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 1.67 2006/03/02 19:07:59 dillon Exp $
44 * External virtual filesystem routines
48 #include <sys/param.h>
49 #include <sys/systm.h>
52 #include <sys/dirent.h>
53 #include <sys/domain.h>
54 #include <sys/eventhandler.h>
55 #include <sys/fcntl.h>
56 #include <sys/kernel.h>
57 #include <sys/kthread.h>
58 #include <sys/malloc.h>
60 #include <sys/mount.h>
62 #include <sys/reboot.h>
63 #include <sys/socket.h>
65 #include <sys/sysctl.h>
66 #include <sys/syslog.h>
67 #include <sys/unistd.h>
68 #include <sys/vmmeter.h>
69 #include <sys/vnode.h>
71 #include <machine/limits.h>
74 #include <vm/vm_object.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_kern.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_pager.h>
81 #include <vm/vnode_pager.h>
82 #include <vm/vm_zone.h>
85 #include <sys/thread2.h>
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;
259 vap->va_fsmid = VNOVAL;
263 * Flush out and invalidate all buffers associated with a vnode.
267 static int vinvalbuf_bp(struct buf *bp, void *data);
269 struct vinvalbuf_bp_info {
277 vinvalbuf(struct vnode *vp, int flags, struct thread *td,
278 int slpflag, int slptimeo)
280 struct vinvalbuf_bp_info info;
285 * If we are being asked to save, call fsync to ensure that the inode
288 if (flags & V_SAVE) {
290 while (vp->v_track_write.bk_active) {
291 vp->v_track_write.bk_waitflag = 1;
292 error = tsleep(&vp->v_track_write, slpflag,
293 "vinvlbuf", slptimeo);
299 if (!RB_EMPTY(&vp->v_rbdirty_tree)) {
301 if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0)
304 if (vp->v_track_write.bk_active > 0 ||
305 !RB_EMPTY(&vp->v_rbdirty_tree))
306 panic("vinvalbuf: dirty bufs");
311 info.slptimeo = slptimeo;
312 info.lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
313 if (slpflag & PCATCH)
314 info.lkflags |= LK_PCATCH;
319 * Flush the buffer cache until nothing is left.
321 while (!RB_EMPTY(&vp->v_rbclean_tree) ||
322 !RB_EMPTY(&vp->v_rbdirty_tree)) {
323 error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, NULL,
324 vinvalbuf_bp, &info);
326 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
327 vinvalbuf_bp, &info);
332 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
333 * have write I/O in-progress but if there is a VM object then the
334 * VM object can also have read-I/O in-progress.
337 while (vp->v_track_write.bk_active > 0) {
338 vp->v_track_write.bk_waitflag = 1;
339 tsleep(&vp->v_track_write, 0, "vnvlbv", 0);
341 if (VOP_GETVOBJECT(vp, &object) == 0) {
342 while (object->paging_in_progress)
343 vm_object_pip_sleep(object, "vnvlbx");
345 } while (vp->v_track_write.bk_active > 0);
350 * Destroy the copy in the VM cache, too.
352 if (VOP_GETVOBJECT(vp, &object) == 0) {
353 vm_object_page_remove(object, 0, 0,
354 (flags & V_SAVE) ? TRUE : FALSE);
357 if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree))
358 panic("vinvalbuf: flush failed");
363 vinvalbuf_bp(struct buf *bp, void *data)
365 struct vinvalbuf_bp_info *info = data;
368 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
369 error = BUF_TIMELOCK(bp, info->lkflags,
370 "vinvalbuf", info->slptimeo);
380 * XXX Since there are no node locks for NFS, I
381 * believe there is a slight chance that a delayed
382 * write will occur while sleeping just above, so
383 * check for it. Note that vfs_bio_awrite expects
384 * buffers to reside on a queue, while VOP_BWRITE and
387 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
388 (info->flags & V_SAVE)) {
389 if (bp->b_vp == info->vp) {
390 if (bp->b_flags & B_CLUSTEROK) {
395 bp->b_flags |= B_ASYNC;
396 VOP_BWRITE(bp->b_vp, bp);
400 VOP_BWRITE(bp->b_vp, bp);
404 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
405 bp->b_flags &= ~B_ASYNC;
412 * Truncate a file's buffer and pages to a specified length. This
413 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
416 * The vnode must be locked.
418 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data);
419 static int vtruncbuf_bp_trunc(struct buf *bp, void *data);
420 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data);
421 static int vtruncbuf_bp_metasync(struct buf *bp, void *data);
424 vtruncbuf(struct vnode *vp, struct thread *td, off_t length, int blksize)
430 * Round up to the *next* lbn, then destroy the buffers in question.
431 * Since we are only removing some of the buffers we must rely on the
432 * scan count to determine whether a loop is necessary.
434 trunclbn = (length + blksize - 1) / blksize;
438 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
439 vtruncbuf_bp_trunc_cmp,
440 vtruncbuf_bp_trunc, &trunclbn);
441 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
442 vtruncbuf_bp_trunc_cmp,
443 vtruncbuf_bp_trunc, &trunclbn);
447 * For safety, fsync any remaining metadata if the file is not being
448 * truncated to 0. Since the metadata does not represent the entire
449 * dirty list we have to rely on the hit count to ensure that we get
454 count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
455 vtruncbuf_bp_metasync_cmp,
456 vtruncbuf_bp_metasync, vp);
461 * Wait for any in-progress I/O to complete before returning (why?)
463 while (vp->v_track_write.bk_active > 0) {
464 vp->v_track_write.bk_waitflag = 1;
465 tsleep(&vp->v_track_write, 0, "vbtrunc", 0);
470 vnode_pager_setsize(vp, length);
476 * The callback buffer is beyond the new file EOF and must be destroyed.
477 * Note that the compare function must conform to the RB_SCAN's requirements.
481 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data)
483 if (bp->b_lblkno >= *(daddr_t *)data)
490 vtruncbuf_bp_trunc(struct buf *bp, void *data)
493 * Do not try to use a buffer we cannot immediately lock, but sleep
494 * anyway to prevent a livelock. The code will loop until all buffers
497 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
498 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
502 bp->b_flags |= (B_INVAL | B_RELBUF);
503 bp->b_flags &= ~B_ASYNC;
510 * Fsync all meta-data after truncating a file to be non-zero. Only metadata
511 * blocks (with a negative lblkno) are scanned.
512 * Note that the compare function must conform to the RB_SCAN's requirements.
515 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data)
517 if (bp->b_lblkno < 0)
523 vtruncbuf_bp_metasync(struct buf *bp, void *data)
525 struct vnode *vp = data;
527 if (bp->b_flags & B_DELWRI) {
529 * Do not try to use a buffer we cannot immediately lock,
530 * but sleep anyway to prevent a livelock. The code will
531 * loop until all buffers can be acted upon.
533 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
534 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
538 if (bp->b_vp == vp) {
539 bp->b_flags |= B_ASYNC;
541 bp->b_flags &= ~B_ASYNC;
543 VOP_BWRITE(bp->b_vp, bp);
552 * vfsync - implements a multipass fsync on a file which understands
553 * dependancies and meta-data. The passed vnode must be locked. The
554 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY.
556 * When fsyncing data asynchronously just do one consolidated pass starting
557 * with the most negative block number. This may not get all the data due
560 * When fsyncing data synchronously do a data pass, then a metadata pass,
561 * then do additional data+metadata passes to try to get all the data out.
563 static int vfsync_wait_output(struct vnode *vp,
564 int (*waitoutput)(struct vnode *, struct thread *));
565 static int vfsync_data_only_cmp(struct buf *bp, void *data);
566 static int vfsync_meta_only_cmp(struct buf *bp, void *data);
567 static int vfsync_lazy_range_cmp(struct buf *bp, void *data);
568 static int vfsync_bp(struct buf *bp, void *data);
577 int (*checkdef)(struct buf *);
581 vfsync(struct vnode *vp, int waitfor, int passes, daddr_t lbn,
582 int (*checkdef)(struct buf *),
583 int (*waitoutput)(struct vnode *, struct thread *))
585 struct vfsync_info info;
588 bzero(&info, sizeof(info));
591 if ((info.checkdef = checkdef) == NULL)
599 * Lazy (filesystem syncer typ) Asynchronous plus limit the
600 * number of data (not meta) pages we try to flush to 1MB.
601 * A non-zero return means that lazy limit was reached.
603 info.lazylimit = 1024 * 1024;
605 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
606 vfsync_lazy_range_cmp, vfsync_bp, &info);
607 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
608 vfsync_meta_only_cmp, vfsync_bp, &info);
611 else if (!RB_EMPTY(&vp->v_rbdirty_tree))
612 vn_syncer_add_to_worklist(vp, 1);
617 * Asynchronous. Do a data-only pass and a meta-only pass.
620 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
622 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp,
628 * Synchronous. Do a data-only pass, then a meta-data+data
629 * pass, then additional integrated passes to try to get
630 * all the dependancies flushed.
632 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
634 error = vfsync_wait_output(vp, waitoutput);
636 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
638 error = vfsync_wait_output(vp, waitoutput);
640 while (error == 0 && passes > 0 &&
641 !RB_EMPTY(&vp->v_rbdirty_tree)) {
643 info.synchronous = 1;
646 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
652 error = vfsync_wait_output(vp, waitoutput);
661 vfsync_wait_output(struct vnode *vp, int (*waitoutput)(struct vnode *, struct thread *))
665 while (vp->v_track_write.bk_active) {
666 vp->v_track_write.bk_waitflag = 1;
667 tsleep(&vp->v_track_write, 0, "fsfsn", 0);
670 error = waitoutput(vp, curthread);
675 vfsync_data_only_cmp(struct buf *bp, void *data)
677 if (bp->b_lblkno < 0)
683 vfsync_meta_only_cmp(struct buf *bp, void *data)
685 if (bp->b_lblkno < 0)
691 vfsync_lazy_range_cmp(struct buf *bp, void *data)
693 struct vfsync_info *info = data;
694 if (bp->b_lblkno < info->vp->v_lazyw)
700 vfsync_bp(struct buf *bp, void *data)
702 struct vfsync_info *info = data;
703 struct vnode *vp = info->vp;
707 * if syncdeps is not set we do not try to write buffers which have
710 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp))
714 * Ignore buffers that we cannot immediately lock. XXX
716 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
718 if ((bp->b_flags & B_DELWRI) == 0)
719 panic("vfsync_bp: buffer not dirty");
721 panic("vfsync_bp: buffer vp mismatch");
724 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
725 * has been written but an additional handshake with the device
726 * is required before we can dispose of the buffer. We have no idea
727 * how to do this so we have to skip these buffers.
729 if (bp->b_flags & B_NEEDCOMMIT) {
735 * (LEGACY FROM UFS, REMOVE WHEN POSSIBLE) - invalidate any dirty
736 * buffers beyond the file EOF.
738 if (info->lbn != (daddr_t)-1 && vp->v_type == VREG &&
739 bp->b_lblkno >= info->lbn) {
741 bp->b_flags |= B_INVAL | B_NOCACHE;
747 if (info->synchronous) {
749 * Synchronous flushing. An error may be returned.
757 * Asynchronous flushing. A negative return value simply
758 * stops the scan and is not considered an error. We use
759 * this to support limited MNT_LAZY flushes.
761 vp->v_lazyw = bp->b_lblkno;
762 if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) {
764 info->lazycount += vfs_bio_awrite(bp);
766 info->lazycount += bp->b_bufsize;
772 if (info->lazylimit && info->lazycount >= info->lazylimit)
781 * Associate a buffer with a vnode.
784 bgetvp(struct vnode *vp, struct buf *bp)
786 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
791 * Insert onto list for new vnode.
794 bp->b_xflags |= BX_VNCLEAN;
795 bp->b_xflags &= ~BX_VNDIRTY;
796 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp))
797 panic("reassignbuf: dup lblk vp %p bp %p", vp, bp);
802 * Disassociate a buffer from a vnode.
805 brelvp(struct buf *bp)
809 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
812 * Delete from old vnode list, if on one.
816 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
817 if (bp->b_xflags & BX_VNDIRTY)
818 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
820 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
821 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
823 if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree)) {
824 vp->v_flag &= ~VONWORKLST;
825 LIST_REMOVE(vp, v_synclist);
833 * Associate a p-buffer with a vnode.
835 * Also sets B_PAGING flag to indicate that vnode is not fully associated
836 * with the buffer. i.e. the bp has not been linked into the vnode or
840 pbgetvp(struct vnode *vp, struct buf *bp)
842 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
845 bp->b_flags |= B_PAGING;
849 * Disassociate a p-buffer from a vnode.
852 pbrelvp(struct buf *bp)
854 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
857 bp->b_flags &= ~B_PAGING;
861 pbreassignbuf(struct buf *bp, struct vnode *newvp)
863 if ((bp->b_flags & B_PAGING) == 0) {
865 "pbreassignbuf() on non phys bp %p",
873 * Reassign a buffer from one vnode to another.
874 * Used to assign file specific control information
875 * (indirect blocks) to the vnode to which they belong.
878 reassignbuf(struct buf *bp, struct vnode *newvp)
883 printf("reassignbuf: NULL");
889 * B_PAGING flagged buffers cannot be reassigned because their vp
890 * is not fully linked in.
892 if (bp->b_flags & B_PAGING)
893 panic("cannot reassign paging buffer");
897 * Delete from old vnode list, if on one.
899 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
900 if (bp->b_xflags & BX_VNDIRTY)
901 buf_rb_tree_RB_REMOVE(&bp->b_vp->v_rbdirty_tree, bp);
903 buf_rb_tree_RB_REMOVE(&bp->b_vp->v_rbclean_tree, bp);
904 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
905 if (bp->b_vp != newvp) {
907 bp->b_vp = NULL; /* for clarification */
911 * If dirty, put on list of dirty buffers; otherwise insert onto list
914 if (bp->b_flags & B_DELWRI) {
915 if ((newvp->v_flag & VONWORKLST) == 0) {
916 switch (newvp->v_type) {
923 newvp->v_rdev->si_mountpoint != NULL) {
931 vn_syncer_add_to_worklist(newvp, delay);
933 bp->b_xflags |= BX_VNDIRTY;
934 if (buf_rb_tree_RB_INSERT(&newvp->v_rbdirty_tree, bp))
935 panic("reassignbuf: dup lblk vp %p bp %p", newvp, bp);
937 bp->b_xflags |= BX_VNCLEAN;
938 if (buf_rb_tree_RB_INSERT(&newvp->v_rbclean_tree, bp))
939 panic("reassignbuf: dup lblk vp %p bp %p", newvp, bp);
940 if ((newvp->v_flag & VONWORKLST) &&
941 RB_EMPTY(&newvp->v_rbdirty_tree)) {
942 newvp->v_flag &= ~VONWORKLST;
943 LIST_REMOVE(newvp, v_synclist);
946 if (bp->b_vp != newvp) {
954 * Create a vnode for a block device.
955 * Used for mounting the root file system.
958 bdevvp(dev_t dev, struct vnode **vpp)
968 error = getspecialvnode(VT_NON, NULL, &spec_vnode_vops, &nvp, 0, 0);
975 vp->v_udev = dev->si_udev;
982 v_associate_rdev(struct vnode *vp, dev_t dev)
986 if (dev == NULL || dev == NODEV)
988 if (dev_is_good(dev) == 0)
990 KKASSERT(vp->v_rdev == NULL);
993 vp->v_rdev = reference_dev(dev);
994 lwkt_gettoken(&ilock, &spechash_token);
995 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_specnext);
996 lwkt_reltoken(&ilock);
1001 v_release_rdev(struct vnode *vp)
1006 if ((dev = vp->v_rdev) != NULL) {
1007 lwkt_gettoken(&ilock, &spechash_token);
1008 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_specnext);
1009 if (dev_ref_debug && vp->v_opencount != 0) {
1010 printf("releasing rdev with non-0 "
1011 "v_opencount(%d) (revoked?)\n",
1015 vp->v_opencount = 0;
1017 lwkt_reltoken(&ilock);
1022 * Add a vnode to the alias list hung off the dev_t. We only associate
1023 * the device number with the vnode. The actual device is not associated
1024 * until the vnode is opened (usually in spec_open()), and will be
1025 * disassociated on last close.
1028 addaliasu(struct vnode *nvp, udev_t nvp_udev)
1030 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1031 panic("addaliasu on non-special vnode");
1032 nvp->v_udev = nvp_udev;
1036 * Disassociate a vnode from its underlying filesystem.
1038 * The vnode must be VX locked and refd
1040 * If there are v_usecount references to the vnode other then ours we have
1041 * to VOP_CLOSE the vnode before we can deactivate and reclaim it.
1044 vclean(struct vnode *vp, int flags, struct thread *td)
1050 * If the vnode has already been reclaimed we have nothing to do.
1052 if (vp->v_flag & VRECLAIMED)
1054 vp->v_flag |= VRECLAIMED;
1057 * Scrap the vfs cache
1059 while (cache_inval_vp(vp, 0, &retflags) != 0) {
1060 printf("Warning: vnode %p clean/cache_resolution race detected\n", vp);
1061 tsleep(vp, 0, "vclninv", 2);
1065 * Check to see if the vnode is in use. If so we have to reference it
1066 * before we clean it out so that its count cannot fall to zero and
1067 * generate a race against ourselves to recycle it.
1069 active = (vp->v_usecount > 1);
1072 * Clean out any buffers associated with the vnode and destroy its
1073 * object, if it has one.
1075 vinvalbuf(vp, V_SAVE, td, 0, 0);
1076 VOP_DESTROYVOBJECT(vp);
1079 * If purging an active vnode, it must be closed and
1080 * deactivated before being reclaimed. XXX
1082 * Note that neither of these routines unlocks the vnode.
1085 if (flags & DOCLOSE)
1086 VOP_CLOSE(vp, FNONBLOCK, td);
1090 * If the vnode has not be deactivated, deactivated it.
1092 if ((vp->v_flag & VINACTIVE) == 0) {
1093 vp->v_flag |= VINACTIVE;
1094 VOP_INACTIVE(vp, td);
1098 * Reclaim the vnode.
1100 if (VOP_RECLAIM(vp, retflags, td))
1101 panic("vclean: cannot reclaim");
1104 * Done with purge, notify sleepers of the grim news.
1106 vp->v_ops = &dead_vnode_vops;
1112 * Eliminate all activity associated with the requested vnode
1113 * and with all vnodes aliased to the requested vnode.
1115 * The vnode must be referenced and vx_lock()'d
1117 * revoke { struct vnode *a_vp, int a_flags }
1120 vop_stdrevoke(struct vop_revoke_args *ap)
1122 struct vnode *vp, *vq;
1126 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
1131 * If the vnode is already dead don't try to revoke it
1133 if (vp->v_flag & VRECLAIMED)
1137 * If the vnode has a device association, scrap all vnodes associated
1138 * with the device. Don't let the device disappear on us while we
1139 * are scrapping the vnodes.
1141 * The passed vp will probably show up in the list, do not VX lock
1144 if (vp->v_type != VCHR && vp->v_type != VBLK)
1146 if ((dev = vp->v_rdev) == NULL) {
1147 if ((dev = udev2dev(vp->v_udev, vp->v_type == VBLK)) == NODEV)
1151 lwkt_gettoken(&ilock, &spechash_token);
1152 while ((vq = SLIST_FIRST(&dev->si_hlist)) != NULL) {
1153 if (vp == vq || vx_get(vq) == 0) {
1154 if (vq == SLIST_FIRST(&dev->si_hlist))
1160 lwkt_reltoken(&ilock);
1166 * Recycle an unused vnode to the front of the free list.
1168 * Returns 1 if we were successfully able to recycle the vnode,
1172 vrecycle(struct vnode *vp, struct thread *td)
1174 if (vp->v_usecount == 1) {
1182 * Eliminate all activity associated with a vnode in preparation for reuse.
1184 * The vnode must be VX locked and refd and will remain VX locked and refd
1185 * on return. This routine may be called with the vnode in any state, as
1186 * long as it is VX locked. The vnode will be cleaned out and marked
1187 * VRECLAIMED but will not actually be reused until all existing refs and
1190 * NOTE: This routine may be called on a vnode which has not yet been
1191 * already been deactivated (VOP_INACTIVE), or on a vnode which has
1192 * already been reclaimed.
1194 * This routine is not responsible for placing us back on the freelist.
1195 * Instead, it happens automatically when the caller releases the VX lock
1196 * (assuming there aren't any other references).
1199 vgone(struct vnode *vp)
1202 * assert that the VX lock is held. This is an absolute requirement
1203 * now for vgone() to be called.
1205 KKASSERT(vp->v_lock.lk_exclusivecount == 1);
1208 * Clean out the filesystem specific data and set the VRECLAIMED
1209 * bit. Also deactivate the vnode if necessary.
1211 vclean(vp, DOCLOSE, curthread);
1214 * Delete from old mount point vnode list, if on one.
1216 if (vp->v_mount != NULL)
1217 insmntque(vp, NULL);
1220 * If special device, remove it from special device alias list
1221 * if it is on one. This should normally only occur if a vnode is
1222 * being revoked as the device should otherwise have been released
1225 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
1236 * Lookup a vnode by device number.
1239 vfinddev(dev_t dev, enum vtype type, struct vnode **vpp)
1244 lwkt_gettoken(&ilock, &spechash_token);
1245 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
1246 if (type == vp->v_type) {
1248 lwkt_reltoken(&ilock);
1252 lwkt_reltoken(&ilock);
1257 * Calculate the total number of references to a special device. This
1258 * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1259 * an overloaded field. Since udev2dev can now return NODEV, we have
1260 * to check for a NULL v_rdev.
1263 count_dev(dev_t dev)
1269 if (SLIST_FIRST(&dev->si_hlist)) {
1270 lwkt_gettoken(&ilock, &spechash_token);
1271 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
1272 count += vp->v_usecount;
1274 lwkt_reltoken(&ilock);
1280 count_udev(udev_t udev)
1284 if ((dev = udev2dev(udev, 0)) == NODEV)
1286 return(count_dev(dev));
1290 vcount(struct vnode *vp)
1292 if (vp->v_rdev == NULL)
1294 return(count_dev(vp->v_rdev));
1298 * Print out a description of a vnode.
1300 static char *typename[] =
1301 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1304 vprint(char *label, struct vnode *vp)
1309 printf("%s: %p: ", label, (void *)vp);
1311 printf("%p: ", (void *)vp);
1312 printf("type %s, usecount %d, writecount %d, refcount %d,",
1313 typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1316 if (vp->v_flag & VROOT)
1317 strcat(buf, "|VROOT");
1318 if (vp->v_flag & VTEXT)
1319 strcat(buf, "|VTEXT");
1320 if (vp->v_flag & VSYSTEM)
1321 strcat(buf, "|VSYSTEM");
1322 if (vp->v_flag & VFREE)
1323 strcat(buf, "|VFREE");
1324 if (vp->v_flag & VOBJBUF)
1325 strcat(buf, "|VOBJBUF");
1327 printf(" flags (%s)", &buf[1]);
1328 if (vp->v_data == NULL) {
1337 #include <ddb/ddb.h>
1339 static int db_show_locked_vnodes(struct mount *mp, void *data);
1342 * List all of the locked vnodes in the system.
1343 * Called when debugging the kernel.
1345 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
1347 printf("Locked vnodes\n");
1348 mountlist_scan(db_show_locked_vnodes, NULL,
1349 MNTSCAN_FORWARD|MNTSCAN_NOBUSY);
1353 db_show_locked_vnodes(struct mount *mp, void *data __unused)
1357 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
1358 if (VOP_ISLOCKED(vp, NULL))
1359 vprint((char *)0, vp);
1366 * Top level filesystem related information gathering.
1368 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
1371 vfs_sysctl(SYSCTL_HANDLER_ARGS)
1373 int *name = (int *)arg1 - 1; /* XXX */
1374 u_int namelen = arg2 + 1; /* XXX */
1375 struct vfsconf *vfsp;
1377 #if 1 || defined(COMPAT_PRELITE2)
1378 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1380 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
1384 /* all sysctl names at this level are at least name and field */
1386 return (ENOTDIR); /* overloaded */
1387 if (name[0] != VFS_GENERIC) {
1388 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1389 if (vfsp->vfc_typenum == name[0])
1392 return (EOPNOTSUPP);
1393 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1394 oldp, oldlenp, newp, newlen, p));
1398 case VFS_MAXTYPENUM:
1401 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
1404 return (ENOTDIR); /* overloaded */
1405 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1406 if (vfsp->vfc_typenum == name[2])
1409 return (EOPNOTSUPP);
1410 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
1412 return (EOPNOTSUPP);
1415 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
1416 "Generic filesystem");
1418 #if 1 || defined(COMPAT_PRELITE2)
1421 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
1424 struct vfsconf *vfsp;
1425 struct ovfsconf ovfs;
1427 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
1428 bzero(&ovfs, sizeof(ovfs));
1429 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
1430 strcpy(ovfs.vfc_name, vfsp->vfc_name);
1431 ovfs.vfc_index = vfsp->vfc_typenum;
1432 ovfs.vfc_refcount = vfsp->vfc_refcount;
1433 ovfs.vfc_flags = vfsp->vfc_flags;
1434 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
1441 #endif /* 1 || COMPAT_PRELITE2 */
1444 * Check to see if a filesystem is mounted on a block device.
1447 vfs_mountedon(struct vnode *vp)
1451 if ((dev = vp->v_rdev) == NULL)
1452 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK));
1453 if (dev != NODEV && dev->si_mountpoint)
1459 * Unmount all filesystems. The list is traversed in reverse order
1460 * of mounting to avoid dependencies.
1463 static int vfs_umountall_callback(struct mount *mp, void *data);
1466 vfs_unmountall(void)
1468 struct thread *td = curthread;
1471 if (td->td_proc == NULL)
1472 td = initproc->p_thread; /* XXX XXX use proc0 instead? */
1475 count = mountlist_scan(vfs_umountall_callback,
1476 &td, MNTSCAN_REVERSE|MNTSCAN_NOBUSY);
1482 vfs_umountall_callback(struct mount *mp, void *data)
1484 struct thread *td = *(struct thread **)data;
1487 error = dounmount(mp, MNT_FORCE, td);
1489 mountlist_remove(mp);
1490 printf("unmount of filesystem mounted from %s failed (",
1491 mp->mnt_stat.f_mntfromname);
1495 printf("%d)\n", error);
1501 * Build hash lists of net addresses and hang them off the mount point.
1502 * Called by ufs_mount() to set up the lists of export addresses.
1505 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1506 struct export_args *argp)
1509 struct radix_node_head *rnh;
1511 struct radix_node *rn;
1512 struct sockaddr *saddr, *smask = 0;
1516 if (argp->ex_addrlen == 0) {
1517 if (mp->mnt_flag & MNT_DEFEXPORTED)
1519 np = &nep->ne_defexported;
1520 np->netc_exflags = argp->ex_flags;
1521 np->netc_anon = argp->ex_anon;
1522 np->netc_anon.cr_ref = 1;
1523 mp->mnt_flag |= MNT_DEFEXPORTED;
1527 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
1529 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
1532 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1533 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
1534 bzero((caddr_t) np, i);
1535 saddr = (struct sockaddr *) (np + 1);
1536 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
1538 if (saddr->sa_len > argp->ex_addrlen)
1539 saddr->sa_len = argp->ex_addrlen;
1540 if (argp->ex_masklen) {
1541 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1542 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
1545 if (smask->sa_len > argp->ex_masklen)
1546 smask->sa_len = argp->ex_masklen;
1548 i = saddr->sa_family;
1549 if ((rnh = nep->ne_rtable[i]) == 0) {
1551 * Seems silly to initialize every AF when most are not used,
1552 * do so on demand here
1554 SLIST_FOREACH(dom, &domains, dom_next)
1555 if (dom->dom_family == i && dom->dom_rtattach) {
1556 dom->dom_rtattach((void **) &nep->ne_rtable[i],
1560 if ((rnh = nep->ne_rtable[i]) == 0) {
1565 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh,
1567 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */
1571 np->netc_exflags = argp->ex_flags;
1572 np->netc_anon = argp->ex_anon;
1573 np->netc_anon.cr_ref = 1;
1576 free(np, M_NETADDR);
1582 vfs_free_netcred(struct radix_node *rn, void *w)
1584 struct radix_node_head *rnh = (struct radix_node_head *) w;
1586 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
1587 free((caddr_t) rn, M_NETADDR);
1592 * Free the net address hash lists that are hanging off the mount points.
1595 vfs_free_addrlist(struct netexport *nep)
1598 struct radix_node_head *rnh;
1600 for (i = 0; i <= AF_MAX; i++)
1601 if ((rnh = nep->ne_rtable[i])) {
1602 (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
1604 free((caddr_t) rnh, M_RTABLE);
1605 nep->ne_rtable[i] = 0;
1610 vfs_export(struct mount *mp, struct netexport *nep, struct export_args *argp)
1614 if (argp->ex_flags & MNT_DELEXPORT) {
1615 if (mp->mnt_flag & MNT_EXPUBLIC) {
1616 vfs_setpublicfs(NULL, NULL, NULL);
1617 mp->mnt_flag &= ~MNT_EXPUBLIC;
1619 vfs_free_addrlist(nep);
1620 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
1622 if (argp->ex_flags & MNT_EXPORTED) {
1623 if (argp->ex_flags & MNT_EXPUBLIC) {
1624 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
1626 mp->mnt_flag |= MNT_EXPUBLIC;
1628 if ((error = vfs_hang_addrlist(mp, nep, argp)))
1630 mp->mnt_flag |= MNT_EXPORTED;
1637 * Set the publicly exported filesystem (WebNFS). Currently, only
1638 * one public filesystem is possible in the spec (RFC 2054 and 2055)
1641 vfs_setpublicfs(struct mount *mp, struct netexport *nep,
1642 struct export_args *argp)
1649 * mp == NULL -> invalidate the current info, the FS is
1650 * no longer exported. May be called from either vfs_export
1651 * or unmount, so check if it hasn't already been done.
1654 if (nfs_pub.np_valid) {
1655 nfs_pub.np_valid = 0;
1656 if (nfs_pub.np_index != NULL) {
1657 FREE(nfs_pub.np_index, M_TEMP);
1658 nfs_pub.np_index = NULL;
1665 * Only one allowed at a time.
1667 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
1671 * Get real filehandle for root of exported FS.
1673 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
1674 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
1676 if ((error = VFS_ROOT(mp, &rvp)))
1679 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
1685 * If an indexfile was specified, pull it in.
1687 if (argp->ex_indexfile != NULL) {
1690 error = vn_get_namelen(rvp, &namelen);
1693 MALLOC(nfs_pub.np_index, char *, namelen, M_TEMP,
1695 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
1696 namelen, (size_t *)0);
1699 * Check for illegal filenames.
1701 for (cp = nfs_pub.np_index; *cp; cp++) {
1709 FREE(nfs_pub.np_index, M_TEMP);
1714 nfs_pub.np_mount = mp;
1715 nfs_pub.np_valid = 1;
1720 vfs_export_lookup(struct mount *mp, struct netexport *nep,
1721 struct sockaddr *nam)
1724 struct radix_node_head *rnh;
1725 struct sockaddr *saddr;
1728 if (mp->mnt_flag & MNT_EXPORTED) {
1730 * Lookup in the export list first.
1734 rnh = nep->ne_rtable[saddr->sa_family];
1736 np = (struct netcred *)
1737 (*rnh->rnh_matchaddr)((char *)saddr,
1739 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
1744 * If no address match, use the default if it exists.
1746 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
1747 np = &nep->ne_defexported;
1753 * perform msync on all vnodes under a mount point. The mount point must
1754 * be locked. This code is also responsible for lazy-freeing unreferenced
1755 * vnodes whos VM objects no longer contain pages.
1757 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
1759 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked,
1760 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it
1761 * way up in this high level function.
1763 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
1764 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data);
1767 vfs_msync(struct mount *mp, int flags)
1771 vmsc_flags = VMSC_GETVP;
1772 if (flags != MNT_WAIT)
1773 vmsc_flags |= VMSC_NOWAIT;
1774 vmntvnodescan(mp, vmsc_flags, vfs_msync_scan1, vfs_msync_scan2,
1779 * scan1 is a fast pre-check. There could be hundreds of thousands of
1780 * vnodes, we cannot afford to do anything heavy weight until we have a
1781 * fairly good indication that there is work to do.
1785 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
1787 int flags = (int)data;
1789 if ((vp->v_flag & VRECLAIMED) == 0) {
1790 if (vshouldfree(vp, 0))
1791 return(0); /* call scan2 */
1792 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
1793 (vp->v_flag & VOBJDIRTY) &&
1794 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
1795 return(0); /* call scan2 */
1800 * do not call scan2, continue the loop
1806 * This callback is handed a locked vnode.
1810 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data)
1813 int flags = (int)data;
1815 if (vp->v_flag & VRECLAIMED)
1818 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
1819 (vp->v_flag & VOBJDIRTY)) {
1820 if (VOP_GETVOBJECT(vp, &obj) == 0) {
1821 vm_object_page_clean(obj, 0, 0,
1822 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
1829 * Create the VM object needed for VMIO and mmap support. This
1830 * is done for all VREG files in the system. Some filesystems might
1831 * afford the additional metadata buffering capability of the
1832 * VMIO code by making the device node be VMIO mode also.
1834 * vp must be locked when vfs_object_create is called.
1837 vfs_object_create(struct vnode *vp, struct thread *td)
1839 return (VOP_CREATEVOBJECT(vp, td));
1843 * Record a process's interest in events which might happen to
1844 * a vnode. Because poll uses the historic select-style interface
1845 * internally, this routine serves as both the ``check for any
1846 * pending events'' and the ``record my interest in future events''
1847 * functions. (These are done together, while the lock is held,
1848 * to avoid race conditions.)
1851 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
1855 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
1856 if (vp->v_pollinfo.vpi_revents & events) {
1858 * This leaves events we are not interested
1859 * in available for the other process which
1860 * which presumably had requested them
1861 * (otherwise they would never have been
1864 events &= vp->v_pollinfo.vpi_revents;
1865 vp->v_pollinfo.vpi_revents &= ~events;
1867 lwkt_reltoken(&ilock);
1870 vp->v_pollinfo.vpi_events |= events;
1871 selrecord(td, &vp->v_pollinfo.vpi_selinfo);
1872 lwkt_reltoken(&ilock);
1877 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
1878 * it is possible for us to miss an event due to race conditions, but
1879 * that condition is expected to be rare, so for the moment it is the
1880 * preferred interface.
1883 vn_pollevent(struct vnode *vp, int events)
1887 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
1888 if (vp->v_pollinfo.vpi_events & events) {
1890 * We clear vpi_events so that we don't
1891 * call selwakeup() twice if two events are
1892 * posted before the polling process(es) is
1893 * awakened. This also ensures that we take at
1894 * most one selwakeup() if the polling process
1895 * is no longer interested. However, it does
1896 * mean that only one event can be noticed at
1897 * a time. (Perhaps we should only clear those
1898 * event bits which we note?) XXX
1900 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */
1901 vp->v_pollinfo.vpi_revents |= events;
1902 selwakeup(&vp->v_pollinfo.vpi_selinfo);
1904 lwkt_reltoken(&ilock);
1908 * Wake up anyone polling on vp because it is being revoked.
1909 * This depends on dead_poll() returning POLLHUP for correct
1913 vn_pollgone(struct vnode *vp)
1917 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
1918 if (vp->v_pollinfo.vpi_events) {
1919 vp->v_pollinfo.vpi_events = 0;
1920 selwakeup(&vp->v_pollinfo.vpi_selinfo);
1922 lwkt_reltoken(&ilock);
1926 * extract the dev_t from a VBLK or VCHR. The vnode must have been opened
1927 * (or v_rdev might be NULL).
1930 vn_todev(struct vnode *vp)
1932 if (vp->v_type != VBLK && vp->v_type != VCHR)
1934 KKASSERT(vp->v_rdev != NULL);
1935 return (vp->v_rdev);
1939 * Check if vnode represents a disk device. The vnode does not need to be
1943 vn_isdisk(struct vnode *vp, int *errp)
1947 if (vp->v_type != VBLK && vp->v_type != VCHR) {
1953 if ((dev = vp->v_rdev) == NULL)
1954 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK));
1955 if (dev == NULL || dev == NODEV) {
1960 if (dev_is_good(dev) == 0) {
1965 if ((dev_dflags(dev) & D_DISK) == 0) {
1975 #ifdef DEBUG_VFS_LOCKS
1978 assert_vop_locked(struct vnode *vp, const char *str)
1980 if (vp && IS_LOCKING_VFS(vp) && !VOP_ISLOCKED(vp, NULL)) {
1981 panic("%s: %p is not locked shared but should be", str, vp);
1986 assert_vop_unlocked(struct vnode *vp, const char *str)
1988 if (vp && IS_LOCKING_VFS(vp)) {
1989 if (VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE) {
1990 panic("%s: %p is locked but should not be", str, vp);
1998 vn_get_namelen(struct vnode *vp, int *namelen)
2000 int error, retval[2];
2002 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval);
2010 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type,
2011 uint16_t d_namlen, const char *d_name)
2016 len = _DIRENT_RECLEN(d_namlen);
2017 if (len > uio->uio_resid)
2020 dp = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
2023 dp->d_namlen = d_namlen;
2024 dp->d_type = d_type;
2025 bcopy(d_name, dp->d_name, d_namlen);
2027 *error = uiomove((caddr_t)dp, len, uio);