]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
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 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
cae5b340 AX |
23 | * Copyright (c) 2011, 2017 by Delphix. All rights reserved. |
24 | * Copyright (c) 2014 Integros [integros.com] | |
34dc7c2f BB |
25 | */ |
26 | ||
428870ff BB |
27 | /* Portions Copyright 2010 Robert Milkowski */ |
28 | ||
34dc7c2f BB |
29 | #include <sys/zfs_context.h> |
30 | #include <sys/spa.h> | |
31 | #include <sys/dmu.h> | |
32 | #include <sys/zap.h> | |
33 | #include <sys/arc.h> | |
34 | #include <sys/stat.h> | |
35 | #include <sys/resource.h> | |
36 | #include <sys/zil.h> | |
37 | #include <sys/zil_impl.h> | |
38 | #include <sys/dsl_dataset.h> | |
572e2857 | 39 | #include <sys/vdev_impl.h> |
34dc7c2f | 40 | #include <sys/dmu_tx.h> |
428870ff | 41 | #include <sys/dsl_pool.h> |
920dd524 | 42 | #include <sys/metaslab.h> |
ea04106b | 43 | #include <sys/trace_zil.h> |
cae5b340 | 44 | #include <sys/abd.h> |
34dc7c2f BB |
45 | |
46 | /* | |
47 | * The zfs intent log (ZIL) saves transaction records of system calls | |
48 | * that change the file system in memory with enough information | |
49 | * to be able to replay them. These are stored in memory until | |
50 | * either the DMU transaction group (txg) commits them to the stable pool | |
51 | * and they can be discarded, or they are flushed to the stable log | |
52 | * (also in the pool) due to a fsync, O_DSYNC or other synchronous | |
53 | * requirement. In the event of a panic or power fail then those log | |
54 | * records (transactions) are replayed. | |
55 | * | |
56 | * There is one ZIL per file system. Its on-disk (pool) format consists | |
57 | * of 3 parts: | |
58 | * | |
59 | * - ZIL header | |
60 | * - ZIL blocks | |
61 | * - ZIL records | |
62 | * | |
63 | * A log record holds a system call transaction. Log blocks can | |
64 | * hold many log records and the blocks are chained together. | |
65 | * Each ZIL block contains a block pointer (blkptr_t) to the next | |
66 | * ZIL block in the chain. The ZIL header points to the first | |
67 | * block in the chain. Note there is not a fixed place in the pool | |
68 | * to hold blocks. They are dynamically allocated and freed as | |
69 | * needed from the blocks available. Figure X shows the ZIL structure: | |
70 | */ | |
71 | ||
b6ad9671 ED |
72 | /* |
73 | * See zil.h for more information about these fields. | |
74 | */ | |
75 | zil_stats_t zil_stats = { | |
a08ee875 LG |
76 | { "zil_commit_count", KSTAT_DATA_UINT64 }, |
77 | { "zil_commit_writer_count", KSTAT_DATA_UINT64 }, | |
78 | { "zil_itx_count", KSTAT_DATA_UINT64 }, | |
79 | { "zil_itx_indirect_count", KSTAT_DATA_UINT64 }, | |
80 | { "zil_itx_indirect_bytes", KSTAT_DATA_UINT64 }, | |
81 | { "zil_itx_copied_count", KSTAT_DATA_UINT64 }, | |
82 | { "zil_itx_copied_bytes", KSTAT_DATA_UINT64 }, | |
83 | { "zil_itx_needcopy_count", KSTAT_DATA_UINT64 }, | |
84 | { "zil_itx_needcopy_bytes", KSTAT_DATA_UINT64 }, | |
85 | { "zil_itx_metaslab_normal_count", KSTAT_DATA_UINT64 }, | |
86 | { "zil_itx_metaslab_normal_bytes", KSTAT_DATA_UINT64 }, | |
87 | { "zil_itx_metaslab_slog_count", KSTAT_DATA_UINT64 }, | |
88 | { "zil_itx_metaslab_slog_bytes", KSTAT_DATA_UINT64 }, | |
b6ad9671 ED |
89 | }; |
90 | ||
91 | static kstat_t *zil_ksp; | |
92 | ||
34dc7c2f | 93 | /* |
a08ee875 | 94 | * Disable intent logging replay. This global ZIL switch affects all pools. |
34dc7c2f | 95 | */ |
a08ee875 | 96 | int zil_replay_disable = 0; |
34dc7c2f BB |
97 | |
98 | /* | |
99 | * Tunable parameter for debugging or performance analysis. Setting | |
100 | * zfs_nocacheflush will cause corruption on power loss if a volatile | |
101 | * out-of-order write cache is enabled. | |
102 | */ | |
c409e464 | 103 | int zfs_nocacheflush = 0; |
34dc7c2f | 104 | |
cae5b340 AX |
105 | /* |
106 | * Limit SLOG write size per commit executed with synchronous priority. | |
107 | * Any writes above that will be executed with lower (asynchronous) priority | |
108 | * to limit potential SLOG device abuse by single active ZIL writer. | |
109 | */ | |
110 | unsigned long zil_slog_bulk = 768 * 1024; | |
111 | ||
34dc7c2f BB |
112 | static kmem_cache_t *zil_lwb_cache; |
113 | ||
572e2857 | 114 | static void zil_async_to_sync(zilog_t *zilog, uint64_t foid); |
428870ff BB |
115 | |
116 | #define LWB_EMPTY(lwb) ((BP_GET_LSIZE(&lwb->lwb_blk) - \ | |
117 | sizeof (zil_chain_t)) == (lwb->lwb_sz - lwb->lwb_nused)) | |
118 | ||
34dc7c2f | 119 | static int |
428870ff | 120 | zil_bp_compare(const void *x1, const void *x2) |
34dc7c2f | 121 | { |
428870ff BB |
122 | const dva_t *dva1 = &((zil_bp_node_t *)x1)->zn_dva; |
123 | const dva_t *dva2 = &((zil_bp_node_t *)x2)->zn_dva; | |
34dc7c2f | 124 | |
cae5b340 AX |
125 | int cmp = AVL_CMP(DVA_GET_VDEV(dva1), DVA_GET_VDEV(dva2)); |
126 | if (likely(cmp)) | |
127 | return (cmp); | |
34dc7c2f | 128 | |
cae5b340 | 129 | return (AVL_CMP(DVA_GET_OFFSET(dva1), DVA_GET_OFFSET(dva2))); |
34dc7c2f BB |
130 | } |
131 | ||
132 | static void | |
428870ff | 133 | zil_bp_tree_init(zilog_t *zilog) |
34dc7c2f | 134 | { |
428870ff BB |
135 | avl_create(&zilog->zl_bp_tree, zil_bp_compare, |
136 | sizeof (zil_bp_node_t), offsetof(zil_bp_node_t, zn_node)); | |
34dc7c2f BB |
137 | } |
138 | ||
139 | static void | |
428870ff | 140 | zil_bp_tree_fini(zilog_t *zilog) |
34dc7c2f | 141 | { |
428870ff BB |
142 | avl_tree_t *t = &zilog->zl_bp_tree; |
143 | zil_bp_node_t *zn; | |
34dc7c2f BB |
144 | void *cookie = NULL; |
145 | ||
146 | while ((zn = avl_destroy_nodes(t, &cookie)) != NULL) | |
428870ff | 147 | kmem_free(zn, sizeof (zil_bp_node_t)); |
34dc7c2f BB |
148 | |
149 | avl_destroy(t); | |
150 | } | |
151 | ||
428870ff BB |
152 | int |
153 | zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp) | |
34dc7c2f | 154 | { |
428870ff | 155 | avl_tree_t *t = &zilog->zl_bp_tree; |
ea04106b | 156 | const dva_t *dva; |
428870ff | 157 | zil_bp_node_t *zn; |
34dc7c2f BB |
158 | avl_index_t where; |
159 | ||
ea04106b AX |
160 | if (BP_IS_EMBEDDED(bp)) |
161 | return (0); | |
162 | ||
163 | dva = BP_IDENTITY(bp); | |
164 | ||
34dc7c2f | 165 | if (avl_find(t, dva, &where) != NULL) |
a08ee875 | 166 | return (SET_ERROR(EEXIST)); |
34dc7c2f | 167 | |
ea04106b | 168 | zn = kmem_alloc(sizeof (zil_bp_node_t), KM_SLEEP); |
34dc7c2f BB |
169 | zn->zn_dva = *dva; |
170 | avl_insert(t, zn, where); | |
171 | ||
172 | return (0); | |
173 | } | |
174 | ||
175 | static zil_header_t * | |
176 | zil_header_in_syncing_context(zilog_t *zilog) | |
177 | { | |
178 | return ((zil_header_t *)zilog->zl_header); | |
179 | } | |
180 | ||
181 | static void | |
182 | zil_init_log_chain(zilog_t *zilog, blkptr_t *bp) | |
183 | { | |
184 | zio_cksum_t *zc = &bp->blk_cksum; | |
185 | ||
186 | zc->zc_word[ZIL_ZC_GUID_0] = spa_get_random(-1ULL); | |
187 | zc->zc_word[ZIL_ZC_GUID_1] = spa_get_random(-1ULL); | |
188 | zc->zc_word[ZIL_ZC_OBJSET] = dmu_objset_id(zilog->zl_os); | |
189 | zc->zc_word[ZIL_ZC_SEQ] = 1ULL; | |
190 | } | |
191 | ||
192 | /* | |
428870ff | 193 | * Read a log block and make sure it's valid. |
34dc7c2f BB |
194 | */ |
195 | static int | |
428870ff BB |
196 | zil_read_log_block(zilog_t *zilog, const blkptr_t *bp, blkptr_t *nbp, void *dst, |
197 | char **end) | |
34dc7c2f | 198 | { |
428870ff | 199 | enum zio_flag zio_flags = ZIO_FLAG_CANFAIL; |
e10b0808 | 200 | arc_flags_t aflags = ARC_FLAG_WAIT; |
428870ff | 201 | arc_buf_t *abuf = NULL; |
ea04106b | 202 | zbookmark_phys_t zb; |
34dc7c2f BB |
203 | int error; |
204 | ||
428870ff BB |
205 | if (zilog->zl_header->zh_claim_txg == 0) |
206 | zio_flags |= ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB; | |
34dc7c2f | 207 | |
428870ff BB |
208 | if (!(zilog->zl_header->zh_flags & ZIL_CLAIM_LR_SEQ_VALID)) |
209 | zio_flags |= ZIO_FLAG_SPECULATIVE; | |
34dc7c2f | 210 | |
428870ff BB |
211 | SET_BOOKMARK(&zb, bp->blk_cksum.zc_word[ZIL_ZC_OBJSET], |
212 | ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, bp->blk_cksum.zc_word[ZIL_ZC_SEQ]); | |
213 | ||
c06d4368 | 214 | error = arc_read(NULL, zilog->zl_spa, bp, arc_getbuf_func, &abuf, |
428870ff | 215 | ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb); |
34dc7c2f BB |
216 | |
217 | if (error == 0) { | |
34dc7c2f BB |
218 | zio_cksum_t cksum = bp->blk_cksum; |
219 | ||
220 | /* | |
b128c09f BB |
221 | * Validate the checksummed log block. |
222 | * | |
34dc7c2f BB |
223 | * Sequence numbers should be... sequential. The checksum |
224 | * verifier for the next block should be bp's checksum plus 1. | |
b128c09f BB |
225 | * |
226 | * Also check the log chain linkage and size used. | |
34dc7c2f BB |
227 | */ |
228 | cksum.zc_word[ZIL_ZC_SEQ]++; | |
229 | ||
428870ff BB |
230 | if (BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_ZILOG2) { |
231 | zil_chain_t *zilc = abuf->b_data; | |
232 | char *lr = (char *)(zilc + 1); | |
233 | uint64_t len = zilc->zc_nused - sizeof (zil_chain_t); | |
34dc7c2f | 234 | |
428870ff BB |
235 | if (bcmp(&cksum, &zilc->zc_next_blk.blk_cksum, |
236 | sizeof (cksum)) || BP_IS_HOLE(&zilc->zc_next_blk)) { | |
a08ee875 | 237 | error = SET_ERROR(ECKSUM); |
428870ff | 238 | } else { |
e10b0808 | 239 | ASSERT3U(len, <=, SPA_OLD_MAXBLOCKSIZE); |
428870ff BB |
240 | bcopy(lr, dst, len); |
241 | *end = (char *)dst + len; | |
242 | *nbp = zilc->zc_next_blk; | |
243 | } | |
244 | } else { | |
245 | char *lr = abuf->b_data; | |
246 | uint64_t size = BP_GET_LSIZE(bp); | |
247 | zil_chain_t *zilc = (zil_chain_t *)(lr + size) - 1; | |
248 | ||
249 | if (bcmp(&cksum, &zilc->zc_next_blk.blk_cksum, | |
250 | sizeof (cksum)) || BP_IS_HOLE(&zilc->zc_next_blk) || | |
251 | (zilc->zc_nused > (size - sizeof (*zilc)))) { | |
a08ee875 | 252 | error = SET_ERROR(ECKSUM); |
428870ff | 253 | } else { |
e10b0808 AX |
254 | ASSERT3U(zilc->zc_nused, <=, |
255 | SPA_OLD_MAXBLOCKSIZE); | |
428870ff BB |
256 | bcopy(lr, dst, zilc->zc_nused); |
257 | *end = (char *)dst + zilc->zc_nused; | |
258 | *nbp = zilc->zc_next_blk; | |
259 | } | |
34dc7c2f | 260 | } |
428870ff | 261 | |
cae5b340 | 262 | arc_buf_destroy(abuf, &abuf); |
428870ff BB |
263 | } |
264 | ||
265 | return (error); | |
266 | } | |
267 | ||
268 | /* | |
269 | * Read a TX_WRITE log data block. | |
270 | */ | |
271 | static int | |
272 | zil_read_log_data(zilog_t *zilog, const lr_write_t *lr, void *wbuf) | |
273 | { | |
274 | enum zio_flag zio_flags = ZIO_FLAG_CANFAIL; | |
275 | const blkptr_t *bp = &lr->lr_blkptr; | |
e10b0808 | 276 | arc_flags_t aflags = ARC_FLAG_WAIT; |
428870ff | 277 | arc_buf_t *abuf = NULL; |
ea04106b | 278 | zbookmark_phys_t zb; |
428870ff BB |
279 | int error; |
280 | ||
281 | if (BP_IS_HOLE(bp)) { | |
282 | if (wbuf != NULL) | |
283 | bzero(wbuf, MAX(BP_GET_LSIZE(bp), lr->lr_length)); | |
284 | return (0); | |
34dc7c2f BB |
285 | } |
286 | ||
428870ff BB |
287 | if (zilog->zl_header->zh_claim_txg == 0) |
288 | zio_flags |= ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB; | |
289 | ||
290 | SET_BOOKMARK(&zb, dmu_objset_id(zilog->zl_os), lr->lr_foid, | |
291 | ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp)); | |
292 | ||
c06d4368 | 293 | error = arc_read(NULL, zilog->zl_spa, bp, arc_getbuf_func, &abuf, |
428870ff BB |
294 | ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb); |
295 | ||
296 | if (error == 0) { | |
297 | if (wbuf != NULL) | |
298 | bcopy(abuf->b_data, wbuf, arc_buf_size(abuf)); | |
cae5b340 | 299 | arc_buf_destroy(abuf, &abuf); |
428870ff | 300 | } |
34dc7c2f BB |
301 | |
302 | return (error); | |
303 | } | |
304 | ||
305 | /* | |
306 | * Parse the intent log, and call parse_func for each valid record within. | |
34dc7c2f | 307 | */ |
428870ff | 308 | int |
34dc7c2f BB |
309 | zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func, |
310 | zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg) | |
311 | { | |
312 | const zil_header_t *zh = zilog->zl_header; | |
428870ff BB |
313 | boolean_t claimed = !!zh->zh_claim_txg; |
314 | uint64_t claim_blk_seq = claimed ? zh->zh_claim_blk_seq : UINT64_MAX; | |
315 | uint64_t claim_lr_seq = claimed ? zh->zh_claim_lr_seq : UINT64_MAX; | |
316 | uint64_t max_blk_seq = 0; | |
317 | uint64_t max_lr_seq = 0; | |
318 | uint64_t blk_count = 0; | |
319 | uint64_t lr_count = 0; | |
320 | blkptr_t blk, next_blk; | |
34dc7c2f | 321 | char *lrbuf, *lrp; |
428870ff | 322 | int error = 0; |
34dc7c2f | 323 | |
a08ee875 | 324 | bzero(&next_blk, sizeof (blkptr_t)); |
d4ed6673 | 325 | |
428870ff BB |
326 | /* |
327 | * Old logs didn't record the maximum zh_claim_lr_seq. | |
328 | */ | |
329 | if (!(zh->zh_flags & ZIL_CLAIM_LR_SEQ_VALID)) | |
330 | claim_lr_seq = UINT64_MAX; | |
34dc7c2f BB |
331 | |
332 | /* | |
333 | * Starting at the block pointed to by zh_log we read the log chain. | |
334 | * For each block in the chain we strongly check that block to | |
335 | * ensure its validity. We stop when an invalid block is found. | |
336 | * For each block pointer in the chain we call parse_blk_func(). | |
337 | * For each record in each valid block we call parse_lr_func(). | |
338 | * If the log has been claimed, stop if we encounter a sequence | |
339 | * number greater than the highest claimed sequence number. | |
340 | */ | |
e10b0808 | 341 | lrbuf = zio_buf_alloc(SPA_OLD_MAXBLOCKSIZE); |
428870ff | 342 | zil_bp_tree_init(zilog); |
34dc7c2f | 343 | |
428870ff BB |
344 | for (blk = zh->zh_log; !BP_IS_HOLE(&blk); blk = next_blk) { |
345 | uint64_t blk_seq = blk.blk_cksum.zc_word[ZIL_ZC_SEQ]; | |
346 | int reclen; | |
d4ed6673 | 347 | char *end = NULL; |
34dc7c2f | 348 | |
428870ff BB |
349 | if (blk_seq > claim_blk_seq) |
350 | break; | |
351 | if ((error = parse_blk_func(zilog, &blk, arg, txg)) != 0) | |
352 | break; | |
353 | ASSERT3U(max_blk_seq, <, blk_seq); | |
354 | max_blk_seq = blk_seq; | |
355 | blk_count++; | |
34dc7c2f | 356 | |
428870ff BB |
357 | if (max_lr_seq == claim_lr_seq && max_blk_seq == claim_blk_seq) |
358 | break; | |
34dc7c2f | 359 | |
428870ff | 360 | error = zil_read_log_block(zilog, &blk, &next_blk, lrbuf, &end); |
a08ee875 | 361 | if (error != 0) |
34dc7c2f BB |
362 | break; |
363 | ||
428870ff | 364 | for (lrp = lrbuf; lrp < end; lrp += reclen) { |
34dc7c2f BB |
365 | lr_t *lr = (lr_t *)lrp; |
366 | reclen = lr->lrc_reclen; | |
367 | ASSERT3U(reclen, >=, sizeof (lr_t)); | |
428870ff BB |
368 | if (lr->lrc_seq > claim_lr_seq) |
369 | goto done; | |
370 | if ((error = parse_lr_func(zilog, lr, arg, txg)) != 0) | |
371 | goto done; | |
372 | ASSERT3U(max_lr_seq, <, lr->lrc_seq); | |
373 | max_lr_seq = lr->lrc_seq; | |
374 | lr_count++; | |
34dc7c2f | 375 | } |
34dc7c2f | 376 | } |
428870ff BB |
377 | done: |
378 | zilog->zl_parse_error = error; | |
379 | zilog->zl_parse_blk_seq = max_blk_seq; | |
380 | zilog->zl_parse_lr_seq = max_lr_seq; | |
381 | zilog->zl_parse_blk_count = blk_count; | |
382 | zilog->zl_parse_lr_count = lr_count; | |
383 | ||
384 | ASSERT(!claimed || !(zh->zh_flags & ZIL_CLAIM_LR_SEQ_VALID) || | |
385 | (max_blk_seq == claim_blk_seq && max_lr_seq == claim_lr_seq)); | |
386 | ||
387 | zil_bp_tree_fini(zilog); | |
e10b0808 | 388 | zio_buf_free(lrbuf, SPA_OLD_MAXBLOCKSIZE); |
34dc7c2f | 389 | |
428870ff | 390 | return (error); |
34dc7c2f BB |
391 | } |
392 | ||
428870ff | 393 | static int |
34dc7c2f BB |
394 | zil_claim_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t first_txg) |
395 | { | |
34dc7c2f BB |
396 | /* |
397 | * Claim log block if not already committed and not already claimed. | |
428870ff | 398 | * If tx == NULL, just verify that the block is claimable. |
34dc7c2f | 399 | */ |
ea04106b AX |
400 | if (BP_IS_HOLE(bp) || bp->blk_birth < first_txg || |
401 | zil_bp_tree_add(zilog, bp) != 0) | |
428870ff BB |
402 | return (0); |
403 | ||
404 | return (zio_wait(zio_claim(NULL, zilog->zl_spa, | |
405 | tx == NULL ? 0 : first_txg, bp, spa_claim_notify, NULL, | |
406 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB))); | |
34dc7c2f BB |
407 | } |
408 | ||
428870ff | 409 | static int |
34dc7c2f BB |
410 | zil_claim_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t first_txg) |
411 | { | |
428870ff BB |
412 | lr_write_t *lr = (lr_write_t *)lrc; |
413 | int error; | |
414 | ||
415 | if (lrc->lrc_txtype != TX_WRITE) | |
416 | return (0); | |
417 | ||
418 | /* | |
419 | * If the block is not readable, don't claim it. This can happen | |
420 | * in normal operation when a log block is written to disk before | |
421 | * some of the dmu_sync() blocks it points to. In this case, the | |
422 | * transaction cannot have been committed to anyone (we would have | |
423 | * waited for all writes to be stable first), so it is semantically | |
424 | * correct to declare this the end of the log. | |
425 | */ | |
426 | if (lr->lr_blkptr.blk_birth >= first_txg && | |
427 | (error = zil_read_log_data(zilog, lr, NULL)) != 0) | |
428 | return (error); | |
429 | return (zil_claim_log_block(zilog, &lr->lr_blkptr, tx, first_txg)); | |
34dc7c2f BB |
430 | } |
431 | ||
432 | /* ARGSUSED */ | |
428870ff | 433 | static int |
34dc7c2f BB |
434 | zil_free_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t claim_txg) |
435 | { | |
428870ff BB |
436 | zio_free_zil(zilog->zl_spa, dmu_tx_get_txg(tx), bp); |
437 | ||
438 | return (0); | |
34dc7c2f BB |
439 | } |
440 | ||
428870ff | 441 | static int |
34dc7c2f BB |
442 | zil_free_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t claim_txg) |
443 | { | |
428870ff BB |
444 | lr_write_t *lr = (lr_write_t *)lrc; |
445 | blkptr_t *bp = &lr->lr_blkptr; | |
446 | ||
34dc7c2f BB |
447 | /* |
448 | * If we previously claimed it, we need to free it. | |
449 | */ | |
428870ff | 450 | if (claim_txg != 0 && lrc->lrc_txtype == TX_WRITE && |
ea04106b AX |
451 | bp->blk_birth >= claim_txg && zil_bp_tree_add(zilog, bp) == 0 && |
452 | !BP_IS_HOLE(bp)) | |
428870ff BB |
453 | zio_free(zilog->zl_spa, dmu_tx_get_txg(tx), bp); |
454 | ||
455 | return (0); | |
456 | } | |
457 | ||
458 | static lwb_t * | |
cae5b340 AX |
459 | zil_alloc_lwb(zilog_t *zilog, blkptr_t *bp, boolean_t slog, uint64_t txg, |
460 | boolean_t fastwrite) | |
428870ff BB |
461 | { |
462 | lwb_t *lwb; | |
463 | ||
ea04106b | 464 | lwb = kmem_cache_alloc(zil_lwb_cache, KM_SLEEP); |
428870ff BB |
465 | lwb->lwb_zilog = zilog; |
466 | lwb->lwb_blk = *bp; | |
920dd524 | 467 | lwb->lwb_fastwrite = fastwrite; |
cae5b340 | 468 | lwb->lwb_slog = slog; |
428870ff BB |
469 | lwb->lwb_buf = zio_buf_alloc(BP_GET_LSIZE(bp)); |
470 | lwb->lwb_max_txg = txg; | |
471 | lwb->lwb_zio = NULL; | |
472 | lwb->lwb_tx = NULL; | |
473 | if (BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_ZILOG2) { | |
474 | lwb->lwb_nused = sizeof (zil_chain_t); | |
475 | lwb->lwb_sz = BP_GET_LSIZE(bp); | |
476 | } else { | |
477 | lwb->lwb_nused = 0; | |
478 | lwb->lwb_sz = BP_GET_LSIZE(bp) - sizeof (zil_chain_t); | |
34dc7c2f | 479 | } |
428870ff BB |
480 | |
481 | mutex_enter(&zilog->zl_lock); | |
482 | list_insert_tail(&zilog->zl_lwb_list, lwb); | |
483 | mutex_exit(&zilog->zl_lock); | |
484 | ||
485 | return (lwb); | |
34dc7c2f BB |
486 | } |
487 | ||
29809a6c MA |
488 | /* |
489 | * Called when we create in-memory log transactions so that we know | |
490 | * to cleanup the itxs at the end of spa_sync(). | |
491 | */ | |
492 | void | |
493 | zilog_dirty(zilog_t *zilog, uint64_t txg) | |
494 | { | |
495 | dsl_pool_t *dp = zilog->zl_dmu_pool; | |
496 | dsl_dataset_t *ds = dmu_objset_ds(zilog->zl_os); | |
497 | ||
e10b0808 | 498 | if (ds->ds_is_snapshot) |
29809a6c MA |
499 | panic("dirtying snapshot!"); |
500 | ||
a08ee875 | 501 | if (txg_list_add(&dp->dp_dirty_zilogs, zilog, txg)) { |
29809a6c MA |
502 | /* up the hold count until we can be written out */ |
503 | dmu_buf_add_ref(ds->ds_dbuf, zilog); | |
504 | } | |
505 | } | |
506 | ||
cae5b340 AX |
507 | /* |
508 | * Determine if the zil is dirty in the specified txg. Callers wanting to | |
509 | * ensure that the dirty state does not change must hold the itxg_lock for | |
510 | * the specified txg. Holding the lock will ensure that the zil cannot be | |
511 | * dirtied (zil_itx_assign) or cleaned (zil_clean) while we check its current | |
512 | * state. | |
513 | */ | |
514 | boolean_t | |
515 | zilog_is_dirty_in_txg(zilog_t *zilog, uint64_t txg) | |
516 | { | |
517 | dsl_pool_t *dp = zilog->zl_dmu_pool; | |
518 | ||
519 | if (txg_list_member(&dp->dp_dirty_zilogs, zilog, txg & TXG_MASK)) | |
520 | return (B_TRUE); | |
521 | return (B_FALSE); | |
522 | } | |
523 | ||
524 | /* | |
525 | * Determine if the zil is dirty. The zil is considered dirty if it has | |
526 | * any pending itx records that have not been cleaned by zil_clean(). | |
527 | */ | |
29809a6c MA |
528 | boolean_t |
529 | zilog_is_dirty(zilog_t *zilog) | |
530 | { | |
531 | dsl_pool_t *dp = zilog->zl_dmu_pool; | |
532 | int t; | |
533 | ||
534 | for (t = 0; t < TXG_SIZE; t++) { | |
535 | if (txg_list_member(&dp->dp_dirty_zilogs, zilog, t)) | |
536 | return (B_TRUE); | |
537 | } | |
538 | return (B_FALSE); | |
539 | } | |
540 | ||
34dc7c2f BB |
541 | /* |
542 | * Create an on-disk intent log. | |
543 | */ | |
428870ff | 544 | static lwb_t * |
34dc7c2f BB |
545 | zil_create(zilog_t *zilog) |
546 | { | |
547 | const zil_header_t *zh = zilog->zl_header; | |
428870ff | 548 | lwb_t *lwb = NULL; |
34dc7c2f BB |
549 | uint64_t txg = 0; |
550 | dmu_tx_t *tx = NULL; | |
551 | blkptr_t blk; | |
552 | int error = 0; | |
920dd524 | 553 | boolean_t fastwrite = FALSE; |
cae5b340 | 554 | boolean_t slog = FALSE; |
34dc7c2f BB |
555 | |
556 | /* | |
557 | * Wait for any previous destroy to complete. | |
558 | */ | |
559 | txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg); | |
560 | ||
561 | ASSERT(zh->zh_claim_txg == 0); | |
562 | ASSERT(zh->zh_replay_seq == 0); | |
563 | ||
564 | blk = zh->zh_log; | |
565 | ||
566 | /* | |
428870ff BB |
567 | * Allocate an initial log block if: |
568 | * - there isn't one already | |
cae5b340 | 569 | * - the existing block is the wrong endianness |
34dc7c2f | 570 | */ |
fb5f0bc8 | 571 | if (BP_IS_HOLE(&blk) || BP_SHOULD_BYTESWAP(&blk)) { |
34dc7c2f | 572 | tx = dmu_tx_create(zilog->zl_os); |
428870ff | 573 | VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0); |
34dc7c2f BB |
574 | dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); |
575 | txg = dmu_tx_get_txg(tx); | |
576 | ||
fb5f0bc8 | 577 | if (!BP_IS_HOLE(&blk)) { |
428870ff | 578 | zio_free_zil(zilog->zl_spa, txg, &blk); |
fb5f0bc8 BB |
579 | BP_ZERO(&blk); |
580 | } | |
581 | ||
920dd524 | 582 | error = zio_alloc_zil(zilog->zl_spa, txg, &blk, |
cae5b340 | 583 | ZIL_MIN_BLKSZ, &slog); |
920dd524 | 584 | fastwrite = TRUE; |
34dc7c2f BB |
585 | |
586 | if (error == 0) | |
587 | zil_init_log_chain(zilog, &blk); | |
588 | } | |
589 | ||
590 | /* | |
591 | * Allocate a log write buffer (lwb) for the first log block. | |
592 | */ | |
428870ff | 593 | if (error == 0) |
cae5b340 | 594 | lwb = zil_alloc_lwb(zilog, &blk, slog, txg, fastwrite); |
34dc7c2f BB |
595 | |
596 | /* | |
597 | * If we just allocated the first log block, commit our transaction | |
598 | * and wait for zil_sync() to stuff the block poiner into zh_log. | |
599 | * (zh is part of the MOS, so we cannot modify it in open context.) | |
600 | */ | |
601 | if (tx != NULL) { | |
602 | dmu_tx_commit(tx); | |
603 | txg_wait_synced(zilog->zl_dmu_pool, txg); | |
604 | } | |
605 | ||
606 | ASSERT(bcmp(&blk, &zh->zh_log, sizeof (blk)) == 0); | |
428870ff BB |
607 | |
608 | return (lwb); | |
34dc7c2f BB |
609 | } |
610 | ||
611 | /* | |
612 | * In one tx, free all log blocks and clear the log header. | |
613 | * If keep_first is set, then we're replaying a log with no content. | |
614 | * We want to keep the first block, however, so that the first | |
615 | * synchronous transaction doesn't require a txg_wait_synced() | |
616 | * in zil_create(). We don't need to txg_wait_synced() here either | |
617 | * when keep_first is set, because both zil_create() and zil_destroy() | |
618 | * will wait for any in-progress destroys to complete. | |
619 | */ | |
620 | void | |
621 | zil_destroy(zilog_t *zilog, boolean_t keep_first) | |
622 | { | |
623 | const zil_header_t *zh = zilog->zl_header; | |
624 | lwb_t *lwb; | |
625 | dmu_tx_t *tx; | |
626 | uint64_t txg; | |
627 | ||
628 | /* | |
629 | * Wait for any previous destroy to complete. | |
630 | */ | |
631 | txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg); | |
632 | ||
428870ff BB |
633 | zilog->zl_old_header = *zh; /* debugging aid */ |
634 | ||
34dc7c2f BB |
635 | if (BP_IS_HOLE(&zh->zh_log)) |
636 | return; | |
637 | ||
638 | tx = dmu_tx_create(zilog->zl_os); | |
428870ff | 639 | VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0); |
34dc7c2f BB |
640 | dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); |
641 | txg = dmu_tx_get_txg(tx); | |
642 | ||
643 | mutex_enter(&zilog->zl_lock); | |
644 | ||
34dc7c2f BB |
645 | ASSERT3U(zilog->zl_destroy_txg, <, txg); |
646 | zilog->zl_destroy_txg = txg; | |
647 | zilog->zl_keep_first = keep_first; | |
648 | ||
649 | if (!list_is_empty(&zilog->zl_lwb_list)) { | |
650 | ASSERT(zh->zh_claim_txg == 0); | |
3e31d2b0 | 651 | VERIFY(!keep_first); |
34dc7c2f | 652 | while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) { |
920dd524 ED |
653 | ASSERT(lwb->lwb_zio == NULL); |
654 | if (lwb->lwb_fastwrite) | |
655 | metaslab_fastwrite_unmark(zilog->zl_spa, | |
656 | &lwb->lwb_blk); | |
34dc7c2f BB |
657 | list_remove(&zilog->zl_lwb_list, lwb); |
658 | if (lwb->lwb_buf != NULL) | |
659 | zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); | |
428870ff | 660 | zio_free_zil(zilog->zl_spa, txg, &lwb->lwb_blk); |
34dc7c2f BB |
661 | kmem_cache_free(zil_lwb_cache, lwb); |
662 | } | |
428870ff | 663 | } else if (!keep_first) { |
29809a6c | 664 | zil_destroy_sync(zilog, tx); |
34dc7c2f BB |
665 | } |
666 | mutex_exit(&zilog->zl_lock); | |
667 | ||
668 | dmu_tx_commit(tx); | |
669 | } | |
670 | ||
29809a6c MA |
671 | void |
672 | zil_destroy_sync(zilog_t *zilog, dmu_tx_t *tx) | |
673 | { | |
674 | ASSERT(list_is_empty(&zilog->zl_lwb_list)); | |
675 | (void) zil_parse(zilog, zil_free_log_block, | |
676 | zil_free_log_record, tx, zilog->zl_header->zh_claim_txg); | |
677 | } | |
678 | ||
34dc7c2f | 679 | int |
e10b0808 | 680 | zil_claim(dsl_pool_t *dp, dsl_dataset_t *ds, void *txarg) |
34dc7c2f BB |
681 | { |
682 | dmu_tx_t *tx = txarg; | |
683 | uint64_t first_txg = dmu_tx_get_txg(tx); | |
684 | zilog_t *zilog; | |
685 | zil_header_t *zh; | |
686 | objset_t *os; | |
687 | int error; | |
688 | ||
e10b0808 AX |
689 | error = dmu_objset_own_obj(dp, ds->ds_object, |
690 | DMU_OST_ANY, B_FALSE, FTAG, &os); | |
a08ee875 | 691 | if (error != 0) { |
ea04106b AX |
692 | /* |
693 | * EBUSY indicates that the objset is inconsistent, in which | |
694 | * case it can not have a ZIL. | |
695 | */ | |
696 | if (error != EBUSY) { | |
e10b0808 AX |
697 | cmn_err(CE_WARN, "can't open objset for %llu, error %u", |
698 | (unsigned long long)ds->ds_object, error); | |
ea04106b AX |
699 | } |
700 | ||
34dc7c2f BB |
701 | return (0); |
702 | } | |
703 | ||
704 | zilog = dmu_objset_zil(os); | |
705 | zh = zil_header_in_syncing_context(zilog); | |
706 | ||
428870ff | 707 | if (spa_get_log_state(zilog->zl_spa) == SPA_LOG_CLEAR) { |
9babb374 | 708 | if (!BP_IS_HOLE(&zh->zh_log)) |
428870ff | 709 | zio_free_zil(zilog->zl_spa, first_txg, &zh->zh_log); |
9babb374 BB |
710 | BP_ZERO(&zh->zh_log); |
711 | dsl_dataset_dirty(dmu_objset_ds(os), tx); | |
a08ee875 | 712 | dmu_objset_disown(os, FTAG); |
428870ff | 713 | return (0); |
9babb374 BB |
714 | } |
715 | ||
34dc7c2f BB |
716 | /* |
717 | * Claim all log blocks if we haven't already done so, and remember | |
718 | * the highest claimed sequence number. This ensures that if we can | |
719 | * read only part of the log now (e.g. due to a missing device), | |
720 | * but we can read the entire log later, we will not try to replay | |
721 | * or destroy beyond the last block we successfully claimed. | |
722 | */ | |
723 | ASSERT3U(zh->zh_claim_txg, <=, first_txg); | |
724 | if (zh->zh_claim_txg == 0 && !BP_IS_HOLE(&zh->zh_log)) { | |
428870ff | 725 | (void) zil_parse(zilog, zil_claim_log_block, |
34dc7c2f | 726 | zil_claim_log_record, tx, first_txg); |
428870ff BB |
727 | zh->zh_claim_txg = first_txg; |
728 | zh->zh_claim_blk_seq = zilog->zl_parse_blk_seq; | |
729 | zh->zh_claim_lr_seq = zilog->zl_parse_lr_seq; | |
730 | if (zilog->zl_parse_lr_count || zilog->zl_parse_blk_count > 1) | |
731 | zh->zh_flags |= ZIL_REPLAY_NEEDED; | |
732 | zh->zh_flags |= ZIL_CLAIM_LR_SEQ_VALID; | |
34dc7c2f BB |
733 | dsl_dataset_dirty(dmu_objset_ds(os), tx); |
734 | } | |
735 | ||
736 | ASSERT3U(first_txg, ==, (spa_last_synced_txg(zilog->zl_spa) + 1)); | |
a08ee875 | 737 | dmu_objset_disown(os, FTAG); |
34dc7c2f BB |
738 | return (0); |
739 | } | |
740 | ||
b128c09f BB |
741 | /* |
742 | * Check the log by walking the log chain. | |
743 | * Checksum errors are ok as they indicate the end of the chain. | |
744 | * Any other error (no device or read failure) returns an error. | |
745 | */ | |
e10b0808 | 746 | /* ARGSUSED */ |
b128c09f | 747 | int |
e10b0808 | 748 | zil_check_log_chain(dsl_pool_t *dp, dsl_dataset_t *ds, void *tx) |
b128c09f BB |
749 | { |
750 | zilog_t *zilog; | |
b128c09f | 751 | objset_t *os; |
572e2857 | 752 | blkptr_t *bp; |
b128c09f BB |
753 | int error; |
754 | ||
428870ff BB |
755 | ASSERT(tx == NULL); |
756 | ||
e10b0808 | 757 | error = dmu_objset_from_ds(ds, &os); |
a08ee875 | 758 | if (error != 0) { |
e10b0808 AX |
759 | cmn_err(CE_WARN, "can't open objset %llu, error %d", |
760 | (unsigned long long)ds->ds_object, error); | |
b128c09f BB |
761 | return (0); |
762 | } | |
763 | ||
764 | zilog = dmu_objset_zil(os); | |
572e2857 BB |
765 | bp = (blkptr_t *)&zilog->zl_header->zh_log; |
766 | ||
767 | /* | |
768 | * Check the first block and determine if it's on a log device | |
769 | * which may have been removed or faulted prior to loading this | |
770 | * pool. If so, there's no point in checking the rest of the log | |
771 | * as its content should have already been synced to the pool. | |
772 | */ | |
773 | if (!BP_IS_HOLE(bp)) { | |
774 | vdev_t *vd; | |
775 | boolean_t valid = B_TRUE; | |
776 | ||
777 | spa_config_enter(os->os_spa, SCL_STATE, FTAG, RW_READER); | |
778 | vd = vdev_lookup_top(os->os_spa, DVA_GET_VDEV(&bp->blk_dva[0])); | |
779 | if (vd->vdev_islog && vdev_is_dead(vd)) | |
780 | valid = vdev_log_state_valid(vd); | |
781 | spa_config_exit(os->os_spa, SCL_STATE, FTAG); | |
782 | ||
e10b0808 | 783 | if (!valid) |
572e2857 | 784 | return (0); |
572e2857 | 785 | } |
b128c09f | 786 | |
428870ff BB |
787 | /* |
788 | * Because tx == NULL, zil_claim_log_block() will not actually claim | |
789 | * any blocks, but just determine whether it is possible to do so. | |
790 | * In addition to checking the log chain, zil_claim_log_block() | |
791 | * will invoke zio_claim() with a done func of spa_claim_notify(), | |
792 | * which will update spa_max_claim_txg. See spa_load() for details. | |
793 | */ | |
794 | error = zil_parse(zilog, zil_claim_log_block, zil_claim_log_record, tx, | |
795 | zilog->zl_header->zh_claim_txg ? -1ULL : spa_first_txg(os->os_spa)); | |
796 | ||
428870ff | 797 | return ((error == ECKSUM || error == ENOENT) ? 0 : error); |
b128c09f BB |
798 | } |
799 | ||
34dc7c2f BB |
800 | static int |
801 | zil_vdev_compare(const void *x1, const void *x2) | |
802 | { | |
572e2857 BB |
803 | const uint64_t v1 = ((zil_vdev_node_t *)x1)->zv_vdev; |
804 | const uint64_t v2 = ((zil_vdev_node_t *)x2)->zv_vdev; | |
34dc7c2f | 805 | |
cae5b340 | 806 | return (AVL_CMP(v1, v2)); |
34dc7c2f BB |
807 | } |
808 | ||
809 | void | |
428870ff | 810 | zil_add_block(zilog_t *zilog, const blkptr_t *bp) |
34dc7c2f BB |
811 | { |
812 | avl_tree_t *t = &zilog->zl_vdev_tree; | |
813 | avl_index_t where; | |
814 | zil_vdev_node_t *zv, zvsearch; | |
815 | int ndvas = BP_GET_NDVAS(bp); | |
816 | int i; | |
817 | ||
818 | if (zfs_nocacheflush) | |
819 | return; | |
820 | ||
821 | ASSERT(zilog->zl_writer); | |
822 | ||
823 | /* | |
824 | * Even though we're zl_writer, we still need a lock because the | |
825 | * zl_get_data() callbacks may have dmu_sync() done callbacks | |
826 | * that will run concurrently. | |
827 | */ | |
828 | mutex_enter(&zilog->zl_vdev_lock); | |
829 | for (i = 0; i < ndvas; i++) { | |
830 | zvsearch.zv_vdev = DVA_GET_VDEV(&bp->blk_dva[i]); | |
831 | if (avl_find(t, &zvsearch, &where) == NULL) { | |
ea04106b | 832 | zv = kmem_alloc(sizeof (*zv), KM_SLEEP); |
34dc7c2f BB |
833 | zv->zv_vdev = zvsearch.zv_vdev; |
834 | avl_insert(t, zv, where); | |
835 | } | |
836 | } | |
837 | mutex_exit(&zilog->zl_vdev_lock); | |
838 | } | |
839 | ||
572e2857 | 840 | static void |
34dc7c2f BB |
841 | zil_flush_vdevs(zilog_t *zilog) |
842 | { | |
843 | spa_t *spa = zilog->zl_spa; | |
844 | avl_tree_t *t = &zilog->zl_vdev_tree; | |
845 | void *cookie = NULL; | |
846 | zil_vdev_node_t *zv; | |
847 | zio_t *zio; | |
848 | ||
849 | ASSERT(zilog->zl_writer); | |
850 | ||
851 | /* | |
852 | * We don't need zl_vdev_lock here because we're the zl_writer, | |
853 | * and all zl_get_data() callbacks are done. | |
854 | */ | |
855 | if (avl_numnodes(t) == 0) | |
856 | return; | |
857 | ||
b128c09f | 858 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); |
34dc7c2f | 859 | |
b128c09f | 860 | zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL); |
34dc7c2f BB |
861 | |
862 | while ((zv = avl_destroy_nodes(t, &cookie)) != NULL) { | |
863 | vdev_t *vd = vdev_lookup_top(spa, zv->zv_vdev); | |
864 | if (vd != NULL) | |
865 | zio_flush(zio, vd); | |
866 | kmem_free(zv, sizeof (*zv)); | |
867 | } | |
868 | ||
869 | /* | |
870 | * Wait for all the flushes to complete. Not all devices actually | |
871 | * support the DKIOCFLUSHWRITECACHE ioctl, so it's OK if it fails. | |
872 | */ | |
873 | (void) zio_wait(zio); | |
874 | ||
b128c09f | 875 | spa_config_exit(spa, SCL_STATE, FTAG); |
34dc7c2f BB |
876 | } |
877 | ||
878 | /* | |
879 | * Function called when a log block write completes | |
880 | */ | |
881 | static void | |
882 | zil_lwb_write_done(zio_t *zio) | |
883 | { | |
884 | lwb_t *lwb = zio->io_private; | |
885 | zilog_t *zilog = lwb->lwb_zilog; | |
428870ff | 886 | dmu_tx_t *tx = lwb->lwb_tx; |
34dc7c2f | 887 | |
b128c09f | 888 | ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF); |
b128c09f BB |
889 | ASSERT(BP_GET_TYPE(zio->io_bp) == DMU_OT_INTENT_LOG); |
890 | ASSERT(BP_GET_LEVEL(zio->io_bp) == 0); | |
891 | ASSERT(BP_GET_BYTEORDER(zio->io_bp) == ZFS_HOST_BYTEORDER); | |
892 | ASSERT(!BP_IS_GANG(zio->io_bp)); | |
893 | ASSERT(!BP_IS_HOLE(zio->io_bp)); | |
ea04106b | 894 | ASSERT(BP_GET_FILL(zio->io_bp) == 0); |
b128c09f | 895 | |
34dc7c2f | 896 | /* |
9babb374 BB |
897 | * Ensure the lwb buffer pointer is cleared before releasing |
898 | * the txg. If we have had an allocation failure and | |
899 | * the txg is waiting to sync then we want want zil_sync() | |
900 | * to remove the lwb so that it's not picked up as the next new | |
901 | * one in zil_commit_writer(). zil_sync() will only remove | |
902 | * the lwb if lwb_buf is null. | |
34dc7c2f | 903 | */ |
cae5b340 | 904 | abd_put(zio->io_abd); |
34dc7c2f BB |
905 | zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); |
906 | mutex_enter(&zilog->zl_lock); | |
920dd524 ED |
907 | lwb->lwb_zio = NULL; |
908 | lwb->lwb_fastwrite = FALSE; | |
34dc7c2f | 909 | lwb->lwb_buf = NULL; |
428870ff BB |
910 | lwb->lwb_tx = NULL; |
911 | mutex_exit(&zilog->zl_lock); | |
9babb374 BB |
912 | |
913 | /* | |
914 | * Now that we've written this log block, we have a stable pointer | |
915 | * to the next block in the chain, so it's OK to let the txg in | |
428870ff | 916 | * which we allocated the next block sync. |
9babb374 | 917 | */ |
428870ff | 918 | dmu_tx_commit(tx); |
34dc7c2f BB |
919 | } |
920 | ||
921 | /* | |
922 | * Initialize the io for a log block. | |
34dc7c2f BB |
923 | */ |
924 | static void | |
925 | zil_lwb_write_init(zilog_t *zilog, lwb_t *lwb) | |
926 | { | |
ea04106b | 927 | zbookmark_phys_t zb; |
cae5b340 | 928 | zio_priority_t prio; |
34dc7c2f | 929 | |
428870ff BB |
930 | SET_BOOKMARK(&zb, lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_OBJSET], |
931 | ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, | |
932 | lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_SEQ]); | |
34dc7c2f BB |
933 | |
934 | if (zilog->zl_root_zio == NULL) { | |
935 | zilog->zl_root_zio = zio_root(zilog->zl_spa, NULL, NULL, | |
936 | ZIO_FLAG_CANFAIL); | |
937 | } | |
920dd524 ED |
938 | |
939 | /* Lock so zil_sync() doesn't fastwrite_unmark after zio is created */ | |
940 | mutex_enter(&zilog->zl_lock); | |
34dc7c2f | 941 | if (lwb->lwb_zio == NULL) { |
cae5b340 AX |
942 | abd_t *lwb_abd = abd_get_from_buf(lwb->lwb_buf, |
943 | BP_GET_LSIZE(&lwb->lwb_blk)); | |
920dd524 ED |
944 | if (!lwb->lwb_fastwrite) { |
945 | metaslab_fastwrite_mark(zilog->zl_spa, &lwb->lwb_blk); | |
946 | lwb->lwb_fastwrite = 1; | |
947 | } | |
cae5b340 AX |
948 | if (!lwb->lwb_slog || zilog->zl_cur_used <= zil_slog_bulk) |
949 | prio = ZIO_PRIORITY_SYNC_WRITE; | |
950 | else | |
951 | prio = ZIO_PRIORITY_ASYNC_WRITE; | |
34dc7c2f | 952 | lwb->lwb_zio = zio_rewrite(zilog->zl_root_zio, zilog->zl_spa, |
cae5b340 AX |
953 | 0, &lwb->lwb_blk, lwb_abd, BP_GET_LSIZE(&lwb->lwb_blk), |
954 | zil_lwb_write_done, lwb, prio, | |
920dd524 ED |
955 | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | |
956 | ZIO_FLAG_FASTWRITE, &zb); | |
34dc7c2f | 957 | } |
920dd524 | 958 | mutex_exit(&zilog->zl_lock); |
34dc7c2f BB |
959 | } |
960 | ||
428870ff BB |
961 | /* |
962 | * Define a limited set of intent log block sizes. | |
a08ee875 | 963 | * |
428870ff BB |
964 | * These must be a multiple of 4KB. Note only the amount used (again |
965 | * aligned to 4KB) actually gets written. However, we can't always just | |
e10b0808 | 966 | * allocate SPA_OLD_MAXBLOCKSIZE as the slog space could be exhausted. |
428870ff BB |
967 | */ |
968 | uint64_t zil_block_buckets[] = { | |
969 | 4096, /* non TX_WRITE */ | |
970 | 8192+4096, /* data base */ | |
971 | 32*1024 + 4096, /* NFS writes */ | |
972 | UINT64_MAX | |
973 | }; | |
974 | ||
34dc7c2f BB |
975 | /* |
976 | * Start a log block write and advance to the next log block. | |
977 | * Calls are serialized. | |
978 | */ | |
979 | static lwb_t * | |
980 | zil_lwb_write_start(zilog_t *zilog, lwb_t *lwb) | |
981 | { | |
428870ff BB |
982 | lwb_t *nlwb = NULL; |
983 | zil_chain_t *zilc; | |
34dc7c2f | 984 | spa_t *spa = zilog->zl_spa; |
428870ff BB |
985 | blkptr_t *bp; |
986 | dmu_tx_t *tx; | |
34dc7c2f | 987 | uint64_t txg; |
428870ff BB |
988 | uint64_t zil_blksz, wsz; |
989 | int i, error; | |
cae5b340 | 990 | boolean_t slog; |
428870ff BB |
991 | |
992 | if (BP_GET_CHECKSUM(&lwb->lwb_blk) == ZIO_CHECKSUM_ZILOG2) { | |
993 | zilc = (zil_chain_t *)lwb->lwb_buf; | |
994 | bp = &zilc->zc_next_blk; | |
995 | } else { | |
996 | zilc = (zil_chain_t *)(lwb->lwb_buf + lwb->lwb_sz); | |
997 | bp = &zilc->zc_next_blk; | |
998 | } | |
34dc7c2f | 999 | |
428870ff | 1000 | ASSERT(lwb->lwb_nused <= lwb->lwb_sz); |
34dc7c2f BB |
1001 | |
1002 | /* | |
1003 | * Allocate the next block and save its address in this block | |
1004 | * before writing it in order to establish the log chain. | |
1005 | * Note that if the allocation of nlwb synced before we wrote | |
1006 | * the block that points at it (lwb), we'd leak it if we crashed. | |
428870ff BB |
1007 | * Therefore, we don't do dmu_tx_commit() until zil_lwb_write_done(). |
1008 | * We dirty the dataset to ensure that zil_sync() will be called | |
1009 | * to clean up in the event of allocation failure or I/O failure. | |
34dc7c2f | 1010 | */ |
428870ff BB |
1011 | tx = dmu_tx_create(zilog->zl_os); |
1012 | VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0); | |
1013 | dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); | |
1014 | txg = dmu_tx_get_txg(tx); | |
1015 | ||
1016 | lwb->lwb_tx = tx; | |
34dc7c2f BB |
1017 | |
1018 | /* | |
428870ff BB |
1019 | * Log blocks are pre-allocated. Here we select the size of the next |
1020 | * block, based on size used in the last block. | |
1021 | * - first find the smallest bucket that will fit the block from a | |
1022 | * limited set of block sizes. This is because it's faster to write | |
1023 | * blocks allocated from the same metaslab as they are adjacent or | |
1024 | * close. | |
1025 | * - next find the maximum from the new suggested size and an array of | |
1026 | * previous sizes. This lessens a picket fence effect of wrongly | |
1027 | * guesssing the size if we have a stream of say 2k, 64k, 2k, 64k | |
1028 | * requests. | |
1029 | * | |
1030 | * Note we only write what is used, but we can't just allocate | |
1031 | * the maximum block size because we can exhaust the available | |
1032 | * pool log space. | |
34dc7c2f | 1033 | */ |
428870ff BB |
1034 | zil_blksz = zilog->zl_cur_used + sizeof (zil_chain_t); |
1035 | for (i = 0; zil_blksz > zil_block_buckets[i]; i++) | |
1036 | continue; | |
1037 | zil_blksz = zil_block_buckets[i]; | |
1038 | if (zil_blksz == UINT64_MAX) | |
e10b0808 | 1039 | zil_blksz = SPA_OLD_MAXBLOCKSIZE; |
428870ff BB |
1040 | zilog->zl_prev_blks[zilog->zl_prev_rotor] = zil_blksz; |
1041 | for (i = 0; i < ZIL_PREV_BLKS; i++) | |
1042 | zil_blksz = MAX(zil_blksz, zilog->zl_prev_blks[i]); | |
1043 | zilog->zl_prev_rotor = (zilog->zl_prev_rotor + 1) & (ZIL_PREV_BLKS - 1); | |
34dc7c2f BB |
1044 | |
1045 | BP_ZERO(bp); | |
cae5b340 AX |
1046 | error = zio_alloc_zil(spa, txg, bp, zil_blksz, &slog); |
1047 | if (slog) { | |
b6ad9671 ED |
1048 | ZIL_STAT_BUMP(zil_itx_metaslab_slog_count); |
1049 | ZIL_STAT_INCR(zil_itx_metaslab_slog_bytes, lwb->lwb_nused); | |
a08ee875 | 1050 | } else { |
b6ad9671 ED |
1051 | ZIL_STAT_BUMP(zil_itx_metaslab_normal_count); |
1052 | ZIL_STAT_INCR(zil_itx_metaslab_normal_bytes, lwb->lwb_nused); | |
1053 | } | |
a08ee875 | 1054 | if (error == 0) { |
428870ff BB |
1055 | ASSERT3U(bp->blk_birth, ==, txg); |
1056 | bp->blk_cksum = lwb->lwb_blk.blk_cksum; | |
1057 | bp->blk_cksum.zc_word[ZIL_ZC_SEQ]++; | |
34dc7c2f BB |
1058 | |
1059 | /* | |
428870ff | 1060 | * Allocate a new log write buffer (lwb). |
34dc7c2f | 1061 | */ |
cae5b340 | 1062 | nlwb = zil_alloc_lwb(zilog, bp, slog, txg, TRUE); |
34dc7c2f | 1063 | |
428870ff BB |
1064 | /* Record the block for later vdev flushing */ |
1065 | zil_add_block(zilog, &lwb->lwb_blk); | |
34dc7c2f BB |
1066 | } |
1067 | ||
428870ff BB |
1068 | if (BP_GET_CHECKSUM(&lwb->lwb_blk) == ZIO_CHECKSUM_ZILOG2) { |
1069 | /* For Slim ZIL only write what is used. */ | |
1070 | wsz = P2ROUNDUP_TYPED(lwb->lwb_nused, ZIL_MIN_BLKSZ, uint64_t); | |
1071 | ASSERT3U(wsz, <=, lwb->lwb_sz); | |
1072 | zio_shrink(lwb->lwb_zio, wsz); | |
34dc7c2f | 1073 | |
428870ff BB |
1074 | } else { |
1075 | wsz = lwb->lwb_sz; | |
1076 | } | |
34dc7c2f | 1077 | |
428870ff BB |
1078 | zilc->zc_pad = 0; |
1079 | zilc->zc_nused = lwb->lwb_nused; | |
1080 | zilc->zc_eck.zec_cksum = lwb->lwb_blk.blk_cksum; | |
34dc7c2f BB |
1081 | |
1082 | /* | |
428870ff | 1083 | * clear unused data for security |
34dc7c2f | 1084 | */ |
428870ff | 1085 | bzero(lwb->lwb_buf + lwb->lwb_nused, wsz - lwb->lwb_nused); |
34dc7c2f | 1086 | |
428870ff | 1087 | zio_nowait(lwb->lwb_zio); /* Kick off the write for the old log block */ |
34dc7c2f BB |
1088 | |
1089 | /* | |
428870ff BB |
1090 | * If there was an allocation failure then nlwb will be null which |
1091 | * forces a txg_wait_synced(). | |
34dc7c2f | 1092 | */ |
34dc7c2f BB |
1093 | return (nlwb); |
1094 | } | |
1095 | ||
1096 | static lwb_t * | |
1097 | zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb) | |
1098 | { | |
cae5b340 AX |
1099 | lr_t *lrcb, *lrc; |
1100 | lr_write_t *lrwb, *lrw; | |
428870ff | 1101 | char *lr_buf; |
cae5b340 | 1102 | uint64_t dlen, dnow, lwb_sp, reclen, txg; |
34dc7c2f BB |
1103 | |
1104 | if (lwb == NULL) | |
1105 | return (NULL); | |
428870ff | 1106 | |
34dc7c2f BB |
1107 | ASSERT(lwb->lwb_buf != NULL); |
1108 | ||
cae5b340 AX |
1109 | lrc = &itx->itx_lr; /* Common log record inside itx. */ |
1110 | lrw = (lr_write_t *)lrc; /* Write log record inside itx. */ | |
1111 | if (lrc->lrc_txtype == TX_WRITE && itx->itx_wr_state == WR_NEED_COPY) { | |
34dc7c2f | 1112 | dlen = P2ROUNDUP_TYPED( |
428870ff | 1113 | lrw->lr_length, sizeof (uint64_t), uint64_t); |
cae5b340 AX |
1114 | } else { |
1115 | dlen = 0; | |
1116 | } | |
1117 | reclen = lrc->lrc_reclen; | |
34dc7c2f | 1118 | zilog->zl_cur_used += (reclen + dlen); |
cae5b340 | 1119 | txg = lrc->lrc_txg; |
34dc7c2f BB |
1120 | |
1121 | zil_lwb_write_init(zilog, lwb); | |
1122 | ||
cae5b340 | 1123 | cont: |
34dc7c2f BB |
1124 | /* |
1125 | * If this record won't fit in the current log block, start a new one. | |
cae5b340 | 1126 | * For WR_NEED_COPY optimize layout for minimal number of chunks. |
34dc7c2f | 1127 | */ |
cae5b340 AX |
1128 | lwb_sp = lwb->lwb_sz - lwb->lwb_nused; |
1129 | if (reclen > lwb_sp || (reclen + dlen > lwb_sp && | |
1130 | lwb_sp < ZIL_MAX_WASTE_SPACE && (dlen % ZIL_MAX_LOG_DATA == 0 || | |
1131 | lwb_sp < reclen + dlen % ZIL_MAX_LOG_DATA))) { | |
34dc7c2f BB |
1132 | lwb = zil_lwb_write_start(zilog, lwb); |
1133 | if (lwb == NULL) | |
1134 | return (NULL); | |
1135 | zil_lwb_write_init(zilog, lwb); | |
428870ff | 1136 | ASSERT(LWB_EMPTY(lwb)); |
cae5b340 AX |
1137 | lwb_sp = lwb->lwb_sz - lwb->lwb_nused; |
1138 | ASSERT3U(reclen + MIN(dlen, sizeof (uint64_t)), <=, lwb_sp); | |
34dc7c2f BB |
1139 | } |
1140 | ||
cae5b340 | 1141 | dnow = MIN(dlen, lwb_sp - reclen); |
428870ff BB |
1142 | lr_buf = lwb->lwb_buf + lwb->lwb_nused; |
1143 | bcopy(lrc, lr_buf, reclen); | |
cae5b340 AX |
1144 | lrcb = (lr_t *)lr_buf; /* Like lrc, but inside lwb. */ |
1145 | lrwb = (lr_write_t *)lrcb; /* Like lrw, but inside lwb. */ | |
34dc7c2f | 1146 | |
b6ad9671 ED |
1147 | ZIL_STAT_BUMP(zil_itx_count); |
1148 | ||
34dc7c2f BB |
1149 | /* |
1150 | * If it's a write, fetch the data or get its blkptr as appropriate. | |
1151 | */ | |
1152 | if (lrc->lrc_txtype == TX_WRITE) { | |
1153 | if (txg > spa_freeze_txg(zilog->zl_spa)) | |
1154 | txg_wait_synced(zilog->zl_dmu_pool, txg); | |
b6ad9671 ED |
1155 | if (itx->itx_wr_state == WR_COPIED) { |
1156 | ZIL_STAT_BUMP(zil_itx_copied_count); | |
1157 | ZIL_STAT_INCR(zil_itx_copied_bytes, lrw->lr_length); | |
1158 | } else { | |
34dc7c2f BB |
1159 | char *dbuf; |
1160 | int error; | |
1161 | ||
cae5b340 | 1162 | if (itx->itx_wr_state == WR_NEED_COPY) { |
428870ff | 1163 | dbuf = lr_buf + reclen; |
cae5b340 AX |
1164 | lrcb->lrc_reclen += dnow; |
1165 | if (lrwb->lr_length > dnow) | |
1166 | lrwb->lr_length = dnow; | |
1167 | lrw->lr_offset += dnow; | |
1168 | lrw->lr_length -= dnow; | |
b6ad9671 | 1169 | ZIL_STAT_BUMP(zil_itx_needcopy_count); |
a08ee875 LG |
1170 | ZIL_STAT_INCR(zil_itx_needcopy_bytes, |
1171 | lrw->lr_length); | |
34dc7c2f BB |
1172 | } else { |
1173 | ASSERT(itx->itx_wr_state == WR_INDIRECT); | |
1174 | dbuf = NULL; | |
b6ad9671 | 1175 | ZIL_STAT_BUMP(zil_itx_indirect_count); |
a08ee875 LG |
1176 | ZIL_STAT_INCR(zil_itx_indirect_bytes, |
1177 | lrw->lr_length); | |
34dc7c2f BB |
1178 | } |
1179 | error = zilog->zl_get_data( | |
cae5b340 | 1180 | itx->itx_private, lrwb, dbuf, lwb->lwb_zio); |
45d1cae3 BB |
1181 | if (error == EIO) { |
1182 | txg_wait_synced(zilog->zl_dmu_pool, txg); | |
1183 | return (lwb); | |
1184 | } | |
a08ee875 | 1185 | if (error != 0) { |
34dc7c2f BB |
1186 | ASSERT(error == ENOENT || error == EEXIST || |
1187 | error == EALREADY); | |
1188 | return (lwb); | |
1189 | } | |
1190 | } | |
1191 | } | |
1192 | ||
428870ff BB |
1193 | /* |
1194 | * We're actually making an entry, so update lrc_seq to be the | |
1195 | * log record sequence number. Note that this is generally not | |
1196 | * equal to the itx sequence number because not all transactions | |
1197 | * are synchronous, and sometimes spa_sync() gets there first. | |
1198 | */ | |
cae5b340 AX |
1199 | lrcb->lrc_seq = ++zilog->zl_lr_seq; /* we are single threaded */ |
1200 | lwb->lwb_nused += reclen + dnow; | |
34dc7c2f | 1201 | lwb->lwb_max_txg = MAX(lwb->lwb_max_txg, txg); |
428870ff | 1202 | ASSERT3U(lwb->lwb_nused, <=, lwb->lwb_sz); |
c06d4368 | 1203 | ASSERT0(P2PHASE(lwb->lwb_nused, sizeof (uint64_t))); |
34dc7c2f | 1204 | |
cae5b340 AX |
1205 | dlen -= dnow; |
1206 | if (dlen > 0) { | |
1207 | zilog->zl_cur_used += reclen; | |
1208 | goto cont; | |
1209 | } | |
1210 | ||
34dc7c2f BB |
1211 | return (lwb); |
1212 | } | |
1213 | ||
1214 | itx_t * | |
1215 | zil_itx_create(uint64_t txtype, size_t lrsize) | |
1216 | { | |
41d74433 | 1217 | size_t itxsize; |
34dc7c2f BB |
1218 | itx_t *itx; |
1219 | ||
1220 | lrsize = P2ROUNDUP_TYPED(lrsize, sizeof (uint64_t), size_t); | |
41d74433 | 1221 | itxsize = offsetof(itx_t, itx_lr) + lrsize; |
34dc7c2f | 1222 | |
41d74433 | 1223 | itx = zio_data_buf_alloc(itxsize); |
34dc7c2f BB |
1224 | itx->itx_lr.lrc_txtype = txtype; |
1225 | itx->itx_lr.lrc_reclen = lrsize; | |
34dc7c2f | 1226 | itx->itx_lr.lrc_seq = 0; /* defensive */ |
572e2857 | 1227 | itx->itx_sync = B_TRUE; /* default is synchronous */ |
a08ee875 LG |
1228 | itx->itx_callback = NULL; |
1229 | itx->itx_callback_data = NULL; | |
41d74433 | 1230 | itx->itx_size = itxsize; |
34dc7c2f BB |
1231 | |
1232 | return (itx); | |
1233 | } | |
1234 | ||
428870ff BB |
1235 | void |
1236 | zil_itx_destroy(itx_t *itx) | |
1237 | { | |
41d74433 | 1238 | zio_data_buf_free(itx, itx->itx_size); |
428870ff BB |
1239 | } |
1240 | ||
572e2857 BB |
1241 | /* |
1242 | * Free up the sync and async itxs. The itxs_t has already been detached | |
1243 | * so no locks are needed. | |
1244 | */ | |
1245 | static void | |
1246 | zil_itxg_clean(itxs_t *itxs) | |
34dc7c2f | 1247 | { |
572e2857 BB |
1248 | itx_t *itx; |
1249 | list_t *list; | |
1250 | avl_tree_t *t; | |
1251 | void *cookie; | |
1252 | itx_async_node_t *ian; | |
1253 | ||
1254 | list = &itxs->i_sync_list; | |
1255 | while ((itx = list_head(list)) != NULL) { | |
a08ee875 LG |
1256 | if (itx->itx_callback != NULL) |
1257 | itx->itx_callback(itx->itx_callback_data); | |
572e2857 | 1258 | list_remove(list, itx); |
ea04106b | 1259 | zil_itx_destroy(itx); |
572e2857 | 1260 | } |
34dc7c2f | 1261 | |
572e2857 BB |
1262 | cookie = NULL; |
1263 | t = &itxs->i_async_tree; | |
1264 | while ((ian = avl_destroy_nodes(t, &cookie)) != NULL) { | |
1265 | list = &ian->ia_list; | |
1266 | while ((itx = list_head(list)) != NULL) { | |
a08ee875 LG |
1267 | if (itx->itx_callback != NULL) |
1268 | itx->itx_callback(itx->itx_callback_data); | |
572e2857 | 1269 | list_remove(list, itx); |
ea04106b | 1270 | zil_itx_destroy(itx); |
572e2857 BB |
1271 | } |
1272 | list_destroy(list); | |
1273 | kmem_free(ian, sizeof (itx_async_node_t)); | |
1274 | } | |
1275 | avl_destroy(t); | |
34dc7c2f | 1276 | |
572e2857 BB |
1277 | kmem_free(itxs, sizeof (itxs_t)); |
1278 | } | |
34dc7c2f | 1279 | |
572e2857 BB |
1280 | static int |
1281 | zil_aitx_compare(const void *x1, const void *x2) | |
1282 | { | |
1283 | const uint64_t o1 = ((itx_async_node_t *)x1)->ia_foid; | |
1284 | const uint64_t o2 = ((itx_async_node_t *)x2)->ia_foid; | |
1285 | ||
cae5b340 | 1286 | return (AVL_CMP(o1, o2)); |
34dc7c2f BB |
1287 | } |
1288 | ||
1289 | /* | |
572e2857 | 1290 | * Remove all async itx with the given oid. |
34dc7c2f BB |
1291 | */ |
1292 | static void | |
572e2857 | 1293 | zil_remove_async(zilog_t *zilog, uint64_t oid) |
34dc7c2f | 1294 | { |
572e2857 BB |
1295 | uint64_t otxg, txg; |
1296 | itx_async_node_t *ian; | |
1297 | avl_tree_t *t; | |
1298 | avl_index_t where; | |
34dc7c2f BB |
1299 | list_t clean_list; |
1300 | itx_t *itx; | |
1301 | ||
572e2857 | 1302 | ASSERT(oid != 0); |
34dc7c2f BB |
1303 | list_create(&clean_list, sizeof (itx_t), offsetof(itx_t, itx_node)); |
1304 | ||
572e2857 BB |
1305 | if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */ |
1306 | otxg = ZILTEST_TXG; | |
1307 | else | |
1308 | otxg = spa_last_synced_txg(zilog->zl_spa) + 1; | |
34dc7c2f | 1309 | |
572e2857 BB |
1310 | for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) { |
1311 | itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK]; | |
1312 | ||
1313 | mutex_enter(&itxg->itxg_lock); | |
1314 | if (itxg->itxg_txg != txg) { | |
1315 | mutex_exit(&itxg->itxg_lock); | |
1316 | continue; | |
1317 | } | |
34dc7c2f | 1318 | |
572e2857 BB |
1319 | /* |
1320 | * Locate the object node and append its list. | |
1321 | */ | |
1322 | t = &itxg->itxg_itxs->i_async_tree; | |
1323 | ian = avl_find(t, &oid, &where); | |
1324 | if (ian != NULL) | |
1325 | list_move_tail(&clean_list, &ian->ia_list); | |
1326 | mutex_exit(&itxg->itxg_lock); | |
1327 | } | |
34dc7c2f | 1328 | while ((itx = list_head(&clean_list)) != NULL) { |
a08ee875 LG |
1329 | if (itx->itx_callback != NULL) |
1330 | itx->itx_callback(itx->itx_callback_data); | |
34dc7c2f | 1331 | list_remove(&clean_list, itx); |
ea04106b | 1332 | zil_itx_destroy(itx); |
34dc7c2f BB |
1333 | } |
1334 | list_destroy(&clean_list); | |
1335 | } | |
1336 | ||
572e2857 BB |
1337 | void |
1338 | zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx) | |
1339 | { | |
1340 | uint64_t txg; | |
1341 | itxg_t *itxg; | |
1342 | itxs_t *itxs, *clean = NULL; | |
1343 | ||
1344 | /* | |
1345 | * Object ids can be re-instantiated in the next txg so | |
1346 | * remove any async transactions to avoid future leaks. | |
1347 | * This can happen if a fsync occurs on the re-instantiated | |
1348 | * object for a WR_INDIRECT or WR_NEED_COPY write, which gets | |
1349 | * the new file data and flushes a write record for the old object. | |
1350 | */ | |
1351 | if ((itx->itx_lr.lrc_txtype & ~TX_CI) == TX_REMOVE) | |
1352 | zil_remove_async(zilog, itx->itx_oid); | |
1353 | ||
1354 | /* | |
1355 | * Ensure the data of a renamed file is committed before the rename. | |
1356 | */ | |
1357 | if ((itx->itx_lr.lrc_txtype & ~TX_CI) == TX_RENAME) | |
1358 | zil_async_to_sync(zilog, itx->itx_oid); | |
1359 | ||
29809a6c | 1360 | if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) |
572e2857 BB |
1361 | txg = ZILTEST_TXG; |
1362 | else | |
1363 | txg = dmu_tx_get_txg(tx); | |
1364 | ||
1365 | itxg = &zilog->zl_itxg[txg & TXG_MASK]; | |
1366 | mutex_enter(&itxg->itxg_lock); | |
1367 | itxs = itxg->itxg_itxs; | |
1368 | if (itxg->itxg_txg != txg) { | |
1369 | if (itxs != NULL) { | |
1370 | /* | |
1371 | * The zil_clean callback hasn't got around to cleaning | |
1372 | * this itxg. Save the itxs for release below. | |
1373 | * This should be rare. | |
1374 | */ | |
cae5b340 AX |
1375 | zfs_dbgmsg("zil_itx_assign: missed itx cleanup for " |
1376 | "txg %llu", itxg->itxg_txg); | |
572e2857 BB |
1377 | clean = itxg->itxg_itxs; |
1378 | } | |
572e2857 | 1379 | itxg->itxg_txg = txg; |
a08ee875 | 1380 | itxs = itxg->itxg_itxs = kmem_zalloc(sizeof (itxs_t), |
ea04106b | 1381 | KM_SLEEP); |
572e2857 BB |
1382 | |
1383 | list_create(&itxs->i_sync_list, sizeof (itx_t), | |
1384 | offsetof(itx_t, itx_node)); | |
1385 | avl_create(&itxs->i_async_tree, zil_aitx_compare, | |
1386 | sizeof (itx_async_node_t), | |
1387 | offsetof(itx_async_node_t, ia_node)); | |
1388 | } | |
1389 | if (itx->itx_sync) { | |
1390 | list_insert_tail(&itxs->i_sync_list, itx); | |
572e2857 BB |
1391 | } else { |
1392 | avl_tree_t *t = &itxs->i_async_tree; | |
cae5b340 AX |
1393 | uint64_t foid = |
1394 | LR_FOID_GET_OBJ(((lr_ooo_t *)&itx->itx_lr)->lr_foid); | |
572e2857 BB |
1395 | itx_async_node_t *ian; |
1396 | avl_index_t where; | |
1397 | ||
1398 | ian = avl_find(t, &foid, &where); | |
1399 | if (ian == NULL) { | |
a08ee875 | 1400 | ian = kmem_alloc(sizeof (itx_async_node_t), |
ea04106b | 1401 | KM_SLEEP); |
572e2857 BB |
1402 | list_create(&ian->ia_list, sizeof (itx_t), |
1403 | offsetof(itx_t, itx_node)); | |
1404 | ian->ia_foid = foid; | |
1405 | avl_insert(t, ian, where); | |
1406 | } | |
1407 | list_insert_tail(&ian->ia_list, itx); | |
1408 | } | |
1409 | ||
1410 | itx->itx_lr.lrc_txg = dmu_tx_get_txg(tx); | |
29809a6c | 1411 | zilog_dirty(zilog, txg); |
572e2857 BB |
1412 | mutex_exit(&itxg->itxg_lock); |
1413 | ||
1414 | /* Release the old itxs now we've dropped the lock */ | |
1415 | if (clean != NULL) | |
1416 | zil_itxg_clean(clean); | |
1417 | } | |
1418 | ||
34dc7c2f BB |
1419 | /* |
1420 | * If there are any in-memory intent log transactions which have now been | |
29809a6c MA |
1421 | * synced then start up a taskq to free them. We should only do this after we |
1422 | * have written out the uberblocks (i.e. txg has been comitted) so that | |
1423 | * don't inadvertently clean out in-memory log records that would be required | |
1424 | * by zil_commit(). | |
34dc7c2f BB |
1425 | */ |
1426 | void | |
572e2857 | 1427 | zil_clean(zilog_t *zilog, uint64_t synced_txg) |
34dc7c2f | 1428 | { |
572e2857 BB |
1429 | itxg_t *itxg = &zilog->zl_itxg[synced_txg & TXG_MASK]; |
1430 | itxs_t *clean_me; | |
34dc7c2f | 1431 | |
572e2857 BB |
1432 | mutex_enter(&itxg->itxg_lock); |
1433 | if (itxg->itxg_itxs == NULL || itxg->itxg_txg == ZILTEST_TXG) { | |
1434 | mutex_exit(&itxg->itxg_lock); | |
1435 | return; | |
1436 | } | |
1437 | ASSERT3U(itxg->itxg_txg, <=, synced_txg); | |
1438 | ASSERT(itxg->itxg_txg != 0); | |
1439 | ASSERT(zilog->zl_clean_taskq != NULL); | |
572e2857 BB |
1440 | clean_me = itxg->itxg_itxs; |
1441 | itxg->itxg_itxs = NULL; | |
1442 | itxg->itxg_txg = 0; | |
1443 | mutex_exit(&itxg->itxg_lock); | |
1444 | /* | |
1445 | * Preferably start a task queue to free up the old itxs but | |
1446 | * if taskq_dispatch can't allocate resources to do that then | |
1447 | * free it in-line. This should be rare. Note, using TQ_SLEEP | |
1448 | * created a bad performance problem. | |
1449 | */ | |
1450 | if (taskq_dispatch(zilog->zl_clean_taskq, | |
b8864a23 | 1451 | (void (*)(void *))zil_itxg_clean, clean_me, TQ_NOSLEEP) == 0) |
572e2857 BB |
1452 | zil_itxg_clean(clean_me); |
1453 | } | |
1454 | ||
1455 | /* | |
1456 | * Get the list of itxs to commit into zl_itx_commit_list. | |
1457 | */ | |
1458 | static void | |
1459 | zil_get_commit_list(zilog_t *zilog) | |
1460 | { | |
1461 | uint64_t otxg, txg; | |
1462 | list_t *commit_list = &zilog->zl_itx_commit_list; | |
572e2857 BB |
1463 | |
1464 | if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */ | |
1465 | otxg = ZILTEST_TXG; | |
1466 | else | |
1467 | otxg = spa_last_synced_txg(zilog->zl_spa) + 1; | |
1468 | ||
cae5b340 AX |
1469 | /* |
1470 | * This is inherently racy, since there is nothing to prevent | |
1471 | * the last synced txg from changing. That's okay since we'll | |
1472 | * only commit things in the future. | |
1473 | */ | |
572e2857 BB |
1474 | for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) { |
1475 | itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK]; | |
1476 | ||
1477 | mutex_enter(&itxg->itxg_lock); | |
1478 | if (itxg->itxg_txg != txg) { | |
1479 | mutex_exit(&itxg->itxg_lock); | |
1480 | continue; | |
1481 | } | |
1482 | ||
cae5b340 AX |
1483 | /* |
1484 | * If we're adding itx records to the zl_itx_commit_list, | |
1485 | * then the zil better be dirty in this "txg". We can assert | |
1486 | * that here since we're holding the itxg_lock which will | |
1487 | * prevent spa_sync from cleaning it. Once we add the itxs | |
1488 | * to the zl_itx_commit_list we must commit it to disk even | |
1489 | * if it's unnecessary (i.e. the txg was synced). | |
1490 | */ | |
1491 | ASSERT(zilog_is_dirty_in_txg(zilog, txg) || | |
1492 | spa_freeze_txg(zilog->zl_spa) != UINT64_MAX); | |
572e2857 | 1493 | list_move_tail(commit_list, &itxg->itxg_itxs->i_sync_list); |
572e2857 BB |
1494 | |
1495 | mutex_exit(&itxg->itxg_lock); | |
1496 | } | |
572e2857 BB |
1497 | } |
1498 | ||
1499 | /* | |
1500 | * Move the async itxs for a specified object to commit into sync lists. | |
1501 | */ | |
1502 | static void | |
1503 | zil_async_to_sync(zilog_t *zilog, uint64_t foid) | |
1504 | { | |
1505 | uint64_t otxg, txg; | |
1506 | itx_async_node_t *ian; | |
1507 | avl_tree_t *t; | |
1508 | avl_index_t where; | |
1509 | ||
1510 | if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */ | |
1511 | otxg = ZILTEST_TXG; | |
1512 | else | |
1513 | otxg = spa_last_synced_txg(zilog->zl_spa) + 1; | |
1514 | ||
cae5b340 AX |
1515 | /* |
1516 | * This is inherently racy, since there is nothing to prevent | |
1517 | * the last synced txg from changing. | |
1518 | */ | |
572e2857 BB |
1519 | for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) { |
1520 | itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK]; | |
1521 | ||
1522 | mutex_enter(&itxg->itxg_lock); | |
1523 | if (itxg->itxg_txg != txg) { | |
1524 | mutex_exit(&itxg->itxg_lock); | |
1525 | continue; | |
1526 | } | |
1527 | ||
1528 | /* | |
1529 | * If a foid is specified then find that node and append its | |
1530 | * list. Otherwise walk the tree appending all the lists | |
1531 | * to the sync list. We add to the end rather than the | |
1532 | * beginning to ensure the create has happened. | |
1533 | */ | |
1534 | t = &itxg->itxg_itxs->i_async_tree; | |
1535 | if (foid != 0) { | |
1536 | ian = avl_find(t, &foid, &where); | |
1537 | if (ian != NULL) { | |
1538 | list_move_tail(&itxg->itxg_itxs->i_sync_list, | |
1539 | &ian->ia_list); | |
1540 | } | |
1541 | } else { | |
1542 | void *cookie = NULL; | |
1543 | ||
1544 | while ((ian = avl_destroy_nodes(t, &cookie)) != NULL) { | |
1545 | list_move_tail(&itxg->itxg_itxs->i_sync_list, | |
1546 | &ian->ia_list); | |
1547 | list_destroy(&ian->ia_list); | |
1548 | kmem_free(ian, sizeof (itx_async_node_t)); | |
1549 | } | |
1550 | } | |
1551 | mutex_exit(&itxg->itxg_lock); | |
34dc7c2f | 1552 | } |
34dc7c2f BB |
1553 | } |
1554 | ||
b128c09f | 1555 | static void |
572e2857 | 1556 | zil_commit_writer(zilog_t *zilog) |
34dc7c2f BB |
1557 | { |
1558 | uint64_t txg; | |
572e2857 | 1559 | itx_t *itx; |
34dc7c2f | 1560 | lwb_t *lwb; |
572e2857 | 1561 | spa_t *spa = zilog->zl_spa; |
428870ff | 1562 | int error = 0; |
34dc7c2f | 1563 | |
b128c09f | 1564 | ASSERT(zilog->zl_root_zio == NULL); |
572e2857 BB |
1565 | |
1566 | mutex_exit(&zilog->zl_lock); | |
1567 | ||
1568 | zil_get_commit_list(zilog); | |
1569 | ||
1570 | /* | |
1571 | * Return if there's nothing to commit before we dirty the fs by | |
1572 | * calling zil_create(). | |
1573 | */ | |
1574 | if (list_head(&zilog->zl_itx_commit_list) == NULL) { | |
1575 | mutex_enter(&zilog->zl_lock); | |
1576 | return; | |
1577 | } | |
34dc7c2f BB |
1578 | |
1579 | if (zilog->zl_suspend) { | |
1580 | lwb = NULL; | |
1581 | } else { | |
1582 | lwb = list_tail(&zilog->zl_lwb_list); | |
572e2857 | 1583 | if (lwb == NULL) |
428870ff | 1584 | lwb = zil_create(zilog); |
34dc7c2f BB |
1585 | } |
1586 | ||
34dc7c2f | 1587 | DTRACE_PROBE1(zil__cw1, zilog_t *, zilog); |
a08ee875 LG |
1588 | for (itx = list_head(&zilog->zl_itx_commit_list); itx != NULL; |
1589 | itx = list_next(&zilog->zl_itx_commit_list, itx)) { | |
34dc7c2f | 1590 | txg = itx->itx_lr.lrc_txg; |
cae5b340 | 1591 | ASSERT3U(txg, !=, 0); |
34dc7c2f | 1592 | |
cae5b340 AX |
1593 | /* |
1594 | * This is inherently racy and may result in us writing | |
1595 | * out a log block for a txg that was just synced. This is | |
1596 | * ok since we'll end cleaning up that log block the next | |
1597 | * time we call zil_sync(). | |
1598 | */ | |
572e2857 | 1599 | if (txg > spa_last_synced_txg(spa) || txg > spa_freeze_txg(spa)) |
34dc7c2f | 1600 | lwb = zil_lwb_commit(zilog, itx, lwb); |
34dc7c2f BB |
1601 | } |
1602 | DTRACE_PROBE1(zil__cw2, zilog_t *, zilog); | |
34dc7c2f BB |
1603 | |
1604 | /* write the last block out */ | |
1605 | if (lwb != NULL && lwb->lwb_zio != NULL) | |
1606 | lwb = zil_lwb_write_start(zilog, lwb); | |
1607 | ||
34dc7c2f BB |
1608 | zilog->zl_cur_used = 0; |
1609 | ||
1610 | /* | |
1611 | * Wait if necessary for the log blocks to be on stable storage. | |
1612 | */ | |
1613 | if (zilog->zl_root_zio) { | |
428870ff | 1614 | error = zio_wait(zilog->zl_root_zio); |
b128c09f | 1615 | zilog->zl_root_zio = NULL; |
34dc7c2f BB |
1616 | zil_flush_vdevs(zilog); |
1617 | } | |
1618 | ||
428870ff | 1619 | if (error || lwb == NULL) |
34dc7c2f | 1620 | txg_wait_synced(zilog->zl_dmu_pool, 0); |
34dc7c2f | 1621 | |
a08ee875 LG |
1622 | while ((itx = list_head(&zilog->zl_itx_commit_list))) { |
1623 | txg = itx->itx_lr.lrc_txg; | |
1624 | ASSERT(txg); | |
1625 | ||
1626 | if (itx->itx_callback != NULL) | |
1627 | itx->itx_callback(itx->itx_callback_data); | |
1628 | list_remove(&zilog->zl_itx_commit_list, itx); | |
ea04106b | 1629 | zil_itx_destroy(itx); |
a08ee875 LG |
1630 | } |
1631 | ||
34dc7c2f | 1632 | mutex_enter(&zilog->zl_lock); |
428870ff BB |
1633 | |
1634 | /* | |
1635 | * Remember the highest committed log sequence number for ztest. | |
1636 | * We only update this value when all the log writes succeeded, | |
1637 | * because ztest wants to ASSERT that it got the whole log chain. | |
1638 | */ | |
1639 | if (error == 0 && lwb != NULL) | |
1640 | zilog->zl_commit_lr_seq = zilog->zl_lr_seq; | |
34dc7c2f BB |
1641 | } |
1642 | ||
1643 | /* | |
572e2857 | 1644 | * Commit zfs transactions to stable storage. |
34dc7c2f | 1645 | * If foid is 0 push out all transactions, otherwise push only those |
572e2857 BB |
1646 | * for that object or might reference that object. |
1647 | * | |
1648 | * itxs are committed in batches. In a heavily stressed zil there will be | |
1649 | * a commit writer thread who is writing out a bunch of itxs to the log | |
1650 | * for a set of committing threads (cthreads) in the same batch as the writer. | |
1651 | * Those cthreads are all waiting on the same cv for that batch. | |
1652 | * | |
1653 | * There will also be a different and growing batch of threads that are | |
1654 | * waiting to commit (qthreads). When the committing batch completes | |
1655 | * a transition occurs such that the cthreads exit and the qthreads become | |
1656 | * cthreads. One of the new cthreads becomes the writer thread for the | |
1657 | * batch. Any new threads arriving become new qthreads. | |
1658 | * | |
1659 | * Only 2 condition variables are needed and there's no transition | |
1660 | * between the two cvs needed. They just flip-flop between qthreads | |
1661 | * and cthreads. | |
1662 | * | |
1663 | * Using this scheme we can efficiently wakeup up only those threads | |
1664 | * that have been committed. | |
34dc7c2f BB |
1665 | */ |
1666 | void | |
572e2857 | 1667 | zil_commit(zilog_t *zilog, uint64_t foid) |
34dc7c2f | 1668 | { |
572e2857 | 1669 | uint64_t mybatch; |
34dc7c2f | 1670 | |
572e2857 BB |
1671 | if (zilog->zl_sync == ZFS_SYNC_DISABLED) |
1672 | return; | |
34dc7c2f | 1673 | |
b6ad9671 ED |
1674 | ZIL_STAT_BUMP(zil_commit_count); |
1675 | ||
572e2857 BB |
1676 | /* move the async itxs for the foid to the sync queues */ |
1677 | zil_async_to_sync(zilog, foid); | |
34dc7c2f | 1678 | |
572e2857 BB |
1679 | mutex_enter(&zilog->zl_lock); |
1680 | mybatch = zilog->zl_next_batch; | |
34dc7c2f | 1681 | while (zilog->zl_writer) { |
572e2857 BB |
1682 | cv_wait(&zilog->zl_cv_batch[mybatch & 1], &zilog->zl_lock); |
1683 | if (mybatch <= zilog->zl_com_batch) { | |
34dc7c2f BB |
1684 | mutex_exit(&zilog->zl_lock); |
1685 | return; | |
1686 | } | |
1687 | } | |
428870ff | 1688 | |
572e2857 BB |
1689 | zilog->zl_next_batch++; |
1690 | zilog->zl_writer = B_TRUE; | |
b6ad9671 | 1691 | ZIL_STAT_BUMP(zil_commit_writer_count); |
572e2857 BB |
1692 | zil_commit_writer(zilog); |
1693 | zilog->zl_com_batch = mybatch; | |
1694 | zilog->zl_writer = B_FALSE; | |
428870ff | 1695 | |
572e2857 BB |
1696 | /* wake up one thread to become the next writer */ |
1697 | cv_signal(&zilog->zl_cv_batch[(mybatch+1) & 1]); | |
428870ff | 1698 | |
572e2857 BB |
1699 | /* wake up all threads waiting for this batch to be committed */ |
1700 | cv_broadcast(&zilog->zl_cv_batch[mybatch & 1]); | |
8c0712fd BB |
1701 | |
1702 | mutex_exit(&zilog->zl_lock); | |
428870ff BB |
1703 | } |
1704 | ||
34dc7c2f BB |
1705 | /* |
1706 | * Called in syncing context to free committed log blocks and update log header. | |
1707 | */ | |
1708 | void | |
1709 | zil_sync(zilog_t *zilog, dmu_tx_t *tx) | |
1710 | { | |
1711 | zil_header_t *zh = zil_header_in_syncing_context(zilog); | |
1712 | uint64_t txg = dmu_tx_get_txg(tx); | |
1713 | spa_t *spa = zilog->zl_spa; | |
428870ff | 1714 | uint64_t *replayed_seq = &zilog->zl_replayed_seq[txg & TXG_MASK]; |
34dc7c2f BB |
1715 | lwb_t *lwb; |
1716 | ||
9babb374 BB |
1717 | /* |
1718 | * We don't zero out zl_destroy_txg, so make sure we don't try | |
1719 | * to destroy it twice. | |
1720 | */ | |
1721 | if (spa_sync_pass(spa) != 1) | |
1722 | return; | |
1723 | ||
34dc7c2f BB |
1724 | mutex_enter(&zilog->zl_lock); |
1725 | ||
1726 | ASSERT(zilog->zl_stop_sync == 0); | |
1727 | ||
428870ff BB |
1728 | if (*replayed_seq != 0) { |
1729 | ASSERT(zh->zh_replay_seq < *replayed_seq); | |
1730 | zh->zh_replay_seq = *replayed_seq; | |
1731 | *replayed_seq = 0; | |
1732 | } | |
34dc7c2f BB |
1733 | |
1734 | if (zilog->zl_destroy_txg == txg) { | |
1735 | blkptr_t blk = zh->zh_log; | |
1736 | ||
1737 | ASSERT(list_head(&zilog->zl_lwb_list) == NULL); | |
34dc7c2f BB |
1738 | |
1739 | bzero(zh, sizeof (zil_header_t)); | |
fb5f0bc8 | 1740 | bzero(zilog->zl_replayed_seq, sizeof (zilog->zl_replayed_seq)); |
34dc7c2f BB |
1741 | |
1742 | if (zilog->zl_keep_first) { | |
1743 | /* | |
1744 | * If this block was part of log chain that couldn't | |
1745 | * be claimed because a device was missing during | |
1746 | * zil_claim(), but that device later returns, | |
1747 | * then this block could erroneously appear valid. | |
1748 | * To guard against this, assign a new GUID to the new | |
1749 | * log chain so it doesn't matter what blk points to. | |
1750 | */ | |
1751 | zil_init_log_chain(zilog, &blk); | |
1752 | zh->zh_log = blk; | |
1753 | } | |
1754 | } | |
1755 | ||
9babb374 | 1756 | while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) { |
34dc7c2f BB |
1757 | zh->zh_log = lwb->lwb_blk; |
1758 | if (lwb->lwb_buf != NULL || lwb->lwb_max_txg > txg) | |
1759 | break; | |
920dd524 ED |
1760 | |
1761 | ASSERT(lwb->lwb_zio == NULL); | |
1762 | ||
34dc7c2f | 1763 | list_remove(&zilog->zl_lwb_list, lwb); |
428870ff | 1764 | zio_free_zil(spa, txg, &lwb->lwb_blk); |
34dc7c2f BB |
1765 | kmem_cache_free(zil_lwb_cache, lwb); |
1766 | ||
1767 | /* | |
1768 | * If we don't have anything left in the lwb list then | |
1769 | * we've had an allocation failure and we need to zero | |
1770 | * out the zil_header blkptr so that we don't end | |
1771 | * up freeing the same block twice. | |
1772 | */ | |
1773 | if (list_head(&zilog->zl_lwb_list) == NULL) | |
1774 | BP_ZERO(&zh->zh_log); | |
1775 | } | |
920dd524 ED |
1776 | |
1777 | /* | |
1778 | * Remove fastwrite on any blocks that have been pre-allocated for | |
1779 | * the next commit. This prevents fastwrite counter pollution by | |
1780 | * unused, long-lived LWBs. | |
1781 | */ | |
1782 | for (; lwb != NULL; lwb = list_next(&zilog->zl_lwb_list, lwb)) { | |
1783 | if (lwb->lwb_fastwrite && !lwb->lwb_zio) { | |
1784 | metaslab_fastwrite_unmark(zilog->zl_spa, &lwb->lwb_blk); | |
1785 | lwb->lwb_fastwrite = 0; | |
1786 | } | |
1787 | } | |
1788 | ||
34dc7c2f BB |
1789 | mutex_exit(&zilog->zl_lock); |
1790 | } | |
1791 | ||
1792 | void | |
1793 | zil_init(void) | |
1794 | { | |
1795 | zil_lwb_cache = kmem_cache_create("zil_lwb_cache", | |
1796 | sizeof (struct lwb), 0, NULL, NULL, NULL, NULL, NULL, 0); | |
b6ad9671 ED |
1797 | |
1798 | zil_ksp = kstat_create("zfs", 0, "zil", "misc", | |
a08ee875 | 1799 | KSTAT_TYPE_NAMED, sizeof (zil_stats) / sizeof (kstat_named_t), |
b6ad9671 ED |
1800 | KSTAT_FLAG_VIRTUAL); |
1801 | ||
1802 | if (zil_ksp != NULL) { | |
1803 | zil_ksp->ks_data = &zil_stats; | |
1804 | kstat_install(zil_ksp); | |
1805 | } | |
34dc7c2f BB |
1806 | } |
1807 | ||
1808 | void | |
1809 | zil_fini(void) | |
1810 | { | |
1811 | kmem_cache_destroy(zil_lwb_cache); | |
b6ad9671 ED |
1812 | |
1813 | if (zil_ksp != NULL) { | |
1814 | kstat_delete(zil_ksp); | |
1815 | zil_ksp = NULL; | |
1816 | } | |
34dc7c2f BB |
1817 | } |
1818 | ||
428870ff BB |
1819 | void |
1820 | zil_set_sync(zilog_t *zilog, uint64_t sync) | |
1821 | { | |
1822 | zilog->zl_sync = sync; | |
1823 | } | |
1824 | ||
1825 | void | |
1826 | zil_set_logbias(zilog_t *zilog, uint64_t logbias) | |
1827 | { | |
1828 | zilog->zl_logbias = logbias; | |
1829 | } | |
1830 | ||
34dc7c2f BB |
1831 | zilog_t * |
1832 | zil_alloc(objset_t *os, zil_header_t *zh_phys) | |
1833 | { | |
1834 | zilog_t *zilog; | |
d6320ddb | 1835 | int i; |
34dc7c2f | 1836 | |
ea04106b | 1837 | zilog = kmem_zalloc(sizeof (zilog_t), KM_SLEEP); |
34dc7c2f BB |
1838 | |
1839 | zilog->zl_header = zh_phys; | |
1840 | zilog->zl_os = os; | |
1841 | zilog->zl_spa = dmu_objset_spa(os); | |
1842 | zilog->zl_dmu_pool = dmu_objset_pool(os); | |
1843 | zilog->zl_destroy_txg = TXG_INITIAL - 1; | |
428870ff BB |
1844 | zilog->zl_logbias = dmu_objset_logbias(os); |
1845 | zilog->zl_sync = dmu_objset_syncprop(os); | |
572e2857 | 1846 | zilog->zl_next_batch = 1; |
34dc7c2f BB |
1847 | |
1848 | mutex_init(&zilog->zl_lock, NULL, MUTEX_DEFAULT, NULL); | |
1849 | ||
d6320ddb | 1850 | for (i = 0; i < TXG_SIZE; i++) { |
572e2857 BB |
1851 | mutex_init(&zilog->zl_itxg[i].itxg_lock, NULL, |
1852 | MUTEX_DEFAULT, NULL); | |
1853 | } | |
34dc7c2f BB |
1854 | |
1855 | list_create(&zilog->zl_lwb_list, sizeof (lwb_t), | |
1856 | offsetof(lwb_t, lwb_node)); | |
1857 | ||
572e2857 BB |
1858 | list_create(&zilog->zl_itx_commit_list, sizeof (itx_t), |
1859 | offsetof(itx_t, itx_node)); | |
1860 | ||
34dc7c2f BB |
1861 | mutex_init(&zilog->zl_vdev_lock, NULL, MUTEX_DEFAULT, NULL); |
1862 | ||
1863 | avl_create(&zilog->zl_vdev_tree, zil_vdev_compare, | |
1864 | sizeof (zil_vdev_node_t), offsetof(zil_vdev_node_t, zv_node)); | |
1865 | ||
1866 | cv_init(&zilog->zl_cv_writer, NULL, CV_DEFAULT, NULL); | |
1867 | cv_init(&zilog->zl_cv_suspend, NULL, CV_DEFAULT, NULL); | |
572e2857 BB |
1868 | cv_init(&zilog->zl_cv_batch[0], NULL, CV_DEFAULT, NULL); |
1869 | cv_init(&zilog->zl_cv_batch[1], NULL, CV_DEFAULT, NULL); | |
34dc7c2f BB |
1870 | |
1871 | return (zilog); | |
1872 | } | |
1873 | ||
1874 | void | |
1875 | zil_free(zilog_t *zilog) | |
1876 | { | |
d6320ddb | 1877 | int i; |
34dc7c2f BB |
1878 | |
1879 | zilog->zl_stop_sync = 1; | |
1880 | ||
a08ee875 LG |
1881 | ASSERT0(zilog->zl_suspend); |
1882 | ASSERT0(zilog->zl_suspending); | |
1883 | ||
3e31d2b0 | 1884 | ASSERT(list_is_empty(&zilog->zl_lwb_list)); |
34dc7c2f BB |
1885 | list_destroy(&zilog->zl_lwb_list); |
1886 | ||
1887 | avl_destroy(&zilog->zl_vdev_tree); | |
1888 | mutex_destroy(&zilog->zl_vdev_lock); | |
1889 | ||
572e2857 BB |
1890 | ASSERT(list_is_empty(&zilog->zl_itx_commit_list)); |
1891 | list_destroy(&zilog->zl_itx_commit_list); | |
1892 | ||
d6320ddb | 1893 | for (i = 0; i < TXG_SIZE; i++) { |
572e2857 BB |
1894 | /* |
1895 | * It's possible for an itx to be generated that doesn't dirty | |
1896 | * a txg (e.g. ztest TX_TRUNCATE). So there's no zil_clean() | |
1897 | * callback to remove the entry. We remove those here. | |
1898 | * | |
1899 | * Also free up the ziltest itxs. | |
1900 | */ | |
1901 | if (zilog->zl_itxg[i].itxg_itxs) | |
1902 | zil_itxg_clean(zilog->zl_itxg[i].itxg_itxs); | |
1903 | mutex_destroy(&zilog->zl_itxg[i].itxg_lock); | |
1904 | } | |
1905 | ||
34dc7c2f BB |
1906 | mutex_destroy(&zilog->zl_lock); |
1907 | ||
1908 | cv_destroy(&zilog->zl_cv_writer); | |
1909 | cv_destroy(&zilog->zl_cv_suspend); | |
572e2857 BB |
1910 | cv_destroy(&zilog->zl_cv_batch[0]); |
1911 | cv_destroy(&zilog->zl_cv_batch[1]); | |
34dc7c2f BB |
1912 | |
1913 | kmem_free(zilog, sizeof (zilog_t)); | |
1914 | } | |
1915 | ||
34dc7c2f BB |
1916 | /* |
1917 | * Open an intent log. | |
1918 | */ | |
1919 | zilog_t * | |
1920 | zil_open(objset_t *os, zil_get_data_t *get_data) | |
1921 | { | |
1922 | zilog_t *zilog = dmu_objset_zil(os); | |
1923 | ||
3e31d2b0 ES |
1924 | ASSERT(zilog->zl_clean_taskq == NULL); |
1925 | ASSERT(zilog->zl_get_data == NULL); | |
1926 | ASSERT(list_is_empty(&zilog->zl_lwb_list)); | |
1927 | ||
34dc7c2f | 1928 | zilog->zl_get_data = get_data; |
e10b0808 | 1929 | zilog->zl_clean_taskq = taskq_create("zil_clean", 1, defclsyspri, |
34dc7c2f BB |
1930 | 2, 2, TASKQ_PREPOPULATE); |
1931 | ||
1932 | return (zilog); | |
1933 | } | |
1934 | ||
1935 | /* | |
1936 | * Close an intent log. | |
1937 | */ | |
1938 | void | |
1939 | zil_close(zilog_t *zilog) | |
1940 | { | |
3e31d2b0 | 1941 | lwb_t *lwb; |
572e2857 BB |
1942 | uint64_t txg = 0; |
1943 | ||
1944 | zil_commit(zilog, 0); /* commit all itx */ | |
1945 | ||
34dc7c2f | 1946 | /* |
572e2857 BB |
1947 | * The lwb_max_txg for the stubby lwb will reflect the last activity |
1948 | * for the zil. After a txg_wait_synced() on the txg we know all the | |
1949 | * callbacks have occurred that may clean the zil. Only then can we | |
1950 | * destroy the zl_clean_taskq. | |
34dc7c2f | 1951 | */ |
572e2857 | 1952 | mutex_enter(&zilog->zl_lock); |
3e31d2b0 ES |
1953 | lwb = list_tail(&zilog->zl_lwb_list); |
1954 | if (lwb != NULL) | |
1955 | txg = lwb->lwb_max_txg; | |
572e2857 BB |
1956 | mutex_exit(&zilog->zl_lock); |
1957 | if (txg) | |
34dc7c2f | 1958 | txg_wait_synced(zilog->zl_dmu_pool, txg); |
cae5b340 AX |
1959 | |
1960 | if (zilog_is_dirty(zilog)) | |
1961 | zfs_dbgmsg("zil (%p) is dirty, txg %llu", zilog, txg); | |
1962 | if (txg < spa_freeze_txg(zilog->zl_spa)) | |
1963 | VERIFY(!zilog_is_dirty(zilog)); | |
34dc7c2f BB |
1964 | |
1965 | taskq_destroy(zilog->zl_clean_taskq); | |
1966 | zilog->zl_clean_taskq = NULL; | |
1967 | zilog->zl_get_data = NULL; | |
3e31d2b0 ES |
1968 | |
1969 | /* | |
1970 | * We should have only one LWB left on the list; remove it now. | |
1971 | */ | |
1972 | mutex_enter(&zilog->zl_lock); | |
1973 | lwb = list_head(&zilog->zl_lwb_list); | |
1974 | if (lwb != NULL) { | |
1975 | ASSERT(lwb == list_tail(&zilog->zl_lwb_list)); | |
920dd524 ED |
1976 | ASSERT(lwb->lwb_zio == NULL); |
1977 | if (lwb->lwb_fastwrite) | |
1978 | metaslab_fastwrite_unmark(zilog->zl_spa, &lwb->lwb_blk); | |
3e31d2b0 ES |
1979 | list_remove(&zilog->zl_lwb_list, lwb); |
1980 | zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); | |
1981 | kmem_cache_free(zil_lwb_cache, lwb); | |
1982 | } | |
1983 | mutex_exit(&zilog->zl_lock); | |
34dc7c2f BB |
1984 | } |
1985 | ||
a08ee875 LG |
1986 | static char *suspend_tag = "zil suspending"; |
1987 | ||
34dc7c2f BB |
1988 | /* |
1989 | * Suspend an intent log. While in suspended mode, we still honor | |
1990 | * synchronous semantics, but we rely on txg_wait_synced() to do it. | |
a08ee875 LG |
1991 | * On old version pools, we suspend the log briefly when taking a |
1992 | * snapshot so that it will have an empty intent log. | |
1993 | * | |
1994 | * Long holds are not really intended to be used the way we do here -- | |
1995 | * held for such a short time. A concurrent caller of dsl_dataset_long_held() | |
1996 | * could fail. Therefore we take pains to only put a long hold if it is | |
1997 | * actually necessary. Fortunately, it will only be necessary if the | |
1998 | * objset is currently mounted (or the ZVOL equivalent). In that case it | |
1999 | * will already have a long hold, so we are not really making things any worse. | |
2000 | * | |
2001 | * Ideally, we would locate the existing long-holder (i.e. the zfsvfs_t or | |
2002 | * zvol_state_t), and use their mechanism to prevent their hold from being | |
2003 | * dropped (e.g. VFS_HOLD()). However, that would be even more pain for | |
2004 | * very little gain. | |
2005 | * | |
2006 | * if cookiep == NULL, this does both the suspend & resume. | |
2007 | * Otherwise, it returns with the dataset "long held", and the cookie | |
2008 | * should be passed into zil_resume(). | |
34dc7c2f BB |
2009 | */ |
2010 | int | |
a08ee875 | 2011 | zil_suspend(const char *osname, void **cookiep) |
34dc7c2f | 2012 | { |
a08ee875 LG |
2013 | objset_t *os; |
2014 | zilog_t *zilog; | |
2015 | const zil_header_t *zh; | |
2016 | int error; | |
2017 | ||
2018 | error = dmu_objset_hold(osname, suspend_tag, &os); | |
2019 | if (error != 0) | |
2020 | return (error); | |
2021 | zilog = dmu_objset_zil(os); | |
34dc7c2f BB |
2022 | |
2023 | mutex_enter(&zilog->zl_lock); | |
a08ee875 LG |
2024 | zh = zilog->zl_header; |
2025 | ||
9babb374 | 2026 | if (zh->zh_flags & ZIL_REPLAY_NEEDED) { /* unplayed log */ |
34dc7c2f | 2027 | mutex_exit(&zilog->zl_lock); |
a08ee875 LG |
2028 | dmu_objset_rele(os, suspend_tag); |
2029 | return (SET_ERROR(EBUSY)); | |
34dc7c2f | 2030 | } |
a08ee875 LG |
2031 | |
2032 | /* | |
2033 | * Don't put a long hold in the cases where we can avoid it. This | |
2034 | * is when there is no cookie so we are doing a suspend & resume | |
2035 | * (i.e. called from zil_vdev_offline()), and there's nothing to do | |
2036 | * for the suspend because it's already suspended, or there's no ZIL. | |
2037 | */ | |
2038 | if (cookiep == NULL && !zilog->zl_suspending && | |
2039 | (zilog->zl_suspend > 0 || BP_IS_HOLE(&zh->zh_log))) { | |
2040 | mutex_exit(&zilog->zl_lock); | |
2041 | dmu_objset_rele(os, suspend_tag); | |
2042 | return (0); | |
2043 | } | |
2044 | ||
2045 | dsl_dataset_long_hold(dmu_objset_ds(os), suspend_tag); | |
2046 | dsl_pool_rele(dmu_objset_pool(os), suspend_tag); | |
2047 | ||
2048 | zilog->zl_suspend++; | |
2049 | ||
2050 | if (zilog->zl_suspend > 1) { | |
34dc7c2f | 2051 | /* |
a08ee875 | 2052 | * Someone else is already suspending it. |
34dc7c2f BB |
2053 | * Just wait for them to finish. |
2054 | */ | |
a08ee875 | 2055 | |
34dc7c2f BB |
2056 | while (zilog->zl_suspending) |
2057 | cv_wait(&zilog->zl_cv_suspend, &zilog->zl_lock); | |
34dc7c2f | 2058 | mutex_exit(&zilog->zl_lock); |
a08ee875 LG |
2059 | |
2060 | if (cookiep == NULL) | |
2061 | zil_resume(os); | |
2062 | else | |
2063 | *cookiep = os; | |
2064 | return (0); | |
2065 | } | |
2066 | ||
2067 | /* | |
2068 | * If there is no pointer to an on-disk block, this ZIL must not | |
2069 | * be active (e.g. filesystem not mounted), so there's nothing | |
2070 | * to clean up. | |
2071 | */ | |
2072 | if (BP_IS_HOLE(&zh->zh_log)) { | |
2073 | ASSERT(cookiep != NULL); /* fast path already handled */ | |
2074 | ||
2075 | *cookiep = os; | |
2076 | mutex_exit(&zilog->zl_lock); | |
34dc7c2f BB |
2077 | return (0); |
2078 | } | |
a08ee875 | 2079 | |
34dc7c2f BB |
2080 | zilog->zl_suspending = B_TRUE; |
2081 | mutex_exit(&zilog->zl_lock); | |
2082 | ||
572e2857 | 2083 | zil_commit(zilog, 0); |
34dc7c2f BB |
2084 | |
2085 | zil_destroy(zilog, B_FALSE); | |
2086 | ||
2087 | mutex_enter(&zilog->zl_lock); | |
2088 | zilog->zl_suspending = B_FALSE; | |
2089 | cv_broadcast(&zilog->zl_cv_suspend); | |
2090 | mutex_exit(&zilog->zl_lock); | |
2091 | ||
a08ee875 LG |
2092 | if (cookiep == NULL) |
2093 | zil_resume(os); | |
2094 | else | |
2095 | *cookiep = os; | |
34dc7c2f BB |
2096 | return (0); |
2097 | } | |
2098 | ||
2099 | void | |
a08ee875 | 2100 | zil_resume(void *cookie) |
34dc7c2f | 2101 | { |
a08ee875 LG |
2102 | objset_t *os = cookie; |
2103 | zilog_t *zilog = dmu_objset_zil(os); | |
2104 | ||
34dc7c2f BB |
2105 | mutex_enter(&zilog->zl_lock); |
2106 | ASSERT(zilog->zl_suspend != 0); | |
2107 | zilog->zl_suspend--; | |
2108 | mutex_exit(&zilog->zl_lock); | |
a08ee875 LG |
2109 | dsl_dataset_long_rele(dmu_objset_ds(os), suspend_tag); |
2110 | dsl_dataset_rele(dmu_objset_ds(os), suspend_tag); | |
34dc7c2f BB |
2111 | } |
2112 | ||
2113 | typedef struct zil_replay_arg { | |
b01615d5 | 2114 | zil_replay_func_t *zr_replay; |
34dc7c2f | 2115 | void *zr_arg; |
34dc7c2f | 2116 | boolean_t zr_byteswap; |
428870ff | 2117 | char *zr_lr; |
34dc7c2f BB |
2118 | } zil_replay_arg_t; |
2119 | ||
428870ff BB |
2120 | static int |
2121 | zil_replay_error(zilog_t *zilog, lr_t *lr, int error) | |
2122 | { | |
cae5b340 | 2123 | char name[ZFS_MAX_DATASET_NAME_LEN]; |
428870ff BB |
2124 | |
2125 | zilog->zl_replaying_seq--; /* didn't actually replay this one */ | |
2126 | ||
2127 | dmu_objset_name(zilog->zl_os, name); | |
2128 | ||
2129 | cmn_err(CE_WARN, "ZFS replay transaction error %d, " | |
2130 | "dataset %s, seq 0x%llx, txtype %llu %s\n", error, name, | |
2131 | (u_longlong_t)lr->lrc_seq, | |
2132 | (u_longlong_t)(lr->lrc_txtype & ~TX_CI), | |
2133 | (lr->lrc_txtype & TX_CI) ? "CI" : ""); | |
2134 | ||
2135 | return (error); | |
2136 | } | |
2137 | ||
2138 | static int | |
34dc7c2f BB |
2139 | zil_replay_log_record(zilog_t *zilog, lr_t *lr, void *zra, uint64_t claim_txg) |
2140 | { | |
2141 | zil_replay_arg_t *zr = zra; | |
2142 | const zil_header_t *zh = zilog->zl_header; | |
2143 | uint64_t reclen = lr->lrc_reclen; | |
2144 | uint64_t txtype = lr->lrc_txtype; | |
428870ff | 2145 | int error = 0; |
34dc7c2f | 2146 | |
428870ff | 2147 | zilog->zl_replaying_seq = lr->lrc_seq; |
34dc7c2f BB |
2148 | |
2149 | if (lr->lrc_seq <= zh->zh_replay_seq) /* already replayed */ | |
428870ff BB |
2150 | return (0); |
2151 | ||
2152 | if (lr->lrc_txg < claim_txg) /* already committed */ | |
2153 | return (0); | |
34dc7c2f BB |
2154 | |
2155 | /* Strip case-insensitive bit, still present in log record */ | |
2156 | txtype &= ~TX_CI; | |
2157 | ||
428870ff BB |
2158 | if (txtype == 0 || txtype >= TX_MAX_TYPE) |
2159 | return (zil_replay_error(zilog, lr, EINVAL)); | |
2160 | ||
2161 | /* | |
2162 | * If this record type can be logged out of order, the object | |
2163 | * (lr_foid) may no longer exist. That's legitimate, not an error. | |
2164 | */ | |
2165 | if (TX_OOO(txtype)) { | |
2166 | error = dmu_object_info(zilog->zl_os, | |
cae5b340 | 2167 | LR_FOID_GET_OBJ(((lr_ooo_t *)lr)->lr_foid), NULL); |
428870ff BB |
2168 | if (error == ENOENT || error == EEXIST) |
2169 | return (0); | |
fb5f0bc8 BB |
2170 | } |
2171 | ||
34dc7c2f BB |
2172 | /* |
2173 | * Make a copy of the data so we can revise and extend it. | |
2174 | */ | |
428870ff BB |
2175 | bcopy(lr, zr->zr_lr, reclen); |
2176 | ||
2177 | /* | |
2178 | * If this is a TX_WRITE with a blkptr, suck in the data. | |
2179 | */ | |
2180 | if (txtype == TX_WRITE && reclen == sizeof (lr_write_t)) { | |
2181 | error = zil_read_log_data(zilog, (lr_write_t *)lr, | |
2182 | zr->zr_lr + reclen); | |
a08ee875 | 2183 | if (error != 0) |
428870ff BB |
2184 | return (zil_replay_error(zilog, lr, error)); |
2185 | } | |
34dc7c2f BB |
2186 | |
2187 | /* | |
2188 | * The log block containing this lr may have been byteswapped | |
2189 | * so that we can easily examine common fields like lrc_txtype. | |
428870ff | 2190 | * However, the log is a mix of different record types, and only the |
34dc7c2f BB |
2191 | * replay vectors know how to byteswap their records. Therefore, if |
2192 | * the lr was byteswapped, undo it before invoking the replay vector. | |
2193 | */ | |
2194 | if (zr->zr_byteswap) | |
428870ff | 2195 | byteswap_uint64_array(zr->zr_lr, reclen); |
34dc7c2f BB |
2196 | |
2197 | /* | |
2198 | * We must now do two things atomically: replay this log record, | |
fb5f0bc8 BB |
2199 | * and update the log header sequence number to reflect the fact that |
2200 | * we did so. At the end of each replay function the sequence number | |
2201 | * is updated if we are in replay mode. | |
34dc7c2f | 2202 | */ |
428870ff | 2203 | error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lr, zr->zr_byteswap); |
a08ee875 | 2204 | if (error != 0) { |
34dc7c2f BB |
2205 | /* |
2206 | * The DMU's dnode layer doesn't see removes until the txg | |
2207 | * commits, so a subsequent claim can spuriously fail with | |
fb5f0bc8 | 2208 | * EEXIST. So if we receive any error we try syncing out |
428870ff BB |
2209 | * any removes then retry the transaction. Note that we |
2210 | * specify B_FALSE for byteswap now, so we don't do it twice. | |
34dc7c2f | 2211 | */ |
428870ff BB |
2212 | txg_wait_synced(spa_get_dsl(zilog->zl_spa), 0); |
2213 | error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lr, B_FALSE); | |
a08ee875 | 2214 | if (error != 0) |
428870ff | 2215 | return (zil_replay_error(zilog, lr, error)); |
34dc7c2f | 2216 | } |
428870ff | 2217 | return (0); |
34dc7c2f BB |
2218 | } |
2219 | ||
2220 | /* ARGSUSED */ | |
428870ff | 2221 | static int |
34dc7c2f BB |
2222 | zil_incr_blks(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg) |
2223 | { | |
2224 | zilog->zl_replay_blks++; | |
428870ff BB |
2225 | |
2226 | return (0); | |
34dc7c2f BB |
2227 | } |
2228 | ||
2229 | /* | |
2230 | * If this dataset has a non-empty intent log, replay it and destroy it. | |
2231 | */ | |
2232 | void | |
b01615d5 | 2233 | zil_replay(objset_t *os, void *arg, zil_replay_func_t replay_func[TX_MAX_TYPE]) |
34dc7c2f BB |
2234 | { |
2235 | zilog_t *zilog = dmu_objset_zil(os); | |
2236 | const zil_header_t *zh = zilog->zl_header; | |
2237 | zil_replay_arg_t zr; | |
2238 | ||
9babb374 | 2239 | if ((zh->zh_flags & ZIL_REPLAY_NEEDED) == 0) { |
34dc7c2f BB |
2240 | zil_destroy(zilog, B_TRUE); |
2241 | return; | |
2242 | } | |
2243 | ||
34dc7c2f BB |
2244 | zr.zr_replay = replay_func; |
2245 | zr.zr_arg = arg; | |
34dc7c2f | 2246 | zr.zr_byteswap = BP_SHOULD_BYTESWAP(&zh->zh_log); |
ea04106b | 2247 | zr.zr_lr = vmem_alloc(2 * SPA_MAXBLOCKSIZE, KM_SLEEP); |
34dc7c2f BB |
2248 | |
2249 | /* | |
2250 | * Wait for in-progress removes to sync before starting replay. | |
2251 | */ | |
2252 | txg_wait_synced(zilog->zl_dmu_pool, 0); | |
2253 | ||
fb5f0bc8 | 2254 | zilog->zl_replay = B_TRUE; |
428870ff | 2255 | zilog->zl_replay_time = ddi_get_lbolt(); |
34dc7c2f BB |
2256 | ASSERT(zilog->zl_replay_blks == 0); |
2257 | (void) zil_parse(zilog, zil_incr_blks, zil_replay_log_record, &zr, | |
2258 | zh->zh_claim_txg); | |
00b46022 | 2259 | vmem_free(zr.zr_lr, 2 * SPA_MAXBLOCKSIZE); |
34dc7c2f BB |
2260 | |
2261 | zil_destroy(zilog, B_FALSE); | |
2262 | txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg); | |
fb5f0bc8 | 2263 | zilog->zl_replay = B_FALSE; |
34dc7c2f BB |
2264 | } |
2265 | ||
428870ff BB |
2266 | boolean_t |
2267 | zil_replaying(zilog_t *zilog, dmu_tx_t *tx) | |
34dc7c2f | 2268 | { |
428870ff BB |
2269 | if (zilog->zl_sync == ZFS_SYNC_DISABLED) |
2270 | return (B_TRUE); | |
34dc7c2f | 2271 | |
428870ff BB |
2272 | if (zilog->zl_replay) { |
2273 | dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); | |
2274 | zilog->zl_replayed_seq[dmu_tx_get_txg(tx) & TXG_MASK] = | |
2275 | zilog->zl_replaying_seq; | |
2276 | return (B_TRUE); | |
34dc7c2f BB |
2277 | } |
2278 | ||
428870ff | 2279 | return (B_FALSE); |
34dc7c2f | 2280 | } |
9babb374 BB |
2281 | |
2282 | /* ARGSUSED */ | |
2283 | int | |
428870ff | 2284 | zil_vdev_offline(const char *osname, void *arg) |
9babb374 | 2285 | { |
9babb374 BB |
2286 | int error; |
2287 | ||
a08ee875 LG |
2288 | error = zil_suspend(osname, NULL); |
2289 | if (error != 0) | |
2290 | return (SET_ERROR(EEXIST)); | |
2291 | return (0); | |
9babb374 | 2292 | } |
c409e464 BB |
2293 | |
2294 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
ea04106b AX |
2295 | EXPORT_SYMBOL(zil_alloc); |
2296 | EXPORT_SYMBOL(zil_free); | |
2297 | EXPORT_SYMBOL(zil_open); | |
2298 | EXPORT_SYMBOL(zil_close); | |
2299 | EXPORT_SYMBOL(zil_replay); | |
2300 | EXPORT_SYMBOL(zil_replaying); | |
2301 | EXPORT_SYMBOL(zil_destroy); | |
2302 | EXPORT_SYMBOL(zil_destroy_sync); | |
2303 | EXPORT_SYMBOL(zil_itx_create); | |
2304 | EXPORT_SYMBOL(zil_itx_destroy); | |
2305 | EXPORT_SYMBOL(zil_itx_assign); | |
2306 | EXPORT_SYMBOL(zil_commit); | |
2307 | EXPORT_SYMBOL(zil_vdev_offline); | |
2308 | EXPORT_SYMBOL(zil_claim); | |
2309 | EXPORT_SYMBOL(zil_check_log_chain); | |
2310 | EXPORT_SYMBOL(zil_sync); | |
2311 | EXPORT_SYMBOL(zil_clean); | |
2312 | EXPORT_SYMBOL(zil_suspend); | |
2313 | EXPORT_SYMBOL(zil_resume); | |
2314 | EXPORT_SYMBOL(zil_add_block); | |
2315 | EXPORT_SYMBOL(zil_bp_tree_add); | |
2316 | EXPORT_SYMBOL(zil_set_sync); | |
2317 | EXPORT_SYMBOL(zil_set_logbias); | |
2318 | ||
cae5b340 | 2319 | /* BEGIN CSTYLED */ |
c409e464 BB |
2320 | module_param(zil_replay_disable, int, 0644); |
2321 | MODULE_PARM_DESC(zil_replay_disable, "Disable intent logging replay"); | |
2322 | ||
2323 | module_param(zfs_nocacheflush, int, 0644); | |
2324 | MODULE_PARM_DESC(zfs_nocacheflush, "Disable cache flushes"); | |
ee191e80 | 2325 | |
cae5b340 AX |
2326 | module_param(zil_slog_bulk, ulong, 0644); |
2327 | MODULE_PARM_DESC(zil_slog_bulk, "Limit in bytes slog sync writes per commit"); | |
2328 | /* END CSTYLED */ | |
c409e464 | 2329 | #endif |