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)
1422 return (kern_fpathconf(td, uap->fd, uap->name));
1426 kern_fpathconf(struct thread *td, int fd, int name)
1430 cap_rights_t rights;
1433 error = fget(td, fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp);
1437 if (name == _PC_ASYNC_IO) {
1438 td->td_retval[0] = _POSIX_ASYNCHRONOUS_IO;
1443 vn_lock(vp, LK_SHARED | LK_RETRY);
1444 error = VOP_PATHCONF(vp, name, td->td_retval);
1446 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1447 if (name != _PC_PIPE_BUF) {
1450 td->td_retval[0] = PIPE_BUF;
1462 * Initialize filecaps structure.
1465 filecaps_init(struct filecaps *fcaps)
1468 bzero(fcaps, sizeof(*fcaps));
1469 fcaps->fc_nioctls = -1;
1473 * Copy filecaps structure allocating memory for ioctls array if needed.
1475 * The last parameter indicates whether the fdtable is locked. If it is not and
1476 * ioctls are encountered, copying fails and the caller must lock the table.
1478 * Note that if the table was not locked, the caller has to check the relevant
1479 * sequence counter to determine whether the operation was successful.
1482 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1487 if (src->fc_ioctls == NULL)
1492 KASSERT(src->fc_nioctls > 0,
1493 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1495 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1496 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1497 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1502 * Move filecaps structure to the new place and clear the old place.
1505 filecaps_move(struct filecaps *src, struct filecaps *dst)
1509 bzero(src, sizeof(*src));
1513 * Fill the given filecaps structure with full rights.
1516 filecaps_fill(struct filecaps *fcaps)
1519 CAP_ALL(&fcaps->fc_rights);
1520 fcaps->fc_ioctls = NULL;
1521 fcaps->fc_nioctls = -1;
1522 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1526 * Free memory allocated within filecaps structure.
1529 filecaps_free(struct filecaps *fcaps)
1532 free(fcaps->fc_ioctls, M_FILECAPS);
1533 bzero(fcaps, sizeof(*fcaps));
1537 * Validate the given filecaps structure.
1540 filecaps_validate(const struct filecaps *fcaps, const char *func)
1543 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1544 ("%s: invalid rights", func));
1545 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1546 ("%s: invalid fcntls", func));
1547 KASSERT(fcaps->fc_fcntls == 0 ||
1548 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1549 ("%s: fcntls without CAP_FCNTL", func));
1550 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1551 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1552 ("%s: invalid ioctls", func));
1553 KASSERT(fcaps->fc_nioctls == 0 ||
1554 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1555 ("%s: ioctls without CAP_IOCTL", func));
1559 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1563 FILEDESC_XLOCK_ASSERT(fdp);
1565 nfd1 = fdp->fd_nfiles * 2;
1568 fdgrowtable(fdp, nfd1);
1572 * Grow the file table to accommodate (at least) nfd descriptors.
1575 fdgrowtable(struct filedesc *fdp, int nfd)
1577 struct filedesc0 *fdp0;
1578 struct freetable *ft;
1579 struct fdescenttbl *ntable;
1580 struct fdescenttbl *otable;
1581 int nnfiles, onfiles;
1582 NDSLOTTYPE *nmap, *omap;
1585 * If lastfile is -1 this struct filedesc was just allocated and we are
1586 * growing it to accommodate for the one we are going to copy from. There
1587 * is no need to have a lock on this one as it's not visible to anyone.
1589 if (fdp->fd_lastfile != -1)
1590 FILEDESC_XLOCK_ASSERT(fdp);
1592 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1594 /* save old values */
1595 onfiles = fdp->fd_nfiles;
1596 otable = fdp->fd_files;
1599 /* compute the size of the new table */
1600 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1601 if (nnfiles <= onfiles)
1602 /* the table is already large enough */
1606 * Allocate a new table. We need enough space for the number of
1607 * entries, file entries themselves and the struct freetable we will use
1608 * when we decommission the table and place it on the freelist.
1609 * We place the struct freetable in the middle so we don't have
1610 * to worry about padding.
1612 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1613 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1614 sizeof(struct freetable),
1615 M_FILEDESC, M_ZERO | M_WAITOK);
1616 /* copy the old data */
1617 ntable->fdt_nfiles = nnfiles;
1618 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1619 onfiles * sizeof(ntable->fdt_ofiles[0]));
1622 * Allocate a new map only if the old is not large enough. It will
1623 * grow at a slower rate than the table as it can map more
1624 * entries than the table can hold.
1626 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1627 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1629 /* copy over the old data and update the pointer */
1630 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1635 * Make sure that ntable is correctly initialized before we replace
1636 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1639 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1642 * Do not free the old file table, as some threads may still
1643 * reference entries within it. Instead, place it on a freelist
1644 * which will be processed when the struct filedesc is released.
1646 * Note that if onfiles == NDFILE, we're dealing with the original
1647 * static allocation contained within (struct filedesc0 *)fdp,
1648 * which must not be freed.
1650 if (onfiles > NDFILE) {
1651 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1652 fdp0 = (struct filedesc0 *)fdp;
1653 ft->ft_table = otable;
1654 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1657 * The map does not have the same possibility of threads still
1658 * holding references to it. So always free it as long as it
1659 * does not reference the original static allocation.
1661 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1662 free(omap, M_FILEDESC);
1666 * Allocate a file descriptor for the process.
1669 fdalloc(struct thread *td, int minfd, int *result)
1671 struct proc *p = td->td_proc;
1672 struct filedesc *fdp = p->p_fd;
1673 int fd, maxfd, allocfd;
1678 FILEDESC_XLOCK_ASSERT(fdp);
1680 if (fdp->fd_freefile > minfd)
1681 minfd = fdp->fd_freefile;
1683 maxfd = getmaxfd(td);
1686 * Search the bitmap for a free descriptor starting at minfd.
1687 * If none is found, grow the file table.
1689 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1692 if (fd >= fdp->fd_nfiles) {
1693 allocfd = min(fd * 2, maxfd);
1697 error = racct_set(p, RACCT_NOFILE, allocfd);
1704 * fd is already equal to first free descriptor >= minfd, so
1705 * we only need to grow the table and we are done.
1707 fdgrowtable_exp(fdp, allocfd);
1711 * Perform some sanity checks, then mark the file descriptor as
1712 * used and return it to the caller.
1714 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1715 ("invalid descriptor %d", fd));
1716 KASSERT(!fdisused(fdp, fd),
1717 ("fd_first_free() returned non-free descriptor"));
1718 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1719 ("file descriptor isn't free"));
1726 * Allocate n file descriptors for the process.
1729 fdallocn(struct thread *td, int minfd, int *fds, int n)
1731 struct proc *p = td->td_proc;
1732 struct filedesc *fdp = p->p_fd;
1735 FILEDESC_XLOCK_ASSERT(fdp);
1737 for (i = 0; i < n; i++)
1738 if (fdalloc(td, 0, &fds[i]) != 0)
1742 for (i--; i >= 0; i--)
1743 fdunused(fdp, fds[i]);
1751 * Create a new open file structure and allocate a file descriptor for the
1752 * process that refers to it. We add one reference to the file for the
1753 * descriptor table and one reference for resultfp. This is to prevent us
1754 * being preempted and the entry in the descriptor table closed after we
1755 * release the FILEDESC lock.
1758 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1759 struct filecaps *fcaps)
1764 error = falloc_noinstall(td, &fp);
1766 return (error); /* no reference held on error */
1768 error = finstall(td, fp, &fd, flags, fcaps);
1770 fdrop(fp, td); /* one reference (fp only) */
1774 if (resultfp != NULL)
1775 *resultfp = fp; /* copy out result */
1777 fdrop(fp, td); /* release local reference */
1779 if (resultfd != NULL)
1786 * Create a new open file structure without allocating a file descriptor.
1789 falloc_noinstall(struct thread *td, struct file **resultfp)
1792 int maxuserfiles = maxfiles - (maxfiles / 20);
1794 static struct timeval lastfail;
1797 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1799 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1800 if ((openfiles_new >= maxuserfiles &&
1801 priv_check(td, PRIV_MAXFILES) != 0) ||
1802 openfiles_new >= maxfiles) {
1803 atomic_subtract_int(&openfiles, 1);
1804 if (ppsratecheck(&lastfail, &curfail, 1)) {
1805 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1806 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1810 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1811 refcount_init(&fp->f_count, 1);
1812 fp->f_cred = crhold(td->td_ucred);
1813 fp->f_ops = &badfileops;
1819 * Install a file in a file descriptor table.
1822 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1823 struct filecaps *fcaps)
1825 struct filedescent *fde;
1829 filecaps_validate(fcaps, __func__);
1830 FILEDESC_XLOCK_ASSERT(fdp);
1832 fde = &fdp->fd_ofiles[fd];
1834 seq_write_begin(&fde->fde_seq);
1837 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1839 filecaps_move(fcaps, &fde->fde_caps);
1841 filecaps_fill(&fde->fde_caps);
1843 seq_write_end(&fde->fde_seq);
1848 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1849 struct filecaps *fcaps)
1851 struct filedesc *fdp = td->td_proc->p_fd;
1856 FILEDESC_XLOCK(fdp);
1857 if ((error = fdalloc(td, 0, fd))) {
1858 FILEDESC_XUNLOCK(fdp);
1862 _finstall(fdp, fp, *fd, flags, fcaps);
1863 FILEDESC_XUNLOCK(fdp);
1868 * Build a new filedesc structure from another.
1869 * Copy the current, root, and jail root vnode references.
1871 * If fdp is not NULL, return with it shared locked.
1874 fdinit(struct filedesc *fdp, bool prepfiles)
1876 struct filedesc0 *newfdp0;
1877 struct filedesc *newfdp;
1879 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1880 newfdp = &newfdp0->fd_fd;
1882 /* Create the file descriptor table. */
1883 FILEDESC_LOCK_INIT(newfdp);
1884 refcount_init(&newfdp->fd_refcnt, 1);
1885 refcount_init(&newfdp->fd_holdcnt, 1);
1886 newfdp->fd_cmask = CMASK;
1887 newfdp->fd_map = newfdp0->fd_dmap;
1888 newfdp->fd_lastfile = -1;
1889 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1890 newfdp->fd_files->fdt_nfiles = NDFILE;
1895 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1896 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1898 FILEDESC_SLOCK(fdp);
1899 newfdp->fd_cdir = fdp->fd_cdir;
1900 if (newfdp->fd_cdir)
1901 vrefact(newfdp->fd_cdir);
1902 newfdp->fd_rdir = fdp->fd_rdir;
1903 if (newfdp->fd_rdir)
1904 vrefact(newfdp->fd_rdir);
1905 newfdp->fd_jdir = fdp->fd_jdir;
1906 if (newfdp->fd_jdir)
1907 vrefact(newfdp->fd_jdir);
1910 FILEDESC_SUNLOCK(fdp);
1912 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1913 FILEDESC_SUNLOCK(fdp);
1914 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1915 FILEDESC_SLOCK(fdp);
1922 static struct filedesc *
1923 fdhold(struct proc *p)
1925 struct filedesc *fdp;
1927 PROC_LOCK_ASSERT(p, MA_OWNED);
1930 refcount_acquire(&fdp->fd_holdcnt);
1935 fddrop(struct filedesc *fdp)
1938 if (fdp->fd_holdcnt > 1) {
1939 if (refcount_release(&fdp->fd_holdcnt) == 0)
1943 FILEDESC_LOCK_DESTROY(fdp);
1944 uma_zfree(filedesc0_zone, fdp);
1948 * Share a filedesc structure.
1951 fdshare(struct filedesc *fdp)
1954 refcount_acquire(&fdp->fd_refcnt);
1959 * Unshare a filedesc structure, if necessary by making a copy
1962 fdunshare(struct thread *td)
1964 struct filedesc *tmp;
1965 struct proc *p = td->td_proc;
1967 if (p->p_fd->fd_refcnt == 1)
1970 tmp = fdcopy(p->p_fd);
1976 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
1980 td->td_proc->p_fd = fdp;
1984 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
1985 * this is to ease callers, not catch errors.
1988 fdcopy(struct filedesc *fdp)
1990 struct filedesc *newfdp;
1991 struct filedescent *nfde, *ofde;
1996 newfdp = fdinit(fdp, true);
1997 /* copy all passable descriptors (i.e. not kqueue) */
1998 newfdp->fd_freefile = -1;
1999 for (i = 0; i <= fdp->fd_lastfile; ++i) {
2000 ofde = &fdp->fd_ofiles[i];
2001 if (ofde->fde_file == NULL ||
2002 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2003 if (newfdp->fd_freefile == -1)
2004 newfdp->fd_freefile = i;
2007 nfde = &newfdp->fd_ofiles[i];
2009 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2010 fhold(nfde->fde_file);
2011 fdused_init(newfdp, i);
2012 newfdp->fd_lastfile = i;
2014 if (newfdp->fd_freefile == -1)
2015 newfdp->fd_freefile = i;
2016 newfdp->fd_cmask = fdp->fd_cmask;
2017 FILEDESC_SUNLOCK(fdp);
2022 * Copies a filedesc structure, while remapping all file descriptors
2023 * stored inside using a translation table.
2025 * File descriptors are copied over to the new file descriptor table,
2026 * regardless of whether the close-on-exec flag is set.
2029 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2030 struct filedesc **ret)
2032 struct filedesc *newfdp;
2033 struct filedescent *nfde, *ofde;
2038 newfdp = fdinit(fdp, true);
2039 if (nfds > fdp->fd_lastfile + 1) {
2040 /* New table cannot be larger than the old one. */
2044 /* Copy all passable descriptors (i.e. not kqueue). */
2045 newfdp->fd_freefile = nfds;
2046 for (i = 0; i < nfds; ++i) {
2047 if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2048 /* File descriptor out of bounds. */
2052 ofde = &fdp->fd_ofiles[fds[i]];
2053 if (ofde->fde_file == NULL) {
2054 /* Unused file descriptor. */
2058 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2059 /* File descriptor cannot be passed. */
2063 nfde = &newfdp->fd_ofiles[i];
2065 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2066 fhold(nfde->fde_file);
2067 fdused_init(newfdp, i);
2068 newfdp->fd_lastfile = i;
2070 newfdp->fd_cmask = fdp->fd_cmask;
2071 FILEDESC_SUNLOCK(fdp);
2075 FILEDESC_SUNLOCK(fdp);
2076 fdescfree_remapped(newfdp);
2081 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2082 * one of processes using it exits) and the table used to be shared.
2085 fdclearlocks(struct thread *td)
2087 struct filedesc *fdp;
2088 struct filedesc_to_leader *fdtol;
2098 MPASS(fdtol != NULL);
2100 FILEDESC_XLOCK(fdp);
2101 KASSERT(fdtol->fdl_refcount > 0,
2102 ("filedesc_to_refcount botch: fdl_refcount=%d",
2103 fdtol->fdl_refcount));
2104 if (fdtol->fdl_refcount == 1 &&
2105 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2106 for (i = 0; i <= fdp->fd_lastfile; i++) {
2107 fp = fdp->fd_ofiles[i].fde_file;
2108 if (fp == NULL || fp->f_type != DTYPE_VNODE)
2111 FILEDESC_XUNLOCK(fdp);
2112 lf.l_whence = SEEK_SET;
2115 lf.l_type = F_UNLCK;
2117 (void) VOP_ADVLOCK(vp,
2118 (caddr_t)p->p_leader, F_UNLCK,
2120 FILEDESC_XLOCK(fdp);
2125 if (fdtol->fdl_refcount == 1) {
2126 if (fdp->fd_holdleaderscount > 0 &&
2127 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2129 * close() or kern_dup() has cleared a reference
2130 * in a shared file descriptor table.
2132 fdp->fd_holdleaderswakeup = 1;
2133 sx_sleep(&fdp->fd_holdleaderscount,
2134 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2137 if (fdtol->fdl_holdcount > 0) {
2139 * Ensure that fdtol->fdl_leader remains
2140 * valid in closef().
2142 fdtol->fdl_wakeup = 1;
2143 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2148 fdtol->fdl_refcount--;
2149 if (fdtol->fdl_refcount == 0 &&
2150 fdtol->fdl_holdcount == 0) {
2151 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2152 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2156 FILEDESC_XUNLOCK(fdp);
2158 free(fdtol, M_FILEDESC_TO_LEADER);
2162 * Release a filedesc structure.
2165 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2167 struct filedesc0 *fdp0;
2168 struct freetable *ft, *tft;
2169 struct filedescent *fde;
2173 for (i = 0; i <= fdp->fd_lastfile; i++) {
2174 fde = &fdp->fd_ofiles[i];
2179 (void) closef(fp, td);
2185 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2186 free(fdp->fd_map, M_FILEDESC);
2187 if (fdp->fd_nfiles > NDFILE)
2188 free(fdp->fd_files, M_FILEDESC);
2190 fdp0 = (struct filedesc0 *)fdp;
2191 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2192 free(ft->ft_table, M_FILEDESC);
2198 fdescfree(struct thread *td)
2201 struct filedesc *fdp;
2202 struct vnode *cdir, *jdir, *rdir;
2211 racct_set(p, RACCT_NOFILE, 0);
2216 if (p->p_fdtol != NULL)
2223 if (refcount_release(&fdp->fd_refcnt) == 0)
2226 FILEDESC_XLOCK(fdp);
2227 cdir = fdp->fd_cdir;
2228 fdp->fd_cdir = NULL;
2229 rdir = fdp->fd_rdir;
2230 fdp->fd_rdir = NULL;
2231 jdir = fdp->fd_jdir;
2232 fdp->fd_jdir = NULL;
2233 FILEDESC_XUNLOCK(fdp);
2242 fdescfree_fds(td, fdp, 1);
2246 fdescfree_remapped(struct filedesc *fdp)
2249 if (fdp->fd_cdir != NULL)
2250 vrele(fdp->fd_cdir);
2251 if (fdp->fd_rdir != NULL)
2252 vrele(fdp->fd_rdir);
2253 if (fdp->fd_jdir != NULL)
2254 vrele(fdp->fd_jdir);
2256 fdescfree_fds(curthread, fdp, 0);
2260 * For setugid programs, we don't want to people to use that setugidness
2261 * to generate error messages which write to a file which otherwise would
2262 * otherwise be off-limits to the process. We check for filesystems where
2263 * the vnode can change out from under us after execve (like [lin]procfs).
2265 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2266 * sufficient. We also don't check for setugidness since we know we are.
2269 is_unsafe(struct file *fp)
2273 if (fp->f_type != DTYPE_VNODE)
2277 return ((vp->v_vflag & VV_PROCDEP) != 0);
2281 * Make this setguid thing safe, if at all possible.
2284 fdsetugidsafety(struct thread *td)
2286 struct filedesc *fdp;
2290 fdp = td->td_proc->p_fd;
2291 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2292 MPASS(fdp->fd_nfiles >= 3);
2293 for (i = 0; i <= 2; i++) {
2294 fp = fdp->fd_ofiles[i].fde_file;
2295 if (fp != NULL && is_unsafe(fp)) {
2296 FILEDESC_XLOCK(fdp);
2297 knote_fdclose(td, i);
2299 * NULL-out descriptor prior to close to avoid
2300 * a race while close blocks.
2303 FILEDESC_XUNLOCK(fdp);
2304 (void) closef(fp, td);
2310 * If a specific file object occupies a specific file descriptor, close the
2311 * file descriptor entry and drop a reference on the file object. This is a
2312 * convenience function to handle a subsequent error in a function that calls
2313 * falloc() that handles the race that another thread might have closed the
2314 * file descriptor out from under the thread creating the file object.
2317 fdclose(struct thread *td, struct file *fp, int idx)
2319 struct filedesc *fdp = td->td_proc->p_fd;
2321 FILEDESC_XLOCK(fdp);
2322 if (fdp->fd_ofiles[idx].fde_file == fp) {
2324 FILEDESC_XUNLOCK(fdp);
2327 FILEDESC_XUNLOCK(fdp);
2331 * Close any files on exec?
2334 fdcloseexec(struct thread *td)
2336 struct filedesc *fdp;
2337 struct filedescent *fde;
2341 fdp = td->td_proc->p_fd;
2342 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2343 for (i = 0; i <= fdp->fd_lastfile; i++) {
2344 fde = &fdp->fd_ofiles[i];
2346 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2347 (fde->fde_flags & UF_EXCLOSE))) {
2348 FILEDESC_XLOCK(fdp);
2350 (void) closefp(fdp, i, fp, td, 0);
2351 FILEDESC_UNLOCK_ASSERT(fdp);
2357 * It is unsafe for set[ug]id processes to be started with file
2358 * descriptors 0..2 closed, as these descriptors are given implicit
2359 * significance in the Standard C library. fdcheckstd() will create a
2360 * descriptor referencing /dev/null for each of stdin, stdout, and
2361 * stderr that is not already open.
2364 fdcheckstd(struct thread *td)
2366 struct filedesc *fdp;
2368 int i, error, devnull;
2370 fdp = td->td_proc->p_fd;
2371 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2372 MPASS(fdp->fd_nfiles >= 3);
2374 for (i = 0; i <= 2; i++) {
2375 if (fdp->fd_ofiles[i].fde_file != NULL)
2378 save = td->td_retval[0];
2379 if (devnull != -1) {
2380 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2382 error = kern_openat(td, AT_FDCWD, "/dev/null",
2383 UIO_SYSSPACE, O_RDWR, 0);
2385 devnull = td->td_retval[0];
2386 KASSERT(devnull == i, ("we didn't get our fd"));
2389 td->td_retval[0] = save;
2397 * Internal form of close. Decrement reference count on file structure.
2398 * Note: td may be NULL when closing a file that was being passed in a
2401 * XXXRW: Giant is not required for the caller, but often will be held; this
2402 * makes it moderately likely the Giant will be recursed in the VFS case.
2405 closef(struct file *fp, struct thread *td)
2409 struct filedesc_to_leader *fdtol;
2410 struct filedesc *fdp;
2413 * POSIX record locking dictates that any close releases ALL
2414 * locks owned by this process. This is handled by setting
2415 * a flag in the unlock to free ONLY locks obeying POSIX
2416 * semantics, and not to free BSD-style file locks.
2417 * If the descriptor was in a message, POSIX-style locks
2418 * aren't passed with the descriptor, and the thread pointer
2419 * will be NULL. Callers should be careful only to pass a
2420 * NULL thread pointer when there really is no owning
2421 * context that might have locks, or the locks will be
2424 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2426 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2427 lf.l_whence = SEEK_SET;
2430 lf.l_type = F_UNLCK;
2431 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2432 F_UNLCK, &lf, F_POSIX);
2434 fdtol = td->td_proc->p_fdtol;
2435 if (fdtol != NULL) {
2437 * Handle special case where file descriptor table is
2438 * shared between multiple process leaders.
2440 fdp = td->td_proc->p_fd;
2441 FILEDESC_XLOCK(fdp);
2442 for (fdtol = fdtol->fdl_next;
2443 fdtol != td->td_proc->p_fdtol;
2444 fdtol = fdtol->fdl_next) {
2445 if ((fdtol->fdl_leader->p_flag &
2448 fdtol->fdl_holdcount++;
2449 FILEDESC_XUNLOCK(fdp);
2450 lf.l_whence = SEEK_SET;
2453 lf.l_type = F_UNLCK;
2455 (void) VOP_ADVLOCK(vp,
2456 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2458 FILEDESC_XLOCK(fdp);
2459 fdtol->fdl_holdcount--;
2460 if (fdtol->fdl_holdcount == 0 &&
2461 fdtol->fdl_wakeup != 0) {
2462 fdtol->fdl_wakeup = 0;
2466 FILEDESC_XUNLOCK(fdp);
2469 return (fdrop(fp, td));
2473 * Initialize the file pointer with the specified properties.
2475 * The ops are set with release semantics to be certain that the flags, type,
2476 * and data are visible when ops is. This is to prevent ops methods from being
2477 * called with bad data.
2480 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2485 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2489 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2490 struct file **fpp, struct filecaps *havecapsp)
2492 struct filedescent *fde;
2495 FILEDESC_LOCK_ASSERT(fdp);
2497 fde = fdeget_locked(fdp, fd);
2504 error = cap_check(cap_rights_fde(fde), needrightsp);
2509 if (havecapsp != NULL)
2510 filecaps_copy(&fde->fde_caps, havecapsp, true);
2512 *fpp = fde->fde_file;
2520 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2521 struct file **fpp, struct filecaps *havecapsp)
2523 struct filedesc *fdp = td->td_proc->p_fd;
2525 #ifndef CAPABILITIES
2526 error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2527 if (error == 0 && havecapsp != NULL)
2528 filecaps_fill(havecapsp);
2534 error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2538 if (havecapsp != NULL) {
2539 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2540 havecapsp, false)) {
2546 if (!fd_modified(fdp, fd, seq))
2555 FILEDESC_SLOCK(fdp);
2556 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2559 FILEDESC_SUNLOCK(fdp);
2565 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2566 struct file **fpp, seq_t *seqp)
2569 struct filedescent *fde;
2571 struct fdescenttbl *fdt;
2576 cap_rights_t haverights;
2580 fdt = fdp->fd_files;
2581 if ((u_int)fd >= fdt->fdt_nfiles)
2584 * Fetch the descriptor locklessly. We avoid fdrop() races by
2585 * never raising a refcount above 0. To accomplish this we have
2586 * to use a cmpset loop rather than an atomic_add. The descriptor
2587 * must be re-verified once we acquire a reference to be certain
2588 * that the identity is still correct and we did not lose a race
2589 * due to preemption.
2593 seq = seq_read(fd_seq(fdt, fd));
2594 fde = &fdt->fdt_ofiles[fd];
2595 haverights = *cap_rights_fde(fde);
2597 if (!seq_consistent(fd_seq(fdt, fd), seq))
2600 fp = fdt->fdt_ofiles[fd].fde_file;
2605 error = cap_check(&haverights, needrightsp);
2609 count = fp->f_count;
2613 * Force a reload. Other thread could reallocate the
2614 * table before this fd was closed, so it possible that
2615 * there is a stale fp pointer in cached version.
2617 fdt = *(struct fdescenttbl * volatile *)&(fdp->fd_files);
2621 * Use an acquire barrier to force re-reading of fdt so it is
2622 * refreshed for verification.
2624 if (atomic_fcmpset_acq_int(&fp->f_count, &count, count + 1) == 0)
2626 fdt = fdp->fd_files;
2628 if (seq_consistent_nomb(fd_seq(fdt, fd), seq))
2630 if (fp == fdt->fdt_ofiles[fd].fde_file)
2633 fdrop(fp, curthread);
2645 * Extract the file pointer associated with the specified descriptor for the
2646 * current user process.
2648 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2651 * File's rights will be checked against the capability rights mask.
2653 * If an error occurred the non-zero error is returned and *fpp is set to
2654 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
2655 * responsible for fdrop().
2658 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2659 cap_rights_t *needrightsp, seq_t *seqp)
2661 struct filedesc *fdp;
2666 fdp = td->td_proc->p_fd;
2667 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2670 if (fp->f_ops == &badfileops) {
2676 * FREAD and FWRITE failure return EBADF as per POSIX.
2682 if ((fp->f_flag & flags) == 0)
2686 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2687 ((fp->f_flag & FWRITE) != 0))
2693 KASSERT(0, ("wrong flags"));
2706 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2709 return (_fget(td, fd, fpp, 0, rightsp, NULL));
2713 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2717 #ifndef CAPABILITIES
2718 error = _fget(td, fd, fpp, 0, rightsp, NULL);
2719 if (maxprotp != NULL)
2720 *maxprotp = VM_PROT_ALL;
2722 struct filedesc *fdp = td->td_proc->p_fd;
2725 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2727 error = _fget(td, fd, fpp, 0, rightsp, &seq);
2731 * If requested, convert capability rights to access flags.
2733 if (maxprotp != NULL)
2734 *maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd));
2735 if (!fd_modified(fdp, fd, seq))
2744 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2747 return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2751 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2754 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2758 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2761 struct filedesc *fdp = td->td_proc->p_fd;
2762 #ifndef CAPABILITIES
2763 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2768 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2770 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2773 error = cap_fcntl_check(fdp, fd, needfcntl);
2774 if (!fd_modified(fdp, fd, seq))
2787 * Like fget() but loads the underlying vnode, or returns an error if the
2788 * descriptor does not represent a vnode. Note that pipes use vnodes but
2789 * never have VM objects. The returned vnode will be vref()'d.
2791 * XXX: what about the unused flags ?
2794 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2801 error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2804 if (fp->f_vnode == NULL) {
2816 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2819 return (_fgetvp(td, fd, 0, rightsp, vpp));
2823 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2824 struct filecaps *havecaps, struct vnode **vpp)
2826 struct filedesc *fdp;
2827 struct filecaps caps;
2831 fdp = td->td_proc->p_fd;
2832 error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2835 if (fp->f_ops == &badfileops) {
2839 if (fp->f_vnode == NULL) {
2850 filecaps_free(&caps);
2855 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2858 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2862 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2865 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2870 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2874 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2879 * Handle the last reference to a file being closed.
2882 _fdrop(struct file *fp, struct thread *td)
2886 if (fp->f_count != 0)
2887 panic("fdrop: count %d", fp->f_count);
2888 error = fo_close(fp, td);
2889 atomic_subtract_int(&openfiles, 1);
2891 free(fp->f_advice, M_FADVISE);
2892 uma_zfree(file_zone, fp);
2898 * Apply an advisory lock on a file descriptor.
2900 * Just attempt to get a record lock of the requested type on the entire file
2901 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2903 #ifndef _SYS_SYSPROTO_H_
2911 sys_flock(struct thread *td, struct flock_args *uap)
2916 cap_rights_t rights;
2919 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp);
2922 if (fp->f_type != DTYPE_VNODE) {
2924 return (EOPNOTSUPP);
2928 lf.l_whence = SEEK_SET;
2931 if (uap->how & LOCK_UN) {
2932 lf.l_type = F_UNLCK;
2933 atomic_clear_int(&fp->f_flag, FHASLOCK);
2934 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2937 if (uap->how & LOCK_EX)
2938 lf.l_type = F_WRLCK;
2939 else if (uap->how & LOCK_SH)
2940 lf.l_type = F_RDLCK;
2945 atomic_set_int(&fp->f_flag, FHASLOCK);
2946 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2947 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2953 * Duplicate the specified descriptor to a free descriptor.
2956 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
2957 int openerror, int *indxp)
2959 struct filedescent *newfde, *oldfde;
2963 KASSERT(openerror == ENODEV || openerror == ENXIO,
2964 ("unexpected error %d in %s", openerror, __func__));
2967 * If the to-be-dup'd fd number is greater than the allowed number
2968 * of file descriptors, or the fd to be dup'd has already been
2969 * closed, then reject.
2971 FILEDESC_XLOCK(fdp);
2972 if ((fp = fget_locked(fdp, dfd)) == NULL) {
2973 FILEDESC_XUNLOCK(fdp);
2977 error = fdalloc(td, 0, &indx);
2979 FILEDESC_XUNLOCK(fdp);
2984 * There are two cases of interest here.
2986 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2988 * For ENXIO steal away the file structure from (dfd) and store it in
2989 * (indx). (dfd) is effectively closed by this operation.
2991 switch (openerror) {
2994 * Check that the mode the file is being opened for is a
2995 * subset of the mode of the existing descriptor.
2997 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
2998 fdunused(fdp, indx);
2999 FILEDESC_XUNLOCK(fdp);
3003 newfde = &fdp->fd_ofiles[indx];
3004 oldfde = &fdp->fd_ofiles[dfd];
3006 seq_write_begin(&newfde->fde_seq);
3008 memcpy(newfde, oldfde, fde_change_size);
3009 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps, true);
3011 seq_write_end(&newfde->fde_seq);
3016 * Steal away the file pointer from dfd and stuff it into indx.
3018 newfde = &fdp->fd_ofiles[indx];
3019 oldfde = &fdp->fd_ofiles[dfd];
3021 seq_write_begin(&newfde->fde_seq);
3023 memcpy(newfde, oldfde, fde_change_size);
3024 oldfde->fde_file = NULL;
3027 seq_write_end(&newfde->fde_seq);
3031 FILEDESC_XUNLOCK(fdp);
3037 * This sysctl determines if we will allow a process to chroot(2) if it
3038 * has a directory open:
3039 * 0: disallowed for all processes.
3040 * 1: allowed for processes that were not already chroot(2)'ed.
3041 * 2: allowed for all processes.
3044 static int chroot_allow_open_directories = 1;
3046 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3047 &chroot_allow_open_directories, 0,
3048 "Allow a process to chroot(2) if it has a directory open");
3051 * Helper function for raised chroot(2) security function: Refuse if
3052 * any filedescriptors are open directories.
3055 chroot_refuse_vdir_fds(struct filedesc *fdp)
3061 FILEDESC_LOCK_ASSERT(fdp);
3063 for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3064 fp = fget_locked(fdp, fd);
3067 if (fp->f_type == DTYPE_VNODE) {
3069 if (vp->v_type == VDIR)
3077 * Common routine for kern_chroot() and jail_attach(). The caller is
3078 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3079 * authorize this operation.
3082 pwd_chroot(struct thread *td, struct vnode *vp)
3084 struct filedesc *fdp;
3085 struct vnode *oldvp;
3088 fdp = td->td_proc->p_fd;
3089 FILEDESC_XLOCK(fdp);
3090 if (chroot_allow_open_directories == 0 ||
3091 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3092 error = chroot_refuse_vdir_fds(fdp);
3094 FILEDESC_XUNLOCK(fdp);
3098 oldvp = fdp->fd_rdir;
3101 if (fdp->fd_jdir == NULL) {
3105 FILEDESC_XUNLOCK(fdp);
3111 pwd_chdir(struct thread *td, struct vnode *vp)
3113 struct filedesc *fdp;
3114 struct vnode *oldvp;
3116 fdp = td->td_proc->p_fd;
3117 FILEDESC_XLOCK(fdp);
3118 VNASSERT(vp->v_usecount > 0, vp,
3119 ("chdir to a vnode with zero usecount"));
3120 oldvp = fdp->fd_cdir;
3122 FILEDESC_XUNLOCK(fdp);
3127 * Scan all active processes and prisons to see if any of them have a current
3128 * or root directory of `olddp'. If so, replace them with the new mount point.
3131 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3133 struct filedesc *fdp;
3138 if (vrefcnt(olddp) == 1)
3141 sx_slock(&allproc_lock);
3142 FOREACH_PROC_IN_SYSTEM(p) {
3148 FILEDESC_XLOCK(fdp);
3149 if (fdp->fd_cdir == olddp) {
3151 fdp->fd_cdir = newdp;
3154 if (fdp->fd_rdir == olddp) {
3156 fdp->fd_rdir = newdp;
3159 if (fdp->fd_jdir == olddp) {
3161 fdp->fd_jdir = newdp;
3164 FILEDESC_XUNLOCK(fdp);
3167 sx_sunlock(&allproc_lock);
3168 if (rootvnode == olddp) {
3173 mtx_lock(&prison0.pr_mtx);
3174 if (prison0.pr_root == olddp) {
3176 prison0.pr_root = newdp;
3179 mtx_unlock(&prison0.pr_mtx);
3180 sx_slock(&allprison_lock);
3181 TAILQ_FOREACH(pr, &allprison, pr_list) {
3182 mtx_lock(&pr->pr_mtx);
3183 if (pr->pr_root == olddp) {
3185 pr->pr_root = newdp;
3188 mtx_unlock(&pr->pr_mtx);
3190 sx_sunlock(&allprison_lock);
3195 struct filedesc_to_leader *
3196 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3198 struct filedesc_to_leader *fdtol;
3200 fdtol = malloc(sizeof(struct filedesc_to_leader),
3201 M_FILEDESC_TO_LEADER, M_WAITOK);
3202 fdtol->fdl_refcount = 1;
3203 fdtol->fdl_holdcount = 0;
3204 fdtol->fdl_wakeup = 0;
3205 fdtol->fdl_leader = leader;
3207 FILEDESC_XLOCK(fdp);
3208 fdtol->fdl_next = old->fdl_next;
3209 fdtol->fdl_prev = old;
3210 old->fdl_next = fdtol;
3211 fdtol->fdl_next->fdl_prev = fdtol;
3212 FILEDESC_XUNLOCK(fdp);
3214 fdtol->fdl_next = fdtol;
3215 fdtol->fdl_prev = fdtol;
3221 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3223 struct filedesc *fdp;
3224 int i, count, slots;
3226 if (*(int *)arg1 != 0)
3229 fdp = curproc->p_fd;
3231 FILEDESC_SLOCK(fdp);
3232 slots = NDSLOTS(fdp->fd_lastfile + 1);
3233 for (i = 0; i < slots; i++)
3234 count += bitcountl(fdp->fd_map[i]);
3235 FILEDESC_SUNLOCK(fdp);
3237 return (SYSCTL_OUT(req, &count, sizeof(count)));
3240 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3241 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3242 "Number of open file descriptors");
3245 * Get file structures globally.
3248 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3251 struct filedesc *fdp;
3256 error = sysctl_wire_old_buffer(req, 0);
3259 if (req->oldptr == NULL) {
3261 sx_slock(&allproc_lock);
3262 FOREACH_PROC_IN_SYSTEM(p) {
3264 if (p->p_state == PRS_NEW) {
3272 /* overestimates sparse tables. */
3273 if (fdp->fd_lastfile > 0)
3274 n += fdp->fd_lastfile;
3277 sx_sunlock(&allproc_lock);
3278 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3281 bzero(&xf, sizeof(xf));
3282 xf.xf_size = sizeof(xf);
3283 sx_slock(&allproc_lock);
3284 FOREACH_PROC_IN_SYSTEM(p) {
3286 if (p->p_state == PRS_NEW) {
3290 if (p_cansee(req->td, p) != 0) {
3294 xf.xf_pid = p->p_pid;
3295 xf.xf_uid = p->p_ucred->cr_uid;
3300 FILEDESC_SLOCK(fdp);
3301 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3302 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3306 xf.xf_data = fp->f_data;
3307 xf.xf_vnode = fp->f_vnode;
3308 xf.xf_type = fp->f_type;
3309 xf.xf_count = fp->f_count;
3311 xf.xf_offset = foffset_get(fp);
3312 xf.xf_flag = fp->f_flag;
3313 error = SYSCTL_OUT(req, &xf, sizeof(xf));
3317 FILEDESC_SUNLOCK(fdp);
3322 sx_sunlock(&allproc_lock);
3326 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3327 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3329 #ifdef KINFO_FILE_SIZE
3330 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3334 xlate_fflags(int fflags)
3336 static const struct {
3339 } fflags_table[] = {
3340 { FAPPEND, KF_FLAG_APPEND },
3341 { FASYNC, KF_FLAG_ASYNC },
3342 { FFSYNC, KF_FLAG_FSYNC },
3343 { FHASLOCK, KF_FLAG_HASLOCK },
3344 { FNONBLOCK, KF_FLAG_NONBLOCK },
3345 { FREAD, KF_FLAG_READ },
3346 { FWRITE, KF_FLAG_WRITE },
3347 { O_CREAT, KF_FLAG_CREAT },
3348 { O_DIRECT, KF_FLAG_DIRECT },
3349 { O_EXCL, KF_FLAG_EXCL },
3350 { O_EXEC, KF_FLAG_EXEC },
3351 { O_EXLOCK, KF_FLAG_EXLOCK },
3352 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3353 { O_SHLOCK, KF_FLAG_SHLOCK },
3354 { O_TRUNC, KF_FLAG_TRUNC }
3360 for (i = 0; i < nitems(fflags_table); i++)
3361 if (fflags & fflags_table[i].fflag)
3362 kflags |= fflags_table[i].kf_fflag;
3366 /* Trim unused data from kf_path by truncating the structure size. */
3368 pack_kinfo(struct kinfo_file *kif)
3371 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3372 strlen(kif->kf_path) + 1;
3373 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3377 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3378 struct kinfo_file *kif, struct filedesc *fdp, int flags)
3382 bzero(kif, sizeof(*kif));
3384 /* Set a default type to allow for empty fill_kinfo() methods. */
3385 kif->kf_type = KF_TYPE_UNKNOWN;
3386 kif->kf_flags = xlate_fflags(fp->f_flag);
3387 if (rightsp != NULL)
3388 kif->kf_cap_rights = *rightsp;
3390 cap_rights_init(&kif->kf_cap_rights);
3392 kif->kf_ref_count = fp->f_count;
3393 kif->kf_offset = foffset_get(fp);
3396 * This may drop the filedesc lock, so the 'fp' cannot be
3397 * accessed after this call.
3399 error = fo_fill_kinfo(fp, kif, fdp);
3401 kif->kf_status |= KF_ATTR_VALID;
3402 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3405 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3409 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3410 struct kinfo_file *kif, int flags)
3414 bzero(kif, sizeof(*kif));
3416 kif->kf_type = KF_TYPE_VNODE;
3417 error = vn_fill_kinfo_vnode(vp, kif);
3419 kif->kf_status |= KF_ATTR_VALID;
3420 kif->kf_flags = xlate_fflags(fflags);
3421 cap_rights_init(&kif->kf_cap_rights);
3423 kif->kf_ref_count = -1;
3424 kif->kf_offset = -1;
3425 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3428 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3432 struct export_fd_buf {
3433 struct filedesc *fdp;
3436 struct kinfo_file kif;
3441 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3443 struct kinfo_file *kif;
3446 if (efbuf->remainder != -1) {
3447 if (efbuf->remainder < kif->kf_structsize) {
3448 /* Terminate export. */
3449 efbuf->remainder = 0;
3452 efbuf->remainder -= kif->kf_structsize;
3454 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3458 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3459 struct export_fd_buf *efbuf)
3463 if (efbuf->remainder == 0)
3465 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3467 FILEDESC_SUNLOCK(efbuf->fdp);
3468 error = export_kinfo_to_sb(efbuf);
3469 FILEDESC_SLOCK(efbuf->fdp);
3474 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3475 struct export_fd_buf *efbuf)
3479 if (efbuf->remainder == 0)
3481 if (efbuf->fdp != NULL)
3482 FILEDESC_SUNLOCK(efbuf->fdp);
3483 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3484 error = export_kinfo_to_sb(efbuf);
3485 if (efbuf->fdp != NULL)
3486 FILEDESC_SLOCK(efbuf->fdp);
3491 * Store a process file descriptor information to sbuf.
3493 * Takes a locked proc as argument, and returns with the proc unlocked.
3496 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
3500 struct filedesc *fdp;
3501 struct export_fd_buf *efbuf;
3502 struct vnode *cttyvp, *textvp, *tracevp;
3504 cap_rights_t rights;
3506 PROC_LOCK_ASSERT(p, MA_OWNED);
3509 tracevp = p->p_tracevp;
3510 if (tracevp != NULL)
3513 textvp = p->p_textvp;
3516 /* Controlling tty. */
3518 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3519 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3525 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3528 efbuf->remainder = maxlen;
3529 efbuf->flags = flags;
3530 if (tracevp != NULL)
3531 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3534 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3536 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3542 FILEDESC_SLOCK(fdp);
3543 /* working directory */
3544 if (fdp->fd_cdir != NULL) {
3545 vrefact(fdp->fd_cdir);
3546 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3548 /* root directory */
3549 if (fdp->fd_rdir != NULL) {
3550 vrefact(fdp->fd_rdir);
3551 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3553 /* jail directory */
3554 if (fdp->fd_jdir != NULL) {
3555 vrefact(fdp->fd_jdir);
3556 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3558 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3559 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3562 rights = *cap_rights(fdp, i);
3563 #else /* !CAPABILITIES */
3564 cap_rights_init(&rights);
3567 * Create sysctl entry. It is OK to drop the filedesc
3568 * lock inside of export_file_to_sb() as we will
3569 * re-validate and re-evaluate its properties when the
3572 error = export_file_to_sb(fp, i, &rights, efbuf);
3573 if (error != 0 || efbuf->remainder == 0)
3576 FILEDESC_SUNLOCK(fdp);
3579 free(efbuf, M_TEMP);
3583 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
3586 * Get per-process file descriptors for use by procstat(1), et al.
3589 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3594 int error, error2, *name;
3598 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3599 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3600 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3605 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3606 error = kern_proc_filedesc_out(p, &sb, maxlen,
3607 KERN_FILEDESC_PACK_KINFO);
3608 error2 = sbuf_finish(&sb);
3610 return (error != 0 ? error : error2);
3613 #ifdef COMPAT_FREEBSD7
3614 #ifdef KINFO_OFILE_SIZE
3615 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3619 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3622 okif->kf_structsize = sizeof(*okif);
3623 okif->kf_type = kif->kf_type;
3624 okif->kf_fd = kif->kf_fd;
3625 okif->kf_ref_count = kif->kf_ref_count;
3626 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3627 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3628 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3629 okif->kf_offset = kif->kf_offset;
3630 if (kif->kf_type == KF_TYPE_VNODE)
3631 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
3633 okif->kf_vnode_type = KF_VTYPE_VNON;
3634 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3635 if (kif->kf_type == KF_TYPE_SOCKET) {
3636 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
3637 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
3638 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
3639 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
3640 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
3642 okif->kf_sa_local.ss_family = AF_UNSPEC;
3643 okif->kf_sa_peer.ss_family = AF_UNSPEC;
3648 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3649 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3654 FILEDESC_SUNLOCK(fdp);
3655 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3656 kinfo_to_okinfo(kif, okif);
3657 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3658 FILEDESC_SLOCK(fdp);
3663 * Get per-process file descriptors for use by procstat(1), et al.
3666 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3668 struct kinfo_ofile *okif;
3669 struct kinfo_file *kif;
3670 struct filedesc *fdp;
3671 int error, i, *name;
3676 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3683 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3684 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3685 FILEDESC_SLOCK(fdp);
3686 if (fdp->fd_cdir != NULL)
3687 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3689 if (fdp->fd_rdir != NULL)
3690 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3692 if (fdp->fd_jdir != NULL)
3693 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3695 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3696 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3698 export_file_to_kinfo(fp, i, NULL, kif, fdp,
3699 KERN_FILEDESC_PACK_KINFO);
3700 FILEDESC_SUNLOCK(fdp);
3701 kinfo_to_okinfo(kif, okif);
3702 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3703 FILEDESC_SLOCK(fdp);
3707 FILEDESC_SUNLOCK(fdp);
3714 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3715 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3716 "Process ofiledesc entries");
3717 #endif /* COMPAT_FREEBSD7 */
3720 vntype_to_kinfo(int vtype)
3725 } vtypes_table[] = {
3726 { VBAD, KF_VTYPE_VBAD },
3727 { VBLK, KF_VTYPE_VBLK },
3728 { VCHR, KF_VTYPE_VCHR },
3729 { VDIR, KF_VTYPE_VDIR },
3730 { VFIFO, KF_VTYPE_VFIFO },
3731 { VLNK, KF_VTYPE_VLNK },
3732 { VNON, KF_VTYPE_VNON },
3733 { VREG, KF_VTYPE_VREG },
3734 { VSOCK, KF_VTYPE_VSOCK }
3739 * Perform vtype translation.
3741 for (i = 0; i < nitems(vtypes_table); i++)
3742 if (vtypes_table[i].vtype == vtype)
3743 return (vtypes_table[i].kf_vtype);
3745 return (KF_VTYPE_UNKNOWN);
3748 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3749 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3750 "Process filedesc entries");
3753 * Store a process current working directory information to sbuf.
3755 * Takes a locked proc as argument, and returns with the proc unlocked.
3758 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
3760 struct filedesc *fdp;
3761 struct export_fd_buf *efbuf;
3764 PROC_LOCK_ASSERT(p, MA_OWNED);
3771 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3774 efbuf->remainder = maxlen;
3776 FILEDESC_SLOCK(fdp);
3777 if (fdp->fd_cdir == NULL)
3780 vrefact(fdp->fd_cdir);
3781 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3784 FILEDESC_SUNLOCK(fdp);
3786 free(efbuf, M_TEMP);
3791 * Get per-process current working directory.
3794 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3799 int error, error2, *name;
3803 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3804 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3805 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3810 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3811 error = kern_proc_cwd_out(p, &sb, maxlen);
3812 error2 = sbuf_finish(&sb);
3814 return (error != 0 ? error : error2);
3817 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3818 sysctl_kern_proc_cwd, "Process current working directory");
3822 * For the purposes of debugging, generate a human-readable string for the
3826 file_type_to_name(short type)
3854 case DTYPE_PROCDESC:
3856 case DTYPE_LINUXEFD:
3858 case DTYPE_LINUXTFD:
3866 * For the purposes of debugging, identify a process (if any, perhaps one of
3867 * many) that references the passed file in its file descriptor array. Return
3870 static struct proc *
3871 file_to_first_proc(struct file *fp)
3873 struct filedesc *fdp;
3877 FOREACH_PROC_IN_SYSTEM(p) {
3878 if (p->p_state == PRS_NEW)
3883 for (n = 0; n <= fdp->fd_lastfile; n++) {
3884 if (fp == fdp->fd_ofiles[n].fde_file)
3892 db_print_file(struct file *fp, int header)
3894 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
3898 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
3899 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
3900 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
3902 p = file_to_first_proc(fp);
3903 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
3904 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
3905 fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode,
3906 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3911 DB_SHOW_COMMAND(file, db_show_file)
3916 db_printf("usage: show file <addr>\n");
3919 fp = (struct file *)addr;
3920 db_print_file(fp, 1);
3923 DB_SHOW_COMMAND(files, db_show_files)
3925 struct filedesc *fdp;
3932 FOREACH_PROC_IN_SYSTEM(p) {
3933 if (p->p_state == PRS_NEW)
3935 if ((fdp = p->p_fd) == NULL)
3937 for (n = 0; n <= fdp->fd_lastfile; ++n) {
3938 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3940 db_print_file(fp, header);
3947 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3948 &maxfilesperproc, 0, "Maximum files allowed open per process");
3950 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3951 &maxfiles, 0, "Maximum number of files");
3953 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3954 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
3958 filelistinit(void *dummy)
3961 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3962 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
3963 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
3964 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
3965 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3967 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3969 /*-------------------------------------------------------------------*/
3972 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
3973 int flags, struct thread *td)
3980 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
3988 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
3996 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4004 badfo_kqfilter(struct file *fp, struct knote *kn)
4011 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4019 badfo_close(struct file *fp, struct thread *td)
4026 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4034 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4042 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4043 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4051 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4057 struct fileops badfileops = {
4058 .fo_read = badfo_readwrite,
4059 .fo_write = badfo_readwrite,
4060 .fo_truncate = badfo_truncate,
4061 .fo_ioctl = badfo_ioctl,
4062 .fo_poll = badfo_poll,
4063 .fo_kqfilter = badfo_kqfilter,
4064 .fo_stat = badfo_stat,
4065 .fo_close = badfo_close,
4066 .fo_chmod = badfo_chmod,
4067 .fo_chown = badfo_chown,
4068 .fo_sendfile = badfo_sendfile,
4069 .fo_fill_kinfo = badfo_fill_kinfo,
4073 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4074 int flags, struct thread *td)
4077 return (EOPNOTSUPP);
4081 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4089 invfo_ioctl(struct file *fp, u_long com, void *data,
4090 struct ucred *active_cred, struct thread *td)
4097 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4101 return (poll_no_poll(events));
4105 invfo_kqfilter(struct file *fp, struct knote *kn)
4112 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4120 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4128 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4129 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4136 /*-------------------------------------------------------------------*/
4139 * File Descriptor pseudo-device driver (/dev/fd/).
4141 * Opening minor device N dup()s the file (if any) connected to file
4142 * descriptor N belonging to the calling process. Note that this driver
4143 * consists of only the ``open()'' routine, because all subsequent
4144 * references to this file will be direct to the other driver.
4146 * XXX: we could give this one a cloning event handler if necessary.
4151 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4155 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4156 * the file descriptor being sought for duplication. The error
4157 * return ensures that the vnode for this device will be released
4158 * by vn_open. Open will detect this special error and take the
4159 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4160 * will simply report the error.
4162 td->td_dupfd = dev2unit(dev);
4166 static struct cdevsw fildesc_cdevsw = {
4167 .d_version = D_VERSION,
4173 fildesc_drvinit(void *unused)
4177 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4178 UID_ROOT, GID_WHEEL, 0666, "fd/0");
4179 make_dev_alias(dev, "stdin");
4180 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4181 UID_ROOT, GID_WHEEL, 0666, "fd/1");
4182 make_dev_alias(dev, "stdout");
4183 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4184 UID_ROOT, GID_WHEEL, 0666, "fd/2");
4185 make_dev_alias(dev, "stderr");
4188 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);