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 $
73 * $DragonFly: src/sys/kern/kern_descrip.c,v 1.79 2008/08/31 13:18:28 aggelos Exp $
76 #include "opt_compat.h"
77 #include <sys/param.h>
78 #include <sys/systm.h>
79 #include <sys/malloc.h>
80 #include <sys/sysproto.h>
82 #include <sys/device.h>
84 #include <sys/filedesc.h>
85 #include <sys/kernel.h>
86 #include <sys/sysctl.h>
87 #include <sys/vnode.h>
89 #include <sys/nlookup.h>
92 #include <sys/filio.h>
93 #include <sys/fcntl.h>
94 #include <sys/unistd.h>
95 #include <sys/resourcevar.h>
96 #include <sys/event.h>
97 #include <sys/kern_syscall.h>
98 #include <sys/kcore.h>
99 #include <sys/kinfo.h>
103 #include <vm/vm_extern.h>
105 #include <sys/thread2.h>
106 #include <sys/file2.h>
107 #include <sys/spinlock2.h>
108 #include <sys/mplock2.h>
110 static void fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd);
111 static void fdreserve_locked (struct filedesc *fdp, int fd0, int incr);
112 static struct file *funsetfd_locked (struct filedesc *fdp, int fd);
113 static void ffree(struct file *fp);
115 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
116 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
117 "file desc to leader structures");
118 MALLOC_DEFINE(M_FILE, "file", "Open file structure");
119 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
121 static struct krate krate_uidinfo = { .freq = 1 };
123 static d_open_t fdopen;
126 #define CDEV_MAJOR 22
127 static struct dev_ops fildesc_ops = {
128 { "FD", CDEV_MAJOR, 0 },
133 * Descriptor management.
135 static struct filelist filehead = LIST_HEAD_INITIALIZER(&filehead);
136 static struct spinlock filehead_spin = SPINLOCK_INITIALIZER(&filehead_spin);
137 static int nfiles; /* actual number of open files */
141 * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared.
143 * MPSAFE - must be called with fdp->fd_spin exclusively held
147 fdfixup_locked(struct filedesc *fdp, int fd)
149 if (fd < fdp->fd_freefile) {
150 fdp->fd_freefile = fd;
152 while (fdp->fd_lastfile >= 0 &&
153 fdp->fd_files[fdp->fd_lastfile].fp == NULL &&
154 fdp->fd_files[fdp->fd_lastfile].reserved == 0
161 * System calls on descriptors.
166 sys_getdtablesize(struct getdtablesize_args *uap)
168 struct proc *p = curproc;
169 struct plimit *limit = p->p_limit;
172 spin_lock_rd(&limit->p_spin);
173 if (limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
176 dtsize = (int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur;
177 spin_unlock_rd(&limit->p_spin);
178 if (dtsize > maxfilesperproc)
179 dtsize = maxfilesperproc;
180 if (dtsize < minfilesperproc)
181 dtsize = minfilesperproc;
182 if (p->p_ucred->cr_uid && dtsize > maxfilesperuser)
183 dtsize = maxfilesperuser;
184 uap->sysmsg_result = dtsize;
189 * Duplicate a file descriptor to a particular value.
191 * note: keep in mind that a potential race condition exists when closing
192 * descriptors from a shared descriptor table (via rfork).
197 sys_dup2(struct dup2_args *uap)
202 error = kern_dup(DUP_FIXED, uap->from, uap->to, &fd);
203 uap->sysmsg_fds[0] = fd;
209 * Duplicate a file descriptor.
214 sys_dup(struct dup_args *uap)
219 error = kern_dup(DUP_VARIABLE, uap->fd, 0, &fd);
220 uap->sysmsg_fds[0] = fd;
226 * MPALMOSTSAFE - acquires mplock for fp operations
229 kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred)
231 struct thread *td = curthread;
232 struct proc *p = td->td_proc;
238 int tmp, error, flg = F_POSIX;
243 * Operations on file descriptors that do not require a file pointer.
247 error = fgetfdflags(p->p_fd, fd, &tmp);
249 dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0;
253 if (dat->fc_cloexec & FD_CLOEXEC)
254 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
256 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
260 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd);
267 * Operations on file pointers
269 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
275 dat->fc_flags = OFLAGS(fp->f_flag);
281 nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS;
282 nflags |= oflags & ~FCNTLFLAGS;
285 if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY))
287 if (error == 0 && ((nflags ^ oflags) & FASYNC)) {
288 tmp = nflags & FASYNC;
289 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp,
297 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner,
302 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner,
308 /* Fall into F_SETLK */
311 if (fp->f_type != DTYPE_VNODE) {
315 vp = (struct vnode *)fp->f_data;
318 * copyin/lockop may block
320 if (dat->fc_flock.l_whence == SEEK_CUR)
321 dat->fc_flock.l_start += fp->f_offset;
323 switch (dat->fc_flock.l_type) {
325 if ((fp->f_flag & FREAD) == 0) {
329 p->p_leader->p_flag |= P_ADVLOCK;
330 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
331 &dat->fc_flock, flg);
334 if ((fp->f_flag & FWRITE) == 0) {
338 p->p_leader->p_flag |= P_ADVLOCK;
339 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
340 &dat->fc_flock, flg);
343 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
344 &dat->fc_flock, F_POSIX);
352 * It is possible to race a close() on the descriptor while
353 * we were blocked getting the lock. If this occurs the
354 * close might not have caught the lock.
356 if (checkfdclosed(p->p_fd, fd, fp)) {
357 dat->fc_flock.l_whence = SEEK_SET;
358 dat->fc_flock.l_start = 0;
359 dat->fc_flock.l_len = 0;
360 dat->fc_flock.l_type = F_UNLCK;
361 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
362 F_UNLCK, &dat->fc_flock, F_POSIX);
367 if (fp->f_type != DTYPE_VNODE) {
371 vp = (struct vnode *)fp->f_data;
373 * copyin/lockop may block
375 if (dat->fc_flock.l_type != F_RDLCK &&
376 dat->fc_flock.l_type != F_WRLCK &&
377 dat->fc_flock.l_type != F_UNLCK) {
381 if (dat->fc_flock.l_whence == SEEK_CUR)
382 dat->fc_flock.l_start += fp->f_offset;
383 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK,
384 &dat->fc_flock, F_POSIX);
397 * The file control system call.
402 sys_fcntl(struct fcntl_args *uap)
409 dat.fc_fd = uap->arg;
412 dat.fc_cloexec = uap->arg;
415 dat.fc_flags = uap->arg;
418 dat.fc_owner = uap->arg;
423 error = copyin((caddr_t)uap->arg, &dat.fc_flock,
424 sizeof(struct flock));
430 error = kern_fcntl(uap->fd, uap->cmd, &dat, curthread->td_ucred);
435 uap->sysmsg_result = dat.fc_fd;
438 uap->sysmsg_result = dat.fc_cloexec;
441 uap->sysmsg_result = dat.fc_flags;
444 uap->sysmsg_result = dat.fc_owner;
446 error = copyout(&dat.fc_flock, (caddr_t)uap->arg,
447 sizeof(struct flock));
456 * Common code for dup, dup2, and fcntl(F_DUPFD).
458 * The type flag can be either DUP_FIXED or DUP_VARIABLE. DUP_FIXED tells
459 * kern_dup() to destructively dup over an existing file descriptor if new
460 * is already open. DUP_VARIABLE tells kern_dup() to find the lowest
461 * unused file descriptor that is greater than or equal to new.
466 kern_dup(enum dup_type type, int old, int new, int *res)
468 struct thread *td = curthread;
469 struct proc *p = td->td_proc;
470 struct filedesc *fdp = p->p_fd;
479 * Verify that we have a valid descriptor to dup from and
480 * possibly to dup to.
482 * NOTE: maxfilesperuser is not applicable to dup()
485 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
488 dtsize = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
489 if (dtsize > maxfilesperproc)
490 dtsize = maxfilesperproc;
491 if (dtsize < minfilesperproc)
492 dtsize = minfilesperproc;
494 if (new < 0 || new > dtsize)
497 spin_lock_wr(&fdp->fd_spin);
498 if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) {
499 spin_unlock_wr(&fdp->fd_spin);
502 if (type == DUP_FIXED && old == new) {
504 spin_unlock_wr(&fdp->fd_spin);
507 fp = fdp->fd_files[old].fp;
508 oldflags = fdp->fd_files[old].fileflags;
509 fhold(fp); /* MPSAFE - can be called with a spinlock held */
512 * Allocate a new descriptor if DUP_VARIABLE, or expand the table
513 * if the requested descriptor is beyond the current table size.
515 * This can block. Retry if the source descriptor no longer matches
516 * or if our expectation in the expansion case races.
518 * If we are not expanding or allocating a new decriptor, then reset
519 * the target descriptor to a reserved state so we have a uniform
520 * setup for the next code block.
522 if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) {
523 spin_unlock_wr(&fdp->fd_spin);
524 error = fdalloc(p, new, &newfd);
525 spin_lock_wr(&fdp->fd_spin);
527 spin_unlock_wr(&fdp->fd_spin);
534 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) {
535 fsetfd_locked(fdp, NULL, newfd);
536 spin_unlock_wr(&fdp->fd_spin);
541 * Check for expansion race
543 if (type != DUP_VARIABLE && new != newfd) {
544 fsetfd_locked(fdp, NULL, newfd);
545 spin_unlock_wr(&fdp->fd_spin);
550 * Check for ripout, newfd reused old (this case probably
554 fsetfd_locked(fdp, NULL, newfd);
555 spin_unlock_wr(&fdp->fd_spin);
562 if (fdp->fd_files[new].reserved) {
563 spin_unlock_wr(&fdp->fd_spin);
565 kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new);
566 tsleep(fdp, 0, "fdres", hz);
571 * If the target descriptor was never allocated we have
572 * to allocate it. If it was we have to clean out the
573 * old descriptor. delfp inherits the ref from the
576 delfp = fdp->fd_files[new].fp;
577 fdp->fd_files[new].fp = NULL;
578 fdp->fd_files[new].reserved = 1;
580 fdreserve_locked(fdp, new, 1);
581 if (new > fdp->fd_lastfile)
582 fdp->fd_lastfile = new;
588 * NOTE: still holding an exclusive spinlock
592 * If a descriptor is being overwritten we may hve to tell
593 * fdfree() to sleep to ensure that all relevant process
594 * leaders can be traversed in closef().
596 if (delfp != NULL && p->p_fdtol != NULL) {
597 fdp->fd_holdleaderscount++;
602 KASSERT(delfp == NULL || type == DUP_FIXED,
603 ("dup() picked an open file"));
606 * Duplicate the source descriptor, update lastfile. If the new
607 * descriptor was not allocated and we aren't replacing an existing
608 * descriptor we have to mark the descriptor as being in use.
610 * The fd_files[] array inherits fp's hold reference.
612 fsetfd_locked(fdp, fp, new);
613 fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE;
614 spin_unlock_wr(&fdp->fd_spin);
619 * If we dup'd over a valid file, we now own the reference to it
620 * and must dispose of it using closef() semantics (as if a
621 * close() were performed on it).
624 if (SLIST_FIRST(&delfp->f_klist))
625 knote_fdclose(delfp, fdp, new);
628 spin_lock_wr(&fdp->fd_spin);
629 fdp->fd_holdleaderscount--;
630 if (fdp->fd_holdleaderscount == 0 &&
631 fdp->fd_holdleaderswakeup != 0) {
632 fdp->fd_holdleaderswakeup = 0;
633 spin_unlock_wr(&fdp->fd_spin);
634 wakeup(&fdp->fd_holdleaderscount);
636 spin_unlock_wr(&fdp->fd_spin);
644 * If sigio is on the list associated with a process or process group,
645 * disable signalling from the device, remove sigio from the list and
649 funsetown(struct sigio *sigio)
654 *(sigio->sio_myref) = NULL;
656 if (sigio->sio_pgid < 0) {
657 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
659 } else /* if ((*sigiop)->sio_pgid > 0) */ {
660 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
663 crfree(sigio->sio_ucred);
664 kfree(sigio, M_SIGIO);
667 /* Free a list of sigio structures. */
669 funsetownlst(struct sigiolst *sigiolst)
673 while ((sigio = SLIST_FIRST(sigiolst)) != NULL)
678 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
680 * After permission checking, add a sigio structure to the sigio list for
681 * the process or process group.
684 fsetown(pid_t pgid, struct sigio **sigiop)
700 * Policy - Don't allow a process to FSETOWN a process
701 * in another session.
703 * Remove this test to allow maximum flexibility or
704 * restrict FSETOWN to the current process or process
705 * group for maximum safety.
707 if (proc->p_session != curproc->p_session)
711 } else /* if (pgid < 0) */ {
712 pgrp = pgfind(-pgid);
717 * Policy - Don't allow a process to FSETOWN a process
718 * in another session.
720 * Remove this test to allow maximum flexibility or
721 * restrict FSETOWN to the current process or process
722 * group for maximum safety.
724 if (pgrp->pg_session != curproc->p_session)
730 sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
732 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
733 sigio->sio_proc = proc;
735 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
736 sigio->sio_pgrp = pgrp;
738 sigio->sio_pgid = pgid;
739 sigio->sio_ucred = crhold(curthread->td_ucred);
740 /* It would be convenient if p_ruid was in ucred. */
741 sigio->sio_ruid = sigio->sio_ucred->cr_ruid;
742 sigio->sio_myref = sigiop;
750 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
753 fgetown(struct sigio *sigio)
755 return (sigio != NULL ? sigio->sio_pgid : 0);
759 * Close many file descriptors.
764 sys_closefrom(struct closefrom_args *uap)
766 return(kern_closefrom(uap->fd));
770 * Close all file descriptors greater then or equal to fd
775 kern_closefrom(int fd)
777 struct thread *td = curthread;
778 struct proc *p = td->td_proc;
779 struct filedesc *fdp;
788 * NOTE: This function will skip unassociated descriptors and
789 * reserved descriptors that have not yet been assigned.
790 * fd_lastfile can change as a side effect of kern_close().
792 spin_lock_wr(&fdp->fd_spin);
793 while (fd <= fdp->fd_lastfile) {
794 if (fdp->fd_files[fd].fp != NULL) {
795 spin_unlock_wr(&fdp->fd_spin);
796 /* ok if this races another close */
797 if (kern_close(fd) == EINTR)
799 spin_lock_wr(&fdp->fd_spin);
803 spin_unlock_wr(&fdp->fd_spin);
808 * Close a file descriptor.
813 sys_close(struct close_args *uap)
815 return(kern_close(uap->fd));
819 * MPALMOSTSAFE - acquires mplock around knote_fdclose() calls
824 struct thread *td = curthread;
825 struct proc *p = td->td_proc;
826 struct filedesc *fdp;
834 spin_lock_wr(&fdp->fd_spin);
835 if ((fp = funsetfd_locked(fdp, fd)) == NULL) {
836 spin_unlock_wr(&fdp->fd_spin);
840 if (p->p_fdtol != NULL) {
842 * Ask fdfree() to sleep to ensure that all relevant
843 * process leaders can be traversed in closef().
845 fdp->fd_holdleaderscount++;
850 * we now hold the fp reference that used to be owned by the descriptor
853 spin_unlock_wr(&fdp->fd_spin);
854 if (SLIST_FIRST(&fp->f_klist))
855 knote_fdclose(fp, fdp, fd);
856 error = closef(fp, p);
858 spin_lock_wr(&fdp->fd_spin);
859 fdp->fd_holdleaderscount--;
860 if (fdp->fd_holdleaderscount == 0 &&
861 fdp->fd_holdleaderswakeup != 0) {
862 fdp->fd_holdleaderswakeup = 0;
863 spin_unlock_wr(&fdp->fd_spin);
864 wakeup(&fdp->fd_holdleaderscount);
866 spin_unlock_wr(&fdp->fd_spin);
873 * shutdown_args(int fd, int how)
876 kern_shutdown(int fd, int how)
878 struct thread *td = curthread;
879 struct proc *p = td->td_proc;
885 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
887 error = fo_shutdown(fp, how);
897 sys_shutdown(struct shutdown_args *uap)
902 error = kern_shutdown(uap->s, uap->how);
912 kern_fstat(int fd, struct stat *ub)
914 struct thread *td = curthread;
915 struct proc *p = td->td_proc;
921 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
923 error = fo_stat(fp, ub, td->td_ucred);
930 * Return status information about a file descriptor.
935 sys_fstat(struct fstat_args *uap)
940 error = kern_fstat(uap->fd, &st);
943 error = copyout(&st, uap->sb, sizeof(st));
948 * Return pathconf information about a file descriptor.
953 sys_fpathconf(struct fpathconf_args *uap)
955 struct thread *td = curthread;
956 struct proc *p = td->td_proc;
961 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
964 switch (fp->f_type) {
967 if (uap->name != _PC_PIPE_BUF) {
970 uap->sysmsg_result = PIPE_BUF;
976 vp = (struct vnode *)fp->f_data;
978 error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg);
990 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand,
994 * Grow the file table so it can hold through descriptor (want).
996 * The fdp's spinlock must be held exclusively on entry and may be held
997 * exclusively on return. The spinlock may be cycled by the routine.
1002 fdgrow_locked(struct filedesc *fdp, int want)
1004 struct fdnode *newfiles;
1005 struct fdnode *oldfiles;
1008 nf = fdp->fd_nfiles;
1010 /* nf has to be of the form 2^n - 1 */
1012 } while (nf <= want);
1014 spin_unlock_wr(&fdp->fd_spin);
1015 newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK);
1016 spin_lock_wr(&fdp->fd_spin);
1019 * We could have raced another extend while we were not holding
1022 if (fdp->fd_nfiles >= nf) {
1023 spin_unlock_wr(&fdp->fd_spin);
1024 kfree(newfiles, M_FILEDESC);
1025 spin_lock_wr(&fdp->fd_spin);
1029 * Copy the existing ofile and ofileflags arrays
1030 * and zero the new portion of each array.
1032 extra = nf - fdp->fd_nfiles;
1033 bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode));
1034 bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode));
1036 oldfiles = fdp->fd_files;
1037 fdp->fd_files = newfiles;
1038 fdp->fd_nfiles = nf;
1040 if (oldfiles != fdp->fd_builtin_files) {
1041 spin_unlock_wr(&fdp->fd_spin);
1042 kfree(oldfiles, M_FILEDESC);
1043 spin_lock_wr(&fdp->fd_spin);
1049 * Number of nodes in right subtree, including the root.
1052 right_subtree_size(int n)
1054 return (n ^ (n | (n + 1)));
1061 right_ancestor(int n)
1063 return (n | (n + 1));
1070 left_ancestor(int n)
1072 return ((n & (n + 1)) - 1);
1076 * Traverse the in-place binary tree buttom-up adjusting the allocation
1077 * count so scans can determine where free descriptors are located.
1079 * MPSAFE - caller must be holding an exclusive spinlock on fdp
1083 fdreserve_locked(struct filedesc *fdp, int fd, int incr)
1086 fdp->fd_files[fd].allocated += incr;
1087 KKASSERT(fdp->fd_files[fd].allocated >= 0);
1088 fd = left_ancestor(fd);
1093 * Reserve a file descriptor for the process. If no error occurs, the
1094 * caller MUST at some point call fsetfd() or assign a file pointer
1095 * or dispose of the reservation.
1100 fdalloc(struct proc *p, int want, int *result)
1102 struct filedesc *fdp = p->p_fd;
1103 struct uidinfo *uip;
1104 int fd, rsize, rsum, node, lim;
1107 * Check dtable size limit
1109 spin_lock_rd(&p->p_limit->p_spin);
1110 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
1113 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
1114 spin_unlock_rd(&p->p_limit->p_spin);
1115 if (lim > maxfilesperproc)
1116 lim = maxfilesperproc;
1117 if (lim < minfilesperproc)
1118 lim = minfilesperproc;
1123 * Check that the user has not run out of descriptors (non-root only).
1124 * As a safety measure the dtable is allowed to have at least
1125 * minfilesperproc open fds regardless of the maxfilesperuser limit.
1127 if (p->p_ucred->cr_uid && fdp->fd_nfiles >= minfilesperproc) {
1128 uip = p->p_ucred->cr_uidinfo;
1129 if (uip->ui_openfiles > maxfilesperuser) {
1130 krateprintf(&krate_uidinfo,
1131 "Warning: user %d pid %d (%s) ran out of "
1132 "file descriptors (%d/%d)\n",
1133 p->p_ucred->cr_uid, (int)p->p_pid,
1135 uip->ui_openfiles, maxfilesperuser);
1141 * Grow the dtable if necessary
1143 spin_lock_wr(&fdp->fd_spin);
1144 if (want >= fdp->fd_nfiles)
1145 fdgrow_locked(fdp, want);
1148 * Search for a free descriptor starting at the higher
1149 * of want or fd_freefile. If that fails, consider
1150 * expanding the ofile array.
1152 * NOTE! the 'allocated' field is a cumulative recursive allocation
1153 * count. If we happen to see a value of 0 then we can shortcut
1154 * our search. Otherwise we run through through the tree going
1155 * down branches we know have free descriptor(s) until we hit a
1156 * leaf node. The leaf node will be free but will not necessarily
1157 * have an allocated field of 0.
1160 /* move up the tree looking for a subtree with a free node */
1161 for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim);
1162 fd = right_ancestor(fd)) {
1163 if (fdp->fd_files[fd].allocated == 0)
1166 rsize = right_subtree_size(fd);
1167 if (fdp->fd_files[fd].allocated == rsize)
1168 continue; /* right subtree full */
1171 * Free fd is in the right subtree of the tree rooted at fd.
1172 * Call that subtree R. Look for the smallest (leftmost)
1173 * subtree of R with an unallocated fd: continue moving
1174 * down the left branch until encountering a full left
1175 * subtree, then move to the right.
1177 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) {
1179 rsum += fdp->fd_files[node].allocated;
1180 if (fdp->fd_files[fd].allocated == rsum + rsize) {
1181 fd = node; /* move to the right */
1182 if (fdp->fd_files[node].allocated == 0)
1191 * No space in current array. Expand?
1193 if (fdp->fd_nfiles >= lim) {
1194 spin_unlock_wr(&fdp->fd_spin);
1197 fdgrow_locked(fdp, want);
1201 KKASSERT(fd < fdp->fd_nfiles);
1202 if (fd > fdp->fd_lastfile)
1203 fdp->fd_lastfile = fd;
1204 if (want <= fdp->fd_freefile)
1205 fdp->fd_freefile = fd;
1207 KKASSERT(fdp->fd_files[fd].fp == NULL);
1208 KKASSERT(fdp->fd_files[fd].reserved == 0);
1209 fdp->fd_files[fd].fileflags = 0;
1210 fdp->fd_files[fd].reserved = 1;
1211 fdreserve_locked(fdp, fd, 1);
1212 spin_unlock_wr(&fdp->fd_spin);
1217 * Check to see whether n user file descriptors
1218 * are available to the process p.
1223 fdavail(struct proc *p, int n)
1225 struct filedesc *fdp = p->p_fd;
1226 struct fdnode *fdnode;
1229 spin_lock_rd(&p->p_limit->p_spin);
1230 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
1233 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
1234 spin_unlock_rd(&p->p_limit->p_spin);
1235 if (lim > maxfilesperproc)
1236 lim = maxfilesperproc;
1237 if (lim < minfilesperproc)
1238 lim = minfilesperproc;
1240 spin_lock_rd(&fdp->fd_spin);
1241 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) {
1242 spin_unlock_rd(&fdp->fd_spin);
1245 last = min(fdp->fd_nfiles, lim);
1246 fdnode = &fdp->fd_files[fdp->fd_freefile];
1247 for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) {
1248 if (fdnode->fp == NULL && --n <= 0) {
1249 spin_unlock_rd(&fdp->fd_spin);
1253 spin_unlock_rd(&fdp->fd_spin);
1258 * Revoke open descriptors referencing (f_data, f_type)
1260 * Any revoke executed within a prison is only able to
1261 * revoke descriptors for processes within that prison.
1263 * Returns 0 on success or an error code.
1265 struct fdrevoke_info {
1275 static int fdrevoke_check_callback(struct file *fp, void *vinfo);
1276 static int fdrevoke_proc_callback(struct proc *p, void *vinfo);
1279 fdrevoke(void *f_data, short f_type, struct ucred *cred)
1281 struct fdrevoke_info info;
1284 bzero(&info, sizeof(info));
1288 error = falloc(NULL, &info.nfp, NULL);
1293 * Scan the file pointer table once. dups do not dup file pointers,
1294 * only descriptors, so there is no leak. Set FREVOKED on the fps
1297 allfiles_scan_exclusive(fdrevoke_check_callback, &info);
1300 * If any fps were marked track down the related descriptors
1301 * and close them. Any dup()s at this point will notice
1302 * the FREVOKED already set in the fp and do the right thing.
1304 * Any fps with non-zero msgcounts (aka sent over a unix-domain
1305 * socket) bumped the intransit counter and will require a
1306 * scan. Races against fps leaving the socket are closed by
1307 * the socket code checking for FREVOKED.
1310 allproc_scan(fdrevoke_proc_callback, &info);
1312 unp_revoke_gc(info.nfp);
1318 * Locate matching file pointers directly.
1321 fdrevoke_check_callback(struct file *fp, void *vinfo)
1323 struct fdrevoke_info *info = vinfo;
1326 * File pointers already flagged for revokation are skipped.
1328 if (fp->f_flag & FREVOKED)
1332 * If revoking from a prison file pointers created outside of
1333 * that prison, or file pointers without creds, cannot be revoked.
1335 if (info->cred->cr_prison &&
1336 (fp->f_cred == NULL ||
1337 info->cred->cr_prison != fp->f_cred->cr_prison)) {
1342 * If the file pointer matches then mark it for revocation. The
1343 * flag is currently only used by unp_revoke_gc().
1345 * info->count is a heuristic and can race in a SMP environment.
1347 if (info->data == fp->f_data && info->type == fp->f_type) {
1348 atomic_set_int(&fp->f_flag, FREVOKED);
1349 info->count += fp->f_count;
1357 * Locate matching file pointers via process descriptor tables.
1360 fdrevoke_proc_callback(struct proc *p, void *vinfo)
1362 struct fdrevoke_info *info = vinfo;
1363 struct filedesc *fdp;
1367 if (p->p_stat == SIDL || p->p_stat == SZOMB)
1369 if (info->cred->cr_prison &&
1370 info->cred->cr_prison != p->p_ucred->cr_prison) {
1375 * If the controlling terminal of the process matches the
1376 * vnode being revoked we clear the controlling terminal.
1378 * The normal spec_close() may not catch this because it
1379 * uses curproc instead of p.
1381 if (p->p_session && info->type == DTYPE_VNODE &&
1382 info->data == p->p_session->s_ttyvp) {
1383 p->p_session->s_ttyvp = NULL;
1388 * Softref the fdp to prevent it from being destroyed
1390 spin_lock_wr(&p->p_spin);
1391 if ((fdp = p->p_fd) == NULL) {
1392 spin_unlock_wr(&p->p_spin);
1395 atomic_add_int(&fdp->fd_softrefs, 1);
1396 spin_unlock_wr(&p->p_spin);
1399 * Locate and close any matching file descriptors.
1401 spin_lock_wr(&fdp->fd_spin);
1402 for (n = 0; n < fdp->fd_nfiles; ++n) {
1403 if ((fp = fdp->fd_files[n].fp) == NULL)
1405 if (fp->f_flag & FREVOKED) {
1407 fdp->fd_files[n].fp = info->nfp;
1408 spin_unlock_wr(&fdp->fd_spin);
1409 knote_fdclose(fp, fdp, n); /* XXX */
1411 spin_lock_wr(&fdp->fd_spin);
1415 spin_unlock_wr(&fdp->fd_spin);
1416 atomic_subtract_int(&fdp->fd_softrefs, 1);
1422 * Create a new open file structure and reserve a file decriptor
1423 * for the process that refers to it.
1425 * Root creds are checked using lp, or assumed if lp is NULL. If
1426 * resultfd is non-NULL then lp must also be non-NULL. No file
1427 * descriptor is reserved (and no process context is needed) if
1430 * A file pointer with a refcount of 1 is returned. Note that the
1431 * file pointer is NOT associated with the descriptor. If falloc
1432 * returns success, fsetfd() MUST be called to either associate the
1433 * file pointer or clear the reservation.
1438 falloc(struct lwp *lp, struct file **resultfp, int *resultfd)
1440 static struct timeval lastfail;
1443 struct ucred *cred = lp ? lp->lwp_thread->td_ucred : proc0.p_ucred;
1449 * Handle filetable full issues and root overfill.
1451 if (nfiles >= maxfiles - maxfilesrootres &&
1452 (cred->cr_ruid != 0 || nfiles >= maxfiles)) {
1453 if (ppsratecheck(&lastfail, &curfail, 1)) {
1454 kprintf("kern.maxfiles limit exceeded by uid %d, "
1455 "please see tuning(7).\n",
1463 * Allocate a new file descriptor.
1465 fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO);
1466 spin_init(&fp->f_spin);
1467 SLIST_INIT(&fp->f_klist);
1469 fp->f_ops = &badfileops;
1472 spin_lock_wr(&filehead_spin);
1474 LIST_INSERT_HEAD(&filehead, fp, f_list);
1475 spin_unlock_wr(&filehead_spin);
1477 if ((error = fdalloc(lp->lwp_proc, 0, resultfd)) != 0) {
1490 * Check for races against a file descriptor by determining that the
1491 * file pointer is still associated with the specified file descriptor,
1492 * and a close is not currently in progress.
1497 checkfdclosed(struct filedesc *fdp, int fd, struct file *fp)
1501 spin_lock_rd(&fdp->fd_spin);
1502 if ((unsigned)fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp)
1506 spin_unlock_rd(&fdp->fd_spin);
1511 * Associate a file pointer with a previously reserved file descriptor.
1512 * This function always succeeds.
1514 * If fp is NULL, the file descriptor is returned to the pool.
1518 * MPSAFE (exclusive spinlock must be held on call)
1521 fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd)
1523 KKASSERT((unsigned)fd < fdp->fd_nfiles);
1524 KKASSERT(fdp->fd_files[fd].reserved != 0);
1527 fdp->fd_files[fd].fp = fp;
1528 fdp->fd_files[fd].reserved = 0;
1530 fdp->fd_files[fd].reserved = 0;
1531 fdreserve_locked(fdp, fd, -1);
1532 fdfixup_locked(fdp, fd);
1540 fsetfd(struct filedesc *fdp, struct file *fp, int fd)
1542 spin_lock_wr(&fdp->fd_spin);
1543 fsetfd_locked(fdp, fp, fd);
1544 spin_unlock_wr(&fdp->fd_spin);
1548 * MPSAFE (exclusive spinlock must be held on call)
1552 funsetfd_locked(struct filedesc *fdp, int fd)
1556 if ((unsigned)fd >= fdp->fd_nfiles)
1558 if ((fp = fdp->fd_files[fd].fp) == NULL)
1560 fdp->fd_files[fd].fp = NULL;
1561 fdp->fd_files[fd].fileflags = 0;
1563 fdreserve_locked(fdp, fd, -1);
1564 fdfixup_locked(fdp, fd);
1572 fgetfdflags(struct filedesc *fdp, int fd, int *flagsp)
1576 spin_lock_rd(&fdp->fd_spin);
1577 if (((u_int)fd) >= fdp->fd_nfiles) {
1579 } else if (fdp->fd_files[fd].fp == NULL) {
1582 *flagsp = fdp->fd_files[fd].fileflags;
1585 spin_unlock_rd(&fdp->fd_spin);
1593 fsetfdflags(struct filedesc *fdp, int fd, int add_flags)
1597 spin_lock_wr(&fdp->fd_spin);
1598 if (((u_int)fd) >= fdp->fd_nfiles) {
1600 } else if (fdp->fd_files[fd].fp == NULL) {
1603 fdp->fd_files[fd].fileflags |= add_flags;
1606 spin_unlock_wr(&fdp->fd_spin);
1614 fclrfdflags(struct filedesc *fdp, int fd, int rem_flags)
1618 spin_lock_wr(&fdp->fd_spin);
1619 if (((u_int)fd) >= fdp->fd_nfiles) {
1621 } else if (fdp->fd_files[fd].fp == NULL) {
1624 fdp->fd_files[fd].fileflags &= ~rem_flags;
1627 spin_unlock_wr(&fdp->fd_spin);
1632 * Set/Change/Clear the creds for a fp and synchronize the uidinfo.
1635 fsetcred(struct file *fp, struct ucred *ncr)
1638 struct uidinfo *uip;
1641 if (ocr == NULL || ncr == NULL || ocr->cr_uidinfo != ncr->cr_uidinfo) {
1643 uip = ocr->cr_uidinfo;
1644 atomic_add_int(&uip->ui_openfiles, -1);
1647 uip = ncr->cr_uidinfo;
1648 atomic_add_int(&uip->ui_openfiles, 1);
1659 * Free a file descriptor.
1663 ffree(struct file *fp)
1665 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!"));
1666 spin_lock_wr(&filehead_spin);
1667 LIST_REMOVE(fp, f_list);
1669 spin_unlock_wr(&filehead_spin);
1671 if (fp->f_nchandle.ncp)
1672 cache_drop(&fp->f_nchandle);
1677 * called from init_main, initialize filedesc0 for proc0.
1680 fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask)
1684 fdp0->fd_refcnt = 1;
1685 fdp0->fd_cmask = cmask;
1686 fdp0->fd_files = fdp0->fd_builtin_files;
1687 fdp0->fd_nfiles = NDFILE;
1688 fdp0->fd_lastfile = -1;
1689 spin_init(&fdp0->fd_spin);
1693 * Build a new filedesc structure.
1698 fdinit(struct proc *p)
1700 struct filedesc *newfdp;
1701 struct filedesc *fdp = p->p_fd;
1703 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO);
1704 spin_lock_rd(&fdp->fd_spin);
1706 newfdp->fd_cdir = fdp->fd_cdir;
1707 vref(newfdp->fd_cdir);
1708 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1712 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of
1713 * proc0, but should unconditionally exist in other processes.
1716 newfdp->fd_rdir = fdp->fd_rdir;
1717 vref(newfdp->fd_rdir);
1718 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1721 newfdp->fd_jdir = fdp->fd_jdir;
1722 vref(newfdp->fd_jdir);
1723 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1725 spin_unlock_rd(&fdp->fd_spin);
1727 /* Create the file descriptor table. */
1728 newfdp->fd_refcnt = 1;
1729 newfdp->fd_cmask = cmask;
1730 newfdp->fd_files = newfdp->fd_builtin_files;
1731 newfdp->fd_nfiles = NDFILE;
1732 newfdp->fd_lastfile = -1;
1733 spin_init(&newfdp->fd_spin);
1739 * Share a filedesc structure.
1744 fdshare(struct proc *p)
1746 struct filedesc *fdp;
1749 spin_lock_wr(&fdp->fd_spin);
1751 spin_unlock_wr(&fdp->fd_spin);
1756 * Copy a filedesc structure.
1761 fdcopy(struct proc *p)
1763 struct filedesc *fdp = p->p_fd;
1764 struct filedesc *newfdp;
1765 struct fdnode *fdnode;
1770 * Certain daemons might not have file descriptors.
1776 * Allocate the new filedesc and fd_files[] array. This can race
1777 * with operations by other threads on the fdp so we have to be
1780 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK | M_ZERO);
1782 spin_lock_rd(&fdp->fd_spin);
1783 if (fdp->fd_lastfile < NDFILE) {
1784 newfdp->fd_files = newfdp->fd_builtin_files;
1788 * We have to allocate (N^2-1) entries for our in-place
1789 * binary tree. Allow the table to shrink.
1793 while (ni > fdp->fd_lastfile && ni > NDFILE) {
1797 spin_unlock_rd(&fdp->fd_spin);
1798 newfdp->fd_files = kmalloc(i * sizeof(struct fdnode),
1799 M_FILEDESC, M_WAITOK | M_ZERO);
1802 * Check for race, retry
1804 spin_lock_rd(&fdp->fd_spin);
1805 if (i <= fdp->fd_lastfile) {
1806 spin_unlock_rd(&fdp->fd_spin);
1807 kfree(newfdp->fd_files, M_FILEDESC);
1813 * Dup the remaining fields. vref() and cache_hold() can be
1814 * safely called while holding the read spinlock on fdp.
1816 * The read spinlock on fdp is still being held.
1818 * NOTE: vref and cache_hold calls for the case where the vnode
1819 * or cache entry already has at least one ref may be called
1820 * while holding spin locks.
1822 if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) {
1823 vref(newfdp->fd_cdir);
1824 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1827 * We must check for fd_rdir here, at least for now because
1828 * the init process is created before we have access to the
1829 * rootvode to take a reference to it.
1831 if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) {
1832 vref(newfdp->fd_rdir);
1833 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1835 if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) {
1836 vref(newfdp->fd_jdir);
1837 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1839 newfdp->fd_refcnt = 1;
1840 newfdp->fd_nfiles = i;
1841 newfdp->fd_lastfile = fdp->fd_lastfile;
1842 newfdp->fd_freefile = fdp->fd_freefile;
1843 newfdp->fd_cmask = fdp->fd_cmask;
1844 spin_init(&newfdp->fd_spin);
1847 * Copy the descriptor table through (i). This also copies the
1848 * allocation state. Then go through and ref the file pointers
1849 * and clean up any KQ descriptors.
1851 * kq descriptors cannot be copied. Since we haven't ref'd the
1852 * copied files yet we can ignore the return value from funsetfd().
1854 * The read spinlock on fdp is still being held.
1856 bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode));
1857 for (i = 0 ; i < newfdp->fd_nfiles; ++i) {
1858 fdnode = &newfdp->fd_files[i];
1859 if (fdnode->reserved) {
1860 fdreserve_locked(newfdp, i, -1);
1861 fdnode->reserved = 0;
1862 fdfixup_locked(newfdp, i);
1863 } else if (fdnode->fp) {
1864 if (fdnode->fp->f_type == DTYPE_KQUEUE) {
1865 (void)funsetfd_locked(newfdp, i);
1871 spin_unlock_rd(&fdp->fd_spin);
1876 * Release a filedesc structure.
1878 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE)
1881 fdfree(struct proc *p, struct filedesc *repl)
1883 struct filedesc *fdp;
1884 struct fdnode *fdnode;
1886 struct filedesc_to_leader *fdtol;
1892 * Certain daemons might not have file descriptors.
1901 * Severe messing around to follow.
1903 spin_lock_wr(&fdp->fd_spin);
1905 /* Check for special need to clear POSIX style locks */
1907 if (fdtol != NULL) {
1908 KASSERT(fdtol->fdl_refcount > 0,
1909 ("filedesc_to_refcount botch: fdl_refcount=%d",
1910 fdtol->fdl_refcount));
1911 if (fdtol->fdl_refcount == 1 &&
1912 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1913 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1914 fdnode = &fdp->fd_files[i];
1915 if (fdnode->fp == NULL ||
1916 fdnode->fp->f_type != DTYPE_VNODE) {
1921 spin_unlock_wr(&fdp->fd_spin);
1923 lf.l_whence = SEEK_SET;
1926 lf.l_type = F_UNLCK;
1927 vp = (struct vnode *)fp->f_data;
1928 (void) VOP_ADVLOCK(vp,
1929 (caddr_t)p->p_leader,
1934 spin_lock_wr(&fdp->fd_spin);
1938 if (fdtol->fdl_refcount == 1) {
1939 if (fdp->fd_holdleaderscount > 0 &&
1940 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1942 * close() or do_dup() has cleared a reference
1943 * in a shared file descriptor table.
1945 fdp->fd_holdleaderswakeup = 1;
1946 ssleep(&fdp->fd_holdleaderscount,
1947 &fdp->fd_spin, 0, "fdlhold", 0);
1950 if (fdtol->fdl_holdcount > 0) {
1952 * Ensure that fdtol->fdl_leader
1953 * remains valid in closef().
1955 fdtol->fdl_wakeup = 1;
1956 ssleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0);
1960 fdtol->fdl_refcount--;
1961 if (fdtol->fdl_refcount == 0 &&
1962 fdtol->fdl_holdcount == 0) {
1963 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1964 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1969 if (fdtol != NULL) {
1970 spin_unlock_wr(&fdp->fd_spin);
1971 kfree(fdtol, M_FILEDESC_TO_LEADER);
1972 spin_lock_wr(&fdp->fd_spin);
1975 if (--fdp->fd_refcnt > 0) {
1976 spin_unlock_wr(&fdp->fd_spin);
1977 spin_lock_wr(&p->p_spin);
1979 spin_unlock_wr(&p->p_spin);
1984 * Even though we are the last reference to the structure allproc
1985 * scans may still reference the structure. Maintain proper
1986 * locks until we can replace p->p_fd.
1988 * Also note that kqueue's closef still needs to reference the
1989 * fdp via p->p_fd, so we have to close the descriptors before
1990 * we replace p->p_fd.
1992 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1993 if (fdp->fd_files[i].fp) {
1994 fp = funsetfd_locked(fdp, i);
1996 spin_unlock_wr(&fdp->fd_spin);
1997 if (SLIST_FIRST(&fp->f_klist))
1998 knote_fdclose(fp, fdp, i);
2000 spin_lock_wr(&fdp->fd_spin);
2004 spin_unlock_wr(&fdp->fd_spin);
2007 * Interlock against an allproc scan operations (typically frevoke).
2009 spin_lock_wr(&p->p_spin);
2011 spin_unlock_wr(&p->p_spin);
2014 * Wait for any softrefs to go away. This race rarely occurs so
2015 * we can use a non-critical-path style poll/sleep loop. The
2016 * race only occurs against allproc scans.
2018 * No new softrefs can occur with the fdp disconnected from the
2021 if (fdp->fd_softrefs) {
2022 kprintf("pid %d: Warning, fdp race avoided\n", p->p_pid);
2023 while (fdp->fd_softrefs)
2024 tsleep(&fdp->fd_softrefs, 0, "fdsoft", 1);
2027 if (fdp->fd_files != fdp->fd_builtin_files)
2028 kfree(fdp->fd_files, M_FILEDESC);
2030 cache_drop(&fdp->fd_ncdir);
2031 vrele(fdp->fd_cdir);
2034 cache_drop(&fdp->fd_nrdir);
2035 vrele(fdp->fd_rdir);
2038 cache_drop(&fdp->fd_njdir);
2039 vrele(fdp->fd_jdir);
2041 kfree(fdp, M_FILEDESC);
2045 * Retrieve and reference the file pointer associated with a descriptor.
2050 holdfp(struct filedesc *fdp, int fd, int flag)
2054 spin_lock_rd(&fdp->fd_spin);
2055 if (((u_int)fd) >= fdp->fd_nfiles) {
2059 if ((fp = fdp->fd_files[fd].fp) == NULL)
2061 if ((fp->f_flag & flag) == 0 && flag != -1) {
2067 spin_unlock_rd(&fdp->fd_spin);
2072 * holdsock() - load the struct file pointer associated
2073 * with a socket into *fpp. If an error occurs, non-zero
2074 * will be returned and *fpp will be set to NULL.
2079 holdsock(struct filedesc *fdp, int fd, struct file **fpp)
2084 spin_lock_rd(&fdp->fd_spin);
2085 if ((unsigned)fd >= fdp->fd_nfiles) {
2090 if ((fp = fdp->fd_files[fd].fp) == NULL) {
2094 if (fp->f_type != DTYPE_SOCKET) {
2101 spin_unlock_rd(&fdp->fd_spin);
2107 * Convert a user file descriptor to a held file pointer.
2112 holdvnode(struct filedesc *fdp, int fd, struct file **fpp)
2117 spin_lock_rd(&fdp->fd_spin);
2118 if ((unsigned)fd >= fdp->fd_nfiles) {
2123 if ((fp = fdp->fd_files[fd].fp) == NULL) {
2127 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
2135 spin_unlock_rd(&fdp->fd_spin);
2141 * For setugid programs, we don't want to people to use that setugidness
2142 * to generate error messages which write to a file which otherwise would
2143 * otherwise be off-limits to the process.
2145 * This is a gross hack to plug the hole. A better solution would involve
2146 * a special vop or other form of generalized access control mechanism. We
2147 * go ahead and just reject all procfs file systems accesses as dangerous.
2149 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2150 * sufficient. We also don't for check setugidness since we know we are.
2153 is_unsafe(struct file *fp)
2155 if (fp->f_type == DTYPE_VNODE &&
2156 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS)
2162 * Make this setguid thing safe, if at all possible.
2164 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2167 setugidsafety(struct proc *p)
2169 struct filedesc *fdp = p->p_fd;
2172 /* Certain daemons might not have file descriptors. */
2177 * note: fdp->fd_files may be reallocated out from under us while
2178 * we are blocked in a close. Be careful!
2180 for (i = 0; i <= fdp->fd_lastfile; i++) {
2183 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) {
2187 * NULL-out descriptor prior to close to avoid
2188 * a race while close blocks.
2190 if ((fp = funsetfd_locked(fdp, i)) != NULL) {
2191 knote_fdclose(fp, fdp, i);
2199 * Close any files on exec?
2201 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2204 fdcloseexec(struct proc *p)
2206 struct filedesc *fdp = p->p_fd;
2209 /* Certain daemons might not have file descriptors. */
2214 * We cannot cache fd_files since operations may block and rip
2215 * them out from under us.
2217 for (i = 0; i <= fdp->fd_lastfile; i++) {
2218 if (fdp->fd_files[i].fp != NULL &&
2219 (fdp->fd_files[i].fileflags & UF_EXCLOSE)) {
2223 * NULL-out descriptor prior to close to avoid
2224 * a race while close blocks.
2226 if ((fp = funsetfd_locked(fdp, i)) != NULL) {
2227 knote_fdclose(fp, fdp, i);
2235 * It is unsafe for set[ug]id processes to be started with file
2236 * descriptors 0..2 closed, as these descriptors are given implicit
2237 * significance in the Standard C library. fdcheckstd() will create a
2238 * descriptor referencing /dev/null for each of stdin, stdout, and
2239 * stderr that is not already open.
2241 * NOT MPSAFE - calls falloc, vn_open, etc
2244 fdcheckstd(struct lwp *lp)
2246 struct nlookupdata nd;
2247 struct filedesc *fdp;
2250 int i, error, flags, devnull;
2252 fdp = lp->lwp_proc->p_fd;
2257 for (i = 0; i < 3; i++) {
2258 if (fdp->fd_files[i].fp != NULL)
2261 if ((error = falloc(lp, &fp, &devnull)) != 0)
2264 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE,
2265 NLC_FOLLOW|NLC_LOCKVP);
2266 flags = FREAD | FWRITE;
2268 error = vn_open(&nd, fp, flags, 0);
2270 fsetfd(fdp, fp, devnull);
2272 fsetfd(fdp, NULL, devnull);
2277 KKASSERT(i == devnull);
2279 error = kern_dup(DUP_FIXED, devnull, i, &retval);
2288 * Internal form of close.
2289 * Decrement reference count on file structure.
2290 * Note: td and/or p may be NULL when closing a file
2291 * that was being passed in a message.
2293 * MPALMOSTSAFE - acquires mplock for VOP operations
2296 closef(struct file *fp, struct proc *p)
2300 struct filedesc_to_leader *fdtol;
2306 * POSIX record locking dictates that any close releases ALL
2307 * locks owned by this process. This is handled by setting
2308 * a flag in the unlock to free ONLY locks obeying POSIX
2309 * semantics, and not to free BSD-style file locks.
2310 * If the descriptor was in a message, POSIX-style locks
2311 * aren't passed with the descriptor.
2313 if (p != NULL && fp->f_type == DTYPE_VNODE &&
2314 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2317 if ((p->p_leader->p_flag & P_ADVLOCK) != 0) {
2318 lf.l_whence = SEEK_SET;
2321 lf.l_type = F_UNLCK;
2322 vp = (struct vnode *)fp->f_data;
2323 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
2327 if (fdtol != NULL) {
2329 * Handle special case where file descriptor table
2330 * is shared between multiple process leaders.
2332 for (fdtol = fdtol->fdl_next;
2333 fdtol != p->p_fdtol;
2334 fdtol = fdtol->fdl_next) {
2335 if ((fdtol->fdl_leader->p_flag &
2338 fdtol->fdl_holdcount++;
2339 lf.l_whence = SEEK_SET;
2342 lf.l_type = F_UNLCK;
2343 vp = (struct vnode *)fp->f_data;
2344 (void) VOP_ADVLOCK(vp,
2345 (caddr_t)fdtol->fdl_leader,
2346 F_UNLCK, &lf, F_POSIX);
2347 fdtol->fdl_holdcount--;
2348 if (fdtol->fdl_holdcount == 0 &&
2349 fdtol->fdl_wakeup != 0) {
2350 fdtol->fdl_wakeup = 0;
2363 * fhold() can only be called if f_count is already at least 1 (i.e. the
2364 * caller of fhold() already has a reference to the file pointer in some
2367 * f_count is not spin-locked. Instead, atomic ops are used for
2368 * incrementing, decrementing, and handling the 1->0 transition.
2371 fhold(struct file *fp)
2373 atomic_add_int(&fp->f_count, 1);
2377 * fdrop() - drop a reference to a descriptor
2379 * MPALMOSTSAFE - acquires mplock for final close sequence
2382 fdrop(struct file *fp)
2389 * A combined fetch and subtract is needed to properly detect
2390 * 1->0 transitions, otherwise two cpus dropping from a ref
2391 * count of 2 might both try to run the 1->0 code.
2393 if (atomic_fetchadd_int(&fp->f_count, -1) > 1)
2396 KKASSERT(SLIST_FIRST(&fp->f_klist) == NULL);
2400 * The last reference has gone away, we own the fp structure free
2403 if (fp->f_count < 0)
2404 panic("fdrop: count < 0");
2405 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE &&
2406 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2408 lf.l_whence = SEEK_SET;
2411 lf.l_type = F_UNLCK;
2412 vp = (struct vnode *)fp->f_data;
2413 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2415 if (fp->f_ops != &badfileops)
2416 error = fo_close(fp);
2425 * Apply an advisory lock on a file descriptor.
2427 * Just attempt to get a record lock of the requested type on
2428 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2433 sys_flock(struct flock_args *uap)
2435 struct proc *p = curproc;
2441 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
2444 if (fp->f_type != DTYPE_VNODE) {
2448 vp = (struct vnode *)fp->f_data;
2449 lf.l_whence = SEEK_SET;
2452 if (uap->how & LOCK_UN) {
2453 lf.l_type = F_UNLCK;
2454 fp->f_flag &= ~FHASLOCK;
2455 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2458 if (uap->how & LOCK_EX)
2459 lf.l_type = F_WRLCK;
2460 else if (uap->how & LOCK_SH)
2461 lf.l_type = F_RDLCK;
2466 fp->f_flag |= FHASLOCK;
2467 if (uap->how & LOCK_NB)
2468 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0);
2470 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT);
2478 * File Descriptor pseudo-device driver (/dev/fd/).
2480 * Opening minor device N dup()s the file (if any) connected to file
2481 * descriptor N belonging to the calling process. Note that this driver
2482 * consists of only the ``open()'' routine, because all subsequent
2483 * references to this file will be direct to the other driver.
2486 fdopen(struct dev_open_args *ap)
2488 thread_t td = curthread;
2490 KKASSERT(td->td_lwp != NULL);
2493 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the
2494 * the file descriptor being sought for duplication. The error
2495 * return ensures that the vnode for this device will be released
2496 * by vn_open. Open will detect this special error and take the
2497 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2498 * will simply report the error.
2500 td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev);
2505 * The caller has reserved the file descriptor dfd for us. On success we
2506 * must fsetfd() it. On failure the caller will clean it up.
2511 dupfdopen(struct filedesc *fdp, int dfd, int sfd, int mode, int error)
2517 if ((wfp = holdfp(fdp, sfd, -1)) == NULL)
2521 * Close a revoke/dup race. Duping a descriptor marked as revoked
2522 * will dup a dummy descriptor instead of the real one.
2524 if (wfp->f_flag & FREVOKED) {
2525 kprintf("Warning: attempt to dup() a revoked descriptor\n");
2528 werror = falloc(NULL, &wfp, NULL);
2534 * There are two cases of interest here.
2536 * For ENODEV simply dup sfd to file descriptor dfd and return.
2538 * For ENXIO steal away the file structure from sfd and store it
2539 * dfd. sfd is effectively closed by this operation.
2541 * Any other error code is just returned.
2546 * Check that the mode the file is being opened for is a
2547 * subset of the mode of the existing descriptor.
2549 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2553 spin_lock_wr(&fdp->fd_spin);
2554 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2555 fsetfd_locked(fdp, wfp, dfd);
2556 spin_unlock_wr(&fdp->fd_spin);
2561 * Steal away the file pointer from dfd, and stuff it into indx.
2563 spin_lock_wr(&fdp->fd_spin);
2564 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2565 fsetfd(fdp, wfp, dfd);
2566 if ((xfp = funsetfd_locked(fdp, sfd)) != NULL) {
2567 spin_unlock_wr(&fdp->fd_spin);
2570 spin_unlock_wr(&fdp->fd_spin);
2582 * NOT MPSAFE - I think these refer to a common file descriptor table
2583 * and we need to spinlock that to link fdtol in.
2585 struct filedesc_to_leader *
2586 filedesc_to_leader_alloc(struct filedesc_to_leader *old,
2587 struct proc *leader)
2589 struct filedesc_to_leader *fdtol;
2591 fdtol = kmalloc(sizeof(struct filedesc_to_leader),
2592 M_FILEDESC_TO_LEADER, M_WAITOK);
2593 fdtol->fdl_refcount = 1;
2594 fdtol->fdl_holdcount = 0;
2595 fdtol->fdl_wakeup = 0;
2596 fdtol->fdl_leader = leader;
2598 fdtol->fdl_next = old->fdl_next;
2599 fdtol->fdl_prev = old;
2600 old->fdl_next = fdtol;
2601 fdtol->fdl_next->fdl_prev = fdtol;
2603 fdtol->fdl_next = fdtol;
2604 fdtol->fdl_prev = fdtol;
2610 * Scan all file pointers in the system. The callback is made with
2611 * the master list spinlock held exclusively.
2616 allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data)
2621 spin_lock_wr(&filehead_spin);
2622 LIST_FOREACH(fp, &filehead, f_list) {
2623 res = callback(fp, data);
2627 spin_unlock_wr(&filehead_spin);
2631 * Get file structures.
2633 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe)
2636 struct sysctl_kern_file_info {
2639 struct sysctl_req *req;
2642 static int sysctl_kern_file_callback(struct proc *p, void *data);
2645 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2647 struct sysctl_kern_file_info info;
2650 * Note: because the number of file descriptors is calculated
2651 * in different ways for sizing vs returning the data,
2652 * there is information leakage from the first loop. However,
2653 * it is of a similar order of magnitude to the leakage from
2654 * global system statistics such as kern.openfiles.
2656 * When just doing a count, note that we cannot just count
2657 * the elements and add f_count via the filehead list because
2658 * threaded processes share their descriptor table and f_count might
2659 * still be '1' in that case.
2661 * Since the SYSCTL op can block, we must hold the process to
2662 * prevent it being ripped out from under us either in the
2663 * file descriptor loop or in the greater LIST_FOREACH. The
2664 * process may be in varying states of disrepair. If the process
2665 * is in SZOMB we may have caught it just as it is being removed
2666 * from the allproc list, we must skip it in that case to maintain
2667 * an unbroken chain through the allproc list.
2672 allproc_scan(sysctl_kern_file_callback, &info);
2675 * When just calculating the size, overestimate a bit to try to
2676 * prevent system activity from causing the buffer-fill call
2679 if (req->oldptr == NULL) {
2680 info.count = (info.count + 16) + (info.count / 10);
2681 info.error = SYSCTL_OUT(req, NULL,
2682 info.count * sizeof(struct kinfo_file));
2684 return (info.error);
2688 sysctl_kern_file_callback(struct proc *p, void *data)
2690 struct sysctl_kern_file_info *info = data;
2691 struct kinfo_file kf;
2692 struct filedesc *fdp;
2697 if (p->p_stat == SIDL || p->p_stat == SZOMB)
2699 if (!PRISON_CHECK(info->req->td->td_ucred, p->p_ucred) != 0)
2703 * Softref the fdp to prevent it from being destroyed
2705 spin_lock_wr(&p->p_spin);
2706 if ((fdp = p->p_fd) == NULL) {
2707 spin_unlock_wr(&p->p_spin);
2710 atomic_add_int(&fdp->fd_softrefs, 1);
2711 spin_unlock_wr(&p->p_spin);
2714 * The fdp's own spinlock prevents the contents from being
2717 spin_lock_rd(&fdp->fd_spin);
2718 for (n = 0; n < fdp->fd_nfiles; ++n) {
2719 if ((fp = fdp->fd_files[n].fp) == NULL)
2721 if (info->req->oldptr == NULL) {
2724 uid = p->p_ucred ? p->p_ucred->cr_uid : -1;
2725 kcore_make_file(&kf, fp, p->p_pid, uid, n);
2726 spin_unlock_rd(&fdp->fd_spin);
2727 info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf));
2728 spin_lock_rd(&fdp->fd_spin);
2733 spin_unlock_rd(&fdp->fd_spin);
2734 atomic_subtract_int(&fdp->fd_softrefs, 1);
2740 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2741 0, 0, sysctl_kern_file, "S,file", "Entire file table");
2743 SYSCTL_INT(_kern, OID_AUTO, minfilesperproc, CTLFLAG_RW,
2744 &minfilesperproc, 0, "Minimum files allowed open per process");
2745 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2746 &maxfilesperproc, 0, "Maximum files allowed open per process");
2747 SYSCTL_INT(_kern, OID_AUTO, maxfilesperuser, CTLFLAG_RW,
2748 &maxfilesperuser, 0, "Maximum files allowed open per user");
2750 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2751 &maxfiles, 0, "Maximum number of files");
2753 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW,
2754 &maxfilesrootres, 0, "Descriptors reserved for root use");
2756 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2757 &nfiles, 0, "System-wide number of open files");
2760 fildesc_drvinit(void *unused)
2764 for (fd = 0; fd < NUMFDESC; fd++) {
2765 make_dev(&fildesc_ops, fd,
2766 UID_BIN, GID_BIN, 0666, "fd/%d", fd);
2769 make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin");
2770 make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout");
2771 make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr");
2777 struct fileops badfileops = {
2778 .fo_read = badfo_readwrite,
2779 .fo_write = badfo_readwrite,
2780 .fo_ioctl = badfo_ioctl,
2781 .fo_poll = badfo_poll,
2782 .fo_kqfilter = badfo_kqfilter,
2783 .fo_stat = badfo_stat,
2784 .fo_close = badfo_close,
2785 .fo_shutdown = badfo_shutdown
2799 badfo_ioctl(struct file *fp, u_long com, caddr_t data,
2800 struct ucred *cred, struct sysmsg *msgv)
2806 badfo_poll(struct file *fp, int events, struct ucred *cred)
2812 * Must return an error to prevent registration, typically
2813 * due to a revoked descriptor (file_filtops assigned).
2816 badfo_kqfilter(struct file *fp, struct knote *kn)
2818 return (EOPNOTSUPP);
2825 badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
2834 badfo_close(struct file *fp)
2843 badfo_shutdown(struct file *fp, int how)
2852 nofo_shutdown(struct file *fp, int how)
2854 return (EOPNOTSUPP);
2857 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,
2858 fildesc_drvinit,NULL)