2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1989, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
42 #include "opt_capsicum.h"
44 #include "opt_ktrace.h"
46 #include <sys/param.h>
47 #include <sys/systm.h>
49 #include <sys/capsicum.h>
51 #include <sys/fcntl.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
67 #include <sys/protosw.h>
68 #include <sys/racct.h>
69 #include <sys/resourcevar.h>
71 #include <sys/signalvar.h>
76 #include <sys/syscallsubr.h>
77 #include <sys/sysctl.h>
78 #include <sys/sysproto.h>
79 #include <sys/unistd.h>
81 #include <sys/vnode.h>
82 #include <sys/ktrace.h>
86 #include <security/audit/audit.h>
93 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
94 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
95 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
96 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
97 "file desc to leader structures");
98 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
99 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
101 MALLOC_DECLARE(M_FADVISE);
103 static __read_mostly uma_zone_t file_zone;
104 static __read_mostly uma_zone_t filedesc0_zone;
105 __read_mostly uma_zone_t pwd_zone;
108 static int closefp(struct filedesc *fdp, int fd, struct file *fp,
109 struct thread *td, bool holdleaders, bool audit);
110 static int fd_first_free(struct filedesc *fdp, int low, int size);
111 static void fdgrowtable(struct filedesc *fdp, int nfd);
112 static void fdgrowtable_exp(struct filedesc *fdp, int nfd);
113 static void fdunused(struct filedesc *fdp, int fd);
114 static void fdused(struct filedesc *fdp, int fd);
115 static int getmaxfd(struct thread *td);
116 static u_long *filecaps_copy_prep(const struct filecaps *src);
117 static void filecaps_copy_finish(const struct filecaps *src,
118 struct filecaps *dst, u_long *ioctls);
119 static u_long *filecaps_free_prep(struct filecaps *fcaps);
120 static void filecaps_free_finish(u_long *ioctls);
122 static struct pwd *pwd_alloc(void);
127 * - An array of open file descriptors (fd_ofiles)
128 * - An array of file flags (fd_ofileflags)
129 * - A bitmap recording which descriptors are in use (fd_map)
131 * A process starts out with NDFILE descriptors. The value of NDFILE has
132 * been selected based the historical limit of 20 open files, and an
133 * assumption that the majority of processes, especially short-lived
134 * processes like shells, will never need more.
136 * If this initial allocation is exhausted, a larger descriptor table and
137 * map are allocated dynamically, and the pointers in the process's struct
138 * filedesc are updated to point to those. This is repeated every time
139 * the process runs out of file descriptors (provided it hasn't hit its
142 * Since threads may hold references to individual descriptor table
143 * entries, the tables are never freed. Instead, they are placed on a
144 * linked list and freed only when the struct filedesc is released.
147 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
148 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
149 #define NDSLOT(x) ((x) / NDENTRIES)
150 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
151 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
154 * SLIST entry used to keep track of ofiles which must be reclaimed when
158 struct fdescenttbl *ft_table;
159 SLIST_ENTRY(freetable) ft_next;
163 * Initial allocation: a filedesc structure + the head of SLIST used to
164 * keep track of old ofiles + enough space for NDFILE descriptors.
167 struct fdescenttbl0 {
169 struct filedescent fdt_ofiles[NDFILE];
173 struct filedesc fd_fd;
174 SLIST_HEAD(, freetable) fd_free;
175 struct fdescenttbl0 fd_dfiles;
176 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
180 * Descriptor management.
182 static int __exclusive_cache_line openfiles; /* actual number of open files */
183 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
184 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
187 * If low >= size, just return low. Otherwise find the first zero bit in the
188 * given bitmap, starting at low and not exceeding size - 1. Return size if
192 fd_first_free(struct filedesc *fdp, int low, int size)
194 NDSLOTTYPE *map = fdp->fd_map;
202 if (low % NDENTRIES) {
203 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
204 if ((mask &= ~map[off]) != 0UL)
205 return (off * NDENTRIES + ffsl(mask) - 1);
208 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
209 if (map[off] != ~0UL)
210 return (off * NDENTRIES + ffsl(~map[off]) - 1);
215 * Find the last used fd.
217 * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
218 * Otherwise use fdlastfile.
221 fdlastfile_single(struct filedesc *fdp)
223 NDSLOTTYPE *map = fdp->fd_map;
226 off = NDSLOT(fdp->fd_nfiles - 1);
227 for (minoff = NDSLOT(0); off >= minoff; --off)
229 return (off * NDENTRIES + flsl(map[off]) - 1);
234 fdlastfile(struct filedesc *fdp)
237 FILEDESC_LOCK_ASSERT(fdp);
238 return (fdlastfile_single(fdp));
242 fdisused(struct filedesc *fdp, int fd)
245 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
246 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
248 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
252 * Mark a file descriptor as used.
255 fdused_init(struct filedesc *fdp, int fd)
258 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
260 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
264 fdused(struct filedesc *fdp, int fd)
267 FILEDESC_XLOCK_ASSERT(fdp);
269 fdused_init(fdp, fd);
270 if (fd == fdp->fd_freefile)
275 * Mark a file descriptor as unused.
278 fdunused(struct filedesc *fdp, int fd)
281 FILEDESC_XLOCK_ASSERT(fdp);
283 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
284 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
285 ("fd=%d is still in use", fd));
287 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
288 if (fd < fdp->fd_freefile)
289 fdp->fd_freefile = fd;
293 * Free a file descriptor.
295 * Avoid some work if fdp is about to be destroyed.
298 fdefree_last(struct filedescent *fde)
301 filecaps_free(&fde->fde_caps);
305 fdfree(struct filedesc *fdp, int fd)
307 struct filedescent *fde;
309 FILEDESC_XLOCK_ASSERT(fdp);
310 fde = &fdp->fd_ofiles[fd];
312 seqc_write_begin(&fde->fde_seqc);
314 fde->fde_file = NULL;
316 seqc_write_end(&fde->fde_seqc);
323 * System calls on descriptors.
325 #ifndef _SYS_SYSPROTO_H_
326 struct getdtablesize_args {
332 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
338 td->td_retval[0] = getmaxfd(td);
340 PROC_LOCK(td->td_proc);
341 lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
342 PROC_UNLOCK(td->td_proc);
343 if (lim < td->td_retval[0])
344 td->td_retval[0] = lim;
350 * Duplicate a file descriptor to a particular value.
352 * Note: keep in mind that a potential race condition exists when closing
353 * descriptors from a shared descriptor table (via rfork).
355 #ifndef _SYS_SYSPROTO_H_
363 sys_dup2(struct thread *td, struct dup2_args *uap)
366 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
370 * Duplicate a file descriptor.
372 #ifndef _SYS_SYSPROTO_H_
379 sys_dup(struct thread *td, struct dup_args *uap)
382 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
386 * The file control system call.
388 #ifndef _SYS_SYSPROTO_H_
397 sys_fcntl(struct thread *td, struct fcntl_args *uap)
400 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
404 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
418 * Convert old flock structure to new.
420 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
421 fl.l_start = ofl.l_start;
422 fl.l_len = ofl.l_len;
423 fl.l_pid = ofl.l_pid;
424 fl.l_type = ofl.l_type;
425 fl.l_whence = ofl.l_whence;
439 arg1 = (intptr_t)&fl;
445 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
446 arg1 = (intptr_t)&fl;
454 error = kern_fcntl(td, fd, newcmd, arg1);
457 if (cmd == F_OGETLK) {
458 ofl.l_start = fl.l_start;
459 ofl.l_len = fl.l_len;
460 ofl.l_pid = fl.l_pid;
461 ofl.l_type = fl.l_type;
462 ofl.l_whence = fl.l_whence;
463 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
464 } else if (cmd == F_GETLK) {
465 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
471 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
473 struct filedesc *fdp;
475 struct file *fp, *fp2;
476 struct filedescent *fde;
480 int error, flg, seals, tmp;
494 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
497 case F_DUPFD_CLOEXEC:
499 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
504 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
507 case F_DUP2FD_CLOEXEC:
509 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
515 fde = fdeget_locked(fdp, fd);
518 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
521 FILEDESC_SUNLOCK(fdp);
527 fde = fdeget_locked(fdp, fd);
529 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
530 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
533 FILEDESC_XUNLOCK(fdp);
537 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
540 td->td_retval[0] = OFLAGS(fp->f_flag);
545 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
548 if (fp->f_ops == &path_fileops) {
554 tmp = flg = fp->f_flag;
556 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
557 } while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
558 tmp = fp->f_flag & FNONBLOCK;
559 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
564 tmp = fp->f_flag & FASYNC;
565 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
570 atomic_clear_int(&fp->f_flag, FNONBLOCK);
572 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
577 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
580 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
582 td->td_retval[0] = tmp;
587 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
591 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
596 error = priv_check(td, PRIV_NFS_LOCKD);
604 /* FALLTHROUGH F_SETLK */
608 flp = (struct flock *)arg;
609 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
614 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
617 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
623 if (flp->l_whence == SEEK_CUR) {
624 foffset = foffset_get(fp);
627 foffset > OFF_MAX - flp->l_start)) {
632 flp->l_start += foffset;
636 switch (flp->l_type) {
638 if ((fp->f_flag & FREAD) == 0) {
642 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
643 PROC_LOCK(p->p_leader);
644 p->p_leader->p_flag |= P_ADVLOCK;
645 PROC_UNLOCK(p->p_leader);
647 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
651 if ((fp->f_flag & FWRITE) == 0) {
655 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
656 PROC_LOCK(p->p_leader);
657 p->p_leader->p_flag |= P_ADVLOCK;
658 PROC_UNLOCK(p->p_leader);
660 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
664 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
668 if (flg != F_REMOTE) {
672 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
673 F_UNLCKSYS, flp, flg);
679 if (error != 0 || flp->l_type == F_UNLCK ||
680 flp->l_type == F_UNLCKSYS) {
686 * Check for a race with close.
688 * The vnode is now advisory locked (or unlocked, but this case
689 * is not really important) as the caller requested.
690 * We had to drop the filedesc lock, so we need to recheck if
691 * the descriptor is still valid, because if it was closed
692 * in the meantime we need to remove advisory lock from the
693 * vnode - close on any descriptor leading to an advisory
694 * locked vnode, removes that lock.
695 * We will return 0 on purpose in that case, as the result of
696 * successful advisory lock might have been externally visible
697 * already. This is fine - effectively we pretend to the caller
698 * that the closing thread was a bit slower and that the
699 * advisory lock succeeded before the close.
701 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
707 flp->l_whence = SEEK_SET;
710 flp->l_type = F_UNLCK;
711 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
712 F_UNLCK, flp, F_POSIX);
719 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
722 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
727 flp = (struct flock *)arg;
728 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
729 flp->l_type != F_UNLCK) {
734 if (flp->l_whence == SEEK_CUR) {
735 foffset = foffset_get(fp);
736 if ((flp->l_start > 0 &&
737 foffset > OFF_MAX - flp->l_start) ||
739 foffset < OFF_MIN - flp->l_start)) {
744 flp->l_start += foffset;
747 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
753 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
756 error = fo_add_seals(fp, arg);
761 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
764 if (fo_get_seals(fp, &seals) == 0)
765 td->td_retval[0] = seals;
772 arg = arg ? 128 * 1024: 0;
775 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
778 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
784 if (vp->v_type != VREG) {
791 * Exclusive lock synchronizes against f_seqcount reads and
792 * writes in sequential_heuristic().
794 error = vn_lock(vp, LK_EXCLUSIVE);
800 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
801 arg = MIN(arg, INT_MAX - bsize + 1);
802 fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
803 (arg + bsize - 1) / bsize);
804 atomic_set_int(&fp->f_flag, FRDAHEAD);
806 atomic_clear_int(&fp->f_flag, FRDAHEAD);
814 * Check if the vnode is part of a union stack (either the
815 * "union" flag from mount(2) or unionfs).
817 * Prior to introduction of this op libc's readdir would call
818 * fstatfs(2), in effect unnecessarily copying kilobytes of
819 * data just to check fs name and a mount flag.
821 * Fixing the code to handle everything in the kernel instead
822 * is a non-trivial endeavor and has low priority, thus this
823 * horrible kludge facilitates the current behavior in a much
824 * cheaper manner until someone(tm) sorts this out.
826 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
829 if (fp->f_type != DTYPE_VNODE) {
836 * Since we don't prevent dooming the vnode even non-null mp
837 * found can become immediately stale. This is tolerable since
838 * mount points are type-stable (providing safe memory access)
839 * and any vfs op on this vnode going forward will return an
840 * error (meaning return value in this case is meaningless).
842 mp = atomic_load_ptr(&vp->v_mount);
843 if (__predict_false(mp == NULL)) {
848 td->td_retval[0] = 0;
849 if (mp->mnt_kern_flag & MNTK_UNIONFS ||
850 mp->mnt_flag & MNT_UNION)
851 td->td_retval[0] = 1;
863 getmaxfd(struct thread *td)
866 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
870 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
873 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
875 struct filedesc *fdp;
876 struct filedescent *oldfde, *newfde;
878 struct file *delfp, *oldfp;
879 u_long *oioctls, *nioctls;
886 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
887 MPASS(mode < FDDUP_LASTMODE);
890 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
893 * Verify we have a valid descriptor to dup from and possibly to
894 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
895 * return EINVAL when the new descriptor is out of bounds.
900 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
901 maxfd = getmaxfd(td);
903 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
907 if (fget_locked(fdp, old) == NULL)
909 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
910 td->td_retval[0] = new;
911 if (flags & FDDUP_FLAG_CLOEXEC)
912 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
917 oldfde = &fdp->fd_ofiles[old];
918 oldfp = oldfde->fde_file;
923 * If the caller specified a file descriptor, make sure the file
924 * table is large enough to hold it, and grab it. Otherwise, just
925 * allocate a new descriptor the usual way.
930 if ((error = fdalloc(td, new, &new)) != 0) {
935 case FDDUP_MUSTREPLACE:
936 /* Target file descriptor must exist. */
937 if (fget_locked(fdp, new) == NULL) {
943 if (new >= fdp->fd_nfiles) {
945 * The resource limits are here instead of e.g.
946 * fdalloc(), because the file descriptor table may be
947 * shared between processes, so we can't really use
948 * racct_add()/racct_sub(). Instead of counting the
949 * number of actually allocated descriptors, just put
950 * the limit on the size of the file descriptor table.
953 if (RACCT_ENABLED()) {
954 error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
962 fdgrowtable_exp(fdp, new + 1);
964 if (!fdisused(fdp, new))
968 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
971 KASSERT(old != new, ("new fd is same as old"));
973 /* Refetch oldfde because the table may have grown and old one freed. */
974 oldfde = &fdp->fd_ofiles[old];
975 KASSERT(oldfp == oldfde->fde_file,
976 ("fdt_ofiles shift from growth observed at fd %d",
979 newfde = &fdp->fd_ofiles[new];
980 delfp = newfde->fde_file;
982 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
985 * Duplicate the source descriptor.
988 seqc_write_begin(&newfde->fde_seqc);
990 oioctls = filecaps_free_prep(&newfde->fde_caps);
991 memcpy(newfde, oldfde, fde_change_size);
992 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
994 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
995 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
997 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
999 seqc_write_end(&newfde->fde_seqc);
1001 td->td_retval[0] = new;
1005 if (delfp != NULL) {
1006 (void) closefp(fdp, new, delfp, td, true, false);
1007 FILEDESC_UNLOCK_ASSERT(fdp);
1010 FILEDESC_XUNLOCK(fdp);
1013 filecaps_free_finish(oioctls);
1018 sigiofree(struct sigio *sigio)
1020 crfree(sigio->sio_ucred);
1021 free(sigio, M_SIGIO);
1024 static struct sigio *
1025 funsetown_locked(struct sigio *sigio)
1030 SIGIO_ASSERT_LOCKED();
1034 *(sigio->sio_myref) = NULL;
1035 if (sigio->sio_pgid < 0) {
1036 pg = sigio->sio_pgrp;
1038 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
1039 sigio, sio_pgsigio);
1042 p = sigio->sio_proc;
1044 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
1045 sigio, sio_pgsigio);
1052 * If sigio is on the list associated with a process or process group,
1053 * disable signalling from the device, remove sigio from the list and
1057 funsetown(struct sigio **sigiop)
1059 struct sigio *sigio;
1061 /* Racy check, consumers must provide synchronization. */
1062 if (*sigiop == NULL)
1066 sigio = funsetown_locked(*sigiop);
1073 * Free a list of sigio structures. The caller must ensure that new sigio
1074 * structures cannot be added after this point. For process groups this is
1075 * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1079 funsetownlst(struct sigiolst *sigiolst)
1083 struct sigio *sigio, *tmp;
1086 sigio = SLIST_FIRST(sigiolst);
1094 sigio = SLIST_FIRST(sigiolst);
1095 if (sigio == NULL) {
1101 * Every entry of the list should belong to a single proc or pgrp.
1103 if (sigio->sio_pgid < 0) {
1104 pg = sigio->sio_pgrp;
1105 sx_assert(&proctree_lock, SX_XLOCKED);
1107 } else /* if (sigio->sio_pgid > 0) */ {
1108 p = sigio->sio_proc;
1110 KASSERT((p->p_flag & P_WEXIT) != 0,
1111 ("%s: process %p is not exiting", __func__, p));
1114 SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1115 *sigio->sio_myref = NULL;
1117 KASSERT(sigio->sio_pgid < 0,
1118 ("Proc sigio in pgrp sigio list"));
1119 KASSERT(sigio->sio_pgrp == pg,
1120 ("Bogus pgrp in sigio list"));
1121 } else /* if (p != NULL) */ {
1122 KASSERT(sigio->sio_pgid > 0,
1123 ("Pgrp sigio in proc sigio list"));
1124 KASSERT(sigio->sio_proc == p,
1125 ("Bogus proc in sigio list"));
1135 SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1140 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1142 * After permission checking, add a sigio structure to the sigio list for
1143 * the process or process group.
1146 fsetown(pid_t pgid, struct sigio **sigiop)
1150 struct sigio *osigio, *sigio;
1160 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1161 sigio->sio_pgid = pgid;
1162 sigio->sio_ucred = crhold(curthread->td_ucred);
1163 sigio->sio_myref = sigiop;
1165 sx_slock(&proctree_lock);
1167 osigio = funsetown_locked(*sigiop);
1176 * Policy - Don't allow a process to FSETOWN a process
1177 * in another session.
1179 * Remove this test to allow maximum flexibility or
1180 * restrict FSETOWN to the current process or process
1181 * group for maximum safety.
1183 if (proc->p_session != curthread->td_proc->p_session) {
1189 sigio->sio_proc = proc;
1190 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1192 } else /* if (pgid < 0) */ {
1193 pgrp = pgfind(-pgid);
1200 * Policy - Don't allow a process to FSETOWN a process
1201 * in another session.
1203 * Remove this test to allow maximum flexibility or
1204 * restrict FSETOWN to the current process or process
1205 * group for maximum safety.
1207 if (pgrp->pg_session != curthread->td_proc->p_session) {
1213 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1214 sigio->sio_pgrp = pgrp;
1217 sx_sunlock(&proctree_lock);
1226 sx_sunlock(&proctree_lock);
1234 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1237 fgetown(struct sigio **sigiop)
1242 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1248 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1253 FILEDESC_XLOCK_ASSERT(fdp);
1256 * We now hold the fp reference that used to be owned by the
1257 * descriptor array. We have to unlock the FILEDESC *AFTER*
1258 * knote_fdclose to prevent a race of the fd getting opened, a knote
1259 * added, and deleteing a knote for the new fd.
1261 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1262 knote_fdclose(td, fd);
1265 * We need to notify mqueue if the object is of type mqueue.
1267 if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1268 mq_fdclose(td, fd, fp);
1269 FILEDESC_XUNLOCK(fdp);
1272 if (AUDITING_TD(td) && audit)
1273 audit_sysclose(td, fd, fp);
1275 error = closef(fp, td);
1278 * All paths leading up to closefp() will have already removed or
1279 * replaced the fd in the filedesc table, so a restart would not
1280 * operate on the same file.
1282 if (error == ERESTART)
1289 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1290 bool holdleaders, bool audit)
1294 FILEDESC_XLOCK_ASSERT(fdp);
1297 if (td->td_proc->p_fdtol != NULL) {
1299 * Ask fdfree() to sleep to ensure that all relevant
1300 * process leaders can be traversed in closef().
1302 fdp->fd_holdleaderscount++;
1304 holdleaders = false;
1308 error = closefp_impl(fdp, fd, fp, td, audit);
1310 FILEDESC_XLOCK(fdp);
1311 fdp->fd_holdleaderscount--;
1312 if (fdp->fd_holdleaderscount == 0 &&
1313 fdp->fd_holdleaderswakeup != 0) {
1314 fdp->fd_holdleaderswakeup = 0;
1315 wakeup(&fdp->fd_holdleaderscount);
1317 FILEDESC_XUNLOCK(fdp);
1323 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1324 bool holdleaders, bool audit)
1327 FILEDESC_XLOCK_ASSERT(fdp);
1329 if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1330 return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1332 return (closefp_impl(fdp, fd, fp, td, audit));
1337 * Close a file descriptor.
1339 #ifndef _SYS_SYSPROTO_H_
1346 sys_close(struct thread *td, struct close_args *uap)
1349 return (kern_close(td, uap->fd));
1353 kern_close(struct thread *td, int fd)
1355 struct filedesc *fdp;
1358 fdp = td->td_proc->p_fd;
1360 FILEDESC_XLOCK(fdp);
1361 if ((fp = fget_locked(fdp, fd)) == NULL) {
1362 FILEDESC_XUNLOCK(fdp);
1367 /* closefp() drops the FILEDESC lock for us. */
1368 return (closefp(fdp, fd, fp, td, true, true));
1372 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1374 struct filedesc *fdp;
1375 const struct fdescenttbl *fdt;
1380 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1381 * open should not be a usage error. From a close_range() perspective,
1382 * close_range(3, ~0U, 0) in the same scenario should also likely not
1383 * be a usage error as all fd above 3 are in-fact already closed.
1385 if (highfd < lowfd) {
1389 fdp = td->td_proc->p_fd;
1390 FILEDESC_XLOCK(fdp);
1391 fdt = atomic_load_ptr(&fdp->fd_files);
1392 highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1394 if (__predict_false(fd > highfd)) {
1398 fp = fdt->fdt_ofiles[fd].fde_file;
1404 (void) closefp(fdp, fd, fp, td, true, true);
1407 FILEDESC_XLOCK(fdp);
1408 fdt = atomic_load_ptr(&fdp->fd_files);
1413 FILEDESC_XUNLOCK(fdp);
1418 #ifndef _SYS_SYSPROTO_H_
1419 struct close_range_args {
1426 sys_close_range(struct thread *td, struct close_range_args *uap)
1429 AUDIT_ARG_FD(uap->lowfd);
1430 AUDIT_ARG_CMD(uap->highfd);
1431 AUDIT_ARG_FFLAGS(uap->flags);
1433 /* No flags currently defined */
1434 if (uap->flags != 0)
1436 return (kern_close_range(td, uap->lowfd, uap->highfd));
1439 #ifdef COMPAT_FREEBSD12
1441 * Close open file descriptors.
1443 #ifndef _SYS_SYSPROTO_H_
1444 struct freebsd12_closefrom_args {
1450 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1454 AUDIT_ARG_FD(uap->lowfd);
1457 * Treat negative starting file descriptor values identical to
1458 * closefrom(0) which closes all files.
1460 lowfd = MAX(0, uap->lowfd);
1461 return (kern_close_range(td, lowfd, ~0U));
1463 #endif /* COMPAT_FREEBSD12 */
1465 #if defined(COMPAT_43)
1467 * Return status information about a file descriptor.
1469 #ifndef _SYS_SYSPROTO_H_
1470 struct ofstat_args {
1477 ofstat(struct thread *td, struct ofstat_args *uap)
1483 error = kern_fstat(td, uap->fd, &ub);
1486 error = copyout(&oub, uap->sb, sizeof(oub));
1490 #endif /* COMPAT_43 */
1492 #if defined(COMPAT_FREEBSD11)
1494 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1497 struct freebsd11_stat osb;
1500 error = kern_fstat(td, uap->fd, &sb);
1503 error = freebsd11_cvtstat(&sb, &osb);
1505 error = copyout(&osb, uap->sb, sizeof(osb));
1508 #endif /* COMPAT_FREEBSD11 */
1511 * Return status information about a file descriptor.
1513 #ifndef _SYS_SYSPROTO_H_
1521 sys_fstat(struct thread *td, struct fstat_args *uap)
1526 error = kern_fstat(td, uap->fd, &ub);
1528 error = copyout(&ub, uap->sb, sizeof(ub));
1533 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1540 error = fget(td, fd, &cap_fstat_rights, &fp);
1541 if (__predict_false(error != 0))
1544 AUDIT_ARG_FILE(td->td_proc, fp);
1546 error = fo_stat(fp, sbp, td->td_ucred, td);
1548 #ifdef __STAT_TIME_T_EXT
1549 sbp->st_atim_ext = 0;
1550 sbp->st_mtim_ext = 0;
1551 sbp->st_ctim_ext = 0;
1552 sbp->st_btim_ext = 0;
1555 if (KTRPOINT(td, KTR_STRUCT))
1556 ktrstat_error(sbp, error);
1561 #if defined(COMPAT_FREEBSD11)
1563 * Return status information about a file descriptor.
1565 #ifndef _SYS_SYSPROTO_H_
1566 struct freebsd11_nfstat_args {
1573 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1579 error = kern_fstat(td, uap->fd, &ub);
1581 freebsd11_cvtnstat(&ub, &nub);
1582 error = copyout(&nub, uap->sb, sizeof(nub));
1586 #endif /* COMPAT_FREEBSD11 */
1589 * Return pathconf information about a file descriptor.
1591 #ifndef _SYS_SYSPROTO_H_
1592 struct fpathconf_args {
1599 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1604 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1606 td->td_retval[0] = value;
1611 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1617 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1621 if (name == _PC_ASYNC_IO) {
1622 *valuep = _POSIX_ASYNCHRONOUS_IO;
1627 vn_lock(vp, LK_SHARED | LK_RETRY);
1628 error = VOP_PATHCONF(vp, name, valuep);
1630 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1631 if (name != _PC_PIPE_BUF) {
1646 * Copy filecaps structure allocating memory for ioctls array if needed.
1648 * The last parameter indicates whether the fdtable is locked. If it is not and
1649 * ioctls are encountered, copying fails and the caller must lock the table.
1651 * Note that if the table was not locked, the caller has to check the relevant
1652 * sequence counter to determine whether the operation was successful.
1655 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1659 if (src->fc_ioctls != NULL && !locked)
1661 memcpy(dst, src, sizeof(*src));
1662 if (src->fc_ioctls == NULL)
1665 KASSERT(src->fc_nioctls > 0,
1666 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1668 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1669 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1670 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1675 filecaps_copy_prep(const struct filecaps *src)
1680 if (__predict_true(src->fc_ioctls == NULL))
1683 KASSERT(src->fc_nioctls > 0,
1684 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1686 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1687 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1692 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1698 if (__predict_true(src->fc_ioctls == NULL)) {
1699 MPASS(ioctls == NULL);
1703 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1704 dst->fc_ioctls = ioctls;
1705 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1709 * Move filecaps structure to the new place and clear the old place.
1712 filecaps_move(struct filecaps *src, struct filecaps *dst)
1716 bzero(src, sizeof(*src));
1720 * Fill the given filecaps structure with full rights.
1723 filecaps_fill(struct filecaps *fcaps)
1726 CAP_ALL(&fcaps->fc_rights);
1727 fcaps->fc_ioctls = NULL;
1728 fcaps->fc_nioctls = -1;
1729 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1733 * Free memory allocated within filecaps structure.
1736 filecaps_free(struct filecaps *fcaps)
1739 free(fcaps->fc_ioctls, M_FILECAPS);
1740 bzero(fcaps, sizeof(*fcaps));
1744 filecaps_free_prep(struct filecaps *fcaps)
1748 ioctls = fcaps->fc_ioctls;
1749 bzero(fcaps, sizeof(*fcaps));
1754 filecaps_free_finish(u_long *ioctls)
1757 free(ioctls, M_FILECAPS);
1761 * Validate the given filecaps structure.
1764 filecaps_validate(const struct filecaps *fcaps, const char *func)
1767 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1768 ("%s: invalid rights", func));
1769 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1770 ("%s: invalid fcntls", func));
1771 KASSERT(fcaps->fc_fcntls == 0 ||
1772 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1773 ("%s: fcntls without CAP_FCNTL", func));
1774 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1775 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1776 ("%s: invalid ioctls", func));
1777 KASSERT(fcaps->fc_nioctls == 0 ||
1778 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1779 ("%s: ioctls without CAP_IOCTL", func));
1783 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1787 FILEDESC_XLOCK_ASSERT(fdp);
1789 nfd1 = fdp->fd_nfiles * 2;
1792 fdgrowtable(fdp, nfd1);
1796 * Grow the file table to accommodate (at least) nfd descriptors.
1799 fdgrowtable(struct filedesc *fdp, int nfd)
1801 struct filedesc0 *fdp0;
1802 struct freetable *ft;
1803 struct fdescenttbl *ntable;
1804 struct fdescenttbl *otable;
1805 int nnfiles, onfiles;
1806 NDSLOTTYPE *nmap, *omap;
1808 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1810 /* save old values */
1811 onfiles = fdp->fd_nfiles;
1812 otable = fdp->fd_files;
1815 /* compute the size of the new table */
1816 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1817 if (nnfiles <= onfiles)
1818 /* the table is already large enough */
1822 * Allocate a new table. We need enough space for the number of
1823 * entries, file entries themselves and the struct freetable we will use
1824 * when we decommission the table and place it on the freelist.
1825 * We place the struct freetable in the middle so we don't have
1826 * to worry about padding.
1828 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1829 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1830 sizeof(struct freetable),
1831 M_FILEDESC, M_ZERO | M_WAITOK);
1832 /* copy the old data */
1833 ntable->fdt_nfiles = nnfiles;
1834 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1835 onfiles * sizeof(ntable->fdt_ofiles[0]));
1838 * Allocate a new map only if the old is not large enough. It will
1839 * grow at a slower rate than the table as it can map more
1840 * entries than the table can hold.
1842 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1843 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1845 /* copy over the old data and update the pointer */
1846 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1851 * Make sure that ntable is correctly initialized before we replace
1852 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1855 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1858 * Free the old file table when not shared by other threads or processes.
1859 * The old file table is considered to be shared when either are true:
1860 * - The process has more than one thread.
1861 * - The file descriptor table has been shared via fdshare().
1863 * When shared, the old file table will be placed on a freelist
1864 * which will be processed when the struct filedesc is released.
1866 * Note that if onfiles == NDFILE, we're dealing with the original
1867 * static allocation contained within (struct filedesc0 *)fdp,
1868 * which must not be freed.
1870 if (onfiles > NDFILE) {
1872 * Note we may be called here from fdinit while allocating a
1873 * table for a new process in which case ->p_fd points
1876 if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) {
1877 free(otable, M_FILEDESC);
1879 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1880 fdp0 = (struct filedesc0 *)fdp;
1881 ft->ft_table = otable;
1882 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1886 * The map does not have the same possibility of threads still
1887 * holding references to it. So always free it as long as it
1888 * does not reference the original static allocation.
1890 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1891 free(omap, M_FILEDESC);
1895 * Allocate a file descriptor for the process.
1898 fdalloc(struct thread *td, int minfd, int *result)
1900 struct proc *p = td->td_proc;
1901 struct filedesc *fdp = p->p_fd;
1902 int fd, maxfd, allocfd;
1907 FILEDESC_XLOCK_ASSERT(fdp);
1909 if (fdp->fd_freefile > minfd)
1910 minfd = fdp->fd_freefile;
1912 maxfd = getmaxfd(td);
1915 * Search the bitmap for a free descriptor starting at minfd.
1916 * If none is found, grow the file table.
1918 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1919 if (__predict_false(fd >= maxfd))
1921 if (__predict_false(fd >= fdp->fd_nfiles)) {
1922 allocfd = min(fd * 2, maxfd);
1924 if (RACCT_ENABLED()) {
1925 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1931 * fd is already equal to first free descriptor >= minfd, so
1932 * we only need to grow the table and we are done.
1934 fdgrowtable_exp(fdp, allocfd);
1938 * Perform some sanity checks, then mark the file descriptor as
1939 * used and return it to the caller.
1941 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1942 ("invalid descriptor %d", fd));
1943 KASSERT(!fdisused(fdp, fd),
1944 ("fd_first_free() returned non-free descriptor"));
1945 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1946 ("file descriptor isn't free"));
1953 * Allocate n file descriptors for the process.
1956 fdallocn(struct thread *td, int minfd, int *fds, int n)
1958 struct proc *p = td->td_proc;
1959 struct filedesc *fdp = p->p_fd;
1962 FILEDESC_XLOCK_ASSERT(fdp);
1964 for (i = 0; i < n; i++)
1965 if (fdalloc(td, 0, &fds[i]) != 0)
1969 for (i--; i >= 0; i--)
1970 fdunused(fdp, fds[i]);
1978 * Create a new open file structure and allocate a file descriptor for the
1979 * process that refers to it. We add one reference to the file for the
1980 * descriptor table and one reference for resultfp. This is to prevent us
1981 * being preempted and the entry in the descriptor table closed after we
1982 * release the FILEDESC lock.
1985 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1986 struct filecaps *fcaps)
1991 MPASS(resultfp != NULL);
1992 MPASS(resultfd != NULL);
1994 error = _falloc_noinstall(td, &fp, 2);
1995 if (__predict_false(error != 0)) {
1999 error = finstall_refed(td, fp, &fd, flags, fcaps);
2000 if (__predict_false(error != 0)) {
2001 falloc_abort(td, fp);
2012 * Create a new open file structure without allocating a file descriptor.
2015 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2018 int maxuserfiles = maxfiles - (maxfiles / 20);
2020 static struct timeval lastfail;
2023 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2026 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2027 if ((openfiles_new >= maxuserfiles &&
2028 priv_check(td, PRIV_MAXFILES) != 0) ||
2029 openfiles_new >= maxfiles) {
2030 atomic_subtract_int(&openfiles, 1);
2031 if (ppsratecheck(&lastfail, &curfail, 1)) {
2032 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2033 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2037 fp = uma_zalloc(file_zone, M_WAITOK);
2038 bzero(fp, sizeof(*fp));
2039 refcount_init(&fp->f_count, n);
2040 fp->f_cred = crhold(td->td_ucred);
2041 fp->f_ops = &badfileops;
2047 falloc_abort(struct thread *td, struct file *fp)
2051 * For assertion purposes.
2053 refcount_init(&fp->f_count, 0);
2058 * Install a file in a file descriptor table.
2061 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2062 struct filecaps *fcaps)
2064 struct filedescent *fde;
2068 filecaps_validate(fcaps, __func__);
2069 FILEDESC_XLOCK_ASSERT(fdp);
2071 fde = &fdp->fd_ofiles[fd];
2073 seqc_write_begin(&fde->fde_seqc);
2076 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2078 filecaps_move(fcaps, &fde->fde_caps);
2080 filecaps_fill(&fde->fde_caps);
2082 seqc_write_end(&fde->fde_seqc);
2087 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2088 struct filecaps *fcaps)
2090 struct filedesc *fdp = td->td_proc->p_fd;
2095 FILEDESC_XLOCK(fdp);
2096 error = fdalloc(td, 0, fd);
2097 if (__predict_true(error == 0)) {
2098 _finstall(fdp, fp, *fd, flags, fcaps);
2100 FILEDESC_XUNLOCK(fdp);
2105 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2106 struct filecaps *fcaps)
2114 error = finstall_refed(td, fp, fd, flags, fcaps);
2115 if (__predict_false(error != 0)) {
2122 * Build a new filedesc structure from another.
2124 * If fdp is not NULL, return with it shared locked.
2127 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2129 struct filedesc0 *newfdp0;
2130 struct filedesc *newfdp;
2133 MPASS(lastfile != NULL);
2135 MPASS(lastfile == NULL);
2137 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2138 newfdp = &newfdp0->fd_fd;
2140 /* Create the file descriptor table. */
2141 FILEDESC_LOCK_INIT(newfdp);
2142 refcount_init(&newfdp->fd_refcnt, 1);
2143 refcount_init(&newfdp->fd_holdcnt, 1);
2144 newfdp->fd_map = newfdp0->fd_dmap;
2145 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2146 newfdp->fd_files->fdt_nfiles = NDFILE;
2151 FILEDESC_SLOCK(fdp);
2153 FILEDESC_SUNLOCK(fdp);
2158 *lastfile = fdlastfile(fdp);
2159 if (*lastfile < newfdp->fd_nfiles)
2161 FILEDESC_SUNLOCK(fdp);
2162 fdgrowtable(newfdp, *lastfile + 1);
2163 FILEDESC_SLOCK(fdp);
2170 * Build a pwddesc structure from another.
2171 * Copy the current, root, and jail root vnode references.
2173 * If pdp is not NULL, return with it shared locked.
2176 pdinit(struct pwddesc *pdp, bool keeplock)
2178 struct pwddesc *newpdp;
2181 newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2183 PWDDESC_LOCK_INIT(newpdp);
2184 refcount_init(&newpdp->pd_refcount, 1);
2185 newpdp->pd_cmask = CMASK;
2188 newpwd = pwd_alloc();
2189 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2194 newpwd = pwd_hold_pwddesc(pdp);
2195 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2197 PWDDESC_XUNLOCK(pdp);
2202 * Hold either filedesc or pwddesc of the passed process.
2204 * The process lock is used to synchronize against the target exiting and
2207 * Clearing can be ilustrated in 3 steps:
2208 * 1. set the pointer to NULL. Either routine can race against it, hence
2210 * 2. observe the process lock as not taken. Until then fdhold/pdhold can
2211 * race to either still see the pointer or find NULL. It is still safe to
2212 * grab a reference as clearing is stalled.
2213 * 3. after the lock is observed as not taken, any fdhold/pdhold calls are
2214 * guaranteed to see NULL, making it safe to finish clearing
2216 static struct filedesc *
2217 fdhold(struct proc *p)
2219 struct filedesc *fdp;
2221 PROC_LOCK_ASSERT(p, MA_OWNED);
2222 fdp = atomic_load_ptr(&p->p_fd);
2224 refcount_acquire(&fdp->fd_holdcnt);
2228 static struct pwddesc *
2229 pdhold(struct proc *p)
2231 struct pwddesc *pdp;
2233 PROC_LOCK_ASSERT(p, MA_OWNED);
2234 pdp = atomic_load_ptr(&p->p_pd);
2236 refcount_acquire(&pdp->pd_refcount);
2241 fddrop(struct filedesc *fdp)
2244 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2245 if (refcount_release(&fdp->fd_holdcnt) == 0)
2249 FILEDESC_LOCK_DESTROY(fdp);
2250 uma_zfree(filedesc0_zone, fdp);
2254 pddrop(struct pwddesc *pdp)
2258 if (refcount_release_if_not_last(&pdp->pd_refcount))
2262 if (refcount_release(&pdp->pd_refcount) == 0) {
2263 PWDDESC_XUNLOCK(pdp);
2266 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2268 PWDDESC_XUNLOCK(pdp);
2271 PWDDESC_LOCK_DESTROY(pdp);
2272 free(pdp, M_PWDDESC);
2276 * Share a filedesc structure.
2279 fdshare(struct filedesc *fdp)
2282 refcount_acquire(&fdp->fd_refcnt);
2287 * Share a pwddesc structure.
2290 pdshare(struct pwddesc *pdp)
2292 refcount_acquire(&pdp->pd_refcount);
2297 * Unshare a filedesc structure, if necessary by making a copy
2300 fdunshare(struct thread *td)
2302 struct filedesc *tmp;
2303 struct proc *p = td->td_proc;
2305 if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2308 tmp = fdcopy(p->p_fd);
2314 * Unshare a pwddesc structure.
2317 pdunshare(struct thread *td)
2319 struct pwddesc *pdp;
2324 if (p->p_pd->pd_refcount == 1)
2327 pdp = pdcopy(p->p_pd);
2333 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2337 td->td_proc->p_fd = fdp;
2341 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2342 * this is to ease callers, not catch errors.
2345 fdcopy(struct filedesc *fdp)
2347 struct filedesc *newfdp;
2348 struct filedescent *nfde, *ofde;
2353 newfdp = fdinit(fdp, true, &lastfile);
2354 /* copy all passable descriptors (i.e. not kqueue) */
2355 newfdp->fd_freefile = -1;
2356 for (i = 0; i <= lastfile; ++i) {
2357 ofde = &fdp->fd_ofiles[i];
2358 if (ofde->fde_file == NULL ||
2359 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2360 !fhold(ofde->fde_file)) {
2361 if (newfdp->fd_freefile == -1)
2362 newfdp->fd_freefile = i;
2365 nfde = &newfdp->fd_ofiles[i];
2367 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2368 fdused_init(newfdp, i);
2370 if (newfdp->fd_freefile == -1)
2371 newfdp->fd_freefile = i;
2372 FILEDESC_SUNLOCK(fdp);
2377 * Copy a pwddesc structure.
2380 pdcopy(struct pwddesc *pdp)
2382 struct pwddesc *newpdp;
2386 newpdp = pdinit(pdp, true);
2387 newpdp->pd_cmask = pdp->pd_cmask;
2388 PWDDESC_XUNLOCK(pdp);
2393 * Copies a filedesc structure, while remapping all file descriptors
2394 * stored inside using a translation table.
2396 * File descriptors are copied over to the new file descriptor table,
2397 * regardless of whether the close-on-exec flag is set.
2400 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2401 struct filedesc **ret)
2403 struct filedesc *newfdp;
2404 struct filedescent *nfde, *ofde;
2405 int error, i, lastfile;
2409 newfdp = fdinit(fdp, true, &lastfile);
2410 if (nfds > lastfile + 1) {
2411 /* New table cannot be larger than the old one. */
2415 /* Copy all passable descriptors (i.e. not kqueue). */
2416 newfdp->fd_freefile = nfds;
2417 for (i = 0; i < nfds; ++i) {
2418 if (fds[i] < 0 || fds[i] > lastfile) {
2419 /* File descriptor out of bounds. */
2423 ofde = &fdp->fd_ofiles[fds[i]];
2424 if (ofde->fde_file == NULL) {
2425 /* Unused file descriptor. */
2429 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2430 /* File descriptor cannot be passed. */
2434 if (!fhold(ofde->fde_file)) {
2438 nfde = &newfdp->fd_ofiles[i];
2440 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2441 fdused_init(newfdp, i);
2443 FILEDESC_SUNLOCK(fdp);
2447 FILEDESC_SUNLOCK(fdp);
2448 fdescfree_remapped(newfdp);
2453 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2454 * one of processes using it exits) and the table used to be shared.
2457 fdclearlocks(struct thread *td)
2459 struct filedesc *fdp;
2460 struct filedesc_to_leader *fdtol;
2470 MPASS(fdtol != NULL);
2472 FILEDESC_XLOCK(fdp);
2473 KASSERT(fdtol->fdl_refcount > 0,
2474 ("filedesc_to_refcount botch: fdl_refcount=%d",
2475 fdtol->fdl_refcount));
2476 if (fdtol->fdl_refcount == 1 &&
2477 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2478 lastfile = fdlastfile(fdp);
2479 for (i = 0; i <= lastfile; i++) {
2480 fp = fdp->fd_ofiles[i].fde_file;
2481 if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2484 FILEDESC_XUNLOCK(fdp);
2485 lf.l_whence = SEEK_SET;
2488 lf.l_type = F_UNLCK;
2490 (void) VOP_ADVLOCK(vp,
2491 (caddr_t)p->p_leader, F_UNLCK,
2493 FILEDESC_XLOCK(fdp);
2498 if (fdtol->fdl_refcount == 1) {
2499 if (fdp->fd_holdleaderscount > 0 &&
2500 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2502 * close() or kern_dup() has cleared a reference
2503 * in a shared file descriptor table.
2505 fdp->fd_holdleaderswakeup = 1;
2506 sx_sleep(&fdp->fd_holdleaderscount,
2507 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2510 if (fdtol->fdl_holdcount > 0) {
2512 * Ensure that fdtol->fdl_leader remains
2513 * valid in closef().
2515 fdtol->fdl_wakeup = 1;
2516 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2521 fdtol->fdl_refcount--;
2522 if (fdtol->fdl_refcount == 0 &&
2523 fdtol->fdl_holdcount == 0) {
2524 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2525 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2529 FILEDESC_XUNLOCK(fdp);
2531 free(fdtol, M_FILEDESC_TO_LEADER);
2535 * Release a filedesc structure.
2538 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2540 struct filedesc0 *fdp0;
2541 struct freetable *ft, *tft;
2542 struct filedescent *fde;
2546 KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2547 ("%s: fd table %p carries references", __func__, fdp));
2550 * Serialize with threads iterating over the table, if any.
2552 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2553 FILEDESC_XLOCK(fdp);
2554 FILEDESC_XUNLOCK(fdp);
2557 lastfile = fdlastfile_single(fdp);
2558 for (i = 0; i <= lastfile; i++) {
2559 fde = &fdp->fd_ofiles[i];
2564 (void) closef(fp, td);
2570 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2571 free(fdp->fd_map, M_FILEDESC);
2572 if (fdp->fd_nfiles > NDFILE)
2573 free(fdp->fd_files, M_FILEDESC);
2575 fdp0 = (struct filedesc0 *)fdp;
2576 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2577 free(ft->ft_table, M_FILEDESC);
2583 fdescfree(struct thread *td)
2586 struct filedesc *fdp;
2593 if (RACCT_ENABLED())
2594 racct_set_unlocked(p, RACCT_NOFILE, 0);
2597 if (p->p_fdtol != NULL)
2601 * Check fdhold for an explanation.
2603 atomic_store_ptr(&p->p_fd, NULL);
2604 atomic_thread_fence_seq_cst();
2605 PROC_WAIT_UNLOCKED(p);
2607 if (refcount_release(&fdp->fd_refcnt) == 0)
2610 fdescfree_fds(td, fdp, 1);
2614 pdescfree(struct thread *td)
2617 struct pwddesc *pdp;
2624 * Check pdhold for an explanation.
2626 atomic_store_ptr(&p->p_pd, NULL);
2627 atomic_thread_fence_seq_cst();
2628 PROC_WAIT_UNLOCKED(p);
2634 fdescfree_remapped(struct filedesc *fdp)
2637 /* fdescfree_fds() asserts that fd_refcnt == 0. */
2638 if (!refcount_release(&fdp->fd_refcnt))
2639 panic("%s: fd table %p has extra references", __func__, fdp);
2641 fdescfree_fds(curthread, fdp, 0);
2645 * For setugid programs, we don't want to people to use that setugidness
2646 * to generate error messages which write to a file which otherwise would
2647 * otherwise be off-limits to the process. We check for filesystems where
2648 * the vnode can change out from under us after execve (like [lin]procfs).
2650 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2651 * sufficient. We also don't check for setugidness since we know we are.
2654 is_unsafe(struct file *fp)
2658 if (fp->f_type != DTYPE_VNODE)
2662 return ((vp->v_vflag & VV_PROCDEP) != 0);
2666 * Make this setguid thing safe, if at all possible.
2669 fdsetugidsafety(struct thread *td)
2671 struct filedesc *fdp;
2675 fdp = td->td_proc->p_fd;
2676 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2677 ("the fdtable should not be shared"));
2678 MPASS(fdp->fd_nfiles >= 3);
2679 for (i = 0; i <= 2; i++) {
2680 fp = fdp->fd_ofiles[i].fde_file;
2681 if (fp != NULL && is_unsafe(fp)) {
2682 FILEDESC_XLOCK(fdp);
2683 knote_fdclose(td, i);
2685 * NULL-out descriptor prior to close to avoid
2686 * a race while close blocks.
2689 FILEDESC_XUNLOCK(fdp);
2690 (void) closef(fp, td);
2696 * If a specific file object occupies a specific file descriptor, close the
2697 * file descriptor entry and drop a reference on the file object. This is a
2698 * convenience function to handle a subsequent error in a function that calls
2699 * falloc() that handles the race that another thread might have closed the
2700 * file descriptor out from under the thread creating the file object.
2703 fdclose(struct thread *td, struct file *fp, int idx)
2705 struct filedesc *fdp = td->td_proc->p_fd;
2707 FILEDESC_XLOCK(fdp);
2708 if (fdp->fd_ofiles[idx].fde_file == fp) {
2710 FILEDESC_XUNLOCK(fdp);
2713 FILEDESC_XUNLOCK(fdp);
2717 * Close any files on exec?
2720 fdcloseexec(struct thread *td)
2722 struct filedesc *fdp;
2723 struct filedescent *fde;
2727 fdp = td->td_proc->p_fd;
2728 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2729 ("the fdtable should not be shared"));
2730 lastfile = fdlastfile_single(fdp);
2731 for (i = 0; i <= lastfile; i++) {
2732 fde = &fdp->fd_ofiles[i];
2734 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2735 (fde->fde_flags & UF_EXCLOSE))) {
2736 FILEDESC_XLOCK(fdp);
2738 (void) closefp(fdp, i, fp, td, false, false);
2739 FILEDESC_UNLOCK_ASSERT(fdp);
2745 * It is unsafe for set[ug]id processes to be started with file
2746 * descriptors 0..2 closed, as these descriptors are given implicit
2747 * significance in the Standard C library. fdcheckstd() will create a
2748 * descriptor referencing /dev/null for each of stdin, stdout, and
2749 * stderr that is not already open.
2752 fdcheckstd(struct thread *td)
2754 struct filedesc *fdp;
2756 int i, error, devnull;
2758 fdp = td->td_proc->p_fd;
2759 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2760 ("the fdtable should not be shared"));
2761 MPASS(fdp->fd_nfiles >= 3);
2763 for (i = 0; i <= 2; i++) {
2764 if (fdp->fd_ofiles[i].fde_file != NULL)
2767 save = td->td_retval[0];
2768 if (devnull != -1) {
2769 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2771 error = kern_openat(td, AT_FDCWD, "/dev/null",
2772 UIO_SYSSPACE, O_RDWR, 0);
2774 devnull = td->td_retval[0];
2775 KASSERT(devnull == i, ("we didn't get our fd"));
2778 td->td_retval[0] = save;
2786 * Internal form of close. Decrement reference count on file structure.
2787 * Note: td may be NULL when closing a file that was being passed in a
2791 closef(struct file *fp, struct thread *td)
2795 struct filedesc_to_leader *fdtol;
2796 struct filedesc *fdp;
2801 * POSIX record locking dictates that any close releases ALL
2802 * locks owned by this process. This is handled by setting
2803 * a flag in the unlock to free ONLY locks obeying POSIX
2804 * semantics, and not to free BSD-style file locks.
2805 * If the descriptor was in a message, POSIX-style locks
2806 * aren't passed with the descriptor, and the thread pointer
2807 * will be NULL. Callers should be careful only to pass a
2808 * NULL thread pointer when there really is no owning
2809 * context that might have locks, or the locks will be
2812 if (fp->f_type == DTYPE_VNODE) {
2814 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2815 lf.l_whence = SEEK_SET;
2818 lf.l_type = F_UNLCK;
2819 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2820 F_UNLCK, &lf, F_POSIX);
2822 fdtol = td->td_proc->p_fdtol;
2823 if (fdtol != NULL) {
2825 * Handle special case where file descriptor table is
2826 * shared between multiple process leaders.
2828 fdp = td->td_proc->p_fd;
2829 FILEDESC_XLOCK(fdp);
2830 for (fdtol = fdtol->fdl_next;
2831 fdtol != td->td_proc->p_fdtol;
2832 fdtol = fdtol->fdl_next) {
2833 if ((fdtol->fdl_leader->p_flag &
2836 fdtol->fdl_holdcount++;
2837 FILEDESC_XUNLOCK(fdp);
2838 lf.l_whence = SEEK_SET;
2841 lf.l_type = F_UNLCK;
2843 (void) VOP_ADVLOCK(vp,
2844 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2846 FILEDESC_XLOCK(fdp);
2847 fdtol->fdl_holdcount--;
2848 if (fdtol->fdl_holdcount == 0 &&
2849 fdtol->fdl_wakeup != 0) {
2850 fdtol->fdl_wakeup = 0;
2854 FILEDESC_XUNLOCK(fdp);
2857 return (fdrop_close(fp, td));
2861 * Hack for file descriptor passing code.
2864 closef_nothread(struct file *fp)
2871 * Initialize the file pointer with the specified properties.
2873 * The ops are set with release semantics to be certain that the flags, type,
2874 * and data are visible when ops is. This is to prevent ops methods from being
2875 * called with bad data.
2878 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2883 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2887 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2889 fp->f_seqcount[UIO_READ] = 1;
2890 fp->f_seqcount[UIO_WRITE] = 1;
2891 finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2896 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2897 struct file **fpp, struct filecaps *havecapsp)
2899 struct filedescent *fde;
2902 FILEDESC_LOCK_ASSERT(fdp);
2904 fde = fdeget_locked(fdp, fd);
2911 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2916 if (havecapsp != NULL)
2917 filecaps_copy(&fde->fde_caps, havecapsp, true);
2919 *fpp = fde->fde_file;
2927 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2928 struct file **fpp, struct filecaps *havecapsp)
2930 struct filedesc *fdp = td->td_proc->p_fd;
2932 #ifndef CAPABILITIES
2933 error = fget_unlocked(fdp, fd, needrightsp, fpp);
2934 if (havecapsp != NULL && error == 0)
2935 filecaps_fill(havecapsp);
2942 error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2946 if (havecapsp != NULL) {
2947 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2948 havecapsp, false)) {
2954 if (!fd_modified(fdp, fd, seq))
2963 FILEDESC_SLOCK(fdp);
2964 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2965 if (error == 0 && !fhold(*fpp))
2967 FILEDESC_SUNLOCK(fdp);
2974 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2976 const struct filedescent *fde;
2977 const struct fdescenttbl *fdt;
2978 struct filedesc *fdp;
2981 const cap_rights_t *haverights;
2982 cap_rights_t rights;
2985 VFS_SMR_ASSERT_ENTERED();
2987 rights = *ndp->ni_rightsneeded;
2988 cap_rights_set_one(&rights, CAP_LOOKUP);
2990 fdp = curproc->p_fd;
2991 fdt = fdp->fd_files;
2992 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2994 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
2995 fde = &fdt->fdt_ofiles[fd];
2996 haverights = cap_rights_fde_inline(fde);
2998 if (__predict_false(fp == NULL))
3000 if (__predict_false(cap_check_inline_transient(haverights, &rights)))
3002 *fsearch = ((fp->f_flag & FSEARCH) != 0);
3004 if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
3007 if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
3011 * Use an acquire barrier to force re-reading of fdt so it is
3012 * refreshed for verification.
3014 atomic_thread_fence_acq();
3015 fdt = fdp->fd_files;
3016 if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3019 * If file descriptor doesn't have all rights,
3020 * all lookups relative to it must also be
3021 * strictly relative.
3023 * Not yet supported by fast path.
3026 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
3027 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
3028 ndp->ni_filecaps.fc_nioctls != -1) {
3030 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
3040 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
3042 const struct fdescenttbl *fdt;
3043 struct filedesc *fdp;
3047 VFS_SMR_ASSERT_ENTERED();
3049 fdp = curproc->p_fd;
3050 fdt = fdp->fd_files;
3051 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3053 fp = fdt->fdt_ofiles[fd].fde_file;
3054 if (__predict_false(fp == NULL))
3056 *fsearch = ((fp->f_flag & FSEARCH) != 0);
3058 if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
3062 * Use an acquire barrier to force re-reading of fdt so it is
3063 * refreshed for verification.
3065 atomic_thread_fence_acq();
3066 fdt = fdp->fd_files;
3067 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3069 filecaps_fill(&ndp->ni_filecaps);
3076 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3077 struct file **fpp, seqc_t *seqp)
3080 const struct filedescent *fde;
3082 const struct fdescenttbl *fdt;
3086 cap_rights_t haverights;
3090 fdt = fdp->fd_files;
3091 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3094 * Fetch the descriptor locklessly. We avoid fdrop() races by
3095 * never raising a refcount above 0. To accomplish this we have
3096 * to use a cmpset loop rather than an atomic_add. The descriptor
3097 * must be re-verified once we acquire a reference to be certain
3098 * that the identity is still correct and we did not lose a race
3099 * due to preemption.
3103 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3104 fde = &fdt->fdt_ofiles[fd];
3105 haverights = *cap_rights_fde_inline(fde);
3107 if (!seqc_consistent(fd_seqc(fdt, fd), seq))
3110 fp = fdt->fdt_ofiles[fd].fde_file;
3115 error = cap_check_inline(&haverights, needrightsp);
3119 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3121 * Force a reload. Other thread could reallocate the
3122 * table before this fd was closed, so it is possible
3123 * that there is a stale fp pointer in cached version.
3125 fdt = atomic_load_ptr(&fdp->fd_files);
3129 * Use an acquire barrier to force re-reading of fdt so it is
3130 * refreshed for verification.
3132 atomic_thread_fence_acq();
3133 fdt = fdp->fd_files;
3135 if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
3137 if (fp == fdt->fdt_ofiles[fd].fde_file)
3140 fdrop(fp, curthread);
3152 * See the comments in fget_unlocked_seq for an explanation of how this works.
3154 * This is a simplified variant which bails out to the aforementioned routine
3155 * if anything goes wrong. In practice this only happens when userspace is
3156 * racing with itself.
3159 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3163 const struct filedescent *fde;
3165 const struct fdescenttbl *fdt;
3169 const cap_rights_t *haverights;
3172 fdt = fdp->fd_files;
3173 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3176 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3177 fde = &fdt->fdt_ofiles[fd];
3178 haverights = cap_rights_fde_inline(fde);
3181 fp = fdt->fdt_ofiles[fd].fde_file;
3183 if (__predict_false(fp == NULL))
3186 if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3189 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3193 * Use an acquire barrier to force re-reading of fdt so it is
3194 * refreshed for verification.
3196 atomic_thread_fence_acq();
3197 fdt = fdp->fd_files;
3199 if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3201 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3207 fdrop(fp, curthread);
3209 return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
3213 * Translate fd -> file when the caller guarantees the file descriptor table
3214 * can't be changed by others.
3216 * Note this does not mean the file object itself is only visible to the caller,
3217 * merely that it wont disappear without having to be referenced.
3219 * Must be paired with fput_only_user.
3223 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3226 const struct filedescent *fde;
3227 const struct fdescenttbl *fdt;
3228 const cap_rights_t *haverights;
3232 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3234 if (__predict_false(fd >= fdp->fd_nfiles))
3237 fdt = fdp->fd_files;
3238 fde = &fdt->fdt_ofiles[fd];
3240 if (__predict_false(fp == NULL))
3242 MPASS(refcount_load(&fp->f_count) > 0);
3243 haverights = cap_rights_fde_inline(fde);
3244 error = cap_check_inline(haverights, needrightsp);
3245 if (__predict_false(error != 0))
3252 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3257 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3259 if (__predict_false(fd >= fdp->fd_nfiles))
3262 fp = fdp->fd_ofiles[fd].fde_file;
3263 if (__predict_false(fp == NULL))
3266 MPASS(refcount_load(&fp->f_count) > 0);
3273 * Extract the file pointer associated with the specified descriptor for the
3274 * current user process.
3276 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3279 * File's rights will be checked against the capability rights mask.
3281 * If an error occurred the non-zero error is returned and *fpp is set to
3282 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
3283 * responsible for fdrop().
3286 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3287 cap_rights_t *needrightsp)
3289 struct filedesc *fdp;
3294 fdp = td->td_proc->p_fd;
3295 error = fget_unlocked(fdp, fd, needrightsp, &fp);
3296 if (__predict_false(error != 0))
3298 if (__predict_false(fp->f_ops == &badfileops)) {
3304 * FREAD and FWRITE failure return EBADF as per POSIX.
3310 if ((fp->f_flag & flags) == 0)
3314 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3315 ((fp->f_flag & FWRITE) != 0))
3321 KASSERT(0, ("wrong flags"));
3334 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3337 return (_fget(td, fd, fpp, 0, rightsp));
3341 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3345 #ifndef CAPABILITIES
3346 error = _fget(td, fd, fpp, 0, rightsp);
3347 if (maxprotp != NULL)
3348 *maxprotp = VM_PROT_ALL;
3351 cap_rights_t fdrights;
3352 struct filedesc *fdp;
3357 fdp = td->td_proc->p_fd;
3358 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3360 error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3361 if (__predict_false(error != 0))
3363 if (__predict_false(fp->f_ops == &badfileops)) {
3367 if (maxprotp != NULL)
3368 fdrights = *cap_rights(fdp, fd);
3369 if (!fd_modified(fdp, fd, seq))
3375 * If requested, convert capability rights to access flags.
3377 if (maxprotp != NULL)
3378 *maxprotp = cap_rights_to_vmprot(&fdrights);
3385 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3388 return (_fget(td, fd, fpp, FREAD, rightsp));
3392 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3395 return (_fget(td, fd, fpp, FWRITE, rightsp));
3399 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3402 struct filedesc *fdp = td->td_proc->p_fd;
3403 #ifndef CAPABILITIES
3404 return (fget_unlocked(fdp, fd, rightsp, fpp));
3411 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3413 error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3416 error = cap_fcntl_check(fdp, fd, needfcntl);
3417 if (!fd_modified(fdp, fd, seq))
3431 * Like fget() but loads the underlying vnode, or returns an error if the
3432 * descriptor does not represent a vnode. Note that pipes use vnodes but
3433 * never have VM objects. The returned vnode will be vref()'d.
3435 * XXX: what about the unused flags ?
3438 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3445 error = _fget(td, fd, &fp, flags, needrightsp);
3448 if (fp->f_vnode == NULL) {
3460 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3463 return (_fgetvp(td, fd, 0, rightsp, vpp));
3467 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3468 struct filecaps *havecaps, struct vnode **vpp)
3470 struct filecaps caps;
3474 error = fget_cap(td, fd, needrightsp, &fp, &caps);
3477 if (fp->f_ops == &badfileops) {
3481 if (fp->f_vnode == NULL) {
3493 filecaps_free(&caps);
3499 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3502 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3506 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3509 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3514 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3518 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3523 * Handle the last reference to a file being closed.
3525 * Without the noinline attribute clang keeps inlining the func thorough this
3526 * file when fdrop is used.
3529 _fdrop(struct file *fp, struct thread *td)
3535 count = refcount_load(&fp->f_count);
3537 panic("fdrop: fp %p count %d", fp, count);
3539 error = fo_close(fp, td);
3540 atomic_subtract_int(&openfiles, 1);
3542 free(fp->f_advice, M_FADVISE);
3543 uma_zfree(file_zone, fp);
3549 * Apply an advisory lock on a file descriptor.
3551 * Just attempt to get a record lock of the requested type on the entire file
3552 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3554 #ifndef _SYS_SYSPROTO_H_
3562 sys_flock(struct thread *td, struct flock_args *uap)
3569 error = fget(td, uap->fd, &cap_flock_rights, &fp);
3572 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
3574 return (EOPNOTSUPP);
3578 lf.l_whence = SEEK_SET;
3581 if (uap->how & LOCK_UN) {
3582 lf.l_type = F_UNLCK;
3583 atomic_clear_int(&fp->f_flag, FHASLOCK);
3584 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3587 if (uap->how & LOCK_EX)
3588 lf.l_type = F_WRLCK;
3589 else if (uap->how & LOCK_SH)
3590 lf.l_type = F_RDLCK;
3595 atomic_set_int(&fp->f_flag, FHASLOCK);
3596 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3597 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3603 * Duplicate the specified descriptor to a free descriptor.
3606 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3607 int openerror, int *indxp)
3609 struct filedescent *newfde, *oldfde;
3614 KASSERT(openerror == ENODEV || openerror == ENXIO,
3615 ("unexpected error %d in %s", openerror, __func__));
3618 * If the to-be-dup'd fd number is greater than the allowed number
3619 * of file descriptors, or the fd to be dup'd has already been
3620 * closed, then reject.
3622 FILEDESC_XLOCK(fdp);
3623 if ((fp = fget_locked(fdp, dfd)) == NULL) {
3624 FILEDESC_XUNLOCK(fdp);
3628 error = fdalloc(td, 0, &indx);
3630 FILEDESC_XUNLOCK(fdp);
3635 * There are two cases of interest here.
3637 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3639 * For ENXIO steal away the file structure from (dfd) and store it in
3640 * (indx). (dfd) is effectively closed by this operation.
3642 switch (openerror) {
3645 * Check that the mode the file is being opened for is a
3646 * subset of the mode of the existing descriptor.
3648 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3649 fdunused(fdp, indx);
3650 FILEDESC_XUNLOCK(fdp);
3654 fdunused(fdp, indx);
3655 FILEDESC_XUNLOCK(fdp);
3658 newfde = &fdp->fd_ofiles[indx];
3659 oldfde = &fdp->fd_ofiles[dfd];
3660 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3662 seqc_write_begin(&newfde->fde_seqc);
3664 memcpy(newfde, oldfde, fde_change_size);
3665 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3668 seqc_write_end(&newfde->fde_seqc);
3673 * Steal away the file pointer from dfd and stuff it into indx.
3675 newfde = &fdp->fd_ofiles[indx];
3676 oldfde = &fdp->fd_ofiles[dfd];
3678 seqc_write_begin(&newfde->fde_seqc);
3680 memcpy(newfde, oldfde, fde_change_size);
3681 oldfde->fde_file = NULL;
3684 seqc_write_end(&newfde->fde_seqc);
3688 FILEDESC_XUNLOCK(fdp);
3694 * This sysctl determines if we will allow a process to chroot(2) if it
3695 * has a directory open:
3696 * 0: disallowed for all processes.
3697 * 1: allowed for processes that were not already chroot(2)'ed.
3698 * 2: allowed for all processes.
3701 static int chroot_allow_open_directories = 1;
3703 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3704 &chroot_allow_open_directories, 0,
3705 "Allow a process to chroot(2) if it has a directory open");
3708 * Helper function for raised chroot(2) security function: Refuse if
3709 * any filedescriptors are open directories.
3712 chroot_refuse_vdir_fds(struct filedesc *fdp)
3718 FILEDESC_LOCK_ASSERT(fdp);
3720 lastfile = fdlastfile(fdp);
3721 for (fd = 0; fd <= lastfile; fd++) {
3722 fp = fget_locked(fdp, fd);
3725 if (fp->f_type == DTYPE_VNODE) {
3727 if (vp->v_type == VDIR)
3735 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3738 if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3739 vrefact(oldpwd->pwd_cdir);
3740 newpwd->pwd_cdir = oldpwd->pwd_cdir;
3743 if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3744 vrefact(oldpwd->pwd_rdir);
3745 newpwd->pwd_rdir = oldpwd->pwd_rdir;
3748 if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3749 vrefact(oldpwd->pwd_jdir);
3750 newpwd->pwd_jdir = oldpwd->pwd_jdir;
3755 pwd_hold_pwddesc(struct pwddesc *pdp)
3759 PWDDESC_ASSERT_XLOCKED(pdp);
3760 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3762 refcount_acquire(&pwd->pwd_refcount);
3767 pwd_hold_smr(struct pwd *pwd)
3771 if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3778 pwd_hold(struct thread *td)
3780 struct pwddesc *pdp;
3783 pdp = td->td_proc->p_pd;
3786 pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3787 if (pwd_hold_smr(pwd)) {
3793 pwd = pwd_hold_pwddesc(pdp);
3795 PWDDESC_XUNLOCK(pdp);
3804 pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3805 bzero(pwd, sizeof(*pwd));
3806 refcount_init(&pwd->pwd_refcount, 1);
3811 pwd_drop(struct pwd *pwd)
3814 if (!refcount_release(&pwd->pwd_refcount))
3817 if (pwd->pwd_cdir != NULL)
3818 vrele(pwd->pwd_cdir);
3819 if (pwd->pwd_rdir != NULL)
3820 vrele(pwd->pwd_rdir);
3821 if (pwd->pwd_jdir != NULL)
3822 vrele(pwd->pwd_jdir);
3823 uma_zfree_smr(pwd_zone, pwd);
3827 * The caller is responsible for invoking priv_check() and
3828 * mac_vnode_check_chroot() to authorize this operation.
3831 pwd_chroot(struct thread *td, struct vnode *vp)
3833 struct pwddesc *pdp;
3834 struct filedesc *fdp;
3835 struct pwd *newpwd, *oldpwd;
3838 fdp = td->td_proc->p_fd;
3839 pdp = td->td_proc->p_pd;
3840 newpwd = pwd_alloc();
3841 FILEDESC_SLOCK(fdp);
3843 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3844 if (chroot_allow_open_directories == 0 ||
3845 (chroot_allow_open_directories == 1 &&
3846 oldpwd->pwd_rdir != rootvnode)) {
3847 error = chroot_refuse_vdir_fds(fdp);
3848 FILEDESC_SUNLOCK(fdp);
3850 PWDDESC_XUNLOCK(pdp);
3855 FILEDESC_SUNLOCK(fdp);
3859 newpwd->pwd_rdir = vp;
3860 if (oldpwd->pwd_jdir == NULL) {
3862 newpwd->pwd_jdir = vp;
3864 pwd_fill(oldpwd, newpwd);
3865 pwd_set(pdp, newpwd);
3866 PWDDESC_XUNLOCK(pdp);
3872 pwd_chdir(struct thread *td, struct vnode *vp)
3874 struct pwddesc *pdp;
3875 struct pwd *newpwd, *oldpwd;
3877 VNPASS(vp->v_usecount > 0, vp);
3879 newpwd = pwd_alloc();
3880 pdp = td->td_proc->p_pd;
3882 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3883 newpwd->pwd_cdir = vp;
3884 pwd_fill(oldpwd, newpwd);
3885 pwd_set(pdp, newpwd);
3886 PWDDESC_XUNLOCK(pdp);
3891 * jail_attach(2) changes both root and working directories.
3894 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
3896 struct pwddesc *pdp;
3897 struct filedesc *fdp;
3898 struct pwd *newpwd, *oldpwd;
3901 fdp = td->td_proc->p_fd;
3902 pdp = td->td_proc->p_pd;
3903 newpwd = pwd_alloc();
3904 FILEDESC_SLOCK(fdp);
3906 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3907 error = chroot_refuse_vdir_fds(fdp);
3908 FILEDESC_SUNLOCK(fdp);
3910 PWDDESC_XUNLOCK(pdp);
3916 newpwd->pwd_rdir = vp;
3918 newpwd->pwd_cdir = vp;
3919 if (oldpwd->pwd_jdir == NULL) {
3921 newpwd->pwd_jdir = vp;
3923 pwd_fill(oldpwd, newpwd);
3924 pwd_set(pdp, newpwd);
3925 PWDDESC_XUNLOCK(pdp);
3931 pwd_ensure_dirs(void)
3933 struct pwddesc *pdp;
3934 struct pwd *oldpwd, *newpwd;
3936 pdp = curproc->p_pd;
3938 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3939 if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3940 PWDDESC_XUNLOCK(pdp);
3943 PWDDESC_XUNLOCK(pdp);
3945 newpwd = pwd_alloc();
3947 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3948 pwd_fill(oldpwd, newpwd);
3949 if (newpwd->pwd_cdir == NULL) {
3951 newpwd->pwd_cdir = rootvnode;
3953 if (newpwd->pwd_rdir == NULL) {
3955 newpwd->pwd_rdir = rootvnode;
3957 pwd_set(pdp, newpwd);
3958 PWDDESC_XUNLOCK(pdp);
3963 pwd_set_rootvnode(void)
3965 struct pwddesc *pdp;
3966 struct pwd *oldpwd, *newpwd;
3968 pdp = curproc->p_pd;
3970 newpwd = pwd_alloc();
3972 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3974 newpwd->pwd_cdir = rootvnode;
3976 newpwd->pwd_rdir = rootvnode;
3977 pwd_fill(oldpwd, newpwd);
3978 pwd_set(pdp, newpwd);
3979 PWDDESC_XUNLOCK(pdp);
3984 * Scan all active processes and prisons to see if any of them have a current
3985 * or root directory of `olddp'. If so, replace them with the new mount point.
3988 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3990 struct pwddesc *pdp;
3991 struct pwd *newpwd, *oldpwd;
3996 if (vrefcnt(olddp) == 1)
3999 newpwd = pwd_alloc();
4000 sx_slock(&allproc_lock);
4001 FOREACH_PROC_IN_SYSTEM(p) {
4008 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4009 if (oldpwd == NULL ||
4010 (oldpwd->pwd_cdir != olddp &&
4011 oldpwd->pwd_rdir != olddp &&
4012 oldpwd->pwd_jdir != olddp)) {
4013 PWDDESC_XUNLOCK(pdp);
4017 if (oldpwd->pwd_cdir == olddp) {
4019 newpwd->pwd_cdir = newdp;
4021 if (oldpwd->pwd_rdir == olddp) {
4023 newpwd->pwd_rdir = newdp;
4025 if (oldpwd->pwd_jdir == olddp) {
4027 newpwd->pwd_jdir = newdp;
4029 pwd_fill(oldpwd, newpwd);
4030 pwd_set(pdp, newpwd);
4031 PWDDESC_XUNLOCK(pdp);
4034 newpwd = pwd_alloc();
4036 sx_sunlock(&allproc_lock);
4038 if (rootvnode == olddp) {
4043 mtx_lock(&prison0.pr_mtx);
4044 if (prison0.pr_root == olddp) {
4046 prison0.pr_root = newdp;
4049 mtx_unlock(&prison0.pr_mtx);
4050 sx_slock(&allprison_lock);
4051 TAILQ_FOREACH(pr, &allprison, pr_list) {
4052 mtx_lock(&pr->pr_mtx);
4053 if (pr->pr_root == olddp) {
4055 pr->pr_root = newdp;
4058 mtx_unlock(&pr->pr_mtx);
4060 sx_sunlock(&allprison_lock);
4065 struct filedesc_to_leader *
4066 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
4068 struct filedesc_to_leader *fdtol;
4070 fdtol = malloc(sizeof(struct filedesc_to_leader),
4071 M_FILEDESC_TO_LEADER, M_WAITOK);
4072 fdtol->fdl_refcount = 1;
4073 fdtol->fdl_holdcount = 0;
4074 fdtol->fdl_wakeup = 0;
4075 fdtol->fdl_leader = leader;
4077 FILEDESC_XLOCK(fdp);
4078 fdtol->fdl_next = old->fdl_next;
4079 fdtol->fdl_prev = old;
4080 old->fdl_next = fdtol;
4081 fdtol->fdl_next->fdl_prev = fdtol;
4082 FILEDESC_XUNLOCK(fdp);
4084 fdtol->fdl_next = fdtol;
4085 fdtol->fdl_prev = fdtol;
4091 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
4094 struct filedesc *fdp;
4096 int count, off, minoff;
4102 if (*(int *)arg1 != 0)
4105 fdp = curproc->p_fd;
4107 FILEDESC_SLOCK(fdp);
4109 off = NDSLOT(fdp->fd_nfiles - 1);
4110 for (minoff = NDSLOT(0); off >= minoff; --off)
4111 count += bitcountl(map[off]);
4112 FILEDESC_SUNLOCK(fdp);
4114 return (SYSCTL_OUT(req, &count, sizeof(count)));
4117 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
4118 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
4119 "Number of open file descriptors");
4122 * Get file structures globally.
4125 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
4128 struct filedesc *fdp;
4131 int error, n, lastfile;
4133 error = sysctl_wire_old_buffer(req, 0);
4136 if (req->oldptr == NULL) {
4138 sx_slock(&allproc_lock);
4139 FOREACH_PROC_IN_SYSTEM(p) {
4141 if (p->p_state == PRS_NEW) {
4149 /* overestimates sparse tables. */
4150 n += fdp->fd_nfiles;
4153 sx_sunlock(&allproc_lock);
4154 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4157 bzero(&xf, sizeof(xf));
4158 xf.xf_size = sizeof(xf);
4159 sx_slock(&allproc_lock);
4160 FOREACH_PROC_IN_SYSTEM(p) {
4162 if (p->p_state == PRS_NEW) {
4166 if (p_cansee(req->td, p) != 0) {
4170 xf.xf_pid = p->p_pid;
4171 xf.xf_uid = p->p_ucred->cr_uid;
4176 FILEDESC_SLOCK(fdp);
4177 lastfile = fdlastfile(fdp);
4178 for (n = 0; refcount_load(&fdp->fd_refcnt) > 0 && n <= lastfile;
4180 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4183 xf.xf_file = (uintptr_t)fp;
4184 xf.xf_data = (uintptr_t)fp->f_data;
4185 xf.xf_vnode = (uintptr_t)fp->f_vnode;
4186 xf.xf_type = (uintptr_t)fp->f_type;
4187 xf.xf_count = refcount_load(&fp->f_count);
4189 xf.xf_offset = foffset_get(fp);
4190 xf.xf_flag = fp->f_flag;
4191 error = SYSCTL_OUT(req, &xf, sizeof(xf));
4195 FILEDESC_SUNLOCK(fdp);
4200 sx_sunlock(&allproc_lock);
4204 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4205 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4207 #ifdef KINFO_FILE_SIZE
4208 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4212 xlate_fflags(int fflags)
4214 static const struct {
4217 } fflags_table[] = {
4218 { FAPPEND, KF_FLAG_APPEND },
4219 { FASYNC, KF_FLAG_ASYNC },
4220 { FFSYNC, KF_FLAG_FSYNC },
4221 { FHASLOCK, KF_FLAG_HASLOCK },
4222 { FNONBLOCK, KF_FLAG_NONBLOCK },
4223 { FREAD, KF_FLAG_READ },
4224 { FWRITE, KF_FLAG_WRITE },
4225 { O_CREAT, KF_FLAG_CREAT },
4226 { O_DIRECT, KF_FLAG_DIRECT },
4227 { O_EXCL, KF_FLAG_EXCL },
4228 { O_EXEC, KF_FLAG_EXEC },
4229 { O_EXLOCK, KF_FLAG_EXLOCK },
4230 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4231 { O_SHLOCK, KF_FLAG_SHLOCK },
4232 { O_TRUNC, KF_FLAG_TRUNC }
4238 for (i = 0; i < nitems(fflags_table); i++)
4239 if (fflags & fflags_table[i].fflag)
4240 kflags |= fflags_table[i].kf_fflag;
4244 /* Trim unused data from kf_path by truncating the structure size. */
4246 pack_kinfo(struct kinfo_file *kif)
4249 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4250 strlen(kif->kf_path) + 1;
4251 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4255 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4256 struct kinfo_file *kif, struct filedesc *fdp, int flags)
4260 bzero(kif, sizeof(*kif));
4262 /* Set a default type to allow for empty fill_kinfo() methods. */
4263 kif->kf_type = KF_TYPE_UNKNOWN;
4264 kif->kf_flags = xlate_fflags(fp->f_flag);
4265 if (rightsp != NULL)
4266 kif->kf_cap_rights = *rightsp;
4268 cap_rights_init_zero(&kif->kf_cap_rights);
4270 kif->kf_ref_count = refcount_load(&fp->f_count);
4271 kif->kf_offset = foffset_get(fp);
4274 * This may drop the filedesc lock, so the 'fp' cannot be
4275 * accessed after this call.
4277 error = fo_fill_kinfo(fp, kif, fdp);
4279 kif->kf_status |= KF_ATTR_VALID;
4280 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4283 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4287 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4288 struct kinfo_file *kif, int flags)
4292 bzero(kif, sizeof(*kif));
4294 kif->kf_type = KF_TYPE_VNODE;
4295 error = vn_fill_kinfo_vnode(vp, kif);
4297 kif->kf_status |= KF_ATTR_VALID;
4298 kif->kf_flags = xlate_fflags(fflags);
4299 cap_rights_init_zero(&kif->kf_cap_rights);
4301 kif->kf_ref_count = -1;
4302 kif->kf_offset = -1;
4303 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4306 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4310 struct export_fd_buf {
4311 struct filedesc *fdp;
4312 struct pwddesc *pdp;
4315 struct kinfo_file kif;
4320 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4322 struct kinfo_file *kif;
4325 if (efbuf->remainder != -1) {
4326 if (efbuf->remainder < kif->kf_structsize) {
4327 /* Terminate export. */
4328 efbuf->remainder = 0;
4331 efbuf->remainder -= kif->kf_structsize;
4333 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
4337 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4338 struct export_fd_buf *efbuf)
4342 if (efbuf->remainder == 0)
4344 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4346 FILEDESC_SUNLOCK(efbuf->fdp);
4347 error = export_kinfo_to_sb(efbuf);
4348 FILEDESC_SLOCK(efbuf->fdp);
4353 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4354 struct export_fd_buf *efbuf)
4358 if (efbuf->remainder == 0)
4360 if (efbuf->pdp != NULL)
4361 PWDDESC_XUNLOCK(efbuf->pdp);
4362 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4363 error = export_kinfo_to_sb(efbuf);
4364 if (efbuf->pdp != NULL)
4365 PWDDESC_XLOCK(efbuf->pdp);
4370 * Store a process file descriptor information to sbuf.
4372 * Takes a locked proc as argument, and returns with the proc unlocked.
4375 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
4379 struct filedesc *fdp;
4380 struct pwddesc *pdp;
4381 struct export_fd_buf *efbuf;
4382 struct vnode *cttyvp, *textvp, *tracevp;
4384 int error, i, lastfile;
4385 cap_rights_t rights;
4387 PROC_LOCK_ASSERT(p, MA_OWNED);
4390 tracevp = ktr_get_tracevp(p, true);
4392 textvp = p->p_textvp;
4395 /* Controlling tty. */
4397 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4398 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4405 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4409 efbuf->remainder = maxlen;
4410 efbuf->flags = flags;
4411 if (tracevp != NULL)
4412 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
4415 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
4417 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
4420 if (pdp == NULL || fdp == NULL)
4425 pwd = pwd_hold_pwddesc(pdp);
4427 /* working directory */
4428 if (pwd->pwd_cdir != NULL) {
4429 vrefact(pwd->pwd_cdir);
4430 export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD,
4433 /* root directory */
4434 if (pwd->pwd_rdir != NULL) {
4435 vrefact(pwd->pwd_rdir);
4436 export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT,
4439 /* jail directory */
4440 if (pwd->pwd_jdir != NULL) {
4441 vrefact(pwd->pwd_jdir);
4442 export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL,
4446 PWDDESC_XUNLOCK(pdp);
4449 FILEDESC_SLOCK(fdp);
4450 lastfile = fdlastfile(fdp);
4451 for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4452 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4455 rights = *cap_rights(fdp, i);
4456 #else /* !CAPABILITIES */
4457 rights = cap_no_rights;
4460 * Create sysctl entry. It is OK to drop the filedesc
4461 * lock inside of export_file_to_sb() as we will
4462 * re-validate and re-evaluate its properties when the
4465 error = export_file_to_sb(fp, i, &rights, efbuf);
4466 if (error != 0 || efbuf->remainder == 0)
4469 FILEDESC_SUNLOCK(fdp);
4475 free(efbuf, M_TEMP);
4479 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
4482 * Get per-process file descriptors for use by procstat(1), et al.
4485 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4491 int error, error2, *name;
4499 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4500 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4501 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4506 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4507 error = kern_proc_filedesc_out(p, &sb, maxlen,
4508 KERN_FILEDESC_PACK_KINFO);
4509 error2 = sbuf_finish(&sb);
4511 return (error != 0 ? error : error2);
4514 #ifdef COMPAT_FREEBSD7
4515 #ifdef KINFO_OFILE_SIZE
4516 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4520 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4523 okif->kf_structsize = sizeof(*okif);
4524 okif->kf_type = kif->kf_type;
4525 okif->kf_fd = kif->kf_fd;
4526 okif->kf_ref_count = kif->kf_ref_count;
4527 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4528 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4529 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4530 okif->kf_offset = kif->kf_offset;
4531 if (kif->kf_type == KF_TYPE_VNODE)
4532 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4534 okif->kf_vnode_type = KF_VTYPE_VNON;
4535 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4536 if (kif->kf_type == KF_TYPE_SOCKET) {
4537 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4538 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4539 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4540 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4541 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4543 okif->kf_sa_local.ss_family = AF_UNSPEC;
4544 okif->kf_sa_peer.ss_family = AF_UNSPEC;
4549 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4550 struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4555 PWDDESC_XUNLOCK(pdp);
4556 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4557 kinfo_to_okinfo(kif, okif);
4558 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4564 * Get per-process file descriptors for use by procstat(1), et al.
4567 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4569 struct kinfo_ofile *okif;
4570 struct kinfo_file *kif;
4571 struct filedesc *fdp;
4572 struct pwddesc *pdp;
4575 int error, i, lastfile, *name;
4584 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4591 if (fdp == NULL || pdp == NULL) {
4596 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4597 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4599 pwd = pwd_hold_pwddesc(pdp);
4601 if (pwd->pwd_cdir != NULL)
4602 export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4604 if (pwd->pwd_rdir != NULL)
4605 export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4607 if (pwd->pwd_jdir != NULL)
4608 export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4611 PWDDESC_XUNLOCK(pdp);
4614 FILEDESC_SLOCK(fdp);
4615 lastfile = fdlastfile(fdp);
4616 for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4617 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4619 export_file_to_kinfo(fp, i, NULL, kif, fdp,
4620 KERN_FILEDESC_PACK_KINFO);
4621 FILEDESC_SUNLOCK(fdp);
4622 kinfo_to_okinfo(kif, okif);
4623 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4624 FILEDESC_SLOCK(fdp);
4628 FILEDESC_SUNLOCK(fdp);
4636 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4637 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4638 "Process ofiledesc entries");
4639 #endif /* COMPAT_FREEBSD7 */
4642 vntype_to_kinfo(int vtype)
4647 } vtypes_table[] = {
4648 { VBAD, KF_VTYPE_VBAD },
4649 { VBLK, KF_VTYPE_VBLK },
4650 { VCHR, KF_VTYPE_VCHR },
4651 { VDIR, KF_VTYPE_VDIR },
4652 { VFIFO, KF_VTYPE_VFIFO },
4653 { VLNK, KF_VTYPE_VLNK },
4654 { VNON, KF_VTYPE_VNON },
4655 { VREG, KF_VTYPE_VREG },
4656 { VSOCK, KF_VTYPE_VSOCK }
4661 * Perform vtype translation.
4663 for (i = 0; i < nitems(vtypes_table); i++)
4664 if (vtypes_table[i].vtype == vtype)
4665 return (vtypes_table[i].kf_vtype);
4667 return (KF_VTYPE_UNKNOWN);
4670 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4671 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4672 "Process filedesc entries");
4675 * Store a process current working directory information to sbuf.
4677 * Takes a locked proc as argument, and returns with the proc unlocked.
4680 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
4682 struct pwddesc *pdp;
4684 struct export_fd_buf *efbuf;
4688 PROC_LOCK_ASSERT(p, MA_OWNED);
4695 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4698 efbuf->remainder = maxlen;
4701 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4702 cdir = pwd->pwd_cdir;
4707 error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4709 PWDDESC_XUNLOCK(pdp);
4711 free(efbuf, M_TEMP);
4716 * Get per-process current working directory.
4719 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4725 int error, error2, *name;
4733 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4734 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4735 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4740 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4741 error = kern_proc_cwd_out(p, &sb, maxlen);
4742 error2 = sbuf_finish(&sb);
4744 return (error != 0 ? error : error2);
4747 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4748 sysctl_kern_proc_cwd, "Process current working directory");
4752 * For the purposes of debugging, generate a human-readable string for the
4756 file_type_to_name(short type)
4784 case DTYPE_PROCDESC:
4788 case DTYPE_LINUXTFD:
4796 * For the purposes of debugging, identify a process (if any, perhaps one of
4797 * many) that references the passed file in its file descriptor array. Return
4800 static struct proc *
4801 file_to_first_proc(struct file *fp)
4803 struct filedesc *fdp;
4807 FOREACH_PROC_IN_SYSTEM(p) {
4808 if (p->p_state == PRS_NEW)
4813 for (n = 0; n < fdp->fd_nfiles; n++) {
4814 if (fp == fdp->fd_ofiles[n].fde_file)
4822 db_print_file(struct file *fp, int header)
4824 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4828 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4829 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4830 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4832 p = file_to_first_proc(fp);
4833 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4834 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4835 fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4836 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4841 DB_SHOW_COMMAND(file, db_show_file)
4846 db_printf("usage: show file <addr>\n");
4849 fp = (struct file *)addr;
4850 db_print_file(fp, 1);
4853 DB_SHOW_COMMAND(files, db_show_files)
4855 struct filedesc *fdp;
4862 FOREACH_PROC_IN_SYSTEM(p) {
4863 if (p->p_state == PRS_NEW)
4865 if ((fdp = p->p_fd) == NULL)
4867 for (n = 0; n < fdp->fd_nfiles; ++n) {
4868 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4870 db_print_file(fp, header);
4877 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4878 &maxfilesperproc, 0, "Maximum files allowed open per process");
4880 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4881 &maxfiles, 0, "Maximum number of files");
4883 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4884 &openfiles, 0, "System-wide number of open files");
4888 filelistinit(void *dummy)
4891 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4892 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4893 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4894 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4895 pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4896 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4898 * XXXMJG this is a temporary hack due to boot ordering issues against
4901 vfs_smr = uma_zone_get_smr(pwd_zone);
4902 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4904 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4906 /*-------------------------------------------------------------------*/
4909 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4910 int flags, struct thread *td)
4917 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4925 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4933 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4941 badfo_kqfilter(struct file *fp, struct knote *kn)
4948 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4956 badfo_close(struct file *fp, struct thread *td)
4963 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4971 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4979 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4980 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4988 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4994 struct fileops badfileops = {
4995 .fo_read = badfo_readwrite,
4996 .fo_write = badfo_readwrite,
4997 .fo_truncate = badfo_truncate,
4998 .fo_ioctl = badfo_ioctl,
4999 .fo_poll = badfo_poll,
5000 .fo_kqfilter = badfo_kqfilter,
5001 .fo_stat = badfo_stat,
5002 .fo_close = badfo_close,
5003 .fo_chmod = badfo_chmod,
5004 .fo_chown = badfo_chown,
5005 .fo_sendfile = badfo_sendfile,
5006 .fo_fill_kinfo = badfo_fill_kinfo,
5010 path_poll(struct file *fp, int events, struct ucred *active_cred,
5017 path_close(struct file *fp, struct thread *td)
5019 MPASS(fp->f_type == DTYPE_VNODE);
5020 fp->f_ops = &badfileops;
5025 struct fileops path_fileops = {
5026 .fo_read = badfo_readwrite,
5027 .fo_write = badfo_readwrite,
5028 .fo_truncate = badfo_truncate,
5029 .fo_ioctl = badfo_ioctl,
5030 .fo_poll = path_poll,
5031 .fo_kqfilter = vn_kqfilter_opath,
5032 .fo_stat = vn_statfile,
5033 .fo_close = path_close,
5034 .fo_chmod = badfo_chmod,
5035 .fo_chown = badfo_chown,
5036 .fo_sendfile = badfo_sendfile,
5037 .fo_fill_kinfo = vn_fill_kinfo,
5038 .fo_flags = DFLAG_PASSABLE,
5042 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
5043 int flags, struct thread *td)
5046 return (EOPNOTSUPP);
5050 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
5058 invfo_ioctl(struct file *fp, u_long com, void *data,
5059 struct ucred *active_cred, struct thread *td)
5066 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
5070 return (poll_no_poll(events));
5074 invfo_kqfilter(struct file *fp, struct knote *kn)
5081 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5089 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5097 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5098 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5105 /*-------------------------------------------------------------------*/
5108 * File Descriptor pseudo-device driver (/dev/fd/).
5110 * Opening minor device N dup()s the file (if any) connected to file
5111 * descriptor N belonging to the calling process. Note that this driver
5112 * consists of only the ``open()'' routine, because all subsequent
5113 * references to this file will be direct to the other driver.
5115 * XXX: we could give this one a cloning event handler if necessary.
5120 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
5124 * XXX Kludge: set curthread->td_dupfd to contain the value of the
5125 * the file descriptor being sought for duplication. The error
5126 * return ensures that the vnode for this device will be released
5127 * by vn_open. Open will detect this special error and take the
5128 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
5129 * will simply report the error.
5131 td->td_dupfd = dev2unit(dev);
5135 static struct cdevsw fildesc_cdevsw = {
5136 .d_version = D_VERSION,
5142 fildesc_drvinit(void *unused)
5146 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
5147 UID_ROOT, GID_WHEEL, 0666, "fd/0");
5148 make_dev_alias(dev, "stdin");
5149 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
5150 UID_ROOT, GID_WHEEL, 0666, "fd/1");
5151 make_dev_alias(dev, "stdout");
5152 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
5153 UID_ROOT, GID_WHEEL, 0666, "fd/2");
5154 make_dev_alias(dev, "stderr");
5157 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);