2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
40 #include "opt_capsicum.h"
41 #include "opt_compat.h"
43 #include "opt_ktrace.h"
44 #include "opt_procdesc.h"
46 #include <sys/param.h>
47 #include <sys/systm.h>
49 #include <sys/capsicum.h>
51 #include <sys/domain.h>
52 #include <sys/fcntl.h>
54 #include <sys/filedesc.h>
55 #include <sys/filio.h>
57 #include <sys/kernel.h>
59 #include <sys/limits.h>
61 #include <sys/malloc.h>
63 #include <sys/mount.h>
64 #include <sys/mqueue.h>
65 #include <sys/mutex.h>
66 #include <sys/namei.h>
67 #include <sys/selinfo.h>
71 #include <sys/procdesc.h>
72 #include <sys/protosw.h>
73 #include <sys/racct.h>
74 #include <sys/resourcevar.h>
76 #include <sys/signalvar.h>
77 #include <sys/socketvar.h>
80 #include <sys/syscallsubr.h>
81 #include <sys/sysctl.h>
82 #include <sys/sysproto.h>
84 #include <sys/unistd.h>
86 #include <sys/unpcb.h>
88 #include <sys/vnode.h>
90 #include <sys/ktrace.h>
95 #include <netinet/in.h>
96 #include <netinet/in_pcb.h>
98 #include <security/audit/audit.h>
105 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
106 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
107 "file desc to leader structures");
108 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
109 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
111 MALLOC_DECLARE(M_FADVISE);
113 static uma_zone_t file_zone;
115 void (*ksem_info)(struct ksem *ks, char *path, size_t size, uint32_t *value);
117 static int closefp(struct filedesc *fdp, int fd, struct file *fp,
118 struct thread *td, int holdleaders);
119 static int fd_first_free(struct filedesc *fdp, int low, int size);
120 static int fd_last_used(struct filedesc *fdp, int size);
121 static void fdgrowtable(struct filedesc *fdp, int nfd);
122 static void fdgrowtable_exp(struct filedesc *fdp, int nfd);
123 static void fdunused(struct filedesc *fdp, int fd);
124 static void fdused(struct filedesc *fdp, int fd);
125 static int fill_pipe_info(struct pipe *pi, struct kinfo_file *kif);
126 static int fill_procdesc_info(struct procdesc *pdp,
127 struct kinfo_file *kif);
128 static int fill_pts_info(struct tty *tp, struct kinfo_file *kif);
129 static int fill_sem_info(struct file *fp, struct kinfo_file *kif);
130 static int fill_shm_info(struct file *fp, struct kinfo_file *kif);
131 static int fill_socket_info(struct socket *so, struct kinfo_file *kif);
132 static int fill_vnode_info(struct vnode *vp, struct kinfo_file *kif);
133 static int getmaxfd(struct proc *p);
138 * - An array of open file descriptors (fd_ofiles)
139 * - An array of file flags (fd_ofileflags)
140 * - A bitmap recording which descriptors are in use (fd_map)
142 * A process starts out with NDFILE descriptors. The value of NDFILE has
143 * been selected based the historical limit of 20 open files, and an
144 * assumption that the majority of processes, especially short-lived
145 * processes like shells, will never need more.
147 * If this initial allocation is exhausted, a larger descriptor table and
148 * map are allocated dynamically, and the pointers in the process's struct
149 * filedesc are updated to point to those. This is repeated every time
150 * the process runs out of file descriptors (provided it hasn't hit its
153 * Since threads may hold references to individual descriptor table
154 * entries, the tables are never freed. Instead, they are placed on a
155 * linked list and freed only when the struct filedesc is released.
158 #define NDSLOTSIZE sizeof(NDSLOTTYPE)
159 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
160 #define NDSLOT(x) ((x) / NDENTRIES)
161 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
162 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
165 * SLIST entry used to keep track of ofiles which must be reclaimed when
169 struct filedescent *ft_table;
170 SLIST_ENTRY(freetable) ft_next;
174 * Initial allocation: a filedesc structure + the head of SLIST used to
175 * keep track of old ofiles + enough space for NDFILE descriptors.
178 struct filedesc fd_fd;
179 SLIST_HEAD(, freetable) fd_free;
180 struct filedescent fd_dfiles[NDFILE];
181 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
185 * Descriptor management.
187 volatile int openfiles; /* actual number of open files */
188 struct mtx sigio_lock; /* mtx to protect pointers to sigio */
189 void (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
191 /* A mutex to protect the association between a proc and filedesc. */
192 static struct mtx fdesc_mtx;
195 * If low >= size, just return low. Otherwise find the first zero bit in the
196 * given bitmap, starting at low and not exceeding size - 1. Return size if
200 fd_first_free(struct filedesc *fdp, int low, int size)
202 NDSLOTTYPE *map = fdp->fd_map;
210 if (low % NDENTRIES) {
211 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
212 if ((mask &= ~map[off]) != 0UL)
213 return (off * NDENTRIES + ffsl(mask) - 1);
216 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
217 if (map[off] != ~0UL)
218 return (off * NDENTRIES + ffsl(~map[off]) - 1);
223 * Find the highest non-zero bit in the given bitmap, starting at 0 and
224 * not exceeding size - 1. Return -1 if not found.
227 fd_last_used(struct filedesc *fdp, int size)
229 NDSLOTTYPE *map = fdp->fd_map;
234 if (size % NDENTRIES) {
235 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
236 if ((mask &= map[off]) != 0)
237 return (off * NDENTRIES + flsl(mask) - 1);
240 for (minoff = NDSLOT(0); off >= minoff; --off)
242 return (off * NDENTRIES + flsl(map[off]) - 1);
247 fdisused(struct filedesc *fdp, int fd)
250 FILEDESC_LOCK_ASSERT(fdp);
252 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
253 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
255 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
259 * Mark a file descriptor as used.
262 fdused(struct filedesc *fdp, int fd)
265 FILEDESC_XLOCK_ASSERT(fdp);
267 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
269 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
270 if (fd > fdp->fd_lastfile)
271 fdp->fd_lastfile = fd;
272 if (fd == fdp->fd_freefile)
273 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
277 * Mark a file descriptor as unused.
280 fdunused(struct filedesc *fdp, int fd)
283 FILEDESC_XLOCK_ASSERT(fdp);
285 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
286 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
287 ("fd=%d is still in use", fd));
289 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
290 if (fd < fdp->fd_freefile)
291 fdp->fd_freefile = fd;
292 if (fd == fdp->fd_lastfile)
293 fdp->fd_lastfile = fd_last_used(fdp, fd);
297 * Free a file descriptor.
299 * Avoid some work if fdp is about to be destroyed.
302 _fdfree(struct filedesc *fdp, int fd, int last)
304 struct filedescent *fde;
306 fde = &fdp->fd_ofiles[fd];
309 seq_write_begin(&fde->fde_seq);
311 filecaps_free(&fde->fde_caps);
314 bzero(fde, fde_change_size);
317 seq_write_end(&fde->fde_seq);
322 fdfree(struct filedesc *fdp, int fd)
329 fdfree_last(struct filedesc *fdp, int fd)
336 * System calls on descriptors.
338 #ifndef _SYS_SYSPROTO_H_
339 struct getdtablesize_args {
345 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
347 struct proc *p = td->td_proc;
352 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
353 lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
355 if (lim < td->td_retval[0])
356 td->td_retval[0] = lim;
361 * Duplicate a file descriptor to a particular value.
363 * Note: keep in mind that a potential race condition exists when closing
364 * descriptors from a shared descriptor table (via rfork).
366 #ifndef _SYS_SYSPROTO_H_
374 sys_dup2(struct thread *td, struct dup2_args *uap)
377 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
382 * Duplicate a file descriptor.
384 #ifndef _SYS_SYSPROTO_H_
391 sys_dup(struct thread *td, struct dup_args *uap)
394 return (do_dup(td, 0, (int)uap->fd, 0, td->td_retval));
398 * The file control system call.
400 #ifndef _SYS_SYSPROTO_H_
409 sys_fcntl(struct thread *td, struct fcntl_args *uap)
412 return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
416 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
430 * Convert old flock structure to new.
432 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
433 fl.l_start = ofl.l_start;
434 fl.l_len = ofl.l_len;
435 fl.l_pid = ofl.l_pid;
436 fl.l_type = ofl.l_type;
437 fl.l_whence = ofl.l_whence;
451 arg1 = (intptr_t)&fl;
457 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
458 arg1 = (intptr_t)&fl;
466 error = kern_fcntl(td, fd, newcmd, arg1);
469 if (cmd == F_OGETLK) {
470 ofl.l_start = fl.l_start;
471 ofl.l_len = fl.l_len;
472 ofl.l_pid = fl.l_pid;
473 ofl.l_type = fl.l_type;
474 ofl.l_whence = fl.l_whence;
475 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
476 } else if (cmd == F_GETLK) {
477 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
483 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
485 struct filedesc *fdp;
487 struct file *fp, *fp2;
488 struct filedescent *fde;
504 error = do_dup(td, DUP_FCNTL, fd, tmp, td->td_retval);
507 case F_DUPFD_CLOEXEC:
509 error = do_dup(td, DUP_FCNTL | DUP_CLOEXEC, fd, tmp,
515 error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval);
518 case F_DUP2FD_CLOEXEC:
520 error = do_dup(td, DUP_FIXED | DUP_CLOEXEC, fd, tmp,
526 if ((fp = fget_locked(fdp, fd)) == NULL) {
527 FILEDESC_SUNLOCK(fdp);
531 fde = &fdp->fd_ofiles[fd];
533 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
534 FILEDESC_SUNLOCK(fdp);
539 if ((fp = fget_locked(fdp, fd)) == NULL) {
540 FILEDESC_XUNLOCK(fdp);
544 fde = &fdp->fd_ofiles[fd];
545 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
546 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
547 FILEDESC_XUNLOCK(fdp);
551 error = fget_unlocked(fdp, fd,
552 cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp, NULL);
555 td->td_retval[0] = OFLAGS(fp->f_flag);
560 error = fget_unlocked(fdp, fd,
561 cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp, NULL);
565 tmp = flg = fp->f_flag;
567 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
568 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
569 tmp = fp->f_flag & FNONBLOCK;
570 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
575 tmp = fp->f_flag & FASYNC;
576 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
581 atomic_clear_int(&fp->f_flag, FNONBLOCK);
583 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
588 error = fget_unlocked(fdp, fd,
589 cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp, NULL);
592 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
594 td->td_retval[0] = tmp;
599 error = fget_unlocked(fdp, fd,
600 cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp, NULL);
604 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
609 error = priv_check(td, PRIV_NFS_LOCKD);
617 /* FALLTHROUGH F_SETLK */
621 cap_rights_init(&rights, CAP_FLOCK);
622 error = fget_unlocked(fdp, fd, &rights, 0, &fp, NULL);
625 if (fp->f_type != DTYPE_VNODE) {
631 flp = (struct flock *)arg;
632 if (flp->l_whence == SEEK_CUR) {
633 foffset = foffset_get(fp);
636 foffset > OFF_MAX - flp->l_start)) {
637 FILEDESC_SUNLOCK(fdp);
642 flp->l_start += foffset;
646 switch (flp->l_type) {
648 if ((fp->f_flag & FREAD) == 0) {
652 PROC_LOCK(p->p_leader);
653 p->p_leader->p_flag |= P_ADVLOCK;
654 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 PROC_LOCK(p->p_leader);
664 p->p_leader->p_flag |= P_ADVLOCK;
665 PROC_UNLOCK(p->p_leader);
666 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
670 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
675 * Temporary api for testing remote lock
678 if (flg != F_REMOTE) {
682 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
683 F_UNLCKSYS, flp, flg);
689 if (error != 0 || flp->l_type == F_UNLCK ||
690 flp->l_type == F_UNLCKSYS) {
696 * Check for a race with close.
698 * The vnode is now advisory locked (or unlocked, but this case
699 * is not really important) as the caller requested.
700 * We had to drop the filedesc lock, so we need to recheck if
701 * the descriptor is still valid, because if it was closed
702 * in the meantime we need to remove advisory lock from the
703 * vnode - close on any descriptor leading to an advisory
704 * locked vnode, removes that lock.
705 * We will return 0 on purpose in that case, as the result of
706 * successful advisory lock might have been externally visible
707 * already. This is fine - effectively we pretend to the caller
708 * that the closing thread was a bit slower and that the
709 * advisory lock succeeded before the close.
711 error = fget_unlocked(fdp, fd, &rights, 0, &fp2, NULL);
717 flp->l_whence = SEEK_SET;
720 flp->l_type = F_UNLCK;
721 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
722 F_UNLCK, flp, F_POSIX);
729 error = fget_unlocked(fdp, fd,
730 cap_rights_init(&rights, CAP_FLOCK), 0, &fp, NULL);
733 if (fp->f_type != DTYPE_VNODE) {
738 flp = (struct flock *)arg;
739 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
740 flp->l_type != F_UNLCK) {
745 if (flp->l_whence == SEEK_CUR) {
746 foffset = foffset_get(fp);
747 if ((flp->l_start > 0 &&
748 foffset > OFF_MAX - flp->l_start) ||
750 foffset < OFF_MIN - flp->l_start)) {
751 FILEDESC_SUNLOCK(fdp);
756 flp->l_start += foffset;
759 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
765 arg = arg ? 128 * 1024: 0;
768 error = fget_unlocked(fdp, fd, NULL, 0, &fp, NULL);
771 if (fp->f_type != DTYPE_VNODE) {
778 * Exclusive lock synchronizes against f_seqcount reads and
779 * writes in sequential_heuristic().
781 error = vn_lock(vp, LK_EXCLUSIVE);
787 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
788 fp->f_seqcount = (arg + bsize - 1) / bsize;
789 atomic_set_int(&fp->f_flag, FRDAHEAD);
791 atomic_clear_int(&fp->f_flag, FRDAHEAD);
805 getmaxfd(struct proc *p)
810 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
817 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
820 do_dup(struct thread *td, int flags, int old, int new,
823 struct filedesc *fdp;
824 struct filedescent *oldfde, *newfde;
834 * Verify we have a valid descriptor to dup from and possibly to
835 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
836 * return EINVAL when the new descriptor is out of bounds.
841 return (flags & DUP_FCNTL ? EINVAL : EBADF);
844 return (flags & DUP_FCNTL ? EINVAL : EBADF);
847 if (fget_locked(fdp, old) == NULL) {
848 FILEDESC_XUNLOCK(fdp);
851 oldfde = &fdp->fd_ofiles[old];
852 if (flags & DUP_FIXED && old == new) {
854 if (flags & DUP_CLOEXEC)
855 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
856 FILEDESC_XUNLOCK(fdp);
859 fp = oldfde->fde_file;
863 * If the caller specified a file descriptor, make sure the file
864 * table is large enough to hold it, and grab it. Otherwise, just
865 * allocate a new descriptor the usual way.
867 if (flags & DUP_FIXED) {
868 if (new >= fdp->fd_nfiles) {
870 * The resource limits are here instead of e.g.
871 * fdalloc(), because the file descriptor table may be
872 * shared between processes, so we can't really use
873 * racct_add()/racct_sub(). Instead of counting the
874 * number of actually allocated descriptors, just put
875 * the limit on the size of the file descriptor table.
880 error = racct_set(p, RACCT_NOFILE, new + 1);
883 FILEDESC_XUNLOCK(fdp);
889 fdgrowtable_exp(fdp, new + 1);
890 oldfde = &fdp->fd_ofiles[old];
892 newfde = &fdp->fd_ofiles[new];
893 if (newfde->fde_file == NULL)
896 if ((error = fdalloc(td, new, &new)) != 0) {
897 FILEDESC_XUNLOCK(fdp);
901 newfde = &fdp->fd_ofiles[new];
904 KASSERT(fp == oldfde->fde_file, ("old fd has been modified"));
905 KASSERT(old != new, ("new fd is same as old"));
907 delfp = newfde->fde_file;
910 * Duplicate the source descriptor.
913 seq_write_begin(&newfde->fde_seq);
915 filecaps_free(&newfde->fde_caps);
916 memcpy(newfde, oldfde, fde_change_size);
917 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps);
918 if ((flags & DUP_CLOEXEC) != 0)
919 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
921 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
923 seq_write_end(&newfde->fde_seq);
928 (void) closefp(fdp, new, delfp, td, 1);
929 /* closefp() drops the FILEDESC lock for us. */
931 FILEDESC_XUNLOCK(fdp);
938 * If sigio is on the list associated with a process or process group,
939 * disable signalling from the device, remove sigio from the list and
943 funsetown(struct sigio **sigiop)
953 *(sigio->sio_myref) = NULL;
954 if ((sigio)->sio_pgid < 0) {
955 struct pgrp *pg = (sigio)->sio_pgrp;
957 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
961 struct proc *p = (sigio)->sio_proc;
963 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
968 crfree(sigio->sio_ucred);
969 free(sigio, M_SIGIO);
973 * Free a list of sigio structures.
974 * We only need to lock the SIGIO_LOCK because we have made ourselves
975 * inaccessible to callers of fsetown and therefore do not need to lock
976 * the proc or pgrp struct for the list manipulation.
979 funsetownlst(struct sigiolst *sigiolst)
985 sigio = SLIST_FIRST(sigiolst);
992 * Every entry of the list should belong
993 * to a single proc or pgrp.
995 if (sigio->sio_pgid < 0) {
996 pg = sigio->sio_pgrp;
997 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
998 } else /* if (sigio->sio_pgid > 0) */ {
1000 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1004 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1005 *(sigio->sio_myref) = NULL;
1007 KASSERT(sigio->sio_pgid < 0,
1008 ("Proc sigio in pgrp sigio list"));
1009 KASSERT(sigio->sio_pgrp == pg,
1010 ("Bogus pgrp in sigio list"));
1012 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1015 } else /* if (p != NULL) */ {
1016 KASSERT(sigio->sio_pgid > 0,
1017 ("Pgrp sigio in proc sigio list"));
1018 KASSERT(sigio->sio_proc == p,
1019 ("Bogus proc in sigio list"));
1021 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1026 crfree(sigio->sio_ucred);
1027 free(sigio, M_SIGIO);
1034 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1036 * After permission checking, add a sigio structure to the sigio list for
1037 * the process or process group.
1040 fsetown(pid_t pgid, struct sigio **sigiop)
1044 struct sigio *sigio;
1054 /* Allocate and fill in the new sigio out of locks. */
1055 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1056 sigio->sio_pgid = pgid;
1057 sigio->sio_ucred = crhold(curthread->td_ucred);
1058 sigio->sio_myref = sigiop;
1060 sx_slock(&proctree_lock);
1069 * Policy - Don't allow a process to FSETOWN a process
1070 * in another session.
1072 * Remove this test to allow maximum flexibility or
1073 * restrict FSETOWN to the current process or process
1074 * group for maximum safety.
1077 if (proc->p_session != curthread->td_proc->p_session) {
1083 } else /* if (pgid < 0) */ {
1084 pgrp = pgfind(-pgid);
1092 * Policy - Don't allow a process to FSETOWN a process
1093 * in another session.
1095 * Remove this test to allow maximum flexibility or
1096 * restrict FSETOWN to the current process or process
1097 * group for maximum safety.
1099 if (pgrp->pg_session != curthread->td_proc->p_session) {
1110 * Since funsetownlst() is called without the proctree
1111 * locked, we need to check for P_WEXIT.
1112 * XXX: is ESRCH correct?
1114 if ((proc->p_flag & P_WEXIT) != 0) {
1119 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1120 sigio->sio_proc = proc;
1124 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1125 sigio->sio_pgrp = pgrp;
1128 sx_sunlock(&proctree_lock);
1135 sx_sunlock(&proctree_lock);
1136 crfree(sigio->sio_ucred);
1137 free(sigio, M_SIGIO);
1142 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1146 struct sigio **sigiop;
1151 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1157 * Function drops the filedesc lock on return.
1160 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1165 FILEDESC_XLOCK_ASSERT(fdp);
1168 if (td->td_proc->p_fdtol != NULL) {
1170 * Ask fdfree() to sleep to ensure that all relevant
1171 * process leaders can be traversed in closef().
1173 fdp->fd_holdleaderscount++;
1180 * We now hold the fp reference that used to be owned by the
1181 * descriptor array. We have to unlock the FILEDESC *AFTER*
1182 * knote_fdclose to prevent a race of the fd getting opened, a knote
1183 * added, and deleteing a knote for the new fd.
1185 knote_fdclose(td, fd);
1188 * We need to notify mqueue if the object is of type mqueue.
1190 if (fp->f_type == DTYPE_MQUEUE)
1191 mq_fdclose(td, fd, fp);
1192 FILEDESC_XUNLOCK(fdp);
1194 error = closef(fp, td);
1196 FILEDESC_XLOCK(fdp);
1197 fdp->fd_holdleaderscount--;
1198 if (fdp->fd_holdleaderscount == 0 &&
1199 fdp->fd_holdleaderswakeup != 0) {
1200 fdp->fd_holdleaderswakeup = 0;
1201 wakeup(&fdp->fd_holdleaderscount);
1203 FILEDESC_XUNLOCK(fdp);
1209 * Close a file descriptor.
1211 #ifndef _SYS_SYSPROTO_H_
1218 sys_close(struct thread *td, struct close_args *uap)
1221 return (kern_close(td, uap->fd));
1225 kern_close(struct thread *td, int fd)
1227 struct filedesc *fdp;
1230 fdp = td->td_proc->p_fd;
1232 AUDIT_SYSCLOSE(td, fd);
1234 FILEDESC_XLOCK(fdp);
1235 if ((fp = fget_locked(fdp, fd)) == NULL) {
1236 FILEDESC_XUNLOCK(fdp);
1241 /* closefp() drops the FILEDESC lock for us. */
1242 return (closefp(fdp, fd, fp, td, 1));
1246 * Close open file descriptors.
1248 #ifndef _SYS_SYSPROTO_H_
1249 struct closefrom_args {
1255 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1257 struct filedesc *fdp;
1260 fdp = td->td_proc->p_fd;
1261 AUDIT_ARG_FD(uap->lowfd);
1264 * Treat negative starting file descriptor values identical to
1265 * closefrom(0) which closes all files.
1269 FILEDESC_SLOCK(fdp);
1270 for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1271 if (fdp->fd_ofiles[fd].fde_file != NULL) {
1272 FILEDESC_SUNLOCK(fdp);
1273 (void)kern_close(td, fd);
1274 FILEDESC_SLOCK(fdp);
1277 FILEDESC_SUNLOCK(fdp);
1281 #if defined(COMPAT_43)
1283 * Return status information about a file descriptor.
1285 #ifndef _SYS_SYSPROTO_H_
1286 struct ofstat_args {
1293 ofstat(struct thread *td, struct ofstat_args *uap)
1299 error = kern_fstat(td, uap->fd, &ub);
1302 error = copyout(&oub, uap->sb, sizeof(oub));
1306 #endif /* COMPAT_43 */
1309 * Return status information about a file descriptor.
1311 #ifndef _SYS_SYSPROTO_H_
1319 sys_fstat(struct thread *td, struct fstat_args *uap)
1324 error = kern_fstat(td, uap->fd, &ub);
1326 error = copyout(&ub, uap->sb, sizeof(ub));
1331 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1334 cap_rights_t rights;
1339 error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp);
1343 AUDIT_ARG_FILE(td->td_proc, fp);
1345 error = fo_stat(fp, sbp, td->td_ucred, td);
1348 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1355 * Return status information about a file descriptor.
1357 #ifndef _SYS_SYSPROTO_H_
1358 struct nfstat_args {
1365 sys_nfstat(struct thread *td, struct nfstat_args *uap)
1371 error = kern_fstat(td, uap->fd, &ub);
1373 cvtnstat(&ub, &nub);
1374 error = copyout(&nub, uap->sb, sizeof(nub));
1380 * Return pathconf information about a file descriptor.
1382 #ifndef _SYS_SYSPROTO_H_
1383 struct fpathconf_args {
1390 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1394 cap_rights_t rights;
1397 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp);
1401 /* If asynchronous I/O is available, it works for all descriptors. */
1402 if (uap->name == _PC_ASYNC_IO) {
1403 td->td_retval[0] = async_io_version;
1408 vn_lock(vp, LK_SHARED | LK_RETRY);
1409 error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1411 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1412 if (uap->name != _PC_PIPE_BUF) {
1415 td->td_retval[0] = PIPE_BUF;
1427 * Initialize filecaps structure.
1430 filecaps_init(struct filecaps *fcaps)
1433 bzero(fcaps, sizeof(*fcaps));
1434 fcaps->fc_nioctls = -1;
1438 * Copy filecaps structure allocating memory for ioctls array if needed.
1441 filecaps_copy(const struct filecaps *src, struct filecaps *dst)
1446 if (src->fc_ioctls != NULL) {
1447 KASSERT(src->fc_nioctls > 0,
1448 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1450 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1451 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1452 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1457 * Move filecaps structure to the new place and clear the old place.
1460 filecaps_move(struct filecaps *src, struct filecaps *dst)
1464 bzero(src, sizeof(*src));
1468 * Fill the given filecaps structure with full rights.
1471 filecaps_fill(struct filecaps *fcaps)
1474 CAP_ALL(&fcaps->fc_rights);
1475 fcaps->fc_ioctls = NULL;
1476 fcaps->fc_nioctls = -1;
1477 fcaps->fc_fcntls = CAP_FCNTL_ALL;
1481 * Free memory allocated within filecaps structure.
1484 filecaps_free(struct filecaps *fcaps)
1487 free(fcaps->fc_ioctls, M_FILECAPS);
1488 bzero(fcaps, sizeof(*fcaps));
1492 * Validate the given filecaps structure.
1495 filecaps_validate(const struct filecaps *fcaps, const char *func)
1498 KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1499 ("%s: invalid rights", func));
1500 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1501 ("%s: invalid fcntls", func));
1502 KASSERT(fcaps->fc_fcntls == 0 ||
1503 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1504 ("%s: fcntls without CAP_FCNTL", func));
1505 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1506 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1507 ("%s: invalid ioctls", func));
1508 KASSERT(fcaps->fc_nioctls == 0 ||
1509 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1510 ("%s: ioctls without CAP_IOCTL", func));
1514 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1518 FILEDESC_XLOCK_ASSERT(fdp);
1520 nfd1 = fdp->fd_nfiles * 2;
1523 fdgrowtable(fdp, nfd1);
1527 * Grow the file table to accomodate (at least) nfd descriptors.
1530 fdgrowtable(struct filedesc *fdp, int nfd)
1532 struct filedesc0 *fdp0;
1533 struct freetable *ft;
1534 struct filedescent *ntable;
1535 struct filedescent *otable;
1536 int nnfiles, onfiles;
1537 NDSLOTTYPE *nmap, *omap;
1539 FILEDESC_XLOCK_ASSERT(fdp);
1541 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1543 /* save old values */
1544 onfiles = fdp->fd_nfiles;
1545 otable = fdp->fd_ofiles;
1548 /* compute the size of the new table */
1549 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1550 if (nnfiles <= onfiles)
1551 /* the table is already large enough */
1555 * Allocate a new table. We need enough space for the
1556 * file entries themselves and the struct freetable we will use
1557 * when we decommission the table and place it on the freelist.
1558 * We place the struct freetable in the middle so we don't have
1559 * to worry about padding.
1561 ntable = malloc(nnfiles * sizeof(ntable[0]) + sizeof(struct freetable),
1562 M_FILEDESC, M_ZERO | M_WAITOK);
1563 /* copy the old data over and point at the new tables */
1564 memcpy(ntable, otable, onfiles * sizeof(*otable));
1565 fdp->fd_ofiles = ntable;
1568 * Allocate a new map only if the old is not large enough. It will
1569 * grow at a slower rate than the table as it can map more
1570 * entries than the table can hold.
1572 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1573 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1575 /* copy over the old data and update the pointer */
1576 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1581 * In order to have a valid pattern for fget_unlocked()
1582 * fdp->fd_nfiles must be the last member to be updated, otherwise
1583 * fget_unlocked() consumers may reference a new, higher value for
1584 * fdp->fd_nfiles before to access the fdp->fd_ofiles array,
1585 * resulting in OOB accesses.
1587 atomic_store_rel_int(&fdp->fd_nfiles, nnfiles);
1590 * Do not free the old file table, as some threads may still
1591 * reference entries within it. Instead, place it on a freelist
1592 * which will be processed when the struct filedesc is released.
1594 * Note that if onfiles == NDFILE, we're dealing with the original
1595 * static allocation contained within (struct filedesc0 *)fdp,
1596 * which must not be freed.
1598 if (onfiles > NDFILE) {
1599 ft = (struct freetable *)&otable[onfiles];
1600 fdp0 = (struct filedesc0 *)fdp;
1601 ft->ft_table = otable;
1602 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1605 * The map does not have the same possibility of threads still
1606 * holding references to it. So always free it as long as it
1607 * does not reference the original static allocation.
1609 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1610 free(omap, M_FILEDESC);
1614 * Allocate a file descriptor for the process.
1617 fdalloc(struct thread *td, int minfd, int *result)
1619 struct proc *p = td->td_proc;
1620 struct filedesc *fdp = p->p_fd;
1621 int fd = -1, maxfd, allocfd;
1626 FILEDESC_XLOCK_ASSERT(fdp);
1628 if (fdp->fd_freefile > minfd)
1629 minfd = fdp->fd_freefile;
1631 maxfd = getmaxfd(p);
1634 * Search the bitmap for a free descriptor starting at minfd.
1635 * If none is found, grow the file table.
1637 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1640 if (fd >= fdp->fd_nfiles) {
1641 allocfd = min(fd * 2, maxfd);
1645 error = racct_set(p, RACCT_NOFILE, allocfd);
1652 * fd is already equal to first free descriptor >= minfd, so
1653 * we only need to grow the table and we are done.
1655 fdgrowtable_exp(fdp, allocfd);
1659 * Perform some sanity checks, then mark the file descriptor as
1660 * used and return it to the caller.
1662 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1663 ("invalid descriptor %d", fd));
1664 KASSERT(!fdisused(fdp, fd),
1665 ("fd_first_free() returned non-free descriptor"));
1666 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1667 ("file descriptor isn't free"));
1668 KASSERT(fdp->fd_ofiles[fd].fde_flags == 0, ("file flags are set"));
1675 * Allocate n file descriptors for the process.
1678 fdallocn(struct thread *td, int minfd, int *fds, int n)
1680 struct proc *p = td->td_proc;
1681 struct filedesc *fdp = p->p_fd;
1684 FILEDESC_XLOCK_ASSERT(fdp);
1686 if (!fdavail(td, n))
1689 for (i = 0; i < n; i++)
1690 if (fdalloc(td, 0, &fds[i]) != 0)
1694 for (i--; i >= 0; i--)
1695 fdunused(fdp, fds[i]);
1703 * Check to see whether n user file descriptors are available to the process
1707 fdavail(struct thread *td, int n)
1709 struct proc *p = td->td_proc;
1710 struct filedesc *fdp = td->td_proc->p_fd;
1713 FILEDESC_LOCK_ASSERT(fdp);
1716 * XXX: This is only called from uipc_usrreq.c:unp_externalize();
1717 * call racct_add() from there instead of dealing with containers
1721 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1723 last = min(fdp->fd_nfiles, lim);
1724 for (i = fdp->fd_freefile; i < last; i++) {
1725 if (fdp->fd_ofiles[i].fde_file == NULL && --n <= 0)
1732 * Create a new open file structure and allocate a file decriptor for the
1733 * process that refers to it. We add one reference to the file for the
1734 * descriptor table and one reference for resultfp. This is to prevent us
1735 * being preempted and the entry in the descriptor table closed after we
1736 * release the FILEDESC lock.
1739 falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags)
1744 error = falloc_noinstall(td, &fp);
1746 return (error); /* no reference held on error */
1748 error = finstall(td, fp, &fd, flags, NULL);
1750 fdrop(fp, td); /* one reference (fp only) */
1754 if (resultfp != NULL)
1755 *resultfp = fp; /* copy out result */
1757 fdrop(fp, td); /* release local reference */
1759 if (resultfd != NULL)
1766 * Create a new open file structure without allocating a file descriptor.
1769 falloc_noinstall(struct thread *td, struct file **resultfp)
1772 int maxuserfiles = maxfiles - (maxfiles / 20);
1773 static struct timeval lastfail;
1776 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1778 if ((openfiles >= maxuserfiles &&
1779 priv_check(td, PRIV_MAXFILES) != 0) ||
1780 openfiles >= maxfiles) {
1781 if (ppsratecheck(&lastfail, &curfail, 1)) {
1782 printf("kern.maxfiles limit exceeded by uid %i, "
1783 "please see tuning(7).\n", td->td_ucred->cr_ruid);
1787 atomic_add_int(&openfiles, 1);
1788 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1789 refcount_init(&fp->f_count, 1);
1790 fp->f_cred = crhold(td->td_ucred);
1791 fp->f_ops = &badfileops;
1799 * Install a file in a file descriptor table.
1802 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1803 struct filecaps *fcaps)
1805 struct filedesc *fdp = td->td_proc->p_fd;
1806 struct filedescent *fde;
1809 KASSERT(fd != NULL, ("%s: fd == NULL", __func__));
1810 KASSERT(fp != NULL, ("%s: fp == NULL", __func__));
1812 filecaps_validate(fcaps, __func__);
1814 FILEDESC_XLOCK(fdp);
1815 if ((error = fdalloc(td, 0, fd))) {
1816 FILEDESC_XUNLOCK(fdp);
1820 fde = &fdp->fd_ofiles[*fd];
1822 seq_write_begin(&fde->fde_seq);
1825 if ((flags & O_CLOEXEC) != 0)
1826 fde->fde_flags |= UF_EXCLOSE;
1828 filecaps_move(fcaps, &fde->fde_caps);
1830 filecaps_fill(&fde->fde_caps);
1832 seq_write_end(&fde->fde_seq);
1834 FILEDESC_XUNLOCK(fdp);
1839 * Build a new filedesc structure from another.
1840 * Copy the current, root, and jail root vnode references.
1843 fdinit(struct filedesc *fdp)
1845 struct filedesc0 *newfdp;
1847 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1848 FILEDESC_LOCK_INIT(&newfdp->fd_fd);
1850 FILEDESC_SLOCK(fdp);
1851 newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1852 if (newfdp->fd_fd.fd_cdir)
1853 VREF(newfdp->fd_fd.fd_cdir);
1854 newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1855 if (newfdp->fd_fd.fd_rdir)
1856 VREF(newfdp->fd_fd.fd_rdir);
1857 newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1858 if (newfdp->fd_fd.fd_jdir)
1859 VREF(newfdp->fd_fd.fd_jdir);
1860 FILEDESC_SUNLOCK(fdp);
1863 /* Create the file descriptor table. */
1864 newfdp->fd_fd.fd_refcnt = 1;
1865 newfdp->fd_fd.fd_holdcnt = 1;
1866 newfdp->fd_fd.fd_cmask = CMASK;
1867 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1868 newfdp->fd_fd.fd_nfiles = NDFILE;
1869 newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1870 newfdp->fd_fd.fd_lastfile = -1;
1871 return (&newfdp->fd_fd);
1874 static struct filedesc *
1875 fdhold(struct proc *p)
1877 struct filedesc *fdp;
1879 mtx_lock(&fdesc_mtx);
1883 mtx_unlock(&fdesc_mtx);
1888 fddrop(struct filedesc *fdp)
1890 struct filedesc0 *fdp0;
1891 struct freetable *ft;
1894 mtx_lock(&fdesc_mtx);
1895 i = --fdp->fd_holdcnt;
1896 mtx_unlock(&fdesc_mtx);
1900 FILEDESC_LOCK_DESTROY(fdp);
1901 fdp0 = (struct filedesc0 *)fdp;
1902 while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) {
1903 SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next);
1904 free(ft->ft_table, M_FILEDESC);
1906 free(fdp, M_FILEDESC);
1910 * Share a filedesc structure.
1913 fdshare(struct filedesc *fdp)
1916 FILEDESC_XLOCK(fdp);
1918 FILEDESC_XUNLOCK(fdp);
1923 * Unshare a filedesc structure, if necessary by making a copy
1926 fdunshare(struct thread *td)
1928 struct filedesc *tmp;
1929 struct proc *p = td->td_proc;
1931 if (p->p_fd->fd_refcnt == 1)
1934 tmp = fdcopy(p->p_fd);
1940 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
1941 * this is to ease callers, not catch errors.
1944 fdcopy(struct filedesc *fdp)
1946 struct filedesc *newfdp;
1947 struct filedescent *nfde, *ofde;
1950 /* Certain daemons might not have file descriptors. */
1954 newfdp = fdinit(fdp);
1955 FILEDESC_SLOCK(fdp);
1956 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1957 FILEDESC_SUNLOCK(fdp);
1958 FILEDESC_XLOCK(newfdp);
1959 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1960 FILEDESC_XUNLOCK(newfdp);
1961 FILEDESC_SLOCK(fdp);
1963 /* copy all passable descriptors (i.e. not kqueue) */
1964 newfdp->fd_freefile = -1;
1965 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1966 ofde = &fdp->fd_ofiles[i];
1967 if (fdisused(fdp, i) &&
1968 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) &&
1969 ofde->fde_file->f_ops != &badfileops) {
1970 nfde = &newfdp->fd_ofiles[i];
1972 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps);
1973 fhold(nfde->fde_file);
1974 newfdp->fd_lastfile = i;
1976 if (newfdp->fd_freefile == -1)
1977 newfdp->fd_freefile = i;
1980 newfdp->fd_cmask = fdp->fd_cmask;
1981 FILEDESC_SUNLOCK(fdp);
1982 FILEDESC_XLOCK(newfdp);
1983 for (i = 0; i <= newfdp->fd_lastfile; ++i) {
1984 if (newfdp->fd_ofiles[i].fde_file != NULL)
1987 if (newfdp->fd_freefile == -1)
1988 newfdp->fd_freefile = i;
1989 FILEDESC_XUNLOCK(newfdp);
1994 * Release a filedesc structure.
1997 fdescfree(struct thread *td)
1999 struct filedesc *fdp;
2001 struct filedesc_to_leader *fdtol;
2003 struct vnode *cdir, *jdir, *rdir, *vp;
2006 /* Certain daemons might not have file descriptors. */
2007 fdp = td->td_proc->p_fd;
2013 PROC_LOCK(td->td_proc);
2014 racct_set(td->td_proc, RACCT_NOFILE, 0);
2015 PROC_UNLOCK(td->td_proc);
2019 /* Check for special need to clear POSIX style locks */
2020 fdtol = td->td_proc->p_fdtol;
2021 if (fdtol != NULL) {
2022 FILEDESC_XLOCK(fdp);
2023 KASSERT(fdtol->fdl_refcount > 0,
2024 ("filedesc_to_refcount botch: fdl_refcount=%d",
2025 fdtol->fdl_refcount));
2026 if (fdtol->fdl_refcount == 1 &&
2027 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2028 for (i = 0; i <= fdp->fd_lastfile; i++) {
2029 fp = fdp->fd_ofiles[i].fde_file;
2030 if (fp == NULL || fp->f_type != DTYPE_VNODE)
2033 FILEDESC_XUNLOCK(fdp);
2034 lf.l_whence = SEEK_SET;
2037 lf.l_type = F_UNLCK;
2039 (void) VOP_ADVLOCK(vp,
2040 (caddr_t)td->td_proc->p_leader, F_UNLCK,
2042 FILEDESC_XLOCK(fdp);
2047 if (fdtol->fdl_refcount == 1) {
2048 if (fdp->fd_holdleaderscount > 0 &&
2049 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2051 * close() or do_dup() has cleared a reference
2052 * in a shared file descriptor table.
2054 fdp->fd_holdleaderswakeup = 1;
2055 sx_sleep(&fdp->fd_holdleaderscount,
2056 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2059 if (fdtol->fdl_holdcount > 0) {
2061 * Ensure that fdtol->fdl_leader remains
2062 * valid in closef().
2064 fdtol->fdl_wakeup = 1;
2065 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2070 fdtol->fdl_refcount--;
2071 if (fdtol->fdl_refcount == 0 &&
2072 fdtol->fdl_holdcount == 0) {
2073 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2074 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2077 td->td_proc->p_fdtol = NULL;
2078 FILEDESC_XUNLOCK(fdp);
2080 free(fdtol, M_FILEDESC_TO_LEADER);
2083 mtx_lock(&fdesc_mtx);
2084 td->td_proc->p_fd = NULL;
2085 mtx_unlock(&fdesc_mtx);
2087 FILEDESC_XLOCK(fdp);
2088 i = --fdp->fd_refcnt;
2090 FILEDESC_XUNLOCK(fdp);
2094 cdir = fdp->fd_cdir;
2095 fdp->fd_cdir = NULL;
2096 rdir = fdp->fd_rdir;
2097 fdp->fd_rdir = NULL;
2098 jdir = fdp->fd_jdir;
2099 fdp->fd_jdir = NULL;
2100 FILEDESC_XUNLOCK(fdp);
2102 for (i = 0; i <= fdp->fd_lastfile; i++) {
2103 fp = fdp->fd_ofiles[i].fde_file;
2105 fdfree_last(fdp, i);
2106 (void) closef(fp, td);
2110 if (fdp->fd_nfiles > NDFILE)
2111 free(fdp->fd_ofiles, M_FILEDESC);
2112 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2113 free(fdp->fd_map, M_FILEDESC);
2126 * For setugid programs, we don't want to people to use that setugidness
2127 * to generate error messages which write to a file which otherwise would
2128 * otherwise be off-limits to the process. We check for filesystems where
2129 * the vnode can change out from under us after execve (like [lin]procfs).
2131 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2132 * sufficient. We also don't check for setugidness since we know we are.
2135 is_unsafe(struct file *fp)
2137 if (fp->f_type == DTYPE_VNODE) {
2138 struct vnode *vp = fp->f_vnode;
2140 if ((vp->v_vflag & VV_PROCDEP) != 0)
2147 * Make this setguid thing safe, if at all possible.
2150 setugidsafety(struct thread *td)
2152 struct filedesc *fdp;
2156 fdp = td->td_proc->p_fd;
2157 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2158 FILEDESC_XLOCK(fdp);
2159 for (i = 0; i <= fdp->fd_lastfile; i++) {
2162 fp = fdp->fd_ofiles[i].fde_file;
2163 if (fp != NULL && is_unsafe(fp)) {
2164 knote_fdclose(td, i);
2166 * NULL-out descriptor prior to close to avoid
2167 * a race while close blocks.
2170 FILEDESC_XUNLOCK(fdp);
2171 (void) closef(fp, td);
2172 FILEDESC_XLOCK(fdp);
2175 FILEDESC_XUNLOCK(fdp);
2179 * If a specific file object occupies a specific file descriptor, close the
2180 * file descriptor entry and drop a reference on the file object. This is a
2181 * convenience function to handle a subsequent error in a function that calls
2182 * falloc() that handles the race that another thread might have closed the
2183 * file descriptor out from under the thread creating the file object.
2186 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
2189 FILEDESC_XLOCK(fdp);
2190 if (fdp->fd_ofiles[idx].fde_file == fp) {
2192 FILEDESC_XUNLOCK(fdp);
2195 FILEDESC_XUNLOCK(fdp);
2199 * Close any files on exec?
2202 fdcloseexec(struct thread *td)
2204 struct filedesc *fdp;
2205 struct filedescent *fde;
2209 fdp = td->td_proc->p_fd;
2210 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2211 FILEDESC_XLOCK(fdp);
2212 for (i = 0; i <= fdp->fd_lastfile; i++) {
2213 fde = &fdp->fd_ofiles[i];
2215 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2216 (fde->fde_flags & UF_EXCLOSE))) {
2218 (void) closefp(fdp, i, fp, td, 0);
2219 /* closefp() drops the FILEDESC lock. */
2220 FILEDESC_XLOCK(fdp);
2223 FILEDESC_XUNLOCK(fdp);
2227 * It is unsafe for set[ug]id processes to be started with file
2228 * descriptors 0..2 closed, as these descriptors are given implicit
2229 * significance in the Standard C library. fdcheckstd() will create a
2230 * descriptor referencing /dev/null for each of stdin, stdout, and
2231 * stderr that is not already open.
2234 fdcheckstd(struct thread *td)
2236 struct filedesc *fdp;
2237 register_t retval, save;
2238 int i, error, devnull;
2240 fdp = td->td_proc->p_fd;
2241 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2244 for (i = 0; i < 3; i++) {
2245 if (fdp->fd_ofiles[i].fde_file != NULL)
2248 save = td->td_retval[0];
2249 error = kern_open(td, "/dev/null", UIO_SYSSPACE,
2251 devnull = td->td_retval[0];
2252 td->td_retval[0] = save;
2255 KASSERT(devnull == i, ("oof, we didn't get our fd"));
2257 error = do_dup(td, DUP_FIXED, devnull, i, &retval);
2266 * Internal form of close. Decrement reference count on file structure.
2267 * Note: td may be NULL when closing a file that was being passed in a
2270 * XXXRW: Giant is not required for the caller, but often will be held; this
2271 * makes it moderately likely the Giant will be recursed in the VFS case.
2274 closef(struct file *fp, struct thread *td)
2278 struct filedesc_to_leader *fdtol;
2279 struct filedesc *fdp;
2282 * POSIX record locking dictates that any close releases ALL
2283 * locks owned by this process. This is handled by setting
2284 * a flag in the unlock to free ONLY locks obeying POSIX
2285 * semantics, and not to free BSD-style file locks.
2286 * If the descriptor was in a message, POSIX-style locks
2287 * aren't passed with the descriptor, and the thread pointer
2288 * will be NULL. Callers should be careful only to pass a
2289 * NULL thread pointer when there really is no owning
2290 * context that might have locks, or the locks will be
2293 if (fp->f_type == DTYPE_VNODE && td != NULL) {
2295 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2296 lf.l_whence = SEEK_SET;
2299 lf.l_type = F_UNLCK;
2300 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2301 F_UNLCK, &lf, F_POSIX);
2303 fdtol = td->td_proc->p_fdtol;
2304 if (fdtol != NULL) {
2306 * Handle special case where file descriptor table is
2307 * shared between multiple process leaders.
2309 fdp = td->td_proc->p_fd;
2310 FILEDESC_XLOCK(fdp);
2311 for (fdtol = fdtol->fdl_next;
2312 fdtol != td->td_proc->p_fdtol;
2313 fdtol = fdtol->fdl_next) {
2314 if ((fdtol->fdl_leader->p_flag &
2317 fdtol->fdl_holdcount++;
2318 FILEDESC_XUNLOCK(fdp);
2319 lf.l_whence = SEEK_SET;
2322 lf.l_type = F_UNLCK;
2324 (void) VOP_ADVLOCK(vp,
2325 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2327 FILEDESC_XLOCK(fdp);
2328 fdtol->fdl_holdcount--;
2329 if (fdtol->fdl_holdcount == 0 &&
2330 fdtol->fdl_wakeup != 0) {
2331 fdtol->fdl_wakeup = 0;
2335 FILEDESC_XUNLOCK(fdp);
2338 return (fdrop(fp, td));
2342 * Initialize the file pointer with the specified properties.
2344 * The ops are set with release semantics to be certain that the flags, type,
2345 * and data are visible when ops is. This is to prevent ops methods from being
2346 * called with bad data.
2349 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2354 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2358 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2359 int needfcntl, struct file **fpp, cap_rights_t *haverightsp)
2362 struct filedescent fde;
2368 cap_rights_t haverights;
2373 * Avoid reads reordering and then a first access to the
2374 * fdp->fd_ofiles table which could result in OOB operation.
2376 if (fd < 0 || fd >= atomic_load_acq_int(&fdp->fd_nfiles))
2379 * Fetch the descriptor locklessly. We avoid fdrop() races by
2380 * never raising a refcount above 0. To accomplish this we have
2381 * to use a cmpset loop rather than an atomic_add. The descriptor
2382 * must be re-verified once we acquire a reference to be certain
2383 * that the identity is still correct and we did not lose a race
2384 * due to preemption.
2388 seq = seq_read(fd_seq(fdp, fd));
2389 fde = fdp->fd_ofiles[fd];
2390 if (!seq_consistent(fd_seq(fdp, fd), seq)) {
2396 fp = fdp->fd_ofiles[fd].fde_file;
2401 haverights = *cap_rights_fde(&fde);
2402 if (needrightsp != NULL) {
2403 error = cap_check(&haverights, needrightsp);
2406 if (cap_rights_is_set(needrightsp, CAP_FCNTL)) {
2407 error = cap_fcntl_check_fde(&fde, needfcntl);
2413 count = fp->f_count;
2417 * Use an acquire barrier to prevent caching of fd_ofiles
2418 * so it is refreshed for verification.
2420 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) != 1)
2423 if (seq_consistent_nomb(fd_seq(fdp, fd), seq))
2425 if (fp == fdp->fd_ofiles[fd].fde_file)
2428 fdrop(fp, curthread);
2431 if (haverightsp != NULL) {
2433 *haverightsp = haverights;
2435 CAP_ALL(haverightsp);
2442 * Extract the file pointer associated with the specified descriptor for the
2443 * current user process.
2445 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2448 * File's rights will be checked against the capability rights mask.
2450 * If an error occured the non-zero error is returned and *fpp is set to
2451 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is
2452 * responsible for fdrop().
2455 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2456 cap_rights_t *needrightsp, u_char *maxprotp)
2458 struct filedesc *fdp;
2460 cap_rights_t haverights, needrights;
2464 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
2466 if (needrightsp != NULL)
2467 needrights = *needrightsp;
2469 cap_rights_init(&needrights);
2470 if (maxprotp != NULL)
2471 cap_rights_set(&needrights, CAP_MMAP);
2472 error = fget_unlocked(fdp, fd, &needrights, 0, &fp, &haverights);
2475 if (fp->f_ops == &badfileops) {
2482 * If requested, convert capability rights to access flags.
2484 if (maxprotp != NULL)
2485 *maxprotp = cap_rights_to_vmprot(&haverights);
2486 #else /* !CAPABILITIES */
2487 if (maxprotp != NULL)
2488 *maxprotp = VM_PROT_ALL;
2489 #endif /* CAPABILITIES */
2492 * FREAD and FWRITE failure return EBADF as per POSIX.
2498 if ((fp->f_flag & flags) == 0)
2502 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2503 ((fp->f_flag & FWRITE) != 0))
2509 KASSERT(0, ("wrong flags"));
2522 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2525 return(_fget(td, fd, fpp, 0, rightsp, NULL));
2529 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2533 return (_fget(td, fd, fpp, 0, rightsp, maxprotp));
2537 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2540 return(_fget(td, fd, fpp, FREAD, rightsp, NULL));
2544 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2547 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2551 * Like fget() but loads the underlying vnode, or returns an error if the
2552 * descriptor does not represent a vnode. Note that pipes use vnodes but
2553 * never have VM objects. The returned vnode will be vref()'d.
2555 * XXX: what about the unused flags ?
2558 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2565 error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2568 if (fp->f_vnode == NULL) {
2580 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2583 return (_fgetvp(td, fd, 0, rightsp, vpp));
2587 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2588 struct filecaps *havecaps, struct vnode **vpp)
2590 struct filedesc *fdp;
2596 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
2599 fp = fget_locked(fdp, fd);
2600 if (fp == NULL || fp->f_ops == &badfileops)
2604 if (needrightsp != NULL) {
2605 error = cap_check(cap_rights(fdp, fd), needrightsp);
2611 if (fp->f_vnode == NULL)
2616 filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps);
2622 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2625 return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2629 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2632 return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2637 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2641 return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2646 * Like fget() but loads the underlying socket, or returns an error if the
2647 * descriptor does not represent a socket.
2649 * We bump the ref count on the returned socket. XXX Also obtain the SX lock
2652 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely
2653 * on their file descriptor reference to prevent the socket from being free'd
2657 fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp,
2666 if ((error = _fget(td, fd, &fp, 0, rightsp, NULL)) != 0)
2668 if (fp->f_type != DTYPE_SOCKET) {
2673 *fflagp = fp->f_flag;
2684 * Drop the reference count on the socket and XXX release the SX lock in the
2685 * future. The last reference closes the socket.
2687 * Note: fputsock() is deprecated, see comment for fgetsock().
2690 fputsock(struct socket *so)
2695 CURVNET_SET(so->so_vnet);
2701 * Handle the last reference to a file being closed.
2704 _fdrop(struct file *fp, struct thread *td)
2709 if (fp->f_count != 0)
2710 panic("fdrop: count %d", fp->f_count);
2711 if (fp->f_ops != &badfileops)
2712 error = fo_close(fp, td);
2713 atomic_subtract_int(&openfiles, 1);
2715 free(fp->f_advice, M_FADVISE);
2716 uma_zfree(file_zone, fp);
2722 * Apply an advisory lock on a file descriptor.
2724 * Just attempt to get a record lock of the requested type on the entire file
2725 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2727 #ifndef _SYS_SYSPROTO_H_
2735 sys_flock(struct thread *td, struct flock_args *uap)
2740 cap_rights_t rights;
2743 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp);
2746 if (fp->f_type != DTYPE_VNODE) {
2748 return (EOPNOTSUPP);
2752 lf.l_whence = SEEK_SET;
2755 if (uap->how & LOCK_UN) {
2756 lf.l_type = F_UNLCK;
2757 atomic_clear_int(&fp->f_flag, FHASLOCK);
2758 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2761 if (uap->how & LOCK_EX)
2762 lf.l_type = F_WRLCK;
2763 else if (uap->how & LOCK_SH)
2764 lf.l_type = F_RDLCK;
2769 atomic_set_int(&fp->f_flag, FHASLOCK);
2770 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2771 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2777 * Duplicate the specified descriptor to a free descriptor.
2780 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
2781 int openerror, int *indxp)
2783 struct filedescent *newfde, *oldfde;
2787 KASSERT(openerror == ENODEV || openerror == ENXIO,
2788 ("unexpected error %d in %s", openerror, __func__));
2791 * If the to-be-dup'd fd number is greater than the allowed number
2792 * of file descriptors, or the fd to be dup'd has already been
2793 * closed, then reject.
2795 FILEDESC_XLOCK(fdp);
2796 if ((fp = fget_locked(fdp, dfd)) == NULL) {
2797 FILEDESC_XUNLOCK(fdp);
2801 error = fdalloc(td, 0, &indx);
2803 FILEDESC_XUNLOCK(fdp);
2808 * There are two cases of interest here.
2810 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2812 * For ENXIO steal away the file structure from (dfd) and store it in
2813 * (indx). (dfd) is effectively closed by this operation.
2815 switch (openerror) {
2818 * Check that the mode the file is being opened for is a
2819 * subset of the mode of the existing descriptor.
2821 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
2822 fdunused(fdp, indx);
2823 FILEDESC_XUNLOCK(fdp);
2827 newfde = &fdp->fd_ofiles[indx];
2828 oldfde = &fdp->fd_ofiles[dfd];
2830 seq_write_begin(&newfde->fde_seq);
2832 memcpy(newfde, oldfde, fde_change_size);
2833 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps);
2835 seq_write_end(&newfde->fde_seq);
2840 * Steal away the file pointer from dfd and stuff it into indx.
2842 newfde = &fdp->fd_ofiles[indx];
2843 oldfde = &fdp->fd_ofiles[dfd];
2845 seq_write_begin(&newfde->fde_seq);
2847 memcpy(newfde, oldfde, fde_change_size);
2848 bzero(oldfde, fde_change_size);
2851 seq_write_end(&newfde->fde_seq);
2855 FILEDESC_XUNLOCK(fdp);
2861 * Scan all active processes and prisons to see if any of them have a current
2862 * or root directory of `olddp'. If so, replace them with the new mount point.
2865 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2867 struct filedesc *fdp;
2872 if (vrefcnt(olddp) == 1)
2875 sx_slock(&allproc_lock);
2876 FOREACH_PROC_IN_SYSTEM(p) {
2880 FILEDESC_XLOCK(fdp);
2881 if (fdp->fd_cdir == olddp) {
2883 fdp->fd_cdir = newdp;
2886 if (fdp->fd_rdir == olddp) {
2888 fdp->fd_rdir = newdp;
2891 if (fdp->fd_jdir == olddp) {
2893 fdp->fd_jdir = newdp;
2896 FILEDESC_XUNLOCK(fdp);
2899 sx_sunlock(&allproc_lock);
2900 if (rootvnode == olddp) {
2905 mtx_lock(&prison0.pr_mtx);
2906 if (prison0.pr_root == olddp) {
2908 prison0.pr_root = newdp;
2911 mtx_unlock(&prison0.pr_mtx);
2912 sx_slock(&allprison_lock);
2913 TAILQ_FOREACH(pr, &allprison, pr_list) {
2914 mtx_lock(&pr->pr_mtx);
2915 if (pr->pr_root == olddp) {
2917 pr->pr_root = newdp;
2920 mtx_unlock(&pr->pr_mtx);
2922 sx_sunlock(&allprison_lock);
2927 struct filedesc_to_leader *
2928 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2930 struct filedesc_to_leader *fdtol;
2932 fdtol = malloc(sizeof(struct filedesc_to_leader),
2933 M_FILEDESC_TO_LEADER, M_WAITOK);
2934 fdtol->fdl_refcount = 1;
2935 fdtol->fdl_holdcount = 0;
2936 fdtol->fdl_wakeup = 0;
2937 fdtol->fdl_leader = leader;
2939 FILEDESC_XLOCK(fdp);
2940 fdtol->fdl_next = old->fdl_next;
2941 fdtol->fdl_prev = old;
2942 old->fdl_next = fdtol;
2943 fdtol->fdl_next->fdl_prev = fdtol;
2944 FILEDESC_XUNLOCK(fdp);
2946 fdtol->fdl_next = fdtol;
2947 fdtol->fdl_prev = fdtol;
2953 * Get file structures globally.
2956 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2959 struct filedesc *fdp;
2964 error = sysctl_wire_old_buffer(req, 0);
2967 if (req->oldptr == NULL) {
2969 sx_slock(&allproc_lock);
2970 FOREACH_PROC_IN_SYSTEM(p) {
2971 if (p->p_state == PRS_NEW)
2976 /* overestimates sparse tables. */
2977 if (fdp->fd_lastfile > 0)
2978 n += fdp->fd_lastfile;
2981 sx_sunlock(&allproc_lock);
2982 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2985 bzero(&xf, sizeof(xf));
2986 xf.xf_size = sizeof(xf);
2987 sx_slock(&allproc_lock);
2988 FOREACH_PROC_IN_SYSTEM(p) {
2990 if (p->p_state == PRS_NEW) {
2994 if (p_cansee(req->td, p) != 0) {
2998 xf.xf_pid = p->p_pid;
2999 xf.xf_uid = p->p_ucred->cr_uid;
3004 FILEDESC_SLOCK(fdp);
3005 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3006 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3010 xf.xf_data = fp->f_data;
3011 xf.xf_vnode = fp->f_vnode;
3012 xf.xf_type = fp->f_type;
3013 xf.xf_count = fp->f_count;
3015 xf.xf_offset = foffset_get(fp);
3016 xf.xf_flag = fp->f_flag;
3017 error = SYSCTL_OUT(req, &xf, sizeof(xf));
3021 FILEDESC_SUNLOCK(fdp);
3026 sx_sunlock(&allproc_lock);
3030 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3031 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3033 #ifdef KINFO_OFILE_SIZE
3034 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3037 #ifdef COMPAT_FREEBSD7
3039 export_vnode_for_osysctl(struct vnode *vp, int type,
3040 struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req)
3043 char *fullpath, *freepath;
3045 bzero(kif, sizeof(*kif));
3046 kif->kf_structsize = sizeof(*kif);
3050 kif->kf_type = KF_TYPE_VNODE;
3051 /* This function only handles directories. */
3052 if (vp->v_type != VDIR) {
3056 kif->kf_vnode_type = KF_VTYPE_VDIR;
3059 * This is not a true file descriptor, so we set a bogus refcount
3060 * and offset to indicate these fields should be ignored.
3062 kif->kf_ref_count = -1;
3063 kif->kf_offset = -1;
3067 FILEDESC_SUNLOCK(fdp);
3068 vn_fullpath(curthread, vp, &fullpath, &freepath);
3070 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
3071 if (freepath != NULL)
3072 free(freepath, M_TEMP);
3073 error = SYSCTL_OUT(req, kif, sizeof(*kif));
3074 FILEDESC_SLOCK(fdp);
3079 * Get per-process file descriptors for use by procstat(1), et al.
3082 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3084 char *fullpath, *freepath;
3085 struct kinfo_ofile *kif;
3086 struct filedesc *fdp;
3087 int error, i, *name;
3088 struct shmfd *shmfd;
3097 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3104 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3105 FILEDESC_SLOCK(fdp);
3106 if (fdp->fd_cdir != NULL)
3107 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3109 if (fdp->fd_rdir != NULL)
3110 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3112 if (fdp->fd_jdir != NULL)
3113 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3115 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3116 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3118 bzero(kif, sizeof(*kif));
3119 kif->kf_structsize = sizeof(*kif);
3127 switch (fp->f_type) {
3129 kif->kf_type = KF_TYPE_VNODE;
3134 kif->kf_type = KF_TYPE_SOCKET;
3139 kif->kf_type = KF_TYPE_PIPE;
3143 kif->kf_type = KF_TYPE_FIFO;
3148 kif->kf_type = KF_TYPE_KQUEUE;
3152 kif->kf_type = KF_TYPE_CRYPTO;
3156 kif->kf_type = KF_TYPE_MQUEUE;
3160 kif->kf_type = KF_TYPE_SHM;
3165 kif->kf_type = KF_TYPE_SEM;
3170 kif->kf_type = KF_TYPE_PTS;
3175 case DTYPE_PROCDESC:
3176 kif->kf_type = KF_TYPE_PROCDESC;
3181 kif->kf_type = KF_TYPE_UNKNOWN;
3184 kif->kf_ref_count = fp->f_count;
3185 if (fp->f_flag & FREAD)
3186 kif->kf_flags |= KF_FLAG_READ;
3187 if (fp->f_flag & FWRITE)
3188 kif->kf_flags |= KF_FLAG_WRITE;
3189 if (fp->f_flag & FAPPEND)
3190 kif->kf_flags |= KF_FLAG_APPEND;
3191 if (fp->f_flag & FASYNC)
3192 kif->kf_flags |= KF_FLAG_ASYNC;
3193 if (fp->f_flag & FFSYNC)
3194 kif->kf_flags |= KF_FLAG_FSYNC;
3195 if (fp->f_flag & FNONBLOCK)
3196 kif->kf_flags |= KF_FLAG_NONBLOCK;
3197 if (fp->f_flag & O_DIRECT)
3198 kif->kf_flags |= KF_FLAG_DIRECT;
3199 if (fp->f_flag & FHASLOCK)
3200 kif->kf_flags |= KF_FLAG_HASLOCK;
3201 kif->kf_offset = foffset_get(fp);
3204 switch (vp->v_type) {
3206 kif->kf_vnode_type = KF_VTYPE_VNON;
3209 kif->kf_vnode_type = KF_VTYPE_VREG;
3212 kif->kf_vnode_type = KF_VTYPE_VDIR;
3215 kif->kf_vnode_type = KF_VTYPE_VBLK;
3218 kif->kf_vnode_type = KF_VTYPE_VCHR;
3221 kif->kf_vnode_type = KF_VTYPE_VLNK;
3224 kif->kf_vnode_type = KF_VTYPE_VSOCK;
3227 kif->kf_vnode_type = KF_VTYPE_VFIFO;
3230 kif->kf_vnode_type = KF_VTYPE_VBAD;
3233 kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
3237 * It is OK to drop the filedesc lock here as we will
3238 * re-validate and re-evaluate its properties when
3239 * the loop continues.
3243 FILEDESC_SUNLOCK(fdp);
3244 vn_fullpath(curthread, vp, &fullpath, &freepath);
3246 strlcpy(kif->kf_path, fullpath,
3247 sizeof(kif->kf_path));
3248 if (freepath != NULL)
3249 free(freepath, M_TEMP);
3250 FILEDESC_SLOCK(fdp);
3253 struct sockaddr *sa;
3255 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
3256 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
3257 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
3260 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
3261 == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
3262 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
3265 kif->kf_sock_domain =
3266 so->so_proto->pr_domain->dom_family;
3267 kif->kf_sock_type = so->so_type;
3268 kif->kf_sock_protocol = so->so_proto->pr_protocol;
3271 strlcpy(kif->kf_path, tty_devname(tp),
3272 sizeof(kif->kf_path));
3275 shm_path(shmfd, kif->kf_path, sizeof(kif->kf_path));
3276 if (ks != NULL && ksem_info != NULL)
3277 ksem_info(ks, kif->kf_path, sizeof(kif->kf_path), NULL);
3278 error = SYSCTL_OUT(req, kif, sizeof(*kif));
3282 FILEDESC_SUNLOCK(fdp);
3288 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3289 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3290 "Process ofiledesc entries");
3291 #endif /* COMPAT_FREEBSD7 */
3293 #ifdef KINFO_FILE_SIZE
3294 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3297 struct export_fd_buf {
3298 struct filedesc *fdp;
3301 struct kinfo_file kif;
3305 export_fd_to_sb(void *data, int type, int fd, int fflags, int refcnt,
3306 int64_t offset, cap_rights_t *rightsp, struct export_fd_buf *efbuf)
3311 } fflags_table[] = {
3312 { FAPPEND, KF_FLAG_APPEND },
3313 { FASYNC, KF_FLAG_ASYNC },
3314 { FFSYNC, KF_FLAG_FSYNC },
3315 { FHASLOCK, KF_FLAG_HASLOCK },
3316 { FNONBLOCK, KF_FLAG_NONBLOCK },
3317 { FREAD, KF_FLAG_READ },
3318 { FWRITE, KF_FLAG_WRITE },
3319 { O_CREAT, KF_FLAG_CREAT },
3320 { O_DIRECT, KF_FLAG_DIRECT },
3321 { O_EXCL, KF_FLAG_EXCL },
3322 { O_EXEC, KF_FLAG_EXEC },
3323 { O_EXLOCK, KF_FLAG_EXLOCK },
3324 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3325 { O_SHLOCK, KF_FLAG_SHLOCK },
3326 { O_TRUNC, KF_FLAG_TRUNC }
3328 #define NFFLAGS (sizeof(fflags_table) / sizeof(*fflags_table))
3329 struct kinfo_file *kif;
3334 if (efbuf->remainder == 0)
3337 bzero(kif, sizeof(*kif));
3338 locked = efbuf->fdp != NULL;
3343 FILEDESC_SUNLOCK(efbuf->fdp);
3346 vp = (struct vnode *)data;
3347 error = fill_vnode_info(vp, kif);
3350 case KF_TYPE_SOCKET:
3351 error = fill_socket_info((struct socket *)data, kif);
3354 error = fill_pipe_info((struct pipe *)data, kif);
3357 error = fill_pts_info((struct tty *)data, kif);
3359 case KF_TYPE_PROCDESC:
3360 error = fill_procdesc_info((struct procdesc *)data, kif);
3363 error = fill_sem_info((struct file *)data, kif);
3366 error = fill_shm_info((struct file *)data, kif);
3372 kif->kf_status |= KF_ATTR_VALID;
3375 * Translate file access flags.
3377 for (i = 0; i < NFFLAGS; i++)
3378 if (fflags & fflags_table[i].fflag)
3379 kif->kf_flags |= fflags_table[i].kf_fflag;
3380 if (rightsp != NULL)
3381 kif->kf_cap_rights = *rightsp;
3383 cap_rights_init(&kif->kf_cap_rights);
3385 kif->kf_type = type;
3386 kif->kf_ref_count = refcnt;
3387 kif->kf_offset = offset;
3388 /* Pack record size down */
3389 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3390 strlen(kif->kf_path) + 1;
3391 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3392 if (efbuf->remainder != -1) {
3393 if (efbuf->remainder < kif->kf_structsize) {
3394 /* Terminate export. */
3395 efbuf->remainder = 0;
3396 if (efbuf->fdp != NULL && !locked)
3397 FILEDESC_SLOCK(efbuf->fdp);
3400 efbuf->remainder -= kif->kf_structsize;
3403 FILEDESC_SUNLOCK(efbuf->fdp);
3404 error = sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM;
3405 if (efbuf->fdp != NULL)
3406 FILEDESC_SLOCK(efbuf->fdp);
3411 * Store a process file descriptor information to sbuf.
3413 * Takes a locked proc as argument, and returns with the proc unlocked.
3416 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen)
3419 struct filedesc *fdp;
3420 struct export_fd_buf *efbuf;
3421 struct vnode *cttyvp, *textvp, *tracevp;
3425 int type, refcnt, fflags;
3426 cap_rights_t rights;
3428 PROC_LOCK_ASSERT(p, MA_OWNED);
3431 tracevp = p->p_tracevp;
3432 if (tracevp != NULL)
3435 textvp = p->p_textvp;
3438 /* Controlling tty. */
3440 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3441 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3447 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3450 efbuf->remainder = maxlen;
3451 if (tracevp != NULL)
3452 export_fd_to_sb(tracevp, KF_TYPE_VNODE, KF_FD_TYPE_TRACE,
3453 FREAD | FWRITE, -1, -1, NULL, efbuf);
3455 export_fd_to_sb(textvp, KF_TYPE_VNODE, KF_FD_TYPE_TEXT,
3456 FREAD, -1, -1, NULL, efbuf);
3458 export_fd_to_sb(cttyvp, KF_TYPE_VNODE, KF_FD_TYPE_CTTY,
3459 FREAD | FWRITE, -1, -1, NULL, efbuf);
3464 FILEDESC_SLOCK(fdp);
3465 /* working directory */
3466 if (fdp->fd_cdir != NULL) {
3468 data = fdp->fd_cdir;
3469 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_CWD,
3470 FREAD, -1, -1, NULL, efbuf);
3472 /* root directory */
3473 if (fdp->fd_rdir != NULL) {
3475 data = fdp->fd_rdir;
3476 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_ROOT,
3477 FREAD, -1, -1, NULL, efbuf);
3479 /* jail directory */
3480 if (fdp->fd_jdir != NULL) {
3482 data = fdp->fd_jdir;
3483 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_JAIL,
3484 FREAD, -1, -1, NULL, efbuf);
3486 for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3487 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3491 rights = *cap_rights(fdp, i);
3492 #else /* !CAPABILITIES */
3493 cap_rights_init(&rights);
3495 switch (fp->f_type) {
3497 type = KF_TYPE_VNODE;
3503 type = KF_TYPE_SOCKET;
3508 type = KF_TYPE_PIPE;
3513 type = KF_TYPE_FIFO;
3519 type = KF_TYPE_KQUEUE;
3523 type = KF_TYPE_CRYPTO;
3527 type = KF_TYPE_MQUEUE;
3546 case DTYPE_PROCDESC:
3547 type = KF_TYPE_PROCDESC;
3553 type = KF_TYPE_UNKNOWN;
3556 refcnt = fp->f_count;
3557 fflags = fp->f_flag;
3558 offset = foffset_get(fp);
3561 * Create sysctl entry.
3562 * It is OK to drop the filedesc lock here as we will
3563 * re-validate and re-evaluate its properties when
3564 * the loop continues.
3566 error = export_fd_to_sb(data, type, i, fflags, refcnt,
3567 offset, &rights, efbuf);
3571 FILEDESC_SUNLOCK(fdp);
3574 free(efbuf, M_TEMP);
3578 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5)
3581 * Get per-process file descriptors for use by procstat(1), et al.
3584 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3589 int error, error2, *name;
3593 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3594 error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3599 maxlen = req->oldptr != NULL ? req->oldlen : -1;
3600 error = kern_proc_filedesc_out(p, &sb, maxlen);
3601 error2 = sbuf_finish(&sb);
3603 return (error != 0 ? error : error2);
3607 vntype_to_kinfo(int vtype)
3612 } vtypes_table[] = {
3613 { VBAD, KF_VTYPE_VBAD },
3614 { VBLK, KF_VTYPE_VBLK },
3615 { VCHR, KF_VTYPE_VCHR },
3616 { VDIR, KF_VTYPE_VDIR },
3617 { VFIFO, KF_VTYPE_VFIFO },
3618 { VLNK, KF_VTYPE_VLNK },
3619 { VNON, KF_VTYPE_VNON },
3620 { VREG, KF_VTYPE_VREG },
3621 { VSOCK, KF_VTYPE_VSOCK }
3623 #define NVTYPES (sizeof(vtypes_table) / sizeof(*vtypes_table))
3627 * Perform vtype translation.
3629 for (i = 0; i < NVTYPES; i++)
3630 if (vtypes_table[i].vtype == vtype)
3633 return (vtypes_table[i].kf_vtype);
3635 return (KF_VTYPE_UNKNOWN);
3639 fill_vnode_info(struct vnode *vp, struct kinfo_file *kif)
3642 char *fullpath, *freepath;
3647 kif->kf_vnode_type = vntype_to_kinfo(vp->v_type);
3650 error = vn_fullpath(curthread, vp, &fullpath, &freepath);
3652 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
3654 if (freepath != NULL)
3655 free(freepath, M_TEMP);
3658 * Retrieve vnode attributes.
3660 va.va_fsid = VNOVAL;
3662 vn_lock(vp, LK_SHARED | LK_RETRY);
3663 error = VOP_GETATTR(vp, &va, curthread->td_ucred);
3667 if (va.va_fsid != VNOVAL)
3668 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
3670 kif->kf_un.kf_file.kf_file_fsid =
3671 vp->v_mount->mnt_stat.f_fsid.val[0];
3672 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
3673 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
3674 kif->kf_un.kf_file.kf_file_size = va.va_size;
3675 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
3680 fill_socket_info(struct socket *so, struct kinfo_file *kif)
3682 struct sockaddr *sa;
3683 struct inpcb *inpcb;
3684 struct unpcb *unpcb;
3689 kif->kf_sock_domain = so->so_proto->pr_domain->dom_family;
3690 kif->kf_sock_type = so->so_type;
3691 kif->kf_sock_protocol = so->so_proto->pr_protocol;
3692 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
3693 switch(kif->kf_sock_domain) {
3696 if (kif->kf_sock_protocol == IPPROTO_TCP) {
3697 if (so->so_pcb != NULL) {
3698 inpcb = (struct inpcb *)(so->so_pcb);
3699 kif->kf_un.kf_sock.kf_sock_inpcb =
3700 (uintptr_t)inpcb->inp_ppcb;
3705 if (so->so_pcb != NULL) {
3706 unpcb = (struct unpcb *)(so->so_pcb);
3707 if (unpcb->unp_conn) {
3708 kif->kf_un.kf_sock.kf_sock_unpconn =
3709 (uintptr_t)unpcb->unp_conn;
3710 kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
3711 so->so_rcv.sb_state;
3712 kif->kf_un.kf_sock.kf_sock_snd_sb_state =
3713 so->so_snd.sb_state;
3718 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
3719 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
3720 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
3723 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
3724 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
3725 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
3728 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
3729 sizeof(kif->kf_path));
3734 fill_pts_info(struct tty *tp, struct kinfo_file *kif)
3739 kif->kf_un.kf_pts.kf_pts_dev = tty_udev(tp);
3740 strlcpy(kif->kf_path, tty_devname(tp), sizeof(kif->kf_path));
3745 fill_pipe_info(struct pipe *pi, struct kinfo_file *kif)
3750 kif->kf_un.kf_pipe.kf_pipe_addr = (uintptr_t)pi;
3751 kif->kf_un.kf_pipe.kf_pipe_peer = (uintptr_t)pi->pipe_peer;
3752 kif->kf_un.kf_pipe.kf_pipe_buffer_cnt = pi->pipe_buffer.cnt;
3757 fill_procdesc_info(struct procdesc *pdp, struct kinfo_file *kif)
3762 kif->kf_un.kf_proc.kf_pid = pdp->pd_pid;
3767 fill_sem_info(struct file *fp, struct kinfo_file *kif)
3773 if (fp->f_data == NULL)
3775 if (fo_stat(fp, &sb, td->td_ucred, td) != 0)
3777 if (ksem_info == NULL)
3779 ksem_info(fp->f_data, kif->kf_path, sizeof(kif->kf_path),
3780 &kif->kf_un.kf_sem.kf_sem_value);
3781 kif->kf_un.kf_sem.kf_sem_mode = sb.st_mode;
3786 fill_shm_info(struct file *fp, struct kinfo_file *kif)
3792 if (fp->f_data == NULL)
3794 if (fo_stat(fp, &sb, td->td_ucred, td) != 0)
3796 shm_path(fp->f_data, kif->kf_path, sizeof(kif->kf_path));
3797 kif->kf_un.kf_file.kf_file_mode = sb.st_mode;
3798 kif->kf_un.kf_file.kf_file_size = sb.st_size;
3802 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3803 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3804 "Process filedesc entries");
3808 * For the purposes of debugging, generate a human-readable string for the
3812 file_type_to_name(short type)
3842 * For the purposes of debugging, identify a process (if any, perhaps one of
3843 * many) that references the passed file in its file descriptor array. Return
3846 static struct proc *
3847 file_to_first_proc(struct file *fp)
3849 struct filedesc *fdp;
3853 FOREACH_PROC_IN_SYSTEM(p) {
3854 if (p->p_state == PRS_NEW)
3859 for (n = 0; n <= fdp->fd_lastfile; n++) {
3860 if (fp == fdp->fd_ofiles[n].fde_file)
3868 db_print_file(struct file *fp, int header)
3873 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
3874 "File", "Type", "Data", "Flag", "GCFl", "Count",
3875 "MCount", "Vnode", "FPID", "FCmd");
3876 p = file_to_first_proc(fp);
3877 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
3878 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
3879 0, fp->f_count, 0, fp->f_vnode,
3880 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3883 DB_SHOW_COMMAND(file, db_show_file)
3888 db_printf("usage: show file <addr>\n");
3891 fp = (struct file *)addr;
3892 db_print_file(fp, 1);
3895 DB_SHOW_COMMAND(files, db_show_files)
3897 struct filedesc *fdp;
3904 FOREACH_PROC_IN_SYSTEM(p) {
3905 if (p->p_state == PRS_NEW)
3907 if ((fdp = p->p_fd) == NULL)
3909 for (n = 0; n <= fdp->fd_lastfile; ++n) {
3910 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3912 db_print_file(fp, header);
3919 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3920 &maxfilesperproc, 0, "Maximum files allowed open per process");
3922 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3923 &maxfiles, 0, "Maximum number of files");
3925 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3926 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
3930 filelistinit(void *dummy)
3933 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3934 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
3935 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3936 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
3938 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3940 /*-------------------------------------------------------------------*/
3943 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
3944 int flags, struct thread *td)
3951 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
3959 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
3967 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
3975 badfo_kqfilter(struct file *fp, struct knote *kn)
3982 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
3990 badfo_close(struct file *fp, struct thread *td)
3997 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4005 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4013 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4014 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4015 int kflags, struct thread *td)
4021 struct fileops badfileops = {
4022 .fo_read = badfo_readwrite,
4023 .fo_write = badfo_readwrite,
4024 .fo_truncate = badfo_truncate,
4025 .fo_ioctl = badfo_ioctl,
4026 .fo_poll = badfo_poll,
4027 .fo_kqfilter = badfo_kqfilter,
4028 .fo_stat = badfo_stat,
4029 .fo_close = badfo_close,
4030 .fo_chmod = badfo_chmod,
4031 .fo_chown = badfo_chown,
4032 .fo_sendfile = badfo_sendfile,
4036 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4044 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4052 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4053 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4054 int kflags, struct thread *td)
4060 /*-------------------------------------------------------------------*/
4063 * File Descriptor pseudo-device driver (/dev/fd/).
4065 * Opening minor device N dup()s the file (if any) connected to file
4066 * descriptor N belonging to the calling process. Note that this driver
4067 * consists of only the ``open()'' routine, because all subsequent
4068 * references to this file will be direct to the other driver.
4070 * XXX: we could give this one a cloning event handler if necessary.
4075 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4079 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4080 * the file descriptor being sought for duplication. The error
4081 * return ensures that the vnode for this device will be released
4082 * by vn_open. Open will detect this special error and take the
4083 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4084 * will simply report the error.
4086 td->td_dupfd = dev2unit(dev);
4090 static struct cdevsw fildesc_cdevsw = {
4091 .d_version = D_VERSION,
4097 fildesc_drvinit(void *unused)
4101 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4102 UID_ROOT, GID_WHEEL, 0666, "fd/0");
4103 make_dev_alias(dev, "stdin");
4104 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4105 UID_ROOT, GID_WHEEL, 0666, "fd/1");
4106 make_dev_alias(dev, "stdout");
4107 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4108 UID_ROOT, GID_WHEEL, 0666, "fd/2");
4109 make_dev_alias(dev, "stderr");
4112 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);