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>
67 #include <sys/protosw.h>
68 #include <sys/racct.h>
69 #include <sys/resourcevar.h>
71 #include <sys/signalvar.h>
76 #include <sys/syscallsubr.h>
77 #include <sys/sysctl.h>
78 #include <sys/sysproto.h>
79 #include <sys/unistd.h>
81 #include <sys/vnode.h>
82 #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_PWD, "pwd", "Descriptor table vnodes");
95 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
96 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
97 "file desc to leader structures");
98 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
99 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
101 MALLOC_DECLARE(M_FADVISE);
103 static __read_mostly uma_zone_t file_zone;
104 static __read_mostly uma_zone_t filedesc0_zone;
105 __read_mostly uma_zone_t pwd_zone;
108 static int closefp(struct filedesc *fdp, int fd, struct file *fp,
109 struct thread *td, bool holdleaders, bool audit);
110 static int fd_first_free(struct filedesc *fdp, int low, int size);
111 static void fdgrowtable(struct filedesc *fdp, int nfd);
112 static void fdgrowtable_exp(struct filedesc *fdp, int nfd);
113 static void fdunused(struct filedesc *fdp, int fd);
114 static void fdused(struct filedesc *fdp, int fd);
115 static int getmaxfd(struct thread *td);
116 static u_long *filecaps_copy_prep(const struct filecaps *src);
117 static void filecaps_copy_finish(const struct filecaps *src,
118 struct filecaps *dst, u_long *ioctls);
119 static u_long *filecaps_free_prep(struct filecaps *fcaps);
120 static void filecaps_free_finish(u_long *ioctls);
122 static struct pwd *pwd_alloc(void);
127 * - An array of open file descriptors (fd_ofiles)
128 * - An array of file flags (fd_ofileflags)
129 * - A bitmap recording which descriptors are in use (fd_map)
131 * A process starts out with NDFILE descriptors. The value of NDFILE has
132 * been selected based the historical limit of 20 open files, and an
133 * assumption that the majority of processes, especially short-lived
134 * processes like shells, will never need more.
136 * If this initial allocation is exhausted, a larger descriptor table and
137 * map are allocated dynamically, and the pointers in the process's struct
138 * filedesc are updated to point to those. This is repeated every time
139 * the process runs out of file descriptors (provided it hasn't hit its
142 * Since threads may hold references to individual descriptor table
143 * entries, the tables are never freed. Instead, they are placed on a
144 * linked list and freed only when the struct filedesc is released.
147 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
148 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
149 #define NDSLOT(x) ((x) / NDENTRIES)
150 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
151 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
154 * SLIST entry used to keep track of ofiles which must be reclaimed when
158 struct fdescenttbl *ft_table;
159 SLIST_ENTRY(freetable) ft_next;
163 * Initial allocation: a filedesc structure + the head of SLIST used to
164 * keep track of old ofiles + enough space for NDFILE descriptors.
167 struct fdescenttbl0 {
169 struct filedescent fdt_ofiles[NDFILE];
173 struct filedesc fd_fd;
174 SLIST_HEAD(, freetable) fd_free;
175 struct fdescenttbl0 fd_dfiles;
176 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
180 * Descriptor management.
182 static int __exclusive_cache_line openfiles; /* actual number of open files */
183 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
184 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
187 * If low >= size, just return low. Otherwise find the first zero bit in the
188 * given bitmap, starting at low and not exceeding size - 1. Return size if
192 fd_first_free(struct filedesc *fdp, int low, int size)
194 NDSLOTTYPE *map = fdp->fd_map;
202 if (low % NDENTRIES) {
203 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
204 if ((mask &= ~map[off]) != 0UL)
205 return (off * NDENTRIES + ffsl(mask) - 1);
208 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
209 if (map[off] != ~0UL)
210 return (off * NDENTRIES + ffsl(~map[off]) - 1);
215 * Find the last used fd.
217 * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
218 * Otherwise use fdlastfile.
221 fdlastfile_single(struct filedesc *fdp)
223 NDSLOTTYPE *map = fdp->fd_map;
226 off = NDSLOT(fdp->fd_nfiles - 1);
227 for (minoff = NDSLOT(0); off >= minoff; --off)
229 return (off * NDENTRIES + flsl(map[off]) - 1);
234 fdlastfile(struct filedesc *fdp)
237 FILEDESC_LOCK_ASSERT(fdp);
238 return (fdlastfile_single(fdp));
242 fdisused(struct filedesc *fdp, int fd)
245 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
246 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
248 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
252 * Mark a file descriptor as used.
255 fdused_init(struct filedesc *fdp, int fd)
258 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
260 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
264 fdused(struct filedesc *fdp, int fd)
267 FILEDESC_XLOCK_ASSERT(fdp);
269 fdused_init(fdp, fd);
270 if (fd == fdp->fd_freefile)
275 * Mark a file descriptor as unused.
278 fdunused(struct filedesc *fdp, int fd)
281 FILEDESC_XLOCK_ASSERT(fdp);
283 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
284 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
285 ("fd=%d is still in use", fd));
287 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
288 if (fd < fdp->fd_freefile)
289 fdp->fd_freefile = fd;
293 * Free a file descriptor.
295 * Avoid some work if fdp is about to be destroyed.
298 fdefree_last(struct filedescent *fde)
301 filecaps_free(&fde->fde_caps);
305 fdfree(struct filedesc *fdp, int fd)
307 struct filedescent *fde;
309 FILEDESC_XLOCK_ASSERT(fdp);
310 fde = &fdp->fd_ofiles[fd];
312 seqc_write_begin(&fde->fde_seqc);
314 fde->fde_file = NULL;
316 seqc_write_end(&fde->fde_seqc);
323 * System calls on descriptors.
325 #ifndef _SYS_SYSPROTO_H_
326 struct getdtablesize_args {
332 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
338 td->td_retval[0] = getmaxfd(td);
340 PROC_LOCK(td->td_proc);
341 lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
342 PROC_UNLOCK(td->td_proc);
343 if (lim < td->td_retval[0])
344 td->td_retval[0] = lim;
350 * Duplicate a file descriptor to a particular value.
352 * Note: keep in mind that a potential race condition exists when closing
353 * descriptors from a shared descriptor table (via rfork).
355 #ifndef _SYS_SYSPROTO_H_
363 sys_dup2(struct thread *td, struct dup2_args *uap)
366 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
370 * Duplicate a file descriptor.
372 #ifndef _SYS_SYSPROTO_H_
379 sys_dup(struct thread *td, struct dup_args *uap)
382 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
386 * The file control system call.
388 #ifndef _SYS_SYSPROTO_H_
397 sys_fcntl(struct thread *td, struct fcntl_args *uap)
400 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
404 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
418 * Convert old flock structure to new.
420 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
421 fl.l_start = ofl.l_start;
422 fl.l_len = ofl.l_len;
423 fl.l_pid = ofl.l_pid;
424 fl.l_type = ofl.l_type;
425 fl.l_whence = ofl.l_whence;
439 arg1 = (intptr_t)&fl;
445 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
446 arg1 = (intptr_t)&fl;
454 error = kern_fcntl(td, fd, newcmd, arg1);
457 if (cmd == F_OGETLK) {
458 ofl.l_start = fl.l_start;
459 ofl.l_len = fl.l_len;
460 ofl.l_pid = fl.l_pid;
461 ofl.l_type = fl.l_type;
462 ofl.l_whence = fl.l_whence;
463 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
464 } else if (cmd == F_GETLK) {
465 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
471 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
473 struct filedesc *fdp;
475 struct file *fp, *fp2;
476 struct filedescent *fde;
480 int error, flg, seals, tmp;
494 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
497 case F_DUPFD_CLOEXEC:
499 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
504 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
507 case F_DUP2FD_CLOEXEC:
509 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
515 fde = fdeget_locked(fdp, fd);
518 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
521 FILEDESC_SUNLOCK(fdp);
527 fde = fdeget_locked(fdp, fd);
529 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
530 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
533 FILEDESC_XUNLOCK(fdp);
537 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
540 td->td_retval[0] = OFLAGS(fp->f_flag);
545 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
548 if (fp->f_ops == &path_fileops) {
554 tmp = flg = fp->f_flag;
556 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
557 } while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
558 tmp = fp->f_flag & FNONBLOCK;
559 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
564 tmp = fp->f_flag & FASYNC;
565 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
570 atomic_clear_int(&fp->f_flag, FNONBLOCK);
572 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
577 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
580 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
582 td->td_retval[0] = tmp;
587 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
591 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
596 error = priv_check(td, PRIV_NFS_LOCKD);
604 /* FALLTHROUGH F_SETLK */
608 flp = (struct flock *)arg;
609 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
614 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
617 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
623 if (flp->l_whence == SEEK_CUR) {
624 foffset = foffset_get(fp);
627 foffset > OFF_MAX - flp->l_start)) {
632 flp->l_start += foffset;
636 switch (flp->l_type) {
638 if ((fp->f_flag & FREAD) == 0) {
642 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
643 PROC_LOCK(p->p_leader);
644 p->p_leader->p_flag |= P_ADVLOCK;
645 PROC_UNLOCK(p->p_leader);
647 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
651 if ((fp->f_flag & FWRITE) == 0) {
655 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
656 PROC_LOCK(p->p_leader);
657 p->p_leader->p_flag |= P_ADVLOCK;
658 PROC_UNLOCK(p->p_leader);
660 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
664 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
668 if (flg != F_REMOTE) {
672 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
673 F_UNLCKSYS, flp, flg);
679 if (error != 0 || flp->l_type == F_UNLCK ||
680 flp->l_type == F_UNLCKSYS) {
686 * Check for a race with close.
688 * The vnode is now advisory locked (or unlocked, but this case
689 * is not really important) as the caller requested.
690 * We had to drop the filedesc lock, so we need to recheck if
691 * the descriptor is still valid, because if it was closed
692 * in the meantime we need to remove advisory lock from the
693 * vnode - close on any descriptor leading to an advisory
694 * locked vnode, removes that lock.
695 * We will return 0 on purpose in that case, as the result of
696 * successful advisory lock might have been externally visible
697 * already. This is fine - effectively we pretend to the caller
698 * that the closing thread was a bit slower and that the
699 * advisory lock succeeded before the close.
701 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
707 flp->l_whence = SEEK_SET;
710 flp->l_type = F_UNLCK;
711 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
712 F_UNLCK, flp, F_POSIX);
719 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
722 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
727 flp = (struct flock *)arg;
728 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
729 flp->l_type != F_UNLCK) {
734 if (flp->l_whence == SEEK_CUR) {
735 foffset = foffset_get(fp);
736 if ((flp->l_start > 0 &&
737 foffset > OFF_MAX - flp->l_start) ||
739 foffset < OFF_MIN - flp->l_start)) {
744 flp->l_start += foffset;
747 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
753 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
756 error = fo_add_seals(fp, arg);
761 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
764 if (fo_get_seals(fp, &seals) == 0)
765 td->td_retval[0] = seals;
772 arg = arg ? 128 * 1024: 0;
775 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
778 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
784 if (vp->v_type != VREG) {
791 * Exclusive lock synchronizes against f_seqcount reads and
792 * writes in sequential_heuristic().
794 error = vn_lock(vp, LK_EXCLUSIVE);
800 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
801 arg = MIN(arg, INT_MAX - bsize + 1);
802 fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
803 (arg + bsize - 1) / bsize);
804 atomic_set_int(&fp->f_flag, FRDAHEAD);
806 atomic_clear_int(&fp->f_flag, FRDAHEAD);
814 * Check if the vnode is part of a union stack (either the
815 * "union" flag from mount(2) or unionfs).
817 * Prior to introduction of this op libc's readdir would call
818 * fstatfs(2), in effect unnecessarily copying kilobytes of
819 * data just to check fs name and a mount flag.
821 * Fixing the code to handle everything in the kernel instead
822 * is a non-trivial endeavor and has low priority, thus this
823 * horrible kludge facilitates the current behavior in a much
824 * cheaper manner until someone(tm) sorts this out.
826 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
829 if (fp->f_type != DTYPE_VNODE) {
836 * Since we don't prevent dooming the vnode even non-null mp
837 * found can become immediately stale. This is tolerable since
838 * mount points are type-stable (providing safe memory access)
839 * and any vfs op on this vnode going forward will return an
840 * error (meaning return value in this case is meaningless).
842 mp = atomic_load_ptr(&vp->v_mount);
843 if (__predict_false(mp == NULL)) {
848 td->td_retval[0] = 0;
849 if (mp->mnt_kern_flag & MNTK_UNIONFS ||
850 mp->mnt_flag & MNT_UNION)
851 td->td_retval[0] = 1;
863 getmaxfd(struct thread *td)
866 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
870 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
873 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
875 struct filedesc *fdp;
876 struct filedescent *oldfde, *newfde;
878 struct file *delfp, *oldfp;
879 u_long *oioctls, *nioctls;
886 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
887 MPASS(mode < FDDUP_LASTMODE);
890 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
893 * Verify we have a valid descriptor to dup from and possibly to
894 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
895 * return EINVAL when the new descriptor is out of bounds.
900 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
901 maxfd = getmaxfd(td);
903 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
907 if (fget_locked(fdp, old) == NULL)
909 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
910 td->td_retval[0] = new;
911 if (flags & FDDUP_FLAG_CLOEXEC)
912 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
917 oldfde = &fdp->fd_ofiles[old];
918 oldfp = oldfde->fde_file;
923 * If the caller specified a file descriptor, make sure the file
924 * table is large enough to hold it, and grab it. Otherwise, just
925 * allocate a new descriptor the usual way.
930 if ((error = fdalloc(td, new, &new)) != 0) {
935 case FDDUP_MUSTREPLACE:
936 /* Target file descriptor must exist. */
937 if (fget_locked(fdp, new) == NULL) {
943 if (new >= fdp->fd_nfiles) {
945 * The resource limits are here instead of e.g.
946 * fdalloc(), because the file descriptor table may be
947 * shared between processes, so we can't really use
948 * racct_add()/racct_sub(). Instead of counting the
949 * number of actually allocated descriptors, just put
950 * the limit on the size of the file descriptor table.
953 if (RACCT_ENABLED()) {
954 error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
962 fdgrowtable_exp(fdp, new + 1);
964 if (!fdisused(fdp, new))
968 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
971 KASSERT(old != new, ("new fd is same as old"));
973 /* Refetch oldfde because the table may have grown and old one freed. */
974 oldfde = &fdp->fd_ofiles[old];
975 KASSERT(oldfp == oldfde->fde_file,
976 ("fdt_ofiles shift from growth observed at fd %d",
979 newfde = &fdp->fd_ofiles[new];
980 delfp = newfde->fde_file;
982 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
985 * Duplicate the source descriptor.
988 seqc_write_begin(&newfde->fde_seqc);
990 oioctls = filecaps_free_prep(&newfde->fde_caps);
991 memcpy(newfde, oldfde, fde_change_size);
992 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
994 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
995 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
997 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
999 seqc_write_end(&newfde->fde_seqc);
1001 td->td_retval[0] = new;
1005 if (delfp != NULL) {
1006 (void) closefp(fdp, new, delfp, td, true, false);
1007 FILEDESC_UNLOCK_ASSERT(fdp);
1010 FILEDESC_XUNLOCK(fdp);
1013 filecaps_free_finish(oioctls);
1018 sigiofree(struct sigio *sigio)
1020 crfree(sigio->sio_ucred);
1021 free(sigio, M_SIGIO);
1024 static struct sigio *
1025 funsetown_locked(struct sigio *sigio)
1030 SIGIO_ASSERT_LOCKED();
1034 *sigio->sio_myref = NULL;
1035 if (sigio->sio_pgid < 0) {
1036 pg = sigio->sio_pgrp;
1038 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio);
1041 p = sigio->sio_proc;
1043 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio);
1050 * If sigio is on the list associated with a process or process group,
1051 * disable signalling from the device, remove sigio from the list and
1055 funsetown(struct sigio **sigiop)
1057 struct sigio *sigio;
1059 /* Racy check, consumers must provide synchronization. */
1060 if (*sigiop == NULL)
1064 sigio = funsetown_locked(*sigiop);
1071 * Free a list of sigio structures. The caller must ensure that new sigio
1072 * structures cannot be added after this point. For process groups this is
1073 * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1077 funsetownlst(struct sigiolst *sigiolst)
1081 struct sigio *sigio, *tmp;
1084 sigio = SLIST_FIRST(sigiolst);
1092 sigio = SLIST_FIRST(sigiolst);
1093 if (sigio == NULL) {
1099 * Every entry of the list should belong to a single proc or pgrp.
1101 if (sigio->sio_pgid < 0) {
1102 pg = sigio->sio_pgrp;
1103 sx_assert(&proctree_lock, SX_XLOCKED);
1105 } else /* if (sigio->sio_pgid > 0) */ {
1106 p = sigio->sio_proc;
1108 KASSERT((p->p_flag & P_WEXIT) != 0,
1109 ("%s: process %p is not exiting", __func__, p));
1112 SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1113 *sigio->sio_myref = NULL;
1115 KASSERT(sigio->sio_pgid < 0,
1116 ("Proc sigio in pgrp sigio list"));
1117 KASSERT(sigio->sio_pgrp == pg,
1118 ("Bogus pgrp in sigio list"));
1119 } else /* if (p != NULL) */ {
1120 KASSERT(sigio->sio_pgid > 0,
1121 ("Pgrp sigio in proc sigio list"));
1122 KASSERT(sigio->sio_proc == p,
1123 ("Bogus proc in sigio list"));
1133 SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1138 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1140 * After permission checking, add a sigio structure to the sigio list for
1141 * the process or process group.
1144 fsetown(pid_t pgid, struct sigio **sigiop)
1148 struct sigio *osigio, *sigio;
1156 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1157 sigio->sio_pgid = pgid;
1158 sigio->sio_ucred = crhold(curthread->td_ucred);
1159 sigio->sio_myref = sigiop;
1163 ret = pget(pgid, PGET_NOTWEXIT | PGET_NOTID | PGET_HOLD, &proc);
1165 osigio = funsetown_locked(*sigiop);
1169 if ((proc->p_flag & P_WEXIT) != 0) {
1171 } else if (proc->p_session !=
1172 curthread->td_proc->p_session) {
1174 * Policy - Don't allow a process to FSETOWN a
1175 * process in another session.
1177 * Remove this test to allow maximum flexibility
1178 * or restrict FSETOWN to the current process or
1179 * process group for maximum safety.
1183 sigio->sio_proc = proc;
1184 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio,
1189 } else /* if (pgid < 0) */ {
1190 sx_slock(&proctree_lock);
1192 osigio = funsetown_locked(*sigiop);
1193 pgrp = pgfind(-pgid);
1197 if (pgrp->pg_session != curthread->td_proc->p_session) {
1199 * Policy - Don't allow a process to FSETOWN a
1200 * process in another session.
1202 * Remove this test to allow maximum flexibility
1203 * or restrict FSETOWN to the current process or
1204 * process group for maximum safety.
1208 sigio->sio_pgrp = pgrp;
1209 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio,
1214 sx_sunlock(&proctree_lock);
1225 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1228 fgetown(struct sigio **sigiop)
1233 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1239 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1244 FILEDESC_XLOCK_ASSERT(fdp);
1247 * We now hold the fp reference that used to be owned by the
1248 * descriptor array. We have to unlock the FILEDESC *AFTER*
1249 * knote_fdclose to prevent a race of the fd getting opened, a knote
1250 * added, and deleteing a knote for the new fd.
1252 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1253 knote_fdclose(td, fd);
1256 * We need to notify mqueue if the object is of type mqueue.
1258 if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1259 mq_fdclose(td, fd, fp);
1260 FILEDESC_XUNLOCK(fdp);
1263 if (AUDITING_TD(td) && audit)
1264 audit_sysclose(td, fd, fp);
1266 error = closef(fp, td);
1269 * All paths leading up to closefp() will have already removed or
1270 * replaced the fd in the filedesc table, so a restart would not
1271 * operate on the same file.
1273 if (error == ERESTART)
1280 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1281 bool holdleaders, bool audit)
1285 FILEDESC_XLOCK_ASSERT(fdp);
1288 if (td->td_proc->p_fdtol != NULL) {
1290 * Ask fdfree() to sleep to ensure that all relevant
1291 * process leaders can be traversed in closef().
1293 fdp->fd_holdleaderscount++;
1295 holdleaders = false;
1299 error = closefp_impl(fdp, fd, fp, td, audit);
1301 FILEDESC_XLOCK(fdp);
1302 fdp->fd_holdleaderscount--;
1303 if (fdp->fd_holdleaderscount == 0 &&
1304 fdp->fd_holdleaderswakeup != 0) {
1305 fdp->fd_holdleaderswakeup = 0;
1306 wakeup(&fdp->fd_holdleaderscount);
1308 FILEDESC_XUNLOCK(fdp);
1314 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1315 bool holdleaders, bool audit)
1318 FILEDESC_XLOCK_ASSERT(fdp);
1320 if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1321 return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1323 return (closefp_impl(fdp, fd, fp, td, audit));
1328 * Close a file descriptor.
1330 #ifndef _SYS_SYSPROTO_H_
1337 sys_close(struct thread *td, struct close_args *uap)
1340 return (kern_close(td, uap->fd));
1344 kern_close(struct thread *td, int fd)
1346 struct filedesc *fdp;
1349 fdp = td->td_proc->p_fd;
1351 FILEDESC_XLOCK(fdp);
1352 if ((fp = fget_locked(fdp, fd)) == NULL) {
1353 FILEDESC_XUNLOCK(fdp);
1358 /* closefp() drops the FILEDESC lock for us. */
1359 return (closefp(fdp, fd, fp, td, true, true));
1363 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1365 struct filedesc *fdp;
1366 const struct fdescenttbl *fdt;
1371 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1372 * open should not be a usage error. From a close_range() perspective,
1373 * close_range(3, ~0U, 0) in the same scenario should also likely not
1374 * be a usage error as all fd above 3 are in-fact already closed.
1376 if (highfd < lowfd) {
1380 fdp = td->td_proc->p_fd;
1381 FILEDESC_XLOCK(fdp);
1382 fdt = atomic_load_ptr(&fdp->fd_files);
1383 highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1385 if (__predict_false(fd > highfd)) {
1389 fp = fdt->fdt_ofiles[fd].fde_file;
1395 (void) closefp(fdp, fd, fp, td, true, true);
1398 FILEDESC_XLOCK(fdp);
1399 fdt = atomic_load_ptr(&fdp->fd_files);
1404 FILEDESC_XUNLOCK(fdp);
1409 #ifndef _SYS_SYSPROTO_H_
1410 struct close_range_args {
1417 sys_close_range(struct thread *td, struct close_range_args *uap)
1420 AUDIT_ARG_FD(uap->lowfd);
1421 AUDIT_ARG_CMD(uap->highfd);
1422 AUDIT_ARG_FFLAGS(uap->flags);
1424 /* No flags currently defined */
1425 if (uap->flags != 0)
1427 return (kern_close_range(td, uap->lowfd, uap->highfd));
1430 #ifdef COMPAT_FREEBSD12
1432 * Close open file descriptors.
1434 #ifndef _SYS_SYSPROTO_H_
1435 struct freebsd12_closefrom_args {
1441 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1445 AUDIT_ARG_FD(uap->lowfd);
1448 * Treat negative starting file descriptor values identical to
1449 * closefrom(0) which closes all files.
1451 lowfd = MAX(0, uap->lowfd);
1452 return (kern_close_range(td, lowfd, ~0U));
1454 #endif /* COMPAT_FREEBSD12 */
1456 #if defined(COMPAT_43)
1458 * Return status information about a file descriptor.
1460 #ifndef _SYS_SYSPROTO_H_
1461 struct ofstat_args {
1468 ofstat(struct thread *td, struct ofstat_args *uap)
1474 error = kern_fstat(td, uap->fd, &ub);
1477 error = copyout(&oub, uap->sb, sizeof(oub));
1481 #endif /* COMPAT_43 */
1483 #if defined(COMPAT_FREEBSD11)
1485 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1488 struct freebsd11_stat osb;
1491 error = kern_fstat(td, uap->fd, &sb);
1494 error = freebsd11_cvtstat(&sb, &osb);
1496 error = copyout(&osb, uap->sb, sizeof(osb));
1499 #endif /* COMPAT_FREEBSD11 */
1502 * Return status information about a file descriptor.
1504 #ifndef _SYS_SYSPROTO_H_
1512 sys_fstat(struct thread *td, struct fstat_args *uap)
1517 error = kern_fstat(td, uap->fd, &ub);
1519 error = copyout(&ub, uap->sb, sizeof(ub));
1524 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1531 error = fget(td, fd, &cap_fstat_rights, &fp);
1532 if (__predict_false(error != 0))
1535 AUDIT_ARG_FILE(td->td_proc, fp);
1537 error = fo_stat(fp, sbp, td->td_ucred, td);
1539 #ifdef __STAT_TIME_T_EXT
1540 sbp->st_atim_ext = 0;
1541 sbp->st_mtim_ext = 0;
1542 sbp->st_ctim_ext = 0;
1543 sbp->st_btim_ext = 0;
1546 if (KTRPOINT(td, KTR_STRUCT))
1547 ktrstat_error(sbp, error);
1552 #if defined(COMPAT_FREEBSD11)
1554 * Return status information about a file descriptor.
1556 #ifndef _SYS_SYSPROTO_H_
1557 struct freebsd11_nfstat_args {
1564 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1570 error = kern_fstat(td, uap->fd, &ub);
1572 freebsd11_cvtnstat(&ub, &nub);
1573 error = copyout(&nub, uap->sb, sizeof(nub));
1577 #endif /* COMPAT_FREEBSD11 */
1580 * Return pathconf information about a file descriptor.
1582 #ifndef _SYS_SYSPROTO_H_
1583 struct fpathconf_args {
1590 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1595 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1597 td->td_retval[0] = value;
1602 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1608 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1612 if (name == _PC_ASYNC_IO) {
1613 *valuep = _POSIX_ASYNCHRONOUS_IO;
1618 vn_lock(vp, LK_SHARED | LK_RETRY);
1619 error = VOP_PATHCONF(vp, name, valuep);
1621 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1622 if (name != _PC_PIPE_BUF) {
1637 * Copy filecaps structure allocating memory for ioctls array if needed.
1639 * The last parameter indicates whether the fdtable is locked. If it is not and
1640 * ioctls are encountered, copying fails and the caller must lock the table.
1642 * Note that if the table was not locked, the caller has to check the relevant
1643 * sequence counter to determine whether the operation was successful.
1646 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1650 if (src->fc_ioctls != NULL && !locked)
1652 memcpy(dst, src, sizeof(*src));
1653 if (src->fc_ioctls == NULL)
1656 KASSERT(src->fc_nioctls > 0,
1657 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1659 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1660 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1661 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1666 filecaps_copy_prep(const struct filecaps *src)
1671 if (__predict_true(src->fc_ioctls == NULL))
1674 KASSERT(src->fc_nioctls > 0,
1675 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1677 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1678 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1683 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1689 if (__predict_true(src->fc_ioctls == NULL)) {
1690 MPASS(ioctls == NULL);
1694 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1695 dst->fc_ioctls = ioctls;
1696 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1700 * Move filecaps structure to the new place and clear the old place.
1703 filecaps_move(struct filecaps *src, struct filecaps *dst)
1707 bzero(src, sizeof(*src));
1711 * Fill the given filecaps structure with full rights.
1714 filecaps_fill(struct filecaps *fcaps)
1717 CAP_ALL(&fcaps->fc_rights);
1718 fcaps->fc_ioctls = NULL;
1719 fcaps->fc_nioctls = -1;
1720 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1724 * Free memory allocated within filecaps structure.
1727 filecaps_free(struct filecaps *fcaps)
1730 free(fcaps->fc_ioctls, M_FILECAPS);
1731 bzero(fcaps, sizeof(*fcaps));
1735 filecaps_free_prep(struct filecaps *fcaps)
1739 ioctls = fcaps->fc_ioctls;
1740 bzero(fcaps, sizeof(*fcaps));
1745 filecaps_free_finish(u_long *ioctls)
1748 free(ioctls, M_FILECAPS);
1752 * Validate the given filecaps structure.
1755 filecaps_validate(const struct filecaps *fcaps, const char *func)
1758 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1759 ("%s: invalid rights", func));
1760 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1761 ("%s: invalid fcntls", func));
1762 KASSERT(fcaps->fc_fcntls == 0 ||
1763 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1764 ("%s: fcntls without CAP_FCNTL", func));
1765 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1766 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1767 ("%s: invalid ioctls", func));
1768 KASSERT(fcaps->fc_nioctls == 0 ||
1769 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1770 ("%s: ioctls without CAP_IOCTL", func));
1774 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1778 FILEDESC_XLOCK_ASSERT(fdp);
1780 nfd1 = fdp->fd_nfiles * 2;
1783 fdgrowtable(fdp, nfd1);
1787 * Grow the file table to accommodate (at least) nfd descriptors.
1790 fdgrowtable(struct filedesc *fdp, int nfd)
1792 struct filedesc0 *fdp0;
1793 struct freetable *ft;
1794 struct fdescenttbl *ntable;
1795 struct fdescenttbl *otable;
1796 int nnfiles, onfiles;
1797 NDSLOTTYPE *nmap, *omap;
1799 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1801 /* save old values */
1802 onfiles = fdp->fd_nfiles;
1803 otable = fdp->fd_files;
1806 /* compute the size of the new table */
1807 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1808 if (nnfiles <= onfiles)
1809 /* the table is already large enough */
1813 * Allocate a new table. We need enough space for the number of
1814 * entries, file entries themselves and the struct freetable we will use
1815 * when we decommission the table and place it on the freelist.
1816 * We place the struct freetable in the middle so we don't have
1817 * to worry about padding.
1819 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1820 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1821 sizeof(struct freetable),
1822 M_FILEDESC, M_ZERO | M_WAITOK);
1823 /* copy the old data */
1824 ntable->fdt_nfiles = nnfiles;
1825 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1826 onfiles * sizeof(ntable->fdt_ofiles[0]));
1829 * Allocate a new map only if the old is not large enough. It will
1830 * grow at a slower rate than the table as it can map more
1831 * entries than the table can hold.
1833 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1834 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1836 /* copy over the old data and update the pointer */
1837 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1842 * Make sure that ntable is correctly initialized before we replace
1843 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1846 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1849 * Free the old file table when not shared by other threads or processes.
1850 * The old file table is considered to be shared when either are true:
1851 * - The process has more than one thread.
1852 * - The file descriptor table has been shared via fdshare().
1854 * When shared, the old file table will be placed on a freelist
1855 * which will be processed when the struct filedesc is released.
1857 * Note that if onfiles == NDFILE, we're dealing with the original
1858 * static allocation contained within (struct filedesc0 *)fdp,
1859 * which must not be freed.
1861 if (onfiles > NDFILE) {
1863 * Note we may be called here from fdinit while allocating a
1864 * table for a new process in which case ->p_fd points
1867 if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) {
1868 free(otable, M_FILEDESC);
1870 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1871 fdp0 = (struct filedesc0 *)fdp;
1872 ft->ft_table = otable;
1873 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1877 * The map does not have the same possibility of threads still
1878 * holding references to it. So always free it as long as it
1879 * does not reference the original static allocation.
1881 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1882 free(omap, M_FILEDESC);
1886 * Allocate a file descriptor for the process.
1889 fdalloc(struct thread *td, int minfd, int *result)
1891 struct proc *p = td->td_proc;
1892 struct filedesc *fdp = p->p_fd;
1893 int fd, maxfd, allocfd;
1898 FILEDESC_XLOCK_ASSERT(fdp);
1900 if (fdp->fd_freefile > minfd)
1901 minfd = fdp->fd_freefile;
1903 maxfd = getmaxfd(td);
1906 * Search the bitmap for a free descriptor starting at minfd.
1907 * If none is found, grow the file table.
1909 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1910 if (__predict_false(fd >= maxfd))
1912 if (__predict_false(fd >= fdp->fd_nfiles)) {
1913 allocfd = min(fd * 2, maxfd);
1915 if (RACCT_ENABLED()) {
1916 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1922 * fd is already equal to first free descriptor >= minfd, so
1923 * we only need to grow the table and we are done.
1925 fdgrowtable_exp(fdp, allocfd);
1929 * Perform some sanity checks, then mark the file descriptor as
1930 * used and return it to the caller.
1932 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1933 ("invalid descriptor %d", fd));
1934 KASSERT(!fdisused(fdp, fd),
1935 ("fd_first_free() returned non-free descriptor"));
1936 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1937 ("file descriptor isn't free"));
1944 * Allocate n file descriptors for the process.
1947 fdallocn(struct thread *td, int minfd, int *fds, int n)
1949 struct proc *p = td->td_proc;
1950 struct filedesc *fdp = p->p_fd;
1953 FILEDESC_XLOCK_ASSERT(fdp);
1955 for (i = 0; i < n; i++)
1956 if (fdalloc(td, 0, &fds[i]) != 0)
1960 for (i--; i >= 0; i--)
1961 fdunused(fdp, fds[i]);
1969 * Create a new open file structure and allocate a file descriptor for the
1970 * process that refers to it. We add one reference to the file for the
1971 * descriptor table and one reference for resultfp. This is to prevent us
1972 * being preempted and the entry in the descriptor table closed after we
1973 * release the FILEDESC lock.
1976 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1977 struct filecaps *fcaps)
1982 MPASS(resultfp != NULL);
1983 MPASS(resultfd != NULL);
1985 error = _falloc_noinstall(td, &fp, 2);
1986 if (__predict_false(error != 0)) {
1990 error = finstall_refed(td, fp, &fd, flags, fcaps);
1991 if (__predict_false(error != 0)) {
1992 falloc_abort(td, fp);
2003 * Create a new open file structure without allocating a file descriptor.
2006 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2009 int maxuserfiles = maxfiles - (maxfiles / 20);
2011 static struct timeval lastfail;
2014 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2017 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2018 if ((openfiles_new >= maxuserfiles &&
2019 priv_check(td, PRIV_MAXFILES) != 0) ||
2020 openfiles_new >= maxfiles) {
2021 atomic_subtract_int(&openfiles, 1);
2022 if (ppsratecheck(&lastfail, &curfail, 1)) {
2023 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2024 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2028 fp = uma_zalloc(file_zone, M_WAITOK);
2029 bzero(fp, sizeof(*fp));
2030 refcount_init(&fp->f_count, n);
2031 fp->f_cred = crhold(td->td_ucred);
2032 fp->f_ops = &badfileops;
2038 falloc_abort(struct thread *td, struct file *fp)
2042 * For assertion purposes.
2044 refcount_init(&fp->f_count, 0);
2049 * Install a file in a file descriptor table.
2052 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2053 struct filecaps *fcaps)
2055 struct filedescent *fde;
2059 filecaps_validate(fcaps, __func__);
2060 FILEDESC_XLOCK_ASSERT(fdp);
2062 fde = &fdp->fd_ofiles[fd];
2064 seqc_write_begin(&fde->fde_seqc);
2067 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2069 filecaps_move(fcaps, &fde->fde_caps);
2071 filecaps_fill(&fde->fde_caps);
2073 seqc_write_end(&fde->fde_seqc);
2078 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2079 struct filecaps *fcaps)
2081 struct filedesc *fdp = td->td_proc->p_fd;
2086 FILEDESC_XLOCK(fdp);
2087 error = fdalloc(td, 0, fd);
2088 if (__predict_true(error == 0)) {
2089 _finstall(fdp, fp, *fd, flags, fcaps);
2091 FILEDESC_XUNLOCK(fdp);
2096 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2097 struct filecaps *fcaps)
2105 error = finstall_refed(td, fp, fd, flags, fcaps);
2106 if (__predict_false(error != 0)) {
2113 * Build a new filedesc structure from another.
2115 * If fdp is not NULL, return with it shared locked.
2118 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2120 struct filedesc0 *newfdp0;
2121 struct filedesc *newfdp;
2124 MPASS(lastfile != NULL);
2126 MPASS(lastfile == NULL);
2128 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2129 newfdp = &newfdp0->fd_fd;
2131 /* Create the file descriptor table. */
2132 FILEDESC_LOCK_INIT(newfdp);
2133 refcount_init(&newfdp->fd_refcnt, 1);
2134 refcount_init(&newfdp->fd_holdcnt, 1);
2135 newfdp->fd_map = newfdp0->fd_dmap;
2136 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2137 newfdp->fd_files->fdt_nfiles = NDFILE;
2142 FILEDESC_SLOCK(fdp);
2144 FILEDESC_SUNLOCK(fdp);
2149 *lastfile = fdlastfile(fdp);
2150 if (*lastfile < newfdp->fd_nfiles)
2152 FILEDESC_SUNLOCK(fdp);
2153 fdgrowtable(newfdp, *lastfile + 1);
2154 FILEDESC_SLOCK(fdp);
2161 * Build a pwddesc structure from another.
2162 * Copy the current, root, and jail root vnode references.
2164 * If pdp is not NULL, return with it shared locked.
2167 pdinit(struct pwddesc *pdp, bool keeplock)
2169 struct pwddesc *newpdp;
2172 newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2174 PWDDESC_LOCK_INIT(newpdp);
2175 refcount_init(&newpdp->pd_refcount, 1);
2176 newpdp->pd_cmask = CMASK;
2179 newpwd = pwd_alloc();
2180 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2185 newpwd = pwd_hold_pwddesc(pdp);
2186 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2188 PWDDESC_XUNLOCK(pdp);
2193 * Hold either filedesc or pwddesc of the passed process.
2195 * The process lock is used to synchronize against the target exiting and
2198 * Clearing can be ilustrated in 3 steps:
2199 * 1. set the pointer to NULL. Either routine can race against it, hence
2201 * 2. observe the process lock as not taken. Until then fdhold/pdhold can
2202 * race to either still see the pointer or find NULL. It is still safe to
2203 * grab a reference as clearing is stalled.
2204 * 3. after the lock is observed as not taken, any fdhold/pdhold calls are
2205 * guaranteed to see NULL, making it safe to finish clearing
2207 static struct filedesc *
2208 fdhold(struct proc *p)
2210 struct filedesc *fdp;
2212 PROC_LOCK_ASSERT(p, MA_OWNED);
2213 fdp = atomic_load_ptr(&p->p_fd);
2215 refcount_acquire(&fdp->fd_holdcnt);
2219 static struct pwddesc *
2220 pdhold(struct proc *p)
2222 struct pwddesc *pdp;
2224 PROC_LOCK_ASSERT(p, MA_OWNED);
2225 pdp = atomic_load_ptr(&p->p_pd);
2227 refcount_acquire(&pdp->pd_refcount);
2232 fddrop(struct filedesc *fdp)
2235 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2236 if (refcount_release(&fdp->fd_holdcnt) == 0)
2240 FILEDESC_LOCK_DESTROY(fdp);
2241 uma_zfree(filedesc0_zone, fdp);
2245 pddrop(struct pwddesc *pdp)
2249 if (refcount_release_if_not_last(&pdp->pd_refcount))
2253 if (refcount_release(&pdp->pd_refcount) == 0) {
2254 PWDDESC_XUNLOCK(pdp);
2257 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2259 PWDDESC_XUNLOCK(pdp);
2262 PWDDESC_LOCK_DESTROY(pdp);
2263 free(pdp, M_PWDDESC);
2267 * Share a filedesc structure.
2270 fdshare(struct filedesc *fdp)
2273 refcount_acquire(&fdp->fd_refcnt);
2278 * Share a pwddesc structure.
2281 pdshare(struct pwddesc *pdp)
2283 refcount_acquire(&pdp->pd_refcount);
2288 * Unshare a filedesc structure, if necessary by making a copy
2291 fdunshare(struct thread *td)
2293 struct filedesc *tmp;
2294 struct proc *p = td->td_proc;
2296 if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2299 tmp = fdcopy(p->p_fd);
2305 * Unshare a pwddesc structure.
2308 pdunshare(struct thread *td)
2310 struct pwddesc *pdp;
2315 if (p->p_pd->pd_refcount == 1)
2318 pdp = pdcopy(p->p_pd);
2324 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2328 td->td_proc->p_fd = fdp;
2332 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2333 * this is to ease callers, not catch errors.
2336 fdcopy(struct filedesc *fdp)
2338 struct filedesc *newfdp;
2339 struct filedescent *nfde, *ofde;
2344 newfdp = fdinit(fdp, true, &lastfile);
2345 /* copy all passable descriptors (i.e. not kqueue) */
2346 newfdp->fd_freefile = -1;
2347 for (i = 0; i <= lastfile; ++i) {
2348 ofde = &fdp->fd_ofiles[i];
2349 if (ofde->fde_file == NULL ||
2350 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2351 !fhold(ofde->fde_file)) {
2352 if (newfdp->fd_freefile == -1)
2353 newfdp->fd_freefile = i;
2356 nfde = &newfdp->fd_ofiles[i];
2358 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2359 fdused_init(newfdp, i);
2361 if (newfdp->fd_freefile == -1)
2362 newfdp->fd_freefile = i;
2363 FILEDESC_SUNLOCK(fdp);
2368 * Copy a pwddesc structure.
2371 pdcopy(struct pwddesc *pdp)
2373 struct pwddesc *newpdp;
2377 newpdp = pdinit(pdp, true);
2378 newpdp->pd_cmask = pdp->pd_cmask;
2379 PWDDESC_XUNLOCK(pdp);
2384 * Copies a filedesc structure, while remapping all file descriptors
2385 * stored inside using a translation table.
2387 * File descriptors are copied over to the new file descriptor table,
2388 * regardless of whether the close-on-exec flag is set.
2391 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2392 struct filedesc **ret)
2394 struct filedesc *newfdp;
2395 struct filedescent *nfde, *ofde;
2396 int error, i, lastfile;
2400 newfdp = fdinit(fdp, true, &lastfile);
2401 if (nfds > lastfile + 1) {
2402 /* New table cannot be larger than the old one. */
2406 /* Copy all passable descriptors (i.e. not kqueue). */
2407 newfdp->fd_freefile = nfds;
2408 for (i = 0; i < nfds; ++i) {
2409 if (fds[i] < 0 || fds[i] > lastfile) {
2410 /* File descriptor out of bounds. */
2414 ofde = &fdp->fd_ofiles[fds[i]];
2415 if (ofde->fde_file == NULL) {
2416 /* Unused file descriptor. */
2420 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2421 /* File descriptor cannot be passed. */
2425 if (!fhold(ofde->fde_file)) {
2429 nfde = &newfdp->fd_ofiles[i];
2431 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2432 fdused_init(newfdp, i);
2434 FILEDESC_SUNLOCK(fdp);
2438 FILEDESC_SUNLOCK(fdp);
2439 fdescfree_remapped(newfdp);
2444 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2445 * one of processes using it exits) and the table used to be shared.
2448 fdclearlocks(struct thread *td)
2450 struct filedesc *fdp;
2451 struct filedesc_to_leader *fdtol;
2461 MPASS(fdtol != NULL);
2463 FILEDESC_XLOCK(fdp);
2464 KASSERT(fdtol->fdl_refcount > 0,
2465 ("filedesc_to_refcount botch: fdl_refcount=%d",
2466 fdtol->fdl_refcount));
2467 if (fdtol->fdl_refcount == 1 &&
2468 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2469 lastfile = fdlastfile(fdp);
2470 for (i = 0; i <= lastfile; i++) {
2471 fp = fdp->fd_ofiles[i].fde_file;
2472 if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2475 FILEDESC_XUNLOCK(fdp);
2476 lf.l_whence = SEEK_SET;
2479 lf.l_type = F_UNLCK;
2481 (void) VOP_ADVLOCK(vp,
2482 (caddr_t)p->p_leader, F_UNLCK,
2484 FILEDESC_XLOCK(fdp);
2489 if (fdtol->fdl_refcount == 1) {
2490 if (fdp->fd_holdleaderscount > 0 &&
2491 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2493 * close() or kern_dup() has cleared a reference
2494 * in a shared file descriptor table.
2496 fdp->fd_holdleaderswakeup = 1;
2497 sx_sleep(&fdp->fd_holdleaderscount,
2498 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2501 if (fdtol->fdl_holdcount > 0) {
2503 * Ensure that fdtol->fdl_leader remains
2504 * valid in closef().
2506 fdtol->fdl_wakeup = 1;
2507 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2512 fdtol->fdl_refcount--;
2513 if (fdtol->fdl_refcount == 0 &&
2514 fdtol->fdl_holdcount == 0) {
2515 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2516 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2520 FILEDESC_XUNLOCK(fdp);
2522 free(fdtol, M_FILEDESC_TO_LEADER);
2526 * Release a filedesc structure.
2529 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2531 struct filedesc0 *fdp0;
2532 struct freetable *ft, *tft;
2533 struct filedescent *fde;
2537 KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2538 ("%s: fd table %p carries references", __func__, fdp));
2541 * Serialize with threads iterating over the table, if any.
2543 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2544 FILEDESC_XLOCK(fdp);
2545 FILEDESC_XUNLOCK(fdp);
2548 lastfile = fdlastfile_single(fdp);
2549 for (i = 0; i <= lastfile; i++) {
2550 fde = &fdp->fd_ofiles[i];
2555 (void) closef(fp, td);
2561 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2562 free(fdp->fd_map, M_FILEDESC);
2563 if (fdp->fd_nfiles > NDFILE)
2564 free(fdp->fd_files, M_FILEDESC);
2566 fdp0 = (struct filedesc0 *)fdp;
2567 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2568 free(ft->ft_table, M_FILEDESC);
2574 fdescfree(struct thread *td)
2577 struct filedesc *fdp;
2584 if (RACCT_ENABLED())
2585 racct_set_unlocked(p, RACCT_NOFILE, 0);
2588 if (p->p_fdtol != NULL)
2592 * Check fdhold for an explanation.
2594 atomic_store_ptr(&p->p_fd, NULL);
2595 atomic_thread_fence_seq_cst();
2596 PROC_WAIT_UNLOCKED(p);
2598 if (refcount_release(&fdp->fd_refcnt) == 0)
2601 fdescfree_fds(td, fdp, 1);
2605 pdescfree(struct thread *td)
2608 struct pwddesc *pdp;
2615 * Check pdhold for an explanation.
2617 atomic_store_ptr(&p->p_pd, NULL);
2618 atomic_thread_fence_seq_cst();
2619 PROC_WAIT_UNLOCKED(p);
2625 fdescfree_remapped(struct filedesc *fdp)
2628 /* fdescfree_fds() asserts that fd_refcnt == 0. */
2629 if (!refcount_release(&fdp->fd_refcnt))
2630 panic("%s: fd table %p has extra references", __func__, fdp);
2632 fdescfree_fds(curthread, fdp, 0);
2636 * For setugid programs, we don't want to people to use that setugidness
2637 * to generate error messages which write to a file which otherwise would
2638 * otherwise be off-limits to the process. We check for filesystems where
2639 * the vnode can change out from under us after execve (like [lin]procfs).
2641 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2642 * sufficient. We also don't check for setugidness since we know we are.
2645 is_unsafe(struct file *fp)
2649 if (fp->f_type != DTYPE_VNODE)
2653 return ((vp->v_vflag & VV_PROCDEP) != 0);
2657 * Make this setguid thing safe, if at all possible.
2660 fdsetugidsafety(struct thread *td)
2662 struct filedesc *fdp;
2666 fdp = td->td_proc->p_fd;
2667 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2668 ("the fdtable should not be shared"));
2669 MPASS(fdp->fd_nfiles >= 3);
2670 for (i = 0; i <= 2; i++) {
2671 fp = fdp->fd_ofiles[i].fde_file;
2672 if (fp != NULL && is_unsafe(fp)) {
2673 FILEDESC_XLOCK(fdp);
2674 knote_fdclose(td, i);
2676 * NULL-out descriptor prior to close to avoid
2677 * a race while close blocks.
2680 FILEDESC_XUNLOCK(fdp);
2681 (void) closef(fp, td);
2687 * If a specific file object occupies a specific file descriptor, close the
2688 * file descriptor entry and drop a reference on the file object. This is a
2689 * convenience function to handle a subsequent error in a function that calls
2690 * falloc() that handles the race that another thread might have closed the
2691 * file descriptor out from under the thread creating the file object.
2694 fdclose(struct thread *td, struct file *fp, int idx)
2696 struct filedesc *fdp = td->td_proc->p_fd;
2698 FILEDESC_XLOCK(fdp);
2699 if (fdp->fd_ofiles[idx].fde_file == fp) {
2701 FILEDESC_XUNLOCK(fdp);
2704 FILEDESC_XUNLOCK(fdp);
2708 * Close any files on exec?
2711 fdcloseexec(struct thread *td)
2713 struct filedesc *fdp;
2714 struct filedescent *fde;
2718 fdp = td->td_proc->p_fd;
2719 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2720 ("the fdtable should not be shared"));
2721 lastfile = fdlastfile_single(fdp);
2722 for (i = 0; i <= lastfile; i++) {
2723 fde = &fdp->fd_ofiles[i];
2725 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2726 (fde->fde_flags & UF_EXCLOSE))) {
2727 FILEDESC_XLOCK(fdp);
2729 (void) closefp(fdp, i, fp, td, false, false);
2730 FILEDESC_UNLOCK_ASSERT(fdp);
2736 * It is unsafe for set[ug]id processes to be started with file
2737 * descriptors 0..2 closed, as these descriptors are given implicit
2738 * significance in the Standard C library. fdcheckstd() will create a
2739 * descriptor referencing /dev/null for each of stdin, stdout, and
2740 * stderr that is not already open.
2743 fdcheckstd(struct thread *td)
2745 struct filedesc *fdp;
2747 int i, error, devnull;
2749 fdp = td->td_proc->p_fd;
2750 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2751 ("the fdtable should not be shared"));
2752 MPASS(fdp->fd_nfiles >= 3);
2754 for (i = 0; i <= 2; i++) {
2755 if (fdp->fd_ofiles[i].fde_file != NULL)
2758 save = td->td_retval[0];
2759 if (devnull != -1) {
2760 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2762 error = kern_openat(td, AT_FDCWD, "/dev/null",
2763 UIO_SYSSPACE, O_RDWR, 0);
2765 devnull = td->td_retval[0];
2766 KASSERT(devnull == i, ("we didn't get our fd"));
2769 td->td_retval[0] = save;
2777 * Internal form of close. Decrement reference count on file structure.
2778 * Note: td may be NULL when closing a file that was being passed in a
2782 closef(struct file *fp, struct thread *td)
2786 struct filedesc_to_leader *fdtol;
2787 struct filedesc *fdp;
2792 * POSIX record locking dictates that any close releases ALL
2793 * locks owned by this process. This is handled by setting
2794 * a flag in the unlock to free ONLY locks obeying POSIX
2795 * semantics, and not to free BSD-style file locks.
2796 * If the descriptor was in a message, POSIX-style locks
2797 * aren't passed with the descriptor, and the thread pointer
2798 * will be NULL. Callers should be careful only to pass a
2799 * NULL thread pointer when there really is no owning
2800 * context that might have locks, or the locks will be
2803 if (fp->f_type == DTYPE_VNODE) {
2805 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2806 lf.l_whence = SEEK_SET;
2809 lf.l_type = F_UNLCK;
2810 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2811 F_UNLCK, &lf, F_POSIX);
2813 fdtol = td->td_proc->p_fdtol;
2814 if (fdtol != NULL) {
2816 * Handle special case where file descriptor table is
2817 * shared between multiple process leaders.
2819 fdp = td->td_proc->p_fd;
2820 FILEDESC_XLOCK(fdp);
2821 for (fdtol = fdtol->fdl_next;
2822 fdtol != td->td_proc->p_fdtol;
2823 fdtol = fdtol->fdl_next) {
2824 if ((fdtol->fdl_leader->p_flag &
2827 fdtol->fdl_holdcount++;
2828 FILEDESC_XUNLOCK(fdp);
2829 lf.l_whence = SEEK_SET;
2832 lf.l_type = F_UNLCK;
2834 (void) VOP_ADVLOCK(vp,
2835 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2837 FILEDESC_XLOCK(fdp);
2838 fdtol->fdl_holdcount--;
2839 if (fdtol->fdl_holdcount == 0 &&
2840 fdtol->fdl_wakeup != 0) {
2841 fdtol->fdl_wakeup = 0;
2845 FILEDESC_XUNLOCK(fdp);
2848 return (fdrop_close(fp, td));
2852 * Hack for file descriptor passing code.
2855 closef_nothread(struct file *fp)
2862 * Initialize the file pointer with the specified properties.
2864 * The ops are set with release semantics to be certain that the flags, type,
2865 * and data are visible when ops is. This is to prevent ops methods from being
2866 * called with bad data.
2869 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2874 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2878 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2880 fp->f_seqcount[UIO_READ] = 1;
2881 fp->f_seqcount[UIO_WRITE] = 1;
2882 finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2887 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2888 struct file **fpp, struct filecaps *havecapsp)
2890 struct filedescent *fde;
2893 FILEDESC_LOCK_ASSERT(fdp);
2895 fde = fdeget_locked(fdp, fd);
2902 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2907 if (havecapsp != NULL)
2908 filecaps_copy(&fde->fde_caps, havecapsp, true);
2910 *fpp = fde->fde_file;
2918 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2919 struct file **fpp, struct filecaps *havecapsp)
2921 struct filedesc *fdp = td->td_proc->p_fd;
2923 #ifndef CAPABILITIES
2924 error = fget_unlocked(fdp, fd, needrightsp, fpp);
2925 if (havecapsp != NULL && error == 0)
2926 filecaps_fill(havecapsp);
2933 error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2937 if (havecapsp != NULL) {
2938 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2939 havecapsp, false)) {
2945 if (!fd_modified(fdp, fd, seq))
2954 FILEDESC_SLOCK(fdp);
2955 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2956 if (error == 0 && !fhold(*fpp))
2958 FILEDESC_SUNLOCK(fdp);
2965 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2967 const struct filedescent *fde;
2968 const struct fdescenttbl *fdt;
2969 struct filedesc *fdp;
2972 const cap_rights_t *haverights;
2973 cap_rights_t rights;
2976 VFS_SMR_ASSERT_ENTERED();
2978 rights = *ndp->ni_rightsneeded;
2979 cap_rights_set_one(&rights, CAP_LOOKUP);
2981 fdp = curproc->p_fd;
2982 fdt = fdp->fd_files;
2983 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2985 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
2986 fde = &fdt->fdt_ofiles[fd];
2987 haverights = cap_rights_fde_inline(fde);
2989 if (__predict_false(fp == NULL))
2991 if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2993 *fsearch = ((fp->f_flag & FSEARCH) != 0);
2995 if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2998 if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
3002 * Use an acquire barrier to force re-reading of fdt so it is
3003 * refreshed for verification.
3005 atomic_thread_fence_acq();
3006 fdt = fdp->fd_files;
3007 if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3010 * If file descriptor doesn't have all rights,
3011 * all lookups relative to it must also be
3012 * strictly relative.
3014 * Not yet supported by fast path.
3017 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
3018 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
3019 ndp->ni_filecaps.fc_nioctls != -1) {
3021 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
3031 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
3033 const struct fdescenttbl *fdt;
3034 struct filedesc *fdp;
3038 VFS_SMR_ASSERT_ENTERED();
3040 fdp = curproc->p_fd;
3041 fdt = fdp->fd_files;
3042 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3044 fp = fdt->fdt_ofiles[fd].fde_file;
3045 if (__predict_false(fp == NULL))
3047 *fsearch = ((fp->f_flag & FSEARCH) != 0);
3049 if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
3053 * Use an acquire barrier to force re-reading of fdt so it is
3054 * refreshed for verification.
3056 atomic_thread_fence_acq();
3057 fdt = fdp->fd_files;
3058 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3060 filecaps_fill(&ndp->ni_filecaps);
3067 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3068 struct file **fpp, seqc_t *seqp)
3071 const struct filedescent *fde;
3073 const struct fdescenttbl *fdt;
3077 cap_rights_t haverights;
3081 fdt = fdp->fd_files;
3082 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3085 * Fetch the descriptor locklessly. We avoid fdrop() races by
3086 * never raising a refcount above 0. To accomplish this we have
3087 * to use a cmpset loop rather than an atomic_add. The descriptor
3088 * must be re-verified once we acquire a reference to be certain
3089 * that the identity is still correct and we did not lose a race
3090 * due to preemption.
3094 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3095 fde = &fdt->fdt_ofiles[fd];
3096 haverights = *cap_rights_fde_inline(fde);
3098 if (!seqc_consistent(fd_seqc(fdt, fd), seq))
3101 fp = fdt->fdt_ofiles[fd].fde_file;
3106 error = cap_check_inline(&haverights, needrightsp);
3110 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3112 * Force a reload. Other thread could reallocate the
3113 * table before this fd was closed, so it is possible
3114 * that there is a stale fp pointer in cached version.
3116 fdt = atomic_load_ptr(&fdp->fd_files);
3120 * Use an acquire barrier to force re-reading of fdt so it is
3121 * refreshed for verification.
3123 atomic_thread_fence_acq();
3124 fdt = fdp->fd_files;
3126 if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
3128 if (fp == fdt->fdt_ofiles[fd].fde_file)
3131 fdrop(fp, curthread);
3143 * See the comments in fget_unlocked_seq for an explanation of how this works.
3145 * This is a simplified variant which bails out to the aforementioned routine
3146 * if anything goes wrong. In practice this only happens when userspace is
3147 * racing with itself.
3150 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3154 const struct filedescent *fde;
3156 const struct fdescenttbl *fdt;
3160 const cap_rights_t *haverights;
3163 fdt = fdp->fd_files;
3164 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3167 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3168 fde = &fdt->fdt_ofiles[fd];
3169 haverights = cap_rights_fde_inline(fde);
3172 fp = fdt->fdt_ofiles[fd].fde_file;
3174 if (__predict_false(fp == NULL))
3177 if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3180 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3184 * Use an acquire barrier to force re-reading of fdt so it is
3185 * refreshed for verification.
3187 atomic_thread_fence_acq();
3188 fdt = fdp->fd_files;
3190 if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3192 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3198 fdrop(fp, curthread);
3200 return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
3204 * Translate fd -> file when the caller guarantees the file descriptor table
3205 * can't be changed by others.
3207 * Note this does not mean the file object itself is only visible to the caller,
3208 * merely that it wont disappear without having to be referenced.
3210 * Must be paired with fput_only_user.
3214 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3217 const struct filedescent *fde;
3218 const struct fdescenttbl *fdt;
3219 const cap_rights_t *haverights;
3223 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3225 if (__predict_false(fd >= fdp->fd_nfiles))
3228 fdt = fdp->fd_files;
3229 fde = &fdt->fdt_ofiles[fd];
3231 if (__predict_false(fp == NULL))
3233 MPASS(refcount_load(&fp->f_count) > 0);
3234 haverights = cap_rights_fde_inline(fde);
3235 error = cap_check_inline(haverights, needrightsp);
3236 if (__predict_false(error != 0))
3243 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3248 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3250 if (__predict_false(fd >= fdp->fd_nfiles))
3253 fp = fdp->fd_ofiles[fd].fde_file;
3254 if (__predict_false(fp == NULL))
3257 MPASS(refcount_load(&fp->f_count) > 0);
3264 * Extract the file pointer associated with the specified descriptor for the
3265 * current user process.
3267 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3270 * File's rights will be checked against the capability rights mask.
3272 * If an error occurred the non-zero error is returned and *fpp is set to
3273 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
3274 * responsible for fdrop().
3277 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3278 cap_rights_t *needrightsp)
3280 struct filedesc *fdp;
3285 fdp = td->td_proc->p_fd;
3286 error = fget_unlocked(fdp, fd, needrightsp, &fp);
3287 if (__predict_false(error != 0))
3289 if (__predict_false(fp->f_ops == &badfileops)) {
3295 * FREAD and FWRITE failure return EBADF as per POSIX.
3301 if ((fp->f_flag & flags) == 0)
3305 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3306 ((fp->f_flag & FWRITE) != 0))
3312 KASSERT(0, ("wrong flags"));
3325 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3328 return (_fget(td, fd, fpp, 0, rightsp));
3332 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3336 #ifndef CAPABILITIES
3337 error = _fget(td, fd, fpp, 0, rightsp);
3338 if (maxprotp != NULL)
3339 *maxprotp = VM_PROT_ALL;
3342 cap_rights_t fdrights;
3343 struct filedesc *fdp;
3348 fdp = td->td_proc->p_fd;
3349 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3351 error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3352 if (__predict_false(error != 0))
3354 if (__predict_false(fp->f_ops == &badfileops)) {
3358 if (maxprotp != NULL)
3359 fdrights = *cap_rights(fdp, fd);
3360 if (!fd_modified(fdp, fd, seq))
3366 * If requested, convert capability rights to access flags.
3368 if (maxprotp != NULL)
3369 *maxprotp = cap_rights_to_vmprot(&fdrights);
3376 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3379 return (_fget(td, fd, fpp, FREAD, rightsp));
3383 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3386 return (_fget(td, fd, fpp, FWRITE, rightsp));
3390 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3393 struct filedesc *fdp = td->td_proc->p_fd;
3394 #ifndef CAPABILITIES
3395 return (fget_unlocked(fdp, fd, rightsp, fpp));
3402 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3404 error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3407 error = cap_fcntl_check(fdp, fd, needfcntl);
3408 if (!fd_modified(fdp, fd, seq))
3422 * Like fget() but loads the underlying vnode, or returns an error if the
3423 * descriptor does not represent a vnode. Note that pipes use vnodes but
3424 * never have VM objects. The returned vnode will be vref()'d.
3426 * XXX: what about the unused flags ?
3429 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3436 error = _fget(td, fd, &fp, flags, needrightsp);
3439 if (fp->f_vnode == NULL) {
3451 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3454 return (_fgetvp(td, fd, 0, rightsp, vpp));
3458 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3459 struct filecaps *havecaps, struct vnode **vpp)
3461 struct filecaps caps;
3465 error = fget_cap(td, fd, needrightsp, &fp, &caps);
3468 if (fp->f_ops == &badfileops) {
3472 if (fp->f_vnode == NULL) {
3484 filecaps_free(&caps);
3490 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3493 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3497 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3500 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3505 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3509 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3514 * Handle the last reference to a file being closed.
3516 * Without the noinline attribute clang keeps inlining the func thorough this
3517 * file when fdrop is used.
3520 _fdrop(struct file *fp, struct thread *td)
3526 count = refcount_load(&fp->f_count);
3528 panic("fdrop: fp %p count %d", fp, count);
3530 error = fo_close(fp, td);
3531 atomic_subtract_int(&openfiles, 1);
3533 free(fp->f_advice, M_FADVISE);
3534 uma_zfree(file_zone, fp);
3540 * Apply an advisory lock on a file descriptor.
3542 * Just attempt to get a record lock of the requested type on the entire file
3543 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3545 #ifndef _SYS_SYSPROTO_H_
3553 sys_flock(struct thread *td, struct flock_args *uap)
3560 error = fget(td, uap->fd, &cap_flock_rights, &fp);
3563 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
3565 return (EOPNOTSUPP);
3569 lf.l_whence = SEEK_SET;
3572 if (uap->how & LOCK_UN) {
3573 lf.l_type = F_UNLCK;
3574 atomic_clear_int(&fp->f_flag, FHASLOCK);
3575 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3578 if (uap->how & LOCK_EX)
3579 lf.l_type = F_WRLCK;
3580 else if (uap->how & LOCK_SH)
3581 lf.l_type = F_RDLCK;
3586 atomic_set_int(&fp->f_flag, FHASLOCK);
3587 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3588 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3594 * Duplicate the specified descriptor to a free descriptor.
3597 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3598 int openerror, int *indxp)
3600 struct filedescent *newfde, *oldfde;
3605 KASSERT(openerror == ENODEV || openerror == ENXIO,
3606 ("unexpected error %d in %s", openerror, __func__));
3609 * If the to-be-dup'd fd number is greater than the allowed number
3610 * of file descriptors, or the fd to be dup'd has already been
3611 * closed, then reject.
3613 FILEDESC_XLOCK(fdp);
3614 if ((fp = fget_locked(fdp, dfd)) == NULL) {
3615 FILEDESC_XUNLOCK(fdp);
3619 error = fdalloc(td, 0, &indx);
3621 FILEDESC_XUNLOCK(fdp);
3626 * There are two cases of interest here.
3628 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3630 * For ENXIO steal away the file structure from (dfd) and store it in
3631 * (indx). (dfd) is effectively closed by this operation.
3633 switch (openerror) {
3636 * Check that the mode the file is being opened for is a
3637 * subset of the mode of the existing descriptor.
3639 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3640 fdunused(fdp, indx);
3641 FILEDESC_XUNLOCK(fdp);
3645 fdunused(fdp, indx);
3646 FILEDESC_XUNLOCK(fdp);
3649 newfde = &fdp->fd_ofiles[indx];
3650 oldfde = &fdp->fd_ofiles[dfd];
3651 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3653 seqc_write_begin(&newfde->fde_seqc);
3655 memcpy(newfde, oldfde, fde_change_size);
3656 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3659 seqc_write_end(&newfde->fde_seqc);
3664 * Steal away the file pointer from dfd and stuff it into indx.
3666 newfde = &fdp->fd_ofiles[indx];
3667 oldfde = &fdp->fd_ofiles[dfd];
3669 seqc_write_begin(&newfde->fde_seqc);
3671 memcpy(newfde, oldfde, fde_change_size);
3672 oldfde->fde_file = NULL;
3675 seqc_write_end(&newfde->fde_seqc);
3679 FILEDESC_XUNLOCK(fdp);
3685 * This sysctl determines if we will allow a process to chroot(2) if it
3686 * has a directory open:
3687 * 0: disallowed for all processes.
3688 * 1: allowed for processes that were not already chroot(2)'ed.
3689 * 2: allowed for all processes.
3692 static int chroot_allow_open_directories = 1;
3694 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3695 &chroot_allow_open_directories, 0,
3696 "Allow a process to chroot(2) if it has a directory open");
3699 * Helper function for raised chroot(2) security function: Refuse if
3700 * any filedescriptors are open directories.
3703 chroot_refuse_vdir_fds(struct filedesc *fdp)
3709 FILEDESC_LOCK_ASSERT(fdp);
3711 lastfile = fdlastfile(fdp);
3712 for (fd = 0; fd <= lastfile; fd++) {
3713 fp = fget_locked(fdp, fd);
3716 if (fp->f_type == DTYPE_VNODE) {
3718 if (vp->v_type == VDIR)
3726 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3729 if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3730 vrefact(oldpwd->pwd_cdir);
3731 newpwd->pwd_cdir = oldpwd->pwd_cdir;
3734 if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3735 vrefact(oldpwd->pwd_rdir);
3736 newpwd->pwd_rdir = oldpwd->pwd_rdir;
3739 if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3740 vrefact(oldpwd->pwd_jdir);
3741 newpwd->pwd_jdir = oldpwd->pwd_jdir;
3746 pwd_hold_pwddesc(struct pwddesc *pdp)
3750 PWDDESC_ASSERT_XLOCKED(pdp);
3751 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3753 refcount_acquire(&pwd->pwd_refcount);
3758 pwd_hold_smr(struct pwd *pwd)
3762 if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3769 pwd_hold(struct thread *td)
3771 struct pwddesc *pdp;
3774 pdp = td->td_proc->p_pd;
3777 pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3778 if (pwd_hold_smr(pwd)) {
3784 pwd = pwd_hold_pwddesc(pdp);
3786 PWDDESC_XUNLOCK(pdp);
3795 pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3796 bzero(pwd, sizeof(*pwd));
3797 refcount_init(&pwd->pwd_refcount, 1);
3802 pwd_drop(struct pwd *pwd)
3805 if (!refcount_release(&pwd->pwd_refcount))
3808 if (pwd->pwd_cdir != NULL)
3809 vrele(pwd->pwd_cdir);
3810 if (pwd->pwd_rdir != NULL)
3811 vrele(pwd->pwd_rdir);
3812 if (pwd->pwd_jdir != NULL)
3813 vrele(pwd->pwd_jdir);
3814 uma_zfree_smr(pwd_zone, pwd);
3818 * The caller is responsible for invoking priv_check() and
3819 * mac_vnode_check_chroot() to authorize this operation.
3822 pwd_chroot(struct thread *td, struct vnode *vp)
3824 struct pwddesc *pdp;
3825 struct filedesc *fdp;
3826 struct pwd *newpwd, *oldpwd;
3829 fdp = td->td_proc->p_fd;
3830 pdp = td->td_proc->p_pd;
3831 newpwd = pwd_alloc();
3832 FILEDESC_SLOCK(fdp);
3834 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3835 if (chroot_allow_open_directories == 0 ||
3836 (chroot_allow_open_directories == 1 &&
3837 oldpwd->pwd_rdir != rootvnode)) {
3838 error = chroot_refuse_vdir_fds(fdp);
3839 FILEDESC_SUNLOCK(fdp);
3841 PWDDESC_XUNLOCK(pdp);
3846 FILEDESC_SUNLOCK(fdp);
3850 newpwd->pwd_rdir = vp;
3851 if (oldpwd->pwd_jdir == NULL) {
3853 newpwd->pwd_jdir = vp;
3855 pwd_fill(oldpwd, newpwd);
3856 pwd_set(pdp, newpwd);
3857 PWDDESC_XUNLOCK(pdp);
3863 pwd_chdir(struct thread *td, struct vnode *vp)
3865 struct pwddesc *pdp;
3866 struct pwd *newpwd, *oldpwd;
3868 VNPASS(vp->v_usecount > 0, vp);
3870 newpwd = pwd_alloc();
3871 pdp = td->td_proc->p_pd;
3873 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3874 newpwd->pwd_cdir = vp;
3875 pwd_fill(oldpwd, newpwd);
3876 pwd_set(pdp, newpwd);
3877 PWDDESC_XUNLOCK(pdp);
3882 * jail_attach(2) changes both root and working directories.
3885 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
3887 struct pwddesc *pdp;
3888 struct filedesc *fdp;
3889 struct pwd *newpwd, *oldpwd;
3892 fdp = td->td_proc->p_fd;
3893 pdp = td->td_proc->p_pd;
3894 newpwd = pwd_alloc();
3895 FILEDESC_SLOCK(fdp);
3897 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3898 error = chroot_refuse_vdir_fds(fdp);
3899 FILEDESC_SUNLOCK(fdp);
3901 PWDDESC_XUNLOCK(pdp);
3907 newpwd->pwd_rdir = vp;
3909 newpwd->pwd_cdir = vp;
3910 if (oldpwd->pwd_jdir == NULL) {
3912 newpwd->pwd_jdir = vp;
3914 pwd_fill(oldpwd, newpwd);
3915 pwd_set(pdp, newpwd);
3916 PWDDESC_XUNLOCK(pdp);
3922 pwd_ensure_dirs(void)
3924 struct pwddesc *pdp;
3925 struct pwd *oldpwd, *newpwd;
3927 pdp = curproc->p_pd;
3929 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3930 if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3931 PWDDESC_XUNLOCK(pdp);
3934 PWDDESC_XUNLOCK(pdp);
3936 newpwd = pwd_alloc();
3938 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3939 pwd_fill(oldpwd, newpwd);
3940 if (newpwd->pwd_cdir == NULL) {
3942 newpwd->pwd_cdir = rootvnode;
3944 if (newpwd->pwd_rdir == NULL) {
3946 newpwd->pwd_rdir = rootvnode;
3948 pwd_set(pdp, newpwd);
3949 PWDDESC_XUNLOCK(pdp);
3954 pwd_set_rootvnode(void)
3956 struct pwddesc *pdp;
3957 struct pwd *oldpwd, *newpwd;
3959 pdp = curproc->p_pd;
3961 newpwd = pwd_alloc();
3963 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3965 newpwd->pwd_cdir = rootvnode;
3967 newpwd->pwd_rdir = rootvnode;
3968 pwd_fill(oldpwd, newpwd);
3969 pwd_set(pdp, newpwd);
3970 PWDDESC_XUNLOCK(pdp);
3975 * Scan all active processes and prisons to see if any of them have a current
3976 * or root directory of `olddp'. If so, replace them with the new mount point.
3979 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3981 struct pwddesc *pdp;
3982 struct pwd *newpwd, *oldpwd;
3987 if (vrefcnt(olddp) == 1)
3990 newpwd = pwd_alloc();
3991 sx_slock(&allproc_lock);
3992 FOREACH_PROC_IN_SYSTEM(p) {
3999 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4000 if (oldpwd == NULL ||
4001 (oldpwd->pwd_cdir != olddp &&
4002 oldpwd->pwd_rdir != olddp &&
4003 oldpwd->pwd_jdir != olddp)) {
4004 PWDDESC_XUNLOCK(pdp);
4008 if (oldpwd->pwd_cdir == olddp) {
4010 newpwd->pwd_cdir = newdp;
4012 if (oldpwd->pwd_rdir == olddp) {
4014 newpwd->pwd_rdir = newdp;
4016 if (oldpwd->pwd_jdir == olddp) {
4018 newpwd->pwd_jdir = newdp;
4020 pwd_fill(oldpwd, newpwd);
4021 pwd_set(pdp, newpwd);
4022 PWDDESC_XUNLOCK(pdp);
4025 newpwd = pwd_alloc();
4027 sx_sunlock(&allproc_lock);
4029 if (rootvnode == olddp) {
4034 mtx_lock(&prison0.pr_mtx);
4035 if (prison0.pr_root == olddp) {
4037 prison0.pr_root = newdp;
4040 mtx_unlock(&prison0.pr_mtx);
4041 sx_slock(&allprison_lock);
4042 TAILQ_FOREACH(pr, &allprison, pr_list) {
4043 mtx_lock(&pr->pr_mtx);
4044 if (pr->pr_root == olddp) {
4046 pr->pr_root = newdp;
4049 mtx_unlock(&pr->pr_mtx);
4051 sx_sunlock(&allprison_lock);
4056 struct filedesc_to_leader *
4057 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
4059 struct filedesc_to_leader *fdtol;
4061 fdtol = malloc(sizeof(struct filedesc_to_leader),
4062 M_FILEDESC_TO_LEADER, M_WAITOK);
4063 fdtol->fdl_refcount = 1;
4064 fdtol->fdl_holdcount = 0;
4065 fdtol->fdl_wakeup = 0;
4066 fdtol->fdl_leader = leader;
4068 FILEDESC_XLOCK(fdp);
4069 fdtol->fdl_next = old->fdl_next;
4070 fdtol->fdl_prev = old;
4071 old->fdl_next = fdtol;
4072 fdtol->fdl_next->fdl_prev = fdtol;
4073 FILEDESC_XUNLOCK(fdp);
4075 fdtol->fdl_next = fdtol;
4076 fdtol->fdl_prev = fdtol;
4082 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
4085 struct filedesc *fdp;
4087 int count, off, minoff;
4093 if (*(int *)arg1 != 0)
4096 fdp = curproc->p_fd;
4098 FILEDESC_SLOCK(fdp);
4100 off = NDSLOT(fdp->fd_nfiles - 1);
4101 for (minoff = NDSLOT(0); off >= minoff; --off)
4102 count += bitcountl(map[off]);
4103 FILEDESC_SUNLOCK(fdp);
4105 return (SYSCTL_OUT(req, &count, sizeof(count)));
4108 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
4109 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
4110 "Number of open file descriptors");
4113 * Get file structures globally.
4116 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
4119 struct filedesc *fdp;
4122 int error, n, lastfile;
4124 error = sysctl_wire_old_buffer(req, 0);
4127 if (req->oldptr == NULL) {
4129 sx_slock(&allproc_lock);
4130 FOREACH_PROC_IN_SYSTEM(p) {
4132 if (p->p_state == PRS_NEW) {
4140 /* overestimates sparse tables. */
4141 n += fdp->fd_nfiles;
4144 sx_sunlock(&allproc_lock);
4145 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4148 bzero(&xf, sizeof(xf));
4149 xf.xf_size = sizeof(xf);
4150 sx_slock(&allproc_lock);
4151 FOREACH_PROC_IN_SYSTEM(p) {
4153 if (p->p_state == PRS_NEW) {
4157 if (p_cansee(req->td, p) != 0) {
4161 xf.xf_pid = p->p_pid;
4162 xf.xf_uid = p->p_ucred->cr_uid;
4167 FILEDESC_SLOCK(fdp);
4168 lastfile = fdlastfile(fdp);
4169 for (n = 0; refcount_load(&fdp->fd_refcnt) > 0 && n <= lastfile;
4171 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4174 xf.xf_file = (uintptr_t)fp;
4175 xf.xf_data = (uintptr_t)fp->f_data;
4176 xf.xf_vnode = (uintptr_t)fp->f_vnode;
4177 xf.xf_type = (uintptr_t)fp->f_type;
4178 xf.xf_count = refcount_load(&fp->f_count);
4180 xf.xf_offset = foffset_get(fp);
4181 xf.xf_flag = fp->f_flag;
4182 error = SYSCTL_OUT(req, &xf, sizeof(xf));
4186 FILEDESC_SUNLOCK(fdp);
4191 sx_sunlock(&allproc_lock);
4195 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4196 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4198 #ifdef KINFO_FILE_SIZE
4199 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4203 xlate_fflags(int fflags)
4205 static const struct {
4208 } fflags_table[] = {
4209 { FAPPEND, KF_FLAG_APPEND },
4210 { FASYNC, KF_FLAG_ASYNC },
4211 { FFSYNC, KF_FLAG_FSYNC },
4212 { FHASLOCK, KF_FLAG_HASLOCK },
4213 { FNONBLOCK, KF_FLAG_NONBLOCK },
4214 { FREAD, KF_FLAG_READ },
4215 { FWRITE, KF_FLAG_WRITE },
4216 { O_CREAT, KF_FLAG_CREAT },
4217 { O_DIRECT, KF_FLAG_DIRECT },
4218 { O_EXCL, KF_FLAG_EXCL },
4219 { O_EXEC, KF_FLAG_EXEC },
4220 { O_EXLOCK, KF_FLAG_EXLOCK },
4221 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4222 { O_SHLOCK, KF_FLAG_SHLOCK },
4223 { O_TRUNC, KF_FLAG_TRUNC }
4229 for (i = 0; i < nitems(fflags_table); i++)
4230 if (fflags & fflags_table[i].fflag)
4231 kflags |= fflags_table[i].kf_fflag;
4235 /* Trim unused data from kf_path by truncating the structure size. */
4237 pack_kinfo(struct kinfo_file *kif)
4240 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4241 strlen(kif->kf_path) + 1;
4242 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4246 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4247 struct kinfo_file *kif, struct filedesc *fdp, int flags)
4251 bzero(kif, sizeof(*kif));
4253 /* Set a default type to allow for empty fill_kinfo() methods. */
4254 kif->kf_type = KF_TYPE_UNKNOWN;
4255 kif->kf_flags = xlate_fflags(fp->f_flag);
4256 if (rightsp != NULL)
4257 kif->kf_cap_rights = *rightsp;
4259 cap_rights_init_zero(&kif->kf_cap_rights);
4261 kif->kf_ref_count = refcount_load(&fp->f_count);
4262 kif->kf_offset = foffset_get(fp);
4265 * This may drop the filedesc lock, so the 'fp' cannot be
4266 * accessed after this call.
4268 error = fo_fill_kinfo(fp, kif, fdp);
4270 kif->kf_status |= KF_ATTR_VALID;
4271 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4274 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4278 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4279 struct kinfo_file *kif, int flags)
4283 bzero(kif, sizeof(*kif));
4285 kif->kf_type = KF_TYPE_VNODE;
4286 error = vn_fill_kinfo_vnode(vp, kif);
4288 kif->kf_status |= KF_ATTR_VALID;
4289 kif->kf_flags = xlate_fflags(fflags);
4290 cap_rights_init_zero(&kif->kf_cap_rights);
4292 kif->kf_ref_count = -1;
4293 kif->kf_offset = -1;
4294 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4297 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4301 struct export_fd_buf {
4302 struct filedesc *fdp;
4303 struct pwddesc *pdp;
4306 struct kinfo_file kif;
4311 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4313 struct kinfo_file *kif;
4316 if (efbuf->remainder != -1) {
4317 if (efbuf->remainder < kif->kf_structsize) {
4318 /* Terminate export. */
4319 efbuf->remainder = 0;
4322 efbuf->remainder -= kif->kf_structsize;
4324 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
4328 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4329 struct export_fd_buf *efbuf)
4333 if (efbuf->remainder == 0)
4335 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4337 FILEDESC_SUNLOCK(efbuf->fdp);
4338 error = export_kinfo_to_sb(efbuf);
4339 FILEDESC_SLOCK(efbuf->fdp);
4344 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4345 struct export_fd_buf *efbuf)
4349 if (efbuf->remainder == 0)
4351 if (efbuf->pdp != NULL)
4352 PWDDESC_XUNLOCK(efbuf->pdp);
4353 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4354 error = export_kinfo_to_sb(efbuf);
4355 if (efbuf->pdp != NULL)
4356 PWDDESC_XLOCK(efbuf->pdp);
4361 * Store a process file descriptor information to sbuf.
4363 * Takes a locked proc as argument, and returns with the proc unlocked.
4366 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
4370 struct filedesc *fdp;
4371 struct pwddesc *pdp;
4372 struct export_fd_buf *efbuf;
4373 struct vnode *cttyvp, *textvp, *tracevp;
4375 int error, i, lastfile;
4376 cap_rights_t rights;
4378 PROC_LOCK_ASSERT(p, MA_OWNED);
4381 tracevp = ktr_get_tracevp(p, true);
4383 textvp = p->p_textvp;
4386 /* Controlling tty. */
4388 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4389 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4396 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4400 efbuf->remainder = maxlen;
4401 efbuf->flags = flags;
4402 if (tracevp != NULL)
4403 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
4406 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
4408 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
4411 if (pdp == NULL || fdp == NULL)
4416 pwd = pwd_hold_pwddesc(pdp);
4418 /* working directory */
4419 if (pwd->pwd_cdir != NULL) {
4420 vrefact(pwd->pwd_cdir);
4421 export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD,
4424 /* root directory */
4425 if (pwd->pwd_rdir != NULL) {
4426 vrefact(pwd->pwd_rdir);
4427 export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT,
4430 /* jail directory */
4431 if (pwd->pwd_jdir != NULL) {
4432 vrefact(pwd->pwd_jdir);
4433 export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL,
4437 PWDDESC_XUNLOCK(pdp);
4440 FILEDESC_SLOCK(fdp);
4441 lastfile = fdlastfile(fdp);
4442 for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4443 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4446 rights = *cap_rights(fdp, i);
4447 #else /* !CAPABILITIES */
4448 rights = cap_no_rights;
4451 * Create sysctl entry. It is OK to drop the filedesc
4452 * lock inside of export_file_to_sb() as we will
4453 * re-validate and re-evaluate its properties when the
4456 error = export_file_to_sb(fp, i, &rights, efbuf);
4457 if (error != 0 || efbuf->remainder == 0)
4460 FILEDESC_SUNLOCK(fdp);
4466 free(efbuf, M_TEMP);
4470 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
4473 * Get per-process file descriptors for use by procstat(1), et al.
4476 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4482 int error, error2, *name;
4490 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4491 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4492 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4497 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4498 error = kern_proc_filedesc_out(p, &sb, maxlen,
4499 KERN_FILEDESC_PACK_KINFO);
4500 error2 = sbuf_finish(&sb);
4502 return (error != 0 ? error : error2);
4505 #ifdef COMPAT_FREEBSD7
4506 #ifdef KINFO_OFILE_SIZE
4507 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4511 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4514 okif->kf_structsize = sizeof(*okif);
4515 okif->kf_type = kif->kf_type;
4516 okif->kf_fd = kif->kf_fd;
4517 okif->kf_ref_count = kif->kf_ref_count;
4518 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4519 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4520 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4521 okif->kf_offset = kif->kf_offset;
4522 if (kif->kf_type == KF_TYPE_VNODE)
4523 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4525 okif->kf_vnode_type = KF_VTYPE_VNON;
4526 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4527 if (kif->kf_type == KF_TYPE_SOCKET) {
4528 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4529 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4530 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4531 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4532 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4534 okif->kf_sa_local.ss_family = AF_UNSPEC;
4535 okif->kf_sa_peer.ss_family = AF_UNSPEC;
4540 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4541 struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4546 PWDDESC_XUNLOCK(pdp);
4547 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4548 kinfo_to_okinfo(kif, okif);
4549 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4555 * Get per-process file descriptors for use by procstat(1), et al.
4558 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4560 struct kinfo_ofile *okif;
4561 struct kinfo_file *kif;
4562 struct filedesc *fdp;
4563 struct pwddesc *pdp;
4566 int error, i, lastfile, *name;
4575 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4582 if (fdp == NULL || pdp == NULL) {
4587 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4588 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4590 pwd = pwd_hold_pwddesc(pdp);
4592 if (pwd->pwd_cdir != NULL)
4593 export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4595 if (pwd->pwd_rdir != NULL)
4596 export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4598 if (pwd->pwd_jdir != NULL)
4599 export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4602 PWDDESC_XUNLOCK(pdp);
4605 FILEDESC_SLOCK(fdp);
4606 lastfile = fdlastfile(fdp);
4607 for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4608 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4610 export_file_to_kinfo(fp, i, NULL, kif, fdp,
4611 KERN_FILEDESC_PACK_KINFO);
4612 FILEDESC_SUNLOCK(fdp);
4613 kinfo_to_okinfo(kif, okif);
4614 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4615 FILEDESC_SLOCK(fdp);
4619 FILEDESC_SUNLOCK(fdp);
4627 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4628 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4629 "Process ofiledesc entries");
4630 #endif /* COMPAT_FREEBSD7 */
4633 vntype_to_kinfo(int vtype)
4638 } vtypes_table[] = {
4639 { VBAD, KF_VTYPE_VBAD },
4640 { VBLK, KF_VTYPE_VBLK },
4641 { VCHR, KF_VTYPE_VCHR },
4642 { VDIR, KF_VTYPE_VDIR },
4643 { VFIFO, KF_VTYPE_VFIFO },
4644 { VLNK, KF_VTYPE_VLNK },
4645 { VNON, KF_VTYPE_VNON },
4646 { VREG, KF_VTYPE_VREG },
4647 { VSOCK, KF_VTYPE_VSOCK }
4652 * Perform vtype translation.
4654 for (i = 0; i < nitems(vtypes_table); i++)
4655 if (vtypes_table[i].vtype == vtype)
4656 return (vtypes_table[i].kf_vtype);
4658 return (KF_VTYPE_UNKNOWN);
4661 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4662 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4663 "Process filedesc entries");
4666 * Store a process current working directory information to sbuf.
4668 * Takes a locked proc as argument, and returns with the proc unlocked.
4671 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
4673 struct pwddesc *pdp;
4675 struct export_fd_buf *efbuf;
4679 PROC_LOCK_ASSERT(p, MA_OWNED);
4686 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4689 efbuf->remainder = maxlen;
4692 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4693 cdir = pwd->pwd_cdir;
4698 error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4700 PWDDESC_XUNLOCK(pdp);
4702 free(efbuf, M_TEMP);
4707 * Get per-process current working directory.
4710 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4716 int error, error2, *name;
4724 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4725 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4726 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4731 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4732 error = kern_proc_cwd_out(p, &sb, maxlen);
4733 error2 = sbuf_finish(&sb);
4735 return (error != 0 ? error : error2);
4738 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4739 sysctl_kern_proc_cwd, "Process current working directory");
4743 * For the purposes of debugging, generate a human-readable string for the
4747 file_type_to_name(short type)
4775 case DTYPE_PROCDESC:
4779 case DTYPE_LINUXTFD:
4787 * For the purposes of debugging, identify a process (if any, perhaps one of
4788 * many) that references the passed file in its file descriptor array. Return
4791 static struct proc *
4792 file_to_first_proc(struct file *fp)
4794 struct filedesc *fdp;
4798 FOREACH_PROC_IN_SYSTEM(p) {
4799 if (p->p_state == PRS_NEW)
4804 for (n = 0; n < fdp->fd_nfiles; n++) {
4805 if (fp == fdp->fd_ofiles[n].fde_file)
4813 db_print_file(struct file *fp, int header)
4815 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4819 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4820 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4821 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4823 p = file_to_first_proc(fp);
4824 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4825 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4826 fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4827 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4832 DB_SHOW_COMMAND(file, db_show_file)
4837 db_printf("usage: show file <addr>\n");
4840 fp = (struct file *)addr;
4841 db_print_file(fp, 1);
4844 DB_SHOW_COMMAND(files, db_show_files)
4846 struct filedesc *fdp;
4853 FOREACH_PROC_IN_SYSTEM(p) {
4854 if (p->p_state == PRS_NEW)
4856 if ((fdp = p->p_fd) == NULL)
4858 for (n = 0; n < fdp->fd_nfiles; ++n) {
4859 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4861 db_print_file(fp, header);
4868 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4869 &maxfilesperproc, 0, "Maximum files allowed open per process");
4871 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4872 &maxfiles, 0, "Maximum number of files");
4874 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4875 &openfiles, 0, "System-wide number of open files");
4879 filelistinit(void *dummy)
4882 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4883 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4884 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4885 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4886 pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4887 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4889 * XXXMJG this is a temporary hack due to boot ordering issues against
4892 vfs_smr = uma_zone_get_smr(pwd_zone);
4893 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4895 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4897 /*-------------------------------------------------------------------*/
4900 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4901 int flags, struct thread *td)
4908 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4916 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4924 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4932 badfo_kqfilter(struct file *fp, struct knote *kn)
4939 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4947 badfo_close(struct file *fp, struct thread *td)
4954 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4962 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4970 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4971 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4979 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4985 struct fileops badfileops = {
4986 .fo_read = badfo_readwrite,
4987 .fo_write = badfo_readwrite,
4988 .fo_truncate = badfo_truncate,
4989 .fo_ioctl = badfo_ioctl,
4990 .fo_poll = badfo_poll,
4991 .fo_kqfilter = badfo_kqfilter,
4992 .fo_stat = badfo_stat,
4993 .fo_close = badfo_close,
4994 .fo_chmod = badfo_chmod,
4995 .fo_chown = badfo_chown,
4996 .fo_sendfile = badfo_sendfile,
4997 .fo_fill_kinfo = badfo_fill_kinfo,
5001 path_poll(struct file *fp, int events, struct ucred *active_cred,
5008 path_close(struct file *fp, struct thread *td)
5010 MPASS(fp->f_type == DTYPE_VNODE);
5011 fp->f_ops = &badfileops;
5016 struct fileops path_fileops = {
5017 .fo_read = badfo_readwrite,
5018 .fo_write = badfo_readwrite,
5019 .fo_truncate = badfo_truncate,
5020 .fo_ioctl = badfo_ioctl,
5021 .fo_poll = path_poll,
5022 .fo_kqfilter = vn_kqfilter_opath,
5023 .fo_stat = vn_statfile,
5024 .fo_close = path_close,
5025 .fo_chmod = badfo_chmod,
5026 .fo_chown = badfo_chown,
5027 .fo_sendfile = badfo_sendfile,
5028 .fo_fill_kinfo = vn_fill_kinfo,
5029 .fo_flags = DFLAG_PASSABLE,
5033 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
5034 int flags, struct thread *td)
5037 return (EOPNOTSUPP);
5041 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
5049 invfo_ioctl(struct file *fp, u_long com, void *data,
5050 struct ucred *active_cred, struct thread *td)
5057 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
5061 return (poll_no_poll(events));
5065 invfo_kqfilter(struct file *fp, struct knote *kn)
5072 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5080 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5088 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5089 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5096 /*-------------------------------------------------------------------*/
5099 * File Descriptor pseudo-device driver (/dev/fd/).
5101 * Opening minor device N dup()s the file (if any) connected to file
5102 * descriptor N belonging to the calling process. Note that this driver
5103 * consists of only the ``open()'' routine, because all subsequent
5104 * references to this file will be direct to the other driver.
5106 * XXX: we could give this one a cloning event handler if necessary.
5111 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
5115 * XXX Kludge: set curthread->td_dupfd to contain the value of the
5116 * the file descriptor being sought for duplication. The error
5117 * return ensures that the vnode for this device will be released
5118 * by vn_open. Open will detect this special error and take the
5119 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
5120 * will simply report the error.
5122 td->td_dupfd = dev2unit(dev);
5126 static struct cdevsw fildesc_cdevsw = {
5127 .d_version = D_VERSION,
5133 fildesc_drvinit(void *unused)
5137 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
5138 UID_ROOT, GID_WHEEL, 0666, "fd/0");
5139 make_dev_alias(dev, "stdin");
5140 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
5141 UID_ROOT, GID_WHEEL, 0666, "fd/1");
5142 make_dev_alias(dev, "stdout");
5143 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
5144 UID_ROOT, GID_WHEEL, 0666, "fd/2");
5145 make_dev_alias(dev, "stderr");
5148 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);