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) {
774 * Exclusive lock synchronizes against f_seqcount reads and
775 * writes in sequential_heuristic().
777 error = vn_lock(vp, LK_EXCLUSIVE);
783 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
784 arg = MIN(arg, INT_MAX - bsize + 1);
785 fp->f_seqcount = MIN(IO_SEQMAX,
786 (arg + bsize - 1) / bsize);
787 atomic_set_int(&fp->f_flag, FRDAHEAD);
789 atomic_clear_int(&fp->f_flag, FRDAHEAD);
803 getmaxfd(struct thread *td)
806 return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
810 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
813 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
815 struct filedesc *fdp;
816 struct filedescent *oldfde, *newfde;
819 u_long *oioctls, *nioctls;
825 MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
826 MPASS(mode < FDDUP_LASTMODE);
829 /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
832 * Verify we have a valid descriptor to dup from and possibly to
833 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
834 * return EINVAL when the new descriptor is out of bounds.
839 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
840 maxfd = getmaxfd(td);
842 return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
846 if (fget_locked(fdp, old) == NULL)
848 if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
849 td->td_retval[0] = new;
850 if (flags & FDDUP_FLAG_CLOEXEC)
851 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
857 * If the caller specified a file descriptor, make sure the file
858 * table is large enough to hold it, and grab it. Otherwise, just
859 * allocate a new descriptor the usual way.
864 if ((error = fdalloc(td, new, &new)) != 0)
867 case FDDUP_MUSTREPLACE:
868 /* Target file descriptor must exist. */
869 if (fget_locked(fdp, new) == NULL)
873 if (new >= fdp->fd_nfiles) {
875 * The resource limits are here instead of e.g.
876 * fdalloc(), because the file descriptor table may be
877 * shared between processes, so we can't really use
878 * racct_add()/racct_sub(). Instead of counting the
879 * number of actually allocated descriptors, just put
880 * the limit on the size of the file descriptor table.
885 error = racct_set(p, RACCT_NOFILE, new + 1);
893 fdgrowtable_exp(fdp, new + 1);
895 if (!fdisused(fdp, new))
899 KASSERT(0, ("%s unsupported mode %d", __func__, mode));
902 KASSERT(old != new, ("new fd is same as old"));
904 oldfde = &fdp->fd_ofiles[old];
905 fhold(oldfde->fde_file);
906 newfde = &fdp->fd_ofiles[new];
907 delfp = newfde->fde_file;
909 oioctls = filecaps_free_prep(&newfde->fde_caps);
910 nioctls = filecaps_copy_prep(&oldfde->fde_caps);
913 * Duplicate the source descriptor.
916 seq_write_begin(&newfde->fde_seq);
918 memcpy(newfde, oldfde, fde_change_size);
919 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
921 if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
922 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
924 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
926 seq_write_end(&newfde->fde_seq);
928 filecaps_free_finish(oioctls);
929 td->td_retval[0] = new;
934 (void) closefp(fdp, new, delfp, td, 1);
935 FILEDESC_UNLOCK_ASSERT(fdp);
938 FILEDESC_XUNLOCK(fdp);
945 * If sigio is on the list associated with a process or process group,
946 * disable signalling from the device, remove sigio from the list and
950 funsetown(struct sigio **sigiop)
962 *(sigio->sio_myref) = NULL;
963 if ((sigio)->sio_pgid < 0) {
964 struct pgrp *pg = (sigio)->sio_pgrp;
966 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
970 struct proc *p = (sigio)->sio_proc;
972 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
977 crfree(sigio->sio_ucred);
978 free(sigio, M_SIGIO);
982 * Free a list of sigio structures.
983 * We only need to lock the SIGIO_LOCK because we have made ourselves
984 * inaccessible to callers of fsetown and therefore do not need to lock
985 * the proc or pgrp struct for the list manipulation.
988 funsetownlst(struct sigiolst *sigiolst)
994 sigio = SLIST_FIRST(sigiolst);
1001 * Every entry of the list should belong
1002 * to a single proc or pgrp.
1004 if (sigio->sio_pgid < 0) {
1005 pg = sigio->sio_pgrp;
1006 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
1007 } else /* if (sigio->sio_pgid > 0) */ {
1008 p = sigio->sio_proc;
1009 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1013 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1014 *(sigio->sio_myref) = NULL;
1016 KASSERT(sigio->sio_pgid < 0,
1017 ("Proc sigio in pgrp sigio list"));
1018 KASSERT(sigio->sio_pgrp == pg,
1019 ("Bogus pgrp in sigio list"));
1021 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1024 } else /* if (p != NULL) */ {
1025 KASSERT(sigio->sio_pgid > 0,
1026 ("Pgrp sigio in proc sigio list"));
1027 KASSERT(sigio->sio_proc == p,
1028 ("Bogus proc in sigio list"));
1030 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1035 crfree(sigio->sio_ucred);
1036 free(sigio, M_SIGIO);
1043 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1045 * After permission checking, add a sigio structure to the sigio list for
1046 * the process or process group.
1049 fsetown(pid_t pgid, struct sigio **sigiop)
1053 struct sigio *sigio;
1063 /* Allocate and fill in the new sigio out of locks. */
1064 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1065 sigio->sio_pgid = pgid;
1066 sigio->sio_ucred = crhold(curthread->td_ucred);
1067 sigio->sio_myref = sigiop;
1069 sx_slock(&proctree_lock);
1078 * Policy - Don't allow a process to FSETOWN a process
1079 * in another session.
1081 * Remove this test to allow maximum flexibility or
1082 * restrict FSETOWN to the current process or process
1083 * group for maximum safety.
1086 if (proc->p_session != curthread->td_proc->p_session) {
1092 } else /* if (pgid < 0) */ {
1093 pgrp = pgfind(-pgid);
1101 * Policy - Don't allow a process to FSETOWN a process
1102 * in another session.
1104 * Remove this test to allow maximum flexibility or
1105 * restrict FSETOWN to the current process or process
1106 * group for maximum safety.
1108 if (pgrp->pg_session != curthread->td_proc->p_session) {
1119 * Since funsetownlst() is called without the proctree
1120 * locked, we need to check for P_WEXIT.
1121 * XXX: is ESRCH correct?
1123 if ((proc->p_flag & P_WEXIT) != 0) {
1128 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1129 sigio->sio_proc = proc;
1133 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1134 sigio->sio_pgrp = pgrp;
1137 sx_sunlock(&proctree_lock);
1144 sx_sunlock(&proctree_lock);
1145 crfree(sigio->sio_ucred);
1146 free(sigio, M_SIGIO);
1151 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1154 fgetown(struct sigio **sigiop)
1159 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1165 * Function drops the filedesc lock on return.
1168 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1173 FILEDESC_XLOCK_ASSERT(fdp);
1176 if (td->td_proc->p_fdtol != NULL) {
1178 * Ask fdfree() to sleep to ensure that all relevant
1179 * process leaders can be traversed in closef().
1181 fdp->fd_holdleaderscount++;
1188 * We now hold the fp reference that used to be owned by the
1189 * descriptor array. We have to unlock the FILEDESC *AFTER*
1190 * knote_fdclose to prevent a race of the fd getting opened, a knote
1191 * added, and deleteing a knote for the new fd.
1193 knote_fdclose(td, fd);
1196 * We need to notify mqueue if the object is of type mqueue.
1198 if (fp->f_type == DTYPE_MQUEUE)
1199 mq_fdclose(td, fd, fp);
1200 FILEDESC_XUNLOCK(fdp);
1202 error = closef(fp, td);
1204 FILEDESC_XLOCK(fdp);
1205 fdp->fd_holdleaderscount--;
1206 if (fdp->fd_holdleaderscount == 0 &&
1207 fdp->fd_holdleaderswakeup != 0) {
1208 fdp->fd_holdleaderswakeup = 0;
1209 wakeup(&fdp->fd_holdleaderscount);
1211 FILEDESC_XUNLOCK(fdp);
1217 * Close a file descriptor.
1219 #ifndef _SYS_SYSPROTO_H_
1226 sys_close(struct thread *td, struct close_args *uap)
1229 return (kern_close(td, uap->fd));
1233 kern_close(struct thread *td, int fd)
1235 struct filedesc *fdp;
1238 fdp = td->td_proc->p_fd;
1240 AUDIT_SYSCLOSE(td, fd);
1242 FILEDESC_XLOCK(fdp);
1243 if ((fp = fget_locked(fdp, fd)) == NULL) {
1244 FILEDESC_XUNLOCK(fdp);
1249 /* closefp() drops the FILEDESC lock for us. */
1250 return (closefp(fdp, fd, fp, td, 1));
1254 * Close open file descriptors.
1256 #ifndef _SYS_SYSPROTO_H_
1257 struct closefrom_args {
1263 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1265 struct filedesc *fdp;
1268 fdp = td->td_proc->p_fd;
1269 AUDIT_ARG_FD(uap->lowfd);
1272 * Treat negative starting file descriptor values identical to
1273 * closefrom(0) which closes all files.
1277 FILEDESC_SLOCK(fdp);
1278 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1279 if (fdp->fd_ofiles[fd].fde_file != NULL) {
1280 FILEDESC_SUNLOCK(fdp);
1281 (void)kern_close(td, fd);
1282 FILEDESC_SLOCK(fdp);
1285 FILEDESC_SUNLOCK(fdp);
1289 #if defined(COMPAT_43)
1291 * Return status information about a file descriptor.
1293 #ifndef _SYS_SYSPROTO_H_
1294 struct ofstat_args {
1301 ofstat(struct thread *td, struct ofstat_args *uap)
1307 error = kern_fstat(td, uap->fd, &ub);
1310 error = copyout(&oub, uap->sb, sizeof(oub));
1314 #endif /* COMPAT_43 */
1316 #if defined(COMPAT_FREEBSD11)
1318 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1321 struct freebsd11_stat osb;
1324 error = kern_fstat(td, uap->fd, &sb);
1327 error = freebsd11_cvtstat(&sb, &osb);
1329 error = copyout(&osb, uap->sb, sizeof(osb));
1332 #endif /* COMPAT_FREEBSD11 */
1335 * Return status information about a file descriptor.
1337 #ifndef _SYS_SYSPROTO_H_
1345 sys_fstat(struct thread *td, struct fstat_args *uap)
1350 error = kern_fstat(td, uap->fd, &ub);
1352 error = copyout(&ub, uap->sb, sizeof(ub));
1357 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1364 error = fget(td, fd, &cap_fstat_rights, &fp);
1368 AUDIT_ARG_FILE(td->td_proc, fp);
1370 error = fo_stat(fp, sbp, td->td_ucred, td);
1372 #ifdef __STAT_TIME_T_EXT
1374 sbp->st_atim_ext = 0;
1375 sbp->st_mtim_ext = 0;
1376 sbp->st_ctim_ext = 0;
1377 sbp->st_btim_ext = 0;
1381 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1387 #if defined(COMPAT_FREEBSD11)
1389 * Return status information about a file descriptor.
1391 #ifndef _SYS_SYSPROTO_H_
1392 struct freebsd11_nfstat_args {
1399 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1405 error = kern_fstat(td, uap->fd, &ub);
1407 freebsd11_cvtnstat(&ub, &nub);
1408 error = copyout(&nub, uap->sb, sizeof(nub));
1412 #endif /* COMPAT_FREEBSD11 */
1415 * Return pathconf information about a file descriptor.
1417 #ifndef _SYS_SYSPROTO_H_
1418 struct fpathconf_args {
1425 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1430 error = kern_fpathconf(td, uap->fd, uap->name, &value);
1432 td->td_retval[0] = value;
1437 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1443 error = fget(td, fd, &cap_fpathconf_rights, &fp);
1447 if (name == _PC_ASYNC_IO) {
1448 *valuep = _POSIX_ASYNCHRONOUS_IO;
1453 vn_lock(vp, LK_SHARED | LK_RETRY);
1454 error = VOP_PATHCONF(vp, name, valuep);
1456 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1457 if (name != _PC_PIPE_BUF) {
1472 * Initialize filecaps structure.
1475 filecaps_init(struct filecaps *fcaps)
1478 bzero(fcaps, sizeof(*fcaps));
1479 fcaps->fc_nioctls = -1;
1483 * Copy filecaps structure allocating memory for ioctls array if needed.
1485 * The last parameter indicates whether the fdtable is locked. If it is not and
1486 * ioctls are encountered, copying fails and the caller must lock the table.
1488 * Note that if the table was not locked, the caller has to check the relevant
1489 * sequence counter to determine whether the operation was successful.
1492 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1496 if (src->fc_ioctls != NULL && !locked)
1498 memcpy(dst, src, sizeof(*src));
1499 if (src->fc_ioctls == NULL)
1502 KASSERT(src->fc_nioctls > 0,
1503 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1505 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1506 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1507 memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1512 filecaps_copy_prep(const struct filecaps *src)
1517 if (src->fc_ioctls == NULL)
1520 KASSERT(src->fc_nioctls > 0,
1521 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1523 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1524 ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1529 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1535 if (src->fc_ioctls == NULL) {
1536 MPASS(ioctls == NULL);
1540 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1541 dst->fc_ioctls = ioctls;
1542 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1546 * Move filecaps structure to the new place and clear the old place.
1549 filecaps_move(struct filecaps *src, struct filecaps *dst)
1553 bzero(src, sizeof(*src));
1557 * Fill the given filecaps structure with full rights.
1560 filecaps_fill(struct filecaps *fcaps)
1563 CAP_ALL(&fcaps->fc_rights);
1564 fcaps->fc_ioctls = NULL;
1565 fcaps->fc_nioctls = -1;
1566 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1570 * Free memory allocated within filecaps structure.
1573 filecaps_free(struct filecaps *fcaps)
1576 free(fcaps->fc_ioctls, M_FILECAPS);
1577 bzero(fcaps, sizeof(*fcaps));
1581 filecaps_free_prep(struct filecaps *fcaps)
1585 ioctls = fcaps->fc_ioctls;
1586 bzero(fcaps, sizeof(*fcaps));
1591 filecaps_free_finish(u_long *ioctls)
1594 free(ioctls, M_FILECAPS);
1598 * Validate the given filecaps structure.
1601 filecaps_validate(const struct filecaps *fcaps, const char *func)
1604 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1605 ("%s: invalid rights", func));
1606 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1607 ("%s: invalid fcntls", func));
1608 KASSERT(fcaps->fc_fcntls == 0 ||
1609 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1610 ("%s: fcntls without CAP_FCNTL", func));
1611 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1612 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1613 ("%s: invalid ioctls", func));
1614 KASSERT(fcaps->fc_nioctls == 0 ||
1615 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1616 ("%s: ioctls without CAP_IOCTL", func));
1620 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1624 FILEDESC_XLOCK_ASSERT(fdp);
1626 nfd1 = fdp->fd_nfiles * 2;
1629 fdgrowtable(fdp, nfd1);
1633 * Grow the file table to accommodate (at least) nfd descriptors.
1636 fdgrowtable(struct filedesc *fdp, int nfd)
1638 struct filedesc0 *fdp0;
1639 struct freetable *ft;
1640 struct fdescenttbl *ntable;
1641 struct fdescenttbl *otable;
1642 int nnfiles, onfiles;
1643 NDSLOTTYPE *nmap, *omap;
1646 * If lastfile is -1 this struct filedesc was just allocated and we are
1647 * growing it to accommodate for the one we are going to copy from. There
1648 * is no need to have a lock on this one as it's not visible to anyone.
1650 if (fdp->fd_lastfile != -1)
1651 FILEDESC_XLOCK_ASSERT(fdp);
1653 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1655 /* save old values */
1656 onfiles = fdp->fd_nfiles;
1657 otable = fdp->fd_files;
1660 /* compute the size of the new table */
1661 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1662 if (nnfiles <= onfiles)
1663 /* the table is already large enough */
1667 * Allocate a new table. We need enough space for the number of
1668 * entries, file entries themselves and the struct freetable we will use
1669 * when we decommission the table and place it on the freelist.
1670 * We place the struct freetable in the middle so we don't have
1671 * to worry about padding.
1673 ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1674 nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1675 sizeof(struct freetable),
1676 M_FILEDESC, M_ZERO | M_WAITOK);
1677 /* copy the old data */
1678 ntable->fdt_nfiles = nnfiles;
1679 memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1680 onfiles * sizeof(ntable->fdt_ofiles[0]));
1683 * Allocate a new map only if the old is not large enough. It will
1684 * grow at a slower rate than the table as it can map more
1685 * entries than the table can hold.
1687 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1688 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1690 /* copy over the old data and update the pointer */
1691 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1696 * Make sure that ntable is correctly initialized before we replace
1697 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1700 atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1703 * Do not free the old file table, as some threads may still
1704 * reference entries within it. Instead, place it on a freelist
1705 * which will be processed when the struct filedesc is released.
1707 * Note that if onfiles == NDFILE, we're dealing with the original
1708 * static allocation contained within (struct filedesc0 *)fdp,
1709 * which must not be freed.
1711 if (onfiles > NDFILE) {
1712 ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1713 fdp0 = (struct filedesc0 *)fdp;
1714 ft->ft_table = otable;
1715 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1718 * The map does not have the same possibility of threads still
1719 * holding references to it. So always free it as long as it
1720 * does not reference the original static allocation.
1722 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1723 free(omap, M_FILEDESC);
1727 * Allocate a file descriptor for the process.
1730 fdalloc(struct thread *td, int minfd, int *result)
1732 struct proc *p = td->td_proc;
1733 struct filedesc *fdp = p->p_fd;
1734 int fd, maxfd, allocfd;
1739 FILEDESC_XLOCK_ASSERT(fdp);
1741 if (fdp->fd_freefile > minfd)
1742 minfd = fdp->fd_freefile;
1744 maxfd = getmaxfd(td);
1747 * Search the bitmap for a free descriptor starting at minfd.
1748 * If none is found, grow the file table.
1750 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1753 if (fd >= fdp->fd_nfiles) {
1754 allocfd = min(fd * 2, maxfd);
1758 error = racct_set(p, RACCT_NOFILE, allocfd);
1765 * fd is already equal to first free descriptor >= minfd, so
1766 * we only need to grow the table and we are done.
1768 fdgrowtable_exp(fdp, allocfd);
1772 * Perform some sanity checks, then mark the file descriptor as
1773 * used and return it to the caller.
1775 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1776 ("invalid descriptor %d", fd));
1777 KASSERT(!fdisused(fdp, fd),
1778 ("fd_first_free() returned non-free descriptor"));
1779 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1780 ("file descriptor isn't free"));
1787 * Allocate n file descriptors for the process.
1790 fdallocn(struct thread *td, int minfd, int *fds, int n)
1792 struct proc *p = td->td_proc;
1793 struct filedesc *fdp = p->p_fd;
1796 FILEDESC_XLOCK_ASSERT(fdp);
1798 for (i = 0; i < n; i++)
1799 if (fdalloc(td, 0, &fds[i]) != 0)
1803 for (i--; i >= 0; i--)
1804 fdunused(fdp, fds[i]);
1812 * Create a new open file structure and allocate a file descriptor for the
1813 * process that refers to it. We add one reference to the file for the
1814 * descriptor table and one reference for resultfp. This is to prevent us
1815 * being preempted and the entry in the descriptor table closed after we
1816 * release the FILEDESC lock.
1819 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1820 struct filecaps *fcaps)
1825 error = falloc_noinstall(td, &fp);
1827 return (error); /* no reference held on error */
1829 error = finstall(td, fp, &fd, flags, fcaps);
1831 fdrop(fp, td); /* one reference (fp only) */
1835 if (resultfp != NULL)
1836 *resultfp = fp; /* copy out result */
1838 fdrop(fp, td); /* release local reference */
1840 if (resultfd != NULL)
1847 * Create a new open file structure without allocating a file descriptor.
1850 falloc_noinstall(struct thread *td, struct file **resultfp)
1853 int maxuserfiles = maxfiles - (maxfiles / 20);
1855 static struct timeval lastfail;
1858 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1860 openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1861 if ((openfiles_new >= maxuserfiles &&
1862 priv_check(td, PRIV_MAXFILES) != 0) ||
1863 openfiles_new >= maxfiles) {
1864 atomic_subtract_int(&openfiles, 1);
1865 if (ppsratecheck(&lastfail, &curfail, 1)) {
1866 printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1867 "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1871 fp = uma_zalloc(file_zone, M_WAITOK);
1872 bzero(fp, sizeof(*fp));
1873 refcount_init(&fp->f_count, 1);
1874 fp->f_cred = crhold(td->td_ucred);
1875 fp->f_ops = &badfileops;
1881 * Install a file in a file descriptor table.
1884 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1885 struct filecaps *fcaps)
1887 struct filedescent *fde;
1891 filecaps_validate(fcaps, __func__);
1892 FILEDESC_XLOCK_ASSERT(fdp);
1894 fde = &fdp->fd_ofiles[fd];
1896 seq_write_begin(&fde->fde_seq);
1899 fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1901 filecaps_move(fcaps, &fde->fde_caps);
1903 filecaps_fill(&fde->fde_caps);
1905 seq_write_end(&fde->fde_seq);
1910 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1911 struct filecaps *fcaps)
1913 struct filedesc *fdp = td->td_proc->p_fd;
1918 FILEDESC_XLOCK(fdp);
1919 if ((error = fdalloc(td, 0, fd))) {
1920 FILEDESC_XUNLOCK(fdp);
1924 _finstall(fdp, fp, *fd, flags, fcaps);
1925 FILEDESC_XUNLOCK(fdp);
1930 * Build a new filedesc structure from another.
1931 * Copy the current, root, and jail root vnode references.
1933 * If fdp is not NULL, return with it shared locked.
1936 fdinit(struct filedesc *fdp, bool prepfiles)
1938 struct filedesc0 *newfdp0;
1939 struct filedesc *newfdp;
1941 newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1942 newfdp = &newfdp0->fd_fd;
1944 /* Create the file descriptor table. */
1945 FILEDESC_LOCK_INIT(newfdp);
1946 refcount_init(&newfdp->fd_refcnt, 1);
1947 refcount_init(&newfdp->fd_holdcnt, 1);
1948 newfdp->fd_cmask = CMASK;
1949 newfdp->fd_map = newfdp0->fd_dmap;
1950 newfdp->fd_lastfile = -1;
1951 newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1952 newfdp->fd_files->fdt_nfiles = NDFILE;
1957 if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1958 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1960 FILEDESC_SLOCK(fdp);
1961 newfdp->fd_cdir = fdp->fd_cdir;
1962 if (newfdp->fd_cdir)
1963 vrefact(newfdp->fd_cdir);
1964 newfdp->fd_rdir = fdp->fd_rdir;
1965 if (newfdp->fd_rdir)
1966 vrefact(newfdp->fd_rdir);
1967 newfdp->fd_jdir = fdp->fd_jdir;
1968 if (newfdp->fd_jdir)
1969 vrefact(newfdp->fd_jdir);
1972 FILEDESC_SUNLOCK(fdp);
1974 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1975 FILEDESC_SUNLOCK(fdp);
1976 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1977 FILEDESC_SLOCK(fdp);
1984 static struct filedesc *
1985 fdhold(struct proc *p)
1987 struct filedesc *fdp;
1989 PROC_LOCK_ASSERT(p, MA_OWNED);
1992 refcount_acquire(&fdp->fd_holdcnt);
1997 fddrop(struct filedesc *fdp)
2000 if (fdp->fd_holdcnt > 1) {
2001 if (refcount_release(&fdp->fd_holdcnt) == 0)
2005 FILEDESC_LOCK_DESTROY(fdp);
2006 uma_zfree(filedesc0_zone, fdp);
2010 * Share a filedesc structure.
2013 fdshare(struct filedesc *fdp)
2016 refcount_acquire(&fdp->fd_refcnt);
2021 * Unshare a filedesc structure, if necessary by making a copy
2024 fdunshare(struct thread *td)
2026 struct filedesc *tmp;
2027 struct proc *p = td->td_proc;
2029 if (p->p_fd->fd_refcnt == 1)
2032 tmp = fdcopy(p->p_fd);
2038 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2042 td->td_proc->p_fd = fdp;
2046 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
2047 * this is to ease callers, not catch errors.
2050 fdcopy(struct filedesc *fdp)
2052 struct filedesc *newfdp;
2053 struct filedescent *nfde, *ofde;
2058 newfdp = fdinit(fdp, true);
2059 /* copy all passable descriptors (i.e. not kqueue) */
2060 newfdp->fd_freefile = -1;
2061 for (i = 0; i <= fdp->fd_lastfile; ++i) {
2062 ofde = &fdp->fd_ofiles[i];
2063 if (ofde->fde_file == NULL ||
2064 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2065 if (newfdp->fd_freefile == -1)
2066 newfdp->fd_freefile = i;
2069 nfde = &newfdp->fd_ofiles[i];
2071 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2072 fhold(nfde->fde_file);
2073 fdused_init(newfdp, i);
2074 newfdp->fd_lastfile = i;
2076 if (newfdp->fd_freefile == -1)
2077 newfdp->fd_freefile = i;
2078 newfdp->fd_cmask = fdp->fd_cmask;
2079 FILEDESC_SUNLOCK(fdp);
2084 * Copies a filedesc structure, while remapping all file descriptors
2085 * stored inside using a translation table.
2087 * File descriptors are copied over to the new file descriptor table,
2088 * regardless of whether the close-on-exec flag is set.
2091 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2092 struct filedesc **ret)
2094 struct filedesc *newfdp;
2095 struct filedescent *nfde, *ofde;
2100 newfdp = fdinit(fdp, true);
2101 if (nfds > fdp->fd_lastfile + 1) {
2102 /* New table cannot be larger than the old one. */
2106 /* Copy all passable descriptors (i.e. not kqueue). */
2107 newfdp->fd_freefile = nfds;
2108 for (i = 0; i < nfds; ++i) {
2109 if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2110 /* File descriptor out of bounds. */
2114 ofde = &fdp->fd_ofiles[fds[i]];
2115 if (ofde->fde_file == NULL) {
2116 /* Unused file descriptor. */
2120 if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2121 /* File descriptor cannot be passed. */
2125 nfde = &newfdp->fd_ofiles[i];
2127 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2128 fhold(nfde->fde_file);
2129 fdused_init(newfdp, i);
2130 newfdp->fd_lastfile = i;
2132 newfdp->fd_cmask = fdp->fd_cmask;
2133 FILEDESC_SUNLOCK(fdp);
2137 FILEDESC_SUNLOCK(fdp);
2138 fdescfree_remapped(newfdp);
2143 * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2144 * one of processes using it exits) and the table used to be shared.
2147 fdclearlocks(struct thread *td)
2149 struct filedesc *fdp;
2150 struct filedesc_to_leader *fdtol;
2160 MPASS(fdtol != NULL);
2162 FILEDESC_XLOCK(fdp);
2163 KASSERT(fdtol->fdl_refcount > 0,
2164 ("filedesc_to_refcount botch: fdl_refcount=%d",
2165 fdtol->fdl_refcount));
2166 if (fdtol->fdl_refcount == 1 &&
2167 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2168 for (i = 0; i <= fdp->fd_lastfile; i++) {
2169 fp = fdp->fd_ofiles[i].fde_file;
2170 if (fp == NULL || fp->f_type != DTYPE_VNODE)
2173 FILEDESC_XUNLOCK(fdp);
2174 lf.l_whence = SEEK_SET;
2177 lf.l_type = F_UNLCK;
2179 (void) VOP_ADVLOCK(vp,
2180 (caddr_t)p->p_leader, F_UNLCK,
2182 FILEDESC_XLOCK(fdp);
2187 if (fdtol->fdl_refcount == 1) {
2188 if (fdp->fd_holdleaderscount > 0 &&
2189 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2191 * close() or kern_dup() has cleared a reference
2192 * in a shared file descriptor table.
2194 fdp->fd_holdleaderswakeup = 1;
2195 sx_sleep(&fdp->fd_holdleaderscount,
2196 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2199 if (fdtol->fdl_holdcount > 0) {
2201 * Ensure that fdtol->fdl_leader remains
2202 * valid in closef().
2204 fdtol->fdl_wakeup = 1;
2205 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2210 fdtol->fdl_refcount--;
2211 if (fdtol->fdl_refcount == 0 &&
2212 fdtol->fdl_holdcount == 0) {
2213 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2214 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2218 FILEDESC_XUNLOCK(fdp);
2220 free(fdtol, M_FILEDESC_TO_LEADER);
2224 * Release a filedesc structure.
2227 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2229 struct filedesc0 *fdp0;
2230 struct freetable *ft, *tft;
2231 struct filedescent *fde;
2235 for (i = 0; i <= fdp->fd_lastfile; i++) {
2236 fde = &fdp->fd_ofiles[i];
2241 (void) closef(fp, td);
2247 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2248 free(fdp->fd_map, M_FILEDESC);
2249 if (fdp->fd_nfiles > NDFILE)
2250 free(fdp->fd_files, M_FILEDESC);
2252 fdp0 = (struct filedesc0 *)fdp;
2253 SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2254 free(ft->ft_table, M_FILEDESC);
2260 fdescfree(struct thread *td)
2263 struct filedesc *fdp;
2264 struct vnode *cdir, *jdir, *rdir;
2273 racct_set(p, RACCT_NOFILE, 0);
2278 if (p->p_fdtol != NULL)
2285 if (refcount_release(&fdp->fd_refcnt) == 0)
2288 FILEDESC_XLOCK(fdp);
2289 cdir = fdp->fd_cdir;
2290 fdp->fd_cdir = NULL;
2291 rdir = fdp->fd_rdir;
2292 fdp->fd_rdir = NULL;
2293 jdir = fdp->fd_jdir;
2294 fdp->fd_jdir = NULL;
2295 FILEDESC_XUNLOCK(fdp);
2304 fdescfree_fds(td, fdp, 1);
2308 fdescfree_remapped(struct filedesc *fdp)
2311 if (fdp->fd_cdir != NULL)
2312 vrele(fdp->fd_cdir);
2313 if (fdp->fd_rdir != NULL)
2314 vrele(fdp->fd_rdir);
2315 if (fdp->fd_jdir != NULL)
2316 vrele(fdp->fd_jdir);
2318 fdescfree_fds(curthread, fdp, 0);
2322 * For setugid programs, we don't want to people to use that setugidness
2323 * to generate error messages which write to a file which otherwise would
2324 * otherwise be off-limits to the process. We check for filesystems where
2325 * the vnode can change out from under us after execve (like [lin]procfs).
2327 * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2328 * sufficient. We also don't check for setugidness since we know we are.
2331 is_unsafe(struct file *fp)
2335 if (fp->f_type != DTYPE_VNODE)
2339 return ((vp->v_vflag & VV_PROCDEP) != 0);
2343 * Make this setguid thing safe, if at all possible.
2346 fdsetugidsafety(struct thread *td)
2348 struct filedesc *fdp;
2352 fdp = td->td_proc->p_fd;
2353 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2354 MPASS(fdp->fd_nfiles >= 3);
2355 for (i = 0; i <= 2; i++) {
2356 fp = fdp->fd_ofiles[i].fde_file;
2357 if (fp != NULL && is_unsafe(fp)) {
2358 FILEDESC_XLOCK(fdp);
2359 knote_fdclose(td, i);
2361 * NULL-out descriptor prior to close to avoid
2362 * a race while close blocks.
2365 FILEDESC_XUNLOCK(fdp);
2366 (void) closef(fp, td);
2372 * If a specific file object occupies a specific file descriptor, close the
2373 * file descriptor entry and drop a reference on the file object. This is a
2374 * convenience function to handle a subsequent error in a function that calls
2375 * falloc() that handles the race that another thread might have closed the
2376 * file descriptor out from under the thread creating the file object.
2379 fdclose(struct thread *td, struct file *fp, int idx)
2381 struct filedesc *fdp = td->td_proc->p_fd;
2383 FILEDESC_XLOCK(fdp);
2384 if (fdp->fd_ofiles[idx].fde_file == fp) {
2386 FILEDESC_XUNLOCK(fdp);
2389 FILEDESC_XUNLOCK(fdp);
2393 * Close any files on exec?
2396 fdcloseexec(struct thread *td)
2398 struct filedesc *fdp;
2399 struct filedescent *fde;
2403 fdp = td->td_proc->p_fd;
2404 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2405 for (i = 0; i <= fdp->fd_lastfile; i++) {
2406 fde = &fdp->fd_ofiles[i];
2408 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2409 (fde->fde_flags & UF_EXCLOSE))) {
2410 FILEDESC_XLOCK(fdp);
2412 (void) closefp(fdp, i, fp, td, 0);
2413 FILEDESC_UNLOCK_ASSERT(fdp);
2419 * It is unsafe for set[ug]id processes to be started with file
2420 * descriptors 0..2 closed, as these descriptors are given implicit
2421 * significance in the Standard C library. fdcheckstd() will create a
2422 * descriptor referencing /dev/null for each of stdin, stdout, and
2423 * stderr that is not already open.
2426 fdcheckstd(struct thread *td)
2428 struct filedesc *fdp;
2430 int i, error, devnull;
2432 fdp = td->td_proc->p_fd;
2433 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2434 MPASS(fdp->fd_nfiles >= 3);
2436 for (i = 0; i <= 2; i++) {
2437 if (fdp->fd_ofiles[i].fde_file != NULL)
2440 save = td->td_retval[0];
2441 if (devnull != -1) {
2442 error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2444 error = kern_openat(td, AT_FDCWD, "/dev/null",
2445 UIO_SYSSPACE, O_RDWR, 0);
2447 devnull = td->td_retval[0];
2448 KASSERT(devnull == i, ("we didn't get our fd"));
2451 td->td_retval[0] = save;
2459 * Internal form of close. Decrement reference count on file structure.
2460 * Note: td may be NULL when closing a file that was being passed in a
2464 closef(struct file *fp, struct thread *td)
2468 struct filedesc_to_leader *fdtol;
2469 struct filedesc *fdp;
2472 * POSIX record locking dictates that any close releases ALL
2473 * locks owned by this process. This is handled by setting
2474 * a flag in the unlock to free ONLY locks obeying POSIX
2475 * semantics, and not to free BSD-style file locks.
2476 * If the descriptor was in a message, POSIX-style locks
2477 * aren't passed with the descriptor, and the thread pointer
2478 * will be NULL. Callers should be careful only to pass a
2479 * NULL thread pointer when there really is no owning
2480 * context that might have locks, or the locks will be
2483 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2485 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2486 lf.l_whence = SEEK_SET;
2489 lf.l_type = F_UNLCK;
2490 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2491 F_UNLCK, &lf, F_POSIX);
2493 fdtol = td->td_proc->p_fdtol;
2494 if (fdtol != NULL) {
2496 * Handle special case where file descriptor table is
2497 * shared between multiple process leaders.
2499 fdp = td->td_proc->p_fd;
2500 FILEDESC_XLOCK(fdp);
2501 for (fdtol = fdtol->fdl_next;
2502 fdtol != td->td_proc->p_fdtol;
2503 fdtol = fdtol->fdl_next) {
2504 if ((fdtol->fdl_leader->p_flag &
2507 fdtol->fdl_holdcount++;
2508 FILEDESC_XUNLOCK(fdp);
2509 lf.l_whence = SEEK_SET;
2512 lf.l_type = F_UNLCK;
2514 (void) VOP_ADVLOCK(vp,
2515 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2517 FILEDESC_XLOCK(fdp);
2518 fdtol->fdl_holdcount--;
2519 if (fdtol->fdl_holdcount == 0 &&
2520 fdtol->fdl_wakeup != 0) {
2521 fdtol->fdl_wakeup = 0;
2525 FILEDESC_XUNLOCK(fdp);
2528 return (fdrop(fp, td));
2532 * Initialize the file pointer with the specified properties.
2534 * The ops are set with release semantics to be certain that the flags, type,
2535 * and data are visible when ops is. This is to prevent ops methods from being
2536 * called with bad data.
2539 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2544 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2548 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2549 struct file **fpp, struct filecaps *havecapsp)
2551 struct filedescent *fde;
2554 FILEDESC_LOCK_ASSERT(fdp);
2556 fde = fdeget_locked(fdp, fd);
2563 error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2568 if (havecapsp != NULL)
2569 filecaps_copy(&fde->fde_caps, havecapsp, true);
2571 *fpp = fde->fde_file;
2579 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2580 struct file **fpp, struct filecaps *havecapsp)
2582 struct filedesc *fdp = td->td_proc->p_fd;
2584 #ifndef CAPABILITIES
2585 error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2586 if (error == 0 && havecapsp != NULL)
2587 filecaps_fill(havecapsp);
2593 error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2597 if (havecapsp != NULL) {
2598 if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2599 havecapsp, false)) {
2605 if (!fd_modified(fdp, fd, seq))
2614 FILEDESC_SLOCK(fdp);
2615 error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2618 FILEDESC_SUNLOCK(fdp);
2624 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2625 struct file **fpp, seq_t *seqp)
2628 const struct filedescent *fde;
2630 const struct fdescenttbl *fdt;
2635 cap_rights_t haverights;
2639 fdt = fdp->fd_files;
2640 if ((u_int)fd >= fdt->fdt_nfiles)
2643 * Fetch the descriptor locklessly. We avoid fdrop() races by
2644 * never raising a refcount above 0. To accomplish this we have
2645 * to use a cmpset loop rather than an atomic_add. The descriptor
2646 * must be re-verified once we acquire a reference to be certain
2647 * that the identity is still correct and we did not lose a race
2648 * due to preemption.
2652 seq = seq_load(fd_seq(fdt, fd));
2653 fde = &fdt->fdt_ofiles[fd];
2654 haverights = *cap_rights_fde_inline(fde);
2656 if (!seq_consistent(fd_seq(fdt, fd), seq))
2659 fp = fdt->fdt_ofiles[fd].fde_file;
2664 error = cap_check(&haverights, needrightsp);
2668 count = fp->f_count;
2672 * Force a reload. Other thread could reallocate the
2673 * table before this fd was closed, so it possible that
2674 * there is a stale fp pointer in cached version.
2676 fdt = *(const struct fdescenttbl * const volatile *)&(fdp->fd_files);
2680 * Use an acquire barrier to force re-reading of fdt so it is
2681 * refreshed for verification.
2683 if (atomic_fcmpset_acq_int(&fp->f_count, &count, count + 1) == 0)
2685 fdt = fdp->fd_files;
2687 if (seq_consistent_nomb(fd_seq(fdt, fd), seq))
2689 if (fp == fdt->fdt_ofiles[fd].fde_file)
2692 fdrop(fp, curthread);
2704 * Extract the file pointer associated with the specified descriptor for the
2705 * current user process.
2707 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2710 * File's rights will be checked against the capability rights mask.
2712 * If an error occurred the non-zero error is returned and *fpp is set to
2713 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
2714 * responsible for fdrop().
2717 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2718 cap_rights_t *needrightsp, seq_t *seqp)
2720 struct filedesc *fdp;
2725 fdp = td->td_proc->p_fd;
2726 error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2729 if (fp->f_ops == &badfileops) {
2735 * FREAD and FWRITE failure return EBADF as per POSIX.
2741 if ((fp->f_flag & flags) == 0)
2745 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2746 ((fp->f_flag & FWRITE) != 0))
2752 KASSERT(0, ("wrong flags"));
2765 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2768 return (_fget(td, fd, fpp, 0, rightsp, NULL));
2772 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2776 #ifndef CAPABILITIES
2777 error = _fget(td, fd, fpp, 0, rightsp, NULL);
2778 if (maxprotp != NULL)
2779 *maxprotp = VM_PROT_ALL;
2781 cap_rights_t fdrights;
2782 struct filedesc *fdp = td->td_proc->p_fd;
2785 MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2787 error = _fget(td, fd, fpp, 0, rightsp, &seq);
2790 if (maxprotp != NULL)
2791 fdrights = *cap_rights(fdp, fd);
2792 if (!fd_modified(fdp, fd, seq))
2798 * If requested, convert capability rights to access flags.
2800 if (maxprotp != NULL)
2801 *maxprotp = cap_rights_to_vmprot(&fdrights);
2807 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2810 return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2814 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2817 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2821 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2824 struct filedesc *fdp = td->td_proc->p_fd;
2825 #ifndef CAPABILITIES
2826 return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2831 MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2833 error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2836 error = cap_fcntl_check(fdp, fd, needfcntl);
2837 if (!fd_modified(fdp, fd, seq))
2850 * Like fget() but loads the underlying vnode, or returns an error if the
2851 * descriptor does not represent a vnode. Note that pipes use vnodes but
2852 * never have VM objects. The returned vnode will be vref()'d.
2854 * XXX: what about the unused flags ?
2857 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2864 error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2867 if (fp->f_vnode == NULL) {
2879 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2882 return (_fgetvp(td, fd, 0, rightsp, vpp));
2886 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2887 struct filecaps *havecaps, struct vnode **vpp)
2889 struct filedesc *fdp;
2890 struct filecaps caps;
2894 fdp = td->td_proc->p_fd;
2895 error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2898 if (fp->f_ops == &badfileops) {
2902 if (fp->f_vnode == NULL) {
2913 filecaps_free(&caps);
2918 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2921 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2925 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2928 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2933 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2937 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2942 * Handle the last reference to a file being closed.
2944 * Without the noinline attribute clang keeps inlining the func thorough this
2945 * file when fdrop is used.
2948 _fdrop(struct file *fp, struct thread *td)
2952 if (fp->f_count != 0)
2953 panic("fdrop: count %d", fp->f_count);
2954 error = fo_close(fp, td);
2955 atomic_subtract_int(&openfiles, 1);
2957 free(fp->f_advice, M_FADVISE);
2958 uma_zfree(file_zone, fp);
2964 * Apply an advisory lock on a file descriptor.
2966 * Just attempt to get a record lock of the requested type on the entire file
2967 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2969 #ifndef _SYS_SYSPROTO_H_
2977 sys_flock(struct thread *td, struct flock_args *uap)
2984 error = fget(td, uap->fd, &cap_flock_rights, &fp);
2987 if (fp->f_type != DTYPE_VNODE) {
2989 return (EOPNOTSUPP);
2993 lf.l_whence = SEEK_SET;
2996 if (uap->how & LOCK_UN) {
2997 lf.l_type = F_UNLCK;
2998 atomic_clear_int(&fp->f_flag, FHASLOCK);
2999 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3002 if (uap->how & LOCK_EX)
3003 lf.l_type = F_WRLCK;
3004 else if (uap->how & LOCK_SH)
3005 lf.l_type = F_RDLCK;
3010 atomic_set_int(&fp->f_flag, FHASLOCK);
3011 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3012 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3018 * Duplicate the specified descriptor to a free descriptor.
3021 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3022 int openerror, int *indxp)
3024 struct filedescent *newfde, *oldfde;
3029 KASSERT(openerror == ENODEV || openerror == ENXIO,
3030 ("unexpected error %d in %s", openerror, __func__));
3033 * If the to-be-dup'd fd number is greater than the allowed number
3034 * of file descriptors, or the fd to be dup'd has already been
3035 * closed, then reject.
3037 FILEDESC_XLOCK(fdp);
3038 if ((fp = fget_locked(fdp, dfd)) == NULL) {
3039 FILEDESC_XUNLOCK(fdp);
3043 error = fdalloc(td, 0, &indx);
3045 FILEDESC_XUNLOCK(fdp);
3050 * There are two cases of interest here.
3052 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3054 * For ENXIO steal away the file structure from (dfd) and store it in
3055 * (indx). (dfd) is effectively closed by this operation.
3057 switch (openerror) {
3060 * Check that the mode the file is being opened for is a
3061 * subset of the mode of the existing descriptor.
3063 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3064 fdunused(fdp, indx);
3065 FILEDESC_XUNLOCK(fdp);
3069 newfde = &fdp->fd_ofiles[indx];
3070 oldfde = &fdp->fd_ofiles[dfd];
3071 ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3073 seq_write_begin(&newfde->fde_seq);
3075 memcpy(newfde, oldfde, fde_change_size);
3076 filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3079 seq_write_end(&newfde->fde_seq);
3084 * Steal away the file pointer from dfd and stuff it into indx.
3086 newfde = &fdp->fd_ofiles[indx];
3087 oldfde = &fdp->fd_ofiles[dfd];
3089 seq_write_begin(&newfde->fde_seq);
3091 memcpy(newfde, oldfde, fde_change_size);
3092 oldfde->fde_file = NULL;
3095 seq_write_end(&newfde->fde_seq);
3099 FILEDESC_XUNLOCK(fdp);
3105 * This sysctl determines if we will allow a process to chroot(2) if it
3106 * has a directory open:
3107 * 0: disallowed for all processes.
3108 * 1: allowed for processes that were not already chroot(2)'ed.
3109 * 2: allowed for all processes.
3112 static int chroot_allow_open_directories = 1;
3114 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3115 &chroot_allow_open_directories, 0,
3116 "Allow a process to chroot(2) if it has a directory open");
3119 * Helper function for raised chroot(2) security function: Refuse if
3120 * any filedescriptors are open directories.
3123 chroot_refuse_vdir_fds(struct filedesc *fdp)
3129 FILEDESC_LOCK_ASSERT(fdp);
3131 for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3132 fp = fget_locked(fdp, fd);
3135 if (fp->f_type == DTYPE_VNODE) {
3137 if (vp->v_type == VDIR)
3145 * Common routine for kern_chroot() and jail_attach(). The caller is
3146 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3147 * authorize this operation.
3150 pwd_chroot(struct thread *td, struct vnode *vp)
3152 struct filedesc *fdp;
3153 struct vnode *oldvp;
3156 fdp = td->td_proc->p_fd;
3157 FILEDESC_XLOCK(fdp);
3158 if (chroot_allow_open_directories == 0 ||
3159 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3160 error = chroot_refuse_vdir_fds(fdp);
3162 FILEDESC_XUNLOCK(fdp);
3166 oldvp = fdp->fd_rdir;
3169 if (fdp->fd_jdir == NULL) {
3173 FILEDESC_XUNLOCK(fdp);
3179 pwd_chdir(struct thread *td, struct vnode *vp)
3181 struct filedesc *fdp;
3182 struct vnode *oldvp;
3184 fdp = td->td_proc->p_fd;
3185 FILEDESC_XLOCK(fdp);
3186 VNASSERT(vp->v_usecount > 0, vp,
3187 ("chdir to a vnode with zero usecount"));
3188 oldvp = fdp->fd_cdir;
3190 FILEDESC_XUNLOCK(fdp);
3195 * Scan all active processes and prisons to see if any of them have a current
3196 * or root directory of `olddp'. If so, replace them with the new mount point.
3199 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3201 struct filedesc *fdp;
3206 if (vrefcnt(olddp) == 1)
3209 sx_slock(&allproc_lock);
3210 FOREACH_PROC_IN_SYSTEM(p) {
3216 FILEDESC_XLOCK(fdp);
3217 if (fdp->fd_cdir == olddp) {
3219 fdp->fd_cdir = newdp;
3222 if (fdp->fd_rdir == olddp) {
3224 fdp->fd_rdir = newdp;
3227 if (fdp->fd_jdir == olddp) {
3229 fdp->fd_jdir = newdp;
3232 FILEDESC_XUNLOCK(fdp);
3235 sx_sunlock(&allproc_lock);
3236 if (rootvnode == olddp) {
3241 mtx_lock(&prison0.pr_mtx);
3242 if (prison0.pr_root == olddp) {
3244 prison0.pr_root = newdp;
3247 mtx_unlock(&prison0.pr_mtx);
3248 sx_slock(&allprison_lock);
3249 TAILQ_FOREACH(pr, &allprison, pr_list) {
3250 mtx_lock(&pr->pr_mtx);
3251 if (pr->pr_root == olddp) {
3253 pr->pr_root = newdp;
3256 mtx_unlock(&pr->pr_mtx);
3258 sx_sunlock(&allprison_lock);
3263 struct filedesc_to_leader *
3264 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3266 struct filedesc_to_leader *fdtol;
3268 fdtol = malloc(sizeof(struct filedesc_to_leader),
3269 M_FILEDESC_TO_LEADER, M_WAITOK);
3270 fdtol->fdl_refcount = 1;
3271 fdtol->fdl_holdcount = 0;
3272 fdtol->fdl_wakeup = 0;
3273 fdtol->fdl_leader = leader;
3275 FILEDESC_XLOCK(fdp);
3276 fdtol->fdl_next = old->fdl_next;
3277 fdtol->fdl_prev = old;
3278 old->fdl_next = fdtol;
3279 fdtol->fdl_next->fdl_prev = fdtol;
3280 FILEDESC_XUNLOCK(fdp);
3282 fdtol->fdl_next = fdtol;
3283 fdtol->fdl_prev = fdtol;
3289 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3291 struct filedesc *fdp;
3292 int i, count, slots;
3294 if (*(int *)arg1 != 0)
3297 fdp = curproc->p_fd;
3299 FILEDESC_SLOCK(fdp);
3300 slots = NDSLOTS(fdp->fd_lastfile + 1);
3301 for (i = 0; i < slots; i++)
3302 count += bitcountl(fdp->fd_map[i]);
3303 FILEDESC_SUNLOCK(fdp);
3305 return (SYSCTL_OUT(req, &count, sizeof(count)));
3308 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3309 CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3310 "Number of open file descriptors");
3313 * Get file structures globally.
3316 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3319 struct filedesc *fdp;
3324 error = sysctl_wire_old_buffer(req, 0);
3327 if (req->oldptr == NULL) {
3329 sx_slock(&allproc_lock);
3330 FOREACH_PROC_IN_SYSTEM(p) {
3332 if (p->p_state == PRS_NEW) {
3340 /* overestimates sparse tables. */
3341 if (fdp->fd_lastfile > 0)
3342 n += fdp->fd_lastfile;
3345 sx_sunlock(&allproc_lock);
3346 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3349 bzero(&xf, sizeof(xf));
3350 xf.xf_size = sizeof(xf);
3351 sx_slock(&allproc_lock);
3352 FOREACH_PROC_IN_SYSTEM(p) {
3354 if (p->p_state == PRS_NEW) {
3358 if (p_cansee(req->td, p) != 0) {
3362 xf.xf_pid = p->p_pid;
3363 xf.xf_uid = p->p_ucred->cr_uid;
3368 FILEDESC_SLOCK(fdp);
3369 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3370 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3373 xf.xf_file = (uintptr_t)fp;
3374 xf.xf_data = (uintptr_t)fp->f_data;
3375 xf.xf_vnode = (uintptr_t)fp->f_vnode;
3376 xf.xf_type = (uintptr_t)fp->f_type;
3377 xf.xf_count = fp->f_count;
3379 xf.xf_offset = foffset_get(fp);
3380 xf.xf_flag = fp->f_flag;
3381 error = SYSCTL_OUT(req, &xf, sizeof(xf));
3385 FILEDESC_SUNLOCK(fdp);
3390 sx_sunlock(&allproc_lock);
3394 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3395 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3397 #ifdef KINFO_FILE_SIZE
3398 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3402 xlate_fflags(int fflags)
3404 static const struct {
3407 } fflags_table[] = {
3408 { FAPPEND, KF_FLAG_APPEND },
3409 { FASYNC, KF_FLAG_ASYNC },
3410 { FFSYNC, KF_FLAG_FSYNC },
3411 { FHASLOCK, KF_FLAG_HASLOCK },
3412 { FNONBLOCK, KF_FLAG_NONBLOCK },
3413 { FREAD, KF_FLAG_READ },
3414 { FWRITE, KF_FLAG_WRITE },
3415 { O_CREAT, KF_FLAG_CREAT },
3416 { O_DIRECT, KF_FLAG_DIRECT },
3417 { O_EXCL, KF_FLAG_EXCL },
3418 { O_EXEC, KF_FLAG_EXEC },
3419 { O_EXLOCK, KF_FLAG_EXLOCK },
3420 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3421 { O_SHLOCK, KF_FLAG_SHLOCK },
3422 { O_TRUNC, KF_FLAG_TRUNC }
3428 for (i = 0; i < nitems(fflags_table); i++)
3429 if (fflags & fflags_table[i].fflag)
3430 kflags |= fflags_table[i].kf_fflag;
3434 /* Trim unused data from kf_path by truncating the structure size. */
3436 pack_kinfo(struct kinfo_file *kif)
3439 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3440 strlen(kif->kf_path) + 1;
3441 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3445 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3446 struct kinfo_file *kif, struct filedesc *fdp, int flags)
3450 bzero(kif, sizeof(*kif));
3452 /* Set a default type to allow for empty fill_kinfo() methods. */
3453 kif->kf_type = KF_TYPE_UNKNOWN;
3454 kif->kf_flags = xlate_fflags(fp->f_flag);
3455 if (rightsp != NULL)
3456 kif->kf_cap_rights = *rightsp;
3458 cap_rights_init(&kif->kf_cap_rights);
3460 kif->kf_ref_count = fp->f_count;
3461 kif->kf_offset = foffset_get(fp);
3464 * This may drop the filedesc lock, so the 'fp' cannot be
3465 * accessed after this call.
3467 error = fo_fill_kinfo(fp, kif, fdp);
3469 kif->kf_status |= KF_ATTR_VALID;
3470 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3473 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3477 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3478 struct kinfo_file *kif, int flags)
3482 bzero(kif, sizeof(*kif));
3484 kif->kf_type = KF_TYPE_VNODE;
3485 error = vn_fill_kinfo_vnode(vp, kif);
3487 kif->kf_status |= KF_ATTR_VALID;
3488 kif->kf_flags = xlate_fflags(fflags);
3489 cap_rights_init(&kif->kf_cap_rights);
3491 kif->kf_ref_count = -1;
3492 kif->kf_offset = -1;
3493 if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3496 kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3500 struct export_fd_buf {
3501 struct filedesc *fdp;
3504 struct kinfo_file kif;
3509 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3511 struct kinfo_file *kif;
3514 if (efbuf->remainder != -1) {
3515 if (efbuf->remainder < kif->kf_structsize) {
3516 /* Terminate export. */
3517 efbuf->remainder = 0;
3520 efbuf->remainder -= kif->kf_structsize;
3522 return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3526 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3527 struct export_fd_buf *efbuf)
3531 if (efbuf->remainder == 0)
3533 export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3535 FILEDESC_SUNLOCK(efbuf->fdp);
3536 error = export_kinfo_to_sb(efbuf);
3537 FILEDESC_SLOCK(efbuf->fdp);
3542 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3543 struct export_fd_buf *efbuf)
3547 if (efbuf->remainder == 0)
3549 if (efbuf->fdp != NULL)
3550 FILEDESC_SUNLOCK(efbuf->fdp);
3551 export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3552 error = export_kinfo_to_sb(efbuf);
3553 if (efbuf->fdp != NULL)
3554 FILEDESC_SLOCK(efbuf->fdp);
3559 * Store a process file descriptor information to sbuf.
3561 * Takes a locked proc as argument, and returns with the proc unlocked.
3564 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen,
3568 struct filedesc *fdp;
3569 struct export_fd_buf *efbuf;
3570 struct vnode *cttyvp, *textvp, *tracevp;
3572 cap_rights_t rights;
3574 PROC_LOCK_ASSERT(p, MA_OWNED);
3577 tracevp = p->p_tracevp;
3578 if (tracevp != NULL)
3581 textvp = p->p_textvp;
3584 /* Controlling tty. */
3586 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3587 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3593 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3596 efbuf->remainder = maxlen;
3597 efbuf->flags = flags;
3598 if (tracevp != NULL)
3599 export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3602 export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3604 export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3610 FILEDESC_SLOCK(fdp);
3611 /* working directory */
3612 if (fdp->fd_cdir != NULL) {
3613 vrefact(fdp->fd_cdir);
3614 export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3616 /* root directory */
3617 if (fdp->fd_rdir != NULL) {
3618 vrefact(fdp->fd_rdir);
3619 export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3621 /* jail directory */
3622 if (fdp->fd_jdir != NULL) {
3623 vrefact(fdp->fd_jdir);
3624 export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3626 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3627 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3630 rights = *cap_rights(fdp, i);
3631 #else /* !CAPABILITIES */
3632 rights = cap_no_rights;
3635 * Create sysctl entry. It is OK to drop the filedesc
3636 * lock inside of export_file_to_sb() as we will
3637 * re-validate and re-evaluate its properties when the
3640 error = export_file_to_sb(fp, i, &rights, efbuf);
3641 if (error != 0 || efbuf->remainder == 0)
3644 FILEDESC_SUNLOCK(fdp);
3647 free(efbuf, M_TEMP);
3651 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
3654 * Get per-process file descriptors for use by procstat(1), et al.
3657 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3662 int error, error2, *name;
3666 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3667 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3668 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3673 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3674 error = kern_proc_filedesc_out(p, &sb, maxlen,
3675 KERN_FILEDESC_PACK_KINFO);
3676 error2 = sbuf_finish(&sb);
3678 return (error != 0 ? error : error2);
3681 #ifdef COMPAT_FREEBSD7
3682 #ifdef KINFO_OFILE_SIZE
3683 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3687 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3690 okif->kf_structsize = sizeof(*okif);
3691 okif->kf_type = kif->kf_type;
3692 okif->kf_fd = kif->kf_fd;
3693 okif->kf_ref_count = kif->kf_ref_count;
3694 okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3695 KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3696 KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3697 okif->kf_offset = kif->kf_offset;
3698 if (kif->kf_type == KF_TYPE_VNODE)
3699 okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
3701 okif->kf_vnode_type = KF_VTYPE_VNON;
3702 strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3703 if (kif->kf_type == KF_TYPE_SOCKET) {
3704 okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
3705 okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
3706 okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
3707 okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
3708 okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
3710 okif->kf_sa_local.ss_family = AF_UNSPEC;
3711 okif->kf_sa_peer.ss_family = AF_UNSPEC;
3716 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3717 struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3722 FILEDESC_SUNLOCK(fdp);
3723 export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3724 kinfo_to_okinfo(kif, okif);
3725 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3726 FILEDESC_SLOCK(fdp);
3731 * Get per-process file descriptors for use by procstat(1), et al.
3734 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3736 struct kinfo_ofile *okif;
3737 struct kinfo_file *kif;
3738 struct filedesc *fdp;
3739 int error, i, *name;
3744 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3751 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3752 okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3753 FILEDESC_SLOCK(fdp);
3754 if (fdp->fd_cdir != NULL)
3755 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3757 if (fdp->fd_rdir != NULL)
3758 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3760 if (fdp->fd_jdir != NULL)
3761 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3763 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3764 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3766 export_file_to_kinfo(fp, i, NULL, kif, fdp,
3767 KERN_FILEDESC_PACK_KINFO);
3768 FILEDESC_SUNLOCK(fdp);
3769 kinfo_to_okinfo(kif, okif);
3770 error = SYSCTL_OUT(req, okif, sizeof(*okif));
3771 FILEDESC_SLOCK(fdp);
3775 FILEDESC_SUNLOCK(fdp);
3782 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3783 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3784 "Process ofiledesc entries");
3785 #endif /* COMPAT_FREEBSD7 */
3788 vntype_to_kinfo(int vtype)
3793 } vtypes_table[] = {
3794 { VBAD, KF_VTYPE_VBAD },
3795 { VBLK, KF_VTYPE_VBLK },
3796 { VCHR, KF_VTYPE_VCHR },
3797 { VDIR, KF_VTYPE_VDIR },
3798 { VFIFO, KF_VTYPE_VFIFO },
3799 { VLNK, KF_VTYPE_VLNK },
3800 { VNON, KF_VTYPE_VNON },
3801 { VREG, KF_VTYPE_VREG },
3802 { VSOCK, KF_VTYPE_VSOCK }
3807 * Perform vtype translation.
3809 for (i = 0; i < nitems(vtypes_table); i++)
3810 if (vtypes_table[i].vtype == vtype)
3811 return (vtypes_table[i].kf_vtype);
3813 return (KF_VTYPE_UNKNOWN);
3816 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3817 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3818 "Process filedesc entries");
3821 * Store a process current working directory information to sbuf.
3823 * Takes a locked proc as argument, and returns with the proc unlocked.
3826 kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
3828 struct filedesc *fdp;
3829 struct export_fd_buf *efbuf;
3832 PROC_LOCK_ASSERT(p, MA_OWNED);
3839 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3842 efbuf->remainder = maxlen;
3844 FILEDESC_SLOCK(fdp);
3845 if (fdp->fd_cdir == NULL)
3848 vrefact(fdp->fd_cdir);
3849 error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3852 FILEDESC_SUNLOCK(fdp);
3854 free(efbuf, M_TEMP);
3859 * Get per-process current working directory.
3862 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3867 int error, error2, *name;
3871 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3872 sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3873 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3878 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3879 error = kern_proc_cwd_out(p, &sb, maxlen);
3880 error2 = sbuf_finish(&sb);
3882 return (error != 0 ? error : error2);
3885 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3886 sysctl_kern_proc_cwd, "Process current working directory");
3890 * For the purposes of debugging, generate a human-readable string for the
3894 file_type_to_name(short type)
3922 case DTYPE_PROCDESC:
3924 case DTYPE_LINUXEFD:
3926 case DTYPE_LINUXTFD:
3934 * For the purposes of debugging, identify a process (if any, perhaps one of
3935 * many) that references the passed file in its file descriptor array. Return
3938 static struct proc *
3939 file_to_first_proc(struct file *fp)
3941 struct filedesc *fdp;
3945 FOREACH_PROC_IN_SYSTEM(p) {
3946 if (p->p_state == PRS_NEW)
3951 for (n = 0; n <= fdp->fd_lastfile; n++) {
3952 if (fp == fdp->fd_ofiles[n].fde_file)
3960 db_print_file(struct file *fp, int header)
3962 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
3966 db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
3967 XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
3968 "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
3970 p = file_to_first_proc(fp);
3971 db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
3972 fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
3973 fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode,
3974 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3979 DB_SHOW_COMMAND(file, db_show_file)
3984 db_printf("usage: show file <addr>\n");
3987 fp = (struct file *)addr;
3988 db_print_file(fp, 1);
3991 DB_SHOW_COMMAND(files, db_show_files)
3993 struct filedesc *fdp;
4000 FOREACH_PROC_IN_SYSTEM(p) {
4001 if (p->p_state == PRS_NEW)
4003 if ((fdp = p->p_fd) == NULL)
4005 for (n = 0; n <= fdp->fd_lastfile; ++n) {
4006 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4008 db_print_file(fp, header);
4015 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4016 &maxfilesperproc, 0, "Maximum files allowed open per process");
4018 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4019 &maxfiles, 0, "Maximum number of files");
4021 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4022 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
4026 filelistinit(void *dummy)
4029 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4030 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4031 filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4032 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4033 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4035 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4037 /*-------------------------------------------------------------------*/
4040 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4041 int flags, struct thread *td)
4048 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4056 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4064 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4072 badfo_kqfilter(struct file *fp, struct knote *kn)
4079 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4087 badfo_close(struct file *fp, struct thread *td)
4094 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4102 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4110 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4111 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4119 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4125 struct fileops badfileops = {
4126 .fo_read = badfo_readwrite,
4127 .fo_write = badfo_readwrite,
4128 .fo_truncate = badfo_truncate,
4129 .fo_ioctl = badfo_ioctl,
4130 .fo_poll = badfo_poll,
4131 .fo_kqfilter = badfo_kqfilter,
4132 .fo_stat = badfo_stat,
4133 .fo_close = badfo_close,
4134 .fo_chmod = badfo_chmod,
4135 .fo_chown = badfo_chown,
4136 .fo_sendfile = badfo_sendfile,
4137 .fo_fill_kinfo = badfo_fill_kinfo,
4141 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4142 int flags, struct thread *td)
4145 return (EOPNOTSUPP);
4149 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4157 invfo_ioctl(struct file *fp, u_long com, void *data,
4158 struct ucred *active_cred, struct thread *td)
4165 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4169 return (poll_no_poll(events));
4173 invfo_kqfilter(struct file *fp, struct knote *kn)
4180 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4188 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4196 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4197 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4204 /*-------------------------------------------------------------------*/
4207 * File Descriptor pseudo-device driver (/dev/fd/).
4209 * Opening minor device N dup()s the file (if any) connected to file
4210 * descriptor N belonging to the calling process. Note that this driver
4211 * consists of only the ``open()'' routine, because all subsequent
4212 * references to this file will be direct to the other driver.
4214 * XXX: we could give this one a cloning event handler if necessary.
4219 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4223 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4224 * the file descriptor being sought for duplication. The error
4225 * return ensures that the vnode for this device will be released
4226 * by vn_open. Open will detect this special error and take the
4227 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4228 * will simply report the error.
4230 td->td_dupfd = dev2unit(dev);
4234 static struct cdevsw fildesc_cdevsw = {
4235 .d_version = D_VERSION,
4241 fildesc_drvinit(void *unused)
4245 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4246 UID_ROOT, GID_WHEEL, 0666, "fd/0");
4247 make_dev_alias(dev, "stdin");
4248 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4249 UID_ROOT, GID_WHEEL, 0666, "fd/1");
4250 make_dev_alias(dev, "stdout");
4251 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4252 UID_ROOT, GID_WHEEL, 0666, "fd/2");
4253 make_dev_alias(dev, "stderr");
4256 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);