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 void export_file_to_kinfo(struct file *fp, int fd,
111 cap_rights_t *rightsp, struct kinfo_file *kif,
112 struct filedesc *fdp, int flags);
113 static int fd_first_free(struct filedesc *fdp, int low, int size);
114 static void fdgrowtable(struct filedesc *fdp, int nfd);
115 static void fdgrowtable_exp(struct filedesc *fdp, int nfd);
116 static void fdunused(struct filedesc *fdp, int fd);
117 static void fdused(struct filedesc *fdp, int fd);
118 static int getmaxfd(struct thread *td);
119 static u_long *filecaps_copy_prep(const struct filecaps *src);
120 static void filecaps_copy_finish(const struct filecaps *src,
121 struct filecaps *dst, u_long *ioctls);
122 static u_long *filecaps_free_prep(struct filecaps *fcaps);
123 static void filecaps_free_finish(u_long *ioctls);
125 static struct pwd *pwd_alloc(void);
130 * - An array of open file descriptors (fd_ofiles)
131 * - An array of file flags (fd_ofileflags)
132 * - A bitmap recording which descriptors are in use (fd_map)
134 * A process starts out with NDFILE descriptors. The value of NDFILE has
135 * been selected based the historical limit of 20 open files, and an
136 * assumption that the majority of processes, especially short-lived
137 * processes like shells, will never need more.
139 * If this initial allocation is exhausted, a larger descriptor table and
140 * map are allocated dynamically, and the pointers in the process's struct
141 * filedesc are updated to point to those. This is repeated every time
142 * the process runs out of file descriptors (provided it hasn't hit its
145 * Since threads may hold references to individual descriptor table
146 * entries, the tables are never freed. Instead, they are placed on a
147 * linked list and freed only when the struct filedesc is released.
150 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
151 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
152 #define NDSLOT(x) ((x) / NDENTRIES)
153 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
154 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
157 * SLIST entry used to keep track of ofiles which must be reclaimed when
161 struct fdescenttbl *ft_table;
162 SLIST_ENTRY(freetable) ft_next;
166 * Initial allocation: a filedesc structure + the head of SLIST used to
167 * keep track of old ofiles + enough space for NDFILE descriptors.
170 struct fdescenttbl0 {
172 struct filedescent fdt_ofiles[NDFILE];
176 struct filedesc fd_fd;
177 SLIST_HEAD(, freetable) fd_free;
178 struct fdescenttbl0 fd_dfiles;
179 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
183 * Descriptor management.
185 static int __exclusive_cache_line openfiles; /* actual number of open files */
186 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
187 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
190 * If low >= size, just return low. Otherwise find the first zero bit in the
191 * given bitmap, starting at low and not exceeding size - 1. Return size if
195 fd_first_free(struct filedesc *fdp, int low, int size)
197 NDSLOTTYPE *map = fdp->fd_map;
205 if (low % NDENTRIES) {
206 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
207 if ((mask &= ~map[off]) != 0UL)
208 return (off * NDENTRIES + ffsl(mask) - 1);
211 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
212 if (map[off] != ~0UL)
213 return (off * NDENTRIES + ffsl(~map[off]) - 1);
218 * Find the last used fd.
220 * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
221 * Otherwise use fdlastfile.
224 fdlastfile_single(struct filedesc *fdp)
226 NDSLOTTYPE *map = fdp->fd_map;
229 off = NDSLOT(fdp->fd_nfiles - 1);
230 for (minoff = NDSLOT(0); off >= minoff; --off)
232 return (off * NDENTRIES + flsl(map[off]) - 1);
237 fdlastfile(struct filedesc *fdp)
240 FILEDESC_LOCK_ASSERT(fdp);
241 return (fdlastfile_single(fdp));
245 fdisused(struct filedesc *fdp, int fd)
248 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
249 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
251 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
255 * Mark a file descriptor as used.
258 fdused_init(struct filedesc *fdp, int fd)
261 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
263 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
267 fdused(struct filedesc *fdp, int fd)
270 FILEDESC_XLOCK_ASSERT(fdp);
272 fdused_init(fdp, fd);
273 if (fd == fdp->fd_freefile)
278 * Mark a file descriptor as unused.
281 fdunused(struct filedesc *fdp, int fd)
284 FILEDESC_XLOCK_ASSERT(fdp);
286 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
287 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
288 ("fd=%d is still in use", fd));
290 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
291 if (fd < fdp->fd_freefile)
292 fdp->fd_freefile = fd;
296 * Free a file descriptor.
298 * Avoid some work if fdp is about to be destroyed.
301 fdefree_last(struct filedescent *fde)
304 filecaps_free(&fde->fde_caps);
308 fdfree(struct filedesc *fdp, int fd)
310 struct filedescent *fde;
312 FILEDESC_XLOCK_ASSERT(fdp);
313 fde = &fdp->fd_ofiles[fd];
315 seqc_write_begin(&fde->fde_seqc);
317 fde->fde_file = NULL;
319 seqc_write_end(&fde->fde_seqc);
326 * System calls on descriptors.
328 #ifndef _SYS_SYSPROTO_H_
329 struct getdtablesize_args {
335 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
341 td->td_retval[0] = getmaxfd(td);
343 PROC_LOCK(td->td_proc);
344 lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
345 PROC_UNLOCK(td->td_proc);
346 if (lim < td->td_retval[0])
347 td->td_retval[0] = lim;
353 * Duplicate a file descriptor to a particular value.
355 * Note: keep in mind that a potential race condition exists when closing
356 * descriptors from a shared descriptor table (via rfork).
358 #ifndef _SYS_SYSPROTO_H_
366 sys_dup2(struct thread *td, struct dup2_args *uap)
369 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
373 * Duplicate a file descriptor.
375 #ifndef _SYS_SYSPROTO_H_
382 sys_dup(struct thread *td, struct dup_args *uap)
385 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
389 * The file control system call.
391 #ifndef _SYS_SYSPROTO_H_
400 sys_fcntl(struct thread *td, struct fcntl_args *uap)
403 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
407 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
421 * Convert old flock structure to new.
423 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
424 fl.l_start = ofl.l_start;
425 fl.l_len = ofl.l_len;
426 fl.l_pid = ofl.l_pid;
427 fl.l_type = ofl.l_type;
428 fl.l_whence = ofl.l_whence;
442 arg1 = (intptr_t)&fl;
448 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
449 arg1 = (intptr_t)&fl;
457 error = kern_fcntl(td, fd, newcmd, arg1);
460 if (cmd == F_OGETLK) {
461 ofl.l_start = fl.l_start;
462 ofl.l_len = fl.l_len;
463 ofl.l_pid = fl.l_pid;
464 ofl.l_type = fl.l_type;
465 ofl.l_whence = fl.l_whence;
466 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
467 } else if (cmd == F_GETLK) {
468 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
474 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
476 struct filedesc *fdp;
478 struct file *fp, *fp2;
479 struct filedescent *fde;
483 struct kinfo_file *kif;
484 int error, flg, kif_sz, seals, tmp;
498 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
501 case F_DUPFD_CLOEXEC:
503 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
508 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
511 case F_DUP2FD_CLOEXEC:
513 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
519 fde = fdeget_locked(fdp, fd);
522 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
525 FILEDESC_SUNLOCK(fdp);
531 fde = fdeget_locked(fdp, fd);
533 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
534 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
537 FILEDESC_XUNLOCK(fdp);
541 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
544 td->td_retval[0] = OFLAGS(fp->f_flag);
549 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
552 if (fp->f_ops == &path_fileops) {
558 tmp = flg = fp->f_flag;
560 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
561 } while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
562 tmp = fp->f_flag & FNONBLOCK;
563 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
568 tmp = fp->f_flag & FASYNC;
569 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
574 atomic_clear_int(&fp->f_flag, FNONBLOCK);
576 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
581 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
584 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
586 td->td_retval[0] = tmp;
591 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
595 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
600 error = priv_check(td, PRIV_NFS_LOCKD);
608 /* FALLTHROUGH F_SETLK */
612 flp = (struct flock *)arg;
613 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
618 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
621 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
627 if (flp->l_whence == SEEK_CUR) {
628 foffset = foffset_get(fp);
631 foffset > OFF_MAX - flp->l_start)) {
636 flp->l_start += foffset;
640 switch (flp->l_type) {
642 if ((fp->f_flag & FREAD) == 0) {
646 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
647 PROC_LOCK(p->p_leader);
648 p->p_leader->p_flag |= P_ADVLOCK;
649 PROC_UNLOCK(p->p_leader);
651 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
655 if ((fp->f_flag & FWRITE) == 0) {
659 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
660 PROC_LOCK(p->p_leader);
661 p->p_leader->p_flag |= P_ADVLOCK;
662 PROC_UNLOCK(p->p_leader);
664 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
668 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
672 if (flg != F_REMOTE) {
676 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
677 F_UNLCKSYS, flp, flg);
683 if (error != 0 || flp->l_type == F_UNLCK ||
684 flp->l_type == F_UNLCKSYS) {
690 * Check for a race with close.
692 * The vnode is now advisory locked (or unlocked, but this case
693 * is not really important) as the caller requested.
694 * We had to drop the filedesc lock, so we need to recheck if
695 * the descriptor is still valid, because if it was closed
696 * in the meantime we need to remove advisory lock from the
697 * vnode - close on any descriptor leading to an advisory
698 * locked vnode, removes that lock.
699 * We will return 0 on purpose in that case, as the result of
700 * successful advisory lock might have been externally visible
701 * already. This is fine - effectively we pretend to the caller
702 * that the closing thread was a bit slower and that the
703 * advisory lock succeeded before the close.
705 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
711 flp->l_whence = SEEK_SET;
714 flp->l_type = F_UNLCK;
715 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
716 F_UNLCK, flp, F_POSIX);
723 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
726 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
731 flp = (struct flock *)arg;
732 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
733 flp->l_type != F_UNLCK) {
738 if (flp->l_whence == SEEK_CUR) {
739 foffset = foffset_get(fp);
740 if ((flp->l_start > 0 &&
741 foffset > OFF_MAX - flp->l_start) ||
743 foffset < OFF_MIN - flp->l_start)) {
748 flp->l_start += foffset;
751 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
757 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
760 error = fo_add_seals(fp, arg);
765 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
768 if (fo_get_seals(fp, &seals) == 0)
769 td->td_retval[0] = seals;
776 arg = arg ? 128 * 1024: 0;
779 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
782 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
788 if (vp->v_type != VREG) {
795 * Exclusive lock synchronizes against f_seqcount reads and
796 * writes in sequential_heuristic().
798 error = vn_lock(vp, LK_EXCLUSIVE);
804 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
805 arg = MIN(arg, INT_MAX - bsize + 1);
806 fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
807 (arg + bsize - 1) / bsize);
808 atomic_set_int(&fp->f_flag, FRDAHEAD);
810 atomic_clear_int(&fp->f_flag, FRDAHEAD);
818 * Check if the vnode is part of a union stack (either the
819 * "union" flag from mount(2) or unionfs).
821 * Prior to introduction of this op libc's readdir would call
822 * fstatfs(2), in effect unnecessarily copying kilobytes of
823 * data just to check fs name and a mount flag.
825 * Fixing the code to handle everything in the kernel instead
826 * is a non-trivial endeavor and has low priority, thus this
827 * horrible kludge facilitates the current behavior in a much
828 * cheaper manner until someone(tm) sorts this out.
830 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
833 if (fp->f_type != DTYPE_VNODE) {
840 * Since we don't prevent dooming the vnode even non-null mp
841 * found can become immediately stale. This is tolerable since
842 * mount points are type-stable (providing safe memory access)
843 * and any vfs op on this vnode going forward will return an
844 * error (meaning return value in this case is meaningless).
846 mp = atomic_load_ptr(&vp->v_mount);
847 if (__predict_false(mp == NULL)) {
852 td->td_retval[0] = 0;
853 if (mp->mnt_kern_flag & MNTK_UNIONFS ||
854 mp->mnt_flag & MNT_UNION)
855 td->td_retval[0] = 1;
860 #ifdef CAPABILITY_MODE
861 if (IN_CAPABILITY_MODE(td)) {
866 error = copyin((void *)arg, &kif_sz, sizeof(kif_sz));
869 if (kif_sz != sizeof(*kif)) {
873 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK | M_ZERO);
875 error = fget_cap_locked(fdp, fd, &cap_fcntl_rights, &fp, NULL);
876 if (error == 0 && fhold(fp)) {
877 export_file_to_kinfo(fp, fd, NULL, kif, fdp, 0);
878 FILEDESC_SUNLOCK(fdp);
880 if ((kif->kf_status & KF_ATTR_VALID) != 0) {
881 kif->kf_structsize = sizeof(*kif);
882 error = copyout(kif, (void *)arg, sizeof(*kif));
887 FILEDESC_SUNLOCK(fdp);
902 getmaxfd(struct thread *td)
905 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
909 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
912 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
914 struct filedesc *fdp;
915 struct filedescent *oldfde, *newfde;
917 struct file *delfp, *oldfp;
918 u_long *oioctls, *nioctls;
925 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
926 MPASS(mode < FDDUP_LASTMODE);
929 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
932 * Verify we have a valid descriptor to dup from and possibly to
933 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
934 * return EINVAL when the new descriptor is out of bounds.
939 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
940 maxfd = getmaxfd(td);
942 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
946 if (fget_locked(fdp, old) == NULL)
948 if (mode == FDDUP_FIXED && old == new) {
949 td->td_retval[0] = new;
950 if (flags & FDDUP_FLAG_CLOEXEC)
951 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
956 oldfde = &fdp->fd_ofiles[old];
957 oldfp = oldfde->fde_file;
962 * If the caller specified a file descriptor, make sure the file
963 * table is large enough to hold it, and grab it. Otherwise, just
964 * allocate a new descriptor the usual way.
969 if ((error = fdalloc(td, new, &new)) != 0) {
975 if (new >= fdp->fd_nfiles) {
977 * The resource limits are here instead of e.g.
978 * fdalloc(), because the file descriptor table may be
979 * shared between processes, so we can't really use
980 * racct_add()/racct_sub(). Instead of counting the
981 * number of actually allocated descriptors, just put
982 * the limit on the size of the file descriptor table.
985 if (RACCT_ENABLED()) {
986 error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
994 fdgrowtable_exp(fdp, new + 1);
996 if (!fdisused(fdp, new))
1000 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
1003 KASSERT(old != new, ("new fd is same as old"));
1005 /* Refetch oldfde because the table may have grown and old one freed. */
1006 oldfde = &fdp->fd_ofiles[old];
1007 KASSERT(oldfp == oldfde->fde_file,
1008 ("fdt_ofiles shift from growth observed at fd %d",
1011 newfde = &fdp->fd_ofiles[new];
1012 delfp = newfde->fde_file;
1014 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
1017 * Duplicate the source descriptor.
1020 seqc_write_begin(&newfde->fde_seqc);
1022 oioctls = filecaps_free_prep(&newfde->fde_caps);
1023 memcpy(newfde, oldfde, fde_change_size);
1024 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
1026 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
1027 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
1029 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
1031 seqc_write_end(&newfde->fde_seqc);
1033 td->td_retval[0] = new;
1037 if (delfp != NULL) {
1038 (void) closefp(fdp, new, delfp, td, true, false);
1039 FILEDESC_UNLOCK_ASSERT(fdp);
1042 FILEDESC_XUNLOCK(fdp);
1045 filecaps_free_finish(oioctls);
1050 sigiofree(struct sigio *sigio)
1052 crfree(sigio->sio_ucred);
1053 free(sigio, M_SIGIO);
1056 static struct sigio *
1057 funsetown_locked(struct sigio *sigio)
1062 SIGIO_ASSERT_LOCKED();
1066 *sigio->sio_myref = NULL;
1067 if (sigio->sio_pgid < 0) {
1068 pg = sigio->sio_pgrp;
1070 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio);
1073 p = sigio->sio_proc;
1075 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio);
1082 * If sigio is on the list associated with a process or process group,
1083 * disable signalling from the device, remove sigio from the list and
1087 funsetown(struct sigio **sigiop)
1089 struct sigio *sigio;
1091 /* Racy check, consumers must provide synchronization. */
1092 if (*sigiop == NULL)
1096 sigio = funsetown_locked(*sigiop);
1103 * Free a list of sigio structures. The caller must ensure that new sigio
1104 * structures cannot be added after this point. For process groups this is
1105 * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1109 funsetownlst(struct sigiolst *sigiolst)
1113 struct sigio *sigio, *tmp;
1116 sigio = SLIST_FIRST(sigiolst);
1124 sigio = SLIST_FIRST(sigiolst);
1125 if (sigio == NULL) {
1131 * Every entry of the list should belong to a single proc or pgrp.
1133 if (sigio->sio_pgid < 0) {
1134 pg = sigio->sio_pgrp;
1135 sx_assert(&proctree_lock, SX_XLOCKED);
1137 } else /* if (sigio->sio_pgid > 0) */ {
1138 p = sigio->sio_proc;
1140 KASSERT((p->p_flag & P_WEXIT) != 0,
1141 ("%s: process %p is not exiting", __func__, p));
1144 SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1145 *sigio->sio_myref = NULL;
1147 KASSERT(sigio->sio_pgid < 0,
1148 ("Proc sigio in pgrp sigio list"));
1149 KASSERT(sigio->sio_pgrp == pg,
1150 ("Bogus pgrp in sigio list"));
1151 } else /* if (p != NULL) */ {
1152 KASSERT(sigio->sio_pgid > 0,
1153 ("Pgrp sigio in proc sigio list"));
1154 KASSERT(sigio->sio_proc == p,
1155 ("Bogus proc in sigio list"));
1165 SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1170 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1172 * After permission checking, add a sigio structure to the sigio list for
1173 * the process or process group.
1176 fsetown(pid_t pgid, struct sigio **sigiop)
1180 struct sigio *osigio, *sigio;
1188 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1189 sigio->sio_pgid = pgid;
1190 sigio->sio_ucred = crhold(curthread->td_ucred);
1191 sigio->sio_myref = sigiop;
1195 ret = pget(pgid, PGET_NOTWEXIT | PGET_NOTID | PGET_HOLD, &proc);
1197 osigio = funsetown_locked(*sigiop);
1201 if ((proc->p_flag & P_WEXIT) != 0) {
1203 } else if (proc->p_session !=
1204 curthread->td_proc->p_session) {
1206 * Policy - Don't allow a process to FSETOWN a
1207 * process in another session.
1209 * Remove this test to allow maximum flexibility
1210 * or restrict FSETOWN to the current process or
1211 * process group for maximum safety.
1215 sigio->sio_proc = proc;
1216 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio,
1221 } else /* if (pgid < 0) */ {
1222 sx_slock(&proctree_lock);
1224 osigio = funsetown_locked(*sigiop);
1225 pgrp = pgfind(-pgid);
1229 if (pgrp->pg_session != curthread->td_proc->p_session) {
1231 * Policy - Don't allow a process to FSETOWN a
1232 * process in another session.
1234 * Remove this test to allow maximum flexibility
1235 * or restrict FSETOWN to the current process or
1236 * process group for maximum safety.
1240 sigio->sio_pgrp = pgrp;
1241 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio,
1246 sx_sunlock(&proctree_lock);
1257 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1260 fgetown(struct sigio **sigiop)
1265 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1271 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1276 FILEDESC_XLOCK_ASSERT(fdp);
1279 * We now hold the fp reference that used to be owned by the
1280 * descriptor array. We have to unlock the FILEDESC *AFTER*
1281 * knote_fdclose to prevent a race of the fd getting opened, a knote
1282 * added, and deleteing a knote for the new fd.
1284 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1285 knote_fdclose(td, fd);
1288 * We need to notify mqueue if the object is of type mqueue.
1290 if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1291 mq_fdclose(td, fd, fp);
1292 FILEDESC_XUNLOCK(fdp);
1295 if (AUDITING_TD(td) && audit)
1296 audit_sysclose(td, fd, fp);
1298 error = closef(fp, td);
1301 * All paths leading up to closefp() will have already removed or
1302 * replaced the fd in the filedesc table, so a restart would not
1303 * operate on the same file.
1305 if (error == ERESTART)
1312 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1313 bool holdleaders, bool audit)
1317 FILEDESC_XLOCK_ASSERT(fdp);
1320 if (td->td_proc->p_fdtol != NULL) {
1322 * Ask fdfree() to sleep to ensure that all relevant
1323 * process leaders can be traversed in closef().
1325 fdp->fd_holdleaderscount++;
1327 holdleaders = false;
1331 error = closefp_impl(fdp, fd, fp, td, audit);
1333 FILEDESC_XLOCK(fdp);
1334 fdp->fd_holdleaderscount--;
1335 if (fdp->fd_holdleaderscount == 0 &&
1336 fdp->fd_holdleaderswakeup != 0) {
1337 fdp->fd_holdleaderswakeup = 0;
1338 wakeup(&fdp->fd_holdleaderscount);
1340 FILEDESC_XUNLOCK(fdp);
1346 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1347 bool holdleaders, bool audit)
1350 FILEDESC_XLOCK_ASSERT(fdp);
1352 if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1353 return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1355 return (closefp_impl(fdp, fd, fp, td, audit));
1360 * Close a file descriptor.
1362 #ifndef _SYS_SYSPROTO_H_
1369 sys_close(struct thread *td, struct close_args *uap)
1372 return (kern_close(td, uap->fd));
1376 kern_close(struct thread *td, int fd)
1378 struct filedesc *fdp;
1381 fdp = td->td_proc->p_fd;
1383 FILEDESC_XLOCK(fdp);
1384 if ((fp = fget_locked(fdp, fd)) == NULL) {
1385 FILEDESC_XUNLOCK(fdp);
1390 /* closefp() drops the FILEDESC lock for us. */
1391 return (closefp(fdp, fd, fp, td, true, true));
1395 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1397 struct filedesc *fdp;
1398 const struct fdescenttbl *fdt;
1403 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1404 * open should not be a usage error. From a close_range() perspective,
1405 * close_range(3, ~0U, 0) in the same scenario should also likely not
1406 * be a usage error as all fd above 3 are in-fact already closed.
1408 if (highfd < lowfd) {
1412 fdp = td->td_proc->p_fd;
1413 FILEDESC_XLOCK(fdp);
1414 fdt = atomic_load_ptr(&fdp->fd_files);
1415 highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1417 if (__predict_false(fd > highfd)) {
1421 fp = fdt->fdt_ofiles[fd].fde_file;
1427 (void) closefp(fdp, fd, fp, td, true, true);
1430 FILEDESC_XLOCK(fdp);
1431 fdt = atomic_load_ptr(&fdp->fd_files);
1436 FILEDESC_XUNLOCK(fdp);
1441 #ifndef _SYS_SYSPROTO_H_
1442 struct close_range_args {
1449 sys_close_range(struct thread *td, struct close_range_args *uap)
1452 AUDIT_ARG_FD(uap->lowfd);
1453 AUDIT_ARG_CMD(uap->highfd);
1454 AUDIT_ARG_FFLAGS(uap->flags);
1456 /* No flags currently defined */
1457 if (uap->flags != 0)
1459 return (kern_close_range(td, uap->lowfd, uap->highfd));
1462 #ifdef COMPAT_FREEBSD12
1464 * Close open file descriptors.
1466 #ifndef _SYS_SYSPROTO_H_
1467 struct freebsd12_closefrom_args {
1473 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1477 AUDIT_ARG_FD(uap->lowfd);
1480 * Treat negative starting file descriptor values identical to
1481 * closefrom(0) which closes all files.
1483 lowfd = MAX(0, uap->lowfd);
1484 return (kern_close_range(td, lowfd, ~0U));
1486 #endif /* COMPAT_FREEBSD12 */
1488 #if defined(COMPAT_43)
1490 * Return status information about a file descriptor.
1492 #ifndef _SYS_SYSPROTO_H_
1493 struct ofstat_args {
1500 ofstat(struct thread *td, struct ofstat_args *uap)
1506 error = kern_fstat(td, uap->fd, &ub);
1509 error = copyout(&oub, uap->sb, sizeof(oub));
1513 #endif /* COMPAT_43 */
1515 #if defined(COMPAT_FREEBSD11)
1517 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1520 struct freebsd11_stat osb;
1523 error = kern_fstat(td, uap->fd, &sb);
1526 error = freebsd11_cvtstat(&sb, &osb);
1528 error = copyout(&osb, uap->sb, sizeof(osb));
1531 #endif /* COMPAT_FREEBSD11 */
1534 * Return status information about a file descriptor.
1536 #ifndef _SYS_SYSPROTO_H_
1544 sys_fstat(struct thread *td, struct fstat_args *uap)
1549 error = kern_fstat(td, uap->fd, &ub);
1551 error = copyout(&ub, uap->sb, sizeof(ub));
1556 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1563 error = fget(td, fd, &cap_fstat_rights, &fp);
1564 if (__predict_false(error != 0))
1567 AUDIT_ARG_FILE(td->td_proc, fp);
1569 error = fo_stat(fp, sbp, td->td_ucred);
1571 #ifdef __STAT_TIME_T_EXT
1572 sbp->st_atim_ext = 0;
1573 sbp->st_mtim_ext = 0;
1574 sbp->st_ctim_ext = 0;
1575 sbp->st_btim_ext = 0;
1578 if (KTRPOINT(td, KTR_STRUCT))
1579 ktrstat_error(sbp, error);
1584 #if defined(COMPAT_FREEBSD11)
1586 * Return status information about a file descriptor.
1588 #ifndef _SYS_SYSPROTO_H_
1589 struct freebsd11_nfstat_args {
1596 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1602 error = kern_fstat(td, uap->fd, &ub);
1605 error = freebsd11_cvtnstat(&ub, &nub);
1607 error = copyout(&nub, uap->sb, sizeof(nub));
1610 #endif /* COMPAT_FREEBSD11 */
1613 * Return pathconf information about a file descriptor.
1615 #ifndef _SYS_SYSPROTO_H_
1616 struct fpathconf_args {
1623 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1628 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1630 td->td_retval[0] = value;
1635 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1641 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1645 if (name == _PC_ASYNC_IO) {
1646 *valuep = _POSIX_ASYNCHRONOUS_IO;
1651 vn_lock(vp, LK_SHARED | LK_RETRY);
1652 error = VOP_PATHCONF(vp, name, valuep);
1654 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1655 if (name != _PC_PIPE_BUF) {
1670 * Copy filecaps structure allocating memory for ioctls array if needed.
1672 * The last parameter indicates whether the fdtable is locked. If it is not and
1673 * ioctls are encountered, copying fails and the caller must lock the table.
1675 * Note that if the table was not locked, the caller has to check the relevant
1676 * sequence counter to determine whether the operation was successful.
1679 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1683 if (src->fc_ioctls != NULL && !locked)
1685 memcpy(dst, src, sizeof(*src));
1686 if (src->fc_ioctls == NULL)
1689 KASSERT(src->fc_nioctls > 0,
1690 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1692 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1693 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1694 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1699 filecaps_copy_prep(const struct filecaps *src)
1704 if (__predict_true(src->fc_ioctls == NULL))
1707 KASSERT(src->fc_nioctls > 0,
1708 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1710 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1711 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1716 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1722 if (__predict_true(src->fc_ioctls == NULL)) {
1723 MPASS(ioctls == NULL);
1727 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1728 dst->fc_ioctls = ioctls;
1729 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1733 * Move filecaps structure to the new place and clear the old place.
1736 filecaps_move(struct filecaps *src, struct filecaps *dst)
1740 bzero(src, sizeof(*src));
1744 * Fill the given filecaps structure with full rights.
1747 filecaps_fill(struct filecaps *fcaps)
1750 CAP_ALL(&fcaps->fc_rights);
1751 fcaps->fc_ioctls = NULL;
1752 fcaps->fc_nioctls = -1;
1753 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1757 * Free memory allocated within filecaps structure.
1760 filecaps_free(struct filecaps *fcaps)
1763 free(fcaps->fc_ioctls, M_FILECAPS);
1764 bzero(fcaps, sizeof(*fcaps));
1768 filecaps_free_prep(struct filecaps *fcaps)
1772 ioctls = fcaps->fc_ioctls;
1773 bzero(fcaps, sizeof(*fcaps));
1778 filecaps_free_finish(u_long *ioctls)
1781 free(ioctls, M_FILECAPS);
1785 * Validate the given filecaps structure.
1788 filecaps_validate(const struct filecaps *fcaps, const char *func)
1791 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1792 ("%s: invalid rights", func));
1793 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1794 ("%s: invalid fcntls", func));
1795 KASSERT(fcaps->fc_fcntls == 0 ||
1796 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1797 ("%s: fcntls without CAP_FCNTL", func));
1798 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1799 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1800 ("%s: invalid ioctls", func));
1801 KASSERT(fcaps->fc_nioctls == 0 ||
1802 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1803 ("%s: ioctls without CAP_IOCTL", func));
1807 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1811 FILEDESC_XLOCK_ASSERT(fdp);
1813 nfd1 = fdp->fd_nfiles * 2;
1816 fdgrowtable(fdp, nfd1);
1820 * Grow the file table to accommodate (at least) nfd descriptors.
1823 fdgrowtable(struct filedesc *fdp, int nfd)
1825 struct filedesc0 *fdp0;
1826 struct freetable *ft;
1827 struct fdescenttbl *ntable;
1828 struct fdescenttbl *otable;
1829 int nnfiles, onfiles;
1830 NDSLOTTYPE *nmap, *omap;
1832 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1834 /* save old values */
1835 onfiles = fdp->fd_nfiles;
1836 otable = fdp->fd_files;
1839 /* compute the size of the new table */
1840 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1841 if (nnfiles <= onfiles)
1842 /* the table is already large enough */
1846 * Allocate a new table. We need enough space for the number of
1847 * entries, file entries themselves and the struct freetable we will use
1848 * when we decommission the table and place it on the freelist.
1849 * We place the struct freetable in the middle so we don't have
1850 * to worry about padding.
1852 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1853 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1854 sizeof(struct freetable),
1855 M_FILEDESC, M_ZERO | M_WAITOK);
1856 /* copy the old data */
1857 ntable->fdt_nfiles = nnfiles;
1858 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1859 onfiles * sizeof(ntable->fdt_ofiles[0]));
1862 * Allocate a new map only if the old is not large enough. It will
1863 * grow at a slower rate than the table as it can map more
1864 * entries than the table can hold.
1866 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1867 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1869 /* copy over the old data and update the pointer */
1870 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1875 * Make sure that ntable is correctly initialized before we replace
1876 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1879 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1882 * Free the old file table when not shared by other threads or processes.
1883 * The old file table is considered to be shared when either are true:
1884 * - The process has more than one thread.
1885 * - The file descriptor table has been shared via fdshare().
1887 * When shared, the old file table will be placed on a freelist
1888 * which will be processed when the struct filedesc is released.
1890 * Note that if onfiles == NDFILE, we're dealing with the original
1891 * static allocation contained within (struct filedesc0 *)fdp,
1892 * which must not be freed.
1894 if (onfiles > NDFILE) {
1896 * Note we may be called here from fdinit while allocating a
1897 * table for a new process in which case ->p_fd points
1900 if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) {
1901 free(otable, M_FILEDESC);
1903 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1904 fdp0 = (struct filedesc0 *)fdp;
1905 ft->ft_table = otable;
1906 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1910 * The map does not have the same possibility of threads still
1911 * holding references to it. So always free it as long as it
1912 * does not reference the original static allocation.
1914 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1915 free(omap, M_FILEDESC);
1919 * Allocate a file descriptor for the process.
1922 fdalloc(struct thread *td, int minfd, int *result)
1924 struct proc *p = td->td_proc;
1925 struct filedesc *fdp = p->p_fd;
1926 int fd, maxfd, allocfd;
1931 FILEDESC_XLOCK_ASSERT(fdp);
1933 if (fdp->fd_freefile > minfd)
1934 minfd = fdp->fd_freefile;
1936 maxfd = getmaxfd(td);
1939 * Search the bitmap for a free descriptor starting at minfd.
1940 * If none is found, grow the file table.
1942 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1943 if (__predict_false(fd >= maxfd))
1945 if (__predict_false(fd >= fdp->fd_nfiles)) {
1946 allocfd = min(fd * 2, maxfd);
1948 if (RACCT_ENABLED()) {
1949 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1955 * fd is already equal to first free descriptor >= minfd, so
1956 * we only need to grow the table and we are done.
1958 fdgrowtable_exp(fdp, allocfd);
1962 * Perform some sanity checks, then mark the file descriptor as
1963 * used and return it to the caller.
1965 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1966 ("invalid descriptor %d", fd));
1967 KASSERT(!fdisused(fdp, fd),
1968 ("fd_first_free() returned non-free descriptor"));
1969 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1970 ("file descriptor isn't free"));
1977 * Allocate n file descriptors for the process.
1980 fdallocn(struct thread *td, int minfd, int *fds, int n)
1982 struct proc *p = td->td_proc;
1983 struct filedesc *fdp = p->p_fd;
1986 FILEDESC_XLOCK_ASSERT(fdp);
1988 for (i = 0; i < n; i++)
1989 if (fdalloc(td, 0, &fds[i]) != 0)
1993 for (i--; i >= 0; i--)
1994 fdunused(fdp, fds[i]);
2002 * Create a new open file structure and allocate a file descriptor for the
2003 * process that refers to it. We add one reference to the file for the
2004 * descriptor table and one reference for resultfp. This is to prevent us
2005 * being preempted and the entry in the descriptor table closed after we
2006 * release the FILEDESC lock.
2009 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
2010 struct filecaps *fcaps)
2015 MPASS(resultfp != NULL);
2016 MPASS(resultfd != NULL);
2018 error = _falloc_noinstall(td, &fp, 2);
2019 if (__predict_false(error != 0)) {
2023 error = finstall_refed(td, fp, &fd, flags, fcaps);
2024 if (__predict_false(error != 0)) {
2025 falloc_abort(td, fp);
2036 * Create a new open file structure without allocating a file descriptor.
2039 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2042 int maxuserfiles = maxfiles - (maxfiles / 20);
2044 static struct timeval lastfail;
2047 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2050 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2051 if ((openfiles_new >= maxuserfiles &&
2052 priv_check(td, PRIV_MAXFILES) != 0) ||
2053 openfiles_new >= maxfiles) {
2054 atomic_subtract_int(&openfiles, 1);
2055 if (ppsratecheck(&lastfail, &curfail, 1)) {
2056 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2057 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2061 fp = uma_zalloc(file_zone, M_WAITOK);
2062 bzero(fp, sizeof(*fp));
2063 refcount_init(&fp->f_count, n);
2064 fp->f_cred = crhold(td->td_ucred);
2065 fp->f_ops = &badfileops;
2071 falloc_abort(struct thread *td, struct file *fp)
2075 * For assertion purposes.
2077 refcount_init(&fp->f_count, 0);
2082 * Install a file in a file descriptor table.
2085 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2086 struct filecaps *fcaps)
2088 struct filedescent *fde;
2092 filecaps_validate(fcaps, __func__);
2093 FILEDESC_XLOCK_ASSERT(fdp);
2095 fde = &fdp->fd_ofiles[fd];
2097 seqc_write_begin(&fde->fde_seqc);
2100 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2102 filecaps_move(fcaps, &fde->fde_caps);
2104 filecaps_fill(&fde->fde_caps);
2106 seqc_write_end(&fde->fde_seqc);
2111 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2112 struct filecaps *fcaps)
2114 struct filedesc *fdp = td->td_proc->p_fd;
2119 FILEDESC_XLOCK(fdp);
2120 error = fdalloc(td, 0, fd);
2121 if (__predict_true(error == 0)) {
2122 _finstall(fdp, fp, *fd, flags, fcaps);
2124 FILEDESC_XUNLOCK(fdp);
2129 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2130 struct filecaps *fcaps)
2138 error = finstall_refed(td, fp, fd, flags, fcaps);
2139 if (__predict_false(error != 0)) {
2146 * Build a new filedesc structure from another.
2148 * If fdp is not NULL, return with it shared locked.
2151 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2153 struct filedesc0 *newfdp0;
2154 struct filedesc *newfdp;
2157 MPASS(lastfile != NULL);
2159 MPASS(lastfile == NULL);
2161 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2162 newfdp = &newfdp0->fd_fd;
2164 /* Create the file descriptor table. */
2165 FILEDESC_LOCK_INIT(newfdp);
2166 refcount_init(&newfdp->fd_refcnt, 1);
2167 refcount_init(&newfdp->fd_holdcnt, 1);
2168 newfdp->fd_map = newfdp0->fd_dmap;
2169 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2170 newfdp->fd_files->fdt_nfiles = NDFILE;
2175 FILEDESC_SLOCK(fdp);
2177 FILEDESC_SUNLOCK(fdp);
2182 *lastfile = fdlastfile(fdp);
2183 if (*lastfile < newfdp->fd_nfiles)
2185 FILEDESC_SUNLOCK(fdp);
2186 fdgrowtable(newfdp, *lastfile + 1);
2187 FILEDESC_SLOCK(fdp);
2194 * Build a pwddesc structure from another.
2195 * Copy the current, root, and jail root vnode references.
2197 * If pdp is not NULL, return with it shared locked.
2200 pdinit(struct pwddesc *pdp, bool keeplock)
2202 struct pwddesc *newpdp;
2205 newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2207 PWDDESC_LOCK_INIT(newpdp);
2208 refcount_init(&newpdp->pd_refcount, 1);
2209 newpdp->pd_cmask = CMASK;
2212 newpwd = pwd_alloc();
2213 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2218 newpwd = pwd_hold_pwddesc(pdp);
2219 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2221 PWDDESC_XUNLOCK(pdp);
2226 * Hold either filedesc or pwddesc of the passed process.
2228 * The process lock is used to synchronize against the target exiting and
2231 * Clearing can be ilustrated in 3 steps:
2232 * 1. set the pointer to NULL. Either routine can race against it, hence
2234 * 2. observe the process lock as not taken. Until then fdhold/pdhold can
2235 * race to either still see the pointer or find NULL. It is still safe to
2236 * grab a reference as clearing is stalled.
2237 * 3. after the lock is observed as not taken, any fdhold/pdhold calls are
2238 * guaranteed to see NULL, making it safe to finish clearing
2240 static struct filedesc *
2241 fdhold(struct proc *p)
2243 struct filedesc *fdp;
2245 PROC_LOCK_ASSERT(p, MA_OWNED);
2246 fdp = atomic_load_ptr(&p->p_fd);
2248 refcount_acquire(&fdp->fd_holdcnt);
2252 static struct pwddesc *
2253 pdhold(struct proc *p)
2255 struct pwddesc *pdp;
2257 PROC_LOCK_ASSERT(p, MA_OWNED);
2258 pdp = atomic_load_ptr(&p->p_pd);
2260 refcount_acquire(&pdp->pd_refcount);
2265 fddrop(struct filedesc *fdp)
2268 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2269 if (refcount_release(&fdp->fd_holdcnt) == 0)
2273 FILEDESC_LOCK_DESTROY(fdp);
2274 uma_zfree(filedesc0_zone, fdp);
2278 pddrop(struct pwddesc *pdp)
2282 if (refcount_release_if_not_last(&pdp->pd_refcount))
2286 if (refcount_release(&pdp->pd_refcount) == 0) {
2287 PWDDESC_XUNLOCK(pdp);
2290 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2292 PWDDESC_XUNLOCK(pdp);
2295 PWDDESC_LOCK_DESTROY(pdp);
2296 free(pdp, M_PWDDESC);
2300 * Share a filedesc structure.
2303 fdshare(struct filedesc *fdp)
2306 refcount_acquire(&fdp->fd_refcnt);
2311 * Share a pwddesc structure.
2314 pdshare(struct pwddesc *pdp)
2316 refcount_acquire(&pdp->pd_refcount);
2321 * Unshare a filedesc structure, if necessary by making a copy
2324 fdunshare(struct thread *td)
2326 struct filedesc *tmp;
2327 struct proc *p = td->td_proc;
2329 if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2332 tmp = fdcopy(p->p_fd);
2338 * Unshare a pwddesc structure.
2341 pdunshare(struct thread *td)
2343 struct pwddesc *pdp;
2348 if (p->p_pd->pd_refcount == 1)
2351 pdp = pdcopy(p->p_pd);
2357 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2358 * this is to ease callers, not catch errors.
2361 fdcopy(struct filedesc *fdp)
2363 struct filedesc *newfdp;
2364 struct filedescent *nfde, *ofde;
2369 newfdp = fdinit(fdp, true, &lastfile);
2370 /* copy all passable descriptors (i.e. not kqueue) */
2371 newfdp->fd_freefile = -1;
2372 for (i = 0; i <= lastfile; ++i) {
2373 ofde = &fdp->fd_ofiles[i];
2374 if (ofde->fde_file == NULL ||
2375 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2376 !fhold(ofde->fde_file)) {
2377 if (newfdp->fd_freefile == -1)
2378 newfdp->fd_freefile = i;
2381 nfde = &newfdp->fd_ofiles[i];
2383 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2384 fdused_init(newfdp, i);
2386 if (newfdp->fd_freefile == -1)
2387 newfdp->fd_freefile = i;
2388 FILEDESC_SUNLOCK(fdp);
2393 * Copy a pwddesc structure.
2396 pdcopy(struct pwddesc *pdp)
2398 struct pwddesc *newpdp;
2402 newpdp = pdinit(pdp, true);
2403 newpdp->pd_cmask = pdp->pd_cmask;
2404 PWDDESC_XUNLOCK(pdp);
2409 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2410 * one of processes using it exits) and the table used to be shared.
2413 fdclearlocks(struct thread *td)
2415 struct filedesc *fdp;
2416 struct filedesc_to_leader *fdtol;
2426 MPASS(fdtol != NULL);
2428 FILEDESC_XLOCK(fdp);
2429 KASSERT(fdtol->fdl_refcount > 0,
2430 ("filedesc_to_refcount botch: fdl_refcount=%d",
2431 fdtol->fdl_refcount));
2432 if (fdtol->fdl_refcount == 1 &&
2433 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2434 lastfile = fdlastfile(fdp);
2435 for (i = 0; i <= lastfile; i++) {
2436 fp = fdp->fd_ofiles[i].fde_file;
2437 if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2440 FILEDESC_XUNLOCK(fdp);
2441 lf.l_whence = SEEK_SET;
2444 lf.l_type = F_UNLCK;
2446 (void) VOP_ADVLOCK(vp,
2447 (caddr_t)p->p_leader, F_UNLCK,
2449 FILEDESC_XLOCK(fdp);
2454 if (fdtol->fdl_refcount == 1) {
2455 if (fdp->fd_holdleaderscount > 0 &&
2456 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2458 * close() or kern_dup() has cleared a reference
2459 * in a shared file descriptor table.
2461 fdp->fd_holdleaderswakeup = 1;
2462 sx_sleep(&fdp->fd_holdleaderscount,
2463 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2466 if (fdtol->fdl_holdcount > 0) {
2468 * Ensure that fdtol->fdl_leader remains
2469 * valid in closef().
2471 fdtol->fdl_wakeup = 1;
2472 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2477 fdtol->fdl_refcount--;
2478 if (fdtol->fdl_refcount == 0 &&
2479 fdtol->fdl_holdcount == 0) {
2480 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2481 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2485 FILEDESC_XUNLOCK(fdp);
2487 free(fdtol, M_FILEDESC_TO_LEADER);
2491 * Release a filedesc structure.
2494 fdescfree_fds(struct thread *td, struct filedesc *fdp)
2496 struct filedesc0 *fdp0;
2497 struct freetable *ft, *tft;
2498 struct filedescent *fde;
2502 KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2503 ("%s: fd table %p carries references", __func__, fdp));
2506 * Serialize with threads iterating over the table, if any.
2508 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2509 FILEDESC_XLOCK(fdp);
2510 FILEDESC_XUNLOCK(fdp);
2513 lastfile = fdlastfile_single(fdp);
2514 for (i = 0; i <= lastfile; i++) {
2515 fde = &fdp->fd_ofiles[i];
2519 (void) closef(fp, td);
2523 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2524 free(fdp->fd_map, M_FILEDESC);
2525 if (fdp->fd_nfiles > NDFILE)
2526 free(fdp->fd_files, M_FILEDESC);
2528 fdp0 = (struct filedesc0 *)fdp;
2529 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2530 free(ft->ft_table, M_FILEDESC);
2536 fdescfree(struct thread *td)
2539 struct filedesc *fdp;
2546 if (RACCT_ENABLED())
2547 racct_set_unlocked(p, RACCT_NOFILE, 0);
2550 if (p->p_fdtol != NULL)
2554 * Check fdhold for an explanation.
2556 atomic_store_ptr(&p->p_fd, NULL);
2557 atomic_thread_fence_seq_cst();
2558 PROC_WAIT_UNLOCKED(p);
2560 if (refcount_release(&fdp->fd_refcnt) == 0)
2563 fdescfree_fds(td, fdp);
2567 pdescfree(struct thread *td)
2570 struct pwddesc *pdp;
2577 * Check pdhold for an explanation.
2579 atomic_store_ptr(&p->p_pd, NULL);
2580 atomic_thread_fence_seq_cst();
2581 PROC_WAIT_UNLOCKED(p);
2587 * For setugid programs, we don't want to people to use that setugidness
2588 * to generate error messages which write to a file which otherwise would
2589 * otherwise be off-limits to the process. We check for filesystems where
2590 * the vnode can change out from under us after execve (like [lin]procfs).
2592 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2593 * sufficient. We also don't check for setugidness since we know we are.
2596 is_unsafe(struct file *fp)
2600 if (fp->f_type != DTYPE_VNODE)
2604 return ((vp->v_vflag & VV_PROCDEP) != 0);
2608 * Make this setguid thing safe, if at all possible.
2611 fdsetugidsafety(struct thread *td)
2613 struct filedesc *fdp;
2617 fdp = td->td_proc->p_fd;
2618 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2619 ("the fdtable should not be shared"));
2620 MPASS(fdp->fd_nfiles >= 3);
2621 for (i = 0; i <= 2; i++) {
2622 fp = fdp->fd_ofiles[i].fde_file;
2623 if (fp != NULL && is_unsafe(fp)) {
2624 FILEDESC_XLOCK(fdp);
2625 knote_fdclose(td, i);
2627 * NULL-out descriptor prior to close to avoid
2628 * a race while close blocks.
2631 FILEDESC_XUNLOCK(fdp);
2632 (void) closef(fp, td);
2638 * If a specific file object occupies a specific file descriptor, close the
2639 * file descriptor entry and drop a reference on the file object. This is a
2640 * convenience function to handle a subsequent error in a function that calls
2641 * falloc() that handles the race that another thread might have closed the
2642 * file descriptor out from under the thread creating the file object.
2645 fdclose(struct thread *td, struct file *fp, int idx)
2647 struct filedesc *fdp = td->td_proc->p_fd;
2649 FILEDESC_XLOCK(fdp);
2650 if (fdp->fd_ofiles[idx].fde_file == fp) {
2652 FILEDESC_XUNLOCK(fdp);
2655 FILEDESC_XUNLOCK(fdp);
2659 * Close any files on exec?
2662 fdcloseexec(struct thread *td)
2664 struct filedesc *fdp;
2665 struct filedescent *fde;
2669 fdp = td->td_proc->p_fd;
2670 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2671 ("the fdtable should not be shared"));
2672 lastfile = fdlastfile_single(fdp);
2673 for (i = 0; i <= lastfile; i++) {
2674 fde = &fdp->fd_ofiles[i];
2676 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2677 (fde->fde_flags & UF_EXCLOSE))) {
2678 FILEDESC_XLOCK(fdp);
2680 (void) closefp(fdp, i, fp, td, false, false);
2681 FILEDESC_UNLOCK_ASSERT(fdp);
2687 * It is unsafe for set[ug]id processes to be started with file
2688 * descriptors 0..2 closed, as these descriptors are given implicit
2689 * significance in the Standard C library. fdcheckstd() will create a
2690 * descriptor referencing /dev/null for each of stdin, stdout, and
2691 * stderr that is not already open.
2694 fdcheckstd(struct thread *td)
2696 struct filedesc *fdp;
2698 int i, error, devnull;
2700 fdp = td->td_proc->p_fd;
2701 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2702 ("the fdtable should not be shared"));
2703 MPASS(fdp->fd_nfiles >= 3);
2705 for (i = 0; i <= 2; i++) {
2706 if (fdp->fd_ofiles[i].fde_file != NULL)
2709 save = td->td_retval[0];
2710 if (devnull != -1) {
2711 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2713 error = kern_openat(td, AT_FDCWD, "/dev/null",
2714 UIO_SYSSPACE, O_RDWR, 0);
2716 devnull = td->td_retval[0];
2717 KASSERT(devnull == i, ("we didn't get our fd"));
2720 td->td_retval[0] = save;
2728 * Internal form of close. Decrement reference count on file structure.
2729 * Note: td may be NULL when closing a file that was being passed in a
2733 closef(struct file *fp, struct thread *td)
2737 struct filedesc_to_leader *fdtol;
2738 struct filedesc *fdp;
2743 * POSIX record locking dictates that any close releases ALL
2744 * locks owned by this process. This is handled by setting
2745 * a flag in the unlock to free ONLY locks obeying POSIX
2746 * semantics, and not to free BSD-style file locks.
2747 * If the descriptor was in a message, POSIX-style locks
2748 * aren't passed with the descriptor, and the thread pointer
2749 * will be NULL. Callers should be careful only to pass a
2750 * NULL thread pointer when there really is no owning
2751 * context that might have locks, or the locks will be
2754 if (fp->f_type == DTYPE_VNODE) {
2756 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2757 lf.l_whence = SEEK_SET;
2760 lf.l_type = F_UNLCK;
2761 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2762 F_UNLCK, &lf, F_POSIX);
2764 fdtol = td->td_proc->p_fdtol;
2765 if (fdtol != NULL) {
2767 * Handle special case where file descriptor table is
2768 * shared between multiple process leaders.
2770 fdp = td->td_proc->p_fd;
2771 FILEDESC_XLOCK(fdp);
2772 for (fdtol = fdtol->fdl_next;
2773 fdtol != td->td_proc->p_fdtol;
2774 fdtol = fdtol->fdl_next) {
2775 if ((fdtol->fdl_leader->p_flag &
2778 fdtol->fdl_holdcount++;
2779 FILEDESC_XUNLOCK(fdp);
2780 lf.l_whence = SEEK_SET;
2783 lf.l_type = F_UNLCK;
2785 (void) VOP_ADVLOCK(vp,
2786 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2788 FILEDESC_XLOCK(fdp);
2789 fdtol->fdl_holdcount--;
2790 if (fdtol->fdl_holdcount == 0 &&
2791 fdtol->fdl_wakeup != 0) {
2792 fdtol->fdl_wakeup = 0;
2796 FILEDESC_XUNLOCK(fdp);
2799 return (fdrop_close(fp, td));
2803 * Hack for file descriptor passing code.
2806 closef_nothread(struct file *fp)
2813 * Initialize the file pointer with the specified properties.
2815 * The ops are set with release semantics to be certain that the flags, type,
2816 * and data are visible when ops is. This is to prevent ops methods from being
2817 * called with bad data.
2820 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2825 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2829 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2831 fp->f_seqcount[UIO_READ] = 1;
2832 fp->f_seqcount[UIO_WRITE] = 1;
2833 finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2838 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2839 struct file **fpp, struct filecaps *havecapsp)
2841 struct filedescent *fde;
2844 FILEDESC_LOCK_ASSERT(fdp);
2846 fde = fdeget_locked(fdp, fd);
2853 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2858 if (havecapsp != NULL)
2859 filecaps_copy(&fde->fde_caps, havecapsp, true);
2861 *fpp = fde->fde_file;
2869 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2870 struct file **fpp, struct filecaps *havecapsp)
2872 struct filedesc *fdp = td->td_proc->p_fd;
2874 #ifndef CAPABILITIES
2875 error = fget_unlocked(fdp, fd, needrightsp, fpp);
2876 if (havecapsp != NULL && error == 0)
2877 filecaps_fill(havecapsp);
2884 error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2888 if (havecapsp != NULL) {
2889 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2890 havecapsp, false)) {
2896 if (!fd_modified(fdp, fd, seq))
2905 FILEDESC_SLOCK(fdp);
2906 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2907 if (error == 0 && !fhold(*fpp))
2909 FILEDESC_SUNLOCK(fdp);
2916 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2918 const struct filedescent *fde;
2919 const struct fdescenttbl *fdt;
2920 struct filedesc *fdp;
2923 const cap_rights_t *haverights;
2924 cap_rights_t rights;
2927 VFS_SMR_ASSERT_ENTERED();
2929 rights = *ndp->ni_rightsneeded;
2930 cap_rights_set_one(&rights, CAP_LOOKUP);
2932 fdp = curproc->p_fd;
2933 fdt = fdp->fd_files;
2934 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2936 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
2937 fde = &fdt->fdt_ofiles[fd];
2938 haverights = cap_rights_fde_inline(fde);
2940 if (__predict_false(fp == NULL))
2942 if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2944 *fsearch = ((fp->f_flag & FSEARCH) != 0);
2946 if (__predict_false(vp == NULL)) {
2949 if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2953 * Use an acquire barrier to force re-reading of fdt so it is
2954 * refreshed for verification.
2956 atomic_thread_fence_acq();
2957 fdt = fdp->fd_files;
2958 if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
2961 * If file descriptor doesn't have all rights,
2962 * all lookups relative to it must also be
2963 * strictly relative.
2965 * Not yet supported by fast path.
2968 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
2969 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
2970 ndp->ni_filecaps.fc_nioctls != -1) {
2972 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
2982 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2984 const struct fdescenttbl *fdt;
2985 struct filedesc *fdp;
2989 VFS_SMR_ASSERT_ENTERED();
2991 fdp = curproc->p_fd;
2992 fdt = fdp->fd_files;
2993 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2995 fp = fdt->fdt_ofiles[fd].fde_file;
2996 if (__predict_false(fp == NULL))
2998 *fsearch = ((fp->f_flag & FSEARCH) != 0);
3000 if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
3004 * Use an acquire barrier to force re-reading of fdt so it is
3005 * refreshed for verification.
3007 atomic_thread_fence_acq();
3008 fdt = fdp->fd_files;
3009 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3011 filecaps_fill(&ndp->ni_filecaps);
3018 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3019 struct file **fpp, seqc_t *seqp)
3022 const struct filedescent *fde;
3024 const struct fdescenttbl *fdt;
3028 cap_rights_t haverights;
3032 fdt = fdp->fd_files;
3033 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3036 * Fetch the descriptor locklessly. We avoid fdrop() races by
3037 * never raising a refcount above 0. To accomplish this we have
3038 * to use a cmpset loop rather than an atomic_add. The descriptor
3039 * must be re-verified once we acquire a reference to be certain
3040 * that the identity is still correct and we did not lose a race
3041 * due to preemption.
3045 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3046 fde = &fdt->fdt_ofiles[fd];
3047 haverights = *cap_rights_fde_inline(fde);
3049 if (!seqc_consistent(fd_seqc(fdt, fd), seq))
3052 fp = fdt->fdt_ofiles[fd].fde_file;
3057 error = cap_check_inline(&haverights, needrightsp);
3061 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3063 * Force a reload. Other thread could reallocate the
3064 * table before this fd was closed, so it is possible
3065 * that there is a stale fp pointer in cached version.
3067 fdt = atomic_load_ptr(&fdp->fd_files);
3071 * Use an acquire barrier to force re-reading of fdt so it is
3072 * refreshed for verification.
3074 atomic_thread_fence_acq();
3075 fdt = fdp->fd_files;
3077 if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
3079 if (fp == fdt->fdt_ofiles[fd].fde_file)
3082 fdrop(fp, curthread);
3094 * See the comments in fget_unlocked_seq for an explanation of how this works.
3096 * This is a simplified variant which bails out to the aforementioned routine
3097 * if anything goes wrong. In practice this only happens when userspace is
3098 * racing with itself.
3101 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3105 const struct filedescent *fde;
3107 const struct fdescenttbl *fdt;
3111 const cap_rights_t *haverights;
3114 fdt = fdp->fd_files;
3115 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3118 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3119 fde = &fdt->fdt_ofiles[fd];
3120 haverights = cap_rights_fde_inline(fde);
3123 fp = fdt->fdt_ofiles[fd].fde_file;
3125 if (__predict_false(fp == NULL))
3128 if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3131 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3135 * Use an acquire barrier to force re-reading of fdt so it is
3136 * refreshed for verification.
3138 atomic_thread_fence_acq();
3139 fdt = fdp->fd_files;
3141 if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3143 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3149 fdrop(fp, curthread);
3151 return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
3155 * Translate fd -> file when the caller guarantees the file descriptor table
3156 * can't be changed by others.
3158 * Note this does not mean the file object itself is only visible to the caller,
3159 * merely that it wont disappear without having to be referenced.
3161 * Must be paired with fput_only_user.
3165 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3168 const struct filedescent *fde;
3169 const struct fdescenttbl *fdt;
3170 const cap_rights_t *haverights;
3174 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3176 if (__predict_false(fd >= fdp->fd_nfiles))
3179 fdt = fdp->fd_files;
3180 fde = &fdt->fdt_ofiles[fd];
3182 if (__predict_false(fp == NULL))
3184 MPASS(refcount_load(&fp->f_count) > 0);
3185 haverights = cap_rights_fde_inline(fde);
3186 error = cap_check_inline(haverights, needrightsp);
3187 if (__predict_false(error != 0))
3194 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3199 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3201 if (__predict_false(fd >= fdp->fd_nfiles))
3204 fp = fdp->fd_ofiles[fd].fde_file;
3205 if (__predict_false(fp == NULL))
3208 MPASS(refcount_load(&fp->f_count) > 0);
3215 * Extract the file pointer associated with the specified descriptor for the
3216 * current user process.
3218 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3221 * File's rights will be checked against the capability rights mask.
3223 * If an error occurred the non-zero error is returned and *fpp is set to
3224 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
3225 * responsible for fdrop().
3228 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3229 cap_rights_t *needrightsp)
3231 struct filedesc *fdp;
3236 fdp = td->td_proc->p_fd;
3237 error = fget_unlocked(fdp, fd, needrightsp, &fp);
3238 if (__predict_false(error != 0))
3240 if (__predict_false(fp->f_ops == &badfileops)) {
3246 * FREAD and FWRITE failure return EBADF as per POSIX.
3252 if ((fp->f_flag & flags) == 0)
3256 if (fp->f_ops != &path_fileops &&
3257 ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3258 (fp->f_flag & FWRITE) != 0))
3264 KASSERT(0, ("wrong flags"));
3277 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3280 return (_fget(td, fd, fpp, 0, rightsp));
3284 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3288 #ifndef CAPABILITIES
3289 error = _fget(td, fd, fpp, 0, rightsp);
3290 if (maxprotp != NULL)
3291 *maxprotp = VM_PROT_ALL;
3294 cap_rights_t fdrights;
3295 struct filedesc *fdp;
3300 fdp = td->td_proc->p_fd;
3301 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3303 error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3304 if (__predict_false(error != 0))
3306 if (__predict_false(fp->f_ops == &badfileops)) {
3310 if (maxprotp != NULL)
3311 fdrights = *cap_rights(fdp, fd);
3312 if (!fd_modified(fdp, fd, seq))
3318 * If requested, convert capability rights to access flags.
3320 if (maxprotp != NULL)
3321 *maxprotp = cap_rights_to_vmprot(&fdrights);
3328 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3331 return (_fget(td, fd, fpp, FREAD, rightsp));
3335 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3338 return (_fget(td, fd, fpp, FWRITE, rightsp));
3342 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3345 struct filedesc *fdp = td->td_proc->p_fd;
3346 #ifndef CAPABILITIES
3347 return (fget_unlocked(fdp, fd, rightsp, fpp));
3354 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3356 error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3359 error = cap_fcntl_check(fdp, fd, needfcntl);
3360 if (!fd_modified(fdp, fd, seq))
3374 * Like fget() but loads the underlying vnode, or returns an error if the
3375 * descriptor does not represent a vnode. Note that pipes use vnodes but
3376 * never have VM objects. The returned vnode will be vref()'d.
3378 * XXX: what about the unused flags ?
3381 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3388 error = _fget(td, fd, &fp, flags, needrightsp);
3391 if (fp->f_vnode == NULL) {
3403 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3406 return (_fgetvp(td, fd, 0, rightsp, vpp));
3410 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3411 struct filecaps *havecaps, struct vnode **vpp)
3413 struct filecaps caps;
3417 error = fget_cap(td, fd, needrightsp, &fp, &caps);
3420 if (fp->f_ops == &badfileops) {
3424 if (fp->f_vnode == NULL) {
3436 filecaps_free(&caps);
3442 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3445 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3449 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3452 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3457 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3461 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3466 * Handle the last reference to a file being closed.
3468 * Without the noinline attribute clang keeps inlining the func thorough this
3469 * file when fdrop is used.
3472 _fdrop(struct file *fp, struct thread *td)
3478 count = refcount_load(&fp->f_count);
3480 panic("fdrop: fp %p count %d", fp, count);
3482 error = fo_close(fp, td);
3483 atomic_subtract_int(&openfiles, 1);
3485 free(fp->f_advice, M_FADVISE);
3486 uma_zfree(file_zone, fp);
3492 * Apply an advisory lock on a file descriptor.
3494 * Just attempt to get a record lock of the requested type on the entire file
3495 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3497 #ifndef _SYS_SYSPROTO_H_
3505 sys_flock(struct thread *td, struct flock_args *uap)
3512 error = fget(td, uap->fd, &cap_flock_rights, &fp);
3516 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
3519 if (fp->f_ops == &path_fileops) {
3525 lf.l_whence = SEEK_SET;
3528 if (uap->how & LOCK_UN) {
3529 lf.l_type = F_UNLCK;
3530 atomic_clear_int(&fp->f_flag, FHASLOCK);
3531 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3534 if (uap->how & LOCK_EX)
3535 lf.l_type = F_WRLCK;
3536 else if (uap->how & LOCK_SH)
3537 lf.l_type = F_RDLCK;
3542 atomic_set_int(&fp->f_flag, FHASLOCK);
3543 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3544 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3550 * Duplicate the specified descriptor to a free descriptor.
3553 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3554 int openerror, int *indxp)
3556 struct filedescent *newfde, *oldfde;
3561 KASSERT(openerror == ENODEV || openerror == ENXIO,
3562 ("unexpected error %d in %s", openerror, __func__));
3565 * If the to-be-dup'd fd number is greater than the allowed number
3566 * of file descriptors, or the fd to be dup'd has already been
3567 * closed, then reject.
3569 FILEDESC_XLOCK(fdp);
3570 if ((fp = fget_locked(fdp, dfd)) == NULL) {
3571 FILEDESC_XUNLOCK(fdp);
3575 error = fdalloc(td, 0, &indx);
3577 FILEDESC_XUNLOCK(fdp);
3582 * There are two cases of interest here.
3584 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3586 * For ENXIO steal away the file structure from (dfd) and store it in
3587 * (indx). (dfd) is effectively closed by this operation.
3589 switch (openerror) {
3592 * Check that the mode the file is being opened for is a
3593 * subset of the mode of the existing descriptor.
3595 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3596 fdunused(fdp, indx);
3597 FILEDESC_XUNLOCK(fdp);
3601 fdunused(fdp, indx);
3602 FILEDESC_XUNLOCK(fdp);
3605 newfde = &fdp->fd_ofiles[indx];
3606 oldfde = &fdp->fd_ofiles[dfd];
3607 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3609 seqc_write_begin(&newfde->fde_seqc);
3611 memcpy(newfde, oldfde, fde_change_size);
3612 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3615 seqc_write_end(&newfde->fde_seqc);
3620 * Steal away the file pointer from dfd and stuff it into indx.
3622 newfde = &fdp->fd_ofiles[indx];
3623 oldfde = &fdp->fd_ofiles[dfd];
3625 seqc_write_begin(&newfde->fde_seqc);
3627 memcpy(newfde, oldfde, fde_change_size);
3628 oldfde->fde_file = NULL;
3631 seqc_write_end(&newfde->fde_seqc);
3635 FILEDESC_XUNLOCK(fdp);
3641 * This sysctl determines if we will allow a process to chroot(2) if it
3642 * has a directory open:
3643 * 0: disallowed for all processes.
3644 * 1: allowed for processes that were not already chroot(2)'ed.
3645 * 2: allowed for all processes.
3648 static int chroot_allow_open_directories = 1;
3650 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3651 &chroot_allow_open_directories, 0,
3652 "Allow a process to chroot(2) if it has a directory open");
3655 * Helper function for raised chroot(2) security function: Refuse if
3656 * any filedescriptors are open directories.
3659 chroot_refuse_vdir_fds(struct filedesc *fdp)
3665 FILEDESC_LOCK_ASSERT(fdp);
3667 lastfile = fdlastfile(fdp);
3668 for (fd = 0; fd <= lastfile; fd++) {
3669 fp = fget_locked(fdp, fd);
3672 if (fp->f_type == DTYPE_VNODE) {
3674 if (vp->v_type == VDIR)
3682 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3685 if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3686 vrefact(oldpwd->pwd_cdir);
3687 newpwd->pwd_cdir = oldpwd->pwd_cdir;
3690 if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3691 vrefact(oldpwd->pwd_rdir);
3692 newpwd->pwd_rdir = oldpwd->pwd_rdir;
3695 if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3696 vrefact(oldpwd->pwd_jdir);
3697 newpwd->pwd_jdir = oldpwd->pwd_jdir;
3702 pwd_hold_pwddesc(struct pwddesc *pdp)
3706 PWDDESC_ASSERT_XLOCKED(pdp);
3707 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3709 refcount_acquire(&pwd->pwd_refcount);
3714 pwd_hold_smr(struct pwd *pwd)
3718 if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3725 pwd_hold(struct thread *td)
3727 struct pwddesc *pdp;
3730 pdp = td->td_proc->p_pd;
3733 pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3734 if (pwd_hold_smr(pwd)) {
3740 pwd = pwd_hold_pwddesc(pdp);
3742 PWDDESC_XUNLOCK(pdp);
3747 pwd_hold_proc(struct proc *p)
3749 struct pwddesc *pdp;
3752 PROC_ASSERT_HELD(p);
3759 pwd = pwd_hold_pwddesc(pdp);
3761 PWDDESC_XUNLOCK(pdp);
3771 pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3772 bzero(pwd, sizeof(*pwd));
3773 refcount_init(&pwd->pwd_refcount, 1);
3778 pwd_drop(struct pwd *pwd)
3781 if (!refcount_release(&pwd->pwd_refcount))
3784 if (pwd->pwd_cdir != NULL)
3785 vrele(pwd->pwd_cdir);
3786 if (pwd->pwd_rdir != NULL)
3787 vrele(pwd->pwd_rdir);
3788 if (pwd->pwd_jdir != NULL)
3789 vrele(pwd->pwd_jdir);
3790 uma_zfree_smr(pwd_zone, pwd);
3794 * The caller is responsible for invoking priv_check() and
3795 * mac_vnode_check_chroot() to authorize this operation.
3798 pwd_chroot(struct thread *td, struct vnode *vp)
3800 struct pwddesc *pdp;
3801 struct filedesc *fdp;
3802 struct pwd *newpwd, *oldpwd;
3805 fdp = td->td_proc->p_fd;
3806 pdp = td->td_proc->p_pd;
3807 newpwd = pwd_alloc();
3808 FILEDESC_SLOCK(fdp);
3810 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3811 if (chroot_allow_open_directories == 0 ||
3812 (chroot_allow_open_directories == 1 &&
3813 oldpwd->pwd_rdir != rootvnode)) {
3814 error = chroot_refuse_vdir_fds(fdp);
3815 FILEDESC_SUNLOCK(fdp);
3817 PWDDESC_XUNLOCK(pdp);
3822 FILEDESC_SUNLOCK(fdp);
3826 newpwd->pwd_rdir = vp;
3827 if (oldpwd->pwd_jdir == NULL) {
3829 newpwd->pwd_jdir = vp;
3831 pwd_fill(oldpwd, newpwd);
3832 pwd_set(pdp, newpwd);
3833 PWDDESC_XUNLOCK(pdp);
3839 pwd_chdir(struct thread *td, struct vnode *vp)
3841 struct pwddesc *pdp;
3842 struct pwd *newpwd, *oldpwd;
3844 VNPASS(vp->v_usecount > 0, vp);
3846 newpwd = pwd_alloc();
3847 pdp = td->td_proc->p_pd;
3849 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3850 newpwd->pwd_cdir = vp;
3851 pwd_fill(oldpwd, newpwd);
3852 pwd_set(pdp, newpwd);
3853 PWDDESC_XUNLOCK(pdp);
3858 * jail_attach(2) changes both root and working directories.
3861 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
3863 struct pwddesc *pdp;
3864 struct filedesc *fdp;
3865 struct pwd *newpwd, *oldpwd;
3868 fdp = td->td_proc->p_fd;
3869 pdp = td->td_proc->p_pd;
3870 newpwd = pwd_alloc();
3871 FILEDESC_SLOCK(fdp);
3873 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3874 error = chroot_refuse_vdir_fds(fdp);
3875 FILEDESC_SUNLOCK(fdp);
3877 PWDDESC_XUNLOCK(pdp);
3883 newpwd->pwd_rdir = vp;
3885 newpwd->pwd_cdir = vp;
3886 if (oldpwd->pwd_jdir == NULL) {
3888 newpwd->pwd_jdir = vp;
3890 pwd_fill(oldpwd, newpwd);
3891 pwd_set(pdp, newpwd);
3892 PWDDESC_XUNLOCK(pdp);
3898 pwd_ensure_dirs(void)
3900 struct pwddesc *pdp;
3901 struct pwd *oldpwd, *newpwd;
3903 pdp = curproc->p_pd;
3905 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3906 if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3907 PWDDESC_XUNLOCK(pdp);
3910 PWDDESC_XUNLOCK(pdp);
3912 newpwd = pwd_alloc();
3914 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3915 pwd_fill(oldpwd, newpwd);
3916 if (newpwd->pwd_cdir == NULL) {
3918 newpwd->pwd_cdir = rootvnode;
3920 if (newpwd->pwd_rdir == NULL) {
3922 newpwd->pwd_rdir = rootvnode;
3924 pwd_set(pdp, newpwd);
3925 PWDDESC_XUNLOCK(pdp);
3930 pwd_set_rootvnode(void)
3932 struct pwddesc *pdp;
3933 struct pwd *oldpwd, *newpwd;
3935 pdp = curproc->p_pd;
3937 newpwd = pwd_alloc();
3939 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3941 newpwd->pwd_cdir = rootvnode;
3943 newpwd->pwd_rdir = rootvnode;
3944 pwd_fill(oldpwd, newpwd);
3945 pwd_set(pdp, newpwd);
3946 PWDDESC_XUNLOCK(pdp);
3951 * Scan all active processes and prisons to see if any of them have a current
3952 * or root directory of `olddp'. If so, replace them with the new mount point.
3955 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3957 struct pwddesc *pdp;
3958 struct pwd *newpwd, *oldpwd;
3963 if (vrefcnt(olddp) == 1)
3966 newpwd = pwd_alloc();
3967 sx_slock(&allproc_lock);
3968 FOREACH_PROC_IN_SYSTEM(p) {
3975 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3976 if (oldpwd == NULL ||
3977 (oldpwd->pwd_cdir != olddp &&
3978 oldpwd->pwd_rdir != olddp &&
3979 oldpwd->pwd_jdir != olddp)) {
3980 PWDDESC_XUNLOCK(pdp);
3984 if (oldpwd->pwd_cdir == olddp) {
3986 newpwd->pwd_cdir = newdp;
3988 if (oldpwd->pwd_rdir == olddp) {
3990 newpwd->pwd_rdir = newdp;
3992 if (oldpwd->pwd_jdir == olddp) {
3994 newpwd->pwd_jdir = newdp;
3996 pwd_fill(oldpwd, newpwd);
3997 pwd_set(pdp, newpwd);
3998 PWDDESC_XUNLOCK(pdp);
4001 newpwd = pwd_alloc();
4003 sx_sunlock(&allproc_lock);
4005 if (rootvnode == olddp) {
4010 mtx_lock(&prison0.pr_mtx);
4011 if (prison0.pr_root == olddp) {
4013 prison0.pr_root = newdp;
4016 mtx_unlock(&prison0.pr_mtx);
4017 sx_slock(&allprison_lock);
4018 TAILQ_FOREACH(pr, &allprison, pr_list) {
4019 mtx_lock(&pr->pr_mtx);
4020 if (pr->pr_root == olddp) {
4022 pr->pr_root = newdp;
4025 mtx_unlock(&pr->pr_mtx);
4027 sx_sunlock(&allprison_lock);
4032 struct filedesc_to_leader *
4033 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
4035 struct filedesc_to_leader *fdtol;
4037 fdtol = malloc(sizeof(struct filedesc_to_leader),
4038 M_FILEDESC_TO_LEADER, M_WAITOK);
4039 fdtol->fdl_refcount = 1;
4040 fdtol->fdl_holdcount = 0;
4041 fdtol->fdl_wakeup = 0;
4042 fdtol->fdl_leader = leader;
4044 FILEDESC_XLOCK(fdp);
4045 fdtol->fdl_next = old->fdl_next;
4046 fdtol->fdl_prev = old;
4047 old->fdl_next = fdtol;
4048 fdtol->fdl_next->fdl_prev = fdtol;
4049 FILEDESC_XUNLOCK(fdp);
4051 fdtol->fdl_next = fdtol;
4052 fdtol->fdl_prev = fdtol;
4058 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
4061 struct filedesc *fdp;
4063 int count, off, minoff;
4069 if (*(int *)arg1 != 0)
4072 fdp = curproc->p_fd;
4074 FILEDESC_SLOCK(fdp);
4076 off = NDSLOT(fdp->fd_nfiles - 1);
4077 for (minoff = NDSLOT(0); off >= minoff; --off)
4078 count += bitcountl(map[off]);
4079 FILEDESC_SUNLOCK(fdp);
4081 return (SYSCTL_OUT(req, &count, sizeof(count)));
4084 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
4085 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
4086 "Number of open file descriptors");
4089 * Get file structures globally.
4092 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
4095 struct filedesc *fdp;
4098 int error, n, lastfile;
4100 error = sysctl_wire_old_buffer(req, 0);
4103 if (req->oldptr == NULL) {
4105 sx_slock(&allproc_lock);
4106 FOREACH_PROC_IN_SYSTEM(p) {
4108 if (p->p_state == PRS_NEW) {
4116 /* overestimates sparse tables. */
4117 n += fdp->fd_nfiles;
4120 sx_sunlock(&allproc_lock);
4121 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4124 bzero(&xf, sizeof(xf));
4125 xf.xf_size = sizeof(xf);
4126 sx_slock(&allproc_lock);
4127 FOREACH_PROC_IN_SYSTEM(p) {
4129 if (p->p_state == PRS_NEW) {
4133 if (p_cansee(req->td, p) != 0) {
4137 xf.xf_pid = p->p_pid;
4138 xf.xf_uid = p->p_ucred->cr_uid;
4143 FILEDESC_SLOCK(fdp);
4144 lastfile = fdlastfile(fdp);
4145 for (n = 0; refcount_load(&fdp->fd_refcnt) > 0 && n <= lastfile;
4147 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4150 xf.xf_file = (uintptr_t)fp;
4151 xf.xf_data = (uintptr_t)fp->f_data;
4152 xf.xf_vnode = (uintptr_t)fp->f_vnode;
4153 xf.xf_type = (uintptr_t)fp->f_type;
4154 xf.xf_count = refcount_load(&fp->f_count);
4156 xf.xf_offset = foffset_get(fp);
4157 xf.xf_flag = fp->f_flag;
4158 error = SYSCTL_OUT(req, &xf, sizeof(xf));
4162 FILEDESC_SUNLOCK(fdp);
4167 sx_sunlock(&allproc_lock);
4171 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4172 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4174 #ifdef KINFO_FILE_SIZE
4175 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4179 xlate_fflags(int fflags)
4181 static const struct {
4184 } fflags_table[] = {
4185 { FAPPEND, KF_FLAG_APPEND },
4186 { FASYNC, KF_FLAG_ASYNC },
4187 { FFSYNC, KF_FLAG_FSYNC },
4188 { FHASLOCK, KF_FLAG_HASLOCK },
4189 { FNONBLOCK, KF_FLAG_NONBLOCK },
4190 { FREAD, KF_FLAG_READ },
4191 { FWRITE, KF_FLAG_WRITE },
4192 { O_CREAT, KF_FLAG_CREAT },
4193 { O_DIRECT, KF_FLAG_DIRECT },
4194 { O_EXCL, KF_FLAG_EXCL },
4195 { O_EXEC, KF_FLAG_EXEC },
4196 { O_EXLOCK, KF_FLAG_EXLOCK },
4197 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4198 { O_SHLOCK, KF_FLAG_SHLOCK },
4199 { O_TRUNC, KF_FLAG_TRUNC }
4205 for (i = 0; i < nitems(fflags_table); i++)
4206 if (fflags & fflags_table[i].fflag)
4207 kflags |= fflags_table[i].kf_fflag;
4211 /* Trim unused data from kf_path by truncating the structure size. */
4213 pack_kinfo(struct kinfo_file *kif)
4216 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4217 strlen(kif->kf_path) + 1;
4218 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4222 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4223 struct kinfo_file *kif, struct filedesc *fdp, int flags)
4227 bzero(kif, sizeof(*kif));
4229 /* Set a default type to allow for empty fill_kinfo() methods. */
4230 kif->kf_type = KF_TYPE_UNKNOWN;
4231 kif->kf_flags = xlate_fflags(fp->f_flag);
4232 if (rightsp != NULL)
4233 kif->kf_cap_rights = *rightsp;
4235 cap_rights_init_zero(&kif->kf_cap_rights);
4237 kif->kf_ref_count = refcount_load(&fp->f_count);
4238 kif->kf_offset = foffset_get(fp);
4241 * This may drop the filedesc lock, so the 'fp' cannot be
4242 * accessed after this call.
4244 error = fo_fill_kinfo(fp, kif, fdp);
4246 kif->kf_status |= KF_ATTR_VALID;
4247 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4250 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4254 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4255 struct kinfo_file *kif, int flags)
4259 bzero(kif, sizeof(*kif));
4261 kif->kf_type = KF_TYPE_VNODE;
4262 error = vn_fill_kinfo_vnode(vp, kif);
4264 kif->kf_status |= KF_ATTR_VALID;
4265 kif->kf_flags = xlate_fflags(fflags);
4266 cap_rights_init_zero(&kif->kf_cap_rights);
4268 kif->kf_ref_count = -1;
4269 kif->kf_offset = -1;
4270 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4273 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4277 struct export_fd_buf {
4278 struct filedesc *fdp;
4279 struct pwddesc *pdp;
4282 struct kinfo_file kif;
4287 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4289 struct kinfo_file *kif;
4292 if (efbuf->remainder != -1) {
4293 if (efbuf->remainder < kif->kf_structsize)
4295 efbuf->remainder -= kif->kf_structsize;
4297 if (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) != 0)
4298 return (sbuf_error(efbuf->sb));
4303 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4304 struct export_fd_buf *efbuf)
4308 if (efbuf->remainder == 0)
4310 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4312 FILEDESC_SUNLOCK(efbuf->fdp);
4313 error = export_kinfo_to_sb(efbuf);
4314 FILEDESC_SLOCK(efbuf->fdp);
4319 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4320 struct export_fd_buf *efbuf)
4324 if (efbuf->remainder == 0)
4326 if (efbuf->pdp != NULL)
4327 PWDDESC_XUNLOCK(efbuf->pdp);
4328 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4329 error = export_kinfo_to_sb(efbuf);
4330 if (efbuf->pdp != NULL)
4331 PWDDESC_XLOCK(efbuf->pdp);
4336 * Store a process file descriptor information to sbuf.
4338 * Takes a locked proc as argument, and returns with the proc unlocked.
4341 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
4345 struct filedesc *fdp;
4346 struct pwddesc *pdp;
4347 struct export_fd_buf *efbuf;
4348 struct vnode *cttyvp, *textvp, *tracevp;
4350 int error, i, lastfile;
4351 cap_rights_t rights;
4353 PROC_LOCK_ASSERT(p, MA_OWNED);
4356 tracevp = ktr_get_tracevp(p, true);
4358 textvp = p->p_textvp;
4361 /* Controlling tty. */
4363 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4364 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4372 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4376 efbuf->remainder = maxlen;
4377 efbuf->flags = flags;
4380 if (tracevp != NULL)
4381 error = export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE,
4382 FREAD | FWRITE, efbuf);
4383 if (error == 0 && textvp != NULL)
4384 error = export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD,
4386 if (error == 0 && cttyvp != NULL)
4387 error = export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY,
4388 FREAD | FWRITE, efbuf);
4389 if (error != 0 || pdp == NULL || fdp == NULL)
4394 pwd = pwd_hold_pwddesc(pdp);
4396 /* working directory */
4397 if (pwd->pwd_cdir != NULL) {
4398 vrefact(pwd->pwd_cdir);
4399 error = export_vnode_to_sb(pwd->pwd_cdir,
4400 KF_FD_TYPE_CWD, FREAD, efbuf);
4402 /* root directory */
4403 if (error == 0 && pwd->pwd_rdir != NULL) {
4404 vrefact(pwd->pwd_rdir);
4405 error = export_vnode_to_sb(pwd->pwd_rdir,
4406 KF_FD_TYPE_ROOT, FREAD, efbuf);
4408 /* jail directory */
4409 if (error == 0 && pwd->pwd_jdir != NULL) {
4410 vrefact(pwd->pwd_jdir);
4411 error = export_vnode_to_sb(pwd->pwd_jdir,
4412 KF_FD_TYPE_JAIL, FREAD, efbuf);
4415 PWDDESC_XUNLOCK(pdp);
4420 FILEDESC_SLOCK(fdp);
4421 lastfile = fdlastfile(fdp);
4422 for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4423 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4426 rights = *cap_rights(fdp, i);
4427 #else /* !CAPABILITIES */
4428 rights = cap_no_rights;
4431 * Create sysctl entry. It is OK to drop the filedesc
4432 * lock inside of export_file_to_sb() as we will
4433 * re-validate and re-evaluate its properties when the
4436 error = export_file_to_sb(fp, i, &rights, efbuf);
4440 FILEDESC_SUNLOCK(fdp);
4446 free(efbuf, M_TEMP);
4450 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
4453 * Get per-process file descriptors for use by procstat(1), et al.
4456 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4462 int error, error2, *name;
4470 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4471 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4472 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4477 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4478 error = kern_proc_filedesc_out(p, &sb, maxlen,
4479 KERN_FILEDESC_PACK_KINFO);
4480 error2 = sbuf_finish(&sb);
4482 return (error != 0 ? error : error2);
4485 #ifdef COMPAT_FREEBSD7
4486 #ifdef KINFO_OFILE_SIZE
4487 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4491 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4494 okif->kf_structsize = sizeof(*okif);
4495 okif->kf_type = kif->kf_type;
4496 okif->kf_fd = kif->kf_fd;
4497 okif->kf_ref_count = kif->kf_ref_count;
4498 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4499 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4500 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4501 okif->kf_offset = kif->kf_offset;
4502 if (kif->kf_type == KF_TYPE_VNODE)
4503 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4505 okif->kf_vnode_type = KF_VTYPE_VNON;
4506 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4507 if (kif->kf_type == KF_TYPE_SOCKET) {
4508 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4509 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4510 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4511 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4512 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4514 okif->kf_sa_local.ss_family = AF_UNSPEC;
4515 okif->kf_sa_peer.ss_family = AF_UNSPEC;
4520 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4521 struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4526 PWDDESC_XUNLOCK(pdp);
4527 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4528 kinfo_to_okinfo(kif, okif);
4529 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4535 * Get per-process file descriptors for use by procstat(1), et al.
4538 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4540 struct kinfo_ofile *okif;
4541 struct kinfo_file *kif;
4542 struct filedesc *fdp;
4543 struct pwddesc *pdp;
4546 int error, i, lastfile, *name;
4555 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4562 if (fdp == NULL || pdp == NULL) {
4567 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4568 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4570 pwd = pwd_hold_pwddesc(pdp);
4572 if (pwd->pwd_cdir != NULL)
4573 export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4575 if (pwd->pwd_rdir != NULL)
4576 export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4578 if (pwd->pwd_jdir != NULL)
4579 export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4582 PWDDESC_XUNLOCK(pdp);
4585 FILEDESC_SLOCK(fdp);
4586 lastfile = fdlastfile(fdp);
4587 for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4588 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4590 export_file_to_kinfo(fp, i, NULL, kif, fdp,
4591 KERN_FILEDESC_PACK_KINFO);
4592 FILEDESC_SUNLOCK(fdp);
4593 kinfo_to_okinfo(kif, okif);
4594 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4595 FILEDESC_SLOCK(fdp);
4599 FILEDESC_SUNLOCK(fdp);
4607 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4608 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4609 "Process ofiledesc entries");
4610 #endif /* COMPAT_FREEBSD7 */
4613 vntype_to_kinfo(int vtype)
4618 } vtypes_table[] = {
4619 { VBAD, KF_VTYPE_VBAD },
4620 { VBLK, KF_VTYPE_VBLK },
4621 { VCHR, KF_VTYPE_VCHR },
4622 { VDIR, KF_VTYPE_VDIR },
4623 { VFIFO, KF_VTYPE_VFIFO },
4624 { VLNK, KF_VTYPE_VLNK },
4625 { VNON, KF_VTYPE_VNON },
4626 { VREG, KF_VTYPE_VREG },
4627 { VSOCK, KF_VTYPE_VSOCK }
4632 * Perform vtype translation.
4634 for (i = 0; i < nitems(vtypes_table); i++)
4635 if (vtypes_table[i].vtype == vtype)
4636 return (vtypes_table[i].kf_vtype);
4638 return (KF_VTYPE_UNKNOWN);
4641 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4642 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4643 "Process filedesc entries");
4646 * Store a process current working directory information to sbuf.
4648 * Takes a locked proc as argument, and returns with the proc unlocked.
4651 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
4653 struct pwddesc *pdp;
4655 struct export_fd_buf *efbuf;
4659 PROC_LOCK_ASSERT(p, MA_OWNED);
4666 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4670 efbuf->remainder = maxlen;
4674 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4675 cdir = pwd->pwd_cdir;
4680 error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4682 PWDDESC_XUNLOCK(pdp);
4684 free(efbuf, M_TEMP);
4689 * Get per-process current working directory.
4692 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4698 int error, error2, *name;
4706 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4707 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4708 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4713 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4714 error = kern_proc_cwd_out(p, &sb, maxlen);
4715 error2 = sbuf_finish(&sb);
4717 return (error != 0 ? error : error2);
4720 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4721 sysctl_kern_proc_cwd, "Process current working directory");
4725 * For the purposes of debugging, generate a human-readable string for the
4729 file_type_to_name(short type)
4757 case DTYPE_PROCDESC:
4761 case DTYPE_LINUXTFD:
4769 * For the purposes of debugging, identify a process (if any, perhaps one of
4770 * many) that references the passed file in its file descriptor array. Return
4773 static struct proc *
4774 file_to_first_proc(struct file *fp)
4776 struct filedesc *fdp;
4780 FOREACH_PROC_IN_SYSTEM(p) {
4781 if (p->p_state == PRS_NEW)
4786 for (n = 0; n < fdp->fd_nfiles; n++) {
4787 if (fp == fdp->fd_ofiles[n].fde_file)
4795 db_print_file(struct file *fp, int header)
4797 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4801 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4802 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4803 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4805 p = file_to_first_proc(fp);
4806 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4807 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4808 fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4809 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4814 DB_SHOW_COMMAND(file, db_show_file)
4819 db_printf("usage: show file <addr>\n");
4822 fp = (struct file *)addr;
4823 db_print_file(fp, 1);
4826 DB_SHOW_COMMAND(files, db_show_files)
4828 struct filedesc *fdp;
4835 FOREACH_PROC_IN_SYSTEM(p) {
4836 if (p->p_state == PRS_NEW)
4838 if ((fdp = p->p_fd) == NULL)
4840 for (n = 0; n < fdp->fd_nfiles; ++n) {
4841 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4843 db_print_file(fp, header);
4850 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4851 &maxfilesperproc, 0, "Maximum files allowed open per process");
4853 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4854 &maxfiles, 0, "Maximum number of files");
4856 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4857 &openfiles, 0, "System-wide number of open files");
4861 filelistinit(void *dummy)
4864 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4865 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4866 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4867 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4868 pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4869 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4871 * XXXMJG this is a temporary hack due to boot ordering issues against
4874 vfs_smr = uma_zone_get_smr(pwd_zone);
4875 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4877 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4879 /*-------------------------------------------------------------------*/
4882 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4883 int flags, struct thread *td)
4890 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4898 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4906 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4914 badfo_kqfilter(struct file *fp, struct knote *kn)
4921 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
4928 badfo_close(struct file *fp, struct thread *td)
4935 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4943 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4951 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4952 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4960 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4966 struct fileops badfileops = {
4967 .fo_read = badfo_readwrite,
4968 .fo_write = badfo_readwrite,
4969 .fo_truncate = badfo_truncate,
4970 .fo_ioctl = badfo_ioctl,
4971 .fo_poll = badfo_poll,
4972 .fo_kqfilter = badfo_kqfilter,
4973 .fo_stat = badfo_stat,
4974 .fo_close = badfo_close,
4975 .fo_chmod = badfo_chmod,
4976 .fo_chown = badfo_chown,
4977 .fo_sendfile = badfo_sendfile,
4978 .fo_fill_kinfo = badfo_fill_kinfo,
4982 path_poll(struct file *fp, int events, struct ucred *active_cred,
4989 path_close(struct file *fp, struct thread *td)
4991 MPASS(fp->f_type == DTYPE_VNODE);
4992 fp->f_ops = &badfileops;
4997 struct fileops path_fileops = {
4998 .fo_read = badfo_readwrite,
4999 .fo_write = badfo_readwrite,
5000 .fo_truncate = badfo_truncate,
5001 .fo_ioctl = badfo_ioctl,
5002 .fo_poll = path_poll,
5003 .fo_kqfilter = vn_kqfilter_opath,
5004 .fo_stat = vn_statfile,
5005 .fo_close = path_close,
5006 .fo_chmod = badfo_chmod,
5007 .fo_chown = badfo_chown,
5008 .fo_sendfile = badfo_sendfile,
5009 .fo_fill_kinfo = vn_fill_kinfo,
5010 .fo_flags = DFLAG_PASSABLE,
5014 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
5015 int flags, struct thread *td)
5018 return (EOPNOTSUPP);
5022 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
5030 invfo_ioctl(struct file *fp, u_long com, void *data,
5031 struct ucred *active_cred, struct thread *td)
5038 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
5042 return (poll_no_poll(events));
5046 invfo_kqfilter(struct file *fp, struct knote *kn)
5053 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5061 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5069 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5070 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5077 /*-------------------------------------------------------------------*/
5080 * File Descriptor pseudo-device driver (/dev/fd/).
5082 * Opening minor device N dup()s the file (if any) connected to file
5083 * descriptor N belonging to the calling process. Note that this driver
5084 * consists of only the ``open()'' routine, because all subsequent
5085 * references to this file will be direct to the other driver.
5087 * XXX: we could give this one a cloning event handler if necessary.
5092 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
5096 * XXX Kludge: set curthread->td_dupfd to contain the value of the
5097 * the file descriptor being sought for duplication. The error
5098 * return ensures that the vnode for this device will be released
5099 * by vn_open. Open will detect this special error and take the
5100 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
5101 * will simply report the error.
5103 td->td_dupfd = dev2unit(dev);
5107 static struct cdevsw fildesc_cdevsw = {
5108 .d_version = D_VERSION,
5114 fildesc_drvinit(void *unused)
5118 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
5119 UID_ROOT, GID_WHEEL, 0666, "fd/0");
5120 make_dev_alias(dev, "stdin");
5121 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
5122 UID_ROOT, GID_WHEEL, 0666, "fd/1");
5123 make_dev_alias(dev, "stdout");
5124 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
5125 UID_ROOT, GID_WHEEL, 0666, "fd/2");
5126 make_dev_alias(dev, "stderr");
5129 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);