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) 2013 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 * Disable intent logging replay. This global ZIL switch affects all pools.
70 */
71 int zil_replay_disable = 0;
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;
141 zil_bp_node_t *zn;
142 avl_index_t where;
143
144 if (BP_IS_EMBEDDED(bp))
145 return (0);
146
147 dva = BP_IDENTITY(bp);
148
149 if (avl_find(t, dva, &where) != NULL)
150 return (SET_ERROR(EEXIST));
151
152 zn = kmem_alloc(sizeof (zil_bp_node_t), KM_SLEEP);
153 zn->zn_dva = *dva;
154 avl_insert(t, zn, where);
155
156 return (0);
157 }
158
159 static zil_header_t *
160 zil_header_in_syncing_context(zilog_t *zilog)
161 {
162 return ((zil_header_t *)zilog->zl_header);
163 }
164
165 static void
166 zil_init_log_chain(zilog_t *zilog, blkptr_t *bp)
167 {
168 zio_cksum_t *zc = &bp->blk_cksum;
169
170 zc->zc_word[ZIL_ZC_GUID_0] = spa_get_random(-1ULL);
171 zc->zc_word[ZIL_ZC_GUID_1] = spa_get_random(-1ULL);
172 zc->zc_word[ZIL_ZC_OBJSET] = dmu_objset_id(zilog->zl_os);
173 zc->zc_word[ZIL_ZC_SEQ] = 1ULL;
174 }
175
176 /*
177 * Read a log block and make sure it's valid.
178 */
179 static int
180 zil_read_log_block(zilog_t *zilog, const blkptr_t *bp, blkptr_t *nbp, void *dst,
181 char **end)
182 {
183 enum zio_flag zio_flags = ZIO_FLAG_CANFAIL;
184 uint32_t aflags = ARC_WAIT;
185 arc_buf_t *abuf = NULL;
186 zbookmark_t zb;
187 int error;
188
189 if (zilog->zl_header->zh_claim_txg == 0)
190 zio_flags |= ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB;
191
192 if (!(zilog->zl_header->zh_flags & ZIL_CLAIM_LR_SEQ_VALID))
193 zio_flags |= ZIO_FLAG_SPECULATIVE;
194
195 SET_BOOKMARK(&zb, bp->blk_cksum.zc_word[ZIL_ZC_OBJSET],
196 ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, bp->blk_cksum.zc_word[ZIL_ZC_SEQ]);
197
198 error = arc_read(NULL, zilog->zl_spa, bp, arc_getbuf_func, &abuf,
199 ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb);
200
201 if (error == 0) {
202 zio_cksum_t cksum = bp->blk_cksum;
203
204 /*
205 * Validate the checksummed log block.
206 *
207 * Sequence numbers should be... sequential. The checksum
208 * verifier for the next block should be bp's checksum plus 1.
209 *
210 * Also check the log chain linkage and size used.
211 */
212 cksum.zc_word[ZIL_ZC_SEQ]++;
213
214 if (BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_ZILOG2) {
215 zil_chain_t *zilc = abuf->b_data;
216 char *lr = (char *)(zilc + 1);
217 uint64_t len = zilc->zc_nused - sizeof (zil_chain_t);
218
219 if (bcmp(&cksum, &zilc->zc_next_blk.blk_cksum,
220 sizeof (cksum)) || BP_IS_HOLE(&zilc->zc_next_blk)) {
221 error = SET_ERROR(ECKSUM);
222 } else {
223 bcopy(lr, dst, len);
224 *end = (char *)dst + len;
225 *nbp = zilc->zc_next_blk;
226 }
227 } else {
228 char *lr = abuf->b_data;
229 uint64_t size = BP_GET_LSIZE(bp);
230 zil_chain_t *zilc = (zil_chain_t *)(lr + size) - 1;
231
232 if (bcmp(&cksum, &zilc->zc_next_blk.blk_cksum,
233 sizeof (cksum)) || BP_IS_HOLE(&zilc->zc_next_blk) ||
234 (zilc->zc_nused > (size - sizeof (*zilc)))) {
235 error = SET_ERROR(ECKSUM);
236 } else {
237 bcopy(lr, dst, zilc->zc_nused);
238 *end = (char *)dst + zilc->zc_nused;
239 *nbp = zilc->zc_next_blk;
240 }
241 }
242
243 VERIFY(arc_buf_remove_ref(abuf, &abuf));
244 }
245
246 return (error);
247 }
248
249 /*
250 * Read a TX_WRITE log data block.
251 */
252 static int
253 zil_read_log_data(zilog_t *zilog, const lr_write_t *lr, void *wbuf)
254 {
255 enum zio_flag zio_flags = ZIO_FLAG_CANFAIL;
256 const blkptr_t *bp = &lr->lr_blkptr;
257 uint32_t aflags = ARC_WAIT;
258 arc_buf_t *abuf = NULL;
259 zbookmark_t zb;
260 int error;
261
262 if (BP_IS_HOLE(bp)) {
263 if (wbuf != NULL)
264 bzero(wbuf, MAX(BP_GET_LSIZE(bp), lr->lr_length));
265 return (0);
266 }
267
268 if (zilog->zl_header->zh_claim_txg == 0)
269 zio_flags |= ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB;
270
271 SET_BOOKMARK(&zb, dmu_objset_id(zilog->zl_os), lr->lr_foid,
272 ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp));
273
274 error = arc_read(NULL, zilog->zl_spa, bp, arc_getbuf_func, &abuf,
275 ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb);
276
277 if (error == 0) {
278 if (wbuf != NULL)
279 bcopy(abuf->b_data, wbuf, arc_buf_size(abuf));
280 (void) arc_buf_remove_ref(abuf, &abuf);
281 }
282
283 return (error);
284 }
285
286 /*
287 * Parse the intent log, and call parse_func for each valid record within.
288 */
289 int
290 zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func,
291 zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg)
292 {
293 const zil_header_t *zh = zilog->zl_header;
294 boolean_t claimed = !!zh->zh_claim_txg;
295 uint64_t claim_blk_seq = claimed ? zh->zh_claim_blk_seq : UINT64_MAX;
296 uint64_t claim_lr_seq = claimed ? zh->zh_claim_lr_seq : UINT64_MAX;
297 uint64_t max_blk_seq = 0;
298 uint64_t max_lr_seq = 0;
299 uint64_t blk_count = 0;
300 uint64_t lr_count = 0;
301 blkptr_t blk, next_blk;
302 char *lrbuf, *lrp;
303 int error = 0;
304
305 /*
306 * Old logs didn't record the maximum zh_claim_lr_seq.
307 */
308 if (!(zh->zh_flags & ZIL_CLAIM_LR_SEQ_VALID))
309 claim_lr_seq = UINT64_MAX;
310
311 /*
312 * Starting at the block pointed to by zh_log we read the log chain.
313 * For each block in the chain we strongly check that block to
314 * ensure its validity. We stop when an invalid block is found.
315 * For each block pointer in the chain we call parse_blk_func().
316 * For each record in each valid block we call parse_lr_func().
317 * If the log has been claimed, stop if we encounter a sequence
318 * number greater than the highest claimed sequence number.
319 */
320 lrbuf = zio_buf_alloc(SPA_MAXBLOCKSIZE);
321 zil_bp_tree_init(zilog);
322
323 for (blk = zh->zh_log; !BP_IS_HOLE(&blk); blk = next_blk) {
324 uint64_t blk_seq = blk.blk_cksum.zc_word[ZIL_ZC_SEQ];
325 int reclen;
326 char *end;
327
328 if (blk_seq > claim_blk_seq)
329 break;
330 if ((error = parse_blk_func(zilog, &blk, arg, txg)) != 0)
331 break;
332 ASSERT3U(max_blk_seq, <, blk_seq);
333 max_blk_seq = blk_seq;
334 blk_count++;
335
336 if (max_lr_seq == claim_lr_seq && max_blk_seq == claim_blk_seq)
337 break;
338
339 error = zil_read_log_block(zilog, &blk, &next_blk, lrbuf, &end);
340 if (error != 0)
341 break;
342
343 for (lrp = lrbuf; lrp < end; lrp += reclen) {
344 lr_t *lr = (lr_t *)lrp;
345 reclen = lr->lrc_reclen;
346 ASSERT3U(reclen, >=, sizeof (lr_t));
347 if (lr->lrc_seq > claim_lr_seq)
348 goto done;
349 if ((error = parse_lr_func(zilog, lr, arg, txg)) != 0)
350 goto done;
351 ASSERT3U(max_lr_seq, <, lr->lrc_seq);
352 max_lr_seq = lr->lrc_seq;
353 lr_count++;
354 }
355 }
356 done:
357 zilog->zl_parse_error = error;
358 zilog->zl_parse_blk_seq = max_blk_seq;
359 zilog->zl_parse_lr_seq = max_lr_seq;
360 zilog->zl_parse_blk_count = blk_count;
361 zilog->zl_parse_lr_count = lr_count;
362
363 ASSERT(!claimed || !(zh->zh_flags & ZIL_CLAIM_LR_SEQ_VALID) ||
364 (max_blk_seq == claim_blk_seq && max_lr_seq == claim_lr_seq));
365
366 zil_bp_tree_fini(zilog);
367 zio_buf_free(lrbuf, SPA_MAXBLOCKSIZE);
368
369 return (error);
370 }
371
372 static int
373 zil_claim_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t first_txg)
374 {
375 /*
376 * Claim log block if not already committed and not already claimed.
377 * If tx == NULL, just verify that the block is claimable.
378 */
379 if (BP_IS_HOLE(bp) || bp->blk_birth < first_txg ||
380 zil_bp_tree_add(zilog, bp) != 0)
381 return (0);
382
383 return (zio_wait(zio_claim(NULL, zilog->zl_spa,
384 tx == NULL ? 0 : first_txg, bp, spa_claim_notify, NULL,
385 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB)));
386 }
387
388 static int
389 zil_claim_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t first_txg)
390 {
391 lr_write_t *lr = (lr_write_t *)lrc;
392 int error;
393
394 if (lrc->lrc_txtype != TX_WRITE)
395 return (0);
396
397 /*
398 * If the block is not readable, don't claim it. This can happen
399 * in normal operation when a log block is written to disk before
400 * some of the dmu_sync() blocks it points to. In this case, the
401 * transaction cannot have been committed to anyone (we would have
402 * waited for all writes to be stable first), so it is semantically
403 * correct to declare this the end of the log.
404 */
405 if (lr->lr_blkptr.blk_birth >= first_txg &&
406 (error = zil_read_log_data(zilog, lr, NULL)) != 0)
407 return (error);
408 return (zil_claim_log_block(zilog, &lr->lr_blkptr, tx, first_txg));
409 }
410
411 /* ARGSUSED */
412 static int
413 zil_free_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t claim_txg)
414 {
415 zio_free_zil(zilog->zl_spa, dmu_tx_get_txg(tx), bp);
416
417 return (0);
418 }
419
420 static int
421 zil_free_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t claim_txg)
422 {
423 lr_write_t *lr = (lr_write_t *)lrc;
424 blkptr_t *bp = &lr->lr_blkptr;
425
426 /*
427 * If we previously claimed it, we need to free it.
428 */
429 if (claim_txg != 0 && lrc->lrc_txtype == TX_WRITE &&
430 bp->blk_birth >= claim_txg && zil_bp_tree_add(zilog, bp) == 0 &&
431 !BP_IS_HOLE(bp))
432 zio_free(zilog->zl_spa, dmu_tx_get_txg(tx), bp);
433
434 return (0);
435 }
436
437 static lwb_t *
438 zil_alloc_lwb(zilog_t *zilog, blkptr_t *bp, uint64_t txg)
439 {
440 lwb_t *lwb;
441
442 lwb = kmem_cache_alloc(zil_lwb_cache, KM_SLEEP);
443 lwb->lwb_zilog = zilog;
444 lwb->lwb_blk = *bp;
445 lwb->lwb_buf = zio_buf_alloc(BP_GET_LSIZE(bp));
446 lwb->lwb_max_txg = txg;
447 lwb->lwb_zio = NULL;
448 lwb->lwb_tx = NULL;
449 if (BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_ZILOG2) {
450 lwb->lwb_nused = sizeof (zil_chain_t);
451 lwb->lwb_sz = BP_GET_LSIZE(bp);
452 } else {
453 lwb->lwb_nused = 0;
454 lwb->lwb_sz = BP_GET_LSIZE(bp) - sizeof (zil_chain_t);
455 }
456
457 mutex_enter(&zilog->zl_lock);
458 list_insert_tail(&zilog->zl_lwb_list, lwb);
459 mutex_exit(&zilog->zl_lock);
460
461 return (lwb);
462 }
463
464 /*
465 * Called when we create in-memory log transactions so that we know
466 * to cleanup the itxs at the end of spa_sync().
467 */
468 void
469 zilog_dirty(zilog_t *zilog, uint64_t txg)
470 {
471 dsl_pool_t *dp = zilog->zl_dmu_pool;
472 dsl_dataset_t *ds = dmu_objset_ds(zilog->zl_os);
473
474 if (dsl_dataset_is_snapshot(ds))
475 panic("dirtying snapshot!");
476
477 if (txg_list_add(&dp->dp_dirty_zilogs, zilog, txg)) {
478 /* up the hold count until we can be written out */
479 dmu_buf_add_ref(ds->ds_dbuf, zilog);
480 }
481 }
482
483 boolean_t
484 zilog_is_dirty(zilog_t *zilog)
485 {
486 dsl_pool_t *dp = zilog->zl_dmu_pool;
487
488 for (int t = 0; t < TXG_SIZE; t++) {
489 if (txg_list_member(&dp->dp_dirty_zilogs, zilog, t))
490 return (B_TRUE);
491 }
492 return (B_FALSE);
493 }
494
495 /*
496 * Create an on-disk intent log.
497 */
498 static lwb_t *
499 zil_create(zilog_t *zilog)
500 {
501 const zil_header_t *zh = zilog->zl_header;
502 lwb_t *lwb = NULL;
503 uint64_t txg = 0;
504 dmu_tx_t *tx = NULL;
505 blkptr_t blk;
506 int error = 0;
507
508 /*
509 * Wait for any previous destroy to complete.
510 */
511 txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg);
512
513 ASSERT(zh->zh_claim_txg == 0);
514 ASSERT(zh->zh_replay_seq == 0);
515
516 blk = zh->zh_log;
517
518 /*
519 * Allocate an initial log block if:
520 * - there isn't one already
521 * - the existing block is the wrong endianess
522 */
523 if (BP_IS_HOLE(&blk) || BP_SHOULD_BYTESWAP(&blk)) {
524 tx = dmu_tx_create(zilog->zl_os);
525 VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0);
526 dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
527 txg = dmu_tx_get_txg(tx);
528
529 if (!BP_IS_HOLE(&blk)) {
530 zio_free_zil(zilog->zl_spa, txg, &blk);
531 BP_ZERO(&blk);
532 }
533
534 error = zio_alloc_zil(zilog->zl_spa, txg, &blk, NULL,
535 ZIL_MIN_BLKSZ, zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
536
537 if (error == 0)
538 zil_init_log_chain(zilog, &blk);
539 }
540
541 /*
542 * Allocate a log write buffer (lwb) for the first log block.
543 */
544 if (error == 0)
545 lwb = zil_alloc_lwb(zilog, &blk, txg);
546
547 /*
548 * If we just allocated the first log block, commit our transaction
549 * and wait for zil_sync() to stuff the block poiner into zh_log.
550 * (zh is part of the MOS, so we cannot modify it in open context.)
551 */
552 if (tx != NULL) {
553 dmu_tx_commit(tx);
554 txg_wait_synced(zilog->zl_dmu_pool, txg);
555 }
556
557 ASSERT(bcmp(&blk, &zh->zh_log, sizeof (blk)) == 0);
558
559 return (lwb);
560 }
561
562 /*
563 * In one tx, free all log blocks and clear the log header.
564 * If keep_first is set, then we're replaying a log with no content.
565 * We want to keep the first block, however, so that the first
566 * synchronous transaction doesn't require a txg_wait_synced()
567 * in zil_create(). We don't need to txg_wait_synced() here either
568 * when keep_first is set, because both zil_create() and zil_destroy()
569 * will wait for any in-progress destroys to complete.
570 */
571 void
572 zil_destroy(zilog_t *zilog, boolean_t keep_first)
573 {
574 const zil_header_t *zh = zilog->zl_header;
575 lwb_t *lwb;
576 dmu_tx_t *tx;
577 uint64_t txg;
578
579 /*
580 * Wait for any previous destroy to complete.
581 */
582 txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg);
583
584 zilog->zl_old_header = *zh; /* debugging aid */
585
586 if (BP_IS_HOLE(&zh->zh_log))
587 return;
588
589 tx = dmu_tx_create(zilog->zl_os);
590 VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0);
591 dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
592 txg = dmu_tx_get_txg(tx);
593
594 mutex_enter(&zilog->zl_lock);
595
596 ASSERT3U(zilog->zl_destroy_txg, <, txg);
597 zilog->zl_destroy_txg = txg;
598 zilog->zl_keep_first = keep_first;
599
600 if (!list_is_empty(&zilog->zl_lwb_list)) {
601 ASSERT(zh->zh_claim_txg == 0);
602 VERIFY(!keep_first);
603 while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) {
604 list_remove(&zilog->zl_lwb_list, lwb);
605 if (lwb->lwb_buf != NULL)
606 zio_buf_free(lwb->lwb_buf, lwb->lwb_sz);
607 zio_free_zil(zilog->zl_spa, txg, &lwb->lwb_blk);
608 kmem_cache_free(zil_lwb_cache, lwb);
609 }
610 } else if (!keep_first) {
611 zil_destroy_sync(zilog, tx);
612 }
613 mutex_exit(&zilog->zl_lock);
614
615 dmu_tx_commit(tx);
616 }
617
618 void
619 zil_destroy_sync(zilog_t *zilog, dmu_tx_t *tx)
620 {
621 ASSERT(list_is_empty(&zilog->zl_lwb_list));
622 (void) zil_parse(zilog, zil_free_log_block,
623 zil_free_log_record, tx, zilog->zl_header->zh_claim_txg);
624 }
625
626 int
627 zil_claim(dsl_pool_t *dp, dsl_dataset_t *ds, void *txarg)
628 {
629 dmu_tx_t *tx = txarg;
630 uint64_t first_txg = dmu_tx_get_txg(tx);
631 zilog_t *zilog;
632 zil_header_t *zh;
633 objset_t *os;
634 int error;
635
636 error = dmu_objset_own_obj(dp, ds->ds_object,
637 DMU_OST_ANY, B_FALSE, FTAG, &os);
638 if (error != 0) {
639 cmn_err(CE_WARN, "can't open objset %llu, error %d",
640 (unsigned long long)ds->ds_object, error);
641 return (0);
642 }
643
644 zilog = dmu_objset_zil(os);
645 zh = zil_header_in_syncing_context(zilog);
646
647 if (spa_get_log_state(zilog->zl_spa) == SPA_LOG_CLEAR) {
648 if (!BP_IS_HOLE(&zh->zh_log))
649 zio_free_zil(zilog->zl_spa, first_txg, &zh->zh_log);
650 BP_ZERO(&zh->zh_log);
651 dsl_dataset_dirty(dmu_objset_ds(os), tx);
652 dmu_objset_disown(os, FTAG);
653 return (0);
654 }
655
656 /*
657 * Claim all log blocks if we haven't already done so, and remember
658 * the highest claimed sequence number. This ensures that if we can
659 * read only part of the log now (e.g. due to a missing device),
660 * but we can read the entire log later, we will not try to replay
661 * or destroy beyond the last block we successfully claimed.
662 */
663 ASSERT3U(zh->zh_claim_txg, <=, first_txg);
664 if (zh->zh_claim_txg == 0 && !BP_IS_HOLE(&zh->zh_log)) {
665 (void) zil_parse(zilog, zil_claim_log_block,
666 zil_claim_log_record, tx, first_txg);
667 zh->zh_claim_txg = first_txg;
668 zh->zh_claim_blk_seq = zilog->zl_parse_blk_seq;
669 zh->zh_claim_lr_seq = zilog->zl_parse_lr_seq;
670 if (zilog->zl_parse_lr_count || zilog->zl_parse_blk_count > 1)
671 zh->zh_flags |= ZIL_REPLAY_NEEDED;
672 zh->zh_flags |= ZIL_CLAIM_LR_SEQ_VALID;
673 dsl_dataset_dirty(dmu_objset_ds(os), tx);
674 }
675
676 ASSERT3U(first_txg, ==, (spa_last_synced_txg(zilog->zl_spa) + 1));
677 dmu_objset_disown(os, FTAG);
678 return (0);
679 }
680
681 /*
682 * Check the log by walking the log chain.
683 * Checksum errors are ok as they indicate the end of the chain.
684 * Any other error (no device or read failure) returns an error.
685 */
686 int
687 zil_check_log_chain(dsl_pool_t *dp, dsl_dataset_t *ds, void *tx)
688 {
689 zilog_t *zilog;
690 objset_t *os;
691 blkptr_t *bp;
692 int error;
693
694 ASSERT(tx == NULL);
695
696 error = dmu_objset_from_ds(ds, &os);
697 if (error != 0) {
698 cmn_err(CE_WARN, "can't open objset %llu, error %d",
699 (unsigned long long)ds->ds_object, error);
700 return (0);
701 }
702
703 zilog = dmu_objset_zil(os);
704 bp = (blkptr_t *)&zilog->zl_header->zh_log;
705
706 /*
707 * Check the first block and determine if it's on a log device
708 * which may have been removed or faulted prior to loading this
709 * pool. If so, there's no point in checking the rest of the log
710 * as its content should have already been synced to the pool.
711 */
712 if (!BP_IS_HOLE(bp)) {
713 vdev_t *vd;
714 boolean_t valid = B_TRUE;
715
716 spa_config_enter(os->os_spa, SCL_STATE, FTAG, RW_READER);
717 vd = vdev_lookup_top(os->os_spa, DVA_GET_VDEV(&bp->blk_dva[0]));
718 if (vd->vdev_islog && vdev_is_dead(vd))
719 valid = vdev_log_state_valid(vd);
720 spa_config_exit(os->os_spa, SCL_STATE, FTAG);
721
722 if (!valid)
723 return (0);
724 }
725
726 /*
727 * Because tx == NULL, zil_claim_log_block() will not actually claim
728 * any blocks, but just determine whether it is possible to do so.
729 * In addition to checking the log chain, zil_claim_log_block()
730 * will invoke zio_claim() with a done func of spa_claim_notify(),
731 * which will update spa_max_claim_txg. See spa_load() for details.
732 */
733 error = zil_parse(zilog, zil_claim_log_block, zil_claim_log_record, tx,
734 zilog->zl_header->zh_claim_txg ? -1ULL : spa_first_txg(os->os_spa));
735
736 return ((error == ECKSUM || error == ENOENT) ? 0 : error);
737 }
738
739 static int
740 zil_vdev_compare(const void *x1, const void *x2)
741 {
742 const uint64_t v1 = ((zil_vdev_node_t *)x1)->zv_vdev;
743 const uint64_t v2 = ((zil_vdev_node_t *)x2)->zv_vdev;
744
745 if (v1 < v2)
746 return (-1);
747 if (v1 > v2)
748 return (1);
749
750 return (0);
751 }
752
753 void
754 zil_add_block(zilog_t *zilog, const blkptr_t *bp)
755 {
756 avl_tree_t *t = &zilog->zl_vdev_tree;
757 avl_index_t where;
758 zil_vdev_node_t *zv, zvsearch;
759 int ndvas = BP_GET_NDVAS(bp);
760 int i;
761
762 if (zfs_nocacheflush)
763 return;
764
765 ASSERT(zilog->zl_writer);
766
767 /*
768 * Even though we're zl_writer, we still need a lock because the
769 * zl_get_data() callbacks may have dmu_sync() done callbacks
770 * that will run concurrently.
771 */
772 mutex_enter(&zilog->zl_vdev_lock);
773 for (i = 0; i < ndvas; i++) {
774 zvsearch.zv_vdev = DVA_GET_VDEV(&bp->blk_dva[i]);
775 if (avl_find(t, &zvsearch, &where) == NULL) {
776 zv = kmem_alloc(sizeof (*zv), KM_SLEEP);
777 zv->zv_vdev = zvsearch.zv_vdev;
778 avl_insert(t, zv, where);
779 }
780 }
781 mutex_exit(&zilog->zl_vdev_lock);
782 }
783
784 static void
785 zil_flush_vdevs(zilog_t *zilog)
786 {
787 spa_t *spa = zilog->zl_spa;
788 avl_tree_t *t = &zilog->zl_vdev_tree;
789 void *cookie = NULL;
790 zil_vdev_node_t *zv;
791 zio_t *zio;
792
793 ASSERT(zilog->zl_writer);
794
795 /*
796 * We don't need zl_vdev_lock here because we're the zl_writer,
797 * and all zl_get_data() callbacks are done.
798 */
799 if (avl_numnodes(t) == 0)
800 return;
801
802 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
803
804 zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
805
806 while ((zv = avl_destroy_nodes(t, &cookie)) != NULL) {
807 vdev_t *vd = vdev_lookup_top(spa, zv->zv_vdev);
808 if (vd != NULL)
809 zio_flush(zio, vd);
810 kmem_free(zv, sizeof (*zv));
811 }
812
813 /*
814 * Wait for all the flushes to complete. Not all devices actually
815 * support the DKIOCFLUSHWRITECACHE ioctl, so it's OK if it fails.
816 */
817 (void) zio_wait(zio);
818
819 spa_config_exit(spa, SCL_STATE, FTAG);
820 }
821
822 /*
823 * Function called when a log block write completes
824 */
825 static void
826 zil_lwb_write_done(zio_t *zio)
827 {
828 lwb_t *lwb = zio->io_private;
829 zilog_t *zilog = lwb->lwb_zilog;
830 dmu_tx_t *tx = lwb->lwb_tx;
831
832 ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF);
833 ASSERT(BP_GET_TYPE(zio->io_bp) == DMU_OT_INTENT_LOG);
834 ASSERT(BP_GET_LEVEL(zio->io_bp) == 0);
835 ASSERT(BP_GET_BYTEORDER(zio->io_bp) == ZFS_HOST_BYTEORDER);
836 ASSERT(!BP_IS_GANG(zio->io_bp));
837 ASSERT(!BP_IS_HOLE(zio->io_bp));
838 ASSERT(BP_GET_FILL(zio->io_bp) == 0);
839
840 /*
841 * Ensure the lwb buffer pointer is cleared before releasing
842 * the txg. If we have had an allocation failure and
843 * the txg is waiting to sync then we want want zil_sync()
844 * to remove the lwb so that it's not picked up as the next new
845 * one in zil_commit_writer(). zil_sync() will only remove
846 * the lwb if lwb_buf is null.
847 */
848 zio_buf_free(lwb->lwb_buf, lwb->lwb_sz);
849 mutex_enter(&zilog->zl_lock);
850 lwb->lwb_buf = NULL;
851 lwb->lwb_tx = NULL;
852 mutex_exit(&zilog->zl_lock);
853
854 /*
855 * Now that we've written this log block, we have a stable pointer
856 * to the next block in the chain, so it's OK to let the txg in
857 * which we allocated the next block sync.
858 */
859 dmu_tx_commit(tx);
860 }
861
862 /*
863 * Initialize the io for a log block.
864 */
865 static void
866 zil_lwb_write_init(zilog_t *zilog, lwb_t *lwb)
867 {
868 zbookmark_t zb;
869
870 SET_BOOKMARK(&zb, lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_OBJSET],
871 ZB_ZIL_OBJECT, ZB_ZIL_LEVEL,
872 lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_SEQ]);
873
874 if (zilog->zl_root_zio == NULL) {
875 zilog->zl_root_zio = zio_root(zilog->zl_spa, NULL, NULL,
876 ZIO_FLAG_CANFAIL);
877 }
878 if (lwb->lwb_zio == NULL) {
879 lwb->lwb_zio = zio_rewrite(zilog->zl_root_zio, zilog->zl_spa,
880 0, &lwb->lwb_blk, lwb->lwb_buf, BP_GET_LSIZE(&lwb->lwb_blk),
881 zil_lwb_write_done, lwb, ZIO_PRIORITY_SYNC_WRITE,
882 ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE, &zb);
883 }
884 }
885
886 /*
887 * Define a limited set of intent log block sizes.
888 *
889 * These must be a multiple of 4KB. Note only the amount used (again
890 * aligned to 4KB) actually gets written. However, we can't always just
891 * allocate SPA_MAXBLOCKSIZE as the slog space could be exhausted.
892 */
893 uint64_t zil_block_buckets[] = {
894 4096, /* non TX_WRITE */
895 8192+4096, /* data base */
896 32*1024 + 4096, /* NFS writes */
897 UINT64_MAX
898 };
899
900 /*
901 * Use the slog as long as the logbias is 'latency' and the current commit size
902 * is less than the limit or the total list size is less than 2X the limit.
903 * Limit checking is disabled by setting zil_slog_limit to UINT64_MAX.
904 */
905 uint64_t zil_slog_limit = 1024 * 1024;
906 #define USE_SLOG(zilog) (((zilog)->zl_logbias == ZFS_LOGBIAS_LATENCY) && \
907 (((zilog)->zl_cur_used < zil_slog_limit) || \
908 ((zilog)->zl_itx_list_sz < (zil_slog_limit << 1))))
909
910 /*
911 * Start a log block write and advance to the next log block.
912 * Calls are serialized.
913 */
914 static lwb_t *
915 zil_lwb_write_start(zilog_t *zilog, lwb_t *lwb)
916 {
917 lwb_t *nlwb = NULL;
918 zil_chain_t *zilc;
919 spa_t *spa = zilog->zl_spa;
920 blkptr_t *bp;
921 dmu_tx_t *tx;
922 uint64_t txg;
923 uint64_t zil_blksz, wsz;
924 int i, error;
925
926 if (BP_GET_CHECKSUM(&lwb->lwb_blk) == ZIO_CHECKSUM_ZILOG2) {
927 zilc = (zil_chain_t *)lwb->lwb_buf;
928 bp = &zilc->zc_next_blk;
929 } else {
930 zilc = (zil_chain_t *)(lwb->lwb_buf + lwb->lwb_sz);
931 bp = &zilc->zc_next_blk;
932 }
933
934 ASSERT(lwb->lwb_nused <= lwb->lwb_sz);
935
936 /*
937 * Allocate the next block and save its address in this block
938 * before writing it in order to establish the log chain.
939 * Note that if the allocation of nlwb synced before we wrote
940 * the block that points at it (lwb), we'd leak it if we crashed.
941 * Therefore, we don't do dmu_tx_commit() until zil_lwb_write_done().
942 * We dirty the dataset to ensure that zil_sync() will be called
943 * to clean up in the event of allocation failure or I/O failure.
944 */
945 tx = dmu_tx_create(zilog->zl_os);
946 VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0);
947 dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
948 txg = dmu_tx_get_txg(tx);
949
950 lwb->lwb_tx = tx;
951
952 /*
953 * Log blocks are pre-allocated. Here we select the size of the next
954 * block, based on size used in the last block.
955 * - first find the smallest bucket that will fit the block from a
956 * limited set of block sizes. This is because it's faster to write
957 * blocks allocated from the same metaslab as they are adjacent or
958 * close.
959 * - next find the maximum from the new suggested size and an array of
960 * previous sizes. This lessens a picket fence effect of wrongly
961 * guesssing the size if we have a stream of say 2k, 64k, 2k, 64k
962 * requests.
963 *
964 * Note we only write what is used, but we can't just allocate
965 * the maximum block size because we can exhaust the available
966 * pool log space.
967 */
968 zil_blksz = zilog->zl_cur_used + sizeof (zil_chain_t);
969 for (i = 0; zil_blksz > zil_block_buckets[i]; i++)
970 continue;
971 zil_blksz = zil_block_buckets[i];
972 if (zil_blksz == UINT64_MAX)
973 zil_blksz = SPA_MAXBLOCKSIZE;
974 zilog->zl_prev_blks[zilog->zl_prev_rotor] = zil_blksz;
975 for (i = 0; i < ZIL_PREV_BLKS; i++)
976 zil_blksz = MAX(zil_blksz, zilog->zl_prev_blks[i]);
977 zilog->zl_prev_rotor = (zilog->zl_prev_rotor + 1) & (ZIL_PREV_BLKS - 1);
978
979 BP_ZERO(bp);
980 /* pass the old blkptr in order to spread log blocks across devs */
981 error = zio_alloc_zil(spa, txg, bp, &lwb->lwb_blk, zil_blksz,
982 USE_SLOG(zilog));
983 if (error == 0) {
984 ASSERT3U(bp->blk_birth, ==, txg);
985 bp->blk_cksum = lwb->lwb_blk.blk_cksum;
986 bp->blk_cksum.zc_word[ZIL_ZC_SEQ]++;
987
988 /*
989 * Allocate a new log write buffer (lwb).
990 */
991 nlwb = zil_alloc_lwb(zilog, bp, txg);
992
993 /* Record the block for later vdev flushing */
994 zil_add_block(zilog, &lwb->lwb_blk);
995 }
996
997 if (BP_GET_CHECKSUM(&lwb->lwb_blk) == ZIO_CHECKSUM_ZILOG2) {
998 /* For Slim ZIL only write what is used. */
999 wsz = P2ROUNDUP_TYPED(lwb->lwb_nused, ZIL_MIN_BLKSZ, uint64_t);
1000 ASSERT3U(wsz, <=, lwb->lwb_sz);
1001 zio_shrink(lwb->lwb_zio, wsz);
1002
1003 } else {
1004 wsz = lwb->lwb_sz;
1005 }
1006
1007 zilc->zc_pad = 0;
1008 zilc->zc_nused = lwb->lwb_nused;
1009 zilc->zc_eck.zec_cksum = lwb->lwb_blk.blk_cksum;
1010
1011 /*
1012 * clear unused data for security
1013 */
1014 bzero(lwb->lwb_buf + lwb->lwb_nused, wsz - lwb->lwb_nused);
1015
1016 zio_nowait(lwb->lwb_zio); /* Kick off the write for the old log block */
1017
1018 /*
1019 * If there was an allocation failure then nlwb will be null which
1020 * forces a txg_wait_synced().
1021 */
1022 return (nlwb);
1023 }
1024
1025 static lwb_t *
1026 zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb)
1027 {
1028 lr_t *lrc = &itx->itx_lr; /* common log record */
1029 lr_write_t *lrw = (lr_write_t *)lrc;
1030 char *lr_buf;
1031 uint64_t txg = lrc->lrc_txg;
1032 uint64_t reclen = lrc->lrc_reclen;
1033 uint64_t dlen = 0;
1034
1035 if (lwb == NULL)
1036 return (NULL);
1037
1038 ASSERT(lwb->lwb_buf != NULL);
1039 ASSERT(zilog_is_dirty(zilog) ||
1040 spa_freeze_txg(zilog->zl_spa) != UINT64_MAX);
1041
1042 if (lrc->lrc_txtype == TX_WRITE && itx->itx_wr_state == WR_NEED_COPY)
1043 dlen = P2ROUNDUP_TYPED(
1044 lrw->lr_length, sizeof (uint64_t), uint64_t);
1045
1046 zilog->zl_cur_used += (reclen + dlen);
1047
1048 zil_lwb_write_init(zilog, lwb);
1049
1050 /*
1051 * If this record won't fit in the current log block, start a new one.
1052 */
1053 if (lwb->lwb_nused + reclen + dlen > lwb->lwb_sz) {
1054 lwb = zil_lwb_write_start(zilog, lwb);
1055 if (lwb == NULL)
1056 return (NULL);
1057 zil_lwb_write_init(zilog, lwb);
1058 ASSERT(LWB_EMPTY(lwb));
1059 if (lwb->lwb_nused + reclen + dlen > lwb->lwb_sz) {
1060 txg_wait_synced(zilog->zl_dmu_pool, txg);
1061 return (lwb);
1062 }
1063 }
1064
1065 lr_buf = lwb->lwb_buf + lwb->lwb_nused;
1066 bcopy(lrc, lr_buf, reclen);
1067 lrc = (lr_t *)lr_buf;
1068 lrw = (lr_write_t *)lrc;
1069
1070 /*
1071 * If it's a write, fetch the data or get its blkptr as appropriate.
1072 */
1073 if (lrc->lrc_txtype == TX_WRITE) {
1074 if (txg > spa_freeze_txg(zilog->zl_spa))
1075 txg_wait_synced(zilog->zl_dmu_pool, txg);
1076 if (itx->itx_wr_state != WR_COPIED) {
1077 char *dbuf;
1078 int error;
1079
1080 if (dlen) {
1081 ASSERT(itx->itx_wr_state == WR_NEED_COPY);
1082 dbuf = lr_buf + reclen;
1083 lrw->lr_common.lrc_reclen += dlen;
1084 } else {
1085 ASSERT(itx->itx_wr_state == WR_INDIRECT);
1086 dbuf = NULL;
1087 }
1088 error = zilog->zl_get_data(
1089 itx->itx_private, lrw, dbuf, lwb->lwb_zio);
1090 if (error == EIO) {
1091 txg_wait_synced(zilog->zl_dmu_pool, txg);
1092 return (lwb);
1093 }
1094 if (error != 0) {
1095 ASSERT(error == ENOENT || error == EEXIST ||
1096 error == EALREADY);
1097 return (lwb);
1098 }
1099 }
1100 }
1101
1102 /*
1103 * We're actually making an entry, so update lrc_seq to be the
1104 * log record sequence number. Note that this is generally not
1105 * equal to the itx sequence number because not all transactions
1106 * are synchronous, and sometimes spa_sync() gets there first.
1107 */
1108 lrc->lrc_seq = ++zilog->zl_lr_seq; /* we are single threaded */
1109 lwb->lwb_nused += reclen + dlen;
1110 lwb->lwb_max_txg = MAX(lwb->lwb_max_txg, txg);
1111 ASSERT3U(lwb->lwb_nused, <=, lwb->lwb_sz);
1112 ASSERT0(P2PHASE(lwb->lwb_nused, sizeof (uint64_t)));
1113
1114 return (lwb);
1115 }
1116
1117 itx_t *
1118 zil_itx_create(uint64_t txtype, size_t lrsize)
1119 {
1120 itx_t *itx;
1121
1122 lrsize = P2ROUNDUP_TYPED(lrsize, sizeof (uint64_t), size_t);
1123
1124 itx = kmem_alloc(offsetof(itx_t, itx_lr) + lrsize, KM_SLEEP);
1125 itx->itx_lr.lrc_txtype = txtype;
1126 itx->itx_lr.lrc_reclen = lrsize;
1127 itx->itx_sod = lrsize; /* if write & WR_NEED_COPY will be increased */
1128 itx->itx_lr.lrc_seq = 0; /* defensive */
1129 itx->itx_sync = B_TRUE; /* default is synchronous */
1130
1131 return (itx);
1132 }
1133
1134 void
1135 zil_itx_destroy(itx_t *itx)
1136 {
1137 kmem_free(itx, offsetof(itx_t, itx_lr) + itx->itx_lr.lrc_reclen);
1138 }
1139
1140 /*
1141 * Free up the sync and async itxs. The itxs_t has already been detached
1142 * so no locks are needed.
1143 */
1144 static void
1145 zil_itxg_clean(itxs_t *itxs)
1146 {
1147 itx_t *itx;
1148 list_t *list;
1149 avl_tree_t *t;
1150 void *cookie;
1151 itx_async_node_t *ian;
1152
1153 list = &itxs->i_sync_list;
1154 while ((itx = list_head(list)) != NULL) {
1155 list_remove(list, itx);
1156 kmem_free(itx, offsetof(itx_t, itx_lr) +
1157 itx->itx_lr.lrc_reclen);
1158 }
1159
1160 cookie = NULL;
1161 t = &itxs->i_async_tree;
1162 while ((ian = avl_destroy_nodes(t, &cookie)) != NULL) {
1163 list = &ian->ia_list;
1164 while ((itx = list_head(list)) != NULL) {
1165 list_remove(list, itx);
1166 kmem_free(itx, offsetof(itx_t, itx_lr) +
1167 itx->itx_lr.lrc_reclen);
1168 }
1169 list_destroy(list);
1170 kmem_free(ian, sizeof (itx_async_node_t));
1171 }
1172 avl_destroy(t);
1173
1174 kmem_free(itxs, sizeof (itxs_t));
1175 }
1176
1177 static int
1178 zil_aitx_compare(const void *x1, const void *x2)
1179 {
1180 const uint64_t o1 = ((itx_async_node_t *)x1)->ia_foid;
1181 const uint64_t o2 = ((itx_async_node_t *)x2)->ia_foid;
1182
1183 if (o1 < o2)
1184 return (-1);
1185 if (o1 > o2)
1186 return (1);
1187
1188 return (0);
1189 }
1190
1191 /*
1192 * Remove all async itx with the given oid.
1193 */
1194 static void
1195 zil_remove_async(zilog_t *zilog, uint64_t oid)
1196 {
1197 uint64_t otxg, txg;
1198 itx_async_node_t *ian;
1199 avl_tree_t *t;
1200 avl_index_t where;
1201 list_t clean_list;
1202 itx_t *itx;
1203
1204 ASSERT(oid != 0);
1205 list_create(&clean_list, sizeof (itx_t), offsetof(itx_t, itx_node));
1206
1207 if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */
1208 otxg = ZILTEST_TXG;
1209 else
1210 otxg = spa_last_synced_txg(zilog->zl_spa) + 1;
1211
1212 for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) {
1213 itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK];
1214
1215 mutex_enter(&itxg->itxg_lock);
1216 if (itxg->itxg_txg != txg) {
1217 mutex_exit(&itxg->itxg_lock);
1218 continue;
1219 }
1220
1221 /*
1222 * Locate the object node and append its list.
1223 */
1224 t = &itxg->itxg_itxs->i_async_tree;
1225 ian = avl_find(t, &oid, &where);
1226 if (ian != NULL)
1227 list_move_tail(&clean_list, &ian->ia_list);
1228 mutex_exit(&itxg->itxg_lock);
1229 }
1230 while ((itx = list_head(&clean_list)) != NULL) {
1231 list_remove(&clean_list, itx);
1232 kmem_free(itx, offsetof(itx_t, itx_lr) +
1233 itx->itx_lr.lrc_reclen);
1234 }
1235 list_destroy(&clean_list);
1236 }
1237
1238 void
1239 zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx)
1240 {
1241 uint64_t txg;
1242 itxg_t *itxg;
1243 itxs_t *itxs, *clean = NULL;
1244
1245 /*
1246 * Object ids can be re-instantiated in the next txg so
1247 * remove any async transactions to avoid future leaks.
1248 * This can happen if a fsync occurs on the re-instantiated
1249 * object for a WR_INDIRECT or WR_NEED_COPY write, which gets
1250 * the new file data and flushes a write record for the old object.
1251 */
1252 if ((itx->itx_lr.lrc_txtype & ~TX_CI) == TX_REMOVE)
1253 zil_remove_async(zilog, itx->itx_oid);
1254
1255 /*
1256 * Ensure the data of a renamed file is committed before the rename.
1257 */
1258 if ((itx->itx_lr.lrc_txtype & ~TX_CI) == TX_RENAME)
1259 zil_async_to_sync(zilog, itx->itx_oid);
1260
1261 if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX)
1262 txg = ZILTEST_TXG;
1263 else
1264 txg = dmu_tx_get_txg(tx);
1265
1266 itxg = &zilog->zl_itxg[txg & TXG_MASK];
1267 mutex_enter(&itxg->itxg_lock);
1268 itxs = itxg->itxg_itxs;
1269 if (itxg->itxg_txg != txg) {
1270 if (itxs != NULL) {
1271 /*
1272 * The zil_clean callback hasn't got around to cleaning
1273 * this itxg. Save the itxs for release below.
1274 * This should be rare.
1275 */
1276 atomic_add_64(&zilog->zl_itx_list_sz, -itxg->itxg_sod);
1277 itxg->itxg_sod = 0;
1278 clean = itxg->itxg_itxs;
1279 }
1280 ASSERT(itxg->itxg_sod == 0);
1281 itxg->itxg_txg = txg;
1282 itxs = itxg->itxg_itxs = kmem_zalloc(sizeof (itxs_t), KM_SLEEP);
1283
1284 list_create(&itxs->i_sync_list, sizeof (itx_t),
1285 offsetof(itx_t, itx_node));
1286 avl_create(&itxs->i_async_tree, zil_aitx_compare,
1287 sizeof (itx_async_node_t),
1288 offsetof(itx_async_node_t, ia_node));
1289 }
1290 if (itx->itx_sync) {
1291 list_insert_tail(&itxs->i_sync_list, itx);
1292 atomic_add_64(&zilog->zl_itx_list_sz, itx->itx_sod);
1293 itxg->itxg_sod += itx->itx_sod;
1294 } else {
1295 avl_tree_t *t = &itxs->i_async_tree;
1296 uint64_t foid = ((lr_ooo_t *)&itx->itx_lr)->lr_foid;
1297 itx_async_node_t *ian;
1298 avl_index_t where;
1299
1300 ian = avl_find(t, &foid, &where);
1301 if (ian == NULL) {
1302 ian = kmem_alloc(sizeof (itx_async_node_t), KM_SLEEP);
1303 list_create(&ian->ia_list, sizeof (itx_t),
1304 offsetof(itx_t, itx_node));
1305 ian->ia_foid = foid;
1306 avl_insert(t, ian, where);
1307 }
1308 list_insert_tail(&ian->ia_list, itx);
1309 }
1310
1311 itx->itx_lr.lrc_txg = dmu_tx_get_txg(tx);
1312 zilog_dirty(zilog, txg);
1313 mutex_exit(&itxg->itxg_lock);
1314
1315 /* Release the old itxs now we've dropped the lock */
1316 if (clean != NULL)
1317 zil_itxg_clean(clean);
1318 }
1319
1320 /*
1321 * If there are any in-memory intent log transactions which have now been
1322 * synced then start up a taskq to free them. We should only do this after we
1323 * have written out the uberblocks (i.e. txg has been comitted) so that
1324 * don't inadvertently clean out in-memory log records that would be required
1325 * by zil_commit().
1326 */
1327 void
1328 zil_clean(zilog_t *zilog, uint64_t synced_txg)
1329 {
1330 itxg_t *itxg = &zilog->zl_itxg[synced_txg & TXG_MASK];
1331 itxs_t *clean_me;
1332
1333 mutex_enter(&itxg->itxg_lock);
1334 if (itxg->itxg_itxs == NULL || itxg->itxg_txg == ZILTEST_TXG) {
1335 mutex_exit(&itxg->itxg_lock);
1336 return;
1337 }
1338 ASSERT3U(itxg->itxg_txg, <=, synced_txg);
1339 ASSERT(itxg->itxg_txg != 0);
1340 ASSERT(zilog->zl_clean_taskq != NULL);
1341 atomic_add_64(&zilog->zl_itx_list_sz, -itxg->itxg_sod);
1342 itxg->itxg_sod = 0;
1343 clean_me = itxg->itxg_itxs;
1344 itxg->itxg_itxs = NULL;
1345 itxg->itxg_txg = 0;
1346 mutex_exit(&itxg->itxg_lock);
1347 /*
1348 * Preferably start a task queue to free up the old itxs but
1349 * if taskq_dispatch can't allocate resources to do that then
1350 * free it in-line. This should be rare. Note, using TQ_SLEEP
1351 * created a bad performance problem.
1352 */
1353 if (taskq_dispatch(zilog->zl_clean_taskq,
1354 (void (*)(void *))zil_itxg_clean, clean_me, TQ_NOSLEEP) == NULL)
1355 zil_itxg_clean(clean_me);
1356 }
1357
1358 /*
1359 * Get the list of itxs to commit into zl_itx_commit_list.
1360 */
1361 static void
1362 zil_get_commit_list(zilog_t *zilog)
1363 {
1364 uint64_t otxg, txg;
1365 list_t *commit_list = &zilog->zl_itx_commit_list;
1366 uint64_t push_sod = 0;
1367
1368 if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */
1369 otxg = ZILTEST_TXG;
1370 else
1371 otxg = spa_last_synced_txg(zilog->zl_spa) + 1;
1372
1373 for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) {
1374 itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK];
1375
1376 mutex_enter(&itxg->itxg_lock);
1377 if (itxg->itxg_txg != txg) {
1378 mutex_exit(&itxg->itxg_lock);
1379 continue;
1380 }
1381
1382 list_move_tail(commit_list, &itxg->itxg_itxs->i_sync_list);
1383 push_sod += itxg->itxg_sod;
1384 itxg->itxg_sod = 0;
1385
1386 mutex_exit(&itxg->itxg_lock);
1387 }
1388 atomic_add_64(&zilog->zl_itx_list_sz, -push_sod);
1389 }
1390
1391 /*
1392 * Move the async itxs for a specified object to commit into sync lists.
1393 */
1394 static void
1395 zil_async_to_sync(zilog_t *zilog, uint64_t foid)
1396 {
1397 uint64_t otxg, txg;
1398 itx_async_node_t *ian;
1399 avl_tree_t *t;
1400 avl_index_t where;
1401
1402 if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */
1403 otxg = ZILTEST_TXG;
1404 else
1405 otxg = spa_last_synced_txg(zilog->zl_spa) + 1;
1406
1407 for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) {
1408 itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK];
1409
1410 mutex_enter(&itxg->itxg_lock);
1411 if (itxg->itxg_txg != txg) {
1412 mutex_exit(&itxg->itxg_lock);
1413 continue;
1414 }
1415
1416 /*
1417 * If a foid is specified then find that node and append its
1418 * list. Otherwise walk the tree appending all the lists
1419 * to the sync list. We add to the end rather than the
1420 * beginning to ensure the create has happened.
1421 */
1422 t = &itxg->itxg_itxs->i_async_tree;
1423 if (foid != 0) {
1424 ian = avl_find(t, &foid, &where);
1425 if (ian != NULL) {
1426 list_move_tail(&itxg->itxg_itxs->i_sync_list,
1427 &ian->ia_list);
1428 }
1429 } else {
1430 void *cookie = NULL;
1431
1432 while ((ian = avl_destroy_nodes(t, &cookie)) != NULL) {
1433 list_move_tail(&itxg->itxg_itxs->i_sync_list,
1434 &ian->ia_list);
1435 list_destroy(&ian->ia_list);
1436 kmem_free(ian, sizeof (itx_async_node_t));
1437 }
1438 }
1439 mutex_exit(&itxg->itxg_lock);
1440 }
1441 }
1442
1443 static void
1444 zil_commit_writer(zilog_t *zilog)
1445 {
1446 uint64_t txg;
1447 itx_t *itx;
1448 lwb_t *lwb;
1449 spa_t *spa = zilog->zl_spa;
1450 int error = 0;
1451
1452 ASSERT(zilog->zl_root_zio == NULL);
1453
1454 mutex_exit(&zilog->zl_lock);
1455
1456 zil_get_commit_list(zilog);
1457
1458 /*
1459 * Return if there's nothing to commit before we dirty the fs by
1460 * calling zil_create().
1461 */
1462 if (list_head(&zilog->zl_itx_commit_list) == NULL) {
1463 mutex_enter(&zilog->zl_lock);
1464 return;
1465 }
1466
1467 if (zilog->zl_suspend) {
1468 lwb = NULL;
1469 } else {
1470 lwb = list_tail(&zilog->zl_lwb_list);
1471 if (lwb == NULL)
1472 lwb = zil_create(zilog);
1473 }
1474
1475 DTRACE_PROBE1(zil__cw1, zilog_t *, zilog);
1476 while (itx = list_head(&zilog->zl_itx_commit_list)) {
1477 txg = itx->itx_lr.lrc_txg;
1478 ASSERT(txg);
1479
1480 if (txg > spa_last_synced_txg(spa) || txg > spa_freeze_txg(spa))
1481 lwb = zil_lwb_commit(zilog, itx, lwb);
1482 list_remove(&zilog->zl_itx_commit_list, itx);
1483 kmem_free(itx, offsetof(itx_t, itx_lr)
1484 + itx->itx_lr.lrc_reclen);
1485 }
1486 DTRACE_PROBE1(zil__cw2, zilog_t *, zilog);
1487
1488 /* write the last block out */
1489 if (lwb != NULL && lwb->lwb_zio != NULL)
1490 lwb = zil_lwb_write_start(zilog, lwb);
1491
1492 zilog->zl_cur_used = 0;
1493
1494 /*
1495 * Wait if necessary for the log blocks to be on stable storage.
1496 */
1497 if (zilog->zl_root_zio) {
1498 error = zio_wait(zilog->zl_root_zio);
1499 zilog->zl_root_zio = NULL;
1500 zil_flush_vdevs(zilog);
1501 }
1502
1503 if (error || lwb == NULL)
1504 txg_wait_synced(zilog->zl_dmu_pool, 0);
1505
1506 mutex_enter(&zilog->zl_lock);
1507
1508 /*
1509 * Remember the highest committed log sequence number for ztest.
1510 * We only update this value when all the log writes succeeded,
1511 * because ztest wants to ASSERT that it got the whole log chain.
1512 */
1513 if (error == 0 && lwb != NULL)
1514 zilog->zl_commit_lr_seq = zilog->zl_lr_seq;
1515 }
1516
1517 /*
1518 * Commit zfs transactions to stable storage.
1519 * If foid is 0 push out all transactions, otherwise push only those
1520 * for that object or might reference that object.
1521 *
1522 * itxs are committed in batches. In a heavily stressed zil there will be
1523 * a commit writer thread who is writing out a bunch of itxs to the log
1524 * for a set of committing threads (cthreads) in the same batch as the writer.
1525 * Those cthreads are all waiting on the same cv for that batch.
1526 *
1527 * There will also be a different and growing batch of threads that are
1528 * waiting to commit (qthreads). When the committing batch completes
1529 * a transition occurs such that the cthreads exit and the qthreads become
1530 * cthreads. One of the new cthreads becomes the writer thread for the
1531 * batch. Any new threads arriving become new qthreads.
1532 *
1533 * Only 2 condition variables are needed and there's no transition
1534 * between the two cvs needed. They just flip-flop between qthreads
1535 * and cthreads.
1536 *
1537 * Using this scheme we can efficiently wakeup up only those threads
1538 * that have been committed.
1539 */
1540 void
1541 zil_commit(zilog_t *zilog, uint64_t foid)
1542 {
1543 uint64_t mybatch;
1544
1545 if (zilog->zl_sync == ZFS_SYNC_DISABLED)
1546 return;
1547
1548 /* move the async itxs for the foid to the sync queues */
1549 zil_async_to_sync(zilog, foid);
1550
1551 mutex_enter(&zilog->zl_lock);
1552 mybatch = zilog->zl_next_batch;
1553 while (zilog->zl_writer) {
1554 cv_wait(&zilog->zl_cv_batch[mybatch & 1], &zilog->zl_lock);
1555 if (mybatch <= zilog->zl_com_batch) {
1556 mutex_exit(&zilog->zl_lock);
1557 return;
1558 }
1559 }
1560
1561 zilog->zl_next_batch++;
1562 zilog->zl_writer = B_TRUE;
1563 zil_commit_writer(zilog);
1564 zilog->zl_com_batch = mybatch;
1565 zilog->zl_writer = B_FALSE;
1566 mutex_exit(&zilog->zl_lock);
1567
1568 /* wake up one thread to become the next writer */
1569 cv_signal(&zilog->zl_cv_batch[(mybatch+1) & 1]);
1570
1571 /* wake up all threads waiting for this batch to be committed */
1572 cv_broadcast(&zilog->zl_cv_batch[mybatch & 1]);
1573 }
1574
1575 /*
1576 * Called in syncing context to free committed log blocks and update log header.
1577 */
1578 void
1579 zil_sync(zilog_t *zilog, dmu_tx_t *tx)
1580 {
1581 zil_header_t *zh = zil_header_in_syncing_context(zilog);
1582 uint64_t txg = dmu_tx_get_txg(tx);
1583 spa_t *spa = zilog->zl_spa;
1584 uint64_t *replayed_seq = &zilog->zl_replayed_seq[txg & TXG_MASK];
1585 lwb_t *lwb;
1586
1587 /*
1588 * We don't zero out zl_destroy_txg, so make sure we don't try
1589 * to destroy it twice.
1590 */
1591 if (spa_sync_pass(spa) != 1)
1592 return;
1593
1594 mutex_enter(&zilog->zl_lock);
1595
1596 ASSERT(zilog->zl_stop_sync == 0);
1597
1598 if (*replayed_seq != 0) {
1599 ASSERT(zh->zh_replay_seq < *replayed_seq);
1600 zh->zh_replay_seq = *replayed_seq;
1601 *replayed_seq = 0;
1602 }
1603
1604 if (zilog->zl_destroy_txg == txg) {
1605 blkptr_t blk = zh->zh_log;
1606
1607 ASSERT(list_head(&zilog->zl_lwb_list) == NULL);
1608
1609 bzero(zh, sizeof (zil_header_t));
1610 bzero(zilog->zl_replayed_seq, sizeof (zilog->zl_replayed_seq));
1611
1612 if (zilog->zl_keep_first) {
1613 /*
1614 * If this block was part of log chain that couldn't
1615 * be claimed because a device was missing during
1616 * zil_claim(), but that device later returns,
1617 * then this block could erroneously appear valid.
1618 * To guard against this, assign a new GUID to the new
1619 * log chain so it doesn't matter what blk points to.
1620 */
1621 zil_init_log_chain(zilog, &blk);
1622 zh->zh_log = blk;
1623 }
1624 }
1625
1626 while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) {
1627 zh->zh_log = lwb->lwb_blk;
1628 if (lwb->lwb_buf != NULL || lwb->lwb_max_txg > txg)
1629 break;
1630 list_remove(&zilog->zl_lwb_list, lwb);
1631 zio_free_zil(spa, txg, &lwb->lwb_blk);
1632 kmem_cache_free(zil_lwb_cache, lwb);
1633
1634 /*
1635 * If we don't have anything left in the lwb list then
1636 * we've had an allocation failure and we need to zero
1637 * out the zil_header blkptr so that we don't end
1638 * up freeing the same block twice.
1639 */
1640 if (list_head(&zilog->zl_lwb_list) == NULL)
1641 BP_ZERO(&zh->zh_log);
1642 }
1643 mutex_exit(&zilog->zl_lock);
1644 }
1645
1646 void
1647 zil_init(void)
1648 {
1649 zil_lwb_cache = kmem_cache_create("zil_lwb_cache",
1650 sizeof (struct lwb), 0, NULL, NULL, NULL, NULL, NULL, 0);
1651 }
1652
1653 void
1654 zil_fini(void)
1655 {
1656 kmem_cache_destroy(zil_lwb_cache);
1657 }
1658
1659 void
1660 zil_set_sync(zilog_t *zilog, uint64_t sync)
1661 {
1662 zilog->zl_sync = sync;
1663 }
1664
1665 void
1666 zil_set_logbias(zilog_t *zilog, uint64_t logbias)
1667 {
1668 zilog->zl_logbias = logbias;
1669 }
1670
1671 zilog_t *
1672 zil_alloc(objset_t *os, zil_header_t *zh_phys)
1673 {
1674 zilog_t *zilog;
1675
1676 zilog = kmem_zalloc(sizeof (zilog_t), KM_SLEEP);
1677
1678 zilog->zl_header = zh_phys;
1679 zilog->zl_os = os;
1680 zilog->zl_spa = dmu_objset_spa(os);
1681 zilog->zl_dmu_pool = dmu_objset_pool(os);
1682 zilog->zl_destroy_txg = TXG_INITIAL - 1;
1683 zilog->zl_logbias = dmu_objset_logbias(os);
1684 zilog->zl_sync = dmu_objset_syncprop(os);
1685 zilog->zl_next_batch = 1;
1686
1687 mutex_init(&zilog->zl_lock, NULL, MUTEX_DEFAULT, NULL);
1688
1689 for (int i = 0; i < TXG_SIZE; i++) {
1690 mutex_init(&zilog->zl_itxg[i].itxg_lock, NULL,
1691 MUTEX_DEFAULT, NULL);
1692 }
1693
1694 list_create(&zilog->zl_lwb_list, sizeof (lwb_t),
1695 offsetof(lwb_t, lwb_node));
1696
1697 list_create(&zilog->zl_itx_commit_list, sizeof (itx_t),
1698 offsetof(itx_t, itx_node));
1699
1700 mutex_init(&zilog->zl_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
1701
1702 avl_create(&zilog->zl_vdev_tree, zil_vdev_compare,
1703 sizeof (zil_vdev_node_t), offsetof(zil_vdev_node_t, zv_node));
1704
1705 cv_init(&zilog->zl_cv_writer, NULL, CV_DEFAULT, NULL);
1706 cv_init(&zilog->zl_cv_suspend, NULL, CV_DEFAULT, NULL);
1707 cv_init(&zilog->zl_cv_batch[0], NULL, CV_DEFAULT, NULL);
1708 cv_init(&zilog->zl_cv_batch[1], NULL, CV_DEFAULT, NULL);
1709
1710 return (zilog);
1711 }
1712
1713 void
1714 zil_free(zilog_t *zilog)
1715 {
1716 zilog->zl_stop_sync = 1;
1717
1718 ASSERT0(zilog->zl_suspend);
1719 ASSERT0(zilog->zl_suspending);
1720
1721 ASSERT(list_is_empty(&zilog->zl_lwb_list));
1722 list_destroy(&zilog->zl_lwb_list);
1723
1724 avl_destroy(&zilog->zl_vdev_tree);
1725 mutex_destroy(&zilog->zl_vdev_lock);
1726
1727 ASSERT(list_is_empty(&zilog->zl_itx_commit_list));
1728 list_destroy(&zilog->zl_itx_commit_list);
1729
1730 for (int i = 0; i < TXG_SIZE; i++) {
1731 /*
1732 * It's possible for an itx to be generated that doesn't dirty
1733 * a txg (e.g. ztest TX_TRUNCATE). So there's no zil_clean()
1734 * callback to remove the entry. We remove those here.
1735 *
1736 * Also free up the ziltest itxs.
1737 */
1738 if (zilog->zl_itxg[i].itxg_itxs)
1739 zil_itxg_clean(zilog->zl_itxg[i].itxg_itxs);
1740 mutex_destroy(&zilog->zl_itxg[i].itxg_lock);
1741 }
1742
1743 mutex_destroy(&zilog->zl_lock);
1744
1745 cv_destroy(&zilog->zl_cv_writer);
1746 cv_destroy(&zilog->zl_cv_suspend);
1747 cv_destroy(&zilog->zl_cv_batch[0]);
1748 cv_destroy(&zilog->zl_cv_batch[1]);
1749
1750 kmem_free(zilog, sizeof (zilog_t));
1751 }
1752
1753 /*
1754 * Open an intent log.
1755 */
1756 zilog_t *
1757 zil_open(objset_t *os, zil_get_data_t *get_data)
1758 {
1759 zilog_t *zilog = dmu_objset_zil(os);
1760
1761 ASSERT(zilog->zl_clean_taskq == NULL);
1762 ASSERT(zilog->zl_get_data == NULL);
1763 ASSERT(list_is_empty(&zilog->zl_lwb_list));
1764
1765 zilog->zl_get_data = get_data;
1766 zilog->zl_clean_taskq = taskq_create("zil_clean", 1, minclsyspri,
1767 2, 2, TASKQ_PREPOPULATE);
1768
1769 return (zilog);
1770 }
1771
1772 /*
1773 * Close an intent log.
1774 */
1775 void
1776 zil_close(zilog_t *zilog)
1777 {
1778 lwb_t *lwb;
1779 uint64_t txg = 0;
1780
1781 zil_commit(zilog, 0); /* commit all itx */
1782
1783 /*
1784 * The lwb_max_txg for the stubby lwb will reflect the last activity
1785 * for the zil. After a txg_wait_synced() on the txg we know all the
1786 * callbacks have occurred that may clean the zil. Only then can we
1787 * destroy the zl_clean_taskq.
1788 */
1789 mutex_enter(&zilog->zl_lock);
1790 lwb = list_tail(&zilog->zl_lwb_list);
1791 if (lwb != NULL)
1792 txg = lwb->lwb_max_txg;
1793 mutex_exit(&zilog->zl_lock);
1794 if (txg)
1795 txg_wait_synced(zilog->zl_dmu_pool, txg);
1796 ASSERT(!zilog_is_dirty(zilog));
1797
1798 taskq_destroy(zilog->zl_clean_taskq);
1799 zilog->zl_clean_taskq = NULL;
1800 zilog->zl_get_data = NULL;
1801
1802 /*
1803 * We should have only one LWB left on the list; remove it now.
1804 */
1805 mutex_enter(&zilog->zl_lock);
1806 lwb = list_head(&zilog->zl_lwb_list);
1807 if (lwb != NULL) {
1808 ASSERT(lwb == list_tail(&zilog->zl_lwb_list));
1809 list_remove(&zilog->zl_lwb_list, lwb);
1810 zio_buf_free(lwb->lwb_buf, lwb->lwb_sz);
1811 kmem_cache_free(zil_lwb_cache, lwb);
1812 }
1813 mutex_exit(&zilog->zl_lock);
1814 }
1815
1816 static char *suspend_tag = "zil suspending";
1817
1818 /*
1819 * Suspend an intent log. While in suspended mode, we still honor
1820 * synchronous semantics, but we rely on txg_wait_synced() to do it.
1821 * On old version pools, we suspend the log briefly when taking a
1822 * snapshot so that it will have an empty intent log.
1823 *
1824 * Long holds are not really intended to be used the way we do here --
1825 * held for such a short time. A concurrent caller of dsl_dataset_long_held()
1826 * could fail. Therefore we take pains to only put a long hold if it is
1827 * actually necessary. Fortunately, it will only be necessary if the
1828 * objset is currently mounted (or the ZVOL equivalent). In that case it
1829 * will already have a long hold, so we are not really making things any worse.
1830 *
1831 * Ideally, we would locate the existing long-holder (i.e. the zfsvfs_t or
1832 * zvol_state_t), and use their mechanism to prevent their hold from being
1833 * dropped (e.g. VFS_HOLD()). However, that would be even more pain for
1834 * very little gain.
1835 *
1836 * if cookiep == NULL, this does both the suspend & resume.
1837 * Otherwise, it returns with the dataset "long held", and the cookie
1838 * should be passed into zil_resume().
1839 */
1840 int
1841 zil_suspend(const char *osname, void **cookiep)
1842 {
1843 objset_t *os;
1844 zilog_t *zilog;
1845 const zil_header_t *zh;
1846 int error;
1847
1848 error = dmu_objset_hold(osname, suspend_tag, &os);
1849 if (error != 0)
1850 return (error);
1851 zilog = dmu_objset_zil(os);
1852
1853 mutex_enter(&zilog->zl_lock);
1854 zh = zilog->zl_header;
1855
1856 if (zh->zh_flags & ZIL_REPLAY_NEEDED) { /* unplayed log */
1857 mutex_exit(&zilog->zl_lock);
1858 dmu_objset_rele(os, suspend_tag);
1859 return (SET_ERROR(EBUSY));
1860 }
1861
1862 /*
1863 * Don't put a long hold in the cases where we can avoid it. This
1864 * is when there is no cookie so we are doing a suspend & resume
1865 * (i.e. called from zil_vdev_offline()), and there's nothing to do
1866 * for the suspend because it's already suspended, or there's no ZIL.
1867 */
1868 if (cookiep == NULL && !zilog->zl_suspending &&
1869 (zilog->zl_suspend > 0 || BP_IS_HOLE(&zh->zh_log))) {
1870 mutex_exit(&zilog->zl_lock);
1871 dmu_objset_rele(os, suspend_tag);
1872 return (0);
1873 }
1874
1875 dsl_dataset_long_hold(dmu_objset_ds(os), suspend_tag);
1876 dsl_pool_rele(dmu_objset_pool(os), suspend_tag);
1877
1878 zilog->zl_suspend++;
1879
1880 if (zilog->zl_suspend > 1) {
1881 /*
1882 * Someone else is already suspending it.
1883 * Just wait for them to finish.
1884 */
1885
1886 while (zilog->zl_suspending)
1887 cv_wait(&zilog->zl_cv_suspend, &zilog->zl_lock);
1888 mutex_exit(&zilog->zl_lock);
1889
1890 if (cookiep == NULL)
1891 zil_resume(os);
1892 else
1893 *cookiep = os;
1894 return (0);
1895 }
1896
1897 /*
1898 * If there is no pointer to an on-disk block, this ZIL must not
1899 * be active (e.g. filesystem not mounted), so there's nothing
1900 * to clean up.
1901 */
1902 if (BP_IS_HOLE(&zh->zh_log)) {
1903 ASSERT(cookiep != NULL); /* fast path already handled */
1904
1905 *cookiep = os;
1906 mutex_exit(&zilog->zl_lock);
1907 return (0);
1908 }
1909
1910 zilog->zl_suspending = B_TRUE;
1911 mutex_exit(&zilog->zl_lock);
1912
1913 zil_commit(zilog, 0);
1914
1915 zil_destroy(zilog, B_FALSE);
1916
1917 mutex_enter(&zilog->zl_lock);
1918 zilog->zl_suspending = B_FALSE;
1919 cv_broadcast(&zilog->zl_cv_suspend);
1920 mutex_exit(&zilog->zl_lock);
1921
1922 if (cookiep == NULL)
1923 zil_resume(os);
1924 else
1925 *cookiep = os;
1926 return (0);
1927 }
1928
1929 void
1930 zil_resume(void *cookie)
1931 {
1932 objset_t *os = cookie;
1933 zilog_t *zilog = dmu_objset_zil(os);
1934
1935 mutex_enter(&zilog->zl_lock);
1936 ASSERT(zilog->zl_suspend != 0);
1937 zilog->zl_suspend--;
1938 mutex_exit(&zilog->zl_lock);
1939 dsl_dataset_long_rele(dmu_objset_ds(os), suspend_tag);
1940 dsl_dataset_rele(dmu_objset_ds(os), suspend_tag);
1941 }
1942
1943 typedef struct zil_replay_arg {
1944 zil_replay_func_t **zr_replay;
1945 void *zr_arg;
1946 boolean_t zr_byteswap;
1947 char *zr_lr;
1948 } zil_replay_arg_t;
1949
1950 static int
1951 zil_replay_error(zilog_t *zilog, lr_t *lr, int error)
1952 {
1953 char name[MAXNAMELEN];
1954
1955 zilog->zl_replaying_seq--; /* didn't actually replay this one */
1956
1957 dmu_objset_name(zilog->zl_os, name);
1958
1959 cmn_err(CE_WARN, "ZFS replay transaction error %d, "
1960 "dataset %s, seq 0x%llx, txtype %llu %s\n", error, name,
1961 (u_longlong_t)lr->lrc_seq,
1962 (u_longlong_t)(lr->lrc_txtype & ~TX_CI),
1963 (lr->lrc_txtype & TX_CI) ? "CI" : "");
1964
1965 return (error);
1966 }
1967
1968 static int
1969 zil_replay_log_record(zilog_t *zilog, lr_t *lr, void *zra, uint64_t claim_txg)
1970 {
1971 zil_replay_arg_t *zr = zra;
1972 const zil_header_t *zh = zilog->zl_header;
1973 uint64_t reclen = lr->lrc_reclen;
1974 uint64_t txtype = lr->lrc_txtype;
1975 int error = 0;
1976
1977 zilog->zl_replaying_seq = lr->lrc_seq;
1978
1979 if (lr->lrc_seq <= zh->zh_replay_seq) /* already replayed */
1980 return (0);
1981
1982 if (lr->lrc_txg < claim_txg) /* already committed */
1983 return (0);
1984
1985 /* Strip case-insensitive bit, still present in log record */
1986 txtype &= ~TX_CI;
1987
1988 if (txtype == 0 || txtype >= TX_MAX_TYPE)
1989 return (zil_replay_error(zilog, lr, EINVAL));
1990
1991 /*
1992 * If this record type can be logged out of order, the object
1993 * (lr_foid) may no longer exist. That's legitimate, not an error.
1994 */
1995 if (TX_OOO(txtype)) {
1996 error = dmu_object_info(zilog->zl_os,
1997 ((lr_ooo_t *)lr)->lr_foid, NULL);
1998 if (error == ENOENT || error == EEXIST)
1999 return (0);
2000 }
2001
2002 /*
2003 * Make a copy of the data so we can revise and extend it.
2004 */
2005 bcopy(lr, zr->zr_lr, reclen);
2006
2007 /*
2008 * If this is a TX_WRITE with a blkptr, suck in the data.
2009 */
2010 if (txtype == TX_WRITE && reclen == sizeof (lr_write_t)) {
2011 error = zil_read_log_data(zilog, (lr_write_t *)lr,
2012 zr->zr_lr + reclen);
2013 if (error != 0)
2014 return (zil_replay_error(zilog, lr, error));
2015 }
2016
2017 /*
2018 * The log block containing this lr may have been byteswapped
2019 * so that we can easily examine common fields like lrc_txtype.
2020 * However, the log is a mix of different record types, and only the
2021 * replay vectors know how to byteswap their records. Therefore, if
2022 * the lr was byteswapped, undo it before invoking the replay vector.
2023 */
2024 if (zr->zr_byteswap)
2025 byteswap_uint64_array(zr->zr_lr, reclen);
2026
2027 /*
2028 * We must now do two things atomically: replay this log record,
2029 * and update the log header sequence number to reflect the fact that
2030 * we did so. At the end of each replay function the sequence number
2031 * is updated if we are in replay mode.
2032 */
2033 error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lr, zr->zr_byteswap);
2034 if (error != 0) {
2035 /*
2036 * The DMU's dnode layer doesn't see removes until the txg
2037 * commits, so a subsequent claim can spuriously fail with
2038 * EEXIST. So if we receive any error we try syncing out
2039 * any removes then retry the transaction. Note that we
2040 * specify B_FALSE for byteswap now, so we don't do it twice.
2041 */
2042 txg_wait_synced(spa_get_dsl(zilog->zl_spa), 0);
2043 error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lr, B_FALSE);
2044 if (error != 0)
2045 return (zil_replay_error(zilog, lr, error));
2046 }
2047 return (0);
2048 }
2049
2050 /* ARGSUSED */
2051 static int
2052 zil_incr_blks(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg)
2053 {
2054 zilog->zl_replay_blks++;
2055
2056 return (0);
2057 }
2058
2059 /*
2060 * If this dataset has a non-empty intent log, replay it and destroy it.
2061 */
2062 void
2063 zil_replay(objset_t *os, void *arg, zil_replay_func_t *replay_func[TX_MAX_TYPE])
2064 {
2065 zilog_t *zilog = dmu_objset_zil(os);
2066 const zil_header_t *zh = zilog->zl_header;
2067 zil_replay_arg_t zr;
2068
2069 if ((zh->zh_flags & ZIL_REPLAY_NEEDED) == 0) {
2070 zil_destroy(zilog, B_TRUE);
2071 return;
2072 }
2073
2074 zr.zr_replay = replay_func;
2075 zr.zr_arg = arg;
2076 zr.zr_byteswap = BP_SHOULD_BYTESWAP(&zh->zh_log);
2077 zr.zr_lr = kmem_alloc(2 * SPA_MAXBLOCKSIZE, KM_SLEEP);
2078
2079 /*
2080 * Wait for in-progress removes to sync before starting replay.
2081 */
2082 txg_wait_synced(zilog->zl_dmu_pool, 0);
2083
2084 zilog->zl_replay = B_TRUE;
2085 zilog->zl_replay_time = ddi_get_lbolt();
2086 ASSERT(zilog->zl_replay_blks == 0);
2087 (void) zil_parse(zilog, zil_incr_blks, zil_replay_log_record, &zr,
2088 zh->zh_claim_txg);
2089 kmem_free(zr.zr_lr, 2 * SPA_MAXBLOCKSIZE);
2090
2091 zil_destroy(zilog, B_FALSE);
2092 txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg);
2093 zilog->zl_replay = B_FALSE;
2094 }
2095
2096 boolean_t
2097 zil_replaying(zilog_t *zilog, dmu_tx_t *tx)
2098 {
2099 if (zilog->zl_sync == ZFS_SYNC_DISABLED)
2100 return (B_TRUE);
2101
2102 if (zilog->zl_replay) {
2103 dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
2104 zilog->zl_replayed_seq[dmu_tx_get_txg(tx) & TXG_MASK] =
2105 zilog->zl_replaying_seq;
2106 return (B_TRUE);
2107 }
2108
2109 return (B_FALSE);
2110 }
2111
2112 /* ARGSUSED */
2113 int
2114 zil_vdev_offline(const char *osname, void *arg)
2115 {
2116 int error;
2117
2118 error = zil_suspend(osname, NULL);
2119 if (error != 0)
2120 return (SET_ERROR(EEXIST));
2121 return (0);
2122 }