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 1.91 2006/07/18 22:22:12 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;
135 * Red black tree functions
137 static int rb_buf_compare(struct buf *b1, struct buf *b2);
138 RB_GENERATE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t, b_loffset);
139 RB_GENERATE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t, b_loffset);
142 rb_buf_compare(struct buf *b1, struct buf *b2)
144 if (b1->b_loffset < b2->b_loffset)
146 if (b1->b_loffset > b2->b_loffset)
152 * 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 vupdatefsmid(struct vnode *vp)
279 atomic_set_int(&vp->v_flag, VFSMID);
283 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
285 struct vinvalbuf_bp_info info;
290 * If we are being asked to save, call fsync to ensure that the inode
293 if (flags & V_SAVE) {
295 while (vp->v_track_write.bk_active) {
296 vp->v_track_write.bk_waitflag = 1;
297 error = tsleep(&vp->v_track_write, slpflag,
298 "vinvlbuf", slptimeo);
304 if (!RB_EMPTY(&vp->v_rbdirty_tree)) {
306 if ((error = VOP_FSYNC(vp, MNT_WAIT)) != 0)
309 if (vp->v_track_write.bk_active > 0 ||
310 !RB_EMPTY(&vp->v_rbdirty_tree))
311 panic("vinvalbuf: dirty bufs");
316 info.slptimeo = slptimeo;
317 info.lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
318 if (slpflag & PCATCH)
319 info.lkflags |= LK_PCATCH;
324 * Flush the buffer cache until nothing is left.
326 while (!RB_EMPTY(&vp->v_rbclean_tree) ||
327 !RB_EMPTY(&vp->v_rbdirty_tree)) {
328 error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, NULL,
329 vinvalbuf_bp, &info);
331 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
332 vinvalbuf_bp, &info);
337 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
338 * have write I/O in-progress but if there is a VM object then the
339 * VM object can also have read-I/O in-progress.
342 while (vp->v_track_write.bk_active > 0) {
343 vp->v_track_write.bk_waitflag = 1;
344 tsleep(&vp->v_track_write, 0, "vnvlbv", 0);
346 if ((object = vp->v_object) != NULL) {
347 while (object->paging_in_progress)
348 vm_object_pip_sleep(object, "vnvlbx");
350 } while (vp->v_track_write.bk_active > 0);
355 * Destroy the copy in the VM cache, too.
357 if ((object = vp->v_object) != NULL) {
358 vm_object_page_remove(object, 0, 0,
359 (flags & V_SAVE) ? TRUE : FALSE);
362 if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree))
363 panic("vinvalbuf: flush failed");
364 if (!RB_EMPTY(&vp->v_rbhash_tree))
365 panic("vinvalbuf: flush failed, buffers still present");
370 vinvalbuf_bp(struct buf *bp, void *data)
372 struct vinvalbuf_bp_info *info = data;
375 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
376 error = BUF_TIMELOCK(bp, info->lkflags,
377 "vinvalbuf", info->slptimeo);
387 KKASSERT(bp->b_vp == info->vp);
390 * XXX Since there are no node locks for NFS, I
391 * believe there is a slight chance that a delayed
392 * write will occur while sleeping just above, so
393 * check for it. Note that vfs_bio_awrite expects
394 * buffers to reside on a queue, while bwrite() and
397 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
398 (info->flags & V_SAVE)) {
399 if (bp->b_vp == info->vp) {
400 if (bp->b_flags & B_CLUSTEROK) {
404 bp->b_flags |= B_ASYNC;
411 } else if (info->flags & V_SAVE) {
413 * Cannot set B_NOCACHE on a clean buffer as this will
414 * destroy the VM backing store which might actually
415 * be dirty (and unsynchronized).
418 bp->b_flags |= (B_INVAL | B_RELBUF);
419 bp->b_flags &= ~B_ASYNC;
423 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
424 bp->b_flags &= ~B_ASYNC;
431 * Truncate a file's buffer and pages to a specified length. This
432 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
435 * The vnode must be locked.
437 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data);
438 static int vtruncbuf_bp_trunc(struct buf *bp, void *data);
439 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data);
440 static int vtruncbuf_bp_metasync(struct buf *bp, void *data);
443 vtruncbuf(struct vnode *vp, off_t length, int blksize)
447 const char *filename;
450 * Round up to the *next* block, then destroy the buffers in question.
451 * Since we are only removing some of the buffers we must rely on the
452 * scan count to determine whether a loop is necessary.
454 if ((count = (int)(length % blksize)) != 0)
455 truncloffset = length + (blksize - count);
457 truncloffset = length;
461 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
462 vtruncbuf_bp_trunc_cmp,
463 vtruncbuf_bp_trunc, &truncloffset);
464 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
465 vtruncbuf_bp_trunc_cmp,
466 vtruncbuf_bp_trunc, &truncloffset);
470 * For safety, fsync any remaining metadata if the file is not being
471 * truncated to 0. Since the metadata does not represent the entire
472 * dirty list we have to rely on the hit count to ensure that we get
477 count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
478 vtruncbuf_bp_metasync_cmp,
479 vtruncbuf_bp_metasync, vp);
484 * Clean out any left over VM backing store.
488 vnode_pager_setsize(vp, length);
493 * It is possible to have in-progress I/O from buffers that were
494 * not part of the truncation. This should not happen if we
495 * are truncating to 0-length.
497 filename = TAILQ_FIRST(&vp->v_namecache) ?
498 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?";
500 while ((count = vp->v_track_write.bk_active) > 0) {
501 vp->v_track_write.bk_waitflag = 1;
502 tsleep(&vp->v_track_write, 0, "vbtrunc", 0);
504 printf("Warning: vtruncbuf(): Had to wait for "
505 "%d buffer I/Os to finish in %s\n",
511 * Make sure no buffers were instantiated while we were trying
512 * to clean out the remaining VM pages. This could occur due
513 * to busy dirty VM pages being flushed out to disk.
516 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
517 vtruncbuf_bp_trunc_cmp,
518 vtruncbuf_bp_trunc, &truncloffset);
519 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
520 vtruncbuf_bp_trunc_cmp,
521 vtruncbuf_bp_trunc, &truncloffset);
523 printf("Warning: vtruncbuf(): Had to re-clean %d "
524 "left over buffers in %s\n", count, filename);
534 * The callback buffer is beyond the new file EOF and must be destroyed.
535 * Note that the compare function must conform to the RB_SCAN's requirements.
539 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data)
541 if (bp->b_loffset >= *(off_t *)data)
548 vtruncbuf_bp_trunc(struct buf *bp, void *data)
551 * Do not try to use a buffer we cannot immediately lock, but sleep
552 * anyway to prevent a livelock. The code will loop until all buffers
555 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
556 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
560 bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE);
561 bp->b_flags &= ~B_ASYNC;
568 * Fsync all meta-data after truncating a file to be non-zero. Only metadata
569 * blocks (with a negative loffset) are scanned.
570 * Note that the compare function must conform to the RB_SCAN's requirements.
573 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data)
575 if (bp->b_loffset < 0)
581 vtruncbuf_bp_metasync(struct buf *bp, void *data)
583 struct vnode *vp = data;
585 if (bp->b_flags & B_DELWRI) {
587 * Do not try to use a buffer we cannot immediately lock,
588 * but sleep anyway to prevent a livelock. The code will
589 * loop until all buffers can be acted upon.
591 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
592 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
596 if (bp->b_vp == vp) {
597 bp->b_flags |= B_ASYNC;
599 bp->b_flags &= ~B_ASYNC;
610 * vfsync - implements a multipass fsync on a file which understands
611 * dependancies and meta-data. The passed vnode must be locked. The
612 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY.
614 * When fsyncing data asynchronously just do one consolidated pass starting
615 * with the most negative block number. This may not get all the data due
618 * When fsyncing data synchronously do a data pass, then a metadata pass,
619 * then do additional data+metadata passes to try to get all the data out.
621 static int vfsync_wait_output(struct vnode *vp,
622 int (*waitoutput)(struct vnode *, struct thread *));
623 static int vfsync_data_only_cmp(struct buf *bp, void *data);
624 static int vfsync_meta_only_cmp(struct buf *bp, void *data);
625 static int vfsync_lazy_range_cmp(struct buf *bp, void *data);
626 static int vfsync_bp(struct buf *bp, void *data);
635 int (*checkdef)(struct buf *);
639 vfsync(struct vnode *vp, int waitfor, int passes,
640 int (*checkdef)(struct buf *),
641 int (*waitoutput)(struct vnode *, struct thread *))
643 struct vfsync_info info;
646 bzero(&info, sizeof(info));
648 if ((info.checkdef = checkdef) == NULL)
651 crit_enter_id("vfsync");
656 * Lazy (filesystem syncer typ) Asynchronous plus limit the
657 * number of data (not meta) pages we try to flush to 1MB.
658 * A non-zero return means that lazy limit was reached.
660 info.lazylimit = 1024 * 1024;
662 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
663 vfsync_lazy_range_cmp, vfsync_bp, &info);
664 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
665 vfsync_meta_only_cmp, vfsync_bp, &info);
668 else if (!RB_EMPTY(&vp->v_rbdirty_tree))
669 vn_syncer_add_to_worklist(vp, 1);
674 * Asynchronous. Do a data-only pass and a meta-only pass.
677 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
679 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp,
685 * Synchronous. Do a data-only pass, then a meta-data+data
686 * pass, then additional integrated passes to try to get
687 * all the dependancies flushed.
689 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
691 error = vfsync_wait_output(vp, waitoutput);
693 info.skippedbufs = 0;
694 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
696 error = vfsync_wait_output(vp, waitoutput);
697 if (info.skippedbufs)
698 printf("Warning: vfsync skipped %d dirty bufs in pass2!\n", info.skippedbufs);
700 while (error == 0 && passes > 0 &&
701 !RB_EMPTY(&vp->v_rbdirty_tree)) {
703 info.synchronous = 1;
706 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
712 error = vfsync_wait_output(vp, waitoutput);
716 crit_exit_id("vfsync");
721 vfsync_wait_output(struct vnode *vp, int (*waitoutput)(struct vnode *, struct thread *))
725 while (vp->v_track_write.bk_active) {
726 vp->v_track_write.bk_waitflag = 1;
727 tsleep(&vp->v_track_write, 0, "fsfsn", 0);
730 error = waitoutput(vp, curthread);
735 vfsync_data_only_cmp(struct buf *bp, void *data)
737 if (bp->b_loffset < 0)
743 vfsync_meta_only_cmp(struct buf *bp, void *data)
745 if (bp->b_loffset < 0)
751 vfsync_lazy_range_cmp(struct buf *bp, void *data)
753 struct vfsync_info *info = data;
754 if (bp->b_loffset < info->vp->v_lazyw)
760 vfsync_bp(struct buf *bp, void *data)
762 struct vfsync_info *info = data;
763 struct vnode *vp = info->vp;
767 * if syncdeps is not set we do not try to write buffers which have
770 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp))
774 * Ignore buffers that we cannot immediately lock. XXX
776 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
777 printf("Warning: vfsync_bp skipping dirty buffer %p\n", bp);
781 if ((bp->b_flags & B_DELWRI) == 0)
782 panic("vfsync_bp: buffer not dirty");
784 panic("vfsync_bp: buffer vp mismatch");
787 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
788 * has been written but an additional handshake with the device
789 * is required before we can dispose of the buffer. We have no idea
790 * how to do this so we have to skip these buffers.
792 if (bp->b_flags & B_NEEDCOMMIT) {
797 if (info->synchronous) {
799 * Synchronous flushing. An error may be returned.
802 crit_exit_id("vfsync");
804 crit_enter_id("vfsync");
807 * Asynchronous flushing. A negative return value simply
808 * stops the scan and is not considered an error. We use
809 * this to support limited MNT_LAZY flushes.
811 vp->v_lazyw = bp->b_loffset;
812 if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) {
813 info->lazycount += vfs_bio_awrite(bp);
815 info->lazycount += bp->b_bufsize;
817 crit_exit_id("vfsync");
819 crit_enter_id("vfsync");
821 if (info->lazylimit && info->lazycount >= info->lazylimit)
830 * Associate a buffer with a vnode.
833 bgetvp(struct vnode *vp, struct buf *bp)
835 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
836 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0);
840 * Insert onto list for new vnode.
844 bp->b_flags |= B_HASHED;
845 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp))
846 panic("reassignbuf: dup lblk vp %p bp %p", vp, bp);
848 bp->b_flags |= B_VNCLEAN;
849 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp))
850 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp);
855 * Disassociate a buffer from a vnode.
858 brelvp(struct buf *bp)
862 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
865 * Delete from old vnode list, if on one.
869 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) {
870 if (bp->b_flags & B_VNDIRTY)
871 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
873 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
874 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN);
876 if (bp->b_flags & B_HASHED) {
877 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp);
878 bp->b_flags &= ~B_HASHED;
880 if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree)) {
881 vp->v_flag &= ~VONWORKLST;
882 LIST_REMOVE(vp, v_synclist);
890 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI.
891 * This routine is called when the state of the B_DELWRI bit is changed.
894 reassignbuf(struct buf *bp)
896 struct vnode *vp = bp->b_vp;
899 KKASSERT(vp != NULL);
903 * B_PAGING flagged buffers cannot be reassigned because their vp
904 * is not fully linked in.
906 if (bp->b_flags & B_PAGING)
907 panic("cannot reassign paging buffer");
910 if (bp->b_flags & B_DELWRI) {
912 * Move to the dirty list, add the vnode to the worklist
914 if (bp->b_flags & B_VNCLEAN) {
915 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
916 bp->b_flags &= ~B_VNCLEAN;
918 if ((bp->b_flags & B_VNDIRTY) == 0) {
919 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) {
920 panic("reassignbuf: dup lblk vp %p bp %p",
923 bp->b_flags |= B_VNDIRTY;
925 if ((vp->v_flag & VONWORKLST) == 0) {
926 switch (vp->v_type) {
933 vp->v_rdev->si_mountpoint != NULL) {
941 vn_syncer_add_to_worklist(vp, delay);
945 * Move to the clean list, remove the vnode from the worklist
946 * if no dirty blocks remain.
948 if (bp->b_flags & B_VNDIRTY) {
949 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
950 bp->b_flags &= ~B_VNDIRTY;
952 if ((bp->b_flags & B_VNCLEAN) == 0) {
953 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) {
954 panic("reassignbuf: dup lblk vp %p bp %p",
957 bp->b_flags |= B_VNCLEAN;
959 if ((vp->v_flag & VONWORKLST) &&
960 RB_EMPTY(&vp->v_rbdirty_tree)) {
961 vp->v_flag &= ~VONWORKLST;
962 LIST_REMOVE(vp, v_synclist);
969 * Create a vnode for a block device.
970 * Used for mounting the root file system.
973 bdevvp(dev_t dev, struct vnode **vpp)
983 error = getspecialvnode(VT_NON, NULL, &spec_vnode_vops_p, &nvp, 0, 0);
990 vp->v_udev = dev->si_udev;
997 v_associate_rdev(struct vnode *vp, dev_t dev)
1001 if (dev == NULL || dev == NODEV)
1003 if (dev_is_good(dev) == 0)
1005 KKASSERT(vp->v_rdev == NULL);
1008 vp->v_rdev = reference_dev(dev);
1009 lwkt_gettoken(&ilock, &spechash_token);
1010 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_specnext);
1011 lwkt_reltoken(&ilock);
1016 v_release_rdev(struct vnode *vp)
1021 if ((dev = vp->v_rdev) != NULL) {
1022 lwkt_gettoken(&ilock, &spechash_token);
1023 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_specnext);
1026 lwkt_reltoken(&ilock);
1031 * Add a vnode to the alias list hung off the dev_t. We only associate
1032 * the device number with the vnode. The actual device is not associated
1033 * until the vnode is opened (usually in spec_open()), and will be
1034 * disassociated on last close.
1037 addaliasu(struct vnode *nvp, udev_t nvp_udev)
1039 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1040 panic("addaliasu on non-special vnode");
1041 nvp->v_udev = nvp_udev;
1045 * Disassociate a vnode from its underlying filesystem.
1047 * The vnode must be VX locked and refd
1049 * If there are v_usecount references to the vnode other then ours we have
1050 * to VOP_CLOSE the vnode before we can deactivate and reclaim it.
1053 vclean(struct vnode *vp, int flags)
1060 * If the vnode has already been reclaimed we have nothing to do.
1062 if (vp->v_flag & VRECLAIMED)
1064 vp->v_flag |= VRECLAIMED;
1067 * Scrap the vfs cache
1069 while (cache_inval_vp(vp, 0) != 0) {
1070 printf("Warning: vnode %p clean/cache_resolution race detected\n", vp);
1071 tsleep(vp, 0, "vclninv", 2);
1075 * Check to see if the vnode is in use. If so we have to reference it
1076 * before we clean it out so that its count cannot fall to zero and
1077 * generate a race against ourselves to recycle it.
1079 active = (vp->v_usecount > 1);
1082 * Clean out any buffers associated with the vnode and destroy its
1083 * object, if it has one.
1085 vinvalbuf(vp, V_SAVE, 0, 0);
1088 * If purging an active vnode (typically during a forced unmount
1089 * or reboot), it must be closed and deactivated before being
1090 * reclaimed. This isn't really all that safe, but what can
1093 * Note that neither of these routines unlocks the vnode.
1095 if (active && (flags & DOCLOSE)) {
1096 while ((n = vp->v_opencount) != 0) {
1097 if (vp->v_writecount)
1098 VOP_CLOSE(vp, FWRITE|FNONBLOCK);
1100 VOP_CLOSE(vp, FNONBLOCK);
1101 if (vp->v_opencount == n) {
1102 printf("Warning: unable to force-close"
1110 * If the vnode has not be deactivated, deactivated it. Deactivation
1111 * can create new buffers and VM pages so we have to call vinvalbuf()
1112 * again to make sure they all get flushed.
1114 * This can occur if a file with a link count of 0 needs to be
1117 if ((vp->v_flag & VINACTIVE) == 0) {
1118 vp->v_flag |= VINACTIVE;
1120 vinvalbuf(vp, V_SAVE, 0, 0);
1124 * If the vnode has an object, destroy it.
1126 if ((object = vp->v_object) != NULL) {
1127 if (object->ref_count == 0) {
1128 if ((object->flags & OBJ_DEAD) == 0)
1129 vm_object_terminate(object);
1131 vm_pager_deallocate(object);
1133 vp->v_flag &= ~VOBJBUF;
1135 KKASSERT((vp->v_flag & VOBJBUF) == 0);
1139 * Reclaim the vnode.
1141 if (VOP_RECLAIM(vp))
1142 panic("vclean: cannot reclaim");
1145 * Done with purge, notify sleepers of the grim news.
1147 vp->v_ops = &dead_vnode_vops_p;
1153 * Eliminate all activity associated with the requested vnode
1154 * and with all vnodes aliased to the requested vnode.
1156 * The vnode must be referenced and vx_lock()'d
1158 * revoke { struct vnode *a_vp, int a_flags }
1161 vop_stdrevoke(struct vop_revoke_args *ap)
1163 struct vnode *vp, *vq;
1167 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
1172 * If the vnode is already dead don't try to revoke it
1174 if (vp->v_flag & VRECLAIMED)
1178 * If the vnode has a device association, scrap all vnodes associated
1179 * with the device. Don't let the device disappear on us while we
1180 * are scrapping the vnodes.
1182 * The passed vp will probably show up in the list, do not VX lock
1185 if (vp->v_type != VCHR && vp->v_type != VBLK)
1187 if ((dev = vp->v_rdev) == NULL) {
1188 if ((dev = udev2dev(vp->v_udev, vp->v_type == VBLK)) == NODEV)
1192 lwkt_gettoken(&ilock, &spechash_token);
1193 while ((vq = SLIST_FIRST(&dev->si_hlist)) != NULL) {
1194 if (vp == vq || vx_get(vq) == 0) {
1195 if (vq == SLIST_FIRST(&dev->si_hlist))
1201 lwkt_reltoken(&ilock);
1207 * Recycle an unused vnode to the front of the free list.
1209 * Returns 1 if we were successfully able to recycle the vnode,
1213 vrecycle(struct vnode *vp)
1215 if (vp->v_usecount == 1) {
1223 * Eliminate all activity associated with a vnode in preparation for reuse.
1225 * The vnode must be VX locked and refd and will remain VX locked and refd
1226 * on return. This routine may be called with the vnode in any state, as
1227 * long as it is VX locked. The vnode will be cleaned out and marked
1228 * VRECLAIMED but will not actually be reused until all existing refs and
1231 * NOTE: This routine may be called on a vnode which has not yet been
1232 * already been deactivated (VOP_INACTIVE), or on a vnode which has
1233 * already been reclaimed.
1235 * This routine is not responsible for placing us back on the freelist.
1236 * Instead, it happens automatically when the caller releases the VX lock
1237 * (assuming there aren't any other references).
1240 vgone(struct vnode *vp)
1243 * assert that the VX lock is held. This is an absolute requirement
1244 * now for vgone() to be called.
1246 KKASSERT(vp->v_lock.lk_exclusivecount == 1);
1249 * Clean out the filesystem specific data and set the VRECLAIMED
1250 * bit. Also deactivate the vnode if necessary.
1252 vclean(vp, DOCLOSE);
1255 * Delete from old mount point vnode list, if on one.
1257 if (vp->v_mount != NULL)
1258 insmntque(vp, NULL);
1261 * If special device, remove it from special device alias list
1262 * if it is on one. This should normally only occur if a vnode is
1263 * being revoked as the device should otherwise have been released
1266 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
1277 * Lookup a vnode by device number.
1280 vfinddev(dev_t dev, enum vtype type, struct vnode **vpp)
1285 lwkt_gettoken(&ilock, &spechash_token);
1286 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
1287 if (type == vp->v_type) {
1289 lwkt_reltoken(&ilock);
1293 lwkt_reltoken(&ilock);
1298 * Calculate the total number of references to a special device. This
1299 * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1300 * an overloaded field. Since udev2dev can now return NODEV, we have
1301 * to check for a NULL v_rdev.
1304 count_dev(dev_t dev)
1310 if (SLIST_FIRST(&dev->si_hlist)) {
1311 lwkt_gettoken(&ilock, &spechash_token);
1312 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
1313 count += vp->v_usecount;
1315 lwkt_reltoken(&ilock);
1321 count_udev(udev_t udev)
1325 if ((dev = udev2dev(udev, 0)) == NODEV)
1327 return(count_dev(dev));
1331 vcount(struct vnode *vp)
1333 if (vp->v_rdev == NULL)
1335 return(count_dev(vp->v_rdev));
1339 * Initialize VMIO for a vnode. This routine MUST be called before a
1340 * VFS can issue buffer cache ops on a vnode. It is typically called
1341 * when a vnode is initialized from its inode.
1344 vinitvmio(struct vnode *vp, off_t filesize)
1350 if ((object = vp->v_object) == NULL) {
1351 object = vnode_pager_alloc(vp, filesize, 0, 0);
1353 * Dereference the reference we just created. This assumes
1354 * that the object is associated with the vp.
1356 object->ref_count--;
1359 if (object->flags & OBJ_DEAD) {
1361 tsleep(object, 0, "vodead", 0);
1362 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1366 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object"));
1367 vp->v_flag |= VOBJBUF;
1373 * Print out a description of a vnode.
1375 static char *typename[] =
1376 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1379 vprint(char *label, struct vnode *vp)
1384 printf("%s: %p: ", label, (void *)vp);
1386 printf("%p: ", (void *)vp);
1387 printf("type %s, usecount %d, writecount %d, refcount %d,",
1388 typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1391 if (vp->v_flag & VROOT)
1392 strcat(buf, "|VROOT");
1393 if (vp->v_flag & VTEXT)
1394 strcat(buf, "|VTEXT");
1395 if (vp->v_flag & VSYSTEM)
1396 strcat(buf, "|VSYSTEM");
1397 if (vp->v_flag & VFREE)
1398 strcat(buf, "|VFREE");
1399 if (vp->v_flag & VOBJBUF)
1400 strcat(buf, "|VOBJBUF");
1402 printf(" flags (%s)", &buf[1]);
1403 if (vp->v_data == NULL) {
1412 #include <ddb/ddb.h>
1414 static int db_show_locked_vnodes(struct mount *mp, void *data);
1417 * List all of the locked vnodes in the system.
1418 * Called when debugging the kernel.
1420 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
1422 printf("Locked vnodes\n");
1423 mountlist_scan(db_show_locked_vnodes, NULL,
1424 MNTSCAN_FORWARD|MNTSCAN_NOBUSY);
1428 db_show_locked_vnodes(struct mount *mp, void *data __unused)
1432 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
1433 if (VOP_ISLOCKED(vp, NULL))
1434 vprint((char *)0, vp);
1441 * Top level filesystem related information gathering.
1443 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
1446 vfs_sysctl(SYSCTL_HANDLER_ARGS)
1448 int *name = (int *)arg1 - 1; /* XXX */
1449 u_int namelen = arg2 + 1; /* XXX */
1450 struct vfsconf *vfsp;
1452 #if 1 || defined(COMPAT_PRELITE2)
1453 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1455 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
1459 /* all sysctl names at this level are at least name and field */
1461 return (ENOTDIR); /* overloaded */
1462 if (name[0] != VFS_GENERIC) {
1463 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1464 if (vfsp->vfc_typenum == name[0])
1467 return (EOPNOTSUPP);
1468 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1469 oldp, oldlenp, newp, newlen, p));
1473 case VFS_MAXTYPENUM:
1476 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
1479 return (ENOTDIR); /* overloaded */
1480 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1481 if (vfsp->vfc_typenum == name[2])
1484 return (EOPNOTSUPP);
1485 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
1487 return (EOPNOTSUPP);
1490 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
1491 "Generic filesystem");
1493 #if 1 || defined(COMPAT_PRELITE2)
1496 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
1499 struct vfsconf *vfsp;
1500 struct ovfsconf ovfs;
1502 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
1503 bzero(&ovfs, sizeof(ovfs));
1504 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
1505 strcpy(ovfs.vfc_name, vfsp->vfc_name);
1506 ovfs.vfc_index = vfsp->vfc_typenum;
1507 ovfs.vfc_refcount = vfsp->vfc_refcount;
1508 ovfs.vfc_flags = vfsp->vfc_flags;
1509 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
1516 #endif /* 1 || COMPAT_PRELITE2 */
1519 * Check to see if a filesystem is mounted on a block device.
1522 vfs_mountedon(struct vnode *vp)
1526 if ((dev = vp->v_rdev) == NULL)
1527 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK));
1528 if (dev != NODEV && dev->si_mountpoint)
1534 * Unmount all filesystems. The list is traversed in reverse order
1535 * of mounting to avoid dependencies.
1538 static int vfs_umountall_callback(struct mount *mp, void *data);
1541 vfs_unmountall(void)
1543 struct thread *td = curthread;
1546 if (td->td_proc == NULL)
1547 td = initproc->p_thread; /* XXX XXX use proc0 instead? */
1550 count = mountlist_scan(vfs_umountall_callback,
1551 NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY);
1557 vfs_umountall_callback(struct mount *mp, void *data)
1561 error = dounmount(mp, MNT_FORCE);
1563 mountlist_remove(mp);
1564 printf("unmount of filesystem mounted from %s failed (",
1565 mp->mnt_stat.f_mntfromname);
1569 printf("%d)\n", error);
1575 * Build hash lists of net addresses and hang them off the mount point.
1576 * Called by ufs_mount() to set up the lists of export addresses.
1579 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1580 struct export_args *argp)
1583 struct radix_node_head *rnh;
1585 struct radix_node *rn;
1586 struct sockaddr *saddr, *smask = 0;
1590 if (argp->ex_addrlen == 0) {
1591 if (mp->mnt_flag & MNT_DEFEXPORTED)
1593 np = &nep->ne_defexported;
1594 np->netc_exflags = argp->ex_flags;
1595 np->netc_anon = argp->ex_anon;
1596 np->netc_anon.cr_ref = 1;
1597 mp->mnt_flag |= MNT_DEFEXPORTED;
1601 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
1603 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
1606 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1607 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
1608 bzero((caddr_t) np, i);
1609 saddr = (struct sockaddr *) (np + 1);
1610 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
1612 if (saddr->sa_len > argp->ex_addrlen)
1613 saddr->sa_len = argp->ex_addrlen;
1614 if (argp->ex_masklen) {
1615 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1616 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
1619 if (smask->sa_len > argp->ex_masklen)
1620 smask->sa_len = argp->ex_masklen;
1622 i = saddr->sa_family;
1623 if ((rnh = nep->ne_rtable[i]) == 0) {
1625 * Seems silly to initialize every AF when most are not used,
1626 * do so on demand here
1628 SLIST_FOREACH(dom, &domains, dom_next)
1629 if (dom->dom_family == i && dom->dom_rtattach) {
1630 dom->dom_rtattach((void **) &nep->ne_rtable[i],
1634 if ((rnh = nep->ne_rtable[i]) == 0) {
1639 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh,
1641 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */
1645 np->netc_exflags = argp->ex_flags;
1646 np->netc_anon = argp->ex_anon;
1647 np->netc_anon.cr_ref = 1;
1650 free(np, M_NETADDR);
1656 vfs_free_netcred(struct radix_node *rn, void *w)
1658 struct radix_node_head *rnh = (struct radix_node_head *) w;
1660 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
1661 free((caddr_t) rn, M_NETADDR);
1666 * Free the net address hash lists that are hanging off the mount points.
1669 vfs_free_addrlist(struct netexport *nep)
1672 struct radix_node_head *rnh;
1674 for (i = 0; i <= AF_MAX; i++)
1675 if ((rnh = nep->ne_rtable[i])) {
1676 (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
1678 free((caddr_t) rnh, M_RTABLE);
1679 nep->ne_rtable[i] = 0;
1684 vfs_export(struct mount *mp, struct netexport *nep, struct export_args *argp)
1688 if (argp->ex_flags & MNT_DELEXPORT) {
1689 if (mp->mnt_flag & MNT_EXPUBLIC) {
1690 vfs_setpublicfs(NULL, NULL, NULL);
1691 mp->mnt_flag &= ~MNT_EXPUBLIC;
1693 vfs_free_addrlist(nep);
1694 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
1696 if (argp->ex_flags & MNT_EXPORTED) {
1697 if (argp->ex_flags & MNT_EXPUBLIC) {
1698 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
1700 mp->mnt_flag |= MNT_EXPUBLIC;
1702 if ((error = vfs_hang_addrlist(mp, nep, argp)))
1704 mp->mnt_flag |= MNT_EXPORTED;
1711 * Set the publicly exported filesystem (WebNFS). Currently, only
1712 * one public filesystem is possible in the spec (RFC 2054 and 2055)
1715 vfs_setpublicfs(struct mount *mp, struct netexport *nep,
1716 struct export_args *argp)
1723 * mp == NULL -> invalidate the current info, the FS is
1724 * no longer exported. May be called from either vfs_export
1725 * or unmount, so check if it hasn't already been done.
1728 if (nfs_pub.np_valid) {
1729 nfs_pub.np_valid = 0;
1730 if (nfs_pub.np_index != NULL) {
1731 FREE(nfs_pub.np_index, M_TEMP);
1732 nfs_pub.np_index = NULL;
1739 * Only one allowed at a time.
1741 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
1745 * Get real filehandle for root of exported FS.
1747 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
1748 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
1750 if ((error = VFS_ROOT(mp, &rvp)))
1753 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
1759 * If an indexfile was specified, pull it in.
1761 if (argp->ex_indexfile != NULL) {
1764 error = vn_get_namelen(rvp, &namelen);
1767 MALLOC(nfs_pub.np_index, char *, namelen, M_TEMP,
1769 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
1770 namelen, (size_t *)0);
1773 * Check for illegal filenames.
1775 for (cp = nfs_pub.np_index; *cp; cp++) {
1783 FREE(nfs_pub.np_index, M_TEMP);
1788 nfs_pub.np_mount = mp;
1789 nfs_pub.np_valid = 1;
1794 vfs_export_lookup(struct mount *mp, struct netexport *nep,
1795 struct sockaddr *nam)
1798 struct radix_node_head *rnh;
1799 struct sockaddr *saddr;
1802 if (mp->mnt_flag & MNT_EXPORTED) {
1804 * Lookup in the export list first.
1808 rnh = nep->ne_rtable[saddr->sa_family];
1810 np = (struct netcred *)
1811 (*rnh->rnh_matchaddr)((char *)saddr,
1813 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
1818 * If no address match, use the default if it exists.
1820 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
1821 np = &nep->ne_defexported;
1827 * perform msync on all vnodes under a mount point. The mount point must
1828 * be locked. This code is also responsible for lazy-freeing unreferenced
1829 * vnodes whos VM objects no longer contain pages.
1831 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
1833 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked,
1834 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it
1835 * way up in this high level function.
1837 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
1838 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data);
1841 vfs_msync(struct mount *mp, int flags)
1845 vmsc_flags = VMSC_GETVP;
1846 if (flags != MNT_WAIT)
1847 vmsc_flags |= VMSC_NOWAIT;
1848 vmntvnodescan(mp, vmsc_flags, vfs_msync_scan1, vfs_msync_scan2,
1853 * scan1 is a fast pre-check. There could be hundreds of thousands of
1854 * vnodes, we cannot afford to do anything heavy weight until we have a
1855 * fairly good indication that there is work to do.
1859 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
1861 int flags = (int)data;
1863 if ((vp->v_flag & VRECLAIMED) == 0) {
1864 if (vshouldfree(vp, 0))
1865 return(0); /* call scan2 */
1866 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
1867 (vp->v_flag & VOBJDIRTY) &&
1868 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
1869 return(0); /* call scan2 */
1874 * do not call scan2, continue the loop
1880 * This callback is handed a locked vnode.
1884 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data)
1887 int flags = (int)data;
1889 if (vp->v_flag & VRECLAIMED)
1892 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) {
1893 if ((obj = vp->v_object) != NULL) {
1894 vm_object_page_clean(obj, 0, 0,
1895 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
1902 * Record a process's interest in events which might happen to
1903 * a vnode. Because poll uses the historic select-style interface
1904 * internally, this routine serves as both the ``check for any
1905 * pending events'' and the ``record my interest in future events''
1906 * functions. (These are done together, while the lock is held,
1907 * to avoid race conditions.)
1910 vn_pollrecord(struct vnode *vp, int events)
1914 KKASSERT(curthread->td_proc != NULL);
1916 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
1917 if (vp->v_pollinfo.vpi_revents & events) {
1919 * This leaves events we are not interested
1920 * in available for the other process which
1921 * which presumably had requested them
1922 * (otherwise they would never have been
1925 events &= vp->v_pollinfo.vpi_revents;
1926 vp->v_pollinfo.vpi_revents &= ~events;
1928 lwkt_reltoken(&ilock);
1931 vp->v_pollinfo.vpi_events |= events;
1932 selrecord(curthread, &vp->v_pollinfo.vpi_selinfo);
1933 lwkt_reltoken(&ilock);
1938 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
1939 * it is possible for us to miss an event due to race conditions, but
1940 * that condition is expected to be rare, so for the moment it is the
1941 * preferred interface.
1944 vn_pollevent(struct vnode *vp, int events)
1948 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
1949 if (vp->v_pollinfo.vpi_events & events) {
1951 * We clear vpi_events so that we don't
1952 * call selwakeup() twice if two events are
1953 * posted before the polling process(es) is
1954 * awakened. This also ensures that we take at
1955 * most one selwakeup() if the polling process
1956 * is no longer interested. However, it does
1957 * mean that only one event can be noticed at
1958 * a time. (Perhaps we should only clear those
1959 * event bits which we note?) XXX
1961 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */
1962 vp->v_pollinfo.vpi_revents |= events;
1963 selwakeup(&vp->v_pollinfo.vpi_selinfo);
1965 lwkt_reltoken(&ilock);
1969 * Wake up anyone polling on vp because it is being revoked.
1970 * This depends on dead_poll() returning POLLHUP for correct
1974 vn_pollgone(struct vnode *vp)
1978 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
1979 if (vp->v_pollinfo.vpi_events) {
1980 vp->v_pollinfo.vpi_events = 0;
1981 selwakeup(&vp->v_pollinfo.vpi_selinfo);
1983 lwkt_reltoken(&ilock);
1987 * extract the dev_t from a VBLK or VCHR. The vnode must have been opened
1988 * (or v_rdev might be NULL).
1991 vn_todev(struct vnode *vp)
1993 if (vp->v_type != VBLK && vp->v_type != VCHR)
1995 KKASSERT(vp->v_rdev != NULL);
1996 return (vp->v_rdev);
2000 * Check if vnode represents a disk device. The vnode does not need to be
2004 vn_isdisk(struct vnode *vp, int *errp)
2008 if (vp->v_type != VBLK && vp->v_type != VCHR) {
2014 if ((dev = vp->v_rdev) == NULL)
2015 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK));
2016 if (dev == NULL || dev == NODEV) {
2021 if (dev_is_good(dev) == 0) {
2026 if ((dev_dflags(dev) & D_DISK) == 0) {
2036 #ifdef DEBUG_VFS_LOCKS
2039 assert_vop_locked(struct vnode *vp, const char *str)
2041 if (vp && !VOP_ISLOCKED(vp, NULL)) {
2042 panic("%s: %p is not locked shared but should be", str, vp);
2047 assert_vop_unlocked(struct vnode *vp, const char *str)
2050 if (VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE) {
2051 panic("%s: %p is locked but should not be", str, vp);
2059 vn_get_namelen(struct vnode *vp, int *namelen)
2061 int error, retval[2];
2063 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval);
2071 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type,
2072 uint16_t d_namlen, const char *d_name)
2077 len = _DIRENT_RECLEN(d_namlen);
2078 if (len > uio->uio_resid)
2081 dp = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
2084 dp->d_namlen = d_namlen;
2085 dp->d_type = d_type;
2086 bcopy(d_name, dp->d_name, d_namlen);
2088 *error = uiomove((caddr_t)dp, len, uio);