xref: /titanic_44/usr/src/uts/common/fs/ufs/lufs.c (revision 7c478bd95313f5f23a4c958a745db2134aa03244)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #include <sys/systm.h>
30 #include <sys/types.h>
31 #include <sys/vnode.h>
32 #include <sys/buf.h>
33 #include <sys/errno.h>
34 #include <sys/fssnap_if.h>
35 #include <sys/fs/ufs_inode.h>
36 #include <sys/fs/ufs_filio.h>
37 #include <sys/sysmacros.h>
38 #include <sys/modctl.h>
39 #include <sys/fs/ufs_log.h>
40 #include <sys/fs/ufs_bio.h>
41 #include <sys/fs/ufs_fsdir.h>
42 #include <sys/debug.h>
43 #include <sys/kmem.h>
44 #include <sys/inttypes.h>
45 #include <sys/vfs.h>
46 #include <sys/mntent.h>
47 #include <sys/conf.h>
48 #include <sys/param.h>
49 #include <sys/kstat.h>
50 #include <sys/cmn_err.h>
51 
52 static kmutex_t	log_mutex;	/* general purpose log layer lock */
53 kmutex_t	ml_scan;	/* Scan thread syncronization */
54 kcondvar_t	ml_scan_cv;	/* Scan thread syncronization */
55 
56 struct kmem_cache	*lufs_sv;
57 struct kmem_cache	*lufs_bp;
58 
59 /* Tunables */
60 uint_t		ldl_maxlogsize	= LDL_MAXLOGSIZE;
61 uint_t		ldl_minlogsize	= LDL_MINLOGSIZE;
62 uint32_t	ldl_divisor	= LDL_DIVISOR;
63 uint32_t	ldl_mintransfer	= LDL_MINTRANSFER;
64 uint32_t	ldl_maxtransfer	= LDL_MAXTRANSFER;
65 uint32_t	ldl_minbufsize	= LDL_MINBUFSIZE;
66 
67 uint32_t	last_loghead_ident = 0;
68 
69 /*
70  * Logging delta and roll statistics
71  */
72 struct delta_kstats {
73 	kstat_named_t ds_superblock_deltas;
74 	kstat_named_t ds_bitmap_deltas;
75 	kstat_named_t ds_suminfo_deltas;
76 	kstat_named_t ds_allocblk_deltas;
77 	kstat_named_t ds_ab0_deltas;
78 	kstat_named_t ds_dir_deltas;
79 	kstat_named_t ds_inode_deltas;
80 	kstat_named_t ds_fbiwrite_deltas;
81 	kstat_named_t ds_quota_deltas;
82 	kstat_named_t ds_shadow_deltas;
83 
84 	kstat_named_t ds_superblock_rolled;
85 	kstat_named_t ds_bitmap_rolled;
86 	kstat_named_t ds_suminfo_rolled;
87 	kstat_named_t ds_allocblk_rolled;
88 	kstat_named_t ds_ab0_rolled;
89 	kstat_named_t ds_dir_rolled;
90 	kstat_named_t ds_inode_rolled;
91 	kstat_named_t ds_fbiwrite_rolled;
92 	kstat_named_t ds_quota_rolled;
93 	kstat_named_t ds_shadow_rolled;
94 } dkstats = {
95 	{ "superblock_deltas",	KSTAT_DATA_UINT64 },
96 	{ "bitmap_deltas",	KSTAT_DATA_UINT64 },
97 	{ "suminfo_deltas",	KSTAT_DATA_UINT64 },
98 	{ "allocblk_deltas",	KSTAT_DATA_UINT64 },
99 	{ "ab0_deltas",		KSTAT_DATA_UINT64 },
100 	{ "dir_deltas",		KSTAT_DATA_UINT64 },
101 	{ "inode_deltas",	KSTAT_DATA_UINT64 },
102 	{ "fbiwrite_deltas",	KSTAT_DATA_UINT64 },
103 	{ "quota_deltas",	KSTAT_DATA_UINT64 },
104 	{ "shadow_deltas",	KSTAT_DATA_UINT64 },
105 
106 	{ "superblock_rolled",	KSTAT_DATA_UINT64 },
107 	{ "bitmap_rolled",	KSTAT_DATA_UINT64 },
108 	{ "suminfo_rolled",	KSTAT_DATA_UINT64 },
109 	{ "allocblk_rolled",	KSTAT_DATA_UINT64 },
110 	{ "ab0_rolled",		KSTAT_DATA_UINT64 },
111 	{ "dir_rolled",		KSTAT_DATA_UINT64 },
112 	{ "inode_rolled",	KSTAT_DATA_UINT64 },
113 	{ "fbiwrite_rolled",	KSTAT_DATA_UINT64 },
114 	{ "quota_rolled",	KSTAT_DATA_UINT64 },
115 	{ "shadow_rolled",	KSTAT_DATA_UINT64 }
116 };
117 
118 uint64_t delta_stats[DT_MAX];
119 uint64_t roll_stats[DT_MAX];
120 
121 /*
122  * General logging kstats
123  */
124 struct logstats logstats = {
125 	{ "master_reads",		KSTAT_DATA_UINT64 },
126 	{ "master_writes",		KSTAT_DATA_UINT64 },
127 	{ "log_reads_inmem",		KSTAT_DATA_UINT64 },
128 	{ "log_reads",			KSTAT_DATA_UINT64 },
129 	{ "log_writes",			KSTAT_DATA_UINT64 },
130 	{ "log_master_reads",		KSTAT_DATA_UINT64 },
131 	{ "log_roll_reads",		KSTAT_DATA_UINT64 },
132 	{ "log_roll_writes",		KSTAT_DATA_UINT64 }
133 };
134 
135 int
136 trans_not_done(struct buf *cb)
137 {
138 	sema_v(&cb->b_io);
139 	return (0);
140 }
141 
142 static void
143 trans_wait_panic(struct buf *cb)
144 {
145 	while ((cb->b_flags & B_DONE) == 0)
146 		drv_usecwait(10);
147 }
148 
149 int
150 trans_not_wait(struct buf *cb)
151 {
152 	/*
153 	 * In case of panic, busy wait for completion
154 	 */
155 	if (panicstr)
156 		trans_wait_panic(cb);
157 	else
158 		sema_p(&cb->b_io);
159 
160 	return (geterror(cb));
161 }
162 
163 int
164 trans_wait(struct buf *cb)
165 {
166 	/*
167 	 * In case of panic, busy wait for completion and run md daemon queues
168 	 */
169 	if (panicstr)
170 		trans_wait_panic(cb);
171 	return (biowait(cb));
172 }
173 
174 static void
175 setsum(int32_t *sp, int32_t *lp, int nb)
176 {
177 	int32_t csum = 0;
178 
179 	*sp = 0;
180 	nb /= sizeof (int32_t);
181 	while (nb--)
182 		csum += *lp++;
183 	*sp = csum;
184 }
185 
186 static int
187 checksum(int32_t *sp, int32_t *lp, int nb)
188 {
189 	int32_t ssum = *sp;
190 
191 	setsum(sp, lp, nb);
192 	if (ssum != *sp) {
193 		*sp = ssum;
194 		return (0);
195 	}
196 	return (1);
197 }
198 
199 void
200 lufs_unsnarf(ufsvfs_t *ufsvfsp)
201 {
202 	ml_unit_t *ul;
203 	mt_map_t *mtm;
204 
205 	ul = ufsvfsp->vfs_log;
206 	if (ul == NULL)
207 		return;
208 
209 	mtm = ul->un_logmap;
210 
211 	/*
212 	 * Wait for a pending top_issue_sync which is
213 	 * dispatched (via taskq_dispatch()) but hasnt completed yet.
214 	 */
215 
216 	mutex_enter(&mtm->mtm_lock);
217 
218 	while (mtm->mtm_taskq_sync_count != 0) {
219 		cv_wait(&mtm->mtm_cv, &mtm->mtm_lock);
220 	}
221 
222 	mutex_exit(&mtm->mtm_lock);
223 
224 	/* Roll committed transactions */
225 	logmap_roll_dev(ul);
226 
227 	/* Kill the roll thread */
228 	logmap_kill_roll(ul);
229 
230 	/* release saved alloction info */
231 	if (ul->un_ebp)
232 		kmem_free(ul->un_ebp, ul->un_nbeb);
233 
234 	/* release circular bufs */
235 	free_cirbuf(&ul->un_rdbuf);
236 	free_cirbuf(&ul->un_wrbuf);
237 
238 	/* release maps */
239 	if (ul->un_logmap)
240 		ul->un_logmap = map_put(ul->un_logmap);
241 	if (ul->un_deltamap)
242 		ul->un_deltamap = map_put(ul->un_deltamap);
243 	if (ul->un_matamap)
244 		ul->un_matamap = map_put(ul->un_matamap);
245 
246 	mutex_destroy(&ul->un_log_mutex);
247 	mutex_destroy(&ul->un_state_mutex);
248 
249 	/* release state buffer MUST BE LAST!! (contains our ondisk data) */
250 	if (ul->un_bp)
251 		brelse(ul->un_bp);
252 	kmem_free(ul, sizeof (*ul));
253 
254 	ufsvfsp->vfs_log = NULL;
255 }
256 
257 int
258 lufs_snarf(ufsvfs_t *ufsvfsp, struct fs *fs, int ronly)
259 {
260 	buf_t		*bp, *tbp;
261 	ml_unit_t	*ul;
262 	extent_block_t	*ebp;
263 	ic_extent_block_t  *nebp;
264 	size_t		nb;
265 	daddr_t		bno;	/* in disk blocks */
266 	int		i;
267 
268 	/* LINTED: warning: logical expression always true: op "||" */
269 	ASSERT(sizeof (ml_odunit_t) < DEV_BSIZE);
270 
271 	/*
272 	 * Get the allocation table
273 	 *	During a remount the superblock pointed to by the ufsvfsp
274 	 *	is out of date.  Hence the need for the ``new'' superblock
275 	 *	pointer, fs, passed in as a parameter.
276 	 */
277 	bp = UFS_BREAD(ufsvfsp, ufsvfsp->vfs_dev, logbtodb(fs, fs->fs_logbno),
278 	    fs->fs_bsize);
279 	if (bp->b_flags & B_ERROR) {
280 		brelse(bp);
281 		return (EIO);
282 	}
283 	ebp = (void *)bp->b_un.b_addr;
284 	if (!checksum(&ebp->chksum, (int32_t *)bp->b_un.b_addr,
285 		fs->fs_bsize)) {
286 		brelse(bp);
287 		return (ENODEV);
288 	}
289 
290 	/*
291 	 * It is possible to get log blocks with all zeros.
292 	 * We should also check for nextents to be zero in such case.
293 	 */
294 	if (ebp->type != LUFS_EXTENTS || ebp->nextents == 0) {
295 		brelse(bp);
296 		return (EDOM);
297 	}
298 	/*
299 	 * Put allocation into memory.  This requires conversion between
300 	 * on the ondisk format of the extent (type extent_t) and the
301 	 * in-core format of the extent (type ic_extent_t).  The
302 	 * difference is the in-core form of the extent block stores
303 	 * the physical offset of the extent in disk blocks, which
304 	 * can require more than a 32-bit field.
305 	 */
306 	nb = (size_t)(sizeof (ic_extent_block_t) +
307 			((ebp->nextents - 1) * sizeof (ic_extent_t)));
308 	nebp = kmem_alloc(nb, KM_SLEEP);
309 	nebp->ic_nextents = ebp->nextents;
310 	nebp->ic_nbytes = ebp->nbytes;
311 	nebp->ic_nextbno = ebp->nextbno;
312 	for (i = 0; i < ebp->nextents; i++) {
313 		nebp->ic_extents[i].ic_lbno = ebp->extents[i].lbno;
314 		nebp->ic_extents[i].ic_nbno = ebp->extents[i].nbno;
315 		nebp->ic_extents[i].ic_pbno =
316 		    logbtodb(fs, ebp->extents[i].pbno);
317 	}
318 	brelse(bp);
319 
320 	/*
321 	 * Get the log state
322 	 */
323 	bno = nebp->ic_extents[0].ic_pbno;
324 	bp = UFS_BREAD(ufsvfsp, ufsvfsp->vfs_dev, bno, DEV_BSIZE);
325 	if (bp->b_flags & B_ERROR) {
326 		brelse(bp);
327 		bp = UFS_BREAD(ufsvfsp, ufsvfsp->vfs_dev, bno + 1, DEV_BSIZE);
328 		if (bp->b_flags & B_ERROR) {
329 			brelse(bp);
330 			kmem_free(nebp, nb);
331 			return (EIO);
332 		}
333 	}
334 
335 	/*
336 	 * Put ondisk struct into an anonymous buffer
337 	 *	This buffer will contain the memory for the ml_odunit struct
338 	 */
339 	tbp = ngeteblk(dbtob(LS_SECTORS));
340 	tbp->b_edev = bp->b_edev;
341 	tbp->b_dev = bp->b_dev;
342 	tbp->b_blkno = bno;
343 	bcopy(bp->b_un.b_addr, tbp->b_un.b_addr, DEV_BSIZE);
344 	bcopy(bp->b_un.b_addr, tbp->b_un.b_addr + DEV_BSIZE, DEV_BSIZE);
345 	bp->b_flags |= (B_STALE | B_AGE);
346 	brelse(bp);
347 	bp = tbp;
348 
349 	/*
350 	 * Verify the log state
351 	 *
352 	 * read/only mounts w/bad logs are allowed.  umount will
353 	 * eventually roll the bad log until the first IO error.
354 	 * fsck will then repair the file system.
355 	 *
356 	 * read/write mounts with bad logs are not allowed.
357 	 *
358 	 */
359 	ul = (ml_unit_t *)kmem_zalloc(sizeof (*ul), KM_SLEEP);
360 	bcopy(bp->b_un.b_addr, &ul->un_ondisk, sizeof (ml_odunit_t));
361 	if ((ul->un_chksum != ul->un_head_ident + ul->un_tail_ident) ||
362 	    (ul->un_version != LUFS_VERSION_LATEST) ||
363 	    (!ronly && ul->un_badlog)) {
364 		kmem_free(ul, sizeof (*ul));
365 		brelse(bp);
366 		kmem_free(nebp, nb);
367 		return (EIO);
368 	}
369 	/*
370 	 * Initialize the incore-only fields
371 	 */
372 	if (ronly)
373 		ul->un_flags |= LDL_NOROLL;
374 	ul->un_bp = bp;
375 	ul->un_ufsvfs = ufsvfsp;
376 	ul->un_dev = ufsvfsp->vfs_dev;
377 	ul->un_ebp = nebp;
378 	ul->un_nbeb = nb;
379 	ul->un_maxresv = btodb(ul->un_logsize) * LDL_USABLE_BSIZE;
380 	ul->un_deltamap = map_get(ul, deltamaptype, DELTAMAP_NHASH);
381 	ul->un_logmap = map_get(ul, logmaptype, LOGMAP_NHASH);
382 	if (ul->un_debug & MT_MATAMAP)
383 		ul->un_matamap = map_get(ul, matamaptype, DELTAMAP_NHASH);
384 	mutex_init(&ul->un_log_mutex, NULL, MUTEX_DEFAULT, NULL);
385 	mutex_init(&ul->un_state_mutex, NULL, MUTEX_DEFAULT, NULL);
386 	ufsvfsp->vfs_log = ul;
387 
388 	/* remember the state of the log before the log scan */
389 	logmap_logscan(ul);
390 
391 	/*
392 	 * Error during scan
393 	 *
394 	 * If this is a read/only mount; ignore the error.
395 	 * At a later time umount/fsck will repair the fs.
396 	 *
397 	 */
398 	if (ul->un_flags & LDL_ERROR) {
399 		if (!ronly) {
400 			lufs_unsnarf(ufsvfsp);
401 			return (EIO);
402 		}
403 		ul->un_flags &= ~LDL_ERROR;
404 	}
405 	if (!ronly)
406 		logmap_start_roll(ul);
407 	return (0);
408 }
409 
410 static int
411 lufs_initialize(
412 	ufsvfs_t *ufsvfsp,
413 	daddr_t bno,
414 	size_t nb,
415 	struct fiolog *flp)
416 {
417 	ml_odunit_t	*ud, *ud2;
418 	buf_t		*bp;
419 	struct timeval	tv;
420 
421 	/* LINTED: warning: logical expression always true: op "||" */
422 	ASSERT(sizeof (ml_odunit_t) < DEV_BSIZE);
423 	ASSERT(nb >= ldl_minlogsize);
424 
425 	bp = UFS_GETBLK(ufsvfsp, ufsvfsp->vfs_dev, bno, dbtob(LS_SECTORS));
426 	bzero(bp->b_un.b_addr, bp->b_bcount);
427 
428 	ud = (void *)bp->b_un.b_addr;
429 	ud->od_version = LUFS_VERSION_LATEST;
430 	ud->od_maxtransfer = MIN(ufsvfsp->vfs_iotransz, ldl_maxtransfer);
431 	if (ud->od_maxtransfer < ldl_mintransfer)
432 		ud->od_maxtransfer = ldl_mintransfer;
433 	ud->od_devbsize = DEV_BSIZE;
434 
435 	ud->od_requestsize = flp->nbytes_actual;
436 	ud->od_statesize = dbtob(LS_SECTORS);
437 	ud->od_logsize = nb - ud->od_statesize;
438 
439 	ud->od_statebno = INT32_C(0);
440 
441 	uniqtime(&tv);
442 	if (tv.tv_usec == last_loghead_ident) {
443 		tv.tv_usec++;
444 	}
445 	last_loghead_ident = tv.tv_usec;
446 	ud->od_head_ident = tv.tv_usec;
447 	ud->od_tail_ident = ud->od_head_ident;
448 	ud->od_chksum = ud->od_head_ident + ud->od_tail_ident;
449 
450 	ud->od_bol_lof = dbtob(ud->od_statebno) + ud->od_statesize;
451 	ud->od_eol_lof = ud->od_bol_lof + ud->od_logsize;
452 	ud->od_head_lof = ud->od_bol_lof;
453 	ud->od_tail_lof = ud->od_bol_lof;
454 
455 	ASSERT(lufs_initialize_debug(ud));
456 
457 	ud2 = (void *)(bp->b_un.b_addr + DEV_BSIZE);
458 	bcopy(ud, ud2, sizeof (*ud));
459 
460 	UFS_BWRITE2(ufsvfsp, bp);
461 	if (bp->b_flags & B_ERROR) {
462 		brelse(bp);
463 		return (EIO);
464 	}
465 	brelse(bp);
466 
467 	return (0);
468 }
469 
470 /*
471  * Free log space
472  *	Assumes the file system is write locked and is not logging
473  */
474 static int
475 lufs_free(struct ufsvfs *ufsvfsp)
476 {
477 	int		error = 0, i, j;
478 	buf_t		*bp = NULL;
479 	extent_t	*ep;
480 	extent_block_t	*ebp;
481 	struct fs	*fs = ufsvfsp->vfs_fs;
482 	daddr_t		fno;
483 	int32_t		logbno;
484 	long		nfno;
485 	inode_t		*ip = NULL;
486 	char		clean;
487 
488 	/*
489 	 * Nothing to free
490 	 */
491 	if (fs->fs_logbno == 0)
492 		return (0);
493 
494 	/*
495 	 * Mark the file system as FSACTIVE and no log but honor the
496 	 * current value of fs_reclaim.  The reclaim thread could have
497 	 * been active when lufs_disable() was called and if fs_reclaim
498 	 * is reset to zero here it could lead to lost inodes.
499 	 */
500 	ufsvfsp->vfs_ulockfs.ul_sbowner = curthread;
501 	mutex_enter(&ufsvfsp->vfs_lock);
502 	clean = fs->fs_clean;
503 	logbno = fs->fs_logbno;
504 	fs->fs_clean = FSACTIVE;
505 	fs->fs_logbno = INT32_C(0);
506 	ufs_sbwrite(ufsvfsp);
507 	mutex_exit(&ufsvfsp->vfs_lock);
508 	ufsvfsp->vfs_ulockfs.ul_sbowner = (kthread_id_t)-1;
509 	if (ufsvfsp->vfs_bufp->b_flags & B_ERROR) {
510 		error = EIO;
511 		fs->fs_clean = clean;
512 		fs->fs_logbno = logbno;
513 		goto errout;
514 	}
515 
516 	/*
517 	 * fetch the allocation block
518 	 *	superblock -> one block of extents -> log data
519 	 */
520 	bp = UFS_BREAD(ufsvfsp, ufsvfsp->vfs_dev, logbtodb(fs, logbno),
521 	    fs->fs_bsize);
522 	if (bp->b_flags & B_ERROR) {
523 		error = EIO;
524 		goto errout;
525 	}
526 
527 	/*
528 	 * Free up the allocated space (dummy inode needed for free())
529 	 */
530 	ip = ufs_alloc_inode(ufsvfsp, UFSROOTINO);
531 	ebp = (void *)bp->b_un.b_addr;
532 	for (i = 0, ep = &ebp->extents[0]; i < ebp->nextents; ++i, ++ep) {
533 		fno = logbtofrag(fs, ep->pbno);
534 		nfno = dbtofsb(fs, ep->nbno);
535 		for (j = 0; j < nfno; j += fs->fs_frag, fno += fs->fs_frag)
536 			free(ip, fno, fs->fs_bsize, 0);
537 	}
538 	free(ip, logbtofrag(fs, logbno), fs->fs_bsize, 0);
539 	brelse(bp);
540 	bp = NULL;
541 
542 	/*
543 	 * Push the metadata dirtied during the allocations
544 	 */
545 	ufsvfsp->vfs_ulockfs.ul_sbowner = curthread;
546 	sbupdate(ufsvfsp->vfs_vfs);
547 	ufsvfsp->vfs_ulockfs.ul_sbowner = (kthread_id_t)-1;
548 	bflush(ufsvfsp->vfs_dev);
549 	error = bfinval(ufsvfsp->vfs_dev, 0);
550 	if (error)
551 		goto errout;
552 
553 	/*
554 	 * Free the dummy inode
555 	 */
556 	ufs_free_inode(ip);
557 
558 	return (0);
559 
560 errout:
561 	/*
562 	 * Free up all resources
563 	 */
564 	if (bp)
565 		brelse(bp);
566 	if (ip)
567 		ufs_free_inode(ip);
568 	return (error);
569 }
570 
571 /*
572  * Allocate log space
573  *	Assumes the file system is write locked and is not logging
574  */
575 static int
576 lufs_alloc(struct ufsvfs *ufsvfsp, struct fiolog *flp, cred_t *cr)
577 {
578 	int		error = 0;
579 	buf_t		*bp = NULL;
580 	extent_t	*ep, *nep;
581 	extent_block_t	*ebp;
582 	struct fs	*fs = ufsvfsp->vfs_fs;
583 	daddr_t		fno;	/* in frags */
584 	daddr_t		bno;	/* in disk blocks */
585 	int32_t		logbno = INT32_C(0);	/* will be fs_logbno */
586 	struct inode	*ip = NULL;
587 	size_t		nb = flp->nbytes_actual;
588 	size_t		tb = 0;
589 
590 	/*
591 	 * Mark the file system as FSACTIVE
592 	 */
593 	ufsvfsp->vfs_ulockfs.ul_sbowner = curthread;
594 	mutex_enter(&ufsvfsp->vfs_lock);
595 	fs->fs_clean = FSACTIVE;
596 	ufs_sbwrite(ufsvfsp);
597 	mutex_exit(&ufsvfsp->vfs_lock);
598 	ufsvfsp->vfs_ulockfs.ul_sbowner = (kthread_id_t)-1;
599 
600 	/*
601 	 * Allocate the allocation block (need dummy shadow inode;
602 	 * we use a shadow inode so the quota sub-system ignores
603 	 * the block allocations.)
604 	 *	superblock -> one block of extents -> log data
605 	 */
606 	ip = ufs_alloc_inode(ufsvfsp, UFSROOTINO);
607 	ip->i_mode = IFSHAD;		/* make the dummy a shadow inode */
608 	rw_enter(&ip->i_contents, RW_WRITER);
609 	fno = contigpref(ufsvfsp, nb + fs->fs_bsize);
610 	error = alloc(ip, fno, fs->fs_bsize, &fno, cr);
611 	if (error)
612 		goto errout;
613 	bno = fsbtodb(fs, fno);
614 
615 	bp = UFS_BREAD(ufsvfsp, ufsvfsp->vfs_dev, bno, fs->fs_bsize);
616 	if (bp->b_flags & B_ERROR) {
617 		error = EIO;
618 		goto errout;
619 	}
620 
621 	ebp = (void *)bp->b_un.b_addr;
622 	ebp->type = LUFS_EXTENTS;
623 	ebp->nextbno = UINT32_C(0);
624 	ebp->nextents = UINT32_C(0);
625 	ebp->chksum = INT32_C(0);
626 	if (fs->fs_magic == FS_MAGIC)
627 		logbno = bno;
628 	else
629 		logbno = dbtofsb(fs, bno);
630 
631 	/*
632 	 * Initialize the first extent
633 	 */
634 	ep = &ebp->extents[0];
635 	error = alloc(ip, fno + fs->fs_frag, fs->fs_bsize, &fno, cr);
636 	if (error)
637 		goto errout;
638 	bno = fsbtodb(fs, fno);
639 
640 	ep->lbno = UINT32_C(0);
641 	if (fs->fs_magic == FS_MAGIC)
642 		ep->pbno = (uint32_t)bno;
643 	else
644 		ep->pbno = (uint32_t)fno;
645 	ep->nbno = (uint32_t)fsbtodb(fs, fs->fs_frag);
646 	ebp->nextents = UINT32_C(1);
647 	tb = fs->fs_bsize;
648 	nb -= fs->fs_bsize;
649 
650 	while (nb) {
651 		error = alloc(ip, fno + fs->fs_frag, fs->fs_bsize, &fno, cr);
652 		if (error) {
653 			if (tb < ldl_minlogsize)
654 				goto errout;
655 			error = 0;
656 			break;
657 		}
658 		bno = fsbtodb(fs, fno);
659 		if ((daddr_t)((logbtodb(fs, ep->pbno) + ep->nbno) == bno))
660 			ep->nbno += (uint32_t)(fsbtodb(fs, fs->fs_frag));
661 		else {
662 			nep = ep + 1;
663 			if ((caddr_t)(nep + 1) >
664 			    (bp->b_un.b_addr + fs->fs_bsize)) {
665 				free(ip, fno, fs->fs_bsize, 0);
666 				break;
667 			}
668 			nep->lbno = ep->lbno + ep->nbno;
669 			if (fs->fs_magic == FS_MAGIC)
670 				nep->pbno = (uint32_t)bno;
671 			else
672 				nep->pbno = (uint32_t)fno;
673 			nep->nbno = (uint32_t)(fsbtodb(fs, fs->fs_frag));
674 			ebp->nextents++;
675 			ep = nep;
676 		}
677 		tb += fs->fs_bsize;
678 		nb -= fs->fs_bsize;
679 	}
680 	ebp->nbytes = (uint32_t)tb;
681 	setsum(&ebp->chksum, (int32_t *)bp->b_un.b_addr, fs->fs_bsize);
682 	UFS_BWRITE2(ufsvfsp, bp);
683 	if (bp->b_flags & B_ERROR) {
684 		error = EIO;
685 		goto errout;
686 	}
687 	/*
688 	 * Initialize the first two sectors of the log
689 	 */
690 	error = lufs_initialize(ufsvfsp, logbtodb(fs, ebp->extents[0].pbno),
691 	    tb, flp);
692 	if (error)
693 		goto errout;
694 
695 	/*
696 	 * We are done initializing the allocation block and the log
697 	 */
698 	brelse(bp);
699 	bp = NULL;
700 
701 	/*
702 	 * Update the superblock and push the dirty metadata
703 	 */
704 	ufsvfsp->vfs_ulockfs.ul_sbowner = curthread;
705 	sbupdate(ufsvfsp->vfs_vfs);
706 	ufsvfsp->vfs_ulockfs.ul_sbowner = (kthread_id_t)-1;
707 	bflush(ufsvfsp->vfs_dev);
708 	error = bfinval(ufsvfsp->vfs_dev, 1);
709 	if (error)
710 		goto errout;
711 	if (ufsvfsp->vfs_bufp->b_flags & B_ERROR) {
712 		error = EIO;
713 		goto errout;
714 	}
715 
716 	/*
717 	 * Everything is safely on disk; update log space pointer in sb
718 	 */
719 	ufsvfsp->vfs_ulockfs.ul_sbowner = curthread;
720 	mutex_enter(&ufsvfsp->vfs_lock);
721 	fs->fs_logbno = (uint32_t)logbno;
722 	ufs_sbwrite(ufsvfsp);
723 	mutex_exit(&ufsvfsp->vfs_lock);
724 	ufsvfsp->vfs_ulockfs.ul_sbowner = (kthread_id_t)-1;
725 
726 	/*
727 	 * Free the dummy inode
728 	 */
729 	rw_exit(&ip->i_contents);
730 	ufs_free_inode(ip);
731 
732 	/* inform user of real log size */
733 	flp->nbytes_actual = tb;
734 	return (0);
735 
736 errout:
737 	/*
738 	 * Free all resources
739 	 */
740 	if (bp)
741 		brelse(bp);
742 	if (logbno) {
743 		fs->fs_logbno = logbno;
744 		(void) lufs_free(ufsvfsp);
745 	}
746 	if (ip) {
747 		rw_exit(&ip->i_contents);
748 		ufs_free_inode(ip);
749 	}
750 	return (error);
751 }
752 
753 /*
754  * Disable logging
755  */
756 int
757 lufs_disable(vnode_t *vp, struct fiolog *flp)
758 {
759 	int		error = 0;
760 	inode_t		*ip = VTOI(vp);
761 	ufsvfs_t	*ufsvfsp = ip->i_ufsvfs;
762 	struct fs	*fs = ufsvfsp->vfs_fs;
763 	struct lockfs	lf;
764 	struct ulockfs	*ulp;
765 
766 	flp->error = FIOLOG_ENONE;
767 
768 	/*
769 	 * Logging is already disabled; done
770 	 */
771 	if (fs->fs_logbno == 0 || ufsvfsp->vfs_log == NULL)
772 		return (0);
773 
774 	/*
775 	 * Readonly file system
776 	 */
777 	if (fs->fs_ronly) {
778 		flp->error = FIOLOG_EROFS;
779 		return (0);
780 	}
781 
782 	/*
783 	 * File system must be write locked to disable logging
784 	 */
785 	error = ufs_fiolfss(vp, &lf);
786 	if (error) {
787 		return (error);
788 	}
789 	if (!LOCKFS_IS_ULOCK(&lf)) {
790 		flp->error = FIOLOG_EULOCK;
791 		return (0);
792 	}
793 	lf.lf_lock = LOCKFS_WLOCK;
794 	lf.lf_flags = 0;
795 	lf.lf_comment = NULL;
796 	error = ufs_fiolfs(vp, &lf, 1);
797 	if (error) {
798 		flp->error = FIOLOG_EWLOCK;
799 		return (0);
800 	}
801 
802 	if (ufsvfsp->vfs_log == NULL || fs->fs_logbno == 0)
803 		goto errout;
804 
805 	/*
806 	 * WE ARE COMMITTED TO DISABLING LOGGING PAST THIS POINT
807 	 */
808 
809 	/*
810 	 * Disable logging:
811 	 * Suspend the reclaim thread and force the delete thread to exit.
812 	 *	When a nologging mount has completed there may still be
813 	 *	work for reclaim to do so just suspend this thread until
814 	 *	it's [deadlock-] safe for it to continue.  The delete
815 	 *	thread won't be needed as ufs_iinactive() calls
816 	 *	ufs_delete() when logging is disabled.
817 	 * Freeze and drain reader ops.
818 	 *	Commit any outstanding reader transactions (ufs_flush).
819 	 *	Set the ``unmounted'' bit in the ufstrans struct.
820 	 *	If debug, remove metadata from matamap.
821 	 *	Disable matamap processing.
822 	 *	NULL the trans ops table.
823 	 *	Free all of the incore structs related to logging.
824 	 * Allow reader ops.
825 	 */
826 	ufs_thread_suspend(&ufsvfsp->vfs_reclaim);
827 	ufs_thread_exit(&ufsvfsp->vfs_delete);
828 
829 	vfs_lock_wait(ufsvfsp->vfs_vfs);
830 	ulp = &ufsvfsp->vfs_ulockfs;
831 	mutex_enter(&ulp->ul_lock);
832 	(void) ufs_quiesce(ulp);
833 
834 	(void) ufs_flush(ufsvfsp->vfs_vfs);
835 
836 	TRANS_MATA_UMOUNT(ufsvfsp);
837 	ufsvfsp->vfs_domatamap = 0;
838 
839 	/*
840 	 * Free all of the incore structs
841 	 */
842 	(void) lufs_unsnarf(ufsvfsp);
843 
844 	mutex_exit(&ulp->ul_lock);
845 	vfs_setmntopt(ufsvfsp->vfs_vfs, MNTOPT_NOLOGGING, NULL, 0);
846 	vfs_unlock(ufsvfsp->vfs_vfs);
847 
848 	fs->fs_rolled = FS_ALL_ROLLED;
849 	ufsvfsp->vfs_nolog_si = 0;
850 
851 	/*
852 	 * Free the log space and mark the superblock as FSACTIVE
853 	 */
854 	(void) lufs_free(ufsvfsp);
855 
856 	/*
857 	 * Allow the reclaim thread to continue.
858 	 */
859 	ufs_thread_continue(&ufsvfsp->vfs_reclaim);
860 
861 	/*
862 	 * Unlock the file system
863 	 */
864 	lf.lf_lock = LOCKFS_ULOCK;
865 	lf.lf_flags = 0;
866 	error = ufs_fiolfs(vp, &lf, 1);
867 	if (error)
868 		flp->error = FIOLOG_ENOULOCK;
869 
870 	return (0);
871 
872 errout:
873 	lf.lf_lock = LOCKFS_ULOCK;
874 	lf.lf_flags = 0;
875 	(void) ufs_fiolfs(vp, &lf, 1);
876 	return (error);
877 }
878 
879 /*
880  * Enable logging
881  */
882 int
883 lufs_enable(struct vnode *vp, struct fiolog *flp, cred_t *cr)
884 {
885 	int		error;
886 	int		reclaim;
887 	inode_t		*ip = VTOI(vp);
888 	ufsvfs_t	*ufsvfsp = ip->i_ufsvfs;
889 	struct fs	*fs;
890 	ml_unit_t	*ul;
891 	struct lockfs	lf;
892 	struct ulockfs	*ulp;
893 	vfs_t		*vfsp = ufsvfsp->vfs_vfs;
894 	uint64_t	tmp_nbytes_actual;
895 
896 	/*
897 	 * Check if logging is already enabled
898 	 */
899 	if (ufsvfsp->vfs_log) {
900 		flp->error = FIOLOG_ETRANS;
901 		/* for root ensure logging option is set */
902 		vfs_setmntopt(vfsp, MNTOPT_LOGGING, NULL, 0);
903 		return (0);
904 	}
905 	fs = ufsvfsp->vfs_fs;
906 
907 	/*
908 	 * Come back here to recheck if we had to disable the log.
909 	 */
910 recheck:
911 	error = 0;
912 	reclaim = 0;
913 	flp->error = FIOLOG_ENONE;
914 
915 	/*
916 	 * Adjust requested log size
917 	 */
918 	flp->nbytes_actual = flp->nbytes_requested;
919 	if (flp->nbytes_actual == 0) {
920 		tmp_nbytes_actual =
921 		    (((uint64_t)fs->fs_size) / ldl_divisor) << fs->fs_fshift;
922 		flp->nbytes_actual = (uint_t)MIN(tmp_nbytes_actual, INT_MAX);
923 	}
924 	flp->nbytes_actual = MAX(flp->nbytes_actual, ldl_minlogsize);
925 	flp->nbytes_actual = MIN(flp->nbytes_actual, ldl_maxlogsize);
926 	flp->nbytes_actual = blkroundup(fs, flp->nbytes_actual);
927 
928 	/*
929 	 * logging is enabled and the log is the right size; done
930 	 */
931 	ul = ufsvfsp->vfs_log;
932 	if (ul && fs->fs_logbno && (flp->nbytes_actual == ul->un_requestsize))
933 			return (0);
934 
935 	/*
936 	 * Readonly file system
937 	 */
938 	if (fs->fs_ronly) {
939 		flp->error = FIOLOG_EROFS;
940 		return (0);
941 	}
942 
943 	/*
944 	 * File system must be write locked to enable logging
945 	 */
946 	error = ufs_fiolfss(vp, &lf);
947 	if (error) {
948 		return (error);
949 	}
950 	if (!LOCKFS_IS_ULOCK(&lf)) {
951 		flp->error = FIOLOG_EULOCK;
952 		return (0);
953 	}
954 	lf.lf_lock = LOCKFS_WLOCK;
955 	lf.lf_flags = 0;
956 	lf.lf_comment = NULL;
957 	error = ufs_fiolfs(vp, &lf, 1);
958 	if (error) {
959 		flp->error = FIOLOG_EWLOCK;
960 		return (0);
961 	}
962 
963 	/*
964 	 * File system must be fairly consistent to enable logging
965 	 */
966 	if (fs->fs_clean != FSLOG &&
967 	    fs->fs_clean != FSACTIVE &&
968 	    fs->fs_clean != FSSTABLE &&
969 	    fs->fs_clean != FSCLEAN) {
970 		flp->error = FIOLOG_ECLEAN;
971 		goto unlockout;
972 	}
973 
974 	/*
975 	 * A write-locked file system is only active if there are
976 	 * open deleted files; so remember to set FS_RECLAIM later.
977 	 */
978 	if (fs->fs_clean == FSACTIVE)
979 		reclaim = FS_RECLAIM;
980 
981 	/*
982 	 * Logging is already enabled; must be changing the log's size
983 	 */
984 	if (fs->fs_logbno && ufsvfsp->vfs_log) {
985 		/*
986 		 * Before we can disable logging, we must give up our
987 		 * lock.  As a consequence of unlocking and disabling the
988 		 * log, the fs structure may change.  Because of this, when
989 		 * disabling is complete, we will go back to recheck to
990 		 * repeat all of the checks that we performed to get to
991 		 * this point.  Disabling sets fs->fs_logbno to 0, so this
992 		 * will not put us into an infinite loop.
993 		 */
994 		lf.lf_lock = LOCKFS_ULOCK;
995 		lf.lf_flags = 0;
996 		error = ufs_fiolfs(vp, &lf, 1);
997 		if (error) {
998 			flp->error = FIOLOG_ENOULOCK;
999 			return (0);
1000 		}
1001 		error = lufs_disable(vp, flp);
1002 		if (error || (flp->error != FIOLOG_ENONE))
1003 			return (0);
1004 		goto recheck;
1005 	}
1006 
1007 	error = lufs_alloc(ufsvfsp, flp, cr);
1008 	if (error)
1009 		goto errout;
1010 
1011 	/*
1012 	 * Create all of the incore structs
1013 	 */
1014 	error = lufs_snarf(ufsvfsp, fs, 0);
1015 	if (error)
1016 		goto errout;
1017 
1018 	/*
1019 	 * DON'T ``GOTO ERROUT'' PAST THIS POINT
1020 	 */
1021 
1022 	/*
1023 	 * Pretend we were just mounted with logging enabled
1024 	 *	freeze and drain the file system of readers
1025 	 *		Get the ops vector
1026 	 *		If debug, record metadata locations with log subsystem
1027 	 *		Start the delete thread
1028 	 *		Start the reclaim thread, if necessary
1029 	 *	Thaw readers
1030 	 */
1031 	vfs_lock_wait(vfsp);
1032 	vfs_setmntopt(vfsp, MNTOPT_LOGGING, NULL, 0);
1033 	ulp = &ufsvfsp->vfs_ulockfs;
1034 	mutex_enter(&ulp->ul_lock);
1035 	(void) ufs_quiesce(ulp);
1036 
1037 	TRANS_DOMATAMAP(ufsvfsp);
1038 	TRANS_MATA_MOUNT(ufsvfsp);
1039 	TRANS_MATA_SI(ufsvfsp, fs);
1040 	ufs_thread_start(&ufsvfsp->vfs_delete, ufs_thread_delete, vfsp);
1041 	if (fs->fs_reclaim & (FS_RECLAIM|FS_RECLAIMING)) {
1042 		fs->fs_reclaim &= ~FS_RECLAIM;
1043 		fs->fs_reclaim |=  FS_RECLAIMING;
1044 		ufs_thread_start(&ufsvfsp->vfs_reclaim,
1045 					ufs_thread_reclaim, vfsp);
1046 	} else
1047 		fs->fs_reclaim |= reclaim;
1048 
1049 	mutex_exit(&ulp->ul_lock);
1050 	vfs_unlock(vfsp);
1051 
1052 	/*
1053 	 * Unlock the file system
1054 	 */
1055 	lf.lf_lock = LOCKFS_ULOCK;
1056 	lf.lf_flags = 0;
1057 	error = ufs_fiolfs(vp, &lf, 1);
1058 	if (error) {
1059 		flp->error = FIOLOG_ENOULOCK;
1060 		return (0);
1061 	}
1062 
1063 	/*
1064 	 * There's nothing in the log yet (we've just allocated it)
1065 	 * so directly write out the super block.
1066 	 * Note, we have to force this sb out to disk
1067 	 * (not just to the log) so that if we crash we know we are logging
1068 	 */
1069 	mutex_enter(&ufsvfsp->vfs_lock);
1070 	fs->fs_clean = FSLOG;
1071 	fs->fs_rolled = FS_NEED_ROLL; /* Mark the fs as unrolled */
1072 	UFS_BWRITE2(NULL, ufsvfsp->vfs_bufp);
1073 	mutex_exit(&ufsvfsp->vfs_lock);
1074 
1075 	return (0);
1076 
1077 errout:
1078 	(void) lufs_unsnarf(ufsvfsp);
1079 	(void) lufs_free(ufsvfsp);
1080 unlockout:
1081 	lf.lf_lock = LOCKFS_ULOCK;
1082 	lf.lf_flags = 0;
1083 	(void) ufs_fiolfs(vp, &lf, 1);
1084 	return (error);
1085 }
1086 
1087 void
1088 lufs_read_strategy(ml_unit_t *ul, buf_t *bp)
1089 {
1090 	mt_map_t	*logmap	= ul->un_logmap;
1091 	offset_t	mof	= ldbtob(bp->b_blkno);
1092 	off_t		nb	= bp->b_bcount;
1093 	mapentry_t	*age;
1094 	char		*va;
1095 	int		(*saviodone)();
1096 	int		entire_range;
1097 
1098 	/*
1099 	 * get a linked list of overlapping deltas
1100 	 * returns with &mtm->mtm_rwlock held
1101 	 */
1102 	entire_range = logmap_list_get(logmap, mof, nb, &age);
1103 
1104 	/*
1105 	 * no overlapping deltas were found; read master
1106 	 */
1107 	if (age == NULL) {
1108 		rw_exit(&logmap->mtm_rwlock);
1109 		if (ul->un_flags & LDL_ERROR) {
1110 			bp->b_flags |= B_ERROR;
1111 			bp->b_error = EIO;
1112 			biodone(bp);
1113 		} else {
1114 			ul->un_ufsvfs->vfs_iotstamp = lbolt;
1115 			logstats.ls_lreads.value.ui64++;
1116 			(void) bdev_strategy(bp);
1117 			lwp_stat_update(LWP_STAT_INBLK, 1);
1118 		}
1119 		return;
1120 	}
1121 
1122 	va = bp_mapin_common(bp, VM_SLEEP);
1123 	/*
1124 	 * if necessary, sync read the data from master
1125 	 *	errors are returned in bp
1126 	 */
1127 	if (!entire_range) {
1128 		saviodone = bp->b_iodone;
1129 		bp->b_iodone = trans_not_done;
1130 		logstats.ls_mreads.value.ui64++;
1131 		(void) bdev_strategy(bp);
1132 		lwp_stat_update(LWP_STAT_INBLK, 1);
1133 		if (trans_not_wait(bp))
1134 			ldl_seterror(ul, "Error reading master");
1135 		bp->b_iodone = saviodone;
1136 	}
1137 
1138 	/*
1139 	 * sync read the data from the log
1140 	 *	errors are returned inline
1141 	 */
1142 	if (ldl_read(ul, va, mof, nb, age)) {
1143 		bp->b_flags |= B_ERROR;
1144 		bp->b_error = EIO;
1145 	}
1146 
1147 	/*
1148 	 * unlist the deltas
1149 	 */
1150 	logmap_list_put(logmap, age);
1151 
1152 	/*
1153 	 * all done
1154 	 */
1155 	if (ul->un_flags & LDL_ERROR) {
1156 		bp->b_flags |= B_ERROR;
1157 		bp->b_error = EIO;
1158 	}
1159 	biodone(bp);
1160 }
1161 
1162 void
1163 lufs_write_strategy(ml_unit_t *ul, buf_t *bp)
1164 {
1165 	offset_t	mof	= ldbtob(bp->b_blkno);
1166 	off_t		nb	= bp->b_bcount;
1167 	char		*va;
1168 	mapentry_t	*me;
1169 
1170 	ASSERT((nb & DEV_BMASK) == 0);
1171 	ul->un_logmap->mtm_ref = 1;
1172 
1173 	/*
1174 	 * if there are deltas, move into log
1175 	 */
1176 	me = deltamap_remove(ul->un_deltamap, mof, nb);
1177 	if (me) {
1178 
1179 		va = bp_mapin_common(bp, VM_SLEEP);
1180 
1181 		ASSERT(((ul->un_debug & MT_WRITE_CHECK) == 0) ||
1182 			(ul->un_matamap == NULL)||
1183 			matamap_within(ul->un_matamap, mof, nb));
1184 
1185 		/*
1186 		 * move to logmap
1187 		 */
1188 		if (ufs_crb_enable) {
1189 			logmap_add_buf(ul, va, mof, me,
1190 			    bp->b_un.b_addr, nb);
1191 		} else {
1192 			logmap_add(ul, va, mof, me);
1193 		}
1194 
1195 		if (ul->un_flags & LDL_ERROR) {
1196 			bp->b_flags |= B_ERROR;
1197 			bp->b_error = EIO;
1198 		}
1199 		biodone(bp);
1200 		return;
1201 	}
1202 	if (ul->un_flags & LDL_ERROR) {
1203 		bp->b_flags |= B_ERROR;
1204 		bp->b_error = EIO;
1205 		biodone(bp);
1206 		return;
1207 	}
1208 
1209 	/*
1210 	 * Check that we are not updating metadata, or if so then via B_PHYS.
1211 	 */
1212 	ASSERT((ul->un_matamap == NULL) ||
1213 		!(matamap_overlap(ul->un_matamap, mof, nb) &&
1214 		((bp->b_flags & B_PHYS) == 0)));
1215 
1216 	ul->un_ufsvfs->vfs_iotstamp = lbolt;
1217 	logstats.ls_lwrites.value.ui64++;
1218 
1219 	/* If snapshots are enabled, write through the snapshot driver */
1220 	if (ul->un_ufsvfs->vfs_snapshot)
1221 		fssnap_strategy(&ul->un_ufsvfs->vfs_snapshot, bp);
1222 	else
1223 		(void) bdev_strategy(bp);
1224 
1225 	lwp_stat_update(LWP_STAT_OUBLK, 1);
1226 }
1227 
1228 void
1229 lufs_strategy(ml_unit_t *ul, buf_t *bp)
1230 {
1231 	if (bp->b_flags & B_READ)
1232 		lufs_read_strategy(ul, bp);
1233 	else
1234 		lufs_write_strategy(ul, bp);
1235 }
1236 
1237 /* ARGSUSED */
1238 static int
1239 delta_stats_update(kstat_t *ksp, int rw)
1240 {
1241 	if (rw == KSTAT_WRITE) {
1242 		delta_stats[DT_SB] = dkstats.ds_superblock_deltas.value.ui64;
1243 		delta_stats[DT_CG] = dkstats.ds_bitmap_deltas.value.ui64;
1244 		delta_stats[DT_SI] = dkstats.ds_suminfo_deltas.value.ui64;
1245 		delta_stats[DT_AB] = dkstats.ds_allocblk_deltas.value.ui64;
1246 		delta_stats[DT_ABZERO] = dkstats.ds_ab0_deltas.value.ui64;
1247 		delta_stats[DT_DIR] = dkstats.ds_dir_deltas.value.ui64;
1248 		delta_stats[DT_INODE] = dkstats.ds_inode_deltas.value.ui64;
1249 		delta_stats[DT_FBI] = dkstats.ds_fbiwrite_deltas.value.ui64;
1250 		delta_stats[DT_QR] = dkstats.ds_quota_deltas.value.ui64;
1251 		delta_stats[DT_SHAD] = dkstats.ds_shadow_deltas.value.ui64;
1252 
1253 		roll_stats[DT_SB] = dkstats.ds_superblock_rolled.value.ui64;
1254 		roll_stats[DT_CG] = dkstats.ds_bitmap_rolled.value.ui64;
1255 		roll_stats[DT_SI] = dkstats.ds_suminfo_rolled.value.ui64;
1256 		roll_stats[DT_AB] = dkstats.ds_allocblk_rolled.value.ui64;
1257 		roll_stats[DT_ABZERO] = dkstats.ds_ab0_rolled.value.ui64;
1258 		roll_stats[DT_DIR] = dkstats.ds_dir_rolled.value.ui64;
1259 		roll_stats[DT_INODE] = dkstats.ds_inode_rolled.value.ui64;
1260 		roll_stats[DT_FBI] = dkstats.ds_fbiwrite_rolled.value.ui64;
1261 		roll_stats[DT_QR] = dkstats.ds_quota_rolled.value.ui64;
1262 		roll_stats[DT_SHAD] = dkstats.ds_shadow_rolled.value.ui64;
1263 	} else {
1264 		dkstats.ds_superblock_deltas.value.ui64 = delta_stats[DT_SB];
1265 		dkstats.ds_bitmap_deltas.value.ui64 = delta_stats[DT_CG];
1266 		dkstats.ds_suminfo_deltas.value.ui64 = delta_stats[DT_SI];
1267 		dkstats.ds_allocblk_deltas.value.ui64 = delta_stats[DT_AB];
1268 		dkstats.ds_ab0_deltas.value.ui64 = delta_stats[DT_ABZERO];
1269 		dkstats.ds_dir_deltas.value.ui64 = delta_stats[DT_DIR];
1270 		dkstats.ds_inode_deltas.value.ui64 = delta_stats[DT_INODE];
1271 		dkstats.ds_fbiwrite_deltas.value.ui64 = delta_stats[DT_FBI];
1272 		dkstats.ds_quota_deltas.value.ui64 = delta_stats[DT_QR];
1273 		dkstats.ds_shadow_deltas.value.ui64 = delta_stats[DT_SHAD];
1274 
1275 		dkstats.ds_superblock_rolled.value.ui64 = roll_stats[DT_SB];
1276 		dkstats.ds_bitmap_rolled.value.ui64 = roll_stats[DT_CG];
1277 		dkstats.ds_suminfo_rolled.value.ui64 = roll_stats[DT_SI];
1278 		dkstats.ds_allocblk_rolled.value.ui64 = roll_stats[DT_AB];
1279 		dkstats.ds_ab0_rolled.value.ui64 = roll_stats[DT_ABZERO];
1280 		dkstats.ds_dir_rolled.value.ui64 = roll_stats[DT_DIR];
1281 		dkstats.ds_inode_rolled.value.ui64 = roll_stats[DT_INODE];
1282 		dkstats.ds_fbiwrite_rolled.value.ui64 = roll_stats[DT_FBI];
1283 		dkstats.ds_quota_rolled.value.ui64 = roll_stats[DT_QR];
1284 		dkstats.ds_shadow_rolled.value.ui64 = roll_stats[DT_SHAD];
1285 	}
1286 	return (0);
1287 }
1288 
1289 extern size_t ufs_crb_limit;
1290 extern int ufs_max_crb_divisor;
1291 
1292 void
1293 lufs_init(void)
1294 {
1295 	kstat_t *ksp;
1296 
1297 	/* Create kmem caches */
1298 	lufs_sv = kmem_cache_create("lufs_save", sizeof (lufs_save_t), 0,
1299 	    NULL, NULL, NULL, NULL, NULL, 0);
1300 	lufs_bp = kmem_cache_create("lufs_bufs", sizeof (lufs_buf_t), 0,
1301 	    NULL, NULL, NULL, NULL, NULL, 0);
1302 
1303 	mutex_init(&log_mutex, NULL, MUTEX_DEFAULT, NULL);
1304 
1305 	_init_top();
1306 
1307 	if (&bio_lufs_strategy != NULL)
1308 		bio_lufs_strategy = (void (*) (void *, buf_t *)) lufs_strategy;
1309 
1310 	/*
1311 	 * Initialise general logging and delta kstats
1312 	 */
1313 	ksp = kstat_create("ufs_log", 0, "logstats", "ufs", KSTAT_TYPE_NAMED,
1314 	    sizeof (logstats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL);
1315 	if (ksp) {
1316 		ksp->ks_data = (void *) &logstats;
1317 		kstat_install(ksp);
1318 	}
1319 
1320 	ksp = kstat_create("ufs_log", 0, "deltastats", "ufs", KSTAT_TYPE_NAMED,
1321 	    sizeof (dkstats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL);
1322 	if (ksp) {
1323 		ksp->ks_data = (void *) &dkstats;
1324 		ksp->ks_update = delta_stats_update;
1325 		kstat_install(ksp);
1326 	}
1327 
1328 	/*
1329 	 * Set up the maximum amount of kmem that the crbs (system wide)
1330 	 * can use.
1331 	 */
1332 	ufs_crb_limit = kmem_maxavail() / ufs_max_crb_divisor;
1333 }
1334