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>
66 #include <sys/protosw.h>
67 #include <sys/racct.h>
68 #include <sys/resourcevar.h>
70 #include <sys/signalvar.h>
74 #include <sys/syscallsubr.h>
75 #include <sys/sysctl.h>
76 #include <sys/sysproto.h>
77 #include <sys/unistd.h>
79 #include <sys/vnode.h>
81 #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_FILEDESC_TO_LEADER, "filedesc_to_leader",
95 "file desc to leader structures");
96 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
97 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
99 MALLOC_DECLARE(M_FADVISE);
101 static __read_mostly uma_zone_t file_zone;
102 static __read_mostly uma_zone_t filedesc0_zone;
104 static int closefp(struct filedesc *fdp, int fd, struct file *fp,
105 struct thread *td, int holdleaders);
106 static int fd_first_free(struct filedesc *fdp, int low, int size);
107 static int fd_last_used(struct filedesc *fdp, int size);
108 static void fdgrowtable(struct filedesc *fdp, int nfd);
109 static void fdgrowtable_exp(struct filedesc *fdp, int nfd);
110 static void fdunused(struct filedesc *fdp, int fd);
111 static void fdused(struct filedesc *fdp, int fd);
112 static int getmaxfd(struct thread *td);
113 static u_long *filecaps_copy_prep(const struct filecaps *src);
114 static void filecaps_copy_finish(const struct filecaps *src,
115 struct filecaps *dst, u_long *ioctls);
116 static u_long *filecaps_free_prep(struct filecaps *fcaps);
117 static void filecaps_free_finish(u_long *ioctls);
122 * - An array of open file descriptors (fd_ofiles)
123 * - An array of file flags (fd_ofileflags)
124 * - A bitmap recording which descriptors are in use (fd_map)
126 * A process starts out with NDFILE descriptors. The value of NDFILE has
127 * been selected based the historical limit of 20 open files, and an
128 * assumption that the majority of processes, especially short-lived
129 * processes like shells, will never need more.
131 * If this initial allocation is exhausted, a larger descriptor table and
132 * map are allocated dynamically, and the pointers in the process's struct
133 * filedesc are updated to point to those. This is repeated every time
134 * the process runs out of file descriptors (provided it hasn't hit its
137 * Since threads may hold references to individual descriptor table
138 * entries, the tables are never freed. Instead, they are placed on a
139 * linked list and freed only when the struct filedesc is released.
142 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
143 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
144 #define NDSLOT(x) ((x) / NDENTRIES)
145 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
146 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
149 * SLIST entry used to keep track of ofiles which must be reclaimed when
153 struct fdescenttbl *ft_table;
154 SLIST_ENTRY(freetable) ft_next;
158 * Initial allocation: a filedesc structure + the head of SLIST used to
159 * keep track of old ofiles + enough space for NDFILE descriptors.
162 struct fdescenttbl0 {
164 struct filedescent fdt_ofiles[NDFILE];
168 struct filedesc fd_fd;
169 SLIST_HEAD(, freetable) fd_free;
170 struct fdescenttbl0 fd_dfiles;
171 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
175 * Descriptor management.
177 volatile int __exclusive_cache_line openfiles; /* actual number of open files */
178 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
179 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
182 * If low >= size, just return low. Otherwise find the first zero bit in the
183 * given bitmap, starting at low and not exceeding size - 1. Return size if
187 fd_first_free(struct filedesc *fdp, int low, int size)
189 NDSLOTTYPE *map = fdp->fd_map;
197 if (low % NDENTRIES) {
198 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
199 if ((mask &= ~map[off]) != 0UL)
200 return (off * NDENTRIES + ffsl(mask) - 1);
203 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
204 if (map[off] != ~0UL)
205 return (off * NDENTRIES + ffsl(~map[off]) - 1);
210 * Find the highest non-zero bit in the given bitmap, starting at 0 and
211 * not exceeding size - 1. Return -1 if not found.
214 fd_last_used(struct filedesc *fdp, int size)
216 NDSLOTTYPE *map = fdp->fd_map;
221 if (size % NDENTRIES) {
222 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
223 if ((mask &= map[off]) != 0)
224 return (off * NDENTRIES + flsl(mask) - 1);
227 for (minoff = NDSLOT(0); off >= minoff; --off)
229 return (off * NDENTRIES + flsl(map[off]) - 1);
234 fdisused(struct filedesc *fdp, int fd)
237 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
238 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
240 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
244 * Mark a file descriptor as used.
247 fdused_init(struct filedesc *fdp, int fd)
250 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
252 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
256 fdused(struct filedesc *fdp, int fd)
259 FILEDESC_XLOCK_ASSERT(fdp);
261 fdused_init(fdp, fd);
262 if (fd > fdp->fd_lastfile)
263 fdp->fd_lastfile = fd;
264 if (fd == fdp->fd_freefile)
265 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
269 * Mark a file descriptor as unused.
272 fdunused(struct filedesc *fdp, int fd)
275 FILEDESC_XLOCK_ASSERT(fdp);
277 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
278 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
279 ("fd=%d is still in use", fd));
281 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
282 if (fd < fdp->fd_freefile)
283 fdp->fd_freefile = fd;
284 if (fd == fdp->fd_lastfile)
285 fdp->fd_lastfile = fd_last_used(fdp, fd);
289 * Free a file descriptor.
291 * Avoid some work if fdp is about to be destroyed.
294 fdefree_last(struct filedescent *fde)
297 filecaps_free(&fde->fde_caps);
301 fdfree(struct filedesc *fdp, int fd)
303 struct filedescent *fde;
305 fde = &fdp->fd_ofiles[fd];
307 seq_write_begin(&fde->fde_seq);
309 fde->fde_file = NULL;
311 seq_write_end(&fde->fde_seq);
318 pwd_ensure_dirs(void)
320 struct filedesc *fdp;
324 if (fdp->fd_cdir == NULL) {
325 fdp->fd_cdir = rootvnode;
328 if (fdp->fd_rdir == NULL) {
329 fdp->fd_rdir = rootvnode;
332 FILEDESC_XUNLOCK(fdp);
336 * System calls on descriptors.
338 #ifndef _SYS_SYSPROTO_H_
339 struct getdtablesize_args {
345 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
352 min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc);
354 PROC_LOCK(td->td_proc);
355 lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
356 PROC_UNLOCK(td->td_proc);
357 if (lim < td->td_retval[0])
358 td->td_retval[0] = lim;
364 * Duplicate a file descriptor to a particular value.
366 * Note: keep in mind that a potential race condition exists when closing
367 * descriptors from a shared descriptor table (via rfork).
369 #ifndef _SYS_SYSPROTO_H_
377 sys_dup2(struct thread *td, struct dup2_args *uap)
380 return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
384 * Duplicate a file descriptor.
386 #ifndef _SYS_SYSPROTO_H_
393 sys_dup(struct thread *td, struct dup_args *uap)
396 return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
400 * The file control system call.
402 #ifndef _SYS_SYSPROTO_H_
411 sys_fcntl(struct thread *td, struct fcntl_args *uap)
414 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
418 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
432 * Convert old flock structure to new.
434 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
435 fl.l_start = ofl.l_start;
436 fl.l_len = ofl.l_len;
437 fl.l_pid = ofl.l_pid;
438 fl.l_type = ofl.l_type;
439 fl.l_whence = ofl.l_whence;
453 arg1 = (intptr_t)&fl;
459 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
460 arg1 = (intptr_t)&fl;
468 error = kern_fcntl(td, fd, newcmd, arg1);
471 if (cmd == F_OGETLK) {
472 ofl.l_start = fl.l_start;
473 ofl.l_len = fl.l_len;
474 ofl.l_pid = fl.l_pid;
475 ofl.l_type = fl.l_type;
476 ofl.l_whence = fl.l_whence;
477 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
478 } else if (cmd == F_GETLK) {
479 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
485 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
487 struct filedesc *fdp;
489 struct file *fp, *fp2;
490 struct filedescent *fde;
507 error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
510 case F_DUPFD_CLOEXEC:
512 error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
517 error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
520 case F_DUP2FD_CLOEXEC:
522 error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
528 fde = fdeget_locked(fdp, fd);
531 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
534 FILEDESC_SUNLOCK(fdp);
540 fde = fdeget_locked(fdp, fd);
542 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
543 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
546 FILEDESC_XUNLOCK(fdp);
550 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
553 td->td_retval[0] = OFLAGS(fp->f_flag);
558 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
562 tmp = flg = fp->f_flag;
564 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
565 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
566 tmp = fp->f_flag & FNONBLOCK;
567 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
572 tmp = fp->f_flag & FASYNC;
573 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
578 atomic_clear_int(&fp->f_flag, FNONBLOCK);
580 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
585 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
588 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
590 td->td_retval[0] = tmp;
595 error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
599 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
604 error = priv_check(td, PRIV_NFS_LOCKD);
612 /* FALLTHROUGH F_SETLK */
616 flp = (struct flock *)arg;
617 if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
622 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL);
625 if (fp->f_type != DTYPE_VNODE) {
631 if (flp->l_whence == SEEK_CUR) {
632 foffset = foffset_get(fp);
635 foffset > OFF_MAX - flp->l_start)) {
640 flp->l_start += foffset;
644 switch (flp->l_type) {
646 if ((fp->f_flag & FREAD) == 0) {
650 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
651 PROC_LOCK(p->p_leader);
652 p->p_leader->p_flag |= P_ADVLOCK;
653 PROC_UNLOCK(p->p_leader);
655 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
659 if ((fp->f_flag & FWRITE) == 0) {
663 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
664 PROC_LOCK(p->p_leader);
665 p->p_leader->p_flag |= P_ADVLOCK;
666 PROC_UNLOCK(p->p_leader);
668 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
672 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
676 if (flg != F_REMOTE) {
680 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
681 F_UNLCKSYS, flp, flg);
687 if (error != 0 || flp->l_type == F_UNLCK ||
688 flp->l_type == F_UNLCKSYS) {
694 * Check for a race with close.
696 * The vnode is now advisory locked (or unlocked, but this case
697 * is not really important) as the caller requested.
698 * We had to drop the filedesc lock, so we need to recheck if
699 * the descriptor is still valid, because if it was closed
700 * in the meantime we need to remove advisory lock from the
701 * vnode - close on any descriptor leading to an advisory
702 * locked vnode, removes that lock.
703 * We will return 0 on purpose in that case, as the result of
704 * successful advisory lock might have been externally visible
705 * already. This is fine - effectively we pretend to the caller
706 * that the closing thread was a bit slower and that the
707 * advisory lock succeeded before the close.
709 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2, NULL);
715 flp->l_whence = SEEK_SET;
718 flp->l_type = F_UNLCK;
719 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
720 F_UNLCK, flp, F_POSIX);
727 error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL);
730 if (fp->f_type != DTYPE_VNODE) {
735 flp = (struct flock *)arg;
736 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
737 flp->l_type != F_UNLCK) {
742 if (flp->l_whence == SEEK_CUR) {
743 foffset = foffset_get(fp);
744 if ((flp->l_start > 0 &&
745 foffset > OFF_MAX - flp->l_start) ||
747 foffset < OFF_MIN - flp->l_start)) {
752 flp->l_start += foffset;
755 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
761 arg = arg ? 128 * 1024: 0;
764 error = fget_unlocked(fdp, fd, &cap_no_rights, &fp, NULL);
767 if (fp->f_type != DTYPE_VNODE) {
773 if (vp->v_type != VREG) {
780 * Exclusive lock synchronizes against f_seqcount reads and
781 * writes in sequential_heuristic().
783 error = vn_lock(vp, LK_EXCLUSIVE);
789 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
790 arg = MIN(arg, INT_MAX - bsize + 1);
791 fp->f_seqcount = MIN(IO_SEQMAX,
792 (arg + bsize - 1) / bsize);
793 atomic_set_int(&fp->f_flag, FRDAHEAD);
795 atomic_clear_int(&fp->f_flag, FRDAHEAD);
809 getmaxfd(struct thread *td)
812 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
816 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
819 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
821 struct filedesc *fdp;
822 struct filedescent *oldfde, *newfde;
825 u_long *oioctls, *nioctls;
831 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
832 MPASS(mode < FDDUP_LASTMODE);
835 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
838 * Verify we have a valid descriptor to dup from and possibly to
839 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
840 * return EINVAL when the new descriptor is out of bounds.
845 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
846 maxfd = getmaxfd(td);
848 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
852 if (fget_locked(fdp, old) == NULL)
854 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
855 td->td_retval[0] = new;
856 if (flags & FDDUP_FLAG_CLOEXEC)
857 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
862 oldfde = &fdp->fd_ofiles[old];
863 if (!fhold(oldfde->fde_file))
867 * If the caller specified a file descriptor, make sure the file
868 * table is large enough to hold it, and grab it. Otherwise, just
869 * allocate a new descriptor the usual way.
874 if ((error = fdalloc(td, new, &new)) != 0) {
875 fdrop(oldfde->fde_file, td);
879 case FDDUP_MUSTREPLACE:
880 /* Target file descriptor must exist. */
881 if (fget_locked(fdp, new) == NULL) {
882 fdrop(oldfde->fde_file, td);
887 if (new >= fdp->fd_nfiles) {
889 * The resource limits are here instead of e.g.
890 * fdalloc(), because the file descriptor table may be
891 * shared between processes, so we can't really use
892 * racct_add()/racct_sub(). Instead of counting the
893 * number of actually allocated descriptors, just put
894 * the limit on the size of the file descriptor table.
899 error = racct_set(p, RACCT_NOFILE, new + 1);
903 fdrop(oldfde->fde_file, td);
908 fdgrowtable_exp(fdp, new + 1);
910 if (!fdisused(fdp, new))
914 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
917 KASSERT(old != new, ("new fd is same as old"));
919 newfde = &fdp->fd_ofiles[new];
920 delfp = newfde->fde_file;
922 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
925 * Duplicate the source descriptor.
928 seq_write_begin(&newfde->fde_seq);
930 oioctls = filecaps_free_prep(&newfde->fde_caps);
931 memcpy(newfde, oldfde, fde_change_size);
932 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
934 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
935 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
937 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
939 seq_write_end(&newfde->fde_seq);
941 filecaps_free_finish(oioctls);
942 td->td_retval[0] = new;
947 (void) closefp(fdp, new, delfp, td, 1);
948 FILEDESC_UNLOCK_ASSERT(fdp);
951 FILEDESC_XUNLOCK(fdp);
958 * If sigio is on the list associated with a process or process group,
959 * disable signalling from the device, remove sigio from the list and
963 funsetown(struct sigio **sigiop)
975 *(sigio->sio_myref) = NULL;
976 if ((sigio)->sio_pgid < 0) {
977 struct pgrp *pg = (sigio)->sio_pgrp;
979 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
983 struct proc *p = (sigio)->sio_proc;
985 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
990 crfree(sigio->sio_ucred);
991 free(sigio, M_SIGIO);
995 * Free a list of sigio structures.
996 * We only need to lock the SIGIO_LOCK because we have made ourselves
997 * inaccessible to callers of fsetown and therefore do not need to lock
998 * the proc or pgrp struct for the list manipulation.
1001 funsetownlst(struct sigiolst *sigiolst)
1005 struct sigio *sigio;
1007 sigio = SLIST_FIRST(sigiolst);
1014 * Every entry of the list should belong
1015 * to a single proc or pgrp.
1017 if (sigio->sio_pgid < 0) {
1018 pg = sigio->sio_pgrp;
1019 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
1020 } else /* if (sigio->sio_pgid > 0) */ {
1021 p = sigio->sio_proc;
1022 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1026 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1027 *(sigio->sio_myref) = NULL;
1029 KASSERT(sigio->sio_pgid < 0,
1030 ("Proc sigio in pgrp sigio list"));
1031 KASSERT(sigio->sio_pgrp == pg,
1032 ("Bogus pgrp in sigio list"));
1034 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1037 } else /* if (p != NULL) */ {
1038 KASSERT(sigio->sio_pgid > 0,
1039 ("Pgrp sigio in proc sigio list"));
1040 KASSERT(sigio->sio_proc == p,
1041 ("Bogus proc in sigio list"));
1043 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1048 crfree(sigio->sio_ucred);
1049 free(sigio, M_SIGIO);
1056 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1058 * After permission checking, add a sigio structure to the sigio list for
1059 * the process or process group.
1062 fsetown(pid_t pgid, struct sigio **sigiop)
1066 struct sigio *sigio;
1076 /* Allocate and fill in the new sigio out of locks. */
1077 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1078 sigio->sio_pgid = pgid;
1079 sigio->sio_ucred = crhold(curthread->td_ucred);
1080 sigio->sio_myref = sigiop;
1082 sx_slock(&proctree_lock);
1091 * Policy - Don't allow a process to FSETOWN a process
1092 * in another session.
1094 * Remove this test to allow maximum flexibility or
1095 * restrict FSETOWN to the current process or process
1096 * group for maximum safety.
1099 if (proc->p_session != curthread->td_proc->p_session) {
1105 } else /* if (pgid < 0) */ {
1106 pgrp = pgfind(-pgid);
1114 * Policy - Don't allow a process to FSETOWN a process
1115 * in another session.
1117 * Remove this test to allow maximum flexibility or
1118 * restrict FSETOWN to the current process or process
1119 * group for maximum safety.
1121 if (pgrp->pg_session != curthread->td_proc->p_session) {
1132 * Since funsetownlst() is called without the proctree
1133 * locked, we need to check for P_WEXIT.
1134 * XXX: is ESRCH correct?
1136 if ((proc->p_flag & P_WEXIT) != 0) {
1141 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1142 sigio->sio_proc = proc;
1146 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1147 sigio->sio_pgrp = pgrp;
1150 sx_sunlock(&proctree_lock);
1157 sx_sunlock(&proctree_lock);
1158 crfree(sigio->sio_ucred);
1159 free(sigio, M_SIGIO);
1164 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1167 fgetown(struct sigio **sigiop)
1172 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1178 * Function drops the filedesc lock on return.
1181 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1186 FILEDESC_XLOCK_ASSERT(fdp);
1189 if (td->td_proc->p_fdtol != NULL) {
1191 * Ask fdfree() to sleep to ensure that all relevant
1192 * process leaders can be traversed in closef().
1194 fdp->fd_holdleaderscount++;
1201 * We now hold the fp reference that used to be owned by the
1202 * descriptor array. We have to unlock the FILEDESC *AFTER*
1203 * knote_fdclose to prevent a race of the fd getting opened, a knote
1204 * added, and deleteing a knote for the new fd.
1206 knote_fdclose(td, fd);
1209 * We need to notify mqueue if the object is of type mqueue.
1211 if (fp->f_type == DTYPE_MQUEUE)
1212 mq_fdclose(td, fd, fp);
1213 FILEDESC_XUNLOCK(fdp);
1215 error = closef(fp, td);
1217 FILEDESC_XLOCK(fdp);
1218 fdp->fd_holdleaderscount--;
1219 if (fdp->fd_holdleaderscount == 0 &&
1220 fdp->fd_holdleaderswakeup != 0) {
1221 fdp->fd_holdleaderswakeup = 0;
1222 wakeup(&fdp->fd_holdleaderscount);
1224 FILEDESC_XUNLOCK(fdp);
1230 * Close a file descriptor.
1232 #ifndef _SYS_SYSPROTO_H_
1239 sys_close(struct thread *td, struct close_args *uap)
1242 return (kern_close(td, uap->fd));
1246 kern_close(struct thread *td, int fd)
1248 struct filedesc *fdp;
1251 fdp = td->td_proc->p_fd;
1253 AUDIT_SYSCLOSE(td, fd);
1255 FILEDESC_XLOCK(fdp);
1256 if ((fp = fget_locked(fdp, fd)) == NULL) {
1257 FILEDESC_XUNLOCK(fdp);
1262 /* closefp() drops the FILEDESC lock for us. */
1263 return (closefp(fdp, fd, fp, td, 1));
1267 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1269 struct filedesc *fdp;
1273 fdp = td->td_proc->p_fd;
1274 FILEDESC_SLOCK(fdp);
1277 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1278 * open should not be a usage error. From a close_range() perspective,
1279 * close_range(3, ~0U, 0) in the same scenario should also likely not
1280 * be a usage error as all fd above 3 are in-fact already closed.
1282 if (highfd < lowfd) {
1288 * If fdp->fd_lastfile == -1, we're dealing with either a fresh file
1289 * table or one in which every fd has been closed. Just return
1290 * successful; there's nothing left to do.
1292 if (fdp->fd_lastfile == -1)
1294 /* Clamped to [lowfd, fd_lastfile] */
1295 highfd = MIN(highfd, fdp->fd_lastfile);
1296 for (fd = lowfd; fd <= highfd; fd++) {
1297 if (fdp->fd_ofiles[fd].fde_file != NULL) {
1298 FILEDESC_SUNLOCK(fdp);
1299 (void)kern_close(td, fd);
1300 FILEDESC_SLOCK(fdp);
1304 FILEDESC_SUNLOCK(fdp);
1308 #ifndef _SYS_SYSPROTO_H_
1309 struct close_range_args {
1316 sys_close_range(struct thread *td, struct close_range_args *uap)
1319 /* No flags currently defined */
1320 if (uap->flags != 0)
1322 return (kern_close_range(td, uap->lowfd, uap->highfd));
1326 * Close open file descriptors.
1328 #ifndef _SYS_SYSPROTO_H_
1329 struct closefrom_args {
1335 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1339 AUDIT_ARG_FD(uap->lowfd);
1342 * Treat negative starting file descriptor values identical to
1343 * closefrom(0) which closes all files.
1345 lowfd = MAX(0, uap->lowfd);
1346 return (kern_close_range(td, lowfd, ~0U));
1349 #if defined(COMPAT_43)
1351 * Return status information about a file descriptor.
1353 #ifndef _SYS_SYSPROTO_H_
1354 struct ofstat_args {
1361 ofstat(struct thread *td, struct ofstat_args *uap)
1367 error = kern_fstat(td, uap->fd, &ub);
1370 error = copyout(&oub, uap->sb, sizeof(oub));
1374 #endif /* COMPAT_43 */
1376 #if defined(COMPAT_FREEBSD11)
1378 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1381 struct freebsd11_stat osb;
1384 error = kern_fstat(td, uap->fd, &sb);
1387 error = freebsd11_cvtstat(&sb, &osb);
1389 error = copyout(&osb, uap->sb, sizeof(osb));
1392 #endif /* COMPAT_FREEBSD11 */
1395 * Return status information about a file descriptor.
1397 #ifndef _SYS_SYSPROTO_H_
1405 sys_fstat(struct thread *td, struct fstat_args *uap)
1410 error = kern_fstat(td, uap->fd, &ub);
1412 error = copyout(&ub, uap->sb, sizeof(ub));
1417 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1424 error = fget(td, fd, &cap_fstat_rights, &fp);
1428 AUDIT_ARG_FILE(td->td_proc, fp);
1430 error = fo_stat(fp, sbp, td->td_ucred, td);
1432 #ifdef __STAT_TIME_T_EXT
1434 sbp->st_atim_ext = 0;
1435 sbp->st_mtim_ext = 0;
1436 sbp->st_ctim_ext = 0;
1437 sbp->st_btim_ext = 0;
1441 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1447 #if defined(COMPAT_FREEBSD11)
1449 * Return status information about a file descriptor.
1451 #ifndef _SYS_SYSPROTO_H_
1452 struct freebsd11_nfstat_args {
1459 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1465 error = kern_fstat(td, uap->fd, &ub);
1467 freebsd11_cvtnstat(&ub, &nub);
1468 error = copyout(&nub, uap->sb, sizeof(nub));
1472 #endif /* COMPAT_FREEBSD11 */
1475 * Return pathconf information about a file descriptor.
1477 #ifndef _SYS_SYSPROTO_H_
1478 struct fpathconf_args {
1485 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1490 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1492 td->td_retval[0] = value;
1497 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1503 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1507 if (name == _PC_ASYNC_IO) {
1508 *valuep = _POSIX_ASYNCHRONOUS_IO;
1513 vn_lock(vp, LK_SHARED | LK_RETRY);
1514 error = VOP_PATHCONF(vp, name, valuep);
1516 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1517 if (name != _PC_PIPE_BUF) {
1532 * Initialize filecaps structure.
1535 filecaps_init(struct filecaps *fcaps)
1538 bzero(fcaps, sizeof(*fcaps));
1539 fcaps->fc_nioctls = -1;
1543 * Copy filecaps structure allocating memory for ioctls array if needed.
1545 * The last parameter indicates whether the fdtable is locked. If it is not and
1546 * ioctls are encountered, copying fails and the caller must lock the table.
1548 * Note that if the table was not locked, the caller has to check the relevant
1549 * sequence counter to determine whether the operation was successful.
1552 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1556 if (src->fc_ioctls != NULL && !locked)
1558 memcpy(dst, src, sizeof(*src));
1559 if (src->fc_ioctls == NULL)
1562 KASSERT(src->fc_nioctls > 0,
1563 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1565 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1566 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1567 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1572 filecaps_copy_prep(const struct filecaps *src)
1577 if (src->fc_ioctls == NULL)
1580 KASSERT(src->fc_nioctls > 0,
1581 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1583 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1584 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1589 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1595 if (src->fc_ioctls == NULL) {
1596 MPASS(ioctls == NULL);
1600 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1601 dst->fc_ioctls = ioctls;
1602 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1606 * Move filecaps structure to the new place and clear the old place.
1609 filecaps_move(struct filecaps *src, struct filecaps *dst)
1613 bzero(src, sizeof(*src));
1617 * Fill the given filecaps structure with full rights.
1620 filecaps_fill(struct filecaps *fcaps)
1623 CAP_ALL(&fcaps->fc_rights);
1624 fcaps->fc_ioctls = NULL;
1625 fcaps->fc_nioctls = -1;
1626 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1630 * Free memory allocated within filecaps structure.
1633 filecaps_free(struct filecaps *fcaps)
1636 free(fcaps->fc_ioctls, M_FILECAPS);
1637 bzero(fcaps, sizeof(*fcaps));
1641 filecaps_free_prep(struct filecaps *fcaps)
1645 ioctls = fcaps->fc_ioctls;
1646 bzero(fcaps, sizeof(*fcaps));
1651 filecaps_free_finish(u_long *ioctls)
1654 free(ioctls, M_FILECAPS);
1658 * Validate the given filecaps structure.
1661 filecaps_validate(const struct filecaps *fcaps, const char *func)
1664 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1665 ("%s: invalid rights", func));
1666 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1667 ("%s: invalid fcntls", func));
1668 KASSERT(fcaps->fc_fcntls == 0 ||
1669 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1670 ("%s: fcntls without CAP_FCNTL", func));
1671 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1672 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1673 ("%s: invalid ioctls", func));
1674 KASSERT(fcaps->fc_nioctls == 0 ||
1675 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1676 ("%s: ioctls without CAP_IOCTL", func));
1680 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1684 FILEDESC_XLOCK_ASSERT(fdp);
1686 nfd1 = fdp->fd_nfiles * 2;
1689 fdgrowtable(fdp, nfd1);
1693 * Grow the file table to accommodate (at least) nfd descriptors.
1696 fdgrowtable(struct filedesc *fdp, int nfd)
1698 struct filedesc0 *fdp0;
1699 struct freetable *ft;
1700 struct fdescenttbl *ntable;
1701 struct fdescenttbl *otable;
1702 int nnfiles, onfiles;
1703 NDSLOTTYPE *nmap, *omap;
1706 * If lastfile is -1 this struct filedesc was just allocated and we are
1707 * growing it to accommodate for the one we are going to copy from. There
1708 * is no need to have a lock on this one as it's not visible to anyone.
1710 if (fdp->fd_lastfile != -1)
1711 FILEDESC_XLOCK_ASSERT(fdp);
1713 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1715 /* save old values */
1716 onfiles = fdp->fd_nfiles;
1717 otable = fdp->fd_files;
1720 /* compute the size of the new table */
1721 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1722 if (nnfiles <= onfiles)
1723 /* the table is already large enough */
1727 * Allocate a new table. We need enough space for the number of
1728 * entries, file entries themselves and the struct freetable we will use
1729 * when we decommission the table and place it on the freelist.
1730 * We place the struct freetable in the middle so we don't have
1731 * to worry about padding.
1733 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1734 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1735 sizeof(struct freetable),
1736 M_FILEDESC, M_ZERO | M_WAITOK);
1737 /* copy the old data */
1738 ntable->fdt_nfiles = nnfiles;
1739 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1740 onfiles * sizeof(ntable->fdt_ofiles[0]));
1743 * Allocate a new map only if the old is not large enough. It will
1744 * grow at a slower rate than the table as it can map more
1745 * entries than the table can hold.
1747 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1748 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1750 /* copy over the old data and update the pointer */
1751 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1756 * Make sure that ntable is correctly initialized before we replace
1757 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1760 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1763 * Do not free the old file table, as some threads may still
1764 * reference entries within it. Instead, place it on a freelist
1765 * which will be processed when the struct filedesc is released.
1767 * Note that if onfiles == NDFILE, we're dealing with the original
1768 * static allocation contained within (struct filedesc0 *)fdp,
1769 * which must not be freed.
1771 if (onfiles > NDFILE) {
1772 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1773 fdp0 = (struct filedesc0 *)fdp;
1774 ft->ft_table = otable;
1775 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1778 * The map does not have the same possibility of threads still
1779 * holding references to it. So always free it as long as it
1780 * does not reference the original static allocation.
1782 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1783 free(omap, M_FILEDESC);
1787 * Allocate a file descriptor for the process.
1790 fdalloc(struct thread *td, int minfd, int *result)
1792 struct proc *p = td->td_proc;
1793 struct filedesc *fdp = p->p_fd;
1794 int fd, maxfd, allocfd;
1799 FILEDESC_XLOCK_ASSERT(fdp);
1801 if (fdp->fd_freefile > minfd)
1802 minfd = fdp->fd_freefile;
1804 maxfd = getmaxfd(td);
1807 * Search the bitmap for a free descriptor starting at minfd.
1808 * If none is found, grow the file table.
1810 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1813 if (fd >= fdp->fd_nfiles) {
1814 allocfd = min(fd * 2, maxfd);
1818 error = racct_set(p, RACCT_NOFILE, allocfd);
1825 * fd is already equal to first free descriptor >= minfd, so
1826 * we only need to grow the table and we are done.
1828 fdgrowtable_exp(fdp, allocfd);
1832 * Perform some sanity checks, then mark the file descriptor as
1833 * used and return it to the caller.
1835 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1836 ("invalid descriptor %d", fd));
1837 KASSERT(!fdisused(fdp, fd),
1838 ("fd_first_free() returned non-free descriptor"));
1839 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1840 ("file descriptor isn't free"));
1847 * Allocate n file descriptors for the process.
1850 fdallocn(struct thread *td, int minfd, int *fds, int n)
1852 struct proc *p = td->td_proc;
1853 struct filedesc *fdp = p->p_fd;
1856 FILEDESC_XLOCK_ASSERT(fdp);
1858 for (i = 0; i < n; i++)
1859 if (fdalloc(td, 0, &fds[i]) != 0)
1863 for (i--; i >= 0; i--)
1864 fdunused(fdp, fds[i]);
1872 * Create a new open file structure and allocate a file descriptor for the
1873 * process that refers to it. We add one reference to the file for the
1874 * descriptor table and one reference for resultfp. This is to prevent us
1875 * being preempted and the entry in the descriptor table closed after we
1876 * release the FILEDESC lock.
1879 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1880 struct filecaps *fcaps)
1885 error = falloc_noinstall(td, &fp);
1887 return (error); /* no reference held on error */
1889 error = finstall(td, fp, &fd, flags, fcaps);
1891 fdrop(fp, td); /* one reference (fp only) */
1895 if (resultfp != NULL)
1896 *resultfp = fp; /* copy out result */
1898 fdrop(fp, td); /* release local reference */
1900 if (resultfd != NULL)
1907 * Create a new open file structure without allocating a file descriptor.
1910 falloc_noinstall(struct thread *td, struct file **resultfp)
1913 int maxuserfiles = maxfiles - (maxfiles / 20);
1915 static struct timeval lastfail;
1918 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1920 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1921 if ((openfiles_new >= maxuserfiles &&
1922 priv_check(td, PRIV_MAXFILES) != 0) ||
1923 openfiles_new >= maxfiles) {
1924 atomic_subtract_int(&openfiles, 1);
1925 if (ppsratecheck(&lastfail, &curfail, 1)) {
1926 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1927 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1931 fp = uma_zalloc(file_zone, M_WAITOK);
1932 bzero(fp, sizeof(*fp));
1933 refcount_init(&fp->f_count, 1);
1934 fp->f_cred = crhold(td->td_ucred);
1935 fp->f_ops = &badfileops;
1941 * Install a file in a file descriptor table.
1944 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1945 struct filecaps *fcaps)
1947 struct filedescent *fde;
1951 filecaps_validate(fcaps, __func__);
1952 FILEDESC_XLOCK_ASSERT(fdp);
1954 fde = &fdp->fd_ofiles[fd];
1956 seq_write_begin(&fde->fde_seq);
1959 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1961 filecaps_move(fcaps, &fde->fde_caps);
1963 filecaps_fill(&fde->fde_caps);
1965 seq_write_end(&fde->fde_seq);
1970 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1971 struct filecaps *fcaps)
1973 struct filedesc *fdp = td->td_proc->p_fd;
1980 FILEDESC_XLOCK(fdp);
1981 if ((error = fdalloc(td, 0, fd))) {
1982 FILEDESC_XUNLOCK(fdp);
1986 _finstall(fdp, fp, *fd, flags, fcaps);
1987 FILEDESC_XUNLOCK(fdp);
1992 * Build a new filedesc structure from another.
1993 * Copy the current, root, and jail root vnode references.
1995 * If fdp is not NULL, return with it shared locked.
1998 fdinit(struct filedesc *fdp, bool prepfiles)
2000 struct filedesc0 *newfdp0;
2001 struct filedesc *newfdp;
2003 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2004 newfdp = &newfdp0->fd_fd;
2006 /* Create the file descriptor table. */
2007 FILEDESC_LOCK_INIT(newfdp);
2008 refcount_init(&newfdp->fd_refcnt, 1);
2009 refcount_init(&newfdp->fd_holdcnt, 1);
2010 newfdp->fd_cmask = CMASK;
2011 newfdp->fd_map = newfdp0->fd_dmap;
2012 newfdp->fd_lastfile = -1;
2013 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2014 newfdp->fd_files->fdt_nfiles = NDFILE;
2019 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
2020 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
2022 FILEDESC_SLOCK(fdp);
2023 newfdp->fd_cdir = fdp->fd_cdir;
2024 if (newfdp->fd_cdir)
2025 vrefact(newfdp->fd_cdir);
2026 newfdp->fd_rdir = fdp->fd_rdir;
2027 if (newfdp->fd_rdir)
2028 vrefact(newfdp->fd_rdir);
2029 newfdp->fd_jdir = fdp->fd_jdir;
2030 if (newfdp->fd_jdir)
2031 vrefact(newfdp->fd_jdir);
2034 FILEDESC_SUNLOCK(fdp);
2036 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
2037 FILEDESC_SUNLOCK(fdp);
2038 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
2039 FILEDESC_SLOCK(fdp);
2046 static struct filedesc *
2047 fdhold(struct proc *p)
2049 struct filedesc *fdp;
2051 PROC_LOCK_ASSERT(p, MA_OWNED);
2054 refcount_acquire(&fdp->fd_holdcnt);
2059 fddrop(struct filedesc *fdp)
2062 if (fdp->fd_holdcnt > 1) {
2063 if (refcount_release(&fdp->fd_holdcnt) == 0)
2067 FILEDESC_LOCK_DESTROY(fdp);
2068 uma_zfree(filedesc0_zone, fdp);
2072 * Share a filedesc structure.
2075 fdshare(struct filedesc *fdp)
2078 refcount_acquire(&fdp->fd_refcnt);
2083 * Unshare a filedesc structure, if necessary by making a copy
2086 fdunshare(struct thread *td)
2088 struct filedesc *tmp;
2089 struct proc *p = td->td_proc;
2091 if (p->p_fd->fd_refcnt == 1)
2094 tmp = fdcopy(p->p_fd);
2100 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2104 td->td_proc->p_fd = fdp;
2108 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2109 * this is to ease callers, not catch errors.
2112 fdcopy(struct filedesc *fdp)
2114 struct filedesc *newfdp;
2115 struct filedescent *nfde, *ofde;
2120 newfdp = fdinit(fdp, true);
2121 /* copy all passable descriptors (i.e. not kqueue) */
2122 newfdp->fd_freefile = -1;
2123 for (i = 0; i <= fdp->fd_lastfile; ++i) {
2124 ofde = &fdp->fd_ofiles[i];
2125 if (ofde->fde_file == NULL ||
2126 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2127 !fhold(ofde->fde_file)) {
2128 if (newfdp->fd_freefile == -1)
2129 newfdp->fd_freefile = i;
2132 nfde = &newfdp->fd_ofiles[i];
2134 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2135 fdused_init(newfdp, i);
2136 newfdp->fd_lastfile = i;
2138 if (newfdp->fd_freefile == -1)
2139 newfdp->fd_freefile = i;
2140 newfdp->fd_cmask = fdp->fd_cmask;
2141 FILEDESC_SUNLOCK(fdp);
2146 * Copies a filedesc structure, while remapping all file descriptors
2147 * stored inside using a translation table.
2149 * File descriptors are copied over to the new file descriptor table,
2150 * regardless of whether the close-on-exec flag is set.
2153 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2154 struct filedesc **ret)
2156 struct filedesc *newfdp;
2157 struct filedescent *nfde, *ofde;
2162 newfdp = fdinit(fdp, true);
2163 if (nfds > fdp->fd_lastfile + 1) {
2164 /* New table cannot be larger than the old one. */
2168 /* Copy all passable descriptors (i.e. not kqueue). */
2169 newfdp->fd_freefile = nfds;
2170 for (i = 0; i < nfds; ++i) {
2171 if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2172 /* File descriptor out of bounds. */
2176 ofde = &fdp->fd_ofiles[fds[i]];
2177 if (ofde->fde_file == NULL) {
2178 /* Unused file descriptor. */
2182 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2183 /* File descriptor cannot be passed. */
2187 if (!fhold(nfde->fde_file)) {
2191 nfde = &newfdp->fd_ofiles[i];
2193 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2194 fdused_init(newfdp, i);
2195 newfdp->fd_lastfile = i;
2197 newfdp->fd_cmask = fdp->fd_cmask;
2198 FILEDESC_SUNLOCK(fdp);
2202 FILEDESC_SUNLOCK(fdp);
2203 fdescfree_remapped(newfdp);
2208 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2209 * one of processes using it exits) and the table used to be shared.
2212 fdclearlocks(struct thread *td)
2214 struct filedesc *fdp;
2215 struct filedesc_to_leader *fdtol;
2225 MPASS(fdtol != NULL);
2227 FILEDESC_XLOCK(fdp);
2228 KASSERT(fdtol->fdl_refcount > 0,
2229 ("filedesc_to_refcount botch: fdl_refcount=%d",
2230 fdtol->fdl_refcount));
2231 if (fdtol->fdl_refcount == 1 &&
2232 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2233 for (i = 0; i <= fdp->fd_lastfile; i++) {
2234 fp = fdp->fd_ofiles[i].fde_file;
2235 if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2238 FILEDESC_XUNLOCK(fdp);
2239 lf.l_whence = SEEK_SET;
2242 lf.l_type = F_UNLCK;
2244 (void) VOP_ADVLOCK(vp,
2245 (caddr_t)p->p_leader, F_UNLCK,
2247 FILEDESC_XLOCK(fdp);
2252 if (fdtol->fdl_refcount == 1) {
2253 if (fdp->fd_holdleaderscount > 0 &&
2254 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2256 * close() or kern_dup() has cleared a reference
2257 * in a shared file descriptor table.
2259 fdp->fd_holdleaderswakeup = 1;
2260 sx_sleep(&fdp->fd_holdleaderscount,
2261 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2264 if (fdtol->fdl_holdcount > 0) {
2266 * Ensure that fdtol->fdl_leader remains
2267 * valid in closef().
2269 fdtol->fdl_wakeup = 1;
2270 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2275 fdtol->fdl_refcount--;
2276 if (fdtol->fdl_refcount == 0 &&
2277 fdtol->fdl_holdcount == 0) {
2278 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2279 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2283 FILEDESC_XUNLOCK(fdp);
2285 free(fdtol, M_FILEDESC_TO_LEADER);
2289 * Release a filedesc structure.
2292 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2294 struct filedesc0 *fdp0;
2295 struct freetable *ft, *tft;
2296 struct filedescent *fde;
2300 for (i = 0; i <= fdp->fd_lastfile; i++) {
2301 fde = &fdp->fd_ofiles[i];
2306 (void) closef(fp, td);
2312 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2313 free(fdp->fd_map, M_FILEDESC);
2314 if (fdp->fd_nfiles > NDFILE)
2315 free(fdp->fd_files, M_FILEDESC);
2317 fdp0 = (struct filedesc0 *)fdp;
2318 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2319 free(ft->ft_table, M_FILEDESC);
2325 fdescfree(struct thread *td)
2328 struct filedesc *fdp;
2329 struct vnode *cdir, *jdir, *rdir;
2338 racct_set(p, RACCT_NOFILE, 0);
2343 if (p->p_fdtol != NULL)
2350 if (refcount_release(&fdp->fd_refcnt) == 0)
2353 FILEDESC_XLOCK(fdp);
2354 cdir = fdp->fd_cdir;
2355 fdp->fd_cdir = NULL;
2356 rdir = fdp->fd_rdir;
2357 fdp->fd_rdir = NULL;
2358 jdir = fdp->fd_jdir;
2359 fdp->fd_jdir = NULL;
2360 FILEDESC_XUNLOCK(fdp);
2369 fdescfree_fds(td, fdp, 1);
2373 fdescfree_remapped(struct filedesc *fdp)
2376 if (fdp->fd_cdir != NULL)
2377 vrele(fdp->fd_cdir);
2378 if (fdp->fd_rdir != NULL)
2379 vrele(fdp->fd_rdir);
2380 if (fdp->fd_jdir != NULL)
2381 vrele(fdp->fd_jdir);
2383 fdescfree_fds(curthread, fdp, 0);
2387 * For setugid programs, we don't want to people to use that setugidness
2388 * to generate error messages which write to a file which otherwise would
2389 * otherwise be off-limits to the process. We check for filesystems where
2390 * the vnode can change out from under us after execve (like [lin]procfs).
2392 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2393 * sufficient. We also don't check for setugidness since we know we are.
2396 is_unsafe(struct file *fp)
2400 if (fp->f_type != DTYPE_VNODE)
2404 return ((vp->v_vflag & VV_PROCDEP) != 0);
2408 * Make this setguid thing safe, if at all possible.
2411 fdsetugidsafety(struct thread *td)
2413 struct filedesc *fdp;
2417 fdp = td->td_proc->p_fd;
2418 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2419 MPASS(fdp->fd_nfiles >= 3);
2420 for (i = 0; i <= 2; i++) {
2421 fp = fdp->fd_ofiles[i].fde_file;
2422 if (fp != NULL && is_unsafe(fp)) {
2423 FILEDESC_XLOCK(fdp);
2424 knote_fdclose(td, i);
2426 * NULL-out descriptor prior to close to avoid
2427 * a race while close blocks.
2430 FILEDESC_XUNLOCK(fdp);
2431 (void) closef(fp, td);
2437 * If a specific file object occupies a specific file descriptor, close the
2438 * file descriptor entry and drop a reference on the file object. This is a
2439 * convenience function to handle a subsequent error in a function that calls
2440 * falloc() that handles the race that another thread might have closed the
2441 * file descriptor out from under the thread creating the file object.
2444 fdclose(struct thread *td, struct file *fp, int idx)
2446 struct filedesc *fdp = td->td_proc->p_fd;
2448 FILEDESC_XLOCK(fdp);
2449 if (fdp->fd_ofiles[idx].fde_file == fp) {
2451 FILEDESC_XUNLOCK(fdp);
2454 FILEDESC_XUNLOCK(fdp);
2458 * Close any files on exec?
2461 fdcloseexec(struct thread *td)
2463 struct filedesc *fdp;
2464 struct filedescent *fde;
2468 fdp = td->td_proc->p_fd;
2469 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2470 for (i = 0; i <= fdp->fd_lastfile; i++) {
2471 fde = &fdp->fd_ofiles[i];
2473 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2474 (fde->fde_flags & UF_EXCLOSE))) {
2475 FILEDESC_XLOCK(fdp);
2477 (void) closefp(fdp, i, fp, td, 0);
2478 FILEDESC_UNLOCK_ASSERT(fdp);
2484 * It is unsafe for set[ug]id processes to be started with file
2485 * descriptors 0..2 closed, as these descriptors are given implicit
2486 * significance in the Standard C library. fdcheckstd() will create a
2487 * descriptor referencing /dev/null for each of stdin, stdout, and
2488 * stderr that is not already open.
2491 fdcheckstd(struct thread *td)
2493 struct filedesc *fdp;
2495 int i, error, devnull;
2497 fdp = td->td_proc->p_fd;
2498 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2499 MPASS(fdp->fd_nfiles >= 3);
2501 for (i = 0; i <= 2; i++) {
2502 if (fdp->fd_ofiles[i].fde_file != NULL)
2505 save = td->td_retval[0];
2506 if (devnull != -1) {
2507 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2509 error = kern_openat(td, AT_FDCWD, "/dev/null",
2510 UIO_SYSSPACE, O_RDWR, 0);
2512 devnull = td->td_retval[0];
2513 KASSERT(devnull == i, ("we didn't get our fd"));
2516 td->td_retval[0] = save;
2524 * Internal form of close. Decrement reference count on file structure.
2525 * Note: td may be NULL when closing a file that was being passed in a
2529 closef(struct file *fp, struct thread *td)
2533 struct filedesc_to_leader *fdtol;
2534 struct filedesc *fdp;
2537 * POSIX record locking dictates that any close releases ALL
2538 * locks owned by this process. This is handled by setting
2539 * a flag in the unlock to free ONLY locks obeying POSIX
2540 * semantics, and not to free BSD-style file locks.
2541 * If the descriptor was in a message, POSIX-style locks
2542 * aren't passed with the descriptor, and the thread pointer
2543 * will be NULL. Callers should be careful only to pass a
2544 * NULL thread pointer when there really is no owning
2545 * context that might have locks, or the locks will be
2548 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2550 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2551 lf.l_whence = SEEK_SET;
2554 lf.l_type = F_UNLCK;
2555 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2556 F_UNLCK, &lf, F_POSIX);
2558 fdtol = td->td_proc->p_fdtol;
2559 if (fdtol != NULL) {
2561 * Handle special case where file descriptor table is
2562 * shared between multiple process leaders.
2564 fdp = td->td_proc->p_fd;
2565 FILEDESC_XLOCK(fdp);
2566 for (fdtol = fdtol->fdl_next;
2567 fdtol != td->td_proc->p_fdtol;
2568 fdtol = fdtol->fdl_next) {
2569 if ((fdtol->fdl_leader->p_flag &
2572 fdtol->fdl_holdcount++;
2573 FILEDESC_XUNLOCK(fdp);
2574 lf.l_whence = SEEK_SET;
2577 lf.l_type = F_UNLCK;
2579 (void) VOP_ADVLOCK(vp,
2580 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2582 FILEDESC_XLOCK(fdp);
2583 fdtol->fdl_holdcount--;
2584 if (fdtol->fdl_holdcount == 0 &&
2585 fdtol->fdl_wakeup != 0) {
2586 fdtol->fdl_wakeup = 0;
2590 FILEDESC_XUNLOCK(fdp);
2593 return (fdrop(fp, td));
2597 * Initialize the file pointer with the specified properties.
2599 * The ops are set with release semantics to be certain that the flags, type,
2600 * and data are visible when ops is. This is to prevent ops methods from being
2601 * called with bad data.
2604 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2609 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2613 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2614 struct file **fpp, struct filecaps *havecapsp)
2616 struct filedescent *fde;
2619 FILEDESC_LOCK_ASSERT(fdp);
2621 fde = fdeget_locked(fdp, fd);
2628 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2633 if (havecapsp != NULL)
2634 filecaps_copy(&fde->fde_caps, havecapsp, true);
2636 *fpp = fde->fde_file;
2644 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2645 struct file **fpp, struct filecaps *havecapsp)
2647 struct filedesc *fdp = td->td_proc->p_fd;
2649 #ifndef CAPABILITIES
2650 error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2651 if (error == 0 && havecapsp != NULL)
2652 filecaps_fill(havecapsp);
2658 error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2662 if (havecapsp != NULL) {
2663 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2664 havecapsp, false)) {
2670 if (!fd_modified(fdp, fd, seq))
2679 FILEDESC_SLOCK(fdp);
2680 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2681 if (error == 0 && !fhold(*fpp))
2683 FILEDESC_SUNLOCK(fdp);
2689 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2690 struct file **fpp, seq_t *seqp)
2693 const struct filedescent *fde;
2695 const struct fdescenttbl *fdt;
2700 cap_rights_t haverights;
2704 fdt = fdp->fd_files;
2705 if ((u_int)fd >= fdt->fdt_nfiles)
2708 * Fetch the descriptor locklessly. We avoid fdrop() races by
2709 * never raising a refcount above 0. To accomplish this we have
2710 * to use a cmpset loop rather than an atomic_add. The descriptor
2711 * must be re-verified once we acquire a reference to be certain
2712 * that the identity is still correct and we did not lose a race
2713 * due to preemption.
2717 seq = seq_load(fd_seq(fdt, fd));
2718 fde = &fdt->fdt_ofiles[fd];
2719 haverights = *cap_rights_fde_inline(fde);
2721 if (!seq_consistent(fd_seq(fdt, fd), seq))
2724 fp = fdt->fdt_ofiles[fd].fde_file;
2729 error = cap_check(&haverights, needrightsp);
2733 count = fp->f_count;
2737 * Force a reload. Other thread could reallocate the
2738 * table before this fd was closed, so it possible that
2739 * there is a stale fp pointer in cached version.
2741 fdt = *(const struct fdescenttbl * const volatile *)
2745 if (__predict_false(count + 1 < count))
2749 * Use an acquire barrier to force re-reading of fdt so it is
2750 * refreshed for verification.
2752 if (__predict_false(atomic_fcmpset_acq_int(&fp->f_count,
2753 &count, count + 1) == 0))
2755 fdt = fdp->fd_files;
2757 if (seq_consistent_nomb(fd_seq(fdt, fd), seq))
2759 if (fp == fdt->fdt_ofiles[fd].fde_file)
2762 fdrop(fp, curthread);
2774 * Extract the file pointer associated with the specified descriptor for the
2775 * current user process.
2777 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2780 * File's rights will be checked against the capability rights mask.
2782 * If an error occurred the non-zero error is returned and *fpp is set to
2783 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
2784 * responsible for fdrop().
2787 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2788 cap_rights_t *needrightsp, seq_t *seqp)
2790 struct filedesc *fdp;
2795 fdp = td->td_proc->p_fd;
2796 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2799 if (fp->f_ops == &badfileops) {
2805 * FREAD and FWRITE failure return EBADF as per POSIX.
2811 if ((fp->f_flag & flags) == 0)
2815 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2816 ((fp->f_flag & FWRITE) != 0))
2822 KASSERT(0, ("wrong flags"));
2835 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2838 return (_fget(td, fd, fpp, 0, rightsp, NULL));
2842 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
2846 #ifndef CAPABILITIES
2847 error = _fget(td, fd, fpp, 0, rightsp, NULL);
2848 if (maxprotp != NULL)
2849 *maxprotp = VM_PROT_ALL;
2851 cap_rights_t fdrights;
2852 struct filedesc *fdp = td->td_proc->p_fd;
2855 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2857 error = _fget(td, fd, fpp, 0, rightsp, &seq);
2860 if (maxprotp != NULL)
2861 fdrights = *cap_rights(fdp, fd);
2862 if (!fd_modified(fdp, fd, seq))
2868 * If requested, convert capability rights to access flags.
2870 if (maxprotp != NULL)
2871 *maxprotp = cap_rights_to_vmprot(&fdrights);
2877 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2880 return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2884 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2887 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2891 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2894 struct filedesc *fdp = td->td_proc->p_fd;
2895 #ifndef CAPABILITIES
2896 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2901 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2903 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2906 error = cap_fcntl_check(fdp, fd, needfcntl);
2907 if (!fd_modified(fdp, fd, seq))
2920 * Like fget() but loads the underlying vnode, or returns an error if the
2921 * descriptor does not represent a vnode. Note that pipes use vnodes but
2922 * never have VM objects. The returned vnode will be vref()'d.
2924 * XXX: what about the unused flags ?
2927 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2934 error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2937 if (fp->f_vnode == NULL) {
2949 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2952 return (_fgetvp(td, fd, 0, rightsp, vpp));
2956 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2957 struct filecaps *havecaps, struct vnode **vpp)
2959 struct filedesc *fdp;
2960 struct filecaps caps;
2964 fdp = td->td_proc->p_fd;
2965 error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2968 if (fp->f_ops == &badfileops) {
2972 if (fp->f_vnode == NULL) {
2983 filecaps_free(&caps);
2988 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2991 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2995 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2998 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3003 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3007 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3012 * Handle the last reference to a file being closed.
3014 * Without the noinline attribute clang keeps inlining the func thorough this
3015 * file when fdrop is used.
3018 _fdrop(struct file *fp, struct thread *td)
3022 if (fp->f_count != 0)
3023 panic("fdrop: count %d", fp->f_count);
3024 error = fo_close(fp, td);
3025 atomic_subtract_int(&openfiles, 1);
3027 free(fp->f_advice, M_FADVISE);
3028 uma_zfree(file_zone, fp);
3034 * Apply an advisory lock on a file descriptor.
3036 * Just attempt to get a record lock of the requested type on the entire file
3037 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3039 #ifndef _SYS_SYSPROTO_H_
3047 sys_flock(struct thread *td, struct flock_args *uap)
3054 error = fget(td, uap->fd, &cap_flock_rights, &fp);
3057 if (fp->f_type != DTYPE_VNODE) {
3059 return (EOPNOTSUPP);
3063 lf.l_whence = SEEK_SET;
3066 if (uap->how & LOCK_UN) {
3067 lf.l_type = F_UNLCK;
3068 atomic_clear_int(&fp->f_flag, FHASLOCK);
3069 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3072 if (uap->how & LOCK_EX)
3073 lf.l_type = F_WRLCK;
3074 else if (uap->how & LOCK_SH)
3075 lf.l_type = F_RDLCK;
3080 atomic_set_int(&fp->f_flag, FHASLOCK);
3081 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3082 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3088 * Duplicate the specified descriptor to a free descriptor.
3091 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3092 int openerror, int *indxp)
3094 struct filedescent *newfde, *oldfde;
3099 KASSERT(openerror == ENODEV || openerror == ENXIO,
3100 ("unexpected error %d in %s", openerror, __func__));
3103 * If the to-be-dup'd fd number is greater than the allowed number
3104 * of file descriptors, or the fd to be dup'd has already been
3105 * closed, then reject.
3107 FILEDESC_XLOCK(fdp);
3108 if ((fp = fget_locked(fdp, dfd)) == NULL) {
3109 FILEDESC_XUNLOCK(fdp);
3113 error = fdalloc(td, 0, &indx);
3115 FILEDESC_XUNLOCK(fdp);
3120 * There are two cases of interest here.
3122 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3124 * For ENXIO steal away the file structure from (dfd) and store it in
3125 * (indx). (dfd) is effectively closed by this operation.
3127 switch (openerror) {
3130 * Check that the mode the file is being opened for is a
3131 * subset of the mode of the existing descriptor.
3133 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3134 fdunused(fdp, indx);
3135 FILEDESC_XUNLOCK(fdp);
3139 fdunused(fdp, indx);
3140 FILEDESC_XUNLOCK(fdp);
3143 newfde = &fdp->fd_ofiles[indx];
3144 oldfde = &fdp->fd_ofiles[dfd];
3145 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3147 seq_write_begin(&newfde->fde_seq);
3149 memcpy(newfde, oldfde, fde_change_size);
3150 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3153 seq_write_end(&newfde->fde_seq);
3158 * Steal away the file pointer from dfd and stuff it into indx.
3160 newfde = &fdp->fd_ofiles[indx];
3161 oldfde = &fdp->fd_ofiles[dfd];
3163 seq_write_begin(&newfde->fde_seq);
3165 memcpy(newfde, oldfde, fde_change_size);
3166 oldfde->fde_file = NULL;
3169 seq_write_end(&newfde->fde_seq);
3173 FILEDESC_XUNLOCK(fdp);
3179 * This sysctl determines if we will allow a process to chroot(2) if it
3180 * has a directory open:
3181 * 0: disallowed for all processes.
3182 * 1: allowed for processes that were not already chroot(2)'ed.
3183 * 2: allowed for all processes.
3186 static int chroot_allow_open_directories = 1;
3188 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3189 &chroot_allow_open_directories, 0,
3190 "Allow a process to chroot(2) if it has a directory open");
3193 * Helper function for raised chroot(2) security function: Refuse if
3194 * any filedescriptors are open directories.
3197 chroot_refuse_vdir_fds(struct filedesc *fdp)
3203 FILEDESC_LOCK_ASSERT(fdp);
3205 for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3206 fp = fget_locked(fdp, fd);
3209 if (fp->f_type == DTYPE_VNODE) {
3211 if (vp->v_type == VDIR)
3219 * Common routine for kern_chroot() and jail_attach(). The caller is
3220 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3221 * authorize this operation.
3224 pwd_chroot(struct thread *td, struct vnode *vp)
3226 struct filedesc *fdp;
3227 struct vnode *oldvp;
3230 fdp = td->td_proc->p_fd;
3231 FILEDESC_XLOCK(fdp);
3232 if (chroot_allow_open_directories == 0 ||
3233 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3234 error = chroot_refuse_vdir_fds(fdp);
3236 FILEDESC_XUNLOCK(fdp);
3240 oldvp = fdp->fd_rdir;
3243 if (fdp->fd_jdir == NULL) {
3247 FILEDESC_XUNLOCK(fdp);
3253 pwd_chdir(struct thread *td, struct vnode *vp)
3255 struct filedesc *fdp;
3256 struct vnode *oldvp;
3258 fdp = td->td_proc->p_fd;
3259 FILEDESC_XLOCK(fdp);
3260 VNASSERT(vp->v_usecount > 0, vp,
3261 ("chdir to a vnode with zero usecount"));
3262 oldvp = fdp->fd_cdir;
3264 FILEDESC_XUNLOCK(fdp);
3269 * Scan all active processes and prisons to see if any of them have a current
3270 * or root directory of `olddp'. If so, replace them with the new mount point.
3273 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3275 struct filedesc *fdp;
3280 if (vrefcnt(olddp) == 1)
3283 sx_slock(&allproc_lock);
3284 FOREACH_PROC_IN_SYSTEM(p) {
3290 FILEDESC_XLOCK(fdp);
3291 if (fdp->fd_cdir == olddp) {
3293 fdp->fd_cdir = newdp;
3296 if (fdp->fd_rdir == olddp) {
3298 fdp->fd_rdir = newdp;
3301 if (fdp->fd_jdir == olddp) {
3303 fdp->fd_jdir = newdp;
3306 FILEDESC_XUNLOCK(fdp);
3309 sx_sunlock(&allproc_lock);
3310 if (rootvnode == olddp) {
3315 mtx_lock(&prison0.pr_mtx);
3316 if (prison0.pr_root == olddp) {
3318 prison0.pr_root = newdp;
3321 mtx_unlock(&prison0.pr_mtx);
3322 sx_slock(&allprison_lock);
3323 TAILQ_FOREACH(pr, &allprison, pr_list) {
3324 mtx_lock(&pr->pr_mtx);
3325 if (pr->pr_root == olddp) {
3327 pr->pr_root = newdp;
3330 mtx_unlock(&pr->pr_mtx);
3332 sx_sunlock(&allprison_lock);
3337 struct filedesc_to_leader *
3338 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3340 struct filedesc_to_leader *fdtol;
3342 fdtol = malloc(sizeof(struct filedesc_to_leader),
3343 M_FILEDESC_TO_LEADER, M_WAITOK);
3344 fdtol->fdl_refcount = 1;
3345 fdtol->fdl_holdcount = 0;
3346 fdtol->fdl_wakeup = 0;
3347 fdtol->fdl_leader = leader;
3349 FILEDESC_XLOCK(fdp);
3350 fdtol->fdl_next = old->fdl_next;
3351 fdtol->fdl_prev = old;
3352 old->fdl_next = fdtol;
3353 fdtol->fdl_next->fdl_prev = fdtol;
3354 FILEDESC_XUNLOCK(fdp);
3356 fdtol->fdl_next = fdtol;
3357 fdtol->fdl_prev = fdtol;
3363 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3365 struct filedesc *fdp;
3366 int i, count, slots;
3368 if (*(int *)arg1 != 0)
3371 fdp = curproc->p_fd;
3373 FILEDESC_SLOCK(fdp);
3374 slots = NDSLOTS(fdp->fd_lastfile + 1);
3375 for (i = 0; i < slots; i++)
3376 count += bitcountl(fdp->fd_map[i]);
3377 FILEDESC_SUNLOCK(fdp);
3379 return (SYSCTL_OUT(req, &count, sizeof(count)));
3382 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3383 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3384 "Number of open file descriptors");
3387 * Get file structures globally.
3390 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3393 struct filedesc *fdp;
3398 error = sysctl_wire_old_buffer(req, 0);
3401 if (req->oldptr == NULL) {
3403 sx_slock(&allproc_lock);
3404 FOREACH_PROC_IN_SYSTEM(p) {
3406 if (p->p_state == PRS_NEW) {
3414 /* overestimates sparse tables. */
3415 if (fdp->fd_lastfile > 0)
3416 n += fdp->fd_lastfile;
3419 sx_sunlock(&allproc_lock);
3420 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3423 bzero(&xf, sizeof(xf));
3424 xf.xf_size = sizeof(xf);
3425 sx_slock(&allproc_lock);
3426 FOREACH_PROC_IN_SYSTEM(p) {
3428 if (p->p_state == PRS_NEW) {
3432 if (p_cansee(req->td, p) != 0) {
3436 xf.xf_pid = p->p_pid;
3437 xf.xf_uid = p->p_ucred->cr_uid;
3442 FILEDESC_SLOCK(fdp);
3443 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3444 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3447 xf.xf_file = (uintptr_t)fp;
3448 xf.xf_data = (uintptr_t)fp->f_data;
3449 xf.xf_vnode = (uintptr_t)fp->f_vnode;
3450 xf.xf_type = (uintptr_t)fp->f_type;
3451 xf.xf_count = fp->f_count;
3453 xf.xf_offset = foffset_get(fp);
3454 xf.xf_flag = fp->f_flag;
3455 error = SYSCTL_OUT(req, &xf, sizeof(xf));
3459 FILEDESC_SUNLOCK(fdp);
3464 sx_sunlock(&allproc_lock);
3468 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3469 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3471 #ifdef KINFO_FILE_SIZE
3472 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3476 xlate_fflags(int fflags)
3478 static const struct {
3481 } fflags_table[] = {
3482 { FAPPEND, KF_FLAG_APPEND },
3483 { FASYNC, KF_FLAG_ASYNC },
3484 { FFSYNC, KF_FLAG_FSYNC },
3485 { FHASLOCK, KF_FLAG_HASLOCK },
3486 { FNONBLOCK, KF_FLAG_NONBLOCK },
3487 { FREAD, KF_FLAG_READ },
3488 { FWRITE, KF_FLAG_WRITE },
3489 { O_CREAT, KF_FLAG_CREAT },
3490 { O_DIRECT, KF_FLAG_DIRECT },
3491 { O_EXCL, KF_FLAG_EXCL },
3492 { O_EXEC, KF_FLAG_EXEC },
3493 { O_EXLOCK, KF_FLAG_EXLOCK },
3494 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3495 { O_SHLOCK, KF_FLAG_SHLOCK },
3496 { O_TRUNC, KF_FLAG_TRUNC }
3502 for (i = 0; i < nitems(fflags_table); i++)
3503 if (fflags & fflags_table[i].fflag)
3504 kflags |= fflags_table[i].kf_fflag;
3508 /* Trim unused data from kf_path by truncating the structure size. */
3510 pack_kinfo(struct kinfo_file *kif)
3513 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3514 strlen(kif->kf_path) + 1;
3515 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3519 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3520 struct kinfo_file *kif, struct filedesc *fdp, int flags)
3524 bzero(kif, sizeof(*kif));
3526 /* Set a default type to allow for empty fill_kinfo() methods. */
3527 kif->kf_type = KF_TYPE_UNKNOWN;
3528 kif->kf_flags = xlate_fflags(fp->f_flag);
3529 if (rightsp != NULL)
3530 kif->kf_cap_rights = *rightsp;
3532 cap_rights_init(&kif->kf_cap_rights);
3534 kif->kf_ref_count = fp->f_count;
3535 kif->kf_offset = foffset_get(fp);
3538 * This may drop the filedesc lock, so the 'fp' cannot be
3539 * accessed after this call.
3541 error = fo_fill_kinfo(fp, kif, fdp);
3543 kif->kf_status |= KF_ATTR_VALID;
3544 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3547 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3551 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3552 struct kinfo_file *kif, int flags)
3556 bzero(kif, sizeof(*kif));
3558 kif->kf_type = KF_TYPE_VNODE;
3559 error = vn_fill_kinfo_vnode(vp, kif);
3561 kif->kf_status |= KF_ATTR_VALID;
3562 kif->kf_flags = xlate_fflags(fflags);
3563 cap_rights_init(&kif->kf_cap_rights);
3565 kif->kf_ref_count = -1;
3566 kif->kf_offset = -1;
3567 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3570 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3574 struct export_fd_buf {
3575 struct filedesc *fdp;
3578 struct kinfo_file kif;
3583 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3585 struct kinfo_file *kif;
3588 if (efbuf->remainder != -1) {
3589 if (efbuf->remainder < kif->kf_structsize) {
3590 /* Terminate export. */
3591 efbuf->remainder = 0;
3594 efbuf->remainder -= kif->kf_structsize;
3596 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3600 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3601 struct export_fd_buf *efbuf)
3605 if (efbuf->remainder == 0)
3607 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3609 FILEDESC_SUNLOCK(efbuf->fdp);
3610 error = export_kinfo_to_sb(efbuf);
3611 FILEDESC_SLOCK(efbuf->fdp);
3616 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3617 struct export_fd_buf *efbuf)
3621 if (efbuf->remainder == 0)
3623 if (efbuf->fdp != NULL)
3624 FILEDESC_SUNLOCK(efbuf->fdp);
3625 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3626 error = export_kinfo_to_sb(efbuf);
3627 if (efbuf->fdp != NULL)
3628 FILEDESC_SLOCK(efbuf->fdp);
3633 * Store a process file descriptor information to sbuf.
3635 * Takes a locked proc as argument, and returns with the proc unlocked.
3638 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
3642 struct filedesc *fdp;
3643 struct export_fd_buf *efbuf;
3644 struct vnode *cttyvp, *textvp, *tracevp;
3646 cap_rights_t rights;
3648 PROC_LOCK_ASSERT(p, MA_OWNED);
3651 tracevp = p->p_tracevp;
3652 if (tracevp != NULL)
3655 textvp = p->p_textvp;
3658 /* Controlling tty. */
3660 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3661 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3667 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3670 efbuf->remainder = maxlen;
3671 efbuf->flags = flags;
3672 if (tracevp != NULL)
3673 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3676 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3678 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3684 FILEDESC_SLOCK(fdp);
3685 /* working directory */
3686 if (fdp->fd_cdir != NULL) {
3687 vrefact(fdp->fd_cdir);
3688 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3690 /* root directory */
3691 if (fdp->fd_rdir != NULL) {
3692 vrefact(fdp->fd_rdir);
3693 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3695 /* jail directory */
3696 if (fdp->fd_jdir != NULL) {
3697 vrefact(fdp->fd_jdir);
3698 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3700 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3701 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3704 rights = *cap_rights(fdp, i);
3705 #else /* !CAPABILITIES */
3706 rights = cap_no_rights;
3709 * Create sysctl entry. It is OK to drop the filedesc
3710 * lock inside of export_file_to_sb() as we will
3711 * re-validate and re-evaluate its properties when the
3714 error = export_file_to_sb(fp, i, &rights, efbuf);
3715 if (error != 0 || efbuf->remainder == 0)
3718 FILEDESC_SUNLOCK(fdp);
3721 free(efbuf, M_TEMP);
3725 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
3728 * Get per-process file descriptors for use by procstat(1), et al.
3731 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3736 int error, error2, *name;
3740 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3741 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3742 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3747 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3748 error = kern_proc_filedesc_out(p, &sb, maxlen,
3749 KERN_FILEDESC_PACK_KINFO);
3750 error2 = sbuf_finish(&sb);
3752 return (error != 0 ? error : error2);
3755 #ifdef COMPAT_FREEBSD7
3756 #ifdef KINFO_OFILE_SIZE
3757 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3761 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3764 okif->kf_structsize = sizeof(*okif);
3765 okif->kf_type = kif->kf_type;
3766 okif->kf_fd = kif->kf_fd;
3767 okif->kf_ref_count = kif->kf_ref_count;
3768 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3769 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3770 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3771 okif->kf_offset = kif->kf_offset;
3772 if (kif->kf_type == KF_TYPE_VNODE)
3773 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
3775 okif->kf_vnode_type = KF_VTYPE_VNON;
3776 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3777 if (kif->kf_type == KF_TYPE_SOCKET) {
3778 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
3779 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
3780 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
3781 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
3782 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
3784 okif->kf_sa_local.ss_family = AF_UNSPEC;
3785 okif->kf_sa_peer.ss_family = AF_UNSPEC;
3790 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3791 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3796 FILEDESC_SUNLOCK(fdp);
3797 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3798 kinfo_to_okinfo(kif, okif);
3799 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3800 FILEDESC_SLOCK(fdp);
3805 * Get per-process file descriptors for use by procstat(1), et al.
3808 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3810 struct kinfo_ofile *okif;
3811 struct kinfo_file *kif;
3812 struct filedesc *fdp;
3813 int error, i, *name;
3818 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3825 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3826 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3827 FILEDESC_SLOCK(fdp);
3828 if (fdp->fd_cdir != NULL)
3829 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3831 if (fdp->fd_rdir != NULL)
3832 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3834 if (fdp->fd_jdir != NULL)
3835 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3837 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3838 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3840 export_file_to_kinfo(fp, i, NULL, kif, fdp,
3841 KERN_FILEDESC_PACK_KINFO);
3842 FILEDESC_SUNLOCK(fdp);
3843 kinfo_to_okinfo(kif, okif);
3844 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3845 FILEDESC_SLOCK(fdp);
3849 FILEDESC_SUNLOCK(fdp);
3856 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3857 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3858 "Process ofiledesc entries");
3859 #endif /* COMPAT_FREEBSD7 */
3862 vntype_to_kinfo(int vtype)
3867 } vtypes_table[] = {
3868 { VBAD, KF_VTYPE_VBAD },
3869 { VBLK, KF_VTYPE_VBLK },
3870 { VCHR, KF_VTYPE_VCHR },
3871 { VDIR, KF_VTYPE_VDIR },
3872 { VFIFO, KF_VTYPE_VFIFO },
3873 { VLNK, KF_VTYPE_VLNK },
3874 { VNON, KF_VTYPE_VNON },
3875 { VREG, KF_VTYPE_VREG },
3876 { VSOCK, KF_VTYPE_VSOCK }
3881 * Perform vtype translation.
3883 for (i = 0; i < nitems(vtypes_table); i++)
3884 if (vtypes_table[i].vtype == vtype)
3885 return (vtypes_table[i].kf_vtype);
3887 return (KF_VTYPE_UNKNOWN);
3890 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3891 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3892 "Process filedesc entries");
3895 * Store a process current working directory information to sbuf.
3897 * Takes a locked proc as argument, and returns with the proc unlocked.
3900 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
3902 struct filedesc *fdp;
3903 struct export_fd_buf *efbuf;
3906 PROC_LOCK_ASSERT(p, MA_OWNED);
3913 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3916 efbuf->remainder = maxlen;
3918 FILEDESC_SLOCK(fdp);
3919 if (fdp->fd_cdir == NULL)
3922 vrefact(fdp->fd_cdir);
3923 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3926 FILEDESC_SUNLOCK(fdp);
3928 free(efbuf, M_TEMP);
3933 * Get per-process current working directory.
3936 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3941 int error, error2, *name;
3945 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3946 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3947 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3952 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3953 error = kern_proc_cwd_out(p, &sb, maxlen);
3954 error2 = sbuf_finish(&sb);
3956 return (error != 0 ? error : error2);
3959 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3960 sysctl_kern_proc_cwd, "Process current working directory");
3964 * For the purposes of debugging, generate a human-readable string for the
3968 file_type_to_name(short type)
3996 case DTYPE_PROCDESC:
3998 case DTYPE_LINUXEFD:
4000 case DTYPE_LINUXTFD:
4008 * For the purposes of debugging, identify a process (if any, perhaps one of
4009 * many) that references the passed file in its file descriptor array. Return
4012 static struct proc *
4013 file_to_first_proc(struct file *fp)
4015 struct filedesc *fdp;
4019 FOREACH_PROC_IN_SYSTEM(p) {
4020 if (p->p_state == PRS_NEW)
4025 for (n = 0; n <= fdp->fd_lastfile; n++) {
4026 if (fp == fdp->fd_ofiles[n].fde_file)
4034 db_print_file(struct file *fp, int header)
4036 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4040 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4041 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4042 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4044 p = file_to_first_proc(fp);
4045 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4046 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4047 fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode,
4048 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4053 DB_SHOW_COMMAND(file, db_show_file)
4058 db_printf("usage: show file <addr>\n");
4061 fp = (struct file *)addr;
4062 db_print_file(fp, 1);
4065 DB_SHOW_COMMAND(files, db_show_files)
4067 struct filedesc *fdp;
4074 FOREACH_PROC_IN_SYSTEM(p) {
4075 if (p->p_state == PRS_NEW)
4077 if ((fdp = p->p_fd) == NULL)
4079 for (n = 0; n <= fdp->fd_lastfile; ++n) {
4080 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4082 db_print_file(fp, header);
4089 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4090 &maxfilesperproc, 0, "Maximum files allowed open per process");
4092 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4093 &maxfiles, 0, "Maximum number of files");
4095 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4096 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
4100 filelistinit(void *dummy)
4103 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4104 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4105 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4106 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4107 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4109 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4111 /*-------------------------------------------------------------------*/
4114 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4115 int flags, struct thread *td)
4122 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4130 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4138 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4146 badfo_kqfilter(struct file *fp, struct knote *kn)
4153 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4161 badfo_close(struct file *fp, struct thread *td)
4168 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4176 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4184 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4185 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4193 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4199 struct fileops badfileops = {
4200 .fo_read = badfo_readwrite,
4201 .fo_write = badfo_readwrite,
4202 .fo_truncate = badfo_truncate,
4203 .fo_ioctl = badfo_ioctl,
4204 .fo_poll = badfo_poll,
4205 .fo_kqfilter = badfo_kqfilter,
4206 .fo_stat = badfo_stat,
4207 .fo_close = badfo_close,
4208 .fo_chmod = badfo_chmod,
4209 .fo_chown = badfo_chown,
4210 .fo_sendfile = badfo_sendfile,
4211 .fo_fill_kinfo = badfo_fill_kinfo,
4215 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4216 int flags, struct thread *td)
4219 return (EOPNOTSUPP);
4223 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4231 invfo_ioctl(struct file *fp, u_long com, void *data,
4232 struct ucred *active_cred, struct thread *td)
4239 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4243 return (poll_no_poll(events));
4247 invfo_kqfilter(struct file *fp, struct knote *kn)
4254 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4262 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4270 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4271 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4278 /*-------------------------------------------------------------------*/
4281 * File Descriptor pseudo-device driver (/dev/fd/).
4283 * Opening minor device N dup()s the file (if any) connected to file
4284 * descriptor N belonging to the calling process. Note that this driver
4285 * consists of only the ``open()'' routine, because all subsequent
4286 * references to this file will be direct to the other driver.
4288 * XXX: we could give this one a cloning event handler if necessary.
4293 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4297 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4298 * the file descriptor being sought for duplication. The error
4299 * return ensures that the vnode for this device will be released
4300 * by vn_open. Open will detect this special error and take the
4301 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4302 * will simply report the error.
4304 td->td_dupfd = dev2unit(dev);
4308 static struct cdevsw fildesc_cdevsw = {
4309 .d_version = D_VERSION,
4315 fildesc_drvinit(void *unused)
4319 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4320 UID_ROOT, GID_WHEEL, 0666, "fd/0");
4321 make_dev_alias(dev, "stdin");
4322 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4323 UID_ROOT, GID_WHEEL, 0666, "fd/1");
4324 make_dev_alias(dev, "stdout");
4325 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4326 UID_ROOT, GID_WHEEL, 0666, "fd/2");
4327 make_dev_alias(dev, "stderr");
4330 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);