2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1989, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
42 #include "opt_capsicum.h"
44 #include "opt_ktrace.h"
46 #include <sys/param.h>
47 #include <sys/systm.h>
49 #include <sys/capsicum.h>
51 #include <sys/fcntl.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
66 #include <sys/protosw.h>
67 #include <sys/racct.h>
68 #include <sys/resourcevar.h>
70 #include <sys/signalvar.h>
74 #include <sys/syscallsubr.h>
75 #include <sys/sysctl.h>
76 #include <sys/sysproto.h>
77 #include <sys/unistd.h>
79 #include <sys/vnode.h>
81 #include <sys/ktrace.h>
86 #include <security/audit/audit.h>
93 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
94 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
95 "file desc to leader structures");
96 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
97 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
99 MALLOC_DECLARE(M_FADVISE);
101 static __read_mostly uma_zone_t file_zone;
102 static __read_mostly uma_zone_t filedesc0_zone;
104 static int closefp(struct filedesc *fdp, int fd, struct file *fp,
105 struct thread *td, int holdleaders);
106 static int fd_first_free(struct filedesc *fdp, int low, int size);
107 static int fd_last_used(struct filedesc *fdp, int size);
108 static void fdgrowtable(struct filedesc *fdp, int nfd);
109 static void fdgrowtable_exp(struct filedesc *fdp, int nfd);
110 static void fdunused(struct filedesc *fdp, int fd);
111 static void fdused(struct filedesc *fdp, int fd);
112 static int getmaxfd(struct thread *td);
113 static u_long *filecaps_copy_prep(const struct filecaps *src);
114 static void filecaps_copy_finish(const struct filecaps *src,
115 struct filecaps *dst, u_long *ioctls);
116 static u_long *filecaps_free_prep(struct filecaps *fcaps);
117 static void filecaps_free_finish(u_long *ioctls);
122 * - An array of open file descriptors (fd_ofiles)
123 * - An array of file flags (fd_ofileflags)
124 * - A bitmap recording which descriptors are in use (fd_map)
126 * A process starts out with NDFILE descriptors. The value of NDFILE has
127 * been selected based the historical limit of 20 open files, and an
128 * assumption that the majority of processes, especially short-lived
129 * processes like shells, will never need more.
131 * If this initial allocation is exhausted, a larger descriptor table and
132 * map are allocated dynamically, and the pointers in the process's struct
133 * filedesc are updated to point to those. This is repeated every time
134 * the process runs out of file descriptors (provided it hasn't hit its
137 * Since threads may hold references to individual descriptor table
138 * entries, the tables are never freed. Instead, they are placed on a
139 * linked list and freed only when the struct filedesc is released.
142 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
143 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
144 #define NDSLOT(x) ((x) / NDENTRIES)
145 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
146 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
149 * SLIST entry used to keep track of ofiles which must be reclaimed when
153 struct fdescenttbl *ft_table;
154 SLIST_ENTRY(freetable) ft_next;
158 * Initial allocation: a filedesc structure + the head of SLIST used to
159 * keep track of old ofiles + enough space for NDFILE descriptors.
162 struct fdescenttbl0 {
164 struct filedescent fdt_ofiles[NDFILE];
168 struct filedesc fd_fd;
169 SLIST_HEAD(, freetable) fd_free;
170 struct fdescenttbl0 fd_dfiles;
171 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
175 * Descriptor management.
177 volatile int __exclusive_cache_line openfiles; /* actual number of open files */
178 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
179 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
182 * If low >= size, just return low. Otherwise find the first zero bit in the
183 * given bitmap, starting at low and not exceeding size - 1. Return size if
187 fd_first_free(struct filedesc *fdp, int low, int size)
189 NDSLOTTYPE *map = fdp->fd_map;
197 if (low % NDENTRIES) {
198 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
199 if ((mask &= ~map[off]) != 0UL)
200 return (off * NDENTRIES + ffsl(mask) - 1);
203 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
204 if (map[off] != ~0UL)
205 return (off * NDENTRIES + ffsl(~map[off]) - 1);
210 * Find the highest non-zero bit in the given bitmap, starting at 0 and
211 * not exceeding size - 1. Return -1 if not found.
214 fd_last_used(struct filedesc *fdp, int size)
216 NDSLOTTYPE *map = fdp->fd_map;
221 if (size % NDENTRIES) {
222 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
223 if ((mask &= map[off]) != 0)
224 return (off * NDENTRIES + flsl(mask) - 1);
227 for (minoff = NDSLOT(0); off >= minoff; --off)
229 return (off * NDENTRIES + flsl(map[off]) - 1);
234 fdisused(struct filedesc *fdp, int fd)
237 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
238 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
240 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
244 * Mark a file descriptor as used.
247 fdused_init(struct filedesc *fdp, int fd)
250 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
252 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
256 fdused(struct filedesc *fdp, int fd)
259 FILEDESC_XLOCK_ASSERT(fdp);
261 fdused_init(fdp, fd);
262 if (fd > fdp->fd_lastfile)
263 fdp->fd_lastfile = fd;
264 if (fd == fdp->fd_freefile)
265 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
269 * Mark a file descriptor as unused.
272 fdunused(struct filedesc *fdp, int fd)
275 FILEDESC_XLOCK_ASSERT(fdp);
277 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
278 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
279 ("fd=%d is still in use", fd));
281 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
282 if (fd < fdp->fd_freefile)
283 fdp->fd_freefile = fd;
284 if (fd == fdp->fd_lastfile)
285 fdp->fd_lastfile = fd_last_used(fdp, fd);
289 * Free a file descriptor.
291 * Avoid some work if fdp is about to be destroyed.
294 fdefree_last(struct filedescent *fde)
297 filecaps_free(&fde->fde_caps);
301 fdfree(struct filedesc *fdp, int fd)
303 struct filedescent *fde;
305 fde = &fdp->fd_ofiles[fd];
307 seq_write_begin(&fde->fde_seq);
309 fde->fde_file = NULL;
311 seq_write_end(&fde->fde_seq);
318 pwd_ensure_dirs(void)
320 struct filedesc *fdp;
324 if (fdp->fd_cdir == NULL) {
325 fdp->fd_cdir = rootvnode;
328 if (fdp->fd_rdir == NULL) {
329 fdp->fd_rdir = rootvnode;
332 FILEDESC_XUNLOCK(fdp);
336 * System calls on descriptors.
338 #ifndef _SYS_SYSPROTO_H_
339 struct getdtablesize_args {
345 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
352 min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc);
354 PROC_LOCK(td->td_proc);
355 lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
356 PROC_UNLOCK(td->td_proc);
357 if (lim < td->td_retval[0])
358 td->td_retval[0] = lim;
364 * Duplicate a file descriptor to a particular value.
366 * Note: keep in mind that a potential race condition exists when closing
367 * descriptors from a shared descriptor table (via rfork).
369 #ifndef _SYS_SYSPROTO_H_
377 sys_dup2(struct thread *td, struct dup2_args *uap)
380 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
384 * Duplicate a file descriptor.
386 #ifndef _SYS_SYSPROTO_H_
393 sys_dup(struct thread *td, struct dup_args *uap)
396 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
400 * The file control system call.
402 #ifndef _SYS_SYSPROTO_H_
411 sys_fcntl(struct thread *td, struct fcntl_args *uap)
414 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
418 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
432 * Convert old flock structure to new.
434 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
435 fl.l_start = ofl.l_start;
436 fl.l_len = ofl.l_len;
437 fl.l_pid = ofl.l_pid;
438 fl.l_type = ofl.l_type;
439 fl.l_whence = ofl.l_whence;
453 arg1 = (intptr_t)&fl;
459 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
460 arg1 = (intptr_t)&fl;
468 error = kern_fcntl(td, fd, newcmd, arg1);
471 if (cmd == F_OGETLK) {
472 ofl.l_start = fl.l_start;
473 ofl.l_len = fl.l_len;
474 ofl.l_pid = fl.l_pid;
475 ofl.l_type = fl.l_type;
476 ofl.l_whence = fl.l_whence;
477 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
478 } else if (cmd == F_GETLK) {
479 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
485 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
487 struct filedesc *fdp;
489 struct file *fp, *fp2;
490 struct filedescent *fde;
507 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
510 case F_DUPFD_CLOEXEC:
512 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
517 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
520 case F_DUP2FD_CLOEXEC:
522 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
528 fde = fdeget_locked(fdp, fd);
531 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
534 FILEDESC_SUNLOCK(fdp);
540 fde = fdeget_locked(fdp, fd);
542 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
543 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
546 FILEDESC_XUNLOCK(fdp);
550 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
553 td->td_retval[0] = OFLAGS(fp->f_flag);
558 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
562 tmp = flg = fp->f_flag;
564 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
565 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
566 tmp = fp->f_flag & FNONBLOCK;
567 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
572 tmp = fp->f_flag & FASYNC;
573 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
578 atomic_clear_int(&fp->f_flag, FNONBLOCK);
580 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
585 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
588 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
590 td->td_retval[0] = tmp;
595 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
599 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
604 error = priv_check(td, PRIV_NFS_LOCKD);
612 /* FALLTHROUGH F_SETLK */
616 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL);
619 if (fp->f_type != DTYPE_VNODE) {
625 flp = (struct flock *)arg;
626 if (flp->l_whence == SEEK_CUR) {
627 foffset = foffset_get(fp);
630 foffset > OFF_MAX - flp->l_start)) {
635 flp->l_start += foffset;
639 switch (flp->l_type) {
641 if ((fp->f_flag & FREAD) == 0) {
645 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
646 PROC_LOCK(p->p_leader);
647 p->p_leader->p_flag |= P_ADVLOCK;
648 PROC_UNLOCK(p->p_leader);
650 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
654 if ((fp->f_flag & FWRITE) == 0) {
658 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
659 PROC_LOCK(p->p_leader);
660 p->p_leader->p_flag |= P_ADVLOCK;
661 PROC_UNLOCK(p->p_leader);
663 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
667 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
672 * Temporary api for testing remote lock
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 fp->f_seqcount = (arg + bsize - 1) / bsize;
784 atomic_set_int(&fp->f_flag, FRDAHEAD);
786 atomic_clear_int(&fp->f_flag, FRDAHEAD);
800 getmaxfd(struct thread *td)
803 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
807 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
810 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
812 struct filedesc *fdp;
813 struct filedescent *oldfde, *newfde;
816 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.
882 error = racct_set(p, RACCT_NOFILE, new + 1);
890 fdgrowtable_exp(fdp, new + 1);
892 if (!fdisused(fdp, new))
896 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
899 KASSERT(old != new, ("new fd is same as old"));
901 oldfde = &fdp->fd_ofiles[old];
902 fhold(oldfde->fde_file);
903 newfde = &fdp->fd_ofiles[new];
904 delfp = newfde->fde_file;
906 oioctls = filecaps_free_prep(&newfde->fde_caps);
907 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
910 * Duplicate the source descriptor.
913 seq_write_begin(&newfde->fde_seq);
915 memcpy(newfde, oldfde, fde_change_size);
916 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
918 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
919 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
921 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
923 seq_write_end(&newfde->fde_seq);
925 filecaps_free_finish(oioctls);
926 td->td_retval[0] = new;
931 (void) closefp(fdp, new, delfp, td, 1);
932 FILEDESC_UNLOCK_ASSERT(fdp);
935 FILEDESC_XUNLOCK(fdp);
942 * If sigio is on the list associated with a process or process group,
943 * disable signalling from the device, remove sigio from the list and
947 funsetown(struct sigio **sigiop)
959 *(sigio->sio_myref) = NULL;
960 if ((sigio)->sio_pgid < 0) {
961 struct pgrp *pg = (sigio)->sio_pgrp;
963 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
967 struct proc *p = (sigio)->sio_proc;
969 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
974 crfree(sigio->sio_ucred);
975 free(sigio, M_SIGIO);
979 * Free a list of sigio structures.
980 * We only need to lock the SIGIO_LOCK because we have made ourselves
981 * inaccessible to callers of fsetown and therefore do not need to lock
982 * the proc or pgrp struct for the list manipulation.
985 funsetownlst(struct sigiolst *sigiolst)
991 sigio = SLIST_FIRST(sigiolst);
998 * Every entry of the list should belong
999 * to a single proc or pgrp.
1001 if (sigio->sio_pgid < 0) {
1002 pg = sigio->sio_pgrp;
1003 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
1004 } else /* if (sigio->sio_pgid > 0) */ {
1005 p = sigio->sio_proc;
1006 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1010 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1011 *(sigio->sio_myref) = NULL;
1013 KASSERT(sigio->sio_pgid < 0,
1014 ("Proc sigio in pgrp sigio list"));
1015 KASSERT(sigio->sio_pgrp == pg,
1016 ("Bogus pgrp in sigio list"));
1018 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1021 } else /* if (p != NULL) */ {
1022 KASSERT(sigio->sio_pgid > 0,
1023 ("Pgrp sigio in proc sigio list"));
1024 KASSERT(sigio->sio_proc == p,
1025 ("Bogus proc in sigio list"));
1027 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1032 crfree(sigio->sio_ucred);
1033 free(sigio, M_SIGIO);
1040 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1042 * After permission checking, add a sigio structure to the sigio list for
1043 * the process or process group.
1046 fsetown(pid_t pgid, struct sigio **sigiop)
1050 struct sigio *sigio;
1060 /* Allocate and fill in the new sigio out of locks. */
1061 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1062 sigio->sio_pgid = pgid;
1063 sigio->sio_ucred = crhold(curthread->td_ucred);
1064 sigio->sio_myref = sigiop;
1066 sx_slock(&proctree_lock);
1075 * Policy - Don't allow a process to FSETOWN a process
1076 * in another session.
1078 * Remove this test to allow maximum flexibility or
1079 * restrict FSETOWN to the current process or process
1080 * group for maximum safety.
1083 if (proc->p_session != curthread->td_proc->p_session) {
1089 } else /* if (pgid < 0) */ {
1090 pgrp = pgfind(-pgid);
1098 * Policy - Don't allow a process to FSETOWN a process
1099 * in another session.
1101 * Remove this test to allow maximum flexibility or
1102 * restrict FSETOWN to the current process or process
1103 * group for maximum safety.
1105 if (pgrp->pg_session != curthread->td_proc->p_session) {
1116 * Since funsetownlst() is called without the proctree
1117 * locked, we need to check for P_WEXIT.
1118 * XXX: is ESRCH correct?
1120 if ((proc->p_flag & P_WEXIT) != 0) {
1125 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1126 sigio->sio_proc = proc;
1130 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1131 sigio->sio_pgrp = pgrp;
1134 sx_sunlock(&proctree_lock);
1141 sx_sunlock(&proctree_lock);
1142 crfree(sigio->sio_ucred);
1143 free(sigio, M_SIGIO);
1148 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1151 fgetown(struct sigio **sigiop)
1156 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1162 * Function drops the filedesc lock on return.
1165 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1170 FILEDESC_XLOCK_ASSERT(fdp);
1173 if (td->td_proc->p_fdtol != NULL) {
1175 * Ask fdfree() to sleep to ensure that all relevant
1176 * process leaders can be traversed in closef().
1178 fdp->fd_holdleaderscount++;
1185 * We now hold the fp reference that used to be owned by the
1186 * descriptor array. We have to unlock the FILEDESC *AFTER*
1187 * knote_fdclose to prevent a race of the fd getting opened, a knote
1188 * added, and deleteing a knote for the new fd.
1190 knote_fdclose(td, fd);
1193 * We need to notify mqueue if the object is of type mqueue.
1195 if (fp->f_type == DTYPE_MQUEUE)
1196 mq_fdclose(td, fd, fp);
1197 FILEDESC_XUNLOCK(fdp);
1199 error = closef(fp, td);
1201 FILEDESC_XLOCK(fdp);
1202 fdp->fd_holdleaderscount--;
1203 if (fdp->fd_holdleaderscount == 0 &&
1204 fdp->fd_holdleaderswakeup != 0) {
1205 fdp->fd_holdleaderswakeup = 0;
1206 wakeup(&fdp->fd_holdleaderscount);
1208 FILEDESC_XUNLOCK(fdp);
1214 * Close a file descriptor.
1216 #ifndef _SYS_SYSPROTO_H_
1223 sys_close(struct thread *td, struct close_args *uap)
1226 return (kern_close(td, uap->fd));
1230 kern_close(struct thread *td, int fd)
1232 struct filedesc *fdp;
1235 fdp = td->td_proc->p_fd;
1237 AUDIT_SYSCLOSE(td, fd);
1239 FILEDESC_XLOCK(fdp);
1240 if ((fp = fget_locked(fdp, fd)) == NULL) {
1241 FILEDESC_XUNLOCK(fdp);
1246 /* closefp() drops the FILEDESC lock for us. */
1247 return (closefp(fdp, fd, fp, td, 1));
1251 * Close open file descriptors.
1253 #ifndef _SYS_SYSPROTO_H_
1254 struct closefrom_args {
1260 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1262 struct filedesc *fdp;
1265 fdp = td->td_proc->p_fd;
1266 AUDIT_ARG_FD(uap->lowfd);
1269 * Treat negative starting file descriptor values identical to
1270 * closefrom(0) which closes all files.
1274 FILEDESC_SLOCK(fdp);
1275 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1276 if (fdp->fd_ofiles[fd].fde_file != NULL) {
1277 FILEDESC_SUNLOCK(fdp);
1278 (void)kern_close(td, fd);
1279 FILEDESC_SLOCK(fdp);
1282 FILEDESC_SUNLOCK(fdp);
1286 #if defined(COMPAT_43)
1288 * Return status information about a file descriptor.
1290 #ifndef _SYS_SYSPROTO_H_
1291 struct ofstat_args {
1298 ofstat(struct thread *td, struct ofstat_args *uap)
1304 error = kern_fstat(td, uap->fd, &ub);
1307 error = copyout(&oub, uap->sb, sizeof(oub));
1311 #endif /* COMPAT_43 */
1313 #if defined(COMPAT_FREEBSD11)
1315 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1318 struct freebsd11_stat osb;
1321 error = kern_fstat(td, uap->fd, &sb);
1324 error = freebsd11_cvtstat(&sb, &osb);
1326 error = copyout(&osb, uap->sb, sizeof(osb));
1329 #endif /* COMPAT_FREEBSD11 */
1332 * Return status information about a file descriptor.
1334 #ifndef _SYS_SYSPROTO_H_
1342 sys_fstat(struct thread *td, struct fstat_args *uap)
1347 error = kern_fstat(td, uap->fd, &ub);
1349 error = copyout(&ub, uap->sb, sizeof(ub));
1354 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1361 error = fget(td, fd, &cap_fstat_rights, &fp);
1365 AUDIT_ARG_FILE(td->td_proc, fp);
1367 error = fo_stat(fp, sbp, td->td_ucred, td);
1369 #ifdef __STAT_TIME_T_EXT
1371 sbp->st_atim_ext = 0;
1372 sbp->st_mtim_ext = 0;
1373 sbp->st_ctim_ext = 0;
1374 sbp->st_btim_ext = 0;
1378 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1384 #if defined(COMPAT_FREEBSD11)
1386 * Return status information about a file descriptor.
1388 #ifndef _SYS_SYSPROTO_H_
1389 struct freebsd11_nfstat_args {
1396 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1402 error = kern_fstat(td, uap->fd, &ub);
1404 freebsd11_cvtnstat(&ub, &nub);
1405 error = copyout(&nub, uap->sb, sizeof(nub));
1409 #endif /* COMPAT_FREEBSD11 */
1412 * Return pathconf information about a file descriptor.
1414 #ifndef _SYS_SYSPROTO_H_
1415 struct fpathconf_args {
1422 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1427 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1429 td->td_retval[0] = value;
1434 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1440 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1444 if (name == _PC_ASYNC_IO) {
1445 *valuep = _POSIX_ASYNCHRONOUS_IO;
1450 vn_lock(vp, LK_SHARED | LK_RETRY);
1451 error = VOP_PATHCONF(vp, name, valuep);
1453 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1454 if (name != _PC_PIPE_BUF) {
1469 * Initialize filecaps structure.
1472 filecaps_init(struct filecaps *fcaps)
1475 bzero(fcaps, sizeof(*fcaps));
1476 fcaps->fc_nioctls = -1;
1480 * Copy filecaps structure allocating memory for ioctls array if needed.
1482 * The last parameter indicates whether the fdtable is locked. If it is not and
1483 * ioctls are encountered, copying fails and the caller must lock the table.
1485 * Note that if the table was not locked, the caller has to check the relevant
1486 * sequence counter to determine whether the operation was successful.
1489 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1493 if (src->fc_ioctls != NULL && !locked)
1495 memcpy(dst, src, sizeof(*src));
1496 if (src->fc_ioctls == NULL)
1499 KASSERT(src->fc_nioctls > 0,
1500 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1502 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1503 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1504 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1509 filecaps_copy_prep(const struct filecaps *src)
1514 if (src->fc_ioctls == NULL)
1517 KASSERT(src->fc_nioctls > 0,
1518 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1520 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1521 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1526 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1532 if (src->fc_ioctls == NULL) {
1533 MPASS(ioctls == NULL);
1537 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1538 dst->fc_ioctls = ioctls;
1539 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1543 * Move filecaps structure to the new place and clear the old place.
1546 filecaps_move(struct filecaps *src, struct filecaps *dst)
1550 bzero(src, sizeof(*src));
1554 * Fill the given filecaps structure with full rights.
1557 filecaps_fill(struct filecaps *fcaps)
1560 CAP_ALL(&fcaps->fc_rights);
1561 fcaps->fc_ioctls = NULL;
1562 fcaps->fc_nioctls = -1;
1563 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1567 * Free memory allocated within filecaps structure.
1570 filecaps_free(struct filecaps *fcaps)
1573 free(fcaps->fc_ioctls, M_FILECAPS);
1574 bzero(fcaps, sizeof(*fcaps));
1578 filecaps_free_prep(struct filecaps *fcaps)
1582 ioctls = fcaps->fc_ioctls;
1583 bzero(fcaps, sizeof(*fcaps));
1588 filecaps_free_finish(u_long *ioctls)
1591 free(ioctls, M_FILECAPS);
1595 * Validate the given filecaps structure.
1598 filecaps_validate(const struct filecaps *fcaps, const char *func)
1601 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1602 ("%s: invalid rights", func));
1603 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1604 ("%s: invalid fcntls", func));
1605 KASSERT(fcaps->fc_fcntls == 0 ||
1606 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1607 ("%s: fcntls without CAP_FCNTL", func));
1608 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1609 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1610 ("%s: invalid ioctls", func));
1611 KASSERT(fcaps->fc_nioctls == 0 ||
1612 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1613 ("%s: ioctls without CAP_IOCTL", func));
1617 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1621 FILEDESC_XLOCK_ASSERT(fdp);
1623 nfd1 = fdp->fd_nfiles * 2;
1626 fdgrowtable(fdp, nfd1);
1630 * Grow the file table to accommodate (at least) nfd descriptors.
1633 fdgrowtable(struct filedesc *fdp, int nfd)
1635 struct filedesc0 *fdp0;
1636 struct freetable *ft;
1637 struct fdescenttbl *ntable;
1638 struct fdescenttbl *otable;
1639 int nnfiles, onfiles;
1640 NDSLOTTYPE *nmap, *omap;
1643 * If lastfile is -1 this struct filedesc was just allocated and we are
1644 * growing it to accommodate for the one we are going to copy from. There
1645 * is no need to have a lock on this one as it's not visible to anyone.
1647 if (fdp->fd_lastfile != -1)
1648 FILEDESC_XLOCK_ASSERT(fdp);
1650 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1652 /* save old values */
1653 onfiles = fdp->fd_nfiles;
1654 otable = fdp->fd_files;
1657 /* compute the size of the new table */
1658 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1659 if (nnfiles <= onfiles)
1660 /* the table is already large enough */
1664 * Allocate a new table. We need enough space for the number of
1665 * entries, file entries themselves and the struct freetable we will use
1666 * when we decommission the table and place it on the freelist.
1667 * We place the struct freetable in the middle so we don't have
1668 * to worry about padding.
1670 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1671 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1672 sizeof(struct freetable),
1673 M_FILEDESC, M_ZERO | M_WAITOK);
1674 /* copy the old data */
1675 ntable->fdt_nfiles = nnfiles;
1676 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1677 onfiles * sizeof(ntable->fdt_ofiles[0]));
1680 * Allocate a new map only if the old is not large enough. It will
1681 * grow at a slower rate than the table as it can map more
1682 * entries than the table can hold.
1684 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1685 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1687 /* copy over the old data and update the pointer */
1688 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1693 * Make sure that ntable is correctly initialized before we replace
1694 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1697 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1700 * Do not free the old file table, as some threads may still
1701 * reference entries within it. Instead, place it on a freelist
1702 * which will be processed when the struct filedesc is released.
1704 * Note that if onfiles == NDFILE, we're dealing with the original
1705 * static allocation contained within (struct filedesc0 *)fdp,
1706 * which must not be freed.
1708 if (onfiles > NDFILE) {
1709 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1710 fdp0 = (struct filedesc0 *)fdp;
1711 ft->ft_table = otable;
1712 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1715 * The map does not have the same possibility of threads still
1716 * holding references to it. So always free it as long as it
1717 * does not reference the original static allocation.
1719 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1720 free(omap, M_FILEDESC);
1724 * Allocate a file descriptor for the process.
1727 fdalloc(struct thread *td, int minfd, int *result)
1729 struct proc *p = td->td_proc;
1730 struct filedesc *fdp = p->p_fd;
1731 int fd, maxfd, allocfd;
1736 FILEDESC_XLOCK_ASSERT(fdp);
1738 if (fdp->fd_freefile > minfd)
1739 minfd = fdp->fd_freefile;
1741 maxfd = getmaxfd(td);
1744 * Search the bitmap for a free descriptor starting at minfd.
1745 * If none is found, grow the file table.
1747 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1750 if (fd >= fdp->fd_nfiles) {
1751 allocfd = min(fd * 2, maxfd);
1755 error = racct_set(p, RACCT_NOFILE, allocfd);
1762 * fd is already equal to first free descriptor >= minfd, so
1763 * we only need to grow the table and we are done.
1765 fdgrowtable_exp(fdp, allocfd);
1769 * Perform some sanity checks, then mark the file descriptor as
1770 * used and return it to the caller.
1772 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1773 ("invalid descriptor %d", fd));
1774 KASSERT(!fdisused(fdp, fd),
1775 ("fd_first_free() returned non-free descriptor"));
1776 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1777 ("file descriptor isn't free"));
1784 * Allocate n file descriptors for the process.
1787 fdallocn(struct thread *td, int minfd, int *fds, int n)
1789 struct proc *p = td->td_proc;
1790 struct filedesc *fdp = p->p_fd;
1793 FILEDESC_XLOCK_ASSERT(fdp);
1795 for (i = 0; i < n; i++)
1796 if (fdalloc(td, 0, &fds[i]) != 0)
1800 for (i--; i >= 0; i--)
1801 fdunused(fdp, fds[i]);
1809 * Create a new open file structure and allocate a file descriptor for the
1810 * process that refers to it. We add one reference to the file for the
1811 * descriptor table and one reference for resultfp. This is to prevent us
1812 * being preempted and the entry in the descriptor table closed after we
1813 * release the FILEDESC lock.
1816 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1817 struct filecaps *fcaps)
1822 error = falloc_noinstall(td, &fp);
1824 return (error); /* no reference held on error */
1826 error = finstall(td, fp, &fd, flags, fcaps);
1828 fdrop(fp, td); /* one reference (fp only) */
1832 if (resultfp != NULL)
1833 *resultfp = fp; /* copy out result */
1835 fdrop(fp, td); /* release local reference */
1837 if (resultfd != NULL)
1844 * Create a new open file structure without allocating a file descriptor.
1847 falloc_noinstall(struct thread *td, struct file **resultfp)
1850 int maxuserfiles = maxfiles - (maxfiles / 20);
1852 static struct timeval lastfail;
1855 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1857 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1858 if ((openfiles_new >= maxuserfiles &&
1859 priv_check(td, PRIV_MAXFILES) != 0) ||
1860 openfiles_new >= maxfiles) {
1861 atomic_subtract_int(&openfiles, 1);
1862 if (ppsratecheck(&lastfail, &curfail, 1)) {
1863 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1864 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1868 fp = uma_zalloc(file_zone, M_WAITOK);
1869 bzero(fp, sizeof(*fp));
1870 refcount_init(&fp->f_count, 1);
1871 fp->f_cred = crhold(td->td_ucred);
1872 fp->f_ops = &badfileops;
1878 * Install a file in a file descriptor table.
1881 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1882 struct filecaps *fcaps)
1884 struct filedescent *fde;
1888 filecaps_validate(fcaps, __func__);
1889 FILEDESC_XLOCK_ASSERT(fdp);
1891 fde = &fdp->fd_ofiles[fd];
1893 seq_write_begin(&fde->fde_seq);
1896 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1898 filecaps_move(fcaps, &fde->fde_caps);
1900 filecaps_fill(&fde->fde_caps);
1902 seq_write_end(&fde->fde_seq);
1907 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1908 struct filecaps *fcaps)
1910 struct filedesc *fdp = td->td_proc->p_fd;
1915 FILEDESC_XLOCK(fdp);
1916 if ((error = fdalloc(td, 0, fd))) {
1917 FILEDESC_XUNLOCK(fdp);
1921 _finstall(fdp, fp, *fd, flags, fcaps);
1922 FILEDESC_XUNLOCK(fdp);
1927 * Build a new filedesc structure from another.
1928 * Copy the current, root, and jail root vnode references.
1930 * If fdp is not NULL, return with it shared locked.
1933 fdinit(struct filedesc *fdp, bool prepfiles)
1935 struct filedesc0 *newfdp0;
1936 struct filedesc *newfdp;
1938 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1939 newfdp = &newfdp0->fd_fd;
1941 /* Create the file descriptor table. */
1942 FILEDESC_LOCK_INIT(newfdp);
1943 refcount_init(&newfdp->fd_refcnt, 1);
1944 refcount_init(&newfdp->fd_holdcnt, 1);
1945 newfdp->fd_cmask = CMASK;
1946 newfdp->fd_map = newfdp0->fd_dmap;
1947 newfdp->fd_lastfile = -1;
1948 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1949 newfdp->fd_files->fdt_nfiles = NDFILE;
1954 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1955 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1957 FILEDESC_SLOCK(fdp);
1958 newfdp->fd_cdir = fdp->fd_cdir;
1959 if (newfdp->fd_cdir)
1960 vrefact(newfdp->fd_cdir);
1961 newfdp->fd_rdir = fdp->fd_rdir;
1962 if (newfdp->fd_rdir)
1963 vrefact(newfdp->fd_rdir);
1964 newfdp->fd_jdir = fdp->fd_jdir;
1965 if (newfdp->fd_jdir)
1966 vrefact(newfdp->fd_jdir);
1969 FILEDESC_SUNLOCK(fdp);
1971 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1972 FILEDESC_SUNLOCK(fdp);
1973 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1974 FILEDESC_SLOCK(fdp);
1981 static struct filedesc *
1982 fdhold(struct proc *p)
1984 struct filedesc *fdp;
1986 PROC_LOCK_ASSERT(p, MA_OWNED);
1989 refcount_acquire(&fdp->fd_holdcnt);
1994 fddrop(struct filedesc *fdp)
1997 if (fdp->fd_holdcnt > 1) {
1998 if (refcount_release(&fdp->fd_holdcnt) == 0)
2002 FILEDESC_LOCK_DESTROY(fdp);
2003 uma_zfree(filedesc0_zone, fdp);
2007 * Share a filedesc structure.
2010 fdshare(struct filedesc *fdp)
2013 refcount_acquire(&fdp->fd_refcnt);
2018 * Unshare a filedesc structure, if necessary by making a copy
2021 fdunshare(struct thread *td)
2023 struct filedesc *tmp;
2024 struct proc *p = td->td_proc;
2026 if (p->p_fd->fd_refcnt == 1)
2029 tmp = fdcopy(p->p_fd);
2035 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2039 td->td_proc->p_fd = fdp;
2043 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2044 * this is to ease callers, not catch errors.
2047 fdcopy(struct filedesc *fdp)
2049 struct filedesc *newfdp;
2050 struct filedescent *nfde, *ofde;
2055 newfdp = fdinit(fdp, true);
2056 /* copy all passable descriptors (i.e. not kqueue) */
2057 newfdp->fd_freefile = -1;
2058 for (i = 0; i <= fdp->fd_lastfile; ++i) {
2059 ofde = &fdp->fd_ofiles[i];
2060 if (ofde->fde_file == NULL ||
2061 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2062 if (newfdp->fd_freefile == -1)
2063 newfdp->fd_freefile = i;
2066 nfde = &newfdp->fd_ofiles[i];
2068 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2069 fhold(nfde->fde_file);
2070 fdused_init(newfdp, i);
2071 newfdp->fd_lastfile = i;
2073 if (newfdp->fd_freefile == -1)
2074 newfdp->fd_freefile = i;
2075 newfdp->fd_cmask = fdp->fd_cmask;
2076 FILEDESC_SUNLOCK(fdp);
2081 * Copies a filedesc structure, while remapping all file descriptors
2082 * stored inside using a translation table.
2084 * File descriptors are copied over to the new file descriptor table,
2085 * regardless of whether the close-on-exec flag is set.
2088 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2089 struct filedesc **ret)
2091 struct filedesc *newfdp;
2092 struct filedescent *nfde, *ofde;
2097 newfdp = fdinit(fdp, true);
2098 if (nfds > fdp->fd_lastfile + 1) {
2099 /* New table cannot be larger than the old one. */
2103 /* Copy all passable descriptors (i.e. not kqueue). */
2104 newfdp->fd_freefile = nfds;
2105 for (i = 0; i < nfds; ++i) {
2106 if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2107 /* File descriptor out of bounds. */
2111 ofde = &fdp->fd_ofiles[fds[i]];
2112 if (ofde->fde_file == NULL) {
2113 /* Unused file descriptor. */
2117 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2118 /* File descriptor cannot be passed. */
2122 nfde = &newfdp->fd_ofiles[i];
2124 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2125 fhold(nfde->fde_file);
2126 fdused_init(newfdp, i);
2127 newfdp->fd_lastfile = i;
2129 newfdp->fd_cmask = fdp->fd_cmask;
2130 FILEDESC_SUNLOCK(fdp);
2134 FILEDESC_SUNLOCK(fdp);
2135 fdescfree_remapped(newfdp);
2140 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2141 * one of processes using it exits) and the table used to be shared.
2144 fdclearlocks(struct thread *td)
2146 struct filedesc *fdp;
2147 struct filedesc_to_leader *fdtol;
2157 MPASS(fdtol != NULL);
2159 FILEDESC_XLOCK(fdp);
2160 KASSERT(fdtol->fdl_refcount > 0,
2161 ("filedesc_to_refcount botch: fdl_refcount=%d",
2162 fdtol->fdl_refcount));
2163 if (fdtol->fdl_refcount == 1 &&
2164 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2165 for (i = 0; i <= fdp->fd_lastfile; i++) {
2166 fp = fdp->fd_ofiles[i].fde_file;
2167 if (fp == NULL || fp->f_type != DTYPE_VNODE)
2170 FILEDESC_XUNLOCK(fdp);
2171 lf.l_whence = SEEK_SET;
2174 lf.l_type = F_UNLCK;
2176 (void) VOP_ADVLOCK(vp,
2177 (caddr_t)p->p_leader, F_UNLCK,
2179 FILEDESC_XLOCK(fdp);
2184 if (fdtol->fdl_refcount == 1) {
2185 if (fdp->fd_holdleaderscount > 0 &&
2186 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2188 * close() or kern_dup() has cleared a reference
2189 * in a shared file descriptor table.
2191 fdp->fd_holdleaderswakeup = 1;
2192 sx_sleep(&fdp->fd_holdleaderscount,
2193 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2196 if (fdtol->fdl_holdcount > 0) {
2198 * Ensure that fdtol->fdl_leader remains
2199 * valid in closef().
2201 fdtol->fdl_wakeup = 1;
2202 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2207 fdtol->fdl_refcount--;
2208 if (fdtol->fdl_refcount == 0 &&
2209 fdtol->fdl_holdcount == 0) {
2210 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2211 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2215 FILEDESC_XUNLOCK(fdp);
2217 free(fdtol, M_FILEDESC_TO_LEADER);
2221 * Release a filedesc structure.
2224 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2226 struct filedesc0 *fdp0;
2227 struct freetable *ft, *tft;
2228 struct filedescent *fde;
2232 for (i = 0; i <= fdp->fd_lastfile; i++) {
2233 fde = &fdp->fd_ofiles[i];
2238 (void) closef(fp, td);
2244 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2245 free(fdp->fd_map, M_FILEDESC);
2246 if (fdp->fd_nfiles > NDFILE)
2247 free(fdp->fd_files, M_FILEDESC);
2249 fdp0 = (struct filedesc0 *)fdp;
2250 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2251 free(ft->ft_table, M_FILEDESC);
2257 fdescfree(struct thread *td)
2260 struct filedesc *fdp;
2261 struct vnode *cdir, *jdir, *rdir;
2270 racct_set(p, RACCT_NOFILE, 0);
2275 if (p->p_fdtol != NULL)
2282 if (refcount_release(&fdp->fd_refcnt) == 0)
2285 FILEDESC_XLOCK(fdp);
2286 cdir = fdp->fd_cdir;
2287 fdp->fd_cdir = NULL;
2288 rdir = fdp->fd_rdir;
2289 fdp->fd_rdir = NULL;
2290 jdir = fdp->fd_jdir;
2291 fdp->fd_jdir = NULL;
2292 FILEDESC_XUNLOCK(fdp);
2301 fdescfree_fds(td, fdp, 1);
2305 fdescfree_remapped(struct filedesc *fdp)
2308 if (fdp->fd_cdir != NULL)
2309 vrele(fdp->fd_cdir);
2310 if (fdp->fd_rdir != NULL)
2311 vrele(fdp->fd_rdir);
2312 if (fdp->fd_jdir != NULL)
2313 vrele(fdp->fd_jdir);
2315 fdescfree_fds(curthread, fdp, 0);
2319 * For setugid programs, we don't want to people to use that setugidness
2320 * to generate error messages which write to a file which otherwise would
2321 * otherwise be off-limits to the process. We check for filesystems where
2322 * the vnode can change out from under us after execve (like [lin]procfs).
2324 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2325 * sufficient. We also don't check for setugidness since we know we are.
2328 is_unsafe(struct file *fp)
2332 if (fp->f_type != DTYPE_VNODE)
2336 return ((vp->v_vflag & VV_PROCDEP) != 0);
2340 * Make this setguid thing safe, if at all possible.
2343 fdsetugidsafety(struct thread *td)
2345 struct filedesc *fdp;
2349 fdp = td->td_proc->p_fd;
2350 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2351 MPASS(fdp->fd_nfiles >= 3);
2352 for (i = 0; i <= 2; i++) {
2353 fp = fdp->fd_ofiles[i].fde_file;
2354 if (fp != NULL && is_unsafe(fp)) {
2355 FILEDESC_XLOCK(fdp);
2356 knote_fdclose(td, i);
2358 * NULL-out descriptor prior to close to avoid
2359 * a race while close blocks.
2362 FILEDESC_XUNLOCK(fdp);
2363 (void) closef(fp, td);
2369 * If a specific file object occupies a specific file descriptor, close the
2370 * file descriptor entry and drop a reference on the file object. This is a
2371 * convenience function to handle a subsequent error in a function that calls
2372 * falloc() that handles the race that another thread might have closed the
2373 * file descriptor out from under the thread creating the file object.
2376 fdclose(struct thread *td, struct file *fp, int idx)
2378 struct filedesc *fdp = td->td_proc->p_fd;
2380 FILEDESC_XLOCK(fdp);
2381 if (fdp->fd_ofiles[idx].fde_file == fp) {
2383 FILEDESC_XUNLOCK(fdp);
2386 FILEDESC_XUNLOCK(fdp);
2390 * Close any files on exec?
2393 fdcloseexec(struct thread *td)
2395 struct filedesc *fdp;
2396 struct filedescent *fde;
2400 fdp = td->td_proc->p_fd;
2401 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2402 for (i = 0; i <= fdp->fd_lastfile; i++) {
2403 fde = &fdp->fd_ofiles[i];
2405 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2406 (fde->fde_flags & UF_EXCLOSE))) {
2407 FILEDESC_XLOCK(fdp);
2409 (void) closefp(fdp, i, fp, td, 0);
2410 FILEDESC_UNLOCK_ASSERT(fdp);
2416 * It is unsafe for set[ug]id processes to be started with file
2417 * descriptors 0..2 closed, as these descriptors are given implicit
2418 * significance in the Standard C library. fdcheckstd() will create a
2419 * descriptor referencing /dev/null for each of stdin, stdout, and
2420 * stderr that is not already open.
2423 fdcheckstd(struct thread *td)
2425 struct filedesc *fdp;
2427 int i, error, devnull;
2429 fdp = td->td_proc->p_fd;
2430 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2431 MPASS(fdp->fd_nfiles >= 3);
2433 for (i = 0; i <= 2; i++) {
2434 if (fdp->fd_ofiles[i].fde_file != NULL)
2437 save = td->td_retval[0];
2438 if (devnull != -1) {
2439 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2441 error = kern_openat(td, AT_FDCWD, "/dev/null",
2442 UIO_SYSSPACE, O_RDWR, 0);
2444 devnull = td->td_retval[0];
2445 KASSERT(devnull == i, ("we didn't get our fd"));
2448 td->td_retval[0] = save;
2456 * Internal form of close. Decrement reference count on file structure.
2457 * Note: td may be NULL when closing a file that was being passed in a
2460 * XXXRW: Giant is not required for the caller, but often will be held; this
2461 * makes it moderately likely the Giant will be recursed in the VFS case.
2464 closef(struct file *fp, struct thread *td)
2468 struct filedesc_to_leader *fdtol;
2469 struct filedesc *fdp;
2472 * POSIX record locking dictates that any close releases ALL
2473 * locks owned by this process. This is handled by setting
2474 * a flag in the unlock to free ONLY locks obeying POSIX
2475 * semantics, and not to free BSD-style file locks.
2476 * If the descriptor was in a message, POSIX-style locks
2477 * aren't passed with the descriptor, and the thread pointer
2478 * will be NULL. Callers should be careful only to pass a
2479 * NULL thread pointer when there really is no owning
2480 * context that might have locks, or the locks will be
2483 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2485 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2486 lf.l_whence = SEEK_SET;
2489 lf.l_type = F_UNLCK;
2490 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2491 F_UNLCK, &lf, F_POSIX);
2493 fdtol = td->td_proc->p_fdtol;
2494 if (fdtol != NULL) {
2496 * Handle special case where file descriptor table is
2497 * shared between multiple process leaders.
2499 fdp = td->td_proc->p_fd;
2500 FILEDESC_XLOCK(fdp);
2501 for (fdtol = fdtol->fdl_next;
2502 fdtol != td->td_proc->p_fdtol;
2503 fdtol = fdtol->fdl_next) {
2504 if ((fdtol->fdl_leader->p_flag &
2507 fdtol->fdl_holdcount++;
2508 FILEDESC_XUNLOCK(fdp);
2509 lf.l_whence = SEEK_SET;
2512 lf.l_type = F_UNLCK;
2514 (void) VOP_ADVLOCK(vp,
2515 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2517 FILEDESC_XLOCK(fdp);
2518 fdtol->fdl_holdcount--;
2519 if (fdtol->fdl_holdcount == 0 &&
2520 fdtol->fdl_wakeup != 0) {
2521 fdtol->fdl_wakeup = 0;
2525 FILEDESC_XUNLOCK(fdp);
2528 return (fdrop(fp, td));
2532 * Initialize the file pointer with the specified properties.
2534 * The ops are set with release semantics to be certain that the flags, type,
2535 * and data are visible when ops is. This is to prevent ops methods from being
2536 * called with bad data.
2539 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2544 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2548 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2549 struct file **fpp, struct filecaps *havecapsp)
2551 struct filedescent *fde;
2554 FILEDESC_LOCK_ASSERT(fdp);
2556 fde = fdeget_locked(fdp, fd);
2563 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2568 if (havecapsp != NULL)
2569 filecaps_copy(&fde->fde_caps, havecapsp, true);
2571 *fpp = fde->fde_file;
2579 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2580 struct file **fpp, struct filecaps *havecapsp)
2582 struct filedesc *fdp = td->td_proc->p_fd;
2584 #ifndef CAPABILITIES
2585 error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2586 if (error == 0 && havecapsp != NULL)
2587 filecaps_fill(havecapsp);
2593 error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2597 if (havecapsp != NULL) {
2598 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2599 havecapsp, false)) {
2605 if (!fd_modified(fdp, fd, seq))
2614 FILEDESC_SLOCK(fdp);
2615 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2618 FILEDESC_SUNLOCK(fdp);
2624 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2625 struct file **fpp, seq_t *seqp)
2628 const struct filedescent *fde;
2630 const struct fdescenttbl *fdt;
2635 cap_rights_t haverights;
2639 fdt = fdp->fd_files;
2640 if ((u_int)fd >= fdt->fdt_nfiles)
2643 * Fetch the descriptor locklessly. We avoid fdrop() races by
2644 * never raising a refcount above 0. To accomplish this we have
2645 * to use a cmpset loop rather than an atomic_add. The descriptor
2646 * must be re-verified once we acquire a reference to be certain
2647 * that the identity is still correct and we did not lose a race
2648 * due to preemption.
2652 seq = seq_read(fd_seq(fdt, fd));
2653 fde = &fdt->fdt_ofiles[fd];
2654 haverights = *cap_rights_fde_inline(fde);
2656 if (!seq_consistent(fd_seq(fdt, fd), seq))
2659 fp = fdt->fdt_ofiles[fd].fde_file;
2664 error = cap_check(&haverights, needrightsp);
2668 count = fp->f_count;
2672 * Force a reload. Other thread could reallocate the
2673 * table before this fd was closed, so it possible that
2674 * there is a stale fp pointer in cached version.
2676 fdt = *(const struct fdescenttbl * const volatile *)&(fdp->fd_files);
2680 * Use an acquire barrier to force re-reading of fdt so it is
2681 * refreshed for verification.
2683 if (atomic_fcmpset_acq_int(&fp->f_count, &count, count + 1) == 0)
2685 fdt = fdp->fd_files;
2687 if (seq_consistent_nomb(fd_seq(fdt, fd), seq))
2689 if (fp == fdt->fdt_ofiles[fd].fde_file)
2692 fdrop(fp, curthread);
2704 * Extract the file pointer associated with the specified descriptor for the
2705 * current user process.
2707 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2710 * File's rights will be checked against the capability rights mask.
2712 * If an error occurred the non-zero error is returned and *fpp is set to
2713 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
2714 * responsible for fdrop().
2717 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2718 cap_rights_t *needrightsp, seq_t *seqp)
2720 struct filedesc *fdp;
2725 fdp = td->td_proc->p_fd;
2726 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2729 if (fp->f_ops == &badfileops) {
2735 * FREAD and FWRITE failure return EBADF as per POSIX.
2741 if ((fp->f_flag & flags) == 0)
2745 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2746 ((fp->f_flag & FWRITE) != 0))
2752 KASSERT(0, ("wrong flags"));
2765 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2768 return (_fget(td, fd, fpp, 0, rightsp, NULL));
2772 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2776 #ifndef CAPABILITIES
2777 error = _fget(td, fd, fpp, 0, rightsp, NULL);
2778 if (maxprotp != NULL)
2779 *maxprotp = VM_PROT_ALL;
2781 struct filedesc *fdp = td->td_proc->p_fd;
2784 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2786 error = _fget(td, fd, fpp, 0, rightsp, &seq);
2790 * If requested, convert capability rights to access flags.
2792 if (maxprotp != NULL)
2793 *maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd));
2794 if (!fd_modified(fdp, fd, seq))
2803 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2806 return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2810 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2813 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2817 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2820 struct filedesc *fdp = td->td_proc->p_fd;
2821 #ifndef CAPABILITIES
2822 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2827 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2829 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2832 error = cap_fcntl_check(fdp, fd, needfcntl);
2833 if (!fd_modified(fdp, fd, seq))
2846 * Like fget() but loads the underlying vnode, or returns an error if the
2847 * descriptor does not represent a vnode. Note that pipes use vnodes but
2848 * never have VM objects. The returned vnode will be vref()'d.
2850 * XXX: what about the unused flags ?
2853 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2860 error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2863 if (fp->f_vnode == NULL) {
2875 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2878 return (_fgetvp(td, fd, 0, rightsp, vpp));
2882 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2883 struct filecaps *havecaps, struct vnode **vpp)
2885 struct filedesc *fdp;
2886 struct filecaps caps;
2890 fdp = td->td_proc->p_fd;
2891 error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2894 if (fp->f_ops == &badfileops) {
2898 if (fp->f_vnode == NULL) {
2909 filecaps_free(&caps);
2914 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2917 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2921 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2924 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2929 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2933 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2938 * Handle the last reference to a file being closed.
2940 * Without the noinline attribute clang keeps inlining the func thorough this
2941 * file when fdrop is used.
2944 _fdrop(struct file *fp, struct thread *td)
2948 if (fp->f_count != 0)
2949 panic("fdrop: count %d", fp->f_count);
2950 error = fo_close(fp, td);
2951 atomic_subtract_int(&openfiles, 1);
2953 free(fp->f_advice, M_FADVISE);
2954 uma_zfree(file_zone, fp);
2960 * Apply an advisory lock on a file descriptor.
2962 * Just attempt to get a record lock of the requested type on the entire file
2963 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2965 #ifndef _SYS_SYSPROTO_H_
2973 sys_flock(struct thread *td, struct flock_args *uap)
2980 error = fget(td, uap->fd, &cap_flock_rights, &fp);
2983 if (fp->f_type != DTYPE_VNODE) {
2985 return (EOPNOTSUPP);
2989 lf.l_whence = SEEK_SET;
2992 if (uap->how & LOCK_UN) {
2993 lf.l_type = F_UNLCK;
2994 atomic_clear_int(&fp->f_flag, FHASLOCK);
2995 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2998 if (uap->how & LOCK_EX)
2999 lf.l_type = F_WRLCK;
3000 else if (uap->how & LOCK_SH)
3001 lf.l_type = F_RDLCK;
3006 atomic_set_int(&fp->f_flag, FHASLOCK);
3007 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3008 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3014 * Duplicate the specified descriptor to a free descriptor.
3017 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3018 int openerror, int *indxp)
3020 struct filedescent *newfde, *oldfde;
3025 KASSERT(openerror == ENODEV || openerror == ENXIO,
3026 ("unexpected error %d in %s", openerror, __func__));
3029 * If the to-be-dup'd fd number is greater than the allowed number
3030 * of file descriptors, or the fd to be dup'd has already been
3031 * closed, then reject.
3033 FILEDESC_XLOCK(fdp);
3034 if ((fp = fget_locked(fdp, dfd)) == NULL) {
3035 FILEDESC_XUNLOCK(fdp);
3039 error = fdalloc(td, 0, &indx);
3041 FILEDESC_XUNLOCK(fdp);
3046 * There are two cases of interest here.
3048 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3050 * For ENXIO steal away the file structure from (dfd) and store it in
3051 * (indx). (dfd) is effectively closed by this operation.
3053 switch (openerror) {
3056 * Check that the mode the file is being opened for is a
3057 * subset of the mode of the existing descriptor.
3059 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3060 fdunused(fdp, indx);
3061 FILEDESC_XUNLOCK(fdp);
3065 newfde = &fdp->fd_ofiles[indx];
3066 oldfde = &fdp->fd_ofiles[dfd];
3067 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3069 seq_write_begin(&newfde->fde_seq);
3071 memcpy(newfde, oldfde, fde_change_size);
3072 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3075 seq_write_end(&newfde->fde_seq);
3080 * Steal away the file pointer from dfd and stuff it into indx.
3082 newfde = &fdp->fd_ofiles[indx];
3083 oldfde = &fdp->fd_ofiles[dfd];
3085 seq_write_begin(&newfde->fde_seq);
3087 memcpy(newfde, oldfde, fde_change_size);
3088 oldfde->fde_file = NULL;
3091 seq_write_end(&newfde->fde_seq);
3095 FILEDESC_XUNLOCK(fdp);
3101 * This sysctl determines if we will allow a process to chroot(2) if it
3102 * has a directory open:
3103 * 0: disallowed for all processes.
3104 * 1: allowed for processes that were not already chroot(2)'ed.
3105 * 2: allowed for all processes.
3108 static int chroot_allow_open_directories = 1;
3110 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3111 &chroot_allow_open_directories, 0,
3112 "Allow a process to chroot(2) if it has a directory open");
3115 * Helper function for raised chroot(2) security function: Refuse if
3116 * any filedescriptors are open directories.
3119 chroot_refuse_vdir_fds(struct filedesc *fdp)
3125 FILEDESC_LOCK_ASSERT(fdp);
3127 for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3128 fp = fget_locked(fdp, fd);
3131 if (fp->f_type == DTYPE_VNODE) {
3133 if (vp->v_type == VDIR)
3141 * Common routine for kern_chroot() and jail_attach(). The caller is
3142 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3143 * authorize this operation.
3146 pwd_chroot(struct thread *td, struct vnode *vp)
3148 struct filedesc *fdp;
3149 struct vnode *oldvp;
3152 fdp = td->td_proc->p_fd;
3153 FILEDESC_XLOCK(fdp);
3154 if (chroot_allow_open_directories == 0 ||
3155 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3156 error = chroot_refuse_vdir_fds(fdp);
3158 FILEDESC_XUNLOCK(fdp);
3162 oldvp = fdp->fd_rdir;
3165 if (fdp->fd_jdir == NULL) {
3169 FILEDESC_XUNLOCK(fdp);
3175 pwd_chdir(struct thread *td, struct vnode *vp)
3177 struct filedesc *fdp;
3178 struct vnode *oldvp;
3180 fdp = td->td_proc->p_fd;
3181 FILEDESC_XLOCK(fdp);
3182 VNASSERT(vp->v_usecount > 0, vp,
3183 ("chdir to a vnode with zero usecount"));
3184 oldvp = fdp->fd_cdir;
3186 FILEDESC_XUNLOCK(fdp);
3191 * Scan all active processes and prisons to see if any of them have a current
3192 * or root directory of `olddp'. If so, replace them with the new mount point.
3195 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3197 struct filedesc *fdp;
3202 if (vrefcnt(olddp) == 1)
3205 sx_slock(&allproc_lock);
3206 FOREACH_PROC_IN_SYSTEM(p) {
3212 FILEDESC_XLOCK(fdp);
3213 if (fdp->fd_cdir == olddp) {
3215 fdp->fd_cdir = newdp;
3218 if (fdp->fd_rdir == olddp) {
3220 fdp->fd_rdir = newdp;
3223 if (fdp->fd_jdir == olddp) {
3225 fdp->fd_jdir = newdp;
3228 FILEDESC_XUNLOCK(fdp);
3231 sx_sunlock(&allproc_lock);
3232 if (rootvnode == olddp) {
3237 mtx_lock(&prison0.pr_mtx);
3238 if (prison0.pr_root == olddp) {
3240 prison0.pr_root = newdp;
3243 mtx_unlock(&prison0.pr_mtx);
3244 sx_slock(&allprison_lock);
3245 TAILQ_FOREACH(pr, &allprison, pr_list) {
3246 mtx_lock(&pr->pr_mtx);
3247 if (pr->pr_root == olddp) {
3249 pr->pr_root = newdp;
3252 mtx_unlock(&pr->pr_mtx);
3254 sx_sunlock(&allprison_lock);
3259 struct filedesc_to_leader *
3260 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3262 struct filedesc_to_leader *fdtol;
3264 fdtol = malloc(sizeof(struct filedesc_to_leader),
3265 M_FILEDESC_TO_LEADER, M_WAITOK);
3266 fdtol->fdl_refcount = 1;
3267 fdtol->fdl_holdcount = 0;
3268 fdtol->fdl_wakeup = 0;
3269 fdtol->fdl_leader = leader;
3271 FILEDESC_XLOCK(fdp);
3272 fdtol->fdl_next = old->fdl_next;
3273 fdtol->fdl_prev = old;
3274 old->fdl_next = fdtol;
3275 fdtol->fdl_next->fdl_prev = fdtol;
3276 FILEDESC_XUNLOCK(fdp);
3278 fdtol->fdl_next = fdtol;
3279 fdtol->fdl_prev = fdtol;
3285 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3287 struct filedesc *fdp;
3288 int i, count, slots;
3290 if (*(int *)arg1 != 0)
3293 fdp = curproc->p_fd;
3295 FILEDESC_SLOCK(fdp);
3296 slots = NDSLOTS(fdp->fd_lastfile + 1);
3297 for (i = 0; i < slots; i++)
3298 count += bitcountl(fdp->fd_map[i]);
3299 FILEDESC_SUNLOCK(fdp);
3301 return (SYSCTL_OUT(req, &count, sizeof(count)));
3304 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3305 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3306 "Number of open file descriptors");
3309 * Get file structures globally.
3312 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3315 struct filedesc *fdp;
3320 error = sysctl_wire_old_buffer(req, 0);
3323 if (req->oldptr == NULL) {
3325 sx_slock(&allproc_lock);
3326 FOREACH_PROC_IN_SYSTEM(p) {
3328 if (p->p_state == PRS_NEW) {
3336 /* overestimates sparse tables. */
3337 if (fdp->fd_lastfile > 0)
3338 n += fdp->fd_lastfile;
3341 sx_sunlock(&allproc_lock);
3342 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3345 bzero(&xf, sizeof(xf));
3346 xf.xf_size = sizeof(xf);
3347 sx_slock(&allproc_lock);
3348 FOREACH_PROC_IN_SYSTEM(p) {
3350 if (p->p_state == PRS_NEW) {
3354 if (p_cansee(req->td, p) != 0) {
3358 xf.xf_pid = p->p_pid;
3359 xf.xf_uid = p->p_ucred->cr_uid;
3364 FILEDESC_SLOCK(fdp);
3365 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3366 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3369 xf.xf_file = (uintptr_t)fp;
3370 xf.xf_data = (uintptr_t)fp->f_data;
3371 xf.xf_vnode = (uintptr_t)fp->f_vnode;
3372 xf.xf_type = (uintptr_t)fp->f_type;
3373 xf.xf_count = fp->f_count;
3375 xf.xf_offset = foffset_get(fp);
3376 xf.xf_flag = fp->f_flag;
3377 error = SYSCTL_OUT(req, &xf, sizeof(xf));
3381 FILEDESC_SUNLOCK(fdp);
3386 sx_sunlock(&allproc_lock);
3390 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3391 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3393 #ifdef KINFO_FILE_SIZE
3394 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3398 xlate_fflags(int fflags)
3400 static const struct {
3403 } fflags_table[] = {
3404 { FAPPEND, KF_FLAG_APPEND },
3405 { FASYNC, KF_FLAG_ASYNC },
3406 { FFSYNC, KF_FLAG_FSYNC },
3407 { FHASLOCK, KF_FLAG_HASLOCK },
3408 { FNONBLOCK, KF_FLAG_NONBLOCK },
3409 { FREAD, KF_FLAG_READ },
3410 { FWRITE, KF_FLAG_WRITE },
3411 { O_CREAT, KF_FLAG_CREAT },
3412 { O_DIRECT, KF_FLAG_DIRECT },
3413 { O_EXCL, KF_FLAG_EXCL },
3414 { O_EXEC, KF_FLAG_EXEC },
3415 { O_EXLOCK, KF_FLAG_EXLOCK },
3416 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3417 { O_SHLOCK, KF_FLAG_SHLOCK },
3418 { O_TRUNC, KF_FLAG_TRUNC }
3424 for (i = 0; i < nitems(fflags_table); i++)
3425 if (fflags & fflags_table[i].fflag)
3426 kflags |= fflags_table[i].kf_fflag;
3430 /* Trim unused data from kf_path by truncating the structure size. */
3432 pack_kinfo(struct kinfo_file *kif)
3435 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3436 strlen(kif->kf_path) + 1;
3437 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3441 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3442 struct kinfo_file *kif, struct filedesc *fdp, int flags)
3446 bzero(kif, sizeof(*kif));
3448 /* Set a default type to allow for empty fill_kinfo() methods. */
3449 kif->kf_type = KF_TYPE_UNKNOWN;
3450 kif->kf_flags = xlate_fflags(fp->f_flag);
3451 if (rightsp != NULL)
3452 kif->kf_cap_rights = *rightsp;
3454 cap_rights_init(&kif->kf_cap_rights);
3456 kif->kf_ref_count = fp->f_count;
3457 kif->kf_offset = foffset_get(fp);
3460 * This may drop the filedesc lock, so the 'fp' cannot be
3461 * accessed after this call.
3463 error = fo_fill_kinfo(fp, kif, fdp);
3465 kif->kf_status |= KF_ATTR_VALID;
3466 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3469 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3473 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3474 struct kinfo_file *kif, int flags)
3478 bzero(kif, sizeof(*kif));
3480 kif->kf_type = KF_TYPE_VNODE;
3481 error = vn_fill_kinfo_vnode(vp, kif);
3483 kif->kf_status |= KF_ATTR_VALID;
3484 kif->kf_flags = xlate_fflags(fflags);
3485 cap_rights_init(&kif->kf_cap_rights);
3487 kif->kf_ref_count = -1;
3488 kif->kf_offset = -1;
3489 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3492 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3496 struct export_fd_buf {
3497 struct filedesc *fdp;
3500 struct kinfo_file kif;
3505 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3507 struct kinfo_file *kif;
3510 if (efbuf->remainder != -1) {
3511 if (efbuf->remainder < kif->kf_structsize) {
3512 /* Terminate export. */
3513 efbuf->remainder = 0;
3516 efbuf->remainder -= kif->kf_structsize;
3518 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3522 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3523 struct export_fd_buf *efbuf)
3527 if (efbuf->remainder == 0)
3529 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3531 FILEDESC_SUNLOCK(efbuf->fdp);
3532 error = export_kinfo_to_sb(efbuf);
3533 FILEDESC_SLOCK(efbuf->fdp);
3538 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3539 struct export_fd_buf *efbuf)
3543 if (efbuf->remainder == 0)
3545 if (efbuf->fdp != NULL)
3546 FILEDESC_SUNLOCK(efbuf->fdp);
3547 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3548 error = export_kinfo_to_sb(efbuf);
3549 if (efbuf->fdp != NULL)
3550 FILEDESC_SLOCK(efbuf->fdp);
3555 * Store a process file descriptor information to sbuf.
3557 * Takes a locked proc as argument, and returns with the proc unlocked.
3560 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
3564 struct filedesc *fdp;
3565 struct export_fd_buf *efbuf;
3566 struct vnode *cttyvp, *textvp, *tracevp;
3568 cap_rights_t rights;
3570 PROC_LOCK_ASSERT(p, MA_OWNED);
3573 tracevp = p->p_tracevp;
3574 if (tracevp != NULL)
3577 textvp = p->p_textvp;
3580 /* Controlling tty. */
3582 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3583 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3589 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3592 efbuf->remainder = maxlen;
3593 efbuf->flags = flags;
3594 if (tracevp != NULL)
3595 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3598 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3600 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3606 FILEDESC_SLOCK(fdp);
3607 /* working directory */
3608 if (fdp->fd_cdir != NULL) {
3609 vrefact(fdp->fd_cdir);
3610 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3612 /* root directory */
3613 if (fdp->fd_rdir != NULL) {
3614 vrefact(fdp->fd_rdir);
3615 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3617 /* jail directory */
3618 if (fdp->fd_jdir != NULL) {
3619 vrefact(fdp->fd_jdir);
3620 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3622 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3623 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3626 rights = *cap_rights(fdp, i);
3627 #else /* !CAPABILITIES */
3628 rights = cap_no_rights;
3631 * Create sysctl entry. It is OK to drop the filedesc
3632 * lock inside of export_file_to_sb() as we will
3633 * re-validate and re-evaluate its properties when the
3636 error = export_file_to_sb(fp, i, &rights, efbuf);
3637 if (error != 0 || efbuf->remainder == 0)
3640 FILEDESC_SUNLOCK(fdp);
3643 free(efbuf, M_TEMP);
3647 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
3650 * Get per-process file descriptors for use by procstat(1), et al.
3653 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3658 int error, error2, *name;
3662 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3663 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3664 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3669 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3670 error = kern_proc_filedesc_out(p, &sb, maxlen,
3671 KERN_FILEDESC_PACK_KINFO);
3672 error2 = sbuf_finish(&sb);
3674 return (error != 0 ? error : error2);
3677 #ifdef COMPAT_FREEBSD7
3678 #ifdef KINFO_OFILE_SIZE
3679 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3683 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3686 okif->kf_structsize = sizeof(*okif);
3687 okif->kf_type = kif->kf_type;
3688 okif->kf_fd = kif->kf_fd;
3689 okif->kf_ref_count = kif->kf_ref_count;
3690 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3691 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3692 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3693 okif->kf_offset = kif->kf_offset;
3694 if (kif->kf_type == KF_TYPE_VNODE)
3695 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
3697 okif->kf_vnode_type = KF_VTYPE_VNON;
3698 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3699 if (kif->kf_type == KF_TYPE_SOCKET) {
3700 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
3701 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
3702 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
3703 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
3704 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
3706 okif->kf_sa_local.ss_family = AF_UNSPEC;
3707 okif->kf_sa_peer.ss_family = AF_UNSPEC;
3712 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3713 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3718 FILEDESC_SUNLOCK(fdp);
3719 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3720 kinfo_to_okinfo(kif, okif);
3721 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3722 FILEDESC_SLOCK(fdp);
3727 * Get per-process file descriptors for use by procstat(1), et al.
3730 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3732 struct kinfo_ofile *okif;
3733 struct kinfo_file *kif;
3734 struct filedesc *fdp;
3735 int error, i, *name;
3740 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3747 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3748 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3749 FILEDESC_SLOCK(fdp);
3750 if (fdp->fd_cdir != NULL)
3751 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3753 if (fdp->fd_rdir != NULL)
3754 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3756 if (fdp->fd_jdir != NULL)
3757 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3759 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3760 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3762 export_file_to_kinfo(fp, i, NULL, kif, fdp,
3763 KERN_FILEDESC_PACK_KINFO);
3764 FILEDESC_SUNLOCK(fdp);
3765 kinfo_to_okinfo(kif, okif);
3766 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3767 FILEDESC_SLOCK(fdp);
3771 FILEDESC_SUNLOCK(fdp);
3778 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3779 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3780 "Process ofiledesc entries");
3781 #endif /* COMPAT_FREEBSD7 */
3784 vntype_to_kinfo(int vtype)
3789 } vtypes_table[] = {
3790 { VBAD, KF_VTYPE_VBAD },
3791 { VBLK, KF_VTYPE_VBLK },
3792 { VCHR, KF_VTYPE_VCHR },
3793 { VDIR, KF_VTYPE_VDIR },
3794 { VFIFO, KF_VTYPE_VFIFO },
3795 { VLNK, KF_VTYPE_VLNK },
3796 { VNON, KF_VTYPE_VNON },
3797 { VREG, KF_VTYPE_VREG },
3798 { VSOCK, KF_VTYPE_VSOCK }
3803 * Perform vtype translation.
3805 for (i = 0; i < nitems(vtypes_table); i++)
3806 if (vtypes_table[i].vtype == vtype)
3807 return (vtypes_table[i].kf_vtype);
3809 return (KF_VTYPE_UNKNOWN);
3812 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3813 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3814 "Process filedesc entries");
3817 * Store a process current working directory information to sbuf.
3819 * Takes a locked proc as argument, and returns with the proc unlocked.
3822 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
3824 struct filedesc *fdp;
3825 struct export_fd_buf *efbuf;
3828 PROC_LOCK_ASSERT(p, MA_OWNED);
3835 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3838 efbuf->remainder = maxlen;
3840 FILEDESC_SLOCK(fdp);
3841 if (fdp->fd_cdir == NULL)
3844 vrefact(fdp->fd_cdir);
3845 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3848 FILEDESC_SUNLOCK(fdp);
3850 free(efbuf, M_TEMP);
3855 * Get per-process current working directory.
3858 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3863 int error, error2, *name;
3867 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3868 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3869 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3874 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3875 error = kern_proc_cwd_out(p, &sb, maxlen);
3876 error2 = sbuf_finish(&sb);
3878 return (error != 0 ? error : error2);
3881 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3882 sysctl_kern_proc_cwd, "Process current working directory");
3886 * For the purposes of debugging, generate a human-readable string for the
3890 file_type_to_name(short type)
3918 case DTYPE_PROCDESC:
3920 case DTYPE_LINUXEFD:
3922 case DTYPE_LINUXTFD:
3930 * For the purposes of debugging, identify a process (if any, perhaps one of
3931 * many) that references the passed file in its file descriptor array. Return
3934 static struct proc *
3935 file_to_first_proc(struct file *fp)
3937 struct filedesc *fdp;
3941 FOREACH_PROC_IN_SYSTEM(p) {
3942 if (p->p_state == PRS_NEW)
3947 for (n = 0; n <= fdp->fd_lastfile; n++) {
3948 if (fp == fdp->fd_ofiles[n].fde_file)
3956 db_print_file(struct file *fp, int header)
3958 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
3962 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
3963 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
3964 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
3966 p = file_to_first_proc(fp);
3967 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
3968 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
3969 fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode,
3970 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3975 DB_SHOW_COMMAND(file, db_show_file)
3980 db_printf("usage: show file <addr>\n");
3983 fp = (struct file *)addr;
3984 db_print_file(fp, 1);
3987 DB_SHOW_COMMAND(files, db_show_files)
3989 struct filedesc *fdp;
3996 FOREACH_PROC_IN_SYSTEM(p) {
3997 if (p->p_state == PRS_NEW)
3999 if ((fdp = p->p_fd) == NULL)
4001 for (n = 0; n <= fdp->fd_lastfile; ++n) {
4002 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4004 db_print_file(fp, header);
4011 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4012 &maxfilesperproc, 0, "Maximum files allowed open per process");
4014 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4015 &maxfiles, 0, "Maximum number of files");
4017 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4018 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
4022 filelistinit(void *dummy)
4025 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4026 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4027 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4028 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4029 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4031 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4033 /*-------------------------------------------------------------------*/
4036 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4037 int flags, struct thread *td)
4044 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4052 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4060 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4068 badfo_kqfilter(struct file *fp, struct knote *kn)
4075 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4083 badfo_close(struct file *fp, struct thread *td)
4090 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4098 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4106 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4107 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4115 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4121 struct fileops badfileops = {
4122 .fo_read = badfo_readwrite,
4123 .fo_write = badfo_readwrite,
4124 .fo_truncate = badfo_truncate,
4125 .fo_ioctl = badfo_ioctl,
4126 .fo_poll = badfo_poll,
4127 .fo_kqfilter = badfo_kqfilter,
4128 .fo_stat = badfo_stat,
4129 .fo_close = badfo_close,
4130 .fo_chmod = badfo_chmod,
4131 .fo_chown = badfo_chown,
4132 .fo_sendfile = badfo_sendfile,
4133 .fo_fill_kinfo = badfo_fill_kinfo,
4137 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4138 int flags, struct thread *td)
4141 return (EOPNOTSUPP);
4145 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4153 invfo_ioctl(struct file *fp, u_long com, void *data,
4154 struct ucred *active_cred, struct thread *td)
4161 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4165 return (poll_no_poll(events));
4169 invfo_kqfilter(struct file *fp, struct knote *kn)
4176 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4184 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4192 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4193 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4200 /*-------------------------------------------------------------------*/
4203 * File Descriptor pseudo-device driver (/dev/fd/).
4205 * Opening minor device N dup()s the file (if any) connected to file
4206 * descriptor N belonging to the calling process. Note that this driver
4207 * consists of only the ``open()'' routine, because all subsequent
4208 * references to this file will be direct to the other driver.
4210 * XXX: we could give this one a cloning event handler if necessary.
4215 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4219 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4220 * the file descriptor being sought for duplication. The error
4221 * return ensures that the vnode for this device will be released
4222 * by vn_open. Open will detect this special error and take the
4223 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4224 * will simply report the error.
4226 td->td_dupfd = dev2unit(dev);
4230 static struct cdevsw fildesc_cdevsw = {
4231 .d_version = D_VERSION,
4237 fildesc_drvinit(void *unused)
4241 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4242 UID_ROOT, GID_WHEEL, 0666, "fd/0");
4243 make_dev_alias(dev, "stdin");
4244 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4245 UID_ROOT, GID_WHEEL, 0666, "fd/1");
4246 make_dev_alias(dev, "stdout");
4247 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4248 UID_ROOT, GID_WHEEL, 0666, "fd/2");
4249 make_dev_alias(dev, "stderr");
4252 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);