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 int error, flg, seals, tmp;
497 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
500 case F_DUPFD_CLOEXEC:
502 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
507 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
510 case F_DUP2FD_CLOEXEC:
512 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
518 fde = fdeget_locked(fdp, fd);
521 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
524 FILEDESC_SUNLOCK(fdp);
530 fde = fdeget_locked(fdp, fd);
532 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
533 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
536 FILEDESC_XUNLOCK(fdp);
540 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
543 td->td_retval[0] = OFLAGS(fp->f_flag);
548 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
551 if (fp->f_ops == &path_fileops) {
557 tmp = flg = fp->f_flag;
559 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
560 } while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
561 tmp = fp->f_flag & FNONBLOCK;
562 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
567 tmp = fp->f_flag & FASYNC;
568 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
573 atomic_clear_int(&fp->f_flag, FNONBLOCK);
575 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
580 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
583 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
585 td->td_retval[0] = tmp;
590 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
594 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
599 error = priv_check(td, PRIV_NFS_LOCKD);
607 /* FALLTHROUGH F_SETLK */
611 flp = (struct flock *)arg;
612 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
617 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
620 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
626 if (flp->l_whence == SEEK_CUR) {
627 foffset = foffset_get(fp);
630 foffset > OFF_MAX - flp->l_start)) {
635 flp->l_start += foffset;
639 switch (flp->l_type) {
641 if ((fp->f_flag & FREAD) == 0) {
645 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
646 PROC_LOCK(p->p_leader);
647 p->p_leader->p_flag |= P_ADVLOCK;
648 PROC_UNLOCK(p->p_leader);
650 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
654 if ((fp->f_flag & FWRITE) == 0) {
658 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
659 PROC_LOCK(p->p_leader);
660 p->p_leader->p_flag |= P_ADVLOCK;
661 PROC_UNLOCK(p->p_leader);
663 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
667 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
671 if (flg != F_REMOTE) {
675 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
676 F_UNLCKSYS, flp, flg);
682 if (error != 0 || flp->l_type == F_UNLCK ||
683 flp->l_type == F_UNLCKSYS) {
689 * Check for a race with close.
691 * The vnode is now advisory locked (or unlocked, but this case
692 * is not really important) as the caller requested.
693 * We had to drop the filedesc lock, so we need to recheck if
694 * the descriptor is still valid, because if it was closed
695 * in the meantime we need to remove advisory lock from the
696 * vnode - close on any descriptor leading to an advisory
697 * locked vnode, removes that lock.
698 * We will return 0 on purpose in that case, as the result of
699 * successful advisory lock might have been externally visible
700 * already. This is fine - effectively we pretend to the caller
701 * that the closing thread was a bit slower and that the
702 * advisory lock succeeded before the close.
704 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
710 flp->l_whence = SEEK_SET;
713 flp->l_type = F_UNLCK;
714 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
715 F_UNLCK, flp, F_POSIX);
722 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
725 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
730 flp = (struct flock *)arg;
731 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
732 flp->l_type != F_UNLCK) {
737 if (flp->l_whence == SEEK_CUR) {
738 foffset = foffset_get(fp);
739 if ((flp->l_start > 0 &&
740 foffset > OFF_MAX - flp->l_start) ||
742 foffset < OFF_MIN - flp->l_start)) {
747 flp->l_start += foffset;
750 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
756 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
759 error = fo_add_seals(fp, arg);
764 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
767 if (fo_get_seals(fp, &seals) == 0)
768 td->td_retval[0] = seals;
775 arg = arg ? 128 * 1024: 0;
778 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
781 if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
787 if (vp->v_type != VREG) {
794 * Exclusive lock synchronizes against f_seqcount reads and
795 * writes in sequential_heuristic().
797 error = vn_lock(vp, LK_EXCLUSIVE);
803 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
804 arg = MIN(arg, INT_MAX - bsize + 1);
805 fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
806 (arg + bsize - 1) / bsize);
807 atomic_set_int(&fp->f_flag, FRDAHEAD);
809 atomic_clear_int(&fp->f_flag, FRDAHEAD);
817 * Check if the vnode is part of a union stack (either the
818 * "union" flag from mount(2) or unionfs).
820 * Prior to introduction of this op libc's readdir would call
821 * fstatfs(2), in effect unnecessarily copying kilobytes of
822 * data just to check fs name and a mount flag.
824 * Fixing the code to handle everything in the kernel instead
825 * is a non-trivial endeavor and has low priority, thus this
826 * horrible kludge facilitates the current behavior in a much
827 * cheaper manner until someone(tm) sorts this out.
829 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
832 if (fp->f_type != DTYPE_VNODE) {
839 * Since we don't prevent dooming the vnode even non-null mp
840 * found can become immediately stale. This is tolerable since
841 * mount points are type-stable (providing safe memory access)
842 * and any vfs op on this vnode going forward will return an
843 * error (meaning return value in this case is meaningless).
845 mp = atomic_load_ptr(&vp->v_mount);
846 if (__predict_false(mp == NULL)) {
851 td->td_retval[0] = 0;
852 if (mp->mnt_kern_flag & MNTK_UNIONFS ||
853 mp->mnt_flag & MNT_UNION)
854 td->td_retval[0] = 1;
866 getmaxfd(struct thread *td)
869 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
873 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
876 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
878 struct filedesc *fdp;
879 struct filedescent *oldfde, *newfde;
881 struct file *delfp, *oldfp;
882 u_long *oioctls, *nioctls;
889 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
890 MPASS(mode < FDDUP_LASTMODE);
893 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
896 * Verify we have a valid descriptor to dup from and possibly to
897 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
898 * return EINVAL when the new descriptor is out of bounds.
903 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
904 maxfd = getmaxfd(td);
906 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
910 if (fget_locked(fdp, old) == NULL)
912 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
913 td->td_retval[0] = new;
914 if (flags & FDDUP_FLAG_CLOEXEC)
915 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
920 oldfde = &fdp->fd_ofiles[old];
921 oldfp = oldfde->fde_file;
926 * If the caller specified a file descriptor, make sure the file
927 * table is large enough to hold it, and grab it. Otherwise, just
928 * allocate a new descriptor the usual way.
933 if ((error = fdalloc(td, new, &new)) != 0) {
938 case FDDUP_MUSTREPLACE:
939 /* Target file descriptor must exist. */
940 if (fget_locked(fdp, new) == NULL) {
946 if (new >= fdp->fd_nfiles) {
948 * The resource limits are here instead of e.g.
949 * fdalloc(), because the file descriptor table may be
950 * shared between processes, so we can't really use
951 * racct_add()/racct_sub(). Instead of counting the
952 * number of actually allocated descriptors, just put
953 * the limit on the size of the file descriptor table.
956 if (RACCT_ENABLED()) {
957 error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
965 fdgrowtable_exp(fdp, new + 1);
967 if (!fdisused(fdp, new))
971 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
974 KASSERT(old != new, ("new fd is same as old"));
976 /* Refetch oldfde because the table may have grown and old one freed. */
977 oldfde = &fdp->fd_ofiles[old];
978 KASSERT(oldfp == oldfde->fde_file,
979 ("fdt_ofiles shift from growth observed at fd %d",
982 newfde = &fdp->fd_ofiles[new];
983 delfp = newfde->fde_file;
985 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
988 * Duplicate the source descriptor.
991 seqc_write_begin(&newfde->fde_seqc);
993 oioctls = filecaps_free_prep(&newfde->fde_caps);
994 memcpy(newfde, oldfde, fde_change_size);
995 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
997 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
998 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
1000 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
1002 seqc_write_end(&newfde->fde_seqc);
1004 td->td_retval[0] = new;
1008 if (delfp != NULL) {
1009 (void) closefp(fdp, new, delfp, td, true, false);
1010 FILEDESC_UNLOCK_ASSERT(fdp);
1013 FILEDESC_XUNLOCK(fdp);
1016 filecaps_free_finish(oioctls);
1021 sigiofree(struct sigio *sigio)
1023 crfree(sigio->sio_ucred);
1024 free(sigio, M_SIGIO);
1027 static struct sigio *
1028 funsetown_locked(struct sigio *sigio)
1033 SIGIO_ASSERT_LOCKED();
1037 *sigio->sio_myref = NULL;
1038 if (sigio->sio_pgid < 0) {
1039 pg = sigio->sio_pgrp;
1041 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio);
1044 p = sigio->sio_proc;
1046 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio);
1053 * If sigio is on the list associated with a process or process group,
1054 * disable signalling from the device, remove sigio from the list and
1058 funsetown(struct sigio **sigiop)
1060 struct sigio *sigio;
1062 /* Racy check, consumers must provide synchronization. */
1063 if (*sigiop == NULL)
1067 sigio = funsetown_locked(*sigiop);
1074 * Free a list of sigio structures. The caller must ensure that new sigio
1075 * structures cannot be added after this point. For process groups this is
1076 * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1080 funsetownlst(struct sigiolst *sigiolst)
1084 struct sigio *sigio, *tmp;
1087 sigio = SLIST_FIRST(sigiolst);
1095 sigio = SLIST_FIRST(sigiolst);
1096 if (sigio == NULL) {
1102 * Every entry of the list should belong to a single proc or pgrp.
1104 if (sigio->sio_pgid < 0) {
1105 pg = sigio->sio_pgrp;
1106 sx_assert(&proctree_lock, SX_XLOCKED);
1108 } else /* if (sigio->sio_pgid > 0) */ {
1109 p = sigio->sio_proc;
1111 KASSERT((p->p_flag & P_WEXIT) != 0,
1112 ("%s: process %p is not exiting", __func__, p));
1115 SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1116 *sigio->sio_myref = NULL;
1118 KASSERT(sigio->sio_pgid < 0,
1119 ("Proc sigio in pgrp sigio list"));
1120 KASSERT(sigio->sio_pgrp == pg,
1121 ("Bogus pgrp in sigio list"));
1122 } else /* if (p != NULL) */ {
1123 KASSERT(sigio->sio_pgid > 0,
1124 ("Pgrp sigio in proc sigio list"));
1125 KASSERT(sigio->sio_proc == p,
1126 ("Bogus proc in sigio list"));
1136 SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1141 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1143 * After permission checking, add a sigio structure to the sigio list for
1144 * the process or process group.
1147 fsetown(pid_t pgid, struct sigio **sigiop)
1151 struct sigio *osigio, *sigio;
1159 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1160 sigio->sio_pgid = pgid;
1161 sigio->sio_ucred = crhold(curthread->td_ucred);
1162 sigio->sio_myref = sigiop;
1166 ret = pget(pgid, PGET_NOTWEXIT | PGET_NOTID | PGET_HOLD, &proc);
1168 osigio = funsetown_locked(*sigiop);
1172 if ((proc->p_flag & P_WEXIT) != 0) {
1174 } else if (proc->p_session !=
1175 curthread->td_proc->p_session) {
1177 * Policy - Don't allow a process to FSETOWN a
1178 * process in another session.
1180 * Remove this test to allow maximum flexibility
1181 * or restrict FSETOWN to the current process or
1182 * process group for maximum safety.
1186 sigio->sio_proc = proc;
1187 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio,
1192 } else /* if (pgid < 0) */ {
1193 sx_slock(&proctree_lock);
1195 osigio = funsetown_locked(*sigiop);
1196 pgrp = pgfind(-pgid);
1200 if (pgrp->pg_session != curthread->td_proc->p_session) {
1202 * Policy - Don't allow a process to FSETOWN a
1203 * process in another session.
1205 * Remove this test to allow maximum flexibility
1206 * or restrict FSETOWN to the current process or
1207 * process group for maximum safety.
1211 sigio->sio_pgrp = pgrp;
1212 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio,
1217 sx_sunlock(&proctree_lock);
1228 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1231 fgetown(struct sigio **sigiop)
1236 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1242 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1247 FILEDESC_XLOCK_ASSERT(fdp);
1250 * We now hold the fp reference that used to be owned by the
1251 * descriptor array. We have to unlock the FILEDESC *AFTER*
1252 * knote_fdclose to prevent a race of the fd getting opened, a knote
1253 * added, and deleteing a knote for the new fd.
1255 if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1256 knote_fdclose(td, fd);
1259 * We need to notify mqueue if the object is of type mqueue.
1261 if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1262 mq_fdclose(td, fd, fp);
1263 FILEDESC_XUNLOCK(fdp);
1266 if (AUDITING_TD(td) && audit)
1267 audit_sysclose(td, fd, fp);
1269 error = closef(fp, td);
1272 * All paths leading up to closefp() will have already removed or
1273 * replaced the fd in the filedesc table, so a restart would not
1274 * operate on the same file.
1276 if (error == ERESTART)
1283 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1284 bool holdleaders, bool audit)
1288 FILEDESC_XLOCK_ASSERT(fdp);
1291 if (td->td_proc->p_fdtol != NULL) {
1293 * Ask fdfree() to sleep to ensure that all relevant
1294 * process leaders can be traversed in closef().
1296 fdp->fd_holdleaderscount++;
1298 holdleaders = false;
1302 error = closefp_impl(fdp, fd, fp, td, audit);
1304 FILEDESC_XLOCK(fdp);
1305 fdp->fd_holdleaderscount--;
1306 if (fdp->fd_holdleaderscount == 0 &&
1307 fdp->fd_holdleaderswakeup != 0) {
1308 fdp->fd_holdleaderswakeup = 0;
1309 wakeup(&fdp->fd_holdleaderscount);
1311 FILEDESC_XUNLOCK(fdp);
1317 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1318 bool holdleaders, bool audit)
1321 FILEDESC_XLOCK_ASSERT(fdp);
1323 if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1324 return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1326 return (closefp_impl(fdp, fd, fp, td, audit));
1331 * Close a file descriptor.
1333 #ifndef _SYS_SYSPROTO_H_
1340 sys_close(struct thread *td, struct close_args *uap)
1343 return (kern_close(td, uap->fd));
1347 kern_close(struct thread *td, int fd)
1349 struct filedesc *fdp;
1352 fdp = td->td_proc->p_fd;
1354 FILEDESC_XLOCK(fdp);
1355 if ((fp = fget_locked(fdp, fd)) == NULL) {
1356 FILEDESC_XUNLOCK(fdp);
1361 /* closefp() drops the FILEDESC lock for us. */
1362 return (closefp(fdp, fd, fp, td, true, true));
1366 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1368 struct filedesc *fdp;
1369 const struct fdescenttbl *fdt;
1374 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1375 * open should not be a usage error. From a close_range() perspective,
1376 * close_range(3, ~0U, 0) in the same scenario should also likely not
1377 * be a usage error as all fd above 3 are in-fact already closed.
1379 if (highfd < lowfd) {
1383 fdp = td->td_proc->p_fd;
1384 FILEDESC_XLOCK(fdp);
1385 fdt = atomic_load_ptr(&fdp->fd_files);
1386 highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1388 if (__predict_false(fd > highfd)) {
1392 fp = fdt->fdt_ofiles[fd].fde_file;
1398 (void) closefp(fdp, fd, fp, td, true, true);
1401 FILEDESC_XLOCK(fdp);
1402 fdt = atomic_load_ptr(&fdp->fd_files);
1407 FILEDESC_XUNLOCK(fdp);
1412 #ifndef _SYS_SYSPROTO_H_
1413 struct close_range_args {
1420 sys_close_range(struct thread *td, struct close_range_args *uap)
1423 AUDIT_ARG_FD(uap->lowfd);
1424 AUDIT_ARG_CMD(uap->highfd);
1425 AUDIT_ARG_FFLAGS(uap->flags);
1427 /* No flags currently defined */
1428 if (uap->flags != 0)
1430 return (kern_close_range(td, uap->lowfd, uap->highfd));
1433 #ifdef COMPAT_FREEBSD12
1435 * Close open file descriptors.
1437 #ifndef _SYS_SYSPROTO_H_
1438 struct freebsd12_closefrom_args {
1444 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1448 AUDIT_ARG_FD(uap->lowfd);
1451 * Treat negative starting file descriptor values identical to
1452 * closefrom(0) which closes all files.
1454 lowfd = MAX(0, uap->lowfd);
1455 return (kern_close_range(td, lowfd, ~0U));
1457 #endif /* COMPAT_FREEBSD12 */
1459 #if defined(COMPAT_43)
1461 * Return status information about a file descriptor.
1463 #ifndef _SYS_SYSPROTO_H_
1464 struct ofstat_args {
1471 ofstat(struct thread *td, struct ofstat_args *uap)
1477 error = kern_fstat(td, uap->fd, &ub);
1480 error = copyout(&oub, uap->sb, sizeof(oub));
1484 #endif /* COMPAT_43 */
1486 #if defined(COMPAT_FREEBSD11)
1488 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1491 struct freebsd11_stat osb;
1494 error = kern_fstat(td, uap->fd, &sb);
1497 error = freebsd11_cvtstat(&sb, &osb);
1499 error = copyout(&osb, uap->sb, sizeof(osb));
1502 #endif /* COMPAT_FREEBSD11 */
1505 * Return status information about a file descriptor.
1507 #ifndef _SYS_SYSPROTO_H_
1515 sys_fstat(struct thread *td, struct fstat_args *uap)
1520 error = kern_fstat(td, uap->fd, &ub);
1522 error = copyout(&ub, uap->sb, sizeof(ub));
1527 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1534 error = fget(td, fd, &cap_fstat_rights, &fp);
1535 if (__predict_false(error != 0))
1538 AUDIT_ARG_FILE(td->td_proc, fp);
1540 error = fo_stat(fp, sbp, td->td_ucred, td);
1542 #ifdef __STAT_TIME_T_EXT
1543 sbp->st_atim_ext = 0;
1544 sbp->st_mtim_ext = 0;
1545 sbp->st_ctim_ext = 0;
1546 sbp->st_btim_ext = 0;
1549 if (KTRPOINT(td, KTR_STRUCT))
1550 ktrstat_error(sbp, error);
1555 #if defined(COMPAT_FREEBSD11)
1557 * Return status information about a file descriptor.
1559 #ifndef _SYS_SYSPROTO_H_
1560 struct freebsd11_nfstat_args {
1567 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1573 error = kern_fstat(td, uap->fd, &ub);
1575 freebsd11_cvtnstat(&ub, &nub);
1576 error = copyout(&nub, uap->sb, sizeof(nub));
1580 #endif /* COMPAT_FREEBSD11 */
1583 * Return pathconf information about a file descriptor.
1585 #ifndef _SYS_SYSPROTO_H_
1586 struct fpathconf_args {
1593 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1598 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1600 td->td_retval[0] = value;
1605 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1611 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1615 if (name == _PC_ASYNC_IO) {
1616 *valuep = _POSIX_ASYNCHRONOUS_IO;
1621 vn_lock(vp, LK_SHARED | LK_RETRY);
1622 error = VOP_PATHCONF(vp, name, valuep);
1624 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1625 if (name != _PC_PIPE_BUF) {
1640 * Copy filecaps structure allocating memory for ioctls array if needed.
1642 * The last parameter indicates whether the fdtable is locked. If it is not and
1643 * ioctls are encountered, copying fails and the caller must lock the table.
1645 * Note that if the table was not locked, the caller has to check the relevant
1646 * sequence counter to determine whether the operation was successful.
1649 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1653 if (src->fc_ioctls != NULL && !locked)
1655 memcpy(dst, src, sizeof(*src));
1656 if (src->fc_ioctls == NULL)
1659 KASSERT(src->fc_nioctls > 0,
1660 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1662 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1663 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1664 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1669 filecaps_copy_prep(const struct filecaps *src)
1674 if (__predict_true(src->fc_ioctls == NULL))
1677 KASSERT(src->fc_nioctls > 0,
1678 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1680 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1681 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1686 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1692 if (__predict_true(src->fc_ioctls == NULL)) {
1693 MPASS(ioctls == NULL);
1697 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1698 dst->fc_ioctls = ioctls;
1699 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1703 * Move filecaps structure to the new place and clear the old place.
1706 filecaps_move(struct filecaps *src, struct filecaps *dst)
1710 bzero(src, sizeof(*src));
1714 * Fill the given filecaps structure with full rights.
1717 filecaps_fill(struct filecaps *fcaps)
1720 CAP_ALL(&fcaps->fc_rights);
1721 fcaps->fc_ioctls = NULL;
1722 fcaps->fc_nioctls = -1;
1723 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1727 * Free memory allocated within filecaps structure.
1730 filecaps_free(struct filecaps *fcaps)
1733 free(fcaps->fc_ioctls, M_FILECAPS);
1734 bzero(fcaps, sizeof(*fcaps));
1738 filecaps_free_prep(struct filecaps *fcaps)
1742 ioctls = fcaps->fc_ioctls;
1743 bzero(fcaps, sizeof(*fcaps));
1748 filecaps_free_finish(u_long *ioctls)
1751 free(ioctls, M_FILECAPS);
1755 * Validate the given filecaps structure.
1758 filecaps_validate(const struct filecaps *fcaps, const char *func)
1761 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1762 ("%s: invalid rights", func));
1763 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1764 ("%s: invalid fcntls", func));
1765 KASSERT(fcaps->fc_fcntls == 0 ||
1766 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1767 ("%s: fcntls without CAP_FCNTL", func));
1768 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1769 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1770 ("%s: invalid ioctls", func));
1771 KASSERT(fcaps->fc_nioctls == 0 ||
1772 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1773 ("%s: ioctls without CAP_IOCTL", func));
1777 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1781 FILEDESC_XLOCK_ASSERT(fdp);
1783 nfd1 = fdp->fd_nfiles * 2;
1786 fdgrowtable(fdp, nfd1);
1790 * Grow the file table to accommodate (at least) nfd descriptors.
1793 fdgrowtable(struct filedesc *fdp, int nfd)
1795 struct filedesc0 *fdp0;
1796 struct freetable *ft;
1797 struct fdescenttbl *ntable;
1798 struct fdescenttbl *otable;
1799 int nnfiles, onfiles;
1800 NDSLOTTYPE *nmap, *omap;
1802 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1804 /* save old values */
1805 onfiles = fdp->fd_nfiles;
1806 otable = fdp->fd_files;
1809 /* compute the size of the new table */
1810 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1811 if (nnfiles <= onfiles)
1812 /* the table is already large enough */
1816 * Allocate a new table. We need enough space for the number of
1817 * entries, file entries themselves and the struct freetable we will use
1818 * when we decommission the table and place it on the freelist.
1819 * We place the struct freetable in the middle so we don't have
1820 * to worry about padding.
1822 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1823 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1824 sizeof(struct freetable),
1825 M_FILEDESC, M_ZERO | M_WAITOK);
1826 /* copy the old data */
1827 ntable->fdt_nfiles = nnfiles;
1828 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1829 onfiles * sizeof(ntable->fdt_ofiles[0]));
1832 * Allocate a new map only if the old is not large enough. It will
1833 * grow at a slower rate than the table as it can map more
1834 * entries than the table can hold.
1836 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1837 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1839 /* copy over the old data and update the pointer */
1840 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1845 * Make sure that ntable is correctly initialized before we replace
1846 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1849 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1852 * Free the old file table when not shared by other threads or processes.
1853 * The old file table is considered to be shared when either are true:
1854 * - The process has more than one thread.
1855 * - The file descriptor table has been shared via fdshare().
1857 * When shared, the old file table will be placed on a freelist
1858 * which will be processed when the struct filedesc is released.
1860 * Note that if onfiles == NDFILE, we're dealing with the original
1861 * static allocation contained within (struct filedesc0 *)fdp,
1862 * which must not be freed.
1864 if (onfiles > NDFILE) {
1866 * Note we may be called here from fdinit while allocating a
1867 * table for a new process in which case ->p_fd points
1870 if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) {
1871 free(otable, M_FILEDESC);
1873 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1874 fdp0 = (struct filedesc0 *)fdp;
1875 ft->ft_table = otable;
1876 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1880 * The map does not have the same possibility of threads still
1881 * holding references to it. So always free it as long as it
1882 * does not reference the original static allocation.
1884 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1885 free(omap, M_FILEDESC);
1889 * Allocate a file descriptor for the process.
1892 fdalloc(struct thread *td, int minfd, int *result)
1894 struct proc *p = td->td_proc;
1895 struct filedesc *fdp = p->p_fd;
1896 int fd, maxfd, allocfd;
1901 FILEDESC_XLOCK_ASSERT(fdp);
1903 if (fdp->fd_freefile > minfd)
1904 minfd = fdp->fd_freefile;
1906 maxfd = getmaxfd(td);
1909 * Search the bitmap for a free descriptor starting at minfd.
1910 * If none is found, grow the file table.
1912 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1913 if (__predict_false(fd >= maxfd))
1915 if (__predict_false(fd >= fdp->fd_nfiles)) {
1916 allocfd = min(fd * 2, maxfd);
1918 if (RACCT_ENABLED()) {
1919 error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1925 * fd is already equal to first free descriptor >= minfd, so
1926 * we only need to grow the table and we are done.
1928 fdgrowtable_exp(fdp, allocfd);
1932 * Perform some sanity checks, then mark the file descriptor as
1933 * used and return it to the caller.
1935 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1936 ("invalid descriptor %d", fd));
1937 KASSERT(!fdisused(fdp, fd),
1938 ("fd_first_free() returned non-free descriptor"));
1939 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1940 ("file descriptor isn't free"));
1947 * Allocate n file descriptors for the process.
1950 fdallocn(struct thread *td, int minfd, int *fds, int n)
1952 struct proc *p = td->td_proc;
1953 struct filedesc *fdp = p->p_fd;
1956 FILEDESC_XLOCK_ASSERT(fdp);
1958 for (i = 0; i < n; i++)
1959 if (fdalloc(td, 0, &fds[i]) != 0)
1963 for (i--; i >= 0; i--)
1964 fdunused(fdp, fds[i]);
1972 * Create a new open file structure and allocate a file descriptor for the
1973 * process that refers to it. We add one reference to the file for the
1974 * descriptor table and one reference for resultfp. This is to prevent us
1975 * being preempted and the entry in the descriptor table closed after we
1976 * release the FILEDESC lock.
1979 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1980 struct filecaps *fcaps)
1985 MPASS(resultfp != NULL);
1986 MPASS(resultfd != NULL);
1988 error = _falloc_noinstall(td, &fp, 2);
1989 if (__predict_false(error != 0)) {
1993 error = finstall_refed(td, fp, &fd, flags, fcaps);
1994 if (__predict_false(error != 0)) {
1995 falloc_abort(td, fp);
2006 * Create a new open file structure without allocating a file descriptor.
2009 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2012 int maxuserfiles = maxfiles - (maxfiles / 20);
2014 static struct timeval lastfail;
2017 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2020 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2021 if ((openfiles_new >= maxuserfiles &&
2022 priv_check(td, PRIV_MAXFILES) != 0) ||
2023 openfiles_new >= maxfiles) {
2024 atomic_subtract_int(&openfiles, 1);
2025 if (ppsratecheck(&lastfail, &curfail, 1)) {
2026 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2027 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2031 fp = uma_zalloc(file_zone, M_WAITOK);
2032 bzero(fp, sizeof(*fp));
2033 refcount_init(&fp->f_count, n);
2034 fp->f_cred = crhold(td->td_ucred);
2035 fp->f_ops = &badfileops;
2041 falloc_abort(struct thread *td, struct file *fp)
2045 * For assertion purposes.
2047 refcount_init(&fp->f_count, 0);
2052 * Install a file in a file descriptor table.
2055 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2056 struct filecaps *fcaps)
2058 struct filedescent *fde;
2062 filecaps_validate(fcaps, __func__);
2063 FILEDESC_XLOCK_ASSERT(fdp);
2065 fde = &fdp->fd_ofiles[fd];
2067 seqc_write_begin(&fde->fde_seqc);
2070 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2072 filecaps_move(fcaps, &fde->fde_caps);
2074 filecaps_fill(&fde->fde_caps);
2076 seqc_write_end(&fde->fde_seqc);
2081 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2082 struct filecaps *fcaps)
2084 struct filedesc *fdp = td->td_proc->p_fd;
2089 FILEDESC_XLOCK(fdp);
2090 error = fdalloc(td, 0, fd);
2091 if (__predict_true(error == 0)) {
2092 _finstall(fdp, fp, *fd, flags, fcaps);
2094 FILEDESC_XUNLOCK(fdp);
2099 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2100 struct filecaps *fcaps)
2108 error = finstall_refed(td, fp, fd, flags, fcaps);
2109 if (__predict_false(error != 0)) {
2116 * Build a new filedesc structure from another.
2118 * If fdp is not NULL, return with it shared locked.
2121 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2123 struct filedesc0 *newfdp0;
2124 struct filedesc *newfdp;
2127 MPASS(lastfile != NULL);
2129 MPASS(lastfile == NULL);
2131 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2132 newfdp = &newfdp0->fd_fd;
2134 /* Create the file descriptor table. */
2135 FILEDESC_LOCK_INIT(newfdp);
2136 refcount_init(&newfdp->fd_refcnt, 1);
2137 refcount_init(&newfdp->fd_holdcnt, 1);
2138 newfdp->fd_map = newfdp0->fd_dmap;
2139 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2140 newfdp->fd_files->fdt_nfiles = NDFILE;
2145 FILEDESC_SLOCK(fdp);
2147 FILEDESC_SUNLOCK(fdp);
2152 *lastfile = fdlastfile(fdp);
2153 if (*lastfile < newfdp->fd_nfiles)
2155 FILEDESC_SUNLOCK(fdp);
2156 fdgrowtable(newfdp, *lastfile + 1);
2157 FILEDESC_SLOCK(fdp);
2164 * Build a pwddesc structure from another.
2165 * Copy the current, root, and jail root vnode references.
2167 * If pdp is not NULL, return with it shared locked.
2170 pdinit(struct pwddesc *pdp, bool keeplock)
2172 struct pwddesc *newpdp;
2175 newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2177 PWDDESC_LOCK_INIT(newpdp);
2178 refcount_init(&newpdp->pd_refcount, 1);
2179 newpdp->pd_cmask = CMASK;
2182 newpwd = pwd_alloc();
2183 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2188 newpwd = pwd_hold_pwddesc(pdp);
2189 smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2191 PWDDESC_XUNLOCK(pdp);
2196 * Hold either filedesc or pwddesc of the passed process.
2198 * The process lock is used to synchronize against the target exiting and
2201 * Clearing can be ilustrated in 3 steps:
2202 * 1. set the pointer to NULL. Either routine can race against it, hence
2204 * 2. observe the process lock as not taken. Until then fdhold/pdhold can
2205 * race to either still see the pointer or find NULL. It is still safe to
2206 * grab a reference as clearing is stalled.
2207 * 3. after the lock is observed as not taken, any fdhold/pdhold calls are
2208 * guaranteed to see NULL, making it safe to finish clearing
2210 static struct filedesc *
2211 fdhold(struct proc *p)
2213 struct filedesc *fdp;
2215 PROC_LOCK_ASSERT(p, MA_OWNED);
2216 fdp = atomic_load_ptr(&p->p_fd);
2218 refcount_acquire(&fdp->fd_holdcnt);
2222 static struct pwddesc *
2223 pdhold(struct proc *p)
2225 struct pwddesc *pdp;
2227 PROC_LOCK_ASSERT(p, MA_OWNED);
2228 pdp = atomic_load_ptr(&p->p_pd);
2230 refcount_acquire(&pdp->pd_refcount);
2235 fddrop(struct filedesc *fdp)
2238 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2239 if (refcount_release(&fdp->fd_holdcnt) == 0)
2243 FILEDESC_LOCK_DESTROY(fdp);
2244 uma_zfree(filedesc0_zone, fdp);
2248 pddrop(struct pwddesc *pdp)
2252 if (refcount_release_if_not_last(&pdp->pd_refcount))
2256 if (refcount_release(&pdp->pd_refcount) == 0) {
2257 PWDDESC_XUNLOCK(pdp);
2260 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2262 PWDDESC_XUNLOCK(pdp);
2265 PWDDESC_LOCK_DESTROY(pdp);
2266 free(pdp, M_PWDDESC);
2270 * Share a filedesc structure.
2273 fdshare(struct filedesc *fdp)
2276 refcount_acquire(&fdp->fd_refcnt);
2281 * Share a pwddesc structure.
2284 pdshare(struct pwddesc *pdp)
2286 refcount_acquire(&pdp->pd_refcount);
2291 * Unshare a filedesc structure, if necessary by making a copy
2294 fdunshare(struct thread *td)
2296 struct filedesc *tmp;
2297 struct proc *p = td->td_proc;
2299 if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2302 tmp = fdcopy(p->p_fd);
2308 * Unshare a pwddesc structure.
2311 pdunshare(struct thread *td)
2313 struct pwddesc *pdp;
2318 if (p->p_pd->pd_refcount == 1)
2321 pdp = pdcopy(p->p_pd);
2327 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2331 td->td_proc->p_fd = fdp;
2335 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2336 * this is to ease callers, not catch errors.
2339 fdcopy(struct filedesc *fdp)
2341 struct filedesc *newfdp;
2342 struct filedescent *nfde, *ofde;
2347 newfdp = fdinit(fdp, true, &lastfile);
2348 /* copy all passable descriptors (i.e. not kqueue) */
2349 newfdp->fd_freefile = -1;
2350 for (i = 0; i <= lastfile; ++i) {
2351 ofde = &fdp->fd_ofiles[i];
2352 if (ofde->fde_file == NULL ||
2353 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2354 !fhold(ofde->fde_file)) {
2355 if (newfdp->fd_freefile == -1)
2356 newfdp->fd_freefile = i;
2359 nfde = &newfdp->fd_ofiles[i];
2361 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2362 fdused_init(newfdp, i);
2364 if (newfdp->fd_freefile == -1)
2365 newfdp->fd_freefile = i;
2366 FILEDESC_SUNLOCK(fdp);
2371 * Copy a pwddesc structure.
2374 pdcopy(struct pwddesc *pdp)
2376 struct pwddesc *newpdp;
2380 newpdp = pdinit(pdp, true);
2381 newpdp->pd_cmask = pdp->pd_cmask;
2382 PWDDESC_XUNLOCK(pdp);
2387 * Copies a filedesc structure, while remapping all file descriptors
2388 * stored inside using a translation table.
2390 * File descriptors are copied over to the new file descriptor table,
2391 * regardless of whether the close-on-exec flag is set.
2394 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2395 struct filedesc **ret)
2397 struct filedesc *newfdp;
2398 struct filedescent *nfde, *ofde;
2399 int error, i, lastfile;
2403 newfdp = fdinit(fdp, true, &lastfile);
2404 if (nfds > lastfile + 1) {
2405 /* New table cannot be larger than the old one. */
2409 /* Copy all passable descriptors (i.e. not kqueue). */
2410 newfdp->fd_freefile = nfds;
2411 for (i = 0; i < nfds; ++i) {
2412 if (fds[i] < 0 || fds[i] > lastfile) {
2413 /* File descriptor out of bounds. */
2417 ofde = &fdp->fd_ofiles[fds[i]];
2418 if (ofde->fde_file == NULL) {
2419 /* Unused file descriptor. */
2423 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2424 /* File descriptor cannot be passed. */
2428 if (!fhold(ofde->fde_file)) {
2432 nfde = &newfdp->fd_ofiles[i];
2434 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2435 fdused_init(newfdp, i);
2437 FILEDESC_SUNLOCK(fdp);
2441 FILEDESC_SUNLOCK(fdp);
2442 fdescfree_remapped(newfdp);
2447 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2448 * one of processes using it exits) and the table used to be shared.
2451 fdclearlocks(struct thread *td)
2453 struct filedesc *fdp;
2454 struct filedesc_to_leader *fdtol;
2464 MPASS(fdtol != NULL);
2466 FILEDESC_XLOCK(fdp);
2467 KASSERT(fdtol->fdl_refcount > 0,
2468 ("filedesc_to_refcount botch: fdl_refcount=%d",
2469 fdtol->fdl_refcount));
2470 if (fdtol->fdl_refcount == 1 &&
2471 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2472 lastfile = fdlastfile(fdp);
2473 for (i = 0; i <= lastfile; i++) {
2474 fp = fdp->fd_ofiles[i].fde_file;
2475 if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2478 FILEDESC_XUNLOCK(fdp);
2479 lf.l_whence = SEEK_SET;
2482 lf.l_type = F_UNLCK;
2484 (void) VOP_ADVLOCK(vp,
2485 (caddr_t)p->p_leader, F_UNLCK,
2487 FILEDESC_XLOCK(fdp);
2492 if (fdtol->fdl_refcount == 1) {
2493 if (fdp->fd_holdleaderscount > 0 &&
2494 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2496 * close() or kern_dup() has cleared a reference
2497 * in a shared file descriptor table.
2499 fdp->fd_holdleaderswakeup = 1;
2500 sx_sleep(&fdp->fd_holdleaderscount,
2501 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2504 if (fdtol->fdl_holdcount > 0) {
2506 * Ensure that fdtol->fdl_leader remains
2507 * valid in closef().
2509 fdtol->fdl_wakeup = 1;
2510 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2515 fdtol->fdl_refcount--;
2516 if (fdtol->fdl_refcount == 0 &&
2517 fdtol->fdl_holdcount == 0) {
2518 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2519 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2523 FILEDESC_XUNLOCK(fdp);
2525 free(fdtol, M_FILEDESC_TO_LEADER);
2529 * Release a filedesc structure.
2532 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2534 struct filedesc0 *fdp0;
2535 struct freetable *ft, *tft;
2536 struct filedescent *fde;
2540 KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2541 ("%s: fd table %p carries references", __func__, fdp));
2544 * Serialize with threads iterating over the table, if any.
2546 if (refcount_load(&fdp->fd_holdcnt) > 1) {
2547 FILEDESC_XLOCK(fdp);
2548 FILEDESC_XUNLOCK(fdp);
2551 lastfile = fdlastfile_single(fdp);
2552 for (i = 0; i <= lastfile; i++) {
2553 fde = &fdp->fd_ofiles[i];
2558 (void) closef(fp, td);
2564 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2565 free(fdp->fd_map, M_FILEDESC);
2566 if (fdp->fd_nfiles > NDFILE)
2567 free(fdp->fd_files, M_FILEDESC);
2569 fdp0 = (struct filedesc0 *)fdp;
2570 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2571 free(ft->ft_table, M_FILEDESC);
2577 fdescfree(struct thread *td)
2580 struct filedesc *fdp;
2587 if (RACCT_ENABLED())
2588 racct_set_unlocked(p, RACCT_NOFILE, 0);
2591 if (p->p_fdtol != NULL)
2595 * Check fdhold for an explanation.
2597 atomic_store_ptr(&p->p_fd, NULL);
2598 atomic_thread_fence_seq_cst();
2599 PROC_WAIT_UNLOCKED(p);
2601 if (refcount_release(&fdp->fd_refcnt) == 0)
2604 fdescfree_fds(td, fdp, 1);
2608 pdescfree(struct thread *td)
2611 struct pwddesc *pdp;
2618 * Check pdhold for an explanation.
2620 atomic_store_ptr(&p->p_pd, NULL);
2621 atomic_thread_fence_seq_cst();
2622 PROC_WAIT_UNLOCKED(p);
2628 fdescfree_remapped(struct filedesc *fdp)
2631 /* fdescfree_fds() asserts that fd_refcnt == 0. */
2632 if (!refcount_release(&fdp->fd_refcnt))
2633 panic("%s: fd table %p has extra references", __func__, fdp);
2635 fdescfree_fds(curthread, fdp, 0);
2639 * For setugid programs, we don't want to people to use that setugidness
2640 * to generate error messages which write to a file which otherwise would
2641 * otherwise be off-limits to the process. We check for filesystems where
2642 * the vnode can change out from under us after execve (like [lin]procfs).
2644 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2645 * sufficient. We also don't check for setugidness since we know we are.
2648 is_unsafe(struct file *fp)
2652 if (fp->f_type != DTYPE_VNODE)
2656 return ((vp->v_vflag & VV_PROCDEP) != 0);
2660 * Make this setguid thing safe, if at all possible.
2663 fdsetugidsafety(struct thread *td)
2665 struct filedesc *fdp;
2669 fdp = td->td_proc->p_fd;
2670 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2671 ("the fdtable should not be shared"));
2672 MPASS(fdp->fd_nfiles >= 3);
2673 for (i = 0; i <= 2; i++) {
2674 fp = fdp->fd_ofiles[i].fde_file;
2675 if (fp != NULL && is_unsafe(fp)) {
2676 FILEDESC_XLOCK(fdp);
2677 knote_fdclose(td, i);
2679 * NULL-out descriptor prior to close to avoid
2680 * a race while close blocks.
2683 FILEDESC_XUNLOCK(fdp);
2684 (void) closef(fp, td);
2690 * If a specific file object occupies a specific file descriptor, close the
2691 * file descriptor entry and drop a reference on the file object. This is a
2692 * convenience function to handle a subsequent error in a function that calls
2693 * falloc() that handles the race that another thread might have closed the
2694 * file descriptor out from under the thread creating the file object.
2697 fdclose(struct thread *td, struct file *fp, int idx)
2699 struct filedesc *fdp = td->td_proc->p_fd;
2701 FILEDESC_XLOCK(fdp);
2702 if (fdp->fd_ofiles[idx].fde_file == fp) {
2704 FILEDESC_XUNLOCK(fdp);
2707 FILEDESC_XUNLOCK(fdp);
2711 * Close any files on exec?
2714 fdcloseexec(struct thread *td)
2716 struct filedesc *fdp;
2717 struct filedescent *fde;
2721 fdp = td->td_proc->p_fd;
2722 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2723 ("the fdtable should not be shared"));
2724 lastfile = fdlastfile_single(fdp);
2725 for (i = 0; i <= lastfile; i++) {
2726 fde = &fdp->fd_ofiles[i];
2728 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2729 (fde->fde_flags & UF_EXCLOSE))) {
2730 FILEDESC_XLOCK(fdp);
2732 (void) closefp(fdp, i, fp, td, false, false);
2733 FILEDESC_UNLOCK_ASSERT(fdp);
2739 * It is unsafe for set[ug]id processes to be started with file
2740 * descriptors 0..2 closed, as these descriptors are given implicit
2741 * significance in the Standard C library. fdcheckstd() will create a
2742 * descriptor referencing /dev/null for each of stdin, stdout, and
2743 * stderr that is not already open.
2746 fdcheckstd(struct thread *td)
2748 struct filedesc *fdp;
2750 int i, error, devnull;
2752 fdp = td->td_proc->p_fd;
2753 KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2754 ("the fdtable should not be shared"));
2755 MPASS(fdp->fd_nfiles >= 3);
2757 for (i = 0; i <= 2; i++) {
2758 if (fdp->fd_ofiles[i].fde_file != NULL)
2761 save = td->td_retval[0];
2762 if (devnull != -1) {
2763 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2765 error = kern_openat(td, AT_FDCWD, "/dev/null",
2766 UIO_SYSSPACE, O_RDWR, 0);
2768 devnull = td->td_retval[0];
2769 KASSERT(devnull == i, ("we didn't get our fd"));
2772 td->td_retval[0] = save;
2780 * Internal form of close. Decrement reference count on file structure.
2781 * Note: td may be NULL when closing a file that was being passed in a
2785 closef(struct file *fp, struct thread *td)
2789 struct filedesc_to_leader *fdtol;
2790 struct filedesc *fdp;
2795 * POSIX record locking dictates that any close releases ALL
2796 * locks owned by this process. This is handled by setting
2797 * a flag in the unlock to free ONLY locks obeying POSIX
2798 * semantics, and not to free BSD-style file locks.
2799 * If the descriptor was in a message, POSIX-style locks
2800 * aren't passed with the descriptor, and the thread pointer
2801 * will be NULL. Callers should be careful only to pass a
2802 * NULL thread pointer when there really is no owning
2803 * context that might have locks, or the locks will be
2806 if (fp->f_type == DTYPE_VNODE) {
2808 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2809 lf.l_whence = SEEK_SET;
2812 lf.l_type = F_UNLCK;
2813 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2814 F_UNLCK, &lf, F_POSIX);
2816 fdtol = td->td_proc->p_fdtol;
2817 if (fdtol != NULL) {
2819 * Handle special case where file descriptor table is
2820 * shared between multiple process leaders.
2822 fdp = td->td_proc->p_fd;
2823 FILEDESC_XLOCK(fdp);
2824 for (fdtol = fdtol->fdl_next;
2825 fdtol != td->td_proc->p_fdtol;
2826 fdtol = fdtol->fdl_next) {
2827 if ((fdtol->fdl_leader->p_flag &
2830 fdtol->fdl_holdcount++;
2831 FILEDESC_XUNLOCK(fdp);
2832 lf.l_whence = SEEK_SET;
2835 lf.l_type = F_UNLCK;
2837 (void) VOP_ADVLOCK(vp,
2838 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2840 FILEDESC_XLOCK(fdp);
2841 fdtol->fdl_holdcount--;
2842 if (fdtol->fdl_holdcount == 0 &&
2843 fdtol->fdl_wakeup != 0) {
2844 fdtol->fdl_wakeup = 0;
2848 FILEDESC_XUNLOCK(fdp);
2851 return (fdrop_close(fp, td));
2855 * Hack for file descriptor passing code.
2858 closef_nothread(struct file *fp)
2865 * Initialize the file pointer with the specified properties.
2867 * The ops are set with release semantics to be certain that the flags, type,
2868 * and data are visible when ops is. This is to prevent ops methods from being
2869 * called with bad data.
2872 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2877 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2881 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2883 fp->f_seqcount[UIO_READ] = 1;
2884 fp->f_seqcount[UIO_WRITE] = 1;
2885 finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2890 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2891 struct file **fpp, struct filecaps *havecapsp)
2893 struct filedescent *fde;
2896 FILEDESC_LOCK_ASSERT(fdp);
2898 fde = fdeget_locked(fdp, fd);
2905 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2910 if (havecapsp != NULL)
2911 filecaps_copy(&fde->fde_caps, havecapsp, true);
2913 *fpp = fde->fde_file;
2921 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2922 struct file **fpp, struct filecaps *havecapsp)
2924 struct filedesc *fdp = td->td_proc->p_fd;
2926 #ifndef CAPABILITIES
2927 error = fget_unlocked(fdp, fd, needrightsp, fpp);
2928 if (havecapsp != NULL && error == 0)
2929 filecaps_fill(havecapsp);
2936 error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2940 if (havecapsp != NULL) {
2941 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2942 havecapsp, false)) {
2948 if (!fd_modified(fdp, fd, seq))
2957 FILEDESC_SLOCK(fdp);
2958 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2959 if (error == 0 && !fhold(*fpp))
2961 FILEDESC_SUNLOCK(fdp);
2968 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2970 const struct filedescent *fde;
2971 const struct fdescenttbl *fdt;
2972 struct filedesc *fdp;
2975 const cap_rights_t *haverights;
2976 cap_rights_t rights;
2979 VFS_SMR_ASSERT_ENTERED();
2981 rights = *ndp->ni_rightsneeded;
2982 cap_rights_set_one(&rights, CAP_LOOKUP);
2984 fdp = curproc->p_fd;
2985 fdt = fdp->fd_files;
2986 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2988 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
2989 fde = &fdt->fdt_ofiles[fd];
2990 haverights = cap_rights_fde_inline(fde);
2992 if (__predict_false(fp == NULL))
2994 if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2996 *fsearch = ((fp->f_flag & FSEARCH) != 0);
2998 if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
3001 if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
3005 * Use an acquire barrier to force re-reading of fdt so it is
3006 * refreshed for verification.
3008 atomic_thread_fence_acq();
3009 fdt = fdp->fd_files;
3010 if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3013 * If file descriptor doesn't have all rights,
3014 * all lookups relative to it must also be
3015 * strictly relative.
3017 * Not yet supported by fast path.
3020 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
3021 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
3022 ndp->ni_filecaps.fc_nioctls != -1) {
3024 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
3034 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
3036 const struct fdescenttbl *fdt;
3037 struct filedesc *fdp;
3041 VFS_SMR_ASSERT_ENTERED();
3043 fdp = curproc->p_fd;
3044 fdt = fdp->fd_files;
3045 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3047 fp = fdt->fdt_ofiles[fd].fde_file;
3048 if (__predict_false(fp == NULL))
3050 *fsearch = ((fp->f_flag & FSEARCH) != 0);
3052 if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
3056 * Use an acquire barrier to force re-reading of fdt so it is
3057 * refreshed for verification.
3059 atomic_thread_fence_acq();
3060 fdt = fdp->fd_files;
3061 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3063 filecaps_fill(&ndp->ni_filecaps);
3070 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3071 struct file **fpp, seqc_t *seqp)
3074 const struct filedescent *fde;
3076 const struct fdescenttbl *fdt;
3080 cap_rights_t haverights;
3084 fdt = fdp->fd_files;
3085 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3088 * Fetch the descriptor locklessly. We avoid fdrop() races by
3089 * never raising a refcount above 0. To accomplish this we have
3090 * to use a cmpset loop rather than an atomic_add. The descriptor
3091 * must be re-verified once we acquire a reference to be certain
3092 * that the identity is still correct and we did not lose a race
3093 * due to preemption.
3097 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3098 fde = &fdt->fdt_ofiles[fd];
3099 haverights = *cap_rights_fde_inline(fde);
3101 if (!seqc_consistent(fd_seqc(fdt, fd), seq))
3104 fp = fdt->fdt_ofiles[fd].fde_file;
3109 error = cap_check_inline(&haverights, needrightsp);
3113 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3115 * Force a reload. Other thread could reallocate the
3116 * table before this fd was closed, so it is possible
3117 * that there is a stale fp pointer in cached version.
3119 fdt = atomic_load_ptr(&fdp->fd_files);
3123 * Use an acquire barrier to force re-reading of fdt so it is
3124 * refreshed for verification.
3126 atomic_thread_fence_acq();
3127 fdt = fdp->fd_files;
3129 if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
3131 if (fp == fdt->fdt_ofiles[fd].fde_file)
3134 fdrop(fp, curthread);
3146 * See the comments in fget_unlocked_seq for an explanation of how this works.
3148 * This is a simplified variant which bails out to the aforementioned routine
3149 * if anything goes wrong. In practice this only happens when userspace is
3150 * racing with itself.
3153 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3157 const struct filedescent *fde;
3159 const struct fdescenttbl *fdt;
3163 const cap_rights_t *haverights;
3166 fdt = fdp->fd_files;
3167 if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3170 seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3171 fde = &fdt->fdt_ofiles[fd];
3172 haverights = cap_rights_fde_inline(fde);
3175 fp = fdt->fdt_ofiles[fd].fde_file;
3177 if (__predict_false(fp == NULL))
3180 if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3183 if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3187 * Use an acquire barrier to force re-reading of fdt so it is
3188 * refreshed for verification.
3190 atomic_thread_fence_acq();
3191 fdt = fdp->fd_files;
3193 if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3195 if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3201 fdrop(fp, curthread);
3203 return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
3207 * Translate fd -> file when the caller guarantees the file descriptor table
3208 * can't be changed by others.
3210 * Note this does not mean the file object itself is only visible to the caller,
3211 * merely that it wont disappear without having to be referenced.
3213 * Must be paired with fput_only_user.
3217 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3220 const struct filedescent *fde;
3221 const struct fdescenttbl *fdt;
3222 const cap_rights_t *haverights;
3226 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3228 if (__predict_false(fd >= fdp->fd_nfiles))
3231 fdt = fdp->fd_files;
3232 fde = &fdt->fdt_ofiles[fd];
3234 if (__predict_false(fp == NULL))
3236 MPASS(refcount_load(&fp->f_count) > 0);
3237 haverights = cap_rights_fde_inline(fde);
3238 error = cap_check_inline(haverights, needrightsp);
3239 if (__predict_false(error != 0))
3246 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3251 MPASS(FILEDESC_IS_ONLY_USER(fdp));
3253 if (__predict_false(fd >= fdp->fd_nfiles))
3256 fp = fdp->fd_ofiles[fd].fde_file;
3257 if (__predict_false(fp == NULL))
3260 MPASS(refcount_load(&fp->f_count) > 0);
3267 * Extract the file pointer associated with the specified descriptor for the
3268 * current user process.
3270 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3273 * File's rights will be checked against the capability rights mask.
3275 * If an error occurred the non-zero error is returned and *fpp is set to
3276 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
3277 * responsible for fdrop().
3280 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3281 cap_rights_t *needrightsp)
3283 struct filedesc *fdp;
3288 fdp = td->td_proc->p_fd;
3289 error = fget_unlocked(fdp, fd, needrightsp, &fp);
3290 if (__predict_false(error != 0))
3292 if (__predict_false(fp->f_ops == &badfileops)) {
3298 * FREAD and FWRITE failure return EBADF as per POSIX.
3304 if ((fp->f_flag & flags) == 0)
3308 if (fp->f_ops != &path_fileops &&
3309 ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3310 (fp->f_flag & FWRITE) != 0))
3316 KASSERT(0, ("wrong flags"));
3329 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3332 return (_fget(td, fd, fpp, 0, rightsp));
3336 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3340 #ifndef CAPABILITIES
3341 error = _fget(td, fd, fpp, 0, rightsp);
3342 if (maxprotp != NULL)
3343 *maxprotp = VM_PROT_ALL;
3346 cap_rights_t fdrights;
3347 struct filedesc *fdp;
3352 fdp = td->td_proc->p_fd;
3353 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3355 error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3356 if (__predict_false(error != 0))
3358 if (__predict_false(fp->f_ops == &badfileops)) {
3362 if (maxprotp != NULL)
3363 fdrights = *cap_rights(fdp, fd);
3364 if (!fd_modified(fdp, fd, seq))
3370 * If requested, convert capability rights to access flags.
3372 if (maxprotp != NULL)
3373 *maxprotp = cap_rights_to_vmprot(&fdrights);
3380 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3383 return (_fget(td, fd, fpp, FREAD, rightsp));
3387 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3390 return (_fget(td, fd, fpp, FWRITE, rightsp));
3394 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3397 struct filedesc *fdp = td->td_proc->p_fd;
3398 #ifndef CAPABILITIES
3399 return (fget_unlocked(fdp, fd, rightsp, fpp));
3406 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3408 error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3411 error = cap_fcntl_check(fdp, fd, needfcntl);
3412 if (!fd_modified(fdp, fd, seq))
3426 * Like fget() but loads the underlying vnode, or returns an error if the
3427 * descriptor does not represent a vnode. Note that pipes use vnodes but
3428 * never have VM objects. The returned vnode will be vref()'d.
3430 * XXX: what about the unused flags ?
3433 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3440 error = _fget(td, fd, &fp, flags, needrightsp);
3443 if (fp->f_vnode == NULL) {
3455 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3458 return (_fgetvp(td, fd, 0, rightsp, vpp));
3462 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3463 struct filecaps *havecaps, struct vnode **vpp)
3465 struct filecaps caps;
3469 error = fget_cap(td, fd, needrightsp, &fp, &caps);
3472 if (fp->f_ops == &badfileops) {
3476 if (fp->f_vnode == NULL) {
3488 filecaps_free(&caps);
3494 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3497 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3501 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3504 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3509 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3513 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3518 * Handle the last reference to a file being closed.
3520 * Without the noinline attribute clang keeps inlining the func thorough this
3521 * file when fdrop is used.
3524 _fdrop(struct file *fp, struct thread *td)
3530 count = refcount_load(&fp->f_count);
3532 panic("fdrop: fp %p count %d", fp, count);
3534 error = fo_close(fp, td);
3535 atomic_subtract_int(&openfiles, 1);
3537 free(fp->f_advice, M_FADVISE);
3538 uma_zfree(file_zone, fp);
3544 * Apply an advisory lock on a file descriptor.
3546 * Just attempt to get a record lock of the requested type on the entire file
3547 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3549 #ifndef _SYS_SYSPROTO_H_
3557 sys_flock(struct thread *td, struct flock_args *uap)
3564 error = fget(td, uap->fd, &cap_flock_rights, &fp);
3568 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
3571 if (fp->f_ops == &path_fileops) {
3577 lf.l_whence = SEEK_SET;
3580 if (uap->how & LOCK_UN) {
3581 lf.l_type = F_UNLCK;
3582 atomic_clear_int(&fp->f_flag, FHASLOCK);
3583 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3586 if (uap->how & LOCK_EX)
3587 lf.l_type = F_WRLCK;
3588 else if (uap->how & LOCK_SH)
3589 lf.l_type = F_RDLCK;
3594 atomic_set_int(&fp->f_flag, FHASLOCK);
3595 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3596 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3602 * Duplicate the specified descriptor to a free descriptor.
3605 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3606 int openerror, int *indxp)
3608 struct filedescent *newfde, *oldfde;
3613 KASSERT(openerror == ENODEV || openerror == ENXIO,
3614 ("unexpected error %d in %s", openerror, __func__));
3617 * If the to-be-dup'd fd number is greater than the allowed number
3618 * of file descriptors, or the fd to be dup'd has already been
3619 * closed, then reject.
3621 FILEDESC_XLOCK(fdp);
3622 if ((fp = fget_locked(fdp, dfd)) == NULL) {
3623 FILEDESC_XUNLOCK(fdp);
3627 error = fdalloc(td, 0, &indx);
3629 FILEDESC_XUNLOCK(fdp);
3634 * There are two cases of interest here.
3636 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3638 * For ENXIO steal away the file structure from (dfd) and store it in
3639 * (indx). (dfd) is effectively closed by this operation.
3641 switch (openerror) {
3644 * Check that the mode the file is being opened for is a
3645 * subset of the mode of the existing descriptor.
3647 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3648 fdunused(fdp, indx);
3649 FILEDESC_XUNLOCK(fdp);
3653 fdunused(fdp, indx);
3654 FILEDESC_XUNLOCK(fdp);
3657 newfde = &fdp->fd_ofiles[indx];
3658 oldfde = &fdp->fd_ofiles[dfd];
3659 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3661 seqc_write_begin(&newfde->fde_seqc);
3663 memcpy(newfde, oldfde, fde_change_size);
3664 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3667 seqc_write_end(&newfde->fde_seqc);
3672 * Steal away the file pointer from dfd and stuff it into indx.
3674 newfde = &fdp->fd_ofiles[indx];
3675 oldfde = &fdp->fd_ofiles[dfd];
3677 seqc_write_begin(&newfde->fde_seqc);
3679 memcpy(newfde, oldfde, fde_change_size);
3680 oldfde->fde_file = NULL;
3683 seqc_write_end(&newfde->fde_seqc);
3687 FILEDESC_XUNLOCK(fdp);
3693 * This sysctl determines if we will allow a process to chroot(2) if it
3694 * has a directory open:
3695 * 0: disallowed for all processes.
3696 * 1: allowed for processes that were not already chroot(2)'ed.
3697 * 2: allowed for all processes.
3700 static int chroot_allow_open_directories = 1;
3702 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3703 &chroot_allow_open_directories, 0,
3704 "Allow a process to chroot(2) if it has a directory open");
3707 * Helper function for raised chroot(2) security function: Refuse if
3708 * any filedescriptors are open directories.
3711 chroot_refuse_vdir_fds(struct filedesc *fdp)
3717 FILEDESC_LOCK_ASSERT(fdp);
3719 lastfile = fdlastfile(fdp);
3720 for (fd = 0; fd <= lastfile; fd++) {
3721 fp = fget_locked(fdp, fd);
3724 if (fp->f_type == DTYPE_VNODE) {
3726 if (vp->v_type == VDIR)
3734 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3737 if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3738 vrefact(oldpwd->pwd_cdir);
3739 newpwd->pwd_cdir = oldpwd->pwd_cdir;
3742 if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3743 vrefact(oldpwd->pwd_rdir);
3744 newpwd->pwd_rdir = oldpwd->pwd_rdir;
3747 if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3748 vrefact(oldpwd->pwd_jdir);
3749 newpwd->pwd_jdir = oldpwd->pwd_jdir;
3754 pwd_hold_pwddesc(struct pwddesc *pdp)
3758 PWDDESC_ASSERT_XLOCKED(pdp);
3759 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3761 refcount_acquire(&pwd->pwd_refcount);
3766 pwd_hold_smr(struct pwd *pwd)
3770 if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3777 pwd_hold(struct thread *td)
3779 struct pwddesc *pdp;
3782 pdp = td->td_proc->p_pd;
3785 pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3786 if (pwd_hold_smr(pwd)) {
3792 pwd = pwd_hold_pwddesc(pdp);
3794 PWDDESC_XUNLOCK(pdp);
3799 pwd_hold_proc(struct proc *p)
3801 struct pwddesc *pdp;
3804 PROC_ASSERT_HELD(p);
3811 pwd = pwd_hold_pwddesc(pdp);
3813 PWDDESC_XUNLOCK(pdp);
3823 pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3824 bzero(pwd, sizeof(*pwd));
3825 refcount_init(&pwd->pwd_refcount, 1);
3830 pwd_drop(struct pwd *pwd)
3833 if (!refcount_release(&pwd->pwd_refcount))
3836 if (pwd->pwd_cdir != NULL)
3837 vrele(pwd->pwd_cdir);
3838 if (pwd->pwd_rdir != NULL)
3839 vrele(pwd->pwd_rdir);
3840 if (pwd->pwd_jdir != NULL)
3841 vrele(pwd->pwd_jdir);
3842 uma_zfree_smr(pwd_zone, pwd);
3846 * The caller is responsible for invoking priv_check() and
3847 * mac_vnode_check_chroot() to authorize this operation.
3850 pwd_chroot(struct thread *td, struct vnode *vp)
3852 struct pwddesc *pdp;
3853 struct filedesc *fdp;
3854 struct pwd *newpwd, *oldpwd;
3857 fdp = td->td_proc->p_fd;
3858 pdp = td->td_proc->p_pd;
3859 newpwd = pwd_alloc();
3860 FILEDESC_SLOCK(fdp);
3862 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3863 if (chroot_allow_open_directories == 0 ||
3864 (chroot_allow_open_directories == 1 &&
3865 oldpwd->pwd_rdir != rootvnode)) {
3866 error = chroot_refuse_vdir_fds(fdp);
3867 FILEDESC_SUNLOCK(fdp);
3869 PWDDESC_XUNLOCK(pdp);
3874 FILEDESC_SUNLOCK(fdp);
3878 newpwd->pwd_rdir = vp;
3879 if (oldpwd->pwd_jdir == NULL) {
3881 newpwd->pwd_jdir = vp;
3883 pwd_fill(oldpwd, newpwd);
3884 pwd_set(pdp, newpwd);
3885 PWDDESC_XUNLOCK(pdp);
3891 pwd_chdir(struct thread *td, struct vnode *vp)
3893 struct pwddesc *pdp;
3894 struct pwd *newpwd, *oldpwd;
3896 VNPASS(vp->v_usecount > 0, vp);
3898 newpwd = pwd_alloc();
3899 pdp = td->td_proc->p_pd;
3901 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3902 newpwd->pwd_cdir = vp;
3903 pwd_fill(oldpwd, newpwd);
3904 pwd_set(pdp, newpwd);
3905 PWDDESC_XUNLOCK(pdp);
3910 * jail_attach(2) changes both root and working directories.
3913 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
3915 struct pwddesc *pdp;
3916 struct filedesc *fdp;
3917 struct pwd *newpwd, *oldpwd;
3920 fdp = td->td_proc->p_fd;
3921 pdp = td->td_proc->p_pd;
3922 newpwd = pwd_alloc();
3923 FILEDESC_SLOCK(fdp);
3925 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3926 error = chroot_refuse_vdir_fds(fdp);
3927 FILEDESC_SUNLOCK(fdp);
3929 PWDDESC_XUNLOCK(pdp);
3935 newpwd->pwd_rdir = vp;
3937 newpwd->pwd_cdir = vp;
3938 if (oldpwd->pwd_jdir == NULL) {
3940 newpwd->pwd_jdir = vp;
3942 pwd_fill(oldpwd, newpwd);
3943 pwd_set(pdp, newpwd);
3944 PWDDESC_XUNLOCK(pdp);
3950 pwd_ensure_dirs(void)
3952 struct pwddesc *pdp;
3953 struct pwd *oldpwd, *newpwd;
3955 pdp = curproc->p_pd;
3957 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3958 if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3959 PWDDESC_XUNLOCK(pdp);
3962 PWDDESC_XUNLOCK(pdp);
3964 newpwd = pwd_alloc();
3966 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3967 pwd_fill(oldpwd, newpwd);
3968 if (newpwd->pwd_cdir == NULL) {
3970 newpwd->pwd_cdir = rootvnode;
3972 if (newpwd->pwd_rdir == NULL) {
3974 newpwd->pwd_rdir = rootvnode;
3976 pwd_set(pdp, newpwd);
3977 PWDDESC_XUNLOCK(pdp);
3982 pwd_set_rootvnode(void)
3984 struct pwddesc *pdp;
3985 struct pwd *oldpwd, *newpwd;
3987 pdp = curproc->p_pd;
3989 newpwd = pwd_alloc();
3991 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3993 newpwd->pwd_cdir = rootvnode;
3995 newpwd->pwd_rdir = rootvnode;
3996 pwd_fill(oldpwd, newpwd);
3997 pwd_set(pdp, newpwd);
3998 PWDDESC_XUNLOCK(pdp);
4003 * Scan all active processes and prisons to see if any of them have a current
4004 * or root directory of `olddp'. If so, replace them with the new mount point.
4007 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
4009 struct pwddesc *pdp;
4010 struct pwd *newpwd, *oldpwd;
4015 if (vrefcnt(olddp) == 1)
4018 newpwd = pwd_alloc();
4019 sx_slock(&allproc_lock);
4020 FOREACH_PROC_IN_SYSTEM(p) {
4027 oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4028 if (oldpwd == NULL ||
4029 (oldpwd->pwd_cdir != olddp &&
4030 oldpwd->pwd_rdir != olddp &&
4031 oldpwd->pwd_jdir != olddp)) {
4032 PWDDESC_XUNLOCK(pdp);
4036 if (oldpwd->pwd_cdir == olddp) {
4038 newpwd->pwd_cdir = newdp;
4040 if (oldpwd->pwd_rdir == olddp) {
4042 newpwd->pwd_rdir = newdp;
4044 if (oldpwd->pwd_jdir == olddp) {
4046 newpwd->pwd_jdir = newdp;
4048 pwd_fill(oldpwd, newpwd);
4049 pwd_set(pdp, newpwd);
4050 PWDDESC_XUNLOCK(pdp);
4053 newpwd = pwd_alloc();
4055 sx_sunlock(&allproc_lock);
4057 if (rootvnode == olddp) {
4062 mtx_lock(&prison0.pr_mtx);
4063 if (prison0.pr_root == olddp) {
4065 prison0.pr_root = newdp;
4068 mtx_unlock(&prison0.pr_mtx);
4069 sx_slock(&allprison_lock);
4070 TAILQ_FOREACH(pr, &allprison, pr_list) {
4071 mtx_lock(&pr->pr_mtx);
4072 if (pr->pr_root == olddp) {
4074 pr->pr_root = newdp;
4077 mtx_unlock(&pr->pr_mtx);
4079 sx_sunlock(&allprison_lock);
4084 struct filedesc_to_leader *
4085 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
4087 struct filedesc_to_leader *fdtol;
4089 fdtol = malloc(sizeof(struct filedesc_to_leader),
4090 M_FILEDESC_TO_LEADER, M_WAITOK);
4091 fdtol->fdl_refcount = 1;
4092 fdtol->fdl_holdcount = 0;
4093 fdtol->fdl_wakeup = 0;
4094 fdtol->fdl_leader = leader;
4096 FILEDESC_XLOCK(fdp);
4097 fdtol->fdl_next = old->fdl_next;
4098 fdtol->fdl_prev = old;
4099 old->fdl_next = fdtol;
4100 fdtol->fdl_next->fdl_prev = fdtol;
4101 FILEDESC_XUNLOCK(fdp);
4103 fdtol->fdl_next = fdtol;
4104 fdtol->fdl_prev = fdtol;
4110 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
4113 struct filedesc *fdp;
4115 int count, off, minoff;
4121 if (*(int *)arg1 != 0)
4124 fdp = curproc->p_fd;
4126 FILEDESC_SLOCK(fdp);
4128 off = NDSLOT(fdp->fd_nfiles - 1);
4129 for (minoff = NDSLOT(0); off >= minoff; --off)
4130 count += bitcountl(map[off]);
4131 FILEDESC_SUNLOCK(fdp);
4133 return (SYSCTL_OUT(req, &count, sizeof(count)));
4136 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
4137 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
4138 "Number of open file descriptors");
4141 * Get file structures globally.
4144 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
4147 struct filedesc *fdp;
4150 int error, n, lastfile;
4152 error = sysctl_wire_old_buffer(req, 0);
4155 if (req->oldptr == NULL) {
4157 sx_slock(&allproc_lock);
4158 FOREACH_PROC_IN_SYSTEM(p) {
4160 if (p->p_state == PRS_NEW) {
4168 /* overestimates sparse tables. */
4169 n += fdp->fd_nfiles;
4172 sx_sunlock(&allproc_lock);
4173 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4176 bzero(&xf, sizeof(xf));
4177 xf.xf_size = sizeof(xf);
4178 sx_slock(&allproc_lock);
4179 FOREACH_PROC_IN_SYSTEM(p) {
4181 if (p->p_state == PRS_NEW) {
4185 if (p_cansee(req->td, p) != 0) {
4189 xf.xf_pid = p->p_pid;
4190 xf.xf_uid = p->p_ucred->cr_uid;
4195 FILEDESC_SLOCK(fdp);
4196 lastfile = fdlastfile(fdp);
4197 for (n = 0; refcount_load(&fdp->fd_refcnt) > 0 && n <= lastfile;
4199 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4202 xf.xf_file = (uintptr_t)fp;
4203 xf.xf_data = (uintptr_t)fp->f_data;
4204 xf.xf_vnode = (uintptr_t)fp->f_vnode;
4205 xf.xf_type = (uintptr_t)fp->f_type;
4206 xf.xf_count = refcount_load(&fp->f_count);
4208 xf.xf_offset = foffset_get(fp);
4209 xf.xf_flag = fp->f_flag;
4210 error = SYSCTL_OUT(req, &xf, sizeof(xf));
4214 FILEDESC_SUNLOCK(fdp);
4219 sx_sunlock(&allproc_lock);
4223 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4224 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4226 #ifdef KINFO_FILE_SIZE
4227 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4231 xlate_fflags(int fflags)
4233 static const struct {
4236 } fflags_table[] = {
4237 { FAPPEND, KF_FLAG_APPEND },
4238 { FASYNC, KF_FLAG_ASYNC },
4239 { FFSYNC, KF_FLAG_FSYNC },
4240 { FHASLOCK, KF_FLAG_HASLOCK },
4241 { FNONBLOCK, KF_FLAG_NONBLOCK },
4242 { FREAD, KF_FLAG_READ },
4243 { FWRITE, KF_FLAG_WRITE },
4244 { O_CREAT, KF_FLAG_CREAT },
4245 { O_DIRECT, KF_FLAG_DIRECT },
4246 { O_EXCL, KF_FLAG_EXCL },
4247 { O_EXEC, KF_FLAG_EXEC },
4248 { O_EXLOCK, KF_FLAG_EXLOCK },
4249 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4250 { O_SHLOCK, KF_FLAG_SHLOCK },
4251 { O_TRUNC, KF_FLAG_TRUNC }
4257 for (i = 0; i < nitems(fflags_table); i++)
4258 if (fflags & fflags_table[i].fflag)
4259 kflags |= fflags_table[i].kf_fflag;
4263 /* Trim unused data from kf_path by truncating the structure size. */
4265 pack_kinfo(struct kinfo_file *kif)
4268 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4269 strlen(kif->kf_path) + 1;
4270 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4274 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4275 struct kinfo_file *kif, struct filedesc *fdp, int flags)
4279 bzero(kif, sizeof(*kif));
4281 /* Set a default type to allow for empty fill_kinfo() methods. */
4282 kif->kf_type = KF_TYPE_UNKNOWN;
4283 kif->kf_flags = xlate_fflags(fp->f_flag);
4284 if (rightsp != NULL)
4285 kif->kf_cap_rights = *rightsp;
4287 cap_rights_init_zero(&kif->kf_cap_rights);
4289 kif->kf_ref_count = refcount_load(&fp->f_count);
4290 kif->kf_offset = foffset_get(fp);
4293 * This may drop the filedesc lock, so the 'fp' cannot be
4294 * accessed after this call.
4296 error = fo_fill_kinfo(fp, kif, fdp);
4298 kif->kf_status |= KF_ATTR_VALID;
4299 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4302 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4306 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4307 struct kinfo_file *kif, int flags)
4311 bzero(kif, sizeof(*kif));
4313 kif->kf_type = KF_TYPE_VNODE;
4314 error = vn_fill_kinfo_vnode(vp, kif);
4316 kif->kf_status |= KF_ATTR_VALID;
4317 kif->kf_flags = xlate_fflags(fflags);
4318 cap_rights_init_zero(&kif->kf_cap_rights);
4320 kif->kf_ref_count = -1;
4321 kif->kf_offset = -1;
4322 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4325 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4329 struct export_fd_buf {
4330 struct filedesc *fdp;
4331 struct pwddesc *pdp;
4334 struct kinfo_file kif;
4339 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4341 struct kinfo_file *kif;
4344 if (efbuf->remainder != -1) {
4345 if (efbuf->remainder < kif->kf_structsize) {
4346 /* Terminate export. */
4347 efbuf->remainder = 0;
4350 efbuf->remainder -= kif->kf_structsize;
4352 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
4356 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4357 struct export_fd_buf *efbuf)
4361 if (efbuf->remainder == 0)
4363 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4365 FILEDESC_SUNLOCK(efbuf->fdp);
4366 error = export_kinfo_to_sb(efbuf);
4367 FILEDESC_SLOCK(efbuf->fdp);
4372 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4373 struct export_fd_buf *efbuf)
4377 if (efbuf->remainder == 0)
4379 if (efbuf->pdp != NULL)
4380 PWDDESC_XUNLOCK(efbuf->pdp);
4381 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4382 error = export_kinfo_to_sb(efbuf);
4383 if (efbuf->pdp != NULL)
4384 PWDDESC_XLOCK(efbuf->pdp);
4389 * Store a process file descriptor information to sbuf.
4391 * Takes a locked proc as argument, and returns with the proc unlocked.
4394 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
4398 struct filedesc *fdp;
4399 struct pwddesc *pdp;
4400 struct export_fd_buf *efbuf;
4401 struct vnode *cttyvp, *textvp, *tracevp;
4403 int error, i, lastfile;
4404 cap_rights_t rights;
4406 PROC_LOCK_ASSERT(p, MA_OWNED);
4409 tracevp = ktr_get_tracevp(p, true);
4411 textvp = p->p_textvp;
4414 /* Controlling tty. */
4416 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4417 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4424 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4428 efbuf->remainder = maxlen;
4429 efbuf->flags = flags;
4430 if (tracevp != NULL)
4431 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
4434 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
4436 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
4439 if (pdp == NULL || fdp == NULL)
4444 pwd = pwd_hold_pwddesc(pdp);
4446 /* working directory */
4447 if (pwd->pwd_cdir != NULL) {
4448 vrefact(pwd->pwd_cdir);
4449 export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD,
4452 /* root directory */
4453 if (pwd->pwd_rdir != NULL) {
4454 vrefact(pwd->pwd_rdir);
4455 export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT,
4458 /* jail directory */
4459 if (pwd->pwd_jdir != NULL) {
4460 vrefact(pwd->pwd_jdir);
4461 export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL,
4465 PWDDESC_XUNLOCK(pdp);
4468 FILEDESC_SLOCK(fdp);
4469 lastfile = fdlastfile(fdp);
4470 for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4471 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4474 rights = *cap_rights(fdp, i);
4475 #else /* !CAPABILITIES */
4476 rights = cap_no_rights;
4479 * Create sysctl entry. It is OK to drop the filedesc
4480 * lock inside of export_file_to_sb() as we will
4481 * re-validate and re-evaluate its properties when the
4484 error = export_file_to_sb(fp, i, &rights, efbuf);
4485 if (error != 0 || efbuf->remainder == 0)
4488 FILEDESC_SUNLOCK(fdp);
4494 free(efbuf, M_TEMP);
4498 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
4501 * Get per-process file descriptors for use by procstat(1), et al.
4504 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4510 int error, error2, *name;
4518 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4519 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4520 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4525 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4526 error = kern_proc_filedesc_out(p, &sb, maxlen,
4527 KERN_FILEDESC_PACK_KINFO);
4528 error2 = sbuf_finish(&sb);
4530 return (error != 0 ? error : error2);
4533 #ifdef COMPAT_FREEBSD7
4534 #ifdef KINFO_OFILE_SIZE
4535 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4539 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4542 okif->kf_structsize = sizeof(*okif);
4543 okif->kf_type = kif->kf_type;
4544 okif->kf_fd = kif->kf_fd;
4545 okif->kf_ref_count = kif->kf_ref_count;
4546 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4547 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4548 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4549 okif->kf_offset = kif->kf_offset;
4550 if (kif->kf_type == KF_TYPE_VNODE)
4551 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4553 okif->kf_vnode_type = KF_VTYPE_VNON;
4554 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4555 if (kif->kf_type == KF_TYPE_SOCKET) {
4556 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4557 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4558 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4559 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4560 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4562 okif->kf_sa_local.ss_family = AF_UNSPEC;
4563 okif->kf_sa_peer.ss_family = AF_UNSPEC;
4568 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4569 struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4574 PWDDESC_XUNLOCK(pdp);
4575 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4576 kinfo_to_okinfo(kif, okif);
4577 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4583 * Get per-process file descriptors for use by procstat(1), et al.
4586 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4588 struct kinfo_ofile *okif;
4589 struct kinfo_file *kif;
4590 struct filedesc *fdp;
4591 struct pwddesc *pdp;
4594 int error, i, lastfile, *name;
4603 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4610 if (fdp == NULL || pdp == NULL) {
4615 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4616 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4618 pwd = pwd_hold_pwddesc(pdp);
4620 if (pwd->pwd_cdir != NULL)
4621 export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4623 if (pwd->pwd_rdir != NULL)
4624 export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4626 if (pwd->pwd_jdir != NULL)
4627 export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4630 PWDDESC_XUNLOCK(pdp);
4633 FILEDESC_SLOCK(fdp);
4634 lastfile = fdlastfile(fdp);
4635 for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4636 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4638 export_file_to_kinfo(fp, i, NULL, kif, fdp,
4639 KERN_FILEDESC_PACK_KINFO);
4640 FILEDESC_SUNLOCK(fdp);
4641 kinfo_to_okinfo(kif, okif);
4642 error = SYSCTL_OUT(req, okif, sizeof(*okif));
4643 FILEDESC_SLOCK(fdp);
4647 FILEDESC_SUNLOCK(fdp);
4655 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4656 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4657 "Process ofiledesc entries");
4658 #endif /* COMPAT_FREEBSD7 */
4661 vntype_to_kinfo(int vtype)
4666 } vtypes_table[] = {
4667 { VBAD, KF_VTYPE_VBAD },
4668 { VBLK, KF_VTYPE_VBLK },
4669 { VCHR, KF_VTYPE_VCHR },
4670 { VDIR, KF_VTYPE_VDIR },
4671 { VFIFO, KF_VTYPE_VFIFO },
4672 { VLNK, KF_VTYPE_VLNK },
4673 { VNON, KF_VTYPE_VNON },
4674 { VREG, KF_VTYPE_VREG },
4675 { VSOCK, KF_VTYPE_VSOCK }
4680 * Perform vtype translation.
4682 for (i = 0; i < nitems(vtypes_table); i++)
4683 if (vtypes_table[i].vtype == vtype)
4684 return (vtypes_table[i].kf_vtype);
4686 return (KF_VTYPE_UNKNOWN);
4689 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4690 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4691 "Process filedesc entries");
4694 * Store a process current working directory information to sbuf.
4696 * Takes a locked proc as argument, and returns with the proc unlocked.
4699 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
4701 struct pwddesc *pdp;
4703 struct export_fd_buf *efbuf;
4707 PROC_LOCK_ASSERT(p, MA_OWNED);
4714 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4717 efbuf->remainder = maxlen;
4720 pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4721 cdir = pwd->pwd_cdir;
4726 error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4728 PWDDESC_XUNLOCK(pdp);
4730 free(efbuf, M_TEMP);
4735 * Get per-process current working directory.
4738 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4744 int error, error2, *name;
4752 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4753 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4754 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4759 maxlen = req->oldptr != NULL ? req->oldlen : -1;
4760 error = kern_proc_cwd_out(p, &sb, maxlen);
4761 error2 = sbuf_finish(&sb);
4763 return (error != 0 ? error : error2);
4766 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4767 sysctl_kern_proc_cwd, "Process current working directory");
4771 * For the purposes of debugging, generate a human-readable string for the
4775 file_type_to_name(short type)
4803 case DTYPE_PROCDESC:
4807 case DTYPE_LINUXTFD:
4815 * For the purposes of debugging, identify a process (if any, perhaps one of
4816 * many) that references the passed file in its file descriptor array. Return
4819 static struct proc *
4820 file_to_first_proc(struct file *fp)
4822 struct filedesc *fdp;
4826 FOREACH_PROC_IN_SYSTEM(p) {
4827 if (p->p_state == PRS_NEW)
4832 for (n = 0; n < fdp->fd_nfiles; n++) {
4833 if (fp == fdp->fd_ofiles[n].fde_file)
4841 db_print_file(struct file *fp, int header)
4843 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4847 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4848 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4849 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4851 p = file_to_first_proc(fp);
4852 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4853 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4854 fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4855 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4860 DB_SHOW_COMMAND(file, db_show_file)
4865 db_printf("usage: show file <addr>\n");
4868 fp = (struct file *)addr;
4869 db_print_file(fp, 1);
4872 DB_SHOW_COMMAND(files, db_show_files)
4874 struct filedesc *fdp;
4881 FOREACH_PROC_IN_SYSTEM(p) {
4882 if (p->p_state == PRS_NEW)
4884 if ((fdp = p->p_fd) == NULL)
4886 for (n = 0; n < fdp->fd_nfiles; ++n) {
4887 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4889 db_print_file(fp, header);
4896 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4897 &maxfilesperproc, 0, "Maximum files allowed open per process");
4899 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4900 &maxfiles, 0, "Maximum number of files");
4902 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4903 &openfiles, 0, "System-wide number of open files");
4907 filelistinit(void *dummy)
4910 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4911 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4912 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4913 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4914 pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4915 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4917 * XXXMJG this is a temporary hack due to boot ordering issues against
4920 vfs_smr = uma_zone_get_smr(pwd_zone);
4921 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4923 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4925 /*-------------------------------------------------------------------*/
4928 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4929 int flags, struct thread *td)
4936 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4944 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4952 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4960 badfo_kqfilter(struct file *fp, struct knote *kn)
4967 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4975 badfo_close(struct file *fp, struct thread *td)
4982 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4990 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4998 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4999 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5007 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
5013 struct fileops badfileops = {
5014 .fo_read = badfo_readwrite,
5015 .fo_write = badfo_readwrite,
5016 .fo_truncate = badfo_truncate,
5017 .fo_ioctl = badfo_ioctl,
5018 .fo_poll = badfo_poll,
5019 .fo_kqfilter = badfo_kqfilter,
5020 .fo_stat = badfo_stat,
5021 .fo_close = badfo_close,
5022 .fo_chmod = badfo_chmod,
5023 .fo_chown = badfo_chown,
5024 .fo_sendfile = badfo_sendfile,
5025 .fo_fill_kinfo = badfo_fill_kinfo,
5029 path_poll(struct file *fp, int events, struct ucred *active_cred,
5036 path_close(struct file *fp, struct thread *td)
5038 MPASS(fp->f_type == DTYPE_VNODE);
5039 fp->f_ops = &badfileops;
5044 struct fileops path_fileops = {
5045 .fo_read = badfo_readwrite,
5046 .fo_write = badfo_readwrite,
5047 .fo_truncate = badfo_truncate,
5048 .fo_ioctl = badfo_ioctl,
5049 .fo_poll = path_poll,
5050 .fo_kqfilter = vn_kqfilter_opath,
5051 .fo_stat = vn_statfile,
5052 .fo_close = path_close,
5053 .fo_chmod = badfo_chmod,
5054 .fo_chown = badfo_chown,
5055 .fo_sendfile = badfo_sendfile,
5056 .fo_fill_kinfo = vn_fill_kinfo,
5057 .fo_flags = DFLAG_PASSABLE,
5061 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
5062 int flags, struct thread *td)
5065 return (EOPNOTSUPP);
5069 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
5077 invfo_ioctl(struct file *fp, u_long com, void *data,
5078 struct ucred *active_cred, struct thread *td)
5085 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
5089 return (poll_no_poll(events));
5093 invfo_kqfilter(struct file *fp, struct knote *kn)
5100 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5108 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5116 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5117 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5124 /*-------------------------------------------------------------------*/
5127 * File Descriptor pseudo-device driver (/dev/fd/).
5129 * Opening minor device N dup()s the file (if any) connected to file
5130 * descriptor N belonging to the calling process. Note that this driver
5131 * consists of only the ``open()'' routine, because all subsequent
5132 * references to this file will be direct to the other driver.
5134 * XXX: we could give this one a cloning event handler if necessary.
5139 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
5143 * XXX Kludge: set curthread->td_dupfd to contain the value of the
5144 * the file descriptor being sought for duplication. The error
5145 * return ensures that the vnode for this device will be released
5146 * by vn_open. Open will detect this special error and take the
5147 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
5148 * will simply report the error.
5150 td->td_dupfd = dev2unit(dev);
5154 static struct cdevsw fildesc_cdevsw = {
5155 .d_version = D_VERSION,
5161 fildesc_drvinit(void *unused)
5165 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
5166 UID_ROOT, GID_WHEEL, 0666, "fd/0");
5167 make_dev_alias(dev, "stdin");
5168 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
5169 UID_ROOT, GID_WHEEL, 0666, "fd/1");
5170 make_dev_alias(dev, "stdout");
5171 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
5172 UID_ROOT, GID_WHEEL, 0666, "fd/2");
5173 make_dev_alias(dev, "stderr");
5176 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);