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 * 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.
157 sys_getdtablesize(struct getdtablesize_args
*uap
)
159 struct proc
*p
= curproc
;
160 struct plimit
*limit
= p
->p_limit
;
163 spin_lock(&limit
->p_spin
);
164 if (limit
->pl_rlimit
[RLIMIT_NOFILE
].rlim_cur
> INT_MAX
)
167 dtsize
= (int)limit
->pl_rlimit
[RLIMIT_NOFILE
].rlim_cur
;
168 spin_unlock(&limit
->p_spin
);
170 if (dtsize
> maxfilesperproc
)
171 dtsize
= maxfilesperproc
;
172 if (dtsize
< minfilesperproc
)
173 dtsize
= minfilesperproc
;
174 if (p
->p_ucred
->cr_uid
&& dtsize
> maxfilesperuser
)
175 dtsize
= maxfilesperuser
;
176 uap
->sysmsg_result
= dtsize
;
181 * Duplicate a file descriptor to a particular value.
183 * note: keep in mind that a potential race condition exists when closing
184 * descriptors from a shared descriptor table (via rfork).
187 sys_dup2(struct dup2_args
*uap
)
192 error
= kern_dup(DUP_FIXED
, uap
->from
, uap
->to
, &fd
);
193 uap
->sysmsg_fds
[0] = fd
;
199 * Duplicate a file descriptor.
202 sys_dup(struct dup_args
*uap
)
207 error
= kern_dup(DUP_VARIABLE
, uap
->fd
, 0, &fd
);
208 uap
->sysmsg_fds
[0] = fd
;
214 * MPALMOSTSAFE - acquires mplock for fp operations
217 kern_fcntl(int fd
, int cmd
, union fcntl_dat
*dat
, struct ucred
*cred
)
219 struct thread
*td
= curthread
;
220 struct proc
*p
= td
->td_proc
;
226 int tmp
, error
, flg
= F_POSIX
;
231 * Operations on file descriptors that do not require a file pointer.
235 error
= fgetfdflags(p
->p_fd
, fd
, &tmp
);
237 dat
->fc_cloexec
= (tmp
& UF_EXCLOSE
) ? FD_CLOEXEC
: 0;
241 if (dat
->fc_cloexec
& FD_CLOEXEC
)
242 error
= fsetfdflags(p
->p_fd
, fd
, UF_EXCLOSE
);
244 error
= fclrfdflags(p
->p_fd
, fd
, UF_EXCLOSE
);
248 error
= kern_dup(DUP_VARIABLE
| DUP_FCNTL
, fd
, newmin
,
251 case F_DUPFD_CLOEXEC
:
253 error
= kern_dup(DUP_VARIABLE
| DUP_CLOEXEC
| DUP_FCNTL
,
254 fd
, newmin
, &dat
->fc_fd
);
258 error
= kern_dup(DUP_FIXED
, fd
, newmin
, &dat
->fc_fd
);
260 case F_DUP2FD_CLOEXEC
:
262 error
= kern_dup(DUP_FIXED
| DUP_CLOEXEC
, fd
, newmin
,
270 * Operations on file pointers
272 if ((fp
= holdfp(p
->p_fd
, fd
, -1)) == NULL
)
277 dat
->fc_flags
= OFLAGS(fp
->f_flag
);
283 nflags
= FFLAGS(dat
->fc_flags
& ~O_ACCMODE
) & FCNTLFLAGS
;
284 nflags
|= oflags
& ~FCNTLFLAGS
;
287 if (((nflags
^ oflags
) & O_APPEND
) && (oflags
& FAPPENDONLY
))
289 if (error
== 0 && ((nflags
^ oflags
) & FASYNC
)) {
290 tmp
= nflags
& FASYNC
;
291 error
= fo_ioctl(fp
, FIOASYNC
, (caddr_t
)&tmp
,
296 * If no error, must be atomically set.
301 nflags
= (oflags
& ~FCNTLFLAGS
) | (nflags
& FCNTLFLAGS
);
302 if (atomic_cmpset_int(&fp
->f_flag
, oflags
, nflags
))
309 error
= fo_ioctl(fp
, FIOGETOWN
, (caddr_t
)&dat
->fc_owner
,
314 error
= fo_ioctl(fp
, FIOSETOWN
, (caddr_t
)&dat
->fc_owner
,
320 /* Fall into F_SETLK */
323 if (fp
->f_type
!= DTYPE_VNODE
) {
327 vp
= (struct vnode
*)fp
->f_data
;
330 * copyin/lockop may block
332 if (dat
->fc_flock
.l_whence
== SEEK_CUR
)
333 dat
->fc_flock
.l_start
+= fp
->f_offset
;
335 switch (dat
->fc_flock
.l_type
) {
337 if ((fp
->f_flag
& FREAD
) == 0) {
341 if ((p
->p_leader
->p_flags
& P_ADVLOCK
) == 0) {
342 lwkt_gettoken(&p
->p_leader
->p_token
);
343 p
->p_leader
->p_flags
|= P_ADVLOCK
;
344 lwkt_reltoken(&p
->p_leader
->p_token
);
346 error
= VOP_ADVLOCK(vp
, (caddr_t
)p
->p_leader
, F_SETLK
,
347 &dat
->fc_flock
, flg
);
350 if ((fp
->f_flag
& FWRITE
) == 0) {
354 if ((p
->p_leader
->p_flags
& P_ADVLOCK
) == 0) {
355 lwkt_gettoken(&p
->p_leader
->p_token
);
356 p
->p_leader
->p_flags
|= P_ADVLOCK
;
357 lwkt_reltoken(&p
->p_leader
->p_token
);
359 error
= VOP_ADVLOCK(vp
, (caddr_t
)p
->p_leader
, F_SETLK
,
360 &dat
->fc_flock
, flg
);
363 error
= VOP_ADVLOCK(vp
, (caddr_t
)p
->p_leader
, F_UNLCK
,
364 &dat
->fc_flock
, F_POSIX
);
372 * It is possible to race a close() on the descriptor while
373 * we were blocked getting the lock. If this occurs the
374 * close might not have caught the lock.
376 if (checkfdclosed(p
->p_fd
, fd
, fp
)) {
377 dat
->fc_flock
.l_whence
= SEEK_SET
;
378 dat
->fc_flock
.l_start
= 0;
379 dat
->fc_flock
.l_len
= 0;
380 dat
->fc_flock
.l_type
= F_UNLCK
;
381 (void) VOP_ADVLOCK(vp
, (caddr_t
)p
->p_leader
,
382 F_UNLCK
, &dat
->fc_flock
, F_POSIX
);
387 if (fp
->f_type
!= DTYPE_VNODE
) {
391 vp
= (struct vnode
*)fp
->f_data
;
393 * copyin/lockop may block
395 if (dat
->fc_flock
.l_type
!= F_RDLCK
&&
396 dat
->fc_flock
.l_type
!= F_WRLCK
&&
397 dat
->fc_flock
.l_type
!= F_UNLCK
) {
401 if (dat
->fc_flock
.l_whence
== SEEK_CUR
)
402 dat
->fc_flock
.l_start
+= fp
->f_offset
;
403 error
= VOP_ADVLOCK(vp
, (caddr_t
)p
->p_leader
, F_GETLK
,
404 &dat
->fc_flock
, F_POSIX
);
416 * The file control system call.
419 sys_fcntl(struct fcntl_args
*uap
)
427 case F_DUPFD_CLOEXEC
:
428 case F_DUP2FD_CLOEXEC
:
429 dat
.fc_fd
= uap
->arg
;
432 dat
.fc_cloexec
= uap
->arg
;
435 dat
.fc_flags
= uap
->arg
;
438 dat
.fc_owner
= uap
->arg
;
443 error
= copyin((caddr_t
)uap
->arg
, &dat
.fc_flock
,
444 sizeof(struct flock
));
450 error
= kern_fcntl(uap
->fd
, uap
->cmd
, &dat
, curthread
->td_ucred
);
456 case F_DUPFD_CLOEXEC
:
457 case F_DUP2FD_CLOEXEC
:
458 uap
->sysmsg_result
= dat
.fc_fd
;
461 uap
->sysmsg_result
= dat
.fc_cloexec
;
464 uap
->sysmsg_result
= dat
.fc_flags
;
467 uap
->sysmsg_result
= dat
.fc_owner
;
470 error
= copyout(&dat
.fc_flock
, (caddr_t
)uap
->arg
,
471 sizeof(struct flock
));
480 * Common code for dup, dup2, and fcntl(F_DUPFD).
482 * There are four type flags: DUP_FCNTL, DUP_FIXED, DUP_VARIABLE, and
485 * DUP_FCNTL is for handling EINVAL vs. EBADF differences between
486 * fcntl()'s F_DUPFD and F_DUPFD_CLOEXEC and dup2() (per POSIX).
487 * The next two flags are mutually exclusive, and the fourth is optional.
488 * DUP_FIXED tells kern_dup() to destructively dup over an existing file
489 * descriptor if "new" is already open. DUP_VARIABLE tells kern_dup()
490 * to find the lowest unused file descriptor that is greater than or
491 * equal to "new". DUP_CLOEXEC, which works with either of the first
492 * 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
;
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
686 funsetown(struct sigio
**sigiop
)
692 if ((sigio
= *sigiop
) != NULL
) {
693 lwkt_gettoken(&sigio_token
); /* protect sigio */
694 KKASSERT(sigiop
== sigio
->sio_myref
);
697 lwkt_reltoken(&sigio_token
);
702 if (sigio
->sio_pgid
< 0) {
703 pgrp
= sigio
->sio_pgrp
;
704 sigio
->sio_pgrp
= NULL
;
705 lwkt_gettoken(&pgrp
->pg_token
);
706 SLIST_REMOVE(&pgrp
->pg_sigiolst
, sigio
, sigio
, sio_pgsigio
);
707 lwkt_reltoken(&pgrp
->pg_token
);
709 } else /* if ((*sigiop)->sio_pgid > 0) */ {
711 sigio
->sio_proc
= NULL
;
713 lwkt_gettoken(&p
->p_token
);
714 SLIST_REMOVE(&p
->p_sigiolst
, sigio
, sigio
, sio_pgsigio
);
715 lwkt_reltoken(&p
->p_token
);
718 crfree(sigio
->sio_ucred
);
719 sigio
->sio_ucred
= NULL
;
720 kfree(sigio
, M_SIGIO
);
724 * Free a list of sigio structures. Caller is responsible for ensuring
725 * that the list is MPSAFE.
728 funsetownlst(struct sigiolst
*sigiolst
)
732 while ((sigio
= SLIST_FIRST(sigiolst
)) != NULL
)
733 funsetown(sigio
->sio_myref
);
737 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
739 * After permission checking, add a sigio structure to the sigio list for
740 * the process or process group.
743 fsetown(pid_t pgid
, struct sigio
**sigiop
)
745 struct proc
*proc
= NULL
;
746 struct pgrp
*pgrp
= NULL
;
763 * Policy - Don't allow a process to FSETOWN a process
764 * in another session.
766 * Remove this test to allow maximum flexibility or
767 * restrict FSETOWN to the current process or process
768 * group for maximum safety.
770 if (proc
->p_session
!= curproc
->p_session
) {
774 } else /* if (pgid < 0) */ {
775 pgrp
= pgfind(-pgid
);
782 * Policy - Don't allow a process to FSETOWN a process
783 * in another session.
785 * Remove this test to allow maximum flexibility or
786 * restrict FSETOWN to the current process or process
787 * group for maximum safety.
789 if (pgrp
->pg_session
!= curproc
->p_session
) {
794 sigio
= kmalloc(sizeof(struct sigio
), M_SIGIO
, M_WAITOK
| M_ZERO
);
796 KKASSERT(pgrp
== NULL
);
797 lwkt_gettoken(&proc
->p_token
);
798 SLIST_INSERT_HEAD(&proc
->p_sigiolst
, sigio
, sio_pgsigio
);
799 sigio
->sio_proc
= proc
;
800 lwkt_reltoken(&proc
->p_token
);
802 KKASSERT(proc
== NULL
);
803 lwkt_gettoken(&pgrp
->pg_token
);
804 SLIST_INSERT_HEAD(&pgrp
->pg_sigiolst
, sigio
, sio_pgsigio
);
805 sigio
->sio_pgrp
= pgrp
;
806 lwkt_reltoken(&pgrp
->pg_token
);
809 sigio
->sio_pgid
= pgid
;
810 sigio
->sio_ucred
= crhold(curthread
->td_ucred
);
811 /* It would be convenient if p_ruid was in ucred. */
812 sigio
->sio_ruid
= sigio
->sio_ucred
->cr_ruid
;
813 sigio
->sio_myref
= sigiop
;
815 lwkt_gettoken(&sigio_token
);
819 lwkt_reltoken(&sigio_token
);
830 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
833 fgetown(struct sigio
**sigiop
)
838 lwkt_gettoken_shared(&sigio_token
);
840 own
= (sigio
!= NULL
? sigio
->sio_pgid
: 0);
841 lwkt_reltoken(&sigio_token
);
847 * Close many file descriptors.
850 sys_closefrom(struct closefrom_args
*uap
)
852 return(kern_closefrom(uap
->fd
));
856 * Close all file descriptors greater then or equal to fd
859 kern_closefrom(int fd
)
861 struct thread
*td
= curthread
;
862 struct proc
*p
= td
->td_proc
;
863 struct filedesc
*fdp
;
872 * NOTE: This function will skip unassociated descriptors and
873 * reserved descriptors that have not yet been assigned.
874 * fd_lastfile can change as a side effect of kern_close().
876 spin_lock(&fdp
->fd_spin
);
877 while (fd
<= fdp
->fd_lastfile
) {
878 if (fdp
->fd_files
[fd
].fp
!= NULL
) {
879 spin_unlock(&fdp
->fd_spin
);
880 /* ok if this races another close */
881 if (kern_close(fd
) == EINTR
)
883 spin_lock(&fdp
->fd_spin
);
887 spin_unlock(&fdp
->fd_spin
);
892 * Close a file descriptor.
895 sys_close(struct close_args
*uap
)
897 return(kern_close(uap
->fd
));
906 struct thread
*td
= curthread
;
907 struct proc
*p
= td
->td_proc
;
908 struct filedesc
*fdp
;
916 spin_lock(&fdp
->fd_spin
);
917 if ((fp
= funsetfd_locked(fdp
, fd
)) == NULL
) {
918 spin_unlock(&fdp
->fd_spin
);
922 if (p
->p_fdtol
!= NULL
) {
924 * Ask fdfree() to sleep to ensure that all relevant
925 * process leaders can be traversed in closef().
927 fdp
->fd_holdleaderscount
++;
932 * we now hold the fp reference that used to be owned by the descriptor
935 spin_unlock(&fdp
->fd_spin
);
936 if (SLIST_FIRST(&fp
->f_klist
))
937 knote_fdclose(fp
, fdp
, fd
);
938 error
= closef(fp
, p
);
940 spin_lock(&fdp
->fd_spin
);
941 fdp
->fd_holdleaderscount
--;
942 if (fdp
->fd_holdleaderscount
== 0 &&
943 fdp
->fd_holdleaderswakeup
!= 0) {
944 fdp
->fd_holdleaderswakeup
= 0;
945 spin_unlock(&fdp
->fd_spin
);
946 wakeup(&fdp
->fd_holdleaderscount
);
948 spin_unlock(&fdp
->fd_spin
);
955 * shutdown_args(int fd, int how)
958 kern_shutdown(int fd
, int how
)
960 struct thread
*td
= curthread
;
961 struct proc
*p
= td
->td_proc
;
967 if ((fp
= holdfp(p
->p_fd
, fd
, -1)) == NULL
)
969 error
= fo_shutdown(fp
, how
);
979 sys_shutdown(struct shutdown_args
*uap
)
983 error
= kern_shutdown(uap
->s
, uap
->how
);
992 kern_fstat(int fd
, struct stat
*ub
)
994 struct thread
*td
= curthread
;
995 struct proc
*p
= td
->td_proc
;
1001 if ((fp
= holdfp(p
->p_fd
, fd
, -1)) == NULL
)
1003 error
= fo_stat(fp
, ub
, td
->td_ucred
);
1010 * Return status information about a file descriptor.
1013 sys_fstat(struct fstat_args
*uap
)
1018 error
= kern_fstat(uap
->fd
, &st
);
1021 error
= copyout(&st
, uap
->sb
, sizeof(st
));
1026 * Return pathconf information about a file descriptor.
1031 sys_fpathconf(struct fpathconf_args
*uap
)
1033 struct thread
*td
= curthread
;
1034 struct proc
*p
= td
->td_proc
;
1039 if ((fp
= holdfp(p
->p_fd
, uap
->fd
, -1)) == NULL
)
1042 switch (fp
->f_type
) {
1045 if (uap
->name
!= _PC_PIPE_BUF
) {
1048 uap
->sysmsg_result
= PIPE_BUF
;
1054 vp
= (struct vnode
*)fp
->f_data
;
1055 error
= VOP_PATHCONF(vp
, uap
->name
, &uap
->sysmsg_reg
);
1065 static int fdexpand
;
1066 SYSCTL_INT(_debug
, OID_AUTO
, fdexpand
, CTLFLAG_RD
, &fdexpand
, 0,
1067 "Number of times a file table has been expanded");
1070 * Grow the file table so it can hold through descriptor (want).
1072 * The fdp's spinlock must be held exclusively on entry and may be held
1073 * exclusively on return. The spinlock may be cycled by the routine.
1076 fdgrow_locked(struct filedesc
*fdp
, int want
)
1078 struct fdnode
*newfiles
;
1079 struct fdnode
*oldfiles
;
1082 nf
= fdp
->fd_nfiles
;
1084 /* nf has to be of the form 2^n - 1 */
1086 } while (nf
<= want
);
1088 spin_unlock(&fdp
->fd_spin
);
1089 newfiles
= kmalloc(nf
* sizeof(struct fdnode
), M_FILEDESC
, M_WAITOK
);
1090 spin_lock(&fdp
->fd_spin
);
1093 * We could have raced another extend while we were not holding
1096 if (fdp
->fd_nfiles
>= nf
) {
1097 spin_unlock(&fdp
->fd_spin
);
1098 kfree(newfiles
, M_FILEDESC
);
1099 spin_lock(&fdp
->fd_spin
);
1103 * Copy the existing ofile and ofileflags arrays
1104 * and zero the new portion of each array.
1106 extra
= nf
- fdp
->fd_nfiles
;
1107 bcopy(fdp
->fd_files
, newfiles
, fdp
->fd_nfiles
* sizeof(struct fdnode
));
1108 bzero(&newfiles
[fdp
->fd_nfiles
], extra
* sizeof(struct fdnode
));
1110 oldfiles
= fdp
->fd_files
;
1111 fdp
->fd_files
= newfiles
;
1112 fdp
->fd_nfiles
= nf
;
1114 if (oldfiles
!= fdp
->fd_builtin_files
) {
1115 spin_unlock(&fdp
->fd_spin
);
1116 kfree(oldfiles
, M_FILEDESC
);
1117 spin_lock(&fdp
->fd_spin
);
1123 * Number of nodes in right subtree, including the root.
1126 right_subtree_size(int n
)
1128 return (n
^ (n
| (n
+ 1)));
1135 right_ancestor(int n
)
1137 return (n
| (n
+ 1));
1144 left_ancestor(int n
)
1146 return ((n
& (n
+ 1)) - 1);
1150 * Traverse the in-place binary tree buttom-up adjusting the allocation
1151 * count so scans can determine where free descriptors are located.
1153 * caller must be holding an exclusive spinlock on fdp
1157 fdreserve_locked(struct filedesc
*fdp
, int fd
, int incr
)
1160 fdp
->fd_files
[fd
].allocated
+= incr
;
1161 KKASSERT(fdp
->fd_files
[fd
].allocated
>= 0);
1162 fd
= left_ancestor(fd
);
1167 * Reserve a file descriptor for the process. If no error occurs, the
1168 * caller MUST at some point call fsetfd() or assign a file pointer
1169 * or dispose of the reservation.
1172 fdalloc(struct proc
*p
, int want
, int *result
)
1174 struct filedesc
*fdp
= p
->p_fd
;
1175 struct uidinfo
*uip
;
1176 int fd
, rsize
, rsum
, node
, lim
;
1179 * Check dtable size limit
1181 spin_lock(&p
->p_limit
->p_spin
);
1182 if (p
->p_rlimit
[RLIMIT_NOFILE
].rlim_cur
> INT_MAX
)
1185 lim
= (int)p
->p_rlimit
[RLIMIT_NOFILE
].rlim_cur
;
1186 spin_unlock(&p
->p_limit
->p_spin
);
1188 if (lim
> maxfilesperproc
)
1189 lim
= maxfilesperproc
;
1190 if (lim
< minfilesperproc
)
1191 lim
= minfilesperproc
;
1196 * Check that the user has not run out of descriptors (non-root only).
1197 * As a safety measure the dtable is allowed to have at least
1198 * minfilesperproc open fds regardless of the maxfilesperuser limit.
1200 if (p
->p_ucred
->cr_uid
&& fdp
->fd_nfiles
>= minfilesperproc
) {
1201 uip
= p
->p_ucred
->cr_uidinfo
;
1202 if (uip
->ui_openfiles
> maxfilesperuser
) {
1203 krateprintf(&krate_uidinfo
,
1204 "Warning: user %d pid %d (%s) ran out of "
1205 "file descriptors (%d/%d)\n",
1206 p
->p_ucred
->cr_uid
, (int)p
->p_pid
,
1208 uip
->ui_openfiles
, maxfilesperuser
);
1214 * Grow the dtable if necessary
1216 spin_lock(&fdp
->fd_spin
);
1217 if (want
>= fdp
->fd_nfiles
)
1218 fdgrow_locked(fdp
, want
);
1221 * Search for a free descriptor starting at the higher
1222 * of want or fd_freefile. If that fails, consider
1223 * expanding the ofile array.
1225 * NOTE! the 'allocated' field is a cumulative recursive allocation
1226 * count. If we happen to see a value of 0 then we can shortcut
1227 * our search. Otherwise we run through through the tree going
1228 * down branches we know have free descriptor(s) until we hit a
1229 * leaf node. The leaf node will be free but will not necessarily
1230 * have an allocated field of 0.
1233 /* move up the tree looking for a subtree with a free node */
1234 for (fd
= max(want
, fdp
->fd_freefile
); fd
< min(fdp
->fd_nfiles
, lim
);
1235 fd
= right_ancestor(fd
)) {
1236 if (fdp
->fd_files
[fd
].allocated
== 0)
1239 rsize
= right_subtree_size(fd
);
1240 if (fdp
->fd_files
[fd
].allocated
== rsize
)
1241 continue; /* right subtree full */
1244 * Free fd is in the right subtree of the tree rooted at fd.
1245 * Call that subtree R. Look for the smallest (leftmost)
1246 * subtree of R with an unallocated fd: continue moving
1247 * down the left branch until encountering a full left
1248 * subtree, then move to the right.
1250 for (rsum
= 0, rsize
/= 2; rsize
> 0; rsize
/= 2) {
1252 rsum
+= fdp
->fd_files
[node
].allocated
;
1253 if (fdp
->fd_files
[fd
].allocated
== rsum
+ rsize
) {
1254 fd
= node
; /* move to the right */
1255 if (fdp
->fd_files
[node
].allocated
== 0)
1264 * No space in current array. Expand?
1266 if (fdp
->fd_nfiles
>= lim
) {
1267 spin_unlock(&fdp
->fd_spin
);
1270 fdgrow_locked(fdp
, want
);
1274 KKASSERT(fd
< fdp
->fd_nfiles
);
1275 if (fd
> fdp
->fd_lastfile
)
1276 fdp
->fd_lastfile
= fd
;
1277 if (want
<= fdp
->fd_freefile
)
1278 fdp
->fd_freefile
= fd
;
1280 KKASSERT(fdp
->fd_files
[fd
].fp
== NULL
);
1281 KKASSERT(fdp
->fd_files
[fd
].reserved
== 0);
1282 fdp
->fd_files
[fd
].fileflags
= 0;
1283 fdp
->fd_files
[fd
].reserved
= 1;
1284 fdreserve_locked(fdp
, fd
, 1);
1285 spin_unlock(&fdp
->fd_spin
);
1290 * Check to see whether n user file descriptors
1291 * are available to the process p.
1294 fdavail(struct proc
*p
, int n
)
1296 struct filedesc
*fdp
= p
->p_fd
;
1297 struct fdnode
*fdnode
;
1300 spin_lock(&p
->p_limit
->p_spin
);
1301 if (p
->p_rlimit
[RLIMIT_NOFILE
].rlim_cur
> INT_MAX
)
1304 lim
= (int)p
->p_rlimit
[RLIMIT_NOFILE
].rlim_cur
;
1305 spin_unlock(&p
->p_limit
->p_spin
);
1307 if (lim
> maxfilesperproc
)
1308 lim
= maxfilesperproc
;
1309 if (lim
< minfilesperproc
)
1310 lim
= minfilesperproc
;
1312 spin_lock(&fdp
->fd_spin
);
1313 if ((i
= lim
- fdp
->fd_nfiles
) > 0 && (n
-= i
) <= 0) {
1314 spin_unlock(&fdp
->fd_spin
);
1317 last
= min(fdp
->fd_nfiles
, lim
);
1318 fdnode
= &fdp
->fd_files
[fdp
->fd_freefile
];
1319 for (i
= last
- fdp
->fd_freefile
; --i
>= 0; ++fdnode
) {
1320 if (fdnode
->fp
== NULL
&& --n
<= 0) {
1321 spin_unlock(&fdp
->fd_spin
);
1325 spin_unlock(&fdp
->fd_spin
);
1330 * Revoke open descriptors referencing (f_data, f_type)
1332 * Any revoke executed within a prison is only able to
1333 * revoke descriptors for processes within that prison.
1335 * Returns 0 on success or an error code.
1337 struct fdrevoke_info
{
1347 static int fdrevoke_check_callback(struct file
*fp
, void *vinfo
);
1348 static int fdrevoke_proc_callback(struct proc
*p
, void *vinfo
);
1351 fdrevoke(void *f_data
, short f_type
, struct ucred
*cred
)
1353 struct fdrevoke_info info
;
1356 bzero(&info
, sizeof(info
));
1360 error
= falloc(NULL
, &info
.nfp
, NULL
);
1365 * Scan the file pointer table once. dups do not dup file pointers,
1366 * only descriptors, so there is no leak. Set FREVOKED on the fps
1369 allfiles_scan_exclusive(fdrevoke_check_callback
, &info
);
1372 * If any fps were marked track down the related descriptors
1373 * and close them. Any dup()s at this point will notice
1374 * the FREVOKED already set in the fp and do the right thing.
1376 * Any fps with non-zero msgcounts (aka sent over a unix-domain
1377 * socket) bumped the intransit counter and will require a
1378 * scan. Races against fps leaving the socket are closed by
1379 * the socket code checking for FREVOKED.
1382 allproc_scan(fdrevoke_proc_callback
, &info
);
1384 unp_revoke_gc(info
.nfp
);
1390 * Locate matching file pointers directly.
1392 * WARNING: allfiles_scan_exclusive() holds a spinlock through these calls!
1395 fdrevoke_check_callback(struct file
*fp
, void *vinfo
)
1397 struct fdrevoke_info
*info
= vinfo
;
1400 * File pointers already flagged for revokation are skipped.
1402 if (fp
->f_flag
& FREVOKED
)
1406 * If revoking from a prison file pointers created outside of
1407 * that prison, or file pointers without creds, cannot be revoked.
1409 if (info
->cred
->cr_prison
&&
1410 (fp
->f_cred
== NULL
||
1411 info
->cred
->cr_prison
!= fp
->f_cred
->cr_prison
)) {
1416 * If the file pointer matches then mark it for revocation. The
1417 * flag is currently only used by unp_revoke_gc().
1419 * info->count is a heuristic and can race in a SMP environment.
1421 if (info
->data
== fp
->f_data
&& info
->type
== fp
->f_type
) {
1422 atomic_set_int(&fp
->f_flag
, FREVOKED
);
1423 info
->count
+= fp
->f_count
;
1431 * Locate matching file pointers via process descriptor tables.
1434 fdrevoke_proc_callback(struct proc
*p
, void *vinfo
)
1436 struct fdrevoke_info
*info
= vinfo
;
1437 struct filedesc
*fdp
;
1441 if (p
->p_stat
== SIDL
|| p
->p_stat
== SZOMB
)
1443 if (info
->cred
->cr_prison
&&
1444 info
->cred
->cr_prison
!= p
->p_ucred
->cr_prison
) {
1449 * If the controlling terminal of the process matches the
1450 * vnode being revoked we clear the controlling terminal.
1452 * The normal spec_close() may not catch this because it
1453 * uses curproc instead of p.
1455 if (p
->p_session
&& info
->type
== DTYPE_VNODE
&&
1456 info
->data
== p
->p_session
->s_ttyvp
) {
1457 p
->p_session
->s_ttyvp
= NULL
;
1462 * Softref the fdp to prevent it from being destroyed
1464 spin_lock(&p
->p_spin
);
1465 if ((fdp
= p
->p_fd
) == NULL
) {
1466 spin_unlock(&p
->p_spin
);
1469 atomic_add_int(&fdp
->fd_softrefs
, 1);
1470 spin_unlock(&p
->p_spin
);
1473 * Locate and close any matching file descriptors.
1475 spin_lock(&fdp
->fd_spin
);
1476 for (n
= 0; n
< fdp
->fd_nfiles
; ++n
) {
1477 if ((fp
= fdp
->fd_files
[n
].fp
) == NULL
)
1479 if (fp
->f_flag
& FREVOKED
) {
1481 fdp
->fd_files
[n
].fp
= info
->nfp
;
1482 spin_unlock(&fdp
->fd_spin
);
1483 knote_fdclose(fp
, fdp
, n
); /* XXX */
1485 spin_lock(&fdp
->fd_spin
);
1489 spin_unlock(&fdp
->fd_spin
);
1490 atomic_subtract_int(&fdp
->fd_softrefs
, 1);
1496 * Create a new open file structure and reserve a file decriptor
1497 * for the process that refers to it.
1499 * Root creds are checked using lp, or assumed if lp is NULL. If
1500 * resultfd is non-NULL then lp must also be non-NULL. No file
1501 * descriptor is reserved (and no process context is needed) if
1504 * A file pointer with a refcount of 1 is returned. Note that the
1505 * file pointer is NOT associated with the descriptor. If falloc
1506 * returns success, fsetfd() MUST be called to either associate the
1507 * file pointer or clear the reservation.
1510 falloc(struct lwp
*lp
, struct file
**resultfp
, int *resultfd
)
1512 static struct timeval lastfail
;
1515 struct ucred
*cred
= lp
? lp
->lwp_thread
->td_ucred
: proc0
.p_ucred
;
1521 * Handle filetable full issues and root overfill.
1523 if (nfiles
>= maxfiles
- maxfilesrootres
&&
1524 (cred
->cr_ruid
!= 0 || nfiles
>= maxfiles
)) {
1525 if (ppsratecheck(&lastfail
, &curfail
, 1)) {
1526 kprintf("kern.maxfiles limit exceeded by uid %d, "
1527 "please see tuning(7).\n",
1535 * Allocate a new file descriptor.
1537 fp
= kmalloc(sizeof(struct file
), M_FILE
, M_WAITOK
| M_ZERO
);
1538 spin_init(&fp
->f_spin
, "falloc");
1539 SLIST_INIT(&fp
->f_klist
);
1541 fp
->f_ops
= &badfileops
;
1544 spin_lock(&filehead_spin
);
1546 LIST_INSERT_HEAD(&filehead
, fp
, f_list
);
1547 spin_unlock(&filehead_spin
);
1549 if ((error
= fdalloc(lp
->lwp_proc
, 0, resultfd
)) != 0) {
1562 * Check for races against a file descriptor by determining that the
1563 * file pointer is still associated with the specified file descriptor,
1564 * and a close is not currently in progress.
1567 checkfdclosed(struct filedesc
*fdp
, int fd
, struct file
*fp
)
1571 spin_lock_shared(&fdp
->fd_spin
);
1572 if ((unsigned)fd
>= fdp
->fd_nfiles
|| fp
!= fdp
->fd_files
[fd
].fp
)
1576 spin_unlock_shared(&fdp
->fd_spin
);
1581 * Associate a file pointer with a previously reserved file descriptor.
1582 * This function always succeeds.
1584 * If fp is NULL, the file descriptor is returned to the pool.
1588 * (exclusive spinlock must be held on call)
1591 fsetfd_locked(struct filedesc
*fdp
, struct file
*fp
, int fd
)
1593 KKASSERT((unsigned)fd
< fdp
->fd_nfiles
);
1594 KKASSERT(fdp
->fd_files
[fd
].reserved
!= 0);
1597 fdp
->fd_files
[fd
].fp
= fp
;
1598 fdp
->fd_files
[fd
].reserved
= 0;
1600 fdp
->fd_files
[fd
].reserved
= 0;
1601 fdreserve_locked(fdp
, fd
, -1);
1602 fdfixup_locked(fdp
, fd
);
1607 fsetfd(struct filedesc
*fdp
, struct file
*fp
, int fd
)
1609 spin_lock(&fdp
->fd_spin
);
1610 fsetfd_locked(fdp
, fp
, fd
);
1611 spin_unlock(&fdp
->fd_spin
);
1615 * (exclusive spinlock must be held on call)
1619 funsetfd_locked(struct filedesc
*fdp
, int fd
)
1623 if ((unsigned)fd
>= fdp
->fd_nfiles
)
1625 if ((fp
= fdp
->fd_files
[fd
].fp
) == NULL
)
1627 fdp
->fd_files
[fd
].fp
= NULL
;
1628 fdp
->fd_files
[fd
].fileflags
= 0;
1630 fdreserve_locked(fdp
, fd
, -1);
1631 fdfixup_locked(fdp
, fd
);
1636 * WARNING: May not be called before initial fsetfd().
1639 fgetfdflags(struct filedesc
*fdp
, int fd
, int *flagsp
)
1643 spin_lock(&fdp
->fd_spin
);
1644 if (((u_int
)fd
) >= fdp
->fd_nfiles
) {
1646 } else if (fdp
->fd_files
[fd
].fp
== NULL
) {
1649 *flagsp
= fdp
->fd_files
[fd
].fileflags
;
1652 spin_unlock(&fdp
->fd_spin
);
1657 * WARNING: May not be called before initial fsetfd().
1660 fsetfdflags(struct filedesc
*fdp
, int fd
, int add_flags
)
1664 spin_lock(&fdp
->fd_spin
);
1665 if (((u_int
)fd
) >= fdp
->fd_nfiles
) {
1667 } else if (fdp
->fd_files
[fd
].fp
== NULL
) {
1670 fdp
->fd_files
[fd
].fileflags
|= add_flags
;
1673 spin_unlock(&fdp
->fd_spin
);
1678 * WARNING: May not be called before initial fsetfd().
1681 fclrfdflags(struct filedesc
*fdp
, int fd
, int rem_flags
)
1685 spin_lock(&fdp
->fd_spin
);
1686 if (((u_int
)fd
) >= fdp
->fd_nfiles
) {
1688 } else if (fdp
->fd_files
[fd
].fp
== NULL
) {
1691 fdp
->fd_files
[fd
].fileflags
&= ~rem_flags
;
1694 spin_unlock(&fdp
->fd_spin
);
1699 * Set/Change/Clear the creds for a fp and synchronize the uidinfo.
1702 fsetcred(struct file
*fp
, struct ucred
*ncr
)
1705 struct uidinfo
*uip
;
1708 if (ocr
== NULL
|| ncr
== NULL
|| ocr
->cr_uidinfo
!= ncr
->cr_uidinfo
) {
1710 uip
= ocr
->cr_uidinfo
;
1711 atomic_add_int(&uip
->ui_openfiles
, -1);
1714 uip
= ncr
->cr_uidinfo
;
1715 atomic_add_int(&uip
->ui_openfiles
, 1);
1726 * Free a file descriptor.
1730 ffree(struct file
*fp
)
1732 KASSERT((fp
->f_count
== 0), ("ffree: fp_fcount not 0!"));
1733 spin_lock(&filehead_spin
);
1734 LIST_REMOVE(fp
, f_list
);
1736 spin_unlock(&filehead_spin
);
1738 if (fp
->f_nchandle
.ncp
)
1739 cache_drop(&fp
->f_nchandle
);
1744 * called from init_main, initialize filedesc0 for proc0.
1747 fdinit_bootstrap(struct proc
*p0
, struct filedesc
*fdp0
, int cmask
)
1751 fdp0
->fd_refcnt
= 1;
1752 fdp0
->fd_cmask
= cmask
;
1753 fdp0
->fd_files
= fdp0
->fd_builtin_files
;
1754 fdp0
->fd_nfiles
= NDFILE
;
1755 fdp0
->fd_lastfile
= -1;
1756 spin_init(&fdp0
->fd_spin
, "fdinitbootstrap");
1760 * Build a new filedesc structure.
1763 fdinit(struct proc
*p
)
1765 struct filedesc
*newfdp
;
1766 struct filedesc
*fdp
= p
->p_fd
;
1768 newfdp
= kmalloc(sizeof(struct filedesc
), M_FILEDESC
, M_WAITOK
|M_ZERO
);
1769 spin_lock(&fdp
->fd_spin
);
1771 newfdp
->fd_cdir
= fdp
->fd_cdir
;
1772 vref(newfdp
->fd_cdir
);
1773 cache_copy(&fdp
->fd_ncdir
, &newfdp
->fd_ncdir
);
1777 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of
1778 * proc0, but should unconditionally exist in other processes.
1781 newfdp
->fd_rdir
= fdp
->fd_rdir
;
1782 vref(newfdp
->fd_rdir
);
1783 cache_copy(&fdp
->fd_nrdir
, &newfdp
->fd_nrdir
);
1786 newfdp
->fd_jdir
= fdp
->fd_jdir
;
1787 vref(newfdp
->fd_jdir
);
1788 cache_copy(&fdp
->fd_njdir
, &newfdp
->fd_njdir
);
1790 spin_unlock(&fdp
->fd_spin
);
1792 /* Create the file descriptor table. */
1793 newfdp
->fd_refcnt
= 1;
1794 newfdp
->fd_cmask
= cmask
;
1795 newfdp
->fd_files
= newfdp
->fd_builtin_files
;
1796 newfdp
->fd_nfiles
= NDFILE
;
1797 newfdp
->fd_lastfile
= -1;
1798 spin_init(&newfdp
->fd_spin
, "fdinit");
1804 * Share a filedesc structure.
1807 fdshare(struct proc
*p
)
1809 struct filedesc
*fdp
;
1812 spin_lock(&fdp
->fd_spin
);
1814 spin_unlock(&fdp
->fd_spin
);
1819 * Copy a filedesc structure.
1822 fdcopy(struct proc
*p
, struct filedesc
**fpp
)
1824 struct filedesc
*fdp
= p
->p_fd
;
1825 struct filedesc
*newfdp
;
1826 struct fdnode
*fdnode
;
1831 * Certain daemons might not have file descriptors.
1837 * Allocate the new filedesc and fd_files[] array. This can race
1838 * with operations by other threads on the fdp so we have to be
1841 newfdp
= kmalloc(sizeof(struct filedesc
),
1842 M_FILEDESC
, M_WAITOK
| M_ZERO
| M_NULLOK
);
1843 if (newfdp
== NULL
) {
1848 spin_lock(&fdp
->fd_spin
);
1849 if (fdp
->fd_lastfile
< NDFILE
) {
1850 newfdp
->fd_files
= newfdp
->fd_builtin_files
;
1854 * We have to allocate (N^2-1) entries for our in-place
1855 * binary tree. Allow the table to shrink.
1859 while (ni
> fdp
->fd_lastfile
&& ni
> NDFILE
) {
1863 spin_unlock(&fdp
->fd_spin
);
1864 newfdp
->fd_files
= kmalloc(i
* sizeof(struct fdnode
),
1865 M_FILEDESC
, M_WAITOK
| M_ZERO
);
1868 * Check for race, retry
1870 spin_lock(&fdp
->fd_spin
);
1871 if (i
<= fdp
->fd_lastfile
) {
1872 spin_unlock(&fdp
->fd_spin
);
1873 kfree(newfdp
->fd_files
, M_FILEDESC
);
1879 * Dup the remaining fields. vref() and cache_hold() can be
1880 * safely called while holding the read spinlock on fdp.
1882 * The read spinlock on fdp is still being held.
1884 * NOTE: vref and cache_hold calls for the case where the vnode
1885 * or cache entry already has at least one ref may be called
1886 * while holding spin locks.
1888 if ((newfdp
->fd_cdir
= fdp
->fd_cdir
) != NULL
) {
1889 vref(newfdp
->fd_cdir
);
1890 cache_copy(&fdp
->fd_ncdir
, &newfdp
->fd_ncdir
);
1893 * We must check for fd_rdir here, at least for now because
1894 * the init process is created before we have access to the
1895 * rootvode to take a reference to it.
1897 if ((newfdp
->fd_rdir
= fdp
->fd_rdir
) != NULL
) {
1898 vref(newfdp
->fd_rdir
);
1899 cache_copy(&fdp
->fd_nrdir
, &newfdp
->fd_nrdir
);
1901 if ((newfdp
->fd_jdir
= fdp
->fd_jdir
) != NULL
) {
1902 vref(newfdp
->fd_jdir
);
1903 cache_copy(&fdp
->fd_njdir
, &newfdp
->fd_njdir
);
1905 newfdp
->fd_refcnt
= 1;
1906 newfdp
->fd_nfiles
= i
;
1907 newfdp
->fd_lastfile
= fdp
->fd_lastfile
;
1908 newfdp
->fd_freefile
= fdp
->fd_freefile
;
1909 newfdp
->fd_cmask
= fdp
->fd_cmask
;
1910 spin_init(&newfdp
->fd_spin
, "fdcopy");
1913 * Copy the descriptor table through (i). This also copies the
1914 * allocation state. Then go through and ref the file pointers
1915 * and clean up any KQ descriptors.
1917 * kq descriptors cannot be copied. Since we haven't ref'd the
1918 * copied files yet we can ignore the return value from funsetfd().
1920 * The read spinlock on fdp is still being held.
1922 bcopy(fdp
->fd_files
, newfdp
->fd_files
, i
* sizeof(struct fdnode
));
1923 for (i
= 0 ; i
< newfdp
->fd_nfiles
; ++i
) {
1924 fdnode
= &newfdp
->fd_files
[i
];
1925 if (fdnode
->reserved
) {
1926 fdreserve_locked(newfdp
, i
, -1);
1927 fdnode
->reserved
= 0;
1928 fdfixup_locked(newfdp
, i
);
1929 } else if (fdnode
->fp
) {
1930 if (fdnode
->fp
->f_type
== DTYPE_KQUEUE
) {
1931 (void)funsetfd_locked(newfdp
, i
);
1937 spin_unlock(&fdp
->fd_spin
);
1943 * Release a filedesc structure.
1945 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE)
1948 fdfree(struct proc
*p
, struct filedesc
*repl
)
1950 struct filedesc
*fdp
;
1951 struct fdnode
*fdnode
;
1953 struct filedesc_to_leader
*fdtol
;
1959 * Certain daemons might not have file descriptors.
1968 * Severe messing around to follow.
1970 spin_lock(&fdp
->fd_spin
);
1972 /* Check for special need to clear POSIX style locks */
1974 if (fdtol
!= NULL
) {
1975 KASSERT(fdtol
->fdl_refcount
> 0,
1976 ("filedesc_to_refcount botch: fdl_refcount=%d",
1977 fdtol
->fdl_refcount
));
1978 if (fdtol
->fdl_refcount
== 1 &&
1979 (p
->p_leader
->p_flags
& P_ADVLOCK
) != 0) {
1980 for (i
= 0; i
<= fdp
->fd_lastfile
; ++i
) {
1981 fdnode
= &fdp
->fd_files
[i
];
1982 if (fdnode
->fp
== NULL
||
1983 fdnode
->fp
->f_type
!= DTYPE_VNODE
) {
1988 spin_unlock(&fdp
->fd_spin
);
1990 lf
.l_whence
= SEEK_SET
;
1993 lf
.l_type
= F_UNLCK
;
1994 vp
= (struct vnode
*)fp
->f_data
;
1995 (void) VOP_ADVLOCK(vp
,
1996 (caddr_t
)p
->p_leader
,
2001 spin_lock(&fdp
->fd_spin
);
2005 if (fdtol
->fdl_refcount
== 1) {
2006 if (fdp
->fd_holdleaderscount
> 0 &&
2007 (p
->p_leader
->p_flags
& P_ADVLOCK
) != 0) {
2009 * close() or do_dup() has cleared a reference
2010 * in a shared file descriptor table.
2012 fdp
->fd_holdleaderswakeup
= 1;
2013 ssleep(&fdp
->fd_holdleaderscount
,
2014 &fdp
->fd_spin
, 0, "fdlhold", 0);
2017 if (fdtol
->fdl_holdcount
> 0) {
2019 * Ensure that fdtol->fdl_leader
2020 * remains valid in closef().
2022 fdtol
->fdl_wakeup
= 1;
2023 ssleep(fdtol
, &fdp
->fd_spin
, 0, "fdlhold", 0);
2027 fdtol
->fdl_refcount
--;
2028 if (fdtol
->fdl_refcount
== 0 &&
2029 fdtol
->fdl_holdcount
== 0) {
2030 fdtol
->fdl_next
->fdl_prev
= fdtol
->fdl_prev
;
2031 fdtol
->fdl_prev
->fdl_next
= fdtol
->fdl_next
;
2036 if (fdtol
!= NULL
) {
2037 spin_unlock(&fdp
->fd_spin
);
2038 kfree(fdtol
, M_FILEDESC_TO_LEADER
);
2039 spin_lock(&fdp
->fd_spin
);
2042 if (--fdp
->fd_refcnt
> 0) {
2043 spin_unlock(&fdp
->fd_spin
);
2044 spin_lock(&p
->p_spin
);
2046 spin_unlock(&p
->p_spin
);
2051 * Even though we are the last reference to the structure allproc
2052 * scans may still reference the structure. Maintain proper
2053 * locks until we can replace p->p_fd.
2055 * Also note that kqueue's closef still needs to reference the
2056 * fdp via p->p_fd, so we have to close the descriptors before
2057 * we replace p->p_fd.
2059 for (i
= 0; i
<= fdp
->fd_lastfile
; ++i
) {
2060 if (fdp
->fd_files
[i
].fp
) {
2061 fp
= funsetfd_locked(fdp
, i
);
2063 spin_unlock(&fdp
->fd_spin
);
2064 if (SLIST_FIRST(&fp
->f_klist
))
2065 knote_fdclose(fp
, fdp
, i
);
2067 spin_lock(&fdp
->fd_spin
);
2071 spin_unlock(&fdp
->fd_spin
);
2074 * Interlock against an allproc scan operations (typically frevoke).
2076 spin_lock(&p
->p_spin
);
2078 spin_unlock(&p
->p_spin
);
2081 * Wait for any softrefs to go away. This race rarely occurs so
2082 * we can use a non-critical-path style poll/sleep loop. The
2083 * race only occurs against allproc scans.
2085 * No new softrefs can occur with the fdp disconnected from the
2088 if (fdp
->fd_softrefs
) {
2089 kprintf("pid %d: Warning, fdp race avoided\n", p
->p_pid
);
2090 while (fdp
->fd_softrefs
)
2091 tsleep(&fdp
->fd_softrefs
, 0, "fdsoft", 1);
2094 if (fdp
->fd_files
!= fdp
->fd_builtin_files
)
2095 kfree(fdp
->fd_files
, M_FILEDESC
);
2097 cache_drop(&fdp
->fd_ncdir
);
2098 vrele(fdp
->fd_cdir
);
2101 cache_drop(&fdp
->fd_nrdir
);
2102 vrele(fdp
->fd_rdir
);
2105 cache_drop(&fdp
->fd_njdir
);
2106 vrele(fdp
->fd_jdir
);
2108 kfree(fdp
, M_FILEDESC
);
2112 * Retrieve and reference the file pointer associated with a descriptor.
2115 holdfp(struct filedesc
*fdp
, int fd
, int flag
)
2119 spin_lock_shared(&fdp
->fd_spin
);
2120 if (((u_int
)fd
) >= fdp
->fd_nfiles
) {
2124 if ((fp
= fdp
->fd_files
[fd
].fp
) == NULL
)
2126 if ((fp
->f_flag
& flag
) == 0 && flag
!= -1) {
2132 spin_unlock_shared(&fdp
->fd_spin
);
2137 * holdsock() - load the struct file pointer associated
2138 * with a socket into *fpp. If an error occurs, non-zero
2139 * will be returned and *fpp will be set to NULL.
2142 holdsock(struct filedesc
*fdp
, int fd
, struct file
**fpp
)
2147 spin_lock_shared(&fdp
->fd_spin
);
2148 if ((unsigned)fd
>= fdp
->fd_nfiles
) {
2153 if ((fp
= fdp
->fd_files
[fd
].fp
) == NULL
) {
2157 if (fp
->f_type
!= DTYPE_SOCKET
) {
2164 spin_unlock_shared(&fdp
->fd_spin
);
2170 * Convert a user file descriptor to a held file pointer.
2173 holdvnode(struct filedesc
*fdp
, int fd
, struct file
**fpp
)
2178 spin_lock_shared(&fdp
->fd_spin
);
2179 if ((unsigned)fd
>= fdp
->fd_nfiles
) {
2184 if ((fp
= fdp
->fd_files
[fd
].fp
) == NULL
) {
2188 if (fp
->f_type
!= DTYPE_VNODE
&& fp
->f_type
!= DTYPE_FIFO
) {
2196 spin_unlock_shared(&fdp
->fd_spin
);
2202 * For setugid programs, we don't want to people to use that setugidness
2203 * to generate error messages which write to a file which otherwise would
2204 * otherwise be off-limits to the process.
2206 * This is a gross hack to plug the hole. A better solution would involve
2207 * a special vop or other form of generalized access control mechanism. We
2208 * go ahead and just reject all procfs file systems accesses as dangerous.
2210 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2211 * sufficient. We also don't for check setugidness since we know we are.
2214 is_unsafe(struct file
*fp
)
2216 if (fp
->f_type
== DTYPE_VNODE
&&
2217 ((struct vnode
*)(fp
->f_data
))->v_tag
== VT_PROCFS
)
2223 * Make this setguid thing safe, if at all possible.
2225 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2228 setugidsafety(struct proc
*p
)
2230 struct filedesc
*fdp
= p
->p_fd
;
2233 /* Certain daemons might not have file descriptors. */
2238 * note: fdp->fd_files may be reallocated out from under us while
2239 * we are blocked in a close. Be careful!
2241 for (i
= 0; i
<= fdp
->fd_lastfile
; i
++) {
2244 if (fdp
->fd_files
[i
].fp
&& is_unsafe(fdp
->fd_files
[i
].fp
)) {
2248 * NULL-out descriptor prior to close to avoid
2249 * a race while close blocks.
2251 if ((fp
= funsetfd_locked(fdp
, i
)) != NULL
) {
2252 knote_fdclose(fp
, fdp
, i
);
2260 * Close any files on exec?
2262 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2265 fdcloseexec(struct proc
*p
)
2267 struct filedesc
*fdp
= p
->p_fd
;
2270 /* Certain daemons might not have file descriptors. */
2275 * We cannot cache fd_files since operations may block and rip
2276 * them out from under us.
2278 for (i
= 0; i
<= fdp
->fd_lastfile
; i
++) {
2279 if (fdp
->fd_files
[i
].fp
!= NULL
&&
2280 (fdp
->fd_files
[i
].fileflags
& UF_EXCLOSE
)) {
2284 * NULL-out descriptor prior to close to avoid
2285 * a race while close blocks.
2287 if ((fp
= funsetfd_locked(fdp
, i
)) != NULL
) {
2288 knote_fdclose(fp
, fdp
, i
);
2296 * It is unsafe for set[ug]id processes to be started with file
2297 * descriptors 0..2 closed, as these descriptors are given implicit
2298 * significance in the Standard C library. fdcheckstd() will create a
2299 * descriptor referencing /dev/null for each of stdin, stdout, and
2300 * stderr that is not already open.
2302 * NOT MPSAFE - calls falloc, vn_open, etc
2305 fdcheckstd(struct lwp
*lp
)
2307 struct nlookupdata nd
;
2308 struct filedesc
*fdp
;
2311 int i
, error
, flags
, devnull
;
2313 fdp
= lp
->lwp_proc
->p_fd
;
2318 for (i
= 0; i
< 3; i
++) {
2319 if (fdp
->fd_files
[i
].fp
!= NULL
)
2322 if ((error
= falloc(lp
, &fp
, &devnull
)) != 0)
2325 error
= nlookup_init(&nd
, "/dev/null", UIO_SYSSPACE
,
2326 NLC_FOLLOW
|NLC_LOCKVP
);
2327 flags
= FREAD
| FWRITE
;
2329 error
= vn_open(&nd
, fp
, flags
, 0);
2331 fsetfd(fdp
, fp
, devnull
);
2333 fsetfd(fdp
, NULL
, devnull
);
2338 KKASSERT(i
== devnull
);
2340 error
= kern_dup(DUP_FIXED
, devnull
, i
, &retval
);
2349 * Internal form of close.
2350 * Decrement reference count on file structure.
2351 * Note: td and/or p may be NULL when closing a file
2352 * that was being passed in a message.
2354 * MPALMOSTSAFE - acquires mplock for VOP operations
2357 closef(struct file
*fp
, struct proc
*p
)
2361 struct filedesc_to_leader
*fdtol
;
2367 * POSIX record locking dictates that any close releases ALL
2368 * locks owned by this process. This is handled by setting
2369 * a flag in the unlock to free ONLY locks obeying POSIX
2370 * semantics, and not to free BSD-style file locks.
2371 * If the descriptor was in a message, POSIX-style locks
2372 * aren't passed with the descriptor.
2374 if (p
!= NULL
&& fp
->f_type
== DTYPE_VNODE
&&
2375 (((struct vnode
*)fp
->f_data
)->v_flag
& VMAYHAVELOCKS
)
2377 if ((p
->p_leader
->p_flags
& P_ADVLOCK
) != 0) {
2378 lf
.l_whence
= SEEK_SET
;
2381 lf
.l_type
= F_UNLCK
;
2382 vp
= (struct vnode
*)fp
->f_data
;
2383 (void) VOP_ADVLOCK(vp
, (caddr_t
)p
->p_leader
, F_UNLCK
,
2387 if (fdtol
!= NULL
) {
2388 lwkt_gettoken(&p
->p_token
);
2390 * Handle special case where file descriptor table
2391 * is shared between multiple process leaders.
2393 for (fdtol
= fdtol
->fdl_next
;
2394 fdtol
!= p
->p_fdtol
;
2395 fdtol
= fdtol
->fdl_next
) {
2396 if ((fdtol
->fdl_leader
->p_flags
&
2399 fdtol
->fdl_holdcount
++;
2400 lf
.l_whence
= SEEK_SET
;
2403 lf
.l_type
= F_UNLCK
;
2404 vp
= (struct vnode
*)fp
->f_data
;
2405 (void) VOP_ADVLOCK(vp
,
2406 (caddr_t
)fdtol
->fdl_leader
,
2407 F_UNLCK
, &lf
, F_POSIX
);
2408 fdtol
->fdl_holdcount
--;
2409 if (fdtol
->fdl_holdcount
== 0 &&
2410 fdtol
->fdl_wakeup
!= 0) {
2411 fdtol
->fdl_wakeup
= 0;
2415 lwkt_reltoken(&p
->p_token
);
2422 * fhold() can only be called if f_count is already at least 1 (i.e. the
2423 * caller of fhold() already has a reference to the file pointer in some
2426 * f_count is not spin-locked. Instead, atomic ops are used for
2427 * incrementing, decrementing, and handling the 1->0 transition.
2430 fhold(struct file
*fp
)
2432 atomic_add_int(&fp
->f_count
, 1);
2436 * fdrop() - drop a reference to a descriptor
2438 * MPALMOSTSAFE - acquires mplock for final close sequence
2441 fdrop(struct file
*fp
)
2448 * A combined fetch and subtract is needed to properly detect
2449 * 1->0 transitions, otherwise two cpus dropping from a ref
2450 * count of 2 might both try to run the 1->0 code.
2452 if (atomic_fetchadd_int(&fp
->f_count
, -1) > 1)
2455 KKASSERT(SLIST_FIRST(&fp
->f_klist
) == NULL
);
2458 * The last reference has gone away, we own the fp structure free
2461 if (fp
->f_count
< 0)
2462 panic("fdrop: count < 0");
2463 if ((fp
->f_flag
& FHASLOCK
) && fp
->f_type
== DTYPE_VNODE
&&
2464 (((struct vnode
*)fp
->f_data
)->v_flag
& VMAYHAVELOCKS
)
2466 lf
.l_whence
= SEEK_SET
;
2469 lf
.l_type
= F_UNLCK
;
2470 vp
= (struct vnode
*)fp
->f_data
;
2471 (void) VOP_ADVLOCK(vp
, (caddr_t
)fp
, F_UNLCK
, &lf
, 0);
2473 if (fp
->f_ops
!= &badfileops
)
2474 error
= fo_close(fp
);
2482 * Apply an advisory lock on a file descriptor.
2484 * Just attempt to get a record lock of the requested type on
2485 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2490 sys_flock(struct flock_args
*uap
)
2492 struct proc
*p
= curproc
;
2498 if ((fp
= holdfp(p
->p_fd
, uap
->fd
, -1)) == NULL
)
2500 if (fp
->f_type
!= DTYPE_VNODE
) {
2504 vp
= (struct vnode
*)fp
->f_data
;
2505 lf
.l_whence
= SEEK_SET
;
2508 if (uap
->how
& LOCK_UN
) {
2509 lf
.l_type
= F_UNLCK
;
2510 atomic_clear_int(&fp
->f_flag
, FHASLOCK
); /* race ok */
2511 error
= VOP_ADVLOCK(vp
, (caddr_t
)fp
, F_UNLCK
, &lf
, 0);
2514 if (uap
->how
& LOCK_EX
)
2515 lf
.l_type
= F_WRLCK
;
2516 else if (uap
->how
& LOCK_SH
)
2517 lf
.l_type
= F_RDLCK
;
2522 if (uap
->how
& LOCK_NB
)
2523 error
= VOP_ADVLOCK(vp
, (caddr_t
)fp
, F_SETLK
, &lf
, 0);
2525 error
= VOP_ADVLOCK(vp
, (caddr_t
)fp
, F_SETLK
, &lf
, F_WAIT
);
2526 atomic_set_int(&fp
->f_flag
, FHASLOCK
); /* race ok */
2533 * File Descriptor pseudo-device driver (/dev/fd/).
2535 * Opening minor device N dup()s the file (if any) connected to file
2536 * descriptor N belonging to the calling process. Note that this driver
2537 * consists of only the ``open()'' routine, because all subsequent
2538 * references to this file will be direct to the other driver.
2541 fdopen(struct dev_open_args
*ap
)
2543 thread_t td
= curthread
;
2545 KKASSERT(td
->td_lwp
!= NULL
);
2548 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the
2549 * the file descriptor being sought for duplication. The error
2550 * return ensures that the vnode for this device will be released
2551 * by vn_open. Open will detect this special error and take the
2552 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2553 * will simply report the error.
2555 td
->td_lwp
->lwp_dupfd
= minor(ap
->a_head
.a_dev
);
2560 * The caller has reserved the file descriptor dfd for us. On success we
2561 * must fsetfd() it. On failure the caller will clean it up.
2564 dupfdopen(struct filedesc
*fdp
, int dfd
, int sfd
, int mode
, int error
)
2570 if ((wfp
= holdfp(fdp
, sfd
, -1)) == NULL
)
2574 * Close a revoke/dup race. Duping a descriptor marked as revoked
2575 * will dup a dummy descriptor instead of the real one.
2577 if (wfp
->f_flag
& FREVOKED
) {
2578 kprintf("Warning: attempt to dup() a revoked descriptor\n");
2581 werror
= falloc(NULL
, &wfp
, NULL
);
2587 * There are two cases of interest here.
2589 * For ENODEV simply dup sfd to file descriptor dfd and return.
2591 * For ENXIO steal away the file structure from sfd and store it
2592 * dfd. sfd is effectively closed by this operation.
2594 * Any other error code is just returned.
2599 * Check that the mode the file is being opened for is a
2600 * subset of the mode of the existing descriptor.
2602 if (((mode
& (FREAD
|FWRITE
)) | wfp
->f_flag
) != wfp
->f_flag
) {
2606 spin_lock(&fdp
->fd_spin
);
2607 fdp
->fd_files
[dfd
].fileflags
= fdp
->fd_files
[sfd
].fileflags
;
2608 fsetfd_locked(fdp
, wfp
, dfd
);
2609 spin_unlock(&fdp
->fd_spin
);
2614 * Steal away the file pointer from dfd, and stuff it into indx.
2616 spin_lock(&fdp
->fd_spin
);
2617 fdp
->fd_files
[dfd
].fileflags
= fdp
->fd_files
[sfd
].fileflags
;
2618 fsetfd(fdp
, wfp
, dfd
);
2619 if ((xfp
= funsetfd_locked(fdp
, sfd
)) != NULL
) {
2620 spin_unlock(&fdp
->fd_spin
);
2623 spin_unlock(&fdp
->fd_spin
);
2635 * NOT MPSAFE - I think these refer to a common file descriptor table
2636 * and we need to spinlock that to link fdtol in.
2638 struct filedesc_to_leader
*
2639 filedesc_to_leader_alloc(struct filedesc_to_leader
*old
,
2640 struct proc
*leader
)
2642 struct filedesc_to_leader
*fdtol
;
2644 fdtol
= kmalloc(sizeof(struct filedesc_to_leader
),
2645 M_FILEDESC_TO_LEADER
, M_WAITOK
| M_ZERO
);
2646 fdtol
->fdl_refcount
= 1;
2647 fdtol
->fdl_holdcount
= 0;
2648 fdtol
->fdl_wakeup
= 0;
2649 fdtol
->fdl_leader
= leader
;
2651 fdtol
->fdl_next
= old
->fdl_next
;
2652 fdtol
->fdl_prev
= old
;
2653 old
->fdl_next
= fdtol
;
2654 fdtol
->fdl_next
->fdl_prev
= fdtol
;
2656 fdtol
->fdl_next
= fdtol
;
2657 fdtol
->fdl_prev
= fdtol
;
2663 * Scan all file pointers in the system. The callback is made with
2664 * the master list spinlock held exclusively.
2667 allfiles_scan_exclusive(int (*callback
)(struct file
*, void *), void *data
)
2672 spin_lock(&filehead_spin
);
2673 LIST_FOREACH(fp
, &filehead
, f_list
) {
2674 res
= callback(fp
, data
);
2678 spin_unlock(&filehead_spin
);
2682 * Get file structures.
2684 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe)
2687 struct sysctl_kern_file_info
{
2690 struct sysctl_req
*req
;
2693 static int sysctl_kern_file_callback(struct proc
*p
, void *data
);
2696 sysctl_kern_file(SYSCTL_HANDLER_ARGS
)
2698 struct sysctl_kern_file_info info
;
2701 * Note: because the number of file descriptors is calculated
2702 * in different ways for sizing vs returning the data,
2703 * there is information leakage from the first loop. However,
2704 * it is of a similar order of magnitude to the leakage from
2705 * global system statistics such as kern.openfiles.
2707 * When just doing a count, note that we cannot just count
2708 * the elements and add f_count via the filehead list because
2709 * threaded processes share their descriptor table and f_count might
2710 * still be '1' in that case.
2712 * Since the SYSCTL op can block, we must hold the process to
2713 * prevent it being ripped out from under us either in the
2714 * file descriptor loop or in the greater LIST_FOREACH. The
2715 * process may be in varying states of disrepair. If the process
2716 * is in SZOMB we may have caught it just as it is being removed
2717 * from the allproc list, we must skip it in that case to maintain
2718 * an unbroken chain through the allproc list.
2723 allproc_scan(sysctl_kern_file_callback
, &info
);
2726 * When just calculating the size, overestimate a bit to try to
2727 * prevent system activity from causing the buffer-fill call
2730 if (req
->oldptr
== NULL
) {
2731 info
.count
= (info
.count
+ 16) + (info
.count
/ 10);
2732 info
.error
= SYSCTL_OUT(req
, NULL
,
2733 info
.count
* sizeof(struct kinfo_file
));
2735 return (info
.error
);
2739 sysctl_kern_file_callback(struct proc
*p
, void *data
)
2741 struct sysctl_kern_file_info
*info
= data
;
2742 struct kinfo_file kf
;
2743 struct filedesc
*fdp
;
2748 if (p
->p_stat
== SIDL
|| p
->p_stat
== SZOMB
)
2750 if (!(PRISON_CHECK(info
->req
->td
->td_ucred
, p
->p_ucred
) != 0))
2754 * Softref the fdp to prevent it from being destroyed
2756 spin_lock(&p
->p_spin
);
2757 if ((fdp
= p
->p_fd
) == NULL
) {
2758 spin_unlock(&p
->p_spin
);
2761 atomic_add_int(&fdp
->fd_softrefs
, 1);
2762 spin_unlock(&p
->p_spin
);
2765 * The fdp's own spinlock prevents the contents from being
2768 spin_lock_shared(&fdp
->fd_spin
);
2769 for (n
= 0; n
< fdp
->fd_nfiles
; ++n
) {
2770 if ((fp
= fdp
->fd_files
[n
].fp
) == NULL
)
2772 if (info
->req
->oldptr
== NULL
) {
2775 uid
= p
->p_ucred
? p
->p_ucred
->cr_uid
: -1;
2776 kcore_make_file(&kf
, fp
, p
->p_pid
, uid
, n
);
2777 spin_unlock_shared(&fdp
->fd_spin
);
2778 info
->error
= SYSCTL_OUT(info
->req
, &kf
, sizeof(kf
));
2779 spin_lock_shared(&fdp
->fd_spin
);
2784 spin_unlock_shared(&fdp
->fd_spin
);
2785 atomic_subtract_int(&fdp
->fd_softrefs
, 1);
2791 SYSCTL_PROC(_kern
, KERN_FILE
, file
, CTLTYPE_OPAQUE
|CTLFLAG_RD
,
2792 0, 0, sysctl_kern_file
, "S,file", "Entire file table");
2794 SYSCTL_INT(_kern
, OID_AUTO
, minfilesperproc
, CTLFLAG_RW
,
2795 &minfilesperproc
, 0, "Minimum files allowed open per process");
2796 SYSCTL_INT(_kern
, KERN_MAXFILESPERPROC
, maxfilesperproc
, CTLFLAG_RW
,
2797 &maxfilesperproc
, 0, "Maximum files allowed open per process");
2798 SYSCTL_INT(_kern
, OID_AUTO
, maxfilesperuser
, CTLFLAG_RW
,
2799 &maxfilesperuser
, 0, "Maximum files allowed open per user");
2801 SYSCTL_INT(_kern
, KERN_MAXFILES
, maxfiles
, CTLFLAG_RW
,
2802 &maxfiles
, 0, "Maximum number of files");
2804 SYSCTL_INT(_kern
, OID_AUTO
, maxfilesrootres
, CTLFLAG_RW
,
2805 &maxfilesrootres
, 0, "Descriptors reserved for root use");
2807 SYSCTL_INT(_kern
, OID_AUTO
, openfiles
, CTLFLAG_RD
,
2808 &nfiles
, 0, "System-wide number of open files");
2811 fildesc_drvinit(void *unused
)
2815 for (fd
= 0; fd
< NUMFDESC
; fd
++) {
2816 make_dev(&fildesc_ops
, fd
,
2817 UID_BIN
, GID_BIN
, 0666, "fd/%d", fd
);
2820 make_dev(&fildesc_ops
, 0, UID_ROOT
, GID_WHEEL
, 0666, "stdin");
2821 make_dev(&fildesc_ops
, 1, UID_ROOT
, GID_WHEEL
, 0666, "stdout");
2822 make_dev(&fildesc_ops
, 2, UID_ROOT
, GID_WHEEL
, 0666, "stderr");
2825 struct fileops badfileops
= {
2826 .fo_read
= badfo_readwrite
,
2827 .fo_write
= badfo_readwrite
,
2828 .fo_ioctl
= badfo_ioctl
,
2829 .fo_kqfilter
= badfo_kqfilter
,
2830 .fo_stat
= badfo_stat
,
2831 .fo_close
= badfo_close
,
2832 .fo_shutdown
= badfo_shutdown
2846 badfo_ioctl(struct file
*fp
, u_long com
, caddr_t data
,
2847 struct ucred
*cred
, struct sysmsg
*msgv
)
2853 * Must return an error to prevent registration, typically
2854 * due to a revoked descriptor (file_filtops assigned).
2857 badfo_kqfilter(struct file
*fp
, struct knote
*kn
)
2859 return (EOPNOTSUPP
);
2863 badfo_stat(struct file
*fp
, struct stat
*sb
, struct ucred
*cred
)
2869 badfo_close(struct file
*fp
)
2875 badfo_shutdown(struct file
*fp
, int how
)
2881 nofo_shutdown(struct file
*fp
, int how
)
2883 return (EOPNOTSUPP
);
2886 SYSINIT(fildescdev
, SI_SUB_DRIVERS
, SI_ORDER_MIDDLE
+ CDEV_MAJOR
,
2887 fildesc_drvinit
,NULL
);