]>
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. |
d3c2ae1c | 23 | * Copyright (c) 2011, 2015 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> |
a6255b7f | 43 | #include <sys/abd.h> |
34dc7c2f BB |
44 | |
45 | /* | |
46 | * The zfs intent log (ZIL) saves transaction records of system calls | |
47 | * that change the file system in memory with enough information | |
48 | * to be able to replay them. These are stored in memory until | |
49 | * either the DMU transaction group (txg) commits them to the stable pool | |
50 | * and they can be discarded, or they are flushed to the stable log | |
51 | * (also in the pool) due to a fsync, O_DSYNC or other synchronous | |
52 | * requirement. In the event of a panic or power fail then those log | |
53 | * records (transactions) are replayed. | |
54 | * | |
55 | * There is one ZIL per file system. Its on-disk (pool) format consists | |
56 | * of 3 parts: | |
57 | * | |
58 | * - ZIL header | |
59 | * - ZIL blocks | |
60 | * - ZIL records | |
61 | * | |
62 | * A log record holds a system call transaction. Log blocks can | |
63 | * hold many log records and the blocks are chained together. | |
64 | * Each ZIL block contains a block pointer (blkptr_t) to the next | |
65 | * ZIL block in the chain. The ZIL header points to the first | |
66 | * block in the chain. Note there is not a fixed place in the pool | |
67 | * to hold blocks. They are dynamically allocated and freed as | |
68 | * needed from the blocks available. Figure X shows the ZIL structure: | |
69 | */ | |
70 | ||
b6ad9671 ED |
71 | /* |
72 | * See zil.h for more information about these fields. | |
73 | */ | |
74 | zil_stats_t zil_stats = { | |
d1d7e268 MK |
75 | { "zil_commit_count", KSTAT_DATA_UINT64 }, |
76 | { "zil_commit_writer_count", KSTAT_DATA_UINT64 }, | |
77 | { "zil_itx_count", KSTAT_DATA_UINT64 }, | |
78 | { "zil_itx_indirect_count", KSTAT_DATA_UINT64 }, | |
79 | { "zil_itx_indirect_bytes", KSTAT_DATA_UINT64 }, | |
80 | { "zil_itx_copied_count", KSTAT_DATA_UINT64 }, | |
81 | { "zil_itx_copied_bytes", KSTAT_DATA_UINT64 }, | |
82 | { "zil_itx_needcopy_count", KSTAT_DATA_UINT64 }, | |
83 | { "zil_itx_needcopy_bytes", KSTAT_DATA_UINT64 }, | |
84 | { "zil_itx_metaslab_normal_count", KSTAT_DATA_UINT64 }, | |
85 | { "zil_itx_metaslab_normal_bytes", KSTAT_DATA_UINT64 }, | |
86 | { "zil_itx_metaslab_slog_count", KSTAT_DATA_UINT64 }, | |
87 | { "zil_itx_metaslab_slog_bytes", KSTAT_DATA_UINT64 }, | |
b6ad9671 ED |
88 | }; |
89 | ||
90 | static kstat_t *zil_ksp; | |
91 | ||
34dc7c2f | 92 | /* |
d3cc8b15 | 93 | * Disable intent logging replay. This global ZIL switch affects all pools. |
34dc7c2f | 94 | */ |
d3cc8b15 | 95 | int zil_replay_disable = 0; |
34dc7c2f BB |
96 | |
97 | /* | |
98 | * Tunable parameter for debugging or performance analysis. Setting | |
99 | * zfs_nocacheflush will cause corruption on power loss if a volatile | |
100 | * out-of-order write cache is enabled. | |
101 | */ | |
c409e464 | 102 | int zfs_nocacheflush = 0; |
34dc7c2f BB |
103 | |
104 | static kmem_cache_t *zil_lwb_cache; | |
105 | ||
572e2857 | 106 | static void zil_async_to_sync(zilog_t *zilog, uint64_t foid); |
428870ff BB |
107 | |
108 | #define LWB_EMPTY(lwb) ((BP_GET_LSIZE(&lwb->lwb_blk) - \ | |
109 | sizeof (zil_chain_t)) == (lwb->lwb_sz - lwb->lwb_nused)) | |
110 | ||
111 | ||
572e2857 BB |
112 | /* |
113 | * ziltest is by and large an ugly hack, but very useful in | |
114 | * checking replay without tedious work. | |
115 | * When running ziltest we want to keep all itx's and so maintain | |
116 | * a single list in the zl_itxg[] that uses a high txg: ZILTEST_TXG | |
117 | * We subtract TXG_CONCURRENT_STATES to allow for common code. | |
118 | */ | |
119 | #define ZILTEST_TXG (UINT64_MAX - TXG_CONCURRENT_STATES) | |
120 | ||
34dc7c2f | 121 | static int |
428870ff | 122 | zil_bp_compare(const void *x1, const void *x2) |
34dc7c2f | 123 | { |
428870ff BB |
124 | const dva_t *dva1 = &((zil_bp_node_t *)x1)->zn_dva; |
125 | const dva_t *dva2 = &((zil_bp_node_t *)x2)->zn_dva; | |
34dc7c2f | 126 | |
ee36c709 GN |
127 | int cmp = AVL_CMP(DVA_GET_VDEV(dva1), DVA_GET_VDEV(dva2)); |
128 | if (likely(cmp)) | |
129 | return (cmp); | |
34dc7c2f | 130 | |
ee36c709 | 131 | return (AVL_CMP(DVA_GET_OFFSET(dva1), DVA_GET_OFFSET(dva2))); |
34dc7c2f BB |
132 | } |
133 | ||
134 | static void | |
428870ff | 135 | zil_bp_tree_init(zilog_t *zilog) |
34dc7c2f | 136 | { |
428870ff BB |
137 | avl_create(&zilog->zl_bp_tree, zil_bp_compare, |
138 | sizeof (zil_bp_node_t), offsetof(zil_bp_node_t, zn_node)); | |
34dc7c2f BB |
139 | } |
140 | ||
141 | static void | |
428870ff | 142 | zil_bp_tree_fini(zilog_t *zilog) |
34dc7c2f | 143 | { |
428870ff BB |
144 | avl_tree_t *t = &zilog->zl_bp_tree; |
145 | zil_bp_node_t *zn; | |
34dc7c2f BB |
146 | void *cookie = NULL; |
147 | ||
148 | while ((zn = avl_destroy_nodes(t, &cookie)) != NULL) | |
428870ff | 149 | kmem_free(zn, sizeof (zil_bp_node_t)); |
34dc7c2f BB |
150 | |
151 | avl_destroy(t); | |
152 | } | |
153 | ||
428870ff BB |
154 | int |
155 | zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp) | |
34dc7c2f | 156 | { |
428870ff | 157 | avl_tree_t *t = &zilog->zl_bp_tree; |
9b67f605 | 158 | const dva_t *dva; |
428870ff | 159 | zil_bp_node_t *zn; |
34dc7c2f BB |
160 | avl_index_t where; |
161 | ||
9b67f605 MA |
162 | if (BP_IS_EMBEDDED(bp)) |
163 | return (0); | |
164 | ||
165 | dva = BP_IDENTITY(bp); | |
166 | ||
34dc7c2f | 167 | if (avl_find(t, dva, &where) != NULL) |
2e528b49 | 168 | return (SET_ERROR(EEXIST)); |
34dc7c2f | 169 | |
79c76d5b | 170 | zn = kmem_alloc(sizeof (zil_bp_node_t), KM_SLEEP); |
34dc7c2f BB |
171 | zn->zn_dva = *dva; |
172 | avl_insert(t, zn, where); | |
173 | ||
174 | return (0); | |
175 | } | |
176 | ||
177 | static zil_header_t * | |
178 | zil_header_in_syncing_context(zilog_t *zilog) | |
179 | { | |
180 | return ((zil_header_t *)zilog->zl_header); | |
181 | } | |
182 | ||
183 | static void | |
184 | zil_init_log_chain(zilog_t *zilog, blkptr_t *bp) | |
185 | { | |
186 | zio_cksum_t *zc = &bp->blk_cksum; | |
187 | ||
188 | zc->zc_word[ZIL_ZC_GUID_0] = spa_get_random(-1ULL); | |
189 | zc->zc_word[ZIL_ZC_GUID_1] = spa_get_random(-1ULL); | |
190 | zc->zc_word[ZIL_ZC_OBJSET] = dmu_objset_id(zilog->zl_os); | |
191 | zc->zc_word[ZIL_ZC_SEQ] = 1ULL; | |
192 | } | |
193 | ||
194 | /* | |
428870ff | 195 | * Read a log block and make sure it's valid. |
34dc7c2f BB |
196 | */ |
197 | static int | |
428870ff BB |
198 | zil_read_log_block(zilog_t *zilog, const blkptr_t *bp, blkptr_t *nbp, void *dst, |
199 | char **end) | |
34dc7c2f | 200 | { |
428870ff | 201 | enum zio_flag zio_flags = ZIO_FLAG_CANFAIL; |
2a432414 | 202 | arc_flags_t aflags = ARC_FLAG_WAIT; |
428870ff | 203 | arc_buf_t *abuf = NULL; |
5dbd68a3 | 204 | zbookmark_phys_t zb; |
34dc7c2f BB |
205 | int error; |
206 | ||
428870ff BB |
207 | if (zilog->zl_header->zh_claim_txg == 0) |
208 | zio_flags |= ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB; | |
34dc7c2f | 209 | |
428870ff BB |
210 | if (!(zilog->zl_header->zh_flags & ZIL_CLAIM_LR_SEQ_VALID)) |
211 | zio_flags |= ZIO_FLAG_SPECULATIVE; | |
34dc7c2f | 212 | |
428870ff BB |
213 | SET_BOOKMARK(&zb, bp->blk_cksum.zc_word[ZIL_ZC_OBJSET], |
214 | ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, bp->blk_cksum.zc_word[ZIL_ZC_SEQ]); | |
215 | ||
294f6806 | 216 | error = arc_read(NULL, zilog->zl_spa, bp, arc_getbuf_func, &abuf, |
428870ff | 217 | ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb); |
34dc7c2f BB |
218 | |
219 | if (error == 0) { | |
34dc7c2f BB |
220 | zio_cksum_t cksum = bp->blk_cksum; |
221 | ||
222 | /* | |
b128c09f BB |
223 | * Validate the checksummed log block. |
224 | * | |
34dc7c2f BB |
225 | * Sequence numbers should be... sequential. The checksum |
226 | * verifier for the next block should be bp's checksum plus 1. | |
b128c09f BB |
227 | * |
228 | * Also check the log chain linkage and size used. | |
34dc7c2f BB |
229 | */ |
230 | cksum.zc_word[ZIL_ZC_SEQ]++; | |
231 | ||
428870ff BB |
232 | if (BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_ZILOG2) { |
233 | zil_chain_t *zilc = abuf->b_data; | |
234 | char *lr = (char *)(zilc + 1); | |
235 | uint64_t len = zilc->zc_nused - sizeof (zil_chain_t); | |
34dc7c2f | 236 | |
428870ff BB |
237 | if (bcmp(&cksum, &zilc->zc_next_blk.blk_cksum, |
238 | sizeof (cksum)) || BP_IS_HOLE(&zilc->zc_next_blk)) { | |
2e528b49 | 239 | error = SET_ERROR(ECKSUM); |
428870ff | 240 | } else { |
f1512ee6 | 241 | ASSERT3U(len, <=, SPA_OLD_MAXBLOCKSIZE); |
428870ff BB |
242 | bcopy(lr, dst, len); |
243 | *end = (char *)dst + len; | |
244 | *nbp = zilc->zc_next_blk; | |
245 | } | |
246 | } else { | |
247 | char *lr = abuf->b_data; | |
248 | uint64_t size = BP_GET_LSIZE(bp); | |
249 | zil_chain_t *zilc = (zil_chain_t *)(lr + size) - 1; | |
250 | ||
251 | if (bcmp(&cksum, &zilc->zc_next_blk.blk_cksum, | |
252 | sizeof (cksum)) || BP_IS_HOLE(&zilc->zc_next_blk) || | |
253 | (zilc->zc_nused > (size - sizeof (*zilc)))) { | |
2e528b49 | 254 | error = SET_ERROR(ECKSUM); |
428870ff | 255 | } else { |
f1512ee6 MA |
256 | ASSERT3U(zilc->zc_nused, <=, |
257 | SPA_OLD_MAXBLOCKSIZE); | |
428870ff BB |
258 | bcopy(lr, dst, zilc->zc_nused); |
259 | *end = (char *)dst + zilc->zc_nused; | |
260 | *nbp = zilc->zc_next_blk; | |
261 | } | |
34dc7c2f | 262 | } |
428870ff | 263 | |
d3c2ae1c | 264 | arc_buf_destroy(abuf, &abuf); |
428870ff BB |
265 | } |
266 | ||
267 | return (error); | |
268 | } | |
269 | ||
270 | /* | |
271 | * Read a TX_WRITE log data block. | |
272 | */ | |
273 | static int | |
274 | zil_read_log_data(zilog_t *zilog, const lr_write_t *lr, void *wbuf) | |
275 | { | |
276 | enum zio_flag zio_flags = ZIO_FLAG_CANFAIL; | |
277 | const blkptr_t *bp = &lr->lr_blkptr; | |
2a432414 | 278 | arc_flags_t aflags = ARC_FLAG_WAIT; |
428870ff | 279 | arc_buf_t *abuf = NULL; |
5dbd68a3 | 280 | zbookmark_phys_t zb; |
428870ff BB |
281 | int error; |
282 | ||
283 | if (BP_IS_HOLE(bp)) { | |
284 | if (wbuf != NULL) | |
285 | bzero(wbuf, MAX(BP_GET_LSIZE(bp), lr->lr_length)); | |
286 | return (0); | |
34dc7c2f BB |
287 | } |
288 | ||
428870ff BB |
289 | if (zilog->zl_header->zh_claim_txg == 0) |
290 | zio_flags |= ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB; | |
291 | ||
292 | SET_BOOKMARK(&zb, dmu_objset_id(zilog->zl_os), lr->lr_foid, | |
293 | ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp)); | |
294 | ||
294f6806 | 295 | error = arc_read(NULL, zilog->zl_spa, bp, arc_getbuf_func, &abuf, |
428870ff BB |
296 | ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb); |
297 | ||
298 | if (error == 0) { | |
299 | if (wbuf != NULL) | |
300 | bcopy(abuf->b_data, wbuf, arc_buf_size(abuf)); | |
d3c2ae1c | 301 | arc_buf_destroy(abuf, &abuf); |
428870ff | 302 | } |
34dc7c2f BB |
303 | |
304 | return (error); | |
305 | } | |
306 | ||
307 | /* | |
308 | * Parse the intent log, and call parse_func for each valid record within. | |
34dc7c2f | 309 | */ |
428870ff | 310 | int |
34dc7c2f BB |
311 | zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func, |
312 | zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg) | |
313 | { | |
314 | const zil_header_t *zh = zilog->zl_header; | |
428870ff BB |
315 | boolean_t claimed = !!zh->zh_claim_txg; |
316 | uint64_t claim_blk_seq = claimed ? zh->zh_claim_blk_seq : UINT64_MAX; | |
317 | uint64_t claim_lr_seq = claimed ? zh->zh_claim_lr_seq : UINT64_MAX; | |
318 | uint64_t max_blk_seq = 0; | |
319 | uint64_t max_lr_seq = 0; | |
320 | uint64_t blk_count = 0; | |
321 | uint64_t lr_count = 0; | |
322 | blkptr_t blk, next_blk; | |
34dc7c2f | 323 | char *lrbuf, *lrp; |
428870ff | 324 | int error = 0; |
34dc7c2f | 325 | |
d1d7e268 | 326 | bzero(&next_blk, sizeof (blkptr_t)); |
d4ed6673 | 327 | |
428870ff BB |
328 | /* |
329 | * Old logs didn't record the maximum zh_claim_lr_seq. | |
330 | */ | |
331 | if (!(zh->zh_flags & ZIL_CLAIM_LR_SEQ_VALID)) | |
332 | claim_lr_seq = UINT64_MAX; | |
34dc7c2f BB |
333 | |
334 | /* | |
335 | * Starting at the block pointed to by zh_log we read the log chain. | |
336 | * For each block in the chain we strongly check that block to | |
337 | * ensure its validity. We stop when an invalid block is found. | |
338 | * For each block pointer in the chain we call parse_blk_func(). | |
339 | * For each record in each valid block we call parse_lr_func(). | |
340 | * If the log has been claimed, stop if we encounter a sequence | |
341 | * number greater than the highest claimed sequence number. | |
342 | */ | |
f1512ee6 | 343 | lrbuf = zio_buf_alloc(SPA_OLD_MAXBLOCKSIZE); |
428870ff | 344 | zil_bp_tree_init(zilog); |
34dc7c2f | 345 | |
428870ff BB |
346 | for (blk = zh->zh_log; !BP_IS_HOLE(&blk); blk = next_blk) { |
347 | uint64_t blk_seq = blk.blk_cksum.zc_word[ZIL_ZC_SEQ]; | |
348 | int reclen; | |
d4ed6673 | 349 | char *end = NULL; |
34dc7c2f | 350 | |
428870ff BB |
351 | if (blk_seq > claim_blk_seq) |
352 | break; | |
353 | if ((error = parse_blk_func(zilog, &blk, arg, txg)) != 0) | |
354 | break; | |
355 | ASSERT3U(max_blk_seq, <, blk_seq); | |
356 | max_blk_seq = blk_seq; | |
357 | blk_count++; | |
34dc7c2f | 358 | |
428870ff BB |
359 | if (max_lr_seq == claim_lr_seq && max_blk_seq == claim_blk_seq) |
360 | break; | |
34dc7c2f | 361 | |
428870ff | 362 | error = zil_read_log_block(zilog, &blk, &next_blk, lrbuf, &end); |
13fe0198 | 363 | if (error != 0) |
34dc7c2f BB |
364 | break; |
365 | ||
428870ff | 366 | for (lrp = lrbuf; lrp < end; lrp += reclen) { |
34dc7c2f BB |
367 | lr_t *lr = (lr_t *)lrp; |
368 | reclen = lr->lrc_reclen; | |
369 | ASSERT3U(reclen, >=, sizeof (lr_t)); | |
428870ff BB |
370 | if (lr->lrc_seq > claim_lr_seq) |
371 | goto done; | |
372 | if ((error = parse_lr_func(zilog, lr, arg, txg)) != 0) | |
373 | goto done; | |
374 | ASSERT3U(max_lr_seq, <, lr->lrc_seq); | |
375 | max_lr_seq = lr->lrc_seq; | |
376 | lr_count++; | |
34dc7c2f | 377 | } |
34dc7c2f | 378 | } |
428870ff BB |
379 | done: |
380 | zilog->zl_parse_error = error; | |
381 | zilog->zl_parse_blk_seq = max_blk_seq; | |
382 | zilog->zl_parse_lr_seq = max_lr_seq; | |
383 | zilog->zl_parse_blk_count = blk_count; | |
384 | zilog->zl_parse_lr_count = lr_count; | |
385 | ||
386 | ASSERT(!claimed || !(zh->zh_flags & ZIL_CLAIM_LR_SEQ_VALID) || | |
387 | (max_blk_seq == claim_blk_seq && max_lr_seq == claim_lr_seq)); | |
388 | ||
389 | zil_bp_tree_fini(zilog); | |
f1512ee6 | 390 | zio_buf_free(lrbuf, SPA_OLD_MAXBLOCKSIZE); |
34dc7c2f | 391 | |
428870ff | 392 | return (error); |
34dc7c2f BB |
393 | } |
394 | ||
428870ff | 395 | static int |
34dc7c2f BB |
396 | zil_claim_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t first_txg) |
397 | { | |
34dc7c2f BB |
398 | /* |
399 | * Claim log block if not already committed and not already claimed. | |
428870ff | 400 | * If tx == NULL, just verify that the block is claimable. |
34dc7c2f | 401 | */ |
b0bc7a84 MG |
402 | if (BP_IS_HOLE(bp) || bp->blk_birth < first_txg || |
403 | zil_bp_tree_add(zilog, bp) != 0) | |
428870ff BB |
404 | return (0); |
405 | ||
406 | return (zio_wait(zio_claim(NULL, zilog->zl_spa, | |
407 | tx == NULL ? 0 : first_txg, bp, spa_claim_notify, NULL, | |
408 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB))); | |
34dc7c2f BB |
409 | } |
410 | ||
428870ff | 411 | static int |
34dc7c2f BB |
412 | zil_claim_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t first_txg) |
413 | { | |
428870ff BB |
414 | lr_write_t *lr = (lr_write_t *)lrc; |
415 | int error; | |
416 | ||
417 | if (lrc->lrc_txtype != TX_WRITE) | |
418 | return (0); | |
419 | ||
420 | /* | |
421 | * If the block is not readable, don't claim it. This can happen | |
422 | * in normal operation when a log block is written to disk before | |
423 | * some of the dmu_sync() blocks it points to. In this case, the | |
424 | * transaction cannot have been committed to anyone (we would have | |
425 | * waited for all writes to be stable first), so it is semantically | |
426 | * correct to declare this the end of the log. | |
427 | */ | |
428 | if (lr->lr_blkptr.blk_birth >= first_txg && | |
429 | (error = zil_read_log_data(zilog, lr, NULL)) != 0) | |
430 | return (error); | |
431 | return (zil_claim_log_block(zilog, &lr->lr_blkptr, tx, first_txg)); | |
34dc7c2f BB |
432 | } |
433 | ||
434 | /* ARGSUSED */ | |
428870ff | 435 | static int |
34dc7c2f BB |
436 | zil_free_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t claim_txg) |
437 | { | |
428870ff BB |
438 | zio_free_zil(zilog->zl_spa, dmu_tx_get_txg(tx), bp); |
439 | ||
440 | return (0); | |
34dc7c2f BB |
441 | } |
442 | ||
428870ff | 443 | static int |
34dc7c2f BB |
444 | zil_free_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t claim_txg) |
445 | { | |
428870ff BB |
446 | lr_write_t *lr = (lr_write_t *)lrc; |
447 | blkptr_t *bp = &lr->lr_blkptr; | |
448 | ||
34dc7c2f BB |
449 | /* |
450 | * If we previously claimed it, we need to free it. | |
451 | */ | |
428870ff | 452 | if (claim_txg != 0 && lrc->lrc_txtype == TX_WRITE && |
b0bc7a84 MG |
453 | bp->blk_birth >= claim_txg && zil_bp_tree_add(zilog, bp) == 0 && |
454 | !BP_IS_HOLE(bp)) | |
428870ff BB |
455 | zio_free(zilog->zl_spa, dmu_tx_get_txg(tx), bp); |
456 | ||
457 | return (0); | |
458 | } | |
459 | ||
460 | static lwb_t * | |
920dd524 | 461 | zil_alloc_lwb(zilog_t *zilog, blkptr_t *bp, uint64_t txg, boolean_t fastwrite) |
428870ff BB |
462 | { |
463 | lwb_t *lwb; | |
464 | ||
79c76d5b | 465 | lwb = kmem_cache_alloc(zil_lwb_cache, KM_SLEEP); |
428870ff BB |
466 | lwb->lwb_zilog = zilog; |
467 | lwb->lwb_blk = *bp; | |
920dd524 | 468 | lwb->lwb_fastwrite = fastwrite; |
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 | ||
0c66c32d | 498 | if (ds->ds_is_snapshot) |
29809a6c MA |
499 | panic("dirtying snapshot!"); |
500 | ||
13fe0198 | 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 | ||
507 | boolean_t | |
508 | zilog_is_dirty(zilog_t *zilog) | |
509 | { | |
510 | dsl_pool_t *dp = zilog->zl_dmu_pool; | |
511 | int t; | |
512 | ||
513 | for (t = 0; t < TXG_SIZE; t++) { | |
514 | if (txg_list_member(&dp->dp_dirty_zilogs, zilog, t)) | |
515 | return (B_TRUE); | |
516 | } | |
517 | return (B_FALSE); | |
518 | } | |
519 | ||
34dc7c2f BB |
520 | /* |
521 | * Create an on-disk intent log. | |
522 | */ | |
428870ff | 523 | static lwb_t * |
34dc7c2f BB |
524 | zil_create(zilog_t *zilog) |
525 | { | |
526 | const zil_header_t *zh = zilog->zl_header; | |
428870ff | 527 | lwb_t *lwb = NULL; |
34dc7c2f BB |
528 | uint64_t txg = 0; |
529 | dmu_tx_t *tx = NULL; | |
530 | blkptr_t blk; | |
531 | int error = 0; | |
920dd524 | 532 | boolean_t fastwrite = FALSE; |
34dc7c2f BB |
533 | |
534 | /* | |
535 | * Wait for any previous destroy to complete. | |
536 | */ | |
537 | txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg); | |
538 | ||
539 | ASSERT(zh->zh_claim_txg == 0); | |
540 | ASSERT(zh->zh_replay_seq == 0); | |
541 | ||
542 | blk = zh->zh_log; | |
543 | ||
544 | /* | |
428870ff BB |
545 | * Allocate an initial log block if: |
546 | * - there isn't one already | |
547 | * - the existing block is the wrong endianess | |
34dc7c2f | 548 | */ |
fb5f0bc8 | 549 | if (BP_IS_HOLE(&blk) || BP_SHOULD_BYTESWAP(&blk)) { |
34dc7c2f | 550 | tx = dmu_tx_create(zilog->zl_os); |
428870ff | 551 | VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0); |
34dc7c2f BB |
552 | dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); |
553 | txg = dmu_tx_get_txg(tx); | |
554 | ||
fb5f0bc8 | 555 | if (!BP_IS_HOLE(&blk)) { |
428870ff | 556 | zio_free_zil(zilog->zl_spa, txg, &blk); |
fb5f0bc8 BB |
557 | BP_ZERO(&blk); |
558 | } | |
559 | ||
920dd524 | 560 | error = zio_alloc_zil(zilog->zl_spa, txg, &blk, |
5d7a86d1 | 561 | ZIL_MIN_BLKSZ, B_TRUE); |
920dd524 | 562 | fastwrite = TRUE; |
34dc7c2f BB |
563 | |
564 | if (error == 0) | |
565 | zil_init_log_chain(zilog, &blk); | |
566 | } | |
567 | ||
568 | /* | |
569 | * Allocate a log write buffer (lwb) for the first log block. | |
570 | */ | |
428870ff | 571 | if (error == 0) |
920dd524 | 572 | lwb = zil_alloc_lwb(zilog, &blk, txg, fastwrite); |
34dc7c2f BB |
573 | |
574 | /* | |
575 | * If we just allocated the first log block, commit our transaction | |
576 | * and wait for zil_sync() to stuff the block poiner into zh_log. | |
577 | * (zh is part of the MOS, so we cannot modify it in open context.) | |
578 | */ | |
579 | if (tx != NULL) { | |
580 | dmu_tx_commit(tx); | |
581 | txg_wait_synced(zilog->zl_dmu_pool, txg); | |
582 | } | |
583 | ||
584 | ASSERT(bcmp(&blk, &zh->zh_log, sizeof (blk)) == 0); | |
428870ff BB |
585 | |
586 | return (lwb); | |
34dc7c2f BB |
587 | } |
588 | ||
589 | /* | |
590 | * In one tx, free all log blocks and clear the log header. | |
591 | * If keep_first is set, then we're replaying a log with no content. | |
592 | * We want to keep the first block, however, so that the first | |
593 | * synchronous transaction doesn't require a txg_wait_synced() | |
594 | * in zil_create(). We don't need to txg_wait_synced() here either | |
595 | * when keep_first is set, because both zil_create() and zil_destroy() | |
596 | * will wait for any in-progress destroys to complete. | |
597 | */ | |
598 | void | |
599 | zil_destroy(zilog_t *zilog, boolean_t keep_first) | |
600 | { | |
601 | const zil_header_t *zh = zilog->zl_header; | |
602 | lwb_t *lwb; | |
603 | dmu_tx_t *tx; | |
604 | uint64_t txg; | |
605 | ||
606 | /* | |
607 | * Wait for any previous destroy to complete. | |
608 | */ | |
609 | txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg); | |
610 | ||
428870ff BB |
611 | zilog->zl_old_header = *zh; /* debugging aid */ |
612 | ||
34dc7c2f BB |
613 | if (BP_IS_HOLE(&zh->zh_log)) |
614 | return; | |
615 | ||
616 | tx = dmu_tx_create(zilog->zl_os); | |
428870ff | 617 | VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0); |
34dc7c2f BB |
618 | dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); |
619 | txg = dmu_tx_get_txg(tx); | |
620 | ||
621 | mutex_enter(&zilog->zl_lock); | |
622 | ||
34dc7c2f BB |
623 | ASSERT3U(zilog->zl_destroy_txg, <, txg); |
624 | zilog->zl_destroy_txg = txg; | |
625 | zilog->zl_keep_first = keep_first; | |
626 | ||
627 | if (!list_is_empty(&zilog->zl_lwb_list)) { | |
628 | ASSERT(zh->zh_claim_txg == 0); | |
3e31d2b0 | 629 | VERIFY(!keep_first); |
34dc7c2f | 630 | while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) { |
920dd524 ED |
631 | ASSERT(lwb->lwb_zio == NULL); |
632 | if (lwb->lwb_fastwrite) | |
633 | metaslab_fastwrite_unmark(zilog->zl_spa, | |
634 | &lwb->lwb_blk); | |
34dc7c2f BB |
635 | list_remove(&zilog->zl_lwb_list, lwb); |
636 | if (lwb->lwb_buf != NULL) | |
637 | zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); | |
428870ff | 638 | zio_free_zil(zilog->zl_spa, txg, &lwb->lwb_blk); |
34dc7c2f BB |
639 | kmem_cache_free(zil_lwb_cache, lwb); |
640 | } | |
428870ff | 641 | } else if (!keep_first) { |
29809a6c | 642 | zil_destroy_sync(zilog, tx); |
34dc7c2f BB |
643 | } |
644 | mutex_exit(&zilog->zl_lock); | |
645 | ||
646 | dmu_tx_commit(tx); | |
647 | } | |
648 | ||
29809a6c MA |
649 | void |
650 | zil_destroy_sync(zilog_t *zilog, dmu_tx_t *tx) | |
651 | { | |
652 | ASSERT(list_is_empty(&zilog->zl_lwb_list)); | |
653 | (void) zil_parse(zilog, zil_free_log_block, | |
654 | zil_free_log_record, tx, zilog->zl_header->zh_claim_txg); | |
655 | } | |
656 | ||
34dc7c2f | 657 | int |
9c43027b | 658 | zil_claim(dsl_pool_t *dp, dsl_dataset_t *ds, void *txarg) |
34dc7c2f BB |
659 | { |
660 | dmu_tx_t *tx = txarg; | |
661 | uint64_t first_txg = dmu_tx_get_txg(tx); | |
662 | zilog_t *zilog; | |
663 | zil_header_t *zh; | |
664 | objset_t *os; | |
665 | int error; | |
666 | ||
9c43027b AJ |
667 | error = dmu_objset_own_obj(dp, ds->ds_object, |
668 | DMU_OST_ANY, B_FALSE, FTAG, &os); | |
13fe0198 | 669 | if (error != 0) { |
6d9036f3 MA |
670 | /* |
671 | * EBUSY indicates that the objset is inconsistent, in which | |
672 | * case it can not have a ZIL. | |
673 | */ | |
674 | if (error != EBUSY) { | |
9c43027b AJ |
675 | cmn_err(CE_WARN, "can't open objset for %llu, error %u", |
676 | (unsigned long long)ds->ds_object, error); | |
6d9036f3 MA |
677 | } |
678 | ||
34dc7c2f BB |
679 | return (0); |
680 | } | |
681 | ||
682 | zilog = dmu_objset_zil(os); | |
683 | zh = zil_header_in_syncing_context(zilog); | |
684 | ||
428870ff | 685 | if (spa_get_log_state(zilog->zl_spa) == SPA_LOG_CLEAR) { |
9babb374 | 686 | if (!BP_IS_HOLE(&zh->zh_log)) |
428870ff | 687 | zio_free_zil(zilog->zl_spa, first_txg, &zh->zh_log); |
9babb374 BB |
688 | BP_ZERO(&zh->zh_log); |
689 | dsl_dataset_dirty(dmu_objset_ds(os), tx); | |
13fe0198 | 690 | dmu_objset_disown(os, FTAG); |
428870ff | 691 | return (0); |
9babb374 BB |
692 | } |
693 | ||
34dc7c2f BB |
694 | /* |
695 | * Claim all log blocks if we haven't already done so, and remember | |
696 | * the highest claimed sequence number. This ensures that if we can | |
697 | * read only part of the log now (e.g. due to a missing device), | |
698 | * but we can read the entire log later, we will not try to replay | |
699 | * or destroy beyond the last block we successfully claimed. | |
700 | */ | |
701 | ASSERT3U(zh->zh_claim_txg, <=, first_txg); | |
702 | if (zh->zh_claim_txg == 0 && !BP_IS_HOLE(&zh->zh_log)) { | |
428870ff | 703 | (void) zil_parse(zilog, zil_claim_log_block, |
34dc7c2f | 704 | zil_claim_log_record, tx, first_txg); |
428870ff BB |
705 | zh->zh_claim_txg = first_txg; |
706 | zh->zh_claim_blk_seq = zilog->zl_parse_blk_seq; | |
707 | zh->zh_claim_lr_seq = zilog->zl_parse_lr_seq; | |
708 | if (zilog->zl_parse_lr_count || zilog->zl_parse_blk_count > 1) | |
709 | zh->zh_flags |= ZIL_REPLAY_NEEDED; | |
710 | zh->zh_flags |= ZIL_CLAIM_LR_SEQ_VALID; | |
34dc7c2f BB |
711 | dsl_dataset_dirty(dmu_objset_ds(os), tx); |
712 | } | |
713 | ||
714 | ASSERT3U(first_txg, ==, (spa_last_synced_txg(zilog->zl_spa) + 1)); | |
13fe0198 | 715 | dmu_objset_disown(os, FTAG); |
34dc7c2f BB |
716 | return (0); |
717 | } | |
718 | ||
b128c09f BB |
719 | /* |
720 | * Check the log by walking the log chain. | |
721 | * Checksum errors are ok as they indicate the end of the chain. | |
722 | * Any other error (no device or read failure) returns an error. | |
723 | */ | |
9c43027b | 724 | /* ARGSUSED */ |
b128c09f | 725 | int |
9c43027b | 726 | zil_check_log_chain(dsl_pool_t *dp, dsl_dataset_t *ds, 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 | ||
9c43027b | 735 | error = dmu_objset_from_ds(ds, &os); |
13fe0198 | 736 | if (error != 0) { |
9c43027b AJ |
737 | cmn_err(CE_WARN, "can't open objset %llu, error %d", |
738 | (unsigned long long)ds->ds_object, error); | |
b128c09f BB |
739 | return (0); |
740 | } | |
741 | ||
742 | zilog = dmu_objset_zil(os); | |
572e2857 BB |
743 | bp = (blkptr_t *)&zilog->zl_header->zh_log; |
744 | ||
745 | /* | |
746 | * Check the first block and determine if it's on a log device | |
747 | * which may have been removed or faulted prior to loading this | |
748 | * pool. If so, there's no point in checking the rest of the log | |
749 | * as its content should have already been synced to the pool. | |
750 | */ | |
751 | if (!BP_IS_HOLE(bp)) { | |
752 | vdev_t *vd; | |
753 | boolean_t valid = B_TRUE; | |
754 | ||
755 | spa_config_enter(os->os_spa, SCL_STATE, FTAG, RW_READER); | |
756 | vd = vdev_lookup_top(os->os_spa, DVA_GET_VDEV(&bp->blk_dva[0])); | |
757 | if (vd->vdev_islog && vdev_is_dead(vd)) | |
758 | valid = vdev_log_state_valid(vd); | |
759 | spa_config_exit(os->os_spa, SCL_STATE, FTAG); | |
760 | ||
9c43027b | 761 | if (!valid) |
572e2857 | 762 | return (0); |
572e2857 | 763 | } |
b128c09f | 764 | |
428870ff BB |
765 | /* |
766 | * Because tx == NULL, zil_claim_log_block() will not actually claim | |
767 | * any blocks, but just determine whether it is possible to do so. | |
768 | * In addition to checking the log chain, zil_claim_log_block() | |
769 | * will invoke zio_claim() with a done func of spa_claim_notify(), | |
770 | * which will update spa_max_claim_txg. See spa_load() for details. | |
771 | */ | |
772 | error = zil_parse(zilog, zil_claim_log_block, zil_claim_log_record, tx, | |
773 | zilog->zl_header->zh_claim_txg ? -1ULL : spa_first_txg(os->os_spa)); | |
774 | ||
428870ff | 775 | return ((error == ECKSUM || error == ENOENT) ? 0 : error); |
b128c09f BB |
776 | } |
777 | ||
34dc7c2f BB |
778 | static int |
779 | zil_vdev_compare(const void *x1, const void *x2) | |
780 | { | |
572e2857 BB |
781 | const uint64_t v1 = ((zil_vdev_node_t *)x1)->zv_vdev; |
782 | const uint64_t v2 = ((zil_vdev_node_t *)x2)->zv_vdev; | |
34dc7c2f | 783 | |
ee36c709 | 784 | return (AVL_CMP(v1, v2)); |
34dc7c2f BB |
785 | } |
786 | ||
787 | void | |
428870ff | 788 | zil_add_block(zilog_t *zilog, const blkptr_t *bp) |
34dc7c2f BB |
789 | { |
790 | avl_tree_t *t = &zilog->zl_vdev_tree; | |
791 | avl_index_t where; | |
792 | zil_vdev_node_t *zv, zvsearch; | |
793 | int ndvas = BP_GET_NDVAS(bp); | |
794 | int i; | |
795 | ||
796 | if (zfs_nocacheflush) | |
797 | return; | |
798 | ||
799 | ASSERT(zilog->zl_writer); | |
800 | ||
801 | /* | |
802 | * Even though we're zl_writer, we still need a lock because the | |
803 | * zl_get_data() callbacks may have dmu_sync() done callbacks | |
804 | * that will run concurrently. | |
805 | */ | |
806 | mutex_enter(&zilog->zl_vdev_lock); | |
807 | for (i = 0; i < ndvas; i++) { | |
808 | zvsearch.zv_vdev = DVA_GET_VDEV(&bp->blk_dva[i]); | |
809 | if (avl_find(t, &zvsearch, &where) == NULL) { | |
79c76d5b | 810 | zv = kmem_alloc(sizeof (*zv), KM_SLEEP); |
34dc7c2f BB |
811 | zv->zv_vdev = zvsearch.zv_vdev; |
812 | avl_insert(t, zv, where); | |
813 | } | |
814 | } | |
815 | mutex_exit(&zilog->zl_vdev_lock); | |
816 | } | |
817 | ||
572e2857 | 818 | static void |
34dc7c2f BB |
819 | zil_flush_vdevs(zilog_t *zilog) |
820 | { | |
821 | spa_t *spa = zilog->zl_spa; | |
822 | avl_tree_t *t = &zilog->zl_vdev_tree; | |
823 | void *cookie = NULL; | |
824 | zil_vdev_node_t *zv; | |
825 | zio_t *zio; | |
826 | ||
827 | ASSERT(zilog->zl_writer); | |
828 | ||
829 | /* | |
830 | * We don't need zl_vdev_lock here because we're the zl_writer, | |
831 | * and all zl_get_data() callbacks are done. | |
832 | */ | |
833 | if (avl_numnodes(t) == 0) | |
834 | return; | |
835 | ||
b128c09f | 836 | spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); |
34dc7c2f | 837 | |
b128c09f | 838 | zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL); |
34dc7c2f BB |
839 | |
840 | while ((zv = avl_destroy_nodes(t, &cookie)) != NULL) { | |
841 | vdev_t *vd = vdev_lookup_top(spa, zv->zv_vdev); | |
842 | if (vd != NULL) | |
843 | zio_flush(zio, vd); | |
844 | kmem_free(zv, sizeof (*zv)); | |
845 | } | |
846 | ||
847 | /* | |
848 | * Wait for all the flushes to complete. Not all devices actually | |
849 | * support the DKIOCFLUSHWRITECACHE ioctl, so it's OK if it fails. | |
850 | */ | |
851 | (void) zio_wait(zio); | |
852 | ||
b128c09f | 853 | spa_config_exit(spa, SCL_STATE, FTAG); |
34dc7c2f BB |
854 | } |
855 | ||
856 | /* | |
857 | * Function called when a log block write completes | |
858 | */ | |
859 | static void | |
860 | zil_lwb_write_done(zio_t *zio) | |
861 | { | |
862 | lwb_t *lwb = zio->io_private; | |
863 | zilog_t *zilog = lwb->lwb_zilog; | |
428870ff | 864 | dmu_tx_t *tx = lwb->lwb_tx; |
34dc7c2f | 865 | |
b128c09f | 866 | ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF); |
b128c09f BB |
867 | ASSERT(BP_GET_TYPE(zio->io_bp) == DMU_OT_INTENT_LOG); |
868 | ASSERT(BP_GET_LEVEL(zio->io_bp) == 0); | |
869 | ASSERT(BP_GET_BYTEORDER(zio->io_bp) == ZFS_HOST_BYTEORDER); | |
870 | ASSERT(!BP_IS_GANG(zio->io_bp)); | |
871 | ASSERT(!BP_IS_HOLE(zio->io_bp)); | |
9b67f605 | 872 | ASSERT(BP_GET_FILL(zio->io_bp) == 0); |
b128c09f | 873 | |
34dc7c2f | 874 | /* |
9babb374 BB |
875 | * Ensure the lwb buffer pointer is cleared before releasing |
876 | * the txg. If we have had an allocation failure and | |
877 | * the txg is waiting to sync then we want want zil_sync() | |
878 | * to remove the lwb so that it's not picked up as the next new | |
879 | * one in zil_commit_writer(). zil_sync() will only remove | |
880 | * the lwb if lwb_buf is null. | |
34dc7c2f | 881 | */ |
a6255b7f | 882 | abd_put(zio->io_abd); |
34dc7c2f BB |
883 | zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); |
884 | mutex_enter(&zilog->zl_lock); | |
920dd524 ED |
885 | lwb->lwb_zio = NULL; |
886 | lwb->lwb_fastwrite = FALSE; | |
34dc7c2f | 887 | lwb->lwb_buf = NULL; |
428870ff BB |
888 | lwb->lwb_tx = NULL; |
889 | mutex_exit(&zilog->zl_lock); | |
9babb374 BB |
890 | |
891 | /* | |
892 | * Now that we've written this log block, we have a stable pointer | |
893 | * to the next block in the chain, so it's OK to let the txg in | |
428870ff | 894 | * which we allocated the next block sync. |
9babb374 | 895 | */ |
428870ff | 896 | dmu_tx_commit(tx); |
34dc7c2f BB |
897 | } |
898 | ||
899 | /* | |
900 | * Initialize the io for a log block. | |
34dc7c2f BB |
901 | */ |
902 | static void | |
903 | zil_lwb_write_init(zilog_t *zilog, lwb_t *lwb) | |
904 | { | |
5dbd68a3 | 905 | zbookmark_phys_t zb; |
34dc7c2f | 906 | |
428870ff BB |
907 | SET_BOOKMARK(&zb, lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_OBJSET], |
908 | ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, | |
909 | lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_SEQ]); | |
34dc7c2f BB |
910 | |
911 | if (zilog->zl_root_zio == NULL) { | |
912 | zilog->zl_root_zio = zio_root(zilog->zl_spa, NULL, NULL, | |
913 | ZIO_FLAG_CANFAIL); | |
914 | } | |
920dd524 ED |
915 | |
916 | /* Lock so zil_sync() doesn't fastwrite_unmark after zio is created */ | |
917 | mutex_enter(&zilog->zl_lock); | |
34dc7c2f | 918 | if (lwb->lwb_zio == NULL) { |
a6255b7f DQ |
919 | abd_t *lwb_abd = abd_get_from_buf(lwb->lwb_buf, |
920 | BP_GET_LSIZE(&lwb->lwb_blk)); | |
920dd524 ED |
921 | if (!lwb->lwb_fastwrite) { |
922 | metaslab_fastwrite_mark(zilog->zl_spa, &lwb->lwb_blk); | |
923 | lwb->lwb_fastwrite = 1; | |
924 | } | |
34dc7c2f | 925 | lwb->lwb_zio = zio_rewrite(zilog->zl_root_zio, zilog->zl_spa, |
a6255b7f | 926 | 0, &lwb->lwb_blk, lwb_abd, BP_GET_LSIZE(&lwb->lwb_blk), |
e8b96c60 | 927 | zil_lwb_write_done, lwb, ZIO_PRIORITY_SYNC_WRITE, |
920dd524 ED |
928 | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | |
929 | ZIO_FLAG_FASTWRITE, &zb); | |
34dc7c2f | 930 | } |
920dd524 | 931 | mutex_exit(&zilog->zl_lock); |
34dc7c2f BB |
932 | } |
933 | ||
428870ff BB |
934 | /* |
935 | * Define a limited set of intent log block sizes. | |
d3cc8b15 | 936 | * |
428870ff BB |
937 | * These must be a multiple of 4KB. Note only the amount used (again |
938 | * aligned to 4KB) actually gets written. However, we can't always just | |
f1512ee6 | 939 | * allocate SPA_OLD_MAXBLOCKSIZE as the slog space could be exhausted. |
428870ff BB |
940 | */ |
941 | uint64_t zil_block_buckets[] = { | |
942 | 4096, /* non TX_WRITE */ | |
943 | 8192+4096, /* data base */ | |
944 | 32*1024 + 4096, /* NFS writes */ | |
945 | UINT64_MAX | |
946 | }; | |
947 | ||
948 | /* | |
5d7a86d1 ED |
949 | * Use the slog as long as the current commit size is less than the |
950 | * limit or the total list size is less than 2X the limit. Limit | |
951 | * checking is disabled by setting zil_slog_limit to UINT64_MAX. | |
428870ff | 952 | */ |
ee191e80 | 953 | unsigned long zil_slog_limit = 1024 * 1024; |
5d7a86d1 ED |
954 | #define USE_SLOG(zilog) (((zilog)->zl_cur_used < zil_slog_limit) || \ |
955 | ((zilog)->zl_itx_list_sz < (zil_slog_limit << 1))) | |
428870ff | 956 | |
34dc7c2f BB |
957 | /* |
958 | * Start a log block write and advance to the next log block. | |
959 | * Calls are serialized. | |
960 | */ | |
961 | static lwb_t * | |
962 | zil_lwb_write_start(zilog_t *zilog, lwb_t *lwb) | |
963 | { | |
428870ff BB |
964 | lwb_t *nlwb = NULL; |
965 | zil_chain_t *zilc; | |
34dc7c2f | 966 | spa_t *spa = zilog->zl_spa; |
428870ff BB |
967 | blkptr_t *bp; |
968 | dmu_tx_t *tx; | |
34dc7c2f | 969 | uint64_t txg; |
428870ff BB |
970 | uint64_t zil_blksz, wsz; |
971 | int i, error; | |
b6ad9671 | 972 | boolean_t use_slog; |
428870ff BB |
973 | |
974 | if (BP_GET_CHECKSUM(&lwb->lwb_blk) == ZIO_CHECKSUM_ZILOG2) { | |
975 | zilc = (zil_chain_t *)lwb->lwb_buf; | |
976 | bp = &zilc->zc_next_blk; | |
977 | } else { | |
978 | zilc = (zil_chain_t *)(lwb->lwb_buf + lwb->lwb_sz); | |
979 | bp = &zilc->zc_next_blk; | |
980 | } | |
34dc7c2f | 981 | |
428870ff | 982 | ASSERT(lwb->lwb_nused <= lwb->lwb_sz); |
34dc7c2f BB |
983 | |
984 | /* | |
985 | * Allocate the next block and save its address in this block | |
986 | * before writing it in order to establish the log chain. | |
987 | * Note that if the allocation of nlwb synced before we wrote | |
988 | * the block that points at it (lwb), we'd leak it if we crashed. | |
428870ff BB |
989 | * Therefore, we don't do dmu_tx_commit() until zil_lwb_write_done(). |
990 | * We dirty the dataset to ensure that zil_sync() will be called | |
991 | * to clean up in the event of allocation failure or I/O failure. | |
34dc7c2f | 992 | */ |
428870ff BB |
993 | tx = dmu_tx_create(zilog->zl_os); |
994 | VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0); | |
995 | dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); | |
996 | txg = dmu_tx_get_txg(tx); | |
997 | ||
998 | lwb->lwb_tx = tx; | |
34dc7c2f BB |
999 | |
1000 | /* | |
428870ff BB |
1001 | * Log blocks are pre-allocated. Here we select the size of the next |
1002 | * block, based on size used in the last block. | |
1003 | * - first find the smallest bucket that will fit the block from a | |
1004 | * limited set of block sizes. This is because it's faster to write | |
1005 | * blocks allocated from the same metaslab as they are adjacent or | |
1006 | * close. | |
1007 | * - next find the maximum from the new suggested size and an array of | |
1008 | * previous sizes. This lessens a picket fence effect of wrongly | |
1009 | * guesssing the size if we have a stream of say 2k, 64k, 2k, 64k | |
1010 | * requests. | |
1011 | * | |
1012 | * Note we only write what is used, but we can't just allocate | |
1013 | * the maximum block size because we can exhaust the available | |
1014 | * pool log space. | |
34dc7c2f | 1015 | */ |
428870ff BB |
1016 | zil_blksz = zilog->zl_cur_used + sizeof (zil_chain_t); |
1017 | for (i = 0; zil_blksz > zil_block_buckets[i]; i++) | |
1018 | continue; | |
1019 | zil_blksz = zil_block_buckets[i]; | |
1020 | if (zil_blksz == UINT64_MAX) | |
f1512ee6 | 1021 | zil_blksz = SPA_OLD_MAXBLOCKSIZE; |
428870ff BB |
1022 | zilog->zl_prev_blks[zilog->zl_prev_rotor] = zil_blksz; |
1023 | for (i = 0; i < ZIL_PREV_BLKS; i++) | |
1024 | zil_blksz = MAX(zil_blksz, zilog->zl_prev_blks[i]); | |
1025 | zilog->zl_prev_rotor = (zilog->zl_prev_rotor + 1) & (ZIL_PREV_BLKS - 1); | |
34dc7c2f BB |
1026 | |
1027 | BP_ZERO(bp); | |
b6ad9671 | 1028 | use_slog = USE_SLOG(zilog); |
13fe0198 MA |
1029 | error = zio_alloc_zil(spa, txg, bp, zil_blksz, |
1030 | USE_SLOG(zilog)); | |
d1d7e268 | 1031 | if (use_slog) { |
b6ad9671 ED |
1032 | ZIL_STAT_BUMP(zil_itx_metaslab_slog_count); |
1033 | ZIL_STAT_INCR(zil_itx_metaslab_slog_bytes, lwb->lwb_nused); | |
d1d7e268 | 1034 | } else { |
b6ad9671 ED |
1035 | ZIL_STAT_BUMP(zil_itx_metaslab_normal_count); |
1036 | ZIL_STAT_INCR(zil_itx_metaslab_normal_bytes, lwb->lwb_nused); | |
1037 | } | |
13fe0198 | 1038 | if (error == 0) { |
428870ff BB |
1039 | ASSERT3U(bp->blk_birth, ==, txg); |
1040 | bp->blk_cksum = lwb->lwb_blk.blk_cksum; | |
1041 | bp->blk_cksum.zc_word[ZIL_ZC_SEQ]++; | |
34dc7c2f BB |
1042 | |
1043 | /* | |
428870ff | 1044 | * Allocate a new log write buffer (lwb). |
34dc7c2f | 1045 | */ |
920dd524 | 1046 | nlwb = zil_alloc_lwb(zilog, bp, txg, TRUE); |
34dc7c2f | 1047 | |
428870ff BB |
1048 | /* Record the block for later vdev flushing */ |
1049 | zil_add_block(zilog, &lwb->lwb_blk); | |
34dc7c2f BB |
1050 | } |
1051 | ||
428870ff BB |
1052 | if (BP_GET_CHECKSUM(&lwb->lwb_blk) == ZIO_CHECKSUM_ZILOG2) { |
1053 | /* For Slim ZIL only write what is used. */ | |
1054 | wsz = P2ROUNDUP_TYPED(lwb->lwb_nused, ZIL_MIN_BLKSZ, uint64_t); | |
1055 | ASSERT3U(wsz, <=, lwb->lwb_sz); | |
1056 | zio_shrink(lwb->lwb_zio, wsz); | |
34dc7c2f | 1057 | |
428870ff BB |
1058 | } else { |
1059 | wsz = lwb->lwb_sz; | |
1060 | } | |
34dc7c2f | 1061 | |
428870ff BB |
1062 | zilc->zc_pad = 0; |
1063 | zilc->zc_nused = lwb->lwb_nused; | |
1064 | zilc->zc_eck.zec_cksum = lwb->lwb_blk.blk_cksum; | |
34dc7c2f BB |
1065 | |
1066 | /* | |
428870ff | 1067 | * clear unused data for security |
34dc7c2f | 1068 | */ |
428870ff | 1069 | bzero(lwb->lwb_buf + lwb->lwb_nused, wsz - lwb->lwb_nused); |
34dc7c2f | 1070 | |
428870ff | 1071 | zio_nowait(lwb->lwb_zio); /* Kick off the write for the old log block */ |
34dc7c2f BB |
1072 | |
1073 | /* | |
428870ff BB |
1074 | * If there was an allocation failure then nlwb will be null which |
1075 | * forces a txg_wait_synced(). | |
34dc7c2f | 1076 | */ |
34dc7c2f BB |
1077 | return (nlwb); |
1078 | } | |
1079 | ||
1080 | static lwb_t * | |
1081 | zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb) | |
1082 | { | |
1083 | lr_t *lrc = &itx->itx_lr; /* common log record */ | |
428870ff BB |
1084 | lr_write_t *lrw = (lr_write_t *)lrc; |
1085 | char *lr_buf; | |
34dc7c2f BB |
1086 | uint64_t txg = lrc->lrc_txg; |
1087 | uint64_t reclen = lrc->lrc_reclen; | |
428870ff | 1088 | uint64_t dlen = 0; |
34dc7c2f BB |
1089 | |
1090 | if (lwb == NULL) | |
1091 | return (NULL); | |
428870ff | 1092 | |
34dc7c2f | 1093 | ASSERT(lwb->lwb_buf != NULL); |
29809a6c MA |
1094 | ASSERT(zilog_is_dirty(zilog) || |
1095 | spa_freeze_txg(zilog->zl_spa) != UINT64_MAX); | |
34dc7c2f BB |
1096 | |
1097 | if (lrc->lrc_txtype == TX_WRITE && itx->itx_wr_state == WR_NEED_COPY) | |
1098 | dlen = P2ROUNDUP_TYPED( | |
428870ff | 1099 | lrw->lr_length, sizeof (uint64_t), uint64_t); |
34dc7c2f BB |
1100 | |
1101 | zilog->zl_cur_used += (reclen + dlen); | |
1102 | ||
1103 | zil_lwb_write_init(zilog, lwb); | |
1104 | ||
1105 | /* | |
1106 | * If this record won't fit in the current log block, start a new one. | |
1107 | */ | |
428870ff | 1108 | if (lwb->lwb_nused + reclen + dlen > lwb->lwb_sz) { |
34dc7c2f BB |
1109 | lwb = zil_lwb_write_start(zilog, lwb); |
1110 | if (lwb == NULL) | |
1111 | return (NULL); | |
1112 | zil_lwb_write_init(zilog, lwb); | |
428870ff BB |
1113 | ASSERT(LWB_EMPTY(lwb)); |
1114 | if (lwb->lwb_nused + reclen + dlen > lwb->lwb_sz) { | |
34dc7c2f BB |
1115 | txg_wait_synced(zilog->zl_dmu_pool, txg); |
1116 | return (lwb); | |
1117 | } | |
1118 | } | |
1119 | ||
428870ff BB |
1120 | lr_buf = lwb->lwb_buf + lwb->lwb_nused; |
1121 | bcopy(lrc, lr_buf, reclen); | |
1122 | lrc = (lr_t *)lr_buf; | |
1123 | lrw = (lr_write_t *)lrc; | |
34dc7c2f | 1124 | |
b6ad9671 ED |
1125 | ZIL_STAT_BUMP(zil_itx_count); |
1126 | ||
34dc7c2f BB |
1127 | /* |
1128 | * If it's a write, fetch the data or get its blkptr as appropriate. | |
1129 | */ | |
1130 | if (lrc->lrc_txtype == TX_WRITE) { | |
1131 | if (txg > spa_freeze_txg(zilog->zl_spa)) | |
1132 | txg_wait_synced(zilog->zl_dmu_pool, txg); | |
b6ad9671 ED |
1133 | if (itx->itx_wr_state == WR_COPIED) { |
1134 | ZIL_STAT_BUMP(zil_itx_copied_count); | |
1135 | ZIL_STAT_INCR(zil_itx_copied_bytes, lrw->lr_length); | |
1136 | } else { | |
34dc7c2f BB |
1137 | char *dbuf; |
1138 | int error; | |
1139 | ||
34dc7c2f BB |
1140 | if (dlen) { |
1141 | ASSERT(itx->itx_wr_state == WR_NEED_COPY); | |
428870ff BB |
1142 | dbuf = lr_buf + reclen; |
1143 | lrw->lr_common.lrc_reclen += dlen; | |
b6ad9671 | 1144 | ZIL_STAT_BUMP(zil_itx_needcopy_count); |
d1d7e268 MK |
1145 | ZIL_STAT_INCR(zil_itx_needcopy_bytes, |
1146 | lrw->lr_length); | |
34dc7c2f BB |
1147 | } else { |
1148 | ASSERT(itx->itx_wr_state == WR_INDIRECT); | |
1149 | dbuf = NULL; | |
b6ad9671 | 1150 | ZIL_STAT_BUMP(zil_itx_indirect_count); |
d1d7e268 MK |
1151 | ZIL_STAT_INCR(zil_itx_indirect_bytes, |
1152 | lrw->lr_length); | |
34dc7c2f BB |
1153 | } |
1154 | error = zilog->zl_get_data( | |
428870ff | 1155 | itx->itx_private, lrw, dbuf, lwb->lwb_zio); |
45d1cae3 BB |
1156 | if (error == EIO) { |
1157 | txg_wait_synced(zilog->zl_dmu_pool, txg); | |
1158 | return (lwb); | |
1159 | } | |
13fe0198 | 1160 | if (error != 0) { |
34dc7c2f BB |
1161 | ASSERT(error == ENOENT || error == EEXIST || |
1162 | error == EALREADY); | |
1163 | return (lwb); | |
1164 | } | |
1165 | } | |
1166 | } | |
1167 | ||
428870ff BB |
1168 | /* |
1169 | * We're actually making an entry, so update lrc_seq to be the | |
1170 | * log record sequence number. Note that this is generally not | |
1171 | * equal to the itx sequence number because not all transactions | |
1172 | * are synchronous, and sometimes spa_sync() gets there first. | |
1173 | */ | |
1174 | lrc->lrc_seq = ++zilog->zl_lr_seq; /* we are single threaded */ | |
34dc7c2f BB |
1175 | lwb->lwb_nused += reclen + dlen; |
1176 | lwb->lwb_max_txg = MAX(lwb->lwb_max_txg, txg); | |
428870ff | 1177 | ASSERT3U(lwb->lwb_nused, <=, lwb->lwb_sz); |
c99c9001 | 1178 | ASSERT0(P2PHASE(lwb->lwb_nused, sizeof (uint64_t))); |
34dc7c2f BB |
1179 | |
1180 | return (lwb); | |
1181 | } | |
1182 | ||
1183 | itx_t * | |
1184 | zil_itx_create(uint64_t txtype, size_t lrsize) | |
1185 | { | |
1186 | itx_t *itx; | |
1187 | ||
1188 | lrsize = P2ROUNDUP_TYPED(lrsize, sizeof (uint64_t), size_t); | |
1189 | ||
19ea3d25 | 1190 | itx = zio_data_buf_alloc(offsetof(itx_t, itx_lr) + lrsize); |
34dc7c2f BB |
1191 | itx->itx_lr.lrc_txtype = txtype; |
1192 | itx->itx_lr.lrc_reclen = lrsize; | |
1193 | itx->itx_sod = lrsize; /* if write & WR_NEED_COPY will be increased */ | |
1194 | itx->itx_lr.lrc_seq = 0; /* defensive */ | |
572e2857 | 1195 | itx->itx_sync = B_TRUE; /* default is synchronous */ |
119a394a ED |
1196 | itx->itx_callback = NULL; |
1197 | itx->itx_callback_data = NULL; | |
34dc7c2f BB |
1198 | |
1199 | return (itx); | |
1200 | } | |
1201 | ||
428870ff BB |
1202 | void |
1203 | zil_itx_destroy(itx_t *itx) | |
1204 | { | |
19ea3d25 | 1205 | zio_data_buf_free(itx, offsetof(itx_t, itx_lr)+itx->itx_lr.lrc_reclen); |
428870ff BB |
1206 | } |
1207 | ||
572e2857 BB |
1208 | /* |
1209 | * Free up the sync and async itxs. The itxs_t has already been detached | |
1210 | * so no locks are needed. | |
1211 | */ | |
1212 | static void | |
1213 | zil_itxg_clean(itxs_t *itxs) | |
34dc7c2f | 1214 | { |
572e2857 BB |
1215 | itx_t *itx; |
1216 | list_t *list; | |
1217 | avl_tree_t *t; | |
1218 | void *cookie; | |
1219 | itx_async_node_t *ian; | |
1220 | ||
1221 | list = &itxs->i_sync_list; | |
1222 | while ((itx = list_head(list)) != NULL) { | |
119a394a ED |
1223 | if (itx->itx_callback != NULL) |
1224 | itx->itx_callback(itx->itx_callback_data); | |
572e2857 | 1225 | list_remove(list, itx); |
19ea3d25 | 1226 | zil_itx_destroy(itx); |
572e2857 | 1227 | } |
34dc7c2f | 1228 | |
572e2857 BB |
1229 | cookie = NULL; |
1230 | t = &itxs->i_async_tree; | |
1231 | while ((ian = avl_destroy_nodes(t, &cookie)) != NULL) { | |
1232 | list = &ian->ia_list; | |
1233 | while ((itx = list_head(list)) != NULL) { | |
119a394a ED |
1234 | if (itx->itx_callback != NULL) |
1235 | itx->itx_callback(itx->itx_callback_data); | |
572e2857 | 1236 | list_remove(list, itx); |
19ea3d25 | 1237 | zil_itx_destroy(itx); |
572e2857 BB |
1238 | } |
1239 | list_destroy(list); | |
1240 | kmem_free(ian, sizeof (itx_async_node_t)); | |
1241 | } | |
1242 | avl_destroy(t); | |
34dc7c2f | 1243 | |
572e2857 BB |
1244 | kmem_free(itxs, sizeof (itxs_t)); |
1245 | } | |
34dc7c2f | 1246 | |
572e2857 BB |
1247 | static int |
1248 | zil_aitx_compare(const void *x1, const void *x2) | |
1249 | { | |
1250 | const uint64_t o1 = ((itx_async_node_t *)x1)->ia_foid; | |
1251 | const uint64_t o2 = ((itx_async_node_t *)x2)->ia_foid; | |
1252 | ||
ee36c709 | 1253 | return (AVL_CMP(o1, o2)); |
34dc7c2f BB |
1254 | } |
1255 | ||
1256 | /* | |
572e2857 | 1257 | * Remove all async itx with the given oid. |
34dc7c2f BB |
1258 | */ |
1259 | static void | |
572e2857 | 1260 | zil_remove_async(zilog_t *zilog, uint64_t oid) |
34dc7c2f | 1261 | { |
572e2857 BB |
1262 | uint64_t otxg, txg; |
1263 | itx_async_node_t *ian; | |
1264 | avl_tree_t *t; | |
1265 | avl_index_t where; | |
34dc7c2f BB |
1266 | list_t clean_list; |
1267 | itx_t *itx; | |
1268 | ||
572e2857 | 1269 | ASSERT(oid != 0); |
34dc7c2f BB |
1270 | list_create(&clean_list, sizeof (itx_t), offsetof(itx_t, itx_node)); |
1271 | ||
572e2857 BB |
1272 | if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */ |
1273 | otxg = ZILTEST_TXG; | |
1274 | else | |
1275 | otxg = spa_last_synced_txg(zilog->zl_spa) + 1; | |
34dc7c2f | 1276 | |
572e2857 BB |
1277 | for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) { |
1278 | itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK]; | |
1279 | ||
1280 | mutex_enter(&itxg->itxg_lock); | |
1281 | if (itxg->itxg_txg != txg) { | |
1282 | mutex_exit(&itxg->itxg_lock); | |
1283 | continue; | |
1284 | } | |
34dc7c2f | 1285 | |
572e2857 BB |
1286 | /* |
1287 | * Locate the object node and append its list. | |
1288 | */ | |
1289 | t = &itxg->itxg_itxs->i_async_tree; | |
1290 | ian = avl_find(t, &oid, &where); | |
1291 | if (ian != NULL) | |
1292 | list_move_tail(&clean_list, &ian->ia_list); | |
1293 | mutex_exit(&itxg->itxg_lock); | |
1294 | } | |
34dc7c2f | 1295 | while ((itx = list_head(&clean_list)) != NULL) { |
119a394a ED |
1296 | if (itx->itx_callback != NULL) |
1297 | itx->itx_callback(itx->itx_callback_data); | |
34dc7c2f | 1298 | list_remove(&clean_list, itx); |
19ea3d25 | 1299 | zil_itx_destroy(itx); |
34dc7c2f BB |
1300 | } |
1301 | list_destroy(&clean_list); | |
1302 | } | |
1303 | ||
572e2857 BB |
1304 | void |
1305 | zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx) | |
1306 | { | |
1307 | uint64_t txg; | |
1308 | itxg_t *itxg; | |
1309 | itxs_t *itxs, *clean = NULL; | |
1310 | ||
1311 | /* | |
1312 | * Object ids can be re-instantiated in the next txg so | |
1313 | * remove any async transactions to avoid future leaks. | |
1314 | * This can happen if a fsync occurs on the re-instantiated | |
1315 | * object for a WR_INDIRECT or WR_NEED_COPY write, which gets | |
1316 | * the new file data and flushes a write record for the old object. | |
1317 | */ | |
1318 | if ((itx->itx_lr.lrc_txtype & ~TX_CI) == TX_REMOVE) | |
1319 | zil_remove_async(zilog, itx->itx_oid); | |
1320 | ||
1321 | /* | |
1322 | * Ensure the data of a renamed file is committed before the rename. | |
1323 | */ | |
1324 | if ((itx->itx_lr.lrc_txtype & ~TX_CI) == TX_RENAME) | |
1325 | zil_async_to_sync(zilog, itx->itx_oid); | |
1326 | ||
29809a6c | 1327 | if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) |
572e2857 BB |
1328 | txg = ZILTEST_TXG; |
1329 | else | |
1330 | txg = dmu_tx_get_txg(tx); | |
1331 | ||
1332 | itxg = &zilog->zl_itxg[txg & TXG_MASK]; | |
1333 | mutex_enter(&itxg->itxg_lock); | |
1334 | itxs = itxg->itxg_itxs; | |
1335 | if (itxg->itxg_txg != txg) { | |
1336 | if (itxs != NULL) { | |
1337 | /* | |
1338 | * The zil_clean callback hasn't got around to cleaning | |
1339 | * this itxg. Save the itxs for release below. | |
1340 | * This should be rare. | |
1341 | */ | |
1342 | atomic_add_64(&zilog->zl_itx_list_sz, -itxg->itxg_sod); | |
1343 | itxg->itxg_sod = 0; | |
1344 | clean = itxg->itxg_itxs; | |
1345 | } | |
1346 | ASSERT(itxg->itxg_sod == 0); | |
1347 | itxg->itxg_txg = txg; | |
d1d7e268 | 1348 | itxs = itxg->itxg_itxs = kmem_zalloc(sizeof (itxs_t), |
79c76d5b | 1349 | KM_SLEEP); |
572e2857 BB |
1350 | |
1351 | list_create(&itxs->i_sync_list, sizeof (itx_t), | |
1352 | offsetof(itx_t, itx_node)); | |
1353 | avl_create(&itxs->i_async_tree, zil_aitx_compare, | |
1354 | sizeof (itx_async_node_t), | |
1355 | offsetof(itx_async_node_t, ia_node)); | |
1356 | } | |
1357 | if (itx->itx_sync) { | |
1358 | list_insert_tail(&itxs->i_sync_list, itx); | |
1359 | atomic_add_64(&zilog->zl_itx_list_sz, itx->itx_sod); | |
1360 | itxg->itxg_sod += itx->itx_sod; | |
1361 | } else { | |
1362 | avl_tree_t *t = &itxs->i_async_tree; | |
50c957f7 NB |
1363 | uint64_t foid = |
1364 | LR_FOID_GET_OBJ(((lr_ooo_t *)&itx->itx_lr)->lr_foid); | |
572e2857 BB |
1365 | itx_async_node_t *ian; |
1366 | avl_index_t where; | |
1367 | ||
1368 | ian = avl_find(t, &foid, &where); | |
1369 | if (ian == NULL) { | |
d1d7e268 | 1370 | ian = kmem_alloc(sizeof (itx_async_node_t), |
79c76d5b | 1371 | KM_SLEEP); |
572e2857 BB |
1372 | list_create(&ian->ia_list, sizeof (itx_t), |
1373 | offsetof(itx_t, itx_node)); | |
1374 | ian->ia_foid = foid; | |
1375 | avl_insert(t, ian, where); | |
1376 | } | |
1377 | list_insert_tail(&ian->ia_list, itx); | |
1378 | } | |
1379 | ||
1380 | itx->itx_lr.lrc_txg = dmu_tx_get_txg(tx); | |
29809a6c | 1381 | zilog_dirty(zilog, txg); |
572e2857 BB |
1382 | mutex_exit(&itxg->itxg_lock); |
1383 | ||
1384 | /* Release the old itxs now we've dropped the lock */ | |
1385 | if (clean != NULL) | |
1386 | zil_itxg_clean(clean); | |
1387 | } | |
1388 | ||
34dc7c2f BB |
1389 | /* |
1390 | * If there are any in-memory intent log transactions which have now been | |
29809a6c MA |
1391 | * synced then start up a taskq to free them. We should only do this after we |
1392 | * have written out the uberblocks (i.e. txg has been comitted) so that | |
1393 | * don't inadvertently clean out in-memory log records that would be required | |
1394 | * by zil_commit(). | |
34dc7c2f BB |
1395 | */ |
1396 | void | |
572e2857 | 1397 | zil_clean(zilog_t *zilog, uint64_t synced_txg) |
34dc7c2f | 1398 | { |
572e2857 BB |
1399 | itxg_t *itxg = &zilog->zl_itxg[synced_txg & TXG_MASK]; |
1400 | itxs_t *clean_me; | |
34dc7c2f | 1401 | |
572e2857 BB |
1402 | mutex_enter(&itxg->itxg_lock); |
1403 | if (itxg->itxg_itxs == NULL || itxg->itxg_txg == ZILTEST_TXG) { | |
1404 | mutex_exit(&itxg->itxg_lock); | |
1405 | return; | |
1406 | } | |
1407 | ASSERT3U(itxg->itxg_txg, <=, synced_txg); | |
1408 | ASSERT(itxg->itxg_txg != 0); | |
1409 | ASSERT(zilog->zl_clean_taskq != NULL); | |
1410 | atomic_add_64(&zilog->zl_itx_list_sz, -itxg->itxg_sod); | |
1411 | itxg->itxg_sod = 0; | |
1412 | clean_me = itxg->itxg_itxs; | |
1413 | itxg->itxg_itxs = NULL; | |
1414 | itxg->itxg_txg = 0; | |
1415 | mutex_exit(&itxg->itxg_lock); | |
1416 | /* | |
1417 | * Preferably start a task queue to free up the old itxs but | |
1418 | * if taskq_dispatch can't allocate resources to do that then | |
1419 | * free it in-line. This should be rare. Note, using TQ_SLEEP | |
1420 | * created a bad performance problem. | |
1421 | */ | |
1422 | if (taskq_dispatch(zilog->zl_clean_taskq, | |
b8864a23 | 1423 | (void (*)(void *))zil_itxg_clean, clean_me, TQ_NOSLEEP) == 0) |
572e2857 BB |
1424 | zil_itxg_clean(clean_me); |
1425 | } | |
1426 | ||
1427 | /* | |
1428 | * Get the list of itxs to commit into zl_itx_commit_list. | |
1429 | */ | |
1430 | static void | |
1431 | zil_get_commit_list(zilog_t *zilog) | |
1432 | { | |
1433 | uint64_t otxg, txg; | |
1434 | list_t *commit_list = &zilog->zl_itx_commit_list; | |
1435 | uint64_t push_sod = 0; | |
1436 | ||
1437 | if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */ | |
1438 | otxg = ZILTEST_TXG; | |
1439 | else | |
1440 | otxg = spa_last_synced_txg(zilog->zl_spa) + 1; | |
1441 | ||
1442 | for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) { | |
1443 | itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK]; | |
1444 | ||
1445 | mutex_enter(&itxg->itxg_lock); | |
1446 | if (itxg->itxg_txg != txg) { | |
1447 | mutex_exit(&itxg->itxg_lock); | |
1448 | continue; | |
1449 | } | |
1450 | ||
1451 | list_move_tail(commit_list, &itxg->itxg_itxs->i_sync_list); | |
1452 | push_sod += itxg->itxg_sod; | |
1453 | itxg->itxg_sod = 0; | |
1454 | ||
1455 | mutex_exit(&itxg->itxg_lock); | |
1456 | } | |
1457 | atomic_add_64(&zilog->zl_itx_list_sz, -push_sod); | |
1458 | } | |
1459 | ||
1460 | /* | |
1461 | * Move the async itxs for a specified object to commit into sync lists. | |
1462 | */ | |
1463 | static void | |
1464 | zil_async_to_sync(zilog_t *zilog, uint64_t foid) | |
1465 | { | |
1466 | uint64_t otxg, txg; | |
1467 | itx_async_node_t *ian; | |
1468 | avl_tree_t *t; | |
1469 | avl_index_t where; | |
1470 | ||
1471 | if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */ | |
1472 | otxg = ZILTEST_TXG; | |
1473 | else | |
1474 | otxg = spa_last_synced_txg(zilog->zl_spa) + 1; | |
1475 | ||
1476 | for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) { | |
1477 | itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK]; | |
1478 | ||
1479 | mutex_enter(&itxg->itxg_lock); | |
1480 | if (itxg->itxg_txg != txg) { | |
1481 | mutex_exit(&itxg->itxg_lock); | |
1482 | continue; | |
1483 | } | |
1484 | ||
1485 | /* | |
1486 | * If a foid is specified then find that node and append its | |
1487 | * list. Otherwise walk the tree appending all the lists | |
1488 | * to the sync list. We add to the end rather than the | |
1489 | * beginning to ensure the create has happened. | |
1490 | */ | |
1491 | t = &itxg->itxg_itxs->i_async_tree; | |
1492 | if (foid != 0) { | |
1493 | ian = avl_find(t, &foid, &where); | |
1494 | if (ian != NULL) { | |
1495 | list_move_tail(&itxg->itxg_itxs->i_sync_list, | |
1496 | &ian->ia_list); | |
1497 | } | |
1498 | } else { | |
1499 | void *cookie = NULL; | |
1500 | ||
1501 | while ((ian = avl_destroy_nodes(t, &cookie)) != NULL) { | |
1502 | list_move_tail(&itxg->itxg_itxs->i_sync_list, | |
1503 | &ian->ia_list); | |
1504 | list_destroy(&ian->ia_list); | |
1505 | kmem_free(ian, sizeof (itx_async_node_t)); | |
1506 | } | |
1507 | } | |
1508 | mutex_exit(&itxg->itxg_lock); | |
34dc7c2f | 1509 | } |
34dc7c2f BB |
1510 | } |
1511 | ||
b128c09f | 1512 | static void |
572e2857 | 1513 | zil_commit_writer(zilog_t *zilog) |
34dc7c2f BB |
1514 | { |
1515 | uint64_t txg; | |
572e2857 | 1516 | itx_t *itx; |
34dc7c2f | 1517 | lwb_t *lwb; |
572e2857 | 1518 | spa_t *spa = zilog->zl_spa; |
428870ff | 1519 | int error = 0; |
34dc7c2f | 1520 | |
b128c09f | 1521 | ASSERT(zilog->zl_root_zio == NULL); |
572e2857 BB |
1522 | |
1523 | mutex_exit(&zilog->zl_lock); | |
1524 | ||
1525 | zil_get_commit_list(zilog); | |
1526 | ||
1527 | /* | |
1528 | * Return if there's nothing to commit before we dirty the fs by | |
1529 | * calling zil_create(). | |
1530 | */ | |
1531 | if (list_head(&zilog->zl_itx_commit_list) == NULL) { | |
1532 | mutex_enter(&zilog->zl_lock); | |
1533 | return; | |
1534 | } | |
34dc7c2f BB |
1535 | |
1536 | if (zilog->zl_suspend) { | |
1537 | lwb = NULL; | |
1538 | } else { | |
1539 | lwb = list_tail(&zilog->zl_lwb_list); | |
572e2857 | 1540 | if (lwb == NULL) |
428870ff | 1541 | lwb = zil_create(zilog); |
34dc7c2f BB |
1542 | } |
1543 | ||
34dc7c2f | 1544 | DTRACE_PROBE1(zil__cw1, zilog_t *, zilog); |
119a394a | 1545 | for (itx = list_head(&zilog->zl_itx_commit_list); itx != NULL; |
d1d7e268 | 1546 | itx = list_next(&zilog->zl_itx_commit_list, itx)) { |
34dc7c2f BB |
1547 | txg = itx->itx_lr.lrc_txg; |
1548 | ASSERT(txg); | |
1549 | ||
572e2857 | 1550 | if (txg > spa_last_synced_txg(spa) || txg > spa_freeze_txg(spa)) |
34dc7c2f | 1551 | lwb = zil_lwb_commit(zilog, itx, lwb); |
34dc7c2f BB |
1552 | } |
1553 | DTRACE_PROBE1(zil__cw2, zilog_t *, zilog); | |
34dc7c2f BB |
1554 | |
1555 | /* write the last block out */ | |
1556 | if (lwb != NULL && lwb->lwb_zio != NULL) | |
1557 | lwb = zil_lwb_write_start(zilog, lwb); | |
1558 | ||
34dc7c2f BB |
1559 | zilog->zl_cur_used = 0; |
1560 | ||
1561 | /* | |
1562 | * Wait if necessary for the log blocks to be on stable storage. | |
1563 | */ | |
1564 | if (zilog->zl_root_zio) { | |
428870ff | 1565 | error = zio_wait(zilog->zl_root_zio); |
b128c09f | 1566 | zilog->zl_root_zio = NULL; |
34dc7c2f BB |
1567 | zil_flush_vdevs(zilog); |
1568 | } | |
1569 | ||
428870ff | 1570 | if (error || lwb == NULL) |
34dc7c2f | 1571 | txg_wait_synced(zilog->zl_dmu_pool, 0); |
34dc7c2f | 1572 | |
119a394a ED |
1573 | while ((itx = list_head(&zilog->zl_itx_commit_list))) { |
1574 | txg = itx->itx_lr.lrc_txg; | |
1575 | ASSERT(txg); | |
1576 | ||
1577 | if (itx->itx_callback != NULL) | |
1578 | itx->itx_callback(itx->itx_callback_data); | |
1579 | list_remove(&zilog->zl_itx_commit_list, itx); | |
19ea3d25 | 1580 | zil_itx_destroy(itx); |
119a394a ED |
1581 | } |
1582 | ||
34dc7c2f | 1583 | mutex_enter(&zilog->zl_lock); |
428870ff BB |
1584 | |
1585 | /* | |
1586 | * Remember the highest committed log sequence number for ztest. | |
1587 | * We only update this value when all the log writes succeeded, | |
1588 | * because ztest wants to ASSERT that it got the whole log chain. | |
1589 | */ | |
1590 | if (error == 0 && lwb != NULL) | |
1591 | zilog->zl_commit_lr_seq = zilog->zl_lr_seq; | |
34dc7c2f BB |
1592 | } |
1593 | ||
1594 | /* | |
572e2857 | 1595 | * Commit zfs transactions to stable storage. |
34dc7c2f | 1596 | * If foid is 0 push out all transactions, otherwise push only those |
572e2857 BB |
1597 | * for that object or might reference that object. |
1598 | * | |
1599 | * itxs are committed in batches. In a heavily stressed zil there will be | |
1600 | * a commit writer thread who is writing out a bunch of itxs to the log | |
1601 | * for a set of committing threads (cthreads) in the same batch as the writer. | |
1602 | * Those cthreads are all waiting on the same cv for that batch. | |
1603 | * | |
1604 | * There will also be a different and growing batch of threads that are | |
1605 | * waiting to commit (qthreads). When the committing batch completes | |
1606 | * a transition occurs such that the cthreads exit and the qthreads become | |
1607 | * cthreads. One of the new cthreads becomes the writer thread for the | |
1608 | * batch. Any new threads arriving become new qthreads. | |
1609 | * | |
1610 | * Only 2 condition variables are needed and there's no transition | |
1611 | * between the two cvs needed. They just flip-flop between qthreads | |
1612 | * and cthreads. | |
1613 | * | |
1614 | * Using this scheme we can efficiently wakeup up only those threads | |
1615 | * that have been committed. | |
34dc7c2f BB |
1616 | */ |
1617 | void | |
572e2857 | 1618 | zil_commit(zilog_t *zilog, uint64_t foid) |
34dc7c2f | 1619 | { |
572e2857 | 1620 | uint64_t mybatch; |
34dc7c2f | 1621 | |
572e2857 BB |
1622 | if (zilog->zl_sync == ZFS_SYNC_DISABLED) |
1623 | return; | |
34dc7c2f | 1624 | |
b6ad9671 ED |
1625 | ZIL_STAT_BUMP(zil_commit_count); |
1626 | ||
572e2857 BB |
1627 | /* move the async itxs for the foid to the sync queues */ |
1628 | zil_async_to_sync(zilog, foid); | |
34dc7c2f | 1629 | |
572e2857 BB |
1630 | mutex_enter(&zilog->zl_lock); |
1631 | mybatch = zilog->zl_next_batch; | |
34dc7c2f | 1632 | while (zilog->zl_writer) { |
572e2857 BB |
1633 | cv_wait(&zilog->zl_cv_batch[mybatch & 1], &zilog->zl_lock); |
1634 | if (mybatch <= zilog->zl_com_batch) { | |
34dc7c2f BB |
1635 | mutex_exit(&zilog->zl_lock); |
1636 | return; | |
1637 | } | |
1638 | } | |
428870ff | 1639 | |
572e2857 BB |
1640 | zilog->zl_next_batch++; |
1641 | zilog->zl_writer = B_TRUE; | |
b6ad9671 | 1642 | ZIL_STAT_BUMP(zil_commit_writer_count); |
572e2857 BB |
1643 | zil_commit_writer(zilog); |
1644 | zilog->zl_com_batch = mybatch; | |
1645 | zilog->zl_writer = B_FALSE; | |
428870ff | 1646 | |
572e2857 BB |
1647 | /* wake up one thread to become the next writer */ |
1648 | cv_signal(&zilog->zl_cv_batch[(mybatch+1) & 1]); | |
428870ff | 1649 | |
572e2857 BB |
1650 | /* wake up all threads waiting for this batch to be committed */ |
1651 | cv_broadcast(&zilog->zl_cv_batch[mybatch & 1]); | |
8c0712fd BB |
1652 | |
1653 | mutex_exit(&zilog->zl_lock); | |
428870ff BB |
1654 | } |
1655 | ||
34dc7c2f BB |
1656 | /* |
1657 | * Called in syncing context to free committed log blocks and update log header. | |
1658 | */ | |
1659 | void | |
1660 | zil_sync(zilog_t *zilog, dmu_tx_t *tx) | |
1661 | { | |
1662 | zil_header_t *zh = zil_header_in_syncing_context(zilog); | |
1663 | uint64_t txg = dmu_tx_get_txg(tx); | |
1664 | spa_t *spa = zilog->zl_spa; | |
428870ff | 1665 | uint64_t *replayed_seq = &zilog->zl_replayed_seq[txg & TXG_MASK]; |
34dc7c2f BB |
1666 | lwb_t *lwb; |
1667 | ||
9babb374 BB |
1668 | /* |
1669 | * We don't zero out zl_destroy_txg, so make sure we don't try | |
1670 | * to destroy it twice. | |
1671 | */ | |
1672 | if (spa_sync_pass(spa) != 1) | |
1673 | return; | |
1674 | ||
34dc7c2f BB |
1675 | mutex_enter(&zilog->zl_lock); |
1676 | ||
1677 | ASSERT(zilog->zl_stop_sync == 0); | |
1678 | ||
428870ff BB |
1679 | if (*replayed_seq != 0) { |
1680 | ASSERT(zh->zh_replay_seq < *replayed_seq); | |
1681 | zh->zh_replay_seq = *replayed_seq; | |
1682 | *replayed_seq = 0; | |
1683 | } | |
34dc7c2f BB |
1684 | |
1685 | if (zilog->zl_destroy_txg == txg) { | |
1686 | blkptr_t blk = zh->zh_log; | |
1687 | ||
1688 | ASSERT(list_head(&zilog->zl_lwb_list) == NULL); | |
34dc7c2f BB |
1689 | |
1690 | bzero(zh, sizeof (zil_header_t)); | |
fb5f0bc8 | 1691 | bzero(zilog->zl_replayed_seq, sizeof (zilog->zl_replayed_seq)); |
34dc7c2f BB |
1692 | |
1693 | if (zilog->zl_keep_first) { | |
1694 | /* | |
1695 | * If this block was part of log chain that couldn't | |
1696 | * be claimed because a device was missing during | |
1697 | * zil_claim(), but that device later returns, | |
1698 | * then this block could erroneously appear valid. | |
1699 | * To guard against this, assign a new GUID to the new | |
1700 | * log chain so it doesn't matter what blk points to. | |
1701 | */ | |
1702 | zil_init_log_chain(zilog, &blk); | |
1703 | zh->zh_log = blk; | |
1704 | } | |
1705 | } | |
1706 | ||
9babb374 | 1707 | while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) { |
34dc7c2f BB |
1708 | zh->zh_log = lwb->lwb_blk; |
1709 | if (lwb->lwb_buf != NULL || lwb->lwb_max_txg > txg) | |
1710 | break; | |
920dd524 ED |
1711 | |
1712 | ASSERT(lwb->lwb_zio == NULL); | |
1713 | ||
34dc7c2f | 1714 | list_remove(&zilog->zl_lwb_list, lwb); |
428870ff | 1715 | zio_free_zil(spa, txg, &lwb->lwb_blk); |
34dc7c2f BB |
1716 | kmem_cache_free(zil_lwb_cache, lwb); |
1717 | ||
1718 | /* | |
1719 | * If we don't have anything left in the lwb list then | |
1720 | * we've had an allocation failure and we need to zero | |
1721 | * out the zil_header blkptr so that we don't end | |
1722 | * up freeing the same block twice. | |
1723 | */ | |
1724 | if (list_head(&zilog->zl_lwb_list) == NULL) | |
1725 | BP_ZERO(&zh->zh_log); | |
1726 | } | |
920dd524 ED |
1727 | |
1728 | /* | |
1729 | * Remove fastwrite on any blocks that have been pre-allocated for | |
1730 | * the next commit. This prevents fastwrite counter pollution by | |
1731 | * unused, long-lived LWBs. | |
1732 | */ | |
1733 | for (; lwb != NULL; lwb = list_next(&zilog->zl_lwb_list, lwb)) { | |
1734 | if (lwb->lwb_fastwrite && !lwb->lwb_zio) { | |
1735 | metaslab_fastwrite_unmark(zilog->zl_spa, &lwb->lwb_blk); | |
1736 | lwb->lwb_fastwrite = 0; | |
1737 | } | |
1738 | } | |
1739 | ||
34dc7c2f BB |
1740 | mutex_exit(&zilog->zl_lock); |
1741 | } | |
1742 | ||
1743 | void | |
1744 | zil_init(void) | |
1745 | { | |
1746 | zil_lwb_cache = kmem_cache_create("zil_lwb_cache", | |
1747 | sizeof (struct lwb), 0, NULL, NULL, NULL, NULL, NULL, 0); | |
b6ad9671 ED |
1748 | |
1749 | zil_ksp = kstat_create("zfs", 0, "zil", "misc", | |
d1d7e268 | 1750 | KSTAT_TYPE_NAMED, sizeof (zil_stats) / sizeof (kstat_named_t), |
b6ad9671 ED |
1751 | KSTAT_FLAG_VIRTUAL); |
1752 | ||
1753 | if (zil_ksp != NULL) { | |
1754 | zil_ksp->ks_data = &zil_stats; | |
1755 | kstat_install(zil_ksp); | |
1756 | } | |
34dc7c2f BB |
1757 | } |
1758 | ||
1759 | void | |
1760 | zil_fini(void) | |
1761 | { | |
1762 | kmem_cache_destroy(zil_lwb_cache); | |
b6ad9671 ED |
1763 | |
1764 | if (zil_ksp != NULL) { | |
1765 | kstat_delete(zil_ksp); | |
1766 | zil_ksp = NULL; | |
1767 | } | |
34dc7c2f BB |
1768 | } |
1769 | ||
428870ff BB |
1770 | void |
1771 | zil_set_sync(zilog_t *zilog, uint64_t sync) | |
1772 | { | |
1773 | zilog->zl_sync = sync; | |
1774 | } | |
1775 | ||
1776 | void | |
1777 | zil_set_logbias(zilog_t *zilog, uint64_t logbias) | |
1778 | { | |
1779 | zilog->zl_logbias = logbias; | |
1780 | } | |
1781 | ||
34dc7c2f BB |
1782 | zilog_t * |
1783 | zil_alloc(objset_t *os, zil_header_t *zh_phys) | |
1784 | { | |
1785 | zilog_t *zilog; | |
d6320ddb | 1786 | int i; |
34dc7c2f | 1787 | |
79c76d5b | 1788 | zilog = kmem_zalloc(sizeof (zilog_t), KM_SLEEP); |
34dc7c2f BB |
1789 | |
1790 | zilog->zl_header = zh_phys; | |
1791 | zilog->zl_os = os; | |
1792 | zilog->zl_spa = dmu_objset_spa(os); | |
1793 | zilog->zl_dmu_pool = dmu_objset_pool(os); | |
1794 | zilog->zl_destroy_txg = TXG_INITIAL - 1; | |
428870ff BB |
1795 | zilog->zl_logbias = dmu_objset_logbias(os); |
1796 | zilog->zl_sync = dmu_objset_syncprop(os); | |
572e2857 | 1797 | zilog->zl_next_batch = 1; |
34dc7c2f BB |
1798 | |
1799 | mutex_init(&zilog->zl_lock, NULL, MUTEX_DEFAULT, NULL); | |
1800 | ||
d6320ddb | 1801 | for (i = 0; i < TXG_SIZE; i++) { |
572e2857 BB |
1802 | mutex_init(&zilog->zl_itxg[i].itxg_lock, NULL, |
1803 | MUTEX_DEFAULT, NULL); | |
1804 | } | |
34dc7c2f BB |
1805 | |
1806 | list_create(&zilog->zl_lwb_list, sizeof (lwb_t), | |
1807 | offsetof(lwb_t, lwb_node)); | |
1808 | ||
572e2857 BB |
1809 | list_create(&zilog->zl_itx_commit_list, sizeof (itx_t), |
1810 | offsetof(itx_t, itx_node)); | |
1811 | ||
34dc7c2f BB |
1812 | mutex_init(&zilog->zl_vdev_lock, NULL, MUTEX_DEFAULT, NULL); |
1813 | ||
1814 | avl_create(&zilog->zl_vdev_tree, zil_vdev_compare, | |
1815 | sizeof (zil_vdev_node_t), offsetof(zil_vdev_node_t, zv_node)); | |
1816 | ||
1817 | cv_init(&zilog->zl_cv_writer, NULL, CV_DEFAULT, NULL); | |
1818 | cv_init(&zilog->zl_cv_suspend, NULL, CV_DEFAULT, NULL); | |
572e2857 BB |
1819 | cv_init(&zilog->zl_cv_batch[0], NULL, CV_DEFAULT, NULL); |
1820 | cv_init(&zilog->zl_cv_batch[1], NULL, CV_DEFAULT, NULL); | |
34dc7c2f BB |
1821 | |
1822 | return (zilog); | |
1823 | } | |
1824 | ||
1825 | void | |
1826 | zil_free(zilog_t *zilog) | |
1827 | { | |
d6320ddb | 1828 | int i; |
34dc7c2f BB |
1829 | |
1830 | zilog->zl_stop_sync = 1; | |
1831 | ||
13fe0198 MA |
1832 | ASSERT0(zilog->zl_suspend); |
1833 | ASSERT0(zilog->zl_suspending); | |
1834 | ||
3e31d2b0 | 1835 | ASSERT(list_is_empty(&zilog->zl_lwb_list)); |
34dc7c2f BB |
1836 | list_destroy(&zilog->zl_lwb_list); |
1837 | ||
1838 | avl_destroy(&zilog->zl_vdev_tree); | |
1839 | mutex_destroy(&zilog->zl_vdev_lock); | |
1840 | ||
572e2857 BB |
1841 | ASSERT(list_is_empty(&zilog->zl_itx_commit_list)); |
1842 | list_destroy(&zilog->zl_itx_commit_list); | |
1843 | ||
d6320ddb | 1844 | for (i = 0; i < TXG_SIZE; i++) { |
572e2857 BB |
1845 | /* |
1846 | * It's possible for an itx to be generated that doesn't dirty | |
1847 | * a txg (e.g. ztest TX_TRUNCATE). So there's no zil_clean() | |
1848 | * callback to remove the entry. We remove those here. | |
1849 | * | |
1850 | * Also free up the ziltest itxs. | |
1851 | */ | |
1852 | if (zilog->zl_itxg[i].itxg_itxs) | |
1853 | zil_itxg_clean(zilog->zl_itxg[i].itxg_itxs); | |
1854 | mutex_destroy(&zilog->zl_itxg[i].itxg_lock); | |
1855 | } | |
1856 | ||
34dc7c2f BB |
1857 | mutex_destroy(&zilog->zl_lock); |
1858 | ||
1859 | cv_destroy(&zilog->zl_cv_writer); | |
1860 | cv_destroy(&zilog->zl_cv_suspend); | |
572e2857 BB |
1861 | cv_destroy(&zilog->zl_cv_batch[0]); |
1862 | cv_destroy(&zilog->zl_cv_batch[1]); | |
34dc7c2f BB |
1863 | |
1864 | kmem_free(zilog, sizeof (zilog_t)); | |
1865 | } | |
1866 | ||
34dc7c2f BB |
1867 | /* |
1868 | * Open an intent log. | |
1869 | */ | |
1870 | zilog_t * | |
1871 | zil_open(objset_t *os, zil_get_data_t *get_data) | |
1872 | { | |
1873 | zilog_t *zilog = dmu_objset_zil(os); | |
1874 | ||
3e31d2b0 ES |
1875 | ASSERT(zilog->zl_clean_taskq == NULL); |
1876 | ASSERT(zilog->zl_get_data == NULL); | |
1877 | ASSERT(list_is_empty(&zilog->zl_lwb_list)); | |
1878 | ||
34dc7c2f | 1879 | zilog->zl_get_data = get_data; |
1229323d | 1880 | zilog->zl_clean_taskq = taskq_create("zil_clean", 1, defclsyspri, |
34dc7c2f BB |
1881 | 2, 2, TASKQ_PREPOPULATE); |
1882 | ||
1883 | return (zilog); | |
1884 | } | |
1885 | ||
1886 | /* | |
1887 | * Close an intent log. | |
1888 | */ | |
1889 | void | |
1890 | zil_close(zilog_t *zilog) | |
1891 | { | |
3e31d2b0 | 1892 | lwb_t *lwb; |
572e2857 BB |
1893 | uint64_t txg = 0; |
1894 | ||
1895 | zil_commit(zilog, 0); /* commit all itx */ | |
1896 | ||
34dc7c2f | 1897 | /* |
572e2857 BB |
1898 | * The lwb_max_txg for the stubby lwb will reflect the last activity |
1899 | * for the zil. After a txg_wait_synced() on the txg we know all the | |
1900 | * callbacks have occurred that may clean the zil. Only then can we | |
1901 | * destroy the zl_clean_taskq. | |
34dc7c2f | 1902 | */ |
572e2857 | 1903 | mutex_enter(&zilog->zl_lock); |
3e31d2b0 ES |
1904 | lwb = list_tail(&zilog->zl_lwb_list); |
1905 | if (lwb != NULL) | |
1906 | txg = lwb->lwb_max_txg; | |
572e2857 BB |
1907 | mutex_exit(&zilog->zl_lock); |
1908 | if (txg) | |
34dc7c2f | 1909 | txg_wait_synced(zilog->zl_dmu_pool, txg); |
50c957f7 NB |
1910 | if (txg < spa_freeze_txg(zilog->zl_spa)) |
1911 | ASSERT(!zilog_is_dirty(zilog)); | |
34dc7c2f BB |
1912 | |
1913 | taskq_destroy(zilog->zl_clean_taskq); | |
1914 | zilog->zl_clean_taskq = NULL; | |
1915 | zilog->zl_get_data = NULL; | |
3e31d2b0 ES |
1916 | |
1917 | /* | |
1918 | * We should have only one LWB left on the list; remove it now. | |
1919 | */ | |
1920 | mutex_enter(&zilog->zl_lock); | |
1921 | lwb = list_head(&zilog->zl_lwb_list); | |
1922 | if (lwb != NULL) { | |
1923 | ASSERT(lwb == list_tail(&zilog->zl_lwb_list)); | |
920dd524 ED |
1924 | ASSERT(lwb->lwb_zio == NULL); |
1925 | if (lwb->lwb_fastwrite) | |
1926 | metaslab_fastwrite_unmark(zilog->zl_spa, &lwb->lwb_blk); | |
3e31d2b0 ES |
1927 | list_remove(&zilog->zl_lwb_list, lwb); |
1928 | zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); | |
1929 | kmem_cache_free(zil_lwb_cache, lwb); | |
1930 | } | |
1931 | mutex_exit(&zilog->zl_lock); | |
34dc7c2f BB |
1932 | } |
1933 | ||
13fe0198 MA |
1934 | static char *suspend_tag = "zil suspending"; |
1935 | ||
34dc7c2f BB |
1936 | /* |
1937 | * Suspend an intent log. While in suspended mode, we still honor | |
1938 | * synchronous semantics, but we rely on txg_wait_synced() to do it. | |
13fe0198 MA |
1939 | * On old version pools, we suspend the log briefly when taking a |
1940 | * snapshot so that it will have an empty intent log. | |
1941 | * | |
1942 | * Long holds are not really intended to be used the way we do here -- | |
1943 | * held for such a short time. A concurrent caller of dsl_dataset_long_held() | |
1944 | * could fail. Therefore we take pains to only put a long hold if it is | |
1945 | * actually necessary. Fortunately, it will only be necessary if the | |
1946 | * objset is currently mounted (or the ZVOL equivalent). In that case it | |
1947 | * will already have a long hold, so we are not really making things any worse. | |
1948 | * | |
1949 | * Ideally, we would locate the existing long-holder (i.e. the zfsvfs_t or | |
1950 | * zvol_state_t), and use their mechanism to prevent their hold from being | |
1951 | * dropped (e.g. VFS_HOLD()). However, that would be even more pain for | |
1952 | * very little gain. | |
1953 | * | |
1954 | * if cookiep == NULL, this does both the suspend & resume. | |
1955 | * Otherwise, it returns with the dataset "long held", and the cookie | |
1956 | * should be passed into zil_resume(). | |
34dc7c2f BB |
1957 | */ |
1958 | int | |
13fe0198 | 1959 | zil_suspend(const char *osname, void **cookiep) |
34dc7c2f | 1960 | { |
13fe0198 MA |
1961 | objset_t *os; |
1962 | zilog_t *zilog; | |
1963 | const zil_header_t *zh; | |
1964 | int error; | |
1965 | ||
1966 | error = dmu_objset_hold(osname, suspend_tag, &os); | |
1967 | if (error != 0) | |
1968 | return (error); | |
1969 | zilog = dmu_objset_zil(os); | |
34dc7c2f BB |
1970 | |
1971 | mutex_enter(&zilog->zl_lock); | |
13fe0198 MA |
1972 | zh = zilog->zl_header; |
1973 | ||
9babb374 | 1974 | if (zh->zh_flags & ZIL_REPLAY_NEEDED) { /* unplayed log */ |
34dc7c2f | 1975 | mutex_exit(&zilog->zl_lock); |
13fe0198 | 1976 | dmu_objset_rele(os, suspend_tag); |
2e528b49 | 1977 | return (SET_ERROR(EBUSY)); |
34dc7c2f | 1978 | } |
13fe0198 MA |
1979 | |
1980 | /* | |
1981 | * Don't put a long hold in the cases where we can avoid it. This | |
1982 | * is when there is no cookie so we are doing a suspend & resume | |
1983 | * (i.e. called from zil_vdev_offline()), and there's nothing to do | |
1984 | * for the suspend because it's already suspended, or there's no ZIL. | |
1985 | */ | |
1986 | if (cookiep == NULL && !zilog->zl_suspending && | |
1987 | (zilog->zl_suspend > 0 || BP_IS_HOLE(&zh->zh_log))) { | |
1988 | mutex_exit(&zilog->zl_lock); | |
1989 | dmu_objset_rele(os, suspend_tag); | |
1990 | return (0); | |
1991 | } | |
1992 | ||
1993 | dsl_dataset_long_hold(dmu_objset_ds(os), suspend_tag); | |
1994 | dsl_pool_rele(dmu_objset_pool(os), suspend_tag); | |
1995 | ||
1996 | zilog->zl_suspend++; | |
1997 | ||
1998 | if (zilog->zl_suspend > 1) { | |
34dc7c2f | 1999 | /* |
13fe0198 | 2000 | * Someone else is already suspending it. |
34dc7c2f BB |
2001 | * Just wait for them to finish. |
2002 | */ | |
13fe0198 | 2003 | |
34dc7c2f BB |
2004 | while (zilog->zl_suspending) |
2005 | cv_wait(&zilog->zl_cv_suspend, &zilog->zl_lock); | |
34dc7c2f | 2006 | mutex_exit(&zilog->zl_lock); |
13fe0198 MA |
2007 | |
2008 | if (cookiep == NULL) | |
2009 | zil_resume(os); | |
2010 | else | |
2011 | *cookiep = os; | |
2012 | return (0); | |
2013 | } | |
2014 | ||
2015 | /* | |
2016 | * If there is no pointer to an on-disk block, this ZIL must not | |
2017 | * be active (e.g. filesystem not mounted), so there's nothing | |
2018 | * to clean up. | |
2019 | */ | |
2020 | if (BP_IS_HOLE(&zh->zh_log)) { | |
2021 | ASSERT(cookiep != NULL); /* fast path already handled */ | |
2022 | ||
2023 | *cookiep = os; | |
2024 | mutex_exit(&zilog->zl_lock); | |
34dc7c2f BB |
2025 | return (0); |
2026 | } | |
13fe0198 | 2027 | |
34dc7c2f BB |
2028 | zilog->zl_suspending = B_TRUE; |
2029 | mutex_exit(&zilog->zl_lock); | |
2030 | ||
572e2857 | 2031 | zil_commit(zilog, 0); |
34dc7c2f BB |
2032 | |
2033 | zil_destroy(zilog, B_FALSE); | |
2034 | ||
2035 | mutex_enter(&zilog->zl_lock); | |
2036 | zilog->zl_suspending = B_FALSE; | |
2037 | cv_broadcast(&zilog->zl_cv_suspend); | |
2038 | mutex_exit(&zilog->zl_lock); | |
2039 | ||
13fe0198 MA |
2040 | if (cookiep == NULL) |
2041 | zil_resume(os); | |
2042 | else | |
2043 | *cookiep = os; | |
34dc7c2f BB |
2044 | return (0); |
2045 | } | |
2046 | ||
2047 | void | |
13fe0198 | 2048 | zil_resume(void *cookie) |
34dc7c2f | 2049 | { |
13fe0198 MA |
2050 | objset_t *os = cookie; |
2051 | zilog_t *zilog = dmu_objset_zil(os); | |
2052 | ||
34dc7c2f BB |
2053 | mutex_enter(&zilog->zl_lock); |
2054 | ASSERT(zilog->zl_suspend != 0); | |
2055 | zilog->zl_suspend--; | |
2056 | mutex_exit(&zilog->zl_lock); | |
13fe0198 MA |
2057 | dsl_dataset_long_rele(dmu_objset_ds(os), suspend_tag); |
2058 | dsl_dataset_rele(dmu_objset_ds(os), suspend_tag); | |
34dc7c2f BB |
2059 | } |
2060 | ||
2061 | typedef struct zil_replay_arg { | |
b01615d5 | 2062 | zil_replay_func_t *zr_replay; |
34dc7c2f | 2063 | void *zr_arg; |
34dc7c2f | 2064 | boolean_t zr_byteswap; |
428870ff | 2065 | char *zr_lr; |
34dc7c2f BB |
2066 | } zil_replay_arg_t; |
2067 | ||
428870ff BB |
2068 | static int |
2069 | zil_replay_error(zilog_t *zilog, lr_t *lr, int error) | |
2070 | { | |
eca7b760 | 2071 | char name[ZFS_MAX_DATASET_NAME_LEN]; |
428870ff BB |
2072 | |
2073 | zilog->zl_replaying_seq--; /* didn't actually replay this one */ | |
2074 | ||
2075 | dmu_objset_name(zilog->zl_os, name); | |
2076 | ||
2077 | cmn_err(CE_WARN, "ZFS replay transaction error %d, " | |
2078 | "dataset %s, seq 0x%llx, txtype %llu %s\n", error, name, | |
2079 | (u_longlong_t)lr->lrc_seq, | |
2080 | (u_longlong_t)(lr->lrc_txtype & ~TX_CI), | |
2081 | (lr->lrc_txtype & TX_CI) ? "CI" : ""); | |
2082 | ||
2083 | return (error); | |
2084 | } | |
2085 | ||
2086 | static int | |
34dc7c2f BB |
2087 | zil_replay_log_record(zilog_t *zilog, lr_t *lr, void *zra, uint64_t claim_txg) |
2088 | { | |
2089 | zil_replay_arg_t *zr = zra; | |
2090 | const zil_header_t *zh = zilog->zl_header; | |
2091 | uint64_t reclen = lr->lrc_reclen; | |
2092 | uint64_t txtype = lr->lrc_txtype; | |
428870ff | 2093 | int error = 0; |
34dc7c2f | 2094 | |
428870ff | 2095 | zilog->zl_replaying_seq = lr->lrc_seq; |
34dc7c2f BB |
2096 | |
2097 | if (lr->lrc_seq <= zh->zh_replay_seq) /* already replayed */ | |
428870ff BB |
2098 | return (0); |
2099 | ||
2100 | if (lr->lrc_txg < claim_txg) /* already committed */ | |
2101 | return (0); | |
34dc7c2f BB |
2102 | |
2103 | /* Strip case-insensitive bit, still present in log record */ | |
2104 | txtype &= ~TX_CI; | |
2105 | ||
428870ff BB |
2106 | if (txtype == 0 || txtype >= TX_MAX_TYPE) |
2107 | return (zil_replay_error(zilog, lr, EINVAL)); | |
2108 | ||
2109 | /* | |
2110 | * If this record type can be logged out of order, the object | |
2111 | * (lr_foid) may no longer exist. That's legitimate, not an error. | |
2112 | */ | |
2113 | if (TX_OOO(txtype)) { | |
2114 | error = dmu_object_info(zilog->zl_os, | |
50c957f7 | 2115 | LR_FOID_GET_OBJ(((lr_ooo_t *)lr)->lr_foid), NULL); |
428870ff BB |
2116 | if (error == ENOENT || error == EEXIST) |
2117 | return (0); | |
fb5f0bc8 BB |
2118 | } |
2119 | ||
34dc7c2f BB |
2120 | /* |
2121 | * Make a copy of the data so we can revise and extend it. | |
2122 | */ | |
428870ff BB |
2123 | bcopy(lr, zr->zr_lr, reclen); |
2124 | ||
2125 | /* | |
2126 | * If this is a TX_WRITE with a blkptr, suck in the data. | |
2127 | */ | |
2128 | if (txtype == TX_WRITE && reclen == sizeof (lr_write_t)) { | |
2129 | error = zil_read_log_data(zilog, (lr_write_t *)lr, | |
2130 | zr->zr_lr + reclen); | |
13fe0198 | 2131 | if (error != 0) |
428870ff BB |
2132 | return (zil_replay_error(zilog, lr, error)); |
2133 | } | |
34dc7c2f BB |
2134 | |
2135 | /* | |
2136 | * The log block containing this lr may have been byteswapped | |
2137 | * so that we can easily examine common fields like lrc_txtype. | |
428870ff | 2138 | * However, the log is a mix of different record types, and only the |
34dc7c2f BB |
2139 | * replay vectors know how to byteswap their records. Therefore, if |
2140 | * the lr was byteswapped, undo it before invoking the replay vector. | |
2141 | */ | |
2142 | if (zr->zr_byteswap) | |
428870ff | 2143 | byteswap_uint64_array(zr->zr_lr, reclen); |
34dc7c2f BB |
2144 | |
2145 | /* | |
2146 | * We must now do two things atomically: replay this log record, | |
fb5f0bc8 BB |
2147 | * and update the log header sequence number to reflect the fact that |
2148 | * we did so. At the end of each replay function the sequence number | |
2149 | * is updated if we are in replay mode. | |
34dc7c2f | 2150 | */ |
428870ff | 2151 | error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lr, zr->zr_byteswap); |
13fe0198 | 2152 | if (error != 0) { |
34dc7c2f BB |
2153 | /* |
2154 | * The DMU's dnode layer doesn't see removes until the txg | |
2155 | * commits, so a subsequent claim can spuriously fail with | |
fb5f0bc8 | 2156 | * EEXIST. So if we receive any error we try syncing out |
428870ff BB |
2157 | * any removes then retry the transaction. Note that we |
2158 | * specify B_FALSE for byteswap now, so we don't do it twice. | |
34dc7c2f | 2159 | */ |
428870ff BB |
2160 | txg_wait_synced(spa_get_dsl(zilog->zl_spa), 0); |
2161 | error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lr, B_FALSE); | |
13fe0198 | 2162 | if (error != 0) |
428870ff | 2163 | return (zil_replay_error(zilog, lr, error)); |
34dc7c2f | 2164 | } |
428870ff | 2165 | return (0); |
34dc7c2f BB |
2166 | } |
2167 | ||
2168 | /* ARGSUSED */ | |
428870ff | 2169 | static int |
34dc7c2f BB |
2170 | zil_incr_blks(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg) |
2171 | { | |
2172 | zilog->zl_replay_blks++; | |
428870ff BB |
2173 | |
2174 | return (0); | |
34dc7c2f BB |
2175 | } |
2176 | ||
2177 | /* | |
2178 | * If this dataset has a non-empty intent log, replay it and destroy it. | |
2179 | */ | |
2180 | void | |
b01615d5 | 2181 | zil_replay(objset_t *os, void *arg, zil_replay_func_t replay_func[TX_MAX_TYPE]) |
34dc7c2f BB |
2182 | { |
2183 | zilog_t *zilog = dmu_objset_zil(os); | |
2184 | const zil_header_t *zh = zilog->zl_header; | |
2185 | zil_replay_arg_t zr; | |
2186 | ||
9babb374 | 2187 | if ((zh->zh_flags & ZIL_REPLAY_NEEDED) == 0) { |
34dc7c2f BB |
2188 | zil_destroy(zilog, B_TRUE); |
2189 | return; | |
2190 | } | |
2191 | ||
34dc7c2f BB |
2192 | zr.zr_replay = replay_func; |
2193 | zr.zr_arg = arg; | |
34dc7c2f | 2194 | zr.zr_byteswap = BP_SHOULD_BYTESWAP(&zh->zh_log); |
79c76d5b | 2195 | zr.zr_lr = vmem_alloc(2 * SPA_MAXBLOCKSIZE, KM_SLEEP); |
34dc7c2f BB |
2196 | |
2197 | /* | |
2198 | * Wait for in-progress removes to sync before starting replay. | |
2199 | */ | |
2200 | txg_wait_synced(zilog->zl_dmu_pool, 0); | |
2201 | ||
fb5f0bc8 | 2202 | zilog->zl_replay = B_TRUE; |
428870ff | 2203 | zilog->zl_replay_time = ddi_get_lbolt(); |
34dc7c2f BB |
2204 | ASSERT(zilog->zl_replay_blks == 0); |
2205 | (void) zil_parse(zilog, zil_incr_blks, zil_replay_log_record, &zr, | |
2206 | zh->zh_claim_txg); | |
00b46022 | 2207 | vmem_free(zr.zr_lr, 2 * SPA_MAXBLOCKSIZE); |
34dc7c2f BB |
2208 | |
2209 | zil_destroy(zilog, B_FALSE); | |
2210 | txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg); | |
fb5f0bc8 | 2211 | zilog->zl_replay = B_FALSE; |
34dc7c2f BB |
2212 | } |
2213 | ||
428870ff BB |
2214 | boolean_t |
2215 | zil_replaying(zilog_t *zilog, dmu_tx_t *tx) | |
34dc7c2f | 2216 | { |
428870ff BB |
2217 | if (zilog->zl_sync == ZFS_SYNC_DISABLED) |
2218 | return (B_TRUE); | |
34dc7c2f | 2219 | |
428870ff BB |
2220 | if (zilog->zl_replay) { |
2221 | dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); | |
2222 | zilog->zl_replayed_seq[dmu_tx_get_txg(tx) & TXG_MASK] = | |
2223 | zilog->zl_replaying_seq; | |
2224 | return (B_TRUE); | |
34dc7c2f BB |
2225 | } |
2226 | ||
428870ff | 2227 | return (B_FALSE); |
34dc7c2f | 2228 | } |
9babb374 BB |
2229 | |
2230 | /* ARGSUSED */ | |
2231 | int | |
428870ff | 2232 | zil_vdev_offline(const char *osname, void *arg) |
9babb374 | 2233 | { |
9babb374 BB |
2234 | int error; |
2235 | ||
13fe0198 MA |
2236 | error = zil_suspend(osname, NULL); |
2237 | if (error != 0) | |
2e528b49 | 2238 | return (SET_ERROR(EEXIST)); |
13fe0198 | 2239 | return (0); |
9babb374 | 2240 | } |
c409e464 BB |
2241 | |
2242 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
0f699108 AZ |
2243 | EXPORT_SYMBOL(zil_alloc); |
2244 | EXPORT_SYMBOL(zil_free); | |
2245 | EXPORT_SYMBOL(zil_open); | |
2246 | EXPORT_SYMBOL(zil_close); | |
2247 | EXPORT_SYMBOL(zil_replay); | |
2248 | EXPORT_SYMBOL(zil_replaying); | |
2249 | EXPORT_SYMBOL(zil_destroy); | |
2250 | EXPORT_SYMBOL(zil_destroy_sync); | |
2251 | EXPORT_SYMBOL(zil_itx_create); | |
2252 | EXPORT_SYMBOL(zil_itx_destroy); | |
2253 | EXPORT_SYMBOL(zil_itx_assign); | |
2254 | EXPORT_SYMBOL(zil_commit); | |
2255 | EXPORT_SYMBOL(zil_vdev_offline); | |
2256 | EXPORT_SYMBOL(zil_claim); | |
2257 | EXPORT_SYMBOL(zil_check_log_chain); | |
2258 | EXPORT_SYMBOL(zil_sync); | |
2259 | EXPORT_SYMBOL(zil_clean); | |
2260 | EXPORT_SYMBOL(zil_suspend); | |
2261 | EXPORT_SYMBOL(zil_resume); | |
2262 | EXPORT_SYMBOL(zil_add_block); | |
2263 | EXPORT_SYMBOL(zil_bp_tree_add); | |
2264 | EXPORT_SYMBOL(zil_set_sync); | |
2265 | EXPORT_SYMBOL(zil_set_logbias); | |
2266 | ||
c409e464 BB |
2267 | module_param(zil_replay_disable, int, 0644); |
2268 | MODULE_PARM_DESC(zil_replay_disable, "Disable intent logging replay"); | |
2269 | ||
2270 | module_param(zfs_nocacheflush, int, 0644); | |
2271 | MODULE_PARM_DESC(zfs_nocacheflush, "Disable cache flushes"); | |
ee191e80 ED |
2272 | |
2273 | module_param(zil_slog_limit, ulong, 0644); | |
2274 | MODULE_PARM_DESC(zil_slog_limit, "Max commit bytes to separate log device"); | |
c409e464 | 2275 | #endif |