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 struct krate krate_uidinfo = { .freq = 1 };
124 static d_open_t fdopen;
127 #define CDEV_MAJOR 22
128 static struct dev_ops fildesc_ops = {
129 { "FD", CDEV_MAJOR, 0 },
134 * Descriptor management.
136 static struct filelist filehead = LIST_HEAD_INITIALIZER(&filehead);
137 static struct spinlock filehead_spin = SPINLOCK_INITIALIZER(&filehead_spin);
138 static int nfiles; /* actual number of open files */
142 * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared.
144 * MPSAFE - must be called with fdp->fd_spin exclusively held
148 fdfixup_locked(struct filedesc *fdp, int fd)
150 if (fd < fdp->fd_freefile) {
151 fdp->fd_freefile = fd;
153 while (fdp->fd_lastfile >= 0 &&
154 fdp->fd_files[fdp->fd_lastfile].fp == NULL &&
155 fdp->fd_files[fdp->fd_lastfile].reserved == 0
162 * System calls on descriptors.
167 sys_getdtablesize(struct getdtablesize_args *uap)
169 struct proc *p = curproc;
170 struct plimit *limit = p->p_limit;
173 spin_lock_rd(&limit->p_spin);
174 if (limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
177 dtsize = (int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur;
178 spin_unlock_rd(&limit->p_spin);
179 if (dtsize > maxfilesperproc)
180 dtsize = maxfilesperproc;
181 if (dtsize < minfilesperproc)
182 dtsize = minfilesperproc;
183 if (p->p_ucred->cr_uid && dtsize > maxfilesperuser)
184 dtsize = maxfilesperuser;
185 uap->sysmsg_result = dtsize;
190 * Duplicate a file descriptor to a particular value.
192 * note: keep in mind that a potential race condition exists when closing
193 * descriptors from a shared descriptor table (via rfork).
198 sys_dup2(struct dup2_args *uap)
203 error = kern_dup(DUP_FIXED, uap->from, uap->to, &fd);
204 uap->sysmsg_fds[0] = fd;
210 * Duplicate a file descriptor.
215 sys_dup(struct dup_args *uap)
220 error = kern_dup(DUP_VARIABLE, uap->fd, 0, &fd);
221 uap->sysmsg_fds[0] = fd;
227 * MPALMOSTSAFE - acquires mplock for fp operations
230 kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred)
232 struct thread *td = curthread;
233 struct proc *p = td->td_proc;
239 int tmp, error, flg = F_POSIX;
244 * Operations on file descriptors that do not require a file pointer.
248 error = fgetfdflags(p->p_fd, fd, &tmp);
250 dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0;
254 if (dat->fc_cloexec & FD_CLOEXEC)
255 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
257 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
261 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd);
268 * Operations on file pointers
270 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
276 dat->fc_flags = OFLAGS(fp->f_flag);
282 nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS;
283 nflags |= oflags & ~FCNTLFLAGS;
286 if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY))
288 if (error == 0 && ((nflags ^ oflags) & FASYNC)) {
289 tmp = nflags & FASYNC;
290 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp,
298 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner,
303 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner,
309 /* Fall into F_SETLK */
312 if (fp->f_type != DTYPE_VNODE) {
316 vp = (struct vnode *)fp->f_data;
319 * copyin/lockop may block
321 if (dat->fc_flock.l_whence == SEEK_CUR)
322 dat->fc_flock.l_start += fp->f_offset;
324 switch (dat->fc_flock.l_type) {
326 if ((fp->f_flag & FREAD) == 0) {
330 p->p_leader->p_flag |= P_ADVLOCK;
331 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
332 &dat->fc_flock, flg);
335 if ((fp->f_flag & FWRITE) == 0) {
339 p->p_leader->p_flag |= P_ADVLOCK;
340 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
341 &dat->fc_flock, flg);
344 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
345 &dat->fc_flock, F_POSIX);
353 * It is possible to race a close() on the descriptor while
354 * we were blocked getting the lock. If this occurs the
355 * close might not have caught the lock.
357 if (checkfpclosed(p->p_fd, fd, fp)) {
358 dat->fc_flock.l_whence = SEEK_SET;
359 dat->fc_flock.l_start = 0;
360 dat->fc_flock.l_len = 0;
361 dat->fc_flock.l_type = F_UNLCK;
362 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
363 F_UNLCK, &dat->fc_flock, F_POSIX);
368 if (fp->f_type != DTYPE_VNODE) {
372 vp = (struct vnode *)fp->f_data;
374 * copyin/lockop may block
376 if (dat->fc_flock.l_type != F_RDLCK &&
377 dat->fc_flock.l_type != F_WRLCK &&
378 dat->fc_flock.l_type != F_UNLCK) {
382 if (dat->fc_flock.l_whence == SEEK_CUR)
383 dat->fc_flock.l_start += fp->f_offset;
384 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK,
385 &dat->fc_flock, F_POSIX);
398 * The file control system call.
403 sys_fcntl(struct fcntl_args *uap)
410 dat.fc_fd = uap->arg;
413 dat.fc_cloexec = uap->arg;
416 dat.fc_flags = uap->arg;
419 dat.fc_owner = uap->arg;
424 error = copyin((caddr_t)uap->arg, &dat.fc_flock,
425 sizeof(struct flock));
431 error = kern_fcntl(uap->fd, uap->cmd, &dat, curthread->td_ucred);
436 uap->sysmsg_result = dat.fc_fd;
439 uap->sysmsg_result = dat.fc_cloexec;
442 uap->sysmsg_result = dat.fc_flags;
445 uap->sysmsg_result = dat.fc_owner;
447 error = copyout(&dat.fc_flock, (caddr_t)uap->arg,
448 sizeof(struct flock));
457 * Common code for dup, dup2, and fcntl(F_DUPFD).
459 * The type flag can be either DUP_FIXED or DUP_VARIABLE. DUP_FIXED tells
460 * kern_dup() to destructively dup over an existing file descriptor if new
461 * is already open. DUP_VARIABLE tells kern_dup() to find the lowest
462 * unused file descriptor that is greater than or equal to new.
467 kern_dup(enum dup_type type, int old, int new, int *res)
469 struct thread *td = curthread;
470 struct proc *p = td->td_proc;
471 struct filedesc *fdp = p->p_fd;
480 * Verify that we have a valid descriptor to dup from and
481 * possibly to dup to.
483 * NOTE: maxfilesperuser is not applicable to dup()
486 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
489 dtsize = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
490 if (dtsize > maxfilesperproc)
491 dtsize = maxfilesperproc;
492 if (dtsize < minfilesperproc)
493 dtsize = minfilesperproc;
495 if (new < 0 || new > dtsize)
498 spin_lock_wr(&fdp->fd_spin);
499 if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) {
500 spin_unlock_wr(&fdp->fd_spin);
503 if (type == DUP_FIXED && old == new) {
505 spin_unlock_wr(&fdp->fd_spin);
508 fp = fdp->fd_files[old].fp;
509 oldflags = fdp->fd_files[old].fileflags;
510 fhold(fp); /* MPSAFE - can be called with a spinlock held */
513 * Allocate a new descriptor if DUP_VARIABLE, or expand the table
514 * if the requested descriptor is beyond the current table size.
516 * This can block. Retry if the source descriptor no longer matches
517 * or if our expectation in the expansion case races.
519 * If we are not expanding or allocating a new decriptor, then reset
520 * the target descriptor to a reserved state so we have a uniform
521 * setup for the next code block.
523 if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) {
524 spin_unlock_wr(&fdp->fd_spin);
525 error = fdalloc(p, new, &newfd);
526 spin_lock_wr(&fdp->fd_spin);
528 spin_unlock_wr(&fdp->fd_spin);
535 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) {
536 fsetfd_locked(fdp, NULL, newfd);
537 spin_unlock_wr(&fdp->fd_spin);
542 * Check for expansion race
544 if (type != DUP_VARIABLE && new != newfd) {
545 fsetfd_locked(fdp, NULL, newfd);
546 spin_unlock_wr(&fdp->fd_spin);
551 * Check for ripout, newfd reused old (this case probably
555 fsetfd_locked(fdp, NULL, newfd);
556 spin_unlock_wr(&fdp->fd_spin);
563 if (fdp->fd_files[new].reserved) {
564 spin_unlock_wr(&fdp->fd_spin);
566 kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new);
567 tsleep(fdp, 0, "fdres", hz);
572 * If the target descriptor was never allocated we have
573 * to allocate it. If it was we have to clean out the
574 * old descriptor. delfp inherits the ref from the
577 delfp = fdp->fd_files[new].fp;
578 fdp->fd_files[new].fp = NULL;
579 fdp->fd_files[new].reserved = 1;
581 fdreserve_locked(fdp, new, 1);
582 if (new > fdp->fd_lastfile)
583 fdp->fd_lastfile = new;
589 * NOTE: still holding an exclusive spinlock
593 * If a descriptor is being overwritten we may hve to tell
594 * fdfree() to sleep to ensure that all relevant process
595 * leaders can be traversed in closef().
597 if (delfp != NULL && p->p_fdtol != NULL) {
598 fdp->fd_holdleaderscount++;
603 KASSERT(delfp == NULL || type == DUP_FIXED,
604 ("dup() picked an open file"));
607 * Duplicate the source descriptor, update lastfile. If the new
608 * descriptor was not allocated and we aren't replacing an existing
609 * descriptor we have to mark the descriptor as being in use.
611 * The fd_files[] array inherits fp's hold reference.
613 fsetfd_locked(fdp, fp, new);
614 fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE;
615 spin_unlock_wr(&fdp->fd_spin);
620 * If we dup'd over a valid file, we now own the reference to it
621 * and must dispose of it using closef() semantics (as if a
622 * close() were performed on it).
627 spin_lock_wr(&fdp->fd_spin);
628 fdp->fd_holdleaderscount--;
629 if (fdp->fd_holdleaderscount == 0 &&
630 fdp->fd_holdleaderswakeup != 0) {
631 fdp->fd_holdleaderswakeup = 0;
632 spin_unlock_wr(&fdp->fd_spin);
633 wakeup(&fdp->fd_holdleaderscount);
635 spin_unlock_wr(&fdp->fd_spin);
643 * If sigio is on the list associated with a process or process group,
644 * disable signalling from the device, remove sigio from the list and
648 funsetown(struct sigio *sigio)
653 *(sigio->sio_myref) = NULL;
655 if (sigio->sio_pgid < 0) {
656 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
658 } else /* if ((*sigiop)->sio_pgid > 0) */ {
659 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
662 crfree(sigio->sio_ucred);
663 kfree(sigio, M_SIGIO);
666 /* Free a list of sigio structures. */
668 funsetownlst(struct sigiolst *sigiolst)
672 while ((sigio = SLIST_FIRST(sigiolst)) != NULL)
677 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
679 * After permission checking, add a sigio structure to the sigio list for
680 * the process or process group.
683 fsetown(pid_t pgid, struct sigio **sigiop)
699 * Policy - Don't allow a process to FSETOWN a process
700 * in another session.
702 * Remove this test to allow maximum flexibility or
703 * restrict FSETOWN to the current process or process
704 * group for maximum safety.
706 if (proc->p_session != curproc->p_session)
710 } else /* if (pgid < 0) */ {
711 pgrp = pgfind(-pgid);
716 * Policy - Don't allow a process to FSETOWN a process
717 * in another session.
719 * Remove this test to allow maximum flexibility or
720 * restrict FSETOWN to the current process or process
721 * group for maximum safety.
723 if (pgrp->pg_session != curproc->p_session)
729 sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
731 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
732 sigio->sio_proc = proc;
734 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
735 sigio->sio_pgrp = pgrp;
737 sigio->sio_pgid = pgid;
738 sigio->sio_ucred = crhold(curthread->td_ucred);
739 /* It would be convenient if p_ruid was in ucred. */
740 sigio->sio_ruid = sigio->sio_ucred->cr_ruid;
741 sigio->sio_myref = sigiop;
749 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
752 fgetown(struct sigio *sigio)
754 return (sigio != NULL ? sigio->sio_pgid : 0);
758 * Close many file descriptors.
763 sys_closefrom(struct closefrom_args *uap)
765 return(kern_closefrom(uap->fd));
769 * Close all file descriptors greater then or equal to fd
774 kern_closefrom(int fd)
776 struct thread *td = curthread;
777 struct proc *p = td->td_proc;
778 struct filedesc *fdp;
787 * NOTE: This function will skip unassociated descriptors and
788 * reserved descriptors that have not yet been assigned.
789 * fd_lastfile can change as a side effect of kern_close().
791 spin_lock_wr(&fdp->fd_spin);
792 while (fd <= fdp->fd_lastfile) {
793 if (fdp->fd_files[fd].fp != NULL) {
794 spin_unlock_wr(&fdp->fd_spin);
795 /* ok if this races another close */
796 if (kern_close(fd) == EINTR)
798 spin_lock_wr(&fdp->fd_spin);
802 spin_unlock_wr(&fdp->fd_spin);
807 * Close a file descriptor.
812 sys_close(struct close_args *uap)
814 return(kern_close(uap->fd));
818 * MPALMOSTSAFE - acquires mplock around knote_fdclose() calls
823 struct thread *td = curthread;
824 struct proc *p = td->td_proc;
825 struct filedesc *fdp;
833 spin_lock_wr(&fdp->fd_spin);
834 if ((fp = funsetfd_locked(fdp, fd)) == NULL) {
835 spin_unlock_wr(&fdp->fd_spin);
839 if (p->p_fdtol != NULL) {
841 * Ask fdfree() to sleep to ensure that all relevant
842 * process leaders can be traversed in closef().
844 fdp->fd_holdleaderscount++;
849 * we now hold the fp reference that used to be owned by the descriptor
852 spin_unlock_wr(&fdp->fd_spin);
853 if (SLIST_FIRST(&fp->f_klist)) {
855 knote_fdclose(fp, fdp, fd);
858 error = closef(fp, p);
860 spin_lock_wr(&fdp->fd_spin);
861 fdp->fd_holdleaderscount--;
862 if (fdp->fd_holdleaderscount == 0 &&
863 fdp->fd_holdleaderswakeup != 0) {
864 fdp->fd_holdleaderswakeup = 0;
865 spin_unlock_wr(&fdp->fd_spin);
866 wakeup(&fdp->fd_holdleaderscount);
868 spin_unlock_wr(&fdp->fd_spin);
875 * shutdown_args(int fd, int how)
878 kern_shutdown(int fd, int how)
880 struct thread *td = curthread;
881 struct proc *p = td->td_proc;
887 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
889 error = fo_shutdown(fp, how);
899 sys_shutdown(struct shutdown_args *uap)
904 error = kern_shutdown(uap->s, uap->how);
914 kern_fstat(int fd, struct stat *ub)
916 struct thread *td = curthread;
917 struct proc *p = td->td_proc;
923 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
925 error = fo_stat(fp, ub, td->td_ucred);
932 * Return status information about a file descriptor.
937 sys_fstat(struct fstat_args *uap)
942 error = kern_fstat(uap->fd, &st);
945 error = copyout(&st, uap->sb, sizeof(st));
950 * Return pathconf information about a file descriptor.
955 sys_fpathconf(struct fpathconf_args *uap)
957 struct thread *td = curthread;
958 struct proc *p = td->td_proc;
963 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
966 switch (fp->f_type) {
969 if (uap->name != _PC_PIPE_BUF) {
972 uap->sysmsg_result = PIPE_BUF;
978 vp = (struct vnode *)fp->f_data;
980 error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg);
992 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand,
996 * Grow the file table so it can hold through descriptor (want).
998 * The fdp's spinlock must be held exclusively on entry and may be held
999 * exclusively on return. The spinlock may be cycled by the routine.
1004 fdgrow_locked(struct filedesc *fdp, int want)
1006 struct fdnode *newfiles;
1007 struct fdnode *oldfiles;
1010 nf = fdp->fd_nfiles;
1012 /* nf has to be of the form 2^n - 1 */
1014 } while (nf <= want);
1016 spin_unlock_wr(&fdp->fd_spin);
1017 newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK);
1018 spin_lock_wr(&fdp->fd_spin);
1021 * We could have raced another extend while we were not holding
1024 if (fdp->fd_nfiles >= nf) {
1025 spin_unlock_wr(&fdp->fd_spin);
1026 kfree(newfiles, M_FILEDESC);
1027 spin_lock_wr(&fdp->fd_spin);
1031 * Copy the existing ofile and ofileflags arrays
1032 * and zero the new portion of each array.
1034 extra = nf - fdp->fd_nfiles;
1035 bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode));
1036 bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode));
1038 oldfiles = fdp->fd_files;
1039 fdp->fd_files = newfiles;
1040 fdp->fd_nfiles = nf;
1042 if (oldfiles != fdp->fd_builtin_files) {
1043 spin_unlock_wr(&fdp->fd_spin);
1044 kfree(oldfiles, M_FILEDESC);
1045 spin_lock_wr(&fdp->fd_spin);
1051 * Number of nodes in right subtree, including the root.
1054 right_subtree_size(int n)
1056 return (n ^ (n | (n + 1)));
1063 right_ancestor(int n)
1065 return (n | (n + 1));
1072 left_ancestor(int n)
1074 return ((n & (n + 1)) - 1);
1078 * Traverse the in-place binary tree buttom-up adjusting the allocation
1079 * count so scans can determine where free descriptors are located.
1081 * MPSAFE - caller must be holding an exclusive spinlock on fdp
1085 fdreserve_locked(struct filedesc *fdp, int fd, int incr)
1088 fdp->fd_files[fd].allocated += incr;
1089 KKASSERT(fdp->fd_files[fd].allocated >= 0);
1090 fd = left_ancestor(fd);
1095 * Reserve a file descriptor for the process. If no error occurs, the
1096 * caller MUST at some point call fsetfd() or assign a file pointer
1097 * or dispose of the reservation.
1102 fdalloc(struct proc *p, int want, int *result)
1104 struct filedesc *fdp = p->p_fd;
1105 struct uidinfo *uip;
1106 int fd, rsize, rsum, node, lim;
1109 * Check dtable size limit
1111 spin_lock_rd(&p->p_limit->p_spin);
1112 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
1115 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
1116 spin_unlock_rd(&p->p_limit->p_spin);
1117 if (lim > maxfilesperproc)
1118 lim = maxfilesperproc;
1119 if (lim < minfilesperproc)
1120 lim = minfilesperproc;
1125 * Check that the user has not run out of descriptors (non-root only).
1126 * As a safety measure the dtable is allowed to have at least
1127 * minfilesperproc open fds regardless of the maxfilesperuser limit.
1129 if (p->p_ucred->cr_uid && fdp->fd_nfiles >= minfilesperproc) {
1130 uip = p->p_ucred->cr_uidinfo;
1131 if (uip->ui_openfiles > maxfilesperuser) {
1132 krateprintf(&krate_uidinfo,
1133 "Warning: user %d pid %d (%s) ran out of "
1134 "file descriptors (%d/%d)\n",
1135 p->p_ucred->cr_uid, (int)p->p_pid,
1137 uip->ui_openfiles, maxfilesperuser);
1143 * Grow the dtable if necessary
1145 spin_lock_wr(&fdp->fd_spin);
1146 if (want >= fdp->fd_nfiles)
1147 fdgrow_locked(fdp, want);
1150 * Search for a free descriptor starting at the higher
1151 * of want or fd_freefile. If that fails, consider
1152 * expanding the ofile array.
1154 * NOTE! the 'allocated' field is a cumulative recursive allocation
1155 * count. If we happen to see a value of 0 then we can shortcut
1156 * our search. Otherwise we run through through the tree going
1157 * down branches we know have free descriptor(s) until we hit a
1158 * leaf node. The leaf node will be free but will not necessarily
1159 * have an allocated field of 0.
1162 /* move up the tree looking for a subtree with a free node */
1163 for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim);
1164 fd = right_ancestor(fd)) {
1165 if (fdp->fd_files[fd].allocated == 0)
1168 rsize = right_subtree_size(fd);
1169 if (fdp->fd_files[fd].allocated == rsize)
1170 continue; /* right subtree full */
1173 * Free fd is in the right subtree of the tree rooted at fd.
1174 * Call that subtree R. Look for the smallest (leftmost)
1175 * subtree of R with an unallocated fd: continue moving
1176 * down the left branch until encountering a full left
1177 * subtree, then move to the right.
1179 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) {
1181 rsum += fdp->fd_files[node].allocated;
1182 if (fdp->fd_files[fd].allocated == rsum + rsize) {
1183 fd = node; /* move to the right */
1184 if (fdp->fd_files[node].allocated == 0)
1193 * No space in current array. Expand?
1195 if (fdp->fd_nfiles >= lim) {
1196 spin_unlock_wr(&fdp->fd_spin);
1199 fdgrow_locked(fdp, want);
1203 KKASSERT(fd < fdp->fd_nfiles);
1204 if (fd > fdp->fd_lastfile)
1205 fdp->fd_lastfile = fd;
1206 if (want <= fdp->fd_freefile)
1207 fdp->fd_freefile = fd;
1209 KKASSERT(fdp->fd_files[fd].fp == NULL);
1210 KKASSERT(fdp->fd_files[fd].reserved == 0);
1211 fdp->fd_files[fd].fileflags = 0;
1212 fdp->fd_files[fd].reserved = 1;
1213 fdreserve_locked(fdp, fd, 1);
1214 spin_unlock_wr(&fdp->fd_spin);
1219 * Check to see whether n user file descriptors
1220 * are available to the process p.
1225 fdavail(struct proc *p, int n)
1227 struct filedesc *fdp = p->p_fd;
1228 struct fdnode *fdnode;
1231 spin_lock_rd(&p->p_limit->p_spin);
1232 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
1235 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
1236 spin_unlock_rd(&p->p_limit->p_spin);
1237 if (lim > maxfilesperproc)
1238 lim = maxfilesperproc;
1239 if (lim < minfilesperproc)
1240 lim = minfilesperproc;
1242 spin_lock_rd(&fdp->fd_spin);
1243 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) {
1244 spin_unlock_rd(&fdp->fd_spin);
1247 last = min(fdp->fd_nfiles, lim);
1248 fdnode = &fdp->fd_files[fdp->fd_freefile];
1249 for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) {
1250 if (fdnode->fp == NULL && --n <= 0) {
1251 spin_unlock_rd(&fdp->fd_spin);
1255 spin_unlock_rd(&fdp->fd_spin);
1260 * Revoke open descriptors referencing (f_data, f_type)
1262 * Any revoke executed within a prison is only able to
1263 * revoke descriptors for processes within that prison.
1265 * Returns 0 on success or an error code.
1267 struct fdrevoke_info {
1277 static int fdrevoke_check_callback(struct file *fp, void *vinfo);
1278 static int fdrevoke_proc_callback(struct proc *p, void *vinfo);
1281 fdrevoke(void *f_data, short f_type, struct ucred *cred)
1283 struct fdrevoke_info info;
1286 bzero(&info, sizeof(info));
1290 error = falloc(NULL, &info.nfp, NULL);
1295 * Scan the file pointer table once. dups do not dup file pointers,
1296 * only descriptors, so there is no leak. Set FREVOKED on the fps
1299 allfiles_scan_exclusive(fdrevoke_check_callback, &info);
1302 * If any fps were marked track down the related descriptors
1303 * and close them. Any dup()s at this point will notice
1304 * the FREVOKED already set in the fp and do the right thing.
1306 * Any fps with non-zero msgcounts (aka sent over a unix-domain
1307 * socket) bumped the intransit counter and will require a
1308 * scan. Races against fps leaving the socket are closed by
1309 * the socket code checking for FREVOKED.
1312 allproc_scan(fdrevoke_proc_callback, &info);
1314 unp_revoke_gc(info.nfp);
1320 * Locate matching file pointers directly.
1323 fdrevoke_check_callback(struct file *fp, void *vinfo)
1325 struct fdrevoke_info *info = vinfo;
1328 * File pointers already flagged for revokation are skipped.
1330 if (fp->f_flag & FREVOKED)
1334 * If revoking from a prison file pointers created outside of
1335 * that prison, or file pointers without creds, cannot be revoked.
1337 if (info->cred->cr_prison &&
1338 (fp->f_cred == NULL ||
1339 info->cred->cr_prison != fp->f_cred->cr_prison)) {
1344 * If the file pointer matches then mark it for revocation. The
1345 * flag is currently only used by unp_revoke_gc().
1347 * info->count is a heuristic and can race in a SMP environment.
1349 if (info->data == fp->f_data && info->type == fp->f_type) {
1350 atomic_set_int(&fp->f_flag, FREVOKED);
1351 info->count += fp->f_count;
1359 * Locate matching file pointers via process descriptor tables.
1362 fdrevoke_proc_callback(struct proc *p, void *vinfo)
1364 struct fdrevoke_info *info = vinfo;
1365 struct filedesc *fdp;
1369 if (p->p_stat == SIDL || p->p_stat == SZOMB)
1371 if (info->cred->cr_prison &&
1372 info->cred->cr_prison != p->p_ucred->cr_prison) {
1377 * If the controlling terminal of the process matches the
1378 * vnode being revoked we clear the controlling terminal.
1380 * The normal spec_close() may not catch this because it
1381 * uses curproc instead of p.
1383 if (p->p_session && info->type == DTYPE_VNODE &&
1384 info->data == p->p_session->s_ttyvp) {
1385 p->p_session->s_ttyvp = NULL;
1390 * Softref the fdp to prevent it from being destroyed
1392 spin_lock_wr(&p->p_spin);
1393 if ((fdp = p->p_fd) == NULL) {
1394 spin_unlock_wr(&p->p_spin);
1397 atomic_add_int(&fdp->fd_softrefs, 1);
1398 spin_unlock_wr(&p->p_spin);
1401 * Locate and close any matching file descriptors.
1403 spin_lock_wr(&fdp->fd_spin);
1404 for (n = 0; n < fdp->fd_nfiles; ++n) {
1405 if ((fp = fdp->fd_files[n].fp) == NULL)
1407 if (fp->f_flag & FREVOKED) {
1409 fdp->fd_files[n].fp = info->nfp;
1410 spin_unlock_wr(&fdp->fd_spin);
1411 knote_fdclose(fp, fdp, n); /* XXX */
1413 spin_lock_wr(&fdp->fd_spin);
1417 spin_unlock_wr(&fdp->fd_spin);
1418 atomic_subtract_int(&fdp->fd_softrefs, 1);
1424 * Create a new open file structure and reserve a file decriptor
1425 * for the process that refers to it.
1427 * Root creds are checked using lp, or assumed if lp is NULL. If
1428 * resultfd is non-NULL then lp must also be non-NULL. No file
1429 * descriptor is reserved (and no process context is needed) if
1432 * A file pointer with a refcount of 1 is returned. Note that the
1433 * file pointer is NOT associated with the descriptor. If falloc
1434 * returns success, fsetfd() MUST be called to either associate the
1435 * file pointer or clear the reservation.
1440 falloc(struct lwp *lp, struct file **resultfp, int *resultfd)
1442 static struct timeval lastfail;
1445 struct ucred *cred = lp ? lp->lwp_thread->td_ucred : proc0.p_ucred;
1451 * Handle filetable full issues and root overfill.
1453 if (nfiles >= maxfiles - maxfilesrootres &&
1454 (cred->cr_ruid != 0 || nfiles >= maxfiles)) {
1455 if (ppsratecheck(&lastfail, &curfail, 1)) {
1456 kprintf("kern.maxfiles limit exceeded by uid %d, "
1457 "please see tuning(7).\n",
1465 * Allocate a new file descriptor.
1467 fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO);
1468 spin_init(&fp->f_spin);
1469 SLIST_INIT(&fp->f_klist);
1471 fp->f_ops = &badfileops;
1474 spin_lock_wr(&filehead_spin);
1476 LIST_INSERT_HEAD(&filehead, fp, f_list);
1477 spin_unlock_wr(&filehead_spin);
1479 if ((error = fdalloc(lp->lwp_proc, 0, resultfd)) != 0) {
1496 checkfpclosed(struct filedesc *fdp, int fd, struct file *fp)
1500 spin_lock_rd(&fdp->fd_spin);
1501 if ((unsigned) fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp)
1505 spin_unlock_rd(&fdp->fd_spin);
1510 * Associate a file pointer with a previously reserved file descriptor.
1511 * This function always succeeds.
1513 * If fp is NULL, the file descriptor is returned to the pool.
1517 * MPSAFE (exclusive spinlock must be held on call)
1520 fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd)
1522 KKASSERT((unsigned)fd < fdp->fd_nfiles);
1523 KKASSERT(fdp->fd_files[fd].reserved != 0);
1526 fdp->fd_files[fd].fp = fp;
1527 fdp->fd_files[fd].reserved = 0;
1529 fdp->fd_files[fd].reserved = 0;
1530 fdreserve_locked(fdp, fd, -1);
1531 fdfixup_locked(fdp, fd);
1539 fsetfd(struct filedesc *fdp, struct file *fp, int fd)
1541 spin_lock_wr(&fdp->fd_spin);
1542 fsetfd_locked(fdp, fp, fd);
1543 spin_unlock_wr(&fdp->fd_spin);
1547 * MPSAFE (exclusive spinlock must be held on call)
1551 funsetfd_locked(struct filedesc *fdp, int fd)
1555 if ((unsigned)fd >= fdp->fd_nfiles)
1557 if ((fp = fdp->fd_files[fd].fp) == NULL)
1559 fdp->fd_files[fd].fp = NULL;
1560 fdp->fd_files[fd].fileflags = 0;
1562 fdreserve_locked(fdp, fd, -1);
1563 fdfixup_locked(fdp, fd);
1571 fgetfdflags(struct filedesc *fdp, int fd, int *flagsp)
1575 spin_lock_rd(&fdp->fd_spin);
1576 if (((u_int)fd) >= fdp->fd_nfiles) {
1578 } else if (fdp->fd_files[fd].fp == NULL) {
1581 *flagsp = fdp->fd_files[fd].fileflags;
1584 spin_unlock_rd(&fdp->fd_spin);
1592 fsetfdflags(struct filedesc *fdp, int fd, int add_flags)
1596 spin_lock_wr(&fdp->fd_spin);
1597 if (((u_int)fd) >= fdp->fd_nfiles) {
1599 } else if (fdp->fd_files[fd].fp == NULL) {
1602 fdp->fd_files[fd].fileflags |= add_flags;
1605 spin_unlock_wr(&fdp->fd_spin);
1613 fclrfdflags(struct filedesc *fdp, int fd, int rem_flags)
1617 spin_lock_wr(&fdp->fd_spin);
1618 if (((u_int)fd) >= fdp->fd_nfiles) {
1620 } else if (fdp->fd_files[fd].fp == NULL) {
1623 fdp->fd_files[fd].fileflags &= ~rem_flags;
1626 spin_unlock_wr(&fdp->fd_spin);
1631 * Set/Change/Clear the creds for a fp and synchronize the uidinfo.
1634 fsetcred(struct file *fp, struct ucred *ncr)
1637 struct uidinfo *uip;
1640 if (ocr == NULL || ncr == NULL || ocr->cr_uidinfo != ncr->cr_uidinfo) {
1642 uip = ocr->cr_uidinfo;
1643 atomic_add_int(&uip->ui_openfiles, -1);
1646 uip = ncr->cr_uidinfo;
1647 atomic_add_int(&uip->ui_openfiles, 1);
1658 * Free a file descriptor.
1662 ffree(struct file *fp)
1664 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!"));
1665 spin_lock_wr(&filehead_spin);
1666 LIST_REMOVE(fp, f_list);
1668 spin_unlock_wr(&filehead_spin);
1670 if (fp->f_nchandle.ncp)
1671 cache_drop(&fp->f_nchandle);
1676 * called from init_main, initialize filedesc0 for proc0.
1679 fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask)
1683 fdp0->fd_refcnt = 1;
1684 fdp0->fd_cmask = cmask;
1685 fdp0->fd_files = fdp0->fd_builtin_files;
1686 fdp0->fd_nfiles = NDFILE;
1687 fdp0->fd_lastfile = -1;
1688 spin_init(&fdp0->fd_spin);
1692 * Build a new filedesc structure.
1697 fdinit(struct proc *p)
1699 struct filedesc *newfdp;
1700 struct filedesc *fdp = p->p_fd;
1702 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO);
1703 spin_lock_rd(&fdp->fd_spin);
1705 newfdp->fd_cdir = fdp->fd_cdir;
1706 vref(newfdp->fd_cdir);
1707 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1711 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of
1712 * proc0, but should unconditionally exist in other processes.
1715 newfdp->fd_rdir = fdp->fd_rdir;
1716 vref(newfdp->fd_rdir);
1717 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1720 newfdp->fd_jdir = fdp->fd_jdir;
1721 vref(newfdp->fd_jdir);
1722 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1724 spin_unlock_rd(&fdp->fd_spin);
1726 /* Create the file descriptor table. */
1727 newfdp->fd_refcnt = 1;
1728 newfdp->fd_cmask = cmask;
1729 newfdp->fd_files = newfdp->fd_builtin_files;
1730 newfdp->fd_nfiles = NDFILE;
1731 newfdp->fd_lastfile = -1;
1732 spin_init(&newfdp->fd_spin);
1738 * Share a filedesc structure.
1743 fdshare(struct proc *p)
1745 struct filedesc *fdp;
1748 spin_lock_wr(&fdp->fd_spin);
1750 spin_unlock_wr(&fdp->fd_spin);
1755 * Copy a filedesc structure.
1760 fdcopy(struct proc *p)
1762 struct filedesc *fdp = p->p_fd;
1763 struct filedesc *newfdp;
1764 struct fdnode *fdnode;
1769 * Certain daemons might not have file descriptors.
1775 * Allocate the new filedesc and fd_files[] array. This can race
1776 * with operations by other threads on the fdp so we have to be
1779 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK | M_ZERO);
1781 spin_lock_rd(&fdp->fd_spin);
1782 if (fdp->fd_lastfile < NDFILE) {
1783 newfdp->fd_files = newfdp->fd_builtin_files;
1787 * We have to allocate (N^2-1) entries for our in-place
1788 * binary tree. Allow the table to shrink.
1792 while (ni > fdp->fd_lastfile && ni > NDFILE) {
1796 spin_unlock_rd(&fdp->fd_spin);
1797 newfdp->fd_files = kmalloc(i * sizeof(struct fdnode),
1798 M_FILEDESC, M_WAITOK | M_ZERO);
1801 * Check for race, retry
1803 spin_lock_rd(&fdp->fd_spin);
1804 if (i <= fdp->fd_lastfile) {
1805 spin_unlock_rd(&fdp->fd_spin);
1806 kfree(newfdp->fd_files, M_FILEDESC);
1812 * Dup the remaining fields. vref() and cache_hold() can be
1813 * safely called while holding the read spinlock on fdp.
1815 * The read spinlock on fdp is still being held.
1817 * NOTE: vref and cache_hold calls for the case where the vnode
1818 * or cache entry already has at least one ref may be called
1819 * while holding spin locks.
1821 if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) {
1822 vref(newfdp->fd_cdir);
1823 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1826 * We must check for fd_rdir here, at least for now because
1827 * the init process is created before we have access to the
1828 * rootvode to take a reference to it.
1830 if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) {
1831 vref(newfdp->fd_rdir);
1832 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1834 if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) {
1835 vref(newfdp->fd_jdir);
1836 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1838 newfdp->fd_refcnt = 1;
1839 newfdp->fd_nfiles = i;
1840 newfdp->fd_lastfile = fdp->fd_lastfile;
1841 newfdp->fd_freefile = fdp->fd_freefile;
1842 newfdp->fd_cmask = fdp->fd_cmask;
1843 spin_init(&newfdp->fd_spin);
1846 * Copy the descriptor table through (i). This also copies the
1847 * allocation state. Then go through and ref the file pointers
1848 * and clean up any KQ descriptors.
1850 * kq descriptors cannot be copied. Since we haven't ref'd the
1851 * copied files yet we can ignore the return value from funsetfd().
1853 * The read spinlock on fdp is still being held.
1855 bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode));
1856 for (i = 0 ; i < newfdp->fd_nfiles; ++i) {
1857 fdnode = &newfdp->fd_files[i];
1858 if (fdnode->reserved) {
1859 fdreserve_locked(newfdp, i, -1);
1860 fdnode->reserved = 0;
1861 fdfixup_locked(newfdp, i);
1862 } else if (fdnode->fp) {
1863 if (fdnode->fp->f_type == DTYPE_KQUEUE) {
1864 (void)funsetfd_locked(newfdp, i);
1870 spin_unlock_rd(&fdp->fd_spin);
1875 * Release a filedesc structure.
1877 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE)
1880 fdfree(struct proc *p, struct filedesc *repl)
1882 struct filedesc *fdp;
1883 struct fdnode *fdnode;
1885 struct filedesc_to_leader *fdtol;
1891 * Certain daemons might not have file descriptors.
1900 * Severe messing around to follow.
1902 spin_lock_wr(&fdp->fd_spin);
1904 /* Check for special need to clear POSIX style locks */
1906 if (fdtol != NULL) {
1907 KASSERT(fdtol->fdl_refcount > 0,
1908 ("filedesc_to_refcount botch: fdl_refcount=%d",
1909 fdtol->fdl_refcount));
1910 if (fdtol->fdl_refcount == 1 &&
1911 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1912 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1913 fdnode = &fdp->fd_files[i];
1914 if (fdnode->fp == NULL ||
1915 fdnode->fp->f_type != DTYPE_VNODE) {
1920 spin_unlock_wr(&fdp->fd_spin);
1922 lf.l_whence = SEEK_SET;
1925 lf.l_type = F_UNLCK;
1926 vp = (struct vnode *)fp->f_data;
1927 (void) VOP_ADVLOCK(vp,
1928 (caddr_t)p->p_leader,
1933 spin_lock_wr(&fdp->fd_spin);
1937 if (fdtol->fdl_refcount == 1) {
1938 if (fdp->fd_holdleaderscount > 0 &&
1939 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1941 * close() or do_dup() has cleared a reference
1942 * in a shared file descriptor table.
1944 fdp->fd_holdleaderswakeup = 1;
1945 ssleep(&fdp->fd_holdleaderscount,
1946 &fdp->fd_spin, 0, "fdlhold", 0);
1949 if (fdtol->fdl_holdcount > 0) {
1951 * Ensure that fdtol->fdl_leader
1952 * remains valid in closef().
1954 fdtol->fdl_wakeup = 1;
1955 ssleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0);
1959 fdtol->fdl_refcount--;
1960 if (fdtol->fdl_refcount == 0 &&
1961 fdtol->fdl_holdcount == 0) {
1962 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1963 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1968 if (fdtol != NULL) {
1969 spin_unlock_wr(&fdp->fd_spin);
1970 kfree(fdtol, M_FILEDESC_TO_LEADER);
1971 spin_lock_wr(&fdp->fd_spin);
1974 if (--fdp->fd_refcnt > 0) {
1975 spin_unlock_wr(&fdp->fd_spin);
1976 spin_lock_wr(&p->p_spin);
1978 spin_unlock_wr(&p->p_spin);
1983 * Even though we are the last reference to the structure allproc
1984 * scans may still reference the structure. Maintain proper
1985 * locks until we can replace p->p_fd.
1987 * Also note that kqueue's closef still needs to reference the
1988 * fdp via p->p_fd, so we have to close the descriptors before
1989 * we replace p->p_fd.
1991 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1992 if (fdp->fd_files[i].fp) {
1993 fp = funsetfd_locked(fdp, i);
1995 spin_unlock_wr(&fdp->fd_spin);
1997 spin_lock_wr(&fdp->fd_spin);
2001 spin_unlock_wr(&fdp->fd_spin);
2004 * Interlock against an allproc scan operations (typically frevoke).
2006 spin_lock_wr(&p->p_spin);
2008 spin_unlock_wr(&p->p_spin);
2011 * Wait for any softrefs to go away. This race rarely occurs so
2012 * we can use a non-critical-path style poll/sleep loop. The
2013 * race only occurs against allproc scans.
2015 * No new softrefs can occur with the fdp disconnected from the
2018 if (fdp->fd_softrefs) {
2019 kprintf("pid %d: Warning, fdp race avoided\n", p->p_pid);
2020 while (fdp->fd_softrefs)
2021 tsleep(&fdp->fd_softrefs, 0, "fdsoft", 1);
2024 if (fdp->fd_files != fdp->fd_builtin_files)
2025 kfree(fdp->fd_files, M_FILEDESC);
2027 cache_drop(&fdp->fd_ncdir);
2028 vrele(fdp->fd_cdir);
2031 cache_drop(&fdp->fd_nrdir);
2032 vrele(fdp->fd_rdir);
2035 cache_drop(&fdp->fd_njdir);
2036 vrele(fdp->fd_jdir);
2038 kfree(fdp, M_FILEDESC);
2042 * Retrieve and reference the file pointer associated with a descriptor.
2047 holdfp(struct filedesc *fdp, int fd, int flag)
2051 spin_lock_rd(&fdp->fd_spin);
2052 if (((u_int)fd) >= fdp->fd_nfiles) {
2056 if ((fp = fdp->fd_files[fd].fp) == NULL)
2058 if ((fp->f_flag & flag) == 0 && flag != -1) {
2064 spin_unlock_rd(&fdp->fd_spin);
2069 * holdsock() - load the struct file pointer associated
2070 * with a socket into *fpp. If an error occurs, non-zero
2071 * will be returned and *fpp will be set to NULL.
2076 holdsock(struct filedesc *fdp, int fd, struct file **fpp)
2081 spin_lock_rd(&fdp->fd_spin);
2082 if ((unsigned)fd >= fdp->fd_nfiles) {
2087 if ((fp = fdp->fd_files[fd].fp) == NULL) {
2091 if (fp->f_type != DTYPE_SOCKET) {
2098 spin_unlock_rd(&fdp->fd_spin);
2104 * Convert a user file descriptor to a held file pointer.
2109 holdvnode(struct filedesc *fdp, int fd, struct file **fpp)
2114 spin_lock_rd(&fdp->fd_spin);
2115 if ((unsigned)fd >= fdp->fd_nfiles) {
2120 if ((fp = fdp->fd_files[fd].fp) == NULL) {
2124 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
2132 spin_unlock_rd(&fdp->fd_spin);
2138 * For setugid programs, we don't want to people to use that setugidness
2139 * to generate error messages which write to a file which otherwise would
2140 * otherwise be off-limits to the process.
2142 * This is a gross hack to plug the hole. A better solution would involve
2143 * a special vop or other form of generalized access control mechanism. We
2144 * go ahead and just reject all procfs file systems accesses as dangerous.
2146 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2147 * sufficient. We also don't for check setugidness since we know we are.
2150 is_unsafe(struct file *fp)
2152 if (fp->f_type == DTYPE_VNODE &&
2153 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS)
2159 * Make this setguid thing safe, if at all possible.
2161 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2164 setugidsafety(struct proc *p)
2166 struct filedesc *fdp = p->p_fd;
2169 /* Certain daemons might not have file descriptors. */
2174 * note: fdp->fd_files may be reallocated out from under us while
2175 * we are blocked in a close. Be careful!
2177 for (i = 0; i <= fdp->fd_lastfile; i++) {
2180 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) {
2184 * NULL-out descriptor prior to close to avoid
2185 * a race while close blocks.
2187 if ((fp = funsetfd_locked(fdp, i)) != NULL) {
2188 knote_fdclose(fp, fdp, i);
2196 * Close any files on exec?
2198 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2201 fdcloseexec(struct proc *p)
2203 struct filedesc *fdp = p->p_fd;
2206 /* Certain daemons might not have file descriptors. */
2211 * We cannot cache fd_files since operations may block and rip
2212 * them out from under us.
2214 for (i = 0; i <= fdp->fd_lastfile; i++) {
2215 if (fdp->fd_files[i].fp != NULL &&
2216 (fdp->fd_files[i].fileflags & UF_EXCLOSE)) {
2220 * NULL-out descriptor prior to close to avoid
2221 * a race while close blocks.
2223 if ((fp = funsetfd_locked(fdp, i)) != NULL) {
2224 knote_fdclose(fp, fdp, i);
2232 * It is unsafe for set[ug]id processes to be started with file
2233 * descriptors 0..2 closed, as these descriptors are given implicit
2234 * significance in the Standard C library. fdcheckstd() will create a
2235 * descriptor referencing /dev/null for each of stdin, stdout, and
2236 * stderr that is not already open.
2238 * NOT MPSAFE - calls falloc, vn_open, etc
2241 fdcheckstd(struct lwp *lp)
2243 struct nlookupdata nd;
2244 struct filedesc *fdp;
2247 int i, error, flags, devnull;
2249 fdp = lp->lwp_proc->p_fd;
2254 for (i = 0; i < 3; i++) {
2255 if (fdp->fd_files[i].fp != NULL)
2258 if ((error = falloc(lp, &fp, &devnull)) != 0)
2261 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE,
2262 NLC_FOLLOW|NLC_LOCKVP);
2263 flags = FREAD | FWRITE;
2265 error = vn_open(&nd, fp, flags, 0);
2267 fsetfd(fdp, fp, devnull);
2269 fsetfd(fdp, NULL, devnull);
2274 KKASSERT(i == devnull);
2276 error = kern_dup(DUP_FIXED, devnull, i, &retval);
2285 * Internal form of close.
2286 * Decrement reference count on file structure.
2287 * Note: td and/or p may be NULL when closing a file
2288 * that was being passed in a message.
2290 * MPALMOSTSAFE - acquires mplock for VOP operations
2293 closef(struct file *fp, struct proc *p)
2297 struct filedesc_to_leader *fdtol;
2303 * POSIX record locking dictates that any close releases ALL
2304 * locks owned by this process. This is handled by setting
2305 * a flag in the unlock to free ONLY locks obeying POSIX
2306 * semantics, and not to free BSD-style file locks.
2307 * If the descriptor was in a message, POSIX-style locks
2308 * aren't passed with the descriptor.
2310 if (p != NULL && fp->f_type == DTYPE_VNODE &&
2311 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2314 if ((p->p_leader->p_flag & P_ADVLOCK) != 0) {
2315 lf.l_whence = SEEK_SET;
2318 lf.l_type = F_UNLCK;
2319 vp = (struct vnode *)fp->f_data;
2320 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
2324 if (fdtol != NULL) {
2326 * Handle special case where file descriptor table
2327 * is shared between multiple process leaders.
2329 for (fdtol = fdtol->fdl_next;
2330 fdtol != p->p_fdtol;
2331 fdtol = fdtol->fdl_next) {
2332 if ((fdtol->fdl_leader->p_flag &
2335 fdtol->fdl_holdcount++;
2336 lf.l_whence = SEEK_SET;
2339 lf.l_type = F_UNLCK;
2340 vp = (struct vnode *)fp->f_data;
2341 (void) VOP_ADVLOCK(vp,
2342 (caddr_t)fdtol->fdl_leader,
2343 F_UNLCK, &lf, F_POSIX);
2344 fdtol->fdl_holdcount--;
2345 if (fdtol->fdl_holdcount == 0 &&
2346 fdtol->fdl_wakeup != 0) {
2347 fdtol->fdl_wakeup = 0;
2360 * fhold() can only be called if f_count is already at least 1 (i.e. the
2361 * caller of fhold() already has a reference to the file pointer in some
2364 * f_count is not spin-locked. Instead, atomic ops are used for
2365 * incrementing, decrementing, and handling the 1->0 transition.
2368 fhold(struct file *fp)
2370 atomic_add_int(&fp->f_count, 1);
2374 * fdrop() - drop a reference to a descriptor
2376 * MPALMOSTSAFE - acquires mplock for final close sequence
2379 fdrop(struct file *fp)
2386 * A combined fetch and subtract is needed to properly detect
2387 * 1->0 transitions, otherwise two cpus dropping from a ref
2388 * count of 2 might both try to run the 1->0 code.
2390 if (atomic_fetchadd_int(&fp->f_count, -1) > 1)
2396 * The last reference has gone away, we own the fp structure free
2399 if (fp->f_count < 0)
2400 panic("fdrop: count < 0");
2401 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE &&
2402 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2404 lf.l_whence = SEEK_SET;
2407 lf.l_type = F_UNLCK;
2408 vp = (struct vnode *)fp->f_data;
2409 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2411 if (fp->f_ops != &badfileops)
2412 error = fo_close(fp);
2421 * Apply an advisory lock on a file descriptor.
2423 * Just attempt to get a record lock of the requested type on
2424 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2429 sys_flock(struct flock_args *uap)
2431 struct proc *p = curproc;
2437 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
2440 if (fp->f_type != DTYPE_VNODE) {
2444 vp = (struct vnode *)fp->f_data;
2445 lf.l_whence = SEEK_SET;
2448 if (uap->how & LOCK_UN) {
2449 lf.l_type = F_UNLCK;
2450 fp->f_flag &= ~FHASLOCK;
2451 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2454 if (uap->how & LOCK_EX)
2455 lf.l_type = F_WRLCK;
2456 else if (uap->how & LOCK_SH)
2457 lf.l_type = F_RDLCK;
2462 fp->f_flag |= FHASLOCK;
2463 if (uap->how & LOCK_NB)
2464 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0);
2466 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT);
2474 * File Descriptor pseudo-device driver (/dev/fd/).
2476 * Opening minor device N dup()s the file (if any) connected to file
2477 * descriptor N belonging to the calling process. Note that this driver
2478 * consists of only the ``open()'' routine, because all subsequent
2479 * references to this file will be direct to the other driver.
2482 fdopen(struct dev_open_args *ap)
2484 thread_t td = curthread;
2486 KKASSERT(td->td_lwp != NULL);
2489 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the
2490 * the file descriptor being sought for duplication. The error
2491 * return ensures that the vnode for this device will be released
2492 * by vn_open. Open will detect this special error and take the
2493 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2494 * will simply report the error.
2496 td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev);
2501 * The caller has reserved the file descriptor dfd for us. On success we
2502 * must fsetfd() it. On failure the caller will clean it up.
2507 dupfdopen(struct filedesc *fdp, int dfd, int sfd, int mode, int error)
2513 if ((wfp = holdfp(fdp, sfd, -1)) == NULL)
2517 * Close a revoke/dup race. Duping a descriptor marked as revoked
2518 * will dup a dummy descriptor instead of the real one.
2520 if (wfp->f_flag & FREVOKED) {
2521 kprintf("Warning: attempt to dup() a revoked descriptor\n");
2524 werror = falloc(NULL, &wfp, NULL);
2530 * There are two cases of interest here.
2532 * For ENODEV simply dup sfd to file descriptor dfd and return.
2534 * For ENXIO steal away the file structure from sfd and store it
2535 * dfd. sfd is effectively closed by this operation.
2537 * Any other error code is just returned.
2542 * Check that the mode the file is being opened for is a
2543 * subset of the mode of the existing descriptor.
2545 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2549 spin_lock_wr(&fdp->fd_spin);
2550 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2551 fsetfd_locked(fdp, wfp, dfd);
2552 spin_unlock_wr(&fdp->fd_spin);
2557 * Steal away the file pointer from dfd, and stuff it into indx.
2559 spin_lock_wr(&fdp->fd_spin);
2560 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2561 fsetfd(fdp, wfp, dfd);
2562 if ((xfp = funsetfd_locked(fdp, sfd)) != NULL) {
2563 spin_unlock_wr(&fdp->fd_spin);
2566 spin_unlock_wr(&fdp->fd_spin);
2578 * NOT MPSAFE - I think these refer to a common file descriptor table
2579 * and we need to spinlock that to link fdtol in.
2581 struct filedesc_to_leader *
2582 filedesc_to_leader_alloc(struct filedesc_to_leader *old,
2583 struct proc *leader)
2585 struct filedesc_to_leader *fdtol;
2587 fdtol = kmalloc(sizeof(struct filedesc_to_leader),
2588 M_FILEDESC_TO_LEADER, M_WAITOK);
2589 fdtol->fdl_refcount = 1;
2590 fdtol->fdl_holdcount = 0;
2591 fdtol->fdl_wakeup = 0;
2592 fdtol->fdl_leader = leader;
2594 fdtol->fdl_next = old->fdl_next;
2595 fdtol->fdl_prev = old;
2596 old->fdl_next = fdtol;
2597 fdtol->fdl_next->fdl_prev = fdtol;
2599 fdtol->fdl_next = fdtol;
2600 fdtol->fdl_prev = fdtol;
2606 * Scan all file pointers in the system. The callback is made with
2607 * the master list spinlock held exclusively.
2612 allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data)
2617 spin_lock_wr(&filehead_spin);
2618 LIST_FOREACH(fp, &filehead, f_list) {
2619 res = callback(fp, data);
2623 spin_unlock_wr(&filehead_spin);
2627 * Get file structures.
2629 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe)
2632 struct sysctl_kern_file_info {
2635 struct sysctl_req *req;
2638 static int sysctl_kern_file_callback(struct proc *p, void *data);
2641 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2643 struct sysctl_kern_file_info info;
2646 * Note: because the number of file descriptors is calculated
2647 * in different ways for sizing vs returning the data,
2648 * there is information leakage from the first loop. However,
2649 * it is of a similar order of magnitude to the leakage from
2650 * global system statistics such as kern.openfiles.
2652 * When just doing a count, note that we cannot just count
2653 * the elements and add f_count via the filehead list because
2654 * threaded processes share their descriptor table and f_count might
2655 * still be '1' in that case.
2657 * Since the SYSCTL op can block, we must hold the process to
2658 * prevent it being ripped out from under us either in the
2659 * file descriptor loop or in the greater LIST_FOREACH. The
2660 * process may be in varying states of disrepair. If the process
2661 * is in SZOMB we may have caught it just as it is being removed
2662 * from the allproc list, we must skip it in that case to maintain
2663 * an unbroken chain through the allproc list.
2668 allproc_scan(sysctl_kern_file_callback, &info);
2671 * When just calculating the size, overestimate a bit to try to
2672 * prevent system activity from causing the buffer-fill call
2675 if (req->oldptr == NULL) {
2676 info.count = (info.count + 16) + (info.count / 10);
2677 info.error = SYSCTL_OUT(req, NULL,
2678 info.count * sizeof(struct kinfo_file));
2680 return (info.error);
2684 sysctl_kern_file_callback(struct proc *p, void *data)
2686 struct sysctl_kern_file_info *info = data;
2687 struct kinfo_file kf;
2688 struct filedesc *fdp;
2693 if (p->p_stat == SIDL || p->p_stat == SZOMB)
2695 if (!PRISON_CHECK(info->req->td->td_ucred, p->p_ucred) != 0)
2699 * Softref the fdp to prevent it from being destroyed
2701 spin_lock_wr(&p->p_spin);
2702 if ((fdp = p->p_fd) == NULL) {
2703 spin_unlock_wr(&p->p_spin);
2706 atomic_add_int(&fdp->fd_softrefs, 1);
2707 spin_unlock_wr(&p->p_spin);
2710 * The fdp's own spinlock prevents the contents from being
2713 spin_lock_rd(&fdp->fd_spin);
2714 for (n = 0; n < fdp->fd_nfiles; ++n) {
2715 if ((fp = fdp->fd_files[n].fp) == NULL)
2717 if (info->req->oldptr == NULL) {
2720 uid = p->p_ucred ? p->p_ucred->cr_uid : -1;
2721 kcore_make_file(&kf, fp, p->p_pid, uid, n);
2722 spin_unlock_rd(&fdp->fd_spin);
2723 info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf));
2724 spin_lock_rd(&fdp->fd_spin);
2729 spin_unlock_rd(&fdp->fd_spin);
2730 atomic_subtract_int(&fdp->fd_softrefs, 1);
2736 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2737 0, 0, sysctl_kern_file, "S,file", "Entire file table");
2739 SYSCTL_INT(_kern, OID_AUTO, minfilesperproc, CTLFLAG_RW,
2740 &minfilesperproc, 0, "Minimum files allowed open per process");
2741 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2742 &maxfilesperproc, 0, "Maximum files allowed open per process");
2743 SYSCTL_INT(_kern, OID_AUTO, maxfilesperuser, CTLFLAG_RW,
2744 &maxfilesperuser, 0, "Maximum files allowed open per user");
2746 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2747 &maxfiles, 0, "Maximum number of files");
2749 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW,
2750 &maxfilesrootres, 0, "Descriptors reserved for root use");
2752 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2753 &nfiles, 0, "System-wide number of open files");
2756 fildesc_drvinit(void *unused)
2760 for (fd = 0; fd < NUMFDESC; fd++) {
2761 make_dev(&fildesc_ops, fd,
2762 UID_BIN, GID_BIN, 0666, "fd/%d", fd);
2765 make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin");
2766 make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout");
2767 make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr");
2773 struct fileops badfileops = {
2774 .fo_read = badfo_readwrite,
2775 .fo_write = badfo_readwrite,
2776 .fo_ioctl = badfo_ioctl,
2777 .fo_poll = badfo_poll,
2778 .fo_kqfilter = badfo_kqfilter,
2779 .fo_stat = badfo_stat,
2780 .fo_close = badfo_close,
2781 .fo_shutdown = badfo_shutdown
2801 badfo_ioctl(struct file *fp, u_long com, caddr_t data,
2802 struct ucred *cred, struct sysmsg *msgv)
2811 badfo_poll(struct file *fp, int events, struct ucred *cred)
2820 badfo_kqfilter(struct file *fp, struct knote *kn)
2826 badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
2835 badfo_close(struct file *fp)
2844 badfo_shutdown(struct file *fp, int how)
2853 nofo_shutdown(struct file *fp, int how)
2855 return (EOPNOTSUPP);
2858 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,
2859 fildesc_drvinit,NULL)