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>
87 #include <netinet/in.h>
89 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
92 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
93 "Number of vnodes allocated");
95 SYSCTL_INT(_debug, OID_AUTO, verbose_reclaims, CTLFLAG_RD, &verbose_reclaims, 0,
96 "Output filename of reclaimed vnode(s)");
98 enum vtype iftovt_tab[16] = {
99 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
100 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
102 int vttoif_tab[9] = {
103 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
104 S_IFSOCK, S_IFIFO, S_IFMT,
107 static int reassignbufcalls;
108 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls,
109 0, "Number of times buffers have been reassigned to the proper list");
111 static int check_buf_overlap = 2; /* invasive check */
112 SYSCTL_INT(_vfs, OID_AUTO, check_buf_overlap, CTLFLAG_RW, &check_buf_overlap,
113 0, "Enable overlapping buffer checks");
115 int nfs_mount_type = -1;
116 static struct lwkt_token spechash_token;
117 struct nfs_public nfs_pub; /* publicly exported FS */
120 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
121 &desiredvnodes, 0, "Maximum number of vnodes");
123 static struct radix_node_head *vfs_create_addrlist_af(int af,
124 struct netexport *nep);
125 static void vfs_free_addrlist (struct netexport *nep);
126 static int vfs_free_netcred (struct radix_node *rn, void *w);
127 static void vfs_free_addrlist_af (struct radix_node_head **prnh);
128 static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
129 const struct export_args *argp);
131 int prtactive = 0; /* 1 => print out reclaim of active vnodes */
134 * Red black tree functions
136 static int rb_buf_compare(struct buf *b1, struct buf *b2);
137 RB_GENERATE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t, b_loffset);
138 RB_GENERATE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t, b_loffset);
141 rb_buf_compare(struct buf *b1, struct buf *b2)
143 if (b1->b_loffset < b2->b_loffset)
145 if (b1->b_loffset > b2->b_loffset)
151 * Initialize the vnode management data structures.
153 * Called from vfsinit()
162 * Desiredvnodes is kern.maxvnodes. We want to scale it
163 * according to available system memory but we may also have
164 * to limit it based on available KVM, which is capped on 32 bit
165 * systems, to ~80K vnodes or so.
167 * WARNING! For machines with 64-256M of ram we have to be sure
168 * that the default limit scales down well due to HAMMER
169 * taking up significantly more memory per-vnode vs UFS.
170 * We want around ~5800 on a 128M machine.
172 factor1 = 20 * (sizeof(struct vm_object) + sizeof(struct vnode));
173 factor2 = 25 * (sizeof(struct vm_object) + sizeof(struct vnode));
175 imin((int64_t)vmstats.v_page_count * PAGE_SIZE / factor1,
177 desiredvnodes = imax(desiredvnodes, 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" : ""));
735 if (!RB_EMPTY(&vp->v_rbdirty_tree))
736 kprintf("dirty bufs left after final pass\n");
739 lwkt_reltoken(&vp->v_token);
745 vfsync_wait_output(struct vnode *vp,
746 int (*waitoutput)(struct vnode *, struct thread *))
750 error = bio_track_wait(&vp->v_track_write, 0, 0);
752 error = waitoutput(vp, curthread);
757 vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused)
763 vfsync_data_only_cmp(struct buf *bp, void *data)
765 if (bp->b_loffset < 0)
771 vfsync_meta_only_cmp(struct buf *bp, void *data)
773 if (bp->b_loffset < 0)
779 vfsync_lazy_range_cmp(struct buf *bp, void *data)
781 struct vfsync_info *info = data;
783 if (bp->b_loffset < info->vp->v_lazyw)
789 vfsync_bp(struct buf *bp, void *data)
791 struct vfsync_info *info = data;
792 struct vnode *vp = info->vp;
795 if (info->fastpass) {
797 * Ignore buffers that we cannot immediately lock.
799 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
800 if (BUF_TIMELOCK(bp, LK_EXCLUSIVE, "bflst1", 1)) {
805 } else if (info->synchronous == 0) {
807 * Normal pass, give the buffer a little time to become
810 if (BUF_TIMELOCK(bp, LK_EXCLUSIVE, "bflst2", hz / 10)) {
816 * Synchronous pass, give the buffer a lot of time before
819 if (BUF_TIMELOCK(bp, LK_EXCLUSIVE, "bflst3", hz * 10)) {
826 * We must revalidate the buffer after locking.
828 if ((bp->b_flags & B_DELWRI) == 0 ||
829 bp->b_vp != info->vp ||
830 info->cmpfunc(bp, data)) {
836 * If syncdeps is not set we do not try to write buffers which have
839 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) {
845 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
846 * has been written but an additional handshake with the device
847 * is required before we can dispose of the buffer. We have no idea
848 * how to do this so we have to skip these buffers.
850 if (bp->b_flags & B_NEEDCOMMIT) {
856 * Ask bioops if it is ok to sync. If not the VFS may have
857 * set B_LOCKED so we have to cycle the buffer.
859 if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) {
865 if (info->synchronous) {
867 * Synchronous flushing. An error may be returned.
873 * Asynchronous flushing. A negative return value simply
874 * stops the scan and is not considered an error. We use
875 * this to support limited MNT_LAZY flushes.
877 vp->v_lazyw = bp->b_loffset;
879 info->lazycount += cluster_awrite(bp);
880 waitrunningbufspace();
882 if (info->lazylimit && info->lazycount >= info->lazylimit)
891 * Associate a buffer with a vnode.
896 bgetvp(struct vnode *vp, struct buf *bp, int testsize)
898 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
899 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0);
902 * Insert onto list for new vnode.
904 lwkt_gettoken(&vp->v_token);
906 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) {
907 lwkt_reltoken(&vp->v_token);
912 * Diagnostics (mainly for HAMMER debugging). Check for
913 * overlapping buffers.
915 if (check_buf_overlap) {
917 bx = buf_rb_hash_RB_PREV(bp);
919 if (bx->b_loffset + bx->b_bufsize > bp->b_loffset) {
920 kprintf("bgetvp: overlapl %016jx/%d %016jx "
922 (intmax_t)bx->b_loffset,
924 (intmax_t)bp->b_loffset,
926 if (check_buf_overlap > 1)
927 panic("bgetvp - overlapping buffer");
930 bx = buf_rb_hash_RB_NEXT(bp);
932 if (bp->b_loffset + testsize > bx->b_loffset) {
933 kprintf("bgetvp: overlapr %016jx/%d %016jx "
935 (intmax_t)bp->b_loffset,
937 (intmax_t)bx->b_loffset,
939 if (check_buf_overlap > 1)
940 panic("bgetvp - overlapping buffer");
945 bp->b_flags |= B_HASHED;
946 bp->b_flags |= B_VNCLEAN;
947 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp))
948 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp);
950 lwkt_reltoken(&vp->v_token);
955 * Disassociate a buffer from a vnode.
960 brelvp(struct buf *bp)
964 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
967 * Delete from old vnode list, if on one.
970 lwkt_gettoken(&vp->v_token);
971 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) {
972 if (bp->b_flags & B_VNDIRTY)
973 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
975 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
976 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN);
978 if (bp->b_flags & B_HASHED) {
979 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp);
980 bp->b_flags &= ~B_HASHED;
984 * Only remove from synclist when no dirty buffers are left AND
985 * the VFS has not flagged the vnode's inode as being dirty.
987 if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) == VONWORKLST &&
988 RB_EMPTY(&vp->v_rbdirty_tree)) {
989 vn_syncer_remove(vp, 0);
993 lwkt_reltoken(&vp->v_token);
999 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI.
1000 * This routine is called when the state of the B_DELWRI bit is changed.
1002 * Must be called with vp->v_token held.
1006 reassignbuf(struct buf *bp)
1008 struct vnode *vp = bp->b_vp;
1011 ASSERT_LWKT_TOKEN_HELD(&vp->v_token);
1015 * B_PAGING flagged buffers cannot be reassigned because their vp
1016 * is not fully linked in.
1018 if (bp->b_flags & B_PAGING)
1019 panic("cannot reassign paging buffer");
1021 if (bp->b_flags & B_DELWRI) {
1023 * Move to the dirty list, add the vnode to the worklist
1025 if (bp->b_flags & B_VNCLEAN) {
1026 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
1027 bp->b_flags &= ~B_VNCLEAN;
1029 if ((bp->b_flags & B_VNDIRTY) == 0) {
1030 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) {
1031 panic("reassignbuf: dup lblk vp %p bp %p",
1034 bp->b_flags |= B_VNDIRTY;
1036 if ((vp->v_flag & VONWORKLST) == 0) {
1037 switch (vp->v_type) {
1044 vp->v_rdev->si_mountpoint != NULL) {
1052 vn_syncer_add(vp, delay);
1056 * Move to the clean list, remove the vnode from the worklist
1057 * if no dirty blocks remain.
1059 if (bp->b_flags & B_VNDIRTY) {
1060 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
1061 bp->b_flags &= ~B_VNDIRTY;
1063 if ((bp->b_flags & B_VNCLEAN) == 0) {
1064 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) {
1065 panic("reassignbuf: dup lblk vp %p bp %p",
1068 bp->b_flags |= B_VNCLEAN;
1072 * Only remove from synclist when no dirty buffers are left
1073 * AND the VFS has not flagged the vnode's inode as being
1076 if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) ==
1078 RB_EMPTY(&vp->v_rbdirty_tree)) {
1079 vn_syncer_remove(vp, 0);
1085 * Create a vnode for a block device. Used for mounting the root file
1088 * A vref()'d vnode is returned.
1090 extern struct vop_ops *devfs_vnode_dev_vops_p;
1092 bdevvp(cdev_t dev, struct vnode **vpp)
1102 error = getspecialvnode(VT_NON, NULL, &devfs_vnode_dev_vops_p,
1113 v_associate_rdev(vp, dev);
1114 vp->v_umajor = dev->si_umajor;
1115 vp->v_uminor = dev->si_uminor;
1122 v_associate_rdev(struct vnode *vp, cdev_t dev)
1126 if (dev_is_good(dev) == 0)
1128 KKASSERT(vp->v_rdev == NULL);
1129 vp->v_rdev = reference_dev(dev);
1130 lwkt_gettoken(&spechash_token);
1131 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext);
1132 lwkt_reltoken(&spechash_token);
1137 v_release_rdev(struct vnode *vp)
1141 if ((dev = vp->v_rdev) != NULL) {
1142 lwkt_gettoken(&spechash_token);
1143 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext);
1146 lwkt_reltoken(&spechash_token);
1151 * Add a vnode to the alias list hung off the cdev_t. We only associate
1152 * the device number with the vnode. The actual device is not associated
1153 * until the vnode is opened (usually in spec_open()), and will be
1154 * disassociated on last close.
1157 addaliasu(struct vnode *nvp, int x, int y)
1159 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1160 panic("addaliasu on non-special vnode");
1166 * Simple call that a filesystem can make to try to get rid of a
1167 * vnode. It will fail if anyone is referencing the vnode (including
1170 * The filesystem can check whether its in-memory inode structure still
1171 * references the vp on return.
1173 * May only be called if the vnode is in a known state (i.e. being prevented
1174 * from being deallocated by some other condition such as a vfs inode hold).
1177 vclean_unlocked(struct vnode *vp)
1180 if (VREFCNT(vp) <= 1)
1186 * Disassociate a vnode from its underlying filesystem.
1188 * The vnode must be VX locked and referenced. In all normal situations
1189 * there are no active references. If vclean_vxlocked() is called while
1190 * there are active references, the vnode is being ripped out and we have
1191 * to call VOP_CLOSE() as appropriate before we can reclaim it.
1194 vclean_vxlocked(struct vnode *vp, int flags)
1199 struct namecache *ncp;
1202 * If the vnode has already been reclaimed we have nothing to do.
1204 if (vp->v_flag & VRECLAIMED)
1208 * Set flag to interlock operation, flag finalization to ensure
1209 * that the vnode winds up on the inactive list, and set v_act to 0.
1211 vsetflags(vp, VRECLAIMED);
1212 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE);
1215 if (verbose_reclaims) {
1216 if ((ncp = TAILQ_FIRST(&vp->v_namecache)) != NULL)
1217 kprintf("Debug: reclaim %p %s\n", vp, ncp->nc_name);
1221 * Scrap the vfs cache
1223 while (cache_inval_vp(vp, 0) != 0) {
1224 kprintf("Warning: vnode %p clean/cache_resolution "
1225 "race detected\n", vp);
1226 tsleep(vp, 0, "vclninv", 2);
1230 * Check to see if the vnode is in use. If so we have to reference it
1231 * before we clean it out so that its count cannot fall to zero and
1232 * generate a race against ourselves to recycle it.
1234 active = (VREFCNT(vp) > 0);
1237 * Clean out any buffers associated with the vnode and destroy its
1238 * object, if it has one.
1240 vinvalbuf(vp, V_SAVE, 0, 0);
1241 KKASSERT(lockcountnb(&vp->v_lock) == 1);
1244 * If purging an active vnode (typically during a forced unmount
1245 * or reboot), it must be closed and deactivated before being
1246 * reclaimed. This isn't really all that safe, but what can
1249 * Note that neither of these routines unlocks the vnode.
1251 if (active && (flags & DOCLOSE)) {
1252 while ((n = vp->v_opencount) != 0) {
1253 if (vp->v_writecount)
1254 VOP_CLOSE(vp, FWRITE|FNONBLOCK, NULL);
1256 VOP_CLOSE(vp, FNONBLOCK, NULL);
1257 if (vp->v_opencount == n) {
1258 kprintf("Warning: unable to force-close"
1266 * If the vnode has not been deactivated, deactivated it. Deactivation
1267 * can create new buffers and VM pages so we have to call vinvalbuf()
1268 * again to make sure they all get flushed.
1270 * This can occur if a file with a link count of 0 needs to be
1273 * If the vnode is already dead don't try to deactivate it.
1275 if ((vp->v_flag & VINACTIVE) == 0) {
1276 vsetflags(vp, VINACTIVE);
1279 vinvalbuf(vp, V_SAVE, 0, 0);
1281 KKASSERT(lockcountnb(&vp->v_lock) == 1);
1284 * If the vnode has an object, destroy it.
1286 while ((object = vp->v_object) != NULL) {
1287 vm_object_hold(object);
1288 if (object == vp->v_object)
1290 vm_object_drop(object);
1293 if (object != NULL) {
1294 if (object->ref_count == 0) {
1295 if ((object->flags & OBJ_DEAD) == 0)
1296 vm_object_terminate(object);
1297 vm_object_drop(object);
1298 vclrflags(vp, VOBJBUF);
1300 vm_pager_deallocate(object);
1301 vclrflags(vp, VOBJBUF);
1302 vm_object_drop(object);
1305 KKASSERT((vp->v_flag & VOBJBUF) == 0);
1308 * Reclaim the vnode if not already dead.
1310 if (vp->v_mount && VOP_RECLAIM(vp))
1311 panic("vclean: cannot reclaim");
1314 * Done with purge, notify sleepers of the grim news.
1316 vp->v_ops = &dead_vnode_vops_p;
1321 * If we are destroying an active vnode, reactivate it now that
1322 * we have reassociated it with deadfs. This prevents the system
1323 * from crashing on the vnode due to it being unexpectedly marked
1324 * as inactive or reclaimed.
1326 if (active && (flags & DOCLOSE)) {
1327 vclrflags(vp, VINACTIVE | VRECLAIMED);
1332 * Eliminate all activity associated with the requested vnode
1333 * and with all vnodes aliased to the requested vnode.
1335 * The vnode must be referenced but should not be locked.
1338 vrevoke(struct vnode *vp, struct ucred *cred)
1346 * If the vnode has a device association, scrap all vnodes associated
1347 * with the device. Don't let the device disappear on us while we
1348 * are scrapping the vnodes.
1350 * The passed vp will probably show up in the list, do not VX lock
1353 * Releasing the vnode's rdev here can mess up specfs's call to
1354 * device close, so don't do it. The vnode has been disassociated
1355 * and the device will be closed after the last ref on the related
1356 * fp goes away (if not still open by e.g. the kernel).
1358 if (vp->v_type != VCHR) {
1359 error = fdrevoke(vp, DTYPE_VNODE, cred);
1362 if ((dev = vp->v_rdev) == NULL) {
1366 lwkt_gettoken(&spechash_token);
1369 vqn = SLIST_FIRST(&dev->si_hlist);
1372 while ((vq = vqn) != NULL) {
1373 if (VREFCNT(vq) > 0) {
1375 fdrevoke(vq, DTYPE_VNODE, cred);
1376 /*v_release_rdev(vq);*/
1378 if (vq->v_rdev != dev) {
1383 vqn = SLIST_NEXT(vq, v_cdevnext);
1388 lwkt_reltoken(&spechash_token);
1395 * This is called when the object underlying a vnode is being destroyed,
1396 * such as in a remove(). Try to recycle the vnode immediately if the
1397 * only active reference is our reference.
1399 * Directory vnodes in the namecache with children cannot be immediately
1400 * recycled because numerous VOP_N*() ops require them to be stable.
1402 * To avoid recursive recycling from VOP_INACTIVE implemenetations this
1403 * function is a NOP if VRECLAIMED is already set.
1406 vrecycle(struct vnode *vp)
1408 if (VREFCNT(vp) <= 1 && (vp->v_flag & VRECLAIMED) == 0) {
1409 if (cache_inval_vp_nonblock(vp))
1418 * Return the maximum I/O size allowed for strategy calls on VP.
1420 * If vp is VCHR or VBLK we dive the device, otherwise we use
1421 * the vp's mount info.
1423 * The returned value is clamped at MAXPHYS as most callers cannot use
1424 * buffers larger than that size.
1427 vmaxiosize(struct vnode *vp)
1431 if (vp->v_type == VBLK || vp->v_type == VCHR)
1432 maxiosize = vp->v_rdev->si_iosize_max;
1434 maxiosize = vp->v_mount->mnt_iosize_max;
1436 if (maxiosize > MAXPHYS)
1437 maxiosize = MAXPHYS;
1442 * Eliminate all activity associated with a vnode in preparation for
1445 * The vnode must be VX locked and refd and will remain VX locked and refd
1446 * on return. This routine may be called with the vnode in any state, as
1447 * long as it is VX locked. The vnode will be cleaned out and marked
1448 * VRECLAIMED but will not actually be reused until all existing refs and
1451 * NOTE: This routine may be called on a vnode which has not yet been
1452 * already been deactivated (VOP_INACTIVE), or on a vnode which has
1453 * already been reclaimed.
1455 * This routine is not responsible for placing us back on the freelist.
1456 * Instead, it happens automatically when the caller releases the VX lock
1457 * (assuming there aren't any other references).
1460 vgone_vxlocked(struct vnode *vp)
1463 * assert that the VX lock is held. This is an absolute requirement
1464 * now for vgone_vxlocked() to be called.
1466 KKASSERT(lockcountnb(&vp->v_lock) == 1);
1469 * Clean out the filesystem specific data and set the VRECLAIMED
1470 * bit. Also deactivate the vnode if necessary.
1472 * The vnode should have automatically been removed from the syncer
1473 * list as syncer/dirty flags cleared during the cleaning.
1475 vclean_vxlocked(vp, DOCLOSE);
1478 * Normally panic if the vnode is still dirty, unless we are doing
1479 * a forced unmount (tmpfs typically).
1481 if (vp->v_flag & VONWORKLST) {
1482 if (vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF) {
1484 vn_syncer_remove(vp, 1);
1486 panic("vp %p still dirty in vgone after flush", vp);
1491 * Delete from old mount point vnode list, if on one.
1493 if (vp->v_mount != NULL) {
1494 KKASSERT(vp->v_data == NULL);
1495 insmntque(vp, NULL);
1499 * If special device, remove it from special device alias list
1500 * if it is on one. This should normally only occur if a vnode is
1501 * being revoked as the device should otherwise have been released
1504 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
1515 * Lookup a vnode by device number.
1517 * Returns non-zero and *vpp set to a vref'd vnode on success.
1518 * Returns zero on failure.
1521 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp)
1525 lwkt_gettoken(&spechash_token);
1526 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1527 if (type == vp->v_type) {
1530 lwkt_reltoken(&spechash_token);
1534 lwkt_reltoken(&spechash_token);
1539 * Calculate the total number of references to a special device. This
1540 * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1541 * an overloaded field. Since udev2dev can now return NULL, we have
1542 * to check for a NULL v_rdev.
1545 count_dev(cdev_t dev)
1550 if (SLIST_FIRST(&dev->si_hlist)) {
1551 lwkt_gettoken(&spechash_token);
1552 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1553 count += vp->v_opencount;
1555 lwkt_reltoken(&spechash_token);
1561 vcount(struct vnode *vp)
1563 if (vp->v_rdev == NULL)
1565 return(count_dev(vp->v_rdev));
1569 * Initialize VMIO for a vnode. This routine MUST be called before a
1570 * VFS can issue buffer cache ops on a vnode. It is typically called
1571 * when a vnode is initialized from its inode.
1574 vinitvmio(struct vnode *vp, off_t filesize, int blksize, int boff)
1579 object = vp->v_object;
1581 vm_object_hold(object);
1582 KKASSERT(vp->v_object == object);
1585 if (object == NULL) {
1586 object = vnode_pager_alloc(vp, filesize, 0, 0, blksize, boff);
1589 * Dereference the reference we just created. This assumes
1590 * that the object is associated with the vp. Allow it to
1591 * have zero refs. It cannot be destroyed as long as it
1592 * is associated with the vnode.
1594 vm_object_hold(object);
1595 atomic_add_int(&object->ref_count, -1);
1598 KKASSERT((object->flags & OBJ_DEAD) == 0);
1600 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object"));
1601 vsetflags(vp, VOBJBUF);
1602 vm_object_drop(object);
1609 * Print out a description of a vnode.
1611 static char *typename[] =
1612 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1615 vprint(char *label, struct vnode *vp)
1620 kprintf("%s: %p: ", label, (void *)vp);
1622 kprintf("%p: ", (void *)vp);
1623 kprintf("type %s, refcnt %08x, writecount %d, holdcnt %d,",
1624 typename[vp->v_type],
1625 vp->v_refcnt, vp->v_writecount, vp->v_auxrefs);
1627 if (vp->v_flag & VROOT)
1628 strcat(buf, "|VROOT");
1629 if (vp->v_flag & VPFSROOT)
1630 strcat(buf, "|VPFSROOT");
1631 if (vp->v_flag & VTEXT)
1632 strcat(buf, "|VTEXT");
1633 if (vp->v_flag & VSYSTEM)
1634 strcat(buf, "|VSYSTEM");
1635 if (vp->v_flag & VOBJBUF)
1636 strcat(buf, "|VOBJBUF");
1638 kprintf(" flags (%s)", &buf[1]);
1639 if (vp->v_data == NULL) {
1648 * Do the usual access checking.
1649 * file_mode, uid and gid are from the vnode in question,
1650 * while acc_mode and cred are from the VOP_ACCESS parameter list
1653 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
1654 mode_t acc_mode, struct ucred *cred)
1660 * Super-user always gets read/write access, but execute access depends
1661 * on at least one execute bit being set.
1663 if (priv_check_cred(cred, PRIV_ROOT, 0) == 0) {
1664 if ((acc_mode & VEXEC) && type != VDIR &&
1665 (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)
1672 /* Otherwise, check the owner. */
1673 if (cred->cr_uid == uid) {
1674 if (acc_mode & VEXEC)
1676 if (acc_mode & VREAD)
1678 if (acc_mode & VWRITE)
1680 return ((file_mode & mask) == mask ? 0 : EACCES);
1683 /* Otherwise, check the groups. */
1684 ismember = groupmember(gid, cred);
1685 if (cred->cr_svgid == gid || ismember) {
1686 if (acc_mode & VEXEC)
1688 if (acc_mode & VREAD)
1690 if (acc_mode & VWRITE)
1692 return ((file_mode & mask) == mask ? 0 : EACCES);
1695 /* Otherwise, check everyone else. */
1696 if (acc_mode & VEXEC)
1698 if (acc_mode & VREAD)
1700 if (acc_mode & VWRITE)
1702 return ((file_mode & mask) == mask ? 0 : EACCES);
1706 #include <ddb/ddb.h>
1708 static int db_show_locked_vnodes(struct mount *mp, void *data);
1711 * List all of the locked vnodes in the system.
1712 * Called when debugging the kernel.
1714 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
1716 kprintf("Locked vnodes\n");
1717 mountlist_scan(db_show_locked_vnodes, NULL,
1718 MNTSCAN_FORWARD|MNTSCAN_NOBUSY);
1722 db_show_locked_vnodes(struct mount *mp, void *data __unused)
1726 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
1727 if (vn_islocked(vp))
1735 * Top level filesystem related information gathering.
1737 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
1740 vfs_sysctl(SYSCTL_HANDLER_ARGS)
1742 int *name = (int *)arg1 - 1; /* XXX */
1743 u_int namelen = arg2 + 1; /* XXX */
1744 struct vfsconf *vfsp;
1747 #if 1 || defined(COMPAT_PRELITE2)
1748 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1750 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
1754 /* all sysctl names at this level are at least name and field */
1756 return (ENOTDIR); /* overloaded */
1757 if (name[0] != VFS_GENERIC) {
1758 vfsp = vfsconf_find_by_typenum(name[0]);
1760 return (EOPNOTSUPP);
1761 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1762 oldp, oldlenp, newp, newlen, p));
1766 case VFS_MAXTYPENUM:
1769 maxtypenum = vfsconf_get_maxtypenum();
1770 return (SYSCTL_OUT(req, &maxtypenum, sizeof(maxtypenum)));
1773 return (ENOTDIR); /* overloaded */
1774 vfsp = vfsconf_find_by_typenum(name[2]);
1776 return (EOPNOTSUPP);
1777 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
1779 return (EOPNOTSUPP);
1782 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
1783 "Generic filesystem");
1785 #if 1 || defined(COMPAT_PRELITE2)
1788 sysctl_ovfs_conf_iter(struct vfsconf *vfsp, void *data)
1791 struct ovfsconf ovfs;
1792 struct sysctl_req *req = (struct sysctl_req*) data;
1794 bzero(&ovfs, sizeof(ovfs));
1795 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
1796 strcpy(ovfs.vfc_name, vfsp->vfc_name);
1797 ovfs.vfc_index = vfsp->vfc_typenum;
1798 ovfs.vfc_refcount = vfsp->vfc_refcount;
1799 ovfs.vfc_flags = vfsp->vfc_flags;
1800 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
1802 return error; /* abort iteration with error code */
1804 return 0; /* continue iterating with next element */
1808 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
1810 return vfsconf_each(sysctl_ovfs_conf_iter, (void*)req);
1813 #endif /* 1 || COMPAT_PRELITE2 */
1816 * Check to see if a filesystem is mounted on a block device.
1819 vfs_mountedon(struct vnode *vp)
1823 if ((dev = vp->v_rdev) == NULL) {
1824 /* if (vp->v_type != VBLK)
1825 dev = get_dev(vp->v_uminor, vp->v_umajor); */
1827 if (dev != NULL && dev->si_mountpoint)
1833 * Unmount all filesystems. The list is traversed in reverse order
1834 * of mounting to avoid dependencies.
1837 static int vfs_umountall_callback(struct mount *mp, void *data);
1840 vfs_unmountall(void)
1845 count = mountlist_scan(vfs_umountall_callback,
1846 NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY);
1852 vfs_umountall_callback(struct mount *mp, void *data)
1856 error = dounmount(mp, MNT_FORCE);
1858 mountlist_remove(mp);
1859 kprintf("unmount of filesystem mounted from %s failed (",
1860 mp->mnt_stat.f_mntfromname);
1864 kprintf("%d)\n", error);
1870 * Checks the mount flags for parameter mp and put the names comma-separated
1871 * into a string buffer buf with a size limit specified by len.
1873 * It returns the number of bytes written into buf, and (*errorp) will be
1874 * set to 0, EINVAL (if passed length is 0), or ENOSPC (supplied buffer was
1875 * not large enough). The buffer will be 0-terminated if len was not 0.
1878 vfs_flagstostr(int flags, const struct mountctl_opt *optp,
1879 char *buf, size_t len, int *errorp)
1881 static const struct mountctl_opt optnames[] = {
1882 { MNT_RDONLY, "read-only" },
1883 { MNT_SYNCHRONOUS, "synchronous" },
1884 { MNT_NOEXEC, "noexec" },
1885 { MNT_NOSUID, "nosuid" },
1886 { MNT_NODEV, "nodev" },
1887 { MNT_AUTOMOUNTED, "automounted" },
1888 { MNT_ASYNC, "asynchronous" },
1889 { MNT_SUIDDIR, "suiddir" },
1890 { MNT_SOFTDEP, "soft-updates" },
1891 { MNT_NOSYMFOLLOW, "nosymfollow" },
1892 { MNT_TRIM, "trim" },
1893 { MNT_NOATIME, "noatime" },
1894 { MNT_NOCLUSTERR, "noclusterr" },
1895 { MNT_NOCLUSTERW, "noclusterw" },
1896 { MNT_EXRDONLY, "NFS read-only" },
1897 { MNT_EXPORTED, "NFS exported" },
1898 /* Remaining NFS flags could come here */
1899 { MNT_LOCAL, "local" },
1900 { MNT_QUOTA, "with-quotas" },
1901 /* { MNT_ROOTFS, "rootfs" }, */
1902 /* { MNT_IGNORE, "ignore" }, */
1912 bleft = len - 1; /* leave room for trailing \0 */
1915 * Checks the size of the string. If it contains
1916 * any data, then we will append the new flags to
1919 actsize = strlen(buf);
1923 /* Default flags if no flags passed */
1927 if (bleft < 0) { /* degenerate case, 0-length buffer */
1932 for (; flags && optp->o_opt; ++optp) {
1933 if ((flags & optp->o_opt) == 0)
1935 optlen = strlen(optp->o_name);
1936 if (bwritten || actsize > 0) {
1941 buf[bwritten++] = ',';
1942 buf[bwritten++] = ' ';
1945 if (bleft < optlen) {
1949 bcopy(optp->o_name, buf + bwritten, optlen);
1952 flags &= ~optp->o_opt;
1956 * Space already reserved for trailing \0
1963 * Build hash lists of net addresses and hang them off the mount point.
1964 * Called by ufs_mount() to set up the lists of export addresses.
1967 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1968 const struct export_args *argp)
1971 struct radix_node_head *rnh;
1973 struct radix_node *rn;
1974 struct sockaddr *saddr, *smask = NULL;
1977 if (argp->ex_addrlen == 0) {
1978 if (mp->mnt_flag & MNT_DEFEXPORTED)
1980 np = &nep->ne_defexported;
1981 np->netc_exflags = argp->ex_flags;
1982 np->netc_anon = argp->ex_anon;
1983 np->netc_anon.cr_ref = 1;
1984 mp->mnt_flag |= MNT_DEFEXPORTED;
1988 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
1990 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
1993 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1994 np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK | M_ZERO);
1995 saddr = (struct sockaddr *) (np + 1);
1996 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
1998 if (saddr->sa_len > argp->ex_addrlen)
1999 saddr->sa_len = argp->ex_addrlen;
2000 if (argp->ex_masklen) {
2001 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
2002 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
2005 if (smask->sa_len > argp->ex_masklen)
2006 smask->sa_len = argp->ex_masklen;
2009 if (nep->ne_maskhead == NULL) {
2010 if (!rn_inithead((void **)&nep->ne_maskhead, NULL, 0)) {
2015 if((rnh = vfs_create_addrlist_af(saddr->sa_family, nep)) == NULL) {
2019 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh,
2022 if (rn == NULL || np != (struct netcred *) rn) { /* already exists */
2026 np->netc_exflags = argp->ex_flags;
2027 np->netc_anon = argp->ex_anon;
2028 np->netc_anon.cr_ref = 1;
2031 kfree(np, M_NETADDR);
2037 vfs_free_netcred(struct radix_node *rn, void *w)
2039 struct radix_node_head *rnh = (struct radix_node_head *) w;
2041 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2042 kfree((caddr_t) rn, M_NETADDR);
2046 static struct radix_node_head *
2047 vfs_create_addrlist_af(int af, struct netexport *nep)
2049 struct radix_node_head *rnh = NULL;
2050 #if defined(INET) || defined(INET6)
2051 struct radix_node_head *maskhead = nep->ne_maskhead;
2055 NE_ASSERT_LOCKED(nep);
2056 KKASSERT(maskhead != NULL);
2060 if ((rnh = nep->ne_inethead) == NULL) {
2061 off = offsetof(struct sockaddr_in, sin_addr) << 3;
2062 if (!rn_inithead((void **)&rnh, maskhead, off))
2064 nep->ne_inethead = rnh;
2070 if ((rnh = nep->ne_inet6head) == NULL) {
2071 off = offsetof(struct sockaddr_in6, sin6_addr) << 3;
2072 if (!rn_inithead((void **)&rnh, maskhead, off))
2074 nep->ne_inet6head = rnh;
2083 vfs_free_addrlist_af(struct radix_node_head **prnh)
2085 struct radix_node_head *rnh = *prnh;
2087 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, rnh);
2088 kfree(rnh, M_RTABLE);
2093 * Free the net address hash lists that are hanging off the mount points.
2096 vfs_free_addrlist(struct netexport *nep)
2099 if (nep->ne_inethead != NULL)
2100 vfs_free_addrlist_af(&nep->ne_inethead);
2101 if (nep->ne_inet6head != NULL)
2102 vfs_free_addrlist_af(&nep->ne_inet6head);
2103 if (nep->ne_maskhead)
2104 vfs_free_addrlist_af(&nep->ne_maskhead);
2109 vfs_export(struct mount *mp, struct netexport *nep,
2110 const struct export_args *argp)
2114 if (argp->ex_flags & MNT_DELEXPORT) {
2115 if (mp->mnt_flag & MNT_EXPUBLIC) {
2116 vfs_setpublicfs(NULL, NULL, NULL);
2117 mp->mnt_flag &= ~MNT_EXPUBLIC;
2119 vfs_free_addrlist(nep);
2120 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2122 if (argp->ex_flags & MNT_EXPORTED) {
2123 if (argp->ex_flags & MNT_EXPUBLIC) {
2124 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2126 mp->mnt_flag |= MNT_EXPUBLIC;
2128 if ((error = vfs_hang_addrlist(mp, nep, argp)))
2130 mp->mnt_flag |= MNT_EXPORTED;
2137 * Set the publicly exported filesystem (WebNFS). Currently, only
2138 * one public filesystem is possible in the spec (RFC 2054 and 2055)
2141 vfs_setpublicfs(struct mount *mp, struct netexport *nep,
2142 const struct export_args *argp)
2149 * mp == NULL -> invalidate the current info, the FS is
2150 * no longer exported. May be called from either vfs_export
2151 * or unmount, so check if it hasn't already been done.
2154 if (nfs_pub.np_valid) {
2155 nfs_pub.np_valid = 0;
2156 if (nfs_pub.np_index != NULL) {
2157 kfree(nfs_pub.np_index, M_TEMP);
2158 nfs_pub.np_index = NULL;
2165 * Only one allowed at a time.
2167 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2171 * Get real filehandle for root of exported FS.
2173 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2174 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2176 if ((error = VFS_ROOT(mp, &rvp)))
2179 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2185 * If an indexfile was specified, pull it in.
2187 if (argp->ex_indexfile != NULL) {
2190 error = vn_get_namelen(rvp, &namelen);
2193 nfs_pub.np_index = kmalloc(namelen, M_TEMP, M_WAITOK);
2194 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2198 * Check for illegal filenames.
2200 for (cp = nfs_pub.np_index; *cp; cp++) {
2208 kfree(nfs_pub.np_index, M_TEMP);
2213 nfs_pub.np_mount = mp;
2214 nfs_pub.np_valid = 1;
2219 vfs_export_lookup(struct mount *mp, struct netexport *nep,
2220 struct sockaddr *nam)
2223 struct radix_node_head *rnh;
2224 struct sockaddr *saddr;
2227 if (mp->mnt_flag & MNT_EXPORTED) {
2229 * Lookup in the export list first.
2234 switch (saddr->sa_family) {
2237 rnh = nep->ne_inethead;
2242 rnh = nep->ne_inet6head;
2249 np = (struct netcred *)
2250 (*rnh->rnh_matchaddr)((char *)saddr,
2252 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2258 * If no address match, use the default if it exists.
2260 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2261 np = &nep->ne_defexported;
2267 * perform msync on all vnodes under a mount point. The mount point must
2268 * be locked. This code is also responsible for lazy-freeing unreferenced
2269 * vnodes whos VM objects no longer contain pages.
2271 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
2273 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked,
2274 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it
2275 * way up in this high level function.
2277 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
2278 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data);
2281 vfs_msync(struct mount *mp, int flags)
2286 * tmpfs sets this flag to prevent msync(), sync, and the
2287 * filesystem periodic syncer from trying to flush VM pages
2288 * to swap. Only pure memory pressure flushes tmpfs VM pages
2291 if (mp->mnt_kern_flag & MNTK_NOMSYNC)
2295 * Ok, scan the vnodes for work. If the filesystem is using the
2296 * syncer thread feature we can use vsyncscan() instead of
2297 * vmntvnodescan(), which is much faster.
2299 vmsc_flags = VMSC_GETVP;
2300 if (flags != MNT_WAIT)
2301 vmsc_flags |= VMSC_NOWAIT;
2303 if (mp->mnt_kern_flag & MNTK_THR_SYNC) {
2304 vsyncscan(mp, vmsc_flags, vfs_msync_scan2,
2305 (void *)(intptr_t)flags);
2307 vmntvnodescan(mp, vmsc_flags,
2308 vfs_msync_scan1, vfs_msync_scan2,
2309 (void *)(intptr_t)flags);
2314 * scan1 is a fast pre-check. There could be hundreds of thousands of
2315 * vnodes, we cannot afford to do anything heavy weight until we have a
2316 * fairly good indication that there is work to do.
2320 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
2322 int flags = (int)(intptr_t)data;
2324 if ((vp->v_flag & VRECLAIMED) == 0) {
2325 if (vp->v_auxrefs == 0 && VREFCNT(vp) <= 0 &&
2327 return(0); /* call scan2 */
2329 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
2330 (vp->v_flag & VOBJDIRTY) &&
2331 (flags == MNT_WAIT || vn_islocked(vp) == 0)) {
2332 return(0); /* call scan2 */
2337 * do not call scan2, continue the loop
2343 * This callback is handed a locked vnode.
2347 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data)
2350 int flags = (int)(intptr_t)data;
2352 if (vp->v_flag & VRECLAIMED)
2355 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) {
2356 if ((obj = vp->v_object) != NULL) {
2357 vm_object_page_clean(obj, 0, 0,
2358 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2365 * Wake up anyone interested in vp because it is being revoked.
2368 vn_gone(struct vnode *vp)
2370 lwkt_gettoken(&vp->v_token);
2371 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, NOTE_REVOKE);
2372 lwkt_reltoken(&vp->v_token);
2376 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened
2377 * (or v_rdev might be NULL).
2380 vn_todev(struct vnode *vp)
2382 if (vp->v_type != VBLK && vp->v_type != VCHR)
2384 KKASSERT(vp->v_rdev != NULL);
2385 return (vp->v_rdev);
2389 * Check if vnode represents a disk device. The vnode does not need to be
2395 vn_isdisk(struct vnode *vp, int *errp)
2399 if (vp->v_type != VCHR) {
2412 if (dev_is_good(dev) == 0) {
2417 if ((dev_dflags(dev) & D_DISK) == 0) {
2428 vn_get_namelen(struct vnode *vp, int *namelen)
2431 register_t retval[2];
2433 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval);
2436 *namelen = (int)retval[0];
2441 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type,
2442 uint16_t d_namlen, const char *d_name)
2447 len = _DIRENT_RECLEN(d_namlen);
2448 if (len > uio->uio_resid)
2451 dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO);
2454 dp->d_namlen = d_namlen;
2455 dp->d_type = d_type;
2456 bcopy(d_name, dp->d_name, d_namlen);
2458 *error = uiomove((caddr_t)dp, len, uio);
2466 vn_mark_atime(struct vnode *vp, struct thread *td)
2468 struct proc *p = td->td_proc;
2469 struct ucred *cred = p ? p->p_ucred : proc0.p_ucred;
2471 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
2472 VOP_MARKATIME(vp, cred);