]>
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. |
5dbd68a3 | 23 | * Copyright (c) 2011, 2014 by Delphix. All rights reserved. |
34dc7c2f BB |
24 | */ |
25 | ||
428870ff BB |
26 | /* Portions Copyright 2010 Robert Milkowski */ |
27 | ||
34dc7c2f BB |
28 | #include <sys/zfs_context.h> |
29 | #include <sys/spa.h> | |
30 | #include <sys/dmu.h> | |
31 | #include <sys/zap.h> | |
32 | #include <sys/arc.h> | |
33 | #include <sys/stat.h> | |
34 | #include <sys/resource.h> | |
35 | #include <sys/zil.h> | |
36 | #include <sys/zil_impl.h> | |
37 | #include <sys/dsl_dataset.h> | |
572e2857 | 38 | #include <sys/vdev_impl.h> |
34dc7c2f | 39 | #include <sys/dmu_tx.h> |
428870ff | 40 | #include <sys/dsl_pool.h> |
920dd524 | 41 | #include <sys/metaslab.h> |
49ee64e5 | 42 | #include <sys/trace_zil.h> |
34dc7c2f BB |
43 | |
44 | /* | |
45 | * The zfs intent log (ZIL) saves transaction records of system calls | |
46 | * that change the file system in memory with enough information | |
47 | * to be able to replay them. These are stored in memory until | |
48 | * either the DMU transaction group (txg) commits them to the stable pool | |
49 | * and they can be discarded, or they are flushed to the stable log | |
50 | * (also in the pool) due to a fsync, O_DSYNC or other synchronous | |
51 | * requirement. In the event of a panic or power fail then those log | |
52 | * records (transactions) are replayed. | |
53 | * | |
54 | * There is one ZIL per file system. Its on-disk (pool) format consists | |
55 | * of 3 parts: | |
56 | * | |
57 | * - ZIL header | |
58 | * - ZIL blocks | |
59 | * - ZIL records | |
60 | * | |
61 | * A log record holds a system call transaction. Log blocks can | |
62 | * hold many log records and the blocks are chained together. | |
63 | * Each ZIL block contains a block pointer (blkptr_t) to the next | |
64 | * ZIL block in the chain. The ZIL header points to the first | |
65 | * block in the chain. Note there is not a fixed place in the pool | |
66 | * to hold blocks. They are dynamically allocated and freed as | |
67 | * needed from the blocks available. Figure X shows the ZIL structure: | |
68 | */ | |
69 | ||
b6ad9671 ED |
70 | /* |
71 | * See zil.h for more information about these fields. | |
72 | */ | |
73 | zil_stats_t zil_stats = { | |
d1d7e268 MK |
74 | { "zil_commit_count", KSTAT_DATA_UINT64 }, |
75 | { "zil_commit_writer_count", KSTAT_DATA_UINT64 }, | |
76 | { "zil_itx_count", KSTAT_DATA_UINT64 }, | |
77 | { "zil_itx_indirect_count", KSTAT_DATA_UINT64 }, | |
78 | { "zil_itx_indirect_bytes", KSTAT_DATA_UINT64 }, | |
79 | { "zil_itx_copied_count", KSTAT_DATA_UINT64 }, | |
80 | { "zil_itx_copied_bytes", KSTAT_DATA_UINT64 }, | |
81 | { "zil_itx_needcopy_count", KSTAT_DATA_UINT64 }, | |
82 | { "zil_itx_needcopy_bytes", KSTAT_DATA_UINT64 }, | |
83 | { "zil_itx_metaslab_normal_count", KSTAT_DATA_UINT64 }, | |
84 | { "zil_itx_metaslab_normal_bytes", KSTAT_DATA_UINT64 }, | |
85 | { "zil_itx_metaslab_slog_count", KSTAT_DATA_UINT64 }, | |
86 | { "zil_itx_metaslab_slog_bytes", KSTAT_DATA_UINT64 }, | |
b6ad9671 ED |
87 | }; |
88 | ||
89 | static kstat_t *zil_ksp; | |
90 | ||
34dc7c2f | 91 | /* |
d3cc8b15 | 92 | * Disable intent logging replay. This global ZIL switch affects all pools. |
34dc7c2f | 93 | */ |
d3cc8b15 | 94 | int zil_replay_disable = 0; |
34dc7c2f BB |
95 | |
96 | /* | |
97 | * Tunable parameter for debugging or performance analysis. Setting | |
98 | * zfs_nocacheflush will cause corruption on power loss if a volatile | |
99 | * out-of-order write cache is enabled. | |
100 | */ | |
c409e464 | 101 | int zfs_nocacheflush = 0; |
34dc7c2f BB |
102 | |
103 | static kmem_cache_t *zil_lwb_cache; | |
104 | ||
572e2857 | 105 | static void zil_async_to_sync(zilog_t *zilog, uint64_t foid); |
428870ff BB |
106 | |
107 | #define LWB_EMPTY(lwb) ((BP_GET_LSIZE(&lwb->lwb_blk) - \ | |
108 | sizeof (zil_chain_t)) == (lwb->lwb_sz - lwb->lwb_nused)) | |
109 | ||
110 | ||
572e2857 BB |
111 | /* |
112 | * ziltest is by and large an ugly hack, but very useful in | |
113 | * checking replay without tedious work. | |
114 | * When running ziltest we want to keep all itx's and so maintain | |
115 | * a single list in the zl_itxg[] that uses a high txg: ZILTEST_TXG | |
116 | * We subtract TXG_CONCURRENT_STATES to allow for common code. | |
117 | */ | |
118 | #define ZILTEST_TXG (UINT64_MAX - TXG_CONCURRENT_STATES) | |
119 | ||
34dc7c2f | 120 | static int |
428870ff | 121 | zil_bp_compare(const void *x1, const void *x2) |
34dc7c2f | 122 | { |
428870ff BB |
123 | const dva_t *dva1 = &((zil_bp_node_t *)x1)->zn_dva; |
124 | const dva_t *dva2 = &((zil_bp_node_t *)x2)->zn_dva; | |
34dc7c2f BB |
125 | |
126 | if (DVA_GET_VDEV(dva1) < DVA_GET_VDEV(dva2)) | |
127 | return (-1); | |
128 | if (DVA_GET_VDEV(dva1) > DVA_GET_VDEV(dva2)) | |
129 | return (1); | |
130 | ||
131 | if (DVA_GET_OFFSET(dva1) < DVA_GET_OFFSET(dva2)) | |
132 | return (-1); | |
133 | if (DVA_GET_OFFSET(dva1) > DVA_GET_OFFSET(dva2)) | |
134 | return (1); | |
135 | ||
136 | return (0); | |
137 | } | |
138 | ||
139 | static void | |
428870ff | 140 | zil_bp_tree_init(zilog_t *zilog) |
34dc7c2f | 141 | { |
428870ff BB |
142 | avl_create(&zilog->zl_bp_tree, zil_bp_compare, |
143 | sizeof (zil_bp_node_t), offsetof(zil_bp_node_t, zn_node)); | |
34dc7c2f BB |
144 | } |
145 | ||
146 | static void | |
428870ff | 147 | zil_bp_tree_fini(zilog_t *zilog) |
34dc7c2f | 148 | { |
428870ff BB |
149 | avl_tree_t *t = &zilog->zl_bp_tree; |
150 | zil_bp_node_t *zn; | |
34dc7c2f BB |
151 | void *cookie = NULL; |
152 | ||
153 | while ((zn = avl_destroy_nodes(t, &cookie)) != NULL) | |
428870ff | 154 | kmem_free(zn, sizeof (zil_bp_node_t)); |
34dc7c2f BB |
155 | |
156 | avl_destroy(t); | |
157 | } | |
158 | ||
428870ff BB |
159 | int |
160 | zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp) | |
34dc7c2f | 161 | { |
428870ff | 162 | avl_tree_t *t = &zilog->zl_bp_tree; |
9b67f605 | 163 | const dva_t *dva; |
428870ff | 164 | zil_bp_node_t *zn; |
34dc7c2f BB |
165 | avl_index_t where; |
166 | ||
9b67f605 MA |
167 | if (BP_IS_EMBEDDED(bp)) |
168 | return (0); | |
169 | ||
170 | dva = BP_IDENTITY(bp); | |
171 | ||
34dc7c2f | 172 | if (avl_find(t, dva, &where) != NULL) |
2e528b49 | 173 | return (SET_ERROR(EEXIST)); |
34dc7c2f | 174 | |
79c76d5b | 175 | zn = kmem_alloc(sizeof (zil_bp_node_t), KM_SLEEP); |
34dc7c2f BB |
176 | zn->zn_dva = *dva; |
177 | avl_insert(t, zn, where); | |
178 | ||
179 | return (0); | |
180 | } | |
181 | ||
182 | static zil_header_t * | |
183 | zil_header_in_syncing_context(zilog_t *zilog) | |
184 | { | |
185 | return ((zil_header_t *)zilog->zl_header); | |
186 | } | |
187 | ||
188 | static void | |
189 | zil_init_log_chain(zilog_t *zilog, blkptr_t *bp) | |
190 | { | |
191 | zio_cksum_t *zc = &bp->blk_cksum; | |
192 | ||
193 | zc->zc_word[ZIL_ZC_GUID_0] = spa_get_random(-1ULL); | |
194 | zc->zc_word[ZIL_ZC_GUID_1] = spa_get_random(-1ULL); | |
195 | zc->zc_word[ZIL_ZC_OBJSET] = dmu_objset_id(zilog->zl_os); | |
196 | zc->zc_word[ZIL_ZC_SEQ] = 1ULL; | |
197 | } | |
198 | ||
199 | /* | |
428870ff | 200 | * Read a log block and make sure it's valid. |
34dc7c2f BB |
201 | */ |
202 | static int | |
428870ff BB |
203 | zil_read_log_block(zilog_t *zilog, const blkptr_t *bp, blkptr_t *nbp, void *dst, |
204 | char **end) | |
34dc7c2f | 205 | { |
428870ff | 206 | enum zio_flag zio_flags = ZIO_FLAG_CANFAIL; |
34dc7c2f | 207 | uint32_t aflags = ARC_WAIT; |
428870ff | 208 | arc_buf_t *abuf = NULL; |
5dbd68a3 | 209 | zbookmark_phys_t zb; |
34dc7c2f BB |
210 | int error; |
211 | ||
428870ff BB |
212 | if (zilog->zl_header->zh_claim_txg == 0) |
213 | zio_flags |= ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB; | |
34dc7c2f | 214 | |
428870ff BB |
215 | if (!(zilog->zl_header->zh_flags & ZIL_CLAIM_LR_SEQ_VALID)) |
216 | zio_flags |= ZIO_FLAG_SPECULATIVE; | |
34dc7c2f | 217 | |
428870ff BB |
218 | SET_BOOKMARK(&zb, bp->blk_cksum.zc_word[ZIL_ZC_OBJSET], |
219 | ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, bp->blk_cksum.zc_word[ZIL_ZC_SEQ]); | |
220 | ||
294f6806 | 221 | error = arc_read(NULL, zilog->zl_spa, bp, arc_getbuf_func, &abuf, |
428870ff | 222 | ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb); |
34dc7c2f BB |
223 | |
224 | if (error == 0) { | |
34dc7c2f BB |
225 | zio_cksum_t cksum = bp->blk_cksum; |
226 | ||
227 | /* | |
b128c09f BB |
228 | * Validate the checksummed log block. |
229 | * | |
34dc7c2f BB |
230 | * Sequence numbers should be... sequential. The checksum |
231 | * verifier for the next block should be bp's checksum plus 1. | |
b128c09f BB |
232 | * |
233 | * Also check the log chain linkage and size used. | |
34dc7c2f BB |
234 | */ |
235 | cksum.zc_word[ZIL_ZC_SEQ]++; | |
236 | ||
428870ff BB |
237 | if (BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_ZILOG2) { |
238 | zil_chain_t *zilc = abuf->b_data; | |
239 | char *lr = (char *)(zilc + 1); | |
240 | uint64_t len = zilc->zc_nused - sizeof (zil_chain_t); | |
34dc7c2f | 241 | |
428870ff BB |
242 | if (bcmp(&cksum, &zilc->zc_next_blk.blk_cksum, |
243 | sizeof (cksum)) || BP_IS_HOLE(&zilc->zc_next_blk)) { | |
2e528b49 | 244 | error = SET_ERROR(ECKSUM); |
428870ff BB |
245 | } else { |
246 | bcopy(lr, dst, len); | |
247 | *end = (char *)dst + len; | |
248 | *nbp = zilc->zc_next_blk; | |
249 | } | |
250 | } else { | |
251 | char *lr = abuf->b_data; | |
252 | uint64_t size = BP_GET_LSIZE(bp); | |
253 | zil_chain_t *zilc = (zil_chain_t *)(lr + size) - 1; | |
254 | ||
255 | if (bcmp(&cksum, &zilc->zc_next_blk.blk_cksum, | |
256 | sizeof (cksum)) || BP_IS_HOLE(&zilc->zc_next_blk) || | |
257 | (zilc->zc_nused > (size - sizeof (*zilc)))) { | |
2e528b49 | 258 | error = SET_ERROR(ECKSUM); |
428870ff BB |
259 | } else { |
260 | bcopy(lr, dst, zilc->zc_nused); | |
261 | *end = (char *)dst + zilc->zc_nused; | |
262 | *nbp = zilc->zc_next_blk; | |
263 | } | |
34dc7c2f | 264 | } |
428870ff | 265 | |
13fe0198 | 266 | VERIFY(arc_buf_remove_ref(abuf, &abuf)); |
428870ff BB |
267 | } |
268 | ||
269 | return (error); | |
270 | } | |
271 | ||
272 | /* | |
273 | * Read a TX_WRITE log data block. | |
274 | */ | |
275 | static int | |
276 | zil_read_log_data(zilog_t *zilog, const lr_write_t *lr, void *wbuf) | |
277 | { | |
278 | enum zio_flag zio_flags = ZIO_FLAG_CANFAIL; | |
279 | const blkptr_t *bp = &lr->lr_blkptr; | |
280 | uint32_t aflags = ARC_WAIT; | |
281 | arc_buf_t *abuf = NULL; | |
5dbd68a3 | 282 | zbookmark_phys_t zb; |
428870ff BB |
283 | int error; |
284 | ||
285 | if (BP_IS_HOLE(bp)) { | |
286 | if (wbuf != NULL) | |
287 | bzero(wbuf, MAX(BP_GET_LSIZE(bp), lr->lr_length)); | |
288 | return (0); | |
34dc7c2f BB |
289 | } |
290 | ||
428870ff BB |
291 | if (zilog->zl_header->zh_claim_txg == 0) |
292 | zio_flags |= ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB; | |
293 | ||
294 | SET_BOOKMARK(&zb, dmu_objset_id(zilog->zl_os), lr->lr_foid, | |
295 | ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp)); | |
296 | ||
294f6806 | 297 | error = arc_read(NULL, zilog->zl_spa, bp, arc_getbuf_func, &abuf, |
428870ff BB |
298 | ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb); |
299 | ||
300 | if (error == 0) { | |
301 | if (wbuf != NULL) | |
302 | bcopy(abuf->b_data, wbuf, arc_buf_size(abuf)); | |
303 | (void) arc_buf_remove_ref(abuf, &abuf); | |
304 | } | |
34dc7c2f BB |
305 | |
306 | return (error); | |
307 | } | |
308 | ||
309 | /* | |
310 | * Parse the intent log, and call parse_func for each valid record within. | |
34dc7c2f | 311 | */ |
428870ff | 312 | int |
34dc7c2f BB |
313 | zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func, |
314 | zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg) | |
315 | { | |
316 | const zil_header_t *zh = zilog->zl_header; | |
428870ff BB |
317 | boolean_t claimed = !!zh->zh_claim_txg; |
318 | uint64_t claim_blk_seq = claimed ? zh->zh_claim_blk_seq : UINT64_MAX; | |
319 | uint64_t claim_lr_seq = claimed ? zh->zh_claim_lr_seq : UINT64_MAX; | |
320 | uint64_t max_blk_seq = 0; | |
321 | uint64_t max_lr_seq = 0; | |
322 | uint64_t blk_count = 0; | |
323 | uint64_t lr_count = 0; | |
324 | blkptr_t blk, next_blk; | |
34dc7c2f | 325 | char *lrbuf, *lrp; |
428870ff | 326 | int error = 0; |
34dc7c2f | 327 | |
d1d7e268 | 328 | bzero(&next_blk, sizeof (blkptr_t)); |
d4ed6673 | 329 | |
428870ff BB |
330 | /* |
331 | * Old logs didn't record the maximum zh_claim_lr_seq. | |
332 | */ | |
333 | if (!(zh->zh_flags & ZIL_CLAIM_LR_SEQ_VALID)) | |
334 | claim_lr_seq = UINT64_MAX; | |
34dc7c2f BB |
335 | |
336 | /* | |
337 | * Starting at the block pointed to by zh_log we read the log chain. | |
338 | * For each block in the chain we strongly check that block to | |
339 | * ensure its validity. We stop when an invalid block is found. | |
340 | * For each block pointer in the chain we call parse_blk_func(). | |
341 | * For each record in each valid block we call parse_lr_func(). | |
342 | * If the log has been claimed, stop if we encounter a sequence | |
343 | * number greater than the highest claimed sequence number. | |
344 | */ | |
428870ff BB |
345 | lrbuf = zio_buf_alloc(SPA_MAXBLOCKSIZE); |
346 | zil_bp_tree_init(zilog); | |
34dc7c2f | 347 | |
428870ff BB |
348 | for (blk = zh->zh_log; !BP_IS_HOLE(&blk); blk = next_blk) { |
349 | uint64_t blk_seq = blk.blk_cksum.zc_word[ZIL_ZC_SEQ]; | |
350 | int reclen; | |
d4ed6673 | 351 | char *end = NULL; |
34dc7c2f | 352 | |
428870ff BB |
353 | if (blk_seq > claim_blk_seq) |
354 | break; | |
355 | if ((error = parse_blk_func(zilog, &blk, arg, txg)) != 0) | |
356 | break; | |
357 | ASSERT3U(max_blk_seq, <, blk_seq); | |
358 | max_blk_seq = blk_seq; | |
359 | blk_count++; | |
34dc7c2f | 360 | |
428870ff BB |
361 | if (max_lr_seq == claim_lr_seq && max_blk_seq == claim_blk_seq) |
362 | break; | |
34dc7c2f | 363 | |
428870ff | 364 | error = zil_read_log_block(zilog, &blk, &next_blk, lrbuf, &end); |
13fe0198 | 365 | if (error != 0) |
34dc7c2f BB |
366 | break; |
367 | ||
428870ff | 368 | for (lrp = lrbuf; lrp < end; lrp += reclen) { |
34dc7c2f BB |
369 | lr_t *lr = (lr_t *)lrp; |
370 | reclen = lr->lrc_reclen; | |
371 | ASSERT3U(reclen, >=, sizeof (lr_t)); | |
428870ff BB |
372 | if (lr->lrc_seq > claim_lr_seq) |
373 | goto done; | |
374 | if ((error = parse_lr_func(zilog, lr, arg, txg)) != 0) | |
375 | goto done; | |
376 | ASSERT3U(max_lr_seq, <, lr->lrc_seq); | |
377 | max_lr_seq = lr->lrc_seq; | |
378 | lr_count++; | |
34dc7c2f | 379 | } |
34dc7c2f | 380 | } |
428870ff BB |
381 | done: |
382 | zilog->zl_parse_error = error; | |
383 | zilog->zl_parse_blk_seq = max_blk_seq; | |
384 | zilog->zl_parse_lr_seq = max_lr_seq; | |
385 | zilog->zl_parse_blk_count = blk_count; | |
386 | zilog->zl_parse_lr_count = lr_count; | |
387 | ||
388 | ASSERT(!claimed || !(zh->zh_flags & ZIL_CLAIM_LR_SEQ_VALID) || | |
389 | (max_blk_seq == claim_blk_seq && max_lr_seq == claim_lr_seq)); | |
390 | ||
391 | zil_bp_tree_fini(zilog); | |
392 | zio_buf_free(lrbuf, SPA_MAXBLOCKSIZE); | |
34dc7c2f | 393 | |
428870ff | 394 | return (error); |
34dc7c2f BB |
395 | } |
396 | ||
428870ff | 397 | static int |
34dc7c2f BB |
398 | zil_claim_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t first_txg) |
399 | { | |
34dc7c2f BB |
400 | /* |
401 | * Claim log block if not already committed and not already claimed. | |
428870ff | 402 | * If tx == NULL, just verify that the block is claimable. |
34dc7c2f | 403 | */ |
b0bc7a84 MG |
404 | if (BP_IS_HOLE(bp) || bp->blk_birth < first_txg || |
405 | zil_bp_tree_add(zilog, bp) != 0) | |
428870ff BB |
406 | return (0); |
407 | ||
408 | return (zio_wait(zio_claim(NULL, zilog->zl_spa, | |
409 | tx == NULL ? 0 : first_txg, bp, spa_claim_notify, NULL, | |
410 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB))); | |
34dc7c2f BB |
411 | } |
412 | ||
428870ff | 413 | static int |
34dc7c2f BB |
414 | zil_claim_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t first_txg) |
415 | { | |
428870ff BB |
416 | lr_write_t *lr = (lr_write_t *)lrc; |
417 | int error; | |
418 | ||
419 | if (lrc->lrc_txtype != TX_WRITE) | |
420 | return (0); | |
421 | ||
422 | /* | |
423 | * If the block is not readable, don't claim it. This can happen | |
424 | * in normal operation when a log block is written to disk before | |
425 | * some of the dmu_sync() blocks it points to. In this case, the | |
426 | * transaction cannot have been committed to anyone (we would have | |
427 | * waited for all writes to be stable first), so it is semantically | |
428 | * correct to declare this the end of the log. | |
429 | */ | |
430 | if (lr->lr_blkptr.blk_birth >= first_txg && | |
431 | (error = zil_read_log_data(zilog, lr, NULL)) != 0) | |
432 | return (error); | |
433 | return (zil_claim_log_block(zilog, &lr->lr_blkptr, tx, first_txg)); | |
34dc7c2f BB |
434 | } |
435 | ||
436 | /* ARGSUSED */ | |
428870ff | 437 | static int |
34dc7c2f BB |
438 | zil_free_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t claim_txg) |
439 | { | |
428870ff BB |
440 | zio_free_zil(zilog->zl_spa, dmu_tx_get_txg(tx), bp); |
441 | ||
442 | return (0); | |
34dc7c2f BB |
443 | } |
444 | ||
428870ff | 445 | static int |
34dc7c2f BB |
446 | zil_free_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t claim_txg) |
447 | { | |
428870ff BB |
448 | lr_write_t *lr = (lr_write_t *)lrc; |
449 | blkptr_t *bp = &lr->lr_blkptr; | |
450 | ||
34dc7c2f BB |
451 | /* |
452 | * If we previously claimed it, we need to free it. | |
453 | */ | |
428870ff | 454 | if (claim_txg != 0 && lrc->lrc_txtype == TX_WRITE && |
b0bc7a84 MG |
455 | bp->blk_birth >= claim_txg && zil_bp_tree_add(zilog, bp) == 0 && |
456 | !BP_IS_HOLE(bp)) | |
428870ff BB |
457 | zio_free(zilog->zl_spa, dmu_tx_get_txg(tx), bp); |
458 | ||
459 | return (0); | |
460 | } | |
461 | ||
462 | static lwb_t * | |
920dd524 | 463 | zil_alloc_lwb(zilog_t *zilog, blkptr_t *bp, uint64_t txg, boolean_t fastwrite) |
428870ff BB |
464 | { |
465 | lwb_t *lwb; | |
466 | ||
79c76d5b | 467 | lwb = kmem_cache_alloc(zil_lwb_cache, KM_SLEEP); |
428870ff BB |
468 | lwb->lwb_zilog = zilog; |
469 | lwb->lwb_blk = *bp; | |
920dd524 | 470 | lwb->lwb_fastwrite = fastwrite; |
428870ff BB |
471 | lwb->lwb_buf = zio_buf_alloc(BP_GET_LSIZE(bp)); |
472 | lwb->lwb_max_txg = txg; | |
473 | lwb->lwb_zio = NULL; | |
474 | lwb->lwb_tx = NULL; | |
475 | if (BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_ZILOG2) { | |
476 | lwb->lwb_nused = sizeof (zil_chain_t); | |
477 | lwb->lwb_sz = BP_GET_LSIZE(bp); | |
478 | } else { | |
479 | lwb->lwb_nused = 0; | |
480 | lwb->lwb_sz = BP_GET_LSIZE(bp) - sizeof (zil_chain_t); | |
34dc7c2f | 481 | } |
428870ff BB |
482 | |
483 | mutex_enter(&zilog->zl_lock); | |
484 | list_insert_tail(&zilog->zl_lwb_list, lwb); | |
485 | mutex_exit(&zilog->zl_lock); | |
486 | ||
487 | return (lwb); | |
34dc7c2f BB |
488 | } |
489 | ||
29809a6c MA |
490 | /* |
491 | * Called when we create in-memory log transactions so that we know | |
492 | * to cleanup the itxs at the end of spa_sync(). | |
493 | */ | |
494 | void | |
495 | zilog_dirty(zilog_t *zilog, uint64_t txg) | |
496 | { | |
497 | dsl_pool_t *dp = zilog->zl_dmu_pool; | |
498 | dsl_dataset_t *ds = dmu_objset_ds(zilog->zl_os); | |
499 | ||
0c66c32d | 500 | if (ds->ds_is_snapshot) |
29809a6c MA |
501 | panic("dirtying snapshot!"); |
502 | ||
13fe0198 | 503 | if (txg_list_add(&dp->dp_dirty_zilogs, zilog, txg)) { |
29809a6c MA |
504 | /* up the hold count until we can be written out */ |
505 | dmu_buf_add_ref(ds->ds_dbuf, zilog); | |
506 | } | |
507 | } | |
508 | ||
509 | boolean_t | |
510 | zilog_is_dirty(zilog_t *zilog) | |
511 | { | |
512 | dsl_pool_t *dp = zilog->zl_dmu_pool; | |
513 | int t; | |
514 | ||
515 | for (t = 0; t < TXG_SIZE; t++) { | |
516 | if (txg_list_member(&dp->dp_dirty_zilogs, zilog, t)) | |
517 | return (B_TRUE); | |
518 | } | |
519 | return (B_FALSE); | |
520 | } | |
521 | ||
34dc7c2f BB |
522 | /* |
523 | * Create an on-disk intent log. | |
524 | */ | |
428870ff | 525 | static lwb_t * |
34dc7c2f BB |
526 | zil_create(zilog_t *zilog) |
527 | { | |
528 | const zil_header_t *zh = zilog->zl_header; | |
428870ff | 529 | lwb_t *lwb = NULL; |
34dc7c2f BB |
530 | uint64_t txg = 0; |
531 | dmu_tx_t *tx = NULL; | |
532 | blkptr_t blk; | |
533 | int error = 0; | |
920dd524 | 534 | boolean_t fastwrite = FALSE; |
34dc7c2f BB |
535 | |
536 | /* | |
537 | * Wait for any previous destroy to complete. | |
538 | */ | |
539 | txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg); | |
540 | ||
541 | ASSERT(zh->zh_claim_txg == 0); | |
542 | ASSERT(zh->zh_replay_seq == 0); | |
543 | ||
544 | blk = zh->zh_log; | |
545 | ||
546 | /* | |
428870ff BB |
547 | * Allocate an initial log block if: |
548 | * - there isn't one already | |
549 | * - the existing block is the wrong endianess | |
34dc7c2f | 550 | */ |
fb5f0bc8 | 551 | if (BP_IS_HOLE(&blk) || BP_SHOULD_BYTESWAP(&blk)) { |
34dc7c2f | 552 | tx = dmu_tx_create(zilog->zl_os); |
428870ff | 553 | VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0); |
34dc7c2f BB |
554 | dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); |
555 | txg = dmu_tx_get_txg(tx); | |
556 | ||
fb5f0bc8 | 557 | if (!BP_IS_HOLE(&blk)) { |
428870ff | 558 | zio_free_zil(zilog->zl_spa, txg, &blk); |
fb5f0bc8 BB |
559 | BP_ZERO(&blk); |
560 | } | |
561 | ||
920dd524 | 562 | error = zio_alloc_zil(zilog->zl_spa, txg, &blk, |
5d7a86d1 | 563 | ZIL_MIN_BLKSZ, B_TRUE); |
920dd524 | 564 | fastwrite = TRUE; |
34dc7c2f BB |
565 | |
566 | if (error == 0) | |
567 | zil_init_log_chain(zilog, &blk); | |
568 | } | |
569 | ||
570 | /* | |
571 | * Allocate a log write buffer (lwb) for the first log block. | |
572 | */ | |
428870ff | 573 | if (error == 0) |
920dd524 | 574 | lwb = zil_alloc_lwb(zilog, &blk, txg, fastwrite); |
34dc7c2f BB |
575 | |
576 | /* | |
577 | * If we just allocated the first log block, commit our transaction | |
578 | * and wait for zil_sync() to stuff the block poiner into zh_log. | |
579 | * (zh is part of the MOS, so we cannot modify it in open context.) | |
580 | */ | |
581 | if (tx != NULL) { | |
582 | dmu_tx_commit(tx); | |
583 | txg_wait_synced(zilog->zl_dmu_pool, txg); | |
584 | } | |
585 | ||
586 | ASSERT(bcmp(&blk, &zh->zh_log, sizeof (blk)) == 0); | |
428870ff BB |
587 | |
588 | return (lwb); | |
34dc7c2f BB |
589 | } |
590 | ||
591 | /* | |
592 | * In one tx, free all log blocks and clear the log header. | |
593 | * If keep_first is set, then we're replaying a log with no content. | |
594 | * We want to keep the first block, however, so that the first | |
595 | * synchronous transaction doesn't require a txg_wait_synced() | |
596 | * in zil_create(). We don't need to txg_wait_synced() here either | |
597 | * when keep_first is set, because both zil_create() and zil_destroy() | |
598 | * will wait for any in-progress destroys to complete. | |
599 | */ | |
600 | void | |
601 | zil_destroy(zilog_t *zilog, boolean_t keep_first) | |
602 | { | |
603 | const zil_header_t *zh = zilog->zl_header; | |
604 | lwb_t *lwb; | |
605 | dmu_tx_t *tx; | |
606 | uint64_t txg; | |
607 | ||
608 | /* | |
609 | * Wait for any previous destroy to complete. | |
610 | */ | |
611 | txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg); | |
612 | ||
428870ff BB |
613 | zilog->zl_old_header = *zh; /* debugging aid */ |
614 | ||
34dc7c2f BB |
615 | if (BP_IS_HOLE(&zh->zh_log)) |
616 | return; | |
617 | ||
618 | tx = dmu_tx_create(zilog->zl_os); | |
428870ff | 619 | VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0); |
34dc7c2f BB |
620 | dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); |
621 | txg = dmu_tx_get_txg(tx); | |
622 | ||
623 | mutex_enter(&zilog->zl_lock); | |
624 | ||
34dc7c2f BB |
625 | ASSERT3U(zilog->zl_destroy_txg, <, txg); |
626 | zilog->zl_destroy_txg = txg; | |
627 | zilog->zl_keep_first = keep_first; | |
628 | ||
629 | if (!list_is_empty(&zilog->zl_lwb_list)) { | |
630 | ASSERT(zh->zh_claim_txg == 0); | |
3e31d2b0 | 631 | VERIFY(!keep_first); |
34dc7c2f | 632 | while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) { |
920dd524 ED |
633 | ASSERT(lwb->lwb_zio == NULL); |
634 | if (lwb->lwb_fastwrite) | |
635 | metaslab_fastwrite_unmark(zilog->zl_spa, | |
636 | &lwb->lwb_blk); | |
34dc7c2f BB |
637 | list_remove(&zilog->zl_lwb_list, lwb); |
638 | if (lwb->lwb_buf != NULL) | |
639 | zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); | |
428870ff | 640 | zio_free_zil(zilog->zl_spa, txg, &lwb->lwb_blk); |
34dc7c2f BB |
641 | kmem_cache_free(zil_lwb_cache, lwb); |
642 | } | |
428870ff | 643 | } else if (!keep_first) { |
29809a6c | 644 | zil_destroy_sync(zilog, tx); |
34dc7c2f BB |
645 | } |
646 | mutex_exit(&zilog->zl_lock); | |
647 | ||
648 | dmu_tx_commit(tx); | |
649 | } | |
650 | ||
29809a6c MA |
651 | void |
652 | zil_destroy_sync(zilog_t *zilog, dmu_tx_t *tx) | |
653 | { | |
654 | ASSERT(list_is_empty(&zilog->zl_lwb_list)); | |
655 | (void) zil_parse(zilog, zil_free_log_block, | |
656 | zil_free_log_record, tx, zilog->zl_header->zh_claim_txg); | |
657 | } | |
658 | ||
34dc7c2f | 659 | int |
428870ff | 660 | zil_claim(const char *osname, void *txarg) |
34dc7c2f BB |
661 | { |
662 | dmu_tx_t *tx = txarg; | |
663 | uint64_t first_txg = dmu_tx_get_txg(tx); | |
664 | zilog_t *zilog; | |
665 | zil_header_t *zh; | |
666 | objset_t *os; | |
667 | int error; | |
668 | ||
13fe0198 MA |
669 | error = dmu_objset_own(osname, DMU_OST_ANY, B_FALSE, FTAG, &os); |
670 | if (error != 0) { | |
6d9036f3 MA |
671 | /* |
672 | * EBUSY indicates that the objset is inconsistent, in which | |
673 | * case it can not have a ZIL. | |
674 | */ | |
675 | if (error != EBUSY) { | |
676 | cmn_err(CE_WARN, "can't open objset for %s, error %u", | |
677 | osname, error); | |
678 | } | |
679 | ||
34dc7c2f BB |
680 | return (0); |
681 | } | |
682 | ||
683 | zilog = dmu_objset_zil(os); | |
684 | zh = zil_header_in_syncing_context(zilog); | |
685 | ||
428870ff | 686 | if (spa_get_log_state(zilog->zl_spa) == SPA_LOG_CLEAR) { |
9babb374 | 687 | if (!BP_IS_HOLE(&zh->zh_log)) |
428870ff | 688 | zio_free_zil(zilog->zl_spa, first_txg, &zh->zh_log); |
9babb374 BB |
689 | BP_ZERO(&zh->zh_log); |
690 | dsl_dataset_dirty(dmu_objset_ds(os), tx); | |
13fe0198 | 691 | dmu_objset_disown(os, FTAG); |
428870ff | 692 | return (0); |
9babb374 BB |
693 | } |
694 | ||
34dc7c2f BB |
695 | /* |
696 | * Claim all log blocks if we haven't already done so, and remember | |
697 | * the highest claimed sequence number. This ensures that if we can | |
698 | * read only part of the log now (e.g. due to a missing device), | |
699 | * but we can read the entire log later, we will not try to replay | |
700 | * or destroy beyond the last block we successfully claimed. | |
701 | */ | |
702 | ASSERT3U(zh->zh_claim_txg, <=, first_txg); | |
703 | if (zh->zh_claim_txg == 0 && !BP_IS_HOLE(&zh->zh_log)) { | |
428870ff | 704 | (void) zil_parse(zilog, zil_claim_log_block, |
34dc7c2f | 705 | zil_claim_log_record, tx, first_txg); |
428870ff BB |
706 | zh->zh_claim_txg = first_txg; |
707 | zh->zh_claim_blk_seq = zilog->zl_parse_blk_seq; | |
708 | zh->zh_claim_lr_seq = zilog->zl_parse_lr_seq; | |
709 | if (zilog->zl_parse_lr_count || zilog->zl_parse_blk_count > 1) | |
710 | zh->zh_flags |= ZIL_REPLAY_NEEDED; | |
711 | zh->zh_flags |= ZIL_CLAIM_LR_SEQ_VALID; | |
34dc7c2f BB |
712 | dsl_dataset_dirty(dmu_objset_ds(os), tx); |
713 | } | |
714 | ||
715 | ASSERT3U(first_txg, ==, (spa_last_synced_txg(zilog->zl_spa) + 1)); | |
13fe0198 | 716 | dmu_objset_disown(os, FTAG); |
34dc7c2f BB |
717 | return (0); |
718 | } | |
719 | ||
b128c09f BB |
720 | /* |
721 | * Check the log by walking the log chain. | |
722 | * Checksum errors are ok as they indicate the end of the chain. | |
723 | * Any other error (no device or read failure) returns an error. | |
724 | */ | |
b128c09f | 725 | int |
428870ff | 726 | zil_check_log_chain(const char *osname, void *tx) |
b128c09f BB |
727 | { |
728 | zilog_t *zilog; | |
b128c09f | 729 | objset_t *os; |
572e2857 | 730 | blkptr_t *bp; |
b128c09f BB |
731 | int error; |
732 | ||
428870ff BB |
733 | ASSERT(tx == NULL); |
734 | ||
735 | error = dmu_objset_hold(osname, FTAG, &os); | |
13fe0198 | 736 | if (error != 0) { |
b128c09f BB |
737 | cmn_err(CE_WARN, "can't open objset for %s", osname); |
738 | return (0); | |
739 | } | |
740 | ||
741 | zilog = dmu_objset_zil(os); | |
572e2857 BB |
742 | bp = (blkptr_t *)&zilog->zl_header->zh_log; |
743 | ||
744 | /* | |
745 | * Check the first block and determine if it's on a log device | |
746 | * which may have been removed or faulted prior to loading this | |
747 | * pool. If so, there's no point in checking the rest of the log | |
748 | * as its content should have already been synced to the pool. | |
749 | */ | |
750 | if (!BP_IS_HOLE(bp)) { | |
751 | vdev_t *vd; | |
752 | boolean_t valid = B_TRUE; | |
753 | ||
754 | spa_config_enter(os->os_spa, SCL_STATE, FTAG, RW_READER); | |
755 | vd = vdev_lookup_top(os->os_spa, DVA_GET_VDEV(&bp->blk_dva[0])); | |
756 | if (vd->vdev_islog && vdev_is_dead(vd)) | |
757 | valid = vdev_log_state_valid(vd); | |
758 | spa_config_exit(os->os_spa, SCL_STATE, FTAG); | |
759 | ||
760 | if (!valid) { | |
761 | dmu_objset_rele(os, FTAG); | |
762 | return (0); | |
763 | } | |
764 | } | |
b128c09f | 765 | |
428870ff BB |
766 | /* |
767 | * Because tx == NULL, zil_claim_log_block() will not actually claim | |
768 | * any blocks, but just determine whether it is possible to do so. | |
769 | * In addition to checking the log chain, zil_claim_log_block() | |
770 | * will invoke zio_claim() with a done func of spa_claim_notify(), | |
771 | * which will update spa_max_claim_txg. See spa_load() for details. | |
772 | */ | |
773 | error = zil_parse(zilog, zil_claim_log_block, zil_claim_log_record, tx, | |
774 | zilog->zl_header->zh_claim_txg ? -1ULL : spa_first_txg(os->os_spa)); | |
775 | ||
776 | dmu_objset_rele(os, FTAG); | |
777 | ||
778 | return ((error == ECKSUM || error == ENOENT) ? 0 : error); | |
b128c09f BB |
779 | } |
780 | ||
34dc7c2f BB |
781 | static int |
782 | zil_vdev_compare(const void *x1, const void *x2) | |
783 | { | |
572e2857 BB |
784 | const uint64_t v1 = ((zil_vdev_node_t *)x1)->zv_vdev; |
785 | const uint64_t v2 = ((zil_vdev_node_t *)x2)->zv_vdev; | |
34dc7c2f BB |
786 | |
787 | if (v1 < v2) | |
788 | return (-1); | |
789 | if (v1 > v2) | |
790 | return (1); | |
791 | ||
792 | return (0); | |
793 | } | |
794 | ||
795 | void | |
428870ff | 796 | zil_add_block(zilog_t *zilog, const blkptr_t *bp) |
34dc7c2f BB |
797 | { |
798 | avl_tree_t *t = &zilog->zl_vdev_tree; | |
799 | avl_index_t where; | |
800 | zil_vdev_node_t *zv, zvsearch; | |
801 | int ndvas = BP_GET_NDVAS(bp); | |
802 | int i; | |
803 | ||
804 | if (zfs_nocacheflush) | |
805 | return; | |
806 | ||
807 | ASSERT(zilog->zl_writer); | |
808 | ||
809 | /* | |
810 | * Even though we're zl_writer, we still need a lock because the | |
811 | * zl_get_data() callbacks may have dmu_sync() done callbacks | |
812 | * that will run concurrently. | |
813 | */ | |
814 | mutex_enter(&zilog->zl_vdev_lock); | |
815 | for (i = 0; i < ndvas; i++) { | |
816 | zvsearch.zv_vdev = DVA_GET_VDEV(&bp->blk_dva[i]); | |
817 | if (avl_find(t, &zvsearch, &where) == NULL) { | |
79c76d5b | 818 | zv = kmem_alloc(sizeof (*zv), KM_SLEEP); |
34dc7c2f BB |
819 | zv->zv_vdev = zvsearch.zv_vdev; |
820 | avl_insert(t, zv, where); | |
821 | } | |
822 | } | |
823 | mutex_exit(&zilog->zl_vdev_lock); | |
824 | } | |
825 | ||
572e2857 | 826 | static void |
34dc7c2f BB |
827 | zil_flush_vdevs(zilog_t *zilog) |
828 | { | |
829 | spa_t *spa = zilog->zl_spa; | |
830 | avl_tree_t *t = &zilog->zl_vdev_tree; | |
831 | void *cookie = NULL; | |
832 | zil_vdev_node_t *zv; | |
833 | zio_t *zio; | |
834 | ||
835 | ASSERT(zilog->zl_writer); | |
836 | ||
837 | /* | |
838 | * We don't need zl_vdev_lock here because we're the zl_writer, | |
839 | * and all zl_get_data() callbacks are done. | |
840 | */ | |
841 | if (avl_numnodes(t) == 0) | |
842 | return; | |
843 | ||
b128c09f | 844 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); |
34dc7c2f | 845 | |
b128c09f | 846 | zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL); |
34dc7c2f BB |
847 | |
848 | while ((zv = avl_destroy_nodes(t, &cookie)) != NULL) { | |
849 | vdev_t *vd = vdev_lookup_top(spa, zv->zv_vdev); | |
850 | if (vd != NULL) | |
851 | zio_flush(zio, vd); | |
852 | kmem_free(zv, sizeof (*zv)); | |
853 | } | |
854 | ||
855 | /* | |
856 | * Wait for all the flushes to complete. Not all devices actually | |
857 | * support the DKIOCFLUSHWRITECACHE ioctl, so it's OK if it fails. | |
858 | */ | |
859 | (void) zio_wait(zio); | |
860 | ||
b128c09f | 861 | spa_config_exit(spa, SCL_STATE, FTAG); |
34dc7c2f BB |
862 | } |
863 | ||
864 | /* | |
865 | * Function called when a log block write completes | |
866 | */ | |
867 | static void | |
868 | zil_lwb_write_done(zio_t *zio) | |
869 | { | |
870 | lwb_t *lwb = zio->io_private; | |
871 | zilog_t *zilog = lwb->lwb_zilog; | |
428870ff | 872 | dmu_tx_t *tx = lwb->lwb_tx; |
34dc7c2f | 873 | |
b128c09f | 874 | ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF); |
b128c09f BB |
875 | ASSERT(BP_GET_TYPE(zio->io_bp) == DMU_OT_INTENT_LOG); |
876 | ASSERT(BP_GET_LEVEL(zio->io_bp) == 0); | |
877 | ASSERT(BP_GET_BYTEORDER(zio->io_bp) == ZFS_HOST_BYTEORDER); | |
878 | ASSERT(!BP_IS_GANG(zio->io_bp)); | |
879 | ASSERT(!BP_IS_HOLE(zio->io_bp)); | |
9b67f605 | 880 | ASSERT(BP_GET_FILL(zio->io_bp) == 0); |
b128c09f | 881 | |
34dc7c2f | 882 | /* |
9babb374 BB |
883 | * Ensure the lwb buffer pointer is cleared before releasing |
884 | * the txg. If we have had an allocation failure and | |
885 | * the txg is waiting to sync then we want want zil_sync() | |
886 | * to remove the lwb so that it's not picked up as the next new | |
887 | * one in zil_commit_writer(). zil_sync() will only remove | |
888 | * the lwb if lwb_buf is null. | |
34dc7c2f | 889 | */ |
34dc7c2f BB |
890 | zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); |
891 | mutex_enter(&zilog->zl_lock); | |
920dd524 ED |
892 | lwb->lwb_zio = NULL; |
893 | lwb->lwb_fastwrite = FALSE; | |
34dc7c2f | 894 | lwb->lwb_buf = NULL; |
428870ff BB |
895 | lwb->lwb_tx = NULL; |
896 | mutex_exit(&zilog->zl_lock); | |
9babb374 BB |
897 | |
898 | /* | |
899 | * Now that we've written this log block, we have a stable pointer | |
900 | * to the next block in the chain, so it's OK to let the txg in | |
428870ff | 901 | * which we allocated the next block sync. |
9babb374 | 902 | */ |
428870ff | 903 | dmu_tx_commit(tx); |
34dc7c2f BB |
904 | } |
905 | ||
906 | /* | |
907 | * Initialize the io for a log block. | |
34dc7c2f BB |
908 | */ |
909 | static void | |
910 | zil_lwb_write_init(zilog_t *zilog, lwb_t *lwb) | |
911 | { | |
5dbd68a3 | 912 | zbookmark_phys_t zb; |
34dc7c2f | 913 | |
428870ff BB |
914 | SET_BOOKMARK(&zb, lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_OBJSET], |
915 | ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, | |
916 | lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_SEQ]); | |
34dc7c2f BB |
917 | |
918 | if (zilog->zl_root_zio == NULL) { | |
919 | zilog->zl_root_zio = zio_root(zilog->zl_spa, NULL, NULL, | |
920 | ZIO_FLAG_CANFAIL); | |
921 | } | |
920dd524 ED |
922 | |
923 | /* Lock so zil_sync() doesn't fastwrite_unmark after zio is created */ | |
924 | mutex_enter(&zilog->zl_lock); | |
34dc7c2f | 925 | if (lwb->lwb_zio == NULL) { |
920dd524 ED |
926 | if (!lwb->lwb_fastwrite) { |
927 | metaslab_fastwrite_mark(zilog->zl_spa, &lwb->lwb_blk); | |
928 | lwb->lwb_fastwrite = 1; | |
929 | } | |
34dc7c2f | 930 | lwb->lwb_zio = zio_rewrite(zilog->zl_root_zio, zilog->zl_spa, |
428870ff | 931 | 0, &lwb->lwb_blk, lwb->lwb_buf, BP_GET_LSIZE(&lwb->lwb_blk), |
e8b96c60 | 932 | zil_lwb_write_done, lwb, ZIO_PRIORITY_SYNC_WRITE, |
920dd524 ED |
933 | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | |
934 | ZIO_FLAG_FASTWRITE, &zb); | |
34dc7c2f | 935 | } |
920dd524 | 936 | mutex_exit(&zilog->zl_lock); |
34dc7c2f BB |
937 | } |
938 | ||
428870ff BB |
939 | /* |
940 | * Define a limited set of intent log block sizes. | |
d3cc8b15 | 941 | * |
428870ff BB |
942 | * These must be a multiple of 4KB. Note only the amount used (again |
943 | * aligned to 4KB) actually gets written. However, we can't always just | |
944 | * allocate SPA_MAXBLOCKSIZE as the slog space could be exhausted. | |
945 | */ | |
946 | uint64_t zil_block_buckets[] = { | |
947 | 4096, /* non TX_WRITE */ | |
948 | 8192+4096, /* data base */ | |
949 | 32*1024 + 4096, /* NFS writes */ | |
950 | UINT64_MAX | |
951 | }; | |
952 | ||
953 | /* | |
5d7a86d1 ED |
954 | * Use the slog as long as the current commit size is less than the |
955 | * limit or the total list size is less than 2X the limit. Limit | |
956 | * checking is disabled by setting zil_slog_limit to UINT64_MAX. | |
428870ff | 957 | */ |
ee191e80 | 958 | unsigned long zil_slog_limit = 1024 * 1024; |
5d7a86d1 ED |
959 | #define USE_SLOG(zilog) (((zilog)->zl_cur_used < zil_slog_limit) || \ |
960 | ((zilog)->zl_itx_list_sz < (zil_slog_limit << 1))) | |
428870ff | 961 | |
34dc7c2f BB |
962 | /* |
963 | * Start a log block write and advance to the next log block. | |
964 | * Calls are serialized. | |
965 | */ | |
966 | static lwb_t * | |
967 | zil_lwb_write_start(zilog_t *zilog, lwb_t *lwb) | |
968 | { | |
428870ff BB |
969 | lwb_t *nlwb = NULL; |
970 | zil_chain_t *zilc; | |
34dc7c2f | 971 | spa_t *spa = zilog->zl_spa; |
428870ff BB |
972 | blkptr_t *bp; |
973 | dmu_tx_t *tx; | |
34dc7c2f | 974 | uint64_t txg; |
428870ff BB |
975 | uint64_t zil_blksz, wsz; |
976 | int i, error; | |
b6ad9671 | 977 | boolean_t use_slog; |
428870ff BB |
978 | |
979 | if (BP_GET_CHECKSUM(&lwb->lwb_blk) == ZIO_CHECKSUM_ZILOG2) { | |
980 | zilc = (zil_chain_t *)lwb->lwb_buf; | |
981 | bp = &zilc->zc_next_blk; | |
982 | } else { | |
983 | zilc = (zil_chain_t *)(lwb->lwb_buf + lwb->lwb_sz); | |
984 | bp = &zilc->zc_next_blk; | |
985 | } | |
34dc7c2f | 986 | |
428870ff | 987 | ASSERT(lwb->lwb_nused <= lwb->lwb_sz); |
34dc7c2f BB |
988 | |
989 | /* | |
990 | * Allocate the next block and save its address in this block | |
991 | * before writing it in order to establish the log chain. | |
992 | * Note that if the allocation of nlwb synced before we wrote | |
993 | * the block that points at it (lwb), we'd leak it if we crashed. | |
428870ff BB |
994 | * Therefore, we don't do dmu_tx_commit() until zil_lwb_write_done(). |
995 | * We dirty the dataset to ensure that zil_sync() will be called | |
996 | * to clean up in the event of allocation failure or I/O failure. | |
34dc7c2f | 997 | */ |
428870ff BB |
998 | tx = dmu_tx_create(zilog->zl_os); |
999 | VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0); | |
1000 | dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); | |
1001 | txg = dmu_tx_get_txg(tx); | |
1002 | ||
1003 | lwb->lwb_tx = tx; | |
34dc7c2f BB |
1004 | |
1005 | /* | |
428870ff BB |
1006 | * Log blocks are pre-allocated. Here we select the size of the next |
1007 | * block, based on size used in the last block. | |
1008 | * - first find the smallest bucket that will fit the block from a | |
1009 | * limited set of block sizes. This is because it's faster to write | |
1010 | * blocks allocated from the same metaslab as they are adjacent or | |
1011 | * close. | |
1012 | * - next find the maximum from the new suggested size and an array of | |
1013 | * previous sizes. This lessens a picket fence effect of wrongly | |
1014 | * guesssing the size if we have a stream of say 2k, 64k, 2k, 64k | |
1015 | * requests. | |
1016 | * | |
1017 | * Note we only write what is used, but we can't just allocate | |
1018 | * the maximum block size because we can exhaust the available | |
1019 | * pool log space. | |
34dc7c2f | 1020 | */ |
428870ff BB |
1021 | zil_blksz = zilog->zl_cur_used + sizeof (zil_chain_t); |
1022 | for (i = 0; zil_blksz > zil_block_buckets[i]; i++) | |
1023 | continue; | |
1024 | zil_blksz = zil_block_buckets[i]; | |
1025 | if (zil_blksz == UINT64_MAX) | |
1026 | zil_blksz = SPA_MAXBLOCKSIZE; | |
1027 | zilog->zl_prev_blks[zilog->zl_prev_rotor] = zil_blksz; | |
1028 | for (i = 0; i < ZIL_PREV_BLKS; i++) | |
1029 | zil_blksz = MAX(zil_blksz, zilog->zl_prev_blks[i]); | |
1030 | zilog->zl_prev_rotor = (zilog->zl_prev_rotor + 1) & (ZIL_PREV_BLKS - 1); | |
34dc7c2f BB |
1031 | |
1032 | BP_ZERO(bp); | |
b6ad9671 | 1033 | use_slog = USE_SLOG(zilog); |
13fe0198 MA |
1034 | error = zio_alloc_zil(spa, txg, bp, zil_blksz, |
1035 | USE_SLOG(zilog)); | |
d1d7e268 | 1036 | if (use_slog) { |
b6ad9671 ED |
1037 | ZIL_STAT_BUMP(zil_itx_metaslab_slog_count); |
1038 | ZIL_STAT_INCR(zil_itx_metaslab_slog_bytes, lwb->lwb_nused); | |
d1d7e268 | 1039 | } else { |
b6ad9671 ED |
1040 | ZIL_STAT_BUMP(zil_itx_metaslab_normal_count); |
1041 | ZIL_STAT_INCR(zil_itx_metaslab_normal_bytes, lwb->lwb_nused); | |
1042 | } | |
13fe0198 | 1043 | if (error == 0) { |
428870ff BB |
1044 | ASSERT3U(bp->blk_birth, ==, txg); |
1045 | bp->blk_cksum = lwb->lwb_blk.blk_cksum; | |
1046 | bp->blk_cksum.zc_word[ZIL_ZC_SEQ]++; | |
34dc7c2f BB |
1047 | |
1048 | /* | |
428870ff | 1049 | * Allocate a new log write buffer (lwb). |
34dc7c2f | 1050 | */ |
920dd524 | 1051 | nlwb = zil_alloc_lwb(zilog, bp, txg, TRUE); |
34dc7c2f | 1052 | |
428870ff BB |
1053 | /* Record the block for later vdev flushing */ |
1054 | zil_add_block(zilog, &lwb->lwb_blk); | |
34dc7c2f BB |
1055 | } |
1056 | ||
428870ff BB |
1057 | if (BP_GET_CHECKSUM(&lwb->lwb_blk) == ZIO_CHECKSUM_ZILOG2) { |
1058 | /* For Slim ZIL only write what is used. */ | |
1059 | wsz = P2ROUNDUP_TYPED(lwb->lwb_nused, ZIL_MIN_BLKSZ, uint64_t); | |
1060 | ASSERT3U(wsz, <=, lwb->lwb_sz); | |
1061 | zio_shrink(lwb->lwb_zio, wsz); | |
34dc7c2f | 1062 | |
428870ff BB |
1063 | } else { |
1064 | wsz = lwb->lwb_sz; | |
1065 | } | |
34dc7c2f | 1066 | |
428870ff BB |
1067 | zilc->zc_pad = 0; |
1068 | zilc->zc_nused = lwb->lwb_nused; | |
1069 | zilc->zc_eck.zec_cksum = lwb->lwb_blk.blk_cksum; | |
34dc7c2f BB |
1070 | |
1071 | /* | |
428870ff | 1072 | * clear unused data for security |
34dc7c2f | 1073 | */ |
428870ff | 1074 | bzero(lwb->lwb_buf + lwb->lwb_nused, wsz - lwb->lwb_nused); |
34dc7c2f | 1075 | |
428870ff | 1076 | zio_nowait(lwb->lwb_zio); /* Kick off the write for the old log block */ |
34dc7c2f BB |
1077 | |
1078 | /* | |
428870ff BB |
1079 | * If there was an allocation failure then nlwb will be null which |
1080 | * forces a txg_wait_synced(). | |
34dc7c2f | 1081 | */ |
34dc7c2f BB |
1082 | return (nlwb); |
1083 | } | |
1084 | ||
1085 | static lwb_t * | |
1086 | zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb) | |
1087 | { | |
1088 | lr_t *lrc = &itx->itx_lr; /* common log record */ | |
428870ff BB |
1089 | lr_write_t *lrw = (lr_write_t *)lrc; |
1090 | char *lr_buf; | |
34dc7c2f BB |
1091 | uint64_t txg = lrc->lrc_txg; |
1092 | uint64_t reclen = lrc->lrc_reclen; | |
428870ff | 1093 | uint64_t dlen = 0; |
34dc7c2f BB |
1094 | |
1095 | if (lwb == NULL) | |
1096 | return (NULL); | |
428870ff | 1097 | |
34dc7c2f | 1098 | ASSERT(lwb->lwb_buf != NULL); |
29809a6c MA |
1099 | ASSERT(zilog_is_dirty(zilog) || |
1100 | spa_freeze_txg(zilog->zl_spa) != UINT64_MAX); | |
34dc7c2f BB |
1101 | |
1102 | if (lrc->lrc_txtype == TX_WRITE && itx->itx_wr_state == WR_NEED_COPY) | |
1103 | dlen = P2ROUNDUP_TYPED( | |
428870ff | 1104 | lrw->lr_length, sizeof (uint64_t), uint64_t); |
34dc7c2f BB |
1105 | |
1106 | zilog->zl_cur_used += (reclen + dlen); | |
1107 | ||
1108 | zil_lwb_write_init(zilog, lwb); | |
1109 | ||
1110 | /* | |
1111 | * If this record won't fit in the current log block, start a new one. | |
1112 | */ | |
428870ff | 1113 | if (lwb->lwb_nused + reclen + dlen > lwb->lwb_sz) { |
34dc7c2f BB |
1114 | lwb = zil_lwb_write_start(zilog, lwb); |
1115 | if (lwb == NULL) | |
1116 | return (NULL); | |
1117 | zil_lwb_write_init(zilog, lwb); | |
428870ff BB |
1118 | ASSERT(LWB_EMPTY(lwb)); |
1119 | if (lwb->lwb_nused + reclen + dlen > lwb->lwb_sz) { | |
34dc7c2f BB |
1120 | txg_wait_synced(zilog->zl_dmu_pool, txg); |
1121 | return (lwb); | |
1122 | } | |
1123 | } | |
1124 | ||
428870ff BB |
1125 | lr_buf = lwb->lwb_buf + lwb->lwb_nused; |
1126 | bcopy(lrc, lr_buf, reclen); | |
1127 | lrc = (lr_t *)lr_buf; | |
1128 | lrw = (lr_write_t *)lrc; | |
34dc7c2f | 1129 | |
b6ad9671 ED |
1130 | ZIL_STAT_BUMP(zil_itx_count); |
1131 | ||
34dc7c2f BB |
1132 | /* |
1133 | * If it's a write, fetch the data or get its blkptr as appropriate. | |
1134 | */ | |
1135 | if (lrc->lrc_txtype == TX_WRITE) { | |
1136 | if (txg > spa_freeze_txg(zilog->zl_spa)) | |
1137 | txg_wait_synced(zilog->zl_dmu_pool, txg); | |
b6ad9671 ED |
1138 | if (itx->itx_wr_state == WR_COPIED) { |
1139 | ZIL_STAT_BUMP(zil_itx_copied_count); | |
1140 | ZIL_STAT_INCR(zil_itx_copied_bytes, lrw->lr_length); | |
1141 | } else { | |
34dc7c2f BB |
1142 | char *dbuf; |
1143 | int error; | |
1144 | ||
34dc7c2f BB |
1145 | if (dlen) { |
1146 | ASSERT(itx->itx_wr_state == WR_NEED_COPY); | |
428870ff BB |
1147 | dbuf = lr_buf + reclen; |
1148 | lrw->lr_common.lrc_reclen += dlen; | |
b6ad9671 | 1149 | ZIL_STAT_BUMP(zil_itx_needcopy_count); |
d1d7e268 MK |
1150 | ZIL_STAT_INCR(zil_itx_needcopy_bytes, |
1151 | lrw->lr_length); | |
34dc7c2f BB |
1152 | } else { |
1153 | ASSERT(itx->itx_wr_state == WR_INDIRECT); | |
1154 | dbuf = NULL; | |
b6ad9671 | 1155 | ZIL_STAT_BUMP(zil_itx_indirect_count); |
d1d7e268 MK |
1156 | ZIL_STAT_INCR(zil_itx_indirect_bytes, |
1157 | lrw->lr_length); | |
34dc7c2f BB |
1158 | } |
1159 | error = zilog->zl_get_data( | |
428870ff | 1160 | itx->itx_private, lrw, dbuf, lwb->lwb_zio); |
45d1cae3 BB |
1161 | if (error == EIO) { |
1162 | txg_wait_synced(zilog->zl_dmu_pool, txg); | |
1163 | return (lwb); | |
1164 | } | |
13fe0198 | 1165 | if (error != 0) { |
34dc7c2f BB |
1166 | ASSERT(error == ENOENT || error == EEXIST || |
1167 | error == EALREADY); | |
1168 | return (lwb); | |
1169 | } | |
1170 | } | |
1171 | } | |
1172 | ||
428870ff BB |
1173 | /* |
1174 | * We're actually making an entry, so update lrc_seq to be the | |
1175 | * log record sequence number. Note that this is generally not | |
1176 | * equal to the itx sequence number because not all transactions | |
1177 | * are synchronous, and sometimes spa_sync() gets there first. | |
1178 | */ | |
1179 | lrc->lrc_seq = ++zilog->zl_lr_seq; /* we are single threaded */ | |
34dc7c2f BB |
1180 | lwb->lwb_nused += reclen + dlen; |
1181 | lwb->lwb_max_txg = MAX(lwb->lwb_max_txg, txg); | |
428870ff | 1182 | ASSERT3U(lwb->lwb_nused, <=, lwb->lwb_sz); |
c99c9001 | 1183 | ASSERT0(P2PHASE(lwb->lwb_nused, sizeof (uint64_t))); |
34dc7c2f BB |
1184 | |
1185 | return (lwb); | |
1186 | } | |
1187 | ||
1188 | itx_t * | |
1189 | zil_itx_create(uint64_t txtype, size_t lrsize) | |
1190 | { | |
1191 | itx_t *itx; | |
1192 | ||
1193 | lrsize = P2ROUNDUP_TYPED(lrsize, sizeof (uint64_t), size_t); | |
1194 | ||
19ea3d25 | 1195 | itx = zio_data_buf_alloc(offsetof(itx_t, itx_lr) + lrsize); |
34dc7c2f BB |
1196 | itx->itx_lr.lrc_txtype = txtype; |
1197 | itx->itx_lr.lrc_reclen = lrsize; | |
1198 | itx->itx_sod = lrsize; /* if write & WR_NEED_COPY will be increased */ | |
1199 | itx->itx_lr.lrc_seq = 0; /* defensive */ | |
572e2857 | 1200 | itx->itx_sync = B_TRUE; /* default is synchronous */ |
119a394a ED |
1201 | itx->itx_callback = NULL; |
1202 | itx->itx_callback_data = NULL; | |
34dc7c2f BB |
1203 | |
1204 | return (itx); | |
1205 | } | |
1206 | ||
428870ff BB |
1207 | void |
1208 | zil_itx_destroy(itx_t *itx) | |
1209 | { | |
19ea3d25 | 1210 | zio_data_buf_free(itx, offsetof(itx_t, itx_lr)+itx->itx_lr.lrc_reclen); |
428870ff BB |
1211 | } |
1212 | ||
572e2857 BB |
1213 | /* |
1214 | * Free up the sync and async itxs. The itxs_t has already been detached | |
1215 | * so no locks are needed. | |
1216 | */ | |
1217 | static void | |
1218 | zil_itxg_clean(itxs_t *itxs) | |
34dc7c2f | 1219 | { |
572e2857 BB |
1220 | itx_t *itx; |
1221 | list_t *list; | |
1222 | avl_tree_t *t; | |
1223 | void *cookie; | |
1224 | itx_async_node_t *ian; | |
1225 | ||
1226 | list = &itxs->i_sync_list; | |
1227 | while ((itx = list_head(list)) != NULL) { | |
119a394a ED |
1228 | if (itx->itx_callback != NULL) |
1229 | itx->itx_callback(itx->itx_callback_data); | |
572e2857 | 1230 | list_remove(list, itx); |
19ea3d25 | 1231 | zil_itx_destroy(itx); |
572e2857 | 1232 | } |
34dc7c2f | 1233 | |
572e2857 BB |
1234 | cookie = NULL; |
1235 | t = &itxs->i_async_tree; | |
1236 | while ((ian = avl_destroy_nodes(t, &cookie)) != NULL) { | |
1237 | list = &ian->ia_list; | |
1238 | while ((itx = list_head(list)) != NULL) { | |
119a394a ED |
1239 | if (itx->itx_callback != NULL) |
1240 | itx->itx_callback(itx->itx_callback_data); | |
572e2857 | 1241 | list_remove(list, itx); |
19ea3d25 | 1242 | zil_itx_destroy(itx); |
572e2857 BB |
1243 | } |
1244 | list_destroy(list); | |
1245 | kmem_free(ian, sizeof (itx_async_node_t)); | |
1246 | } | |
1247 | avl_destroy(t); | |
34dc7c2f | 1248 | |
572e2857 BB |
1249 | kmem_free(itxs, sizeof (itxs_t)); |
1250 | } | |
34dc7c2f | 1251 | |
572e2857 BB |
1252 | static int |
1253 | zil_aitx_compare(const void *x1, const void *x2) | |
1254 | { | |
1255 | const uint64_t o1 = ((itx_async_node_t *)x1)->ia_foid; | |
1256 | const uint64_t o2 = ((itx_async_node_t *)x2)->ia_foid; | |
1257 | ||
1258 | if (o1 < o2) | |
1259 | return (-1); | |
1260 | if (o1 > o2) | |
1261 | return (1); | |
1262 | ||
1263 | return (0); | |
34dc7c2f BB |
1264 | } |
1265 | ||
1266 | /* | |
572e2857 | 1267 | * Remove all async itx with the given oid. |
34dc7c2f BB |
1268 | */ |
1269 | static void | |
572e2857 | 1270 | zil_remove_async(zilog_t *zilog, uint64_t oid) |
34dc7c2f | 1271 | { |
572e2857 BB |
1272 | uint64_t otxg, txg; |
1273 | itx_async_node_t *ian; | |
1274 | avl_tree_t *t; | |
1275 | avl_index_t where; | |
34dc7c2f BB |
1276 | list_t clean_list; |
1277 | itx_t *itx; | |
1278 | ||
572e2857 | 1279 | ASSERT(oid != 0); |
34dc7c2f BB |
1280 | list_create(&clean_list, sizeof (itx_t), offsetof(itx_t, itx_node)); |
1281 | ||
572e2857 BB |
1282 | if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */ |
1283 | otxg = ZILTEST_TXG; | |
1284 | else | |
1285 | otxg = spa_last_synced_txg(zilog->zl_spa) + 1; | |
34dc7c2f | 1286 | |
572e2857 BB |
1287 | for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) { |
1288 | itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK]; | |
1289 | ||
1290 | mutex_enter(&itxg->itxg_lock); | |
1291 | if (itxg->itxg_txg != txg) { | |
1292 | mutex_exit(&itxg->itxg_lock); | |
1293 | continue; | |
1294 | } | |
34dc7c2f | 1295 | |
572e2857 BB |
1296 | /* |
1297 | * Locate the object node and append its list. | |
1298 | */ | |
1299 | t = &itxg->itxg_itxs->i_async_tree; | |
1300 | ian = avl_find(t, &oid, &where); | |
1301 | if (ian != NULL) | |
1302 | list_move_tail(&clean_list, &ian->ia_list); | |
1303 | mutex_exit(&itxg->itxg_lock); | |
1304 | } | |
34dc7c2f | 1305 | while ((itx = list_head(&clean_list)) != NULL) { |
119a394a ED |
1306 | if (itx->itx_callback != NULL) |
1307 | itx->itx_callback(itx->itx_callback_data); | |
34dc7c2f | 1308 | list_remove(&clean_list, itx); |
19ea3d25 | 1309 | zil_itx_destroy(itx); |
34dc7c2f BB |
1310 | } |
1311 | list_destroy(&clean_list); | |
1312 | } | |
1313 | ||
572e2857 BB |
1314 | void |
1315 | zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx) | |
1316 | { | |
1317 | uint64_t txg; | |
1318 | itxg_t *itxg; | |
1319 | itxs_t *itxs, *clean = NULL; | |
1320 | ||
1321 | /* | |
1322 | * Object ids can be re-instantiated in the next txg so | |
1323 | * remove any async transactions to avoid future leaks. | |
1324 | * This can happen if a fsync occurs on the re-instantiated | |
1325 | * object for a WR_INDIRECT or WR_NEED_COPY write, which gets | |
1326 | * the new file data and flushes a write record for the old object. | |
1327 | */ | |
1328 | if ((itx->itx_lr.lrc_txtype & ~TX_CI) == TX_REMOVE) | |
1329 | zil_remove_async(zilog, itx->itx_oid); | |
1330 | ||
1331 | /* | |
1332 | * Ensure the data of a renamed file is committed before the rename. | |
1333 | */ | |
1334 | if ((itx->itx_lr.lrc_txtype & ~TX_CI) == TX_RENAME) | |
1335 | zil_async_to_sync(zilog, itx->itx_oid); | |
1336 | ||
29809a6c | 1337 | if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) |
572e2857 BB |
1338 | txg = ZILTEST_TXG; |
1339 | else | |
1340 | txg = dmu_tx_get_txg(tx); | |
1341 | ||
1342 | itxg = &zilog->zl_itxg[txg & TXG_MASK]; | |
1343 | mutex_enter(&itxg->itxg_lock); | |
1344 | itxs = itxg->itxg_itxs; | |
1345 | if (itxg->itxg_txg != txg) { | |
1346 | if (itxs != NULL) { | |
1347 | /* | |
1348 | * The zil_clean callback hasn't got around to cleaning | |
1349 | * this itxg. Save the itxs for release below. | |
1350 | * This should be rare. | |
1351 | */ | |
1352 | atomic_add_64(&zilog->zl_itx_list_sz, -itxg->itxg_sod); | |
1353 | itxg->itxg_sod = 0; | |
1354 | clean = itxg->itxg_itxs; | |
1355 | } | |
1356 | ASSERT(itxg->itxg_sod == 0); | |
1357 | itxg->itxg_txg = txg; | |
d1d7e268 | 1358 | itxs = itxg->itxg_itxs = kmem_zalloc(sizeof (itxs_t), |
79c76d5b | 1359 | KM_SLEEP); |
572e2857 BB |
1360 | |
1361 | list_create(&itxs->i_sync_list, sizeof (itx_t), | |
1362 | offsetof(itx_t, itx_node)); | |
1363 | avl_create(&itxs->i_async_tree, zil_aitx_compare, | |
1364 | sizeof (itx_async_node_t), | |
1365 | offsetof(itx_async_node_t, ia_node)); | |
1366 | } | |
1367 | if (itx->itx_sync) { | |
1368 | list_insert_tail(&itxs->i_sync_list, itx); | |
1369 | atomic_add_64(&zilog->zl_itx_list_sz, itx->itx_sod); | |
1370 | itxg->itxg_sod += itx->itx_sod; | |
1371 | } else { | |
1372 | avl_tree_t *t = &itxs->i_async_tree; | |
1373 | uint64_t foid = ((lr_ooo_t *)&itx->itx_lr)->lr_foid; | |
1374 | itx_async_node_t *ian; | |
1375 | avl_index_t where; | |
1376 | ||
1377 | ian = avl_find(t, &foid, &where); | |
1378 | if (ian == NULL) { | |
d1d7e268 | 1379 | ian = kmem_alloc(sizeof (itx_async_node_t), |
79c76d5b | 1380 | KM_SLEEP); |
572e2857 BB |
1381 | list_create(&ian->ia_list, sizeof (itx_t), |
1382 | offsetof(itx_t, itx_node)); | |
1383 | ian->ia_foid = foid; | |
1384 | avl_insert(t, ian, where); | |
1385 | } | |
1386 | list_insert_tail(&ian->ia_list, itx); | |
1387 | } | |
1388 | ||
1389 | itx->itx_lr.lrc_txg = dmu_tx_get_txg(tx); | |
29809a6c | 1390 | zilog_dirty(zilog, txg); |
572e2857 BB |
1391 | mutex_exit(&itxg->itxg_lock); |
1392 | ||
1393 | /* Release the old itxs now we've dropped the lock */ | |
1394 | if (clean != NULL) | |
1395 | zil_itxg_clean(clean); | |
1396 | } | |
1397 | ||
34dc7c2f BB |
1398 | /* |
1399 | * If there are any in-memory intent log transactions which have now been | |
29809a6c MA |
1400 | * synced then start up a taskq to free them. We should only do this after we |
1401 | * have written out the uberblocks (i.e. txg has been comitted) so that | |
1402 | * don't inadvertently clean out in-memory log records that would be required | |
1403 | * by zil_commit(). | |
34dc7c2f BB |
1404 | */ |
1405 | void | |
572e2857 | 1406 | zil_clean(zilog_t *zilog, uint64_t synced_txg) |
34dc7c2f | 1407 | { |
572e2857 BB |
1408 | itxg_t *itxg = &zilog->zl_itxg[synced_txg & TXG_MASK]; |
1409 | itxs_t *clean_me; | |
34dc7c2f | 1410 | |
572e2857 BB |
1411 | mutex_enter(&itxg->itxg_lock); |
1412 | if (itxg->itxg_itxs == NULL || itxg->itxg_txg == ZILTEST_TXG) { | |
1413 | mutex_exit(&itxg->itxg_lock); | |
1414 | return; | |
1415 | } | |
1416 | ASSERT3U(itxg->itxg_txg, <=, synced_txg); | |
1417 | ASSERT(itxg->itxg_txg != 0); | |
1418 | ASSERT(zilog->zl_clean_taskq != NULL); | |
1419 | atomic_add_64(&zilog->zl_itx_list_sz, -itxg->itxg_sod); | |
1420 | itxg->itxg_sod = 0; | |
1421 | clean_me = itxg->itxg_itxs; | |
1422 | itxg->itxg_itxs = NULL; | |
1423 | itxg->itxg_txg = 0; | |
1424 | mutex_exit(&itxg->itxg_lock); | |
1425 | /* | |
1426 | * Preferably start a task queue to free up the old itxs but | |
1427 | * if taskq_dispatch can't allocate resources to do that then | |
1428 | * free it in-line. This should be rare. Note, using TQ_SLEEP | |
1429 | * created a bad performance problem. | |
1430 | */ | |
1431 | if (taskq_dispatch(zilog->zl_clean_taskq, | |
b8864a23 | 1432 | (void (*)(void *))zil_itxg_clean, clean_me, TQ_NOSLEEP) == 0) |
572e2857 BB |
1433 | zil_itxg_clean(clean_me); |
1434 | } | |
1435 | ||
1436 | /* | |
1437 | * Get the list of itxs to commit into zl_itx_commit_list. | |
1438 | */ | |
1439 | static void | |
1440 | zil_get_commit_list(zilog_t *zilog) | |
1441 | { | |
1442 | uint64_t otxg, txg; | |
1443 | list_t *commit_list = &zilog->zl_itx_commit_list; | |
1444 | uint64_t push_sod = 0; | |
1445 | ||
1446 | if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */ | |
1447 | otxg = ZILTEST_TXG; | |
1448 | else | |
1449 | otxg = spa_last_synced_txg(zilog->zl_spa) + 1; | |
1450 | ||
1451 | for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) { | |
1452 | itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK]; | |
1453 | ||
1454 | mutex_enter(&itxg->itxg_lock); | |
1455 | if (itxg->itxg_txg != txg) { | |
1456 | mutex_exit(&itxg->itxg_lock); | |
1457 | continue; | |
1458 | } | |
1459 | ||
1460 | list_move_tail(commit_list, &itxg->itxg_itxs->i_sync_list); | |
1461 | push_sod += itxg->itxg_sod; | |
1462 | itxg->itxg_sod = 0; | |
1463 | ||
1464 | mutex_exit(&itxg->itxg_lock); | |
1465 | } | |
1466 | atomic_add_64(&zilog->zl_itx_list_sz, -push_sod); | |
1467 | } | |
1468 | ||
1469 | /* | |
1470 | * Move the async itxs for a specified object to commit into sync lists. | |
1471 | */ | |
1472 | static void | |
1473 | zil_async_to_sync(zilog_t *zilog, uint64_t foid) | |
1474 | { | |
1475 | uint64_t otxg, txg; | |
1476 | itx_async_node_t *ian; | |
1477 | avl_tree_t *t; | |
1478 | avl_index_t where; | |
1479 | ||
1480 | if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */ | |
1481 | otxg = ZILTEST_TXG; | |
1482 | else | |
1483 | otxg = spa_last_synced_txg(zilog->zl_spa) + 1; | |
1484 | ||
1485 | for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) { | |
1486 | itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK]; | |
1487 | ||
1488 | mutex_enter(&itxg->itxg_lock); | |
1489 | if (itxg->itxg_txg != txg) { | |
1490 | mutex_exit(&itxg->itxg_lock); | |
1491 | continue; | |
1492 | } | |
1493 | ||
1494 | /* | |
1495 | * If a foid is specified then find that node and append its | |
1496 | * list. Otherwise walk the tree appending all the lists | |
1497 | * to the sync list. We add to the end rather than the | |
1498 | * beginning to ensure the create has happened. | |
1499 | */ | |
1500 | t = &itxg->itxg_itxs->i_async_tree; | |
1501 | if (foid != 0) { | |
1502 | ian = avl_find(t, &foid, &where); | |
1503 | if (ian != NULL) { | |
1504 | list_move_tail(&itxg->itxg_itxs->i_sync_list, | |
1505 | &ian->ia_list); | |
1506 | } | |
1507 | } else { | |
1508 | void *cookie = NULL; | |
1509 | ||
1510 | while ((ian = avl_destroy_nodes(t, &cookie)) != NULL) { | |
1511 | list_move_tail(&itxg->itxg_itxs->i_sync_list, | |
1512 | &ian->ia_list); | |
1513 | list_destroy(&ian->ia_list); | |
1514 | kmem_free(ian, sizeof (itx_async_node_t)); | |
1515 | } | |
1516 | } | |
1517 | mutex_exit(&itxg->itxg_lock); | |
34dc7c2f | 1518 | } |
34dc7c2f BB |
1519 | } |
1520 | ||
b128c09f | 1521 | static void |
572e2857 | 1522 | zil_commit_writer(zilog_t *zilog) |
34dc7c2f BB |
1523 | { |
1524 | uint64_t txg; | |
572e2857 | 1525 | itx_t *itx; |
34dc7c2f | 1526 | lwb_t *lwb; |
572e2857 | 1527 | spa_t *spa = zilog->zl_spa; |
428870ff | 1528 | int error = 0; |
34dc7c2f | 1529 | |
b128c09f | 1530 | ASSERT(zilog->zl_root_zio == NULL); |
572e2857 BB |
1531 | |
1532 | mutex_exit(&zilog->zl_lock); | |
1533 | ||
1534 | zil_get_commit_list(zilog); | |
1535 | ||
1536 | /* | |
1537 | * Return if there's nothing to commit before we dirty the fs by | |
1538 | * calling zil_create(). | |
1539 | */ | |
1540 | if (list_head(&zilog->zl_itx_commit_list) == NULL) { | |
1541 | mutex_enter(&zilog->zl_lock); | |
1542 | return; | |
1543 | } | |
34dc7c2f BB |
1544 | |
1545 | if (zilog->zl_suspend) { | |
1546 | lwb = NULL; | |
1547 | } else { | |
1548 | lwb = list_tail(&zilog->zl_lwb_list); | |
572e2857 | 1549 | if (lwb == NULL) |
428870ff | 1550 | lwb = zil_create(zilog); |
34dc7c2f BB |
1551 | } |
1552 | ||
34dc7c2f | 1553 | DTRACE_PROBE1(zil__cw1, zilog_t *, zilog); |
119a394a | 1554 | for (itx = list_head(&zilog->zl_itx_commit_list); itx != NULL; |
d1d7e268 | 1555 | itx = list_next(&zilog->zl_itx_commit_list, itx)) { |
34dc7c2f BB |
1556 | txg = itx->itx_lr.lrc_txg; |
1557 | ASSERT(txg); | |
1558 | ||
572e2857 | 1559 | if (txg > spa_last_synced_txg(spa) || txg > spa_freeze_txg(spa)) |
34dc7c2f | 1560 | lwb = zil_lwb_commit(zilog, itx, lwb); |
34dc7c2f BB |
1561 | } |
1562 | DTRACE_PROBE1(zil__cw2, zilog_t *, zilog); | |
34dc7c2f BB |
1563 | |
1564 | /* write the last block out */ | |
1565 | if (lwb != NULL && lwb->lwb_zio != NULL) | |
1566 | lwb = zil_lwb_write_start(zilog, lwb); | |
1567 | ||
34dc7c2f BB |
1568 | zilog->zl_cur_used = 0; |
1569 | ||
1570 | /* | |
1571 | * Wait if necessary for the log blocks to be on stable storage. | |
1572 | */ | |
1573 | if (zilog->zl_root_zio) { | |
428870ff | 1574 | error = zio_wait(zilog->zl_root_zio); |
b128c09f | 1575 | zilog->zl_root_zio = NULL; |
34dc7c2f BB |
1576 | zil_flush_vdevs(zilog); |
1577 | } | |
1578 | ||
428870ff | 1579 | if (error || lwb == NULL) |
34dc7c2f | 1580 | txg_wait_synced(zilog->zl_dmu_pool, 0); |
34dc7c2f | 1581 | |
119a394a ED |
1582 | while ((itx = list_head(&zilog->zl_itx_commit_list))) { |
1583 | txg = itx->itx_lr.lrc_txg; | |
1584 | ASSERT(txg); | |
1585 | ||
1586 | if (itx->itx_callback != NULL) | |
1587 | itx->itx_callback(itx->itx_callback_data); | |
1588 | list_remove(&zilog->zl_itx_commit_list, itx); | |
19ea3d25 | 1589 | zil_itx_destroy(itx); |
119a394a ED |
1590 | } |
1591 | ||
34dc7c2f | 1592 | mutex_enter(&zilog->zl_lock); |
428870ff BB |
1593 | |
1594 | /* | |
1595 | * Remember the highest committed log sequence number for ztest. | |
1596 | * We only update this value when all the log writes succeeded, | |
1597 | * because ztest wants to ASSERT that it got the whole log chain. | |
1598 | */ | |
1599 | if (error == 0 && lwb != NULL) | |
1600 | zilog->zl_commit_lr_seq = zilog->zl_lr_seq; | |
34dc7c2f BB |
1601 | } |
1602 | ||
1603 | /* | |
572e2857 | 1604 | * Commit zfs transactions to stable storage. |
34dc7c2f | 1605 | * If foid is 0 push out all transactions, otherwise push only those |
572e2857 BB |
1606 | * for that object or might reference that object. |
1607 | * | |
1608 | * itxs are committed in batches. In a heavily stressed zil there will be | |
1609 | * a commit writer thread who is writing out a bunch of itxs to the log | |
1610 | * for a set of committing threads (cthreads) in the same batch as the writer. | |
1611 | * Those cthreads are all waiting on the same cv for that batch. | |
1612 | * | |
1613 | * There will also be a different and growing batch of threads that are | |
1614 | * waiting to commit (qthreads). When the committing batch completes | |
1615 | * a transition occurs such that the cthreads exit and the qthreads become | |
1616 | * cthreads. One of the new cthreads becomes the writer thread for the | |
1617 | * batch. Any new threads arriving become new qthreads. | |
1618 | * | |
1619 | * Only 2 condition variables are needed and there's no transition | |
1620 | * between the two cvs needed. They just flip-flop between qthreads | |
1621 | * and cthreads. | |
1622 | * | |
1623 | * Using this scheme we can efficiently wakeup up only those threads | |
1624 | * that have been committed. | |
34dc7c2f BB |
1625 | */ |
1626 | void | |
572e2857 | 1627 | zil_commit(zilog_t *zilog, uint64_t foid) |
34dc7c2f | 1628 | { |
572e2857 | 1629 | uint64_t mybatch; |
34dc7c2f | 1630 | |
572e2857 BB |
1631 | if (zilog->zl_sync == ZFS_SYNC_DISABLED) |
1632 | return; | |
34dc7c2f | 1633 | |
b6ad9671 ED |
1634 | ZIL_STAT_BUMP(zil_commit_count); |
1635 | ||
572e2857 BB |
1636 | /* move the async itxs for the foid to the sync queues */ |
1637 | zil_async_to_sync(zilog, foid); | |
34dc7c2f | 1638 | |
572e2857 BB |
1639 | mutex_enter(&zilog->zl_lock); |
1640 | mybatch = zilog->zl_next_batch; | |
34dc7c2f | 1641 | while (zilog->zl_writer) { |
572e2857 BB |
1642 | cv_wait(&zilog->zl_cv_batch[mybatch & 1], &zilog->zl_lock); |
1643 | if (mybatch <= zilog->zl_com_batch) { | |
34dc7c2f BB |
1644 | mutex_exit(&zilog->zl_lock); |
1645 | return; | |
1646 | } | |
1647 | } | |
428870ff | 1648 | |
572e2857 BB |
1649 | zilog->zl_next_batch++; |
1650 | zilog->zl_writer = B_TRUE; | |
b6ad9671 | 1651 | ZIL_STAT_BUMP(zil_commit_writer_count); |
572e2857 BB |
1652 | zil_commit_writer(zilog); |
1653 | zilog->zl_com_batch = mybatch; | |
1654 | zilog->zl_writer = B_FALSE; | |
428870ff | 1655 | |
572e2857 BB |
1656 | /* wake up one thread to become the next writer */ |
1657 | cv_signal(&zilog->zl_cv_batch[(mybatch+1) & 1]); | |
428870ff | 1658 | |
572e2857 BB |
1659 | /* wake up all threads waiting for this batch to be committed */ |
1660 | cv_broadcast(&zilog->zl_cv_batch[mybatch & 1]); | |
8c0712fd BB |
1661 | |
1662 | mutex_exit(&zilog->zl_lock); | |
428870ff BB |
1663 | } |
1664 | ||
34dc7c2f BB |
1665 | /* |
1666 | * Called in syncing context to free committed log blocks and update log header. | |
1667 | */ | |
1668 | void | |
1669 | zil_sync(zilog_t *zilog, dmu_tx_t *tx) | |
1670 | { | |
1671 | zil_header_t *zh = zil_header_in_syncing_context(zilog); | |
1672 | uint64_t txg = dmu_tx_get_txg(tx); | |
1673 | spa_t *spa = zilog->zl_spa; | |
428870ff | 1674 | uint64_t *replayed_seq = &zilog->zl_replayed_seq[txg & TXG_MASK]; |
34dc7c2f BB |
1675 | lwb_t *lwb; |
1676 | ||
9babb374 BB |
1677 | /* |
1678 | * We don't zero out zl_destroy_txg, so make sure we don't try | |
1679 | * to destroy it twice. | |
1680 | */ | |
1681 | if (spa_sync_pass(spa) != 1) | |
1682 | return; | |
1683 | ||
34dc7c2f BB |
1684 | mutex_enter(&zilog->zl_lock); |
1685 | ||
1686 | ASSERT(zilog->zl_stop_sync == 0); | |
1687 | ||
428870ff BB |
1688 | if (*replayed_seq != 0) { |
1689 | ASSERT(zh->zh_replay_seq < *replayed_seq); | |
1690 | zh->zh_replay_seq = *replayed_seq; | |
1691 | *replayed_seq = 0; | |
1692 | } | |
34dc7c2f BB |
1693 | |
1694 | if (zilog->zl_destroy_txg == txg) { | |
1695 | blkptr_t blk = zh->zh_log; | |
1696 | ||
1697 | ASSERT(list_head(&zilog->zl_lwb_list) == NULL); | |
34dc7c2f BB |
1698 | |
1699 | bzero(zh, sizeof (zil_header_t)); | |
fb5f0bc8 | 1700 | bzero(zilog->zl_replayed_seq, sizeof (zilog->zl_replayed_seq)); |
34dc7c2f BB |
1701 | |
1702 | if (zilog->zl_keep_first) { | |
1703 | /* | |
1704 | * If this block was part of log chain that couldn't | |
1705 | * be claimed because a device was missing during | |
1706 | * zil_claim(), but that device later returns, | |
1707 | * then this block could erroneously appear valid. | |
1708 | * To guard against this, assign a new GUID to the new | |
1709 | * log chain so it doesn't matter what blk points to. | |
1710 | */ | |
1711 | zil_init_log_chain(zilog, &blk); | |
1712 | zh->zh_log = blk; | |
1713 | } | |
1714 | } | |
1715 | ||
9babb374 | 1716 | while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) { |
34dc7c2f BB |
1717 | zh->zh_log = lwb->lwb_blk; |
1718 | if (lwb->lwb_buf != NULL || lwb->lwb_max_txg > txg) | |
1719 | break; | |
920dd524 ED |
1720 | |
1721 | ASSERT(lwb->lwb_zio == NULL); | |
1722 | ||
34dc7c2f | 1723 | list_remove(&zilog->zl_lwb_list, lwb); |
428870ff | 1724 | zio_free_zil(spa, txg, &lwb->lwb_blk); |
34dc7c2f BB |
1725 | kmem_cache_free(zil_lwb_cache, lwb); |
1726 | ||
1727 | /* | |
1728 | * If we don't have anything left in the lwb list then | |
1729 | * we've had an allocation failure and we need to zero | |
1730 | * out the zil_header blkptr so that we don't end | |
1731 | * up freeing the same block twice. | |
1732 | */ | |
1733 | if (list_head(&zilog->zl_lwb_list) == NULL) | |
1734 | BP_ZERO(&zh->zh_log); | |
1735 | } | |
920dd524 ED |
1736 | |
1737 | /* | |
1738 | * Remove fastwrite on any blocks that have been pre-allocated for | |
1739 | * the next commit. This prevents fastwrite counter pollution by | |
1740 | * unused, long-lived LWBs. | |
1741 | */ | |
1742 | for (; lwb != NULL; lwb = list_next(&zilog->zl_lwb_list, lwb)) { | |
1743 | if (lwb->lwb_fastwrite && !lwb->lwb_zio) { | |
1744 | metaslab_fastwrite_unmark(zilog->zl_spa, &lwb->lwb_blk); | |
1745 | lwb->lwb_fastwrite = 0; | |
1746 | } | |
1747 | } | |
1748 | ||
34dc7c2f BB |
1749 | mutex_exit(&zilog->zl_lock); |
1750 | } | |
1751 | ||
1752 | void | |
1753 | zil_init(void) | |
1754 | { | |
1755 | zil_lwb_cache = kmem_cache_create("zil_lwb_cache", | |
1756 | sizeof (struct lwb), 0, NULL, NULL, NULL, NULL, NULL, 0); | |
b6ad9671 ED |
1757 | |
1758 | zil_ksp = kstat_create("zfs", 0, "zil", "misc", | |
d1d7e268 | 1759 | KSTAT_TYPE_NAMED, sizeof (zil_stats) / sizeof (kstat_named_t), |
b6ad9671 ED |
1760 | KSTAT_FLAG_VIRTUAL); |
1761 | ||
1762 | if (zil_ksp != NULL) { | |
1763 | zil_ksp->ks_data = &zil_stats; | |
1764 | kstat_install(zil_ksp); | |
1765 | } | |
34dc7c2f BB |
1766 | } |
1767 | ||
1768 | void | |
1769 | zil_fini(void) | |
1770 | { | |
1771 | kmem_cache_destroy(zil_lwb_cache); | |
b6ad9671 ED |
1772 | |
1773 | if (zil_ksp != NULL) { | |
1774 | kstat_delete(zil_ksp); | |
1775 | zil_ksp = NULL; | |
1776 | } | |
34dc7c2f BB |
1777 | } |
1778 | ||
428870ff BB |
1779 | void |
1780 | zil_set_sync(zilog_t *zilog, uint64_t sync) | |
1781 | { | |
1782 | zilog->zl_sync = sync; | |
1783 | } | |
1784 | ||
1785 | void | |
1786 | zil_set_logbias(zilog_t *zilog, uint64_t logbias) | |
1787 | { | |
1788 | zilog->zl_logbias = logbias; | |
1789 | } | |
1790 | ||
34dc7c2f BB |
1791 | zilog_t * |
1792 | zil_alloc(objset_t *os, zil_header_t *zh_phys) | |
1793 | { | |
1794 | zilog_t *zilog; | |
d6320ddb | 1795 | int i; |
34dc7c2f | 1796 | |
79c76d5b | 1797 | zilog = kmem_zalloc(sizeof (zilog_t), KM_SLEEP); |
34dc7c2f BB |
1798 | |
1799 | zilog->zl_header = zh_phys; | |
1800 | zilog->zl_os = os; | |
1801 | zilog->zl_spa = dmu_objset_spa(os); | |
1802 | zilog->zl_dmu_pool = dmu_objset_pool(os); | |
1803 | zilog->zl_destroy_txg = TXG_INITIAL - 1; | |
428870ff BB |
1804 | zilog->zl_logbias = dmu_objset_logbias(os); |
1805 | zilog->zl_sync = dmu_objset_syncprop(os); | |
572e2857 | 1806 | zilog->zl_next_batch = 1; |
34dc7c2f BB |
1807 | |
1808 | mutex_init(&zilog->zl_lock, NULL, MUTEX_DEFAULT, NULL); | |
1809 | ||
d6320ddb | 1810 | for (i = 0; i < TXG_SIZE; i++) { |
572e2857 BB |
1811 | mutex_init(&zilog->zl_itxg[i].itxg_lock, NULL, |
1812 | MUTEX_DEFAULT, NULL); | |
1813 | } | |
34dc7c2f BB |
1814 | |
1815 | list_create(&zilog->zl_lwb_list, sizeof (lwb_t), | |
1816 | offsetof(lwb_t, lwb_node)); | |
1817 | ||
572e2857 BB |
1818 | list_create(&zilog->zl_itx_commit_list, sizeof (itx_t), |
1819 | offsetof(itx_t, itx_node)); | |
1820 | ||
34dc7c2f BB |
1821 | mutex_init(&zilog->zl_vdev_lock, NULL, MUTEX_DEFAULT, NULL); |
1822 | ||
1823 | avl_create(&zilog->zl_vdev_tree, zil_vdev_compare, | |
1824 | sizeof (zil_vdev_node_t), offsetof(zil_vdev_node_t, zv_node)); | |
1825 | ||
1826 | cv_init(&zilog->zl_cv_writer, NULL, CV_DEFAULT, NULL); | |
1827 | cv_init(&zilog->zl_cv_suspend, NULL, CV_DEFAULT, NULL); | |
572e2857 BB |
1828 | cv_init(&zilog->zl_cv_batch[0], NULL, CV_DEFAULT, NULL); |
1829 | cv_init(&zilog->zl_cv_batch[1], NULL, CV_DEFAULT, NULL); | |
34dc7c2f BB |
1830 | |
1831 | return (zilog); | |
1832 | } | |
1833 | ||
1834 | void | |
1835 | zil_free(zilog_t *zilog) | |
1836 | { | |
d6320ddb | 1837 | int i; |
34dc7c2f BB |
1838 | |
1839 | zilog->zl_stop_sync = 1; | |
1840 | ||
13fe0198 MA |
1841 | ASSERT0(zilog->zl_suspend); |
1842 | ASSERT0(zilog->zl_suspending); | |
1843 | ||
3e31d2b0 | 1844 | ASSERT(list_is_empty(&zilog->zl_lwb_list)); |
34dc7c2f BB |
1845 | list_destroy(&zilog->zl_lwb_list); |
1846 | ||
1847 | avl_destroy(&zilog->zl_vdev_tree); | |
1848 | mutex_destroy(&zilog->zl_vdev_lock); | |
1849 | ||
572e2857 BB |
1850 | ASSERT(list_is_empty(&zilog->zl_itx_commit_list)); |
1851 | list_destroy(&zilog->zl_itx_commit_list); | |
1852 | ||
d6320ddb | 1853 | for (i = 0; i < TXG_SIZE; i++) { |
572e2857 BB |
1854 | /* |
1855 | * It's possible for an itx to be generated that doesn't dirty | |
1856 | * a txg (e.g. ztest TX_TRUNCATE). So there's no zil_clean() | |
1857 | * callback to remove the entry. We remove those here. | |
1858 | * | |
1859 | * Also free up the ziltest itxs. | |
1860 | */ | |
1861 | if (zilog->zl_itxg[i].itxg_itxs) | |
1862 | zil_itxg_clean(zilog->zl_itxg[i].itxg_itxs); | |
1863 | mutex_destroy(&zilog->zl_itxg[i].itxg_lock); | |
1864 | } | |
1865 | ||
34dc7c2f BB |
1866 | mutex_destroy(&zilog->zl_lock); |
1867 | ||
1868 | cv_destroy(&zilog->zl_cv_writer); | |
1869 | cv_destroy(&zilog->zl_cv_suspend); | |
572e2857 BB |
1870 | cv_destroy(&zilog->zl_cv_batch[0]); |
1871 | cv_destroy(&zilog->zl_cv_batch[1]); | |
34dc7c2f BB |
1872 | |
1873 | kmem_free(zilog, sizeof (zilog_t)); | |
1874 | } | |
1875 | ||
34dc7c2f BB |
1876 | /* |
1877 | * Open an intent log. | |
1878 | */ | |
1879 | zilog_t * | |
1880 | zil_open(objset_t *os, zil_get_data_t *get_data) | |
1881 | { | |
1882 | zilog_t *zilog = dmu_objset_zil(os); | |
1883 | ||
3e31d2b0 ES |
1884 | ASSERT(zilog->zl_clean_taskq == NULL); |
1885 | ASSERT(zilog->zl_get_data == NULL); | |
1886 | ASSERT(list_is_empty(&zilog->zl_lwb_list)); | |
1887 | ||
34dc7c2f BB |
1888 | zilog->zl_get_data = get_data; |
1889 | zilog->zl_clean_taskq = taskq_create("zil_clean", 1, minclsyspri, | |
1890 | 2, 2, TASKQ_PREPOPULATE); | |
1891 | ||
1892 | return (zilog); | |
1893 | } | |
1894 | ||
1895 | /* | |
1896 | * Close an intent log. | |
1897 | */ | |
1898 | void | |
1899 | zil_close(zilog_t *zilog) | |
1900 | { | |
3e31d2b0 | 1901 | lwb_t *lwb; |
572e2857 BB |
1902 | uint64_t txg = 0; |
1903 | ||
1904 | zil_commit(zilog, 0); /* commit all itx */ | |
1905 | ||
34dc7c2f | 1906 | /* |
572e2857 BB |
1907 | * The lwb_max_txg for the stubby lwb will reflect the last activity |
1908 | * for the zil. After a txg_wait_synced() on the txg we know all the | |
1909 | * callbacks have occurred that may clean the zil. Only then can we | |
1910 | * destroy the zl_clean_taskq. | |
34dc7c2f | 1911 | */ |
572e2857 | 1912 | mutex_enter(&zilog->zl_lock); |
3e31d2b0 ES |
1913 | lwb = list_tail(&zilog->zl_lwb_list); |
1914 | if (lwb != NULL) | |
1915 | txg = lwb->lwb_max_txg; | |
572e2857 BB |
1916 | mutex_exit(&zilog->zl_lock); |
1917 | if (txg) | |
34dc7c2f | 1918 | txg_wait_synced(zilog->zl_dmu_pool, txg); |
29809a6c | 1919 | ASSERT(!zilog_is_dirty(zilog)); |
34dc7c2f BB |
1920 | |
1921 | taskq_destroy(zilog->zl_clean_taskq); | |
1922 | zilog->zl_clean_taskq = NULL; | |
1923 | zilog->zl_get_data = NULL; | |
3e31d2b0 ES |
1924 | |
1925 | /* | |
1926 | * We should have only one LWB left on the list; remove it now. | |
1927 | */ | |
1928 | mutex_enter(&zilog->zl_lock); | |
1929 | lwb = list_head(&zilog->zl_lwb_list); | |
1930 | if (lwb != NULL) { | |
1931 | ASSERT(lwb == list_tail(&zilog->zl_lwb_list)); | |
920dd524 ED |
1932 | ASSERT(lwb->lwb_zio == NULL); |
1933 | if (lwb->lwb_fastwrite) | |
1934 | metaslab_fastwrite_unmark(zilog->zl_spa, &lwb->lwb_blk); | |
3e31d2b0 ES |
1935 | list_remove(&zilog->zl_lwb_list, lwb); |
1936 | zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); | |
1937 | kmem_cache_free(zil_lwb_cache, lwb); | |
1938 | } | |
1939 | mutex_exit(&zilog->zl_lock); | |
34dc7c2f BB |
1940 | } |
1941 | ||
13fe0198 MA |
1942 | static char *suspend_tag = "zil suspending"; |
1943 | ||
34dc7c2f BB |
1944 | /* |
1945 | * Suspend an intent log. While in suspended mode, we still honor | |
1946 | * synchronous semantics, but we rely on txg_wait_synced() to do it. | |
13fe0198 MA |
1947 | * On old version pools, we suspend the log briefly when taking a |
1948 | * snapshot so that it will have an empty intent log. | |
1949 | * | |
1950 | * Long holds are not really intended to be used the way we do here -- | |
1951 | * held for such a short time. A concurrent caller of dsl_dataset_long_held() | |
1952 | * could fail. Therefore we take pains to only put a long hold if it is | |
1953 | * actually necessary. Fortunately, it will only be necessary if the | |
1954 | * objset is currently mounted (or the ZVOL equivalent). In that case it | |
1955 | * will already have a long hold, so we are not really making things any worse. | |
1956 | * | |
1957 | * Ideally, we would locate the existing long-holder (i.e. the zfsvfs_t or | |
1958 | * zvol_state_t), and use their mechanism to prevent their hold from being | |
1959 | * dropped (e.g. VFS_HOLD()). However, that would be even more pain for | |
1960 | * very little gain. | |
1961 | * | |
1962 | * if cookiep == NULL, this does both the suspend & resume. | |
1963 | * Otherwise, it returns with the dataset "long held", and the cookie | |
1964 | * should be passed into zil_resume(). | |
34dc7c2f BB |
1965 | */ |
1966 | int | |
13fe0198 | 1967 | zil_suspend(const char *osname, void **cookiep) |
34dc7c2f | 1968 | { |
13fe0198 MA |
1969 | objset_t *os; |
1970 | zilog_t *zilog; | |
1971 | const zil_header_t *zh; | |
1972 | int error; | |
1973 | ||
1974 | error = dmu_objset_hold(osname, suspend_tag, &os); | |
1975 | if (error != 0) | |
1976 | return (error); | |
1977 | zilog = dmu_objset_zil(os); | |
34dc7c2f BB |
1978 | |
1979 | mutex_enter(&zilog->zl_lock); | |
13fe0198 MA |
1980 | zh = zilog->zl_header; |
1981 | ||
9babb374 | 1982 | if (zh->zh_flags & ZIL_REPLAY_NEEDED) { /* unplayed log */ |
34dc7c2f | 1983 | mutex_exit(&zilog->zl_lock); |
13fe0198 | 1984 | dmu_objset_rele(os, suspend_tag); |
2e528b49 | 1985 | return (SET_ERROR(EBUSY)); |
34dc7c2f | 1986 | } |
13fe0198 MA |
1987 | |
1988 | /* | |
1989 | * Don't put a long hold in the cases where we can avoid it. This | |
1990 | * is when there is no cookie so we are doing a suspend & resume | |
1991 | * (i.e. called from zil_vdev_offline()), and there's nothing to do | |
1992 | * for the suspend because it's already suspended, or there's no ZIL. | |
1993 | */ | |
1994 | if (cookiep == NULL && !zilog->zl_suspending && | |
1995 | (zilog->zl_suspend > 0 || BP_IS_HOLE(&zh->zh_log))) { | |
1996 | mutex_exit(&zilog->zl_lock); | |
1997 | dmu_objset_rele(os, suspend_tag); | |
1998 | return (0); | |
1999 | } | |
2000 | ||
2001 | dsl_dataset_long_hold(dmu_objset_ds(os), suspend_tag); | |
2002 | dsl_pool_rele(dmu_objset_pool(os), suspend_tag); | |
2003 | ||
2004 | zilog->zl_suspend++; | |
2005 | ||
2006 | if (zilog->zl_suspend > 1) { | |
34dc7c2f | 2007 | /* |
13fe0198 | 2008 | * Someone else is already suspending it. |
34dc7c2f BB |
2009 | * Just wait for them to finish. |
2010 | */ | |
13fe0198 | 2011 | |
34dc7c2f BB |
2012 | while (zilog->zl_suspending) |
2013 | cv_wait(&zilog->zl_cv_suspend, &zilog->zl_lock); | |
34dc7c2f | 2014 | mutex_exit(&zilog->zl_lock); |
13fe0198 MA |
2015 | |
2016 | if (cookiep == NULL) | |
2017 | zil_resume(os); | |
2018 | else | |
2019 | *cookiep = os; | |
2020 | return (0); | |
2021 | } | |
2022 | ||
2023 | /* | |
2024 | * If there is no pointer to an on-disk block, this ZIL must not | |
2025 | * be active (e.g. filesystem not mounted), so there's nothing | |
2026 | * to clean up. | |
2027 | */ | |
2028 | if (BP_IS_HOLE(&zh->zh_log)) { | |
2029 | ASSERT(cookiep != NULL); /* fast path already handled */ | |
2030 | ||
2031 | *cookiep = os; | |
2032 | mutex_exit(&zilog->zl_lock); | |
34dc7c2f BB |
2033 | return (0); |
2034 | } | |
13fe0198 | 2035 | |
34dc7c2f BB |
2036 | zilog->zl_suspending = B_TRUE; |
2037 | mutex_exit(&zilog->zl_lock); | |
2038 | ||
572e2857 | 2039 | zil_commit(zilog, 0); |
34dc7c2f BB |
2040 | |
2041 | zil_destroy(zilog, B_FALSE); | |
2042 | ||
2043 | mutex_enter(&zilog->zl_lock); | |
2044 | zilog->zl_suspending = B_FALSE; | |
2045 | cv_broadcast(&zilog->zl_cv_suspend); | |
2046 | mutex_exit(&zilog->zl_lock); | |
2047 | ||
13fe0198 MA |
2048 | if (cookiep == NULL) |
2049 | zil_resume(os); | |
2050 | else | |
2051 | *cookiep = os; | |
34dc7c2f BB |
2052 | return (0); |
2053 | } | |
2054 | ||
2055 | void | |
13fe0198 | 2056 | zil_resume(void *cookie) |
34dc7c2f | 2057 | { |
13fe0198 MA |
2058 | objset_t *os = cookie; |
2059 | zilog_t *zilog = dmu_objset_zil(os); | |
2060 | ||
34dc7c2f BB |
2061 | mutex_enter(&zilog->zl_lock); |
2062 | ASSERT(zilog->zl_suspend != 0); | |
2063 | zilog->zl_suspend--; | |
2064 | mutex_exit(&zilog->zl_lock); | |
13fe0198 MA |
2065 | dsl_dataset_long_rele(dmu_objset_ds(os), suspend_tag); |
2066 | dsl_dataset_rele(dmu_objset_ds(os), suspend_tag); | |
34dc7c2f BB |
2067 | } |
2068 | ||
2069 | typedef struct zil_replay_arg { | |
b01615d5 | 2070 | zil_replay_func_t *zr_replay; |
34dc7c2f | 2071 | void *zr_arg; |
34dc7c2f | 2072 | boolean_t zr_byteswap; |
428870ff | 2073 | char *zr_lr; |
34dc7c2f BB |
2074 | } zil_replay_arg_t; |
2075 | ||
428870ff BB |
2076 | static int |
2077 | zil_replay_error(zilog_t *zilog, lr_t *lr, int error) | |
2078 | { | |
2079 | char name[MAXNAMELEN]; | |
2080 | ||
2081 | zilog->zl_replaying_seq--; /* didn't actually replay this one */ | |
2082 | ||
2083 | dmu_objset_name(zilog->zl_os, name); | |
2084 | ||
2085 | cmn_err(CE_WARN, "ZFS replay transaction error %d, " | |
2086 | "dataset %s, seq 0x%llx, txtype %llu %s\n", error, name, | |
2087 | (u_longlong_t)lr->lrc_seq, | |
2088 | (u_longlong_t)(lr->lrc_txtype & ~TX_CI), | |
2089 | (lr->lrc_txtype & TX_CI) ? "CI" : ""); | |
2090 | ||
2091 | return (error); | |
2092 | } | |
2093 | ||
2094 | static int | |
34dc7c2f BB |
2095 | zil_replay_log_record(zilog_t *zilog, lr_t *lr, void *zra, uint64_t claim_txg) |
2096 | { | |
2097 | zil_replay_arg_t *zr = zra; | |
2098 | const zil_header_t *zh = zilog->zl_header; | |
2099 | uint64_t reclen = lr->lrc_reclen; | |
2100 | uint64_t txtype = lr->lrc_txtype; | |
428870ff | 2101 | int error = 0; |
34dc7c2f | 2102 | |
428870ff | 2103 | zilog->zl_replaying_seq = lr->lrc_seq; |
34dc7c2f BB |
2104 | |
2105 | if (lr->lrc_seq <= zh->zh_replay_seq) /* already replayed */ | |
428870ff BB |
2106 | return (0); |
2107 | ||
2108 | if (lr->lrc_txg < claim_txg) /* already committed */ | |
2109 | return (0); | |
34dc7c2f BB |
2110 | |
2111 | /* Strip case-insensitive bit, still present in log record */ | |
2112 | txtype &= ~TX_CI; | |
2113 | ||
428870ff BB |
2114 | if (txtype == 0 || txtype >= TX_MAX_TYPE) |
2115 | return (zil_replay_error(zilog, lr, EINVAL)); | |
2116 | ||
2117 | /* | |
2118 | * If this record type can be logged out of order, the object | |
2119 | * (lr_foid) may no longer exist. That's legitimate, not an error. | |
2120 | */ | |
2121 | if (TX_OOO(txtype)) { | |
2122 | error = dmu_object_info(zilog->zl_os, | |
2123 | ((lr_ooo_t *)lr)->lr_foid, NULL); | |
2124 | if (error == ENOENT || error == EEXIST) | |
2125 | return (0); | |
fb5f0bc8 BB |
2126 | } |
2127 | ||
34dc7c2f BB |
2128 | /* |
2129 | * Make a copy of the data so we can revise and extend it. | |
2130 | */ | |
428870ff BB |
2131 | bcopy(lr, zr->zr_lr, reclen); |
2132 | ||
2133 | /* | |
2134 | * If this is a TX_WRITE with a blkptr, suck in the data. | |
2135 | */ | |
2136 | if (txtype == TX_WRITE && reclen == sizeof (lr_write_t)) { | |
2137 | error = zil_read_log_data(zilog, (lr_write_t *)lr, | |
2138 | zr->zr_lr + reclen); | |
13fe0198 | 2139 | if (error != 0) |
428870ff BB |
2140 | return (zil_replay_error(zilog, lr, error)); |
2141 | } | |
34dc7c2f BB |
2142 | |
2143 | /* | |
2144 | * The log block containing this lr may have been byteswapped | |
2145 | * so that we can easily examine common fields like lrc_txtype. | |
428870ff | 2146 | * However, the log is a mix of different record types, and only the |
34dc7c2f BB |
2147 | * replay vectors know how to byteswap their records. Therefore, if |
2148 | * the lr was byteswapped, undo it before invoking the replay vector. | |
2149 | */ | |
2150 | if (zr->zr_byteswap) | |
428870ff | 2151 | byteswap_uint64_array(zr->zr_lr, reclen); |
34dc7c2f BB |
2152 | |
2153 | /* | |
2154 | * We must now do two things atomically: replay this log record, | |
fb5f0bc8 BB |
2155 | * and update the log header sequence number to reflect the fact that |
2156 | * we did so. At the end of each replay function the sequence number | |
2157 | * is updated if we are in replay mode. | |
34dc7c2f | 2158 | */ |
428870ff | 2159 | error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lr, zr->zr_byteswap); |
13fe0198 | 2160 | if (error != 0) { |
34dc7c2f BB |
2161 | /* |
2162 | * The DMU's dnode layer doesn't see removes until the txg | |
2163 | * commits, so a subsequent claim can spuriously fail with | |
fb5f0bc8 | 2164 | * EEXIST. So if we receive any error we try syncing out |
428870ff BB |
2165 | * any removes then retry the transaction. Note that we |
2166 | * specify B_FALSE for byteswap now, so we don't do it twice. | |
34dc7c2f | 2167 | */ |
428870ff BB |
2168 | txg_wait_synced(spa_get_dsl(zilog->zl_spa), 0); |
2169 | error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lr, B_FALSE); | |
13fe0198 | 2170 | if (error != 0) |
428870ff | 2171 | return (zil_replay_error(zilog, lr, error)); |
34dc7c2f | 2172 | } |
428870ff | 2173 | return (0); |
34dc7c2f BB |
2174 | } |
2175 | ||
2176 | /* ARGSUSED */ | |
428870ff | 2177 | static int |
34dc7c2f BB |
2178 | zil_incr_blks(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg) |
2179 | { | |
2180 | zilog->zl_replay_blks++; | |
428870ff BB |
2181 | |
2182 | return (0); | |
34dc7c2f BB |
2183 | } |
2184 | ||
2185 | /* | |
2186 | * If this dataset has a non-empty intent log, replay it and destroy it. | |
2187 | */ | |
2188 | void | |
b01615d5 | 2189 | zil_replay(objset_t *os, void *arg, zil_replay_func_t replay_func[TX_MAX_TYPE]) |
34dc7c2f BB |
2190 | { |
2191 | zilog_t *zilog = dmu_objset_zil(os); | |
2192 | const zil_header_t *zh = zilog->zl_header; | |
2193 | zil_replay_arg_t zr; | |
2194 | ||
9babb374 | 2195 | if ((zh->zh_flags & ZIL_REPLAY_NEEDED) == 0) { |
34dc7c2f BB |
2196 | zil_destroy(zilog, B_TRUE); |
2197 | return; | |
2198 | } | |
2199 | ||
34dc7c2f BB |
2200 | zr.zr_replay = replay_func; |
2201 | zr.zr_arg = arg; | |
34dc7c2f | 2202 | zr.zr_byteswap = BP_SHOULD_BYTESWAP(&zh->zh_log); |
79c76d5b | 2203 | zr.zr_lr = vmem_alloc(2 * SPA_MAXBLOCKSIZE, KM_SLEEP); |
34dc7c2f BB |
2204 | |
2205 | /* | |
2206 | * Wait for in-progress removes to sync before starting replay. | |
2207 | */ | |
2208 | txg_wait_synced(zilog->zl_dmu_pool, 0); | |
2209 | ||
fb5f0bc8 | 2210 | zilog->zl_replay = B_TRUE; |
428870ff | 2211 | zilog->zl_replay_time = ddi_get_lbolt(); |
34dc7c2f BB |
2212 | ASSERT(zilog->zl_replay_blks == 0); |
2213 | (void) zil_parse(zilog, zil_incr_blks, zil_replay_log_record, &zr, | |
2214 | zh->zh_claim_txg); | |
00b46022 | 2215 | vmem_free(zr.zr_lr, 2 * SPA_MAXBLOCKSIZE); |
34dc7c2f BB |
2216 | |
2217 | zil_destroy(zilog, B_FALSE); | |
2218 | txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg); | |
fb5f0bc8 | 2219 | zilog->zl_replay = B_FALSE; |
34dc7c2f BB |
2220 | } |
2221 | ||
428870ff BB |
2222 | boolean_t |
2223 | zil_replaying(zilog_t *zilog, dmu_tx_t *tx) | |
34dc7c2f | 2224 | { |
428870ff BB |
2225 | if (zilog->zl_sync == ZFS_SYNC_DISABLED) |
2226 | return (B_TRUE); | |
34dc7c2f | 2227 | |
428870ff BB |
2228 | if (zilog->zl_replay) { |
2229 | dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); | |
2230 | zilog->zl_replayed_seq[dmu_tx_get_txg(tx) & TXG_MASK] = | |
2231 | zilog->zl_replaying_seq; | |
2232 | return (B_TRUE); | |
34dc7c2f BB |
2233 | } |
2234 | ||
428870ff | 2235 | return (B_FALSE); |
34dc7c2f | 2236 | } |
9babb374 BB |
2237 | |
2238 | /* ARGSUSED */ | |
2239 | int | |
428870ff | 2240 | zil_vdev_offline(const char *osname, void *arg) |
9babb374 | 2241 | { |
9babb374 BB |
2242 | int error; |
2243 | ||
13fe0198 MA |
2244 | error = zil_suspend(osname, NULL); |
2245 | if (error != 0) | |
2e528b49 | 2246 | return (SET_ERROR(EEXIST)); |
13fe0198 | 2247 | return (0); |
9babb374 | 2248 | } |
c409e464 BB |
2249 | |
2250 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
0f699108 AZ |
2251 | EXPORT_SYMBOL(zil_alloc); |
2252 | EXPORT_SYMBOL(zil_free); | |
2253 | EXPORT_SYMBOL(zil_open); | |
2254 | EXPORT_SYMBOL(zil_close); | |
2255 | EXPORT_SYMBOL(zil_replay); | |
2256 | EXPORT_SYMBOL(zil_replaying); | |
2257 | EXPORT_SYMBOL(zil_destroy); | |
2258 | EXPORT_SYMBOL(zil_destroy_sync); | |
2259 | EXPORT_SYMBOL(zil_itx_create); | |
2260 | EXPORT_SYMBOL(zil_itx_destroy); | |
2261 | EXPORT_SYMBOL(zil_itx_assign); | |
2262 | EXPORT_SYMBOL(zil_commit); | |
2263 | EXPORT_SYMBOL(zil_vdev_offline); | |
2264 | EXPORT_SYMBOL(zil_claim); | |
2265 | EXPORT_SYMBOL(zil_check_log_chain); | |
2266 | EXPORT_SYMBOL(zil_sync); | |
2267 | EXPORT_SYMBOL(zil_clean); | |
2268 | EXPORT_SYMBOL(zil_suspend); | |
2269 | EXPORT_SYMBOL(zil_resume); | |
2270 | EXPORT_SYMBOL(zil_add_block); | |
2271 | EXPORT_SYMBOL(zil_bp_tree_add); | |
2272 | EXPORT_SYMBOL(zil_set_sync); | |
2273 | EXPORT_SYMBOL(zil_set_logbias); | |
2274 | ||
c409e464 BB |
2275 | module_param(zil_replay_disable, int, 0644); |
2276 | MODULE_PARM_DESC(zil_replay_disable, "Disable intent logging replay"); | |
2277 | ||
2278 | module_param(zfs_nocacheflush, int, 0644); | |
2279 | MODULE_PARM_DESC(zfs_nocacheflush, "Disable cache flushes"); | |
ee191e80 ED |
2280 | |
2281 | module_param(zil_slog_limit, ulong, 0644); | |
2282 | MODULE_PARM_DESC(zil_slog_limit, "Max commit bytes to separate log device"); | |
c409e464 | 2283 | #endif |