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>
83 #include <sys/ktrace.h>
88 #include <security/audit/audit.h>
95 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
96 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
97 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
98 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
99 "file desc to leader structures");
100 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
101 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
103 MALLOC_DECLARE(M_FADVISE);
105 static __read_mostly uma_zone_t file_zone;
106 static __read_mostly uma_zone_t filedesc0_zone;
107 __read_mostly uma_zone_t pwd_zone;
110 static int closefp(struct filedesc *fdp, int fd, struct file *fp,
111 struct thread *td, bool holdleaders, bool audit);
112 static int fd_first_free(struct filedesc *fdp, int low, int size);
113 static void fdgrowtable(struct filedesc *fdp, int nfd);
114 static void fdgrowtable_exp(struct filedesc *fdp, int nfd);
115 static void fdunused(struct filedesc *fdp, int fd);
116 static void fdused(struct filedesc *fdp, int fd);
117 static int getmaxfd(struct thread *td);
118 static u_long *filecaps_copy_prep(const struct filecaps *src);
119 static void filecaps_copy_finish(const struct filecaps *src,
120 struct filecaps *dst, u_long *ioctls);
121 static u_long *filecaps_free_prep(struct filecaps *fcaps);
122 static void filecaps_free_finish(u_long *ioctls);
124 static struct pwd *pwd_alloc(void);
129 * - An array of open file descriptors (fd_ofiles)
130 * - An array of file flags (fd_ofileflags)
131 * - A bitmap recording which descriptors are in use (fd_map)
133 * A process starts out with NDFILE descriptors. The value of NDFILE has
134 * been selected based the historical limit of 20 open files, and an
135 * assumption that the majority of processes, especially short-lived
136 * processes like shells, will never need more.
138 * If this initial allocation is exhausted, a larger descriptor table and
139 * map are allocated dynamically, and the pointers in the process's struct
140 * filedesc are updated to point to those. This is repeated every time
141 * the process runs out of file descriptors (provided it hasn't hit its
144 * Since threads may hold references to individual descriptor table
145 * entries, the tables are never freed. Instead, they are placed on a
146 * linked list and freed only when the struct filedesc is released.
149 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
150 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
151 #define NDSLOT(x) ((x) / NDENTRIES)
152 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
153 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
156 * SLIST entry used to keep track of ofiles which must be reclaimed when
160 struct fdescenttbl *ft_table;
161 SLIST_ENTRY(freetable) ft_next;
165 * Initial allocation: a filedesc structure + the head of SLIST used to
166 * keep track of old ofiles + enough space for NDFILE descriptors.
169 struct fdescenttbl0 {
171 struct filedescent fdt_ofiles[NDFILE];
175 struct filedesc fd_fd;
176 SLIST_HEAD(, freetable) fd_free;
177 struct fdescenttbl0 fd_dfiles;
178 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
182 * Descriptor management.
184 static int __exclusive_cache_line openfiles; /* actual number of open files */
185 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
186 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
189 * If low >= size, just return low. Otherwise find the first zero bit in the
190 * given bitmap, starting at low and not exceeding size - 1. Return size if
194 fd_first_free(struct filedesc *fdp, int low, int size)
196 NDSLOTTYPE *map = fdp->fd_map;
204 if (low % NDENTRIES) {
205 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
206 if ((mask &= ~map[off]) != 0UL)
207 return (off * NDENTRIES + ffsl(mask) - 1);
210 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
211 if (map[off] != ~0UL)
212 return (off * NDENTRIES + ffsl(~map[off]) - 1);
217 * Find the last used fd.
219 * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
220 * Otherwise use fdlastfile.
223 fdlastfile_single(struct filedesc *fdp)
225 NDSLOTTYPE *map = fdp->fd_map;
228 off = NDSLOT(fdp->fd_nfiles - 1);
229 for (minoff = NDSLOT(0); off >= minoff; --off)
231 return (off * NDENTRIES + flsl(map[off]) - 1);
236 fdlastfile(struct filedesc *fdp)
239 FILEDESC_LOCK_ASSERT(fdp);
240 return (fdlastfile_single(fdp));
244 fdisused(struct filedesc *fdp, int fd)
247 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
248 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
250 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
254 * Mark a file descriptor as used.
257 fdused_init(struct filedesc *fdp, int fd)
260 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
262 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
266 fdused(struct filedesc *fdp, int fd)
269 FILEDESC_XLOCK_ASSERT(fdp);
271 fdused_init(fdp, fd);
272 if (fd == fdp->fd_freefile)
277 * Mark a file descriptor as unused.
280 fdunused(struct filedesc *fdp, int fd)
283 FILEDESC_XLOCK_ASSERT(fdp);
285 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
286 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
287 ("fd=%d is still in use", fd));
289 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
290 if (fd < fdp->fd_freefile)
291 fdp->fd_freefile = fd;
295 * Free a file descriptor.
297 * Avoid some work if fdp is about to be destroyed.
300 fdefree_last(struct filedescent *fde)
303 filecaps_free(&fde->fde_caps);
307 fdfree(struct filedesc *fdp, int fd)
309 struct filedescent *fde;
311 FILEDESC_XLOCK_ASSERT(fdp);
312 fde = &fdp->fd_ofiles[fd];
314 seqc_write_begin(&fde->fde_seqc);
316 fde->fde_file = NULL;
318 seqc_write_end(&fde->fde_seqc);
325 * System calls on descriptors.
327 #ifndef _SYS_SYSPROTO_H_
328 struct getdtablesize_args {
334 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
340 td->td_retval[0] = getmaxfd(td);
342 PROC_LOCK(td->td_proc);
343 lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
344 PROC_UNLOCK(td->td_proc);
345 if (lim < td->td_retval[0])
346 td->td_retval[0] = lim;
352 * Duplicate a file descriptor to a particular value.
354 * Note: keep in mind that a potential race condition exists when closing
355 * descriptors from a shared descriptor table (via rfork).
357 #ifndef _SYS_SYSPROTO_H_
365 sys_dup2(struct thread *td, struct dup2_args *uap)
368 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
372 * Duplicate a file descriptor.
374 #ifndef _SYS_SYSPROTO_H_
381 sys_dup(struct thread *td, struct dup_args *uap)
384 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
388 * The file control system call.
390 #ifndef _SYS_SYSPROTO_H_
399 sys_fcntl(struct thread *td, struct fcntl_args *uap)
402 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
406 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
420 * Convert old flock structure to new.
422 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
423 fl.l_start = ofl.l_start;
424 fl.l_len = ofl.l_len;
425 fl.l_pid = ofl.l_pid;
426 fl.l_type = ofl.l_type;
427 fl.l_whence = ofl.l_whence;
441 arg1 = (intptr_t)&fl;
447 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
448 arg1 = (intptr_t)&fl;
456 error = kern_fcntl(td, fd, newcmd, arg1);
459 if (cmd == F_OGETLK) {
460 ofl.l_start = fl.l_start;
461 ofl.l_len = fl.l_len;
462 ofl.l_pid = fl.l_pid;
463 ofl.l_type = fl.l_type;
464 ofl.l_whence = fl.l_whence;
465 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
466 } else if (cmd == F_GETLK) {
467 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
473 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
475 struct filedesc *fdp;
477 struct file *fp, *fp2;
478 struct filedescent *fde;
482 int error, flg, seals, tmp;
496 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
499 case F_DUPFD_CLOEXEC:
501 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
506 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
509 case F_DUP2FD_CLOEXEC:
511 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
517 fde = fdeget_locked(fdp, fd);
520 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
523 FILEDESC_SUNLOCK(fdp);
529 fde = fdeget_locked(fdp, fd);
531 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
532 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
535 FILEDESC_XUNLOCK(fdp);
539 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
542 td->td_retval[0] = OFLAGS(fp->f_flag);
547 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
550 if (fp->f_ops == &path_fileops) {
556 tmp = flg = fp->f_flag;
558 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
559 } while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
560 tmp = fp->f_flag & FNONBLOCK;
561 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
566 tmp = fp->f_flag & FASYNC;
567 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
572 atomic_clear_int(&fp->f_flag, FNONBLOCK);
574 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
579 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
582 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
584 td->td_retval[0] = tmp;
589 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
593 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
598 error = priv_check(td, PRIV_NFS_LOCKD);
606 /* FALLTHROUGH F_SETLK */
610 flp = (struct flock *)arg;
611 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
616 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
619 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
625 if (flp->l_whence == SEEK_CUR) {
626 foffset = foffset_get(fp);
629 foffset > OFF_MAX - flp->l_start)) {
634 flp->l_start += foffset;
638 switch (flp->l_type) {
640 if ((fp->f_flag & FREAD) == 0) {
644 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
645 PROC_LOCK(p->p_leader);
646 p->p_leader->p_flag |= P_ADVLOCK;
647 PROC_UNLOCK(p->p_leader);
649 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
653 if ((fp->f_flag & FWRITE) == 0) {
657 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
658 PROC_LOCK(p->p_leader);
659 p->p_leader->p_flag |= P_ADVLOCK;
660 PROC_UNLOCK(p->p_leader);
662 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
666 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
670 if (flg != F_REMOTE) {
674 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
675 F_UNLCKSYS, flp, flg);
681 if (error != 0 || flp->l_type == F_UNLCK ||
682 flp->l_type == F_UNLCKSYS) {
688 * Check for a race with close.
690 * The vnode is now advisory locked (or unlocked, but this case
691 * is not really important) as the caller requested.
692 * We had to drop the filedesc lock, so we need to recheck if
693 * the descriptor is still valid, because if it was closed
694 * in the meantime we need to remove advisory lock from the
695 * vnode - close on any descriptor leading to an advisory
696 * locked vnode, removes that lock.
697 * We will return 0 on purpose in that case, as the result of
698 * successful advisory lock might have been externally visible
699 * already. This is fine - effectively we pretend to the caller
700 * that the closing thread was a bit slower and that the
701 * advisory lock succeeded before the close.
703 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
709 flp->l_whence = SEEK_SET;
712 flp->l_type = F_UNLCK;
713 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
714 F_UNLCK, flp, F_POSIX);
721 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
724 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
729 flp = (struct flock *)arg;
730 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
731 flp->l_type != F_UNLCK) {
736 if (flp->l_whence == SEEK_CUR) {
737 foffset = foffset_get(fp);
738 if ((flp->l_start > 0 &&
739 foffset > OFF_MAX - flp->l_start) ||
741 foffset < OFF_MIN - flp->l_start)) {
746 flp->l_start += foffset;
749 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
755 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
758 error = fo_add_seals(fp, arg);
763 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
766 if (fo_get_seals(fp, &seals) == 0)
767 td->td_retval[0] = seals;
774 arg = arg ? 128 * 1024: 0;
777 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
780 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
786 if (vp->v_type != VREG) {
793 * Exclusive lock synchronizes against f_seqcount reads and
794 * writes in sequential_heuristic().
796 error = vn_lock(vp, LK_EXCLUSIVE);
802 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
803 arg = MIN(arg, INT_MAX - bsize + 1);
804 fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
805 (arg + bsize - 1) / bsize);
806 atomic_set_int(&fp->f_flag, FRDAHEAD);
808 atomic_clear_int(&fp->f_flag, FRDAHEAD);
816 * Check if the vnode is part of a union stack (either the
817 * "union" flag from mount(2) or unionfs).
819 * Prior to introduction of this op libc's readdir would call
820 * fstatfs(2), in effect unnecessarily copying kilobytes of
821 * data just to check fs name and a mount flag.
823 * Fixing the code to handle everything in the kernel instead
824 * is a non-trivial endeavor and has low priority, thus this
825 * horrible kludge facilitates the current behavior in a much
826 * cheaper manner until someone(tm) sorts this out.
828 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
831 if (fp->f_type != DTYPE_VNODE) {
838 * Since we don't prevent dooming the vnode even non-null mp
839 * found can become immediately stale. This is tolerable since
840 * mount points are type-stable (providing safe memory access)
841 * and any vfs op on this vnode going forward will return an
842 * error (meaning return value in this case is meaningless).
844 mp = atomic_load_ptr(&vp->v_mount);
845 if (__predict_false(mp == NULL)) {
850 td->td_retval[0] = 0;
851 if (mp->mnt_kern_flag & MNTK_UNIONFS ||
852 mp->mnt_flag & MNT_UNION)
853 td->td_retval[0] = 1;
865 getmaxfd(struct thread *td)
868 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
872 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
875 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
877 struct filedesc *fdp;
878 struct filedescent *oldfde, *newfde;
880 struct file *delfp, *oldfp;
881 u_long *oioctls, *nioctls;
888 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
889 MPASS(mode < FDDUP_LASTMODE);
892 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
895 * Verify we have a valid descriptor to dup from and possibly to
896 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
897 * return EINVAL when the new descriptor is out of bounds.
902 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
903 maxfd = getmaxfd(td);
905 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
909 if (fget_locked(fdp, old) == NULL)
911 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
912 td->td_retval[0] = new;
913 if (flags & FDDUP_FLAG_CLOEXEC)
914 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
919 oldfde = &fdp->fd_ofiles[old];
920 oldfp = oldfde->fde_file;
925 * If the caller specified a file descriptor, make sure the file
926 * table is large enough to hold it, and grab it. Otherwise, just
927 * allocate a new descriptor the usual way.
932 if ((error = fdalloc(td, new, &new)) != 0) {
937 case FDDUP_MUSTREPLACE:
938 /* Target file descriptor must exist. */
939 if (fget_locked(fdp, new) == NULL) {
945 if (new >= fdp->fd_nfiles) {
947 * The resource limits are here instead of e.g.
948 * fdalloc(), because the file descriptor table may be
949 * shared between processes, so we can't really use
950 * racct_add()/racct_sub(). Instead of counting the
951 * number of actually allocated descriptors, just put
952 * the limit on the size of the file descriptor table.
955 if (RACCT_ENABLED()) {
956 error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
964 fdgrowtable_exp(fdp, new + 1);
966 if (!fdisused(fdp, new))
970 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
973 KASSERT(old != new, ("new fd is same as old"));
975 /* Refetch oldfde because the table may have grown and old one freed. */
976 oldfde = &fdp->fd_ofiles[old];
977 KASSERT(oldfp == oldfde->fde_file,
978 ("fdt_ofiles shift from growth observed at fd %d",
981 newfde = &fdp->fd_ofiles[new];
982 delfp = newfde->fde_file;
984 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
987 * Duplicate the source descriptor.
990 seqc_write_begin(&newfde->fde_seqc);
992 oioctls = filecaps_free_prep(&newfde->fde_caps);
993 memcpy(newfde, oldfde, fde_change_size);
994 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
996 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
997 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
999 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
1001 seqc_write_end(&newfde->fde_seqc);
1003 td->td_retval[0] = new;
1007 if (delfp != NULL) {
1008 (void) closefp(fdp, new, delfp, td, true, false);
1009 FILEDESC_UNLOCK_ASSERT(fdp);
1012 FILEDESC_XUNLOCK(fdp);
1015 filecaps_free_finish(oioctls);
1020 sigiofree(struct sigio *sigio)
1022 crfree(sigio->sio_ucred);
1023 free(sigio, M_SIGIO);
1026 static struct sigio *
1027 funsetown_locked(struct sigio *sigio)
1032 SIGIO_ASSERT_LOCKED();
1036 *(sigio->sio_myref) = NULL;
1037 if (sigio->sio_pgid < 0) {
1038 pg = sigio->sio_pgrp;
1040 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
1041 sigio, sio_pgsigio);
1044 p = sigio->sio_proc;
1046 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
1047 sigio, sio_pgsigio);
1054 * If sigio is on the list associated with a process or process group,
1055 * disable signalling from the device, remove sigio from the list and
1059 funsetown(struct sigio **sigiop)
1061 struct sigio *sigio;
1063 /* Racy check, consumers must provide synchronization. */
1064 if (*sigiop == NULL)
1068 sigio = funsetown_locked(*sigiop);
1075 * Free a list of sigio structures. The caller must ensure that new sigio
1076 * structures cannot be added after this point. For process groups this is
1077 * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1081 funsetownlst(struct sigiolst *sigiolst)
1085 struct sigio *sigio, *tmp;
1088 sigio = SLIST_FIRST(sigiolst);
1096 sigio = SLIST_FIRST(sigiolst);
1097 if (sigio == NULL) {
1103 * Every entry of the list should belong to a single proc or pgrp.
1105 if (sigio->sio_pgid < 0) {
1106 pg = sigio->sio_pgrp;
1107 sx_assert(&proctree_lock, SX_XLOCKED);
1109 } else /* if (sigio->sio_pgid > 0) */ {
1110 p = sigio->sio_proc;
1112 KASSERT((p->p_flag & P_WEXIT) != 0,
1113 ("%s: process %p is not exiting", __func__, p));
1116 SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1117 *sigio->sio_myref = NULL;
1119 KASSERT(sigio->sio_pgid < 0,
1120 ("Proc sigio in pgrp sigio list"));
1121 KASSERT(sigio->sio_pgrp == pg,
1122 ("Bogus pgrp in sigio list"));
1123 } else /* if (p != NULL) */ {
1124 KASSERT(sigio->sio_pgid > 0,
1125 ("Pgrp sigio in proc sigio list"));
1126 KASSERT(sigio->sio_proc == p,
1127 ("Bogus proc in sigio list"));
1137 SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1142 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1144 * After permission checking, add a sigio structure to the sigio list for
1145 * the process or process group.
1148 fsetown(pid_t pgid, struct sigio **sigiop)
1152 struct sigio *osigio, *sigio;
1162 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1163 sigio->sio_pgid = pgid;
1164 sigio->sio_ucred = crhold(curthread->td_ucred);
1165 sigio->sio_myref = sigiop;
1167 sx_slock(&proctree_lock);
1169 osigio = funsetown_locked(*sigiop);
1178 * Policy - Don't allow a process to FSETOWN a process
1179 * in another session.
1181 * Remove this test to allow maximum flexibility or
1182 * restrict FSETOWN to the current process or process
1183 * group for maximum safety.
1185 if (proc->p_session != curthread->td_proc->p_session) {
1191 sigio->sio_proc = proc;
1192 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1194 } else /* if (pgid < 0) */ {
1195 pgrp = pgfind(-pgid);
1202 * Policy - Don't allow a process to FSETOWN a process
1203 * in another session.
1205 * Remove this test to allow maximum flexibility or
1206 * restrict FSETOWN to the current process or process
1207 * group for maximum safety.
1209 if (pgrp->pg_session != curthread->td_proc->p_session) {
1215 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1216 sigio->sio_pgrp = pgrp;
1219 sx_sunlock(&proctree_lock);
1228 sx_sunlock(&proctree_lock);
1236 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1239 fgetown(struct sigio **sigiop)
1244 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1250 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1255 FILEDESC_XLOCK_ASSERT(fdp);
1258 * We now hold the fp reference that used to be owned by the
1259 * descriptor array. We have to unlock the FILEDESC *AFTER*
1260 * knote_fdclose to prevent a race of the fd getting opened, a knote
1261 * added, and deleteing a knote for the new fd.
1263 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1264 knote_fdclose(td, fd);
1267 * We need to notify mqueue if the object is of type mqueue.
1269 if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1270 mq_fdclose(td, fd, fp);
1271 FILEDESC_XUNLOCK(fdp);
1274 if (AUDITING_TD(td) && audit)
1275 audit_sysclose(td, fd, fp);
1277 error = closef(fp, td);
1280 * All paths leading up to closefp() will have already removed or
1281 * replaced the fd in the filedesc table, so a restart would not
1282 * operate on the same file.
1284 if (error == ERESTART)
1291 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1292 bool holdleaders, bool audit)
1296 FILEDESC_XLOCK_ASSERT(fdp);
1299 if (td->td_proc->p_fdtol != NULL) {
1301 * Ask fdfree() to sleep to ensure that all relevant
1302 * process leaders can be traversed in closef().
1304 fdp->fd_holdleaderscount++;
1306 holdleaders = false;
1310 error = closefp_impl(fdp, fd, fp, td, audit);
1312 FILEDESC_XLOCK(fdp);
1313 fdp->fd_holdleaderscount--;
1314 if (fdp->fd_holdleaderscount == 0 &&
1315 fdp->fd_holdleaderswakeup != 0) {
1316 fdp->fd_holdleaderswakeup = 0;
1317 wakeup(&fdp->fd_holdleaderscount);
1319 FILEDESC_XUNLOCK(fdp);
1325 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1326 bool holdleaders, bool audit)
1329 FILEDESC_XLOCK_ASSERT(fdp);
1331 if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1332 return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1334 return (closefp_impl(fdp, fd, fp, td, audit));
1339 * Close a file descriptor.
1341 #ifndef _SYS_SYSPROTO_H_
1348 sys_close(struct thread *td, struct close_args *uap)
1351 return (kern_close(td, uap->fd));
1355 kern_close(struct thread *td, int fd)
1357 struct filedesc *fdp;
1360 fdp = td->td_proc->p_fd;
1362 FILEDESC_XLOCK(fdp);
1363 if ((fp = fget_locked(fdp, fd)) == NULL) {
1364 FILEDESC_XUNLOCK(fdp);
1369 /* closefp() drops the FILEDESC lock for us. */
1370 return (closefp(fdp, fd, fp, td, true, true));
1374 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1376 struct filedesc *fdp;
1377 const struct fdescenttbl *fdt;
1382 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1383 * open should not be a usage error. From a close_range() perspective,
1384 * close_range(3, ~0U, 0) in the same scenario should also likely not
1385 * be a usage error as all fd above 3 are in-fact already closed.
1387 if (highfd < lowfd) {
1391 fdp = td->td_proc->p_fd;
1392 FILEDESC_XLOCK(fdp);
1393 fdt = atomic_load_ptr(&fdp->fd_files);
1394 highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1396 if (__predict_false(fd > highfd)) {
1400 fp = fdt->fdt_ofiles[fd].fde_file;
1406 (void) closefp(fdp, fd, fp, td, true, true);
1409 FILEDESC_XLOCK(fdp);
1410 fdt = atomic_load_ptr(&fdp->fd_files);
1415 FILEDESC_XUNLOCK(fdp);
1420 #ifndef _SYS_SYSPROTO_H_
1421 struct close_range_args {
1428 sys_close_range(struct thread *td, struct close_range_args *uap)
1431 AUDIT_ARG_FD(uap->lowfd);
1432 AUDIT_ARG_CMD(uap->highfd);
1433 AUDIT_ARG_FFLAGS(uap->flags);
1435 /* No flags currently defined */
1436 if (uap->flags != 0)
1438 return (kern_close_range(td, uap->lowfd, uap->highfd));
1441 #ifdef COMPAT_FREEBSD12
1443 * Close open file descriptors.
1445 #ifndef _SYS_SYSPROTO_H_
1446 struct freebsd12_closefrom_args {
1452 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1456 AUDIT_ARG_FD(uap->lowfd);
1459 * Treat negative starting file descriptor values identical to
1460 * closefrom(0) which closes all files.
1462 lowfd = MAX(0, uap->lowfd);
1463 return (kern_close_range(td, lowfd, ~0U));
1465 #endif /* COMPAT_FREEBSD12 */
1467 #if defined(COMPAT_43)
1469 * Return status information about a file descriptor.
1471 #ifndef _SYS_SYSPROTO_H_
1472 struct ofstat_args {
1479 ofstat(struct thread *td, struct ofstat_args *uap)
1485 error = kern_fstat(td, uap->fd, &ub);
1488 error = copyout(&oub, uap->sb, sizeof(oub));
1492 #endif /* COMPAT_43 */
1494 #if defined(COMPAT_FREEBSD11)
1496 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1499 struct freebsd11_stat osb;
1502 error = kern_fstat(td, uap->fd, &sb);
1505 error = freebsd11_cvtstat(&sb, &osb);
1507 error = copyout(&osb, uap->sb, sizeof(osb));
1510 #endif /* COMPAT_FREEBSD11 */
1513 * Return status information about a file descriptor.
1515 #ifndef _SYS_SYSPROTO_H_
1523 sys_fstat(struct thread *td, struct fstat_args *uap)
1528 error = kern_fstat(td, uap->fd, &ub);
1530 error = copyout(&ub, uap->sb, sizeof(ub));
1535 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1542 error = fget(td, fd, &cap_fstat_rights, &fp);
1543 if (__predict_false(error != 0))
1546 AUDIT_ARG_FILE(td->td_proc, fp);
1548 error = fo_stat(fp, sbp, td->td_ucred, td);
1550 #ifdef __STAT_TIME_T_EXT
1551 sbp->st_atim_ext = 0;
1552 sbp->st_mtim_ext = 0;
1553 sbp->st_ctim_ext = 0;
1554 sbp->st_btim_ext = 0;
1557 if (KTRPOINT(td, KTR_STRUCT))
1558 ktrstat_error(sbp, error);
1563 #if defined(COMPAT_FREEBSD11)
1565 * Return status information about a file descriptor.
1567 #ifndef _SYS_SYSPROTO_H_
1568 struct freebsd11_nfstat_args {
1575 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1581 error = kern_fstat(td, uap->fd, &ub);
1583 freebsd11_cvtnstat(&ub, &nub);
1584 error = copyout(&nub, uap->sb, sizeof(nub));
1588 #endif /* COMPAT_FREEBSD11 */
1591 * Return pathconf information about a file descriptor.
1593 #ifndef _SYS_SYSPROTO_H_
1594 struct fpathconf_args {
1601 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1606 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1608 td->td_retval[0] = value;
1613 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1619 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1623 if (name == _PC_ASYNC_IO) {
1624 *valuep = _POSIX_ASYNCHRONOUS_IO;
1629 vn_lock(vp, LK_SHARED | LK_RETRY);
1630 error = VOP_PATHCONF(vp, name, valuep);
1632 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1633 if (name != _PC_PIPE_BUF) {
1648 * Copy filecaps structure allocating memory for ioctls array if needed.
1650 * The last parameter indicates whether the fdtable is locked. If it is not and
1651 * ioctls are encountered, copying fails and the caller must lock the table.
1653 * Note that if the table was not locked, the caller has to check the relevant
1654 * sequence counter to determine whether the operation was successful.
1657 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1661 if (src->fc_ioctls != NULL && !locked)
1663 memcpy(dst, src, sizeof(*src));
1664 if (src->fc_ioctls == NULL)
1667 KASSERT(src->fc_nioctls > 0,
1668 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1670 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1671 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1672 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1677 filecaps_copy_prep(const struct filecaps *src)
1682 if (__predict_true(src->fc_ioctls == NULL))
1685 KASSERT(src->fc_nioctls > 0,
1686 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1688 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1689 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1694 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1700 if (__predict_true(src->fc_ioctls == NULL)) {
1701 MPASS(ioctls == NULL);
1705 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1706 dst->fc_ioctls = ioctls;
1707 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1711 * Move filecaps structure to the new place and clear the old place.
1714 filecaps_move(struct filecaps *src, struct filecaps *dst)
1718 bzero(src, sizeof(*src));
1722 * Fill the given filecaps structure with full rights.
1725 filecaps_fill(struct filecaps *fcaps)
1728 CAP_ALL(&fcaps->fc_rights);
1729 fcaps->fc_ioctls = NULL;
1730 fcaps->fc_nioctls = -1;
1731 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1735 * Free memory allocated within filecaps structure.
1738 filecaps_free(struct filecaps *fcaps)
1741 free(fcaps->fc_ioctls, M_FILECAPS);
1742 bzero(fcaps, sizeof(*fcaps));
1746 filecaps_free_prep(struct filecaps *fcaps)
1750 ioctls = fcaps->fc_ioctls;
1751 bzero(fcaps, sizeof(*fcaps));
1756 filecaps_free_finish(u_long *ioctls)
1759 free(ioctls, M_FILECAPS);
1763 * Validate the given filecaps structure.
1766 filecaps_validate(const struct filecaps *fcaps, const char *func)
1769 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1770 ("%s: invalid rights", func));
1771 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1772 ("%s: invalid fcntls", func));
1773 KASSERT(fcaps->fc_fcntls == 0 ||
1774 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1775 ("%s: fcntls without CAP_FCNTL", func));
1776 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1777 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1778 ("%s: invalid ioctls", func));
1779 KASSERT(fcaps->fc_nioctls == 0 ||
1780 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1781 ("%s: ioctls without CAP_IOCTL", func));
1785 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1789 FILEDESC_XLOCK_ASSERT(fdp);
1791 nfd1 = fdp->fd_nfiles * 2;
1794 fdgrowtable(fdp, nfd1);
1798 * Grow the file table to accommodate (at least) nfd descriptors.
1801 fdgrowtable(struct filedesc *fdp, int nfd)
1803 struct filedesc0 *fdp0;
1804 struct freetable *ft;
1805 struct fdescenttbl *ntable;
1806 struct fdescenttbl *otable;
1807 int nnfiles, onfiles;
1808 NDSLOTTYPE *nmap, *omap;
1810 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1812 /* save old values */
1813 onfiles = fdp->fd_nfiles;
1814 otable = fdp->fd_files;
1817 /* compute the size of the new table */
1818 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1819 if (nnfiles <= onfiles)
1820 /* the table is already large enough */
1824 * Allocate a new table. We need enough space for the number of
1825 * entries, file entries themselves and the struct freetable we will use
1826 * when we decommission the table and place it on the freelist.
1827 * We place the struct freetable in the middle so we don't have
1828 * to worry about padding.
1830 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1831 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1832 sizeof(struct freetable),
1833 M_FILEDESC, M_ZERO | M_WAITOK);
1834 /* copy the old data */
1835 ntable->fdt_nfiles = nnfiles;
1836 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1837 onfiles * sizeof(ntable->fdt_ofiles[0]));
1840 * Allocate a new map only if the old is not large enough. It will
1841 * grow at a slower rate than the table as it can map more
1842 * entries than the table can hold.
1844 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1845 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1847 /* copy over the old data and update the pointer */
1848 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1853 * Make sure that ntable is correctly initialized before we replace
1854 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1857 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1860 * Free the old file table when not shared by other threads or processes.
1861 * The old file table is considered to be shared when either are true:
1862 * - The process has more than one thread.
1863 * - The file descriptor table has been shared via fdshare().
1865 * When shared, the old file table will be placed on a freelist
1866 * which will be processed when the struct filedesc is released.
1868 * Note that if onfiles == NDFILE, we're dealing with the original
1869 * static allocation contained within (struct filedesc0 *)fdp,
1870 * which must not be freed.
1872 if (onfiles > NDFILE) {
1874 * Note we may be called here from fdinit while allocating a
1875 * table for a new process in which case ->p_fd points
1878 if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) {
1879 free(otable, M_FILEDESC);
1881 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1882 fdp0 = (struct filedesc0 *)fdp;
1883 ft->ft_table = otable;
1884 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1888 * The map does not have the same possibility of threads still
1889 * holding references to it. So always free it as long as it
1890 * does not reference the original static allocation.
1892 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1893 free(omap, M_FILEDESC);
1897 * Allocate a file descriptor for the process.
1900 fdalloc(struct thread *td, int minfd, int *result)
1902 struct proc *p = td->td_proc;
1903 struct filedesc *fdp = p->p_fd;
1904 int fd, maxfd, allocfd;
1909 FILEDESC_XLOCK_ASSERT(fdp);
1911 if (fdp->fd_freefile > minfd)
1912 minfd = fdp->fd_freefile;
1914 maxfd = getmaxfd(td);
1917 * Search the bitmap for a free descriptor starting at minfd.
1918 * If none is found, grow the file table.
1920 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1921 if (__predict_false(fd >= maxfd))
1923 if (__predict_false(fd >= fdp->fd_nfiles)) {
1924 allocfd = min(fd * 2, maxfd);
1926 if (RACCT_ENABLED()) {
1927 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1933 * fd is already equal to first free descriptor >= minfd, so
1934 * we only need to grow the table and we are done.
1936 fdgrowtable_exp(fdp, allocfd);
1940 * Perform some sanity checks, then mark the file descriptor as
1941 * used and return it to the caller.
1943 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1944 ("invalid descriptor %d", fd));
1945 KASSERT(!fdisused(fdp, fd),
1946 ("fd_first_free() returned non-free descriptor"));
1947 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1948 ("file descriptor isn't free"));
1955 * Allocate n file descriptors for the process.
1958 fdallocn(struct thread *td, int minfd, int *fds, int n)
1960 struct proc *p = td->td_proc;
1961 struct filedesc *fdp = p->p_fd;
1964 FILEDESC_XLOCK_ASSERT(fdp);
1966 for (i = 0; i < n; i++)
1967 if (fdalloc(td, 0, &fds[i]) != 0)
1971 for (i--; i >= 0; i--)
1972 fdunused(fdp, fds[i]);
1980 * Create a new open file structure and allocate a file descriptor for the
1981 * process that refers to it. We add one reference to the file for the
1982 * descriptor table and one reference for resultfp. This is to prevent us
1983 * being preempted and the entry in the descriptor table closed after we
1984 * release the FILEDESC lock.
1987 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1988 struct filecaps *fcaps)
1993 MPASS(resultfp != NULL);
1994 MPASS(resultfd != NULL);
1996 error = _falloc_noinstall(td, &fp, 2);
1997 if (__predict_false(error != 0)) {
2001 error = finstall_refed(td, fp, &fd, flags, fcaps);
2002 if (__predict_false(error != 0)) {
2003 falloc_abort(td, fp);
2014 * Create a new open file structure without allocating a file descriptor.
2017 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2020 int maxuserfiles = maxfiles - (maxfiles / 20);
2022 static struct timeval lastfail;
2025 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2028 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2029 if ((openfiles_new >= maxuserfiles &&
2030 priv_check(td, PRIV_MAXFILES) != 0) ||
2031 openfiles_new >= maxfiles) {
2032 atomic_subtract_int(&openfiles, 1);
2033 if (ppsratecheck(&lastfail, &curfail, 1)) {
2034 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2035 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2039 fp = uma_zalloc(file_zone, M_WAITOK);
2040 bzero(fp, sizeof(*fp));
2041 refcount_init(&fp->f_count, n);
2042 fp->f_cred = crhold(td->td_ucred);
2043 fp->f_ops = &badfileops;
2049 falloc_abort(struct thread *td, struct file *fp)
2053 * For assertion purposes.
2055 refcount_init(&fp->f_count, 0);
2060 * Install a file in a file descriptor table.
2063 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2064 struct filecaps *fcaps)
2066 struct filedescent *fde;
2070 filecaps_validate(fcaps, __func__);
2071 FILEDESC_XLOCK_ASSERT(fdp);
2073 fde = &fdp->fd_ofiles[fd];
2075 seqc_write_begin(&fde->fde_seqc);
2078 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2080 filecaps_move(fcaps, &fde->fde_caps);
2082 filecaps_fill(&fde->fde_caps);
2084 seqc_write_end(&fde->fde_seqc);
2089 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2090 struct filecaps *fcaps)
2092 struct filedesc *fdp = td->td_proc->p_fd;
2097 FILEDESC_XLOCK(fdp);
2098 error = fdalloc(td, 0, fd);
2099 if (__predict_true(error == 0)) {
2100 _finstall(fdp, fp, *fd, flags, fcaps);
2102 FILEDESC_XUNLOCK(fdp);
2107 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2108 struct filecaps *fcaps)
2116 error = finstall_refed(td, fp, fd, flags, fcaps);
2117 if (__predict_false(error != 0)) {
2124 * Build a new filedesc structure from another.
2126 * If fdp is not NULL, return with it shared locked.
2129 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2131 struct filedesc0 *newfdp0;
2132 struct filedesc *newfdp;
2135 MPASS(lastfile != NULL);
2137 MPASS(lastfile == NULL);
2139 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2140 newfdp = &newfdp0->fd_fd;
2142 /* Create the file descriptor table. */
2143 FILEDESC_LOCK_INIT(newfdp);
2144 refcount_init(&newfdp->fd_refcnt, 1);
2145 refcount_init(&newfdp->fd_holdcnt, 1);
2146 newfdp->fd_map = newfdp0->fd_dmap;
2147 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2148 newfdp->fd_files->fdt_nfiles = NDFILE;
2153 FILEDESC_SLOCK(fdp);
2155 FILEDESC_SUNLOCK(fdp);
2160 *lastfile = fdlastfile(fdp);
2161 if (*lastfile < newfdp->fd_nfiles)
2163 FILEDESC_SUNLOCK(fdp);
2164 fdgrowtable(newfdp, *lastfile + 1);
2165 FILEDESC_SLOCK(fdp);
2172 * Build a pwddesc structure from another.
2173 * Copy the current, root, and jail root vnode references.
2175 * If pdp is not NULL, return with it shared locked.
2178 pdinit(struct pwddesc *pdp, bool keeplock)
2180 struct pwddesc *newpdp;
2183 newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2185 PWDDESC_LOCK_INIT(newpdp);
2186 refcount_init(&newpdp->pd_refcount, 1);
2187 newpdp->pd_cmask = CMASK;
2190 newpwd = pwd_alloc();
2191 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2196 newpwd = pwd_hold_pwddesc(pdp);
2197 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2199 PWDDESC_XUNLOCK(pdp);
2203 static struct filedesc *
2204 fdhold(struct proc *p)
2206 struct filedesc *fdp;
2208 PROC_LOCK_ASSERT(p, MA_OWNED);
2211 refcount_acquire(&fdp->fd_holdcnt);
2215 static struct pwddesc *
2216 pdhold(struct proc *p)
2218 struct pwddesc *pdp;
2220 PROC_LOCK_ASSERT(p, MA_OWNED);
2223 refcount_acquire(&pdp->pd_refcount);
2228 fddrop(struct filedesc *fdp)
2231 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2232 if (refcount_release(&fdp->fd_holdcnt) == 0)
2236 FILEDESC_LOCK_DESTROY(fdp);
2237 uma_zfree(filedesc0_zone, fdp);
2241 pddrop(struct pwddesc *pdp)
2245 if (refcount_release_if_not_last(&pdp->pd_refcount))
2249 if (refcount_release(&pdp->pd_refcount) == 0) {
2250 PWDDESC_XUNLOCK(pdp);
2253 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2255 PWDDESC_XUNLOCK(pdp);
2258 PWDDESC_LOCK_DESTROY(pdp);
2259 free(pdp, M_PWDDESC);
2263 * Share a filedesc structure.
2266 fdshare(struct filedesc *fdp)
2269 refcount_acquire(&fdp->fd_refcnt);
2274 * Share a pwddesc structure.
2277 pdshare(struct pwddesc *pdp)
2279 refcount_acquire(&pdp->pd_refcount);
2284 * Unshare a filedesc structure, if necessary by making a copy
2287 fdunshare(struct thread *td)
2289 struct filedesc *tmp;
2290 struct proc *p = td->td_proc;
2292 if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2295 tmp = fdcopy(p->p_fd);
2301 * Unshare a pwddesc structure.
2304 pdunshare(struct thread *td)
2306 struct pwddesc *pdp;
2311 if (p->p_pd->pd_refcount == 1)
2314 pdp = pdcopy(p->p_pd);
2320 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2324 td->td_proc->p_fd = fdp;
2328 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2329 * this is to ease callers, not catch errors.
2332 fdcopy(struct filedesc *fdp)
2334 struct filedesc *newfdp;
2335 struct filedescent *nfde, *ofde;
2340 newfdp = fdinit(fdp, true, &lastfile);
2341 /* copy all passable descriptors (i.e. not kqueue) */
2342 newfdp->fd_freefile = -1;
2343 for (i = 0; i <= lastfile; ++i) {
2344 ofde = &fdp->fd_ofiles[i];
2345 if (ofde->fde_file == NULL ||
2346 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2347 !fhold(ofde->fde_file)) {
2348 if (newfdp->fd_freefile == -1)
2349 newfdp->fd_freefile = i;
2352 nfde = &newfdp->fd_ofiles[i];
2354 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2355 fdused_init(newfdp, i);
2357 if (newfdp->fd_freefile == -1)
2358 newfdp->fd_freefile = i;
2359 FILEDESC_SUNLOCK(fdp);
2364 * Copy a pwddesc structure.
2367 pdcopy(struct pwddesc *pdp)
2369 struct pwddesc *newpdp;
2373 newpdp = pdinit(pdp, true);
2374 newpdp->pd_cmask = pdp->pd_cmask;
2375 PWDDESC_XUNLOCK(pdp);
2380 * Copies a filedesc structure, while remapping all file descriptors
2381 * stored inside using a translation table.
2383 * File descriptors are copied over to the new file descriptor table,
2384 * regardless of whether the close-on-exec flag is set.
2387 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2388 struct filedesc **ret)
2390 struct filedesc *newfdp;
2391 struct filedescent *nfde, *ofde;
2392 int error, i, lastfile;
2396 newfdp = fdinit(fdp, true, &lastfile);
2397 if (nfds > lastfile + 1) {
2398 /* New table cannot be larger than the old one. */
2402 /* Copy all passable descriptors (i.e. not kqueue). */
2403 newfdp->fd_freefile = nfds;
2404 for (i = 0; i < nfds; ++i) {
2405 if (fds[i] < 0 || fds[i] > lastfile) {
2406 /* File descriptor out of bounds. */
2410 ofde = &fdp->fd_ofiles[fds[i]];
2411 if (ofde->fde_file == NULL) {
2412 /* Unused file descriptor. */
2416 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2417 /* File descriptor cannot be passed. */
2421 if (!fhold(ofde->fde_file)) {
2425 nfde = &newfdp->fd_ofiles[i];
2427 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2428 fdused_init(newfdp, i);
2430 FILEDESC_SUNLOCK(fdp);
2434 FILEDESC_SUNLOCK(fdp);
2435 fdescfree_remapped(newfdp);
2440 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2441 * one of processes using it exits) and the table used to be shared.
2444 fdclearlocks(struct thread *td)
2446 struct filedesc *fdp;
2447 struct filedesc_to_leader *fdtol;
2457 MPASS(fdtol != NULL);
2459 FILEDESC_XLOCK(fdp);
2460 KASSERT(fdtol->fdl_refcount > 0,
2461 ("filedesc_to_refcount botch: fdl_refcount=%d",
2462 fdtol->fdl_refcount));
2463 if (fdtol->fdl_refcount == 1 &&
2464 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2465 lastfile = fdlastfile(fdp);
2466 for (i = 0; i <= lastfile; i++) {
2467 fp = fdp->fd_ofiles[i].fde_file;
2468 if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2471 FILEDESC_XUNLOCK(fdp);
2472 lf.l_whence = SEEK_SET;
2475 lf.l_type = F_UNLCK;
2477 (void) VOP_ADVLOCK(vp,
2478 (caddr_t)p->p_leader, F_UNLCK,
2480 FILEDESC_XLOCK(fdp);
2485 if (fdtol->fdl_refcount == 1) {
2486 if (fdp->fd_holdleaderscount > 0 &&
2487 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2489 * close() or kern_dup() has cleared a reference
2490 * in a shared file descriptor table.
2492 fdp->fd_holdleaderswakeup = 1;
2493 sx_sleep(&fdp->fd_holdleaderscount,
2494 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2497 if (fdtol->fdl_holdcount > 0) {
2499 * Ensure that fdtol->fdl_leader remains
2500 * valid in closef().
2502 fdtol->fdl_wakeup = 1;
2503 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2508 fdtol->fdl_refcount--;
2509 if (fdtol->fdl_refcount == 0 &&
2510 fdtol->fdl_holdcount == 0) {
2511 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2512 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2516 FILEDESC_XUNLOCK(fdp);
2518 free(fdtol, M_FILEDESC_TO_LEADER);
2522 * Release a filedesc structure.
2525 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2527 struct filedesc0 *fdp0;
2528 struct freetable *ft, *tft;
2529 struct filedescent *fde;
2533 KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2534 ("%s: fd table %p carries references", __func__, fdp));
2537 * Serialize with threads iterating over the table, if any.
2539 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2540 FILEDESC_XLOCK(fdp);
2541 FILEDESC_XUNLOCK(fdp);
2544 lastfile = fdlastfile_single(fdp);
2545 for (i = 0; i <= lastfile; i++) {
2546 fde = &fdp->fd_ofiles[i];
2551 (void) closef(fp, td);
2557 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2558 free(fdp->fd_map, M_FILEDESC);
2559 if (fdp->fd_nfiles > NDFILE)
2560 free(fdp->fd_files, M_FILEDESC);
2562 fdp0 = (struct filedesc0 *)fdp;
2563 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2564 free(ft->ft_table, M_FILEDESC);
2570 fdescfree(struct thread *td)
2573 struct filedesc *fdp;
2580 if (RACCT_ENABLED())
2581 racct_set_unlocked(p, RACCT_NOFILE, 0);
2584 if (p->p_fdtol != NULL)
2591 if (refcount_release(&fdp->fd_refcnt) == 0)
2594 fdescfree_fds(td, fdp, 1);
2598 pdescfree(struct thread *td)
2601 struct pwddesc *pdp;
2615 fdescfree_remapped(struct filedesc *fdp)
2618 /* fdescfree_fds() asserts that fd_refcnt == 0. */
2619 if (!refcount_release(&fdp->fd_refcnt))
2620 panic("%s: fd table %p has extra references", __func__, fdp);
2622 fdescfree_fds(curthread, fdp, 0);
2626 * For setugid programs, we don't want to people to use that setugidness
2627 * to generate error messages which write to a file which otherwise would
2628 * otherwise be off-limits to the process. We check for filesystems where
2629 * the vnode can change out from under us after execve (like [lin]procfs).
2631 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2632 * sufficient. We also don't check for setugidness since we know we are.
2635 is_unsafe(struct file *fp)
2639 if (fp->f_type != DTYPE_VNODE)
2643 return ((vp->v_vflag & VV_PROCDEP) != 0);
2647 * Make this setguid thing safe, if at all possible.
2650 fdsetugidsafety(struct thread *td)
2652 struct filedesc *fdp;
2656 fdp = td->td_proc->p_fd;
2657 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2658 ("the fdtable should not be shared"));
2659 MPASS(fdp->fd_nfiles >= 3);
2660 for (i = 0; i <= 2; i++) {
2661 fp = fdp->fd_ofiles[i].fde_file;
2662 if (fp != NULL && is_unsafe(fp)) {
2663 FILEDESC_XLOCK(fdp);
2664 knote_fdclose(td, i);
2666 * NULL-out descriptor prior to close to avoid
2667 * a race while close blocks.
2670 FILEDESC_XUNLOCK(fdp);
2671 (void) closef(fp, td);
2677 * If a specific file object occupies a specific file descriptor, close the
2678 * file descriptor entry and drop a reference on the file object. This is a
2679 * convenience function to handle a subsequent error in a function that calls
2680 * falloc() that handles the race that another thread might have closed the
2681 * file descriptor out from under the thread creating the file object.
2684 fdclose(struct thread *td, struct file *fp, int idx)
2686 struct filedesc *fdp = td->td_proc->p_fd;
2688 FILEDESC_XLOCK(fdp);
2689 if (fdp->fd_ofiles[idx].fde_file == fp) {
2691 FILEDESC_XUNLOCK(fdp);
2694 FILEDESC_XUNLOCK(fdp);
2698 * Close any files on exec?
2701 fdcloseexec(struct thread *td)
2703 struct filedesc *fdp;
2704 struct filedescent *fde;
2708 fdp = td->td_proc->p_fd;
2709 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2710 ("the fdtable should not be shared"));
2711 lastfile = fdlastfile_single(fdp);
2712 for (i = 0; i <= lastfile; i++) {
2713 fde = &fdp->fd_ofiles[i];
2715 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2716 (fde->fde_flags & UF_EXCLOSE))) {
2717 FILEDESC_XLOCK(fdp);
2719 (void) closefp(fdp, i, fp, td, false, false);
2720 FILEDESC_UNLOCK_ASSERT(fdp);
2726 * It is unsafe for set[ug]id processes to be started with file
2727 * descriptors 0..2 closed, as these descriptors are given implicit
2728 * significance in the Standard C library. fdcheckstd() will create a
2729 * descriptor referencing /dev/null for each of stdin, stdout, and
2730 * stderr that is not already open.
2733 fdcheckstd(struct thread *td)
2735 struct filedesc *fdp;
2737 int i, error, devnull;
2739 fdp = td->td_proc->p_fd;
2740 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2741 ("the fdtable should not be shared"));
2742 MPASS(fdp->fd_nfiles >= 3);
2744 for (i = 0; i <= 2; i++) {
2745 if (fdp->fd_ofiles[i].fde_file != NULL)
2748 save = td->td_retval[0];
2749 if (devnull != -1) {
2750 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2752 error = kern_openat(td, AT_FDCWD, "/dev/null",
2753 UIO_SYSSPACE, O_RDWR, 0);
2755 devnull = td->td_retval[0];
2756 KASSERT(devnull == i, ("we didn't get our fd"));
2759 td->td_retval[0] = save;
2767 * Internal form of close. Decrement reference count on file structure.
2768 * Note: td may be NULL when closing a file that was being passed in a
2772 closef(struct file *fp, struct thread *td)
2776 struct filedesc_to_leader *fdtol;
2777 struct filedesc *fdp;
2782 * POSIX record locking dictates that any close releases ALL
2783 * locks owned by this process. This is handled by setting
2784 * a flag in the unlock to free ONLY locks obeying POSIX
2785 * semantics, and not to free BSD-style file locks.
2786 * If the descriptor was in a message, POSIX-style locks
2787 * aren't passed with the descriptor, and the thread pointer
2788 * will be NULL. Callers should be careful only to pass a
2789 * NULL thread pointer when there really is no owning
2790 * context that might have locks, or the locks will be
2793 if (fp->f_type == DTYPE_VNODE) {
2795 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2796 lf.l_whence = SEEK_SET;
2799 lf.l_type = F_UNLCK;
2800 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2801 F_UNLCK, &lf, F_POSIX);
2803 fdtol = td->td_proc->p_fdtol;
2804 if (fdtol != NULL) {
2806 * Handle special case where file descriptor table is
2807 * shared between multiple process leaders.
2809 fdp = td->td_proc->p_fd;
2810 FILEDESC_XLOCK(fdp);
2811 for (fdtol = fdtol->fdl_next;
2812 fdtol != td->td_proc->p_fdtol;
2813 fdtol = fdtol->fdl_next) {
2814 if ((fdtol->fdl_leader->p_flag &
2817 fdtol->fdl_holdcount++;
2818 FILEDESC_XUNLOCK(fdp);
2819 lf.l_whence = SEEK_SET;
2822 lf.l_type = F_UNLCK;
2824 (void) VOP_ADVLOCK(vp,
2825 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2827 FILEDESC_XLOCK(fdp);
2828 fdtol->fdl_holdcount--;
2829 if (fdtol->fdl_holdcount == 0 &&
2830 fdtol->fdl_wakeup != 0) {
2831 fdtol->fdl_wakeup = 0;
2835 FILEDESC_XUNLOCK(fdp);
2838 return (fdrop_close(fp, td));
2842 * Hack for file descriptor passing code.
2845 closef_nothread(struct file *fp)
2852 * Initialize the file pointer with the specified properties.
2854 * The ops are set with release semantics to be certain that the flags, type,
2855 * and data are visible when ops is. This is to prevent ops methods from being
2856 * called with bad data.
2859 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2864 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2868 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2870 fp->f_seqcount[UIO_READ] = 1;
2871 fp->f_seqcount[UIO_WRITE] = 1;
2872 finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2877 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2878 struct file **fpp, struct filecaps *havecapsp)
2880 struct filedescent *fde;
2883 FILEDESC_LOCK_ASSERT(fdp);
2885 fde = fdeget_locked(fdp, fd);
2892 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2897 if (havecapsp != NULL)
2898 filecaps_copy(&fde->fde_caps, havecapsp, true);
2900 *fpp = fde->fde_file;
2908 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2909 struct file **fpp, struct filecaps *havecapsp)
2911 struct filedesc *fdp = td->td_proc->p_fd;
2913 #ifndef CAPABILITIES
2914 error = fget_unlocked(fdp, fd, needrightsp, fpp);
2915 if (havecapsp != NULL && error == 0)
2916 filecaps_fill(havecapsp);
2923 error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2927 if (havecapsp != NULL) {
2928 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2929 havecapsp, false)) {
2935 if (!fd_modified(fdp, fd, seq))
2944 FILEDESC_SLOCK(fdp);
2945 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2946 if (error == 0 && !fhold(*fpp))
2948 FILEDESC_SUNLOCK(fdp);
2955 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2957 const struct filedescent *fde;
2958 const struct fdescenttbl *fdt;
2959 struct filedesc *fdp;
2962 const cap_rights_t *haverights;
2963 cap_rights_t rights;
2966 VFS_SMR_ASSERT_ENTERED();
2968 rights = *ndp->ni_rightsneeded;
2969 cap_rights_set_one(&rights, CAP_LOOKUP);
2971 fdp = curproc->p_fd;
2972 fdt = fdp->fd_files;
2973 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2975 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
2976 fde = &fdt->fdt_ofiles[fd];
2977 haverights = cap_rights_fde_inline(fde);
2979 if (__predict_false(fp == NULL))
2981 if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2983 *fsearch = ((fp->f_flag & FSEARCH) != 0);
2985 if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2988 if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2992 * Use an acquire barrier to force re-reading of fdt so it is
2993 * refreshed for verification.
2995 atomic_thread_fence_acq();
2996 fdt = fdp->fd_files;
2997 if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3000 * If file descriptor doesn't have all rights,
3001 * all lookups relative to it must also be
3002 * strictly relative.
3004 * Not yet supported by fast path.
3007 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
3008 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
3009 ndp->ni_filecaps.fc_nioctls != -1) {
3011 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
3021 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
3023 const struct fdescenttbl *fdt;
3024 struct filedesc *fdp;
3028 VFS_SMR_ASSERT_ENTERED();
3030 fdp = curproc->p_fd;
3031 fdt = fdp->fd_files;
3032 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3034 fp = fdt->fdt_ofiles[fd].fde_file;
3035 if (__predict_false(fp == NULL))
3037 *fsearch = ((fp->f_flag & FSEARCH) != 0);
3039 if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
3043 * Use an acquire barrier to force re-reading of fdt so it is
3044 * refreshed for verification.
3046 atomic_thread_fence_acq();
3047 fdt = fdp->fd_files;
3048 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3050 filecaps_fill(&ndp->ni_filecaps);
3057 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3058 struct file **fpp, seqc_t *seqp)
3061 const struct filedescent *fde;
3063 const struct fdescenttbl *fdt;
3067 cap_rights_t haverights;
3071 fdt = fdp->fd_files;
3072 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3075 * Fetch the descriptor locklessly. We avoid fdrop() races by
3076 * never raising a refcount above 0. To accomplish this we have
3077 * to use a cmpset loop rather than an atomic_add. The descriptor
3078 * must be re-verified once we acquire a reference to be certain
3079 * that the identity is still correct and we did not lose a race
3080 * due to preemption.
3084 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3085 fde = &fdt->fdt_ofiles[fd];
3086 haverights = *cap_rights_fde_inline(fde);
3088 if (!seqc_consistent(fd_seqc(fdt, fd), seq))
3091 fp = fdt->fdt_ofiles[fd].fde_file;
3096 error = cap_check_inline(&haverights, needrightsp);
3100 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3102 * Force a reload. Other thread could reallocate the
3103 * table before this fd was closed, so it is possible
3104 * that there is a stale fp pointer in cached version.
3106 fdt = atomic_load_ptr(&fdp->fd_files);
3110 * Use an acquire barrier to force re-reading of fdt so it is
3111 * refreshed for verification.
3113 atomic_thread_fence_acq();
3114 fdt = fdp->fd_files;
3116 if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
3118 if (fp == fdt->fdt_ofiles[fd].fde_file)
3121 fdrop(fp, curthread);
3133 * See the comments in fget_unlocked_seq for an explanation of how this works.
3135 * This is a simplified variant which bails out to the aforementioned routine
3136 * if anything goes wrong. In practice this only happens when userspace is
3137 * racing with itself.
3140 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3144 const struct filedescent *fde;
3146 const struct fdescenttbl *fdt;
3150 const cap_rights_t *haverights;
3153 fdt = fdp->fd_files;
3154 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3157 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3158 fde = &fdt->fdt_ofiles[fd];
3159 haverights = cap_rights_fde_inline(fde);
3162 fp = fdt->fdt_ofiles[fd].fde_file;
3164 if (__predict_false(fp == NULL))
3167 if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3170 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3174 * Use an acquire barrier to force re-reading of fdt so it is
3175 * refreshed for verification.
3177 atomic_thread_fence_acq();
3178 fdt = fdp->fd_files;
3180 if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3182 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3188 fdrop(fp, curthread);
3190 return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
3194 * Translate fd -> file when the caller guarantees the file descriptor table
3195 * can't be changed by others.
3197 * Note this does not mean the file object itself is only visible to the caller,
3198 * merely that it wont disappear without having to be referenced.
3200 * Must be paired with fput_only_user.
3204 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3207 const struct filedescent *fde;
3208 const struct fdescenttbl *fdt;
3209 const cap_rights_t *haverights;
3213 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3215 if (__predict_false(fd >= fdp->fd_nfiles))
3218 fdt = fdp->fd_files;
3219 fde = &fdt->fdt_ofiles[fd];
3221 if (__predict_false(fp == NULL))
3223 MPASS(refcount_load(&fp->f_count) > 0);
3224 haverights = cap_rights_fde_inline(fde);
3225 error = cap_check_inline(haverights, needrightsp);
3226 if (__predict_false(error != 0))
3233 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3238 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3240 if (__predict_false(fd >= fdp->fd_nfiles))
3243 fp = fdp->fd_ofiles[fd].fde_file;
3244 if (__predict_false(fp == NULL))
3247 MPASS(refcount_load(&fp->f_count) > 0);
3254 * Extract the file pointer associated with the specified descriptor for the
3255 * current user process.
3257 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3260 * File's rights will be checked against the capability rights mask.
3262 * If an error occurred the non-zero error is returned and *fpp is set to
3263 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
3264 * responsible for fdrop().
3267 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3268 cap_rights_t *needrightsp)
3270 struct filedesc *fdp;
3275 fdp = td->td_proc->p_fd;
3276 error = fget_unlocked(fdp, fd, needrightsp, &fp);
3277 if (__predict_false(error != 0))
3279 if (__predict_false(fp->f_ops == &badfileops)) {
3285 * FREAD and FWRITE failure return EBADF as per POSIX.
3291 if ((fp->f_flag & flags) == 0)
3295 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3296 ((fp->f_flag & FWRITE) != 0))
3302 KASSERT(0, ("wrong flags"));
3315 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3318 return (_fget(td, fd, fpp, 0, rightsp));
3322 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3326 #ifndef CAPABILITIES
3327 error = _fget(td, fd, fpp, 0, rightsp);
3328 if (maxprotp != NULL)
3329 *maxprotp = VM_PROT_ALL;
3332 cap_rights_t fdrights;
3333 struct filedesc *fdp;
3338 fdp = td->td_proc->p_fd;
3339 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3341 error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3342 if (__predict_false(error != 0))
3344 if (__predict_false(fp->f_ops == &badfileops)) {
3348 if (maxprotp != NULL)
3349 fdrights = *cap_rights(fdp, fd);
3350 if (!fd_modified(fdp, fd, seq))
3356 * If requested, convert capability rights to access flags.
3358 if (maxprotp != NULL)
3359 *maxprotp = cap_rights_to_vmprot(&fdrights);
3366 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3369 return (_fget(td, fd, fpp, FREAD, rightsp));
3373 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3376 return (_fget(td, fd, fpp, FWRITE, rightsp));
3380 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3383 struct filedesc *fdp = td->td_proc->p_fd;
3384 #ifndef CAPABILITIES
3385 return (fget_unlocked(fdp, fd, rightsp, fpp));
3392 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3394 error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3397 error = cap_fcntl_check(fdp, fd, needfcntl);
3398 if (!fd_modified(fdp, fd, seq))
3412 * Like fget() but loads the underlying vnode, or returns an error if the
3413 * descriptor does not represent a vnode. Note that pipes use vnodes but
3414 * never have VM objects. The returned vnode will be vref()'d.
3416 * XXX: what about the unused flags ?
3419 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3426 error = _fget(td, fd, &fp, flags, needrightsp);
3429 if (fp->f_vnode == NULL) {
3441 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3444 return (_fgetvp(td, fd, 0, rightsp, vpp));
3448 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3449 struct filecaps *havecaps, struct vnode **vpp)
3451 struct filecaps caps;
3455 error = fget_cap(td, fd, needrightsp, &fp, &caps);
3458 if (fp->f_ops == &badfileops) {
3462 if (fp->f_vnode == NULL) {
3474 filecaps_free(&caps);
3480 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3483 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3487 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3490 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3495 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3499 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3504 * Handle the last reference to a file being closed.
3506 * Without the noinline attribute clang keeps inlining the func thorough this
3507 * file when fdrop is used.
3510 _fdrop(struct file *fp, struct thread *td)
3516 count = refcount_load(&fp->f_count);
3518 panic("fdrop: fp %p count %d", fp, count);
3520 error = fo_close(fp, td);
3521 atomic_subtract_int(&openfiles, 1);
3523 free(fp->f_advice, M_FADVISE);
3524 uma_zfree(file_zone, fp);
3530 * Apply an advisory lock on a file descriptor.
3532 * Just attempt to get a record lock of the requested type on the entire file
3533 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3535 #ifndef _SYS_SYSPROTO_H_
3543 sys_flock(struct thread *td, struct flock_args *uap)
3550 error = fget(td, uap->fd, &cap_flock_rights, &fp);
3553 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
3555 return (EOPNOTSUPP);
3559 lf.l_whence = SEEK_SET;
3562 if (uap->how & LOCK_UN) {
3563 lf.l_type = F_UNLCK;
3564 atomic_clear_int(&fp->f_flag, FHASLOCK);
3565 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3568 if (uap->how & LOCK_EX)
3569 lf.l_type = F_WRLCK;
3570 else if (uap->how & LOCK_SH)
3571 lf.l_type = F_RDLCK;
3576 atomic_set_int(&fp->f_flag, FHASLOCK);
3577 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3578 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3584 * Duplicate the specified descriptor to a free descriptor.
3587 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3588 int openerror, int *indxp)
3590 struct filedescent *newfde, *oldfde;
3595 KASSERT(openerror == ENODEV || openerror == ENXIO,
3596 ("unexpected error %d in %s", openerror, __func__));
3599 * If the to-be-dup'd fd number is greater than the allowed number
3600 * of file descriptors, or the fd to be dup'd has already been
3601 * closed, then reject.
3603 FILEDESC_XLOCK(fdp);
3604 if ((fp = fget_locked(fdp, dfd)) == NULL) {
3605 FILEDESC_XUNLOCK(fdp);
3609 error = fdalloc(td, 0, &indx);
3611 FILEDESC_XUNLOCK(fdp);
3616 * There are two cases of interest here.
3618 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3620 * For ENXIO steal away the file structure from (dfd) and store it in
3621 * (indx). (dfd) is effectively closed by this operation.
3623 switch (openerror) {
3626 * Check that the mode the file is being opened for is a
3627 * subset of the mode of the existing descriptor.
3629 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3630 fdunused(fdp, indx);
3631 FILEDESC_XUNLOCK(fdp);
3635 fdunused(fdp, indx);
3636 FILEDESC_XUNLOCK(fdp);
3639 newfde = &fdp->fd_ofiles[indx];
3640 oldfde = &fdp->fd_ofiles[dfd];
3641 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3643 seqc_write_begin(&newfde->fde_seqc);
3645 memcpy(newfde, oldfde, fde_change_size);
3646 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3649 seqc_write_end(&newfde->fde_seqc);
3654 * Steal away the file pointer from dfd and stuff it into indx.
3656 newfde = &fdp->fd_ofiles[indx];
3657 oldfde = &fdp->fd_ofiles[dfd];
3659 seqc_write_begin(&newfde->fde_seqc);
3661 memcpy(newfde, oldfde, fde_change_size);
3662 oldfde->fde_file = NULL;
3665 seqc_write_end(&newfde->fde_seqc);
3669 FILEDESC_XUNLOCK(fdp);
3675 * This sysctl determines if we will allow a process to chroot(2) if it
3676 * has a directory open:
3677 * 0: disallowed for all processes.
3678 * 1: allowed for processes that were not already chroot(2)'ed.
3679 * 2: allowed for all processes.
3682 static int chroot_allow_open_directories = 1;
3684 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3685 &chroot_allow_open_directories, 0,
3686 "Allow a process to chroot(2) if it has a directory open");
3689 * Helper function for raised chroot(2) security function: Refuse if
3690 * any filedescriptors are open directories.
3693 chroot_refuse_vdir_fds(struct filedesc *fdp)
3699 FILEDESC_LOCK_ASSERT(fdp);
3701 lastfile = fdlastfile(fdp);
3702 for (fd = 0; fd <= lastfile; fd++) {
3703 fp = fget_locked(fdp, fd);
3706 if (fp->f_type == DTYPE_VNODE) {
3708 if (vp->v_type == VDIR)
3716 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3719 if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3720 vrefact(oldpwd->pwd_cdir);
3721 newpwd->pwd_cdir = oldpwd->pwd_cdir;
3724 if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3725 vrefact(oldpwd->pwd_rdir);
3726 newpwd->pwd_rdir = oldpwd->pwd_rdir;
3729 if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3730 vrefact(oldpwd->pwd_jdir);
3731 newpwd->pwd_jdir = oldpwd->pwd_jdir;
3736 pwd_hold_pwddesc(struct pwddesc *pdp)
3740 PWDDESC_ASSERT_XLOCKED(pdp);
3741 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3743 refcount_acquire(&pwd->pwd_refcount);
3748 pwd_hold_smr(struct pwd *pwd)
3752 if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3759 pwd_hold(struct thread *td)
3761 struct pwddesc *pdp;
3764 pdp = td->td_proc->p_pd;
3767 pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3768 if (pwd_hold_smr(pwd)) {
3774 pwd = pwd_hold_pwddesc(pdp);
3776 PWDDESC_XUNLOCK(pdp);
3785 pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3786 bzero(pwd, sizeof(*pwd));
3787 refcount_init(&pwd->pwd_refcount, 1);
3792 pwd_drop(struct pwd *pwd)
3795 if (!refcount_release(&pwd->pwd_refcount))
3798 if (pwd->pwd_cdir != NULL)
3799 vrele(pwd->pwd_cdir);
3800 if (pwd->pwd_rdir != NULL)
3801 vrele(pwd->pwd_rdir);
3802 if (pwd->pwd_jdir != NULL)
3803 vrele(pwd->pwd_jdir);
3804 uma_zfree_smr(pwd_zone, pwd);
3808 * The caller is responsible for invoking priv_check() and
3809 * mac_vnode_check_chroot() to authorize this operation.
3812 pwd_chroot(struct thread *td, struct vnode *vp)
3814 struct pwddesc *pdp;
3815 struct filedesc *fdp;
3816 struct pwd *newpwd, *oldpwd;
3819 fdp = td->td_proc->p_fd;
3820 pdp = td->td_proc->p_pd;
3821 newpwd = pwd_alloc();
3822 FILEDESC_SLOCK(fdp);
3824 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3825 if (chroot_allow_open_directories == 0 ||
3826 (chroot_allow_open_directories == 1 &&
3827 oldpwd->pwd_rdir != rootvnode)) {
3828 error = chroot_refuse_vdir_fds(fdp);
3829 FILEDESC_SUNLOCK(fdp);
3831 PWDDESC_XUNLOCK(pdp);
3836 FILEDESC_SUNLOCK(fdp);
3840 newpwd->pwd_rdir = vp;
3841 if (oldpwd->pwd_jdir == NULL) {
3843 newpwd->pwd_jdir = vp;
3845 pwd_fill(oldpwd, newpwd);
3846 pwd_set(pdp, newpwd);
3847 PWDDESC_XUNLOCK(pdp);
3853 pwd_chdir(struct thread *td, struct vnode *vp)
3855 struct pwddesc *pdp;
3856 struct pwd *newpwd, *oldpwd;
3858 VNPASS(vp->v_usecount > 0, vp);
3860 newpwd = pwd_alloc();
3861 pdp = td->td_proc->p_pd;
3863 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3864 newpwd->pwd_cdir = vp;
3865 pwd_fill(oldpwd, newpwd);
3866 pwd_set(pdp, newpwd);
3867 PWDDESC_XUNLOCK(pdp);
3872 * jail_attach(2) changes both root and working directories.
3875 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
3877 struct pwddesc *pdp;
3878 struct filedesc *fdp;
3879 struct pwd *newpwd, *oldpwd;
3882 fdp = td->td_proc->p_fd;
3883 pdp = td->td_proc->p_pd;
3884 newpwd = pwd_alloc();
3885 FILEDESC_SLOCK(fdp);
3887 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3888 error = chroot_refuse_vdir_fds(fdp);
3889 FILEDESC_SUNLOCK(fdp);
3891 PWDDESC_XUNLOCK(pdp);
3897 newpwd->pwd_rdir = vp;
3899 newpwd->pwd_cdir = vp;
3900 if (oldpwd->pwd_jdir == NULL) {
3902 newpwd->pwd_jdir = vp;
3904 pwd_fill(oldpwd, newpwd);
3905 pwd_set(pdp, newpwd);
3906 PWDDESC_XUNLOCK(pdp);
3912 pwd_ensure_dirs(void)
3914 struct pwddesc *pdp;
3915 struct pwd *oldpwd, *newpwd;
3917 pdp = curproc->p_pd;
3919 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3920 if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3921 PWDDESC_XUNLOCK(pdp);
3924 PWDDESC_XUNLOCK(pdp);
3926 newpwd = pwd_alloc();
3928 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3929 pwd_fill(oldpwd, newpwd);
3930 if (newpwd->pwd_cdir == NULL) {
3932 newpwd->pwd_cdir = rootvnode;
3934 if (newpwd->pwd_rdir == NULL) {
3936 newpwd->pwd_rdir = rootvnode;
3938 pwd_set(pdp, newpwd);
3939 PWDDESC_XUNLOCK(pdp);
3944 pwd_set_rootvnode(void)
3946 struct pwddesc *pdp;
3947 struct pwd *oldpwd, *newpwd;
3949 pdp = curproc->p_pd;
3951 newpwd = pwd_alloc();
3953 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3955 newpwd->pwd_cdir = rootvnode;
3957 newpwd->pwd_rdir = rootvnode;
3958 pwd_fill(oldpwd, newpwd);
3959 pwd_set(pdp, newpwd);
3960 PWDDESC_XUNLOCK(pdp);
3965 * Scan all active processes and prisons to see if any of them have a current
3966 * or root directory of `olddp'. If so, replace them with the new mount point.
3969 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3971 struct pwddesc *pdp;
3972 struct pwd *newpwd, *oldpwd;
3977 if (vrefcnt(olddp) == 1)
3980 newpwd = pwd_alloc();
3981 sx_slock(&allproc_lock);
3982 FOREACH_PROC_IN_SYSTEM(p) {
3989 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3990 if (oldpwd == NULL ||
3991 (oldpwd->pwd_cdir != olddp &&
3992 oldpwd->pwd_rdir != olddp &&
3993 oldpwd->pwd_jdir != olddp)) {
3994 PWDDESC_XUNLOCK(pdp);
3998 if (oldpwd->pwd_cdir == olddp) {
4000 newpwd->pwd_cdir = newdp;
4002 if (oldpwd->pwd_rdir == olddp) {
4004 newpwd->pwd_rdir = newdp;
4006 if (oldpwd->pwd_jdir == olddp) {
4008 newpwd->pwd_jdir = newdp;
4010 pwd_fill(oldpwd, newpwd);
4011 pwd_set(pdp, newpwd);
4012 PWDDESC_XUNLOCK(pdp);
4015 newpwd = pwd_alloc();
4017 sx_sunlock(&allproc_lock);
4019 if (rootvnode == olddp) {
4024 mtx_lock(&prison0.pr_mtx);
4025 if (prison0.pr_root == olddp) {
4027 prison0.pr_root = newdp;
4030 mtx_unlock(&prison0.pr_mtx);
4031 sx_slock(&allprison_lock);
4032 TAILQ_FOREACH(pr, &allprison, pr_list) {
4033 mtx_lock(&pr->pr_mtx);
4034 if (pr->pr_root == olddp) {
4036 pr->pr_root = newdp;
4039 mtx_unlock(&pr->pr_mtx);
4041 sx_sunlock(&allprison_lock);
4046 struct filedesc_to_leader *
4047 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
4049 struct filedesc_to_leader *fdtol;
4051 fdtol = malloc(sizeof(struct filedesc_to_leader),
4052 M_FILEDESC_TO_LEADER, M_WAITOK);
4053 fdtol->fdl_refcount = 1;
4054 fdtol->fdl_holdcount = 0;
4055 fdtol->fdl_wakeup = 0;
4056 fdtol->fdl_leader = leader;
4058 FILEDESC_XLOCK(fdp);
4059 fdtol->fdl_next = old->fdl_next;
4060 fdtol->fdl_prev = old;
4061 old->fdl_next = fdtol;
4062 fdtol->fdl_next->fdl_prev = fdtol;
4063 FILEDESC_XUNLOCK(fdp);
4065 fdtol->fdl_next = fdtol;
4066 fdtol->fdl_prev = fdtol;
4072 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
4075 struct filedesc *fdp;
4076 int count, off, minoff;
4078 if (*(int *)arg1 != 0)
4081 fdp = curproc->p_fd;
4083 FILEDESC_SLOCK(fdp);
4085 off = NDSLOT(fdp->fd_nfiles - 1);
4086 for (minoff = NDSLOT(0); off >= minoff; --off)
4087 count += bitcountl(map[off]);
4088 FILEDESC_SUNLOCK(fdp);
4090 return (SYSCTL_OUT(req, &count, sizeof(count)));
4093 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
4094 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
4095 "Number of open file descriptors");
4098 * Get file structures globally.
4101 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
4104 struct filedesc *fdp;
4107 int error, n, lastfile;
4109 error = sysctl_wire_old_buffer(req, 0);
4112 if (req->oldptr == NULL) {
4114 sx_slock(&allproc_lock);
4115 FOREACH_PROC_IN_SYSTEM(p) {
4117 if (p->p_state == PRS_NEW) {
4125 /* overestimates sparse tables. */
4126 n += fdp->fd_nfiles;
4129 sx_sunlock(&allproc_lock);
4130 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4133 bzero(&xf, sizeof(xf));
4134 xf.xf_size = sizeof(xf);
4135 sx_slock(&allproc_lock);
4136 FOREACH_PROC_IN_SYSTEM(p) {
4138 if (p->p_state == PRS_NEW) {
4142 if (p_cansee(req->td, p) != 0) {
4146 xf.xf_pid = p->p_pid;
4147 xf.xf_uid = p->p_ucred->cr_uid;
4152 FILEDESC_SLOCK(fdp);
4153 lastfile = fdlastfile(fdp);
4154 for (n = 0; refcount_load(&fdp->fd_refcnt) > 0 && n <= lastfile;
4156 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4159 xf.xf_file = (uintptr_t)fp;
4160 xf.xf_data = (uintptr_t)fp->f_data;
4161 xf.xf_vnode = (uintptr_t)fp->f_vnode;
4162 xf.xf_type = (uintptr_t)fp->f_type;
4163 xf.xf_count = refcount_load(&fp->f_count);
4165 xf.xf_offset = foffset_get(fp);
4166 xf.xf_flag = fp->f_flag;
4167 error = SYSCTL_OUT(req, &xf, sizeof(xf));
4171 FILEDESC_SUNLOCK(fdp);
4176 sx_sunlock(&allproc_lock);
4180 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4181 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4183 #ifdef KINFO_FILE_SIZE
4184 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4188 xlate_fflags(int fflags)
4190 static const struct {
4193 } fflags_table[] = {
4194 { FAPPEND, KF_FLAG_APPEND },
4195 { FASYNC, KF_FLAG_ASYNC },
4196 { FFSYNC, KF_FLAG_FSYNC },
4197 { FHASLOCK, KF_FLAG_HASLOCK },
4198 { FNONBLOCK, KF_FLAG_NONBLOCK },
4199 { FREAD, KF_FLAG_READ },
4200 { FWRITE, KF_FLAG_WRITE },
4201 { O_CREAT, KF_FLAG_CREAT },
4202 { O_DIRECT, KF_FLAG_DIRECT },
4203 { O_EXCL, KF_FLAG_EXCL },
4204 { O_EXEC, KF_FLAG_EXEC },
4205 { O_EXLOCK, KF_FLAG_EXLOCK },
4206 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4207 { O_SHLOCK, KF_FLAG_SHLOCK },
4208 { O_TRUNC, KF_FLAG_TRUNC }
4214 for (i = 0; i < nitems(fflags_table); i++)
4215 if (fflags & fflags_table[i].fflag)
4216 kflags |= fflags_table[i].kf_fflag;
4220 /* Trim unused data from kf_path by truncating the structure size. */
4222 pack_kinfo(struct kinfo_file *kif)
4225 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4226 strlen(kif->kf_path) + 1;
4227 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4231 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4232 struct kinfo_file *kif, struct filedesc *fdp, int flags)
4236 bzero(kif, sizeof(*kif));
4238 /* Set a default type to allow for empty fill_kinfo() methods. */
4239 kif->kf_type = KF_TYPE_UNKNOWN;
4240 kif->kf_flags = xlate_fflags(fp->f_flag);
4241 if (rightsp != NULL)
4242 kif->kf_cap_rights = *rightsp;
4244 cap_rights_init_zero(&kif->kf_cap_rights);
4246 kif->kf_ref_count = refcount_load(&fp->f_count);
4247 kif->kf_offset = foffset_get(fp);
4250 * This may drop the filedesc lock, so the 'fp' cannot be
4251 * accessed after this call.
4253 error = fo_fill_kinfo(fp, kif, fdp);
4255 kif->kf_status |= KF_ATTR_VALID;
4256 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4259 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4263 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4264 struct kinfo_file *kif, int flags)
4268 bzero(kif, sizeof(*kif));
4270 kif->kf_type = KF_TYPE_VNODE;
4271 error = vn_fill_kinfo_vnode(vp, kif);
4273 kif->kf_status |= KF_ATTR_VALID;
4274 kif->kf_flags = xlate_fflags(fflags);
4275 cap_rights_init_zero(&kif->kf_cap_rights);
4277 kif->kf_ref_count = -1;
4278 kif->kf_offset = -1;
4279 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4282 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4286 struct export_fd_buf {
4287 struct filedesc *fdp;
4288 struct pwddesc *pdp;
4291 struct kinfo_file kif;
4296 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4298 struct kinfo_file *kif;
4301 if (efbuf->remainder != -1) {
4302 if (efbuf->remainder < kif->kf_structsize) {
4303 /* Terminate export. */
4304 efbuf->remainder = 0;
4307 efbuf->remainder -= kif->kf_structsize;
4309 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
4313 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4314 struct export_fd_buf *efbuf)
4318 if (efbuf->remainder == 0)
4320 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4322 FILEDESC_SUNLOCK(efbuf->fdp);
4323 error = export_kinfo_to_sb(efbuf);
4324 FILEDESC_SLOCK(efbuf->fdp);
4329 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4330 struct export_fd_buf *efbuf)
4334 if (efbuf->remainder == 0)
4336 if (efbuf->pdp != NULL)
4337 PWDDESC_XUNLOCK(efbuf->pdp);
4338 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4339 error = export_kinfo_to_sb(efbuf);
4340 if (efbuf->pdp != NULL)
4341 PWDDESC_XLOCK(efbuf->pdp);
4346 * Store a process file descriptor information to sbuf.
4348 * Takes a locked proc as argument, and returns with the proc unlocked.
4351 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
4355 struct filedesc *fdp;
4356 struct pwddesc *pdp;
4357 struct export_fd_buf *efbuf;
4358 struct vnode *cttyvp, *textvp, *tracevp;
4360 int error, i, lastfile;
4361 cap_rights_t rights;
4363 PROC_LOCK_ASSERT(p, MA_OWNED);
4366 tracevp = p->p_tracevp;
4367 if (tracevp != NULL)
4370 textvp = p->p_textvp;
4373 /* Controlling tty. */
4375 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4376 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4383 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4387 efbuf->remainder = maxlen;
4388 efbuf->flags = flags;
4389 if (tracevp != NULL)
4390 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
4393 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
4395 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
4398 if (pdp == NULL || fdp == NULL)
4403 pwd = pwd_hold_pwddesc(pdp);
4405 /* working directory */
4406 if (pwd->pwd_cdir != NULL) {
4407 vrefact(pwd->pwd_cdir);
4408 export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD,
4411 /* root directory */
4412 if (pwd->pwd_rdir != NULL) {
4413 vrefact(pwd->pwd_rdir);
4414 export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT,
4417 /* jail directory */
4418 if (pwd->pwd_jdir != NULL) {
4419 vrefact(pwd->pwd_jdir);
4420 export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL,
4424 PWDDESC_XUNLOCK(pdp);
4427 FILEDESC_SLOCK(fdp);
4428 lastfile = fdlastfile(fdp);
4429 for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4430 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4433 rights = *cap_rights(fdp, i);
4434 #else /* !CAPABILITIES */
4435 rights = cap_no_rights;
4438 * Create sysctl entry. It is OK to drop the filedesc
4439 * lock inside of export_file_to_sb() as we will
4440 * re-validate and re-evaluate its properties when the
4443 error = export_file_to_sb(fp, i, &rights, efbuf);
4444 if (error != 0 || efbuf->remainder == 0)
4447 FILEDESC_SUNLOCK(fdp);
4453 free(efbuf, M_TEMP);
4457 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
4460 * Get per-process file descriptors for use by procstat(1), et al.
4463 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4468 int error, error2, *name;
4472 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4473 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4474 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4479 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4480 error = kern_proc_filedesc_out(p, &sb, maxlen,
4481 KERN_FILEDESC_PACK_KINFO);
4482 error2 = sbuf_finish(&sb);
4484 return (error != 0 ? error : error2);
4487 #ifdef COMPAT_FREEBSD7
4488 #ifdef KINFO_OFILE_SIZE
4489 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4493 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4496 okif->kf_structsize = sizeof(*okif);
4497 okif->kf_type = kif->kf_type;
4498 okif->kf_fd = kif->kf_fd;
4499 okif->kf_ref_count = kif->kf_ref_count;
4500 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4501 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4502 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4503 okif->kf_offset = kif->kf_offset;
4504 if (kif->kf_type == KF_TYPE_VNODE)
4505 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4507 okif->kf_vnode_type = KF_VTYPE_VNON;
4508 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4509 if (kif->kf_type == KF_TYPE_SOCKET) {
4510 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4511 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4512 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4513 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4514 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4516 okif->kf_sa_local.ss_family = AF_UNSPEC;
4517 okif->kf_sa_peer.ss_family = AF_UNSPEC;
4522 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4523 struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4528 PWDDESC_XUNLOCK(pdp);
4529 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4530 kinfo_to_okinfo(kif, okif);
4531 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4537 * Get per-process file descriptors for use by procstat(1), et al.
4540 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4542 struct kinfo_ofile *okif;
4543 struct kinfo_file *kif;
4544 struct filedesc *fdp;
4545 struct pwddesc *pdp;
4547 int error, i, lastfile, *name;
4552 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4559 if (fdp == NULL || pdp == NULL) {
4564 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4565 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4567 pwd = pwd_hold_pwddesc(pdp);
4569 if (pwd->pwd_cdir != NULL)
4570 export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4572 if (pwd->pwd_rdir != NULL)
4573 export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4575 if (pwd->pwd_jdir != NULL)
4576 export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4579 PWDDESC_XUNLOCK(pdp);
4582 FILEDESC_SLOCK(fdp);
4583 lastfile = fdlastfile(fdp);
4584 for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4585 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4587 export_file_to_kinfo(fp, i, NULL, kif, fdp,
4588 KERN_FILEDESC_PACK_KINFO);
4589 FILEDESC_SUNLOCK(fdp);
4590 kinfo_to_okinfo(kif, okif);
4591 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4592 FILEDESC_SLOCK(fdp);
4596 FILEDESC_SUNLOCK(fdp);
4604 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4605 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4606 "Process ofiledesc entries");
4607 #endif /* COMPAT_FREEBSD7 */
4610 vntype_to_kinfo(int vtype)
4615 } vtypes_table[] = {
4616 { VBAD, KF_VTYPE_VBAD },
4617 { VBLK, KF_VTYPE_VBLK },
4618 { VCHR, KF_VTYPE_VCHR },
4619 { VDIR, KF_VTYPE_VDIR },
4620 { VFIFO, KF_VTYPE_VFIFO },
4621 { VLNK, KF_VTYPE_VLNK },
4622 { VNON, KF_VTYPE_VNON },
4623 { VREG, KF_VTYPE_VREG },
4624 { VSOCK, KF_VTYPE_VSOCK }
4629 * Perform vtype translation.
4631 for (i = 0; i < nitems(vtypes_table); i++)
4632 if (vtypes_table[i].vtype == vtype)
4633 return (vtypes_table[i].kf_vtype);
4635 return (KF_VTYPE_UNKNOWN);
4638 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4639 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4640 "Process filedesc entries");
4643 * Store a process current working directory information to sbuf.
4645 * Takes a locked proc as argument, and returns with the proc unlocked.
4648 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
4650 struct pwddesc *pdp;
4652 struct export_fd_buf *efbuf;
4656 PROC_LOCK_ASSERT(p, MA_OWNED);
4663 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4666 efbuf->remainder = maxlen;
4669 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4670 cdir = pwd->pwd_cdir;
4675 error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4677 PWDDESC_XUNLOCK(pdp);
4679 free(efbuf, M_TEMP);
4684 * Get per-process current working directory.
4687 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4692 int error, error2, *name;
4696 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4697 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4698 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4703 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4704 error = kern_proc_cwd_out(p, &sb, maxlen);
4705 error2 = sbuf_finish(&sb);
4707 return (error != 0 ? error : error2);
4710 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4711 sysctl_kern_proc_cwd, "Process current working directory");
4715 * For the purposes of debugging, generate a human-readable string for the
4719 file_type_to_name(short type)
4747 case DTYPE_PROCDESC:
4751 case DTYPE_LINUXTFD:
4759 * For the purposes of debugging, identify a process (if any, perhaps one of
4760 * many) that references the passed file in its file descriptor array. Return
4763 static struct proc *
4764 file_to_first_proc(struct file *fp)
4766 struct filedesc *fdp;
4770 FOREACH_PROC_IN_SYSTEM(p) {
4771 if (p->p_state == PRS_NEW)
4776 for (n = 0; n < fdp->fd_nfiles; n++) {
4777 if (fp == fdp->fd_ofiles[n].fde_file)
4785 db_print_file(struct file *fp, int header)
4787 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4791 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4792 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4793 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4795 p = file_to_first_proc(fp);
4796 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4797 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4798 fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4799 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4804 DB_SHOW_COMMAND(file, db_show_file)
4809 db_printf("usage: show file <addr>\n");
4812 fp = (struct file *)addr;
4813 db_print_file(fp, 1);
4816 DB_SHOW_COMMAND(files, db_show_files)
4818 struct filedesc *fdp;
4825 FOREACH_PROC_IN_SYSTEM(p) {
4826 if (p->p_state == PRS_NEW)
4828 if ((fdp = p->p_fd) == NULL)
4830 for (n = 0; n < fdp->fd_nfiles; ++n) {
4831 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4833 db_print_file(fp, header);
4840 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4841 &maxfilesperproc, 0, "Maximum files allowed open per process");
4843 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4844 &maxfiles, 0, "Maximum number of files");
4846 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4847 &openfiles, 0, "System-wide number of open files");
4851 filelistinit(void *dummy)
4854 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4855 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4856 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4857 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4858 pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4859 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4861 * XXXMJG this is a temporary hack due to boot ordering issues against
4864 vfs_smr = uma_zone_get_smr(pwd_zone);
4865 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4867 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4869 /*-------------------------------------------------------------------*/
4872 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4873 int flags, struct thread *td)
4880 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4888 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4896 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4904 badfo_kqfilter(struct file *fp, struct knote *kn)
4911 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4919 badfo_close(struct file *fp, struct thread *td)
4926 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4934 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4942 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4943 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4951 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4957 struct fileops badfileops = {
4958 .fo_read = badfo_readwrite,
4959 .fo_write = badfo_readwrite,
4960 .fo_truncate = badfo_truncate,
4961 .fo_ioctl = badfo_ioctl,
4962 .fo_poll = badfo_poll,
4963 .fo_kqfilter = badfo_kqfilter,
4964 .fo_stat = badfo_stat,
4965 .fo_close = badfo_close,
4966 .fo_chmod = badfo_chmod,
4967 .fo_chown = badfo_chown,
4968 .fo_sendfile = badfo_sendfile,
4969 .fo_fill_kinfo = badfo_fill_kinfo,
4973 path_poll(struct file *fp, int events, struct ucred *active_cred,
4980 path_close(struct file *fp, struct thread *td)
4982 MPASS(fp->f_type == DTYPE_VNODE);
4983 fp->f_ops = &badfileops;
4988 struct fileops path_fileops = {
4989 .fo_read = badfo_readwrite,
4990 .fo_write = badfo_readwrite,
4991 .fo_truncate = badfo_truncate,
4992 .fo_ioctl = badfo_ioctl,
4993 .fo_poll = path_poll,
4994 .fo_kqfilter = vn_kqfilter_opath,
4995 .fo_stat = vn_statfile,
4996 .fo_close = path_close,
4997 .fo_chmod = badfo_chmod,
4998 .fo_chown = badfo_chown,
4999 .fo_sendfile = badfo_sendfile,
5000 .fo_fill_kinfo = vn_fill_kinfo,
5001 .fo_flags = DFLAG_PASSABLE,
5005 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
5006 int flags, struct thread *td)
5009 return (EOPNOTSUPP);
5013 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
5021 invfo_ioctl(struct file *fp, u_long com, void *data,
5022 struct ucred *active_cred, struct thread *td)
5029 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
5033 return (poll_no_poll(events));
5037 invfo_kqfilter(struct file *fp, struct knote *kn)
5044 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5052 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5060 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5061 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5068 /*-------------------------------------------------------------------*/
5071 * File Descriptor pseudo-device driver (/dev/fd/).
5073 * Opening minor device N dup()s the file (if any) connected to file
5074 * descriptor N belonging to the calling process. Note that this driver
5075 * consists of only the ``open()'' routine, because all subsequent
5076 * references to this file will be direct to the other driver.
5078 * XXX: we could give this one a cloning event handler if necessary.
5083 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
5087 * XXX Kludge: set curthread->td_dupfd to contain the value of the
5088 * the file descriptor being sought for duplication. The error
5089 * return ensures that the vnode for this device will be released
5090 * by vn_open. Open will detect this special error and take the
5091 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
5092 * will simply report the error.
5094 td->td_dupfd = dev2unit(dev);
5098 static struct cdevsw fildesc_cdevsw = {
5099 .d_version = D_VERSION,
5105 fildesc_drvinit(void *unused)
5109 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
5110 UID_ROOT, GID_WHEEL, 0666, "fd/0");
5111 make_dev_alias(dev, "stdin");
5112 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
5113 UID_ROOT, GID_WHEEL, 0666, "fd/1");
5114 make_dev_alias(dev, "stdout");
5115 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
5116 UID_ROOT, GID_WHEEL, 0666, "fd/2");
5117 make_dev_alias(dev, "stderr");
5120 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);