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 seqc_write_begin(&fde->fde_seqc);
309 fde->fde_file = NULL;
311 seqc_write_end(&fde->fde_seqc);
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 flp = (struct flock *)arg;
616 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
621 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL);
624 if (fp->f_type != DTYPE_VNODE) {
630 if (flp->l_whence == SEEK_CUR) {
631 foffset = foffset_get(fp);
634 foffset > OFF_MAX - flp->l_start)) {
639 flp->l_start += foffset;
643 switch (flp->l_type) {
645 if ((fp->f_flag & FREAD) == 0) {
649 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
650 PROC_LOCK(p->p_leader);
651 p->p_leader->p_flag |= P_ADVLOCK;
652 PROC_UNLOCK(p->p_leader);
654 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
658 if ((fp->f_flag & FWRITE) == 0) {
662 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
663 PROC_LOCK(p->p_leader);
664 p->p_leader->p_flag |= P_ADVLOCK;
665 PROC_UNLOCK(p->p_leader);
667 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
671 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
675 if (flg != F_REMOTE) {
679 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
680 F_UNLCKSYS, flp, flg);
686 if (error != 0 || flp->l_type == F_UNLCK ||
687 flp->l_type == F_UNLCKSYS) {
693 * Check for a race with close.
695 * The vnode is now advisory locked (or unlocked, but this case
696 * is not really important) as the caller requested.
697 * We had to drop the filedesc lock, so we need to recheck if
698 * the descriptor is still valid, because if it was closed
699 * in the meantime we need to remove advisory lock from the
700 * vnode - close on any descriptor leading to an advisory
701 * locked vnode, removes that lock.
702 * We will return 0 on purpose in that case, as the result of
703 * successful advisory lock might have been externally visible
704 * already. This is fine - effectively we pretend to the caller
705 * that the closing thread was a bit slower and that the
706 * advisory lock succeeded before the close.
708 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2, NULL);
714 flp->l_whence = SEEK_SET;
717 flp->l_type = F_UNLCK;
718 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
719 F_UNLCK, flp, F_POSIX);
726 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL);
729 if (fp->f_type != DTYPE_VNODE) {
734 flp = (struct flock *)arg;
735 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
736 flp->l_type != F_UNLCK) {
741 if (flp->l_whence == SEEK_CUR) {
742 foffset = foffset_get(fp);
743 if ((flp->l_start > 0 &&
744 foffset > OFF_MAX - flp->l_start) ||
746 foffset < OFF_MIN - flp->l_start)) {
751 flp->l_start += foffset;
754 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
760 arg = arg ? 128 * 1024: 0;
763 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp, NULL);
766 if (fp->f_type != DTYPE_VNODE) {
773 * Exclusive lock synchronizes against f_seqcount reads and
774 * writes in sequential_heuristic().
776 error = vn_lock(vp, LK_EXCLUSIVE);
782 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
783 arg = MIN(arg, INT_MAX - bsize + 1);
784 fp->f_seqcount = MIN(IO_SEQMAX,
785 (arg + bsize - 1) / bsize);
786 atomic_set_int(&fp->f_flag, FRDAHEAD);
788 atomic_clear_int(&fp->f_flag, FRDAHEAD);
802 getmaxfd(struct thread *td)
805 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
809 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
812 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
814 struct filedesc *fdp;
815 struct filedescent *oldfde, *newfde;
818 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;
857 * If the caller specified a file descriptor, make sure the file
858 * table is large enough to hold it, and grab it. Otherwise, just
859 * allocate a new descriptor the usual way.
864 if ((error = fdalloc(td, new, &new)) != 0)
867 case FDDUP_MUSTREPLACE:
868 /* Target file descriptor must exist. */
869 if (fget_locked(fdp, new) == NULL)
873 if (new >= fdp->fd_nfiles) {
875 * The resource limits are here instead of e.g.
876 * fdalloc(), because the file descriptor table may be
877 * shared between processes, so we can't really use
878 * racct_add()/racct_sub(). Instead of counting the
879 * number of actually allocated descriptors, just put
880 * the limit on the size of the file descriptor table.
883 if (RACCT_ENABLED()) {
884 error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
891 fdgrowtable_exp(fdp, new + 1);
893 if (!fdisused(fdp, new))
897 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
900 KASSERT(old != new, ("new fd is same as old"));
902 oldfde = &fdp->fd_ofiles[old];
903 fhold(oldfde->fde_file);
904 newfde = &fdp->fd_ofiles[new];
905 delfp = newfde->fde_file;
907 oioctls = filecaps_free_prep(&newfde->fde_caps);
908 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
911 * Duplicate the source descriptor.
914 seqc_write_begin(&newfde->fde_seqc);
916 memcpy(newfde, oldfde, fde_change_size);
917 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
919 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
920 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
922 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
924 seqc_write_end(&newfde->fde_seqc);
926 td->td_retval[0] = new;
931 (void) closefp(fdp, new, delfp, td, 1);
932 FILEDESC_UNLOCK_ASSERT(fdp);
935 FILEDESC_XUNLOCK(fdp);
938 filecaps_free_finish(oioctls);
943 * If sigio is on the list associated with a process or process group,
944 * disable signalling from the device, remove sigio from the list and
948 funsetown(struct sigio **sigiop)
960 *(sigio->sio_myref) = NULL;
961 if ((sigio)->sio_pgid < 0) {
962 struct pgrp *pg = (sigio)->sio_pgrp;
964 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
968 struct proc *p = (sigio)->sio_proc;
970 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
975 crfree(sigio->sio_ucred);
976 free(sigio, M_SIGIO);
980 * Free a list of sigio structures.
981 * We only need to lock the SIGIO_LOCK because we have made ourselves
982 * inaccessible to callers of fsetown and therefore do not need to lock
983 * the proc or pgrp struct for the list manipulation.
986 funsetownlst(struct sigiolst *sigiolst)
992 sigio = SLIST_FIRST(sigiolst);
999 * Every entry of the list should belong
1000 * to a single proc or pgrp.
1002 if (sigio->sio_pgid < 0) {
1003 pg = sigio->sio_pgrp;
1004 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
1005 } else /* if (sigio->sio_pgid > 0) */ {
1006 p = sigio->sio_proc;
1007 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1011 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1012 *(sigio->sio_myref) = NULL;
1014 KASSERT(sigio->sio_pgid < 0,
1015 ("Proc sigio in pgrp sigio list"));
1016 KASSERT(sigio->sio_pgrp == pg,
1017 ("Bogus pgrp in sigio list"));
1019 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1022 } else /* if (p != NULL) */ {
1023 KASSERT(sigio->sio_pgid > 0,
1024 ("Pgrp sigio in proc sigio list"));
1025 KASSERT(sigio->sio_proc == p,
1026 ("Bogus proc in sigio list"));
1028 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1033 crfree(sigio->sio_ucred);
1034 free(sigio, M_SIGIO);
1041 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1043 * After permission checking, add a sigio structure to the sigio list for
1044 * the process or process group.
1047 fsetown(pid_t pgid, struct sigio **sigiop)
1051 struct sigio *sigio;
1061 /* Allocate and fill in the new sigio out of locks. */
1062 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1063 sigio->sio_pgid = pgid;
1064 sigio->sio_ucred = crhold(curthread->td_ucred);
1065 sigio->sio_myref = sigiop;
1067 sx_slock(&proctree_lock);
1076 * Policy - Don't allow a process to FSETOWN a process
1077 * in another session.
1079 * Remove this test to allow maximum flexibility or
1080 * restrict FSETOWN to the current process or process
1081 * group for maximum safety.
1084 if (proc->p_session != curthread->td_proc->p_session) {
1090 } else /* if (pgid < 0) */ {
1091 pgrp = pgfind(-pgid);
1099 * Policy - Don't allow a process to FSETOWN a process
1100 * in another session.
1102 * Remove this test to allow maximum flexibility or
1103 * restrict FSETOWN to the current process or process
1104 * group for maximum safety.
1106 if (pgrp->pg_session != curthread->td_proc->p_session) {
1117 * Since funsetownlst() is called without the proctree
1118 * locked, we need to check for P_WEXIT.
1119 * XXX: is ESRCH correct?
1121 if ((proc->p_flag & P_WEXIT) != 0) {
1126 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1127 sigio->sio_proc = proc;
1131 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1132 sigio->sio_pgrp = pgrp;
1135 sx_sunlock(&proctree_lock);
1142 sx_sunlock(&proctree_lock);
1143 crfree(sigio->sio_ucred);
1144 free(sigio, M_SIGIO);
1149 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1152 fgetown(struct sigio **sigiop)
1157 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1163 * Function drops the filedesc lock on return.
1166 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1171 FILEDESC_XLOCK_ASSERT(fdp);
1174 if (td->td_proc->p_fdtol != NULL) {
1176 * Ask fdfree() to sleep to ensure that all relevant
1177 * process leaders can be traversed in closef().
1179 fdp->fd_holdleaderscount++;
1186 * We now hold the fp reference that used to be owned by the
1187 * descriptor array. We have to unlock the FILEDESC *AFTER*
1188 * knote_fdclose to prevent a race of the fd getting opened, a knote
1189 * added, and deleteing a knote for the new fd.
1191 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1192 knote_fdclose(td, fd);
1195 * We need to notify mqueue if the object is of type mqueue.
1197 if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1198 mq_fdclose(td, fd, fp);
1199 FILEDESC_XUNLOCK(fdp);
1201 error = closef(fp, td);
1203 FILEDESC_XLOCK(fdp);
1204 fdp->fd_holdleaderscount--;
1205 if (fdp->fd_holdleaderscount == 0 &&
1206 fdp->fd_holdleaderswakeup != 0) {
1207 fdp->fd_holdleaderswakeup = 0;
1208 wakeup(&fdp->fd_holdleaderscount);
1210 FILEDESC_XUNLOCK(fdp);
1216 * Close a file descriptor.
1218 #ifndef _SYS_SYSPROTO_H_
1225 sys_close(struct thread *td, struct close_args *uap)
1228 return (kern_close(td, uap->fd));
1232 kern_close(struct thread *td, int fd)
1234 struct filedesc *fdp;
1237 fdp = td->td_proc->p_fd;
1239 AUDIT_SYSCLOSE(td, fd);
1241 FILEDESC_XLOCK(fdp);
1242 if ((fp = fget_locked(fdp, fd)) == NULL) {
1243 FILEDESC_XUNLOCK(fdp);
1248 /* closefp() drops the FILEDESC lock for us. */
1249 return (closefp(fdp, fd, fp, td, 1));
1253 * Close open file descriptors.
1255 #ifndef _SYS_SYSPROTO_H_
1256 struct closefrom_args {
1262 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1264 struct filedesc *fdp;
1267 fdp = td->td_proc->p_fd;
1268 AUDIT_ARG_FD(uap->lowfd);
1271 * Treat negative starting file descriptor values identical to
1272 * closefrom(0) which closes all files.
1276 FILEDESC_SLOCK(fdp);
1277 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1278 if (fdp->fd_ofiles[fd].fde_file != NULL) {
1279 FILEDESC_SUNLOCK(fdp);
1280 (void)kern_close(td, fd);
1281 FILEDESC_SLOCK(fdp);
1284 FILEDESC_SUNLOCK(fdp);
1288 #if defined(COMPAT_43)
1290 * Return status information about a file descriptor.
1292 #ifndef _SYS_SYSPROTO_H_
1293 struct ofstat_args {
1300 ofstat(struct thread *td, struct ofstat_args *uap)
1306 error = kern_fstat(td, uap->fd, &ub);
1309 error = copyout(&oub, uap->sb, sizeof(oub));
1313 #endif /* COMPAT_43 */
1315 #if defined(COMPAT_FREEBSD11)
1317 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1320 struct freebsd11_stat osb;
1323 error = kern_fstat(td, uap->fd, &sb);
1326 error = freebsd11_cvtstat(&sb, &osb);
1328 error = copyout(&osb, uap->sb, sizeof(osb));
1331 #endif /* COMPAT_FREEBSD11 */
1334 * Return status information about a file descriptor.
1336 #ifndef _SYS_SYSPROTO_H_
1344 sys_fstat(struct thread *td, struct fstat_args *uap)
1349 error = kern_fstat(td, uap->fd, &ub);
1351 error = copyout(&ub, uap->sb, sizeof(ub));
1356 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1363 error = fget(td, fd, &cap_fstat_rights, &fp);
1367 AUDIT_ARG_FILE(td->td_proc, fp);
1369 error = fo_stat(fp, sbp, td->td_ucred, td);
1371 #ifdef __STAT_TIME_T_EXT
1373 sbp->st_atim_ext = 0;
1374 sbp->st_mtim_ext = 0;
1375 sbp->st_ctim_ext = 0;
1376 sbp->st_btim_ext = 0;
1380 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1386 #if defined(COMPAT_FREEBSD11)
1388 * Return status information about a file descriptor.
1390 #ifndef _SYS_SYSPROTO_H_
1391 struct freebsd11_nfstat_args {
1398 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1404 error = kern_fstat(td, uap->fd, &ub);
1406 freebsd11_cvtnstat(&ub, &nub);
1407 error = copyout(&nub, uap->sb, sizeof(nub));
1411 #endif /* COMPAT_FREEBSD11 */
1414 * Return pathconf information about a file descriptor.
1416 #ifndef _SYS_SYSPROTO_H_
1417 struct fpathconf_args {
1424 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1429 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1431 td->td_retval[0] = value;
1436 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1442 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1446 if (name == _PC_ASYNC_IO) {
1447 *valuep = _POSIX_ASYNCHRONOUS_IO;
1452 vn_lock(vp, LK_SHARED | LK_RETRY);
1453 error = VOP_PATHCONF(vp, name, valuep);
1455 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1456 if (name != _PC_PIPE_BUF) {
1471 * Copy filecaps structure allocating memory for ioctls array if needed.
1473 * The last parameter indicates whether the fdtable is locked. If it is not and
1474 * ioctls are encountered, copying fails and the caller must lock the table.
1476 * Note that if the table was not locked, the caller has to check the relevant
1477 * sequence counter to determine whether the operation was successful.
1480 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1484 if (src->fc_ioctls != NULL && !locked)
1486 memcpy(dst, src, sizeof(*src));
1487 if (src->fc_ioctls == NULL)
1490 KASSERT(src->fc_nioctls > 0,
1491 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1493 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1494 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1495 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1500 filecaps_copy_prep(const struct filecaps *src)
1505 if (__predict_true(src->fc_ioctls == NULL))
1508 KASSERT(src->fc_nioctls > 0,
1509 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1511 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1512 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1517 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1523 if (__predict_true(src->fc_ioctls == NULL)) {
1524 MPASS(ioctls == NULL);
1528 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1529 dst->fc_ioctls = ioctls;
1530 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1534 * Move filecaps structure to the new place and clear the old place.
1537 filecaps_move(struct filecaps *src, struct filecaps *dst)
1541 bzero(src, sizeof(*src));
1545 * Fill the given filecaps structure with full rights.
1548 filecaps_fill(struct filecaps *fcaps)
1551 CAP_ALL(&fcaps->fc_rights);
1552 fcaps->fc_ioctls = NULL;
1553 fcaps->fc_nioctls = -1;
1554 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1558 * Free memory allocated within filecaps structure.
1561 filecaps_free(struct filecaps *fcaps)
1564 free(fcaps->fc_ioctls, M_FILECAPS);
1565 bzero(fcaps, sizeof(*fcaps));
1569 filecaps_free_prep(struct filecaps *fcaps)
1573 ioctls = fcaps->fc_ioctls;
1574 bzero(fcaps, sizeof(*fcaps));
1579 filecaps_free_finish(u_long *ioctls)
1582 free(ioctls, M_FILECAPS);
1586 * Validate the given filecaps structure.
1589 filecaps_validate(const struct filecaps *fcaps, const char *func)
1592 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1593 ("%s: invalid rights", func));
1594 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1595 ("%s: invalid fcntls", func));
1596 KASSERT(fcaps->fc_fcntls == 0 ||
1597 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1598 ("%s: fcntls without CAP_FCNTL", func));
1599 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1600 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1601 ("%s: invalid ioctls", func));
1602 KASSERT(fcaps->fc_nioctls == 0 ||
1603 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1604 ("%s: ioctls without CAP_IOCTL", func));
1608 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1612 FILEDESC_XLOCK_ASSERT(fdp);
1614 nfd1 = fdp->fd_nfiles * 2;
1617 fdgrowtable(fdp, nfd1);
1621 * Grow the file table to accommodate (at least) nfd descriptors.
1624 fdgrowtable(struct filedesc *fdp, int nfd)
1626 struct filedesc0 *fdp0;
1627 struct freetable *ft;
1628 struct fdescenttbl *ntable;
1629 struct fdescenttbl *otable;
1630 int nnfiles, onfiles;
1631 NDSLOTTYPE *nmap, *omap;
1634 * If lastfile is -1 this struct filedesc was just allocated and we are
1635 * growing it to accommodate for the one we are going to copy from. There
1636 * is no need to have a lock on this one as it's not visible to anyone.
1638 if (fdp->fd_lastfile != -1)
1639 FILEDESC_XLOCK_ASSERT(fdp);
1641 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1643 /* save old values */
1644 onfiles = fdp->fd_nfiles;
1645 otable = fdp->fd_files;
1648 /* compute the size of the new table */
1649 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1650 if (nnfiles <= onfiles)
1651 /* the table is already large enough */
1655 * Allocate a new table. We need enough space for the number of
1656 * entries, file entries themselves and the struct freetable we will use
1657 * when we decommission the table and place it on the freelist.
1658 * We place the struct freetable in the middle so we don't have
1659 * to worry about padding.
1661 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1662 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1663 sizeof(struct freetable),
1664 M_FILEDESC, M_ZERO | M_WAITOK);
1665 /* copy the old data */
1666 ntable->fdt_nfiles = nnfiles;
1667 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1668 onfiles * sizeof(ntable->fdt_ofiles[0]));
1671 * Allocate a new map only if the old is not large enough. It will
1672 * grow at a slower rate than the table as it can map more
1673 * entries than the table can hold.
1675 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1676 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1678 /* copy over the old data and update the pointer */
1679 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1684 * Make sure that ntable is correctly initialized before we replace
1685 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1688 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1691 * Do not free the old file table, as some threads may still
1692 * reference entries within it. Instead, place it on a freelist
1693 * which will be processed when the struct filedesc is released.
1695 * Note that if onfiles == NDFILE, we're dealing with the original
1696 * static allocation contained within (struct filedesc0 *)fdp,
1697 * which must not be freed.
1699 if (onfiles > NDFILE) {
1700 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1701 fdp0 = (struct filedesc0 *)fdp;
1702 ft->ft_table = otable;
1703 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1706 * The map does not have the same possibility of threads still
1707 * holding references to it. So always free it as long as it
1708 * does not reference the original static allocation.
1710 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1711 free(omap, M_FILEDESC);
1715 * Allocate a file descriptor for the process.
1718 fdalloc(struct thread *td, int minfd, int *result)
1720 struct proc *p = td->td_proc;
1721 struct filedesc *fdp = p->p_fd;
1722 int fd, maxfd, allocfd;
1727 FILEDESC_XLOCK_ASSERT(fdp);
1729 if (fdp->fd_freefile > minfd)
1730 minfd = fdp->fd_freefile;
1732 maxfd = getmaxfd(td);
1735 * Search the bitmap for a free descriptor starting at minfd.
1736 * If none is found, grow the file table.
1738 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1741 if (fd >= fdp->fd_nfiles) {
1742 allocfd = min(fd * 2, maxfd);
1744 if (RACCT_ENABLED()) {
1745 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1751 * fd is already equal to first free descriptor >= minfd, so
1752 * we only need to grow the table and we are done.
1754 fdgrowtable_exp(fdp, allocfd);
1758 * Perform some sanity checks, then mark the file descriptor as
1759 * used and return it to the caller.
1761 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1762 ("invalid descriptor %d", fd));
1763 KASSERT(!fdisused(fdp, fd),
1764 ("fd_first_free() returned non-free descriptor"));
1765 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1766 ("file descriptor isn't free"));
1773 * Allocate n file descriptors for the process.
1776 fdallocn(struct thread *td, int minfd, int *fds, int n)
1778 struct proc *p = td->td_proc;
1779 struct filedesc *fdp = p->p_fd;
1782 FILEDESC_XLOCK_ASSERT(fdp);
1784 for (i = 0; i < n; i++)
1785 if (fdalloc(td, 0, &fds[i]) != 0)
1789 for (i--; i >= 0; i--)
1790 fdunused(fdp, fds[i]);
1798 * Create a new open file structure and allocate a file descriptor for the
1799 * process that refers to it. We add one reference to the file for the
1800 * descriptor table and one reference for resultfp. This is to prevent us
1801 * being preempted and the entry in the descriptor table closed after we
1802 * release the FILEDESC lock.
1805 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1806 struct filecaps *fcaps)
1811 error = falloc_noinstall(td, &fp);
1813 return (error); /* no reference held on error */
1815 error = finstall(td, fp, &fd, flags, fcaps);
1817 fdrop(fp, td); /* one reference (fp only) */
1821 if (resultfp != NULL)
1822 *resultfp = fp; /* copy out result */
1824 fdrop(fp, td); /* release local reference */
1826 if (resultfd != NULL)
1833 * Create a new open file structure without allocating a file descriptor.
1836 falloc_noinstall(struct thread *td, struct file **resultfp)
1839 int maxuserfiles = maxfiles - (maxfiles / 20);
1841 static struct timeval lastfail;
1844 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1846 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1847 if ((openfiles_new >= maxuserfiles &&
1848 priv_check(td, PRIV_MAXFILES) != 0) ||
1849 openfiles_new >= maxfiles) {
1850 atomic_subtract_int(&openfiles, 1);
1851 if (ppsratecheck(&lastfail, &curfail, 1)) {
1852 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1853 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1857 fp = uma_zalloc(file_zone, M_WAITOK);
1858 bzero(fp, sizeof(*fp));
1859 refcount_init(&fp->f_count, 1);
1860 fp->f_cred = crhold(td->td_ucred);
1861 fp->f_ops = &badfileops;
1867 * Install a file in a file descriptor table.
1870 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1871 struct filecaps *fcaps)
1873 struct filedescent *fde;
1877 filecaps_validate(fcaps, __func__);
1878 FILEDESC_XLOCK_ASSERT(fdp);
1880 fde = &fdp->fd_ofiles[fd];
1882 seqc_write_begin(&fde->fde_seqc);
1885 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1887 filecaps_move(fcaps, &fde->fde_caps);
1889 filecaps_fill(&fde->fde_caps);
1891 seqc_write_end(&fde->fde_seqc);
1896 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1897 struct filecaps *fcaps)
1899 struct filedesc *fdp = td->td_proc->p_fd;
1904 FILEDESC_XLOCK(fdp);
1905 if ((error = fdalloc(td, 0, fd))) {
1906 FILEDESC_XUNLOCK(fdp);
1910 _finstall(fdp, fp, *fd, flags, fcaps);
1911 FILEDESC_XUNLOCK(fdp);
1916 * Build a new filedesc structure from another.
1917 * Copy the current, root, and jail root vnode references.
1919 * If fdp is not NULL, return with it shared locked.
1922 fdinit(struct filedesc *fdp, bool prepfiles)
1924 struct filedesc0 *newfdp0;
1925 struct filedesc *newfdp;
1927 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1928 newfdp = &newfdp0->fd_fd;
1930 /* Create the file descriptor table. */
1931 FILEDESC_LOCK_INIT(newfdp);
1932 refcount_init(&newfdp->fd_refcnt, 1);
1933 refcount_init(&newfdp->fd_holdcnt, 1);
1934 newfdp->fd_cmask = CMASK;
1935 newfdp->fd_map = newfdp0->fd_dmap;
1936 newfdp->fd_lastfile = -1;
1937 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1938 newfdp->fd_files->fdt_nfiles = NDFILE;
1943 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1944 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1946 FILEDESC_SLOCK(fdp);
1947 newfdp->fd_cdir = fdp->fd_cdir;
1948 if (newfdp->fd_cdir)
1949 vrefact(newfdp->fd_cdir);
1950 newfdp->fd_rdir = fdp->fd_rdir;
1951 if (newfdp->fd_rdir)
1952 vrefact(newfdp->fd_rdir);
1953 newfdp->fd_jdir = fdp->fd_jdir;
1954 if (newfdp->fd_jdir)
1955 vrefact(newfdp->fd_jdir);
1958 FILEDESC_SUNLOCK(fdp);
1960 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1961 FILEDESC_SUNLOCK(fdp);
1962 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1963 FILEDESC_SLOCK(fdp);
1970 static struct filedesc *
1971 fdhold(struct proc *p)
1973 struct filedesc *fdp;
1975 PROC_LOCK_ASSERT(p, MA_OWNED);
1978 refcount_acquire(&fdp->fd_holdcnt);
1983 fddrop(struct filedesc *fdp)
1986 if (fdp->fd_holdcnt > 1) {
1987 if (refcount_release(&fdp->fd_holdcnt) == 0)
1991 FILEDESC_LOCK_DESTROY(fdp);
1992 uma_zfree(filedesc0_zone, fdp);
1996 * Share a filedesc structure.
1999 fdshare(struct filedesc *fdp)
2002 refcount_acquire(&fdp->fd_refcnt);
2007 * Unshare a filedesc structure, if necessary by making a copy
2010 fdunshare(struct thread *td)
2012 struct filedesc *tmp;
2013 struct proc *p = td->td_proc;
2015 if (p->p_fd->fd_refcnt == 1)
2018 tmp = fdcopy(p->p_fd);
2024 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2028 td->td_proc->p_fd = fdp;
2032 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2033 * this is to ease callers, not catch errors.
2036 fdcopy(struct filedesc *fdp)
2038 struct filedesc *newfdp;
2039 struct filedescent *nfde, *ofde;
2044 newfdp = fdinit(fdp, true);
2045 /* copy all passable descriptors (i.e. not kqueue) */
2046 newfdp->fd_freefile = -1;
2047 for (i = 0; i <= fdp->fd_lastfile; ++i) {
2048 ofde = &fdp->fd_ofiles[i];
2049 if (ofde->fde_file == NULL ||
2050 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2051 if (newfdp->fd_freefile == -1)
2052 newfdp->fd_freefile = i;
2055 nfde = &newfdp->fd_ofiles[i];
2057 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2058 fhold(nfde->fde_file);
2059 fdused_init(newfdp, i);
2060 newfdp->fd_lastfile = i;
2062 if (newfdp->fd_freefile == -1)
2063 newfdp->fd_freefile = i;
2064 newfdp->fd_cmask = fdp->fd_cmask;
2065 FILEDESC_SUNLOCK(fdp);
2070 * Copies a filedesc structure, while remapping all file descriptors
2071 * stored inside using a translation table.
2073 * File descriptors are copied over to the new file descriptor table,
2074 * regardless of whether the close-on-exec flag is set.
2077 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2078 struct filedesc **ret)
2080 struct filedesc *newfdp;
2081 struct filedescent *nfde, *ofde;
2086 newfdp = fdinit(fdp, true);
2087 if (nfds > fdp->fd_lastfile + 1) {
2088 /* New table cannot be larger than the old one. */
2092 /* Copy all passable descriptors (i.e. not kqueue). */
2093 newfdp->fd_freefile = nfds;
2094 for (i = 0; i < nfds; ++i) {
2095 if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2096 /* File descriptor out of bounds. */
2100 ofde = &fdp->fd_ofiles[fds[i]];
2101 if (ofde->fde_file == NULL) {
2102 /* Unused file descriptor. */
2106 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2107 /* File descriptor cannot be passed. */
2111 nfde = &newfdp->fd_ofiles[i];
2113 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2114 fhold(nfde->fde_file);
2115 fdused_init(newfdp, i);
2116 newfdp->fd_lastfile = i;
2118 newfdp->fd_cmask = fdp->fd_cmask;
2119 FILEDESC_SUNLOCK(fdp);
2123 FILEDESC_SUNLOCK(fdp);
2124 fdescfree_remapped(newfdp);
2129 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2130 * one of processes using it exits) and the table used to be shared.
2133 fdclearlocks(struct thread *td)
2135 struct filedesc *fdp;
2136 struct filedesc_to_leader *fdtol;
2146 MPASS(fdtol != NULL);
2148 FILEDESC_XLOCK(fdp);
2149 KASSERT(fdtol->fdl_refcount > 0,
2150 ("filedesc_to_refcount botch: fdl_refcount=%d",
2151 fdtol->fdl_refcount));
2152 if (fdtol->fdl_refcount == 1 &&
2153 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2154 for (i = 0; i <= fdp->fd_lastfile; i++) {
2155 fp = fdp->fd_ofiles[i].fde_file;
2156 if (fp == NULL || fp->f_type != DTYPE_VNODE)
2159 FILEDESC_XUNLOCK(fdp);
2160 lf.l_whence = SEEK_SET;
2163 lf.l_type = F_UNLCK;
2165 (void) VOP_ADVLOCK(vp,
2166 (caddr_t)p->p_leader, F_UNLCK,
2168 FILEDESC_XLOCK(fdp);
2173 if (fdtol->fdl_refcount == 1) {
2174 if (fdp->fd_holdleaderscount > 0 &&
2175 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2177 * close() or kern_dup() has cleared a reference
2178 * in a shared file descriptor table.
2180 fdp->fd_holdleaderswakeup = 1;
2181 sx_sleep(&fdp->fd_holdleaderscount,
2182 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2185 if (fdtol->fdl_holdcount > 0) {
2187 * Ensure that fdtol->fdl_leader remains
2188 * valid in closef().
2190 fdtol->fdl_wakeup = 1;
2191 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2196 fdtol->fdl_refcount--;
2197 if (fdtol->fdl_refcount == 0 &&
2198 fdtol->fdl_holdcount == 0) {
2199 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2200 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2204 FILEDESC_XUNLOCK(fdp);
2206 free(fdtol, M_FILEDESC_TO_LEADER);
2210 * Release a filedesc structure.
2213 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2215 struct filedesc0 *fdp0;
2216 struct freetable *ft, *tft;
2217 struct filedescent *fde;
2221 for (i = 0; i <= fdp->fd_lastfile; i++) {
2222 fde = &fdp->fd_ofiles[i];
2227 (void) closef(fp, td);
2233 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2234 free(fdp->fd_map, M_FILEDESC);
2235 if (fdp->fd_nfiles > NDFILE)
2236 free(fdp->fd_files, M_FILEDESC);
2238 fdp0 = (struct filedesc0 *)fdp;
2239 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2240 free(ft->ft_table, M_FILEDESC);
2246 fdescfree(struct thread *td)
2249 struct filedesc *fdp;
2250 struct vnode *cdir, *jdir, *rdir;
2257 if (RACCT_ENABLED())
2258 racct_set_unlocked(p, RACCT_NOFILE, 0);
2261 if (p->p_fdtol != NULL)
2268 if (refcount_release(&fdp->fd_refcnt) == 0)
2271 FILEDESC_XLOCK(fdp);
2272 cdir = fdp->fd_cdir;
2273 fdp->fd_cdir = NULL;
2274 rdir = fdp->fd_rdir;
2275 fdp->fd_rdir = NULL;
2276 jdir = fdp->fd_jdir;
2277 fdp->fd_jdir = NULL;
2278 FILEDESC_XUNLOCK(fdp);
2287 fdescfree_fds(td, fdp, 1);
2291 fdescfree_remapped(struct filedesc *fdp)
2294 if (fdp->fd_cdir != NULL)
2295 vrele(fdp->fd_cdir);
2296 if (fdp->fd_rdir != NULL)
2297 vrele(fdp->fd_rdir);
2298 if (fdp->fd_jdir != NULL)
2299 vrele(fdp->fd_jdir);
2301 fdescfree_fds(curthread, fdp, 0);
2305 * For setugid programs, we don't want to people to use that setugidness
2306 * to generate error messages which write to a file which otherwise would
2307 * otherwise be off-limits to the process. We check for filesystems where
2308 * the vnode can change out from under us after execve (like [lin]procfs).
2310 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2311 * sufficient. We also don't check for setugidness since we know we are.
2314 is_unsafe(struct file *fp)
2318 if (fp->f_type != DTYPE_VNODE)
2322 return ((vp->v_vflag & VV_PROCDEP) != 0);
2326 * Make this setguid thing safe, if at all possible.
2329 fdsetugidsafety(struct thread *td)
2331 struct filedesc *fdp;
2335 fdp = td->td_proc->p_fd;
2336 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2337 MPASS(fdp->fd_nfiles >= 3);
2338 for (i = 0; i <= 2; i++) {
2339 fp = fdp->fd_ofiles[i].fde_file;
2340 if (fp != NULL && is_unsafe(fp)) {
2341 FILEDESC_XLOCK(fdp);
2342 knote_fdclose(td, i);
2344 * NULL-out descriptor prior to close to avoid
2345 * a race while close blocks.
2348 FILEDESC_XUNLOCK(fdp);
2349 (void) closef(fp, td);
2355 * If a specific file object occupies a specific file descriptor, close the
2356 * file descriptor entry and drop a reference on the file object. This is a
2357 * convenience function to handle a subsequent error in a function that calls
2358 * falloc() that handles the race that another thread might have closed the
2359 * file descriptor out from under the thread creating the file object.
2362 fdclose(struct thread *td, struct file *fp, int idx)
2364 struct filedesc *fdp = td->td_proc->p_fd;
2366 FILEDESC_XLOCK(fdp);
2367 if (fdp->fd_ofiles[idx].fde_file == fp) {
2369 FILEDESC_XUNLOCK(fdp);
2372 FILEDESC_XUNLOCK(fdp);
2376 * Close any files on exec?
2379 fdcloseexec(struct thread *td)
2381 struct filedesc *fdp;
2382 struct filedescent *fde;
2386 fdp = td->td_proc->p_fd;
2387 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2388 for (i = 0; i <= fdp->fd_lastfile; i++) {
2389 fde = &fdp->fd_ofiles[i];
2391 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2392 (fde->fde_flags & UF_EXCLOSE))) {
2393 FILEDESC_XLOCK(fdp);
2395 (void) closefp(fdp, i, fp, td, 0);
2396 FILEDESC_UNLOCK_ASSERT(fdp);
2402 * It is unsafe for set[ug]id processes to be started with file
2403 * descriptors 0..2 closed, as these descriptors are given implicit
2404 * significance in the Standard C library. fdcheckstd() will create a
2405 * descriptor referencing /dev/null for each of stdin, stdout, and
2406 * stderr that is not already open.
2409 fdcheckstd(struct thread *td)
2411 struct filedesc *fdp;
2413 int i, error, devnull;
2415 fdp = td->td_proc->p_fd;
2416 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2417 MPASS(fdp->fd_nfiles >= 3);
2419 for (i = 0; i <= 2; i++) {
2420 if (fdp->fd_ofiles[i].fde_file != NULL)
2423 save = td->td_retval[0];
2424 if (devnull != -1) {
2425 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2427 error = kern_openat(td, AT_FDCWD, "/dev/null",
2428 UIO_SYSSPACE, O_RDWR, 0);
2430 devnull = td->td_retval[0];
2431 KASSERT(devnull == i, ("we didn't get our fd"));
2434 td->td_retval[0] = save;
2442 * Internal form of close. Decrement reference count on file structure.
2443 * Note: td may be NULL when closing a file that was being passed in a
2447 closef(struct file *fp, struct thread *td)
2451 struct filedesc_to_leader *fdtol;
2452 struct filedesc *fdp;
2455 * POSIX record locking dictates that any close releases ALL
2456 * locks owned by this process. This is handled by setting
2457 * a flag in the unlock to free ONLY locks obeying POSIX
2458 * semantics, and not to free BSD-style file locks.
2459 * If the descriptor was in a message, POSIX-style locks
2460 * aren't passed with the descriptor, and the thread pointer
2461 * will be NULL. Callers should be careful only to pass a
2462 * NULL thread pointer when there really is no owning
2463 * context that might have locks, or the locks will be
2466 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2468 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2469 lf.l_whence = SEEK_SET;
2472 lf.l_type = F_UNLCK;
2473 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2474 F_UNLCK, &lf, F_POSIX);
2476 fdtol = td->td_proc->p_fdtol;
2477 if (fdtol != NULL) {
2479 * Handle special case where file descriptor table is
2480 * shared between multiple process leaders.
2482 fdp = td->td_proc->p_fd;
2483 FILEDESC_XLOCK(fdp);
2484 for (fdtol = fdtol->fdl_next;
2485 fdtol != td->td_proc->p_fdtol;
2486 fdtol = fdtol->fdl_next) {
2487 if ((fdtol->fdl_leader->p_flag &
2490 fdtol->fdl_holdcount++;
2491 FILEDESC_XUNLOCK(fdp);
2492 lf.l_whence = SEEK_SET;
2495 lf.l_type = F_UNLCK;
2497 (void) VOP_ADVLOCK(vp,
2498 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2500 FILEDESC_XLOCK(fdp);
2501 fdtol->fdl_holdcount--;
2502 if (fdtol->fdl_holdcount == 0 &&
2503 fdtol->fdl_wakeup != 0) {
2504 fdtol->fdl_wakeup = 0;
2508 FILEDESC_XUNLOCK(fdp);
2511 return (fdrop(fp, td));
2515 * Initialize the file pointer with the specified properties.
2517 * The ops are set with release semantics to be certain that the flags, type,
2518 * and data are visible when ops is. This is to prevent ops methods from being
2519 * called with bad data.
2522 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2527 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2531 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2532 struct file **fpp, struct filecaps *havecapsp)
2534 struct filedescent *fde;
2537 FILEDESC_LOCK_ASSERT(fdp);
2539 fde = fdeget_locked(fdp, fd);
2546 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2551 if (havecapsp != NULL)
2552 filecaps_copy(&fde->fde_caps, havecapsp, true);
2554 *fpp = fde->fde_file;
2562 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2563 struct file **fpp, struct filecaps *havecapsp)
2565 struct filedesc *fdp = td->td_proc->p_fd;
2567 #ifndef CAPABILITIES
2568 error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2569 if (error == 0 && havecapsp != NULL)
2570 filecaps_fill(havecapsp);
2576 error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2580 if (havecapsp != NULL) {
2581 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2582 havecapsp, false)) {
2588 if (!fd_modified(fdp, fd, seq))
2597 FILEDESC_SLOCK(fdp);
2598 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2601 FILEDESC_SUNLOCK(fdp);
2607 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2608 struct file **fpp, seqc_t *seqp)
2611 const struct filedescent *fde;
2613 const struct fdescenttbl *fdt;
2618 cap_rights_t haverights;
2622 fdt = fdp->fd_files;
2623 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2626 * Fetch the descriptor locklessly. We avoid fdrop() races by
2627 * never raising a refcount above 0. To accomplish this we have
2628 * to use a cmpset loop rather than an atomic_add. The descriptor
2629 * must be re-verified once we acquire a reference to be certain
2630 * that the identity is still correct and we did not lose a race
2631 * due to preemption.
2635 seq = seqc_read(fd_seqc(fdt, fd));
2636 fde = &fdt->fdt_ofiles[fd];
2637 haverights = *cap_rights_fde_inline(fde);
2639 if (!seqc_consistent(fd_seqc(fdt, fd), seq))
2642 fp = fdt->fdt_ofiles[fd].fde_file;
2647 error = cap_check(&haverights, needrightsp);
2651 count = fp->f_count;
2655 * Force a reload. Other thread could reallocate the
2656 * table before this fd was closed, so it possible that
2657 * there is a stale fp pointer in cached version.
2659 fdt = *(const struct fdescenttbl * const volatile *)&(fdp->fd_files);
2663 * Use an acquire barrier to force re-reading of fdt so it is
2664 * refreshed for verification.
2666 if (atomic_fcmpset_acq_int(&fp->f_count, &count, count + 1) == 0)
2668 fdt = fdp->fd_files;
2670 if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
2672 if (fp == fdt->fdt_ofiles[fd].fde_file)
2675 fdrop(fp, curthread);
2687 * Extract the file pointer associated with the specified descriptor for the
2688 * current user process.
2690 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2693 * File's rights will be checked against the capability rights mask.
2695 * If an error occurred the non-zero error is returned and *fpp is set to
2696 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
2697 * responsible for fdrop().
2700 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2701 cap_rights_t *needrightsp, seqc_t *seqp)
2703 struct filedesc *fdp;
2708 fdp = td->td_proc->p_fd;
2709 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2712 if (fp->f_ops == &badfileops) {
2718 * FREAD and FWRITE failure return EBADF as per POSIX.
2724 if ((fp->f_flag & flags) == 0)
2728 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2729 ((fp->f_flag & FWRITE) != 0))
2735 KASSERT(0, ("wrong flags"));
2748 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2751 return (_fget(td, fd, fpp, 0, rightsp, NULL));
2755 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2759 #ifndef CAPABILITIES
2760 error = _fget(td, fd, fpp, 0, rightsp, NULL);
2761 if (maxprotp != NULL)
2762 *maxprotp = VM_PROT_ALL;
2764 cap_rights_t fdrights;
2765 struct filedesc *fdp = td->td_proc->p_fd;
2768 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2770 error = _fget(td, fd, fpp, 0, rightsp, &seq);
2773 if (maxprotp != NULL)
2774 fdrights = *cap_rights(fdp, fd);
2775 if (!fd_modified(fdp, fd, seq))
2781 * If requested, convert capability rights to access flags.
2783 if (maxprotp != NULL)
2784 *maxprotp = cap_rights_to_vmprot(&fdrights);
2790 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2793 return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2797 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2800 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2804 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2807 struct filedesc *fdp = td->td_proc->p_fd;
2808 #ifndef CAPABILITIES
2809 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2814 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2816 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2819 error = cap_fcntl_check(fdp, fd, needfcntl);
2820 if (!fd_modified(fdp, fd, seq))
2833 * Like fget() but loads the underlying vnode, or returns an error if the
2834 * descriptor does not represent a vnode. Note that pipes use vnodes but
2835 * never have VM objects. The returned vnode will be vref()'d.
2837 * XXX: what about the unused flags ?
2840 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2847 error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2850 if (fp->f_vnode == NULL) {
2862 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2865 return (_fgetvp(td, fd, 0, rightsp, vpp));
2869 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2870 struct filecaps *havecaps, struct vnode **vpp)
2872 struct filedesc *fdp;
2873 struct filecaps caps;
2877 fdp = td->td_proc->p_fd;
2878 error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2881 if (fp->f_ops == &badfileops) {
2885 if (fp->f_vnode == NULL) {
2896 filecaps_free(&caps);
2901 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2904 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2908 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2911 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2916 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2920 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2925 * Handle the last reference to a file being closed.
2927 * Without the noinline attribute clang keeps inlining the func thorough this
2928 * file when fdrop is used.
2931 _fdrop(struct file *fp, struct thread *td)
2935 if (fp->f_count != 0)
2936 panic("fdrop: count %d", fp->f_count);
2937 error = fo_close(fp, td);
2938 atomic_subtract_int(&openfiles, 1);
2940 free(fp->f_advice, M_FADVISE);
2941 uma_zfree(file_zone, fp);
2947 * Apply an advisory lock on a file descriptor.
2949 * Just attempt to get a record lock of the requested type on the entire file
2950 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2952 #ifndef _SYS_SYSPROTO_H_
2960 sys_flock(struct thread *td, struct flock_args *uap)
2967 error = fget(td, uap->fd, &cap_flock_rights, &fp);
2970 if (fp->f_type != DTYPE_VNODE) {
2972 return (EOPNOTSUPP);
2976 lf.l_whence = SEEK_SET;
2979 if (uap->how & LOCK_UN) {
2980 lf.l_type = F_UNLCK;
2981 atomic_clear_int(&fp->f_flag, FHASLOCK);
2982 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2985 if (uap->how & LOCK_EX)
2986 lf.l_type = F_WRLCK;
2987 else if (uap->how & LOCK_SH)
2988 lf.l_type = F_RDLCK;
2993 atomic_set_int(&fp->f_flag, FHASLOCK);
2994 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2995 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3001 * Duplicate the specified descriptor to a free descriptor.
3004 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3005 int openerror, int *indxp)
3007 struct filedescent *newfde, *oldfde;
3012 KASSERT(openerror == ENODEV || openerror == ENXIO,
3013 ("unexpected error %d in %s", openerror, __func__));
3016 * If the to-be-dup'd fd number is greater than the allowed number
3017 * of file descriptors, or the fd to be dup'd has already been
3018 * closed, then reject.
3020 FILEDESC_XLOCK(fdp);
3021 if ((fp = fget_locked(fdp, dfd)) == NULL) {
3022 FILEDESC_XUNLOCK(fdp);
3026 error = fdalloc(td, 0, &indx);
3028 FILEDESC_XUNLOCK(fdp);
3033 * There are two cases of interest here.
3035 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3037 * For ENXIO steal away the file structure from (dfd) and store it in
3038 * (indx). (dfd) is effectively closed by this operation.
3040 switch (openerror) {
3043 * Check that the mode the file is being opened for is a
3044 * subset of the mode of the existing descriptor.
3046 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3047 fdunused(fdp, indx);
3048 FILEDESC_XUNLOCK(fdp);
3052 newfde = &fdp->fd_ofiles[indx];
3053 oldfde = &fdp->fd_ofiles[dfd];
3054 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3056 seqc_write_begin(&newfde->fde_seqc);
3058 memcpy(newfde, oldfde, fde_change_size);
3059 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3062 seqc_write_end(&newfde->fde_seqc);
3067 * Steal away the file pointer from dfd and stuff it into indx.
3069 newfde = &fdp->fd_ofiles[indx];
3070 oldfde = &fdp->fd_ofiles[dfd];
3072 seqc_write_begin(&newfde->fde_seqc);
3074 memcpy(newfde, oldfde, fde_change_size);
3075 oldfde->fde_file = NULL;
3078 seqc_write_end(&newfde->fde_seqc);
3082 FILEDESC_XUNLOCK(fdp);
3088 * This sysctl determines if we will allow a process to chroot(2) if it
3089 * has a directory open:
3090 * 0: disallowed for all processes.
3091 * 1: allowed for processes that were not already chroot(2)'ed.
3092 * 2: allowed for all processes.
3095 static int chroot_allow_open_directories = 1;
3097 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3098 &chroot_allow_open_directories, 0,
3099 "Allow a process to chroot(2) if it has a directory open");
3102 * Helper function for raised chroot(2) security function: Refuse if
3103 * any filedescriptors are open directories.
3106 chroot_refuse_vdir_fds(struct filedesc *fdp)
3112 FILEDESC_LOCK_ASSERT(fdp);
3114 for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3115 fp = fget_locked(fdp, fd);
3118 if (fp->f_type == DTYPE_VNODE) {
3120 if (vp->v_type == VDIR)
3128 * Common routine for kern_chroot() and jail_attach(). The caller is
3129 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3130 * authorize this operation.
3133 pwd_chroot(struct thread *td, struct vnode *vp)
3135 struct filedesc *fdp;
3136 struct vnode *oldvp;
3139 fdp = td->td_proc->p_fd;
3140 FILEDESC_XLOCK(fdp);
3141 if (chroot_allow_open_directories == 0 ||
3142 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3143 error = chroot_refuse_vdir_fds(fdp);
3145 FILEDESC_XUNLOCK(fdp);
3149 oldvp = fdp->fd_rdir;
3152 if (fdp->fd_jdir == NULL) {
3156 FILEDESC_XUNLOCK(fdp);
3162 pwd_chdir(struct thread *td, struct vnode *vp)
3164 struct filedesc *fdp;
3165 struct vnode *oldvp;
3167 fdp = td->td_proc->p_fd;
3168 FILEDESC_XLOCK(fdp);
3169 VNASSERT(vp->v_usecount > 0, vp,
3170 ("chdir to a vnode with zero usecount"));
3171 oldvp = fdp->fd_cdir;
3173 FILEDESC_XUNLOCK(fdp);
3178 * Scan all active processes and prisons to see if any of them have a current
3179 * or root directory of `olddp'. If so, replace them with the new mount point.
3182 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3184 struct filedesc *fdp;
3189 if (vrefcnt(olddp) == 1)
3192 sx_slock(&allproc_lock);
3193 FOREACH_PROC_IN_SYSTEM(p) {
3199 FILEDESC_XLOCK(fdp);
3200 if (fdp->fd_cdir == olddp) {
3202 fdp->fd_cdir = newdp;
3205 if (fdp->fd_rdir == olddp) {
3207 fdp->fd_rdir = newdp;
3210 if (fdp->fd_jdir == olddp) {
3212 fdp->fd_jdir = newdp;
3215 FILEDESC_XUNLOCK(fdp);
3218 sx_sunlock(&allproc_lock);
3219 if (rootvnode == olddp) {
3224 mtx_lock(&prison0.pr_mtx);
3225 if (prison0.pr_root == olddp) {
3227 prison0.pr_root = newdp;
3230 mtx_unlock(&prison0.pr_mtx);
3231 sx_slock(&allprison_lock);
3232 TAILQ_FOREACH(pr, &allprison, pr_list) {
3233 mtx_lock(&pr->pr_mtx);
3234 if (pr->pr_root == olddp) {
3236 pr->pr_root = newdp;
3239 mtx_unlock(&pr->pr_mtx);
3241 sx_sunlock(&allprison_lock);
3246 struct filedesc_to_leader *
3247 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3249 struct filedesc_to_leader *fdtol;
3251 fdtol = malloc(sizeof(struct filedesc_to_leader),
3252 M_FILEDESC_TO_LEADER, M_WAITOK);
3253 fdtol->fdl_refcount = 1;
3254 fdtol->fdl_holdcount = 0;
3255 fdtol->fdl_wakeup = 0;
3256 fdtol->fdl_leader = leader;
3258 FILEDESC_XLOCK(fdp);
3259 fdtol->fdl_next = old->fdl_next;
3260 fdtol->fdl_prev = old;
3261 old->fdl_next = fdtol;
3262 fdtol->fdl_next->fdl_prev = fdtol;
3263 FILEDESC_XUNLOCK(fdp);
3265 fdtol->fdl_next = fdtol;
3266 fdtol->fdl_prev = fdtol;
3272 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3274 struct filedesc *fdp;
3275 int i, count, slots;
3277 if (*(int *)arg1 != 0)
3280 fdp = curproc->p_fd;
3282 FILEDESC_SLOCK(fdp);
3283 slots = NDSLOTS(fdp->fd_lastfile + 1);
3284 for (i = 0; i < slots; i++)
3285 count += bitcountl(fdp->fd_map[i]);
3286 FILEDESC_SUNLOCK(fdp);
3288 return (SYSCTL_OUT(req, &count, sizeof(count)));
3291 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3292 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3293 "Number of open file descriptors");
3296 * Get file structures globally.
3299 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3302 struct filedesc *fdp;
3307 error = sysctl_wire_old_buffer(req, 0);
3310 if (req->oldptr == NULL) {
3312 sx_slock(&allproc_lock);
3313 FOREACH_PROC_IN_SYSTEM(p) {
3315 if (p->p_state == PRS_NEW) {
3323 /* overestimates sparse tables. */
3324 if (fdp->fd_lastfile > 0)
3325 n += fdp->fd_lastfile;
3328 sx_sunlock(&allproc_lock);
3329 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3332 bzero(&xf, sizeof(xf));
3333 xf.xf_size = sizeof(xf);
3334 sx_slock(&allproc_lock);
3335 FOREACH_PROC_IN_SYSTEM(p) {
3337 if (p->p_state == PRS_NEW) {
3341 if (p_cansee(req->td, p) != 0) {
3345 xf.xf_pid = p->p_pid;
3346 xf.xf_uid = p->p_ucred->cr_uid;
3351 FILEDESC_SLOCK(fdp);
3352 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3353 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3356 xf.xf_file = (uintptr_t)fp;
3357 xf.xf_data = (uintptr_t)fp->f_data;
3358 xf.xf_vnode = (uintptr_t)fp->f_vnode;
3359 xf.xf_type = (uintptr_t)fp->f_type;
3360 xf.xf_count = fp->f_count;
3362 xf.xf_offset = foffset_get(fp);
3363 xf.xf_flag = fp->f_flag;
3364 error = SYSCTL_OUT(req, &xf, sizeof(xf));
3368 FILEDESC_SUNLOCK(fdp);
3373 sx_sunlock(&allproc_lock);
3377 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3378 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3380 #ifdef KINFO_FILE_SIZE
3381 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3385 xlate_fflags(int fflags)
3387 static const struct {
3390 } fflags_table[] = {
3391 { FAPPEND, KF_FLAG_APPEND },
3392 { FASYNC, KF_FLAG_ASYNC },
3393 { FFSYNC, KF_FLAG_FSYNC },
3394 { FHASLOCK, KF_FLAG_HASLOCK },
3395 { FNONBLOCK, KF_FLAG_NONBLOCK },
3396 { FREAD, KF_FLAG_READ },
3397 { FWRITE, KF_FLAG_WRITE },
3398 { O_CREAT, KF_FLAG_CREAT },
3399 { O_DIRECT, KF_FLAG_DIRECT },
3400 { O_EXCL, KF_FLAG_EXCL },
3401 { O_EXEC, KF_FLAG_EXEC },
3402 { O_EXLOCK, KF_FLAG_EXLOCK },
3403 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3404 { O_SHLOCK, KF_FLAG_SHLOCK },
3405 { O_TRUNC, KF_FLAG_TRUNC }
3411 for (i = 0; i < nitems(fflags_table); i++)
3412 if (fflags & fflags_table[i].fflag)
3413 kflags |= fflags_table[i].kf_fflag;
3417 /* Trim unused data from kf_path by truncating the structure size. */
3419 pack_kinfo(struct kinfo_file *kif)
3422 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3423 strlen(kif->kf_path) + 1;
3424 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3428 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3429 struct kinfo_file *kif, struct filedesc *fdp, int flags)
3433 bzero(kif, sizeof(*kif));
3435 /* Set a default type to allow for empty fill_kinfo() methods. */
3436 kif->kf_type = KF_TYPE_UNKNOWN;
3437 kif->kf_flags = xlate_fflags(fp->f_flag);
3438 if (rightsp != NULL)
3439 kif->kf_cap_rights = *rightsp;
3441 cap_rights_init(&kif->kf_cap_rights);
3443 kif->kf_ref_count = fp->f_count;
3444 kif->kf_offset = foffset_get(fp);
3447 * This may drop the filedesc lock, so the 'fp' cannot be
3448 * accessed after this call.
3450 error = fo_fill_kinfo(fp, kif, fdp);
3452 kif->kf_status |= KF_ATTR_VALID;
3453 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3456 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3460 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3461 struct kinfo_file *kif, int flags)
3465 bzero(kif, sizeof(*kif));
3467 kif->kf_type = KF_TYPE_VNODE;
3468 error = vn_fill_kinfo_vnode(vp, kif);
3470 kif->kf_status |= KF_ATTR_VALID;
3471 kif->kf_flags = xlate_fflags(fflags);
3472 cap_rights_init(&kif->kf_cap_rights);
3474 kif->kf_ref_count = -1;
3475 kif->kf_offset = -1;
3476 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3479 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3483 struct export_fd_buf {
3484 struct filedesc *fdp;
3487 struct kinfo_file kif;
3492 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3494 struct kinfo_file *kif;
3497 if (efbuf->remainder != -1) {
3498 if (efbuf->remainder < kif->kf_structsize) {
3499 /* Terminate export. */
3500 efbuf->remainder = 0;
3503 efbuf->remainder -= kif->kf_structsize;
3505 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3509 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3510 struct export_fd_buf *efbuf)
3514 if (efbuf->remainder == 0)
3516 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3518 FILEDESC_SUNLOCK(efbuf->fdp);
3519 error = export_kinfo_to_sb(efbuf);
3520 FILEDESC_SLOCK(efbuf->fdp);
3525 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3526 struct export_fd_buf *efbuf)
3530 if (efbuf->remainder == 0)
3532 if (efbuf->fdp != NULL)
3533 FILEDESC_SUNLOCK(efbuf->fdp);
3534 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3535 error = export_kinfo_to_sb(efbuf);
3536 if (efbuf->fdp != NULL)
3537 FILEDESC_SLOCK(efbuf->fdp);
3542 * Store a process file descriptor information to sbuf.
3544 * Takes a locked proc as argument, and returns with the proc unlocked.
3547 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
3551 struct filedesc *fdp;
3552 struct export_fd_buf *efbuf;
3553 struct vnode *cttyvp, *textvp, *tracevp;
3555 cap_rights_t rights;
3557 PROC_LOCK_ASSERT(p, MA_OWNED);
3560 tracevp = p->p_tracevp;
3561 if (tracevp != NULL)
3564 textvp = p->p_textvp;
3567 /* Controlling tty. */
3569 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3570 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3576 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3579 efbuf->remainder = maxlen;
3580 efbuf->flags = flags;
3581 if (tracevp != NULL)
3582 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3585 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3587 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3593 FILEDESC_SLOCK(fdp);
3594 /* working directory */
3595 if (fdp->fd_cdir != NULL) {
3596 vrefact(fdp->fd_cdir);
3597 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3599 /* root directory */
3600 if (fdp->fd_rdir != NULL) {
3601 vrefact(fdp->fd_rdir);
3602 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3604 /* jail directory */
3605 if (fdp->fd_jdir != NULL) {
3606 vrefact(fdp->fd_jdir);
3607 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3609 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3610 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3613 rights = *cap_rights(fdp, i);
3614 #else /* !CAPABILITIES */
3615 rights = cap_no_rights;
3618 * Create sysctl entry. It is OK to drop the filedesc
3619 * lock inside of export_file_to_sb() as we will
3620 * re-validate and re-evaluate its properties when the
3623 error = export_file_to_sb(fp, i, &rights, efbuf);
3624 if (error != 0 || efbuf->remainder == 0)
3627 FILEDESC_SUNLOCK(fdp);
3630 free(efbuf, M_TEMP);
3634 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
3637 * Get per-process file descriptors for use by procstat(1), et al.
3640 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3645 int error, error2, *name;
3649 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3650 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3651 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3656 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3657 error = kern_proc_filedesc_out(p, &sb, maxlen,
3658 KERN_FILEDESC_PACK_KINFO);
3659 error2 = sbuf_finish(&sb);
3661 return (error != 0 ? error : error2);
3664 #ifdef COMPAT_FREEBSD7
3665 #ifdef KINFO_OFILE_SIZE
3666 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3670 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3673 okif->kf_structsize = sizeof(*okif);
3674 okif->kf_type = kif->kf_type;
3675 okif->kf_fd = kif->kf_fd;
3676 okif->kf_ref_count = kif->kf_ref_count;
3677 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3678 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3679 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3680 okif->kf_offset = kif->kf_offset;
3681 if (kif->kf_type == KF_TYPE_VNODE)
3682 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
3684 okif->kf_vnode_type = KF_VTYPE_VNON;
3685 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3686 if (kif->kf_type == KF_TYPE_SOCKET) {
3687 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
3688 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
3689 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
3690 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
3691 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
3693 okif->kf_sa_local.ss_family = AF_UNSPEC;
3694 okif->kf_sa_peer.ss_family = AF_UNSPEC;
3699 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3700 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3705 FILEDESC_SUNLOCK(fdp);
3706 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3707 kinfo_to_okinfo(kif, okif);
3708 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3709 FILEDESC_SLOCK(fdp);
3714 * Get per-process file descriptors for use by procstat(1), et al.
3717 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3719 struct kinfo_ofile *okif;
3720 struct kinfo_file *kif;
3721 struct filedesc *fdp;
3722 int error, i, *name;
3727 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3734 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3735 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3736 FILEDESC_SLOCK(fdp);
3737 if (fdp->fd_cdir != NULL)
3738 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3740 if (fdp->fd_rdir != NULL)
3741 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3743 if (fdp->fd_jdir != NULL)
3744 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3746 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3747 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3749 export_file_to_kinfo(fp, i, NULL, kif, fdp,
3750 KERN_FILEDESC_PACK_KINFO);
3751 FILEDESC_SUNLOCK(fdp);
3752 kinfo_to_okinfo(kif, okif);
3753 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3754 FILEDESC_SLOCK(fdp);
3758 FILEDESC_SUNLOCK(fdp);
3765 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3766 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3767 "Process ofiledesc entries");
3768 #endif /* COMPAT_FREEBSD7 */
3771 vntype_to_kinfo(int vtype)
3776 } vtypes_table[] = {
3777 { VBAD, KF_VTYPE_VBAD },
3778 { VBLK, KF_VTYPE_VBLK },
3779 { VCHR, KF_VTYPE_VCHR },
3780 { VDIR, KF_VTYPE_VDIR },
3781 { VFIFO, KF_VTYPE_VFIFO },
3782 { VLNK, KF_VTYPE_VLNK },
3783 { VNON, KF_VTYPE_VNON },
3784 { VREG, KF_VTYPE_VREG },
3785 { VSOCK, KF_VTYPE_VSOCK }
3790 * Perform vtype translation.
3792 for (i = 0; i < nitems(vtypes_table); i++)
3793 if (vtypes_table[i].vtype == vtype)
3794 return (vtypes_table[i].kf_vtype);
3796 return (KF_VTYPE_UNKNOWN);
3799 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3800 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3801 "Process filedesc entries");
3804 * Store a process current working directory information to sbuf.
3806 * Takes a locked proc as argument, and returns with the proc unlocked.
3809 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
3811 struct filedesc *fdp;
3812 struct export_fd_buf *efbuf;
3815 PROC_LOCK_ASSERT(p, MA_OWNED);
3822 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3825 efbuf->remainder = maxlen;
3827 FILEDESC_SLOCK(fdp);
3828 if (fdp->fd_cdir == NULL)
3831 vrefact(fdp->fd_cdir);
3832 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3835 FILEDESC_SUNLOCK(fdp);
3837 free(efbuf, M_TEMP);
3842 * Get per-process current working directory.
3845 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3850 int error, error2, *name;
3854 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3855 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3856 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3861 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3862 error = kern_proc_cwd_out(p, &sb, maxlen);
3863 error2 = sbuf_finish(&sb);
3865 return (error != 0 ? error : error2);
3868 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3869 sysctl_kern_proc_cwd, "Process current working directory");
3873 * For the purposes of debugging, generate a human-readable string for the
3877 file_type_to_name(short type)
3905 case DTYPE_PROCDESC:
3907 case DTYPE_LINUXEFD:
3909 case DTYPE_LINUXTFD:
3917 * For the purposes of debugging, identify a process (if any, perhaps one of
3918 * many) that references the passed file in its file descriptor array. Return
3921 static struct proc *
3922 file_to_first_proc(struct file *fp)
3924 struct filedesc *fdp;
3928 FOREACH_PROC_IN_SYSTEM(p) {
3929 if (p->p_state == PRS_NEW)
3934 for (n = 0; n <= fdp->fd_lastfile; n++) {
3935 if (fp == fdp->fd_ofiles[n].fde_file)
3943 db_print_file(struct file *fp, int header)
3945 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
3949 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
3950 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
3951 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
3953 p = file_to_first_proc(fp);
3954 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
3955 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
3956 fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode,
3957 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3962 DB_SHOW_COMMAND(file, db_show_file)
3967 db_printf("usage: show file <addr>\n");
3970 fp = (struct file *)addr;
3971 db_print_file(fp, 1);
3974 DB_SHOW_COMMAND(files, db_show_files)
3976 struct filedesc *fdp;
3983 FOREACH_PROC_IN_SYSTEM(p) {
3984 if (p->p_state == PRS_NEW)
3986 if ((fdp = p->p_fd) == NULL)
3988 for (n = 0; n <= fdp->fd_lastfile; ++n) {
3989 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3991 db_print_file(fp, header);
3998 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3999 &maxfilesperproc, 0, "Maximum files allowed open per process");
4001 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4002 &maxfiles, 0, "Maximum number of files");
4004 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4005 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
4009 filelistinit(void *dummy)
4012 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4013 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4014 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4015 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4016 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4018 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4020 /*-------------------------------------------------------------------*/
4023 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4024 int flags, struct thread *td)
4031 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4039 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4047 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4055 badfo_kqfilter(struct file *fp, struct knote *kn)
4062 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4070 badfo_close(struct file *fp, struct thread *td)
4077 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4085 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4093 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4094 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4102 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4108 struct fileops badfileops = {
4109 .fo_read = badfo_readwrite,
4110 .fo_write = badfo_readwrite,
4111 .fo_truncate = badfo_truncate,
4112 .fo_ioctl = badfo_ioctl,
4113 .fo_poll = badfo_poll,
4114 .fo_kqfilter = badfo_kqfilter,
4115 .fo_stat = badfo_stat,
4116 .fo_close = badfo_close,
4117 .fo_chmod = badfo_chmod,
4118 .fo_chown = badfo_chown,
4119 .fo_sendfile = badfo_sendfile,
4120 .fo_fill_kinfo = badfo_fill_kinfo,
4124 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4125 int flags, struct thread *td)
4128 return (EOPNOTSUPP);
4132 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4140 invfo_ioctl(struct file *fp, u_long com, void *data,
4141 struct ucred *active_cred, struct thread *td)
4148 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4152 return (poll_no_poll(events));
4156 invfo_kqfilter(struct file *fp, struct knote *kn)
4163 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4171 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4179 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4180 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4187 /*-------------------------------------------------------------------*/
4190 * File Descriptor pseudo-device driver (/dev/fd/).
4192 * Opening minor device N dup()s the file (if any) connected to file
4193 * descriptor N belonging to the calling process. Note that this driver
4194 * consists of only the ``open()'' routine, because all subsequent
4195 * references to this file will be direct to the other driver.
4197 * XXX: we could give this one a cloning event handler if necessary.
4202 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4206 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4207 * the file descriptor being sought for duplication. The error
4208 * return ensures that the vnode for this device will be released
4209 * by vn_open. Open will detect this special error and take the
4210 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4211 * will simply report the error.
4213 td->td_dupfd = dev2unit(dev);
4217 static struct cdevsw fildesc_cdevsw = {
4218 .d_version = D_VERSION,
4224 fildesc_drvinit(void *unused)
4228 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4229 UID_ROOT, GID_WHEEL, 0666, "fd/0");
4230 make_dev_alias(dev, "stdin");
4231 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4232 UID_ROOT, GID_WHEEL, 0666, "fd/1");
4233 make_dev_alias(dev, "stdout");
4234 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4235 UID_ROOT, GID_WHEEL, 0666, "fd/2");
4236 make_dev_alias(dev, "stderr");
4239 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);