ffs_vfsops.c revision 181345
1/*-
2 * Copyright (c) 1989, 1991, 1993, 1994
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 *    may be used to endorse or promote products derived from this software
15 *    without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 *	@(#)ffs_vfsops.c	8.31 (Berkeley) 5/20/95
30 */
31
32#include <sys/cdefs.h>
33__FBSDID("$FreeBSD: head/sys/ufs/ffs/ffs_vfsops.c 181345 2008-08-06 14:47:19Z kib $");
34
35#include "opt_mac.h"
36#include "opt_quota.h"
37#include "opt_ufs.h"
38#include "opt_ffs.h"
39
40#include <sys/param.h>
41#include <sys/systm.h>
42#include <sys/namei.h>
43#include <sys/priv.h>
44#include <sys/proc.h>
45#include <sys/kernel.h>
46#include <sys/vnode.h>
47#include <sys/mount.h>
48#include <sys/bio.h>
49#include <sys/buf.h>
50#include <sys/conf.h>
51#include <sys/fcntl.h>
52#include <sys/malloc.h>
53#include <sys/mutex.h>
54
55#include <security/mac/mac_framework.h>
56
57#include <ufs/ufs/extattr.h>
58#include <ufs/ufs/gjournal.h>
59#include <ufs/ufs/quota.h>
60#include <ufs/ufs/ufsmount.h>
61#include <ufs/ufs/inode.h>
62#include <ufs/ufs/ufs_extern.h>
63
64#include <ufs/ffs/fs.h>
65#include <ufs/ffs/ffs_extern.h>
66
67#include <vm/vm.h>
68#include <vm/uma.h>
69#include <vm/vm_page.h>
70
71#include <geom/geom.h>
72#include <geom/geom_vfs.h>
73
74static uma_zone_t uma_inode, uma_ufs1, uma_ufs2;
75
76static int	ffs_reload(struct mount *, struct thread *);
77static int	ffs_mountfs(struct vnode *, struct mount *, struct thread *);
78static void	ffs_oldfscompat_read(struct fs *, struct ufsmount *,
79		    ufs2_daddr_t);
80static void	ffs_oldfscompat_write(struct fs *, struct ufsmount *);
81static void	ffs_ifree(struct ufsmount *ump, struct inode *ip);
82static vfs_init_t ffs_init;
83static vfs_uninit_t ffs_uninit;
84static vfs_extattrctl_t ffs_extattrctl;
85static vfs_cmount_t ffs_cmount;
86static vfs_unmount_t ffs_unmount;
87static vfs_mount_t ffs_mount;
88static vfs_statfs_t ffs_statfs;
89static vfs_fhtovp_t ffs_fhtovp;
90static vfs_sync_t ffs_sync;
91
92static struct vfsops ufs_vfsops = {
93	.vfs_extattrctl =	ffs_extattrctl,
94	.vfs_fhtovp =		ffs_fhtovp,
95	.vfs_init =		ffs_init,
96	.vfs_mount =		ffs_mount,
97	.vfs_cmount =		ffs_cmount,
98	.vfs_quotactl =		ufs_quotactl,
99	.vfs_root =		ufs_root,
100	.vfs_statfs =		ffs_statfs,
101	.vfs_sync =		ffs_sync,
102	.vfs_uninit =		ffs_uninit,
103	.vfs_unmount =		ffs_unmount,
104	.vfs_vget =		ffs_vget,
105};
106
107VFS_SET(ufs_vfsops, ufs, 0);
108MODULE_VERSION(ufs, 1);
109
110static b_strategy_t ffs_geom_strategy;
111static b_write_t ffs_bufwrite;
112
113static struct buf_ops ffs_ops = {
114	.bop_name =	"FFS",
115	.bop_write =	ffs_bufwrite,
116	.bop_strategy =	ffs_geom_strategy,
117	.bop_sync =	bufsync,
118#ifdef NO_FFS_SNAPSHOT
119	.bop_bdflush =	bufbdflush,
120#else
121	.bop_bdflush =	ffs_bdflush,
122#endif
123};
124
125static const char *ffs_opts[] = { "acls", "async", "noatime", "noclusterr",
126    "noclusterw", "noexec", "export", "force", "from", "multilabel",
127    "snapshot", "nosuid", "suiddir", "nosymfollow", "sync",
128    "union", NULL };
129
130static int
131ffs_mount(struct mount *mp, struct thread *td)
132{
133	struct vnode *devvp;
134	struct ufsmount *ump = 0;
135	struct fs *fs;
136	int error, flags;
137	u_int mntorflags, mntandnotflags;
138	mode_t accessmode;
139	struct nameidata ndp;
140	char *fspec;
141
142	if (vfs_filteropt(mp->mnt_optnew, ffs_opts))
143		return (EINVAL);
144	if (uma_inode == NULL) {
145		uma_inode = uma_zcreate("FFS inode",
146		    sizeof(struct inode), NULL, NULL, NULL, NULL,
147		    UMA_ALIGN_PTR, 0);
148		uma_ufs1 = uma_zcreate("FFS1 dinode",
149		    sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL,
150		    UMA_ALIGN_PTR, 0);
151		uma_ufs2 = uma_zcreate("FFS2 dinode",
152		    sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL,
153		    UMA_ALIGN_PTR, 0);
154	}
155
156	fspec = vfs_getopts(mp->mnt_optnew, "from", &error);
157	if (error)
158		return (error);
159
160	mntorflags = 0;
161	mntandnotflags = 0;
162	if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0)
163		mntorflags |= MNT_ACLS;
164
165	if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) {
166		mntorflags |= MNT_SNAPSHOT;
167		/*
168		 * Once we have set the MNT_SNAPSHOT flag, do not
169		 * persist "snapshot" in the options list.
170		 */
171		vfs_deleteopt(mp->mnt_optnew, "snapshot");
172		if (mp->mnt_opt != NULL)
173			vfs_deleteopt(mp->mnt_opt, "snapshot");
174	}
175
176	MNT_ILOCK(mp);
177	mp->mnt_flag = (mp->mnt_flag | mntorflags) & ~mntandnotflags;
178	MNT_IUNLOCK(mp);
179	/*
180	 * If updating, check whether changing from read-only to
181	 * read/write; if there is no device name, that's all we do.
182	 */
183	if (mp->mnt_flag & MNT_UPDATE) {
184		ump = VFSTOUFS(mp);
185		fs = ump->um_fs;
186		devvp = ump->um_devvp;
187		if (fs->fs_ronly == 0 &&
188		    vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
189			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
190				return (error);
191			/*
192			 * Flush any dirty data.
193			 */
194			if ((error = ffs_sync(mp, MNT_WAIT, td)) != 0) {
195				vn_finished_write(mp);
196				return (error);
197			}
198			/*
199			 * Check for and optionally get rid of files open
200			 * for writing.
201			 */
202			flags = WRITECLOSE;
203			if (mp->mnt_flag & MNT_FORCE)
204				flags |= FORCECLOSE;
205			if (mp->mnt_flag & MNT_SOFTDEP) {
206				error = softdep_flushfiles(mp, flags, td);
207			} else {
208				error = ffs_flushfiles(mp, flags, td);
209			}
210			if (error) {
211				vn_finished_write(mp);
212				return (error);
213			}
214			if (fs->fs_pendingblocks != 0 ||
215			    fs->fs_pendinginodes != 0) {
216				printf("%s: %s: blocks %jd files %d\n",
217				    fs->fs_fsmnt, "update error",
218				    (intmax_t)fs->fs_pendingblocks,
219				    fs->fs_pendinginodes);
220				fs->fs_pendingblocks = 0;
221				fs->fs_pendinginodes = 0;
222			}
223			if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
224				fs->fs_clean = 1;
225			if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
226				fs->fs_ronly = 0;
227				fs->fs_clean = 0;
228				vn_finished_write(mp);
229				return (error);
230			}
231			vn_finished_write(mp);
232			DROP_GIANT();
233			g_topology_lock();
234			g_access(ump->um_cp, 0, -1, 0);
235			g_topology_unlock();
236			PICKUP_GIANT();
237			fs->fs_ronly = 1;
238			MNT_ILOCK(mp);
239			mp->mnt_flag |= MNT_RDONLY;
240			MNT_IUNLOCK(mp);
241		}
242		if ((mp->mnt_flag & MNT_RELOAD) &&
243		    (error = ffs_reload(mp, td)) != 0)
244			return (error);
245		if (fs->fs_ronly &&
246		    !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
247			/*
248			 * If upgrade to read-write by non-root, then verify
249			 * that user has necessary permissions on the device.
250			 */
251			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
252			error = VOP_ACCESS(devvp, VREAD | VWRITE,
253			    td->td_ucred, td);
254			if (error)
255				error = priv_check(td, PRIV_VFS_MOUNT_PERM);
256			if (error) {
257				VOP_UNLOCK(devvp, 0);
258				return (error);
259			}
260			VOP_UNLOCK(devvp, 0);
261			fs->fs_flags &= ~FS_UNCLEAN;
262			if (fs->fs_clean == 0) {
263				fs->fs_flags |= FS_UNCLEAN;
264				if ((mp->mnt_flag & MNT_FORCE) ||
265				    ((fs->fs_flags & FS_NEEDSFSCK) == 0 &&
266				     (fs->fs_flags & FS_DOSOFTDEP))) {
267					printf("WARNING: %s was not %s\n",
268					   fs->fs_fsmnt, "properly dismounted");
269				} else {
270					printf(
271"WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
272					    fs->fs_fsmnt);
273					return (EPERM);
274				}
275			}
276			DROP_GIANT();
277			g_topology_lock();
278			/*
279			 * If we're the root device, we may not have an E count
280			 * yet, get it now.
281			 */
282			if (ump->um_cp->ace == 0)
283				error = g_access(ump->um_cp, 0, 1, 1);
284			else
285				error = g_access(ump->um_cp, 0, 1, 0);
286			g_topology_unlock();
287			PICKUP_GIANT();
288			if (error)
289				return (error);
290			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
291				return (error);
292			fs->fs_ronly = 0;
293			MNT_ILOCK(mp);
294			mp->mnt_flag &= ~MNT_RDONLY;
295			MNT_IUNLOCK(mp);
296			fs->fs_clean = 0;
297			if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
298				vn_finished_write(mp);
299				return (error);
300			}
301			/* check to see if we need to start softdep */
302			if ((fs->fs_flags & FS_DOSOFTDEP) &&
303			    (error = softdep_mount(devvp, mp, fs, td->td_ucred))){
304				vn_finished_write(mp);
305				return (error);
306			}
307			if (fs->fs_snapinum[0] != 0)
308				ffs_snapshot_mount(mp);
309			vn_finished_write(mp);
310		}
311		/*
312		 * Soft updates is incompatible with "async",
313		 * so if we are doing softupdates stop the user
314		 * from setting the async flag in an update.
315		 * Softdep_mount() clears it in an initial mount
316		 * or ro->rw remount.
317		 */
318		if (mp->mnt_flag & MNT_SOFTDEP) {
319			/* XXX: Reset too late ? */
320			MNT_ILOCK(mp);
321			mp->mnt_flag &= ~MNT_ASYNC;
322			MNT_IUNLOCK(mp);
323		}
324		/*
325		 * Keep MNT_ACLS flag if it is stored in superblock.
326		 */
327		if ((fs->fs_flags & FS_ACLS) != 0) {
328			/* XXX: Set too late ? */
329			MNT_ILOCK(mp);
330			mp->mnt_flag |= MNT_ACLS;
331			MNT_IUNLOCK(mp);
332		}
333
334		/*
335		 * If this is a snapshot request, take the snapshot.
336		 */
337		if (mp->mnt_flag & MNT_SNAPSHOT)
338			return (ffs_snapshot(mp, fspec));
339	}
340
341	/*
342	 * Not an update, or updating the name: look up the name
343	 * and verify that it refers to a sensible disk device.
344	 */
345	NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td);
346	if ((error = namei(&ndp)) != 0)
347		return (error);
348	NDFREE(&ndp, NDF_ONLY_PNBUF);
349	devvp = ndp.ni_vp;
350	if (!vn_isdisk(devvp, &error)) {
351		vput(devvp);
352		return (error);
353	}
354
355	/*
356	 * If mount by non-root, then verify that user has necessary
357	 * permissions on the device.
358	 */
359	accessmode = VREAD;
360	if ((mp->mnt_flag & MNT_RDONLY) == 0)
361		accessmode |= VWRITE;
362	error = VOP_ACCESS(devvp, accessmode, td->td_ucred, td);
363	if (error)
364		error = priv_check(td, PRIV_VFS_MOUNT_PERM);
365	if (error) {
366		vput(devvp);
367		return (error);
368	}
369
370	if (mp->mnt_flag & MNT_UPDATE) {
371		/*
372		 * Update only
373		 *
374		 * If it's not the same vnode, or at least the same device
375		 * then it's not correct.
376		 */
377
378		if (devvp->v_rdev != ump->um_devvp->v_rdev)
379			error = EINVAL;	/* needs translation */
380		vput(devvp);
381		if (error)
382			return (error);
383	} else {
384		/*
385		 * New mount
386		 *
387		 * We need the name for the mount point (also used for
388		 * "last mounted on") copied in. If an error occurs,
389		 * the mount point is discarded by the upper level code.
390		 * Note that vfs_mount() populates f_mntonname for us.
391		 */
392		if ((error = ffs_mountfs(devvp, mp, td)) != 0) {
393			vrele(devvp);
394			return (error);
395		}
396	}
397	vfs_mountedfrom(mp, fspec);
398	return (0);
399}
400
401/*
402 * Compatibility with old mount system call.
403 */
404
405static int
406ffs_cmount(struct mntarg *ma, void *data, int flags, struct thread *td)
407{
408	struct ufs_args args;
409	int error;
410
411	if (data == NULL)
412		return (EINVAL);
413	error = copyin(data, &args, sizeof args);
414	if (error)
415		return (error);
416
417	ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN);
418	ma = mount_arg(ma, "export", &args.export, sizeof args.export);
419	error = kernel_mount(ma, flags);
420
421	return (error);
422}
423
424/*
425 * Reload all incore data for a filesystem (used after running fsck on
426 * the root filesystem and finding things to fix). The filesystem must
427 * be mounted read-only.
428 *
429 * Things to do to update the mount:
430 *	1) invalidate all cached meta-data.
431 *	2) re-read superblock from disk.
432 *	3) re-read summary information from disk.
433 *	4) invalidate all inactive vnodes.
434 *	5) invalidate all cached file data.
435 *	6) re-read inode data for all active vnodes.
436 */
437static int
438ffs_reload(struct mount *mp, struct thread *td)
439{
440	struct vnode *vp, *mvp, *devvp;
441	struct inode *ip;
442	void *space;
443	struct buf *bp;
444	struct fs *fs, *newfs;
445	struct ufsmount *ump;
446	ufs2_daddr_t sblockloc;
447	int i, blks, size, error;
448	int32_t *lp;
449
450	if ((mp->mnt_flag & MNT_RDONLY) == 0)
451		return (EINVAL);
452	ump = VFSTOUFS(mp);
453	/*
454	 * Step 1: invalidate all cached meta-data.
455	 */
456	devvp = VFSTOUFS(mp)->um_devvp;
457	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
458	if (vinvalbuf(devvp, 0, td, 0, 0) != 0)
459		panic("ffs_reload: dirty1");
460	VOP_UNLOCK(devvp, 0);
461
462	/*
463	 * Step 2: re-read superblock from disk.
464	 */
465	fs = VFSTOUFS(mp)->um_fs;
466	if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize,
467	    NOCRED, &bp)) != 0)
468		return (error);
469	newfs = (struct fs *)bp->b_data;
470	if ((newfs->fs_magic != FS_UFS1_MAGIC &&
471	     newfs->fs_magic != FS_UFS2_MAGIC) ||
472	    newfs->fs_bsize > MAXBSIZE ||
473	    newfs->fs_bsize < sizeof(struct fs)) {
474			brelse(bp);
475			return (EIO);		/* XXX needs translation */
476	}
477	/*
478	 * Copy pointer fields back into superblock before copying in	XXX
479	 * new superblock. These should really be in the ufsmount.	XXX
480	 * Note that important parameters (eg fs_ncg) are unchanged.
481	 */
482	newfs->fs_csp = fs->fs_csp;
483	newfs->fs_maxcluster = fs->fs_maxcluster;
484	newfs->fs_contigdirs = fs->fs_contigdirs;
485	newfs->fs_active = fs->fs_active;
486	/* The file system is still read-only. */
487	newfs->fs_ronly = 1;
488	sblockloc = fs->fs_sblockloc;
489	bcopy(newfs, fs, (u_int)fs->fs_sbsize);
490	brelse(bp);
491	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
492	ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc);
493	UFS_LOCK(ump);
494	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
495		printf("%s: reload pending error: blocks %jd files %d\n",
496		    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
497		    fs->fs_pendinginodes);
498		fs->fs_pendingblocks = 0;
499		fs->fs_pendinginodes = 0;
500	}
501	UFS_UNLOCK(ump);
502
503	/*
504	 * Step 3: re-read summary information from disk.
505	 */
506	blks = howmany(fs->fs_cssize, fs->fs_fsize);
507	space = fs->fs_csp;
508	for (i = 0; i < blks; i += fs->fs_frag) {
509		size = fs->fs_bsize;
510		if (i + fs->fs_frag > blks)
511			size = (blks - i) * fs->fs_fsize;
512		error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
513		    NOCRED, &bp);
514		if (error)
515			return (error);
516		bcopy(bp->b_data, space, (u_int)size);
517		space = (char *)space + size;
518		brelse(bp);
519	}
520	/*
521	 * We no longer know anything about clusters per cylinder group.
522	 */
523	if (fs->fs_contigsumsize > 0) {
524		lp = fs->fs_maxcluster;
525		for (i = 0; i < fs->fs_ncg; i++)
526			*lp++ = fs->fs_contigsumsize;
527	}
528
529loop:
530	MNT_ILOCK(mp);
531	MNT_VNODE_FOREACH(vp, mp, mvp) {
532		VI_LOCK(vp);
533		if (vp->v_iflag & VI_DOOMED) {
534			VI_UNLOCK(vp);
535			continue;
536		}
537		MNT_IUNLOCK(mp);
538		/*
539		 * Step 4: invalidate all cached file data.
540		 */
541		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
542			MNT_VNODE_FOREACH_ABORT(mp, mvp);
543			goto loop;
544		}
545		if (vinvalbuf(vp, 0, td, 0, 0))
546			panic("ffs_reload: dirty2");
547		/*
548		 * Step 5: re-read inode data for all active vnodes.
549		 */
550		ip = VTOI(vp);
551		error =
552		    bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
553		    (int)fs->fs_bsize, NOCRED, &bp);
554		if (error) {
555			VOP_UNLOCK(vp, 0);
556			vrele(vp);
557			MNT_VNODE_FOREACH_ABORT(mp, mvp);
558			return (error);
559		}
560		ffs_load_inode(bp, ip, fs, ip->i_number);
561		ip->i_effnlink = ip->i_nlink;
562		brelse(bp);
563		VOP_UNLOCK(vp, 0);
564		vrele(vp);
565		MNT_ILOCK(mp);
566	}
567	MNT_IUNLOCK(mp);
568	return (0);
569}
570
571/*
572 * Possible superblock locations ordered from most to least likely.
573 */
574static int sblock_try[] = SBLOCKSEARCH;
575
576/*
577 * Common code for mount and mountroot
578 */
579static int
580ffs_mountfs(devvp, mp, td)
581	struct vnode *devvp;
582	struct mount *mp;
583	struct thread *td;
584{
585	struct ufsmount *ump;
586	struct buf *bp;
587	struct fs *fs;
588	struct cdev *dev;
589	void *space;
590	ufs2_daddr_t sblockloc;
591	int error, i, blks, size, ronly;
592	int32_t *lp;
593	struct ucred *cred;
594	struct g_consumer *cp;
595	struct mount *nmp;
596
597	dev = devvp->v_rdev;
598	cred = td ? td->td_ucred : NOCRED;
599
600	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
601	DROP_GIANT();
602	g_topology_lock();
603	error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1);
604
605	/*
606	 * If we are a root mount, drop the E flag so fsck can do its magic.
607	 * We will pick it up again when we remount R/W.
608	 */
609	if (error == 0 && ronly && (mp->mnt_flag & MNT_ROOTFS))
610		error = g_access(cp, 0, 0, -1);
611	g_topology_unlock();
612	PICKUP_GIANT();
613	VOP_UNLOCK(devvp, 0);
614	if (error)
615		return (error);
616	if (devvp->v_rdev->si_iosize_max != 0)
617		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
618	if (mp->mnt_iosize_max > MAXPHYS)
619		mp->mnt_iosize_max = MAXPHYS;
620
621	devvp->v_bufobj.bo_private = cp;
622	devvp->v_bufobj.bo_ops = &ffs_ops;
623
624	bp = NULL;
625	ump = NULL;
626	fs = NULL;
627	sblockloc = 0;
628	/*
629	 * Try reading the superblock in each of its possible locations.
630	 */
631	for (i = 0; sblock_try[i] != -1; i++) {
632		if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) {
633			error = EINVAL;
634			vfs_mount_error(mp,
635			    "Invalid sectorsize %d for superblock size %d",
636			    cp->provider->sectorsize, SBLOCKSIZE);
637			goto out;
638		}
639		if ((error = bread(devvp, btodb(sblock_try[i]), SBLOCKSIZE,
640		    cred, &bp)) != 0)
641			goto out;
642		fs = (struct fs *)bp->b_data;
643		sblockloc = sblock_try[i];
644		if ((fs->fs_magic == FS_UFS1_MAGIC ||
645		     (fs->fs_magic == FS_UFS2_MAGIC &&
646		      (fs->fs_sblockloc == sblockloc ||
647		       (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) &&
648		    fs->fs_bsize <= MAXBSIZE &&
649		    fs->fs_bsize >= sizeof(struct fs))
650			break;
651		brelse(bp);
652		bp = NULL;
653	}
654	if (sblock_try[i] == -1) {
655		error = EINVAL;		/* XXX needs translation */
656		goto out;
657	}
658	fs->fs_fmod = 0;
659	fs->fs_flags &= ~FS_INDEXDIRS;	/* no support for directory indicies */
660	fs->fs_flags &= ~FS_UNCLEAN;
661	if (fs->fs_clean == 0) {
662		fs->fs_flags |= FS_UNCLEAN;
663		if (ronly || (mp->mnt_flag & MNT_FORCE) ||
664		    ((fs->fs_flags & FS_NEEDSFSCK) == 0 &&
665		     (fs->fs_flags & FS_DOSOFTDEP))) {
666			printf(
667"WARNING: %s was not properly dismounted\n",
668			    fs->fs_fsmnt);
669		} else {
670			printf(
671"WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
672			    fs->fs_fsmnt);
673			error = EPERM;
674			goto out;
675		}
676		if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) &&
677		    (mp->mnt_flag & MNT_FORCE)) {
678			printf("%s: lost blocks %jd files %d\n", fs->fs_fsmnt,
679			    (intmax_t)fs->fs_pendingblocks,
680			    fs->fs_pendinginodes);
681			fs->fs_pendingblocks = 0;
682			fs->fs_pendinginodes = 0;
683		}
684	}
685	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
686		printf("%s: mount pending error: blocks %jd files %d\n",
687		    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
688		    fs->fs_pendinginodes);
689		fs->fs_pendingblocks = 0;
690		fs->fs_pendinginodes = 0;
691	}
692	if ((fs->fs_flags & FS_GJOURNAL) != 0) {
693#ifdef UFS_GJOURNAL
694		/*
695		 * Get journal provider name.
696		 */
697		size = 1024;
698		mp->mnt_gjprovider = malloc(size, M_UFSMNT, M_WAITOK);
699		if (g_io_getattr("GJOURNAL::provider", cp, &size,
700		    mp->mnt_gjprovider) == 0) {
701			mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, size,
702			    M_UFSMNT, M_WAITOK);
703			MNT_ILOCK(mp);
704			mp->mnt_flag |= MNT_GJOURNAL;
705			MNT_IUNLOCK(mp);
706		} else {
707			printf(
708"WARNING: %s: GJOURNAL flag on fs but no gjournal provider below\n",
709			    mp->mnt_stat.f_mntonname);
710			free(mp->mnt_gjprovider, M_UFSMNT);
711			mp->mnt_gjprovider = NULL;
712		}
713#else
714		printf(
715"WARNING: %s: GJOURNAL flag on fs but no UFS_GJOURNAL support\n",
716		    mp->mnt_stat.f_mntonname);
717#endif
718	} else {
719		mp->mnt_gjprovider = NULL;
720	}
721	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
722	ump->um_cp = cp;
723	ump->um_bo = &devvp->v_bufobj;
724	ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT, M_WAITOK);
725	if (fs->fs_magic == FS_UFS1_MAGIC) {
726		ump->um_fstype = UFS1;
727		ump->um_balloc = ffs_balloc_ufs1;
728	} else {
729		ump->um_fstype = UFS2;
730		ump->um_balloc = ffs_balloc_ufs2;
731	}
732	ump->um_blkatoff = ffs_blkatoff;
733	ump->um_truncate = ffs_truncate;
734	ump->um_update = ffs_update;
735	ump->um_valloc = ffs_valloc;
736	ump->um_vfree = ffs_vfree;
737	ump->um_ifree = ffs_ifree;
738	mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF);
739	bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize);
740	if (fs->fs_sbsize < SBLOCKSIZE)
741		bp->b_flags |= B_INVAL | B_NOCACHE;
742	brelse(bp);
743	bp = NULL;
744	fs = ump->um_fs;
745	ffs_oldfscompat_read(fs, ump, sblockloc);
746	fs->fs_ronly = ronly;
747	size = fs->fs_cssize;
748	blks = howmany(size, fs->fs_fsize);
749	if (fs->fs_contigsumsize > 0)
750		size += fs->fs_ncg * sizeof(int32_t);
751	size += fs->fs_ncg * sizeof(u_int8_t);
752	space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
753	fs->fs_csp = space;
754	for (i = 0; i < blks; i += fs->fs_frag) {
755		size = fs->fs_bsize;
756		if (i + fs->fs_frag > blks)
757			size = (blks - i) * fs->fs_fsize;
758		if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
759		    cred, &bp)) != 0) {
760			free(fs->fs_csp, M_UFSMNT);
761			goto out;
762		}
763		bcopy(bp->b_data, space, (u_int)size);
764		space = (char *)space + size;
765		brelse(bp);
766		bp = NULL;
767	}
768	if (fs->fs_contigsumsize > 0) {
769		fs->fs_maxcluster = lp = space;
770		for (i = 0; i < fs->fs_ncg; i++)
771			*lp++ = fs->fs_contigsumsize;
772		space = lp;
773	}
774	size = fs->fs_ncg * sizeof(u_int8_t);
775	fs->fs_contigdirs = (u_int8_t *)space;
776	bzero(fs->fs_contigdirs, size);
777	fs->fs_active = NULL;
778	mp->mnt_data = ump;
779	mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
780	mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
781	nmp = NULL;
782	if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
783	    (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) {
784		if (nmp)
785			vfs_rel(nmp);
786		vfs_getnewfsid(mp);
787	}
788	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
789	MNT_ILOCK(mp);
790	mp->mnt_flag |= MNT_LOCAL;
791	MNT_IUNLOCK(mp);
792	if ((fs->fs_flags & FS_MULTILABEL) != 0) {
793#ifdef MAC
794		MNT_ILOCK(mp);
795		mp->mnt_flag |= MNT_MULTILABEL;
796		MNT_IUNLOCK(mp);
797#else
798		printf(
799"WARNING: %s: multilabel flag on fs but no MAC support\n",
800		    mp->mnt_stat.f_mntonname);
801#endif
802	}
803	if ((fs->fs_flags & FS_ACLS) != 0) {
804#ifdef UFS_ACL
805		MNT_ILOCK(mp);
806		mp->mnt_flag |= MNT_ACLS;
807		MNT_IUNLOCK(mp);
808#else
809		printf(
810"WARNING: %s: ACLs flag on fs but no ACLs support\n",
811		    mp->mnt_stat.f_mntonname);
812#endif
813	}
814	ump->um_mountp = mp;
815	ump->um_dev = dev;
816	ump->um_devvp = devvp;
817	ump->um_nindir = fs->fs_nindir;
818	ump->um_bptrtodb = fs->fs_fsbtodb;
819	ump->um_seqinc = fs->fs_frag;
820	for (i = 0; i < MAXQUOTAS; i++)
821		ump->um_quotas[i] = NULLVP;
822#ifdef UFS_EXTATTR
823	ufs_extattr_uepm_init(&ump->um_extattr);
824#endif
825	/*
826	 * Set FS local "last mounted on" information (NULL pad)
827	 */
828	bzero(fs->fs_fsmnt, MAXMNTLEN);
829	strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN);
830
831	if( mp->mnt_flag & MNT_ROOTFS) {
832		/*
833		 * Root mount; update timestamp in mount structure.
834		 * this will be used by the common root mount code
835		 * to update the system clock.
836		 */
837		mp->mnt_time = fs->fs_time;
838	}
839
840	if (ronly == 0) {
841		if ((fs->fs_flags & FS_DOSOFTDEP) &&
842		    (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
843			free(fs->fs_csp, M_UFSMNT);
844			goto out;
845		}
846		if (fs->fs_snapinum[0] != 0)
847			ffs_snapshot_mount(mp);
848		fs->fs_fmod = 1;
849		fs->fs_clean = 0;
850		(void) ffs_sbupdate(ump, MNT_WAIT, 0);
851	}
852	/*
853	 * Initialize filesystem stat information in mount struct.
854	 */
855	MNT_ILOCK(mp);
856	mp->mnt_kern_flag |= MNTK_MPSAFE;
857	MNT_IUNLOCK(mp);
858#ifdef UFS_EXTATTR
859#ifdef UFS_EXTATTR_AUTOSTART
860	/*
861	 *
862	 * Auto-starting does the following:
863	 *	- check for /.attribute in the fs, and extattr_start if so
864	 *	- for each file in .attribute, enable that file with
865	 * 	  an attribute of the same name.
866	 * Not clear how to report errors -- probably eat them.
867	 * This would all happen while the filesystem was busy/not
868	 * available, so would effectively be "atomic".
869	 */
870	mp->mnt_stat.f_iosize = fs->fs_bsize;
871	(void) ufs_extattr_autostart(mp, td);
872#endif /* !UFS_EXTATTR_AUTOSTART */
873#endif /* !UFS_EXTATTR */
874	return (0);
875out:
876	if (bp)
877		brelse(bp);
878	if (cp != NULL) {
879		DROP_GIANT();
880		g_topology_lock();
881		g_vfs_close(cp, td);
882		g_topology_unlock();
883		PICKUP_GIANT();
884	}
885	if (ump) {
886		mtx_destroy(UFS_MTX(ump));
887		if (mp->mnt_gjprovider != NULL) {
888			free(mp->mnt_gjprovider, M_UFSMNT);
889			mp->mnt_gjprovider = NULL;
890		}
891		free(ump->um_fs, M_UFSMNT);
892		free(ump, M_UFSMNT);
893		mp->mnt_data = NULL;
894	}
895	return (error);
896}
897
898#include <sys/sysctl.h>
899static int bigcgs = 0;
900SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, "");
901
902/*
903 * Sanity checks for loading old filesystem superblocks.
904 * See ffs_oldfscompat_write below for unwound actions.
905 *
906 * XXX - Parts get retired eventually.
907 * Unfortunately new bits get added.
908 */
909static void
910ffs_oldfscompat_read(fs, ump, sblockloc)
911	struct fs *fs;
912	struct ufsmount *ump;
913	ufs2_daddr_t sblockloc;
914{
915	off_t maxfilesize;
916
917	/*
918	 * If not yet done, update fs_flags location and value of fs_sblockloc.
919	 */
920	if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
921		fs->fs_flags = fs->fs_old_flags;
922		fs->fs_old_flags |= FS_FLAGS_UPDATED;
923		fs->fs_sblockloc = sblockloc;
924	}
925	/*
926	 * If not yet done, update UFS1 superblock with new wider fields.
927	 */
928	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) {
929		fs->fs_maxbsize = fs->fs_bsize;
930		fs->fs_time = fs->fs_old_time;
931		fs->fs_size = fs->fs_old_size;
932		fs->fs_dsize = fs->fs_old_dsize;
933		fs->fs_csaddr = fs->fs_old_csaddr;
934		fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir;
935		fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree;
936		fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree;
937		fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree;
938	}
939	if (fs->fs_magic == FS_UFS1_MAGIC &&
940	    fs->fs_old_inodefmt < FS_44INODEFMT) {
941		fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1;
942		fs->fs_qbmask = ~fs->fs_bmask;
943		fs->fs_qfmask = ~fs->fs_fmask;
944	}
945	if (fs->fs_magic == FS_UFS1_MAGIC) {
946		ump->um_savedmaxfilesize = fs->fs_maxfilesize;
947		maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1;
948		if (fs->fs_maxfilesize > maxfilesize)
949			fs->fs_maxfilesize = maxfilesize;
950	}
951	/* Compatibility for old filesystems */
952	if (fs->fs_avgfilesize <= 0)
953		fs->fs_avgfilesize = AVFILESIZ;
954	if (fs->fs_avgfpdir <= 0)
955		fs->fs_avgfpdir = AFPDIR;
956	if (bigcgs) {
957		fs->fs_save_cgsize = fs->fs_cgsize;
958		fs->fs_cgsize = fs->fs_bsize;
959	}
960}
961
962/*
963 * Unwinding superblock updates for old filesystems.
964 * See ffs_oldfscompat_read above for details.
965 *
966 * XXX - Parts get retired eventually.
967 * Unfortunately new bits get added.
968 */
969static void
970ffs_oldfscompat_write(fs, ump)
971	struct fs *fs;
972	struct ufsmount *ump;
973{
974
975	/*
976	 * Copy back UFS2 updated fields that UFS1 inspects.
977	 */
978	if (fs->fs_magic == FS_UFS1_MAGIC) {
979		fs->fs_old_time = fs->fs_time;
980		fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir;
981		fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree;
982		fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree;
983		fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree;
984		fs->fs_maxfilesize = ump->um_savedmaxfilesize;
985	}
986	if (bigcgs) {
987		fs->fs_cgsize = fs->fs_save_cgsize;
988		fs->fs_save_cgsize = 0;
989	}
990}
991
992/*
993 * unmount system call
994 */
995static int
996ffs_unmount(mp, mntflags, td)
997	struct mount *mp;
998	int mntflags;
999	struct thread *td;
1000{
1001	struct ufsmount *ump = VFSTOUFS(mp);
1002	struct fs *fs;
1003	int error, flags;
1004
1005	flags = 0;
1006	if (mntflags & MNT_FORCE) {
1007		flags |= FORCECLOSE;
1008	}
1009#ifdef UFS_EXTATTR
1010	if ((error = ufs_extattr_stop(mp, td))) {
1011		if (error != EOPNOTSUPP)
1012			printf("ffs_unmount: ufs_extattr_stop returned %d\n",
1013			    error);
1014	} else {
1015		ufs_extattr_uepm_destroy(&ump->um_extattr);
1016	}
1017#endif
1018	if (mp->mnt_flag & MNT_SOFTDEP) {
1019		if ((error = softdep_flushfiles(mp, flags, td)) != 0)
1020			return (error);
1021	} else {
1022		if ((error = ffs_flushfiles(mp, flags, td)) != 0)
1023			return (error);
1024	}
1025	fs = ump->um_fs;
1026	UFS_LOCK(ump);
1027	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
1028		printf("%s: unmount pending error: blocks %jd files %d\n",
1029		    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
1030		    fs->fs_pendinginodes);
1031		fs->fs_pendingblocks = 0;
1032		fs->fs_pendinginodes = 0;
1033	}
1034	UFS_UNLOCK(ump);
1035	if (fs->fs_ronly == 0) {
1036		fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1;
1037		error = ffs_sbupdate(ump, MNT_WAIT, 0);
1038		if (error) {
1039			fs->fs_clean = 0;
1040			return (error);
1041		}
1042	}
1043	DROP_GIANT();
1044	g_topology_lock();
1045	g_vfs_close(ump->um_cp, td);
1046	g_topology_unlock();
1047	PICKUP_GIANT();
1048	vrele(ump->um_devvp);
1049	mtx_destroy(UFS_MTX(ump));
1050	if (mp->mnt_gjprovider != NULL) {
1051		free(mp->mnt_gjprovider, M_UFSMNT);
1052		mp->mnt_gjprovider = NULL;
1053	}
1054	free(fs->fs_csp, M_UFSMNT);
1055	free(fs, M_UFSMNT);
1056	free(ump, M_UFSMNT);
1057	mp->mnt_data = NULL;
1058	MNT_ILOCK(mp);
1059	mp->mnt_flag &= ~MNT_LOCAL;
1060	MNT_IUNLOCK(mp);
1061	return (error);
1062}
1063
1064/*
1065 * Flush out all the files in a filesystem.
1066 */
1067int
1068ffs_flushfiles(mp, flags, td)
1069	struct mount *mp;
1070	int flags;
1071	struct thread *td;
1072{
1073	struct ufsmount *ump;
1074	int error;
1075
1076	ump = VFSTOUFS(mp);
1077#ifdef QUOTA
1078	if (mp->mnt_flag & MNT_QUOTA) {
1079		int i;
1080		error = vflush(mp, 0, SKIPSYSTEM|flags, td);
1081		if (error)
1082			return (error);
1083		for (i = 0; i < MAXQUOTAS; i++) {
1084			quotaoff(td, mp, i);
1085		}
1086		/*
1087		 * Here we fall through to vflush again to ensure
1088		 * that we have gotten rid of all the system vnodes.
1089		 */
1090	}
1091#endif
1092	ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles");
1093	if (ump->um_devvp->v_vflag & VV_COPYONWRITE) {
1094		if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0)
1095			return (error);
1096		ffs_snapshot_unmount(mp);
1097		flags |= FORCECLOSE;
1098		/*
1099		 * Here we fall through to vflush again to ensure
1100		 * that we have gotten rid of all the system vnodes.
1101		 */
1102	}
1103        /*
1104	 * Flush all the files.
1105	 */
1106	if ((error = vflush(mp, 0, flags, td)) != 0)
1107		return (error);
1108	/*
1109	 * Flush filesystem metadata.
1110	 */
1111	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1112	error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td);
1113	VOP_UNLOCK(ump->um_devvp, 0);
1114	return (error);
1115}
1116
1117/*
1118 * Get filesystem statistics.
1119 */
1120static int
1121ffs_statfs(mp, sbp, td)
1122	struct mount *mp;
1123	struct statfs *sbp;
1124	struct thread *td;
1125{
1126	struct ufsmount *ump;
1127	struct fs *fs;
1128
1129	ump = VFSTOUFS(mp);
1130	fs = ump->um_fs;
1131	if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC)
1132		panic("ffs_statfs");
1133	sbp->f_version = STATFS_VERSION;
1134	sbp->f_bsize = fs->fs_fsize;
1135	sbp->f_iosize = fs->fs_bsize;
1136	sbp->f_blocks = fs->fs_dsize;
1137	UFS_LOCK(ump);
1138	sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
1139	    fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
1140	sbp->f_bavail = freespace(fs, fs->fs_minfree) +
1141	    dbtofsb(fs, fs->fs_pendingblocks);
1142	sbp->f_files =  fs->fs_ncg * fs->fs_ipg - ROOTINO;
1143	sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
1144	UFS_UNLOCK(ump);
1145	sbp->f_namemax = NAME_MAX;
1146	return (0);
1147}
1148
1149/*
1150 * Go through the disk queues to initiate sandbagged IO;
1151 * go through the inodes to write those that have been modified;
1152 * initiate the writing of the super block if it has been modified.
1153 *
1154 * Note: we are always called with the filesystem marked `MPBUSY'.
1155 */
1156static int
1157ffs_sync(mp, waitfor, td)
1158	struct mount *mp;
1159	int waitfor;
1160	struct thread *td;
1161{
1162	struct vnode *mvp, *vp, *devvp;
1163	struct inode *ip;
1164	struct ufsmount *ump = VFSTOUFS(mp);
1165	struct fs *fs;
1166	int error, count, wait, lockreq, allerror = 0;
1167	int suspend;
1168	int suspended;
1169	int secondary_writes;
1170	int secondary_accwrites;
1171	int softdep_deps;
1172	int softdep_accdeps;
1173	struct bufobj *bo;
1174
1175	fs = ump->um_fs;
1176	if (fs->fs_fmod != 0 && fs->fs_ronly != 0) {		/* XXX */
1177		printf("fs = %s\n", fs->fs_fsmnt);
1178		panic("ffs_sync: rofs mod");
1179	}
1180	/*
1181	 * Write back each (modified) inode.
1182	 */
1183	wait = 0;
1184	suspend = 0;
1185	suspended = 0;
1186	lockreq = LK_EXCLUSIVE | LK_NOWAIT;
1187	if (waitfor == MNT_SUSPEND) {
1188		suspend = 1;
1189		waitfor = MNT_WAIT;
1190	}
1191	if (waitfor == MNT_WAIT) {
1192		wait = 1;
1193		lockreq = LK_EXCLUSIVE;
1194	}
1195	lockreq |= LK_INTERLOCK | LK_SLEEPFAIL;
1196	MNT_ILOCK(mp);
1197loop:
1198	/* Grab snapshot of secondary write counts */
1199	secondary_writes = mp->mnt_secondary_writes;
1200	secondary_accwrites = mp->mnt_secondary_accwrites;
1201
1202	/* Grab snapshot of softdep dependency counts */
1203	MNT_IUNLOCK(mp);
1204	softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps);
1205	MNT_ILOCK(mp);
1206
1207	MNT_VNODE_FOREACH(vp, mp, mvp) {
1208		/*
1209		 * Depend on the mntvnode_slock to keep things stable enough
1210		 * for a quick test.  Since there might be hundreds of
1211		 * thousands of vnodes, we cannot afford even a subroutine
1212		 * call unless there's a good chance that we have work to do.
1213		 */
1214		VI_LOCK(vp);
1215		if (vp->v_iflag & VI_DOOMED) {
1216			VI_UNLOCK(vp);
1217			continue;
1218		}
1219		ip = VTOI(vp);
1220		if (vp->v_type == VNON || ((ip->i_flag &
1221		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1222		    vp->v_bufobj.bo_dirty.bv_cnt == 0)) {
1223			VI_UNLOCK(vp);
1224			continue;
1225		}
1226		MNT_IUNLOCK(mp);
1227		if ((error = vget(vp, lockreq, td)) != 0) {
1228			MNT_ILOCK(mp);
1229			if (error == ENOENT || error == ENOLCK) {
1230				MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
1231				goto loop;
1232			}
1233			continue;
1234		}
1235		if ((error = ffs_syncvnode(vp, waitfor)) != 0)
1236			allerror = error;
1237		vput(vp);
1238		MNT_ILOCK(mp);
1239	}
1240	MNT_IUNLOCK(mp);
1241	/*
1242	 * Force stale filesystem control information to be flushed.
1243	 */
1244	if (waitfor == MNT_WAIT) {
1245		if ((error = softdep_flushworklist(ump->um_mountp, &count, td)))
1246			allerror = error;
1247		/* Flushed work items may create new vnodes to clean */
1248		if (allerror == 0 && count) {
1249			MNT_ILOCK(mp);
1250			goto loop;
1251		}
1252	}
1253#ifdef QUOTA
1254	qsync(mp);
1255#endif
1256	devvp = ump->um_devvp;
1257	bo = &devvp->v_bufobj;
1258	BO_LOCK(bo);
1259	if (waitfor != MNT_LAZY &&
1260	    (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) {
1261		BO_UNLOCK(bo);
1262		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1263		if ((error = VOP_FSYNC(devvp, waitfor, td)) != 0)
1264			allerror = error;
1265		VOP_UNLOCK(devvp, 0);
1266		if (allerror == 0 && waitfor == MNT_WAIT) {
1267			MNT_ILOCK(mp);
1268			goto loop;
1269		}
1270	} else if (suspend != 0) {
1271		if (softdep_check_suspend(mp,
1272					  devvp,
1273					  softdep_deps,
1274					  softdep_accdeps,
1275					  secondary_writes,
1276					  secondary_accwrites) != 0)
1277			goto loop;	/* More work needed */
1278		mtx_assert(MNT_MTX(mp), MA_OWNED);
1279		mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
1280		MNT_IUNLOCK(mp);
1281		suspended = 1;
1282	} else
1283		BO_UNLOCK(bo);
1284	/*
1285	 * Write back modified superblock.
1286	 */
1287	if (fs->fs_fmod != 0 &&
1288	    (error = ffs_sbupdate(ump, waitfor, suspended)) != 0)
1289		allerror = error;
1290	return (allerror);
1291}
1292
1293int
1294ffs_vget(mp, ino, flags, vpp)
1295	struct mount *mp;
1296	ino_t ino;
1297	int flags;
1298	struct vnode **vpp;
1299{
1300	struct fs *fs;
1301	struct inode *ip;
1302	struct ufsmount *ump;
1303	struct buf *bp;
1304	struct vnode *vp;
1305	struct cdev *dev;
1306	int error;
1307
1308	error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL);
1309	if (error || *vpp != NULL)
1310		return (error);
1311
1312	/*
1313	 * We must promote to an exclusive lock for vnode creation.  This
1314	 * can happen if lookup is passed LOCKSHARED.
1315 	 */
1316	if ((flags & LK_TYPE_MASK) == LK_SHARED) {
1317		flags &= ~LK_TYPE_MASK;
1318		flags |= LK_EXCLUSIVE;
1319	}
1320
1321	/*
1322	 * We do not lock vnode creation as it is believed to be too
1323	 * expensive for such rare case as simultaneous creation of vnode
1324	 * for same ino by different processes. We just allow them to race
1325	 * and check later to decide who wins. Let the race begin!
1326	 */
1327
1328	ump = VFSTOUFS(mp);
1329	dev = ump->um_dev;
1330	fs = ump->um_fs;
1331
1332	/*
1333	 * If this MALLOC() is performed after the getnewvnode()
1334	 * it might block, leaving a vnode with a NULL v_data to be
1335	 * found by ffs_sync() if a sync happens to fire right then,
1336	 * which will cause a panic because ffs_sync() blindly
1337	 * dereferences vp->v_data (as well it should).
1338	 */
1339	ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO);
1340
1341	/* Allocate a new vnode/inode. */
1342	if (fs->fs_magic == FS_UFS1_MAGIC)
1343		error = getnewvnode("ufs", mp, &ffs_vnodeops1, &vp);
1344	else
1345		error = getnewvnode("ufs", mp, &ffs_vnodeops2, &vp);
1346	if (error) {
1347		*vpp = NULL;
1348		uma_zfree(uma_inode, ip);
1349		return (error);
1350	}
1351	/*
1352	 * FFS supports recursive and shared locking.
1353	 */
1354	VN_LOCK_AREC(vp);
1355	VN_LOCK_ASHARE(vp);
1356	vp->v_data = ip;
1357	vp->v_bufobj.bo_bsize = fs->fs_bsize;
1358	ip->i_vnode = vp;
1359	ip->i_ump = ump;
1360	ip->i_fs = fs;
1361	ip->i_dev = dev;
1362	ip->i_number = ino;
1363#ifdef QUOTA
1364	{
1365		int i;
1366		for (i = 0; i < MAXQUOTAS; i++)
1367			ip->i_dquot[i] = NODQUOT;
1368	}
1369#endif
1370
1371	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
1372	error = insmntque(vp, mp);
1373	if (error != 0) {
1374		uma_zfree(uma_inode, ip);
1375		*vpp = NULL;
1376		return (error);
1377	}
1378	error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL);
1379	if (error || *vpp != NULL)
1380		return (error);
1381
1382	/* Read in the disk contents for the inode, copy into the inode. */
1383	error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1384	    (int)fs->fs_bsize, NOCRED, &bp);
1385	if (error) {
1386		/*
1387		 * The inode does not contain anything useful, so it would
1388		 * be misleading to leave it on its hash chain. With mode
1389		 * still zero, it will be unlinked and returned to the free
1390		 * list by vput().
1391		 */
1392		brelse(bp);
1393		vput(vp);
1394		*vpp = NULL;
1395		return (error);
1396	}
1397	if (ip->i_ump->um_fstype == UFS1)
1398		ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK);
1399	else
1400		ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK);
1401	ffs_load_inode(bp, ip, fs, ino);
1402	if (DOINGSOFTDEP(vp))
1403		softdep_load_inodeblock(ip);
1404	else
1405		ip->i_effnlink = ip->i_nlink;
1406	bqrelse(bp);
1407
1408	/*
1409	 * Initialize the vnode from the inode, check for aliases.
1410	 * Note that the underlying vnode may have changed.
1411	 */
1412	if (ip->i_ump->um_fstype == UFS1)
1413		error = ufs_vinit(mp, &ffs_fifoops1, &vp);
1414	else
1415		error = ufs_vinit(mp, &ffs_fifoops2, &vp);
1416	if (error) {
1417		vput(vp);
1418		*vpp = NULL;
1419		return (error);
1420	}
1421
1422	/*
1423	 * Finish inode initialization.
1424	 */
1425
1426	/*
1427	 * Set up a generation number for this inode if it does not
1428	 * already have one. This should only happen on old filesystems.
1429	 */
1430	if (ip->i_gen == 0) {
1431		ip->i_gen = arc4random() / 2 + 1;
1432		if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
1433			ip->i_flag |= IN_MODIFIED;
1434			DIP_SET(ip, i_gen, ip->i_gen);
1435		}
1436	}
1437	/*
1438	 * Ensure that uid and gid are correct. This is a temporary
1439	 * fix until fsck has been changed to do the update.
1440	 */
1441	if (fs->fs_magic == FS_UFS1_MAGIC &&		/* XXX */
1442	    fs->fs_old_inodefmt < FS_44INODEFMT) {	/* XXX */
1443		ip->i_uid = ip->i_din1->di_ouid;	/* XXX */
1444		ip->i_gid = ip->i_din1->di_ogid;	/* XXX */
1445	}						/* XXX */
1446
1447#ifdef MAC
1448	if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) {
1449		/*
1450		 * If this vnode is already allocated, and we're running
1451		 * multi-label, attempt to perform a label association
1452		 * from the extended attributes on the inode.
1453		 */
1454		error = mac_vnode_associate_extattr(mp, vp);
1455		if (error) {
1456			/* ufs_inactive will release ip->i_devvp ref. */
1457			vput(vp);
1458			*vpp = NULL;
1459			return (error);
1460		}
1461	}
1462#endif
1463
1464	*vpp = vp;
1465	return (0);
1466}
1467
1468/*
1469 * File handle to vnode
1470 *
1471 * Have to be really careful about stale file handles:
1472 * - check that the inode number is valid
1473 * - call ffs_vget() to get the locked inode
1474 * - check for an unallocated inode (i_mode == 0)
1475 * - check that the given client host has export rights and return
1476 *   those rights via. exflagsp and credanonp
1477 */
1478static int
1479ffs_fhtovp(mp, fhp, vpp)
1480	struct mount *mp;
1481	struct fid *fhp;
1482	struct vnode **vpp;
1483{
1484	struct ufid *ufhp;
1485	struct fs *fs;
1486
1487	ufhp = (struct ufid *)fhp;
1488	fs = VFSTOUFS(mp)->um_fs;
1489	if (ufhp->ufid_ino < ROOTINO ||
1490	    ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg)
1491		return (ESTALE);
1492	return (ufs_fhtovp(mp, ufhp, vpp));
1493}
1494
1495/*
1496 * Initialize the filesystem.
1497 */
1498static int
1499ffs_init(vfsp)
1500	struct vfsconf *vfsp;
1501{
1502
1503	softdep_initialize();
1504	return (ufs_init(vfsp));
1505}
1506
1507/*
1508 * Undo the work of ffs_init().
1509 */
1510static int
1511ffs_uninit(vfsp)
1512	struct vfsconf *vfsp;
1513{
1514	int ret;
1515
1516	ret = ufs_uninit(vfsp);
1517	softdep_uninitialize();
1518	return (ret);
1519}
1520
1521/*
1522 * Write a superblock and associated information back to disk.
1523 */
1524int
1525ffs_sbupdate(mp, waitfor, suspended)
1526	struct ufsmount *mp;
1527	int waitfor;
1528	int suspended;
1529{
1530	struct fs *fs = mp->um_fs;
1531	struct buf *sbbp;
1532	struct buf *bp;
1533	int blks;
1534	void *space;
1535	int i, size, error, allerror = 0;
1536
1537	if (fs->fs_ronly == 1 &&
1538	    (mp->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) !=
1539	    (MNT_RDONLY | MNT_UPDATE))
1540		panic("ffs_sbupdate: write read-only filesystem");
1541	/*
1542	 * We use the superblock's buf to serialize calls to ffs_sbupdate().
1543	 */
1544	sbbp = getblk(mp->um_devvp, btodb(fs->fs_sblockloc), (int)fs->fs_sbsize,
1545	    0, 0, 0);
1546	/*
1547	 * First write back the summary information.
1548	 */
1549	blks = howmany(fs->fs_cssize, fs->fs_fsize);
1550	space = fs->fs_csp;
1551	for (i = 0; i < blks; i += fs->fs_frag) {
1552		size = fs->fs_bsize;
1553		if (i + fs->fs_frag > blks)
1554			size = (blks - i) * fs->fs_fsize;
1555		bp = getblk(mp->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
1556		    size, 0, 0, 0);
1557		bcopy(space, bp->b_data, (u_int)size);
1558		space = (char *)space + size;
1559		if (suspended)
1560			bp->b_flags |= B_VALIDSUSPWRT;
1561		if (waitfor != MNT_WAIT)
1562			bawrite(bp);
1563		else if ((error = bwrite(bp)) != 0)
1564			allerror = error;
1565	}
1566	/*
1567	 * Now write back the superblock itself. If any errors occurred
1568	 * up to this point, then fail so that the superblock avoids
1569	 * being written out as clean.
1570	 */
1571	if (allerror) {
1572		brelse(sbbp);
1573		return (allerror);
1574	}
1575	bp = sbbp;
1576	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 &&
1577	    (fs->fs_flags & FS_FLAGS_UPDATED) == 0) {
1578		printf("%s: correcting fs_sblockloc from %jd to %d\n",
1579		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1);
1580		fs->fs_sblockloc = SBLOCK_UFS1;
1581	}
1582	if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 &&
1583	    (fs->fs_flags & FS_FLAGS_UPDATED) == 0) {
1584		printf("%s: correcting fs_sblockloc from %jd to %d\n",
1585		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2);
1586		fs->fs_sblockloc = SBLOCK_UFS2;
1587	}
1588	fs->fs_fmod = 0;
1589	fs->fs_time = time_second;
1590	bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
1591	ffs_oldfscompat_write((struct fs *)bp->b_data, mp);
1592	if (suspended)
1593		bp->b_flags |= B_VALIDSUSPWRT;
1594	if (waitfor != MNT_WAIT)
1595		bawrite(bp);
1596	else if ((error = bwrite(bp)) != 0)
1597		allerror = error;
1598	return (allerror);
1599}
1600
1601static int
1602ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
1603	int attrnamespace, const char *attrname, struct thread *td)
1604{
1605
1606#ifdef UFS_EXTATTR
1607	return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace,
1608	    attrname, td));
1609#else
1610	return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace,
1611	    attrname, td));
1612#endif
1613}
1614
1615static void
1616ffs_ifree(struct ufsmount *ump, struct inode *ip)
1617{
1618
1619	if (ump->um_fstype == UFS1 && ip->i_din1 != NULL)
1620		uma_zfree(uma_ufs1, ip->i_din1);
1621	else if (ip->i_din2 != NULL)
1622		uma_zfree(uma_ufs2, ip->i_din2);
1623	uma_zfree(uma_inode, ip);
1624}
1625
1626static int dobkgrdwrite = 1;
1627SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0,
1628    "Do background writes (honoring the BV_BKGRDWRITE flag)?");
1629
1630/*
1631 * Complete a background write started from bwrite.
1632 */
1633static void
1634ffs_backgroundwritedone(struct buf *bp)
1635{
1636	struct bufobj *bufobj;
1637	struct buf *origbp;
1638
1639	/*
1640	 * Find the original buffer that we are writing.
1641	 */
1642	bufobj = bp->b_bufobj;
1643	BO_LOCK(bufobj);
1644	if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL)
1645		panic("backgroundwritedone: lost buffer");
1646	/* Grab an extra reference to be dropped by the bufdone() below. */
1647	bufobj_wrefl(bufobj);
1648	BO_UNLOCK(bufobj);
1649	/*
1650	 * Process dependencies then return any unfinished ones.
1651	 */
1652	if (!LIST_EMPTY(&bp->b_dep))
1653		buf_complete(bp);
1654#ifdef SOFTUPDATES
1655	if (!LIST_EMPTY(&bp->b_dep))
1656		softdep_move_dependencies(bp, origbp);
1657#endif
1658	/*
1659	 * This buffer is marked B_NOCACHE so when it is released
1660	 * by biodone it will be tossed.
1661	 */
1662	bp->b_flags |= B_NOCACHE;
1663	bp->b_flags &= ~B_CACHE;
1664	bufdone(bp);
1665	BO_LOCK(bufobj);
1666	/*
1667	 * Clear the BV_BKGRDINPROG flag in the original buffer
1668	 * and awaken it if it is waiting for the write to complete.
1669	 * If BV_BKGRDINPROG is not set in the original buffer it must
1670	 * have been released and re-instantiated - which is not legal.
1671	 */
1672	KASSERT((origbp->b_vflags & BV_BKGRDINPROG),
1673	    ("backgroundwritedone: lost buffer2"));
1674	origbp->b_vflags &= ~BV_BKGRDINPROG;
1675	if (origbp->b_vflags & BV_BKGRDWAIT) {
1676		origbp->b_vflags &= ~BV_BKGRDWAIT;
1677		wakeup(&origbp->b_xflags);
1678	}
1679	BO_UNLOCK(bufobj);
1680}
1681
1682
1683/*
1684 * Write, release buffer on completion.  (Done by iodone
1685 * if async).  Do not bother writing anything if the buffer
1686 * is invalid.
1687 *
1688 * Note that we set B_CACHE here, indicating that buffer is
1689 * fully valid and thus cacheable.  This is true even of NFS
1690 * now so we set it generally.  This could be set either here
1691 * or in biodone() since the I/O is synchronous.  We put it
1692 * here.
1693 */
1694static int
1695ffs_bufwrite(struct buf *bp)
1696{
1697	int oldflags, s;
1698	struct buf *newbp;
1699
1700	CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1701	if (bp->b_flags & B_INVAL) {
1702		brelse(bp);
1703		return (0);
1704	}
1705
1706	oldflags = bp->b_flags;
1707
1708	if (!BUF_ISLOCKED(bp))
1709		panic("bufwrite: buffer is not busy???");
1710	s = splbio();
1711	/*
1712	 * If a background write is already in progress, delay
1713	 * writing this block if it is asynchronous. Otherwise
1714	 * wait for the background write to complete.
1715	 */
1716	BO_LOCK(bp->b_bufobj);
1717	if (bp->b_vflags & BV_BKGRDINPROG) {
1718		if (bp->b_flags & B_ASYNC) {
1719			BO_UNLOCK(bp->b_bufobj);
1720			splx(s);
1721			bdwrite(bp);
1722			return (0);
1723		}
1724		bp->b_vflags |= BV_BKGRDWAIT;
1725		msleep(&bp->b_xflags, BO_MTX(bp->b_bufobj), PRIBIO, "bwrbg", 0);
1726		if (bp->b_vflags & BV_BKGRDINPROG)
1727			panic("bufwrite: still writing");
1728	}
1729	BO_UNLOCK(bp->b_bufobj);
1730
1731	/* Mark the buffer clean */
1732	bundirty(bp);
1733
1734	/*
1735	 * If this buffer is marked for background writing and we
1736	 * do not have to wait for it, make a copy and write the
1737	 * copy so as to leave this buffer ready for further use.
1738	 *
1739	 * This optimization eats a lot of memory.  If we have a page
1740	 * or buffer shortfall we can't do it.
1741	 */
1742	if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) &&
1743	    (bp->b_flags & B_ASYNC) &&
1744	    !vm_page_count_severe() &&
1745	    !buf_dirty_count_severe()) {
1746		KASSERT(bp->b_iodone == NULL,
1747		    ("bufwrite: needs chained iodone (%p)", bp->b_iodone));
1748
1749		/* get a new block */
1750		newbp = geteblk(bp->b_bufsize);
1751
1752		/*
1753		 * set it to be identical to the old block.  We have to
1754		 * set b_lblkno and BKGRDMARKER before calling bgetvp()
1755		 * to avoid confusing the splay tree and gbincore().
1756		 */
1757		memcpy(newbp->b_data, bp->b_data, bp->b_bufsize);
1758		newbp->b_lblkno = bp->b_lblkno;
1759		newbp->b_xflags |= BX_BKGRDMARKER;
1760		BO_LOCK(bp->b_bufobj);
1761		bp->b_vflags |= BV_BKGRDINPROG;
1762		bgetvp(bp->b_vp, newbp);
1763		BO_UNLOCK(bp->b_bufobj);
1764		newbp->b_bufobj = &bp->b_vp->v_bufobj;
1765		newbp->b_blkno = bp->b_blkno;
1766		newbp->b_offset = bp->b_offset;
1767		newbp->b_iodone = ffs_backgroundwritedone;
1768		newbp->b_flags |= B_ASYNC;
1769		newbp->b_flags &= ~B_INVAL;
1770
1771#ifdef SOFTUPDATES
1772		/* move over the dependencies */
1773		if (!LIST_EMPTY(&bp->b_dep))
1774			softdep_move_dependencies(bp, newbp);
1775#endif
1776
1777		/*
1778		 * Initiate write on the copy, release the original to
1779		 * the B_LOCKED queue so that it cannot go away until
1780		 * the background write completes. If not locked it could go
1781		 * away and then be reconstituted while it was being written.
1782		 * If the reconstituted buffer were written, we could end up
1783		 * with two background copies being written at the same time.
1784		 */
1785		bqrelse(bp);
1786		bp = newbp;
1787	}
1788
1789	/* Let the normal bufwrite do the rest for us */
1790	return (bufwrite(bp));
1791}
1792
1793
1794static void
1795ffs_geom_strategy(struct bufobj *bo, struct buf *bp)
1796{
1797	struct vnode *vp;
1798	int error;
1799	struct buf *tbp;
1800
1801	vp = bo->__bo_vnode;
1802	if (bp->b_iocmd == BIO_WRITE) {
1803		if ((bp->b_flags & B_VALIDSUSPWRT) == 0 &&
1804		    bp->b_vp != NULL && bp->b_vp->v_mount != NULL &&
1805		    (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0)
1806			panic("ffs_geom_strategy: bad I/O");
1807		bp->b_flags &= ~B_VALIDSUSPWRT;
1808		if ((vp->v_vflag & VV_COPYONWRITE) &&
1809		    vp->v_rdev->si_snapdata != NULL) {
1810			if ((bp->b_flags & B_CLUSTER) != 0) {
1811				runningbufwakeup(bp);
1812				TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
1813					      b_cluster.cluster_entry) {
1814					error = ffs_copyonwrite(vp, tbp);
1815					if (error != 0 &&
1816					    error != EOPNOTSUPP) {
1817						bp->b_error = error;
1818						bp->b_ioflags |= BIO_ERROR;
1819						bufdone(bp);
1820						return;
1821					}
1822				}
1823				bp->b_runningbufspace = bp->b_bufsize;
1824				atomic_add_int(&runningbufspace,
1825					       bp->b_runningbufspace);
1826			} else {
1827				error = ffs_copyonwrite(vp, bp);
1828				if (error != 0 && error != EOPNOTSUPP) {
1829					bp->b_error = error;
1830					bp->b_ioflags |= BIO_ERROR;
1831					bufdone(bp);
1832					return;
1833				}
1834			}
1835		}
1836#ifdef SOFTUPDATES
1837		if ((bp->b_flags & B_CLUSTER) != 0) {
1838			TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
1839				      b_cluster.cluster_entry) {
1840				if (!LIST_EMPTY(&tbp->b_dep))
1841					buf_start(tbp);
1842			}
1843		} else {
1844			if (!LIST_EMPTY(&bp->b_dep))
1845				buf_start(bp);
1846		}
1847
1848#endif
1849	}
1850	g_vfs_strategy(bo, bp);
1851}
1852