Print this page
3006 VERIFY[S,U,P] and ASSERT[S,U,P] frequently check if first argument is zero
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/zil.c
+++ new/usr/src/uts/common/fs/zfs/zil.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
|
↓ open down ↓ |
12 lines elided |
↑ open up ↑ |
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 - * Copyright (c) 2011 by Delphix. All rights reserved.
23 + * Copyright (c) 2012 by Delphix. All rights reserved.
24 24 */
25 25
26 26 /* Portions Copyright 2010 Robert Milkowski */
27 27
28 28 #include <sys/zfs_context.h>
29 29 #include <sys/spa.h>
30 30 #include <sys/dmu.h>
31 31 #include <sys/zap.h>
32 32 #include <sys/arc.h>
33 33 #include <sys/stat.h>
34 34 #include <sys/resource.h>
35 35 #include <sys/zil.h>
36 36 #include <sys/zil_impl.h>
37 37 #include <sys/dsl_dataset.h>
38 38 #include <sys/vdev_impl.h>
39 39 #include <sys/dmu_tx.h>
40 40 #include <sys/dsl_pool.h>
41 41
42 42 /*
43 43 * The zfs intent log (ZIL) saves transaction records of system calls
44 44 * that change the file system in memory with enough information
45 45 * to be able to replay them. These are stored in memory until
46 46 * either the DMU transaction group (txg) commits them to the stable pool
47 47 * and they can be discarded, or they are flushed to the stable log
48 48 * (also in the pool) due to a fsync, O_DSYNC or other synchronous
49 49 * requirement. In the event of a panic or power fail then those log
50 50 * records (transactions) are replayed.
51 51 *
52 52 * There is one ZIL per file system. Its on-disk (pool) format consists
53 53 * of 3 parts:
54 54 *
55 55 * - ZIL header
56 56 * - ZIL blocks
57 57 * - ZIL records
58 58 *
59 59 * A log record holds a system call transaction. Log blocks can
60 60 * hold many log records and the blocks are chained together.
61 61 * Each ZIL block contains a block pointer (blkptr_t) to the next
62 62 * ZIL block in the chain. The ZIL header points to the first
63 63 * block in the chain. Note there is not a fixed place in the pool
64 64 * to hold blocks. They are dynamically allocated and freed as
65 65 * needed from the blocks available. Figure X shows the ZIL structure:
66 66 */
67 67
68 68 /*
69 69 * This global ZIL switch affects all pools
70 70 */
71 71 int zil_replay_disable = 0; /* disable intent logging replay */
72 72
73 73 /*
74 74 * Tunable parameter for debugging or performance analysis. Setting
75 75 * zfs_nocacheflush will cause corruption on power loss if a volatile
76 76 * out-of-order write cache is enabled.
77 77 */
78 78 boolean_t zfs_nocacheflush = B_FALSE;
79 79
80 80 static kmem_cache_t *zil_lwb_cache;
81 81
82 82 static void zil_async_to_sync(zilog_t *zilog, uint64_t foid);
83 83
84 84 #define LWB_EMPTY(lwb) ((BP_GET_LSIZE(&lwb->lwb_blk) - \
85 85 sizeof (zil_chain_t)) == (lwb->lwb_sz - lwb->lwb_nused))
86 86
87 87
88 88 /*
89 89 * ziltest is by and large an ugly hack, but very useful in
90 90 * checking replay without tedious work.
91 91 * When running ziltest we want to keep all itx's and so maintain
92 92 * a single list in the zl_itxg[] that uses a high txg: ZILTEST_TXG
93 93 * We subtract TXG_CONCURRENT_STATES to allow for common code.
94 94 */
95 95 #define ZILTEST_TXG (UINT64_MAX - TXG_CONCURRENT_STATES)
96 96
97 97 static int
98 98 zil_bp_compare(const void *x1, const void *x2)
99 99 {
100 100 const dva_t *dva1 = &((zil_bp_node_t *)x1)->zn_dva;
101 101 const dva_t *dva2 = &((zil_bp_node_t *)x2)->zn_dva;
102 102
103 103 if (DVA_GET_VDEV(dva1) < DVA_GET_VDEV(dva2))
104 104 return (-1);
105 105 if (DVA_GET_VDEV(dva1) > DVA_GET_VDEV(dva2))
106 106 return (1);
107 107
108 108 if (DVA_GET_OFFSET(dva1) < DVA_GET_OFFSET(dva2))
109 109 return (-1);
110 110 if (DVA_GET_OFFSET(dva1) > DVA_GET_OFFSET(dva2))
111 111 return (1);
112 112
113 113 return (0);
114 114 }
115 115
116 116 static void
117 117 zil_bp_tree_init(zilog_t *zilog)
118 118 {
119 119 avl_create(&zilog->zl_bp_tree, zil_bp_compare,
120 120 sizeof (zil_bp_node_t), offsetof(zil_bp_node_t, zn_node));
121 121 }
122 122
123 123 static void
124 124 zil_bp_tree_fini(zilog_t *zilog)
125 125 {
126 126 avl_tree_t *t = &zilog->zl_bp_tree;
127 127 zil_bp_node_t *zn;
128 128 void *cookie = NULL;
129 129
130 130 while ((zn = avl_destroy_nodes(t, &cookie)) != NULL)
131 131 kmem_free(zn, sizeof (zil_bp_node_t));
132 132
133 133 avl_destroy(t);
134 134 }
135 135
136 136 int
137 137 zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp)
138 138 {
139 139 avl_tree_t *t = &zilog->zl_bp_tree;
140 140 const dva_t *dva = BP_IDENTITY(bp);
141 141 zil_bp_node_t *zn;
142 142 avl_index_t where;
143 143
144 144 if (avl_find(t, dva, &where) != NULL)
145 145 return (EEXIST);
146 146
147 147 zn = kmem_alloc(sizeof (zil_bp_node_t), KM_SLEEP);
148 148 zn->zn_dva = *dva;
149 149 avl_insert(t, zn, where);
150 150
151 151 return (0);
152 152 }
153 153
154 154 static zil_header_t *
155 155 zil_header_in_syncing_context(zilog_t *zilog)
156 156 {
157 157 return ((zil_header_t *)zilog->zl_header);
158 158 }
159 159
160 160 static void
161 161 zil_init_log_chain(zilog_t *zilog, blkptr_t *bp)
162 162 {
163 163 zio_cksum_t *zc = &bp->blk_cksum;
164 164
165 165 zc->zc_word[ZIL_ZC_GUID_0] = spa_get_random(-1ULL);
166 166 zc->zc_word[ZIL_ZC_GUID_1] = spa_get_random(-1ULL);
167 167 zc->zc_word[ZIL_ZC_OBJSET] = dmu_objset_id(zilog->zl_os);
168 168 zc->zc_word[ZIL_ZC_SEQ] = 1ULL;
169 169 }
170 170
171 171 /*
172 172 * Read a log block and make sure it's valid.
173 173 */
174 174 static int
175 175 zil_read_log_block(zilog_t *zilog, const blkptr_t *bp, blkptr_t *nbp, void *dst,
176 176 char **end)
177 177 {
178 178 enum zio_flag zio_flags = ZIO_FLAG_CANFAIL;
179 179 uint32_t aflags = ARC_WAIT;
180 180 arc_buf_t *abuf = NULL;
181 181 zbookmark_t zb;
182 182 int error;
183 183
184 184 if (zilog->zl_header->zh_claim_txg == 0)
185 185 zio_flags |= ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB;
186 186
187 187 if (!(zilog->zl_header->zh_flags & ZIL_CLAIM_LR_SEQ_VALID))
188 188 zio_flags |= ZIO_FLAG_SPECULATIVE;
189 189
190 190 SET_BOOKMARK(&zb, bp->blk_cksum.zc_word[ZIL_ZC_OBJSET],
191 191 ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, bp->blk_cksum.zc_word[ZIL_ZC_SEQ]);
192 192
193 193 error = dsl_read_nolock(NULL, zilog->zl_spa, bp, arc_getbuf_func, &abuf,
194 194 ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb);
195 195
196 196 if (error == 0) {
197 197 zio_cksum_t cksum = bp->blk_cksum;
198 198
199 199 /*
200 200 * Validate the checksummed log block.
201 201 *
202 202 * Sequence numbers should be... sequential. The checksum
203 203 * verifier for the next block should be bp's checksum plus 1.
204 204 *
205 205 * Also check the log chain linkage and size used.
206 206 */
207 207 cksum.zc_word[ZIL_ZC_SEQ]++;
208 208
209 209 if (BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_ZILOG2) {
210 210 zil_chain_t *zilc = abuf->b_data;
211 211 char *lr = (char *)(zilc + 1);
212 212 uint64_t len = zilc->zc_nused - sizeof (zil_chain_t);
213 213
214 214 if (bcmp(&cksum, &zilc->zc_next_blk.blk_cksum,
215 215 sizeof (cksum)) || BP_IS_HOLE(&zilc->zc_next_blk)) {
216 216 error = ECKSUM;
217 217 } else {
218 218 bcopy(lr, dst, len);
219 219 *end = (char *)dst + len;
220 220 *nbp = zilc->zc_next_blk;
221 221 }
222 222 } else {
223 223 char *lr = abuf->b_data;
224 224 uint64_t size = BP_GET_LSIZE(bp);
225 225 zil_chain_t *zilc = (zil_chain_t *)(lr + size) - 1;
226 226
227 227 if (bcmp(&cksum, &zilc->zc_next_blk.blk_cksum,
228 228 sizeof (cksum)) || BP_IS_HOLE(&zilc->zc_next_blk) ||
229 229 (zilc->zc_nused > (size - sizeof (*zilc)))) {
230 230 error = ECKSUM;
231 231 } else {
232 232 bcopy(lr, dst, zilc->zc_nused);
233 233 *end = (char *)dst + zilc->zc_nused;
234 234 *nbp = zilc->zc_next_blk;
235 235 }
236 236 }
237 237
238 238 VERIFY(arc_buf_remove_ref(abuf, &abuf) == 1);
239 239 }
240 240
241 241 return (error);
242 242 }
243 243
244 244 /*
245 245 * Read a TX_WRITE log data block.
246 246 */
247 247 static int
248 248 zil_read_log_data(zilog_t *zilog, const lr_write_t *lr, void *wbuf)
249 249 {
250 250 enum zio_flag zio_flags = ZIO_FLAG_CANFAIL;
251 251 const blkptr_t *bp = &lr->lr_blkptr;
252 252 uint32_t aflags = ARC_WAIT;
253 253 arc_buf_t *abuf = NULL;
254 254 zbookmark_t zb;
255 255 int error;
256 256
257 257 if (BP_IS_HOLE(bp)) {
258 258 if (wbuf != NULL)
259 259 bzero(wbuf, MAX(BP_GET_LSIZE(bp), lr->lr_length));
260 260 return (0);
261 261 }
262 262
263 263 if (zilog->zl_header->zh_claim_txg == 0)
264 264 zio_flags |= ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB;
265 265
266 266 SET_BOOKMARK(&zb, dmu_objset_id(zilog->zl_os), lr->lr_foid,
267 267 ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp));
268 268
269 269 error = arc_read_nolock(NULL, zilog->zl_spa, bp, arc_getbuf_func, &abuf,
270 270 ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb);
271 271
272 272 if (error == 0) {
273 273 if (wbuf != NULL)
274 274 bcopy(abuf->b_data, wbuf, arc_buf_size(abuf));
275 275 (void) arc_buf_remove_ref(abuf, &abuf);
276 276 }
277 277
278 278 return (error);
279 279 }
280 280
281 281 /*
282 282 * Parse the intent log, and call parse_func for each valid record within.
283 283 */
284 284 int
285 285 zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func,
286 286 zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg)
287 287 {
288 288 const zil_header_t *zh = zilog->zl_header;
289 289 boolean_t claimed = !!zh->zh_claim_txg;
290 290 uint64_t claim_blk_seq = claimed ? zh->zh_claim_blk_seq : UINT64_MAX;
291 291 uint64_t claim_lr_seq = claimed ? zh->zh_claim_lr_seq : UINT64_MAX;
292 292 uint64_t max_blk_seq = 0;
293 293 uint64_t max_lr_seq = 0;
294 294 uint64_t blk_count = 0;
295 295 uint64_t lr_count = 0;
296 296 blkptr_t blk, next_blk;
297 297 char *lrbuf, *lrp;
298 298 int error = 0;
299 299
300 300 /*
301 301 * Old logs didn't record the maximum zh_claim_lr_seq.
302 302 */
303 303 if (!(zh->zh_flags & ZIL_CLAIM_LR_SEQ_VALID))
304 304 claim_lr_seq = UINT64_MAX;
305 305
306 306 /*
307 307 * Starting at the block pointed to by zh_log we read the log chain.
308 308 * For each block in the chain we strongly check that block to
309 309 * ensure its validity. We stop when an invalid block is found.
310 310 * For each block pointer in the chain we call parse_blk_func().
311 311 * For each record in each valid block we call parse_lr_func().
312 312 * If the log has been claimed, stop if we encounter a sequence
313 313 * number greater than the highest claimed sequence number.
314 314 */
315 315 lrbuf = zio_buf_alloc(SPA_MAXBLOCKSIZE);
316 316 zil_bp_tree_init(zilog);
317 317
318 318 for (blk = zh->zh_log; !BP_IS_HOLE(&blk); blk = next_blk) {
319 319 uint64_t blk_seq = blk.blk_cksum.zc_word[ZIL_ZC_SEQ];
320 320 int reclen;
321 321 char *end;
322 322
323 323 if (blk_seq > claim_blk_seq)
324 324 break;
325 325 if ((error = parse_blk_func(zilog, &blk, arg, txg)) != 0)
326 326 break;
327 327 ASSERT3U(max_blk_seq, <, blk_seq);
328 328 max_blk_seq = blk_seq;
329 329 blk_count++;
330 330
331 331 if (max_lr_seq == claim_lr_seq && max_blk_seq == claim_blk_seq)
332 332 break;
333 333
334 334 error = zil_read_log_block(zilog, &blk, &next_blk, lrbuf, &end);
335 335 if (error)
336 336 break;
337 337
338 338 for (lrp = lrbuf; lrp < end; lrp += reclen) {
339 339 lr_t *lr = (lr_t *)lrp;
340 340 reclen = lr->lrc_reclen;
341 341 ASSERT3U(reclen, >=, sizeof (lr_t));
342 342 if (lr->lrc_seq > claim_lr_seq)
343 343 goto done;
344 344 if ((error = parse_lr_func(zilog, lr, arg, txg)) != 0)
345 345 goto done;
346 346 ASSERT3U(max_lr_seq, <, lr->lrc_seq);
347 347 max_lr_seq = lr->lrc_seq;
348 348 lr_count++;
349 349 }
350 350 }
351 351 done:
352 352 zilog->zl_parse_error = error;
353 353 zilog->zl_parse_blk_seq = max_blk_seq;
354 354 zilog->zl_parse_lr_seq = max_lr_seq;
355 355 zilog->zl_parse_blk_count = blk_count;
356 356 zilog->zl_parse_lr_count = lr_count;
357 357
358 358 ASSERT(!claimed || !(zh->zh_flags & ZIL_CLAIM_LR_SEQ_VALID) ||
359 359 (max_blk_seq == claim_blk_seq && max_lr_seq == claim_lr_seq));
360 360
361 361 zil_bp_tree_fini(zilog);
362 362 zio_buf_free(lrbuf, SPA_MAXBLOCKSIZE);
363 363
364 364 return (error);
365 365 }
366 366
367 367 static int
368 368 zil_claim_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t first_txg)
369 369 {
370 370 /*
371 371 * Claim log block if not already committed and not already claimed.
372 372 * If tx == NULL, just verify that the block is claimable.
373 373 */
374 374 if (bp->blk_birth < first_txg || zil_bp_tree_add(zilog, bp) != 0)
375 375 return (0);
376 376
377 377 return (zio_wait(zio_claim(NULL, zilog->zl_spa,
378 378 tx == NULL ? 0 : first_txg, bp, spa_claim_notify, NULL,
379 379 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB)));
380 380 }
381 381
382 382 static int
383 383 zil_claim_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t first_txg)
384 384 {
385 385 lr_write_t *lr = (lr_write_t *)lrc;
386 386 int error;
387 387
388 388 if (lrc->lrc_txtype != TX_WRITE)
389 389 return (0);
390 390
391 391 /*
392 392 * If the block is not readable, don't claim it. This can happen
393 393 * in normal operation when a log block is written to disk before
394 394 * some of the dmu_sync() blocks it points to. In this case, the
395 395 * transaction cannot have been committed to anyone (we would have
396 396 * waited for all writes to be stable first), so it is semantically
397 397 * correct to declare this the end of the log.
398 398 */
399 399 if (lr->lr_blkptr.blk_birth >= first_txg &&
400 400 (error = zil_read_log_data(zilog, lr, NULL)) != 0)
401 401 return (error);
402 402 return (zil_claim_log_block(zilog, &lr->lr_blkptr, tx, first_txg));
403 403 }
404 404
405 405 /* ARGSUSED */
406 406 static int
407 407 zil_free_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t claim_txg)
408 408 {
409 409 zio_free_zil(zilog->zl_spa, dmu_tx_get_txg(tx), bp);
410 410
411 411 return (0);
412 412 }
413 413
414 414 static int
415 415 zil_free_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t claim_txg)
416 416 {
417 417 lr_write_t *lr = (lr_write_t *)lrc;
418 418 blkptr_t *bp = &lr->lr_blkptr;
419 419
420 420 /*
421 421 * If we previously claimed it, we need to free it.
422 422 */
423 423 if (claim_txg != 0 && lrc->lrc_txtype == TX_WRITE &&
424 424 bp->blk_birth >= claim_txg && zil_bp_tree_add(zilog, bp) == 0)
425 425 zio_free(zilog->zl_spa, dmu_tx_get_txg(tx), bp);
426 426
427 427 return (0);
428 428 }
429 429
430 430 static lwb_t *
431 431 zil_alloc_lwb(zilog_t *zilog, blkptr_t *bp, uint64_t txg)
432 432 {
433 433 lwb_t *lwb;
434 434
435 435 lwb = kmem_cache_alloc(zil_lwb_cache, KM_SLEEP);
436 436 lwb->lwb_zilog = zilog;
437 437 lwb->lwb_blk = *bp;
438 438 lwb->lwb_buf = zio_buf_alloc(BP_GET_LSIZE(bp));
439 439 lwb->lwb_max_txg = txg;
440 440 lwb->lwb_zio = NULL;
441 441 lwb->lwb_tx = NULL;
442 442 if (BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_ZILOG2) {
443 443 lwb->lwb_nused = sizeof (zil_chain_t);
444 444 lwb->lwb_sz = BP_GET_LSIZE(bp);
445 445 } else {
446 446 lwb->lwb_nused = 0;
447 447 lwb->lwb_sz = BP_GET_LSIZE(bp) - sizeof (zil_chain_t);
448 448 }
449 449
450 450 mutex_enter(&zilog->zl_lock);
451 451 list_insert_tail(&zilog->zl_lwb_list, lwb);
452 452 mutex_exit(&zilog->zl_lock);
453 453
454 454 return (lwb);
455 455 }
456 456
457 457 /*
458 458 * Create an on-disk intent log.
459 459 */
460 460 static lwb_t *
461 461 zil_create(zilog_t *zilog)
462 462 {
463 463 const zil_header_t *zh = zilog->zl_header;
464 464 lwb_t *lwb = NULL;
465 465 uint64_t txg = 0;
466 466 dmu_tx_t *tx = NULL;
467 467 blkptr_t blk;
468 468 int error = 0;
469 469
470 470 /*
471 471 * Wait for any previous destroy to complete.
472 472 */
473 473 txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg);
474 474
475 475 ASSERT(zh->zh_claim_txg == 0);
476 476 ASSERT(zh->zh_replay_seq == 0);
477 477
478 478 blk = zh->zh_log;
479 479
480 480 /*
481 481 * Allocate an initial log block if:
482 482 * - there isn't one already
483 483 * - the existing block is the wrong endianess
484 484 */
485 485 if (BP_IS_HOLE(&blk) || BP_SHOULD_BYTESWAP(&blk)) {
486 486 tx = dmu_tx_create(zilog->zl_os);
487 487 VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0);
488 488 dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
489 489 txg = dmu_tx_get_txg(tx);
490 490
491 491 if (!BP_IS_HOLE(&blk)) {
492 492 zio_free_zil(zilog->zl_spa, txg, &blk);
493 493 BP_ZERO(&blk);
494 494 }
495 495
496 496 error = zio_alloc_zil(zilog->zl_spa, txg, &blk, NULL,
497 497 ZIL_MIN_BLKSZ, zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
498 498
499 499 if (error == 0)
500 500 zil_init_log_chain(zilog, &blk);
501 501 }
502 502
503 503 /*
504 504 * Allocate a log write buffer (lwb) for the first log block.
505 505 */
506 506 if (error == 0)
507 507 lwb = zil_alloc_lwb(zilog, &blk, txg);
508 508
509 509 /*
510 510 * If we just allocated the first log block, commit our transaction
511 511 * and wait for zil_sync() to stuff the block poiner into zh_log.
512 512 * (zh is part of the MOS, so we cannot modify it in open context.)
513 513 */
514 514 if (tx != NULL) {
515 515 dmu_tx_commit(tx);
516 516 txg_wait_synced(zilog->zl_dmu_pool, txg);
517 517 }
518 518
519 519 ASSERT(bcmp(&blk, &zh->zh_log, sizeof (blk)) == 0);
520 520
521 521 return (lwb);
522 522 }
523 523
524 524 /*
525 525 * In one tx, free all log blocks and clear the log header.
526 526 * If keep_first is set, then we're replaying a log with no content.
527 527 * We want to keep the first block, however, so that the first
528 528 * synchronous transaction doesn't require a txg_wait_synced()
529 529 * in zil_create(). We don't need to txg_wait_synced() here either
530 530 * when keep_first is set, because both zil_create() and zil_destroy()
531 531 * will wait for any in-progress destroys to complete.
532 532 */
533 533 void
534 534 zil_destroy(zilog_t *zilog, boolean_t keep_first)
535 535 {
536 536 const zil_header_t *zh = zilog->zl_header;
537 537 lwb_t *lwb;
538 538 dmu_tx_t *tx;
539 539 uint64_t txg;
540 540
541 541 /*
542 542 * Wait for any previous destroy to complete.
543 543 */
544 544 txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg);
545 545
546 546 zilog->zl_old_header = *zh; /* debugging aid */
547 547
548 548 if (BP_IS_HOLE(&zh->zh_log))
549 549 return;
550 550
551 551 tx = dmu_tx_create(zilog->zl_os);
552 552 VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0);
553 553 dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
554 554 txg = dmu_tx_get_txg(tx);
555 555
556 556 mutex_enter(&zilog->zl_lock);
557 557
558 558 ASSERT3U(zilog->zl_destroy_txg, <, txg);
559 559 zilog->zl_destroy_txg = txg;
560 560 zilog->zl_keep_first = keep_first;
561 561
562 562 if (!list_is_empty(&zilog->zl_lwb_list)) {
563 563 ASSERT(zh->zh_claim_txg == 0);
564 564 VERIFY(!keep_first);
565 565 while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) {
566 566 list_remove(&zilog->zl_lwb_list, lwb);
567 567 if (lwb->lwb_buf != NULL)
568 568 zio_buf_free(lwb->lwb_buf, lwb->lwb_sz);
569 569 zio_free_zil(zilog->zl_spa, txg, &lwb->lwb_blk);
570 570 kmem_cache_free(zil_lwb_cache, lwb);
571 571 }
572 572 } else if (!keep_first) {
573 573 (void) zil_parse(zilog, zil_free_log_block,
574 574 zil_free_log_record, tx, zh->zh_claim_txg);
575 575 }
576 576 mutex_exit(&zilog->zl_lock);
577 577
578 578 dmu_tx_commit(tx);
579 579 }
580 580
581 581 int
582 582 zil_claim(const char *osname, void *txarg)
583 583 {
584 584 dmu_tx_t *tx = txarg;
585 585 uint64_t first_txg = dmu_tx_get_txg(tx);
586 586 zilog_t *zilog;
587 587 zil_header_t *zh;
588 588 objset_t *os;
589 589 int error;
590 590
591 591 error = dmu_objset_hold(osname, FTAG, &os);
592 592 if (error) {
593 593 cmn_err(CE_WARN, "can't open objset for %s", osname);
594 594 return (0);
595 595 }
596 596
597 597 zilog = dmu_objset_zil(os);
598 598 zh = zil_header_in_syncing_context(zilog);
599 599
600 600 if (spa_get_log_state(zilog->zl_spa) == SPA_LOG_CLEAR) {
601 601 if (!BP_IS_HOLE(&zh->zh_log))
602 602 zio_free_zil(zilog->zl_spa, first_txg, &zh->zh_log);
603 603 BP_ZERO(&zh->zh_log);
604 604 dsl_dataset_dirty(dmu_objset_ds(os), tx);
605 605 dmu_objset_rele(os, FTAG);
606 606 return (0);
607 607 }
608 608
609 609 /*
610 610 * Claim all log blocks if we haven't already done so, and remember
611 611 * the highest claimed sequence number. This ensures that if we can
612 612 * read only part of the log now (e.g. due to a missing device),
613 613 * but we can read the entire log later, we will not try to replay
614 614 * or destroy beyond the last block we successfully claimed.
615 615 */
616 616 ASSERT3U(zh->zh_claim_txg, <=, first_txg);
617 617 if (zh->zh_claim_txg == 0 && !BP_IS_HOLE(&zh->zh_log)) {
618 618 (void) zil_parse(zilog, zil_claim_log_block,
619 619 zil_claim_log_record, tx, first_txg);
620 620 zh->zh_claim_txg = first_txg;
621 621 zh->zh_claim_blk_seq = zilog->zl_parse_blk_seq;
622 622 zh->zh_claim_lr_seq = zilog->zl_parse_lr_seq;
623 623 if (zilog->zl_parse_lr_count || zilog->zl_parse_blk_count > 1)
624 624 zh->zh_flags |= ZIL_REPLAY_NEEDED;
625 625 zh->zh_flags |= ZIL_CLAIM_LR_SEQ_VALID;
626 626 dsl_dataset_dirty(dmu_objset_ds(os), tx);
627 627 }
628 628
629 629 ASSERT3U(first_txg, ==, (spa_last_synced_txg(zilog->zl_spa) + 1));
630 630 dmu_objset_rele(os, FTAG);
631 631 return (0);
632 632 }
633 633
634 634 /*
635 635 * Check the log by walking the log chain.
636 636 * Checksum errors are ok as they indicate the end of the chain.
637 637 * Any other error (no device or read failure) returns an error.
638 638 */
639 639 int
640 640 zil_check_log_chain(const char *osname, void *tx)
641 641 {
642 642 zilog_t *zilog;
643 643 objset_t *os;
644 644 blkptr_t *bp;
645 645 int error;
646 646
647 647 ASSERT(tx == NULL);
648 648
649 649 error = dmu_objset_hold(osname, FTAG, &os);
650 650 if (error) {
651 651 cmn_err(CE_WARN, "can't open objset for %s", osname);
652 652 return (0);
653 653 }
654 654
655 655 zilog = dmu_objset_zil(os);
656 656 bp = (blkptr_t *)&zilog->zl_header->zh_log;
657 657
658 658 /*
659 659 * Check the first block and determine if it's on a log device
660 660 * which may have been removed or faulted prior to loading this
661 661 * pool. If so, there's no point in checking the rest of the log
662 662 * as its content should have already been synced to the pool.
663 663 */
664 664 if (!BP_IS_HOLE(bp)) {
665 665 vdev_t *vd;
666 666 boolean_t valid = B_TRUE;
667 667
668 668 spa_config_enter(os->os_spa, SCL_STATE, FTAG, RW_READER);
669 669 vd = vdev_lookup_top(os->os_spa, DVA_GET_VDEV(&bp->blk_dva[0]));
670 670 if (vd->vdev_islog && vdev_is_dead(vd))
671 671 valid = vdev_log_state_valid(vd);
672 672 spa_config_exit(os->os_spa, SCL_STATE, FTAG);
673 673
674 674 if (!valid) {
675 675 dmu_objset_rele(os, FTAG);
676 676 return (0);
677 677 }
678 678 }
679 679
680 680 /*
681 681 * Because tx == NULL, zil_claim_log_block() will not actually claim
682 682 * any blocks, but just determine whether it is possible to do so.
683 683 * In addition to checking the log chain, zil_claim_log_block()
684 684 * will invoke zio_claim() with a done func of spa_claim_notify(),
685 685 * which will update spa_max_claim_txg. See spa_load() for details.
686 686 */
687 687 error = zil_parse(zilog, zil_claim_log_block, zil_claim_log_record, tx,
688 688 zilog->zl_header->zh_claim_txg ? -1ULL : spa_first_txg(os->os_spa));
689 689
690 690 dmu_objset_rele(os, FTAG);
691 691
692 692 return ((error == ECKSUM || error == ENOENT) ? 0 : error);
693 693 }
694 694
695 695 static int
696 696 zil_vdev_compare(const void *x1, const void *x2)
697 697 {
698 698 const uint64_t v1 = ((zil_vdev_node_t *)x1)->zv_vdev;
699 699 const uint64_t v2 = ((zil_vdev_node_t *)x2)->zv_vdev;
700 700
701 701 if (v1 < v2)
702 702 return (-1);
703 703 if (v1 > v2)
704 704 return (1);
705 705
706 706 return (0);
707 707 }
708 708
709 709 void
710 710 zil_add_block(zilog_t *zilog, const blkptr_t *bp)
711 711 {
712 712 avl_tree_t *t = &zilog->zl_vdev_tree;
713 713 avl_index_t where;
714 714 zil_vdev_node_t *zv, zvsearch;
715 715 int ndvas = BP_GET_NDVAS(bp);
716 716 int i;
717 717
718 718 if (zfs_nocacheflush)
719 719 return;
720 720
721 721 ASSERT(zilog->zl_writer);
722 722
723 723 /*
724 724 * Even though we're zl_writer, we still need a lock because the
725 725 * zl_get_data() callbacks may have dmu_sync() done callbacks
726 726 * that will run concurrently.
727 727 */
728 728 mutex_enter(&zilog->zl_vdev_lock);
729 729 for (i = 0; i < ndvas; i++) {
730 730 zvsearch.zv_vdev = DVA_GET_VDEV(&bp->blk_dva[i]);
731 731 if (avl_find(t, &zvsearch, &where) == NULL) {
732 732 zv = kmem_alloc(sizeof (*zv), KM_SLEEP);
733 733 zv->zv_vdev = zvsearch.zv_vdev;
734 734 avl_insert(t, zv, where);
735 735 }
736 736 }
737 737 mutex_exit(&zilog->zl_vdev_lock);
738 738 }
739 739
740 740 static void
741 741 zil_flush_vdevs(zilog_t *zilog)
742 742 {
743 743 spa_t *spa = zilog->zl_spa;
744 744 avl_tree_t *t = &zilog->zl_vdev_tree;
745 745 void *cookie = NULL;
746 746 zil_vdev_node_t *zv;
747 747 zio_t *zio;
748 748
749 749 ASSERT(zilog->zl_writer);
750 750
751 751 /*
752 752 * We don't need zl_vdev_lock here because we're the zl_writer,
753 753 * and all zl_get_data() callbacks are done.
754 754 */
755 755 if (avl_numnodes(t) == 0)
756 756 return;
757 757
758 758 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
759 759
760 760 zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
761 761
762 762 while ((zv = avl_destroy_nodes(t, &cookie)) != NULL) {
763 763 vdev_t *vd = vdev_lookup_top(spa, zv->zv_vdev);
764 764 if (vd != NULL)
765 765 zio_flush(zio, vd);
766 766 kmem_free(zv, sizeof (*zv));
767 767 }
768 768
769 769 /*
770 770 * Wait for all the flushes to complete. Not all devices actually
771 771 * support the DKIOCFLUSHWRITECACHE ioctl, so it's OK if it fails.
772 772 */
773 773 (void) zio_wait(zio);
774 774
775 775 spa_config_exit(spa, SCL_STATE, FTAG);
776 776 }
777 777
778 778 /*
779 779 * Function called when a log block write completes
780 780 */
781 781 static void
782 782 zil_lwb_write_done(zio_t *zio)
783 783 {
784 784 lwb_t *lwb = zio->io_private;
785 785 zilog_t *zilog = lwb->lwb_zilog;
786 786 dmu_tx_t *tx = lwb->lwb_tx;
787 787
788 788 ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF);
789 789 ASSERT(BP_GET_TYPE(zio->io_bp) == DMU_OT_INTENT_LOG);
790 790 ASSERT(BP_GET_LEVEL(zio->io_bp) == 0);
791 791 ASSERT(BP_GET_BYTEORDER(zio->io_bp) == ZFS_HOST_BYTEORDER);
792 792 ASSERT(!BP_IS_GANG(zio->io_bp));
793 793 ASSERT(!BP_IS_HOLE(zio->io_bp));
794 794 ASSERT(zio->io_bp->blk_fill == 0);
795 795
796 796 /*
797 797 * Ensure the lwb buffer pointer is cleared before releasing
798 798 * the txg. If we have had an allocation failure and
799 799 * the txg is waiting to sync then we want want zil_sync()
800 800 * to remove the lwb so that it's not picked up as the next new
801 801 * one in zil_commit_writer(). zil_sync() will only remove
802 802 * the lwb if lwb_buf is null.
803 803 */
804 804 zio_buf_free(lwb->lwb_buf, lwb->lwb_sz);
805 805 mutex_enter(&zilog->zl_lock);
806 806 lwb->lwb_buf = NULL;
807 807 lwb->lwb_tx = NULL;
808 808 mutex_exit(&zilog->zl_lock);
809 809
810 810 /*
811 811 * Now that we've written this log block, we have a stable pointer
812 812 * to the next block in the chain, so it's OK to let the txg in
813 813 * which we allocated the next block sync.
814 814 */
815 815 dmu_tx_commit(tx);
816 816 }
817 817
818 818 /*
819 819 * Initialize the io for a log block.
820 820 */
821 821 static void
822 822 zil_lwb_write_init(zilog_t *zilog, lwb_t *lwb)
823 823 {
824 824 zbookmark_t zb;
825 825
826 826 SET_BOOKMARK(&zb, lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_OBJSET],
827 827 ZB_ZIL_OBJECT, ZB_ZIL_LEVEL,
828 828 lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_SEQ]);
829 829
830 830 if (zilog->zl_root_zio == NULL) {
831 831 zilog->zl_root_zio = zio_root(zilog->zl_spa, NULL, NULL,
832 832 ZIO_FLAG_CANFAIL);
833 833 }
834 834 if (lwb->lwb_zio == NULL) {
835 835 lwb->lwb_zio = zio_rewrite(zilog->zl_root_zio, zilog->zl_spa,
836 836 0, &lwb->lwb_blk, lwb->lwb_buf, BP_GET_LSIZE(&lwb->lwb_blk),
837 837 zil_lwb_write_done, lwb, ZIO_PRIORITY_LOG_WRITE,
838 838 ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE, &zb);
839 839 }
840 840 }
841 841
842 842 /*
843 843 * Define a limited set of intent log block sizes.
844 844 * These must be a multiple of 4KB. Note only the amount used (again
845 845 * aligned to 4KB) actually gets written. However, we can't always just
846 846 * allocate SPA_MAXBLOCKSIZE as the slog space could be exhausted.
847 847 */
848 848 uint64_t zil_block_buckets[] = {
849 849 4096, /* non TX_WRITE */
850 850 8192+4096, /* data base */
851 851 32*1024 + 4096, /* NFS writes */
852 852 UINT64_MAX
853 853 };
854 854
855 855 /*
856 856 * Use the slog as long as the logbias is 'latency' and the current commit size
857 857 * is less than the limit or the total list size is less than 2X the limit.
858 858 * Limit checking is disabled by setting zil_slog_limit to UINT64_MAX.
859 859 */
860 860 uint64_t zil_slog_limit = 1024 * 1024;
861 861 #define USE_SLOG(zilog) (((zilog)->zl_logbias == ZFS_LOGBIAS_LATENCY) && \
862 862 (((zilog)->zl_cur_used < zil_slog_limit) || \
863 863 ((zilog)->zl_itx_list_sz < (zil_slog_limit << 1))))
864 864
865 865 /*
866 866 * Start a log block write and advance to the next log block.
867 867 * Calls are serialized.
868 868 */
869 869 static lwb_t *
870 870 zil_lwb_write_start(zilog_t *zilog, lwb_t *lwb)
871 871 {
872 872 lwb_t *nlwb = NULL;
873 873 zil_chain_t *zilc;
874 874 spa_t *spa = zilog->zl_spa;
875 875 blkptr_t *bp;
876 876 dmu_tx_t *tx;
877 877 uint64_t txg;
878 878 uint64_t zil_blksz, wsz;
879 879 int i, error;
880 880
881 881 if (BP_GET_CHECKSUM(&lwb->lwb_blk) == ZIO_CHECKSUM_ZILOG2) {
882 882 zilc = (zil_chain_t *)lwb->lwb_buf;
883 883 bp = &zilc->zc_next_blk;
884 884 } else {
885 885 zilc = (zil_chain_t *)(lwb->lwb_buf + lwb->lwb_sz);
886 886 bp = &zilc->zc_next_blk;
887 887 }
888 888
889 889 ASSERT(lwb->lwb_nused <= lwb->lwb_sz);
890 890
891 891 /*
892 892 * Allocate the next block and save its address in this block
893 893 * before writing it in order to establish the log chain.
894 894 * Note that if the allocation of nlwb synced before we wrote
895 895 * the block that points at it (lwb), we'd leak it if we crashed.
896 896 * Therefore, we don't do dmu_tx_commit() until zil_lwb_write_done().
897 897 * We dirty the dataset to ensure that zil_sync() will be called
898 898 * to clean up in the event of allocation failure or I/O failure.
899 899 */
900 900 tx = dmu_tx_create(zilog->zl_os);
901 901 VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0);
902 902 dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
903 903 txg = dmu_tx_get_txg(tx);
904 904
905 905 lwb->lwb_tx = tx;
906 906
907 907 /*
908 908 * Log blocks are pre-allocated. Here we select the size of the next
909 909 * block, based on size used in the last block.
910 910 * - first find the smallest bucket that will fit the block from a
911 911 * limited set of block sizes. This is because it's faster to write
912 912 * blocks allocated from the same metaslab as they are adjacent or
913 913 * close.
914 914 * - next find the maximum from the new suggested size and an array of
915 915 * previous sizes. This lessens a picket fence effect of wrongly
916 916 * guesssing the size if we have a stream of say 2k, 64k, 2k, 64k
917 917 * requests.
918 918 *
919 919 * Note we only write what is used, but we can't just allocate
920 920 * the maximum block size because we can exhaust the available
921 921 * pool log space.
922 922 */
923 923 zil_blksz = zilog->zl_cur_used + sizeof (zil_chain_t);
924 924 for (i = 0; zil_blksz > zil_block_buckets[i]; i++)
925 925 continue;
926 926 zil_blksz = zil_block_buckets[i];
927 927 if (zil_blksz == UINT64_MAX)
928 928 zil_blksz = SPA_MAXBLOCKSIZE;
929 929 zilog->zl_prev_blks[zilog->zl_prev_rotor] = zil_blksz;
930 930 for (i = 0; i < ZIL_PREV_BLKS; i++)
931 931 zil_blksz = MAX(zil_blksz, zilog->zl_prev_blks[i]);
932 932 zilog->zl_prev_rotor = (zilog->zl_prev_rotor + 1) & (ZIL_PREV_BLKS - 1);
933 933
934 934 BP_ZERO(bp);
935 935 /* pass the old blkptr in order to spread log blocks across devs */
936 936 error = zio_alloc_zil(spa, txg, bp, &lwb->lwb_blk, zil_blksz,
937 937 USE_SLOG(zilog));
938 938 if (!error) {
939 939 ASSERT3U(bp->blk_birth, ==, txg);
940 940 bp->blk_cksum = lwb->lwb_blk.blk_cksum;
941 941 bp->blk_cksum.zc_word[ZIL_ZC_SEQ]++;
942 942
943 943 /*
944 944 * Allocate a new log write buffer (lwb).
945 945 */
946 946 nlwb = zil_alloc_lwb(zilog, bp, txg);
947 947
948 948 /* Record the block for later vdev flushing */
949 949 zil_add_block(zilog, &lwb->lwb_blk);
950 950 }
951 951
952 952 if (BP_GET_CHECKSUM(&lwb->lwb_blk) == ZIO_CHECKSUM_ZILOG2) {
953 953 /* For Slim ZIL only write what is used. */
954 954 wsz = P2ROUNDUP_TYPED(lwb->lwb_nused, ZIL_MIN_BLKSZ, uint64_t);
955 955 ASSERT3U(wsz, <=, lwb->lwb_sz);
956 956 zio_shrink(lwb->lwb_zio, wsz);
957 957
958 958 } else {
959 959 wsz = lwb->lwb_sz;
960 960 }
961 961
962 962 zilc->zc_pad = 0;
963 963 zilc->zc_nused = lwb->lwb_nused;
964 964 zilc->zc_eck.zec_cksum = lwb->lwb_blk.blk_cksum;
965 965
966 966 /*
967 967 * clear unused data for security
968 968 */
969 969 bzero(lwb->lwb_buf + lwb->lwb_nused, wsz - lwb->lwb_nused);
970 970
971 971 zio_nowait(lwb->lwb_zio); /* Kick off the write for the old log block */
972 972
973 973 /*
974 974 * If there was an allocation failure then nlwb will be null which
975 975 * forces a txg_wait_synced().
976 976 */
977 977 return (nlwb);
978 978 }
979 979
980 980 static lwb_t *
981 981 zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb)
982 982 {
983 983 lr_t *lrc = &itx->itx_lr; /* common log record */
984 984 lr_write_t *lrw = (lr_write_t *)lrc;
985 985 char *lr_buf;
986 986 uint64_t txg = lrc->lrc_txg;
987 987 uint64_t reclen = lrc->lrc_reclen;
988 988 uint64_t dlen = 0;
989 989
990 990 if (lwb == NULL)
991 991 return (NULL);
992 992
993 993 ASSERT(lwb->lwb_buf != NULL);
994 994
995 995 if (lrc->lrc_txtype == TX_WRITE && itx->itx_wr_state == WR_NEED_COPY)
996 996 dlen = P2ROUNDUP_TYPED(
997 997 lrw->lr_length, sizeof (uint64_t), uint64_t);
998 998
999 999 zilog->zl_cur_used += (reclen + dlen);
1000 1000
1001 1001 zil_lwb_write_init(zilog, lwb);
1002 1002
1003 1003 /*
1004 1004 * If this record won't fit in the current log block, start a new one.
1005 1005 */
1006 1006 if (lwb->lwb_nused + reclen + dlen > lwb->lwb_sz) {
1007 1007 lwb = zil_lwb_write_start(zilog, lwb);
1008 1008 if (lwb == NULL)
1009 1009 return (NULL);
1010 1010 zil_lwb_write_init(zilog, lwb);
1011 1011 ASSERT(LWB_EMPTY(lwb));
1012 1012 if (lwb->lwb_nused + reclen + dlen > lwb->lwb_sz) {
1013 1013 txg_wait_synced(zilog->zl_dmu_pool, txg);
1014 1014 return (lwb);
1015 1015 }
1016 1016 }
1017 1017
1018 1018 lr_buf = lwb->lwb_buf + lwb->lwb_nused;
1019 1019 bcopy(lrc, lr_buf, reclen);
1020 1020 lrc = (lr_t *)lr_buf;
1021 1021 lrw = (lr_write_t *)lrc;
1022 1022
1023 1023 /*
1024 1024 * If it's a write, fetch the data or get its blkptr as appropriate.
1025 1025 */
1026 1026 if (lrc->lrc_txtype == TX_WRITE) {
1027 1027 if (txg > spa_freeze_txg(zilog->zl_spa))
1028 1028 txg_wait_synced(zilog->zl_dmu_pool, txg);
1029 1029 if (itx->itx_wr_state != WR_COPIED) {
1030 1030 char *dbuf;
1031 1031 int error;
1032 1032
1033 1033 if (dlen) {
1034 1034 ASSERT(itx->itx_wr_state == WR_NEED_COPY);
1035 1035 dbuf = lr_buf + reclen;
1036 1036 lrw->lr_common.lrc_reclen += dlen;
1037 1037 } else {
1038 1038 ASSERT(itx->itx_wr_state == WR_INDIRECT);
1039 1039 dbuf = NULL;
1040 1040 }
1041 1041 error = zilog->zl_get_data(
1042 1042 itx->itx_private, lrw, dbuf, lwb->lwb_zio);
1043 1043 if (error == EIO) {
1044 1044 txg_wait_synced(zilog->zl_dmu_pool, txg);
1045 1045 return (lwb);
1046 1046 }
1047 1047 if (error) {
1048 1048 ASSERT(error == ENOENT || error == EEXIST ||
1049 1049 error == EALREADY);
1050 1050 return (lwb);
1051 1051 }
1052 1052 }
1053 1053 }
1054 1054
|
↓ open down ↓ |
1021 lines elided |
↑ open up ↑ |
1055 1055 /*
1056 1056 * We're actually making an entry, so update lrc_seq to be the
1057 1057 * log record sequence number. Note that this is generally not
1058 1058 * equal to the itx sequence number because not all transactions
1059 1059 * are synchronous, and sometimes spa_sync() gets there first.
1060 1060 */
1061 1061 lrc->lrc_seq = ++zilog->zl_lr_seq; /* we are single threaded */
1062 1062 lwb->lwb_nused += reclen + dlen;
1063 1063 lwb->lwb_max_txg = MAX(lwb->lwb_max_txg, txg);
1064 1064 ASSERT3U(lwb->lwb_nused, <=, lwb->lwb_sz);
1065 - ASSERT3U(P2PHASE(lwb->lwb_nused, sizeof (uint64_t)), ==, 0);
1065 + ASSERT0(P2PHASE(lwb->lwb_nused, sizeof (uint64_t)));
1066 1066
1067 1067 return (lwb);
1068 1068 }
1069 1069
1070 1070 itx_t *
1071 1071 zil_itx_create(uint64_t txtype, size_t lrsize)
1072 1072 {
1073 1073 itx_t *itx;
1074 1074
1075 1075 lrsize = P2ROUNDUP_TYPED(lrsize, sizeof (uint64_t), size_t);
1076 1076
1077 1077 itx = kmem_alloc(offsetof(itx_t, itx_lr) + lrsize, KM_SLEEP);
1078 1078 itx->itx_lr.lrc_txtype = txtype;
1079 1079 itx->itx_lr.lrc_reclen = lrsize;
1080 1080 itx->itx_sod = lrsize; /* if write & WR_NEED_COPY will be increased */
1081 1081 itx->itx_lr.lrc_seq = 0; /* defensive */
1082 1082 itx->itx_sync = B_TRUE; /* default is synchronous */
1083 1083
1084 1084 return (itx);
1085 1085 }
1086 1086
1087 1087 void
1088 1088 zil_itx_destroy(itx_t *itx)
1089 1089 {
1090 1090 kmem_free(itx, offsetof(itx_t, itx_lr) + itx->itx_lr.lrc_reclen);
1091 1091 }
1092 1092
1093 1093 /*
1094 1094 * Free up the sync and async itxs. The itxs_t has already been detached
1095 1095 * so no locks are needed.
1096 1096 */
1097 1097 static void
1098 1098 zil_itxg_clean(itxs_t *itxs)
1099 1099 {
1100 1100 itx_t *itx;
1101 1101 list_t *list;
1102 1102 avl_tree_t *t;
1103 1103 void *cookie;
1104 1104 itx_async_node_t *ian;
1105 1105
1106 1106 list = &itxs->i_sync_list;
1107 1107 while ((itx = list_head(list)) != NULL) {
1108 1108 list_remove(list, itx);
1109 1109 kmem_free(itx, offsetof(itx_t, itx_lr) +
1110 1110 itx->itx_lr.lrc_reclen);
1111 1111 }
1112 1112
1113 1113 cookie = NULL;
1114 1114 t = &itxs->i_async_tree;
1115 1115 while ((ian = avl_destroy_nodes(t, &cookie)) != NULL) {
1116 1116 list = &ian->ia_list;
1117 1117 while ((itx = list_head(list)) != NULL) {
1118 1118 list_remove(list, itx);
1119 1119 kmem_free(itx, offsetof(itx_t, itx_lr) +
1120 1120 itx->itx_lr.lrc_reclen);
1121 1121 }
1122 1122 list_destroy(list);
1123 1123 kmem_free(ian, sizeof (itx_async_node_t));
1124 1124 }
1125 1125 avl_destroy(t);
1126 1126
1127 1127 kmem_free(itxs, sizeof (itxs_t));
1128 1128 }
1129 1129
1130 1130 static int
1131 1131 zil_aitx_compare(const void *x1, const void *x2)
1132 1132 {
1133 1133 const uint64_t o1 = ((itx_async_node_t *)x1)->ia_foid;
1134 1134 const uint64_t o2 = ((itx_async_node_t *)x2)->ia_foid;
1135 1135
1136 1136 if (o1 < o2)
1137 1137 return (-1);
1138 1138 if (o1 > o2)
1139 1139 return (1);
1140 1140
1141 1141 return (0);
1142 1142 }
1143 1143
1144 1144 /*
1145 1145 * Remove all async itx with the given oid.
1146 1146 */
1147 1147 static void
1148 1148 zil_remove_async(zilog_t *zilog, uint64_t oid)
1149 1149 {
1150 1150 uint64_t otxg, txg;
1151 1151 itx_async_node_t *ian;
1152 1152 avl_tree_t *t;
1153 1153 avl_index_t where;
1154 1154 list_t clean_list;
1155 1155 itx_t *itx;
1156 1156
1157 1157 ASSERT(oid != 0);
1158 1158 list_create(&clean_list, sizeof (itx_t), offsetof(itx_t, itx_node));
1159 1159
1160 1160 if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */
1161 1161 otxg = ZILTEST_TXG;
1162 1162 else
1163 1163 otxg = spa_last_synced_txg(zilog->zl_spa) + 1;
1164 1164
1165 1165 for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) {
1166 1166 itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK];
1167 1167
1168 1168 mutex_enter(&itxg->itxg_lock);
1169 1169 if (itxg->itxg_txg != txg) {
1170 1170 mutex_exit(&itxg->itxg_lock);
1171 1171 continue;
1172 1172 }
1173 1173
1174 1174 /*
1175 1175 * Locate the object node and append its list.
1176 1176 */
1177 1177 t = &itxg->itxg_itxs->i_async_tree;
1178 1178 ian = avl_find(t, &oid, &where);
1179 1179 if (ian != NULL)
1180 1180 list_move_tail(&clean_list, &ian->ia_list);
1181 1181 mutex_exit(&itxg->itxg_lock);
1182 1182 }
1183 1183 while ((itx = list_head(&clean_list)) != NULL) {
1184 1184 list_remove(&clean_list, itx);
1185 1185 kmem_free(itx, offsetof(itx_t, itx_lr) +
1186 1186 itx->itx_lr.lrc_reclen);
1187 1187 }
1188 1188 list_destroy(&clean_list);
1189 1189 }
1190 1190
1191 1191 void
1192 1192 zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx)
1193 1193 {
1194 1194 uint64_t txg;
1195 1195 itxg_t *itxg;
1196 1196 itxs_t *itxs, *clean = NULL;
1197 1197
1198 1198 /*
1199 1199 * Object ids can be re-instantiated in the next txg so
1200 1200 * remove any async transactions to avoid future leaks.
1201 1201 * This can happen if a fsync occurs on the re-instantiated
1202 1202 * object for a WR_INDIRECT or WR_NEED_COPY write, which gets
1203 1203 * the new file data and flushes a write record for the old object.
1204 1204 */
1205 1205 if ((itx->itx_lr.lrc_txtype & ~TX_CI) == TX_REMOVE)
1206 1206 zil_remove_async(zilog, itx->itx_oid);
1207 1207
1208 1208 /*
1209 1209 * Ensure the data of a renamed file is committed before the rename.
1210 1210 */
1211 1211 if ((itx->itx_lr.lrc_txtype & ~TX_CI) == TX_RENAME)
1212 1212 zil_async_to_sync(zilog, itx->itx_oid);
1213 1213
1214 1214 if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX)
1215 1215 txg = ZILTEST_TXG;
1216 1216 else
1217 1217 txg = dmu_tx_get_txg(tx);
1218 1218
1219 1219 itxg = &zilog->zl_itxg[txg & TXG_MASK];
1220 1220 mutex_enter(&itxg->itxg_lock);
1221 1221 itxs = itxg->itxg_itxs;
1222 1222 if (itxg->itxg_txg != txg) {
1223 1223 if (itxs != NULL) {
1224 1224 /*
1225 1225 * The zil_clean callback hasn't got around to cleaning
1226 1226 * this itxg. Save the itxs for release below.
1227 1227 * This should be rare.
1228 1228 */
1229 1229 atomic_add_64(&zilog->zl_itx_list_sz, -itxg->itxg_sod);
1230 1230 itxg->itxg_sod = 0;
1231 1231 clean = itxg->itxg_itxs;
1232 1232 }
1233 1233 ASSERT(itxg->itxg_sod == 0);
1234 1234 itxg->itxg_txg = txg;
1235 1235 itxs = itxg->itxg_itxs = kmem_zalloc(sizeof (itxs_t), KM_SLEEP);
1236 1236
1237 1237 list_create(&itxs->i_sync_list, sizeof (itx_t),
1238 1238 offsetof(itx_t, itx_node));
1239 1239 avl_create(&itxs->i_async_tree, zil_aitx_compare,
1240 1240 sizeof (itx_async_node_t),
1241 1241 offsetof(itx_async_node_t, ia_node));
1242 1242 }
1243 1243 if (itx->itx_sync) {
1244 1244 list_insert_tail(&itxs->i_sync_list, itx);
1245 1245 atomic_add_64(&zilog->zl_itx_list_sz, itx->itx_sod);
1246 1246 itxg->itxg_sod += itx->itx_sod;
1247 1247 } else {
1248 1248 avl_tree_t *t = &itxs->i_async_tree;
1249 1249 uint64_t foid = ((lr_ooo_t *)&itx->itx_lr)->lr_foid;
1250 1250 itx_async_node_t *ian;
1251 1251 avl_index_t where;
1252 1252
1253 1253 ian = avl_find(t, &foid, &where);
1254 1254 if (ian == NULL) {
1255 1255 ian = kmem_alloc(sizeof (itx_async_node_t), KM_SLEEP);
1256 1256 list_create(&ian->ia_list, sizeof (itx_t),
1257 1257 offsetof(itx_t, itx_node));
1258 1258 ian->ia_foid = foid;
1259 1259 avl_insert(t, ian, where);
1260 1260 }
1261 1261 list_insert_tail(&ian->ia_list, itx);
1262 1262 }
1263 1263
1264 1264 itx->itx_lr.lrc_txg = dmu_tx_get_txg(tx);
1265 1265 mutex_exit(&itxg->itxg_lock);
1266 1266
1267 1267 /* Release the old itxs now we've dropped the lock */
1268 1268 if (clean != NULL)
1269 1269 zil_itxg_clean(clean);
1270 1270 }
1271 1271
1272 1272 /*
1273 1273 * If there are any in-memory intent log transactions which have now been
1274 1274 * synced then start up a taskq to free them.
1275 1275 */
1276 1276 void
1277 1277 zil_clean(zilog_t *zilog, uint64_t synced_txg)
1278 1278 {
1279 1279 itxg_t *itxg = &zilog->zl_itxg[synced_txg & TXG_MASK];
1280 1280 itxs_t *clean_me;
1281 1281
1282 1282 mutex_enter(&itxg->itxg_lock);
1283 1283 if (itxg->itxg_itxs == NULL || itxg->itxg_txg == ZILTEST_TXG) {
1284 1284 mutex_exit(&itxg->itxg_lock);
1285 1285 return;
1286 1286 }
1287 1287 ASSERT3U(itxg->itxg_txg, <=, synced_txg);
1288 1288 ASSERT(itxg->itxg_txg != 0);
1289 1289 ASSERT(zilog->zl_clean_taskq != NULL);
1290 1290 atomic_add_64(&zilog->zl_itx_list_sz, -itxg->itxg_sod);
1291 1291 itxg->itxg_sod = 0;
1292 1292 clean_me = itxg->itxg_itxs;
1293 1293 itxg->itxg_itxs = NULL;
1294 1294 itxg->itxg_txg = 0;
1295 1295 mutex_exit(&itxg->itxg_lock);
1296 1296 /*
1297 1297 * Preferably start a task queue to free up the old itxs but
1298 1298 * if taskq_dispatch can't allocate resources to do that then
1299 1299 * free it in-line. This should be rare. Note, using TQ_SLEEP
1300 1300 * created a bad performance problem.
1301 1301 */
1302 1302 if (taskq_dispatch(zilog->zl_clean_taskq,
1303 1303 (void (*)(void *))zil_itxg_clean, clean_me, TQ_NOSLEEP) == NULL)
1304 1304 zil_itxg_clean(clean_me);
1305 1305 }
1306 1306
1307 1307 /*
1308 1308 * Get the list of itxs to commit into zl_itx_commit_list.
1309 1309 */
1310 1310 static void
1311 1311 zil_get_commit_list(zilog_t *zilog)
1312 1312 {
1313 1313 uint64_t otxg, txg;
1314 1314 list_t *commit_list = &zilog->zl_itx_commit_list;
1315 1315 uint64_t push_sod = 0;
1316 1316
1317 1317 if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */
1318 1318 otxg = ZILTEST_TXG;
1319 1319 else
1320 1320 otxg = spa_last_synced_txg(zilog->zl_spa) + 1;
1321 1321
1322 1322 for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) {
1323 1323 itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK];
1324 1324
1325 1325 mutex_enter(&itxg->itxg_lock);
1326 1326 if (itxg->itxg_txg != txg) {
1327 1327 mutex_exit(&itxg->itxg_lock);
1328 1328 continue;
1329 1329 }
1330 1330
1331 1331 list_move_tail(commit_list, &itxg->itxg_itxs->i_sync_list);
1332 1332 push_sod += itxg->itxg_sod;
1333 1333 itxg->itxg_sod = 0;
1334 1334
1335 1335 mutex_exit(&itxg->itxg_lock);
1336 1336 }
1337 1337 atomic_add_64(&zilog->zl_itx_list_sz, -push_sod);
1338 1338 }
1339 1339
1340 1340 /*
1341 1341 * Move the async itxs for a specified object to commit into sync lists.
1342 1342 */
1343 1343 static void
1344 1344 zil_async_to_sync(zilog_t *zilog, uint64_t foid)
1345 1345 {
1346 1346 uint64_t otxg, txg;
1347 1347 itx_async_node_t *ian;
1348 1348 avl_tree_t *t;
1349 1349 avl_index_t where;
1350 1350
1351 1351 if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */
1352 1352 otxg = ZILTEST_TXG;
1353 1353 else
1354 1354 otxg = spa_last_synced_txg(zilog->zl_spa) + 1;
1355 1355
1356 1356 for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) {
1357 1357 itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK];
1358 1358
1359 1359 mutex_enter(&itxg->itxg_lock);
1360 1360 if (itxg->itxg_txg != txg) {
1361 1361 mutex_exit(&itxg->itxg_lock);
1362 1362 continue;
1363 1363 }
1364 1364
1365 1365 /*
1366 1366 * If a foid is specified then find that node and append its
1367 1367 * list. Otherwise walk the tree appending all the lists
1368 1368 * to the sync list. We add to the end rather than the
1369 1369 * beginning to ensure the create has happened.
1370 1370 */
1371 1371 t = &itxg->itxg_itxs->i_async_tree;
1372 1372 if (foid != 0) {
1373 1373 ian = avl_find(t, &foid, &where);
1374 1374 if (ian != NULL) {
1375 1375 list_move_tail(&itxg->itxg_itxs->i_sync_list,
1376 1376 &ian->ia_list);
1377 1377 }
1378 1378 } else {
1379 1379 void *cookie = NULL;
1380 1380
1381 1381 while ((ian = avl_destroy_nodes(t, &cookie)) != NULL) {
1382 1382 list_move_tail(&itxg->itxg_itxs->i_sync_list,
1383 1383 &ian->ia_list);
1384 1384 list_destroy(&ian->ia_list);
1385 1385 kmem_free(ian, sizeof (itx_async_node_t));
1386 1386 }
1387 1387 }
1388 1388 mutex_exit(&itxg->itxg_lock);
1389 1389 }
1390 1390 }
1391 1391
1392 1392 static void
1393 1393 zil_commit_writer(zilog_t *zilog)
1394 1394 {
1395 1395 uint64_t txg;
1396 1396 itx_t *itx;
1397 1397 lwb_t *lwb;
1398 1398 spa_t *spa = zilog->zl_spa;
1399 1399 int error = 0;
1400 1400
1401 1401 ASSERT(zilog->zl_root_zio == NULL);
1402 1402
1403 1403 mutex_exit(&zilog->zl_lock);
1404 1404
1405 1405 zil_get_commit_list(zilog);
1406 1406
1407 1407 /*
1408 1408 * Return if there's nothing to commit before we dirty the fs by
1409 1409 * calling zil_create().
1410 1410 */
1411 1411 if (list_head(&zilog->zl_itx_commit_list) == NULL) {
1412 1412 mutex_enter(&zilog->zl_lock);
1413 1413 return;
1414 1414 }
1415 1415
1416 1416 if (zilog->zl_suspend) {
1417 1417 lwb = NULL;
1418 1418 } else {
1419 1419 lwb = list_tail(&zilog->zl_lwb_list);
1420 1420 if (lwb == NULL)
1421 1421 lwb = zil_create(zilog);
1422 1422 }
1423 1423
1424 1424 DTRACE_PROBE1(zil__cw1, zilog_t *, zilog);
1425 1425 while (itx = list_head(&zilog->zl_itx_commit_list)) {
1426 1426 txg = itx->itx_lr.lrc_txg;
1427 1427 ASSERT(txg);
1428 1428
1429 1429 if (txg > spa_last_synced_txg(spa) || txg > spa_freeze_txg(spa))
1430 1430 lwb = zil_lwb_commit(zilog, itx, lwb);
1431 1431 list_remove(&zilog->zl_itx_commit_list, itx);
1432 1432 kmem_free(itx, offsetof(itx_t, itx_lr)
1433 1433 + itx->itx_lr.lrc_reclen);
1434 1434 }
1435 1435 DTRACE_PROBE1(zil__cw2, zilog_t *, zilog);
1436 1436
1437 1437 /* write the last block out */
1438 1438 if (lwb != NULL && lwb->lwb_zio != NULL)
1439 1439 lwb = zil_lwb_write_start(zilog, lwb);
1440 1440
1441 1441 zilog->zl_cur_used = 0;
1442 1442
1443 1443 /*
1444 1444 * Wait if necessary for the log blocks to be on stable storage.
1445 1445 */
1446 1446 if (zilog->zl_root_zio) {
1447 1447 error = zio_wait(zilog->zl_root_zio);
1448 1448 zilog->zl_root_zio = NULL;
1449 1449 zil_flush_vdevs(zilog);
1450 1450 }
1451 1451
1452 1452 if (error || lwb == NULL)
1453 1453 txg_wait_synced(zilog->zl_dmu_pool, 0);
1454 1454
1455 1455 mutex_enter(&zilog->zl_lock);
1456 1456
1457 1457 /*
1458 1458 * Remember the highest committed log sequence number for ztest.
1459 1459 * We only update this value when all the log writes succeeded,
1460 1460 * because ztest wants to ASSERT that it got the whole log chain.
1461 1461 */
1462 1462 if (error == 0 && lwb != NULL)
1463 1463 zilog->zl_commit_lr_seq = zilog->zl_lr_seq;
1464 1464 }
1465 1465
1466 1466 /*
1467 1467 * Commit zfs transactions to stable storage.
1468 1468 * If foid is 0 push out all transactions, otherwise push only those
1469 1469 * for that object or might reference that object.
1470 1470 *
1471 1471 * itxs are committed in batches. In a heavily stressed zil there will be
1472 1472 * a commit writer thread who is writing out a bunch of itxs to the log
1473 1473 * for a set of committing threads (cthreads) in the same batch as the writer.
1474 1474 * Those cthreads are all waiting on the same cv for that batch.
1475 1475 *
1476 1476 * There will also be a different and growing batch of threads that are
1477 1477 * waiting to commit (qthreads). When the committing batch completes
1478 1478 * a transition occurs such that the cthreads exit and the qthreads become
1479 1479 * cthreads. One of the new cthreads becomes the writer thread for the
1480 1480 * batch. Any new threads arriving become new qthreads.
1481 1481 *
1482 1482 * Only 2 condition variables are needed and there's no transition
1483 1483 * between the two cvs needed. They just flip-flop between qthreads
1484 1484 * and cthreads.
1485 1485 *
1486 1486 * Using this scheme we can efficiently wakeup up only those threads
1487 1487 * that have been committed.
1488 1488 */
1489 1489 void
1490 1490 zil_commit(zilog_t *zilog, uint64_t foid)
1491 1491 {
1492 1492 uint64_t mybatch;
1493 1493
1494 1494 if (zilog->zl_sync == ZFS_SYNC_DISABLED)
1495 1495 return;
1496 1496
1497 1497 /* move the async itxs for the foid to the sync queues */
1498 1498 zil_async_to_sync(zilog, foid);
1499 1499
1500 1500 mutex_enter(&zilog->zl_lock);
1501 1501 mybatch = zilog->zl_next_batch;
1502 1502 while (zilog->zl_writer) {
1503 1503 cv_wait(&zilog->zl_cv_batch[mybatch & 1], &zilog->zl_lock);
1504 1504 if (mybatch <= zilog->zl_com_batch) {
1505 1505 mutex_exit(&zilog->zl_lock);
1506 1506 return;
1507 1507 }
1508 1508 }
1509 1509
1510 1510 zilog->zl_next_batch++;
1511 1511 zilog->zl_writer = B_TRUE;
1512 1512 zil_commit_writer(zilog);
1513 1513 zilog->zl_com_batch = mybatch;
1514 1514 zilog->zl_writer = B_FALSE;
1515 1515 mutex_exit(&zilog->zl_lock);
1516 1516
1517 1517 /* wake up one thread to become the next writer */
1518 1518 cv_signal(&zilog->zl_cv_batch[(mybatch+1) & 1]);
1519 1519
1520 1520 /* wake up all threads waiting for this batch to be committed */
1521 1521 cv_broadcast(&zilog->zl_cv_batch[mybatch & 1]);
1522 1522 }
1523 1523
1524 1524 /*
1525 1525 * Called in syncing context to free committed log blocks and update log header.
1526 1526 */
1527 1527 void
1528 1528 zil_sync(zilog_t *zilog, dmu_tx_t *tx)
1529 1529 {
1530 1530 zil_header_t *zh = zil_header_in_syncing_context(zilog);
1531 1531 uint64_t txg = dmu_tx_get_txg(tx);
1532 1532 spa_t *spa = zilog->zl_spa;
1533 1533 uint64_t *replayed_seq = &zilog->zl_replayed_seq[txg & TXG_MASK];
1534 1534 lwb_t *lwb;
1535 1535
1536 1536 /*
1537 1537 * We don't zero out zl_destroy_txg, so make sure we don't try
1538 1538 * to destroy it twice.
1539 1539 */
1540 1540 if (spa_sync_pass(spa) != 1)
1541 1541 return;
1542 1542
1543 1543 mutex_enter(&zilog->zl_lock);
1544 1544
1545 1545 ASSERT(zilog->zl_stop_sync == 0);
1546 1546
1547 1547 if (*replayed_seq != 0) {
1548 1548 ASSERT(zh->zh_replay_seq < *replayed_seq);
1549 1549 zh->zh_replay_seq = *replayed_seq;
1550 1550 *replayed_seq = 0;
1551 1551 }
1552 1552
1553 1553 if (zilog->zl_destroy_txg == txg) {
1554 1554 blkptr_t blk = zh->zh_log;
1555 1555
1556 1556 ASSERT(list_head(&zilog->zl_lwb_list) == NULL);
1557 1557
1558 1558 bzero(zh, sizeof (zil_header_t));
1559 1559 bzero(zilog->zl_replayed_seq, sizeof (zilog->zl_replayed_seq));
1560 1560
1561 1561 if (zilog->zl_keep_first) {
1562 1562 /*
1563 1563 * If this block was part of log chain that couldn't
1564 1564 * be claimed because a device was missing during
1565 1565 * zil_claim(), but that device later returns,
1566 1566 * then this block could erroneously appear valid.
1567 1567 * To guard against this, assign a new GUID to the new
1568 1568 * log chain so it doesn't matter what blk points to.
1569 1569 */
1570 1570 zil_init_log_chain(zilog, &blk);
1571 1571 zh->zh_log = blk;
1572 1572 }
1573 1573 }
1574 1574
1575 1575 while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) {
1576 1576 zh->zh_log = lwb->lwb_blk;
1577 1577 if (lwb->lwb_buf != NULL || lwb->lwb_max_txg > txg)
1578 1578 break;
1579 1579 list_remove(&zilog->zl_lwb_list, lwb);
1580 1580 zio_free_zil(spa, txg, &lwb->lwb_blk);
1581 1581 kmem_cache_free(zil_lwb_cache, lwb);
1582 1582
1583 1583 /*
1584 1584 * If we don't have anything left in the lwb list then
1585 1585 * we've had an allocation failure and we need to zero
1586 1586 * out the zil_header blkptr so that we don't end
1587 1587 * up freeing the same block twice.
1588 1588 */
1589 1589 if (list_head(&zilog->zl_lwb_list) == NULL)
1590 1590 BP_ZERO(&zh->zh_log);
1591 1591 }
1592 1592 mutex_exit(&zilog->zl_lock);
1593 1593 }
1594 1594
1595 1595 void
1596 1596 zil_init(void)
1597 1597 {
1598 1598 zil_lwb_cache = kmem_cache_create("zil_lwb_cache",
1599 1599 sizeof (struct lwb), 0, NULL, NULL, NULL, NULL, NULL, 0);
1600 1600 }
1601 1601
1602 1602 void
1603 1603 zil_fini(void)
1604 1604 {
1605 1605 kmem_cache_destroy(zil_lwb_cache);
1606 1606 }
1607 1607
1608 1608 void
1609 1609 zil_set_sync(zilog_t *zilog, uint64_t sync)
1610 1610 {
1611 1611 zilog->zl_sync = sync;
1612 1612 }
1613 1613
1614 1614 void
1615 1615 zil_set_logbias(zilog_t *zilog, uint64_t logbias)
1616 1616 {
1617 1617 zilog->zl_logbias = logbias;
1618 1618 }
1619 1619
1620 1620 zilog_t *
1621 1621 zil_alloc(objset_t *os, zil_header_t *zh_phys)
1622 1622 {
1623 1623 zilog_t *zilog;
1624 1624
1625 1625 zilog = kmem_zalloc(sizeof (zilog_t), KM_SLEEP);
1626 1626
1627 1627 zilog->zl_header = zh_phys;
1628 1628 zilog->zl_os = os;
1629 1629 zilog->zl_spa = dmu_objset_spa(os);
1630 1630 zilog->zl_dmu_pool = dmu_objset_pool(os);
1631 1631 zilog->zl_destroy_txg = TXG_INITIAL - 1;
1632 1632 zilog->zl_logbias = dmu_objset_logbias(os);
1633 1633 zilog->zl_sync = dmu_objset_syncprop(os);
1634 1634 zilog->zl_next_batch = 1;
1635 1635
1636 1636 mutex_init(&zilog->zl_lock, NULL, MUTEX_DEFAULT, NULL);
1637 1637
1638 1638 for (int i = 0; i < TXG_SIZE; i++) {
1639 1639 mutex_init(&zilog->zl_itxg[i].itxg_lock, NULL,
1640 1640 MUTEX_DEFAULT, NULL);
1641 1641 }
1642 1642
1643 1643 list_create(&zilog->zl_lwb_list, sizeof (lwb_t),
1644 1644 offsetof(lwb_t, lwb_node));
1645 1645
1646 1646 list_create(&zilog->zl_itx_commit_list, sizeof (itx_t),
1647 1647 offsetof(itx_t, itx_node));
1648 1648
1649 1649 mutex_init(&zilog->zl_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
1650 1650
1651 1651 avl_create(&zilog->zl_vdev_tree, zil_vdev_compare,
1652 1652 sizeof (zil_vdev_node_t), offsetof(zil_vdev_node_t, zv_node));
1653 1653
1654 1654 cv_init(&zilog->zl_cv_writer, NULL, CV_DEFAULT, NULL);
1655 1655 cv_init(&zilog->zl_cv_suspend, NULL, CV_DEFAULT, NULL);
1656 1656 cv_init(&zilog->zl_cv_batch[0], NULL, CV_DEFAULT, NULL);
1657 1657 cv_init(&zilog->zl_cv_batch[1], NULL, CV_DEFAULT, NULL);
1658 1658
1659 1659 return (zilog);
1660 1660 }
1661 1661
1662 1662 void
1663 1663 zil_free(zilog_t *zilog)
1664 1664 {
1665 1665 zilog->zl_stop_sync = 1;
1666 1666
1667 1667 ASSERT(list_is_empty(&zilog->zl_lwb_list));
1668 1668 list_destroy(&zilog->zl_lwb_list);
1669 1669
1670 1670 avl_destroy(&zilog->zl_vdev_tree);
1671 1671 mutex_destroy(&zilog->zl_vdev_lock);
1672 1672
1673 1673 ASSERT(list_is_empty(&zilog->zl_itx_commit_list));
1674 1674 list_destroy(&zilog->zl_itx_commit_list);
1675 1675
1676 1676 for (int i = 0; i < TXG_SIZE; i++) {
1677 1677 /*
1678 1678 * It's possible for an itx to be generated that doesn't dirty
1679 1679 * a txg (e.g. ztest TX_TRUNCATE). So there's no zil_clean()
1680 1680 * callback to remove the entry. We remove those here.
1681 1681 *
1682 1682 * Also free up the ziltest itxs.
1683 1683 */
1684 1684 if (zilog->zl_itxg[i].itxg_itxs)
1685 1685 zil_itxg_clean(zilog->zl_itxg[i].itxg_itxs);
1686 1686 mutex_destroy(&zilog->zl_itxg[i].itxg_lock);
1687 1687 }
1688 1688
1689 1689 mutex_destroy(&zilog->zl_lock);
1690 1690
1691 1691 cv_destroy(&zilog->zl_cv_writer);
1692 1692 cv_destroy(&zilog->zl_cv_suspend);
1693 1693 cv_destroy(&zilog->zl_cv_batch[0]);
1694 1694 cv_destroy(&zilog->zl_cv_batch[1]);
1695 1695
1696 1696 kmem_free(zilog, sizeof (zilog_t));
1697 1697 }
1698 1698
1699 1699 /*
1700 1700 * Open an intent log.
1701 1701 */
1702 1702 zilog_t *
1703 1703 zil_open(objset_t *os, zil_get_data_t *get_data)
1704 1704 {
1705 1705 zilog_t *zilog = dmu_objset_zil(os);
1706 1706
1707 1707 ASSERT(zilog->zl_clean_taskq == NULL);
1708 1708 ASSERT(zilog->zl_get_data == NULL);
1709 1709 ASSERT(list_is_empty(&zilog->zl_lwb_list));
1710 1710
1711 1711 zilog->zl_get_data = get_data;
1712 1712 zilog->zl_clean_taskq = taskq_create("zil_clean", 1, minclsyspri,
1713 1713 2, 2, TASKQ_PREPOPULATE);
1714 1714
1715 1715 return (zilog);
1716 1716 }
1717 1717
1718 1718 /*
1719 1719 * Close an intent log.
1720 1720 */
1721 1721 void
1722 1722 zil_close(zilog_t *zilog)
1723 1723 {
1724 1724 lwb_t *lwb;
1725 1725 uint64_t txg = 0;
1726 1726
1727 1727 zil_commit(zilog, 0); /* commit all itx */
1728 1728
1729 1729 /*
1730 1730 * The lwb_max_txg for the stubby lwb will reflect the last activity
1731 1731 * for the zil. After a txg_wait_synced() on the txg we know all the
1732 1732 * callbacks have occurred that may clean the zil. Only then can we
1733 1733 * destroy the zl_clean_taskq.
1734 1734 */
1735 1735 mutex_enter(&zilog->zl_lock);
1736 1736 lwb = list_tail(&zilog->zl_lwb_list);
1737 1737 if (lwb != NULL)
1738 1738 txg = lwb->lwb_max_txg;
1739 1739 mutex_exit(&zilog->zl_lock);
1740 1740 if (txg)
1741 1741 txg_wait_synced(zilog->zl_dmu_pool, txg);
1742 1742
1743 1743 taskq_destroy(zilog->zl_clean_taskq);
1744 1744 zilog->zl_clean_taskq = NULL;
1745 1745 zilog->zl_get_data = NULL;
1746 1746
1747 1747 /*
1748 1748 * We should have only one LWB left on the list; remove it now.
1749 1749 */
1750 1750 mutex_enter(&zilog->zl_lock);
1751 1751 lwb = list_head(&zilog->zl_lwb_list);
1752 1752 if (lwb != NULL) {
1753 1753 ASSERT(lwb == list_tail(&zilog->zl_lwb_list));
1754 1754 list_remove(&zilog->zl_lwb_list, lwb);
1755 1755 zio_buf_free(lwb->lwb_buf, lwb->lwb_sz);
1756 1756 kmem_cache_free(zil_lwb_cache, lwb);
1757 1757 }
1758 1758 mutex_exit(&zilog->zl_lock);
1759 1759 }
1760 1760
1761 1761 /*
1762 1762 * Suspend an intent log. While in suspended mode, we still honor
1763 1763 * synchronous semantics, but we rely on txg_wait_synced() to do it.
1764 1764 * We suspend the log briefly when taking a snapshot so that the snapshot
1765 1765 * contains all the data it's supposed to, and has an empty intent log.
1766 1766 */
1767 1767 int
1768 1768 zil_suspend(zilog_t *zilog)
1769 1769 {
1770 1770 const zil_header_t *zh = zilog->zl_header;
1771 1771
1772 1772 mutex_enter(&zilog->zl_lock);
1773 1773 if (zh->zh_flags & ZIL_REPLAY_NEEDED) { /* unplayed log */
1774 1774 mutex_exit(&zilog->zl_lock);
1775 1775 return (EBUSY);
1776 1776 }
1777 1777 if (zilog->zl_suspend++ != 0) {
1778 1778 /*
1779 1779 * Someone else already began a suspend.
1780 1780 * Just wait for them to finish.
1781 1781 */
1782 1782 while (zilog->zl_suspending)
1783 1783 cv_wait(&zilog->zl_cv_suspend, &zilog->zl_lock);
1784 1784 mutex_exit(&zilog->zl_lock);
1785 1785 return (0);
1786 1786 }
1787 1787 zilog->zl_suspending = B_TRUE;
1788 1788 mutex_exit(&zilog->zl_lock);
1789 1789
1790 1790 zil_commit(zilog, 0);
1791 1791
1792 1792 zil_destroy(zilog, B_FALSE);
1793 1793
1794 1794 mutex_enter(&zilog->zl_lock);
1795 1795 zilog->zl_suspending = B_FALSE;
1796 1796 cv_broadcast(&zilog->zl_cv_suspend);
1797 1797 mutex_exit(&zilog->zl_lock);
1798 1798
1799 1799 return (0);
1800 1800 }
1801 1801
1802 1802 void
1803 1803 zil_resume(zilog_t *zilog)
1804 1804 {
1805 1805 mutex_enter(&zilog->zl_lock);
1806 1806 ASSERT(zilog->zl_suspend != 0);
1807 1807 zilog->zl_suspend--;
1808 1808 mutex_exit(&zilog->zl_lock);
1809 1809 }
1810 1810
1811 1811 typedef struct zil_replay_arg {
1812 1812 zil_replay_func_t **zr_replay;
1813 1813 void *zr_arg;
1814 1814 boolean_t zr_byteswap;
1815 1815 char *zr_lr;
1816 1816 } zil_replay_arg_t;
1817 1817
1818 1818 static int
1819 1819 zil_replay_error(zilog_t *zilog, lr_t *lr, int error)
1820 1820 {
1821 1821 char name[MAXNAMELEN];
1822 1822
1823 1823 zilog->zl_replaying_seq--; /* didn't actually replay this one */
1824 1824
1825 1825 dmu_objset_name(zilog->zl_os, name);
1826 1826
1827 1827 cmn_err(CE_WARN, "ZFS replay transaction error %d, "
1828 1828 "dataset %s, seq 0x%llx, txtype %llu %s\n", error, name,
1829 1829 (u_longlong_t)lr->lrc_seq,
1830 1830 (u_longlong_t)(lr->lrc_txtype & ~TX_CI),
1831 1831 (lr->lrc_txtype & TX_CI) ? "CI" : "");
1832 1832
1833 1833 return (error);
1834 1834 }
1835 1835
1836 1836 static int
1837 1837 zil_replay_log_record(zilog_t *zilog, lr_t *lr, void *zra, uint64_t claim_txg)
1838 1838 {
1839 1839 zil_replay_arg_t *zr = zra;
1840 1840 const zil_header_t *zh = zilog->zl_header;
1841 1841 uint64_t reclen = lr->lrc_reclen;
1842 1842 uint64_t txtype = lr->lrc_txtype;
1843 1843 int error = 0;
1844 1844
1845 1845 zilog->zl_replaying_seq = lr->lrc_seq;
1846 1846
1847 1847 if (lr->lrc_seq <= zh->zh_replay_seq) /* already replayed */
1848 1848 return (0);
1849 1849
1850 1850 if (lr->lrc_txg < claim_txg) /* already committed */
1851 1851 return (0);
1852 1852
1853 1853 /* Strip case-insensitive bit, still present in log record */
1854 1854 txtype &= ~TX_CI;
1855 1855
1856 1856 if (txtype == 0 || txtype >= TX_MAX_TYPE)
1857 1857 return (zil_replay_error(zilog, lr, EINVAL));
1858 1858
1859 1859 /*
1860 1860 * If this record type can be logged out of order, the object
1861 1861 * (lr_foid) may no longer exist. That's legitimate, not an error.
1862 1862 */
1863 1863 if (TX_OOO(txtype)) {
1864 1864 error = dmu_object_info(zilog->zl_os,
1865 1865 ((lr_ooo_t *)lr)->lr_foid, NULL);
1866 1866 if (error == ENOENT || error == EEXIST)
1867 1867 return (0);
1868 1868 }
1869 1869
1870 1870 /*
1871 1871 * Make a copy of the data so we can revise and extend it.
1872 1872 */
1873 1873 bcopy(lr, zr->zr_lr, reclen);
1874 1874
1875 1875 /*
1876 1876 * If this is a TX_WRITE with a blkptr, suck in the data.
1877 1877 */
1878 1878 if (txtype == TX_WRITE && reclen == sizeof (lr_write_t)) {
1879 1879 error = zil_read_log_data(zilog, (lr_write_t *)lr,
1880 1880 zr->zr_lr + reclen);
1881 1881 if (error)
1882 1882 return (zil_replay_error(zilog, lr, error));
1883 1883 }
1884 1884
1885 1885 /*
1886 1886 * The log block containing this lr may have been byteswapped
1887 1887 * so that we can easily examine common fields like lrc_txtype.
1888 1888 * However, the log is a mix of different record types, and only the
1889 1889 * replay vectors know how to byteswap their records. Therefore, if
1890 1890 * the lr was byteswapped, undo it before invoking the replay vector.
1891 1891 */
1892 1892 if (zr->zr_byteswap)
1893 1893 byteswap_uint64_array(zr->zr_lr, reclen);
1894 1894
1895 1895 /*
1896 1896 * We must now do two things atomically: replay this log record,
1897 1897 * and update the log header sequence number to reflect the fact that
1898 1898 * we did so. At the end of each replay function the sequence number
1899 1899 * is updated if we are in replay mode.
1900 1900 */
1901 1901 error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lr, zr->zr_byteswap);
1902 1902 if (error) {
1903 1903 /*
1904 1904 * The DMU's dnode layer doesn't see removes until the txg
1905 1905 * commits, so a subsequent claim can spuriously fail with
1906 1906 * EEXIST. So if we receive any error we try syncing out
1907 1907 * any removes then retry the transaction. Note that we
1908 1908 * specify B_FALSE for byteswap now, so we don't do it twice.
1909 1909 */
1910 1910 txg_wait_synced(spa_get_dsl(zilog->zl_spa), 0);
1911 1911 error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lr, B_FALSE);
1912 1912 if (error)
1913 1913 return (zil_replay_error(zilog, lr, error));
1914 1914 }
1915 1915 return (0);
1916 1916 }
1917 1917
1918 1918 /* ARGSUSED */
1919 1919 static int
1920 1920 zil_incr_blks(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg)
1921 1921 {
1922 1922 zilog->zl_replay_blks++;
1923 1923
1924 1924 return (0);
1925 1925 }
1926 1926
1927 1927 /*
1928 1928 * If this dataset has a non-empty intent log, replay it and destroy it.
1929 1929 */
1930 1930 void
1931 1931 zil_replay(objset_t *os, void *arg, zil_replay_func_t *replay_func[TX_MAX_TYPE])
1932 1932 {
1933 1933 zilog_t *zilog = dmu_objset_zil(os);
1934 1934 const zil_header_t *zh = zilog->zl_header;
1935 1935 zil_replay_arg_t zr;
1936 1936
1937 1937 if ((zh->zh_flags & ZIL_REPLAY_NEEDED) == 0) {
1938 1938 zil_destroy(zilog, B_TRUE);
1939 1939 return;
1940 1940 }
1941 1941
1942 1942 zr.zr_replay = replay_func;
1943 1943 zr.zr_arg = arg;
1944 1944 zr.zr_byteswap = BP_SHOULD_BYTESWAP(&zh->zh_log);
1945 1945 zr.zr_lr = kmem_alloc(2 * SPA_MAXBLOCKSIZE, KM_SLEEP);
1946 1946
1947 1947 /*
1948 1948 * Wait for in-progress removes to sync before starting replay.
1949 1949 */
1950 1950 txg_wait_synced(zilog->zl_dmu_pool, 0);
1951 1951
1952 1952 zilog->zl_replay = B_TRUE;
1953 1953 zilog->zl_replay_time = ddi_get_lbolt();
1954 1954 ASSERT(zilog->zl_replay_blks == 0);
1955 1955 (void) zil_parse(zilog, zil_incr_blks, zil_replay_log_record, &zr,
1956 1956 zh->zh_claim_txg);
1957 1957 kmem_free(zr.zr_lr, 2 * SPA_MAXBLOCKSIZE);
1958 1958
1959 1959 zil_destroy(zilog, B_FALSE);
1960 1960 txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg);
1961 1961 zilog->zl_replay = B_FALSE;
1962 1962 }
1963 1963
1964 1964 boolean_t
1965 1965 zil_replaying(zilog_t *zilog, dmu_tx_t *tx)
1966 1966 {
1967 1967 if (zilog->zl_sync == ZFS_SYNC_DISABLED)
1968 1968 return (B_TRUE);
1969 1969
1970 1970 if (zilog->zl_replay) {
1971 1971 dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
1972 1972 zilog->zl_replayed_seq[dmu_tx_get_txg(tx) & TXG_MASK] =
1973 1973 zilog->zl_replaying_seq;
1974 1974 return (B_TRUE);
1975 1975 }
1976 1976
1977 1977 return (B_FALSE);
1978 1978 }
1979 1979
1980 1980 /* ARGSUSED */
1981 1981 int
1982 1982 zil_vdev_offline(const char *osname, void *arg)
1983 1983 {
1984 1984 objset_t *os;
1985 1985 zilog_t *zilog;
1986 1986 int error;
1987 1987
1988 1988 error = dmu_objset_hold(osname, FTAG, &os);
1989 1989 if (error)
1990 1990 return (error);
1991 1991
1992 1992 zilog = dmu_objset_zil(os);
1993 1993 if (zil_suspend(zilog) != 0)
1994 1994 error = EEXIST;
1995 1995 else
1996 1996 zil_resume(zilog);
1997 1997 dmu_objset_rele(os, FTAG);
1998 1998 return (error);
1999 1999 }
|
↓ open down ↓ |
924 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX