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 $
43 * External virtual filesystem routines
47 #include <sys/param.h>
48 #include <sys/systm.h>
51 #include <sys/dirent.h>
52 #include <sys/domain.h>
53 #include <sys/eventhandler.h>
54 #include <sys/fcntl.h>
56 #include <sys/kernel.h>
57 #include <sys/kthread.h>
58 #include <sys/malloc.h>
60 #include <sys/mount.h>
63 #include <sys/reboot.h>
64 #include <sys/socket.h>
66 #include <sys/sysctl.h>
67 #include <sys/syslog.h>
68 #include <sys/unistd.h>
69 #include <sys/vmmeter.h>
70 #include <sys/vnode.h>
72 #include <machine/limits.h>
75 #include <vm/vm_object.h>
76 #include <vm/vm_extern.h>
77 #include <vm/vm_kern.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_page.h>
81 #include <vm/vm_pager.h>
82 #include <vm/vnode_pager.h>
83 #include <vm/vm_zone.h>
86 #include <sys/thread2.h>
87 #include <sys/sysref2.h>
88 #include <sys/mplock2.h>
90 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
93 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
94 "Number of vnodes allocated");
96 SYSCTL_INT(_debug, OID_AUTO, verbose_reclaims, CTLFLAG_RD, &verbose_reclaims, 0,
97 "Output filename of reclaimed vnode(s)");
99 enum vtype iftovt_tab[16] = {
100 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
101 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
103 int vttoif_tab[9] = {
104 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
105 S_IFSOCK, S_IFIFO, S_IFMT,
108 static int reassignbufcalls;
109 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls,
110 0, "Number of times buffers have been reassigned to the proper list");
112 static int check_buf_overlap = 2; /* invasive check */
113 SYSCTL_INT(_vfs, OID_AUTO, check_buf_overlap, CTLFLAG_RW, &check_buf_overlap,
114 0, "Enable overlapping buffer checks");
116 int nfs_mount_type = -1;
117 static struct lwkt_token spechash_token;
118 struct nfs_public nfs_pub; /* publicly exported FS */
121 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
122 &desiredvnodes, 0, "Maximum number of vnodes");
124 static void vfs_free_addrlist (struct netexport *nep);
125 static int vfs_free_netcred (struct radix_node *rn, void *w);
126 static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
127 const struct export_args *argp);
130 * Red black tree functions
132 static int rb_buf_compare(struct buf *b1, struct buf *b2);
133 RB_GENERATE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t, b_loffset);
134 RB_GENERATE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t, b_loffset);
137 rb_buf_compare(struct buf *b1, struct buf *b2)
139 if (b1->b_loffset < b2->b_loffset)
141 if (b1->b_loffset > b2->b_loffset)
147 * Returns non-zero if the vnode is a candidate for lazy msyncing.
149 * NOTE: v_object is not stable (this scan can race), however the
150 * mntvnodescan code holds vmobj_token so any VM object we
151 * do find will remain stable storage.
154 vshouldmsync(struct vnode *vp)
158 if (vp->v_auxrefs != 0 || vp->v_sysref.refcnt > 0)
159 return (0); /* other holders */
160 object = vp->v_object;
162 if (object && (object->ref_count || object->resident_page_count))
168 * Initialize the vnode management data structures.
170 * Called from vfsinit()
179 * Desiredvnodes is kern.maxvnodes. We want to scale it
180 * according to available system memory but we may also have
181 * to limit it based on available KVM, which is capped on 32 bit
184 * WARNING! For machines with 64-256M of ram we have to be sure
185 * that the default limit scales down well due to HAMMER
186 * taking up significantly more memory per-vnode vs UFS.
187 * We want around ~5800 on a 128M machine.
189 factor1 = 20 * (sizeof(struct vm_object) + sizeof(struct vnode));
190 factor2 = 22 * (sizeof(struct vm_object) + sizeof(struct vnode));
192 imin((int64_t)vmstats.v_page_count * PAGE_SIZE / factor1,
194 desiredvnodes = imax(desiredvnodes, maxproc * 8);
196 lwkt_token_init(&spechash_token, "spechash");
200 * Knob to control the precision of file timestamps:
202 * 0 = seconds only; nanoseconds zeroed.
203 * 1 = seconds and nanoseconds, accurate within 1/HZ.
204 * 2 = seconds and nanoseconds, truncated to microseconds.
205 * >=3 = seconds and nanoseconds, maximum precision.
207 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
209 static int timestamp_precision = TSP_SEC;
210 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
211 ×tamp_precision, 0, "Precision of file timestamps");
214 * Get a current timestamp.
219 vfs_timestamp(struct timespec *tsp)
223 switch (timestamp_precision) {
225 tsp->tv_sec = time_second;
233 TIMEVAL_TO_TIMESPEC(&tv, tsp);
243 * Set vnode attributes to VNOVAL
246 vattr_null(struct vattr *vap)
249 vap->va_size = VNOVAL;
250 vap->va_bytes = VNOVAL;
251 vap->va_mode = VNOVAL;
252 vap->va_nlink = VNOVAL;
253 vap->va_uid = VNOVAL;
254 vap->va_gid = VNOVAL;
255 vap->va_fsid = VNOVAL;
256 vap->va_fileid = VNOVAL;
257 vap->va_blocksize = VNOVAL;
258 vap->va_rmajor = VNOVAL;
259 vap->va_rminor = VNOVAL;
260 vap->va_atime.tv_sec = VNOVAL;
261 vap->va_atime.tv_nsec = VNOVAL;
262 vap->va_mtime.tv_sec = VNOVAL;
263 vap->va_mtime.tv_nsec = VNOVAL;
264 vap->va_ctime.tv_sec = VNOVAL;
265 vap->va_ctime.tv_nsec = VNOVAL;
266 vap->va_flags = VNOVAL;
267 vap->va_gen = VNOVAL;
269 /* va_*_uuid fields are only valid if related flags are set */
273 * Flush out and invalidate all buffers associated with a vnode.
277 static int vinvalbuf_bp(struct buf *bp, void *data);
279 struct vinvalbuf_bp_info {
288 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
290 struct vinvalbuf_bp_info info;
294 lwkt_gettoken(&vp->v_token);
297 * If we are being asked to save, call fsync to ensure that the inode
300 if (flags & V_SAVE) {
301 error = bio_track_wait(&vp->v_track_write, slpflag, slptimeo);
304 if (!RB_EMPTY(&vp->v_rbdirty_tree)) {
305 if ((error = VOP_FSYNC(vp, MNT_WAIT, 0)) != 0)
309 * Dirty bufs may be left or generated via races
310 * in circumstances where vinvalbuf() is called on
311 * a vnode not undergoing reclamation. Only
312 * panic if we are trying to reclaim the vnode.
314 if ((vp->v_flag & VRECLAIMED) &&
315 (bio_track_active(&vp->v_track_write) ||
316 !RB_EMPTY(&vp->v_rbdirty_tree))) {
317 panic("vinvalbuf: dirty bufs");
321 info.slptimeo = slptimeo;
322 info.lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
323 if (slpflag & PCATCH)
324 info.lkflags |= LK_PCATCH;
329 * Flush the buffer cache until nothing is left.
331 while (!RB_EMPTY(&vp->v_rbclean_tree) ||
332 !RB_EMPTY(&vp->v_rbdirty_tree)) {
334 error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, NULL,
335 vinvalbuf_bp, &info);
338 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
339 vinvalbuf_bp, &info);
344 * Wait for I/O completion. We may block in the pip code so we have
348 bio_track_wait(&vp->v_track_write, 0, 0);
349 if ((object = vp->v_object) != NULL) {
350 refcount_wait(&object->paging_in_progress, "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 atomic_add_int(&bp->b_refs, 1);
380 error = BUF_TIMELOCK(bp, info->lkflags,
381 "vinvalbuf", info->slptimeo);
382 atomic_subtract_int(&bp->b_refs, 1);
391 KKASSERT(bp->b_vp == info->vp);
394 * Must check clean/dirty status after successfully locking as
397 if ((info->clean && (bp->b_flags & B_DELWRI)) ||
398 (info->clean == 0 && (bp->b_flags & B_DELWRI) == 0)) {
404 * NOTE: NO B_LOCKED CHECK. Also no buf_checkwrite()
405 * check. This code will write out the buffer, period.
408 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
409 (info->flags & V_SAVE)) {
411 } else if (info->flags & V_SAVE) {
413 * Cannot set B_NOCACHE on a clean buffer as this will
414 * destroy the VM backing store which might actually
415 * be dirty (and unsynchronized).
417 bp->b_flags |= (B_INVAL | B_RELBUF);
420 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
427 * Truncate a file's buffer and pages to a specified length. This
428 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
431 * The vnode must be locked.
433 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data);
434 static int vtruncbuf_bp_trunc(struct buf *bp, void *data);
435 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data);
436 static int vtruncbuf_bp_metasync(struct buf *bp, void *data);
438 struct vtruncbuf_info {
445 vtruncbuf(struct vnode *vp, off_t length, int blksize)
447 struct vtruncbuf_info info;
448 const char *filename;
452 * Round up to the *next* block, then destroy the buffers in question.
453 * Since we are only removing some of the buffers we must rely on the
454 * scan count to determine whether a loop is necessary.
456 if ((count = (int)(length % blksize)) != 0)
457 info.truncloffset = length + (blksize - count);
459 info.truncloffset = length;
462 lwkt_gettoken(&vp->v_token);
465 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
466 vtruncbuf_bp_trunc_cmp,
467 vtruncbuf_bp_trunc, &info);
469 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
470 vtruncbuf_bp_trunc_cmp,
471 vtruncbuf_bp_trunc, &info);
475 * For safety, fsync any remaining metadata if the file is not being
476 * truncated to 0. Since the metadata does not represent the entire
477 * dirty list we have to rely on the hit count to ensure that we get
482 count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
483 vtruncbuf_bp_metasync_cmp,
484 vtruncbuf_bp_metasync, &info);
489 * Clean out any left over VM backing store.
491 * It is possible to have in-progress I/O from buffers that were
492 * not part of the truncation. This should not happen if we
493 * are truncating to 0-length.
495 vnode_pager_setsize(vp, length);
496 bio_track_wait(&vp->v_track_write, 0, 0);
501 spin_lock(&vp->v_spin);
502 filename = TAILQ_FIRST(&vp->v_namecache) ?
503 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?";
504 spin_unlock(&vp->v_spin);
507 * Make sure no buffers were instantiated while we were trying
508 * to clean out the remaining VM pages. This could occur due
509 * to busy dirty VM pages being flushed out to disk.
513 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
514 vtruncbuf_bp_trunc_cmp,
515 vtruncbuf_bp_trunc, &info);
517 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
518 vtruncbuf_bp_trunc_cmp,
519 vtruncbuf_bp_trunc, &info);
521 kprintf("Warning: vtruncbuf(): Had to re-clean %d "
522 "left over buffers in %s\n", count, filename);
526 lwkt_reltoken(&vp->v_token);
532 * The callback buffer is beyond the new file EOF and must be destroyed.
533 * Note that the compare function must conform to the RB_SCAN's requirements.
537 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data)
539 struct vtruncbuf_info *info = data;
541 if (bp->b_loffset >= info->truncloffset)
548 vtruncbuf_bp_trunc(struct buf *bp, void *data)
550 struct vtruncbuf_info *info = data;
553 * Do not try to use a buffer we cannot immediately lock, but sleep
554 * anyway to prevent a livelock. The code will loop until all buffers
557 * We must always revalidate the buffer after locking it to deal
560 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
561 atomic_add_int(&bp->b_refs, 1);
562 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
564 atomic_subtract_int(&bp->b_refs, 1);
565 } else if ((info->clean && (bp->b_flags & B_DELWRI)) ||
566 (info->clean == 0 && (bp->b_flags & B_DELWRI) == 0) ||
567 bp->b_vp != info->vp ||
568 vtruncbuf_bp_trunc_cmp(bp, data)) {
572 bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE);
579 * Fsync all meta-data after truncating a file to be non-zero. Only metadata
580 * blocks (with a negative loffset) are scanned.
581 * Note that the compare function must conform to the RB_SCAN's requirements.
584 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data __unused)
586 if (bp->b_loffset < 0)
592 vtruncbuf_bp_metasync(struct buf *bp, void *data)
594 struct vtruncbuf_info *info = data;
596 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
597 atomic_add_int(&bp->b_refs, 1);
598 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
600 atomic_subtract_int(&bp->b_refs, 1);
601 } else if ((bp->b_flags & B_DELWRI) == 0 ||
602 bp->b_vp != info->vp ||
603 vtruncbuf_bp_metasync_cmp(bp, data)) {
607 if (bp->b_vp == info->vp)
616 * vfsync - implements a multipass fsync on a file which understands
617 * dependancies and meta-data. The passed vnode must be locked. The
618 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY.
620 * When fsyncing data asynchronously just do one consolidated pass starting
621 * with the most negative block number. This may not get all the data due
624 * When fsyncing data synchronously do a data pass, then a metadata pass,
625 * then do additional data+metadata passes to try to get all the data out.
627 static int vfsync_wait_output(struct vnode *vp,
628 int (*waitoutput)(struct vnode *, struct thread *));
629 static int vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused);
630 static int vfsync_data_only_cmp(struct buf *bp, void *data);
631 static int vfsync_meta_only_cmp(struct buf *bp, void *data);
632 static int vfsync_lazy_range_cmp(struct buf *bp, void *data);
633 static int vfsync_bp(struct buf *bp, void *data);
642 int (*checkdef)(struct buf *);
643 int (*cmpfunc)(struct buf *, void *);
647 vfsync(struct vnode *vp, int waitfor, int passes,
648 int (*checkdef)(struct buf *),
649 int (*waitoutput)(struct vnode *, struct thread *))
651 struct vfsync_info info;
654 bzero(&info, sizeof(info));
656 if ((info.checkdef = checkdef) == NULL)
659 lwkt_gettoken(&vp->v_token);
662 case MNT_LAZY | MNT_NOWAIT:
665 * Lazy (filesystem syncer typ) Asynchronous plus limit the
666 * number of data (not meta) pages we try to flush to 1MB.
667 * A non-zero return means that lazy limit was reached.
669 info.lazylimit = 1024 * 1024;
671 info.cmpfunc = vfsync_lazy_range_cmp;
672 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
673 vfsync_lazy_range_cmp, vfsync_bp, &info);
674 info.cmpfunc = vfsync_meta_only_cmp;
675 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
676 vfsync_meta_only_cmp, vfsync_bp, &info);
679 else if (!RB_EMPTY(&vp->v_rbdirty_tree))
680 vn_syncer_add(vp, 1);
685 * Asynchronous. Do a data-only pass and a meta-only pass.
688 info.cmpfunc = vfsync_data_only_cmp;
689 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
691 info.cmpfunc = vfsync_meta_only_cmp;
692 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp,
698 * Synchronous. Do a data-only pass, then a meta-data+data
699 * pass, then additional integrated passes to try to get
700 * all the dependancies flushed.
702 info.cmpfunc = vfsync_data_only_cmp;
703 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
705 error = vfsync_wait_output(vp, waitoutput);
707 info.skippedbufs = 0;
708 info.cmpfunc = vfsync_dummy_cmp;
709 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
711 error = vfsync_wait_output(vp, waitoutput);
712 if (info.skippedbufs) {
713 kprintf("Warning: vfsync skipped %d dirty "
714 "bufs in pass2!\n", info.skippedbufs);
717 while (error == 0 && passes > 0 &&
718 !RB_EMPTY(&vp->v_rbdirty_tree)
721 info.synchronous = 1;
724 info.cmpfunc = vfsync_dummy_cmp;
725 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
731 error = vfsync_wait_output(vp, waitoutput);
735 lwkt_reltoken(&vp->v_token);
740 vfsync_wait_output(struct vnode *vp,
741 int (*waitoutput)(struct vnode *, struct thread *))
745 error = bio_track_wait(&vp->v_track_write, 0, 0);
747 error = waitoutput(vp, curthread);
752 vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused)
758 vfsync_data_only_cmp(struct buf *bp, void *data)
760 if (bp->b_loffset < 0)
766 vfsync_meta_only_cmp(struct buf *bp, void *data)
768 if (bp->b_loffset < 0)
774 vfsync_lazy_range_cmp(struct buf *bp, void *data)
776 struct vfsync_info *info = data;
778 if (bp->b_loffset < info->vp->v_lazyw)
784 vfsync_bp(struct buf *bp, void *data)
786 struct vfsync_info *info = data;
787 struct vnode *vp = info->vp;
791 * Ignore buffers that we cannot immediately lock.
793 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
799 * We must revalidate the buffer after locking.
801 if ((bp->b_flags & B_DELWRI) == 0 ||
802 bp->b_vp != info->vp ||
803 info->cmpfunc(bp, data)) {
809 * If syncdeps is not set we do not try to write buffers which have
812 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) {
818 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
819 * has been written but an additional handshake with the device
820 * is required before we can dispose of the buffer. We have no idea
821 * how to do this so we have to skip these buffers.
823 if (bp->b_flags & B_NEEDCOMMIT) {
829 * Ask bioops if it is ok to sync. If not the VFS may have
830 * set B_LOCKED so we have to cycle the buffer.
832 if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) {
838 if (info->synchronous) {
840 * Synchronous flushing. An error may be returned.
846 * Asynchronous flushing. A negative return value simply
847 * stops the scan and is not considered an error. We use
848 * this to support limited MNT_LAZY flushes.
850 vp->v_lazyw = bp->b_loffset;
852 info->lazycount += cluster_awrite(bp);
853 waitrunningbufspace();
855 if (info->lazylimit && info->lazycount >= info->lazylimit)
864 * Associate a buffer with a vnode.
869 bgetvp(struct vnode *vp, struct buf *bp, int testsize)
871 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
872 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0);
875 * Insert onto list for new vnode.
877 lwkt_gettoken(&vp->v_token);
879 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) {
880 lwkt_reltoken(&vp->v_token);
885 * Diagnostics (mainly for HAMMER debugging). Check for
886 * overlapping buffers.
888 if (check_buf_overlap) {
890 bx = buf_rb_hash_RB_PREV(bp);
892 if (bx->b_loffset + bx->b_bufsize > bp->b_loffset) {
893 kprintf("bgetvp: overlapl %016jx/%d %016jx "
895 (intmax_t)bx->b_loffset,
897 (intmax_t)bp->b_loffset,
899 if (check_buf_overlap > 1)
900 panic("bgetvp - overlapping buffer");
903 bx = buf_rb_hash_RB_NEXT(bp);
905 if (bp->b_loffset + testsize > bx->b_loffset) {
906 kprintf("bgetvp: overlapr %016jx/%d %016jx "
908 (intmax_t)bp->b_loffset,
910 (intmax_t)bx->b_loffset,
912 if (check_buf_overlap > 1)
913 panic("bgetvp - overlapping buffer");
918 bp->b_flags |= B_HASHED;
919 bp->b_flags |= B_VNCLEAN;
920 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp))
921 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp);
923 lwkt_reltoken(&vp->v_token);
928 * Disassociate a buffer from a vnode.
933 brelvp(struct buf *bp)
937 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
940 * Delete from old vnode list, if on one.
943 lwkt_gettoken(&vp->v_token);
944 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) {
945 if (bp->b_flags & B_VNDIRTY)
946 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
948 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
949 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN);
951 if (bp->b_flags & B_HASHED) {
952 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp);
953 bp->b_flags &= ~B_HASHED;
955 if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree))
956 vn_syncer_remove(vp);
959 lwkt_reltoken(&vp->v_token);
965 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI.
966 * This routine is called when the state of the B_DELWRI bit is changed.
968 * Must be called with vp->v_token held.
972 reassignbuf(struct buf *bp)
974 struct vnode *vp = bp->b_vp;
977 ASSERT_LWKT_TOKEN_HELD(&vp->v_token);
981 * B_PAGING flagged buffers cannot be reassigned because their vp
982 * is not fully linked in.
984 if (bp->b_flags & B_PAGING)
985 panic("cannot reassign paging buffer");
987 if (bp->b_flags & B_DELWRI) {
989 * Move to the dirty list, add the vnode to the worklist
991 if (bp->b_flags & B_VNCLEAN) {
992 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
993 bp->b_flags &= ~B_VNCLEAN;
995 if ((bp->b_flags & B_VNDIRTY) == 0) {
996 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) {
997 panic("reassignbuf: dup lblk vp %p bp %p",
1000 bp->b_flags |= B_VNDIRTY;
1002 if ((vp->v_flag & VONWORKLST) == 0) {
1003 switch (vp->v_type) {
1010 vp->v_rdev->si_mountpoint != NULL) {
1018 vn_syncer_add(vp, delay);
1022 * Move to the clean list, remove the vnode from the worklist
1023 * if no dirty blocks remain.
1025 if (bp->b_flags & B_VNDIRTY) {
1026 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
1027 bp->b_flags &= ~B_VNDIRTY;
1029 if ((bp->b_flags & B_VNCLEAN) == 0) {
1030 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) {
1031 panic("reassignbuf: dup lblk vp %p bp %p",
1034 bp->b_flags |= B_VNCLEAN;
1036 if ((vp->v_flag & VONWORKLST) &&
1037 RB_EMPTY(&vp->v_rbdirty_tree)) {
1038 vn_syncer_remove(vp);
1044 * Create a vnode for a block device. Used for mounting the root file
1047 * A vref()'d vnode is returned.
1049 extern struct vop_ops *devfs_vnode_dev_vops_p;
1051 bdevvp(cdev_t dev, struct vnode **vpp)
1061 error = getspecialvnode(VT_NON, NULL, &devfs_vnode_dev_vops_p,
1072 v_associate_rdev(vp, dev);
1073 vp->v_umajor = dev->si_umajor;
1074 vp->v_uminor = dev->si_uminor;
1081 v_associate_rdev(struct vnode *vp, cdev_t dev)
1085 if (dev_is_good(dev) == 0)
1087 KKASSERT(vp->v_rdev == NULL);
1088 vp->v_rdev = reference_dev(dev);
1089 lwkt_gettoken(&spechash_token);
1090 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext);
1091 lwkt_reltoken(&spechash_token);
1096 v_release_rdev(struct vnode *vp)
1100 if ((dev = vp->v_rdev) != NULL) {
1101 lwkt_gettoken(&spechash_token);
1102 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext);
1105 lwkt_reltoken(&spechash_token);
1110 * Add a vnode to the alias list hung off the cdev_t. We only associate
1111 * the device number with the vnode. The actual device is not associated
1112 * until the vnode is opened (usually in spec_open()), and will be
1113 * disassociated on last close.
1116 addaliasu(struct vnode *nvp, int x, int y)
1118 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1119 panic("addaliasu on non-special vnode");
1125 * Simple call that a filesystem can make to try to get rid of a
1126 * vnode. It will fail if anyone is referencing the vnode (including
1129 * The filesystem can check whether its in-memory inode structure still
1130 * references the vp on return.
1133 vclean_unlocked(struct vnode *vp)
1136 if (sysref_isactive(&vp->v_sysref) == 0)
1142 * Disassociate a vnode from its underlying filesystem.
1144 * The vnode must be VX locked and referenced. In all normal situations
1145 * there are no active references. If vclean_vxlocked() is called while
1146 * there are active references, the vnode is being ripped out and we have
1147 * to call VOP_CLOSE() as appropriate before we can reclaim it.
1150 vclean_vxlocked(struct vnode *vp, int flags)
1155 struct namecache *ncp;
1158 * If the vnode has already been reclaimed we have nothing to do.
1160 if (vp->v_flag & VRECLAIMED)
1162 vsetflags(vp, VRECLAIMED);
1164 if (verbose_reclaims) {
1165 if ((ncp = TAILQ_FIRST(&vp->v_namecache)) != NULL)
1166 kprintf("Debug: reclaim %p %s\n", vp, ncp->nc_name);
1170 * Scrap the vfs cache
1172 while (cache_inval_vp(vp, 0) != 0) {
1173 kprintf("Warning: vnode %p clean/cache_resolution "
1174 "race detected\n", vp);
1175 tsleep(vp, 0, "vclninv", 2);
1179 * Check to see if the vnode is in use. If so we have to reference it
1180 * before we clean it out so that its count cannot fall to zero and
1181 * generate a race against ourselves to recycle it.
1183 active = sysref_isactive(&vp->v_sysref);
1186 * Clean out any buffers associated with the vnode and destroy its
1187 * object, if it has one.
1189 vinvalbuf(vp, V_SAVE, 0, 0);
1192 * If purging an active vnode (typically during a forced unmount
1193 * or reboot), it must be closed and deactivated before being
1194 * reclaimed. This isn't really all that safe, but what can
1197 * Note that neither of these routines unlocks the vnode.
1199 if (active && (flags & DOCLOSE)) {
1200 while ((n = vp->v_opencount) != 0) {
1201 if (vp->v_writecount)
1202 VOP_CLOSE(vp, FWRITE|FNONBLOCK);
1204 VOP_CLOSE(vp, FNONBLOCK);
1205 if (vp->v_opencount == n) {
1206 kprintf("Warning: unable to force-close"
1214 * If the vnode has not been deactivated, deactivated it. Deactivation
1215 * can create new buffers and VM pages so we have to call vinvalbuf()
1216 * again to make sure they all get flushed.
1218 * This can occur if a file with a link count of 0 needs to be
1221 * If the vnode is already dead don't try to deactivate it.
1223 if ((vp->v_flag & VINACTIVE) == 0) {
1224 vsetflags(vp, VINACTIVE);
1227 vinvalbuf(vp, V_SAVE, 0, 0);
1231 * If the vnode has an object, destroy it.
1233 while ((object = vp->v_object) != NULL) {
1234 vm_object_hold(object);
1235 if (object == vp->v_object)
1237 vm_object_drop(object);
1240 if (object != NULL) {
1241 if (object->ref_count == 0) {
1242 if ((object->flags & OBJ_DEAD) == 0)
1243 vm_object_terminate(object);
1244 vm_object_drop(object);
1245 vclrflags(vp, VOBJBUF);
1247 vm_pager_deallocate(object);
1248 vclrflags(vp, VOBJBUF);
1249 vm_object_drop(object);
1252 KKASSERT((vp->v_flag & VOBJBUF) == 0);
1255 * Reclaim the vnode if not already dead.
1257 if (vp->v_mount && VOP_RECLAIM(vp))
1258 panic("vclean: cannot reclaim");
1261 * Done with purge, notify sleepers of the grim news.
1263 vp->v_ops = &dead_vnode_vops_p;
1268 * If we are destroying an active vnode, reactivate it now that
1269 * we have reassociated it with deadfs. This prevents the system
1270 * from crashing on the vnode due to it being unexpectedly marked
1271 * as inactive or reclaimed.
1273 if (active && (flags & DOCLOSE)) {
1274 vclrflags(vp, VINACTIVE | VRECLAIMED);
1279 * Eliminate all activity associated with the requested vnode
1280 * and with all vnodes aliased to the requested vnode.
1282 * The vnode must be referenced but should not be locked.
1285 vrevoke(struct vnode *vp, struct ucred *cred)
1293 * If the vnode has a device association, scrap all vnodes associated
1294 * with the device. Don't let the device disappear on us while we
1295 * are scrapping the vnodes.
1297 * The passed vp will probably show up in the list, do not VX lock
1300 * Releasing the vnode's rdev here can mess up specfs's call to
1301 * device close, so don't do it. The vnode has been disassociated
1302 * and the device will be closed after the last ref on the related
1303 * fp goes away (if not still open by e.g. the kernel).
1305 if (vp->v_type != VCHR) {
1306 error = fdrevoke(vp, DTYPE_VNODE, cred);
1309 if ((dev = vp->v_rdev) == NULL) {
1313 lwkt_gettoken(&spechash_token);
1316 vqn = SLIST_FIRST(&dev->si_hlist);
1319 while ((vq = vqn) != NULL) {
1320 if (sysref_isactive(&vq->v_sysref)) {
1322 fdrevoke(vq, DTYPE_VNODE, cred);
1323 /*v_release_rdev(vq);*/
1325 if (vq->v_rdev != dev) {
1330 vqn = SLIST_NEXT(vq, v_cdevnext);
1335 lwkt_reltoken(&spechash_token);
1342 * This is called when the object underlying a vnode is being destroyed,
1343 * such as in a remove(). Try to recycle the vnode immediately if the
1344 * only active reference is our reference.
1346 * Directory vnodes in the namecache with children cannot be immediately
1347 * recycled because numerous VOP_N*() ops require them to be stable.
1349 * To avoid recursive recycling from VOP_INACTIVE implemenetations this
1350 * function is a NOP if VRECLAIMED is already set.
1353 vrecycle(struct vnode *vp)
1355 if (vp->v_sysref.refcnt <= 1 && (vp->v_flag & VRECLAIMED) == 0) {
1356 if (cache_inval_vp_nonblock(vp))
1365 * Return the maximum I/O size allowed for strategy calls on VP.
1367 * If vp is VCHR or VBLK we dive the device, otherwise we use
1368 * the vp's mount info.
1370 * The returned value is clamped at MAXPHYS as most callers cannot use
1371 * buffers larger than that size.
1374 vmaxiosize(struct vnode *vp)
1378 if (vp->v_type == VBLK || vp->v_type == VCHR)
1379 maxiosize = vp->v_rdev->si_iosize_max;
1381 maxiosize = vp->v_mount->mnt_iosize_max;
1383 if (maxiosize > MAXPHYS)
1384 maxiosize = MAXPHYS;
1389 * Eliminate all activity associated with a vnode in preparation for reuse.
1391 * The vnode must be VX locked and refd and will remain VX locked and refd
1392 * on return. This routine may be called with the vnode in any state, as
1393 * long as it is VX locked. The vnode will be cleaned out and marked
1394 * VRECLAIMED but will not actually be reused until all existing refs and
1397 * NOTE: This routine may be called on a vnode which has not yet been
1398 * already been deactivated (VOP_INACTIVE), or on a vnode which has
1399 * already been reclaimed.
1401 * This routine is not responsible for placing us back on the freelist.
1402 * Instead, it happens automatically when the caller releases the VX lock
1403 * (assuming there aren't any other references).
1406 vgone_vxlocked(struct vnode *vp)
1409 * assert that the VX lock is held. This is an absolute requirement
1410 * now for vgone_vxlocked() to be called.
1412 KKASSERT(vp->v_lock.lk_exclusivecount == 1);
1415 * Clean out the filesystem specific data and set the VRECLAIMED
1416 * bit. Also deactivate the vnode if necessary.
1418 vclean_vxlocked(vp, DOCLOSE);
1421 * Delete from old mount point vnode list, if on one.
1423 if (vp->v_mount != NULL) {
1424 KKASSERT(vp->v_data == NULL);
1425 insmntque(vp, NULL);
1429 * If special device, remove it from special device alias list
1430 * if it is on one. This should normally only occur if a vnode is
1431 * being revoked as the device should otherwise have been released
1434 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
1445 * Lookup a vnode by device number.
1447 * Returns non-zero and *vpp set to a vref'd vnode on success.
1448 * Returns zero on failure.
1451 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp)
1455 lwkt_gettoken(&spechash_token);
1456 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1457 if (type == vp->v_type) {
1460 lwkt_reltoken(&spechash_token);
1464 lwkt_reltoken(&spechash_token);
1469 * Calculate the total number of references to a special device. This
1470 * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1471 * an overloaded field. Since udev2dev can now return NULL, we have
1472 * to check for a NULL v_rdev.
1475 count_dev(cdev_t dev)
1480 if (SLIST_FIRST(&dev->si_hlist)) {
1481 lwkt_gettoken(&spechash_token);
1482 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1483 count += vp->v_opencount;
1485 lwkt_reltoken(&spechash_token);
1491 vcount(struct vnode *vp)
1493 if (vp->v_rdev == NULL)
1495 return(count_dev(vp->v_rdev));
1499 * Initialize VMIO for a vnode. This routine MUST be called before a
1500 * VFS can issue buffer cache ops on a vnode. It is typically called
1501 * when a vnode is initialized from its inode.
1504 vinitvmio(struct vnode *vp, off_t filesize, int blksize, int boff)
1510 while ((object = vp->v_object) != NULL) {
1511 vm_object_hold(object);
1512 if (object == vp->v_object)
1514 vm_object_drop(object);
1517 if (object == NULL) {
1518 object = vnode_pager_alloc(vp, filesize, 0, 0, blksize, boff);
1521 * Dereference the reference we just created. This assumes
1522 * that the object is associated with the vp.
1524 vm_object_hold(object);
1525 object->ref_count--;
1528 if (object->flags & OBJ_DEAD) {
1530 if (vp->v_object == object)
1531 vm_object_dead_sleep(object, "vodead");
1532 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1533 vm_object_drop(object);
1537 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object"));
1538 vsetflags(vp, VOBJBUF);
1539 vm_object_drop(object);
1546 * Print out a description of a vnode.
1548 static char *typename[] =
1549 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1552 vprint(char *label, struct vnode *vp)
1557 kprintf("%s: %p: ", label, (void *)vp);
1559 kprintf("%p: ", (void *)vp);
1560 kprintf("type %s, sysrefs %d, writecount %d, holdcnt %d,",
1561 typename[vp->v_type],
1562 vp->v_sysref.refcnt, vp->v_writecount, vp->v_auxrefs);
1564 if (vp->v_flag & VROOT)
1565 strcat(buf, "|VROOT");
1566 if (vp->v_flag & VPFSROOT)
1567 strcat(buf, "|VPFSROOT");
1568 if (vp->v_flag & VTEXT)
1569 strcat(buf, "|VTEXT");
1570 if (vp->v_flag & VSYSTEM)
1571 strcat(buf, "|VSYSTEM");
1572 if (vp->v_flag & VFREE)
1573 strcat(buf, "|VFREE");
1574 if (vp->v_flag & VOBJBUF)
1575 strcat(buf, "|VOBJBUF");
1577 kprintf(" flags (%s)", &buf[1]);
1578 if (vp->v_data == NULL) {
1587 * Do the usual access checking.
1588 * file_mode, uid and gid are from the vnode in question,
1589 * while acc_mode and cred are from the VOP_ACCESS parameter list
1592 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
1593 mode_t acc_mode, struct ucred *cred)
1599 * Super-user always gets read/write access, but execute access depends
1600 * on at least one execute bit being set.
1602 if (priv_check_cred(cred, PRIV_ROOT, 0) == 0) {
1603 if ((acc_mode & VEXEC) && type != VDIR &&
1604 (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)
1611 /* Otherwise, check the owner. */
1612 if (cred->cr_uid == uid) {
1613 if (acc_mode & VEXEC)
1615 if (acc_mode & VREAD)
1617 if (acc_mode & VWRITE)
1619 return ((file_mode & mask) == mask ? 0 : EACCES);
1622 /* Otherwise, check the groups. */
1623 ismember = groupmember(gid, cred);
1624 if (cred->cr_svgid == gid || ismember) {
1625 if (acc_mode & VEXEC)
1627 if (acc_mode & VREAD)
1629 if (acc_mode & VWRITE)
1631 return ((file_mode & mask) == mask ? 0 : EACCES);
1634 /* Otherwise, check everyone else. */
1635 if (acc_mode & VEXEC)
1637 if (acc_mode & VREAD)
1639 if (acc_mode & VWRITE)
1641 return ((file_mode & mask) == mask ? 0 : EACCES);
1645 #include <ddb/ddb.h>
1647 static int db_show_locked_vnodes(struct mount *mp, void *data);
1650 * List all of the locked vnodes in the system.
1651 * Called when debugging the kernel.
1653 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
1655 kprintf("Locked vnodes\n");
1656 mountlist_scan(db_show_locked_vnodes, NULL,
1657 MNTSCAN_FORWARD|MNTSCAN_NOBUSY);
1661 db_show_locked_vnodes(struct mount *mp, void *data __unused)
1665 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
1666 if (vn_islocked(vp))
1674 * Top level filesystem related information gathering.
1676 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
1679 vfs_sysctl(SYSCTL_HANDLER_ARGS)
1681 int *name = (int *)arg1 - 1; /* XXX */
1682 u_int namelen = arg2 + 1; /* XXX */
1683 struct vfsconf *vfsp;
1686 #if 1 || defined(COMPAT_PRELITE2)
1687 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1689 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
1693 /* all sysctl names at this level are at least name and field */
1695 return (ENOTDIR); /* overloaded */
1696 if (name[0] != VFS_GENERIC) {
1697 vfsp = vfsconf_find_by_typenum(name[0]);
1699 return (EOPNOTSUPP);
1700 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1701 oldp, oldlenp, newp, newlen, p));
1705 case VFS_MAXTYPENUM:
1708 maxtypenum = vfsconf_get_maxtypenum();
1709 return (SYSCTL_OUT(req, &maxtypenum, sizeof(maxtypenum)));
1712 return (ENOTDIR); /* overloaded */
1713 vfsp = vfsconf_find_by_typenum(name[2]);
1715 return (EOPNOTSUPP);
1716 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
1718 return (EOPNOTSUPP);
1721 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
1722 "Generic filesystem");
1724 #if 1 || defined(COMPAT_PRELITE2)
1727 sysctl_ovfs_conf_iter(struct vfsconf *vfsp, void *data)
1730 struct ovfsconf ovfs;
1731 struct sysctl_req *req = (struct sysctl_req*) data;
1733 bzero(&ovfs, sizeof(ovfs));
1734 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
1735 strcpy(ovfs.vfc_name, vfsp->vfc_name);
1736 ovfs.vfc_index = vfsp->vfc_typenum;
1737 ovfs.vfc_refcount = vfsp->vfc_refcount;
1738 ovfs.vfc_flags = vfsp->vfc_flags;
1739 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
1741 return error; /* abort iteration with error code */
1743 return 0; /* continue iterating with next element */
1747 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
1749 return vfsconf_each(sysctl_ovfs_conf_iter, (void*)req);
1752 #endif /* 1 || COMPAT_PRELITE2 */
1755 * Check to see if a filesystem is mounted on a block device.
1758 vfs_mountedon(struct vnode *vp)
1762 if ((dev = vp->v_rdev) == NULL) {
1763 /* if (vp->v_type != VBLK)
1764 dev = get_dev(vp->v_uminor, vp->v_umajor); */
1766 if (dev != NULL && dev->si_mountpoint)
1772 * Unmount all filesystems. The list is traversed in reverse order
1773 * of mounting to avoid dependencies.
1776 static int vfs_umountall_callback(struct mount *mp, void *data);
1779 vfs_unmountall(void)
1784 count = mountlist_scan(vfs_umountall_callback,
1785 NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY);
1791 vfs_umountall_callback(struct mount *mp, void *data)
1795 error = dounmount(mp, MNT_FORCE);
1797 mountlist_remove(mp);
1798 kprintf("unmount of filesystem mounted from %s failed (",
1799 mp->mnt_stat.f_mntfromname);
1803 kprintf("%d)\n", error);
1809 * Checks the mount flags for parameter mp and put the names comma-separated
1810 * into a string buffer buf with a size limit specified by len.
1812 * It returns the number of bytes written into buf, and (*errorp) will be
1813 * set to 0, EINVAL (if passed length is 0), or ENOSPC (supplied buffer was
1814 * not large enough). The buffer will be 0-terminated if len was not 0.
1817 vfs_flagstostr(int flags, const struct mountctl_opt *optp,
1818 char *buf, size_t len, int *errorp)
1820 static const struct mountctl_opt optnames[] = {
1821 { MNT_ASYNC, "asynchronous" },
1822 { MNT_EXPORTED, "NFS exported" },
1823 { MNT_LOCAL, "local" },
1824 { MNT_NOATIME, "noatime" },
1825 { MNT_NODEV, "nodev" },
1826 { MNT_NOEXEC, "noexec" },
1827 { MNT_NOSUID, "nosuid" },
1828 { MNT_NOSYMFOLLOW, "nosymfollow" },
1829 { MNT_QUOTA, "with-quotas" },
1830 { MNT_RDONLY, "read-only" },
1831 { MNT_SYNCHRONOUS, "synchronous" },
1832 { MNT_UNION, "union" },
1833 { MNT_NOCLUSTERR, "noclusterr" },
1834 { MNT_NOCLUSTERW, "noclusterw" },
1835 { MNT_SUIDDIR, "suiddir" },
1836 { MNT_SOFTDEP, "soft-updates" },
1837 { MNT_IGNORE, "ignore" },
1847 bleft = len - 1; /* leave room for trailing \0 */
1850 * Checks the size of the string. If it contains
1851 * any data, then we will append the new flags to
1854 actsize = strlen(buf);
1858 /* Default flags if no flags passed */
1862 if (bleft < 0) { /* degenerate case, 0-length buffer */
1867 for (; flags && optp->o_opt; ++optp) {
1868 if ((flags & optp->o_opt) == 0)
1870 optlen = strlen(optp->o_name);
1871 if (bwritten || actsize > 0) {
1876 buf[bwritten++] = ',';
1877 buf[bwritten++] = ' ';
1880 if (bleft < optlen) {
1884 bcopy(optp->o_name, buf + bwritten, optlen);
1887 flags &= ~optp->o_opt;
1891 * Space already reserved for trailing \0
1898 * Build hash lists of net addresses and hang them off the mount point.
1899 * Called by ufs_mount() to set up the lists of export addresses.
1902 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1903 const struct export_args *argp)
1906 struct radix_node_head *rnh;
1908 struct radix_node *rn;
1909 struct sockaddr *saddr, *smask = NULL;
1913 if (argp->ex_addrlen == 0) {
1914 if (mp->mnt_flag & MNT_DEFEXPORTED)
1916 np = &nep->ne_defexported;
1917 np->netc_exflags = argp->ex_flags;
1918 np->netc_anon = argp->ex_anon;
1919 np->netc_anon.cr_ref = 1;
1920 mp->mnt_flag |= MNT_DEFEXPORTED;
1924 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
1926 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
1929 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1930 np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK | M_ZERO);
1931 saddr = (struct sockaddr *) (np + 1);
1932 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
1934 if (saddr->sa_len > argp->ex_addrlen)
1935 saddr->sa_len = argp->ex_addrlen;
1936 if (argp->ex_masklen) {
1937 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1938 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
1941 if (smask->sa_len > argp->ex_masklen)
1942 smask->sa_len = argp->ex_masklen;
1944 i = saddr->sa_family;
1945 if ((rnh = nep->ne_rtable[i]) == NULL) {
1947 * Seems silly to initialize every AF when most are not used,
1948 * do so on demand here
1950 SLIST_FOREACH(dom, &domains, dom_next)
1951 if (dom->dom_family == i && dom->dom_rtattach) {
1952 dom->dom_rtattach((void **) &nep->ne_rtable[i],
1956 if ((rnh = nep->ne_rtable[i]) == NULL) {
1961 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh,
1963 if (rn == NULL || np != (struct netcred *) rn) { /* already exists */
1967 np->netc_exflags = argp->ex_flags;
1968 np->netc_anon = argp->ex_anon;
1969 np->netc_anon.cr_ref = 1;
1972 kfree(np, M_NETADDR);
1978 vfs_free_netcred(struct radix_node *rn, void *w)
1980 struct radix_node_head *rnh = (struct radix_node_head *) w;
1982 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
1983 kfree((caddr_t) rn, M_NETADDR);
1988 * Free the net address hash lists that are hanging off the mount points.
1991 vfs_free_addrlist(struct netexport *nep)
1994 struct radix_node_head *rnh;
1996 for (i = 0; i <= AF_MAX; i++)
1997 if ((rnh = nep->ne_rtable[i])) {
1998 (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2000 kfree((caddr_t) rnh, M_RTABLE);
2001 nep->ne_rtable[i] = 0;
2006 vfs_export(struct mount *mp, struct netexport *nep,
2007 const struct export_args *argp)
2011 if (argp->ex_flags & MNT_DELEXPORT) {
2012 if (mp->mnt_flag & MNT_EXPUBLIC) {
2013 vfs_setpublicfs(NULL, NULL, NULL);
2014 mp->mnt_flag &= ~MNT_EXPUBLIC;
2016 vfs_free_addrlist(nep);
2017 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2019 if (argp->ex_flags & MNT_EXPORTED) {
2020 if (argp->ex_flags & MNT_EXPUBLIC) {
2021 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2023 mp->mnt_flag |= MNT_EXPUBLIC;
2025 if ((error = vfs_hang_addrlist(mp, nep, argp)))
2027 mp->mnt_flag |= MNT_EXPORTED;
2034 * Set the publicly exported filesystem (WebNFS). Currently, only
2035 * one public filesystem is possible in the spec (RFC 2054 and 2055)
2038 vfs_setpublicfs(struct mount *mp, struct netexport *nep,
2039 const struct export_args *argp)
2046 * mp == NULL -> invalidate the current info, the FS is
2047 * no longer exported. May be called from either vfs_export
2048 * or unmount, so check if it hasn't already been done.
2051 if (nfs_pub.np_valid) {
2052 nfs_pub.np_valid = 0;
2053 if (nfs_pub.np_index != NULL) {
2054 kfree(nfs_pub.np_index, M_TEMP);
2055 nfs_pub.np_index = NULL;
2062 * Only one allowed at a time.
2064 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2068 * Get real filehandle for root of exported FS.
2070 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2071 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2073 if ((error = VFS_ROOT(mp, &rvp)))
2076 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2082 * If an indexfile was specified, pull it in.
2084 if (argp->ex_indexfile != NULL) {
2087 error = vn_get_namelen(rvp, &namelen);
2090 nfs_pub.np_index = kmalloc(namelen, M_TEMP, M_WAITOK);
2091 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2095 * Check for illegal filenames.
2097 for (cp = nfs_pub.np_index; *cp; cp++) {
2105 kfree(nfs_pub.np_index, M_TEMP);
2110 nfs_pub.np_mount = mp;
2111 nfs_pub.np_valid = 1;
2116 vfs_export_lookup(struct mount *mp, struct netexport *nep,
2117 struct sockaddr *nam)
2120 struct radix_node_head *rnh;
2121 struct sockaddr *saddr;
2124 if (mp->mnt_flag & MNT_EXPORTED) {
2126 * Lookup in the export list first.
2130 rnh = nep->ne_rtable[saddr->sa_family];
2132 np = (struct netcred *)
2133 (*rnh->rnh_matchaddr)((char *)saddr,
2135 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2140 * If no address match, use the default if it exists.
2142 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2143 np = &nep->ne_defexported;
2149 * perform msync on all vnodes under a mount point. The mount point must
2150 * be locked. This code is also responsible for lazy-freeing unreferenced
2151 * vnodes whos VM objects no longer contain pages.
2153 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
2155 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked,
2156 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it
2157 * way up in this high level function.
2159 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
2160 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data);
2163 vfs_msync(struct mount *mp, int flags)
2168 * tmpfs sets this flag to prevent msync(), sync, and the
2169 * filesystem periodic syncer from trying to flush VM pages
2170 * to swap. Only pure memory pressure flushes tmpfs VM pages
2173 if (mp->mnt_kern_flag & MNTK_NOMSYNC)
2177 * Ok, scan the vnodes for work.
2179 vmsc_flags = VMSC_GETVP;
2180 if (flags != MNT_WAIT)
2181 vmsc_flags |= VMSC_NOWAIT;
2182 vmntvnodescan(mp, vmsc_flags,
2183 vfs_msync_scan1, vfs_msync_scan2,
2184 (void *)(intptr_t)flags);
2188 * scan1 is a fast pre-check. There could be hundreds of thousands of
2189 * vnodes, we cannot afford to do anything heavy weight until we have a
2190 * fairly good indication that there is work to do.
2194 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
2196 int flags = (int)(intptr_t)data;
2198 if ((vp->v_flag & VRECLAIMED) == 0) {
2199 if (vshouldmsync(vp))
2200 return(0); /* call scan2 */
2201 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
2202 (vp->v_flag & VOBJDIRTY) &&
2203 (flags == MNT_WAIT || vn_islocked(vp) == 0)) {
2204 return(0); /* call scan2 */
2209 * do not call scan2, continue the loop
2215 * This callback is handed a locked vnode.
2219 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data)
2222 int flags = (int)(intptr_t)data;
2224 if (vp->v_flag & VRECLAIMED)
2227 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) {
2228 if ((obj = vp->v_object) != NULL) {
2229 vm_object_page_clean(obj, 0, 0,
2230 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2237 * Wake up anyone interested in vp because it is being revoked.
2240 vn_gone(struct vnode *vp)
2242 lwkt_gettoken(&vp->v_token);
2243 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, NOTE_REVOKE);
2244 lwkt_reltoken(&vp->v_token);
2248 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened
2249 * (or v_rdev might be NULL).
2252 vn_todev(struct vnode *vp)
2254 if (vp->v_type != VBLK && vp->v_type != VCHR)
2256 KKASSERT(vp->v_rdev != NULL);
2257 return (vp->v_rdev);
2261 * Check if vnode represents a disk device. The vnode does not need to be
2267 vn_isdisk(struct vnode *vp, int *errp)
2271 if (vp->v_type != VCHR) {
2284 if (dev_is_good(dev) == 0) {
2289 if ((dev_dflags(dev) & D_DISK) == 0) {
2300 vn_get_namelen(struct vnode *vp, int *namelen)
2303 register_t retval[2];
2305 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval);
2308 *namelen = (int)retval[0];
2313 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type,
2314 uint16_t d_namlen, const char *d_name)
2319 len = _DIRENT_RECLEN(d_namlen);
2320 if (len > uio->uio_resid)
2323 dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO);
2326 dp->d_namlen = d_namlen;
2327 dp->d_type = d_type;
2328 bcopy(d_name, dp->d_name, d_namlen);
2330 *error = uiomove((caddr_t)dp, len, uio);
2338 vn_mark_atime(struct vnode *vp, struct thread *td)
2340 struct proc *p = td->td_proc;
2341 struct ucred *cred = p ? p->p_ucred : proc0.p_ucred;
2343 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
2344 VOP_MARKATIME(vp, cred);