structure.
2128 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2130 struct filedesc0 *fdp0;
2131 struct freetable *ft, *tft;
2132 struct filedescent *fde;
2136 for (i = 0; i <= fdp->fd_lastfile; i++) {
2137 fde = &fdp->fd_ofiles[i];
2142 (void) closef(fp, td);
2148 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2149 free(fdp->fd_map, M_FILEDESC);
2150 if (fdp->fd_nfiles > NDFILE)
2151 free(fdp->fd_files, M_FILEDESC);
2153 fdp0 = (struct filedesc0 *)fdp;
2154 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2155 free(ft->ft_table, M_FILEDESC);
2161 fdescfree(struct thread *td)
2164 struct filedesc *fdp;
2165 struct vnode *cdir, *jdir, *rdir;
2174 racct_set(p, RACCT_NOFILE, 0);
2179 if (p->p_fdtol != NULL)
2186 if (refcount_release(&fdp->fd_refcnt) == 0)
2189 FILEDESC_XLOCK(fdp);
2190 cdir = fdp->fd_cdir;
2191 fdp->fd_cdir = NULL;
2192 rdir = fdp->fd_rdir;
2193 fdp->fd_rdir = NULL;
2194 jdir = fdp->fd_jdir;
2195 fdp->fd_jdir = NULL;
2196 FILEDESC_XUNLOCK(fdp);
2205 fdescfree_fds(td, fdp, 1);
2209 fdescfree_remapped(struct filedesc *fdp)
2212 if (fdp->fd_cdir != NULL)
2213 vrele(fdp->fd_cdir);
2214 if (fdp->fd_rdir != NULL)
2215 vrele(fdp->fd_rdir);
2216 if (fdp->fd_jdir != NULL)
2217 vrele(fdp->fd_jdir);
2219 fdescfree_fds(curthread, fdp, 0);
2223 * For setugid programs, we don't want to people to use that setugidness
2224 * to generate error messages which write to a file which otherwise would
2225 * otherwise be off-limits to the process. We check for filesystems where
2226 * the vnode can change out from under us after execve (like [lin]procfs).
2228 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2229 * sufficient. We also don't check for setugidness since we know we are.
2232 is_unsafe(struct file *fp)
2236 if (fp->f_type != DTYPE_VNODE)
2240 return ((vp->v_vflag & VV_PROCDEP) != 0);
2244 * Make this setguid thing safe, if at all possible.
2247 fdsetugidsafety(struct thread *td)
2249 struct filedesc *fdp;
2253 fdp = td->td_proc->p_fd;
2254 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2255 MPASS(fdp->fd_nfiles >= 3);
2256 for (i = 0; i <= 2; i++) {
2257 fp = fdp->fd_ofiles[i].fde_file;
2258 if (fp != NULL && is_unsafe(fp)) {
2259 FILEDESC_XLOCK(fdp);
2260 knote_fdclose(td, i);
2262 * NULL-out descriptor prior to close to avoid
2263 * a race while close blocks.
2266 FILEDESC_XUNLOCK(fdp);
2267 (void) closef(fp, td);
2273 * If a specific file object occupies a specific file descriptor, close the
2274 * file descriptor entry and drop a reference on the file object. This is a
2275 * convenience function to handle a subsequent error in a function that calls
2276 * falloc() that handles the race that another thread might have closed the
2277 * file descriptor out from under the thread creating the file object.
2280 fdclose(struct thread *td, struct file *fp, int idx)
2282 struct filedesc *fdp = td->td_proc->p_fd;
2284 FILEDESC_XLOCK(fdp);
2285 if (fdp->fd_ofiles[idx].fde_file == fp) {
2287 FILEDESC_XUNLOCK(fdp);
2290 FILEDESC_XUNLOCK(fdp);
2294 * Close any files on exec?
2297 fdcloseexec(struct thread *td)
2299 struct filedesc *fdp;
2300 struct filedescent *fde;
2304 fdp = td->td_proc->p_fd;
2305 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2306 for (i = 0; i <= fdp->fd_lastfile; i++) {
2307 fde = &fdp->fd_ofiles[i];
2309 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2310 (fde->fde_flags & UF_EXCLOSE))) {
2311 FILEDESC_XLOCK(fdp);
2313 (void) closefp(fdp, i, fp, td, 0);
2314 FILEDESC_UNLOCK_ASSERT(fdp);
2320 * It is unsafe for set[ug]id processes to be started with file
2321 * descriptors 0..2 closed, as these descriptors are given implicit
2322 * significance in the Standard C library. fdcheckstd() will create a
2323 * descriptor referencing /dev/null for each of stdin, stdout, and
2324 * stderr that is not already open.
2327 fdcheckstd(struct thread *td)
2329 struct filedesc *fdp;
2331 int i, error, devnull;
2333 fdp = td->td_proc->p_fd;
2334 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2335 MPASS(fdp->fd_nfiles >= 3);
2337 for (i = 0; i <= 2; i++) {
2338 if (fdp->fd_ofiles[i].fde_file != NULL)
2341 save = td->td_retval[0];
2342 if (devnull != -1) {
2343 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2345 error = kern_openat(td, AT_FDCWD, "/dev/null",
2346 UIO_SYSSPACE, O_RDWR, 0);
2348 devnull = td->td_retval[0];
2349 KASSERT(devnull == i, ("we didn't get our fd"));
2352 td->td_retval[0] = save;
2360 * Internal form of close. Decrement reference count on file structure.
2361 * Note: td may be NULL when closing a file that was being passed in a
2364 * XXXRW: Giant is not required for the caller, but often will be held; this
2365 * makes it moderately likely the Giant will be recursed in the VFS case.
2368 closef(struct file *fp, struct thread *td)
2372 struct filedesc_to_leader *fdtol;
2373 struct filedesc *fdp;
2376 * POSIX record locking dictates that any close releases ALL
2377 * locks owned by this process. This is handled by setting
2378 * a flag in the unlock to free ONLY locks obeying POSIX
2379 * semantics, and not to free BSD-style file locks.
2380 * If the descriptor was in a message, POSIX-style locks
2381 * aren't passed with the descriptor, and the thread pointer
2382 * will be NULL. Callers should be careful only to pass a
2383 * NULL thread pointer when there really is no owning
2384 * context that might have locks, or the locks will be
2387 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2389 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2390 lf.l_whence = SEEK_SET;
2393 lf.l_type = F_UNLCK;
2394 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2395 F_UNLCK, &lf, F_POSIX);
2397 fdtol = td->td_proc->p_fdtol;
2398 if (fdtol != NULL) {
2400 * Handle special case where file descriptor table is
2401 * shared between multiple process leaders.
2403 fdp = td->td_proc->p_fd;
2404 FILEDESC_XLOCK(fdp);
2405 for (fdtol = fdtol->fdl_next;
2406 fdtol != td->td_proc->p_fdtol;
2407 fdtol = fdtol->fdl_next) {
2408 if ((fdtol->fdl_leader->p_flag &
2411 fdtol->fdl_holdcount++;
2412 FILEDESC_XUNLOCK(fdp);
2413 lf.l_whence = SEEK_SET;
2416 lf.l_type = F_UNLCK;
2418 (void) VOP_ADVLOCK(vp,
2419 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2421 FILEDESC_XLOCK(fdp);
2422 fdtol->fdl_holdcount--;
2423 if (fdtol->fdl_holdcount == 0 &&
2424 fdtol->fdl_wakeup != 0) {
2425 fdtol->fdl_wakeup = 0;
2429 FILEDESC_XUNLOCK(fdp);
2432 return (fdrop(fp, td));
2436 * Initialize the file pointer with the specified properties.
2438 * The ops are set with release semantics to be certain that the flags, type,
2439 * and data are visible when ops is. This is to prevent ops methods from being
2440 * called with bad data.
2443 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2448 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2452 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2453 struct file **fpp, struct filecaps *havecapsp)
2455 struct filedescent *fde;
2458 FILEDESC_LOCK_ASSERT(fdp);
2460 fde = fdeget_locked(fdp, fd);
2467 error = cap_check(cap_rights_fde(fde), needrightsp);
2472 if (havecapsp != NULL)
2473 filecaps_copy(&fde->fde_caps, havecapsp, true);
2475 *fpp = fde->fde_file;
2483 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2484 struct file **fpp, struct filecaps *havecapsp)
2486 struct filedesc *fdp = td->td_proc->p_fd;
2488 #ifndef CAPABILITIES
2489 error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2490 if (error == 0 && havecapsp != NULL)
2491 filecaps_fill(havecapsp);
2497 error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2501 if (havecapsp != NULL) {
2502 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2503 havecapsp, false)) {
2509 if (!fd_modified(fdp, fd, seq))
2518 FILEDESC_SLOCK(fdp);
2519 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2522 FILEDESC_SUNLOCK(fdp);
2528 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2529 struct file **fpp, seq_t *seqp)
2532 struct filedescent *fde;
2534 struct fdescenttbl *fdt;
2539 cap_rights_t haverights;
2543 fdt = fdp->fd_files;
2544 if ((u_int)fd >= fdt->fdt_nfiles)
2547 * Fetch the descriptor locklessly. We avoid fdrop() races by
2548 * never raising a refcount above 0. To accomplish this we have
2549 * to use a cmpset loop rather than an atomic_add. The descriptor
2550 * must be re-verified once we acquire a reference to be certain
2551 * that the identity is still correct and we did not lose a race
2552 * due to preemption.
2556 seq = seq_read(fd_seq(fdt, fd));
2557 fde = &fdt->fdt_ofiles[fd];
2558 haverights = *cap_rights_fde(fde);
2560 if (!seq_consistent(fd_seq(fdt, fd), seq))
2563 fp = fdt->fdt_ofiles[fd].fde_file;
2568 error = cap_check(&haverights, needrightsp);
2573 count = fp->f_count;
2576 * Force a reload. Other thread could reallocate the
2577 * table before this fd was closed, so it possible that
2578 * there is a stale fp pointer in cached version.
2580 fdt = *(struct fdescenttbl * volatile *)&(fdp->fd_files);
2584 * Use an acquire barrier to force re-reading of fdt so it is
2585 * refreshed for verification.
2587 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) == 0)
2589 fdt = fdp->fd_files;
2591 if (seq_consistent_nomb(fd_seq(fdt, fd), seq))
2593 if (fp == fdt->fdt_ofiles[fd].fde_file)
2596 fdrop(fp, curthread);
2608 * Extract the file pointer associated with the specified descriptor for the
2609 * current user process.
2611 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2614 * File's rights will be checked against the capability rights mask.
2616 * If an error occurred the non-zero error is returned and *fpp is set to
2617 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
2618 * responsible for fdrop().
2621 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2622 cap_rights_t *needrightsp, seq_t *seqp)
2624 struct filedesc *fdp;
2629 fdp = td->td_proc->p_fd;
2630 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2633 if (fp->f_ops == &badfileops) {
2639 * FREAD and FWRITE failure return EBADF as per POSIX.
2645 if ((fp->f_flag & flags) == 0)
2649 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2650 ((fp->f_flag & FWRITE) != 0))
2656 KASSERT(0, ("wrong flags"));
2669 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2672 return (_fget(td, fd, fpp, 0, rightsp, NULL));
2676 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2680 #ifndef CAPABILITIES
2681 error = _fget(td, fd, fpp, 0, rightsp, NULL);
2682 if (maxprotp != NULL)
2683 *maxprotp = VM_PROT_ALL;
2685 struct filedesc *fdp = td->td_proc->p_fd;
2688 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2690 error = _fget(td, fd, fpp, 0, rightsp, &seq);
2694 * If requested, convert capability rights to access flags.
2696 if (maxprotp != NULL)
2697 *maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd));
2698 if (!fd_modified(fdp, fd, seq))
2707 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2710 return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2714 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2717 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2721 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2724 struct filedesc *fdp = td->td_proc->p_fd;
2725 #ifndef CAPABILITIES
2726 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2731 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2733 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2736 error = cap_fcntl_check(fdp, fd, needfcntl);
2737 if (!fd_modified(fdp, fd, seq))
2750 * Like fget() but loads the underlying vnode, or returns an error if the
2751 * descriptor does not represent a vnode. Note that pipes use vnodes but
2752 * never have VM objects. The returned vnode will be vref()'d.
2754 * XXX: what about the unused flags ?
2757 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2764 error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2767 if (fp->f_vnode == NULL) {
2779 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2782 return (_fgetvp(td, fd, 0, rightsp, vpp));
2786 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2787 struct filecaps *havecaps, struct vnode **vpp)
2789 struct filedesc *fdp;
2790 struct filecaps caps;
2794 fdp = td->td_proc->p_fd;
2795 error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2798 if (fp->f_ops == &badfileops) {
2802 if (fp->f_vnode == NULL) {
2813 filecaps_free(&caps);
2818 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2821 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2825 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2828 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2833 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2837 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2842 * Handle the last reference to a file being closed.
2845 _fdrop(struct file *fp, struct thread *td)
2849 if (fp->f_count != 0)
2850 panic("fdrop: count %d", fp->f_count);
2851 error = fo_close(fp, td);
2852 atomic_subtract_int(&openfiles, 1);
2854 free(fp->f_advice, M_FADVISE);
2855 uma_zfree(file_zone, fp);
2861 * Apply an advisory lock on a file descriptor.
2863 * Just attempt to get a record lock of the requested type on the entire file
2864 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2866 #ifndef _SYS_SYSPROTO_H_
2874 sys_flock(struct thread *td, struct flock_args *uap)
2879 cap_rights_t rights;
2882 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp);
2885 if (fp->f_type != DTYPE_VNODE) {
2887 return (EOPNOTSUPP);
2891 lf.l_whence = SEEK_SET;
2894 if (uap->how & LOCK_UN) {
2895 lf.l_type = F_UNLCK;
2896 atomic_clear_int(&fp->f_flag, FHASLOCK);
2897 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2900 if (uap->how & LOCK_EX)
2901 lf.l_type = F_WRLCK;
2902 else if (uap->how & LOCK_SH)
2903 lf.l_type = F_RDLCK;
2908 atomic_set_int(&fp->f_flag, FHASLOCK);
2909 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2910 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2916 * Duplicate the specified descriptor to a free descriptor.
2919 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
2920 int openerror, int *indxp)
2922 struct filedescent *newfde, *oldfde;
2926 KASSERT(openerror == ENODEV || openerror == ENXIO,
2927 ("unexpected error %d in %s", openerror, __func__));
2930 * If the to-be-dup'd fd number is greater than the allowed number
2931 * of file descriptors, or the fd to be dup'd has already been
2932 * closed, then reject.
2934 FILEDESC_XLOCK(fdp);
2935 if ((fp = fget_locked(fdp, dfd)) == NULL) {
2936 FILEDESC_XUNLOCK(fdp);
2940 error = fdalloc(td, 0, &indx);
2942 FILEDESC_XUNLOCK(fdp);
2947 * There are two cases of interest here.
2949 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2951 * For ENXIO steal away the file structure from (dfd) and store it in
2952 * (indx). (dfd) is effectively closed by this operation.
2954 switch (openerror) {
2957 * Check that the mode the file is being opened for is a
2958 * subset of the mode of the existing descriptor.
2960 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
2961 fdunused(fdp, indx);
2962 FILEDESC_XUNLOCK(fdp);
2966 newfde = &fdp->fd_ofiles[indx];
2967 oldfde = &fdp->fd_ofiles[dfd];
2969 seq_write_begin(&newfde->fde_seq);
2971 memcpy(newfde, oldfde, fde_change_size);
2972 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps, true);
2974 seq_write_end(&newfde->fde_seq);
2979 * Steal away the file pointer from dfd and stuff it into indx.
2981 newfde = &fdp->fd_ofiles[indx];
2982 oldfde = &fdp->fd_ofiles[dfd];
2984 seq_write_begin(&newfde->fde_seq);
2986 memcpy(newfde, oldfde, fde_change_size);
2987 oldfde->fde_file = NULL;
2990 seq_write_end(&newfde->fde_seq);
2994 FILEDESC_XUNLOCK(fdp);
3000 * This sysctl determines if we will allow a process to chroot(2) if it
3001 * has a directory open:
3002 * 0: disallowed for all processes.
3003 * 1: allowed for processes that were not already chroot(2)'ed.
3004 * 2: allowed for all processes.
3007 static int chroot_allow_open_directories = 1;
3009 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3010 &chroot_allow_open_directories, 0,
3011 "Allow a process to chroot(2) if it has a directory open");
3014 * Helper function for raised chroot(2) security function: Refuse if
3015 * any filedescriptors are open directories.
3018 chroot_refuse_vdir_fds(struct filedesc *fdp)
3024 FILEDESC_LOCK_ASSERT(fdp);
3026 for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3027 fp = fget_locked(fdp, fd);
3030 if (fp->f_type == DTYPE_VNODE) {
3032 if (vp->v_type == VDIR)
3040 * Common routine for kern_chroot() and jail_attach(). The caller is
3041 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3042 * authorize this operation.
3045 pwd_chroot(struct thread *td, struct vnode *vp)
3047 struct filedesc *fdp;
3048 struct vnode *oldvp;
3051 fdp = td->td_proc->p_fd;
3052 FILEDESC_XLOCK(fdp);
3053 if (chroot_allow_open_directories == 0 ||
3054 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3055 error = chroot_refuse_vdir_fds(fdp);
3057 FILEDESC_XUNLOCK(fdp);
3061 oldvp = fdp->fd_rdir;
3064 if (fdp->fd_jdir == NULL) {
3068 FILEDESC_XUNLOCK(fdp);
3074 pwd_chdir(struct thread *td, struct vnode *vp)
3076 struct filedesc *fdp;
3077 struct vnode *oldvp;
3079 fdp = td->td_proc->p_fd;
3080 FILEDESC_XLOCK(fdp);
3081 VNASSERT(vp->v_usecount > 0, vp,
3082 ("chdir to a vnode with zero usecount"));
3083 oldvp = fdp->fd_cdir;
3085 FILEDESC_XUNLOCK(fdp);
3090 * Scan all active processes and prisons to see if any of them have a current
3091 * or root directory of `olddp'. If so, replace them with the new mount point.
3094 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3096 struct filedesc *fdp;
3101 if (vrefcnt(olddp) == 1)
3104 sx_slock(&allproc_lock);
3105 FOREACH_PROC_IN_SYSTEM(p) {
3111 FILEDESC_XLOCK(fdp);
3112 if (fdp->fd_cdir == olddp) {
3114 fdp->fd_cdir = newdp;
3117 if (fdp->fd_rdir == olddp) {
3119 fdp->fd_rdir = newdp;
3122 if (fdp->fd_jdir == olddp) {
3124 fdp->fd_jdir = newdp;
3127 FILEDESC_XUNLOCK(fdp);
3130 sx_sunlock(&allproc_lock);
3131 if (rootvnode == olddp) {
3136 mtx_lock(&prison0.pr_mtx);
3137 if (prison0.pr_root == olddp) {
3139 prison0.pr_root = newdp;
3142 mtx_unlock(&prison0.pr_mtx);
3143 sx_slock(&allprison_lock);
3144 TAILQ_FOREACH(pr, &allprison, pr_list) {
3145 mtx_lock(&pr->pr_mtx);
3146 if (pr->pr_root == olddp) {
3148 pr->pr_root = newdp;
3151 mtx_unlock(&pr->pr_mtx);
3153 sx_sunlock(&allprison_lock);
3158 struct filedesc_to_leader *
3159 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3161 struct filedesc_to_leader *fdtol;
3163 fdtol = malloc(sizeof(struct filedesc_to_leader),
3164 M_FILEDESC_TO_LEADER, M_WAITOK);
3165 fdtol->fdl_refcount = 1;
3166 fdtol->fdl_holdcount = 0;
3167 fdtol->fdl_wakeup = 0;
3168 fdtol->fdl_leader = leader;
3170 FILEDESC_XLOCK(fdp);
3171 fdtol->fdl_next = old->fdl_next;
3172 fdtol->fdl_prev = old;
3173 old->fdl_next = fdtol;
3174 fdtol->fdl_next->fdl_prev = fdtol;
3175 FILEDESC_XUNLOCK(fdp);
3177 fdtol->fdl_next = fdtol;
3178 fdtol->fdl_prev = fdtol;
3184 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3186 struct filedesc *fdp;
3187 int i, count, slots;
3189 if (*(int *)arg1 != 0)
3192 fdp = curproc->p_fd;
3194 FILEDESC_SLOCK(fdp);
3195 slots = NDSLOTS(fdp->fd_lastfile + 1);
3196 for (i = 0; i < slots; i++)
3197 count += bitcountl(fdp->fd_map[i]);
3198 FILEDESC_SUNLOCK(fdp);
3200 return (SYSCTL_OUT(req, &count, sizeof(count)));
3203 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3204 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3205 "Number of open file descriptors");
3208 * Get file structures globally.
3211 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3214 struct filedesc *fdp;
3219 error = sysctl_wire_old_buffer(req, 0);
3222 if (req->oldptr == NULL) {
3224 sx_slock(&allproc_lock);
3225 FOREACH_PROC_IN_SYSTEM(p) {
3227 if (p->p_state == PRS_NEW) {
3235 /* overestimates sparse tables. */
3236 if (fdp->fd_lastfile > 0)
3237 n += fdp->fd_lastfile;
3240 sx_sunlock(&allproc_lock);
3241 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3244 bzero(&xf, sizeof(xf));
3245 xf.xf_size = sizeof(xf);
3246 sx_slock(&allproc_lock);
3247 FOREACH_PROC_IN_SYSTEM(p) {
3249 if (p->p_state == PRS_NEW) {
3253 if (p_cansee(req->td, p) != 0) {
3257 xf.xf_pid = p->p_pid;
3258 xf.xf_uid = p->p_ucred->cr_uid;
3263 FILEDESC_SLOCK(fdp);
3264 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3265 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3269 xf.xf_data = fp->f_data;
3270 xf.xf_vnode = fp->f_vnode;
3271 xf.xf_type = fp->f_type;
3272 xf.xf_count = fp->f_count;
3274 xf.xf_offset = foffset_get(fp);
3275 xf.xf_flag = fp->f_flag;
3276 error = SYSCTL_OUT(req, &xf, sizeof(xf));
3280 FILEDESC_SUNLOCK(fdp);
3285 sx_sunlock(&allproc_lock);
3289 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3290 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3292 #ifdef KINFO_FILE_SIZE
3293 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3297 xlate_fflags(int fflags)
3299 static const struct {
3302 } fflags_table[] = {
3303 { FAPPEND, KF_FLAG_APPEND },
3304 { FASYNC, KF_FLAG_ASYNC },
3305 { FFSYNC, KF_FLAG_FSYNC },
3306 { FHASLOCK, KF_FLAG_HASLOCK },
3307 { FNONBLOCK, KF_FLAG_NONBLOCK },
3308 { FREAD, KF_FLAG_READ },
3309 { FWRITE, KF_FLAG_WRITE },
3310 { O_CREAT, KF_FLAG_CREAT },
3311 { O_DIRECT, KF_FLAG_DIRECT },
3312 { O_EXCL, KF_FLAG_EXCL },
3313 { O_EXEC, KF_FLAG_EXEC },
3314 { O_EXLOCK, KF_FLAG_EXLOCK },
3315 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3316 { O_SHLOCK, KF_FLAG_SHLOCK },
3317 { O_TRUNC, KF_FLAG_TRUNC }
3323 for (i = 0; i < nitems(fflags_table); i++)
3324 if (fflags & fflags_table[i].fflag)
3325 kflags |= fflags_table[i].kf_fflag;
3329 /* Trim unused data from kf_path by truncating the structure size. */
3331 pack_kinfo(struct kinfo_file *kif)
3334 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3335 strlen(kif->kf_path) + 1;
3336 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3340 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3341 struct kinfo_file *kif, struct filedesc *fdp, int flags)
3345 bzero(kif, sizeof(*kif));
3347 /* Set a default type to allow for empty fill_kinfo() methods. */
3348 kif->kf_type = KF_TYPE_UNKNOWN;
3349 kif->kf_flags = xlate_fflags(fp->f_flag);
3350 if (rightsp != NULL)
3351 kif->kf_cap_rights = *rightsp;
3353 cap_rights_init(&kif->kf_cap_rights);
3355 kif->kf_ref_count = fp->f_count;
3356 kif->kf_offset = foffset_get(fp);
3359 * This may drop the filedesc lock, so the 'fp' cannot be
3360 * accessed after this call.
3362 error = fo_fill_kinfo(fp, kif, fdp);
3364 kif->kf_status |= KF_ATTR_VALID;
3365 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3368 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3372 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3373 struct kinfo_file *kif, int flags)
3377 bzero(kif, sizeof(*kif));
3379 kif->kf_type = KF_TYPE_VNODE;
3380 error = vn_fill_kinfo_vnode(vp, kif);
3382 kif->kf_status |= KF_ATTR_VALID;
3383 kif->kf_flags = xlate_fflags(fflags);
3384 cap_rights_init(&kif->kf_cap_rights);
3386 kif->kf_ref_count = -1;
3387 kif->kf_offset = -1;
3388 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3391 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3395 struct export_fd_buf {
3396 struct filedesc *fdp;
3399 struct kinfo_file kif;
3404 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3406 struct kinfo_file *kif;
3409 if (efbuf->remainder != -1) {
3410 if (efbuf->remainder < kif->kf_structsize) {
3411 /* Terminate export. */
3412 efbuf->remainder = 0;
3415 efbuf->remainder -= kif->kf_structsize;
3417 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3421 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3422 struct export_fd_buf *efbuf)
3426 if (efbuf->remainder == 0)
3428 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3430 FILEDESC_SUNLOCK(efbuf->fdp);
3431 error = export_kinfo_to_sb(efbuf);
3432 FILEDESC_SLOCK(efbuf->fdp);
3437 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3438 struct export_fd_buf *efbuf)
3442 if (efbuf->remainder == 0)
3444 if (efbuf->fdp != NULL)
3445 FILEDESC_SUNLOCK(efbuf->fdp);
3446 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3447 error = export_kinfo_to_sb(efbuf);
3448 if (efbuf->fdp != NULL)
3449 FILEDESC_SLOCK(efbuf->fdp);
3454 * Store a process file descriptor information to sbuf.
3456 * Takes a locked proc as argument, and returns with the proc unlocked.
3459 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
3463 struct filedesc *fdp;
3464 struct export_fd_buf *efbuf;
3465 struct vnode *cttyvp, *textvp, *tracevp;
3467 cap_rights_t rights;
3469 PROC_LOCK_ASSERT(p, MA_OWNED);
3472 tracevp = p->p_tracevp;
3473 if (tracevp != NULL)
3476 textvp = p->p_textvp;
3479 /* Controlling tty. */
3481 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3482 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3488 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3491 efbuf->remainder = maxlen;
3492 efbuf->flags = flags;
3493 if (tracevp != NULL)
3494 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3497 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3499 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3505 FILEDESC_SLOCK(fdp);
3506 /* working directory */
3507 if (fdp->fd_cdir != NULL) {
3508 vrefact(fdp->fd_cdir);
3509 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3511 /* root directory */
3512 if (fdp->fd_rdir != NULL) {
3513 vrefact(fdp->fd_rdir);
3514 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3516 /* jail directory */
3517 if (fdp->fd_jdir != NULL) {
3518 vrefact(fdp->fd_jdir);
3519 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3521 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3522 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3525 rights = *cap_rights(fdp, i);
3526 #else /* !CAPABILITIES */
3527 cap_rights_init(&rights);
3530 * Create sysctl entry. It is OK to drop the filedesc
3531 * lock inside of export_file_to_sb() as we will
3532 * re-validate and re-evaluate its properties when the
3535 error = export_file_to_sb(fp, i, &rights, efbuf);
3536 if (error != 0 || efbuf->remainder == 0)
3539 FILEDESC_SUNLOCK(fdp);
3542 free(efbuf, M_TEMP);
3546 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
3549 * Get per-process file descriptors for use by procstat(1), et al.
3552 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3557 int error, error2, *name;
3561 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3562 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3563 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3568 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3569 error = kern_proc_filedesc_out(p, &sb, maxlen,
3570 KERN_FILEDESC_PACK_KINFO);
3571 error2 = sbuf_finish(&sb);
3573 return (error != 0 ? error : error2);
3576 #ifdef KINFO_OFILE_SIZE
3577 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3580 #ifdef COMPAT_FREEBSD7
3582 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3585 okif->kf_structsize = sizeof(*okif);
3586 okif->kf_type = kif->kf_type;
3587 okif->kf_fd = kif->kf_fd;
3588 okif->kf_ref_count = kif->kf_ref_count;
3589 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3590 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3591 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3592 okif->kf_offset = kif->kf_offset;
3593 okif->kf_vnode_type = kif->kf_vnode_type;
3594 okif->kf_sock_domain = kif->kf_sock_domain;
3595 okif->kf_sock_type = kif->kf_sock_type;
3596 okif->kf_sock_protocol = kif->kf_sock_protocol;
3597 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3598 okif->kf_sa_local = kif->kf_sa_local;
3599 okif->kf_sa_peer = kif->kf_sa_peer;
3603 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3604 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3609 FILEDESC_SUNLOCK(fdp);
3610 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3611 kinfo_to_okinfo(kif, okif);
3612 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3613 FILEDESC_SLOCK(fdp);
3618 * Get per-process file descriptors for use by procstat(1), et al.
3621 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3623 struct kinfo_ofile *okif;
3624 struct kinfo_file *kif;
3625 struct filedesc *fdp;
3626 int error, i, *name;
3631 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3638 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3639 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3640 FILEDESC_SLOCK(fdp);
3641 if (fdp->fd_cdir != NULL)
3642 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3644 if (fdp->fd_rdir != NULL)
3645 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3647 if (fdp->fd_jdir != NULL)
3648 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3650 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3651 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3653 export_file_to_kinfo(fp, i, NULL, kif, fdp,
3654 KERN_FILEDESC_PACK_KINFO);
3655 FILEDESC_SUNLOCK(fdp);
3656 kinfo_to_okinfo(kif, okif);
3657 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3658 FILEDESC_SLOCK(fdp);
3662 FILEDESC_SUNLOCK(fdp);
3669 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3670 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3671 "Process ofiledesc entries");
3672 #endif /* COMPAT_FREEBSD7 */
3675 vntype_to_kinfo(int vtype)
3680 } vtypes_table[] = {
3681 { VBAD, KF_VTYPE_VBAD },
3682 { VBLK, KF_VTYPE_VBLK },
3683 { VCHR, KF_VTYPE_VCHR },
3684 { VDIR, KF_VTYPE_VDIR },
3685 { VFIFO, KF_VTYPE_VFIFO },
3686 { VLNK, KF_VTYPE_VLNK },
3687 { VNON, KF_VTYPE_VNON },
3688 { VREG, KF_VTYPE_VREG },
3689 { VSOCK, KF_VTYPE_VSOCK }
3694 * Perform vtype translation.
3696 for (i = 0; i < nitems(vtypes_table); i++)
3697 if (vtypes_table[i].vtype == vtype)
3698 return (vtypes_table[i].kf_vtype);
3700 return (KF_VTYPE_UNKNOWN);
3703 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3704 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3705 "Process filedesc entries");
3708 * Store a process current working directory information to sbuf.
3710 * Takes a locked proc as argument, and returns with the proc unlocked.
3713 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
3715 struct filedesc *fdp;
3716 struct export_fd_buf *efbuf;
3719 PROC_LOCK_ASSERT(p, MA_OWNED);
3726 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3729 efbuf->remainder = maxlen;
3731 FILEDESC_SLOCK(fdp);
3732 if (fdp->fd_cdir == NULL)
3735 vrefact(fdp->fd_cdir);
3736 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3739 FILEDESC_SUNLOCK(fdp);
3741 free(efbuf, M_TEMP);
3746 * Get per-process current working directory.
3749 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3754 int error, error2, *name;
3758 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3759 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3760 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3765 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3766 error = kern_proc_cwd_out(p, &sb, maxlen);
3767 error2 = sbuf_finish(&sb);
3769 return (error != 0 ? error : error2);
3772 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3773 sysctl_kern_proc_cwd, "Process current working directory");
3777 * For the purposes of debugging, generate a human-readable string for the
3781 file_type_to_name(short type)
3811 * For the purposes of debugging, identify a process (if any, perhaps one of
3812 * many) that references the passed file in its file descriptor array. Return
3815 static struct proc *
3816 file_to_first_proc(struct file *fp)
3818 struct filedesc *fdp;
3822 FOREACH_PROC_IN_SYSTEM(p) {
3823 if (p->p_state == PRS_NEW)
3828 for (n = 0; n <= fdp->fd_lastfile; n++) {
3829 if (fp == fdp->fd_ofiles[n].fde_file)
3837 db_print_file(struct file *fp, int header)
3842 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
3843 "File", "Type", "Data", "Flag", "GCFl", "Count",
3844 "MCount", "Vnode", "FPID", "FCmd");
3845 p = file_to_first_proc(fp);
3846 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
3847 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
3848 0, fp->f_count, 0, fp->f_vnode,
3849 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3852 DB_SHOW_COMMAND(file, db_show_file)
3857 db_printf("usage: show file <addr>\n");
3860 fp = (struct file *)addr;
3861 db_print_file(fp, 1);
3864 DB_SHOW_COMMAND(files, db_show_files)
3866 struct filedesc *fdp;
3873 FOREACH_PROC_IN_SYSTEM(p) {
3874 if (p->p_state == PRS_NEW)
3876 if ((fdp = p->p_fd) == NULL)
3878 for (n = 0; n <= fdp->fd_lastfile; ++n) {
3879 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3881 db_print_file(fp, header);
3888 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3889 &maxfilesperproc, 0, "Maximum files allowed open per process");
3891 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3892 &maxfiles, 0, "Maximum number of files");
3894 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3895 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
3899 filelistinit(void *dummy)
3902 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3903 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
3904 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
3905 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
3906 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3908 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3910 /*-------------------------------------------------------------------*/
3913 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
3914 int flags, struct thread *td)
3921 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
3929 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
3937 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
3945 badfo_kqfilter(struct file *fp, struct knote *kn)
3952 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
3960 badfo_close(struct file *fp, struct thread *td)
3967 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
3975 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
3983 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
3984 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
3992 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
3998 struct fileops badfileops = {
3999 .fo_read = badfo_readwrite,
4000 .fo_write = badfo_readwrite,
4001 .fo_truncate = badfo_truncate,
4002 .fo_ioctl = badfo_ioctl,
4003 .fo_poll = badfo_poll,
4004 .fo_kqfilter = badfo_kqfilter,
4005 .fo_stat = badfo_stat,
4006 .fo_close = badfo_close,
4007 .fo_chmod = badfo_chmod,
4008 .fo_chown = badfo_chown,
4009 .fo_sendfile = badfo_sendfile,
4010 .fo_fill_kinfo = badfo_fill_kinfo,
4014 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4015 int flags, struct thread *td)
4018 return (EOPNOTSUPP);
4022 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4030 invfo_ioctl(struct file *fp, u_long com, void *data,
4031 struct ucred *active_cred, struct thread *td)
4038 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4042 return (poll_no_poll(events));
4046 invfo_kqfilter(struct file *fp, struct knote *kn)
4053 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4061 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4069 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4070 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4077 /*-------------------------------------------------------------------*/
4080 * File Descriptor pseudo-device driver (/dev/fd/).
4082 * Opening minor device N dup()s the file (if any) connected to file
4083 * descriptor N belonging to the calling process. Note that this driver
4084 * consists of only the ``open()'' routine, because all subsequent
4085 * references to this file will be direct to the other driver.
4087 * XXX: we could give this one a cloning event handler if necessary.
4092 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4096 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4097 * the file descriptor being sought for duplication. The error
4098 * return ensures that the vnode for this device will be released
4099 * by vn_open. Open will detect this special error and take the
4100 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4101 * will simply report the error.
4103 td->td_dupfd = dev2unit(dev);
4107 static struct cdevsw fildesc_cdevsw = {
4108 .d_version = D_VERSION,
4114 fildesc_drvinit(void *unused)
4118 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4119 UID_ROOT, GID_WHEEL, 0666, "fd/0");
4120 make_dev_alias(dev, "stdin");
4121 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4122 UID_ROOT, GID_WHEEL, 0666, "fd/1");
4123 make_dev_alias(dev, "stdout");
4124 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4125 UID_ROOT, GID_WHEEL, 0666, "fd/2");
4126 make_dev_alias(dev, "stderr");
4129 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);