2 * Copyright (c) 2005 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * Copyright (c) 1982, 1986, 1989, 1991, 1993
36 * The Regents of the University of California. All rights reserved.
37 * (c) UNIX System Laboratories, Inc.
38 * All or some portions of this file are derived from material licensed
39 * to the University of California by American Telephone and Telegraph
40 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
41 * the permission of UNIX System Laboratories, Inc.
43 * Redistribution and use in source and binary forms, with or without
44 * modification, are permitted provided that the following conditions
46 * 1. Redistributions of source code must retain the above copyright
47 * notice, this list of conditions and the following disclaimer.
48 * 2. Redistributions in binary form must reproduce the above copyright
49 * notice, this list of conditions and the following disclaimer in the
50 * documentation and/or other materials provided with the distribution.
51 * 3. All advertising materials mentioning features or use of this software
52 * must display the following acknowledgement:
53 * This product includes software developed by the University of
54 * California, Berkeley and its contributors.
55 * 4. Neither the name of the University nor the names of its contributors
56 * may be used to endorse or promote products derived from this software
57 * without specific prior written permission.
59 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
60 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
61 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
62 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
63 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
64 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
65 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
66 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
67 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
71 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94
72 * $FreeBSD: src/sys/kern/kern_descrip.c,v 1.81.2.19 2004/02/28 00:43:31 tegge Exp $
75 #include "opt_compat.h"
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/malloc.h>
79 #include <sys/sysproto.h>
81 #include <sys/device.h>
83 #include <sys/filedesc.h>
84 #include <sys/kernel.h>
85 #include <sys/sysctl.h>
86 #include <sys/vnode.h>
88 #include <sys/nlookup.h>
90 #include <sys/filio.h>
91 #include <sys/fcntl.h>
92 #include <sys/unistd.h>
93 #include <sys/resourcevar.h>
94 #include <sys/event.h>
95 #include <sys/kern_syscall.h>
96 #include <sys/kcore.h>
97 #include <sys/kinfo.h>
101 #include <vm/vm_extern.h>
103 #include <sys/thread2.h>
104 #include <sys/file2.h>
105 #include <sys/spinlock2.h>
107 static void fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd);
108 static void fdreserve_locked (struct filedesc *fdp, int fd0, int incr);
109 static struct file *funsetfd_locked (struct filedesc *fdp, int fd);
110 static void ffree(struct file *fp);
112 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
113 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
114 "file desc to leader structures");
115 MALLOC_DEFINE(M_FILE, "file", "Open file structure");
116 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
118 static struct krate krate_uidinfo = { .freq = 1 };
120 static d_open_t fdopen;
123 #define CDEV_MAJOR 22
124 static struct dev_ops fildesc_ops = {
130 * Descriptor management.
132 static struct filelist filehead = LIST_HEAD_INITIALIZER(&filehead);
133 static struct spinlock filehead_spin = SPINLOCK_INITIALIZER(&filehead_spin);
134 static int nfiles; /* actual number of open files */
138 * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared.
140 * MPSAFE - must be called with fdp->fd_spin exclusively held
144 fdfixup_locked(struct filedesc *fdp, int fd)
146 if (fd < fdp->fd_freefile) {
147 fdp->fd_freefile = fd;
149 while (fdp->fd_lastfile >= 0 &&
150 fdp->fd_files[fdp->fd_lastfile].fp == NULL &&
151 fdp->fd_files[fdp->fd_lastfile].reserved == 0
158 * System calls on descriptors.
163 sys_getdtablesize(struct getdtablesize_args *uap)
165 struct proc *p = curproc;
166 struct plimit *limit = p->p_limit;
169 spin_lock(&limit->p_spin);
170 if (limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
173 dtsize = (int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur;
174 spin_unlock(&limit->p_spin);
176 if (dtsize > maxfilesperproc)
177 dtsize = maxfilesperproc;
178 if (dtsize < minfilesperproc)
179 dtsize = minfilesperproc;
180 if (p->p_ucred->cr_uid && dtsize > maxfilesperuser)
181 dtsize = maxfilesperuser;
182 uap->sysmsg_result = dtsize;
187 * Duplicate a file descriptor to a particular value.
189 * note: keep in mind that a potential race condition exists when closing
190 * descriptors from a shared descriptor table (via rfork).
195 sys_dup2(struct dup2_args *uap)
200 error = kern_dup(DUP_FIXED, uap->from, uap->to, &fd);
201 uap->sysmsg_fds[0] = fd;
207 * Duplicate a file descriptor.
212 sys_dup(struct dup_args *uap)
217 error = kern_dup(DUP_VARIABLE, uap->fd, 0, &fd);
218 uap->sysmsg_fds[0] = fd;
224 * MPALMOSTSAFE - acquires mplock for fp operations
227 kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred)
229 struct thread *td = curthread;
230 struct proc *p = td->td_proc;
236 int tmp, error, flg = F_POSIX;
241 * Operations on file descriptors that do not require a file pointer.
245 error = fgetfdflags(p->p_fd, fd, &tmp);
247 dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0;
251 if (dat->fc_cloexec & FD_CLOEXEC)
252 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
254 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
258 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd);
265 * Operations on file pointers
267 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
272 dat->fc_flags = OFLAGS(fp->f_flag);
278 nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS;
279 nflags |= oflags & ~FCNTLFLAGS;
282 if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY))
284 if (error == 0 && ((nflags ^ oflags) & FASYNC)) {
285 tmp = nflags & FASYNC;
286 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp,
294 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner,
299 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner,
305 /* Fall into F_SETLK */
308 if (fp->f_type != DTYPE_VNODE) {
312 vp = (struct vnode *)fp->f_data;
315 * copyin/lockop may block
317 if (dat->fc_flock.l_whence == SEEK_CUR)
318 dat->fc_flock.l_start += fp->f_offset;
320 switch (dat->fc_flock.l_type) {
322 if ((fp->f_flag & FREAD) == 0) {
326 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
327 lwkt_gettoken(&p->p_leader->p_token);
328 p->p_leader->p_flag |= P_ADVLOCK;
329 lwkt_reltoken(&p->p_leader->p_token);
331 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
332 &dat->fc_flock, flg);
335 if ((fp->f_flag & FWRITE) == 0) {
339 if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
340 lwkt_gettoken(&p->p_leader->p_token);
341 p->p_leader->p_flag |= P_ADVLOCK;
342 lwkt_reltoken(&p->p_leader->p_token);
344 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
345 &dat->fc_flock, flg);
348 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
349 &dat->fc_flock, F_POSIX);
357 * It is possible to race a close() on the descriptor while
358 * we were blocked getting the lock. If this occurs the
359 * close might not have caught the lock.
361 if (checkfdclosed(p->p_fd, fd, fp)) {
362 dat->fc_flock.l_whence = SEEK_SET;
363 dat->fc_flock.l_start = 0;
364 dat->fc_flock.l_len = 0;
365 dat->fc_flock.l_type = F_UNLCK;
366 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
367 F_UNLCK, &dat->fc_flock, F_POSIX);
372 if (fp->f_type != DTYPE_VNODE) {
376 vp = (struct vnode *)fp->f_data;
378 * copyin/lockop may block
380 if (dat->fc_flock.l_type != F_RDLCK &&
381 dat->fc_flock.l_type != F_WRLCK &&
382 dat->fc_flock.l_type != F_UNLCK) {
386 if (dat->fc_flock.l_whence == SEEK_CUR)
387 dat->fc_flock.l_start += fp->f_offset;
388 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK,
389 &dat->fc_flock, F_POSIX);
401 * The file control system call.
406 sys_fcntl(struct fcntl_args *uap)
413 dat.fc_fd = uap->arg;
416 dat.fc_cloexec = uap->arg;
419 dat.fc_flags = uap->arg;
422 dat.fc_owner = uap->arg;
427 error = copyin((caddr_t)uap->arg, &dat.fc_flock,
428 sizeof(struct flock));
434 error = kern_fcntl(uap->fd, uap->cmd, &dat, curthread->td_ucred);
439 uap->sysmsg_result = dat.fc_fd;
442 uap->sysmsg_result = dat.fc_cloexec;
445 uap->sysmsg_result = dat.fc_flags;
448 uap->sysmsg_result = dat.fc_owner;
450 error = copyout(&dat.fc_flock, (caddr_t)uap->arg,
451 sizeof(struct flock));
460 * Common code for dup, dup2, and fcntl(F_DUPFD).
462 * The type flag can be either DUP_FIXED or DUP_VARIABLE. DUP_FIXED tells
463 * kern_dup() to destructively dup over an existing file descriptor if new
464 * is already open. DUP_VARIABLE tells kern_dup() to find the lowest
465 * unused file descriptor that is greater than or equal to new.
470 kern_dup(enum dup_type type, int old, int new, int *res)
472 struct thread *td = curthread;
473 struct proc *p = td->td_proc;
474 struct filedesc *fdp = p->p_fd;
483 * Verify that we have a valid descriptor to dup from and
484 * possibly to dup to.
486 * NOTE: maxfilesperuser is not applicable to dup()
489 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
492 dtsize = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
493 if (dtsize > maxfilesperproc)
494 dtsize = maxfilesperproc;
495 if (dtsize < minfilesperproc)
496 dtsize = minfilesperproc;
498 if (new < 0 || new > dtsize)
501 spin_lock(&fdp->fd_spin);
502 if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) {
503 spin_unlock(&fdp->fd_spin);
506 if (type == DUP_FIXED && old == new) {
508 spin_unlock(&fdp->fd_spin);
511 fp = fdp->fd_files[old].fp;
512 oldflags = fdp->fd_files[old].fileflags;
513 fhold(fp); /* MPSAFE - can be called with a spinlock held */
516 * Allocate a new descriptor if DUP_VARIABLE, or expand the table
517 * if the requested descriptor is beyond the current table size.
519 * This can block. Retry if the source descriptor no longer matches
520 * or if our expectation in the expansion case races.
522 * If we are not expanding or allocating a new decriptor, then reset
523 * the target descriptor to a reserved state so we have a uniform
524 * setup for the next code block.
526 if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) {
527 spin_unlock(&fdp->fd_spin);
528 error = fdalloc(p, new, &newfd);
529 spin_lock(&fdp->fd_spin);
531 spin_unlock(&fdp->fd_spin);
538 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) {
539 fsetfd_locked(fdp, NULL, newfd);
540 spin_unlock(&fdp->fd_spin);
545 * Check for expansion race
547 if (type != DUP_VARIABLE && new != newfd) {
548 fsetfd_locked(fdp, NULL, newfd);
549 spin_unlock(&fdp->fd_spin);
554 * Check for ripout, newfd reused old (this case probably
558 fsetfd_locked(fdp, NULL, newfd);
559 spin_unlock(&fdp->fd_spin);
566 if (fdp->fd_files[new].reserved) {
567 spin_unlock(&fdp->fd_spin);
569 kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new);
570 tsleep(fdp, 0, "fdres", hz);
575 * If the target descriptor was never allocated we have
576 * to allocate it. If it was we have to clean out the
577 * old descriptor. delfp inherits the ref from the
580 delfp = fdp->fd_files[new].fp;
581 fdp->fd_files[new].fp = NULL;
582 fdp->fd_files[new].reserved = 1;
584 fdreserve_locked(fdp, new, 1);
585 if (new > fdp->fd_lastfile)
586 fdp->fd_lastfile = new;
592 * NOTE: still holding an exclusive spinlock
596 * If a descriptor is being overwritten we may hve to tell
597 * fdfree() to sleep to ensure that all relevant process
598 * leaders can be traversed in closef().
600 if (delfp != NULL && p->p_fdtol != NULL) {
601 fdp->fd_holdleaderscount++;
606 KASSERT(delfp == NULL || type == DUP_FIXED,
607 ("dup() picked an open file"));
610 * Duplicate the source descriptor, update lastfile. If the new
611 * descriptor was not allocated and we aren't replacing an existing
612 * descriptor we have to mark the descriptor as being in use.
614 * The fd_files[] array inherits fp's hold reference.
616 fsetfd_locked(fdp, fp, new);
617 fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE;
618 spin_unlock(&fdp->fd_spin);
623 * If we dup'd over a valid file, we now own the reference to it
624 * and must dispose of it using closef() semantics (as if a
625 * close() were performed on it).
628 if (SLIST_FIRST(&delfp->f_klist))
629 knote_fdclose(delfp, fdp, new);
632 spin_lock(&fdp->fd_spin);
633 fdp->fd_holdleaderscount--;
634 if (fdp->fd_holdleaderscount == 0 &&
635 fdp->fd_holdleaderswakeup != 0) {
636 fdp->fd_holdleaderswakeup = 0;
637 spin_unlock(&fdp->fd_spin);
638 wakeup(&fdp->fd_holdleaderscount);
640 spin_unlock(&fdp->fd_spin);
648 * If sigio is on the list associated with a process or process group,
649 * disable signalling from the device, remove sigio from the list and
655 funsetown(struct sigio **sigiop)
661 if ((sigio = *sigiop) != NULL) {
662 lwkt_gettoken(&proc_token); /* protect sigio */
663 KKASSERT(sigiop == sigio->sio_myref);
666 lwkt_reltoken(&proc_token);
671 if (sigio->sio_pgid < 0) {
672 pgrp = sigio->sio_pgrp;
673 sigio->sio_pgrp = NULL;
674 lwkt_gettoken(&pgrp->pg_token);
675 SLIST_REMOVE(&pgrp->pg_sigiolst, sigio, sigio, sio_pgsigio);
676 lwkt_reltoken(&pgrp->pg_token);
678 } else /* if ((*sigiop)->sio_pgid > 0) */ {
680 sigio->sio_proc = NULL;
682 lwkt_gettoken(&p->p_token);
683 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio);
684 lwkt_reltoken(&p->p_token);
687 crfree(sigio->sio_ucred);
688 sigio->sio_ucred = NULL;
689 kfree(sigio, M_SIGIO);
693 * Free a list of sigio structures. Caller is responsible for ensuring
694 * that the list is MPSAFE.
699 funsetownlst(struct sigiolst *sigiolst)
703 while ((sigio = SLIST_FIRST(sigiolst)) != NULL)
704 funsetown(sigio->sio_myref);
708 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
710 * After permission checking, add a sigio structure to the sigio list for
711 * the process or process group.
716 fsetown(pid_t pgid, struct sigio **sigiop)
718 struct proc *proc = NULL;
719 struct pgrp *pgrp = NULL;
736 * Policy - Don't allow a process to FSETOWN a process
737 * in another session.
739 * Remove this test to allow maximum flexibility or
740 * restrict FSETOWN to the current process or process
741 * group for maximum safety.
743 if (proc->p_session != curproc->p_session) {
747 } else /* if (pgid < 0) */ {
748 pgrp = pgfind(-pgid);
755 * Policy - Don't allow a process to FSETOWN a process
756 * in another session.
758 * Remove this test to allow maximum flexibility or
759 * restrict FSETOWN to the current process or process
760 * group for maximum safety.
762 if (pgrp->pg_session != curproc->p_session) {
767 sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK | M_ZERO);
769 KKASSERT(pgrp == NULL);
770 lwkt_gettoken(&proc->p_token);
771 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
772 sigio->sio_proc = proc;
773 lwkt_reltoken(&proc->p_token);
775 KKASSERT(proc == NULL);
776 lwkt_gettoken(&pgrp->pg_token);
777 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
778 sigio->sio_pgrp = pgrp;
779 lwkt_reltoken(&pgrp->pg_token);
782 sigio->sio_pgid = pgid;
783 sigio->sio_ucred = crhold(curthread->td_ucred);
784 /* It would be convenient if p_ruid was in ucred. */
785 sigio->sio_ruid = sigio->sio_ucred->cr_ruid;
786 sigio->sio_myref = sigiop;
788 lwkt_gettoken(&proc_token);
792 lwkt_reltoken(&proc_token);
803 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
808 fgetown(struct sigio **sigiop)
813 lwkt_gettoken(&proc_token);
815 own = (sigio != NULL ? sigio->sio_pgid : 0);
816 lwkt_reltoken(&proc_token);
822 * Close many file descriptors.
827 sys_closefrom(struct closefrom_args *uap)
829 return(kern_closefrom(uap->fd));
833 * Close all file descriptors greater then or equal to fd
838 kern_closefrom(int fd)
840 struct thread *td = curthread;
841 struct proc *p = td->td_proc;
842 struct filedesc *fdp;
851 * NOTE: This function will skip unassociated descriptors and
852 * reserved descriptors that have not yet been assigned.
853 * fd_lastfile can change as a side effect of kern_close().
855 spin_lock(&fdp->fd_spin);
856 while (fd <= fdp->fd_lastfile) {
857 if (fdp->fd_files[fd].fp != NULL) {
858 spin_unlock(&fdp->fd_spin);
859 /* ok if this races another close */
860 if (kern_close(fd) == EINTR)
862 spin_lock(&fdp->fd_spin);
866 spin_unlock(&fdp->fd_spin);
871 * Close a file descriptor.
876 sys_close(struct close_args *uap)
878 return(kern_close(uap->fd));
887 struct thread *td = curthread;
888 struct proc *p = td->td_proc;
889 struct filedesc *fdp;
897 spin_lock(&fdp->fd_spin);
898 if ((fp = funsetfd_locked(fdp, fd)) == NULL) {
899 spin_unlock(&fdp->fd_spin);
903 if (p->p_fdtol != NULL) {
905 * Ask fdfree() to sleep to ensure that all relevant
906 * process leaders can be traversed in closef().
908 fdp->fd_holdleaderscount++;
913 * we now hold the fp reference that used to be owned by the descriptor
916 spin_unlock(&fdp->fd_spin);
917 if (SLIST_FIRST(&fp->f_klist))
918 knote_fdclose(fp, fdp, fd);
919 error = closef(fp, p);
921 spin_lock(&fdp->fd_spin);
922 fdp->fd_holdleaderscount--;
923 if (fdp->fd_holdleaderscount == 0 &&
924 fdp->fd_holdleaderswakeup != 0) {
925 fdp->fd_holdleaderswakeup = 0;
926 spin_unlock(&fdp->fd_spin);
927 wakeup(&fdp->fd_holdleaderscount);
929 spin_unlock(&fdp->fd_spin);
936 * shutdown_args(int fd, int how)
939 kern_shutdown(int fd, int how)
941 struct thread *td = curthread;
942 struct proc *p = td->td_proc;
948 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
950 error = fo_shutdown(fp, how);
960 sys_shutdown(struct shutdown_args *uap)
964 error = kern_shutdown(uap->s, uap->how);
973 kern_fstat(int fd, struct stat *ub)
975 struct thread *td = curthread;
976 struct proc *p = td->td_proc;
982 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
984 error = fo_stat(fp, ub, td->td_ucred);
991 * Return status information about a file descriptor.
996 sys_fstat(struct fstat_args *uap)
1001 error = kern_fstat(uap->fd, &st);
1004 error = copyout(&st, uap->sb, sizeof(st));
1009 * Return pathconf information about a file descriptor.
1014 sys_fpathconf(struct fpathconf_args *uap)
1016 struct thread *td = curthread;
1017 struct proc *p = td->td_proc;
1022 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
1025 switch (fp->f_type) {
1028 if (uap->name != _PC_PIPE_BUF) {
1031 uap->sysmsg_result = PIPE_BUF;
1037 vp = (struct vnode *)fp->f_data;
1038 error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg);
1048 static int fdexpand;
1049 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0,
1050 "Number of times a file table has been expanded");
1053 * Grow the file table so it can hold through descriptor (want).
1055 * The fdp's spinlock must be held exclusively on entry and may be held
1056 * exclusively on return. The spinlock may be cycled by the routine.
1061 fdgrow_locked(struct filedesc *fdp, int want)
1063 struct fdnode *newfiles;
1064 struct fdnode *oldfiles;
1067 nf = fdp->fd_nfiles;
1069 /* nf has to be of the form 2^n - 1 */
1071 } while (nf <= want);
1073 spin_unlock(&fdp->fd_spin);
1074 newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK);
1075 spin_lock(&fdp->fd_spin);
1078 * We could have raced another extend while we were not holding
1081 if (fdp->fd_nfiles >= nf) {
1082 spin_unlock(&fdp->fd_spin);
1083 kfree(newfiles, M_FILEDESC);
1084 spin_lock(&fdp->fd_spin);
1088 * Copy the existing ofile and ofileflags arrays
1089 * and zero the new portion of each array.
1091 extra = nf - fdp->fd_nfiles;
1092 bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode));
1093 bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode));
1095 oldfiles = fdp->fd_files;
1096 fdp->fd_files = newfiles;
1097 fdp->fd_nfiles = nf;
1099 if (oldfiles != fdp->fd_builtin_files) {
1100 spin_unlock(&fdp->fd_spin);
1101 kfree(oldfiles, M_FILEDESC);
1102 spin_lock(&fdp->fd_spin);
1108 * Number of nodes in right subtree, including the root.
1111 right_subtree_size(int n)
1113 return (n ^ (n | (n + 1)));
1120 right_ancestor(int n)
1122 return (n | (n + 1));
1129 left_ancestor(int n)
1131 return ((n & (n + 1)) - 1);
1135 * Traverse the in-place binary tree buttom-up adjusting the allocation
1136 * count so scans can determine where free descriptors are located.
1138 * MPSAFE - caller must be holding an exclusive spinlock on fdp
1142 fdreserve_locked(struct filedesc *fdp, int fd, int incr)
1145 fdp->fd_files[fd].allocated += incr;
1146 KKASSERT(fdp->fd_files[fd].allocated >= 0);
1147 fd = left_ancestor(fd);
1152 * Reserve a file descriptor for the process. If no error occurs, the
1153 * caller MUST at some point call fsetfd() or assign a file pointer
1154 * or dispose of the reservation.
1159 fdalloc(struct proc *p, int want, int *result)
1161 struct filedesc *fdp = p->p_fd;
1162 struct uidinfo *uip;
1163 int fd, rsize, rsum, node, lim;
1166 * Check dtable size limit
1168 spin_lock(&p->p_limit->p_spin);
1169 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
1172 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
1173 spin_unlock(&p->p_limit->p_spin);
1175 if (lim > maxfilesperproc)
1176 lim = maxfilesperproc;
1177 if (lim < minfilesperproc)
1178 lim = minfilesperproc;
1183 * Check that the user has not run out of descriptors (non-root only).
1184 * As a safety measure the dtable is allowed to have at least
1185 * minfilesperproc open fds regardless of the maxfilesperuser limit.
1187 if (p->p_ucred->cr_uid && fdp->fd_nfiles >= minfilesperproc) {
1188 uip = p->p_ucred->cr_uidinfo;
1189 if (uip->ui_openfiles > maxfilesperuser) {
1190 krateprintf(&krate_uidinfo,
1191 "Warning: user %d pid %d (%s) ran out of "
1192 "file descriptors (%d/%d)\n",
1193 p->p_ucred->cr_uid, (int)p->p_pid,
1195 uip->ui_openfiles, maxfilesperuser);
1201 * Grow the dtable if necessary
1203 spin_lock(&fdp->fd_spin);
1204 if (want >= fdp->fd_nfiles)
1205 fdgrow_locked(fdp, want);
1208 * Search for a free descriptor starting at the higher
1209 * of want or fd_freefile. If that fails, consider
1210 * expanding the ofile array.
1212 * NOTE! the 'allocated' field is a cumulative recursive allocation
1213 * count. If we happen to see a value of 0 then we can shortcut
1214 * our search. Otherwise we run through through the tree going
1215 * down branches we know have free descriptor(s) until we hit a
1216 * leaf node. The leaf node will be free but will not necessarily
1217 * have an allocated field of 0.
1220 /* move up the tree looking for a subtree with a free node */
1221 for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim);
1222 fd = right_ancestor(fd)) {
1223 if (fdp->fd_files[fd].allocated == 0)
1226 rsize = right_subtree_size(fd);
1227 if (fdp->fd_files[fd].allocated == rsize)
1228 continue; /* right subtree full */
1231 * Free fd is in the right subtree of the tree rooted at fd.
1232 * Call that subtree R. Look for the smallest (leftmost)
1233 * subtree of R with an unallocated fd: continue moving
1234 * down the left branch until encountering a full left
1235 * subtree, then move to the right.
1237 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) {
1239 rsum += fdp->fd_files[node].allocated;
1240 if (fdp->fd_files[fd].allocated == rsum + rsize) {
1241 fd = node; /* move to the right */
1242 if (fdp->fd_files[node].allocated == 0)
1251 * No space in current array. Expand?
1253 if (fdp->fd_nfiles >= lim) {
1254 spin_unlock(&fdp->fd_spin);
1257 fdgrow_locked(fdp, want);
1261 KKASSERT(fd < fdp->fd_nfiles);
1262 if (fd > fdp->fd_lastfile)
1263 fdp->fd_lastfile = fd;
1264 if (want <= fdp->fd_freefile)
1265 fdp->fd_freefile = fd;
1267 KKASSERT(fdp->fd_files[fd].fp == NULL);
1268 KKASSERT(fdp->fd_files[fd].reserved == 0);
1269 fdp->fd_files[fd].fileflags = 0;
1270 fdp->fd_files[fd].reserved = 1;
1271 fdreserve_locked(fdp, fd, 1);
1272 spin_unlock(&fdp->fd_spin);
1277 * Check to see whether n user file descriptors
1278 * are available to the process p.
1283 fdavail(struct proc *p, int n)
1285 struct filedesc *fdp = p->p_fd;
1286 struct fdnode *fdnode;
1289 spin_lock(&p->p_limit->p_spin);
1290 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
1293 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
1294 spin_unlock(&p->p_limit->p_spin);
1296 if (lim > maxfilesperproc)
1297 lim = maxfilesperproc;
1298 if (lim < minfilesperproc)
1299 lim = minfilesperproc;
1301 spin_lock(&fdp->fd_spin);
1302 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) {
1303 spin_unlock(&fdp->fd_spin);
1306 last = min(fdp->fd_nfiles, lim);
1307 fdnode = &fdp->fd_files[fdp->fd_freefile];
1308 for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) {
1309 if (fdnode->fp == NULL && --n <= 0) {
1310 spin_unlock(&fdp->fd_spin);
1314 spin_unlock(&fdp->fd_spin);
1319 * Revoke open descriptors referencing (f_data, f_type)
1321 * Any revoke executed within a prison is only able to
1322 * revoke descriptors for processes within that prison.
1324 * Returns 0 on success or an error code.
1326 struct fdrevoke_info {
1336 static int fdrevoke_check_callback(struct file *fp, void *vinfo);
1337 static int fdrevoke_proc_callback(struct proc *p, void *vinfo);
1340 fdrevoke(void *f_data, short f_type, struct ucred *cred)
1342 struct fdrevoke_info info;
1345 bzero(&info, sizeof(info));
1349 error = falloc(NULL, &info.nfp, NULL);
1354 * Scan the file pointer table once. dups do not dup file pointers,
1355 * only descriptors, so there is no leak. Set FREVOKED on the fps
1358 allfiles_scan_exclusive(fdrevoke_check_callback, &info);
1361 * If any fps were marked track down the related descriptors
1362 * and close them. Any dup()s at this point will notice
1363 * the FREVOKED already set in the fp and do the right thing.
1365 * Any fps with non-zero msgcounts (aka sent over a unix-domain
1366 * socket) bumped the intransit counter and will require a
1367 * scan. Races against fps leaving the socket are closed by
1368 * the socket code checking for FREVOKED.
1371 allproc_scan(fdrevoke_proc_callback, &info);
1373 unp_revoke_gc(info.nfp);
1379 * Locate matching file pointers directly.
1381 * WARNING: allfiles_scan_exclusive() holds a spinlock through these calls!
1384 fdrevoke_check_callback(struct file *fp, void *vinfo)
1386 struct fdrevoke_info *info = vinfo;
1389 * File pointers already flagged for revokation are skipped.
1391 if (fp->f_flag & FREVOKED)
1395 * If revoking from a prison file pointers created outside of
1396 * that prison, or file pointers without creds, cannot be revoked.
1398 if (info->cred->cr_prison &&
1399 (fp->f_cred == NULL ||
1400 info->cred->cr_prison != fp->f_cred->cr_prison)) {
1405 * If the file pointer matches then mark it for revocation. The
1406 * flag is currently only used by unp_revoke_gc().
1408 * info->count is a heuristic and can race in a SMP environment.
1410 if (info->data == fp->f_data && info->type == fp->f_type) {
1411 atomic_set_int(&fp->f_flag, FREVOKED);
1412 info->count += fp->f_count;
1420 * Locate matching file pointers via process descriptor tables.
1423 fdrevoke_proc_callback(struct proc *p, void *vinfo)
1425 struct fdrevoke_info *info = vinfo;
1426 struct filedesc *fdp;
1430 if (p->p_stat == SIDL || p->p_stat == SZOMB)
1432 if (info->cred->cr_prison &&
1433 info->cred->cr_prison != p->p_ucred->cr_prison) {
1438 * If the controlling terminal of the process matches the
1439 * vnode being revoked we clear the controlling terminal.
1441 * The normal spec_close() may not catch this because it
1442 * uses curproc instead of p.
1444 if (p->p_session && info->type == DTYPE_VNODE &&
1445 info->data == p->p_session->s_ttyvp) {
1446 p->p_session->s_ttyvp = NULL;
1451 * Softref the fdp to prevent it from being destroyed
1453 spin_lock(&p->p_spin);
1454 if ((fdp = p->p_fd) == NULL) {
1455 spin_unlock(&p->p_spin);
1458 atomic_add_int(&fdp->fd_softrefs, 1);
1459 spin_unlock(&p->p_spin);
1462 * Locate and close any matching file descriptors.
1464 spin_lock(&fdp->fd_spin);
1465 for (n = 0; n < fdp->fd_nfiles; ++n) {
1466 if ((fp = fdp->fd_files[n].fp) == NULL)
1468 if (fp->f_flag & FREVOKED) {
1470 fdp->fd_files[n].fp = info->nfp;
1471 spin_unlock(&fdp->fd_spin);
1472 knote_fdclose(fp, fdp, n); /* XXX */
1474 spin_lock(&fdp->fd_spin);
1478 spin_unlock(&fdp->fd_spin);
1479 atomic_subtract_int(&fdp->fd_softrefs, 1);
1485 * Create a new open file structure and reserve a file decriptor
1486 * for the process that refers to it.
1488 * Root creds are checked using lp, or assumed if lp is NULL. If
1489 * resultfd is non-NULL then lp must also be non-NULL. No file
1490 * descriptor is reserved (and no process context is needed) if
1493 * A file pointer with a refcount of 1 is returned. Note that the
1494 * file pointer is NOT associated with the descriptor. If falloc
1495 * returns success, fsetfd() MUST be called to either associate the
1496 * file pointer or clear the reservation.
1501 falloc(struct lwp *lp, struct file **resultfp, int *resultfd)
1503 static struct timeval lastfail;
1506 struct ucred *cred = lp ? lp->lwp_thread->td_ucred : proc0.p_ucred;
1512 * Handle filetable full issues and root overfill.
1514 if (nfiles >= maxfiles - maxfilesrootres &&
1515 (cred->cr_ruid != 0 || nfiles >= maxfiles)) {
1516 if (ppsratecheck(&lastfail, &curfail, 1)) {
1517 kprintf("kern.maxfiles limit exceeded by uid %d, "
1518 "please see tuning(7).\n",
1526 * Allocate a new file descriptor.
1528 fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO);
1529 spin_init(&fp->f_spin);
1530 SLIST_INIT(&fp->f_klist);
1532 fp->f_ops = &badfileops;
1535 spin_lock(&filehead_spin);
1537 LIST_INSERT_HEAD(&filehead, fp, f_list);
1538 spin_unlock(&filehead_spin);
1540 if ((error = fdalloc(lp->lwp_proc, 0, resultfd)) != 0) {
1553 * Check for races against a file descriptor by determining that the
1554 * file pointer is still associated with the specified file descriptor,
1555 * and a close is not currently in progress.
1560 checkfdclosed(struct filedesc *fdp, int fd, struct file *fp)
1564 spin_lock(&fdp->fd_spin);
1565 if ((unsigned)fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp)
1569 spin_unlock(&fdp->fd_spin);
1574 * Associate a file pointer with a previously reserved file descriptor.
1575 * This function always succeeds.
1577 * If fp is NULL, the file descriptor is returned to the pool.
1581 * MPSAFE (exclusive spinlock must be held on call)
1584 fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd)
1586 KKASSERT((unsigned)fd < fdp->fd_nfiles);
1587 KKASSERT(fdp->fd_files[fd].reserved != 0);
1590 fdp->fd_files[fd].fp = fp;
1591 fdp->fd_files[fd].reserved = 0;
1593 fdp->fd_files[fd].reserved = 0;
1594 fdreserve_locked(fdp, fd, -1);
1595 fdfixup_locked(fdp, fd);
1603 fsetfd(struct filedesc *fdp, struct file *fp, int fd)
1605 spin_lock(&fdp->fd_spin);
1606 fsetfd_locked(fdp, fp, fd);
1607 spin_unlock(&fdp->fd_spin);
1611 * MPSAFE (exclusive spinlock must be held on call)
1615 funsetfd_locked(struct filedesc *fdp, int fd)
1619 if ((unsigned)fd >= fdp->fd_nfiles)
1621 if ((fp = fdp->fd_files[fd].fp) == NULL)
1623 fdp->fd_files[fd].fp = NULL;
1624 fdp->fd_files[fd].fileflags = 0;
1626 fdreserve_locked(fdp, fd, -1);
1627 fdfixup_locked(fdp, fd);
1635 fgetfdflags(struct filedesc *fdp, int fd, int *flagsp)
1639 spin_lock(&fdp->fd_spin);
1640 if (((u_int)fd) >= fdp->fd_nfiles) {
1642 } else if (fdp->fd_files[fd].fp == NULL) {
1645 *flagsp = fdp->fd_files[fd].fileflags;
1648 spin_unlock(&fdp->fd_spin);
1656 fsetfdflags(struct filedesc *fdp, int fd, int add_flags)
1660 spin_lock(&fdp->fd_spin);
1661 if (((u_int)fd) >= fdp->fd_nfiles) {
1663 } else if (fdp->fd_files[fd].fp == NULL) {
1666 fdp->fd_files[fd].fileflags |= add_flags;
1669 spin_unlock(&fdp->fd_spin);
1677 fclrfdflags(struct filedesc *fdp, int fd, int rem_flags)
1681 spin_lock(&fdp->fd_spin);
1682 if (((u_int)fd) >= fdp->fd_nfiles) {
1684 } else if (fdp->fd_files[fd].fp == NULL) {
1687 fdp->fd_files[fd].fileflags &= ~rem_flags;
1690 spin_unlock(&fdp->fd_spin);
1695 * Set/Change/Clear the creds for a fp and synchronize the uidinfo.
1698 fsetcred(struct file *fp, struct ucred *ncr)
1701 struct uidinfo *uip;
1704 if (ocr == NULL || ncr == NULL || ocr->cr_uidinfo != ncr->cr_uidinfo) {
1706 uip = ocr->cr_uidinfo;
1707 atomic_add_int(&uip->ui_openfiles, -1);
1710 uip = ncr->cr_uidinfo;
1711 atomic_add_int(&uip->ui_openfiles, 1);
1722 * Free a file descriptor.
1726 ffree(struct file *fp)
1728 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!"));
1729 spin_lock(&filehead_spin);
1730 LIST_REMOVE(fp, f_list);
1732 spin_unlock(&filehead_spin);
1734 if (fp->f_nchandle.ncp)
1735 cache_drop(&fp->f_nchandle);
1740 * called from init_main, initialize filedesc0 for proc0.
1743 fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask)
1747 fdp0->fd_refcnt = 1;
1748 fdp0->fd_cmask = cmask;
1749 fdp0->fd_files = fdp0->fd_builtin_files;
1750 fdp0->fd_nfiles = NDFILE;
1751 fdp0->fd_lastfile = -1;
1752 spin_init(&fdp0->fd_spin);
1756 * Build a new filedesc structure.
1761 fdinit(struct proc *p)
1763 struct filedesc *newfdp;
1764 struct filedesc *fdp = p->p_fd;
1766 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO);
1767 spin_lock(&fdp->fd_spin);
1769 newfdp->fd_cdir = fdp->fd_cdir;
1770 vref(newfdp->fd_cdir);
1771 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1775 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of
1776 * proc0, but should unconditionally exist in other processes.
1779 newfdp->fd_rdir = fdp->fd_rdir;
1780 vref(newfdp->fd_rdir);
1781 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1784 newfdp->fd_jdir = fdp->fd_jdir;
1785 vref(newfdp->fd_jdir);
1786 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1788 spin_unlock(&fdp->fd_spin);
1790 /* Create the file descriptor table. */
1791 newfdp->fd_refcnt = 1;
1792 newfdp->fd_cmask = cmask;
1793 newfdp->fd_files = newfdp->fd_builtin_files;
1794 newfdp->fd_nfiles = NDFILE;
1795 newfdp->fd_lastfile = -1;
1796 spin_init(&newfdp->fd_spin);
1802 * Share a filedesc structure.
1807 fdshare(struct proc *p)
1809 struct filedesc *fdp;
1812 spin_lock(&fdp->fd_spin);
1814 spin_unlock(&fdp->fd_spin);
1819 * Copy a filedesc structure.
1824 fdcopy(struct proc *p, struct filedesc **fpp)
1826 struct filedesc *fdp = p->p_fd;
1827 struct filedesc *newfdp;
1828 struct fdnode *fdnode;
1833 * Certain daemons might not have file descriptors.
1839 * Allocate the new filedesc and fd_files[] array. This can race
1840 * with operations by other threads on the fdp so we have to be
1843 newfdp = kmalloc(sizeof(struct filedesc),
1844 M_FILEDESC, M_WAITOK | M_ZERO | M_NULLOK);
1845 if (newfdp == NULL) {
1850 spin_lock(&fdp->fd_spin);
1851 if (fdp->fd_lastfile < NDFILE) {
1852 newfdp->fd_files = newfdp->fd_builtin_files;
1856 * We have to allocate (N^2-1) entries for our in-place
1857 * binary tree. Allow the table to shrink.
1861 while (ni > fdp->fd_lastfile && ni > NDFILE) {
1865 spin_unlock(&fdp->fd_spin);
1866 newfdp->fd_files = kmalloc(i * sizeof(struct fdnode),
1867 M_FILEDESC, M_WAITOK | M_ZERO);
1870 * Check for race, retry
1872 spin_lock(&fdp->fd_spin);
1873 if (i <= fdp->fd_lastfile) {
1874 spin_unlock(&fdp->fd_spin);
1875 kfree(newfdp->fd_files, M_FILEDESC);
1881 * Dup the remaining fields. vref() and cache_hold() can be
1882 * safely called while holding the read spinlock on fdp.
1884 * The read spinlock on fdp is still being held.
1886 * NOTE: vref and cache_hold calls for the case where the vnode
1887 * or cache entry already has at least one ref may be called
1888 * while holding spin locks.
1890 if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) {
1891 vref(newfdp->fd_cdir);
1892 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1895 * We must check for fd_rdir here, at least for now because
1896 * the init process is created before we have access to the
1897 * rootvode to take a reference to it.
1899 if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) {
1900 vref(newfdp->fd_rdir);
1901 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1903 if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) {
1904 vref(newfdp->fd_jdir);
1905 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1907 newfdp->fd_refcnt = 1;
1908 newfdp->fd_nfiles = i;
1909 newfdp->fd_lastfile = fdp->fd_lastfile;
1910 newfdp->fd_freefile = fdp->fd_freefile;
1911 newfdp->fd_cmask = fdp->fd_cmask;
1912 spin_init(&newfdp->fd_spin);
1915 * Copy the descriptor table through (i). This also copies the
1916 * allocation state. Then go through and ref the file pointers
1917 * and clean up any KQ descriptors.
1919 * kq descriptors cannot be copied. Since we haven't ref'd the
1920 * copied files yet we can ignore the return value from funsetfd().
1922 * The read spinlock on fdp is still being held.
1924 bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode));
1925 for (i = 0 ; i < newfdp->fd_nfiles; ++i) {
1926 fdnode = &newfdp->fd_files[i];
1927 if (fdnode->reserved) {
1928 fdreserve_locked(newfdp, i, -1);
1929 fdnode->reserved = 0;
1930 fdfixup_locked(newfdp, i);
1931 } else if (fdnode->fp) {
1932 if (fdnode->fp->f_type == DTYPE_KQUEUE) {
1933 (void)funsetfd_locked(newfdp, i);
1939 spin_unlock(&fdp->fd_spin);
1945 * Release a filedesc structure.
1947 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE)
1950 fdfree(struct proc *p, struct filedesc *repl)
1952 struct filedesc *fdp;
1953 struct fdnode *fdnode;
1955 struct filedesc_to_leader *fdtol;
1961 * Certain daemons might not have file descriptors.
1970 * Severe messing around to follow.
1972 spin_lock(&fdp->fd_spin);
1974 /* Check for special need to clear POSIX style locks */
1976 if (fdtol != NULL) {
1977 KASSERT(fdtol->fdl_refcount > 0,
1978 ("filedesc_to_refcount botch: fdl_refcount=%d",
1979 fdtol->fdl_refcount));
1980 if (fdtol->fdl_refcount == 1 &&
1981 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1982 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1983 fdnode = &fdp->fd_files[i];
1984 if (fdnode->fp == NULL ||
1985 fdnode->fp->f_type != DTYPE_VNODE) {
1990 spin_unlock(&fdp->fd_spin);
1992 lf.l_whence = SEEK_SET;
1995 lf.l_type = F_UNLCK;
1996 vp = (struct vnode *)fp->f_data;
1997 (void) VOP_ADVLOCK(vp,
1998 (caddr_t)p->p_leader,
2003 spin_lock(&fdp->fd_spin);
2007 if (fdtol->fdl_refcount == 1) {
2008 if (fdp->fd_holdleaderscount > 0 &&
2009 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2011 * close() or do_dup() has cleared a reference
2012 * in a shared file descriptor table.
2014 fdp->fd_holdleaderswakeup = 1;
2015 ssleep(&fdp->fd_holdleaderscount,
2016 &fdp->fd_spin, 0, "fdlhold", 0);
2019 if (fdtol->fdl_holdcount > 0) {
2021 * Ensure that fdtol->fdl_leader
2022 * remains valid in closef().
2024 fdtol->fdl_wakeup = 1;
2025 ssleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0);
2029 fdtol->fdl_refcount--;
2030 if (fdtol->fdl_refcount == 0 &&
2031 fdtol->fdl_holdcount == 0) {
2032 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2033 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2038 if (fdtol != NULL) {
2039 spin_unlock(&fdp->fd_spin);
2040 kfree(fdtol, M_FILEDESC_TO_LEADER);
2041 spin_lock(&fdp->fd_spin);
2044 if (--fdp->fd_refcnt > 0) {
2045 spin_unlock(&fdp->fd_spin);
2046 spin_lock(&p->p_spin);
2048 spin_unlock(&p->p_spin);
2053 * Even though we are the last reference to the structure allproc
2054 * scans may still reference the structure. Maintain proper
2055 * locks until we can replace p->p_fd.
2057 * Also note that kqueue's closef still needs to reference the
2058 * fdp via p->p_fd, so we have to close the descriptors before
2059 * we replace p->p_fd.
2061 for (i = 0; i <= fdp->fd_lastfile; ++i) {
2062 if (fdp->fd_files[i].fp) {
2063 fp = funsetfd_locked(fdp, i);
2065 spin_unlock(&fdp->fd_spin);
2066 if (SLIST_FIRST(&fp->f_klist))
2067 knote_fdclose(fp, fdp, i);
2069 spin_lock(&fdp->fd_spin);
2073 spin_unlock(&fdp->fd_spin);
2076 * Interlock against an allproc scan operations (typically frevoke).
2078 spin_lock(&p->p_spin);
2080 spin_unlock(&p->p_spin);
2083 * Wait for any softrefs to go away. This race rarely occurs so
2084 * we can use a non-critical-path style poll/sleep loop. The
2085 * race only occurs against allproc scans.
2087 * No new softrefs can occur with the fdp disconnected from the
2090 if (fdp->fd_softrefs) {
2091 kprintf("pid %d: Warning, fdp race avoided\n", p->p_pid);
2092 while (fdp->fd_softrefs)
2093 tsleep(&fdp->fd_softrefs, 0, "fdsoft", 1);
2096 if (fdp->fd_files != fdp->fd_builtin_files)
2097 kfree(fdp->fd_files, M_FILEDESC);
2099 cache_drop(&fdp->fd_ncdir);
2100 vrele(fdp->fd_cdir);
2103 cache_drop(&fdp->fd_nrdir);
2104 vrele(fdp->fd_rdir);
2107 cache_drop(&fdp->fd_njdir);
2108 vrele(fdp->fd_jdir);
2110 kfree(fdp, M_FILEDESC);
2114 * Retrieve and reference the file pointer associated with a descriptor.
2119 holdfp(struct filedesc *fdp, int fd, int flag)
2123 spin_lock(&fdp->fd_spin);
2124 if (((u_int)fd) >= fdp->fd_nfiles) {
2128 if ((fp = fdp->fd_files[fd].fp) == NULL)
2130 if ((fp->f_flag & flag) == 0 && flag != -1) {
2136 spin_unlock(&fdp->fd_spin);
2141 * holdsock() - load the struct file pointer associated
2142 * with a socket into *fpp. If an error occurs, non-zero
2143 * will be returned and *fpp will be set to NULL.
2148 holdsock(struct filedesc *fdp, int fd, struct file **fpp)
2153 spin_lock(&fdp->fd_spin);
2154 if ((unsigned)fd >= fdp->fd_nfiles) {
2159 if ((fp = fdp->fd_files[fd].fp) == NULL) {
2163 if (fp->f_type != DTYPE_SOCKET) {
2170 spin_unlock(&fdp->fd_spin);
2176 * Convert a user file descriptor to a held file pointer.
2181 holdvnode(struct filedesc *fdp, int fd, struct file **fpp)
2186 spin_lock(&fdp->fd_spin);
2187 if ((unsigned)fd >= fdp->fd_nfiles) {
2192 if ((fp = fdp->fd_files[fd].fp) == NULL) {
2196 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
2204 spin_unlock(&fdp->fd_spin);
2210 * For setugid programs, we don't want to people to use that setugidness
2211 * to generate error messages which write to a file which otherwise would
2212 * otherwise be off-limits to the process.
2214 * This is a gross hack to plug the hole. A better solution would involve
2215 * a special vop or other form of generalized access control mechanism. We
2216 * go ahead and just reject all procfs file systems accesses as dangerous.
2218 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2219 * sufficient. We also don't for check setugidness since we know we are.
2222 is_unsafe(struct file *fp)
2224 if (fp->f_type == DTYPE_VNODE &&
2225 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS)
2231 * Make this setguid thing safe, if at all possible.
2233 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2236 setugidsafety(struct proc *p)
2238 struct filedesc *fdp = p->p_fd;
2241 /* Certain daemons might not have file descriptors. */
2246 * note: fdp->fd_files may be reallocated out from under us while
2247 * we are blocked in a close. Be careful!
2249 for (i = 0; i <= fdp->fd_lastfile; i++) {
2252 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) {
2256 * NULL-out descriptor prior to close to avoid
2257 * a race while close blocks.
2259 if ((fp = funsetfd_locked(fdp, i)) != NULL) {
2260 knote_fdclose(fp, fdp, i);
2268 * Close any files on exec?
2270 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2273 fdcloseexec(struct proc *p)
2275 struct filedesc *fdp = p->p_fd;
2278 /* Certain daemons might not have file descriptors. */
2283 * We cannot cache fd_files since operations may block and rip
2284 * them out from under us.
2286 for (i = 0; i <= fdp->fd_lastfile; i++) {
2287 if (fdp->fd_files[i].fp != NULL &&
2288 (fdp->fd_files[i].fileflags & UF_EXCLOSE)) {
2292 * NULL-out descriptor prior to close to avoid
2293 * a race while close blocks.
2295 if ((fp = funsetfd_locked(fdp, i)) != NULL) {
2296 knote_fdclose(fp, fdp, i);
2304 * It is unsafe for set[ug]id processes to be started with file
2305 * descriptors 0..2 closed, as these descriptors are given implicit
2306 * significance in the Standard C library. fdcheckstd() will create a
2307 * descriptor referencing /dev/null for each of stdin, stdout, and
2308 * stderr that is not already open.
2310 * NOT MPSAFE - calls falloc, vn_open, etc
2313 fdcheckstd(struct lwp *lp)
2315 struct nlookupdata nd;
2316 struct filedesc *fdp;
2319 int i, error, flags, devnull;
2321 fdp = lp->lwp_proc->p_fd;
2326 for (i = 0; i < 3; i++) {
2327 if (fdp->fd_files[i].fp != NULL)
2330 if ((error = falloc(lp, &fp, &devnull)) != 0)
2333 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE,
2334 NLC_FOLLOW|NLC_LOCKVP);
2335 flags = FREAD | FWRITE;
2337 error = vn_open(&nd, fp, flags, 0);
2339 fsetfd(fdp, fp, devnull);
2341 fsetfd(fdp, NULL, devnull);
2346 KKASSERT(i == devnull);
2348 error = kern_dup(DUP_FIXED, devnull, i, &retval);
2357 * Internal form of close.
2358 * Decrement reference count on file structure.
2359 * Note: td and/or p may be NULL when closing a file
2360 * that was being passed in a message.
2362 * MPALMOSTSAFE - acquires mplock for VOP operations
2365 closef(struct file *fp, struct proc *p)
2369 struct filedesc_to_leader *fdtol;
2375 * POSIX record locking dictates that any close releases ALL
2376 * locks owned by this process. This is handled by setting
2377 * a flag in the unlock to free ONLY locks obeying POSIX
2378 * semantics, and not to free BSD-style file locks.
2379 * If the descriptor was in a message, POSIX-style locks
2380 * aren't passed with the descriptor.
2382 if (p != NULL && fp->f_type == DTYPE_VNODE &&
2383 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2385 if ((p->p_leader->p_flag & P_ADVLOCK) != 0) {
2386 lf.l_whence = SEEK_SET;
2389 lf.l_type = F_UNLCK;
2390 vp = (struct vnode *)fp->f_data;
2391 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
2395 if (fdtol != NULL) {
2396 lwkt_gettoken(&p->p_token);
2398 * Handle special case where file descriptor table
2399 * is shared between multiple process leaders.
2401 for (fdtol = fdtol->fdl_next;
2402 fdtol != p->p_fdtol;
2403 fdtol = fdtol->fdl_next) {
2404 if ((fdtol->fdl_leader->p_flag &
2407 fdtol->fdl_holdcount++;
2408 lf.l_whence = SEEK_SET;
2411 lf.l_type = F_UNLCK;
2412 vp = (struct vnode *)fp->f_data;
2413 (void) VOP_ADVLOCK(vp,
2414 (caddr_t)fdtol->fdl_leader,
2415 F_UNLCK, &lf, F_POSIX);
2416 fdtol->fdl_holdcount--;
2417 if (fdtol->fdl_holdcount == 0 &&
2418 fdtol->fdl_wakeup != 0) {
2419 fdtol->fdl_wakeup = 0;
2423 lwkt_reltoken(&p->p_token);
2432 * fhold() can only be called if f_count is already at least 1 (i.e. the
2433 * caller of fhold() already has a reference to the file pointer in some
2436 * f_count is not spin-locked. Instead, atomic ops are used for
2437 * incrementing, decrementing, and handling the 1->0 transition.
2440 fhold(struct file *fp)
2442 atomic_add_int(&fp->f_count, 1);
2446 * fdrop() - drop a reference to a descriptor
2448 * MPALMOSTSAFE - acquires mplock for final close sequence
2451 fdrop(struct file *fp)
2458 * A combined fetch and subtract is needed to properly detect
2459 * 1->0 transitions, otherwise two cpus dropping from a ref
2460 * count of 2 might both try to run the 1->0 code.
2462 if (atomic_fetchadd_int(&fp->f_count, -1) > 1)
2465 KKASSERT(SLIST_FIRST(&fp->f_klist) == NULL);
2468 * The last reference has gone away, we own the fp structure free
2471 if (fp->f_count < 0)
2472 panic("fdrop: count < 0");
2473 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE &&
2474 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2476 lf.l_whence = SEEK_SET;
2479 lf.l_type = F_UNLCK;
2480 vp = (struct vnode *)fp->f_data;
2481 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2483 if (fp->f_ops != &badfileops)
2484 error = fo_close(fp);
2492 * Apply an advisory lock on a file descriptor.
2494 * Just attempt to get a record lock of the requested type on
2495 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2500 sys_flock(struct flock_args *uap)
2502 struct proc *p = curproc;
2508 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
2510 if (fp->f_type != DTYPE_VNODE) {
2514 vp = (struct vnode *)fp->f_data;
2515 lf.l_whence = SEEK_SET;
2518 if (uap->how & LOCK_UN) {
2519 lf.l_type = F_UNLCK;
2520 fp->f_flag &= ~FHASLOCK;
2521 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2524 if (uap->how & LOCK_EX)
2525 lf.l_type = F_WRLCK;
2526 else if (uap->how & LOCK_SH)
2527 lf.l_type = F_RDLCK;
2532 fp->f_flag |= FHASLOCK;
2533 if (uap->how & LOCK_NB)
2534 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0);
2536 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT);
2543 * File Descriptor pseudo-device driver (/dev/fd/).
2545 * Opening minor device N dup()s the file (if any) connected to file
2546 * descriptor N belonging to the calling process. Note that this driver
2547 * consists of only the ``open()'' routine, because all subsequent
2548 * references to this file will be direct to the other driver.
2551 fdopen(struct dev_open_args *ap)
2553 thread_t td = curthread;
2555 KKASSERT(td->td_lwp != NULL);
2558 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the
2559 * the file descriptor being sought for duplication. The error
2560 * return ensures that the vnode for this device will be released
2561 * by vn_open. Open will detect this special error and take the
2562 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2563 * will simply report the error.
2565 td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev);
2570 * The caller has reserved the file descriptor dfd for us. On success we
2571 * must fsetfd() it. On failure the caller will clean it up.
2576 dupfdopen(struct filedesc *fdp, int dfd, int sfd, int mode, int error)
2582 if ((wfp = holdfp(fdp, sfd, -1)) == NULL)
2586 * Close a revoke/dup race. Duping a descriptor marked as revoked
2587 * will dup a dummy descriptor instead of the real one.
2589 if (wfp->f_flag & FREVOKED) {
2590 kprintf("Warning: attempt to dup() a revoked descriptor\n");
2593 werror = falloc(NULL, &wfp, NULL);
2599 * There are two cases of interest here.
2601 * For ENODEV simply dup sfd to file descriptor dfd and return.
2603 * For ENXIO steal away the file structure from sfd and store it
2604 * dfd. sfd is effectively closed by this operation.
2606 * Any other error code is just returned.
2611 * Check that the mode the file is being opened for is a
2612 * subset of the mode of the existing descriptor.
2614 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2618 spin_lock(&fdp->fd_spin);
2619 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2620 fsetfd_locked(fdp, wfp, dfd);
2621 spin_unlock(&fdp->fd_spin);
2626 * Steal away the file pointer from dfd, and stuff it into indx.
2628 spin_lock(&fdp->fd_spin);
2629 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2630 fsetfd(fdp, wfp, dfd);
2631 if ((xfp = funsetfd_locked(fdp, sfd)) != NULL) {
2632 spin_unlock(&fdp->fd_spin);
2635 spin_unlock(&fdp->fd_spin);
2647 * NOT MPSAFE - I think these refer to a common file descriptor table
2648 * and we need to spinlock that to link fdtol in.
2650 struct filedesc_to_leader *
2651 filedesc_to_leader_alloc(struct filedesc_to_leader *old,
2652 struct proc *leader)
2654 struct filedesc_to_leader *fdtol;
2656 fdtol = kmalloc(sizeof(struct filedesc_to_leader),
2657 M_FILEDESC_TO_LEADER, M_WAITOK | M_ZERO);
2658 fdtol->fdl_refcount = 1;
2659 fdtol->fdl_holdcount = 0;
2660 fdtol->fdl_wakeup = 0;
2661 fdtol->fdl_leader = leader;
2663 fdtol->fdl_next = old->fdl_next;
2664 fdtol->fdl_prev = old;
2665 old->fdl_next = fdtol;
2666 fdtol->fdl_next->fdl_prev = fdtol;
2668 fdtol->fdl_next = fdtol;
2669 fdtol->fdl_prev = fdtol;
2675 * Scan all file pointers in the system. The callback is made with
2676 * the master list spinlock held exclusively.
2681 allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data)
2686 spin_lock(&filehead_spin);
2687 LIST_FOREACH(fp, &filehead, f_list) {
2688 res = callback(fp, data);
2692 spin_unlock(&filehead_spin);
2696 * Get file structures.
2698 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe)
2701 struct sysctl_kern_file_info {
2704 struct sysctl_req *req;
2707 static int sysctl_kern_file_callback(struct proc *p, void *data);
2710 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2712 struct sysctl_kern_file_info info;
2715 * Note: because the number of file descriptors is calculated
2716 * in different ways for sizing vs returning the data,
2717 * there is information leakage from the first loop. However,
2718 * it is of a similar order of magnitude to the leakage from
2719 * global system statistics such as kern.openfiles.
2721 * When just doing a count, note that we cannot just count
2722 * the elements and add f_count via the filehead list because
2723 * threaded processes share their descriptor table and f_count might
2724 * still be '1' in that case.
2726 * Since the SYSCTL op can block, we must hold the process to
2727 * prevent it being ripped out from under us either in the
2728 * file descriptor loop or in the greater LIST_FOREACH. The
2729 * process may be in varying states of disrepair. If the process
2730 * is in SZOMB we may have caught it just as it is being removed
2731 * from the allproc list, we must skip it in that case to maintain
2732 * an unbroken chain through the allproc list.
2737 allproc_scan(sysctl_kern_file_callback, &info);
2740 * When just calculating the size, overestimate a bit to try to
2741 * prevent system activity from causing the buffer-fill call
2744 if (req->oldptr == NULL) {
2745 info.count = (info.count + 16) + (info.count / 10);
2746 info.error = SYSCTL_OUT(req, NULL,
2747 info.count * sizeof(struct kinfo_file));
2749 return (info.error);
2753 sysctl_kern_file_callback(struct proc *p, void *data)
2755 struct sysctl_kern_file_info *info = data;
2756 struct kinfo_file kf;
2757 struct filedesc *fdp;
2762 if (p->p_stat == SIDL || p->p_stat == SZOMB)
2764 if (!PRISON_CHECK(info->req->td->td_ucred, p->p_ucred) != 0)
2768 * Softref the fdp to prevent it from being destroyed
2770 spin_lock(&p->p_spin);
2771 if ((fdp = p->p_fd) == NULL) {
2772 spin_unlock(&p->p_spin);
2775 atomic_add_int(&fdp->fd_softrefs, 1);
2776 spin_unlock(&p->p_spin);
2779 * The fdp's own spinlock prevents the contents from being
2782 spin_lock(&fdp->fd_spin);
2783 for (n = 0; n < fdp->fd_nfiles; ++n) {
2784 if ((fp = fdp->fd_files[n].fp) == NULL)
2786 if (info->req->oldptr == NULL) {
2789 uid = p->p_ucred ? p->p_ucred->cr_uid : -1;
2790 kcore_make_file(&kf, fp, p->p_pid, uid, n);
2791 spin_unlock(&fdp->fd_spin);
2792 info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf));
2793 spin_lock(&fdp->fd_spin);
2798 spin_unlock(&fdp->fd_spin);
2799 atomic_subtract_int(&fdp->fd_softrefs, 1);
2805 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2806 0, 0, sysctl_kern_file, "S,file", "Entire file table");
2808 SYSCTL_INT(_kern, OID_AUTO, minfilesperproc, CTLFLAG_RW,
2809 &minfilesperproc, 0, "Minimum files allowed open per process");
2810 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2811 &maxfilesperproc, 0, "Maximum files allowed open per process");
2812 SYSCTL_INT(_kern, OID_AUTO, maxfilesperuser, CTLFLAG_RW,
2813 &maxfilesperuser, 0, "Maximum files allowed open per user");
2815 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2816 &maxfiles, 0, "Maximum number of files");
2818 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW,
2819 &maxfilesrootres, 0, "Descriptors reserved for root use");
2821 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2822 &nfiles, 0, "System-wide number of open files");
2825 fildesc_drvinit(void *unused)
2829 for (fd = 0; fd < NUMFDESC; fd++) {
2830 make_dev(&fildesc_ops, fd,
2831 UID_BIN, GID_BIN, 0666, "fd/%d", fd);
2834 make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin");
2835 make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout");
2836 make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr");
2842 struct fileops badfileops = {
2843 .fo_read = badfo_readwrite,
2844 .fo_write = badfo_readwrite,
2845 .fo_ioctl = badfo_ioctl,
2846 .fo_kqfilter = badfo_kqfilter,
2847 .fo_stat = badfo_stat,
2848 .fo_close = badfo_close,
2849 .fo_shutdown = badfo_shutdown
2863 badfo_ioctl(struct file *fp, u_long com, caddr_t data,
2864 struct ucred *cred, struct sysmsg *msgv)
2870 * Must return an error to prevent registration, typically
2871 * due to a revoked descriptor (file_filtops assigned).
2874 badfo_kqfilter(struct file *fp, struct knote *kn)
2876 return (EOPNOTSUPP);
2883 badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
2892 badfo_close(struct file *fp)
2901 badfo_shutdown(struct file *fp, int how)
2910 nofo_shutdown(struct file *fp, int how)
2912 return (EOPNOTSUPP);
2915 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,
2916 fildesc_drvinit,NULL)