2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1989, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
42 #include "opt_capsicum.h"
43 #include "opt_compat.h"
45 #include "opt_ktrace.h"
47 #include <sys/param.h>
48 #include <sys/systm.h>
50 #include <sys/capsicum.h>
52 #include <sys/fcntl.h>
54 #include <sys/filedesc.h>
55 #include <sys/filio.h>
57 #include <sys/kernel.h>
58 #include <sys/limits.h>
60 #include <sys/malloc.h>
61 #include <sys/mount.h>
62 #include <sys/mutex.h>
63 #include <sys/namei.h>
64 #include <sys/selinfo.h>
67 #include <sys/protosw.h>
68 #include <sys/racct.h>
69 #include <sys/resourcevar.h>
71 #include <sys/signalvar.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysctl.h>
77 #include <sys/sysproto.h>
78 #include <sys/unistd.h>
80 #include <sys/vnode.h>
82 #include <sys/ktrace.h>
87 #include <security/audit/audit.h>
94 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
95 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
96 "file desc to leader structures");
97 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
98 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
100 MALLOC_DECLARE(M_FADVISE);
102 static __read_mostly uma_zone_t file_zone;
103 static __read_mostly uma_zone_t filedesc0_zone;
105 static int closefp(struct filedesc *fdp, int fd, struct file *fp,
106 struct thread *td, int holdleaders);
107 static int fd_first_free(struct filedesc *fdp, int low, int size);
108 static int fd_last_used(struct filedesc *fdp, int size);
109 static void fdgrowtable(struct filedesc *fdp, int nfd);
110 static void fdgrowtable_exp(struct filedesc *fdp, int nfd);
111 static void fdunused(struct filedesc *fdp, int fd);
112 static void fdused(struct filedesc *fdp, int fd);
113 static int getmaxfd(struct thread *td);
118 * - An array of open file descriptors (fd_ofiles)
119 * - An array of file flags (fd_ofileflags)
120 * - A bitmap recording which descriptors are in use (fd_map)
122 * A process starts out with NDFILE descriptors. The value of NDFILE has
123 * been selected based the historical limit of 20 open files, and an
124 * assumption that the majority of processes, especially short-lived
125 * processes like shells, will never need more.
127 * If this initial allocation is exhausted, a larger descriptor table and
128 * map are allocated dynamically, and the pointers in the process's struct
129 * filedesc are updated to point to those. This is repeated every time
130 * the process runs out of file descriptors (provided it hasn't hit its
133 * Since threads may hold references to individual descriptor table
134 * entries, the tables are never freed. Instead, they are placed on a
135 * linked list and freed only when the struct filedesc is released.
138 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
139 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
140 #define NDSLOT(x) ((x) / NDENTRIES)
141 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
142 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
145 * SLIST entry used to keep track of ofiles which must be reclaimed when
149 struct fdescenttbl *ft_table;
150 SLIST_ENTRY(freetable) ft_next;
154 * Initial allocation: a filedesc structure + the head of SLIST used to
155 * keep track of old ofiles + enough space for NDFILE descriptors.
158 struct fdescenttbl0 {
160 struct filedescent fdt_ofiles[NDFILE];
164 struct filedesc fd_fd;
165 SLIST_HEAD(, freetable) fd_free;
166 struct fdescenttbl0 fd_dfiles;
167 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
171 * Descriptor management.
173 volatile int __exclusive_cache_line openfiles; /* actual number of open files */
174 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
175 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
178 * If low >= size, just return low. Otherwise find the first zero bit in the
179 * given bitmap, starting at low and not exceeding size - 1. Return size if
183 fd_first_free(struct filedesc *fdp, int low, int size)
185 NDSLOTTYPE *map = fdp->fd_map;
193 if (low % NDENTRIES) {
194 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
195 if ((mask &= ~map[off]) != 0UL)
196 return (off * NDENTRIES + ffsl(mask) - 1);
199 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
200 if (map[off] != ~0UL)
201 return (off * NDENTRIES + ffsl(~map[off]) - 1);
206 * Find the highest non-zero bit in the given bitmap, starting at 0 and
207 * not exceeding size - 1. Return -1 if not found.
210 fd_last_used(struct filedesc *fdp, int size)
212 NDSLOTTYPE *map = fdp->fd_map;
217 if (size % NDENTRIES) {
218 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
219 if ((mask &= map[off]) != 0)
220 return (off * NDENTRIES + flsl(mask) - 1);
223 for (minoff = NDSLOT(0); off >= minoff; --off)
225 return (off * NDENTRIES + flsl(map[off]) - 1);
230 fdisused(struct filedesc *fdp, int fd)
233 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
234 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
236 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
240 * Mark a file descriptor as used.
243 fdused_init(struct filedesc *fdp, int fd)
246 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
248 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
252 fdused(struct filedesc *fdp, int fd)
255 FILEDESC_XLOCK_ASSERT(fdp);
257 fdused_init(fdp, fd);
258 if (fd > fdp->fd_lastfile)
259 fdp->fd_lastfile = fd;
260 if (fd == fdp->fd_freefile)
261 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
265 * Mark a file descriptor as unused.
268 fdunused(struct filedesc *fdp, int fd)
271 FILEDESC_XLOCK_ASSERT(fdp);
273 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
274 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
275 ("fd=%d is still in use", fd));
277 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
278 if (fd < fdp->fd_freefile)
279 fdp->fd_freefile = fd;
280 if (fd == fdp->fd_lastfile)
281 fdp->fd_lastfile = fd_last_used(fdp, fd);
285 * Free a file descriptor.
287 * Avoid some work if fdp is about to be destroyed.
290 fdefree_last(struct filedescent *fde)
293 filecaps_free(&fde->fde_caps);
297 fdfree(struct filedesc *fdp, int fd)
299 struct filedescent *fde;
301 fde = &fdp->fd_ofiles[fd];
303 seq_write_begin(&fde->fde_seq);
306 fde->fde_file = NULL;
309 seq_write_end(&fde->fde_seq);
314 pwd_ensure_dirs(void)
316 struct filedesc *fdp;
320 if (fdp->fd_cdir == NULL) {
321 fdp->fd_cdir = rootvnode;
324 if (fdp->fd_rdir == NULL) {
325 fdp->fd_rdir = rootvnode;
328 FILEDESC_XUNLOCK(fdp);
332 * System calls on descriptors.
334 #ifndef _SYS_SYSPROTO_H_
335 struct getdtablesize_args {
341 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
348 min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc);
350 PROC_LOCK(td->td_proc);
351 lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
352 PROC_UNLOCK(td->td_proc);
353 if (lim < td->td_retval[0])
354 td->td_retval[0] = lim;
360 * Duplicate a file descriptor to a particular value.
362 * Note: keep in mind that a potential race condition exists when closing
363 * descriptors from a shared descriptor table (via rfork).
365 #ifndef _SYS_SYSPROTO_H_
373 sys_dup2(struct thread *td, struct dup2_args *uap)
376 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
380 * Duplicate a file descriptor.
382 #ifndef _SYS_SYSPROTO_H_
389 sys_dup(struct thread *td, struct dup_args *uap)
392 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
396 * The file control system call.
398 #ifndef _SYS_SYSPROTO_H_
407 sys_fcntl(struct thread *td, struct fcntl_args *uap)
410 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
414 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
428 * Convert old flock structure to new.
430 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
431 fl.l_start = ofl.l_start;
432 fl.l_len = ofl.l_len;
433 fl.l_pid = ofl.l_pid;
434 fl.l_type = ofl.l_type;
435 fl.l_whence = ofl.l_whence;
449 arg1 = (intptr_t)&fl;
455 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
456 arg1 = (intptr_t)&fl;
464 error = kern_fcntl(td, fd, newcmd, arg1);
467 if (cmd == F_OGETLK) {
468 ofl.l_start = fl.l_start;
469 ofl.l_len = fl.l_len;
470 ofl.l_pid = fl.l_pid;
471 ofl.l_type = fl.l_type;
472 ofl.l_whence = fl.l_whence;
473 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
474 } else if (cmd == F_GETLK) {
475 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
481 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
483 struct filedesc *fdp;
485 struct file *fp, *fp2;
486 struct filedescent *fde;
504 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
507 case F_DUPFD_CLOEXEC:
509 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
514 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
517 case F_DUP2FD_CLOEXEC:
519 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
525 fde = fdeget_locked(fdp, fd);
528 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
531 FILEDESC_SUNLOCK(fdp);
537 fde = fdeget_locked(fdp, fd);
539 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
540 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
543 FILEDESC_XUNLOCK(fdp);
547 error = fget_fcntl(td, fd,
548 cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp);
551 td->td_retval[0] = OFLAGS(fp->f_flag);
556 error = fget_fcntl(td, fd,
557 cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp);
561 tmp = flg = fp->f_flag;
563 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
564 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
565 tmp = fp->f_flag & FNONBLOCK;
566 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
571 tmp = fp->f_flag & FASYNC;
572 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
577 atomic_clear_int(&fp->f_flag, FNONBLOCK);
579 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
584 error = fget_fcntl(td, fd,
585 cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp);
588 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
590 td->td_retval[0] = tmp;
595 error = fget_fcntl(td, fd,
596 cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp);
600 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
605 error = priv_check(td, PRIV_NFS_LOCKD);
613 /* FALLTHROUGH F_SETLK */
617 cap_rights_init(&rights, CAP_FLOCK);
618 error = fget_unlocked(fdp, fd, &rights, &fp, NULL);
621 if (fp->f_type != DTYPE_VNODE) {
627 flp = (struct flock *)arg;
628 if (flp->l_whence == SEEK_CUR) {
629 foffset = foffset_get(fp);
632 foffset > OFF_MAX - flp->l_start)) {
637 flp->l_start += foffset;
641 switch (flp->l_type) {
643 if ((fp->f_flag & FREAD) == 0) {
647 PROC_LOCK(p->p_leader);
648 p->p_leader->p_flag |= P_ADVLOCK;
649 PROC_UNLOCK(p->p_leader);
650 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
654 if ((fp->f_flag & FWRITE) == 0) {
658 PROC_LOCK(p->p_leader);
659 p->p_leader->p_flag |= P_ADVLOCK;
660 PROC_UNLOCK(p->p_leader);
661 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
665 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
670 * Temporary api for testing remote lock
673 if (flg != F_REMOTE) {
677 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
678 F_UNLCKSYS, flp, flg);
684 if (error != 0 || flp->l_type == F_UNLCK ||
685 flp->l_type == F_UNLCKSYS) {
691 * Check for a race with close.
693 * The vnode is now advisory locked (or unlocked, but this case
694 * is not really important) as the caller requested.
695 * We had to drop the filedesc lock, so we need to recheck if
696 * the descriptor is still valid, because if it was closed
697 * in the meantime we need to remove advisory lock from the
698 * vnode - close on any descriptor leading to an advisory
699 * locked vnode, removes that lock.
700 * We will return 0 on purpose in that case, as the result of
701 * successful advisory lock might have been externally visible
702 * already. This is fine - effectively we pretend to the caller
703 * that the closing thread was a bit slower and that the
704 * advisory lock succeeded before the close.
706 error = fget_unlocked(fdp, fd, &rights, &fp2, NULL);
712 flp->l_whence = SEEK_SET;
715 flp->l_type = F_UNLCK;
716 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
717 F_UNLCK, flp, F_POSIX);
724 error = fget_unlocked(fdp, fd,
725 cap_rights_init(&rights, CAP_FLOCK), &fp, NULL);
728 if (fp->f_type != DTYPE_VNODE) {
733 flp = (struct flock *)arg;
734 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
735 flp->l_type != F_UNLCK) {
740 if (flp->l_whence == SEEK_CUR) {
741 foffset = foffset_get(fp);
742 if ((flp->l_start > 0 &&
743 foffset > OFF_MAX - flp->l_start) ||
745 foffset < OFF_MIN - flp->l_start)) {
750 flp->l_start += foffset;
753 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
759 arg = arg ? 128 * 1024: 0;
762 error = fget_unlocked(fdp, fd,
763 cap_rights_init(&rights), &fp, NULL);
766 if (fp->f_type != DTYPE_VNODE) {
773 * Exclusive lock synchronizes against f_seqcount reads and
774 * writes in sequential_heuristic().
776 error = vn_lock(vp, LK_EXCLUSIVE);
782 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
783 fp->f_seqcount = (arg + bsize - 1) / bsize;
784 atomic_set_int(&fp->f_flag, FRDAHEAD);
786 atomic_clear_int(&fp->f_flag, FRDAHEAD);
800 getmaxfd(struct thread *td)
803 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
807 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
810 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
812 struct filedesc *fdp;
813 struct filedescent *oldfde, *newfde;
821 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
822 MPASS(mode < FDDUP_LASTMODE);
825 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
828 * Verify we have a valid descriptor to dup from and possibly to
829 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
830 * return EINVAL when the new descriptor is out of bounds.
835 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
836 maxfd = getmaxfd(td);
838 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
842 if (fget_locked(fdp, old) == NULL)
844 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
845 td->td_retval[0] = new;
846 if (flags & FDDUP_FLAG_CLOEXEC)
847 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
853 * If the caller specified a file descriptor, make sure the file
854 * table is large enough to hold it, and grab it. Otherwise, just
855 * allocate a new descriptor the usual way.
860 if ((error = fdalloc(td, new, &new)) != 0)
863 case FDDUP_MUSTREPLACE:
864 /* Target file descriptor must exist. */
865 if (fget_locked(fdp, new) == NULL)
869 if (new >= fdp->fd_nfiles) {
871 * The resource limits are here instead of e.g.
872 * fdalloc(), because the file descriptor table may be
873 * shared between processes, so we can't really use
874 * racct_add()/racct_sub(). Instead of counting the
875 * number of actually allocated descriptors, just put
876 * the limit on the size of the file descriptor table.
881 error = racct_set(p, RACCT_NOFILE, new + 1);
889 fdgrowtable_exp(fdp, new + 1);
891 if (!fdisused(fdp, new))
895 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
898 KASSERT(old != new, ("new fd is same as old"));
900 oldfde = &fdp->fd_ofiles[old];
901 fhold(oldfde->fde_file);
902 newfde = &fdp->fd_ofiles[new];
903 delfp = newfde->fde_file;
906 * Duplicate the source descriptor.
909 seq_write_begin(&newfde->fde_seq);
911 filecaps_free(&newfde->fde_caps);
912 memcpy(newfde, oldfde, fde_change_size);
913 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps, true);
914 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
915 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
917 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
919 seq_write_end(&newfde->fde_seq);
921 td->td_retval[0] = new;
926 (void) closefp(fdp, new, delfp, td, 1);
927 FILEDESC_UNLOCK_ASSERT(fdp);
930 FILEDESC_XUNLOCK(fdp);
937 * If sigio is on the list associated with a process or process group,
938 * disable signalling from the device, remove sigio from the list and
942 funsetown(struct sigio **sigiop)
954 *(sigio->sio_myref) = NULL;
955 if ((sigio)->sio_pgid < 0) {
956 struct pgrp *pg = (sigio)->sio_pgrp;
958 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
962 struct proc *p = (sigio)->sio_proc;
964 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
969 crfree(sigio->sio_ucred);
970 free(sigio, M_SIGIO);
974 * Free a list of sigio structures.
975 * We only need to lock the SIGIO_LOCK because we have made ourselves
976 * inaccessible to callers of fsetown and therefore do not need to lock
977 * the proc or pgrp struct for the list manipulation.
980 funsetownlst(struct sigiolst *sigiolst)
986 sigio = SLIST_FIRST(sigiolst);
993 * Every entry of the list should belong
994 * to a single proc or pgrp.
996 if (sigio->sio_pgid < 0) {
997 pg = sigio->sio_pgrp;
998 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
999 } else /* if (sigio->sio_pgid > 0) */ {
1000 p = sigio->sio_proc;
1001 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1005 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1006 *(sigio->sio_myref) = NULL;
1008 KASSERT(sigio->sio_pgid < 0,
1009 ("Proc sigio in pgrp sigio list"));
1010 KASSERT(sigio->sio_pgrp == pg,
1011 ("Bogus pgrp in sigio list"));
1013 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1016 } else /* if (p != NULL) */ {
1017 KASSERT(sigio->sio_pgid > 0,
1018 ("Pgrp sigio in proc sigio list"));
1019 KASSERT(sigio->sio_proc == p,
1020 ("Bogus proc in sigio list"));
1022 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1027 crfree(sigio->sio_ucred);
1028 free(sigio, M_SIGIO);
1035 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1037 * After permission checking, add a sigio structure to the sigio list for
1038 * the process or process group.
1041 fsetown(pid_t pgid, struct sigio **sigiop)
1045 struct sigio *sigio;
1055 /* Allocate and fill in the new sigio out of locks. */
1056 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1057 sigio->sio_pgid = pgid;
1058 sigio->sio_ucred = crhold(curthread->td_ucred);
1059 sigio->sio_myref = sigiop;
1061 sx_slock(&proctree_lock);
1070 * Policy - Don't allow a process to FSETOWN a process
1071 * in another session.
1073 * Remove this test to allow maximum flexibility or
1074 * restrict FSETOWN to the current process or process
1075 * group for maximum safety.
1078 if (proc->p_session != curthread->td_proc->p_session) {
1084 } else /* if (pgid < 0) */ {
1085 pgrp = pgfind(-pgid);
1093 * Policy - Don't allow a process to FSETOWN a process
1094 * in another session.
1096 * Remove this test to allow maximum flexibility or
1097 * restrict FSETOWN to the current process or process
1098 * group for maximum safety.
1100 if (pgrp->pg_session != curthread->td_proc->p_session) {
1111 * Since funsetownlst() is called without the proctree
1112 * locked, we need to check for P_WEXIT.
1113 * XXX: is ESRCH correct?
1115 if ((proc->p_flag & P_WEXIT) != 0) {
1120 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1121 sigio->sio_proc = proc;
1125 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1126 sigio->sio_pgrp = pgrp;
1129 sx_sunlock(&proctree_lock);
1136 sx_sunlock(&proctree_lock);
1137 crfree(sigio->sio_ucred);
1138 free(sigio, M_SIGIO);
1143 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1147 struct sigio **sigiop;
1152 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1158 * Function drops the filedesc lock on return.
1161 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1166 FILEDESC_XLOCK_ASSERT(fdp);
1169 if (td->td_proc->p_fdtol != NULL) {
1171 * Ask fdfree() to sleep to ensure that all relevant
1172 * process leaders can be traversed in closef().
1174 fdp->fd_holdleaderscount++;
1181 * We now hold the fp reference that used to be owned by the
1182 * descriptor array. We have to unlock the FILEDESC *AFTER*
1183 * knote_fdclose to prevent a race of the fd getting opened, a knote
1184 * added, and deleteing a knote for the new fd.
1186 knote_fdclose(td, fd);
1189 * We need to notify mqueue if the object is of type mqueue.
1191 if (fp->f_type == DTYPE_MQUEUE)
1192 mq_fdclose(td, fd, fp);
1193 FILEDESC_XUNLOCK(fdp);
1195 error = closef(fp, td);
1197 FILEDESC_XLOCK(fdp);
1198 fdp->fd_holdleaderscount--;
1199 if (fdp->fd_holdleaderscount == 0 &&
1200 fdp->fd_holdleaderswakeup != 0) {
1201 fdp->fd_holdleaderswakeup = 0;
1202 wakeup(&fdp->fd_holdleaderscount);
1204 FILEDESC_XUNLOCK(fdp);
1210 * Close a file descriptor.
1212 #ifndef _SYS_SYSPROTO_H_
1219 sys_close(struct thread *td, struct close_args *uap)
1222 return (kern_close(td, uap->fd));
1226 kern_close(struct thread *td, int fd)
1228 struct filedesc *fdp;
1231 fdp = td->td_proc->p_fd;
1233 AUDIT_SYSCLOSE(td, fd);
1235 FILEDESC_XLOCK(fdp);
1236 if ((fp = fget_locked(fdp, fd)) == NULL) {
1237 FILEDESC_XUNLOCK(fdp);
1242 /* closefp() drops the FILEDESC lock for us. */
1243 return (closefp(fdp, fd, fp, td, 1));
1247 * Close open file descriptors.
1249 #ifndef _SYS_SYSPROTO_H_
1250 struct closefrom_args {
1256 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1258 struct filedesc *fdp;
1261 fdp = td->td_proc->p_fd;
1262 AUDIT_ARG_FD(uap->lowfd);
1265 * Treat negative starting file descriptor values identical to
1266 * closefrom(0) which closes all files.
1270 FILEDESC_SLOCK(fdp);
1271 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1272 if (fdp->fd_ofiles[fd].fde_file != NULL) {
1273 FILEDESC_SUNLOCK(fdp);
1274 (void)kern_close(td, fd);
1275 FILEDESC_SLOCK(fdp);
1278 FILEDESC_SUNLOCK(fdp);
1282 #if defined(COMPAT_43)
1284 * Return status information about a file descriptor.
1286 #ifndef _SYS_SYSPROTO_H_
1287 struct ofstat_args {
1294 ofstat(struct thread *td, struct ofstat_args *uap)
1300 error = kern_fstat(td, uap->fd, &ub);
1303 error = copyout(&oub, uap->sb, sizeof(oub));
1307 #endif /* COMPAT_43 */
1309 #if defined(COMPAT_FREEBSD11)
1311 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1314 struct freebsd11_stat osb;
1317 error = kern_fstat(td, uap->fd, &sb);
1320 error = freebsd11_cvtstat(&sb, &osb);
1322 error = copyout(&osb, uap->sb, sizeof(osb));
1325 #endif /* COMPAT_FREEBSD11 */
1328 * Return status information about a file descriptor.
1330 #ifndef _SYS_SYSPROTO_H_
1338 sys_fstat(struct thread *td, struct fstat_args *uap)
1343 error = kern_fstat(td, uap->fd, &ub);
1345 error = copyout(&ub, uap->sb, sizeof(ub));
1350 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1353 cap_rights_t rights;
1358 error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp);
1362 AUDIT_ARG_FILE(td->td_proc, fp);
1364 error = fo_stat(fp, sbp, td->td_ucred, td);
1366 #ifdef __STAT_TIME_T_EXT
1368 sbp->st_atim_ext = 0;
1369 sbp->st_mtim_ext = 0;
1370 sbp->st_ctim_ext = 0;
1371 sbp->st_btim_ext = 0;
1375 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1381 #if defined(COMPAT_FREEBSD11)
1383 * Return status information about a file descriptor.
1385 #ifndef _SYS_SYSPROTO_H_
1386 struct freebsd11_nfstat_args {
1393 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1399 error = kern_fstat(td, uap->fd, &ub);
1401 freebsd11_cvtnstat(&ub, &nub);
1402 error = copyout(&nub, uap->sb, sizeof(nub));
1406 #endif /* COMPAT_FREEBSD11 */
1409 * Return pathconf information about a file descriptor.
1411 #ifndef _SYS_SYSPROTO_H_
1412 struct fpathconf_args {
1419 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1423 cap_rights_t rights;
1426 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp);
1430 if (uap->name == _PC_ASYNC_IO) {
1431 td->td_retval[0] = _POSIX_ASYNCHRONOUS_IO;
1436 vn_lock(vp, LK_SHARED | LK_RETRY);
1437 error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1439 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1440 if (uap->name != _PC_PIPE_BUF) {
1443 td->td_retval[0] = PIPE_BUF;
1455 * Initialize filecaps structure.
1458 filecaps_init(struct filecaps *fcaps)
1461 bzero(fcaps, sizeof(*fcaps));
1462 fcaps->fc_nioctls = -1;
1466 * Copy filecaps structure allocating memory for ioctls array if needed.
1468 * The last parameter indicates whether the fdtable is locked. If it is not and
1469 * ioctls are encountered, copying fails and the caller must lock the table.
1471 * Note that if the table was not locked, the caller has to check the relevant
1472 * sequence counter to determine whether the operation was successful.
1475 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1480 if (src->fc_ioctls == NULL)
1485 KASSERT(src->fc_nioctls > 0,
1486 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1488 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1489 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1490 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1495 * Move filecaps structure to the new place and clear the old place.
1498 filecaps_move(struct filecaps *src, struct filecaps *dst)
1502 bzero(src, sizeof(*src));
1506 * Fill the given filecaps structure with full rights.
1509 filecaps_fill(struct filecaps *fcaps)
1512 CAP_ALL(&fcaps->fc_rights);
1513 fcaps->fc_ioctls = NULL;
1514 fcaps->fc_nioctls = -1;
1515 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1519 * Free memory allocated within filecaps structure.
1522 filecaps_free(struct filecaps *fcaps)
1525 free(fcaps->fc_ioctls, M_FILECAPS);
1526 bzero(fcaps, sizeof(*fcaps));
1530 * Validate the given filecaps structure.
1533 filecaps_validate(const struct filecaps *fcaps, const char *func)
1536 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1537 ("%s: invalid rights", func));
1538 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1539 ("%s: invalid fcntls", func));
1540 KASSERT(fcaps->fc_fcntls == 0 ||
1541 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1542 ("%s: fcntls without CAP_FCNTL", func));
1543 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1544 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1545 ("%s: invalid ioctls", func));
1546 KASSERT(fcaps->fc_nioctls == 0 ||
1547 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1548 ("%s: ioctls without CAP_IOCTL", func));
1552 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1556 FILEDESC_XLOCK_ASSERT(fdp);
1558 nfd1 = fdp->fd_nfiles * 2;
1561 fdgrowtable(fdp, nfd1);
1565 * Grow the file table to accommodate (at least) nfd descriptors.
1568 fdgrowtable(struct filedesc *fdp, int nfd)
1570 struct filedesc0 *fdp0;
1571 struct freetable *ft;
1572 struct fdescenttbl *ntable;
1573 struct fdescenttbl *otable;
1574 int nnfiles, onfiles;
1575 NDSLOTTYPE *nmap, *omap;
1578 * If lastfile is -1 this struct filedesc was just allocated and we are
1579 * growing it to accommodate for the one we are going to copy from. There
1580 * is no need to have a lock on this one as it's not visible to anyone.
1582 if (fdp->fd_lastfile != -1)
1583 FILEDESC_XLOCK_ASSERT(fdp);
1585 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1587 /* save old values */
1588 onfiles = fdp->fd_nfiles;
1589 otable = fdp->fd_files;
1592 /* compute the size of the new table */
1593 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1594 if (nnfiles <= onfiles)
1595 /* the table is already large enough */
1599 * Allocate a new table. We need enough space for the number of
1600 * entries, file entries themselves and the struct freetable we will use
1601 * when we decommission the table and place it on the freelist.
1602 * We place the struct freetable in the middle so we don't have
1603 * to worry about padding.
1605 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1606 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1607 sizeof(struct freetable),
1608 M_FILEDESC, M_ZERO | M_WAITOK);
1609 /* copy the old data */
1610 ntable->fdt_nfiles = nnfiles;
1611 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1612 onfiles * sizeof(ntable->fdt_ofiles[0]));
1615 * Allocate a new map only if the old is not large enough. It will
1616 * grow at a slower rate than the table as it can map more
1617 * entries than the table can hold.
1619 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1620 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1622 /* copy over the old data and update the pointer */
1623 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1628 * Make sure that ntable is correctly initialized before we replace
1629 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1632 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1635 * Do not free the old file table, as some threads may still
1636 * reference entries within it. Instead, place it on a freelist
1637 * which will be processed when the struct filedesc is released.
1639 * Note that if onfiles == NDFILE, we're dealing with the original
1640 * static allocation contained within (struct filedesc0 *)fdp,
1641 * which must not be freed.
1643 if (onfiles > NDFILE) {
1644 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1645 fdp0 = (struct filedesc0 *)fdp;
1646 ft->ft_table = otable;
1647 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1650 * The map does not have the same possibility of threads still
1651 * holding references to it. So always free it as long as it
1652 * does not reference the original static allocation.
1654 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1655 free(omap, M_FILEDESC);
1659 * Allocate a file descriptor for the process.
1662 fdalloc(struct thread *td, int minfd, int *result)
1664 struct proc *p = td->td_proc;
1665 struct filedesc *fdp = p->p_fd;
1666 int fd, maxfd, allocfd;
1671 FILEDESC_XLOCK_ASSERT(fdp);
1673 if (fdp->fd_freefile > minfd)
1674 minfd = fdp->fd_freefile;
1676 maxfd = getmaxfd(td);
1679 * Search the bitmap for a free descriptor starting at minfd.
1680 * If none is found, grow the file table.
1682 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1685 if (fd >= fdp->fd_nfiles) {
1686 allocfd = min(fd * 2, maxfd);
1690 error = racct_set(p, RACCT_NOFILE, allocfd);
1697 * fd is already equal to first free descriptor >= minfd, so
1698 * we only need to grow the table and we are done.
1700 fdgrowtable_exp(fdp, allocfd);
1704 * Perform some sanity checks, then mark the file descriptor as
1705 * used and return it to the caller.
1707 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1708 ("invalid descriptor %d", fd));
1709 KASSERT(!fdisused(fdp, fd),
1710 ("fd_first_free() returned non-free descriptor"));
1711 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1712 ("file descriptor isn't free"));
1719 * Allocate n file descriptors for the process.
1722 fdallocn(struct thread *td, int minfd, int *fds, int n)
1724 struct proc *p = td->td_proc;
1725 struct filedesc *fdp = p->p_fd;
1728 FILEDESC_XLOCK_ASSERT(fdp);
1730 for (i = 0; i < n; i++)
1731 if (fdalloc(td, 0, &fds[i]) != 0)
1735 for (i--; i >= 0; i--)
1736 fdunused(fdp, fds[i]);
1744 * Create a new open file structure and allocate a file descriptor for the
1745 * process that refers to it. We add one reference to the file for the
1746 * descriptor table and one reference for resultfp. This is to prevent us
1747 * being preempted and the entry in the descriptor table closed after we
1748 * release the FILEDESC lock.
1751 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1752 struct filecaps *fcaps)
1757 error = falloc_noinstall(td, &fp);
1759 return (error); /* no reference held on error */
1761 error = finstall(td, fp, &fd, flags, fcaps);
1763 fdrop(fp, td); /* one reference (fp only) */
1767 if (resultfp != NULL)
1768 *resultfp = fp; /* copy out result */
1770 fdrop(fp, td); /* release local reference */
1772 if (resultfd != NULL)
1779 * Create a new open file structure without allocating a file descriptor.
1782 falloc_noinstall(struct thread *td, struct file **resultfp)
1785 int maxuserfiles = maxfiles - (maxfiles / 20);
1787 static struct timeval lastfail;
1790 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1792 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1793 if ((openfiles_new >= maxuserfiles &&
1794 priv_check(td, PRIV_MAXFILES) != 0) ||
1795 openfiles_new >= maxfiles) {
1796 atomic_subtract_int(&openfiles, 1);
1797 if (ppsratecheck(&lastfail, &curfail, 1)) {
1798 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1799 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1803 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1804 refcount_init(&fp->f_count, 1);
1805 fp->f_cred = crhold(td->td_ucred);
1806 fp->f_ops = &badfileops;
1812 * Install a file in a file descriptor table.
1815 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1816 struct filecaps *fcaps)
1818 struct filedescent *fde;
1822 filecaps_validate(fcaps, __func__);
1823 FILEDESC_XLOCK_ASSERT(fdp);
1825 fde = &fdp->fd_ofiles[fd];
1827 seq_write_begin(&fde->fde_seq);
1830 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1832 filecaps_move(fcaps, &fde->fde_caps);
1834 filecaps_fill(&fde->fde_caps);
1836 seq_write_end(&fde->fde_seq);
1841 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1842 struct filecaps *fcaps)
1844 struct filedesc *fdp = td->td_proc->p_fd;
1849 FILEDESC_XLOCK(fdp);
1850 if ((error = fdalloc(td, 0, fd))) {
1851 FILEDESC_XUNLOCK(fdp);
1855 _finstall(fdp, fp, *fd, flags, fcaps);
1856 FILEDESC_XUNLOCK(fdp);
1861 * Build a new filedesc structure from another.
1862 * Copy the current, root, and jail root vnode references.
1864 * If fdp is not NULL, return with it shared locked.
1867 fdinit(struct filedesc *fdp, bool prepfiles)
1869 struct filedesc0 *newfdp0;
1870 struct filedesc *newfdp;
1872 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1873 newfdp = &newfdp0->fd_fd;
1875 /* Create the file descriptor table. */
1876 FILEDESC_LOCK_INIT(newfdp);
1877 refcount_init(&newfdp->fd_refcnt, 1);
1878 refcount_init(&newfdp->fd_holdcnt, 1);
1879 newfdp->fd_cmask = CMASK;
1880 newfdp->fd_map = newfdp0->fd_dmap;
1881 newfdp->fd_lastfile = -1;
1882 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1883 newfdp->fd_files->fdt_nfiles = NDFILE;
1888 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1889 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1891 FILEDESC_SLOCK(fdp);
1892 newfdp->fd_cdir = fdp->fd_cdir;
1893 if (newfdp->fd_cdir)
1894 vrefact(newfdp->fd_cdir);
1895 newfdp->fd_rdir = fdp->fd_rdir;
1896 if (newfdp->fd_rdir)
1897 vrefact(newfdp->fd_rdir);
1898 newfdp->fd_jdir = fdp->fd_jdir;
1899 if (newfdp->fd_jdir)
1900 vrefact(newfdp->fd_jdir);
1903 FILEDESC_SUNLOCK(fdp);
1905 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1906 FILEDESC_SUNLOCK(fdp);
1907 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1908 FILEDESC_SLOCK(fdp);
1915 static struct filedesc *
1916 fdhold(struct proc *p)
1918 struct filedesc *fdp;
1920 PROC_LOCK_ASSERT(p, MA_OWNED);
1923 refcount_acquire(&fdp->fd_holdcnt);
1928 fddrop(struct filedesc *fdp)
1931 if (fdp->fd_holdcnt > 1) {
1932 if (refcount_release(&fdp->fd_holdcnt) == 0)
1936 FILEDESC_LOCK_DESTROY(fdp);
1937 uma_zfree(filedesc0_zone, fdp);
1941 * Share a filedesc structure.
1944 fdshare(struct filedesc *fdp)
1947 refcount_acquire(&fdp->fd_refcnt);
1952 * Unshare a filedesc structure, if necessary by making a copy
1955 fdunshare(struct thread *td)
1957 struct filedesc *tmp;
1958 struct proc *p = td->td_proc;
1960 if (p->p_fd->fd_refcnt == 1)
1963 tmp = fdcopy(p->p_fd);
1969 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
1973 td->td_proc->p_fd = fdp;
1977 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
1978 * this is to ease callers, not catch errors.
1981 fdcopy(struct filedesc *fdp)
1983 struct filedesc *newfdp;
1984 struct filedescent *nfde, *ofde;
1989 newfdp = fdinit(fdp, true);
1990 /* copy all passable descriptors (i.e. not kqueue) */
1991 newfdp->fd_freefile = -1;
1992 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1993 ofde = &fdp->fd_ofiles[i];
1994 if (ofde->fde_file == NULL ||
1995 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
1996 if (newfdp->fd_freefile == -1)
1997 newfdp->fd_freefile = i;
2000 nfde = &newfdp->fd_ofiles[i];
2002 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2003 fhold(nfde->fde_file);
2004 fdused_init(newfdp, i);
2005 newfdp->fd_lastfile = i;
2007 if (newfdp->fd_freefile == -1)
2008 newfdp->fd_freefile = i;
2009 newfdp->fd_cmask = fdp->fd_cmask;
2010 FILEDESC_SUNLOCK(fdp);
2015 * Copies a filedesc structure, while remapping all file descriptors
2016 * stored inside using a translation table.
2018 * File descriptors are copied over to the new file descriptor table,
2019 * regardless of whether the close-on-exec flag is set.
2022 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2023 struct filedesc **ret)
2025 struct filedesc *newfdp;
2026 struct filedescent *nfde, *ofde;
2031 newfdp = fdinit(fdp, true);
2032 if (nfds > fdp->fd_lastfile + 1) {
2033 /* New table cannot be larger than the old one. */
2037 /* Copy all passable descriptors (i.e. not kqueue). */
2038 newfdp->fd_freefile = nfds;
2039 for (i = 0; i < nfds; ++i) {
2040 if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2041 /* File descriptor out of bounds. */
2045 ofde = &fdp->fd_ofiles[fds[i]];
2046 if (ofde->fde_file == NULL) {
2047 /* Unused file descriptor. */
2051 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2052 /* File descriptor cannot be passed. */
2056 nfde = &newfdp->fd_ofiles[i];
2058 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2059 fhold(nfde->fde_file);
2060 fdused_init(newfdp, i);
2061 newfdp->fd_lastfile = i;
2063 newfdp->fd_cmask = fdp->fd_cmask;
2064 FILEDESC_SUNLOCK(fdp);
2068 FILEDESC_SUNLOCK(fdp);
2069 fdescfree_remapped(newfdp);
2074 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2075 * one of processes using it exits) and the table used to be shared.
2078 fdclearlocks(struct thread *td)
2080 struct filedesc *fdp;
2081 struct filedesc_to_leader *fdtol;
2091 MPASS(fdtol != NULL);
2093 FILEDESC_XLOCK(fdp);
2094 KASSERT(fdtol->fdl_refcount > 0,
2095 ("filedesc_to_refcount botch: fdl_refcount=%d",
2096 fdtol->fdl_refcount));
2097 if (fdtol->fdl_refcount == 1 &&
2098 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2099 for (i = 0; i <= fdp->fd_lastfile; i++) {
2100 fp = fdp->fd_ofiles[i].fde_file;
2101 if (fp == NULL || fp->f_type != DTYPE_VNODE)
2104 FILEDESC_XUNLOCK(fdp);
2105 lf.l_whence = SEEK_SET;
2108 lf.l_type = F_UNLCK;
2110 (void) VOP_ADVLOCK(vp,
2111 (caddr_t)p->p_leader, F_UNLCK,
2113 FILEDESC_XLOCK(fdp);
2118 if (fdtol->fdl_refcount == 1) {
2119 if (fdp->fd_holdleaderscount > 0 &&
2120 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2122 * close() or kern_dup() has cleared a reference
2123 * in a shared file descriptor table.
2125 fdp->fd_holdleaderswakeup = 1;
2126 sx_sleep(&fdp->fd_holdleaderscount,
2127 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2130 if (fdtol->fdl_holdcount > 0) {
2132 * Ensure that fdtol->fdl_leader remains
2133 * valid in closef().
2135 fdtol->fdl_wakeup = 1;
2136 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2141 fdtol->fdl_refcount--;
2142 if (fdtol->fdl_refcount == 0 &&
2143 fdtol->fdl_holdcount == 0) {
2144 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2145 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2149 FILEDESC_XUNLOCK(fdp);
2151 free(fdtol, M_FILEDESC_TO_LEADER);
2155 * Release a filedesc structure.
2158 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2160 struct filedesc0 *fdp0;
2161 struct freetable *ft, *tft;
2162 struct filedescent *fde;
2166 for (i = 0; i <= fdp->fd_lastfile; i++) {
2167 fde = &fdp->fd_ofiles[i];
2172 (void) closef(fp, td);
2178 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2179 free(fdp->fd_map, M_FILEDESC);
2180 if (fdp->fd_nfiles > NDFILE)
2181 free(fdp->fd_files, M_FILEDESC);
2183 fdp0 = (struct filedesc0 *)fdp;
2184 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2185 free(ft->ft_table, M_FILEDESC);
2191 fdescfree(struct thread *td)
2194 struct filedesc *fdp;
2195 struct vnode *cdir, *jdir, *rdir;
2204 racct_set(p, RACCT_NOFILE, 0);
2209 if (p->p_fdtol != NULL)
2216 if (refcount_release(&fdp->fd_refcnt) == 0)
2219 FILEDESC_XLOCK(fdp);
2220 cdir = fdp->fd_cdir;
2221 fdp->fd_cdir = NULL;
2222 rdir = fdp->fd_rdir;
2223 fdp->fd_rdir = NULL;
2224 jdir = fdp->fd_jdir;
2225 fdp->fd_jdir = NULL;
2226 FILEDESC_XUNLOCK(fdp);
2235 fdescfree_fds(td, fdp, 1);
2239 fdescfree_remapped(struct filedesc *fdp)
2242 if (fdp->fd_cdir != NULL)
2243 vrele(fdp->fd_cdir);
2244 if (fdp->fd_rdir != NULL)
2245 vrele(fdp->fd_rdir);
2246 if (fdp->fd_jdir != NULL)
2247 vrele(fdp->fd_jdir);
2249 fdescfree_fds(curthread, fdp, 0);
2253 * For setugid programs, we don't want to people to use that setugidness
2254 * to generate error messages which write to a file which otherwise would
2255 * otherwise be off-limits to the process. We check for filesystems where
2256 * the vnode can change out from under us after execve (like [lin]procfs).
2258 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2259 * sufficient. We also don't check for setugidness since we know we are.
2262 is_unsafe(struct file *fp)
2266 if (fp->f_type != DTYPE_VNODE)
2270 return ((vp->v_vflag & VV_PROCDEP) != 0);
2274 * Make this setguid thing safe, if at all possible.
2277 fdsetugidsafety(struct thread *td)
2279 struct filedesc *fdp;
2283 fdp = td->td_proc->p_fd;
2284 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2285 MPASS(fdp->fd_nfiles >= 3);
2286 for (i = 0; i <= 2; i++) {
2287 fp = fdp->fd_ofiles[i].fde_file;
2288 if (fp != NULL && is_unsafe(fp)) {
2289 FILEDESC_XLOCK(fdp);
2290 knote_fdclose(td, i);
2292 * NULL-out descriptor prior to close to avoid
2293 * a race while close blocks.
2296 FILEDESC_XUNLOCK(fdp);
2297 (void) closef(fp, td);
2303 * If a specific file object occupies a specific file descriptor, close the
2304 * file descriptor entry and drop a reference on the file object. This is a
2305 * convenience function to handle a subsequent error in a function that calls
2306 * falloc() that handles the race that another thread might have closed the
2307 * file descriptor out from under the thread creating the file object.
2310 fdclose(struct thread *td, struct file *fp, int idx)
2312 struct filedesc *fdp = td->td_proc->p_fd;
2314 FILEDESC_XLOCK(fdp);
2315 if (fdp->fd_ofiles[idx].fde_file == fp) {
2317 FILEDESC_XUNLOCK(fdp);
2320 FILEDESC_XUNLOCK(fdp);
2324 * Close any files on exec?
2327 fdcloseexec(struct thread *td)
2329 struct filedesc *fdp;
2330 struct filedescent *fde;
2334 fdp = td->td_proc->p_fd;
2335 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2336 for (i = 0; i <= fdp->fd_lastfile; i++) {
2337 fde = &fdp->fd_ofiles[i];
2339 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2340 (fde->fde_flags & UF_EXCLOSE))) {
2341 FILEDESC_XLOCK(fdp);
2343 (void) closefp(fdp, i, fp, td, 0);
2344 FILEDESC_UNLOCK_ASSERT(fdp);
2350 * It is unsafe for set[ug]id processes to be started with file
2351 * descriptors 0..2 closed, as these descriptors are given implicit
2352 * significance in the Standard C library. fdcheckstd() will create a
2353 * descriptor referencing /dev/null for each of stdin, stdout, and
2354 * stderr that is not already open.
2357 fdcheckstd(struct thread *td)
2359 struct filedesc *fdp;
2361 int i, error, devnull;
2363 fdp = td->td_proc->p_fd;
2364 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2365 MPASS(fdp->fd_nfiles >= 3);
2367 for (i = 0; i <= 2; i++) {
2368 if (fdp->fd_ofiles[i].fde_file != NULL)
2371 save = td->td_retval[0];
2372 if (devnull != -1) {
2373 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2375 error = kern_openat(td, AT_FDCWD, "/dev/null",
2376 UIO_SYSSPACE, O_RDWR, 0);
2378 devnull = td->td_retval[0];
2379 KASSERT(devnull == i, ("we didn't get our fd"));
2382 td->td_retval[0] = save;
2390 * Internal form of close. Decrement reference count on file structure.
2391 * Note: td may be NULL when closing a file that was being passed in a
2394 * XXXRW: Giant is not required for the caller, but often will be held; this
2395 * makes it moderately likely the Giant will be recursed in the VFS case.
2398 closef(struct file *fp, struct thread *td)
2402 struct filedesc_to_leader *fdtol;
2403 struct filedesc *fdp;
2406 * POSIX record locking dictates that any close releases ALL
2407 * locks owned by this process. This is handled by setting
2408 * a flag in the unlock to free ONLY locks obeying POSIX
2409 * semantics, and not to free BSD-style file locks.
2410 * If the descriptor was in a message, POSIX-style locks
2411 * aren't passed with the descriptor, and the thread pointer
2412 * will be NULL. Callers should be careful only to pass a
2413 * NULL thread pointer when there really is no owning
2414 * context that might have locks, or the locks will be
2417 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2419 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2420 lf.l_whence = SEEK_SET;
2423 lf.l_type = F_UNLCK;
2424 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2425 F_UNLCK, &lf, F_POSIX);
2427 fdtol = td->td_proc->p_fdtol;
2428 if (fdtol != NULL) {
2430 * Handle special case where file descriptor table is
2431 * shared between multiple process leaders.
2433 fdp = td->td_proc->p_fd;
2434 FILEDESC_XLOCK(fdp);
2435 for (fdtol = fdtol->fdl_next;
2436 fdtol != td->td_proc->p_fdtol;
2437 fdtol = fdtol->fdl_next) {
2438 if ((fdtol->fdl_leader->p_flag &
2441 fdtol->fdl_holdcount++;
2442 FILEDESC_XUNLOCK(fdp);
2443 lf.l_whence = SEEK_SET;
2446 lf.l_type = F_UNLCK;
2448 (void) VOP_ADVLOCK(vp,
2449 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2451 FILEDESC_XLOCK(fdp);
2452 fdtol->fdl_holdcount--;
2453 if (fdtol->fdl_holdcount == 0 &&
2454 fdtol->fdl_wakeup != 0) {
2455 fdtol->fdl_wakeup = 0;
2459 FILEDESC_XUNLOCK(fdp);
2462 return (fdrop(fp, td));
2466 * Initialize the file pointer with the specified properties.
2468 * The ops are set with release semantics to be certain that the flags, type,
2469 * and data are visible when ops is. This is to prevent ops methods from being
2470 * called with bad data.
2473 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2478 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2482 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2483 struct file **fpp, struct filecaps *havecapsp)
2485 struct filedescent *fde;
2488 FILEDESC_LOCK_ASSERT(fdp);
2490 fde = fdeget_locked(fdp, fd);
2497 error = cap_check(cap_rights_fde(fde), needrightsp);
2502 if (havecapsp != NULL)
2503 filecaps_copy(&fde->fde_caps, havecapsp, true);
2505 *fpp = fde->fde_file;
2513 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2514 struct file **fpp, struct filecaps *havecapsp)
2516 struct filedesc *fdp = td->td_proc->p_fd;
2518 #ifndef CAPABILITIES
2519 error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2520 if (error == 0 && havecapsp != NULL)
2521 filecaps_fill(havecapsp);
2527 error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2531 if (havecapsp != NULL) {
2532 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2533 havecapsp, false)) {
2539 if (!fd_modified(fdp, fd, seq))
2548 FILEDESC_SLOCK(fdp);
2549 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2552 FILEDESC_SUNLOCK(fdp);
2558 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2559 struct file **fpp, seq_t *seqp)
2562 struct filedescent *fde;
2564 struct fdescenttbl *fdt;
2569 cap_rights_t haverights;
2573 fdt = fdp->fd_files;
2574 if ((u_int)fd >= fdt->fdt_nfiles)
2577 * Fetch the descriptor locklessly. We avoid fdrop() races by
2578 * never raising a refcount above 0. To accomplish this we have
2579 * to use a cmpset loop rather than an atomic_add. The descriptor
2580 * must be re-verified once we acquire a reference to be certain
2581 * that the identity is still correct and we did not lose a race
2582 * due to preemption.
2586 seq = seq_read(fd_seq(fdt, fd));
2587 fde = &fdt->fdt_ofiles[fd];
2588 haverights = *cap_rights_fde(fde);
2590 if (!seq_consistent(fd_seq(fdt, fd), seq))
2593 fp = fdt->fdt_ofiles[fd].fde_file;
2598 error = cap_check(&haverights, needrightsp);
2602 count = fp->f_count;
2606 * Force a reload. Other thread could reallocate the
2607 * table before this fd was closed, so it possible that
2608 * there is a stale fp pointer in cached version.
2610 fdt = *(struct fdescenttbl * volatile *)&(fdp->fd_files);
2614 * Use an acquire barrier to force re-reading of fdt so it is
2615 * refreshed for verification.
2617 if (atomic_fcmpset_acq_int(&fp->f_count, &count, count + 1) == 0)
2619 fdt = fdp->fd_files;
2621 if (seq_consistent_nomb(fd_seq(fdt, fd), seq))
2623 if (fp == fdt->fdt_ofiles[fd].fde_file)
2626 fdrop(fp, curthread);
2638 * Extract the file pointer associated with the specified descriptor for the
2639 * current user process.
2641 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2644 * File's rights will be checked against the capability rights mask.
2646 * If an error occurred the non-zero error is returned and *fpp is set to
2647 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
2648 * responsible for fdrop().
2651 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2652 cap_rights_t *needrightsp, seq_t *seqp)
2654 struct filedesc *fdp;
2659 fdp = td->td_proc->p_fd;
2660 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2663 if (fp->f_ops == &badfileops) {
2669 * FREAD and FWRITE failure return EBADF as per POSIX.
2675 if ((fp->f_flag & flags) == 0)
2679 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2680 ((fp->f_flag & FWRITE) != 0))
2686 KASSERT(0, ("wrong flags"));
2699 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2702 return (_fget(td, fd, fpp, 0, rightsp, NULL));
2706 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2710 #ifndef CAPABILITIES
2711 error = _fget(td, fd, fpp, 0, rightsp, NULL);
2712 if (maxprotp != NULL)
2713 *maxprotp = VM_PROT_ALL;
2715 struct filedesc *fdp = td->td_proc->p_fd;
2718 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2720 error = _fget(td, fd, fpp, 0, rightsp, &seq);
2724 * If requested, convert capability rights to access flags.
2726 if (maxprotp != NULL)
2727 *maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd));
2728 if (!fd_modified(fdp, fd, seq))
2737 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2740 return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2744 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2747 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2751 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2754 struct filedesc *fdp = td->td_proc->p_fd;
2755 #ifndef CAPABILITIES
2756 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2761 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2763 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2766 error = cap_fcntl_check(fdp, fd, needfcntl);
2767 if (!fd_modified(fdp, fd, seq))
2780 * Like fget() but loads the underlying vnode, or returns an error if the
2781 * descriptor does not represent a vnode. Note that pipes use vnodes but
2782 * never have VM objects. The returned vnode will be vref()'d.
2784 * XXX: what about the unused flags ?
2787 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2794 error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2797 if (fp->f_vnode == NULL) {
2809 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2812 return (_fgetvp(td, fd, 0, rightsp, vpp));
2816 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2817 struct filecaps *havecaps, struct vnode **vpp)
2819 struct filedesc *fdp;
2820 struct filecaps caps;
2824 fdp = td->td_proc->p_fd;
2825 error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2828 if (fp->f_ops == &badfileops) {
2832 if (fp->f_vnode == NULL) {
2843 filecaps_free(&caps);
2848 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2851 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2855 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2858 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2863 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2867 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2872 * Handle the last reference to a file being closed.
2875 _fdrop(struct file *fp, struct thread *td)
2879 if (fp->f_count != 0)
2880 panic("fdrop: count %d", fp->f_count);
2881 error = fo_close(fp, td);
2882 atomic_subtract_int(&openfiles, 1);
2884 free(fp->f_advice, M_FADVISE);
2885 uma_zfree(file_zone, fp);
2891 * Apply an advisory lock on a file descriptor.
2893 * Just attempt to get a record lock of the requested type on the entire file
2894 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2896 #ifndef _SYS_SYSPROTO_H_
2904 sys_flock(struct thread *td, struct flock_args *uap)
2909 cap_rights_t rights;
2912 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp);
2915 if (fp->f_type != DTYPE_VNODE) {
2917 return (EOPNOTSUPP);
2921 lf.l_whence = SEEK_SET;
2924 if (uap->how & LOCK_UN) {
2925 lf.l_type = F_UNLCK;
2926 atomic_clear_int(&fp->f_flag, FHASLOCK);
2927 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2930 if (uap->how & LOCK_EX)
2931 lf.l_type = F_WRLCK;
2932 else if (uap->how & LOCK_SH)
2933 lf.l_type = F_RDLCK;
2938 atomic_set_int(&fp->f_flag, FHASLOCK);
2939 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2940 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2946 * Duplicate the specified descriptor to a free descriptor.
2949 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
2950 int openerror, int *indxp)
2952 struct filedescent *newfde, *oldfde;
2956 KASSERT(openerror == ENODEV || openerror == ENXIO,
2957 ("unexpected error %d in %s", openerror, __func__));
2960 * If the to-be-dup'd fd number is greater than the allowed number
2961 * of file descriptors, or the fd to be dup'd has already been
2962 * closed, then reject.
2964 FILEDESC_XLOCK(fdp);
2965 if ((fp = fget_locked(fdp, dfd)) == NULL) {
2966 FILEDESC_XUNLOCK(fdp);
2970 error = fdalloc(td, 0, &indx);
2972 FILEDESC_XUNLOCK(fdp);
2977 * There are two cases of interest here.
2979 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2981 * For ENXIO steal away the file structure from (dfd) and store it in
2982 * (indx). (dfd) is effectively closed by this operation.
2984 switch (openerror) {
2987 * Check that the mode the file is being opened for is a
2988 * subset of the mode of the existing descriptor.
2990 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
2991 fdunused(fdp, indx);
2992 FILEDESC_XUNLOCK(fdp);
2996 newfde = &fdp->fd_ofiles[indx];
2997 oldfde = &fdp->fd_ofiles[dfd];
2999 seq_write_begin(&newfde->fde_seq);
3001 memcpy(newfde, oldfde, fde_change_size);
3002 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps, true);
3004 seq_write_end(&newfde->fde_seq);
3009 * Steal away the file pointer from dfd and stuff it into indx.
3011 newfde = &fdp->fd_ofiles[indx];
3012 oldfde = &fdp->fd_ofiles[dfd];
3014 seq_write_begin(&newfde->fde_seq);
3016 memcpy(newfde, oldfde, fde_change_size);
3017 oldfde->fde_file = NULL;
3020 seq_write_end(&newfde->fde_seq);
3024 FILEDESC_XUNLOCK(fdp);
3030 * This sysctl determines if we will allow a process to chroot(2) if it
3031 * has a directory open:
3032 * 0: disallowed for all processes.
3033 * 1: allowed for processes that were not already chroot(2)'ed.
3034 * 2: allowed for all processes.
3037 static int chroot_allow_open_directories = 1;
3039 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3040 &chroot_allow_open_directories, 0,
3041 "Allow a process to chroot(2) if it has a directory open");
3044 * Helper function for raised chroot(2) security function: Refuse if
3045 * any filedescriptors are open directories.
3048 chroot_refuse_vdir_fds(struct filedesc *fdp)
3054 FILEDESC_LOCK_ASSERT(fdp);
3056 for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3057 fp = fget_locked(fdp, fd);
3060 if (fp->f_type == DTYPE_VNODE) {
3062 if (vp->v_type == VDIR)
3070 * Common routine for kern_chroot() and jail_attach(). The caller is
3071 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3072 * authorize this operation.
3075 pwd_chroot(struct thread *td, struct vnode *vp)
3077 struct filedesc *fdp;
3078 struct vnode *oldvp;
3081 fdp = td->td_proc->p_fd;
3082 FILEDESC_XLOCK(fdp);
3083 if (chroot_allow_open_directories == 0 ||
3084 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3085 error = chroot_refuse_vdir_fds(fdp);
3087 FILEDESC_XUNLOCK(fdp);
3091 oldvp = fdp->fd_rdir;
3094 if (fdp->fd_jdir == NULL) {
3098 FILEDESC_XUNLOCK(fdp);
3104 pwd_chdir(struct thread *td, struct vnode *vp)
3106 struct filedesc *fdp;
3107 struct vnode *oldvp;
3109 fdp = td->td_proc->p_fd;
3110 FILEDESC_XLOCK(fdp);
3111 VNASSERT(vp->v_usecount > 0, vp,
3112 ("chdir to a vnode with zero usecount"));
3113 oldvp = fdp->fd_cdir;
3115 FILEDESC_XUNLOCK(fdp);
3120 * Scan all active processes and prisons to see if any of them have a current
3121 * or root directory of `olddp'. If so, replace them with the new mount point.
3124 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3126 struct filedesc *fdp;
3131 if (vrefcnt(olddp) == 1)
3134 sx_slock(&allproc_lock);
3135 FOREACH_PROC_IN_SYSTEM(p) {
3141 FILEDESC_XLOCK(fdp);
3142 if (fdp->fd_cdir == olddp) {
3144 fdp->fd_cdir = newdp;
3147 if (fdp->fd_rdir == olddp) {
3149 fdp->fd_rdir = newdp;
3152 if (fdp->fd_jdir == olddp) {
3154 fdp->fd_jdir = newdp;
3157 FILEDESC_XUNLOCK(fdp);
3160 sx_sunlock(&allproc_lock);
3161 if (rootvnode == olddp) {
3166 mtx_lock(&prison0.pr_mtx);
3167 if (prison0.pr_root == olddp) {
3169 prison0.pr_root = newdp;
3172 mtx_unlock(&prison0.pr_mtx);
3173 sx_slock(&allprison_lock);
3174 TAILQ_FOREACH(pr, &allprison, pr_list) {
3175 mtx_lock(&pr->pr_mtx);
3176 if (pr->pr_root == olddp) {
3178 pr->pr_root = newdp;
3181 mtx_unlock(&pr->pr_mtx);
3183 sx_sunlock(&allprison_lock);
3188 struct filedesc_to_leader *
3189 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3191 struct filedesc_to_leader *fdtol;
3193 fdtol = malloc(sizeof(struct filedesc_to_leader),
3194 M_FILEDESC_TO_LEADER, M_WAITOK);
3195 fdtol->fdl_refcount = 1;
3196 fdtol->fdl_holdcount = 0;
3197 fdtol->fdl_wakeup = 0;
3198 fdtol->fdl_leader = leader;
3200 FILEDESC_XLOCK(fdp);
3201 fdtol->fdl_next = old->fdl_next;
3202 fdtol->fdl_prev = old;
3203 old->fdl_next = fdtol;
3204 fdtol->fdl_next->fdl_prev = fdtol;
3205 FILEDESC_XUNLOCK(fdp);
3207 fdtol->fdl_next = fdtol;
3208 fdtol->fdl_prev = fdtol;
3214 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3216 struct filedesc *fdp;
3217 int i, count, slots;
3219 if (*(int *)arg1 != 0)
3222 fdp = curproc->p_fd;
3224 FILEDESC_SLOCK(fdp);
3225 slots = NDSLOTS(fdp->fd_lastfile + 1);
3226 for (i = 0; i < slots; i++)
3227 count += bitcountl(fdp->fd_map[i]);
3228 FILEDESC_SUNLOCK(fdp);
3230 return (SYSCTL_OUT(req, &count, sizeof(count)));
3233 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3234 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3235 "Number of open file descriptors");
3238 * Get file structures globally.
3241 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3244 struct filedesc *fdp;
3249 error = sysctl_wire_old_buffer(req, 0);
3252 if (req->oldptr == NULL) {
3254 sx_slock(&allproc_lock);
3255 FOREACH_PROC_IN_SYSTEM(p) {
3257 if (p->p_state == PRS_NEW) {
3265 /* overestimates sparse tables. */
3266 if (fdp->fd_lastfile > 0)
3267 n += fdp->fd_lastfile;
3270 sx_sunlock(&allproc_lock);
3271 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3274 bzero(&xf, sizeof(xf));
3275 xf.xf_size = sizeof(xf);
3276 sx_slock(&allproc_lock);
3277 FOREACH_PROC_IN_SYSTEM(p) {
3279 if (p->p_state == PRS_NEW) {
3283 if (p_cansee(req->td, p) != 0) {
3287 xf.xf_pid = p->p_pid;
3288 xf.xf_uid = p->p_ucred->cr_uid;
3293 FILEDESC_SLOCK(fdp);
3294 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3295 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3299 xf.xf_data = fp->f_data;
3300 xf.xf_vnode = fp->f_vnode;
3301 xf.xf_type = fp->f_type;
3302 xf.xf_count = fp->f_count;
3304 xf.xf_offset = foffset_get(fp);
3305 xf.xf_flag = fp->f_flag;
3306 error = SYSCTL_OUT(req, &xf, sizeof(xf));
3310 FILEDESC_SUNLOCK(fdp);
3315 sx_sunlock(&allproc_lock);
3319 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3320 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3322 #ifdef KINFO_FILE_SIZE
3323 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3327 xlate_fflags(int fflags)
3329 static const struct {
3332 } fflags_table[] = {
3333 { FAPPEND, KF_FLAG_APPEND },
3334 { FASYNC, KF_FLAG_ASYNC },
3335 { FFSYNC, KF_FLAG_FSYNC },
3336 { FHASLOCK, KF_FLAG_HASLOCK },
3337 { FNONBLOCK, KF_FLAG_NONBLOCK },
3338 { FREAD, KF_FLAG_READ },
3339 { FWRITE, KF_FLAG_WRITE },
3340 { O_CREAT, KF_FLAG_CREAT },
3341 { O_DIRECT, KF_FLAG_DIRECT },
3342 { O_EXCL, KF_FLAG_EXCL },
3343 { O_EXEC, KF_FLAG_EXEC },
3344 { O_EXLOCK, KF_FLAG_EXLOCK },
3345 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3346 { O_SHLOCK, KF_FLAG_SHLOCK },
3347 { O_TRUNC, KF_FLAG_TRUNC }
3353 for (i = 0; i < nitems(fflags_table); i++)
3354 if (fflags & fflags_table[i].fflag)
3355 kflags |= fflags_table[i].kf_fflag;
3359 /* Trim unused data from kf_path by truncating the structure size. */
3361 pack_kinfo(struct kinfo_file *kif)
3364 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3365 strlen(kif->kf_path) + 1;
3366 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3370 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3371 struct kinfo_file *kif, struct filedesc *fdp, int flags)
3375 bzero(kif, sizeof(*kif));
3377 /* Set a default type to allow for empty fill_kinfo() methods. */
3378 kif->kf_type = KF_TYPE_UNKNOWN;
3379 kif->kf_flags = xlate_fflags(fp->f_flag);
3380 if (rightsp != NULL)
3381 kif->kf_cap_rights = *rightsp;
3383 cap_rights_init(&kif->kf_cap_rights);
3385 kif->kf_ref_count = fp->f_count;
3386 kif->kf_offset = foffset_get(fp);
3389 * This may drop the filedesc lock, so the 'fp' cannot be
3390 * accessed after this call.
3392 error = fo_fill_kinfo(fp, kif, fdp);
3394 kif->kf_status |= KF_ATTR_VALID;
3395 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3398 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3402 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3403 struct kinfo_file *kif, int flags)
3407 bzero(kif, sizeof(*kif));
3409 kif->kf_type = KF_TYPE_VNODE;
3410 error = vn_fill_kinfo_vnode(vp, kif);
3412 kif->kf_status |= KF_ATTR_VALID;
3413 kif->kf_flags = xlate_fflags(fflags);
3414 cap_rights_init(&kif->kf_cap_rights);
3416 kif->kf_ref_count = -1;
3417 kif->kf_offset = -1;
3418 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3421 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3425 struct export_fd_buf {
3426 struct filedesc *fdp;
3429 struct kinfo_file kif;
3434 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3436 struct kinfo_file *kif;
3439 if (efbuf->remainder != -1) {
3440 if (efbuf->remainder < kif->kf_structsize) {
3441 /* Terminate export. */
3442 efbuf->remainder = 0;
3445 efbuf->remainder -= kif->kf_structsize;
3447 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3451 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3452 struct export_fd_buf *efbuf)
3456 if (efbuf->remainder == 0)
3458 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3460 FILEDESC_SUNLOCK(efbuf->fdp);
3461 error = export_kinfo_to_sb(efbuf);
3462 FILEDESC_SLOCK(efbuf->fdp);
3467 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3468 struct export_fd_buf *efbuf)
3472 if (efbuf->remainder == 0)
3474 if (efbuf->fdp != NULL)
3475 FILEDESC_SUNLOCK(efbuf->fdp);
3476 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3477 error = export_kinfo_to_sb(efbuf);
3478 if (efbuf->fdp != NULL)
3479 FILEDESC_SLOCK(efbuf->fdp);
3484 * Store a process file descriptor information to sbuf.
3486 * Takes a locked proc as argument, and returns with the proc unlocked.
3489 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
3493 struct filedesc *fdp;
3494 struct export_fd_buf *efbuf;
3495 struct vnode *cttyvp, *textvp, *tracevp;
3497 cap_rights_t rights;
3499 PROC_LOCK_ASSERT(p, MA_OWNED);
3502 tracevp = p->p_tracevp;
3503 if (tracevp != NULL)
3506 textvp = p->p_textvp;
3509 /* Controlling tty. */
3511 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3512 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3518 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3521 efbuf->remainder = maxlen;
3522 efbuf->flags = flags;
3523 if (tracevp != NULL)
3524 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3527 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3529 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3535 FILEDESC_SLOCK(fdp);
3536 /* working directory */
3537 if (fdp->fd_cdir != NULL) {
3538 vrefact(fdp->fd_cdir);
3539 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3541 /* root directory */
3542 if (fdp->fd_rdir != NULL) {
3543 vrefact(fdp->fd_rdir);
3544 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3546 /* jail directory */
3547 if (fdp->fd_jdir != NULL) {
3548 vrefact(fdp->fd_jdir);
3549 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3551 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3552 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3555 rights = *cap_rights(fdp, i);
3556 #else /* !CAPABILITIES */
3557 cap_rights_init(&rights);
3560 * Create sysctl entry. It is OK to drop the filedesc
3561 * lock inside of export_file_to_sb() as we will
3562 * re-validate and re-evaluate its properties when the
3565 error = export_file_to_sb(fp, i, &rights, efbuf);
3566 if (error != 0 || efbuf->remainder == 0)
3569 FILEDESC_SUNLOCK(fdp);
3572 free(efbuf, M_TEMP);
3576 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
3579 * Get per-process file descriptors for use by procstat(1), et al.
3582 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3587 int error, error2, *name;
3591 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3592 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3593 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3598 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3599 error = kern_proc_filedesc_out(p, &sb, maxlen,
3600 KERN_FILEDESC_PACK_KINFO);
3601 error2 = sbuf_finish(&sb);
3603 return (error != 0 ? error : error2);
3606 #ifdef COMPAT_FREEBSD7
3607 #ifdef KINFO_OFILE_SIZE
3608 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3612 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3615 okif->kf_structsize = sizeof(*okif);
3616 okif->kf_type = kif->kf_type;
3617 okif->kf_fd = kif->kf_fd;
3618 okif->kf_ref_count = kif->kf_ref_count;
3619 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3620 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3621 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3622 okif->kf_offset = kif->kf_offset;
3623 if (kif->kf_type == KF_TYPE_VNODE)
3624 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
3626 okif->kf_vnode_type = KF_VTYPE_VNON;
3627 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3628 if (kif->kf_type == KF_TYPE_SOCKET) {
3629 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
3630 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
3631 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
3632 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
3633 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
3635 okif->kf_sa_local.ss_family = AF_UNSPEC;
3636 okif->kf_sa_peer.ss_family = AF_UNSPEC;
3641 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3642 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3647 FILEDESC_SUNLOCK(fdp);
3648 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3649 kinfo_to_okinfo(kif, okif);
3650 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3651 FILEDESC_SLOCK(fdp);
3656 * Get per-process file descriptors for use by procstat(1), et al.
3659 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3661 struct kinfo_ofile *okif;
3662 struct kinfo_file *kif;
3663 struct filedesc *fdp;
3664 int error, i, *name;
3669 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3676 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3677 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3678 FILEDESC_SLOCK(fdp);
3679 if (fdp->fd_cdir != NULL)
3680 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3682 if (fdp->fd_rdir != NULL)
3683 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3685 if (fdp->fd_jdir != NULL)
3686 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3688 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3689 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3691 export_file_to_kinfo(fp, i, NULL, kif, fdp,
3692 KERN_FILEDESC_PACK_KINFO);
3693 FILEDESC_SUNLOCK(fdp);
3694 kinfo_to_okinfo(kif, okif);
3695 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3696 FILEDESC_SLOCK(fdp);
3700 FILEDESC_SUNLOCK(fdp);
3707 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3708 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3709 "Process ofiledesc entries");
3710 #endif /* COMPAT_FREEBSD7 */
3713 vntype_to_kinfo(int vtype)
3718 } vtypes_table[] = {
3719 { VBAD, KF_VTYPE_VBAD },
3720 { VBLK, KF_VTYPE_VBLK },
3721 { VCHR, KF_VTYPE_VCHR },
3722 { VDIR, KF_VTYPE_VDIR },
3723 { VFIFO, KF_VTYPE_VFIFO },
3724 { VLNK, KF_VTYPE_VLNK },
3725 { VNON, KF_VTYPE_VNON },
3726 { VREG, KF_VTYPE_VREG },
3727 { VSOCK, KF_VTYPE_VSOCK }
3732 * Perform vtype translation.
3734 for (i = 0; i < nitems(vtypes_table); i++)
3735 if (vtypes_table[i].vtype == vtype)
3736 return (vtypes_table[i].kf_vtype);
3738 return (KF_VTYPE_UNKNOWN);
3741 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3742 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3743 "Process filedesc entries");
3746 * Store a process current working directory information to sbuf.
3748 * Takes a locked proc as argument, and returns with the proc unlocked.
3751 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
3753 struct filedesc *fdp;
3754 struct export_fd_buf *efbuf;
3757 PROC_LOCK_ASSERT(p, MA_OWNED);
3764 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3767 efbuf->remainder = maxlen;
3769 FILEDESC_SLOCK(fdp);
3770 if (fdp->fd_cdir == NULL)
3773 vrefact(fdp->fd_cdir);
3774 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3777 FILEDESC_SUNLOCK(fdp);
3779 free(efbuf, M_TEMP);
3784 * Get per-process current working directory.
3787 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3792 int error, error2, *name;
3796 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3797 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3798 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3803 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3804 error = kern_proc_cwd_out(p, &sb, maxlen);
3805 error2 = sbuf_finish(&sb);
3807 return (error != 0 ? error : error2);
3810 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3811 sysctl_kern_proc_cwd, "Process current working directory");
3815 * For the purposes of debugging, generate a human-readable string for the
3819 file_type_to_name(short type)
3847 case DTYPE_PROCDESC:
3849 case DTYPE_LINUXEFD:
3851 case DTYPE_LINUXTFD:
3859 * For the purposes of debugging, identify a process (if any, perhaps one of
3860 * many) that references the passed file in its file descriptor array. Return
3863 static struct proc *
3864 file_to_first_proc(struct file *fp)
3866 struct filedesc *fdp;
3870 FOREACH_PROC_IN_SYSTEM(p) {
3871 if (p->p_state == PRS_NEW)
3876 for (n = 0; n <= fdp->fd_lastfile; n++) {
3877 if (fp == fdp->fd_ofiles[n].fde_file)
3885 db_print_file(struct file *fp, int header)
3887 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
3891 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
3892 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
3893 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
3895 p = file_to_first_proc(fp);
3896 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
3897 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
3898 fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode,
3899 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3904 DB_SHOW_COMMAND(file, db_show_file)
3909 db_printf("usage: show file <addr>\n");
3912 fp = (struct file *)addr;
3913 db_print_file(fp, 1);
3916 DB_SHOW_COMMAND(files, db_show_files)
3918 struct filedesc *fdp;
3925 FOREACH_PROC_IN_SYSTEM(p) {
3926 if (p->p_state == PRS_NEW)
3928 if ((fdp = p->p_fd) == NULL)
3930 for (n = 0; n <= fdp->fd_lastfile; ++n) {
3931 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3933 db_print_file(fp, header);
3940 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3941 &maxfilesperproc, 0, "Maximum files allowed open per process");
3943 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3944 &maxfiles, 0, "Maximum number of files");
3946 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3947 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
3951 filelistinit(void *dummy)
3954 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3955 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
3956 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
3957 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
3958 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3960 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3962 /*-------------------------------------------------------------------*/
3965 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
3966 int flags, struct thread *td)
3973 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
3981 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
3989 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
3997 badfo_kqfilter(struct file *fp, struct knote *kn)
4004 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4012 badfo_close(struct file *fp, struct thread *td)
4019 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4027 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4035 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4036 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4044 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4050 struct fileops badfileops = {
4051 .fo_read = badfo_readwrite,
4052 .fo_write = badfo_readwrite,
4053 .fo_truncate = badfo_truncate,
4054 .fo_ioctl = badfo_ioctl,
4055 .fo_poll = badfo_poll,
4056 .fo_kqfilter = badfo_kqfilter,
4057 .fo_stat = badfo_stat,
4058 .fo_close = badfo_close,
4059 .fo_chmod = badfo_chmod,
4060 .fo_chown = badfo_chown,
4061 .fo_sendfile = badfo_sendfile,
4062 .fo_fill_kinfo = badfo_fill_kinfo,
4066 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4067 int flags, struct thread *td)
4070 return (EOPNOTSUPP);
4074 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4082 invfo_ioctl(struct file *fp, u_long com, void *data,
4083 struct ucred *active_cred, struct thread *td)
4090 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4094 return (poll_no_poll(events));
4098 invfo_kqfilter(struct file *fp, struct knote *kn)
4105 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4113 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4121 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4122 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4129 /*-------------------------------------------------------------------*/
4132 * File Descriptor pseudo-device driver (/dev/fd/).
4134 * Opening minor device N dup()s the file (if any) connected to file
4135 * descriptor N belonging to the calling process. Note that this driver
4136 * consists of only the ``open()'' routine, because all subsequent
4137 * references to this file will be direct to the other driver.
4139 * XXX: we could give this one a cloning event handler if necessary.
4144 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4148 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4149 * the file descriptor being sought for duplication. The error
4150 * return ensures that the vnode for this device will be released
4151 * by vn_open. Open will detect this special error and take the
4152 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4153 * will simply report the error.
4155 td->td_dupfd = dev2unit(dev);
4159 static struct cdevsw fildesc_cdevsw = {
4160 .d_version = D_VERSION,
4166 fildesc_drvinit(void *unused)
4170 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4171 UID_ROOT, GID_WHEEL, 0666, "fd/0");
4172 make_dev_alias(dev, "stdin");
4173 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4174 UID_ROOT, GID_WHEEL, 0666, "fd/1");
4175 make_dev_alias(dev, "stdout");
4176 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4177 UID_ROOT, GID_WHEEL, 0666, "fd/2");
4178 make_dev_alias(dev, "stderr");
4181 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);