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 1.57 2005/06/06 15:02:28 dillon Exp $
44 * External virtual filesystem routines
48 #include <sys/param.h>
49 #include <sys/systm.h>
52 #include <sys/dirent.h>
53 #include <sys/domain.h>
54 #include <sys/eventhandler.h>
55 #include <sys/fcntl.h>
56 #include <sys/kernel.h>
57 #include <sys/kthread.h>
58 #include <sys/malloc.h>
60 #include <sys/mount.h>
62 #include <sys/reboot.h>
63 #include <sys/socket.h>
65 #include <sys/sysctl.h>
66 #include <sys/syslog.h>
67 #include <sys/vmmeter.h>
68 #include <sys/vnode.h>
70 #include <machine/limits.h>
73 #include <vm/vm_object.h>
74 #include <vm/vm_extern.h>
75 #include <vm/vm_kern.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_page.h>
79 #include <vm/vm_pager.h>
80 #include <vm/vnode_pager.h>
81 #include <vm/vm_zone.h>
84 #include <sys/thread2.h>
86 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
89 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
91 SYSCTL_INT(_vfs, OID_AUTO, fastdev, CTLFLAG_RW, &vfs_fastdev, 0, "");
93 enum vtype iftovt_tab[16] = {
94 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
95 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
98 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
99 S_IFSOCK, S_IFIFO, S_IFMT,
102 static int reassignbufcalls;
103 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW,
104 &reassignbufcalls, 0, "");
105 static int reassignbufloops;
106 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW,
107 &reassignbufloops, 0, "");
108 static int reassignbufsortgood;
109 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW,
110 &reassignbufsortgood, 0, "");
111 static int reassignbufsortbad;
112 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW,
113 &reassignbufsortbad, 0, "");
114 static int reassignbufmethod = 1;
115 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW,
116 &reassignbufmethod, 0, "");
118 int nfs_mount_type = -1;
119 static struct lwkt_token spechash_token;
120 struct nfs_public nfs_pub; /* publicly exported FS */
123 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
124 &desiredvnodes, 0, "Maximum number of vnodes");
126 static void vfs_free_addrlist (struct netexport *nep);
127 static int vfs_free_netcred (struct radix_node *rn, void *w);
128 static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
129 struct export_args *argp);
131 extern int dev_ref_debug;
132 extern struct vnodeopv_entry_desc spec_vnodeop_entries[];
135 * Red black tree functions
137 static int rb_buf_compare(struct buf *b1, struct buf *b2);
138 RB_GENERATE(buf_rb_tree, buf, b_rbnode, rb_buf_compare);
141 rb_buf_compare(struct buf *b1, struct buf *b2)
143 if (b1->b_lblkno < b2->b_lblkno)
145 if (b1->b_lblkno > b2->b_lblkno)
151 * Return 0 if the vnode is already on the free list or cannot be placed
152 * on the free list. Return 1 if the vnode can be placed on the free list.
155 vshouldfree(struct vnode *vp, int usecount)
157 if (vp->v_flag & VFREE)
158 return (0); /* already free */
159 if (vp->v_holdcnt != 0 || vp->v_usecount != usecount)
160 return (0); /* other holderse */
162 (vp->v_object->ref_count || vp->v_object->resident_page_count)) {
169 * Initialize the vnode management data structures.
171 * Called from vfsinit()
177 * Desired vnodes is a result of the physical page count
178 * and the size of kernel's heap. It scales in proportion
179 * to the amount of available physical memory. This can
180 * cause trouble on 64-bit and large memory platforms.
182 /* desiredvnodes = maxproc + vmstats.v_page_count / 4; */
184 min(maxproc + vmstats.v_page_count /4,
185 2 * (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) /
186 (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
188 lwkt_token_init(&spechash_token);
192 * Knob to control the precision of file timestamps:
194 * 0 = seconds only; nanoseconds zeroed.
195 * 1 = seconds and nanoseconds, accurate within 1/HZ.
196 * 2 = seconds and nanoseconds, truncated to microseconds.
197 * >=3 = seconds and nanoseconds, maximum precision.
199 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
201 static int timestamp_precision = TSP_SEC;
202 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
203 ×tamp_precision, 0, "");
206 * Get a current timestamp.
209 vfs_timestamp(struct timespec *tsp)
213 switch (timestamp_precision) {
215 tsp->tv_sec = time_second;
223 TIMEVAL_TO_TIMESPEC(&tv, tsp);
233 * Set vnode attributes to VNOVAL
236 vattr_null(struct vattr *vap)
239 vap->va_size = VNOVAL;
240 vap->va_bytes = VNOVAL;
241 vap->va_mode = VNOVAL;
242 vap->va_nlink = VNOVAL;
243 vap->va_uid = VNOVAL;
244 vap->va_gid = VNOVAL;
245 vap->va_fsid = VNOVAL;
246 vap->va_fileid = VNOVAL;
247 vap->va_blocksize = VNOVAL;
248 vap->va_rdev = VNOVAL;
249 vap->va_atime.tv_sec = VNOVAL;
250 vap->va_atime.tv_nsec = VNOVAL;
251 vap->va_mtime.tv_sec = VNOVAL;
252 vap->va_mtime.tv_nsec = VNOVAL;
253 vap->va_ctime.tv_sec = VNOVAL;
254 vap->va_ctime.tv_nsec = VNOVAL;
255 vap->va_flags = VNOVAL;
256 vap->va_gen = VNOVAL;
261 * Update outstanding I/O count and do wakeup if requested.
264 vwakeup(struct buf *bp)
268 if ((vp = bp->b_vp)) {
270 if (vp->v_numoutput < 0)
271 panic("vwakeup: neg numoutput");
272 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
273 vp->v_flag &= ~VBWAIT;
274 wakeup((caddr_t) &vp->v_numoutput);
280 * Flush out and invalidate all buffers associated with a vnode.
284 static int vinvalbuf_bp(struct buf *bp, void *data);
286 struct vinvalbuf_bp_info {
294 vinvalbuf(struct vnode *vp, int flags, struct thread *td,
295 int slpflag, int slptimeo)
297 struct vinvalbuf_bp_info info;
302 * If we are being asked to save, call fsync to ensure that the inode
305 if (flags & V_SAVE) {
307 while (vp->v_numoutput) {
308 vp->v_flag |= VBWAIT;
309 error = tsleep((caddr_t)&vp->v_numoutput,
310 slpflag, "vinvlbuf", slptimeo);
316 if (!RB_EMPTY(&vp->v_rbdirty_tree)) {
318 if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0)
321 if (vp->v_numoutput > 0 ||
322 !RB_EMPTY(&vp->v_rbdirty_tree))
323 panic("vinvalbuf: dirty bufs");
328 info.slptimeo = slptimeo;
329 info.slpflag = slpflag;
334 * Flush the buffer cache until nothing is left.
336 while (!RB_EMPTY(&vp->v_rbclean_tree) ||
337 !RB_EMPTY(&vp->v_rbdirty_tree)) {
338 error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, NULL,
339 vinvalbuf_bp, &info);
341 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
342 vinvalbuf_bp, &info);
347 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
348 * have write I/O in-progress but if there is a VM object then the
349 * VM object can also have read-I/O in-progress.
352 while (vp->v_numoutput > 0) {
353 vp->v_flag |= VBWAIT;
354 tsleep(&vp->v_numoutput, 0, "vnvlbv", 0);
356 if (VOP_GETVOBJECT(vp, &object) == 0) {
357 while (object->paging_in_progress)
358 vm_object_pip_sleep(object, "vnvlbx");
360 } while (vp->v_numoutput > 0);
365 * Destroy the copy in the VM cache, too.
367 if (VOP_GETVOBJECT(vp, &object) == 0) {
368 vm_object_page_remove(object, 0, 0,
369 (flags & V_SAVE) ? TRUE : FALSE);
372 if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree))
373 panic("vinvalbuf: flush failed");
378 vinvalbuf_bp(struct buf *bp, void *data)
380 struct vinvalbuf_bp_info *info = data;
383 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
384 error = BUF_TIMELOCK(bp,
385 LK_EXCLUSIVE | LK_SLEEPFAIL,
386 "vinvalbuf", info->slpflag, info->slptimeo);
396 * XXX Since there are no node locks for NFS, I
397 * believe there is a slight chance that a delayed
398 * write will occur while sleeping just above, so
399 * check for it. Note that vfs_bio_awrite expects
400 * buffers to reside on a queue, while VOP_BWRITE and
403 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
404 (info->flags & V_SAVE)) {
405 if (bp->b_vp == info->vp) {
406 if (bp->b_flags & B_CLUSTEROK) {
411 bp->b_flags |= B_ASYNC;
412 VOP_BWRITE(bp->b_vp, bp);
416 VOP_BWRITE(bp->b_vp, bp);
420 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
421 bp->b_flags &= ~B_ASYNC;
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, struct thread *td, off_t length, int blksize)
446 * Round up to the *next* lbn, then destroy the buffers in question.
447 * Since we are only removing some of the buffers we must rely on the
448 * scan count to determine whether a loop is necessary.
450 trunclbn = (length + blksize - 1) / blksize;
454 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
455 vtruncbuf_bp_trunc_cmp,
456 vtruncbuf_bp_trunc, &trunclbn);
457 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
458 vtruncbuf_bp_trunc_cmp,
459 vtruncbuf_bp_trunc, &trunclbn);
463 * For safety, fsync any remaining metadata if the file is not being
464 * truncated to 0. Since the metadata does not represent the entire
465 * dirty list we have to rely on the hit count to ensure that we get
470 count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
471 vtruncbuf_bp_metasync_cmp,
472 vtruncbuf_bp_metasync, vp);
477 * Wait for any in-progress I/O to complete before returning (why?)
479 while (vp->v_numoutput > 0) {
480 vp->v_flag |= VBWAIT;
481 tsleep(&vp->v_numoutput, 0, "vbtrunc", 0);
486 vnode_pager_setsize(vp, length);
492 * The callback buffer is beyond the new file EOF and must be destroyed.
493 * Note that the compare function must conform to the RB_SCAN's requirements.
497 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data)
499 if (bp->b_lblkno >= *(daddr_t *)data)
506 vtruncbuf_bp_trunc(struct buf *bp, void *data)
509 * Do not try to use a buffer we cannot immediately lock, but sleep
510 * anyway to prevent a livelock. The code will loop until all buffers
513 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
514 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
518 bp->b_flags |= (B_INVAL | B_RELBUF);
519 bp->b_flags &= ~B_ASYNC;
526 * Fsync all meta-data after truncating a file to be non-zero. Only metadata
527 * blocks (with a negative lblkno) are scanned.
528 * Note that the compare function must conform to the RB_SCAN's requirements.
531 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data)
533 if (bp->b_lblkno < 0)
539 vtruncbuf_bp_metasync(struct buf *bp, void *data)
541 struct vnode *vp = data;
543 if (bp->b_flags & B_DELWRI) {
545 * Do not try to use a buffer we cannot immediately lock,
546 * but sleep anyway to prevent a livelock. The code will
547 * loop until all buffers can be acted upon.
549 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
550 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
554 if (bp->b_vp == vp) {
555 bp->b_flags |= B_ASYNC;
557 bp->b_flags &= ~B_ASYNC;
559 VOP_BWRITE(bp->b_vp, bp);
568 * vfsync - implements a multipass fsync on a file which understands
569 * dependancies and meta-data. The passed vnode must be locked. The
570 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY.
572 * When fsyncing data asynchronously just do one consolidated pass starting
573 * with the most negative block number. This may not get all the data due
576 * When fsyncing data synchronously do a data pass, then a metadata pass,
577 * then do additional data+metadata passes to try to get all the data out.
579 static int vfsync_wait_output(struct vnode *vp,
580 int (*waitoutput)(struct vnode *, struct thread *));
581 static int vfsync_data_only_cmp(struct buf *bp, void *data);
582 static int vfsync_meta_only_cmp(struct buf *bp, void *data);
583 static int vfsync_lazy_range_cmp(struct buf *bp, void *data);
584 static int vfsync_bp(struct buf *bp, void *data);
593 int (*checkdef)(struct buf *);
597 vfsync(struct vnode *vp, int waitfor, int passes, daddr_t lbn,
598 int (*checkdef)(struct buf *),
599 int (*waitoutput)(struct vnode *, struct thread *))
601 struct vfsync_info info;
604 bzero(&info, sizeof(info));
607 if ((info.checkdef = checkdef) == NULL)
615 * Lazy (filesystem syncer typ) Asynchronous plus limit the
616 * number of data (not meta) pages we try to flush to 1MB.
617 * A non-zero return means that lazy limit was reached.
619 info.lazylimit = 1024 * 1024;
621 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
622 vfsync_lazy_range_cmp, vfsync_bp, &info);
623 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
624 vfsync_meta_only_cmp, vfsync_bp, &info);
627 else if (!RB_EMPTY(&vp->v_rbdirty_tree))
628 vn_syncer_add_to_worklist(vp, 1);
633 * Asynchronous. Do a data-only pass and a meta-only pass.
636 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
638 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp,
644 * Synchronous. Do a data-only pass, then a meta-data+data
645 * pass, then additional integrated passes to try to get
646 * all the dependancies flushed.
648 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
650 error = vfsync_wait_output(vp, waitoutput);
652 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
654 error = vfsync_wait_output(vp, waitoutput);
656 while (error == 0 && passes > 0 &&
657 !RB_EMPTY(&vp->v_rbdirty_tree)) {
659 info.synchronous = 1;
662 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
668 error = vfsync_wait_output(vp, waitoutput);
677 vfsync_wait_output(struct vnode *vp, int (*waitoutput)(struct vnode *, struct thread *))
681 while (vp->v_numoutput) {
682 vp->v_flag |= VBWAIT;
683 tsleep(&vp->v_numoutput, 0, "fsfsn", 0);
686 error = waitoutput(vp, curthread);
691 vfsync_data_only_cmp(struct buf *bp, void *data)
693 if (bp->b_lblkno < 0)
699 vfsync_meta_only_cmp(struct buf *bp, void *data)
701 if (bp->b_lblkno < 0)
707 vfsync_lazy_range_cmp(struct buf *bp, void *data)
709 struct vfsync_info *info = data;
710 if (bp->b_lblkno < info->vp->v_lazyw)
716 vfsync_bp(struct buf *bp, void *data)
718 struct vfsync_info *info = data;
719 struct vnode *vp = info->vp;
723 * if syncdeps is not set we do not try to write buffers which have
726 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp))
730 * Ignore buffers that we cannot immediately lock. XXX
732 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
734 if ((bp->b_flags & B_DELWRI) == 0)
735 panic("vfsync_bp: buffer not dirty");
737 panic("vfsync_bp: buffer vp mismatch");
740 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
741 * has been written but an additional handshake with the device
742 * is required before we can dispose of the buffer. We have no idea
743 * how to do this so we have to skip these buffers.
745 if (bp->b_flags & B_NEEDCOMMIT) {
751 * (LEGACY FROM UFS, REMOVE WHEN POSSIBLE) - invalidate any dirty
752 * buffers beyond the file EOF.
754 if (info->lbn != (daddr_t)-1 && vp->v_type == VREG &&
755 bp->b_lblkno >= info->lbn) {
757 bp->b_flags |= B_INVAL | B_NOCACHE;
763 if (info->synchronous) {
765 * Synchronous flushing. An error may be returned.
773 * Asynchronous flushing. A negative return value simply
774 * stops the scan and is not considered an error. We use
775 * this to support limited MNT_LAZY flushes.
777 vp->v_lazyw = bp->b_lblkno;
778 if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) {
780 info->lazycount += vfs_bio_awrite(bp);
782 info->lazycount += bp->b_bufsize;
788 if (info->lazylimit && info->lazycount >= info->lazylimit)
797 * Associate a buffer with a vnode.
800 bgetvp(struct vnode *vp, struct buf *bp)
802 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
806 bp->b_dev = vn_todev(vp);
808 * Insert onto list for new vnode.
811 bp->b_xflags |= BX_VNCLEAN;
812 bp->b_xflags &= ~BX_VNDIRTY;
813 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp))
814 panic("reassignbuf: dup lblk vp %p bp %p", vp, bp);
819 * Disassociate a buffer from a vnode.
822 brelvp(struct buf *bp)
826 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
829 * Delete from old vnode list, if on one.
833 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
834 if (bp->b_xflags & BX_VNDIRTY)
835 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
837 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
838 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
840 if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree)) {
841 vp->v_flag &= ~VONWORKLST;
842 LIST_REMOVE(vp, v_synclist);
850 * Associate a p-buffer with a vnode.
852 * Also sets B_PAGING flag to indicate that vnode is not fully associated
853 * with the buffer. i.e. the bp has not been linked into the vnode or
857 pbgetvp(struct vnode *vp, struct buf *bp)
859 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
862 bp->b_flags |= B_PAGING;
863 bp->b_dev = vn_todev(vp);
867 * Disassociate a p-buffer from a vnode.
870 pbrelvp(struct buf *bp)
872 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
875 bp->b_flags &= ~B_PAGING;
879 pbreassignbuf(struct buf *bp, struct vnode *newvp)
881 if ((bp->b_flags & B_PAGING) == 0) {
883 "pbreassignbuf() on non phys bp %p",
891 * Reassign a buffer from one vnode to another.
892 * Used to assign file specific control information
893 * (indirect blocks) to the vnode to which they belong.
896 reassignbuf(struct buf *bp, struct vnode *newvp)
901 printf("reassignbuf: NULL");
907 * B_PAGING flagged buffers cannot be reassigned because their vp
908 * is not fully linked in.
910 if (bp->b_flags & B_PAGING)
911 panic("cannot reassign paging buffer");
915 * Delete from old vnode list, if on one.
917 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
918 if (bp->b_xflags & BX_VNDIRTY)
919 buf_rb_tree_RB_REMOVE(&bp->b_vp->v_rbdirty_tree, bp);
921 buf_rb_tree_RB_REMOVE(&bp->b_vp->v_rbclean_tree, bp);
922 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
923 if (bp->b_vp != newvp) {
925 bp->b_vp = NULL; /* for clarification */
929 * If dirty, put on list of dirty buffers; otherwise insert onto list
932 if (bp->b_flags & B_DELWRI) {
933 if ((newvp->v_flag & VONWORKLST) == 0) {
934 switch (newvp->v_type) {
941 newvp->v_rdev->si_mountpoint != NULL) {
949 vn_syncer_add_to_worklist(newvp, delay);
951 bp->b_xflags |= BX_VNDIRTY;
952 if (buf_rb_tree_RB_INSERT(&newvp->v_rbdirty_tree, bp))
953 panic("reassignbuf: dup lblk vp %p bp %p", newvp, bp);
955 bp->b_xflags |= BX_VNCLEAN;
956 if (buf_rb_tree_RB_INSERT(&newvp->v_rbclean_tree, bp))
957 panic("reassignbuf: dup lblk vp %p bp %p", newvp, bp);
958 if ((newvp->v_flag & VONWORKLST) &&
959 RB_EMPTY(&newvp->v_rbdirty_tree)) {
960 newvp->v_flag &= ~VONWORKLST;
961 LIST_REMOVE(newvp, v_synclist);
964 if (bp->b_vp != newvp) {
972 * Create a vnode for a block device.
973 * Used for mounting the root file system.
976 bdevvp(dev_t dev, struct vnode **vpp)
986 error = getspecialvnode(VT_NON, NULL, &spec_vnode_vops, &nvp, 0, 0);
993 vp->v_udev = dev->si_udev;
1000 v_associate_rdev(struct vnode *vp, dev_t dev)
1004 if (dev == NULL || dev == NODEV)
1006 if (dev_is_good(dev) == 0)
1008 KKASSERT(vp->v_rdev == NULL);
1011 vp->v_rdev = reference_dev(dev);
1012 lwkt_gettoken(&ilock, &spechash_token);
1013 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_specnext);
1014 lwkt_reltoken(&ilock);
1019 v_release_rdev(struct vnode *vp)
1024 if ((dev = vp->v_rdev) != NULL) {
1025 lwkt_gettoken(&ilock, &spechash_token);
1026 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_specnext);
1027 if (dev_ref_debug && vp->v_opencount != 0) {
1028 printf("releasing rdev with non-0 "
1029 "v_opencount(%d) (revoked?)\n",
1033 vp->v_opencount = 0;
1035 lwkt_reltoken(&ilock);
1040 * Add a vnode to the alias list hung off the dev_t. We only associate
1041 * the device number with the vnode. The actual device is not associated
1042 * until the vnode is opened (usually in spec_open()), and will be
1043 * disassociated on last close.
1046 addaliasu(struct vnode *nvp, udev_t nvp_udev)
1048 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1049 panic("addaliasu on non-special vnode");
1050 nvp->v_udev = nvp_udev;
1054 * Disassociate a vnode from its underlying filesystem.
1056 * The vnode must be VX locked and refd
1058 * If there are v_usecount references to the vnode other then ours we have
1059 * to VOP_CLOSE the vnode before we can deactivate and reclaim it.
1062 vclean(struct vnode *vp, int flags, struct thread *td)
1067 * If the vnode has already been reclaimed we have nothing to do.
1069 if (vp->v_flag & VRECLAIMED)
1071 vp->v_flag |= VRECLAIMED;
1074 * Scrap the vfs cache
1076 while (cache_inval_vp(vp, 0) != 0) {
1077 printf("Warning: vnode %p clean/cache_resolution race detected\n", vp);
1078 tsleep(vp, 0, "vclninv", 2);
1082 * Check to see if the vnode is in use. If so we have to reference it
1083 * before we clean it out so that its count cannot fall to zero and
1084 * generate a race against ourselves to recycle it.
1086 active = (vp->v_usecount > 1);
1089 * Clean out any buffers associated with the vnode and destroy its
1090 * object, if it has one.
1092 vinvalbuf(vp, V_SAVE, td, 0, 0);
1093 VOP_DESTROYVOBJECT(vp);
1096 * If purging an active vnode, it must be closed and
1097 * deactivated before being reclaimed. XXX
1099 * Note that neither of these routines unlocks the vnode.
1102 if (flags & DOCLOSE)
1103 VOP_CLOSE(vp, FNONBLOCK, td);
1107 * If the vnode has not be deactivated, deactivated it.
1109 if ((vp->v_flag & VINACTIVE) == 0) {
1110 vp->v_flag |= VINACTIVE;
1111 VOP_INACTIVE(vp, td);
1115 * Reclaim the vnode.
1117 if (VOP_RECLAIM(vp, td))
1118 panic("vclean: cannot reclaim");
1121 * Done with purge, notify sleepers of the grim news.
1123 vp->v_ops = &dead_vnode_vops;
1129 * Eliminate all activity associated with the requested vnode
1130 * and with all vnodes aliased to the requested vnode.
1132 * The vnode must be referenced and vx_lock()'d
1134 * revoke { struct vnode *a_vp, int a_flags }
1137 vop_stdrevoke(struct vop_revoke_args *ap)
1139 struct vnode *vp, *vq;
1143 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
1148 * If the vnode is already dead don't try to revoke it
1150 if (vp->v_flag & VRECLAIMED)
1154 * If the vnode has a device association, scrap all vnodes associated
1155 * with the device. Don't let the device disappear on us while we
1156 * are scrapping the vnodes.
1158 * The passed vp will probably show up in the list, do not VX lock
1161 if (vp->v_type != VCHR && vp->v_type != VBLK)
1163 if ((dev = vp->v_rdev) == NULL) {
1164 if ((dev = udev2dev(vp->v_udev, vp->v_type == VBLK)) == NODEV)
1168 lwkt_gettoken(&ilock, &spechash_token);
1169 while ((vq = SLIST_FIRST(&dev->si_hlist)) != NULL) {
1170 if (vp == vq || vx_get(vq) == 0) {
1171 if (vq == SLIST_FIRST(&dev->si_hlist))
1177 lwkt_reltoken(&ilock);
1183 * Recycle an unused vnode to the front of the free list.
1185 * Returns 1 if we were successfully able to recycle the vnode,
1189 vrecycle(struct vnode *vp, struct thread *td)
1191 if (vp->v_usecount == 1) {
1199 * Eliminate all activity associated with a vnode in preparation for reuse.
1201 * The vnode must be VX locked and refd and will remain VX locked and refd
1202 * on return. This routine may be called with the vnode in any state, as
1203 * long as it is VX locked. The vnode will be cleaned out and marked
1204 * VRECLAIMED but will not actually be reused until all existing refs and
1207 * NOTE: This routine may be called on a vnode which has not yet been
1208 * already been deactivated (VOP_INACTIVE), or on a vnode which has
1209 * already been reclaimed.
1211 * This routine is not responsible for placing us back on the freelist.
1212 * Instead, it happens automatically when the caller releases the VX lock
1213 * (assuming there aren't any other references).
1216 vgone(struct vnode *vp)
1219 * assert that the VX lock is held. This is an absolute requirement
1220 * now for vgone() to be called.
1222 KKASSERT(vp->v_lock.lk_exclusivecount == 1);
1225 * Clean out the filesystem specific data and set the VRECLAIMED
1226 * bit. Also deactivate the vnode if necessary.
1228 vclean(vp, DOCLOSE, curthread);
1231 * Delete from old mount point vnode list, if on one.
1233 if (vp->v_mount != NULL)
1234 insmntque(vp, NULL);
1237 * If special device, remove it from special device alias list
1238 * if it is on one. This should normally only occur if a vnode is
1239 * being revoked as the device should otherwise have been released
1242 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
1253 * Lookup a vnode by device number.
1256 vfinddev(dev_t dev, enum vtype type, struct vnode **vpp)
1261 lwkt_gettoken(&ilock, &spechash_token);
1262 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
1263 if (type == vp->v_type) {
1265 lwkt_reltoken(&ilock);
1269 lwkt_reltoken(&ilock);
1274 * Calculate the total number of references to a special device. This
1275 * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1276 * an overloaded field. Since udev2dev can now return NODEV, we have
1277 * to check for a NULL v_rdev.
1280 count_dev(dev_t dev)
1286 if (SLIST_FIRST(&dev->si_hlist)) {
1287 lwkt_gettoken(&ilock, &spechash_token);
1288 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
1289 count += vp->v_usecount;
1291 lwkt_reltoken(&ilock);
1297 count_udev(udev_t udev)
1301 if ((dev = udev2dev(udev, 0)) == NODEV)
1303 return(count_dev(dev));
1307 vcount(struct vnode *vp)
1309 if (vp->v_rdev == NULL)
1311 return(count_dev(vp->v_rdev));
1315 * Print out a description of a vnode.
1317 static char *typename[] =
1318 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1321 vprint(char *label, struct vnode *vp)
1326 printf("%s: %p: ", label, (void *)vp);
1328 printf("%p: ", (void *)vp);
1329 printf("type %s, usecount %d, writecount %d, refcount %d,",
1330 typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1333 if (vp->v_flag & VROOT)
1334 strcat(buf, "|VROOT");
1335 if (vp->v_flag & VTEXT)
1336 strcat(buf, "|VTEXT");
1337 if (vp->v_flag & VSYSTEM)
1338 strcat(buf, "|VSYSTEM");
1339 if (vp->v_flag & VBWAIT)
1340 strcat(buf, "|VBWAIT");
1341 if (vp->v_flag & VFREE)
1342 strcat(buf, "|VFREE");
1343 if (vp->v_flag & VOBJBUF)
1344 strcat(buf, "|VOBJBUF");
1346 printf(" flags (%s)", &buf[1]);
1347 if (vp->v_data == NULL) {
1356 #include <ddb/ddb.h>
1358 static int db_show_locked_vnodes(struct mount *mp, void *data);
1361 * List all of the locked vnodes in the system.
1362 * Called when debugging the kernel.
1364 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
1366 printf("Locked vnodes\n");
1367 mountlist_scan(db_show_locked_vnodes, NULL,
1368 MNTSCAN_FORWARD|MNTSCAN_NOBUSY);
1372 db_show_locked_vnodes(struct mount *mp, void *data __unused)
1376 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
1377 if (VOP_ISLOCKED(vp, NULL))
1378 vprint((char *)0, vp);
1385 * Top level filesystem related information gathering.
1387 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
1390 vfs_sysctl(SYSCTL_HANDLER_ARGS)
1392 int *name = (int *)arg1 - 1; /* XXX */
1393 u_int namelen = arg2 + 1; /* XXX */
1394 struct vfsconf *vfsp;
1396 #if 1 || defined(COMPAT_PRELITE2)
1397 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1399 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
1403 /* all sysctl names at this level are at least name and field */
1405 return (ENOTDIR); /* overloaded */
1406 if (name[0] != VFS_GENERIC) {
1407 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1408 if (vfsp->vfc_typenum == name[0])
1411 return (EOPNOTSUPP);
1412 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1413 oldp, oldlenp, newp, newlen, p));
1417 case VFS_MAXTYPENUM:
1420 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
1423 return (ENOTDIR); /* overloaded */
1424 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1425 if (vfsp->vfc_typenum == name[2])
1428 return (EOPNOTSUPP);
1429 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
1431 return (EOPNOTSUPP);
1434 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
1435 "Generic filesystem");
1437 #if 1 || defined(COMPAT_PRELITE2)
1440 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
1443 struct vfsconf *vfsp;
1444 struct ovfsconf ovfs;
1446 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
1447 bzero(&ovfs, sizeof(ovfs));
1448 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
1449 strcpy(ovfs.vfc_name, vfsp->vfc_name);
1450 ovfs.vfc_index = vfsp->vfc_typenum;
1451 ovfs.vfc_refcount = vfsp->vfc_refcount;
1452 ovfs.vfc_flags = vfsp->vfc_flags;
1453 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
1460 #endif /* 1 || COMPAT_PRELITE2 */
1463 * Check to see if a filesystem is mounted on a block device.
1466 vfs_mountedon(struct vnode *vp)
1470 if ((dev = vp->v_rdev) == NULL)
1471 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK));
1472 if (dev != NODEV && dev->si_mountpoint)
1478 * Unmount all filesystems. The list is traversed in reverse order
1479 * of mounting to avoid dependencies.
1482 static int vfs_umountall_callback(struct mount *mp, void *data);
1485 vfs_unmountall(void)
1487 struct thread *td = curthread;
1490 if (td->td_proc == NULL)
1491 td = initproc->p_thread; /* XXX XXX use proc0 instead? */
1494 count = mountlist_scan(vfs_umountall_callback,
1495 &td, MNTSCAN_REVERSE|MNTSCAN_NOBUSY);
1501 vfs_umountall_callback(struct mount *mp, void *data)
1503 struct thread *td = *(struct thread **)data;
1506 error = dounmount(mp, MNT_FORCE, td);
1508 mountlist_remove(mp);
1509 printf("unmount of filesystem mounted from %s failed (",
1510 mp->mnt_stat.f_mntfromname);
1514 printf("%d)\n", error);
1520 * Build hash lists of net addresses and hang them off the mount point.
1521 * Called by ufs_mount() to set up the lists of export addresses.
1524 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1525 struct export_args *argp)
1528 struct radix_node_head *rnh;
1530 struct radix_node *rn;
1531 struct sockaddr *saddr, *smask = 0;
1535 if (argp->ex_addrlen == 0) {
1536 if (mp->mnt_flag & MNT_DEFEXPORTED)
1538 np = &nep->ne_defexported;
1539 np->netc_exflags = argp->ex_flags;
1540 np->netc_anon = argp->ex_anon;
1541 np->netc_anon.cr_ref = 1;
1542 mp->mnt_flag |= MNT_DEFEXPORTED;
1546 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
1548 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
1551 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1552 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
1553 bzero((caddr_t) np, i);
1554 saddr = (struct sockaddr *) (np + 1);
1555 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
1557 if (saddr->sa_len > argp->ex_addrlen)
1558 saddr->sa_len = argp->ex_addrlen;
1559 if (argp->ex_masklen) {
1560 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1561 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
1564 if (smask->sa_len > argp->ex_masklen)
1565 smask->sa_len = argp->ex_masklen;
1567 i = saddr->sa_family;
1568 if ((rnh = nep->ne_rtable[i]) == 0) {
1570 * Seems silly to initialize every AF when most are not used,
1571 * do so on demand here
1573 SLIST_FOREACH(dom, &domains, dom_next)
1574 if (dom->dom_family == i && dom->dom_rtattach) {
1575 dom->dom_rtattach((void **) &nep->ne_rtable[i],
1579 if ((rnh = nep->ne_rtable[i]) == 0) {
1584 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh,
1586 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */
1590 np->netc_exflags = argp->ex_flags;
1591 np->netc_anon = argp->ex_anon;
1592 np->netc_anon.cr_ref = 1;
1595 free(np, M_NETADDR);
1601 vfs_free_netcred(struct radix_node *rn, void *w)
1603 struct radix_node_head *rnh = (struct radix_node_head *) w;
1605 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
1606 free((caddr_t) rn, M_NETADDR);
1611 * Free the net address hash lists that are hanging off the mount points.
1614 vfs_free_addrlist(struct netexport *nep)
1617 struct radix_node_head *rnh;
1619 for (i = 0; i <= AF_MAX; i++)
1620 if ((rnh = nep->ne_rtable[i])) {
1621 (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
1623 free((caddr_t) rnh, M_RTABLE);
1624 nep->ne_rtable[i] = 0;
1629 vfs_export(struct mount *mp, struct netexport *nep, struct export_args *argp)
1633 if (argp->ex_flags & MNT_DELEXPORT) {
1634 if (mp->mnt_flag & MNT_EXPUBLIC) {
1635 vfs_setpublicfs(NULL, NULL, NULL);
1636 mp->mnt_flag &= ~MNT_EXPUBLIC;
1638 vfs_free_addrlist(nep);
1639 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
1641 if (argp->ex_flags & MNT_EXPORTED) {
1642 if (argp->ex_flags & MNT_EXPUBLIC) {
1643 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
1645 mp->mnt_flag |= MNT_EXPUBLIC;
1647 if ((error = vfs_hang_addrlist(mp, nep, argp)))
1649 mp->mnt_flag |= MNT_EXPORTED;
1656 * Set the publicly exported filesystem (WebNFS). Currently, only
1657 * one public filesystem is possible in the spec (RFC 2054 and 2055)
1660 vfs_setpublicfs(struct mount *mp, struct netexport *nep,
1661 struct export_args *argp)
1668 * mp == NULL -> invalidate the current info, the FS is
1669 * no longer exported. May be called from either vfs_export
1670 * or unmount, so check if it hasn't already been done.
1673 if (nfs_pub.np_valid) {
1674 nfs_pub.np_valid = 0;
1675 if (nfs_pub.np_index != NULL) {
1676 FREE(nfs_pub.np_index, M_TEMP);
1677 nfs_pub.np_index = NULL;
1684 * Only one allowed at a time.
1686 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
1690 * Get real filehandle for root of exported FS.
1692 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
1693 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
1695 if ((error = VFS_ROOT(mp, &rvp)))
1698 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
1704 * If an indexfile was specified, pull it in.
1706 if (argp->ex_indexfile != NULL) {
1707 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
1709 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
1710 MAXNAMLEN, (size_t *)0);
1713 * Check for illegal filenames.
1715 for (cp = nfs_pub.np_index; *cp; cp++) {
1723 FREE(nfs_pub.np_index, M_TEMP);
1728 nfs_pub.np_mount = mp;
1729 nfs_pub.np_valid = 1;
1734 vfs_export_lookup(struct mount *mp, struct netexport *nep,
1735 struct sockaddr *nam)
1738 struct radix_node_head *rnh;
1739 struct sockaddr *saddr;
1742 if (mp->mnt_flag & MNT_EXPORTED) {
1744 * Lookup in the export list first.
1748 rnh = nep->ne_rtable[saddr->sa_family];
1750 np = (struct netcred *)
1751 (*rnh->rnh_matchaddr)((char *)saddr,
1753 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
1758 * If no address match, use the default if it exists.
1760 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
1761 np = &nep->ne_defexported;
1767 * perform msync on all vnodes under a mount point. The mount point must
1768 * be locked. This code is also responsible for lazy-freeing unreferenced
1769 * vnodes whos VM objects no longer contain pages.
1771 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
1773 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
1774 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data);
1777 vfs_msync(struct mount *mp, int flags)
1779 vmntvnodescan(mp, VMSC_REFVP, vfs_msync_scan1, vfs_msync_scan2,
1784 * scan1 is a fast pre-check. There could be hundreds of thousands of
1785 * vnodes, we cannot afford to do anything heavy weight until we have a
1786 * fairly good indication that there is work to do.
1790 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
1792 int flags = (int)data;
1794 if ((vp->v_flag & VRECLAIMED) == 0) {
1795 if (vshouldfree(vp, 0))
1796 return(0); /* call scan2 */
1797 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
1798 (vp->v_flag & VOBJDIRTY) &&
1799 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
1800 return(0); /* call scan2 */
1805 * do not call scan2, continue the loop
1812 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data)
1815 int flags = (int)data;
1817 if (vp->v_flag & VRECLAIMED)
1820 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
1821 (vp->v_flag & VOBJDIRTY) &&
1822 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
1823 if (VOP_GETVOBJECT(vp, &obj) == 0) {
1824 vm_object_page_clean(obj, 0, 0,
1825 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
1832 * Create the VM object needed for VMIO and mmap support. This
1833 * is done for all VREG files in the system. Some filesystems might
1834 * afford the additional metadata buffering capability of the
1835 * VMIO code by making the device node be VMIO mode also.
1837 * vp must be locked when vfs_object_create is called.
1840 vfs_object_create(struct vnode *vp, struct thread *td)
1842 return (VOP_CREATEVOBJECT(vp, td));
1846 * Record a process's interest in events which might happen to
1847 * a vnode. Because poll uses the historic select-style interface
1848 * internally, this routine serves as both the ``check for any
1849 * pending events'' and the ``record my interest in future events''
1850 * functions. (These are done together, while the lock is held,
1851 * to avoid race conditions.)
1854 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
1858 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
1859 if (vp->v_pollinfo.vpi_revents & events) {
1861 * This leaves events we are not interested
1862 * in available for the other process which
1863 * which presumably had requested them
1864 * (otherwise they would never have been
1867 events &= vp->v_pollinfo.vpi_revents;
1868 vp->v_pollinfo.vpi_revents &= ~events;
1870 lwkt_reltoken(&ilock);
1873 vp->v_pollinfo.vpi_events |= events;
1874 selrecord(td, &vp->v_pollinfo.vpi_selinfo);
1875 lwkt_reltoken(&ilock);
1880 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
1881 * it is possible for us to miss an event due to race conditions, but
1882 * that condition is expected to be rare, so for the moment it is the
1883 * preferred interface.
1886 vn_pollevent(struct vnode *vp, int events)
1890 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
1891 if (vp->v_pollinfo.vpi_events & events) {
1893 * We clear vpi_events so that we don't
1894 * call selwakeup() twice if two events are
1895 * posted before the polling process(es) is
1896 * awakened. This also ensures that we take at
1897 * most one selwakeup() if the polling process
1898 * is no longer interested. However, it does
1899 * mean that only one event can be noticed at
1900 * a time. (Perhaps we should only clear those
1901 * event bits which we note?) XXX
1903 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */
1904 vp->v_pollinfo.vpi_revents |= events;
1905 selwakeup(&vp->v_pollinfo.vpi_selinfo);
1907 lwkt_reltoken(&ilock);
1911 * Wake up anyone polling on vp because it is being revoked.
1912 * This depends on dead_poll() returning POLLHUP for correct
1916 vn_pollgone(struct vnode *vp)
1920 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
1921 if (vp->v_pollinfo.vpi_events) {
1922 vp->v_pollinfo.vpi_events = 0;
1923 selwakeup(&vp->v_pollinfo.vpi_selinfo);
1925 lwkt_reltoken(&ilock);
1929 * extract the dev_t from a VBLK or VCHR. The vnode must have been opened
1930 * (or v_rdev might be NULL).
1933 vn_todev(struct vnode *vp)
1935 if (vp->v_type != VBLK && vp->v_type != VCHR)
1937 KKASSERT(vp->v_rdev != NULL);
1938 return (vp->v_rdev);
1942 * Check if vnode represents a disk device. The vnode does not need to be
1946 vn_isdisk(struct vnode *vp, int *errp)
1950 if (vp->v_type != VBLK && vp->v_type != VCHR) {
1956 if ((dev = vp->v_rdev) == NULL)
1957 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK));
1958 if (dev == NULL || dev == NODEV) {
1963 if (dev_is_good(dev) == 0) {
1968 if ((dev_dflags(dev) & D_DISK) == 0) {
1978 #ifdef DEBUG_VFS_LOCKS
1981 assert_vop_locked(struct vnode *vp, const char *str)
1983 if (vp && IS_LOCKING_VFS(vp) && !VOP_ISLOCKED(vp, NULL)) {
1984 panic("%s: %p is not locked shared but should be", str, vp);
1989 assert_vop_unlocked(struct vnode *vp, const char *str)
1991 if (vp && IS_LOCKING_VFS(vp)) {
1992 if (VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE) {
1993 panic("%s: %p is locked but should not be", str, vp);