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 oioctls = filecaps_free_prep(&newfde->fde_caps);
923 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
926 * Duplicate the source descriptor.
929 seq_write_begin(&newfde->fde_seq);
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 * Close open file descriptors.
1269 #ifndef _SYS_SYSPROTO_H_
1270 struct closefrom_args {
1276 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1278 struct filedesc *fdp;
1281 fdp = td->td_proc->p_fd;
1282 AUDIT_ARG_FD(uap->lowfd);
1285 * Treat negative starting file descriptor values identical to
1286 * closefrom(0) which closes all files.
1290 FILEDESC_SLOCK(fdp);
1291 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1292 if (fdp->fd_ofiles[fd].fde_file != NULL) {
1293 FILEDESC_SUNLOCK(fdp);
1294 (void)kern_close(td, fd);
1295 FILEDESC_SLOCK(fdp);
1298 FILEDESC_SUNLOCK(fdp);
1302 #if defined(COMPAT_43)
1304 * Return status information about a file descriptor.
1306 #ifndef _SYS_SYSPROTO_H_
1307 struct ofstat_args {
1314 ofstat(struct thread *td, struct ofstat_args *uap)
1320 error = kern_fstat(td, uap->fd, &ub);
1323 error = copyout(&oub, uap->sb, sizeof(oub));
1327 #endif /* COMPAT_43 */
1329 #if defined(COMPAT_FREEBSD11)
1331 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1334 struct freebsd11_stat osb;
1337 error = kern_fstat(td, uap->fd, &sb);
1340 error = freebsd11_cvtstat(&sb, &osb);
1342 error = copyout(&osb, uap->sb, sizeof(osb));
1345 #endif /* COMPAT_FREEBSD11 */
1348 * Return status information about a file descriptor.
1350 #ifndef _SYS_SYSPROTO_H_
1358 sys_fstat(struct thread *td, struct fstat_args *uap)
1363 error = kern_fstat(td, uap->fd, &ub);
1365 error = copyout(&ub, uap->sb, sizeof(ub));
1370 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1377 error = fget(td, fd, &cap_fstat_rights, &fp);
1381 AUDIT_ARG_FILE(td->td_proc, fp);
1383 error = fo_stat(fp, sbp, td->td_ucred, td);
1385 #ifdef __STAT_TIME_T_EXT
1387 sbp->st_atim_ext = 0;
1388 sbp->st_mtim_ext = 0;
1389 sbp->st_ctim_ext = 0;
1390 sbp->st_btim_ext = 0;
1394 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1400 #if defined(COMPAT_FREEBSD11)
1402 * Return status information about a file descriptor.
1404 #ifndef _SYS_SYSPROTO_H_
1405 struct freebsd11_nfstat_args {
1412 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1418 error = kern_fstat(td, uap->fd, &ub);
1420 freebsd11_cvtnstat(&ub, &nub);
1421 error = copyout(&nub, uap->sb, sizeof(nub));
1425 #endif /* COMPAT_FREEBSD11 */
1428 * Return pathconf information about a file descriptor.
1430 #ifndef _SYS_SYSPROTO_H_
1431 struct fpathconf_args {
1438 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1443 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1445 td->td_retval[0] = value;
1450 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1456 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1460 if (name == _PC_ASYNC_IO) {
1461 *valuep = _POSIX_ASYNCHRONOUS_IO;
1466 vn_lock(vp, LK_SHARED | LK_RETRY);
1467 error = VOP_PATHCONF(vp, name, valuep);
1469 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1470 if (name != _PC_PIPE_BUF) {
1485 * Initialize filecaps structure.
1488 filecaps_init(struct filecaps *fcaps)
1491 bzero(fcaps, sizeof(*fcaps));
1492 fcaps->fc_nioctls = -1;
1496 * Copy filecaps structure allocating memory for ioctls array if needed.
1498 * The last parameter indicates whether the fdtable is locked. If it is not and
1499 * ioctls are encountered, copying fails and the caller must lock the table.
1501 * Note that if the table was not locked, the caller has to check the relevant
1502 * sequence counter to determine whether the operation was successful.
1505 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1509 if (src->fc_ioctls != NULL && !locked)
1511 memcpy(dst, src, sizeof(*src));
1512 if (src->fc_ioctls == NULL)
1515 KASSERT(src->fc_nioctls > 0,
1516 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1518 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1519 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1520 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1525 filecaps_copy_prep(const struct filecaps *src)
1530 if (src->fc_ioctls == NULL)
1533 KASSERT(src->fc_nioctls > 0,
1534 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1536 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1537 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1542 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1548 if (src->fc_ioctls == NULL) {
1549 MPASS(ioctls == NULL);
1553 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1554 dst->fc_ioctls = ioctls;
1555 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1559 * Move filecaps structure to the new place and clear the old place.
1562 filecaps_move(struct filecaps *src, struct filecaps *dst)
1566 bzero(src, sizeof(*src));
1570 * Fill the given filecaps structure with full rights.
1573 filecaps_fill(struct filecaps *fcaps)
1576 CAP_ALL(&fcaps->fc_rights);
1577 fcaps->fc_ioctls = NULL;
1578 fcaps->fc_nioctls = -1;
1579 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1583 * Free memory allocated within filecaps structure.
1586 filecaps_free(struct filecaps *fcaps)
1589 free(fcaps->fc_ioctls, M_FILECAPS);
1590 bzero(fcaps, sizeof(*fcaps));
1594 filecaps_free_prep(struct filecaps *fcaps)
1598 ioctls = fcaps->fc_ioctls;
1599 bzero(fcaps, sizeof(*fcaps));
1604 filecaps_free_finish(u_long *ioctls)
1607 free(ioctls, M_FILECAPS);
1611 * Validate the given filecaps structure.
1614 filecaps_validate(const struct filecaps *fcaps, const char *func)
1617 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1618 ("%s: invalid rights", func));
1619 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1620 ("%s: invalid fcntls", func));
1621 KASSERT(fcaps->fc_fcntls == 0 ||
1622 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1623 ("%s: fcntls without CAP_FCNTL", func));
1624 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1625 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1626 ("%s: invalid ioctls", func));
1627 KASSERT(fcaps->fc_nioctls == 0 ||
1628 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1629 ("%s: ioctls without CAP_IOCTL", func));
1633 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1637 FILEDESC_XLOCK_ASSERT(fdp);
1639 nfd1 = fdp->fd_nfiles * 2;
1642 fdgrowtable(fdp, nfd1);
1646 * Grow the file table to accommodate (at least) nfd descriptors.
1649 fdgrowtable(struct filedesc *fdp, int nfd)
1651 struct filedesc0 *fdp0;
1652 struct freetable *ft;
1653 struct fdescenttbl *ntable;
1654 struct fdescenttbl *otable;
1655 int nnfiles, onfiles;
1656 NDSLOTTYPE *nmap, *omap;
1659 * If lastfile is -1 this struct filedesc was just allocated and we are
1660 * growing it to accommodate for the one we are going to copy from. There
1661 * is no need to have a lock on this one as it's not visible to anyone.
1663 if (fdp->fd_lastfile != -1)
1664 FILEDESC_XLOCK_ASSERT(fdp);
1666 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1668 /* save old values */
1669 onfiles = fdp->fd_nfiles;
1670 otable = fdp->fd_files;
1673 /* compute the size of the new table */
1674 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1675 if (nnfiles <= onfiles)
1676 /* the table is already large enough */
1680 * Allocate a new table. We need enough space for the number of
1681 * entries, file entries themselves and the struct freetable we will use
1682 * when we decommission the table and place it on the freelist.
1683 * We place the struct freetable in the middle so we don't have
1684 * to worry about padding.
1686 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1687 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1688 sizeof(struct freetable),
1689 M_FILEDESC, M_ZERO | M_WAITOK);
1690 /* copy the old data */
1691 ntable->fdt_nfiles = nnfiles;
1692 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1693 onfiles * sizeof(ntable->fdt_ofiles[0]));
1696 * Allocate a new map only if the old is not large enough. It will
1697 * grow at a slower rate than the table as it can map more
1698 * entries than the table can hold.
1700 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1701 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1703 /* copy over the old data and update the pointer */
1704 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1709 * Make sure that ntable is correctly initialized before we replace
1710 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1713 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1716 * Do not free the old file table, as some threads may still
1717 * reference entries within it. Instead, place it on a freelist
1718 * which will be processed when the struct filedesc is released.
1720 * Note that if onfiles == NDFILE, we're dealing with the original
1721 * static allocation contained within (struct filedesc0 *)fdp,
1722 * which must not be freed.
1724 if (onfiles > NDFILE) {
1725 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1726 fdp0 = (struct filedesc0 *)fdp;
1727 ft->ft_table = otable;
1728 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1731 * The map does not have the same possibility of threads still
1732 * holding references to it. So always free it as long as it
1733 * does not reference the original static allocation.
1735 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1736 free(omap, M_FILEDESC);
1740 * Allocate a file descriptor for the process.
1743 fdalloc(struct thread *td, int minfd, int *result)
1745 struct proc *p = td->td_proc;
1746 struct filedesc *fdp = p->p_fd;
1747 int fd, maxfd, allocfd;
1752 FILEDESC_XLOCK_ASSERT(fdp);
1754 if (fdp->fd_freefile > minfd)
1755 minfd = fdp->fd_freefile;
1757 maxfd = getmaxfd(td);
1760 * Search the bitmap for a free descriptor starting at minfd.
1761 * If none is found, grow the file table.
1763 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1766 if (fd >= fdp->fd_nfiles) {
1767 allocfd = min(fd * 2, maxfd);
1771 error = racct_set(p, RACCT_NOFILE, allocfd);
1778 * fd is already equal to first free descriptor >= minfd, so
1779 * we only need to grow the table and we are done.
1781 fdgrowtable_exp(fdp, allocfd);
1785 * Perform some sanity checks, then mark the file descriptor as
1786 * used and return it to the caller.
1788 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1789 ("invalid descriptor %d", fd));
1790 KASSERT(!fdisused(fdp, fd),
1791 ("fd_first_free() returned non-free descriptor"));
1792 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1793 ("file descriptor isn't free"));
1800 * Allocate n file descriptors for the process.
1803 fdallocn(struct thread *td, int minfd, int *fds, int n)
1805 struct proc *p = td->td_proc;
1806 struct filedesc *fdp = p->p_fd;
1809 FILEDESC_XLOCK_ASSERT(fdp);
1811 for (i = 0; i < n; i++)
1812 if (fdalloc(td, 0, &fds[i]) != 0)
1816 for (i--; i >= 0; i--)
1817 fdunused(fdp, fds[i]);
1825 * Create a new open file structure and allocate a file descriptor for the
1826 * process that refers to it. We add one reference to the file for the
1827 * descriptor table and one reference for resultfp. This is to prevent us
1828 * being preempted and the entry in the descriptor table closed after we
1829 * release the FILEDESC lock.
1832 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1833 struct filecaps *fcaps)
1838 error = falloc_noinstall(td, &fp);
1840 return (error); /* no reference held on error */
1842 error = finstall(td, fp, &fd, flags, fcaps);
1844 fdrop(fp, td); /* one reference (fp only) */
1848 if (resultfp != NULL)
1849 *resultfp = fp; /* copy out result */
1851 fdrop(fp, td); /* release local reference */
1853 if (resultfd != NULL)
1860 * Create a new open file structure without allocating a file descriptor.
1863 falloc_noinstall(struct thread *td, struct file **resultfp)
1866 int maxuserfiles = maxfiles - (maxfiles / 20);
1868 static struct timeval lastfail;
1871 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1873 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1874 if ((openfiles_new >= maxuserfiles &&
1875 priv_check(td, PRIV_MAXFILES) != 0) ||
1876 openfiles_new >= maxfiles) {
1877 atomic_subtract_int(&openfiles, 1);
1878 if (ppsratecheck(&lastfail, &curfail, 1)) {
1879 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1880 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1884 fp = uma_zalloc(file_zone, M_WAITOK);
1885 bzero(fp, sizeof(*fp));
1886 refcount_init(&fp->f_count, 1);
1887 fp->f_cred = crhold(td->td_ucred);
1888 fp->f_ops = &badfileops;
1894 * Install a file in a file descriptor table.
1897 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1898 struct filecaps *fcaps)
1900 struct filedescent *fde;
1904 filecaps_validate(fcaps, __func__);
1905 FILEDESC_XLOCK_ASSERT(fdp);
1907 fde = &fdp->fd_ofiles[fd];
1909 seq_write_begin(&fde->fde_seq);
1912 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1914 filecaps_move(fcaps, &fde->fde_caps);
1916 filecaps_fill(&fde->fde_caps);
1918 seq_write_end(&fde->fde_seq);
1923 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1924 struct filecaps *fcaps)
1926 struct filedesc *fdp = td->td_proc->p_fd;
1933 FILEDESC_XLOCK(fdp);
1934 if ((error = fdalloc(td, 0, fd))) {
1935 FILEDESC_XUNLOCK(fdp);
1939 _finstall(fdp, fp, *fd, flags, fcaps);
1940 FILEDESC_XUNLOCK(fdp);
1945 * Build a new filedesc structure from another.
1946 * Copy the current, root, and jail root vnode references.
1948 * If fdp is not NULL, return with it shared locked.
1951 fdinit(struct filedesc *fdp, bool prepfiles)
1953 struct filedesc0 *newfdp0;
1954 struct filedesc *newfdp;
1956 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1957 newfdp = &newfdp0->fd_fd;
1959 /* Create the file descriptor table. */
1960 FILEDESC_LOCK_INIT(newfdp);
1961 refcount_init(&newfdp->fd_refcnt, 1);
1962 refcount_init(&newfdp->fd_holdcnt, 1);
1963 newfdp->fd_cmask = CMASK;
1964 newfdp->fd_map = newfdp0->fd_dmap;
1965 newfdp->fd_lastfile = -1;
1966 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1967 newfdp->fd_files->fdt_nfiles = NDFILE;
1972 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1973 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1975 FILEDESC_SLOCK(fdp);
1976 newfdp->fd_cdir = fdp->fd_cdir;
1977 if (newfdp->fd_cdir)
1978 vrefact(newfdp->fd_cdir);
1979 newfdp->fd_rdir = fdp->fd_rdir;
1980 if (newfdp->fd_rdir)
1981 vrefact(newfdp->fd_rdir);
1982 newfdp->fd_jdir = fdp->fd_jdir;
1983 if (newfdp->fd_jdir)
1984 vrefact(newfdp->fd_jdir);
1987 FILEDESC_SUNLOCK(fdp);
1989 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1990 FILEDESC_SUNLOCK(fdp);
1991 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1992 FILEDESC_SLOCK(fdp);
1999 static struct filedesc *
2000 fdhold(struct proc *p)
2002 struct filedesc *fdp;
2004 PROC_LOCK_ASSERT(p, MA_OWNED);
2007 refcount_acquire(&fdp->fd_holdcnt);
2012 fddrop(struct filedesc *fdp)
2015 if (fdp->fd_holdcnt > 1) {
2016 if (refcount_release(&fdp->fd_holdcnt) == 0)
2020 FILEDESC_LOCK_DESTROY(fdp);
2021 uma_zfree(filedesc0_zone, fdp);
2025 * Share a filedesc structure.
2028 fdshare(struct filedesc *fdp)
2031 refcount_acquire(&fdp->fd_refcnt);
2036 * Unshare a filedesc structure, if necessary by making a copy
2039 fdunshare(struct thread *td)
2041 struct filedesc *tmp;
2042 struct proc *p = td->td_proc;
2044 if (p->p_fd->fd_refcnt == 1)
2047 tmp = fdcopy(p->p_fd);
2053 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2057 td->td_proc->p_fd = fdp;
2061 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2062 * this is to ease callers, not catch errors.
2065 fdcopy(struct filedesc *fdp)
2067 struct filedesc *newfdp;
2068 struct filedescent *nfde, *ofde;
2073 newfdp = fdinit(fdp, true);
2074 /* copy all passable descriptors (i.e. not kqueue) */
2075 newfdp->fd_freefile = -1;
2076 for (i = 0; i <= fdp->fd_lastfile; ++i) {
2077 ofde = &fdp->fd_ofiles[i];
2078 if (ofde->fde_file == NULL ||
2079 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2080 !fhold(ofde->fde_file)) {
2081 if (newfdp->fd_freefile == -1)
2082 newfdp->fd_freefile = i;
2085 nfde = &newfdp->fd_ofiles[i];
2087 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2088 fdused_init(newfdp, i);
2089 newfdp->fd_lastfile = i;
2091 if (newfdp->fd_freefile == -1)
2092 newfdp->fd_freefile = i;
2093 newfdp->fd_cmask = fdp->fd_cmask;
2094 FILEDESC_SUNLOCK(fdp);
2099 * Copies a filedesc structure, while remapping all file descriptors
2100 * stored inside using a translation table.
2102 * File descriptors are copied over to the new file descriptor table,
2103 * regardless of whether the close-on-exec flag is set.
2106 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2107 struct filedesc **ret)
2109 struct filedesc *newfdp;
2110 struct filedescent *nfde, *ofde;
2115 newfdp = fdinit(fdp, true);
2116 if (nfds > fdp->fd_lastfile + 1) {
2117 /* New table cannot be larger than the old one. */
2121 /* Copy all passable descriptors (i.e. not kqueue). */
2122 newfdp->fd_freefile = nfds;
2123 for (i = 0; i < nfds; ++i) {
2124 if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2125 /* File descriptor out of bounds. */
2129 ofde = &fdp->fd_ofiles[fds[i]];
2130 if (ofde->fde_file == NULL) {
2131 /* Unused file descriptor. */
2135 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2136 /* File descriptor cannot be passed. */
2140 if (!fhold(nfde->fde_file)) {
2144 nfde = &newfdp->fd_ofiles[i];
2146 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2147 fdused_init(newfdp, i);
2148 newfdp->fd_lastfile = i;
2150 newfdp->fd_cmask = fdp->fd_cmask;
2151 FILEDESC_SUNLOCK(fdp);
2155 FILEDESC_SUNLOCK(fdp);
2156 fdescfree_remapped(newfdp);
2161 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2162 * one of processes using it exits) and the table used to be shared.
2165 fdclearlocks(struct thread *td)
2167 struct filedesc *fdp;
2168 struct filedesc_to_leader *fdtol;
2178 MPASS(fdtol != NULL);
2180 FILEDESC_XLOCK(fdp);
2181 KASSERT(fdtol->fdl_refcount > 0,
2182 ("filedesc_to_refcount botch: fdl_refcount=%d",
2183 fdtol->fdl_refcount));
2184 if (fdtol->fdl_refcount == 1 &&
2185 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2186 for (i = 0; i <= fdp->fd_lastfile; i++) {
2187 fp = fdp->fd_ofiles[i].fde_file;
2188 if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2191 FILEDESC_XUNLOCK(fdp);
2192 lf.l_whence = SEEK_SET;
2195 lf.l_type = F_UNLCK;
2197 (void) VOP_ADVLOCK(vp,
2198 (caddr_t)p->p_leader, F_UNLCK,
2200 FILEDESC_XLOCK(fdp);
2205 if (fdtol->fdl_refcount == 1) {
2206 if (fdp->fd_holdleaderscount > 0 &&
2207 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2209 * close() or kern_dup() has cleared a reference
2210 * in a shared file descriptor table.
2212 fdp->fd_holdleaderswakeup = 1;
2213 sx_sleep(&fdp->fd_holdleaderscount,
2214 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2217 if (fdtol->fdl_holdcount > 0) {
2219 * Ensure that fdtol->fdl_leader remains
2220 * valid in closef().
2222 fdtol->fdl_wakeup = 1;
2223 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2228 fdtol->fdl_refcount--;
2229 if (fdtol->fdl_refcount == 0 &&
2230 fdtol->fdl_holdcount == 0) {
2231 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2232 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2236 FILEDESC_XUNLOCK(fdp);
2238 free(fdtol, M_FILEDESC_TO_LEADER);
2242 * Release a filedesc structure.
2245 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2247 struct filedesc0 *fdp0;
2248 struct freetable *ft, *tft;
2249 struct filedescent *fde;
2253 for (i = 0; i <= fdp->fd_lastfile; i++) {
2254 fde = &fdp->fd_ofiles[i];
2259 (void) closef(fp, td);
2265 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2266 free(fdp->fd_map, M_FILEDESC);
2267 if (fdp->fd_nfiles > NDFILE)
2268 free(fdp->fd_files, M_FILEDESC);
2270 fdp0 = (struct filedesc0 *)fdp;
2271 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2272 free(ft->ft_table, M_FILEDESC);
2278 fdescfree(struct thread *td)
2281 struct filedesc *fdp;
2282 struct vnode *cdir, *jdir, *rdir;
2291 racct_set(p, RACCT_NOFILE, 0);
2296 if (p->p_fdtol != NULL)
2303 if (refcount_release(&fdp->fd_refcnt) == 0)
2306 FILEDESC_XLOCK(fdp);
2307 cdir = fdp->fd_cdir;
2308 fdp->fd_cdir = NULL;
2309 rdir = fdp->fd_rdir;
2310 fdp->fd_rdir = NULL;
2311 jdir = fdp->fd_jdir;
2312 fdp->fd_jdir = NULL;
2313 FILEDESC_XUNLOCK(fdp);
2322 fdescfree_fds(td, fdp, 1);
2326 fdescfree_remapped(struct filedesc *fdp)
2329 if (fdp->fd_cdir != NULL)
2330 vrele(fdp->fd_cdir);
2331 if (fdp->fd_rdir != NULL)
2332 vrele(fdp->fd_rdir);
2333 if (fdp->fd_jdir != NULL)
2334 vrele(fdp->fd_jdir);
2336 fdescfree_fds(curthread, fdp, 0);
2340 * For setugid programs, we don't want to people to use that setugidness
2341 * to generate error messages which write to a file which otherwise would
2342 * otherwise be off-limits to the process. We check for filesystems where
2343 * the vnode can change out from under us after execve (like [lin]procfs).
2345 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2346 * sufficient. We also don't check for setugidness since we know we are.
2349 is_unsafe(struct file *fp)
2353 if (fp->f_type != DTYPE_VNODE)
2357 return ((vp->v_vflag & VV_PROCDEP) != 0);
2361 * Make this setguid thing safe, if at all possible.
2364 fdsetugidsafety(struct thread *td)
2366 struct filedesc *fdp;
2370 fdp = td->td_proc->p_fd;
2371 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2372 MPASS(fdp->fd_nfiles >= 3);
2373 for (i = 0; i <= 2; i++) {
2374 fp = fdp->fd_ofiles[i].fde_file;
2375 if (fp != NULL && is_unsafe(fp)) {
2376 FILEDESC_XLOCK(fdp);
2377 knote_fdclose(td, i);
2379 * NULL-out descriptor prior to close to avoid
2380 * a race while close blocks.
2383 FILEDESC_XUNLOCK(fdp);
2384 (void) closef(fp, td);
2390 * If a specific file object occupies a specific file descriptor, close the
2391 * file descriptor entry and drop a reference on the file object. This is a
2392 * convenience function to handle a subsequent error in a function that calls
2393 * falloc() that handles the race that another thread might have closed the
2394 * file descriptor out from under the thread creating the file object.
2397 fdclose(struct thread *td, struct file *fp, int idx)
2399 struct filedesc *fdp = td->td_proc->p_fd;
2401 FILEDESC_XLOCK(fdp);
2402 if (fdp->fd_ofiles[idx].fde_file == fp) {
2404 FILEDESC_XUNLOCK(fdp);
2407 FILEDESC_XUNLOCK(fdp);
2411 * Close any files on exec?
2414 fdcloseexec(struct thread *td)
2416 struct filedesc *fdp;
2417 struct filedescent *fde;
2421 fdp = td->td_proc->p_fd;
2422 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2423 for (i = 0; i <= fdp->fd_lastfile; i++) {
2424 fde = &fdp->fd_ofiles[i];
2426 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2427 (fde->fde_flags & UF_EXCLOSE))) {
2428 FILEDESC_XLOCK(fdp);
2430 (void) closefp(fdp, i, fp, td, 0);
2431 FILEDESC_UNLOCK_ASSERT(fdp);
2437 * It is unsafe for set[ug]id processes to be started with file
2438 * descriptors 0..2 closed, as these descriptors are given implicit
2439 * significance in the Standard C library. fdcheckstd() will create a
2440 * descriptor referencing /dev/null for each of stdin, stdout, and
2441 * stderr that is not already open.
2444 fdcheckstd(struct thread *td)
2446 struct filedesc *fdp;
2448 int i, error, devnull;
2450 fdp = td->td_proc->p_fd;
2451 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2452 MPASS(fdp->fd_nfiles >= 3);
2454 for (i = 0; i <= 2; i++) {
2455 if (fdp->fd_ofiles[i].fde_file != NULL)
2458 save = td->td_retval[0];
2459 if (devnull != -1) {
2460 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2462 error = kern_openat(td, AT_FDCWD, "/dev/null",
2463 UIO_SYSSPACE, O_RDWR, 0);
2465 devnull = td->td_retval[0];
2466 KASSERT(devnull == i, ("we didn't get our fd"));
2469 td->td_retval[0] = save;
2477 * Internal form of close. Decrement reference count on file structure.
2478 * Note: td may be NULL when closing a file that was being passed in a
2482 closef(struct file *fp, struct thread *td)
2486 struct filedesc_to_leader *fdtol;
2487 struct filedesc *fdp;
2490 * POSIX record locking dictates that any close releases ALL
2491 * locks owned by this process. This is handled by setting
2492 * a flag in the unlock to free ONLY locks obeying POSIX
2493 * semantics, and not to free BSD-style file locks.
2494 * If the descriptor was in a message, POSIX-style locks
2495 * aren't passed with the descriptor, and the thread pointer
2496 * will be NULL. Callers should be careful only to pass a
2497 * NULL thread pointer when there really is no owning
2498 * context that might have locks, or the locks will be
2501 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2503 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2504 lf.l_whence = SEEK_SET;
2507 lf.l_type = F_UNLCK;
2508 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2509 F_UNLCK, &lf, F_POSIX);
2511 fdtol = td->td_proc->p_fdtol;
2512 if (fdtol != NULL) {
2514 * Handle special case where file descriptor table is
2515 * shared between multiple process leaders.
2517 fdp = td->td_proc->p_fd;
2518 FILEDESC_XLOCK(fdp);
2519 for (fdtol = fdtol->fdl_next;
2520 fdtol != td->td_proc->p_fdtol;
2521 fdtol = fdtol->fdl_next) {
2522 if ((fdtol->fdl_leader->p_flag &
2525 fdtol->fdl_holdcount++;
2526 FILEDESC_XUNLOCK(fdp);
2527 lf.l_whence = SEEK_SET;
2530 lf.l_type = F_UNLCK;
2532 (void) VOP_ADVLOCK(vp,
2533 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2535 FILEDESC_XLOCK(fdp);
2536 fdtol->fdl_holdcount--;
2537 if (fdtol->fdl_holdcount == 0 &&
2538 fdtol->fdl_wakeup != 0) {
2539 fdtol->fdl_wakeup = 0;
2543 FILEDESC_XUNLOCK(fdp);
2546 return (fdrop(fp, td));
2550 * Initialize the file pointer with the specified properties.
2552 * The ops are set with release semantics to be certain that the flags, type,
2553 * and data are visible when ops is. This is to prevent ops methods from being
2554 * called with bad data.
2557 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2562 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2566 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2567 struct file **fpp, struct filecaps *havecapsp)
2569 struct filedescent *fde;
2572 FILEDESC_LOCK_ASSERT(fdp);
2574 fde = fdeget_locked(fdp, fd);
2581 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2586 if (havecapsp != NULL)
2587 filecaps_copy(&fde->fde_caps, havecapsp, true);
2589 *fpp = fde->fde_file;
2597 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2598 struct file **fpp, struct filecaps *havecapsp)
2600 struct filedesc *fdp = td->td_proc->p_fd;
2602 #ifndef CAPABILITIES
2603 error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2604 if (error == 0 && havecapsp != NULL)
2605 filecaps_fill(havecapsp);
2611 error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2615 if (havecapsp != NULL) {
2616 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2617 havecapsp, false)) {
2623 if (!fd_modified(fdp, fd, seq))
2632 FILEDESC_SLOCK(fdp);
2633 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2634 if (error == 0 && !fhold(*fpp))
2636 FILEDESC_SUNLOCK(fdp);
2642 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2643 struct file **fpp, seq_t *seqp)
2646 const struct filedescent *fde;
2648 const struct fdescenttbl *fdt;
2653 cap_rights_t haverights;
2657 fdt = fdp->fd_files;
2658 if ((u_int)fd >= fdt->fdt_nfiles)
2661 * Fetch the descriptor locklessly. We avoid fdrop() races by
2662 * never raising a refcount above 0. To accomplish this we have
2663 * to use a cmpset loop rather than an atomic_add. The descriptor
2664 * must be re-verified once we acquire a reference to be certain
2665 * that the identity is still correct and we did not lose a race
2666 * due to preemption.
2670 seq = seq_load(fd_seq(fdt, fd));
2671 fde = &fdt->fdt_ofiles[fd];
2672 haverights = *cap_rights_fde_inline(fde);
2674 if (!seq_consistent(fd_seq(fdt, fd), seq))
2677 fp = fdt->fdt_ofiles[fd].fde_file;
2682 error = cap_check(&haverights, needrightsp);
2686 count = fp->f_count;
2690 * Force a reload. Other thread could reallocate the
2691 * table before this fd was closed, so it possible that
2692 * there is a stale fp pointer in cached version.
2694 fdt = *(const struct fdescenttbl * const volatile *)
2698 if (__predict_false(count + 1 < count))
2702 * Use an acquire barrier to force re-reading of fdt so it is
2703 * refreshed for verification.
2705 if (__predict_false(atomic_fcmpset_acq_int(&fp->f_count,
2706 &count, count + 1) == 0))
2708 fdt = fdp->fd_files;
2710 if (seq_consistent_nomb(fd_seq(fdt, fd), seq))
2712 if (fp == fdt->fdt_ofiles[fd].fde_file)
2715 fdrop(fp, curthread);
2727 * Extract the file pointer associated with the specified descriptor for the
2728 * current user process.
2730 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2733 * File's rights will be checked against the capability rights mask.
2735 * If an error occurred the non-zero error is returned and *fpp is set to
2736 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
2737 * responsible for fdrop().
2740 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2741 cap_rights_t *needrightsp, seq_t *seqp)
2743 struct filedesc *fdp;
2748 fdp = td->td_proc->p_fd;
2749 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2752 if (fp->f_ops == &badfileops) {
2758 * FREAD and FWRITE failure return EBADF as per POSIX.
2764 if ((fp->f_flag & flags) == 0)
2768 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2769 ((fp->f_flag & FWRITE) != 0))
2775 KASSERT(0, ("wrong flags"));
2788 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2791 return (_fget(td, fd, fpp, 0, rightsp, NULL));
2795 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
2799 #ifndef CAPABILITIES
2800 error = _fget(td, fd, fpp, 0, rightsp, NULL);
2801 if (maxprotp != NULL)
2802 *maxprotp = VM_PROT_ALL;
2804 cap_rights_t fdrights;
2805 struct filedesc *fdp = td->td_proc->p_fd;
2808 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2810 error = _fget(td, fd, fpp, 0, rightsp, &seq);
2813 if (maxprotp != NULL)
2814 fdrights = *cap_rights(fdp, fd);
2815 if (!fd_modified(fdp, fd, seq))
2821 * If requested, convert capability rights to access flags.
2823 if (maxprotp != NULL)
2824 *maxprotp = cap_rights_to_vmprot(&fdrights);
2830 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2833 return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2837 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2840 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2844 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2847 struct filedesc *fdp = td->td_proc->p_fd;
2848 #ifndef CAPABILITIES
2849 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2854 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2856 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2859 error = cap_fcntl_check(fdp, fd, needfcntl);
2860 if (!fd_modified(fdp, fd, seq))
2873 * Like fget() but loads the underlying vnode, or returns an error if the
2874 * descriptor does not represent a vnode. Note that pipes use vnodes but
2875 * never have VM objects. The returned vnode will be vref()'d.
2877 * XXX: what about the unused flags ?
2880 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2887 error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2890 if (fp->f_vnode == NULL) {
2902 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2905 return (_fgetvp(td, fd, 0, rightsp, vpp));
2909 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2910 struct filecaps *havecaps, struct vnode **vpp)
2912 struct filedesc *fdp;
2913 struct filecaps caps;
2917 fdp = td->td_proc->p_fd;
2918 error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2921 if (fp->f_ops == &badfileops) {
2925 if (fp->f_vnode == NULL) {
2936 filecaps_free(&caps);
2941 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2944 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2948 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2951 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2956 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2960 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2965 * Handle the last reference to a file being closed.
2967 * Without the noinline attribute clang keeps inlining the func thorough this
2968 * file when fdrop is used.
2971 _fdrop(struct file *fp, struct thread *td)
2975 if (fp->f_count != 0)
2976 panic("fdrop: count %d", fp->f_count);
2977 error = fo_close(fp, td);
2978 atomic_subtract_int(&openfiles, 1);
2980 free(fp->f_advice, M_FADVISE);
2981 uma_zfree(file_zone, fp);
2987 * Apply an advisory lock on a file descriptor.
2989 * Just attempt to get a record lock of the requested type on the entire file
2990 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2992 #ifndef _SYS_SYSPROTO_H_
3000 sys_flock(struct thread *td, struct flock_args *uap)
3007 error = fget(td, uap->fd, &cap_flock_rights, &fp);
3010 if (fp->f_type != DTYPE_VNODE) {
3012 return (EOPNOTSUPP);
3016 lf.l_whence = SEEK_SET;
3019 if (uap->how & LOCK_UN) {
3020 lf.l_type = F_UNLCK;
3021 atomic_clear_int(&fp->f_flag, FHASLOCK);
3022 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3025 if (uap->how & LOCK_EX)
3026 lf.l_type = F_WRLCK;
3027 else if (uap->how & LOCK_SH)
3028 lf.l_type = F_RDLCK;
3033 atomic_set_int(&fp->f_flag, FHASLOCK);
3034 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3035 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3041 * Duplicate the specified descriptor to a free descriptor.
3044 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3045 int openerror, int *indxp)
3047 struct filedescent *newfde, *oldfde;
3052 KASSERT(openerror == ENODEV || openerror == ENXIO,
3053 ("unexpected error %d in %s", openerror, __func__));
3056 * If the to-be-dup'd fd number is greater than the allowed number
3057 * of file descriptors, or the fd to be dup'd has already been
3058 * closed, then reject.
3060 FILEDESC_XLOCK(fdp);
3061 if ((fp = fget_locked(fdp, dfd)) == NULL) {
3062 FILEDESC_XUNLOCK(fdp);
3066 error = fdalloc(td, 0, &indx);
3068 FILEDESC_XUNLOCK(fdp);
3073 * There are two cases of interest here.
3075 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3077 * For ENXIO steal away the file structure from (dfd) and store it in
3078 * (indx). (dfd) is effectively closed by this operation.
3080 switch (openerror) {
3083 * Check that the mode the file is being opened for is a
3084 * subset of the mode of the existing descriptor.
3086 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3087 fdunused(fdp, indx);
3088 FILEDESC_XUNLOCK(fdp);
3092 fdunused(fdp, indx);
3093 FILEDESC_XUNLOCK(fdp);
3096 newfde = &fdp->fd_ofiles[indx];
3097 oldfde = &fdp->fd_ofiles[dfd];
3098 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3100 seq_write_begin(&newfde->fde_seq);
3102 memcpy(newfde, oldfde, fde_change_size);
3103 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3106 seq_write_end(&newfde->fde_seq);
3111 * Steal away the file pointer from dfd and stuff it into indx.
3113 newfde = &fdp->fd_ofiles[indx];
3114 oldfde = &fdp->fd_ofiles[dfd];
3116 seq_write_begin(&newfde->fde_seq);
3118 memcpy(newfde, oldfde, fde_change_size);
3119 oldfde->fde_file = NULL;
3122 seq_write_end(&newfde->fde_seq);
3126 FILEDESC_XUNLOCK(fdp);
3132 * This sysctl determines if we will allow a process to chroot(2) if it
3133 * has a directory open:
3134 * 0: disallowed for all processes.
3135 * 1: allowed for processes that were not already chroot(2)'ed.
3136 * 2: allowed for all processes.
3139 static int chroot_allow_open_directories = 1;
3141 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3142 &chroot_allow_open_directories, 0,
3143 "Allow a process to chroot(2) if it has a directory open");
3146 * Helper function for raised chroot(2) security function: Refuse if
3147 * any filedescriptors are open directories.
3150 chroot_refuse_vdir_fds(struct filedesc *fdp)
3156 FILEDESC_LOCK_ASSERT(fdp);
3158 for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3159 fp = fget_locked(fdp, fd);
3162 if (fp->f_type == DTYPE_VNODE) {
3164 if (vp->v_type == VDIR)
3172 * Common routine for kern_chroot() and jail_attach(). The caller is
3173 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3174 * authorize this operation.
3177 pwd_chroot(struct thread *td, struct vnode *vp)
3179 struct filedesc *fdp;
3180 struct vnode *oldvp;
3183 fdp = td->td_proc->p_fd;
3184 FILEDESC_XLOCK(fdp);
3185 if (chroot_allow_open_directories == 0 ||
3186 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3187 error = chroot_refuse_vdir_fds(fdp);
3189 FILEDESC_XUNLOCK(fdp);
3193 oldvp = fdp->fd_rdir;
3196 if (fdp->fd_jdir == NULL) {
3200 FILEDESC_XUNLOCK(fdp);
3206 pwd_chdir(struct thread *td, struct vnode *vp)
3208 struct filedesc *fdp;
3209 struct vnode *oldvp;
3211 fdp = td->td_proc->p_fd;
3212 FILEDESC_XLOCK(fdp);
3213 VNASSERT(vp->v_usecount > 0, vp,
3214 ("chdir to a vnode with zero usecount"));
3215 oldvp = fdp->fd_cdir;
3217 FILEDESC_XUNLOCK(fdp);
3222 * Scan all active processes and prisons to see if any of them have a current
3223 * or root directory of `olddp'. If so, replace them with the new mount point.
3226 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3228 struct filedesc *fdp;
3233 if (vrefcnt(olddp) == 1)
3236 sx_slock(&allproc_lock);
3237 FOREACH_PROC_IN_SYSTEM(p) {
3243 FILEDESC_XLOCK(fdp);
3244 if (fdp->fd_cdir == olddp) {
3246 fdp->fd_cdir = newdp;
3249 if (fdp->fd_rdir == olddp) {
3251 fdp->fd_rdir = newdp;
3254 if (fdp->fd_jdir == olddp) {
3256 fdp->fd_jdir = newdp;
3259 FILEDESC_XUNLOCK(fdp);
3262 sx_sunlock(&allproc_lock);
3263 if (rootvnode == olddp) {
3268 mtx_lock(&prison0.pr_mtx);
3269 if (prison0.pr_root == olddp) {
3271 prison0.pr_root = newdp;
3274 mtx_unlock(&prison0.pr_mtx);
3275 sx_slock(&allprison_lock);
3276 TAILQ_FOREACH(pr, &allprison, pr_list) {
3277 mtx_lock(&pr->pr_mtx);
3278 if (pr->pr_root == olddp) {
3280 pr->pr_root = newdp;
3283 mtx_unlock(&pr->pr_mtx);
3285 sx_sunlock(&allprison_lock);
3290 struct filedesc_to_leader *
3291 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3293 struct filedesc_to_leader *fdtol;
3295 fdtol = malloc(sizeof(struct filedesc_to_leader),
3296 M_FILEDESC_TO_LEADER, M_WAITOK);
3297 fdtol->fdl_refcount = 1;
3298 fdtol->fdl_holdcount = 0;
3299 fdtol->fdl_wakeup = 0;
3300 fdtol->fdl_leader = leader;
3302 FILEDESC_XLOCK(fdp);
3303 fdtol->fdl_next = old->fdl_next;
3304 fdtol->fdl_prev = old;
3305 old->fdl_next = fdtol;
3306 fdtol->fdl_next->fdl_prev = fdtol;
3307 FILEDESC_XUNLOCK(fdp);
3309 fdtol->fdl_next = fdtol;
3310 fdtol->fdl_prev = fdtol;
3316 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3318 struct filedesc *fdp;
3319 int i, count, slots;
3321 if (*(int *)arg1 != 0)
3324 fdp = curproc->p_fd;
3326 FILEDESC_SLOCK(fdp);
3327 slots = NDSLOTS(fdp->fd_lastfile + 1);
3328 for (i = 0; i < slots; i++)
3329 count += bitcountl(fdp->fd_map[i]);
3330 FILEDESC_SUNLOCK(fdp);
3332 return (SYSCTL_OUT(req, &count, sizeof(count)));
3335 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3336 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3337 "Number of open file descriptors");
3340 * Get file structures globally.
3343 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3346 struct filedesc *fdp;
3351 error = sysctl_wire_old_buffer(req, 0);
3354 if (req->oldptr == NULL) {
3356 sx_slock(&allproc_lock);
3357 FOREACH_PROC_IN_SYSTEM(p) {
3359 if (p->p_state == PRS_NEW) {
3367 /* overestimates sparse tables. */
3368 if (fdp->fd_lastfile > 0)
3369 n += fdp->fd_lastfile;
3372 sx_sunlock(&allproc_lock);
3373 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3376 bzero(&xf, sizeof(xf));
3377 xf.xf_size = sizeof(xf);
3378 sx_slock(&allproc_lock);
3379 FOREACH_PROC_IN_SYSTEM(p) {
3381 if (p->p_state == PRS_NEW) {
3385 if (p_cansee(req->td, p) != 0) {
3389 xf.xf_pid = p->p_pid;
3390 xf.xf_uid = p->p_ucred->cr_uid;
3395 FILEDESC_SLOCK(fdp);
3396 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3397 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3400 xf.xf_file = (uintptr_t)fp;
3401 xf.xf_data = (uintptr_t)fp->f_data;
3402 xf.xf_vnode = (uintptr_t)fp->f_vnode;
3403 xf.xf_type = (uintptr_t)fp->f_type;
3404 xf.xf_count = fp->f_count;
3406 xf.xf_offset = foffset_get(fp);
3407 xf.xf_flag = fp->f_flag;
3408 error = SYSCTL_OUT(req, &xf, sizeof(xf));
3412 FILEDESC_SUNLOCK(fdp);
3417 sx_sunlock(&allproc_lock);
3421 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3422 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3424 #ifdef KINFO_FILE_SIZE
3425 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3429 xlate_fflags(int fflags)
3431 static const struct {
3434 } fflags_table[] = {
3435 { FAPPEND, KF_FLAG_APPEND },
3436 { FASYNC, KF_FLAG_ASYNC },
3437 { FFSYNC, KF_FLAG_FSYNC },
3438 { FHASLOCK, KF_FLAG_HASLOCK },
3439 { FNONBLOCK, KF_FLAG_NONBLOCK },
3440 { FREAD, KF_FLAG_READ },
3441 { FWRITE, KF_FLAG_WRITE },
3442 { O_CREAT, KF_FLAG_CREAT },
3443 { O_DIRECT, KF_FLAG_DIRECT },
3444 { O_EXCL, KF_FLAG_EXCL },
3445 { O_EXEC, KF_FLAG_EXEC },
3446 { O_EXLOCK, KF_FLAG_EXLOCK },
3447 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3448 { O_SHLOCK, KF_FLAG_SHLOCK },
3449 { O_TRUNC, KF_FLAG_TRUNC }
3455 for (i = 0; i < nitems(fflags_table); i++)
3456 if (fflags & fflags_table[i].fflag)
3457 kflags |= fflags_table[i].kf_fflag;
3461 /* Trim unused data from kf_path by truncating the structure size. */
3463 pack_kinfo(struct kinfo_file *kif)
3466 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3467 strlen(kif->kf_path) + 1;
3468 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3472 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3473 struct kinfo_file *kif, struct filedesc *fdp, int flags)
3477 bzero(kif, sizeof(*kif));
3479 /* Set a default type to allow for empty fill_kinfo() methods. */
3480 kif->kf_type = KF_TYPE_UNKNOWN;
3481 kif->kf_flags = xlate_fflags(fp->f_flag);
3482 if (rightsp != NULL)
3483 kif->kf_cap_rights = *rightsp;
3485 cap_rights_init(&kif->kf_cap_rights);
3487 kif->kf_ref_count = fp->f_count;
3488 kif->kf_offset = foffset_get(fp);
3491 * This may drop the filedesc lock, so the 'fp' cannot be
3492 * accessed after this call.
3494 error = fo_fill_kinfo(fp, kif, fdp);
3496 kif->kf_status |= KF_ATTR_VALID;
3497 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3500 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3504 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3505 struct kinfo_file *kif, int flags)
3509 bzero(kif, sizeof(*kif));
3511 kif->kf_type = KF_TYPE_VNODE;
3512 error = vn_fill_kinfo_vnode(vp, kif);
3514 kif->kf_status |= KF_ATTR_VALID;
3515 kif->kf_flags = xlate_fflags(fflags);
3516 cap_rights_init(&kif->kf_cap_rights);
3518 kif->kf_ref_count = -1;
3519 kif->kf_offset = -1;
3520 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3523 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3527 struct export_fd_buf {
3528 struct filedesc *fdp;
3531 struct kinfo_file kif;
3536 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3538 struct kinfo_file *kif;
3541 if (efbuf->remainder != -1) {
3542 if (efbuf->remainder < kif->kf_structsize) {
3543 /* Terminate export. */
3544 efbuf->remainder = 0;
3547 efbuf->remainder -= kif->kf_structsize;
3549 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3553 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3554 struct export_fd_buf *efbuf)
3558 if (efbuf->remainder == 0)
3560 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3562 FILEDESC_SUNLOCK(efbuf->fdp);
3563 error = export_kinfo_to_sb(efbuf);
3564 FILEDESC_SLOCK(efbuf->fdp);
3569 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3570 struct export_fd_buf *efbuf)
3574 if (efbuf->remainder == 0)
3576 if (efbuf->fdp != NULL)
3577 FILEDESC_SUNLOCK(efbuf->fdp);
3578 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3579 error = export_kinfo_to_sb(efbuf);
3580 if (efbuf->fdp != NULL)
3581 FILEDESC_SLOCK(efbuf->fdp);
3586 * Store a process file descriptor information to sbuf.
3588 * Takes a locked proc as argument, and returns with the proc unlocked.
3591 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
3595 struct filedesc *fdp;
3596 struct export_fd_buf *efbuf;
3597 struct vnode *cttyvp, *textvp, *tracevp;
3599 cap_rights_t rights;
3601 PROC_LOCK_ASSERT(p, MA_OWNED);
3604 tracevp = p->p_tracevp;
3605 if (tracevp != NULL)
3608 textvp = p->p_textvp;
3611 /* Controlling tty. */
3613 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3614 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3620 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3623 efbuf->remainder = maxlen;
3624 efbuf->flags = flags;
3625 if (tracevp != NULL)
3626 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3629 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3631 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3637 FILEDESC_SLOCK(fdp);
3638 /* working directory */
3639 if (fdp->fd_cdir != NULL) {
3640 vrefact(fdp->fd_cdir);
3641 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3643 /* root directory */
3644 if (fdp->fd_rdir != NULL) {
3645 vrefact(fdp->fd_rdir);
3646 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3648 /* jail directory */
3649 if (fdp->fd_jdir != NULL) {
3650 vrefact(fdp->fd_jdir);
3651 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3653 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3654 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3657 rights = *cap_rights(fdp, i);
3658 #else /* !CAPABILITIES */
3659 rights = cap_no_rights;
3662 * Create sysctl entry. It is OK to drop the filedesc
3663 * lock inside of export_file_to_sb() as we will
3664 * re-validate and re-evaluate its properties when the
3667 error = export_file_to_sb(fp, i, &rights, efbuf);
3668 if (error != 0 || efbuf->remainder == 0)
3671 FILEDESC_SUNLOCK(fdp);
3674 free(efbuf, M_TEMP);
3678 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
3681 * Get per-process file descriptors for use by procstat(1), et al.
3684 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3689 int error, error2, *name;
3693 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3694 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3695 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3700 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3701 error = kern_proc_filedesc_out(p, &sb, maxlen,
3702 KERN_FILEDESC_PACK_KINFO);
3703 error2 = sbuf_finish(&sb);
3705 return (error != 0 ? error : error2);
3708 #ifdef COMPAT_FREEBSD7
3709 #ifdef KINFO_OFILE_SIZE
3710 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3714 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3717 okif->kf_structsize = sizeof(*okif);
3718 okif->kf_type = kif->kf_type;
3719 okif->kf_fd = kif->kf_fd;
3720 okif->kf_ref_count = kif->kf_ref_count;
3721 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3722 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3723 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3724 okif->kf_offset = kif->kf_offset;
3725 if (kif->kf_type == KF_TYPE_VNODE)
3726 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
3728 okif->kf_vnode_type = KF_VTYPE_VNON;
3729 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3730 if (kif->kf_type == KF_TYPE_SOCKET) {
3731 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
3732 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
3733 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
3734 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
3735 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
3737 okif->kf_sa_local.ss_family = AF_UNSPEC;
3738 okif->kf_sa_peer.ss_family = AF_UNSPEC;
3743 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3744 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3749 FILEDESC_SUNLOCK(fdp);
3750 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3751 kinfo_to_okinfo(kif, okif);
3752 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3753 FILEDESC_SLOCK(fdp);
3758 * Get per-process file descriptors for use by procstat(1), et al.
3761 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3763 struct kinfo_ofile *okif;
3764 struct kinfo_file *kif;
3765 struct filedesc *fdp;
3766 int error, i, *name;
3771 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3778 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3779 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3780 FILEDESC_SLOCK(fdp);
3781 if (fdp->fd_cdir != NULL)
3782 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3784 if (fdp->fd_rdir != NULL)
3785 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3787 if (fdp->fd_jdir != NULL)
3788 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3790 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3791 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3793 export_file_to_kinfo(fp, i, NULL, kif, fdp,
3794 KERN_FILEDESC_PACK_KINFO);
3795 FILEDESC_SUNLOCK(fdp);
3796 kinfo_to_okinfo(kif, okif);
3797 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3798 FILEDESC_SLOCK(fdp);
3802 FILEDESC_SUNLOCK(fdp);
3809 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3810 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3811 "Process ofiledesc entries");
3812 #endif /* COMPAT_FREEBSD7 */
3815 vntype_to_kinfo(int vtype)
3820 } vtypes_table[] = {
3821 { VBAD, KF_VTYPE_VBAD },
3822 { VBLK, KF_VTYPE_VBLK },
3823 { VCHR, KF_VTYPE_VCHR },
3824 { VDIR, KF_VTYPE_VDIR },
3825 { VFIFO, KF_VTYPE_VFIFO },
3826 { VLNK, KF_VTYPE_VLNK },
3827 { VNON, KF_VTYPE_VNON },
3828 { VREG, KF_VTYPE_VREG },
3829 { VSOCK, KF_VTYPE_VSOCK }
3834 * Perform vtype translation.
3836 for (i = 0; i < nitems(vtypes_table); i++)
3837 if (vtypes_table[i].vtype == vtype)
3838 return (vtypes_table[i].kf_vtype);
3840 return (KF_VTYPE_UNKNOWN);
3843 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3844 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3845 "Process filedesc entries");
3848 * Store a process current working directory information to sbuf.
3850 * Takes a locked proc as argument, and returns with the proc unlocked.
3853 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
3855 struct filedesc *fdp;
3856 struct export_fd_buf *efbuf;
3859 PROC_LOCK_ASSERT(p, MA_OWNED);
3866 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3869 efbuf->remainder = maxlen;
3871 FILEDESC_SLOCK(fdp);
3872 if (fdp->fd_cdir == NULL)
3875 vrefact(fdp->fd_cdir);
3876 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3879 FILEDESC_SUNLOCK(fdp);
3881 free(efbuf, M_TEMP);
3886 * Get per-process current working directory.
3889 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3894 int error, error2, *name;
3898 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3899 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3900 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3905 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3906 error = kern_proc_cwd_out(p, &sb, maxlen);
3907 error2 = sbuf_finish(&sb);
3909 return (error != 0 ? error : error2);
3912 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3913 sysctl_kern_proc_cwd, "Process current working directory");
3917 * For the purposes of debugging, generate a human-readable string for the
3921 file_type_to_name(short type)
3949 case DTYPE_PROCDESC:
3951 case DTYPE_LINUXEFD:
3953 case DTYPE_LINUXTFD:
3961 * For the purposes of debugging, identify a process (if any, perhaps one of
3962 * many) that references the passed file in its file descriptor array. Return
3965 static struct proc *
3966 file_to_first_proc(struct file *fp)
3968 struct filedesc *fdp;
3972 FOREACH_PROC_IN_SYSTEM(p) {
3973 if (p->p_state == PRS_NEW)
3978 for (n = 0; n <= fdp->fd_lastfile; n++) {
3979 if (fp == fdp->fd_ofiles[n].fde_file)
3987 db_print_file(struct file *fp, int header)
3989 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
3993 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
3994 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
3995 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
3997 p = file_to_first_proc(fp);
3998 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
3999 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4000 fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode,
4001 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4006 DB_SHOW_COMMAND(file, db_show_file)
4011 db_printf("usage: show file <addr>\n");
4014 fp = (struct file *)addr;
4015 db_print_file(fp, 1);
4018 DB_SHOW_COMMAND(files, db_show_files)
4020 struct filedesc *fdp;
4027 FOREACH_PROC_IN_SYSTEM(p) {
4028 if (p->p_state == PRS_NEW)
4030 if ((fdp = p->p_fd) == NULL)
4032 for (n = 0; n <= fdp->fd_lastfile; ++n) {
4033 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4035 db_print_file(fp, header);
4042 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4043 &maxfilesperproc, 0, "Maximum files allowed open per process");
4045 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4046 &maxfiles, 0, "Maximum number of files");
4048 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4049 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
4053 filelistinit(void *dummy)
4056 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4057 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4058 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4059 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4060 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4062 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4064 /*-------------------------------------------------------------------*/
4067 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4068 int flags, struct thread *td)
4075 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4083 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4091 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4099 badfo_kqfilter(struct file *fp, struct knote *kn)
4106 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4114 badfo_close(struct file *fp, struct thread *td)
4121 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4129 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4137 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4138 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4146 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4152 struct fileops badfileops = {
4153 .fo_read = badfo_readwrite,
4154 .fo_write = badfo_readwrite,
4155 .fo_truncate = badfo_truncate,
4156 .fo_ioctl = badfo_ioctl,
4157 .fo_poll = badfo_poll,
4158 .fo_kqfilter = badfo_kqfilter,
4159 .fo_stat = badfo_stat,
4160 .fo_close = badfo_close,
4161 .fo_chmod = badfo_chmod,
4162 .fo_chown = badfo_chown,
4163 .fo_sendfile = badfo_sendfile,
4164 .fo_fill_kinfo = badfo_fill_kinfo,
4168 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4169 int flags, struct thread *td)
4172 return (EOPNOTSUPP);
4176 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4184 invfo_ioctl(struct file *fp, u_long com, void *data,
4185 struct ucred *active_cred, struct thread *td)
4192 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4196 return (poll_no_poll(events));
4200 invfo_kqfilter(struct file *fp, struct knote *kn)
4207 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4215 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4223 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4224 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4231 /*-------------------------------------------------------------------*/
4234 * File Descriptor pseudo-device driver (/dev/fd/).
4236 * Opening minor device N dup()s the file (if any) connected to file
4237 * descriptor N belonging to the calling process. Note that this driver
4238 * consists of only the ``open()'' routine, because all subsequent
4239 * references to this file will be direct to the other driver.
4241 * XXX: we could give this one a cloning event handler if necessary.
4246 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4250 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4251 * the file descriptor being sought for duplication. The error
4252 * return ensures that the vnode for this device will be released
4253 * by vn_open. Open will detect this special error and take the
4254 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4255 * will simply report the error.
4257 td->td_dupfd = dev2unit(dev);
4261 static struct cdevsw fildesc_cdevsw = {
4262 .d_version = D_VERSION,
4268 fildesc_drvinit(void *unused)
4272 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4273 UID_ROOT, GID_WHEEL, 0666, "fd/0");
4274 make_dev_alias(dev, "stdin");
4275 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4276 UID_ROOT, GID_WHEEL, 0666, "fd/1");
4277 make_dev_alias(dev, "stdout");
4278 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4279 UID_ROOT, GID_WHEEL, 0666, "fd/2");
4280 make_dev_alias(dev, "stderr");
4283 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);