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 1.118 2008/09/17 21:44:18 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>
57 #include <sys/kernel.h>
58 #include <sys/kthread.h>
59 #include <sys/malloc.h>
61 #include <sys/mount.h>
64 #include <sys/reboot.h>
65 #include <sys/socket.h>
67 #include <sys/sysctl.h>
68 #include <sys/syslog.h>
69 #include <sys/unistd.h>
70 #include <sys/vmmeter.h>
71 #include <sys/vnode.h>
73 #include <machine/limits.h>
76 #include <vm/vm_object.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_kern.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_page.h>
82 #include <vm/vm_pager.h>
83 #include <vm/vnode_pager.h>
84 #include <vm/vm_zone.h>
87 #include <sys/thread2.h>
88 #include <sys/sysref2.h>
90 static MALLOC_DEFINE(M_NETADDR
, "Export Host", "Export host address structure");
93 SYSCTL_INT(_debug
, OID_AUTO
, numvnodes
, CTLFLAG_RD
, &numvnodes
, 0, "");
95 SYSCTL_INT(_vfs
, OID_AUTO
, fastdev
, CTLFLAG_RW
, &vfs_fastdev
, 0, "");
97 enum vtype iftovt_tab
[16] = {
98 VNON
, VFIFO
, VCHR
, VNON
, VDIR
, VNON
, VBLK
, VNON
,
99 VREG
, VNON
, VLNK
, VNON
, VSOCK
, VNON
, VNON
, VBAD
,
101 int vttoif_tab
[9] = {
102 0, S_IFREG
, S_IFDIR
, S_IFBLK
, S_IFCHR
, S_IFLNK
,
103 S_IFSOCK
, S_IFIFO
, S_IFMT
,
106 static int reassignbufcalls
;
107 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufcalls
, CTLFLAG_RW
,
108 &reassignbufcalls
, 0, "");
109 static int reassignbufloops
;
110 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufloops
, CTLFLAG_RW
,
111 &reassignbufloops
, 0, "");
112 static int reassignbufsortgood
;
113 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufsortgood
, CTLFLAG_RW
,
114 &reassignbufsortgood
, 0, "");
115 static int reassignbufsortbad
;
116 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufsortbad
, CTLFLAG_RW
,
117 &reassignbufsortbad
, 0, "");
118 static int reassignbufmethod
= 1;
119 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufmethod
, CTLFLAG_RW
,
120 &reassignbufmethod
, 0, "");
122 int nfs_mount_type
= -1;
123 static struct lwkt_token spechash_token
;
124 struct nfs_public nfs_pub
; /* publicly exported FS */
127 SYSCTL_INT(_kern
, KERN_MAXVNODES
, maxvnodes
, CTLFLAG_RW
,
128 &desiredvnodes
, 0, "Maximum number of vnodes");
130 static void vfs_free_addrlist (struct netexport
*nep
);
131 static int vfs_free_netcred (struct radix_node
*rn
, void *w
);
132 static int vfs_hang_addrlist (struct mount
*mp
, struct netexport
*nep
,
133 const struct export_args
*argp
);
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 * Desiredvnodes is kern.maxvnodes. We want to scale it
177 * according to available system memory but we may also have
178 * to limit it based on available KVM, which is capped on 32 bit
181 desiredvnodes
= min(maxproc
+ vmstats
.v_page_count
/ 4,
183 (sizeof(struct vm_object
) + sizeof(struct vnode
))));
185 lwkt_token_init(&spechash_token
);
189 * Knob to control the precision of file timestamps:
191 * 0 = seconds only; nanoseconds zeroed.
192 * 1 = seconds and nanoseconds, accurate within 1/HZ.
193 * 2 = seconds and nanoseconds, truncated to microseconds.
194 * >=3 = seconds and nanoseconds, maximum precision.
196 enum { TSP_SEC
, TSP_HZ
, TSP_USEC
, TSP_NSEC
};
198 static int timestamp_precision
= TSP_SEC
;
199 SYSCTL_INT(_vfs
, OID_AUTO
, timestamp_precision
, CTLFLAG_RW
,
200 ×tamp_precision
, 0, "");
203 * 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 /* va_*_uuid fields are only valid if related flags are set */
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 vinvalbuf(struct vnode
*vp
, int flags
, int slpflag
, int slptimeo
)
278 struct vinvalbuf_bp_info info
;
283 lwkt_gettoken(&vlock
, &vp
->v_token
);
286 * If we are being asked to save, call fsync to ensure that the inode
289 if (flags
& V_SAVE
) {
290 error
= bio_track_wait(&vp
->v_track_write
, slpflag
, slptimeo
);
293 if (!RB_EMPTY(&vp
->v_rbdirty_tree
)) {
294 if ((error
= VOP_FSYNC(vp
, MNT_WAIT
, 0)) != 0)
298 * Dirty bufs may be left or generated via races
299 * in circumstances where vinvalbuf() is called on
300 * a vnode not undergoing reclamation. Only
301 * panic if we are trying to reclaim the vnode.
303 if ((vp
->v_flag
& VRECLAIMED
) &&
304 (bio_track_active(&vp
->v_track_write
) ||
305 !RB_EMPTY(&vp
->v_rbdirty_tree
))) {
306 panic("vinvalbuf: dirty bufs");
310 info
.slptimeo
= slptimeo
;
311 info
.lkflags
= LK_EXCLUSIVE
| LK_SLEEPFAIL
;
312 if (slpflag
& PCATCH
)
313 info
.lkflags
|= LK_PCATCH
;
318 * Flush the buffer cache until nothing is left.
320 while (!RB_EMPTY(&vp
->v_rbclean_tree
) ||
321 !RB_EMPTY(&vp
->v_rbdirty_tree
)) {
322 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbclean_tree
, NULL
,
323 vinvalbuf_bp
, &info
);
325 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, NULL
,
326 vinvalbuf_bp
, &info
);
331 * Wait for I/O completion. We may block in the pip code so we have
335 bio_track_wait(&vp
->v_track_write
, 0, 0);
336 if ((object
= vp
->v_object
) != NULL
) {
337 while (object
->paging_in_progress
)
338 vm_object_pip_sleep(object
, "vnvlbx");
340 } while (bio_track_active(&vp
->v_track_write
));
343 * Destroy the copy in the VM cache, too.
345 if ((object
= vp
->v_object
) != NULL
) {
346 vm_object_page_remove(object
, 0, 0,
347 (flags
& V_SAVE
) ? TRUE
: FALSE
);
350 if (!RB_EMPTY(&vp
->v_rbdirty_tree
) || !RB_EMPTY(&vp
->v_rbclean_tree
))
351 panic("vinvalbuf: flush failed");
352 if (!RB_EMPTY(&vp
->v_rbhash_tree
))
353 panic("vinvalbuf: flush failed, buffers still present");
356 lwkt_reltoken(&vlock
);
361 vinvalbuf_bp(struct buf
*bp
, void *data
)
363 struct vinvalbuf_bp_info
*info
= data
;
366 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
)) {
367 error
= BUF_TIMELOCK(bp
, info
->lkflags
,
368 "vinvalbuf", info
->slptimeo
);
378 KKASSERT(bp
->b_vp
== info
->vp
);
381 * XXX Since there are no node locks for NFS, I
382 * believe there is a slight chance that a delayed
383 * write will occur while sleeping just above, so
384 * check for it. Note that vfs_bio_awrite expects
385 * buffers to reside on a queue, while bwrite() and
388 * NOTE: NO B_LOCKED CHECK. Also no buf_checkwrite()
389 * check. This code will write out the buffer, period.
391 if (((bp
->b_flags
& (B_DELWRI
| B_INVAL
)) == B_DELWRI
) &&
392 (info
->flags
& V_SAVE
)) {
393 if (bp
->b_vp
== info
->vp
) {
394 if (bp
->b_flags
& B_CLUSTEROK
) {
404 } else if (info
->flags
& V_SAVE
) {
406 * Cannot set B_NOCACHE on a clean buffer as this will
407 * destroy the VM backing store which might actually
408 * be dirty (and unsynchronized).
411 bp
->b_flags
|= (B_INVAL
| B_RELBUF
);
415 bp
->b_flags
|= (B_INVAL
| B_NOCACHE
| B_RELBUF
);
422 * Truncate a file's buffer and pages to a specified length. This
423 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
426 * The vnode must be locked.
428 static int vtruncbuf_bp_trunc_cmp(struct buf
*bp
, void *data
);
429 static int vtruncbuf_bp_trunc(struct buf
*bp
, void *data
);
430 static int vtruncbuf_bp_metasync_cmp(struct buf
*bp
, void *data
);
431 static int vtruncbuf_bp_metasync(struct buf
*bp
, void *data
);
434 vtruncbuf(struct vnode
*vp
, off_t length
, int blksize
)
437 const char *filename
;
442 * Round up to the *next* block, then destroy the buffers in question.
443 * Since we are only removing some of the buffers we must rely on the
444 * scan count to determine whether a loop is necessary.
446 if ((count
= (int)(length
% blksize
)) != 0)
447 truncloffset
= length
+ (blksize
- count
);
449 truncloffset
= length
;
451 lwkt_gettoken(&vlock
, &vp
->v_token
);
453 count
= RB_SCAN(buf_rb_tree
, &vp
->v_rbclean_tree
,
454 vtruncbuf_bp_trunc_cmp
,
455 vtruncbuf_bp_trunc
, &truncloffset
);
456 count
+= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
457 vtruncbuf_bp_trunc_cmp
,
458 vtruncbuf_bp_trunc
, &truncloffset
);
462 * For safety, fsync any remaining metadata if the file is not being
463 * truncated to 0. Since the metadata does not represent the entire
464 * dirty list we have to rely on the hit count to ensure that we get
469 count
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
470 vtruncbuf_bp_metasync_cmp
,
471 vtruncbuf_bp_metasync
, vp
);
476 * Clean out any left over VM backing store.
478 * It is possible to have in-progress I/O from buffers that were
479 * not part of the truncation. This should not happen if we
480 * are truncating to 0-length.
482 vnode_pager_setsize(vp
, length
);
483 bio_track_wait(&vp
->v_track_write
, 0, 0);
485 filename
= TAILQ_FIRST(&vp
->v_namecache
) ?
486 TAILQ_FIRST(&vp
->v_namecache
)->nc_name
: "?";
489 * Make sure no buffers were instantiated while we were trying
490 * to clean out the remaining VM pages. This could occur due
491 * to busy dirty VM pages being flushed out to disk.
494 count
= RB_SCAN(buf_rb_tree
, &vp
->v_rbclean_tree
,
495 vtruncbuf_bp_trunc_cmp
,
496 vtruncbuf_bp_trunc
, &truncloffset
);
497 count
+= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
498 vtruncbuf_bp_trunc_cmp
,
499 vtruncbuf_bp_trunc
, &truncloffset
);
501 kprintf("Warning: vtruncbuf(): Had to re-clean %d "
502 "left over buffers in %s\n", count
, filename
);
506 lwkt_reltoken(&vlock
);
512 * The callback buffer is beyond the new file EOF and must be destroyed.
513 * Note that the compare function must conform to the RB_SCAN's requirements.
517 vtruncbuf_bp_trunc_cmp(struct buf
*bp
, void *data
)
519 if (bp
->b_loffset
>= *(off_t
*)data
)
526 vtruncbuf_bp_trunc(struct buf
*bp
, void *data
)
529 * Do not try to use a buffer we cannot immediately lock, but sleep
530 * anyway to prevent a livelock. The code will loop until all buffers
533 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
)) {
534 if (BUF_LOCK(bp
, LK_EXCLUSIVE
|LK_SLEEPFAIL
) == 0)
538 bp
->b_flags
|= (B_INVAL
| B_RELBUF
| B_NOCACHE
);
545 * Fsync all meta-data after truncating a file to be non-zero. Only metadata
546 * blocks (with a negative loffset) are scanned.
547 * Note that the compare function must conform to the RB_SCAN's requirements.
550 vtruncbuf_bp_metasync_cmp(struct buf
*bp
, void *data
)
552 if (bp
->b_loffset
< 0)
558 vtruncbuf_bp_metasync(struct buf
*bp
, void *data
)
560 struct vnode
*vp
= data
;
562 if (bp
->b_flags
& B_DELWRI
) {
564 * Do not try to use a buffer we cannot immediately lock,
565 * but sleep anyway to prevent a livelock. The code will
566 * loop until all buffers can be acted upon.
568 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
)) {
569 if (BUF_LOCK(bp
, LK_EXCLUSIVE
|LK_SLEEPFAIL
) == 0)
585 * vfsync - implements a multipass fsync on a file which understands
586 * dependancies and meta-data. The passed vnode must be locked. The
587 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY.
589 * When fsyncing data asynchronously just do one consolidated pass starting
590 * with the most negative block number. This may not get all the data due
593 * When fsyncing data synchronously do a data pass, then a metadata pass,
594 * then do additional data+metadata passes to try to get all the data out.
596 static int vfsync_wait_output(struct vnode
*vp
,
597 int (*waitoutput
)(struct vnode
*, struct thread
*));
598 static int vfsync_data_only_cmp(struct buf
*bp
, void *data
);
599 static int vfsync_meta_only_cmp(struct buf
*bp
, void *data
);
600 static int vfsync_lazy_range_cmp(struct buf
*bp
, void *data
);
601 static int vfsync_bp(struct buf
*bp
, void *data
);
610 int (*checkdef
)(struct buf
*);
614 vfsync(struct vnode
*vp
, int waitfor
, int passes
,
615 int (*checkdef
)(struct buf
*),
616 int (*waitoutput
)(struct vnode
*, struct thread
*))
618 struct vfsync_info info
;
622 bzero(&info
, sizeof(info
));
624 if ((info
.checkdef
= checkdef
) == NULL
)
627 lwkt_gettoken(&vlock
, &vp
->v_token
);
632 * Lazy (filesystem syncer typ) Asynchronous plus limit the
633 * number of data (not meta) pages we try to flush to 1MB.
634 * A non-zero return means that lazy limit was reached.
636 info
.lazylimit
= 1024 * 1024;
638 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
639 vfsync_lazy_range_cmp
, vfsync_bp
, &info
);
640 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
641 vfsync_meta_only_cmp
, vfsync_bp
, &info
);
644 else if (!RB_EMPTY(&vp
->v_rbdirty_tree
))
645 vn_syncer_add_to_worklist(vp
, 1);
650 * Asynchronous. Do a data-only pass and a meta-only pass.
653 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, vfsync_data_only_cmp
,
655 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, vfsync_meta_only_cmp
,
661 * Synchronous. Do a data-only pass, then a meta-data+data
662 * pass, then additional integrated passes to try to get
663 * all the dependancies flushed.
665 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, vfsync_data_only_cmp
,
667 error
= vfsync_wait_output(vp
, waitoutput
);
669 info
.skippedbufs
= 0;
670 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, NULL
,
672 error
= vfsync_wait_output(vp
, waitoutput
);
673 if (info
.skippedbufs
)
674 kprintf("Warning: vfsync skipped %d dirty bufs in pass2!\n", info
.skippedbufs
);
676 while (error
== 0 && passes
> 0 &&
677 !RB_EMPTY(&vp
->v_rbdirty_tree
)
680 info
.synchronous
= 1;
683 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, NULL
,
689 error
= vfsync_wait_output(vp
, waitoutput
);
693 lwkt_reltoken(&vlock
);
698 vfsync_wait_output(struct vnode
*vp
,
699 int (*waitoutput
)(struct vnode
*, struct thread
*))
703 error
= bio_track_wait(&vp
->v_track_write
, 0, 0);
705 error
= waitoutput(vp
, curthread
);
710 vfsync_data_only_cmp(struct buf
*bp
, void *data
)
712 if (bp
->b_loffset
< 0)
718 vfsync_meta_only_cmp(struct buf
*bp
, void *data
)
720 if (bp
->b_loffset
< 0)
726 vfsync_lazy_range_cmp(struct buf
*bp
, void *data
)
728 struct vfsync_info
*info
= data
;
729 if (bp
->b_loffset
< info
->vp
->v_lazyw
)
735 vfsync_bp(struct buf
*bp
, void *data
)
737 struct vfsync_info
*info
= data
;
738 struct vnode
*vp
= info
->vp
;
742 * if syncdeps is not set we do not try to write buffers which have
745 if (!info
->synchronous
&& info
->syncdeps
== 0 && info
->checkdef(bp
))
749 * Ignore buffers that we cannot immediately lock. XXX
751 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
)) {
752 kprintf("Warning: vfsync_bp skipping dirty buffer %p\n", bp
);
756 if ((bp
->b_flags
& B_DELWRI
) == 0)
757 panic("vfsync_bp: buffer not dirty");
759 panic("vfsync_bp: buffer vp mismatch");
762 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
763 * has been written but an additional handshake with the device
764 * is required before we can dispose of the buffer. We have no idea
765 * how to do this so we have to skip these buffers.
767 if (bp
->b_flags
& B_NEEDCOMMIT
) {
773 * Ask bioops if it is ok to sync
775 if (LIST_FIRST(&bp
->b_dep
) != NULL
&& buf_checkwrite(bp
)) {
781 if (info
->synchronous
) {
783 * Synchronous flushing. An error may be returned.
789 * Asynchronous flushing. A negative return value simply
790 * stops the scan and is not considered an error. We use
791 * this to support limited MNT_LAZY flushes.
793 vp
->v_lazyw
= bp
->b_loffset
;
794 if ((vp
->v_flag
& VOBJBUF
) && (bp
->b_flags
& B_CLUSTEROK
)) {
795 info
->lazycount
+= vfs_bio_awrite(bp
);
797 info
->lazycount
+= bp
->b_bufsize
;
801 if (info
->lazylimit
&& info
->lazycount
>= info
->lazylimit
)
810 * Associate a buffer with a vnode.
815 bgetvp(struct vnode
*vp
, struct buf
*bp
)
819 KASSERT(bp
->b_vp
== NULL
, ("bgetvp: not free"));
820 KKASSERT((bp
->b_flags
& (B_HASHED
|B_DELWRI
|B_VNCLEAN
|B_VNDIRTY
)) == 0);
823 * Insert onto list for new vnode.
825 lwkt_gettoken(&vlock
, &vp
->v_token
);
826 if (buf_rb_hash_RB_INSERT(&vp
->v_rbhash_tree
, bp
)) {
827 lwkt_reltoken(&vlock
);
831 bp
->b_flags
|= B_HASHED
;
832 bp
->b_flags
|= B_VNCLEAN
;
833 if (buf_rb_tree_RB_INSERT(&vp
->v_rbclean_tree
, bp
))
834 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp
, bp
);
836 lwkt_reltoken(&vlock
);
841 * Disassociate a buffer from a vnode.
844 brelvp(struct buf
*bp
)
849 KASSERT(bp
->b_vp
!= NULL
, ("brelvp: NULL"));
852 * Delete from old vnode list, if on one.
855 lwkt_gettoken(&vlock
, &vp
->v_token
);
856 if (bp
->b_flags
& (B_VNDIRTY
| B_VNCLEAN
)) {
857 if (bp
->b_flags
& B_VNDIRTY
)
858 buf_rb_tree_RB_REMOVE(&vp
->v_rbdirty_tree
, bp
);
860 buf_rb_tree_RB_REMOVE(&vp
->v_rbclean_tree
, bp
);
861 bp
->b_flags
&= ~(B_VNDIRTY
| B_VNCLEAN
);
863 if (bp
->b_flags
& B_HASHED
) {
864 buf_rb_hash_RB_REMOVE(&vp
->v_rbhash_tree
, bp
);
865 bp
->b_flags
&= ~B_HASHED
;
867 if ((vp
->v_flag
& VONWORKLST
) && RB_EMPTY(&vp
->v_rbdirty_tree
)) {
868 vp
->v_flag
&= ~VONWORKLST
;
869 LIST_REMOVE(vp
, v_synclist
);
872 lwkt_reltoken(&vlock
);
878 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI.
879 * This routine is called when the state of the B_DELWRI bit is changed.
884 reassignbuf(struct buf
*bp
)
886 struct vnode
*vp
= bp
->b_vp
;
890 KKASSERT(vp
!= NULL
);
894 * B_PAGING flagged buffers cannot be reassigned because their vp
895 * is not fully linked in.
897 if (bp
->b_flags
& B_PAGING
)
898 panic("cannot reassign paging buffer");
900 lwkt_gettoken(&vlock
, &vp
->v_token
);
901 if (bp
->b_flags
& B_DELWRI
) {
903 * Move to the dirty list, add the vnode to the worklist
905 if (bp
->b_flags
& B_VNCLEAN
) {
906 buf_rb_tree_RB_REMOVE(&vp
->v_rbclean_tree
, bp
);
907 bp
->b_flags
&= ~B_VNCLEAN
;
909 if ((bp
->b_flags
& B_VNDIRTY
) == 0) {
910 if (buf_rb_tree_RB_INSERT(&vp
->v_rbdirty_tree
, bp
)) {
911 panic("reassignbuf: dup lblk vp %p bp %p",
914 bp
->b_flags
|= B_VNDIRTY
;
916 if ((vp
->v_flag
& VONWORKLST
) == 0) {
917 switch (vp
->v_type
) {
924 vp
->v_rdev
->si_mountpoint
!= NULL
) {
932 vn_syncer_add_to_worklist(vp
, delay
);
936 * Move to the clean list, remove the vnode from the worklist
937 * if no dirty blocks remain.
939 if (bp
->b_flags
& B_VNDIRTY
) {
940 buf_rb_tree_RB_REMOVE(&vp
->v_rbdirty_tree
, bp
);
941 bp
->b_flags
&= ~B_VNDIRTY
;
943 if ((bp
->b_flags
& B_VNCLEAN
) == 0) {
944 if (buf_rb_tree_RB_INSERT(&vp
->v_rbclean_tree
, bp
)) {
945 panic("reassignbuf: dup lblk vp %p bp %p",
948 bp
->b_flags
|= B_VNCLEAN
;
950 if ((vp
->v_flag
& VONWORKLST
) &&
951 RB_EMPTY(&vp
->v_rbdirty_tree
)) {
952 vp
->v_flag
&= ~VONWORKLST
;
953 LIST_REMOVE(vp
, v_synclist
);
956 lwkt_reltoken(&vlock
);
960 * Create a vnode for a block device.
961 * Used for mounting the root file system.
963 extern struct vop_ops
*devfs_vnode_dev_vops_p
;
965 bdevvp(cdev_t dev
, struct vnode
**vpp
)
975 error
= getspecialvnode(VT_NON
, NULL
, &devfs_vnode_dev_vops_p
,
986 v_associate_rdev(vp
, dev
);
987 vp
->v_umajor
= dev
->si_umajor
;
988 vp
->v_uminor
= dev
->si_uminor
;
995 v_associate_rdev(struct vnode
*vp
, cdev_t dev
)
1001 if (dev_is_good(dev
) == 0)
1003 KKASSERT(vp
->v_rdev
== NULL
);
1004 vp
->v_rdev
= reference_dev(dev
);
1005 lwkt_gettoken(&ilock
, &spechash_token
);
1006 SLIST_INSERT_HEAD(&dev
->si_hlist
, vp
, v_cdevnext
);
1007 lwkt_reltoken(&ilock
);
1012 v_release_rdev(struct vnode
*vp
)
1017 if ((dev
= vp
->v_rdev
) != NULL
) {
1018 lwkt_gettoken(&ilock
, &spechash_token
);
1019 SLIST_REMOVE(&dev
->si_hlist
, vp
, vnode
, v_cdevnext
);
1022 lwkt_reltoken(&ilock
);
1027 * Add a vnode to the alias list hung off the cdev_t. We only associate
1028 * the device number with the vnode. The actual device is not associated
1029 * until the vnode is opened (usually in spec_open()), and will be
1030 * disassociated on last close.
1033 addaliasu(struct vnode
*nvp
, int x
, int y
)
1035 if (nvp
->v_type
!= VBLK
&& nvp
->v_type
!= VCHR
)
1036 panic("addaliasu on non-special vnode");
1042 * Simple call that a filesystem can make to try to get rid of a
1043 * vnode. It will fail if anyone is referencing the vnode (including
1046 * The filesystem can check whether its in-memory inode structure still
1047 * references the vp on return.
1050 vclean_unlocked(struct vnode
*vp
)
1053 if (sysref_isactive(&vp
->v_sysref
) == 0)
1059 * Disassociate a vnode from its underlying filesystem.
1061 * The vnode must be VX locked and referenced. In all normal situations
1062 * there are no active references. If vclean_vxlocked() is called while
1063 * there are active references, the vnode is being ripped out and we have
1064 * to call VOP_CLOSE() as appropriate before we can reclaim it.
1067 vclean_vxlocked(struct vnode
*vp
, int flags
)
1074 * If the vnode has already been reclaimed we have nothing to do.
1076 if (vp
->v_flag
& VRECLAIMED
)
1078 vp
->v_flag
|= VRECLAIMED
;
1081 * Scrap the vfs cache
1083 while (cache_inval_vp(vp
, 0) != 0) {
1084 kprintf("Warning: vnode %p clean/cache_resolution race detected\n", vp
);
1085 tsleep(vp
, 0, "vclninv", 2);
1089 * Check to see if the vnode is in use. If so we have to reference it
1090 * before we clean it out so that its count cannot fall to zero and
1091 * generate a race against ourselves to recycle it.
1093 active
= sysref_isactive(&vp
->v_sysref
);
1096 * Clean out any buffers associated with the vnode and destroy its
1097 * object, if it has one.
1099 vinvalbuf(vp
, V_SAVE
, 0, 0);
1102 * If purging an active vnode (typically during a forced unmount
1103 * or reboot), it must be closed and deactivated before being
1104 * reclaimed. This isn't really all that safe, but what can
1107 * Note that neither of these routines unlocks the vnode.
1109 if (active
&& (flags
& DOCLOSE
)) {
1110 while ((n
= vp
->v_opencount
) != 0) {
1111 if (vp
->v_writecount
)
1112 VOP_CLOSE(vp
, FWRITE
|FNONBLOCK
);
1114 VOP_CLOSE(vp
, FNONBLOCK
);
1115 if (vp
->v_opencount
== n
) {
1116 kprintf("Warning: unable to force-close"
1124 * If the vnode has not been deactivated, deactivated it. Deactivation
1125 * can create new buffers and VM pages so we have to call vinvalbuf()
1126 * again to make sure they all get flushed.
1128 * This can occur if a file with a link count of 0 needs to be
1131 if ((vp
->v_flag
& VINACTIVE
) == 0) {
1132 vp
->v_flag
|= VINACTIVE
;
1134 vinvalbuf(vp
, V_SAVE
, 0, 0);
1138 * If the vnode has an object, destroy it.
1140 if ((object
= vp
->v_object
) != NULL
) {
1141 if (object
->ref_count
== 0) {
1142 if ((object
->flags
& OBJ_DEAD
) == 0)
1143 vm_object_terminate(object
);
1145 vm_pager_deallocate(object
);
1147 vp
->v_flag
&= ~VOBJBUF
;
1149 KKASSERT((vp
->v_flag
& VOBJBUF
) == 0);
1152 * Reclaim the vnode.
1154 if (VOP_RECLAIM(vp
))
1155 panic("vclean: cannot reclaim");
1158 * Done with purge, notify sleepers of the grim news.
1160 vp
->v_ops
= &dead_vnode_vops_p
;
1165 * If we are destroying an active vnode, reactivate it now that
1166 * we have reassociated it with deadfs. This prevents the system
1167 * from crashing on the vnode due to it being unexpectedly marked
1168 * as inactive or reclaimed.
1170 if (active
&& (flags
& DOCLOSE
)) {
1171 vp
->v_flag
&= ~(VINACTIVE
|VRECLAIMED
);
1176 * Eliminate all activity associated with the requested vnode
1177 * and with all vnodes aliased to the requested vnode.
1179 * The vnode must be referenced but should not be locked.
1182 vrevoke(struct vnode
*vp
, struct ucred
*cred
)
1191 * If the vnode has a device association, scrap all vnodes associated
1192 * with the device. Don't let the device disappear on us while we
1193 * are scrapping the vnodes.
1195 * The passed vp will probably show up in the list, do not VX lock
1198 * Releasing the vnode's rdev here can mess up specfs's call to
1199 * device close, so don't do it. The vnode has been disassociated
1200 * and the device will be closed after the last ref on the related
1201 * fp goes away (if not still open by e.g. the kernel).
1203 if (vp
->v_type
!= VCHR
) {
1204 error
= fdrevoke(vp
, DTYPE_VNODE
, cred
);
1207 if ((dev
= vp
->v_rdev
) == NULL
) {
1211 lwkt_gettoken(&ilock
, &spechash_token
);
1213 vqn
= SLIST_FIRST(&dev
->si_hlist
);
1216 while ((vq
= vqn
) != NULL
) {
1217 vqn
= SLIST_NEXT(vqn
, v_cdevnext
);
1220 fdrevoke(vq
, DTYPE_VNODE
, cred
);
1221 /*v_release_rdev(vq);*/
1224 lwkt_reltoken(&ilock
);
1231 * This is called when the object underlying a vnode is being destroyed,
1232 * such as in a remove(). Try to recycle the vnode immediately if the
1233 * only active reference is our reference.
1235 * Directory vnodes in the namecache with children cannot be immediately
1236 * recycled because numerous VOP_N*() ops require them to be stable.
1239 vrecycle(struct vnode
*vp
)
1241 if (vp
->v_sysref
.refcnt
<= 1) {
1242 if (cache_inval_vp_nonblock(vp
))
1251 * Return the maximum I/O size allowed for strategy calls on VP.
1253 * If vp is VCHR or VBLK we dive the device, otherwise we use
1254 * the vp's mount info.
1257 vmaxiosize(struct vnode
*vp
)
1259 if (vp
->v_type
== VBLK
|| vp
->v_type
== VCHR
) {
1260 return(vp
->v_rdev
->si_iosize_max
);
1262 return(vp
->v_mount
->mnt_iosize_max
);
1267 * Eliminate all activity associated with a vnode in preparation for reuse.
1269 * The vnode must be VX locked and refd and will remain VX locked and refd
1270 * on return. This routine may be called with the vnode in any state, as
1271 * long as it is VX locked. The vnode will be cleaned out and marked
1272 * VRECLAIMED but will not actually be reused until all existing refs and
1275 * NOTE: This routine may be called on a vnode which has not yet been
1276 * already been deactivated (VOP_INACTIVE), or on a vnode which has
1277 * already been reclaimed.
1279 * This routine is not responsible for placing us back on the freelist.
1280 * Instead, it happens automatically when the caller releases the VX lock
1281 * (assuming there aren't any other references).
1285 vgone_vxlocked(struct vnode
*vp
)
1288 * assert that the VX lock is held. This is an absolute requirement
1289 * now for vgone_vxlocked() to be called.
1291 KKASSERT(vp
->v_lock
.lk_exclusivecount
== 1);
1294 * Clean out the filesystem specific data and set the VRECLAIMED
1295 * bit. Also deactivate the vnode if necessary.
1297 vclean_vxlocked(vp
, DOCLOSE
);
1300 * Delete from old mount point vnode list, if on one.
1302 if (vp
->v_mount
!= NULL
)
1303 insmntque(vp
, NULL
);
1306 * If special device, remove it from special device alias list
1307 * if it is on one. This should normally only occur if a vnode is
1308 * being revoked as the device should otherwise have been released
1311 if ((vp
->v_type
== VBLK
|| vp
->v_type
== VCHR
) && vp
->v_rdev
!= NULL
) {
1322 * Lookup a vnode by device number.
1324 * Returns non-zero and *vpp set to a vref'd vnode on success.
1325 * Returns zero on failure.
1328 vfinddev(cdev_t dev
, enum vtype type
, struct vnode
**vpp
)
1333 lwkt_gettoken(&ilock
, &spechash_token
);
1334 SLIST_FOREACH(vp
, &dev
->si_hlist
, v_cdevnext
) {
1335 if (type
== vp
->v_type
) {
1338 lwkt_reltoken(&ilock
);
1342 lwkt_reltoken(&ilock
);
1347 * Calculate the total number of references to a special device. This
1348 * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1349 * an overloaded field. Since udev2dev can now return NULL, we have
1350 * to check for a NULL v_rdev.
1353 count_dev(cdev_t dev
)
1359 if (SLIST_FIRST(&dev
->si_hlist
)) {
1360 lwkt_gettoken(&ilock
, &spechash_token
);
1361 SLIST_FOREACH(vp
, &dev
->si_hlist
, v_cdevnext
) {
1362 count
+= vp
->v_opencount
;
1364 lwkt_reltoken(&ilock
);
1370 vcount(struct vnode
*vp
)
1372 if (vp
->v_rdev
== NULL
)
1374 return(count_dev(vp
->v_rdev
));
1378 * Initialize VMIO for a vnode. This routine MUST be called before a
1379 * VFS can issue buffer cache ops on a vnode. It is typically called
1380 * when a vnode is initialized from its inode.
1383 vinitvmio(struct vnode
*vp
, off_t filesize
)
1389 if ((object
= vp
->v_object
) == NULL
) {
1390 object
= vnode_pager_alloc(vp
, filesize
, 0, 0);
1392 * Dereference the reference we just created. This assumes
1393 * that the object is associated with the vp.
1395 object
->ref_count
--;
1398 if (object
->flags
& OBJ_DEAD
) {
1400 vm_object_dead_sleep(object
, "vodead");
1401 vn_lock(vp
, LK_EXCLUSIVE
| LK_RETRY
);
1405 KASSERT(vp
->v_object
!= NULL
, ("vinitvmio: NULL object"));
1406 vp
->v_flag
|= VOBJBUF
;
1412 * Print out a description of a vnode.
1414 static char *typename
[] =
1415 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1418 vprint(char *label
, struct vnode
*vp
)
1423 kprintf("%s: %p: ", label
, (void *)vp
);
1425 kprintf("%p: ", (void *)vp
);
1426 kprintf("type %s, sysrefs %d, writecount %d, holdcnt %d,",
1427 typename
[vp
->v_type
],
1428 vp
->v_sysref
.refcnt
, vp
->v_writecount
, vp
->v_auxrefs
);
1430 if (vp
->v_flag
& VROOT
)
1431 strcat(buf
, "|VROOT");
1432 if (vp
->v_flag
& VPFSROOT
)
1433 strcat(buf
, "|VPFSROOT");
1434 if (vp
->v_flag
& VTEXT
)
1435 strcat(buf
, "|VTEXT");
1436 if (vp
->v_flag
& VSYSTEM
)
1437 strcat(buf
, "|VSYSTEM");
1438 if (vp
->v_flag
& VFREE
)
1439 strcat(buf
, "|VFREE");
1440 if (vp
->v_flag
& VOBJBUF
)
1441 strcat(buf
, "|VOBJBUF");
1443 kprintf(" flags (%s)", &buf
[1]);
1444 if (vp
->v_data
== NULL
) {
1453 * Do the usual access checking.
1454 * file_mode, uid and gid are from the vnode in question,
1455 * while acc_mode and cred are from the VOP_ACCESS parameter list
1458 vaccess(enum vtype type
, mode_t file_mode
, uid_t uid
, gid_t gid
,
1459 mode_t acc_mode
, struct ucred
*cred
)
1465 * Super-user always gets read/write access, but execute access depends
1466 * on at least one execute bit being set.
1468 if (priv_check_cred(cred
, PRIV_ROOT
, 0) == 0) {
1469 if ((acc_mode
& VEXEC
) && type
!= VDIR
&&
1470 (file_mode
& (S_IXUSR
|S_IXGRP
|S_IXOTH
)) == 0)
1477 /* Otherwise, check the owner. */
1478 if (cred
->cr_uid
== uid
) {
1479 if (acc_mode
& VEXEC
)
1481 if (acc_mode
& VREAD
)
1483 if (acc_mode
& VWRITE
)
1485 return ((file_mode
& mask
) == mask
? 0 : EACCES
);
1488 /* Otherwise, check the groups. */
1489 ismember
= groupmember(gid
, cred
);
1490 if (cred
->cr_svgid
== gid
|| ismember
) {
1491 if (acc_mode
& VEXEC
)
1493 if (acc_mode
& VREAD
)
1495 if (acc_mode
& VWRITE
)
1497 return ((file_mode
& mask
) == mask
? 0 : EACCES
);
1500 /* Otherwise, check everyone else. */
1501 if (acc_mode
& VEXEC
)
1503 if (acc_mode
& VREAD
)
1505 if (acc_mode
& VWRITE
)
1507 return ((file_mode
& mask
) == mask
? 0 : EACCES
);
1511 #include <ddb/ddb.h>
1513 static int db_show_locked_vnodes(struct mount
*mp
, void *data
);
1516 * List all of the locked vnodes in the system.
1517 * Called when debugging the kernel.
1519 DB_SHOW_COMMAND(lockedvnodes
, lockedvnodes
)
1521 kprintf("Locked vnodes\n");
1522 mountlist_scan(db_show_locked_vnodes
, NULL
,
1523 MNTSCAN_FORWARD
|MNTSCAN_NOBUSY
);
1527 db_show_locked_vnodes(struct mount
*mp
, void *data __unused
)
1531 TAILQ_FOREACH(vp
, &mp
->mnt_nvnodelist
, v_nmntvnodes
) {
1532 if (vn_islocked(vp
))
1540 * Top level filesystem related information gathering.
1542 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS
);
1545 vfs_sysctl(SYSCTL_HANDLER_ARGS
)
1547 int *name
= (int *)arg1
- 1; /* XXX */
1548 u_int namelen
= arg2
+ 1; /* XXX */
1549 struct vfsconf
*vfsp
;
1552 #if 1 || defined(COMPAT_PRELITE2)
1553 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1555 return (sysctl_ovfs_conf(oidp
, arg1
, arg2
, req
));
1559 /* all sysctl names at this level are at least name and field */
1561 return (ENOTDIR
); /* overloaded */
1562 if (name
[0] != VFS_GENERIC
) {
1563 vfsp
= vfsconf_find_by_typenum(name
[0]);
1565 return (EOPNOTSUPP
);
1566 return ((*vfsp
->vfc_vfsops
->vfs_sysctl
)(&name
[1], namelen
- 1,
1567 oldp
, oldlenp
, newp
, newlen
, p
));
1571 case VFS_MAXTYPENUM
:
1574 maxtypenum
= vfsconf_get_maxtypenum();
1575 return (SYSCTL_OUT(req
, &maxtypenum
, sizeof(maxtypenum
)));
1578 return (ENOTDIR
); /* overloaded */
1579 vfsp
= vfsconf_find_by_typenum(name
[2]);
1581 return (EOPNOTSUPP
);
1582 return (SYSCTL_OUT(req
, vfsp
, sizeof *vfsp
));
1584 return (EOPNOTSUPP
);
1587 SYSCTL_NODE(_vfs
, VFS_GENERIC
, generic
, CTLFLAG_RD
, vfs_sysctl
,
1588 "Generic filesystem");
1590 #if 1 || defined(COMPAT_PRELITE2)
1593 sysctl_ovfs_conf_iter(struct vfsconf
*vfsp
, void *data
)
1596 struct ovfsconf ovfs
;
1597 struct sysctl_req
*req
= (struct sysctl_req
*) data
;
1599 bzero(&ovfs
, sizeof(ovfs
));
1600 ovfs
.vfc_vfsops
= vfsp
->vfc_vfsops
; /* XXX used as flag */
1601 strcpy(ovfs
.vfc_name
, vfsp
->vfc_name
);
1602 ovfs
.vfc_index
= vfsp
->vfc_typenum
;
1603 ovfs
.vfc_refcount
= vfsp
->vfc_refcount
;
1604 ovfs
.vfc_flags
= vfsp
->vfc_flags
;
1605 error
= SYSCTL_OUT(req
, &ovfs
, sizeof ovfs
);
1607 return error
; /* abort iteration with error code */
1609 return 0; /* continue iterating with next element */
1613 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS
)
1615 return vfsconf_each(sysctl_ovfs_conf_iter
, (void*)req
);
1618 #endif /* 1 || COMPAT_PRELITE2 */
1621 * Check to see if a filesystem is mounted on a block device.
1624 vfs_mountedon(struct vnode
*vp
)
1628 if ((dev
= vp
->v_rdev
) == NULL
) {
1629 /* if (vp->v_type != VBLK)
1630 dev = get_dev(vp->v_uminor, vp->v_umajor); */
1632 if (dev
!= NULL
&& dev
->si_mountpoint
)
1638 * Unmount all filesystems. The list is traversed in reverse order
1639 * of mounting to avoid dependencies.
1642 static int vfs_umountall_callback(struct mount
*mp
, void *data
);
1645 vfs_unmountall(void)
1650 count
= mountlist_scan(vfs_umountall_callback
,
1651 NULL
, MNTSCAN_REVERSE
|MNTSCAN_NOBUSY
);
1657 vfs_umountall_callback(struct mount
*mp
, void *data
)
1661 error
= dounmount(mp
, MNT_FORCE
);
1663 mountlist_remove(mp
);
1664 kprintf("unmount of filesystem mounted from %s failed (",
1665 mp
->mnt_stat
.f_mntfromname
);
1669 kprintf("%d)\n", error
);
1675 * Checks the mount flags for parameter mp and put the names comma-separated
1676 * into a string buffer buf with a size limit specified by len.
1678 * It returns the number of bytes written into buf, and (*errorp) will be
1679 * set to 0, EINVAL (if passed length is 0), or ENOSPC (supplied buffer was
1680 * not large enough). The buffer will be 0-terminated if len was not 0.
1683 vfs_flagstostr(int flags
, const struct mountctl_opt
*optp
,
1684 char *buf
, size_t len
, int *errorp
)
1686 static const struct mountctl_opt optnames
[] = {
1687 { MNT_ASYNC
, "asynchronous" },
1688 { MNT_EXPORTED
, "NFS exported" },
1689 { MNT_LOCAL
, "local" },
1690 { MNT_NOATIME
, "noatime" },
1691 { MNT_NODEV
, "nodev" },
1692 { MNT_NOEXEC
, "noexec" },
1693 { MNT_NOSUID
, "nosuid" },
1694 { MNT_NOSYMFOLLOW
, "nosymfollow" },
1695 { MNT_QUOTA
, "with-quotas" },
1696 { MNT_RDONLY
, "read-only" },
1697 { MNT_SYNCHRONOUS
, "synchronous" },
1698 { MNT_UNION
, "union" },
1699 { MNT_NOCLUSTERR
, "noclusterr" },
1700 { MNT_NOCLUSTERW
, "noclusterw" },
1701 { MNT_SUIDDIR
, "suiddir" },
1702 { MNT_SOFTDEP
, "soft-updates" },
1703 { MNT_IGNORE
, "ignore" },
1713 bleft
= len
- 1; /* leave room for trailing \0 */
1716 * Checks the size of the string. If it contains
1717 * any data, then we will append the new flags to
1720 actsize
= strlen(buf
);
1724 /* Default flags if no flags passed */
1728 if (bleft
< 0) { /* degenerate case, 0-length buffer */
1733 for (; flags
&& optp
->o_opt
; ++optp
) {
1734 if ((flags
& optp
->o_opt
) == 0)
1736 optlen
= strlen(optp
->o_name
);
1737 if (bwritten
|| actsize
> 0) {
1742 buf
[bwritten
++] = ',';
1743 buf
[bwritten
++] = ' ';
1746 if (bleft
< optlen
) {
1750 bcopy(optp
->o_name
, buf
+ bwritten
, optlen
);
1753 flags
&= ~optp
->o_opt
;
1757 * Space already reserved for trailing \0
1764 * Build hash lists of net addresses and hang them off the mount point.
1765 * Called by ufs_mount() to set up the lists of export addresses.
1768 vfs_hang_addrlist(struct mount
*mp
, struct netexport
*nep
,
1769 const struct export_args
*argp
)
1772 struct radix_node_head
*rnh
;
1774 struct radix_node
*rn
;
1775 struct sockaddr
*saddr
, *smask
= 0;
1779 if (argp
->ex_addrlen
== 0) {
1780 if (mp
->mnt_flag
& MNT_DEFEXPORTED
)
1782 np
= &nep
->ne_defexported
;
1783 np
->netc_exflags
= argp
->ex_flags
;
1784 np
->netc_anon
= argp
->ex_anon
;
1785 np
->netc_anon
.cr_ref
= 1;
1786 mp
->mnt_flag
|= MNT_DEFEXPORTED
;
1790 if (argp
->ex_addrlen
< 0 || argp
->ex_addrlen
> MLEN
)
1792 if (argp
->ex_masklen
< 0 || argp
->ex_masklen
> MLEN
)
1795 i
= sizeof(struct netcred
) + argp
->ex_addrlen
+ argp
->ex_masklen
;
1796 np
= (struct netcred
*) kmalloc(i
, M_NETADDR
, M_WAITOK
| M_ZERO
);
1797 saddr
= (struct sockaddr
*) (np
+ 1);
1798 if ((error
= copyin(argp
->ex_addr
, (caddr_t
) saddr
, argp
->ex_addrlen
)))
1800 if (saddr
->sa_len
> argp
->ex_addrlen
)
1801 saddr
->sa_len
= argp
->ex_addrlen
;
1802 if (argp
->ex_masklen
) {
1803 smask
= (struct sockaddr
*)((caddr_t
)saddr
+ argp
->ex_addrlen
);
1804 error
= copyin(argp
->ex_mask
, (caddr_t
)smask
, argp
->ex_masklen
);
1807 if (smask
->sa_len
> argp
->ex_masklen
)
1808 smask
->sa_len
= argp
->ex_masklen
;
1810 i
= saddr
->sa_family
;
1811 if ((rnh
= nep
->ne_rtable
[i
]) == 0) {
1813 * Seems silly to initialize every AF when most are not used,
1814 * do so on demand here
1816 SLIST_FOREACH(dom
, &domains
, dom_next
)
1817 if (dom
->dom_family
== i
&& dom
->dom_rtattach
) {
1818 dom
->dom_rtattach((void **) &nep
->ne_rtable
[i
],
1822 if ((rnh
= nep
->ne_rtable
[i
]) == 0) {
1827 rn
= (*rnh
->rnh_addaddr
) ((char *) saddr
, (char *) smask
, rnh
,
1829 if (rn
== 0 || np
!= (struct netcred
*) rn
) { /* already exists */
1833 np
->netc_exflags
= argp
->ex_flags
;
1834 np
->netc_anon
= argp
->ex_anon
;
1835 np
->netc_anon
.cr_ref
= 1;
1838 kfree(np
, M_NETADDR
);
1844 vfs_free_netcred(struct radix_node
*rn
, void *w
)
1846 struct radix_node_head
*rnh
= (struct radix_node_head
*) w
;
1848 (*rnh
->rnh_deladdr
) (rn
->rn_key
, rn
->rn_mask
, rnh
);
1849 kfree((caddr_t
) rn
, M_NETADDR
);
1854 * Free the net address hash lists that are hanging off the mount points.
1857 vfs_free_addrlist(struct netexport
*nep
)
1860 struct radix_node_head
*rnh
;
1862 for (i
= 0; i
<= AF_MAX
; i
++)
1863 if ((rnh
= nep
->ne_rtable
[i
])) {
1864 (*rnh
->rnh_walktree
) (rnh
, vfs_free_netcred
,
1866 kfree((caddr_t
) rnh
, M_RTABLE
);
1867 nep
->ne_rtable
[i
] = 0;
1872 vfs_export(struct mount
*mp
, struct netexport
*nep
,
1873 const struct export_args
*argp
)
1877 if (argp
->ex_flags
& MNT_DELEXPORT
) {
1878 if (mp
->mnt_flag
& MNT_EXPUBLIC
) {
1879 vfs_setpublicfs(NULL
, NULL
, NULL
);
1880 mp
->mnt_flag
&= ~MNT_EXPUBLIC
;
1882 vfs_free_addrlist(nep
);
1883 mp
->mnt_flag
&= ~(MNT_EXPORTED
| MNT_DEFEXPORTED
);
1885 if (argp
->ex_flags
& MNT_EXPORTED
) {
1886 if (argp
->ex_flags
& MNT_EXPUBLIC
) {
1887 if ((error
= vfs_setpublicfs(mp
, nep
, argp
)) != 0)
1889 mp
->mnt_flag
|= MNT_EXPUBLIC
;
1891 if ((error
= vfs_hang_addrlist(mp
, nep
, argp
)))
1893 mp
->mnt_flag
|= MNT_EXPORTED
;
1900 * Set the publicly exported filesystem (WebNFS). Currently, only
1901 * one public filesystem is possible in the spec (RFC 2054 and 2055)
1904 vfs_setpublicfs(struct mount
*mp
, struct netexport
*nep
,
1905 const struct export_args
*argp
)
1912 * mp == NULL -> invalidate the current info, the FS is
1913 * no longer exported. May be called from either vfs_export
1914 * or unmount, so check if it hasn't already been done.
1917 if (nfs_pub
.np_valid
) {
1918 nfs_pub
.np_valid
= 0;
1919 if (nfs_pub
.np_index
!= NULL
) {
1920 FREE(nfs_pub
.np_index
, M_TEMP
);
1921 nfs_pub
.np_index
= NULL
;
1928 * Only one allowed at a time.
1930 if (nfs_pub
.np_valid
!= 0 && mp
!= nfs_pub
.np_mount
)
1934 * Get real filehandle for root of exported FS.
1936 bzero((caddr_t
)&nfs_pub
.np_handle
, sizeof(nfs_pub
.np_handle
));
1937 nfs_pub
.np_handle
.fh_fsid
= mp
->mnt_stat
.f_fsid
;
1939 if ((error
= VFS_ROOT(mp
, &rvp
)))
1942 if ((error
= VFS_VPTOFH(rvp
, &nfs_pub
.np_handle
.fh_fid
)))
1948 * If an indexfile was specified, pull it in.
1950 if (argp
->ex_indexfile
!= NULL
) {
1953 error
= vn_get_namelen(rvp
, &namelen
);
1956 MALLOC(nfs_pub
.np_index
, char *, namelen
, M_TEMP
,
1958 error
= copyinstr(argp
->ex_indexfile
, nfs_pub
.np_index
,
1962 * Check for illegal filenames.
1964 for (cp
= nfs_pub
.np_index
; *cp
; cp
++) {
1972 FREE(nfs_pub
.np_index
, M_TEMP
);
1977 nfs_pub
.np_mount
= mp
;
1978 nfs_pub
.np_valid
= 1;
1983 vfs_export_lookup(struct mount
*mp
, struct netexport
*nep
,
1984 struct sockaddr
*nam
)
1987 struct radix_node_head
*rnh
;
1988 struct sockaddr
*saddr
;
1991 if (mp
->mnt_flag
& MNT_EXPORTED
) {
1993 * Lookup in the export list first.
1997 rnh
= nep
->ne_rtable
[saddr
->sa_family
];
1999 np
= (struct netcred
*)
2000 (*rnh
->rnh_matchaddr
)((char *)saddr
,
2002 if (np
&& np
->netc_rnodes
->rn_flags
& RNF_ROOT
)
2007 * If no address match, use the default if it exists.
2009 if (np
== NULL
&& mp
->mnt_flag
& MNT_DEFEXPORTED
)
2010 np
= &nep
->ne_defexported
;
2016 * perform msync on all vnodes under a mount point. The mount point must
2017 * be locked. This code is also responsible for lazy-freeing unreferenced
2018 * vnodes whos VM objects no longer contain pages.
2020 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
2022 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked,
2023 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it
2024 * way up in this high level function.
2026 static int vfs_msync_scan1(struct mount
*mp
, struct vnode
*vp
, void *data
);
2027 static int vfs_msync_scan2(struct mount
*mp
, struct vnode
*vp
, void *data
);
2030 vfs_msync(struct mount
*mp
, int flags
)
2034 vmsc_flags
= VMSC_GETVP
;
2035 if (flags
!= MNT_WAIT
)
2036 vmsc_flags
|= VMSC_NOWAIT
;
2037 vmntvnodescan(mp
, vmsc_flags
, vfs_msync_scan1
, vfs_msync_scan2
,
2038 (void *)(intptr_t)flags
);
2042 * scan1 is a fast pre-check. There could be hundreds of thousands of
2043 * vnodes, we cannot afford to do anything heavy weight until we have a
2044 * fairly good indication that there is work to do.
2048 vfs_msync_scan1(struct mount
*mp
, struct vnode
*vp
, void *data
)
2050 int flags
= (int)(intptr_t)data
;
2052 if ((vp
->v_flag
& VRECLAIMED
) == 0) {
2053 if (vshouldmsync(vp
))
2054 return(0); /* call scan2 */
2055 if ((mp
->mnt_flag
& MNT_RDONLY
) == 0 &&
2056 (vp
->v_flag
& VOBJDIRTY
) &&
2057 (flags
== MNT_WAIT
|| vn_islocked(vp
) == 0)) {
2058 return(0); /* call scan2 */
2063 * do not call scan2, continue the loop
2069 * This callback is handed a locked vnode.
2073 vfs_msync_scan2(struct mount
*mp
, struct vnode
*vp
, void *data
)
2076 int flags
= (int)(intptr_t)data
;
2078 if (vp
->v_flag
& VRECLAIMED
)
2081 if ((mp
->mnt_flag
& MNT_RDONLY
) == 0 && (vp
->v_flag
& VOBJDIRTY
)) {
2082 if ((obj
= vp
->v_object
) != NULL
) {
2083 vm_object_page_clean(obj
, 0, 0,
2084 flags
== MNT_WAIT
? OBJPC_SYNC
: OBJPC_NOSYNC
);
2091 * Record a process's interest in events which might happen to
2092 * a vnode. Because poll uses the historic select-style interface
2093 * internally, this routine serves as both the ``check for any
2094 * pending events'' and the ``record my interest in future events''
2095 * functions. (These are done together, while the lock is held,
2096 * to avoid race conditions.)
2099 vn_pollrecord(struct vnode
*vp
, int events
)
2103 KKASSERT(curthread
->td_proc
!= NULL
);
2105 lwkt_gettoken(&vlock
, &vp
->v_token
);
2106 if (vp
->v_pollinfo
.vpi_revents
& events
) {
2108 * This leaves events we are not interested
2109 * in available for the other process which
2110 * which presumably had requested them
2111 * (otherwise they would never have been
2114 events
&= vp
->v_pollinfo
.vpi_revents
;
2115 vp
->v_pollinfo
.vpi_revents
&= ~events
;
2117 lwkt_reltoken(&vlock
);
2120 vp
->v_pollinfo
.vpi_events
|= events
;
2121 selrecord(curthread
, &vp
->v_pollinfo
.vpi_selinfo
);
2122 lwkt_reltoken(&vlock
);
2127 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
2128 * it is possible for us to miss an event due to race conditions, but
2129 * that condition is expected to be rare, so for the moment it is the
2130 * preferred interface.
2133 vn_pollevent(struct vnode
*vp
, int events
)
2137 lwkt_gettoken(&vlock
, &vp
->v_token
);
2138 if (vp
->v_pollinfo
.vpi_events
& events
) {
2140 * We clear vpi_events so that we don't
2141 * call selwakeup() twice if two events are
2142 * posted before the polling process(es) is
2143 * awakened. This also ensures that we take at
2144 * most one selwakeup() if the polling process
2145 * is no longer interested. However, it does
2146 * mean that only one event can be noticed at
2147 * a time. (Perhaps we should only clear those
2148 * event bits which we note?) XXX
2150 vp
->v_pollinfo
.vpi_events
= 0; /* &= ~events ??? */
2151 vp
->v_pollinfo
.vpi_revents
|= events
;
2152 selwakeup(&vp
->v_pollinfo
.vpi_selinfo
);
2154 lwkt_reltoken(&vlock
);
2158 * Wake up anyone polling on vp because it is being revoked.
2159 * This depends on dead_poll() returning POLLHUP for correct
2163 vn_pollgone(struct vnode
*vp
)
2167 lwkt_gettoken(&vlock
, &vp
->v_token
);
2168 if (vp
->v_pollinfo
.vpi_events
) {
2169 vp
->v_pollinfo
.vpi_events
= 0;
2170 selwakeup(&vp
->v_pollinfo
.vpi_selinfo
);
2172 lwkt_reltoken(&vlock
);
2176 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened
2177 * (or v_rdev might be NULL).
2180 vn_todev(struct vnode
*vp
)
2182 if (vp
->v_type
!= VBLK
&& vp
->v_type
!= VCHR
)
2184 KKASSERT(vp
->v_rdev
!= NULL
);
2185 return (vp
->v_rdev
);
2189 * Check if vnode represents a disk device. The vnode does not need to be
2195 vn_isdisk(struct vnode
*vp
, int *errp
)
2199 if (vp
->v_type
!= VCHR
) {
2212 if (dev_is_good(dev
) == 0) {
2217 if ((dev_dflags(dev
) & D_DISK
) == 0) {
2228 vn_get_namelen(struct vnode
*vp
, int *namelen
)
2231 register_t retval
[2];
2233 error
= VOP_PATHCONF(vp
, _PC_NAME_MAX
, retval
);
2236 *namelen
= (int)retval
[0];
2241 vop_write_dirent(int *error
, struct uio
*uio
, ino_t d_ino
, uint8_t d_type
,
2242 uint16_t d_namlen
, const char *d_name
)
2247 len
= _DIRENT_RECLEN(d_namlen
);
2248 if (len
> uio
->uio_resid
)
2251 dp
= kmalloc(len
, M_TEMP
, M_WAITOK
| M_ZERO
);
2254 dp
->d_namlen
= d_namlen
;
2255 dp
->d_type
= d_type
;
2256 bcopy(d_name
, dp
->d_name
, d_namlen
);
2258 *error
= uiomove((caddr_t
)dp
, len
, uio
);
2266 vn_mark_atime(struct vnode
*vp
, struct thread
*td
)
2268 struct proc
*p
= td
->td_proc
;
2269 struct ucred
*cred
= p
? p
->p_ucred
: proc0
.p_ucred
;
2271 if ((vp
->v_mount
->mnt_flag
& (MNT_NOATIME
| MNT_RDONLY
)) == 0) {
2272 VOP_MARKATIME(vp
, cred
);