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 1.105 2007/06/08 02:00:45 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>
86 #include <sys/sysref2.h>
88 static MALLOC_DEFINE(M_NETADDR
, "Export Host", "Export host address structure");
91 SYSCTL_INT(_debug
, OID_AUTO
, numvnodes
, CTLFLAG_RD
, &numvnodes
, 0, "");
93 SYSCTL_INT(_vfs
, OID_AUTO
, fastdev
, CTLFLAG_RW
, &vfs_fastdev
, 0, "");
95 enum vtype iftovt_tab
[16] = {
96 VNON
, VFIFO
, VCHR
, VNON
, VDIR
, VNON
, VBLK
, VNON
,
97 VREG
, VNON
, VLNK
, VNON
, VSOCK
, VNON
, VNON
, VBAD
,
100 0, S_IFREG
, S_IFDIR
, S_IFBLK
, S_IFCHR
, S_IFLNK
,
101 S_IFSOCK
, S_IFIFO
, S_IFMT
,
104 static int reassignbufcalls
;
105 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufcalls
, CTLFLAG_RW
,
106 &reassignbufcalls
, 0, "");
107 static int reassignbufloops
;
108 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufloops
, CTLFLAG_RW
,
109 &reassignbufloops
, 0, "");
110 static int reassignbufsortgood
;
111 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufsortgood
, CTLFLAG_RW
,
112 &reassignbufsortgood
, 0, "");
113 static int reassignbufsortbad
;
114 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufsortbad
, CTLFLAG_RW
,
115 &reassignbufsortbad
, 0, "");
116 static int reassignbufmethod
= 1;
117 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufmethod
, CTLFLAG_RW
,
118 &reassignbufmethod
, 0, "");
120 int nfs_mount_type
= -1;
121 static struct lwkt_token spechash_token
;
122 struct nfs_public nfs_pub
; /* publicly exported FS */
125 SYSCTL_INT(_kern
, KERN_MAXVNODES
, maxvnodes
, CTLFLAG_RW
,
126 &desiredvnodes
, 0, "Maximum number of vnodes");
128 static void vfs_free_addrlist (struct netexport
*nep
);
129 static int vfs_free_netcred (struct radix_node
*rn
, void *w
);
130 static int vfs_hang_addrlist (struct mount
*mp
, struct netexport
*nep
,
131 struct export_args
*argp
);
133 extern int dev_ref_debug
;
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
, off_t
, b_loffset
);
140 RB_GENERATE2(buf_rb_hash
, buf
, b_rbhash
, rb_buf_compare
, off_t
, b_loffset
);
143 rb_buf_compare(struct buf
*b1
, struct buf
*b2
)
145 if (b1
->b_loffset
< b2
->b_loffset
)
147 if (b1
->b_loffset
> b2
->b_loffset
)
153 * Returns non-zero if the vnode is a candidate for lazy msyncing.
156 vshouldmsync(struct vnode
*vp
)
158 if (vp
->v_auxrefs
!= 0 || vp
->v_sysref
.refcnt
> 0)
159 return (0); /* other holders */
161 (vp
->v_object
->ref_count
|| vp
->v_object
->resident_page_count
)) {
168 * Initialize the vnode management data structures.
170 * Called from vfsinit()
176 * Desired vnodes is a result of the physical page count
177 * and the size of kernel's heap. It scales in proportion
178 * to the amount of available physical memory. This can
179 * cause trouble on 64-bit and large memory platforms.
181 /* desiredvnodes = maxproc + vmstats.v_page_count / 4; */
183 min(maxproc
+ vmstats
.v_page_count
/ 4,
185 (5 * (sizeof(struct vm_object
) + sizeof(struct vnode
))));
187 lwkt_token_init(&spechash_token
);
191 * Knob to control the precision of file timestamps:
193 * 0 = seconds only; nanoseconds zeroed.
194 * 1 = seconds and nanoseconds, accurate within 1/HZ.
195 * 2 = seconds and nanoseconds, truncated to microseconds.
196 * >=3 = seconds and nanoseconds, maximum precision.
198 enum { TSP_SEC
, TSP_HZ
, TSP_USEC
, TSP_NSEC
};
200 static int timestamp_precision
= TSP_SEC
;
201 SYSCTL_INT(_vfs
, OID_AUTO
, timestamp_precision
, CTLFLAG_RW
,
202 ×tamp_precision
, 0, "");
205 * Get a current timestamp.
208 vfs_timestamp(struct timespec
*tsp
)
212 switch (timestamp_precision
) {
214 tsp
->tv_sec
= time_second
;
222 TIMEVAL_TO_TIMESPEC(&tv
, tsp
);
232 * Set vnode attributes to VNOVAL
235 vattr_null(struct vattr
*vap
)
238 vap
->va_size
= VNOVAL
;
239 vap
->va_bytes
= VNOVAL
;
240 vap
->va_mode
= VNOVAL
;
241 vap
->va_nlink
= VNOVAL
;
242 vap
->va_uid
= VNOVAL
;
243 vap
->va_gid
= VNOVAL
;
244 vap
->va_fsid
= VNOVAL
;
245 vap
->va_fileid
= VNOVAL
;
246 vap
->va_blocksize
= VNOVAL
;
247 vap
->va_rmajor
= VNOVAL
;
248 vap
->va_rminor
= VNOVAL
;
249 vap
->va_atime
.tv_sec
= VNOVAL
;
250 vap
->va_atime
.tv_nsec
= VNOVAL
;
251 vap
->va_mtime
.tv_sec
= VNOVAL
;
252 vap
->va_mtime
.tv_nsec
= VNOVAL
;
253 vap
->va_ctime
.tv_sec
= VNOVAL
;
254 vap
->va_ctime
.tv_nsec
= VNOVAL
;
255 vap
->va_flags
= VNOVAL
;
256 vap
->va_gen
= VNOVAL
;
258 vap
->va_fsmid
= VNOVAL
;
262 * Flush out and invalidate all buffers associated with a vnode.
266 static int vinvalbuf_bp(struct buf
*bp
, void *data
);
268 struct vinvalbuf_bp_info
{
276 vupdatefsmid(struct vnode
*vp
)
278 atomic_set_int(&vp
->v_flag
, VFSMID
);
282 vinvalbuf(struct vnode
*vp
, int flags
, int slpflag
, int slptimeo
)
284 struct vinvalbuf_bp_info info
;
289 * If we are being asked to save, call fsync to ensure that the inode
292 if (flags
& V_SAVE
) {
294 while (vp
->v_track_write
.bk_active
) {
295 vp
->v_track_write
.bk_waitflag
= 1;
296 error
= tsleep(&vp
->v_track_write
, slpflag
,
297 "vinvlbuf", slptimeo
);
303 if (!RB_EMPTY(&vp
->v_rbdirty_tree
)) {
305 if ((error
= VOP_FSYNC(vp
, MNT_WAIT
)) != 0)
308 if (vp
->v_track_write
.bk_active
> 0 ||
309 !RB_EMPTY(&vp
->v_rbdirty_tree
))
310 panic("vinvalbuf: dirty bufs");
315 info
.slptimeo
= slptimeo
;
316 info
.lkflags
= LK_EXCLUSIVE
| LK_SLEEPFAIL
;
317 if (slpflag
& PCATCH
)
318 info
.lkflags
|= LK_PCATCH
;
323 * Flush the buffer cache until nothing is left.
325 while (!RB_EMPTY(&vp
->v_rbclean_tree
) ||
326 !RB_EMPTY(&vp
->v_rbdirty_tree
)) {
327 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbclean_tree
, NULL
,
328 vinvalbuf_bp
, &info
);
330 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, NULL
,
331 vinvalbuf_bp
, &info
);
336 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
337 * have write I/O in-progress but if there is a VM object then the
338 * VM object can also have read-I/O in-progress.
341 while (vp
->v_track_write
.bk_active
> 0) {
342 vp
->v_track_write
.bk_waitflag
= 1;
343 tsleep(&vp
->v_track_write
, 0, "vnvlbv", 0);
345 if ((object
= vp
->v_object
) != NULL
) {
346 while (object
->paging_in_progress
)
347 vm_object_pip_sleep(object
, "vnvlbx");
349 } while (vp
->v_track_write
.bk_active
> 0);
354 * Destroy the copy in the VM cache, too.
356 if ((object
= vp
->v_object
) != NULL
) {
357 vm_object_page_remove(object
, 0, 0,
358 (flags
& V_SAVE
) ? TRUE
: FALSE
);
361 if (!RB_EMPTY(&vp
->v_rbdirty_tree
) || !RB_EMPTY(&vp
->v_rbclean_tree
))
362 panic("vinvalbuf: flush failed");
363 if (!RB_EMPTY(&vp
->v_rbhash_tree
))
364 panic("vinvalbuf: flush failed, buffers still present");
369 vinvalbuf_bp(struct buf
*bp
, void *data
)
371 struct vinvalbuf_bp_info
*info
= data
;
374 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
)) {
375 error
= BUF_TIMELOCK(bp
, info
->lkflags
,
376 "vinvalbuf", info
->slptimeo
);
386 KKASSERT(bp
->b_vp
== info
->vp
);
389 * XXX Since there are no node locks for NFS, I
390 * believe there is a slight chance that a delayed
391 * write will occur while sleeping just above, so
392 * check for it. Note that vfs_bio_awrite expects
393 * buffers to reside on a queue, while bwrite() and
396 if (((bp
->b_flags
& (B_DELWRI
| B_INVAL
)) == B_DELWRI
) &&
397 (info
->flags
& V_SAVE
)) {
398 if (bp
->b_vp
== info
->vp
) {
399 if (bp
->b_flags
& B_CLUSTEROK
) {
403 bp
->b_flags
|= B_ASYNC
;
410 } else if (info
->flags
& V_SAVE
) {
412 * Cannot set B_NOCACHE on a clean buffer as this will
413 * destroy the VM backing store which might actually
414 * be dirty (and unsynchronized).
417 bp
->b_flags
|= (B_INVAL
| B_RELBUF
);
418 bp
->b_flags
&= ~B_ASYNC
;
422 bp
->b_flags
|= (B_INVAL
| B_NOCACHE
| B_RELBUF
);
423 bp
->b_flags
&= ~B_ASYNC
;
430 * Truncate a file's buffer and pages to a specified length. This
431 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
434 * The vnode must be locked.
436 static int vtruncbuf_bp_trunc_cmp(struct buf
*bp
, void *data
);
437 static int vtruncbuf_bp_trunc(struct buf
*bp
, void *data
);
438 static int vtruncbuf_bp_metasync_cmp(struct buf
*bp
, void *data
);
439 static int vtruncbuf_bp_metasync(struct buf
*bp
, void *data
);
442 vtruncbuf(struct vnode
*vp
, off_t length
, int blksize
)
446 const char *filename
;
449 * Round up to the *next* block, then destroy the buffers in question.
450 * Since we are only removing some of the buffers we must rely on the
451 * scan count to determine whether a loop is necessary.
453 if ((count
= (int)(length
% blksize
)) != 0)
454 truncloffset
= length
+ (blksize
- count
);
456 truncloffset
= length
;
460 count
= RB_SCAN(buf_rb_tree
, &vp
->v_rbclean_tree
,
461 vtruncbuf_bp_trunc_cmp
,
462 vtruncbuf_bp_trunc
, &truncloffset
);
463 count
+= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
464 vtruncbuf_bp_trunc_cmp
,
465 vtruncbuf_bp_trunc
, &truncloffset
);
469 * For safety, fsync any remaining metadata if the file is not being
470 * truncated to 0. Since the metadata does not represent the entire
471 * dirty list we have to rely on the hit count to ensure that we get
476 count
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
477 vtruncbuf_bp_metasync_cmp
,
478 vtruncbuf_bp_metasync
, vp
);
483 * Clean out any left over VM backing store.
487 vnode_pager_setsize(vp
, length
);
492 * It is possible to have in-progress I/O from buffers that were
493 * not part of the truncation. This should not happen if we
494 * are truncating to 0-length.
496 filename
= TAILQ_FIRST(&vp
->v_namecache
) ?
497 TAILQ_FIRST(&vp
->v_namecache
)->nc_name
: "?";
499 while ((count
= vp
->v_track_write
.bk_active
) > 0) {
500 vp
->v_track_write
.bk_waitflag
= 1;
501 tsleep(&vp
->v_track_write
, 0, "vbtrunc", 0);
503 kprintf("Warning: vtruncbuf(): Had to wait for "
504 "%d buffer I/Os to finish in %s\n",
510 * Make sure no buffers were instantiated while we were trying
511 * to clean out the remaining VM pages. This could occur due
512 * to busy dirty VM pages being flushed out to disk.
515 count
= RB_SCAN(buf_rb_tree
, &vp
->v_rbclean_tree
,
516 vtruncbuf_bp_trunc_cmp
,
517 vtruncbuf_bp_trunc
, &truncloffset
);
518 count
+= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
519 vtruncbuf_bp_trunc_cmp
,
520 vtruncbuf_bp_trunc
, &truncloffset
);
522 kprintf("Warning: vtruncbuf(): Had to re-clean %d "
523 "left over buffers in %s\n", count
, filename
);
533 * The callback buffer is beyond the new file EOF and must be destroyed.
534 * Note that the compare function must conform to the RB_SCAN's requirements.
538 vtruncbuf_bp_trunc_cmp(struct buf
*bp
, void *data
)
540 if (bp
->b_loffset
>= *(off_t
*)data
)
547 vtruncbuf_bp_trunc(struct buf
*bp
, void *data
)
550 * Do not try to use a buffer we cannot immediately lock, but sleep
551 * anyway to prevent a livelock. The code will loop until all buffers
554 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
)) {
555 if (BUF_LOCK(bp
, LK_EXCLUSIVE
|LK_SLEEPFAIL
) == 0)
559 bp
->b_flags
|= (B_INVAL
| B_RELBUF
| B_NOCACHE
);
560 bp
->b_flags
&= ~B_ASYNC
;
567 * Fsync all meta-data after truncating a file to be non-zero. Only metadata
568 * blocks (with a negative loffset) are scanned.
569 * Note that the compare function must conform to the RB_SCAN's requirements.
572 vtruncbuf_bp_metasync_cmp(struct buf
*bp
, void *data
)
574 if (bp
->b_loffset
< 0)
580 vtruncbuf_bp_metasync(struct buf
*bp
, void *data
)
582 struct vnode
*vp
= data
;
584 if (bp
->b_flags
& B_DELWRI
) {
586 * Do not try to use a buffer we cannot immediately lock,
587 * but sleep anyway to prevent a livelock. The code will
588 * loop until all buffers can be acted upon.
590 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
)) {
591 if (BUF_LOCK(bp
, LK_EXCLUSIVE
|LK_SLEEPFAIL
) == 0)
595 if (bp
->b_vp
== vp
) {
596 bp
->b_flags
|= B_ASYNC
;
598 bp
->b_flags
&= ~B_ASYNC
;
609 * vfsync - implements a multipass fsync on a file which understands
610 * dependancies and meta-data. The passed vnode must be locked. The
611 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY.
613 * When fsyncing data asynchronously just do one consolidated pass starting
614 * with the most negative block number. This may not get all the data due
617 * When fsyncing data synchronously do a data pass, then a metadata pass,
618 * then do additional data+metadata passes to try to get all the data out.
620 static int vfsync_wait_output(struct vnode
*vp
,
621 int (*waitoutput
)(struct vnode
*, struct thread
*));
622 static int vfsync_data_only_cmp(struct buf
*bp
, void *data
);
623 static int vfsync_meta_only_cmp(struct buf
*bp
, void *data
);
624 static int vfsync_lazy_range_cmp(struct buf
*bp
, void *data
);
625 static int vfsync_bp(struct buf
*bp
, void *data
);
634 int (*checkdef
)(struct buf
*);
638 vfsync(struct vnode
*vp
, int waitfor
, int passes
,
639 int (*checkdef
)(struct buf
*),
640 int (*waitoutput
)(struct vnode
*, struct thread
*))
642 struct vfsync_info info
;
645 bzero(&info
, sizeof(info
));
647 if ((info
.checkdef
= checkdef
) == NULL
)
650 crit_enter_id("vfsync");
655 * Lazy (filesystem syncer typ) Asynchronous plus limit the
656 * number of data (not meta) pages we try to flush to 1MB.
657 * A non-zero return means that lazy limit was reached.
659 info
.lazylimit
= 1024 * 1024;
661 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
662 vfsync_lazy_range_cmp
, vfsync_bp
, &info
);
663 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
664 vfsync_meta_only_cmp
, vfsync_bp
, &info
);
667 else if (!RB_EMPTY(&vp
->v_rbdirty_tree
))
668 vn_syncer_add_to_worklist(vp
, 1);
673 * Asynchronous. Do a data-only pass and a meta-only pass.
676 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, vfsync_data_only_cmp
,
678 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, vfsync_meta_only_cmp
,
684 * Synchronous. Do a data-only pass, then a meta-data+data
685 * pass, then additional integrated passes to try to get
686 * all the dependancies flushed.
688 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, vfsync_data_only_cmp
,
690 error
= vfsync_wait_output(vp
, waitoutput
);
692 info
.skippedbufs
= 0;
693 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, NULL
,
695 error
= vfsync_wait_output(vp
, waitoutput
);
696 if (info
.skippedbufs
)
697 kprintf("Warning: vfsync skipped %d dirty bufs in pass2!\n", info
.skippedbufs
);
699 while (error
== 0 && passes
> 0 &&
700 !RB_EMPTY(&vp
->v_rbdirty_tree
)) {
702 info
.synchronous
= 1;
705 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, NULL
,
711 error
= vfsync_wait_output(vp
, waitoutput
);
715 crit_exit_id("vfsync");
720 vfsync_wait_output(struct vnode
*vp
, int (*waitoutput
)(struct vnode
*, struct thread
*))
724 while (vp
->v_track_write
.bk_active
) {
725 vp
->v_track_write
.bk_waitflag
= 1;
726 tsleep(&vp
->v_track_write
, 0, "fsfsn", 0);
729 error
= waitoutput(vp
, curthread
);
734 vfsync_data_only_cmp(struct buf
*bp
, void *data
)
736 if (bp
->b_loffset
< 0)
742 vfsync_meta_only_cmp(struct buf
*bp
, void *data
)
744 if (bp
->b_loffset
< 0)
750 vfsync_lazy_range_cmp(struct buf
*bp
, void *data
)
752 struct vfsync_info
*info
= data
;
753 if (bp
->b_loffset
< info
->vp
->v_lazyw
)
759 vfsync_bp(struct buf
*bp
, void *data
)
761 struct vfsync_info
*info
= data
;
762 struct vnode
*vp
= info
->vp
;
766 * if syncdeps is not set we do not try to write buffers which have
769 if (!info
->synchronous
&& info
->syncdeps
== 0 && info
->checkdef(bp
))
773 * Ignore buffers that we cannot immediately lock. XXX
775 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
)) {
776 kprintf("Warning: vfsync_bp skipping dirty buffer %p\n", bp
);
780 if ((bp
->b_flags
& B_DELWRI
) == 0)
781 panic("vfsync_bp: buffer not dirty");
783 panic("vfsync_bp: buffer vp mismatch");
786 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
787 * has been written but an additional handshake with the device
788 * is required before we can dispose of the buffer. We have no idea
789 * how to do this so we have to skip these buffers.
791 if (bp
->b_flags
& B_NEEDCOMMIT
) {
796 if (info
->synchronous
) {
798 * Synchronous flushing. An error may be returned.
801 crit_exit_id("vfsync");
803 crit_enter_id("vfsync");
806 * Asynchronous flushing. A negative return value simply
807 * stops the scan and is not considered an error. We use
808 * this to support limited MNT_LAZY flushes.
810 vp
->v_lazyw
= bp
->b_loffset
;
811 if ((vp
->v_flag
& VOBJBUF
) && (bp
->b_flags
& B_CLUSTEROK
)) {
812 info
->lazycount
+= vfs_bio_awrite(bp
);
814 info
->lazycount
+= bp
->b_bufsize
;
816 crit_exit_id("vfsync");
818 crit_enter_id("vfsync");
820 if (info
->lazylimit
&& info
->lazycount
>= info
->lazylimit
)
829 * Associate a buffer with a vnode.
832 bgetvp(struct vnode
*vp
, struct buf
*bp
)
834 KASSERT(bp
->b_vp
== NULL
, ("bgetvp: not free"));
835 KKASSERT((bp
->b_flags
& (B_HASHED
|B_DELWRI
|B_VNCLEAN
|B_VNDIRTY
)) == 0);
839 * Insert onto list for new vnode.
843 bp
->b_flags
|= B_HASHED
;
844 if (buf_rb_hash_RB_INSERT(&vp
->v_rbhash_tree
, bp
))
845 panic("reassignbuf: dup lblk vp %p bp %p", vp
, bp
);
847 bp
->b_flags
|= B_VNCLEAN
;
848 if (buf_rb_tree_RB_INSERT(&vp
->v_rbclean_tree
, bp
))
849 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp
, bp
);
854 * Disassociate a buffer from a vnode.
857 brelvp(struct buf
*bp
)
861 KASSERT(bp
->b_vp
!= NULL
, ("brelvp: NULL"));
864 * Delete from old vnode list, if on one.
868 if (bp
->b_flags
& (B_VNDIRTY
| B_VNCLEAN
)) {
869 if (bp
->b_flags
& B_VNDIRTY
)
870 buf_rb_tree_RB_REMOVE(&vp
->v_rbdirty_tree
, bp
);
872 buf_rb_tree_RB_REMOVE(&vp
->v_rbclean_tree
, bp
);
873 bp
->b_flags
&= ~(B_VNDIRTY
| B_VNCLEAN
);
875 if (bp
->b_flags
& B_HASHED
) {
876 buf_rb_hash_RB_REMOVE(&vp
->v_rbhash_tree
, bp
);
877 bp
->b_flags
&= ~B_HASHED
;
879 if ((vp
->v_flag
& VONWORKLST
) && RB_EMPTY(&vp
->v_rbdirty_tree
)) {
880 vp
->v_flag
&= ~VONWORKLST
;
881 LIST_REMOVE(vp
, v_synclist
);
889 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI.
890 * This routine is called when the state of the B_DELWRI bit is changed.
893 reassignbuf(struct buf
*bp
)
895 struct vnode
*vp
= bp
->b_vp
;
898 KKASSERT(vp
!= NULL
);
902 * B_PAGING flagged buffers cannot be reassigned because their vp
903 * is not fully linked in.
905 if (bp
->b_flags
& B_PAGING
)
906 panic("cannot reassign paging buffer");
909 if (bp
->b_flags
& B_DELWRI
) {
911 * Move to the dirty list, add the vnode to the worklist
913 if (bp
->b_flags
& B_VNCLEAN
) {
914 buf_rb_tree_RB_REMOVE(&vp
->v_rbclean_tree
, bp
);
915 bp
->b_flags
&= ~B_VNCLEAN
;
917 if ((bp
->b_flags
& B_VNDIRTY
) == 0) {
918 if (buf_rb_tree_RB_INSERT(&vp
->v_rbdirty_tree
, bp
)) {
919 panic("reassignbuf: dup lblk vp %p bp %p",
922 bp
->b_flags
|= B_VNDIRTY
;
924 if ((vp
->v_flag
& VONWORKLST
) == 0) {
925 switch (vp
->v_type
) {
932 vp
->v_rdev
->si_mountpoint
!= NULL
) {
940 vn_syncer_add_to_worklist(vp
, delay
);
944 * Move to the clean list, remove the vnode from the worklist
945 * if no dirty blocks remain.
947 if (bp
->b_flags
& B_VNDIRTY
) {
948 buf_rb_tree_RB_REMOVE(&vp
->v_rbdirty_tree
, bp
);
949 bp
->b_flags
&= ~B_VNDIRTY
;
951 if ((bp
->b_flags
& B_VNCLEAN
) == 0) {
952 if (buf_rb_tree_RB_INSERT(&vp
->v_rbclean_tree
, bp
)) {
953 panic("reassignbuf: dup lblk vp %p bp %p",
956 bp
->b_flags
|= B_VNCLEAN
;
958 if ((vp
->v_flag
& VONWORKLST
) &&
959 RB_EMPTY(&vp
->v_rbdirty_tree
)) {
960 vp
->v_flag
&= ~VONWORKLST
;
961 LIST_REMOVE(vp
, v_synclist
);
968 * Create a vnode for a block device.
969 * Used for mounting the root file system.
972 bdevvp(cdev_t dev
, struct vnode
**vpp
)
982 error
= getspecialvnode(VT_NON
, NULL
, &spec_vnode_vops_p
, &nvp
, 0, 0);
989 vp
->v_umajor
= dev
->si_umajor
;
990 vp
->v_uminor
= dev
->si_uminor
;
997 v_associate_rdev(struct vnode
*vp
, cdev_t dev
)
1003 if (dev_is_good(dev
) == 0)
1005 KKASSERT(vp
->v_rdev
== NULL
);
1008 vp
->v_rdev
= reference_dev(dev
);
1009 lwkt_gettoken(&ilock
, &spechash_token
);
1010 SLIST_INSERT_HEAD(&dev
->si_hlist
, vp
, v_cdevnext
);
1011 lwkt_reltoken(&ilock
);
1016 v_release_rdev(struct vnode
*vp
)
1021 if ((dev
= vp
->v_rdev
) != NULL
) {
1022 lwkt_gettoken(&ilock
, &spechash_token
);
1023 SLIST_REMOVE(&dev
->si_hlist
, vp
, vnode
, v_cdevnext
);
1026 lwkt_reltoken(&ilock
);
1031 * Add a vnode to the alias list hung off the cdev_t. We only associate
1032 * the device number with the vnode. The actual device is not associated
1033 * until the vnode is opened (usually in spec_open()), and will be
1034 * disassociated on last close.
1037 addaliasu(struct vnode
*nvp
, int x
, int y
)
1039 if (nvp
->v_type
!= VBLK
&& nvp
->v_type
!= VCHR
)
1040 panic("addaliasu on non-special vnode");
1046 * Disassociate a vnode from its underlying filesystem.
1048 * The vnode must be VX locked and referenced. In all normal situations
1049 * there are no active references. If vclean_vxlocked() is called while
1050 * there are active references, the vnode is being ripped out and we have
1051 * to call VOP_CLOSE() as appropriate before we can reclaim it.
1054 vclean_vxlocked(struct vnode
*vp
, int flags
)
1061 * If the vnode has already been reclaimed we have nothing to do.
1063 if (vp
->v_flag
& VRECLAIMED
)
1065 vp
->v_flag
|= VRECLAIMED
;
1068 * Scrap the vfs cache
1070 while (cache_inval_vp(vp
, 0) != 0) {
1071 kprintf("Warning: vnode %p clean/cache_resolution race detected\n", vp
);
1072 tsleep(vp
, 0, "vclninv", 2);
1076 * Check to see if the vnode is in use. If so we have to reference it
1077 * before we clean it out so that its count cannot fall to zero and
1078 * generate a race against ourselves to recycle it.
1080 active
= sysref_isactive(&vp
->v_sysref
);
1083 * Clean out any buffers associated with the vnode and destroy its
1084 * object, if it has one.
1086 vinvalbuf(vp
, V_SAVE
, 0, 0);
1089 * If purging an active vnode (typically during a forced unmount
1090 * or reboot), it must be closed and deactivated before being
1091 * reclaimed. This isn't really all that safe, but what can
1094 * Note that neither of these routines unlocks the vnode.
1096 if (active
&& (flags
& DOCLOSE
)) {
1097 while ((n
= vp
->v_opencount
) != 0) {
1098 if (vp
->v_writecount
)
1099 VOP_CLOSE(vp
, FWRITE
|FNONBLOCK
);
1101 VOP_CLOSE(vp
, FNONBLOCK
);
1102 if (vp
->v_opencount
== n
) {
1103 kprintf("Warning: unable to force-close"
1111 * If the vnode has not be deactivated, deactivated it. Deactivation
1112 * can create new buffers and VM pages so we have to call vinvalbuf()
1113 * again to make sure they all get flushed.
1115 * This can occur if a file with a link count of 0 needs to be
1118 if ((vp
->v_flag
& VINACTIVE
) == 0) {
1119 vp
->v_flag
|= VINACTIVE
;
1121 vinvalbuf(vp
, V_SAVE
, 0, 0);
1125 * If the vnode has an object, destroy it.
1127 if ((object
= vp
->v_object
) != NULL
) {
1128 if (object
->ref_count
== 0) {
1129 if ((object
->flags
& OBJ_DEAD
) == 0)
1130 vm_object_terminate(object
);
1132 vm_pager_deallocate(object
);
1134 vp
->v_flag
&= ~VOBJBUF
;
1136 KKASSERT((vp
->v_flag
& VOBJBUF
) == 0);
1140 * Reclaim the vnode.
1142 if (VOP_RECLAIM(vp
))
1143 panic("vclean: cannot reclaim");
1146 * Done with purge, notify sleepers of the grim news.
1148 vp
->v_ops
= &dead_vnode_vops_p
;
1154 * Eliminate all activity associated with the requested vnode
1155 * and with all vnodes aliased to the requested vnode.
1157 * The vnode must be referenced and vx_lock()'d
1159 * revoke { struct vnode *a_vp, int a_flags }
1162 vop_stdrevoke(struct vop_revoke_args
*ap
)
1164 struct vnode
*vp
, *vq
;
1168 KASSERT((ap
->a_flags
& REVOKEALL
) != 0, ("vop_revoke"));
1173 * If the vnode is already dead don't try to revoke it
1175 if (vp
->v_flag
& VRECLAIMED
)
1179 * If the vnode has a device association, scrap all vnodes associated
1180 * with the device. Don't let the device disappear on us while we
1181 * are scrapping the vnodes.
1183 * The passed vp will probably show up in the list, do not VX lock
1186 if (vp
->v_type
!= VCHR
)
1188 if ((dev
= vp
->v_rdev
) == NULL
) {
1189 if ((dev
= get_dev(vp
->v_umajor
, vp
->v_uminor
)) == NULL
)
1193 lwkt_gettoken(&ilock
, &spechash_token
);
1194 while ((vq
= SLIST_FIRST(&dev
->si_hlist
)) != NULL
) {
1197 if (vq
== SLIST_FIRST(&dev
->si_hlist
))
1202 lwkt_reltoken(&ilock
);
1208 * This is called when the object underlying a vnode is being destroyed,
1209 * such as in a remove(). Try to recycle the vnode immediately if the
1210 * only active reference is our reference.
1213 vrecycle(struct vnode
*vp
)
1215 if (vp
->v_sysref
.refcnt
== 1) {
1223 * Eliminate all activity associated with a vnode in preparation for reuse.
1225 * The vnode must be VX locked and refd and will remain VX locked and refd
1226 * on return. This routine may be called with the vnode in any state, as
1227 * long as it is VX locked. The vnode will be cleaned out and marked
1228 * VRECLAIMED but will not actually be reused until all existing refs and
1231 * NOTE: This routine may be called on a vnode which has not yet been
1232 * already been deactivated (VOP_INACTIVE), or on a vnode which has
1233 * already been reclaimed.
1235 * This routine is not responsible for placing us back on the freelist.
1236 * Instead, it happens automatically when the caller releases the VX lock
1237 * (assuming there aren't any other references).
1241 vgone_vxlocked(struct vnode
*vp
)
1244 * assert that the VX lock is held. This is an absolute requirement
1245 * now for vgone_vxlocked() to be called.
1247 KKASSERT(vp
->v_lock
.lk_exclusivecount
== 1);
1250 * Clean out the filesystem specific data and set the VRECLAIMED
1251 * bit. Also deactivate the vnode if necessary.
1253 vclean_vxlocked(vp
, DOCLOSE
);
1256 * Delete from old mount point vnode list, if on one.
1258 if (vp
->v_mount
!= NULL
)
1259 insmntque(vp
, NULL
);
1262 * If special device, remove it from special device alias list
1263 * if it is on one. This should normally only occur if a vnode is
1264 * being revoked as the device should otherwise have been released
1267 if ((vp
->v_type
== VBLK
|| vp
->v_type
== VCHR
) && vp
->v_rdev
!= NULL
) {
1278 * Lookup a vnode by device number.
1281 vfinddev(cdev_t dev
, enum vtype type
, struct vnode
**vpp
)
1286 lwkt_gettoken(&ilock
, &spechash_token
);
1287 SLIST_FOREACH(vp
, &dev
->si_hlist
, v_cdevnext
) {
1288 if (type
== vp
->v_type
) {
1290 lwkt_reltoken(&ilock
);
1294 lwkt_reltoken(&ilock
);
1299 * Calculate the total number of references to a special device. This
1300 * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1301 * an overloaded field. Since udev2dev can now return NULL, we have
1302 * to check for a NULL v_rdev.
1305 count_dev(cdev_t dev
)
1311 if (SLIST_FIRST(&dev
->si_hlist
)) {
1312 lwkt_gettoken(&ilock
, &spechash_token
);
1313 SLIST_FOREACH(vp
, &dev
->si_hlist
, v_cdevnext
) {
1314 if (vp
->v_sysref
.refcnt
> 0)
1315 count
+= vp
->v_sysref
.refcnt
;
1317 lwkt_reltoken(&ilock
);
1323 count_udev(int x
, int y
)
1327 if ((dev
= get_dev(x
, y
)) == NULL
)
1329 return(count_dev(dev
));
1333 vcount(struct vnode
*vp
)
1335 if (vp
->v_rdev
== NULL
)
1337 return(count_dev(vp
->v_rdev
));
1341 * Initialize VMIO for a vnode. This routine MUST be called before a
1342 * VFS can issue buffer cache ops on a vnode. It is typically called
1343 * when a vnode is initialized from its inode.
1346 vinitvmio(struct vnode
*vp
, off_t filesize
)
1352 if ((object
= vp
->v_object
) == NULL
) {
1353 object
= vnode_pager_alloc(vp
, filesize
, 0, 0);
1355 * Dereference the reference we just created. This assumes
1356 * that the object is associated with the vp.
1358 object
->ref_count
--;
1361 if (object
->flags
& OBJ_DEAD
) {
1363 vm_object_dead_sleep(object
, "vodead");
1364 vn_lock(vp
, LK_EXCLUSIVE
| LK_RETRY
);
1368 KASSERT(vp
->v_object
!= NULL
, ("vinitvmio: NULL object"));
1369 vp
->v_flag
|= VOBJBUF
;
1375 * Print out a description of a vnode.
1377 static char *typename
[] =
1378 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1381 vprint(char *label
, struct vnode
*vp
)
1386 kprintf("%s: %p: ", label
, (void *)vp
);
1388 kprintf("%p: ", (void *)vp
);
1389 kprintf("type %s, sysrefs %d, writecount %d, holdcnt %d,",
1390 typename
[vp
->v_type
],
1391 vp
->v_sysref
.refcnt
, vp
->v_writecount
, vp
->v_auxrefs
);
1393 if (vp
->v_flag
& VROOT
)
1394 strcat(buf
, "|VROOT");
1395 if (vp
->v_flag
& VTEXT
)
1396 strcat(buf
, "|VTEXT");
1397 if (vp
->v_flag
& VSYSTEM
)
1398 strcat(buf
, "|VSYSTEM");
1399 if (vp
->v_flag
& VFREE
)
1400 strcat(buf
, "|VFREE");
1401 if (vp
->v_flag
& VOBJBUF
)
1402 strcat(buf
, "|VOBJBUF");
1404 kprintf(" flags (%s)", &buf
[1]);
1405 if (vp
->v_data
== NULL
) {
1414 #include <ddb/ddb.h>
1416 static int db_show_locked_vnodes(struct mount
*mp
, void *data
);
1419 * List all of the locked vnodes in the system.
1420 * Called when debugging the kernel.
1422 DB_SHOW_COMMAND(lockedvnodes
, lockedvnodes
)
1424 kprintf("Locked vnodes\n");
1425 mountlist_scan(db_show_locked_vnodes
, NULL
,
1426 MNTSCAN_FORWARD
|MNTSCAN_NOBUSY
);
1430 db_show_locked_vnodes(struct mount
*mp
, void *data __unused
)
1434 TAILQ_FOREACH(vp
, &mp
->mnt_nvnodelist
, v_nmntvnodes
) {
1435 if (vn_islocked(vp
))
1436 vprint((char *)0, vp
);
1443 * Top level filesystem related information gathering.
1445 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS
);
1448 vfs_sysctl(SYSCTL_HANDLER_ARGS
)
1450 int *name
= (int *)arg1
- 1; /* XXX */
1451 u_int namelen
= arg2
+ 1; /* XXX */
1452 struct vfsconf
*vfsp
;
1454 #if 1 || defined(COMPAT_PRELITE2)
1455 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1457 return (sysctl_ovfs_conf(oidp
, arg1
, arg2
, req
));
1461 /* all sysctl names at this level are at least name and field */
1463 return (ENOTDIR
); /* overloaded */
1464 if (name
[0] != VFS_GENERIC
) {
1465 for (vfsp
= vfsconf
; vfsp
; vfsp
= vfsp
->vfc_next
)
1466 if (vfsp
->vfc_typenum
== name
[0])
1469 return (EOPNOTSUPP
);
1470 return ((*vfsp
->vfc_vfsops
->vfs_sysctl
)(&name
[1], namelen
- 1,
1471 oldp
, oldlenp
, newp
, newlen
, p
));
1475 case VFS_MAXTYPENUM
:
1478 return (SYSCTL_OUT(req
, &maxvfsconf
, sizeof(int)));
1481 return (ENOTDIR
); /* overloaded */
1482 for (vfsp
= vfsconf
; vfsp
; vfsp
= vfsp
->vfc_next
)
1483 if (vfsp
->vfc_typenum
== name
[2])
1486 return (EOPNOTSUPP
);
1487 return (SYSCTL_OUT(req
, vfsp
, sizeof *vfsp
));
1489 return (EOPNOTSUPP
);
1492 SYSCTL_NODE(_vfs
, VFS_GENERIC
, generic
, CTLFLAG_RD
, vfs_sysctl
,
1493 "Generic filesystem");
1495 #if 1 || defined(COMPAT_PRELITE2)
1498 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS
)
1501 struct vfsconf
*vfsp
;
1502 struct ovfsconf ovfs
;
1504 for (vfsp
= vfsconf
; vfsp
; vfsp
= vfsp
->vfc_next
) {
1505 bzero(&ovfs
, sizeof(ovfs
));
1506 ovfs
.vfc_vfsops
= vfsp
->vfc_vfsops
; /* XXX used as flag */
1507 strcpy(ovfs
.vfc_name
, vfsp
->vfc_name
);
1508 ovfs
.vfc_index
= vfsp
->vfc_typenum
;
1509 ovfs
.vfc_refcount
= vfsp
->vfc_refcount
;
1510 ovfs
.vfc_flags
= vfsp
->vfc_flags
;
1511 error
= SYSCTL_OUT(req
, &ovfs
, sizeof ovfs
);
1518 #endif /* 1 || COMPAT_PRELITE2 */
1521 * Check to see if a filesystem is mounted on a block device.
1524 vfs_mountedon(struct vnode
*vp
)
1528 if ((dev
= vp
->v_rdev
) == NULL
) {
1529 if (vp
->v_type
!= VBLK
)
1530 dev
= get_dev(vp
->v_uminor
, vp
->v_umajor
);
1532 if (dev
!= NULL
&& dev
->si_mountpoint
)
1538 * Unmount all filesystems. The list is traversed in reverse order
1539 * of mounting to avoid dependencies.
1542 static int vfs_umountall_callback(struct mount
*mp
, void *data
);
1545 vfs_unmountall(void)
1550 count
= mountlist_scan(vfs_umountall_callback
,
1551 NULL
, MNTSCAN_REVERSE
|MNTSCAN_NOBUSY
);
1557 vfs_umountall_callback(struct mount
*mp
, void *data
)
1561 error
= dounmount(mp
, MNT_FORCE
);
1563 mountlist_remove(mp
);
1564 kprintf("unmount of filesystem mounted from %s failed (",
1565 mp
->mnt_stat
.f_mntfromname
);
1569 kprintf("%d)\n", error
);
1575 * Build hash lists of net addresses and hang them off the mount point.
1576 * Called by ufs_mount() to set up the lists of export addresses.
1579 vfs_hang_addrlist(struct mount
*mp
, struct netexport
*nep
,
1580 struct export_args
*argp
)
1583 struct radix_node_head
*rnh
;
1585 struct radix_node
*rn
;
1586 struct sockaddr
*saddr
, *smask
= 0;
1590 if (argp
->ex_addrlen
== 0) {
1591 if (mp
->mnt_flag
& MNT_DEFEXPORTED
)
1593 np
= &nep
->ne_defexported
;
1594 np
->netc_exflags
= argp
->ex_flags
;
1595 np
->netc_anon
= argp
->ex_anon
;
1596 np
->netc_anon
.cr_ref
= 1;
1597 mp
->mnt_flag
|= MNT_DEFEXPORTED
;
1601 if (argp
->ex_addrlen
< 0 || argp
->ex_addrlen
> MLEN
)
1603 if (argp
->ex_masklen
< 0 || argp
->ex_masklen
> MLEN
)
1606 i
= sizeof(struct netcred
) + argp
->ex_addrlen
+ argp
->ex_masklen
;
1607 np
= (struct netcred
*) kmalloc(i
, M_NETADDR
, M_WAITOK
);
1608 bzero((caddr_t
) np
, i
);
1609 saddr
= (struct sockaddr
*) (np
+ 1);
1610 if ((error
= copyin(argp
->ex_addr
, (caddr_t
) saddr
, argp
->ex_addrlen
)))
1612 if (saddr
->sa_len
> argp
->ex_addrlen
)
1613 saddr
->sa_len
= argp
->ex_addrlen
;
1614 if (argp
->ex_masklen
) {
1615 smask
= (struct sockaddr
*)((caddr_t
)saddr
+ argp
->ex_addrlen
);
1616 error
= copyin(argp
->ex_mask
, (caddr_t
)smask
, argp
->ex_masklen
);
1619 if (smask
->sa_len
> argp
->ex_masklen
)
1620 smask
->sa_len
= argp
->ex_masklen
;
1622 i
= saddr
->sa_family
;
1623 if ((rnh
= nep
->ne_rtable
[i
]) == 0) {
1625 * Seems silly to initialize every AF when most are not used,
1626 * do so on demand here
1628 SLIST_FOREACH(dom
, &domains
, dom_next
)
1629 if (dom
->dom_family
== i
&& dom
->dom_rtattach
) {
1630 dom
->dom_rtattach((void **) &nep
->ne_rtable
[i
],
1634 if ((rnh
= nep
->ne_rtable
[i
]) == 0) {
1639 rn
= (*rnh
->rnh_addaddr
) ((char *) saddr
, (char *) smask
, rnh
,
1641 if (rn
== 0 || np
!= (struct netcred
*) rn
) { /* already exists */
1645 np
->netc_exflags
= argp
->ex_flags
;
1646 np
->netc_anon
= argp
->ex_anon
;
1647 np
->netc_anon
.cr_ref
= 1;
1650 kfree(np
, M_NETADDR
);
1656 vfs_free_netcred(struct radix_node
*rn
, void *w
)
1658 struct radix_node_head
*rnh
= (struct radix_node_head
*) w
;
1660 (*rnh
->rnh_deladdr
) (rn
->rn_key
, rn
->rn_mask
, rnh
);
1661 kfree((caddr_t
) rn
, M_NETADDR
);
1666 * Free the net address hash lists that are hanging off the mount points.
1669 vfs_free_addrlist(struct netexport
*nep
)
1672 struct radix_node_head
*rnh
;
1674 for (i
= 0; i
<= AF_MAX
; i
++)
1675 if ((rnh
= nep
->ne_rtable
[i
])) {
1676 (*rnh
->rnh_walktree
) (rnh
, vfs_free_netcred
,
1678 kfree((caddr_t
) rnh
, M_RTABLE
);
1679 nep
->ne_rtable
[i
] = 0;
1684 vfs_export(struct mount
*mp
, struct netexport
*nep
, struct export_args
*argp
)
1688 if (argp
->ex_flags
& MNT_DELEXPORT
) {
1689 if (mp
->mnt_flag
& MNT_EXPUBLIC
) {
1690 vfs_setpublicfs(NULL
, NULL
, NULL
);
1691 mp
->mnt_flag
&= ~MNT_EXPUBLIC
;
1693 vfs_free_addrlist(nep
);
1694 mp
->mnt_flag
&= ~(MNT_EXPORTED
| MNT_DEFEXPORTED
);
1696 if (argp
->ex_flags
& MNT_EXPORTED
) {
1697 if (argp
->ex_flags
& MNT_EXPUBLIC
) {
1698 if ((error
= vfs_setpublicfs(mp
, nep
, argp
)) != 0)
1700 mp
->mnt_flag
|= MNT_EXPUBLIC
;
1702 if ((error
= vfs_hang_addrlist(mp
, nep
, argp
)))
1704 mp
->mnt_flag
|= MNT_EXPORTED
;
1711 * Set the publicly exported filesystem (WebNFS). Currently, only
1712 * one public filesystem is possible in the spec (RFC 2054 and 2055)
1715 vfs_setpublicfs(struct mount
*mp
, struct netexport
*nep
,
1716 struct export_args
*argp
)
1723 * mp == NULL -> invalidate the current info, the FS is
1724 * no longer exported. May be called from either vfs_export
1725 * or unmount, so check if it hasn't already been done.
1728 if (nfs_pub
.np_valid
) {
1729 nfs_pub
.np_valid
= 0;
1730 if (nfs_pub
.np_index
!= NULL
) {
1731 FREE(nfs_pub
.np_index
, M_TEMP
);
1732 nfs_pub
.np_index
= NULL
;
1739 * Only one allowed at a time.
1741 if (nfs_pub
.np_valid
!= 0 && mp
!= nfs_pub
.np_mount
)
1745 * Get real filehandle for root of exported FS.
1747 bzero((caddr_t
)&nfs_pub
.np_handle
, sizeof(nfs_pub
.np_handle
));
1748 nfs_pub
.np_handle
.fh_fsid
= mp
->mnt_stat
.f_fsid
;
1750 if ((error
= VFS_ROOT(mp
, &rvp
)))
1753 if ((error
= VFS_VPTOFH(rvp
, &nfs_pub
.np_handle
.fh_fid
)))
1759 * If an indexfile was specified, pull it in.
1761 if (argp
->ex_indexfile
!= NULL
) {
1764 error
= vn_get_namelen(rvp
, &namelen
);
1767 MALLOC(nfs_pub
.np_index
, char *, namelen
, M_TEMP
,
1769 error
= copyinstr(argp
->ex_indexfile
, nfs_pub
.np_index
,
1770 namelen
, (size_t *)0);
1773 * Check for illegal filenames.
1775 for (cp
= nfs_pub
.np_index
; *cp
; cp
++) {
1783 FREE(nfs_pub
.np_index
, M_TEMP
);
1788 nfs_pub
.np_mount
= mp
;
1789 nfs_pub
.np_valid
= 1;
1794 vfs_export_lookup(struct mount
*mp
, struct netexport
*nep
,
1795 struct sockaddr
*nam
)
1798 struct radix_node_head
*rnh
;
1799 struct sockaddr
*saddr
;
1802 if (mp
->mnt_flag
& MNT_EXPORTED
) {
1804 * Lookup in the export list first.
1808 rnh
= nep
->ne_rtable
[saddr
->sa_family
];
1810 np
= (struct netcred
*)
1811 (*rnh
->rnh_matchaddr
)((char *)saddr
,
1813 if (np
&& np
->netc_rnodes
->rn_flags
& RNF_ROOT
)
1818 * If no address match, use the default if it exists.
1820 if (np
== NULL
&& mp
->mnt_flag
& MNT_DEFEXPORTED
)
1821 np
= &nep
->ne_defexported
;
1827 * perform msync on all vnodes under a mount point. The mount point must
1828 * be locked. This code is also responsible for lazy-freeing unreferenced
1829 * vnodes whos VM objects no longer contain pages.
1831 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
1833 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked,
1834 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it
1835 * way up in this high level function.
1837 static int vfs_msync_scan1(struct mount
*mp
, struct vnode
*vp
, void *data
);
1838 static int vfs_msync_scan2(struct mount
*mp
, struct vnode
*vp
, void *data
);
1841 vfs_msync(struct mount
*mp
, int flags
)
1845 vmsc_flags
= VMSC_GETVP
;
1846 if (flags
!= MNT_WAIT
)
1847 vmsc_flags
|= VMSC_NOWAIT
;
1848 vmntvnodescan(mp
, vmsc_flags
, vfs_msync_scan1
, vfs_msync_scan2
,
1853 * scan1 is a fast pre-check. There could be hundreds of thousands of
1854 * vnodes, we cannot afford to do anything heavy weight until we have a
1855 * fairly good indication that there is work to do.
1859 vfs_msync_scan1(struct mount
*mp
, struct vnode
*vp
, void *data
)
1861 int flags
= (int)data
;
1863 if ((vp
->v_flag
& VRECLAIMED
) == 0) {
1864 if (vshouldmsync(vp
))
1865 return(0); /* call scan2 */
1866 if ((mp
->mnt_flag
& MNT_RDONLY
) == 0 &&
1867 (vp
->v_flag
& VOBJDIRTY
) &&
1868 (flags
== MNT_WAIT
|| vn_islocked(vp
) == 0)) {
1869 return(0); /* call scan2 */
1874 * do not call scan2, continue the loop
1880 * This callback is handed a locked vnode.
1884 vfs_msync_scan2(struct mount
*mp
, struct vnode
*vp
, void *data
)
1887 int flags
= (int)data
;
1889 if (vp
->v_flag
& VRECLAIMED
)
1892 if ((mp
->mnt_flag
& MNT_RDONLY
) == 0 && (vp
->v_flag
& VOBJDIRTY
)) {
1893 if ((obj
= vp
->v_object
) != NULL
) {
1894 vm_object_page_clean(obj
, 0, 0,
1895 flags
== MNT_WAIT
? OBJPC_SYNC
: OBJPC_NOSYNC
);
1902 * Record a process's interest in events which might happen to
1903 * a vnode. Because poll uses the historic select-style interface
1904 * internally, this routine serves as both the ``check for any
1905 * pending events'' and the ``record my interest in future events''
1906 * functions. (These are done together, while the lock is held,
1907 * to avoid race conditions.)
1910 vn_pollrecord(struct vnode
*vp
, int events
)
1914 KKASSERT(curthread
->td_proc
!= NULL
);
1916 lwkt_gettoken(&ilock
, &vp
->v_pollinfo
.vpi_token
);
1917 if (vp
->v_pollinfo
.vpi_revents
& events
) {
1919 * This leaves events we are not interested
1920 * in available for the other process which
1921 * which presumably had requested them
1922 * (otherwise they would never have been
1925 events
&= vp
->v_pollinfo
.vpi_revents
;
1926 vp
->v_pollinfo
.vpi_revents
&= ~events
;
1928 lwkt_reltoken(&ilock
);
1931 vp
->v_pollinfo
.vpi_events
|= events
;
1932 selrecord(curthread
, &vp
->v_pollinfo
.vpi_selinfo
);
1933 lwkt_reltoken(&ilock
);
1938 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
1939 * it is possible for us to miss an event due to race conditions, but
1940 * that condition is expected to be rare, so for the moment it is the
1941 * preferred interface.
1944 vn_pollevent(struct vnode
*vp
, int events
)
1948 lwkt_gettoken(&ilock
, &vp
->v_pollinfo
.vpi_token
);
1949 if (vp
->v_pollinfo
.vpi_events
& events
) {
1951 * We clear vpi_events so that we don't
1952 * call selwakeup() twice if two events are
1953 * posted before the polling process(es) is
1954 * awakened. This also ensures that we take at
1955 * most one selwakeup() if the polling process
1956 * is no longer interested. However, it does
1957 * mean that only one event can be noticed at
1958 * a time. (Perhaps we should only clear those
1959 * event bits which we note?) XXX
1961 vp
->v_pollinfo
.vpi_events
= 0; /* &= ~events ??? */
1962 vp
->v_pollinfo
.vpi_revents
|= events
;
1963 selwakeup(&vp
->v_pollinfo
.vpi_selinfo
);
1965 lwkt_reltoken(&ilock
);
1969 * Wake up anyone polling on vp because it is being revoked.
1970 * This depends on dead_poll() returning POLLHUP for correct
1974 vn_pollgone(struct vnode
*vp
)
1978 lwkt_gettoken(&ilock
, &vp
->v_pollinfo
.vpi_token
);
1979 if (vp
->v_pollinfo
.vpi_events
) {
1980 vp
->v_pollinfo
.vpi_events
= 0;
1981 selwakeup(&vp
->v_pollinfo
.vpi_selinfo
);
1983 lwkt_reltoken(&ilock
);
1987 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened
1988 * (or v_rdev might be NULL).
1991 vn_todev(struct vnode
*vp
)
1993 if (vp
->v_type
!= VBLK
&& vp
->v_type
!= VCHR
)
1995 KKASSERT(vp
->v_rdev
!= NULL
);
1996 return (vp
->v_rdev
);
2000 * Check if vnode represents a disk device. The vnode does not need to be
2004 vn_isdisk(struct vnode
*vp
, int *errp
)
2008 if (vp
->v_type
!= VCHR
) {
2014 if ((dev
= vp
->v_rdev
) == NULL
)
2015 dev
= get_dev(vp
->v_umajor
, vp
->v_uminor
);
2022 if (dev_is_good(dev
) == 0) {
2027 if ((dev_dflags(dev
) & D_DISK
) == 0) {
2038 vn_get_namelen(struct vnode
*vp
, int *namelen
)
2040 int error
, retval
[2];
2042 error
= VOP_PATHCONF(vp
, _PC_NAME_MAX
, retval
);
2050 vop_write_dirent(int *error
, struct uio
*uio
, ino_t d_ino
, uint8_t d_type
,
2051 uint16_t d_namlen
, const char *d_name
)
2056 len
= _DIRENT_RECLEN(d_namlen
);
2057 if (len
> uio
->uio_resid
)
2060 dp
= kmalloc(len
, M_TEMP
, M_WAITOK
| M_ZERO
);
2063 dp
->d_namlen
= d_namlen
;
2064 dp
->d_type
= d_type
;
2065 bcopy(d_name
, dp
->d_name
, d_namlen
);
2067 *error
= uiomove((caddr_t
)dp
, len
, uio
);