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"
44 #include "opt_ktrace.h"
46 #include <sys/param.h>
47 #include <sys/systm.h>
49 #include <sys/capsicum.h>
51 #include <sys/fcntl.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
66 #include <sys/protosw.h>
67 #include <sys/racct.h>
68 #include <sys/resourcevar.h>
70 #include <sys/signalvar.h>
74 #include <sys/syscallsubr.h>
75 #include <sys/sysctl.h>
76 #include <sys/sysproto.h>
77 #include <sys/unistd.h>
79 #include <sys/vnode.h>
81 #include <sys/ktrace.h>
86 #include <security/audit/audit.h>
93 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
94 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
95 "file desc to leader structures");
96 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
97 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
99 MALLOC_DECLARE(M_FADVISE);
101 static __read_mostly uma_zone_t file_zone;
102 static __read_mostly uma_zone_t filedesc0_zone;
104 static int closefp(struct filedesc *fdp, int fd, struct file *fp,
105 struct thread *td, int holdleaders);
106 static int fd_first_free(struct filedesc *fdp, int low, int size);
107 static int fd_last_used(struct filedesc *fdp, int size);
108 static void fdgrowtable(struct filedesc *fdp, int nfd);
109 static void fdgrowtable_exp(struct filedesc *fdp, int nfd);
110 static void fdunused(struct filedesc *fdp, int fd);
111 static void fdused(struct filedesc *fdp, int fd);
112 static int getmaxfd(struct thread *td);
113 static u_long *filecaps_copy_prep(const struct filecaps *src);
114 static void filecaps_copy_finish(const struct filecaps *src,
115 struct filecaps *dst, u_long *ioctls);
116 static u_long *filecaps_free_prep(struct filecaps *fcaps);
117 static void filecaps_free_finish(u_long *ioctls);
122 * - An array of open file descriptors (fd_ofiles)
123 * - An array of file flags (fd_ofileflags)
124 * - A bitmap recording which descriptors are in use (fd_map)
126 * A process starts out with NDFILE descriptors. The value of NDFILE has
127 * been selected based the historical limit of 20 open files, and an
128 * assumption that the majority of processes, especially short-lived
129 * processes like shells, will never need more.
131 * If this initial allocation is exhausted, a larger descriptor table and
132 * map are allocated dynamically, and the pointers in the process's struct
133 * filedesc are updated to point to those. This is repeated every time
134 * the process runs out of file descriptors (provided it hasn't hit its
137 * Since threads may hold references to individual descriptor table
138 * entries, the tables are never freed. Instead, they are placed on a
139 * linked list and freed only when the struct filedesc is released.
142 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
143 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
144 #define NDSLOT(x) ((x) / NDENTRIES)
145 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
146 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
149 * SLIST entry used to keep track of ofiles which must be reclaimed when
153 struct fdescenttbl *ft_table;
154 SLIST_ENTRY(freetable) ft_next;
158 * Initial allocation: a filedesc structure + the head of SLIST used to
159 * keep track of old ofiles + enough space for NDFILE descriptors.
162 struct fdescenttbl0 {
164 struct filedescent fdt_ofiles[NDFILE];
168 struct filedesc fd_fd;
169 SLIST_HEAD(, freetable) fd_free;
170 struct fdescenttbl0 fd_dfiles;
171 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
175 * Descriptor management.
177 volatile int __exclusive_cache_line openfiles; /* actual number of open files */
178 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
179 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
182 * If low >= size, just return low. Otherwise find the first zero bit in the
183 * given bitmap, starting at low and not exceeding size - 1. Return size if
187 fd_first_free(struct filedesc *fdp, int low, int size)
189 NDSLOTTYPE *map = fdp->fd_map;
197 if (low % NDENTRIES) {
198 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
199 if ((mask &= ~map[off]) != 0UL)
200 return (off * NDENTRIES + ffsl(mask) - 1);
203 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
204 if (map[off] != ~0UL)
205 return (off * NDENTRIES + ffsl(~map[off]) - 1);
210 * Find the highest non-zero bit in the given bitmap, starting at 0 and
211 * not exceeding size - 1. Return -1 if not found.
214 fd_last_used(struct filedesc *fdp, int size)
216 NDSLOTTYPE *map = fdp->fd_map;
221 if (size % NDENTRIES) {
222 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
223 if ((mask &= map[off]) != 0)
224 return (off * NDENTRIES + flsl(mask) - 1);
227 for (minoff = NDSLOT(0); off >= minoff; --off)
229 return (off * NDENTRIES + flsl(map[off]) - 1);
234 fdisused(struct filedesc *fdp, int fd)
237 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
238 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
240 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
244 * Mark a file descriptor as used.
247 fdused_init(struct filedesc *fdp, int fd)
250 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
252 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
256 fdused(struct filedesc *fdp, int fd)
259 FILEDESC_XLOCK_ASSERT(fdp);
261 fdused_init(fdp, fd);
262 if (fd > fdp->fd_lastfile)
263 fdp->fd_lastfile = fd;
264 if (fd == fdp->fd_freefile)
265 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
269 * Mark a file descriptor as unused.
272 fdunused(struct filedesc *fdp, int fd)
275 FILEDESC_XLOCK_ASSERT(fdp);
277 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
278 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
279 ("fd=%d is still in use", fd));
281 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
282 if (fd < fdp->fd_freefile)
283 fdp->fd_freefile = fd;
284 if (fd == fdp->fd_lastfile)
285 fdp->fd_lastfile = fd_last_used(fdp, fd);
289 * Free a file descriptor.
291 * Avoid some work if fdp is about to be destroyed.
294 fdefree_last(struct filedescent *fde)
297 filecaps_free(&fde->fde_caps);
301 fdfree(struct filedesc *fdp, int fd)
303 struct filedescent *fde;
305 fde = &fdp->fd_ofiles[fd];
307 seq_write_begin(&fde->fde_seq);
309 fde->fde_file = NULL;
311 seq_write_end(&fde->fde_seq);
318 pwd_ensure_dirs(void)
320 struct filedesc *fdp;
324 if (fdp->fd_cdir == NULL) {
325 fdp->fd_cdir = rootvnode;
328 if (fdp->fd_rdir == NULL) {
329 fdp->fd_rdir = rootvnode;
332 FILEDESC_XUNLOCK(fdp);
336 * System calls on descriptors.
338 #ifndef _SYS_SYSPROTO_H_
339 struct getdtablesize_args {
345 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
352 min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc);
354 PROC_LOCK(td->td_proc);
355 lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
356 PROC_UNLOCK(td->td_proc);
357 if (lim < td->td_retval[0])
358 td->td_retval[0] = lim;
364 * Duplicate a file descriptor to a particular value.
366 * Note: keep in mind that a potential race condition exists when closing
367 * descriptors from a shared descriptor table (via rfork).
369 #ifndef _SYS_SYSPROTO_H_
377 sys_dup2(struct thread *td, struct dup2_args *uap)
380 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
384 * Duplicate a file descriptor.
386 #ifndef _SYS_SYSPROTO_H_
393 sys_dup(struct thread *td, struct dup_args *uap)
396 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
400 * The file control system call.
402 #ifndef _SYS_SYSPROTO_H_
411 sys_fcntl(struct thread *td, struct fcntl_args *uap)
414 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
418 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
432 * Convert old flock structure to new.
434 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
435 fl.l_start = ofl.l_start;
436 fl.l_len = ofl.l_len;
437 fl.l_pid = ofl.l_pid;
438 fl.l_type = ofl.l_type;
439 fl.l_whence = ofl.l_whence;
453 arg1 = (intptr_t)&fl;
459 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
460 arg1 = (intptr_t)&fl;
468 error = kern_fcntl(td, fd, newcmd, arg1);
471 if (cmd == F_OGETLK) {
472 ofl.l_start = fl.l_start;
473 ofl.l_len = fl.l_len;
474 ofl.l_pid = fl.l_pid;
475 ofl.l_type = fl.l_type;
476 ofl.l_whence = fl.l_whence;
477 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
478 } else if (cmd == F_GETLK) {
479 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
485 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
487 struct filedesc *fdp;
489 struct file *fp, *fp2;
490 struct filedescent *fde;
507 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
510 case F_DUPFD_CLOEXEC:
512 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
517 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
520 case F_DUP2FD_CLOEXEC:
522 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
528 fde = fdeget_locked(fdp, fd);
531 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
534 FILEDESC_SUNLOCK(fdp);
540 fde = fdeget_locked(fdp, fd);
542 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
543 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
546 FILEDESC_XUNLOCK(fdp);
550 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
553 td->td_retval[0] = OFLAGS(fp->f_flag);
558 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
562 tmp = flg = fp->f_flag;
564 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
565 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
566 tmp = fp->f_flag & FNONBLOCK;
567 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
572 tmp = fp->f_flag & FASYNC;
573 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
578 atomic_clear_int(&fp->f_flag, FNONBLOCK);
580 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
585 error = fget_fcntl(td, fd, &cap_fcntl_rights, 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, &cap_fcntl_rights, F_SETOWN, &fp);
599 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
604 error = priv_check(td, PRIV_NFS_LOCKD);
612 /* FALLTHROUGH F_SETLK */
616 flp = (struct flock *)arg;
617 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
622 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL);
625 if (fp->f_type != DTYPE_VNODE) {
631 if (flp->l_whence == SEEK_CUR) {
632 foffset = foffset_get(fp);
635 foffset > OFF_MAX - flp->l_start)) {
640 flp->l_start += foffset;
644 switch (flp->l_type) {
646 if ((fp->f_flag & FREAD) == 0) {
650 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
651 PROC_LOCK(p->p_leader);
652 p->p_leader->p_flag |= P_ADVLOCK;
653 PROC_UNLOCK(p->p_leader);
655 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
659 if ((fp->f_flag & FWRITE) == 0) {
663 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
664 PROC_LOCK(p->p_leader);
665 p->p_leader->p_flag |= P_ADVLOCK;
666 PROC_UNLOCK(p->p_leader);
668 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
672 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
676 if (flg != F_REMOTE) {
680 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
681 F_UNLCKSYS, flp, flg);
687 if (error != 0 || flp->l_type == F_UNLCK ||
688 flp->l_type == F_UNLCKSYS) {
694 * Check for a race with close.
696 * The vnode is now advisory locked (or unlocked, but this case
697 * is not really important) as the caller requested.
698 * We had to drop the filedesc lock, so we need to recheck if
699 * the descriptor is still valid, because if it was closed
700 * in the meantime we need to remove advisory lock from the
701 * vnode - close on any descriptor leading to an advisory
702 * locked vnode, removes that lock.
703 * We will return 0 on purpose in that case, as the result of
704 * successful advisory lock might have been externally visible
705 * already. This is fine - effectively we pretend to the caller
706 * that the closing thread was a bit slower and that the
707 * advisory lock succeeded before the close.
709 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2, NULL);
715 flp->l_whence = SEEK_SET;
718 flp->l_type = F_UNLCK;
719 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
720 F_UNLCK, flp, F_POSIX);
727 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL);
730 if (fp->f_type != DTYPE_VNODE) {
735 flp = (struct flock *)arg;
736 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
737 flp->l_type != F_UNLCK) {
742 if (flp->l_whence == SEEK_CUR) {
743 foffset = foffset_get(fp);
744 if ((flp->l_start > 0 &&
745 foffset > OFF_MAX - flp->l_start) ||
747 foffset < OFF_MIN - flp->l_start)) {
752 flp->l_start += foffset;
755 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
761 arg = arg ? 128 * 1024: 0;
764 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp, NULL);
767 if (fp->f_type != DTYPE_VNODE) {
774 * Exclusive lock synchronizes against f_seqcount reads and
775 * writes in sequential_heuristic().
777 error = vn_lock(vp, LK_EXCLUSIVE);
783 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
784 fp->f_seqcount = (arg + bsize - 1) / bsize;
785 atomic_set_int(&fp->f_flag, FRDAHEAD);
787 atomic_clear_int(&fp->f_flag, FRDAHEAD);
801 getmaxfd(struct thread *td)
804 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
808 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
811 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
813 struct filedesc *fdp;
814 struct filedescent *oldfde, *newfde;
817 u_long *oioctls, *nioctls;
823 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
824 MPASS(mode < FDDUP_LASTMODE);
827 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
830 * Verify we have a valid descriptor to dup from and possibly to
831 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
832 * return EINVAL when the new descriptor is out of bounds.
837 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
838 maxfd = getmaxfd(td);
840 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
844 if (fget_locked(fdp, old) == NULL)
846 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
847 td->td_retval[0] = new;
848 if (flags & FDDUP_FLAG_CLOEXEC)
849 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
855 * If the caller specified a file descriptor, make sure the file
856 * table is large enough to hold it, and grab it. Otherwise, just
857 * allocate a new descriptor the usual way.
862 if ((error = fdalloc(td, new, &new)) != 0)
865 case FDDUP_MUSTREPLACE:
866 /* Target file descriptor must exist. */
867 if (fget_locked(fdp, new) == NULL)
871 if (new >= fdp->fd_nfiles) {
873 * The resource limits are here instead of e.g.
874 * fdalloc(), because the file descriptor table may be
875 * shared between processes, so we can't really use
876 * racct_add()/racct_sub(). Instead of counting the
877 * number of actually allocated descriptors, just put
878 * the limit on the size of the file descriptor table.
883 error = racct_set(p, RACCT_NOFILE, new + 1);
891 fdgrowtable_exp(fdp, new + 1);
893 if (!fdisused(fdp, new))
897 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
900 KASSERT(old != new, ("new fd is same as old"));
902 oldfde = &fdp->fd_ofiles[old];
903 fhold(oldfde->fde_file);
904 newfde = &fdp->fd_ofiles[new];
905 delfp = newfde->fde_file;
907 oioctls = filecaps_free_prep(&newfde->fde_caps);
908 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
911 * Duplicate the source descriptor.
914 seq_write_begin(&newfde->fde_seq);
916 memcpy(newfde, oldfde, fde_change_size);
917 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
919 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
920 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
922 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
924 seq_write_end(&newfde->fde_seq);
926 filecaps_free_finish(oioctls);
927 td->td_retval[0] = new;
932 (void) closefp(fdp, new, delfp, td, 1);
933 FILEDESC_UNLOCK_ASSERT(fdp);
936 FILEDESC_XUNLOCK(fdp);
943 * If sigio is on the list associated with a process or process group,
944 * disable signalling from the device, remove sigio from the list and
948 funsetown(struct sigio **sigiop)
960 *(sigio->sio_myref) = NULL;
961 if ((sigio)->sio_pgid < 0) {
962 struct pgrp *pg = (sigio)->sio_pgrp;
964 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
968 struct proc *p = (sigio)->sio_proc;
970 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
975 crfree(sigio->sio_ucred);
976 free(sigio, M_SIGIO);
980 * Free a list of sigio structures.
981 * We only need to lock the SIGIO_LOCK because we have made ourselves
982 * inaccessible to callers of fsetown and therefore do not need to lock
983 * the proc or pgrp struct for the list manipulation.
986 funsetownlst(struct sigiolst *sigiolst)
992 sigio = SLIST_FIRST(sigiolst);
999 * Every entry of the list should belong
1000 * to a single proc or pgrp.
1002 if (sigio->sio_pgid < 0) {
1003 pg = sigio->sio_pgrp;
1004 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
1005 } else /* if (sigio->sio_pgid > 0) */ {
1006 p = sigio->sio_proc;
1007 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1011 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1012 *(sigio->sio_myref) = NULL;
1014 KASSERT(sigio->sio_pgid < 0,
1015 ("Proc sigio in pgrp sigio list"));
1016 KASSERT(sigio->sio_pgrp == pg,
1017 ("Bogus pgrp in sigio list"));
1019 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1022 } else /* if (p != NULL) */ {
1023 KASSERT(sigio->sio_pgid > 0,
1024 ("Pgrp sigio in proc sigio list"));
1025 KASSERT(sigio->sio_proc == p,
1026 ("Bogus proc in sigio list"));
1028 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1033 crfree(sigio->sio_ucred);
1034 free(sigio, M_SIGIO);
1041 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1043 * After permission checking, add a sigio structure to the sigio list for
1044 * the process or process group.
1047 fsetown(pid_t pgid, struct sigio **sigiop)
1051 struct sigio *sigio;
1061 /* Allocate and fill in the new sigio out of locks. */
1062 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1063 sigio->sio_pgid = pgid;
1064 sigio->sio_ucred = crhold(curthread->td_ucred);
1065 sigio->sio_myref = sigiop;
1067 sx_slock(&proctree_lock);
1076 * Policy - Don't allow a process to FSETOWN a process
1077 * in another session.
1079 * Remove this test to allow maximum flexibility or
1080 * restrict FSETOWN to the current process or process
1081 * group for maximum safety.
1084 if (proc->p_session != curthread->td_proc->p_session) {
1090 } else /* if (pgid < 0) */ {
1091 pgrp = pgfind(-pgid);
1099 * Policy - Don't allow a process to FSETOWN a process
1100 * in another session.
1102 * Remove this test to allow maximum flexibility or
1103 * restrict FSETOWN to the current process or process
1104 * group for maximum safety.
1106 if (pgrp->pg_session != curthread->td_proc->p_session) {
1117 * Since funsetownlst() is called without the proctree
1118 * locked, we need to check for P_WEXIT.
1119 * XXX: is ESRCH correct?
1121 if ((proc->p_flag & P_WEXIT) != 0) {
1126 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1127 sigio->sio_proc = proc;
1131 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1132 sigio->sio_pgrp = pgrp;
1135 sx_sunlock(&proctree_lock);
1142 sx_sunlock(&proctree_lock);
1143 crfree(sigio->sio_ucred);
1144 free(sigio, M_SIGIO);
1149 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1152 fgetown(struct sigio **sigiop)
1157 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1163 * Function drops the filedesc lock on return.
1166 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1171 FILEDESC_XLOCK_ASSERT(fdp);
1174 if (td->td_proc->p_fdtol != NULL) {
1176 * Ask fdfree() to sleep to ensure that all relevant
1177 * process leaders can be traversed in closef().
1179 fdp->fd_holdleaderscount++;
1186 * We now hold the fp reference that used to be owned by the
1187 * descriptor array. We have to unlock the FILEDESC *AFTER*
1188 * knote_fdclose to prevent a race of the fd getting opened, a knote
1189 * added, and deleteing a knote for the new fd.
1191 knote_fdclose(td, fd);
1194 * We need to notify mqueue if the object is of type mqueue.
1196 if (fp->f_type == DTYPE_MQUEUE)
1197 mq_fdclose(td, fd, fp);
1198 FILEDESC_XUNLOCK(fdp);
1200 error = closef(fp, td);
1202 FILEDESC_XLOCK(fdp);
1203 fdp->fd_holdleaderscount--;
1204 if (fdp->fd_holdleaderscount == 0 &&
1205 fdp->fd_holdleaderswakeup != 0) {
1206 fdp->fd_holdleaderswakeup = 0;
1207 wakeup(&fdp->fd_holdleaderscount);
1209 FILEDESC_XUNLOCK(fdp);
1215 * Close a file descriptor.
1217 #ifndef _SYS_SYSPROTO_H_
1224 sys_close(struct thread *td, struct close_args *uap)
1227 return (kern_close(td, uap->fd));
1231 kern_close(struct thread *td, int fd)
1233 struct filedesc *fdp;
1236 fdp = td->td_proc->p_fd;
1238 AUDIT_SYSCLOSE(td, fd);
1240 FILEDESC_XLOCK(fdp);
1241 if ((fp = fget_locked(fdp, fd)) == NULL) {
1242 FILEDESC_XUNLOCK(fdp);
1247 /* closefp() drops the FILEDESC lock for us. */
1248 return (closefp(fdp, fd, fp, td, 1));
1252 * Close open file descriptors.
1254 #ifndef _SYS_SYSPROTO_H_
1255 struct closefrom_args {
1261 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1263 struct filedesc *fdp;
1266 fdp = td->td_proc->p_fd;
1267 AUDIT_ARG_FD(uap->lowfd);
1270 * Treat negative starting file descriptor values identical to
1271 * closefrom(0) which closes all files.
1275 FILEDESC_SLOCK(fdp);
1276 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1277 if (fdp->fd_ofiles[fd].fde_file != NULL) {
1278 FILEDESC_SUNLOCK(fdp);
1279 (void)kern_close(td, fd);
1280 FILEDESC_SLOCK(fdp);
1283 FILEDESC_SUNLOCK(fdp);
1287 #if defined(COMPAT_43)
1289 * Return status information about a file descriptor.
1291 #ifndef _SYS_SYSPROTO_H_
1292 struct ofstat_args {
1299 ofstat(struct thread *td, struct ofstat_args *uap)
1305 error = kern_fstat(td, uap->fd, &ub);
1308 error = copyout(&oub, uap->sb, sizeof(oub));
1312 #endif /* COMPAT_43 */
1314 #if defined(COMPAT_FREEBSD11)
1316 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1319 struct freebsd11_stat osb;
1322 error = kern_fstat(td, uap->fd, &sb);
1325 error = freebsd11_cvtstat(&sb, &osb);
1327 error = copyout(&osb, uap->sb, sizeof(osb));
1330 #endif /* COMPAT_FREEBSD11 */
1333 * Return status information about a file descriptor.
1335 #ifndef _SYS_SYSPROTO_H_
1343 sys_fstat(struct thread *td, struct fstat_args *uap)
1348 error = kern_fstat(td, uap->fd, &ub);
1350 error = copyout(&ub, uap->sb, sizeof(ub));
1355 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1362 error = fget(td, fd, &cap_fstat_rights, &fp);
1366 AUDIT_ARG_FILE(td->td_proc, fp);
1368 error = fo_stat(fp, sbp, td->td_ucred, td);
1370 #ifdef __STAT_TIME_T_EXT
1372 sbp->st_atim_ext = 0;
1373 sbp->st_mtim_ext = 0;
1374 sbp->st_ctim_ext = 0;
1375 sbp->st_btim_ext = 0;
1379 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1385 #if defined(COMPAT_FREEBSD11)
1387 * Return status information about a file descriptor.
1389 #ifndef _SYS_SYSPROTO_H_
1390 struct freebsd11_nfstat_args {
1397 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1403 error = kern_fstat(td, uap->fd, &ub);
1405 freebsd11_cvtnstat(&ub, &nub);
1406 error = copyout(&nub, uap->sb, sizeof(nub));
1410 #endif /* COMPAT_FREEBSD11 */
1413 * Return pathconf information about a file descriptor.
1415 #ifndef _SYS_SYSPROTO_H_
1416 struct fpathconf_args {
1423 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1428 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1430 td->td_retval[0] = value;
1435 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1441 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1445 if (name == _PC_ASYNC_IO) {
1446 *valuep = _POSIX_ASYNCHRONOUS_IO;
1451 vn_lock(vp, LK_SHARED | LK_RETRY);
1452 error = VOP_PATHCONF(vp, name, valuep);
1454 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1455 if (name != _PC_PIPE_BUF) {
1470 * Initialize filecaps structure.
1473 filecaps_init(struct filecaps *fcaps)
1476 bzero(fcaps, sizeof(*fcaps));
1477 fcaps->fc_nioctls = -1;
1481 * Copy filecaps structure allocating memory for ioctls array if needed.
1483 * The last parameter indicates whether the fdtable is locked. If it is not and
1484 * ioctls are encountered, copying fails and the caller must lock the table.
1486 * Note that if the table was not locked, the caller has to check the relevant
1487 * sequence counter to determine whether the operation was successful.
1490 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1494 if (src->fc_ioctls != NULL && !locked)
1496 memcpy(dst, src, sizeof(*src));
1497 if (src->fc_ioctls == NULL)
1500 KASSERT(src->fc_nioctls > 0,
1501 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1503 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1504 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1505 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1510 filecaps_copy_prep(const struct filecaps *src)
1515 if (src->fc_ioctls == NULL)
1518 KASSERT(src->fc_nioctls > 0,
1519 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1521 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1522 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1527 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1533 if (src->fc_ioctls == NULL) {
1534 MPASS(ioctls == NULL);
1538 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1539 dst->fc_ioctls = ioctls;
1540 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1544 * Move filecaps structure to the new place and clear the old place.
1547 filecaps_move(struct filecaps *src, struct filecaps *dst)
1551 bzero(src, sizeof(*src));
1555 * Fill the given filecaps structure with full rights.
1558 filecaps_fill(struct filecaps *fcaps)
1561 CAP_ALL(&fcaps->fc_rights);
1562 fcaps->fc_ioctls = NULL;
1563 fcaps->fc_nioctls = -1;
1564 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1568 * Free memory allocated within filecaps structure.
1571 filecaps_free(struct filecaps *fcaps)
1574 free(fcaps->fc_ioctls, M_FILECAPS);
1575 bzero(fcaps, sizeof(*fcaps));
1579 filecaps_free_prep(struct filecaps *fcaps)
1583 ioctls = fcaps->fc_ioctls;
1584 bzero(fcaps, sizeof(*fcaps));
1589 filecaps_free_finish(u_long *ioctls)
1592 free(ioctls, M_FILECAPS);
1596 * Validate the given filecaps structure.
1599 filecaps_validate(const struct filecaps *fcaps, const char *func)
1602 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1603 ("%s: invalid rights", func));
1604 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1605 ("%s: invalid fcntls", func));
1606 KASSERT(fcaps->fc_fcntls == 0 ||
1607 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1608 ("%s: fcntls without CAP_FCNTL", func));
1609 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1610 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1611 ("%s: invalid ioctls", func));
1612 KASSERT(fcaps->fc_nioctls == 0 ||
1613 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1614 ("%s: ioctls without CAP_IOCTL", func));
1618 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1622 FILEDESC_XLOCK_ASSERT(fdp);
1624 nfd1 = fdp->fd_nfiles * 2;
1627 fdgrowtable(fdp, nfd1);
1631 * Grow the file table to accommodate (at least) nfd descriptors.
1634 fdgrowtable(struct filedesc *fdp, int nfd)
1636 struct filedesc0 *fdp0;
1637 struct freetable *ft;
1638 struct fdescenttbl *ntable;
1639 struct fdescenttbl *otable;
1640 int nnfiles, onfiles;
1641 NDSLOTTYPE *nmap, *omap;
1644 * If lastfile is -1 this struct filedesc was just allocated and we are
1645 * growing it to accommodate for the one we are going to copy from. There
1646 * is no need to have a lock on this one as it's not visible to anyone.
1648 if (fdp->fd_lastfile != -1)
1649 FILEDESC_XLOCK_ASSERT(fdp);
1651 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1653 /* save old values */
1654 onfiles = fdp->fd_nfiles;
1655 otable = fdp->fd_files;
1658 /* compute the size of the new table */
1659 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1660 if (nnfiles <= onfiles)
1661 /* the table is already large enough */
1665 * Allocate a new table. We need enough space for the number of
1666 * entries, file entries themselves and the struct freetable we will use
1667 * when we decommission the table and place it on the freelist.
1668 * We place the struct freetable in the middle so we don't have
1669 * to worry about padding.
1671 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1672 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1673 sizeof(struct freetable),
1674 M_FILEDESC, M_ZERO | M_WAITOK);
1675 /* copy the old data */
1676 ntable->fdt_nfiles = nnfiles;
1677 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1678 onfiles * sizeof(ntable->fdt_ofiles[0]));
1681 * Allocate a new map only if the old is not large enough. It will
1682 * grow at a slower rate than the table as it can map more
1683 * entries than the table can hold.
1685 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1686 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1688 /* copy over the old data and update the pointer */
1689 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1694 * Make sure that ntable is correctly initialized before we replace
1695 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1698 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1701 * Do not free the old file table, as some threads may still
1702 * reference entries within it. Instead, place it on a freelist
1703 * which will be processed when the struct filedesc is released.
1705 * Note that if onfiles == NDFILE, we're dealing with the original
1706 * static allocation contained within (struct filedesc0 *)fdp,
1707 * which must not be freed.
1709 if (onfiles > NDFILE) {
1710 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1711 fdp0 = (struct filedesc0 *)fdp;
1712 ft->ft_table = otable;
1713 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1716 * The map does not have the same possibility of threads still
1717 * holding references to it. So always free it as long as it
1718 * does not reference the original static allocation.
1720 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1721 free(omap, M_FILEDESC);
1725 * Allocate a file descriptor for the process.
1728 fdalloc(struct thread *td, int minfd, int *result)
1730 struct proc *p = td->td_proc;
1731 struct filedesc *fdp = p->p_fd;
1732 int fd, maxfd, allocfd;
1737 FILEDESC_XLOCK_ASSERT(fdp);
1739 if (fdp->fd_freefile > minfd)
1740 minfd = fdp->fd_freefile;
1742 maxfd = getmaxfd(td);
1745 * Search the bitmap for a free descriptor starting at minfd.
1746 * If none is found, grow the file table.
1748 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1751 if (fd >= fdp->fd_nfiles) {
1752 allocfd = min(fd * 2, maxfd);
1756 error = racct_set(p, RACCT_NOFILE, allocfd);
1763 * fd is already equal to first free descriptor >= minfd, so
1764 * we only need to grow the table and we are done.
1766 fdgrowtable_exp(fdp, allocfd);
1770 * Perform some sanity checks, then mark the file descriptor as
1771 * used and return it to the caller.
1773 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1774 ("invalid descriptor %d", fd));
1775 KASSERT(!fdisused(fdp, fd),
1776 ("fd_first_free() returned non-free descriptor"));
1777 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1778 ("file descriptor isn't free"));
1785 * Allocate n file descriptors for the process.
1788 fdallocn(struct thread *td, int minfd, int *fds, int n)
1790 struct proc *p = td->td_proc;
1791 struct filedesc *fdp = p->p_fd;
1794 FILEDESC_XLOCK_ASSERT(fdp);
1796 for (i = 0; i < n; i++)
1797 if (fdalloc(td, 0, &fds[i]) != 0)
1801 for (i--; i >= 0; i--)
1802 fdunused(fdp, fds[i]);
1810 * Create a new open file structure and allocate a file descriptor for the
1811 * process that refers to it. We add one reference to the file for the
1812 * descriptor table and one reference for resultfp. This is to prevent us
1813 * being preempted and the entry in the descriptor table closed after we
1814 * release the FILEDESC lock.
1817 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1818 struct filecaps *fcaps)
1823 error = falloc_noinstall(td, &fp);
1825 return (error); /* no reference held on error */
1827 error = finstall(td, fp, &fd, flags, fcaps);
1829 fdrop(fp, td); /* one reference (fp only) */
1833 if (resultfp != NULL)
1834 *resultfp = fp; /* copy out result */
1836 fdrop(fp, td); /* release local reference */
1838 if (resultfd != NULL)
1845 * Create a new open file structure without allocating a file descriptor.
1848 falloc_noinstall(struct thread *td, struct file **resultfp)
1851 int maxuserfiles = maxfiles - (maxfiles / 20);
1853 static struct timeval lastfail;
1856 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1858 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1859 if ((openfiles_new >= maxuserfiles &&
1860 priv_check(td, PRIV_MAXFILES) != 0) ||
1861 openfiles_new >= maxfiles) {
1862 atomic_subtract_int(&openfiles, 1);
1863 if (ppsratecheck(&lastfail, &curfail, 1)) {
1864 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1865 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1869 fp = uma_zalloc(file_zone, M_WAITOK);
1870 bzero(fp, sizeof(*fp));
1871 refcount_init(&fp->f_count, 1);
1872 fp->f_cred = crhold(td->td_ucred);
1873 fp->f_ops = &badfileops;
1879 * Install a file in a file descriptor table.
1882 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1883 struct filecaps *fcaps)
1885 struct filedescent *fde;
1889 filecaps_validate(fcaps, __func__);
1890 FILEDESC_XLOCK_ASSERT(fdp);
1892 fde = &fdp->fd_ofiles[fd];
1894 seq_write_begin(&fde->fde_seq);
1897 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1899 filecaps_move(fcaps, &fde->fde_caps);
1901 filecaps_fill(&fde->fde_caps);
1903 seq_write_end(&fde->fde_seq);
1908 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1909 struct filecaps *fcaps)
1911 struct filedesc *fdp = td->td_proc->p_fd;
1916 FILEDESC_XLOCK(fdp);
1917 if ((error = fdalloc(td, 0, fd))) {
1918 FILEDESC_XUNLOCK(fdp);
1922 _finstall(fdp, fp, *fd, flags, fcaps);
1923 FILEDESC_XUNLOCK(fdp);
1928 * Build a new filedesc structure from another.
1929 * Copy the current, root, and jail root vnode references.
1931 * If fdp is not NULL, return with it shared locked.
1934 fdinit(struct filedesc *fdp, bool prepfiles)
1936 struct filedesc0 *newfdp0;
1937 struct filedesc *newfdp;
1939 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1940 newfdp = &newfdp0->fd_fd;
1942 /* Create the file descriptor table. */
1943 FILEDESC_LOCK_INIT(newfdp);
1944 refcount_init(&newfdp->fd_refcnt, 1);
1945 refcount_init(&newfdp->fd_holdcnt, 1);
1946 newfdp->fd_cmask = CMASK;
1947 newfdp->fd_map = newfdp0->fd_dmap;
1948 newfdp->fd_lastfile = -1;
1949 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1950 newfdp->fd_files->fdt_nfiles = NDFILE;
1955 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1956 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1958 FILEDESC_SLOCK(fdp);
1959 newfdp->fd_cdir = fdp->fd_cdir;
1960 if (newfdp->fd_cdir)
1961 vrefact(newfdp->fd_cdir);
1962 newfdp->fd_rdir = fdp->fd_rdir;
1963 if (newfdp->fd_rdir)
1964 vrefact(newfdp->fd_rdir);
1965 newfdp->fd_jdir = fdp->fd_jdir;
1966 if (newfdp->fd_jdir)
1967 vrefact(newfdp->fd_jdir);
1970 FILEDESC_SUNLOCK(fdp);
1972 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1973 FILEDESC_SUNLOCK(fdp);
1974 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1975 FILEDESC_SLOCK(fdp);
1982 static struct filedesc *
1983 fdhold(struct proc *p)
1985 struct filedesc *fdp;
1987 PROC_LOCK_ASSERT(p, MA_OWNED);
1990 refcount_acquire(&fdp->fd_holdcnt);
1995 fddrop(struct filedesc *fdp)
1998 if (fdp->fd_holdcnt > 1) {
1999 if (refcount_release(&fdp->fd_holdcnt) == 0)
2003 FILEDESC_LOCK_DESTROY(fdp);
2004 uma_zfree(filedesc0_zone, fdp);
2008 * Share a filedesc structure.
2011 fdshare(struct filedesc *fdp)
2014 refcount_acquire(&fdp->fd_refcnt);
2019 * Unshare a filedesc structure, if necessary by making a copy
2022 fdunshare(struct thread *td)
2024 struct filedesc *tmp;
2025 struct proc *p = td->td_proc;
2027 if (p->p_fd->fd_refcnt == 1)
2030 tmp = fdcopy(p->p_fd);
2036 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2040 td->td_proc->p_fd = fdp;
2044 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2045 * this is to ease callers, not catch errors.
2048 fdcopy(struct filedesc *fdp)
2050 struct filedesc *newfdp;
2051 struct filedescent *nfde, *ofde;
2056 newfdp = fdinit(fdp, true);
2057 /* copy all passable descriptors (i.e. not kqueue) */
2058 newfdp->fd_freefile = -1;
2059 for (i = 0; i <= fdp->fd_lastfile; ++i) {
2060 ofde = &fdp->fd_ofiles[i];
2061 if (ofde->fde_file == NULL ||
2062 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2063 if (newfdp->fd_freefile == -1)
2064 newfdp->fd_freefile = i;
2067 nfde = &newfdp->fd_ofiles[i];
2069 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2070 fhold(nfde->fde_file);
2071 fdused_init(newfdp, i);
2072 newfdp->fd_lastfile = i;
2074 if (newfdp->fd_freefile == -1)
2075 newfdp->fd_freefile = i;
2076 newfdp->fd_cmask = fdp->fd_cmask;
2077 FILEDESC_SUNLOCK(fdp);
2082 * Copies a filedesc structure, while remapping all file descriptors
2083 * stored inside using a translation table.
2085 * File descriptors are copied over to the new file descriptor table,
2086 * regardless of whether the close-on-exec flag is set.
2089 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2090 struct filedesc **ret)
2092 struct filedesc *newfdp;
2093 struct filedescent *nfde, *ofde;
2098 newfdp = fdinit(fdp, true);
2099 if (nfds > fdp->fd_lastfile + 1) {
2100 /* New table cannot be larger than the old one. */
2104 /* Copy all passable descriptors (i.e. not kqueue). */
2105 newfdp->fd_freefile = nfds;
2106 for (i = 0; i < nfds; ++i) {
2107 if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2108 /* File descriptor out of bounds. */
2112 ofde = &fdp->fd_ofiles[fds[i]];
2113 if (ofde->fde_file == NULL) {
2114 /* Unused file descriptor. */
2118 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2119 /* File descriptor cannot be passed. */
2123 nfde = &newfdp->fd_ofiles[i];
2125 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2126 fhold(nfde->fde_file);
2127 fdused_init(newfdp, i);
2128 newfdp->fd_lastfile = i;
2130 newfdp->fd_cmask = fdp->fd_cmask;
2131 FILEDESC_SUNLOCK(fdp);
2135 FILEDESC_SUNLOCK(fdp);
2136 fdescfree_remapped(newfdp);
2141 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2142 * one of processes using it exits) and the table used to be shared.
2145 fdclearlocks(struct thread *td)
2147 struct filedesc *fdp;
2148 struct filedesc_to_leader *fdtol;
2158 MPASS(fdtol != NULL);
2160 FILEDESC_XLOCK(fdp);
2161 KASSERT(fdtol->fdl_refcount > 0,
2162 ("filedesc_to_refcount botch: fdl_refcount=%d",
2163 fdtol->fdl_refcount));
2164 if (fdtol->fdl_refcount == 1 &&
2165 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2166 for (i = 0; i <= fdp->fd_lastfile; i++) {
2167 fp = fdp->fd_ofiles[i].fde_file;
2168 if (fp == NULL || fp->f_type != DTYPE_VNODE)
2171 FILEDESC_XUNLOCK(fdp);
2172 lf.l_whence = SEEK_SET;
2175 lf.l_type = F_UNLCK;
2177 (void) VOP_ADVLOCK(vp,
2178 (caddr_t)p->p_leader, F_UNLCK,
2180 FILEDESC_XLOCK(fdp);
2185 if (fdtol->fdl_refcount == 1) {
2186 if (fdp->fd_holdleaderscount > 0 &&
2187 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2189 * close() or kern_dup() has cleared a reference
2190 * in a shared file descriptor table.
2192 fdp->fd_holdleaderswakeup = 1;
2193 sx_sleep(&fdp->fd_holdleaderscount,
2194 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2197 if (fdtol->fdl_holdcount > 0) {
2199 * Ensure that fdtol->fdl_leader remains
2200 * valid in closef().
2202 fdtol->fdl_wakeup = 1;
2203 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2208 fdtol->fdl_refcount--;
2209 if (fdtol->fdl_refcount == 0 &&
2210 fdtol->fdl_holdcount == 0) {
2211 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2212 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2216 FILEDESC_XUNLOCK(fdp);
2218 free(fdtol, M_FILEDESC_TO_LEADER);
2222 * Release a filedesc structure.
2225 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2227 struct filedesc0 *fdp0;
2228 struct freetable *ft, *tft;
2229 struct filedescent *fde;
2233 for (i = 0; i <= fdp->fd_lastfile; i++) {
2234 fde = &fdp->fd_ofiles[i];
2239 (void) closef(fp, td);
2245 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2246 free(fdp->fd_map, M_FILEDESC);
2247 if (fdp->fd_nfiles > NDFILE)
2248 free(fdp->fd_files, M_FILEDESC);
2250 fdp0 = (struct filedesc0 *)fdp;
2251 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2252 free(ft->ft_table, M_FILEDESC);
2258 fdescfree(struct thread *td)
2261 struct filedesc *fdp;
2262 struct vnode *cdir, *jdir, *rdir;
2271 racct_set(p, RACCT_NOFILE, 0);
2276 if (p->p_fdtol != NULL)
2283 if (refcount_release(&fdp->fd_refcnt) == 0)
2286 FILEDESC_XLOCK(fdp);
2287 cdir = fdp->fd_cdir;
2288 fdp->fd_cdir = NULL;
2289 rdir = fdp->fd_rdir;
2290 fdp->fd_rdir = NULL;
2291 jdir = fdp->fd_jdir;
2292 fdp->fd_jdir = NULL;
2293 FILEDESC_XUNLOCK(fdp);
2302 fdescfree_fds(td, fdp, 1);
2306 fdescfree_remapped(struct filedesc *fdp)
2309 if (fdp->fd_cdir != NULL)
2310 vrele(fdp->fd_cdir);
2311 if (fdp->fd_rdir != NULL)
2312 vrele(fdp->fd_rdir);
2313 if (fdp->fd_jdir != NULL)
2314 vrele(fdp->fd_jdir);
2316 fdescfree_fds(curthread, fdp, 0);
2320 * For setugid programs, we don't want to people to use that setugidness
2321 * to generate error messages which write to a file which otherwise would
2322 * otherwise be off-limits to the process. We check for filesystems where
2323 * the vnode can change out from under us after execve (like [lin]procfs).
2325 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2326 * sufficient. We also don't check for setugidness since we know we are.
2329 is_unsafe(struct file *fp)
2333 if (fp->f_type != DTYPE_VNODE)
2337 return ((vp->v_vflag & VV_PROCDEP) != 0);
2341 * Make this setguid thing safe, if at all possible.
2344 fdsetugidsafety(struct thread *td)
2346 struct filedesc *fdp;
2350 fdp = td->td_proc->p_fd;
2351 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2352 MPASS(fdp->fd_nfiles >= 3);
2353 for (i = 0; i <= 2; i++) {
2354 fp = fdp->fd_ofiles[i].fde_file;
2355 if (fp != NULL && is_unsafe(fp)) {
2356 FILEDESC_XLOCK(fdp);
2357 knote_fdclose(td, i);
2359 * NULL-out descriptor prior to close to avoid
2360 * a race while close blocks.
2363 FILEDESC_XUNLOCK(fdp);
2364 (void) closef(fp, td);
2370 * If a specific file object occupies a specific file descriptor, close the
2371 * file descriptor entry and drop a reference on the file object. This is a
2372 * convenience function to handle a subsequent error in a function that calls
2373 * falloc() that handles the race that another thread might have closed the
2374 * file descriptor out from under the thread creating the file object.
2377 fdclose(struct thread *td, struct file *fp, int idx)
2379 struct filedesc *fdp = td->td_proc->p_fd;
2381 FILEDESC_XLOCK(fdp);
2382 if (fdp->fd_ofiles[idx].fde_file == fp) {
2384 FILEDESC_XUNLOCK(fdp);
2387 FILEDESC_XUNLOCK(fdp);
2391 * Close any files on exec?
2394 fdcloseexec(struct thread *td)
2396 struct filedesc *fdp;
2397 struct filedescent *fde;
2401 fdp = td->td_proc->p_fd;
2402 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2403 for (i = 0; i <= fdp->fd_lastfile; i++) {
2404 fde = &fdp->fd_ofiles[i];
2406 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2407 (fde->fde_flags & UF_EXCLOSE))) {
2408 FILEDESC_XLOCK(fdp);
2410 (void) closefp(fdp, i, fp, td, 0);
2411 FILEDESC_UNLOCK_ASSERT(fdp);
2417 * It is unsafe for set[ug]id processes to be started with file
2418 * descriptors 0..2 closed, as these descriptors are given implicit
2419 * significance in the Standard C library. fdcheckstd() will create a
2420 * descriptor referencing /dev/null for each of stdin, stdout, and
2421 * stderr that is not already open.
2424 fdcheckstd(struct thread *td)
2426 struct filedesc *fdp;
2428 int i, error, devnull;
2430 fdp = td->td_proc->p_fd;
2431 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2432 MPASS(fdp->fd_nfiles >= 3);
2434 for (i = 0; i <= 2; i++) {
2435 if (fdp->fd_ofiles[i].fde_file != NULL)
2438 save = td->td_retval[0];
2439 if (devnull != -1) {
2440 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2442 error = kern_openat(td, AT_FDCWD, "/dev/null",
2443 UIO_SYSSPACE, O_RDWR, 0);
2445 devnull = td->td_retval[0];
2446 KASSERT(devnull == i, ("we didn't get our fd"));
2449 td->td_retval[0] = save;
2457 * Internal form of close. Decrement reference count on file structure.
2458 * Note: td may be NULL when closing a file that was being passed in a
2462 closef(struct file *fp, struct thread *td)
2466 struct filedesc_to_leader *fdtol;
2467 struct filedesc *fdp;
2470 * POSIX record locking dictates that any close releases ALL
2471 * locks owned by this process. This is handled by setting
2472 * a flag in the unlock to free ONLY locks obeying POSIX
2473 * semantics, and not to free BSD-style file locks.
2474 * If the descriptor was in a message, POSIX-style locks
2475 * aren't passed with the descriptor, and the thread pointer
2476 * will be NULL. Callers should be careful only to pass a
2477 * NULL thread pointer when there really is no owning
2478 * context that might have locks, or the locks will be
2481 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2483 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2484 lf.l_whence = SEEK_SET;
2487 lf.l_type = F_UNLCK;
2488 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2489 F_UNLCK, &lf, F_POSIX);
2491 fdtol = td->td_proc->p_fdtol;
2492 if (fdtol != NULL) {
2494 * Handle special case where file descriptor table is
2495 * shared between multiple process leaders.
2497 fdp = td->td_proc->p_fd;
2498 FILEDESC_XLOCK(fdp);
2499 for (fdtol = fdtol->fdl_next;
2500 fdtol != td->td_proc->p_fdtol;
2501 fdtol = fdtol->fdl_next) {
2502 if ((fdtol->fdl_leader->p_flag &
2505 fdtol->fdl_holdcount++;
2506 FILEDESC_XUNLOCK(fdp);
2507 lf.l_whence = SEEK_SET;
2510 lf.l_type = F_UNLCK;
2512 (void) VOP_ADVLOCK(vp,
2513 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2515 FILEDESC_XLOCK(fdp);
2516 fdtol->fdl_holdcount--;
2517 if (fdtol->fdl_holdcount == 0 &&
2518 fdtol->fdl_wakeup != 0) {
2519 fdtol->fdl_wakeup = 0;
2523 FILEDESC_XUNLOCK(fdp);
2526 return (fdrop(fp, td));
2530 * Initialize the file pointer with the specified properties.
2532 * The ops are set with release semantics to be certain that the flags, type,
2533 * and data are visible when ops is. This is to prevent ops methods from being
2534 * called with bad data.
2537 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2542 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2546 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2547 struct file **fpp, struct filecaps *havecapsp)
2549 struct filedescent *fde;
2552 FILEDESC_LOCK_ASSERT(fdp);
2554 fde = fdeget_locked(fdp, fd);
2561 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2566 if (havecapsp != NULL)
2567 filecaps_copy(&fde->fde_caps, havecapsp, true);
2569 *fpp = fde->fde_file;
2577 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2578 struct file **fpp, struct filecaps *havecapsp)
2580 struct filedesc *fdp = td->td_proc->p_fd;
2582 #ifndef CAPABILITIES
2583 error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2584 if (error == 0 && havecapsp != NULL)
2585 filecaps_fill(havecapsp);
2591 error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2595 if (havecapsp != NULL) {
2596 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2597 havecapsp, false)) {
2603 if (!fd_modified(fdp, fd, seq))
2612 FILEDESC_SLOCK(fdp);
2613 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2616 FILEDESC_SUNLOCK(fdp);
2622 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2623 struct file **fpp, seq_t *seqp)
2626 const struct filedescent *fde;
2628 const struct fdescenttbl *fdt;
2633 cap_rights_t haverights;
2637 fdt = fdp->fd_files;
2638 if ((u_int)fd >= fdt->fdt_nfiles)
2641 * Fetch the descriptor locklessly. We avoid fdrop() races by
2642 * never raising a refcount above 0. To accomplish this we have
2643 * to use a cmpset loop rather than an atomic_add. The descriptor
2644 * must be re-verified once we acquire a reference to be certain
2645 * that the identity is still correct and we did not lose a race
2646 * due to preemption.
2650 seq = seq_load(fd_seq(fdt, fd));
2651 fde = &fdt->fdt_ofiles[fd];
2652 haverights = *cap_rights_fde_inline(fde);
2654 if (!seq_consistent(fd_seq(fdt, fd), seq))
2657 fp = fdt->fdt_ofiles[fd].fde_file;
2662 error = cap_check(&haverights, needrightsp);
2666 count = fp->f_count;
2670 * Force a reload. Other thread could reallocate the
2671 * table before this fd was closed, so it possible that
2672 * there is a stale fp pointer in cached version.
2674 fdt = *(const struct fdescenttbl * const volatile *)&(fdp->fd_files);
2678 * Use an acquire barrier to force re-reading of fdt so it is
2679 * refreshed for verification.
2681 if (atomic_fcmpset_acq_int(&fp->f_count, &count, count + 1) == 0)
2683 fdt = fdp->fd_files;
2685 if (seq_consistent_nomb(fd_seq(fdt, fd), seq))
2687 if (fp == fdt->fdt_ofiles[fd].fde_file)
2690 fdrop(fp, curthread);
2702 * Extract the file pointer associated with the specified descriptor for the
2703 * current user process.
2705 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2708 * File's rights will be checked against the capability rights mask.
2710 * If an error occurred the non-zero error is returned and *fpp is set to
2711 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
2712 * responsible for fdrop().
2715 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2716 cap_rights_t *needrightsp, seq_t *seqp)
2718 struct filedesc *fdp;
2723 fdp = td->td_proc->p_fd;
2724 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2727 if (fp->f_ops == &badfileops) {
2733 * FREAD and FWRITE failure return EBADF as per POSIX.
2739 if ((fp->f_flag & flags) == 0)
2743 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2744 ((fp->f_flag & FWRITE) != 0))
2750 KASSERT(0, ("wrong flags"));
2763 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2766 return (_fget(td, fd, fpp, 0, rightsp, NULL));
2770 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2774 #ifndef CAPABILITIES
2775 error = _fget(td, fd, fpp, 0, rightsp, NULL);
2776 if (maxprotp != NULL)
2777 *maxprotp = VM_PROT_ALL;
2779 struct filedesc *fdp = td->td_proc->p_fd;
2782 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2784 error = _fget(td, fd, fpp, 0, rightsp, &seq);
2788 * If requested, convert capability rights to access flags.
2790 if (maxprotp != NULL)
2791 *maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd));
2792 if (!fd_modified(fdp, fd, seq))
2801 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2804 return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2808 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2811 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2815 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2818 struct filedesc *fdp = td->td_proc->p_fd;
2819 #ifndef CAPABILITIES
2820 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2825 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2827 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2830 error = cap_fcntl_check(fdp, fd, needfcntl);
2831 if (!fd_modified(fdp, fd, seq))
2844 * Like fget() but loads the underlying vnode, or returns an error if the
2845 * descriptor does not represent a vnode. Note that pipes use vnodes but
2846 * never have VM objects. The returned vnode will be vref()'d.
2848 * XXX: what about the unused flags ?
2851 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2858 error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2861 if (fp->f_vnode == NULL) {
2873 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2876 return (_fgetvp(td, fd, 0, rightsp, vpp));
2880 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2881 struct filecaps *havecaps, struct vnode **vpp)
2883 struct filedesc *fdp;
2884 struct filecaps caps;
2888 fdp = td->td_proc->p_fd;
2889 error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2892 if (fp->f_ops == &badfileops) {
2896 if (fp->f_vnode == NULL) {
2907 filecaps_free(&caps);
2912 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2915 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2919 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2922 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2927 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2931 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2936 * Handle the last reference to a file being closed.
2938 * Without the noinline attribute clang keeps inlining the func thorough this
2939 * file when fdrop is used.
2942 _fdrop(struct file *fp, struct thread *td)
2946 if (fp->f_count != 0)
2947 panic("fdrop: count %d", fp->f_count);
2948 error = fo_close(fp, td);
2949 atomic_subtract_int(&openfiles, 1);
2951 free(fp->f_advice, M_FADVISE);
2952 uma_zfree(file_zone, fp);
2958 * Apply an advisory lock on a file descriptor.
2960 * Just attempt to get a record lock of the requested type on the entire file
2961 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2963 #ifndef _SYS_SYSPROTO_H_
2971 sys_flock(struct thread *td, struct flock_args *uap)
2978 error = fget(td, uap->fd, &cap_flock_rights, &fp);
2981 if (fp->f_type != DTYPE_VNODE) {
2983 return (EOPNOTSUPP);
2987 lf.l_whence = SEEK_SET;
2990 if (uap->how & LOCK_UN) {
2991 lf.l_type = F_UNLCK;
2992 atomic_clear_int(&fp->f_flag, FHASLOCK);
2993 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2996 if (uap->how & LOCK_EX)
2997 lf.l_type = F_WRLCK;
2998 else if (uap->how & LOCK_SH)
2999 lf.l_type = F_RDLCK;
3004 atomic_set_int(&fp->f_flag, FHASLOCK);
3005 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3006 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3012 * Duplicate the specified descriptor to a free descriptor.
3015 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3016 int openerror, int *indxp)
3018 struct filedescent *newfde, *oldfde;
3023 KASSERT(openerror == ENODEV || openerror == ENXIO,
3024 ("unexpected error %d in %s", openerror, __func__));
3027 * If the to-be-dup'd fd number is greater than the allowed number
3028 * of file descriptors, or the fd to be dup'd has already been
3029 * closed, then reject.
3031 FILEDESC_XLOCK(fdp);
3032 if ((fp = fget_locked(fdp, dfd)) == NULL) {
3033 FILEDESC_XUNLOCK(fdp);
3037 error = fdalloc(td, 0, &indx);
3039 FILEDESC_XUNLOCK(fdp);
3044 * There are two cases of interest here.
3046 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3048 * For ENXIO steal away the file structure from (dfd) and store it in
3049 * (indx). (dfd) is effectively closed by this operation.
3051 switch (openerror) {
3054 * Check that the mode the file is being opened for is a
3055 * subset of the mode of the existing descriptor.
3057 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3058 fdunused(fdp, indx);
3059 FILEDESC_XUNLOCK(fdp);
3063 newfde = &fdp->fd_ofiles[indx];
3064 oldfde = &fdp->fd_ofiles[dfd];
3065 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3067 seq_write_begin(&newfde->fde_seq);
3069 memcpy(newfde, oldfde, fde_change_size);
3070 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3073 seq_write_end(&newfde->fde_seq);
3078 * Steal away the file pointer from dfd and stuff it into indx.
3080 newfde = &fdp->fd_ofiles[indx];
3081 oldfde = &fdp->fd_ofiles[dfd];
3083 seq_write_begin(&newfde->fde_seq);
3085 memcpy(newfde, oldfde, fde_change_size);
3086 oldfde->fde_file = NULL;
3089 seq_write_end(&newfde->fde_seq);
3093 FILEDESC_XUNLOCK(fdp);
3099 * This sysctl determines if we will allow a process to chroot(2) if it
3100 * has a directory open:
3101 * 0: disallowed for all processes.
3102 * 1: allowed for processes that were not already chroot(2)'ed.
3103 * 2: allowed for all processes.
3106 static int chroot_allow_open_directories = 1;
3108 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3109 &chroot_allow_open_directories, 0,
3110 "Allow a process to chroot(2) if it has a directory open");
3113 * Helper function for raised chroot(2) security function: Refuse if
3114 * any filedescriptors are open directories.
3117 chroot_refuse_vdir_fds(struct filedesc *fdp)
3123 FILEDESC_LOCK_ASSERT(fdp);
3125 for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3126 fp = fget_locked(fdp, fd);
3129 if (fp->f_type == DTYPE_VNODE) {
3131 if (vp->v_type == VDIR)
3139 * Common routine for kern_chroot() and jail_attach(). The caller is
3140 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3141 * authorize this operation.
3144 pwd_chroot(struct thread *td, struct vnode *vp)
3146 struct filedesc *fdp;
3147 struct vnode *oldvp;
3150 fdp = td->td_proc->p_fd;
3151 FILEDESC_XLOCK(fdp);
3152 if (chroot_allow_open_directories == 0 ||
3153 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3154 error = chroot_refuse_vdir_fds(fdp);
3156 FILEDESC_XUNLOCK(fdp);
3160 oldvp = fdp->fd_rdir;
3163 if (fdp->fd_jdir == NULL) {
3167 FILEDESC_XUNLOCK(fdp);
3173 pwd_chdir(struct thread *td, struct vnode *vp)
3175 struct filedesc *fdp;
3176 struct vnode *oldvp;
3178 fdp = td->td_proc->p_fd;
3179 FILEDESC_XLOCK(fdp);
3180 VNASSERT(vp->v_usecount > 0, vp,
3181 ("chdir to a vnode with zero usecount"));
3182 oldvp = fdp->fd_cdir;
3184 FILEDESC_XUNLOCK(fdp);
3189 * Scan all active processes and prisons to see if any of them have a current
3190 * or root directory of `olddp'. If so, replace them with the new mount point.
3193 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3195 struct filedesc *fdp;
3200 if (vrefcnt(olddp) == 1)
3203 sx_slock(&allproc_lock);
3204 FOREACH_PROC_IN_SYSTEM(p) {
3210 FILEDESC_XLOCK(fdp);
3211 if (fdp->fd_cdir == olddp) {
3213 fdp->fd_cdir = newdp;
3216 if (fdp->fd_rdir == olddp) {
3218 fdp->fd_rdir = newdp;
3221 if (fdp->fd_jdir == olddp) {
3223 fdp->fd_jdir = newdp;
3226 FILEDESC_XUNLOCK(fdp);
3229 sx_sunlock(&allproc_lock);
3230 if (rootvnode == olddp) {
3235 mtx_lock(&prison0.pr_mtx);
3236 if (prison0.pr_root == olddp) {
3238 prison0.pr_root = newdp;
3241 mtx_unlock(&prison0.pr_mtx);
3242 sx_slock(&allprison_lock);
3243 TAILQ_FOREACH(pr, &allprison, pr_list) {
3244 mtx_lock(&pr->pr_mtx);
3245 if (pr->pr_root == olddp) {
3247 pr->pr_root = newdp;
3250 mtx_unlock(&pr->pr_mtx);
3252 sx_sunlock(&allprison_lock);
3257 struct filedesc_to_leader *
3258 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3260 struct filedesc_to_leader *fdtol;
3262 fdtol = malloc(sizeof(struct filedesc_to_leader),
3263 M_FILEDESC_TO_LEADER, M_WAITOK);
3264 fdtol->fdl_refcount = 1;
3265 fdtol->fdl_holdcount = 0;
3266 fdtol->fdl_wakeup = 0;
3267 fdtol->fdl_leader = leader;
3269 FILEDESC_XLOCK(fdp);
3270 fdtol->fdl_next = old->fdl_next;
3271 fdtol->fdl_prev = old;
3272 old->fdl_next = fdtol;
3273 fdtol->fdl_next->fdl_prev = fdtol;
3274 FILEDESC_XUNLOCK(fdp);
3276 fdtol->fdl_next = fdtol;
3277 fdtol->fdl_prev = fdtol;
3283 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3285 struct filedesc *fdp;
3286 int i, count, slots;
3288 if (*(int *)arg1 != 0)
3291 fdp = curproc->p_fd;
3293 FILEDESC_SLOCK(fdp);
3294 slots = NDSLOTS(fdp->fd_lastfile + 1);
3295 for (i = 0; i < slots; i++)
3296 count += bitcountl(fdp->fd_map[i]);
3297 FILEDESC_SUNLOCK(fdp);
3299 return (SYSCTL_OUT(req, &count, sizeof(count)));
3302 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3303 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3304 "Number of open file descriptors");
3307 * Get file structures globally.
3310 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3313 struct filedesc *fdp;
3318 error = sysctl_wire_old_buffer(req, 0);
3321 if (req->oldptr == NULL) {
3323 sx_slock(&allproc_lock);
3324 FOREACH_PROC_IN_SYSTEM(p) {
3326 if (p->p_state == PRS_NEW) {
3334 /* overestimates sparse tables. */
3335 if (fdp->fd_lastfile > 0)
3336 n += fdp->fd_lastfile;
3339 sx_sunlock(&allproc_lock);
3340 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3343 bzero(&xf, sizeof(xf));
3344 xf.xf_size = sizeof(xf);
3345 sx_slock(&allproc_lock);
3346 FOREACH_PROC_IN_SYSTEM(p) {
3348 if (p->p_state == PRS_NEW) {
3352 if (p_cansee(req->td, p) != 0) {
3356 xf.xf_pid = p->p_pid;
3357 xf.xf_uid = p->p_ucred->cr_uid;
3362 FILEDESC_SLOCK(fdp);
3363 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3364 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3367 xf.xf_file = (uintptr_t)fp;
3368 xf.xf_data = (uintptr_t)fp->f_data;
3369 xf.xf_vnode = (uintptr_t)fp->f_vnode;
3370 xf.xf_type = (uintptr_t)fp->f_type;
3371 xf.xf_count = fp->f_count;
3373 xf.xf_offset = foffset_get(fp);
3374 xf.xf_flag = fp->f_flag;
3375 error = SYSCTL_OUT(req, &xf, sizeof(xf));
3379 FILEDESC_SUNLOCK(fdp);
3384 sx_sunlock(&allproc_lock);
3388 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3389 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3391 #ifdef KINFO_FILE_SIZE
3392 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3396 xlate_fflags(int fflags)
3398 static const struct {
3401 } fflags_table[] = {
3402 { FAPPEND, KF_FLAG_APPEND },
3403 { FASYNC, KF_FLAG_ASYNC },
3404 { FFSYNC, KF_FLAG_FSYNC },
3405 { FHASLOCK, KF_FLAG_HASLOCK },
3406 { FNONBLOCK, KF_FLAG_NONBLOCK },
3407 { FREAD, KF_FLAG_READ },
3408 { FWRITE, KF_FLAG_WRITE },
3409 { O_CREAT, KF_FLAG_CREAT },
3410 { O_DIRECT, KF_FLAG_DIRECT },
3411 { O_EXCL, KF_FLAG_EXCL },
3412 { O_EXEC, KF_FLAG_EXEC },
3413 { O_EXLOCK, KF_FLAG_EXLOCK },
3414 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3415 { O_SHLOCK, KF_FLAG_SHLOCK },
3416 { O_TRUNC, KF_FLAG_TRUNC }
3422 for (i = 0; i < nitems(fflags_table); i++)
3423 if (fflags & fflags_table[i].fflag)
3424 kflags |= fflags_table[i].kf_fflag;
3428 /* Trim unused data from kf_path by truncating the structure size. */
3430 pack_kinfo(struct kinfo_file *kif)
3433 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3434 strlen(kif->kf_path) + 1;
3435 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3439 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3440 struct kinfo_file *kif, struct filedesc *fdp, int flags)
3444 bzero(kif, sizeof(*kif));
3446 /* Set a default type to allow for empty fill_kinfo() methods. */
3447 kif->kf_type = KF_TYPE_UNKNOWN;
3448 kif->kf_flags = xlate_fflags(fp->f_flag);
3449 if (rightsp != NULL)
3450 kif->kf_cap_rights = *rightsp;
3452 cap_rights_init(&kif->kf_cap_rights);
3454 kif->kf_ref_count = fp->f_count;
3455 kif->kf_offset = foffset_get(fp);
3458 * This may drop the filedesc lock, so the 'fp' cannot be
3459 * accessed after this call.
3461 error = fo_fill_kinfo(fp, kif, fdp);
3463 kif->kf_status |= KF_ATTR_VALID;
3464 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3467 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3471 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3472 struct kinfo_file *kif, int flags)
3476 bzero(kif, sizeof(*kif));
3478 kif->kf_type = KF_TYPE_VNODE;
3479 error = vn_fill_kinfo_vnode(vp, kif);
3481 kif->kf_status |= KF_ATTR_VALID;
3482 kif->kf_flags = xlate_fflags(fflags);
3483 cap_rights_init(&kif->kf_cap_rights);
3485 kif->kf_ref_count = -1;
3486 kif->kf_offset = -1;
3487 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3490 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3494 struct export_fd_buf {
3495 struct filedesc *fdp;
3498 struct kinfo_file kif;
3503 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3505 struct kinfo_file *kif;
3508 if (efbuf->remainder != -1) {
3509 if (efbuf->remainder < kif->kf_structsize) {
3510 /* Terminate export. */
3511 efbuf->remainder = 0;
3514 efbuf->remainder -= kif->kf_structsize;
3516 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3520 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3521 struct export_fd_buf *efbuf)
3525 if (efbuf->remainder == 0)
3527 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3529 FILEDESC_SUNLOCK(efbuf->fdp);
3530 error = export_kinfo_to_sb(efbuf);
3531 FILEDESC_SLOCK(efbuf->fdp);
3536 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3537 struct export_fd_buf *efbuf)
3541 if (efbuf->remainder == 0)
3543 if (efbuf->fdp != NULL)
3544 FILEDESC_SUNLOCK(efbuf->fdp);
3545 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3546 error = export_kinfo_to_sb(efbuf);
3547 if (efbuf->fdp != NULL)
3548 FILEDESC_SLOCK(efbuf->fdp);
3553 * Store a process file descriptor information to sbuf.
3555 * Takes a locked proc as argument, and returns with the proc unlocked.
3558 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
3562 struct filedesc *fdp;
3563 struct export_fd_buf *efbuf;
3564 struct vnode *cttyvp, *textvp, *tracevp;
3566 cap_rights_t rights;
3568 PROC_LOCK_ASSERT(p, MA_OWNED);
3571 tracevp = p->p_tracevp;
3572 if (tracevp != NULL)
3575 textvp = p->p_textvp;
3578 /* Controlling tty. */
3580 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3581 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3587 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3590 efbuf->remainder = maxlen;
3591 efbuf->flags = flags;
3592 if (tracevp != NULL)
3593 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3596 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3598 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3604 FILEDESC_SLOCK(fdp);
3605 /* working directory */
3606 if (fdp->fd_cdir != NULL) {
3607 vrefact(fdp->fd_cdir);
3608 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3610 /* root directory */
3611 if (fdp->fd_rdir != NULL) {
3612 vrefact(fdp->fd_rdir);
3613 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3615 /* jail directory */
3616 if (fdp->fd_jdir != NULL) {
3617 vrefact(fdp->fd_jdir);
3618 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3620 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3621 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3624 rights = *cap_rights(fdp, i);
3625 #else /* !CAPABILITIES */
3626 rights = cap_no_rights;
3629 * Create sysctl entry. It is OK to drop the filedesc
3630 * lock inside of export_file_to_sb() as we will
3631 * re-validate and re-evaluate its properties when the
3634 error = export_file_to_sb(fp, i, &rights, efbuf);
3635 if (error != 0 || efbuf->remainder == 0)
3638 FILEDESC_SUNLOCK(fdp);
3641 free(efbuf, M_TEMP);
3645 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
3648 * Get per-process file descriptors for use by procstat(1), et al.
3651 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3656 int error, error2, *name;
3660 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3661 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3662 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3667 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3668 error = kern_proc_filedesc_out(p, &sb, maxlen,
3669 KERN_FILEDESC_PACK_KINFO);
3670 error2 = sbuf_finish(&sb);
3672 return (error != 0 ? error : error2);
3675 #ifdef COMPAT_FREEBSD7
3676 #ifdef KINFO_OFILE_SIZE
3677 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3681 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3684 okif->kf_structsize = sizeof(*okif);
3685 okif->kf_type = kif->kf_type;
3686 okif->kf_fd = kif->kf_fd;
3687 okif->kf_ref_count = kif->kf_ref_count;
3688 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3689 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3690 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3691 okif->kf_offset = kif->kf_offset;
3692 if (kif->kf_type == KF_TYPE_VNODE)
3693 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
3695 okif->kf_vnode_type = KF_VTYPE_VNON;
3696 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3697 if (kif->kf_type == KF_TYPE_SOCKET) {
3698 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
3699 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
3700 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
3701 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
3702 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
3704 okif->kf_sa_local.ss_family = AF_UNSPEC;
3705 okif->kf_sa_peer.ss_family = AF_UNSPEC;
3710 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3711 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3716 FILEDESC_SUNLOCK(fdp);
3717 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3718 kinfo_to_okinfo(kif, okif);
3719 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3720 FILEDESC_SLOCK(fdp);
3725 * Get per-process file descriptors for use by procstat(1), et al.
3728 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3730 struct kinfo_ofile *okif;
3731 struct kinfo_file *kif;
3732 struct filedesc *fdp;
3733 int error, i, *name;
3738 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3745 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3746 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3747 FILEDESC_SLOCK(fdp);
3748 if (fdp->fd_cdir != NULL)
3749 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3751 if (fdp->fd_rdir != NULL)
3752 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3754 if (fdp->fd_jdir != NULL)
3755 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3757 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3758 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3760 export_file_to_kinfo(fp, i, NULL, kif, fdp,
3761 KERN_FILEDESC_PACK_KINFO);
3762 FILEDESC_SUNLOCK(fdp);
3763 kinfo_to_okinfo(kif, okif);
3764 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3765 FILEDESC_SLOCK(fdp);
3769 FILEDESC_SUNLOCK(fdp);
3776 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3777 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3778 "Process ofiledesc entries");
3779 #endif /* COMPAT_FREEBSD7 */
3782 vntype_to_kinfo(int vtype)
3787 } vtypes_table[] = {
3788 { VBAD, KF_VTYPE_VBAD },
3789 { VBLK, KF_VTYPE_VBLK },
3790 { VCHR, KF_VTYPE_VCHR },
3791 { VDIR, KF_VTYPE_VDIR },
3792 { VFIFO, KF_VTYPE_VFIFO },
3793 { VLNK, KF_VTYPE_VLNK },
3794 { VNON, KF_VTYPE_VNON },
3795 { VREG, KF_VTYPE_VREG },
3796 { VSOCK, KF_VTYPE_VSOCK }
3801 * Perform vtype translation.
3803 for (i = 0; i < nitems(vtypes_table); i++)
3804 if (vtypes_table[i].vtype == vtype)
3805 return (vtypes_table[i].kf_vtype);
3807 return (KF_VTYPE_UNKNOWN);
3810 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3811 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3812 "Process filedesc entries");
3815 * Store a process current working directory information to sbuf.
3817 * Takes a locked proc as argument, and returns with the proc unlocked.
3820 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
3822 struct filedesc *fdp;
3823 struct export_fd_buf *efbuf;
3826 PROC_LOCK_ASSERT(p, MA_OWNED);
3833 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3836 efbuf->remainder = maxlen;
3838 FILEDESC_SLOCK(fdp);
3839 if (fdp->fd_cdir == NULL)
3842 vrefact(fdp->fd_cdir);
3843 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3846 FILEDESC_SUNLOCK(fdp);
3848 free(efbuf, M_TEMP);
3853 * Get per-process current working directory.
3856 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3861 int error, error2, *name;
3865 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3866 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3867 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3872 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3873 error = kern_proc_cwd_out(p, &sb, maxlen);
3874 error2 = sbuf_finish(&sb);
3876 return (error != 0 ? error : error2);
3879 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3880 sysctl_kern_proc_cwd, "Process current working directory");
3884 * For the purposes of debugging, generate a human-readable string for the
3888 file_type_to_name(short type)
3916 case DTYPE_PROCDESC:
3918 case DTYPE_LINUXEFD:
3920 case DTYPE_LINUXTFD:
3928 * For the purposes of debugging, identify a process (if any, perhaps one of
3929 * many) that references the passed file in its file descriptor array. Return
3932 static struct proc *
3933 file_to_first_proc(struct file *fp)
3935 struct filedesc *fdp;
3939 FOREACH_PROC_IN_SYSTEM(p) {
3940 if (p->p_state == PRS_NEW)
3945 for (n = 0; n <= fdp->fd_lastfile; n++) {
3946 if (fp == fdp->fd_ofiles[n].fde_file)
3954 db_print_file(struct file *fp, int header)
3956 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
3960 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
3961 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
3962 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
3964 p = file_to_first_proc(fp);
3965 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
3966 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
3967 fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode,
3968 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3973 DB_SHOW_COMMAND(file, db_show_file)
3978 db_printf("usage: show file <addr>\n");
3981 fp = (struct file *)addr;
3982 db_print_file(fp, 1);
3985 DB_SHOW_COMMAND(files, db_show_files)
3987 struct filedesc *fdp;
3994 FOREACH_PROC_IN_SYSTEM(p) {
3995 if (p->p_state == PRS_NEW)
3997 if ((fdp = p->p_fd) == NULL)
3999 for (n = 0; n <= fdp->fd_lastfile; ++n) {
4000 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4002 db_print_file(fp, header);
4009 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4010 &maxfilesperproc, 0, "Maximum files allowed open per process");
4012 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4013 &maxfiles, 0, "Maximum number of files");
4015 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4016 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
4020 filelistinit(void *dummy)
4023 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4024 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4025 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4026 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4027 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4029 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4031 /*-------------------------------------------------------------------*/
4034 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4035 int flags, struct thread *td)
4042 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4050 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4058 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4066 badfo_kqfilter(struct file *fp, struct knote *kn)
4073 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4081 badfo_close(struct file *fp, struct thread *td)
4088 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4096 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4104 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4105 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4113 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4119 struct fileops badfileops = {
4120 .fo_read = badfo_readwrite,
4121 .fo_write = badfo_readwrite,
4122 .fo_truncate = badfo_truncate,
4123 .fo_ioctl = badfo_ioctl,
4124 .fo_poll = badfo_poll,
4125 .fo_kqfilter = badfo_kqfilter,
4126 .fo_stat = badfo_stat,
4127 .fo_close = badfo_close,
4128 .fo_chmod = badfo_chmod,
4129 .fo_chown = badfo_chown,
4130 .fo_sendfile = badfo_sendfile,
4131 .fo_fill_kinfo = badfo_fill_kinfo,
4135 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4136 int flags, struct thread *td)
4139 return (EOPNOTSUPP);
4143 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4151 invfo_ioctl(struct file *fp, u_long com, void *data,
4152 struct ucred *active_cred, struct thread *td)
4159 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4163 return (poll_no_poll(events));
4167 invfo_kqfilter(struct file *fp, struct knote *kn)
4174 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4182 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4190 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4191 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4198 /*-------------------------------------------------------------------*/
4201 * File Descriptor pseudo-device driver (/dev/fd/).
4203 * Opening minor device N dup()s the file (if any) connected to file
4204 * descriptor N belonging to the calling process. Note that this driver
4205 * consists of only the ``open()'' routine, because all subsequent
4206 * references to this file will be direct to the other driver.
4208 * XXX: we could give this one a cloning event handler if necessary.
4213 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4217 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4218 * the file descriptor being sought for duplication. The error
4219 * return ensures that the vnode for this device will be released
4220 * by vn_open. Open will detect this special error and take the
4221 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4222 * will simply report the error.
4224 td->td_dupfd = dev2unit(dev);
4228 static struct cdevsw fildesc_cdevsw = {
4229 .d_version = D_VERSION,
4235 fildesc_drvinit(void *unused)
4239 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4240 UID_ROOT, GID_WHEEL, 0666, "fd/0");
4241 make_dev_alias(dev, "stdin");
4242 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4243 UID_ROOT, GID_WHEEL, 0666, "fd/1");
4244 make_dev_alias(dev, "stdout");
4245 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4246 UID_ROOT, GID_WHEEL, 0666, "fd/2");
4247 make_dev_alias(dev, "stderr");
4250 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);