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>
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>
91 #include <sys/filio.h>
92 #include <sys/fcntl.h>
93 #include <sys/unistd.h>
94 #include <sys/resourcevar.h>
95 #include <sys/event.h>
96 #include <sys/kern_syscall.h>
97 #include <sys/kcore.h>
98 #include <sys/kinfo.h>
102 #include <vm/vm_extern.h>
104 #include <sys/thread2.h>
105 #include <sys/file2.h>
106 #include <sys/spinlock2.h>
108 static void fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd);
109 static void fdreserve_locked (struct filedesc *fdp, int fd0, int incr);
110 static struct file *funsetfd_locked (struct filedesc *fdp, int fd);
111 static int checkfpclosed(struct filedesc *fdp, int fd, struct file *fp);
112 static void ffree(struct file *fp);
114 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
115 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
116 "file desc to leader structures");
117 MALLOC_DEFINE(M_FILE, "file", "Open file structure");
118 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
120 static d_open_t fdopen;
123 #define CDEV_MAJOR 22
124 static struct dev_ops fildesc_ops = {
125 { "FD", CDEV_MAJOR, 0 },
129 static int badfo_readwrite (struct file *fp, struct uio *uio,
130 struct ucred *cred, int flags);
131 static int badfo_ioctl (struct file *fp, u_long com, caddr_t data,
133 static int badfo_poll (struct file *fp, int events, struct ucred *cred);
134 static int badfo_kqfilter (struct file *fp, struct knote *kn);
135 static int badfo_stat (struct file *fp, struct stat *sb, struct ucred *cred);
136 static int badfo_close (struct file *fp);
137 static int badfo_shutdown (struct file *fp, int how);
140 * Descriptor management.
142 static struct filelist filehead = LIST_HEAD_INITIALIZER(&filehead);
143 static struct spinlock filehead_spin = SPINLOCK_INITIALIZER(&filehead_spin);
144 static int nfiles; /* actual number of open files */
148 * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared.
150 * MPSAFE - must be called with fdp->fd_spin exclusively held
154 fdfixup_locked(struct filedesc *fdp, int fd)
156 if (fd < fdp->fd_freefile) {
157 fdp->fd_freefile = fd;
159 while (fdp->fd_lastfile >= 0 &&
160 fdp->fd_files[fdp->fd_lastfile].fp == NULL &&
161 fdp->fd_files[fdp->fd_lastfile].reserved == 0
168 * System calls on descriptors.
173 sys_getdtablesize(struct getdtablesize_args *uap)
175 struct proc *p = curproc;
176 struct plimit *limit = p->p_limit;
178 spin_lock_rd(&limit->p_spin);
180 min((int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
181 spin_unlock_rd(&limit->p_spin);
186 * Duplicate a file descriptor to a particular value.
188 * note: keep in mind that a potential race condition exists when closing
189 * descriptors from a shared descriptor table (via rfork).
194 sys_dup2(struct dup2_args *uap)
198 error = kern_dup(DUP_FIXED, uap->from, uap->to, uap->sysmsg_fds);
204 * Duplicate a file descriptor.
209 sys_dup(struct dup_args *uap)
213 error = kern_dup(DUP_VARIABLE, uap->fd, 0, uap->sysmsg_fds);
219 * MPALMOSTSAFE - acquires mplock for fp operations
222 kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred)
224 struct thread *td = curthread;
225 struct proc *p = td->td_proc;
230 int tmp, error, flg = F_POSIX;
235 * Operations on file descriptors that do not require a file pointer.
239 error = fgetfdflags(p->p_fd, fd, &tmp);
241 dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0;
245 if (dat->fc_cloexec & FD_CLOEXEC)
246 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
248 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
252 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd);
259 * Operations on file pointers
261 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
267 dat->fc_flags = OFLAGS(fp->f_flag);
272 oflags = fp->f_flag & FCNTLFLAGS;
273 fp->f_flag &= ~FCNTLFLAGS;
274 fp->f_flag |= FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS;
276 if ((fp->f_flag ^ oflags) & FASYNC) {
277 tmp = fp->f_flag & FASYNC;
278 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, cred);
281 fp->f_flag = (fp->f_flag & ~FCNTLFLAGS) | oflags;
285 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner, cred);
289 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner, cred);
294 /* Fall into F_SETLK */
297 if (fp->f_type != DTYPE_VNODE) {
301 vp = (struct vnode *)fp->f_data;
304 * copyin/lockop may block
306 if (dat->fc_flock.l_whence == SEEK_CUR)
307 dat->fc_flock.l_start += fp->f_offset;
309 switch (dat->fc_flock.l_type) {
311 if ((fp->f_flag & FREAD) == 0) {
315 p->p_leader->p_flag |= P_ADVLOCK;
316 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
317 &dat->fc_flock, flg);
320 if ((fp->f_flag & FWRITE) == 0) {
324 p->p_leader->p_flag |= P_ADVLOCK;
325 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
326 &dat->fc_flock, flg);
329 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
330 &dat->fc_flock, F_POSIX);
338 * It is possible to race a close() on the descriptor while
339 * we were blocked getting the lock. If this occurs the
340 * close might not have caught the lock.
342 if (checkfpclosed(p->p_fd, fd, fp)) {
343 dat->fc_flock.l_whence = SEEK_SET;
344 dat->fc_flock.l_start = 0;
345 dat->fc_flock.l_len = 0;
346 dat->fc_flock.l_type = F_UNLCK;
347 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
348 F_UNLCK, &dat->fc_flock, F_POSIX);
353 if (fp->f_type != DTYPE_VNODE) {
357 vp = (struct vnode *)fp->f_data;
359 * copyin/lockop may block
361 if (dat->fc_flock.l_type != F_RDLCK &&
362 dat->fc_flock.l_type != F_WRLCK &&
363 dat->fc_flock.l_type != F_UNLCK) {
367 if (dat->fc_flock.l_whence == SEEK_CUR)
368 dat->fc_flock.l_start += fp->f_offset;
369 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK,
370 &dat->fc_flock, F_POSIX);
383 * The file control system call.
388 sys_fcntl(struct fcntl_args *uap)
395 dat.fc_fd = uap->arg;
398 dat.fc_cloexec = uap->arg;
401 dat.fc_flags = uap->arg;
404 dat.fc_owner = uap->arg;
409 error = copyin((caddr_t)uap->arg, &dat.fc_flock,
410 sizeof(struct flock));
416 error = kern_fcntl(uap->fd, uap->cmd, &dat, curproc->p_ucred);
421 uap->sysmsg_result = dat.fc_fd;
424 uap->sysmsg_result = dat.fc_cloexec;
427 uap->sysmsg_result = dat.fc_flags;
430 uap->sysmsg_result = dat.fc_owner;
432 error = copyout(&dat.fc_flock, (caddr_t)uap->arg,
433 sizeof(struct flock));
442 * Common code for dup, dup2, and fcntl(F_DUPFD).
444 * The type flag can be either DUP_FIXED or DUP_VARIABLE. DUP_FIXED tells
445 * kern_dup() to destructively dup over an existing file descriptor if new
446 * is already open. DUP_VARIABLE tells kern_dup() to find the lowest
447 * unused file descriptor that is greater than or equal to new.
452 kern_dup(enum dup_type type, int old, int new, int *res)
454 struct thread *td = curthread;
455 struct proc *p = td->td_proc;
456 struct filedesc *fdp = p->p_fd;
464 * Verify that we have a valid descriptor to dup from and
465 * possibly to dup to.
468 spin_lock_wr(&fdp->fd_spin);
469 if (new < 0 || new > p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
470 new >= maxfilesperproc) {
471 spin_unlock_wr(&fdp->fd_spin);
474 if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) {
475 spin_unlock_wr(&fdp->fd_spin);
478 if (type == DUP_FIXED && old == new) {
480 spin_unlock_wr(&fdp->fd_spin);
483 fp = fdp->fd_files[old].fp;
484 oldflags = fdp->fd_files[old].fileflags;
485 fhold(fp); /* MPSAFE - can be called with a spinlock held */
488 * Allocate a new descriptor if DUP_VARIABLE, or expand the table
489 * if the requested descriptor is beyond the current table size.
491 * This can block. Retry if the source descriptor no longer matches
492 * or if our expectation in the expansion case races.
494 * If we are not expanding or allocating a new decriptor, then reset
495 * the target descriptor to a reserved state so we have a uniform
496 * setup for the next code block.
498 if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) {
499 spin_unlock_wr(&fdp->fd_spin);
500 error = fdalloc(p, new, &newfd);
501 spin_lock_wr(&fdp->fd_spin);
503 spin_unlock_wr(&fdp->fd_spin);
510 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) {
511 fsetfd_locked(fdp, NULL, newfd);
512 spin_unlock_wr(&fdp->fd_spin);
517 * Check for expansion race
519 if (type != DUP_VARIABLE && new != newfd) {
520 fsetfd_locked(fdp, NULL, newfd);
521 spin_unlock_wr(&fdp->fd_spin);
526 * Check for ripout, newfd reused old (this case probably
530 fsetfd_locked(fdp, NULL, newfd);
531 spin_unlock_wr(&fdp->fd_spin);
538 if (fdp->fd_files[new].reserved) {
539 spin_unlock_wr(&fdp->fd_spin);
541 kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new);
542 tsleep(fdp, 0, "fdres", hz);
547 * If the target descriptor was never allocated we have
548 * to allocate it. If it was we have to clean out the
549 * old descriptor. delfp inherits the ref from the
552 delfp = fdp->fd_files[new].fp;
553 fdp->fd_files[new].fp = NULL;
554 fdp->fd_files[new].reserved = 1;
556 fdreserve_locked(fdp, new, 1);
557 if (new > fdp->fd_lastfile)
558 fdp->fd_lastfile = new;
564 * NOTE: still holding an exclusive spinlock
568 * If a descriptor is being overwritten we may hve to tell
569 * fdfree() to sleep to ensure that all relevant process
570 * leaders can be traversed in closef().
572 if (delfp != NULL && p->p_fdtol != NULL) {
573 fdp->fd_holdleaderscount++;
578 KASSERT(delfp == NULL || type == DUP_FIXED,
579 ("dup() picked an open file"));
582 * Duplicate the source descriptor, update lastfile. If the new
583 * descriptor was not allocated and we aren't replacing an existing
584 * descriptor we have to mark the descriptor as being in use.
586 * The fd_files[] array inherits fp's hold reference.
588 fsetfd_locked(fdp, fp, new);
589 fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE;
590 spin_unlock_wr(&fdp->fd_spin);
595 * If we dup'd over a valid file, we now own the reference to it
596 * and must dispose of it using closef() semantics (as if a
597 * close() were performed on it).
602 spin_lock_wr(&fdp->fd_spin);
603 fdp->fd_holdleaderscount--;
604 if (fdp->fd_holdleaderscount == 0 &&
605 fdp->fd_holdleaderswakeup != 0) {
606 fdp->fd_holdleaderswakeup = 0;
607 spin_unlock_wr(&fdp->fd_spin);
608 wakeup(&fdp->fd_holdleaderscount);
610 spin_unlock_wr(&fdp->fd_spin);
618 * If sigio is on the list associated with a process or process group,
619 * disable signalling from the device, remove sigio from the list and
623 funsetown(struct sigio *sigio)
628 *(sigio->sio_myref) = NULL;
630 if (sigio->sio_pgid < 0) {
631 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
633 } else /* if ((*sigiop)->sio_pgid > 0) */ {
634 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
637 crfree(sigio->sio_ucred);
638 kfree(sigio, M_SIGIO);
641 /* Free a list of sigio structures. */
643 funsetownlst(struct sigiolst *sigiolst)
647 while ((sigio = SLIST_FIRST(sigiolst)) != NULL)
652 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
654 * After permission checking, add a sigio structure to the sigio list for
655 * the process or process group.
658 fsetown(pid_t pgid, struct sigio **sigiop)
674 * Policy - Don't allow a process to FSETOWN a process
675 * in another session.
677 * Remove this test to allow maximum flexibility or
678 * restrict FSETOWN to the current process or process
679 * group for maximum safety.
681 if (proc->p_session != curproc->p_session)
685 } else /* if (pgid < 0) */ {
686 pgrp = pgfind(-pgid);
691 * Policy - Don't allow a process to FSETOWN a process
692 * in another session.
694 * Remove this test to allow maximum flexibility or
695 * restrict FSETOWN to the current process or process
696 * group for maximum safety.
698 if (pgrp->pg_session != curproc->p_session)
704 sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
706 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
707 sigio->sio_proc = proc;
709 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
710 sigio->sio_pgrp = pgrp;
712 sigio->sio_pgid = pgid;
713 sigio->sio_ucred = crhold(curproc->p_ucred);
714 /* It would be convenient if p_ruid was in ucred. */
715 sigio->sio_ruid = curproc->p_ucred->cr_ruid;
716 sigio->sio_myref = sigiop;
724 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
727 fgetown(struct sigio *sigio)
729 return (sigio != NULL ? sigio->sio_pgid : 0);
733 * Close many file descriptors.
738 sys_closefrom(struct closefrom_args *uap)
740 return(kern_closefrom(uap->fd));
744 * Close all file descriptors greater then or equal to fd
749 kern_closefrom(int fd)
751 struct thread *td = curthread;
752 struct proc *p = td->td_proc;
753 struct filedesc *fdp;
762 * NOTE: This function will skip unassociated descriptors and
763 * reserved descriptors that have not yet been assigned.
764 * fd_lastfile can change as a side effect of kern_close().
766 spin_lock_wr(&fdp->fd_spin);
767 while (fd <= fdp->fd_lastfile) {
768 if (fdp->fd_files[fd].fp != NULL) {
769 spin_unlock_wr(&fdp->fd_spin);
770 /* ok if this races another close */
771 if (kern_close(fd) == EINTR)
773 spin_lock_wr(&fdp->fd_spin);
777 spin_unlock_wr(&fdp->fd_spin);
782 * Close a file descriptor.
787 sys_close(struct close_args *uap)
789 return(kern_close(uap->fd));
793 * MPALMOSTSAFE - acquires mplock around knote_fdclose() calls
798 struct thread *td = curthread;
799 struct proc *p = td->td_proc;
800 struct filedesc *fdp;
808 spin_lock_wr(&fdp->fd_spin);
809 if ((fp = funsetfd_locked(fdp, fd)) == NULL) {
810 spin_unlock_wr(&fdp->fd_spin);
814 if (p->p_fdtol != NULL) {
816 * Ask fdfree() to sleep to ensure that all relevant
817 * process leaders can be traversed in closef().
819 fdp->fd_holdleaderscount++;
824 * we now hold the fp reference that used to be owned by the descriptor
827 spin_unlock_wr(&fdp->fd_spin);
828 if (fd < fdp->fd_knlistsize) {
830 if (fd < fdp->fd_knlistsize)
831 knote_fdclose(p, fd);
834 error = closef(fp, p);
836 spin_lock_wr(&fdp->fd_spin);
837 fdp->fd_holdleaderscount--;
838 if (fdp->fd_holdleaderscount == 0 &&
839 fdp->fd_holdleaderswakeup != 0) {
840 fdp->fd_holdleaderswakeup = 0;
841 spin_unlock_wr(&fdp->fd_spin);
842 wakeup(&fdp->fd_holdleaderscount);
844 spin_unlock_wr(&fdp->fd_spin);
851 * shutdown_args(int fd, int how)
854 kern_shutdown(int fd, int how)
856 struct thread *td = curthread;
857 struct proc *p = td->td_proc;
863 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
865 error = fo_shutdown(fp, how);
872 sys_shutdown(struct shutdown_args *uap)
876 error = kern_shutdown(uap->s, uap->how);
882 kern_fstat(int fd, struct stat *ub)
884 struct thread *td = curthread;
885 struct proc *p = td->td_proc;
891 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
893 error = fo_stat(fp, ub, p->p_ucred);
900 * Return status information about a file descriptor.
903 sys_fstat(struct fstat_args *uap)
908 error = kern_fstat(uap->fd, &st);
911 error = copyout(&st, uap->sb, sizeof(st));
916 * Return pathconf information about a file descriptor.
920 sys_fpathconf(struct fpathconf_args *uap)
922 struct thread *td = curthread;
923 struct proc *p = td->td_proc;
930 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
933 switch (fp->f_type) {
936 if (uap->name != _PC_PIPE_BUF) {
939 uap->sysmsg_result = PIPE_BUF;
945 vp = (struct vnode *)fp->f_data;
946 error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds);
957 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0, "");
960 * Grow the file table so it can hold through descriptor (want).
962 * The fdp's spinlock must be held exclusively on entry and may be held
963 * exclusively on return. The spinlock may be cycled by the routine.
968 fdgrow_locked(struct filedesc *fdp, int want)
970 struct fdnode *newfiles;
971 struct fdnode *oldfiles;
976 /* nf has to be of the form 2^n - 1 */
978 } while (nf <= want);
980 spin_unlock_wr(&fdp->fd_spin);
981 newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK);
982 spin_lock_wr(&fdp->fd_spin);
985 * We could have raced another extend while we were not holding
988 if (fdp->fd_nfiles >= nf) {
989 spin_unlock_wr(&fdp->fd_spin);
990 kfree(newfiles, M_FILEDESC);
991 spin_lock_wr(&fdp->fd_spin);
995 * Copy the existing ofile and ofileflags arrays
996 * and zero the new portion of each array.
998 extra = nf - fdp->fd_nfiles;
999 bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode));
1000 bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode));
1002 oldfiles = fdp->fd_files;
1003 fdp->fd_files = newfiles;
1004 fdp->fd_nfiles = nf;
1006 if (oldfiles != fdp->fd_builtin_files) {
1007 spin_unlock_wr(&fdp->fd_spin);
1008 kfree(oldfiles, M_FILEDESC);
1009 spin_lock_wr(&fdp->fd_spin);
1015 * Number of nodes in right subtree, including the root.
1018 right_subtree_size(int n)
1020 return (n ^ (n | (n + 1)));
1027 right_ancestor(int n)
1029 return (n | (n + 1));
1036 left_ancestor(int n)
1038 return ((n & (n + 1)) - 1);
1042 * Traverse the in-place binary tree buttom-up adjusting the allocation
1043 * count so scans can determine where free descriptors are located.
1045 * MPSAFE - caller must be holding an exclusive spinlock on fdp
1049 fdreserve_locked(struct filedesc *fdp, int fd, int incr)
1052 fdp->fd_files[fd].allocated += incr;
1053 KKASSERT(fdp->fd_files[fd].allocated >= 0);
1054 fd = left_ancestor(fd);
1059 * Reserve a file descriptor for the process. If no error occurs, the
1060 * caller MUST at some point call fsetfd() or assign a file pointer
1061 * or dispose of the reservation.
1066 fdalloc(struct proc *p, int want, int *result)
1068 struct filedesc *fdp = p->p_fd;
1069 int fd, rsize, rsum, node, lim;
1071 spin_lock_rd(&p->p_limit->p_spin);
1072 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
1073 spin_unlock_rd(&p->p_limit->p_spin);
1076 spin_lock_wr(&fdp->fd_spin);
1077 if (want >= fdp->fd_nfiles)
1078 fdgrow_locked(fdp, want);
1081 * Search for a free descriptor starting at the higher
1082 * of want or fd_freefile. If that fails, consider
1083 * expanding the ofile array.
1085 * NOTE! the 'allocated' field is a cumulative recursive allocation
1086 * count. If we happen to see a value of 0 then we can shortcut
1087 * our search. Otherwise we run through through the tree going
1088 * down branches we know have free descriptor(s) until we hit a
1089 * leaf node. The leaf node will be free but will not necessarily
1090 * have an allocated field of 0.
1093 /* move up the tree looking for a subtree with a free node */
1094 for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim);
1095 fd = right_ancestor(fd)) {
1096 if (fdp->fd_files[fd].allocated == 0)
1099 rsize = right_subtree_size(fd);
1100 if (fdp->fd_files[fd].allocated == rsize)
1101 continue; /* right subtree full */
1104 * Free fd is in the right subtree of the tree rooted at fd.
1105 * Call that subtree R. Look for the smallest (leftmost)
1106 * subtree of R with an unallocated fd: continue moving
1107 * down the left branch until encountering a full left
1108 * subtree, then move to the right.
1110 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) {
1112 rsum += fdp->fd_files[node].allocated;
1113 if (fdp->fd_files[fd].allocated == rsum + rsize) {
1114 fd = node; /* move to the right */
1115 if (fdp->fd_files[node].allocated == 0)
1124 * No space in current array. Expand?
1126 if (fdp->fd_nfiles >= lim) {
1127 spin_unlock_wr(&fdp->fd_spin);
1130 fdgrow_locked(fdp, want);
1134 KKASSERT(fd < fdp->fd_nfiles);
1135 if (fd > fdp->fd_lastfile)
1136 fdp->fd_lastfile = fd;
1137 if (want <= fdp->fd_freefile)
1138 fdp->fd_freefile = fd;
1140 KKASSERT(fdp->fd_files[fd].fp == NULL);
1141 KKASSERT(fdp->fd_files[fd].reserved == 0);
1142 fdp->fd_files[fd].fileflags = 0;
1143 fdp->fd_files[fd].reserved = 1;
1144 fdreserve_locked(fdp, fd, 1);
1145 spin_unlock_wr(&fdp->fd_spin);
1150 * Check to see whether n user file descriptors
1151 * are available to the process p.
1156 fdavail(struct proc *p, int n)
1158 struct filedesc *fdp = p->p_fd;
1159 struct fdnode *fdnode;
1162 spin_lock_rd(&p->p_limit->p_spin);
1163 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
1164 spin_unlock_rd(&p->p_limit->p_spin);
1166 spin_lock_rd(&fdp->fd_spin);
1167 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) {
1168 spin_unlock_rd(&fdp->fd_spin);
1171 last = min(fdp->fd_nfiles, lim);
1172 fdnode = &fdp->fd_files[fdp->fd_freefile];
1173 for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) {
1174 if (fdnode->fp == NULL && --n <= 0) {
1175 spin_unlock_rd(&fdp->fd_spin);
1179 spin_unlock_rd(&fdp->fd_spin);
1184 * Revoke open descriptors referencing (f_data, f_type)
1186 * Any revoke executed within a prison is only able to
1187 * revoke descriptors for processes within that prison.
1189 * Returns 0 on success or an error code.
1191 struct fdrevoke_info {
1201 static int fdrevoke_check_callback(struct file *fp, void *vinfo);
1202 static int fdrevoke_proc_callback(struct proc *p, void *vinfo);
1205 fdrevoke(void *f_data, short f_type, struct ucred *cred)
1207 struct fdrevoke_info info;
1210 bzero(&info, sizeof(info));
1214 error = falloc(NULL, &info.nfp, NULL);
1219 * Scan the file pointer table once. dups do not dup file pointers,
1220 * only descriptors, so there is no leak. Set FREVOKED on the fps
1223 allfiles_scan_exclusive(fdrevoke_check_callback, &info);
1226 * If any fps were marked track down the related descriptors
1227 * and close them. Any dup()s at this point will notice
1228 * the FREVOKED already set in the fp and do the right thing.
1230 * Any fps with non-zero msgcounts (aka sent over a unix-domain
1231 * socket) bumped the intransit counter and will require a
1232 * scan. Races against fps leaving the socket are closed by
1233 * the socket code checking for FREVOKED.
1236 allproc_scan(fdrevoke_proc_callback, &info);
1238 unp_revoke_gc(info.nfp);
1244 * Locate matching file pointers directly.
1247 fdrevoke_check_callback(struct file *fp, void *vinfo)
1249 struct fdrevoke_info *info = vinfo;
1252 * File pointers already flagged for revokation are skipped.
1254 if (fp->f_flag & FREVOKED)
1258 * If revoking from a prison file pointers created outside of
1259 * that prison, or file pointers without creds, cannot be revoked.
1261 if (info->cred->cr_prison &&
1262 (fp->f_cred == NULL ||
1263 info->cred->cr_prison != fp->f_cred->cr_prison)) {
1268 * If the file pointer matches then mark it for revocation. The
1269 * flag is currently only used by unp_revoke_gc().
1271 * info->count is a heuristic and can race in a SMP environment.
1273 if (info->data == fp->f_data && info->type == fp->f_type) {
1274 atomic_set_int(&fp->f_flag, FREVOKED);
1275 info->count += fp->f_count;
1283 * Locate matching file pointers via process descriptor tables.
1286 fdrevoke_proc_callback(struct proc *p, void *vinfo)
1288 struct fdrevoke_info *info = vinfo;
1289 struct filedesc *fdp;
1293 if (p->p_stat == SIDL || p->p_stat == SZOMB)
1295 if (info->cred->cr_prison &&
1296 info->cred->cr_prison != p->p_ucred->cr_prison) {
1301 * If the controlling terminal of the process matches the
1302 * vnode being revoked we clear the controlling terminal.
1304 * The normal spec_close() may not catch this because it
1305 * uses curproc instead of p.
1307 if (p->p_session && info->type == DTYPE_VNODE &&
1308 info->data == p->p_session->s_ttyvp) {
1309 p->p_session->s_ttyvp = NULL;
1314 * Locate and close any matching file descriptors.
1316 if ((fdp = p->p_fd) == NULL)
1318 spin_lock_wr(&fdp->fd_spin);
1319 for (n = 0; n < fdp->fd_nfiles; ++n) {
1320 if ((fp = fdp->fd_files[n].fp) == NULL)
1322 if (fp->f_flag & FREVOKED) {
1324 fdp->fd_files[n].fp = info->nfp;
1325 spin_unlock_wr(&fdp->fd_spin);
1327 spin_lock_wr(&fdp->fd_spin);
1331 spin_unlock_wr(&fdp->fd_spin);
1337 * Create a new open file structure and reserve a file decriptor
1338 * for the process that refers to it.
1340 * Root creds are checked using p, or assumed if p is NULL. If
1341 * resultfd is non-NULL then p must also be non-NULL. No file
1342 * descriptor is reserved if resultfd is NULL.
1344 * A file pointer with a refcount of 1 is returned. Note that the
1345 * file pointer is NOT associated with the descriptor. If falloc
1346 * returns success, fsetfd() MUST be called to either associate the
1347 * file pointer or clear the reservation.
1352 falloc(struct proc *p, struct file **resultfp, int *resultfd)
1354 static struct timeval lastfail;
1362 * Handle filetable full issues and root overfill.
1364 if (nfiles >= maxfiles - maxfilesrootres &&
1365 ((p && p->p_ucred->cr_ruid != 0) || nfiles >= maxfiles)) {
1366 if (ppsratecheck(&lastfail, &curfail, 1)) {
1367 kprintf("kern.maxfiles limit exceeded by uid %d, please see tuning(7).\n",
1368 (p ? p->p_ucred->cr_ruid : -1));
1375 * Allocate a new file descriptor.
1377 fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO);
1378 spin_init(&fp->f_spin);
1380 fp->f_ops = &badfileops;
1383 fp->f_cred = crhold(p->p_ucred);
1385 fp->f_cred = crhold(proc0.p_ucred);
1386 spin_lock_wr(&filehead_spin);
1388 LIST_INSERT_HEAD(&filehead, fp, f_list);
1389 spin_unlock_wr(&filehead_spin);
1391 if ((error = fdalloc(p, 0, resultfd)) != 0) {
1408 checkfpclosed(struct filedesc *fdp, int fd, struct file *fp)
1412 spin_lock_rd(&fdp->fd_spin);
1413 if ((unsigned) fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp)
1417 spin_unlock_rd(&fdp->fd_spin);
1422 * Associate a file pointer with a previously reserved file descriptor.
1423 * This function always succeeds.
1425 * If fp is NULL, the file descriptor is returned to the pool.
1429 * MPSAFE (exclusive spinlock must be held on call)
1432 fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd)
1434 KKASSERT((unsigned)fd < fdp->fd_nfiles);
1435 KKASSERT(fdp->fd_files[fd].reserved != 0);
1438 fdp->fd_files[fd].fp = fp;
1439 fdp->fd_files[fd].reserved = 0;
1440 if (fp->f_type == DTYPE_KQUEUE) {
1441 if (fdp->fd_knlistsize < 0)
1442 fdp->fd_knlistsize = 0;
1445 fdp->fd_files[fd].reserved = 0;
1446 fdreserve_locked(fdp, fd, -1);
1447 fdfixup_locked(fdp, fd);
1455 fsetfd(struct proc *p, struct file *fp, int fd)
1457 struct filedesc *fdp = p->p_fd;
1459 spin_lock_wr(&fdp->fd_spin);
1460 fsetfd_locked(fdp, fp, fd);
1461 spin_unlock_wr(&fdp->fd_spin);
1465 * MPSAFE (exclusive spinlock must be held on call)
1469 funsetfd_locked(struct filedesc *fdp, int fd)
1473 if ((unsigned)fd >= fdp->fd_nfiles)
1475 if ((fp = fdp->fd_files[fd].fp) == NULL)
1477 fdp->fd_files[fd].fp = NULL;
1478 fdp->fd_files[fd].fileflags = 0;
1480 fdreserve_locked(fdp, fd, -1);
1481 fdfixup_locked(fdp, fd);
1489 fgetfdflags(struct filedesc *fdp, int fd, int *flagsp)
1493 spin_lock_rd(&fdp->fd_spin);
1494 if (((u_int)fd) >= fdp->fd_nfiles) {
1496 } else if (fdp->fd_files[fd].fp == NULL) {
1499 *flagsp = fdp->fd_files[fd].fileflags;
1502 spin_unlock_rd(&fdp->fd_spin);
1510 fsetfdflags(struct filedesc *fdp, int fd, int add_flags)
1514 spin_lock_wr(&fdp->fd_spin);
1515 if (((u_int)fd) >= fdp->fd_nfiles) {
1517 } else if (fdp->fd_files[fd].fp == NULL) {
1520 fdp->fd_files[fd].fileflags |= add_flags;
1523 spin_unlock_wr(&fdp->fd_spin);
1531 fclrfdflags(struct filedesc *fdp, int fd, int rem_flags)
1535 spin_lock_wr(&fdp->fd_spin);
1536 if (((u_int)fd) >= fdp->fd_nfiles) {
1538 } else if (fdp->fd_files[fd].fp == NULL) {
1541 fdp->fd_files[fd].fileflags &= ~rem_flags;
1544 spin_unlock_wr(&fdp->fd_spin);
1549 fsetcred(struct file *fp, struct ucred *cr)
1557 * Free a file descriptor.
1561 ffree(struct file *fp)
1563 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!"));
1564 spin_lock_wr(&filehead_spin);
1565 LIST_REMOVE(fp, f_list);
1567 spin_unlock_wr(&filehead_spin);
1569 if (fp->f_nchandle.ncp)
1570 cache_drop(&fp->f_nchandle);
1575 * called from init_main, initialize filedesc0 for proc0.
1578 fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask)
1582 fdp0->fd_refcnt = 1;
1583 fdp0->fd_cmask = cmask;
1584 fdp0->fd_files = fdp0->fd_builtin_files;
1585 fdp0->fd_nfiles = NDFILE;
1586 fdp0->fd_lastfile = -1;
1587 spin_init(&fdp0->fd_spin);
1591 * Build a new filedesc structure.
1596 fdinit(struct proc *p)
1598 struct filedesc *newfdp;
1599 struct filedesc *fdp = p->p_fd;
1601 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO);
1602 spin_lock_rd(&fdp->fd_spin);
1604 newfdp->fd_cdir = fdp->fd_cdir;
1605 vref(newfdp->fd_cdir);
1606 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1610 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of
1611 * proc0, but should unconditionally exist in other processes.
1614 newfdp->fd_rdir = fdp->fd_rdir;
1615 vref(newfdp->fd_rdir);
1616 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1619 newfdp->fd_jdir = fdp->fd_jdir;
1620 vref(newfdp->fd_jdir);
1621 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1623 spin_unlock_rd(&fdp->fd_spin);
1625 /* Create the file descriptor table. */
1626 newfdp->fd_refcnt = 1;
1627 newfdp->fd_cmask = cmask;
1628 newfdp->fd_files = newfdp->fd_builtin_files;
1629 newfdp->fd_nfiles = NDFILE;
1630 newfdp->fd_knlistsize = -1;
1631 newfdp->fd_lastfile = -1;
1632 spin_init(&newfdp->fd_spin);
1638 * Share a filedesc structure.
1643 fdshare(struct proc *p)
1645 struct filedesc *fdp;
1648 spin_lock_wr(&fdp->fd_spin);
1650 spin_unlock_wr(&fdp->fd_spin);
1655 * Copy a filedesc structure.
1660 fdcopy(struct proc *p)
1662 struct filedesc *fdp = p->p_fd;
1663 struct filedesc *newfdp;
1664 struct fdnode *fdnode;
1669 * Certain daemons might not have file descriptors.
1675 * Allocate the new filedesc and fd_files[] array. This can race
1676 * with operations by other threads on the fdp so we have to be
1679 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK | M_ZERO);
1681 spin_lock_rd(&fdp->fd_spin);
1682 if (fdp->fd_lastfile < NDFILE) {
1683 newfdp->fd_files = newfdp->fd_builtin_files;
1687 * We have to allocate (N^2-1) entries for our in-place
1688 * binary tree. Allow the table to shrink.
1692 while (ni > fdp->fd_lastfile && ni > NDFILE) {
1696 spin_unlock_rd(&fdp->fd_spin);
1697 newfdp->fd_files = kmalloc(i * sizeof(struct fdnode),
1698 M_FILEDESC, M_WAITOK | M_ZERO);
1701 * Check for race, retry
1703 spin_lock_rd(&fdp->fd_spin);
1704 if (i <= fdp->fd_lastfile) {
1705 spin_unlock_rd(&fdp->fd_spin);
1706 kfree(newfdp->fd_files, M_FILEDESC);
1712 * Dup the remaining fields. vref() and cache_hold() can be
1713 * safely called while holding the read spinlock on fdp.
1715 * The read spinlock on fdp is still being held.
1717 * NOTE: vref and cache_hold calls for the case where the vnode
1718 * or cache entry already has at least one ref may be called
1719 * while holding spin locks.
1721 if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) {
1722 vref(newfdp->fd_cdir);
1723 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1726 * We must check for fd_rdir here, at least for now because
1727 * the init process is created before we have access to the
1728 * rootvode to take a reference to it.
1730 if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) {
1731 vref(newfdp->fd_rdir);
1732 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1734 if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) {
1735 vref(newfdp->fd_jdir);
1736 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1738 newfdp->fd_refcnt = 1;
1739 newfdp->fd_nfiles = i;
1740 newfdp->fd_lastfile = fdp->fd_lastfile;
1741 newfdp->fd_freefile = fdp->fd_freefile;
1742 newfdp->fd_cmask = fdp->fd_cmask;
1743 newfdp->fd_knlist = NULL;
1744 newfdp->fd_knlistsize = -1;
1745 newfdp->fd_knhash = NULL;
1746 newfdp->fd_knhashmask = 0;
1747 spin_init(&newfdp->fd_spin);
1750 * Copy the descriptor table through (i). This also copies the
1751 * allocation state. Then go through and ref the file pointers
1752 * and clean up any KQ descriptors.
1754 * kq descriptors cannot be copied. Since we haven't ref'd the
1755 * copied files yet we can ignore the return value from funsetfd().
1757 * The read spinlock on fdp is still being held.
1759 bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode));
1760 for (i = 0 ; i < newfdp->fd_nfiles; ++i) {
1761 fdnode = &newfdp->fd_files[i];
1762 if (fdnode->reserved) {
1763 fdreserve_locked(newfdp, i, -1);
1764 fdnode->reserved = 0;
1765 fdfixup_locked(newfdp, i);
1766 } else if (fdnode->fp) {
1767 if (fdnode->fp->f_type == DTYPE_KQUEUE) {
1768 (void)funsetfd_locked(newfdp, i);
1774 spin_unlock_rd(&fdp->fd_spin);
1779 * Release a filedesc structure.
1781 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE)
1784 fdfree(struct proc *p)
1786 struct filedesc *fdp = p->p_fd;
1787 struct fdnode *fdnode;
1789 struct filedesc_to_leader *fdtol;
1794 /* Certain daemons might not have file descriptors. */
1799 * Severe messing around to follow
1801 spin_lock_wr(&fdp->fd_spin);
1803 /* Check for special need to clear POSIX style locks */
1805 if (fdtol != NULL) {
1806 KASSERT(fdtol->fdl_refcount > 0,
1807 ("filedesc_to_refcount botch: fdl_refcount=%d",
1808 fdtol->fdl_refcount));
1809 if (fdtol->fdl_refcount == 1 &&
1810 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1811 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1812 fdnode = &fdp->fd_files[i];
1813 if (fdnode->fp == NULL ||
1814 fdnode->fp->f_type != DTYPE_VNODE) {
1819 spin_unlock_wr(&fdp->fd_spin);
1821 lf.l_whence = SEEK_SET;
1824 lf.l_type = F_UNLCK;
1825 vp = (struct vnode *)fp->f_data;
1826 (void) VOP_ADVLOCK(vp,
1827 (caddr_t)p->p_leader,
1832 spin_lock_wr(&fdp->fd_spin);
1836 if (fdtol->fdl_refcount == 1) {
1837 if (fdp->fd_holdleaderscount > 0 &&
1838 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1840 * close() or do_dup() has cleared a reference
1841 * in a shared file descriptor table.
1843 fdp->fd_holdleaderswakeup = 1;
1844 msleep(&fdp->fd_holdleaderscount,
1845 &fdp->fd_spin, 0, "fdlhold", 0);
1848 if (fdtol->fdl_holdcount > 0) {
1850 * Ensure that fdtol->fdl_leader
1851 * remains valid in closef().
1853 fdtol->fdl_wakeup = 1;
1854 msleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0);
1858 fdtol->fdl_refcount--;
1859 if (fdtol->fdl_refcount == 0 &&
1860 fdtol->fdl_holdcount == 0) {
1861 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1862 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1867 if (fdtol != NULL) {
1868 spin_unlock_wr(&fdp->fd_spin);
1869 kfree(fdtol, M_FILEDESC_TO_LEADER);
1870 spin_lock_wr(&fdp->fd_spin);
1873 if (--fdp->fd_refcnt > 0) {
1874 spin_unlock_wr(&fdp->fd_spin);
1877 spin_unlock_wr(&fdp->fd_spin);
1880 * we are the last reference to the structure, we can
1881 * safely assume it will not change out from under us.
1883 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1884 if (fdp->fd_files[i].fp)
1885 closef(fdp->fd_files[i].fp, p);
1887 if (fdp->fd_files != fdp->fd_builtin_files)
1888 kfree(fdp->fd_files, M_FILEDESC);
1890 cache_drop(&fdp->fd_ncdir);
1891 vrele(fdp->fd_cdir);
1894 cache_drop(&fdp->fd_nrdir);
1895 vrele(fdp->fd_rdir);
1898 cache_drop(&fdp->fd_njdir);
1899 vrele(fdp->fd_jdir);
1902 kfree(fdp->fd_knlist, M_KQUEUE);
1904 kfree(fdp->fd_knhash, M_KQUEUE);
1905 kfree(fdp, M_FILEDESC);
1909 * Retrieve and reference the file pointer associated with a descriptor.
1914 holdfp(struct filedesc *fdp, int fd, int flag)
1918 spin_lock_rd(&fdp->fd_spin);
1919 if (((u_int)fd) >= fdp->fd_nfiles) {
1923 if ((fp = fdp->fd_files[fd].fp) == NULL)
1925 if ((fp->f_flag & flag) == 0 && flag != -1) {
1931 spin_unlock_rd(&fdp->fd_spin);
1936 * holdsock() - load the struct file pointer associated
1937 * with a socket into *fpp. If an error occurs, non-zero
1938 * will be returned and *fpp will be set to NULL.
1943 holdsock(struct filedesc *fdp, int fd, struct file **fpp)
1948 spin_lock_rd(&fdp->fd_spin);
1949 if ((unsigned)fd >= fdp->fd_nfiles) {
1954 if ((fp = fdp->fd_files[fd].fp) == NULL) {
1958 if (fp->f_type != DTYPE_SOCKET) {
1965 spin_unlock_rd(&fdp->fd_spin);
1971 * Convert a user file descriptor to a held file pointer.
1976 holdvnode(struct filedesc *fdp, int fd, struct file **fpp)
1981 spin_lock_rd(&fdp->fd_spin);
1982 if ((unsigned)fd >= fdp->fd_nfiles) {
1987 if ((fp = fdp->fd_files[fd].fp) == NULL) {
1991 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
1999 spin_unlock_rd(&fdp->fd_spin);
2005 * For setugid programs, we don't want to people to use that setugidness
2006 * to generate error messages which write to a file which otherwise would
2007 * otherwise be off-limits to the process.
2009 * This is a gross hack to plug the hole. A better solution would involve
2010 * a special vop or other form of generalized access control mechanism. We
2011 * go ahead and just reject all procfs file systems accesses as dangerous.
2013 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2014 * sufficient. We also don't for check setugidness since we know we are.
2017 is_unsafe(struct file *fp)
2019 if (fp->f_type == DTYPE_VNODE &&
2020 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS)
2026 * Make this setguid thing safe, if at all possible.
2028 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2031 setugidsafety(struct proc *p)
2033 struct filedesc *fdp = p->p_fd;
2036 /* Certain daemons might not have file descriptors. */
2041 * note: fdp->fd_files may be reallocated out from under us while
2042 * we are blocked in a close. Be careful!
2044 for (i = 0; i <= fdp->fd_lastfile; i++) {
2047 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) {
2050 if (i < fdp->fd_knlistsize)
2051 knote_fdclose(p, i);
2053 * NULL-out descriptor prior to close to avoid
2054 * a race while close blocks.
2056 if ((fp = funsetfd_locked(fdp, i)) != NULL)
2063 * Close any files on exec?
2065 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2068 fdcloseexec(struct proc *p)
2070 struct filedesc *fdp = p->p_fd;
2073 /* Certain daemons might not have file descriptors. */
2078 * We cannot cache fd_files since operations may block and rip
2079 * them out from under us.
2081 for (i = 0; i <= fdp->fd_lastfile; i++) {
2082 if (fdp->fd_files[i].fp != NULL &&
2083 (fdp->fd_files[i].fileflags & UF_EXCLOSE)) {
2086 if (i < fdp->fd_knlistsize)
2087 knote_fdclose(p, i);
2089 * NULL-out descriptor prior to close to avoid
2090 * a race while close blocks.
2092 if ((fp = funsetfd_locked(fdp, i)) != NULL)
2099 * It is unsafe for set[ug]id processes to be started with file
2100 * descriptors 0..2 closed, as these descriptors are given implicit
2101 * significance in the Standard C library. fdcheckstd() will create a
2102 * descriptor referencing /dev/null for each of stdin, stdout, and
2103 * stderr that is not already open.
2105 * NOT MPSAFE - calls falloc, vn_open, etc
2108 fdcheckstd(struct proc *p)
2110 struct nlookupdata nd;
2111 struct filedesc *fdp;
2114 int i, error, flags, devnull;
2121 for (i = 0; i < 3; i++) {
2122 if (fdp->fd_files[i].fp != NULL)
2125 if ((error = falloc(p, &fp, &devnull)) != 0)
2128 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE,
2129 NLC_FOLLOW|NLC_LOCKVP);
2130 flags = FREAD | FWRITE;
2132 error = vn_open(&nd, fp, flags, 0);
2134 fsetfd(p, fp, devnull);
2136 fsetfd(p, NULL, devnull);
2141 KKASSERT(i == devnull);
2143 error = kern_dup(DUP_FIXED, devnull, i, &retval);
2152 * Internal form of close.
2153 * Decrement reference count on file structure.
2154 * Note: td and/or p may be NULL when closing a file
2155 * that was being passed in a message.
2157 * MPALMOSTSAFE - acquires mplock for VOP operations
2160 closef(struct file *fp, struct proc *p)
2164 struct filedesc_to_leader *fdtol;
2170 * POSIX record locking dictates that any close releases ALL
2171 * locks owned by this process. This is handled by setting
2172 * a flag in the unlock to free ONLY locks obeying POSIX
2173 * semantics, and not to free BSD-style file locks.
2174 * If the descriptor was in a message, POSIX-style locks
2175 * aren't passed with the descriptor.
2177 if (p != NULL && fp->f_type == DTYPE_VNODE &&
2178 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2181 if ((p->p_leader->p_flag & P_ADVLOCK) != 0) {
2182 lf.l_whence = SEEK_SET;
2185 lf.l_type = F_UNLCK;
2186 vp = (struct vnode *)fp->f_data;
2187 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
2191 if (fdtol != NULL) {
2193 * Handle special case where file descriptor table
2194 * is shared between multiple process leaders.
2196 for (fdtol = fdtol->fdl_next;
2197 fdtol != p->p_fdtol;
2198 fdtol = fdtol->fdl_next) {
2199 if ((fdtol->fdl_leader->p_flag &
2202 fdtol->fdl_holdcount++;
2203 lf.l_whence = SEEK_SET;
2206 lf.l_type = F_UNLCK;
2207 vp = (struct vnode *)fp->f_data;
2208 (void) VOP_ADVLOCK(vp,
2209 (caddr_t)fdtol->fdl_leader,
2210 F_UNLCK, &lf, F_POSIX);
2211 fdtol->fdl_holdcount--;
2212 if (fdtol->fdl_holdcount == 0 &&
2213 fdtol->fdl_wakeup != 0) {
2214 fdtol->fdl_wakeup = 0;
2227 * fhold() can only be called if f_count is already at least 1 (i.e. the
2228 * caller of fhold() already has a reference to the file pointer in some
2231 * f_count is not spin-locked. Instead, atomic ops are used for
2232 * incrementing, decrementing, and handling the 1->0 transition.
2235 fhold(struct file *fp)
2237 atomic_add_int(&fp->f_count, 1);
2241 * fdrop() - drop a reference to a descriptor
2243 * MPALMOSTSAFE - acquires mplock for final close sequence
2246 fdrop(struct file *fp)
2253 * A combined fetch and subtract is needed to properly detect
2254 * 1->0 transitions, otherwise two cpus dropping from a ref
2255 * count of 2 might both try to run the 1->0 code.
2257 if (atomic_fetchadd_int(&fp->f_count, -1) > 1)
2263 * The last reference has gone away, we own the fp structure free
2266 if (fp->f_count < 0)
2267 panic("fdrop: count < 0");
2268 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE &&
2269 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2271 lf.l_whence = SEEK_SET;
2274 lf.l_type = F_UNLCK;
2275 vp = (struct vnode *)fp->f_data;
2276 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2278 if (fp->f_ops != &badfileops)
2279 error = fo_close(fp);
2288 * Apply an advisory lock on a file descriptor.
2290 * Just attempt to get a record lock of the requested type on
2291 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2294 sys_flock(struct flock_args *uap)
2296 struct proc *p = curproc;
2302 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
2304 if (fp->f_type != DTYPE_VNODE) {
2308 vp = (struct vnode *)fp->f_data;
2309 lf.l_whence = SEEK_SET;
2312 if (uap->how & LOCK_UN) {
2313 lf.l_type = F_UNLCK;
2314 fp->f_flag &= ~FHASLOCK;
2315 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2318 if (uap->how & LOCK_EX)
2319 lf.l_type = F_WRLCK;
2320 else if (uap->how & LOCK_SH)
2321 lf.l_type = F_RDLCK;
2326 fp->f_flag |= FHASLOCK;
2327 if (uap->how & LOCK_NB)
2328 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0);
2330 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT);
2337 * File Descriptor pseudo-device driver (/dev/fd/).
2339 * Opening minor device N dup()s the file (if any) connected to file
2340 * descriptor N belonging to the calling process. Note that this driver
2341 * consists of only the ``open()'' routine, because all subsequent
2342 * references to this file will be direct to the other driver.
2346 fdopen(struct dev_open_args *ap)
2348 thread_t td = curthread;
2350 KKASSERT(td->td_lwp != NULL);
2353 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the
2354 * the file descriptor being sought for duplication. The error
2355 * return ensures that the vnode for this device will be released
2356 * by vn_open. Open will detect this special error and take the
2357 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2358 * will simply report the error.
2360 td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev);
2365 * The caller has reserved the file descriptor dfd for us. On success we
2366 * must fsetfd() it. On failure the caller will clean it up.
2368 * NOT MPSAFE - isn't getting spinlocks, possibly other things
2371 dupfdopen(struct proc *p, int dfd, int sfd, int mode, int error)
2373 struct filedesc *fdp = p->p_fd;
2378 if ((wfp = holdfp(fdp, sfd, -1)) == NULL)
2382 * Close a revoke/dup race. Duping a descriptor marked as revoked
2383 * will dup a dummy descriptor instead of the real one.
2385 if (wfp->f_flag & FREVOKED) {
2386 kprintf("Warning: attempt to dup() a revoked descriptor\n");
2389 werror = falloc(NULL, &wfp, NULL);
2395 * There are two cases of interest here.
2397 * For ENODEV simply dup sfd to file descriptor dfd and return.
2399 * For ENXIO steal away the file structure from sfd and store it
2400 * dfd. sfd is effectively closed by this operation.
2402 * Any other error code is just returned.
2407 * Check that the mode the file is being opened for is a
2408 * subset of the mode of the existing descriptor.
2410 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2414 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2415 fsetfd(p, wfp, dfd);
2420 * Steal away the file pointer from dfd, and stuff it into indx.
2422 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2423 fsetfd(p, wfp, dfd);
2424 if ((xfp = funsetfd_locked(fdp, sfd)) != NULL)
2436 * NOT MPSAFE - I think these refer to a common file descriptor table
2437 * and we need to spinlock that to link fdtol in.
2439 struct filedesc_to_leader *
2440 filedesc_to_leader_alloc(struct filedesc_to_leader *old,
2441 struct proc *leader)
2443 struct filedesc_to_leader *fdtol;
2445 fdtol = kmalloc(sizeof(struct filedesc_to_leader),
2446 M_FILEDESC_TO_LEADER, M_WAITOK);
2447 fdtol->fdl_refcount = 1;
2448 fdtol->fdl_holdcount = 0;
2449 fdtol->fdl_wakeup = 0;
2450 fdtol->fdl_leader = leader;
2452 fdtol->fdl_next = old->fdl_next;
2453 fdtol->fdl_prev = old;
2454 old->fdl_next = fdtol;
2455 fdtol->fdl_next->fdl_prev = fdtol;
2457 fdtol->fdl_next = fdtol;
2458 fdtol->fdl_prev = fdtol;
2464 * Scan all file pointers in the system. The callback is made with
2465 * the master list spinlock held exclusively.
2470 allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data)
2475 spin_lock_wr(&filehead_spin);
2476 LIST_FOREACH(fp, &filehead, f_list) {
2477 res = callback(fp, data);
2481 spin_unlock_wr(&filehead_spin);
2485 * Get file structures.
2487 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe)
2490 struct sysctl_kern_file_info {
2493 struct sysctl_req *req;
2496 static int sysctl_kern_file_callback(struct proc *p, void *data);
2499 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2501 struct sysctl_kern_file_info info;
2504 * Note: because the number of file descriptors is calculated
2505 * in different ways for sizing vs returning the data,
2506 * there is information leakage from the first loop. However,
2507 * it is of a similar order of magnitude to the leakage from
2508 * global system statistics such as kern.openfiles.
2510 * When just doing a count, note that we cannot just count
2511 * the elements and add f_count via the filehead list because
2512 * threaded processes share their descriptor table and f_count might
2513 * still be '1' in that case.
2515 * Since the SYSCTL op can block, we must hold the process to
2516 * prevent it being ripped out from under us either in the
2517 * file descriptor loop or in the greater LIST_FOREACH. The
2518 * process may be in varying states of disrepair. If the process
2519 * is in SZOMB we may have caught it just as it is being removed
2520 * from the allproc list, we must skip it in that case to maintain
2521 * an unbroken chain through the allproc list.
2526 allproc_scan(sysctl_kern_file_callback, &info);
2529 * When just calculating the size, overestimate a bit to try to
2530 * prevent system activity from causing the buffer-fill call
2533 if (req->oldptr == NULL) {
2534 info.count = (info.count + 16) + (info.count / 10);
2535 info.error = SYSCTL_OUT(req, NULL,
2536 info.count * sizeof(struct kinfo_file));
2538 return (info.error);
2542 sysctl_kern_file_callback(struct proc *p, void *data)
2544 struct sysctl_kern_file_info *info = data;
2545 struct kinfo_file kf;
2546 struct filedesc *fdp;
2551 if (p->p_stat == SIDL || p->p_stat == SZOMB)
2553 if (!PRISON_CHECK(info->req->td->td_proc->p_ucred, p->p_ucred) != 0)
2555 if ((fdp = p->p_fd) == NULL)
2557 spin_lock_rd(&fdp->fd_spin);
2558 for (n = 0; n < fdp->fd_nfiles; ++n) {
2559 if ((fp = fdp->fd_files[n].fp) == NULL)
2561 if (info->req->oldptr == NULL) {
2564 uid = p->p_ucred ? p->p_ucred->cr_uid : -1;
2565 kcore_make_file(&kf, fp, p->p_pid, uid, n);
2566 spin_unlock_rd(&fdp->fd_spin);
2567 info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf));
2568 spin_lock_rd(&fdp->fd_spin);
2573 spin_unlock_rd(&fdp->fd_spin);
2579 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2580 0, 0, sysctl_kern_file, "S,file", "Entire file table");
2582 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2583 &maxfilesperproc, 0, "Maximum files allowed open per process");
2585 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2586 &maxfiles, 0, "Maximum number of files");
2588 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW,
2589 &maxfilesrootres, 0, "Descriptors reserved for root use");
2591 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2592 &nfiles, 0, "System-wide number of open files");
2595 fildesc_drvinit(void *unused)
2599 dev_ops_add(&fildesc_ops, 0, 0);
2600 for (fd = 0; fd < NUMFDESC; fd++) {
2601 make_dev(&fildesc_ops, fd,
2602 UID_BIN, GID_BIN, 0666, "fd/%d", fd);
2604 make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin");
2605 make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout");
2606 make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr");
2612 struct fileops badfileops = {
2613 .fo_read = badfo_readwrite,
2614 .fo_write = badfo_readwrite,
2615 .fo_ioctl = badfo_ioctl,
2616 .fo_poll = badfo_poll,
2617 .fo_kqfilter = badfo_kqfilter,
2618 .fo_stat = badfo_stat,
2619 .fo_close = badfo_close,
2620 .fo_shutdown = badfo_shutdown
2640 badfo_ioctl(struct file *fp, u_long com, caddr_t data, struct ucred *cred)
2649 badfo_poll(struct file *fp, int events, struct ucred *cred)
2658 badfo_kqfilter(struct file *fp, struct knote *kn)
2664 badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
2673 badfo_close(struct file *fp)
2682 badfo_shutdown(struct file *fp, int how)
2691 nofo_shutdown(struct file *fp, int how)
2693 return (EOPNOTSUPP);
2696 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,
2697 fildesc_drvinit,NULL)