2 * Copyright (c) 2005 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * Copyright (c) 1982, 1986, 1989, 1991, 1993
36 * The Regents of the University of California. All rights reserved.
37 * (c) UNIX System Laboratories, Inc.
38 * All or some portions of this file are derived from material licensed
39 * to the University of California by American Telephone and Telegraph
40 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
41 * the permission of UNIX System Laboratories, Inc.
43 * Redistribution and use in source and binary forms, with or without
44 * modification, are permitted provided that the following conditions
46 * 1. Redistributions of source code must retain the above copyright
47 * notice, this list of conditions and the following disclaimer.
48 * 2. Redistributions in binary form must reproduce the above copyright
49 * notice, this list of conditions and the following disclaimer in the
50 * documentation and/or other materials provided with the distribution.
51 * 3. All advertising materials mentioning features or use of this software
52 * must display the following acknowledgement:
53 * This product includes software developed by the University of
54 * California, Berkeley and its contributors.
55 * 4. Neither the name of the University nor the names of its contributors
56 * may be used to endorse or promote products derived from this software
57 * without specific prior written permission.
59 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
60 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
61 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
62 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
63 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
64 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
65 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
66 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
67 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
71 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94
72 * $FreeBSD: src/sys/kern/kern_descrip.c,v 1.81.2.19 2004/02/28 00:43:31 tegge Exp $
73 * $DragonFly: src/sys/kern/kern_descrip.c,v 1.79 2008/08/31 13:18:28 aggelos Exp $
76 #include "opt_compat.h"
77 #include <sys/param.h>
78 #include <sys/systm.h>
79 #include <sys/malloc.h>
80 #include <sys/sysproto.h>
82 #include <sys/device.h>
83 #include <sys/filedesc.h>
84 #include <sys/kernel.h>
85 #include <sys/sysctl.h>
86 #include <sys/vnode.h>
88 #include <sys/nlookup.h>
91 #include <sys/filio.h>
92 #include <sys/fcntl.h>
93 #include <sys/unistd.h>
94 #include <sys/resourcevar.h>
95 #include <sys/event.h>
96 #include <sys/kern_syscall.h>
97 #include <sys/kcore.h>
98 #include <sys/kinfo.h>
102 #include <vm/vm_extern.h>
104 #include <sys/thread2.h>
105 #include <sys/file2.h>
106 #include <sys/spinlock2.h>
108 static void fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd);
109 static void fdreserve_locked (struct filedesc *fdp, int fd0, int incr);
110 static struct file *funsetfd_locked (struct filedesc *fdp, int fd);
111 static int checkfpclosed(struct filedesc *fdp, int fd, struct file *fp);
112 static void ffree(struct file *fp);
114 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
115 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
116 "file desc to leader structures");
117 MALLOC_DEFINE(M_FILE, "file", "Open file structure");
118 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
120 static d_open_t fdopen;
123 #define CDEV_MAJOR 22
124 static struct dev_ops fildesc_ops = {
125 { "FD", CDEV_MAJOR, 0 },
129 static int badfo_readwrite (struct file *fp, struct uio *uio,
130 struct ucred *cred, int flags);
131 static int badfo_ioctl (struct file *fp, u_long com, caddr_t data,
133 static int badfo_poll (struct file *fp, int events, struct ucred *cred);
134 static int badfo_kqfilter (struct file *fp, struct knote *kn);
135 static int badfo_stat (struct file *fp, struct stat *sb, struct ucred *cred);
136 static int badfo_close (struct file *fp);
137 static int badfo_shutdown (struct file *fp, int how);
140 * Descriptor management.
142 static struct filelist filehead = LIST_HEAD_INITIALIZER(&filehead);
143 static struct spinlock filehead_spin = SPINLOCK_INITIALIZER(&filehead_spin);
144 static int nfiles; /* actual number of open files */
148 * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared.
150 * MPSAFE - must be called with fdp->fd_spin exclusively held
154 fdfixup_locked(struct filedesc *fdp, int fd)
156 if (fd < fdp->fd_freefile) {
157 fdp->fd_freefile = fd;
159 while (fdp->fd_lastfile >= 0 &&
160 fdp->fd_files[fdp->fd_lastfile].fp == NULL &&
161 fdp->fd_files[fdp->fd_lastfile].reserved == 0
168 * System calls on descriptors.
173 sys_getdtablesize(struct getdtablesize_args *uap)
175 struct proc *p = curproc;
176 struct plimit *limit = p->p_limit;
178 spin_lock_rd(&limit->p_spin);
180 min((int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
181 spin_unlock_rd(&limit->p_spin);
186 * Duplicate a file descriptor to a particular value.
188 * note: keep in mind that a potential race condition exists when closing
189 * descriptors from a shared descriptor table (via rfork).
194 sys_dup2(struct dup2_args *uap)
198 error = kern_dup(DUP_FIXED, uap->from, uap->to, uap->sysmsg_fds);
204 * Duplicate a file descriptor.
209 sys_dup(struct dup_args *uap)
213 error = kern_dup(DUP_VARIABLE, uap->fd, 0, uap->sysmsg_fds);
219 * MPALMOSTSAFE - acquires mplock for fp operations
222 kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred)
224 struct thread *td = curthread;
225 struct proc *p = td->td_proc;
231 int tmp, error, flg = F_POSIX;
236 * Operations on file descriptors that do not require a file pointer.
240 error = fgetfdflags(p->p_fd, fd, &tmp);
242 dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0;
246 if (dat->fc_cloexec & FD_CLOEXEC)
247 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
249 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
253 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd);
260 * Operations on file pointers
262 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
268 dat->fc_flags = OFLAGS(fp->f_flag);
274 nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS;
275 nflags |= oflags & ~FCNTLFLAGS;
278 if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY))
280 if (error == 0 && ((nflags ^ oflags) & FASYNC)) {
281 tmp = nflags & FASYNC;
282 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, cred);
289 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner, cred);
293 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner, cred);
298 /* Fall into F_SETLK */
301 if (fp->f_type != DTYPE_VNODE) {
305 vp = (struct vnode *)fp->f_data;
308 * copyin/lockop may block
310 if (dat->fc_flock.l_whence == SEEK_CUR)
311 dat->fc_flock.l_start += fp->f_offset;
313 switch (dat->fc_flock.l_type) {
315 if ((fp->f_flag & FREAD) == 0) {
319 p->p_leader->p_flag |= P_ADVLOCK;
320 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
321 &dat->fc_flock, flg);
324 if ((fp->f_flag & FWRITE) == 0) {
328 p->p_leader->p_flag |= P_ADVLOCK;
329 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
330 &dat->fc_flock, flg);
333 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
334 &dat->fc_flock, F_POSIX);
342 * It is possible to race a close() on the descriptor while
343 * we were blocked getting the lock. If this occurs the
344 * close might not have caught the lock.
346 if (checkfpclosed(p->p_fd, fd, fp)) {
347 dat->fc_flock.l_whence = SEEK_SET;
348 dat->fc_flock.l_start = 0;
349 dat->fc_flock.l_len = 0;
350 dat->fc_flock.l_type = F_UNLCK;
351 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
352 F_UNLCK, &dat->fc_flock, F_POSIX);
357 if (fp->f_type != DTYPE_VNODE) {
361 vp = (struct vnode *)fp->f_data;
363 * copyin/lockop may block
365 if (dat->fc_flock.l_type != F_RDLCK &&
366 dat->fc_flock.l_type != F_WRLCK &&
367 dat->fc_flock.l_type != F_UNLCK) {
371 if (dat->fc_flock.l_whence == SEEK_CUR)
372 dat->fc_flock.l_start += fp->f_offset;
373 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK,
374 &dat->fc_flock, F_POSIX);
387 * The file control system call.
392 sys_fcntl(struct fcntl_args *uap)
399 dat.fc_fd = uap->arg;
402 dat.fc_cloexec = uap->arg;
405 dat.fc_flags = uap->arg;
408 dat.fc_owner = uap->arg;
413 error = copyin((caddr_t)uap->arg, &dat.fc_flock,
414 sizeof(struct flock));
420 error = kern_fcntl(uap->fd, uap->cmd, &dat, curproc->p_ucred);
425 uap->sysmsg_result = dat.fc_fd;
428 uap->sysmsg_result = dat.fc_cloexec;
431 uap->sysmsg_result = dat.fc_flags;
434 uap->sysmsg_result = dat.fc_owner;
436 error = copyout(&dat.fc_flock, (caddr_t)uap->arg,
437 sizeof(struct flock));
446 * Common code for dup, dup2, and fcntl(F_DUPFD).
448 * The type flag can be either DUP_FIXED or DUP_VARIABLE. DUP_FIXED tells
449 * kern_dup() to destructively dup over an existing file descriptor if new
450 * is already open. DUP_VARIABLE tells kern_dup() to find the lowest
451 * unused file descriptor that is greater than or equal to new.
456 kern_dup(enum dup_type type, int old, int new, int *res)
458 struct thread *td = curthread;
459 struct proc *p = td->td_proc;
460 struct filedesc *fdp = p->p_fd;
468 * Verify that we have a valid descriptor to dup from and
469 * possibly to dup to.
472 spin_lock_wr(&fdp->fd_spin);
473 if (new < 0 || new > p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
474 new >= maxfilesperproc) {
475 spin_unlock_wr(&fdp->fd_spin);
478 if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) {
479 spin_unlock_wr(&fdp->fd_spin);
482 if (type == DUP_FIXED && old == new) {
484 spin_unlock_wr(&fdp->fd_spin);
487 fp = fdp->fd_files[old].fp;
488 oldflags = fdp->fd_files[old].fileflags;
489 fhold(fp); /* MPSAFE - can be called with a spinlock held */
492 * Allocate a new descriptor if DUP_VARIABLE, or expand the table
493 * if the requested descriptor is beyond the current table size.
495 * This can block. Retry if the source descriptor no longer matches
496 * or if our expectation in the expansion case races.
498 * If we are not expanding or allocating a new decriptor, then reset
499 * the target descriptor to a reserved state so we have a uniform
500 * setup for the next code block.
502 if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) {
503 spin_unlock_wr(&fdp->fd_spin);
504 error = fdalloc(p, new, &newfd);
505 spin_lock_wr(&fdp->fd_spin);
507 spin_unlock_wr(&fdp->fd_spin);
514 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) {
515 fsetfd_locked(fdp, NULL, newfd);
516 spin_unlock_wr(&fdp->fd_spin);
521 * Check for expansion race
523 if (type != DUP_VARIABLE && new != newfd) {
524 fsetfd_locked(fdp, NULL, newfd);
525 spin_unlock_wr(&fdp->fd_spin);
530 * Check for ripout, newfd reused old (this case probably
534 fsetfd_locked(fdp, NULL, newfd);
535 spin_unlock_wr(&fdp->fd_spin);
542 if (fdp->fd_files[new].reserved) {
543 spin_unlock_wr(&fdp->fd_spin);
545 kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new);
546 tsleep(fdp, 0, "fdres", hz);
551 * If the target descriptor was never allocated we have
552 * to allocate it. If it was we have to clean out the
553 * old descriptor. delfp inherits the ref from the
556 delfp = fdp->fd_files[new].fp;
557 fdp->fd_files[new].fp = NULL;
558 fdp->fd_files[new].reserved = 1;
560 fdreserve_locked(fdp, new, 1);
561 if (new > fdp->fd_lastfile)
562 fdp->fd_lastfile = new;
568 * NOTE: still holding an exclusive spinlock
572 * If a descriptor is being overwritten we may hve to tell
573 * fdfree() to sleep to ensure that all relevant process
574 * leaders can be traversed in closef().
576 if (delfp != NULL && p->p_fdtol != NULL) {
577 fdp->fd_holdleaderscount++;
582 KASSERT(delfp == NULL || type == DUP_FIXED,
583 ("dup() picked an open file"));
586 * Duplicate the source descriptor, update lastfile. If the new
587 * descriptor was not allocated and we aren't replacing an existing
588 * descriptor we have to mark the descriptor as being in use.
590 * The fd_files[] array inherits fp's hold reference.
592 fsetfd_locked(fdp, fp, new);
593 fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE;
594 spin_unlock_wr(&fdp->fd_spin);
599 * If we dup'd over a valid file, we now own the reference to it
600 * and must dispose of it using closef() semantics (as if a
601 * close() were performed on it).
606 spin_lock_wr(&fdp->fd_spin);
607 fdp->fd_holdleaderscount--;
608 if (fdp->fd_holdleaderscount == 0 &&
609 fdp->fd_holdleaderswakeup != 0) {
610 fdp->fd_holdleaderswakeup = 0;
611 spin_unlock_wr(&fdp->fd_spin);
612 wakeup(&fdp->fd_holdleaderscount);
614 spin_unlock_wr(&fdp->fd_spin);
622 * If sigio is on the list associated with a process or process group,
623 * disable signalling from the device, remove sigio from the list and
627 funsetown(struct sigio *sigio)
632 *(sigio->sio_myref) = NULL;
634 if (sigio->sio_pgid < 0) {
635 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
637 } else /* if ((*sigiop)->sio_pgid > 0) */ {
638 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
641 crfree(sigio->sio_ucred);
642 kfree(sigio, M_SIGIO);
645 /* Free a list of sigio structures. */
647 funsetownlst(struct sigiolst *sigiolst)
651 while ((sigio = SLIST_FIRST(sigiolst)) != NULL)
656 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
658 * After permission checking, add a sigio structure to the sigio list for
659 * the process or process group.
662 fsetown(pid_t pgid, struct sigio **sigiop)
678 * Policy - Don't allow a process to FSETOWN a process
679 * in another session.
681 * Remove this test to allow maximum flexibility or
682 * restrict FSETOWN to the current process or process
683 * group for maximum safety.
685 if (proc->p_session != curproc->p_session)
689 } else /* if (pgid < 0) */ {
690 pgrp = pgfind(-pgid);
695 * Policy - Don't allow a process to FSETOWN a process
696 * in another session.
698 * Remove this test to allow maximum flexibility or
699 * restrict FSETOWN to the current process or process
700 * group for maximum safety.
702 if (pgrp->pg_session != curproc->p_session)
708 sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
710 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
711 sigio->sio_proc = proc;
713 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
714 sigio->sio_pgrp = pgrp;
716 sigio->sio_pgid = pgid;
717 sigio->sio_ucred = crhold(curproc->p_ucred);
718 /* It would be convenient if p_ruid was in ucred. */
719 sigio->sio_ruid = curproc->p_ucred->cr_ruid;
720 sigio->sio_myref = sigiop;
728 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
731 fgetown(struct sigio *sigio)
733 return (sigio != NULL ? sigio->sio_pgid : 0);
737 * Close many file descriptors.
742 sys_closefrom(struct closefrom_args *uap)
744 return(kern_closefrom(uap->fd));
748 * Close all file descriptors greater then or equal to fd
753 kern_closefrom(int fd)
755 struct thread *td = curthread;
756 struct proc *p = td->td_proc;
757 struct filedesc *fdp;
766 * NOTE: This function will skip unassociated descriptors and
767 * reserved descriptors that have not yet been assigned.
768 * fd_lastfile can change as a side effect of kern_close().
770 spin_lock_wr(&fdp->fd_spin);
771 while (fd <= fdp->fd_lastfile) {
772 if (fdp->fd_files[fd].fp != NULL) {
773 spin_unlock_wr(&fdp->fd_spin);
774 /* ok if this races another close */
775 if (kern_close(fd) == EINTR)
777 spin_lock_wr(&fdp->fd_spin);
781 spin_unlock_wr(&fdp->fd_spin);
786 * Close a file descriptor.
791 sys_close(struct close_args *uap)
793 return(kern_close(uap->fd));
797 * MPALMOSTSAFE - acquires mplock around knote_fdclose() calls
802 struct thread *td = curthread;
803 struct proc *p = td->td_proc;
804 struct filedesc *fdp;
812 spin_lock_wr(&fdp->fd_spin);
813 if ((fp = funsetfd_locked(fdp, fd)) == NULL) {
814 spin_unlock_wr(&fdp->fd_spin);
818 if (p->p_fdtol != NULL) {
820 * Ask fdfree() to sleep to ensure that all relevant
821 * process leaders can be traversed in closef().
823 fdp->fd_holdleaderscount++;
828 * we now hold the fp reference that used to be owned by the descriptor
831 spin_unlock_wr(&fdp->fd_spin);
832 if (fd < fdp->fd_knlistsize) {
834 if (fd < fdp->fd_knlistsize)
835 knote_fdclose(p, fd);
838 error = closef(fp, p);
840 spin_lock_wr(&fdp->fd_spin);
841 fdp->fd_holdleaderscount--;
842 if (fdp->fd_holdleaderscount == 0 &&
843 fdp->fd_holdleaderswakeup != 0) {
844 fdp->fd_holdleaderswakeup = 0;
845 spin_unlock_wr(&fdp->fd_spin);
846 wakeup(&fdp->fd_holdleaderscount);
848 spin_unlock_wr(&fdp->fd_spin);
855 * shutdown_args(int fd, int how)
858 kern_shutdown(int fd, int how)
860 struct thread *td = curthread;
861 struct proc *p = td->td_proc;
867 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
869 error = fo_shutdown(fp, how);
876 sys_shutdown(struct shutdown_args *uap)
880 error = kern_shutdown(uap->s, uap->how);
886 kern_fstat(int fd, struct stat *ub)
888 struct thread *td = curthread;
889 struct proc *p = td->td_proc;
895 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
897 error = fo_stat(fp, ub, p->p_ucred);
904 * Return status information about a file descriptor.
907 sys_fstat(struct fstat_args *uap)
912 error = kern_fstat(uap->fd, &st);
915 error = copyout(&st, uap->sb, sizeof(st));
920 * Return pathconf information about a file descriptor.
924 sys_fpathconf(struct fpathconf_args *uap)
926 struct thread *td = curthread;
927 struct proc *p = td->td_proc;
934 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
937 switch (fp->f_type) {
940 if (uap->name != _PC_PIPE_BUF) {
943 uap->sysmsg_result = PIPE_BUF;
949 vp = (struct vnode *)fp->f_data;
950 error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds);
961 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0, "");
964 * Grow the file table so it can hold through descriptor (want).
966 * The fdp's spinlock must be held exclusively on entry and may be held
967 * exclusively on return. The spinlock may be cycled by the routine.
972 fdgrow_locked(struct filedesc *fdp, int want)
974 struct fdnode *newfiles;
975 struct fdnode *oldfiles;
980 /* nf has to be of the form 2^n - 1 */
982 } while (nf <= want);
984 spin_unlock_wr(&fdp->fd_spin);
985 newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK);
986 spin_lock_wr(&fdp->fd_spin);
989 * We could have raced another extend while we were not holding
992 if (fdp->fd_nfiles >= nf) {
993 spin_unlock_wr(&fdp->fd_spin);
994 kfree(newfiles, M_FILEDESC);
995 spin_lock_wr(&fdp->fd_spin);
999 * Copy the existing ofile and ofileflags arrays
1000 * and zero the new portion of each array.
1002 extra = nf - fdp->fd_nfiles;
1003 bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode));
1004 bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode));
1006 oldfiles = fdp->fd_files;
1007 fdp->fd_files = newfiles;
1008 fdp->fd_nfiles = nf;
1010 if (oldfiles != fdp->fd_builtin_files) {
1011 spin_unlock_wr(&fdp->fd_spin);
1012 kfree(oldfiles, M_FILEDESC);
1013 spin_lock_wr(&fdp->fd_spin);
1019 * Number of nodes in right subtree, including the root.
1022 right_subtree_size(int n)
1024 return (n ^ (n | (n + 1)));
1031 right_ancestor(int n)
1033 return (n | (n + 1));
1040 left_ancestor(int n)
1042 return ((n & (n + 1)) - 1);
1046 * Traverse the in-place binary tree buttom-up adjusting the allocation
1047 * count so scans can determine where free descriptors are located.
1049 * MPSAFE - caller must be holding an exclusive spinlock on fdp
1053 fdreserve_locked(struct filedesc *fdp, int fd, int incr)
1056 fdp->fd_files[fd].allocated += incr;
1057 KKASSERT(fdp->fd_files[fd].allocated >= 0);
1058 fd = left_ancestor(fd);
1063 * Reserve a file descriptor for the process. If no error occurs, the
1064 * caller MUST at some point call fsetfd() or assign a file pointer
1065 * or dispose of the reservation.
1070 fdalloc(struct proc *p, int want, int *result)
1072 struct filedesc *fdp = p->p_fd;
1073 int fd, rsize, rsum, node, lim;
1075 spin_lock_rd(&p->p_limit->p_spin);
1076 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
1077 spin_unlock_rd(&p->p_limit->p_spin);
1080 spin_lock_wr(&fdp->fd_spin);
1081 if (want >= fdp->fd_nfiles)
1082 fdgrow_locked(fdp, want);
1085 * Search for a free descriptor starting at the higher
1086 * of want or fd_freefile. If that fails, consider
1087 * expanding the ofile array.
1089 * NOTE! the 'allocated' field is a cumulative recursive allocation
1090 * count. If we happen to see a value of 0 then we can shortcut
1091 * our search. Otherwise we run through through the tree going
1092 * down branches we know have free descriptor(s) until we hit a
1093 * leaf node. The leaf node will be free but will not necessarily
1094 * have an allocated field of 0.
1097 /* move up the tree looking for a subtree with a free node */
1098 for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim);
1099 fd = right_ancestor(fd)) {
1100 if (fdp->fd_files[fd].allocated == 0)
1103 rsize = right_subtree_size(fd);
1104 if (fdp->fd_files[fd].allocated == rsize)
1105 continue; /* right subtree full */
1108 * Free fd is in the right subtree of the tree rooted at fd.
1109 * Call that subtree R. Look for the smallest (leftmost)
1110 * subtree of R with an unallocated fd: continue moving
1111 * down the left branch until encountering a full left
1112 * subtree, then move to the right.
1114 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) {
1116 rsum += fdp->fd_files[node].allocated;
1117 if (fdp->fd_files[fd].allocated == rsum + rsize) {
1118 fd = node; /* move to the right */
1119 if (fdp->fd_files[node].allocated == 0)
1128 * No space in current array. Expand?
1130 if (fdp->fd_nfiles >= lim) {
1131 spin_unlock_wr(&fdp->fd_spin);
1134 fdgrow_locked(fdp, want);
1138 KKASSERT(fd < fdp->fd_nfiles);
1139 if (fd > fdp->fd_lastfile)
1140 fdp->fd_lastfile = fd;
1141 if (want <= fdp->fd_freefile)
1142 fdp->fd_freefile = fd;
1144 KKASSERT(fdp->fd_files[fd].fp == NULL);
1145 KKASSERT(fdp->fd_files[fd].reserved == 0);
1146 fdp->fd_files[fd].fileflags = 0;
1147 fdp->fd_files[fd].reserved = 1;
1148 fdreserve_locked(fdp, fd, 1);
1149 spin_unlock_wr(&fdp->fd_spin);
1154 * Check to see whether n user file descriptors
1155 * are available to the process p.
1160 fdavail(struct proc *p, int n)
1162 struct filedesc *fdp = p->p_fd;
1163 struct fdnode *fdnode;
1166 spin_lock_rd(&p->p_limit->p_spin);
1167 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
1168 spin_unlock_rd(&p->p_limit->p_spin);
1170 spin_lock_rd(&fdp->fd_spin);
1171 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) {
1172 spin_unlock_rd(&fdp->fd_spin);
1175 last = min(fdp->fd_nfiles, lim);
1176 fdnode = &fdp->fd_files[fdp->fd_freefile];
1177 for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) {
1178 if (fdnode->fp == NULL && --n <= 0) {
1179 spin_unlock_rd(&fdp->fd_spin);
1183 spin_unlock_rd(&fdp->fd_spin);
1188 * Revoke open descriptors referencing (f_data, f_type)
1190 * Any revoke executed within a prison is only able to
1191 * revoke descriptors for processes within that prison.
1193 * Returns 0 on success or an error code.
1195 struct fdrevoke_info {
1205 static int fdrevoke_check_callback(struct file *fp, void *vinfo);
1206 static int fdrevoke_proc_callback(struct proc *p, void *vinfo);
1209 fdrevoke(void *f_data, short f_type, struct ucred *cred)
1211 struct fdrevoke_info info;
1214 bzero(&info, sizeof(info));
1218 error = falloc(NULL, &info.nfp, NULL);
1223 * Scan the file pointer table once. dups do not dup file pointers,
1224 * only descriptors, so there is no leak. Set FREVOKED on the fps
1227 allfiles_scan_exclusive(fdrevoke_check_callback, &info);
1230 * If any fps were marked track down the related descriptors
1231 * and close them. Any dup()s at this point will notice
1232 * the FREVOKED already set in the fp and do the right thing.
1234 * Any fps with non-zero msgcounts (aka sent over a unix-domain
1235 * socket) bumped the intransit counter and will require a
1236 * scan. Races against fps leaving the socket are closed by
1237 * the socket code checking for FREVOKED.
1240 allproc_scan(fdrevoke_proc_callback, &info);
1242 unp_revoke_gc(info.nfp);
1248 * Locate matching file pointers directly.
1251 fdrevoke_check_callback(struct file *fp, void *vinfo)
1253 struct fdrevoke_info *info = vinfo;
1256 * File pointers already flagged for revokation are skipped.
1258 if (fp->f_flag & FREVOKED)
1262 * If revoking from a prison file pointers created outside of
1263 * that prison, or file pointers without creds, cannot be revoked.
1265 if (info->cred->cr_prison &&
1266 (fp->f_cred == NULL ||
1267 info->cred->cr_prison != fp->f_cred->cr_prison)) {
1272 * If the file pointer matches then mark it for revocation. The
1273 * flag is currently only used by unp_revoke_gc().
1275 * info->count is a heuristic and can race in a SMP environment.
1277 if (info->data == fp->f_data && info->type == fp->f_type) {
1278 atomic_set_int(&fp->f_flag, FREVOKED);
1279 info->count += fp->f_count;
1287 * Locate matching file pointers via process descriptor tables.
1290 fdrevoke_proc_callback(struct proc *p, void *vinfo)
1292 struct fdrevoke_info *info = vinfo;
1293 struct filedesc *fdp;
1297 if (p->p_stat == SIDL || p->p_stat == SZOMB)
1299 if (info->cred->cr_prison &&
1300 info->cred->cr_prison != p->p_ucred->cr_prison) {
1305 * If the controlling terminal of the process matches the
1306 * vnode being revoked we clear the controlling terminal.
1308 * The normal spec_close() may not catch this because it
1309 * uses curproc instead of p.
1311 if (p->p_session && info->type == DTYPE_VNODE &&
1312 info->data == p->p_session->s_ttyvp) {
1313 p->p_session->s_ttyvp = NULL;
1318 * Locate and close any matching file descriptors.
1320 if ((fdp = p->p_fd) == NULL)
1322 spin_lock_wr(&fdp->fd_spin);
1323 for (n = 0; n < fdp->fd_nfiles; ++n) {
1324 if ((fp = fdp->fd_files[n].fp) == NULL)
1326 if (fp->f_flag & FREVOKED) {
1328 fdp->fd_files[n].fp = info->nfp;
1329 spin_unlock_wr(&fdp->fd_spin);
1331 spin_lock_wr(&fdp->fd_spin);
1335 spin_unlock_wr(&fdp->fd_spin);
1341 * Create a new open file structure and reserve a file decriptor
1342 * for the process that refers to it.
1344 * Root creds are checked using p, or assumed if p is NULL. If
1345 * resultfd is non-NULL then p must also be non-NULL. No file
1346 * descriptor is reserved if resultfd is NULL.
1348 * A file pointer with a refcount of 1 is returned. Note that the
1349 * file pointer is NOT associated with the descriptor. If falloc
1350 * returns success, fsetfd() MUST be called to either associate the
1351 * file pointer or clear the reservation.
1356 falloc(struct proc *p, struct file **resultfp, int *resultfd)
1358 static struct timeval lastfail;
1366 * Handle filetable full issues and root overfill.
1368 if (nfiles >= maxfiles - maxfilesrootres &&
1369 ((p && p->p_ucred->cr_ruid != 0) || nfiles >= maxfiles)) {
1370 if (ppsratecheck(&lastfail, &curfail, 1)) {
1371 kprintf("kern.maxfiles limit exceeded by uid %d, please see tuning(7).\n",
1372 (p ? p->p_ucred->cr_ruid : -1));
1379 * Allocate a new file descriptor.
1381 fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO);
1382 spin_init(&fp->f_spin);
1384 fp->f_ops = &badfileops;
1387 fp->f_cred = crhold(p->p_ucred);
1389 fp->f_cred = crhold(proc0.p_ucred);
1390 spin_lock_wr(&filehead_spin);
1392 LIST_INSERT_HEAD(&filehead, fp, f_list);
1393 spin_unlock_wr(&filehead_spin);
1395 if ((error = fdalloc(p, 0, resultfd)) != 0) {
1412 checkfpclosed(struct filedesc *fdp, int fd, struct file *fp)
1416 spin_lock_rd(&fdp->fd_spin);
1417 if ((unsigned) fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp)
1421 spin_unlock_rd(&fdp->fd_spin);
1426 * Associate a file pointer with a previously reserved file descriptor.
1427 * This function always succeeds.
1429 * If fp is NULL, the file descriptor is returned to the pool.
1433 * MPSAFE (exclusive spinlock must be held on call)
1436 fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd)
1438 KKASSERT((unsigned)fd < fdp->fd_nfiles);
1439 KKASSERT(fdp->fd_files[fd].reserved != 0);
1442 fdp->fd_files[fd].fp = fp;
1443 fdp->fd_files[fd].reserved = 0;
1444 if (fp->f_type == DTYPE_KQUEUE) {
1445 if (fdp->fd_knlistsize < 0)
1446 fdp->fd_knlistsize = 0;
1449 fdp->fd_files[fd].reserved = 0;
1450 fdreserve_locked(fdp, fd, -1);
1451 fdfixup_locked(fdp, fd);
1459 fsetfd(struct proc *p, struct file *fp, int fd)
1461 struct filedesc *fdp = p->p_fd;
1463 spin_lock_wr(&fdp->fd_spin);
1464 fsetfd_locked(fdp, fp, fd);
1465 spin_unlock_wr(&fdp->fd_spin);
1469 * MPSAFE (exclusive spinlock must be held on call)
1473 funsetfd_locked(struct filedesc *fdp, int fd)
1477 if ((unsigned)fd >= fdp->fd_nfiles)
1479 if ((fp = fdp->fd_files[fd].fp) == NULL)
1481 fdp->fd_files[fd].fp = NULL;
1482 fdp->fd_files[fd].fileflags = 0;
1484 fdreserve_locked(fdp, fd, -1);
1485 fdfixup_locked(fdp, fd);
1493 fgetfdflags(struct filedesc *fdp, int fd, int *flagsp)
1497 spin_lock_rd(&fdp->fd_spin);
1498 if (((u_int)fd) >= fdp->fd_nfiles) {
1500 } else if (fdp->fd_files[fd].fp == NULL) {
1503 *flagsp = fdp->fd_files[fd].fileflags;
1506 spin_unlock_rd(&fdp->fd_spin);
1514 fsetfdflags(struct filedesc *fdp, int fd, int add_flags)
1518 spin_lock_wr(&fdp->fd_spin);
1519 if (((u_int)fd) >= fdp->fd_nfiles) {
1521 } else if (fdp->fd_files[fd].fp == NULL) {
1524 fdp->fd_files[fd].fileflags |= add_flags;
1527 spin_unlock_wr(&fdp->fd_spin);
1535 fclrfdflags(struct filedesc *fdp, int fd, int rem_flags)
1539 spin_lock_wr(&fdp->fd_spin);
1540 if (((u_int)fd) >= fdp->fd_nfiles) {
1542 } else if (fdp->fd_files[fd].fp == NULL) {
1545 fdp->fd_files[fd].fileflags &= ~rem_flags;
1548 spin_unlock_wr(&fdp->fd_spin);
1553 fsetcred(struct file *fp, struct ucred *cr)
1561 * Free a file descriptor.
1565 ffree(struct file *fp)
1567 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!"));
1568 spin_lock_wr(&filehead_spin);
1569 LIST_REMOVE(fp, f_list);
1571 spin_unlock_wr(&filehead_spin);
1573 if (fp->f_nchandle.ncp)
1574 cache_drop(&fp->f_nchandle);
1579 * called from init_main, initialize filedesc0 for proc0.
1582 fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask)
1586 fdp0->fd_refcnt = 1;
1587 fdp0->fd_cmask = cmask;
1588 fdp0->fd_files = fdp0->fd_builtin_files;
1589 fdp0->fd_nfiles = NDFILE;
1590 fdp0->fd_lastfile = -1;
1591 spin_init(&fdp0->fd_spin);
1595 * Build a new filedesc structure.
1600 fdinit(struct proc *p)
1602 struct filedesc *newfdp;
1603 struct filedesc *fdp = p->p_fd;
1605 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO);
1606 spin_lock_rd(&fdp->fd_spin);
1608 newfdp->fd_cdir = fdp->fd_cdir;
1609 vref(newfdp->fd_cdir);
1610 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1614 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of
1615 * proc0, but should unconditionally exist in other processes.
1618 newfdp->fd_rdir = fdp->fd_rdir;
1619 vref(newfdp->fd_rdir);
1620 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1623 newfdp->fd_jdir = fdp->fd_jdir;
1624 vref(newfdp->fd_jdir);
1625 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1627 spin_unlock_rd(&fdp->fd_spin);
1629 /* Create the file descriptor table. */
1630 newfdp->fd_refcnt = 1;
1631 newfdp->fd_cmask = cmask;
1632 newfdp->fd_files = newfdp->fd_builtin_files;
1633 newfdp->fd_nfiles = NDFILE;
1634 newfdp->fd_knlistsize = -1;
1635 newfdp->fd_lastfile = -1;
1636 spin_init(&newfdp->fd_spin);
1642 * Share a filedesc structure.
1647 fdshare(struct proc *p)
1649 struct filedesc *fdp;
1652 spin_lock_wr(&fdp->fd_spin);
1654 spin_unlock_wr(&fdp->fd_spin);
1659 * Copy a filedesc structure.
1664 fdcopy(struct proc *p)
1666 struct filedesc *fdp = p->p_fd;
1667 struct filedesc *newfdp;
1668 struct fdnode *fdnode;
1673 * Certain daemons might not have file descriptors.
1679 * Allocate the new filedesc and fd_files[] array. This can race
1680 * with operations by other threads on the fdp so we have to be
1683 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK | M_ZERO);
1685 spin_lock_rd(&fdp->fd_spin);
1686 if (fdp->fd_lastfile < NDFILE) {
1687 newfdp->fd_files = newfdp->fd_builtin_files;
1691 * We have to allocate (N^2-1) entries for our in-place
1692 * binary tree. Allow the table to shrink.
1696 while (ni > fdp->fd_lastfile && ni > NDFILE) {
1700 spin_unlock_rd(&fdp->fd_spin);
1701 newfdp->fd_files = kmalloc(i * sizeof(struct fdnode),
1702 M_FILEDESC, M_WAITOK | M_ZERO);
1705 * Check for race, retry
1707 spin_lock_rd(&fdp->fd_spin);
1708 if (i <= fdp->fd_lastfile) {
1709 spin_unlock_rd(&fdp->fd_spin);
1710 kfree(newfdp->fd_files, M_FILEDESC);
1716 * Dup the remaining fields. vref() and cache_hold() can be
1717 * safely called while holding the read spinlock on fdp.
1719 * The read spinlock on fdp is still being held.
1721 * NOTE: vref and cache_hold calls for the case where the vnode
1722 * or cache entry already has at least one ref may be called
1723 * while holding spin locks.
1725 if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) {
1726 vref(newfdp->fd_cdir);
1727 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1730 * We must check for fd_rdir here, at least for now because
1731 * the init process is created before we have access to the
1732 * rootvode to take a reference to it.
1734 if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) {
1735 vref(newfdp->fd_rdir);
1736 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1738 if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) {
1739 vref(newfdp->fd_jdir);
1740 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1742 newfdp->fd_refcnt = 1;
1743 newfdp->fd_nfiles = i;
1744 newfdp->fd_lastfile = fdp->fd_lastfile;
1745 newfdp->fd_freefile = fdp->fd_freefile;
1746 newfdp->fd_cmask = fdp->fd_cmask;
1747 newfdp->fd_knlist = NULL;
1748 newfdp->fd_knlistsize = -1;
1749 newfdp->fd_knhash = NULL;
1750 newfdp->fd_knhashmask = 0;
1751 spin_init(&newfdp->fd_spin);
1754 * Copy the descriptor table through (i). This also copies the
1755 * allocation state. Then go through and ref the file pointers
1756 * and clean up any KQ descriptors.
1758 * kq descriptors cannot be copied. Since we haven't ref'd the
1759 * copied files yet we can ignore the return value from funsetfd().
1761 * The read spinlock on fdp is still being held.
1763 bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode));
1764 for (i = 0 ; i < newfdp->fd_nfiles; ++i) {
1765 fdnode = &newfdp->fd_files[i];
1766 if (fdnode->reserved) {
1767 fdreserve_locked(newfdp, i, -1);
1768 fdnode->reserved = 0;
1769 fdfixup_locked(newfdp, i);
1770 } else if (fdnode->fp) {
1771 if (fdnode->fp->f_type == DTYPE_KQUEUE) {
1772 (void)funsetfd_locked(newfdp, i);
1778 spin_unlock_rd(&fdp->fd_spin);
1783 * Release a filedesc structure.
1785 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE)
1788 fdfree(struct proc *p)
1790 struct filedesc *fdp = p->p_fd;
1791 struct fdnode *fdnode;
1793 struct filedesc_to_leader *fdtol;
1798 /* Certain daemons might not have file descriptors. */
1803 * Severe messing around to follow
1805 spin_lock_wr(&fdp->fd_spin);
1807 /* Check for special need to clear POSIX style locks */
1809 if (fdtol != NULL) {
1810 KASSERT(fdtol->fdl_refcount > 0,
1811 ("filedesc_to_refcount botch: fdl_refcount=%d",
1812 fdtol->fdl_refcount));
1813 if (fdtol->fdl_refcount == 1 &&
1814 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1815 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1816 fdnode = &fdp->fd_files[i];
1817 if (fdnode->fp == NULL ||
1818 fdnode->fp->f_type != DTYPE_VNODE) {
1823 spin_unlock_wr(&fdp->fd_spin);
1825 lf.l_whence = SEEK_SET;
1828 lf.l_type = F_UNLCK;
1829 vp = (struct vnode *)fp->f_data;
1830 (void) VOP_ADVLOCK(vp,
1831 (caddr_t)p->p_leader,
1836 spin_lock_wr(&fdp->fd_spin);
1840 if (fdtol->fdl_refcount == 1) {
1841 if (fdp->fd_holdleaderscount > 0 &&
1842 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1844 * close() or do_dup() has cleared a reference
1845 * in a shared file descriptor table.
1847 fdp->fd_holdleaderswakeup = 1;
1848 msleep(&fdp->fd_holdleaderscount,
1849 &fdp->fd_spin, 0, "fdlhold", 0);
1852 if (fdtol->fdl_holdcount > 0) {
1854 * Ensure that fdtol->fdl_leader
1855 * remains valid in closef().
1857 fdtol->fdl_wakeup = 1;
1858 msleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0);
1862 fdtol->fdl_refcount--;
1863 if (fdtol->fdl_refcount == 0 &&
1864 fdtol->fdl_holdcount == 0) {
1865 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1866 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1871 if (fdtol != NULL) {
1872 spin_unlock_wr(&fdp->fd_spin);
1873 kfree(fdtol, M_FILEDESC_TO_LEADER);
1874 spin_lock_wr(&fdp->fd_spin);
1877 if (--fdp->fd_refcnt > 0) {
1878 spin_unlock_wr(&fdp->fd_spin);
1881 spin_unlock_wr(&fdp->fd_spin);
1884 * we are the last reference to the structure, we can
1885 * safely assume it will not change out from under us.
1887 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1888 if (fdp->fd_files[i].fp)
1889 closef(fdp->fd_files[i].fp, p);
1891 if (fdp->fd_files != fdp->fd_builtin_files)
1892 kfree(fdp->fd_files, M_FILEDESC);
1894 cache_drop(&fdp->fd_ncdir);
1895 vrele(fdp->fd_cdir);
1898 cache_drop(&fdp->fd_nrdir);
1899 vrele(fdp->fd_rdir);
1902 cache_drop(&fdp->fd_njdir);
1903 vrele(fdp->fd_jdir);
1906 kfree(fdp->fd_knlist, M_KQUEUE);
1908 kfree(fdp->fd_knhash, M_KQUEUE);
1909 kfree(fdp, M_FILEDESC);
1913 * Retrieve and reference the file pointer associated with a descriptor.
1918 holdfp(struct filedesc *fdp, int fd, int flag)
1922 spin_lock_rd(&fdp->fd_spin);
1923 if (((u_int)fd) >= fdp->fd_nfiles) {
1927 if ((fp = fdp->fd_files[fd].fp) == NULL)
1929 if ((fp->f_flag & flag) == 0 && flag != -1) {
1935 spin_unlock_rd(&fdp->fd_spin);
1940 * holdsock() - load the struct file pointer associated
1941 * with a socket into *fpp. If an error occurs, non-zero
1942 * will be returned and *fpp will be set to NULL.
1947 holdsock(struct filedesc *fdp, int fd, struct file **fpp)
1952 spin_lock_rd(&fdp->fd_spin);
1953 if ((unsigned)fd >= fdp->fd_nfiles) {
1958 if ((fp = fdp->fd_files[fd].fp) == NULL) {
1962 if (fp->f_type != DTYPE_SOCKET) {
1969 spin_unlock_rd(&fdp->fd_spin);
1975 * Convert a user file descriptor to a held file pointer.
1980 holdvnode(struct filedesc *fdp, int fd, struct file **fpp)
1985 spin_lock_rd(&fdp->fd_spin);
1986 if ((unsigned)fd >= fdp->fd_nfiles) {
1991 if ((fp = fdp->fd_files[fd].fp) == NULL) {
1995 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
2003 spin_unlock_rd(&fdp->fd_spin);
2009 * For setugid programs, we don't want to people to use that setugidness
2010 * to generate error messages which write to a file which otherwise would
2011 * otherwise be off-limits to the process.
2013 * This is a gross hack to plug the hole. A better solution would involve
2014 * a special vop or other form of generalized access control mechanism. We
2015 * go ahead and just reject all procfs file systems accesses as dangerous.
2017 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2018 * sufficient. We also don't for check setugidness since we know we are.
2021 is_unsafe(struct file *fp)
2023 if (fp->f_type == DTYPE_VNODE &&
2024 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS)
2030 * Make this setguid thing safe, if at all possible.
2032 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2035 setugidsafety(struct proc *p)
2037 struct filedesc *fdp = p->p_fd;
2040 /* Certain daemons might not have file descriptors. */
2045 * note: fdp->fd_files may be reallocated out from under us while
2046 * we are blocked in a close. Be careful!
2048 for (i = 0; i <= fdp->fd_lastfile; i++) {
2051 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) {
2054 if (i < fdp->fd_knlistsize)
2055 knote_fdclose(p, i);
2057 * NULL-out descriptor prior to close to avoid
2058 * a race while close blocks.
2060 if ((fp = funsetfd_locked(fdp, i)) != NULL)
2067 * Close any files on exec?
2069 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2072 fdcloseexec(struct proc *p)
2074 struct filedesc *fdp = p->p_fd;
2077 /* Certain daemons might not have file descriptors. */
2082 * We cannot cache fd_files since operations may block and rip
2083 * them out from under us.
2085 for (i = 0; i <= fdp->fd_lastfile; i++) {
2086 if (fdp->fd_files[i].fp != NULL &&
2087 (fdp->fd_files[i].fileflags & UF_EXCLOSE)) {
2090 if (i < fdp->fd_knlistsize)
2091 knote_fdclose(p, i);
2093 * NULL-out descriptor prior to close to avoid
2094 * a race while close blocks.
2096 if ((fp = funsetfd_locked(fdp, i)) != NULL)
2103 * It is unsafe for set[ug]id processes to be started with file
2104 * descriptors 0..2 closed, as these descriptors are given implicit
2105 * significance in the Standard C library. fdcheckstd() will create a
2106 * descriptor referencing /dev/null for each of stdin, stdout, and
2107 * stderr that is not already open.
2109 * NOT MPSAFE - calls falloc, vn_open, etc
2112 fdcheckstd(struct proc *p)
2114 struct nlookupdata nd;
2115 struct filedesc *fdp;
2118 int i, error, flags, devnull;
2125 for (i = 0; i < 3; i++) {
2126 if (fdp->fd_files[i].fp != NULL)
2129 if ((error = falloc(p, &fp, &devnull)) != 0)
2132 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE,
2133 NLC_FOLLOW|NLC_LOCKVP);
2134 flags = FREAD | FWRITE;
2136 error = vn_open(&nd, fp, flags, 0);
2138 fsetfd(p, fp, devnull);
2140 fsetfd(p, NULL, devnull);
2145 KKASSERT(i == devnull);
2147 error = kern_dup(DUP_FIXED, devnull, i, &retval);
2156 * Internal form of close.
2157 * Decrement reference count on file structure.
2158 * Note: td and/or p may be NULL when closing a file
2159 * that was being passed in a message.
2161 * MPALMOSTSAFE - acquires mplock for VOP operations
2164 closef(struct file *fp, struct proc *p)
2168 struct filedesc_to_leader *fdtol;
2174 * POSIX record locking dictates that any close releases ALL
2175 * locks owned by this process. This is handled by setting
2176 * a flag in the unlock to free ONLY locks obeying POSIX
2177 * semantics, and not to free BSD-style file locks.
2178 * If the descriptor was in a message, POSIX-style locks
2179 * aren't passed with the descriptor.
2181 if (p != NULL && fp->f_type == DTYPE_VNODE &&
2182 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2185 if ((p->p_leader->p_flag & P_ADVLOCK) != 0) {
2186 lf.l_whence = SEEK_SET;
2189 lf.l_type = F_UNLCK;
2190 vp = (struct vnode *)fp->f_data;
2191 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
2195 if (fdtol != NULL) {
2197 * Handle special case where file descriptor table
2198 * is shared between multiple process leaders.
2200 for (fdtol = fdtol->fdl_next;
2201 fdtol != p->p_fdtol;
2202 fdtol = fdtol->fdl_next) {
2203 if ((fdtol->fdl_leader->p_flag &
2206 fdtol->fdl_holdcount++;
2207 lf.l_whence = SEEK_SET;
2210 lf.l_type = F_UNLCK;
2211 vp = (struct vnode *)fp->f_data;
2212 (void) VOP_ADVLOCK(vp,
2213 (caddr_t)fdtol->fdl_leader,
2214 F_UNLCK, &lf, F_POSIX);
2215 fdtol->fdl_holdcount--;
2216 if (fdtol->fdl_holdcount == 0 &&
2217 fdtol->fdl_wakeup != 0) {
2218 fdtol->fdl_wakeup = 0;
2231 * fhold() can only be called if f_count is already at least 1 (i.e. the
2232 * caller of fhold() already has a reference to the file pointer in some
2235 * f_count is not spin-locked. Instead, atomic ops are used for
2236 * incrementing, decrementing, and handling the 1->0 transition.
2239 fhold(struct file *fp)
2241 atomic_add_int(&fp->f_count, 1);
2245 * fdrop() - drop a reference to a descriptor
2247 * MPALMOSTSAFE - acquires mplock for final close sequence
2250 fdrop(struct file *fp)
2257 * A combined fetch and subtract is needed to properly detect
2258 * 1->0 transitions, otherwise two cpus dropping from a ref
2259 * count of 2 might both try to run the 1->0 code.
2261 if (atomic_fetchadd_int(&fp->f_count, -1) > 1)
2267 * The last reference has gone away, we own the fp structure free
2270 if (fp->f_count < 0)
2271 panic("fdrop: count < 0");
2272 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE &&
2273 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2275 lf.l_whence = SEEK_SET;
2278 lf.l_type = F_UNLCK;
2279 vp = (struct vnode *)fp->f_data;
2280 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2282 if (fp->f_ops != &badfileops)
2283 error = fo_close(fp);
2292 * Apply an advisory lock on a file descriptor.
2294 * Just attempt to get a record lock of the requested type on
2295 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2298 sys_flock(struct flock_args *uap)
2300 struct proc *p = curproc;
2306 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
2308 if (fp->f_type != DTYPE_VNODE) {
2312 vp = (struct vnode *)fp->f_data;
2313 lf.l_whence = SEEK_SET;
2316 if (uap->how & LOCK_UN) {
2317 lf.l_type = F_UNLCK;
2318 fp->f_flag &= ~FHASLOCK;
2319 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2322 if (uap->how & LOCK_EX)
2323 lf.l_type = F_WRLCK;
2324 else if (uap->how & LOCK_SH)
2325 lf.l_type = F_RDLCK;
2330 fp->f_flag |= FHASLOCK;
2331 if (uap->how & LOCK_NB)
2332 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0);
2334 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT);
2341 * File Descriptor pseudo-device driver (/dev/fd/).
2343 * Opening minor device N dup()s the file (if any) connected to file
2344 * descriptor N belonging to the calling process. Note that this driver
2345 * consists of only the ``open()'' routine, because all subsequent
2346 * references to this file will be direct to the other driver.
2350 fdopen(struct dev_open_args *ap)
2352 thread_t td = curthread;
2354 KKASSERT(td->td_lwp != NULL);
2357 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the
2358 * the file descriptor being sought for duplication. The error
2359 * return ensures that the vnode for this device will be released
2360 * by vn_open. Open will detect this special error and take the
2361 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2362 * will simply report the error.
2364 td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev);
2369 * The caller has reserved the file descriptor dfd for us. On success we
2370 * must fsetfd() it. On failure the caller will clean it up.
2372 * NOT MPSAFE - isn't getting spinlocks, possibly other things
2375 dupfdopen(struct proc *p, int dfd, int sfd, int mode, int error)
2377 struct filedesc *fdp = p->p_fd;
2382 if ((wfp = holdfp(fdp, sfd, -1)) == NULL)
2386 * Close a revoke/dup race. Duping a descriptor marked as revoked
2387 * will dup a dummy descriptor instead of the real one.
2389 if (wfp->f_flag & FREVOKED) {
2390 kprintf("Warning: attempt to dup() a revoked descriptor\n");
2393 werror = falloc(NULL, &wfp, NULL);
2399 * There are two cases of interest here.
2401 * For ENODEV simply dup sfd to file descriptor dfd and return.
2403 * For ENXIO steal away the file structure from sfd and store it
2404 * dfd. sfd is effectively closed by this operation.
2406 * Any other error code is just returned.
2411 * Check that the mode the file is being opened for is a
2412 * subset of the mode of the existing descriptor.
2414 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2418 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2419 fsetfd(p, wfp, dfd);
2424 * Steal away the file pointer from dfd, and stuff it into indx.
2426 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2427 fsetfd(p, wfp, dfd);
2428 if ((xfp = funsetfd_locked(fdp, sfd)) != NULL)
2440 * NOT MPSAFE - I think these refer to a common file descriptor table
2441 * and we need to spinlock that to link fdtol in.
2443 struct filedesc_to_leader *
2444 filedesc_to_leader_alloc(struct filedesc_to_leader *old,
2445 struct proc *leader)
2447 struct filedesc_to_leader *fdtol;
2449 fdtol = kmalloc(sizeof(struct filedesc_to_leader),
2450 M_FILEDESC_TO_LEADER, M_WAITOK);
2451 fdtol->fdl_refcount = 1;
2452 fdtol->fdl_holdcount = 0;
2453 fdtol->fdl_wakeup = 0;
2454 fdtol->fdl_leader = leader;
2456 fdtol->fdl_next = old->fdl_next;
2457 fdtol->fdl_prev = old;
2458 old->fdl_next = fdtol;
2459 fdtol->fdl_next->fdl_prev = fdtol;
2461 fdtol->fdl_next = fdtol;
2462 fdtol->fdl_prev = fdtol;
2468 * Scan all file pointers in the system. The callback is made with
2469 * the master list spinlock held exclusively.
2474 allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data)
2479 spin_lock_wr(&filehead_spin);
2480 LIST_FOREACH(fp, &filehead, f_list) {
2481 res = callback(fp, data);
2485 spin_unlock_wr(&filehead_spin);
2489 * Get file structures.
2491 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe)
2494 struct sysctl_kern_file_info {
2497 struct sysctl_req *req;
2500 static int sysctl_kern_file_callback(struct proc *p, void *data);
2503 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2505 struct sysctl_kern_file_info info;
2508 * Note: because the number of file descriptors is calculated
2509 * in different ways for sizing vs returning the data,
2510 * there is information leakage from the first loop. However,
2511 * it is of a similar order of magnitude to the leakage from
2512 * global system statistics such as kern.openfiles.
2514 * When just doing a count, note that we cannot just count
2515 * the elements and add f_count via the filehead list because
2516 * threaded processes share their descriptor table and f_count might
2517 * still be '1' in that case.
2519 * Since the SYSCTL op can block, we must hold the process to
2520 * prevent it being ripped out from under us either in the
2521 * file descriptor loop or in the greater LIST_FOREACH. The
2522 * process may be in varying states of disrepair. If the process
2523 * is in SZOMB we may have caught it just as it is being removed
2524 * from the allproc list, we must skip it in that case to maintain
2525 * an unbroken chain through the allproc list.
2530 allproc_scan(sysctl_kern_file_callback, &info);
2533 * When just calculating the size, overestimate a bit to try to
2534 * prevent system activity from causing the buffer-fill call
2537 if (req->oldptr == NULL) {
2538 info.count = (info.count + 16) + (info.count / 10);
2539 info.error = SYSCTL_OUT(req, NULL,
2540 info.count * sizeof(struct kinfo_file));
2542 return (info.error);
2546 sysctl_kern_file_callback(struct proc *p, void *data)
2548 struct sysctl_kern_file_info *info = data;
2549 struct kinfo_file kf;
2550 struct filedesc *fdp;
2555 if (p->p_stat == SIDL || p->p_stat == SZOMB)
2557 if (!PRISON_CHECK(info->req->td->td_proc->p_ucred, p->p_ucred) != 0)
2559 if ((fdp = p->p_fd) == NULL)
2561 spin_lock_rd(&fdp->fd_spin);
2562 for (n = 0; n < fdp->fd_nfiles; ++n) {
2563 if ((fp = fdp->fd_files[n].fp) == NULL)
2565 if (info->req->oldptr == NULL) {
2568 uid = p->p_ucred ? p->p_ucred->cr_uid : -1;
2569 kcore_make_file(&kf, fp, p->p_pid, uid, n);
2570 spin_unlock_rd(&fdp->fd_spin);
2571 info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf));
2572 spin_lock_rd(&fdp->fd_spin);
2577 spin_unlock_rd(&fdp->fd_spin);
2583 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2584 0, 0, sysctl_kern_file, "S,file", "Entire file table");
2586 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2587 &maxfilesperproc, 0, "Maximum files allowed open per process");
2589 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2590 &maxfiles, 0, "Maximum number of files");
2592 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW,
2593 &maxfilesrootres, 0, "Descriptors reserved for root use");
2595 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2596 &nfiles, 0, "System-wide number of open files");
2599 fildesc_drvinit(void *unused)
2603 dev_ops_add(&fildesc_ops, 0, 0);
2604 for (fd = 0; fd < NUMFDESC; fd++) {
2605 make_dev(&fildesc_ops, fd,
2606 UID_BIN, GID_BIN, 0666, "fd/%d", fd);
2608 make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin");
2609 make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout");
2610 make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr");
2616 struct fileops badfileops = {
2617 .fo_read = badfo_readwrite,
2618 .fo_write = badfo_readwrite,
2619 .fo_ioctl = badfo_ioctl,
2620 .fo_poll = badfo_poll,
2621 .fo_kqfilter = badfo_kqfilter,
2622 .fo_stat = badfo_stat,
2623 .fo_close = badfo_close,
2624 .fo_shutdown = badfo_shutdown
2644 badfo_ioctl(struct file *fp, u_long com, caddr_t data, struct ucred *cred)
2653 badfo_poll(struct file *fp, int events, struct ucred *cred)
2662 badfo_kqfilter(struct file *fp, struct knote *kn)
2668 badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
2677 badfo_close(struct file *fp)
2686 badfo_shutdown(struct file *fp, int how)
2695 nofo_shutdown(struct file *fp, int how)
2697 return (EOPNOTSUPP);
2700 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,
2701 fildesc_drvinit,NULL)