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>
95 #include <sys/objcache.h>
98 #include <vm/vm_extern.h>
100 #include <sys/thread2.h>
101 #include <sys/file2.h>
102 #include <sys/spinlock2.h>
104 static void fsetfd_locked(struct filedesc
*fdp
, struct file
*fp
, int fd
);
105 static void fdreserve_locked (struct filedesc
*fdp
, int fd0
, int incr
);
106 static struct file
*funsetfd_locked (struct filedesc
*fdp
, int fd
);
107 static void ffree(struct file
*fp
);
109 static MALLOC_DEFINE(M_FILEDESC
, "file desc", "Open file descriptor table");
110 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER
, "file desc to leader",
111 "file desc to leader structures");
112 MALLOC_DEFINE(M_FILE
, "file", "Open file structure");
113 static MALLOC_DEFINE(M_SIGIO
, "sigio", "sigio structures");
115 static struct krate krate_uidinfo
= { .freq
= 1 };
117 static d_open_t fdopen
;
120 #define CDEV_MAJOR 22
121 static struct dev_ops fildesc_ops
= {
127 * Descriptor management.
129 #ifndef NFILELIST_HEADS
130 #define NFILELIST_HEADS 257 /* primary number */
133 struct filelist_head
{
134 struct spinlock spin
;
135 struct filelist list
;
138 static struct filelist_head filelist_heads
[NFILELIST_HEADS
];
140 static int nfiles
; /* actual number of open files */
143 struct lwkt_token revoke_token
= LWKT_TOKEN_INITIALIZER(revoke_token
);
145 static struct objcache
*file_objcache
;
147 static struct objcache_malloc_args file_malloc_args
= {
148 .objsize
= sizeof(struct file
),
153 * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared.
155 * must be called with fdp->fd_spin exclusively held
159 fdfixup_locked(struct filedesc
*fdp
, int fd
)
161 if (fd
< fdp
->fd_freefile
) {
162 fdp
->fd_freefile
= fd
;
164 while (fdp
->fd_lastfile
>= 0 &&
165 fdp
->fd_files
[fdp
->fd_lastfile
].fp
== NULL
&&
166 fdp
->fd_files
[fdp
->fd_lastfile
].reserved
== 0
172 static __inline
struct filelist_head
*
173 fp2filelist(const struct file
*fp
)
177 i
= (u_int
)(uintptr_t)fp
% NFILELIST_HEADS
;
178 return &filelist_heads
[i
];
182 * System calls on descriptors.
185 sys_getdtablesize(struct getdtablesize_args
*uap
)
187 struct proc
*p
= curproc
;
188 struct plimit
*limit
= p
->p_limit
;
191 spin_lock(&limit
->p_spin
);
192 if (limit
->pl_rlimit
[RLIMIT_NOFILE
].rlim_cur
> INT_MAX
)
195 dtsize
= (int)limit
->pl_rlimit
[RLIMIT_NOFILE
].rlim_cur
;
196 spin_unlock(&limit
->p_spin
);
198 if (dtsize
> maxfilesperproc
)
199 dtsize
= maxfilesperproc
;
200 if (dtsize
< minfilesperproc
)
201 dtsize
= minfilesperproc
;
202 if (p
->p_ucred
->cr_uid
&& dtsize
> maxfilesperuser
)
203 dtsize
= maxfilesperuser
;
204 uap
->sysmsg_result
= dtsize
;
209 * Duplicate a file descriptor to a particular value.
211 * note: keep in mind that a potential race condition exists when closing
212 * descriptors from a shared descriptor table (via rfork).
215 sys_dup2(struct dup2_args
*uap
)
220 error
= kern_dup(DUP_FIXED
, uap
->from
, uap
->to
, &fd
);
221 uap
->sysmsg_fds
[0] = fd
;
227 * Duplicate a file descriptor.
230 sys_dup(struct dup_args
*uap
)
235 error
= kern_dup(DUP_VARIABLE
, uap
->fd
, 0, &fd
);
236 uap
->sysmsg_fds
[0] = fd
;
242 * MPALMOSTSAFE - acquires mplock for fp operations
245 kern_fcntl(int fd
, int cmd
, union fcntl_dat
*dat
, struct ucred
*cred
)
247 struct thread
*td
= curthread
;
248 struct proc
*p
= td
->td_proc
;
254 int tmp
, error
, flg
= F_POSIX
;
259 * Operations on file descriptors that do not require a file pointer.
263 error
= fgetfdflags(p
->p_fd
, fd
, &tmp
);
265 dat
->fc_cloexec
= (tmp
& UF_EXCLOSE
) ? FD_CLOEXEC
: 0;
269 if (dat
->fc_cloexec
& FD_CLOEXEC
)
270 error
= fsetfdflags(p
->p_fd
, fd
, UF_EXCLOSE
);
272 error
= fclrfdflags(p
->p_fd
, fd
, UF_EXCLOSE
);
276 error
= kern_dup(DUP_VARIABLE
| DUP_FCNTL
, fd
, newmin
,
279 case F_DUPFD_CLOEXEC
:
281 error
= kern_dup(DUP_VARIABLE
| DUP_CLOEXEC
| DUP_FCNTL
,
282 fd
, newmin
, &dat
->fc_fd
);
286 error
= kern_dup(DUP_FIXED
, fd
, newmin
, &dat
->fc_fd
);
288 case F_DUP2FD_CLOEXEC
:
290 error
= kern_dup(DUP_FIXED
| DUP_CLOEXEC
, fd
, newmin
,
298 * Operations on file pointers
300 if ((fp
= holdfp(p
->p_fd
, fd
, -1)) == NULL
)
305 dat
->fc_flags
= OFLAGS(fp
->f_flag
);
311 nflags
= FFLAGS(dat
->fc_flags
& ~O_ACCMODE
) & FCNTLFLAGS
;
312 nflags
|= oflags
& ~FCNTLFLAGS
;
315 if (((nflags
^ oflags
) & O_APPEND
) && (oflags
& FAPPENDONLY
))
317 if (error
== 0 && ((nflags
^ oflags
) & FASYNC
)) {
318 tmp
= nflags
& FASYNC
;
319 error
= fo_ioctl(fp
, FIOASYNC
, (caddr_t
)&tmp
,
324 * If no error, must be atomically set.
329 nflags
= (oflags
& ~FCNTLFLAGS
) | (nflags
& FCNTLFLAGS
);
330 if (atomic_cmpset_int(&fp
->f_flag
, oflags
, nflags
))
337 error
= fo_ioctl(fp
, FIOGETOWN
, (caddr_t
)&dat
->fc_owner
,
342 error
= fo_ioctl(fp
, FIOSETOWN
, (caddr_t
)&dat
->fc_owner
,
348 /* Fall into F_SETLK */
351 if (fp
->f_type
!= DTYPE_VNODE
) {
355 vp
= (struct vnode
*)fp
->f_data
;
358 * copyin/lockop may block
360 if (dat
->fc_flock
.l_whence
== SEEK_CUR
)
361 dat
->fc_flock
.l_start
+= fp
->f_offset
;
363 switch (dat
->fc_flock
.l_type
) {
365 if ((fp
->f_flag
& FREAD
) == 0) {
369 if ((p
->p_leader
->p_flags
& P_ADVLOCK
) == 0) {
370 lwkt_gettoken(&p
->p_leader
->p_token
);
371 p
->p_leader
->p_flags
|= P_ADVLOCK
;
372 lwkt_reltoken(&p
->p_leader
->p_token
);
374 error
= VOP_ADVLOCK(vp
, (caddr_t
)p
->p_leader
, F_SETLK
,
375 &dat
->fc_flock
, flg
);
378 if ((fp
->f_flag
& FWRITE
) == 0) {
382 if ((p
->p_leader
->p_flags
& P_ADVLOCK
) == 0) {
383 lwkt_gettoken(&p
->p_leader
->p_token
);
384 p
->p_leader
->p_flags
|= P_ADVLOCK
;
385 lwkt_reltoken(&p
->p_leader
->p_token
);
387 error
= VOP_ADVLOCK(vp
, (caddr_t
)p
->p_leader
, F_SETLK
,
388 &dat
->fc_flock
, flg
);
391 error
= VOP_ADVLOCK(vp
, (caddr_t
)p
->p_leader
, F_UNLCK
,
392 &dat
->fc_flock
, F_POSIX
);
400 * It is possible to race a close() on the descriptor while
401 * we were blocked getting the lock. If this occurs the
402 * close might not have caught the lock.
404 if (checkfdclosed(p
->p_fd
, fd
, fp
)) {
405 dat
->fc_flock
.l_whence
= SEEK_SET
;
406 dat
->fc_flock
.l_start
= 0;
407 dat
->fc_flock
.l_len
= 0;
408 dat
->fc_flock
.l_type
= F_UNLCK
;
409 (void) VOP_ADVLOCK(vp
, (caddr_t
)p
->p_leader
,
410 F_UNLCK
, &dat
->fc_flock
, F_POSIX
);
415 if (fp
->f_type
!= DTYPE_VNODE
) {
419 vp
= (struct vnode
*)fp
->f_data
;
421 * copyin/lockop may block
423 if (dat
->fc_flock
.l_type
!= F_RDLCK
&&
424 dat
->fc_flock
.l_type
!= F_WRLCK
&&
425 dat
->fc_flock
.l_type
!= F_UNLCK
) {
429 if (dat
->fc_flock
.l_whence
== SEEK_CUR
)
430 dat
->fc_flock
.l_start
+= fp
->f_offset
;
431 error
= VOP_ADVLOCK(vp
, (caddr_t
)p
->p_leader
, F_GETLK
,
432 &dat
->fc_flock
, F_POSIX
);
444 * The file control system call.
447 sys_fcntl(struct fcntl_args
*uap
)
455 case F_DUPFD_CLOEXEC
:
456 case F_DUP2FD_CLOEXEC
:
457 dat
.fc_fd
= uap
->arg
;
460 dat
.fc_cloexec
= uap
->arg
;
463 dat
.fc_flags
= uap
->arg
;
466 dat
.fc_owner
= uap
->arg
;
471 error
= copyin((caddr_t
)uap
->arg
, &dat
.fc_flock
,
472 sizeof(struct flock
));
478 error
= kern_fcntl(uap
->fd
, uap
->cmd
, &dat
, curthread
->td_ucred
);
484 case F_DUPFD_CLOEXEC
:
485 case F_DUP2FD_CLOEXEC
:
486 uap
->sysmsg_result
= dat
.fc_fd
;
489 uap
->sysmsg_result
= dat
.fc_cloexec
;
492 uap
->sysmsg_result
= dat
.fc_flags
;
495 uap
->sysmsg_result
= dat
.fc_owner
;
498 error
= copyout(&dat
.fc_flock
, (caddr_t
)uap
->arg
,
499 sizeof(struct flock
));
508 * Common code for dup, dup2, and fcntl(F_DUPFD).
510 * There are four type flags: DUP_FCNTL, DUP_FIXED, DUP_VARIABLE, and
513 * DUP_FCNTL is for handling EINVAL vs. EBADF differences between
514 * fcntl()'s F_DUPFD and F_DUPFD_CLOEXEC and dup2() (per POSIX).
515 * The next two flags are mutually exclusive, and the fourth is optional.
516 * DUP_FIXED tells kern_dup() to destructively dup over an existing file
517 * descriptor if "new" is already open. DUP_VARIABLE tells kern_dup()
518 * to find the lowest unused file descriptor that is greater than or
519 * equal to "new". DUP_CLOEXEC, which works with either of the first
520 * two flags, sets the close-on-exec flag on the "new" file descriptor.
523 kern_dup(int flags
, int old
, int new, int *res
)
525 struct thread
*td
= curthread
;
526 struct proc
*p
= td
->td_proc
;
527 struct filedesc
*fdp
= p
->p_fd
;
536 * Verify that we have a valid descriptor to dup from and
537 * possibly to dup to. When the new descriptor is out of
538 * bounds, fcntl()'s F_DUPFD and F_DUPFD_CLOEXEC must
539 * return EINVAL, while dup2() returns EBADF in
542 * NOTE: maxfilesperuser is not applicable to dup()
545 if (p
->p_rlimit
[RLIMIT_NOFILE
].rlim_cur
> INT_MAX
)
548 dtsize
= (int)p
->p_rlimit
[RLIMIT_NOFILE
].rlim_cur
;
549 if (dtsize
> maxfilesperproc
)
550 dtsize
= maxfilesperproc
;
551 if (dtsize
< minfilesperproc
)
552 dtsize
= minfilesperproc
;
554 if (new < 0 || new > dtsize
)
555 return (flags
& DUP_FCNTL
? EINVAL
: EBADF
);
557 spin_lock(&fdp
->fd_spin
);
558 if ((unsigned)old
>= fdp
->fd_nfiles
|| fdp
->fd_files
[old
].fp
== NULL
) {
559 spin_unlock(&fdp
->fd_spin
);
562 if ((flags
& DUP_FIXED
) && old
== new) {
564 if (flags
& DUP_CLOEXEC
)
565 fdp
->fd_files
[new].fileflags
|= UF_EXCLOSE
;
566 spin_unlock(&fdp
->fd_spin
);
569 fp
= fdp
->fd_files
[old
].fp
;
570 oldflags
= fdp
->fd_files
[old
].fileflags
;
574 * Allocate a new descriptor if DUP_VARIABLE, or expand the table
575 * if the requested descriptor is beyond the current table size.
577 * This can block. Retry if the source descriptor no longer matches
578 * or if our expectation in the expansion case races.
580 * If we are not expanding or allocating a new decriptor, then reset
581 * the target descriptor to a reserved state so we have a uniform
582 * setup for the next code block.
584 if ((flags
& DUP_VARIABLE
) || new >= fdp
->fd_nfiles
) {
585 spin_unlock(&fdp
->fd_spin
);
586 error
= fdalloc(p
, new, &newfd
);
587 spin_lock(&fdp
->fd_spin
);
589 spin_unlock(&fdp
->fd_spin
);
596 if (old
>= fdp
->fd_nfiles
|| fdp
->fd_files
[old
].fp
!= fp
) {
597 fsetfd_locked(fdp
, NULL
, newfd
);
598 spin_unlock(&fdp
->fd_spin
);
603 * Check for expansion race
605 if ((flags
& DUP_VARIABLE
) == 0 && new != newfd
) {
606 fsetfd_locked(fdp
, NULL
, newfd
);
607 spin_unlock(&fdp
->fd_spin
);
612 * Check for ripout, newfd reused old (this case probably
616 fsetfd_locked(fdp
, NULL
, newfd
);
617 spin_unlock(&fdp
->fd_spin
);
624 if (fdp
->fd_files
[new].reserved
) {
625 spin_unlock(&fdp
->fd_spin
);
627 kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new);
628 tsleep(fdp
, 0, "fdres", hz
);
633 * If the target descriptor was never allocated we have
634 * to allocate it. If it was we have to clean out the
635 * old descriptor. delfp inherits the ref from the
638 delfp
= fdp
->fd_files
[new].fp
;
639 fdp
->fd_files
[new].fp
= NULL
;
640 fdp
->fd_files
[new].reserved
= 1;
642 fdreserve_locked(fdp
, new, 1);
643 if (new > fdp
->fd_lastfile
)
644 fdp
->fd_lastfile
= new;
650 * NOTE: still holding an exclusive spinlock
654 * If a descriptor is being overwritten we may hve to tell
655 * fdfree() to sleep to ensure that all relevant process
656 * leaders can be traversed in closef().
658 if (delfp
!= NULL
&& p
->p_fdtol
!= NULL
) {
659 fdp
->fd_holdleaderscount
++;
664 KASSERT(delfp
== NULL
|| (flags
& DUP_FIXED
),
665 ("dup() picked an open file"));
668 * Duplicate the source descriptor, update lastfile. If the new
669 * descriptor was not allocated and we aren't replacing an existing
670 * descriptor we have to mark the descriptor as being in use.
672 * The fd_files[] array inherits fp's hold reference.
674 fsetfd_locked(fdp
, fp
, new);
675 if ((flags
& DUP_CLOEXEC
) != 0)
676 fdp
->fd_files
[new].fileflags
= oldflags
| UF_EXCLOSE
;
678 fdp
->fd_files
[new].fileflags
= oldflags
& ~UF_EXCLOSE
;
679 spin_unlock(&fdp
->fd_spin
);
684 * If we dup'd over a valid file, we now own the reference to it
685 * and must dispose of it using closef() semantics (as if a
686 * close() were performed on it).
689 if (SLIST_FIRST(&delfp
->f_klist
))
690 knote_fdclose(delfp
, fdp
, new);
693 spin_lock(&fdp
->fd_spin
);
694 fdp
->fd_holdleaderscount
--;
695 if (fdp
->fd_holdleaderscount
== 0 &&
696 fdp
->fd_holdleaderswakeup
!= 0) {
697 fdp
->fd_holdleaderswakeup
= 0;
698 spin_unlock(&fdp
->fd_spin
);
699 wakeup(&fdp
->fd_holdleaderscount
);
701 spin_unlock(&fdp
->fd_spin
);
709 * If sigio is on the list associated with a process or process group,
710 * disable signalling from the device, remove sigio from the list and
714 funsetown(struct sigio
**sigiop
)
720 if ((sigio
= *sigiop
) != NULL
) {
721 lwkt_gettoken(&sigio_token
); /* protect sigio */
722 KKASSERT(sigiop
== sigio
->sio_myref
);
725 lwkt_reltoken(&sigio_token
);
730 if (sigio
->sio_pgid
< 0) {
731 pgrp
= sigio
->sio_pgrp
;
732 sigio
->sio_pgrp
= NULL
;
733 lwkt_gettoken(&pgrp
->pg_token
);
734 SLIST_REMOVE(&pgrp
->pg_sigiolst
, sigio
, sigio
, sio_pgsigio
);
735 lwkt_reltoken(&pgrp
->pg_token
);
737 } else /* if ((*sigiop)->sio_pgid > 0) */ {
739 sigio
->sio_proc
= NULL
;
741 lwkt_gettoken(&p
->p_token
);
742 SLIST_REMOVE(&p
->p_sigiolst
, sigio
, sigio
, sio_pgsigio
);
743 lwkt_reltoken(&p
->p_token
);
746 crfree(sigio
->sio_ucred
);
747 sigio
->sio_ucred
= NULL
;
748 kfree(sigio
, M_SIGIO
);
752 * Free a list of sigio structures. Caller is responsible for ensuring
753 * that the list is MPSAFE.
756 funsetownlst(struct sigiolst
*sigiolst
)
760 while ((sigio
= SLIST_FIRST(sigiolst
)) != NULL
)
761 funsetown(sigio
->sio_myref
);
765 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
767 * After permission checking, add a sigio structure to the sigio list for
768 * the process or process group.
771 fsetown(pid_t pgid
, struct sigio
**sigiop
)
773 struct proc
*proc
= NULL
;
774 struct pgrp
*pgrp
= NULL
;
791 * Policy - Don't allow a process to FSETOWN a process
792 * in another session.
794 * Remove this test to allow maximum flexibility or
795 * restrict FSETOWN to the current process or process
796 * group for maximum safety.
798 if (proc
->p_session
!= curproc
->p_session
) {
802 } else /* if (pgid < 0) */ {
803 pgrp
= pgfind(-pgid
);
810 * Policy - Don't allow a process to FSETOWN a process
811 * in another session.
813 * Remove this test to allow maximum flexibility or
814 * restrict FSETOWN to the current process or process
815 * group for maximum safety.
817 if (pgrp
->pg_session
!= curproc
->p_session
) {
822 sigio
= kmalloc(sizeof(struct sigio
), M_SIGIO
, M_WAITOK
| M_ZERO
);
824 KKASSERT(pgrp
== NULL
);
825 lwkt_gettoken(&proc
->p_token
);
826 SLIST_INSERT_HEAD(&proc
->p_sigiolst
, sigio
, sio_pgsigio
);
827 sigio
->sio_proc
= proc
;
828 lwkt_reltoken(&proc
->p_token
);
830 KKASSERT(proc
== NULL
);
831 lwkt_gettoken(&pgrp
->pg_token
);
832 SLIST_INSERT_HEAD(&pgrp
->pg_sigiolst
, sigio
, sio_pgsigio
);
833 sigio
->sio_pgrp
= pgrp
;
834 lwkt_reltoken(&pgrp
->pg_token
);
837 sigio
->sio_pgid
= pgid
;
838 sigio
->sio_ucred
= crhold(curthread
->td_ucred
);
839 /* It would be convenient if p_ruid was in ucred. */
840 sigio
->sio_ruid
= sigio
->sio_ucred
->cr_ruid
;
841 sigio
->sio_myref
= sigiop
;
843 lwkt_gettoken(&sigio_token
);
847 lwkt_reltoken(&sigio_token
);
858 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
861 fgetown(struct sigio
**sigiop
)
866 lwkt_gettoken_shared(&sigio_token
);
868 own
= (sigio
!= NULL
? sigio
->sio_pgid
: 0);
869 lwkt_reltoken(&sigio_token
);
875 * Close many file descriptors.
878 sys_closefrom(struct closefrom_args
*uap
)
880 return(kern_closefrom(uap
->fd
));
884 * Close all file descriptors greater then or equal to fd
887 kern_closefrom(int fd
)
889 struct thread
*td
= curthread
;
890 struct proc
*p
= td
->td_proc
;
891 struct filedesc
*fdp
;
900 * NOTE: This function will skip unassociated descriptors and
901 * reserved descriptors that have not yet been assigned.
902 * fd_lastfile can change as a side effect of kern_close().
904 spin_lock(&fdp
->fd_spin
);
905 while (fd
<= fdp
->fd_lastfile
) {
906 if (fdp
->fd_files
[fd
].fp
!= NULL
) {
907 spin_unlock(&fdp
->fd_spin
);
908 /* ok if this races another close */
909 if (kern_close(fd
) == EINTR
)
911 spin_lock(&fdp
->fd_spin
);
915 spin_unlock(&fdp
->fd_spin
);
920 * Close a file descriptor.
923 sys_close(struct close_args
*uap
)
925 return(kern_close(uap
->fd
));
934 struct thread
*td
= curthread
;
935 struct proc
*p
= td
->td_proc
;
936 struct filedesc
*fdp
;
944 spin_lock(&fdp
->fd_spin
);
945 if ((fp
= funsetfd_locked(fdp
, fd
)) == NULL
) {
946 spin_unlock(&fdp
->fd_spin
);
950 if (p
->p_fdtol
!= NULL
) {
952 * Ask fdfree() to sleep to ensure that all relevant
953 * process leaders can be traversed in closef().
955 fdp
->fd_holdleaderscount
++;
960 * we now hold the fp reference that used to be owned by the descriptor
963 spin_unlock(&fdp
->fd_spin
);
964 if (SLIST_FIRST(&fp
->f_klist
))
965 knote_fdclose(fp
, fdp
, fd
);
966 error
= closef(fp
, p
);
968 spin_lock(&fdp
->fd_spin
);
969 fdp
->fd_holdleaderscount
--;
970 if (fdp
->fd_holdleaderscount
== 0 &&
971 fdp
->fd_holdleaderswakeup
!= 0) {
972 fdp
->fd_holdleaderswakeup
= 0;
973 spin_unlock(&fdp
->fd_spin
);
974 wakeup(&fdp
->fd_holdleaderscount
);
976 spin_unlock(&fdp
->fd_spin
);
983 * shutdown_args(int fd, int how)
986 kern_shutdown(int fd
, int how
)
988 struct thread
*td
= curthread
;
989 struct proc
*p
= td
->td_proc
;
995 if ((fp
= holdfp(p
->p_fd
, fd
, -1)) == NULL
)
997 error
= fo_shutdown(fp
, how
);
1007 sys_shutdown(struct shutdown_args
*uap
)
1011 error
= kern_shutdown(uap
->s
, uap
->how
);
1020 kern_fstat(int fd
, struct stat
*ub
)
1022 struct thread
*td
= curthread
;
1023 struct proc
*p
= td
->td_proc
;
1029 if ((fp
= holdfp(p
->p_fd
, fd
, -1)) == NULL
)
1031 error
= fo_stat(fp
, ub
, td
->td_ucred
);
1038 * Return status information about a file descriptor.
1041 sys_fstat(struct fstat_args
*uap
)
1046 error
= kern_fstat(uap
->fd
, &st
);
1049 error
= copyout(&st
, uap
->sb
, sizeof(st
));
1054 * Return pathconf information about a file descriptor.
1059 sys_fpathconf(struct fpathconf_args
*uap
)
1061 struct thread
*td
= curthread
;
1062 struct proc
*p
= td
->td_proc
;
1067 if ((fp
= holdfp(p
->p_fd
, uap
->fd
, -1)) == NULL
)
1070 switch (fp
->f_type
) {
1073 if (uap
->name
!= _PC_PIPE_BUF
) {
1076 uap
->sysmsg_result
= PIPE_BUF
;
1082 vp
= (struct vnode
*)fp
->f_data
;
1083 error
= VOP_PATHCONF(vp
, uap
->name
, &uap
->sysmsg_reg
);
1093 static int fdexpand
;
1094 SYSCTL_INT(_debug
, OID_AUTO
, fdexpand
, CTLFLAG_RD
, &fdexpand
, 0,
1095 "Number of times a file table has been expanded");
1098 * Grow the file table so it can hold through descriptor (want).
1100 * The fdp's spinlock must be held exclusively on entry and may be held
1101 * exclusively on return. The spinlock may be cycled by the routine.
1104 fdgrow_locked(struct filedesc
*fdp
, int want
)
1106 struct fdnode
*newfiles
;
1107 struct fdnode
*oldfiles
;
1110 nf
= fdp
->fd_nfiles
;
1112 /* nf has to be of the form 2^n - 1 */
1114 } while (nf
<= want
);
1116 spin_unlock(&fdp
->fd_spin
);
1117 newfiles
= kmalloc(nf
* sizeof(struct fdnode
), M_FILEDESC
, M_WAITOK
);
1118 spin_lock(&fdp
->fd_spin
);
1121 * We could have raced another extend while we were not holding
1124 if (fdp
->fd_nfiles
>= nf
) {
1125 spin_unlock(&fdp
->fd_spin
);
1126 kfree(newfiles
, M_FILEDESC
);
1127 spin_lock(&fdp
->fd_spin
);
1131 * Copy the existing ofile and ofileflags arrays
1132 * and zero the new portion of each array.
1134 extra
= nf
- fdp
->fd_nfiles
;
1135 bcopy(fdp
->fd_files
, newfiles
, fdp
->fd_nfiles
* sizeof(struct fdnode
));
1136 bzero(&newfiles
[fdp
->fd_nfiles
], extra
* sizeof(struct fdnode
));
1138 oldfiles
= fdp
->fd_files
;
1139 fdp
->fd_files
= newfiles
;
1140 fdp
->fd_nfiles
= nf
;
1142 if (oldfiles
!= fdp
->fd_builtin_files
) {
1143 spin_unlock(&fdp
->fd_spin
);
1144 kfree(oldfiles
, M_FILEDESC
);
1145 spin_lock(&fdp
->fd_spin
);
1151 * Number of nodes in right subtree, including the root.
1154 right_subtree_size(int n
)
1156 return (n
^ (n
| (n
+ 1)));
1163 right_ancestor(int n
)
1165 return (n
| (n
+ 1));
1172 left_ancestor(int n
)
1174 return ((n
& (n
+ 1)) - 1);
1178 * Traverse the in-place binary tree buttom-up adjusting the allocation
1179 * count so scans can determine where free descriptors are located.
1181 * caller must be holding an exclusive spinlock on fdp
1185 fdreserve_locked(struct filedesc
*fdp
, int fd
, int incr
)
1188 fdp
->fd_files
[fd
].allocated
+= incr
;
1189 KKASSERT(fdp
->fd_files
[fd
].allocated
>= 0);
1190 fd
= left_ancestor(fd
);
1195 * Reserve a file descriptor for the process. If no error occurs, the
1196 * caller MUST at some point call fsetfd() or assign a file pointer
1197 * or dispose of the reservation.
1200 fdalloc(struct proc
*p
, int want
, int *result
)
1202 struct filedesc
*fdp
= p
->p_fd
;
1203 struct uidinfo
*uip
;
1204 int fd
, rsize
, rsum
, node
, lim
;
1207 * Check dtable size limit
1209 spin_lock(&p
->p_limit
->p_spin
);
1210 if (p
->p_rlimit
[RLIMIT_NOFILE
].rlim_cur
> INT_MAX
)
1213 lim
= (int)p
->p_rlimit
[RLIMIT_NOFILE
].rlim_cur
;
1214 spin_unlock(&p
->p_limit
->p_spin
);
1216 if (lim
> maxfilesperproc
)
1217 lim
= maxfilesperproc
;
1218 if (lim
< minfilesperproc
)
1219 lim
= minfilesperproc
;
1224 * Check that the user has not run out of descriptors (non-root only).
1225 * As a safety measure the dtable is allowed to have at least
1226 * minfilesperproc open fds regardless of the maxfilesperuser limit.
1228 if (p
->p_ucred
->cr_uid
&& fdp
->fd_nfiles
>= minfilesperproc
) {
1229 uip
= p
->p_ucred
->cr_uidinfo
;
1230 if (uip
->ui_openfiles
> maxfilesperuser
) {
1231 krateprintf(&krate_uidinfo
,
1232 "Warning: user %d pid %d (%s) ran out of "
1233 "file descriptors (%d/%d)\n",
1234 p
->p_ucred
->cr_uid
, (int)p
->p_pid
,
1236 uip
->ui_openfiles
, maxfilesperuser
);
1242 * Grow the dtable if necessary
1244 spin_lock(&fdp
->fd_spin
);
1245 if (want
>= fdp
->fd_nfiles
)
1246 fdgrow_locked(fdp
, want
);
1249 * Search for a free descriptor starting at the higher
1250 * of want or fd_freefile. If that fails, consider
1251 * expanding the ofile array.
1253 * NOTE! the 'allocated' field is a cumulative recursive allocation
1254 * count. If we happen to see a value of 0 then we can shortcut
1255 * our search. Otherwise we run through through the tree going
1256 * down branches we know have free descriptor(s) until we hit a
1257 * leaf node. The leaf node will be free but will not necessarily
1258 * have an allocated field of 0.
1261 /* move up the tree looking for a subtree with a free node */
1262 for (fd
= max(want
, fdp
->fd_freefile
); fd
< min(fdp
->fd_nfiles
, lim
);
1263 fd
= right_ancestor(fd
)) {
1264 if (fdp
->fd_files
[fd
].allocated
== 0)
1267 rsize
= right_subtree_size(fd
);
1268 if (fdp
->fd_files
[fd
].allocated
== rsize
)
1269 continue; /* right subtree full */
1272 * Free fd is in the right subtree of the tree rooted at fd.
1273 * Call that subtree R. Look for the smallest (leftmost)
1274 * subtree of R with an unallocated fd: continue moving
1275 * down the left branch until encountering a full left
1276 * subtree, then move to the right.
1278 for (rsum
= 0, rsize
/= 2; rsize
> 0; rsize
/= 2) {
1280 rsum
+= fdp
->fd_files
[node
].allocated
;
1281 if (fdp
->fd_files
[fd
].allocated
== rsum
+ rsize
) {
1282 fd
= node
; /* move to the right */
1283 if (fdp
->fd_files
[node
].allocated
== 0)
1292 * No space in current array. Expand?
1294 if (fdp
->fd_nfiles
>= lim
) {
1295 spin_unlock(&fdp
->fd_spin
);
1298 fdgrow_locked(fdp
, want
);
1302 KKASSERT(fd
< fdp
->fd_nfiles
);
1303 if (fd
> fdp
->fd_lastfile
)
1304 fdp
->fd_lastfile
= fd
;
1305 if (want
<= fdp
->fd_freefile
)
1306 fdp
->fd_freefile
= fd
;
1308 KKASSERT(fdp
->fd_files
[fd
].fp
== NULL
);
1309 KKASSERT(fdp
->fd_files
[fd
].reserved
== 0);
1310 fdp
->fd_files
[fd
].fileflags
= 0;
1311 fdp
->fd_files
[fd
].reserved
= 1;
1312 fdreserve_locked(fdp
, fd
, 1);
1313 spin_unlock(&fdp
->fd_spin
);
1318 * Check to see whether n user file descriptors
1319 * are available to the process p.
1322 fdavail(struct proc
*p
, int n
)
1324 struct filedesc
*fdp
= p
->p_fd
;
1325 struct fdnode
*fdnode
;
1328 spin_lock(&p
->p_limit
->p_spin
);
1329 if (p
->p_rlimit
[RLIMIT_NOFILE
].rlim_cur
> INT_MAX
)
1332 lim
= (int)p
->p_rlimit
[RLIMIT_NOFILE
].rlim_cur
;
1333 spin_unlock(&p
->p_limit
->p_spin
);
1335 if (lim
> maxfilesperproc
)
1336 lim
= maxfilesperproc
;
1337 if (lim
< minfilesperproc
)
1338 lim
= minfilesperproc
;
1340 spin_lock(&fdp
->fd_spin
);
1341 if ((i
= lim
- fdp
->fd_nfiles
) > 0 && (n
-= i
) <= 0) {
1342 spin_unlock(&fdp
->fd_spin
);
1345 last
= min(fdp
->fd_nfiles
, lim
);
1346 fdnode
= &fdp
->fd_files
[fdp
->fd_freefile
];
1347 for (i
= last
- fdp
->fd_freefile
; --i
>= 0; ++fdnode
) {
1348 if (fdnode
->fp
== NULL
&& --n
<= 0) {
1349 spin_unlock(&fdp
->fd_spin
);
1353 spin_unlock(&fdp
->fd_spin
);
1358 * Revoke open descriptors referencing (f_data, f_type)
1360 * Any revoke executed within a prison is only able to
1361 * revoke descriptors for processes within that prison.
1363 * Returns 0 on success or an error code.
1365 struct fdrevoke_info
{
1374 static int fdrevoke_check_callback(struct file
*fp
, void *vinfo
);
1375 static int fdrevoke_proc_callback(struct proc
*p
, void *vinfo
);
1378 fdrevoke(void *f_data
, short f_type
, struct ucred
*cred
)
1380 struct fdrevoke_info info
;
1383 bzero(&info
, sizeof(info
));
1387 error
= falloc(NULL
, &info
.nfp
, NULL
);
1392 * Scan the file pointer table once. dups do not dup file pointers,
1393 * only descriptors, so there is no leak. Set FREVOKED on the fps
1396 * Any fps sent over unix-domain sockets will be revoked by the
1397 * socket code checking for FREVOKED when the fps are externialized.
1398 * revoke_token is used to make sure that fps marked FREVOKED and
1399 * externalized will be picked up by the following allproc_scan().
1401 lwkt_gettoken(&revoke_token
);
1402 allfiles_scan_exclusive(fdrevoke_check_callback
, &info
);
1403 lwkt_reltoken(&revoke_token
);
1406 * If any fps were marked track down the related descriptors
1407 * and close them. Any dup()s at this point will notice
1408 * the FREVOKED already set in the fp and do the right thing.
1411 allproc_scan(fdrevoke_proc_callback
, &info
);
1417 * Locate matching file pointers directly.
1419 * WARNING: allfiles_scan_exclusive() holds a spinlock through these calls!
1422 fdrevoke_check_callback(struct file
*fp
, void *vinfo
)
1424 struct fdrevoke_info
*info
= vinfo
;
1427 * File pointers already flagged for revokation are skipped.
1429 if (fp
->f_flag
& FREVOKED
)
1433 * If revoking from a prison file pointers created outside of
1434 * that prison, or file pointers without creds, cannot be revoked.
1436 if (info
->cred
->cr_prison
&&
1437 (fp
->f_cred
== NULL
||
1438 info
->cred
->cr_prison
!= fp
->f_cred
->cr_prison
)) {
1443 * If the file pointer matches then mark it for revocation. The
1444 * flag is currently only used by unp_revoke_gc().
1446 * info->found is a heuristic and can race in a SMP environment.
1448 if (info
->data
== fp
->f_data
&& info
->type
== fp
->f_type
) {
1449 atomic_set_int(&fp
->f_flag
, FREVOKED
);
1456 * Locate matching file pointers via process descriptor tables.
1459 fdrevoke_proc_callback(struct proc
*p
, void *vinfo
)
1461 struct fdrevoke_info
*info
= vinfo
;
1462 struct filedesc
*fdp
;
1466 if (p
->p_stat
== SIDL
|| p
->p_stat
== SZOMB
)
1468 if (info
->cred
->cr_prison
&&
1469 info
->cred
->cr_prison
!= p
->p_ucred
->cr_prison
) {
1474 * If the controlling terminal of the process matches the
1475 * vnode being revoked we clear the controlling terminal.
1477 * The normal spec_close() may not catch this because it
1478 * uses curproc instead of p.
1480 if (p
->p_session
&& info
->type
== DTYPE_VNODE
&&
1481 info
->data
== p
->p_session
->s_ttyvp
) {
1482 p
->p_session
->s_ttyvp
= NULL
;
1487 * Softref the fdp to prevent it from being destroyed
1489 spin_lock(&p
->p_spin
);
1490 if ((fdp
= p
->p_fd
) == NULL
) {
1491 spin_unlock(&p
->p_spin
);
1494 atomic_add_int(&fdp
->fd_softrefs
, 1);
1495 spin_unlock(&p
->p_spin
);
1498 * Locate and close any matching file descriptors.
1500 spin_lock(&fdp
->fd_spin
);
1501 for (n
= 0; n
< fdp
->fd_nfiles
; ++n
) {
1502 if ((fp
= fdp
->fd_files
[n
].fp
) == NULL
)
1504 if (fp
->f_flag
& FREVOKED
) {
1506 fdp
->fd_files
[n
].fp
= info
->nfp
;
1507 spin_unlock(&fdp
->fd_spin
);
1508 knote_fdclose(fp
, fdp
, n
); /* XXX */
1510 spin_lock(&fdp
->fd_spin
);
1513 spin_unlock(&fdp
->fd_spin
);
1514 atomic_subtract_int(&fdp
->fd_softrefs
, 1);
1520 * Create a new open file structure and reserve a file decriptor
1521 * for the process that refers to it.
1523 * Root creds are checked using lp, or assumed if lp is NULL. If
1524 * resultfd is non-NULL then lp must also be non-NULL. No file
1525 * descriptor is reserved (and no process context is needed) if
1528 * A file pointer with a refcount of 1 is returned. Note that the
1529 * file pointer is NOT associated with the descriptor. If falloc
1530 * returns success, fsetfd() MUST be called to either associate the
1531 * file pointer or clear the reservation.
1534 falloc(struct lwp
*lp
, struct file
**resultfp
, int *resultfd
)
1536 static struct timeval lastfail
;
1538 struct filelist_head
*head
;
1540 struct ucred
*cred
= lp
? lp
->lwp_thread
->td_ucred
: proc0
.p_ucred
;
1546 * Handle filetable full issues and root overfill.
1548 if (nfiles
>= maxfiles
- maxfilesrootres
&&
1549 (cred
->cr_ruid
!= 0 || nfiles
>= maxfiles
)) {
1550 if (ppsratecheck(&lastfail
, &curfail
, 1)) {
1551 kprintf("kern.maxfiles limit exceeded by uid %d, "
1552 "please see tuning(7).\n",
1560 * Allocate a new file descriptor.
1562 fp
= objcache_get(file_objcache
, M_WAITOK
);
1563 bzero(fp
, sizeof(*fp
));
1564 spin_init(&fp
->f_spin
, "falloc");
1565 SLIST_INIT(&fp
->f_klist
);
1567 fp
->f_ops
= &badfileops
;
1570 atomic_add_int(&nfiles
, 1);
1572 head
= fp2filelist(fp
);
1573 spin_lock(&head
->spin
);
1574 LIST_INSERT_HEAD(&head
->list
, fp
, f_list
);
1575 spin_unlock(&head
->spin
);
1578 if ((error
= fdalloc(lp
->lwp_proc
, 0, resultfd
)) != 0) {
1591 * Check for races against a file descriptor by determining that the
1592 * file pointer is still associated with the specified file descriptor,
1593 * and a close is not currently in progress.
1596 checkfdclosed(struct filedesc
*fdp
, int fd
, struct file
*fp
)
1600 spin_lock_shared(&fdp
->fd_spin
);
1601 if ((unsigned)fd
>= fdp
->fd_nfiles
|| fp
!= fdp
->fd_files
[fd
].fp
)
1605 spin_unlock_shared(&fdp
->fd_spin
);
1610 * Associate a file pointer with a previously reserved file descriptor.
1611 * This function always succeeds.
1613 * If fp is NULL, the file descriptor is returned to the pool.
1617 * (exclusive spinlock must be held on call)
1620 fsetfd_locked(struct filedesc
*fdp
, struct file
*fp
, int fd
)
1622 KKASSERT((unsigned)fd
< fdp
->fd_nfiles
);
1623 KKASSERT(fdp
->fd_files
[fd
].reserved
!= 0);
1626 fdp
->fd_files
[fd
].fp
= fp
;
1627 fdp
->fd_files
[fd
].reserved
= 0;
1629 fdp
->fd_files
[fd
].reserved
= 0;
1630 fdreserve_locked(fdp
, fd
, -1);
1631 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 * (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
);
1665 * WARNING: May not be called before initial fsetfd().
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
);
1686 * WARNING: May not be called before initial fsetfd().
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
);
1707 * WARNING: May not be called before initial fsetfd().
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!"));
1763 if (fp
->f_nchandle
.ncp
)
1764 cache_drop(&fp
->f_nchandle
);
1765 objcache_put(file_objcache
, fp
);
1769 * called from init_main, initialize filedesc0 for proc0.
1772 fdinit_bootstrap(struct proc
*p0
, struct filedesc
*fdp0
, int cmask
)
1776 fdp0
->fd_refcnt
= 1;
1777 fdp0
->fd_cmask
= cmask
;
1778 fdp0
->fd_files
= fdp0
->fd_builtin_files
;
1779 fdp0
->fd_nfiles
= NDFILE
;
1780 fdp0
->fd_lastfile
= -1;
1781 spin_init(&fdp0
->fd_spin
, "fdinitbootstrap");
1785 * Build a new filedesc structure.
1788 fdinit(struct proc
*p
)
1790 struct filedesc
*newfdp
;
1791 struct filedesc
*fdp
= p
->p_fd
;
1793 newfdp
= kmalloc(sizeof(struct filedesc
), M_FILEDESC
, M_WAITOK
|M_ZERO
);
1794 spin_lock(&fdp
->fd_spin
);
1796 newfdp
->fd_cdir
= fdp
->fd_cdir
;
1797 vref(newfdp
->fd_cdir
);
1798 cache_copy(&fdp
->fd_ncdir
, &newfdp
->fd_ncdir
);
1802 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of
1803 * proc0, but should unconditionally exist in other processes.
1806 newfdp
->fd_rdir
= fdp
->fd_rdir
;
1807 vref(newfdp
->fd_rdir
);
1808 cache_copy(&fdp
->fd_nrdir
, &newfdp
->fd_nrdir
);
1811 newfdp
->fd_jdir
= fdp
->fd_jdir
;
1812 vref(newfdp
->fd_jdir
);
1813 cache_copy(&fdp
->fd_njdir
, &newfdp
->fd_njdir
);
1815 spin_unlock(&fdp
->fd_spin
);
1817 /* Create the file descriptor table. */
1818 newfdp
->fd_refcnt
= 1;
1819 newfdp
->fd_cmask
= cmask
;
1820 newfdp
->fd_files
= newfdp
->fd_builtin_files
;
1821 newfdp
->fd_nfiles
= NDFILE
;
1822 newfdp
->fd_lastfile
= -1;
1823 spin_init(&newfdp
->fd_spin
, "fdinit");
1829 * Share a filedesc structure.
1832 fdshare(struct proc
*p
)
1834 struct filedesc
*fdp
;
1837 spin_lock(&fdp
->fd_spin
);
1839 spin_unlock(&fdp
->fd_spin
);
1844 * Copy a filedesc structure.
1847 fdcopy(struct proc
*p
, struct filedesc
**fpp
)
1849 struct filedesc
*fdp
= p
->p_fd
;
1850 struct filedesc
*newfdp
;
1851 struct fdnode
*fdnode
;
1856 * Certain daemons might not have file descriptors.
1862 * Allocate the new filedesc and fd_files[] array. This can race
1863 * with operations by other threads on the fdp so we have to be
1866 newfdp
= kmalloc(sizeof(struct filedesc
),
1867 M_FILEDESC
, M_WAITOK
| M_ZERO
| M_NULLOK
);
1868 if (newfdp
== NULL
) {
1873 spin_lock(&fdp
->fd_spin
);
1874 if (fdp
->fd_lastfile
< NDFILE
) {
1875 newfdp
->fd_files
= newfdp
->fd_builtin_files
;
1879 * We have to allocate (N^2-1) entries for our in-place
1880 * binary tree. Allow the table to shrink.
1884 while (ni
> fdp
->fd_lastfile
&& ni
> NDFILE
) {
1888 spin_unlock(&fdp
->fd_spin
);
1889 newfdp
->fd_files
= kmalloc(i
* sizeof(struct fdnode
),
1890 M_FILEDESC
, M_WAITOK
| M_ZERO
);
1893 * Check for race, retry
1895 spin_lock(&fdp
->fd_spin
);
1896 if (i
<= fdp
->fd_lastfile
) {
1897 spin_unlock(&fdp
->fd_spin
);
1898 kfree(newfdp
->fd_files
, M_FILEDESC
);
1904 * Dup the remaining fields. vref() and cache_hold() can be
1905 * safely called while holding the read spinlock on fdp.
1907 * The read spinlock on fdp is still being held.
1909 * NOTE: vref and cache_hold calls for the case where the vnode
1910 * or cache entry already has at least one ref may be called
1911 * while holding spin locks.
1913 if ((newfdp
->fd_cdir
= fdp
->fd_cdir
) != NULL
) {
1914 vref(newfdp
->fd_cdir
);
1915 cache_copy(&fdp
->fd_ncdir
, &newfdp
->fd_ncdir
);
1918 * We must check for fd_rdir here, at least for now because
1919 * the init process is created before we have access to the
1920 * rootvode to take a reference to it.
1922 if ((newfdp
->fd_rdir
= fdp
->fd_rdir
) != NULL
) {
1923 vref(newfdp
->fd_rdir
);
1924 cache_copy(&fdp
->fd_nrdir
, &newfdp
->fd_nrdir
);
1926 if ((newfdp
->fd_jdir
= fdp
->fd_jdir
) != NULL
) {
1927 vref(newfdp
->fd_jdir
);
1928 cache_copy(&fdp
->fd_njdir
, &newfdp
->fd_njdir
);
1930 newfdp
->fd_refcnt
= 1;
1931 newfdp
->fd_nfiles
= i
;
1932 newfdp
->fd_lastfile
= fdp
->fd_lastfile
;
1933 newfdp
->fd_freefile
= fdp
->fd_freefile
;
1934 newfdp
->fd_cmask
= fdp
->fd_cmask
;
1935 spin_init(&newfdp
->fd_spin
, "fdcopy");
1938 * Copy the descriptor table through (i). This also copies the
1939 * allocation state. Then go through and ref the file pointers
1940 * and clean up any KQ descriptors.
1942 * kq descriptors cannot be copied. Since we haven't ref'd the
1943 * copied files yet we can ignore the return value from funsetfd().
1945 * The read spinlock on fdp is still being held.
1947 bcopy(fdp
->fd_files
, newfdp
->fd_files
, i
* sizeof(struct fdnode
));
1948 for (i
= 0 ; i
< newfdp
->fd_nfiles
; ++i
) {
1949 fdnode
= &newfdp
->fd_files
[i
];
1950 if (fdnode
->reserved
) {
1951 fdreserve_locked(newfdp
, i
, -1);
1952 fdnode
->reserved
= 0;
1953 fdfixup_locked(newfdp
, i
);
1954 } else if (fdnode
->fp
) {
1955 if (fdnode
->fp
->f_type
== DTYPE_KQUEUE
) {
1956 (void)funsetfd_locked(newfdp
, i
);
1962 spin_unlock(&fdp
->fd_spin
);
1968 * Release a filedesc structure.
1970 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE)
1973 fdfree(struct proc
*p
, struct filedesc
*repl
)
1975 struct filedesc
*fdp
;
1976 struct fdnode
*fdnode
;
1978 struct filedesc_to_leader
*fdtol
;
1984 * Certain daemons might not have file descriptors.
1993 * Severe messing around to follow.
1995 spin_lock(&fdp
->fd_spin
);
1997 /* Check for special need to clear POSIX style locks */
1999 if (fdtol
!= NULL
) {
2000 KASSERT(fdtol
->fdl_refcount
> 0,
2001 ("filedesc_to_refcount botch: fdl_refcount=%d",
2002 fdtol
->fdl_refcount
));
2003 if (fdtol
->fdl_refcount
== 1 &&
2004 (p
->p_leader
->p_flags
& P_ADVLOCK
) != 0) {
2005 for (i
= 0; i
<= fdp
->fd_lastfile
; ++i
) {
2006 fdnode
= &fdp
->fd_files
[i
];
2007 if (fdnode
->fp
== NULL
||
2008 fdnode
->fp
->f_type
!= DTYPE_VNODE
) {
2013 spin_unlock(&fdp
->fd_spin
);
2015 lf
.l_whence
= SEEK_SET
;
2018 lf
.l_type
= F_UNLCK
;
2019 vp
= (struct vnode
*)fp
->f_data
;
2020 (void) VOP_ADVLOCK(vp
,
2021 (caddr_t
)p
->p_leader
,
2026 spin_lock(&fdp
->fd_spin
);
2030 if (fdtol
->fdl_refcount
== 1) {
2031 if (fdp
->fd_holdleaderscount
> 0 &&
2032 (p
->p_leader
->p_flags
& P_ADVLOCK
) != 0) {
2034 * close() or do_dup() has cleared a reference
2035 * in a shared file descriptor table.
2037 fdp
->fd_holdleaderswakeup
= 1;
2038 ssleep(&fdp
->fd_holdleaderscount
,
2039 &fdp
->fd_spin
, 0, "fdlhold", 0);
2042 if (fdtol
->fdl_holdcount
> 0) {
2044 * Ensure that fdtol->fdl_leader
2045 * remains valid in closef().
2047 fdtol
->fdl_wakeup
= 1;
2048 ssleep(fdtol
, &fdp
->fd_spin
, 0, "fdlhold", 0);
2052 fdtol
->fdl_refcount
--;
2053 if (fdtol
->fdl_refcount
== 0 &&
2054 fdtol
->fdl_holdcount
== 0) {
2055 fdtol
->fdl_next
->fdl_prev
= fdtol
->fdl_prev
;
2056 fdtol
->fdl_prev
->fdl_next
= fdtol
->fdl_next
;
2061 if (fdtol
!= NULL
) {
2062 spin_unlock(&fdp
->fd_spin
);
2063 kfree(fdtol
, M_FILEDESC_TO_LEADER
);
2064 spin_lock(&fdp
->fd_spin
);
2067 if (--fdp
->fd_refcnt
> 0) {
2068 spin_unlock(&fdp
->fd_spin
);
2069 spin_lock(&p
->p_spin
);
2071 spin_unlock(&p
->p_spin
);
2076 * Even though we are the last reference to the structure allproc
2077 * scans may still reference the structure. Maintain proper
2078 * locks until we can replace p->p_fd.
2080 * Also note that kqueue's closef still needs to reference the
2081 * fdp via p->p_fd, so we have to close the descriptors before
2082 * we replace p->p_fd.
2084 for (i
= 0; i
<= fdp
->fd_lastfile
; ++i
) {
2085 if (fdp
->fd_files
[i
].fp
) {
2086 fp
= funsetfd_locked(fdp
, i
);
2088 spin_unlock(&fdp
->fd_spin
);
2089 if (SLIST_FIRST(&fp
->f_klist
))
2090 knote_fdclose(fp
, fdp
, i
);
2092 spin_lock(&fdp
->fd_spin
);
2096 spin_unlock(&fdp
->fd_spin
);
2099 * Interlock against an allproc scan operations (typically frevoke).
2101 spin_lock(&p
->p_spin
);
2103 spin_unlock(&p
->p_spin
);
2106 * Wait for any softrefs to go away. This race rarely occurs so
2107 * we can use a non-critical-path style poll/sleep loop. The
2108 * race only occurs against allproc scans.
2110 * No new softrefs can occur with the fdp disconnected from the
2113 if (fdp
->fd_softrefs
) {
2114 kprintf("pid %d: Warning, fdp race avoided\n", p
->p_pid
);
2115 while (fdp
->fd_softrefs
)
2116 tsleep(&fdp
->fd_softrefs
, 0, "fdsoft", 1);
2119 if (fdp
->fd_files
!= fdp
->fd_builtin_files
)
2120 kfree(fdp
->fd_files
, M_FILEDESC
);
2122 cache_drop(&fdp
->fd_ncdir
);
2123 vrele(fdp
->fd_cdir
);
2126 cache_drop(&fdp
->fd_nrdir
);
2127 vrele(fdp
->fd_rdir
);
2130 cache_drop(&fdp
->fd_njdir
);
2131 vrele(fdp
->fd_jdir
);
2133 kfree(fdp
, M_FILEDESC
);
2137 * Retrieve and reference the file pointer associated with a descriptor.
2140 holdfp(struct filedesc
*fdp
, int fd
, int flag
)
2144 spin_lock_shared(&fdp
->fd_spin
);
2145 if (((u_int
)fd
) >= fdp
->fd_nfiles
) {
2149 if ((fp
= fdp
->fd_files
[fd
].fp
) == NULL
)
2151 if ((fp
->f_flag
& flag
) == 0 && flag
!= -1) {
2157 spin_unlock_shared(&fdp
->fd_spin
);
2162 * holdsock() - load the struct file pointer associated
2163 * with a socket into *fpp. If an error occurs, non-zero
2164 * will be returned and *fpp will be set to NULL.
2167 holdsock(struct filedesc
*fdp
, int fd
, struct file
**fpp
)
2172 spin_lock_shared(&fdp
->fd_spin
);
2173 if ((unsigned)fd
>= fdp
->fd_nfiles
) {
2178 if ((fp
= fdp
->fd_files
[fd
].fp
) == NULL
) {
2182 if (fp
->f_type
!= DTYPE_SOCKET
) {
2189 spin_unlock_shared(&fdp
->fd_spin
);
2195 * Convert a user file descriptor to a held file pointer.
2198 holdvnode(struct filedesc
*fdp
, int fd
, struct file
**fpp
)
2203 spin_lock_shared(&fdp
->fd_spin
);
2204 if ((unsigned)fd
>= fdp
->fd_nfiles
) {
2209 if ((fp
= fdp
->fd_files
[fd
].fp
) == NULL
) {
2213 if (fp
->f_type
!= DTYPE_VNODE
&& fp
->f_type
!= DTYPE_FIFO
) {
2221 spin_unlock_shared(&fdp
->fd_spin
);
2227 * For setugid programs, we don't want to people to use that setugidness
2228 * to generate error messages which write to a file which otherwise would
2229 * otherwise be off-limits to the process.
2231 * This is a gross hack to plug the hole. A better solution would involve
2232 * a special vop or other form of generalized access control mechanism. We
2233 * go ahead and just reject all procfs file systems accesses as dangerous.
2235 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2236 * sufficient. We also don't for check setugidness since we know we are.
2239 is_unsafe(struct file
*fp
)
2241 if (fp
->f_type
== DTYPE_VNODE
&&
2242 ((struct vnode
*)(fp
->f_data
))->v_tag
== VT_PROCFS
)
2248 * Make this setguid thing safe, if at all possible.
2250 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2253 setugidsafety(struct proc
*p
)
2255 struct filedesc
*fdp
= p
->p_fd
;
2258 /* Certain daemons might not have file descriptors. */
2263 * note: fdp->fd_files may be reallocated out from under us while
2264 * we are blocked in a close. Be careful!
2266 for (i
= 0; i
<= fdp
->fd_lastfile
; i
++) {
2269 if (fdp
->fd_files
[i
].fp
&& is_unsafe(fdp
->fd_files
[i
].fp
)) {
2273 * NULL-out descriptor prior to close to avoid
2274 * a race while close blocks.
2276 if ((fp
= funsetfd_locked(fdp
, i
)) != NULL
) {
2277 knote_fdclose(fp
, fdp
, i
);
2285 * Close any files on exec?
2287 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2290 fdcloseexec(struct proc
*p
)
2292 struct filedesc
*fdp
= p
->p_fd
;
2295 /* Certain daemons might not have file descriptors. */
2300 * We cannot cache fd_files since operations may block and rip
2301 * them out from under us.
2303 for (i
= 0; i
<= fdp
->fd_lastfile
; i
++) {
2304 if (fdp
->fd_files
[i
].fp
!= NULL
&&
2305 (fdp
->fd_files
[i
].fileflags
& UF_EXCLOSE
)) {
2309 * NULL-out descriptor prior to close to avoid
2310 * a race while close blocks.
2312 if ((fp
= funsetfd_locked(fdp
, i
)) != NULL
) {
2313 knote_fdclose(fp
, fdp
, i
);
2321 * It is unsafe for set[ug]id processes to be started with file
2322 * descriptors 0..2 closed, as these descriptors are given implicit
2323 * significance in the Standard C library. fdcheckstd() will create a
2324 * descriptor referencing /dev/null for each of stdin, stdout, and
2325 * stderr that is not already open.
2327 * NOT MPSAFE - calls falloc, vn_open, etc
2330 fdcheckstd(struct lwp
*lp
)
2332 struct nlookupdata nd
;
2333 struct filedesc
*fdp
;
2336 int i
, error
, flags
, devnull
;
2338 fdp
= lp
->lwp_proc
->p_fd
;
2343 for (i
= 0; i
< 3; i
++) {
2344 if (fdp
->fd_files
[i
].fp
!= NULL
)
2347 if ((error
= falloc(lp
, &fp
, &devnull
)) != 0)
2350 error
= nlookup_init(&nd
, "/dev/null", UIO_SYSSPACE
,
2351 NLC_FOLLOW
|NLC_LOCKVP
);
2352 flags
= FREAD
| FWRITE
;
2354 error
= vn_open(&nd
, fp
, flags
, 0);
2356 fsetfd(fdp
, fp
, devnull
);
2358 fsetfd(fdp
, NULL
, devnull
);
2363 KKASSERT(i
== devnull
);
2365 error
= kern_dup(DUP_FIXED
, devnull
, i
, &retval
);
2374 * Internal form of close.
2375 * Decrement reference count on file structure.
2376 * Note: td and/or p may be NULL when closing a file
2377 * that was being passed in a message.
2379 * MPALMOSTSAFE - acquires mplock for VOP operations
2382 closef(struct file
*fp
, struct proc
*p
)
2386 struct filedesc_to_leader
*fdtol
;
2392 * POSIX record locking dictates that any close releases ALL
2393 * locks owned by this process. This is handled by setting
2394 * a flag in the unlock to free ONLY locks obeying POSIX
2395 * semantics, and not to free BSD-style file locks.
2396 * If the descriptor was in a message, POSIX-style locks
2397 * aren't passed with the descriptor.
2399 if (p
!= NULL
&& fp
->f_type
== DTYPE_VNODE
&&
2400 (((struct vnode
*)fp
->f_data
)->v_flag
& VMAYHAVELOCKS
)
2402 if ((p
->p_leader
->p_flags
& P_ADVLOCK
) != 0) {
2403 lf
.l_whence
= SEEK_SET
;
2406 lf
.l_type
= F_UNLCK
;
2407 vp
= (struct vnode
*)fp
->f_data
;
2408 (void) VOP_ADVLOCK(vp
, (caddr_t
)p
->p_leader
, F_UNLCK
,
2412 if (fdtol
!= NULL
) {
2413 lwkt_gettoken(&p
->p_token
);
2415 * Handle special case where file descriptor table
2416 * is shared between multiple process leaders.
2418 for (fdtol
= fdtol
->fdl_next
;
2419 fdtol
!= p
->p_fdtol
;
2420 fdtol
= fdtol
->fdl_next
) {
2421 if ((fdtol
->fdl_leader
->p_flags
&
2424 fdtol
->fdl_holdcount
++;
2425 lf
.l_whence
= SEEK_SET
;
2428 lf
.l_type
= F_UNLCK
;
2429 vp
= (struct vnode
*)fp
->f_data
;
2430 (void) VOP_ADVLOCK(vp
,
2431 (caddr_t
)fdtol
->fdl_leader
,
2432 F_UNLCK
, &lf
, F_POSIX
);
2433 fdtol
->fdl_holdcount
--;
2434 if (fdtol
->fdl_holdcount
== 0 &&
2435 fdtol
->fdl_wakeup
!= 0) {
2436 fdtol
->fdl_wakeup
= 0;
2440 lwkt_reltoken(&p
->p_token
);
2447 * fhold() can only be called if f_count is already at least 1 (i.e. the
2448 * caller of fhold() already has a reference to the file pointer in some
2451 * Atomic ops are used for incrementing and decrementing f_count before
2452 * the 1->0 transition. f_count 1->0 transition is special, see the
2453 * comment in fdrop().
2456 fhold(struct file
*fp
)
2458 /* 0->1 transition will never work */
2459 KASSERT(fp
->f_count
> 0, ("fhold: invalid f_count %d", fp
->f_count
));
2460 atomic_add_int(&fp
->f_count
, 1);
2464 * fdrop() - drop a reference to a descriptor
2467 fdrop(struct file
*fp
)
2471 int error
, do_free
= 0;
2475 * Simple atomic_fetchadd_int(f_count, -1) here will cause use-
2476 * after-free or double free (due to f_count 0->1 transition), if
2477 * fhold() is called on the fps found through filehead iteration.
2480 int count
= fp
->f_count
;
2483 KASSERT(count
> 0, ("fdrop: invalid f_count %d", count
));
2485 struct filelist_head
*head
= fp2filelist(fp
);
2488 * About to drop the last reference, hold the
2489 * filehead spin lock and drop it, so that no
2490 * one could see this fp through filehead anymore,
2491 * let alone fhold() this fp.
2493 spin_lock(&head
->spin
);
2494 if (atomic_cmpset_int(&fp
->f_count
, count
, 0)) {
2495 LIST_REMOVE(fp
, f_list
);
2496 spin_unlock(&head
->spin
);
2497 atomic_subtract_int(&nfiles
, 1);
2498 do_free
= 1; /* free this fp */
2501 spin_unlock(&head
->spin
);
2503 } else if (atomic_cmpset_int(&fp
->f_count
, count
, count
- 1)) {
2511 KKASSERT(SLIST_FIRST(&fp
->f_klist
) == NULL
);
2514 * The last reference has gone away, we own the fp structure free
2517 if (fp
->f_count
< 0)
2518 panic("fdrop: count < 0");
2519 if ((fp
->f_flag
& FHASLOCK
) && fp
->f_type
== DTYPE_VNODE
&&
2520 (((struct vnode
*)fp
->f_data
)->v_flag
& VMAYHAVELOCKS
)
2522 lf
.l_whence
= SEEK_SET
;
2525 lf
.l_type
= F_UNLCK
;
2526 vp
= (struct vnode
*)fp
->f_data
;
2527 (void) VOP_ADVLOCK(vp
, (caddr_t
)fp
, F_UNLCK
, &lf
, 0);
2529 if (fp
->f_ops
!= &badfileops
)
2530 error
= fo_close(fp
);
2538 * Apply an advisory lock on a file descriptor.
2540 * Just attempt to get a record lock of the requested type on
2541 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2546 sys_flock(struct flock_args
*uap
)
2548 struct proc
*p
= curproc
;
2554 if ((fp
= holdfp(p
->p_fd
, uap
->fd
, -1)) == NULL
)
2556 if (fp
->f_type
!= DTYPE_VNODE
) {
2560 vp
= (struct vnode
*)fp
->f_data
;
2561 lf
.l_whence
= SEEK_SET
;
2564 if (uap
->how
& LOCK_UN
) {
2565 lf
.l_type
= F_UNLCK
;
2566 atomic_clear_int(&fp
->f_flag
, FHASLOCK
); /* race ok */
2567 error
= VOP_ADVLOCK(vp
, (caddr_t
)fp
, F_UNLCK
, &lf
, 0);
2570 if (uap
->how
& LOCK_EX
)
2571 lf
.l_type
= F_WRLCK
;
2572 else if (uap
->how
& LOCK_SH
)
2573 lf
.l_type
= F_RDLCK
;
2578 if (uap
->how
& LOCK_NB
)
2579 error
= VOP_ADVLOCK(vp
, (caddr_t
)fp
, F_SETLK
, &lf
, 0);
2581 error
= VOP_ADVLOCK(vp
, (caddr_t
)fp
, F_SETLK
, &lf
, F_WAIT
);
2582 atomic_set_int(&fp
->f_flag
, FHASLOCK
); /* race ok */
2589 * File Descriptor pseudo-device driver (/dev/fd/).
2591 * Opening minor device N dup()s the file (if any) connected to file
2592 * descriptor N belonging to the calling process. Note that this driver
2593 * consists of only the ``open()'' routine, because all subsequent
2594 * references to this file will be direct to the other driver.
2597 fdopen(struct dev_open_args
*ap
)
2599 thread_t td
= curthread
;
2601 KKASSERT(td
->td_lwp
!= NULL
);
2604 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the
2605 * the file descriptor being sought for duplication. The error
2606 * return ensures that the vnode for this device will be released
2607 * by vn_open. Open will detect this special error and take the
2608 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2609 * will simply report the error.
2611 td
->td_lwp
->lwp_dupfd
= minor(ap
->a_head
.a_dev
);
2616 * The caller has reserved the file descriptor dfd for us. On success we
2617 * must fsetfd() it. On failure the caller will clean it up.
2620 dupfdopen(struct filedesc
*fdp
, int dfd
, int sfd
, int mode
, int error
)
2626 if ((wfp
= holdfp(fdp
, sfd
, -1)) == NULL
)
2630 * Close a revoke/dup race. Duping a descriptor marked as revoked
2631 * will dup a dummy descriptor instead of the real one.
2633 if (wfp
->f_flag
& FREVOKED
) {
2634 kprintf("Warning: attempt to dup() a revoked descriptor\n");
2637 werror
= falloc(NULL
, &wfp
, NULL
);
2643 * There are two cases of interest here.
2645 * For ENODEV simply dup sfd to file descriptor dfd and return.
2647 * For ENXIO steal away the file structure from sfd and store it
2648 * dfd. sfd is effectively closed by this operation.
2650 * Any other error code is just returned.
2655 * Check that the mode the file is being opened for is a
2656 * subset of the mode of the existing descriptor.
2658 if (((mode
& (FREAD
|FWRITE
)) | wfp
->f_flag
) != wfp
->f_flag
) {
2662 spin_lock(&fdp
->fd_spin
);
2663 fdp
->fd_files
[dfd
].fileflags
= fdp
->fd_files
[sfd
].fileflags
;
2664 fsetfd_locked(fdp
, wfp
, dfd
);
2665 spin_unlock(&fdp
->fd_spin
);
2670 * Steal away the file pointer from dfd, and stuff it into indx.
2672 spin_lock(&fdp
->fd_spin
);
2673 fdp
->fd_files
[dfd
].fileflags
= fdp
->fd_files
[sfd
].fileflags
;
2674 fsetfd(fdp
, wfp
, dfd
);
2675 if ((xfp
= funsetfd_locked(fdp
, sfd
)) != NULL
) {
2676 spin_unlock(&fdp
->fd_spin
);
2679 spin_unlock(&fdp
->fd_spin
);
2691 * NOT MPSAFE - I think these refer to a common file descriptor table
2692 * and we need to spinlock that to link fdtol in.
2694 struct filedesc_to_leader
*
2695 filedesc_to_leader_alloc(struct filedesc_to_leader
*old
,
2696 struct proc
*leader
)
2698 struct filedesc_to_leader
*fdtol
;
2700 fdtol
= kmalloc(sizeof(struct filedesc_to_leader
),
2701 M_FILEDESC_TO_LEADER
, M_WAITOK
| M_ZERO
);
2702 fdtol
->fdl_refcount
= 1;
2703 fdtol
->fdl_holdcount
= 0;
2704 fdtol
->fdl_wakeup
= 0;
2705 fdtol
->fdl_leader
= leader
;
2707 fdtol
->fdl_next
= old
->fdl_next
;
2708 fdtol
->fdl_prev
= old
;
2709 old
->fdl_next
= fdtol
;
2710 fdtol
->fdl_next
->fdl_prev
= fdtol
;
2712 fdtol
->fdl_next
= fdtol
;
2713 fdtol
->fdl_prev
= fdtol
;
2719 * Scan all file pointers in the system. The callback is made with
2720 * the master list spinlock held exclusively.
2723 allfiles_scan_exclusive(int (*callback
)(struct file
*, void *), void *data
)
2727 for (i
= 0; i
< NFILELIST_HEADS
; ++i
) {
2728 struct filelist_head
*head
= &filelist_heads
[i
];
2731 spin_lock(&head
->spin
);
2732 LIST_FOREACH(fp
, &head
->list
, f_list
) {
2735 res
= callback(fp
, data
);
2739 spin_unlock(&head
->spin
);
2744 * Get file structures.
2746 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe)
2749 struct sysctl_kern_file_info
{
2752 struct sysctl_req
*req
;
2755 static int sysctl_kern_file_callback(struct proc
*p
, void *data
);
2758 sysctl_kern_file(SYSCTL_HANDLER_ARGS
)
2760 struct sysctl_kern_file_info info
;
2763 * Note: because the number of file descriptors is calculated
2764 * in different ways for sizing vs returning the data,
2765 * there is information leakage from the first loop. However,
2766 * it is of a similar order of magnitude to the leakage from
2767 * global system statistics such as kern.openfiles.
2769 * When just doing a count, note that we cannot just count
2770 * the elements and add f_count via the filehead list because
2771 * threaded processes share their descriptor table and f_count might
2772 * still be '1' in that case.
2774 * Since the SYSCTL op can block, we must hold the process to
2775 * prevent it being ripped out from under us either in the
2776 * file descriptor loop or in the greater LIST_FOREACH. The
2777 * process may be in varying states of disrepair. If the process
2778 * is in SZOMB we may have caught it just as it is being removed
2779 * from the allproc list, we must skip it in that case to maintain
2780 * an unbroken chain through the allproc list.
2785 allproc_scan(sysctl_kern_file_callback
, &info
);
2788 * When just calculating the size, overestimate a bit to try to
2789 * prevent system activity from causing the buffer-fill call
2792 if (req
->oldptr
== NULL
) {
2793 info
.count
= (info
.count
+ 16) + (info
.count
/ 10);
2794 info
.error
= SYSCTL_OUT(req
, NULL
,
2795 info
.count
* sizeof(struct kinfo_file
));
2797 return (info
.error
);
2801 sysctl_kern_file_callback(struct proc
*p
, void *data
)
2803 struct sysctl_kern_file_info
*info
= data
;
2804 struct kinfo_file kf
;
2805 struct filedesc
*fdp
;
2810 if (p
->p_stat
== SIDL
|| p
->p_stat
== SZOMB
)
2812 if (!(PRISON_CHECK(info
->req
->td
->td_ucred
, p
->p_ucred
) != 0))
2816 * Softref the fdp to prevent it from being destroyed
2818 spin_lock(&p
->p_spin
);
2819 if ((fdp
= p
->p_fd
) == NULL
) {
2820 spin_unlock(&p
->p_spin
);
2823 atomic_add_int(&fdp
->fd_softrefs
, 1);
2824 spin_unlock(&p
->p_spin
);
2827 * The fdp's own spinlock prevents the contents from being
2830 spin_lock_shared(&fdp
->fd_spin
);
2831 for (n
= 0; n
< fdp
->fd_nfiles
; ++n
) {
2832 if ((fp
= fdp
->fd_files
[n
].fp
) == NULL
)
2834 if (info
->req
->oldptr
== NULL
) {
2837 uid
= p
->p_ucred
? p
->p_ucred
->cr_uid
: -1;
2838 kcore_make_file(&kf
, fp
, p
->p_pid
, uid
, n
);
2839 spin_unlock_shared(&fdp
->fd_spin
);
2840 info
->error
= SYSCTL_OUT(info
->req
, &kf
, sizeof(kf
));
2841 spin_lock_shared(&fdp
->fd_spin
);
2846 spin_unlock_shared(&fdp
->fd_spin
);
2847 atomic_subtract_int(&fdp
->fd_softrefs
, 1);
2853 SYSCTL_PROC(_kern
, KERN_FILE
, file
, CTLTYPE_OPAQUE
|CTLFLAG_RD
,
2854 0, 0, sysctl_kern_file
, "S,file", "Entire file table");
2856 SYSCTL_INT(_kern
, OID_AUTO
, minfilesperproc
, CTLFLAG_RW
,
2857 &minfilesperproc
, 0, "Minimum files allowed open per process");
2858 SYSCTL_INT(_kern
, KERN_MAXFILESPERPROC
, maxfilesperproc
, CTLFLAG_RW
,
2859 &maxfilesperproc
, 0, "Maximum files allowed open per process");
2860 SYSCTL_INT(_kern
, OID_AUTO
, maxfilesperuser
, CTLFLAG_RW
,
2861 &maxfilesperuser
, 0, "Maximum files allowed open per user");
2863 SYSCTL_INT(_kern
, KERN_MAXFILES
, maxfiles
, CTLFLAG_RW
,
2864 &maxfiles
, 0, "Maximum number of files");
2866 SYSCTL_INT(_kern
, OID_AUTO
, maxfilesrootres
, CTLFLAG_RW
,
2867 &maxfilesrootres
, 0, "Descriptors reserved for root use");
2869 SYSCTL_INT(_kern
, OID_AUTO
, openfiles
, CTLFLAG_RD
,
2870 &nfiles
, 0, "System-wide number of open files");
2873 fildesc_drvinit(void *unused
)
2877 for (fd
= 0; fd
< NUMFDESC
; fd
++) {
2878 make_dev(&fildesc_ops
, fd
,
2879 UID_BIN
, GID_BIN
, 0666, "fd/%d", fd
);
2882 make_dev(&fildesc_ops
, 0, UID_ROOT
, GID_WHEEL
, 0666, "stdin");
2883 make_dev(&fildesc_ops
, 1, UID_ROOT
, GID_WHEEL
, 0666, "stdout");
2884 make_dev(&fildesc_ops
, 2, UID_ROOT
, GID_WHEEL
, 0666, "stderr");
2887 struct fileops badfileops
= {
2888 .fo_read
= badfo_readwrite
,
2889 .fo_write
= badfo_readwrite
,
2890 .fo_ioctl
= badfo_ioctl
,
2891 .fo_kqfilter
= badfo_kqfilter
,
2892 .fo_stat
= badfo_stat
,
2893 .fo_close
= badfo_close
,
2894 .fo_shutdown
= badfo_shutdown
2908 badfo_ioctl(struct file
*fp
, u_long com
, caddr_t data
,
2909 struct ucred
*cred
, struct sysmsg
*msgv
)
2915 * Must return an error to prevent registration, typically
2916 * due to a revoked descriptor (file_filtops assigned).
2919 badfo_kqfilter(struct file
*fp
, struct knote
*kn
)
2921 return (EOPNOTSUPP
);
2925 badfo_stat(struct file
*fp
, struct stat
*sb
, struct ucred
*cred
)
2931 badfo_close(struct file
*fp
)
2937 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
);
2952 filelist_heads_init(void *arg __unused
)
2956 for (i
= 0; i
< NFILELIST_HEADS
; ++i
) {
2957 struct filelist_head
*head
= &filelist_heads
[i
];
2959 spin_init(&head
->spin
, "filehead_spin");
2960 LIST_INIT(&head
->list
);
2964 SYSINIT(filelistheads
, SI_BOOT1_LOCK
, SI_ORDER_ANY
,
2965 filelist_heads_init
, NULL
);
2968 file_objcache_init(void *dummy __unused
)
2970 file_objcache
= objcache_create("file", maxfiles
, maxfiles
/ 8,
2971 NULL
, NULL
, NULL
, /* TODO: ctor/dtor */
2972 objcache_malloc_alloc
, objcache_malloc_free
, &file_malloc_args
);
2974 SYSINIT(fpobjcache
, SI_BOOT2_POST_SMP
, SI_ORDER_ANY
, file_objcache_init
, NULL
);