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 }