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 arg = MIN(arg, INT_MAX - bsize + 1);
785 fp->f_seqcount = MIN(IO_SEQMAX,
786 (arg + bsize - 1) / bsize);
787 atomic_set_int(&fp->f_flag, FRDAHEAD);
789 atomic_clear_int(&fp->f_flag, FRDAHEAD);
803 getmaxfd(struct thread *td)
806 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
810 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
813 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
815 struct filedesc *fdp;
816 struct filedescent *oldfde, *newfde;
819 u_long *oioctls, *nioctls;
825 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
826 MPASS(mode < FDDUP_LASTMODE);
829 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
832 * Verify we have a valid descriptor to dup from and possibly to
833 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
834 * return EINVAL when the new descriptor is out of bounds.
839 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
840 maxfd = getmaxfd(td);
842 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
846 if (fget_locked(fdp, old) == NULL)
848 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
849 td->td_retval[0] = new;
850 if (flags & FDDUP_FLAG_CLOEXEC)
851 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
856 oldfde = &fdp->fd_ofiles[old];
857 if (!fhold(oldfde->fde_file))
861 * If the caller specified a file descriptor, make sure the file
862 * table is large enough to hold it, and grab it. Otherwise, just
863 * allocate a new descriptor the usual way.
868 if ((error = fdalloc(td, new, &new)) != 0) {
869 fdrop(oldfde->fde_file, td);
873 case FDDUP_MUSTREPLACE:
874 /* Target file descriptor must exist. */
875 if (fget_locked(fdp, new) == NULL) {
876 fdrop(oldfde->fde_file, td);
881 if (new >= fdp->fd_nfiles) {
883 * The resource limits are here instead of e.g.
884 * fdalloc(), because the file descriptor table may be
885 * shared between processes, so we can't really use
886 * racct_add()/racct_sub(). Instead of counting the
887 * number of actually allocated descriptors, just put
888 * the limit on the size of the file descriptor table.
893 error = racct_set(p, RACCT_NOFILE, new + 1);
897 fdrop(oldfde->fde_file, td);
902 fdgrowtable_exp(fdp, new + 1);
904 if (!fdisused(fdp, new))
908 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
911 KASSERT(old != new, ("new fd is same as old"));
913 newfde = &fdp->fd_ofiles[new];
914 delfp = newfde->fde_file;
916 oioctls = filecaps_free_prep(&newfde->fde_caps);
917 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
920 * Duplicate the source descriptor.
923 seq_write_begin(&newfde->fde_seq);
925 memcpy(newfde, oldfde, fde_change_size);
926 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
928 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
929 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
931 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
933 seq_write_end(&newfde->fde_seq);
935 filecaps_free_finish(oioctls);
936 td->td_retval[0] = new;
941 (void) closefp(fdp, new, delfp, td, 1);
942 FILEDESC_UNLOCK_ASSERT(fdp);
945 FILEDESC_XUNLOCK(fdp);
952 * If sigio is on the list associated with a process or process group,
953 * disable signalling from the device, remove sigio from the list and
957 funsetown(struct sigio **sigiop)
969 *(sigio->sio_myref) = NULL;
970 if ((sigio)->sio_pgid < 0) {
971 struct pgrp *pg = (sigio)->sio_pgrp;
973 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
977 struct proc *p = (sigio)->sio_proc;
979 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
984 crfree(sigio->sio_ucred);
985 free(sigio, M_SIGIO);
989 * Free a list of sigio structures.
990 * We only need to lock the SIGIO_LOCK because we have made ourselves
991 * inaccessible to callers of fsetown and therefore do not need to lock
992 * the proc or pgrp struct for the list manipulation.
995 funsetownlst(struct sigiolst *sigiolst)
1001 sigio = SLIST_FIRST(sigiolst);
1008 * Every entry of the list should belong
1009 * to a single proc or pgrp.
1011 if (sigio->sio_pgid < 0) {
1012 pg = sigio->sio_pgrp;
1013 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
1014 } else /* if (sigio->sio_pgid > 0) */ {
1015 p = sigio->sio_proc;
1016 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1020 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1021 *(sigio->sio_myref) = NULL;
1023 KASSERT(sigio->sio_pgid < 0,
1024 ("Proc sigio in pgrp sigio list"));
1025 KASSERT(sigio->sio_pgrp == pg,
1026 ("Bogus pgrp in sigio list"));
1028 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1031 } else /* if (p != NULL) */ {
1032 KASSERT(sigio->sio_pgid > 0,
1033 ("Pgrp sigio in proc sigio list"));
1034 KASSERT(sigio->sio_proc == p,
1035 ("Bogus proc in sigio list"));
1037 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1042 crfree(sigio->sio_ucred);
1043 free(sigio, M_SIGIO);
1050 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1052 * After permission checking, add a sigio structure to the sigio list for
1053 * the process or process group.
1056 fsetown(pid_t pgid, struct sigio **sigiop)
1060 struct sigio *sigio;
1070 /* Allocate and fill in the new sigio out of locks. */
1071 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1072 sigio->sio_pgid = pgid;
1073 sigio->sio_ucred = crhold(curthread->td_ucred);
1074 sigio->sio_myref = sigiop;
1076 sx_slock(&proctree_lock);
1085 * Policy - Don't allow a process to FSETOWN a process
1086 * in another session.
1088 * Remove this test to allow maximum flexibility or
1089 * restrict FSETOWN to the current process or process
1090 * group for maximum safety.
1093 if (proc->p_session != curthread->td_proc->p_session) {
1099 } else /* if (pgid < 0) */ {
1100 pgrp = pgfind(-pgid);
1108 * Policy - Don't allow a process to FSETOWN a process
1109 * in another session.
1111 * Remove this test to allow maximum flexibility or
1112 * restrict FSETOWN to the current process or process
1113 * group for maximum safety.
1115 if (pgrp->pg_session != curthread->td_proc->p_session) {
1126 * Since funsetownlst() is called without the proctree
1127 * locked, we need to check for P_WEXIT.
1128 * XXX: is ESRCH correct?
1130 if ((proc->p_flag & P_WEXIT) != 0) {
1135 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1136 sigio->sio_proc = proc;
1140 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1141 sigio->sio_pgrp = pgrp;
1144 sx_sunlock(&proctree_lock);
1151 sx_sunlock(&proctree_lock);
1152 crfree(sigio->sio_ucred);
1153 free(sigio, M_SIGIO);
1158 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1161 fgetown(struct sigio **sigiop)
1166 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1172 * Function drops the filedesc lock on return.
1175 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1180 FILEDESC_XLOCK_ASSERT(fdp);
1183 if (td->td_proc->p_fdtol != NULL) {
1185 * Ask fdfree() to sleep to ensure that all relevant
1186 * process leaders can be traversed in closef().
1188 fdp->fd_holdleaderscount++;
1195 * We now hold the fp reference that used to be owned by the
1196 * descriptor array. We have to unlock the FILEDESC *AFTER*
1197 * knote_fdclose to prevent a race of the fd getting opened, a knote
1198 * added, and deleteing a knote for the new fd.
1200 knote_fdclose(td, fd);
1203 * We need to notify mqueue if the object is of type mqueue.
1205 if (fp->f_type == DTYPE_MQUEUE)
1206 mq_fdclose(td, fd, fp);
1207 FILEDESC_XUNLOCK(fdp);
1209 error = closef(fp, td);
1211 FILEDESC_XLOCK(fdp);
1212 fdp->fd_holdleaderscount--;
1213 if (fdp->fd_holdleaderscount == 0 &&
1214 fdp->fd_holdleaderswakeup != 0) {
1215 fdp->fd_holdleaderswakeup = 0;
1216 wakeup(&fdp->fd_holdleaderscount);
1218 FILEDESC_XUNLOCK(fdp);
1224 * Close a file descriptor.
1226 #ifndef _SYS_SYSPROTO_H_
1233 sys_close(struct thread *td, struct close_args *uap)
1236 return (kern_close(td, uap->fd));
1240 kern_close(struct thread *td, int fd)
1242 struct filedesc *fdp;
1245 fdp = td->td_proc->p_fd;
1247 AUDIT_SYSCLOSE(td, fd);
1249 FILEDESC_XLOCK(fdp);
1250 if ((fp = fget_locked(fdp, fd)) == NULL) {
1251 FILEDESC_XUNLOCK(fdp);
1256 /* closefp() drops the FILEDESC lock for us. */
1257 return (closefp(fdp, fd, fp, td, 1));
1261 * Close open file descriptors.
1263 #ifndef _SYS_SYSPROTO_H_
1264 struct closefrom_args {
1270 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1272 struct filedesc *fdp;
1275 fdp = td->td_proc->p_fd;
1276 AUDIT_ARG_FD(uap->lowfd);
1279 * Treat negative starting file descriptor values identical to
1280 * closefrom(0) which closes all files.
1284 FILEDESC_SLOCK(fdp);
1285 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1286 if (fdp->fd_ofiles[fd].fde_file != NULL) {
1287 FILEDESC_SUNLOCK(fdp);
1288 (void)kern_close(td, fd);
1289 FILEDESC_SLOCK(fdp);
1292 FILEDESC_SUNLOCK(fdp);
1296 #if defined(COMPAT_43)
1298 * Return status information about a file descriptor.
1300 #ifndef _SYS_SYSPROTO_H_
1301 struct ofstat_args {
1308 ofstat(struct thread *td, struct ofstat_args *uap)
1314 error = kern_fstat(td, uap->fd, &ub);
1317 error = copyout(&oub, uap->sb, sizeof(oub));
1321 #endif /* COMPAT_43 */
1323 #if defined(COMPAT_FREEBSD11)
1325 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1328 struct freebsd11_stat osb;
1331 error = kern_fstat(td, uap->fd, &sb);
1334 error = freebsd11_cvtstat(&sb, &osb);
1336 error = copyout(&osb, uap->sb, sizeof(osb));
1339 #endif /* COMPAT_FREEBSD11 */
1342 * Return status information about a file descriptor.
1344 #ifndef _SYS_SYSPROTO_H_
1352 sys_fstat(struct thread *td, struct fstat_args *uap)
1357 error = kern_fstat(td, uap->fd, &ub);
1359 error = copyout(&ub, uap->sb, sizeof(ub));
1364 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1371 error = fget(td, fd, &cap_fstat_rights, &fp);
1375 AUDIT_ARG_FILE(td->td_proc, fp);
1377 error = fo_stat(fp, sbp, td->td_ucred, td);
1379 #ifdef __STAT_TIME_T_EXT
1381 sbp->st_atim_ext = 0;
1382 sbp->st_mtim_ext = 0;
1383 sbp->st_ctim_ext = 0;
1384 sbp->st_btim_ext = 0;
1388 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1394 #if defined(COMPAT_FREEBSD11)
1396 * Return status information about a file descriptor.
1398 #ifndef _SYS_SYSPROTO_H_
1399 struct freebsd11_nfstat_args {
1406 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1412 error = kern_fstat(td, uap->fd, &ub);
1414 freebsd11_cvtnstat(&ub, &nub);
1415 error = copyout(&nub, uap->sb, sizeof(nub));
1419 #endif /* COMPAT_FREEBSD11 */
1422 * Return pathconf information about a file descriptor.
1424 #ifndef _SYS_SYSPROTO_H_
1425 struct fpathconf_args {
1432 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1437 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1439 td->td_retval[0] = value;
1444 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1450 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1454 if (name == _PC_ASYNC_IO) {
1455 *valuep = _POSIX_ASYNCHRONOUS_IO;
1460 vn_lock(vp, LK_SHARED | LK_RETRY);
1461 error = VOP_PATHCONF(vp, name, valuep);
1463 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1464 if (name != _PC_PIPE_BUF) {
1479 * Initialize filecaps structure.
1482 filecaps_init(struct filecaps *fcaps)
1485 bzero(fcaps, sizeof(*fcaps));
1486 fcaps->fc_nioctls = -1;
1490 * Copy filecaps structure allocating memory for ioctls array if needed.
1492 * The last parameter indicates whether the fdtable is locked. If it is not and
1493 * ioctls are encountered, copying fails and the caller must lock the table.
1495 * Note that if the table was not locked, the caller has to check the relevant
1496 * sequence counter to determine whether the operation was successful.
1499 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1503 if (src->fc_ioctls != NULL && !locked)
1505 memcpy(dst, src, sizeof(*src));
1506 if (src->fc_ioctls == NULL)
1509 KASSERT(src->fc_nioctls > 0,
1510 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1512 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1513 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1514 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1519 filecaps_copy_prep(const struct filecaps *src)
1524 if (src->fc_ioctls == NULL)
1527 KASSERT(src->fc_nioctls > 0,
1528 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1530 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1531 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1536 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1542 if (src->fc_ioctls == NULL) {
1543 MPASS(ioctls == NULL);
1547 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1548 dst->fc_ioctls = ioctls;
1549 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1553 * Move filecaps structure to the new place and clear the old place.
1556 filecaps_move(struct filecaps *src, struct filecaps *dst)
1560 bzero(src, sizeof(*src));
1564 * Fill the given filecaps structure with full rights.
1567 filecaps_fill(struct filecaps *fcaps)
1570 CAP_ALL(&fcaps->fc_rights);
1571 fcaps->fc_ioctls = NULL;
1572 fcaps->fc_nioctls = -1;
1573 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1577 * Free memory allocated within filecaps structure.
1580 filecaps_free(struct filecaps *fcaps)
1583 free(fcaps->fc_ioctls, M_FILECAPS);
1584 bzero(fcaps, sizeof(*fcaps));
1588 filecaps_free_prep(struct filecaps *fcaps)
1592 ioctls = fcaps->fc_ioctls;
1593 bzero(fcaps, sizeof(*fcaps));
1598 filecaps_free_finish(u_long *ioctls)
1601 free(ioctls, M_FILECAPS);
1605 * Validate the given filecaps structure.
1608 filecaps_validate(const struct filecaps *fcaps, const char *func)
1611 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1612 ("%s: invalid rights", func));
1613 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1614 ("%s: invalid fcntls", func));
1615 KASSERT(fcaps->fc_fcntls == 0 ||
1616 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1617 ("%s: fcntls without CAP_FCNTL", func));
1618 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1619 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1620 ("%s: invalid ioctls", func));
1621 KASSERT(fcaps->fc_nioctls == 0 ||
1622 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1623 ("%s: ioctls without CAP_IOCTL", func));
1627 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1631 FILEDESC_XLOCK_ASSERT(fdp);
1633 nfd1 = fdp->fd_nfiles * 2;
1636 fdgrowtable(fdp, nfd1);
1640 * Grow the file table to accommodate (at least) nfd descriptors.
1643 fdgrowtable(struct filedesc *fdp, int nfd)
1645 struct filedesc0 *fdp0;
1646 struct freetable *ft;
1647 struct fdescenttbl *ntable;
1648 struct fdescenttbl *otable;
1649 int nnfiles, onfiles;
1650 NDSLOTTYPE *nmap, *omap;
1653 * If lastfile is -1 this struct filedesc was just allocated and we are
1654 * growing it to accommodate for the one we are going to copy from. There
1655 * is no need to have a lock on this one as it's not visible to anyone.
1657 if (fdp->fd_lastfile != -1)
1658 FILEDESC_XLOCK_ASSERT(fdp);
1660 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1662 /* save old values */
1663 onfiles = fdp->fd_nfiles;
1664 otable = fdp->fd_files;
1667 /* compute the size of the new table */
1668 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1669 if (nnfiles <= onfiles)
1670 /* the table is already large enough */
1674 * Allocate a new table. We need enough space for the number of
1675 * entries, file entries themselves and the struct freetable we will use
1676 * when we decommission the table and place it on the freelist.
1677 * We place the struct freetable in the middle so we don't have
1678 * to worry about padding.
1680 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1681 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1682 sizeof(struct freetable),
1683 M_FILEDESC, M_ZERO | M_WAITOK);
1684 /* copy the old data */
1685 ntable->fdt_nfiles = nnfiles;
1686 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1687 onfiles * sizeof(ntable->fdt_ofiles[0]));
1690 * Allocate a new map only if the old is not large enough. It will
1691 * grow at a slower rate than the table as it can map more
1692 * entries than the table can hold.
1694 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1695 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1697 /* copy over the old data and update the pointer */
1698 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1703 * Make sure that ntable is correctly initialized before we replace
1704 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1707 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1710 * Do not free the old file table, as some threads may still
1711 * reference entries within it. Instead, place it on a freelist
1712 * which will be processed when the struct filedesc is released.
1714 * Note that if onfiles == NDFILE, we're dealing with the original
1715 * static allocation contained within (struct filedesc0 *)fdp,
1716 * which must not be freed.
1718 if (onfiles > NDFILE) {
1719 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1720 fdp0 = (struct filedesc0 *)fdp;
1721 ft->ft_table = otable;
1722 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1725 * The map does not have the same possibility of threads still
1726 * holding references to it. So always free it as long as it
1727 * does not reference the original static allocation.
1729 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1730 free(omap, M_FILEDESC);
1734 * Allocate a file descriptor for the process.
1737 fdalloc(struct thread *td, int minfd, int *result)
1739 struct proc *p = td->td_proc;
1740 struct filedesc *fdp = p->p_fd;
1741 int fd, maxfd, allocfd;
1746 FILEDESC_XLOCK_ASSERT(fdp);
1748 if (fdp->fd_freefile > minfd)
1749 minfd = fdp->fd_freefile;
1751 maxfd = getmaxfd(td);
1754 * Search the bitmap for a free descriptor starting at minfd.
1755 * If none is found, grow the file table.
1757 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1760 if (fd >= fdp->fd_nfiles) {
1761 allocfd = min(fd * 2, maxfd);
1765 error = racct_set(p, RACCT_NOFILE, allocfd);
1772 * fd is already equal to first free descriptor >= minfd, so
1773 * we only need to grow the table and we are done.
1775 fdgrowtable_exp(fdp, allocfd);
1779 * Perform some sanity checks, then mark the file descriptor as
1780 * used and return it to the caller.
1782 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1783 ("invalid descriptor %d", fd));
1784 KASSERT(!fdisused(fdp, fd),
1785 ("fd_first_free() returned non-free descriptor"));
1786 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1787 ("file descriptor isn't free"));
1794 * Allocate n file descriptors for the process.
1797 fdallocn(struct thread *td, int minfd, int *fds, int n)
1799 struct proc *p = td->td_proc;
1800 struct filedesc *fdp = p->p_fd;
1803 FILEDESC_XLOCK_ASSERT(fdp);
1805 for (i = 0; i < n; i++)
1806 if (fdalloc(td, 0, &fds[i]) != 0)
1810 for (i--; i >= 0; i--)
1811 fdunused(fdp, fds[i]);
1819 * Create a new open file structure and allocate a file descriptor for the
1820 * process that refers to it. We add one reference to the file for the
1821 * descriptor table and one reference for resultfp. This is to prevent us
1822 * being preempted and the entry in the descriptor table closed after we
1823 * release the FILEDESC lock.
1826 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1827 struct filecaps *fcaps)
1832 error = falloc_noinstall(td, &fp);
1834 return (error); /* no reference held on error */
1836 error = finstall(td, fp, &fd, flags, fcaps);
1838 fdrop(fp, td); /* one reference (fp only) */
1842 if (resultfp != NULL)
1843 *resultfp = fp; /* copy out result */
1845 fdrop(fp, td); /* release local reference */
1847 if (resultfd != NULL)
1854 * Create a new open file structure without allocating a file descriptor.
1857 falloc_noinstall(struct thread *td, struct file **resultfp)
1860 int maxuserfiles = maxfiles - (maxfiles / 20);
1862 static struct timeval lastfail;
1865 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1867 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1868 if ((openfiles_new >= maxuserfiles &&
1869 priv_check(td, PRIV_MAXFILES) != 0) ||
1870 openfiles_new >= maxfiles) {
1871 atomic_subtract_int(&openfiles, 1);
1872 if (ppsratecheck(&lastfail, &curfail, 1)) {
1873 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1874 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1878 fp = uma_zalloc(file_zone, M_WAITOK);
1879 bzero(fp, sizeof(*fp));
1880 refcount_init(&fp->f_count, 1);
1881 fp->f_cred = crhold(td->td_ucred);
1882 fp->f_ops = &badfileops;
1888 * Install a file in a file descriptor table.
1891 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1892 struct filecaps *fcaps)
1894 struct filedescent *fde;
1898 filecaps_validate(fcaps, __func__);
1899 FILEDESC_XLOCK_ASSERT(fdp);
1901 fde = &fdp->fd_ofiles[fd];
1903 seq_write_begin(&fde->fde_seq);
1906 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1908 filecaps_move(fcaps, &fde->fde_caps);
1910 filecaps_fill(&fde->fde_caps);
1912 seq_write_end(&fde->fde_seq);
1917 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1918 struct filecaps *fcaps)
1920 struct filedesc *fdp = td->td_proc->p_fd;
1927 FILEDESC_XLOCK(fdp);
1928 if ((error = fdalloc(td, 0, fd))) {
1929 FILEDESC_XUNLOCK(fdp);
1933 _finstall(fdp, fp, *fd, flags, fcaps);
1934 FILEDESC_XUNLOCK(fdp);
1939 * Build a new filedesc structure from another.
1940 * Copy the current, root, and jail root vnode references.
1942 * If fdp is not NULL, return with it shared locked.
1945 fdinit(struct filedesc *fdp, bool prepfiles)
1947 struct filedesc0 *newfdp0;
1948 struct filedesc *newfdp;
1950 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1951 newfdp = &newfdp0->fd_fd;
1953 /* Create the file descriptor table. */
1954 FILEDESC_LOCK_INIT(newfdp);
1955 refcount_init(&newfdp->fd_refcnt, 1);
1956 refcount_init(&newfdp->fd_holdcnt, 1);
1957 newfdp->fd_cmask = CMASK;
1958 newfdp->fd_map = newfdp0->fd_dmap;
1959 newfdp->fd_lastfile = -1;
1960 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1961 newfdp->fd_files->fdt_nfiles = NDFILE;
1966 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1967 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1969 FILEDESC_SLOCK(fdp);
1970 newfdp->fd_cdir = fdp->fd_cdir;
1971 if (newfdp->fd_cdir)
1972 vrefact(newfdp->fd_cdir);
1973 newfdp->fd_rdir = fdp->fd_rdir;
1974 if (newfdp->fd_rdir)
1975 vrefact(newfdp->fd_rdir);
1976 newfdp->fd_jdir = fdp->fd_jdir;
1977 if (newfdp->fd_jdir)
1978 vrefact(newfdp->fd_jdir);
1981 FILEDESC_SUNLOCK(fdp);
1983 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1984 FILEDESC_SUNLOCK(fdp);
1985 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1986 FILEDESC_SLOCK(fdp);
1993 static struct filedesc *
1994 fdhold(struct proc *p)
1996 struct filedesc *fdp;
1998 PROC_LOCK_ASSERT(p, MA_OWNED);
2001 refcount_acquire(&fdp->fd_holdcnt);
2006 fddrop(struct filedesc *fdp)
2009 if (fdp->fd_holdcnt > 1) {
2010 if (refcount_release(&fdp->fd_holdcnt) == 0)
2014 FILEDESC_LOCK_DESTROY(fdp);
2015 uma_zfree(filedesc0_zone, fdp);
2019 * Share a filedesc structure.
2022 fdshare(struct filedesc *fdp)
2025 refcount_acquire(&fdp->fd_refcnt);
2030 * Unshare a filedesc structure, if necessary by making a copy
2033 fdunshare(struct thread *td)
2035 struct filedesc *tmp;
2036 struct proc *p = td->td_proc;
2038 if (p->p_fd->fd_refcnt == 1)
2041 tmp = fdcopy(p->p_fd);
2047 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2051 td->td_proc->p_fd = fdp;
2055 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2056 * this is to ease callers, not catch errors.
2059 fdcopy(struct filedesc *fdp)
2061 struct filedesc *newfdp;
2062 struct filedescent *nfde, *ofde;
2067 newfdp = fdinit(fdp, true);
2068 /* copy all passable descriptors (i.e. not kqueue) */
2069 newfdp->fd_freefile = -1;
2070 for (i = 0; i <= fdp->fd_lastfile; ++i) {
2071 ofde = &fdp->fd_ofiles[i];
2072 if (ofde->fde_file == NULL ||
2073 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2074 !fhold(ofde->fde_file)) {
2075 if (newfdp->fd_freefile == -1)
2076 newfdp->fd_freefile = i;
2079 nfde = &newfdp->fd_ofiles[i];
2081 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2082 fdused_init(newfdp, i);
2083 newfdp->fd_lastfile = i;
2085 if (newfdp->fd_freefile == -1)
2086 newfdp->fd_freefile = i;
2087 newfdp->fd_cmask = fdp->fd_cmask;
2088 FILEDESC_SUNLOCK(fdp);
2093 * Copies a filedesc structure, while remapping all file descriptors
2094 * stored inside using a translation table.
2096 * File descriptors are copied over to the new file descriptor table,
2097 * regardless of whether the close-on-exec flag is set.
2100 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2101 struct filedesc **ret)
2103 struct filedesc *newfdp;
2104 struct filedescent *nfde, *ofde;
2109 newfdp = fdinit(fdp, true);
2110 if (nfds > fdp->fd_lastfile + 1) {
2111 /* New table cannot be larger than the old one. */
2115 /* Copy all passable descriptors (i.e. not kqueue). */
2116 newfdp->fd_freefile = nfds;
2117 for (i = 0; i < nfds; ++i) {
2118 if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2119 /* File descriptor out of bounds. */
2123 ofde = &fdp->fd_ofiles[fds[i]];
2124 if (ofde->fde_file == NULL) {
2125 /* Unused file descriptor. */
2129 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2130 /* File descriptor cannot be passed. */
2134 if (!fhold(nfde->fde_file)) {
2138 nfde = &newfdp->fd_ofiles[i];
2140 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2141 fdused_init(newfdp, i);
2142 newfdp->fd_lastfile = i;
2144 newfdp->fd_cmask = fdp->fd_cmask;
2145 FILEDESC_SUNLOCK(fdp);
2149 FILEDESC_SUNLOCK(fdp);
2150 fdescfree_remapped(newfdp);
2155 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2156 * one of processes using it exits) and the table used to be shared.
2159 fdclearlocks(struct thread *td)
2161 struct filedesc *fdp;
2162 struct filedesc_to_leader *fdtol;
2172 MPASS(fdtol != NULL);
2174 FILEDESC_XLOCK(fdp);
2175 KASSERT(fdtol->fdl_refcount > 0,
2176 ("filedesc_to_refcount botch: fdl_refcount=%d",
2177 fdtol->fdl_refcount));
2178 if (fdtol->fdl_refcount == 1 &&
2179 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2180 for (i = 0; i <= fdp->fd_lastfile; i++) {
2181 fp = fdp->fd_ofiles[i].fde_file;
2182 if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2185 FILEDESC_XUNLOCK(fdp);
2186 lf.l_whence = SEEK_SET;
2189 lf.l_type = F_UNLCK;
2191 (void) VOP_ADVLOCK(vp,
2192 (caddr_t)p->p_leader, F_UNLCK,
2194 FILEDESC_XLOCK(fdp);
2199 if (fdtol->fdl_refcount == 1) {
2200 if (fdp->fd_holdleaderscount > 0 &&
2201 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2203 * close() or kern_dup() has cleared a reference
2204 * in a shared file descriptor table.
2206 fdp->fd_holdleaderswakeup = 1;
2207 sx_sleep(&fdp->fd_holdleaderscount,
2208 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2211 if (fdtol->fdl_holdcount > 0) {
2213 * Ensure that fdtol->fdl_leader remains
2214 * valid in closef().
2216 fdtol->fdl_wakeup = 1;
2217 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2222 fdtol->fdl_refcount--;
2223 if (fdtol->fdl_refcount == 0 &&
2224 fdtol->fdl_holdcount == 0) {
2225 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2226 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2230 FILEDESC_XUNLOCK(fdp);
2232 free(fdtol, M_FILEDESC_TO_LEADER);
2236 * Release a filedesc structure.
2239 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2241 struct filedesc0 *fdp0;
2242 struct freetable *ft, *tft;
2243 struct filedescent *fde;
2247 for (i = 0; i <= fdp->fd_lastfile; i++) {
2248 fde = &fdp->fd_ofiles[i];
2253 (void) closef(fp, td);
2259 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2260 free(fdp->fd_map, M_FILEDESC);
2261 if (fdp->fd_nfiles > NDFILE)
2262 free(fdp->fd_files, M_FILEDESC);
2264 fdp0 = (struct filedesc0 *)fdp;
2265 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2266 free(ft->ft_table, M_FILEDESC);
2272 fdescfree(struct thread *td)
2275 struct filedesc *fdp;
2276 struct vnode *cdir, *jdir, *rdir;
2285 racct_set(p, RACCT_NOFILE, 0);
2290 if (p->p_fdtol != NULL)
2297 if (refcount_release(&fdp->fd_refcnt) == 0)
2300 FILEDESC_XLOCK(fdp);
2301 cdir = fdp->fd_cdir;
2302 fdp->fd_cdir = NULL;
2303 rdir = fdp->fd_rdir;
2304 fdp->fd_rdir = NULL;
2305 jdir = fdp->fd_jdir;
2306 fdp->fd_jdir = NULL;
2307 FILEDESC_XUNLOCK(fdp);
2316 fdescfree_fds(td, fdp, 1);
2320 fdescfree_remapped(struct filedesc *fdp)
2323 if (fdp->fd_cdir != NULL)
2324 vrele(fdp->fd_cdir);
2325 if (fdp->fd_rdir != NULL)
2326 vrele(fdp->fd_rdir);
2327 if (fdp->fd_jdir != NULL)
2328 vrele(fdp->fd_jdir);
2330 fdescfree_fds(curthread, fdp, 0);
2334 * For setugid programs, we don't want to people to use that setugidness
2335 * to generate error messages which write to a file which otherwise would
2336 * otherwise be off-limits to the process. We check for filesystems where
2337 * the vnode can change out from under us after execve (like [lin]procfs).
2339 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2340 * sufficient. We also don't check for setugidness since we know we are.
2343 is_unsafe(struct file *fp)
2347 if (fp->f_type != DTYPE_VNODE)
2351 return ((vp->v_vflag & VV_PROCDEP) != 0);
2355 * Make this setguid thing safe, if at all possible.
2358 fdsetugidsafety(struct thread *td)
2360 struct filedesc *fdp;
2364 fdp = td->td_proc->p_fd;
2365 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2366 MPASS(fdp->fd_nfiles >= 3);
2367 for (i = 0; i <= 2; i++) {
2368 fp = fdp->fd_ofiles[i].fde_file;
2369 if (fp != NULL && is_unsafe(fp)) {
2370 FILEDESC_XLOCK(fdp);
2371 knote_fdclose(td, i);
2373 * NULL-out descriptor prior to close to avoid
2374 * a race while close blocks.
2377 FILEDESC_XUNLOCK(fdp);
2378 (void) closef(fp, td);
2384 * If a specific file object occupies a specific file descriptor, close the
2385 * file descriptor entry and drop a reference on the file object. This is a
2386 * convenience function to handle a subsequent error in a function that calls
2387 * falloc() that handles the race that another thread might have closed the
2388 * file descriptor out from under the thread creating the file object.
2391 fdclose(struct thread *td, struct file *fp, int idx)
2393 struct filedesc *fdp = td->td_proc->p_fd;
2395 FILEDESC_XLOCK(fdp);
2396 if (fdp->fd_ofiles[idx].fde_file == fp) {
2398 FILEDESC_XUNLOCK(fdp);
2401 FILEDESC_XUNLOCK(fdp);
2405 * Close any files on exec?
2408 fdcloseexec(struct thread *td)
2410 struct filedesc *fdp;
2411 struct filedescent *fde;
2415 fdp = td->td_proc->p_fd;
2416 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2417 for (i = 0; i <= fdp->fd_lastfile; i++) {
2418 fde = &fdp->fd_ofiles[i];
2420 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2421 (fde->fde_flags & UF_EXCLOSE))) {
2422 FILEDESC_XLOCK(fdp);
2424 (void) closefp(fdp, i, fp, td, 0);
2425 FILEDESC_UNLOCK_ASSERT(fdp);
2431 * It is unsafe for set[ug]id processes to be started with file
2432 * descriptors 0..2 closed, as these descriptors are given implicit
2433 * significance in the Standard C library. fdcheckstd() will create a
2434 * descriptor referencing /dev/null for each of stdin, stdout, and
2435 * stderr that is not already open.
2438 fdcheckstd(struct thread *td)
2440 struct filedesc *fdp;
2442 int i, error, devnull;
2444 fdp = td->td_proc->p_fd;
2445 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2446 MPASS(fdp->fd_nfiles >= 3);
2448 for (i = 0; i <= 2; i++) {
2449 if (fdp->fd_ofiles[i].fde_file != NULL)
2452 save = td->td_retval[0];
2453 if (devnull != -1) {
2454 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2456 error = kern_openat(td, AT_FDCWD, "/dev/null",
2457 UIO_SYSSPACE, O_RDWR, 0);
2459 devnull = td->td_retval[0];
2460 KASSERT(devnull == i, ("we didn't get our fd"));
2463 td->td_retval[0] = save;
2471 * Internal form of close. Decrement reference count on file structure.
2472 * Note: td may be NULL when closing a file that was being passed in a
2476 closef(struct file *fp, struct thread *td)
2480 struct filedesc_to_leader *fdtol;
2481 struct filedesc *fdp;
2484 * POSIX record locking dictates that any close releases ALL
2485 * locks owned by this process. This is handled by setting
2486 * a flag in the unlock to free ONLY locks obeying POSIX
2487 * semantics, and not to free BSD-style file locks.
2488 * If the descriptor was in a message, POSIX-style locks
2489 * aren't passed with the descriptor, and the thread pointer
2490 * will be NULL. Callers should be careful only to pass a
2491 * NULL thread pointer when there really is no owning
2492 * context that might have locks, or the locks will be
2495 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2497 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2498 lf.l_whence = SEEK_SET;
2501 lf.l_type = F_UNLCK;
2502 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2503 F_UNLCK, &lf, F_POSIX);
2505 fdtol = td->td_proc->p_fdtol;
2506 if (fdtol != NULL) {
2508 * Handle special case where file descriptor table is
2509 * shared between multiple process leaders.
2511 fdp = td->td_proc->p_fd;
2512 FILEDESC_XLOCK(fdp);
2513 for (fdtol = fdtol->fdl_next;
2514 fdtol != td->td_proc->p_fdtol;
2515 fdtol = fdtol->fdl_next) {
2516 if ((fdtol->fdl_leader->p_flag &
2519 fdtol->fdl_holdcount++;
2520 FILEDESC_XUNLOCK(fdp);
2521 lf.l_whence = SEEK_SET;
2524 lf.l_type = F_UNLCK;
2526 (void) VOP_ADVLOCK(vp,
2527 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2529 FILEDESC_XLOCK(fdp);
2530 fdtol->fdl_holdcount--;
2531 if (fdtol->fdl_holdcount == 0 &&
2532 fdtol->fdl_wakeup != 0) {
2533 fdtol->fdl_wakeup = 0;
2537 FILEDESC_XUNLOCK(fdp);
2540 return (fdrop(fp, td));
2544 * Initialize the file pointer with the specified properties.
2546 * The ops are set with release semantics to be certain that the flags, type,
2547 * and data are visible when ops is. This is to prevent ops methods from being
2548 * called with bad data.
2551 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2556 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2560 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2561 struct file **fpp, struct filecaps *havecapsp)
2563 struct filedescent *fde;
2566 FILEDESC_LOCK_ASSERT(fdp);
2568 fde = fdeget_locked(fdp, fd);
2575 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2580 if (havecapsp != NULL)
2581 filecaps_copy(&fde->fde_caps, havecapsp, true);
2583 *fpp = fde->fde_file;
2591 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2592 struct file **fpp, struct filecaps *havecapsp)
2594 struct filedesc *fdp = td->td_proc->p_fd;
2596 #ifndef CAPABILITIES
2597 error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2598 if (error == 0 && havecapsp != NULL)
2599 filecaps_fill(havecapsp);
2605 error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2609 if (havecapsp != NULL) {
2610 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2611 havecapsp, false)) {
2617 if (!fd_modified(fdp, fd, seq))
2626 FILEDESC_SLOCK(fdp);
2627 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2628 if (error == 0 && !fhold(*fpp))
2630 FILEDESC_SUNLOCK(fdp);
2636 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2637 struct file **fpp, seq_t *seqp)
2640 const struct filedescent *fde;
2642 const struct fdescenttbl *fdt;
2647 cap_rights_t haverights;
2651 fdt = fdp->fd_files;
2652 if ((u_int)fd >= fdt->fdt_nfiles)
2655 * Fetch the descriptor locklessly. We avoid fdrop() races by
2656 * never raising a refcount above 0. To accomplish this we have
2657 * to use a cmpset loop rather than an atomic_add. The descriptor
2658 * must be re-verified once we acquire a reference to be certain
2659 * that the identity is still correct and we did not lose a race
2660 * due to preemption.
2664 seq = seq_load(fd_seq(fdt, fd));
2665 fde = &fdt->fdt_ofiles[fd];
2666 haverights = *cap_rights_fde_inline(fde);
2668 if (!seq_consistent(fd_seq(fdt, fd), seq))
2671 fp = fdt->fdt_ofiles[fd].fde_file;
2676 error = cap_check(&haverights, needrightsp);
2680 count = fp->f_count;
2684 * Force a reload. Other thread could reallocate the
2685 * table before this fd was closed, so it possible that
2686 * there is a stale fp pointer in cached version.
2688 fdt = *(const struct fdescenttbl * const volatile *)
2692 if (__predict_false(count + 1 < count))
2696 * Use an acquire barrier to force re-reading of fdt so it is
2697 * refreshed for verification.
2699 if (__predict_false(atomic_fcmpset_acq_int(&fp->f_count,
2700 &count, count + 1) == 0))
2702 fdt = fdp->fd_files;
2704 if (seq_consistent_nomb(fd_seq(fdt, fd), seq))
2706 if (fp == fdt->fdt_ofiles[fd].fde_file)
2709 fdrop(fp, curthread);
2721 * Extract the file pointer associated with the specified descriptor for the
2722 * current user process.
2724 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2727 * File's rights will be checked against the capability rights mask.
2729 * If an error occurred the non-zero error is returned and *fpp is set to
2730 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
2731 * responsible for fdrop().
2734 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2735 cap_rights_t *needrightsp, seq_t *seqp)
2737 struct filedesc *fdp;
2742 fdp = td->td_proc->p_fd;
2743 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2746 if (fp->f_ops == &badfileops) {
2752 * FREAD and FWRITE failure return EBADF as per POSIX.
2758 if ((fp->f_flag & flags) == 0)
2762 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2763 ((fp->f_flag & FWRITE) != 0))
2769 KASSERT(0, ("wrong flags"));
2782 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2785 return (_fget(td, fd, fpp, 0, rightsp, NULL));
2789 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2793 #ifndef CAPABILITIES
2794 error = _fget(td, fd, fpp, 0, rightsp, NULL);
2795 if (maxprotp != NULL)
2796 *maxprotp = VM_PROT_ALL;
2798 cap_rights_t fdrights;
2799 struct filedesc *fdp = td->td_proc->p_fd;
2802 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2804 error = _fget(td, fd, fpp, 0, rightsp, &seq);
2807 if (maxprotp != NULL)
2808 fdrights = *cap_rights(fdp, fd);
2809 if (!fd_modified(fdp, fd, seq))
2815 * If requested, convert capability rights to access flags.
2817 if (maxprotp != NULL)
2818 *maxprotp = cap_rights_to_vmprot(&fdrights);
2824 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2827 return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2831 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2834 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2838 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2841 struct filedesc *fdp = td->td_proc->p_fd;
2842 #ifndef CAPABILITIES
2843 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2848 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2850 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2853 error = cap_fcntl_check(fdp, fd, needfcntl);
2854 if (!fd_modified(fdp, fd, seq))
2867 * Like fget() but loads the underlying vnode, or returns an error if the
2868 * descriptor does not represent a vnode. Note that pipes use vnodes but
2869 * never have VM objects. The returned vnode will be vref()'d.
2871 * XXX: what about the unused flags ?
2874 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2881 error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2884 if (fp->f_vnode == NULL) {
2896 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2899 return (_fgetvp(td, fd, 0, rightsp, vpp));
2903 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2904 struct filecaps *havecaps, struct vnode **vpp)
2906 struct filedesc *fdp;
2907 struct filecaps caps;
2911 fdp = td->td_proc->p_fd;
2912 error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2915 if (fp->f_ops == &badfileops) {
2919 if (fp->f_vnode == NULL) {
2930 filecaps_free(&caps);
2935 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2938 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2942 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2945 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2950 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2954 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2959 * Handle the last reference to a file being closed.
2961 * Without the noinline attribute clang keeps inlining the func thorough this
2962 * file when fdrop is used.
2965 _fdrop(struct file *fp, struct thread *td)
2969 if (fp->f_count != 0)
2970 panic("fdrop: count %d", fp->f_count);
2971 error = fo_close(fp, td);
2972 atomic_subtract_int(&openfiles, 1);
2974 free(fp->f_advice, M_FADVISE);
2975 uma_zfree(file_zone, fp);
2981 * Apply an advisory lock on a file descriptor.
2983 * Just attempt to get a record lock of the requested type on the entire file
2984 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2986 #ifndef _SYS_SYSPROTO_H_
2994 sys_flock(struct thread *td, struct flock_args *uap)
3001 error = fget(td, uap->fd, &cap_flock_rights, &fp);
3004 if (fp->f_type != DTYPE_VNODE) {
3006 return (EOPNOTSUPP);
3010 lf.l_whence = SEEK_SET;
3013 if (uap->how & LOCK_UN) {
3014 lf.l_type = F_UNLCK;
3015 atomic_clear_int(&fp->f_flag, FHASLOCK);
3016 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3019 if (uap->how & LOCK_EX)
3020 lf.l_type = F_WRLCK;
3021 else if (uap->how & LOCK_SH)
3022 lf.l_type = F_RDLCK;
3027 atomic_set_int(&fp->f_flag, FHASLOCK);
3028 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3029 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3035 * Duplicate the specified descriptor to a free descriptor.
3038 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3039 int openerror, int *indxp)
3041 struct filedescent *newfde, *oldfde;
3046 KASSERT(openerror == ENODEV || openerror == ENXIO,
3047 ("unexpected error %d in %s", openerror, __func__));
3050 * If the to-be-dup'd fd number is greater than the allowed number
3051 * of file descriptors, or the fd to be dup'd has already been
3052 * closed, then reject.
3054 FILEDESC_XLOCK(fdp);
3055 if ((fp = fget_locked(fdp, dfd)) == NULL) {
3056 FILEDESC_XUNLOCK(fdp);
3060 error = fdalloc(td, 0, &indx);
3062 FILEDESC_XUNLOCK(fdp);
3067 * There are two cases of interest here.
3069 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3071 * For ENXIO steal away the file structure from (dfd) and store it in
3072 * (indx). (dfd) is effectively closed by this operation.
3074 switch (openerror) {
3077 * Check that the mode the file is being opened for is a
3078 * subset of the mode of the existing descriptor.
3080 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3081 fdunused(fdp, indx);
3082 FILEDESC_XUNLOCK(fdp);
3086 fdunused(fdp, indx);
3087 FILEDESC_XUNLOCK(fdp);
3090 newfde = &fdp->fd_ofiles[indx];
3091 oldfde = &fdp->fd_ofiles[dfd];
3092 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3094 seq_write_begin(&newfde->fde_seq);
3096 memcpy(newfde, oldfde, fde_change_size);
3097 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3100 seq_write_end(&newfde->fde_seq);
3105 * Steal away the file pointer from dfd and stuff it into indx.
3107 newfde = &fdp->fd_ofiles[indx];
3108 oldfde = &fdp->fd_ofiles[dfd];
3110 seq_write_begin(&newfde->fde_seq);
3112 memcpy(newfde, oldfde, fde_change_size);
3113 oldfde->fde_file = NULL;
3116 seq_write_end(&newfde->fde_seq);
3120 FILEDESC_XUNLOCK(fdp);
3126 * This sysctl determines if we will allow a process to chroot(2) if it
3127 * has a directory open:
3128 * 0: disallowed for all processes.
3129 * 1: allowed for processes that were not already chroot(2)'ed.
3130 * 2: allowed for all processes.
3133 static int chroot_allow_open_directories = 1;
3135 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3136 &chroot_allow_open_directories, 0,
3137 "Allow a process to chroot(2) if it has a directory open");
3140 * Helper function for raised chroot(2) security function: Refuse if
3141 * any filedescriptors are open directories.
3144 chroot_refuse_vdir_fds(struct filedesc *fdp)
3150 FILEDESC_LOCK_ASSERT(fdp);
3152 for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3153 fp = fget_locked(fdp, fd);
3156 if (fp->f_type == DTYPE_VNODE) {
3158 if (vp->v_type == VDIR)
3166 * Common routine for kern_chroot() and jail_attach(). The caller is
3167 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3168 * authorize this operation.
3171 pwd_chroot(struct thread *td, struct vnode *vp)
3173 struct filedesc *fdp;
3174 struct vnode *oldvp;
3177 fdp = td->td_proc->p_fd;
3178 FILEDESC_XLOCK(fdp);
3179 if (chroot_allow_open_directories == 0 ||
3180 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3181 error = chroot_refuse_vdir_fds(fdp);
3183 FILEDESC_XUNLOCK(fdp);
3187 oldvp = fdp->fd_rdir;
3190 if (fdp->fd_jdir == NULL) {
3194 FILEDESC_XUNLOCK(fdp);
3200 pwd_chdir(struct thread *td, struct vnode *vp)
3202 struct filedesc *fdp;
3203 struct vnode *oldvp;
3205 fdp = td->td_proc->p_fd;
3206 FILEDESC_XLOCK(fdp);
3207 VNASSERT(vp->v_usecount > 0, vp,
3208 ("chdir to a vnode with zero usecount"));
3209 oldvp = fdp->fd_cdir;
3211 FILEDESC_XUNLOCK(fdp);
3216 * Scan all active processes and prisons to see if any of them have a current
3217 * or root directory of `olddp'. If so, replace them with the new mount point.
3220 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3222 struct filedesc *fdp;
3227 if (vrefcnt(olddp) == 1)
3230 sx_slock(&allproc_lock);
3231 FOREACH_PROC_IN_SYSTEM(p) {
3237 FILEDESC_XLOCK(fdp);
3238 if (fdp->fd_cdir == olddp) {
3240 fdp->fd_cdir = newdp;
3243 if (fdp->fd_rdir == olddp) {
3245 fdp->fd_rdir = newdp;
3248 if (fdp->fd_jdir == olddp) {
3250 fdp->fd_jdir = newdp;
3253 FILEDESC_XUNLOCK(fdp);
3256 sx_sunlock(&allproc_lock);
3257 if (rootvnode == olddp) {
3262 mtx_lock(&prison0.pr_mtx);
3263 if (prison0.pr_root == olddp) {
3265 prison0.pr_root = newdp;
3268 mtx_unlock(&prison0.pr_mtx);
3269 sx_slock(&allprison_lock);
3270 TAILQ_FOREACH(pr, &allprison, pr_list) {
3271 mtx_lock(&pr->pr_mtx);
3272 if (pr->pr_root == olddp) {
3274 pr->pr_root = newdp;
3277 mtx_unlock(&pr->pr_mtx);
3279 sx_sunlock(&allprison_lock);
3284 struct filedesc_to_leader *
3285 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3287 struct filedesc_to_leader *fdtol;
3289 fdtol = malloc(sizeof(struct filedesc_to_leader),
3290 M_FILEDESC_TO_LEADER, M_WAITOK);
3291 fdtol->fdl_refcount = 1;
3292 fdtol->fdl_holdcount = 0;
3293 fdtol->fdl_wakeup = 0;
3294 fdtol->fdl_leader = leader;
3296 FILEDESC_XLOCK(fdp);
3297 fdtol->fdl_next = old->fdl_next;
3298 fdtol->fdl_prev = old;
3299 old->fdl_next = fdtol;
3300 fdtol->fdl_next->fdl_prev = fdtol;
3301 FILEDESC_XUNLOCK(fdp);
3303 fdtol->fdl_next = fdtol;
3304 fdtol->fdl_prev = fdtol;
3310 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3312 struct filedesc *fdp;
3313 int i, count, slots;
3315 if (*(int *)arg1 != 0)
3318 fdp = curproc->p_fd;
3320 FILEDESC_SLOCK(fdp);
3321 slots = NDSLOTS(fdp->fd_lastfile + 1);
3322 for (i = 0; i < slots; i++)
3323 count += bitcountl(fdp->fd_map[i]);
3324 FILEDESC_SUNLOCK(fdp);
3326 return (SYSCTL_OUT(req, &count, sizeof(count)));
3329 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3330 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3331 "Number of open file descriptors");
3334 * Get file structures globally.
3337 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3340 struct filedesc *fdp;
3345 error = sysctl_wire_old_buffer(req, 0);
3348 if (req->oldptr == NULL) {
3350 sx_slock(&allproc_lock);
3351 FOREACH_PROC_IN_SYSTEM(p) {
3353 if (p->p_state == PRS_NEW) {
3361 /* overestimates sparse tables. */
3362 if (fdp->fd_lastfile > 0)
3363 n += fdp->fd_lastfile;
3366 sx_sunlock(&allproc_lock);
3367 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3370 bzero(&xf, sizeof(xf));
3371 xf.xf_size = sizeof(xf);
3372 sx_slock(&allproc_lock);
3373 FOREACH_PROC_IN_SYSTEM(p) {
3375 if (p->p_state == PRS_NEW) {
3379 if (p_cansee(req->td, p) != 0) {
3383 xf.xf_pid = p->p_pid;
3384 xf.xf_uid = p->p_ucred->cr_uid;
3389 FILEDESC_SLOCK(fdp);
3390 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3391 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3394 xf.xf_file = (uintptr_t)fp;
3395 xf.xf_data = (uintptr_t)fp->f_data;
3396 xf.xf_vnode = (uintptr_t)fp->f_vnode;
3397 xf.xf_type = (uintptr_t)fp->f_type;
3398 xf.xf_count = fp->f_count;
3400 xf.xf_offset = foffset_get(fp);
3401 xf.xf_flag = fp->f_flag;
3402 error = SYSCTL_OUT(req, &xf, sizeof(xf));
3406 FILEDESC_SUNLOCK(fdp);
3411 sx_sunlock(&allproc_lock);
3415 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3416 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3418 #ifdef KINFO_FILE_SIZE
3419 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3423 xlate_fflags(int fflags)
3425 static const struct {
3428 } fflags_table[] = {
3429 { FAPPEND, KF_FLAG_APPEND },
3430 { FASYNC, KF_FLAG_ASYNC },
3431 { FFSYNC, KF_FLAG_FSYNC },
3432 { FHASLOCK, KF_FLAG_HASLOCK },
3433 { FNONBLOCK, KF_FLAG_NONBLOCK },
3434 { FREAD, KF_FLAG_READ },
3435 { FWRITE, KF_FLAG_WRITE },
3436 { O_CREAT, KF_FLAG_CREAT },
3437 { O_DIRECT, KF_FLAG_DIRECT },
3438 { O_EXCL, KF_FLAG_EXCL },
3439 { O_EXEC, KF_FLAG_EXEC },
3440 { O_EXLOCK, KF_FLAG_EXLOCK },
3441 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3442 { O_SHLOCK, KF_FLAG_SHLOCK },
3443 { O_TRUNC, KF_FLAG_TRUNC }
3449 for (i = 0; i < nitems(fflags_table); i++)
3450 if (fflags & fflags_table[i].fflag)
3451 kflags |= fflags_table[i].kf_fflag;
3455 /* Trim unused data from kf_path by truncating the structure size. */
3457 pack_kinfo(struct kinfo_file *kif)
3460 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3461 strlen(kif->kf_path) + 1;
3462 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3466 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3467 struct kinfo_file *kif, struct filedesc *fdp, int flags)
3471 bzero(kif, sizeof(*kif));
3473 /* Set a default type to allow for empty fill_kinfo() methods. */
3474 kif->kf_type = KF_TYPE_UNKNOWN;
3475 kif->kf_flags = xlate_fflags(fp->f_flag);
3476 if (rightsp != NULL)
3477 kif->kf_cap_rights = *rightsp;
3479 cap_rights_init(&kif->kf_cap_rights);
3481 kif->kf_ref_count = fp->f_count;
3482 kif->kf_offset = foffset_get(fp);
3485 * This may drop the filedesc lock, so the 'fp' cannot be
3486 * accessed after this call.
3488 error = fo_fill_kinfo(fp, kif, fdp);
3490 kif->kf_status |= KF_ATTR_VALID;
3491 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3494 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3498 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3499 struct kinfo_file *kif, int flags)
3503 bzero(kif, sizeof(*kif));
3505 kif->kf_type = KF_TYPE_VNODE;
3506 error = vn_fill_kinfo_vnode(vp, kif);
3508 kif->kf_status |= KF_ATTR_VALID;
3509 kif->kf_flags = xlate_fflags(fflags);
3510 cap_rights_init(&kif->kf_cap_rights);
3512 kif->kf_ref_count = -1;
3513 kif->kf_offset = -1;
3514 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3517 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3521 struct export_fd_buf {
3522 struct filedesc *fdp;
3525 struct kinfo_file kif;
3530 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3532 struct kinfo_file *kif;
3535 if (efbuf->remainder != -1) {
3536 if (efbuf->remainder < kif->kf_structsize) {
3537 /* Terminate export. */
3538 efbuf->remainder = 0;
3541 efbuf->remainder -= kif->kf_structsize;
3543 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3547 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3548 struct export_fd_buf *efbuf)
3552 if (efbuf->remainder == 0)
3554 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3556 FILEDESC_SUNLOCK(efbuf->fdp);
3557 error = export_kinfo_to_sb(efbuf);
3558 FILEDESC_SLOCK(efbuf->fdp);
3563 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3564 struct export_fd_buf *efbuf)
3568 if (efbuf->remainder == 0)
3570 if (efbuf->fdp != NULL)
3571 FILEDESC_SUNLOCK(efbuf->fdp);
3572 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3573 error = export_kinfo_to_sb(efbuf);
3574 if (efbuf->fdp != NULL)
3575 FILEDESC_SLOCK(efbuf->fdp);
3580 * Store a process file descriptor information to sbuf.
3582 * Takes a locked proc as argument, and returns with the proc unlocked.
3585 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
3589 struct filedesc *fdp;
3590 struct export_fd_buf *efbuf;
3591 struct vnode *cttyvp, *textvp, *tracevp;
3593 cap_rights_t rights;
3595 PROC_LOCK_ASSERT(p, MA_OWNED);
3598 tracevp = p->p_tracevp;
3599 if (tracevp != NULL)
3602 textvp = p->p_textvp;
3605 /* Controlling tty. */
3607 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3608 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3614 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3617 efbuf->remainder = maxlen;
3618 efbuf->flags = flags;
3619 if (tracevp != NULL)
3620 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3623 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3625 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3631 FILEDESC_SLOCK(fdp);
3632 /* working directory */
3633 if (fdp->fd_cdir != NULL) {
3634 vrefact(fdp->fd_cdir);
3635 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3637 /* root directory */
3638 if (fdp->fd_rdir != NULL) {
3639 vrefact(fdp->fd_rdir);
3640 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3642 /* jail directory */
3643 if (fdp->fd_jdir != NULL) {
3644 vrefact(fdp->fd_jdir);
3645 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3647 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3648 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3651 rights = *cap_rights(fdp, i);
3652 #else /* !CAPABILITIES */
3653 rights = cap_no_rights;
3656 * Create sysctl entry. It is OK to drop the filedesc
3657 * lock inside of export_file_to_sb() as we will
3658 * re-validate and re-evaluate its properties when the
3661 error = export_file_to_sb(fp, i, &rights, efbuf);
3662 if (error != 0 || efbuf->remainder == 0)
3665 FILEDESC_SUNLOCK(fdp);
3668 free(efbuf, M_TEMP);
3672 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
3675 * Get per-process file descriptors for use by procstat(1), et al.
3678 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3683 int error, error2, *name;
3687 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3688 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3689 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3694 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3695 error = kern_proc_filedesc_out(p, &sb, maxlen,
3696 KERN_FILEDESC_PACK_KINFO);
3697 error2 = sbuf_finish(&sb);
3699 return (error != 0 ? error : error2);
3702 #ifdef COMPAT_FREEBSD7
3703 #ifdef KINFO_OFILE_SIZE
3704 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3708 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3711 okif->kf_structsize = sizeof(*okif);
3712 okif->kf_type = kif->kf_type;
3713 okif->kf_fd = kif->kf_fd;
3714 okif->kf_ref_count = kif->kf_ref_count;
3715 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3716 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3717 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3718 okif->kf_offset = kif->kf_offset;
3719 if (kif->kf_type == KF_TYPE_VNODE)
3720 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
3722 okif->kf_vnode_type = KF_VTYPE_VNON;
3723 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3724 if (kif->kf_type == KF_TYPE_SOCKET) {
3725 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
3726 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
3727 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
3728 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
3729 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
3731 okif->kf_sa_local.ss_family = AF_UNSPEC;
3732 okif->kf_sa_peer.ss_family = AF_UNSPEC;
3737 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3738 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3743 FILEDESC_SUNLOCK(fdp);
3744 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3745 kinfo_to_okinfo(kif, okif);
3746 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3747 FILEDESC_SLOCK(fdp);
3752 * Get per-process file descriptors for use by procstat(1), et al.
3755 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3757 struct kinfo_ofile *okif;
3758 struct kinfo_file *kif;
3759 struct filedesc *fdp;
3760 int error, i, *name;
3765 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3772 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3773 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3774 FILEDESC_SLOCK(fdp);
3775 if (fdp->fd_cdir != NULL)
3776 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3778 if (fdp->fd_rdir != NULL)
3779 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3781 if (fdp->fd_jdir != NULL)
3782 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3784 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3785 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3787 export_file_to_kinfo(fp, i, NULL, kif, fdp,
3788 KERN_FILEDESC_PACK_KINFO);
3789 FILEDESC_SUNLOCK(fdp);
3790 kinfo_to_okinfo(kif, okif);
3791 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3792 FILEDESC_SLOCK(fdp);
3796 FILEDESC_SUNLOCK(fdp);
3803 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3804 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3805 "Process ofiledesc entries");
3806 #endif /* COMPAT_FREEBSD7 */
3809 vntype_to_kinfo(int vtype)
3814 } vtypes_table[] = {
3815 { VBAD, KF_VTYPE_VBAD },
3816 { VBLK, KF_VTYPE_VBLK },
3817 { VCHR, KF_VTYPE_VCHR },
3818 { VDIR, KF_VTYPE_VDIR },
3819 { VFIFO, KF_VTYPE_VFIFO },
3820 { VLNK, KF_VTYPE_VLNK },
3821 { VNON, KF_VTYPE_VNON },
3822 { VREG, KF_VTYPE_VREG },
3823 { VSOCK, KF_VTYPE_VSOCK }
3828 * Perform vtype translation.
3830 for (i = 0; i < nitems(vtypes_table); i++)
3831 if (vtypes_table[i].vtype == vtype)
3832 return (vtypes_table[i].kf_vtype);
3834 return (KF_VTYPE_UNKNOWN);
3837 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3838 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3839 "Process filedesc entries");
3842 * Store a process current working directory information to sbuf.
3844 * Takes a locked proc as argument, and returns with the proc unlocked.
3847 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
3849 struct filedesc *fdp;
3850 struct export_fd_buf *efbuf;
3853 PROC_LOCK_ASSERT(p, MA_OWNED);
3860 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3863 efbuf->remainder = maxlen;
3865 FILEDESC_SLOCK(fdp);
3866 if (fdp->fd_cdir == NULL)
3869 vrefact(fdp->fd_cdir);
3870 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3873 FILEDESC_SUNLOCK(fdp);
3875 free(efbuf, M_TEMP);
3880 * Get per-process current working directory.
3883 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3888 int error, error2, *name;
3892 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3893 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3894 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3899 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3900 error = kern_proc_cwd_out(p, &sb, maxlen);
3901 error2 = sbuf_finish(&sb);
3903 return (error != 0 ? error : error2);
3906 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3907 sysctl_kern_proc_cwd, "Process current working directory");
3911 * For the purposes of debugging, generate a human-readable string for the
3915 file_type_to_name(short type)
3943 case DTYPE_PROCDESC:
3945 case DTYPE_LINUXEFD:
3947 case DTYPE_LINUXTFD:
3955 * For the purposes of debugging, identify a process (if any, perhaps one of
3956 * many) that references the passed file in its file descriptor array. Return
3959 static struct proc *
3960 file_to_first_proc(struct file *fp)
3962 struct filedesc *fdp;
3966 FOREACH_PROC_IN_SYSTEM(p) {
3967 if (p->p_state == PRS_NEW)
3972 for (n = 0; n <= fdp->fd_lastfile; n++) {
3973 if (fp == fdp->fd_ofiles[n].fde_file)
3981 db_print_file(struct file *fp, int header)
3983 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
3987 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
3988 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
3989 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
3991 p = file_to_first_proc(fp);
3992 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
3993 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
3994 fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode,
3995 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4000 DB_SHOW_COMMAND(file, db_show_file)
4005 db_printf("usage: show file <addr>\n");
4008 fp = (struct file *)addr;
4009 db_print_file(fp, 1);
4012 DB_SHOW_COMMAND(files, db_show_files)
4014 struct filedesc *fdp;
4021 FOREACH_PROC_IN_SYSTEM(p) {
4022 if (p->p_state == PRS_NEW)
4024 if ((fdp = p->p_fd) == NULL)
4026 for (n = 0; n <= fdp->fd_lastfile; ++n) {
4027 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4029 db_print_file(fp, header);
4036 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4037 &maxfilesperproc, 0, "Maximum files allowed open per process");
4039 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4040 &maxfiles, 0, "Maximum number of files");
4042 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4043 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
4047 filelistinit(void *dummy)
4050 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4051 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4052 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4053 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4054 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4056 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4058 /*-------------------------------------------------------------------*/
4061 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4062 int flags, struct thread *td)
4069 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4077 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4085 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4093 badfo_kqfilter(struct file *fp, struct knote *kn)
4100 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4108 badfo_close(struct file *fp, struct thread *td)
4115 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4123 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4131 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4132 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4140 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4146 struct fileops badfileops = {
4147 .fo_read = badfo_readwrite,
4148 .fo_write = badfo_readwrite,
4149 .fo_truncate = badfo_truncate,
4150 .fo_ioctl = badfo_ioctl,
4151 .fo_poll = badfo_poll,
4152 .fo_kqfilter = badfo_kqfilter,
4153 .fo_stat = badfo_stat,
4154 .fo_close = badfo_close,
4155 .fo_chmod = badfo_chmod,
4156 .fo_chown = badfo_chown,
4157 .fo_sendfile = badfo_sendfile,
4158 .fo_fill_kinfo = badfo_fill_kinfo,
4162 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4163 int flags, struct thread *td)
4166 return (EOPNOTSUPP);
4170 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4178 invfo_ioctl(struct file *fp, u_long com, void *data,
4179 struct ucred *active_cred, struct thread *td)
4186 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4190 return (poll_no_poll(events));
4194 invfo_kqfilter(struct file *fp, struct knote *kn)
4201 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4209 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4217 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4218 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4225 /*-------------------------------------------------------------------*/
4228 * File Descriptor pseudo-device driver (/dev/fd/).
4230 * Opening minor device N dup()s the file (if any) connected to file
4231 * descriptor N belonging to the calling process. Note that this driver
4232 * consists of only the ``open()'' routine, because all subsequent
4233 * references to this file will be direct to the other driver.
4235 * XXX: we could give this one a cloning event handler if necessary.
4240 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4244 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4245 * the file descriptor being sought for duplication. The error
4246 * return ensures that the vnode for this device will be released
4247 * by vn_open. Open will detect this special error and take the
4248 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4249 * will simply report the error.
4251 td->td_dupfd = dev2unit(dev);
4255 static struct cdevsw fildesc_cdevsw = {
4256 .d_version = D_VERSION,
4262 fildesc_drvinit(void *unused)
4266 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4267 UID_ROOT, GID_WHEEL, 0666, "fd/0");
4268 make_dev_alias(dev, "stdin");
4269 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4270 UID_ROOT, GID_WHEEL, 0666, "fd/1");
4271 make_dev_alias(dev, "stdout");
4272 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4273 UID_ROOT, GID_WHEEL, 0666, "fd/2");
4274 make_dev_alias(dev, "stderr");
4277 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);