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. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95
35 * $FreeBSD: src/sys/kern/vfs_subr.c,v 1.249.2.30 2003/04/04 20:35:57 tegge Exp $
39 * External virtual filesystem routines
43 #include "opt_inet6.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
49 #include <sys/dirent.h>
50 #include <sys/eventhandler.h>
51 #include <sys/fcntl.h>
53 #include <sys/kernel.h>
54 #include <sys/kthread.h>
55 #include <sys/malloc.h>
57 #include <sys/mount.h>
60 #include <sys/reboot.h>
61 #include <sys/socket.h>
63 #include <sys/sysctl.h>
64 #include <sys/syslog.h>
65 #include <sys/unistd.h>
66 #include <sys/vmmeter.h>
67 #include <sys/vnode.h>
69 #include <machine/limits.h>
72 #include <vm/vm_object.h>
73 #include <vm/vm_extern.h>
74 #include <vm/vm_kern.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_page.h>
78 #include <vm/vm_pager.h>
79 #include <vm/vnode_pager.h>
80 #include <vm/vm_zone.h>
83 #include <sys/thread2.h>
84 #include <sys/sysref2.h>
85 #include <sys/mplock2.h>
86 #include <vm/vm_page2.h>
88 #include <netinet/in.h>
90 static MALLOC_DEFINE(M_NETCRED, "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 &maxvnodes, 0, "Maximum number of vnodes");
124 static struct radix_node_head *vfs_create_addrlist_af(int af,
125 struct netexport *nep);
126 static void vfs_free_addrlist (struct netexport *nep);
127 static int vfs_free_netcred (struct radix_node *rn, void *w);
128 static void vfs_free_addrlist_af (struct radix_node_head **prnh);
129 static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
130 const struct export_args *argp);
132 int prtactive = 0; /* 1 => print out reclaim of active vnodes */
135 * Red black tree functions
137 static int rb_buf_compare(struct buf *b1, struct buf *b2);
138 RB_GENERATE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t, b_loffset);
139 RB_GENERATE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t, b_loffset);
142 rb_buf_compare(struct buf *b1, struct buf *b2)
144 if (b1->b_loffset < b2->b_loffset)
146 if (b1->b_loffset > b2->b_loffset)
152 * Initialize the vnode management data structures.
154 * Called from vfsinit()
163 * Desiredvnodes is kern.maxvnodes. We want to scale it
164 * according to available system memory but we may also have
165 * to limit it based on available KVM, which is capped on 32 bit
166 * systems, to ~80K vnodes or so.
168 * WARNING! For machines with 64-256M of ram we have to be sure
169 * that the default limit scales down well due to HAMMER
170 * taking up significantly more memory per-vnode vs UFS.
171 * We want around ~5800 on a 128M machine.
173 factor1 = 25 * (sizeof(struct vm_object) + sizeof(struct vnode));
174 factor2 = 30 * (sizeof(struct vm_object) + sizeof(struct vnode));
175 maxvnodes = imin((int64_t)vmstats.v_page_count * PAGE_SIZE / factor1,
177 maxvnodes = imax(maxvnodes, maxproc * 8);
179 lwkt_token_init(&spechash_token, "spechash");
183 * Knob to control the precision of file timestamps:
185 * 0 = seconds only; nanoseconds zeroed.
186 * 1 = seconds and nanoseconds, accurate within 1/HZ.
187 * 2 = seconds and nanoseconds, truncated to microseconds.
188 * >=3 = seconds and nanoseconds, maximum precision.
190 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
192 static int timestamp_precision = TSP_SEC;
193 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
194 ×tamp_precision, 0, "Precision of file timestamps");
197 * Get a current timestamp.
202 vfs_timestamp(struct timespec *tsp)
206 switch (timestamp_precision) {
208 tsp->tv_sec = time_second;
216 TIMEVAL_TO_TIMESPEC(&tv, tsp);
226 * Set vnode attributes to VNOVAL
229 vattr_null(struct vattr *vap)
232 vap->va_size = VNOVAL;
233 vap->va_bytes = VNOVAL;
234 vap->va_mode = VNOVAL;
235 vap->va_nlink = VNOVAL;
236 vap->va_uid = VNOVAL;
237 vap->va_gid = VNOVAL;
238 vap->va_fsid = VNOVAL;
239 vap->va_fileid = VNOVAL;
240 vap->va_blocksize = VNOVAL;
241 vap->va_rmajor = VNOVAL;
242 vap->va_rminor = VNOVAL;
243 vap->va_atime.tv_sec = VNOVAL;
244 vap->va_atime.tv_nsec = VNOVAL;
245 vap->va_mtime.tv_sec = VNOVAL;
246 vap->va_mtime.tv_nsec = VNOVAL;
247 vap->va_ctime.tv_sec = VNOVAL;
248 vap->va_ctime.tv_nsec = VNOVAL;
249 vap->va_flags = VNOVAL;
250 vap->va_gen = VNOVAL;
252 /* va_*_uuid fields are only valid if related flags are set */
256 * Flush out and invalidate all buffers associated with a vnode.
260 static int vinvalbuf_bp(struct buf *bp, void *data);
262 struct vinvalbuf_bp_info {
271 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
273 struct vinvalbuf_bp_info info;
277 lwkt_gettoken(&vp->v_token);
280 * If we are being asked to save, call fsync to ensure that the inode
283 if (flags & V_SAVE) {
284 error = bio_track_wait(&vp->v_track_write, slpflag, slptimeo);
287 if (!RB_EMPTY(&vp->v_rbdirty_tree)) {
288 if ((error = VOP_FSYNC(vp, MNT_WAIT, 0)) != 0)
292 * Dirty bufs may be left or generated via races
293 * in circumstances where vinvalbuf() is called on
294 * a vnode not undergoing reclamation. Only
295 * panic if we are trying to reclaim the vnode.
297 if ((vp->v_flag & VRECLAIMED) &&
298 (bio_track_active(&vp->v_track_write) ||
299 !RB_EMPTY(&vp->v_rbdirty_tree))) {
300 panic("vinvalbuf: dirty bufs");
305 info.slptimeo = slptimeo;
306 info.lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
307 if (slpflag & PCATCH)
308 info.lkflags |= LK_PCATCH;
313 * Flush the buffer cache until nothing is left, wait for all I/O
314 * to complete. At least one pass is required. We might block
315 * in the pip code so we have to re-check. Order is important.
321 if (!RB_EMPTY(&vp->v_rbclean_tree)) {
323 error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
324 NULL, vinvalbuf_bp, &info);
326 if (!RB_EMPTY(&vp->v_rbdirty_tree)) {
328 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
329 NULL, vinvalbuf_bp, &info);
333 * Wait for I/O completion.
335 bio_track_wait(&vp->v_track_write, 0, 0);
336 if ((object = vp->v_object) != NULL)
337 refcount_wait(&object->paging_in_progress, "vnvlbx");
338 } while (bio_track_active(&vp->v_track_write) ||
339 !RB_EMPTY(&vp->v_rbclean_tree) ||
340 !RB_EMPTY(&vp->v_rbdirty_tree));
343 * Destroy the copy in the VM cache, too.
345 if ((object = vp->v_object) != NULL) {
346 vm_object_page_remove(object, 0, 0,
347 (flags & V_SAVE) ? TRUE : FALSE);
350 if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree))
351 panic("vinvalbuf: flush failed");
352 if (!RB_EMPTY(&vp->v_rbhash_tree))
353 panic("vinvalbuf: flush failed, buffers still present");
356 lwkt_reltoken(&vp->v_token);
361 vinvalbuf_bp(struct buf *bp, void *data)
363 struct vinvalbuf_bp_info *info = data;
366 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
367 atomic_add_int(&bp->b_refs, 1);
368 error = BUF_TIMELOCK(bp, info->lkflags,
369 "vinvalbuf", info->slptimeo);
370 atomic_subtract_int(&bp->b_refs, 1);
379 KKASSERT(bp->b_vp == info->vp);
382 * Must check clean/dirty status after successfully locking as
385 if ((info->clean && (bp->b_flags & B_DELWRI)) ||
386 (info->clean == 0 && (bp->b_flags & B_DELWRI) == 0)) {
392 * NOTE: NO B_LOCKED CHECK. Also no buf_checkwrite()
393 * check. This code will write out the buffer, period.
396 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
397 (info->flags & V_SAVE)) {
399 } else if (info->flags & V_SAVE) {
401 * Cannot set B_NOCACHE on a clean buffer as this will
402 * destroy the VM backing store which might actually
403 * be dirty (and unsynchronized).
405 bp->b_flags |= (B_INVAL | B_RELBUF);
408 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
415 * Truncate a file's buffer and pages to a specified length. This
416 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
419 * The vnode must be locked.
421 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data);
422 static int vtruncbuf_bp_trunc(struct buf *bp, void *data);
423 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data);
424 static int vtruncbuf_bp_metasync(struct buf *bp, void *data);
426 struct vtruncbuf_info {
433 vtruncbuf(struct vnode *vp, off_t length, int blksize)
435 struct vtruncbuf_info info;
436 const char *filename;
440 * Round up to the *next* block, then destroy the buffers in question.
441 * Since we are only removing some of the buffers we must rely on the
442 * scan count to determine whether a loop is necessary.
444 if ((count = (int)(length % blksize)) != 0)
445 info.truncloffset = length + (blksize - count);
447 info.truncloffset = length;
450 lwkt_gettoken(&vp->v_token);
453 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
454 vtruncbuf_bp_trunc_cmp,
455 vtruncbuf_bp_trunc, &info);
457 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
458 vtruncbuf_bp_trunc_cmp,
459 vtruncbuf_bp_trunc, &info);
463 * For safety, fsync any remaining metadata if the file is not being
464 * truncated to 0. Since the metadata does not represent the entire
465 * dirty list we have to rely on the hit count to ensure that we get
470 count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
471 vtruncbuf_bp_metasync_cmp,
472 vtruncbuf_bp_metasync, &info);
477 * Clean out any left over VM backing store.
479 * It is possible to have in-progress I/O from buffers that were
480 * not part of the truncation. This should not happen if we
481 * are truncating to 0-length.
483 vnode_pager_setsize(vp, length);
484 bio_track_wait(&vp->v_track_write, 0, 0);
489 spin_lock(&vp->v_spin);
490 filename = TAILQ_FIRST(&vp->v_namecache) ?
491 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?";
492 spin_unlock(&vp->v_spin);
495 * Make sure no buffers were instantiated while we were trying
496 * to clean out the remaining VM pages. This could occur due
497 * to busy dirty VM pages being flushed out to disk.
501 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
502 vtruncbuf_bp_trunc_cmp,
503 vtruncbuf_bp_trunc, &info);
505 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
506 vtruncbuf_bp_trunc_cmp,
507 vtruncbuf_bp_trunc, &info);
509 kprintf("Warning: vtruncbuf(): Had to re-clean %d "
510 "left over buffers in %s\n", count, filename);
514 lwkt_reltoken(&vp->v_token);
520 * The callback buffer is beyond the new file EOF and must be destroyed.
521 * Note that the compare function must conform to the RB_SCAN's requirements.
525 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data)
527 struct vtruncbuf_info *info = data;
529 if (bp->b_loffset >= info->truncloffset)
536 vtruncbuf_bp_trunc(struct buf *bp, void *data)
538 struct vtruncbuf_info *info = data;
541 * Do not try to use a buffer we cannot immediately lock, but sleep
542 * anyway to prevent a livelock. The code will loop until all buffers
545 * We must always revalidate the buffer after locking it to deal
548 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
549 atomic_add_int(&bp->b_refs, 1);
550 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
552 atomic_subtract_int(&bp->b_refs, 1);
553 } else if ((info->clean && (bp->b_flags & B_DELWRI)) ||
554 (info->clean == 0 && (bp->b_flags & B_DELWRI) == 0) ||
555 bp->b_vp != info->vp ||
556 vtruncbuf_bp_trunc_cmp(bp, data)) {
560 bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE);
567 * Fsync all meta-data after truncating a file to be non-zero. Only metadata
568 * blocks (with a negative loffset) are scanned.
569 * Note that the compare function must conform to the RB_SCAN's requirements.
572 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data __unused)
574 if (bp->b_loffset < 0)
580 vtruncbuf_bp_metasync(struct buf *bp, void *data)
582 struct vtruncbuf_info *info = data;
584 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
585 atomic_add_int(&bp->b_refs, 1);
586 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
588 atomic_subtract_int(&bp->b_refs, 1);
589 } else if ((bp->b_flags & B_DELWRI) == 0 ||
590 bp->b_vp != info->vp ||
591 vtruncbuf_bp_metasync_cmp(bp, data)) {
595 if (bp->b_vp == info->vp)
604 * vfsync - implements a multipass fsync on a file which understands
605 * dependancies and meta-data. The passed vnode must be locked. The
606 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY.
608 * When fsyncing data asynchronously just do one consolidated pass starting
609 * with the most negative block number. This may not get all the data due
612 * When fsyncing data synchronously do a data pass, then a metadata pass,
613 * then do additional data+metadata passes to try to get all the data out.
615 * Caller must ref the vnode but does not have to lock it.
617 static int vfsync_wait_output(struct vnode *vp,
618 int (*waitoutput)(struct vnode *, struct thread *));
619 static int vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused);
620 static int vfsync_data_only_cmp(struct buf *bp, void *data);
621 static int vfsync_meta_only_cmp(struct buf *bp, void *data);
622 static int vfsync_lazy_range_cmp(struct buf *bp, void *data);
623 static int vfsync_bp(struct buf *bp, void *data);
633 int (*checkdef)(struct buf *);
634 int (*cmpfunc)(struct buf *, void *);
638 vfsync(struct vnode *vp, int waitfor, int passes,
639 int (*checkdef)(struct buf *),
640 int (*waitoutput)(struct vnode *, struct thread *))
642 struct vfsync_info info;
645 bzero(&info, sizeof(info));
647 if ((info.checkdef = checkdef) == NULL)
650 lwkt_gettoken(&vp->v_token);
653 case MNT_LAZY | MNT_NOWAIT:
656 * Lazy (filesystem syncer typ) Asynchronous plus limit the
657 * number of data (not meta) pages we try to flush to 1MB.
658 * A non-zero return means that lazy limit was reached.
660 info.lazylimit = 1024 * 1024;
662 info.cmpfunc = vfsync_lazy_range_cmp;
663 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
664 vfsync_lazy_range_cmp, vfsync_bp, &info);
665 info.cmpfunc = vfsync_meta_only_cmp;
666 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
667 vfsync_meta_only_cmp, vfsync_bp, &info);
670 else if (!RB_EMPTY(&vp->v_rbdirty_tree))
671 vn_syncer_add(vp, 1);
676 * Asynchronous. Do a data-only pass and a meta-only pass.
679 info.cmpfunc = vfsync_data_only_cmp;
680 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
682 info.cmpfunc = vfsync_meta_only_cmp;
683 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp,
689 * Synchronous. Do a data-only pass, then a meta-data+data
690 * pass, then additional integrated passes to try to get
691 * all the dependancies flushed.
693 info.cmpfunc = vfsync_data_only_cmp;
695 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
698 error = vfsync_wait_output(vp, waitoutput);
700 info.skippedbufs = 0;
701 info.cmpfunc = vfsync_dummy_cmp;
702 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
704 error = vfsync_wait_output(vp, waitoutput);
705 if (info.skippedbufs) {
706 kprintf("Warning: vfsync skipped %d dirty "
709 ((info.skippedbufs > 1) ? "s" : ""));
712 while (error == 0 && passes > 0 &&
713 !RB_EMPTY(&vp->v_rbdirty_tree)
715 info.skippedbufs = 0;
717 info.synchronous = 1;
720 info.cmpfunc = vfsync_dummy_cmp;
721 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
727 error = vfsync_wait_output(vp, waitoutput);
728 if (info.skippedbufs && passes == 0) {
729 kprintf("Warning: vfsync skipped %d dirty "
730 "buf%s in final pass!\n",
732 ((info.skippedbufs > 1) ? "s" : ""));
737 * This case can occur normally because vnode lock might
740 if (!RB_EMPTY(&vp->v_rbdirty_tree))
741 kprintf("dirty bufs left after final pass\n");
745 lwkt_reltoken(&vp->v_token);
751 vfsync_wait_output(struct vnode *vp,
752 int (*waitoutput)(struct vnode *, struct thread *))
756 error = bio_track_wait(&vp->v_track_write, 0, 0);
758 error = waitoutput(vp, curthread);
763 vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused)
769 vfsync_data_only_cmp(struct buf *bp, void *data)
771 if (bp->b_loffset < 0)
777 vfsync_meta_only_cmp(struct buf *bp, void *data)
779 if (bp->b_loffset < 0)
785 vfsync_lazy_range_cmp(struct buf *bp, void *data)
787 struct vfsync_info *info = data;
789 if (bp->b_loffset < info->vp->v_lazyw)
795 vfsync_bp(struct buf *bp, void *data)
797 struct vfsync_info *info = data;
798 struct vnode *vp = info->vp;
801 if (info->fastpass) {
803 * Ignore buffers that we cannot immediately lock.
805 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
806 if (BUF_TIMELOCK(bp, LK_EXCLUSIVE, "bflst1", 1)) {
811 } else if (info->synchronous == 0) {
813 * Normal pass, give the buffer a little time to become
816 if (BUF_TIMELOCK(bp, LK_EXCLUSIVE, "bflst2", hz / 10)) {
822 * Synchronous pass, give the buffer a lot of time before
825 if (BUF_TIMELOCK(bp, LK_EXCLUSIVE, "bflst3", hz * 10)) {
832 * We must revalidate the buffer after locking.
834 if ((bp->b_flags & B_DELWRI) == 0 ||
835 bp->b_vp != info->vp ||
836 info->cmpfunc(bp, data)) {
842 * If syncdeps is not set we do not try to write buffers which have
845 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) {
851 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
852 * has been written but an additional handshake with the device
853 * is required before we can dispose of the buffer. We have no idea
854 * how to do this so we have to skip these buffers.
856 if (bp->b_flags & B_NEEDCOMMIT) {
862 * Ask bioops if it is ok to sync. If not the VFS may have
863 * set B_LOCKED so we have to cycle the buffer.
865 if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) {
871 if (info->synchronous) {
873 * Synchronous flush. An error may be returned and will
880 * Asynchronous flush. We use the error return to support
883 * In low-memory situations we revert to synchronous
884 * operation. This should theoretically prevent the I/O
885 * path from exhausting memory in a non-recoverable way.
887 vp->v_lazyw = bp->b_loffset;
889 if (vm_page_count_min(0)) {
891 info->lazycount += bp->b_bufsize;
895 info->lazycount += cluster_awrite(bp);
896 waitrunningbufspace();
897 /*vm_wait_nominal();*/
899 if (info->lazylimit && info->lazycount >= info->lazylimit)
908 * Associate a buffer with a vnode.
913 bgetvp(struct vnode *vp, struct buf *bp, int testsize)
915 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
916 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0);
919 * Insert onto list for new vnode.
921 lwkt_gettoken(&vp->v_token);
923 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) {
924 lwkt_reltoken(&vp->v_token);
929 * Diagnostics (mainly for HAMMER debugging). Check for
930 * overlapping buffers.
932 if (check_buf_overlap) {
934 bx = buf_rb_hash_RB_PREV(bp);
936 if (bx->b_loffset + bx->b_bufsize > bp->b_loffset) {
937 kprintf("bgetvp: overlapl %016jx/%d %016jx "
939 (intmax_t)bx->b_loffset,
941 (intmax_t)bp->b_loffset,
943 if (check_buf_overlap > 1)
944 panic("bgetvp - overlapping buffer");
947 bx = buf_rb_hash_RB_NEXT(bp);
949 if (bp->b_loffset + testsize > bx->b_loffset) {
950 kprintf("bgetvp: overlapr %016jx/%d %016jx "
952 (intmax_t)bp->b_loffset,
954 (intmax_t)bx->b_loffset,
956 if (check_buf_overlap > 1)
957 panic("bgetvp - overlapping buffer");
962 bp->b_flags |= B_HASHED;
963 bp->b_flags |= B_VNCLEAN;
964 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp))
965 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp);
967 lwkt_reltoken(&vp->v_token);
972 * Disassociate a buffer from a vnode.
977 brelvp(struct buf *bp)
981 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
984 * Delete from old vnode list, if on one.
987 lwkt_gettoken(&vp->v_token);
988 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) {
989 if (bp->b_flags & B_VNDIRTY)
990 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
992 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
993 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN);
995 if (bp->b_flags & B_HASHED) {
996 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp);
997 bp->b_flags &= ~B_HASHED;
1001 * Only remove from synclist when no dirty buffers are left AND
1002 * the VFS has not flagged the vnode's inode as being dirty.
1004 if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) == VONWORKLST &&
1005 RB_EMPTY(&vp->v_rbdirty_tree)) {
1006 vn_syncer_remove(vp, 0);
1010 lwkt_reltoken(&vp->v_token);
1016 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI.
1017 * This routine is called when the state of the B_DELWRI bit is changed.
1019 * Must be called with vp->v_token held.
1023 reassignbuf(struct buf *bp)
1025 struct vnode *vp = bp->b_vp;
1028 ASSERT_LWKT_TOKEN_HELD(&vp->v_token);
1032 * B_PAGING flagged buffers cannot be reassigned because their vp
1033 * is not fully linked in.
1035 if (bp->b_flags & B_PAGING)
1036 panic("cannot reassign paging buffer");
1038 if (bp->b_flags & B_DELWRI) {
1040 * Move to the dirty list, add the vnode to the worklist
1042 if (bp->b_flags & B_VNCLEAN) {
1043 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
1044 bp->b_flags &= ~B_VNCLEAN;
1046 if ((bp->b_flags & B_VNDIRTY) == 0) {
1047 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) {
1048 panic("reassignbuf: dup lblk vp %p bp %p",
1051 bp->b_flags |= B_VNDIRTY;
1053 if ((vp->v_flag & VONWORKLST) == 0) {
1054 switch (vp->v_type) {
1061 vp->v_rdev->si_mountpoint != NULL) {
1069 vn_syncer_add(vp, delay);
1073 * Move to the clean list, remove the vnode from the worklist
1074 * if no dirty blocks remain.
1076 if (bp->b_flags & B_VNDIRTY) {
1077 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
1078 bp->b_flags &= ~B_VNDIRTY;
1080 if ((bp->b_flags & B_VNCLEAN) == 0) {
1081 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) {
1082 panic("reassignbuf: dup lblk vp %p bp %p",
1085 bp->b_flags |= B_VNCLEAN;
1089 * Only remove from synclist when no dirty buffers are left
1090 * AND the VFS has not flagged the vnode's inode as being
1093 if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) ==
1095 RB_EMPTY(&vp->v_rbdirty_tree)) {
1096 vn_syncer_remove(vp, 0);
1102 * Create a vnode for a block device. Used for mounting the root file
1105 * A vref()'d vnode is returned.
1107 extern struct vop_ops *devfs_vnode_dev_vops_p;
1109 bdevvp(cdev_t dev, struct vnode **vpp)
1119 error = getspecialvnode(VT_NON, NULL, &devfs_vnode_dev_vops_p,
1130 v_associate_rdev(vp, dev);
1131 vp->v_umajor = dev->si_umajor;
1132 vp->v_uminor = dev->si_uminor;
1139 v_associate_rdev(struct vnode *vp, cdev_t dev)
1143 if (dev_is_good(dev) == 0)
1145 KKASSERT(vp->v_rdev == NULL);
1146 vp->v_rdev = reference_dev(dev);
1147 lwkt_gettoken(&spechash_token);
1148 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext);
1149 lwkt_reltoken(&spechash_token);
1154 v_release_rdev(struct vnode *vp)
1158 if ((dev = vp->v_rdev) != NULL) {
1159 lwkt_gettoken(&spechash_token);
1160 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext);
1163 lwkt_reltoken(&spechash_token);
1168 * Add a vnode to the alias list hung off the cdev_t. We only associate
1169 * the device number with the vnode. The actual device is not associated
1170 * until the vnode is opened (usually in spec_open()), and will be
1171 * disassociated on last close.
1174 addaliasu(struct vnode *nvp, int x, int y)
1176 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1177 panic("addaliasu on non-special vnode");
1183 * Simple call that a filesystem can make to try to get rid of a
1184 * vnode. It will fail if anyone is referencing the vnode (including
1187 * The filesystem can check whether its in-memory inode structure still
1188 * references the vp on return.
1190 * May only be called if the vnode is in a known state (i.e. being prevented
1191 * from being deallocated by some other condition such as a vfs inode hold).
1194 vclean_unlocked(struct vnode *vp)
1197 if (VREFCNT(vp) <= 1)
1203 * Disassociate a vnode from its underlying filesystem.
1205 * The vnode must be VX locked and referenced. In all normal situations
1206 * there are no active references. If vclean_vxlocked() is called while
1207 * there are active references, the vnode is being ripped out and we have
1208 * to call VOP_CLOSE() as appropriate before we can reclaim it.
1211 vclean_vxlocked(struct vnode *vp, int flags)
1216 struct namecache *ncp;
1219 * If the vnode has already been reclaimed we have nothing to do.
1221 if (vp->v_flag & VRECLAIMED)
1225 * Set flag to interlock operation, flag finalization to ensure
1226 * that the vnode winds up on the inactive list, and set v_act to 0.
1228 vsetflags(vp, VRECLAIMED);
1229 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE);
1232 if (verbose_reclaims) {
1233 if ((ncp = TAILQ_FIRST(&vp->v_namecache)) != NULL)
1234 kprintf("Debug: reclaim %p %s\n", vp, ncp->nc_name);
1238 * Scrap the vfs cache
1240 while (cache_inval_vp(vp, 0) != 0) {
1241 kprintf("Warning: vnode %p clean/cache_resolution "
1242 "race detected\n", vp);
1243 tsleep(vp, 0, "vclninv", 2);
1247 * Check to see if the vnode is in use. If so we have to reference it
1248 * before we clean it out so that its count cannot fall to zero and
1249 * generate a race against ourselves to recycle it.
1251 active = (VREFCNT(vp) > 0);
1254 * Clean out any buffers associated with the vnode and destroy its
1255 * object, if it has one.
1257 vinvalbuf(vp, V_SAVE, 0, 0);
1258 KKASSERT(lockcountnb(&vp->v_lock) == 1);
1261 * If purging an active vnode (typically during a forced unmount
1262 * or reboot), it must be closed and deactivated before being
1263 * reclaimed. This isn't really all that safe, but what can
1266 * Note that neither of these routines unlocks the vnode.
1268 if (active && (flags & DOCLOSE)) {
1269 while ((n = vp->v_opencount) != 0) {
1270 if (vp->v_writecount)
1271 VOP_CLOSE(vp, FWRITE|FNONBLOCK, NULL);
1273 VOP_CLOSE(vp, FNONBLOCK, NULL);
1274 if (vp->v_opencount == n) {
1275 kprintf("Warning: unable to force-close"
1283 * If the vnode has not been deactivated, deactivated it. Deactivation
1284 * can create new buffers and VM pages so we have to call vinvalbuf()
1285 * again to make sure they all get flushed.
1287 * This can occur if a file with a link count of 0 needs to be
1290 * If the vnode is already dead don't try to deactivate it.
1292 if ((vp->v_flag & VINACTIVE) == 0) {
1293 vsetflags(vp, VINACTIVE);
1296 vinvalbuf(vp, V_SAVE, 0, 0);
1298 KKASSERT(lockcountnb(&vp->v_lock) == 1);
1301 * If the vnode has an object, destroy it.
1303 while ((object = vp->v_object) != NULL) {
1304 vm_object_hold(object);
1305 if (object == vp->v_object)
1307 vm_object_drop(object);
1310 if (object != NULL) {
1311 if (object->ref_count == 0) {
1312 if ((object->flags & OBJ_DEAD) == 0)
1313 vm_object_terminate(object);
1314 vm_object_drop(object);
1315 vclrflags(vp, VOBJBUF);
1317 vm_pager_deallocate(object);
1318 vclrflags(vp, VOBJBUF);
1319 vm_object_drop(object);
1322 KKASSERT((vp->v_flag & VOBJBUF) == 0);
1325 * Reclaim the vnode if not already dead.
1327 if (vp->v_mount && VOP_RECLAIM(vp))
1328 panic("vclean: cannot reclaim");
1331 * Done with purge, notify sleepers of the grim news.
1333 vp->v_ops = &dead_vnode_vops_p;
1338 * If we are destroying an active vnode, reactivate it now that
1339 * we have reassociated it with deadfs. This prevents the system
1340 * from crashing on the vnode due to it being unexpectedly marked
1341 * as inactive or reclaimed.
1343 if (active && (flags & DOCLOSE)) {
1344 vclrflags(vp, VINACTIVE | VRECLAIMED);
1349 * Eliminate all activity associated with the requested vnode
1350 * and with all vnodes aliased to the requested vnode.
1352 * The vnode must be referenced but should not be locked.
1355 vrevoke(struct vnode *vp, struct ucred *cred)
1363 * If the vnode has a device association, scrap all vnodes associated
1364 * with the device. Don't let the device disappear on us while we
1365 * are scrapping the vnodes.
1367 * The passed vp will probably show up in the list, do not VX lock
1370 * Releasing the vnode's rdev here can mess up specfs's call to
1371 * device close, so don't do it. The vnode has been disassociated
1372 * and the device will be closed after the last ref on the related
1373 * fp goes away (if not still open by e.g. the kernel).
1375 if (vp->v_type != VCHR) {
1376 error = fdrevoke(vp, DTYPE_VNODE, cred);
1379 if ((dev = vp->v_rdev) == NULL) {
1383 lwkt_gettoken(&spechash_token);
1386 vqn = SLIST_FIRST(&dev->si_hlist);
1389 while ((vq = vqn) != NULL) {
1390 if (VREFCNT(vq) > 0) {
1392 fdrevoke(vq, DTYPE_VNODE, cred);
1393 /*v_release_rdev(vq);*/
1395 if (vq->v_rdev != dev) {
1400 vqn = SLIST_NEXT(vq, v_cdevnext);
1405 lwkt_reltoken(&spechash_token);
1412 * This is called when the object underlying a vnode is being destroyed,
1413 * such as in a remove(). Try to recycle the vnode immediately if the
1414 * only active reference is our reference.
1416 * Directory vnodes in the namecache with children cannot be immediately
1417 * recycled because numerous VOP_N*() ops require them to be stable.
1419 * To avoid recursive recycling from VOP_INACTIVE implemenetations this
1420 * function is a NOP if VRECLAIMED is already set.
1423 vrecycle(struct vnode *vp)
1425 if (VREFCNT(vp) <= 1 && (vp->v_flag & VRECLAIMED) == 0) {
1426 if (cache_inval_vp_nonblock(vp))
1435 * Return the maximum I/O size allowed for strategy calls on VP.
1437 * If vp is VCHR or VBLK we dive the device, otherwise we use
1438 * the vp's mount info.
1440 * The returned value is clamped at MAXPHYS as most callers cannot use
1441 * buffers larger than that size.
1444 vmaxiosize(struct vnode *vp)
1448 if (vp->v_type == VBLK || vp->v_type == VCHR)
1449 maxiosize = vp->v_rdev->si_iosize_max;
1451 maxiosize = vp->v_mount->mnt_iosize_max;
1453 if (maxiosize > MAXPHYS)
1454 maxiosize = MAXPHYS;
1459 * Eliminate all activity associated with a vnode in preparation for
1462 * The vnode must be VX locked and refd and will remain VX locked and refd
1463 * on return. This routine may be called with the vnode in any state, as
1464 * long as it is VX locked. The vnode will be cleaned out and marked
1465 * VRECLAIMED but will not actually be reused until all existing refs and
1468 * NOTE: This routine may be called on a vnode which has not yet been
1469 * already been deactivated (VOP_INACTIVE), or on a vnode which has
1470 * already been reclaimed.
1472 * This routine is not responsible for placing us back on the freelist.
1473 * Instead, it happens automatically when the caller releases the VX lock
1474 * (assuming there aren't any other references).
1477 vgone_vxlocked(struct vnode *vp)
1480 * assert that the VX lock is held. This is an absolute requirement
1481 * now for vgone_vxlocked() to be called.
1483 KKASSERT(lockcountnb(&vp->v_lock) == 1);
1486 * Clean out the filesystem specific data and set the VRECLAIMED
1487 * bit. Also deactivate the vnode if necessary.
1489 * The vnode should have automatically been removed from the syncer
1490 * list as syncer/dirty flags cleared during the cleaning.
1492 vclean_vxlocked(vp, DOCLOSE);
1495 * Normally panic if the vnode is still dirty, unless we are doing
1496 * a forced unmount (tmpfs typically).
1498 if (vp->v_flag & VONWORKLST) {
1499 if (vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF) {
1501 vn_syncer_remove(vp, 1);
1503 panic("vp %p still dirty in vgone after flush", vp);
1508 * Delete from old mount point vnode list, if on one.
1510 if (vp->v_mount != NULL) {
1511 KKASSERT(vp->v_data == NULL);
1512 insmntque(vp, NULL);
1516 * If special device, remove it from special device alias list
1517 * if it is on one. This should normally only occur if a vnode is
1518 * being revoked as the device should otherwise have been released
1521 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
1532 * Lookup a vnode by device number.
1534 * Returns non-zero and *vpp set to a vref'd vnode on success.
1535 * Returns zero on failure.
1538 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp)
1542 lwkt_gettoken(&spechash_token);
1543 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1544 if (type == vp->v_type) {
1547 lwkt_reltoken(&spechash_token);
1551 lwkt_reltoken(&spechash_token);
1556 * Calculate the total number of references to a special device. This
1557 * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1558 * an overloaded field. Since udev2dev can now return NULL, we have
1559 * to check for a NULL v_rdev.
1562 count_dev(cdev_t dev)
1567 if (SLIST_FIRST(&dev->si_hlist)) {
1568 lwkt_gettoken(&spechash_token);
1569 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1570 count += vp->v_opencount;
1572 lwkt_reltoken(&spechash_token);
1578 vcount(struct vnode *vp)
1580 if (vp->v_rdev == NULL)
1582 return(count_dev(vp->v_rdev));
1586 * Initialize VMIO for a vnode. This routine MUST be called before a
1587 * VFS can issue buffer cache ops on a vnode. It is typically called
1588 * when a vnode is initialized from its inode.
1591 vinitvmio(struct vnode *vp, off_t filesize, int blksize, int boff)
1596 object = vp->v_object;
1598 vm_object_hold(object);
1599 KKASSERT(vp->v_object == object);
1602 if (object == NULL) {
1603 object = vnode_pager_alloc(vp, filesize, 0, 0, blksize, boff);
1606 * Dereference the reference we just created. This assumes
1607 * that the object is associated with the vp. Allow it to
1608 * have zero refs. It cannot be destroyed as long as it
1609 * is associated with the vnode.
1611 vm_object_hold(object);
1612 atomic_add_int(&object->ref_count, -1);
1615 KKASSERT((object->flags & OBJ_DEAD) == 0);
1617 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object"));
1618 vsetflags(vp, VOBJBUF);
1619 vm_object_drop(object);
1626 * Print out a description of a vnode.
1628 static char *typename[] =
1629 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1632 vprint(char *label, struct vnode *vp)
1637 kprintf("%s: %p: ", label, (void *)vp);
1639 kprintf("%p: ", (void *)vp);
1640 kprintf("type %s, refcnt %08x, writecount %d, holdcnt %d,",
1641 typename[vp->v_type],
1642 vp->v_refcnt, vp->v_writecount, vp->v_auxrefs);
1644 if (vp->v_flag & VROOT)
1645 strcat(buf, "|VROOT");
1646 if (vp->v_flag & VPFSROOT)
1647 strcat(buf, "|VPFSROOT");
1648 if (vp->v_flag & VTEXT)
1649 strcat(buf, "|VTEXT");
1650 if (vp->v_flag & VSYSTEM)
1651 strcat(buf, "|VSYSTEM");
1652 if (vp->v_flag & VOBJBUF)
1653 strcat(buf, "|VOBJBUF");
1655 kprintf(" flags (%s)", &buf[1]);
1656 if (vp->v_data == NULL) {
1665 * Do the usual access checking.
1666 * file_mode, uid and gid are from the vnode in question,
1667 * while acc_mode and cred are from the VOP_ACCESS parameter list
1670 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
1671 mode_t acc_mode, struct ucred *cred)
1677 * Super-user always gets read/write access, but execute access depends
1678 * on at least one execute bit being set.
1680 if (priv_check_cred(cred, PRIV_ROOT, 0) == 0) {
1681 if ((acc_mode & VEXEC) && type != VDIR &&
1682 (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)
1689 /* Otherwise, check the owner. */
1690 if (cred->cr_uid == uid) {
1691 if (acc_mode & VEXEC)
1693 if (acc_mode & VREAD)
1695 if (acc_mode & VWRITE)
1697 return ((file_mode & mask) == mask ? 0 : EACCES);
1700 /* Otherwise, check the groups. */
1701 ismember = groupmember(gid, cred);
1702 if (cred->cr_svgid == gid || ismember) {
1703 if (acc_mode & VEXEC)
1705 if (acc_mode & VREAD)
1707 if (acc_mode & VWRITE)
1709 return ((file_mode & mask) == mask ? 0 : EACCES);
1712 /* Otherwise, check everyone else. */
1713 if (acc_mode & VEXEC)
1715 if (acc_mode & VREAD)
1717 if (acc_mode & VWRITE)
1719 return ((file_mode & mask) == mask ? 0 : EACCES);
1723 #include <ddb/ddb.h>
1725 static int db_show_locked_vnodes(struct mount *mp, void *data);
1728 * List all of the locked vnodes in the system.
1729 * Called when debugging the kernel.
1731 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
1733 kprintf("Locked vnodes\n");
1734 mountlist_scan(db_show_locked_vnodes, NULL,
1735 MNTSCAN_FORWARD|MNTSCAN_NOBUSY);
1739 db_show_locked_vnodes(struct mount *mp, void *data __unused)
1743 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
1744 if (vn_islocked(vp))
1752 * Top level filesystem related information gathering.
1754 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
1757 vfs_sysctl(SYSCTL_HANDLER_ARGS)
1759 int *name = (int *)arg1 - 1; /* XXX */
1760 u_int namelen = arg2 + 1; /* XXX */
1761 struct vfsconf *vfsp;
1764 #if 1 || defined(COMPAT_PRELITE2)
1765 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1767 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
1771 /* all sysctl names at this level are at least name and field */
1773 return (ENOTDIR); /* overloaded */
1774 if (name[0] != VFS_GENERIC) {
1775 vfsp = vfsconf_find_by_typenum(name[0]);
1777 return (EOPNOTSUPP);
1778 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1779 oldp, oldlenp, newp, newlen, p));
1783 case VFS_MAXTYPENUM:
1786 maxtypenum = vfsconf_get_maxtypenum();
1787 return (SYSCTL_OUT(req, &maxtypenum, sizeof(maxtypenum)));
1790 return (ENOTDIR); /* overloaded */
1791 vfsp = vfsconf_find_by_typenum(name[2]);
1793 return (EOPNOTSUPP);
1794 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
1796 return (EOPNOTSUPP);
1799 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
1800 "Generic filesystem");
1802 #if 1 || defined(COMPAT_PRELITE2)
1805 sysctl_ovfs_conf_iter(struct vfsconf *vfsp, void *data)
1808 struct ovfsconf ovfs;
1809 struct sysctl_req *req = (struct sysctl_req*) data;
1811 bzero(&ovfs, sizeof(ovfs));
1812 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
1813 strcpy(ovfs.vfc_name, vfsp->vfc_name);
1814 ovfs.vfc_index = vfsp->vfc_typenum;
1815 ovfs.vfc_refcount = vfsp->vfc_refcount;
1816 ovfs.vfc_flags = vfsp->vfc_flags;
1817 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
1819 return error; /* abort iteration with error code */
1821 return 0; /* continue iterating with next element */
1825 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
1827 return vfsconf_each(sysctl_ovfs_conf_iter, (void*)req);
1830 #endif /* 1 || COMPAT_PRELITE2 */
1833 * Check to see if a filesystem is mounted on a block device.
1836 vfs_mountedon(struct vnode *vp)
1840 if ((dev = vp->v_rdev) == NULL) {
1841 /* if (vp->v_type != VBLK)
1842 dev = get_dev(vp->v_uminor, vp->v_umajor); */
1844 if (dev != NULL && dev->si_mountpoint)
1850 * Unmount all filesystems. The list is traversed in reverse order
1851 * of mounting to avoid dependencies.
1853 * We want the umountall to be able to break out of its loop if a
1854 * failure occurs, after scanning all possible mounts, so the callback
1855 * returns 0 on error.
1857 * NOTE: Do not call mountlist_remove(mp) on error any more, this will
1858 * confuse mountlist_scan()'s unbusy check.
1860 static int vfs_umountall_callback(struct mount *mp, void *data);
1863 vfs_unmountall(void)
1868 count = mountlist_scan(vfs_umountall_callback,
1869 NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY);
1875 vfs_umountall_callback(struct mount *mp, void *data)
1879 error = dounmount(mp, MNT_FORCE);
1881 kprintf("unmount of filesystem mounted from %s failed (",
1882 mp->mnt_stat.f_mntfromname);
1886 kprintf("%d)\n", error);
1894 * Checks the mount flags for parameter mp and put the names comma-separated
1895 * into a string buffer buf with a size limit specified by len.
1897 * It returns the number of bytes written into buf, and (*errorp) will be
1898 * set to 0, EINVAL (if passed length is 0), or ENOSPC (supplied buffer was
1899 * not large enough). The buffer will be 0-terminated if len was not 0.
1902 vfs_flagstostr(int flags, const struct mountctl_opt *optp,
1903 char *buf, size_t len, int *errorp)
1905 static const struct mountctl_opt optnames[] = {
1906 { MNT_RDONLY, "read-only" },
1907 { MNT_SYNCHRONOUS, "synchronous" },
1908 { MNT_NOEXEC, "noexec" },
1909 { MNT_NOSUID, "nosuid" },
1910 { MNT_NODEV, "nodev" },
1911 { MNT_AUTOMOUNTED, "automounted" },
1912 { MNT_ASYNC, "asynchronous" },
1913 { MNT_SUIDDIR, "suiddir" },
1914 { MNT_SOFTDEP, "soft-updates" },
1915 { MNT_NOSYMFOLLOW, "nosymfollow" },
1916 { MNT_TRIM, "trim" },
1917 { MNT_NOATIME, "noatime" },
1918 { MNT_NOCLUSTERR, "noclusterr" },
1919 { MNT_NOCLUSTERW, "noclusterw" },
1920 { MNT_EXRDONLY, "NFS read-only" },
1921 { MNT_EXPORTED, "NFS exported" },
1922 /* Remaining NFS flags could come here */
1923 { MNT_LOCAL, "local" },
1924 { MNT_QUOTA, "with-quotas" },
1925 /* { MNT_ROOTFS, "rootfs" }, */
1926 /* { MNT_IGNORE, "ignore" }, */
1936 bleft = len - 1; /* leave room for trailing \0 */
1939 * Checks the size of the string. If it contains
1940 * any data, then we will append the new flags to
1943 actsize = strlen(buf);
1947 /* Default flags if no flags passed */
1951 if (bleft < 0) { /* degenerate case, 0-length buffer */
1956 for (; flags && optp->o_opt; ++optp) {
1957 if ((flags & optp->o_opt) == 0)
1959 optlen = strlen(optp->o_name);
1960 if (bwritten || actsize > 0) {
1965 buf[bwritten++] = ',';
1966 buf[bwritten++] = ' ';
1969 if (bleft < optlen) {
1973 bcopy(optp->o_name, buf + bwritten, optlen);
1976 flags &= ~optp->o_opt;
1980 * Space already reserved for trailing \0
1987 * Build hash lists of net addresses and hang them off the mount point.
1988 * Called by ufs_mount() to set up the lists of export addresses.
1991 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1992 const struct export_args *argp)
1995 struct radix_node_head *rnh;
1997 struct radix_node *rn;
1998 struct sockaddr *saddr, *smask = NULL;
2001 if (argp->ex_addrlen == 0) {
2002 if (mp->mnt_flag & MNT_DEFEXPORTED)
2004 np = &nep->ne_defexported;
2005 np->netc_exflags = argp->ex_flags;
2006 np->netc_anon = argp->ex_anon;
2007 np->netc_anon.cr_ref = 1;
2008 mp->mnt_flag |= MNT_DEFEXPORTED;
2012 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
2014 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
2017 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2018 np = (struct netcred *)kmalloc(i, M_NETCRED, M_WAITOK | M_ZERO);
2019 saddr = (struct sockaddr *) (np + 1);
2020 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2022 if (saddr->sa_len > argp->ex_addrlen)
2023 saddr->sa_len = argp->ex_addrlen;
2024 if (argp->ex_masklen) {
2025 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
2026 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
2029 if (smask->sa_len > argp->ex_masklen)
2030 smask->sa_len = argp->ex_masklen;
2033 if (nep->ne_maskhead == NULL) {
2034 if (!rn_inithead((void **)&nep->ne_maskhead, NULL, 0)) {
2039 if ((rnh = vfs_create_addrlist_af(saddr->sa_family, nep)) == NULL) {
2043 rn = (*rnh->rnh_addaddr)((char *)saddr, (char *)smask, rnh,
2046 if (rn == NULL || np != (struct netcred *)rn) { /* already exists */
2050 np->netc_exflags = argp->ex_flags;
2051 np->netc_anon = argp->ex_anon;
2052 np->netc_anon.cr_ref = 1;
2056 kfree(np, M_NETCRED);
2061 * Free netcred structures installed in the netexport
2064 vfs_free_netcred(struct radix_node *rn, void *w)
2066 struct radix_node_head *rnh = (struct radix_node_head *)w;
2068 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2069 kfree(rn, M_NETCRED);
2075 * callback to free an element of the mask table installed in the
2076 * netexport. These may be created indirectly and are not netcred
2080 vfs_free_netcred_mask(struct radix_node *rn, void *w)
2082 struct radix_node_head *rnh = (struct radix_node_head *)w;
2084 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2085 kfree(rn, M_RTABLE);
2090 static struct radix_node_head *
2091 vfs_create_addrlist_af(int af, struct netexport *nep)
2093 struct radix_node_head *rnh = NULL;
2094 #if defined(INET) || defined(INET6)
2095 struct radix_node_head *maskhead = nep->ne_maskhead;
2099 NE_ASSERT_LOCKED(nep);
2100 KKASSERT(maskhead != NULL);
2104 if ((rnh = nep->ne_inethead) == NULL) {
2105 off = offsetof(struct sockaddr_in, sin_addr) << 3;
2106 if (!rn_inithead((void **)&rnh, maskhead, off))
2108 nep->ne_inethead = rnh;
2114 if ((rnh = nep->ne_inet6head) == NULL) {
2115 off = offsetof(struct sockaddr_in6, sin6_addr) << 3;
2116 if (!rn_inithead((void **)&rnh, maskhead, off))
2118 nep->ne_inet6head = rnh;
2127 * helper function for freeing netcred elements
2130 vfs_free_addrlist_af(struct radix_node_head **prnh)
2132 struct radix_node_head *rnh = *prnh;
2134 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, rnh);
2135 kfree(rnh, M_RTABLE);
2140 * helper function for freeing mask elements
2143 vfs_free_addrlist_masks(struct radix_node_head **prnh)
2145 struct radix_node_head *rnh = *prnh;
2147 (*rnh->rnh_walktree) (rnh, vfs_free_netcred_mask, rnh);
2148 kfree(rnh, M_RTABLE);
2153 * Free the net address hash lists that are hanging off the mount points.
2156 vfs_free_addrlist(struct netexport *nep)
2159 if (nep->ne_inethead != NULL)
2160 vfs_free_addrlist_af(&nep->ne_inethead);
2161 if (nep->ne_inet6head != NULL)
2162 vfs_free_addrlist_af(&nep->ne_inet6head);
2163 if (nep->ne_maskhead)
2164 vfs_free_addrlist_masks(&nep->ne_maskhead);
2169 vfs_export(struct mount *mp, struct netexport *nep,
2170 const struct export_args *argp)
2174 if (argp->ex_flags & MNT_DELEXPORT) {
2175 if (mp->mnt_flag & MNT_EXPUBLIC) {
2176 vfs_setpublicfs(NULL, NULL, NULL);
2177 mp->mnt_flag &= ~MNT_EXPUBLIC;
2179 vfs_free_addrlist(nep);
2180 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2182 if (argp->ex_flags & MNT_EXPORTED) {
2183 if (argp->ex_flags & MNT_EXPUBLIC) {
2184 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2186 mp->mnt_flag |= MNT_EXPUBLIC;
2188 if ((error = vfs_hang_addrlist(mp, nep, argp)))
2190 mp->mnt_flag |= MNT_EXPORTED;
2197 * Set the publicly exported filesystem (WebNFS). Currently, only
2198 * one public filesystem is possible in the spec (RFC 2054 and 2055)
2201 vfs_setpublicfs(struct mount *mp, struct netexport *nep,
2202 const struct export_args *argp)
2209 * mp == NULL -> invalidate the current info, the FS is
2210 * no longer exported. May be called from either vfs_export
2211 * or unmount, so check if it hasn't already been done.
2214 if (nfs_pub.np_valid) {
2215 nfs_pub.np_valid = 0;
2216 if (nfs_pub.np_index != NULL) {
2217 kfree(nfs_pub.np_index, M_TEMP);
2218 nfs_pub.np_index = NULL;
2225 * Only one allowed at a time.
2227 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2231 * Get real filehandle for root of exported FS.
2233 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2234 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2236 if ((error = VFS_ROOT(mp, &rvp)))
2239 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2245 * If an indexfile was specified, pull it in.
2247 if (argp->ex_indexfile != NULL) {
2250 error = vn_get_namelen(rvp, &namelen);
2253 nfs_pub.np_index = kmalloc(namelen, M_TEMP, M_WAITOK);
2254 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2258 * Check for illegal filenames.
2260 for (cp = nfs_pub.np_index; *cp; cp++) {
2268 kfree(nfs_pub.np_index, M_TEMP);
2273 nfs_pub.np_mount = mp;
2274 nfs_pub.np_valid = 1;
2279 vfs_export_lookup(struct mount *mp, struct netexport *nep,
2280 struct sockaddr *nam)
2283 struct radix_node_head *rnh;
2284 struct sockaddr *saddr;
2287 if (mp->mnt_flag & MNT_EXPORTED) {
2289 * Lookup in the export list first.
2294 switch (saddr->sa_family) {
2297 rnh = nep->ne_inethead;
2302 rnh = nep->ne_inet6head;
2309 np = (struct netcred *)
2310 (*rnh->rnh_matchaddr)((char *)saddr,
2312 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2318 * If no address match, use the default if it exists.
2320 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2321 np = &nep->ne_defexported;
2327 * perform msync on all vnodes under a mount point. The mount point must
2328 * be locked. This code is also responsible for lazy-freeing unreferenced
2329 * vnodes whos VM objects no longer contain pages.
2331 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
2333 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked,
2334 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it
2335 * way up in this high level function.
2337 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
2338 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data);
2341 vfs_msync(struct mount *mp, int flags)
2346 * tmpfs sets this flag to prevent msync(), sync, and the
2347 * filesystem periodic syncer from trying to flush VM pages
2348 * to swap. Only pure memory pressure flushes tmpfs VM pages
2351 if (mp->mnt_kern_flag & MNTK_NOMSYNC)
2355 * Ok, scan the vnodes for work. If the filesystem is using the
2356 * syncer thread feature we can use vsyncscan() instead of
2357 * vmntvnodescan(), which is much faster.
2359 vmsc_flags = VMSC_GETVP;
2360 if (flags != MNT_WAIT)
2361 vmsc_flags |= VMSC_NOWAIT;
2363 if (mp->mnt_kern_flag & MNTK_THR_SYNC) {
2364 vsyncscan(mp, vmsc_flags, vfs_msync_scan2,
2365 (void *)(intptr_t)flags);
2367 vmntvnodescan(mp, vmsc_flags,
2368 vfs_msync_scan1, vfs_msync_scan2,
2369 (void *)(intptr_t)flags);
2374 * scan1 is a fast pre-check. There could be hundreds of thousands of
2375 * vnodes, we cannot afford to do anything heavy weight until we have a
2376 * fairly good indication that there is work to do.
2380 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
2382 int flags = (int)(intptr_t)data;
2384 if ((vp->v_flag & VRECLAIMED) == 0) {
2385 if (vp->v_auxrefs == 0 && VREFCNT(vp) <= 0 &&
2387 return(0); /* call scan2 */
2389 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
2390 (vp->v_flag & VOBJDIRTY) &&
2391 (flags == MNT_WAIT || vn_islocked(vp) == 0)) {
2392 return(0); /* call scan2 */
2397 * do not call scan2, continue the loop
2403 * This callback is handed a locked vnode.
2407 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data)
2410 int flags = (int)(intptr_t)data;
2412 if (vp->v_flag & VRECLAIMED)
2415 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) {
2416 if ((obj = vp->v_object) != NULL) {
2417 vm_object_page_clean(obj, 0, 0,
2418 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2425 * Wake up anyone interested in vp because it is being revoked.
2428 vn_gone(struct vnode *vp)
2430 lwkt_gettoken(&vp->v_token);
2431 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, NOTE_REVOKE);
2432 lwkt_reltoken(&vp->v_token);
2436 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened
2437 * (or v_rdev might be NULL).
2440 vn_todev(struct vnode *vp)
2442 if (vp->v_type != VBLK && vp->v_type != VCHR)
2444 KKASSERT(vp->v_rdev != NULL);
2445 return (vp->v_rdev);
2449 * Check if vnode represents a disk device. The vnode does not need to be
2455 vn_isdisk(struct vnode *vp, int *errp)
2459 if (vp->v_type != VCHR) {
2472 if (dev_is_good(dev) == 0) {
2477 if ((dev_dflags(dev) & D_DISK) == 0) {
2488 vn_get_namelen(struct vnode *vp, int *namelen)
2491 register_t retval[2];
2493 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval);
2496 *namelen = (int)retval[0];
2501 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type,
2502 uint16_t d_namlen, const char *d_name)
2507 len = _DIRENT_RECLEN(d_namlen);
2508 if (len > uio->uio_resid)
2511 dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO);
2514 dp->d_namlen = d_namlen;
2515 dp->d_type = d_type;
2516 bcopy(d_name, dp->d_name, d_namlen);
2518 *error = uiomove((caddr_t)dp, len, uio);
2526 vn_mark_atime(struct vnode *vp, struct thread *td)
2528 struct proc *p = td->td_proc;
2529 struct ucred *cred = p ? p->p_ucred : proc0.p_ucred;
2531 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
2532 VOP_MARKATIME(vp, cred);
2537 * Calculate the number of entries in an inode-related chained hash table.
2538 * With today's memory sizes, maxvnodes can wind up being a very large
2539 * number. There is no reason to waste memory, so tolerate some stacking.
2542 vfs_inodehashsize(void)
2547 while (hsize < maxvnodes)
2549 while (hsize > maxvnodes * 2)
2550 hsize >>= 1; /* nominal 2x stacking */
2552 if (maxvnodes > 1024 * 1024)
2553 hsize >>= 1; /* nominal 8x stacking */
2555 if (maxvnodes > 128 * 1024)
2556 hsize >>= 1; /* nominal 4x stacking */