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>
109 static void fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd);
110 static void fdreserve_locked (struct filedesc *fdp, int fd0, int incr);
111 static struct file *funsetfd_locked (struct filedesc *fdp, int fd);
112 static int checkfpclosed(struct filedesc *fdp, int fd, struct file *fp);
113 static void ffree(struct file *fp);
115 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
116 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
117 "file desc to leader structures");
118 MALLOC_DEFINE(M_FILE, "file", "Open file structure");
119 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
121 static d_open_t fdopen;
124 #define CDEV_MAJOR 22
125 static struct dev_ops fildesc_ops = {
126 { "FD", CDEV_MAJOR, 0 },
131 * Descriptor management.
133 static struct filelist filehead = LIST_HEAD_INITIALIZER(&filehead);
134 static struct spinlock filehead_spin = SPINLOCK_INITIALIZER(&filehead_spin);
135 static int nfiles; /* actual number of open files */
139 * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared.
141 * MPSAFE - must be called with fdp->fd_spin exclusively held
145 fdfixup_locked(struct filedesc *fdp, int fd)
147 if (fd < fdp->fd_freefile) {
148 fdp->fd_freefile = fd;
150 while (fdp->fd_lastfile >= 0 &&
151 fdp->fd_files[fdp->fd_lastfile].fp == NULL &&
152 fdp->fd_files[fdp->fd_lastfile].reserved == 0
159 * System calls on descriptors.
164 sys_getdtablesize(struct getdtablesize_args *uap)
166 struct proc *p = curproc;
167 struct plimit *limit = p->p_limit;
169 spin_lock_rd(&limit->p_spin);
171 min((int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
172 spin_unlock_rd(&limit->p_spin);
177 * Duplicate a file descriptor to a particular value.
179 * note: keep in mind that a potential race condition exists when closing
180 * descriptors from a shared descriptor table (via rfork).
185 sys_dup2(struct dup2_args *uap)
190 error = kern_dup(DUP_FIXED, uap->from, uap->to, &fd);
191 uap->sysmsg_fds[0] = fd;
197 * Duplicate a file descriptor.
202 sys_dup(struct dup_args *uap)
207 error = kern_dup(DUP_VARIABLE, uap->fd, 0, &fd);
208 uap->sysmsg_fds[0] = fd;
214 * MPALMOSTSAFE - acquires mplock for fp operations
217 kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred)
219 struct thread *td = curthread;
220 struct proc *p = td->td_proc;
226 int tmp, error, flg = F_POSIX;
231 * Operations on file descriptors that do not require a file pointer.
235 error = fgetfdflags(p->p_fd, fd, &tmp);
237 dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0;
241 if (dat->fc_cloexec & FD_CLOEXEC)
242 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
244 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
248 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd);
255 * Operations on file pointers
257 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
263 dat->fc_flags = OFLAGS(fp->f_flag);
269 nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS;
270 nflags |= oflags & ~FCNTLFLAGS;
273 if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY))
275 if (error == 0 && ((nflags ^ oflags) & FASYNC)) {
276 tmp = nflags & FASYNC;
277 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp,
285 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner,
290 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner,
296 /* Fall into F_SETLK */
299 if (fp->f_type != DTYPE_VNODE) {
303 vp = (struct vnode *)fp->f_data;
306 * copyin/lockop may block
308 if (dat->fc_flock.l_whence == SEEK_CUR)
309 dat->fc_flock.l_start += fp->f_offset;
311 switch (dat->fc_flock.l_type) {
313 if ((fp->f_flag & FREAD) == 0) {
317 p->p_leader->p_flag |= P_ADVLOCK;
318 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
319 &dat->fc_flock, flg);
322 if ((fp->f_flag & FWRITE) == 0) {
326 p->p_leader->p_flag |= P_ADVLOCK;
327 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
328 &dat->fc_flock, flg);
331 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
332 &dat->fc_flock, F_POSIX);
340 * It is possible to race a close() on the descriptor while
341 * we were blocked getting the lock. If this occurs the
342 * close might not have caught the lock.
344 if (checkfpclosed(p->p_fd, fd, fp)) {
345 dat->fc_flock.l_whence = SEEK_SET;
346 dat->fc_flock.l_start = 0;
347 dat->fc_flock.l_len = 0;
348 dat->fc_flock.l_type = F_UNLCK;
349 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
350 F_UNLCK, &dat->fc_flock, F_POSIX);
355 if (fp->f_type != DTYPE_VNODE) {
359 vp = (struct vnode *)fp->f_data;
361 * copyin/lockop may block
363 if (dat->fc_flock.l_type != F_RDLCK &&
364 dat->fc_flock.l_type != F_WRLCK &&
365 dat->fc_flock.l_type != F_UNLCK) {
369 if (dat->fc_flock.l_whence == SEEK_CUR)
370 dat->fc_flock.l_start += fp->f_offset;
371 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK,
372 &dat->fc_flock, F_POSIX);
385 * The file control system call.
390 sys_fcntl(struct fcntl_args *uap)
397 dat.fc_fd = uap->arg;
400 dat.fc_cloexec = uap->arg;
403 dat.fc_flags = uap->arg;
406 dat.fc_owner = uap->arg;
411 error = copyin((caddr_t)uap->arg, &dat.fc_flock,
412 sizeof(struct flock));
418 error = kern_fcntl(uap->fd, uap->cmd, &dat, curthread->td_ucred);
423 uap->sysmsg_result = dat.fc_fd;
426 uap->sysmsg_result = dat.fc_cloexec;
429 uap->sysmsg_result = dat.fc_flags;
432 uap->sysmsg_result = dat.fc_owner;
434 error = copyout(&dat.fc_flock, (caddr_t)uap->arg,
435 sizeof(struct flock));
444 * Common code for dup, dup2, and fcntl(F_DUPFD).
446 * The type flag can be either DUP_FIXED or DUP_VARIABLE. DUP_FIXED tells
447 * kern_dup() to destructively dup over an existing file descriptor if new
448 * is already open. DUP_VARIABLE tells kern_dup() to find the lowest
449 * unused file descriptor that is greater than or equal to new.
454 kern_dup(enum dup_type type, int old, int new, int *res)
456 struct thread *td = curthread;
457 struct proc *p = td->td_proc;
458 struct filedesc *fdp = p->p_fd;
466 * Verify that we have a valid descriptor to dup from and
467 * possibly to dup to.
470 spin_lock_wr(&fdp->fd_spin);
471 if (new < 0 || new > p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
472 new >= maxfilesperproc) {
473 spin_unlock_wr(&fdp->fd_spin);
476 if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) {
477 spin_unlock_wr(&fdp->fd_spin);
480 if (type == DUP_FIXED && old == new) {
482 spin_unlock_wr(&fdp->fd_spin);
485 fp = fdp->fd_files[old].fp;
486 oldflags = fdp->fd_files[old].fileflags;
487 fhold(fp); /* MPSAFE - can be called with a spinlock held */
490 * Allocate a new descriptor if DUP_VARIABLE, or expand the table
491 * if the requested descriptor is beyond the current table size.
493 * This can block. Retry if the source descriptor no longer matches
494 * or if our expectation in the expansion case races.
496 * If we are not expanding or allocating a new decriptor, then reset
497 * the target descriptor to a reserved state so we have a uniform
498 * setup for the next code block.
500 if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) {
501 spin_unlock_wr(&fdp->fd_spin);
502 error = fdalloc(p, new, &newfd);
503 spin_lock_wr(&fdp->fd_spin);
505 spin_unlock_wr(&fdp->fd_spin);
512 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) {
513 fsetfd_locked(fdp, NULL, newfd);
514 spin_unlock_wr(&fdp->fd_spin);
519 * Check for expansion race
521 if (type != DUP_VARIABLE && new != newfd) {
522 fsetfd_locked(fdp, NULL, newfd);
523 spin_unlock_wr(&fdp->fd_spin);
528 * Check for ripout, newfd reused old (this case probably
532 fsetfd_locked(fdp, NULL, newfd);
533 spin_unlock_wr(&fdp->fd_spin);
540 if (fdp->fd_files[new].reserved) {
541 spin_unlock_wr(&fdp->fd_spin);
543 kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new);
544 tsleep(fdp, 0, "fdres", hz);
549 * If the target descriptor was never allocated we have
550 * to allocate it. If it was we have to clean out the
551 * old descriptor. delfp inherits the ref from the
554 delfp = fdp->fd_files[new].fp;
555 fdp->fd_files[new].fp = NULL;
556 fdp->fd_files[new].reserved = 1;
558 fdreserve_locked(fdp, new, 1);
559 if (new > fdp->fd_lastfile)
560 fdp->fd_lastfile = new;
566 * NOTE: still holding an exclusive spinlock
570 * If a descriptor is being overwritten we may hve to tell
571 * fdfree() to sleep to ensure that all relevant process
572 * leaders can be traversed in closef().
574 if (delfp != NULL && p->p_fdtol != NULL) {
575 fdp->fd_holdleaderscount++;
580 KASSERT(delfp == NULL || type == DUP_FIXED,
581 ("dup() picked an open file"));
584 * Duplicate the source descriptor, update lastfile. If the new
585 * descriptor was not allocated and we aren't replacing an existing
586 * descriptor we have to mark the descriptor as being in use.
588 * The fd_files[] array inherits fp's hold reference.
590 fsetfd_locked(fdp, fp, new);
591 fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE;
592 spin_unlock_wr(&fdp->fd_spin);
597 * If we dup'd over a valid file, we now own the reference to it
598 * and must dispose of it using closef() semantics (as if a
599 * close() were performed on it).
604 spin_lock_wr(&fdp->fd_spin);
605 fdp->fd_holdleaderscount--;
606 if (fdp->fd_holdleaderscount == 0 &&
607 fdp->fd_holdleaderswakeup != 0) {
608 fdp->fd_holdleaderswakeup = 0;
609 spin_unlock_wr(&fdp->fd_spin);
610 wakeup(&fdp->fd_holdleaderscount);
612 spin_unlock_wr(&fdp->fd_spin);
620 * If sigio is on the list associated with a process or process group,
621 * disable signalling from the device, remove sigio from the list and
625 funsetown(struct sigio *sigio)
630 *(sigio->sio_myref) = NULL;
632 if (sigio->sio_pgid < 0) {
633 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
635 } else /* if ((*sigiop)->sio_pgid > 0) */ {
636 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
639 crfree(sigio->sio_ucred);
640 kfree(sigio, M_SIGIO);
643 /* Free a list of sigio structures. */
645 funsetownlst(struct sigiolst *sigiolst)
649 while ((sigio = SLIST_FIRST(sigiolst)) != NULL)
654 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
656 * After permission checking, add a sigio structure to the sigio list for
657 * the process or process group.
660 fsetown(pid_t pgid, struct sigio **sigiop)
676 * Policy - Don't allow a process to FSETOWN a process
677 * in another session.
679 * Remove this test to allow maximum flexibility or
680 * restrict FSETOWN to the current process or process
681 * group for maximum safety.
683 if (proc->p_session != curproc->p_session)
687 } else /* if (pgid < 0) */ {
688 pgrp = pgfind(-pgid);
693 * Policy - Don't allow a process to FSETOWN a process
694 * in another session.
696 * Remove this test to allow maximum flexibility or
697 * restrict FSETOWN to the current process or process
698 * group for maximum safety.
700 if (pgrp->pg_session != curproc->p_session)
706 sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
708 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
709 sigio->sio_proc = proc;
711 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
712 sigio->sio_pgrp = pgrp;
714 sigio->sio_pgid = pgid;
715 sigio->sio_ucred = crhold(curthread->td_ucred);
716 /* It would be convenient if p_ruid was in ucred. */
717 sigio->sio_ruid = sigio->sio_ucred->cr_ruid;
718 sigio->sio_myref = sigiop;
726 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
729 fgetown(struct sigio *sigio)
731 return (sigio != NULL ? sigio->sio_pgid : 0);
735 * Close many file descriptors.
740 sys_closefrom(struct closefrom_args *uap)
742 return(kern_closefrom(uap->fd));
746 * Close all file descriptors greater then or equal to fd
751 kern_closefrom(int fd)
753 struct thread *td = curthread;
754 struct proc *p = td->td_proc;
755 struct filedesc *fdp;
764 * NOTE: This function will skip unassociated descriptors and
765 * reserved descriptors that have not yet been assigned.
766 * fd_lastfile can change as a side effect of kern_close().
768 spin_lock_wr(&fdp->fd_spin);
769 while (fd <= fdp->fd_lastfile) {
770 if (fdp->fd_files[fd].fp != NULL) {
771 spin_unlock_wr(&fdp->fd_spin);
772 /* ok if this races another close */
773 if (kern_close(fd) == EINTR)
775 spin_lock_wr(&fdp->fd_spin);
779 spin_unlock_wr(&fdp->fd_spin);
784 * Close a file descriptor.
789 sys_close(struct close_args *uap)
791 return(kern_close(uap->fd));
795 * MPALMOSTSAFE - acquires mplock around knote_fdclose() calls
800 struct thread *td = curthread;
801 struct proc *p = td->td_proc;
802 struct filedesc *fdp;
810 spin_lock_wr(&fdp->fd_spin);
811 if ((fp = funsetfd_locked(fdp, fd)) == NULL) {
812 spin_unlock_wr(&fdp->fd_spin);
816 if (p->p_fdtol != NULL) {
818 * Ask fdfree() to sleep to ensure that all relevant
819 * process leaders can be traversed in closef().
821 fdp->fd_holdleaderscount++;
826 * we now hold the fp reference that used to be owned by the descriptor
829 spin_unlock_wr(&fdp->fd_spin);
830 if (fd < fdp->fd_knlistsize) {
832 if (fd < fdp->fd_knlistsize)
833 knote_fdclose(p, fd);
836 error = closef(fp, p);
838 spin_lock_wr(&fdp->fd_spin);
839 fdp->fd_holdleaderscount--;
840 if (fdp->fd_holdleaderscount == 0 &&
841 fdp->fd_holdleaderswakeup != 0) {
842 fdp->fd_holdleaderswakeup = 0;
843 spin_unlock_wr(&fdp->fd_spin);
844 wakeup(&fdp->fd_holdleaderscount);
846 spin_unlock_wr(&fdp->fd_spin);
853 * shutdown_args(int fd, int how)
856 kern_shutdown(int fd, int how)
858 struct thread *td = curthread;
859 struct proc *p = td->td_proc;
865 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
867 error = fo_shutdown(fp, how);
877 sys_shutdown(struct shutdown_args *uap)
882 error = kern_shutdown(uap->s, uap->how);
892 kern_fstat(int fd, struct stat *ub)
894 struct thread *td = curthread;
895 struct proc *p = td->td_proc;
901 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
903 error = fo_stat(fp, ub, td->td_ucred);
910 * Return status information about a file descriptor.
915 sys_fstat(struct fstat_args *uap)
920 error = kern_fstat(uap->fd, &st);
923 error = copyout(&st, uap->sb, sizeof(st));
928 * Return pathconf information about a file descriptor.
933 sys_fpathconf(struct fpathconf_args *uap)
935 struct thread *td = curthread;
936 struct proc *p = td->td_proc;
941 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
944 switch (fp->f_type) {
947 if (uap->name != _PC_PIPE_BUF) {
950 uap->sysmsg_result = PIPE_BUF;
956 vp = (struct vnode *)fp->f_data;
958 error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg);
970 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0, "");
973 * Grow the file table so it can hold through descriptor (want).
975 * The fdp's spinlock must be held exclusively on entry and may be held
976 * exclusively on return. The spinlock may be cycled by the routine.
981 fdgrow_locked(struct filedesc *fdp, int want)
983 struct fdnode *newfiles;
984 struct fdnode *oldfiles;
989 /* nf has to be of the form 2^n - 1 */
991 } while (nf <= want);
993 spin_unlock_wr(&fdp->fd_spin);
994 newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK);
995 spin_lock_wr(&fdp->fd_spin);
998 * We could have raced another extend while we were not holding
1001 if (fdp->fd_nfiles >= nf) {
1002 spin_unlock_wr(&fdp->fd_spin);
1003 kfree(newfiles, M_FILEDESC);
1004 spin_lock_wr(&fdp->fd_spin);
1008 * Copy the existing ofile and ofileflags arrays
1009 * and zero the new portion of each array.
1011 extra = nf - fdp->fd_nfiles;
1012 bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode));
1013 bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode));
1015 oldfiles = fdp->fd_files;
1016 fdp->fd_files = newfiles;
1017 fdp->fd_nfiles = nf;
1019 if (oldfiles != fdp->fd_builtin_files) {
1020 spin_unlock_wr(&fdp->fd_spin);
1021 kfree(oldfiles, M_FILEDESC);
1022 spin_lock_wr(&fdp->fd_spin);
1028 * Number of nodes in right subtree, including the root.
1031 right_subtree_size(int n)
1033 return (n ^ (n | (n + 1)));
1040 right_ancestor(int n)
1042 return (n | (n + 1));
1049 left_ancestor(int n)
1051 return ((n & (n + 1)) - 1);
1055 * Traverse the in-place binary tree buttom-up adjusting the allocation
1056 * count so scans can determine where free descriptors are located.
1058 * MPSAFE - caller must be holding an exclusive spinlock on fdp
1062 fdreserve_locked(struct filedesc *fdp, int fd, int incr)
1065 fdp->fd_files[fd].allocated += incr;
1066 KKASSERT(fdp->fd_files[fd].allocated >= 0);
1067 fd = left_ancestor(fd);
1072 * Reserve a file descriptor for the process. If no error occurs, the
1073 * caller MUST at some point call fsetfd() or assign a file pointer
1074 * or dispose of the reservation.
1079 fdalloc(struct proc *p, int want, int *result)
1081 struct filedesc *fdp = p->p_fd;
1082 int fd, rsize, rsum, node, lim;
1084 spin_lock_rd(&p->p_limit->p_spin);
1085 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
1086 spin_unlock_rd(&p->p_limit->p_spin);
1089 spin_lock_wr(&fdp->fd_spin);
1090 if (want >= fdp->fd_nfiles)
1091 fdgrow_locked(fdp, want);
1094 * Search for a free descriptor starting at the higher
1095 * of want or fd_freefile. If that fails, consider
1096 * expanding the ofile array.
1098 * NOTE! the 'allocated' field is a cumulative recursive allocation
1099 * count. If we happen to see a value of 0 then we can shortcut
1100 * our search. Otherwise we run through through the tree going
1101 * down branches we know have free descriptor(s) until we hit a
1102 * leaf node. The leaf node will be free but will not necessarily
1103 * have an allocated field of 0.
1106 /* move up the tree looking for a subtree with a free node */
1107 for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim);
1108 fd = right_ancestor(fd)) {
1109 if (fdp->fd_files[fd].allocated == 0)
1112 rsize = right_subtree_size(fd);
1113 if (fdp->fd_files[fd].allocated == rsize)
1114 continue; /* right subtree full */
1117 * Free fd is in the right subtree of the tree rooted at fd.
1118 * Call that subtree R. Look for the smallest (leftmost)
1119 * subtree of R with an unallocated fd: continue moving
1120 * down the left branch until encountering a full left
1121 * subtree, then move to the right.
1123 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) {
1125 rsum += fdp->fd_files[node].allocated;
1126 if (fdp->fd_files[fd].allocated == rsum + rsize) {
1127 fd = node; /* move to the right */
1128 if (fdp->fd_files[node].allocated == 0)
1137 * No space in current array. Expand?
1139 if (fdp->fd_nfiles >= lim) {
1140 spin_unlock_wr(&fdp->fd_spin);
1143 fdgrow_locked(fdp, want);
1147 KKASSERT(fd < fdp->fd_nfiles);
1148 if (fd > fdp->fd_lastfile)
1149 fdp->fd_lastfile = fd;
1150 if (want <= fdp->fd_freefile)
1151 fdp->fd_freefile = fd;
1153 KKASSERT(fdp->fd_files[fd].fp == NULL);
1154 KKASSERT(fdp->fd_files[fd].reserved == 0);
1155 fdp->fd_files[fd].fileflags = 0;
1156 fdp->fd_files[fd].reserved = 1;
1157 fdreserve_locked(fdp, fd, 1);
1158 spin_unlock_wr(&fdp->fd_spin);
1163 * Check to see whether n user file descriptors
1164 * are available to the process p.
1169 fdavail(struct proc *p, int n)
1171 struct filedesc *fdp = p->p_fd;
1172 struct fdnode *fdnode;
1175 spin_lock_rd(&p->p_limit->p_spin);
1176 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
1177 spin_unlock_rd(&p->p_limit->p_spin);
1179 spin_lock_rd(&fdp->fd_spin);
1180 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) {
1181 spin_unlock_rd(&fdp->fd_spin);
1184 last = min(fdp->fd_nfiles, lim);
1185 fdnode = &fdp->fd_files[fdp->fd_freefile];
1186 for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) {
1187 if (fdnode->fp == NULL && --n <= 0) {
1188 spin_unlock_rd(&fdp->fd_spin);
1192 spin_unlock_rd(&fdp->fd_spin);
1197 * Revoke open descriptors referencing (f_data, f_type)
1199 * Any revoke executed within a prison is only able to
1200 * revoke descriptors for processes within that prison.
1202 * Returns 0 on success or an error code.
1204 struct fdrevoke_info {
1214 static int fdrevoke_check_callback(struct file *fp, void *vinfo);
1215 static int fdrevoke_proc_callback(struct proc *p, void *vinfo);
1218 fdrevoke(void *f_data, short f_type, struct ucred *cred)
1220 struct fdrevoke_info info;
1223 bzero(&info, sizeof(info));
1227 error = falloc(NULL, &info.nfp, NULL);
1232 * Scan the file pointer table once. dups do not dup file pointers,
1233 * only descriptors, so there is no leak. Set FREVOKED on the fps
1236 allfiles_scan_exclusive(fdrevoke_check_callback, &info);
1239 * If any fps were marked track down the related descriptors
1240 * and close them. Any dup()s at this point will notice
1241 * the FREVOKED already set in the fp and do the right thing.
1243 * Any fps with non-zero msgcounts (aka sent over a unix-domain
1244 * socket) bumped the intransit counter and will require a
1245 * scan. Races against fps leaving the socket are closed by
1246 * the socket code checking for FREVOKED.
1249 allproc_scan(fdrevoke_proc_callback, &info);
1251 unp_revoke_gc(info.nfp);
1257 * Locate matching file pointers directly.
1260 fdrevoke_check_callback(struct file *fp, void *vinfo)
1262 struct fdrevoke_info *info = vinfo;
1265 * File pointers already flagged for revokation are skipped.
1267 if (fp->f_flag & FREVOKED)
1271 * If revoking from a prison file pointers created outside of
1272 * that prison, or file pointers without creds, cannot be revoked.
1274 if (info->cred->cr_prison &&
1275 (fp->f_cred == NULL ||
1276 info->cred->cr_prison != fp->f_cred->cr_prison)) {
1281 * If the file pointer matches then mark it for revocation. The
1282 * flag is currently only used by unp_revoke_gc().
1284 * info->count is a heuristic and can race in a SMP environment.
1286 if (info->data == fp->f_data && info->type == fp->f_type) {
1287 atomic_set_int(&fp->f_flag, FREVOKED);
1288 info->count += fp->f_count;
1296 * Locate matching file pointers via process descriptor tables.
1299 fdrevoke_proc_callback(struct proc *p, void *vinfo)
1301 struct fdrevoke_info *info = vinfo;
1302 struct filedesc *fdp;
1306 if (p->p_stat == SIDL || p->p_stat == SZOMB)
1308 if (info->cred->cr_prison &&
1309 info->cred->cr_prison != p->p_ucred->cr_prison) {
1314 * If the controlling terminal of the process matches the
1315 * vnode being revoked we clear the controlling terminal.
1317 * The normal spec_close() may not catch this because it
1318 * uses curproc instead of p.
1320 if (p->p_session && info->type == DTYPE_VNODE &&
1321 info->data == p->p_session->s_ttyvp) {
1322 p->p_session->s_ttyvp = NULL;
1327 * Softref the fdp to prevent it from being destroyed
1329 spin_lock_wr(&p->p_spin);
1330 if ((fdp = p->p_fd) == NULL) {
1331 spin_unlock_wr(&p->p_spin);
1334 atomic_add_int(&fdp->fd_softrefs, 1);
1335 spin_unlock_wr(&p->p_spin);
1338 * Locate and close any matching file descriptors.
1340 spin_lock_wr(&fdp->fd_spin);
1341 for (n = 0; n < fdp->fd_nfiles; ++n) {
1342 if ((fp = fdp->fd_files[n].fp) == NULL)
1344 if (fp->f_flag & FREVOKED) {
1346 fdp->fd_files[n].fp = info->nfp;
1347 spin_unlock_wr(&fdp->fd_spin);
1349 spin_lock_wr(&fdp->fd_spin);
1353 spin_unlock_wr(&fdp->fd_spin);
1354 atomic_subtract_int(&fdp->fd_softrefs, 1);
1360 * Create a new open file structure and reserve a file decriptor
1361 * for the process that refers to it.
1363 * Root creds are checked using lp, or assumed if lp is NULL. If
1364 * resultfd is non-NULL then lp must also be non-NULL. No file
1365 * descriptor is reserved (and no process context is needed) if
1368 * A file pointer with a refcount of 1 is returned. Note that the
1369 * file pointer is NOT associated with the descriptor. If falloc
1370 * returns success, fsetfd() MUST be called to either associate the
1371 * file pointer or clear the reservation.
1376 falloc(struct lwp *lp, struct file **resultfp, int *resultfd)
1378 static struct timeval lastfail;
1381 struct ucred *cred = lp ? lp->lwp_thread->td_ucred : proc0.p_ucred;
1387 * Handle filetable full issues and root overfill.
1389 if (nfiles >= maxfiles - maxfilesrootres &&
1390 (cred->cr_ruid != 0 || nfiles >= maxfiles)) {
1391 if (ppsratecheck(&lastfail, &curfail, 1)) {
1392 kprintf("kern.maxfiles limit exceeded by uid %d, "
1393 "please see tuning(7).\n",
1401 * Allocate a new file descriptor.
1403 fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO);
1404 spin_init(&fp->f_spin);
1406 fp->f_ops = &badfileops;
1408 fp->f_cred = crhold(cred);
1409 spin_lock_wr(&filehead_spin);
1411 LIST_INSERT_HEAD(&filehead, fp, f_list);
1412 spin_unlock_wr(&filehead_spin);
1414 if ((error = fdalloc(lp->lwp_proc, 0, resultfd)) != 0) {
1431 checkfpclosed(struct filedesc *fdp, int fd, struct file *fp)
1435 spin_lock_rd(&fdp->fd_spin);
1436 if ((unsigned) fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp)
1440 spin_unlock_rd(&fdp->fd_spin);
1445 * Associate a file pointer with a previously reserved file descriptor.
1446 * This function always succeeds.
1448 * If fp is NULL, the file descriptor is returned to the pool.
1452 * MPSAFE (exclusive spinlock must be held on call)
1455 fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd)
1457 KKASSERT((unsigned)fd < fdp->fd_nfiles);
1458 KKASSERT(fdp->fd_files[fd].reserved != 0);
1461 fdp->fd_files[fd].fp = fp;
1462 fdp->fd_files[fd].reserved = 0;
1463 if (fp->f_type == DTYPE_KQUEUE) {
1464 if (fdp->fd_knlistsize < 0)
1465 fdp->fd_knlistsize = 0;
1468 fdp->fd_files[fd].reserved = 0;
1469 fdreserve_locked(fdp, fd, -1);
1470 fdfixup_locked(fdp, fd);
1478 fsetfd(struct filedesc *fdp, struct file *fp, int fd)
1480 spin_lock_wr(&fdp->fd_spin);
1481 fsetfd_locked(fdp, fp, fd);
1482 spin_unlock_wr(&fdp->fd_spin);
1486 * MPSAFE (exclusive spinlock must be held on call)
1490 funsetfd_locked(struct filedesc *fdp, int fd)
1494 if ((unsigned)fd >= fdp->fd_nfiles)
1496 if ((fp = fdp->fd_files[fd].fp) == NULL)
1498 fdp->fd_files[fd].fp = NULL;
1499 fdp->fd_files[fd].fileflags = 0;
1501 fdreserve_locked(fdp, fd, -1);
1502 fdfixup_locked(fdp, fd);
1510 fgetfdflags(struct filedesc *fdp, int fd, int *flagsp)
1514 spin_lock_rd(&fdp->fd_spin);
1515 if (((u_int)fd) >= fdp->fd_nfiles) {
1517 } else if (fdp->fd_files[fd].fp == NULL) {
1520 *flagsp = fdp->fd_files[fd].fileflags;
1523 spin_unlock_rd(&fdp->fd_spin);
1531 fsetfdflags(struct filedesc *fdp, int fd, int add_flags)
1535 spin_lock_wr(&fdp->fd_spin);
1536 if (((u_int)fd) >= fdp->fd_nfiles) {
1538 } else if (fdp->fd_files[fd].fp == NULL) {
1541 fdp->fd_files[fd].fileflags |= add_flags;
1544 spin_unlock_wr(&fdp->fd_spin);
1552 fclrfdflags(struct filedesc *fdp, int fd, int rem_flags)
1556 spin_lock_wr(&fdp->fd_spin);
1557 if (((u_int)fd) >= fdp->fd_nfiles) {
1559 } else if (fdp->fd_files[fd].fp == NULL) {
1562 fdp->fd_files[fd].fileflags &= ~rem_flags;
1565 spin_unlock_wr(&fdp->fd_spin);
1570 fsetcred(struct file *fp, struct ucred *cr)
1578 * Free a file descriptor.
1582 ffree(struct file *fp)
1584 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!"));
1585 spin_lock_wr(&filehead_spin);
1586 LIST_REMOVE(fp, f_list);
1588 spin_unlock_wr(&filehead_spin);
1590 if (fp->f_nchandle.ncp)
1591 cache_drop(&fp->f_nchandle);
1596 * called from init_main, initialize filedesc0 for proc0.
1599 fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask)
1603 fdp0->fd_refcnt = 1;
1604 fdp0->fd_cmask = cmask;
1605 fdp0->fd_files = fdp0->fd_builtin_files;
1606 fdp0->fd_nfiles = NDFILE;
1607 fdp0->fd_lastfile = -1;
1608 spin_init(&fdp0->fd_spin);
1612 * Build a new filedesc structure.
1617 fdinit(struct proc *p)
1619 struct filedesc *newfdp;
1620 struct filedesc *fdp = p->p_fd;
1622 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO);
1623 spin_lock_rd(&fdp->fd_spin);
1625 newfdp->fd_cdir = fdp->fd_cdir;
1626 vref(newfdp->fd_cdir);
1627 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1631 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of
1632 * proc0, but should unconditionally exist in other processes.
1635 newfdp->fd_rdir = fdp->fd_rdir;
1636 vref(newfdp->fd_rdir);
1637 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1640 newfdp->fd_jdir = fdp->fd_jdir;
1641 vref(newfdp->fd_jdir);
1642 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1644 spin_unlock_rd(&fdp->fd_spin);
1646 /* Create the file descriptor table. */
1647 newfdp->fd_refcnt = 1;
1648 newfdp->fd_cmask = cmask;
1649 newfdp->fd_files = newfdp->fd_builtin_files;
1650 newfdp->fd_nfiles = NDFILE;
1651 newfdp->fd_knlistsize = -1;
1652 newfdp->fd_lastfile = -1;
1653 spin_init(&newfdp->fd_spin);
1659 * Share a filedesc structure.
1664 fdshare(struct proc *p)
1666 struct filedesc *fdp;
1669 spin_lock_wr(&fdp->fd_spin);
1671 spin_unlock_wr(&fdp->fd_spin);
1676 * Copy a filedesc structure.
1681 fdcopy(struct proc *p)
1683 struct filedesc *fdp = p->p_fd;
1684 struct filedesc *newfdp;
1685 struct fdnode *fdnode;
1690 * Certain daemons might not have file descriptors.
1696 * Allocate the new filedesc and fd_files[] array. This can race
1697 * with operations by other threads on the fdp so we have to be
1700 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK | M_ZERO);
1702 spin_lock_rd(&fdp->fd_spin);
1703 if (fdp->fd_lastfile < NDFILE) {
1704 newfdp->fd_files = newfdp->fd_builtin_files;
1708 * We have to allocate (N^2-1) entries for our in-place
1709 * binary tree. Allow the table to shrink.
1713 while (ni > fdp->fd_lastfile && ni > NDFILE) {
1717 spin_unlock_rd(&fdp->fd_spin);
1718 newfdp->fd_files = kmalloc(i * sizeof(struct fdnode),
1719 M_FILEDESC, M_WAITOK | M_ZERO);
1722 * Check for race, retry
1724 spin_lock_rd(&fdp->fd_spin);
1725 if (i <= fdp->fd_lastfile) {
1726 spin_unlock_rd(&fdp->fd_spin);
1727 kfree(newfdp->fd_files, M_FILEDESC);
1733 * Dup the remaining fields. vref() and cache_hold() can be
1734 * safely called while holding the read spinlock on fdp.
1736 * The read spinlock on fdp is still being held.
1738 * NOTE: vref and cache_hold calls for the case where the vnode
1739 * or cache entry already has at least one ref may be called
1740 * while holding spin locks.
1742 if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) {
1743 vref(newfdp->fd_cdir);
1744 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1747 * We must check for fd_rdir here, at least for now because
1748 * the init process is created before we have access to the
1749 * rootvode to take a reference to it.
1751 if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) {
1752 vref(newfdp->fd_rdir);
1753 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1755 if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) {
1756 vref(newfdp->fd_jdir);
1757 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1759 newfdp->fd_refcnt = 1;
1760 newfdp->fd_nfiles = i;
1761 newfdp->fd_lastfile = fdp->fd_lastfile;
1762 newfdp->fd_freefile = fdp->fd_freefile;
1763 newfdp->fd_cmask = fdp->fd_cmask;
1764 newfdp->fd_knlist = NULL;
1765 newfdp->fd_knlistsize = -1;
1766 newfdp->fd_knhash = NULL;
1767 newfdp->fd_knhashmask = 0;
1768 spin_init(&newfdp->fd_spin);
1771 * Copy the descriptor table through (i). This also copies the
1772 * allocation state. Then go through and ref the file pointers
1773 * and clean up any KQ descriptors.
1775 * kq descriptors cannot be copied. Since we haven't ref'd the
1776 * copied files yet we can ignore the return value from funsetfd().
1778 * The read spinlock on fdp is still being held.
1780 bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode));
1781 for (i = 0 ; i < newfdp->fd_nfiles; ++i) {
1782 fdnode = &newfdp->fd_files[i];
1783 if (fdnode->reserved) {
1784 fdreserve_locked(newfdp, i, -1);
1785 fdnode->reserved = 0;
1786 fdfixup_locked(newfdp, i);
1787 } else if (fdnode->fp) {
1788 if (fdnode->fp->f_type == DTYPE_KQUEUE) {
1789 (void)funsetfd_locked(newfdp, i);
1795 spin_unlock_rd(&fdp->fd_spin);
1800 * Release a filedesc structure.
1802 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE)
1805 fdfree(struct proc *p, struct filedesc *repl)
1807 struct filedesc *fdp;
1808 struct fdnode *fdnode;
1810 struct filedesc_to_leader *fdtol;
1816 * Certain daemons might not have file descriptors.
1825 * Severe messing around to follow.
1827 spin_lock_wr(&fdp->fd_spin);
1829 /* Check for special need to clear POSIX style locks */
1831 if (fdtol != NULL) {
1832 KASSERT(fdtol->fdl_refcount > 0,
1833 ("filedesc_to_refcount botch: fdl_refcount=%d",
1834 fdtol->fdl_refcount));
1835 if (fdtol->fdl_refcount == 1 &&
1836 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1837 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1838 fdnode = &fdp->fd_files[i];
1839 if (fdnode->fp == NULL ||
1840 fdnode->fp->f_type != DTYPE_VNODE) {
1845 spin_unlock_wr(&fdp->fd_spin);
1847 lf.l_whence = SEEK_SET;
1850 lf.l_type = F_UNLCK;
1851 vp = (struct vnode *)fp->f_data;
1852 (void) VOP_ADVLOCK(vp,
1853 (caddr_t)p->p_leader,
1858 spin_lock_wr(&fdp->fd_spin);
1862 if (fdtol->fdl_refcount == 1) {
1863 if (fdp->fd_holdleaderscount > 0 &&
1864 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1866 * close() or do_dup() has cleared a reference
1867 * in a shared file descriptor table.
1869 fdp->fd_holdleaderswakeup = 1;
1870 ssleep(&fdp->fd_holdleaderscount,
1871 &fdp->fd_spin, 0, "fdlhold", 0);
1874 if (fdtol->fdl_holdcount > 0) {
1876 * Ensure that fdtol->fdl_leader
1877 * remains valid in closef().
1879 fdtol->fdl_wakeup = 1;
1880 ssleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0);
1884 fdtol->fdl_refcount--;
1885 if (fdtol->fdl_refcount == 0 &&
1886 fdtol->fdl_holdcount == 0) {
1887 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1888 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1893 if (fdtol != NULL) {
1894 spin_unlock_wr(&fdp->fd_spin);
1895 kfree(fdtol, M_FILEDESC_TO_LEADER);
1896 spin_lock_wr(&fdp->fd_spin);
1899 if (--fdp->fd_refcnt > 0) {
1900 spin_unlock_wr(&fdp->fd_spin);
1901 spin_lock_wr(&p->p_spin);
1903 spin_unlock_wr(&p->p_spin);
1908 * Even though we are the last reference to the structure allproc
1909 * scans may still reference the structure. Maintain proper
1910 * locks until we can replace p->p_fd.
1912 * Also note that kqueue's closef still needs to reference the
1913 * fdp via p->p_fd, so we have to close the descriptors before
1914 * we replace p->p_fd.
1916 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1917 if (fdp->fd_files[i].fp) {
1918 fp = funsetfd_locked(fdp, i);
1920 spin_unlock_wr(&fdp->fd_spin);
1922 spin_lock_wr(&fdp->fd_spin);
1926 spin_unlock_wr(&fdp->fd_spin);
1929 * Interlock against an allproc scan operations (typically frevoke).
1931 spin_lock_wr(&p->p_spin);
1933 spin_unlock_wr(&p->p_spin);
1936 * Wait for any softrefs to go away. This race rarely occurs so
1937 * we can use a non-critical-path style poll/sleep loop. The
1938 * race only occurs against allproc scans.
1940 * No new softrefs can occur with the fdp disconnected from the
1943 if (fdp->fd_softrefs) {
1944 kprintf("pid %d: Warning, fdp race avoided\n", p->p_pid);
1945 while (fdp->fd_softrefs)
1946 tsleep(&fdp->fd_softrefs, 0, "fdsoft", 1);
1949 if (fdp->fd_files != fdp->fd_builtin_files)
1950 kfree(fdp->fd_files, M_FILEDESC);
1952 cache_drop(&fdp->fd_ncdir);
1953 vrele(fdp->fd_cdir);
1956 cache_drop(&fdp->fd_nrdir);
1957 vrele(fdp->fd_rdir);
1960 cache_drop(&fdp->fd_njdir);
1961 vrele(fdp->fd_jdir);
1964 kfree(fdp->fd_knlist, M_KQUEUE);
1966 kfree(fdp->fd_knhash, M_KQUEUE);
1967 kfree(fdp, M_FILEDESC);
1971 * Retrieve and reference the file pointer associated with a descriptor.
1976 holdfp(struct filedesc *fdp, int fd, int flag)
1980 spin_lock_rd(&fdp->fd_spin);
1981 if (((u_int)fd) >= fdp->fd_nfiles) {
1985 if ((fp = fdp->fd_files[fd].fp) == NULL)
1987 if ((fp->f_flag & flag) == 0 && flag != -1) {
1993 spin_unlock_rd(&fdp->fd_spin);
1998 * holdsock() - load the struct file pointer associated
1999 * with a socket into *fpp. If an error occurs, non-zero
2000 * will be returned and *fpp will be set to NULL.
2005 holdsock(struct filedesc *fdp, int fd, struct file **fpp)
2010 spin_lock_rd(&fdp->fd_spin);
2011 if ((unsigned)fd >= fdp->fd_nfiles) {
2016 if ((fp = fdp->fd_files[fd].fp) == NULL) {
2020 if (fp->f_type != DTYPE_SOCKET) {
2027 spin_unlock_rd(&fdp->fd_spin);
2033 * Convert a user file descriptor to a held file pointer.
2038 holdvnode(struct filedesc *fdp, int fd, struct file **fpp)
2043 spin_lock_rd(&fdp->fd_spin);
2044 if ((unsigned)fd >= fdp->fd_nfiles) {
2049 if ((fp = fdp->fd_files[fd].fp) == NULL) {
2053 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
2061 spin_unlock_rd(&fdp->fd_spin);
2067 * For setugid programs, we don't want to people to use that setugidness
2068 * to generate error messages which write to a file which otherwise would
2069 * otherwise be off-limits to the process.
2071 * This is a gross hack to plug the hole. A better solution would involve
2072 * a special vop or other form of generalized access control mechanism. We
2073 * go ahead and just reject all procfs file systems accesses as dangerous.
2075 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2076 * sufficient. We also don't for check setugidness since we know we are.
2079 is_unsafe(struct file *fp)
2081 if (fp->f_type == DTYPE_VNODE &&
2082 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS)
2088 * Make this setguid thing safe, if at all possible.
2090 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2093 setugidsafety(struct proc *p)
2095 struct filedesc *fdp = p->p_fd;
2098 /* Certain daemons might not have file descriptors. */
2103 * note: fdp->fd_files may be reallocated out from under us while
2104 * we are blocked in a close. Be careful!
2106 for (i = 0; i <= fdp->fd_lastfile; i++) {
2109 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) {
2112 if (i < fdp->fd_knlistsize)
2113 knote_fdclose(p, i);
2115 * NULL-out descriptor prior to close to avoid
2116 * a race while close blocks.
2118 if ((fp = funsetfd_locked(fdp, i)) != NULL)
2125 * Close any files on exec?
2127 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2130 fdcloseexec(struct proc *p)
2132 struct filedesc *fdp = p->p_fd;
2135 /* Certain daemons might not have file descriptors. */
2140 * We cannot cache fd_files since operations may block and rip
2141 * them out from under us.
2143 for (i = 0; i <= fdp->fd_lastfile; i++) {
2144 if (fdp->fd_files[i].fp != NULL &&
2145 (fdp->fd_files[i].fileflags & UF_EXCLOSE)) {
2148 if (i < fdp->fd_knlistsize)
2149 knote_fdclose(p, i);
2151 * NULL-out descriptor prior to close to avoid
2152 * a race while close blocks.
2154 if ((fp = funsetfd_locked(fdp, i)) != NULL)
2161 * It is unsafe for set[ug]id processes to be started with file
2162 * descriptors 0..2 closed, as these descriptors are given implicit
2163 * significance in the Standard C library. fdcheckstd() will create a
2164 * descriptor referencing /dev/null for each of stdin, stdout, and
2165 * stderr that is not already open.
2167 * NOT MPSAFE - calls falloc, vn_open, etc
2170 fdcheckstd(struct lwp *lp)
2172 struct nlookupdata nd;
2173 struct filedesc *fdp;
2176 int i, error, flags, devnull;
2178 fdp = lp->lwp_proc->p_fd;
2183 for (i = 0; i < 3; i++) {
2184 if (fdp->fd_files[i].fp != NULL)
2187 if ((error = falloc(lp, &fp, &devnull)) != 0)
2190 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE,
2191 NLC_FOLLOW|NLC_LOCKVP);
2192 flags = FREAD | FWRITE;
2194 error = vn_open(&nd, fp, flags, 0);
2196 fsetfd(fdp, fp, devnull);
2198 fsetfd(fdp, NULL, devnull);
2203 KKASSERT(i == devnull);
2205 error = kern_dup(DUP_FIXED, devnull, i, &retval);
2214 * Internal form of close.
2215 * Decrement reference count on file structure.
2216 * Note: td and/or p may be NULL when closing a file
2217 * that was being passed in a message.
2219 * MPALMOSTSAFE - acquires mplock for VOP operations
2222 closef(struct file *fp, struct proc *p)
2226 struct filedesc_to_leader *fdtol;
2232 * POSIX record locking dictates that any close releases ALL
2233 * locks owned by this process. This is handled by setting
2234 * a flag in the unlock to free ONLY locks obeying POSIX
2235 * semantics, and not to free BSD-style file locks.
2236 * If the descriptor was in a message, POSIX-style locks
2237 * aren't passed with the descriptor.
2239 if (p != NULL && fp->f_type == DTYPE_VNODE &&
2240 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2243 if ((p->p_leader->p_flag & P_ADVLOCK) != 0) {
2244 lf.l_whence = SEEK_SET;
2247 lf.l_type = F_UNLCK;
2248 vp = (struct vnode *)fp->f_data;
2249 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
2253 if (fdtol != NULL) {
2255 * Handle special case where file descriptor table
2256 * is shared between multiple process leaders.
2258 for (fdtol = fdtol->fdl_next;
2259 fdtol != p->p_fdtol;
2260 fdtol = fdtol->fdl_next) {
2261 if ((fdtol->fdl_leader->p_flag &
2264 fdtol->fdl_holdcount++;
2265 lf.l_whence = SEEK_SET;
2268 lf.l_type = F_UNLCK;
2269 vp = (struct vnode *)fp->f_data;
2270 (void) VOP_ADVLOCK(vp,
2271 (caddr_t)fdtol->fdl_leader,
2272 F_UNLCK, &lf, F_POSIX);
2273 fdtol->fdl_holdcount--;
2274 if (fdtol->fdl_holdcount == 0 &&
2275 fdtol->fdl_wakeup != 0) {
2276 fdtol->fdl_wakeup = 0;
2289 * fhold() can only be called if f_count is already at least 1 (i.e. the
2290 * caller of fhold() already has a reference to the file pointer in some
2293 * f_count is not spin-locked. Instead, atomic ops are used for
2294 * incrementing, decrementing, and handling the 1->0 transition.
2297 fhold(struct file *fp)
2299 atomic_add_int(&fp->f_count, 1);
2303 * fdrop() - drop a reference to a descriptor
2305 * MPALMOSTSAFE - acquires mplock for final close sequence
2308 fdrop(struct file *fp)
2315 * A combined fetch and subtract is needed to properly detect
2316 * 1->0 transitions, otherwise two cpus dropping from a ref
2317 * count of 2 might both try to run the 1->0 code.
2319 if (atomic_fetchadd_int(&fp->f_count, -1) > 1)
2325 * The last reference has gone away, we own the fp structure free
2328 if (fp->f_count < 0)
2329 panic("fdrop: count < 0");
2330 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE &&
2331 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2333 lf.l_whence = SEEK_SET;
2336 lf.l_type = F_UNLCK;
2337 vp = (struct vnode *)fp->f_data;
2338 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2340 if (fp->f_ops != &badfileops)
2341 error = fo_close(fp);
2350 * Apply an advisory lock on a file descriptor.
2352 * Just attempt to get a record lock of the requested type on
2353 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2358 sys_flock(struct flock_args *uap)
2360 struct proc *p = curproc;
2366 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
2369 if (fp->f_type != DTYPE_VNODE) {
2373 vp = (struct vnode *)fp->f_data;
2374 lf.l_whence = SEEK_SET;
2377 if (uap->how & LOCK_UN) {
2378 lf.l_type = F_UNLCK;
2379 fp->f_flag &= ~FHASLOCK;
2380 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2383 if (uap->how & LOCK_EX)
2384 lf.l_type = F_WRLCK;
2385 else if (uap->how & LOCK_SH)
2386 lf.l_type = F_RDLCK;
2391 fp->f_flag |= FHASLOCK;
2392 if (uap->how & LOCK_NB)
2393 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0);
2395 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT);
2403 * File Descriptor pseudo-device driver (/dev/fd/).
2405 * Opening minor device N dup()s the file (if any) connected to file
2406 * descriptor N belonging to the calling process. Note that this driver
2407 * consists of only the ``open()'' routine, because all subsequent
2408 * references to this file will be direct to the other driver.
2411 fdopen(struct dev_open_args *ap)
2413 thread_t td = curthread;
2415 KKASSERT(td->td_lwp != NULL);
2418 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the
2419 * the file descriptor being sought for duplication. The error
2420 * return ensures that the vnode for this device will be released
2421 * by vn_open. Open will detect this special error and take the
2422 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2423 * will simply report the error.
2425 td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev);
2430 * The caller has reserved the file descriptor dfd for us. On success we
2431 * must fsetfd() it. On failure the caller will clean it up.
2433 * NOT MPSAFE - isn't getting spinlocks, possibly other things
2436 dupfdopen(struct filedesc *fdp, int dfd, int sfd, int mode, int error)
2442 if ((wfp = holdfp(fdp, sfd, -1)) == NULL)
2446 * Close a revoke/dup race. Duping a descriptor marked as revoked
2447 * will dup a dummy descriptor instead of the real one.
2449 if (wfp->f_flag & FREVOKED) {
2450 kprintf("Warning: attempt to dup() a revoked descriptor\n");
2453 werror = falloc(NULL, &wfp, NULL);
2459 * There are two cases of interest here.
2461 * For ENODEV simply dup sfd to file descriptor dfd and return.
2463 * For ENXIO steal away the file structure from sfd and store it
2464 * dfd. sfd is effectively closed by this operation.
2466 * Any other error code is just returned.
2471 * Check that the mode the file is being opened for is a
2472 * subset of the mode of the existing descriptor.
2474 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2478 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2479 fsetfd(fdp, wfp, dfd);
2484 * Steal away the file pointer from dfd, and stuff it into indx.
2486 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2487 fsetfd(fdp, wfp, dfd);
2488 if ((xfp = funsetfd_locked(fdp, sfd)) != NULL)
2500 * NOT MPSAFE - I think these refer to a common file descriptor table
2501 * and we need to spinlock that to link fdtol in.
2503 struct filedesc_to_leader *
2504 filedesc_to_leader_alloc(struct filedesc_to_leader *old,
2505 struct proc *leader)
2507 struct filedesc_to_leader *fdtol;
2509 fdtol = kmalloc(sizeof(struct filedesc_to_leader),
2510 M_FILEDESC_TO_LEADER, M_WAITOK);
2511 fdtol->fdl_refcount = 1;
2512 fdtol->fdl_holdcount = 0;
2513 fdtol->fdl_wakeup = 0;
2514 fdtol->fdl_leader = leader;
2516 fdtol->fdl_next = old->fdl_next;
2517 fdtol->fdl_prev = old;
2518 old->fdl_next = fdtol;
2519 fdtol->fdl_next->fdl_prev = fdtol;
2521 fdtol->fdl_next = fdtol;
2522 fdtol->fdl_prev = fdtol;
2528 * Scan all file pointers in the system. The callback is made with
2529 * the master list spinlock held exclusively.
2534 allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data)
2539 spin_lock_wr(&filehead_spin);
2540 LIST_FOREACH(fp, &filehead, f_list) {
2541 res = callback(fp, data);
2545 spin_unlock_wr(&filehead_spin);
2549 * Get file structures.
2551 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe)
2554 struct sysctl_kern_file_info {
2557 struct sysctl_req *req;
2560 static int sysctl_kern_file_callback(struct proc *p, void *data);
2563 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2565 struct sysctl_kern_file_info info;
2568 * Note: because the number of file descriptors is calculated
2569 * in different ways for sizing vs returning the data,
2570 * there is information leakage from the first loop. However,
2571 * it is of a similar order of magnitude to the leakage from
2572 * global system statistics such as kern.openfiles.
2574 * When just doing a count, note that we cannot just count
2575 * the elements and add f_count via the filehead list because
2576 * threaded processes share their descriptor table and f_count might
2577 * still be '1' in that case.
2579 * Since the SYSCTL op can block, we must hold the process to
2580 * prevent it being ripped out from under us either in the
2581 * file descriptor loop or in the greater LIST_FOREACH. The
2582 * process may be in varying states of disrepair. If the process
2583 * is in SZOMB we may have caught it just as it is being removed
2584 * from the allproc list, we must skip it in that case to maintain
2585 * an unbroken chain through the allproc list.
2590 allproc_scan(sysctl_kern_file_callback, &info);
2593 * When just calculating the size, overestimate a bit to try to
2594 * prevent system activity from causing the buffer-fill call
2597 if (req->oldptr == NULL) {
2598 info.count = (info.count + 16) + (info.count / 10);
2599 info.error = SYSCTL_OUT(req, NULL,
2600 info.count * sizeof(struct kinfo_file));
2602 return (info.error);
2606 sysctl_kern_file_callback(struct proc *p, void *data)
2608 struct sysctl_kern_file_info *info = data;
2609 struct kinfo_file kf;
2610 struct filedesc *fdp;
2615 if (p->p_stat == SIDL || p->p_stat == SZOMB)
2617 if (!PRISON_CHECK(info->req->td->td_ucred, p->p_ucred) != 0)
2621 * Softref the fdp to prevent it from being destroyed
2623 spin_lock_wr(&p->p_spin);
2624 if ((fdp = p->p_fd) == NULL) {
2625 spin_unlock_wr(&p->p_spin);
2628 atomic_add_int(&fdp->fd_softrefs, 1);
2629 spin_unlock_wr(&p->p_spin);
2632 * The fdp's own spinlock prevents the contents from being
2635 spin_lock_rd(&fdp->fd_spin);
2636 for (n = 0; n < fdp->fd_nfiles; ++n) {
2637 if ((fp = fdp->fd_files[n].fp) == NULL)
2639 if (info->req->oldptr == NULL) {
2642 uid = p->p_ucred ? p->p_ucred->cr_uid : -1;
2643 kcore_make_file(&kf, fp, p->p_pid, uid, n);
2644 spin_unlock_rd(&fdp->fd_spin);
2645 info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf));
2646 spin_lock_rd(&fdp->fd_spin);
2651 spin_unlock_rd(&fdp->fd_spin);
2652 atomic_subtract_int(&fdp->fd_softrefs, 1);
2658 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2659 0, 0, sysctl_kern_file, "S,file", "Entire file table");
2661 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2662 &maxfilesperproc, 0, "Maximum files allowed open per process");
2664 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2665 &maxfiles, 0, "Maximum number of files");
2667 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW,
2668 &maxfilesrootres, 0, "Descriptors reserved for root use");
2670 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2671 &nfiles, 0, "System-wide number of open files");
2674 fildesc_drvinit(void *unused)
2678 for (fd = 0; fd < NUMFDESC; fd++) {
2679 make_dev(&fildesc_ops, fd,
2680 UID_BIN, GID_BIN, 0666, "fd/%d", fd);
2683 make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin");
2684 make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout");
2685 make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr");
2691 struct fileops badfileops = {
2692 .fo_read = badfo_readwrite,
2693 .fo_write = badfo_readwrite,
2694 .fo_ioctl = badfo_ioctl,
2695 .fo_poll = badfo_poll,
2696 .fo_kqfilter = badfo_kqfilter,
2697 .fo_stat = badfo_stat,
2698 .fo_close = badfo_close,
2699 .fo_shutdown = badfo_shutdown
2719 badfo_ioctl(struct file *fp, u_long com, caddr_t data,
2720 struct ucred *cred, struct sysmsg *msgv)
2729 badfo_poll(struct file *fp, int events, struct ucred *cred)
2738 badfo_kqfilter(struct file *fp, struct knote *kn)
2744 badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
2753 badfo_close(struct file *fp)
2762 badfo_shutdown(struct file *fp, int how)
2771 nofo_shutdown(struct file *fp, int how)
2773 return (EOPNOTSUPP);
2776 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,
2777 fildesc_drvinit,NULL)