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);
632 int (*checkdef)(struct buf *);
633 int (*cmpfunc)(struct buf *, void *);
637 vfsync(struct vnode *vp, int waitfor, int passes,
638 int (*checkdef)(struct buf *),
639 int (*waitoutput)(struct vnode *, struct thread *))
641 struct vfsync_info info;
644 bzero(&info, sizeof(info));
646 if ((info.checkdef = checkdef) == NULL)
649 lwkt_gettoken(&vp->v_token);
652 case MNT_LAZY | MNT_NOWAIT:
655 * Lazy (filesystem syncer typ) Asynchronous plus limit the
656 * number of data (not meta) pages we try to flush to 1MB.
657 * A non-zero return means that lazy limit was reached.
659 info.lazylimit = 1024 * 1024;
661 info.cmpfunc = vfsync_lazy_range_cmp;
662 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
663 vfsync_lazy_range_cmp, vfsync_bp, &info);
664 info.cmpfunc = vfsync_meta_only_cmp;
665 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
666 vfsync_meta_only_cmp, vfsync_bp, &info);
669 else if (!RB_EMPTY(&vp->v_rbdirty_tree))
670 vn_syncer_add(vp, 1);
675 * Asynchronous. Do a data-only pass and a meta-only pass.
678 info.cmpfunc = vfsync_data_only_cmp;
679 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
681 info.cmpfunc = vfsync_meta_only_cmp;
682 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp,
688 * Synchronous. Do a data-only pass, then a meta-data+data
689 * pass, then additional integrated passes to try to get
690 * all the dependancies flushed.
692 info.cmpfunc = vfsync_data_only_cmp;
693 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
695 error = vfsync_wait_output(vp, waitoutput);
697 info.skippedbufs = 0;
698 info.cmpfunc = vfsync_dummy_cmp;
699 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
701 error = vfsync_wait_output(vp, waitoutput);
702 if (info.skippedbufs) {
703 kprintf("Warning: vfsync skipped %d dirty "
704 "bufs in pass2!\n", info.skippedbufs);
707 while (error == 0 && passes > 0 &&
708 !RB_EMPTY(&vp->v_rbdirty_tree)
711 info.synchronous = 1;
714 info.cmpfunc = vfsync_dummy_cmp;
715 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
721 error = vfsync_wait_output(vp, waitoutput);
725 lwkt_reltoken(&vp->v_token);
730 vfsync_wait_output(struct vnode *vp,
731 int (*waitoutput)(struct vnode *, struct thread *))
735 error = bio_track_wait(&vp->v_track_write, 0, 0);
737 error = waitoutput(vp, curthread);
742 vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused)
748 vfsync_data_only_cmp(struct buf *bp, void *data)
750 if (bp->b_loffset < 0)
756 vfsync_meta_only_cmp(struct buf *bp, void *data)
758 if (bp->b_loffset < 0)
764 vfsync_lazy_range_cmp(struct buf *bp, void *data)
766 struct vfsync_info *info = data;
768 if (bp->b_loffset < info->vp->v_lazyw)
774 vfsync_bp(struct buf *bp, void *data)
776 struct vfsync_info *info = data;
777 struct vnode *vp = info->vp;
781 * Ignore buffers that we cannot immediately lock.
783 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
789 * We must revalidate the buffer after locking.
791 if ((bp->b_flags & B_DELWRI) == 0 ||
792 bp->b_vp != info->vp ||
793 info->cmpfunc(bp, data)) {
799 * If syncdeps is not set we do not try to write buffers which have
802 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) {
808 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
809 * has been written but an additional handshake with the device
810 * is required before we can dispose of the buffer. We have no idea
811 * how to do this so we have to skip these buffers.
813 if (bp->b_flags & B_NEEDCOMMIT) {
819 * Ask bioops if it is ok to sync. If not the VFS may have
820 * set B_LOCKED so we have to cycle the buffer.
822 if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) {
828 if (info->synchronous) {
830 * Synchronous flushing. An error may be returned.
836 * Asynchronous flushing. A negative return value simply
837 * stops the scan and is not considered an error. We use
838 * this to support limited MNT_LAZY flushes.
840 vp->v_lazyw = bp->b_loffset;
842 info->lazycount += cluster_awrite(bp);
843 waitrunningbufspace();
845 if (info->lazylimit && info->lazycount >= info->lazylimit)
854 * Associate a buffer with a vnode.
859 bgetvp(struct vnode *vp, struct buf *bp, int testsize)
861 KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
862 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0);
865 * Insert onto list for new vnode.
867 lwkt_gettoken(&vp->v_token);
869 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) {
870 lwkt_reltoken(&vp->v_token);
875 * Diagnostics (mainly for HAMMER debugging). Check for
876 * overlapping buffers.
878 if (check_buf_overlap) {
880 bx = buf_rb_hash_RB_PREV(bp);
882 if (bx->b_loffset + bx->b_bufsize > bp->b_loffset) {
883 kprintf("bgetvp: overlapl %016jx/%d %016jx "
885 (intmax_t)bx->b_loffset,
887 (intmax_t)bp->b_loffset,
889 if (check_buf_overlap > 1)
890 panic("bgetvp - overlapping buffer");
893 bx = buf_rb_hash_RB_NEXT(bp);
895 if (bp->b_loffset + testsize > bx->b_loffset) {
896 kprintf("bgetvp: overlapr %016jx/%d %016jx "
898 (intmax_t)bp->b_loffset,
900 (intmax_t)bx->b_loffset,
902 if (check_buf_overlap > 1)
903 panic("bgetvp - overlapping buffer");
908 bp->b_flags |= B_HASHED;
909 bp->b_flags |= B_VNCLEAN;
910 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp))
911 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp);
913 lwkt_reltoken(&vp->v_token);
918 * Disassociate a buffer from a vnode.
923 brelvp(struct buf *bp)
927 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
930 * Delete from old vnode list, if on one.
933 lwkt_gettoken(&vp->v_token);
934 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) {
935 if (bp->b_flags & B_VNDIRTY)
936 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
938 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
939 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN);
941 if (bp->b_flags & B_HASHED) {
942 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp);
943 bp->b_flags &= ~B_HASHED;
947 * Only remove from synclist when no dirty buffers are left AND
948 * the VFS has not flagged the vnode's inode as being dirty.
950 if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) == VONWORKLST &&
951 RB_EMPTY(&vp->v_rbdirty_tree)) {
952 vn_syncer_remove(vp);
956 lwkt_reltoken(&vp->v_token);
962 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI.
963 * This routine is called when the state of the B_DELWRI bit is changed.
965 * Must be called with vp->v_token held.
969 reassignbuf(struct buf *bp)
971 struct vnode *vp = bp->b_vp;
974 ASSERT_LWKT_TOKEN_HELD(&vp->v_token);
978 * B_PAGING flagged buffers cannot be reassigned because their vp
979 * is not fully linked in.
981 if (bp->b_flags & B_PAGING)
982 panic("cannot reassign paging buffer");
984 if (bp->b_flags & B_DELWRI) {
986 * Move to the dirty list, add the vnode to the worklist
988 if (bp->b_flags & B_VNCLEAN) {
989 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
990 bp->b_flags &= ~B_VNCLEAN;
992 if ((bp->b_flags & B_VNDIRTY) == 0) {
993 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) {
994 panic("reassignbuf: dup lblk vp %p bp %p",
997 bp->b_flags |= B_VNDIRTY;
999 if ((vp->v_flag & VONWORKLST) == 0) {
1000 switch (vp->v_type) {
1007 vp->v_rdev->si_mountpoint != NULL) {
1015 vn_syncer_add(vp, delay);
1019 * Move to the clean list, remove the vnode from the worklist
1020 * if no dirty blocks remain.
1022 if (bp->b_flags & B_VNDIRTY) {
1023 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
1024 bp->b_flags &= ~B_VNDIRTY;
1026 if ((bp->b_flags & B_VNCLEAN) == 0) {
1027 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) {
1028 panic("reassignbuf: dup lblk vp %p bp %p",
1031 bp->b_flags |= B_VNCLEAN;
1035 * Only remove from synclist when no dirty buffers are left
1036 * AND the VFS has not flagged the vnode's inode as being
1039 if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) ==
1041 RB_EMPTY(&vp->v_rbdirty_tree)) {
1042 vn_syncer_remove(vp);
1048 * Create a vnode for a block device. Used for mounting the root file
1051 * A vref()'d vnode is returned.
1053 extern struct vop_ops *devfs_vnode_dev_vops_p;
1055 bdevvp(cdev_t dev, struct vnode **vpp)
1065 error = getspecialvnode(VT_NON, NULL, &devfs_vnode_dev_vops_p,
1076 v_associate_rdev(vp, dev);
1077 vp->v_umajor = dev->si_umajor;
1078 vp->v_uminor = dev->si_uminor;
1085 v_associate_rdev(struct vnode *vp, cdev_t dev)
1089 if (dev_is_good(dev) == 0)
1091 KKASSERT(vp->v_rdev == NULL);
1092 vp->v_rdev = reference_dev(dev);
1093 lwkt_gettoken(&spechash_token);
1094 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext);
1095 lwkt_reltoken(&spechash_token);
1100 v_release_rdev(struct vnode *vp)
1104 if ((dev = vp->v_rdev) != NULL) {
1105 lwkt_gettoken(&spechash_token);
1106 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext);
1109 lwkt_reltoken(&spechash_token);
1114 * Add a vnode to the alias list hung off the cdev_t. We only associate
1115 * the device number with the vnode. The actual device is not associated
1116 * until the vnode is opened (usually in spec_open()), and will be
1117 * disassociated on last close.
1120 addaliasu(struct vnode *nvp, int x, int y)
1122 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1123 panic("addaliasu on non-special vnode");
1129 * Simple call that a filesystem can make to try to get rid of a
1130 * vnode. It will fail if anyone is referencing the vnode (including
1133 * The filesystem can check whether its in-memory inode structure still
1134 * references the vp on return.
1136 * May only be called if the vnode is in a known state (i.e. being prevented
1137 * from being deallocated by some other condition such as a vfs inode hold).
1140 vclean_unlocked(struct vnode *vp)
1143 if (VREFCNT(vp) <= 1)
1149 * Disassociate a vnode from its underlying filesystem.
1151 * The vnode must be VX locked and referenced. In all normal situations
1152 * there are no active references. If vclean_vxlocked() is called while
1153 * there are active references, the vnode is being ripped out and we have
1154 * to call VOP_CLOSE() as appropriate before we can reclaim it.
1157 vclean_vxlocked(struct vnode *vp, int flags)
1162 struct namecache *ncp;
1165 * If the vnode has already been reclaimed we have nothing to do.
1167 if (vp->v_flag & VRECLAIMED)
1171 * Set flag to interlock operation, flag finalization to ensure
1172 * that the vnode winds up on the inactive list, and set v_act to 0.
1174 vsetflags(vp, VRECLAIMED);
1175 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE);
1178 if (verbose_reclaims) {
1179 if ((ncp = TAILQ_FIRST(&vp->v_namecache)) != NULL)
1180 kprintf("Debug: reclaim %p %s\n", vp, ncp->nc_name);
1184 * Scrap the vfs cache
1186 while (cache_inval_vp(vp, 0) != 0) {
1187 kprintf("Warning: vnode %p clean/cache_resolution "
1188 "race detected\n", vp);
1189 tsleep(vp, 0, "vclninv", 2);
1193 * Check to see if the vnode is in use. If so we have to reference it
1194 * before we clean it out so that its count cannot fall to zero and
1195 * generate a race against ourselves to recycle it.
1197 active = (VREFCNT(vp) > 0);
1200 * Clean out any buffers associated with the vnode and destroy its
1201 * object, if it has one.
1203 vinvalbuf(vp, V_SAVE, 0, 0);
1204 KKASSERT(lockcountnb(&vp->v_lock) == 1);
1207 * If purging an active vnode (typically during a forced unmount
1208 * or reboot), it must be closed and deactivated before being
1209 * reclaimed. This isn't really all that safe, but what can
1212 * Note that neither of these routines unlocks the vnode.
1214 if (active && (flags & DOCLOSE)) {
1215 while ((n = vp->v_opencount) != 0) {
1216 if (vp->v_writecount)
1217 VOP_CLOSE(vp, FWRITE|FNONBLOCK, NULL);
1219 VOP_CLOSE(vp, FNONBLOCK, NULL);
1220 if (vp->v_opencount == n) {
1221 kprintf("Warning: unable to force-close"
1229 * If the vnode has not been deactivated, deactivated it. Deactivation
1230 * can create new buffers and VM pages so we have to call vinvalbuf()
1231 * again to make sure they all get flushed.
1233 * This can occur if a file with a link count of 0 needs to be
1236 * If the vnode is already dead don't try to deactivate it.
1238 if ((vp->v_flag & VINACTIVE) == 0) {
1239 vsetflags(vp, VINACTIVE);
1242 vinvalbuf(vp, V_SAVE, 0, 0);
1244 KKASSERT(lockcountnb(&vp->v_lock) == 1);
1247 * If the vnode has an object, destroy it.
1249 while ((object = vp->v_object) != NULL) {
1250 vm_object_hold(object);
1251 if (object == vp->v_object)
1253 vm_object_drop(object);
1256 if (object != NULL) {
1257 if (object->ref_count == 0) {
1258 if ((object->flags & OBJ_DEAD) == 0)
1259 vm_object_terminate(object);
1260 vm_object_drop(object);
1261 vclrflags(vp, VOBJBUF);
1263 vm_pager_deallocate(object);
1264 vclrflags(vp, VOBJBUF);
1265 vm_object_drop(object);
1268 KKASSERT((vp->v_flag & VOBJBUF) == 0);
1271 * Reclaim the vnode if not already dead.
1273 if (vp->v_mount && VOP_RECLAIM(vp))
1274 panic("vclean: cannot reclaim");
1277 * Done with purge, notify sleepers of the grim news.
1279 vp->v_ops = &dead_vnode_vops_p;
1284 * If we are destroying an active vnode, reactivate it now that
1285 * we have reassociated it with deadfs. This prevents the system
1286 * from crashing on the vnode due to it being unexpectedly marked
1287 * as inactive or reclaimed.
1289 if (active && (flags & DOCLOSE)) {
1290 vclrflags(vp, VINACTIVE | VRECLAIMED);
1295 * Eliminate all activity associated with the requested vnode
1296 * and with all vnodes aliased to the requested vnode.
1298 * The vnode must be referenced but should not be locked.
1301 vrevoke(struct vnode *vp, struct ucred *cred)
1309 * If the vnode has a device association, scrap all vnodes associated
1310 * with the device. Don't let the device disappear on us while we
1311 * are scrapping the vnodes.
1313 * The passed vp will probably show up in the list, do not VX lock
1316 * Releasing the vnode's rdev here can mess up specfs's call to
1317 * device close, so don't do it. The vnode has been disassociated
1318 * and the device will be closed after the last ref on the related
1319 * fp goes away (if not still open by e.g. the kernel).
1321 if (vp->v_type != VCHR) {
1322 error = fdrevoke(vp, DTYPE_VNODE, cred);
1325 if ((dev = vp->v_rdev) == NULL) {
1329 lwkt_gettoken(&spechash_token);
1332 vqn = SLIST_FIRST(&dev->si_hlist);
1335 while ((vq = vqn) != NULL) {
1336 if (VREFCNT(vq) > 0) {
1338 fdrevoke(vq, DTYPE_VNODE, cred);
1339 /*v_release_rdev(vq);*/
1341 if (vq->v_rdev != dev) {
1346 vqn = SLIST_NEXT(vq, v_cdevnext);
1351 lwkt_reltoken(&spechash_token);
1358 * This is called when the object underlying a vnode is being destroyed,
1359 * such as in a remove(). Try to recycle the vnode immediately if the
1360 * only active reference is our reference.
1362 * Directory vnodes in the namecache with children cannot be immediately
1363 * recycled because numerous VOP_N*() ops require them to be stable.
1365 * To avoid recursive recycling from VOP_INACTIVE implemenetations this
1366 * function is a NOP if VRECLAIMED is already set.
1369 vrecycle(struct vnode *vp)
1371 if (VREFCNT(vp) <= 1 && (vp->v_flag & VRECLAIMED) == 0) {
1372 if (cache_inval_vp_nonblock(vp))
1381 * Return the maximum I/O size allowed for strategy calls on VP.
1383 * If vp is VCHR or VBLK we dive the device, otherwise we use
1384 * the vp's mount info.
1386 * The returned value is clamped at MAXPHYS as most callers cannot use
1387 * buffers larger than that size.
1390 vmaxiosize(struct vnode *vp)
1394 if (vp->v_type == VBLK || vp->v_type == VCHR)
1395 maxiosize = vp->v_rdev->si_iosize_max;
1397 maxiosize = vp->v_mount->mnt_iosize_max;
1399 if (maxiosize > MAXPHYS)
1400 maxiosize = MAXPHYS;
1405 * Eliminate all activity associated with a vnode in preparation for
1408 * The vnode must be VX locked and refd and will remain VX locked and refd
1409 * on return. This routine may be called with the vnode in any state, as
1410 * long as it is VX locked. The vnode will be cleaned out and marked
1411 * VRECLAIMED but will not actually be reused until all existing refs and
1414 * NOTE: This routine may be called on a vnode which has not yet been
1415 * already been deactivated (VOP_INACTIVE), or on a vnode which has
1416 * already been reclaimed.
1418 * This routine is not responsible for placing us back on the freelist.
1419 * Instead, it happens automatically when the caller releases the VX lock
1420 * (assuming there aren't any other references).
1423 vgone_vxlocked(struct vnode *vp)
1426 * assert that the VX lock is held. This is an absolute requirement
1427 * now for vgone_vxlocked() to be called.
1429 KKASSERT(lockcountnb(&vp->v_lock) == 1);
1432 * Clean out the filesystem specific data and set the VRECLAIMED
1433 * bit. Also deactivate the vnode if necessary.
1435 * The vnode should have automatically been removed from the syncer
1436 * list as syncer/dirty flags cleared during the cleaning.
1438 vclean_vxlocked(vp, DOCLOSE);
1439 KKASSERT((vp->v_flag & VONWORKLST) == 0);
1442 * Delete from old mount point vnode list, if on one.
1444 if (vp->v_mount != NULL) {
1445 KKASSERT(vp->v_data == NULL);
1446 insmntque(vp, NULL);
1450 * If special device, remove it from special device alias list
1451 * if it is on one. This should normally only occur if a vnode is
1452 * being revoked as the device should otherwise have been released
1455 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
1466 * Lookup a vnode by device number.
1468 * Returns non-zero and *vpp set to a vref'd vnode on success.
1469 * Returns zero on failure.
1472 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp)
1476 lwkt_gettoken(&spechash_token);
1477 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1478 if (type == vp->v_type) {
1481 lwkt_reltoken(&spechash_token);
1485 lwkt_reltoken(&spechash_token);
1490 * Calculate the total number of references to a special device. This
1491 * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1492 * an overloaded field. Since udev2dev can now return NULL, we have
1493 * to check for a NULL v_rdev.
1496 count_dev(cdev_t dev)
1501 if (SLIST_FIRST(&dev->si_hlist)) {
1502 lwkt_gettoken(&spechash_token);
1503 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1504 count += vp->v_opencount;
1506 lwkt_reltoken(&spechash_token);
1512 vcount(struct vnode *vp)
1514 if (vp->v_rdev == NULL)
1516 return(count_dev(vp->v_rdev));
1520 * Initialize VMIO for a vnode. This routine MUST be called before a
1521 * VFS can issue buffer cache ops on a vnode. It is typically called
1522 * when a vnode is initialized from its inode.
1525 vinitvmio(struct vnode *vp, off_t filesize, int blksize, int boff)
1530 object = vp->v_object;
1532 vm_object_hold(object);
1533 KKASSERT(vp->v_object == object);
1536 if (object == NULL) {
1537 object = vnode_pager_alloc(vp, filesize, 0, 0, blksize, boff);
1540 * Dereference the reference we just created. This assumes
1541 * that the object is associated with the vp. Allow it to
1542 * have zero refs. It cannot be destroyed as long as it
1543 * is associated with the vnode.
1545 vm_object_hold(object);
1546 atomic_add_int(&object->ref_count, -1);
1549 KKASSERT((object->flags & OBJ_DEAD) == 0);
1551 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object"));
1552 vsetflags(vp, VOBJBUF);
1553 vm_object_drop(object);
1560 * Print out a description of a vnode.
1562 static char *typename[] =
1563 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1566 vprint(char *label, struct vnode *vp)
1571 kprintf("%s: %p: ", label, (void *)vp);
1573 kprintf("%p: ", (void *)vp);
1574 kprintf("type %s, refcnt %08x, writecount %d, holdcnt %d,",
1575 typename[vp->v_type],
1576 vp->v_refcnt, vp->v_writecount, vp->v_auxrefs);
1578 if (vp->v_flag & VROOT)
1579 strcat(buf, "|VROOT");
1580 if (vp->v_flag & VPFSROOT)
1581 strcat(buf, "|VPFSROOT");
1582 if (vp->v_flag & VTEXT)
1583 strcat(buf, "|VTEXT");
1584 if (vp->v_flag & VSYSTEM)
1585 strcat(buf, "|VSYSTEM");
1586 if (vp->v_flag & VOBJBUF)
1587 strcat(buf, "|VOBJBUF");
1589 kprintf(" flags (%s)", &buf[1]);
1590 if (vp->v_data == NULL) {
1599 * Do the usual access checking.
1600 * file_mode, uid and gid are from the vnode in question,
1601 * while acc_mode and cred are from the VOP_ACCESS parameter list
1604 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
1605 mode_t acc_mode, struct ucred *cred)
1611 * Super-user always gets read/write access, but execute access depends
1612 * on at least one execute bit being set.
1614 if (priv_check_cred(cred, PRIV_ROOT, 0) == 0) {
1615 if ((acc_mode & VEXEC) && type != VDIR &&
1616 (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)
1623 /* Otherwise, check the owner. */
1624 if (cred->cr_uid == uid) {
1625 if (acc_mode & VEXEC)
1627 if (acc_mode & VREAD)
1629 if (acc_mode & VWRITE)
1631 return ((file_mode & mask) == mask ? 0 : EACCES);
1634 /* Otherwise, check the groups. */
1635 ismember = groupmember(gid, cred);
1636 if (cred->cr_svgid == gid || ismember) {
1637 if (acc_mode & VEXEC)
1639 if (acc_mode & VREAD)
1641 if (acc_mode & VWRITE)
1643 return ((file_mode & mask) == mask ? 0 : EACCES);
1646 /* Otherwise, check everyone else. */
1647 if (acc_mode & VEXEC)
1649 if (acc_mode & VREAD)
1651 if (acc_mode & VWRITE)
1653 return ((file_mode & mask) == mask ? 0 : EACCES);
1657 #include <ddb/ddb.h>
1659 static int db_show_locked_vnodes(struct mount *mp, void *data);
1662 * List all of the locked vnodes in the system.
1663 * Called when debugging the kernel.
1665 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
1667 kprintf("Locked vnodes\n");
1668 mountlist_scan(db_show_locked_vnodes, NULL,
1669 MNTSCAN_FORWARD|MNTSCAN_NOBUSY);
1673 db_show_locked_vnodes(struct mount *mp, void *data __unused)
1677 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
1678 if (vn_islocked(vp))
1686 * Top level filesystem related information gathering.
1688 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
1691 vfs_sysctl(SYSCTL_HANDLER_ARGS)
1693 int *name = (int *)arg1 - 1; /* XXX */
1694 u_int namelen = arg2 + 1; /* XXX */
1695 struct vfsconf *vfsp;
1698 #if 1 || defined(COMPAT_PRELITE2)
1699 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1701 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
1705 /* all sysctl names at this level are at least name and field */
1707 return (ENOTDIR); /* overloaded */
1708 if (name[0] != VFS_GENERIC) {
1709 vfsp = vfsconf_find_by_typenum(name[0]);
1711 return (EOPNOTSUPP);
1712 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1713 oldp, oldlenp, newp, newlen, p));
1717 case VFS_MAXTYPENUM:
1720 maxtypenum = vfsconf_get_maxtypenum();
1721 return (SYSCTL_OUT(req, &maxtypenum, sizeof(maxtypenum)));
1724 return (ENOTDIR); /* overloaded */
1725 vfsp = vfsconf_find_by_typenum(name[2]);
1727 return (EOPNOTSUPP);
1728 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
1730 return (EOPNOTSUPP);
1733 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
1734 "Generic filesystem");
1736 #if 1 || defined(COMPAT_PRELITE2)
1739 sysctl_ovfs_conf_iter(struct vfsconf *vfsp, void *data)
1742 struct ovfsconf ovfs;
1743 struct sysctl_req *req = (struct sysctl_req*) data;
1745 bzero(&ovfs, sizeof(ovfs));
1746 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
1747 strcpy(ovfs.vfc_name, vfsp->vfc_name);
1748 ovfs.vfc_index = vfsp->vfc_typenum;
1749 ovfs.vfc_refcount = vfsp->vfc_refcount;
1750 ovfs.vfc_flags = vfsp->vfc_flags;
1751 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
1753 return error; /* abort iteration with error code */
1755 return 0; /* continue iterating with next element */
1759 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
1761 return vfsconf_each(sysctl_ovfs_conf_iter, (void*)req);
1764 #endif /* 1 || COMPAT_PRELITE2 */
1767 * Check to see if a filesystem is mounted on a block device.
1770 vfs_mountedon(struct vnode *vp)
1774 if ((dev = vp->v_rdev) == NULL) {
1775 /* if (vp->v_type != VBLK)
1776 dev = get_dev(vp->v_uminor, vp->v_umajor); */
1778 if (dev != NULL && dev->si_mountpoint)
1784 * Unmount all filesystems. The list is traversed in reverse order
1785 * of mounting to avoid dependencies.
1788 static int vfs_umountall_callback(struct mount *mp, void *data);
1791 vfs_unmountall(void)
1796 count = mountlist_scan(vfs_umountall_callback,
1797 NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY);
1803 vfs_umountall_callback(struct mount *mp, void *data)
1807 error = dounmount(mp, MNT_FORCE);
1809 mountlist_remove(mp);
1810 kprintf("unmount of filesystem mounted from %s failed (",
1811 mp->mnt_stat.f_mntfromname);
1815 kprintf("%d)\n", error);
1821 * Checks the mount flags for parameter mp and put the names comma-separated
1822 * into a string buffer buf with a size limit specified by len.
1824 * It returns the number of bytes written into buf, and (*errorp) will be
1825 * set to 0, EINVAL (if passed length is 0), or ENOSPC (supplied buffer was
1826 * not large enough). The buffer will be 0-terminated if len was not 0.
1829 vfs_flagstostr(int flags, const struct mountctl_opt *optp,
1830 char *buf, size_t len, int *errorp)
1832 static const struct mountctl_opt optnames[] = {
1833 { MNT_RDONLY, "read-only" },
1834 { MNT_SYNCHRONOUS, "synchronous" },
1835 { MNT_NOEXEC, "noexec" },
1836 { MNT_NOSUID, "nosuid" },
1837 { MNT_NODEV, "nodev" },
1838 { MNT_ASYNC, "asynchronous" },
1839 { MNT_SUIDDIR, "suiddir" },
1840 { MNT_SOFTDEP, "soft-updates" },
1841 { MNT_NOSYMFOLLOW, "nosymfollow" },
1842 { MNT_TRIM, "trim" },
1843 { MNT_NOATIME, "noatime" },
1844 { MNT_NOCLUSTERR, "noclusterr" },
1845 { MNT_NOCLUSTERW, "noclusterw" },
1846 { MNT_EXRDONLY, "NFS read-only" },
1847 { MNT_EXPORTED, "NFS exported" },
1848 /* Remaining NFS flags could come here */
1849 { MNT_LOCAL, "local" },
1850 { MNT_QUOTA, "with-quotas" },
1851 /* { MNT_ROOTFS, "rootfs" }, */
1852 /* { MNT_IGNORE, "ignore" }, */
1862 bleft = len - 1; /* leave room for trailing \0 */
1865 * Checks the size of the string. If it contains
1866 * any data, then we will append the new flags to
1869 actsize = strlen(buf);
1873 /* Default flags if no flags passed */
1877 if (bleft < 0) { /* degenerate case, 0-length buffer */
1882 for (; flags && optp->o_opt; ++optp) {
1883 if ((flags & optp->o_opt) == 0)
1885 optlen = strlen(optp->o_name);
1886 if (bwritten || actsize > 0) {
1891 buf[bwritten++] = ',';
1892 buf[bwritten++] = ' ';
1895 if (bleft < optlen) {
1899 bcopy(optp->o_name, buf + bwritten, optlen);
1902 flags &= ~optp->o_opt;
1906 * Space already reserved for trailing \0
1913 * Build hash lists of net addresses and hang them off the mount point.
1914 * Called by ufs_mount() to set up the lists of export addresses.
1917 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1918 const struct export_args *argp)
1921 struct radix_node_head *rnh;
1923 struct radix_node *rn;
1924 struct sockaddr *saddr, *smask = NULL;
1927 if (argp->ex_addrlen == 0) {
1928 if (mp->mnt_flag & MNT_DEFEXPORTED)
1930 np = &nep->ne_defexported;
1931 np->netc_exflags = argp->ex_flags;
1932 np->netc_anon = argp->ex_anon;
1933 np->netc_anon.cr_ref = 1;
1934 mp->mnt_flag |= MNT_DEFEXPORTED;
1938 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
1940 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
1943 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1944 np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK | M_ZERO);
1945 saddr = (struct sockaddr *) (np + 1);
1946 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
1948 if (saddr->sa_len > argp->ex_addrlen)
1949 saddr->sa_len = argp->ex_addrlen;
1950 if (argp->ex_masklen) {
1951 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1952 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
1955 if (smask->sa_len > argp->ex_masklen)
1956 smask->sa_len = argp->ex_masklen;
1959 if (nep->ne_maskhead == NULL) {
1960 if (!rn_inithead((void **)&nep->ne_maskhead, NULL, 0)) {
1965 if((rnh = vfs_create_addrlist_af(saddr->sa_family, nep)) == NULL) {
1969 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh,
1972 if (rn == NULL || np != (struct netcred *) rn) { /* already exists */
1976 np->netc_exflags = argp->ex_flags;
1977 np->netc_anon = argp->ex_anon;
1978 np->netc_anon.cr_ref = 1;
1981 kfree(np, M_NETADDR);
1987 vfs_free_netcred(struct radix_node *rn, void *w)
1989 struct radix_node_head *rnh = (struct radix_node_head *) w;
1991 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
1992 kfree((caddr_t) rn, M_NETADDR);
1996 static struct radix_node_head *
1997 vfs_create_addrlist_af(int af, struct netexport *nep)
1999 struct radix_node_head *rnh = NULL;
2000 #if defined(INET) || defined(INET6)
2001 struct radix_node_head *maskhead = nep->ne_maskhead;
2005 NE_ASSERT_LOCKED(nep);
2006 KKASSERT(maskhead != NULL);
2010 if ((rnh = nep->ne_inethead) == NULL) {
2011 off = offsetof(struct sockaddr_in, sin_addr) << 3;
2012 if (!rn_inithead((void **)&rnh, maskhead, off))
2014 nep->ne_inethead = rnh;
2020 if ((rnh = nep->ne_inet6head) == NULL) {
2021 off = offsetof(struct sockaddr_in6, sin6_addr) << 3;
2022 if (!rn_inithead((void **)&rnh, maskhead, off))
2024 nep->ne_inet6head = rnh;
2033 vfs_free_addrlist_af(struct radix_node_head **prnh)
2035 struct radix_node_head *rnh = *prnh;
2037 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, rnh);
2038 kfree(rnh, M_RTABLE);
2043 * Free the net address hash lists that are hanging off the mount points.
2046 vfs_free_addrlist(struct netexport *nep)
2049 if (nep->ne_inethead != NULL)
2050 vfs_free_addrlist_af(&nep->ne_inethead);
2051 if (nep->ne_inet6head != NULL)
2052 vfs_free_addrlist_af(&nep->ne_inet6head);
2053 if (nep->ne_maskhead)
2054 vfs_free_addrlist_af(&nep->ne_maskhead);
2059 vfs_export(struct mount *mp, struct netexport *nep,
2060 const struct export_args *argp)
2064 if (argp->ex_flags & MNT_DELEXPORT) {
2065 if (mp->mnt_flag & MNT_EXPUBLIC) {
2066 vfs_setpublicfs(NULL, NULL, NULL);
2067 mp->mnt_flag &= ~MNT_EXPUBLIC;
2069 vfs_free_addrlist(nep);
2070 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2072 if (argp->ex_flags & MNT_EXPORTED) {
2073 if (argp->ex_flags & MNT_EXPUBLIC) {
2074 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2076 mp->mnt_flag |= MNT_EXPUBLIC;
2078 if ((error = vfs_hang_addrlist(mp, nep, argp)))
2080 mp->mnt_flag |= MNT_EXPORTED;
2087 * Set the publicly exported filesystem (WebNFS). Currently, only
2088 * one public filesystem is possible in the spec (RFC 2054 and 2055)
2091 vfs_setpublicfs(struct mount *mp, struct netexport *nep,
2092 const struct export_args *argp)
2099 * mp == NULL -> invalidate the current info, the FS is
2100 * no longer exported. May be called from either vfs_export
2101 * or unmount, so check if it hasn't already been done.
2104 if (nfs_pub.np_valid) {
2105 nfs_pub.np_valid = 0;
2106 if (nfs_pub.np_index != NULL) {
2107 kfree(nfs_pub.np_index, M_TEMP);
2108 nfs_pub.np_index = NULL;
2115 * Only one allowed at a time.
2117 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2121 * Get real filehandle for root of exported FS.
2123 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2124 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2126 if ((error = VFS_ROOT(mp, &rvp)))
2129 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2135 * If an indexfile was specified, pull it in.
2137 if (argp->ex_indexfile != NULL) {
2140 error = vn_get_namelen(rvp, &namelen);
2143 nfs_pub.np_index = kmalloc(namelen, M_TEMP, M_WAITOK);
2144 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2148 * Check for illegal filenames.
2150 for (cp = nfs_pub.np_index; *cp; cp++) {
2158 kfree(nfs_pub.np_index, M_TEMP);
2163 nfs_pub.np_mount = mp;
2164 nfs_pub.np_valid = 1;
2169 vfs_export_lookup(struct mount *mp, struct netexport *nep,
2170 struct sockaddr *nam)
2173 struct radix_node_head *rnh;
2174 struct sockaddr *saddr;
2177 if (mp->mnt_flag & MNT_EXPORTED) {
2179 * Lookup in the export list first.
2184 switch (saddr->sa_family) {
2187 rnh = nep->ne_inethead;
2192 rnh = nep->ne_inet6head;
2199 np = (struct netcred *)
2200 (*rnh->rnh_matchaddr)((char *)saddr,
2202 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2208 * If no address match, use the default if it exists.
2210 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2211 np = &nep->ne_defexported;
2217 * perform msync on all vnodes under a mount point. The mount point must
2218 * be locked. This code is also responsible for lazy-freeing unreferenced
2219 * vnodes whos VM objects no longer contain pages.
2221 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
2223 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked,
2224 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it
2225 * way up in this high level function.
2227 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
2228 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data);
2231 vfs_msync(struct mount *mp, int flags)
2236 * tmpfs sets this flag to prevent msync(), sync, and the
2237 * filesystem periodic syncer from trying to flush VM pages
2238 * to swap. Only pure memory pressure flushes tmpfs VM pages
2241 if (mp->mnt_kern_flag & MNTK_NOMSYNC)
2245 * Ok, scan the vnodes for work. If the filesystem is using the
2246 * syncer thread feature we can use vsyncscan() instead of
2247 * vmntvnodescan(), which is much faster.
2249 vmsc_flags = VMSC_GETVP;
2250 if (flags != MNT_WAIT)
2251 vmsc_flags |= VMSC_NOWAIT;
2253 if (mp->mnt_kern_flag & MNTK_THR_SYNC) {
2254 vsyncscan(mp, vmsc_flags, vfs_msync_scan2,
2255 (void *)(intptr_t)flags);
2257 vmntvnodescan(mp, vmsc_flags,
2258 vfs_msync_scan1, vfs_msync_scan2,
2259 (void *)(intptr_t)flags);
2264 * scan1 is a fast pre-check. There could be hundreds of thousands of
2265 * vnodes, we cannot afford to do anything heavy weight until we have a
2266 * fairly good indication that there is work to do.
2270 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
2272 int flags = (int)(intptr_t)data;
2274 if ((vp->v_flag & VRECLAIMED) == 0) {
2275 if (vp->v_auxrefs == 0 && VREFCNT(vp) <= 0 &&
2277 return(0); /* call scan2 */
2279 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
2280 (vp->v_flag & VOBJDIRTY) &&
2281 (flags == MNT_WAIT || vn_islocked(vp) == 0)) {
2282 return(0); /* call scan2 */
2287 * do not call scan2, continue the loop
2293 * This callback is handed a locked vnode.
2297 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data)
2300 int flags = (int)(intptr_t)data;
2302 if (vp->v_flag & VRECLAIMED)
2305 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) {
2306 if ((obj = vp->v_object) != NULL) {
2307 vm_object_page_clean(obj, 0, 0,
2308 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2315 * Wake up anyone interested in vp because it is being revoked.
2318 vn_gone(struct vnode *vp)
2320 lwkt_gettoken(&vp->v_token);
2321 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, NOTE_REVOKE);
2322 lwkt_reltoken(&vp->v_token);
2326 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened
2327 * (or v_rdev might be NULL).
2330 vn_todev(struct vnode *vp)
2332 if (vp->v_type != VBLK && vp->v_type != VCHR)
2334 KKASSERT(vp->v_rdev != NULL);
2335 return (vp->v_rdev);
2339 * Check if vnode represents a disk device. The vnode does not need to be
2345 vn_isdisk(struct vnode *vp, int *errp)
2349 if (vp->v_type != VCHR) {
2362 if (dev_is_good(dev) == 0) {
2367 if ((dev_dflags(dev) & D_DISK) == 0) {
2378 vn_get_namelen(struct vnode *vp, int *namelen)
2381 register_t retval[2];
2383 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval);
2386 *namelen = (int)retval[0];
2391 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type,
2392 uint16_t d_namlen, const char *d_name)
2397 len = _DIRENT_RECLEN(d_namlen);
2398 if (len > uio->uio_resid)
2401 dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO);
2404 dp->d_namlen = d_namlen;
2405 dp->d_type = d_type;
2406 bcopy(d_name, dp->d_name, d_namlen);
2408 *error = uiomove((caddr_t)dp, len, uio);
2416 vn_mark_atime(struct vnode *vp, struct thread *td)
2418 struct proc *p = td->td_proc;
2419 struct ucred *cred = p ? p->p_ucred : proc0.p_ucred;
2421 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
2422 VOP_MARKATIME(vp, cred);