xref: /titanic_44/usr/src/uts/common/fs/zfs/zil.c (revision 70f9559bd0c02885d84a425eaafc8c280df10efb)
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 (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2011 by Delphix. All rights reserved.
24  */
25 
26 /* Portions Copyright 2010 Robert Milkowski */
27 
28 #include <sys/zfs_context.h>
29 #include <sys/spa.h>
30 #include <sys/dmu.h>
31 #include <sys/zap.h>
32 #include <sys/arc.h>
33 #include <sys/stat.h>
34 #include <sys/resource.h>
35 #include <sys/zil.h>
36 #include <sys/zil_impl.h>
37 #include <sys/dsl_dataset.h>
38 #include <sys/vdev_impl.h>
39 #include <sys/dmu_tx.h>
40 #include <sys/dsl_pool.h>
41 
42 /*
43  * The zfs intent log (ZIL) saves transaction records of system calls
44  * that change the file system in memory with enough information
45  * to be able to replay them. These are stored in memory until
46  * either the DMU transaction group (txg) commits them to the stable pool
47  * and they can be discarded, or they are flushed to the stable log
48  * (also in the pool) due to a fsync, O_DSYNC or other synchronous
49  * requirement. In the event of a panic or power fail then those log
50  * records (transactions) are replayed.
51  *
52  * There is one ZIL per file system. Its on-disk (pool) format consists
53  * of 3 parts:
54  *
55  * 	- ZIL header
56  * 	- ZIL blocks
57  * 	- ZIL records
58  *
59  * A log record holds a system call transaction. Log blocks can
60  * hold many log records and the blocks are chained together.
61  * Each ZIL block contains a block pointer (blkptr_t) to the next
62  * ZIL block in the chain. The ZIL header points to the first
63  * block in the chain. Note there is not a fixed place in the pool
64  * to hold blocks. They are dynamically allocated and freed as
65  * needed from the blocks available. Figure X shows the ZIL structure:
66  */
67 
68 /*
69  * This global ZIL switch affects all pools
70  */
71 int zil_replay_disable = 0;    /* disable intent logging replay */
72 
73 /*
74  * Tunable parameter for debugging or performance analysis.  Setting
75  * zfs_nocacheflush will cause corruption on power loss if a volatile
76  * out-of-order write cache is enabled.
77  */
78 boolean_t zfs_nocacheflush = B_FALSE;
79 
80 static kmem_cache_t *zil_lwb_cache;
81 
82 static void zil_async_to_sync(zilog_t *zilog, uint64_t foid);
83 
84 #define	LWB_EMPTY(lwb) ((BP_GET_LSIZE(&lwb->lwb_blk) - \
85     sizeof (zil_chain_t)) == (lwb->lwb_sz - lwb->lwb_nused))
86 
87 
88 /*
89  * ziltest is by and large an ugly hack, but very useful in
90  * checking replay without tedious work.
91  * When running ziltest we want to keep all itx's and so maintain
92  * a single list in the zl_itxg[] that uses a high txg: ZILTEST_TXG
93  * We subtract TXG_CONCURRENT_STATES to allow for common code.
94  */
95 #define	ZILTEST_TXG (UINT64_MAX - TXG_CONCURRENT_STATES)
96 
97 static int
98 zil_bp_compare(const void *x1, const void *x2)
99 {
100 	const dva_t *dva1 = &((zil_bp_node_t *)x1)->zn_dva;
101 	const dva_t *dva2 = &((zil_bp_node_t *)x2)->zn_dva;
102 
103 	if (DVA_GET_VDEV(dva1) < DVA_GET_VDEV(dva2))
104 		return (-1);
105 	if (DVA_GET_VDEV(dva1) > DVA_GET_VDEV(dva2))
106 		return (1);
107 
108 	if (DVA_GET_OFFSET(dva1) < DVA_GET_OFFSET(dva2))
109 		return (-1);
110 	if (DVA_GET_OFFSET(dva1) > DVA_GET_OFFSET(dva2))
111 		return (1);
112 
113 	return (0);
114 }
115 
116 static void
117 zil_bp_tree_init(zilog_t *zilog)
118 {
119 	avl_create(&zilog->zl_bp_tree, zil_bp_compare,
120 	    sizeof (zil_bp_node_t), offsetof(zil_bp_node_t, zn_node));
121 }
122 
123 static void
124 zil_bp_tree_fini(zilog_t *zilog)
125 {
126 	avl_tree_t *t = &zilog->zl_bp_tree;
127 	zil_bp_node_t *zn;
128 	void *cookie = NULL;
129 
130 	while ((zn = avl_destroy_nodes(t, &cookie)) != NULL)
131 		kmem_free(zn, sizeof (zil_bp_node_t));
132 
133 	avl_destroy(t);
134 }
135 
136 int
137 zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp)
138 {
139 	avl_tree_t *t = &zilog->zl_bp_tree;
140 	const dva_t *dva = BP_IDENTITY(bp);
141 	zil_bp_node_t *zn;
142 	avl_index_t where;
143 
144 	if (avl_find(t, dva, &where) != NULL)
145 		return (EEXIST);
146 
147 	zn = kmem_alloc(sizeof (zil_bp_node_t), KM_SLEEP);
148 	zn->zn_dva = *dva;
149 	avl_insert(t, zn, where);
150 
151 	return (0);
152 }
153 
154 static zil_header_t *
155 zil_header_in_syncing_context(zilog_t *zilog)
156 {
157 	return ((zil_header_t *)zilog->zl_header);
158 }
159 
160 static void
161 zil_init_log_chain(zilog_t *zilog, blkptr_t *bp)
162 {
163 	zio_cksum_t *zc = &bp->blk_cksum;
164 
165 	zc->zc_word[ZIL_ZC_GUID_0] = spa_get_random(-1ULL);
166 	zc->zc_word[ZIL_ZC_GUID_1] = spa_get_random(-1ULL);
167 	zc->zc_word[ZIL_ZC_OBJSET] = dmu_objset_id(zilog->zl_os);
168 	zc->zc_word[ZIL_ZC_SEQ] = 1ULL;
169 }
170 
171 /*
172  * Read a log block and make sure it's valid.
173  */
174 static int
175 zil_read_log_block(zilog_t *zilog, const blkptr_t *bp, blkptr_t *nbp, void *dst,
176     char **end)
177 {
178 	enum zio_flag zio_flags = ZIO_FLAG_CANFAIL;
179 	uint32_t aflags = ARC_WAIT;
180 	arc_buf_t *abuf = NULL;
181 	zbookmark_t zb;
182 	int error;
183 
184 	if (zilog->zl_header->zh_claim_txg == 0)
185 		zio_flags |= ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB;
186 
187 	if (!(zilog->zl_header->zh_flags & ZIL_CLAIM_LR_SEQ_VALID))
188 		zio_flags |= ZIO_FLAG_SPECULATIVE;
189 
190 	SET_BOOKMARK(&zb, bp->blk_cksum.zc_word[ZIL_ZC_OBJSET],
191 	    ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, bp->blk_cksum.zc_word[ZIL_ZC_SEQ]);
192 
193 	error = dsl_read_nolock(NULL, zilog->zl_spa, bp, arc_getbuf_func, &abuf,
194 	    ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb);
195 
196 	if (error == 0) {
197 		zio_cksum_t cksum = bp->blk_cksum;
198 
199 		/*
200 		 * Validate the checksummed log block.
201 		 *
202 		 * Sequence numbers should be... sequential.  The checksum
203 		 * verifier for the next block should be bp's checksum plus 1.
204 		 *
205 		 * Also check the log chain linkage and size used.
206 		 */
207 		cksum.zc_word[ZIL_ZC_SEQ]++;
208 
209 		if (BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_ZILOG2) {
210 			zil_chain_t *zilc = abuf->b_data;
211 			char *lr = (char *)(zilc + 1);
212 			uint64_t len = zilc->zc_nused - sizeof (zil_chain_t);
213 
214 			if (bcmp(&cksum, &zilc->zc_next_blk.blk_cksum,
215 			    sizeof (cksum)) || BP_IS_HOLE(&zilc->zc_next_blk)) {
216 				error = ECKSUM;
217 			} else {
218 				bcopy(lr, dst, len);
219 				*end = (char *)dst + len;
220 				*nbp = zilc->zc_next_blk;
221 			}
222 		} else {
223 			char *lr = abuf->b_data;
224 			uint64_t size = BP_GET_LSIZE(bp);
225 			zil_chain_t *zilc = (zil_chain_t *)(lr + size) - 1;
226 
227 			if (bcmp(&cksum, &zilc->zc_next_blk.blk_cksum,
228 			    sizeof (cksum)) || BP_IS_HOLE(&zilc->zc_next_blk) ||
229 			    (zilc->zc_nused > (size - sizeof (*zilc)))) {
230 				error = ECKSUM;
231 			} else {
232 				bcopy(lr, dst, zilc->zc_nused);
233 				*end = (char *)dst + zilc->zc_nused;
234 				*nbp = zilc->zc_next_blk;
235 			}
236 		}
237 
238 		VERIFY(arc_buf_remove_ref(abuf, &abuf) == 1);
239 	}
240 
241 	return (error);
242 }
243 
244 /*
245  * Read a TX_WRITE log data block.
246  */
247 static int
248 zil_read_log_data(zilog_t *zilog, const lr_write_t *lr, void *wbuf)
249 {
250 	enum zio_flag zio_flags = ZIO_FLAG_CANFAIL;
251 	const blkptr_t *bp = &lr->lr_blkptr;
252 	uint32_t aflags = ARC_WAIT;
253 	arc_buf_t *abuf = NULL;
254 	zbookmark_t zb;
255 	int error;
256 
257 	if (BP_IS_HOLE(bp)) {
258 		if (wbuf != NULL)
259 			bzero(wbuf, MAX(BP_GET_LSIZE(bp), lr->lr_length));
260 		return (0);
261 	}
262 
263 	if (zilog->zl_header->zh_claim_txg == 0)
264 		zio_flags |= ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB;
265 
266 	SET_BOOKMARK(&zb, dmu_objset_id(zilog->zl_os), lr->lr_foid,
267 	    ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp));
268 
269 	error = arc_read_nolock(NULL, zilog->zl_spa, bp, arc_getbuf_func, &abuf,
270 	    ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb);
271 
272 	if (error == 0) {
273 		if (wbuf != NULL)
274 			bcopy(abuf->b_data, wbuf, arc_buf_size(abuf));
275 		(void) arc_buf_remove_ref(abuf, &abuf);
276 	}
277 
278 	return (error);
279 }
280 
281 /*
282  * Parse the intent log, and call parse_func for each valid record within.
283  */
284 int
285 zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func,
286     zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg)
287 {
288 	const zil_header_t *zh = zilog->zl_header;
289 	boolean_t claimed = !!zh->zh_claim_txg;
290 	uint64_t claim_blk_seq = claimed ? zh->zh_claim_blk_seq : UINT64_MAX;
291 	uint64_t claim_lr_seq = claimed ? zh->zh_claim_lr_seq : UINT64_MAX;
292 	uint64_t max_blk_seq = 0;
293 	uint64_t max_lr_seq = 0;
294 	uint64_t blk_count = 0;
295 	uint64_t lr_count = 0;
296 	blkptr_t blk, next_blk;
297 	char *lrbuf, *lrp;
298 	int error = 0;
299 
300 	/*
301 	 * Old logs didn't record the maximum zh_claim_lr_seq.
302 	 */
303 	if (!(zh->zh_flags & ZIL_CLAIM_LR_SEQ_VALID))
304 		claim_lr_seq = UINT64_MAX;
305 
306 	/*
307 	 * Starting at the block pointed to by zh_log we read the log chain.
308 	 * For each block in the chain we strongly check that block to
309 	 * ensure its validity.  We stop when an invalid block is found.
310 	 * For each block pointer in the chain we call parse_blk_func().
311 	 * For each record in each valid block we call parse_lr_func().
312 	 * If the log has been claimed, stop if we encounter a sequence
313 	 * number greater than the highest claimed sequence number.
314 	 */
315 	lrbuf = zio_buf_alloc(SPA_MAXBLOCKSIZE);
316 	zil_bp_tree_init(zilog);
317 
318 	for (blk = zh->zh_log; !BP_IS_HOLE(&blk); blk = next_blk) {
319 		uint64_t blk_seq = blk.blk_cksum.zc_word[ZIL_ZC_SEQ];
320 		int reclen;
321 		char *end;
322 
323 		if (blk_seq > claim_blk_seq)
324 			break;
325 		if ((error = parse_blk_func(zilog, &blk, arg, txg)) != 0)
326 			break;
327 		ASSERT3U(max_blk_seq, <, blk_seq);
328 		max_blk_seq = blk_seq;
329 		blk_count++;
330 
331 		if (max_lr_seq == claim_lr_seq && max_blk_seq == claim_blk_seq)
332 			break;
333 
334 		error = zil_read_log_block(zilog, &blk, &next_blk, lrbuf, &end);
335 		if (error)
336 			break;
337 
338 		for (lrp = lrbuf; lrp < end; lrp += reclen) {
339 			lr_t *lr = (lr_t *)lrp;
340 			reclen = lr->lrc_reclen;
341 			ASSERT3U(reclen, >=, sizeof (lr_t));
342 			if (lr->lrc_seq > claim_lr_seq)
343 				goto done;
344 			if ((error = parse_lr_func(zilog, lr, arg, txg)) != 0)
345 				goto done;
346 			ASSERT3U(max_lr_seq, <, lr->lrc_seq);
347 			max_lr_seq = lr->lrc_seq;
348 			lr_count++;
349 		}
350 	}
351 done:
352 	zilog->zl_parse_error = error;
353 	zilog->zl_parse_blk_seq = max_blk_seq;
354 	zilog->zl_parse_lr_seq = max_lr_seq;
355 	zilog->zl_parse_blk_count = blk_count;
356 	zilog->zl_parse_lr_count = lr_count;
357 
358 	ASSERT(!claimed || !(zh->zh_flags & ZIL_CLAIM_LR_SEQ_VALID) ||
359 	    (max_blk_seq == claim_blk_seq && max_lr_seq == claim_lr_seq));
360 
361 	zil_bp_tree_fini(zilog);
362 	zio_buf_free(lrbuf, SPA_MAXBLOCKSIZE);
363 
364 	return (error);
365 }
366 
367 static int
368 zil_claim_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t first_txg)
369 {
370 	/*
371 	 * Claim log block if not already committed and not already claimed.
372 	 * If tx == NULL, just verify that the block is claimable.
373 	 */
374 	if (bp->blk_birth < first_txg || zil_bp_tree_add(zilog, bp) != 0)
375 		return (0);
376 
377 	return (zio_wait(zio_claim(NULL, zilog->zl_spa,
378 	    tx == NULL ? 0 : first_txg, bp, spa_claim_notify, NULL,
379 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB)));
380 }
381 
382 static int
383 zil_claim_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t first_txg)
384 {
385 	lr_write_t *lr = (lr_write_t *)lrc;
386 	int error;
387 
388 	if (lrc->lrc_txtype != TX_WRITE)
389 		return (0);
390 
391 	/*
392 	 * If the block is not readable, don't claim it.  This can happen
393 	 * in normal operation when a log block is written to disk before
394 	 * some of the dmu_sync() blocks it points to.  In this case, the
395 	 * transaction cannot have been committed to anyone (we would have
396 	 * waited for all writes to be stable first), so it is semantically
397 	 * correct to declare this the end of the log.
398 	 */
399 	if (lr->lr_blkptr.blk_birth >= first_txg &&
400 	    (error = zil_read_log_data(zilog, lr, NULL)) != 0)
401 		return (error);
402 	return (zil_claim_log_block(zilog, &lr->lr_blkptr, tx, first_txg));
403 }
404 
405 /* ARGSUSED */
406 static int
407 zil_free_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t claim_txg)
408 {
409 	zio_free_zil(zilog->zl_spa, dmu_tx_get_txg(tx), bp);
410 
411 	return (0);
412 }
413 
414 static int
415 zil_free_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t claim_txg)
416 {
417 	lr_write_t *lr = (lr_write_t *)lrc;
418 	blkptr_t *bp = &lr->lr_blkptr;
419 
420 	/*
421 	 * If we previously claimed it, we need to free it.
422 	 */
423 	if (claim_txg != 0 && lrc->lrc_txtype == TX_WRITE &&
424 	    bp->blk_birth >= claim_txg && zil_bp_tree_add(zilog, bp) == 0)
425 		zio_free(zilog->zl_spa, dmu_tx_get_txg(tx), bp);
426 
427 	return (0);
428 }
429 
430 static lwb_t *
431 zil_alloc_lwb(zilog_t *zilog, blkptr_t *bp, uint64_t txg)
432 {
433 	lwb_t *lwb;
434 
435 	lwb = kmem_cache_alloc(zil_lwb_cache, KM_SLEEP);
436 	lwb->lwb_zilog = zilog;
437 	lwb->lwb_blk = *bp;
438 	lwb->lwb_buf = zio_buf_alloc(BP_GET_LSIZE(bp));
439 	lwb->lwb_max_txg = txg;
440 	lwb->lwb_zio = NULL;
441 	lwb->lwb_tx = NULL;
442 	if (BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_ZILOG2) {
443 		lwb->lwb_nused = sizeof (zil_chain_t);
444 		lwb->lwb_sz = BP_GET_LSIZE(bp);
445 	} else {
446 		lwb->lwb_nused = 0;
447 		lwb->lwb_sz = BP_GET_LSIZE(bp) - sizeof (zil_chain_t);
448 	}
449 
450 	mutex_enter(&zilog->zl_lock);
451 	list_insert_tail(&zilog->zl_lwb_list, lwb);
452 	mutex_exit(&zilog->zl_lock);
453 
454 	return (lwb);
455 }
456 
457 /*
458  * Create an on-disk intent log.
459  */
460 static lwb_t *
461 zil_create(zilog_t *zilog)
462 {
463 	const zil_header_t *zh = zilog->zl_header;
464 	lwb_t *lwb = NULL;
465 	uint64_t txg = 0;
466 	dmu_tx_t *tx = NULL;
467 	blkptr_t blk;
468 	int error = 0;
469 
470 	/*
471 	 * Wait for any previous destroy to complete.
472 	 */
473 	txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg);
474 
475 	ASSERT(zh->zh_claim_txg == 0);
476 	ASSERT(zh->zh_replay_seq == 0);
477 
478 	blk = zh->zh_log;
479 
480 	/*
481 	 * Allocate an initial log block if:
482 	 *    - there isn't one already
483 	 *    - the existing block is the wrong endianess
484 	 */
485 	if (BP_IS_HOLE(&blk) || BP_SHOULD_BYTESWAP(&blk)) {
486 		tx = dmu_tx_create(zilog->zl_os);
487 		VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0);
488 		dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
489 		txg = dmu_tx_get_txg(tx);
490 
491 		if (!BP_IS_HOLE(&blk)) {
492 			zio_free_zil(zilog->zl_spa, txg, &blk);
493 			BP_ZERO(&blk);
494 		}
495 
496 		error = zio_alloc_zil(zilog->zl_spa, txg, &blk, NULL,
497 		    ZIL_MIN_BLKSZ, zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
498 
499 		if (error == 0)
500 			zil_init_log_chain(zilog, &blk);
501 	}
502 
503 	/*
504 	 * Allocate a log write buffer (lwb) for the first log block.
505 	 */
506 	if (error == 0)
507 		lwb = zil_alloc_lwb(zilog, &blk, txg);
508 
509 	/*
510 	 * If we just allocated the first log block, commit our transaction
511 	 * and wait for zil_sync() to stuff the block poiner into zh_log.
512 	 * (zh is part of the MOS, so we cannot modify it in open context.)
513 	 */
514 	if (tx != NULL) {
515 		dmu_tx_commit(tx);
516 		txg_wait_synced(zilog->zl_dmu_pool, txg);
517 	}
518 
519 	ASSERT(bcmp(&blk, &zh->zh_log, sizeof (blk)) == 0);
520 
521 	return (lwb);
522 }
523 
524 /*
525  * In one tx, free all log blocks and clear the log header.
526  * If keep_first is set, then we're replaying a log with no content.
527  * We want to keep the first block, however, so that the first
528  * synchronous transaction doesn't require a txg_wait_synced()
529  * in zil_create().  We don't need to txg_wait_synced() here either
530  * when keep_first is set, because both zil_create() and zil_destroy()
531  * will wait for any in-progress destroys to complete.
532  */
533 void
534 zil_destroy(zilog_t *zilog, boolean_t keep_first)
535 {
536 	const zil_header_t *zh = zilog->zl_header;
537 	lwb_t *lwb;
538 	dmu_tx_t *tx;
539 	uint64_t txg;
540 
541 	/*
542 	 * Wait for any previous destroy to complete.
543 	 */
544 	txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg);
545 
546 	zilog->zl_old_header = *zh;		/* debugging aid */
547 
548 	if (BP_IS_HOLE(&zh->zh_log))
549 		return;
550 
551 	tx = dmu_tx_create(zilog->zl_os);
552 	VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0);
553 	dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
554 	txg = dmu_tx_get_txg(tx);
555 
556 	mutex_enter(&zilog->zl_lock);
557 
558 	ASSERT3U(zilog->zl_destroy_txg, <, txg);
559 	zilog->zl_destroy_txg = txg;
560 	zilog->zl_keep_first = keep_first;
561 
562 	if (!list_is_empty(&zilog->zl_lwb_list)) {
563 		ASSERT(zh->zh_claim_txg == 0);
564 		VERIFY(!keep_first);
565 		while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) {
566 			list_remove(&zilog->zl_lwb_list, lwb);
567 			if (lwb->lwb_buf != NULL)
568 				zio_buf_free(lwb->lwb_buf, lwb->lwb_sz);
569 			zio_free_zil(zilog->zl_spa, txg, &lwb->lwb_blk);
570 			kmem_cache_free(zil_lwb_cache, lwb);
571 		}
572 	} else if (!keep_first) {
573 		(void) zil_parse(zilog, zil_free_log_block,
574 		    zil_free_log_record, tx, zh->zh_claim_txg);
575 	}
576 	mutex_exit(&zilog->zl_lock);
577 
578 	dmu_tx_commit(tx);
579 }
580 
581 int
582 zil_claim(const char *osname, void *txarg)
583 {
584 	dmu_tx_t *tx = txarg;
585 	uint64_t first_txg = dmu_tx_get_txg(tx);
586 	zilog_t *zilog;
587 	zil_header_t *zh;
588 	objset_t *os;
589 	int error;
590 
591 	error = dmu_objset_hold(osname, FTAG, &os);
592 	if (error) {
593 		cmn_err(CE_WARN, "can't open objset for %s", osname);
594 		return (0);
595 	}
596 
597 	zilog = dmu_objset_zil(os);
598 	zh = zil_header_in_syncing_context(zilog);
599 
600 	if (spa_get_log_state(zilog->zl_spa) == SPA_LOG_CLEAR) {
601 		if (!BP_IS_HOLE(&zh->zh_log))
602 			zio_free_zil(zilog->zl_spa, first_txg, &zh->zh_log);
603 		BP_ZERO(&zh->zh_log);
604 		dsl_dataset_dirty(dmu_objset_ds(os), tx);
605 		dmu_objset_rele(os, FTAG);
606 		return (0);
607 	}
608 
609 	/*
610 	 * Claim all log blocks if we haven't already done so, and remember
611 	 * the highest claimed sequence number.  This ensures that if we can
612 	 * read only part of the log now (e.g. due to a missing device),
613 	 * but we can read the entire log later, we will not try to replay
614 	 * or destroy beyond the last block we successfully claimed.
615 	 */
616 	ASSERT3U(zh->zh_claim_txg, <=, first_txg);
617 	if (zh->zh_claim_txg == 0 && !BP_IS_HOLE(&zh->zh_log)) {
618 		(void) zil_parse(zilog, zil_claim_log_block,
619 		    zil_claim_log_record, tx, first_txg);
620 		zh->zh_claim_txg = first_txg;
621 		zh->zh_claim_blk_seq = zilog->zl_parse_blk_seq;
622 		zh->zh_claim_lr_seq = zilog->zl_parse_lr_seq;
623 		if (zilog->zl_parse_lr_count || zilog->zl_parse_blk_count > 1)
624 			zh->zh_flags |= ZIL_REPLAY_NEEDED;
625 		zh->zh_flags |= ZIL_CLAIM_LR_SEQ_VALID;
626 		dsl_dataset_dirty(dmu_objset_ds(os), tx);
627 	}
628 
629 	ASSERT3U(first_txg, ==, (spa_last_synced_txg(zilog->zl_spa) + 1));
630 	dmu_objset_rele(os, FTAG);
631 	return (0);
632 }
633 
634 /*
635  * Check the log by walking the log chain.
636  * Checksum errors are ok as they indicate the end of the chain.
637  * Any other error (no device or read failure) returns an error.
638  */
639 int
640 zil_check_log_chain(const char *osname, void *tx)
641 {
642 	zilog_t *zilog;
643 	objset_t *os;
644 	blkptr_t *bp;
645 	int error;
646 
647 	ASSERT(tx == NULL);
648 
649 	error = dmu_objset_hold(osname, FTAG, &os);
650 	if (error) {
651 		cmn_err(CE_WARN, "can't open objset for %s", osname);
652 		return (0);
653 	}
654 
655 	zilog = dmu_objset_zil(os);
656 	bp = (blkptr_t *)&zilog->zl_header->zh_log;
657 
658 	/*
659 	 * Check the first block and determine if it's on a log device
660 	 * which may have been removed or faulted prior to loading this
661 	 * pool.  If so, there's no point in checking the rest of the log
662 	 * as its content should have already been synced to the pool.
663 	 */
664 	if (!BP_IS_HOLE(bp)) {
665 		vdev_t *vd;
666 		boolean_t valid = B_TRUE;
667 
668 		spa_config_enter(os->os_spa, SCL_STATE, FTAG, RW_READER);
669 		vd = vdev_lookup_top(os->os_spa, DVA_GET_VDEV(&bp->blk_dva[0]));
670 		if (vd->vdev_islog && vdev_is_dead(vd))
671 			valid = vdev_log_state_valid(vd);
672 		spa_config_exit(os->os_spa, SCL_STATE, FTAG);
673 
674 		if (!valid) {
675 			dmu_objset_rele(os, FTAG);
676 			return (0);
677 		}
678 	}
679 
680 	/*
681 	 * Because tx == NULL, zil_claim_log_block() will not actually claim
682 	 * any blocks, but just determine whether it is possible to do so.
683 	 * In addition to checking the log chain, zil_claim_log_block()
684 	 * will invoke zio_claim() with a done func of spa_claim_notify(),
685 	 * which will update spa_max_claim_txg.  See spa_load() for details.
686 	 */
687 	error = zil_parse(zilog, zil_claim_log_block, zil_claim_log_record, tx,
688 	    zilog->zl_header->zh_claim_txg ? -1ULL : spa_first_txg(os->os_spa));
689 
690 	dmu_objset_rele(os, FTAG);
691 
692 	return ((error == ECKSUM || error == ENOENT) ? 0 : error);
693 }
694 
695 static int
696 zil_vdev_compare(const void *x1, const void *x2)
697 {
698 	const uint64_t v1 = ((zil_vdev_node_t *)x1)->zv_vdev;
699 	const uint64_t v2 = ((zil_vdev_node_t *)x2)->zv_vdev;
700 
701 	if (v1 < v2)
702 		return (-1);
703 	if (v1 > v2)
704 		return (1);
705 
706 	return (0);
707 }
708 
709 void
710 zil_add_block(zilog_t *zilog, const blkptr_t *bp)
711 {
712 	avl_tree_t *t = &zilog->zl_vdev_tree;
713 	avl_index_t where;
714 	zil_vdev_node_t *zv, zvsearch;
715 	int ndvas = BP_GET_NDVAS(bp);
716 	int i;
717 
718 	if (zfs_nocacheflush)
719 		return;
720 
721 	ASSERT(zilog->zl_writer);
722 
723 	/*
724 	 * Even though we're zl_writer, we still need a lock because the
725 	 * zl_get_data() callbacks may have dmu_sync() done callbacks
726 	 * that will run concurrently.
727 	 */
728 	mutex_enter(&zilog->zl_vdev_lock);
729 	for (i = 0; i < ndvas; i++) {
730 		zvsearch.zv_vdev = DVA_GET_VDEV(&bp->blk_dva[i]);
731 		if (avl_find(t, &zvsearch, &where) == NULL) {
732 			zv = kmem_alloc(sizeof (*zv), KM_SLEEP);
733 			zv->zv_vdev = zvsearch.zv_vdev;
734 			avl_insert(t, zv, where);
735 		}
736 	}
737 	mutex_exit(&zilog->zl_vdev_lock);
738 }
739 
740 static void
741 zil_flush_vdevs(zilog_t *zilog)
742 {
743 	spa_t *spa = zilog->zl_spa;
744 	avl_tree_t *t = &zilog->zl_vdev_tree;
745 	void *cookie = NULL;
746 	zil_vdev_node_t *zv;
747 	zio_t *zio;
748 
749 	ASSERT(zilog->zl_writer);
750 
751 	/*
752 	 * We don't need zl_vdev_lock here because we're the zl_writer,
753 	 * and all zl_get_data() callbacks are done.
754 	 */
755 	if (avl_numnodes(t) == 0)
756 		return;
757 
758 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
759 
760 	zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
761 
762 	while ((zv = avl_destroy_nodes(t, &cookie)) != NULL) {
763 		vdev_t *vd = vdev_lookup_top(spa, zv->zv_vdev);
764 		if (vd != NULL)
765 			zio_flush(zio, vd);
766 		kmem_free(zv, sizeof (*zv));
767 	}
768 
769 	/*
770 	 * Wait for all the flushes to complete.  Not all devices actually
771 	 * support the DKIOCFLUSHWRITECACHE ioctl, so it's OK if it fails.
772 	 */
773 	(void) zio_wait(zio);
774 
775 	spa_config_exit(spa, SCL_STATE, FTAG);
776 }
777 
778 /*
779  * Function called when a log block write completes
780  */
781 static void
782 zil_lwb_write_done(zio_t *zio)
783 {
784 	lwb_t *lwb = zio->io_private;
785 	zilog_t *zilog = lwb->lwb_zilog;
786 	dmu_tx_t *tx = lwb->lwb_tx;
787 
788 	ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF);
789 	ASSERT(BP_GET_TYPE(zio->io_bp) == DMU_OT_INTENT_LOG);
790 	ASSERT(BP_GET_LEVEL(zio->io_bp) == 0);
791 	ASSERT(BP_GET_BYTEORDER(zio->io_bp) == ZFS_HOST_BYTEORDER);
792 	ASSERT(!BP_IS_GANG(zio->io_bp));
793 	ASSERT(!BP_IS_HOLE(zio->io_bp));
794 	ASSERT(zio->io_bp->blk_fill == 0);
795 
796 	/*
797 	 * Ensure the lwb buffer pointer is cleared before releasing
798 	 * the txg. If we have had an allocation failure and
799 	 * the txg is waiting to sync then we want want zil_sync()
800 	 * to remove the lwb so that it's not picked up as the next new
801 	 * one in zil_commit_writer(). zil_sync() will only remove
802 	 * the lwb if lwb_buf is null.
803 	 */
804 	zio_buf_free(lwb->lwb_buf, lwb->lwb_sz);
805 	mutex_enter(&zilog->zl_lock);
806 	lwb->lwb_buf = NULL;
807 	lwb->lwb_tx = NULL;
808 	mutex_exit(&zilog->zl_lock);
809 
810 	/*
811 	 * Now that we've written this log block, we have a stable pointer
812 	 * to the next block in the chain, so it's OK to let the txg in
813 	 * which we allocated the next block sync.
814 	 */
815 	dmu_tx_commit(tx);
816 }
817 
818 /*
819  * Initialize the io for a log block.
820  */
821 static void
822 zil_lwb_write_init(zilog_t *zilog, lwb_t *lwb)
823 {
824 	zbookmark_t zb;
825 
826 	SET_BOOKMARK(&zb, lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_OBJSET],
827 	    ZB_ZIL_OBJECT, ZB_ZIL_LEVEL,
828 	    lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_SEQ]);
829 
830 	if (zilog->zl_root_zio == NULL) {
831 		zilog->zl_root_zio = zio_root(zilog->zl_spa, NULL, NULL,
832 		    ZIO_FLAG_CANFAIL);
833 	}
834 	if (lwb->lwb_zio == NULL) {
835 		lwb->lwb_zio = zio_rewrite(zilog->zl_root_zio, zilog->zl_spa,
836 		    0, &lwb->lwb_blk, lwb->lwb_buf, BP_GET_LSIZE(&lwb->lwb_blk),
837 		    zil_lwb_write_done, lwb, ZIO_PRIORITY_LOG_WRITE,
838 		    ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE, &zb);
839 	}
840 }
841 
842 /*
843  * Define a limited set of intent log block sizes.
844  * These must be a multiple of 4KB. Note only the amount used (again
845  * aligned to 4KB) actually gets written. However, we can't always just
846  * allocate SPA_MAXBLOCKSIZE as the slog space could be exhausted.
847  */
848 uint64_t zil_block_buckets[] = {
849     4096,		/* non TX_WRITE */
850     8192+4096,		/* data base */
851     32*1024 + 4096, 	/* NFS writes */
852     UINT64_MAX
853 };
854 
855 /*
856  * Use the slog as long as the logbias is 'latency' and the current commit size
857  * is less than the limit or the total list size is less than 2X the limit.
858  * Limit checking is disabled by setting zil_slog_limit to UINT64_MAX.
859  */
860 uint64_t zil_slog_limit = 1024 * 1024;
861 #define	USE_SLOG(zilog) (((zilog)->zl_logbias == ZFS_LOGBIAS_LATENCY) && \
862 	(((zilog)->zl_cur_used < zil_slog_limit) || \
863 	((zilog)->zl_itx_list_sz < (zil_slog_limit << 1))))
864 
865 /*
866  * Start a log block write and advance to the next log block.
867  * Calls are serialized.
868  */
869 static lwb_t *
870 zil_lwb_write_start(zilog_t *zilog, lwb_t *lwb)
871 {
872 	lwb_t *nlwb = NULL;
873 	zil_chain_t *zilc;
874 	spa_t *spa = zilog->zl_spa;
875 	blkptr_t *bp;
876 	dmu_tx_t *tx;
877 	uint64_t txg;
878 	uint64_t zil_blksz, wsz;
879 	int i, error;
880 
881 	if (BP_GET_CHECKSUM(&lwb->lwb_blk) == ZIO_CHECKSUM_ZILOG2) {
882 		zilc = (zil_chain_t *)lwb->lwb_buf;
883 		bp = &zilc->zc_next_blk;
884 	} else {
885 		zilc = (zil_chain_t *)(lwb->lwb_buf + lwb->lwb_sz);
886 		bp = &zilc->zc_next_blk;
887 	}
888 
889 	ASSERT(lwb->lwb_nused <= lwb->lwb_sz);
890 
891 	/*
892 	 * Allocate the next block and save its address in this block
893 	 * before writing it in order to establish the log chain.
894 	 * Note that if the allocation of nlwb synced before we wrote
895 	 * the block that points at it (lwb), we'd leak it if we crashed.
896 	 * Therefore, we don't do dmu_tx_commit() until zil_lwb_write_done().
897 	 * We dirty the dataset to ensure that zil_sync() will be called
898 	 * to clean up in the event of allocation failure or I/O failure.
899 	 */
900 	tx = dmu_tx_create(zilog->zl_os);
901 	VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0);
902 	dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
903 	txg = dmu_tx_get_txg(tx);
904 
905 	lwb->lwb_tx = tx;
906 
907 	/*
908 	 * Log blocks are pre-allocated. Here we select the size of the next
909 	 * block, based on size used in the last block.
910 	 * - first find the smallest bucket that will fit the block from a
911 	 *   limited set of block sizes. This is because it's faster to write
912 	 *   blocks allocated from the same metaslab as they are adjacent or
913 	 *   close.
914 	 * - next find the maximum from the new suggested size and an array of
915 	 *   previous sizes. This lessens a picket fence effect of wrongly
916 	 *   guesssing the size if we have a stream of say 2k, 64k, 2k, 64k
917 	 *   requests.
918 	 *
919 	 * Note we only write what is used, but we can't just allocate
920 	 * the maximum block size because we can exhaust the available
921 	 * pool log space.
922 	 */
923 	zil_blksz = zilog->zl_cur_used + sizeof (zil_chain_t);
924 	for (i = 0; zil_blksz > zil_block_buckets[i]; i++)
925 		continue;
926 	zil_blksz = zil_block_buckets[i];
927 	if (zil_blksz == UINT64_MAX)
928 		zil_blksz = SPA_MAXBLOCKSIZE;
929 	zilog->zl_prev_blks[zilog->zl_prev_rotor] = zil_blksz;
930 	for (i = 0; i < ZIL_PREV_BLKS; i++)
931 		zil_blksz = MAX(zil_blksz, zilog->zl_prev_blks[i]);
932 	zilog->zl_prev_rotor = (zilog->zl_prev_rotor + 1) & (ZIL_PREV_BLKS - 1);
933 
934 	BP_ZERO(bp);
935 	/* pass the old blkptr in order to spread log blocks across devs */
936 	error = zio_alloc_zil(spa, txg, bp, &lwb->lwb_blk, zil_blksz,
937 	    USE_SLOG(zilog));
938 	if (!error) {
939 		ASSERT3U(bp->blk_birth, ==, txg);
940 		bp->blk_cksum = lwb->lwb_blk.blk_cksum;
941 		bp->blk_cksum.zc_word[ZIL_ZC_SEQ]++;
942 
943 		/*
944 		 * Allocate a new log write buffer (lwb).
945 		 */
946 		nlwb = zil_alloc_lwb(zilog, bp, txg);
947 
948 		/* Record the block for later vdev flushing */
949 		zil_add_block(zilog, &lwb->lwb_blk);
950 	}
951 
952 	if (BP_GET_CHECKSUM(&lwb->lwb_blk) == ZIO_CHECKSUM_ZILOG2) {
953 		/* For Slim ZIL only write what is used. */
954 		wsz = P2ROUNDUP_TYPED(lwb->lwb_nused, ZIL_MIN_BLKSZ, uint64_t);
955 		ASSERT3U(wsz, <=, lwb->lwb_sz);
956 		zio_shrink(lwb->lwb_zio, wsz);
957 
958 	} else {
959 		wsz = lwb->lwb_sz;
960 	}
961 
962 	zilc->zc_pad = 0;
963 	zilc->zc_nused = lwb->lwb_nused;
964 	zilc->zc_eck.zec_cksum = lwb->lwb_blk.blk_cksum;
965 
966 	/*
967 	 * clear unused data for security
968 	 */
969 	bzero(lwb->lwb_buf + lwb->lwb_nused, wsz - lwb->lwb_nused);
970 
971 	zio_nowait(lwb->lwb_zio); /* Kick off the write for the old log block */
972 
973 	/*
974 	 * If there was an allocation failure then nlwb will be null which
975 	 * forces a txg_wait_synced().
976 	 */
977 	return (nlwb);
978 }
979 
980 static lwb_t *
981 zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb)
982 {
983 	lr_t *lrc = &itx->itx_lr; /* common log record */
984 	lr_write_t *lrw = (lr_write_t *)lrc;
985 	char *lr_buf;
986 	uint64_t txg = lrc->lrc_txg;
987 	uint64_t reclen = lrc->lrc_reclen;
988 	uint64_t dlen = 0;
989 
990 	if (lwb == NULL)
991 		return (NULL);
992 
993 	ASSERT(lwb->lwb_buf != NULL);
994 
995 	if (lrc->lrc_txtype == TX_WRITE && itx->itx_wr_state == WR_NEED_COPY)
996 		dlen = P2ROUNDUP_TYPED(
997 		    lrw->lr_length, sizeof (uint64_t), uint64_t);
998 
999 	zilog->zl_cur_used += (reclen + dlen);
1000 
1001 	zil_lwb_write_init(zilog, lwb);
1002 
1003 	/*
1004 	 * If this record won't fit in the current log block, start a new one.
1005 	 */
1006 	if (lwb->lwb_nused + reclen + dlen > lwb->lwb_sz) {
1007 		lwb = zil_lwb_write_start(zilog, lwb);
1008 		if (lwb == NULL)
1009 			return (NULL);
1010 		zil_lwb_write_init(zilog, lwb);
1011 		ASSERT(LWB_EMPTY(lwb));
1012 		if (lwb->lwb_nused + reclen + dlen > lwb->lwb_sz) {
1013 			txg_wait_synced(zilog->zl_dmu_pool, txg);
1014 			return (lwb);
1015 		}
1016 	}
1017 
1018 	lr_buf = lwb->lwb_buf + lwb->lwb_nused;
1019 	bcopy(lrc, lr_buf, reclen);
1020 	lrc = (lr_t *)lr_buf;
1021 	lrw = (lr_write_t *)lrc;
1022 
1023 	/*
1024 	 * If it's a write, fetch the data or get its blkptr as appropriate.
1025 	 */
1026 	if (lrc->lrc_txtype == TX_WRITE) {
1027 		if (txg > spa_freeze_txg(zilog->zl_spa))
1028 			txg_wait_synced(zilog->zl_dmu_pool, txg);
1029 		if (itx->itx_wr_state != WR_COPIED) {
1030 			char *dbuf;
1031 			int error;
1032 
1033 			if (dlen) {
1034 				ASSERT(itx->itx_wr_state == WR_NEED_COPY);
1035 				dbuf = lr_buf + reclen;
1036 				lrw->lr_common.lrc_reclen += dlen;
1037 			} else {
1038 				ASSERT(itx->itx_wr_state == WR_INDIRECT);
1039 				dbuf = NULL;
1040 			}
1041 			error = zilog->zl_get_data(
1042 			    itx->itx_private, lrw, dbuf, lwb->lwb_zio);
1043 			if (error == EIO) {
1044 				txg_wait_synced(zilog->zl_dmu_pool, txg);
1045 				return (lwb);
1046 			}
1047 			if (error) {
1048 				ASSERT(error == ENOENT || error == EEXIST ||
1049 				    error == EALREADY);
1050 				return (lwb);
1051 			}
1052 		}
1053 	}
1054 
1055 	/*
1056 	 * We're actually making an entry, so update lrc_seq to be the
1057 	 * log record sequence number.  Note that this is generally not
1058 	 * equal to the itx sequence number because not all transactions
1059 	 * are synchronous, and sometimes spa_sync() gets there first.
1060 	 */
1061 	lrc->lrc_seq = ++zilog->zl_lr_seq; /* we are single threaded */
1062 	lwb->lwb_nused += reclen + dlen;
1063 	lwb->lwb_max_txg = MAX(lwb->lwb_max_txg, txg);
1064 	ASSERT3U(lwb->lwb_nused, <=, lwb->lwb_sz);
1065 	ASSERT3U(P2PHASE(lwb->lwb_nused, sizeof (uint64_t)), ==, 0);
1066 
1067 	return (lwb);
1068 }
1069 
1070 itx_t *
1071 zil_itx_create(uint64_t txtype, size_t lrsize)
1072 {
1073 	itx_t *itx;
1074 
1075 	lrsize = P2ROUNDUP_TYPED(lrsize, sizeof (uint64_t), size_t);
1076 
1077 	itx = kmem_alloc(offsetof(itx_t, itx_lr) + lrsize, KM_SLEEP);
1078 	itx->itx_lr.lrc_txtype = txtype;
1079 	itx->itx_lr.lrc_reclen = lrsize;
1080 	itx->itx_sod = lrsize; /* if write & WR_NEED_COPY will be increased */
1081 	itx->itx_lr.lrc_seq = 0;	/* defensive */
1082 	itx->itx_sync = B_TRUE;		/* default is synchronous */
1083 
1084 	return (itx);
1085 }
1086 
1087 void
1088 zil_itx_destroy(itx_t *itx)
1089 {
1090 	kmem_free(itx, offsetof(itx_t, itx_lr) + itx->itx_lr.lrc_reclen);
1091 }
1092 
1093 /*
1094  * Free up the sync and async itxs. The itxs_t has already been detached
1095  * so no locks are needed.
1096  */
1097 static void
1098 zil_itxg_clean(itxs_t *itxs)
1099 {
1100 	itx_t *itx;
1101 	list_t *list;
1102 	avl_tree_t *t;
1103 	void *cookie;
1104 	itx_async_node_t *ian;
1105 
1106 	list = &itxs->i_sync_list;
1107 	while ((itx = list_head(list)) != NULL) {
1108 		list_remove(list, itx);
1109 		kmem_free(itx, offsetof(itx_t, itx_lr) +
1110 		    itx->itx_lr.lrc_reclen);
1111 	}
1112 
1113 	cookie = NULL;
1114 	t = &itxs->i_async_tree;
1115 	while ((ian = avl_destroy_nodes(t, &cookie)) != NULL) {
1116 		list = &ian->ia_list;
1117 		while ((itx = list_head(list)) != NULL) {
1118 			list_remove(list, itx);
1119 			kmem_free(itx, offsetof(itx_t, itx_lr) +
1120 			    itx->itx_lr.lrc_reclen);
1121 		}
1122 		list_destroy(list);
1123 		kmem_free(ian, sizeof (itx_async_node_t));
1124 	}
1125 	avl_destroy(t);
1126 
1127 	kmem_free(itxs, sizeof (itxs_t));
1128 }
1129 
1130 static int
1131 zil_aitx_compare(const void *x1, const void *x2)
1132 {
1133 	const uint64_t o1 = ((itx_async_node_t *)x1)->ia_foid;
1134 	const uint64_t o2 = ((itx_async_node_t *)x2)->ia_foid;
1135 
1136 	if (o1 < o2)
1137 		return (-1);
1138 	if (o1 > o2)
1139 		return (1);
1140 
1141 	return (0);
1142 }
1143 
1144 /*
1145  * Remove all async itx with the given oid.
1146  */
1147 static void
1148 zil_remove_async(zilog_t *zilog, uint64_t oid)
1149 {
1150 	uint64_t otxg, txg;
1151 	itx_async_node_t *ian;
1152 	avl_tree_t *t;
1153 	avl_index_t where;
1154 	list_t clean_list;
1155 	itx_t *itx;
1156 
1157 	ASSERT(oid != 0);
1158 	list_create(&clean_list, sizeof (itx_t), offsetof(itx_t, itx_node));
1159 
1160 	if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */
1161 		otxg = ZILTEST_TXG;
1162 	else
1163 		otxg = spa_last_synced_txg(zilog->zl_spa) + 1;
1164 
1165 	for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) {
1166 		itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK];
1167 
1168 		mutex_enter(&itxg->itxg_lock);
1169 		if (itxg->itxg_txg != txg) {
1170 			mutex_exit(&itxg->itxg_lock);
1171 			continue;
1172 		}
1173 
1174 		/*
1175 		 * Locate the object node and append its list.
1176 		 */
1177 		t = &itxg->itxg_itxs->i_async_tree;
1178 		ian = avl_find(t, &oid, &where);
1179 		if (ian != NULL)
1180 			list_move_tail(&clean_list, &ian->ia_list);
1181 		mutex_exit(&itxg->itxg_lock);
1182 	}
1183 	while ((itx = list_head(&clean_list)) != NULL) {
1184 		list_remove(&clean_list, itx);
1185 		kmem_free(itx, offsetof(itx_t, itx_lr) +
1186 		    itx->itx_lr.lrc_reclen);
1187 	}
1188 	list_destroy(&clean_list);
1189 }
1190 
1191 void
1192 zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx)
1193 {
1194 	uint64_t txg;
1195 	itxg_t *itxg;
1196 	itxs_t *itxs, *clean = NULL;
1197 
1198 	/*
1199 	 * Object ids can be re-instantiated in the next txg so
1200 	 * remove any async transactions to avoid future leaks.
1201 	 * This can happen if a fsync occurs on the re-instantiated
1202 	 * object for a WR_INDIRECT or WR_NEED_COPY write, which gets
1203 	 * the new file data and flushes a write record for the old object.
1204 	 */
1205 	if ((itx->itx_lr.lrc_txtype & ~TX_CI) == TX_REMOVE)
1206 		zil_remove_async(zilog, itx->itx_oid);
1207 
1208 	/*
1209 	 * Ensure the data of a renamed file is committed before the rename.
1210 	 */
1211 	if ((itx->itx_lr.lrc_txtype & ~TX_CI) == TX_RENAME)
1212 		zil_async_to_sync(zilog, itx->itx_oid);
1213 
1214 	if (spa_freeze_txg(zilog->zl_spa) !=  UINT64_MAX)
1215 		txg = ZILTEST_TXG;
1216 	else
1217 		txg = dmu_tx_get_txg(tx);
1218 
1219 	itxg = &zilog->zl_itxg[txg & TXG_MASK];
1220 	mutex_enter(&itxg->itxg_lock);
1221 	itxs = itxg->itxg_itxs;
1222 	if (itxg->itxg_txg != txg) {
1223 		if (itxs != NULL) {
1224 			/*
1225 			 * The zil_clean callback hasn't got around to cleaning
1226 			 * this itxg. Save the itxs for release below.
1227 			 * This should be rare.
1228 			 */
1229 			atomic_add_64(&zilog->zl_itx_list_sz, -itxg->itxg_sod);
1230 			itxg->itxg_sod = 0;
1231 			clean = itxg->itxg_itxs;
1232 		}
1233 		ASSERT(itxg->itxg_sod == 0);
1234 		itxg->itxg_txg = txg;
1235 		itxs = itxg->itxg_itxs = kmem_zalloc(sizeof (itxs_t), KM_SLEEP);
1236 
1237 		list_create(&itxs->i_sync_list, sizeof (itx_t),
1238 		    offsetof(itx_t, itx_node));
1239 		avl_create(&itxs->i_async_tree, zil_aitx_compare,
1240 		    sizeof (itx_async_node_t),
1241 		    offsetof(itx_async_node_t, ia_node));
1242 	}
1243 	if (itx->itx_sync) {
1244 		list_insert_tail(&itxs->i_sync_list, itx);
1245 		atomic_add_64(&zilog->zl_itx_list_sz, itx->itx_sod);
1246 		itxg->itxg_sod += itx->itx_sod;
1247 	} else {
1248 		avl_tree_t *t = &itxs->i_async_tree;
1249 		uint64_t foid = ((lr_ooo_t *)&itx->itx_lr)->lr_foid;
1250 		itx_async_node_t *ian;
1251 		avl_index_t where;
1252 
1253 		ian = avl_find(t, &foid, &where);
1254 		if (ian == NULL) {
1255 			ian = kmem_alloc(sizeof (itx_async_node_t), KM_SLEEP);
1256 			list_create(&ian->ia_list, sizeof (itx_t),
1257 			    offsetof(itx_t, itx_node));
1258 			ian->ia_foid = foid;
1259 			avl_insert(t, ian, where);
1260 		}
1261 		list_insert_tail(&ian->ia_list, itx);
1262 	}
1263 
1264 	itx->itx_lr.lrc_txg = dmu_tx_get_txg(tx);
1265 	mutex_exit(&itxg->itxg_lock);
1266 
1267 	/* Release the old itxs now we've dropped the lock */
1268 	if (clean != NULL)
1269 		zil_itxg_clean(clean);
1270 }
1271 
1272 /*
1273  * If there are any in-memory intent log transactions which have now been
1274  * synced then start up a taskq to free them.
1275  */
1276 void
1277 zil_clean(zilog_t *zilog, uint64_t synced_txg)
1278 {
1279 	itxg_t *itxg = &zilog->zl_itxg[synced_txg & TXG_MASK];
1280 	itxs_t *clean_me;
1281 
1282 	mutex_enter(&itxg->itxg_lock);
1283 	if (itxg->itxg_itxs == NULL || itxg->itxg_txg == ZILTEST_TXG) {
1284 		mutex_exit(&itxg->itxg_lock);
1285 		return;
1286 	}
1287 	ASSERT3U(itxg->itxg_txg, <=, synced_txg);
1288 	ASSERT(itxg->itxg_txg != 0);
1289 	ASSERT(zilog->zl_clean_taskq != NULL);
1290 	atomic_add_64(&zilog->zl_itx_list_sz, -itxg->itxg_sod);
1291 	itxg->itxg_sod = 0;
1292 	clean_me = itxg->itxg_itxs;
1293 	itxg->itxg_itxs = NULL;
1294 	itxg->itxg_txg = 0;
1295 	mutex_exit(&itxg->itxg_lock);
1296 	/*
1297 	 * Preferably start a task queue to free up the old itxs but
1298 	 * if taskq_dispatch can't allocate resources to do that then
1299 	 * free it in-line. This should be rare. Note, using TQ_SLEEP
1300 	 * created a bad performance problem.
1301 	 */
1302 	if (taskq_dispatch(zilog->zl_clean_taskq,
1303 	    (void (*)(void *))zil_itxg_clean, clean_me, TQ_NOSLEEP) == NULL)
1304 		zil_itxg_clean(clean_me);
1305 }
1306 
1307 /*
1308  * Get the list of itxs to commit into zl_itx_commit_list.
1309  */
1310 static void
1311 zil_get_commit_list(zilog_t *zilog)
1312 {
1313 	uint64_t otxg, txg;
1314 	list_t *commit_list = &zilog->zl_itx_commit_list;
1315 	uint64_t push_sod = 0;
1316 
1317 	if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */
1318 		otxg = ZILTEST_TXG;
1319 	else
1320 		otxg = spa_last_synced_txg(zilog->zl_spa) + 1;
1321 
1322 	for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) {
1323 		itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK];
1324 
1325 		mutex_enter(&itxg->itxg_lock);
1326 		if (itxg->itxg_txg != txg) {
1327 			mutex_exit(&itxg->itxg_lock);
1328 			continue;
1329 		}
1330 
1331 		list_move_tail(commit_list, &itxg->itxg_itxs->i_sync_list);
1332 		push_sod += itxg->itxg_sod;
1333 		itxg->itxg_sod = 0;
1334 
1335 		mutex_exit(&itxg->itxg_lock);
1336 	}
1337 	atomic_add_64(&zilog->zl_itx_list_sz, -push_sod);
1338 }
1339 
1340 /*
1341  * Move the async itxs for a specified object to commit into sync lists.
1342  */
1343 static void
1344 zil_async_to_sync(zilog_t *zilog, uint64_t foid)
1345 {
1346 	uint64_t otxg, txg;
1347 	itx_async_node_t *ian;
1348 	avl_tree_t *t;
1349 	avl_index_t where;
1350 
1351 	if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */
1352 		otxg = ZILTEST_TXG;
1353 	else
1354 		otxg = spa_last_synced_txg(zilog->zl_spa) + 1;
1355 
1356 	for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) {
1357 		itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK];
1358 
1359 		mutex_enter(&itxg->itxg_lock);
1360 		if (itxg->itxg_txg != txg) {
1361 			mutex_exit(&itxg->itxg_lock);
1362 			continue;
1363 		}
1364 
1365 		/*
1366 		 * If a foid is specified then find that node and append its
1367 		 * list. Otherwise walk the tree appending all the lists
1368 		 * to the sync list. We add to the end rather than the
1369 		 * beginning to ensure the create has happened.
1370 		 */
1371 		t = &itxg->itxg_itxs->i_async_tree;
1372 		if (foid != 0) {
1373 			ian = avl_find(t, &foid, &where);
1374 			if (ian != NULL) {
1375 				list_move_tail(&itxg->itxg_itxs->i_sync_list,
1376 				    &ian->ia_list);
1377 			}
1378 		} else {
1379 			void *cookie = NULL;
1380 
1381 			while ((ian = avl_destroy_nodes(t, &cookie)) != NULL) {
1382 				list_move_tail(&itxg->itxg_itxs->i_sync_list,
1383 				    &ian->ia_list);
1384 				list_destroy(&ian->ia_list);
1385 				kmem_free(ian, sizeof (itx_async_node_t));
1386 			}
1387 		}
1388 		mutex_exit(&itxg->itxg_lock);
1389 	}
1390 }
1391 
1392 static void
1393 zil_commit_writer(zilog_t *zilog)
1394 {
1395 	uint64_t txg;
1396 	itx_t *itx;
1397 	lwb_t *lwb;
1398 	spa_t *spa = zilog->zl_spa;
1399 	int error = 0;
1400 
1401 	ASSERT(zilog->zl_root_zio == NULL);
1402 
1403 	mutex_exit(&zilog->zl_lock);
1404 
1405 	zil_get_commit_list(zilog);
1406 
1407 	/*
1408 	 * Return if there's nothing to commit before we dirty the fs by
1409 	 * calling zil_create().
1410 	 */
1411 	if (list_head(&zilog->zl_itx_commit_list) == NULL) {
1412 		mutex_enter(&zilog->zl_lock);
1413 		return;
1414 	}
1415 
1416 	if (zilog->zl_suspend) {
1417 		lwb = NULL;
1418 	} else {
1419 		lwb = list_tail(&zilog->zl_lwb_list);
1420 		if (lwb == NULL)
1421 			lwb = zil_create(zilog);
1422 	}
1423 
1424 	DTRACE_PROBE1(zil__cw1, zilog_t *, zilog);
1425 	while (itx = list_head(&zilog->zl_itx_commit_list)) {
1426 		txg = itx->itx_lr.lrc_txg;
1427 		ASSERT(txg);
1428 
1429 		if (txg > spa_last_synced_txg(spa) || txg > spa_freeze_txg(spa))
1430 			lwb = zil_lwb_commit(zilog, itx, lwb);
1431 		list_remove(&zilog->zl_itx_commit_list, itx);
1432 		kmem_free(itx, offsetof(itx_t, itx_lr)
1433 		    + itx->itx_lr.lrc_reclen);
1434 	}
1435 	DTRACE_PROBE1(zil__cw2, zilog_t *, zilog);
1436 
1437 	/* write the last block out */
1438 	if (lwb != NULL && lwb->lwb_zio != NULL)
1439 		lwb = zil_lwb_write_start(zilog, lwb);
1440 
1441 	zilog->zl_cur_used = 0;
1442 
1443 	/*
1444 	 * Wait if necessary for the log blocks to be on stable storage.
1445 	 */
1446 	if (zilog->zl_root_zio) {
1447 		error = zio_wait(zilog->zl_root_zio);
1448 		zilog->zl_root_zio = NULL;
1449 		zil_flush_vdevs(zilog);
1450 	}
1451 
1452 	if (error || lwb == NULL)
1453 		txg_wait_synced(zilog->zl_dmu_pool, 0);
1454 
1455 	mutex_enter(&zilog->zl_lock);
1456 
1457 	/*
1458 	 * Remember the highest committed log sequence number for ztest.
1459 	 * We only update this value when all the log writes succeeded,
1460 	 * because ztest wants to ASSERT that it got the whole log chain.
1461 	 */
1462 	if (error == 0 && lwb != NULL)
1463 		zilog->zl_commit_lr_seq = zilog->zl_lr_seq;
1464 }
1465 
1466 /*
1467  * Commit zfs transactions to stable storage.
1468  * If foid is 0 push out all transactions, otherwise push only those
1469  * for that object or might reference that object.
1470  *
1471  * itxs are committed in batches. In a heavily stressed zil there will be
1472  * a commit writer thread who is writing out a bunch of itxs to the log
1473  * for a set of committing threads (cthreads) in the same batch as the writer.
1474  * Those cthreads are all waiting on the same cv for that batch.
1475  *
1476  * There will also be a different and growing batch of threads that are
1477  * waiting to commit (qthreads). When the committing batch completes
1478  * a transition occurs such that the cthreads exit and the qthreads become
1479  * cthreads. One of the new cthreads becomes the writer thread for the
1480  * batch. Any new threads arriving become new qthreads.
1481  *
1482  * Only 2 condition variables are needed and there's no transition
1483  * between the two cvs needed. They just flip-flop between qthreads
1484  * and cthreads.
1485  *
1486  * Using this scheme we can efficiently wakeup up only those threads
1487  * that have been committed.
1488  */
1489 void
1490 zil_commit(zilog_t *zilog, uint64_t foid)
1491 {
1492 	uint64_t mybatch;
1493 
1494 	if (zilog->zl_sync == ZFS_SYNC_DISABLED)
1495 		return;
1496 
1497 	/* move the async itxs for the foid to the sync queues */
1498 	zil_async_to_sync(zilog, foid);
1499 
1500 	mutex_enter(&zilog->zl_lock);
1501 	mybatch = zilog->zl_next_batch;
1502 	while (zilog->zl_writer) {
1503 		cv_wait(&zilog->zl_cv_batch[mybatch & 1], &zilog->zl_lock);
1504 		if (mybatch <= zilog->zl_com_batch) {
1505 			mutex_exit(&zilog->zl_lock);
1506 			return;
1507 		}
1508 	}
1509 
1510 	zilog->zl_next_batch++;
1511 	zilog->zl_writer = B_TRUE;
1512 	zil_commit_writer(zilog);
1513 	zilog->zl_com_batch = mybatch;
1514 	zilog->zl_writer = B_FALSE;
1515 	mutex_exit(&zilog->zl_lock);
1516 
1517 	/* wake up one thread to become the next writer */
1518 	cv_signal(&zilog->zl_cv_batch[(mybatch+1) & 1]);
1519 
1520 	/* wake up all threads waiting for this batch to be committed */
1521 	cv_broadcast(&zilog->zl_cv_batch[mybatch & 1]);
1522 }
1523 
1524 /*
1525  * Called in syncing context to free committed log blocks and update log header.
1526  */
1527 void
1528 zil_sync(zilog_t *zilog, dmu_tx_t *tx)
1529 {
1530 	zil_header_t *zh = zil_header_in_syncing_context(zilog);
1531 	uint64_t txg = dmu_tx_get_txg(tx);
1532 	spa_t *spa = zilog->zl_spa;
1533 	uint64_t *replayed_seq = &zilog->zl_replayed_seq[txg & TXG_MASK];
1534 	lwb_t *lwb;
1535 
1536 	/*
1537 	 * We don't zero out zl_destroy_txg, so make sure we don't try
1538 	 * to destroy it twice.
1539 	 */
1540 	if (spa_sync_pass(spa) != 1)
1541 		return;
1542 
1543 	mutex_enter(&zilog->zl_lock);
1544 
1545 	ASSERT(zilog->zl_stop_sync == 0);
1546 
1547 	if (*replayed_seq != 0) {
1548 		ASSERT(zh->zh_replay_seq < *replayed_seq);
1549 		zh->zh_replay_seq = *replayed_seq;
1550 		*replayed_seq = 0;
1551 	}
1552 
1553 	if (zilog->zl_destroy_txg == txg) {
1554 		blkptr_t blk = zh->zh_log;
1555 
1556 		ASSERT(list_head(&zilog->zl_lwb_list) == NULL);
1557 
1558 		bzero(zh, sizeof (zil_header_t));
1559 		bzero(zilog->zl_replayed_seq, sizeof (zilog->zl_replayed_seq));
1560 
1561 		if (zilog->zl_keep_first) {
1562 			/*
1563 			 * If this block was part of log chain that couldn't
1564 			 * be claimed because a device was missing during
1565 			 * zil_claim(), but that device later returns,
1566 			 * then this block could erroneously appear valid.
1567 			 * To guard against this, assign a new GUID to the new
1568 			 * log chain so it doesn't matter what blk points to.
1569 			 */
1570 			zil_init_log_chain(zilog, &blk);
1571 			zh->zh_log = blk;
1572 		}
1573 	}
1574 
1575 	while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) {
1576 		zh->zh_log = lwb->lwb_blk;
1577 		if (lwb->lwb_buf != NULL || lwb->lwb_max_txg > txg)
1578 			break;
1579 		list_remove(&zilog->zl_lwb_list, lwb);
1580 		zio_free_zil(spa, txg, &lwb->lwb_blk);
1581 		kmem_cache_free(zil_lwb_cache, lwb);
1582 
1583 		/*
1584 		 * If we don't have anything left in the lwb list then
1585 		 * we've had an allocation failure and we need to zero
1586 		 * out the zil_header blkptr so that we don't end
1587 		 * up freeing the same block twice.
1588 		 */
1589 		if (list_head(&zilog->zl_lwb_list) == NULL)
1590 			BP_ZERO(&zh->zh_log);
1591 	}
1592 	mutex_exit(&zilog->zl_lock);
1593 }
1594 
1595 void
1596 zil_init(void)
1597 {
1598 	zil_lwb_cache = kmem_cache_create("zil_lwb_cache",
1599 	    sizeof (struct lwb), 0, NULL, NULL, NULL, NULL, NULL, 0);
1600 }
1601 
1602 void
1603 zil_fini(void)
1604 {
1605 	kmem_cache_destroy(zil_lwb_cache);
1606 }
1607 
1608 void
1609 zil_set_sync(zilog_t *zilog, uint64_t sync)
1610 {
1611 	zilog->zl_sync = sync;
1612 }
1613 
1614 void
1615 zil_set_logbias(zilog_t *zilog, uint64_t logbias)
1616 {
1617 	zilog->zl_logbias = logbias;
1618 }
1619 
1620 zilog_t *
1621 zil_alloc(objset_t *os, zil_header_t *zh_phys)
1622 {
1623 	zilog_t *zilog;
1624 
1625 	zilog = kmem_zalloc(sizeof (zilog_t), KM_SLEEP);
1626 
1627 	zilog->zl_header = zh_phys;
1628 	zilog->zl_os = os;
1629 	zilog->zl_spa = dmu_objset_spa(os);
1630 	zilog->zl_dmu_pool = dmu_objset_pool(os);
1631 	zilog->zl_destroy_txg = TXG_INITIAL - 1;
1632 	zilog->zl_logbias = dmu_objset_logbias(os);
1633 	zilog->zl_sync = dmu_objset_syncprop(os);
1634 	zilog->zl_next_batch = 1;
1635 
1636 	mutex_init(&zilog->zl_lock, NULL, MUTEX_DEFAULT, NULL);
1637 
1638 	for (int i = 0; i < TXG_SIZE; i++) {
1639 		mutex_init(&zilog->zl_itxg[i].itxg_lock, NULL,
1640 		    MUTEX_DEFAULT, NULL);
1641 	}
1642 
1643 	list_create(&zilog->zl_lwb_list, sizeof (lwb_t),
1644 	    offsetof(lwb_t, lwb_node));
1645 
1646 	list_create(&zilog->zl_itx_commit_list, sizeof (itx_t),
1647 	    offsetof(itx_t, itx_node));
1648 
1649 	mutex_init(&zilog->zl_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
1650 
1651 	avl_create(&zilog->zl_vdev_tree, zil_vdev_compare,
1652 	    sizeof (zil_vdev_node_t), offsetof(zil_vdev_node_t, zv_node));
1653 
1654 	cv_init(&zilog->zl_cv_writer, NULL, CV_DEFAULT, NULL);
1655 	cv_init(&zilog->zl_cv_suspend, NULL, CV_DEFAULT, NULL);
1656 	cv_init(&zilog->zl_cv_batch[0], NULL, CV_DEFAULT, NULL);
1657 	cv_init(&zilog->zl_cv_batch[1], NULL, CV_DEFAULT, NULL);
1658 
1659 	return (zilog);
1660 }
1661 
1662 void
1663 zil_free(zilog_t *zilog)
1664 {
1665 	zilog->zl_stop_sync = 1;
1666 
1667 	ASSERT(list_is_empty(&zilog->zl_lwb_list));
1668 	list_destroy(&zilog->zl_lwb_list);
1669 
1670 	avl_destroy(&zilog->zl_vdev_tree);
1671 	mutex_destroy(&zilog->zl_vdev_lock);
1672 
1673 	ASSERT(list_is_empty(&zilog->zl_itx_commit_list));
1674 	list_destroy(&zilog->zl_itx_commit_list);
1675 
1676 	for (int i = 0; i < TXG_SIZE; i++) {
1677 		/*
1678 		 * It's possible for an itx to be generated that doesn't dirty
1679 		 * a txg (e.g. ztest TX_TRUNCATE). So there's no zil_clean()
1680 		 * callback to remove the entry. We remove those here.
1681 		 *
1682 		 * Also free up the ziltest itxs.
1683 		 */
1684 		if (zilog->zl_itxg[i].itxg_itxs)
1685 			zil_itxg_clean(zilog->zl_itxg[i].itxg_itxs);
1686 		mutex_destroy(&zilog->zl_itxg[i].itxg_lock);
1687 	}
1688 
1689 	mutex_destroy(&zilog->zl_lock);
1690 
1691 	cv_destroy(&zilog->zl_cv_writer);
1692 	cv_destroy(&zilog->zl_cv_suspend);
1693 	cv_destroy(&zilog->zl_cv_batch[0]);
1694 	cv_destroy(&zilog->zl_cv_batch[1]);
1695 
1696 	kmem_free(zilog, sizeof (zilog_t));
1697 }
1698 
1699 /*
1700  * Open an intent log.
1701  */
1702 zilog_t *
1703 zil_open(objset_t *os, zil_get_data_t *get_data)
1704 {
1705 	zilog_t *zilog = dmu_objset_zil(os);
1706 
1707 	ASSERT(zilog->zl_clean_taskq == NULL);
1708 	ASSERT(zilog->zl_get_data == NULL);
1709 	ASSERT(list_is_empty(&zilog->zl_lwb_list));
1710 
1711 	zilog->zl_get_data = get_data;
1712 	zilog->zl_clean_taskq = taskq_create("zil_clean", 1, minclsyspri,
1713 	    2, 2, TASKQ_PREPOPULATE);
1714 
1715 	return (zilog);
1716 }
1717 
1718 /*
1719  * Close an intent log.
1720  */
1721 void
1722 zil_close(zilog_t *zilog)
1723 {
1724 	lwb_t *lwb;
1725 	uint64_t txg = 0;
1726 
1727 	zil_commit(zilog, 0); /* commit all itx */
1728 
1729 	/*
1730 	 * The lwb_max_txg for the stubby lwb will reflect the last activity
1731 	 * for the zil.  After a txg_wait_synced() on the txg we know all the
1732 	 * callbacks have occurred that may clean the zil.  Only then can we
1733 	 * destroy the zl_clean_taskq.
1734 	 */
1735 	mutex_enter(&zilog->zl_lock);
1736 	lwb = list_tail(&zilog->zl_lwb_list);
1737 	if (lwb != NULL)
1738 		txg = lwb->lwb_max_txg;
1739 	mutex_exit(&zilog->zl_lock);
1740 	if (txg)
1741 		txg_wait_synced(zilog->zl_dmu_pool, txg);
1742 
1743 	taskq_destroy(zilog->zl_clean_taskq);
1744 	zilog->zl_clean_taskq = NULL;
1745 	zilog->zl_get_data = NULL;
1746 
1747 	/*
1748 	 * We should have only one LWB left on the list; remove it now.
1749 	 */
1750 	mutex_enter(&zilog->zl_lock);
1751 	lwb = list_head(&zilog->zl_lwb_list);
1752 	if (lwb != NULL) {
1753 		ASSERT(lwb == list_tail(&zilog->zl_lwb_list));
1754 		list_remove(&zilog->zl_lwb_list, lwb);
1755 		zio_buf_free(lwb->lwb_buf, lwb->lwb_sz);
1756 		kmem_cache_free(zil_lwb_cache, lwb);
1757 	}
1758 	mutex_exit(&zilog->zl_lock);
1759 }
1760 
1761 /*
1762  * Suspend an intent log.  While in suspended mode, we still honor
1763  * synchronous semantics, but we rely on txg_wait_synced() to do it.
1764  * We suspend the log briefly when taking a snapshot so that the snapshot
1765  * contains all the data it's supposed to, and has an empty intent log.
1766  */
1767 int
1768 zil_suspend(zilog_t *zilog)
1769 {
1770 	const zil_header_t *zh = zilog->zl_header;
1771 
1772 	mutex_enter(&zilog->zl_lock);
1773 	if (zh->zh_flags & ZIL_REPLAY_NEEDED) {		/* unplayed log */
1774 		mutex_exit(&zilog->zl_lock);
1775 		return (EBUSY);
1776 	}
1777 	if (zilog->zl_suspend++ != 0) {
1778 		/*
1779 		 * Someone else already began a suspend.
1780 		 * Just wait for them to finish.
1781 		 */
1782 		while (zilog->zl_suspending)
1783 			cv_wait(&zilog->zl_cv_suspend, &zilog->zl_lock);
1784 		mutex_exit(&zilog->zl_lock);
1785 		return (0);
1786 	}
1787 	zilog->zl_suspending = B_TRUE;
1788 	mutex_exit(&zilog->zl_lock);
1789 
1790 	zil_commit(zilog, 0);
1791 
1792 	zil_destroy(zilog, B_FALSE);
1793 
1794 	mutex_enter(&zilog->zl_lock);
1795 	zilog->zl_suspending = B_FALSE;
1796 	cv_broadcast(&zilog->zl_cv_suspend);
1797 	mutex_exit(&zilog->zl_lock);
1798 
1799 	return (0);
1800 }
1801 
1802 void
1803 zil_resume(zilog_t *zilog)
1804 {
1805 	mutex_enter(&zilog->zl_lock);
1806 	ASSERT(zilog->zl_suspend != 0);
1807 	zilog->zl_suspend--;
1808 	mutex_exit(&zilog->zl_lock);
1809 }
1810 
1811 typedef struct zil_replay_arg {
1812 	zil_replay_func_t **zr_replay;
1813 	void		*zr_arg;
1814 	boolean_t	zr_byteswap;
1815 	char		*zr_lr;
1816 } zil_replay_arg_t;
1817 
1818 static int
1819 zil_replay_error(zilog_t *zilog, lr_t *lr, int error)
1820 {
1821 	char name[MAXNAMELEN];
1822 
1823 	zilog->zl_replaying_seq--;	/* didn't actually replay this one */
1824 
1825 	dmu_objset_name(zilog->zl_os, name);
1826 
1827 	cmn_err(CE_WARN, "ZFS replay transaction error %d, "
1828 	    "dataset %s, seq 0x%llx, txtype %llu %s\n", error, name,
1829 	    (u_longlong_t)lr->lrc_seq,
1830 	    (u_longlong_t)(lr->lrc_txtype & ~TX_CI),
1831 	    (lr->lrc_txtype & TX_CI) ? "CI" : "");
1832 
1833 	return (error);
1834 }
1835 
1836 static int
1837 zil_replay_log_record(zilog_t *zilog, lr_t *lr, void *zra, uint64_t claim_txg)
1838 {
1839 	zil_replay_arg_t *zr = zra;
1840 	const zil_header_t *zh = zilog->zl_header;
1841 	uint64_t reclen = lr->lrc_reclen;
1842 	uint64_t txtype = lr->lrc_txtype;
1843 	int error = 0;
1844 
1845 	zilog->zl_replaying_seq = lr->lrc_seq;
1846 
1847 	if (lr->lrc_seq <= zh->zh_replay_seq)	/* already replayed */
1848 		return (0);
1849 
1850 	if (lr->lrc_txg < claim_txg)		/* already committed */
1851 		return (0);
1852 
1853 	/* Strip case-insensitive bit, still present in log record */
1854 	txtype &= ~TX_CI;
1855 
1856 	if (txtype == 0 || txtype >= TX_MAX_TYPE)
1857 		return (zil_replay_error(zilog, lr, EINVAL));
1858 
1859 	/*
1860 	 * If this record type can be logged out of order, the object
1861 	 * (lr_foid) may no longer exist.  That's legitimate, not an error.
1862 	 */
1863 	if (TX_OOO(txtype)) {
1864 		error = dmu_object_info(zilog->zl_os,
1865 		    ((lr_ooo_t *)lr)->lr_foid, NULL);
1866 		if (error == ENOENT || error == EEXIST)
1867 			return (0);
1868 	}
1869 
1870 	/*
1871 	 * Make a copy of the data so we can revise and extend it.
1872 	 */
1873 	bcopy(lr, zr->zr_lr, reclen);
1874 
1875 	/*
1876 	 * If this is a TX_WRITE with a blkptr, suck in the data.
1877 	 */
1878 	if (txtype == TX_WRITE && reclen == sizeof (lr_write_t)) {
1879 		error = zil_read_log_data(zilog, (lr_write_t *)lr,
1880 		    zr->zr_lr + reclen);
1881 		if (error)
1882 			return (zil_replay_error(zilog, lr, error));
1883 	}
1884 
1885 	/*
1886 	 * The log block containing this lr may have been byteswapped
1887 	 * so that we can easily examine common fields like lrc_txtype.
1888 	 * However, the log is a mix of different record types, and only the
1889 	 * replay vectors know how to byteswap their records.  Therefore, if
1890 	 * the lr was byteswapped, undo it before invoking the replay vector.
1891 	 */
1892 	if (zr->zr_byteswap)
1893 		byteswap_uint64_array(zr->zr_lr, reclen);
1894 
1895 	/*
1896 	 * We must now do two things atomically: replay this log record,
1897 	 * and update the log header sequence number to reflect the fact that
1898 	 * we did so. At the end of each replay function the sequence number
1899 	 * is updated if we are in replay mode.
1900 	 */
1901 	error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lr, zr->zr_byteswap);
1902 	if (error) {
1903 		/*
1904 		 * The DMU's dnode layer doesn't see removes until the txg
1905 		 * commits, so a subsequent claim can spuriously fail with
1906 		 * EEXIST. So if we receive any error we try syncing out
1907 		 * any removes then retry the transaction.  Note that we
1908 		 * specify B_FALSE for byteswap now, so we don't do it twice.
1909 		 */
1910 		txg_wait_synced(spa_get_dsl(zilog->zl_spa), 0);
1911 		error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lr, B_FALSE);
1912 		if (error)
1913 			return (zil_replay_error(zilog, lr, error));
1914 	}
1915 	return (0);
1916 }
1917 
1918 /* ARGSUSED */
1919 static int
1920 zil_incr_blks(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg)
1921 {
1922 	zilog->zl_replay_blks++;
1923 
1924 	return (0);
1925 }
1926 
1927 /*
1928  * If this dataset has a non-empty intent log, replay it and destroy it.
1929  */
1930 void
1931 zil_replay(objset_t *os, void *arg, zil_replay_func_t *replay_func[TX_MAX_TYPE])
1932 {
1933 	zilog_t *zilog = dmu_objset_zil(os);
1934 	const zil_header_t *zh = zilog->zl_header;
1935 	zil_replay_arg_t zr;
1936 
1937 	if ((zh->zh_flags & ZIL_REPLAY_NEEDED) == 0) {
1938 		zil_destroy(zilog, B_TRUE);
1939 		return;
1940 	}
1941 
1942 	zr.zr_replay = replay_func;
1943 	zr.zr_arg = arg;
1944 	zr.zr_byteswap = BP_SHOULD_BYTESWAP(&zh->zh_log);
1945 	zr.zr_lr = kmem_alloc(2 * SPA_MAXBLOCKSIZE, KM_SLEEP);
1946 
1947 	/*
1948 	 * Wait for in-progress removes to sync before starting replay.
1949 	 */
1950 	txg_wait_synced(zilog->zl_dmu_pool, 0);
1951 
1952 	zilog->zl_replay = B_TRUE;
1953 	zilog->zl_replay_time = ddi_get_lbolt();
1954 	ASSERT(zilog->zl_replay_blks == 0);
1955 	(void) zil_parse(zilog, zil_incr_blks, zil_replay_log_record, &zr,
1956 	    zh->zh_claim_txg);
1957 	kmem_free(zr.zr_lr, 2 * SPA_MAXBLOCKSIZE);
1958 
1959 	zil_destroy(zilog, B_FALSE);
1960 	txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg);
1961 	zilog->zl_replay = B_FALSE;
1962 }
1963 
1964 boolean_t
1965 zil_replaying(zilog_t *zilog, dmu_tx_t *tx)
1966 {
1967 	if (zilog->zl_sync == ZFS_SYNC_DISABLED)
1968 		return (B_TRUE);
1969 
1970 	if (zilog->zl_replay) {
1971 		dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
1972 		zilog->zl_replayed_seq[dmu_tx_get_txg(tx) & TXG_MASK] =
1973 		    zilog->zl_replaying_seq;
1974 		return (B_TRUE);
1975 	}
1976 
1977 	return (B_FALSE);
1978 }
1979 
1980 /* ARGSUSED */
1981 int
1982 zil_vdev_offline(const char *osname, void *arg)
1983 {
1984 	objset_t *os;
1985 	zilog_t *zilog;
1986 	int error;
1987 
1988 	error = dmu_objset_hold(osname, FTAG, &os);
1989 	if (error)
1990 		return (error);
1991 
1992 	zilog = dmu_objset_zil(os);
1993 	if (zil_suspend(zilog) != 0)
1994 		error = EEXIST;
1995 	else
1996 		zil_resume(zilog);
1997 	dmu_objset_rele(os, FTAG);
1998 	return (error);
1999 }
2000