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 1.58 2005/08/09 16:53:34 joerg Exp $
44 * External virtual filesystem routines
48 #include <sys/param.h>
49 #include <sys/systm.h>
52 #include <sys/dirent.h>
53 #include <sys/domain.h>
54 #include <sys/eventhandler.h>
55 #include <sys/fcntl.h>
56 #include <sys/kernel.h>
57 #include <sys/kthread.h>
58 #include <sys/malloc.h>
60 #include <sys/mount.h>
62 #include <sys/reboot.h>
63 #include <sys/socket.h>
65 #include <sys/sysctl.h>
66 #include <sys/syslog.h>
67 #include <sys/unistd.h>
68 #include <sys/vmmeter.h>
69 #include <sys/vnode.h>
71 #include <machine/limits.h>
74 #include <vm/vm_object.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_kern.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_pager.h>
81 #include <vm/vnode_pager.h>
82 #include <vm/vm_zone.h>
85 #include <sys/thread2.h>
87 static MALLOC_DEFINE(M_NETADDR
, "Export Host", "Export host address structure");
90 SYSCTL_INT(_debug
, OID_AUTO
, numvnodes
, CTLFLAG_RD
, &numvnodes
, 0, "");
92 SYSCTL_INT(_vfs
, OID_AUTO
, fastdev
, CTLFLAG_RW
, &vfs_fastdev
, 0, "");
94 enum vtype iftovt_tab
[16] = {
95 VNON
, VFIFO
, VCHR
, VNON
, VDIR
, VNON
, VBLK
, VNON
,
96 VREG
, VNON
, VLNK
, VNON
, VSOCK
, VNON
, VNON
, VBAD
,
99 0, S_IFREG
, S_IFDIR
, S_IFBLK
, S_IFCHR
, S_IFLNK
,
100 S_IFSOCK
, S_IFIFO
, S_IFMT
,
103 static int reassignbufcalls
;
104 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufcalls
, CTLFLAG_RW
,
105 &reassignbufcalls
, 0, "");
106 static int reassignbufloops
;
107 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufloops
, CTLFLAG_RW
,
108 &reassignbufloops
, 0, "");
109 static int reassignbufsortgood
;
110 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufsortgood
, CTLFLAG_RW
,
111 &reassignbufsortgood
, 0, "");
112 static int reassignbufsortbad
;
113 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufsortbad
, CTLFLAG_RW
,
114 &reassignbufsortbad
, 0, "");
115 static int reassignbufmethod
= 1;
116 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufmethod
, CTLFLAG_RW
,
117 &reassignbufmethod
, 0, "");
119 int nfs_mount_type
= -1;
120 static struct lwkt_token spechash_token
;
121 struct nfs_public nfs_pub
; /* publicly exported FS */
124 SYSCTL_INT(_kern
, KERN_MAXVNODES
, maxvnodes
, CTLFLAG_RW
,
125 &desiredvnodes
, 0, "Maximum number of vnodes");
127 static void vfs_free_addrlist (struct netexport
*nep
);
128 static int vfs_free_netcred (struct radix_node
*rn
, void *w
);
129 static int vfs_hang_addrlist (struct mount
*mp
, struct netexport
*nep
,
130 struct export_args
*argp
);
132 extern int dev_ref_debug
;
133 extern struct vnodeopv_entry_desc spec_vnodeop_entries
[];
136 * Red black tree functions
138 static int rb_buf_compare(struct buf
*b1
, struct buf
*b2
);
139 RB_GENERATE(buf_rb_tree
, buf
, b_rbnode
, rb_buf_compare
);
142 rb_buf_compare(struct buf
*b1
, struct buf
*b2
)
144 if (b1
->b_lblkno
< b2
->b_lblkno
)
146 if (b1
->b_lblkno
> b2
->b_lblkno
)
152 * Return 0 if the vnode is already on the free list or cannot be placed
153 * on the free list. Return 1 if the vnode can be placed on the free list.
156 vshouldfree(struct vnode
*vp
, int usecount
)
158 if (vp
->v_flag
& VFREE
)
159 return (0); /* already free */
160 if (vp
->v_holdcnt
!= 0 || vp
->v_usecount
!= usecount
)
161 return (0); /* other holderse */
163 (vp
->v_object
->ref_count
|| vp
->v_object
->resident_page_count
)) {
170 * Initialize the vnode management data structures.
172 * Called from vfsinit()
178 * Desired vnodes is a result of the physical page count
179 * and the size of kernel's heap. It scales in proportion
180 * to the amount of available physical memory. This can
181 * cause trouble on 64-bit and large memory platforms.
183 /* desiredvnodes = maxproc + vmstats.v_page_count / 4; */
185 min(maxproc
+ vmstats
.v_page_count
/4,
186 2 * (VM_MAX_KERNEL_ADDRESS
- VM_MIN_KERNEL_ADDRESS
) /
187 (5 * (sizeof(struct vm_object
) + sizeof(struct vnode
))));
189 lwkt_token_init(&spechash_token
);
193 * Knob to control the precision of file timestamps:
195 * 0 = seconds only; nanoseconds zeroed.
196 * 1 = seconds and nanoseconds, accurate within 1/HZ.
197 * 2 = seconds and nanoseconds, truncated to microseconds.
198 * >=3 = seconds and nanoseconds, maximum precision.
200 enum { TSP_SEC
, TSP_HZ
, TSP_USEC
, TSP_NSEC
};
202 static int timestamp_precision
= TSP_SEC
;
203 SYSCTL_INT(_vfs
, OID_AUTO
, timestamp_precision
, CTLFLAG_RW
,
204 ×tamp_precision
, 0, "");
207 * Get a current timestamp.
210 vfs_timestamp(struct timespec
*tsp
)
214 switch (timestamp_precision
) {
216 tsp
->tv_sec
= time_second
;
224 TIMEVAL_TO_TIMESPEC(&tv
, tsp
);
234 * Set vnode attributes to VNOVAL
237 vattr_null(struct vattr
*vap
)
240 vap
->va_size
= VNOVAL
;
241 vap
->va_bytes
= VNOVAL
;
242 vap
->va_mode
= VNOVAL
;
243 vap
->va_nlink
= VNOVAL
;
244 vap
->va_uid
= VNOVAL
;
245 vap
->va_gid
= VNOVAL
;
246 vap
->va_fsid
= VNOVAL
;
247 vap
->va_fileid
= VNOVAL
;
248 vap
->va_blocksize
= VNOVAL
;
249 vap
->va_rdev
= VNOVAL
;
250 vap
->va_atime
.tv_sec
= VNOVAL
;
251 vap
->va_atime
.tv_nsec
= VNOVAL
;
252 vap
->va_mtime
.tv_sec
= VNOVAL
;
253 vap
->va_mtime
.tv_nsec
= VNOVAL
;
254 vap
->va_ctime
.tv_sec
= VNOVAL
;
255 vap
->va_ctime
.tv_nsec
= VNOVAL
;
256 vap
->va_flags
= VNOVAL
;
257 vap
->va_gen
= VNOVAL
;
262 * Update outstanding I/O count and do wakeup if requested.
265 vwakeup(struct buf
*bp
)
269 if ((vp
= bp
->b_vp
)) {
271 if (vp
->v_numoutput
< 0)
272 panic("vwakeup: neg numoutput");
273 if ((vp
->v_numoutput
== 0) && (vp
->v_flag
& VBWAIT
)) {
274 vp
->v_flag
&= ~VBWAIT
;
275 wakeup((caddr_t
) &vp
->v_numoutput
);
281 * Flush out and invalidate all buffers associated with a vnode.
285 static int vinvalbuf_bp(struct buf
*bp
, void *data
);
287 struct vinvalbuf_bp_info
{
295 vinvalbuf(struct vnode
*vp
, int flags
, struct thread
*td
,
296 int slpflag
, int slptimeo
)
298 struct vinvalbuf_bp_info info
;
303 * If we are being asked to save, call fsync to ensure that the inode
306 if (flags
& V_SAVE
) {
308 while (vp
->v_numoutput
) {
309 vp
->v_flag
|= VBWAIT
;
310 error
= tsleep((caddr_t
)&vp
->v_numoutput
,
311 slpflag
, "vinvlbuf", slptimeo
);
317 if (!RB_EMPTY(&vp
->v_rbdirty_tree
)) {
319 if ((error
= VOP_FSYNC(vp
, MNT_WAIT
, td
)) != 0)
322 if (vp
->v_numoutput
> 0 ||
323 !RB_EMPTY(&vp
->v_rbdirty_tree
))
324 panic("vinvalbuf: dirty bufs");
329 info
.slptimeo
= slptimeo
;
330 info
.slpflag
= slpflag
;
335 * Flush the buffer cache until nothing is left.
337 while (!RB_EMPTY(&vp
->v_rbclean_tree
) ||
338 !RB_EMPTY(&vp
->v_rbdirty_tree
)) {
339 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbclean_tree
, NULL
,
340 vinvalbuf_bp
, &info
);
342 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, NULL
,
343 vinvalbuf_bp
, &info
);
348 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
349 * have write I/O in-progress but if there is a VM object then the
350 * VM object can also have read-I/O in-progress.
353 while (vp
->v_numoutput
> 0) {
354 vp
->v_flag
|= VBWAIT
;
355 tsleep(&vp
->v_numoutput
, 0, "vnvlbv", 0);
357 if (VOP_GETVOBJECT(vp
, &object
) == 0) {
358 while (object
->paging_in_progress
)
359 vm_object_pip_sleep(object
, "vnvlbx");
361 } while (vp
->v_numoutput
> 0);
366 * Destroy the copy in the VM cache, too.
368 if (VOP_GETVOBJECT(vp
, &object
) == 0) {
369 vm_object_page_remove(object
, 0, 0,
370 (flags
& V_SAVE
) ? TRUE
: FALSE
);
373 if (!RB_EMPTY(&vp
->v_rbdirty_tree
) || !RB_EMPTY(&vp
->v_rbclean_tree
))
374 panic("vinvalbuf: flush failed");
379 vinvalbuf_bp(struct buf
*bp
, void *data
)
381 struct vinvalbuf_bp_info
*info
= data
;
384 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
)) {
385 error
= BUF_TIMELOCK(bp
,
386 LK_EXCLUSIVE
| LK_SLEEPFAIL
,
387 "vinvalbuf", info
->slpflag
, info
->slptimeo
);
397 * XXX Since there are no node locks for NFS, I
398 * believe there is a slight chance that a delayed
399 * write will occur while sleeping just above, so
400 * check for it. Note that vfs_bio_awrite expects
401 * buffers to reside on a queue, while VOP_BWRITE and
404 if (((bp
->b_flags
& (B_DELWRI
| B_INVAL
)) == B_DELWRI
) &&
405 (info
->flags
& V_SAVE
)) {
406 if (bp
->b_vp
== info
->vp
) {
407 if (bp
->b_flags
& B_CLUSTEROK
) {
412 bp
->b_flags
|= B_ASYNC
;
413 VOP_BWRITE(bp
->b_vp
, bp
);
417 VOP_BWRITE(bp
->b_vp
, bp
);
421 bp
->b_flags
|= (B_INVAL
| B_NOCACHE
| B_RELBUF
);
422 bp
->b_flags
&= ~B_ASYNC
;
429 * Truncate a file's buffer and pages to a specified length. This
430 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
433 * The vnode must be locked.
435 static int vtruncbuf_bp_trunc_cmp(struct buf
*bp
, void *data
);
436 static int vtruncbuf_bp_trunc(struct buf
*bp
, void *data
);
437 static int vtruncbuf_bp_metasync_cmp(struct buf
*bp
, void *data
);
438 static int vtruncbuf_bp_metasync(struct buf
*bp
, void *data
);
441 vtruncbuf(struct vnode
*vp
, struct thread
*td
, off_t length
, int blksize
)
447 * Round up to the *next* lbn, then destroy the buffers in question.
448 * Since we are only removing some of the buffers we must rely on the
449 * scan count to determine whether a loop is necessary.
451 trunclbn
= (length
+ blksize
- 1) / blksize
;
455 count
= RB_SCAN(buf_rb_tree
, &vp
->v_rbclean_tree
,
456 vtruncbuf_bp_trunc_cmp
,
457 vtruncbuf_bp_trunc
, &trunclbn
);
458 count
+= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
459 vtruncbuf_bp_trunc_cmp
,
460 vtruncbuf_bp_trunc
, &trunclbn
);
464 * For safety, fsync any remaining metadata if the file is not being
465 * truncated to 0. Since the metadata does not represent the entire
466 * dirty list we have to rely on the hit count to ensure that we get
471 count
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
472 vtruncbuf_bp_metasync_cmp
,
473 vtruncbuf_bp_metasync
, vp
);
478 * Wait for any in-progress I/O to complete before returning (why?)
480 while (vp
->v_numoutput
> 0) {
481 vp
->v_flag
|= VBWAIT
;
482 tsleep(&vp
->v_numoutput
, 0, "vbtrunc", 0);
487 vnode_pager_setsize(vp
, length
);
493 * The callback buffer is beyond the new file EOF and must be destroyed.
494 * Note that the compare function must conform to the RB_SCAN's requirements.
498 vtruncbuf_bp_trunc_cmp(struct buf
*bp
, void *data
)
500 if (bp
->b_lblkno
>= *(daddr_t
*)data
)
507 vtruncbuf_bp_trunc(struct buf
*bp
, void *data
)
510 * Do not try to use a buffer we cannot immediately lock, but sleep
511 * anyway to prevent a livelock. The code will loop until all buffers
514 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
)) {
515 if (BUF_LOCK(bp
, LK_EXCLUSIVE
|LK_SLEEPFAIL
) == 0)
519 bp
->b_flags
|= (B_INVAL
| B_RELBUF
);
520 bp
->b_flags
&= ~B_ASYNC
;
527 * Fsync all meta-data after truncating a file to be non-zero. Only metadata
528 * blocks (with a negative lblkno) are scanned.
529 * Note that the compare function must conform to the RB_SCAN's requirements.
532 vtruncbuf_bp_metasync_cmp(struct buf
*bp
, void *data
)
534 if (bp
->b_lblkno
< 0)
540 vtruncbuf_bp_metasync(struct buf
*bp
, void *data
)
542 struct vnode
*vp
= data
;
544 if (bp
->b_flags
& B_DELWRI
) {
546 * Do not try to use a buffer we cannot immediately lock,
547 * but sleep anyway to prevent a livelock. The code will
548 * loop until all buffers can be acted upon.
550 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
)) {
551 if (BUF_LOCK(bp
, LK_EXCLUSIVE
|LK_SLEEPFAIL
) == 0)
555 if (bp
->b_vp
== vp
) {
556 bp
->b_flags
|= B_ASYNC
;
558 bp
->b_flags
&= ~B_ASYNC
;
560 VOP_BWRITE(bp
->b_vp
, bp
);
569 * vfsync - implements a multipass fsync on a file which understands
570 * dependancies and meta-data. The passed vnode must be locked. The
571 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY.
573 * When fsyncing data asynchronously just do one consolidated pass starting
574 * with the most negative block number. This may not get all the data due
577 * When fsyncing data synchronously do a data pass, then a metadata pass,
578 * then do additional data+metadata passes to try to get all the data out.
580 static int vfsync_wait_output(struct vnode
*vp
,
581 int (*waitoutput
)(struct vnode
*, struct thread
*));
582 static int vfsync_data_only_cmp(struct buf
*bp
, void *data
);
583 static int vfsync_meta_only_cmp(struct buf
*bp
, void *data
);
584 static int vfsync_lazy_range_cmp(struct buf
*bp
, void *data
);
585 static int vfsync_bp(struct buf
*bp
, void *data
);
594 int (*checkdef
)(struct buf
*);
598 vfsync(struct vnode
*vp
, int waitfor
, int passes
, daddr_t lbn
,
599 int (*checkdef
)(struct buf
*),
600 int (*waitoutput
)(struct vnode
*, struct thread
*))
602 struct vfsync_info info
;
605 bzero(&info
, sizeof(info
));
608 if ((info
.checkdef
= checkdef
) == NULL
)
616 * Lazy (filesystem syncer typ) Asynchronous plus limit the
617 * number of data (not meta) pages we try to flush to 1MB.
618 * A non-zero return means that lazy limit was reached.
620 info
.lazylimit
= 1024 * 1024;
622 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
623 vfsync_lazy_range_cmp
, vfsync_bp
, &info
);
624 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
625 vfsync_meta_only_cmp
, vfsync_bp
, &info
);
628 else if (!RB_EMPTY(&vp
->v_rbdirty_tree
))
629 vn_syncer_add_to_worklist(vp
, 1);
634 * Asynchronous. Do a data-only pass and a meta-only pass.
637 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, vfsync_data_only_cmp
,
639 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, vfsync_meta_only_cmp
,
645 * Synchronous. Do a data-only pass, then a meta-data+data
646 * pass, then additional integrated passes to try to get
647 * all the dependancies flushed.
649 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, vfsync_data_only_cmp
,
651 error
= vfsync_wait_output(vp
, waitoutput
);
653 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, NULL
,
655 error
= vfsync_wait_output(vp
, waitoutput
);
657 while (error
== 0 && passes
> 0 &&
658 !RB_EMPTY(&vp
->v_rbdirty_tree
)) {
660 info
.synchronous
= 1;
663 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, NULL
,
669 error
= vfsync_wait_output(vp
, waitoutput
);
678 vfsync_wait_output(struct vnode
*vp
, int (*waitoutput
)(struct vnode
*, struct thread
*))
682 while (vp
->v_numoutput
) {
683 vp
->v_flag
|= VBWAIT
;
684 tsleep(&vp
->v_numoutput
, 0, "fsfsn", 0);
687 error
= waitoutput(vp
, curthread
);
692 vfsync_data_only_cmp(struct buf
*bp
, void *data
)
694 if (bp
->b_lblkno
< 0)
700 vfsync_meta_only_cmp(struct buf
*bp
, void *data
)
702 if (bp
->b_lblkno
< 0)
708 vfsync_lazy_range_cmp(struct buf
*bp
, void *data
)
710 struct vfsync_info
*info
= data
;
711 if (bp
->b_lblkno
< info
->vp
->v_lazyw
)
717 vfsync_bp(struct buf
*bp
, void *data
)
719 struct vfsync_info
*info
= data
;
720 struct vnode
*vp
= info
->vp
;
724 * if syncdeps is not set we do not try to write buffers which have
727 if (!info
->synchronous
&& info
->syncdeps
== 0 && info
->checkdef(bp
))
731 * Ignore buffers that we cannot immediately lock. XXX
733 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
))
735 if ((bp
->b_flags
& B_DELWRI
) == 0)
736 panic("vfsync_bp: buffer not dirty");
738 panic("vfsync_bp: buffer vp mismatch");
741 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
742 * has been written but an additional handshake with the device
743 * is required before we can dispose of the buffer. We have no idea
744 * how to do this so we have to skip these buffers.
746 if (bp
->b_flags
& B_NEEDCOMMIT
) {
752 * (LEGACY FROM UFS, REMOVE WHEN POSSIBLE) - invalidate any dirty
753 * buffers beyond the file EOF.
755 if (info
->lbn
!= (daddr_t
)-1 && vp
->v_type
== VREG
&&
756 bp
->b_lblkno
>= info
->lbn
) {
758 bp
->b_flags
|= B_INVAL
| B_NOCACHE
;
764 if (info
->synchronous
) {
766 * Synchronous flushing. An error may be returned.
774 * Asynchronous flushing. A negative return value simply
775 * stops the scan and is not considered an error. We use
776 * this to support limited MNT_LAZY flushes.
778 vp
->v_lazyw
= bp
->b_lblkno
;
779 if ((vp
->v_flag
& VOBJBUF
) && (bp
->b_flags
& B_CLUSTEROK
)) {
781 info
->lazycount
+= vfs_bio_awrite(bp
);
783 info
->lazycount
+= bp
->b_bufsize
;
789 if (info
->lazylimit
&& info
->lazycount
>= info
->lazylimit
)
798 * Associate a buffer with a vnode.
801 bgetvp(struct vnode
*vp
, struct buf
*bp
)
803 KASSERT(bp
->b_vp
== NULL
, ("bgetvp: not free"));
807 bp
->b_dev
= vn_todev(vp
);
809 * Insert onto list for new vnode.
812 bp
->b_xflags
|= BX_VNCLEAN
;
813 bp
->b_xflags
&= ~BX_VNDIRTY
;
814 if (buf_rb_tree_RB_INSERT(&vp
->v_rbclean_tree
, bp
))
815 panic("reassignbuf: dup lblk vp %p bp %p", vp
, bp
);
820 * Disassociate a buffer from a vnode.
823 brelvp(struct buf
*bp
)
827 KASSERT(bp
->b_vp
!= NULL
, ("brelvp: NULL"));
830 * Delete from old vnode list, if on one.
834 if (bp
->b_xflags
& (BX_VNDIRTY
| BX_VNCLEAN
)) {
835 if (bp
->b_xflags
& BX_VNDIRTY
)
836 buf_rb_tree_RB_REMOVE(&vp
->v_rbdirty_tree
, bp
);
838 buf_rb_tree_RB_REMOVE(&vp
->v_rbclean_tree
, bp
);
839 bp
->b_xflags
&= ~(BX_VNDIRTY
| BX_VNCLEAN
);
841 if ((vp
->v_flag
& VONWORKLST
) && RB_EMPTY(&vp
->v_rbdirty_tree
)) {
842 vp
->v_flag
&= ~VONWORKLST
;
843 LIST_REMOVE(vp
, v_synclist
);
851 * Associate a p-buffer with a vnode.
853 * Also sets B_PAGING flag to indicate that vnode is not fully associated
854 * with the buffer. i.e. the bp has not been linked into the vnode or
858 pbgetvp(struct vnode
*vp
, struct buf
*bp
)
860 KASSERT(bp
->b_vp
== NULL
, ("pbgetvp: not free"));
863 bp
->b_flags
|= B_PAGING
;
864 bp
->b_dev
= vn_todev(vp
);
868 * Disassociate a p-buffer from a vnode.
871 pbrelvp(struct buf
*bp
)
873 KASSERT(bp
->b_vp
!= NULL
, ("pbrelvp: NULL"));
876 bp
->b_flags
&= ~B_PAGING
;
880 pbreassignbuf(struct buf
*bp
, struct vnode
*newvp
)
882 if ((bp
->b_flags
& B_PAGING
) == 0) {
884 "pbreassignbuf() on non phys bp %p",
892 * Reassign a buffer from one vnode to another.
893 * Used to assign file specific control information
894 * (indirect blocks) to the vnode to which they belong.
897 reassignbuf(struct buf
*bp
, struct vnode
*newvp
)
902 printf("reassignbuf: NULL");
908 * B_PAGING flagged buffers cannot be reassigned because their vp
909 * is not fully linked in.
911 if (bp
->b_flags
& B_PAGING
)
912 panic("cannot reassign paging buffer");
916 * Delete from old vnode list, if on one.
918 if (bp
->b_xflags
& (BX_VNDIRTY
| BX_VNCLEAN
)) {
919 if (bp
->b_xflags
& BX_VNDIRTY
)
920 buf_rb_tree_RB_REMOVE(&bp
->b_vp
->v_rbdirty_tree
, bp
);
922 buf_rb_tree_RB_REMOVE(&bp
->b_vp
->v_rbclean_tree
, bp
);
923 bp
->b_xflags
&= ~(BX_VNDIRTY
| BX_VNCLEAN
);
924 if (bp
->b_vp
!= newvp
) {
926 bp
->b_vp
= NULL
; /* for clarification */
930 * If dirty, put on list of dirty buffers; otherwise insert onto list
933 if (bp
->b_flags
& B_DELWRI
) {
934 if ((newvp
->v_flag
& VONWORKLST
) == 0) {
935 switch (newvp
->v_type
) {
942 newvp
->v_rdev
->si_mountpoint
!= NULL
) {
950 vn_syncer_add_to_worklist(newvp
, delay
);
952 bp
->b_xflags
|= BX_VNDIRTY
;
953 if (buf_rb_tree_RB_INSERT(&newvp
->v_rbdirty_tree
, bp
))
954 panic("reassignbuf: dup lblk vp %p bp %p", newvp
, bp
);
956 bp
->b_xflags
|= BX_VNCLEAN
;
957 if (buf_rb_tree_RB_INSERT(&newvp
->v_rbclean_tree
, bp
))
958 panic("reassignbuf: dup lblk vp %p bp %p", newvp
, bp
);
959 if ((newvp
->v_flag
& VONWORKLST
) &&
960 RB_EMPTY(&newvp
->v_rbdirty_tree
)) {
961 newvp
->v_flag
&= ~VONWORKLST
;
962 LIST_REMOVE(newvp
, v_synclist
);
965 if (bp
->b_vp
!= newvp
) {
973 * Create a vnode for a block device.
974 * Used for mounting the root file system.
977 bdevvp(dev_t dev
, struct vnode
**vpp
)
987 error
= getspecialvnode(VT_NON
, NULL
, &spec_vnode_vops
, &nvp
, 0, 0);
994 vp
->v_udev
= dev
->si_udev
;
1001 v_associate_rdev(struct vnode
*vp
, dev_t dev
)
1005 if (dev
== NULL
|| dev
== NODEV
)
1007 if (dev_is_good(dev
) == 0)
1009 KKASSERT(vp
->v_rdev
== NULL
);
1012 vp
->v_rdev
= reference_dev(dev
);
1013 lwkt_gettoken(&ilock
, &spechash_token
);
1014 SLIST_INSERT_HEAD(&dev
->si_hlist
, vp
, v_specnext
);
1015 lwkt_reltoken(&ilock
);
1020 v_release_rdev(struct vnode
*vp
)
1025 if ((dev
= vp
->v_rdev
) != NULL
) {
1026 lwkt_gettoken(&ilock
, &spechash_token
);
1027 SLIST_REMOVE(&dev
->si_hlist
, vp
, vnode
, v_specnext
);
1028 if (dev_ref_debug
&& vp
->v_opencount
!= 0) {
1029 printf("releasing rdev with non-0 "
1030 "v_opencount(%d) (revoked?)\n",
1034 vp
->v_opencount
= 0;
1036 lwkt_reltoken(&ilock
);
1041 * Add a vnode to the alias list hung off the dev_t. We only associate
1042 * the device number with the vnode. The actual device is not associated
1043 * until the vnode is opened (usually in spec_open()), and will be
1044 * disassociated on last close.
1047 addaliasu(struct vnode
*nvp
, udev_t nvp_udev
)
1049 if (nvp
->v_type
!= VBLK
&& nvp
->v_type
!= VCHR
)
1050 panic("addaliasu on non-special vnode");
1051 nvp
->v_udev
= nvp_udev
;
1055 * Disassociate a vnode from its underlying filesystem.
1057 * The vnode must be VX locked and refd
1059 * If there are v_usecount references to the vnode other then ours we have
1060 * to VOP_CLOSE the vnode before we can deactivate and reclaim it.
1063 vclean(struct vnode
*vp
, int flags
, struct thread
*td
)
1068 * If the vnode has already been reclaimed we have nothing to do.
1070 if (vp
->v_flag
& VRECLAIMED
)
1072 vp
->v_flag
|= VRECLAIMED
;
1075 * Scrap the vfs cache
1077 while (cache_inval_vp(vp
, 0) != 0) {
1078 printf("Warning: vnode %p clean/cache_resolution race detected\n", vp
);
1079 tsleep(vp
, 0, "vclninv", 2);
1083 * Check to see if the vnode is in use. If so we have to reference it
1084 * before we clean it out so that its count cannot fall to zero and
1085 * generate a race against ourselves to recycle it.
1087 active
= (vp
->v_usecount
> 1);
1090 * Clean out any buffers associated with the vnode and destroy its
1091 * object, if it has one.
1093 vinvalbuf(vp
, V_SAVE
, td
, 0, 0);
1094 VOP_DESTROYVOBJECT(vp
);
1097 * If purging an active vnode, it must be closed and
1098 * deactivated before being reclaimed. XXX
1100 * Note that neither of these routines unlocks the vnode.
1103 if (flags
& DOCLOSE
)
1104 VOP_CLOSE(vp
, FNONBLOCK
, td
);
1108 * If the vnode has not be deactivated, deactivated it.
1110 if ((vp
->v_flag
& VINACTIVE
) == 0) {
1111 vp
->v_flag
|= VINACTIVE
;
1112 VOP_INACTIVE(vp
, td
);
1116 * Reclaim the vnode.
1118 if (VOP_RECLAIM(vp
, td
))
1119 panic("vclean: cannot reclaim");
1122 * Done with purge, notify sleepers of the grim news.
1124 vp
->v_ops
= &dead_vnode_vops
;
1130 * Eliminate all activity associated with the requested vnode
1131 * and with all vnodes aliased to the requested vnode.
1133 * The vnode must be referenced and vx_lock()'d
1135 * revoke { struct vnode *a_vp, int a_flags }
1138 vop_stdrevoke(struct vop_revoke_args
*ap
)
1140 struct vnode
*vp
, *vq
;
1144 KASSERT((ap
->a_flags
& REVOKEALL
) != 0, ("vop_revoke"));
1149 * If the vnode is already dead don't try to revoke it
1151 if (vp
->v_flag
& VRECLAIMED
)
1155 * If the vnode has a device association, scrap all vnodes associated
1156 * with the device. Don't let the device disappear on us while we
1157 * are scrapping the vnodes.
1159 * The passed vp will probably show up in the list, do not VX lock
1162 if (vp
->v_type
!= VCHR
&& vp
->v_type
!= VBLK
)
1164 if ((dev
= vp
->v_rdev
) == NULL
) {
1165 if ((dev
= udev2dev(vp
->v_udev
, vp
->v_type
== VBLK
)) == NODEV
)
1169 lwkt_gettoken(&ilock
, &spechash_token
);
1170 while ((vq
= SLIST_FIRST(&dev
->si_hlist
)) != NULL
) {
1171 if (vp
== vq
|| vx_get(vq
) == 0) {
1172 if (vq
== SLIST_FIRST(&dev
->si_hlist
))
1178 lwkt_reltoken(&ilock
);
1184 * Recycle an unused vnode to the front of the free list.
1186 * Returns 1 if we were successfully able to recycle the vnode,
1190 vrecycle(struct vnode
*vp
, struct thread
*td
)
1192 if (vp
->v_usecount
== 1) {
1200 * Eliminate all activity associated with a vnode in preparation for reuse.
1202 * The vnode must be VX locked and refd and will remain VX locked and refd
1203 * on return. This routine may be called with the vnode in any state, as
1204 * long as it is VX locked. The vnode will be cleaned out and marked
1205 * VRECLAIMED but will not actually be reused until all existing refs and
1208 * NOTE: This routine may be called on a vnode which has not yet been
1209 * already been deactivated (VOP_INACTIVE), or on a vnode which has
1210 * already been reclaimed.
1212 * This routine is not responsible for placing us back on the freelist.
1213 * Instead, it happens automatically when the caller releases the VX lock
1214 * (assuming there aren't any other references).
1217 vgone(struct vnode
*vp
)
1220 * assert that the VX lock is held. This is an absolute requirement
1221 * now for vgone() to be called.
1223 KKASSERT(vp
->v_lock
.lk_exclusivecount
== 1);
1226 * Clean out the filesystem specific data and set the VRECLAIMED
1227 * bit. Also deactivate the vnode if necessary.
1229 vclean(vp
, DOCLOSE
, curthread
);
1232 * Delete from old mount point vnode list, if on one.
1234 if (vp
->v_mount
!= NULL
)
1235 insmntque(vp
, NULL
);
1238 * If special device, remove it from special device alias list
1239 * if it is on one. This should normally only occur if a vnode is
1240 * being revoked as the device should otherwise have been released
1243 if ((vp
->v_type
== VBLK
|| vp
->v_type
== VCHR
) && vp
->v_rdev
!= NULL
) {
1254 * Lookup a vnode by device number.
1257 vfinddev(dev_t dev
, enum vtype type
, struct vnode
**vpp
)
1262 lwkt_gettoken(&ilock
, &spechash_token
);
1263 SLIST_FOREACH(vp
, &dev
->si_hlist
, v_specnext
) {
1264 if (type
== vp
->v_type
) {
1266 lwkt_reltoken(&ilock
);
1270 lwkt_reltoken(&ilock
);
1275 * Calculate the total number of references to a special device. This
1276 * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1277 * an overloaded field. Since udev2dev can now return NODEV, we have
1278 * to check for a NULL v_rdev.
1281 count_dev(dev_t dev
)
1287 if (SLIST_FIRST(&dev
->si_hlist
)) {
1288 lwkt_gettoken(&ilock
, &spechash_token
);
1289 SLIST_FOREACH(vp
, &dev
->si_hlist
, v_specnext
) {
1290 count
+= vp
->v_usecount
;
1292 lwkt_reltoken(&ilock
);
1298 count_udev(udev_t udev
)
1302 if ((dev
= udev2dev(udev
, 0)) == NODEV
)
1304 return(count_dev(dev
));
1308 vcount(struct vnode
*vp
)
1310 if (vp
->v_rdev
== NULL
)
1312 return(count_dev(vp
->v_rdev
));
1316 * Print out a description of a vnode.
1318 static char *typename
[] =
1319 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1322 vprint(char *label
, struct vnode
*vp
)
1327 printf("%s: %p: ", label
, (void *)vp
);
1329 printf("%p: ", (void *)vp
);
1330 printf("type %s, usecount %d, writecount %d, refcount %d,",
1331 typename
[vp
->v_type
], vp
->v_usecount
, vp
->v_writecount
,
1334 if (vp
->v_flag
& VROOT
)
1335 strcat(buf
, "|VROOT");
1336 if (vp
->v_flag
& VTEXT
)
1337 strcat(buf
, "|VTEXT");
1338 if (vp
->v_flag
& VSYSTEM
)
1339 strcat(buf
, "|VSYSTEM");
1340 if (vp
->v_flag
& VBWAIT
)
1341 strcat(buf
, "|VBWAIT");
1342 if (vp
->v_flag
& VFREE
)
1343 strcat(buf
, "|VFREE");
1344 if (vp
->v_flag
& VOBJBUF
)
1345 strcat(buf
, "|VOBJBUF");
1347 printf(" flags (%s)", &buf
[1]);
1348 if (vp
->v_data
== NULL
) {
1357 #include <ddb/ddb.h>
1359 static int db_show_locked_vnodes(struct mount
*mp
, void *data
);
1362 * List all of the locked vnodes in the system.
1363 * Called when debugging the kernel.
1365 DB_SHOW_COMMAND(lockedvnodes
, lockedvnodes
)
1367 printf("Locked vnodes\n");
1368 mountlist_scan(db_show_locked_vnodes
, NULL
,
1369 MNTSCAN_FORWARD
|MNTSCAN_NOBUSY
);
1373 db_show_locked_vnodes(struct mount
*mp
, void *data __unused
)
1377 TAILQ_FOREACH(vp
, &mp
->mnt_nvnodelist
, v_nmntvnodes
) {
1378 if (VOP_ISLOCKED(vp
, NULL
))
1379 vprint((char *)0, vp
);
1386 * Top level filesystem related information gathering.
1388 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS
);
1391 vfs_sysctl(SYSCTL_HANDLER_ARGS
)
1393 int *name
= (int *)arg1
- 1; /* XXX */
1394 u_int namelen
= arg2
+ 1; /* XXX */
1395 struct vfsconf
*vfsp
;
1397 #if 1 || defined(COMPAT_PRELITE2)
1398 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1400 return (sysctl_ovfs_conf(oidp
, arg1
, arg2
, req
));
1404 /* all sysctl names at this level are at least name and field */
1406 return (ENOTDIR
); /* overloaded */
1407 if (name
[0] != VFS_GENERIC
) {
1408 for (vfsp
= vfsconf
; vfsp
; vfsp
= vfsp
->vfc_next
)
1409 if (vfsp
->vfc_typenum
== name
[0])
1412 return (EOPNOTSUPP
);
1413 return ((*vfsp
->vfc_vfsops
->vfs_sysctl
)(&name
[1], namelen
- 1,
1414 oldp
, oldlenp
, newp
, newlen
, p
));
1418 case VFS_MAXTYPENUM
:
1421 return (SYSCTL_OUT(req
, &maxvfsconf
, sizeof(int)));
1424 return (ENOTDIR
); /* overloaded */
1425 for (vfsp
= vfsconf
; vfsp
; vfsp
= vfsp
->vfc_next
)
1426 if (vfsp
->vfc_typenum
== name
[2])
1429 return (EOPNOTSUPP
);
1430 return (SYSCTL_OUT(req
, vfsp
, sizeof *vfsp
));
1432 return (EOPNOTSUPP
);
1435 SYSCTL_NODE(_vfs
, VFS_GENERIC
, generic
, CTLFLAG_RD
, vfs_sysctl
,
1436 "Generic filesystem");
1438 #if 1 || defined(COMPAT_PRELITE2)
1441 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS
)
1444 struct vfsconf
*vfsp
;
1445 struct ovfsconf ovfs
;
1447 for (vfsp
= vfsconf
; vfsp
; vfsp
= vfsp
->vfc_next
) {
1448 bzero(&ovfs
, sizeof(ovfs
));
1449 ovfs
.vfc_vfsops
= vfsp
->vfc_vfsops
; /* XXX used as flag */
1450 strcpy(ovfs
.vfc_name
, vfsp
->vfc_name
);
1451 ovfs
.vfc_index
= vfsp
->vfc_typenum
;
1452 ovfs
.vfc_refcount
= vfsp
->vfc_refcount
;
1453 ovfs
.vfc_flags
= vfsp
->vfc_flags
;
1454 error
= SYSCTL_OUT(req
, &ovfs
, sizeof ovfs
);
1461 #endif /* 1 || COMPAT_PRELITE2 */
1464 * Check to see if a filesystem is mounted on a block device.
1467 vfs_mountedon(struct vnode
*vp
)
1471 if ((dev
= vp
->v_rdev
) == NULL
)
1472 dev
= udev2dev(vp
->v_udev
, (vp
->v_type
== VBLK
));
1473 if (dev
!= NODEV
&& dev
->si_mountpoint
)
1479 * Unmount all filesystems. The list is traversed in reverse order
1480 * of mounting to avoid dependencies.
1483 static int vfs_umountall_callback(struct mount
*mp
, void *data
);
1486 vfs_unmountall(void)
1488 struct thread
*td
= curthread
;
1491 if (td
->td_proc
== NULL
)
1492 td
= initproc
->p_thread
; /* XXX XXX use proc0 instead? */
1495 count
= mountlist_scan(vfs_umountall_callback
,
1496 &td
, MNTSCAN_REVERSE
|MNTSCAN_NOBUSY
);
1502 vfs_umountall_callback(struct mount
*mp
, void *data
)
1504 struct thread
*td
= *(struct thread
**)data
;
1507 error
= dounmount(mp
, MNT_FORCE
, td
);
1509 mountlist_remove(mp
);
1510 printf("unmount of filesystem mounted from %s failed (",
1511 mp
->mnt_stat
.f_mntfromname
);
1515 printf("%d)\n", error
);
1521 * Build hash lists of net addresses and hang them off the mount point.
1522 * Called by ufs_mount() to set up the lists of export addresses.
1525 vfs_hang_addrlist(struct mount
*mp
, struct netexport
*nep
,
1526 struct export_args
*argp
)
1529 struct radix_node_head
*rnh
;
1531 struct radix_node
*rn
;
1532 struct sockaddr
*saddr
, *smask
= 0;
1536 if (argp
->ex_addrlen
== 0) {
1537 if (mp
->mnt_flag
& MNT_DEFEXPORTED
)
1539 np
= &nep
->ne_defexported
;
1540 np
->netc_exflags
= argp
->ex_flags
;
1541 np
->netc_anon
= argp
->ex_anon
;
1542 np
->netc_anon
.cr_ref
= 1;
1543 mp
->mnt_flag
|= MNT_DEFEXPORTED
;
1547 if (argp
->ex_addrlen
< 0 || argp
->ex_addrlen
> MLEN
)
1549 if (argp
->ex_masklen
< 0 || argp
->ex_masklen
> MLEN
)
1552 i
= sizeof(struct netcred
) + argp
->ex_addrlen
+ argp
->ex_masklen
;
1553 np
= (struct netcred
*) malloc(i
, M_NETADDR
, M_WAITOK
);
1554 bzero((caddr_t
) np
, i
);
1555 saddr
= (struct sockaddr
*) (np
+ 1);
1556 if ((error
= copyin(argp
->ex_addr
, (caddr_t
) saddr
, argp
->ex_addrlen
)))
1558 if (saddr
->sa_len
> argp
->ex_addrlen
)
1559 saddr
->sa_len
= argp
->ex_addrlen
;
1560 if (argp
->ex_masklen
) {
1561 smask
= (struct sockaddr
*)((caddr_t
)saddr
+ argp
->ex_addrlen
);
1562 error
= copyin(argp
->ex_mask
, (caddr_t
)smask
, argp
->ex_masklen
);
1565 if (smask
->sa_len
> argp
->ex_masklen
)
1566 smask
->sa_len
= argp
->ex_masklen
;
1568 i
= saddr
->sa_family
;
1569 if ((rnh
= nep
->ne_rtable
[i
]) == 0) {
1571 * Seems silly to initialize every AF when most are not used,
1572 * do so on demand here
1574 SLIST_FOREACH(dom
, &domains
, dom_next
)
1575 if (dom
->dom_family
== i
&& dom
->dom_rtattach
) {
1576 dom
->dom_rtattach((void **) &nep
->ne_rtable
[i
],
1580 if ((rnh
= nep
->ne_rtable
[i
]) == 0) {
1585 rn
= (*rnh
->rnh_addaddr
) ((char *) saddr
, (char *) smask
, rnh
,
1587 if (rn
== 0 || np
!= (struct netcred
*) rn
) { /* already exists */
1591 np
->netc_exflags
= argp
->ex_flags
;
1592 np
->netc_anon
= argp
->ex_anon
;
1593 np
->netc_anon
.cr_ref
= 1;
1596 free(np
, M_NETADDR
);
1602 vfs_free_netcred(struct radix_node
*rn
, void *w
)
1604 struct radix_node_head
*rnh
= (struct radix_node_head
*) w
;
1606 (*rnh
->rnh_deladdr
) (rn
->rn_key
, rn
->rn_mask
, rnh
);
1607 free((caddr_t
) rn
, M_NETADDR
);
1612 * Free the net address hash lists that are hanging off the mount points.
1615 vfs_free_addrlist(struct netexport
*nep
)
1618 struct radix_node_head
*rnh
;
1620 for (i
= 0; i
<= AF_MAX
; i
++)
1621 if ((rnh
= nep
->ne_rtable
[i
])) {
1622 (*rnh
->rnh_walktree
) (rnh
, vfs_free_netcred
,
1624 free((caddr_t
) rnh
, M_RTABLE
);
1625 nep
->ne_rtable
[i
] = 0;
1630 vfs_export(struct mount
*mp
, struct netexport
*nep
, struct export_args
*argp
)
1634 if (argp
->ex_flags
& MNT_DELEXPORT
) {
1635 if (mp
->mnt_flag
& MNT_EXPUBLIC
) {
1636 vfs_setpublicfs(NULL
, NULL
, NULL
);
1637 mp
->mnt_flag
&= ~MNT_EXPUBLIC
;
1639 vfs_free_addrlist(nep
);
1640 mp
->mnt_flag
&= ~(MNT_EXPORTED
| MNT_DEFEXPORTED
);
1642 if (argp
->ex_flags
& MNT_EXPORTED
) {
1643 if (argp
->ex_flags
& MNT_EXPUBLIC
) {
1644 if ((error
= vfs_setpublicfs(mp
, nep
, argp
)) != 0)
1646 mp
->mnt_flag
|= MNT_EXPUBLIC
;
1648 if ((error
= vfs_hang_addrlist(mp
, nep
, argp
)))
1650 mp
->mnt_flag
|= MNT_EXPORTED
;
1657 * Set the publicly exported filesystem (WebNFS). Currently, only
1658 * one public filesystem is possible in the spec (RFC 2054 and 2055)
1661 vfs_setpublicfs(struct mount
*mp
, struct netexport
*nep
,
1662 struct export_args
*argp
)
1669 * mp == NULL -> invalidate the current info, the FS is
1670 * no longer exported. May be called from either vfs_export
1671 * or unmount, so check if it hasn't already been done.
1674 if (nfs_pub
.np_valid
) {
1675 nfs_pub
.np_valid
= 0;
1676 if (nfs_pub
.np_index
!= NULL
) {
1677 FREE(nfs_pub
.np_index
, M_TEMP
);
1678 nfs_pub
.np_index
= NULL
;
1685 * Only one allowed at a time.
1687 if (nfs_pub
.np_valid
!= 0 && mp
!= nfs_pub
.np_mount
)
1691 * Get real filehandle for root of exported FS.
1693 bzero((caddr_t
)&nfs_pub
.np_handle
, sizeof(nfs_pub
.np_handle
));
1694 nfs_pub
.np_handle
.fh_fsid
= mp
->mnt_stat
.f_fsid
;
1696 if ((error
= VFS_ROOT(mp
, &rvp
)))
1699 if ((error
= VFS_VPTOFH(rvp
, &nfs_pub
.np_handle
.fh_fid
)))
1705 * If an indexfile was specified, pull it in.
1707 if (argp
->ex_indexfile
!= NULL
) {
1708 MALLOC(nfs_pub
.np_index
, char *, MAXNAMLEN
+ 1, M_TEMP
,
1710 error
= copyinstr(argp
->ex_indexfile
, nfs_pub
.np_index
,
1711 MAXNAMLEN
, (size_t *)0);
1714 * Check for illegal filenames.
1716 for (cp
= nfs_pub
.np_index
; *cp
; cp
++) {
1724 FREE(nfs_pub
.np_index
, M_TEMP
);
1729 nfs_pub
.np_mount
= mp
;
1730 nfs_pub
.np_valid
= 1;
1735 vfs_export_lookup(struct mount
*mp
, struct netexport
*nep
,
1736 struct sockaddr
*nam
)
1739 struct radix_node_head
*rnh
;
1740 struct sockaddr
*saddr
;
1743 if (mp
->mnt_flag
& MNT_EXPORTED
) {
1745 * Lookup in the export list first.
1749 rnh
= nep
->ne_rtable
[saddr
->sa_family
];
1751 np
= (struct netcred
*)
1752 (*rnh
->rnh_matchaddr
)((char *)saddr
,
1754 if (np
&& np
->netc_rnodes
->rn_flags
& RNF_ROOT
)
1759 * If no address match, use the default if it exists.
1761 if (np
== NULL
&& mp
->mnt_flag
& MNT_DEFEXPORTED
)
1762 np
= &nep
->ne_defexported
;
1768 * perform msync on all vnodes under a mount point. The mount point must
1769 * be locked. This code is also responsible for lazy-freeing unreferenced
1770 * vnodes whos VM objects no longer contain pages.
1772 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
1774 static int vfs_msync_scan1(struct mount
*mp
, struct vnode
*vp
, void *data
);
1775 static int vfs_msync_scan2(struct mount
*mp
, struct vnode
*vp
, void *data
);
1778 vfs_msync(struct mount
*mp
, int flags
)
1780 vmntvnodescan(mp
, VMSC_REFVP
, vfs_msync_scan1
, vfs_msync_scan2
,
1785 * scan1 is a fast pre-check. There could be hundreds of thousands of
1786 * vnodes, we cannot afford to do anything heavy weight until we have a
1787 * fairly good indication that there is work to do.
1791 vfs_msync_scan1(struct mount
*mp
, struct vnode
*vp
, void *data
)
1793 int flags
= (int)data
;
1795 if ((vp
->v_flag
& VRECLAIMED
) == 0) {
1796 if (vshouldfree(vp
, 0))
1797 return(0); /* call scan2 */
1798 if ((mp
->mnt_flag
& MNT_RDONLY
) == 0 &&
1799 (vp
->v_flag
& VOBJDIRTY
) &&
1800 (flags
== MNT_WAIT
|| VOP_ISLOCKED(vp
, NULL
) == 0)) {
1801 return(0); /* call scan2 */
1806 * do not call scan2, continue the loop
1813 vfs_msync_scan2(struct mount
*mp
, struct vnode
*vp
, void *data
)
1816 int flags
= (int)data
;
1818 if (vp
->v_flag
& VRECLAIMED
)
1821 if ((mp
->mnt_flag
& MNT_RDONLY
) == 0 &&
1822 (vp
->v_flag
& VOBJDIRTY
) &&
1823 (flags
== MNT_WAIT
|| VOP_ISLOCKED(vp
, NULL
) == 0)) {
1824 if (VOP_GETVOBJECT(vp
, &obj
) == 0) {
1825 vm_object_page_clean(obj
, 0, 0,
1826 flags
== MNT_WAIT
? OBJPC_SYNC
: OBJPC_NOSYNC
);
1833 * Create the VM object needed for VMIO and mmap support. This
1834 * is done for all VREG files in the system. Some filesystems might
1835 * afford the additional metadata buffering capability of the
1836 * VMIO code by making the device node be VMIO mode also.
1838 * vp must be locked when vfs_object_create is called.
1841 vfs_object_create(struct vnode
*vp
, struct thread
*td
)
1843 return (VOP_CREATEVOBJECT(vp
, td
));
1847 * Record a process's interest in events which might happen to
1848 * a vnode. Because poll uses the historic select-style interface
1849 * internally, this routine serves as both the ``check for any
1850 * pending events'' and the ``record my interest in future events''
1851 * functions. (These are done together, while the lock is held,
1852 * to avoid race conditions.)
1855 vn_pollrecord(struct vnode
*vp
, struct thread
*td
, int events
)
1859 lwkt_gettoken(&ilock
, &vp
->v_pollinfo
.vpi_token
);
1860 if (vp
->v_pollinfo
.vpi_revents
& events
) {
1862 * This leaves events we are not interested
1863 * in available for the other process which
1864 * which presumably had requested them
1865 * (otherwise they would never have been
1868 events
&= vp
->v_pollinfo
.vpi_revents
;
1869 vp
->v_pollinfo
.vpi_revents
&= ~events
;
1871 lwkt_reltoken(&ilock
);
1874 vp
->v_pollinfo
.vpi_events
|= events
;
1875 selrecord(td
, &vp
->v_pollinfo
.vpi_selinfo
);
1876 lwkt_reltoken(&ilock
);
1881 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
1882 * it is possible for us to miss an event due to race conditions, but
1883 * that condition is expected to be rare, so for the moment it is the
1884 * preferred interface.
1887 vn_pollevent(struct vnode
*vp
, int events
)
1891 lwkt_gettoken(&ilock
, &vp
->v_pollinfo
.vpi_token
);
1892 if (vp
->v_pollinfo
.vpi_events
& events
) {
1894 * We clear vpi_events so that we don't
1895 * call selwakeup() twice if two events are
1896 * posted before the polling process(es) is
1897 * awakened. This also ensures that we take at
1898 * most one selwakeup() if the polling process
1899 * is no longer interested. However, it does
1900 * mean that only one event can be noticed at
1901 * a time. (Perhaps we should only clear those
1902 * event bits which we note?) XXX
1904 vp
->v_pollinfo
.vpi_events
= 0; /* &= ~events ??? */
1905 vp
->v_pollinfo
.vpi_revents
|= events
;
1906 selwakeup(&vp
->v_pollinfo
.vpi_selinfo
);
1908 lwkt_reltoken(&ilock
);
1912 * Wake up anyone polling on vp because it is being revoked.
1913 * This depends on dead_poll() returning POLLHUP for correct
1917 vn_pollgone(struct vnode
*vp
)
1921 lwkt_gettoken(&ilock
, &vp
->v_pollinfo
.vpi_token
);
1922 if (vp
->v_pollinfo
.vpi_events
) {
1923 vp
->v_pollinfo
.vpi_events
= 0;
1924 selwakeup(&vp
->v_pollinfo
.vpi_selinfo
);
1926 lwkt_reltoken(&ilock
);
1930 * extract the dev_t from a VBLK or VCHR. The vnode must have been opened
1931 * (or v_rdev might be NULL).
1934 vn_todev(struct vnode
*vp
)
1936 if (vp
->v_type
!= VBLK
&& vp
->v_type
!= VCHR
)
1938 KKASSERT(vp
->v_rdev
!= NULL
);
1939 return (vp
->v_rdev
);
1943 * Check if vnode represents a disk device. The vnode does not need to be
1947 vn_isdisk(struct vnode
*vp
, int *errp
)
1951 if (vp
->v_type
!= VBLK
&& vp
->v_type
!= VCHR
) {
1957 if ((dev
= vp
->v_rdev
) == NULL
)
1958 dev
= udev2dev(vp
->v_udev
, (vp
->v_type
== VBLK
));
1959 if (dev
== NULL
|| dev
== NODEV
) {
1964 if (dev_is_good(dev
) == 0) {
1969 if ((dev_dflags(dev
) & D_DISK
) == 0) {
1979 #ifdef DEBUG_VFS_LOCKS
1982 assert_vop_locked(struct vnode
*vp
, const char *str
)
1984 if (vp
&& IS_LOCKING_VFS(vp
) && !VOP_ISLOCKED(vp
, NULL
)) {
1985 panic("%s: %p is not locked shared but should be", str
, vp
);
1990 assert_vop_unlocked(struct vnode
*vp
, const char *str
)
1992 if (vp
&& IS_LOCKING_VFS(vp
)) {
1993 if (VOP_ISLOCKED(vp
, curthread
) == LK_EXCLUSIVE
) {
1994 panic("%s: %p is locked but should not be", str
, vp
);
2002 vn_get_namelen(struct vnode
*vp
, int *namelen
)
2004 int error
, retval
[2];
2006 error
= VOP_PATHCONF(vp
, _PC_NAME_MAX
, retval
);