2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95
39 * $FreeBSD: src/sys/kern/vfs_subr.c,v 1.249.2.30 2003/04/04 20:35:57 tegge Exp $
40 * $DragonFly: src/sys/kern/vfs_subr.c,v 1.70 2006/03/05 18:38:34 dillon Exp $
44 * External virtual filesystem routines
48 #include <sys/param.h>
49 #include <sys/systm.h>
52 #include <sys/dirent.h>
53 #include <sys/domain.h>
54 #include <sys/eventhandler.h>
55 #include <sys/fcntl.h>
56 #include <sys/kernel.h>
57 #include <sys/kthread.h>
58 #include <sys/malloc.h>
60 #include <sys/mount.h>
62 #include <sys/reboot.h>
63 #include <sys/socket.h>
65 #include <sys/sysctl.h>
66 #include <sys/syslog.h>
67 #include <sys/unistd.h>
68 #include <sys/vmmeter.h>
69 #include <sys/vnode.h>
71 #include <machine/limits.h>
74 #include <vm/vm_object.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_kern.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_pager.h>
81 #include <vm/vnode_pager.h>
82 #include <vm/vm_zone.h>
85 #include <sys/thread2.h>
87 static MALLOC_DEFINE(M_NETADDR
, "Export Host", "Export host address structure");
90 SYSCTL_INT(_debug
, OID_AUTO
, numvnodes
, CTLFLAG_RD
, &numvnodes
, 0, "");
92 SYSCTL_INT(_vfs
, OID_AUTO
, fastdev
, CTLFLAG_RW
, &vfs_fastdev
, 0, "");
94 enum vtype iftovt_tab
[16] = {
95 VNON
, VFIFO
, VCHR
, VNON
, VDIR
, VNON
, VBLK
, VNON
,
96 VREG
, VNON
, VLNK
, VNON
, VSOCK
, VNON
, VNON
, VBAD
,
99 0, S_IFREG
, S_IFDIR
, S_IFBLK
, S_IFCHR
, S_IFLNK
,
100 S_IFSOCK
, S_IFIFO
, S_IFMT
,
103 static int reassignbufcalls
;
104 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufcalls
, CTLFLAG_RW
,
105 &reassignbufcalls
, 0, "");
106 static int reassignbufloops
;
107 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufloops
, CTLFLAG_RW
,
108 &reassignbufloops
, 0, "");
109 static int reassignbufsortgood
;
110 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufsortgood
, CTLFLAG_RW
,
111 &reassignbufsortgood
, 0, "");
112 static int reassignbufsortbad
;
113 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufsortbad
, CTLFLAG_RW
,
114 &reassignbufsortbad
, 0, "");
115 static int reassignbufmethod
= 1;
116 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufmethod
, CTLFLAG_RW
,
117 &reassignbufmethod
, 0, "");
119 int nfs_mount_type
= -1;
120 static struct lwkt_token spechash_token
;
121 struct nfs_public nfs_pub
; /* publicly exported FS */
124 SYSCTL_INT(_kern
, KERN_MAXVNODES
, maxvnodes
, CTLFLAG_RW
,
125 &desiredvnodes
, 0, "Maximum number of vnodes");
127 static void vfs_free_addrlist (struct netexport
*nep
);
128 static int vfs_free_netcred (struct radix_node
*rn
, void *w
);
129 static int vfs_hang_addrlist (struct mount
*mp
, struct netexport
*nep
,
130 struct export_args
*argp
);
132 extern int dev_ref_debug
;
133 extern struct vnodeopv_entry_desc spec_vnodeop_entries
[];
136 * Red black tree functions
138 static int rb_buf_compare(struct buf
*b1
, struct buf
*b2
);
139 RB_GENERATE2(buf_rb_tree
, buf
, b_rbnode
, rb_buf_compare
, daddr_t
, b_lblkno
);
140 RB_GENERATE2(buf_rb_hash
, buf
, b_rbhash
, rb_buf_compare
, daddr_t
, b_lblkno
);
143 rb_buf_compare(struct buf
*b1
, struct buf
*b2
)
145 if (b1
->b_lblkno
< b2
->b_lblkno
)
147 if (b1
->b_lblkno
> b2
->b_lblkno
)
153 * Return 0 if the vnode is already on the free list or cannot be placed
154 * on the free list. Return 1 if the vnode can be placed on the free list.
157 vshouldfree(struct vnode
*vp
, int usecount
)
159 if (vp
->v_flag
& VFREE
)
160 return (0); /* already free */
161 if (vp
->v_holdcnt
!= 0 || vp
->v_usecount
!= usecount
)
162 return (0); /* other holderse */
164 (vp
->v_object
->ref_count
|| vp
->v_object
->resident_page_count
)) {
171 * Initialize the vnode management data structures.
173 * Called from vfsinit()
179 * Desired vnodes is a result of the physical page count
180 * and the size of kernel's heap. It scales in proportion
181 * to the amount of available physical memory. This can
182 * cause trouble on 64-bit and large memory platforms.
184 /* desiredvnodes = maxproc + vmstats.v_page_count / 4; */
186 min(maxproc
+ vmstats
.v_page_count
/4,
187 2 * (VM_MAX_KERNEL_ADDRESS
- VM_MIN_KERNEL_ADDRESS
) /
188 (5 * (sizeof(struct vm_object
) + sizeof(struct vnode
))));
190 lwkt_token_init(&spechash_token
);
194 * Knob to control the precision of file timestamps:
196 * 0 = seconds only; nanoseconds zeroed.
197 * 1 = seconds and nanoseconds, accurate within 1/HZ.
198 * 2 = seconds and nanoseconds, truncated to microseconds.
199 * >=3 = seconds and nanoseconds, maximum precision.
201 enum { TSP_SEC
, TSP_HZ
, TSP_USEC
, TSP_NSEC
};
203 static int timestamp_precision
= TSP_SEC
;
204 SYSCTL_INT(_vfs
, OID_AUTO
, timestamp_precision
, CTLFLAG_RW
,
205 ×tamp_precision
, 0, "");
208 * Get a current timestamp.
211 vfs_timestamp(struct timespec
*tsp
)
215 switch (timestamp_precision
) {
217 tsp
->tv_sec
= time_second
;
225 TIMEVAL_TO_TIMESPEC(&tv
, tsp
);
235 * Set vnode attributes to VNOVAL
238 vattr_null(struct vattr
*vap
)
241 vap
->va_size
= VNOVAL
;
242 vap
->va_bytes
= VNOVAL
;
243 vap
->va_mode
= VNOVAL
;
244 vap
->va_nlink
= VNOVAL
;
245 vap
->va_uid
= VNOVAL
;
246 vap
->va_gid
= VNOVAL
;
247 vap
->va_fsid
= VNOVAL
;
248 vap
->va_fileid
= VNOVAL
;
249 vap
->va_blocksize
= VNOVAL
;
250 vap
->va_rdev
= VNOVAL
;
251 vap
->va_atime
.tv_sec
= VNOVAL
;
252 vap
->va_atime
.tv_nsec
= VNOVAL
;
253 vap
->va_mtime
.tv_sec
= VNOVAL
;
254 vap
->va_mtime
.tv_nsec
= VNOVAL
;
255 vap
->va_ctime
.tv_sec
= VNOVAL
;
256 vap
->va_ctime
.tv_nsec
= VNOVAL
;
257 vap
->va_flags
= VNOVAL
;
258 vap
->va_gen
= VNOVAL
;
260 vap
->va_fsmid
= VNOVAL
;
264 * Flush out and invalidate all buffers associated with a vnode.
268 static int vinvalbuf_bp(struct buf
*bp
, void *data
);
270 struct vinvalbuf_bp_info
{
278 vinvalbuf(struct vnode
*vp
, int flags
, struct thread
*td
,
279 int slpflag
, int slptimeo
)
281 struct vinvalbuf_bp_info info
;
286 * If we are being asked to save, call fsync to ensure that the inode
289 if (flags
& V_SAVE
) {
291 while (vp
->v_track_write
.bk_active
) {
292 vp
->v_track_write
.bk_waitflag
= 1;
293 error
= tsleep(&vp
->v_track_write
, slpflag
,
294 "vinvlbuf", slptimeo
);
300 if (!RB_EMPTY(&vp
->v_rbdirty_tree
)) {
302 if ((error
= VOP_FSYNC(vp
, MNT_WAIT
, td
)) != 0)
305 if (vp
->v_track_write
.bk_active
> 0 ||
306 !RB_EMPTY(&vp
->v_rbdirty_tree
))
307 panic("vinvalbuf: dirty bufs");
312 info
.slptimeo
= slptimeo
;
313 info
.lkflags
= LK_EXCLUSIVE
| LK_SLEEPFAIL
;
314 if (slpflag
& PCATCH
)
315 info
.lkflags
|= LK_PCATCH
;
320 * Flush the buffer cache until nothing is left.
322 while (!RB_EMPTY(&vp
->v_rbclean_tree
) ||
323 !RB_EMPTY(&vp
->v_rbdirty_tree
)) {
324 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbclean_tree
, NULL
,
325 vinvalbuf_bp
, &info
);
327 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, NULL
,
328 vinvalbuf_bp
, &info
);
333 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
334 * have write I/O in-progress but if there is a VM object then the
335 * VM object can also have read-I/O in-progress.
338 while (vp
->v_track_write
.bk_active
> 0) {
339 vp
->v_track_write
.bk_waitflag
= 1;
340 tsleep(&vp
->v_track_write
, 0, "vnvlbv", 0);
342 if (VOP_GETVOBJECT(vp
, &object
) == 0) {
343 while (object
->paging_in_progress
)
344 vm_object_pip_sleep(object
, "vnvlbx");
346 } while (vp
->v_track_write
.bk_active
> 0);
351 * Destroy the copy in the VM cache, too.
353 if (VOP_GETVOBJECT(vp
, &object
) == 0) {
354 vm_object_page_remove(object
, 0, 0,
355 (flags
& V_SAVE
) ? TRUE
: FALSE
);
358 if (!RB_EMPTY(&vp
->v_rbdirty_tree
) || !RB_EMPTY(&vp
->v_rbclean_tree
))
359 panic("vinvalbuf: flush failed");
360 if (!RB_EMPTY(&vp
->v_rbhash_tree
))
361 panic("vinvalbuf: flush failed, buffers still present");
366 vinvalbuf_bp(struct buf
*bp
, void *data
)
368 struct vinvalbuf_bp_info
*info
= data
;
371 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
)) {
372 error
= BUF_TIMELOCK(bp
, info
->lkflags
,
373 "vinvalbuf", info
->slptimeo
);
383 KKASSERT(bp
->b_vp
== info
->vp
);
386 * XXX Since there are no node locks for NFS, I
387 * believe there is a slight chance that a delayed
388 * write will occur while sleeping just above, so
389 * check for it. Note that vfs_bio_awrite expects
390 * buffers to reside on a queue, while VOP_BWRITE and
393 if (((bp
->b_flags
& (B_DELWRI
| B_INVAL
)) == B_DELWRI
) &&
394 (info
->flags
& V_SAVE
)) {
395 if (bp
->b_vp
== info
->vp
) {
396 if (bp
->b_flags
& B_CLUSTEROK
) {
400 bp
->b_flags
|= B_ASYNC
;
401 VOP_BWRITE(bp
->b_vp
, bp
);
405 VOP_BWRITE(bp
->b_vp
, bp
);
409 bp
->b_flags
|= (B_INVAL
| B_NOCACHE
| B_RELBUF
);
410 bp
->b_flags
&= ~B_ASYNC
;
417 * Truncate a file's buffer and pages to a specified length. This
418 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
421 * The vnode must be locked.
423 static int vtruncbuf_bp_trunc_cmp(struct buf
*bp
, void *data
);
424 static int vtruncbuf_bp_trunc(struct buf
*bp
, void *data
);
425 static int vtruncbuf_bp_metasync_cmp(struct buf
*bp
, void *data
);
426 static int vtruncbuf_bp_metasync(struct buf
*bp
, void *data
);
429 vtruncbuf(struct vnode
*vp
, struct thread
*td
, off_t length
, int blksize
)
435 * Round up to the *next* lbn, then destroy the buffers in question.
436 * Since we are only removing some of the buffers we must rely on the
437 * scan count to determine whether a loop is necessary.
439 trunclbn
= (length
+ blksize
- 1) / blksize
;
443 count
= RB_SCAN(buf_rb_tree
, &vp
->v_rbclean_tree
,
444 vtruncbuf_bp_trunc_cmp
,
445 vtruncbuf_bp_trunc
, &trunclbn
);
446 count
+= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
447 vtruncbuf_bp_trunc_cmp
,
448 vtruncbuf_bp_trunc
, &trunclbn
);
452 * For safety, fsync any remaining metadata if the file is not being
453 * truncated to 0. Since the metadata does not represent the entire
454 * dirty list we have to rely on the hit count to ensure that we get
459 count
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
460 vtruncbuf_bp_metasync_cmp
,
461 vtruncbuf_bp_metasync
, vp
);
466 * Wait for any in-progress I/O to complete before returning (why?)
468 while (vp
->v_track_write
.bk_active
> 0) {
469 vp
->v_track_write
.bk_waitflag
= 1;
470 tsleep(&vp
->v_track_write
, 0, "vbtrunc", 0);
475 vnode_pager_setsize(vp
, length
);
481 * The callback buffer is beyond the new file EOF and must be destroyed.
482 * Note that the compare function must conform to the RB_SCAN's requirements.
486 vtruncbuf_bp_trunc_cmp(struct buf
*bp
, void *data
)
488 if (bp
->b_lblkno
>= *(daddr_t
*)data
)
495 vtruncbuf_bp_trunc(struct buf
*bp
, void *data
)
498 * Do not try to use a buffer we cannot immediately lock, but sleep
499 * anyway to prevent a livelock. The code will loop until all buffers
502 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
)) {
503 if (BUF_LOCK(bp
, LK_EXCLUSIVE
|LK_SLEEPFAIL
) == 0)
507 bp
->b_flags
|= (B_INVAL
| B_RELBUF
);
508 bp
->b_flags
&= ~B_ASYNC
;
515 * Fsync all meta-data after truncating a file to be non-zero. Only metadata
516 * blocks (with a negative lblkno) are scanned.
517 * Note that the compare function must conform to the RB_SCAN's requirements.
520 vtruncbuf_bp_metasync_cmp(struct buf
*bp
, void *data
)
522 if (bp
->b_lblkno
< 0)
528 vtruncbuf_bp_metasync(struct buf
*bp
, void *data
)
530 struct vnode
*vp
= data
;
532 if (bp
->b_flags
& B_DELWRI
) {
534 * Do not try to use a buffer we cannot immediately lock,
535 * but sleep anyway to prevent a livelock. The code will
536 * loop until all buffers can be acted upon.
538 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
)) {
539 if (BUF_LOCK(bp
, LK_EXCLUSIVE
|LK_SLEEPFAIL
) == 0)
543 if (bp
->b_vp
== vp
) {
544 bp
->b_flags
|= B_ASYNC
;
546 bp
->b_flags
&= ~B_ASYNC
;
548 VOP_BWRITE(bp
->b_vp
, bp
);
557 * vfsync - implements a multipass fsync on a file which understands
558 * dependancies and meta-data. The passed vnode must be locked. The
559 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY.
561 * When fsyncing data asynchronously just do one consolidated pass starting
562 * with the most negative block number. This may not get all the data due
565 * When fsyncing data synchronously do a data pass, then a metadata pass,
566 * then do additional data+metadata passes to try to get all the data out.
568 static int vfsync_wait_output(struct vnode
*vp
,
569 int (*waitoutput
)(struct vnode
*, struct thread
*));
570 static int vfsync_data_only_cmp(struct buf
*bp
, void *data
);
571 static int vfsync_meta_only_cmp(struct buf
*bp
, void *data
);
572 static int vfsync_lazy_range_cmp(struct buf
*bp
, void *data
);
573 static int vfsync_bp(struct buf
*bp
, void *data
);
582 int (*checkdef
)(struct buf
*);
586 vfsync(struct vnode
*vp
, int waitfor
, int passes
, daddr_t lbn
,
587 int (*checkdef
)(struct buf
*),
588 int (*waitoutput
)(struct vnode
*, struct thread
*))
590 struct vfsync_info info
;
593 bzero(&info
, sizeof(info
));
596 if ((info
.checkdef
= checkdef
) == NULL
)
604 * Lazy (filesystem syncer typ) Asynchronous plus limit the
605 * number of data (not meta) pages we try to flush to 1MB.
606 * A non-zero return means that lazy limit was reached.
608 info
.lazylimit
= 1024 * 1024;
610 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
611 vfsync_lazy_range_cmp
, vfsync_bp
, &info
);
612 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
613 vfsync_meta_only_cmp
, vfsync_bp
, &info
);
616 else if (!RB_EMPTY(&vp
->v_rbdirty_tree
))
617 vn_syncer_add_to_worklist(vp
, 1);
622 * Asynchronous. Do a data-only pass and a meta-only pass.
625 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, vfsync_data_only_cmp
,
627 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, vfsync_meta_only_cmp
,
633 * Synchronous. Do a data-only pass, then a meta-data+data
634 * pass, then additional integrated passes to try to get
635 * all the dependancies flushed.
637 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, vfsync_data_only_cmp
,
639 error
= vfsync_wait_output(vp
, waitoutput
);
641 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, NULL
,
643 error
= vfsync_wait_output(vp
, waitoutput
);
645 while (error
== 0 && passes
> 0 &&
646 !RB_EMPTY(&vp
->v_rbdirty_tree
)) {
648 info
.synchronous
= 1;
651 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, NULL
,
657 error
= vfsync_wait_output(vp
, waitoutput
);
666 vfsync_wait_output(struct vnode
*vp
, int (*waitoutput
)(struct vnode
*, struct thread
*))
670 while (vp
->v_track_write
.bk_active
) {
671 vp
->v_track_write
.bk_waitflag
= 1;
672 tsleep(&vp
->v_track_write
, 0, "fsfsn", 0);
675 error
= waitoutput(vp
, curthread
);
680 vfsync_data_only_cmp(struct buf
*bp
, void *data
)
682 if (bp
->b_lblkno
< 0)
688 vfsync_meta_only_cmp(struct buf
*bp
, void *data
)
690 if (bp
->b_lblkno
< 0)
696 vfsync_lazy_range_cmp(struct buf
*bp
, void *data
)
698 struct vfsync_info
*info
= data
;
699 if (bp
->b_lblkno
< info
->vp
->v_lazyw
)
705 vfsync_bp(struct buf
*bp
, void *data
)
707 struct vfsync_info
*info
= data
;
708 struct vnode
*vp
= info
->vp
;
712 * if syncdeps is not set we do not try to write buffers which have
715 if (!info
->synchronous
&& info
->syncdeps
== 0 && info
->checkdef(bp
))
719 * Ignore buffers that we cannot immediately lock. XXX
721 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
))
723 if ((bp
->b_flags
& B_DELWRI
) == 0)
724 panic("vfsync_bp: buffer not dirty");
726 panic("vfsync_bp: buffer vp mismatch");
729 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
730 * has been written but an additional handshake with the device
731 * is required before we can dispose of the buffer. We have no idea
732 * how to do this so we have to skip these buffers.
734 if (bp
->b_flags
& B_NEEDCOMMIT
) {
740 * (LEGACY FROM UFS, REMOVE WHEN POSSIBLE) - invalidate any dirty
741 * buffers beyond the file EOF.
743 if (info
->lbn
!= (daddr_t
)-1 && vp
->v_type
== VREG
&&
744 bp
->b_lblkno
>= info
->lbn
) {
746 bp
->b_flags
|= B_INVAL
| B_NOCACHE
;
752 if (info
->synchronous
) {
754 * Synchronous flushing. An error may be returned.
762 * Asynchronous flushing. A negative return value simply
763 * stops the scan and is not considered an error. We use
764 * this to support limited MNT_LAZY flushes.
766 vp
->v_lazyw
= bp
->b_lblkno
;
767 if ((vp
->v_flag
& VOBJBUF
) && (bp
->b_flags
& B_CLUSTEROK
)) {
768 info
->lazycount
+= vfs_bio_awrite(bp
);
770 info
->lazycount
+= bp
->b_bufsize
;
776 if (info
->lazylimit
&& info
->lazycount
>= info
->lazylimit
)
785 * Associate a buffer with a vnode.
788 bgetvp(struct vnode
*vp
, struct buf
*bp
)
790 KASSERT(bp
->b_vp
== NULL
, ("bgetvp: not free"));
791 KKASSERT((bp
->b_flags
& (B_HASHED
|B_DELWRI
)) == 0);
792 KKASSERT((bp
->b_xflags
& (BX_VNCLEAN
|BX_VNDIRTY
)) == 0);
796 * Insert onto list for new vnode.
800 bp
->b_flags
|= B_HASHED
;
801 if (buf_rb_hash_RB_INSERT(&vp
->v_rbhash_tree
, bp
))
802 panic("reassignbuf: dup lblk vp %p bp %p", vp
, bp
);
804 bp
->b_xflags
|= BX_VNCLEAN
;
805 if (buf_rb_tree_RB_INSERT(&vp
->v_rbclean_tree
, bp
))
806 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp
, bp
);
811 * Disassociate a buffer from a vnode.
814 brelvp(struct buf
*bp
)
818 KASSERT(bp
->b_vp
!= NULL
, ("brelvp: NULL"));
821 * Delete from old vnode list, if on one.
825 if (bp
->b_xflags
& (BX_VNDIRTY
| BX_VNCLEAN
)) {
826 if (bp
->b_xflags
& BX_VNDIRTY
)
827 buf_rb_tree_RB_REMOVE(&vp
->v_rbdirty_tree
, bp
);
829 buf_rb_tree_RB_REMOVE(&vp
->v_rbclean_tree
, bp
);
830 bp
->b_xflags
&= ~(BX_VNDIRTY
| BX_VNCLEAN
);
832 if (bp
->b_flags
& B_HASHED
) {
833 buf_rb_hash_RB_REMOVE(&vp
->v_rbhash_tree
, bp
);
834 bp
->b_flags
&= ~B_HASHED
;
836 if ((vp
->v_flag
& VONWORKLST
) && RB_EMPTY(&vp
->v_rbdirty_tree
)) {
837 vp
->v_flag
&= ~VONWORKLST
;
838 LIST_REMOVE(vp
, v_synclist
);
846 * Associate a p-buffer with a vnode.
848 * Also sets B_PAGING flag to indicate that vnode is not fully associated
849 * with the buffer. i.e. the bp has not been linked into the vnode or
853 pbgetvp(struct vnode
*vp
, struct buf
*bp
)
855 KASSERT(bp
->b_vp
== NULL
, ("pbgetvp: not free"));
856 KKASSERT((bp
->b_flags
& B_HASHED
) == 0);
859 bp
->b_flags
|= B_PAGING
;
863 * Disassociate a p-buffer from a vnode.
866 pbrelvp(struct buf
*bp
)
868 KASSERT(bp
->b_vp
!= NULL
, ("pbrelvp: NULL"));
869 KKASSERT((bp
->b_flags
& B_HASHED
) == 0);
872 bp
->b_flags
&= ~B_PAGING
;
876 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI.
877 * This routine is called when the state of the B_DELWRI bit is changed.
880 reassignbuf(struct buf
*bp
)
882 struct vnode
*vp
= bp
->b_vp
;
885 KKASSERT(vp
!= NULL
);
889 * B_PAGING flagged buffers cannot be reassigned because their vp
890 * is not fully linked in.
892 if (bp
->b_flags
& B_PAGING
)
893 panic("cannot reassign paging buffer");
896 if (bp
->b_flags
& B_DELWRI
) {
898 * Move to the dirty list, add the vnode to the worklist
900 if (bp
->b_xflags
& BX_VNCLEAN
) {
901 buf_rb_tree_RB_REMOVE(&vp
->v_rbclean_tree
, bp
);
902 bp
->b_xflags
&= ~BX_VNCLEAN
;
904 if ((bp
->b_xflags
& BX_VNDIRTY
) == 0) {
905 if (buf_rb_tree_RB_INSERT(&vp
->v_rbdirty_tree
, bp
)) {
906 panic("reassignbuf: dup lblk vp %p bp %p",
909 bp
->b_xflags
|= BX_VNDIRTY
;
911 if ((vp
->v_flag
& VONWORKLST
) == 0) {
912 switch (vp
->v_type
) {
919 vp
->v_rdev
->si_mountpoint
!= NULL
) {
927 vn_syncer_add_to_worklist(vp
, delay
);
931 * Move to the clean list, remove the vnode from the worklist
932 * if no dirty blocks remain.
934 if (bp
->b_xflags
& BX_VNDIRTY
) {
935 buf_rb_tree_RB_REMOVE(&vp
->v_rbdirty_tree
, bp
);
936 bp
->b_xflags
&= ~BX_VNDIRTY
;
938 if ((bp
->b_xflags
& BX_VNCLEAN
) == 0) {
939 if (buf_rb_tree_RB_INSERT(&vp
->v_rbclean_tree
, bp
)) {
940 panic("reassignbuf: dup lblk vp %p bp %p",
943 bp
->b_xflags
|= BX_VNCLEAN
;
945 if ((vp
->v_flag
& VONWORKLST
) &&
946 RB_EMPTY(&vp
->v_rbdirty_tree
)) {
947 vp
->v_flag
&= ~VONWORKLST
;
948 LIST_REMOVE(vp
, v_synclist
);
955 * Create a vnode for a block device.
956 * Used for mounting the root file system.
959 bdevvp(dev_t dev
, struct vnode
**vpp
)
969 error
= getspecialvnode(VT_NON
, NULL
, &spec_vnode_vops
, &nvp
, 0, 0);
976 vp
->v_udev
= dev
->si_udev
;
983 v_associate_rdev(struct vnode
*vp
, dev_t dev
)
987 if (dev
== NULL
|| dev
== NODEV
)
989 if (dev_is_good(dev
) == 0)
991 KKASSERT(vp
->v_rdev
== NULL
);
994 vp
->v_rdev
= reference_dev(dev
);
995 lwkt_gettoken(&ilock
, &spechash_token
);
996 SLIST_INSERT_HEAD(&dev
->si_hlist
, vp
, v_specnext
);
997 lwkt_reltoken(&ilock
);
1002 v_release_rdev(struct vnode
*vp
)
1007 if ((dev
= vp
->v_rdev
) != NULL
) {
1008 lwkt_gettoken(&ilock
, &spechash_token
);
1009 SLIST_REMOVE(&dev
->si_hlist
, vp
, vnode
, v_specnext
);
1010 if (dev_ref_debug
&& vp
->v_opencount
!= 0) {
1011 printf("releasing rdev with non-0 "
1012 "v_opencount(%d) (revoked?)\n",
1016 vp
->v_opencount
= 0;
1018 lwkt_reltoken(&ilock
);
1023 * Add a vnode to the alias list hung off the dev_t. We only associate
1024 * the device number with the vnode. The actual device is not associated
1025 * until the vnode is opened (usually in spec_open()), and will be
1026 * disassociated on last close.
1029 addaliasu(struct vnode
*nvp
, udev_t nvp_udev
)
1031 if (nvp
->v_type
!= VBLK
&& nvp
->v_type
!= VCHR
)
1032 panic("addaliasu on non-special vnode");
1033 nvp
->v_udev
= nvp_udev
;
1037 * Disassociate a vnode from its underlying filesystem.
1039 * The vnode must be VX locked and refd
1041 * If there are v_usecount references to the vnode other then ours we have
1042 * to VOP_CLOSE the vnode before we can deactivate and reclaim it.
1045 vclean(struct vnode
*vp
, int flags
, struct thread
*td
)
1051 * If the vnode has already been reclaimed we have nothing to do.
1053 if (vp
->v_flag
& VRECLAIMED
)
1055 vp
->v_flag
|= VRECLAIMED
;
1058 * Scrap the vfs cache
1060 while (cache_inval_vp(vp
, 0, &retflags
) != 0) {
1061 printf("Warning: vnode %p clean/cache_resolution race detected\n", vp
);
1062 tsleep(vp
, 0, "vclninv", 2);
1066 * Check to see if the vnode is in use. If so we have to reference it
1067 * before we clean it out so that its count cannot fall to zero and
1068 * generate a race against ourselves to recycle it.
1070 active
= (vp
->v_usecount
> 1);
1073 * Clean out any buffers associated with the vnode and destroy its
1074 * object, if it has one.
1076 vinvalbuf(vp
, V_SAVE
, td
, 0, 0);
1077 VOP_DESTROYVOBJECT(vp
);
1080 * If purging an active vnode, it must be closed and
1081 * deactivated before being reclaimed. XXX
1083 * Note that neither of these routines unlocks the vnode.
1086 if (flags
& DOCLOSE
)
1087 VOP_CLOSE(vp
, FNONBLOCK
, td
);
1091 * If the vnode has not be deactivated, deactivated it.
1093 if ((vp
->v_flag
& VINACTIVE
) == 0) {
1094 vp
->v_flag
|= VINACTIVE
;
1095 VOP_INACTIVE(vp
, td
);
1099 * Reclaim the vnode.
1101 if (VOP_RECLAIM(vp
, retflags
, td
))
1102 panic("vclean: cannot reclaim");
1105 * Done with purge, notify sleepers of the grim news.
1107 vp
->v_ops
= &dead_vnode_vops
;
1113 * Eliminate all activity associated with the requested vnode
1114 * and with all vnodes aliased to the requested vnode.
1116 * The vnode must be referenced and vx_lock()'d
1118 * revoke { struct vnode *a_vp, int a_flags }
1121 vop_stdrevoke(struct vop_revoke_args
*ap
)
1123 struct vnode
*vp
, *vq
;
1127 KASSERT((ap
->a_flags
& REVOKEALL
) != 0, ("vop_revoke"));
1132 * If the vnode is already dead don't try to revoke it
1134 if (vp
->v_flag
& VRECLAIMED
)
1138 * If the vnode has a device association, scrap all vnodes associated
1139 * with the device. Don't let the device disappear on us while we
1140 * are scrapping the vnodes.
1142 * The passed vp will probably show up in the list, do not VX lock
1145 if (vp
->v_type
!= VCHR
&& vp
->v_type
!= VBLK
)
1147 if ((dev
= vp
->v_rdev
) == NULL
) {
1148 if ((dev
= udev2dev(vp
->v_udev
, vp
->v_type
== VBLK
)) == NODEV
)
1152 lwkt_gettoken(&ilock
, &spechash_token
);
1153 while ((vq
= SLIST_FIRST(&dev
->si_hlist
)) != NULL
) {
1154 if (vp
== vq
|| vx_get(vq
) == 0) {
1155 if (vq
== SLIST_FIRST(&dev
->si_hlist
))
1161 lwkt_reltoken(&ilock
);
1167 * Recycle an unused vnode to the front of the free list.
1169 * Returns 1 if we were successfully able to recycle the vnode,
1173 vrecycle(struct vnode
*vp
, struct thread
*td
)
1175 if (vp
->v_usecount
== 1) {
1183 * Eliminate all activity associated with a vnode in preparation for reuse.
1185 * The vnode must be VX locked and refd and will remain VX locked and refd
1186 * on return. This routine may be called with the vnode in any state, as
1187 * long as it is VX locked. The vnode will be cleaned out and marked
1188 * VRECLAIMED but will not actually be reused until all existing refs and
1191 * NOTE: This routine may be called on a vnode which has not yet been
1192 * already been deactivated (VOP_INACTIVE), or on a vnode which has
1193 * already been reclaimed.
1195 * This routine is not responsible for placing us back on the freelist.
1196 * Instead, it happens automatically when the caller releases the VX lock
1197 * (assuming there aren't any other references).
1200 vgone(struct vnode
*vp
)
1203 * assert that the VX lock is held. This is an absolute requirement
1204 * now for vgone() to be called.
1206 KKASSERT(vp
->v_lock
.lk_exclusivecount
== 1);
1209 * Clean out the filesystem specific data and set the VRECLAIMED
1210 * bit. Also deactivate the vnode if necessary.
1212 vclean(vp
, DOCLOSE
, curthread
);
1215 * Delete from old mount point vnode list, if on one.
1217 if (vp
->v_mount
!= NULL
)
1218 insmntque(vp
, NULL
);
1221 * If special device, remove it from special device alias list
1222 * if it is on one. This should normally only occur if a vnode is
1223 * being revoked as the device should otherwise have been released
1226 if ((vp
->v_type
== VBLK
|| vp
->v_type
== VCHR
) && vp
->v_rdev
!= NULL
) {
1237 * Lookup a vnode by device number.
1240 vfinddev(dev_t dev
, enum vtype type
, struct vnode
**vpp
)
1245 lwkt_gettoken(&ilock
, &spechash_token
);
1246 SLIST_FOREACH(vp
, &dev
->si_hlist
, v_specnext
) {
1247 if (type
== vp
->v_type
) {
1249 lwkt_reltoken(&ilock
);
1253 lwkt_reltoken(&ilock
);
1258 * Calculate the total number of references to a special device. This
1259 * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1260 * an overloaded field. Since udev2dev can now return NODEV, we have
1261 * to check for a NULL v_rdev.
1264 count_dev(dev_t dev
)
1270 if (SLIST_FIRST(&dev
->si_hlist
)) {
1271 lwkt_gettoken(&ilock
, &spechash_token
);
1272 SLIST_FOREACH(vp
, &dev
->si_hlist
, v_specnext
) {
1273 count
+= vp
->v_usecount
;
1275 lwkt_reltoken(&ilock
);
1281 count_udev(udev_t udev
)
1285 if ((dev
= udev2dev(udev
, 0)) == NODEV
)
1287 return(count_dev(dev
));
1291 vcount(struct vnode
*vp
)
1293 if (vp
->v_rdev
== NULL
)
1295 return(count_dev(vp
->v_rdev
));
1299 * Print out a description of a vnode.
1301 static char *typename
[] =
1302 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1305 vprint(char *label
, struct vnode
*vp
)
1310 printf("%s: %p: ", label
, (void *)vp
);
1312 printf("%p: ", (void *)vp
);
1313 printf("type %s, usecount %d, writecount %d, refcount %d,",
1314 typename
[vp
->v_type
], vp
->v_usecount
, vp
->v_writecount
,
1317 if (vp
->v_flag
& VROOT
)
1318 strcat(buf
, "|VROOT");
1319 if (vp
->v_flag
& VTEXT
)
1320 strcat(buf
, "|VTEXT");
1321 if (vp
->v_flag
& VSYSTEM
)
1322 strcat(buf
, "|VSYSTEM");
1323 if (vp
->v_flag
& VFREE
)
1324 strcat(buf
, "|VFREE");
1325 if (vp
->v_flag
& VOBJBUF
)
1326 strcat(buf
, "|VOBJBUF");
1328 printf(" flags (%s)", &buf
[1]);
1329 if (vp
->v_data
== NULL
) {
1338 #include <ddb/ddb.h>
1340 static int db_show_locked_vnodes(struct mount
*mp
, void *data
);
1343 * List all of the locked vnodes in the system.
1344 * Called when debugging the kernel.
1346 DB_SHOW_COMMAND(lockedvnodes
, lockedvnodes
)
1348 printf("Locked vnodes\n");
1349 mountlist_scan(db_show_locked_vnodes
, NULL
,
1350 MNTSCAN_FORWARD
|MNTSCAN_NOBUSY
);
1354 db_show_locked_vnodes(struct mount
*mp
, void *data __unused
)
1358 TAILQ_FOREACH(vp
, &mp
->mnt_nvnodelist
, v_nmntvnodes
) {
1359 if (VOP_ISLOCKED(vp
, NULL
))
1360 vprint((char *)0, vp
);
1367 * Top level filesystem related information gathering.
1369 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS
);
1372 vfs_sysctl(SYSCTL_HANDLER_ARGS
)
1374 int *name
= (int *)arg1
- 1; /* XXX */
1375 u_int namelen
= arg2
+ 1; /* XXX */
1376 struct vfsconf
*vfsp
;
1378 #if 1 || defined(COMPAT_PRELITE2)
1379 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1381 return (sysctl_ovfs_conf(oidp
, arg1
, arg2
, req
));
1385 /* all sysctl names at this level are at least name and field */
1387 return (ENOTDIR
); /* overloaded */
1388 if (name
[0] != VFS_GENERIC
) {
1389 for (vfsp
= vfsconf
; vfsp
; vfsp
= vfsp
->vfc_next
)
1390 if (vfsp
->vfc_typenum
== name
[0])
1393 return (EOPNOTSUPP
);
1394 return ((*vfsp
->vfc_vfsops
->vfs_sysctl
)(&name
[1], namelen
- 1,
1395 oldp
, oldlenp
, newp
, newlen
, p
));
1399 case VFS_MAXTYPENUM
:
1402 return (SYSCTL_OUT(req
, &maxvfsconf
, sizeof(int)));
1405 return (ENOTDIR
); /* overloaded */
1406 for (vfsp
= vfsconf
; vfsp
; vfsp
= vfsp
->vfc_next
)
1407 if (vfsp
->vfc_typenum
== name
[2])
1410 return (EOPNOTSUPP
);
1411 return (SYSCTL_OUT(req
, vfsp
, sizeof *vfsp
));
1413 return (EOPNOTSUPP
);
1416 SYSCTL_NODE(_vfs
, VFS_GENERIC
, generic
, CTLFLAG_RD
, vfs_sysctl
,
1417 "Generic filesystem");
1419 #if 1 || defined(COMPAT_PRELITE2)
1422 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS
)
1425 struct vfsconf
*vfsp
;
1426 struct ovfsconf ovfs
;
1428 for (vfsp
= vfsconf
; vfsp
; vfsp
= vfsp
->vfc_next
) {
1429 bzero(&ovfs
, sizeof(ovfs
));
1430 ovfs
.vfc_vfsops
= vfsp
->vfc_vfsops
; /* XXX used as flag */
1431 strcpy(ovfs
.vfc_name
, vfsp
->vfc_name
);
1432 ovfs
.vfc_index
= vfsp
->vfc_typenum
;
1433 ovfs
.vfc_refcount
= vfsp
->vfc_refcount
;
1434 ovfs
.vfc_flags
= vfsp
->vfc_flags
;
1435 error
= SYSCTL_OUT(req
, &ovfs
, sizeof ovfs
);
1442 #endif /* 1 || COMPAT_PRELITE2 */
1445 * Check to see if a filesystem is mounted on a block device.
1448 vfs_mountedon(struct vnode
*vp
)
1452 if ((dev
= vp
->v_rdev
) == NULL
)
1453 dev
= udev2dev(vp
->v_udev
, (vp
->v_type
== VBLK
));
1454 if (dev
!= NODEV
&& dev
->si_mountpoint
)
1460 * Unmount all filesystems. The list is traversed in reverse order
1461 * of mounting to avoid dependencies.
1464 static int vfs_umountall_callback(struct mount
*mp
, void *data
);
1467 vfs_unmountall(void)
1469 struct thread
*td
= curthread
;
1472 if (td
->td_proc
== NULL
)
1473 td
= initproc
->p_thread
; /* XXX XXX use proc0 instead? */
1476 count
= mountlist_scan(vfs_umountall_callback
,
1477 &td
, MNTSCAN_REVERSE
|MNTSCAN_NOBUSY
);
1483 vfs_umountall_callback(struct mount
*mp
, void *data
)
1485 struct thread
*td
= *(struct thread
**)data
;
1488 error
= dounmount(mp
, MNT_FORCE
, td
);
1490 mountlist_remove(mp
);
1491 printf("unmount of filesystem mounted from %s failed (",
1492 mp
->mnt_stat
.f_mntfromname
);
1496 printf("%d)\n", error
);
1502 * Build hash lists of net addresses and hang them off the mount point.
1503 * Called by ufs_mount() to set up the lists of export addresses.
1506 vfs_hang_addrlist(struct mount
*mp
, struct netexport
*nep
,
1507 struct export_args
*argp
)
1510 struct radix_node_head
*rnh
;
1512 struct radix_node
*rn
;
1513 struct sockaddr
*saddr
, *smask
= 0;
1517 if (argp
->ex_addrlen
== 0) {
1518 if (mp
->mnt_flag
& MNT_DEFEXPORTED
)
1520 np
= &nep
->ne_defexported
;
1521 np
->netc_exflags
= argp
->ex_flags
;
1522 np
->netc_anon
= argp
->ex_anon
;
1523 np
->netc_anon
.cr_ref
= 1;
1524 mp
->mnt_flag
|= MNT_DEFEXPORTED
;
1528 if (argp
->ex_addrlen
< 0 || argp
->ex_addrlen
> MLEN
)
1530 if (argp
->ex_masklen
< 0 || argp
->ex_masklen
> MLEN
)
1533 i
= sizeof(struct netcred
) + argp
->ex_addrlen
+ argp
->ex_masklen
;
1534 np
= (struct netcred
*) malloc(i
, M_NETADDR
, M_WAITOK
);
1535 bzero((caddr_t
) np
, i
);
1536 saddr
= (struct sockaddr
*) (np
+ 1);
1537 if ((error
= copyin(argp
->ex_addr
, (caddr_t
) saddr
, argp
->ex_addrlen
)))
1539 if (saddr
->sa_len
> argp
->ex_addrlen
)
1540 saddr
->sa_len
= argp
->ex_addrlen
;
1541 if (argp
->ex_masklen
) {
1542 smask
= (struct sockaddr
*)((caddr_t
)saddr
+ argp
->ex_addrlen
);
1543 error
= copyin(argp
->ex_mask
, (caddr_t
)smask
, argp
->ex_masklen
);
1546 if (smask
->sa_len
> argp
->ex_masklen
)
1547 smask
->sa_len
= argp
->ex_masklen
;
1549 i
= saddr
->sa_family
;
1550 if ((rnh
= nep
->ne_rtable
[i
]) == 0) {
1552 * Seems silly to initialize every AF when most are not used,
1553 * do so on demand here
1555 SLIST_FOREACH(dom
, &domains
, dom_next
)
1556 if (dom
->dom_family
== i
&& dom
->dom_rtattach
) {
1557 dom
->dom_rtattach((void **) &nep
->ne_rtable
[i
],
1561 if ((rnh
= nep
->ne_rtable
[i
]) == 0) {
1566 rn
= (*rnh
->rnh_addaddr
) ((char *) saddr
, (char *) smask
, rnh
,
1568 if (rn
== 0 || np
!= (struct netcred
*) rn
) { /* already exists */
1572 np
->netc_exflags
= argp
->ex_flags
;
1573 np
->netc_anon
= argp
->ex_anon
;
1574 np
->netc_anon
.cr_ref
= 1;
1577 free(np
, M_NETADDR
);
1583 vfs_free_netcred(struct radix_node
*rn
, void *w
)
1585 struct radix_node_head
*rnh
= (struct radix_node_head
*) w
;
1587 (*rnh
->rnh_deladdr
) (rn
->rn_key
, rn
->rn_mask
, rnh
);
1588 free((caddr_t
) rn
, M_NETADDR
);
1593 * Free the net address hash lists that are hanging off the mount points.
1596 vfs_free_addrlist(struct netexport
*nep
)
1599 struct radix_node_head
*rnh
;
1601 for (i
= 0; i
<= AF_MAX
; i
++)
1602 if ((rnh
= nep
->ne_rtable
[i
])) {
1603 (*rnh
->rnh_walktree
) (rnh
, vfs_free_netcred
,
1605 free((caddr_t
) rnh
, M_RTABLE
);
1606 nep
->ne_rtable
[i
] = 0;
1611 vfs_export(struct mount
*mp
, struct netexport
*nep
, struct export_args
*argp
)
1615 if (argp
->ex_flags
& MNT_DELEXPORT
) {
1616 if (mp
->mnt_flag
& MNT_EXPUBLIC
) {
1617 vfs_setpublicfs(NULL
, NULL
, NULL
);
1618 mp
->mnt_flag
&= ~MNT_EXPUBLIC
;
1620 vfs_free_addrlist(nep
);
1621 mp
->mnt_flag
&= ~(MNT_EXPORTED
| MNT_DEFEXPORTED
);
1623 if (argp
->ex_flags
& MNT_EXPORTED
) {
1624 if (argp
->ex_flags
& MNT_EXPUBLIC
) {
1625 if ((error
= vfs_setpublicfs(mp
, nep
, argp
)) != 0)
1627 mp
->mnt_flag
|= MNT_EXPUBLIC
;
1629 if ((error
= vfs_hang_addrlist(mp
, nep
, argp
)))
1631 mp
->mnt_flag
|= MNT_EXPORTED
;
1638 * Set the publicly exported filesystem (WebNFS). Currently, only
1639 * one public filesystem is possible in the spec (RFC 2054 and 2055)
1642 vfs_setpublicfs(struct mount
*mp
, struct netexport
*nep
,
1643 struct export_args
*argp
)
1650 * mp == NULL -> invalidate the current info, the FS is
1651 * no longer exported. May be called from either vfs_export
1652 * or unmount, so check if it hasn't already been done.
1655 if (nfs_pub
.np_valid
) {
1656 nfs_pub
.np_valid
= 0;
1657 if (nfs_pub
.np_index
!= NULL
) {
1658 FREE(nfs_pub
.np_index
, M_TEMP
);
1659 nfs_pub
.np_index
= NULL
;
1666 * Only one allowed at a time.
1668 if (nfs_pub
.np_valid
!= 0 && mp
!= nfs_pub
.np_mount
)
1672 * Get real filehandle for root of exported FS.
1674 bzero((caddr_t
)&nfs_pub
.np_handle
, sizeof(nfs_pub
.np_handle
));
1675 nfs_pub
.np_handle
.fh_fsid
= mp
->mnt_stat
.f_fsid
;
1677 if ((error
= VFS_ROOT(mp
, &rvp
)))
1680 if ((error
= VFS_VPTOFH(rvp
, &nfs_pub
.np_handle
.fh_fid
)))
1686 * If an indexfile was specified, pull it in.
1688 if (argp
->ex_indexfile
!= NULL
) {
1691 error
= vn_get_namelen(rvp
, &namelen
);
1694 MALLOC(nfs_pub
.np_index
, char *, namelen
, M_TEMP
,
1696 error
= copyinstr(argp
->ex_indexfile
, nfs_pub
.np_index
,
1697 namelen
, (size_t *)0);
1700 * Check for illegal filenames.
1702 for (cp
= nfs_pub
.np_index
; *cp
; cp
++) {
1710 FREE(nfs_pub
.np_index
, M_TEMP
);
1715 nfs_pub
.np_mount
= mp
;
1716 nfs_pub
.np_valid
= 1;
1721 vfs_export_lookup(struct mount
*mp
, struct netexport
*nep
,
1722 struct sockaddr
*nam
)
1725 struct radix_node_head
*rnh
;
1726 struct sockaddr
*saddr
;
1729 if (mp
->mnt_flag
& MNT_EXPORTED
) {
1731 * Lookup in the export list first.
1735 rnh
= nep
->ne_rtable
[saddr
->sa_family
];
1737 np
= (struct netcred
*)
1738 (*rnh
->rnh_matchaddr
)((char *)saddr
,
1740 if (np
&& np
->netc_rnodes
->rn_flags
& RNF_ROOT
)
1745 * If no address match, use the default if it exists.
1747 if (np
== NULL
&& mp
->mnt_flag
& MNT_DEFEXPORTED
)
1748 np
= &nep
->ne_defexported
;
1754 * perform msync on all vnodes under a mount point. The mount point must
1755 * be locked. This code is also responsible for lazy-freeing unreferenced
1756 * vnodes whos VM objects no longer contain pages.
1758 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
1760 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked,
1761 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it
1762 * way up in this high level function.
1764 static int vfs_msync_scan1(struct mount
*mp
, struct vnode
*vp
, void *data
);
1765 static int vfs_msync_scan2(struct mount
*mp
, struct vnode
*vp
, void *data
);
1768 vfs_msync(struct mount
*mp
, int flags
)
1772 vmsc_flags
= VMSC_GETVP
;
1773 if (flags
!= MNT_WAIT
)
1774 vmsc_flags
|= VMSC_NOWAIT
;
1775 vmntvnodescan(mp
, vmsc_flags
, vfs_msync_scan1
, vfs_msync_scan2
,
1780 * scan1 is a fast pre-check. There could be hundreds of thousands of
1781 * vnodes, we cannot afford to do anything heavy weight until we have a
1782 * fairly good indication that there is work to do.
1786 vfs_msync_scan1(struct mount
*mp
, struct vnode
*vp
, void *data
)
1788 int flags
= (int)data
;
1790 if ((vp
->v_flag
& VRECLAIMED
) == 0) {
1791 if (vshouldfree(vp
, 0))
1792 return(0); /* call scan2 */
1793 if ((mp
->mnt_flag
& MNT_RDONLY
) == 0 &&
1794 (vp
->v_flag
& VOBJDIRTY
) &&
1795 (flags
== MNT_WAIT
|| VOP_ISLOCKED(vp
, NULL
) == 0)) {
1796 return(0); /* call scan2 */
1801 * do not call scan2, continue the loop
1807 * This callback is handed a locked vnode.
1811 vfs_msync_scan2(struct mount
*mp
, struct vnode
*vp
, void *data
)
1814 int flags
= (int)data
;
1816 if (vp
->v_flag
& VRECLAIMED
)
1819 if ((mp
->mnt_flag
& MNT_RDONLY
) == 0 &&
1820 (vp
->v_flag
& VOBJDIRTY
)) {
1821 if (VOP_GETVOBJECT(vp
, &obj
) == 0) {
1822 vm_object_page_clean(obj
, 0, 0,
1823 flags
== MNT_WAIT
? OBJPC_SYNC
: OBJPC_NOSYNC
);
1830 * Create the VM object needed for VMIO and mmap support. This
1831 * is done for all VREG files in the system. Some filesystems might
1832 * afford the additional metadata buffering capability of the
1833 * VMIO code by making the device node be VMIO mode also.
1835 * vp must be locked when vfs_object_create is called.
1838 vfs_object_create(struct vnode
*vp
, struct thread
*td
)
1840 return (VOP_CREATEVOBJECT(vp
, td
));
1844 * Record a process's interest in events which might happen to
1845 * a vnode. Because poll uses the historic select-style interface
1846 * internally, this routine serves as both the ``check for any
1847 * pending events'' and the ``record my interest in future events''
1848 * functions. (These are done together, while the lock is held,
1849 * to avoid race conditions.)
1852 vn_pollrecord(struct vnode
*vp
, struct thread
*td
, int events
)
1856 lwkt_gettoken(&ilock
, &vp
->v_pollinfo
.vpi_token
);
1857 if (vp
->v_pollinfo
.vpi_revents
& events
) {
1859 * This leaves events we are not interested
1860 * in available for the other process which
1861 * which presumably had requested them
1862 * (otherwise they would never have been
1865 events
&= vp
->v_pollinfo
.vpi_revents
;
1866 vp
->v_pollinfo
.vpi_revents
&= ~events
;
1868 lwkt_reltoken(&ilock
);
1871 vp
->v_pollinfo
.vpi_events
|= events
;
1872 selrecord(td
, &vp
->v_pollinfo
.vpi_selinfo
);
1873 lwkt_reltoken(&ilock
);
1878 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
1879 * it is possible for us to miss an event due to race conditions, but
1880 * that condition is expected to be rare, so for the moment it is the
1881 * preferred interface.
1884 vn_pollevent(struct vnode
*vp
, int events
)
1888 lwkt_gettoken(&ilock
, &vp
->v_pollinfo
.vpi_token
);
1889 if (vp
->v_pollinfo
.vpi_events
& events
) {
1891 * We clear vpi_events so that we don't
1892 * call selwakeup() twice if two events are
1893 * posted before the polling process(es) is
1894 * awakened. This also ensures that we take at
1895 * most one selwakeup() if the polling process
1896 * is no longer interested. However, it does
1897 * mean that only one event can be noticed at
1898 * a time. (Perhaps we should only clear those
1899 * event bits which we note?) XXX
1901 vp
->v_pollinfo
.vpi_events
= 0; /* &= ~events ??? */
1902 vp
->v_pollinfo
.vpi_revents
|= events
;
1903 selwakeup(&vp
->v_pollinfo
.vpi_selinfo
);
1905 lwkt_reltoken(&ilock
);
1909 * Wake up anyone polling on vp because it is being revoked.
1910 * This depends on dead_poll() returning POLLHUP for correct
1914 vn_pollgone(struct vnode
*vp
)
1918 lwkt_gettoken(&ilock
, &vp
->v_pollinfo
.vpi_token
);
1919 if (vp
->v_pollinfo
.vpi_events
) {
1920 vp
->v_pollinfo
.vpi_events
= 0;
1921 selwakeup(&vp
->v_pollinfo
.vpi_selinfo
);
1923 lwkt_reltoken(&ilock
);
1927 * extract the dev_t from a VBLK or VCHR. The vnode must have been opened
1928 * (or v_rdev might be NULL).
1931 vn_todev(struct vnode
*vp
)
1933 if (vp
->v_type
!= VBLK
&& vp
->v_type
!= VCHR
)
1935 KKASSERT(vp
->v_rdev
!= NULL
);
1936 return (vp
->v_rdev
);
1940 * Check if vnode represents a disk device. The vnode does not need to be
1944 vn_isdisk(struct vnode
*vp
, int *errp
)
1948 if (vp
->v_type
!= VBLK
&& vp
->v_type
!= VCHR
) {
1954 if ((dev
= vp
->v_rdev
) == NULL
)
1955 dev
= udev2dev(vp
->v_udev
, (vp
->v_type
== VBLK
));
1956 if (dev
== NULL
|| dev
== NODEV
) {
1961 if (dev_is_good(dev
) == 0) {
1966 if ((dev_dflags(dev
) & D_DISK
) == 0) {
1976 #ifdef DEBUG_VFS_LOCKS
1979 assert_vop_locked(struct vnode
*vp
, const char *str
)
1981 if (vp
&& IS_LOCKING_VFS(vp
) && !VOP_ISLOCKED(vp
, NULL
)) {
1982 panic("%s: %p is not locked shared but should be", str
, vp
);
1987 assert_vop_unlocked(struct vnode
*vp
, const char *str
)
1989 if (vp
&& IS_LOCKING_VFS(vp
)) {
1990 if (VOP_ISLOCKED(vp
, curthread
) == LK_EXCLUSIVE
) {
1991 panic("%s: %p is locked but should not be", str
, vp
);
1999 vn_get_namelen(struct vnode
*vp
, int *namelen
)
2001 int error
, retval
[2];
2003 error
= VOP_PATHCONF(vp
, _PC_NAME_MAX
, retval
);
2011 vop_write_dirent(int *error
, struct uio
*uio
, ino_t d_ino
, uint8_t d_type
,
2012 uint16_t d_namlen
, const char *d_name
)
2017 len
= _DIRENT_RECLEN(d_namlen
);
2018 if (len
> uio
->uio_resid
)
2021 dp
= malloc(len
, M_TEMP
, M_WAITOK
| M_ZERO
);
2024 dp
->d_namlen
= d_namlen
;
2025 dp
->d_type
= d_type
;
2026 bcopy(d_name
, dp
->d_name
, d_namlen
);
2028 *error
= uiomove((caddr_t
)dp
, len
, uio
);