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 $
75 #include "opt_compat.h"
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/malloc.h>
79 #include <sys/sysproto.h>
81 #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 void ffree(struct file *fp);
113 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
114 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
115 "file desc to leader structures");
116 MALLOC_DEFINE(M_FILE, "file", "Open file structure");
117 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
119 static struct krate krate_uidinfo = { .freq = 1 };
121 static d_open_t fdopen;
124 #define CDEV_MAJOR 22
125 static struct dev_ops fildesc_ops = {
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;
170 spin_lock(&limit->p_spin);
171 if (limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
174 dtsize = (int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur;
175 spin_unlock(&limit->p_spin);
177 if (dtsize > maxfilesperproc)
178 dtsize = maxfilesperproc;
179 if (dtsize < minfilesperproc)
180 dtsize = minfilesperproc;
181 if (p->p_ucred->cr_uid && dtsize > maxfilesperuser)
182 dtsize = maxfilesperuser;
183 uap->sysmsg_result = dtsize;
188 * Duplicate a file descriptor to a particular value.
190 * note: keep in mind that a potential race condition exists when closing
191 * descriptors from a shared descriptor table (via rfork).
196 sys_dup2(struct dup2_args *uap)
201 error = kern_dup(DUP_FIXED, uap->from, uap->to, &fd);
202 uap->sysmsg_fds[0] = fd;
208 * Duplicate a file descriptor.
213 sys_dup(struct dup_args *uap)
218 error = kern_dup(DUP_VARIABLE, uap->fd, 0, &fd);
219 uap->sysmsg_fds[0] = fd;
225 * MPALMOSTSAFE - acquires mplock for fp operations
228 kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred)
230 struct thread *td = curthread;
231 struct proc *p = td->td_proc;
237 int tmp, error, flg = F_POSIX;
242 * Operations on file descriptors that do not require a file pointer.
246 error = fgetfdflags(p->p_fd, fd, &tmp);
248 dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0;
252 if (dat->fc_cloexec & FD_CLOEXEC)
253 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
255 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
259 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd);
266 * Operations on file pointers
268 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
273 dat->fc_flags = OFLAGS(fp->f_flag);
279 nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS;
280 nflags |= oflags & ~FCNTLFLAGS;
283 if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY))
285 if (error == 0 && ((nflags ^ oflags) & FASYNC)) {
286 tmp = nflags & FASYNC;
287 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp,
295 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner,
300 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner,
306 /* Fall into F_SETLK */
309 if (fp->f_type != DTYPE_VNODE) {
313 vp = (struct vnode *)fp->f_data;
316 * copyin/lockop may block
318 if (dat->fc_flock.l_whence == SEEK_CUR)
319 dat->fc_flock.l_start += fp->f_offset;
321 switch (dat->fc_flock.l_type) {
323 if ((fp->f_flag & FREAD) == 0) {
327 p->p_leader->p_flag |= P_ADVLOCK;
328 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
329 &dat->fc_flock, flg);
332 if ((fp->f_flag & FWRITE) == 0) {
336 p->p_leader->p_flag |= P_ADVLOCK;
337 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
338 &dat->fc_flock, flg);
341 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
342 &dat->fc_flock, F_POSIX);
350 * It is possible to race a close() on the descriptor while
351 * we were blocked getting the lock. If this occurs the
352 * close might not have caught the lock.
354 if (checkfdclosed(p->p_fd, fd, fp)) {
355 dat->fc_flock.l_whence = SEEK_SET;
356 dat->fc_flock.l_start = 0;
357 dat->fc_flock.l_len = 0;
358 dat->fc_flock.l_type = F_UNLCK;
359 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
360 F_UNLCK, &dat->fc_flock, F_POSIX);
365 if (fp->f_type != DTYPE_VNODE) {
369 vp = (struct vnode *)fp->f_data;
371 * copyin/lockop may block
373 if (dat->fc_flock.l_type != F_RDLCK &&
374 dat->fc_flock.l_type != F_WRLCK &&
375 dat->fc_flock.l_type != F_UNLCK) {
379 if (dat->fc_flock.l_whence == SEEK_CUR)
380 dat->fc_flock.l_start += fp->f_offset;
381 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK,
382 &dat->fc_flock, F_POSIX);
394 * The file control system call.
399 sys_fcntl(struct fcntl_args *uap)
406 dat.fc_fd = uap->arg;
409 dat.fc_cloexec = uap->arg;
412 dat.fc_flags = uap->arg;
415 dat.fc_owner = uap->arg;
420 error = copyin((caddr_t)uap->arg, &dat.fc_flock,
421 sizeof(struct flock));
427 error = kern_fcntl(uap->fd, uap->cmd, &dat, curthread->td_ucred);
432 uap->sysmsg_result = dat.fc_fd;
435 uap->sysmsg_result = dat.fc_cloexec;
438 uap->sysmsg_result = dat.fc_flags;
441 uap->sysmsg_result = dat.fc_owner;
443 error = copyout(&dat.fc_flock, (caddr_t)uap->arg,
444 sizeof(struct flock));
453 * Common code for dup, dup2, and fcntl(F_DUPFD).
455 * The type flag can be either DUP_FIXED or DUP_VARIABLE. DUP_FIXED tells
456 * kern_dup() to destructively dup over an existing file descriptor if new
457 * is already open. DUP_VARIABLE tells kern_dup() to find the lowest
458 * unused file descriptor that is greater than or equal to new.
463 kern_dup(enum dup_type type, int old, int new, int *res)
465 struct thread *td = curthread;
466 struct proc *p = td->td_proc;
467 struct filedesc *fdp = p->p_fd;
476 * Verify that we have a valid descriptor to dup from and
477 * possibly to dup to.
479 * NOTE: maxfilesperuser is not applicable to dup()
482 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
485 dtsize = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
486 if (dtsize > maxfilesperproc)
487 dtsize = maxfilesperproc;
488 if (dtsize < minfilesperproc)
489 dtsize = minfilesperproc;
491 if (new < 0 || new > dtsize)
494 spin_lock(&fdp->fd_spin);
495 if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) {
496 spin_unlock(&fdp->fd_spin);
499 if (type == DUP_FIXED && old == new) {
501 spin_unlock(&fdp->fd_spin);
504 fp = fdp->fd_files[old].fp;
505 oldflags = fdp->fd_files[old].fileflags;
506 fhold(fp); /* MPSAFE - can be called with a spinlock held */
509 * Allocate a new descriptor if DUP_VARIABLE, or expand the table
510 * if the requested descriptor is beyond the current table size.
512 * This can block. Retry if the source descriptor no longer matches
513 * or if our expectation in the expansion case races.
515 * If we are not expanding or allocating a new decriptor, then reset
516 * the target descriptor to a reserved state so we have a uniform
517 * setup for the next code block.
519 if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) {
520 spin_unlock(&fdp->fd_spin);
521 error = fdalloc(p, new, &newfd);
522 spin_lock(&fdp->fd_spin);
524 spin_unlock(&fdp->fd_spin);
531 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) {
532 fsetfd_locked(fdp, NULL, newfd);
533 spin_unlock(&fdp->fd_spin);
538 * Check for expansion race
540 if (type != DUP_VARIABLE && new != newfd) {
541 fsetfd_locked(fdp, NULL, newfd);
542 spin_unlock(&fdp->fd_spin);
547 * Check for ripout, newfd reused old (this case probably
551 fsetfd_locked(fdp, NULL, newfd);
552 spin_unlock(&fdp->fd_spin);
559 if (fdp->fd_files[new].reserved) {
560 spin_unlock(&fdp->fd_spin);
562 kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new);
563 tsleep(fdp, 0, "fdres", hz);
568 * If the target descriptor was never allocated we have
569 * to allocate it. If it was we have to clean out the
570 * old descriptor. delfp inherits the ref from the
573 delfp = fdp->fd_files[new].fp;
574 fdp->fd_files[new].fp = NULL;
575 fdp->fd_files[new].reserved = 1;
577 fdreserve_locked(fdp, new, 1);
578 if (new > fdp->fd_lastfile)
579 fdp->fd_lastfile = new;
585 * NOTE: still holding an exclusive spinlock
589 * If a descriptor is being overwritten we may hve to tell
590 * fdfree() to sleep to ensure that all relevant process
591 * leaders can be traversed in closef().
593 if (delfp != NULL && p->p_fdtol != NULL) {
594 fdp->fd_holdleaderscount++;
599 KASSERT(delfp == NULL || type == DUP_FIXED,
600 ("dup() picked an open file"));
603 * Duplicate the source descriptor, update lastfile. If the new
604 * descriptor was not allocated and we aren't replacing an existing
605 * descriptor we have to mark the descriptor as being in use.
607 * The fd_files[] array inherits fp's hold reference.
609 fsetfd_locked(fdp, fp, new);
610 fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE;
611 spin_unlock(&fdp->fd_spin);
616 * If we dup'd over a valid file, we now own the reference to it
617 * and must dispose of it using closef() semantics (as if a
618 * close() were performed on it).
621 if (SLIST_FIRST(&delfp->f_klist))
622 knote_fdclose(delfp, fdp, new);
625 spin_lock(&fdp->fd_spin);
626 fdp->fd_holdleaderscount--;
627 if (fdp->fd_holdleaderscount == 0 &&
628 fdp->fd_holdleaderswakeup != 0) {
629 fdp->fd_holdleaderswakeup = 0;
630 spin_unlock(&fdp->fd_spin);
631 wakeup(&fdp->fd_holdleaderscount);
633 spin_unlock(&fdp->fd_spin);
641 * If sigio is on the list associated with a process or process group,
642 * disable signalling from the device, remove sigio from the list and
648 funsetown(struct sigio *sigio)
652 lwkt_gettoken(&proc_token);
653 *(sigio->sio_myref) = NULL;
654 if (sigio->sio_pgid < 0) {
655 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
657 } else /* if ((*sigiop)->sio_pgid > 0) */ {
658 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
661 lwkt_reltoken(&proc_token);
662 crfree(sigio->sio_ucred);
663 kfree(sigio, M_SIGIO);
667 * Free a list of sigio structures.
672 funsetownlst(struct sigiolst *sigiolst)
676 lwkt_gettoken(&proc_token);
677 while ((sigio = SLIST_FIRST(sigiolst)) != NULL)
679 lwkt_reltoken(&proc_token);
683 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
685 * After permission checking, add a sigio structure to the sigio list for
686 * the process or process group.
691 fsetown(pid_t pgid, struct sigio **sigiop)
703 lwkt_gettoken(&proc_token);
712 * Policy - Don't allow a process to FSETOWN a process
713 * in another session.
715 * Remove this test to allow maximum flexibility or
716 * restrict FSETOWN to the current process or process
717 * group for maximum safety.
719 if (proc->p_session != curproc->p_session) {
725 } else /* if (pgid < 0) */ {
726 pgrp = pgfind(-pgid);
733 * Policy - Don't allow a process to FSETOWN a process
734 * in another session.
736 * Remove this test to allow maximum flexibility or
737 * restrict FSETOWN to the current process or process
738 * group for maximum safety.
740 if (pgrp->pg_session != curproc->p_session) {
748 sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
750 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
751 sigio->sio_proc = proc;
753 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
754 sigio->sio_pgrp = pgrp;
756 sigio->sio_pgid = pgid;
757 sigio->sio_ucred = crhold(curthread->td_ucred);
758 /* It would be convenient if p_ruid was in ucred. */
759 sigio->sio_ruid = sigio->sio_ucred->cr_ruid;
760 sigio->sio_myref = sigiop;
764 lwkt_reltoken(&proc_token);
769 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
774 fgetown(struct sigio **sigiop)
779 lwkt_gettoken(&proc_token);
781 own = (sigio != NULL ? sigio->sio_pgid : 0);
782 lwkt_reltoken(&proc_token);
788 * Close many file descriptors.
793 sys_closefrom(struct closefrom_args *uap)
795 return(kern_closefrom(uap->fd));
799 * Close all file descriptors greater then or equal to fd
804 kern_closefrom(int fd)
806 struct thread *td = curthread;
807 struct proc *p = td->td_proc;
808 struct filedesc *fdp;
817 * NOTE: This function will skip unassociated descriptors and
818 * reserved descriptors that have not yet been assigned.
819 * fd_lastfile can change as a side effect of kern_close().
821 spin_lock(&fdp->fd_spin);
822 while (fd <= fdp->fd_lastfile) {
823 if (fdp->fd_files[fd].fp != NULL) {
824 spin_unlock(&fdp->fd_spin);
825 /* ok if this races another close */
826 if (kern_close(fd) == EINTR)
828 spin_lock(&fdp->fd_spin);
832 spin_unlock(&fdp->fd_spin);
837 * Close a file descriptor.
842 sys_close(struct close_args *uap)
844 return(kern_close(uap->fd));
853 struct thread *td = curthread;
854 struct proc *p = td->td_proc;
855 struct filedesc *fdp;
863 spin_lock(&fdp->fd_spin);
864 if ((fp = funsetfd_locked(fdp, fd)) == NULL) {
865 spin_unlock(&fdp->fd_spin);
869 if (p->p_fdtol != NULL) {
871 * Ask fdfree() to sleep to ensure that all relevant
872 * process leaders can be traversed in closef().
874 fdp->fd_holdleaderscount++;
879 * we now hold the fp reference that used to be owned by the descriptor
882 spin_unlock(&fdp->fd_spin);
883 if (SLIST_FIRST(&fp->f_klist))
884 knote_fdclose(fp, fdp, fd);
885 error = closef(fp, p);
887 spin_lock(&fdp->fd_spin);
888 fdp->fd_holdleaderscount--;
889 if (fdp->fd_holdleaderscount == 0 &&
890 fdp->fd_holdleaderswakeup != 0) {
891 fdp->fd_holdleaderswakeup = 0;
892 spin_unlock(&fdp->fd_spin);
893 wakeup(&fdp->fd_holdleaderscount);
895 spin_unlock(&fdp->fd_spin);
902 * shutdown_args(int fd, int how)
905 kern_shutdown(int fd, int how)
907 struct thread *td = curthread;
908 struct proc *p = td->td_proc;
914 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
916 error = fo_shutdown(fp, how);
926 sys_shutdown(struct shutdown_args *uap)
930 error = kern_shutdown(uap->s, uap->how);
939 kern_fstat(int fd, struct stat *ub)
941 struct thread *td = curthread;
942 struct proc *p = td->td_proc;
948 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
950 error = fo_stat(fp, ub, td->td_ucred);
957 * Return status information about a file descriptor.
962 sys_fstat(struct fstat_args *uap)
967 error = kern_fstat(uap->fd, &st);
970 error = copyout(&st, uap->sb, sizeof(st));
975 * Return pathconf information about a file descriptor.
980 sys_fpathconf(struct fpathconf_args *uap)
982 struct thread *td = curthread;
983 struct proc *p = td->td_proc;
988 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
991 switch (fp->f_type) {
994 if (uap->name != _PC_PIPE_BUF) {
997 uap->sysmsg_result = PIPE_BUF;
1003 vp = (struct vnode *)fp->f_data;
1004 error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg);
1014 static int fdexpand;
1015 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0,
1016 "Number of times a file table has been expanded");
1019 * Grow the file table so it can hold through descriptor (want).
1021 * The fdp's spinlock must be held exclusively on entry and may be held
1022 * exclusively on return. The spinlock may be cycled by the routine.
1027 fdgrow_locked(struct filedesc *fdp, int want)
1029 struct fdnode *newfiles;
1030 struct fdnode *oldfiles;
1033 nf = fdp->fd_nfiles;
1035 /* nf has to be of the form 2^n - 1 */
1037 } while (nf <= want);
1039 spin_unlock(&fdp->fd_spin);
1040 newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK);
1041 spin_lock(&fdp->fd_spin);
1044 * We could have raced another extend while we were not holding
1047 if (fdp->fd_nfiles >= nf) {
1048 spin_unlock(&fdp->fd_spin);
1049 kfree(newfiles, M_FILEDESC);
1050 spin_lock(&fdp->fd_spin);
1054 * Copy the existing ofile and ofileflags arrays
1055 * and zero the new portion of each array.
1057 extra = nf - fdp->fd_nfiles;
1058 bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode));
1059 bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode));
1061 oldfiles = fdp->fd_files;
1062 fdp->fd_files = newfiles;
1063 fdp->fd_nfiles = nf;
1065 if (oldfiles != fdp->fd_builtin_files) {
1066 spin_unlock(&fdp->fd_spin);
1067 kfree(oldfiles, M_FILEDESC);
1068 spin_lock(&fdp->fd_spin);
1074 * Number of nodes in right subtree, including the root.
1077 right_subtree_size(int n)
1079 return (n ^ (n | (n + 1)));
1086 right_ancestor(int n)
1088 return (n | (n + 1));
1095 left_ancestor(int n)
1097 return ((n & (n + 1)) - 1);
1101 * Traverse the in-place binary tree buttom-up adjusting the allocation
1102 * count so scans can determine where free descriptors are located.
1104 * MPSAFE - caller must be holding an exclusive spinlock on fdp
1108 fdreserve_locked(struct filedesc *fdp, int fd, int incr)
1111 fdp->fd_files[fd].allocated += incr;
1112 KKASSERT(fdp->fd_files[fd].allocated >= 0);
1113 fd = left_ancestor(fd);
1118 * Reserve a file descriptor for the process. If no error occurs, the
1119 * caller MUST at some point call fsetfd() or assign a file pointer
1120 * or dispose of the reservation.
1125 fdalloc(struct proc *p, int want, int *result)
1127 struct filedesc *fdp = p->p_fd;
1128 struct uidinfo *uip;
1129 int fd, rsize, rsum, node, lim;
1132 * Check dtable size limit
1134 spin_lock(&p->p_limit->p_spin);
1135 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
1138 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
1139 spin_unlock(&p->p_limit->p_spin);
1141 if (lim > maxfilesperproc)
1142 lim = maxfilesperproc;
1143 if (lim < minfilesperproc)
1144 lim = minfilesperproc;
1149 * Check that the user has not run out of descriptors (non-root only).
1150 * As a safety measure the dtable is allowed to have at least
1151 * minfilesperproc open fds regardless of the maxfilesperuser limit.
1153 if (p->p_ucred->cr_uid && fdp->fd_nfiles >= minfilesperproc) {
1154 uip = p->p_ucred->cr_uidinfo;
1155 if (uip->ui_openfiles > maxfilesperuser) {
1156 krateprintf(&krate_uidinfo,
1157 "Warning: user %d pid %d (%s) ran out of "
1158 "file descriptors (%d/%d)\n",
1159 p->p_ucred->cr_uid, (int)p->p_pid,
1161 uip->ui_openfiles, maxfilesperuser);
1167 * Grow the dtable if necessary
1169 spin_lock(&fdp->fd_spin);
1170 if (want >= fdp->fd_nfiles)
1171 fdgrow_locked(fdp, want);
1174 * Search for a free descriptor starting at the higher
1175 * of want or fd_freefile. If that fails, consider
1176 * expanding the ofile array.
1178 * NOTE! the 'allocated' field is a cumulative recursive allocation
1179 * count. If we happen to see a value of 0 then we can shortcut
1180 * our search. Otherwise we run through through the tree going
1181 * down branches we know have free descriptor(s) until we hit a
1182 * leaf node. The leaf node will be free but will not necessarily
1183 * have an allocated field of 0.
1186 /* move up the tree looking for a subtree with a free node */
1187 for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim);
1188 fd = right_ancestor(fd)) {
1189 if (fdp->fd_files[fd].allocated == 0)
1192 rsize = right_subtree_size(fd);
1193 if (fdp->fd_files[fd].allocated == rsize)
1194 continue; /* right subtree full */
1197 * Free fd is in the right subtree of the tree rooted at fd.
1198 * Call that subtree R. Look for the smallest (leftmost)
1199 * subtree of R with an unallocated fd: continue moving
1200 * down the left branch until encountering a full left
1201 * subtree, then move to the right.
1203 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) {
1205 rsum += fdp->fd_files[node].allocated;
1206 if (fdp->fd_files[fd].allocated == rsum + rsize) {
1207 fd = node; /* move to the right */
1208 if (fdp->fd_files[node].allocated == 0)
1217 * No space in current array. Expand?
1219 if (fdp->fd_nfiles >= lim) {
1220 spin_unlock(&fdp->fd_spin);
1223 fdgrow_locked(fdp, want);
1227 KKASSERT(fd < fdp->fd_nfiles);
1228 if (fd > fdp->fd_lastfile)
1229 fdp->fd_lastfile = fd;
1230 if (want <= fdp->fd_freefile)
1231 fdp->fd_freefile = fd;
1233 KKASSERT(fdp->fd_files[fd].fp == NULL);
1234 KKASSERT(fdp->fd_files[fd].reserved == 0);
1235 fdp->fd_files[fd].fileflags = 0;
1236 fdp->fd_files[fd].reserved = 1;
1237 fdreserve_locked(fdp, fd, 1);
1238 spin_unlock(&fdp->fd_spin);
1243 * Check to see whether n user file descriptors
1244 * are available to the process p.
1249 fdavail(struct proc *p, int n)
1251 struct filedesc *fdp = p->p_fd;
1252 struct fdnode *fdnode;
1255 spin_lock(&p->p_limit->p_spin);
1256 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
1259 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
1260 spin_unlock(&p->p_limit->p_spin);
1262 if (lim > maxfilesperproc)
1263 lim = maxfilesperproc;
1264 if (lim < minfilesperproc)
1265 lim = minfilesperproc;
1267 spin_lock(&fdp->fd_spin);
1268 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) {
1269 spin_unlock(&fdp->fd_spin);
1272 last = min(fdp->fd_nfiles, lim);
1273 fdnode = &fdp->fd_files[fdp->fd_freefile];
1274 for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) {
1275 if (fdnode->fp == NULL && --n <= 0) {
1276 spin_unlock(&fdp->fd_spin);
1280 spin_unlock(&fdp->fd_spin);
1285 * Revoke open descriptors referencing (f_data, f_type)
1287 * Any revoke executed within a prison is only able to
1288 * revoke descriptors for processes within that prison.
1290 * Returns 0 on success or an error code.
1292 struct fdrevoke_info {
1302 static int fdrevoke_check_callback(struct file *fp, void *vinfo);
1303 static int fdrevoke_proc_callback(struct proc *p, void *vinfo);
1306 fdrevoke(void *f_data, short f_type, struct ucred *cred)
1308 struct fdrevoke_info info;
1311 bzero(&info, sizeof(info));
1315 error = falloc(NULL, &info.nfp, NULL);
1320 * Scan the file pointer table once. dups do not dup file pointers,
1321 * only descriptors, so there is no leak. Set FREVOKED on the fps
1324 allfiles_scan_exclusive(fdrevoke_check_callback, &info);
1327 * If any fps were marked track down the related descriptors
1328 * and close them. Any dup()s at this point will notice
1329 * the FREVOKED already set in the fp and do the right thing.
1331 * Any fps with non-zero msgcounts (aka sent over a unix-domain
1332 * socket) bumped the intransit counter and will require a
1333 * scan. Races against fps leaving the socket are closed by
1334 * the socket code checking for FREVOKED.
1337 allproc_scan(fdrevoke_proc_callback, &info);
1339 unp_revoke_gc(info.nfp);
1345 * Locate matching file pointers directly.
1347 * WARNING: allfiles_scan_exclusive() holds a spinlock through these calls!
1350 fdrevoke_check_callback(struct file *fp, void *vinfo)
1352 struct fdrevoke_info *info = vinfo;
1355 * File pointers already flagged for revokation are skipped.
1357 if (fp->f_flag & FREVOKED)
1361 * If revoking from a prison file pointers created outside of
1362 * that prison, or file pointers without creds, cannot be revoked.
1364 if (info->cred->cr_prison &&
1365 (fp->f_cred == NULL ||
1366 info->cred->cr_prison != fp->f_cred->cr_prison)) {
1371 * If the file pointer matches then mark it for revocation. The
1372 * flag is currently only used by unp_revoke_gc().
1374 * info->count is a heuristic and can race in a SMP environment.
1376 if (info->data == fp->f_data && info->type == fp->f_type) {
1377 atomic_set_int(&fp->f_flag, FREVOKED);
1378 info->count += fp->f_count;
1386 * Locate matching file pointers via process descriptor tables.
1389 fdrevoke_proc_callback(struct proc *p, void *vinfo)
1391 struct fdrevoke_info *info = vinfo;
1392 struct filedesc *fdp;
1396 if (p->p_stat == SIDL || p->p_stat == SZOMB)
1398 if (info->cred->cr_prison &&
1399 info->cred->cr_prison != p->p_ucred->cr_prison) {
1404 * If the controlling terminal of the process matches the
1405 * vnode being revoked we clear the controlling terminal.
1407 * The normal spec_close() may not catch this because it
1408 * uses curproc instead of p.
1410 if (p->p_session && info->type == DTYPE_VNODE &&
1411 info->data == p->p_session->s_ttyvp) {
1412 p->p_session->s_ttyvp = NULL;
1417 * Softref the fdp to prevent it from being destroyed
1419 spin_lock(&p->p_spin);
1420 if ((fdp = p->p_fd) == NULL) {
1421 spin_unlock(&p->p_spin);
1424 atomic_add_int(&fdp->fd_softrefs, 1);
1425 spin_unlock(&p->p_spin);
1428 * Locate and close any matching file descriptors.
1430 spin_lock(&fdp->fd_spin);
1431 for (n = 0; n < fdp->fd_nfiles; ++n) {
1432 if ((fp = fdp->fd_files[n].fp) == NULL)
1434 if (fp->f_flag & FREVOKED) {
1436 fdp->fd_files[n].fp = info->nfp;
1437 spin_unlock(&fdp->fd_spin);
1438 knote_fdclose(fp, fdp, n); /* XXX */
1440 spin_lock(&fdp->fd_spin);
1444 spin_unlock(&fdp->fd_spin);
1445 atomic_subtract_int(&fdp->fd_softrefs, 1);
1451 * Create a new open file structure and reserve a file decriptor
1452 * for the process that refers to it.
1454 * Root creds are checked using lp, or assumed if lp is NULL. If
1455 * resultfd is non-NULL then lp must also be non-NULL. No file
1456 * descriptor is reserved (and no process context is needed) if
1459 * A file pointer with a refcount of 1 is returned. Note that the
1460 * file pointer is NOT associated with the descriptor. If falloc
1461 * returns success, fsetfd() MUST be called to either associate the
1462 * file pointer or clear the reservation.
1467 falloc(struct lwp *lp, struct file **resultfp, int *resultfd)
1469 static struct timeval lastfail;
1472 struct ucred *cred = lp ? lp->lwp_thread->td_ucred : proc0.p_ucred;
1478 * Handle filetable full issues and root overfill.
1480 if (nfiles >= maxfiles - maxfilesrootres &&
1481 (cred->cr_ruid != 0 || nfiles >= maxfiles)) {
1482 if (ppsratecheck(&lastfail, &curfail, 1)) {
1483 kprintf("kern.maxfiles limit exceeded by uid %d, "
1484 "please see tuning(7).\n",
1492 * Allocate a new file descriptor.
1494 fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO);
1495 spin_init(&fp->f_spin);
1496 SLIST_INIT(&fp->f_klist);
1498 fp->f_ops = &badfileops;
1501 spin_lock(&filehead_spin);
1503 LIST_INSERT_HEAD(&filehead, fp, f_list);
1504 spin_unlock(&filehead_spin);
1506 if ((error = fdalloc(lp->lwp_proc, 0, resultfd)) != 0) {
1519 * Check for races against a file descriptor by determining that the
1520 * file pointer is still associated with the specified file descriptor,
1521 * and a close is not currently in progress.
1526 checkfdclosed(struct filedesc *fdp, int fd, struct file *fp)
1530 spin_lock(&fdp->fd_spin);
1531 if ((unsigned)fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp)
1535 spin_unlock(&fdp->fd_spin);
1540 * Associate a file pointer with a previously reserved file descriptor.
1541 * This function always succeeds.
1543 * If fp is NULL, the file descriptor is returned to the pool.
1547 * MPSAFE (exclusive spinlock must be held on call)
1550 fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd)
1552 KKASSERT((unsigned)fd < fdp->fd_nfiles);
1553 KKASSERT(fdp->fd_files[fd].reserved != 0);
1556 fdp->fd_files[fd].fp = fp;
1557 fdp->fd_files[fd].reserved = 0;
1559 fdp->fd_files[fd].reserved = 0;
1560 fdreserve_locked(fdp, fd, -1);
1561 fdfixup_locked(fdp, fd);
1569 fsetfd(struct filedesc *fdp, struct file *fp, int fd)
1571 spin_lock(&fdp->fd_spin);
1572 fsetfd_locked(fdp, fp, fd);
1573 spin_unlock(&fdp->fd_spin);
1577 * MPSAFE (exclusive spinlock must be held on call)
1581 funsetfd_locked(struct filedesc *fdp, int fd)
1585 if ((unsigned)fd >= fdp->fd_nfiles)
1587 if ((fp = fdp->fd_files[fd].fp) == NULL)
1589 fdp->fd_files[fd].fp = NULL;
1590 fdp->fd_files[fd].fileflags = 0;
1592 fdreserve_locked(fdp, fd, -1);
1593 fdfixup_locked(fdp, fd);
1601 fgetfdflags(struct filedesc *fdp, int fd, int *flagsp)
1605 spin_lock(&fdp->fd_spin);
1606 if (((u_int)fd) >= fdp->fd_nfiles) {
1608 } else if (fdp->fd_files[fd].fp == NULL) {
1611 *flagsp = fdp->fd_files[fd].fileflags;
1614 spin_unlock(&fdp->fd_spin);
1622 fsetfdflags(struct filedesc *fdp, int fd, int add_flags)
1626 spin_lock(&fdp->fd_spin);
1627 if (((u_int)fd) >= fdp->fd_nfiles) {
1629 } else if (fdp->fd_files[fd].fp == NULL) {
1632 fdp->fd_files[fd].fileflags |= add_flags;
1635 spin_unlock(&fdp->fd_spin);
1643 fclrfdflags(struct filedesc *fdp, int fd, int rem_flags)
1647 spin_lock(&fdp->fd_spin);
1648 if (((u_int)fd) >= fdp->fd_nfiles) {
1650 } else if (fdp->fd_files[fd].fp == NULL) {
1653 fdp->fd_files[fd].fileflags &= ~rem_flags;
1656 spin_unlock(&fdp->fd_spin);
1661 * Set/Change/Clear the creds for a fp and synchronize the uidinfo.
1664 fsetcred(struct file *fp, struct ucred *ncr)
1667 struct uidinfo *uip;
1670 if (ocr == NULL || ncr == NULL || ocr->cr_uidinfo != ncr->cr_uidinfo) {
1672 uip = ocr->cr_uidinfo;
1673 atomic_add_int(&uip->ui_openfiles, -1);
1676 uip = ncr->cr_uidinfo;
1677 atomic_add_int(&uip->ui_openfiles, 1);
1688 * Free a file descriptor.
1692 ffree(struct file *fp)
1694 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!"));
1695 spin_lock(&filehead_spin);
1696 LIST_REMOVE(fp, f_list);
1698 spin_unlock(&filehead_spin);
1700 if (fp->f_nchandle.ncp)
1701 cache_drop(&fp->f_nchandle);
1706 * called from init_main, initialize filedesc0 for proc0.
1709 fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask)
1713 fdp0->fd_refcnt = 1;
1714 fdp0->fd_cmask = cmask;
1715 fdp0->fd_files = fdp0->fd_builtin_files;
1716 fdp0->fd_nfiles = NDFILE;
1717 fdp0->fd_lastfile = -1;
1718 spin_init(&fdp0->fd_spin);
1722 * Build a new filedesc structure.
1727 fdinit(struct proc *p)
1729 struct filedesc *newfdp;
1730 struct filedesc *fdp = p->p_fd;
1732 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO);
1733 spin_lock(&fdp->fd_spin);
1735 newfdp->fd_cdir = fdp->fd_cdir;
1736 vref(newfdp->fd_cdir);
1737 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1741 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of
1742 * proc0, but should unconditionally exist in other processes.
1745 newfdp->fd_rdir = fdp->fd_rdir;
1746 vref(newfdp->fd_rdir);
1747 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1750 newfdp->fd_jdir = fdp->fd_jdir;
1751 vref(newfdp->fd_jdir);
1752 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1754 spin_unlock(&fdp->fd_spin);
1756 /* Create the file descriptor table. */
1757 newfdp->fd_refcnt = 1;
1758 newfdp->fd_cmask = cmask;
1759 newfdp->fd_files = newfdp->fd_builtin_files;
1760 newfdp->fd_nfiles = NDFILE;
1761 newfdp->fd_lastfile = -1;
1762 spin_init(&newfdp->fd_spin);
1768 * Share a filedesc structure.
1773 fdshare(struct proc *p)
1775 struct filedesc *fdp;
1778 spin_lock(&fdp->fd_spin);
1780 spin_unlock(&fdp->fd_spin);
1785 * Copy a filedesc structure.
1790 fdcopy(struct proc *p)
1792 struct filedesc *fdp = p->p_fd;
1793 struct filedesc *newfdp;
1794 struct fdnode *fdnode;
1799 * Certain daemons might not have file descriptors.
1805 * Allocate the new filedesc and fd_files[] array. This can race
1806 * with operations by other threads on the fdp so we have to be
1809 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK | M_ZERO);
1811 spin_lock(&fdp->fd_spin);
1812 if (fdp->fd_lastfile < NDFILE) {
1813 newfdp->fd_files = newfdp->fd_builtin_files;
1817 * We have to allocate (N^2-1) entries for our in-place
1818 * binary tree. Allow the table to shrink.
1822 while (ni > fdp->fd_lastfile && ni > NDFILE) {
1826 spin_unlock(&fdp->fd_spin);
1827 newfdp->fd_files = kmalloc(i * sizeof(struct fdnode),
1828 M_FILEDESC, M_WAITOK | M_ZERO);
1831 * Check for race, retry
1833 spin_lock(&fdp->fd_spin);
1834 if (i <= fdp->fd_lastfile) {
1835 spin_unlock(&fdp->fd_spin);
1836 kfree(newfdp->fd_files, M_FILEDESC);
1842 * Dup the remaining fields. vref() and cache_hold() can be
1843 * safely called while holding the read spinlock on fdp.
1845 * The read spinlock on fdp is still being held.
1847 * NOTE: vref and cache_hold calls for the case where the vnode
1848 * or cache entry already has at least one ref may be called
1849 * while holding spin locks.
1851 if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) {
1852 vref(newfdp->fd_cdir);
1853 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1856 * We must check for fd_rdir here, at least for now because
1857 * the init process is created before we have access to the
1858 * rootvode to take a reference to it.
1860 if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) {
1861 vref(newfdp->fd_rdir);
1862 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1864 if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) {
1865 vref(newfdp->fd_jdir);
1866 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1868 newfdp->fd_refcnt = 1;
1869 newfdp->fd_nfiles = i;
1870 newfdp->fd_lastfile = fdp->fd_lastfile;
1871 newfdp->fd_freefile = fdp->fd_freefile;
1872 newfdp->fd_cmask = fdp->fd_cmask;
1873 spin_init(&newfdp->fd_spin);
1876 * Copy the descriptor table through (i). This also copies the
1877 * allocation state. Then go through and ref the file pointers
1878 * and clean up any KQ descriptors.
1880 * kq descriptors cannot be copied. Since we haven't ref'd the
1881 * copied files yet we can ignore the return value from funsetfd().
1883 * The read spinlock on fdp is still being held.
1885 bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode));
1886 for (i = 0 ; i < newfdp->fd_nfiles; ++i) {
1887 fdnode = &newfdp->fd_files[i];
1888 if (fdnode->reserved) {
1889 fdreserve_locked(newfdp, i, -1);
1890 fdnode->reserved = 0;
1891 fdfixup_locked(newfdp, i);
1892 } else if (fdnode->fp) {
1893 if (fdnode->fp->f_type == DTYPE_KQUEUE) {
1894 (void)funsetfd_locked(newfdp, i);
1900 spin_unlock(&fdp->fd_spin);
1905 * Release a filedesc structure.
1907 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE)
1910 fdfree(struct proc *p, struct filedesc *repl)
1912 struct filedesc *fdp;
1913 struct fdnode *fdnode;
1915 struct filedesc_to_leader *fdtol;
1921 * Certain daemons might not have file descriptors.
1930 * Severe messing around to follow.
1932 spin_lock(&fdp->fd_spin);
1934 /* Check for special need to clear POSIX style locks */
1936 if (fdtol != NULL) {
1937 KASSERT(fdtol->fdl_refcount > 0,
1938 ("filedesc_to_refcount botch: fdl_refcount=%d",
1939 fdtol->fdl_refcount));
1940 if (fdtol->fdl_refcount == 1 &&
1941 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1942 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1943 fdnode = &fdp->fd_files[i];
1944 if (fdnode->fp == NULL ||
1945 fdnode->fp->f_type != DTYPE_VNODE) {
1950 spin_unlock(&fdp->fd_spin);
1952 lf.l_whence = SEEK_SET;
1955 lf.l_type = F_UNLCK;
1956 vp = (struct vnode *)fp->f_data;
1957 (void) VOP_ADVLOCK(vp,
1958 (caddr_t)p->p_leader,
1963 spin_lock(&fdp->fd_spin);
1967 if (fdtol->fdl_refcount == 1) {
1968 if (fdp->fd_holdleaderscount > 0 &&
1969 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1971 * close() or do_dup() has cleared a reference
1972 * in a shared file descriptor table.
1974 fdp->fd_holdleaderswakeup = 1;
1975 ssleep(&fdp->fd_holdleaderscount,
1976 &fdp->fd_spin, 0, "fdlhold", 0);
1979 if (fdtol->fdl_holdcount > 0) {
1981 * Ensure that fdtol->fdl_leader
1982 * remains valid in closef().
1984 fdtol->fdl_wakeup = 1;
1985 ssleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0);
1989 fdtol->fdl_refcount--;
1990 if (fdtol->fdl_refcount == 0 &&
1991 fdtol->fdl_holdcount == 0) {
1992 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1993 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1998 if (fdtol != NULL) {
1999 spin_unlock(&fdp->fd_spin);
2000 kfree(fdtol, M_FILEDESC_TO_LEADER);
2001 spin_lock(&fdp->fd_spin);
2004 if (--fdp->fd_refcnt > 0) {
2005 spin_unlock(&fdp->fd_spin);
2006 spin_lock(&p->p_spin);
2008 spin_unlock(&p->p_spin);
2013 * Even though we are the last reference to the structure allproc
2014 * scans may still reference the structure. Maintain proper
2015 * locks until we can replace p->p_fd.
2017 * Also note that kqueue's closef still needs to reference the
2018 * fdp via p->p_fd, so we have to close the descriptors before
2019 * we replace p->p_fd.
2021 for (i = 0; i <= fdp->fd_lastfile; ++i) {
2022 if (fdp->fd_files[i].fp) {
2023 fp = funsetfd_locked(fdp, i);
2025 spin_unlock(&fdp->fd_spin);
2026 if (SLIST_FIRST(&fp->f_klist))
2027 knote_fdclose(fp, fdp, i);
2029 spin_lock(&fdp->fd_spin);
2033 spin_unlock(&fdp->fd_spin);
2036 * Interlock against an allproc scan operations (typically frevoke).
2038 spin_lock(&p->p_spin);
2040 spin_unlock(&p->p_spin);
2043 * Wait for any softrefs to go away. This race rarely occurs so
2044 * we can use a non-critical-path style poll/sleep loop. The
2045 * race only occurs against allproc scans.
2047 * No new softrefs can occur with the fdp disconnected from the
2050 if (fdp->fd_softrefs) {
2051 kprintf("pid %d: Warning, fdp race avoided\n", p->p_pid);
2052 while (fdp->fd_softrefs)
2053 tsleep(&fdp->fd_softrefs, 0, "fdsoft", 1);
2056 if (fdp->fd_files != fdp->fd_builtin_files)
2057 kfree(fdp->fd_files, M_FILEDESC);
2059 cache_drop(&fdp->fd_ncdir);
2060 vrele(fdp->fd_cdir);
2063 cache_drop(&fdp->fd_nrdir);
2064 vrele(fdp->fd_rdir);
2067 cache_drop(&fdp->fd_njdir);
2068 vrele(fdp->fd_jdir);
2070 kfree(fdp, M_FILEDESC);
2074 * Retrieve and reference the file pointer associated with a descriptor.
2079 holdfp(struct filedesc *fdp, int fd, int flag)
2083 spin_lock(&fdp->fd_spin);
2084 if (((u_int)fd) >= fdp->fd_nfiles) {
2088 if ((fp = fdp->fd_files[fd].fp) == NULL)
2090 if ((fp->f_flag & flag) == 0 && flag != -1) {
2096 spin_unlock(&fdp->fd_spin);
2101 * holdsock() - load the struct file pointer associated
2102 * with a socket into *fpp. If an error occurs, non-zero
2103 * will be returned and *fpp will be set to NULL.
2108 holdsock(struct filedesc *fdp, int fd, struct file **fpp)
2113 spin_lock(&fdp->fd_spin);
2114 if ((unsigned)fd >= fdp->fd_nfiles) {
2119 if ((fp = fdp->fd_files[fd].fp) == NULL) {
2123 if (fp->f_type != DTYPE_SOCKET) {
2130 spin_unlock(&fdp->fd_spin);
2136 * Convert a user file descriptor to a held file pointer.
2141 holdvnode(struct filedesc *fdp, int fd, struct file **fpp)
2146 spin_lock(&fdp->fd_spin);
2147 if ((unsigned)fd >= fdp->fd_nfiles) {
2152 if ((fp = fdp->fd_files[fd].fp) == NULL) {
2156 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
2164 spin_unlock(&fdp->fd_spin);
2170 * For setugid programs, we don't want to people to use that setugidness
2171 * to generate error messages which write to a file which otherwise would
2172 * otherwise be off-limits to the process.
2174 * This is a gross hack to plug the hole. A better solution would involve
2175 * a special vop or other form of generalized access control mechanism. We
2176 * go ahead and just reject all procfs file systems accesses as dangerous.
2178 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2179 * sufficient. We also don't for check setugidness since we know we are.
2182 is_unsafe(struct file *fp)
2184 if (fp->f_type == DTYPE_VNODE &&
2185 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS)
2191 * Make this setguid thing safe, if at all possible.
2193 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2196 setugidsafety(struct proc *p)
2198 struct filedesc *fdp = p->p_fd;
2201 /* Certain daemons might not have file descriptors. */
2206 * note: fdp->fd_files may be reallocated out from under us while
2207 * we are blocked in a close. Be careful!
2209 for (i = 0; i <= fdp->fd_lastfile; i++) {
2212 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) {
2216 * NULL-out descriptor prior to close to avoid
2217 * a race while close blocks.
2219 if ((fp = funsetfd_locked(fdp, i)) != NULL) {
2220 knote_fdclose(fp, fdp, i);
2228 * Close any files on exec?
2230 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2233 fdcloseexec(struct proc *p)
2235 struct filedesc *fdp = p->p_fd;
2238 /* Certain daemons might not have file descriptors. */
2243 * We cannot cache fd_files since operations may block and rip
2244 * them out from under us.
2246 for (i = 0; i <= fdp->fd_lastfile; i++) {
2247 if (fdp->fd_files[i].fp != NULL &&
2248 (fdp->fd_files[i].fileflags & UF_EXCLOSE)) {
2252 * NULL-out descriptor prior to close to avoid
2253 * a race while close blocks.
2255 if ((fp = funsetfd_locked(fdp, i)) != NULL) {
2256 knote_fdclose(fp, fdp, i);
2264 * It is unsafe for set[ug]id processes to be started with file
2265 * descriptors 0..2 closed, as these descriptors are given implicit
2266 * significance in the Standard C library. fdcheckstd() will create a
2267 * descriptor referencing /dev/null for each of stdin, stdout, and
2268 * stderr that is not already open.
2270 * NOT MPSAFE - calls falloc, vn_open, etc
2273 fdcheckstd(struct lwp *lp)
2275 struct nlookupdata nd;
2276 struct filedesc *fdp;
2279 int i, error, flags, devnull;
2281 fdp = lp->lwp_proc->p_fd;
2286 for (i = 0; i < 3; i++) {
2287 if (fdp->fd_files[i].fp != NULL)
2290 if ((error = falloc(lp, &fp, &devnull)) != 0)
2293 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE,
2294 NLC_FOLLOW|NLC_LOCKVP);
2295 flags = FREAD | FWRITE;
2297 error = vn_open(&nd, fp, flags, 0);
2299 fsetfd(fdp, fp, devnull);
2301 fsetfd(fdp, NULL, devnull);
2306 KKASSERT(i == devnull);
2308 error = kern_dup(DUP_FIXED, devnull, i, &retval);
2317 * Internal form of close.
2318 * Decrement reference count on file structure.
2319 * Note: td and/or p may be NULL when closing a file
2320 * that was being passed in a message.
2322 * MPALMOSTSAFE - acquires mplock for VOP operations
2325 closef(struct file *fp, struct proc *p)
2329 struct filedesc_to_leader *fdtol;
2335 * POSIX record locking dictates that any close releases ALL
2336 * locks owned by this process. This is handled by setting
2337 * a flag in the unlock to free ONLY locks obeying POSIX
2338 * semantics, and not to free BSD-style file locks.
2339 * If the descriptor was in a message, POSIX-style locks
2340 * aren't passed with the descriptor.
2342 if (p != NULL && fp->f_type == DTYPE_VNODE &&
2343 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2345 if ((p->p_leader->p_flag & P_ADVLOCK) != 0) {
2346 lf.l_whence = SEEK_SET;
2349 lf.l_type = F_UNLCK;
2350 vp = (struct vnode *)fp->f_data;
2351 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
2355 if (fdtol != NULL) {
2356 lwkt_gettoken(&p->p_token);
2358 * Handle special case where file descriptor table
2359 * is shared between multiple process leaders.
2361 for (fdtol = fdtol->fdl_next;
2362 fdtol != p->p_fdtol;
2363 fdtol = fdtol->fdl_next) {
2364 if ((fdtol->fdl_leader->p_flag &
2367 fdtol->fdl_holdcount++;
2368 lf.l_whence = SEEK_SET;
2371 lf.l_type = F_UNLCK;
2372 vp = (struct vnode *)fp->f_data;
2373 (void) VOP_ADVLOCK(vp,
2374 (caddr_t)fdtol->fdl_leader,
2375 F_UNLCK, &lf, F_POSIX);
2376 fdtol->fdl_holdcount--;
2377 if (fdtol->fdl_holdcount == 0 &&
2378 fdtol->fdl_wakeup != 0) {
2379 fdtol->fdl_wakeup = 0;
2383 lwkt_reltoken(&p->p_token);
2392 * fhold() can only be called if f_count is already at least 1 (i.e. the
2393 * caller of fhold() already has a reference to the file pointer in some
2396 * f_count is not spin-locked. Instead, atomic ops are used for
2397 * incrementing, decrementing, and handling the 1->0 transition.
2400 fhold(struct file *fp)
2402 atomic_add_int(&fp->f_count, 1);
2406 * fdrop() - drop a reference to a descriptor
2408 * MPALMOSTSAFE - acquires mplock for final close sequence
2411 fdrop(struct file *fp)
2418 * A combined fetch and subtract is needed to properly detect
2419 * 1->0 transitions, otherwise two cpus dropping from a ref
2420 * count of 2 might both try to run the 1->0 code.
2422 if (atomic_fetchadd_int(&fp->f_count, -1) > 1)
2425 KKASSERT(SLIST_FIRST(&fp->f_klist) == NULL);
2428 * The last reference has gone away, we own the fp structure free
2431 if (fp->f_count < 0)
2432 panic("fdrop: count < 0");
2433 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE &&
2434 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2436 lf.l_whence = SEEK_SET;
2439 lf.l_type = F_UNLCK;
2440 vp = (struct vnode *)fp->f_data;
2441 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2443 if (fp->f_ops != &badfileops)
2444 error = fo_close(fp);
2452 * Apply an advisory lock on a file descriptor.
2454 * Just attempt to get a record lock of the requested type on
2455 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2460 sys_flock(struct flock_args *uap)
2462 struct proc *p = curproc;
2468 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
2470 if (fp->f_type != DTYPE_VNODE) {
2474 vp = (struct vnode *)fp->f_data;
2475 lf.l_whence = SEEK_SET;
2478 if (uap->how & LOCK_UN) {
2479 lf.l_type = F_UNLCK;
2480 fp->f_flag &= ~FHASLOCK;
2481 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2484 if (uap->how & LOCK_EX)
2485 lf.l_type = F_WRLCK;
2486 else if (uap->how & LOCK_SH)
2487 lf.l_type = F_RDLCK;
2492 fp->f_flag |= FHASLOCK;
2493 if (uap->how & LOCK_NB)
2494 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0);
2496 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT);
2503 * File Descriptor pseudo-device driver (/dev/fd/).
2505 * Opening minor device N dup()s the file (if any) connected to file
2506 * descriptor N belonging to the calling process. Note that this driver
2507 * consists of only the ``open()'' routine, because all subsequent
2508 * references to this file will be direct to the other driver.
2511 fdopen(struct dev_open_args *ap)
2513 thread_t td = curthread;
2515 KKASSERT(td->td_lwp != NULL);
2518 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the
2519 * the file descriptor being sought for duplication. The error
2520 * return ensures that the vnode for this device will be released
2521 * by vn_open. Open will detect this special error and take the
2522 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2523 * will simply report the error.
2525 td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev);
2530 * The caller has reserved the file descriptor dfd for us. On success we
2531 * must fsetfd() it. On failure the caller will clean it up.
2536 dupfdopen(struct filedesc *fdp, int dfd, int sfd, int mode, int error)
2542 if ((wfp = holdfp(fdp, sfd, -1)) == NULL)
2546 * Close a revoke/dup race. Duping a descriptor marked as revoked
2547 * will dup a dummy descriptor instead of the real one.
2549 if (wfp->f_flag & FREVOKED) {
2550 kprintf("Warning: attempt to dup() a revoked descriptor\n");
2553 werror = falloc(NULL, &wfp, NULL);
2559 * There are two cases of interest here.
2561 * For ENODEV simply dup sfd to file descriptor dfd and return.
2563 * For ENXIO steal away the file structure from sfd and store it
2564 * dfd. sfd is effectively closed by this operation.
2566 * Any other error code is just returned.
2571 * Check that the mode the file is being opened for is a
2572 * subset of the mode of the existing descriptor.
2574 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2578 spin_lock(&fdp->fd_spin);
2579 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2580 fsetfd_locked(fdp, wfp, dfd);
2581 spin_unlock(&fdp->fd_spin);
2586 * Steal away the file pointer from dfd, and stuff it into indx.
2588 spin_lock(&fdp->fd_spin);
2589 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2590 fsetfd(fdp, wfp, dfd);
2591 if ((xfp = funsetfd_locked(fdp, sfd)) != NULL) {
2592 spin_unlock(&fdp->fd_spin);
2595 spin_unlock(&fdp->fd_spin);
2607 * NOT MPSAFE - I think these refer to a common file descriptor table
2608 * and we need to spinlock that to link fdtol in.
2610 struct filedesc_to_leader *
2611 filedesc_to_leader_alloc(struct filedesc_to_leader *old,
2612 struct proc *leader)
2614 struct filedesc_to_leader *fdtol;
2616 fdtol = kmalloc(sizeof(struct filedesc_to_leader),
2617 M_FILEDESC_TO_LEADER, M_WAITOK | M_ZERO);
2618 fdtol->fdl_refcount = 1;
2619 fdtol->fdl_holdcount = 0;
2620 fdtol->fdl_wakeup = 0;
2621 fdtol->fdl_leader = leader;
2623 fdtol->fdl_next = old->fdl_next;
2624 fdtol->fdl_prev = old;
2625 old->fdl_next = fdtol;
2626 fdtol->fdl_next->fdl_prev = fdtol;
2628 fdtol->fdl_next = fdtol;
2629 fdtol->fdl_prev = fdtol;
2635 * Scan all file pointers in the system. The callback is made with
2636 * the master list spinlock held exclusively.
2641 allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data)
2646 spin_lock(&filehead_spin);
2647 LIST_FOREACH(fp, &filehead, f_list) {
2648 res = callback(fp, data);
2652 spin_unlock(&filehead_spin);
2656 * Get file structures.
2658 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe)
2661 struct sysctl_kern_file_info {
2664 struct sysctl_req *req;
2667 static int sysctl_kern_file_callback(struct proc *p, void *data);
2670 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2672 struct sysctl_kern_file_info info;
2675 * Note: because the number of file descriptors is calculated
2676 * in different ways for sizing vs returning the data,
2677 * there is information leakage from the first loop. However,
2678 * it is of a similar order of magnitude to the leakage from
2679 * global system statistics such as kern.openfiles.
2681 * When just doing a count, note that we cannot just count
2682 * the elements and add f_count via the filehead list because
2683 * threaded processes share their descriptor table and f_count might
2684 * still be '1' in that case.
2686 * Since the SYSCTL op can block, we must hold the process to
2687 * prevent it being ripped out from under us either in the
2688 * file descriptor loop or in the greater LIST_FOREACH. The
2689 * process may be in varying states of disrepair. If the process
2690 * is in SZOMB we may have caught it just as it is being removed
2691 * from the allproc list, we must skip it in that case to maintain
2692 * an unbroken chain through the allproc list.
2697 allproc_scan(sysctl_kern_file_callback, &info);
2700 * When just calculating the size, overestimate a bit to try to
2701 * prevent system activity from causing the buffer-fill call
2704 if (req->oldptr == NULL) {
2705 info.count = (info.count + 16) + (info.count / 10);
2706 info.error = SYSCTL_OUT(req, NULL,
2707 info.count * sizeof(struct kinfo_file));
2709 return (info.error);
2713 sysctl_kern_file_callback(struct proc *p, void *data)
2715 struct sysctl_kern_file_info *info = data;
2716 struct kinfo_file kf;
2717 struct filedesc *fdp;
2722 if (p->p_stat == SIDL || p->p_stat == SZOMB)
2724 if (!PRISON_CHECK(info->req->td->td_ucred, p->p_ucred) != 0)
2728 * Softref the fdp to prevent it from being destroyed
2730 spin_lock(&p->p_spin);
2731 if ((fdp = p->p_fd) == NULL) {
2732 spin_unlock(&p->p_spin);
2735 atomic_add_int(&fdp->fd_softrefs, 1);
2736 spin_unlock(&p->p_spin);
2739 * The fdp's own spinlock prevents the contents from being
2742 spin_lock(&fdp->fd_spin);
2743 for (n = 0; n < fdp->fd_nfiles; ++n) {
2744 if ((fp = fdp->fd_files[n].fp) == NULL)
2746 if (info->req->oldptr == NULL) {
2749 uid = p->p_ucred ? p->p_ucred->cr_uid : -1;
2750 kcore_make_file(&kf, fp, p->p_pid, uid, n);
2751 spin_unlock(&fdp->fd_spin);
2752 info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf));
2753 spin_lock(&fdp->fd_spin);
2758 spin_unlock(&fdp->fd_spin);
2759 atomic_subtract_int(&fdp->fd_softrefs, 1);
2765 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2766 0, 0, sysctl_kern_file, "S,file", "Entire file table");
2768 SYSCTL_INT(_kern, OID_AUTO, minfilesperproc, CTLFLAG_RW,
2769 &minfilesperproc, 0, "Minimum files allowed open per process");
2770 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2771 &maxfilesperproc, 0, "Maximum files allowed open per process");
2772 SYSCTL_INT(_kern, OID_AUTO, maxfilesperuser, CTLFLAG_RW,
2773 &maxfilesperuser, 0, "Maximum files allowed open per user");
2775 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2776 &maxfiles, 0, "Maximum number of files");
2778 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW,
2779 &maxfilesrootres, 0, "Descriptors reserved for root use");
2781 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2782 &nfiles, 0, "System-wide number of open files");
2785 fildesc_drvinit(void *unused)
2789 for (fd = 0; fd < NUMFDESC; fd++) {
2790 make_dev(&fildesc_ops, fd,
2791 UID_BIN, GID_BIN, 0666, "fd/%d", fd);
2794 make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin");
2795 make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout");
2796 make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr");
2802 struct fileops badfileops = {
2803 .fo_read = badfo_readwrite,
2804 .fo_write = badfo_readwrite,
2805 .fo_ioctl = badfo_ioctl,
2806 .fo_kqfilter = badfo_kqfilter,
2807 .fo_stat = badfo_stat,
2808 .fo_close = badfo_close,
2809 .fo_shutdown = badfo_shutdown
2823 badfo_ioctl(struct file *fp, u_long com, caddr_t data,
2824 struct ucred *cred, struct sysmsg *msgv)
2830 * Must return an error to prevent registration, typically
2831 * due to a revoked descriptor (file_filtops assigned).
2834 badfo_kqfilter(struct file *fp, struct knote *kn)
2836 return (EOPNOTSUPP);
2843 badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
2852 badfo_close(struct file *fp)
2861 badfo_shutdown(struct file *fp, int how)
2870 nofo_shutdown(struct file *fp, int how)
2872 return (EOPNOTSUPP);
2875 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,
2876 fildesc_drvinit,NULL)