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 1.118 2008/09/17 21:44:18 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>
57 #include <sys/kernel.h>
58 #include <sys/kthread.h>
59 #include <sys/malloc.h>
61 #include <sys/mount.h>
64 #include <sys/reboot.h>
65 #include <sys/socket.h>
67 #include <sys/sysctl.h>
68 #include <sys/syslog.h>
69 #include <sys/unistd.h>
70 #include <sys/vmmeter.h>
71 #include <sys/vnode.h>
73 #include <machine/limits.h>
76 #include <vm/vm_object.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_kern.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_page.h>
82 #include <vm/vm_pager.h>
83 #include <vm/vnode_pager.h>
84 #include <vm/vm_zone.h>
87 #include <sys/thread2.h>
88 #include <sys/sysref2.h>
89 #include <sys/mplock2.h>
91 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
94 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
96 enum vtype iftovt_tab[16] = {
97 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
98 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
100 int vttoif_tab[9] = {
101 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
102 S_IFSOCK, S_IFIFO, S_IFMT,
105 static int reassignbufcalls;
106 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW,
107 &reassignbufcalls, 0, "");
108 static int reassignbufloops;
109 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW,
110 &reassignbufloops, 0, "");
111 static int reassignbufsortgood;
112 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW,
113 &reassignbufsortgood, 0, "");
114 static int reassignbufsortbad;
115 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW,
116 &reassignbufsortbad, 0, "");
117 static int reassignbufmethod = 1;
118 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW,
119 &reassignbufmethod, 0, "");
120 static int check_buf_overlap = 2; /* invasive check */
121 SYSCTL_INT(_vfs, OID_AUTO, check_buf_overlap, CTLFLAG_RW,
122 &check_buf_overlap, 0, "");
124 int nfs_mount_type = -1;
125 static struct lwkt_token spechash_token;
126 struct nfs_public nfs_pub; /* publicly exported FS */
129 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
130 &desiredvnodes, 0, "Maximum number of vnodes");
132 static void vfs_free_addrlist (struct netexport *nep);
133 static int vfs_free_netcred (struct radix_node *rn, void *w);
134 static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
135 const struct export_args *argp);
138 * Red black tree functions
140 static int rb_buf_compare(struct buf *b1, struct buf *b2);
141 RB_GENERATE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t, b_loffset);
142 RB_GENERATE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t, b_loffset);
145 rb_buf_compare(struct buf *b1, struct buf *b2)
147 if (b1->b_loffset < b2->b_loffset)
149 if (b1->b_loffset > b2->b_loffset)
155 * Returns non-zero if the vnode is a candidate for lazy msyncing.
158 vshouldmsync(struct vnode *vp)
160 if (vp->v_auxrefs != 0 || vp->v_sysref.refcnt > 0)
161 return (0); /* other holders */
163 (vp->v_object->ref_count || vp->v_object->resident_page_count)) {
170 * Initialize the vnode management data structures.
172 * Called from vfsinit()
181 * Desiredvnodes is kern.maxvnodes. We want to scale it
182 * according to available system memory but we may also have
183 * to limit it based on available KVM, which is capped on 32 bit
186 * WARNING! For machines with 64-256M of ram we have to be sure
187 * that the default limit scales down well due to HAMMER
188 * taking up significantly more memory per-vnode vs UFS.
189 * We want around ~5800 on a 128M machine.
191 factor1 = 20 * (sizeof(struct vm_object) + sizeof(struct vnode));
192 factor2 = 22 * (sizeof(struct vm_object) + sizeof(struct vnode));
194 imin((int64_t)vmstats.v_page_count * PAGE_SIZE / factor1,
196 desiredvnodes = imax(desiredvnodes, maxproc * 8);
198 lwkt_token_init(&spechash_token, 1, "spechash");
202 * Knob to control the precision of file timestamps:
204 * 0 = seconds only; nanoseconds zeroed.
205 * 1 = seconds and nanoseconds, accurate within 1/HZ.
206 * 2 = seconds and nanoseconds, truncated to microseconds.
207 * >=3 = seconds and nanoseconds, maximum precision.
209 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
211 static int timestamp_precision = TSP_SEC;
212 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
213 ×tamp_precision, 0, "");
216 * Get a current timestamp.
221 vfs_timestamp(struct timespec *tsp)
225 switch (timestamp_precision) {
227 tsp->tv_sec = time_second;
235 TIMEVAL_TO_TIMESPEC(&tv, tsp);
245 * Set vnode attributes to VNOVAL
248 vattr_null(struct vattr *vap)
251 vap->va_size = VNOVAL;
252 vap->va_bytes = VNOVAL;
253 vap->va_mode = VNOVAL;
254 vap->va_nlink = VNOVAL;
255 vap->va_uid = VNOVAL;
256 vap->va_gid = VNOVAL;
257 vap->va_fsid = VNOVAL;
258 vap->va_fileid = VNOVAL;
259 vap->va_blocksize = VNOVAL;
260 vap->va_rmajor = VNOVAL;
261 vap->va_rminor = VNOVAL;
262 vap->va_atime.tv_sec = VNOVAL;
263 vap->va_atime.tv_nsec = VNOVAL;
264 vap->va_mtime.tv_sec = VNOVAL;
265 vap->va_mtime.tv_nsec = VNOVAL;
266 vap->va_ctime.tv_sec = VNOVAL;
267 vap->va_ctime.tv_nsec = VNOVAL;
268 vap->va_flags = VNOVAL;
269 vap->va_gen = VNOVAL;
271 /* va_*_uuid fields are only valid if related flags are set */
275 * Flush out and invalidate all buffers associated with a vnode.
279 static int vinvalbuf_bp(struct buf *bp, void *data);
281 struct vinvalbuf_bp_info {
289 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
291 struct vinvalbuf_bp_info info;
295 lwkt_gettoken(&vp->v_token);
298 * If we are being asked to save, call fsync to ensure that the inode
301 if (flags & V_SAVE) {
302 error = bio_track_wait(&vp->v_track_write, slpflag, slptimeo);
305 if (!RB_EMPTY(&vp->v_rbdirty_tree)) {
306 if ((error = VOP_FSYNC(vp, MNT_WAIT, 0)) != 0)
310 * Dirty bufs may be left or generated via races
311 * in circumstances where vinvalbuf() is called on
312 * a vnode not undergoing reclamation. Only
313 * panic if we are trying to reclaim the vnode.
315 if ((vp->v_flag & VRECLAIMED) &&
316 (bio_track_active(&vp->v_track_write) ||
317 !RB_EMPTY(&vp->v_rbdirty_tree))) {
318 panic("vinvalbuf: dirty bufs");
322 info.slptimeo = slptimeo;
323 info.lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
324 if (slpflag & PCATCH)
325 info.lkflags |= LK_PCATCH;
330 * Flush the buffer cache until nothing is left.
332 while (!RB_EMPTY(&vp->v_rbclean_tree) ||
333 !RB_EMPTY(&vp->v_rbdirty_tree)) {
334 error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, NULL,
335 vinvalbuf_bp, &info);
337 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
338 vinvalbuf_bp, &info);
343 * Wait for I/O completion. We may block in the pip code so we have
347 bio_track_wait(&vp->v_track_write, 0, 0);
348 if ((object = vp->v_object) != NULL) {
349 while (object->paging_in_progress)
350 vm_object_pip_sleep(object, "vnvlbx");
352 } while (bio_track_active(&vp->v_track_write));
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");
368 lwkt_reltoken(&vp->v_token);
373 vinvalbuf_bp(struct buf *bp, void *data)
375 struct vinvalbuf_bp_info *info = data;
378 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
379 error = BUF_TIMELOCK(bp, info->lkflags,
380 "vinvalbuf", info->slptimeo);
390 KKASSERT(bp->b_vp == info->vp);
393 * XXX Since there are no node locks for NFS, I
394 * believe there is a slight chance that a delayed
395 * write will occur while sleeping just above, so
396 * check for it. Note that vfs_bio_awrite expects
397 * buffers to reside on a queue, while bwrite() and
400 * NOTE: NO B_LOCKED CHECK. Also no buf_checkwrite()
401 * check. This code will write out the buffer, period.
403 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
404 (info->flags & V_SAVE)) {
405 if (bp->b_vp == info->vp) {
406 if (bp->b_flags & B_CLUSTEROK) {
416 } else if (info->flags & V_SAVE) {
418 * Cannot set B_NOCACHE on a clean buffer as this will
419 * destroy the VM backing store which might actually
420 * be dirty (and unsynchronized).
423 bp->b_flags |= (B_INVAL | B_RELBUF);
427 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
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)
449 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;
462 lwkt_gettoken(&vp->v_token);
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.
489 * It is possible to have in-progress I/O from buffers that were
490 * not part of the truncation. This should not happen if we
491 * are truncating to 0-length.
493 vnode_pager_setsize(vp, length);
494 bio_track_wait(&vp->v_track_write, 0, 0);
499 spin_lock_wr(&vp->v_spinlock);
500 filename = TAILQ_FIRST(&vp->v_namecache) ?
501 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?";
502 spin_unlock_wr(&vp->v_spinlock);
505 * Make sure no buffers were instantiated while we were trying
506 * to clean out the remaining VM pages. This could occur due
507 * to busy dirty VM pages being flushed out to disk.
510 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
511 vtruncbuf_bp_trunc_cmp,
512 vtruncbuf_bp_trunc, &truncloffset);
513 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
514 vtruncbuf_bp_trunc_cmp,
515 vtruncbuf_bp_trunc, &truncloffset);
517 kprintf("Warning: vtruncbuf(): Had to re-clean %d "
518 "left over buffers in %s\n", count, filename);
522 lwkt_reltoken(&vp->v_token);
528 * The callback buffer is beyond the new file EOF and must be destroyed.
529 * Note that the compare function must conform to the RB_SCAN's requirements.
533 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data)
535 if (bp->b_loffset >= *(off_t *)data)
542 vtruncbuf_bp_trunc(struct buf *bp, void *data)
545 * Do not try to use a buffer we cannot immediately lock, but sleep
546 * anyway to prevent a livelock. The code will loop until all buffers
549 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
550 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
554 bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE);
561 * Fsync all meta-data after truncating a file to be non-zero. Only metadata
562 * blocks (with a negative loffset) are scanned.
563 * Note that the compare function must conform to the RB_SCAN's requirements.
566 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data)
568 if (bp->b_loffset < 0)
574 vtruncbuf_bp_metasync(struct buf *bp, void *data)
576 struct vnode *vp = data;
578 if (bp->b_flags & B_DELWRI) {
580 * Do not try to use a buffer we cannot immediately lock,
581 * but sleep anyway to prevent a livelock. The code will
582 * loop until all buffers can be acted upon.
584 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
585 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
601 * vfsync - implements a multipass fsync on a file which understands
602 * dependancies and meta-data. The passed vnode must be locked. The
603 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY.
605 * When fsyncing data asynchronously just do one consolidated pass starting
606 * with the most negative block number. This may not get all the data due
609 * When fsyncing data synchronously do a data pass, then a metadata pass,
610 * then do additional data+metadata passes to try to get all the data out.
612 static int vfsync_wait_output(struct vnode *vp,
613 int (*waitoutput)(struct vnode *, struct thread *));
614 static int vfsync_data_only_cmp(struct buf *bp, void *data);
615 static int vfsync_meta_only_cmp(struct buf *bp, void *data);
616 static int vfsync_lazy_range_cmp(struct buf *bp, void *data);
617 static int vfsync_bp(struct buf *bp, void *data);
626 int (*checkdef)(struct buf *);
630 vfsync(struct vnode *vp, int waitfor, int passes,
631 int (*checkdef)(struct buf *),
632 int (*waitoutput)(struct vnode *, struct thread *))
634 struct vfsync_info info;
637 bzero(&info, sizeof(info));
639 if ((info.checkdef = checkdef) == NULL)
642 lwkt_gettoken(&vp->v_token);
647 * Lazy (filesystem syncer typ) Asynchronous plus limit the
648 * number of data (not meta) pages we try to flush to 1MB.
649 * A non-zero return means that lazy limit was reached.
651 info.lazylimit = 1024 * 1024;
653 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
654 vfsync_lazy_range_cmp, vfsync_bp, &info);
655 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
656 vfsync_meta_only_cmp, vfsync_bp, &info);
659 else if (!RB_EMPTY(&vp->v_rbdirty_tree))
660 vn_syncer_add_to_worklist(vp, 1);
665 * Asynchronous. Do a data-only pass and a meta-only pass.
668 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
670 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp,
676 * Synchronous. Do a data-only pass, then a meta-data+data
677 * pass, then additional integrated passes to try to get
678 * all the dependancies flushed.
680 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
682 error = vfsync_wait_output(vp, waitoutput);
684 info.skippedbufs = 0;
685 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
687 error = vfsync_wait_output(vp, waitoutput);
688 if (info.skippedbufs)
689 kprintf("Warning: vfsync skipped %d dirty bufs in pass2!\n", info.skippedbufs);
691 while (error == 0 && passes > 0 &&
692 !RB_EMPTY(&vp->v_rbdirty_tree)
695 info.synchronous = 1;
698 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
704 error = vfsync_wait_output(vp, waitoutput);
708 lwkt_reltoken(&vp->v_token);
713 vfsync_wait_output(struct vnode *vp,
714 int (*waitoutput)(struct vnode *, struct thread *))
718 error = bio_track_wait(&vp->v_track_write, 0, 0);
720 error = waitoutput(vp, curthread);
725 vfsync_data_only_cmp(struct buf *bp, void *data)
727 if (bp->b_loffset < 0)
733 vfsync_meta_only_cmp(struct buf *bp, void *data)
735 if (bp->b_loffset < 0)
741 vfsync_lazy_range_cmp(struct buf *bp, void *data)
743 struct vfsync_info *info = data;
744 if (bp->b_loffset < info->vp->v_lazyw)
750 vfsync_bp(struct buf *bp, void *data)
752 struct vfsync_info *info = data;
753 struct vnode *vp = info->vp;
757 * if syncdeps is not set we do not try to write buffers which have
760 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp))
764 * Ignore buffers that we cannot immediately lock. XXX
766 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
767 kprintf("Warning: vfsync_bp skipping dirty buffer %p\n", bp);
771 if ((bp->b_flags & B_DELWRI) == 0)
772 panic("vfsync_bp: buffer not dirty");
774 panic("vfsync_bp: buffer vp mismatch");
777 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
778 * has been written but an additional handshake with the device
779 * is required before we can dispose of the buffer. We have no idea
780 * how to do this so we have to skip these buffers.
782 if (bp->b_flags & B_NEEDCOMMIT) {
788 * Ask bioops if it is ok to sync
790 if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) {
796 if (info->synchronous) {
798 * Synchronous flushing. An error may be returned.
804 * Asynchronous flushing. A negative return value simply
805 * stops the scan and is not considered an error. We use
806 * this to support limited MNT_LAZY flushes.
808 vp->v_lazyw = bp->b_loffset;
809 if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) {
810 info->lazycount += vfs_bio_awrite(bp);
812 info->lazycount += bp->b_bufsize;
816 waitrunningbufspace();
817 if (info->lazylimit && info->lazycount >= info->lazylimit)
826 * Associate a buffer with a vnode.
831 bgetvp(struct vnode *vp, struct buf *bp, int testsize)
833 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
834 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0);
837 * Insert onto list for new vnode.
839 lwkt_gettoken(&vp->v_token);
840 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) {
841 lwkt_reltoken(&vp->v_token);
846 * Diagnostics (mainly for HAMMER debugging). Check for
847 * overlapping buffers.
849 if (check_buf_overlap) {
851 bx = buf_rb_hash_RB_PREV(bp);
853 if (bx->b_loffset + bx->b_bufsize > bp->b_loffset) {
854 kprintf("bgetvp: overlapl %016jx/%d %016jx "
856 (intmax_t)bx->b_loffset,
858 (intmax_t)bp->b_loffset,
860 if (check_buf_overlap > 1)
861 panic("bgetvp - overlapping buffer");
864 bx = buf_rb_hash_RB_NEXT(bp);
866 if (bp->b_loffset + testsize > bx->b_loffset) {
867 kprintf("bgetvp: overlapr %016jx/%d %016jx "
869 (intmax_t)bp->b_loffset,
871 (intmax_t)bx->b_loffset,
873 if (check_buf_overlap > 1)
874 panic("bgetvp - overlapping buffer");
879 bp->b_flags |= B_HASHED;
880 bp->b_flags |= B_VNCLEAN;
881 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp))
882 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp);
884 lwkt_reltoken(&vp->v_token);
889 * Disassociate a buffer from a vnode.
892 brelvp(struct buf *bp)
896 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
899 * Delete from old vnode list, if on one.
902 lwkt_gettoken(&vp->v_token);
903 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) {
904 if (bp->b_flags & B_VNDIRTY)
905 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
907 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
908 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN);
910 if (bp->b_flags & B_HASHED) {
911 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp);
912 bp->b_flags &= ~B_HASHED;
914 if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree)) {
915 vclrflags(vp, VONWORKLST);
916 LIST_REMOVE(vp, v_synclist);
919 lwkt_reltoken(&vp->v_token);
925 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI.
926 * This routine is called when the state of the B_DELWRI bit is changed.
931 reassignbuf(struct buf *bp)
933 struct vnode *vp = bp->b_vp;
936 KKASSERT(vp != NULL);
940 * B_PAGING flagged buffers cannot be reassigned because their vp
941 * is not fully linked in.
943 if (bp->b_flags & B_PAGING)
944 panic("cannot reassign paging buffer");
946 lwkt_gettoken(&vp->v_token);
947 if (bp->b_flags & B_DELWRI) {
949 * Move to the dirty list, add the vnode to the worklist
951 if (bp->b_flags & B_VNCLEAN) {
952 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
953 bp->b_flags &= ~B_VNCLEAN;
955 if ((bp->b_flags & B_VNDIRTY) == 0) {
956 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) {
957 panic("reassignbuf: dup lblk vp %p bp %p",
960 bp->b_flags |= B_VNDIRTY;
962 if ((vp->v_flag & VONWORKLST) == 0) {
963 switch (vp->v_type) {
970 vp->v_rdev->si_mountpoint != NULL) {
978 vn_syncer_add_to_worklist(vp, delay);
982 * Move to the clean list, remove the vnode from the worklist
983 * if no dirty blocks remain.
985 if (bp->b_flags & B_VNDIRTY) {
986 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
987 bp->b_flags &= ~B_VNDIRTY;
989 if ((bp->b_flags & B_VNCLEAN) == 0) {
990 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) {
991 panic("reassignbuf: dup lblk vp %p bp %p",
994 bp->b_flags |= B_VNCLEAN;
996 if ((vp->v_flag & VONWORKLST) &&
997 RB_EMPTY(&vp->v_rbdirty_tree)) {
998 vclrflags(vp, VONWORKLST);
999 LIST_REMOVE(vp, v_synclist);
1002 lwkt_reltoken(&vp->v_token);
1006 * Create a vnode for a block device.
1007 * Used for mounting the root file system.
1009 extern struct vop_ops *devfs_vnode_dev_vops_p;
1011 bdevvp(cdev_t dev, struct vnode **vpp)
1021 error = getspecialvnode(VT_NON, NULL, &devfs_vnode_dev_vops_p,
1032 v_associate_rdev(vp, dev);
1033 vp->v_umajor = dev->si_umajor;
1034 vp->v_uminor = dev->si_uminor;
1041 v_associate_rdev(struct vnode *vp, cdev_t dev)
1045 if (dev_is_good(dev) == 0)
1047 KKASSERT(vp->v_rdev == NULL);
1048 vp->v_rdev = reference_dev(dev);
1049 lwkt_gettoken(&spechash_token);
1050 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext);
1051 lwkt_reltoken(&spechash_token);
1056 v_release_rdev(struct vnode *vp)
1060 if ((dev = vp->v_rdev) != NULL) {
1061 lwkt_gettoken(&spechash_token);
1062 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext);
1065 lwkt_reltoken(&spechash_token);
1070 * Add a vnode to the alias list hung off the cdev_t. We only associate
1071 * the device number with the vnode. The actual device is not associated
1072 * until the vnode is opened (usually in spec_open()), and will be
1073 * disassociated on last close.
1076 addaliasu(struct vnode *nvp, int x, int y)
1078 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1079 panic("addaliasu on non-special vnode");
1085 * Simple call that a filesystem can make to try to get rid of a
1086 * vnode. It will fail if anyone is referencing the vnode (including
1089 * The filesystem can check whether its in-memory inode structure still
1090 * references the vp on return.
1093 vclean_unlocked(struct vnode *vp)
1096 if (sysref_isactive(&vp->v_sysref) == 0)
1102 * Disassociate a vnode from its underlying filesystem.
1104 * The vnode must be VX locked and referenced. In all normal situations
1105 * there are no active references. If vclean_vxlocked() is called while
1106 * there are active references, the vnode is being ripped out and we have
1107 * to call VOP_CLOSE() as appropriate before we can reclaim it.
1110 vclean_vxlocked(struct vnode *vp, int flags)
1117 * If the vnode has already been reclaimed we have nothing to do.
1119 if (vp->v_flag & VRECLAIMED)
1121 vsetflags(vp, VRECLAIMED);
1124 * Scrap the vfs cache
1126 while (cache_inval_vp(vp, 0) != 0) {
1127 kprintf("Warning: vnode %p clean/cache_resolution race detected\n", vp);
1128 tsleep(vp, 0, "vclninv", 2);
1132 * Check to see if the vnode is in use. If so we have to reference it
1133 * before we clean it out so that its count cannot fall to zero and
1134 * generate a race against ourselves to recycle it.
1136 active = sysref_isactive(&vp->v_sysref);
1139 * Clean out any buffers associated with the vnode and destroy its
1140 * object, if it has one.
1142 vinvalbuf(vp, V_SAVE, 0, 0);
1145 * If purging an active vnode (typically during a forced unmount
1146 * or reboot), it must be closed and deactivated before being
1147 * reclaimed. This isn't really all that safe, but what can
1150 * Note that neither of these routines unlocks the vnode.
1152 if (active && (flags & DOCLOSE)) {
1153 while ((n = vp->v_opencount) != 0) {
1154 if (vp->v_writecount)
1155 VOP_CLOSE(vp, FWRITE|FNONBLOCK);
1157 VOP_CLOSE(vp, FNONBLOCK);
1158 if (vp->v_opencount == n) {
1159 kprintf("Warning: unable to force-close"
1167 * If the vnode has not been deactivated, deactivated it. Deactivation
1168 * can create new buffers and VM pages so we have to call vinvalbuf()
1169 * again to make sure they all get flushed.
1171 * This can occur if a file with a link count of 0 needs to be
1174 * If the vnode is already dead don't try to deactivate it.
1176 if ((vp->v_flag & VINACTIVE) == 0) {
1177 vsetflags(vp, VINACTIVE);
1180 vinvalbuf(vp, V_SAVE, 0, 0);
1184 * If the vnode has an object, destroy it.
1186 if ((object = vp->v_object) != NULL) {
1187 lwkt_gettoken(&vm_token);
1188 KKASSERT(object == vp->v_object);
1189 if (object->ref_count == 0) {
1190 if ((object->flags & OBJ_DEAD) == 0)
1191 vm_object_terminate(object);
1193 vm_pager_deallocate(object);
1195 vclrflags(vp, VOBJBUF);
1196 lwkt_reltoken(&vm_token);
1198 KKASSERT((vp->v_flag & VOBJBUF) == 0);
1201 * Reclaim the vnode if not already dead.
1203 if (vp->v_mount && VOP_RECLAIM(vp))
1204 panic("vclean: cannot reclaim");
1207 * Done with purge, notify sleepers of the grim news.
1209 vp->v_ops = &dead_vnode_vops_p;
1214 * If we are destroying an active vnode, reactivate it now that
1215 * we have reassociated it with deadfs. This prevents the system
1216 * from crashing on the vnode due to it being unexpectedly marked
1217 * as inactive or reclaimed.
1219 if (active && (flags & DOCLOSE)) {
1220 vclrflags(vp, VINACTIVE | VRECLAIMED);
1225 * Eliminate all activity associated with the requested vnode
1226 * and with all vnodes aliased to the requested vnode.
1228 * The vnode must be referenced but should not be locked.
1231 vrevoke(struct vnode *vp, struct ucred *cred)
1239 * If the vnode has a device association, scrap all vnodes associated
1240 * with the device. Don't let the device disappear on us while we
1241 * are scrapping the vnodes.
1243 * The passed vp will probably show up in the list, do not VX lock
1246 * Releasing the vnode's rdev here can mess up specfs's call to
1247 * device close, so don't do it. The vnode has been disassociated
1248 * and the device will be closed after the last ref on the related
1249 * fp goes away (if not still open by e.g. the kernel).
1251 if (vp->v_type != VCHR) {
1252 error = fdrevoke(vp, DTYPE_VNODE, cred);
1255 if ((dev = vp->v_rdev) == NULL) {
1259 lwkt_gettoken(&spechash_token);
1261 vqn = SLIST_FIRST(&dev->si_hlist);
1264 while ((vq = vqn) != NULL) {
1265 vqn = SLIST_NEXT(vqn, v_cdevnext);
1268 fdrevoke(vq, DTYPE_VNODE, cred);
1269 /*v_release_rdev(vq);*/
1272 lwkt_reltoken(&spechash_token);
1279 * This is called when the object underlying a vnode is being destroyed,
1280 * such as in a remove(). Try to recycle the vnode immediately if the
1281 * only active reference is our reference.
1283 * Directory vnodes in the namecache with children cannot be immediately
1284 * recycled because numerous VOP_N*() ops require them to be stable.
1286 * To avoid recursive recycling from VOP_INACTIVE implemenetations this
1287 * function is a NOP if VRECLAIMED is already set.
1290 vrecycle(struct vnode *vp)
1292 if (vp->v_sysref.refcnt <= 1 && (vp->v_flag & VRECLAIMED) == 0) {
1293 if (cache_inval_vp_nonblock(vp))
1302 * Return the maximum I/O size allowed for strategy calls on VP.
1304 * If vp is VCHR or VBLK we dive the device, otherwise we use
1305 * the vp's mount info.
1308 vmaxiosize(struct vnode *vp)
1310 if (vp->v_type == VBLK || vp->v_type == VCHR) {
1311 return(vp->v_rdev->si_iosize_max);
1313 return(vp->v_mount->mnt_iosize_max);
1318 * Eliminate all activity associated with a vnode in preparation for reuse.
1320 * The vnode must be VX locked and refd and will remain VX locked and refd
1321 * on return. This routine may be called with the vnode in any state, as
1322 * long as it is VX locked. The vnode will be cleaned out and marked
1323 * VRECLAIMED but will not actually be reused until all existing refs and
1326 * NOTE: This routine may be called on a vnode which has not yet been
1327 * already been deactivated (VOP_INACTIVE), or on a vnode which has
1328 * already been reclaimed.
1330 * This routine is not responsible for placing us back on the freelist.
1331 * Instead, it happens automatically when the caller releases the VX lock
1332 * (assuming there aren't any other references).
1335 vgone_vxlocked(struct vnode *vp)
1338 * assert that the VX lock is held. This is an absolute requirement
1339 * now for vgone_vxlocked() to be called.
1341 KKASSERT(vp->v_lock.lk_exclusivecount == 1);
1346 * Clean out the filesystem specific data and set the VRECLAIMED
1347 * bit. Also deactivate the vnode if necessary.
1349 vclean_vxlocked(vp, DOCLOSE);
1352 * Delete from old mount point vnode list, if on one.
1354 if (vp->v_mount != NULL) {
1355 KKASSERT(vp->v_data == NULL);
1356 insmntque(vp, NULL);
1360 * If special device, remove it from special device alias list
1361 * if it is on one. This should normally only occur if a vnode is
1362 * being revoked as the device should otherwise have been released
1365 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
1377 * Lookup a vnode by device number.
1379 * Returns non-zero and *vpp set to a vref'd vnode on success.
1380 * Returns zero on failure.
1383 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp)
1387 lwkt_gettoken(&spechash_token);
1388 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1389 if (type == vp->v_type) {
1392 lwkt_reltoken(&spechash_token);
1396 lwkt_reltoken(&spechash_token);
1401 * Calculate the total number of references to a special device. This
1402 * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1403 * an overloaded field. Since udev2dev can now return NULL, we have
1404 * to check for a NULL v_rdev.
1407 count_dev(cdev_t dev)
1412 if (SLIST_FIRST(&dev->si_hlist)) {
1413 lwkt_gettoken(&spechash_token);
1414 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1415 count += vp->v_opencount;
1417 lwkt_reltoken(&spechash_token);
1423 vcount(struct vnode *vp)
1425 if (vp->v_rdev == NULL)
1427 return(count_dev(vp->v_rdev));
1431 * Initialize VMIO for a vnode. This routine MUST be called before a
1432 * VFS can issue buffer cache ops on a vnode. It is typically called
1433 * when a vnode is initialized from its inode.
1436 vinitvmio(struct vnode *vp, off_t filesize, int blksize, int boff)
1442 if ((object = vp->v_object) == NULL) {
1443 object = vnode_pager_alloc(vp, filesize, 0, 0, blksize, boff);
1445 * Dereference the reference we just created. This assumes
1446 * that the object is associated with the vp.
1448 object->ref_count--;
1451 if (object->flags & OBJ_DEAD) {
1453 vm_object_dead_sleep(object, "vodead");
1454 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1458 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object"));
1459 vsetflags(vp, VOBJBUF);
1465 * Print out a description of a vnode.
1467 static char *typename[] =
1468 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1471 vprint(char *label, struct vnode *vp)
1476 kprintf("%s: %p: ", label, (void *)vp);
1478 kprintf("%p: ", (void *)vp);
1479 kprintf("type %s, sysrefs %d, writecount %d, holdcnt %d,",
1480 typename[vp->v_type],
1481 vp->v_sysref.refcnt, vp->v_writecount, vp->v_auxrefs);
1483 if (vp->v_flag & VROOT)
1484 strcat(buf, "|VROOT");
1485 if (vp->v_flag & VPFSROOT)
1486 strcat(buf, "|VPFSROOT");
1487 if (vp->v_flag & VTEXT)
1488 strcat(buf, "|VTEXT");
1489 if (vp->v_flag & VSYSTEM)
1490 strcat(buf, "|VSYSTEM");
1491 if (vp->v_flag & VFREE)
1492 strcat(buf, "|VFREE");
1493 if (vp->v_flag & VOBJBUF)
1494 strcat(buf, "|VOBJBUF");
1496 kprintf(" flags (%s)", &buf[1]);
1497 if (vp->v_data == NULL) {
1506 * Do the usual access checking.
1507 * file_mode, uid and gid are from the vnode in question,
1508 * while acc_mode and cred are from the VOP_ACCESS parameter list
1511 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
1512 mode_t acc_mode, struct ucred *cred)
1518 * Super-user always gets read/write access, but execute access depends
1519 * on at least one execute bit being set.
1521 if (priv_check_cred(cred, PRIV_ROOT, 0) == 0) {
1522 if ((acc_mode & VEXEC) && type != VDIR &&
1523 (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)
1530 /* Otherwise, check the owner. */
1531 if (cred->cr_uid == uid) {
1532 if (acc_mode & VEXEC)
1534 if (acc_mode & VREAD)
1536 if (acc_mode & VWRITE)
1538 return ((file_mode & mask) == mask ? 0 : EACCES);
1541 /* Otherwise, check the groups. */
1542 ismember = groupmember(gid, cred);
1543 if (cred->cr_svgid == gid || ismember) {
1544 if (acc_mode & VEXEC)
1546 if (acc_mode & VREAD)
1548 if (acc_mode & VWRITE)
1550 return ((file_mode & mask) == mask ? 0 : EACCES);
1553 /* Otherwise, check everyone else. */
1554 if (acc_mode & VEXEC)
1556 if (acc_mode & VREAD)
1558 if (acc_mode & VWRITE)
1560 return ((file_mode & mask) == mask ? 0 : EACCES);
1564 #include <ddb/ddb.h>
1566 static int db_show_locked_vnodes(struct mount *mp, void *data);
1569 * List all of the locked vnodes in the system.
1570 * Called when debugging the kernel.
1572 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
1574 kprintf("Locked vnodes\n");
1575 mountlist_scan(db_show_locked_vnodes, NULL,
1576 MNTSCAN_FORWARD|MNTSCAN_NOBUSY);
1580 db_show_locked_vnodes(struct mount *mp, void *data __unused)
1584 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
1585 if (vn_islocked(vp))
1593 * Top level filesystem related information gathering.
1595 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
1598 vfs_sysctl(SYSCTL_HANDLER_ARGS)
1600 int *name = (int *)arg1 - 1; /* XXX */
1601 u_int namelen = arg2 + 1; /* XXX */
1602 struct vfsconf *vfsp;
1605 #if 1 || defined(COMPAT_PRELITE2)
1606 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1608 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
1612 /* all sysctl names at this level are at least name and field */
1614 return (ENOTDIR); /* overloaded */
1615 if (name[0] != VFS_GENERIC) {
1616 vfsp = vfsconf_find_by_typenum(name[0]);
1618 return (EOPNOTSUPP);
1619 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1620 oldp, oldlenp, newp, newlen, p));
1624 case VFS_MAXTYPENUM:
1627 maxtypenum = vfsconf_get_maxtypenum();
1628 return (SYSCTL_OUT(req, &maxtypenum, sizeof(maxtypenum)));
1631 return (ENOTDIR); /* overloaded */
1632 vfsp = vfsconf_find_by_typenum(name[2]);
1634 return (EOPNOTSUPP);
1635 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
1637 return (EOPNOTSUPP);
1640 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
1641 "Generic filesystem");
1643 #if 1 || defined(COMPAT_PRELITE2)
1646 sysctl_ovfs_conf_iter(struct vfsconf *vfsp, void *data)
1649 struct ovfsconf ovfs;
1650 struct sysctl_req *req = (struct sysctl_req*) data;
1652 bzero(&ovfs, sizeof(ovfs));
1653 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
1654 strcpy(ovfs.vfc_name, vfsp->vfc_name);
1655 ovfs.vfc_index = vfsp->vfc_typenum;
1656 ovfs.vfc_refcount = vfsp->vfc_refcount;
1657 ovfs.vfc_flags = vfsp->vfc_flags;
1658 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
1660 return error; /* abort iteration with error code */
1662 return 0; /* continue iterating with next element */
1666 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
1668 return vfsconf_each(sysctl_ovfs_conf_iter, (void*)req);
1671 #endif /* 1 || COMPAT_PRELITE2 */
1674 * Check to see if a filesystem is mounted on a block device.
1677 vfs_mountedon(struct vnode *vp)
1681 if ((dev = vp->v_rdev) == NULL) {
1682 /* if (vp->v_type != VBLK)
1683 dev = get_dev(vp->v_uminor, vp->v_umajor); */
1685 if (dev != NULL && dev->si_mountpoint)
1691 * Unmount all filesystems. The list is traversed in reverse order
1692 * of mounting to avoid dependencies.
1695 static int vfs_umountall_callback(struct mount *mp, void *data);
1698 vfs_unmountall(void)
1703 count = mountlist_scan(vfs_umountall_callback,
1704 NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY);
1710 vfs_umountall_callback(struct mount *mp, void *data)
1714 error = dounmount(mp, MNT_FORCE);
1716 mountlist_remove(mp);
1717 kprintf("unmount of filesystem mounted from %s failed (",
1718 mp->mnt_stat.f_mntfromname);
1722 kprintf("%d)\n", error);
1728 * Checks the mount flags for parameter mp and put the names comma-separated
1729 * into a string buffer buf with a size limit specified by len.
1731 * It returns the number of bytes written into buf, and (*errorp) will be
1732 * set to 0, EINVAL (if passed length is 0), or ENOSPC (supplied buffer was
1733 * not large enough). The buffer will be 0-terminated if len was not 0.
1736 vfs_flagstostr(int flags, const struct mountctl_opt *optp,
1737 char *buf, size_t len, int *errorp)
1739 static const struct mountctl_opt optnames[] = {
1740 { MNT_ASYNC, "asynchronous" },
1741 { MNT_EXPORTED, "NFS exported" },
1742 { MNT_LOCAL, "local" },
1743 { MNT_NOATIME, "noatime" },
1744 { MNT_NODEV, "nodev" },
1745 { MNT_NOEXEC, "noexec" },
1746 { MNT_NOSUID, "nosuid" },
1747 { MNT_NOSYMFOLLOW, "nosymfollow" },
1748 { MNT_QUOTA, "with-quotas" },
1749 { MNT_RDONLY, "read-only" },
1750 { MNT_SYNCHRONOUS, "synchronous" },
1751 { MNT_UNION, "union" },
1752 { MNT_NOCLUSTERR, "noclusterr" },
1753 { MNT_NOCLUSTERW, "noclusterw" },
1754 { MNT_SUIDDIR, "suiddir" },
1755 { MNT_SOFTDEP, "soft-updates" },
1756 { MNT_IGNORE, "ignore" },
1766 bleft = len - 1; /* leave room for trailing \0 */
1769 * Checks the size of the string. If it contains
1770 * any data, then we will append the new flags to
1773 actsize = strlen(buf);
1777 /* Default flags if no flags passed */
1781 if (bleft < 0) { /* degenerate case, 0-length buffer */
1786 for (; flags && optp->o_opt; ++optp) {
1787 if ((flags & optp->o_opt) == 0)
1789 optlen = strlen(optp->o_name);
1790 if (bwritten || actsize > 0) {
1795 buf[bwritten++] = ',';
1796 buf[bwritten++] = ' ';
1799 if (bleft < optlen) {
1803 bcopy(optp->o_name, buf + bwritten, optlen);
1806 flags &= ~optp->o_opt;
1810 * Space already reserved for trailing \0
1817 * Build hash lists of net addresses and hang them off the mount point.
1818 * Called by ufs_mount() to set up the lists of export addresses.
1821 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1822 const struct export_args *argp)
1825 struct radix_node_head *rnh;
1827 struct radix_node *rn;
1828 struct sockaddr *saddr, *smask = 0;
1832 if (argp->ex_addrlen == 0) {
1833 if (mp->mnt_flag & MNT_DEFEXPORTED)
1835 np = &nep->ne_defexported;
1836 np->netc_exflags = argp->ex_flags;
1837 np->netc_anon = argp->ex_anon;
1838 np->netc_anon.cr_ref = 1;
1839 mp->mnt_flag |= MNT_DEFEXPORTED;
1843 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
1845 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
1848 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1849 np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK | M_ZERO);
1850 saddr = (struct sockaddr *) (np + 1);
1851 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
1853 if (saddr->sa_len > argp->ex_addrlen)
1854 saddr->sa_len = argp->ex_addrlen;
1855 if (argp->ex_masklen) {
1856 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1857 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
1860 if (smask->sa_len > argp->ex_masklen)
1861 smask->sa_len = argp->ex_masklen;
1863 i = saddr->sa_family;
1864 if ((rnh = nep->ne_rtable[i]) == 0) {
1866 * Seems silly to initialize every AF when most are not used,
1867 * do so on demand here
1869 SLIST_FOREACH(dom, &domains, dom_next)
1870 if (dom->dom_family == i && dom->dom_rtattach) {
1871 dom->dom_rtattach((void **) &nep->ne_rtable[i],
1875 if ((rnh = nep->ne_rtable[i]) == 0) {
1880 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh,
1882 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */
1886 np->netc_exflags = argp->ex_flags;
1887 np->netc_anon = argp->ex_anon;
1888 np->netc_anon.cr_ref = 1;
1891 kfree(np, M_NETADDR);
1897 vfs_free_netcred(struct radix_node *rn, void *w)
1899 struct radix_node_head *rnh = (struct radix_node_head *) w;
1901 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
1902 kfree((caddr_t) rn, M_NETADDR);
1907 * Free the net address hash lists that are hanging off the mount points.
1910 vfs_free_addrlist(struct netexport *nep)
1913 struct radix_node_head *rnh;
1915 for (i = 0; i <= AF_MAX; i++)
1916 if ((rnh = nep->ne_rtable[i])) {
1917 (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
1919 kfree((caddr_t) rnh, M_RTABLE);
1920 nep->ne_rtable[i] = 0;
1925 vfs_export(struct mount *mp, struct netexport *nep,
1926 const struct export_args *argp)
1930 if (argp->ex_flags & MNT_DELEXPORT) {
1931 if (mp->mnt_flag & MNT_EXPUBLIC) {
1932 vfs_setpublicfs(NULL, NULL, NULL);
1933 mp->mnt_flag &= ~MNT_EXPUBLIC;
1935 vfs_free_addrlist(nep);
1936 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
1938 if (argp->ex_flags & MNT_EXPORTED) {
1939 if (argp->ex_flags & MNT_EXPUBLIC) {
1940 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
1942 mp->mnt_flag |= MNT_EXPUBLIC;
1944 if ((error = vfs_hang_addrlist(mp, nep, argp)))
1946 mp->mnt_flag |= MNT_EXPORTED;
1953 * Set the publicly exported filesystem (WebNFS). Currently, only
1954 * one public filesystem is possible in the spec (RFC 2054 and 2055)
1957 vfs_setpublicfs(struct mount *mp, struct netexport *nep,
1958 const struct export_args *argp)
1965 * mp == NULL -> invalidate the current info, the FS is
1966 * no longer exported. May be called from either vfs_export
1967 * or unmount, so check if it hasn't already been done.
1970 if (nfs_pub.np_valid) {
1971 nfs_pub.np_valid = 0;
1972 if (nfs_pub.np_index != NULL) {
1973 FREE(nfs_pub.np_index, M_TEMP);
1974 nfs_pub.np_index = NULL;
1981 * Only one allowed at a time.
1983 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
1987 * Get real filehandle for root of exported FS.
1989 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
1990 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
1992 if ((error = VFS_ROOT(mp, &rvp)))
1995 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2001 * If an indexfile was specified, pull it in.
2003 if (argp->ex_indexfile != NULL) {
2006 error = vn_get_namelen(rvp, &namelen);
2009 MALLOC(nfs_pub.np_index, char *, namelen, M_TEMP,
2011 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2015 * Check for illegal filenames.
2017 for (cp = nfs_pub.np_index; *cp; cp++) {
2025 FREE(nfs_pub.np_index, M_TEMP);
2030 nfs_pub.np_mount = mp;
2031 nfs_pub.np_valid = 1;
2036 vfs_export_lookup(struct mount *mp, struct netexport *nep,
2037 struct sockaddr *nam)
2040 struct radix_node_head *rnh;
2041 struct sockaddr *saddr;
2044 if (mp->mnt_flag & MNT_EXPORTED) {
2046 * Lookup in the export list first.
2050 rnh = nep->ne_rtable[saddr->sa_family];
2052 np = (struct netcred *)
2053 (*rnh->rnh_matchaddr)((char *)saddr,
2055 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2060 * If no address match, use the default if it exists.
2062 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2063 np = &nep->ne_defexported;
2069 * perform msync on all vnodes under a mount point. The mount point must
2070 * be locked. This code is also responsible for lazy-freeing unreferenced
2071 * vnodes whos VM objects no longer contain pages.
2073 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
2075 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked,
2076 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it
2077 * way up in this high level function.
2079 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
2080 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data);
2083 vfs_msync(struct mount *mp, int flags)
2088 * tmpfs sets this flag to prevent msync(), sync, and the
2089 * filesystem periodic syncer from trying to flush VM pages
2090 * to swap. Only pure memory pressure flushes tmpfs VM pages
2093 if (mp->mnt_kern_flag & MNTK_NOMSYNC)
2097 * Ok, scan the vnodes for work.
2099 vmsc_flags = VMSC_GETVP;
2100 if (flags != MNT_WAIT)
2101 vmsc_flags |= VMSC_NOWAIT;
2102 vmntvnodescan(mp, vmsc_flags, vfs_msync_scan1, vfs_msync_scan2,
2103 (void *)(intptr_t)flags);
2107 * scan1 is a fast pre-check. There could be hundreds of thousands of
2108 * vnodes, we cannot afford to do anything heavy weight until we have a
2109 * fairly good indication that there is work to do.
2113 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
2115 int flags = (int)(intptr_t)data;
2117 if ((vp->v_flag & VRECLAIMED) == 0) {
2118 if (vshouldmsync(vp))
2119 return(0); /* call scan2 */
2120 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
2121 (vp->v_flag & VOBJDIRTY) &&
2122 (flags == MNT_WAIT || vn_islocked(vp) == 0)) {
2123 return(0); /* call scan2 */
2128 * do not call scan2, continue the loop
2134 * This callback is handed a locked vnode.
2138 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data)
2141 int flags = (int)(intptr_t)data;
2143 if (vp->v_flag & VRECLAIMED)
2146 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) {
2147 if ((obj = vp->v_object) != NULL) {
2148 vm_object_page_clean(obj, 0, 0,
2149 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2156 * Wake up anyone interested in vp because it is being revoked.
2159 vn_gone(struct vnode *vp)
2161 lwkt_gettoken(&vp->v_token);
2162 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, NOTE_REVOKE);
2163 lwkt_reltoken(&vp->v_token);
2167 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened
2168 * (or v_rdev might be NULL).
2171 vn_todev(struct vnode *vp)
2173 if (vp->v_type != VBLK && vp->v_type != VCHR)
2175 KKASSERT(vp->v_rdev != NULL);
2176 return (vp->v_rdev);
2180 * Check if vnode represents a disk device. The vnode does not need to be
2186 vn_isdisk(struct vnode *vp, int *errp)
2190 if (vp->v_type != VCHR) {
2203 if (dev_is_good(dev) == 0) {
2208 if ((dev_dflags(dev) & D_DISK) == 0) {
2219 vn_get_namelen(struct vnode *vp, int *namelen)
2222 register_t retval[2];
2224 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval);
2227 *namelen = (int)retval[0];
2232 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type,
2233 uint16_t d_namlen, const char *d_name)
2238 len = _DIRENT_RECLEN(d_namlen);
2239 if (len > uio->uio_resid)
2242 dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO);
2245 dp->d_namlen = d_namlen;
2246 dp->d_type = d_type;
2247 bcopy(d_name, dp->d_name, d_namlen);
2249 *error = uiomove((caddr_t)dp, len, uio);
2257 vn_mark_atime(struct vnode *vp, struct thread *td)
2259 struct proc *p = td->td_proc;
2260 struct ucred *cred = p ? p->p_ucred : proc0.p_ucred;
2262 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
2263 VOP_MARKATIME(vp, cred);