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 1.66 2006/02/17 19:18:06 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.slpflag = slpflag;
317 * Flush the buffer cache until nothing is left.
319 while (!RB_EMPTY(&vp->v_rbclean_tree) ||
320 !RB_EMPTY(&vp->v_rbdirty_tree)) {
321 error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, NULL,
322 vinvalbuf_bp, &info);
324 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
325 vinvalbuf_bp, &info);
330 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
331 * have write I/O in-progress but if there is a VM object then the
332 * VM object can also have read-I/O in-progress.
335 while (vp->v_track_write.bk_active > 0) {
336 vp->v_track_write.bk_waitflag = 1;
337 tsleep(&vp->v_track_write, 0, "vnvlbv", 0);
339 if (VOP_GETVOBJECT(vp, &object) == 0) {
340 while (object->paging_in_progress)
341 vm_object_pip_sleep(object, "vnvlbx");
343 } while (vp->v_track_write.bk_active > 0);
348 * Destroy the copy in the VM cache, too.
350 if (VOP_GETVOBJECT(vp, &object) == 0) {
351 vm_object_page_remove(object, 0, 0,
352 (flags & V_SAVE) ? TRUE : FALSE);
355 if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree))
356 panic("vinvalbuf: flush failed");
361 vinvalbuf_bp(struct buf *bp, void *data)
363 struct vinvalbuf_bp_info *info = data;
366 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
367 error = BUF_TIMELOCK(bp,
368 LK_EXCLUSIVE | LK_SLEEPFAIL,
369 "vinvalbuf", info->slpflag, info->slptimeo);
379 * XXX Since there are no node locks for NFS, I
380 * believe there is a slight chance that a delayed
381 * write will occur while sleeping just above, so
382 * check for it. Note that vfs_bio_awrite expects
383 * buffers to reside on a queue, while VOP_BWRITE and
386 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
387 (info->flags & V_SAVE)) {
388 if (bp->b_vp == info->vp) {
389 if (bp->b_flags & B_CLUSTEROK) {
394 bp->b_flags |= B_ASYNC;
395 VOP_BWRITE(bp->b_vp, bp);
399 VOP_BWRITE(bp->b_vp, bp);
403 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
404 bp->b_flags &= ~B_ASYNC;
411 * Truncate a file's buffer and pages to a specified length. This
412 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
415 * The vnode must be locked.
417 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data);
418 static int vtruncbuf_bp_trunc(struct buf *bp, void *data);
419 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data);
420 static int vtruncbuf_bp_metasync(struct buf *bp, void *data);
423 vtruncbuf(struct vnode *vp, struct thread *td, off_t length, int blksize)
429 * Round up to the *next* lbn, then destroy the buffers in question.
430 * Since we are only removing some of the buffers we must rely on the
431 * scan count to determine whether a loop is necessary.
433 trunclbn = (length + blksize - 1) / blksize;
437 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
438 vtruncbuf_bp_trunc_cmp,
439 vtruncbuf_bp_trunc, &trunclbn);
440 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
441 vtruncbuf_bp_trunc_cmp,
442 vtruncbuf_bp_trunc, &trunclbn);
446 * For safety, fsync any remaining metadata if the file is not being
447 * truncated to 0. Since the metadata does not represent the entire
448 * dirty list we have to rely on the hit count to ensure that we get
453 count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
454 vtruncbuf_bp_metasync_cmp,
455 vtruncbuf_bp_metasync, vp);
460 * Wait for any in-progress I/O to complete before returning (why?)
462 while (vp->v_track_write.bk_active > 0) {
463 vp->v_track_write.bk_waitflag = 1;
464 tsleep(&vp->v_track_write, 0, "vbtrunc", 0);
469 vnode_pager_setsize(vp, length);
475 * The callback buffer is beyond the new file EOF and must be destroyed.
476 * Note that the compare function must conform to the RB_SCAN's requirements.
480 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data)
482 if (bp->b_lblkno >= *(daddr_t *)data)
489 vtruncbuf_bp_trunc(struct buf *bp, void *data)
492 * Do not try to use a buffer we cannot immediately lock, but sleep
493 * anyway to prevent a livelock. The code will loop until all buffers
496 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
497 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
501 bp->b_flags |= (B_INVAL | B_RELBUF);
502 bp->b_flags &= ~B_ASYNC;
509 * Fsync all meta-data after truncating a file to be non-zero. Only metadata
510 * blocks (with a negative lblkno) are scanned.
511 * Note that the compare function must conform to the RB_SCAN's requirements.
514 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data)
516 if (bp->b_lblkno < 0)
522 vtruncbuf_bp_metasync(struct buf *bp, void *data)
524 struct vnode *vp = data;
526 if (bp->b_flags & B_DELWRI) {
528 * Do not try to use a buffer we cannot immediately lock,
529 * but sleep anyway to prevent a livelock. The code will
530 * loop until all buffers can be acted upon.
532 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
533 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
537 if (bp->b_vp == vp) {
538 bp->b_flags |= B_ASYNC;
540 bp->b_flags &= ~B_ASYNC;
542 VOP_BWRITE(bp->b_vp, bp);
551 * vfsync - implements a multipass fsync on a file which understands
552 * dependancies and meta-data. The passed vnode must be locked. The
553 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY.
555 * When fsyncing data asynchronously just do one consolidated pass starting
556 * with the most negative block number. This may not get all the data due
559 * When fsyncing data synchronously do a data pass, then a metadata pass,
560 * then do additional data+metadata passes to try to get all the data out.
562 static int vfsync_wait_output(struct vnode *vp,
563 int (*waitoutput)(struct vnode *, struct thread *));
564 static int vfsync_data_only_cmp(struct buf *bp, void *data);
565 static int vfsync_meta_only_cmp(struct buf *bp, void *data);
566 static int vfsync_lazy_range_cmp(struct buf *bp, void *data);
567 static int vfsync_bp(struct buf *bp, void *data);
576 int (*checkdef)(struct buf *);
580 vfsync(struct vnode *vp, int waitfor, int passes, daddr_t lbn,
581 int (*checkdef)(struct buf *),
582 int (*waitoutput)(struct vnode *, struct thread *))
584 struct vfsync_info info;
587 bzero(&info, sizeof(info));
590 if ((info.checkdef = checkdef) == NULL)
598 * Lazy (filesystem syncer typ) Asynchronous plus limit the
599 * number of data (not meta) pages we try to flush to 1MB.
600 * A non-zero return means that lazy limit was reached.
602 info.lazylimit = 1024 * 1024;
604 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
605 vfsync_lazy_range_cmp, vfsync_bp, &info);
606 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
607 vfsync_meta_only_cmp, vfsync_bp, &info);
610 else if (!RB_EMPTY(&vp->v_rbdirty_tree))
611 vn_syncer_add_to_worklist(vp, 1);
616 * Asynchronous. Do a data-only pass and a meta-only pass.
619 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
621 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp,
627 * Synchronous. Do a data-only pass, then a meta-data+data
628 * pass, then additional integrated passes to try to get
629 * all the dependancies flushed.
631 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
633 error = vfsync_wait_output(vp, waitoutput);
635 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
637 error = vfsync_wait_output(vp, waitoutput);
639 while (error == 0 && passes > 0 &&
640 !RB_EMPTY(&vp->v_rbdirty_tree)) {
642 info.synchronous = 1;
645 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
651 error = vfsync_wait_output(vp, waitoutput);
660 vfsync_wait_output(struct vnode *vp, int (*waitoutput)(struct vnode *, struct thread *))
664 while (vp->v_track_write.bk_active) {
665 vp->v_track_write.bk_waitflag = 1;
666 tsleep(&vp->v_track_write, 0, "fsfsn", 0);
669 error = waitoutput(vp, curthread);
674 vfsync_data_only_cmp(struct buf *bp, void *data)
676 if (bp->b_lblkno < 0)
682 vfsync_meta_only_cmp(struct buf *bp, void *data)
684 if (bp->b_lblkno < 0)
690 vfsync_lazy_range_cmp(struct buf *bp, void *data)
692 struct vfsync_info *info = data;
693 if (bp->b_lblkno < info->vp->v_lazyw)
699 vfsync_bp(struct buf *bp, void *data)
701 struct vfsync_info *info = data;
702 struct vnode *vp = info->vp;
706 * if syncdeps is not set we do not try to write buffers which have
709 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp))
713 * Ignore buffers that we cannot immediately lock. XXX
715 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
717 if ((bp->b_flags & B_DELWRI) == 0)
718 panic("vfsync_bp: buffer not dirty");
720 panic("vfsync_bp: buffer vp mismatch");
723 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
724 * has been written but an additional handshake with the device
725 * is required before we can dispose of the buffer. We have no idea
726 * how to do this so we have to skip these buffers.
728 if (bp->b_flags & B_NEEDCOMMIT) {
734 * (LEGACY FROM UFS, REMOVE WHEN POSSIBLE) - invalidate any dirty
735 * buffers beyond the file EOF.
737 if (info->lbn != (daddr_t)-1 && vp->v_type == VREG &&
738 bp->b_lblkno >= info->lbn) {
740 bp->b_flags |= B_INVAL | B_NOCACHE;
746 if (info->synchronous) {
748 * Synchronous flushing. An error may be returned.
756 * Asynchronous flushing. A negative return value simply
757 * stops the scan and is not considered an error. We use
758 * this to support limited MNT_LAZY flushes.
760 vp->v_lazyw = bp->b_lblkno;
761 if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) {
763 info->lazycount += vfs_bio_awrite(bp);
765 info->lazycount += bp->b_bufsize;
771 if (info->lazylimit && info->lazycount >= info->lazylimit)
780 * Associate a buffer with a vnode.
783 bgetvp(struct vnode *vp, struct buf *bp)
785 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
790 * Insert onto list for new vnode.
793 bp->b_xflags |= BX_VNCLEAN;
794 bp->b_xflags &= ~BX_VNDIRTY;
795 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp))
796 panic("reassignbuf: dup lblk vp %p bp %p", vp, bp);
801 * Disassociate a buffer from a vnode.
804 brelvp(struct buf *bp)
808 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
811 * Delete from old vnode list, if on one.
815 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
816 if (bp->b_xflags & BX_VNDIRTY)
817 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
819 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
820 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
822 if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree)) {
823 vp->v_flag &= ~VONWORKLST;
824 LIST_REMOVE(vp, v_synclist);
832 * Associate a p-buffer with a vnode.
834 * Also sets B_PAGING flag to indicate that vnode is not fully associated
835 * with the buffer. i.e. the bp has not been linked into the vnode or
839 pbgetvp(struct vnode *vp, struct buf *bp)
841 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
844 bp->b_flags |= B_PAGING;
848 * Disassociate a p-buffer from a vnode.
851 pbrelvp(struct buf *bp)
853 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
856 bp->b_flags &= ~B_PAGING;
860 pbreassignbuf(struct buf *bp, struct vnode *newvp)
862 if ((bp->b_flags & B_PAGING) == 0) {
864 "pbreassignbuf() on non phys bp %p",
872 * Reassign a buffer from one vnode to another.
873 * Used to assign file specific control information
874 * (indirect blocks) to the vnode to which they belong.
877 reassignbuf(struct buf *bp, struct vnode *newvp)
882 printf("reassignbuf: NULL");
888 * B_PAGING flagged buffers cannot be reassigned because their vp
889 * is not fully linked in.
891 if (bp->b_flags & B_PAGING)
892 panic("cannot reassign paging buffer");
896 * Delete from old vnode list, if on one.
898 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
899 if (bp->b_xflags & BX_VNDIRTY)
900 buf_rb_tree_RB_REMOVE(&bp->b_vp->v_rbdirty_tree, bp);
902 buf_rb_tree_RB_REMOVE(&bp->b_vp->v_rbclean_tree, bp);
903 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
904 if (bp->b_vp != newvp) {
906 bp->b_vp = NULL; /* for clarification */
910 * If dirty, put on list of dirty buffers; otherwise insert onto list
913 if (bp->b_flags & B_DELWRI) {
914 if ((newvp->v_flag & VONWORKLST) == 0) {
915 switch (newvp->v_type) {
922 newvp->v_rdev->si_mountpoint != NULL) {
930 vn_syncer_add_to_worklist(newvp, delay);
932 bp->b_xflags |= BX_VNDIRTY;
933 if (buf_rb_tree_RB_INSERT(&newvp->v_rbdirty_tree, bp))
934 panic("reassignbuf: dup lblk vp %p bp %p", newvp, bp);
936 bp->b_xflags |= BX_VNCLEAN;
937 if (buf_rb_tree_RB_INSERT(&newvp->v_rbclean_tree, bp))
938 panic("reassignbuf: dup lblk vp %p bp %p", newvp, bp);
939 if ((newvp->v_flag & VONWORKLST) &&
940 RB_EMPTY(&newvp->v_rbdirty_tree)) {
941 newvp->v_flag &= ~VONWORKLST;
942 LIST_REMOVE(newvp, v_synclist);
945 if (bp->b_vp != newvp) {
953 * Create a vnode for a block device.
954 * Used for mounting the root file system.
957 bdevvp(dev_t dev, struct vnode **vpp)
967 error = getspecialvnode(VT_NON, NULL, &spec_vnode_vops, &nvp, 0, 0);
974 vp->v_udev = dev->si_udev;
981 v_associate_rdev(struct vnode *vp, dev_t dev)
985 if (dev == NULL || dev == NODEV)
987 if (dev_is_good(dev) == 0)
989 KKASSERT(vp->v_rdev == NULL);
992 vp->v_rdev = reference_dev(dev);
993 lwkt_gettoken(&ilock, &spechash_token);
994 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_specnext);
995 lwkt_reltoken(&ilock);
1000 v_release_rdev(struct vnode *vp)
1005 if ((dev = vp->v_rdev) != NULL) {
1006 lwkt_gettoken(&ilock, &spechash_token);
1007 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_specnext);
1008 if (dev_ref_debug && vp->v_opencount != 0) {
1009 printf("releasing rdev with non-0 "
1010 "v_opencount(%d) (revoked?)\n",
1014 vp->v_opencount = 0;
1016 lwkt_reltoken(&ilock);
1021 * Add a vnode to the alias list hung off the dev_t. We only associate
1022 * the device number with the vnode. The actual device is not associated
1023 * until the vnode is opened (usually in spec_open()), and will be
1024 * disassociated on last close.
1027 addaliasu(struct vnode *nvp, udev_t nvp_udev)
1029 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1030 panic("addaliasu on non-special vnode");
1031 nvp->v_udev = nvp_udev;
1035 * Disassociate a vnode from its underlying filesystem.
1037 * The vnode must be VX locked and refd
1039 * If there are v_usecount references to the vnode other then ours we have
1040 * to VOP_CLOSE the vnode before we can deactivate and reclaim it.
1043 vclean(struct vnode *vp, int flags, struct thread *td)
1049 * If the vnode has already been reclaimed we have nothing to do.
1051 if (vp->v_flag & VRECLAIMED)
1053 vp->v_flag |= VRECLAIMED;
1056 * Scrap the vfs cache
1058 while (cache_inval_vp(vp, 0, &retflags) != 0) {
1059 printf("Warning: vnode %p clean/cache_resolution race detected\n", vp);
1060 tsleep(vp, 0, "vclninv", 2);
1064 * Check to see if the vnode is in use. If so we have to reference it
1065 * before we clean it out so that its count cannot fall to zero and
1066 * generate a race against ourselves to recycle it.
1068 active = (vp->v_usecount > 1);
1071 * Clean out any buffers associated with the vnode and destroy its
1072 * object, if it has one.
1074 vinvalbuf(vp, V_SAVE, td, 0, 0);
1075 VOP_DESTROYVOBJECT(vp);
1078 * If purging an active vnode, it must be closed and
1079 * deactivated before being reclaimed. XXX
1081 * Note that neither of these routines unlocks the vnode.
1084 if (flags & DOCLOSE)
1085 VOP_CLOSE(vp, FNONBLOCK, td);
1089 * If the vnode has not be deactivated, deactivated it.
1091 if ((vp->v_flag & VINACTIVE) == 0) {
1092 vp->v_flag |= VINACTIVE;
1093 VOP_INACTIVE(vp, td);
1097 * Reclaim the vnode.
1099 if (VOP_RECLAIM(vp, retflags, td))
1100 panic("vclean: cannot reclaim");
1103 * Done with purge, notify sleepers of the grim news.
1105 vp->v_ops = &dead_vnode_vops;
1111 * Eliminate all activity associated with the requested vnode
1112 * and with all vnodes aliased to the requested vnode.
1114 * The vnode must be referenced and vx_lock()'d
1116 * revoke { struct vnode *a_vp, int a_flags }
1119 vop_stdrevoke(struct vop_revoke_args *ap)
1121 struct vnode *vp, *vq;
1125 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
1130 * If the vnode is already dead don't try to revoke it
1132 if (vp->v_flag & VRECLAIMED)
1136 * If the vnode has a device association, scrap all vnodes associated
1137 * with the device. Don't let the device disappear on us while we
1138 * are scrapping the vnodes.
1140 * The passed vp will probably show up in the list, do not VX lock
1143 if (vp->v_type != VCHR && vp->v_type != VBLK)
1145 if ((dev = vp->v_rdev) == NULL) {
1146 if ((dev = udev2dev(vp->v_udev, vp->v_type == VBLK)) == NODEV)
1150 lwkt_gettoken(&ilock, &spechash_token);
1151 while ((vq = SLIST_FIRST(&dev->si_hlist)) != NULL) {
1152 if (vp == vq || vx_get(vq) == 0) {
1153 if (vq == SLIST_FIRST(&dev->si_hlist))
1159 lwkt_reltoken(&ilock);
1165 * Recycle an unused vnode to the front of the free list.
1167 * Returns 1 if we were successfully able to recycle the vnode,
1171 vrecycle(struct vnode *vp, struct thread *td)
1173 if (vp->v_usecount == 1) {
1181 * Eliminate all activity associated with a vnode in preparation for reuse.
1183 * The vnode must be VX locked and refd and will remain VX locked and refd
1184 * on return. This routine may be called with the vnode in any state, as
1185 * long as it is VX locked. The vnode will be cleaned out and marked
1186 * VRECLAIMED but will not actually be reused until all existing refs and
1189 * NOTE: This routine may be called on a vnode which has not yet been
1190 * already been deactivated (VOP_INACTIVE), or on a vnode which has
1191 * already been reclaimed.
1193 * This routine is not responsible for placing us back on the freelist.
1194 * Instead, it happens automatically when the caller releases the VX lock
1195 * (assuming there aren't any other references).
1198 vgone(struct vnode *vp)
1201 * assert that the VX lock is held. This is an absolute requirement
1202 * now for vgone() to be called.
1204 KKASSERT(vp->v_lock.lk_exclusivecount == 1);
1207 * Clean out the filesystem specific data and set the VRECLAIMED
1208 * bit. Also deactivate the vnode if necessary.
1210 vclean(vp, DOCLOSE, curthread);
1213 * Delete from old mount point vnode list, if on one.
1215 if (vp->v_mount != NULL)
1216 insmntque(vp, NULL);
1219 * If special device, remove it from special device alias list
1220 * if it is on one. This should normally only occur if a vnode is
1221 * being revoked as the device should otherwise have been released
1224 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
1235 * Lookup a vnode by device number.
1238 vfinddev(dev_t dev, enum vtype type, struct vnode **vpp)
1243 lwkt_gettoken(&ilock, &spechash_token);
1244 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
1245 if (type == vp->v_type) {
1247 lwkt_reltoken(&ilock);
1251 lwkt_reltoken(&ilock);
1256 * Calculate the total number of references to a special device. This
1257 * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1258 * an overloaded field. Since udev2dev can now return NODEV, we have
1259 * to check for a NULL v_rdev.
1262 count_dev(dev_t dev)
1268 if (SLIST_FIRST(&dev->si_hlist)) {
1269 lwkt_gettoken(&ilock, &spechash_token);
1270 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
1271 count += vp->v_usecount;
1273 lwkt_reltoken(&ilock);
1279 count_udev(udev_t udev)
1283 if ((dev = udev2dev(udev, 0)) == NODEV)
1285 return(count_dev(dev));
1289 vcount(struct vnode *vp)
1291 if (vp->v_rdev == NULL)
1293 return(count_dev(vp->v_rdev));
1297 * Print out a description of a vnode.
1299 static char *typename[] =
1300 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1303 vprint(char *label, struct vnode *vp)
1308 printf("%s: %p: ", label, (void *)vp);
1310 printf("%p: ", (void *)vp);
1311 printf("type %s, usecount %d, writecount %d, refcount %d,",
1312 typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1315 if (vp->v_flag & VROOT)
1316 strcat(buf, "|VROOT");
1317 if (vp->v_flag & VTEXT)
1318 strcat(buf, "|VTEXT");
1319 if (vp->v_flag & VSYSTEM)
1320 strcat(buf, "|VSYSTEM");
1321 if (vp->v_flag & VFREE)
1322 strcat(buf, "|VFREE");
1323 if (vp->v_flag & VOBJBUF)
1324 strcat(buf, "|VOBJBUF");
1326 printf(" flags (%s)", &buf[1]);
1327 if (vp->v_data == NULL) {
1336 #include <ddb/ddb.h>
1338 static int db_show_locked_vnodes(struct mount *mp, void *data);
1341 * List all of the locked vnodes in the system.
1342 * Called when debugging the kernel.
1344 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
1346 printf("Locked vnodes\n");
1347 mountlist_scan(db_show_locked_vnodes, NULL,
1348 MNTSCAN_FORWARD|MNTSCAN_NOBUSY);
1352 db_show_locked_vnodes(struct mount *mp, void *data __unused)
1356 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
1357 if (VOP_ISLOCKED(vp, NULL))
1358 vprint((char *)0, vp);
1365 * Top level filesystem related information gathering.
1367 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
1370 vfs_sysctl(SYSCTL_HANDLER_ARGS)
1372 int *name = (int *)arg1 - 1; /* XXX */
1373 u_int namelen = arg2 + 1; /* XXX */
1374 struct vfsconf *vfsp;
1376 #if 1 || defined(COMPAT_PRELITE2)
1377 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1379 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
1383 /* all sysctl names at this level are at least name and field */
1385 return (ENOTDIR); /* overloaded */
1386 if (name[0] != VFS_GENERIC) {
1387 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1388 if (vfsp->vfc_typenum == name[0])
1391 return (EOPNOTSUPP);
1392 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1393 oldp, oldlenp, newp, newlen, p));
1397 case VFS_MAXTYPENUM:
1400 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
1403 return (ENOTDIR); /* overloaded */
1404 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1405 if (vfsp->vfc_typenum == name[2])
1408 return (EOPNOTSUPP);
1409 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
1411 return (EOPNOTSUPP);
1414 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
1415 "Generic filesystem");
1417 #if 1 || defined(COMPAT_PRELITE2)
1420 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
1423 struct vfsconf *vfsp;
1424 struct ovfsconf ovfs;
1426 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
1427 bzero(&ovfs, sizeof(ovfs));
1428 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
1429 strcpy(ovfs.vfc_name, vfsp->vfc_name);
1430 ovfs.vfc_index = vfsp->vfc_typenum;
1431 ovfs.vfc_refcount = vfsp->vfc_refcount;
1432 ovfs.vfc_flags = vfsp->vfc_flags;
1433 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
1440 #endif /* 1 || COMPAT_PRELITE2 */
1443 * Check to see if a filesystem is mounted on a block device.
1446 vfs_mountedon(struct vnode *vp)
1450 if ((dev = vp->v_rdev) == NULL)
1451 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK));
1452 if (dev != NODEV && dev->si_mountpoint)
1458 * Unmount all filesystems. The list is traversed in reverse order
1459 * of mounting to avoid dependencies.
1462 static int vfs_umountall_callback(struct mount *mp, void *data);
1465 vfs_unmountall(void)
1467 struct thread *td = curthread;
1470 if (td->td_proc == NULL)
1471 td = initproc->p_thread; /* XXX XXX use proc0 instead? */
1474 count = mountlist_scan(vfs_umountall_callback,
1475 &td, MNTSCAN_REVERSE|MNTSCAN_NOBUSY);
1481 vfs_umountall_callback(struct mount *mp, void *data)
1483 struct thread *td = *(struct thread **)data;
1486 error = dounmount(mp, MNT_FORCE, td);
1488 mountlist_remove(mp);
1489 printf("unmount of filesystem mounted from %s failed (",
1490 mp->mnt_stat.f_mntfromname);
1494 printf("%d)\n", error);
1500 * Build hash lists of net addresses and hang them off the mount point.
1501 * Called by ufs_mount() to set up the lists of export addresses.
1504 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1505 struct export_args *argp)
1508 struct radix_node_head *rnh;
1510 struct radix_node *rn;
1511 struct sockaddr *saddr, *smask = 0;
1515 if (argp->ex_addrlen == 0) {
1516 if (mp->mnt_flag & MNT_DEFEXPORTED)
1518 np = &nep->ne_defexported;
1519 np->netc_exflags = argp->ex_flags;
1520 np->netc_anon = argp->ex_anon;
1521 np->netc_anon.cr_ref = 1;
1522 mp->mnt_flag |= MNT_DEFEXPORTED;
1526 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
1528 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
1531 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1532 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
1533 bzero((caddr_t) np, i);
1534 saddr = (struct sockaddr *) (np + 1);
1535 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
1537 if (saddr->sa_len > argp->ex_addrlen)
1538 saddr->sa_len = argp->ex_addrlen;
1539 if (argp->ex_masklen) {
1540 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1541 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
1544 if (smask->sa_len > argp->ex_masklen)
1545 smask->sa_len = argp->ex_masklen;
1547 i = saddr->sa_family;
1548 if ((rnh = nep->ne_rtable[i]) == 0) {
1550 * Seems silly to initialize every AF when most are not used,
1551 * do so on demand here
1553 SLIST_FOREACH(dom, &domains, dom_next)
1554 if (dom->dom_family == i && dom->dom_rtattach) {
1555 dom->dom_rtattach((void **) &nep->ne_rtable[i],
1559 if ((rnh = nep->ne_rtable[i]) == 0) {
1564 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh,
1566 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */
1570 np->netc_exflags = argp->ex_flags;
1571 np->netc_anon = argp->ex_anon;
1572 np->netc_anon.cr_ref = 1;
1575 free(np, M_NETADDR);
1581 vfs_free_netcred(struct radix_node *rn, void *w)
1583 struct radix_node_head *rnh = (struct radix_node_head *) w;
1585 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
1586 free((caddr_t) rn, M_NETADDR);
1591 * Free the net address hash lists that are hanging off the mount points.
1594 vfs_free_addrlist(struct netexport *nep)
1597 struct radix_node_head *rnh;
1599 for (i = 0; i <= AF_MAX; i++)
1600 if ((rnh = nep->ne_rtable[i])) {
1601 (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
1603 free((caddr_t) rnh, M_RTABLE);
1604 nep->ne_rtable[i] = 0;
1609 vfs_export(struct mount *mp, struct netexport *nep, struct export_args *argp)
1613 if (argp->ex_flags & MNT_DELEXPORT) {
1614 if (mp->mnt_flag & MNT_EXPUBLIC) {
1615 vfs_setpublicfs(NULL, NULL, NULL);
1616 mp->mnt_flag &= ~MNT_EXPUBLIC;
1618 vfs_free_addrlist(nep);
1619 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
1621 if (argp->ex_flags & MNT_EXPORTED) {
1622 if (argp->ex_flags & MNT_EXPUBLIC) {
1623 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
1625 mp->mnt_flag |= MNT_EXPUBLIC;
1627 if ((error = vfs_hang_addrlist(mp, nep, argp)))
1629 mp->mnt_flag |= MNT_EXPORTED;
1636 * Set the publicly exported filesystem (WebNFS). Currently, only
1637 * one public filesystem is possible in the spec (RFC 2054 and 2055)
1640 vfs_setpublicfs(struct mount *mp, struct netexport *nep,
1641 struct export_args *argp)
1648 * mp == NULL -> invalidate the current info, the FS is
1649 * no longer exported. May be called from either vfs_export
1650 * or unmount, so check if it hasn't already been done.
1653 if (nfs_pub.np_valid) {
1654 nfs_pub.np_valid = 0;
1655 if (nfs_pub.np_index != NULL) {
1656 FREE(nfs_pub.np_index, M_TEMP);
1657 nfs_pub.np_index = NULL;
1664 * Only one allowed at a time.
1666 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
1670 * Get real filehandle for root of exported FS.
1672 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
1673 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
1675 if ((error = VFS_ROOT(mp, &rvp)))
1678 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
1684 * If an indexfile was specified, pull it in.
1686 if (argp->ex_indexfile != NULL) {
1689 error = vn_get_namelen(rvp, &namelen);
1692 MALLOC(nfs_pub.np_index, char *, namelen, M_TEMP,
1694 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
1695 namelen, (size_t *)0);
1698 * Check for illegal filenames.
1700 for (cp = nfs_pub.np_index; *cp; cp++) {
1708 FREE(nfs_pub.np_index, M_TEMP);
1713 nfs_pub.np_mount = mp;
1714 nfs_pub.np_valid = 1;
1719 vfs_export_lookup(struct mount *mp, struct netexport *nep,
1720 struct sockaddr *nam)
1723 struct radix_node_head *rnh;
1724 struct sockaddr *saddr;
1727 if (mp->mnt_flag & MNT_EXPORTED) {
1729 * Lookup in the export list first.
1733 rnh = nep->ne_rtable[saddr->sa_family];
1735 np = (struct netcred *)
1736 (*rnh->rnh_matchaddr)((char *)saddr,
1738 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
1743 * If no address match, use the default if it exists.
1745 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
1746 np = &nep->ne_defexported;
1752 * perform msync on all vnodes under a mount point. The mount point must
1753 * be locked. This code is also responsible for lazy-freeing unreferenced
1754 * vnodes whos VM objects no longer contain pages.
1756 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
1758 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked,
1759 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it
1760 * way up in this high level function.
1762 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
1763 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data);
1766 vfs_msync(struct mount *mp, int flags)
1770 vmsc_flags = VMSC_GETVP;
1771 if (flags != MNT_WAIT)
1772 vmsc_flags |= VMSC_NOWAIT;
1773 vmntvnodescan(mp, vmsc_flags, vfs_msync_scan1, vfs_msync_scan2,
1778 * scan1 is a fast pre-check. There could be hundreds of thousands of
1779 * vnodes, we cannot afford to do anything heavy weight until we have a
1780 * fairly good indication that there is work to do.
1784 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
1786 int flags = (int)data;
1788 if ((vp->v_flag & VRECLAIMED) == 0) {
1789 if (vshouldfree(vp, 0))
1790 return(0); /* call scan2 */
1791 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
1792 (vp->v_flag & VOBJDIRTY) &&
1793 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
1794 return(0); /* call scan2 */
1799 * do not call scan2, continue the loop
1805 * This callback is handed a locked vnode.
1809 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data)
1812 int flags = (int)data;
1814 if (vp->v_flag & VRECLAIMED)
1817 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
1818 (vp->v_flag & VOBJDIRTY)) {
1819 if (VOP_GETVOBJECT(vp, &obj) == 0) {
1820 vm_object_page_clean(obj, 0, 0,
1821 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
1828 * Create the VM object needed for VMIO and mmap support. This
1829 * is done for all VREG files in the system. Some filesystems might
1830 * afford the additional metadata buffering capability of the
1831 * VMIO code by making the device node be VMIO mode also.
1833 * vp must be locked when vfs_object_create is called.
1836 vfs_object_create(struct vnode *vp, struct thread *td)
1838 return (VOP_CREATEVOBJECT(vp, td));
1842 * Record a process's interest in events which might happen to
1843 * a vnode. Because poll uses the historic select-style interface
1844 * internally, this routine serves as both the ``check for any
1845 * pending events'' and the ``record my interest in future events''
1846 * functions. (These are done together, while the lock is held,
1847 * to avoid race conditions.)
1850 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
1854 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
1855 if (vp->v_pollinfo.vpi_revents & events) {
1857 * This leaves events we are not interested
1858 * in available for the other process which
1859 * which presumably had requested them
1860 * (otherwise they would never have been
1863 events &= vp->v_pollinfo.vpi_revents;
1864 vp->v_pollinfo.vpi_revents &= ~events;
1866 lwkt_reltoken(&ilock);
1869 vp->v_pollinfo.vpi_events |= events;
1870 selrecord(td, &vp->v_pollinfo.vpi_selinfo);
1871 lwkt_reltoken(&ilock);
1876 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
1877 * it is possible for us to miss an event due to race conditions, but
1878 * that condition is expected to be rare, so for the moment it is the
1879 * preferred interface.
1882 vn_pollevent(struct vnode *vp, int events)
1886 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
1887 if (vp->v_pollinfo.vpi_events & events) {
1889 * We clear vpi_events so that we don't
1890 * call selwakeup() twice if two events are
1891 * posted before the polling process(es) is
1892 * awakened. This also ensures that we take at
1893 * most one selwakeup() if the polling process
1894 * is no longer interested. However, it does
1895 * mean that only one event can be noticed at
1896 * a time. (Perhaps we should only clear those
1897 * event bits which we note?) XXX
1899 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */
1900 vp->v_pollinfo.vpi_revents |= events;
1901 selwakeup(&vp->v_pollinfo.vpi_selinfo);
1903 lwkt_reltoken(&ilock);
1907 * Wake up anyone polling on vp because it is being revoked.
1908 * This depends on dead_poll() returning POLLHUP for correct
1912 vn_pollgone(struct vnode *vp)
1916 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
1917 if (vp->v_pollinfo.vpi_events) {
1918 vp->v_pollinfo.vpi_events = 0;
1919 selwakeup(&vp->v_pollinfo.vpi_selinfo);
1921 lwkt_reltoken(&ilock);
1925 * extract the dev_t from a VBLK or VCHR. The vnode must have been opened
1926 * (or v_rdev might be NULL).
1929 vn_todev(struct vnode *vp)
1931 if (vp->v_type != VBLK && vp->v_type != VCHR)
1933 KKASSERT(vp->v_rdev != NULL);
1934 return (vp->v_rdev);
1938 * Check if vnode represents a disk device. The vnode does not need to be
1942 vn_isdisk(struct vnode *vp, int *errp)
1946 if (vp->v_type != VBLK && vp->v_type != VCHR) {
1952 if ((dev = vp->v_rdev) == NULL)
1953 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK));
1954 if (dev == NULL || dev == NODEV) {
1959 if (dev_is_good(dev) == 0) {
1964 if ((dev_dflags(dev) & D_DISK) == 0) {
1974 #ifdef DEBUG_VFS_LOCKS
1977 assert_vop_locked(struct vnode *vp, const char *str)
1979 if (vp && IS_LOCKING_VFS(vp) && !VOP_ISLOCKED(vp, NULL)) {
1980 panic("%s: %p is not locked shared but should be", str, vp);
1985 assert_vop_unlocked(struct vnode *vp, const char *str)
1987 if (vp && IS_LOCKING_VFS(vp)) {
1988 if (VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE) {
1989 panic("%s: %p is locked but should not be", str, vp);
1997 vn_get_namelen(struct vnode *vp, int *namelen)
1999 int error, retval[2];
2001 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval);
2009 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type,
2010 uint16_t d_namlen, const char *d_name)
2015 len = _DIRENT_RECLEN(d_namlen);
2016 if (len > uio->uio_resid)
2019 dp = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
2022 dp->d_namlen = d_namlen;
2023 dp->d_type = d_type;
2024 bcopy(d_name, dp->d_name, d_namlen);
2026 *error = uiomove((caddr_t)dp, len, uio);