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)
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)
351 td->td_retval[0] = getmaxfd(td);
353 PROC_LOCK(td->td_proc);
354 lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
355 PROC_UNLOCK(td->td_proc);
356 if (lim < td->td_retval[0])
357 td->td_retval[0] = lim;
363 * Duplicate a file descriptor to a particular value.
365 * Note: keep in mind that a potential race condition exists when closing
366 * descriptors from a shared descriptor table (via rfork).
368 #ifndef _SYS_SYSPROTO_H_
376 sys_dup2(struct thread *td, struct dup2_args *uap)
379 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
383 * Duplicate a file descriptor.
385 #ifndef _SYS_SYSPROTO_H_
392 sys_dup(struct thread *td, struct dup_args *uap)
395 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
399 * The file control system call.
401 #ifndef _SYS_SYSPROTO_H_
410 sys_fcntl(struct thread *td, struct fcntl_args *uap)
413 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
417 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
431 * Convert old flock structure to new.
433 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
434 fl.l_start = ofl.l_start;
435 fl.l_len = ofl.l_len;
436 fl.l_pid = ofl.l_pid;
437 fl.l_type = ofl.l_type;
438 fl.l_whence = ofl.l_whence;
452 arg1 = (intptr_t)&fl;
458 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
459 arg1 = (intptr_t)&fl;
467 error = kern_fcntl(td, fd, newcmd, arg1);
470 if (cmd == F_OGETLK) {
471 ofl.l_start = fl.l_start;
472 ofl.l_len = fl.l_len;
473 ofl.l_pid = fl.l_pid;
474 ofl.l_type = fl.l_type;
475 ofl.l_whence = fl.l_whence;
476 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
477 } else if (cmd == F_GETLK) {
478 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
484 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
486 struct filedesc *fdp;
488 struct file *fp, *fp2;
489 struct filedescent *fde;
506 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
509 case F_DUPFD_CLOEXEC:
511 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
516 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
519 case F_DUP2FD_CLOEXEC:
521 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
527 fde = fdeget_locked(fdp, fd);
530 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
533 FILEDESC_SUNLOCK(fdp);
539 fde = fdeget_locked(fdp, fd);
541 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
542 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
545 FILEDESC_XUNLOCK(fdp);
549 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
552 td->td_retval[0] = OFLAGS(fp->f_flag);
557 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
561 tmp = flg = fp->f_flag;
563 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
564 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
565 tmp = fp->f_flag & FNONBLOCK;
566 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
571 tmp = fp->f_flag & FASYNC;
572 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
577 atomic_clear_int(&fp->f_flag, FNONBLOCK);
579 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
584 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
587 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
589 td->td_retval[0] = tmp;
594 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
598 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
603 error = priv_check(td, PRIV_NFS_LOCKD);
611 /* FALLTHROUGH F_SETLK */
615 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL);
618 if (fp->f_type != DTYPE_VNODE) {
624 flp = (struct flock *)arg;
625 if (flp->l_whence == SEEK_CUR) {
626 foffset = foffset_get(fp);
629 foffset > OFF_MAX - flp->l_start)) {
634 flp->l_start += foffset;
638 switch (flp->l_type) {
640 if ((fp->f_flag & FREAD) == 0) {
644 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
645 PROC_LOCK(p->p_leader);
646 p->p_leader->p_flag |= P_ADVLOCK;
647 PROC_UNLOCK(p->p_leader);
649 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
653 if ((fp->f_flag & FWRITE) == 0) {
657 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
658 PROC_LOCK(p->p_leader);
659 p->p_leader->p_flag |= P_ADVLOCK;
660 PROC_UNLOCK(p->p_leader);
662 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
666 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
671 * Temporary api for testing remote lock
674 if (flg != F_REMOTE) {
678 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
679 F_UNLCKSYS, flp, flg);
685 if (error != 0 || flp->l_type == F_UNLCK ||
686 flp->l_type == F_UNLCKSYS) {
692 * Check for a race with close.
694 * The vnode is now advisory locked (or unlocked, but this case
695 * is not really important) as the caller requested.
696 * We had to drop the filedesc lock, so we need to recheck if
697 * the descriptor is still valid, because if it was closed
698 * in the meantime we need to remove advisory lock from the
699 * vnode - close on any descriptor leading to an advisory
700 * locked vnode, removes that lock.
701 * We will return 0 on purpose in that case, as the result of
702 * successful advisory lock might have been externally visible
703 * already. This is fine - effectively we pretend to the caller
704 * that the closing thread was a bit slower and that the
705 * advisory lock succeeded before the close.
707 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2, NULL);
713 flp->l_whence = SEEK_SET;
716 flp->l_type = F_UNLCK;
717 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
718 F_UNLCK, flp, F_POSIX);
725 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL);
728 if (fp->f_type != DTYPE_VNODE) {
733 flp = (struct flock *)arg;
734 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
735 flp->l_type != F_UNLCK) {
740 if (flp->l_whence == SEEK_CUR) {
741 foffset = foffset_get(fp);
742 if ((flp->l_start > 0 &&
743 foffset > OFF_MAX - flp->l_start) ||
745 foffset < OFF_MIN - flp->l_start)) {
750 flp->l_start += foffset;
753 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
759 arg = arg ? 128 * 1024: 0;
762 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp, NULL);
765 if (fp->f_type != DTYPE_VNODE) {
772 * Exclusive lock synchronizes against f_seqcount reads and
773 * writes in sequential_heuristic().
775 error = vn_lock(vp, LK_EXCLUSIVE);
781 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
782 fp->f_seqcount = (arg + bsize - 1) / bsize;
783 atomic_set_int(&fp->f_flag, FRDAHEAD);
785 atomic_clear_int(&fp->f_flag, FRDAHEAD);
799 getmaxfd(struct thread *td)
802 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
806 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
809 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
811 struct filedesc *fdp;
812 struct filedescent *oldfde, *newfde;
815 u_long *oioctls, *nioctls;
822 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
823 MPASS(mode < FDDUP_LASTMODE);
826 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
829 * Verify we have a valid descriptor to dup from and possibly to
830 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
831 * return EINVAL when the new descriptor is out of bounds.
836 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
837 maxfd = getmaxfd(td);
839 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
843 if (fget_locked(fdp, old) == NULL)
845 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
846 td->td_retval[0] = new;
847 if (flags & FDDUP_FLAG_CLOEXEC)
848 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
854 * If the caller specified a file descriptor, make sure the file
855 * table is large enough to hold it, and grab it. Otherwise, just
856 * allocate a new descriptor the usual way.
861 if ((error = fdalloc(td, new, &new)) != 0)
864 case FDDUP_MUSTREPLACE:
865 /* Target file descriptor must exist. */
866 if (fget_locked(fdp, new) == NULL)
870 if (new >= fdp->fd_nfiles) {
872 * The resource limits are here instead of e.g.
873 * fdalloc(), because the file descriptor table may be
874 * shared between processes, so we can't really use
875 * racct_add()/racct_sub(). Instead of counting the
876 * number of actually allocated descriptors, just put
877 * the limit on the size of the file descriptor table.
880 if (RACCT_ENABLED()) {
881 error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
888 fdgrowtable_exp(fdp, new + 1);
890 if (!fdisused(fdp, new))
894 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
897 KASSERT(old != new, ("new fd is same as old"));
899 oldfde = &fdp->fd_ofiles[old];
900 fhold(oldfde->fde_file);
901 newfde = &fdp->fd_ofiles[new];
902 delfp = newfde->fde_file;
904 oioctls = filecaps_free_prep(&newfde->fde_caps);
905 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
908 * Duplicate the source descriptor.
911 seq_write_begin(&newfde->fde_seq);
913 memcpy(newfde, oldfde, fde_change_size);
914 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
916 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
917 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
919 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
921 seq_write_end(&newfde->fde_seq);
923 td->td_retval[0] = new;
928 (void) closefp(fdp, new, delfp, td, 1);
929 FILEDESC_UNLOCK_ASSERT(fdp);
932 FILEDESC_XUNLOCK(fdp);
935 filecaps_free_finish(oioctls);
940 * If sigio is on the list associated with a process or process group,
941 * disable signalling from the device, remove sigio from the list and
945 funsetown(struct sigio **sigiop)
957 *(sigio->sio_myref) = NULL;
958 if ((sigio)->sio_pgid < 0) {
959 struct pgrp *pg = (sigio)->sio_pgrp;
961 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
965 struct proc *p = (sigio)->sio_proc;
967 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
972 crfree(sigio->sio_ucred);
973 free(sigio, M_SIGIO);
977 * Free a list of sigio structures.
978 * We only need to lock the SIGIO_LOCK because we have made ourselves
979 * inaccessible to callers of fsetown and therefore do not need to lock
980 * the proc or pgrp struct for the list manipulation.
983 funsetownlst(struct sigiolst *sigiolst)
989 sigio = SLIST_FIRST(sigiolst);
996 * Every entry of the list should belong
997 * to a single proc or pgrp.
999 if (sigio->sio_pgid < 0) {
1000 pg = sigio->sio_pgrp;
1001 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
1002 } else /* if (sigio->sio_pgid > 0) */ {
1003 p = sigio->sio_proc;
1004 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1008 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1009 *(sigio->sio_myref) = NULL;
1011 KASSERT(sigio->sio_pgid < 0,
1012 ("Proc sigio in pgrp sigio list"));
1013 KASSERT(sigio->sio_pgrp == pg,
1014 ("Bogus pgrp in sigio list"));
1016 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1019 } else /* if (p != NULL) */ {
1020 KASSERT(sigio->sio_pgid > 0,
1021 ("Pgrp sigio in proc sigio list"));
1022 KASSERT(sigio->sio_proc == p,
1023 ("Bogus proc in sigio list"));
1025 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1030 crfree(sigio->sio_ucred);
1031 free(sigio, M_SIGIO);
1038 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1040 * After permission checking, add a sigio structure to the sigio list for
1041 * the process or process group.
1044 fsetown(pid_t pgid, struct sigio **sigiop)
1048 struct sigio *sigio;
1058 /* Allocate and fill in the new sigio out of locks. */
1059 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1060 sigio->sio_pgid = pgid;
1061 sigio->sio_ucred = crhold(curthread->td_ucred);
1062 sigio->sio_myref = sigiop;
1064 sx_slock(&proctree_lock);
1073 * Policy - Don't allow a process to FSETOWN a process
1074 * in another session.
1076 * Remove this test to allow maximum flexibility or
1077 * restrict FSETOWN to the current process or process
1078 * group for maximum safety.
1081 if (proc->p_session != curthread->td_proc->p_session) {
1087 } else /* if (pgid < 0) */ {
1088 pgrp = pgfind(-pgid);
1096 * Policy - Don't allow a process to FSETOWN a process
1097 * in another session.
1099 * Remove this test to allow maximum flexibility or
1100 * restrict FSETOWN to the current process or process
1101 * group for maximum safety.
1103 if (pgrp->pg_session != curthread->td_proc->p_session) {
1114 * Since funsetownlst() is called without the proctree
1115 * locked, we need to check for P_WEXIT.
1116 * XXX: is ESRCH correct?
1118 if ((proc->p_flag & P_WEXIT) != 0) {
1123 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1124 sigio->sio_proc = proc;
1128 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1129 sigio->sio_pgrp = pgrp;
1132 sx_sunlock(&proctree_lock);
1139 sx_sunlock(&proctree_lock);
1140 crfree(sigio->sio_ucred);
1141 free(sigio, M_SIGIO);
1146 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1149 fgetown(struct sigio **sigiop)
1154 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1160 * Function drops the filedesc lock on return.
1163 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1168 FILEDESC_XLOCK_ASSERT(fdp);
1171 if (td->td_proc->p_fdtol != NULL) {
1173 * Ask fdfree() to sleep to ensure that all relevant
1174 * process leaders can be traversed in closef().
1176 fdp->fd_holdleaderscount++;
1183 * We now hold the fp reference that used to be owned by the
1184 * descriptor array. We have to unlock the FILEDESC *AFTER*
1185 * knote_fdclose to prevent a race of the fd getting opened, a knote
1186 * added, and deleteing a knote for the new fd.
1188 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1189 knote_fdclose(td, fd);
1192 * We need to notify mqueue if the object is of type mqueue.
1194 if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1195 mq_fdclose(td, fd, fp);
1196 FILEDESC_XUNLOCK(fdp);
1198 error = closef(fp, td);
1200 FILEDESC_XLOCK(fdp);
1201 fdp->fd_holdleaderscount--;
1202 if (fdp->fd_holdleaderscount == 0 &&
1203 fdp->fd_holdleaderswakeup != 0) {
1204 fdp->fd_holdleaderswakeup = 0;
1205 wakeup(&fdp->fd_holdleaderscount);
1207 FILEDESC_XUNLOCK(fdp);
1213 * Close a file descriptor.
1215 #ifndef _SYS_SYSPROTO_H_
1222 sys_close(struct thread *td, struct close_args *uap)
1225 return (kern_close(td, uap->fd));
1229 kern_close(struct thread *td, int fd)
1231 struct filedesc *fdp;
1234 fdp = td->td_proc->p_fd;
1236 AUDIT_SYSCLOSE(td, fd);
1238 FILEDESC_XLOCK(fdp);
1239 if ((fp = fget_locked(fdp, fd)) == NULL) {
1240 FILEDESC_XUNLOCK(fdp);
1245 /* closefp() drops the FILEDESC lock for us. */
1246 return (closefp(fdp, fd, fp, td, 1));
1250 * Close open file descriptors.
1252 #ifndef _SYS_SYSPROTO_H_
1253 struct closefrom_args {
1259 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1261 struct filedesc *fdp;
1264 fdp = td->td_proc->p_fd;
1265 AUDIT_ARG_FD(uap->lowfd);
1268 * Treat negative starting file descriptor values identical to
1269 * closefrom(0) which closes all files.
1273 FILEDESC_SLOCK(fdp);
1274 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1275 if (fdp->fd_ofiles[fd].fde_file != NULL) {
1276 FILEDESC_SUNLOCK(fdp);
1277 (void)kern_close(td, fd);
1278 FILEDESC_SLOCK(fdp);
1281 FILEDESC_SUNLOCK(fdp);
1285 #if defined(COMPAT_43)
1287 * Return status information about a file descriptor.
1289 #ifndef _SYS_SYSPROTO_H_
1290 struct ofstat_args {
1297 ofstat(struct thread *td, struct ofstat_args *uap)
1303 error = kern_fstat(td, uap->fd, &ub);
1306 error = copyout(&oub, uap->sb, sizeof(oub));
1310 #endif /* COMPAT_43 */
1312 #if defined(COMPAT_FREEBSD11)
1314 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1317 struct freebsd11_stat osb;
1320 error = kern_fstat(td, uap->fd, &sb);
1323 error = freebsd11_cvtstat(&sb, &osb);
1325 error = copyout(&osb, uap->sb, sizeof(osb));
1328 #endif /* COMPAT_FREEBSD11 */
1331 * Return status information about a file descriptor.
1333 #ifndef _SYS_SYSPROTO_H_
1341 sys_fstat(struct thread *td, struct fstat_args *uap)
1346 error = kern_fstat(td, uap->fd, &ub);
1348 error = copyout(&ub, uap->sb, sizeof(ub));
1353 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1360 error = fget(td, fd, &cap_fstat_rights, &fp);
1364 AUDIT_ARG_FILE(td->td_proc, fp);
1366 error = fo_stat(fp, sbp, td->td_ucred, td);
1368 #ifdef __STAT_TIME_T_EXT
1370 sbp->st_atim_ext = 0;
1371 sbp->st_mtim_ext = 0;
1372 sbp->st_ctim_ext = 0;
1373 sbp->st_btim_ext = 0;
1377 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1383 #if defined(COMPAT_FREEBSD11)
1385 * Return status information about a file descriptor.
1387 #ifndef _SYS_SYSPROTO_H_
1388 struct freebsd11_nfstat_args {
1395 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1401 error = kern_fstat(td, uap->fd, &ub);
1403 freebsd11_cvtnstat(&ub, &nub);
1404 error = copyout(&nub, uap->sb, sizeof(nub));
1408 #endif /* COMPAT_FREEBSD11 */
1411 * Return pathconf information about a file descriptor.
1413 #ifndef _SYS_SYSPROTO_H_
1414 struct fpathconf_args {
1421 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1426 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1428 td->td_retval[0] = value;
1433 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1439 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1443 if (name == _PC_ASYNC_IO) {
1444 *valuep = _POSIX_ASYNCHRONOUS_IO;
1449 vn_lock(vp, LK_SHARED | LK_RETRY);
1450 error = VOP_PATHCONF(vp, name, valuep);
1452 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1453 if (name != _PC_PIPE_BUF) {
1468 * Copy filecaps structure allocating memory for ioctls array if needed.
1470 * The last parameter indicates whether the fdtable is locked. If it is not and
1471 * ioctls are encountered, copying fails and the caller must lock the table.
1473 * Note that if the table was not locked, the caller has to check the relevant
1474 * sequence counter to determine whether the operation was successful.
1477 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1481 if (src->fc_ioctls != NULL && !locked)
1483 memcpy(dst, src, sizeof(*src));
1484 if (src->fc_ioctls == NULL)
1487 KASSERT(src->fc_nioctls > 0,
1488 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1490 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1491 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1492 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1497 filecaps_copy_prep(const struct filecaps *src)
1502 if (__predict_true(src->fc_ioctls == NULL))
1505 KASSERT(src->fc_nioctls > 0,
1506 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1508 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1509 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1514 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1520 if (__predict_true(src->fc_ioctls == NULL)) {
1521 MPASS(ioctls == NULL);
1525 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1526 dst->fc_ioctls = ioctls;
1527 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1531 * Move filecaps structure to the new place and clear the old place.
1534 filecaps_move(struct filecaps *src, struct filecaps *dst)
1538 bzero(src, sizeof(*src));
1542 * Fill the given filecaps structure with full rights.
1545 filecaps_fill(struct filecaps *fcaps)
1548 CAP_ALL(&fcaps->fc_rights);
1549 fcaps->fc_ioctls = NULL;
1550 fcaps->fc_nioctls = -1;
1551 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1555 * Free memory allocated within filecaps structure.
1558 filecaps_free(struct filecaps *fcaps)
1561 free(fcaps->fc_ioctls, M_FILECAPS);
1562 bzero(fcaps, sizeof(*fcaps));
1566 filecaps_free_prep(struct filecaps *fcaps)
1570 ioctls = fcaps->fc_ioctls;
1571 bzero(fcaps, sizeof(*fcaps));
1576 filecaps_free_finish(u_long *ioctls)
1579 free(ioctls, M_FILECAPS);
1583 * Validate the given filecaps structure.
1586 filecaps_validate(const struct filecaps *fcaps, const char *func)
1589 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1590 ("%s: invalid rights", func));
1591 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1592 ("%s: invalid fcntls", func));
1593 KASSERT(fcaps->fc_fcntls == 0 ||
1594 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1595 ("%s: fcntls without CAP_FCNTL", func));
1596 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1597 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1598 ("%s: invalid ioctls", func));
1599 KASSERT(fcaps->fc_nioctls == 0 ||
1600 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1601 ("%s: ioctls without CAP_IOCTL", func));
1605 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1609 FILEDESC_XLOCK_ASSERT(fdp);
1611 nfd1 = fdp->fd_nfiles * 2;
1614 fdgrowtable(fdp, nfd1);
1618 * Grow the file table to accommodate (at least) nfd descriptors.
1621 fdgrowtable(struct filedesc *fdp, int nfd)
1623 struct filedesc0 *fdp0;
1624 struct freetable *ft;
1625 struct fdescenttbl *ntable;
1626 struct fdescenttbl *otable;
1627 int nnfiles, onfiles;
1628 NDSLOTTYPE *nmap, *omap;
1631 * If lastfile is -1 this struct filedesc was just allocated and we are
1632 * growing it to accommodate for the one we are going to copy from. There
1633 * is no need to have a lock on this one as it's not visible to anyone.
1635 if (fdp->fd_lastfile != -1)
1636 FILEDESC_XLOCK_ASSERT(fdp);
1638 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1640 /* save old values */
1641 onfiles = fdp->fd_nfiles;
1642 otable = fdp->fd_files;
1645 /* compute the size of the new table */
1646 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1647 if (nnfiles <= onfiles)
1648 /* the table is already large enough */
1652 * Allocate a new table. We need enough space for the number of
1653 * entries, file entries themselves and the struct freetable we will use
1654 * when we decommission the table and place it on the freelist.
1655 * We place the struct freetable in the middle so we don't have
1656 * to worry about padding.
1658 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1659 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1660 sizeof(struct freetable),
1661 M_FILEDESC, M_ZERO | M_WAITOK);
1662 /* copy the old data */
1663 ntable->fdt_nfiles = nnfiles;
1664 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1665 onfiles * sizeof(ntable->fdt_ofiles[0]));
1668 * Allocate a new map only if the old is not large enough. It will
1669 * grow at a slower rate than the table as it can map more
1670 * entries than the table can hold.
1672 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1673 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1675 /* copy over the old data and update the pointer */
1676 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1681 * Make sure that ntable is correctly initialized before we replace
1682 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1685 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1688 * Do not free the old file table, as some threads may still
1689 * reference entries within it. Instead, place it on a freelist
1690 * which will be processed when the struct filedesc is released.
1692 * Note that if onfiles == NDFILE, we're dealing with the original
1693 * static allocation contained within (struct filedesc0 *)fdp,
1694 * which must not be freed.
1696 if (onfiles > NDFILE) {
1697 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1698 fdp0 = (struct filedesc0 *)fdp;
1699 ft->ft_table = otable;
1700 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1703 * The map does not have the same possibility of threads still
1704 * holding references to it. So always free it as long as it
1705 * does not reference the original static allocation.
1707 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1708 free(omap, M_FILEDESC);
1712 * Allocate a file descriptor for the process.
1715 fdalloc(struct thread *td, int minfd, int *result)
1717 struct proc *p = td->td_proc;
1718 struct filedesc *fdp = p->p_fd;
1719 int fd, maxfd, allocfd;
1724 FILEDESC_XLOCK_ASSERT(fdp);
1726 if (fdp->fd_freefile > minfd)
1727 minfd = fdp->fd_freefile;
1729 maxfd = getmaxfd(td);
1732 * Search the bitmap for a free descriptor starting at minfd.
1733 * If none is found, grow the file table.
1735 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1738 if (fd >= fdp->fd_nfiles) {
1739 allocfd = min(fd * 2, maxfd);
1741 if (RACCT_ENABLED()) {
1742 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1748 * fd is already equal to first free descriptor >= minfd, so
1749 * we only need to grow the table and we are done.
1751 fdgrowtable_exp(fdp, allocfd);
1755 * Perform some sanity checks, then mark the file descriptor as
1756 * used and return it to the caller.
1758 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1759 ("invalid descriptor %d", fd));
1760 KASSERT(!fdisused(fdp, fd),
1761 ("fd_first_free() returned non-free descriptor"));
1762 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1763 ("file descriptor isn't free"));
1770 * Allocate n file descriptors for the process.
1773 fdallocn(struct thread *td, int minfd, int *fds, int n)
1775 struct proc *p = td->td_proc;
1776 struct filedesc *fdp = p->p_fd;
1779 FILEDESC_XLOCK_ASSERT(fdp);
1781 for (i = 0; i < n; i++)
1782 if (fdalloc(td, 0, &fds[i]) != 0)
1786 for (i--; i >= 0; i--)
1787 fdunused(fdp, fds[i]);
1795 * Create a new open file structure and allocate a file descriptor for the
1796 * process that refers to it. We add one reference to the file for the
1797 * descriptor table and one reference for resultfp. This is to prevent us
1798 * being preempted and the entry in the descriptor table closed after we
1799 * release the FILEDESC lock.
1802 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1803 struct filecaps *fcaps)
1808 error = falloc_noinstall(td, &fp);
1810 return (error); /* no reference held on error */
1812 error = finstall(td, fp, &fd, flags, fcaps);
1814 fdrop(fp, td); /* one reference (fp only) */
1818 if (resultfp != NULL)
1819 *resultfp = fp; /* copy out result */
1821 fdrop(fp, td); /* release local reference */
1823 if (resultfd != NULL)
1830 * Create a new open file structure without allocating a file descriptor.
1833 falloc_noinstall(struct thread *td, struct file **resultfp)
1836 int maxuserfiles = maxfiles - (maxfiles / 20);
1838 static struct timeval lastfail;
1841 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1843 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1844 if ((openfiles_new >= maxuserfiles &&
1845 priv_check(td, PRIV_MAXFILES) != 0) ||
1846 openfiles_new >= maxfiles) {
1847 atomic_subtract_int(&openfiles, 1);
1848 if (ppsratecheck(&lastfail, &curfail, 1)) {
1849 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1850 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1854 fp = uma_zalloc(file_zone, M_WAITOK);
1855 bzero(fp, sizeof(*fp));
1856 refcount_init(&fp->f_count, 1);
1857 fp->f_cred = crhold(td->td_ucred);
1858 fp->f_ops = &badfileops;
1864 * Install a file in a file descriptor table.
1867 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1868 struct filecaps *fcaps)
1870 struct filedescent *fde;
1874 filecaps_validate(fcaps, __func__);
1875 FILEDESC_XLOCK_ASSERT(fdp);
1877 fde = &fdp->fd_ofiles[fd];
1879 seq_write_begin(&fde->fde_seq);
1882 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1884 filecaps_move(fcaps, &fde->fde_caps);
1886 filecaps_fill(&fde->fde_caps);
1888 seq_write_end(&fde->fde_seq);
1893 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1894 struct filecaps *fcaps)
1896 struct filedesc *fdp = td->td_proc->p_fd;
1901 FILEDESC_XLOCK(fdp);
1902 if ((error = fdalloc(td, 0, fd))) {
1903 FILEDESC_XUNLOCK(fdp);
1907 _finstall(fdp, fp, *fd, flags, fcaps);
1908 FILEDESC_XUNLOCK(fdp);
1913 * Build a new filedesc structure from another.
1914 * Copy the current, root, and jail root vnode references.
1916 * If fdp is not NULL, return with it shared locked.
1919 fdinit(struct filedesc *fdp, bool prepfiles)
1921 struct filedesc0 *newfdp0;
1922 struct filedesc *newfdp;
1924 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1925 newfdp = &newfdp0->fd_fd;
1927 /* Create the file descriptor table. */
1928 FILEDESC_LOCK_INIT(newfdp);
1929 refcount_init(&newfdp->fd_refcnt, 1);
1930 refcount_init(&newfdp->fd_holdcnt, 1);
1931 newfdp->fd_cmask = CMASK;
1932 newfdp->fd_map = newfdp0->fd_dmap;
1933 newfdp->fd_lastfile = -1;
1934 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1935 newfdp->fd_files->fdt_nfiles = NDFILE;
1940 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1941 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1943 FILEDESC_SLOCK(fdp);
1944 newfdp->fd_cdir = fdp->fd_cdir;
1945 if (newfdp->fd_cdir)
1946 vrefact(newfdp->fd_cdir);
1947 newfdp->fd_rdir = fdp->fd_rdir;
1948 if (newfdp->fd_rdir)
1949 vrefact(newfdp->fd_rdir);
1950 newfdp->fd_jdir = fdp->fd_jdir;
1951 if (newfdp->fd_jdir)
1952 vrefact(newfdp->fd_jdir);
1955 FILEDESC_SUNLOCK(fdp);
1957 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1958 FILEDESC_SUNLOCK(fdp);
1959 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1960 FILEDESC_SLOCK(fdp);
1967 static struct filedesc *
1968 fdhold(struct proc *p)
1970 struct filedesc *fdp;
1972 PROC_LOCK_ASSERT(p, MA_OWNED);
1975 refcount_acquire(&fdp->fd_holdcnt);
1980 fddrop(struct filedesc *fdp)
1983 if (fdp->fd_holdcnt > 1) {
1984 if (refcount_release(&fdp->fd_holdcnt) == 0)
1988 FILEDESC_LOCK_DESTROY(fdp);
1989 uma_zfree(filedesc0_zone, fdp);
1993 * Share a filedesc structure.
1996 fdshare(struct filedesc *fdp)
1999 refcount_acquire(&fdp->fd_refcnt);
2004 * Unshare a filedesc structure, if necessary by making a copy
2007 fdunshare(struct thread *td)
2009 struct filedesc *tmp;
2010 struct proc *p = td->td_proc;
2012 if (p->p_fd->fd_refcnt == 1)
2015 tmp = fdcopy(p->p_fd);
2021 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2025 td->td_proc->p_fd = fdp;
2029 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2030 * this is to ease callers, not catch errors.
2033 fdcopy(struct filedesc *fdp)
2035 struct filedesc *newfdp;
2036 struct filedescent *nfde, *ofde;
2041 newfdp = fdinit(fdp, true);
2042 /* copy all passable descriptors (i.e. not kqueue) */
2043 newfdp->fd_freefile = -1;
2044 for (i = 0; i <= fdp->fd_lastfile; ++i) {
2045 ofde = &fdp->fd_ofiles[i];
2046 if (ofde->fde_file == NULL ||
2047 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2048 if (newfdp->fd_freefile == -1)
2049 newfdp->fd_freefile = i;
2052 nfde = &newfdp->fd_ofiles[i];
2054 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2055 fhold(nfde->fde_file);
2056 fdused_init(newfdp, i);
2057 newfdp->fd_lastfile = i;
2059 if (newfdp->fd_freefile == -1)
2060 newfdp->fd_freefile = i;
2061 newfdp->fd_cmask = fdp->fd_cmask;
2062 FILEDESC_SUNLOCK(fdp);
2067 * Copies a filedesc structure, while remapping all file descriptors
2068 * stored inside using a translation table.
2070 * File descriptors are copied over to the new file descriptor table,
2071 * regardless of whether the close-on-exec flag is set.
2074 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2075 struct filedesc **ret)
2077 struct filedesc *newfdp;
2078 struct filedescent *nfde, *ofde;
2083 newfdp = fdinit(fdp, true);
2084 if (nfds > fdp->fd_lastfile + 1) {
2085 /* New table cannot be larger than the old one. */
2089 /* Copy all passable descriptors (i.e. not kqueue). */
2090 newfdp->fd_freefile = nfds;
2091 for (i = 0; i < nfds; ++i) {
2092 if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2093 /* File descriptor out of bounds. */
2097 ofde = &fdp->fd_ofiles[fds[i]];
2098 if (ofde->fde_file == NULL) {
2099 /* Unused file descriptor. */
2103 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2104 /* File descriptor cannot be passed. */
2108 nfde = &newfdp->fd_ofiles[i];
2110 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2111 fhold(nfde->fde_file);
2112 fdused_init(newfdp, i);
2113 newfdp->fd_lastfile = i;
2115 newfdp->fd_cmask = fdp->fd_cmask;
2116 FILEDESC_SUNLOCK(fdp);
2120 FILEDESC_SUNLOCK(fdp);
2121 fdescfree_remapped(newfdp);
2126 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2127 * one of processes using it exits) and the table used to be shared.
2130 fdclearlocks(struct thread *td)
2132 struct filedesc *fdp;
2133 struct filedesc_to_leader *fdtol;
2143 MPASS(fdtol != NULL);
2145 FILEDESC_XLOCK(fdp);
2146 KASSERT(fdtol->fdl_refcount > 0,
2147 ("filedesc_to_refcount botch: fdl_refcount=%d",
2148 fdtol->fdl_refcount));
2149 if (fdtol->fdl_refcount == 1 &&
2150 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2151 for (i = 0; i <= fdp->fd_lastfile; i++) {
2152 fp = fdp->fd_ofiles[i].fde_file;
2153 if (fp == NULL || fp->f_type != DTYPE_VNODE)
2156 FILEDESC_XUNLOCK(fdp);
2157 lf.l_whence = SEEK_SET;
2160 lf.l_type = F_UNLCK;
2162 (void) VOP_ADVLOCK(vp,
2163 (caddr_t)p->p_leader, F_UNLCK,
2165 FILEDESC_XLOCK(fdp);
2170 if (fdtol->fdl_refcount == 1) {
2171 if (fdp->fd_holdleaderscount > 0 &&
2172 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2174 * close() or kern_dup() has cleared a reference
2175 * in a shared file descriptor table.
2177 fdp->fd_holdleaderswakeup = 1;
2178 sx_sleep(&fdp->fd_holdleaderscount,
2179 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2182 if (fdtol->fdl_holdcount > 0) {
2184 * Ensure that fdtol->fdl_leader remains
2185 * valid in closef().
2187 fdtol->fdl_wakeup = 1;
2188 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2193 fdtol->fdl_refcount--;
2194 if (fdtol->fdl_refcount == 0 &&
2195 fdtol->fdl_holdcount == 0) {
2196 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2197 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2201 FILEDESC_XUNLOCK(fdp);
2203 free(fdtol, M_FILEDESC_TO_LEADER);
2207 * Release a filedesc structure.
2210 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2212 struct filedesc0 *fdp0;
2213 struct freetable *ft, *tft;
2214 struct filedescent *fde;
2218 for (i = 0; i <= fdp->fd_lastfile; i++) {
2219 fde = &fdp->fd_ofiles[i];
2224 (void) closef(fp, td);
2230 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2231 free(fdp->fd_map, M_FILEDESC);
2232 if (fdp->fd_nfiles > NDFILE)
2233 free(fdp->fd_files, M_FILEDESC);
2235 fdp0 = (struct filedesc0 *)fdp;
2236 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2237 free(ft->ft_table, M_FILEDESC);
2243 fdescfree(struct thread *td)
2246 struct filedesc *fdp;
2247 struct vnode *cdir, *jdir, *rdir;
2254 if (RACCT_ENABLED())
2255 racct_set_unlocked(p, RACCT_NOFILE, 0);
2258 if (p->p_fdtol != NULL)
2265 if (refcount_release(&fdp->fd_refcnt) == 0)
2268 FILEDESC_XLOCK(fdp);
2269 cdir = fdp->fd_cdir;
2270 fdp->fd_cdir = NULL;
2271 rdir = fdp->fd_rdir;
2272 fdp->fd_rdir = NULL;
2273 jdir = fdp->fd_jdir;
2274 fdp->fd_jdir = NULL;
2275 FILEDESC_XUNLOCK(fdp);
2284 fdescfree_fds(td, fdp, 1);
2288 fdescfree_remapped(struct filedesc *fdp)
2291 if (fdp->fd_cdir != NULL)
2292 vrele(fdp->fd_cdir);
2293 if (fdp->fd_rdir != NULL)
2294 vrele(fdp->fd_rdir);
2295 if (fdp->fd_jdir != NULL)
2296 vrele(fdp->fd_jdir);
2298 fdescfree_fds(curthread, fdp, 0);
2302 * For setugid programs, we don't want to people to use that setugidness
2303 * to generate error messages which write to a file which otherwise would
2304 * otherwise be off-limits to the process. We check for filesystems where
2305 * the vnode can change out from under us after execve (like [lin]procfs).
2307 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2308 * sufficient. We also don't check for setugidness since we know we are.
2311 is_unsafe(struct file *fp)
2315 if (fp->f_type != DTYPE_VNODE)
2319 return ((vp->v_vflag & VV_PROCDEP) != 0);
2323 * Make this setguid thing safe, if at all possible.
2326 fdsetugidsafety(struct thread *td)
2328 struct filedesc *fdp;
2332 fdp = td->td_proc->p_fd;
2333 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2334 MPASS(fdp->fd_nfiles >= 3);
2335 for (i = 0; i <= 2; i++) {
2336 fp = fdp->fd_ofiles[i].fde_file;
2337 if (fp != NULL && is_unsafe(fp)) {
2338 FILEDESC_XLOCK(fdp);
2339 knote_fdclose(td, i);
2341 * NULL-out descriptor prior to close to avoid
2342 * a race while close blocks.
2345 FILEDESC_XUNLOCK(fdp);
2346 (void) closef(fp, td);
2352 * If a specific file object occupies a specific file descriptor, close the
2353 * file descriptor entry and drop a reference on the file object. This is a
2354 * convenience function to handle a subsequent error in a function that calls
2355 * falloc() that handles the race that another thread might have closed the
2356 * file descriptor out from under the thread creating the file object.
2359 fdclose(struct thread *td, struct file *fp, int idx)
2361 struct filedesc *fdp = td->td_proc->p_fd;
2363 FILEDESC_XLOCK(fdp);
2364 if (fdp->fd_ofiles[idx].fde_file == fp) {
2366 FILEDESC_XUNLOCK(fdp);
2369 FILEDESC_XUNLOCK(fdp);
2373 * Close any files on exec?
2376 fdcloseexec(struct thread *td)
2378 struct filedesc *fdp;
2379 struct filedescent *fde;
2383 fdp = td->td_proc->p_fd;
2384 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2385 for (i = 0; i <= fdp->fd_lastfile; i++) {
2386 fde = &fdp->fd_ofiles[i];
2388 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2389 (fde->fde_flags & UF_EXCLOSE))) {
2390 FILEDESC_XLOCK(fdp);
2392 (void) closefp(fdp, i, fp, td, 0);
2393 FILEDESC_UNLOCK_ASSERT(fdp);
2399 * It is unsafe for set[ug]id processes to be started with file
2400 * descriptors 0..2 closed, as these descriptors are given implicit
2401 * significance in the Standard C library. fdcheckstd() will create a
2402 * descriptor referencing /dev/null for each of stdin, stdout, and
2403 * stderr that is not already open.
2406 fdcheckstd(struct thread *td)
2408 struct filedesc *fdp;
2410 int i, error, devnull;
2412 fdp = td->td_proc->p_fd;
2413 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2414 MPASS(fdp->fd_nfiles >= 3);
2416 for (i = 0; i <= 2; i++) {
2417 if (fdp->fd_ofiles[i].fde_file != NULL)
2420 save = td->td_retval[0];
2421 if (devnull != -1) {
2422 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2424 error = kern_openat(td, AT_FDCWD, "/dev/null",
2425 UIO_SYSSPACE, O_RDWR, 0);
2427 devnull = td->td_retval[0];
2428 KASSERT(devnull == i, ("we didn't get our fd"));
2431 td->td_retval[0] = save;
2439 * Internal form of close. Decrement reference count on file structure.
2440 * Note: td may be NULL when closing a file that was being passed in a
2444 closef(struct file *fp, struct thread *td)
2448 struct filedesc_to_leader *fdtol;
2449 struct filedesc *fdp;
2452 * POSIX record locking dictates that any close releases ALL
2453 * locks owned by this process. This is handled by setting
2454 * a flag in the unlock to free ONLY locks obeying POSIX
2455 * semantics, and not to free BSD-style file locks.
2456 * If the descriptor was in a message, POSIX-style locks
2457 * aren't passed with the descriptor, and the thread pointer
2458 * will be NULL. Callers should be careful only to pass a
2459 * NULL thread pointer when there really is no owning
2460 * context that might have locks, or the locks will be
2463 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2465 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2466 lf.l_whence = SEEK_SET;
2469 lf.l_type = F_UNLCK;
2470 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2471 F_UNLCK, &lf, F_POSIX);
2473 fdtol = td->td_proc->p_fdtol;
2474 if (fdtol != NULL) {
2476 * Handle special case where file descriptor table is
2477 * shared between multiple process leaders.
2479 fdp = td->td_proc->p_fd;
2480 FILEDESC_XLOCK(fdp);
2481 for (fdtol = fdtol->fdl_next;
2482 fdtol != td->td_proc->p_fdtol;
2483 fdtol = fdtol->fdl_next) {
2484 if ((fdtol->fdl_leader->p_flag &
2487 fdtol->fdl_holdcount++;
2488 FILEDESC_XUNLOCK(fdp);
2489 lf.l_whence = SEEK_SET;
2492 lf.l_type = F_UNLCK;
2494 (void) VOP_ADVLOCK(vp,
2495 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2497 FILEDESC_XLOCK(fdp);
2498 fdtol->fdl_holdcount--;
2499 if (fdtol->fdl_holdcount == 0 &&
2500 fdtol->fdl_wakeup != 0) {
2501 fdtol->fdl_wakeup = 0;
2505 FILEDESC_XUNLOCK(fdp);
2508 return (fdrop(fp, td));
2512 * Initialize the file pointer with the specified properties.
2514 * The ops are set with release semantics to be certain that the flags, type,
2515 * and data are visible when ops is. This is to prevent ops methods from being
2516 * called with bad data.
2519 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2524 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2528 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2529 struct file **fpp, struct filecaps *havecapsp)
2531 struct filedescent *fde;
2534 FILEDESC_LOCK_ASSERT(fdp);
2536 fde = fdeget_locked(fdp, fd);
2543 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2548 if (havecapsp != NULL)
2549 filecaps_copy(&fde->fde_caps, havecapsp, true);
2551 *fpp = fde->fde_file;
2559 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2560 struct file **fpp, struct filecaps *havecapsp)
2562 struct filedesc *fdp = td->td_proc->p_fd;
2564 #ifndef CAPABILITIES
2565 error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2566 if (error == 0 && havecapsp != NULL)
2567 filecaps_fill(havecapsp);
2573 error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2577 if (havecapsp != NULL) {
2578 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2579 havecapsp, false)) {
2585 if (!fd_modified(fdp, fd, seq))
2594 FILEDESC_SLOCK(fdp);
2595 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2598 FILEDESC_SUNLOCK(fdp);
2604 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2605 struct file **fpp, seq_t *seqp)
2608 const struct filedescent *fde;
2610 const struct fdescenttbl *fdt;
2615 cap_rights_t haverights;
2619 fdt = fdp->fd_files;
2620 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2623 * Fetch the descriptor locklessly. We avoid fdrop() races by
2624 * never raising a refcount above 0. To accomplish this we have
2625 * to use a cmpset loop rather than an atomic_add. The descriptor
2626 * must be re-verified once we acquire a reference to be certain
2627 * that the identity is still correct and we did not lose a race
2628 * due to preemption.
2632 seq = seq_load(fd_seq(fdt, fd));
2633 fde = &fdt->fdt_ofiles[fd];
2634 haverights = *cap_rights_fde_inline(fde);
2636 if (!seq_consistent(fd_seq(fdt, fd), seq))
2639 fp = fdt->fdt_ofiles[fd].fde_file;
2644 error = cap_check(&haverights, needrightsp);
2648 count = fp->f_count;
2652 * Force a reload. Other thread could reallocate the
2653 * table before this fd was closed, so it possible that
2654 * there is a stale fp pointer in cached version.
2656 fdt = *(const struct fdescenttbl * const volatile *)&(fdp->fd_files);
2660 * Use an acquire barrier to force re-reading of fdt so it is
2661 * refreshed for verification.
2663 if (atomic_fcmpset_acq_int(&fp->f_count, &count, count + 1) == 0)
2665 fdt = fdp->fd_files;
2667 if (seq_consistent_nomb(fd_seq(fdt, fd), seq))
2669 if (fp == fdt->fdt_ofiles[fd].fde_file)
2672 fdrop(fp, curthread);
2684 * Extract the file pointer associated with the specified descriptor for the
2685 * current user process.
2687 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2690 * File's rights will be checked against the capability rights mask.
2692 * If an error occurred the non-zero error is returned and *fpp is set to
2693 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
2694 * responsible for fdrop().
2697 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2698 cap_rights_t *needrightsp, seq_t *seqp)
2700 struct filedesc *fdp;
2705 fdp = td->td_proc->p_fd;
2706 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2709 if (fp->f_ops == &badfileops) {
2715 * FREAD and FWRITE failure return EBADF as per POSIX.
2721 if ((fp->f_flag & flags) == 0)
2725 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2726 ((fp->f_flag & FWRITE) != 0))
2732 KASSERT(0, ("wrong flags"));
2745 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2748 return (_fget(td, fd, fpp, 0, rightsp, NULL));
2752 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2756 #ifndef CAPABILITIES
2757 error = _fget(td, fd, fpp, 0, rightsp, NULL);
2758 if (maxprotp != NULL)
2759 *maxprotp = VM_PROT_ALL;
2761 struct filedesc *fdp = td->td_proc->p_fd;
2764 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2766 error = _fget(td, fd, fpp, 0, rightsp, &seq);
2770 * If requested, convert capability rights to access flags.
2772 if (maxprotp != NULL)
2773 *maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd));
2774 if (!fd_modified(fdp, fd, seq))
2783 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2786 return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2790 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2793 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2797 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2800 struct filedesc *fdp = td->td_proc->p_fd;
2801 #ifndef CAPABILITIES
2802 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2807 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2809 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2812 error = cap_fcntl_check(fdp, fd, needfcntl);
2813 if (!fd_modified(fdp, fd, seq))
2826 * Like fget() but loads the underlying vnode, or returns an error if the
2827 * descriptor does not represent a vnode. Note that pipes use vnodes but
2828 * never have VM objects. The returned vnode will be vref()'d.
2830 * XXX: what about the unused flags ?
2833 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2840 error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2843 if (fp->f_vnode == NULL) {
2855 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2858 return (_fgetvp(td, fd, 0, rightsp, vpp));
2862 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2863 struct filecaps *havecaps, struct vnode **vpp)
2865 struct filedesc *fdp;
2866 struct filecaps caps;
2870 fdp = td->td_proc->p_fd;
2871 error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2874 if (fp->f_ops == &badfileops) {
2878 if (fp->f_vnode == NULL) {
2889 filecaps_free(&caps);
2894 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2897 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2901 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2904 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2909 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2913 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2918 * Handle the last reference to a file being closed.
2920 * Without the noinline attribute clang keeps inlining the func thorough this
2921 * file when fdrop is used.
2924 _fdrop(struct file *fp, struct thread *td)
2928 if (fp->f_count != 0)
2929 panic("fdrop: count %d", fp->f_count);
2930 error = fo_close(fp, td);
2931 atomic_subtract_int(&openfiles, 1);
2933 free(fp->f_advice, M_FADVISE);
2934 uma_zfree(file_zone, fp);
2940 * Apply an advisory lock on a file descriptor.
2942 * Just attempt to get a record lock of the requested type on the entire file
2943 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2945 #ifndef _SYS_SYSPROTO_H_
2953 sys_flock(struct thread *td, struct flock_args *uap)
2960 error = fget(td, uap->fd, &cap_flock_rights, &fp);
2963 if (fp->f_type != DTYPE_VNODE) {
2965 return (EOPNOTSUPP);
2969 lf.l_whence = SEEK_SET;
2972 if (uap->how & LOCK_UN) {
2973 lf.l_type = F_UNLCK;
2974 atomic_clear_int(&fp->f_flag, FHASLOCK);
2975 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2978 if (uap->how & LOCK_EX)
2979 lf.l_type = F_WRLCK;
2980 else if (uap->how & LOCK_SH)
2981 lf.l_type = F_RDLCK;
2986 atomic_set_int(&fp->f_flag, FHASLOCK);
2987 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2988 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2994 * Duplicate the specified descriptor to a free descriptor.
2997 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
2998 int openerror, int *indxp)
3000 struct filedescent *newfde, *oldfde;
3005 KASSERT(openerror == ENODEV || openerror == ENXIO,
3006 ("unexpected error %d in %s", openerror, __func__));
3009 * If the to-be-dup'd fd number is greater than the allowed number
3010 * of file descriptors, or the fd to be dup'd has already been
3011 * closed, then reject.
3013 FILEDESC_XLOCK(fdp);
3014 if ((fp = fget_locked(fdp, dfd)) == NULL) {
3015 FILEDESC_XUNLOCK(fdp);
3019 error = fdalloc(td, 0, &indx);
3021 FILEDESC_XUNLOCK(fdp);
3026 * There are two cases of interest here.
3028 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3030 * For ENXIO steal away the file structure from (dfd) and store it in
3031 * (indx). (dfd) is effectively closed by this operation.
3033 switch (openerror) {
3036 * Check that the mode the file is being opened for is a
3037 * subset of the mode of the existing descriptor.
3039 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3040 fdunused(fdp, indx);
3041 FILEDESC_XUNLOCK(fdp);
3045 newfde = &fdp->fd_ofiles[indx];
3046 oldfde = &fdp->fd_ofiles[dfd];
3047 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3049 seq_write_begin(&newfde->fde_seq);
3051 memcpy(newfde, oldfde, fde_change_size);
3052 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3055 seq_write_end(&newfde->fde_seq);
3060 * Steal away the file pointer from dfd and stuff it into indx.
3062 newfde = &fdp->fd_ofiles[indx];
3063 oldfde = &fdp->fd_ofiles[dfd];
3065 seq_write_begin(&newfde->fde_seq);
3067 memcpy(newfde, oldfde, fde_change_size);
3068 oldfde->fde_file = NULL;
3071 seq_write_end(&newfde->fde_seq);
3075 FILEDESC_XUNLOCK(fdp);
3081 * This sysctl determines if we will allow a process to chroot(2) if it
3082 * has a directory open:
3083 * 0: disallowed for all processes.
3084 * 1: allowed for processes that were not already chroot(2)'ed.
3085 * 2: allowed for all processes.
3088 static int chroot_allow_open_directories = 1;
3090 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3091 &chroot_allow_open_directories, 0,
3092 "Allow a process to chroot(2) if it has a directory open");
3095 * Helper function for raised chroot(2) security function: Refuse if
3096 * any filedescriptors are open directories.
3099 chroot_refuse_vdir_fds(struct filedesc *fdp)
3105 FILEDESC_LOCK_ASSERT(fdp);
3107 for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3108 fp = fget_locked(fdp, fd);
3111 if (fp->f_type == DTYPE_VNODE) {
3113 if (vp->v_type == VDIR)
3121 * Common routine for kern_chroot() and jail_attach(). The caller is
3122 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3123 * authorize this operation.
3126 pwd_chroot(struct thread *td, struct vnode *vp)
3128 struct filedesc *fdp;
3129 struct vnode *oldvp;
3132 fdp = td->td_proc->p_fd;
3133 FILEDESC_XLOCK(fdp);
3134 if (chroot_allow_open_directories == 0 ||
3135 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3136 error = chroot_refuse_vdir_fds(fdp);
3138 FILEDESC_XUNLOCK(fdp);
3142 oldvp = fdp->fd_rdir;
3145 if (fdp->fd_jdir == NULL) {
3149 FILEDESC_XUNLOCK(fdp);
3155 pwd_chdir(struct thread *td, struct vnode *vp)
3157 struct filedesc *fdp;
3158 struct vnode *oldvp;
3160 fdp = td->td_proc->p_fd;
3161 FILEDESC_XLOCK(fdp);
3162 VNASSERT(vp->v_usecount > 0, vp,
3163 ("chdir to a vnode with zero usecount"));
3164 oldvp = fdp->fd_cdir;
3166 FILEDESC_XUNLOCK(fdp);
3171 * Scan all active processes and prisons to see if any of them have a current
3172 * or root directory of `olddp'. If so, replace them with the new mount point.
3175 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3177 struct filedesc *fdp;
3182 if (vrefcnt(olddp) == 1)
3185 sx_slock(&allproc_lock);
3186 FOREACH_PROC_IN_SYSTEM(p) {
3192 FILEDESC_XLOCK(fdp);
3193 if (fdp->fd_cdir == olddp) {
3195 fdp->fd_cdir = newdp;
3198 if (fdp->fd_rdir == olddp) {
3200 fdp->fd_rdir = newdp;
3203 if (fdp->fd_jdir == olddp) {
3205 fdp->fd_jdir = newdp;
3208 FILEDESC_XUNLOCK(fdp);
3211 sx_sunlock(&allproc_lock);
3212 if (rootvnode == olddp) {
3217 mtx_lock(&prison0.pr_mtx);
3218 if (prison0.pr_root == olddp) {
3220 prison0.pr_root = newdp;
3223 mtx_unlock(&prison0.pr_mtx);
3224 sx_slock(&allprison_lock);
3225 TAILQ_FOREACH(pr, &allprison, pr_list) {
3226 mtx_lock(&pr->pr_mtx);
3227 if (pr->pr_root == olddp) {
3229 pr->pr_root = newdp;
3232 mtx_unlock(&pr->pr_mtx);
3234 sx_sunlock(&allprison_lock);
3239 struct filedesc_to_leader *
3240 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3242 struct filedesc_to_leader *fdtol;
3244 fdtol = malloc(sizeof(struct filedesc_to_leader),
3245 M_FILEDESC_TO_LEADER, M_WAITOK);
3246 fdtol->fdl_refcount = 1;
3247 fdtol->fdl_holdcount = 0;
3248 fdtol->fdl_wakeup = 0;
3249 fdtol->fdl_leader = leader;
3251 FILEDESC_XLOCK(fdp);
3252 fdtol->fdl_next = old->fdl_next;
3253 fdtol->fdl_prev = old;
3254 old->fdl_next = fdtol;
3255 fdtol->fdl_next->fdl_prev = fdtol;
3256 FILEDESC_XUNLOCK(fdp);
3258 fdtol->fdl_next = fdtol;
3259 fdtol->fdl_prev = fdtol;
3265 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3267 struct filedesc *fdp;
3268 int i, count, slots;
3270 if (*(int *)arg1 != 0)
3273 fdp = curproc->p_fd;
3275 FILEDESC_SLOCK(fdp);
3276 slots = NDSLOTS(fdp->fd_lastfile + 1);
3277 for (i = 0; i < slots; i++)
3278 count += bitcountl(fdp->fd_map[i]);
3279 FILEDESC_SUNLOCK(fdp);
3281 return (SYSCTL_OUT(req, &count, sizeof(count)));
3284 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3285 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3286 "Number of open file descriptors");
3289 * Get file structures globally.
3292 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3295 struct filedesc *fdp;
3300 error = sysctl_wire_old_buffer(req, 0);
3303 if (req->oldptr == NULL) {
3305 sx_slock(&allproc_lock);
3306 FOREACH_PROC_IN_SYSTEM(p) {
3308 if (p->p_state == PRS_NEW) {
3316 /* overestimates sparse tables. */
3317 if (fdp->fd_lastfile > 0)
3318 n += fdp->fd_lastfile;
3321 sx_sunlock(&allproc_lock);
3322 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3325 bzero(&xf, sizeof(xf));
3326 xf.xf_size = sizeof(xf);
3327 sx_slock(&allproc_lock);
3328 FOREACH_PROC_IN_SYSTEM(p) {
3330 if (p->p_state == PRS_NEW) {
3334 if (p_cansee(req->td, p) != 0) {
3338 xf.xf_pid = p->p_pid;
3339 xf.xf_uid = p->p_ucred->cr_uid;
3344 FILEDESC_SLOCK(fdp);
3345 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3346 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3349 xf.xf_file = (uintptr_t)fp;
3350 xf.xf_data = (uintptr_t)fp->f_data;
3351 xf.xf_vnode = (uintptr_t)fp->f_vnode;
3352 xf.xf_type = (uintptr_t)fp->f_type;
3353 xf.xf_count = fp->f_count;
3355 xf.xf_offset = foffset_get(fp);
3356 xf.xf_flag = fp->f_flag;
3357 error = SYSCTL_OUT(req, &xf, sizeof(xf));
3361 FILEDESC_SUNLOCK(fdp);
3366 sx_sunlock(&allproc_lock);
3370 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3371 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3373 #ifdef KINFO_FILE_SIZE
3374 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3378 xlate_fflags(int fflags)
3380 static const struct {
3383 } fflags_table[] = {
3384 { FAPPEND, KF_FLAG_APPEND },
3385 { FASYNC, KF_FLAG_ASYNC },
3386 { FFSYNC, KF_FLAG_FSYNC },
3387 { FHASLOCK, KF_FLAG_HASLOCK },
3388 { FNONBLOCK, KF_FLAG_NONBLOCK },
3389 { FREAD, KF_FLAG_READ },
3390 { FWRITE, KF_FLAG_WRITE },
3391 { O_CREAT, KF_FLAG_CREAT },
3392 { O_DIRECT, KF_FLAG_DIRECT },
3393 { O_EXCL, KF_FLAG_EXCL },
3394 { O_EXEC, KF_FLAG_EXEC },
3395 { O_EXLOCK, KF_FLAG_EXLOCK },
3396 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3397 { O_SHLOCK, KF_FLAG_SHLOCK },
3398 { O_TRUNC, KF_FLAG_TRUNC }
3404 for (i = 0; i < nitems(fflags_table); i++)
3405 if (fflags & fflags_table[i].fflag)
3406 kflags |= fflags_table[i].kf_fflag;
3410 /* Trim unused data from kf_path by truncating the structure size. */
3412 pack_kinfo(struct kinfo_file *kif)
3415 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3416 strlen(kif->kf_path) + 1;
3417 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3421 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3422 struct kinfo_file *kif, struct filedesc *fdp, int flags)
3426 bzero(kif, sizeof(*kif));
3428 /* Set a default type to allow for empty fill_kinfo() methods. */
3429 kif->kf_type = KF_TYPE_UNKNOWN;
3430 kif->kf_flags = xlate_fflags(fp->f_flag);
3431 if (rightsp != NULL)
3432 kif->kf_cap_rights = *rightsp;
3434 cap_rights_init(&kif->kf_cap_rights);
3436 kif->kf_ref_count = fp->f_count;
3437 kif->kf_offset = foffset_get(fp);
3440 * This may drop the filedesc lock, so the 'fp' cannot be
3441 * accessed after this call.
3443 error = fo_fill_kinfo(fp, kif, fdp);
3445 kif->kf_status |= KF_ATTR_VALID;
3446 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3449 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3453 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3454 struct kinfo_file *kif, int flags)
3458 bzero(kif, sizeof(*kif));
3460 kif->kf_type = KF_TYPE_VNODE;
3461 error = vn_fill_kinfo_vnode(vp, kif);
3463 kif->kf_status |= KF_ATTR_VALID;
3464 kif->kf_flags = xlate_fflags(fflags);
3465 cap_rights_init(&kif->kf_cap_rights);
3467 kif->kf_ref_count = -1;
3468 kif->kf_offset = -1;
3469 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3472 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3476 struct export_fd_buf {
3477 struct filedesc *fdp;
3480 struct kinfo_file kif;
3485 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3487 struct kinfo_file *kif;
3490 if (efbuf->remainder != -1) {
3491 if (efbuf->remainder < kif->kf_structsize) {
3492 /* Terminate export. */
3493 efbuf->remainder = 0;
3496 efbuf->remainder -= kif->kf_structsize;
3498 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3502 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3503 struct export_fd_buf *efbuf)
3507 if (efbuf->remainder == 0)
3509 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3511 FILEDESC_SUNLOCK(efbuf->fdp);
3512 error = export_kinfo_to_sb(efbuf);
3513 FILEDESC_SLOCK(efbuf->fdp);
3518 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3519 struct export_fd_buf *efbuf)
3523 if (efbuf->remainder == 0)
3525 if (efbuf->fdp != NULL)
3526 FILEDESC_SUNLOCK(efbuf->fdp);
3527 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3528 error = export_kinfo_to_sb(efbuf);
3529 if (efbuf->fdp != NULL)
3530 FILEDESC_SLOCK(efbuf->fdp);
3535 * Store a process file descriptor information to sbuf.
3537 * Takes a locked proc as argument, and returns with the proc unlocked.
3540 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
3544 struct filedesc *fdp;
3545 struct export_fd_buf *efbuf;
3546 struct vnode *cttyvp, *textvp, *tracevp;
3548 cap_rights_t rights;
3550 PROC_LOCK_ASSERT(p, MA_OWNED);
3553 tracevp = p->p_tracevp;
3554 if (tracevp != NULL)
3557 textvp = p->p_textvp;
3560 /* Controlling tty. */
3562 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3563 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3569 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3572 efbuf->remainder = maxlen;
3573 efbuf->flags = flags;
3574 if (tracevp != NULL)
3575 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3578 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3580 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3586 FILEDESC_SLOCK(fdp);
3587 /* working directory */
3588 if (fdp->fd_cdir != NULL) {
3589 vrefact(fdp->fd_cdir);
3590 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3592 /* root directory */
3593 if (fdp->fd_rdir != NULL) {
3594 vrefact(fdp->fd_rdir);
3595 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3597 /* jail directory */
3598 if (fdp->fd_jdir != NULL) {
3599 vrefact(fdp->fd_jdir);
3600 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3602 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3603 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3606 rights = *cap_rights(fdp, i);
3607 #else /* !CAPABILITIES */
3608 rights = cap_no_rights;
3611 * Create sysctl entry. It is OK to drop the filedesc
3612 * lock inside of export_file_to_sb() as we will
3613 * re-validate and re-evaluate its properties when the
3616 error = export_file_to_sb(fp, i, &rights, efbuf);
3617 if (error != 0 || efbuf->remainder == 0)
3620 FILEDESC_SUNLOCK(fdp);
3623 free(efbuf, M_TEMP);
3627 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
3630 * Get per-process file descriptors for use by procstat(1), et al.
3633 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3638 int error, error2, *name;
3642 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3643 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3644 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3649 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3650 error = kern_proc_filedesc_out(p, &sb, maxlen,
3651 KERN_FILEDESC_PACK_KINFO);
3652 error2 = sbuf_finish(&sb);
3654 return (error != 0 ? error : error2);
3657 #ifdef COMPAT_FREEBSD7
3658 #ifdef KINFO_OFILE_SIZE
3659 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3663 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3666 okif->kf_structsize = sizeof(*okif);
3667 okif->kf_type = kif->kf_type;
3668 okif->kf_fd = kif->kf_fd;
3669 okif->kf_ref_count = kif->kf_ref_count;
3670 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3671 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3672 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3673 okif->kf_offset = kif->kf_offset;
3674 if (kif->kf_type == KF_TYPE_VNODE)
3675 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
3677 okif->kf_vnode_type = KF_VTYPE_VNON;
3678 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3679 if (kif->kf_type == KF_TYPE_SOCKET) {
3680 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
3681 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
3682 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
3683 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
3684 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
3686 okif->kf_sa_local.ss_family = AF_UNSPEC;
3687 okif->kf_sa_peer.ss_family = AF_UNSPEC;
3692 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3693 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3698 FILEDESC_SUNLOCK(fdp);
3699 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3700 kinfo_to_okinfo(kif, okif);
3701 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3702 FILEDESC_SLOCK(fdp);
3707 * Get per-process file descriptors for use by procstat(1), et al.
3710 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3712 struct kinfo_ofile *okif;
3713 struct kinfo_file *kif;
3714 struct filedesc *fdp;
3715 int error, i, *name;
3720 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3727 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3728 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3729 FILEDESC_SLOCK(fdp);
3730 if (fdp->fd_cdir != NULL)
3731 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3733 if (fdp->fd_rdir != NULL)
3734 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3736 if (fdp->fd_jdir != NULL)
3737 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3739 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3740 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3742 export_file_to_kinfo(fp, i, NULL, kif, fdp,
3743 KERN_FILEDESC_PACK_KINFO);
3744 FILEDESC_SUNLOCK(fdp);
3745 kinfo_to_okinfo(kif, okif);
3746 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3747 FILEDESC_SLOCK(fdp);
3751 FILEDESC_SUNLOCK(fdp);
3758 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3759 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3760 "Process ofiledesc entries");
3761 #endif /* COMPAT_FREEBSD7 */
3764 vntype_to_kinfo(int vtype)
3769 } vtypes_table[] = {
3770 { VBAD, KF_VTYPE_VBAD },
3771 { VBLK, KF_VTYPE_VBLK },
3772 { VCHR, KF_VTYPE_VCHR },
3773 { VDIR, KF_VTYPE_VDIR },
3774 { VFIFO, KF_VTYPE_VFIFO },
3775 { VLNK, KF_VTYPE_VLNK },
3776 { VNON, KF_VTYPE_VNON },
3777 { VREG, KF_VTYPE_VREG },
3778 { VSOCK, KF_VTYPE_VSOCK }
3783 * Perform vtype translation.
3785 for (i = 0; i < nitems(vtypes_table); i++)
3786 if (vtypes_table[i].vtype == vtype)
3787 return (vtypes_table[i].kf_vtype);
3789 return (KF_VTYPE_UNKNOWN);
3792 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3793 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3794 "Process filedesc entries");
3797 * Store a process current working directory information to sbuf.
3799 * Takes a locked proc as argument, and returns with the proc unlocked.
3802 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
3804 struct filedesc *fdp;
3805 struct export_fd_buf *efbuf;
3808 PROC_LOCK_ASSERT(p, MA_OWNED);
3815 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3818 efbuf->remainder = maxlen;
3820 FILEDESC_SLOCK(fdp);
3821 if (fdp->fd_cdir == NULL)
3824 vrefact(fdp->fd_cdir);
3825 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3828 FILEDESC_SUNLOCK(fdp);
3830 free(efbuf, M_TEMP);
3835 * Get per-process current working directory.
3838 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3843 int error, error2, *name;
3847 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3848 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3849 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3854 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3855 error = kern_proc_cwd_out(p, &sb, maxlen);
3856 error2 = sbuf_finish(&sb);
3858 return (error != 0 ? error : error2);
3861 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3862 sysctl_kern_proc_cwd, "Process current working directory");
3866 * For the purposes of debugging, generate a human-readable string for the
3870 file_type_to_name(short type)
3898 case DTYPE_PROCDESC:
3900 case DTYPE_LINUXEFD:
3902 case DTYPE_LINUXTFD:
3910 * For the purposes of debugging, identify a process (if any, perhaps one of
3911 * many) that references the passed file in its file descriptor array. Return
3914 static struct proc *
3915 file_to_first_proc(struct file *fp)
3917 struct filedesc *fdp;
3921 FOREACH_PROC_IN_SYSTEM(p) {
3922 if (p->p_state == PRS_NEW)
3927 for (n = 0; n <= fdp->fd_lastfile; n++) {
3928 if (fp == fdp->fd_ofiles[n].fde_file)
3936 db_print_file(struct file *fp, int header)
3938 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
3942 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
3943 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
3944 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
3946 p = file_to_first_proc(fp);
3947 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
3948 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
3949 fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode,
3950 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3955 DB_SHOW_COMMAND(file, db_show_file)
3960 db_printf("usage: show file <addr>\n");
3963 fp = (struct file *)addr;
3964 db_print_file(fp, 1);
3967 DB_SHOW_COMMAND(files, db_show_files)
3969 struct filedesc *fdp;
3976 FOREACH_PROC_IN_SYSTEM(p) {
3977 if (p->p_state == PRS_NEW)
3979 if ((fdp = p->p_fd) == NULL)
3981 for (n = 0; n <= fdp->fd_lastfile; ++n) {
3982 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3984 db_print_file(fp, header);
3991 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3992 &maxfilesperproc, 0, "Maximum files allowed open per process");
3994 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3995 &maxfiles, 0, "Maximum number of files");
3997 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3998 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
4002 filelistinit(void *dummy)
4005 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4006 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4007 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4008 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4009 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4011 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4013 /*-------------------------------------------------------------------*/
4016 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4017 int flags, struct thread *td)
4024 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4032 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4040 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4048 badfo_kqfilter(struct file *fp, struct knote *kn)
4055 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4063 badfo_close(struct file *fp, struct thread *td)
4070 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4078 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4086 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4087 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4095 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4101 struct fileops badfileops = {
4102 .fo_read = badfo_readwrite,
4103 .fo_write = badfo_readwrite,
4104 .fo_truncate = badfo_truncate,
4105 .fo_ioctl = badfo_ioctl,
4106 .fo_poll = badfo_poll,
4107 .fo_kqfilter = badfo_kqfilter,
4108 .fo_stat = badfo_stat,
4109 .fo_close = badfo_close,
4110 .fo_chmod = badfo_chmod,
4111 .fo_chown = badfo_chown,
4112 .fo_sendfile = badfo_sendfile,
4113 .fo_fill_kinfo = badfo_fill_kinfo,
4117 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4118 int flags, struct thread *td)
4121 return (EOPNOTSUPP);
4125 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4133 invfo_ioctl(struct file *fp, u_long com, void *data,
4134 struct ucred *active_cred, struct thread *td)
4141 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4145 return (poll_no_poll(events));
4149 invfo_kqfilter(struct file *fp, struct knote *kn)
4156 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4164 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4172 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4173 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4180 /*-------------------------------------------------------------------*/
4183 * File Descriptor pseudo-device driver (/dev/fd/).
4185 * Opening minor device N dup()s the file (if any) connected to file
4186 * descriptor N belonging to the calling process. Note that this driver
4187 * consists of only the ``open()'' routine, because all subsequent
4188 * references to this file will be direct to the other driver.
4190 * XXX: we could give this one a cloning event handler if necessary.
4195 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4199 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4200 * the file descriptor being sought for duplication. The error
4201 * return ensures that the vnode for this device will be released
4202 * by vn_open. Open will detect this special error and take the
4203 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4204 * will simply report the error.
4206 td->td_dupfd = dev2unit(dev);
4210 static struct cdevsw fildesc_cdevsw = {
4211 .d_version = D_VERSION,
4217 fildesc_drvinit(void *unused)
4221 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4222 UID_ROOT, GID_WHEEL, 0666, "fd/0");
4223 make_dev_alias(dev, "stdin");
4224 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4225 UID_ROOT, GID_WHEEL, 0666, "fd/1");
4226 make_dev_alias(dev, "stdout");
4227 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4228 UID_ROOT, GID_WHEEL, 0666, "fd/2");
4229 make_dev_alias(dev, "stderr");
4232 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);