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>
63 #include <sys/reboot.h>
64 #include <sys/socket.h>
66 #include <sys/sysctl.h>
67 #include <sys/syslog.h>
68 #include <sys/unistd.h>
69 #include <sys/vmmeter.h>
70 #include <sys/vnode.h>
72 #include <machine/limits.h>
75 #include <vm/vm_object.h>
76 #include <vm/vm_extern.h>
77 #include <vm/vm_kern.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_page.h>
81 #include <vm/vm_pager.h>
82 #include <vm/vnode_pager.h>
83 #include <vm/vm_zone.h>
86 #include <sys/thread2.h>
87 #include <sys/sysref2.h>
89 static MALLOC_DEFINE(M_NETADDR
, "Export Host", "Export host address structure");
92 SYSCTL_INT(_debug
, OID_AUTO
, numvnodes
, CTLFLAG_RD
, &numvnodes
, 0, "");
94 SYSCTL_INT(_vfs
, OID_AUTO
, fastdev
, CTLFLAG_RW
, &vfs_fastdev
, 0, "");
96 enum vtype iftovt_tab
[16] = {
97 VNON
, VFIFO
, VCHR
, VNON
, VDIR
, VNON
, VBLK
, VNON
,
98 VREG
, VNON
, VLNK
, VNON
, VSOCK
, VNON
, VNON
, VBAD
,
100 int vttoif_tab
[9] = {
101 0, S_IFREG
, S_IFDIR
, S_IFBLK
, S_IFCHR
, S_IFLNK
,
102 S_IFSOCK
, S_IFIFO
, S_IFMT
,
105 static int reassignbufcalls
;
106 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufcalls
, CTLFLAG_RW
,
107 &reassignbufcalls
, 0, "");
108 static int reassignbufloops
;
109 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufloops
, CTLFLAG_RW
,
110 &reassignbufloops
, 0, "");
111 static int reassignbufsortgood
;
112 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufsortgood
, CTLFLAG_RW
,
113 &reassignbufsortgood
, 0, "");
114 static int reassignbufsortbad
;
115 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufsortbad
, CTLFLAG_RW
,
116 &reassignbufsortbad
, 0, "");
117 static int reassignbufmethod
= 1;
118 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufmethod
, CTLFLAG_RW
,
119 &reassignbufmethod
, 0, "");
121 int nfs_mount_type
= -1;
122 static struct lwkt_token spechash_token
;
123 struct nfs_public nfs_pub
; /* publicly exported FS */
126 SYSCTL_INT(_kern
, KERN_MAXVNODES
, maxvnodes
, CTLFLAG_RW
,
127 &desiredvnodes
, 0, "Maximum number of vnodes");
129 static void vfs_free_addrlist (struct netexport
*nep
);
130 static int vfs_free_netcred (struct radix_node
*rn
, void *w
);
131 static int vfs_hang_addrlist (struct mount
*mp
, struct netexport
*nep
,
132 const struct export_args
*argp
);
135 * Red black tree functions
137 static int rb_buf_compare(struct buf
*b1
, struct buf
*b2
);
138 RB_GENERATE2(buf_rb_tree
, buf
, b_rbnode
, rb_buf_compare
, off_t
, b_loffset
);
139 RB_GENERATE2(buf_rb_hash
, buf
, b_rbhash
, rb_buf_compare
, off_t
, b_loffset
);
142 rb_buf_compare(struct buf
*b1
, struct buf
*b2
)
144 if (b1
->b_loffset
< b2
->b_loffset
)
146 if (b1
->b_loffset
> b2
->b_loffset
)
152 * Returns non-zero if the vnode is a candidate for lazy msyncing.
155 vshouldmsync(struct vnode
*vp
)
157 if (vp
->v_auxrefs
!= 0 || vp
->v_sysref
.refcnt
> 0)
158 return (0); /* other holders */
160 (vp
->v_object
->ref_count
|| vp
->v_object
->resident_page_count
)) {
167 * Initialize the vnode management data structures.
169 * Called from vfsinit()
175 * Desiredvnodes is kern.maxvnodes. We want to scale it
176 * according to available system memory but we may also have
177 * to limit it based on available KVM, which is capped on 32 bit
180 desiredvnodes
= min(maxproc
+ vmstats
.v_page_count
/ 4,
182 (sizeof(struct vm_object
) + sizeof(struct vnode
))));
184 lwkt_token_init(&spechash_token
);
188 * Knob to control the precision of file timestamps:
190 * 0 = seconds only; nanoseconds zeroed.
191 * 1 = seconds and nanoseconds, accurate within 1/HZ.
192 * 2 = seconds and nanoseconds, truncated to microseconds.
193 * >=3 = seconds and nanoseconds, maximum precision.
195 enum { TSP_SEC
, TSP_HZ
, TSP_USEC
, TSP_NSEC
};
197 static int timestamp_precision
= TSP_SEC
;
198 SYSCTL_INT(_vfs
, OID_AUTO
, timestamp_precision
, CTLFLAG_RW
,
199 ×tamp_precision
, 0, "");
202 * Get a current timestamp.
207 vfs_timestamp(struct timespec
*tsp
)
211 switch (timestamp_precision
) {
213 tsp
->tv_sec
= time_second
;
221 TIMEVAL_TO_TIMESPEC(&tv
, tsp
);
231 * Set vnode attributes to VNOVAL
234 vattr_null(struct vattr
*vap
)
237 vap
->va_size
= VNOVAL
;
238 vap
->va_bytes
= VNOVAL
;
239 vap
->va_mode
= VNOVAL
;
240 vap
->va_nlink
= VNOVAL
;
241 vap
->va_uid
= VNOVAL
;
242 vap
->va_gid
= VNOVAL
;
243 vap
->va_fsid
= VNOVAL
;
244 vap
->va_fileid
= VNOVAL
;
245 vap
->va_blocksize
= VNOVAL
;
246 vap
->va_rmajor
= VNOVAL
;
247 vap
->va_rminor
= VNOVAL
;
248 vap
->va_atime
.tv_sec
= VNOVAL
;
249 vap
->va_atime
.tv_nsec
= VNOVAL
;
250 vap
->va_mtime
.tv_sec
= VNOVAL
;
251 vap
->va_mtime
.tv_nsec
= VNOVAL
;
252 vap
->va_ctime
.tv_sec
= VNOVAL
;
253 vap
->va_ctime
.tv_nsec
= VNOVAL
;
254 vap
->va_flags
= VNOVAL
;
255 vap
->va_gen
= VNOVAL
;
257 vap
->va_fsmid
= 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 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 lwkt_gettoken(&vlock
, &vp
->v_token
);
292 * If we are being asked to save, call fsync to ensure that the inode
295 if (flags
& V_SAVE
) {
296 error
= bio_track_wait(&vp
->v_track_write
, slpflag
, slptimeo
);
299 if (!RB_EMPTY(&vp
->v_rbdirty_tree
)) {
300 if ((error
= VOP_FSYNC(vp
, MNT_WAIT
, 0)) != 0)
304 * Dirty bufs may be left or generated via races
305 * in circumstances where vinvalbuf() is called on
306 * a vnode not undergoing reclamation. Only
307 * panic if we are trying to reclaim the vnode.
309 if ((vp
->v_flag
& VRECLAIMED
) &&
310 (bio_track_active(&vp
->v_track_write
) ||
311 !RB_EMPTY(&vp
->v_rbdirty_tree
))) {
312 panic("vinvalbuf: dirty bufs");
316 info
.slptimeo
= slptimeo
;
317 info
.lkflags
= LK_EXCLUSIVE
| LK_SLEEPFAIL
;
318 if (slpflag
& PCATCH
)
319 info
.lkflags
|= LK_PCATCH
;
324 * Flush the buffer cache until nothing is left.
326 while (!RB_EMPTY(&vp
->v_rbclean_tree
) ||
327 !RB_EMPTY(&vp
->v_rbdirty_tree
)) {
328 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbclean_tree
, NULL
,
329 vinvalbuf_bp
, &info
);
331 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, NULL
,
332 vinvalbuf_bp
, &info
);
337 * Wait for I/O completion. We may block in the pip code so we have
341 bio_track_wait(&vp
->v_track_write
, 0, 0);
342 if ((object
= vp
->v_object
) != NULL
) {
343 while (object
->paging_in_progress
)
344 vm_object_pip_sleep(object
, "vnvlbx");
346 } while (bio_track_active(&vp
->v_track_write
));
349 * Destroy the copy in the VM cache, too.
351 if ((object
= vp
->v_object
) != NULL
) {
352 vm_object_page_remove(object
, 0, 0,
353 (flags
& V_SAVE
) ? TRUE
: FALSE
);
356 if (!RB_EMPTY(&vp
->v_rbdirty_tree
) || !RB_EMPTY(&vp
->v_rbclean_tree
))
357 panic("vinvalbuf: flush failed");
358 if (!RB_EMPTY(&vp
->v_rbhash_tree
))
359 panic("vinvalbuf: flush failed, buffers still present");
362 lwkt_reltoken(&vlock
);
367 vinvalbuf_bp(struct buf
*bp
, void *data
)
369 struct vinvalbuf_bp_info
*info
= data
;
372 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
)) {
373 error
= BUF_TIMELOCK(bp
, info
->lkflags
,
374 "vinvalbuf", info
->slptimeo
);
384 KKASSERT(bp
->b_vp
== info
->vp
);
387 * XXX Since there are no node locks for NFS, I
388 * believe there is a slight chance that a delayed
389 * write will occur while sleeping just above, so
390 * check for it. Note that vfs_bio_awrite expects
391 * buffers to reside on a queue, while bwrite() and
394 * NOTE: NO B_LOCKED CHECK. Also no buf_checkwrite()
395 * check. This code will write out the buffer, period.
397 if (((bp
->b_flags
& (B_DELWRI
| B_INVAL
)) == B_DELWRI
) &&
398 (info
->flags
& V_SAVE
)) {
399 if (bp
->b_vp
== info
->vp
) {
400 if (bp
->b_flags
& B_CLUSTEROK
) {
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
);
421 bp
->b_flags
|= (B_INVAL
| B_NOCACHE
| B_RELBUF
);
428 * Truncate a file's buffer and pages to a specified length. This
429 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
432 * The vnode must be locked.
434 static int vtruncbuf_bp_trunc_cmp(struct buf
*bp
, void *data
);
435 static int vtruncbuf_bp_trunc(struct buf
*bp
, void *data
);
436 static int vtruncbuf_bp_metasync_cmp(struct buf
*bp
, void *data
);
437 static int vtruncbuf_bp_metasync(struct buf
*bp
, void *data
);
440 vtruncbuf(struct vnode
*vp
, off_t length
, int blksize
)
443 const char *filename
;
448 * Round up to the *next* block, then destroy the buffers in question.
449 * Since we are only removing some of the buffers we must rely on the
450 * scan count to determine whether a loop is necessary.
452 if ((count
= (int)(length
% blksize
)) != 0)
453 truncloffset
= length
+ (blksize
- count
);
455 truncloffset
= length
;
457 lwkt_gettoken(&vlock
, &vp
->v_token
);
459 count
= RB_SCAN(buf_rb_tree
, &vp
->v_rbclean_tree
,
460 vtruncbuf_bp_trunc_cmp
,
461 vtruncbuf_bp_trunc
, &truncloffset
);
462 count
+= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
463 vtruncbuf_bp_trunc_cmp
,
464 vtruncbuf_bp_trunc
, &truncloffset
);
468 * For safety, fsync any remaining metadata if the file is not being
469 * truncated to 0. Since the metadata does not represent the entire
470 * dirty list we have to rely on the hit count to ensure that we get
475 count
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
476 vtruncbuf_bp_metasync_cmp
,
477 vtruncbuf_bp_metasync
, vp
);
482 * Clean out any left over VM backing store.
484 * It is possible to have in-progress I/O from buffers that were
485 * not part of the truncation. This should not happen if we
486 * are truncating to 0-length.
488 vnode_pager_setsize(vp
, length
);
489 bio_track_wait(&vp
->v_track_write
, 0, 0);
491 filename
= TAILQ_FIRST(&vp
->v_namecache
) ?
492 TAILQ_FIRST(&vp
->v_namecache
)->nc_name
: "?";
495 * Make sure no buffers were instantiated while we were trying
496 * to clean out the remaining VM pages. This could occur due
497 * to busy dirty VM pages being flushed out to disk.
500 count
= RB_SCAN(buf_rb_tree
, &vp
->v_rbclean_tree
,
501 vtruncbuf_bp_trunc_cmp
,
502 vtruncbuf_bp_trunc
, &truncloffset
);
503 count
+= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
504 vtruncbuf_bp_trunc_cmp
,
505 vtruncbuf_bp_trunc
, &truncloffset
);
507 kprintf("Warning: vtruncbuf(): Had to re-clean %d "
508 "left over buffers in %s\n", count
, filename
);
512 lwkt_reltoken(&vlock
);
518 * The callback buffer is beyond the new file EOF and must be destroyed.
519 * Note that the compare function must conform to the RB_SCAN's requirements.
523 vtruncbuf_bp_trunc_cmp(struct buf
*bp
, void *data
)
525 if (bp
->b_loffset
>= *(off_t
*)data
)
532 vtruncbuf_bp_trunc(struct buf
*bp
, void *data
)
535 * Do not try to use a buffer we cannot immediately lock, but sleep
536 * anyway to prevent a livelock. The code will loop until all buffers
539 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
)) {
540 if (BUF_LOCK(bp
, LK_EXCLUSIVE
|LK_SLEEPFAIL
) == 0)
544 bp
->b_flags
|= (B_INVAL
| B_RELBUF
| B_NOCACHE
);
551 * Fsync all meta-data after truncating a file to be non-zero. Only metadata
552 * blocks (with a negative loffset) are scanned.
553 * Note that the compare function must conform to the RB_SCAN's requirements.
556 vtruncbuf_bp_metasync_cmp(struct buf
*bp
, void *data
)
558 if (bp
->b_loffset
< 0)
564 vtruncbuf_bp_metasync(struct buf
*bp
, void *data
)
566 struct vnode
*vp
= data
;
568 if (bp
->b_flags
& B_DELWRI
) {
570 * Do not try to use a buffer we cannot immediately lock,
571 * but sleep anyway to prevent a livelock. The code will
572 * loop until all buffers can be acted upon.
574 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
)) {
575 if (BUF_LOCK(bp
, LK_EXCLUSIVE
|LK_SLEEPFAIL
) == 0)
591 * vfsync - implements a multipass fsync on a file which understands
592 * dependancies and meta-data. The passed vnode must be locked. The
593 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY.
595 * When fsyncing data asynchronously just do one consolidated pass starting
596 * with the most negative block number. This may not get all the data due
599 * When fsyncing data synchronously do a data pass, then a metadata pass,
600 * then do additional data+metadata passes to try to get all the data out.
602 static int vfsync_wait_output(struct vnode
*vp
,
603 int (*waitoutput
)(struct vnode
*, struct thread
*));
604 static int vfsync_data_only_cmp(struct buf
*bp
, void *data
);
605 static int vfsync_meta_only_cmp(struct buf
*bp
, void *data
);
606 static int vfsync_lazy_range_cmp(struct buf
*bp
, void *data
);
607 static int vfsync_bp(struct buf
*bp
, void *data
);
616 int (*checkdef
)(struct buf
*);
620 vfsync(struct vnode
*vp
, int waitfor
, int passes
,
621 int (*checkdef
)(struct buf
*),
622 int (*waitoutput
)(struct vnode
*, struct thread
*))
624 struct vfsync_info info
;
628 bzero(&info
, sizeof(info
));
630 if ((info
.checkdef
= checkdef
) == NULL
)
633 lwkt_gettoken(&vlock
, &vp
->v_token
);
638 * Lazy (filesystem syncer typ) Asynchronous plus limit the
639 * number of data (not meta) pages we try to flush to 1MB.
640 * A non-zero return means that lazy limit was reached.
642 info
.lazylimit
= 1024 * 1024;
644 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
645 vfsync_lazy_range_cmp
, vfsync_bp
, &info
);
646 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
647 vfsync_meta_only_cmp
, vfsync_bp
, &info
);
650 else if (!RB_EMPTY(&vp
->v_rbdirty_tree
))
651 vn_syncer_add_to_worklist(vp
, 1);
656 * Asynchronous. Do a data-only pass and a meta-only pass.
659 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, vfsync_data_only_cmp
,
661 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, vfsync_meta_only_cmp
,
667 * Synchronous. Do a data-only pass, then a meta-data+data
668 * pass, then additional integrated passes to try to get
669 * all the dependancies flushed.
671 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, vfsync_data_only_cmp
,
673 error
= vfsync_wait_output(vp
, waitoutput
);
675 info
.skippedbufs
= 0;
676 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, NULL
,
678 error
= vfsync_wait_output(vp
, waitoutput
);
679 if (info
.skippedbufs
)
680 kprintf("Warning: vfsync skipped %d dirty bufs in pass2!\n", info
.skippedbufs
);
682 while (error
== 0 && passes
> 0 &&
683 !RB_EMPTY(&vp
->v_rbdirty_tree
)
686 info
.synchronous
= 1;
689 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, NULL
,
695 error
= vfsync_wait_output(vp
, waitoutput
);
699 lwkt_reltoken(&vlock
);
704 vfsync_wait_output(struct vnode
*vp
,
705 int (*waitoutput
)(struct vnode
*, struct thread
*))
709 error
= bio_track_wait(&vp
->v_track_write
, 0, 0);
711 error
= waitoutput(vp
, curthread
);
716 vfsync_data_only_cmp(struct buf
*bp
, void *data
)
718 if (bp
->b_loffset
< 0)
724 vfsync_meta_only_cmp(struct buf
*bp
, void *data
)
726 if (bp
->b_loffset
< 0)
732 vfsync_lazy_range_cmp(struct buf
*bp
, void *data
)
734 struct vfsync_info
*info
= data
;
735 if (bp
->b_loffset
< info
->vp
->v_lazyw
)
741 vfsync_bp(struct buf
*bp
, void *data
)
743 struct vfsync_info
*info
= data
;
744 struct vnode
*vp
= info
->vp
;
748 * if syncdeps is not set we do not try to write buffers which have
751 if (!info
->synchronous
&& info
->syncdeps
== 0 && info
->checkdef(bp
))
755 * Ignore buffers that we cannot immediately lock. XXX
757 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
)) {
758 kprintf("Warning: vfsync_bp skipping dirty buffer %p\n", bp
);
762 if ((bp
->b_flags
& B_DELWRI
) == 0)
763 panic("vfsync_bp: buffer not dirty");
765 panic("vfsync_bp: buffer vp mismatch");
768 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
769 * has been written but an additional handshake with the device
770 * is required before we can dispose of the buffer. We have no idea
771 * how to do this so we have to skip these buffers.
773 if (bp
->b_flags
& B_NEEDCOMMIT
) {
779 * Ask bioops if it is ok to sync
781 if (LIST_FIRST(&bp
->b_dep
) != NULL
&& buf_checkwrite(bp
)) {
787 if (info
->synchronous
) {
789 * Synchronous flushing. An error may be returned.
795 * Asynchronous flushing. A negative return value simply
796 * stops the scan and is not considered an error. We use
797 * this to support limited MNT_LAZY flushes.
799 vp
->v_lazyw
= bp
->b_loffset
;
800 if ((vp
->v_flag
& VOBJBUF
) && (bp
->b_flags
& B_CLUSTEROK
)) {
801 info
->lazycount
+= vfs_bio_awrite(bp
);
803 info
->lazycount
+= bp
->b_bufsize
;
807 if (info
->lazylimit
&& info
->lazycount
>= info
->lazylimit
)
816 * Associate a buffer with a vnode.
821 bgetvp(struct vnode
*vp
, struct buf
*bp
)
825 KASSERT(bp
->b_vp
== NULL
, ("bgetvp: not free"));
826 KKASSERT((bp
->b_flags
& (B_HASHED
|B_DELWRI
|B_VNCLEAN
|B_VNDIRTY
)) == 0);
829 * Insert onto list for new vnode.
831 lwkt_gettoken(&vlock
, &vp
->v_token
);
832 if (buf_rb_hash_RB_INSERT(&vp
->v_rbhash_tree
, bp
)) {
833 lwkt_reltoken(&vlock
);
837 bp
->b_flags
|= B_HASHED
;
838 bp
->b_flags
|= B_VNCLEAN
;
839 if (buf_rb_tree_RB_INSERT(&vp
->v_rbclean_tree
, bp
))
840 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp
, bp
);
842 lwkt_reltoken(&vlock
);
847 * Disassociate a buffer from a vnode.
850 brelvp(struct buf
*bp
)
855 KASSERT(bp
->b_vp
!= NULL
, ("brelvp: NULL"));
858 * Delete from old vnode list, if on one.
861 lwkt_gettoken(&vlock
, &vp
->v_token
);
862 if (bp
->b_flags
& (B_VNDIRTY
| B_VNCLEAN
)) {
863 if (bp
->b_flags
& B_VNDIRTY
)
864 buf_rb_tree_RB_REMOVE(&vp
->v_rbdirty_tree
, bp
);
866 buf_rb_tree_RB_REMOVE(&vp
->v_rbclean_tree
, bp
);
867 bp
->b_flags
&= ~(B_VNDIRTY
| B_VNCLEAN
);
869 if (bp
->b_flags
& B_HASHED
) {
870 buf_rb_hash_RB_REMOVE(&vp
->v_rbhash_tree
, bp
);
871 bp
->b_flags
&= ~B_HASHED
;
873 if ((vp
->v_flag
& VONWORKLST
) && RB_EMPTY(&vp
->v_rbdirty_tree
)) {
874 vp
->v_flag
&= ~VONWORKLST
;
875 LIST_REMOVE(vp
, v_synclist
);
878 lwkt_reltoken(&vlock
);
884 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI.
885 * This routine is called when the state of the B_DELWRI bit is changed.
890 reassignbuf(struct buf
*bp
)
892 struct vnode
*vp
= bp
->b_vp
;
896 KKASSERT(vp
!= NULL
);
900 * B_PAGING flagged buffers cannot be reassigned because their vp
901 * is not fully linked in.
903 if (bp
->b_flags
& B_PAGING
)
904 panic("cannot reassign paging buffer");
906 lwkt_gettoken(&vlock
, &vp
->v_token
);
907 if (bp
->b_flags
& B_DELWRI
) {
909 * Move to the dirty list, add the vnode to the worklist
911 if (bp
->b_flags
& B_VNCLEAN
) {
912 buf_rb_tree_RB_REMOVE(&vp
->v_rbclean_tree
, bp
);
913 bp
->b_flags
&= ~B_VNCLEAN
;
915 if ((bp
->b_flags
& B_VNDIRTY
) == 0) {
916 if (buf_rb_tree_RB_INSERT(&vp
->v_rbdirty_tree
, bp
)) {
917 panic("reassignbuf: dup lblk vp %p bp %p",
920 bp
->b_flags
|= B_VNDIRTY
;
922 if ((vp
->v_flag
& VONWORKLST
) == 0) {
923 switch (vp
->v_type
) {
930 vp
->v_rdev
->si_mountpoint
!= NULL
) {
938 vn_syncer_add_to_worklist(vp
, delay
);
942 * Move to the clean list, remove the vnode from the worklist
943 * if no dirty blocks remain.
945 if (bp
->b_flags
& B_VNDIRTY
) {
946 buf_rb_tree_RB_REMOVE(&vp
->v_rbdirty_tree
, bp
);
947 bp
->b_flags
&= ~B_VNDIRTY
;
949 if ((bp
->b_flags
& B_VNCLEAN
) == 0) {
950 if (buf_rb_tree_RB_INSERT(&vp
->v_rbclean_tree
, bp
)) {
951 panic("reassignbuf: dup lblk vp %p bp %p",
954 bp
->b_flags
|= B_VNCLEAN
;
956 if ((vp
->v_flag
& VONWORKLST
) &&
957 RB_EMPTY(&vp
->v_rbdirty_tree
)) {
958 vp
->v_flag
&= ~VONWORKLST
;
959 LIST_REMOVE(vp
, v_synclist
);
962 lwkt_reltoken(&vlock
);
966 * Create a vnode for a block device.
967 * Used for mounting the root file system.
969 extern struct vop_ops
*devfs_vnode_dev_vops_p
;
971 bdevvp(cdev_t dev
, struct vnode
**vpp
)
981 error
= getspecialvnode(VT_NON
, NULL
, &devfs_vnode_dev_vops_p
,
992 v_associate_rdev(vp
, dev
);
993 vp
->v_umajor
= dev
->si_umajor
;
994 vp
->v_uminor
= dev
->si_uminor
;
1001 v_associate_rdev(struct vnode
*vp
, cdev_t dev
)
1007 if (dev_is_good(dev
) == 0)
1009 KKASSERT(vp
->v_rdev
== NULL
);
1010 vp
->v_rdev
= reference_dev(dev
);
1011 lwkt_gettoken(&ilock
, &spechash_token
);
1012 SLIST_INSERT_HEAD(&dev
->si_hlist
, vp
, v_cdevnext
);
1013 lwkt_reltoken(&ilock
);
1018 v_release_rdev(struct vnode
*vp
)
1023 if ((dev
= vp
->v_rdev
) != NULL
) {
1024 lwkt_gettoken(&ilock
, &spechash_token
);
1025 SLIST_REMOVE(&dev
->si_hlist
, vp
, vnode
, v_cdevnext
);
1028 lwkt_reltoken(&ilock
);
1033 * Add a vnode to the alias list hung off the cdev_t. We only associate
1034 * the device number with the vnode. The actual device is not associated
1035 * until the vnode is opened (usually in spec_open()), and will be
1036 * disassociated on last close.
1039 addaliasu(struct vnode
*nvp
, int x
, int y
)
1041 if (nvp
->v_type
!= VBLK
&& nvp
->v_type
!= VCHR
)
1042 panic("addaliasu on non-special vnode");
1048 * Simple call that a filesystem can make to try to get rid of a
1049 * vnode. It will fail if anyone is referencing the vnode (including
1052 * The filesystem can check whether its in-memory inode structure still
1053 * references the vp on return.
1056 vclean_unlocked(struct vnode
*vp
)
1059 if (sysref_isactive(&vp
->v_sysref
) == 0)
1065 * Disassociate a vnode from its underlying filesystem.
1067 * The vnode must be VX locked and referenced. In all normal situations
1068 * there are no active references. If vclean_vxlocked() is called while
1069 * there are active references, the vnode is being ripped out and we have
1070 * to call VOP_CLOSE() as appropriate before we can reclaim it.
1073 vclean_vxlocked(struct vnode
*vp
, int flags
)
1080 * If the vnode has already been reclaimed we have nothing to do.
1082 if (vp
->v_flag
& VRECLAIMED
)
1084 vp
->v_flag
|= VRECLAIMED
;
1087 * Scrap the vfs cache
1089 while (cache_inval_vp(vp
, 0) != 0) {
1090 kprintf("Warning: vnode %p clean/cache_resolution race detected\n", vp
);
1091 tsleep(vp
, 0, "vclninv", 2);
1095 * Check to see if the vnode is in use. If so we have to reference it
1096 * before we clean it out so that its count cannot fall to zero and
1097 * generate a race against ourselves to recycle it.
1099 active
= sysref_isactive(&vp
->v_sysref
);
1102 * Clean out any buffers associated with the vnode and destroy its
1103 * object, if it has one.
1105 vinvalbuf(vp
, V_SAVE
, 0, 0);
1108 * If purging an active vnode (typically during a forced unmount
1109 * or reboot), it must be closed and deactivated before being
1110 * reclaimed. This isn't really all that safe, but what can
1113 * Note that neither of these routines unlocks the vnode.
1115 if (active
&& (flags
& DOCLOSE
)) {
1116 while ((n
= vp
->v_opencount
) != 0) {
1117 if (vp
->v_writecount
)
1118 VOP_CLOSE(vp
, FWRITE
|FNONBLOCK
);
1120 VOP_CLOSE(vp
, FNONBLOCK
);
1121 if (vp
->v_opencount
== n
) {
1122 kprintf("Warning: unable to force-close"
1130 * If the vnode has not been deactivated, deactivated it. Deactivation
1131 * can create new buffers and VM pages so we have to call vinvalbuf()
1132 * again to make sure they all get flushed.
1134 * This can occur if a file with a link count of 0 needs to be
1137 if ((vp
->v_flag
& VINACTIVE
) == 0) {
1138 vp
->v_flag
|= VINACTIVE
;
1140 vinvalbuf(vp
, V_SAVE
, 0, 0);
1144 * If the vnode has an object, destroy it.
1146 if ((object
= vp
->v_object
) != NULL
) {
1147 if (object
->ref_count
== 0) {
1148 if ((object
->flags
& OBJ_DEAD
) == 0)
1149 vm_object_terminate(object
);
1151 vm_pager_deallocate(object
);
1153 vp
->v_flag
&= ~VOBJBUF
;
1155 KKASSERT((vp
->v_flag
& VOBJBUF
) == 0);
1158 * Reclaim the vnode.
1160 if (VOP_RECLAIM(vp
))
1161 panic("vclean: cannot reclaim");
1164 * Done with purge, notify sleepers of the grim news.
1166 vp
->v_ops
= &dead_vnode_vops_p
;
1171 * If we are destroying an active vnode, reactivate it now that
1172 * we have reassociated it with deadfs. This prevents the system
1173 * from crashing on the vnode due to it being unexpectedly marked
1174 * as inactive or reclaimed.
1176 if (active
&& (flags
& DOCLOSE
)) {
1177 vp
->v_flag
&= ~(VINACTIVE
|VRECLAIMED
);
1182 * Eliminate all activity associated with the requested vnode
1183 * and with all vnodes aliased to the requested vnode.
1185 * The vnode must be referenced but should not be locked.
1188 vrevoke(struct vnode
*vp
, struct ucred
*cred
)
1197 * If the vnode has a device association, scrap all vnodes associated
1198 * with the device. Don't let the device disappear on us while we
1199 * are scrapping the vnodes.
1201 * The passed vp will probably show up in the list, do not VX lock
1204 * Releasing the vnode's rdev here can mess up specfs's call to
1205 * device close, so don't do it. The vnode has been disassociated
1206 * and the device will be closed after the last ref on the related
1207 * fp goes away (if not still open by e.g. the kernel).
1209 if (vp
->v_type
!= VCHR
) {
1210 error
= fdrevoke(vp
, DTYPE_VNODE
, cred
);
1213 if ((dev
= vp
->v_rdev
) == NULL
) {
1217 lwkt_gettoken(&ilock
, &spechash_token
);
1219 vqn
= SLIST_FIRST(&dev
->si_hlist
);
1222 while ((vq
= vqn
) != NULL
) {
1223 vqn
= SLIST_NEXT(vqn
, v_cdevnext
);
1226 fdrevoke(vq
, DTYPE_VNODE
, cred
);
1227 /*v_release_rdev(vq);*/
1230 lwkt_reltoken(&ilock
);
1237 * This is called when the object underlying a vnode is being destroyed,
1238 * such as in a remove(). Try to recycle the vnode immediately if the
1239 * only active reference is our reference.
1241 * Directory vnodes in the namecache with children cannot be immediately
1242 * recycled because numerous VOP_N*() ops require them to be stable.
1245 vrecycle(struct vnode
*vp
)
1247 if (vp
->v_sysref
.refcnt
<= 1) {
1248 if (cache_inval_vp_nonblock(vp
))
1257 * Return the maximum I/O size allowed for strategy calls on VP.
1259 * If vp is VCHR or VBLK we dive the device, otherwise we use
1260 * the vp's mount info.
1263 vmaxiosize(struct vnode
*vp
)
1265 if (vp
->v_type
== VBLK
|| vp
->v_type
== VCHR
) {
1266 return(vp
->v_rdev
->si_iosize_max
);
1268 return(vp
->v_mount
->mnt_iosize_max
);
1273 * Eliminate all activity associated with a vnode in preparation for reuse.
1275 * The vnode must be VX locked and refd and will remain VX locked and refd
1276 * on return. This routine may be called with the vnode in any state, as
1277 * long as it is VX locked. The vnode will be cleaned out and marked
1278 * VRECLAIMED but will not actually be reused until all existing refs and
1281 * NOTE: This routine may be called on a vnode which has not yet been
1282 * already been deactivated (VOP_INACTIVE), or on a vnode which has
1283 * already been reclaimed.
1285 * This routine is not responsible for placing us back on the freelist.
1286 * Instead, it happens automatically when the caller releases the VX lock
1287 * (assuming there aren't any other references).
1291 vgone_vxlocked(struct vnode
*vp
)
1294 * assert that the VX lock is held. This is an absolute requirement
1295 * now for vgone_vxlocked() to be called.
1297 KKASSERT(vp
->v_lock
.lk_exclusivecount
== 1);
1300 * Clean out the filesystem specific data and set the VRECLAIMED
1301 * bit. Also deactivate the vnode if necessary.
1303 vclean_vxlocked(vp
, DOCLOSE
);
1306 * Delete from old mount point vnode list, if on one.
1308 if (vp
->v_mount
!= NULL
)
1309 insmntque(vp
, NULL
);
1312 * If special device, remove it from special device alias list
1313 * if it is on one. This should normally only occur if a vnode is
1314 * being revoked as the device should otherwise have been released
1317 if ((vp
->v_type
== VBLK
|| vp
->v_type
== VCHR
) && vp
->v_rdev
!= NULL
) {
1328 * Lookup a vnode by device number.
1330 * Returns non-zero and *vpp set to a vref'd vnode on success.
1331 * Returns zero on failure.
1334 vfinddev(cdev_t dev
, enum vtype type
, struct vnode
**vpp
)
1339 lwkt_gettoken(&ilock
, &spechash_token
);
1340 SLIST_FOREACH(vp
, &dev
->si_hlist
, v_cdevnext
) {
1341 if (type
== vp
->v_type
) {
1344 lwkt_reltoken(&ilock
);
1348 lwkt_reltoken(&ilock
);
1353 * Calculate the total number of references to a special device. This
1354 * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1355 * an overloaded field. Since udev2dev can now return NULL, we have
1356 * to check for a NULL v_rdev.
1359 count_dev(cdev_t dev
)
1365 if (SLIST_FIRST(&dev
->si_hlist
)) {
1366 lwkt_gettoken(&ilock
, &spechash_token
);
1367 SLIST_FOREACH(vp
, &dev
->si_hlist
, v_cdevnext
) {
1368 count
+= vp
->v_opencount
;
1370 lwkt_reltoken(&ilock
);
1376 vcount(struct vnode
*vp
)
1378 if (vp
->v_rdev
== NULL
)
1380 return(count_dev(vp
->v_rdev
));
1384 * Initialize VMIO for a vnode. This routine MUST be called before a
1385 * VFS can issue buffer cache ops on a vnode. It is typically called
1386 * when a vnode is initialized from its inode.
1389 vinitvmio(struct vnode
*vp
, off_t filesize
)
1395 if ((object
= vp
->v_object
) == NULL
) {
1396 object
= vnode_pager_alloc(vp
, filesize
, 0, 0);
1398 * Dereference the reference we just created. This assumes
1399 * that the object is associated with the vp.
1401 object
->ref_count
--;
1404 if (object
->flags
& OBJ_DEAD
) {
1406 vm_object_dead_sleep(object
, "vodead");
1407 vn_lock(vp
, LK_EXCLUSIVE
| LK_RETRY
);
1411 KASSERT(vp
->v_object
!= NULL
, ("vinitvmio: NULL object"));
1412 vp
->v_flag
|= VOBJBUF
;
1418 * Print out a description of a vnode.
1420 static char *typename
[] =
1421 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1424 vprint(char *label
, struct vnode
*vp
)
1429 kprintf("%s: %p: ", label
, (void *)vp
);
1431 kprintf("%p: ", (void *)vp
);
1432 kprintf("type %s, sysrefs %d, writecount %d, holdcnt %d,",
1433 typename
[vp
->v_type
],
1434 vp
->v_sysref
.refcnt
, vp
->v_writecount
, vp
->v_auxrefs
);
1436 if (vp
->v_flag
& VROOT
)
1437 strcat(buf
, "|VROOT");
1438 if (vp
->v_flag
& VPFSROOT
)
1439 strcat(buf
, "|VPFSROOT");
1440 if (vp
->v_flag
& VTEXT
)
1441 strcat(buf
, "|VTEXT");
1442 if (vp
->v_flag
& VSYSTEM
)
1443 strcat(buf
, "|VSYSTEM");
1444 if (vp
->v_flag
& VFREE
)
1445 strcat(buf
, "|VFREE");
1446 if (vp
->v_flag
& VOBJBUF
)
1447 strcat(buf
, "|VOBJBUF");
1449 kprintf(" flags (%s)", &buf
[1]);
1450 if (vp
->v_data
== NULL
) {
1459 #include <ddb/ddb.h>
1461 static int db_show_locked_vnodes(struct mount
*mp
, void *data
);
1464 * List all of the locked vnodes in the system.
1465 * Called when debugging the kernel.
1467 DB_SHOW_COMMAND(lockedvnodes
, lockedvnodes
)
1469 kprintf("Locked vnodes\n");
1470 mountlist_scan(db_show_locked_vnodes
, NULL
,
1471 MNTSCAN_FORWARD
|MNTSCAN_NOBUSY
);
1475 db_show_locked_vnodes(struct mount
*mp
, void *data __unused
)
1479 TAILQ_FOREACH(vp
, &mp
->mnt_nvnodelist
, v_nmntvnodes
) {
1480 if (vn_islocked(vp
))
1488 * Top level filesystem related information gathering.
1490 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS
);
1493 vfs_sysctl(SYSCTL_HANDLER_ARGS
)
1495 int *name
= (int *)arg1
- 1; /* XXX */
1496 u_int namelen
= arg2
+ 1; /* XXX */
1497 struct vfsconf
*vfsp
;
1500 #if 1 || defined(COMPAT_PRELITE2)
1501 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1503 return (sysctl_ovfs_conf(oidp
, arg1
, arg2
, req
));
1507 /* all sysctl names at this level are at least name and field */
1509 return (ENOTDIR
); /* overloaded */
1510 if (name
[0] != VFS_GENERIC
) {
1511 vfsp
= vfsconf_find_by_typenum(name
[0]);
1513 return (EOPNOTSUPP
);
1514 return ((*vfsp
->vfc_vfsops
->vfs_sysctl
)(&name
[1], namelen
- 1,
1515 oldp
, oldlenp
, newp
, newlen
, p
));
1519 case VFS_MAXTYPENUM
:
1522 maxtypenum
= vfsconf_get_maxtypenum();
1523 return (SYSCTL_OUT(req
, &maxtypenum
, sizeof(maxtypenum
)));
1526 return (ENOTDIR
); /* overloaded */
1527 vfsp
= vfsconf_find_by_typenum(name
[2]);
1529 return (EOPNOTSUPP
);
1530 return (SYSCTL_OUT(req
, vfsp
, sizeof *vfsp
));
1532 return (EOPNOTSUPP
);
1535 SYSCTL_NODE(_vfs
, VFS_GENERIC
, generic
, CTLFLAG_RD
, vfs_sysctl
,
1536 "Generic filesystem");
1538 #if 1 || defined(COMPAT_PRELITE2)
1541 sysctl_ovfs_conf_iter(struct vfsconf
*vfsp
, void *data
)
1544 struct ovfsconf ovfs
;
1545 struct sysctl_req
*req
= (struct sysctl_req
*) data
;
1547 bzero(&ovfs
, sizeof(ovfs
));
1548 ovfs
.vfc_vfsops
= vfsp
->vfc_vfsops
; /* XXX used as flag */
1549 strcpy(ovfs
.vfc_name
, vfsp
->vfc_name
);
1550 ovfs
.vfc_index
= vfsp
->vfc_typenum
;
1551 ovfs
.vfc_refcount
= vfsp
->vfc_refcount
;
1552 ovfs
.vfc_flags
= vfsp
->vfc_flags
;
1553 error
= SYSCTL_OUT(req
, &ovfs
, sizeof ovfs
);
1555 return error
; /* abort iteration with error code */
1557 return 0; /* continue iterating with next element */
1561 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS
)
1563 return vfsconf_each(sysctl_ovfs_conf_iter
, (void*)req
);
1566 #endif /* 1 || COMPAT_PRELITE2 */
1569 * Check to see if a filesystem is mounted on a block device.
1572 vfs_mountedon(struct vnode
*vp
)
1576 if ((dev
= vp
->v_rdev
) == NULL
) {
1577 /* if (vp->v_type != VBLK)
1578 dev = get_dev(vp->v_uminor, vp->v_umajor); */
1580 if (dev
!= NULL
&& dev
->si_mountpoint
)
1586 * Unmount all filesystems. The list is traversed in reverse order
1587 * of mounting to avoid dependencies.
1590 static int vfs_umountall_callback(struct mount
*mp
, void *data
);
1593 vfs_unmountall(void)
1598 count
= mountlist_scan(vfs_umountall_callback
,
1599 NULL
, MNTSCAN_REVERSE
|MNTSCAN_NOBUSY
);
1605 vfs_umountall_callback(struct mount
*mp
, void *data
)
1609 error
= dounmount(mp
, MNT_FORCE
);
1611 mountlist_remove(mp
);
1612 kprintf("unmount of filesystem mounted from %s failed (",
1613 mp
->mnt_stat
.f_mntfromname
);
1617 kprintf("%d)\n", error
);
1623 * Checks the mount flags for parameter mp and put the names comma-separated
1624 * into a string buffer buf with a size limit specified by len.
1626 * It returns the number of bytes written into buf, and (*errorp) will be
1627 * set to 0, EINVAL (if passed length is 0), or ENOSPC (supplied buffer was
1628 * not large enough). The buffer will be 0-terminated if len was not 0.
1631 vfs_flagstostr(int flags
, const struct mountctl_opt
*optp
,
1632 char *buf
, size_t len
, int *errorp
)
1634 static const struct mountctl_opt optnames
[] = {
1635 { MNT_ASYNC
, "asynchronous" },
1636 { MNT_EXPORTED
, "NFS exported" },
1637 { MNT_LOCAL
, "local" },
1638 { MNT_NOATIME
, "noatime" },
1639 { MNT_NODEV
, "nodev" },
1640 { MNT_NOEXEC
, "noexec" },
1641 { MNT_NOSUID
, "nosuid" },
1642 { MNT_NOSYMFOLLOW
, "nosymfollow" },
1643 { MNT_QUOTA
, "with-quotas" },
1644 { MNT_RDONLY
, "read-only" },
1645 { MNT_SYNCHRONOUS
, "synchronous" },
1646 { MNT_UNION
, "union" },
1647 { MNT_NOCLUSTERR
, "noclusterr" },
1648 { MNT_NOCLUSTERW
, "noclusterw" },
1649 { MNT_SUIDDIR
, "suiddir" },
1650 { MNT_SOFTDEP
, "soft-updates" },
1651 { MNT_IGNORE
, "ignore" },
1661 bleft
= len
- 1; /* leave room for trailing \0 */
1664 * Checks the size of the string. If it contains
1665 * any data, then we will append the new flags to
1668 actsize
= strlen(buf
);
1672 /* Default flags if no flags passed */
1676 if (bleft
< 0) { /* degenerate case, 0-length buffer */
1681 for (; flags
&& optp
->o_opt
; ++optp
) {
1682 if ((flags
& optp
->o_opt
) == 0)
1684 optlen
= strlen(optp
->o_name
);
1685 if (bwritten
|| actsize
> 0) {
1690 buf
[bwritten
++] = ',';
1691 buf
[bwritten
++] = ' ';
1694 if (bleft
< optlen
) {
1698 bcopy(optp
->o_name
, buf
+ bwritten
, optlen
);
1701 flags
&= ~optp
->o_opt
;
1705 * Space already reserved for trailing \0
1712 * Build hash lists of net addresses and hang them off the mount point.
1713 * Called by ufs_mount() to set up the lists of export addresses.
1716 vfs_hang_addrlist(struct mount
*mp
, struct netexport
*nep
,
1717 const struct export_args
*argp
)
1720 struct radix_node_head
*rnh
;
1722 struct radix_node
*rn
;
1723 struct sockaddr
*saddr
, *smask
= 0;
1727 if (argp
->ex_addrlen
== 0) {
1728 if (mp
->mnt_flag
& MNT_DEFEXPORTED
)
1730 np
= &nep
->ne_defexported
;
1731 np
->netc_exflags
= argp
->ex_flags
;
1732 np
->netc_anon
= argp
->ex_anon
;
1733 np
->netc_anon
.cr_ref
= 1;
1734 mp
->mnt_flag
|= MNT_DEFEXPORTED
;
1738 if (argp
->ex_addrlen
< 0 || argp
->ex_addrlen
> MLEN
)
1740 if (argp
->ex_masklen
< 0 || argp
->ex_masklen
> MLEN
)
1743 i
= sizeof(struct netcred
) + argp
->ex_addrlen
+ argp
->ex_masklen
;
1744 np
= (struct netcred
*) kmalloc(i
, M_NETADDR
, M_WAITOK
| M_ZERO
);
1745 saddr
= (struct sockaddr
*) (np
+ 1);
1746 if ((error
= copyin(argp
->ex_addr
, (caddr_t
) saddr
, argp
->ex_addrlen
)))
1748 if (saddr
->sa_len
> argp
->ex_addrlen
)
1749 saddr
->sa_len
= argp
->ex_addrlen
;
1750 if (argp
->ex_masklen
) {
1751 smask
= (struct sockaddr
*)((caddr_t
)saddr
+ argp
->ex_addrlen
);
1752 error
= copyin(argp
->ex_mask
, (caddr_t
)smask
, argp
->ex_masklen
);
1755 if (smask
->sa_len
> argp
->ex_masklen
)
1756 smask
->sa_len
= argp
->ex_masklen
;
1758 i
= saddr
->sa_family
;
1759 if ((rnh
= nep
->ne_rtable
[i
]) == 0) {
1761 * Seems silly to initialize every AF when most are not used,
1762 * do so on demand here
1764 SLIST_FOREACH(dom
, &domains
, dom_next
)
1765 if (dom
->dom_family
== i
&& dom
->dom_rtattach
) {
1766 dom
->dom_rtattach((void **) &nep
->ne_rtable
[i
],
1770 if ((rnh
= nep
->ne_rtable
[i
]) == 0) {
1775 rn
= (*rnh
->rnh_addaddr
) ((char *) saddr
, (char *) smask
, rnh
,
1777 if (rn
== 0 || np
!= (struct netcred
*) rn
) { /* already exists */
1781 np
->netc_exflags
= argp
->ex_flags
;
1782 np
->netc_anon
= argp
->ex_anon
;
1783 np
->netc_anon
.cr_ref
= 1;
1786 kfree(np
, M_NETADDR
);
1792 vfs_free_netcred(struct radix_node
*rn
, void *w
)
1794 struct radix_node_head
*rnh
= (struct radix_node_head
*) w
;
1796 (*rnh
->rnh_deladdr
) (rn
->rn_key
, rn
->rn_mask
, rnh
);
1797 kfree((caddr_t
) rn
, M_NETADDR
);
1802 * Free the net address hash lists that are hanging off the mount points.
1805 vfs_free_addrlist(struct netexport
*nep
)
1808 struct radix_node_head
*rnh
;
1810 for (i
= 0; i
<= AF_MAX
; i
++)
1811 if ((rnh
= nep
->ne_rtable
[i
])) {
1812 (*rnh
->rnh_walktree
) (rnh
, vfs_free_netcred
,
1814 kfree((caddr_t
) rnh
, M_RTABLE
);
1815 nep
->ne_rtable
[i
] = 0;
1820 vfs_export(struct mount
*mp
, struct netexport
*nep
,
1821 const struct export_args
*argp
)
1825 if (argp
->ex_flags
& MNT_DELEXPORT
) {
1826 if (mp
->mnt_flag
& MNT_EXPUBLIC
) {
1827 vfs_setpublicfs(NULL
, NULL
, NULL
);
1828 mp
->mnt_flag
&= ~MNT_EXPUBLIC
;
1830 vfs_free_addrlist(nep
);
1831 mp
->mnt_flag
&= ~(MNT_EXPORTED
| MNT_DEFEXPORTED
);
1833 if (argp
->ex_flags
& MNT_EXPORTED
) {
1834 if (argp
->ex_flags
& MNT_EXPUBLIC
) {
1835 if ((error
= vfs_setpublicfs(mp
, nep
, argp
)) != 0)
1837 mp
->mnt_flag
|= MNT_EXPUBLIC
;
1839 if ((error
= vfs_hang_addrlist(mp
, nep
, argp
)))
1841 mp
->mnt_flag
|= MNT_EXPORTED
;
1848 * Set the publicly exported filesystem (WebNFS). Currently, only
1849 * one public filesystem is possible in the spec (RFC 2054 and 2055)
1852 vfs_setpublicfs(struct mount
*mp
, struct netexport
*nep
,
1853 const struct export_args
*argp
)
1860 * mp == NULL -> invalidate the current info, the FS is
1861 * no longer exported. May be called from either vfs_export
1862 * or unmount, so check if it hasn't already been done.
1865 if (nfs_pub
.np_valid
) {
1866 nfs_pub
.np_valid
= 0;
1867 if (nfs_pub
.np_index
!= NULL
) {
1868 FREE(nfs_pub
.np_index
, M_TEMP
);
1869 nfs_pub
.np_index
= NULL
;
1876 * Only one allowed at a time.
1878 if (nfs_pub
.np_valid
!= 0 && mp
!= nfs_pub
.np_mount
)
1882 * Get real filehandle for root of exported FS.
1884 bzero((caddr_t
)&nfs_pub
.np_handle
, sizeof(nfs_pub
.np_handle
));
1885 nfs_pub
.np_handle
.fh_fsid
= mp
->mnt_stat
.f_fsid
;
1887 if ((error
= VFS_ROOT(mp
, &rvp
)))
1890 if ((error
= VFS_VPTOFH(rvp
, &nfs_pub
.np_handle
.fh_fid
)))
1896 * If an indexfile was specified, pull it in.
1898 if (argp
->ex_indexfile
!= NULL
) {
1901 error
= vn_get_namelen(rvp
, &namelen
);
1904 MALLOC(nfs_pub
.np_index
, char *, namelen
, M_TEMP
,
1906 error
= copyinstr(argp
->ex_indexfile
, nfs_pub
.np_index
,
1910 * Check for illegal filenames.
1912 for (cp
= nfs_pub
.np_index
; *cp
; cp
++) {
1920 FREE(nfs_pub
.np_index
, M_TEMP
);
1925 nfs_pub
.np_mount
= mp
;
1926 nfs_pub
.np_valid
= 1;
1931 vfs_export_lookup(struct mount
*mp
, struct netexport
*nep
,
1932 struct sockaddr
*nam
)
1935 struct radix_node_head
*rnh
;
1936 struct sockaddr
*saddr
;
1939 if (mp
->mnt_flag
& MNT_EXPORTED
) {
1941 * Lookup in the export list first.
1945 rnh
= nep
->ne_rtable
[saddr
->sa_family
];
1947 np
= (struct netcred
*)
1948 (*rnh
->rnh_matchaddr
)((char *)saddr
,
1950 if (np
&& np
->netc_rnodes
->rn_flags
& RNF_ROOT
)
1955 * If no address match, use the default if it exists.
1957 if (np
== NULL
&& mp
->mnt_flag
& MNT_DEFEXPORTED
)
1958 np
= &nep
->ne_defexported
;
1964 * perform msync on all vnodes under a mount point. The mount point must
1965 * be locked. This code is also responsible for lazy-freeing unreferenced
1966 * vnodes whos VM objects no longer contain pages.
1968 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
1970 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked,
1971 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it
1972 * way up in this high level function.
1974 static int vfs_msync_scan1(struct mount
*mp
, struct vnode
*vp
, void *data
);
1975 static int vfs_msync_scan2(struct mount
*mp
, struct vnode
*vp
, void *data
);
1978 vfs_msync(struct mount
*mp
, int flags
)
1982 vmsc_flags
= VMSC_GETVP
;
1983 if (flags
!= MNT_WAIT
)
1984 vmsc_flags
|= VMSC_NOWAIT
;
1985 vmntvnodescan(mp
, vmsc_flags
, vfs_msync_scan1
, vfs_msync_scan2
,
1986 (void *)(intptr_t)flags
);
1990 * scan1 is a fast pre-check. There could be hundreds of thousands of
1991 * vnodes, we cannot afford to do anything heavy weight until we have a
1992 * fairly good indication that there is work to do.
1996 vfs_msync_scan1(struct mount
*mp
, struct vnode
*vp
, void *data
)
1998 int flags
= (int)(intptr_t)data
;
2000 if ((vp
->v_flag
& VRECLAIMED
) == 0) {
2001 if (vshouldmsync(vp
))
2002 return(0); /* call scan2 */
2003 if ((mp
->mnt_flag
& MNT_RDONLY
) == 0 &&
2004 (vp
->v_flag
& VOBJDIRTY
) &&
2005 (flags
== MNT_WAIT
|| vn_islocked(vp
) == 0)) {
2006 return(0); /* call scan2 */
2011 * do not call scan2, continue the loop
2017 * This callback is handed a locked vnode.
2021 vfs_msync_scan2(struct mount
*mp
, struct vnode
*vp
, void *data
)
2024 int flags
= (int)(intptr_t)data
;
2026 if (vp
->v_flag
& VRECLAIMED
)
2029 if ((mp
->mnt_flag
& MNT_RDONLY
) == 0 && (vp
->v_flag
& VOBJDIRTY
)) {
2030 if ((obj
= vp
->v_object
) != NULL
) {
2031 vm_object_page_clean(obj
, 0, 0,
2032 flags
== MNT_WAIT
? OBJPC_SYNC
: OBJPC_NOSYNC
);
2039 * Record a process's interest in events which might happen to
2040 * a vnode. Because poll uses the historic select-style interface
2041 * internally, this routine serves as both the ``check for any
2042 * pending events'' and the ``record my interest in future events''
2043 * functions. (These are done together, while the lock is held,
2044 * to avoid race conditions.)
2047 vn_pollrecord(struct vnode
*vp
, int events
)
2051 KKASSERT(curthread
->td_proc
!= NULL
);
2053 lwkt_gettoken(&vlock
, &vp
->v_token
);
2054 if (vp
->v_pollinfo
.vpi_revents
& events
) {
2056 * This leaves events we are not interested
2057 * in available for the other process which
2058 * which presumably had requested them
2059 * (otherwise they would never have been
2062 events
&= vp
->v_pollinfo
.vpi_revents
;
2063 vp
->v_pollinfo
.vpi_revents
&= ~events
;
2065 lwkt_reltoken(&vlock
);
2068 vp
->v_pollinfo
.vpi_events
|= events
;
2069 selrecord(curthread
, &vp
->v_pollinfo
.vpi_selinfo
);
2070 lwkt_reltoken(&vlock
);
2075 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
2076 * it is possible for us to miss an event due to race conditions, but
2077 * that condition is expected to be rare, so for the moment it is the
2078 * preferred interface.
2081 vn_pollevent(struct vnode
*vp
, int events
)
2085 lwkt_gettoken(&vlock
, &vp
->v_token
);
2086 if (vp
->v_pollinfo
.vpi_events
& events
) {
2088 * We clear vpi_events so that we don't
2089 * call selwakeup() twice if two events are
2090 * posted before the polling process(es) is
2091 * awakened. This also ensures that we take at
2092 * most one selwakeup() if the polling process
2093 * is no longer interested. However, it does
2094 * mean that only one event can be noticed at
2095 * a time. (Perhaps we should only clear those
2096 * event bits which we note?) XXX
2098 vp
->v_pollinfo
.vpi_events
= 0; /* &= ~events ??? */
2099 vp
->v_pollinfo
.vpi_revents
|= events
;
2100 selwakeup(&vp
->v_pollinfo
.vpi_selinfo
);
2102 lwkt_reltoken(&vlock
);
2106 * Wake up anyone polling on vp because it is being revoked.
2107 * This depends on dead_poll() returning POLLHUP for correct
2111 vn_pollgone(struct vnode
*vp
)
2115 lwkt_gettoken(&vlock
, &vp
->v_token
);
2116 if (vp
->v_pollinfo
.vpi_events
) {
2117 vp
->v_pollinfo
.vpi_events
= 0;
2118 selwakeup(&vp
->v_pollinfo
.vpi_selinfo
);
2120 lwkt_reltoken(&vlock
);
2124 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened
2125 * (or v_rdev might be NULL).
2128 vn_todev(struct vnode
*vp
)
2130 if (vp
->v_type
!= VBLK
&& vp
->v_type
!= VCHR
)
2132 KKASSERT(vp
->v_rdev
!= NULL
);
2133 return (vp
->v_rdev
);
2137 * Check if vnode represents a disk device. The vnode does not need to be
2143 vn_isdisk(struct vnode
*vp
, int *errp
)
2147 if (vp
->v_type
!= VCHR
) {
2160 if (dev_is_good(dev
) == 0) {
2165 if ((dev_dflags(dev
) & D_DISK
) == 0) {
2176 vn_get_namelen(struct vnode
*vp
, int *namelen
)
2179 register_t retval
[2];
2181 error
= VOP_PATHCONF(vp
, _PC_NAME_MAX
, retval
);
2184 *namelen
= (int)retval
[0];
2189 vop_write_dirent(int *error
, struct uio
*uio
, ino_t d_ino
, uint8_t d_type
,
2190 uint16_t d_namlen
, const char *d_name
)
2195 len
= _DIRENT_RECLEN(d_namlen
);
2196 if (len
> uio
->uio_resid
)
2199 dp
= kmalloc(len
, M_TEMP
, M_WAITOK
| M_ZERO
);
2202 dp
->d_namlen
= d_namlen
;
2203 dp
->d_type
= d_type
;
2204 bcopy(d_name
, dp
->d_name
, d_namlen
);
2206 *error
= uiomove((caddr_t
)dp
, len
, uio
);
2214 vn_mark_atime(struct vnode
*vp
, struct thread
*td
)
2216 struct proc
*p
= td
->td_proc
;
2217 struct ucred
*cred
= p
? p
->p_ucred
: proc0
.p_ucred
;
2219 if ((vp
->v_mount
->mnt_flag
& (MNT_NOATIME
| MNT_RDONLY
)) == 0) {
2220 VOP_MARKATIME(vp
, cred
);