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. Neither the name of the University nor the names of its contributors
52 * may be used to endorse or promote products derived from this software
53 * without specific prior written permission.
55 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
56 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
57 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
58 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
59 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
60 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
61 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
62 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
63 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
64 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
67 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94
68 * $FreeBSD: src/sys/kern/kern_descrip.c,v 1.81.2.19 2004/02/28 00:43:31 tegge Exp $
71 #include "opt_compat.h"
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/malloc.h>
75 #include <sys/sysproto.h>
77 #include <sys/device.h>
79 #include <sys/filedesc.h>
80 #include <sys/kernel.h>
81 #include <sys/sysctl.h>
82 #include <sys/vnode.h>
84 #include <sys/nlookup.h>
86 #include <sys/filio.h>
87 #include <sys/fcntl.h>
88 #include <sys/unistd.h>
89 #include <sys/resourcevar.h>
90 #include <sys/event.h>
91 #include <sys/kern_syscall.h>
92 #include <sys/kcore.h>
93 #include <sys/kinfo.h>
97 #include <vm/vm_extern.h>
99 #include <sys/thread2.h>
100 #include <sys/file2.h>
101 #include <sys/spinlock2.h>
103 static void fsetfd_locked(struct filedesc
*fdp
, struct file
*fp
, int fd
);
104 static void fdreserve_locked (struct filedesc
*fdp
, int fd0
, int incr
);
105 static struct file
*funsetfd_locked (struct filedesc
*fdp
, int fd
);
106 static void ffree(struct file
*fp
);
108 static MALLOC_DEFINE(M_FILEDESC
, "file desc", "Open file descriptor table");
109 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER
, "file desc to leader",
110 "file desc to leader structures");
111 MALLOC_DEFINE(M_FILE
, "file", "Open file structure");
112 static MALLOC_DEFINE(M_SIGIO
, "sigio", "sigio structures");
114 static struct krate krate_uidinfo
= { .freq
= 1 };
116 static d_open_t fdopen
;
119 #define CDEV_MAJOR 22
120 static struct dev_ops fildesc_ops
= {
126 * Descriptor management.
128 static struct filelist filehead
= LIST_HEAD_INITIALIZER(&filehead
);
129 static struct spinlock filehead_spin
= SPINLOCK_INITIALIZER(&filehead_spin
, "filehead_spin");
130 static int nfiles
; /* actual number of open files */
134 * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared.
136 * MPSAFE - must be called with fdp->fd_spin exclusively held
140 fdfixup_locked(struct filedesc
*fdp
, int fd
)
142 if (fd
< fdp
->fd_freefile
) {
143 fdp
->fd_freefile
= fd
;
145 while (fdp
->fd_lastfile
>= 0 &&
146 fdp
->fd_files
[fdp
->fd_lastfile
].fp
== NULL
&&
147 fdp
->fd_files
[fdp
->fd_lastfile
].reserved
== 0
154 * System calls on descriptors.
159 sys_getdtablesize(struct getdtablesize_args
*uap
)
161 struct proc
*p
= curproc
;
162 struct plimit
*limit
= p
->p_limit
;
165 spin_lock(&limit
->p_spin
);
166 if (limit
->pl_rlimit
[RLIMIT_NOFILE
].rlim_cur
> INT_MAX
)
169 dtsize
= (int)limit
->pl_rlimit
[RLIMIT_NOFILE
].rlim_cur
;
170 spin_unlock(&limit
->p_spin
);
172 if (dtsize
> maxfilesperproc
)
173 dtsize
= maxfilesperproc
;
174 if (dtsize
< minfilesperproc
)
175 dtsize
= minfilesperproc
;
176 if (p
->p_ucred
->cr_uid
&& dtsize
> maxfilesperuser
)
177 dtsize
= maxfilesperuser
;
178 uap
->sysmsg_result
= dtsize
;
183 * Duplicate a file descriptor to a particular value.
185 * note: keep in mind that a potential race condition exists when closing
186 * descriptors from a shared descriptor table (via rfork).
191 sys_dup2(struct dup2_args
*uap
)
196 error
= kern_dup(DUP_FIXED
, uap
->from
, uap
->to
, &fd
);
197 uap
->sysmsg_fds
[0] = fd
;
203 * Duplicate a file descriptor.
208 sys_dup(struct dup_args
*uap
)
213 error
= kern_dup(DUP_VARIABLE
, uap
->fd
, 0, &fd
);
214 uap
->sysmsg_fds
[0] = fd
;
220 * MPALMOSTSAFE - acquires mplock for fp operations
223 kern_fcntl(int fd
, int cmd
, union fcntl_dat
*dat
, struct ucred
*cred
)
225 struct thread
*td
= curthread
;
226 struct proc
*p
= td
->td_proc
;
232 int tmp
, error
, flg
= F_POSIX
;
237 * Operations on file descriptors that do not require a file pointer.
241 error
= fgetfdflags(p
->p_fd
, fd
, &tmp
);
243 dat
->fc_cloexec
= (tmp
& UF_EXCLOSE
) ? FD_CLOEXEC
: 0;
247 if (dat
->fc_cloexec
& FD_CLOEXEC
)
248 error
= fsetfdflags(p
->p_fd
, fd
, UF_EXCLOSE
);
250 error
= fclrfdflags(p
->p_fd
, fd
, UF_EXCLOSE
);
254 error
= kern_dup(DUP_VARIABLE
| DUP_FCNTL
, fd
, newmin
,
257 case F_DUPFD_CLOEXEC
:
259 error
= kern_dup(DUP_VARIABLE
| DUP_CLOEXEC
| DUP_FCNTL
,
260 fd
, newmin
, &dat
->fc_fd
);
264 error
= kern_dup(DUP_FIXED
, fd
, newmin
, &dat
->fc_fd
);
266 case F_DUP2FD_CLOEXEC
:
268 error
= kern_dup(DUP_FIXED
| DUP_CLOEXEC
, fd
, newmin
,
276 * Operations on file pointers
278 if ((fp
= holdfp(p
->p_fd
, fd
, -1)) == NULL
)
283 dat
->fc_flags
= OFLAGS(fp
->f_flag
);
289 nflags
= FFLAGS(dat
->fc_flags
& ~O_ACCMODE
) & FCNTLFLAGS
;
290 nflags
|= oflags
& ~FCNTLFLAGS
;
293 if (((nflags
^ oflags
) & O_APPEND
) && (oflags
& FAPPENDONLY
))
295 if (error
== 0 && ((nflags
^ oflags
) & FASYNC
)) {
296 tmp
= nflags
& FASYNC
;
297 error
= fo_ioctl(fp
, FIOASYNC
, (caddr_t
)&tmp
,
305 error
= fo_ioctl(fp
, FIOGETOWN
, (caddr_t
)&dat
->fc_owner
,
310 error
= fo_ioctl(fp
, FIOSETOWN
, (caddr_t
)&dat
->fc_owner
,
316 /* Fall into F_SETLK */
319 if (fp
->f_type
!= DTYPE_VNODE
) {
323 vp
= (struct vnode
*)fp
->f_data
;
326 * copyin/lockop may block
328 if (dat
->fc_flock
.l_whence
== SEEK_CUR
)
329 dat
->fc_flock
.l_start
+= fp
->f_offset
;
331 switch (dat
->fc_flock
.l_type
) {
333 if ((fp
->f_flag
& FREAD
) == 0) {
337 if ((p
->p_leader
->p_flags
& P_ADVLOCK
) == 0) {
338 lwkt_gettoken(&p
->p_leader
->p_token
);
339 p
->p_leader
->p_flags
|= P_ADVLOCK
;
340 lwkt_reltoken(&p
->p_leader
->p_token
);
342 error
= VOP_ADVLOCK(vp
, (caddr_t
)p
->p_leader
, F_SETLK
,
343 &dat
->fc_flock
, flg
);
346 if ((fp
->f_flag
& FWRITE
) == 0) {
350 if ((p
->p_leader
->p_flags
& P_ADVLOCK
) == 0) {
351 lwkt_gettoken(&p
->p_leader
->p_token
);
352 p
->p_leader
->p_flags
|= P_ADVLOCK
;
353 lwkt_reltoken(&p
->p_leader
->p_token
);
355 error
= VOP_ADVLOCK(vp
, (caddr_t
)p
->p_leader
, F_SETLK
,
356 &dat
->fc_flock
, flg
);
359 error
= VOP_ADVLOCK(vp
, (caddr_t
)p
->p_leader
, F_UNLCK
,
360 &dat
->fc_flock
, F_POSIX
);
368 * It is possible to race a close() on the descriptor while
369 * we were blocked getting the lock. If this occurs the
370 * close might not have caught the lock.
372 if (checkfdclosed(p
->p_fd
, fd
, fp
)) {
373 dat
->fc_flock
.l_whence
= SEEK_SET
;
374 dat
->fc_flock
.l_start
= 0;
375 dat
->fc_flock
.l_len
= 0;
376 dat
->fc_flock
.l_type
= F_UNLCK
;
377 (void) VOP_ADVLOCK(vp
, (caddr_t
)p
->p_leader
,
378 F_UNLCK
, &dat
->fc_flock
, F_POSIX
);
383 if (fp
->f_type
!= DTYPE_VNODE
) {
387 vp
= (struct vnode
*)fp
->f_data
;
389 * copyin/lockop may block
391 if (dat
->fc_flock
.l_type
!= F_RDLCK
&&
392 dat
->fc_flock
.l_type
!= F_WRLCK
&&
393 dat
->fc_flock
.l_type
!= F_UNLCK
) {
397 if (dat
->fc_flock
.l_whence
== SEEK_CUR
)
398 dat
->fc_flock
.l_start
+= fp
->f_offset
;
399 error
= VOP_ADVLOCK(vp
, (caddr_t
)p
->p_leader
, F_GETLK
,
400 &dat
->fc_flock
, F_POSIX
);
412 * The file control system call.
417 sys_fcntl(struct fcntl_args
*uap
)
425 case F_DUPFD_CLOEXEC
:
426 case F_DUP2FD_CLOEXEC
:
427 dat
.fc_fd
= uap
->arg
;
430 dat
.fc_cloexec
= uap
->arg
;
433 dat
.fc_flags
= uap
->arg
;
436 dat
.fc_owner
= uap
->arg
;
441 error
= copyin((caddr_t
)uap
->arg
, &dat
.fc_flock
,
442 sizeof(struct flock
));
448 error
= kern_fcntl(uap
->fd
, uap
->cmd
, &dat
, curthread
->td_ucred
);
454 case F_DUPFD_CLOEXEC
:
455 case F_DUP2FD_CLOEXEC
:
456 uap
->sysmsg_result
= dat
.fc_fd
;
459 uap
->sysmsg_result
= dat
.fc_cloexec
;
462 uap
->sysmsg_result
= dat
.fc_flags
;
465 uap
->sysmsg_result
= dat
.fc_owner
;
468 error
= copyout(&dat
.fc_flock
, (caddr_t
)uap
->arg
,
469 sizeof(struct flock
));
478 * Common code for dup, dup2, and fcntl(F_DUPFD).
480 * There are four type flags: DUP_FCNTL, DUP_FIXED, DUP_VARIABLE, and
483 * DUP_FCNTL is for handling EINVAL vs. EBADF differences between
484 * fcntl()'s F_DUPFD and F_DUPFD_CLOEXEC and dup2() (per POSIX).
485 * The next two flags are mutually exclusive, and the fourth is optional.
486 * DUP_FIXED tells kern_dup() to destructively dup over an existing file
487 * descriptor if "new" is already open. DUP_VARIABLE tells kern_dup()
488 * to find the lowest unused file descriptor that is greater than or
489 * equal to "new". DUP_CLOEXEC, which works with either of the first
490 * two flags, sets the close-on-exec flag on the "new" file descriptor.
495 kern_dup(int flags
, int old
, int new, int *res
)
497 struct thread
*td
= curthread
;
498 struct proc
*p
= td
->td_proc
;
499 struct filedesc
*fdp
= p
->p_fd
;
508 * Verify that we have a valid descriptor to dup from and
509 * possibly to dup to. When the new descriptor is out of
510 * bounds, fcntl()'s F_DUPFD and F_DUPFD_CLOEXEC must
511 * return EINVAL, while dup2() returns EBADF in
514 * NOTE: maxfilesperuser is not applicable to dup()
517 if (p
->p_rlimit
[RLIMIT_NOFILE
].rlim_cur
> INT_MAX
)
520 dtsize
= (int)p
->p_rlimit
[RLIMIT_NOFILE
].rlim_cur
;
521 if (dtsize
> maxfilesperproc
)
522 dtsize
= maxfilesperproc
;
523 if (dtsize
< minfilesperproc
)
524 dtsize
= minfilesperproc
;
526 if (new < 0 || new > dtsize
)
527 return (flags
& DUP_FCNTL
? EINVAL
: EBADF
);
529 spin_lock(&fdp
->fd_spin
);
530 if ((unsigned)old
>= fdp
->fd_nfiles
|| fdp
->fd_files
[old
].fp
== NULL
) {
531 spin_unlock(&fdp
->fd_spin
);
534 if ((flags
& DUP_FIXED
) && old
== new) {
536 if (flags
& DUP_CLOEXEC
)
537 fdp
->fd_files
[new].fileflags
|= UF_EXCLOSE
;
538 spin_unlock(&fdp
->fd_spin
);
541 fp
= fdp
->fd_files
[old
].fp
;
542 oldflags
= fdp
->fd_files
[old
].fileflags
;
543 fhold(fp
); /* MPSAFE - can be called with a spinlock held */
546 * Allocate a new descriptor if DUP_VARIABLE, or expand the table
547 * if the requested descriptor is beyond the current table size.
549 * This can block. Retry if the source descriptor no longer matches
550 * or if our expectation in the expansion case races.
552 * If we are not expanding or allocating a new decriptor, then reset
553 * the target descriptor to a reserved state so we have a uniform
554 * setup for the next code block.
556 if ((flags
& DUP_VARIABLE
) || new >= fdp
->fd_nfiles
) {
557 spin_unlock(&fdp
->fd_spin
);
558 error
= fdalloc(p
, new, &newfd
);
559 spin_lock(&fdp
->fd_spin
);
561 spin_unlock(&fdp
->fd_spin
);
568 if (old
>= fdp
->fd_nfiles
|| fdp
->fd_files
[old
].fp
!= fp
) {
569 fsetfd_locked(fdp
, NULL
, newfd
);
570 spin_unlock(&fdp
->fd_spin
);
575 * Check for expansion race
577 if ((flags
& DUP_VARIABLE
) == 0 && new != newfd
) {
578 fsetfd_locked(fdp
, NULL
, newfd
);
579 spin_unlock(&fdp
->fd_spin
);
584 * Check for ripout, newfd reused old (this case probably
588 fsetfd_locked(fdp
, NULL
, newfd
);
589 spin_unlock(&fdp
->fd_spin
);
596 if (fdp
->fd_files
[new].reserved
) {
597 spin_unlock(&fdp
->fd_spin
);
599 kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new);
600 tsleep(fdp
, 0, "fdres", hz
);
605 * If the target descriptor was never allocated we have
606 * to allocate it. If it was we have to clean out the
607 * old descriptor. delfp inherits the ref from the
610 delfp
= fdp
->fd_files
[new].fp
;
611 fdp
->fd_files
[new].fp
= NULL
;
612 fdp
->fd_files
[new].reserved
= 1;
614 fdreserve_locked(fdp
, new, 1);
615 if (new > fdp
->fd_lastfile
)
616 fdp
->fd_lastfile
= new;
622 * NOTE: still holding an exclusive spinlock
626 * If a descriptor is being overwritten we may hve to tell
627 * fdfree() to sleep to ensure that all relevant process
628 * leaders can be traversed in closef().
630 if (delfp
!= NULL
&& p
->p_fdtol
!= NULL
) {
631 fdp
->fd_holdleaderscount
++;
636 KASSERT(delfp
== NULL
|| (flags
& DUP_FIXED
),
637 ("dup() picked an open file"));
640 * Duplicate the source descriptor, update lastfile. If the new
641 * descriptor was not allocated and we aren't replacing an existing
642 * descriptor we have to mark the descriptor as being in use.
644 * The fd_files[] array inherits fp's hold reference.
646 fsetfd_locked(fdp
, fp
, new);
647 if ((flags
& DUP_CLOEXEC
) != 0)
648 fdp
->fd_files
[new].fileflags
= oldflags
| UF_EXCLOSE
;
650 fdp
->fd_files
[new].fileflags
= oldflags
& ~UF_EXCLOSE
;
651 spin_unlock(&fdp
->fd_spin
);
656 * If we dup'd over a valid file, we now own the reference to it
657 * and must dispose of it using closef() semantics (as if a
658 * close() were performed on it).
661 if (SLIST_FIRST(&delfp
->f_klist
))
662 knote_fdclose(delfp
, fdp
, new);
665 spin_lock(&fdp
->fd_spin
);
666 fdp
->fd_holdleaderscount
--;
667 if (fdp
->fd_holdleaderscount
== 0 &&
668 fdp
->fd_holdleaderswakeup
!= 0) {
669 fdp
->fd_holdleaderswakeup
= 0;
670 spin_unlock(&fdp
->fd_spin
);
671 wakeup(&fdp
->fd_holdleaderscount
);
673 spin_unlock(&fdp
->fd_spin
);
681 * If sigio is on the list associated with a process or process group,
682 * disable signalling from the device, remove sigio from the list and
688 funsetown(struct sigio
**sigiop
)
694 if ((sigio
= *sigiop
) != NULL
) {
695 lwkt_gettoken(&sigio_token
); /* protect sigio */
696 KKASSERT(sigiop
== sigio
->sio_myref
);
699 lwkt_reltoken(&sigio_token
);
704 if (sigio
->sio_pgid
< 0) {
705 pgrp
= sigio
->sio_pgrp
;
706 sigio
->sio_pgrp
= NULL
;
707 lwkt_gettoken(&pgrp
->pg_token
);
708 SLIST_REMOVE(&pgrp
->pg_sigiolst
, sigio
, sigio
, sio_pgsigio
);
709 lwkt_reltoken(&pgrp
->pg_token
);
711 } else /* if ((*sigiop)->sio_pgid > 0) */ {
713 sigio
->sio_proc
= NULL
;
715 lwkt_gettoken(&p
->p_token
);
716 SLIST_REMOVE(&p
->p_sigiolst
, sigio
, sigio
, sio_pgsigio
);
717 lwkt_reltoken(&p
->p_token
);
720 crfree(sigio
->sio_ucred
);
721 sigio
->sio_ucred
= NULL
;
722 kfree(sigio
, M_SIGIO
);
726 * Free a list of sigio structures. Caller is responsible for ensuring
727 * that the list is MPSAFE.
732 funsetownlst(struct sigiolst
*sigiolst
)
736 while ((sigio
= SLIST_FIRST(sigiolst
)) != NULL
)
737 funsetown(sigio
->sio_myref
);
741 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
743 * After permission checking, add a sigio structure to the sigio list for
744 * the process or process group.
749 fsetown(pid_t pgid
, struct sigio
**sigiop
)
751 struct proc
*proc
= NULL
;
752 struct pgrp
*pgrp
= NULL
;
769 * Policy - Don't allow a process to FSETOWN a process
770 * in another session.
772 * Remove this test to allow maximum flexibility or
773 * restrict FSETOWN to the current process or process
774 * group for maximum safety.
776 if (proc
->p_session
!= curproc
->p_session
) {
780 } else /* if (pgid < 0) */ {
781 pgrp
= pgfind(-pgid
);
788 * Policy - Don't allow a process to FSETOWN a process
789 * in another session.
791 * Remove this test to allow maximum flexibility or
792 * restrict FSETOWN to the current process or process
793 * group for maximum safety.
795 if (pgrp
->pg_session
!= curproc
->p_session
) {
800 sigio
= kmalloc(sizeof(struct sigio
), M_SIGIO
, M_WAITOK
| M_ZERO
);
802 KKASSERT(pgrp
== NULL
);
803 lwkt_gettoken(&proc
->p_token
);
804 SLIST_INSERT_HEAD(&proc
->p_sigiolst
, sigio
, sio_pgsigio
);
805 sigio
->sio_proc
= proc
;
806 lwkt_reltoken(&proc
->p_token
);
808 KKASSERT(proc
== NULL
);
809 lwkt_gettoken(&pgrp
->pg_token
);
810 SLIST_INSERT_HEAD(&pgrp
->pg_sigiolst
, sigio
, sio_pgsigio
);
811 sigio
->sio_pgrp
= pgrp
;
812 lwkt_reltoken(&pgrp
->pg_token
);
815 sigio
->sio_pgid
= pgid
;
816 sigio
->sio_ucred
= crhold(curthread
->td_ucred
);
817 /* It would be convenient if p_ruid was in ucred. */
818 sigio
->sio_ruid
= sigio
->sio_ucred
->cr_ruid
;
819 sigio
->sio_myref
= sigiop
;
821 lwkt_gettoken(&sigio_token
);
825 lwkt_reltoken(&sigio_token
);
836 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
841 fgetown(struct sigio
**sigiop
)
846 lwkt_gettoken_shared(&sigio_token
);
848 own
= (sigio
!= NULL
? sigio
->sio_pgid
: 0);
849 lwkt_reltoken(&sigio_token
);
855 * Close many file descriptors.
860 sys_closefrom(struct closefrom_args
*uap
)
862 return(kern_closefrom(uap
->fd
));
866 * Close all file descriptors greater then or equal to fd
871 kern_closefrom(int fd
)
873 struct thread
*td
= curthread
;
874 struct proc
*p
= td
->td_proc
;
875 struct filedesc
*fdp
;
884 * NOTE: This function will skip unassociated descriptors and
885 * reserved descriptors that have not yet been assigned.
886 * fd_lastfile can change as a side effect of kern_close().
888 spin_lock(&fdp
->fd_spin
);
889 while (fd
<= fdp
->fd_lastfile
) {
890 if (fdp
->fd_files
[fd
].fp
!= NULL
) {
891 spin_unlock(&fdp
->fd_spin
);
892 /* ok if this races another close */
893 if (kern_close(fd
) == EINTR
)
895 spin_lock(&fdp
->fd_spin
);
899 spin_unlock(&fdp
->fd_spin
);
904 * Close a file descriptor.
909 sys_close(struct close_args
*uap
)
911 return(kern_close(uap
->fd
));
920 struct thread
*td
= curthread
;
921 struct proc
*p
= td
->td_proc
;
922 struct filedesc
*fdp
;
930 spin_lock(&fdp
->fd_spin
);
931 if ((fp
= funsetfd_locked(fdp
, fd
)) == NULL
) {
932 spin_unlock(&fdp
->fd_spin
);
936 if (p
->p_fdtol
!= NULL
) {
938 * Ask fdfree() to sleep to ensure that all relevant
939 * process leaders can be traversed in closef().
941 fdp
->fd_holdleaderscount
++;
946 * we now hold the fp reference that used to be owned by the descriptor
949 spin_unlock(&fdp
->fd_spin
);
950 if (SLIST_FIRST(&fp
->f_klist
))
951 knote_fdclose(fp
, fdp
, fd
);
952 error
= closef(fp
, p
);
954 spin_lock(&fdp
->fd_spin
);
955 fdp
->fd_holdleaderscount
--;
956 if (fdp
->fd_holdleaderscount
== 0 &&
957 fdp
->fd_holdleaderswakeup
!= 0) {
958 fdp
->fd_holdleaderswakeup
= 0;
959 spin_unlock(&fdp
->fd_spin
);
960 wakeup(&fdp
->fd_holdleaderscount
);
962 spin_unlock(&fdp
->fd_spin
);
969 * shutdown_args(int fd, int how)
972 kern_shutdown(int fd
, int how
)
974 struct thread
*td
= curthread
;
975 struct proc
*p
= td
->td_proc
;
981 if ((fp
= holdfp(p
->p_fd
, fd
, -1)) == NULL
)
983 error
= fo_shutdown(fp
, how
);
993 sys_shutdown(struct shutdown_args
*uap
)
997 error
= kern_shutdown(uap
->s
, uap
->how
);
1006 kern_fstat(int fd
, struct stat
*ub
)
1008 struct thread
*td
= curthread
;
1009 struct proc
*p
= td
->td_proc
;
1015 if ((fp
= holdfp(p
->p_fd
, fd
, -1)) == NULL
)
1017 error
= fo_stat(fp
, ub
, td
->td_ucred
);
1024 * Return status information about a file descriptor.
1029 sys_fstat(struct fstat_args
*uap
)
1034 error
= kern_fstat(uap
->fd
, &st
);
1037 error
= copyout(&st
, uap
->sb
, sizeof(st
));
1042 * Return pathconf information about a file descriptor.
1047 sys_fpathconf(struct fpathconf_args
*uap
)
1049 struct thread
*td
= curthread
;
1050 struct proc
*p
= td
->td_proc
;
1055 if ((fp
= holdfp(p
->p_fd
, uap
->fd
, -1)) == NULL
)
1058 switch (fp
->f_type
) {
1061 if (uap
->name
!= _PC_PIPE_BUF
) {
1064 uap
->sysmsg_result
= PIPE_BUF
;
1070 vp
= (struct vnode
*)fp
->f_data
;
1071 error
= VOP_PATHCONF(vp
, uap
->name
, &uap
->sysmsg_reg
);
1081 static int fdexpand
;
1082 SYSCTL_INT(_debug
, OID_AUTO
, fdexpand
, CTLFLAG_RD
, &fdexpand
, 0,
1083 "Number of times a file table has been expanded");
1086 * Grow the file table so it can hold through descriptor (want).
1088 * The fdp's spinlock must be held exclusively on entry and may be held
1089 * exclusively on return. The spinlock may be cycled by the routine.
1094 fdgrow_locked(struct filedesc
*fdp
, int want
)
1096 struct fdnode
*newfiles
;
1097 struct fdnode
*oldfiles
;
1100 nf
= fdp
->fd_nfiles
;
1102 /* nf has to be of the form 2^n - 1 */
1104 } while (nf
<= want
);
1106 spin_unlock(&fdp
->fd_spin
);
1107 newfiles
= kmalloc(nf
* sizeof(struct fdnode
), M_FILEDESC
, M_WAITOK
);
1108 spin_lock(&fdp
->fd_spin
);
1111 * We could have raced another extend while we were not holding
1114 if (fdp
->fd_nfiles
>= nf
) {
1115 spin_unlock(&fdp
->fd_spin
);
1116 kfree(newfiles
, M_FILEDESC
);
1117 spin_lock(&fdp
->fd_spin
);
1121 * Copy the existing ofile and ofileflags arrays
1122 * and zero the new portion of each array.
1124 extra
= nf
- fdp
->fd_nfiles
;
1125 bcopy(fdp
->fd_files
, newfiles
, fdp
->fd_nfiles
* sizeof(struct fdnode
));
1126 bzero(&newfiles
[fdp
->fd_nfiles
], extra
* sizeof(struct fdnode
));
1128 oldfiles
= fdp
->fd_files
;
1129 fdp
->fd_files
= newfiles
;
1130 fdp
->fd_nfiles
= nf
;
1132 if (oldfiles
!= fdp
->fd_builtin_files
) {
1133 spin_unlock(&fdp
->fd_spin
);
1134 kfree(oldfiles
, M_FILEDESC
);
1135 spin_lock(&fdp
->fd_spin
);
1141 * Number of nodes in right subtree, including the root.
1144 right_subtree_size(int n
)
1146 return (n
^ (n
| (n
+ 1)));
1153 right_ancestor(int n
)
1155 return (n
| (n
+ 1));
1162 left_ancestor(int n
)
1164 return ((n
& (n
+ 1)) - 1);
1168 * Traverse the in-place binary tree buttom-up adjusting the allocation
1169 * count so scans can determine where free descriptors are located.
1171 * MPSAFE - caller must be holding an exclusive spinlock on fdp
1175 fdreserve_locked(struct filedesc
*fdp
, int fd
, int incr
)
1178 fdp
->fd_files
[fd
].allocated
+= incr
;
1179 KKASSERT(fdp
->fd_files
[fd
].allocated
>= 0);
1180 fd
= left_ancestor(fd
);
1185 * Reserve a file descriptor for the process. If no error occurs, the
1186 * caller MUST at some point call fsetfd() or assign a file pointer
1187 * or dispose of the reservation.
1192 fdalloc(struct proc
*p
, int want
, int *result
)
1194 struct filedesc
*fdp
= p
->p_fd
;
1195 struct uidinfo
*uip
;
1196 int fd
, rsize
, rsum
, node
, lim
;
1199 * Check dtable size limit
1201 spin_lock(&p
->p_limit
->p_spin
);
1202 if (p
->p_rlimit
[RLIMIT_NOFILE
].rlim_cur
> INT_MAX
)
1205 lim
= (int)p
->p_rlimit
[RLIMIT_NOFILE
].rlim_cur
;
1206 spin_unlock(&p
->p_limit
->p_spin
);
1208 if (lim
> maxfilesperproc
)
1209 lim
= maxfilesperproc
;
1210 if (lim
< minfilesperproc
)
1211 lim
= minfilesperproc
;
1216 * Check that the user has not run out of descriptors (non-root only).
1217 * As a safety measure the dtable is allowed to have at least
1218 * minfilesperproc open fds regardless of the maxfilesperuser limit.
1220 if (p
->p_ucred
->cr_uid
&& fdp
->fd_nfiles
>= minfilesperproc
) {
1221 uip
= p
->p_ucred
->cr_uidinfo
;
1222 if (uip
->ui_openfiles
> maxfilesperuser
) {
1223 krateprintf(&krate_uidinfo
,
1224 "Warning: user %d pid %d (%s) ran out of "
1225 "file descriptors (%d/%d)\n",
1226 p
->p_ucred
->cr_uid
, (int)p
->p_pid
,
1228 uip
->ui_openfiles
, maxfilesperuser
);
1234 * Grow the dtable if necessary
1236 spin_lock(&fdp
->fd_spin
);
1237 if (want
>= fdp
->fd_nfiles
)
1238 fdgrow_locked(fdp
, want
);
1241 * Search for a free descriptor starting at the higher
1242 * of want or fd_freefile. If that fails, consider
1243 * expanding the ofile array.
1245 * NOTE! the 'allocated' field is a cumulative recursive allocation
1246 * count. If we happen to see a value of 0 then we can shortcut
1247 * our search. Otherwise we run through through the tree going
1248 * down branches we know have free descriptor(s) until we hit a
1249 * leaf node. The leaf node will be free but will not necessarily
1250 * have an allocated field of 0.
1253 /* move up the tree looking for a subtree with a free node */
1254 for (fd
= max(want
, fdp
->fd_freefile
); fd
< min(fdp
->fd_nfiles
, lim
);
1255 fd
= right_ancestor(fd
)) {
1256 if (fdp
->fd_files
[fd
].allocated
== 0)
1259 rsize
= right_subtree_size(fd
);
1260 if (fdp
->fd_files
[fd
].allocated
== rsize
)
1261 continue; /* right subtree full */
1264 * Free fd is in the right subtree of the tree rooted at fd.
1265 * Call that subtree R. Look for the smallest (leftmost)
1266 * subtree of R with an unallocated fd: continue moving
1267 * down the left branch until encountering a full left
1268 * subtree, then move to the right.
1270 for (rsum
= 0, rsize
/= 2; rsize
> 0; rsize
/= 2) {
1272 rsum
+= fdp
->fd_files
[node
].allocated
;
1273 if (fdp
->fd_files
[fd
].allocated
== rsum
+ rsize
) {
1274 fd
= node
; /* move to the right */
1275 if (fdp
->fd_files
[node
].allocated
== 0)
1284 * No space in current array. Expand?
1286 if (fdp
->fd_nfiles
>= lim
) {
1287 spin_unlock(&fdp
->fd_spin
);
1290 fdgrow_locked(fdp
, want
);
1294 KKASSERT(fd
< fdp
->fd_nfiles
);
1295 if (fd
> fdp
->fd_lastfile
)
1296 fdp
->fd_lastfile
= fd
;
1297 if (want
<= fdp
->fd_freefile
)
1298 fdp
->fd_freefile
= fd
;
1300 KKASSERT(fdp
->fd_files
[fd
].fp
== NULL
);
1301 KKASSERT(fdp
->fd_files
[fd
].reserved
== 0);
1302 fdp
->fd_files
[fd
].fileflags
= 0;
1303 fdp
->fd_files
[fd
].reserved
= 1;
1304 fdreserve_locked(fdp
, fd
, 1);
1305 spin_unlock(&fdp
->fd_spin
);
1310 * Check to see whether n user file descriptors
1311 * are available to the process p.
1316 fdavail(struct proc
*p
, int n
)
1318 struct filedesc
*fdp
= p
->p_fd
;
1319 struct fdnode
*fdnode
;
1322 spin_lock(&p
->p_limit
->p_spin
);
1323 if (p
->p_rlimit
[RLIMIT_NOFILE
].rlim_cur
> INT_MAX
)
1326 lim
= (int)p
->p_rlimit
[RLIMIT_NOFILE
].rlim_cur
;
1327 spin_unlock(&p
->p_limit
->p_spin
);
1329 if (lim
> maxfilesperproc
)
1330 lim
= maxfilesperproc
;
1331 if (lim
< minfilesperproc
)
1332 lim
= minfilesperproc
;
1334 spin_lock(&fdp
->fd_spin
);
1335 if ((i
= lim
- fdp
->fd_nfiles
) > 0 && (n
-= i
) <= 0) {
1336 spin_unlock(&fdp
->fd_spin
);
1339 last
= min(fdp
->fd_nfiles
, lim
);
1340 fdnode
= &fdp
->fd_files
[fdp
->fd_freefile
];
1341 for (i
= last
- fdp
->fd_freefile
; --i
>= 0; ++fdnode
) {
1342 if (fdnode
->fp
== NULL
&& --n
<= 0) {
1343 spin_unlock(&fdp
->fd_spin
);
1347 spin_unlock(&fdp
->fd_spin
);
1352 * Revoke open descriptors referencing (f_data, f_type)
1354 * Any revoke executed within a prison is only able to
1355 * revoke descriptors for processes within that prison.
1357 * Returns 0 on success or an error code.
1359 struct fdrevoke_info
{
1369 static int fdrevoke_check_callback(struct file
*fp
, void *vinfo
);
1370 static int fdrevoke_proc_callback(struct proc
*p
, void *vinfo
);
1373 fdrevoke(void *f_data
, short f_type
, struct ucred
*cred
)
1375 struct fdrevoke_info info
;
1378 bzero(&info
, sizeof(info
));
1382 error
= falloc(NULL
, &info
.nfp
, NULL
);
1387 * Scan the file pointer table once. dups do not dup file pointers,
1388 * only descriptors, so there is no leak. Set FREVOKED on the fps
1391 allfiles_scan_exclusive(fdrevoke_check_callback
, &info
);
1394 * If any fps were marked track down the related descriptors
1395 * and close them. Any dup()s at this point will notice
1396 * the FREVOKED already set in the fp and do the right thing.
1398 * Any fps with non-zero msgcounts (aka sent over a unix-domain
1399 * socket) bumped the intransit counter and will require a
1400 * scan. Races against fps leaving the socket are closed by
1401 * the socket code checking for FREVOKED.
1404 allproc_scan(fdrevoke_proc_callback
, &info
);
1406 unp_revoke_gc(info
.nfp
);
1412 * Locate matching file pointers directly.
1414 * WARNING: allfiles_scan_exclusive() holds a spinlock through these calls!
1417 fdrevoke_check_callback(struct file
*fp
, void *vinfo
)
1419 struct fdrevoke_info
*info
= vinfo
;
1422 * File pointers already flagged for revokation are skipped.
1424 if (fp
->f_flag
& FREVOKED
)
1428 * If revoking from a prison file pointers created outside of
1429 * that prison, or file pointers without creds, cannot be revoked.
1431 if (info
->cred
->cr_prison
&&
1432 (fp
->f_cred
== NULL
||
1433 info
->cred
->cr_prison
!= fp
->f_cred
->cr_prison
)) {
1438 * If the file pointer matches then mark it for revocation. The
1439 * flag is currently only used by unp_revoke_gc().
1441 * info->count is a heuristic and can race in a SMP environment.
1443 if (info
->data
== fp
->f_data
&& info
->type
== fp
->f_type
) {
1444 atomic_set_int(&fp
->f_flag
, FREVOKED
);
1445 info
->count
+= fp
->f_count
;
1453 * Locate matching file pointers via process descriptor tables.
1456 fdrevoke_proc_callback(struct proc
*p
, void *vinfo
)
1458 struct fdrevoke_info
*info
= vinfo
;
1459 struct filedesc
*fdp
;
1463 if (p
->p_stat
== SIDL
|| p
->p_stat
== SZOMB
)
1465 if (info
->cred
->cr_prison
&&
1466 info
->cred
->cr_prison
!= p
->p_ucred
->cr_prison
) {
1471 * If the controlling terminal of the process matches the
1472 * vnode being revoked we clear the controlling terminal.
1474 * The normal spec_close() may not catch this because it
1475 * uses curproc instead of p.
1477 if (p
->p_session
&& info
->type
== DTYPE_VNODE
&&
1478 info
->data
== p
->p_session
->s_ttyvp
) {
1479 p
->p_session
->s_ttyvp
= NULL
;
1484 * Softref the fdp to prevent it from being destroyed
1486 spin_lock(&p
->p_spin
);
1487 if ((fdp
= p
->p_fd
) == NULL
) {
1488 spin_unlock(&p
->p_spin
);
1491 atomic_add_int(&fdp
->fd_softrefs
, 1);
1492 spin_unlock(&p
->p_spin
);
1495 * Locate and close any matching file descriptors.
1497 spin_lock(&fdp
->fd_spin
);
1498 for (n
= 0; n
< fdp
->fd_nfiles
; ++n
) {
1499 if ((fp
= fdp
->fd_files
[n
].fp
) == NULL
)
1501 if (fp
->f_flag
& FREVOKED
) {
1503 fdp
->fd_files
[n
].fp
= info
->nfp
;
1504 spin_unlock(&fdp
->fd_spin
);
1505 knote_fdclose(fp
, fdp
, n
); /* XXX */
1507 spin_lock(&fdp
->fd_spin
);
1511 spin_unlock(&fdp
->fd_spin
);
1512 atomic_subtract_int(&fdp
->fd_softrefs
, 1);
1518 * Create a new open file structure and reserve a file decriptor
1519 * for the process that refers to it.
1521 * Root creds are checked using lp, or assumed if lp is NULL. If
1522 * resultfd is non-NULL then lp must also be non-NULL. No file
1523 * descriptor is reserved (and no process context is needed) if
1526 * A file pointer with a refcount of 1 is returned. Note that the
1527 * file pointer is NOT associated with the descriptor. If falloc
1528 * returns success, fsetfd() MUST be called to either associate the
1529 * file pointer or clear the reservation.
1534 falloc(struct lwp
*lp
, struct file
**resultfp
, int *resultfd
)
1536 static struct timeval lastfail
;
1539 struct ucred
*cred
= lp
? lp
->lwp_thread
->td_ucred
: proc0
.p_ucred
;
1545 * Handle filetable full issues and root overfill.
1547 if (nfiles
>= maxfiles
- maxfilesrootres
&&
1548 (cred
->cr_ruid
!= 0 || nfiles
>= maxfiles
)) {
1549 if (ppsratecheck(&lastfail
, &curfail
, 1)) {
1550 kprintf("kern.maxfiles limit exceeded by uid %d, "
1551 "please see tuning(7).\n",
1559 * Allocate a new file descriptor.
1561 fp
= kmalloc(sizeof(struct file
), M_FILE
, M_WAITOK
| M_ZERO
);
1562 spin_init(&fp
->f_spin
, "falloc");
1563 SLIST_INIT(&fp
->f_klist
);
1565 fp
->f_ops
= &badfileops
;
1568 spin_lock(&filehead_spin
);
1570 LIST_INSERT_HEAD(&filehead
, fp
, f_list
);
1571 spin_unlock(&filehead_spin
);
1573 if ((error
= fdalloc(lp
->lwp_proc
, 0, resultfd
)) != 0) {
1586 * Check for races against a file descriptor by determining that the
1587 * file pointer is still associated with the specified file descriptor,
1588 * and a close is not currently in progress.
1593 checkfdclosed(struct filedesc
*fdp
, int fd
, struct file
*fp
)
1597 spin_lock_shared(&fdp
->fd_spin
);
1598 if ((unsigned)fd
>= fdp
->fd_nfiles
|| fp
!= fdp
->fd_files
[fd
].fp
)
1602 spin_unlock_shared(&fdp
->fd_spin
);
1607 * Associate a file pointer with a previously reserved file descriptor.
1608 * This function always succeeds.
1610 * If fp is NULL, the file descriptor is returned to the pool.
1614 * MPSAFE (exclusive spinlock must be held on call)
1617 fsetfd_locked(struct filedesc
*fdp
, struct file
*fp
, int fd
)
1619 KKASSERT((unsigned)fd
< fdp
->fd_nfiles
);
1620 KKASSERT(fdp
->fd_files
[fd
].reserved
!= 0);
1623 fdp
->fd_files
[fd
].fp
= fp
;
1624 fdp
->fd_files
[fd
].reserved
= 0;
1626 fdp
->fd_files
[fd
].reserved
= 0;
1627 fdreserve_locked(fdp
, fd
, -1);
1628 fdfixup_locked(fdp
, fd
);
1636 fsetfd(struct filedesc
*fdp
, struct file
*fp
, int fd
)
1638 spin_lock(&fdp
->fd_spin
);
1639 fsetfd_locked(fdp
, fp
, fd
);
1640 spin_unlock(&fdp
->fd_spin
);
1644 * MPSAFE (exclusive spinlock must be held on call)
1648 funsetfd_locked(struct filedesc
*fdp
, int fd
)
1652 if ((unsigned)fd
>= fdp
->fd_nfiles
)
1654 if ((fp
= fdp
->fd_files
[fd
].fp
) == NULL
)
1656 fdp
->fd_files
[fd
].fp
= NULL
;
1657 fdp
->fd_files
[fd
].fileflags
= 0;
1659 fdreserve_locked(fdp
, fd
, -1);
1660 fdfixup_locked(fdp
, fd
);
1668 fgetfdflags(struct filedesc
*fdp
, int fd
, int *flagsp
)
1672 spin_lock(&fdp
->fd_spin
);
1673 if (((u_int
)fd
) >= fdp
->fd_nfiles
) {
1675 } else if (fdp
->fd_files
[fd
].fp
== NULL
) {
1678 *flagsp
= fdp
->fd_files
[fd
].fileflags
;
1681 spin_unlock(&fdp
->fd_spin
);
1689 fsetfdflags(struct filedesc
*fdp
, int fd
, int add_flags
)
1693 spin_lock(&fdp
->fd_spin
);
1694 if (((u_int
)fd
) >= fdp
->fd_nfiles
) {
1696 } else if (fdp
->fd_files
[fd
].fp
== NULL
) {
1699 fdp
->fd_files
[fd
].fileflags
|= add_flags
;
1702 spin_unlock(&fdp
->fd_spin
);
1710 fclrfdflags(struct filedesc
*fdp
, int fd
, int rem_flags
)
1714 spin_lock(&fdp
->fd_spin
);
1715 if (((u_int
)fd
) >= fdp
->fd_nfiles
) {
1717 } else if (fdp
->fd_files
[fd
].fp
== NULL
) {
1720 fdp
->fd_files
[fd
].fileflags
&= ~rem_flags
;
1723 spin_unlock(&fdp
->fd_spin
);
1728 * Set/Change/Clear the creds for a fp and synchronize the uidinfo.
1731 fsetcred(struct file
*fp
, struct ucred
*ncr
)
1734 struct uidinfo
*uip
;
1737 if (ocr
== NULL
|| ncr
== NULL
|| ocr
->cr_uidinfo
!= ncr
->cr_uidinfo
) {
1739 uip
= ocr
->cr_uidinfo
;
1740 atomic_add_int(&uip
->ui_openfiles
, -1);
1743 uip
= ncr
->cr_uidinfo
;
1744 atomic_add_int(&uip
->ui_openfiles
, 1);
1755 * Free a file descriptor.
1759 ffree(struct file
*fp
)
1761 KASSERT((fp
->f_count
== 0), ("ffree: fp_fcount not 0!"));
1762 spin_lock(&filehead_spin
);
1763 LIST_REMOVE(fp
, f_list
);
1765 spin_unlock(&filehead_spin
);
1767 if (fp
->f_nchandle
.ncp
)
1768 cache_drop(&fp
->f_nchandle
);
1773 * called from init_main, initialize filedesc0 for proc0.
1776 fdinit_bootstrap(struct proc
*p0
, struct filedesc
*fdp0
, int cmask
)
1780 fdp0
->fd_refcnt
= 1;
1781 fdp0
->fd_cmask
= cmask
;
1782 fdp0
->fd_files
= fdp0
->fd_builtin_files
;
1783 fdp0
->fd_nfiles
= NDFILE
;
1784 fdp0
->fd_lastfile
= -1;
1785 spin_init(&fdp0
->fd_spin
, "fdinitbootstrap");
1789 * Build a new filedesc structure.
1794 fdinit(struct proc
*p
)
1796 struct filedesc
*newfdp
;
1797 struct filedesc
*fdp
= p
->p_fd
;
1799 newfdp
= kmalloc(sizeof(struct filedesc
), M_FILEDESC
, M_WAITOK
|M_ZERO
);
1800 spin_lock(&fdp
->fd_spin
);
1802 newfdp
->fd_cdir
= fdp
->fd_cdir
;
1803 vref(newfdp
->fd_cdir
);
1804 cache_copy(&fdp
->fd_ncdir
, &newfdp
->fd_ncdir
);
1808 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of
1809 * proc0, but should unconditionally exist in other processes.
1812 newfdp
->fd_rdir
= fdp
->fd_rdir
;
1813 vref(newfdp
->fd_rdir
);
1814 cache_copy(&fdp
->fd_nrdir
, &newfdp
->fd_nrdir
);
1817 newfdp
->fd_jdir
= fdp
->fd_jdir
;
1818 vref(newfdp
->fd_jdir
);
1819 cache_copy(&fdp
->fd_njdir
, &newfdp
->fd_njdir
);
1821 spin_unlock(&fdp
->fd_spin
);
1823 /* Create the file descriptor table. */
1824 newfdp
->fd_refcnt
= 1;
1825 newfdp
->fd_cmask
= cmask
;
1826 newfdp
->fd_files
= newfdp
->fd_builtin_files
;
1827 newfdp
->fd_nfiles
= NDFILE
;
1828 newfdp
->fd_lastfile
= -1;
1829 spin_init(&newfdp
->fd_spin
, "fdinit");
1835 * Share a filedesc structure.
1840 fdshare(struct proc
*p
)
1842 struct filedesc
*fdp
;
1845 spin_lock(&fdp
->fd_spin
);
1847 spin_unlock(&fdp
->fd_spin
);
1852 * Copy a filedesc structure.
1857 fdcopy(struct proc
*p
, struct filedesc
**fpp
)
1859 struct filedesc
*fdp
= p
->p_fd
;
1860 struct filedesc
*newfdp
;
1861 struct fdnode
*fdnode
;
1866 * Certain daemons might not have file descriptors.
1872 * Allocate the new filedesc and fd_files[] array. This can race
1873 * with operations by other threads on the fdp so we have to be
1876 newfdp
= kmalloc(sizeof(struct filedesc
),
1877 M_FILEDESC
, M_WAITOK
| M_ZERO
| M_NULLOK
);
1878 if (newfdp
== NULL
) {
1883 spin_lock(&fdp
->fd_spin
);
1884 if (fdp
->fd_lastfile
< NDFILE
) {
1885 newfdp
->fd_files
= newfdp
->fd_builtin_files
;
1889 * We have to allocate (N^2-1) entries for our in-place
1890 * binary tree. Allow the table to shrink.
1894 while (ni
> fdp
->fd_lastfile
&& ni
> NDFILE
) {
1898 spin_unlock(&fdp
->fd_spin
);
1899 newfdp
->fd_files
= kmalloc(i
* sizeof(struct fdnode
),
1900 M_FILEDESC
, M_WAITOK
| M_ZERO
);
1903 * Check for race, retry
1905 spin_lock(&fdp
->fd_spin
);
1906 if (i
<= fdp
->fd_lastfile
) {
1907 spin_unlock(&fdp
->fd_spin
);
1908 kfree(newfdp
->fd_files
, M_FILEDESC
);
1914 * Dup the remaining fields. vref() and cache_hold() can be
1915 * safely called while holding the read spinlock on fdp.
1917 * The read spinlock on fdp is still being held.
1919 * NOTE: vref and cache_hold calls for the case where the vnode
1920 * or cache entry already has at least one ref may be called
1921 * while holding spin locks.
1923 if ((newfdp
->fd_cdir
= fdp
->fd_cdir
) != NULL
) {
1924 vref(newfdp
->fd_cdir
);
1925 cache_copy(&fdp
->fd_ncdir
, &newfdp
->fd_ncdir
);
1928 * We must check for fd_rdir here, at least for now because
1929 * the init process is created before we have access to the
1930 * rootvode to take a reference to it.
1932 if ((newfdp
->fd_rdir
= fdp
->fd_rdir
) != NULL
) {
1933 vref(newfdp
->fd_rdir
);
1934 cache_copy(&fdp
->fd_nrdir
, &newfdp
->fd_nrdir
);
1936 if ((newfdp
->fd_jdir
= fdp
->fd_jdir
) != NULL
) {
1937 vref(newfdp
->fd_jdir
);
1938 cache_copy(&fdp
->fd_njdir
, &newfdp
->fd_njdir
);
1940 newfdp
->fd_refcnt
= 1;
1941 newfdp
->fd_nfiles
= i
;
1942 newfdp
->fd_lastfile
= fdp
->fd_lastfile
;
1943 newfdp
->fd_freefile
= fdp
->fd_freefile
;
1944 newfdp
->fd_cmask
= fdp
->fd_cmask
;
1945 spin_init(&newfdp
->fd_spin
, "fdcopy");
1948 * Copy the descriptor table through (i). This also copies the
1949 * allocation state. Then go through and ref the file pointers
1950 * and clean up any KQ descriptors.
1952 * kq descriptors cannot be copied. Since we haven't ref'd the
1953 * copied files yet we can ignore the return value from funsetfd().
1955 * The read spinlock on fdp is still being held.
1957 bcopy(fdp
->fd_files
, newfdp
->fd_files
, i
* sizeof(struct fdnode
));
1958 for (i
= 0 ; i
< newfdp
->fd_nfiles
; ++i
) {
1959 fdnode
= &newfdp
->fd_files
[i
];
1960 if (fdnode
->reserved
) {
1961 fdreserve_locked(newfdp
, i
, -1);
1962 fdnode
->reserved
= 0;
1963 fdfixup_locked(newfdp
, i
);
1964 } else if (fdnode
->fp
) {
1965 if (fdnode
->fp
->f_type
== DTYPE_KQUEUE
) {
1966 (void)funsetfd_locked(newfdp
, i
);
1972 spin_unlock(&fdp
->fd_spin
);
1978 * Release a filedesc structure.
1980 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE)
1983 fdfree(struct proc
*p
, struct filedesc
*repl
)
1985 struct filedesc
*fdp
;
1986 struct fdnode
*fdnode
;
1988 struct filedesc_to_leader
*fdtol
;
1994 * Certain daemons might not have file descriptors.
2003 * Severe messing around to follow.
2005 spin_lock(&fdp
->fd_spin
);
2007 /* Check for special need to clear POSIX style locks */
2009 if (fdtol
!= NULL
) {
2010 KASSERT(fdtol
->fdl_refcount
> 0,
2011 ("filedesc_to_refcount botch: fdl_refcount=%d",
2012 fdtol
->fdl_refcount
));
2013 if (fdtol
->fdl_refcount
== 1 &&
2014 (p
->p_leader
->p_flags
& P_ADVLOCK
) != 0) {
2015 for (i
= 0; i
<= fdp
->fd_lastfile
; ++i
) {
2016 fdnode
= &fdp
->fd_files
[i
];
2017 if (fdnode
->fp
== NULL
||
2018 fdnode
->fp
->f_type
!= DTYPE_VNODE
) {
2023 spin_unlock(&fdp
->fd_spin
);
2025 lf
.l_whence
= SEEK_SET
;
2028 lf
.l_type
= F_UNLCK
;
2029 vp
= (struct vnode
*)fp
->f_data
;
2030 (void) VOP_ADVLOCK(vp
,
2031 (caddr_t
)p
->p_leader
,
2036 spin_lock(&fdp
->fd_spin
);
2040 if (fdtol
->fdl_refcount
== 1) {
2041 if (fdp
->fd_holdleaderscount
> 0 &&
2042 (p
->p_leader
->p_flags
& P_ADVLOCK
) != 0) {
2044 * close() or do_dup() has cleared a reference
2045 * in a shared file descriptor table.
2047 fdp
->fd_holdleaderswakeup
= 1;
2048 ssleep(&fdp
->fd_holdleaderscount
,
2049 &fdp
->fd_spin
, 0, "fdlhold", 0);
2052 if (fdtol
->fdl_holdcount
> 0) {
2054 * Ensure that fdtol->fdl_leader
2055 * remains valid in closef().
2057 fdtol
->fdl_wakeup
= 1;
2058 ssleep(fdtol
, &fdp
->fd_spin
, 0, "fdlhold", 0);
2062 fdtol
->fdl_refcount
--;
2063 if (fdtol
->fdl_refcount
== 0 &&
2064 fdtol
->fdl_holdcount
== 0) {
2065 fdtol
->fdl_next
->fdl_prev
= fdtol
->fdl_prev
;
2066 fdtol
->fdl_prev
->fdl_next
= fdtol
->fdl_next
;
2071 if (fdtol
!= NULL
) {
2072 spin_unlock(&fdp
->fd_spin
);
2073 kfree(fdtol
, M_FILEDESC_TO_LEADER
);
2074 spin_lock(&fdp
->fd_spin
);
2077 if (--fdp
->fd_refcnt
> 0) {
2078 spin_unlock(&fdp
->fd_spin
);
2079 spin_lock(&p
->p_spin
);
2081 spin_unlock(&p
->p_spin
);
2086 * Even though we are the last reference to the structure allproc
2087 * scans may still reference the structure. Maintain proper
2088 * locks until we can replace p->p_fd.
2090 * Also note that kqueue's closef still needs to reference the
2091 * fdp via p->p_fd, so we have to close the descriptors before
2092 * we replace p->p_fd.
2094 for (i
= 0; i
<= fdp
->fd_lastfile
; ++i
) {
2095 if (fdp
->fd_files
[i
].fp
) {
2096 fp
= funsetfd_locked(fdp
, i
);
2098 spin_unlock(&fdp
->fd_spin
);
2099 if (SLIST_FIRST(&fp
->f_klist
))
2100 knote_fdclose(fp
, fdp
, i
);
2102 spin_lock(&fdp
->fd_spin
);
2106 spin_unlock(&fdp
->fd_spin
);
2109 * Interlock against an allproc scan operations (typically frevoke).
2111 spin_lock(&p
->p_spin
);
2113 spin_unlock(&p
->p_spin
);
2116 * Wait for any softrefs to go away. This race rarely occurs so
2117 * we can use a non-critical-path style poll/sleep loop. The
2118 * race only occurs against allproc scans.
2120 * No new softrefs can occur with the fdp disconnected from the
2123 if (fdp
->fd_softrefs
) {
2124 kprintf("pid %d: Warning, fdp race avoided\n", p
->p_pid
);
2125 while (fdp
->fd_softrefs
)
2126 tsleep(&fdp
->fd_softrefs
, 0, "fdsoft", 1);
2129 if (fdp
->fd_files
!= fdp
->fd_builtin_files
)
2130 kfree(fdp
->fd_files
, M_FILEDESC
);
2132 cache_drop(&fdp
->fd_ncdir
);
2133 vrele(fdp
->fd_cdir
);
2136 cache_drop(&fdp
->fd_nrdir
);
2137 vrele(fdp
->fd_rdir
);
2140 cache_drop(&fdp
->fd_njdir
);
2141 vrele(fdp
->fd_jdir
);
2143 kfree(fdp
, M_FILEDESC
);
2147 * Retrieve and reference the file pointer associated with a descriptor.
2152 holdfp(struct filedesc
*fdp
, int fd
, int flag
)
2156 spin_lock_shared(&fdp
->fd_spin
);
2157 if (((u_int
)fd
) >= fdp
->fd_nfiles
) {
2161 if ((fp
= fdp
->fd_files
[fd
].fp
) == NULL
)
2163 if ((fp
->f_flag
& flag
) == 0 && flag
!= -1) {
2169 spin_unlock_shared(&fdp
->fd_spin
);
2174 * holdsock() - load the struct file pointer associated
2175 * with a socket into *fpp. If an error occurs, non-zero
2176 * will be returned and *fpp will be set to NULL.
2181 holdsock(struct filedesc
*fdp
, int fd
, struct file
**fpp
)
2186 spin_lock_shared(&fdp
->fd_spin
);
2187 if ((unsigned)fd
>= fdp
->fd_nfiles
) {
2192 if ((fp
= fdp
->fd_files
[fd
].fp
) == NULL
) {
2196 if (fp
->f_type
!= DTYPE_SOCKET
) {
2203 spin_unlock_shared(&fdp
->fd_spin
);
2209 * Convert a user file descriptor to a held file pointer.
2214 holdvnode(struct filedesc
*fdp
, int fd
, struct file
**fpp
)
2219 spin_lock_shared(&fdp
->fd_spin
);
2220 if ((unsigned)fd
>= fdp
->fd_nfiles
) {
2225 if ((fp
= fdp
->fd_files
[fd
].fp
) == NULL
) {
2229 if (fp
->f_type
!= DTYPE_VNODE
&& fp
->f_type
!= DTYPE_FIFO
) {
2237 spin_unlock_shared(&fdp
->fd_spin
);
2243 * For setugid programs, we don't want to people to use that setugidness
2244 * to generate error messages which write to a file which otherwise would
2245 * otherwise be off-limits to the process.
2247 * This is a gross hack to plug the hole. A better solution would involve
2248 * a special vop or other form of generalized access control mechanism. We
2249 * go ahead and just reject all procfs file systems accesses as dangerous.
2251 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2252 * sufficient. We also don't for check setugidness since we know we are.
2255 is_unsafe(struct file
*fp
)
2257 if (fp
->f_type
== DTYPE_VNODE
&&
2258 ((struct vnode
*)(fp
->f_data
))->v_tag
== VT_PROCFS
)
2264 * Make this setguid thing safe, if at all possible.
2266 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2269 setugidsafety(struct proc
*p
)
2271 struct filedesc
*fdp
= p
->p_fd
;
2274 /* Certain daemons might not have file descriptors. */
2279 * note: fdp->fd_files may be reallocated out from under us while
2280 * we are blocked in a close. Be careful!
2282 for (i
= 0; i
<= fdp
->fd_lastfile
; i
++) {
2285 if (fdp
->fd_files
[i
].fp
&& is_unsafe(fdp
->fd_files
[i
].fp
)) {
2289 * NULL-out descriptor prior to close to avoid
2290 * a race while close blocks.
2292 if ((fp
= funsetfd_locked(fdp
, i
)) != NULL
) {
2293 knote_fdclose(fp
, fdp
, i
);
2301 * Close any files on exec?
2303 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2306 fdcloseexec(struct proc
*p
)
2308 struct filedesc
*fdp
= p
->p_fd
;
2311 /* Certain daemons might not have file descriptors. */
2316 * We cannot cache fd_files since operations may block and rip
2317 * them out from under us.
2319 for (i
= 0; i
<= fdp
->fd_lastfile
; i
++) {
2320 if (fdp
->fd_files
[i
].fp
!= NULL
&&
2321 (fdp
->fd_files
[i
].fileflags
& UF_EXCLOSE
)) {
2325 * NULL-out descriptor prior to close to avoid
2326 * a race while close blocks.
2328 if ((fp
= funsetfd_locked(fdp
, i
)) != NULL
) {
2329 knote_fdclose(fp
, fdp
, i
);
2337 * It is unsafe for set[ug]id processes to be started with file
2338 * descriptors 0..2 closed, as these descriptors are given implicit
2339 * significance in the Standard C library. fdcheckstd() will create a
2340 * descriptor referencing /dev/null for each of stdin, stdout, and
2341 * stderr that is not already open.
2343 * NOT MPSAFE - calls falloc, vn_open, etc
2346 fdcheckstd(struct lwp
*lp
)
2348 struct nlookupdata nd
;
2349 struct filedesc
*fdp
;
2352 int i
, error
, flags
, devnull
;
2354 fdp
= lp
->lwp_proc
->p_fd
;
2359 for (i
= 0; i
< 3; i
++) {
2360 if (fdp
->fd_files
[i
].fp
!= NULL
)
2363 if ((error
= falloc(lp
, &fp
, &devnull
)) != 0)
2366 error
= nlookup_init(&nd
, "/dev/null", UIO_SYSSPACE
,
2367 NLC_FOLLOW
|NLC_LOCKVP
);
2368 flags
= FREAD
| FWRITE
;
2370 error
= vn_open(&nd
, fp
, flags
, 0);
2372 fsetfd(fdp
, fp
, devnull
);
2374 fsetfd(fdp
, NULL
, devnull
);
2379 KKASSERT(i
== devnull
);
2381 error
= kern_dup(DUP_FIXED
, devnull
, i
, &retval
);
2390 * Internal form of close.
2391 * Decrement reference count on file structure.
2392 * Note: td and/or p may be NULL when closing a file
2393 * that was being passed in a message.
2395 * MPALMOSTSAFE - acquires mplock for VOP operations
2398 closef(struct file
*fp
, struct proc
*p
)
2402 struct filedesc_to_leader
*fdtol
;
2408 * POSIX record locking dictates that any close releases ALL
2409 * locks owned by this process. This is handled by setting
2410 * a flag in the unlock to free ONLY locks obeying POSIX
2411 * semantics, and not to free BSD-style file locks.
2412 * If the descriptor was in a message, POSIX-style locks
2413 * aren't passed with the descriptor.
2415 if (p
!= NULL
&& fp
->f_type
== DTYPE_VNODE
&&
2416 (((struct vnode
*)fp
->f_data
)->v_flag
& VMAYHAVELOCKS
)
2418 if ((p
->p_leader
->p_flags
& P_ADVLOCK
) != 0) {
2419 lf
.l_whence
= SEEK_SET
;
2422 lf
.l_type
= F_UNLCK
;
2423 vp
= (struct vnode
*)fp
->f_data
;
2424 (void) VOP_ADVLOCK(vp
, (caddr_t
)p
->p_leader
, F_UNLCK
,
2428 if (fdtol
!= NULL
) {
2429 lwkt_gettoken(&p
->p_token
);
2431 * Handle special case where file descriptor table
2432 * is shared between multiple process leaders.
2434 for (fdtol
= fdtol
->fdl_next
;
2435 fdtol
!= p
->p_fdtol
;
2436 fdtol
= fdtol
->fdl_next
) {
2437 if ((fdtol
->fdl_leader
->p_flags
&
2440 fdtol
->fdl_holdcount
++;
2441 lf
.l_whence
= SEEK_SET
;
2444 lf
.l_type
= F_UNLCK
;
2445 vp
= (struct vnode
*)fp
->f_data
;
2446 (void) VOP_ADVLOCK(vp
,
2447 (caddr_t
)fdtol
->fdl_leader
,
2448 F_UNLCK
, &lf
, F_POSIX
);
2449 fdtol
->fdl_holdcount
--;
2450 if (fdtol
->fdl_holdcount
== 0 &&
2451 fdtol
->fdl_wakeup
!= 0) {
2452 fdtol
->fdl_wakeup
= 0;
2456 lwkt_reltoken(&p
->p_token
);
2465 * fhold() can only be called if f_count is already at least 1 (i.e. the
2466 * caller of fhold() already has a reference to the file pointer in some
2469 * f_count is not spin-locked. Instead, atomic ops are used for
2470 * incrementing, decrementing, and handling the 1->0 transition.
2473 fhold(struct file
*fp
)
2475 atomic_add_int(&fp
->f_count
, 1);
2479 * fdrop() - drop a reference to a descriptor
2481 * MPALMOSTSAFE - acquires mplock for final close sequence
2484 fdrop(struct file
*fp
)
2491 * A combined fetch and subtract is needed to properly detect
2492 * 1->0 transitions, otherwise two cpus dropping from a ref
2493 * count of 2 might both try to run the 1->0 code.
2495 if (atomic_fetchadd_int(&fp
->f_count
, -1) > 1)
2498 KKASSERT(SLIST_FIRST(&fp
->f_klist
) == NULL
);
2501 * The last reference has gone away, we own the fp structure free
2504 if (fp
->f_count
< 0)
2505 panic("fdrop: count < 0");
2506 if ((fp
->f_flag
& FHASLOCK
) && fp
->f_type
== DTYPE_VNODE
&&
2507 (((struct vnode
*)fp
->f_data
)->v_flag
& VMAYHAVELOCKS
)
2509 lf
.l_whence
= SEEK_SET
;
2512 lf
.l_type
= F_UNLCK
;
2513 vp
= (struct vnode
*)fp
->f_data
;
2514 (void) VOP_ADVLOCK(vp
, (caddr_t
)fp
, F_UNLCK
, &lf
, 0);
2516 if (fp
->f_ops
!= &badfileops
)
2517 error
= fo_close(fp
);
2525 * Apply an advisory lock on a file descriptor.
2527 * Just attempt to get a record lock of the requested type on
2528 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2533 sys_flock(struct flock_args
*uap
)
2535 struct proc
*p
= curproc
;
2541 if ((fp
= holdfp(p
->p_fd
, uap
->fd
, -1)) == NULL
)
2543 if (fp
->f_type
!= DTYPE_VNODE
) {
2547 vp
= (struct vnode
*)fp
->f_data
;
2548 lf
.l_whence
= SEEK_SET
;
2551 if (uap
->how
& LOCK_UN
) {
2552 lf
.l_type
= F_UNLCK
;
2553 fp
->f_flag
&= ~FHASLOCK
;
2554 error
= VOP_ADVLOCK(vp
, (caddr_t
)fp
, F_UNLCK
, &lf
, 0);
2557 if (uap
->how
& LOCK_EX
)
2558 lf
.l_type
= F_WRLCK
;
2559 else if (uap
->how
& LOCK_SH
)
2560 lf
.l_type
= F_RDLCK
;
2565 fp
->f_flag
|= FHASLOCK
;
2566 if (uap
->how
& LOCK_NB
)
2567 error
= VOP_ADVLOCK(vp
, (caddr_t
)fp
, F_SETLK
, &lf
, 0);
2569 error
= VOP_ADVLOCK(vp
, (caddr_t
)fp
, F_SETLK
, &lf
, F_WAIT
);
2576 * File Descriptor pseudo-device driver (/dev/fd/).
2578 * Opening minor device N dup()s the file (if any) connected to file
2579 * descriptor N belonging to the calling process. Note that this driver
2580 * consists of only the ``open()'' routine, because all subsequent
2581 * references to this file will be direct to the other driver.
2584 fdopen(struct dev_open_args
*ap
)
2586 thread_t td
= curthread
;
2588 KKASSERT(td
->td_lwp
!= NULL
);
2591 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the
2592 * the file descriptor being sought for duplication. The error
2593 * return ensures that the vnode for this device will be released
2594 * by vn_open. Open will detect this special error and take the
2595 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2596 * will simply report the error.
2598 td
->td_lwp
->lwp_dupfd
= minor(ap
->a_head
.a_dev
);
2603 * The caller has reserved the file descriptor dfd for us. On success we
2604 * must fsetfd() it. On failure the caller will clean it up.
2609 dupfdopen(struct filedesc
*fdp
, int dfd
, int sfd
, int mode
, int error
)
2615 if ((wfp
= holdfp(fdp
, sfd
, -1)) == NULL
)
2619 * Close a revoke/dup race. Duping a descriptor marked as revoked
2620 * will dup a dummy descriptor instead of the real one.
2622 if (wfp
->f_flag
& FREVOKED
) {
2623 kprintf("Warning: attempt to dup() a revoked descriptor\n");
2626 werror
= falloc(NULL
, &wfp
, NULL
);
2632 * There are two cases of interest here.
2634 * For ENODEV simply dup sfd to file descriptor dfd and return.
2636 * For ENXIO steal away the file structure from sfd and store it
2637 * dfd. sfd is effectively closed by this operation.
2639 * Any other error code is just returned.
2644 * Check that the mode the file is being opened for is a
2645 * subset of the mode of the existing descriptor.
2647 if (((mode
& (FREAD
|FWRITE
)) | wfp
->f_flag
) != wfp
->f_flag
) {
2651 spin_lock(&fdp
->fd_spin
);
2652 fdp
->fd_files
[dfd
].fileflags
= fdp
->fd_files
[sfd
].fileflags
;
2653 fsetfd_locked(fdp
, wfp
, dfd
);
2654 spin_unlock(&fdp
->fd_spin
);
2659 * Steal away the file pointer from dfd, and stuff it into indx.
2661 spin_lock(&fdp
->fd_spin
);
2662 fdp
->fd_files
[dfd
].fileflags
= fdp
->fd_files
[sfd
].fileflags
;
2663 fsetfd(fdp
, wfp
, dfd
);
2664 if ((xfp
= funsetfd_locked(fdp
, sfd
)) != NULL
) {
2665 spin_unlock(&fdp
->fd_spin
);
2668 spin_unlock(&fdp
->fd_spin
);
2680 * NOT MPSAFE - I think these refer to a common file descriptor table
2681 * and we need to spinlock that to link fdtol in.
2683 struct filedesc_to_leader
*
2684 filedesc_to_leader_alloc(struct filedesc_to_leader
*old
,
2685 struct proc
*leader
)
2687 struct filedesc_to_leader
*fdtol
;
2689 fdtol
= kmalloc(sizeof(struct filedesc_to_leader
),
2690 M_FILEDESC_TO_LEADER
, M_WAITOK
| M_ZERO
);
2691 fdtol
->fdl_refcount
= 1;
2692 fdtol
->fdl_holdcount
= 0;
2693 fdtol
->fdl_wakeup
= 0;
2694 fdtol
->fdl_leader
= leader
;
2696 fdtol
->fdl_next
= old
->fdl_next
;
2697 fdtol
->fdl_prev
= old
;
2698 old
->fdl_next
= fdtol
;
2699 fdtol
->fdl_next
->fdl_prev
= fdtol
;
2701 fdtol
->fdl_next
= fdtol
;
2702 fdtol
->fdl_prev
= fdtol
;
2708 * Scan all file pointers in the system. The callback is made with
2709 * the master list spinlock held exclusively.
2714 allfiles_scan_exclusive(int (*callback
)(struct file
*, void *), void *data
)
2719 spin_lock(&filehead_spin
);
2720 LIST_FOREACH(fp
, &filehead
, f_list
) {
2721 res
= callback(fp
, data
);
2725 spin_unlock(&filehead_spin
);
2729 * Get file structures.
2731 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe)
2734 struct sysctl_kern_file_info
{
2737 struct sysctl_req
*req
;
2740 static int sysctl_kern_file_callback(struct proc
*p
, void *data
);
2743 sysctl_kern_file(SYSCTL_HANDLER_ARGS
)
2745 struct sysctl_kern_file_info info
;
2748 * Note: because the number of file descriptors is calculated
2749 * in different ways for sizing vs returning the data,
2750 * there is information leakage from the first loop. However,
2751 * it is of a similar order of magnitude to the leakage from
2752 * global system statistics such as kern.openfiles.
2754 * When just doing a count, note that we cannot just count
2755 * the elements and add f_count via the filehead list because
2756 * threaded processes share their descriptor table and f_count might
2757 * still be '1' in that case.
2759 * Since the SYSCTL op can block, we must hold the process to
2760 * prevent it being ripped out from under us either in the
2761 * file descriptor loop or in the greater LIST_FOREACH. The
2762 * process may be in varying states of disrepair. If the process
2763 * is in SZOMB we may have caught it just as it is being removed
2764 * from the allproc list, we must skip it in that case to maintain
2765 * an unbroken chain through the allproc list.
2770 allproc_scan(sysctl_kern_file_callback
, &info
);
2773 * When just calculating the size, overestimate a bit to try to
2774 * prevent system activity from causing the buffer-fill call
2777 if (req
->oldptr
== NULL
) {
2778 info
.count
= (info
.count
+ 16) + (info
.count
/ 10);
2779 info
.error
= SYSCTL_OUT(req
, NULL
,
2780 info
.count
* sizeof(struct kinfo_file
));
2782 return (info
.error
);
2786 sysctl_kern_file_callback(struct proc
*p
, void *data
)
2788 struct sysctl_kern_file_info
*info
= data
;
2789 struct kinfo_file kf
;
2790 struct filedesc
*fdp
;
2795 if (p
->p_stat
== SIDL
|| p
->p_stat
== SZOMB
)
2797 if (!PRISON_CHECK(info
->req
->td
->td_ucred
, p
->p_ucred
) != 0)
2801 * Softref the fdp to prevent it from being destroyed
2803 spin_lock(&p
->p_spin
);
2804 if ((fdp
= p
->p_fd
) == NULL
) {
2805 spin_unlock(&p
->p_spin
);
2808 atomic_add_int(&fdp
->fd_softrefs
, 1);
2809 spin_unlock(&p
->p_spin
);
2812 * The fdp's own spinlock prevents the contents from being
2815 spin_lock_shared(&fdp
->fd_spin
);
2816 for (n
= 0; n
< fdp
->fd_nfiles
; ++n
) {
2817 if ((fp
= fdp
->fd_files
[n
].fp
) == NULL
)
2819 if (info
->req
->oldptr
== NULL
) {
2822 uid
= p
->p_ucred
? p
->p_ucred
->cr_uid
: -1;
2823 kcore_make_file(&kf
, fp
, p
->p_pid
, uid
, n
);
2824 spin_unlock_shared(&fdp
->fd_spin
);
2825 info
->error
= SYSCTL_OUT(info
->req
, &kf
, sizeof(kf
));
2826 spin_lock_shared(&fdp
->fd_spin
);
2831 spin_unlock_shared(&fdp
->fd_spin
);
2832 atomic_subtract_int(&fdp
->fd_softrefs
, 1);
2838 SYSCTL_PROC(_kern
, KERN_FILE
, file
, CTLTYPE_OPAQUE
|CTLFLAG_RD
,
2839 0, 0, sysctl_kern_file
, "S,file", "Entire file table");
2841 SYSCTL_INT(_kern
, OID_AUTO
, minfilesperproc
, CTLFLAG_RW
,
2842 &minfilesperproc
, 0, "Minimum files allowed open per process");
2843 SYSCTL_INT(_kern
, KERN_MAXFILESPERPROC
, maxfilesperproc
, CTLFLAG_RW
,
2844 &maxfilesperproc
, 0, "Maximum files allowed open per process");
2845 SYSCTL_INT(_kern
, OID_AUTO
, maxfilesperuser
, CTLFLAG_RW
,
2846 &maxfilesperuser
, 0, "Maximum files allowed open per user");
2848 SYSCTL_INT(_kern
, KERN_MAXFILES
, maxfiles
, CTLFLAG_RW
,
2849 &maxfiles
, 0, "Maximum number of files");
2851 SYSCTL_INT(_kern
, OID_AUTO
, maxfilesrootres
, CTLFLAG_RW
,
2852 &maxfilesrootres
, 0, "Descriptors reserved for root use");
2854 SYSCTL_INT(_kern
, OID_AUTO
, openfiles
, CTLFLAG_RD
,
2855 &nfiles
, 0, "System-wide number of open files");
2858 fildesc_drvinit(void *unused
)
2862 for (fd
= 0; fd
< NUMFDESC
; fd
++) {
2863 make_dev(&fildesc_ops
, fd
,
2864 UID_BIN
, GID_BIN
, 0666, "fd/%d", fd
);
2867 make_dev(&fildesc_ops
, 0, UID_ROOT
, GID_WHEEL
, 0666, "stdin");
2868 make_dev(&fildesc_ops
, 1, UID_ROOT
, GID_WHEEL
, 0666, "stdout");
2869 make_dev(&fildesc_ops
, 2, UID_ROOT
, GID_WHEEL
, 0666, "stderr");
2875 struct fileops badfileops
= {
2876 .fo_read
= badfo_readwrite
,
2877 .fo_write
= badfo_readwrite
,
2878 .fo_ioctl
= badfo_ioctl
,
2879 .fo_kqfilter
= badfo_kqfilter
,
2880 .fo_stat
= badfo_stat
,
2881 .fo_close
= badfo_close
,
2882 .fo_shutdown
= badfo_shutdown
2896 badfo_ioctl(struct file
*fp
, u_long com
, caddr_t data
,
2897 struct ucred
*cred
, struct sysmsg
*msgv
)
2903 * Must return an error to prevent registration, typically
2904 * due to a revoked descriptor (file_filtops assigned).
2907 badfo_kqfilter(struct file
*fp
, struct knote
*kn
)
2909 return (EOPNOTSUPP
);
2916 badfo_stat(struct file
*fp
, struct stat
*sb
, struct ucred
*cred
)
2925 badfo_close(struct file
*fp
)
2934 badfo_shutdown(struct file
*fp
, int how
)
2943 nofo_shutdown(struct file
*fp
, int how
)
2945 return (EOPNOTSUPP
);
2948 SYSINIT(fildescdev
,SI_SUB_DRIVERS
,SI_ORDER_MIDDLE
+CDEV_MAJOR
,
2949 fildesc_drvinit
,NULL
)