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 int checkfpclosed(struct filedesc *fdp, int fd, struct file *fp);
114 static void ffree(struct file *fp);
116 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
117 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
118 "file desc to leader structures");
119 MALLOC_DEFINE(M_FILE, "file", "Open file structure");
120 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
122 static d_open_t fdopen;
125 #define CDEV_MAJOR 22
126 static struct dev_ops fildesc_ops = {
127 { "FD", CDEV_MAJOR, 0 },
132 * Descriptor management.
134 static struct filelist filehead = LIST_HEAD_INITIALIZER(&filehead);
135 static struct spinlock filehead_spin = SPINLOCK_INITIALIZER(&filehead_spin);
136 static int nfiles; /* actual number of open files */
140 * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared.
142 * MPSAFE - must be called with fdp->fd_spin exclusively held
146 fdfixup_locked(struct filedesc *fdp, int fd)
148 if (fd < fdp->fd_freefile) {
149 fdp->fd_freefile = fd;
151 while (fdp->fd_lastfile >= 0 &&
152 fdp->fd_files[fdp->fd_lastfile].fp == NULL &&
153 fdp->fd_files[fdp->fd_lastfile].reserved == 0
160 * System calls on descriptors.
165 sys_getdtablesize(struct getdtablesize_args *uap)
167 struct proc *p = curproc;
168 struct plimit *limit = p->p_limit;
170 spin_lock_rd(&limit->p_spin);
172 min((int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
173 spin_unlock_rd(&limit->p_spin);
178 * Duplicate a file descriptor to a particular value.
180 * note: keep in mind that a potential race condition exists when closing
181 * descriptors from a shared descriptor table (via rfork).
186 sys_dup2(struct dup2_args *uap)
191 error = kern_dup(DUP_FIXED, uap->from, uap->to, &fd);
192 uap->sysmsg_fds[0] = fd;
198 * Duplicate a file descriptor.
203 sys_dup(struct dup_args *uap)
208 error = kern_dup(DUP_VARIABLE, uap->fd, 0, &fd);
209 uap->sysmsg_fds[0] = fd;
215 * MPALMOSTSAFE - acquires mplock for fp operations
218 kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred)
220 struct thread *td = curthread;
221 struct proc *p = td->td_proc;
227 int tmp, error, flg = F_POSIX;
232 * Operations on file descriptors that do not require a file pointer.
236 error = fgetfdflags(p->p_fd, fd, &tmp);
238 dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0;
242 if (dat->fc_cloexec & FD_CLOEXEC)
243 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
245 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
249 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd);
256 * Operations on file pointers
258 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
264 dat->fc_flags = OFLAGS(fp->f_flag);
270 nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS;
271 nflags |= oflags & ~FCNTLFLAGS;
274 if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY))
276 if (error == 0 && ((nflags ^ oflags) & FASYNC)) {
277 tmp = nflags & FASYNC;
278 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp,
286 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner,
291 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner,
297 /* Fall into F_SETLK */
300 if (fp->f_type != DTYPE_VNODE) {
304 vp = (struct vnode *)fp->f_data;
307 * copyin/lockop may block
309 if (dat->fc_flock.l_whence == SEEK_CUR)
310 dat->fc_flock.l_start += fp->f_offset;
312 switch (dat->fc_flock.l_type) {
314 if ((fp->f_flag & FREAD) == 0) {
318 p->p_leader->p_flag |= P_ADVLOCK;
319 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
320 &dat->fc_flock, flg);
323 if ((fp->f_flag & FWRITE) == 0) {
327 p->p_leader->p_flag |= P_ADVLOCK;
328 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
329 &dat->fc_flock, flg);
332 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
333 &dat->fc_flock, F_POSIX);
341 * It is possible to race a close() on the descriptor while
342 * we were blocked getting the lock. If this occurs the
343 * close might not have caught the lock.
345 if (checkfpclosed(p->p_fd, fd, fp)) {
346 dat->fc_flock.l_whence = SEEK_SET;
347 dat->fc_flock.l_start = 0;
348 dat->fc_flock.l_len = 0;
349 dat->fc_flock.l_type = F_UNLCK;
350 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
351 F_UNLCK, &dat->fc_flock, F_POSIX);
356 if (fp->f_type != DTYPE_VNODE) {
360 vp = (struct vnode *)fp->f_data;
362 * copyin/lockop may block
364 if (dat->fc_flock.l_type != F_RDLCK &&
365 dat->fc_flock.l_type != F_WRLCK &&
366 dat->fc_flock.l_type != F_UNLCK) {
370 if (dat->fc_flock.l_whence == SEEK_CUR)
371 dat->fc_flock.l_start += fp->f_offset;
372 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK,
373 &dat->fc_flock, F_POSIX);
386 * The file control system call.
391 sys_fcntl(struct fcntl_args *uap)
398 dat.fc_fd = uap->arg;
401 dat.fc_cloexec = uap->arg;
404 dat.fc_flags = uap->arg;
407 dat.fc_owner = uap->arg;
412 error = copyin((caddr_t)uap->arg, &dat.fc_flock,
413 sizeof(struct flock));
419 error = kern_fcntl(uap->fd, uap->cmd, &dat, curthread->td_ucred);
424 uap->sysmsg_result = dat.fc_fd;
427 uap->sysmsg_result = dat.fc_cloexec;
430 uap->sysmsg_result = dat.fc_flags;
433 uap->sysmsg_result = dat.fc_owner;
435 error = copyout(&dat.fc_flock, (caddr_t)uap->arg,
436 sizeof(struct flock));
445 * Common code for dup, dup2, and fcntl(F_DUPFD).
447 * The type flag can be either DUP_FIXED or DUP_VARIABLE. DUP_FIXED tells
448 * kern_dup() to destructively dup over an existing file descriptor if new
449 * is already open. DUP_VARIABLE tells kern_dup() to find the lowest
450 * unused file descriptor that is greater than or equal to new.
455 kern_dup(enum dup_type type, int old, int new, int *res)
457 struct thread *td = curthread;
458 struct proc *p = td->td_proc;
459 struct filedesc *fdp = p->p_fd;
467 * Verify that we have a valid descriptor to dup from and
468 * possibly to dup to.
471 spin_lock_wr(&fdp->fd_spin);
472 if (new < 0 || new > p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
473 new >= maxfilesperproc) {
474 spin_unlock_wr(&fdp->fd_spin);
477 if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) {
478 spin_unlock_wr(&fdp->fd_spin);
481 if (type == DUP_FIXED && old == new) {
483 spin_unlock_wr(&fdp->fd_spin);
486 fp = fdp->fd_files[old].fp;
487 oldflags = fdp->fd_files[old].fileflags;
488 fhold(fp); /* MPSAFE - can be called with a spinlock held */
491 * Allocate a new descriptor if DUP_VARIABLE, or expand the table
492 * if the requested descriptor is beyond the current table size.
494 * This can block. Retry if the source descriptor no longer matches
495 * or if our expectation in the expansion case races.
497 * If we are not expanding or allocating a new decriptor, then reset
498 * the target descriptor to a reserved state so we have a uniform
499 * setup for the next code block.
501 if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) {
502 spin_unlock_wr(&fdp->fd_spin);
503 error = fdalloc(p, new, &newfd);
504 spin_lock_wr(&fdp->fd_spin);
506 spin_unlock_wr(&fdp->fd_spin);
513 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) {
514 fsetfd_locked(fdp, NULL, newfd);
515 spin_unlock_wr(&fdp->fd_spin);
520 * Check for expansion race
522 if (type != DUP_VARIABLE && new != newfd) {
523 fsetfd_locked(fdp, NULL, newfd);
524 spin_unlock_wr(&fdp->fd_spin);
529 * Check for ripout, newfd reused old (this case probably
533 fsetfd_locked(fdp, NULL, newfd);
534 spin_unlock_wr(&fdp->fd_spin);
541 if (fdp->fd_files[new].reserved) {
542 spin_unlock_wr(&fdp->fd_spin);
544 kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new);
545 tsleep(fdp, 0, "fdres", hz);
550 * If the target descriptor was never allocated we have
551 * to allocate it. If it was we have to clean out the
552 * old descriptor. delfp inherits the ref from the
555 delfp = fdp->fd_files[new].fp;
556 fdp->fd_files[new].fp = NULL;
557 fdp->fd_files[new].reserved = 1;
559 fdreserve_locked(fdp, new, 1);
560 if (new > fdp->fd_lastfile)
561 fdp->fd_lastfile = new;
567 * NOTE: still holding an exclusive spinlock
571 * If a descriptor is being overwritten we may hve to tell
572 * fdfree() to sleep to ensure that all relevant process
573 * leaders can be traversed in closef().
575 if (delfp != NULL && p->p_fdtol != NULL) {
576 fdp->fd_holdleaderscount++;
581 KASSERT(delfp == NULL || type == DUP_FIXED,
582 ("dup() picked an open file"));
585 * Duplicate the source descriptor, update lastfile. If the new
586 * descriptor was not allocated and we aren't replacing an existing
587 * descriptor we have to mark the descriptor as being in use.
589 * The fd_files[] array inherits fp's hold reference.
591 fsetfd_locked(fdp, fp, new);
592 fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE;
593 spin_unlock_wr(&fdp->fd_spin);
598 * If we dup'd over a valid file, we now own the reference to it
599 * and must dispose of it using closef() semantics (as if a
600 * close() were performed on it).
605 spin_lock_wr(&fdp->fd_spin);
606 fdp->fd_holdleaderscount--;
607 if (fdp->fd_holdleaderscount == 0 &&
608 fdp->fd_holdleaderswakeup != 0) {
609 fdp->fd_holdleaderswakeup = 0;
610 spin_unlock_wr(&fdp->fd_spin);
611 wakeup(&fdp->fd_holdleaderscount);
613 spin_unlock_wr(&fdp->fd_spin);
621 * If sigio is on the list associated with a process or process group,
622 * disable signalling from the device, remove sigio from the list and
626 funsetown(struct sigio *sigio)
631 *(sigio->sio_myref) = NULL;
633 if (sigio->sio_pgid < 0) {
634 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
636 } else /* if ((*sigiop)->sio_pgid > 0) */ {
637 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
640 crfree(sigio->sio_ucred);
641 kfree(sigio, M_SIGIO);
644 /* Free a list of sigio structures. */
646 funsetownlst(struct sigiolst *sigiolst)
650 while ((sigio = SLIST_FIRST(sigiolst)) != NULL)
655 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
657 * After permission checking, add a sigio structure to the sigio list for
658 * the process or process group.
661 fsetown(pid_t pgid, struct sigio **sigiop)
677 * Policy - Don't allow a process to FSETOWN a process
678 * in another session.
680 * Remove this test to allow maximum flexibility or
681 * restrict FSETOWN to the current process or process
682 * group for maximum safety.
684 if (proc->p_session != curproc->p_session)
688 } else /* if (pgid < 0) */ {
689 pgrp = pgfind(-pgid);
694 * Policy - Don't allow a process to FSETOWN a process
695 * in another session.
697 * Remove this test to allow maximum flexibility or
698 * restrict FSETOWN to the current process or process
699 * group for maximum safety.
701 if (pgrp->pg_session != curproc->p_session)
707 sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
709 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
710 sigio->sio_proc = proc;
712 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
713 sigio->sio_pgrp = pgrp;
715 sigio->sio_pgid = pgid;
716 sigio->sio_ucred = crhold(curthread->td_ucred);
717 /* It would be convenient if p_ruid was in ucred. */
718 sigio->sio_ruid = sigio->sio_ucred->cr_ruid;
719 sigio->sio_myref = sigiop;
727 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
730 fgetown(struct sigio *sigio)
732 return (sigio != NULL ? sigio->sio_pgid : 0);
736 * Close many file descriptors.
741 sys_closefrom(struct closefrom_args *uap)
743 return(kern_closefrom(uap->fd));
747 * Close all file descriptors greater then or equal to fd
752 kern_closefrom(int fd)
754 struct thread *td = curthread;
755 struct proc *p = td->td_proc;
756 struct filedesc *fdp;
765 * NOTE: This function will skip unassociated descriptors and
766 * reserved descriptors that have not yet been assigned.
767 * fd_lastfile can change as a side effect of kern_close().
769 spin_lock_wr(&fdp->fd_spin);
770 while (fd <= fdp->fd_lastfile) {
771 if (fdp->fd_files[fd].fp != NULL) {
772 spin_unlock_wr(&fdp->fd_spin);
773 /* ok if this races another close */
774 if (kern_close(fd) == EINTR)
776 spin_lock_wr(&fdp->fd_spin);
780 spin_unlock_wr(&fdp->fd_spin);
785 * Close a file descriptor.
790 sys_close(struct close_args *uap)
792 return(kern_close(uap->fd));
796 * MPALMOSTSAFE - acquires mplock around knote_fdclose() calls
801 struct thread *td = curthread;
802 struct proc *p = td->td_proc;
803 struct filedesc *fdp;
811 spin_lock_wr(&fdp->fd_spin);
812 if ((fp = funsetfd_locked(fdp, fd)) == NULL) {
813 spin_unlock_wr(&fdp->fd_spin);
817 if (p->p_fdtol != NULL) {
819 * Ask fdfree() to sleep to ensure that all relevant
820 * process leaders can be traversed in closef().
822 fdp->fd_holdleaderscount++;
827 * we now hold the fp reference that used to be owned by the descriptor
830 spin_unlock_wr(&fdp->fd_spin);
831 if (fd < fdp->fd_knlistsize) {
833 if (fd < fdp->fd_knlistsize)
834 knote_fdclose(p, fd);
837 error = closef(fp, p);
839 spin_lock_wr(&fdp->fd_spin);
840 fdp->fd_holdleaderscount--;
841 if (fdp->fd_holdleaderscount == 0 &&
842 fdp->fd_holdleaderswakeup != 0) {
843 fdp->fd_holdleaderswakeup = 0;
844 spin_unlock_wr(&fdp->fd_spin);
845 wakeup(&fdp->fd_holdleaderscount);
847 spin_unlock_wr(&fdp->fd_spin);
854 * shutdown_args(int fd, int how)
857 kern_shutdown(int fd, int how)
859 struct thread *td = curthread;
860 struct proc *p = td->td_proc;
866 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
868 error = fo_shutdown(fp, how);
878 sys_shutdown(struct shutdown_args *uap)
883 error = kern_shutdown(uap->s, uap->how);
893 kern_fstat(int fd, struct stat *ub)
895 struct thread *td = curthread;
896 struct proc *p = td->td_proc;
902 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
904 error = fo_stat(fp, ub, td->td_ucred);
911 * Return status information about a file descriptor.
916 sys_fstat(struct fstat_args *uap)
921 error = kern_fstat(uap->fd, &st);
924 error = copyout(&st, uap->sb, sizeof(st));
929 * Return pathconf information about a file descriptor.
934 sys_fpathconf(struct fpathconf_args *uap)
936 struct thread *td = curthread;
937 struct proc *p = td->td_proc;
942 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
945 switch (fp->f_type) {
948 if (uap->name != _PC_PIPE_BUF) {
951 uap->sysmsg_result = PIPE_BUF;
957 vp = (struct vnode *)fp->f_data;
959 error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg);
971 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0, "");
974 * Grow the file table so it can hold through descriptor (want).
976 * The fdp's spinlock must be held exclusively on entry and may be held
977 * exclusively on return. The spinlock may be cycled by the routine.
982 fdgrow_locked(struct filedesc *fdp, int want)
984 struct fdnode *newfiles;
985 struct fdnode *oldfiles;
990 /* nf has to be of the form 2^n - 1 */
992 } while (nf <= want);
994 spin_unlock_wr(&fdp->fd_spin);
995 newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK);
996 spin_lock_wr(&fdp->fd_spin);
999 * We could have raced another extend while we were not holding
1002 if (fdp->fd_nfiles >= nf) {
1003 spin_unlock_wr(&fdp->fd_spin);
1004 kfree(newfiles, M_FILEDESC);
1005 spin_lock_wr(&fdp->fd_spin);
1009 * Copy the existing ofile and ofileflags arrays
1010 * and zero the new portion of each array.
1012 extra = nf - fdp->fd_nfiles;
1013 bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode));
1014 bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode));
1016 oldfiles = fdp->fd_files;
1017 fdp->fd_files = newfiles;
1018 fdp->fd_nfiles = nf;
1020 if (oldfiles != fdp->fd_builtin_files) {
1021 spin_unlock_wr(&fdp->fd_spin);
1022 kfree(oldfiles, M_FILEDESC);
1023 spin_lock_wr(&fdp->fd_spin);
1029 * Number of nodes in right subtree, including the root.
1032 right_subtree_size(int n)
1034 return (n ^ (n | (n + 1)));
1041 right_ancestor(int n)
1043 return (n | (n + 1));
1050 left_ancestor(int n)
1052 return ((n & (n + 1)) - 1);
1056 * Traverse the in-place binary tree buttom-up adjusting the allocation
1057 * count so scans can determine where free descriptors are located.
1059 * MPSAFE - caller must be holding an exclusive spinlock on fdp
1063 fdreserve_locked(struct filedesc *fdp, int fd, int incr)
1066 fdp->fd_files[fd].allocated += incr;
1067 KKASSERT(fdp->fd_files[fd].allocated >= 0);
1068 fd = left_ancestor(fd);
1073 * Reserve a file descriptor for the process. If no error occurs, the
1074 * caller MUST at some point call fsetfd() or assign a file pointer
1075 * or dispose of the reservation.
1080 fdalloc(struct proc *p, int want, int *result)
1082 struct filedesc *fdp = p->p_fd;
1083 int fd, rsize, rsum, node, lim;
1085 spin_lock_rd(&p->p_limit->p_spin);
1086 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
1087 spin_unlock_rd(&p->p_limit->p_spin);
1090 spin_lock_wr(&fdp->fd_spin);
1091 if (want >= fdp->fd_nfiles)
1092 fdgrow_locked(fdp, want);
1095 * Search for a free descriptor starting at the higher
1096 * of want or fd_freefile. If that fails, consider
1097 * expanding the ofile array.
1099 * NOTE! the 'allocated' field is a cumulative recursive allocation
1100 * count. If we happen to see a value of 0 then we can shortcut
1101 * our search. Otherwise we run through through the tree going
1102 * down branches we know have free descriptor(s) until we hit a
1103 * leaf node. The leaf node will be free but will not necessarily
1104 * have an allocated field of 0.
1107 /* move up the tree looking for a subtree with a free node */
1108 for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim);
1109 fd = right_ancestor(fd)) {
1110 if (fdp->fd_files[fd].allocated == 0)
1113 rsize = right_subtree_size(fd);
1114 if (fdp->fd_files[fd].allocated == rsize)
1115 continue; /* right subtree full */
1118 * Free fd is in the right subtree of the tree rooted at fd.
1119 * Call that subtree R. Look for the smallest (leftmost)
1120 * subtree of R with an unallocated fd: continue moving
1121 * down the left branch until encountering a full left
1122 * subtree, then move to the right.
1124 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) {
1126 rsum += fdp->fd_files[node].allocated;
1127 if (fdp->fd_files[fd].allocated == rsum + rsize) {
1128 fd = node; /* move to the right */
1129 if (fdp->fd_files[node].allocated == 0)
1138 * No space in current array. Expand?
1140 if (fdp->fd_nfiles >= lim) {
1141 spin_unlock_wr(&fdp->fd_spin);
1144 fdgrow_locked(fdp, want);
1148 KKASSERT(fd < fdp->fd_nfiles);
1149 if (fd > fdp->fd_lastfile)
1150 fdp->fd_lastfile = fd;
1151 if (want <= fdp->fd_freefile)
1152 fdp->fd_freefile = fd;
1154 KKASSERT(fdp->fd_files[fd].fp == NULL);
1155 KKASSERT(fdp->fd_files[fd].reserved == 0);
1156 fdp->fd_files[fd].fileflags = 0;
1157 fdp->fd_files[fd].reserved = 1;
1158 fdreserve_locked(fdp, fd, 1);
1159 spin_unlock_wr(&fdp->fd_spin);
1164 * Check to see whether n user file descriptors
1165 * are available to the process p.
1170 fdavail(struct proc *p, int n)
1172 struct filedesc *fdp = p->p_fd;
1173 struct fdnode *fdnode;
1176 spin_lock_rd(&p->p_limit->p_spin);
1177 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
1178 spin_unlock_rd(&p->p_limit->p_spin);
1180 spin_lock_rd(&fdp->fd_spin);
1181 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) {
1182 spin_unlock_rd(&fdp->fd_spin);
1185 last = min(fdp->fd_nfiles, lim);
1186 fdnode = &fdp->fd_files[fdp->fd_freefile];
1187 for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) {
1188 if (fdnode->fp == NULL && --n <= 0) {
1189 spin_unlock_rd(&fdp->fd_spin);
1193 spin_unlock_rd(&fdp->fd_spin);
1198 * Revoke open descriptors referencing (f_data, f_type)
1200 * Any revoke executed within a prison is only able to
1201 * revoke descriptors for processes within that prison.
1203 * Returns 0 on success or an error code.
1205 struct fdrevoke_info {
1215 static int fdrevoke_check_callback(struct file *fp, void *vinfo);
1216 static int fdrevoke_proc_callback(struct proc *p, void *vinfo);
1219 fdrevoke(void *f_data, short f_type, struct ucred *cred)
1221 struct fdrevoke_info info;
1224 bzero(&info, sizeof(info));
1228 error = falloc(NULL, &info.nfp, NULL);
1233 * Scan the file pointer table once. dups do not dup file pointers,
1234 * only descriptors, so there is no leak. Set FREVOKED on the fps
1237 allfiles_scan_exclusive(fdrevoke_check_callback, &info);
1240 * If any fps were marked track down the related descriptors
1241 * and close them. Any dup()s at this point will notice
1242 * the FREVOKED already set in the fp and do the right thing.
1244 * Any fps with non-zero msgcounts (aka sent over a unix-domain
1245 * socket) bumped the intransit counter and will require a
1246 * scan. Races against fps leaving the socket are closed by
1247 * the socket code checking for FREVOKED.
1250 allproc_scan(fdrevoke_proc_callback, &info);
1252 unp_revoke_gc(info.nfp);
1258 * Locate matching file pointers directly.
1261 fdrevoke_check_callback(struct file *fp, void *vinfo)
1263 struct fdrevoke_info *info = vinfo;
1266 * File pointers already flagged for revokation are skipped.
1268 if (fp->f_flag & FREVOKED)
1272 * If revoking from a prison file pointers created outside of
1273 * that prison, or file pointers without creds, cannot be revoked.
1275 if (info->cred->cr_prison &&
1276 (fp->f_cred == NULL ||
1277 info->cred->cr_prison != fp->f_cred->cr_prison)) {
1282 * If the file pointer matches then mark it for revocation. The
1283 * flag is currently only used by unp_revoke_gc().
1285 * info->count is a heuristic and can race in a SMP environment.
1287 if (info->data == fp->f_data && info->type == fp->f_type) {
1288 atomic_set_int(&fp->f_flag, FREVOKED);
1289 info->count += fp->f_count;
1297 * Locate matching file pointers via process descriptor tables.
1300 fdrevoke_proc_callback(struct proc *p, void *vinfo)
1302 struct fdrevoke_info *info = vinfo;
1303 struct filedesc *fdp;
1307 if (p->p_stat == SIDL || p->p_stat == SZOMB)
1309 if (info->cred->cr_prison &&
1310 info->cred->cr_prison != p->p_ucred->cr_prison) {
1315 * If the controlling terminal of the process matches the
1316 * vnode being revoked we clear the controlling terminal.
1318 * The normal spec_close() may not catch this because it
1319 * uses curproc instead of p.
1321 if (p->p_session && info->type == DTYPE_VNODE &&
1322 info->data == p->p_session->s_ttyvp) {
1323 p->p_session->s_ttyvp = NULL;
1328 * Softref the fdp to prevent it from being destroyed
1330 spin_lock_wr(&p->p_spin);
1331 if ((fdp = p->p_fd) == NULL) {
1332 spin_unlock_wr(&p->p_spin);
1335 atomic_add_int(&fdp->fd_softrefs, 1);
1336 spin_unlock_wr(&p->p_spin);
1339 * Locate and close any matching file descriptors.
1341 spin_lock_wr(&fdp->fd_spin);
1342 for (n = 0; n < fdp->fd_nfiles; ++n) {
1343 if ((fp = fdp->fd_files[n].fp) == NULL)
1345 if (fp->f_flag & FREVOKED) {
1347 fdp->fd_files[n].fp = info->nfp;
1348 spin_unlock_wr(&fdp->fd_spin);
1350 spin_lock_wr(&fdp->fd_spin);
1354 spin_unlock_wr(&fdp->fd_spin);
1355 atomic_subtract_int(&fdp->fd_softrefs, 1);
1361 * Create a new open file structure and reserve a file decriptor
1362 * for the process that refers to it.
1364 * Root creds are checked using lp, or assumed if lp is NULL. If
1365 * resultfd is non-NULL then lp must also be non-NULL. No file
1366 * descriptor is reserved (and no process context is needed) if
1369 * A file pointer with a refcount of 1 is returned. Note that the
1370 * file pointer is NOT associated with the descriptor. If falloc
1371 * returns success, fsetfd() MUST be called to either associate the
1372 * file pointer or clear the reservation.
1377 falloc(struct lwp *lp, struct file **resultfp, int *resultfd)
1379 static struct timeval lastfail;
1382 struct ucred *cred = lp ? lp->lwp_thread->td_ucred : proc0.p_ucred;
1388 * Handle filetable full issues and root overfill.
1390 if (nfiles >= maxfiles - maxfilesrootres &&
1391 (cred->cr_ruid != 0 || nfiles >= maxfiles)) {
1392 if (ppsratecheck(&lastfail, &curfail, 1)) {
1393 kprintf("kern.maxfiles limit exceeded by uid %d, "
1394 "please see tuning(7).\n",
1402 * Allocate a new file descriptor.
1404 fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO);
1405 spin_init(&fp->f_spin);
1407 fp->f_ops = &badfileops;
1409 fp->f_cred = crhold(cred);
1410 spin_lock_wr(&filehead_spin);
1412 LIST_INSERT_HEAD(&filehead, fp, f_list);
1413 spin_unlock_wr(&filehead_spin);
1415 if ((error = fdalloc(lp->lwp_proc, 0, resultfd)) != 0) {
1432 checkfpclosed(struct filedesc *fdp, int fd, struct file *fp)
1436 spin_lock_rd(&fdp->fd_spin);
1437 if ((unsigned) fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp)
1441 spin_unlock_rd(&fdp->fd_spin);
1446 * Associate a file pointer with a previously reserved file descriptor.
1447 * This function always succeeds.
1449 * If fp is NULL, the file descriptor is returned to the pool.
1453 * MPSAFE (exclusive spinlock must be held on call)
1456 fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd)
1458 KKASSERT((unsigned)fd < fdp->fd_nfiles);
1459 KKASSERT(fdp->fd_files[fd].reserved != 0);
1462 fdp->fd_files[fd].fp = fp;
1463 fdp->fd_files[fd].reserved = 0;
1464 if (fp->f_type == DTYPE_KQUEUE) {
1465 if (fdp->fd_knlistsize < 0)
1466 fdp->fd_knlistsize = 0;
1469 fdp->fd_files[fd].reserved = 0;
1470 fdreserve_locked(fdp, fd, -1);
1471 fdfixup_locked(fdp, fd);
1479 fsetfd(struct filedesc *fdp, struct file *fp, int fd)
1481 spin_lock_wr(&fdp->fd_spin);
1482 fsetfd_locked(fdp, fp, fd);
1483 spin_unlock_wr(&fdp->fd_spin);
1487 * MPSAFE (exclusive spinlock must be held on call)
1491 funsetfd_locked(struct filedesc *fdp, int fd)
1495 if ((unsigned)fd >= fdp->fd_nfiles)
1497 if ((fp = fdp->fd_files[fd].fp) == NULL)
1499 fdp->fd_files[fd].fp = NULL;
1500 fdp->fd_files[fd].fileflags = 0;
1502 fdreserve_locked(fdp, fd, -1);
1503 fdfixup_locked(fdp, fd);
1511 fgetfdflags(struct filedesc *fdp, int fd, int *flagsp)
1515 spin_lock_rd(&fdp->fd_spin);
1516 if (((u_int)fd) >= fdp->fd_nfiles) {
1518 } else if (fdp->fd_files[fd].fp == NULL) {
1521 *flagsp = fdp->fd_files[fd].fileflags;
1524 spin_unlock_rd(&fdp->fd_spin);
1532 fsetfdflags(struct filedesc *fdp, int fd, int add_flags)
1536 spin_lock_wr(&fdp->fd_spin);
1537 if (((u_int)fd) >= fdp->fd_nfiles) {
1539 } else if (fdp->fd_files[fd].fp == NULL) {
1542 fdp->fd_files[fd].fileflags |= add_flags;
1545 spin_unlock_wr(&fdp->fd_spin);
1553 fclrfdflags(struct filedesc *fdp, int fd, int rem_flags)
1557 spin_lock_wr(&fdp->fd_spin);
1558 if (((u_int)fd) >= fdp->fd_nfiles) {
1560 } else if (fdp->fd_files[fd].fp == NULL) {
1563 fdp->fd_files[fd].fileflags &= ~rem_flags;
1566 spin_unlock_wr(&fdp->fd_spin);
1571 fsetcred(struct file *fp, struct ucred *cr)
1579 * Free a file descriptor.
1583 ffree(struct file *fp)
1585 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!"));
1586 spin_lock_wr(&filehead_spin);
1587 LIST_REMOVE(fp, f_list);
1589 spin_unlock_wr(&filehead_spin);
1591 if (fp->f_nchandle.ncp)
1592 cache_drop(&fp->f_nchandle);
1597 * called from init_main, initialize filedesc0 for proc0.
1600 fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask)
1604 fdp0->fd_refcnt = 1;
1605 fdp0->fd_cmask = cmask;
1606 fdp0->fd_files = fdp0->fd_builtin_files;
1607 fdp0->fd_nfiles = NDFILE;
1608 fdp0->fd_lastfile = -1;
1609 spin_init(&fdp0->fd_spin);
1613 * Build a new filedesc structure.
1618 fdinit(struct proc *p)
1620 struct filedesc *newfdp;
1621 struct filedesc *fdp = p->p_fd;
1623 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO);
1624 spin_lock_rd(&fdp->fd_spin);
1626 newfdp->fd_cdir = fdp->fd_cdir;
1627 vref(newfdp->fd_cdir);
1628 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1632 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of
1633 * proc0, but should unconditionally exist in other processes.
1636 newfdp->fd_rdir = fdp->fd_rdir;
1637 vref(newfdp->fd_rdir);
1638 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1641 newfdp->fd_jdir = fdp->fd_jdir;
1642 vref(newfdp->fd_jdir);
1643 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1645 spin_unlock_rd(&fdp->fd_spin);
1647 /* Create the file descriptor table. */
1648 newfdp->fd_refcnt = 1;
1649 newfdp->fd_cmask = cmask;
1650 newfdp->fd_files = newfdp->fd_builtin_files;
1651 newfdp->fd_nfiles = NDFILE;
1652 newfdp->fd_knlistsize = -1;
1653 newfdp->fd_lastfile = -1;
1654 spin_init(&newfdp->fd_spin);
1660 * Share a filedesc structure.
1665 fdshare(struct proc *p)
1667 struct filedesc *fdp;
1670 spin_lock_wr(&fdp->fd_spin);
1672 spin_unlock_wr(&fdp->fd_spin);
1677 * Copy a filedesc structure.
1682 fdcopy(struct proc *p)
1684 struct filedesc *fdp = p->p_fd;
1685 struct filedesc *newfdp;
1686 struct fdnode *fdnode;
1691 * Certain daemons might not have file descriptors.
1697 * Allocate the new filedesc and fd_files[] array. This can race
1698 * with operations by other threads on the fdp so we have to be
1701 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK | M_ZERO);
1703 spin_lock_rd(&fdp->fd_spin);
1704 if (fdp->fd_lastfile < NDFILE) {
1705 newfdp->fd_files = newfdp->fd_builtin_files;
1709 * We have to allocate (N^2-1) entries for our in-place
1710 * binary tree. Allow the table to shrink.
1714 while (ni > fdp->fd_lastfile && ni > NDFILE) {
1718 spin_unlock_rd(&fdp->fd_spin);
1719 newfdp->fd_files = kmalloc(i * sizeof(struct fdnode),
1720 M_FILEDESC, M_WAITOK | M_ZERO);
1723 * Check for race, retry
1725 spin_lock_rd(&fdp->fd_spin);
1726 if (i <= fdp->fd_lastfile) {
1727 spin_unlock_rd(&fdp->fd_spin);
1728 kfree(newfdp->fd_files, M_FILEDESC);
1734 * Dup the remaining fields. vref() and cache_hold() can be
1735 * safely called while holding the read spinlock on fdp.
1737 * The read spinlock on fdp is still being held.
1739 * NOTE: vref and cache_hold calls for the case where the vnode
1740 * or cache entry already has at least one ref may be called
1741 * while holding spin locks.
1743 if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) {
1744 vref(newfdp->fd_cdir);
1745 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1748 * We must check for fd_rdir here, at least for now because
1749 * the init process is created before we have access to the
1750 * rootvode to take a reference to it.
1752 if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) {
1753 vref(newfdp->fd_rdir);
1754 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1756 if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) {
1757 vref(newfdp->fd_jdir);
1758 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1760 newfdp->fd_refcnt = 1;
1761 newfdp->fd_nfiles = i;
1762 newfdp->fd_lastfile = fdp->fd_lastfile;
1763 newfdp->fd_freefile = fdp->fd_freefile;
1764 newfdp->fd_cmask = fdp->fd_cmask;
1765 newfdp->fd_knlist = NULL;
1766 newfdp->fd_knlistsize = -1;
1767 newfdp->fd_knhash = NULL;
1768 newfdp->fd_knhashmask = 0;
1769 spin_init(&newfdp->fd_spin);
1772 * Copy the descriptor table through (i). This also copies the
1773 * allocation state. Then go through and ref the file pointers
1774 * and clean up any KQ descriptors.
1776 * kq descriptors cannot be copied. Since we haven't ref'd the
1777 * copied files yet we can ignore the return value from funsetfd().
1779 * The read spinlock on fdp is still being held.
1781 bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode));
1782 for (i = 0 ; i < newfdp->fd_nfiles; ++i) {
1783 fdnode = &newfdp->fd_files[i];
1784 if (fdnode->reserved) {
1785 fdreserve_locked(newfdp, i, -1);
1786 fdnode->reserved = 0;
1787 fdfixup_locked(newfdp, i);
1788 } else if (fdnode->fp) {
1789 if (fdnode->fp->f_type == DTYPE_KQUEUE) {
1790 (void)funsetfd_locked(newfdp, i);
1796 spin_unlock_rd(&fdp->fd_spin);
1801 * Release a filedesc structure.
1803 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE)
1806 fdfree(struct proc *p, struct filedesc *repl)
1808 struct filedesc *fdp;
1809 struct fdnode *fdnode;
1811 struct filedesc_to_leader *fdtol;
1817 * Certain daemons might not have file descriptors.
1826 * Severe messing around to follow.
1828 spin_lock_wr(&fdp->fd_spin);
1830 /* Check for special need to clear POSIX style locks */
1832 if (fdtol != NULL) {
1833 KASSERT(fdtol->fdl_refcount > 0,
1834 ("filedesc_to_refcount botch: fdl_refcount=%d",
1835 fdtol->fdl_refcount));
1836 if (fdtol->fdl_refcount == 1 &&
1837 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1838 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1839 fdnode = &fdp->fd_files[i];
1840 if (fdnode->fp == NULL ||
1841 fdnode->fp->f_type != DTYPE_VNODE) {
1846 spin_unlock_wr(&fdp->fd_spin);
1848 lf.l_whence = SEEK_SET;
1851 lf.l_type = F_UNLCK;
1852 vp = (struct vnode *)fp->f_data;
1853 (void) VOP_ADVLOCK(vp,
1854 (caddr_t)p->p_leader,
1859 spin_lock_wr(&fdp->fd_spin);
1863 if (fdtol->fdl_refcount == 1) {
1864 if (fdp->fd_holdleaderscount > 0 &&
1865 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1867 * close() or do_dup() has cleared a reference
1868 * in a shared file descriptor table.
1870 fdp->fd_holdleaderswakeup = 1;
1871 ssleep(&fdp->fd_holdleaderscount,
1872 &fdp->fd_spin, 0, "fdlhold", 0);
1875 if (fdtol->fdl_holdcount > 0) {
1877 * Ensure that fdtol->fdl_leader
1878 * remains valid in closef().
1880 fdtol->fdl_wakeup = 1;
1881 ssleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0);
1885 fdtol->fdl_refcount--;
1886 if (fdtol->fdl_refcount == 0 &&
1887 fdtol->fdl_holdcount == 0) {
1888 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1889 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1894 if (fdtol != NULL) {
1895 spin_unlock_wr(&fdp->fd_spin);
1896 kfree(fdtol, M_FILEDESC_TO_LEADER);
1897 spin_lock_wr(&fdp->fd_spin);
1900 if (--fdp->fd_refcnt > 0) {
1901 spin_unlock_wr(&fdp->fd_spin);
1902 spin_lock_wr(&p->p_spin);
1904 spin_unlock_wr(&p->p_spin);
1909 * Even though we are the last reference to the structure allproc
1910 * scans may still reference the structure. Maintain proper
1911 * locks until we can replace p->p_fd.
1913 * Also note that kqueue's closef still needs to reference the
1914 * fdp via p->p_fd, so we have to close the descriptors before
1915 * we replace p->p_fd.
1917 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1918 if (fdp->fd_files[i].fp) {
1919 fp = funsetfd_locked(fdp, i);
1921 spin_unlock_wr(&fdp->fd_spin);
1923 spin_lock_wr(&fdp->fd_spin);
1927 spin_unlock_wr(&fdp->fd_spin);
1930 * Interlock against an allproc scan operations (typically frevoke).
1932 spin_lock_wr(&p->p_spin);
1934 spin_unlock_wr(&p->p_spin);
1937 * Wait for any softrefs to go away. This race rarely occurs so
1938 * we can use a non-critical-path style poll/sleep loop. The
1939 * race only occurs against allproc scans.
1941 * No new softrefs can occur with the fdp disconnected from the
1944 if (fdp->fd_softrefs) {
1945 kprintf("pid %d: Warning, fdp race avoided\n", p->p_pid);
1946 while (fdp->fd_softrefs)
1947 tsleep(&fdp->fd_softrefs, 0, "fdsoft", 1);
1950 if (fdp->fd_files != fdp->fd_builtin_files)
1951 kfree(fdp->fd_files, M_FILEDESC);
1953 cache_drop(&fdp->fd_ncdir);
1954 vrele(fdp->fd_cdir);
1957 cache_drop(&fdp->fd_nrdir);
1958 vrele(fdp->fd_rdir);
1961 cache_drop(&fdp->fd_njdir);
1962 vrele(fdp->fd_jdir);
1965 kfree(fdp->fd_knlist, M_KQUEUE);
1967 kfree(fdp->fd_knhash, M_KQUEUE);
1968 kfree(fdp, M_FILEDESC);
1972 * Retrieve and reference the file pointer associated with a descriptor.
1977 holdfp(struct filedesc *fdp, int fd, int flag)
1981 spin_lock_rd(&fdp->fd_spin);
1982 if (((u_int)fd) >= fdp->fd_nfiles) {
1986 if ((fp = fdp->fd_files[fd].fp) == NULL)
1988 if ((fp->f_flag & flag) == 0 && flag != -1) {
1994 spin_unlock_rd(&fdp->fd_spin);
1999 * holdsock() - load the struct file pointer associated
2000 * with a socket into *fpp. If an error occurs, non-zero
2001 * will be returned and *fpp will be set to NULL.
2006 holdsock(struct filedesc *fdp, int fd, struct file **fpp)
2011 spin_lock_rd(&fdp->fd_spin);
2012 if ((unsigned)fd >= fdp->fd_nfiles) {
2017 if ((fp = fdp->fd_files[fd].fp) == NULL) {
2021 if (fp->f_type != DTYPE_SOCKET) {
2028 spin_unlock_rd(&fdp->fd_spin);
2034 * Convert a user file descriptor to a held file pointer.
2039 holdvnode(struct filedesc *fdp, int fd, struct file **fpp)
2044 spin_lock_rd(&fdp->fd_spin);
2045 if ((unsigned)fd >= fdp->fd_nfiles) {
2050 if ((fp = fdp->fd_files[fd].fp) == NULL) {
2054 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
2062 spin_unlock_rd(&fdp->fd_spin);
2068 * For setugid programs, we don't want to people to use that setugidness
2069 * to generate error messages which write to a file which otherwise would
2070 * otherwise be off-limits to the process.
2072 * This is a gross hack to plug the hole. A better solution would involve
2073 * a special vop or other form of generalized access control mechanism. We
2074 * go ahead and just reject all procfs file systems accesses as dangerous.
2076 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2077 * sufficient. We also don't for check setugidness since we know we are.
2080 is_unsafe(struct file *fp)
2082 if (fp->f_type == DTYPE_VNODE &&
2083 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS)
2089 * Make this setguid thing safe, if at all possible.
2091 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2094 setugidsafety(struct proc *p)
2096 struct filedesc *fdp = p->p_fd;
2099 /* Certain daemons might not have file descriptors. */
2104 * note: fdp->fd_files may be reallocated out from under us while
2105 * we are blocked in a close. Be careful!
2107 for (i = 0; i <= fdp->fd_lastfile; i++) {
2110 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) {
2113 if (i < fdp->fd_knlistsize)
2114 knote_fdclose(p, i);
2116 * NULL-out descriptor prior to close to avoid
2117 * a race while close blocks.
2119 if ((fp = funsetfd_locked(fdp, i)) != NULL)
2126 * Close any files on exec?
2128 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2131 fdcloseexec(struct proc *p)
2133 struct filedesc *fdp = p->p_fd;
2136 /* Certain daemons might not have file descriptors. */
2141 * We cannot cache fd_files since operations may block and rip
2142 * them out from under us.
2144 for (i = 0; i <= fdp->fd_lastfile; i++) {
2145 if (fdp->fd_files[i].fp != NULL &&
2146 (fdp->fd_files[i].fileflags & UF_EXCLOSE)) {
2149 if (i < fdp->fd_knlistsize)
2150 knote_fdclose(p, i);
2152 * NULL-out descriptor prior to close to avoid
2153 * a race while close blocks.
2155 if ((fp = funsetfd_locked(fdp, i)) != NULL)
2162 * It is unsafe for set[ug]id processes to be started with file
2163 * descriptors 0..2 closed, as these descriptors are given implicit
2164 * significance in the Standard C library. fdcheckstd() will create a
2165 * descriptor referencing /dev/null for each of stdin, stdout, and
2166 * stderr that is not already open.
2168 * NOT MPSAFE - calls falloc, vn_open, etc
2171 fdcheckstd(struct lwp *lp)
2173 struct nlookupdata nd;
2174 struct filedesc *fdp;
2177 int i, error, flags, devnull;
2179 fdp = lp->lwp_proc->p_fd;
2184 for (i = 0; i < 3; i++) {
2185 if (fdp->fd_files[i].fp != NULL)
2188 if ((error = falloc(lp, &fp, &devnull)) != 0)
2191 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE,
2192 NLC_FOLLOW|NLC_LOCKVP);
2193 flags = FREAD | FWRITE;
2195 error = vn_open(&nd, fp, flags, 0);
2197 fsetfd(fdp, fp, devnull);
2199 fsetfd(fdp, NULL, devnull);
2204 KKASSERT(i == devnull);
2206 error = kern_dup(DUP_FIXED, devnull, i, &retval);
2215 * Internal form of close.
2216 * Decrement reference count on file structure.
2217 * Note: td and/or p may be NULL when closing a file
2218 * that was being passed in a message.
2220 * MPALMOSTSAFE - acquires mplock for VOP operations
2223 closef(struct file *fp, struct proc *p)
2227 struct filedesc_to_leader *fdtol;
2233 * POSIX record locking dictates that any close releases ALL
2234 * locks owned by this process. This is handled by setting
2235 * a flag in the unlock to free ONLY locks obeying POSIX
2236 * semantics, and not to free BSD-style file locks.
2237 * If the descriptor was in a message, POSIX-style locks
2238 * aren't passed with the descriptor.
2240 if (p != NULL && fp->f_type == DTYPE_VNODE &&
2241 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2244 if ((p->p_leader->p_flag & P_ADVLOCK) != 0) {
2245 lf.l_whence = SEEK_SET;
2248 lf.l_type = F_UNLCK;
2249 vp = (struct vnode *)fp->f_data;
2250 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
2254 if (fdtol != NULL) {
2256 * Handle special case where file descriptor table
2257 * is shared between multiple process leaders.
2259 for (fdtol = fdtol->fdl_next;
2260 fdtol != p->p_fdtol;
2261 fdtol = fdtol->fdl_next) {
2262 if ((fdtol->fdl_leader->p_flag &
2265 fdtol->fdl_holdcount++;
2266 lf.l_whence = SEEK_SET;
2269 lf.l_type = F_UNLCK;
2270 vp = (struct vnode *)fp->f_data;
2271 (void) VOP_ADVLOCK(vp,
2272 (caddr_t)fdtol->fdl_leader,
2273 F_UNLCK, &lf, F_POSIX);
2274 fdtol->fdl_holdcount--;
2275 if (fdtol->fdl_holdcount == 0 &&
2276 fdtol->fdl_wakeup != 0) {
2277 fdtol->fdl_wakeup = 0;
2290 * fhold() can only be called if f_count is already at least 1 (i.e. the
2291 * caller of fhold() already has a reference to the file pointer in some
2294 * f_count is not spin-locked. Instead, atomic ops are used for
2295 * incrementing, decrementing, and handling the 1->0 transition.
2298 fhold(struct file *fp)
2300 atomic_add_int(&fp->f_count, 1);
2304 * fdrop() - drop a reference to a descriptor
2306 * MPALMOSTSAFE - acquires mplock for final close sequence
2309 fdrop(struct file *fp)
2316 * A combined fetch and subtract is needed to properly detect
2317 * 1->0 transitions, otherwise two cpus dropping from a ref
2318 * count of 2 might both try to run the 1->0 code.
2320 if (atomic_fetchadd_int(&fp->f_count, -1) > 1)
2326 * The last reference has gone away, we own the fp structure free
2329 if (fp->f_count < 0)
2330 panic("fdrop: count < 0");
2331 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE &&
2332 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2334 lf.l_whence = SEEK_SET;
2337 lf.l_type = F_UNLCK;
2338 vp = (struct vnode *)fp->f_data;
2339 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2341 if (fp->f_ops != &badfileops)
2342 error = fo_close(fp);
2351 * Apply an advisory lock on a file descriptor.
2353 * Just attempt to get a record lock of the requested type on
2354 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2359 sys_flock(struct flock_args *uap)
2361 struct proc *p = curproc;
2367 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
2370 if (fp->f_type != DTYPE_VNODE) {
2374 vp = (struct vnode *)fp->f_data;
2375 lf.l_whence = SEEK_SET;
2378 if (uap->how & LOCK_UN) {
2379 lf.l_type = F_UNLCK;
2380 fp->f_flag &= ~FHASLOCK;
2381 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2384 if (uap->how & LOCK_EX)
2385 lf.l_type = F_WRLCK;
2386 else if (uap->how & LOCK_SH)
2387 lf.l_type = F_RDLCK;
2392 fp->f_flag |= FHASLOCK;
2393 if (uap->how & LOCK_NB)
2394 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0);
2396 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT);
2404 * File Descriptor pseudo-device driver (/dev/fd/).
2406 * Opening minor device N dup()s the file (if any) connected to file
2407 * descriptor N belonging to the calling process. Note that this driver
2408 * consists of only the ``open()'' routine, because all subsequent
2409 * references to this file will be direct to the other driver.
2412 fdopen(struct dev_open_args *ap)
2414 thread_t td = curthread;
2416 KKASSERT(td->td_lwp != NULL);
2419 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the
2420 * the file descriptor being sought for duplication. The error
2421 * return ensures that the vnode for this device will be released
2422 * by vn_open. Open will detect this special error and take the
2423 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2424 * will simply report the error.
2426 td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev);
2431 * The caller has reserved the file descriptor dfd for us. On success we
2432 * must fsetfd() it. On failure the caller will clean it up.
2434 * NOT MPSAFE - isn't getting spinlocks, possibly other things
2437 dupfdopen(struct filedesc *fdp, int dfd, int sfd, int mode, int error)
2443 if ((wfp = holdfp(fdp, sfd, -1)) == NULL)
2447 * Close a revoke/dup race. Duping a descriptor marked as revoked
2448 * will dup a dummy descriptor instead of the real one.
2450 if (wfp->f_flag & FREVOKED) {
2451 kprintf("Warning: attempt to dup() a revoked descriptor\n");
2454 werror = falloc(NULL, &wfp, NULL);
2460 * There are two cases of interest here.
2462 * For ENODEV simply dup sfd to file descriptor dfd and return.
2464 * For ENXIO steal away the file structure from sfd and store it
2465 * dfd. sfd is effectively closed by this operation.
2467 * Any other error code is just returned.
2472 * Check that the mode the file is being opened for is a
2473 * subset of the mode of the existing descriptor.
2475 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2479 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2480 fsetfd(fdp, wfp, dfd);
2485 * Steal away the file pointer from dfd, and stuff it into indx.
2487 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2488 fsetfd(fdp, wfp, dfd);
2489 if ((xfp = funsetfd_locked(fdp, sfd)) != NULL)
2501 * NOT MPSAFE - I think these refer to a common file descriptor table
2502 * and we need to spinlock that to link fdtol in.
2504 struct filedesc_to_leader *
2505 filedesc_to_leader_alloc(struct filedesc_to_leader *old,
2506 struct proc *leader)
2508 struct filedesc_to_leader *fdtol;
2510 fdtol = kmalloc(sizeof(struct filedesc_to_leader),
2511 M_FILEDESC_TO_LEADER, M_WAITOK);
2512 fdtol->fdl_refcount = 1;
2513 fdtol->fdl_holdcount = 0;
2514 fdtol->fdl_wakeup = 0;
2515 fdtol->fdl_leader = leader;
2517 fdtol->fdl_next = old->fdl_next;
2518 fdtol->fdl_prev = old;
2519 old->fdl_next = fdtol;
2520 fdtol->fdl_next->fdl_prev = fdtol;
2522 fdtol->fdl_next = fdtol;
2523 fdtol->fdl_prev = fdtol;
2529 * Scan all file pointers in the system. The callback is made with
2530 * the master list spinlock held exclusively.
2535 allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data)
2540 spin_lock_wr(&filehead_spin);
2541 LIST_FOREACH(fp, &filehead, f_list) {
2542 res = callback(fp, data);
2546 spin_unlock_wr(&filehead_spin);
2550 * Get file structures.
2552 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe)
2555 struct sysctl_kern_file_info {
2558 struct sysctl_req *req;
2561 static int sysctl_kern_file_callback(struct proc *p, void *data);
2564 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2566 struct sysctl_kern_file_info info;
2569 * Note: because the number of file descriptors is calculated
2570 * in different ways for sizing vs returning the data,
2571 * there is information leakage from the first loop. However,
2572 * it is of a similar order of magnitude to the leakage from
2573 * global system statistics such as kern.openfiles.
2575 * When just doing a count, note that we cannot just count
2576 * the elements and add f_count via the filehead list because
2577 * threaded processes share their descriptor table and f_count might
2578 * still be '1' in that case.
2580 * Since the SYSCTL op can block, we must hold the process to
2581 * prevent it being ripped out from under us either in the
2582 * file descriptor loop or in the greater LIST_FOREACH. The
2583 * process may be in varying states of disrepair. If the process
2584 * is in SZOMB we may have caught it just as it is being removed
2585 * from the allproc list, we must skip it in that case to maintain
2586 * an unbroken chain through the allproc list.
2591 allproc_scan(sysctl_kern_file_callback, &info);
2594 * When just calculating the size, overestimate a bit to try to
2595 * prevent system activity from causing the buffer-fill call
2598 if (req->oldptr == NULL) {
2599 info.count = (info.count + 16) + (info.count / 10);
2600 info.error = SYSCTL_OUT(req, NULL,
2601 info.count * sizeof(struct kinfo_file));
2603 return (info.error);
2607 sysctl_kern_file_callback(struct proc *p, void *data)
2609 struct sysctl_kern_file_info *info = data;
2610 struct kinfo_file kf;
2611 struct filedesc *fdp;
2616 if (p->p_stat == SIDL || p->p_stat == SZOMB)
2618 if (!PRISON_CHECK(info->req->td->td_ucred, p->p_ucred) != 0)
2622 * Softref the fdp to prevent it from being destroyed
2624 spin_lock_wr(&p->p_spin);
2625 if ((fdp = p->p_fd) == NULL) {
2626 spin_unlock_wr(&p->p_spin);
2629 atomic_add_int(&fdp->fd_softrefs, 1);
2630 spin_unlock_wr(&p->p_spin);
2633 * The fdp's own spinlock prevents the contents from being
2636 spin_lock_rd(&fdp->fd_spin);
2637 for (n = 0; n < fdp->fd_nfiles; ++n) {
2638 if ((fp = fdp->fd_files[n].fp) == NULL)
2640 if (info->req->oldptr == NULL) {
2643 uid = p->p_ucred ? p->p_ucred->cr_uid : -1;
2644 kcore_make_file(&kf, fp, p->p_pid, uid, n);
2645 spin_unlock_rd(&fdp->fd_spin);
2646 info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf));
2647 spin_lock_rd(&fdp->fd_spin);
2652 spin_unlock_rd(&fdp->fd_spin);
2653 atomic_subtract_int(&fdp->fd_softrefs, 1);
2659 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2660 0, 0, sysctl_kern_file, "S,file", "Entire file table");
2662 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2663 &maxfilesperproc, 0, "Maximum files allowed open per process");
2665 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2666 &maxfiles, 0, "Maximum number of files");
2668 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW,
2669 &maxfilesrootres, 0, "Descriptors reserved for root use");
2671 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2672 &nfiles, 0, "System-wide number of open files");
2675 fildesc_drvinit(void *unused)
2679 for (fd = 0; fd < NUMFDESC; fd++) {
2680 make_dev(&fildesc_ops, fd,
2681 UID_BIN, GID_BIN, 0666, "fd/%d", fd);
2684 make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin");
2685 make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout");
2686 make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr");
2692 struct fileops badfileops = {
2693 .fo_read = badfo_readwrite,
2694 .fo_write = badfo_readwrite,
2695 .fo_ioctl = badfo_ioctl,
2696 .fo_poll = badfo_poll,
2697 .fo_kqfilter = badfo_kqfilter,
2698 .fo_stat = badfo_stat,
2699 .fo_close = badfo_close,
2700 .fo_shutdown = badfo_shutdown
2720 badfo_ioctl(struct file *fp, u_long com, caddr_t data,
2721 struct ucred *cred, struct sysmsg *msgv)
2730 badfo_poll(struct file *fp, int events, struct ucred *cred)
2739 badfo_kqfilter(struct file *fp, struct knote *kn)
2745 badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
2754 badfo_close(struct file *fp)
2763 badfo_shutdown(struct file *fp, int how)
2772 nofo_shutdown(struct file *fp, int how)
2774 return (EOPNOTSUPP);
2777 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,
2778 fildesc_drvinit,NULL)