xref: /titanic_41/usr/src/uts/common/fs/ufs/lufs.c (revision 355b4669e025ff377602b6fc7caaf30dbc218371)
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/atomic.h>
44 #include <sys/kmem.h>
45 #include <sys/inttypes.h>
46 #include <sys/vfs.h>
47 #include <sys/mntent.h>
48 #include <sys/conf.h>
49 #include <sys/param.h>
50 #include <sys/kstat.h>
51 #include <sys/cmn_err.h>
52 
53 extern	kmutex_t	ufs_scan_lock;
54 
55 static kmutex_t	log_mutex;	/* general purpose log layer lock */
56 kmutex_t	ml_scan;	/* Scan thread syncronization */
57 kcondvar_t	ml_scan_cv;	/* Scan thread syncronization */
58 
59 struct kmem_cache	*lufs_sv;
60 struct kmem_cache	*lufs_bp;
61 
62 /* Tunables */
63 uint_t		ldl_maxlogsize	= LDL_MAXLOGSIZE;
64 uint_t		ldl_minlogsize	= LDL_MINLOGSIZE;
65 uint32_t	ldl_divisor	= LDL_DIVISOR;
66 uint32_t	ldl_mintransfer	= LDL_MINTRANSFER;
67 uint32_t	ldl_maxtransfer	= LDL_MAXTRANSFER;
68 uint32_t	ldl_minbufsize	= LDL_MINBUFSIZE;
69 
70 uint32_t	last_loghead_ident = 0;
71 
72 /*
73  * Logging delta and roll statistics
74  */
75 struct delta_kstats {
76 	kstat_named_t ds_superblock_deltas;
77 	kstat_named_t ds_bitmap_deltas;
78 	kstat_named_t ds_suminfo_deltas;
79 	kstat_named_t ds_allocblk_deltas;
80 	kstat_named_t ds_ab0_deltas;
81 	kstat_named_t ds_dir_deltas;
82 	kstat_named_t ds_inode_deltas;
83 	kstat_named_t ds_fbiwrite_deltas;
84 	kstat_named_t ds_quota_deltas;
85 	kstat_named_t ds_shadow_deltas;
86 
87 	kstat_named_t ds_superblock_rolled;
88 	kstat_named_t ds_bitmap_rolled;
89 	kstat_named_t ds_suminfo_rolled;
90 	kstat_named_t ds_allocblk_rolled;
91 	kstat_named_t ds_ab0_rolled;
92 	kstat_named_t ds_dir_rolled;
93 	kstat_named_t ds_inode_rolled;
94 	kstat_named_t ds_fbiwrite_rolled;
95 	kstat_named_t ds_quota_rolled;
96 	kstat_named_t ds_shadow_rolled;
97 } dkstats = {
98 	{ "superblock_deltas",	KSTAT_DATA_UINT64 },
99 	{ "bitmap_deltas",	KSTAT_DATA_UINT64 },
100 	{ "suminfo_deltas",	KSTAT_DATA_UINT64 },
101 	{ "allocblk_deltas",	KSTAT_DATA_UINT64 },
102 	{ "ab0_deltas",		KSTAT_DATA_UINT64 },
103 	{ "dir_deltas",		KSTAT_DATA_UINT64 },
104 	{ "inode_deltas",	KSTAT_DATA_UINT64 },
105 	{ "fbiwrite_deltas",	KSTAT_DATA_UINT64 },
106 	{ "quota_deltas",	KSTAT_DATA_UINT64 },
107 	{ "shadow_deltas",	KSTAT_DATA_UINT64 },
108 
109 	{ "superblock_rolled",	KSTAT_DATA_UINT64 },
110 	{ "bitmap_rolled",	KSTAT_DATA_UINT64 },
111 	{ "suminfo_rolled",	KSTAT_DATA_UINT64 },
112 	{ "allocblk_rolled",	KSTAT_DATA_UINT64 },
113 	{ "ab0_rolled",		KSTAT_DATA_UINT64 },
114 	{ "dir_rolled",		KSTAT_DATA_UINT64 },
115 	{ "inode_rolled",	KSTAT_DATA_UINT64 },
116 	{ "fbiwrite_rolled",	KSTAT_DATA_UINT64 },
117 	{ "quota_rolled",	KSTAT_DATA_UINT64 },
118 	{ "shadow_rolled",	KSTAT_DATA_UINT64 }
119 };
120 
121 uint64_t delta_stats[DT_MAX];
122 uint64_t roll_stats[DT_MAX];
123 
124 /*
125  * General logging kstats
126  */
127 struct logstats logstats = {
128 	{ "master_reads",		KSTAT_DATA_UINT64 },
129 	{ "master_writes",		KSTAT_DATA_UINT64 },
130 	{ "log_reads_inmem",		KSTAT_DATA_UINT64 },
131 	{ "log_reads",			KSTAT_DATA_UINT64 },
132 	{ "log_writes",			KSTAT_DATA_UINT64 },
133 	{ "log_master_reads",		KSTAT_DATA_UINT64 },
134 	{ "log_roll_reads",		KSTAT_DATA_UINT64 },
135 	{ "log_roll_writes",		KSTAT_DATA_UINT64 }
136 };
137 
138 int
139 trans_not_done(struct buf *cb)
140 {
141 	sema_v(&cb->b_io);
142 	return (0);
143 }
144 
145 static void
146 trans_wait_panic(struct buf *cb)
147 {
148 	while ((cb->b_flags & B_DONE) == 0)
149 		drv_usecwait(10);
150 }
151 
152 int
153 trans_not_wait(struct buf *cb)
154 {
155 	/*
156 	 * In case of panic, busy wait for completion
157 	 */
158 	if (panicstr)
159 		trans_wait_panic(cb);
160 	else
161 		sema_p(&cb->b_io);
162 
163 	return (geterror(cb));
164 }
165 
166 int
167 trans_wait(struct buf *cb)
168 {
169 	/*
170 	 * In case of panic, busy wait for completion and run md daemon queues
171 	 */
172 	if (panicstr)
173 		trans_wait_panic(cb);
174 	return (biowait(cb));
175 }
176 
177 static void
178 setsum(int32_t *sp, int32_t *lp, int nb)
179 {
180 	int32_t csum = 0;
181 
182 	*sp = 0;
183 	nb /= sizeof (int32_t);
184 	while (nb--)
185 		csum += *lp++;
186 	*sp = csum;
187 }
188 
189 static int
190 checksum(int32_t *sp, int32_t *lp, int nb)
191 {
192 	int32_t ssum = *sp;
193 
194 	setsum(sp, lp, nb);
195 	if (ssum != *sp) {
196 		*sp = ssum;
197 		return (0);
198 	}
199 	return (1);
200 }
201 
202 void
203 lufs_unsnarf(ufsvfs_t *ufsvfsp)
204 {
205 	ml_unit_t *ul;
206 	mt_map_t *mtm;
207 
208 	ul = ufsvfsp->vfs_log;
209 	if (ul == NULL)
210 		return;
211 
212 	mtm = ul->un_logmap;
213 
214 	/*
215 	 * Wait for a pending top_issue_sync which is
216 	 * dispatched (via taskq_dispatch()) but hasnt completed yet.
217 	 */
218 
219 	mutex_enter(&mtm->mtm_lock);
220 
221 	while (mtm->mtm_taskq_sync_count != 0) {
222 		cv_wait(&mtm->mtm_cv, &mtm->mtm_lock);
223 	}
224 
225 	mutex_exit(&mtm->mtm_lock);
226 
227 	/* Roll committed transactions */
228 	logmap_roll_dev(ul);
229 
230 	/* Kill the roll thread */
231 	logmap_kill_roll(ul);
232 
233 	/* release saved alloction info */
234 	if (ul->un_ebp)
235 		kmem_free(ul->un_ebp, ul->un_nbeb);
236 
237 	/* release circular bufs */
238 	free_cirbuf(&ul->un_rdbuf);
239 	free_cirbuf(&ul->un_wrbuf);
240 
241 	/* release maps */
242 	if (ul->un_logmap)
243 		ul->un_logmap = map_put(ul->un_logmap);
244 	if (ul->un_deltamap)
245 		ul->un_deltamap = map_put(ul->un_deltamap);
246 	if (ul->un_matamap)
247 		ul->un_matamap = map_put(ul->un_matamap);
248 
249 	mutex_destroy(&ul->un_log_mutex);
250 	mutex_destroy(&ul->un_state_mutex);
251 
252 	/* release state buffer MUST BE LAST!! (contains our ondisk data) */
253 	if (ul->un_bp)
254 		brelse(ul->un_bp);
255 	kmem_free(ul, sizeof (*ul));
256 
257 	ufsvfsp->vfs_log = NULL;
258 }
259 
260 int
261 lufs_snarf(ufsvfs_t *ufsvfsp, struct fs *fs, int ronly)
262 {
263 	buf_t		*bp, *tbp;
264 	ml_unit_t	*ul;
265 	extent_block_t	*ebp;
266 	ic_extent_block_t  *nebp;
267 	size_t		nb;
268 	daddr_t		bno;	/* in disk blocks */
269 	int		i;
270 
271 	/* LINTED: warning: logical expression always true: op "||" */
272 	ASSERT(sizeof (ml_odunit_t) < DEV_BSIZE);
273 
274 	/*
275 	 * Get the allocation table
276 	 *	During a remount the superblock pointed to by the ufsvfsp
277 	 *	is out of date.  Hence the need for the ``new'' superblock
278 	 *	pointer, fs, passed in as a parameter.
279 	 */
280 	bp = UFS_BREAD(ufsvfsp, ufsvfsp->vfs_dev, logbtodb(fs, fs->fs_logbno),
281 	    fs->fs_bsize);
282 	if (bp->b_flags & B_ERROR) {
283 		brelse(bp);
284 		return (EIO);
285 	}
286 	ebp = (void *)bp->b_un.b_addr;
287 	if (!checksum(&ebp->chksum, (int32_t *)bp->b_un.b_addr,
288 		fs->fs_bsize)) {
289 		brelse(bp);
290 		return (ENODEV);
291 	}
292 
293 	/*
294 	 * It is possible to get log blocks with all zeros.
295 	 * We should also check for nextents to be zero in such case.
296 	 */
297 	if (ebp->type != LUFS_EXTENTS || ebp->nextents == 0) {
298 		brelse(bp);
299 		return (EDOM);
300 	}
301 	/*
302 	 * Put allocation into memory.  This requires conversion between
303 	 * on the ondisk format of the extent (type extent_t) and the
304 	 * in-core format of the extent (type ic_extent_t).  The
305 	 * difference is the in-core form of the extent block stores
306 	 * the physical offset of the extent in disk blocks, which
307 	 * can require more than a 32-bit field.
308 	 */
309 	nb = (size_t)(sizeof (ic_extent_block_t) +
310 			((ebp->nextents - 1) * sizeof (ic_extent_t)));
311 	nebp = kmem_alloc(nb, KM_SLEEP);
312 	nebp->ic_nextents = ebp->nextents;
313 	nebp->ic_nbytes = ebp->nbytes;
314 	nebp->ic_nextbno = ebp->nextbno;
315 	for (i = 0; i < ebp->nextents; i++) {
316 		nebp->ic_extents[i].ic_lbno = ebp->extents[i].lbno;
317 		nebp->ic_extents[i].ic_nbno = ebp->extents[i].nbno;
318 		nebp->ic_extents[i].ic_pbno =
319 		    logbtodb(fs, ebp->extents[i].pbno);
320 	}
321 	brelse(bp);
322 
323 	/*
324 	 * Get the log state
325 	 */
326 	bno = nebp->ic_extents[0].ic_pbno;
327 	bp = UFS_BREAD(ufsvfsp, ufsvfsp->vfs_dev, bno, DEV_BSIZE);
328 	if (bp->b_flags & B_ERROR) {
329 		brelse(bp);
330 		bp = UFS_BREAD(ufsvfsp, ufsvfsp->vfs_dev, bno + 1, DEV_BSIZE);
331 		if (bp->b_flags & B_ERROR) {
332 			brelse(bp);
333 			kmem_free(nebp, nb);
334 			return (EIO);
335 		}
336 	}
337 
338 	/*
339 	 * Put ondisk struct into an anonymous buffer
340 	 *	This buffer will contain the memory for the ml_odunit struct
341 	 */
342 	tbp = ngeteblk(dbtob(LS_SECTORS));
343 	tbp->b_edev = bp->b_edev;
344 	tbp->b_dev = bp->b_dev;
345 	tbp->b_blkno = bno;
346 	bcopy(bp->b_un.b_addr, tbp->b_un.b_addr, DEV_BSIZE);
347 	bcopy(bp->b_un.b_addr, tbp->b_un.b_addr + DEV_BSIZE, DEV_BSIZE);
348 	bp->b_flags |= (B_STALE | B_AGE);
349 	brelse(bp);
350 	bp = tbp;
351 
352 	/*
353 	 * Verify the log state
354 	 *
355 	 * read/only mounts w/bad logs are allowed.  umount will
356 	 * eventually roll the bad log until the first IO error.
357 	 * fsck will then repair the file system.
358 	 *
359 	 * read/write mounts with bad logs are not allowed.
360 	 *
361 	 */
362 	ul = (ml_unit_t *)kmem_zalloc(sizeof (*ul), KM_SLEEP);
363 	bcopy(bp->b_un.b_addr, &ul->un_ondisk, sizeof (ml_odunit_t));
364 	if ((ul->un_chksum != ul->un_head_ident + ul->un_tail_ident) ||
365 	    (ul->un_version != LUFS_VERSION_LATEST) ||
366 	    (!ronly && ul->un_badlog)) {
367 		kmem_free(ul, sizeof (*ul));
368 		brelse(bp);
369 		kmem_free(nebp, nb);
370 		return (EIO);
371 	}
372 	/*
373 	 * Initialize the incore-only fields
374 	 */
375 	if (ronly)
376 		ul->un_flags |= LDL_NOROLL;
377 	ul->un_bp = bp;
378 	ul->un_ufsvfs = ufsvfsp;
379 	ul->un_dev = ufsvfsp->vfs_dev;
380 	ul->un_ebp = nebp;
381 	ul->un_nbeb = nb;
382 	ul->un_maxresv = btodb(ul->un_logsize) * LDL_USABLE_BSIZE;
383 	ul->un_deltamap = map_get(ul, deltamaptype, DELTAMAP_NHASH);
384 	ul->un_logmap = map_get(ul, logmaptype, LOGMAP_NHASH);
385 	if (ul->un_debug & MT_MATAMAP)
386 		ul->un_matamap = map_get(ul, matamaptype, DELTAMAP_NHASH);
387 	mutex_init(&ul->un_log_mutex, NULL, MUTEX_DEFAULT, NULL);
388 	mutex_init(&ul->un_state_mutex, NULL, MUTEX_DEFAULT, NULL);
389 
390 	/*
391 	 * Aquire the ufs_scan_lock before linking the mtm data
392 	 * structure so that we keep ufs_sync() and ufs_update() away
393 	 * when they execute the ufs_scan_inodes() run while we're in
394 	 * progress of enabling/disabling logging.
395 	 */
396 	mutex_enter(&ufs_scan_lock);
397 	ufsvfsp->vfs_log = ul;
398 
399 	/* remember the state of the log before the log scan */
400 	logmap_logscan(ul);
401 	mutex_exit(&ufs_scan_lock);
402 
403 	/*
404 	 * Error during scan
405 	 *
406 	 * If this is a read/only mount; ignore the error.
407 	 * At a later time umount/fsck will repair the fs.
408 	 *
409 	 */
410 	if (ul->un_flags & LDL_ERROR) {
411 		if (!ronly) {
412 			/*
413 			 * Aquire the ufs_scan_lock before de-linking
414 			 * the mtm data structure so that we keep ufs_sync()
415 			 * and ufs_update() away when they execute the
416 			 * ufs_scan_inodes() run while we're in progress of
417 			 * enabling/disabling logging.
418 			 */
419 			mutex_enter(&ufs_scan_lock);
420 			lufs_unsnarf(ufsvfsp);
421 			mutex_exit(&ufs_scan_lock);
422 			return (EIO);
423 		}
424 		ul->un_flags &= ~LDL_ERROR;
425 	}
426 	if (!ronly)
427 		logmap_start_roll(ul);
428 	return (0);
429 }
430 
431 static int
432 lufs_initialize(
433 	ufsvfs_t *ufsvfsp,
434 	daddr_t bno,
435 	size_t nb,
436 	struct fiolog *flp)
437 {
438 	ml_odunit_t	*ud, *ud2;
439 	buf_t		*bp;
440 	struct timeval	tv;
441 
442 	/* LINTED: warning: logical expression always true: op "||" */
443 	ASSERT(sizeof (ml_odunit_t) < DEV_BSIZE);
444 	ASSERT(nb >= ldl_minlogsize);
445 
446 	bp = UFS_GETBLK(ufsvfsp, ufsvfsp->vfs_dev, bno, dbtob(LS_SECTORS));
447 	bzero(bp->b_un.b_addr, bp->b_bcount);
448 
449 	ud = (void *)bp->b_un.b_addr;
450 	ud->od_version = LUFS_VERSION_LATEST;
451 	ud->od_maxtransfer = MIN(ufsvfsp->vfs_iotransz, ldl_maxtransfer);
452 	if (ud->od_maxtransfer < ldl_mintransfer)
453 		ud->od_maxtransfer = ldl_mintransfer;
454 	ud->od_devbsize = DEV_BSIZE;
455 
456 	ud->od_requestsize = flp->nbytes_actual;
457 	ud->od_statesize = dbtob(LS_SECTORS);
458 	ud->od_logsize = nb - ud->od_statesize;
459 
460 	ud->od_statebno = INT32_C(0);
461 
462 	uniqtime(&tv);
463 	if (tv.tv_usec == last_loghead_ident) {
464 		tv.tv_usec++;
465 	}
466 	last_loghead_ident = tv.tv_usec;
467 	ud->od_head_ident = tv.tv_usec;
468 	ud->od_tail_ident = ud->od_head_ident;
469 	ud->od_chksum = ud->od_head_ident + ud->od_tail_ident;
470 
471 	ud->od_bol_lof = dbtob(ud->od_statebno) + ud->od_statesize;
472 	ud->od_eol_lof = ud->od_bol_lof + ud->od_logsize;
473 	ud->od_head_lof = ud->od_bol_lof;
474 	ud->od_tail_lof = ud->od_bol_lof;
475 
476 	ASSERT(lufs_initialize_debug(ud));
477 
478 	ud2 = (void *)(bp->b_un.b_addr + DEV_BSIZE);
479 	bcopy(ud, ud2, sizeof (*ud));
480 
481 	UFS_BWRITE2(ufsvfsp, bp);
482 	if (bp->b_flags & B_ERROR) {
483 		brelse(bp);
484 		return (EIO);
485 	}
486 	brelse(bp);
487 
488 	return (0);
489 }
490 
491 /*
492  * Free log space
493  *	Assumes the file system is write locked and is not logging
494  */
495 static int
496 lufs_free(struct ufsvfs *ufsvfsp)
497 {
498 	int		error = 0, i, j;
499 	buf_t		*bp = NULL;
500 	extent_t	*ep;
501 	extent_block_t	*ebp;
502 	struct fs	*fs = ufsvfsp->vfs_fs;
503 	daddr_t		fno;
504 	int32_t		logbno;
505 	long		nfno;
506 	inode_t		*ip = NULL;
507 	char		clean;
508 
509 	/*
510 	 * Nothing to free
511 	 */
512 	if (fs->fs_logbno == 0)
513 		return (0);
514 
515 	/*
516 	 * Mark the file system as FSACTIVE and no log but honor the
517 	 * current value of fs_reclaim.  The reclaim thread could have
518 	 * been active when lufs_disable() was called and if fs_reclaim
519 	 * is reset to zero here it could lead to lost inodes.
520 	 */
521 	ufsvfsp->vfs_ulockfs.ul_sbowner = curthread;
522 	mutex_enter(&ufsvfsp->vfs_lock);
523 	clean = fs->fs_clean;
524 	logbno = fs->fs_logbno;
525 	fs->fs_clean = FSACTIVE;
526 	fs->fs_logbno = INT32_C(0);
527 	ufs_sbwrite(ufsvfsp);
528 	mutex_exit(&ufsvfsp->vfs_lock);
529 	ufsvfsp->vfs_ulockfs.ul_sbowner = (kthread_id_t)-1;
530 	if (ufsvfsp->vfs_bufp->b_flags & B_ERROR) {
531 		error = EIO;
532 		fs->fs_clean = clean;
533 		fs->fs_logbno = logbno;
534 		goto errout;
535 	}
536 
537 	/*
538 	 * fetch the allocation block
539 	 *	superblock -> one block of extents -> log data
540 	 */
541 	bp = UFS_BREAD(ufsvfsp, ufsvfsp->vfs_dev, logbtodb(fs, logbno),
542 	    fs->fs_bsize);
543 	if (bp->b_flags & B_ERROR) {
544 		error = EIO;
545 		goto errout;
546 	}
547 
548 	/*
549 	 * Free up the allocated space (dummy inode needed for free())
550 	 */
551 	ip = ufs_alloc_inode(ufsvfsp, UFSROOTINO);
552 	ebp = (void *)bp->b_un.b_addr;
553 	for (i = 0, ep = &ebp->extents[0]; i < ebp->nextents; ++i, ++ep) {
554 		fno = logbtofrag(fs, ep->pbno);
555 		nfno = dbtofsb(fs, ep->nbno);
556 		for (j = 0; j < nfno; j += fs->fs_frag, fno += fs->fs_frag)
557 			free(ip, fno, fs->fs_bsize, 0);
558 	}
559 	free(ip, logbtofrag(fs, logbno), fs->fs_bsize, 0);
560 	brelse(bp);
561 	bp = NULL;
562 
563 	/*
564 	 * Push the metadata dirtied during the allocations
565 	 */
566 	ufsvfsp->vfs_ulockfs.ul_sbowner = curthread;
567 	sbupdate(ufsvfsp->vfs_vfs);
568 	ufsvfsp->vfs_ulockfs.ul_sbowner = (kthread_id_t)-1;
569 	bflush(ufsvfsp->vfs_dev);
570 	error = bfinval(ufsvfsp->vfs_dev, 0);
571 	if (error)
572 		goto errout;
573 
574 	/*
575 	 * Free the dummy inode
576 	 */
577 	ufs_free_inode(ip);
578 
579 	return (0);
580 
581 errout:
582 	/*
583 	 * Free up all resources
584 	 */
585 	if (bp)
586 		brelse(bp);
587 	if (ip)
588 		ufs_free_inode(ip);
589 	return (error);
590 }
591 
592 /*
593  * Allocate log space
594  *	Assumes the file system is write locked and is not logging
595  */
596 static int
597 lufs_alloc(struct ufsvfs *ufsvfsp, struct fiolog *flp, cred_t *cr)
598 {
599 	int		error = 0;
600 	buf_t		*bp = NULL;
601 	extent_t	*ep, *nep;
602 	extent_block_t	*ebp;
603 	struct fs	*fs = ufsvfsp->vfs_fs;
604 	daddr_t		fno;	/* in frags */
605 	daddr_t		bno;	/* in disk blocks */
606 	int32_t		logbno = INT32_C(0);	/* will be fs_logbno */
607 	struct inode	*ip = NULL;
608 	size_t		nb = flp->nbytes_actual;
609 	size_t		tb = 0;
610 
611 	/*
612 	 * Mark the file system as FSACTIVE
613 	 */
614 	ufsvfsp->vfs_ulockfs.ul_sbowner = curthread;
615 	mutex_enter(&ufsvfsp->vfs_lock);
616 	fs->fs_clean = FSACTIVE;
617 	ufs_sbwrite(ufsvfsp);
618 	mutex_exit(&ufsvfsp->vfs_lock);
619 	ufsvfsp->vfs_ulockfs.ul_sbowner = (kthread_id_t)-1;
620 
621 	/*
622 	 * Allocate the allocation block (need dummy shadow inode;
623 	 * we use a shadow inode so the quota sub-system ignores
624 	 * the block allocations.)
625 	 *	superblock -> one block of extents -> log data
626 	 */
627 	ip = ufs_alloc_inode(ufsvfsp, UFSROOTINO);
628 	ip->i_mode = IFSHAD;		/* make the dummy a shadow inode */
629 	rw_enter(&ip->i_contents, RW_WRITER);
630 	fno = contigpref(ufsvfsp, nb + fs->fs_bsize);
631 	error = alloc(ip, fno, fs->fs_bsize, &fno, cr);
632 	if (error)
633 		goto errout;
634 	bno = fsbtodb(fs, fno);
635 
636 	bp = UFS_BREAD(ufsvfsp, ufsvfsp->vfs_dev, bno, fs->fs_bsize);
637 	if (bp->b_flags & B_ERROR) {
638 		error = EIO;
639 		goto errout;
640 	}
641 
642 	ebp = (void *)bp->b_un.b_addr;
643 	ebp->type = LUFS_EXTENTS;
644 	ebp->nextbno = UINT32_C(0);
645 	ebp->nextents = UINT32_C(0);
646 	ebp->chksum = INT32_C(0);
647 	if (fs->fs_magic == FS_MAGIC)
648 		logbno = bno;
649 	else
650 		logbno = dbtofsb(fs, bno);
651 
652 	/*
653 	 * Initialize the first extent
654 	 */
655 	ep = &ebp->extents[0];
656 	error = alloc(ip, fno + fs->fs_frag, fs->fs_bsize, &fno, cr);
657 	if (error)
658 		goto errout;
659 	bno = fsbtodb(fs, fno);
660 
661 	ep->lbno = UINT32_C(0);
662 	if (fs->fs_magic == FS_MAGIC)
663 		ep->pbno = (uint32_t)bno;
664 	else
665 		ep->pbno = (uint32_t)fno;
666 	ep->nbno = (uint32_t)fsbtodb(fs, fs->fs_frag);
667 	ebp->nextents = UINT32_C(1);
668 	tb = fs->fs_bsize;
669 	nb -= fs->fs_bsize;
670 
671 	while (nb) {
672 		error = alloc(ip, fno + fs->fs_frag, fs->fs_bsize, &fno, cr);
673 		if (error) {
674 			if (tb < ldl_minlogsize)
675 				goto errout;
676 			error = 0;
677 			break;
678 		}
679 		bno = fsbtodb(fs, fno);
680 		if ((daddr_t)((logbtodb(fs, ep->pbno) + ep->nbno) == bno))
681 			ep->nbno += (uint32_t)(fsbtodb(fs, fs->fs_frag));
682 		else {
683 			nep = ep + 1;
684 			if ((caddr_t)(nep + 1) >
685 			    (bp->b_un.b_addr + fs->fs_bsize)) {
686 				free(ip, fno, fs->fs_bsize, 0);
687 				break;
688 			}
689 			nep->lbno = ep->lbno + ep->nbno;
690 			if (fs->fs_magic == FS_MAGIC)
691 				nep->pbno = (uint32_t)bno;
692 			else
693 				nep->pbno = (uint32_t)fno;
694 			nep->nbno = (uint32_t)(fsbtodb(fs, fs->fs_frag));
695 			ebp->nextents++;
696 			ep = nep;
697 		}
698 		tb += fs->fs_bsize;
699 		nb -= fs->fs_bsize;
700 	}
701 	ebp->nbytes = (uint32_t)tb;
702 	setsum(&ebp->chksum, (int32_t *)bp->b_un.b_addr, fs->fs_bsize);
703 	UFS_BWRITE2(ufsvfsp, bp);
704 	if (bp->b_flags & B_ERROR) {
705 		error = EIO;
706 		goto errout;
707 	}
708 	/*
709 	 * Initialize the first two sectors of the log
710 	 */
711 	error = lufs_initialize(ufsvfsp, logbtodb(fs, ebp->extents[0].pbno),
712 	    tb, flp);
713 	if (error)
714 		goto errout;
715 
716 	/*
717 	 * We are done initializing the allocation block and the log
718 	 */
719 	brelse(bp);
720 	bp = NULL;
721 
722 	/*
723 	 * Update the superblock and push the dirty metadata
724 	 */
725 	ufsvfsp->vfs_ulockfs.ul_sbowner = curthread;
726 	sbupdate(ufsvfsp->vfs_vfs);
727 	ufsvfsp->vfs_ulockfs.ul_sbowner = (kthread_id_t)-1;
728 	bflush(ufsvfsp->vfs_dev);
729 	error = bfinval(ufsvfsp->vfs_dev, 1);
730 	if (error)
731 		goto errout;
732 	if (ufsvfsp->vfs_bufp->b_flags & B_ERROR) {
733 		error = EIO;
734 		goto errout;
735 	}
736 
737 	/*
738 	 * Everything is safely on disk; update log space pointer in sb
739 	 */
740 	ufsvfsp->vfs_ulockfs.ul_sbowner = curthread;
741 	mutex_enter(&ufsvfsp->vfs_lock);
742 	fs->fs_logbno = (uint32_t)logbno;
743 	ufs_sbwrite(ufsvfsp);
744 	mutex_exit(&ufsvfsp->vfs_lock);
745 	ufsvfsp->vfs_ulockfs.ul_sbowner = (kthread_id_t)-1;
746 
747 	/*
748 	 * Free the dummy inode
749 	 */
750 	rw_exit(&ip->i_contents);
751 	ufs_free_inode(ip);
752 
753 	/* inform user of real log size */
754 	flp->nbytes_actual = tb;
755 	return (0);
756 
757 errout:
758 	/*
759 	 * Free all resources
760 	 */
761 	if (bp)
762 		brelse(bp);
763 	if (logbno) {
764 		fs->fs_logbno = logbno;
765 		(void) lufs_free(ufsvfsp);
766 	}
767 	if (ip) {
768 		rw_exit(&ip->i_contents);
769 		ufs_free_inode(ip);
770 	}
771 	return (error);
772 }
773 
774 /*
775  * Disable logging
776  */
777 int
778 lufs_disable(vnode_t *vp, struct fiolog *flp)
779 {
780 	int		error = 0;
781 	inode_t		*ip = VTOI(vp);
782 	ufsvfs_t	*ufsvfsp = ip->i_ufsvfs;
783 	struct fs	*fs = ufsvfsp->vfs_fs;
784 	struct lockfs	lf;
785 	struct ulockfs	*ulp;
786 
787 	flp->error = FIOLOG_ENONE;
788 
789 	/*
790 	 * Logging is already disabled; done
791 	 */
792 	if (fs->fs_logbno == 0 || ufsvfsp->vfs_log == NULL)
793 		return (0);
794 
795 	/*
796 	 * Readonly file system
797 	 */
798 	if (fs->fs_ronly) {
799 		flp->error = FIOLOG_EROFS;
800 		return (0);
801 	}
802 
803 	/*
804 	 * File system must be write locked to disable logging
805 	 */
806 	error = ufs_fiolfss(vp, &lf);
807 	if (error) {
808 		return (error);
809 	}
810 	if (!LOCKFS_IS_ULOCK(&lf)) {
811 		flp->error = FIOLOG_EULOCK;
812 		return (0);
813 	}
814 	lf.lf_lock = LOCKFS_WLOCK;
815 	lf.lf_flags = 0;
816 	lf.lf_comment = NULL;
817 	error = ufs_fiolfs(vp, &lf, 1);
818 	if (error) {
819 		flp->error = FIOLOG_EWLOCK;
820 		return (0);
821 	}
822 
823 	if (ufsvfsp->vfs_log == NULL || fs->fs_logbno == 0)
824 		goto errout;
825 
826 	/*
827 	 * WE ARE COMMITTED TO DISABLING LOGGING PAST THIS POINT
828 	 */
829 
830 	/*
831 	 * Disable logging:
832 	 * Suspend the reclaim thread and force the delete thread to exit.
833 	 *	When a nologging mount has completed there may still be
834 	 *	work for reclaim to do so just suspend this thread until
835 	 *	it's [deadlock-] safe for it to continue.  The delete
836 	 *	thread won't be needed as ufs_iinactive() calls
837 	 *	ufs_delete() when logging is disabled.
838 	 * Freeze and drain reader ops.
839 	 *	Commit any outstanding reader transactions (ufs_flush).
840 	 *	Set the ``unmounted'' bit in the ufstrans struct.
841 	 *	If debug, remove metadata from matamap.
842 	 *	Disable matamap processing.
843 	 *	NULL the trans ops table.
844 	 *	Free all of the incore structs related to logging.
845 	 * Allow reader ops.
846 	 */
847 	ufs_thread_suspend(&ufsvfsp->vfs_reclaim);
848 	ufs_thread_exit(&ufsvfsp->vfs_delete);
849 
850 	vfs_lock_wait(ufsvfsp->vfs_vfs);
851 	ulp = &ufsvfsp->vfs_ulockfs;
852 	mutex_enter(&ulp->ul_lock);
853 	atomic_add_long(&ufs_quiesce_pend, 1);
854 	(void) ufs_quiesce(ulp);
855 
856 	(void) ufs_flush(ufsvfsp->vfs_vfs);
857 
858 	TRANS_MATA_UMOUNT(ufsvfsp);
859 	ufsvfsp->vfs_domatamap = 0;
860 
861 	/*
862 	 * Free all of the incore structs
863 	 * Aquire the ufs_scan_lock before de-linking the mtm data
864 	 * structure so that we keep ufs_sync() and ufs_update() away
865 	 * when they execute the ufs_scan_inodes() run while we're in
866 	 * progress of enabling/disabling logging.
867 	 */
868 	mutex_enter(&ufs_scan_lock);
869 	(void) lufs_unsnarf(ufsvfsp);
870 	mutex_exit(&ufs_scan_lock);
871 
872 	atomic_add_long(&ufs_quiesce_pend, -1);
873 	mutex_exit(&ulp->ul_lock);
874 	vfs_setmntopt(ufsvfsp->vfs_vfs, MNTOPT_NOLOGGING, NULL, 0);
875 	vfs_unlock(ufsvfsp->vfs_vfs);
876 
877 	fs->fs_rolled = FS_ALL_ROLLED;
878 	ufsvfsp->vfs_nolog_si = 0;
879 
880 	/*
881 	 * Free the log space and mark the superblock as FSACTIVE
882 	 */
883 	(void) lufs_free(ufsvfsp);
884 
885 	/*
886 	 * Allow the reclaim thread to continue.
887 	 */
888 	ufs_thread_continue(&ufsvfsp->vfs_reclaim);
889 
890 	/*
891 	 * Unlock the file system
892 	 */
893 	lf.lf_lock = LOCKFS_ULOCK;
894 	lf.lf_flags = 0;
895 	error = ufs_fiolfs(vp, &lf, 1);
896 	if (error)
897 		flp->error = FIOLOG_ENOULOCK;
898 
899 	return (0);
900 
901 errout:
902 	lf.lf_lock = LOCKFS_ULOCK;
903 	lf.lf_flags = 0;
904 	(void) ufs_fiolfs(vp, &lf, 1);
905 	return (error);
906 }
907 
908 /*
909  * Enable logging
910  */
911 int
912 lufs_enable(struct vnode *vp, struct fiolog *flp, cred_t *cr)
913 {
914 	int		error;
915 	int		reclaim;
916 	inode_t		*ip = VTOI(vp);
917 	ufsvfs_t	*ufsvfsp = ip->i_ufsvfs;
918 	struct fs	*fs;
919 	ml_unit_t	*ul;
920 	struct lockfs	lf;
921 	struct ulockfs	*ulp;
922 	vfs_t		*vfsp = ufsvfsp->vfs_vfs;
923 	uint64_t	tmp_nbytes_actual;
924 
925 	/*
926 	 * Check if logging is already enabled
927 	 */
928 	if (ufsvfsp->vfs_log) {
929 		flp->error = FIOLOG_ETRANS;
930 		/* for root ensure logging option is set */
931 		vfs_setmntopt(vfsp, MNTOPT_LOGGING, NULL, 0);
932 		return (0);
933 	}
934 	fs = ufsvfsp->vfs_fs;
935 
936 	/*
937 	 * Come back here to recheck if we had to disable the log.
938 	 */
939 recheck:
940 	error = 0;
941 	reclaim = 0;
942 	flp->error = FIOLOG_ENONE;
943 
944 	/*
945 	 * Adjust requested log size
946 	 */
947 	flp->nbytes_actual = flp->nbytes_requested;
948 	if (flp->nbytes_actual == 0) {
949 		tmp_nbytes_actual =
950 		    (((uint64_t)fs->fs_size) / ldl_divisor) << fs->fs_fshift;
951 		flp->nbytes_actual = (uint_t)MIN(tmp_nbytes_actual, INT_MAX);
952 	}
953 	flp->nbytes_actual = MAX(flp->nbytes_actual, ldl_minlogsize);
954 	flp->nbytes_actual = MIN(flp->nbytes_actual, ldl_maxlogsize);
955 	flp->nbytes_actual = blkroundup(fs, flp->nbytes_actual);
956 
957 	/*
958 	 * logging is enabled and the log is the right size; done
959 	 */
960 	ul = ufsvfsp->vfs_log;
961 	if (ul && fs->fs_logbno && (flp->nbytes_actual == ul->un_requestsize))
962 			return (0);
963 
964 	/*
965 	 * Readonly file system
966 	 */
967 	if (fs->fs_ronly) {
968 		flp->error = FIOLOG_EROFS;
969 		return (0);
970 	}
971 
972 	/*
973 	 * File system must be write locked to enable logging
974 	 */
975 	error = ufs_fiolfss(vp, &lf);
976 	if (error) {
977 		return (error);
978 	}
979 	if (!LOCKFS_IS_ULOCK(&lf)) {
980 		flp->error = FIOLOG_EULOCK;
981 		return (0);
982 	}
983 	lf.lf_lock = LOCKFS_WLOCK;
984 	lf.lf_flags = 0;
985 	lf.lf_comment = NULL;
986 	error = ufs_fiolfs(vp, &lf, 1);
987 	if (error) {
988 		flp->error = FIOLOG_EWLOCK;
989 		return (0);
990 	}
991 
992 	/*
993 	 * Grab appropriate locks to synchronize with the rest
994 	 * of the system
995 	 */
996 	vfs_lock_wait(vfsp);
997 	ulp = &ufsvfsp->vfs_ulockfs;
998 	mutex_enter(&ulp->ul_lock);
999 
1000 	/*
1001 	 * File system must be fairly consistent to enable logging
1002 	 */
1003 	if (fs->fs_clean != FSLOG &&
1004 	    fs->fs_clean != FSACTIVE &&
1005 	    fs->fs_clean != FSSTABLE &&
1006 	    fs->fs_clean != FSCLEAN) {
1007 		flp->error = FIOLOG_ECLEAN;
1008 		goto unlockout;
1009 	}
1010 
1011 	/*
1012 	 * A write-locked file system is only active if there are
1013 	 * open deleted files; so remember to set FS_RECLAIM later.
1014 	 */
1015 	if (fs->fs_clean == FSACTIVE)
1016 		reclaim = FS_RECLAIM;
1017 
1018 	/*
1019 	 * Logging is already enabled; must be changing the log's size
1020 	 */
1021 	if (fs->fs_logbno && ufsvfsp->vfs_log) {
1022 		/*
1023 		 * Before we can disable logging, we must give up our
1024 		 * lock.  As a consequence of unlocking and disabling the
1025 		 * log, the fs structure may change.  Because of this, when
1026 		 * disabling is complete, we will go back to recheck to
1027 		 * repeat all of the checks that we performed to get to
1028 		 * this point.  Disabling sets fs->fs_logbno to 0, so this
1029 		 * will not put us into an infinite loop.
1030 		 */
1031 		mutex_exit(&ulp->ul_lock);
1032 		vfs_unlock(vfsp);
1033 
1034 		lf.lf_lock = LOCKFS_ULOCK;
1035 		lf.lf_flags = 0;
1036 		error = ufs_fiolfs(vp, &lf, 1);
1037 		if (error) {
1038 			flp->error = FIOLOG_ENOULOCK;
1039 			return (0);
1040 		}
1041 		error = lufs_disable(vp, flp);
1042 		if (error || (flp->error != FIOLOG_ENONE))
1043 			return (0);
1044 		goto recheck;
1045 	}
1046 
1047 	error = lufs_alloc(ufsvfsp, flp, cr);
1048 	if (error)
1049 		goto errout;
1050 
1051 	/*
1052 	 * Create all of the incore structs
1053 	 */
1054 	error = lufs_snarf(ufsvfsp, fs, 0);
1055 	if (error)
1056 		goto errout;
1057 
1058 	/*
1059 	 * DON'T ``GOTO ERROUT'' PAST THIS POINT
1060 	 */
1061 
1062 	/*
1063 	 * Pretend we were just mounted with logging enabled
1064 	 *		Get the ops vector
1065 	 *		If debug, record metadata locations with log subsystem
1066 	 *		Start the delete thread
1067 	 *		Start the reclaim thread, if necessary
1068 	 */
1069 	vfs_setmntopt(vfsp, MNTOPT_LOGGING, NULL, 0);
1070 
1071 	TRANS_DOMATAMAP(ufsvfsp);
1072 	TRANS_MATA_MOUNT(ufsvfsp);
1073 	TRANS_MATA_SI(ufsvfsp, fs);
1074 	ufs_thread_start(&ufsvfsp->vfs_delete, ufs_thread_delete, vfsp);
1075 	if (fs->fs_reclaim & (FS_RECLAIM|FS_RECLAIMING)) {
1076 		fs->fs_reclaim &= ~FS_RECLAIM;
1077 		fs->fs_reclaim |=  FS_RECLAIMING;
1078 		ufs_thread_start(&ufsvfsp->vfs_reclaim,
1079 					ufs_thread_reclaim, vfsp);
1080 	} else
1081 		fs->fs_reclaim |= reclaim;
1082 
1083 	mutex_exit(&ulp->ul_lock);
1084 	vfs_unlock(vfsp);
1085 
1086 	/*
1087 	 * Unlock the file system
1088 	 */
1089 	lf.lf_lock = LOCKFS_ULOCK;
1090 	lf.lf_flags = 0;
1091 	error = ufs_fiolfs(vp, &lf, 1);
1092 	if (error) {
1093 		flp->error = FIOLOG_ENOULOCK;
1094 		return (0);
1095 	}
1096 
1097 	/*
1098 	 * There's nothing in the log yet (we've just allocated it)
1099 	 * so directly write out the super block.
1100 	 * Note, we have to force this sb out to disk
1101 	 * (not just to the log) so that if we crash we know we are logging
1102 	 */
1103 	mutex_enter(&ufsvfsp->vfs_lock);
1104 	fs->fs_clean = FSLOG;
1105 	fs->fs_rolled = FS_NEED_ROLL; /* Mark the fs as unrolled */
1106 	UFS_BWRITE2(NULL, ufsvfsp->vfs_bufp);
1107 	mutex_exit(&ufsvfsp->vfs_lock);
1108 
1109 	return (0);
1110 
1111 errout:
1112 	/*
1113 	 * Aquire the ufs_scan_lock before de-linking the mtm data
1114 	 * structure so that we keep ufs_sync() and ufs_update() away
1115 	 * when they execute the ufs_scan_inodes() run while we're in
1116 	 * progress of enabling/disabling logging.
1117 	 */
1118 	mutex_enter(&ufs_scan_lock);
1119 	(void) lufs_unsnarf(ufsvfsp);
1120 	mutex_exit(&ufs_scan_lock);
1121 
1122 	(void) lufs_free(ufsvfsp);
1123 unlockout:
1124 	mutex_exit(&ulp->ul_lock);
1125 	vfs_unlock(vfsp);
1126 
1127 	lf.lf_lock = LOCKFS_ULOCK;
1128 	lf.lf_flags = 0;
1129 	(void) ufs_fiolfs(vp, &lf, 1);
1130 	return (error);
1131 }
1132 
1133 void
1134 lufs_read_strategy(ml_unit_t *ul, buf_t *bp)
1135 {
1136 	mt_map_t	*logmap	= ul->un_logmap;
1137 	offset_t	mof	= ldbtob(bp->b_blkno);
1138 	off_t		nb	= bp->b_bcount;
1139 	mapentry_t	*age;
1140 	char		*va;
1141 	int		(*saviodone)();
1142 	int		entire_range;
1143 
1144 	/*
1145 	 * get a linked list of overlapping deltas
1146 	 * returns with &mtm->mtm_rwlock held
1147 	 */
1148 	entire_range = logmap_list_get(logmap, mof, nb, &age);
1149 
1150 	/*
1151 	 * no overlapping deltas were found; read master
1152 	 */
1153 	if (age == NULL) {
1154 		rw_exit(&logmap->mtm_rwlock);
1155 		if (ul->un_flags & LDL_ERROR) {
1156 			bp->b_flags |= B_ERROR;
1157 			bp->b_error = EIO;
1158 			biodone(bp);
1159 		} else {
1160 			ul->un_ufsvfs->vfs_iotstamp = lbolt;
1161 			logstats.ls_lreads.value.ui64++;
1162 			(void) bdev_strategy(bp);
1163 			lwp_stat_update(LWP_STAT_INBLK, 1);
1164 		}
1165 		return;
1166 	}
1167 
1168 	va = bp_mapin_common(bp, VM_SLEEP);
1169 	/*
1170 	 * if necessary, sync read the data from master
1171 	 *	errors are returned in bp
1172 	 */
1173 	if (!entire_range) {
1174 		saviodone = bp->b_iodone;
1175 		bp->b_iodone = trans_not_done;
1176 		logstats.ls_mreads.value.ui64++;
1177 		(void) bdev_strategy(bp);
1178 		lwp_stat_update(LWP_STAT_INBLK, 1);
1179 		if (trans_not_wait(bp))
1180 			ldl_seterror(ul, "Error reading master");
1181 		bp->b_iodone = saviodone;
1182 	}
1183 
1184 	/*
1185 	 * sync read the data from the log
1186 	 *	errors are returned inline
1187 	 */
1188 	if (ldl_read(ul, va, mof, nb, age)) {
1189 		bp->b_flags |= B_ERROR;
1190 		bp->b_error = EIO;
1191 	}
1192 
1193 	/*
1194 	 * unlist the deltas
1195 	 */
1196 	logmap_list_put(logmap, age);
1197 
1198 	/*
1199 	 * all done
1200 	 */
1201 	if (ul->un_flags & LDL_ERROR) {
1202 		bp->b_flags |= B_ERROR;
1203 		bp->b_error = EIO;
1204 	}
1205 	biodone(bp);
1206 }
1207 
1208 void
1209 lufs_write_strategy(ml_unit_t *ul, buf_t *bp)
1210 {
1211 	offset_t	mof	= ldbtob(bp->b_blkno);
1212 	off_t		nb	= bp->b_bcount;
1213 	char		*va;
1214 	mapentry_t	*me;
1215 
1216 	ASSERT((nb & DEV_BMASK) == 0);
1217 	ul->un_logmap->mtm_ref = 1;
1218 
1219 	/*
1220 	 * if there are deltas, move into log
1221 	 */
1222 	me = deltamap_remove(ul->un_deltamap, mof, nb);
1223 	if (me) {
1224 
1225 		va = bp_mapin_common(bp, VM_SLEEP);
1226 
1227 		ASSERT(((ul->un_debug & MT_WRITE_CHECK) == 0) ||
1228 			(ul->un_matamap == NULL)||
1229 			matamap_within(ul->un_matamap, mof, nb));
1230 
1231 		/*
1232 		 * move to logmap
1233 		 */
1234 		if (ufs_crb_enable) {
1235 			logmap_add_buf(ul, va, mof, me,
1236 			    bp->b_un.b_addr, nb);
1237 		} else {
1238 			logmap_add(ul, va, mof, me);
1239 		}
1240 
1241 		if (ul->un_flags & LDL_ERROR) {
1242 			bp->b_flags |= B_ERROR;
1243 			bp->b_error = EIO;
1244 		}
1245 		biodone(bp);
1246 		return;
1247 	}
1248 	if (ul->un_flags & LDL_ERROR) {
1249 		bp->b_flags |= B_ERROR;
1250 		bp->b_error = EIO;
1251 		biodone(bp);
1252 		return;
1253 	}
1254 
1255 	/*
1256 	 * Check that we are not updating metadata, or if so then via B_PHYS.
1257 	 */
1258 	ASSERT((ul->un_matamap == NULL) ||
1259 		!(matamap_overlap(ul->un_matamap, mof, nb) &&
1260 		((bp->b_flags & B_PHYS) == 0)));
1261 
1262 	ul->un_ufsvfs->vfs_iotstamp = lbolt;
1263 	logstats.ls_lwrites.value.ui64++;
1264 
1265 	/* If snapshots are enabled, write through the snapshot driver */
1266 	if (ul->un_ufsvfs->vfs_snapshot)
1267 		fssnap_strategy(&ul->un_ufsvfs->vfs_snapshot, bp);
1268 	else
1269 		(void) bdev_strategy(bp);
1270 
1271 	lwp_stat_update(LWP_STAT_OUBLK, 1);
1272 }
1273 
1274 void
1275 lufs_strategy(ml_unit_t *ul, buf_t *bp)
1276 {
1277 	if (bp->b_flags & B_READ)
1278 		lufs_read_strategy(ul, bp);
1279 	else
1280 		lufs_write_strategy(ul, bp);
1281 }
1282 
1283 /* ARGSUSED */
1284 static int
1285 delta_stats_update(kstat_t *ksp, int rw)
1286 {
1287 	if (rw == KSTAT_WRITE) {
1288 		delta_stats[DT_SB] = dkstats.ds_superblock_deltas.value.ui64;
1289 		delta_stats[DT_CG] = dkstats.ds_bitmap_deltas.value.ui64;
1290 		delta_stats[DT_SI] = dkstats.ds_suminfo_deltas.value.ui64;
1291 		delta_stats[DT_AB] = dkstats.ds_allocblk_deltas.value.ui64;
1292 		delta_stats[DT_ABZERO] = dkstats.ds_ab0_deltas.value.ui64;
1293 		delta_stats[DT_DIR] = dkstats.ds_dir_deltas.value.ui64;
1294 		delta_stats[DT_INODE] = dkstats.ds_inode_deltas.value.ui64;
1295 		delta_stats[DT_FBI] = dkstats.ds_fbiwrite_deltas.value.ui64;
1296 		delta_stats[DT_QR] = dkstats.ds_quota_deltas.value.ui64;
1297 		delta_stats[DT_SHAD] = dkstats.ds_shadow_deltas.value.ui64;
1298 
1299 		roll_stats[DT_SB] = dkstats.ds_superblock_rolled.value.ui64;
1300 		roll_stats[DT_CG] = dkstats.ds_bitmap_rolled.value.ui64;
1301 		roll_stats[DT_SI] = dkstats.ds_suminfo_rolled.value.ui64;
1302 		roll_stats[DT_AB] = dkstats.ds_allocblk_rolled.value.ui64;
1303 		roll_stats[DT_ABZERO] = dkstats.ds_ab0_rolled.value.ui64;
1304 		roll_stats[DT_DIR] = dkstats.ds_dir_rolled.value.ui64;
1305 		roll_stats[DT_INODE] = dkstats.ds_inode_rolled.value.ui64;
1306 		roll_stats[DT_FBI] = dkstats.ds_fbiwrite_rolled.value.ui64;
1307 		roll_stats[DT_QR] = dkstats.ds_quota_rolled.value.ui64;
1308 		roll_stats[DT_SHAD] = dkstats.ds_shadow_rolled.value.ui64;
1309 	} else {
1310 		dkstats.ds_superblock_deltas.value.ui64 = delta_stats[DT_SB];
1311 		dkstats.ds_bitmap_deltas.value.ui64 = delta_stats[DT_CG];
1312 		dkstats.ds_suminfo_deltas.value.ui64 = delta_stats[DT_SI];
1313 		dkstats.ds_allocblk_deltas.value.ui64 = delta_stats[DT_AB];
1314 		dkstats.ds_ab0_deltas.value.ui64 = delta_stats[DT_ABZERO];
1315 		dkstats.ds_dir_deltas.value.ui64 = delta_stats[DT_DIR];
1316 		dkstats.ds_inode_deltas.value.ui64 = delta_stats[DT_INODE];
1317 		dkstats.ds_fbiwrite_deltas.value.ui64 = delta_stats[DT_FBI];
1318 		dkstats.ds_quota_deltas.value.ui64 = delta_stats[DT_QR];
1319 		dkstats.ds_shadow_deltas.value.ui64 = delta_stats[DT_SHAD];
1320 
1321 		dkstats.ds_superblock_rolled.value.ui64 = roll_stats[DT_SB];
1322 		dkstats.ds_bitmap_rolled.value.ui64 = roll_stats[DT_CG];
1323 		dkstats.ds_suminfo_rolled.value.ui64 = roll_stats[DT_SI];
1324 		dkstats.ds_allocblk_rolled.value.ui64 = roll_stats[DT_AB];
1325 		dkstats.ds_ab0_rolled.value.ui64 = roll_stats[DT_ABZERO];
1326 		dkstats.ds_dir_rolled.value.ui64 = roll_stats[DT_DIR];
1327 		dkstats.ds_inode_rolled.value.ui64 = roll_stats[DT_INODE];
1328 		dkstats.ds_fbiwrite_rolled.value.ui64 = roll_stats[DT_FBI];
1329 		dkstats.ds_quota_rolled.value.ui64 = roll_stats[DT_QR];
1330 		dkstats.ds_shadow_rolled.value.ui64 = roll_stats[DT_SHAD];
1331 	}
1332 	return (0);
1333 }
1334 
1335 extern size_t ufs_crb_limit;
1336 extern int ufs_max_crb_divisor;
1337 
1338 void
1339 lufs_init(void)
1340 {
1341 	kstat_t *ksp;
1342 
1343 	/* Create kmem caches */
1344 	lufs_sv = kmem_cache_create("lufs_save", sizeof (lufs_save_t), 0,
1345 	    NULL, NULL, NULL, NULL, NULL, 0);
1346 	lufs_bp = kmem_cache_create("lufs_bufs", sizeof (lufs_buf_t), 0,
1347 	    NULL, NULL, NULL, NULL, NULL, 0);
1348 
1349 	mutex_init(&log_mutex, NULL, MUTEX_DEFAULT, NULL);
1350 
1351 	_init_top();
1352 
1353 	if (&bio_lufs_strategy != NULL)
1354 		bio_lufs_strategy = (void (*) (void *, buf_t *)) lufs_strategy;
1355 
1356 	/*
1357 	 * Initialise general logging and delta kstats
1358 	 */
1359 	ksp = kstat_create("ufs_log", 0, "logstats", "ufs", KSTAT_TYPE_NAMED,
1360 	    sizeof (logstats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL);
1361 	if (ksp) {
1362 		ksp->ks_data = (void *) &logstats;
1363 		kstat_install(ksp);
1364 	}
1365 
1366 	ksp = kstat_create("ufs_log", 0, "deltastats", "ufs", KSTAT_TYPE_NAMED,
1367 	    sizeof (dkstats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL);
1368 	if (ksp) {
1369 		ksp->ks_data = (void *) &dkstats;
1370 		ksp->ks_update = delta_stats_update;
1371 		kstat_install(ksp);
1372 	}
1373 
1374 	/*
1375 	 * Set up the maximum amount of kmem that the crbs (system wide)
1376 	 * can use.
1377 	 */
1378 	ufs_crb_limit = kmem_maxavail() / ufs_max_crb_divisor;
1379 }
1380