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 1.110 2008/01/05 14:02:38 swildner 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>
86 #include <sys/sysref2.h>
88 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
91 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
93 SYSCTL_INT(_vfs, OID_AUTO, fastdev, CTLFLAG_RW, &vfs_fastdev, 0, "");
95 enum vtype iftovt_tab[16] = {
96 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
97 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
100 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
101 S_IFSOCK, S_IFIFO, S_IFMT,
104 static int reassignbufcalls;
105 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW,
106 &reassignbufcalls, 0, "");
107 static int reassignbufloops;
108 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW,
109 &reassignbufloops, 0, "");
110 static int reassignbufsortgood;
111 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW,
112 &reassignbufsortgood, 0, "");
113 static int reassignbufsortbad;
114 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW,
115 &reassignbufsortbad, 0, "");
116 static int reassignbufmethod = 1;
117 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW,
118 &reassignbufmethod, 0, "");
120 int nfs_mount_type = -1;
121 static struct lwkt_token spechash_token;
122 struct nfs_public nfs_pub; /* publicly exported FS */
125 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
126 &desiredvnodes, 0, "Maximum number of vnodes");
128 static void vfs_free_addrlist (struct netexport *nep);
129 static int vfs_free_netcred (struct radix_node *rn, void *w);
130 static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
131 struct export_args *argp);
133 extern int dev_ref_debug;
136 * Red black tree functions
138 static int rb_buf_compare(struct buf *b1, struct buf *b2);
139 RB_GENERATE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t, b_loffset);
140 RB_GENERATE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t, b_loffset);
143 rb_buf_compare(struct buf *b1, struct buf *b2)
145 if (b1->b_loffset < b2->b_loffset)
147 if (b1->b_loffset > b2->b_loffset)
153 * Returns non-zero if the vnode is a candidate for lazy msyncing.
156 vshouldmsync(struct vnode *vp)
158 if (vp->v_auxrefs != 0 || vp->v_sysref.refcnt > 0)
159 return (0); /* other holders */
161 (vp->v_object->ref_count || vp->v_object->resident_page_count)) {
168 * Initialize the vnode management data structures.
170 * Called from vfsinit()
176 * Desired vnodes is a result of the physical page count
177 * and the size of kernel's heap. It scales in proportion
178 * to the amount of available physical memory. This can
179 * cause trouble on 64-bit and large memory platforms.
181 /* desiredvnodes = maxproc + vmstats.v_page_count / 4; */
183 min(maxproc + vmstats.v_page_count / 4,
185 (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
187 lwkt_token_init(&spechash_token);
191 * Knob to control the precision of file timestamps:
193 * 0 = seconds only; nanoseconds zeroed.
194 * 1 = seconds and nanoseconds, accurate within 1/HZ.
195 * 2 = seconds and nanoseconds, truncated to microseconds.
196 * >=3 = seconds and nanoseconds, maximum precision.
198 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
200 static int timestamp_precision = TSP_SEC;
201 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
202 ×tamp_precision, 0, "");
205 * Get a current timestamp.
208 vfs_timestamp(struct timespec *tsp)
212 switch (timestamp_precision) {
214 tsp->tv_sec = time_second;
222 TIMEVAL_TO_TIMESPEC(&tv, tsp);
232 * Set vnode attributes to VNOVAL
235 vattr_null(struct vattr *vap)
238 vap->va_size = VNOVAL;
239 vap->va_bytes = VNOVAL;
240 vap->va_mode = VNOVAL;
241 vap->va_nlink = VNOVAL;
242 vap->va_uid = VNOVAL;
243 vap->va_gid = VNOVAL;
244 vap->va_fsid = VNOVAL;
245 vap->va_fileid = VNOVAL;
246 vap->va_blocksize = VNOVAL;
247 vap->va_rmajor = VNOVAL;
248 vap->va_rminor = VNOVAL;
249 vap->va_atime.tv_sec = VNOVAL;
250 vap->va_atime.tv_nsec = VNOVAL;
251 vap->va_mtime.tv_sec = VNOVAL;
252 vap->va_mtime.tv_nsec = VNOVAL;
253 vap->va_ctime.tv_sec = VNOVAL;
254 vap->va_ctime.tv_nsec = VNOVAL;
255 vap->va_flags = VNOVAL;
256 vap->va_gen = VNOVAL;
258 vap->va_fsmid = VNOVAL;
259 /* va_*_uuid fields are only valid if related flags are set */
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 * NOTE: NO B_LOCKED CHECK. Also no buf_checkwrite()
398 * check. This code will write out the buffer, period.
400 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
401 (info->flags & V_SAVE)) {
402 if (bp->b_vp == info->vp) {
403 if (bp->b_flags & B_CLUSTEROK) {
407 bp->b_flags |= B_ASYNC;
414 } else if (info->flags & V_SAVE) {
416 * Cannot set B_NOCACHE on a clean buffer as this will
417 * destroy the VM backing store which might actually
418 * be dirty (and unsynchronized).
421 bp->b_flags |= (B_INVAL | B_RELBUF);
422 bp->b_flags &= ~B_ASYNC;
426 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
427 bp->b_flags &= ~B_ASYNC;
434 * Truncate a file's buffer and pages to a specified length. This
435 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
438 * The vnode must be locked.
440 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data);
441 static int vtruncbuf_bp_trunc(struct buf *bp, void *data);
442 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data);
443 static int vtruncbuf_bp_metasync(struct buf *bp, void *data);
446 vtruncbuf(struct vnode *vp, off_t length, int blksize)
450 const char *filename;
453 * Round up to the *next* block, then destroy the buffers in question.
454 * Since we are only removing some of the buffers we must rely on the
455 * scan count to determine whether a loop is necessary.
457 if ((count = (int)(length % blksize)) != 0)
458 truncloffset = length + (blksize - count);
460 truncloffset = length;
464 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
465 vtruncbuf_bp_trunc_cmp,
466 vtruncbuf_bp_trunc, &truncloffset);
467 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
468 vtruncbuf_bp_trunc_cmp,
469 vtruncbuf_bp_trunc, &truncloffset);
473 * For safety, fsync any remaining metadata if the file is not being
474 * truncated to 0. Since the metadata does not represent the entire
475 * dirty list we have to rely on the hit count to ensure that we get
480 count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
481 vtruncbuf_bp_metasync_cmp,
482 vtruncbuf_bp_metasync, vp);
487 * Clean out any left over VM backing store.
491 vnode_pager_setsize(vp, length);
496 * It is possible to have in-progress I/O from buffers that were
497 * not part of the truncation. This should not happen if we
498 * are truncating to 0-length.
500 filename = TAILQ_FIRST(&vp->v_namecache) ?
501 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?";
503 while ((count = vp->v_track_write.bk_active) > 0) {
504 vp->v_track_write.bk_waitflag = 1;
505 tsleep(&vp->v_track_write, 0, "vbtrunc", 0);
507 kprintf("Warning: vtruncbuf(): Had to wait for "
508 "%d buffer I/Os to finish in %s\n",
514 * Make sure no buffers were instantiated while we were trying
515 * to clean out the remaining VM pages. This could occur due
516 * to busy dirty VM pages being flushed out to disk.
519 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
520 vtruncbuf_bp_trunc_cmp,
521 vtruncbuf_bp_trunc, &truncloffset);
522 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
523 vtruncbuf_bp_trunc_cmp,
524 vtruncbuf_bp_trunc, &truncloffset);
526 kprintf("Warning: vtruncbuf(): Had to re-clean %d "
527 "left over buffers in %s\n", count, filename);
537 * The callback buffer is beyond the new file EOF and must be destroyed.
538 * Note that the compare function must conform to the RB_SCAN's requirements.
542 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data)
544 if (bp->b_loffset >= *(off_t *)data)
551 vtruncbuf_bp_trunc(struct buf *bp, void *data)
554 * Do not try to use a buffer we cannot immediately lock, but sleep
555 * anyway to prevent a livelock. The code will loop until all buffers
558 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
559 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
563 bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE);
564 bp->b_flags &= ~B_ASYNC;
571 * Fsync all meta-data after truncating a file to be non-zero. Only metadata
572 * blocks (with a negative loffset) are scanned.
573 * Note that the compare function must conform to the RB_SCAN's requirements.
576 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data)
578 if (bp->b_loffset < 0)
584 vtruncbuf_bp_metasync(struct buf *bp, void *data)
586 struct vnode *vp = data;
588 if (bp->b_flags & B_DELWRI) {
590 * Do not try to use a buffer we cannot immediately lock,
591 * but sleep anyway to prevent a livelock. The code will
592 * loop until all buffers can be acted upon.
594 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
595 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
599 if (bp->b_vp == vp) {
600 bp->b_flags |= B_ASYNC;
602 bp->b_flags &= ~B_ASYNC;
613 * vfsync - implements a multipass fsync on a file which understands
614 * dependancies and meta-data. The passed vnode must be locked. The
615 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY.
617 * When fsyncing data asynchronously just do one consolidated pass starting
618 * with the most negative block number. This may not get all the data due
621 * When fsyncing data synchronously do a data pass, then a metadata pass,
622 * then do additional data+metadata passes to try to get all the data out.
624 static int vfsync_wait_output(struct vnode *vp,
625 int (*waitoutput)(struct vnode *, struct thread *));
626 static int vfsync_data_only_cmp(struct buf *bp, void *data);
627 static int vfsync_meta_only_cmp(struct buf *bp, void *data);
628 static int vfsync_lazy_range_cmp(struct buf *bp, void *data);
629 static int vfsync_bp(struct buf *bp, void *data);
638 int (*checkdef)(struct buf *);
642 vfsync(struct vnode *vp, int waitfor, int passes,
643 int (*checkdef)(struct buf *),
644 int (*waitoutput)(struct vnode *, struct thread *))
646 struct vfsync_info info;
649 bzero(&info, sizeof(info));
651 if ((info.checkdef = checkdef) == NULL)
654 crit_enter_id("vfsync");
659 * Lazy (filesystem syncer typ) Asynchronous plus limit the
660 * number of data (not meta) pages we try to flush to 1MB.
661 * A non-zero return means that lazy limit was reached.
663 info.lazylimit = 1024 * 1024;
665 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
666 vfsync_lazy_range_cmp, vfsync_bp, &info);
667 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
668 vfsync_meta_only_cmp, vfsync_bp, &info);
671 else if (!RB_EMPTY(&vp->v_rbdirty_tree))
672 vn_syncer_add_to_worklist(vp, 1);
677 * Asynchronous. Do a data-only pass and a meta-only pass.
680 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
682 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp,
688 * Synchronous. Do a data-only pass, then a meta-data+data
689 * pass, then additional integrated passes to try to get
690 * all the dependancies flushed.
692 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
694 error = vfsync_wait_output(vp, waitoutput);
696 info.skippedbufs = 0;
697 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
699 error = vfsync_wait_output(vp, waitoutput);
700 if (info.skippedbufs)
701 kprintf("Warning: vfsync skipped %d dirty bufs in pass2!\n", info.skippedbufs);
703 while (error == 0 && passes > 0 &&
704 !RB_EMPTY(&vp->v_rbdirty_tree)) {
706 info.synchronous = 1;
709 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
715 error = vfsync_wait_output(vp, waitoutput);
719 crit_exit_id("vfsync");
724 vfsync_wait_output(struct vnode *vp, int (*waitoutput)(struct vnode *, struct thread *))
728 while (vp->v_track_write.bk_active) {
729 vp->v_track_write.bk_waitflag = 1;
730 tsleep(&vp->v_track_write, 0, "fsfsn", 0);
733 error = waitoutput(vp, curthread);
738 vfsync_data_only_cmp(struct buf *bp, void *data)
740 if (bp->b_loffset < 0)
746 vfsync_meta_only_cmp(struct buf *bp, void *data)
748 if (bp->b_loffset < 0)
754 vfsync_lazy_range_cmp(struct buf *bp, void *data)
756 struct vfsync_info *info = data;
757 if (bp->b_loffset < info->vp->v_lazyw)
763 vfsync_bp(struct buf *bp, void *data)
765 struct vfsync_info *info = data;
766 struct vnode *vp = info->vp;
770 * if syncdeps is not set we do not try to write buffers which have
773 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp))
777 * Ignore buffers that we cannot immediately lock. XXX
779 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
780 kprintf("Warning: vfsync_bp skipping dirty buffer %p\n", bp);
784 if ((bp->b_flags & B_DELWRI) == 0)
785 panic("vfsync_bp: buffer not dirty");
787 panic("vfsync_bp: buffer vp mismatch");
790 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
791 * has been written but an additional handshake with the device
792 * is required before we can dispose of the buffer. We have no idea
793 * how to do this so we have to skip these buffers.
795 if (bp->b_flags & B_NEEDCOMMIT) {
800 if (info->synchronous) {
802 * Synchronous flushing. An error may be returned.
805 crit_exit_id("vfsync");
807 crit_enter_id("vfsync");
810 * Asynchronous flushing. A negative return value simply
811 * stops the scan and is not considered an error. We use
812 * this to support limited MNT_LAZY flushes.
814 vp->v_lazyw = bp->b_loffset;
815 if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) {
816 info->lazycount += vfs_bio_awrite(bp);
818 info->lazycount += bp->b_bufsize;
820 crit_exit_id("vfsync");
822 crit_enter_id("vfsync");
824 if (info->lazylimit && info->lazycount >= info->lazylimit)
833 * Associate a buffer with a vnode.
836 bgetvp(struct vnode *vp, struct buf *bp)
838 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
839 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0);
843 * Insert onto list for new vnode.
847 bp->b_flags |= B_HASHED;
848 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp))
849 panic("reassignbuf: dup lblk vp %p bp %p", vp, bp);
851 bp->b_flags |= B_VNCLEAN;
852 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp))
853 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp);
858 * Disassociate a buffer from a vnode.
861 brelvp(struct buf *bp)
865 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
868 * Delete from old vnode list, if on one.
872 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) {
873 if (bp->b_flags & B_VNDIRTY)
874 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
876 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
877 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN);
879 if (bp->b_flags & B_HASHED) {
880 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp);
881 bp->b_flags &= ~B_HASHED;
883 if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree)) {
884 vp->v_flag &= ~VONWORKLST;
885 LIST_REMOVE(vp, v_synclist);
893 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI.
894 * This routine is called when the state of the B_DELWRI bit is changed.
897 reassignbuf(struct buf *bp)
899 struct vnode *vp = bp->b_vp;
902 KKASSERT(vp != NULL);
906 * B_PAGING flagged buffers cannot be reassigned because their vp
907 * is not fully linked in.
909 if (bp->b_flags & B_PAGING)
910 panic("cannot reassign paging buffer");
913 if (bp->b_flags & B_DELWRI) {
915 * Move to the dirty list, add the vnode to the worklist
917 if (bp->b_flags & B_VNCLEAN) {
918 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
919 bp->b_flags &= ~B_VNCLEAN;
921 if ((bp->b_flags & B_VNDIRTY) == 0) {
922 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) {
923 panic("reassignbuf: dup lblk vp %p bp %p",
926 bp->b_flags |= B_VNDIRTY;
928 if ((vp->v_flag & VONWORKLST) == 0) {
929 switch (vp->v_type) {
936 vp->v_rdev->si_mountpoint != NULL) {
944 vn_syncer_add_to_worklist(vp, delay);
948 * Move to the clean list, remove the vnode from the worklist
949 * if no dirty blocks remain.
951 if (bp->b_flags & B_VNDIRTY) {
952 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
953 bp->b_flags &= ~B_VNDIRTY;
955 if ((bp->b_flags & B_VNCLEAN) == 0) {
956 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) {
957 panic("reassignbuf: dup lblk vp %p bp %p",
960 bp->b_flags |= B_VNCLEAN;
962 if ((vp->v_flag & VONWORKLST) &&
963 RB_EMPTY(&vp->v_rbdirty_tree)) {
964 vp->v_flag &= ~VONWORKLST;
965 LIST_REMOVE(vp, v_synclist);
972 * Create a vnode for a block device.
973 * Used for mounting the root file system.
976 bdevvp(cdev_t dev, struct vnode **vpp)
986 error = getspecialvnode(VT_NON, NULL, &spec_vnode_vops_p, &nvp, 0, 0);
993 vp->v_umajor = dev->si_umajor;
994 vp->v_uminor = dev->si_uminor;
1001 v_associate_rdev(struct vnode *vp, cdev_t dev)
1007 if (dev_is_good(dev) == 0)
1009 KKASSERT(vp->v_rdev == NULL);
1012 vp->v_rdev = reference_dev(dev);
1013 lwkt_gettoken(&ilock, &spechash_token);
1014 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext);
1015 lwkt_reltoken(&ilock);
1020 v_release_rdev(struct vnode *vp)
1025 if ((dev = vp->v_rdev) != NULL) {
1026 lwkt_gettoken(&ilock, &spechash_token);
1027 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext);
1030 lwkt_reltoken(&ilock);
1035 * Add a vnode to the alias list hung off the cdev_t. We only associate
1036 * the device number with the vnode. The actual device is not associated
1037 * until the vnode is opened (usually in spec_open()), and will be
1038 * disassociated on last close.
1041 addaliasu(struct vnode *nvp, int x, int y)
1043 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1044 panic("addaliasu on non-special vnode");
1050 * Disassociate a vnode from its underlying filesystem.
1052 * The vnode must be VX locked and referenced. In all normal situations
1053 * there are no active references. If vclean_vxlocked() is called while
1054 * there are active references, the vnode is being ripped out and we have
1055 * to call VOP_CLOSE() as appropriate before we can reclaim it.
1058 vclean_vxlocked(struct vnode *vp, int flags)
1065 * If the vnode has already been reclaimed we have nothing to do.
1067 if (vp->v_flag & VRECLAIMED)
1069 vp->v_flag |= VRECLAIMED;
1072 * Scrap the vfs cache
1074 while (cache_inval_vp(vp, 0) != 0) {
1075 kprintf("Warning: vnode %p clean/cache_resolution race detected\n", vp);
1076 tsleep(vp, 0, "vclninv", 2);
1080 * Check to see if the vnode is in use. If so we have to reference it
1081 * before we clean it out so that its count cannot fall to zero and
1082 * generate a race against ourselves to recycle it.
1084 active = sysref_isactive(&vp->v_sysref);
1087 * Clean out any buffers associated with the vnode and destroy its
1088 * object, if it has one.
1090 vinvalbuf(vp, V_SAVE, 0, 0);
1093 * If purging an active vnode (typically during a forced unmount
1094 * or reboot), it must be closed and deactivated before being
1095 * reclaimed. This isn't really all that safe, but what can
1098 * Note that neither of these routines unlocks the vnode.
1100 if (active && (flags & DOCLOSE)) {
1101 while ((n = vp->v_opencount) != 0) {
1102 if (vp->v_writecount)
1103 VOP_CLOSE(vp, FWRITE|FNONBLOCK);
1105 VOP_CLOSE(vp, FNONBLOCK);
1106 if (vp->v_opencount == n) {
1107 kprintf("Warning: unable to force-close"
1115 * If the vnode has not been deactivated, deactivated it. Deactivation
1116 * can create new buffers and VM pages so we have to call vinvalbuf()
1117 * again to make sure they all get flushed.
1119 * This can occur if a file with a link count of 0 needs to be
1122 if ((vp->v_flag & VINACTIVE) == 0) {
1123 vp->v_flag |= VINACTIVE;
1125 vinvalbuf(vp, V_SAVE, 0, 0);
1129 * If the vnode has an object, destroy it.
1131 if ((object = vp->v_object) != NULL) {
1132 if (object->ref_count == 0) {
1133 if ((object->flags & OBJ_DEAD) == 0)
1134 vm_object_terminate(object);
1136 vm_pager_deallocate(object);
1138 vp->v_flag &= ~VOBJBUF;
1140 KKASSERT((vp->v_flag & VOBJBUF) == 0);
1143 * Reclaim the vnode.
1145 if (VOP_RECLAIM(vp))
1146 panic("vclean: cannot reclaim");
1149 * Done with purge, notify sleepers of the grim news.
1151 vp->v_ops = &dead_vnode_vops_p;
1156 * If we are destroying an active vnode, reactivate it now that
1157 * we have reassociated it with deadfs. This prevents the system
1158 * from crashing on the vnode due to it being unexpectedly marked
1159 * as inactive or reclaimed.
1161 if (active && (flags & DOCLOSE)) {
1162 vp->v_flag &= ~(VINACTIVE|VRECLAIMED);
1167 * Eliminate all activity associated with the requested vnode
1168 * and with all vnodes aliased to the requested vnode.
1170 * The vnode must be referenced and vx_lock()'d
1172 * revoke { struct vnode *a_vp, int a_flags }
1175 vop_stdrevoke(struct vop_revoke_args *ap)
1177 struct vnode *vp, *vq;
1181 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
1186 * If the vnode is already dead don't try to revoke it
1188 if (vp->v_flag & VRECLAIMED)
1192 * If the vnode has a device association, scrap all vnodes associated
1193 * with the device. Don't let the device disappear on us while we
1194 * are scrapping the vnodes.
1196 * The passed vp will probably show up in the list, do not VX lock
1199 if (vp->v_type != VCHR)
1201 if ((dev = vp->v_rdev) == NULL) {
1202 if ((dev = get_dev(vp->v_umajor, vp->v_uminor)) == NULL)
1206 lwkt_gettoken(&ilock, &spechash_token);
1207 while ((vq = SLIST_FIRST(&dev->si_hlist)) != NULL) {
1210 if (vq == SLIST_FIRST(&dev->si_hlist))
1215 lwkt_reltoken(&ilock);
1221 * This is called when the object underlying a vnode is being destroyed,
1222 * such as in a remove(). Try to recycle the vnode immediately if the
1223 * only active reference is our reference.
1226 vrecycle(struct vnode *vp)
1228 if (vp->v_sysref.refcnt <= 1) {
1236 * Eliminate all activity associated with a vnode in preparation for reuse.
1238 * The vnode must be VX locked and refd and will remain VX locked and refd
1239 * on return. This routine may be called with the vnode in any state, as
1240 * long as it is VX locked. The vnode will be cleaned out and marked
1241 * VRECLAIMED but will not actually be reused until all existing refs and
1244 * NOTE: This routine may be called on a vnode which has not yet been
1245 * already been deactivated (VOP_INACTIVE), or on a vnode which has
1246 * already been reclaimed.
1248 * This routine is not responsible for placing us back on the freelist.
1249 * Instead, it happens automatically when the caller releases the VX lock
1250 * (assuming there aren't any other references).
1254 vgone_vxlocked(struct vnode *vp)
1257 * assert that the VX lock is held. This is an absolute requirement
1258 * now for vgone_vxlocked() to be called.
1260 KKASSERT(vp->v_lock.lk_exclusivecount == 1);
1263 * Clean out the filesystem specific data and set the VRECLAIMED
1264 * bit. Also deactivate the vnode if necessary.
1266 vclean_vxlocked(vp, DOCLOSE);
1269 * Delete from old mount point vnode list, if on one.
1271 if (vp->v_mount != NULL)
1272 insmntque(vp, NULL);
1275 * If special device, remove it from special device alias list
1276 * if it is on one. This should normally only occur if a vnode is
1277 * being revoked as the device should otherwise have been released
1280 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
1291 * Lookup a vnode by device number.
1294 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp)
1299 lwkt_gettoken(&ilock, &spechash_token);
1300 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1301 if (type == vp->v_type) {
1303 lwkt_reltoken(&ilock);
1307 lwkt_reltoken(&ilock);
1312 * Calculate the total number of references to a special device. This
1313 * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1314 * an overloaded field. Since udev2dev can now return NULL, we have
1315 * to check for a NULL v_rdev.
1318 count_dev(cdev_t dev)
1324 if (SLIST_FIRST(&dev->si_hlist)) {
1325 lwkt_gettoken(&ilock, &spechash_token);
1326 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1327 if (vp->v_sysref.refcnt > 0)
1328 count += vp->v_sysref.refcnt;
1330 lwkt_reltoken(&ilock);
1336 count_udev(int x, int y)
1340 if ((dev = get_dev(x, y)) == NULL)
1342 return(count_dev(dev));
1346 vcount(struct vnode *vp)
1348 if (vp->v_rdev == NULL)
1350 return(count_dev(vp->v_rdev));
1354 * Initialize VMIO for a vnode. This routine MUST be called before a
1355 * VFS can issue buffer cache ops on a vnode. It is typically called
1356 * when a vnode is initialized from its inode.
1359 vinitvmio(struct vnode *vp, off_t filesize)
1365 if ((object = vp->v_object) == NULL) {
1366 object = vnode_pager_alloc(vp, filesize, 0, 0);
1368 * Dereference the reference we just created. This assumes
1369 * that the object is associated with the vp.
1371 object->ref_count--;
1374 if (object->flags & OBJ_DEAD) {
1376 vm_object_dead_sleep(object, "vodead");
1377 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1381 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object"));
1382 vp->v_flag |= VOBJBUF;
1388 * Print out a description of a vnode.
1390 static char *typename[] =
1391 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1394 vprint(char *label, struct vnode *vp)
1399 kprintf("%s: %p: ", label, (void *)vp);
1401 kprintf("%p: ", (void *)vp);
1402 kprintf("type %s, sysrefs %d, writecount %d, holdcnt %d,",
1403 typename[vp->v_type],
1404 vp->v_sysref.refcnt, vp->v_writecount, vp->v_auxrefs);
1406 if (vp->v_flag & VROOT)
1407 strcat(buf, "|VROOT");
1408 if (vp->v_flag & VTEXT)
1409 strcat(buf, "|VTEXT");
1410 if (vp->v_flag & VSYSTEM)
1411 strcat(buf, "|VSYSTEM");
1412 if (vp->v_flag & VFREE)
1413 strcat(buf, "|VFREE");
1414 if (vp->v_flag & VOBJBUF)
1415 strcat(buf, "|VOBJBUF");
1417 kprintf(" flags (%s)", &buf[1]);
1418 if (vp->v_data == NULL) {
1427 #include <ddb/ddb.h>
1429 static int db_show_locked_vnodes(struct mount *mp, void *data);
1432 * List all of the locked vnodes in the system.
1433 * Called when debugging the kernel.
1435 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
1437 kprintf("Locked vnodes\n");
1438 mountlist_scan(db_show_locked_vnodes, NULL,
1439 MNTSCAN_FORWARD|MNTSCAN_NOBUSY);
1443 db_show_locked_vnodes(struct mount *mp, void *data __unused)
1447 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
1448 if (vn_islocked(vp))
1449 vprint((char *)0, vp);
1456 * Top level filesystem related information gathering.
1458 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
1461 vfs_sysctl(SYSCTL_HANDLER_ARGS)
1463 int *name = (int *)arg1 - 1; /* XXX */
1464 u_int namelen = arg2 + 1; /* XXX */
1465 struct vfsconf *vfsp;
1467 #if 1 || defined(COMPAT_PRELITE2)
1468 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1470 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
1474 /* all sysctl names at this level are at least name and field */
1476 return (ENOTDIR); /* overloaded */
1477 if (name[0] != VFS_GENERIC) {
1478 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1479 if (vfsp->vfc_typenum == name[0])
1482 return (EOPNOTSUPP);
1483 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1484 oldp, oldlenp, newp, newlen, p));
1488 case VFS_MAXTYPENUM:
1491 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
1494 return (ENOTDIR); /* overloaded */
1495 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1496 if (vfsp->vfc_typenum == name[2])
1499 return (EOPNOTSUPP);
1500 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
1502 return (EOPNOTSUPP);
1505 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
1506 "Generic filesystem");
1508 #if 1 || defined(COMPAT_PRELITE2)
1511 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
1514 struct vfsconf *vfsp;
1515 struct ovfsconf ovfs;
1517 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
1518 bzero(&ovfs, sizeof(ovfs));
1519 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
1520 strcpy(ovfs.vfc_name, vfsp->vfc_name);
1521 ovfs.vfc_index = vfsp->vfc_typenum;
1522 ovfs.vfc_refcount = vfsp->vfc_refcount;
1523 ovfs.vfc_flags = vfsp->vfc_flags;
1524 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
1531 #endif /* 1 || COMPAT_PRELITE2 */
1534 * Check to see if a filesystem is mounted on a block device.
1537 vfs_mountedon(struct vnode *vp)
1541 if ((dev = vp->v_rdev) == NULL) {
1542 if (vp->v_type != VBLK)
1543 dev = get_dev(vp->v_uminor, vp->v_umajor);
1545 if (dev != NULL && dev->si_mountpoint)
1551 * Unmount all filesystems. The list is traversed in reverse order
1552 * of mounting to avoid dependencies.
1555 static int vfs_umountall_callback(struct mount *mp, void *data);
1558 vfs_unmountall(void)
1563 count = mountlist_scan(vfs_umountall_callback,
1564 NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY);
1570 vfs_umountall_callback(struct mount *mp, void *data)
1574 error = dounmount(mp, MNT_FORCE);
1576 mountlist_remove(mp);
1577 kprintf("unmount of filesystem mounted from %s failed (",
1578 mp->mnt_stat.f_mntfromname);
1582 kprintf("%d)\n", error);
1588 * Build hash lists of net addresses and hang them off the mount point.
1589 * Called by ufs_mount() to set up the lists of export addresses.
1592 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1593 struct export_args *argp)
1596 struct radix_node_head *rnh;
1598 struct radix_node *rn;
1599 struct sockaddr *saddr, *smask = 0;
1603 if (argp->ex_addrlen == 0) {
1604 if (mp->mnt_flag & MNT_DEFEXPORTED)
1606 np = &nep->ne_defexported;
1607 np->netc_exflags = argp->ex_flags;
1608 np->netc_anon = argp->ex_anon;
1609 np->netc_anon.cr_ref = 1;
1610 mp->mnt_flag |= MNT_DEFEXPORTED;
1614 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
1616 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
1619 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1620 np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK | M_ZERO);
1621 saddr = (struct sockaddr *) (np + 1);
1622 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
1624 if (saddr->sa_len > argp->ex_addrlen)
1625 saddr->sa_len = argp->ex_addrlen;
1626 if (argp->ex_masklen) {
1627 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1628 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
1631 if (smask->sa_len > argp->ex_masklen)
1632 smask->sa_len = argp->ex_masklen;
1634 i = saddr->sa_family;
1635 if ((rnh = nep->ne_rtable[i]) == 0) {
1637 * Seems silly to initialize every AF when most are not used,
1638 * do so on demand here
1640 SLIST_FOREACH(dom, &domains, dom_next)
1641 if (dom->dom_family == i && dom->dom_rtattach) {
1642 dom->dom_rtattach((void **) &nep->ne_rtable[i],
1646 if ((rnh = nep->ne_rtable[i]) == 0) {
1651 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh,
1653 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */
1657 np->netc_exflags = argp->ex_flags;
1658 np->netc_anon = argp->ex_anon;
1659 np->netc_anon.cr_ref = 1;
1662 kfree(np, M_NETADDR);
1668 vfs_free_netcred(struct radix_node *rn, void *w)
1670 struct radix_node_head *rnh = (struct radix_node_head *) w;
1672 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
1673 kfree((caddr_t) rn, M_NETADDR);
1678 * Free the net address hash lists that are hanging off the mount points.
1681 vfs_free_addrlist(struct netexport *nep)
1684 struct radix_node_head *rnh;
1686 for (i = 0; i <= AF_MAX; i++)
1687 if ((rnh = nep->ne_rtable[i])) {
1688 (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
1690 kfree((caddr_t) rnh, M_RTABLE);
1691 nep->ne_rtable[i] = 0;
1696 vfs_export(struct mount *mp, struct netexport *nep, struct export_args *argp)
1700 if (argp->ex_flags & MNT_DELEXPORT) {
1701 if (mp->mnt_flag & MNT_EXPUBLIC) {
1702 vfs_setpublicfs(NULL, NULL, NULL);
1703 mp->mnt_flag &= ~MNT_EXPUBLIC;
1705 vfs_free_addrlist(nep);
1706 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
1708 if (argp->ex_flags & MNT_EXPORTED) {
1709 if (argp->ex_flags & MNT_EXPUBLIC) {
1710 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
1712 mp->mnt_flag |= MNT_EXPUBLIC;
1714 if ((error = vfs_hang_addrlist(mp, nep, argp)))
1716 mp->mnt_flag |= MNT_EXPORTED;
1723 * Set the publicly exported filesystem (WebNFS). Currently, only
1724 * one public filesystem is possible in the spec (RFC 2054 and 2055)
1727 vfs_setpublicfs(struct mount *mp, struct netexport *nep,
1728 struct export_args *argp)
1735 * mp == NULL -> invalidate the current info, the FS is
1736 * no longer exported. May be called from either vfs_export
1737 * or unmount, so check if it hasn't already been done.
1740 if (nfs_pub.np_valid) {
1741 nfs_pub.np_valid = 0;
1742 if (nfs_pub.np_index != NULL) {
1743 FREE(nfs_pub.np_index, M_TEMP);
1744 nfs_pub.np_index = NULL;
1751 * Only one allowed at a time.
1753 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
1757 * Get real filehandle for root of exported FS.
1759 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
1760 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
1762 if ((error = VFS_ROOT(mp, &rvp)))
1765 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
1771 * If an indexfile was specified, pull it in.
1773 if (argp->ex_indexfile != NULL) {
1776 error = vn_get_namelen(rvp, &namelen);
1779 MALLOC(nfs_pub.np_index, char *, namelen, M_TEMP,
1781 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
1782 namelen, (size_t *)0);
1785 * Check for illegal filenames.
1787 for (cp = nfs_pub.np_index; *cp; cp++) {
1795 FREE(nfs_pub.np_index, M_TEMP);
1800 nfs_pub.np_mount = mp;
1801 nfs_pub.np_valid = 1;
1806 vfs_export_lookup(struct mount *mp, struct netexport *nep,
1807 struct sockaddr *nam)
1810 struct radix_node_head *rnh;
1811 struct sockaddr *saddr;
1814 if (mp->mnt_flag & MNT_EXPORTED) {
1816 * Lookup in the export list first.
1820 rnh = nep->ne_rtable[saddr->sa_family];
1822 np = (struct netcred *)
1823 (*rnh->rnh_matchaddr)((char *)saddr,
1825 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
1830 * If no address match, use the default if it exists.
1832 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
1833 np = &nep->ne_defexported;
1839 * perform msync on all vnodes under a mount point. The mount point must
1840 * be locked. This code is also responsible for lazy-freeing unreferenced
1841 * vnodes whos VM objects no longer contain pages.
1843 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
1845 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked,
1846 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it
1847 * way up in this high level function.
1849 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
1850 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data);
1853 vfs_msync(struct mount *mp, int flags)
1857 vmsc_flags = VMSC_GETVP;
1858 if (flags != MNT_WAIT)
1859 vmsc_flags |= VMSC_NOWAIT;
1860 vmntvnodescan(mp, vmsc_flags, vfs_msync_scan1, vfs_msync_scan2,
1865 * scan1 is a fast pre-check. There could be hundreds of thousands of
1866 * vnodes, we cannot afford to do anything heavy weight until we have a
1867 * fairly good indication that there is work to do.
1871 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
1873 int flags = (int)data;
1875 if ((vp->v_flag & VRECLAIMED) == 0) {
1876 if (vshouldmsync(vp))
1877 return(0); /* call scan2 */
1878 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
1879 (vp->v_flag & VOBJDIRTY) &&
1880 (flags == MNT_WAIT || vn_islocked(vp) == 0)) {
1881 return(0); /* call scan2 */
1886 * do not call scan2, continue the loop
1892 * This callback is handed a locked vnode.
1896 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data)
1899 int flags = (int)data;
1901 if (vp->v_flag & VRECLAIMED)
1904 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) {
1905 if ((obj = vp->v_object) != NULL) {
1906 vm_object_page_clean(obj, 0, 0,
1907 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
1914 * Record a process's interest in events which might happen to
1915 * a vnode. Because poll uses the historic select-style interface
1916 * internally, this routine serves as both the ``check for any
1917 * pending events'' and the ``record my interest in future events''
1918 * functions. (These are done together, while the lock is held,
1919 * to avoid race conditions.)
1922 vn_pollrecord(struct vnode *vp, int events)
1926 KKASSERT(curthread->td_proc != NULL);
1928 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
1929 if (vp->v_pollinfo.vpi_revents & events) {
1931 * This leaves events we are not interested
1932 * in available for the other process which
1933 * which presumably had requested them
1934 * (otherwise they would never have been
1937 events &= vp->v_pollinfo.vpi_revents;
1938 vp->v_pollinfo.vpi_revents &= ~events;
1940 lwkt_reltoken(&ilock);
1943 vp->v_pollinfo.vpi_events |= events;
1944 selrecord(curthread, &vp->v_pollinfo.vpi_selinfo);
1945 lwkt_reltoken(&ilock);
1950 * Note the occurrence of an event. If the VN_POLLEVENT macro is used,
1951 * it is possible for us to miss an event due to race conditions, but
1952 * that condition is expected to be rare, so for the moment it is the
1953 * preferred interface.
1956 vn_pollevent(struct vnode *vp, int events)
1960 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
1961 if (vp->v_pollinfo.vpi_events & events) {
1963 * We clear vpi_events so that we don't
1964 * call selwakeup() twice if two events are
1965 * posted before the polling process(es) is
1966 * awakened. This also ensures that we take at
1967 * most one selwakeup() if the polling process
1968 * is no longer interested. However, it does
1969 * mean that only one event can be noticed at
1970 * a time. (Perhaps we should only clear those
1971 * event bits which we note?) XXX
1973 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */
1974 vp->v_pollinfo.vpi_revents |= events;
1975 selwakeup(&vp->v_pollinfo.vpi_selinfo);
1977 lwkt_reltoken(&ilock);
1981 * Wake up anyone polling on vp because it is being revoked.
1982 * This depends on dead_poll() returning POLLHUP for correct
1986 vn_pollgone(struct vnode *vp)
1990 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
1991 if (vp->v_pollinfo.vpi_events) {
1992 vp->v_pollinfo.vpi_events = 0;
1993 selwakeup(&vp->v_pollinfo.vpi_selinfo);
1995 lwkt_reltoken(&ilock);
1999 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened
2000 * (or v_rdev might be NULL).
2003 vn_todev(struct vnode *vp)
2005 if (vp->v_type != VBLK && vp->v_type != VCHR)
2007 KKASSERT(vp->v_rdev != NULL);
2008 return (vp->v_rdev);
2012 * Check if vnode represents a disk device. The vnode does not need to be
2016 vn_isdisk(struct vnode *vp, int *errp)
2020 if (vp->v_type != VCHR) {
2026 if ((dev = vp->v_rdev) == NULL)
2027 dev = get_dev(vp->v_umajor, vp->v_uminor);
2034 if (dev_is_good(dev) == 0) {
2039 if ((dev_dflags(dev) & D_DISK) == 0) {
2050 vn_get_namelen(struct vnode *vp, int *namelen)
2052 int error, retval[2];
2054 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval);
2062 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type,
2063 uint16_t d_namlen, const char *d_name)
2068 len = _DIRENT_RECLEN(d_namlen);
2069 if (len > uio->uio_resid)
2072 dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO);
2075 dp->d_namlen = d_namlen;
2076 dp->d_type = d_type;
2077 bcopy(d_name, dp->d_name, d_namlen);
2079 *error = uiomove((caddr_t)dp, len, uio);