1fa9e4066Sahrens /* 2fa9e4066Sahrens * CDDL HEADER START 3fa9e4066Sahrens * 4fa9e4066Sahrens * The contents of this file are subject to the terms of the 5fe9cf88cSperrin * Common Development and Distribution License (the "License"). 6fe9cf88cSperrin * You may not use this file except in compliance with the License. 7fa9e4066Sahrens * 8fa9e4066Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9fa9e4066Sahrens * or http://www.opensolaris.org/os/licensing. 10fa9e4066Sahrens * See the License for the specific language governing permissions 11fa9e4066Sahrens * and limitations under the License. 12fa9e4066Sahrens * 13fa9e4066Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14fa9e4066Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15fa9e4066Sahrens * If applicable, add the following below this CDDL HEADER, with the 16fa9e4066Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17fa9e4066Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18fa9e4066Sahrens * 19fa9e4066Sahrens * CDDL HEADER END 20fa9e4066Sahrens */ 21fa9e4066Sahrens /* 22ab69d62fSMatthew Ahrens * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 23fa9e4066Sahrens * Use is subject to license terms. 24fa9e4066Sahrens */ 25fa9e4066Sahrens 26fa9e4066Sahrens #include <sys/zfs_context.h> 27fa9e4066Sahrens #include <sys/spa.h> 28fa9e4066Sahrens #include <sys/dmu.h> 29fa9e4066Sahrens #include <sys/zap.h> 30fa9e4066Sahrens #include <sys/arc.h> 31fa9e4066Sahrens #include <sys/stat.h> 32fa9e4066Sahrens #include <sys/resource.h> 33fa9e4066Sahrens #include <sys/zil.h> 34fa9e4066Sahrens #include <sys/zil_impl.h> 35fa9e4066Sahrens #include <sys/dsl_dataset.h> 36fa9e4066Sahrens #include <sys/vdev.h> 37d63d470bSgw #include <sys/dmu_tx.h> 38fa9e4066Sahrens 39fa9e4066Sahrens /* 40fa9e4066Sahrens * The zfs intent log (ZIL) saves transaction records of system calls 41fa9e4066Sahrens * that change the file system in memory with enough information 42fa9e4066Sahrens * to be able to replay them. These are stored in memory until 43fa9e4066Sahrens * either the DMU transaction group (txg) commits them to the stable pool 44fa9e4066Sahrens * and they can be discarded, or they are flushed to the stable log 45fa9e4066Sahrens * (also in the pool) due to a fsync, O_DSYNC or other synchronous 46fa9e4066Sahrens * requirement. In the event of a panic or power fail then those log 47fa9e4066Sahrens * records (transactions) are replayed. 48fa9e4066Sahrens * 49fa9e4066Sahrens * There is one ZIL per file system. Its on-disk (pool) format consists 50fa9e4066Sahrens * of 3 parts: 51fa9e4066Sahrens * 52fa9e4066Sahrens * - ZIL header 53fa9e4066Sahrens * - ZIL blocks 54fa9e4066Sahrens * - ZIL records 55fa9e4066Sahrens * 56fa9e4066Sahrens * A log record holds a system call transaction. Log blocks can 57fa9e4066Sahrens * hold many log records and the blocks are chained together. 58fa9e4066Sahrens * Each ZIL block contains a block pointer (blkptr_t) to the next 59fa9e4066Sahrens * ZIL block in the chain. The ZIL header points to the first 60fa9e4066Sahrens * block in the chain. Note there is not a fixed place in the pool 61fa9e4066Sahrens * to hold blocks. They are dynamically allocated and freed as 62fa9e4066Sahrens * needed from the blocks available. Figure X shows the ZIL structure: 63fa9e4066Sahrens */ 64fa9e4066Sahrens 65fa9e4066Sahrens /* 66416e0cd8Sek * This global ZIL switch affects all pools 67fa9e4066Sahrens */ 68fa9e4066Sahrens int zil_disable = 0; /* disable intent logging */ 69416e0cd8Sek 70416e0cd8Sek /* 71416e0cd8Sek * Tunable parameter for debugging or performance analysis. Setting 72416e0cd8Sek * zfs_nocacheflush will cause corruption on power loss if a volatile 73416e0cd8Sek * out-of-order write cache is enabled. 74416e0cd8Sek */ 75416e0cd8Sek boolean_t zfs_nocacheflush = B_FALSE; 76fa9e4066Sahrens 77fa9e4066Sahrens static kmem_cache_t *zil_lwb_cache; 78fa9e4066Sahrens 79fa9e4066Sahrens static int 80fa9e4066Sahrens zil_dva_compare(const void *x1, const void *x2) 81fa9e4066Sahrens { 82fa9e4066Sahrens const dva_t *dva1 = x1; 83fa9e4066Sahrens const dva_t *dva2 = x2; 84fa9e4066Sahrens 85fa9e4066Sahrens if (DVA_GET_VDEV(dva1) < DVA_GET_VDEV(dva2)) 86fa9e4066Sahrens return (-1); 87fa9e4066Sahrens if (DVA_GET_VDEV(dva1) > DVA_GET_VDEV(dva2)) 88fa9e4066Sahrens return (1); 89fa9e4066Sahrens 90fa9e4066Sahrens if (DVA_GET_OFFSET(dva1) < DVA_GET_OFFSET(dva2)) 91fa9e4066Sahrens return (-1); 92fa9e4066Sahrens if (DVA_GET_OFFSET(dva1) > DVA_GET_OFFSET(dva2)) 93fa9e4066Sahrens return (1); 94fa9e4066Sahrens 95fa9e4066Sahrens return (0); 96fa9e4066Sahrens } 97fa9e4066Sahrens 98fa9e4066Sahrens static void 99fa9e4066Sahrens zil_dva_tree_init(avl_tree_t *t) 100fa9e4066Sahrens { 101fa9e4066Sahrens avl_create(t, zil_dva_compare, sizeof (zil_dva_node_t), 102fa9e4066Sahrens offsetof(zil_dva_node_t, zn_node)); 103fa9e4066Sahrens } 104fa9e4066Sahrens 105fa9e4066Sahrens static void 106fa9e4066Sahrens zil_dva_tree_fini(avl_tree_t *t) 107fa9e4066Sahrens { 108fa9e4066Sahrens zil_dva_node_t *zn; 109fa9e4066Sahrens void *cookie = NULL; 110fa9e4066Sahrens 111fa9e4066Sahrens while ((zn = avl_destroy_nodes(t, &cookie)) != NULL) 112fa9e4066Sahrens kmem_free(zn, sizeof (zil_dva_node_t)); 113fa9e4066Sahrens 114fa9e4066Sahrens avl_destroy(t); 115fa9e4066Sahrens } 116fa9e4066Sahrens 117fa9e4066Sahrens static int 118fa9e4066Sahrens zil_dva_tree_add(avl_tree_t *t, dva_t *dva) 119fa9e4066Sahrens { 120fa9e4066Sahrens zil_dva_node_t *zn; 121fa9e4066Sahrens avl_index_t where; 122fa9e4066Sahrens 123fa9e4066Sahrens if (avl_find(t, dva, &where) != NULL) 124fa9e4066Sahrens return (EEXIST); 125fa9e4066Sahrens 126fa9e4066Sahrens zn = kmem_alloc(sizeof (zil_dva_node_t), KM_SLEEP); 127fa9e4066Sahrens zn->zn_dva = *dva; 128fa9e4066Sahrens avl_insert(t, zn, where); 129fa9e4066Sahrens 130fa9e4066Sahrens return (0); 131fa9e4066Sahrens } 132fa9e4066Sahrens 133d80c45e0Sbonwick static zil_header_t * 134d80c45e0Sbonwick zil_header_in_syncing_context(zilog_t *zilog) 135d80c45e0Sbonwick { 136d80c45e0Sbonwick return ((zil_header_t *)zilog->zl_header); 137d80c45e0Sbonwick } 138d80c45e0Sbonwick 139d80c45e0Sbonwick static void 140d80c45e0Sbonwick zil_init_log_chain(zilog_t *zilog, blkptr_t *bp) 141d80c45e0Sbonwick { 142d80c45e0Sbonwick zio_cksum_t *zc = &bp->blk_cksum; 143d80c45e0Sbonwick 144d80c45e0Sbonwick zc->zc_word[ZIL_ZC_GUID_0] = spa_get_random(-1ULL); 145d80c45e0Sbonwick zc->zc_word[ZIL_ZC_GUID_1] = spa_get_random(-1ULL); 146d80c45e0Sbonwick zc->zc_word[ZIL_ZC_OBJSET] = dmu_objset_id(zilog->zl_os); 147d80c45e0Sbonwick zc->zc_word[ZIL_ZC_SEQ] = 1ULL; 148d80c45e0Sbonwick } 149d80c45e0Sbonwick 150fa9e4066Sahrens /* 151fa9e4066Sahrens * Read a log block, make sure it's valid, and byteswap it if necessary. 152fa9e4066Sahrens */ 153fa9e4066Sahrens static int 154d80c45e0Sbonwick zil_read_log_block(zilog_t *zilog, const blkptr_t *bp, arc_buf_t **abufpp) 155fa9e4066Sahrens { 156d80c45e0Sbonwick blkptr_t blk = *bp; 157ea8dc4b6Seschrock zbookmark_t zb; 15813506d1eSmaybee uint32_t aflags = ARC_WAIT; 159fa9e4066Sahrens int error; 160fa9e4066Sahrens 161d80c45e0Sbonwick zb.zb_objset = bp->blk_cksum.zc_word[ZIL_ZC_OBJSET]; 162ea8dc4b6Seschrock zb.zb_object = 0; 163ea8dc4b6Seschrock zb.zb_level = -1; 164d80c45e0Sbonwick zb.zb_blkid = bp->blk_cksum.zc_word[ZIL_ZC_SEQ]; 165ea8dc4b6Seschrock 166d80c45e0Sbonwick *abufpp = NULL; 167fa9e4066Sahrens 168088f3894Sahrens /* 169088f3894Sahrens * We shouldn't be doing any scrubbing while we're doing log 170088f3894Sahrens * replay, it's OK to not lock. 171088f3894Sahrens */ 172088f3894Sahrens error = arc_read_nolock(NULL, zilog->zl_spa, &blk, 173d80c45e0Sbonwick arc_getbuf_func, abufpp, ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL | 17413506d1eSmaybee ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB, &aflags, &zb); 175fa9e4066Sahrens 176d80c45e0Sbonwick if (error == 0) { 177d80c45e0Sbonwick char *data = (*abufpp)->b_data; 178d80c45e0Sbonwick uint64_t blksz = BP_GET_LSIZE(bp); 179d80c45e0Sbonwick zil_trailer_t *ztp = (zil_trailer_t *)(data + blksz) - 1; 180d80c45e0Sbonwick zio_cksum_t cksum = bp->blk_cksum; 181fa9e4066Sahrens 182d80c45e0Sbonwick /* 183f5e6e722SNeil Perrin * Validate the checksummed log block. 184f5e6e722SNeil Perrin * 185d80c45e0Sbonwick * Sequence numbers should be... sequential. The checksum 186d80c45e0Sbonwick * verifier for the next block should be bp's checksum plus 1. 187f5e6e722SNeil Perrin * 188f5e6e722SNeil Perrin * Also check the log chain linkage and size used. 189d80c45e0Sbonwick */ 190d80c45e0Sbonwick cksum.zc_word[ZIL_ZC_SEQ]++; 191d80c45e0Sbonwick 192f5e6e722SNeil Perrin if (bcmp(&cksum, &ztp->zit_next_blk.blk_cksum, 193f5e6e722SNeil Perrin sizeof (cksum)) || BP_IS_HOLE(&ztp->zit_next_blk) || 194f5e6e722SNeil Perrin (ztp->zit_nused > (blksz - sizeof (zil_trailer_t)))) { 195f5e6e722SNeil Perrin error = ECKSUM; 196f5e6e722SNeil Perrin } 197fa9e4066Sahrens 198d80c45e0Sbonwick if (error) { 199d80c45e0Sbonwick VERIFY(arc_buf_remove_ref(*abufpp, abufpp) == 1); 200d80c45e0Sbonwick *abufpp = NULL; 201d80c45e0Sbonwick } 202fa9e4066Sahrens } 203fa9e4066Sahrens 204d80c45e0Sbonwick dprintf("error %d on %llu:%llu\n", error, zb.zb_objset, zb.zb_blkid); 205fa9e4066Sahrens 206d80c45e0Sbonwick return (error); 207fa9e4066Sahrens } 208fa9e4066Sahrens 209fa9e4066Sahrens /* 210fa9e4066Sahrens * Parse the intent log, and call parse_func for each valid record within. 211d80c45e0Sbonwick * Return the highest sequence number. 212fa9e4066Sahrens */ 213d80c45e0Sbonwick uint64_t 214fa9e4066Sahrens zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func, 215fa9e4066Sahrens zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg) 216fa9e4066Sahrens { 217d80c45e0Sbonwick const zil_header_t *zh = zilog->zl_header; 218d80c45e0Sbonwick uint64_t claim_seq = zh->zh_claim_seq; 219d80c45e0Sbonwick uint64_t seq = 0; 220d80c45e0Sbonwick uint64_t max_seq = 0; 221d80c45e0Sbonwick blkptr_t blk = zh->zh_log; 222d80c45e0Sbonwick arc_buf_t *abuf; 223fa9e4066Sahrens char *lrbuf, *lrp; 224fa9e4066Sahrens zil_trailer_t *ztp; 225fa9e4066Sahrens int reclen, error; 226fa9e4066Sahrens 227fa9e4066Sahrens if (BP_IS_HOLE(&blk)) 228d80c45e0Sbonwick return (max_seq); 229fa9e4066Sahrens 230fa9e4066Sahrens /* 231fa9e4066Sahrens * Starting at the block pointed to by zh_log we read the log chain. 232fa9e4066Sahrens * For each block in the chain we strongly check that block to 233fa9e4066Sahrens * ensure its validity. We stop when an invalid block is found. 234fa9e4066Sahrens * For each block pointer in the chain we call parse_blk_func(). 235fa9e4066Sahrens * For each record in each valid block we call parse_lr_func(). 236d80c45e0Sbonwick * If the log has been claimed, stop if we encounter a sequence 237d80c45e0Sbonwick * number greater than the highest claimed sequence number. 238fa9e4066Sahrens */ 239fa9e4066Sahrens zil_dva_tree_init(&zilog->zl_dva_tree); 240fa9e4066Sahrens for (;;) { 241d80c45e0Sbonwick seq = blk.blk_cksum.zc_word[ZIL_ZC_SEQ]; 242d80c45e0Sbonwick 243d80c45e0Sbonwick if (claim_seq != 0 && seq > claim_seq) 244d80c45e0Sbonwick break; 245d80c45e0Sbonwick 246d80c45e0Sbonwick ASSERT(max_seq < seq); 247d80c45e0Sbonwick max_seq = seq; 248d80c45e0Sbonwick 249d80c45e0Sbonwick error = zil_read_log_block(zilog, &blk, &abuf); 250fa9e4066Sahrens 251fa9e4066Sahrens if (parse_blk_func != NULL) 252fa9e4066Sahrens parse_blk_func(zilog, &blk, arg, txg); 253fa9e4066Sahrens 254fa9e4066Sahrens if (error) 255fa9e4066Sahrens break; 256fa9e4066Sahrens 257d80c45e0Sbonwick lrbuf = abuf->b_data; 258fa9e4066Sahrens ztp = (zil_trailer_t *)(lrbuf + BP_GET_LSIZE(&blk)) - 1; 259fa9e4066Sahrens blk = ztp->zit_next_blk; 260fa9e4066Sahrens 261d80c45e0Sbonwick if (parse_lr_func == NULL) { 262d80c45e0Sbonwick VERIFY(arc_buf_remove_ref(abuf, &abuf) == 1); 263fa9e4066Sahrens continue; 264d80c45e0Sbonwick } 265fa9e4066Sahrens 266fa9e4066Sahrens for (lrp = lrbuf; lrp < lrbuf + ztp->zit_nused; lrp += reclen) { 267fa9e4066Sahrens lr_t *lr = (lr_t *)lrp; 268fa9e4066Sahrens reclen = lr->lrc_reclen; 269fa9e4066Sahrens ASSERT3U(reclen, >=, sizeof (lr_t)); 270fa9e4066Sahrens parse_lr_func(zilog, lr, arg, txg); 271fa9e4066Sahrens } 272d80c45e0Sbonwick VERIFY(arc_buf_remove_ref(abuf, &abuf) == 1); 273fa9e4066Sahrens } 274fa9e4066Sahrens zil_dva_tree_fini(&zilog->zl_dva_tree); 275d80c45e0Sbonwick 276d80c45e0Sbonwick return (max_seq); 277fa9e4066Sahrens } 278fa9e4066Sahrens 279fa9e4066Sahrens /* ARGSUSED */ 280fa9e4066Sahrens static void 281fa9e4066Sahrens zil_claim_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t first_txg) 282fa9e4066Sahrens { 283fa9e4066Sahrens spa_t *spa = zilog->zl_spa; 284fa9e4066Sahrens int err; 285fa9e4066Sahrens 286fa9e4066Sahrens /* 287fa9e4066Sahrens * Claim log block if not already committed and not already claimed. 288fa9e4066Sahrens */ 289fa9e4066Sahrens if (bp->blk_birth >= first_txg && 290fa9e4066Sahrens zil_dva_tree_add(&zilog->zl_dva_tree, BP_IDENTITY(bp)) == 0) { 291e14bb325SJeff Bonwick err = zio_wait(zio_claim(NULL, spa, first_txg, bp, NULL, NULL, 292e14bb325SJeff Bonwick ZIO_FLAG_MUSTSUCCEED)); 293fa9e4066Sahrens ASSERT(err == 0); 294fa9e4066Sahrens } 295fa9e4066Sahrens } 296fa9e4066Sahrens 297fa9e4066Sahrens static void 298fa9e4066Sahrens zil_claim_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t first_txg) 299fa9e4066Sahrens { 300fa9e4066Sahrens if (lrc->lrc_txtype == TX_WRITE) { 301fa9e4066Sahrens lr_write_t *lr = (lr_write_t *)lrc; 302fa9e4066Sahrens zil_claim_log_block(zilog, &lr->lr_blkptr, tx, first_txg); 303fa9e4066Sahrens } 304fa9e4066Sahrens } 305fa9e4066Sahrens 306fa9e4066Sahrens /* ARGSUSED */ 307fa9e4066Sahrens static void 308fa9e4066Sahrens zil_free_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t claim_txg) 309fa9e4066Sahrens { 310fa9e4066Sahrens zio_free_blk(zilog->zl_spa, bp, dmu_tx_get_txg(tx)); 311fa9e4066Sahrens } 312fa9e4066Sahrens 313fa9e4066Sahrens static void 314fa9e4066Sahrens zil_free_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t claim_txg) 315fa9e4066Sahrens { 316fa9e4066Sahrens /* 317fa9e4066Sahrens * If we previously claimed it, we need to free it. 318fa9e4066Sahrens */ 319fa9e4066Sahrens if (claim_txg != 0 && lrc->lrc_txtype == TX_WRITE) { 320fa9e4066Sahrens lr_write_t *lr = (lr_write_t *)lrc; 321fa9e4066Sahrens blkptr_t *bp = &lr->lr_blkptr; 322fa9e4066Sahrens if (bp->blk_birth >= claim_txg && 323fa9e4066Sahrens !zil_dva_tree_add(&zilog->zl_dva_tree, BP_IDENTITY(bp))) { 324fa9e4066Sahrens (void) arc_free(NULL, zilog->zl_spa, 325fa9e4066Sahrens dmu_tx_get_txg(tx), bp, NULL, NULL, ARC_WAIT); 326fa9e4066Sahrens } 327fa9e4066Sahrens } 328fa9e4066Sahrens } 329fa9e4066Sahrens 330fa9e4066Sahrens /* 331fa9e4066Sahrens * Create an on-disk intent log. 332fa9e4066Sahrens */ 333fa9e4066Sahrens static void 334fa9e4066Sahrens zil_create(zilog_t *zilog) 335fa9e4066Sahrens { 336d80c45e0Sbonwick const zil_header_t *zh = zilog->zl_header; 337fa9e4066Sahrens lwb_t *lwb; 338d80c45e0Sbonwick uint64_t txg = 0; 339d80c45e0Sbonwick dmu_tx_t *tx = NULL; 340fa9e4066Sahrens blkptr_t blk; 341d80c45e0Sbonwick int error = 0; 342fa9e4066Sahrens 343fa9e4066Sahrens /* 344d80c45e0Sbonwick * Wait for any previous destroy to complete. 345fa9e4066Sahrens */ 346d80c45e0Sbonwick txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg); 347d80c45e0Sbonwick 348d80c45e0Sbonwick ASSERT(zh->zh_claim_txg == 0); 349d80c45e0Sbonwick ASSERT(zh->zh_replay_seq == 0); 350d80c45e0Sbonwick 351d80c45e0Sbonwick blk = zh->zh_log; 352fa9e4066Sahrens 353fa9e4066Sahrens /* 354899217ddSNeil Perrin * If we don't already have an initial log block or we have one 355899217ddSNeil Perrin * but it's the wrong endianness then allocate one. 356fa9e4066Sahrens */ 357899217ddSNeil Perrin if (BP_IS_HOLE(&blk) || BP_SHOULD_BYTESWAP(&blk)) { 358d80c45e0Sbonwick tx = dmu_tx_create(zilog->zl_os); 359d80c45e0Sbonwick (void) dmu_tx_assign(tx, TXG_WAIT); 360d80c45e0Sbonwick dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); 361d80c45e0Sbonwick txg = dmu_tx_get_txg(tx); 362d80c45e0Sbonwick 363899217ddSNeil Perrin if (!BP_IS_HOLE(&blk)) { 364899217ddSNeil Perrin zio_free_blk(zilog->zl_spa, &blk, txg); 365899217ddSNeil Perrin BP_ZERO(&blk); 366899217ddSNeil Perrin } 367899217ddSNeil Perrin 36867bd71c6Sperrin error = zio_alloc_blk(zilog->zl_spa, ZIL_MIN_BLKSZ, &blk, 36967bd71c6Sperrin NULL, txg); 370d80c45e0Sbonwick 371d80c45e0Sbonwick if (error == 0) 372d80c45e0Sbonwick zil_init_log_chain(zilog, &blk); 37313f5297eSperrin } 374fa9e4066Sahrens 375d80c45e0Sbonwick /* 376d80c45e0Sbonwick * Allocate a log write buffer (lwb) for the first log block. 377d80c45e0Sbonwick */ 378d80c45e0Sbonwick if (error == 0) { 379fa9e4066Sahrens lwb = kmem_cache_alloc(zil_lwb_cache, KM_SLEEP); 380fa9e4066Sahrens lwb->lwb_zilog = zilog; 381fa9e4066Sahrens lwb->lwb_blk = blk; 382fa9e4066Sahrens lwb->lwb_nused = 0; 383fa9e4066Sahrens lwb->lwb_sz = BP_GET_LSIZE(&lwb->lwb_blk); 384fa9e4066Sahrens lwb->lwb_buf = zio_buf_alloc(lwb->lwb_sz); 385fa9e4066Sahrens lwb->lwb_max_txg = txg; 386c5c6ffa0Smaybee lwb->lwb_zio = NULL; 387c5c6ffa0Smaybee 388fa9e4066Sahrens mutex_enter(&zilog->zl_lock); 389fa9e4066Sahrens list_insert_tail(&zilog->zl_lwb_list, lwb); 390fa9e4066Sahrens mutex_exit(&zilog->zl_lock); 391fa9e4066Sahrens } 392fa9e4066Sahrens 393d80c45e0Sbonwick /* 394d80c45e0Sbonwick * If we just allocated the first log block, commit our transaction 395d80c45e0Sbonwick * and wait for zil_sync() to stuff the block poiner into zh_log. 396d80c45e0Sbonwick * (zh is part of the MOS, so we cannot modify it in open context.) 397d80c45e0Sbonwick */ 398d80c45e0Sbonwick if (tx != NULL) { 399d80c45e0Sbonwick dmu_tx_commit(tx); 40013f5297eSperrin txg_wait_synced(zilog->zl_dmu_pool, txg); 401d80c45e0Sbonwick } 402d80c45e0Sbonwick 403d80c45e0Sbonwick ASSERT(bcmp(&blk, &zh->zh_log, sizeof (blk)) == 0); 404fa9e4066Sahrens } 405fa9e4066Sahrens 406fa9e4066Sahrens /* 407fa9e4066Sahrens * In one tx, free all log blocks and clear the log header. 408d80c45e0Sbonwick * If keep_first is set, then we're replaying a log with no content. 409d80c45e0Sbonwick * We want to keep the first block, however, so that the first 410d80c45e0Sbonwick * synchronous transaction doesn't require a txg_wait_synced() 411d80c45e0Sbonwick * in zil_create(). We don't need to txg_wait_synced() here either 412d80c45e0Sbonwick * when keep_first is set, because both zil_create() and zil_destroy() 413d80c45e0Sbonwick * will wait for any in-progress destroys to complete. 414fa9e4066Sahrens */ 415fa9e4066Sahrens void 416d80c45e0Sbonwick zil_destroy(zilog_t *zilog, boolean_t keep_first) 417fa9e4066Sahrens { 418d80c45e0Sbonwick const zil_header_t *zh = zilog->zl_header; 419d80c45e0Sbonwick lwb_t *lwb; 420fa9e4066Sahrens dmu_tx_t *tx; 421fa9e4066Sahrens uint64_t txg; 422fa9e4066Sahrens 423d80c45e0Sbonwick /* 424d80c45e0Sbonwick * Wait for any previous destroy to complete. 425d80c45e0Sbonwick */ 426d80c45e0Sbonwick txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg); 427fa9e4066Sahrens 428d80c45e0Sbonwick if (BP_IS_HOLE(&zh->zh_log)) 429fa9e4066Sahrens return; 430fa9e4066Sahrens 431fa9e4066Sahrens tx = dmu_tx_create(zilog->zl_os); 432fa9e4066Sahrens (void) dmu_tx_assign(tx, TXG_WAIT); 433fa9e4066Sahrens dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); 434fa9e4066Sahrens txg = dmu_tx_get_txg(tx); 435fa9e4066Sahrens 436d80c45e0Sbonwick mutex_enter(&zilog->zl_lock); 437d80c45e0Sbonwick 438587c2a35Sperrin /* 439587c2a35Sperrin * It is possible for the ZIL to get the previously mounted zilog 440587c2a35Sperrin * structure of the same dataset if quickly remounted and the dbuf 441587c2a35Sperrin * eviction has not completed. In this case we can see a non 442587c2a35Sperrin * empty lwb list and keep_first will be set. We fix this by 443587c2a35Sperrin * clearing the keep_first. This will be slower but it's very rare. 444587c2a35Sperrin */ 445587c2a35Sperrin if (!list_is_empty(&zilog->zl_lwb_list) && keep_first) 446587c2a35Sperrin keep_first = B_FALSE; 447587c2a35Sperrin 448d80c45e0Sbonwick ASSERT3U(zilog->zl_destroy_txg, <, txg); 449fa9e4066Sahrens zilog->zl_destroy_txg = txg; 450d80c45e0Sbonwick zilog->zl_keep_first = keep_first; 451d80c45e0Sbonwick 452d80c45e0Sbonwick if (!list_is_empty(&zilog->zl_lwb_list)) { 453d80c45e0Sbonwick ASSERT(zh->zh_claim_txg == 0); 454d80c45e0Sbonwick ASSERT(!keep_first); 455d80c45e0Sbonwick while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) { 456d80c45e0Sbonwick list_remove(&zilog->zl_lwb_list, lwb); 457d80c45e0Sbonwick if (lwb->lwb_buf != NULL) 458d80c45e0Sbonwick zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); 459d80c45e0Sbonwick zio_free_blk(zilog->zl_spa, &lwb->lwb_blk, txg); 460d80c45e0Sbonwick kmem_cache_free(zil_lwb_cache, lwb); 461d80c45e0Sbonwick } 462d80c45e0Sbonwick } else { 463d80c45e0Sbonwick if (!keep_first) { 464d80c45e0Sbonwick (void) zil_parse(zilog, zil_free_log_block, 465d80c45e0Sbonwick zil_free_log_record, tx, zh->zh_claim_txg); 466d80c45e0Sbonwick } 467d80c45e0Sbonwick } 468b19a79ecSperrin mutex_exit(&zilog->zl_lock); 469fa9e4066Sahrens 470fa9e4066Sahrens dmu_tx_commit(tx); 471fa9e4066Sahrens } 472fa9e4066Sahrens 4733589c4f0SNeil Perrin /* 4743589c4f0SNeil Perrin * return true if the initial log block is not valid 4753589c4f0SNeil Perrin */ 4763589c4f0SNeil Perrin static boolean_t 4773589c4f0SNeil Perrin zil_empty(zilog_t *zilog) 4783589c4f0SNeil Perrin { 4793589c4f0SNeil Perrin const zil_header_t *zh = zilog->zl_header; 4803589c4f0SNeil Perrin arc_buf_t *abuf = NULL; 4813589c4f0SNeil Perrin 4823589c4f0SNeil Perrin if (BP_IS_HOLE(&zh->zh_log)) 4833589c4f0SNeil Perrin return (B_TRUE); 4843589c4f0SNeil Perrin 4853589c4f0SNeil Perrin if (zil_read_log_block(zilog, &zh->zh_log, &abuf) != 0) 4863589c4f0SNeil Perrin return (B_TRUE); 4873589c4f0SNeil Perrin 4883589c4f0SNeil Perrin VERIFY(arc_buf_remove_ref(abuf, &abuf) == 1); 4893589c4f0SNeil Perrin return (B_FALSE); 4903589c4f0SNeil Perrin } 4913589c4f0SNeil Perrin 4921d452cf5Sahrens int 493fa9e4066Sahrens zil_claim(char *osname, void *txarg) 494fa9e4066Sahrens { 495fa9e4066Sahrens dmu_tx_t *tx = txarg; 496fa9e4066Sahrens uint64_t first_txg = dmu_tx_get_txg(tx); 497fa9e4066Sahrens zilog_t *zilog; 498fa9e4066Sahrens zil_header_t *zh; 499fa9e4066Sahrens objset_t *os; 500fa9e4066Sahrens int error; 501fa9e4066Sahrens 502745cd3c5Smaybee error = dmu_objset_open(osname, DMU_OST_ANY, DS_MODE_USER, &os); 503fa9e4066Sahrens if (error) { 504b87f3af3Sperrin cmn_err(CE_WARN, "can't open objset for %s", osname); 5051d452cf5Sahrens return (0); 506fa9e4066Sahrens } 507fa9e4066Sahrens 508fa9e4066Sahrens zilog = dmu_objset_zil(os); 509d80c45e0Sbonwick zh = zil_header_in_syncing_context(zilog); 510fa9e4066Sahrens 5113589c4f0SNeil Perrin /* 5123589c4f0SNeil Perrin * Record here whether the zil has any records to replay. 5133589c4f0SNeil Perrin * If the header block pointer is null or the block points 5143589c4f0SNeil Perrin * to the stubby then we know there are no valid log records. 5153589c4f0SNeil Perrin * We use the header to store this state as the the zilog gets 5163589c4f0SNeil Perrin * freed later in dmu_objset_close(). 5173589c4f0SNeil Perrin * The flags (and the rest of the header fields) are cleared in 5183589c4f0SNeil Perrin * zil_sync() as a result of a zil_destroy(), after replaying the log. 5193589c4f0SNeil Perrin * 5203589c4f0SNeil Perrin * Note, the intent log can be empty but still need the 5213589c4f0SNeil Perrin * stubby to be claimed. 5223589c4f0SNeil Perrin */ 5233589c4f0SNeil Perrin if (!zil_empty(zilog)) 5243589c4f0SNeil Perrin zh->zh_flags |= ZIL_REPLAY_NEEDED; 5253589c4f0SNeil Perrin 526fa9e4066Sahrens /* 527d80c45e0Sbonwick * Claim all log blocks if we haven't already done so, and remember 528d80c45e0Sbonwick * the highest claimed sequence number. This ensures that if we can 529d80c45e0Sbonwick * read only part of the log now (e.g. due to a missing device), 530d80c45e0Sbonwick * but we can read the entire log later, we will not try to replay 531d80c45e0Sbonwick * or destroy beyond the last block we successfully claimed. 532fa9e4066Sahrens */ 533fa9e4066Sahrens ASSERT3U(zh->zh_claim_txg, <=, first_txg); 534fa9e4066Sahrens if (zh->zh_claim_txg == 0 && !BP_IS_HOLE(&zh->zh_log)) { 535fa9e4066Sahrens zh->zh_claim_txg = first_txg; 536d80c45e0Sbonwick zh->zh_claim_seq = zil_parse(zilog, zil_claim_log_block, 537d80c45e0Sbonwick zil_claim_log_record, tx, first_txg); 538fa9e4066Sahrens dsl_dataset_dirty(dmu_objset_ds(os), tx); 539fa9e4066Sahrens } 540d80c45e0Sbonwick 541fa9e4066Sahrens ASSERT3U(first_txg, ==, (spa_last_synced_txg(zilog->zl_spa) + 1)); 542fa9e4066Sahrens dmu_objset_close(os); 5431d452cf5Sahrens return (0); 544b87f3af3Sperrin } 545b87f3af3Sperrin 546b87f3af3Sperrin /* 547b87f3af3Sperrin * Check the log by walking the log chain. 548b87f3af3Sperrin * Checksum errors are ok as they indicate the end of the chain. 549b87f3af3Sperrin * Any other error (no device or read failure) returns an error. 550b87f3af3Sperrin */ 551b87f3af3Sperrin /* ARGSUSED */ 552b87f3af3Sperrin int 553b87f3af3Sperrin zil_check_log_chain(char *osname, void *txarg) 554b87f3af3Sperrin { 555b87f3af3Sperrin zilog_t *zilog; 556b87f3af3Sperrin zil_header_t *zh; 557b87f3af3Sperrin blkptr_t blk; 558b87f3af3Sperrin arc_buf_t *abuf; 559b87f3af3Sperrin objset_t *os; 560b87f3af3Sperrin char *lrbuf; 561b87f3af3Sperrin zil_trailer_t *ztp; 562b87f3af3Sperrin int error; 563b87f3af3Sperrin 564b87f3af3Sperrin error = dmu_objset_open(osname, DMU_OST_ANY, DS_MODE_USER, &os); 565b87f3af3Sperrin if (error) { 566b87f3af3Sperrin cmn_err(CE_WARN, "can't open objset for %s", osname); 567b87f3af3Sperrin return (0); 568b87f3af3Sperrin } 569b87f3af3Sperrin 570b87f3af3Sperrin zilog = dmu_objset_zil(os); 571b87f3af3Sperrin zh = zil_header_in_syncing_context(zilog); 572b87f3af3Sperrin blk = zh->zh_log; 573b87f3af3Sperrin if (BP_IS_HOLE(&blk)) { 574b87f3af3Sperrin dmu_objset_close(os); 575b87f3af3Sperrin return (0); /* no chain */ 576b87f3af3Sperrin } 577b87f3af3Sperrin 578b87f3af3Sperrin for (;;) { 579b87f3af3Sperrin error = zil_read_log_block(zilog, &blk, &abuf); 580b87f3af3Sperrin if (error) 581b87f3af3Sperrin break; 582b87f3af3Sperrin lrbuf = abuf->b_data; 583b87f3af3Sperrin ztp = (zil_trailer_t *)(lrbuf + BP_GET_LSIZE(&blk)) - 1; 584b87f3af3Sperrin blk = ztp->zit_next_blk; 585b87f3af3Sperrin VERIFY(arc_buf_remove_ref(abuf, &abuf) == 1); 586b87f3af3Sperrin } 587b87f3af3Sperrin dmu_objset_close(os); 588b87f3af3Sperrin if (error == ECKSUM) 589b87f3af3Sperrin return (0); /* normal end of chain */ 590b87f3af3Sperrin return (error); 591b87f3af3Sperrin } 592b87f3af3Sperrin 593b87f3af3Sperrin /* 594b87f3af3Sperrin * Clear a log chain 595b87f3af3Sperrin */ 596b87f3af3Sperrin /* ARGSUSED */ 597b87f3af3Sperrin int 598b87f3af3Sperrin zil_clear_log_chain(char *osname, void *txarg) 599b87f3af3Sperrin { 600b87f3af3Sperrin zilog_t *zilog; 601b87f3af3Sperrin zil_header_t *zh; 602b87f3af3Sperrin objset_t *os; 603b87f3af3Sperrin dmu_tx_t *tx; 604b87f3af3Sperrin int error; 605b87f3af3Sperrin 606b87f3af3Sperrin error = dmu_objset_open(osname, DMU_OST_ANY, DS_MODE_USER, &os); 607b87f3af3Sperrin if (error) { 608b87f3af3Sperrin cmn_err(CE_WARN, "can't open objset for %s", osname); 609b87f3af3Sperrin return (0); 610b87f3af3Sperrin } 611b87f3af3Sperrin 612b87f3af3Sperrin zilog = dmu_objset_zil(os); 613b87f3af3Sperrin tx = dmu_tx_create(zilog->zl_os); 614b87f3af3Sperrin (void) dmu_tx_assign(tx, TXG_WAIT); 615b87f3af3Sperrin zh = zil_header_in_syncing_context(zilog); 616b87f3af3Sperrin BP_ZERO(&zh->zh_log); 617b87f3af3Sperrin dsl_dataset_dirty(dmu_objset_ds(os), tx); 618b87f3af3Sperrin dmu_tx_commit(tx); 619b87f3af3Sperrin dmu_objset_close(os); 620b87f3af3Sperrin return (0); 621fa9e4066Sahrens } 622fa9e4066Sahrens 62317f17c2dSbonwick static int 62417f17c2dSbonwick zil_vdev_compare(const void *x1, const void *x2) 62517f17c2dSbonwick { 626c0d19b33Sperrin uint64_t v1 = ((zil_vdev_node_t *)x1)->zv_vdev; 627c0d19b33Sperrin uint64_t v2 = ((zil_vdev_node_t *)x2)->zv_vdev; 62817f17c2dSbonwick 62917f17c2dSbonwick if (v1 < v2) 63017f17c2dSbonwick return (-1); 63117f17c2dSbonwick if (v1 > v2) 63217f17c2dSbonwick return (1); 63317f17c2dSbonwick 63417f17c2dSbonwick return (0); 63517f17c2dSbonwick } 63617f17c2dSbonwick 637fa9e4066Sahrens void 63817f17c2dSbonwick zil_add_block(zilog_t *zilog, blkptr_t *bp) 639fa9e4066Sahrens { 64017f17c2dSbonwick avl_tree_t *t = &zilog->zl_vdev_tree; 64117f17c2dSbonwick avl_index_t where; 64217f17c2dSbonwick zil_vdev_node_t *zv, zvsearch; 64317f17c2dSbonwick int ndvas = BP_GET_NDVAS(bp); 64417f17c2dSbonwick int i; 645fa9e4066Sahrens 646416e0cd8Sek if (zfs_nocacheflush) 647fa9e4066Sahrens return; 648fa9e4066Sahrens 64917f17c2dSbonwick ASSERT(zilog->zl_writer); 65017f17c2dSbonwick 65117f17c2dSbonwick /* 65217f17c2dSbonwick * Even though we're zl_writer, we still need a lock because the 65317f17c2dSbonwick * zl_get_data() callbacks may have dmu_sync() done callbacks 65417f17c2dSbonwick * that will run concurrently. 65517f17c2dSbonwick */ 65617f17c2dSbonwick mutex_enter(&zilog->zl_vdev_lock); 65717f17c2dSbonwick for (i = 0; i < ndvas; i++) { 65817f17c2dSbonwick zvsearch.zv_vdev = DVA_GET_VDEV(&bp->blk_dva[i]); 65917f17c2dSbonwick if (avl_find(t, &zvsearch, &where) == NULL) { 66017f17c2dSbonwick zv = kmem_alloc(sizeof (*zv), KM_SLEEP); 66117f17c2dSbonwick zv->zv_vdev = zvsearch.zv_vdev; 66217f17c2dSbonwick avl_insert(t, zv, where); 66367bd71c6Sperrin } 66467bd71c6Sperrin } 66517f17c2dSbonwick mutex_exit(&zilog->zl_vdev_lock); 666fa9e4066Sahrens } 667fa9e4066Sahrens 66867bd71c6Sperrin void 66967bd71c6Sperrin zil_flush_vdevs(zilog_t *zilog) 67067bd71c6Sperrin { 67167bd71c6Sperrin spa_t *spa = zilog->zl_spa; 67217f17c2dSbonwick avl_tree_t *t = &zilog->zl_vdev_tree; 67317f17c2dSbonwick void *cookie = NULL; 67417f17c2dSbonwick zil_vdev_node_t *zv; 67517f17c2dSbonwick zio_t *zio; 676fa9e4066Sahrens 67767bd71c6Sperrin ASSERT(zilog->zl_writer); 67867bd71c6Sperrin 67917f17c2dSbonwick /* 68017f17c2dSbonwick * We don't need zl_vdev_lock here because we're the zl_writer, 68117f17c2dSbonwick * and all zl_get_data() callbacks are done. 68217f17c2dSbonwick */ 68317f17c2dSbonwick if (avl_numnodes(t) == 0) 68417f17c2dSbonwick return; 68517f17c2dSbonwick 686e14bb325SJeff Bonwick spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 68717f17c2dSbonwick 688e14bb325SJeff Bonwick zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL); 689fa9e4066Sahrens 69017f17c2dSbonwick while ((zv = avl_destroy_nodes(t, &cookie)) != NULL) { 69117f17c2dSbonwick vdev_t *vd = vdev_lookup_top(spa, zv->zv_vdev); 69217f17c2dSbonwick if (vd != NULL) 69317f17c2dSbonwick zio_flush(zio, vd); 69417f17c2dSbonwick kmem_free(zv, sizeof (*zv)); 69567bd71c6Sperrin } 69617f17c2dSbonwick 697fa9e4066Sahrens /* 698fa9e4066Sahrens * Wait for all the flushes to complete. Not all devices actually 699fa9e4066Sahrens * support the DKIOCFLUSHWRITECACHE ioctl, so it's OK if it fails. 700fa9e4066Sahrens */ 70117f17c2dSbonwick (void) zio_wait(zio); 70217f17c2dSbonwick 703e14bb325SJeff Bonwick spa_config_exit(spa, SCL_STATE, FTAG); 704fa9e4066Sahrens } 705fa9e4066Sahrens 706fa9e4066Sahrens /* 707fa9e4066Sahrens * Function called when a log block write completes 708fa9e4066Sahrens */ 709fa9e4066Sahrens static void 710fa9e4066Sahrens zil_lwb_write_done(zio_t *zio) 711fa9e4066Sahrens { 712fa9e4066Sahrens lwb_t *lwb = zio->io_private; 713fa9e4066Sahrens zilog_t *zilog = lwb->lwb_zilog; 714fa9e4066Sahrens 715e14bb325SJeff Bonwick ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF); 716e14bb325SJeff Bonwick ASSERT(BP_GET_CHECKSUM(zio->io_bp) == ZIO_CHECKSUM_ZILOG); 717e14bb325SJeff Bonwick ASSERT(BP_GET_TYPE(zio->io_bp) == DMU_OT_INTENT_LOG); 718e14bb325SJeff Bonwick ASSERT(BP_GET_LEVEL(zio->io_bp) == 0); 719e14bb325SJeff Bonwick ASSERT(BP_GET_BYTEORDER(zio->io_bp) == ZFS_HOST_BYTEORDER); 720e14bb325SJeff Bonwick ASSERT(!BP_IS_GANG(zio->io_bp)); 721e14bb325SJeff Bonwick ASSERT(!BP_IS_HOLE(zio->io_bp)); 722e14bb325SJeff Bonwick ASSERT(zio->io_bp->blk_fill == 0); 723e14bb325SJeff Bonwick 724fa9e4066Sahrens /* 725*ef0d8e11SNeil Perrin * Ensure the lwb buffer pointer is cleared before releasing 726*ef0d8e11SNeil Perrin * the txg. If we have had an allocation failure and 727*ef0d8e11SNeil Perrin * the txg is waiting to sync then we want want zil_sync() 728*ef0d8e11SNeil Perrin * to remove the lwb so that it's not picked up as the next new 729*ef0d8e11SNeil Perrin * one in zil_commit_writer(). zil_sync() will only remove 730*ef0d8e11SNeil Perrin * the lwb if lwb_buf is null. 731fa9e4066Sahrens */ 732fa9e4066Sahrens zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); 733fa9e4066Sahrens mutex_enter(&zilog->zl_lock); 734fa9e4066Sahrens lwb->lwb_buf = NULL; 7358654d025Sperrin if (zio->io_error) 736fa9e4066Sahrens zilog->zl_log_error = B_TRUE; 737fa9e4066Sahrens mutex_exit(&zilog->zl_lock); 738*ef0d8e11SNeil Perrin 739*ef0d8e11SNeil Perrin /* 740*ef0d8e11SNeil Perrin * Now that we've written this log block, we have a stable pointer 741*ef0d8e11SNeil Perrin * to the next block in the chain, so it's OK to let the txg in 742*ef0d8e11SNeil Perrin * which we allocated the next block sync. 743*ef0d8e11SNeil Perrin */ 744*ef0d8e11SNeil Perrin txg_rele_to_sync(&lwb->lwb_txgh); 745fa9e4066Sahrens } 746fa9e4066Sahrens 747c5c6ffa0Smaybee /* 748c5c6ffa0Smaybee * Initialize the io for a log block. 749c5c6ffa0Smaybee */ 750c5c6ffa0Smaybee static void 751c5c6ffa0Smaybee zil_lwb_write_init(zilog_t *zilog, lwb_t *lwb) 752c5c6ffa0Smaybee { 753c5c6ffa0Smaybee zbookmark_t zb; 754c5c6ffa0Smaybee 755c5c6ffa0Smaybee zb.zb_objset = lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_OBJSET]; 756c5c6ffa0Smaybee zb.zb_object = 0; 757c5c6ffa0Smaybee zb.zb_level = -1; 758c5c6ffa0Smaybee zb.zb_blkid = lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_SEQ]; 759c5c6ffa0Smaybee 760b19a79ecSperrin if (zilog->zl_root_zio == NULL) { 761b19a79ecSperrin zilog->zl_root_zio = zio_root(zilog->zl_spa, NULL, NULL, 762b19a79ecSperrin ZIO_FLAG_CANFAIL); 763b19a79ecSperrin } 76467bd71c6Sperrin if (lwb->lwb_zio == NULL) { 76567bd71c6Sperrin lwb->lwb_zio = zio_rewrite(zilog->zl_root_zio, zilog->zl_spa, 766e14bb325SJeff Bonwick 0, &lwb->lwb_blk, lwb->lwb_buf, 76767bd71c6Sperrin lwb->lwb_sz, zil_lwb_write_done, lwb, 7688654d025Sperrin ZIO_PRIORITY_LOG_WRITE, ZIO_FLAG_CANFAIL, &zb); 76967bd71c6Sperrin } 770c5c6ffa0Smaybee } 771c5c6ffa0Smaybee 772fa9e4066Sahrens /* 773fa9e4066Sahrens * Start a log block write and advance to the next log block. 774fa9e4066Sahrens * Calls are serialized. 775fa9e4066Sahrens */ 776fa9e4066Sahrens static lwb_t * 777fa9e4066Sahrens zil_lwb_write_start(zilog_t *zilog, lwb_t *lwb) 778fa9e4066Sahrens { 779fa9e4066Sahrens lwb_t *nlwb; 780fa9e4066Sahrens zil_trailer_t *ztp = (zil_trailer_t *)(lwb->lwb_buf + lwb->lwb_sz) - 1; 781d80c45e0Sbonwick spa_t *spa = zilog->zl_spa; 782d80c45e0Sbonwick blkptr_t *bp = &ztp->zit_next_blk; 783fa9e4066Sahrens uint64_t txg; 784fa9e4066Sahrens uint64_t zil_blksz; 785fa9e4066Sahrens int error; 786fa9e4066Sahrens 787fa9e4066Sahrens ASSERT(lwb->lwb_nused <= ZIL_BLK_DATA_SZ(lwb)); 788fa9e4066Sahrens 789fa9e4066Sahrens /* 790fa9e4066Sahrens * Allocate the next block and save its address in this block 791fa9e4066Sahrens * before writing it in order to establish the log chain. 792fa9e4066Sahrens * Note that if the allocation of nlwb synced before we wrote 793fa9e4066Sahrens * the block that points at it (lwb), we'd leak it if we crashed. 794fa9e4066Sahrens * Therefore, we don't do txg_rele_to_sync() until zil_lwb_write_done(). 795fa9e4066Sahrens */ 796fa9e4066Sahrens txg = txg_hold_open(zilog->zl_dmu_pool, &lwb->lwb_txgh); 797fa9e4066Sahrens txg_rele_to_quiesce(&lwb->lwb_txgh); 798fa9e4066Sahrens 799fa9e4066Sahrens /* 80022ac5be4Sperrin * Pick a ZIL blocksize. We request a size that is the 80122ac5be4Sperrin * maximum of the previous used size, the current used size and 80222ac5be4Sperrin * the amount waiting in the queue. 803fa9e4066Sahrens */ 804c5c6ffa0Smaybee zil_blksz = MAX(zilog->zl_prev_used, 805c5c6ffa0Smaybee zilog->zl_cur_used + sizeof (*ztp)); 80622ac5be4Sperrin zil_blksz = MAX(zil_blksz, zilog->zl_itx_list_sz + sizeof (*ztp)); 807b4d654b0Sperrin zil_blksz = P2ROUNDUP_TYPED(zil_blksz, ZIL_MIN_BLKSZ, uint64_t); 80822ac5be4Sperrin if (zil_blksz > ZIL_MAX_BLKSZ) 80922ac5be4Sperrin zil_blksz = ZIL_MAX_BLKSZ; 810fa9e4066Sahrens 81167bd71c6Sperrin BP_ZERO(bp); 81267bd71c6Sperrin /* pass the old blkptr in order to spread log blocks across devs */ 81367bd71c6Sperrin error = zio_alloc_blk(spa, zil_blksz, bp, &lwb->lwb_blk, txg); 814fa9e4066Sahrens if (error) { 815d63d470bSgw dmu_tx_t *tx = dmu_tx_create_assigned(zilog->zl_dmu_pool, txg); 816d63d470bSgw 817d63d470bSgw /* 818d63d470bSgw * We dirty the dataset to ensure that zil_sync() will 819d63d470bSgw * be called to remove this lwb from our zl_lwb_list. 820d63d470bSgw * Failing to do so, may leave an lwb with a NULL lwb_buf 821d63d470bSgw * hanging around on the zl_lwb_list. 822d63d470bSgw */ 823d63d470bSgw dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); 824aeb1c1b6Sgw dmu_tx_commit(tx); 825d63d470bSgw 826d63d470bSgw /* 827d63d470bSgw * Since we've just experienced an allocation failure so we 828d63d470bSgw * terminate the current lwb and send it on its way. 829d63d470bSgw */ 830d63d470bSgw ztp->zit_pad = 0; 831d63d470bSgw ztp->zit_nused = lwb->lwb_nused; 832d63d470bSgw ztp->zit_bt.zbt_cksum = lwb->lwb_blk.blk_cksum; 833d63d470bSgw zio_nowait(lwb->lwb_zio); 834d63d470bSgw 835ea8dc4b6Seschrock /* 836ea8dc4b6Seschrock * By returning NULL the caller will call tx_wait_synced() 837ea8dc4b6Seschrock */ 838fa9e4066Sahrens return (NULL); 839fa9e4066Sahrens } 840fa9e4066Sahrens 841d80c45e0Sbonwick ASSERT3U(bp->blk_birth, ==, txg); 842ea8dc4b6Seschrock ztp->zit_pad = 0; 843fa9e4066Sahrens ztp->zit_nused = lwb->lwb_nused; 844fa9e4066Sahrens ztp->zit_bt.zbt_cksum = lwb->lwb_blk.blk_cksum; 845d80c45e0Sbonwick bp->blk_cksum = lwb->lwb_blk.blk_cksum; 846d80c45e0Sbonwick bp->blk_cksum.zc_word[ZIL_ZC_SEQ]++; 847fa9e4066Sahrens 848fa9e4066Sahrens /* 849fa9e4066Sahrens * Allocate a new log write buffer (lwb). 850fa9e4066Sahrens */ 851fa9e4066Sahrens nlwb = kmem_cache_alloc(zil_lwb_cache, KM_SLEEP); 852fa9e4066Sahrens 853fa9e4066Sahrens nlwb->lwb_zilog = zilog; 854d80c45e0Sbonwick nlwb->lwb_blk = *bp; 855fa9e4066Sahrens nlwb->lwb_nused = 0; 856fa9e4066Sahrens nlwb->lwb_sz = BP_GET_LSIZE(&nlwb->lwb_blk); 857fa9e4066Sahrens nlwb->lwb_buf = zio_buf_alloc(nlwb->lwb_sz); 858fa9e4066Sahrens nlwb->lwb_max_txg = txg; 859c5c6ffa0Smaybee nlwb->lwb_zio = NULL; 860fa9e4066Sahrens 861fa9e4066Sahrens /* 86267bd71c6Sperrin * Put new lwb at the end of the log chain 863fa9e4066Sahrens */ 864fa9e4066Sahrens mutex_enter(&zilog->zl_lock); 865fa9e4066Sahrens list_insert_tail(&zilog->zl_lwb_list, nlwb); 866fa9e4066Sahrens mutex_exit(&zilog->zl_lock); 867fa9e4066Sahrens 86817f17c2dSbonwick /* Record the block for later vdev flushing */ 86917f17c2dSbonwick zil_add_block(zilog, &lwb->lwb_blk); 87067bd71c6Sperrin 871fa9e4066Sahrens /* 872c5c6ffa0Smaybee * kick off the write for the old log block 873fa9e4066Sahrens */ 874c5c6ffa0Smaybee dprintf_bp(&lwb->lwb_blk, "lwb %p txg %llu: ", lwb, txg); 87567bd71c6Sperrin ASSERT(lwb->lwb_zio); 876c5c6ffa0Smaybee zio_nowait(lwb->lwb_zio); 877fa9e4066Sahrens 878fa9e4066Sahrens return (nlwb); 879fa9e4066Sahrens } 880fa9e4066Sahrens 881fa9e4066Sahrens static lwb_t * 882fa9e4066Sahrens zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb) 883fa9e4066Sahrens { 884fa9e4066Sahrens lr_t *lrc = &itx->itx_lr; /* common log record */ 885c5c6ffa0Smaybee lr_write_t *lr = (lr_write_t *)lrc; 886fa9e4066Sahrens uint64_t txg = lrc->lrc_txg; 887fa9e4066Sahrens uint64_t reclen = lrc->lrc_reclen; 888c5c6ffa0Smaybee uint64_t dlen; 889fa9e4066Sahrens 890fa9e4066Sahrens if (lwb == NULL) 891fa9e4066Sahrens return (NULL); 892fa9e4066Sahrens ASSERT(lwb->lwb_buf != NULL); 893fa9e4066Sahrens 894c5c6ffa0Smaybee if (lrc->lrc_txtype == TX_WRITE && itx->itx_wr_state == WR_NEED_COPY) 895c5c6ffa0Smaybee dlen = P2ROUNDUP_TYPED( 896c5c6ffa0Smaybee lr->lr_length, sizeof (uint64_t), uint64_t); 897c5c6ffa0Smaybee else 898c5c6ffa0Smaybee dlen = 0; 899fa9e4066Sahrens 900104e2ed7Sperrin zilog->zl_cur_used += (reclen + dlen); 90122ac5be4Sperrin 90267bd71c6Sperrin zil_lwb_write_init(zilog, lwb); 90367bd71c6Sperrin 904fa9e4066Sahrens /* 905fa9e4066Sahrens * If this record won't fit in the current log block, start a new one. 906fa9e4066Sahrens */ 907104e2ed7Sperrin if (lwb->lwb_nused + reclen + dlen > ZIL_BLK_DATA_SZ(lwb)) { 908fa9e4066Sahrens lwb = zil_lwb_write_start(zilog, lwb); 909c5c6ffa0Smaybee if (lwb == NULL) 910fa9e4066Sahrens return (NULL); 91167bd71c6Sperrin zil_lwb_write_init(zilog, lwb); 912ea8dc4b6Seschrock ASSERT(lwb->lwb_nused == 0); 913104e2ed7Sperrin if (reclen + dlen > ZIL_BLK_DATA_SZ(lwb)) { 914fa9e4066Sahrens txg_wait_synced(zilog->zl_dmu_pool, txg); 915fa9e4066Sahrens return (lwb); 916fa9e4066Sahrens } 917fa9e4066Sahrens } 918fa9e4066Sahrens 919b19a79ecSperrin /* 920b19a79ecSperrin * Update the lrc_seq, to be log record sequence number. See zil.h 921b19a79ecSperrin * Then copy the record to the log buffer. 922b19a79ecSperrin */ 923b19a79ecSperrin lrc->lrc_seq = ++zilog->zl_lr_seq; /* we are single threaded */ 924fa9e4066Sahrens bcopy(lrc, lwb->lwb_buf + lwb->lwb_nused, reclen); 925c5c6ffa0Smaybee 926c5c6ffa0Smaybee /* 927c5c6ffa0Smaybee * If it's a write, fetch the data or get its blkptr as appropriate. 928c5c6ffa0Smaybee */ 929c5c6ffa0Smaybee if (lrc->lrc_txtype == TX_WRITE) { 930c5c6ffa0Smaybee if (txg > spa_freeze_txg(zilog->zl_spa)) 931c5c6ffa0Smaybee txg_wait_synced(zilog->zl_dmu_pool, txg); 932c5c6ffa0Smaybee if (itx->itx_wr_state != WR_COPIED) { 933c5c6ffa0Smaybee char *dbuf; 934c5c6ffa0Smaybee int error; 935c5c6ffa0Smaybee 936c5c6ffa0Smaybee /* alignment is guaranteed */ 937c5c6ffa0Smaybee lr = (lr_write_t *)(lwb->lwb_buf + lwb->lwb_nused); 938c5c6ffa0Smaybee if (dlen) { 939c5c6ffa0Smaybee ASSERT(itx->itx_wr_state == WR_NEED_COPY); 940c5c6ffa0Smaybee dbuf = lwb->lwb_buf + lwb->lwb_nused + reclen; 941c5c6ffa0Smaybee lr->lr_common.lrc_reclen += dlen; 942c5c6ffa0Smaybee } else { 943c5c6ffa0Smaybee ASSERT(itx->itx_wr_state == WR_INDIRECT); 944c5c6ffa0Smaybee dbuf = NULL; 945c5c6ffa0Smaybee } 946c5c6ffa0Smaybee error = zilog->zl_get_data( 947c5c6ffa0Smaybee itx->itx_private, lr, dbuf, lwb->lwb_zio); 948c5c6ffa0Smaybee if (error) { 949c5c6ffa0Smaybee ASSERT(error == ENOENT || error == EEXIST || 950c5c6ffa0Smaybee error == EALREADY); 951c5c6ffa0Smaybee return (lwb); 952c5c6ffa0Smaybee } 953c5c6ffa0Smaybee } 954104e2ed7Sperrin } 955c5c6ffa0Smaybee 956c5c6ffa0Smaybee lwb->lwb_nused += reclen + dlen; 957fa9e4066Sahrens lwb->lwb_max_txg = MAX(lwb->lwb_max_txg, txg); 958fa9e4066Sahrens ASSERT3U(lwb->lwb_nused, <=, ZIL_BLK_DATA_SZ(lwb)); 959fa9e4066Sahrens ASSERT3U(P2PHASE(lwb->lwb_nused, sizeof (uint64_t)), ==, 0); 960fa9e4066Sahrens 961fa9e4066Sahrens return (lwb); 962fa9e4066Sahrens } 963fa9e4066Sahrens 964fa9e4066Sahrens itx_t * 965da6c28aaSamw zil_itx_create(uint64_t txtype, size_t lrsize) 966fa9e4066Sahrens { 967fa9e4066Sahrens itx_t *itx; 968fa9e4066Sahrens 969b4d654b0Sperrin lrsize = P2ROUNDUP_TYPED(lrsize, sizeof (uint64_t), size_t); 970fa9e4066Sahrens 971fa9e4066Sahrens itx = kmem_alloc(offsetof(itx_t, itx_lr) + lrsize, KM_SLEEP); 972fa9e4066Sahrens itx->itx_lr.lrc_txtype = txtype; 973fa9e4066Sahrens itx->itx_lr.lrc_reclen = lrsize; 974abf76b6eSperrin itx->itx_sod = lrsize; /* if write & WR_NEED_COPY will be increased */ 975fa9e4066Sahrens itx->itx_lr.lrc_seq = 0; /* defensive */ 976fa9e4066Sahrens 977fa9e4066Sahrens return (itx); 978fa9e4066Sahrens } 979fa9e4066Sahrens 980fa9e4066Sahrens uint64_t 981fa9e4066Sahrens zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx) 982fa9e4066Sahrens { 983fa9e4066Sahrens uint64_t seq; 984fa9e4066Sahrens 985fa9e4066Sahrens ASSERT(itx->itx_lr.lrc_seq == 0); 986fa9e4066Sahrens 987fa9e4066Sahrens mutex_enter(&zilog->zl_lock); 988fa9e4066Sahrens list_insert_tail(&zilog->zl_itx_list, itx); 989abf76b6eSperrin zilog->zl_itx_list_sz += itx->itx_sod; 990fa9e4066Sahrens itx->itx_lr.lrc_txg = dmu_tx_get_txg(tx); 991fa9e4066Sahrens itx->itx_lr.lrc_seq = seq = ++zilog->zl_itx_seq; 992fa9e4066Sahrens mutex_exit(&zilog->zl_lock); 993fa9e4066Sahrens 994fa9e4066Sahrens return (seq); 995fa9e4066Sahrens } 996fa9e4066Sahrens 997fa9e4066Sahrens /* 998fa9e4066Sahrens * Free up all in-memory intent log transactions that have now been synced. 999fa9e4066Sahrens */ 1000fa9e4066Sahrens static void 1001fa9e4066Sahrens zil_itx_clean(zilog_t *zilog) 1002fa9e4066Sahrens { 1003fa9e4066Sahrens uint64_t synced_txg = spa_last_synced_txg(zilog->zl_spa); 1004fa9e4066Sahrens uint64_t freeze_txg = spa_freeze_txg(zilog->zl_spa); 1005a584ef65Sjohansen list_t clean_list; 1006fa9e4066Sahrens itx_t *itx; 1007fa9e4066Sahrens 1008a584ef65Sjohansen list_create(&clean_list, sizeof (itx_t), offsetof(itx_t, itx_node)); 1009a584ef65Sjohansen 1010fa9e4066Sahrens mutex_enter(&zilog->zl_lock); 1011b19a79ecSperrin /* wait for a log writer to finish walking list */ 1012b19a79ecSperrin while (zilog->zl_writer) { 1013b19a79ecSperrin cv_wait(&zilog->zl_cv_writer, &zilog->zl_lock); 1014b19a79ecSperrin } 1015a584ef65Sjohansen 1016a584ef65Sjohansen /* 1017a584ef65Sjohansen * Move the sync'd log transactions to a separate list so we can call 1018a584ef65Sjohansen * kmem_free without holding the zl_lock. 1019a584ef65Sjohansen * 1020a584ef65Sjohansen * There is no need to set zl_writer as we don't drop zl_lock here 1021a584ef65Sjohansen */ 1022fa9e4066Sahrens while ((itx = list_head(&zilog->zl_itx_list)) != NULL && 1023fa9e4066Sahrens itx->itx_lr.lrc_txg <= MIN(synced_txg, freeze_txg)) { 1024fa9e4066Sahrens list_remove(&zilog->zl_itx_list, itx); 1025abf76b6eSperrin zilog->zl_itx_list_sz -= itx->itx_sod; 1026a584ef65Sjohansen list_insert_tail(&clean_list, itx); 1027fa9e4066Sahrens } 102867bd71c6Sperrin cv_broadcast(&zilog->zl_cv_writer); 1029fa9e4066Sahrens mutex_exit(&zilog->zl_lock); 1030a584ef65Sjohansen 1031a584ef65Sjohansen /* destroy sync'd log transactions */ 1032a584ef65Sjohansen while ((itx = list_head(&clean_list)) != NULL) { 1033a584ef65Sjohansen list_remove(&clean_list, itx); 1034a584ef65Sjohansen kmem_free(itx, offsetof(itx_t, itx_lr) 1035a584ef65Sjohansen + itx->itx_lr.lrc_reclen); 1036a584ef65Sjohansen } 1037a584ef65Sjohansen list_destroy(&clean_list); 1038fa9e4066Sahrens } 1039fa9e4066Sahrens 1040b19a79ecSperrin /* 104167bd71c6Sperrin * If there are any in-memory intent log transactions which have now been 104267bd71c6Sperrin * synced then start up a taskq to free them. 1043b19a79ecSperrin */ 1044fa9e4066Sahrens void 1045fa9e4066Sahrens zil_clean(zilog_t *zilog) 1046fa9e4066Sahrens { 104767bd71c6Sperrin itx_t *itx; 104867bd71c6Sperrin 1049fa9e4066Sahrens mutex_enter(&zilog->zl_lock); 105067bd71c6Sperrin itx = list_head(&zilog->zl_itx_list); 105167bd71c6Sperrin if ((itx != NULL) && 105267bd71c6Sperrin (itx->itx_lr.lrc_txg <= spa_last_synced_txg(zilog->zl_spa))) { 1053fa9e4066Sahrens (void) taskq_dispatch(zilog->zl_clean_taskq, 10549d3574bfSNeil Perrin (task_func_t *)zil_itx_clean, zilog, TQ_SLEEP); 105567bd71c6Sperrin } 1056fa9e4066Sahrens mutex_exit(&zilog->zl_lock); 1057fa9e4066Sahrens } 1058fa9e4066Sahrens 1059e14bb325SJeff Bonwick static void 1060b19a79ecSperrin zil_commit_writer(zilog_t *zilog, uint64_t seq, uint64_t foid) 1061fa9e4066Sahrens { 1062fa9e4066Sahrens uint64_t txg; 106367bd71c6Sperrin uint64_t commit_seq = 0; 1064b19a79ecSperrin itx_t *itx, *itx_next = (itx_t *)-1; 1065fa9e4066Sahrens lwb_t *lwb; 1066fa9e4066Sahrens spa_t *spa; 1067fa9e4066Sahrens 1068fa9e4066Sahrens zilog->zl_writer = B_TRUE; 1069e14bb325SJeff Bonwick ASSERT(zilog->zl_root_zio == NULL); 1070b19a79ecSperrin spa = zilog->zl_spa; 1071fa9e4066Sahrens 1072fa9e4066Sahrens if (zilog->zl_suspend) { 1073fa9e4066Sahrens lwb = NULL; 1074fa9e4066Sahrens } else { 1075fa9e4066Sahrens lwb = list_tail(&zilog->zl_lwb_list); 1076fa9e4066Sahrens if (lwb == NULL) { 1077b19a79ecSperrin /* 1078b19a79ecSperrin * Return if there's nothing to flush before we 1079b19a79ecSperrin * dirty the fs by calling zil_create() 1080b19a79ecSperrin */ 1081b19a79ecSperrin if (list_is_empty(&zilog->zl_itx_list)) { 1082b19a79ecSperrin zilog->zl_writer = B_FALSE; 1083b19a79ecSperrin return; 1084b19a79ecSperrin } 1085fa9e4066Sahrens mutex_exit(&zilog->zl_lock); 1086fa9e4066Sahrens zil_create(zilog); 1087fa9e4066Sahrens mutex_enter(&zilog->zl_lock); 1088fa9e4066Sahrens lwb = list_tail(&zilog->zl_lwb_list); 1089fa9e4066Sahrens } 1090fa9e4066Sahrens } 1091fa9e4066Sahrens 109267bd71c6Sperrin /* Loop through in-memory log transactions filling log blocks. */ 1093b19a79ecSperrin DTRACE_PROBE1(zil__cw1, zilog_t *, zilog); 1094fa9e4066Sahrens for (;;) { 1095b19a79ecSperrin /* 1096b19a79ecSperrin * Find the next itx to push: 1097b19a79ecSperrin * Push all transactions related to specified foid and all 1098b19a79ecSperrin * other transactions except TX_WRITE, TX_TRUNCATE, 1099b19a79ecSperrin * TX_SETATTR and TX_ACL for all other files. 1100b19a79ecSperrin */ 1101b19a79ecSperrin if (itx_next != (itx_t *)-1) 1102b19a79ecSperrin itx = itx_next; 1103b19a79ecSperrin else 1104b19a79ecSperrin itx = list_head(&zilog->zl_itx_list); 1105b19a79ecSperrin for (; itx != NULL; itx = list_next(&zilog->zl_itx_list, itx)) { 1106b19a79ecSperrin if (foid == 0) /* push all foids? */ 1107b19a79ecSperrin break; 110867bd71c6Sperrin if (itx->itx_sync) /* push all O_[D]SYNC */ 110967bd71c6Sperrin break; 1110b19a79ecSperrin switch (itx->itx_lr.lrc_txtype) { 1111b19a79ecSperrin case TX_SETATTR: 1112b19a79ecSperrin case TX_WRITE: 1113b19a79ecSperrin case TX_TRUNCATE: 1114b19a79ecSperrin case TX_ACL: 1115b19a79ecSperrin /* lr_foid is same offset for these records */ 1116b19a79ecSperrin if (((lr_write_t *)&itx->itx_lr)->lr_foid 1117b19a79ecSperrin != foid) { 1118b19a79ecSperrin continue; /* skip this record */ 1119b19a79ecSperrin } 1120b19a79ecSperrin } 1121b19a79ecSperrin break; 1122b19a79ecSperrin } 1123fa9e4066Sahrens if (itx == NULL) 1124fa9e4066Sahrens break; 1125fa9e4066Sahrens 1126fa9e4066Sahrens if ((itx->itx_lr.lrc_seq > seq) && 1127b19a79ecSperrin ((lwb == NULL) || (lwb->lwb_nused == 0) || 1128abf76b6eSperrin (lwb->lwb_nused + itx->itx_sod > ZIL_BLK_DATA_SZ(lwb)))) { 1129fa9e4066Sahrens break; 113067bd71c6Sperrin } 1131fa9e4066Sahrens 1132b19a79ecSperrin /* 1133b19a79ecSperrin * Save the next pointer. Even though we soon drop 1134b19a79ecSperrin * zl_lock all threads that may change the list 1135b19a79ecSperrin * (another writer or zil_itx_clean) can't do so until 1136b19a79ecSperrin * they have zl_writer. 1137b19a79ecSperrin */ 1138b19a79ecSperrin itx_next = list_next(&zilog->zl_itx_list, itx); 1139fa9e4066Sahrens list_remove(&zilog->zl_itx_list, itx); 1140abf76b6eSperrin zilog->zl_itx_list_sz -= itx->itx_sod; 114167bd71c6Sperrin mutex_exit(&zilog->zl_lock); 1142fa9e4066Sahrens txg = itx->itx_lr.lrc_txg; 1143fa9e4066Sahrens ASSERT(txg); 1144fa9e4066Sahrens 1145fa9e4066Sahrens if (txg > spa_last_synced_txg(spa) || 1146fa9e4066Sahrens txg > spa_freeze_txg(spa)) 1147fa9e4066Sahrens lwb = zil_lwb_commit(zilog, itx, lwb); 1148fa9e4066Sahrens kmem_free(itx, offsetof(itx_t, itx_lr) 1149fa9e4066Sahrens + itx->itx_lr.lrc_reclen); 1150fa9e4066Sahrens mutex_enter(&zilog->zl_lock); 1151fa9e4066Sahrens } 1152b19a79ecSperrin DTRACE_PROBE1(zil__cw2, zilog_t *, zilog); 115367bd71c6Sperrin /* determine commit sequence number */ 115467bd71c6Sperrin itx = list_head(&zilog->zl_itx_list); 115567bd71c6Sperrin if (itx) 115667bd71c6Sperrin commit_seq = itx->itx_lr.lrc_seq; 115767bd71c6Sperrin else 115867bd71c6Sperrin commit_seq = zilog->zl_itx_seq; 1159fa9e4066Sahrens mutex_exit(&zilog->zl_lock); 1160fa9e4066Sahrens 1161fa9e4066Sahrens /* write the last block out */ 116267bd71c6Sperrin if (lwb != NULL && lwb->lwb_zio != NULL) 1163fa9e4066Sahrens lwb = zil_lwb_write_start(zilog, lwb); 1164fa9e4066Sahrens 116522ac5be4Sperrin zilog->zl_prev_used = zilog->zl_cur_used; 116622ac5be4Sperrin zilog->zl_cur_used = 0; 1167fa9e4066Sahrens 1168fa9e4066Sahrens /* 1169b19a79ecSperrin * Wait if necessary for the log blocks to be on stable storage. 1170fa9e4066Sahrens */ 1171b19a79ecSperrin if (zilog->zl_root_zio) { 1172b19a79ecSperrin DTRACE_PROBE1(zil__cw3, zilog_t *, zilog); 1173b19a79ecSperrin (void) zio_wait(zilog->zl_root_zio); 1174e14bb325SJeff Bonwick zilog->zl_root_zio = NULL; 1175b19a79ecSperrin DTRACE_PROBE1(zil__cw4, zilog_t *, zilog); 117617f17c2dSbonwick zil_flush_vdevs(zilog); 1177fa9e4066Sahrens } 117822ac5be4Sperrin 1179fa9e4066Sahrens if (zilog->zl_log_error || lwb == NULL) { 1180fa9e4066Sahrens zilog->zl_log_error = 0; 1181fa9e4066Sahrens txg_wait_synced(zilog->zl_dmu_pool, 0); 1182fa9e4066Sahrens } 118367bd71c6Sperrin 118467bd71c6Sperrin mutex_enter(&zilog->zl_lock); 118522ac5be4Sperrin zilog->zl_writer = B_FALSE; 118667bd71c6Sperrin 118767bd71c6Sperrin ASSERT3U(commit_seq, >=, zilog->zl_commit_seq); 118867bd71c6Sperrin zilog->zl_commit_seq = commit_seq; 1189b19a79ecSperrin } 1190b19a79ecSperrin 1191b19a79ecSperrin /* 1192b19a79ecSperrin * Push zfs transactions to stable storage up to the supplied sequence number. 1193b19a79ecSperrin * If foid is 0 push out all transactions, otherwise push only those 1194b19a79ecSperrin * for that file or might have been used to create that file. 1195b19a79ecSperrin */ 1196b19a79ecSperrin void 1197b19a79ecSperrin zil_commit(zilog_t *zilog, uint64_t seq, uint64_t foid) 1198b19a79ecSperrin { 1199b19a79ecSperrin if (zilog == NULL || seq == 0) 1200b19a79ecSperrin return; 1201b19a79ecSperrin 1202b19a79ecSperrin mutex_enter(&zilog->zl_lock); 1203b19a79ecSperrin 1204b19a79ecSperrin seq = MIN(seq, zilog->zl_itx_seq); /* cap seq at largest itx seq */ 1205b19a79ecSperrin 120667bd71c6Sperrin while (zilog->zl_writer) { 1207b19a79ecSperrin cv_wait(&zilog->zl_cv_writer, &zilog->zl_lock); 120867bd71c6Sperrin if (seq < zilog->zl_commit_seq) { 120967bd71c6Sperrin mutex_exit(&zilog->zl_lock); 121067bd71c6Sperrin return; 121167bd71c6Sperrin } 121267bd71c6Sperrin } 1213b19a79ecSperrin zil_commit_writer(zilog, seq, foid); /* drops zl_lock */ 121467bd71c6Sperrin /* wake up others waiting on the commit */ 121567bd71c6Sperrin cv_broadcast(&zilog->zl_cv_writer); 121667bd71c6Sperrin mutex_exit(&zilog->zl_lock); 1217fa9e4066Sahrens } 1218fa9e4066Sahrens 1219fa9e4066Sahrens /* 1220fa9e4066Sahrens * Called in syncing context to free committed log blocks and update log header. 1221fa9e4066Sahrens */ 1222fa9e4066Sahrens void 1223fa9e4066Sahrens zil_sync(zilog_t *zilog, dmu_tx_t *tx) 1224fa9e4066Sahrens { 1225d80c45e0Sbonwick zil_header_t *zh = zil_header_in_syncing_context(zilog); 1226fa9e4066Sahrens uint64_t txg = dmu_tx_get_txg(tx); 1227fa9e4066Sahrens spa_t *spa = zilog->zl_spa; 1228fa9e4066Sahrens lwb_t *lwb; 1229fa9e4066Sahrens 123014843421SMatthew Ahrens /* 123114843421SMatthew Ahrens * We don't zero out zl_destroy_txg, so make sure we don't try 123214843421SMatthew Ahrens * to destroy it twice. 123314843421SMatthew Ahrens */ 123414843421SMatthew Ahrens if (spa_sync_pass(spa) != 1) 123514843421SMatthew Ahrens return; 123614843421SMatthew Ahrens 1237d80c45e0Sbonwick mutex_enter(&zilog->zl_lock); 1238d80c45e0Sbonwick 1239fa9e4066Sahrens ASSERT(zilog->zl_stop_sync == 0); 1240fa9e4066Sahrens 12411209a471SNeil Perrin zh->zh_replay_seq = zilog->zl_replayed_seq[txg & TXG_MASK]; 1242fa9e4066Sahrens 1243fa9e4066Sahrens if (zilog->zl_destroy_txg == txg) { 1244d80c45e0Sbonwick blkptr_t blk = zh->zh_log; 1245d80c45e0Sbonwick 1246d80c45e0Sbonwick ASSERT(list_head(&zilog->zl_lwb_list) == NULL); 1247d80c45e0Sbonwick 1248d80c45e0Sbonwick bzero(zh, sizeof (zil_header_t)); 12491209a471SNeil Perrin bzero(zilog->zl_replayed_seq, sizeof (zilog->zl_replayed_seq)); 1250d80c45e0Sbonwick 1251d80c45e0Sbonwick if (zilog->zl_keep_first) { 1252d80c45e0Sbonwick /* 1253d80c45e0Sbonwick * If this block was part of log chain that couldn't 1254d80c45e0Sbonwick * be claimed because a device was missing during 1255d80c45e0Sbonwick * zil_claim(), but that device later returns, 1256d80c45e0Sbonwick * then this block could erroneously appear valid. 1257d80c45e0Sbonwick * To guard against this, assign a new GUID to the new 1258d80c45e0Sbonwick * log chain so it doesn't matter what blk points to. 1259d80c45e0Sbonwick */ 1260d80c45e0Sbonwick zil_init_log_chain(zilog, &blk); 1261d80c45e0Sbonwick zh->zh_log = blk; 1262d80c45e0Sbonwick } 1263fa9e4066Sahrens } 1264fa9e4066Sahrens 1265fa9e4066Sahrens for (;;) { 1266fa9e4066Sahrens lwb = list_head(&zilog->zl_lwb_list); 1267fa9e4066Sahrens if (lwb == NULL) { 1268fa9e4066Sahrens mutex_exit(&zilog->zl_lock); 1269fa9e4066Sahrens return; 1270fa9e4066Sahrens } 1271b19a79ecSperrin zh->zh_log = lwb->lwb_blk; 1272fa9e4066Sahrens if (lwb->lwb_buf != NULL || lwb->lwb_max_txg > txg) 1273fa9e4066Sahrens break; 1274fa9e4066Sahrens list_remove(&zilog->zl_lwb_list, lwb); 1275fa9e4066Sahrens zio_free_blk(spa, &lwb->lwb_blk, txg); 1276fa9e4066Sahrens kmem_cache_free(zil_lwb_cache, lwb); 1277d63d470bSgw 1278d63d470bSgw /* 1279d63d470bSgw * If we don't have anything left in the lwb list then 1280d63d470bSgw * we've had an allocation failure and we need to zero 1281d63d470bSgw * out the zil_header blkptr so that we don't end 1282d63d470bSgw * up freeing the same block twice. 1283d63d470bSgw */ 1284d63d470bSgw if (list_head(&zilog->zl_lwb_list) == NULL) 1285d63d470bSgw BP_ZERO(&zh->zh_log); 1286fa9e4066Sahrens } 1287fa9e4066Sahrens mutex_exit(&zilog->zl_lock); 1288fa9e4066Sahrens } 1289fa9e4066Sahrens 1290fa9e4066Sahrens void 1291fa9e4066Sahrens zil_init(void) 1292fa9e4066Sahrens { 1293fa9e4066Sahrens zil_lwb_cache = kmem_cache_create("zil_lwb_cache", 12945ad82045Snd sizeof (struct lwb), 0, NULL, NULL, NULL, NULL, NULL, 0); 1295fa9e4066Sahrens } 1296fa9e4066Sahrens 1297fa9e4066Sahrens void 1298fa9e4066Sahrens zil_fini(void) 1299fa9e4066Sahrens { 1300fa9e4066Sahrens kmem_cache_destroy(zil_lwb_cache); 1301fa9e4066Sahrens } 1302fa9e4066Sahrens 1303fa9e4066Sahrens zilog_t * 1304fa9e4066Sahrens zil_alloc(objset_t *os, zil_header_t *zh_phys) 1305fa9e4066Sahrens { 1306fa9e4066Sahrens zilog_t *zilog; 1307fa9e4066Sahrens 1308fa9e4066Sahrens zilog = kmem_zalloc(sizeof (zilog_t), KM_SLEEP); 1309fa9e4066Sahrens 1310fa9e4066Sahrens zilog->zl_header = zh_phys; 1311fa9e4066Sahrens zilog->zl_os = os; 1312fa9e4066Sahrens zilog->zl_spa = dmu_objset_spa(os); 1313fa9e4066Sahrens zilog->zl_dmu_pool = dmu_objset_pool(os); 1314d80c45e0Sbonwick zilog->zl_destroy_txg = TXG_INITIAL - 1; 1315fa9e4066Sahrens 13165ad82045Snd mutex_init(&zilog->zl_lock, NULL, MUTEX_DEFAULT, NULL); 13175ad82045Snd 1318fa9e4066Sahrens list_create(&zilog->zl_itx_list, sizeof (itx_t), 1319fa9e4066Sahrens offsetof(itx_t, itx_node)); 1320fa9e4066Sahrens 1321fa9e4066Sahrens list_create(&zilog->zl_lwb_list, sizeof (lwb_t), 1322fa9e4066Sahrens offsetof(lwb_t, lwb_node)); 1323fa9e4066Sahrens 132417f17c2dSbonwick mutex_init(&zilog->zl_vdev_lock, NULL, MUTEX_DEFAULT, NULL); 132517f17c2dSbonwick 132617f17c2dSbonwick avl_create(&zilog->zl_vdev_tree, zil_vdev_compare, 132717f17c2dSbonwick sizeof (zil_vdev_node_t), offsetof(zil_vdev_node_t, zv_node)); 1328fa9e4066Sahrens 1329b7b97454Sperrin cv_init(&zilog->zl_cv_writer, NULL, CV_DEFAULT, NULL); 1330b7b97454Sperrin cv_init(&zilog->zl_cv_suspend, NULL, CV_DEFAULT, NULL); 1331b7b97454Sperrin 1332fa9e4066Sahrens return (zilog); 1333fa9e4066Sahrens } 1334fa9e4066Sahrens 1335fa9e4066Sahrens void 1336fa9e4066Sahrens zil_free(zilog_t *zilog) 1337fa9e4066Sahrens { 1338fa9e4066Sahrens lwb_t *lwb; 1339fa9e4066Sahrens 1340fa9e4066Sahrens zilog->zl_stop_sync = 1; 1341fa9e4066Sahrens 1342fa9e4066Sahrens while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) { 1343fa9e4066Sahrens list_remove(&zilog->zl_lwb_list, lwb); 1344fa9e4066Sahrens if (lwb->lwb_buf != NULL) 1345fa9e4066Sahrens zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); 1346fa9e4066Sahrens kmem_cache_free(zil_lwb_cache, lwb); 1347fa9e4066Sahrens } 1348fa9e4066Sahrens list_destroy(&zilog->zl_lwb_list); 1349fa9e4066Sahrens 135017f17c2dSbonwick avl_destroy(&zilog->zl_vdev_tree); 135117f17c2dSbonwick mutex_destroy(&zilog->zl_vdev_lock); 1352fa9e4066Sahrens 1353fa9e4066Sahrens ASSERT(list_head(&zilog->zl_itx_list) == NULL); 1354fa9e4066Sahrens list_destroy(&zilog->zl_itx_list); 13555ad82045Snd mutex_destroy(&zilog->zl_lock); 1356fa9e4066Sahrens 1357b7b97454Sperrin cv_destroy(&zilog->zl_cv_writer); 1358b7b97454Sperrin cv_destroy(&zilog->zl_cv_suspend); 1359b7b97454Sperrin 1360fa9e4066Sahrens kmem_free(zilog, sizeof (zilog_t)); 1361fa9e4066Sahrens } 1362fa9e4066Sahrens 1363fa9e4066Sahrens /* 1364fa9e4066Sahrens * Open an intent log. 1365fa9e4066Sahrens */ 1366fa9e4066Sahrens zilog_t * 1367fa9e4066Sahrens zil_open(objset_t *os, zil_get_data_t *get_data) 1368fa9e4066Sahrens { 1369fa9e4066Sahrens zilog_t *zilog = dmu_objset_zil(os); 1370fa9e4066Sahrens 1371fa9e4066Sahrens zilog->zl_get_data = get_data; 1372fa9e4066Sahrens zilog->zl_clean_taskq = taskq_create("zil_clean", 1, minclsyspri, 1373fa9e4066Sahrens 2, 2, TASKQ_PREPOPULATE); 1374fa9e4066Sahrens 1375fa9e4066Sahrens return (zilog); 1376fa9e4066Sahrens } 1377fa9e4066Sahrens 1378fa9e4066Sahrens /* 1379fa9e4066Sahrens * Close an intent log. 1380fa9e4066Sahrens */ 1381fa9e4066Sahrens void 1382fa9e4066Sahrens zil_close(zilog_t *zilog) 1383fa9e4066Sahrens { 1384d80c45e0Sbonwick /* 1385d80c45e0Sbonwick * If the log isn't already committed, mark the objset dirty 1386d80c45e0Sbonwick * (so zil_sync() will be called) and wait for that txg to sync. 1387d80c45e0Sbonwick */ 1388d80c45e0Sbonwick if (!zil_is_committed(zilog)) { 1389d80c45e0Sbonwick uint64_t txg; 1390d80c45e0Sbonwick dmu_tx_t *tx = dmu_tx_create(zilog->zl_os); 1391d80c45e0Sbonwick (void) dmu_tx_assign(tx, TXG_WAIT); 1392d80c45e0Sbonwick dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); 1393d80c45e0Sbonwick txg = dmu_tx_get_txg(tx); 1394d80c45e0Sbonwick dmu_tx_commit(tx); 1395d80c45e0Sbonwick txg_wait_synced(zilog->zl_dmu_pool, txg); 1396d80c45e0Sbonwick } 1397d80c45e0Sbonwick 1398fa9e4066Sahrens taskq_destroy(zilog->zl_clean_taskq); 1399fa9e4066Sahrens zilog->zl_clean_taskq = NULL; 1400fa9e4066Sahrens zilog->zl_get_data = NULL; 1401fa9e4066Sahrens 1402fa9e4066Sahrens zil_itx_clean(zilog); 1403fa9e4066Sahrens ASSERT(list_head(&zilog->zl_itx_list) == NULL); 1404fa9e4066Sahrens } 1405fa9e4066Sahrens 1406fa9e4066Sahrens /* 1407fa9e4066Sahrens * Suspend an intent log. While in suspended mode, we still honor 1408fa9e4066Sahrens * synchronous semantics, but we rely on txg_wait_synced() to do it. 1409fa9e4066Sahrens * We suspend the log briefly when taking a snapshot so that the snapshot 1410fa9e4066Sahrens * contains all the data it's supposed to, and has an empty intent log. 1411fa9e4066Sahrens */ 1412fa9e4066Sahrens int 1413fa9e4066Sahrens zil_suspend(zilog_t *zilog) 1414fa9e4066Sahrens { 1415d80c45e0Sbonwick const zil_header_t *zh = zilog->zl_header; 1416fa9e4066Sahrens 1417fa9e4066Sahrens mutex_enter(&zilog->zl_lock); 14183589c4f0SNeil Perrin if (zh->zh_flags & ZIL_REPLAY_NEEDED) { /* unplayed log */ 1419fa9e4066Sahrens mutex_exit(&zilog->zl_lock); 1420fa9e4066Sahrens return (EBUSY); 1421fa9e4066Sahrens } 1422d80c45e0Sbonwick if (zilog->zl_suspend++ != 0) { 1423d80c45e0Sbonwick /* 1424d80c45e0Sbonwick * Someone else already began a suspend. 1425d80c45e0Sbonwick * Just wait for them to finish. 1426d80c45e0Sbonwick */ 1427d80c45e0Sbonwick while (zilog->zl_suspending) 1428d80c45e0Sbonwick cv_wait(&zilog->zl_cv_suspend, &zilog->zl_lock); 1429d80c45e0Sbonwick mutex_exit(&zilog->zl_lock); 1430d80c45e0Sbonwick return (0); 1431d80c45e0Sbonwick } 1432d80c45e0Sbonwick zilog->zl_suspending = B_TRUE; 1433fa9e4066Sahrens mutex_exit(&zilog->zl_lock); 1434fa9e4066Sahrens 1435b19a79ecSperrin zil_commit(zilog, UINT64_MAX, 0); 1436fa9e4066Sahrens 1437b19a79ecSperrin /* 1438b19a79ecSperrin * Wait for any in-flight log writes to complete. 1439b19a79ecSperrin */ 1440fa9e4066Sahrens mutex_enter(&zilog->zl_lock); 1441b19a79ecSperrin while (zilog->zl_writer) 1442b19a79ecSperrin cv_wait(&zilog->zl_cv_writer, &zilog->zl_lock); 1443fa9e4066Sahrens mutex_exit(&zilog->zl_lock); 1444fa9e4066Sahrens 1445d80c45e0Sbonwick zil_destroy(zilog, B_FALSE); 1446d80c45e0Sbonwick 1447d80c45e0Sbonwick mutex_enter(&zilog->zl_lock); 1448d80c45e0Sbonwick zilog->zl_suspending = B_FALSE; 1449d80c45e0Sbonwick cv_broadcast(&zilog->zl_cv_suspend); 1450d80c45e0Sbonwick mutex_exit(&zilog->zl_lock); 1451fa9e4066Sahrens 1452fa9e4066Sahrens return (0); 1453fa9e4066Sahrens } 1454fa9e4066Sahrens 1455fa9e4066Sahrens void 1456fa9e4066Sahrens zil_resume(zilog_t *zilog) 1457fa9e4066Sahrens { 1458fa9e4066Sahrens mutex_enter(&zilog->zl_lock); 1459fa9e4066Sahrens ASSERT(zilog->zl_suspend != 0); 1460fa9e4066Sahrens zilog->zl_suspend--; 1461fa9e4066Sahrens mutex_exit(&zilog->zl_lock); 1462fa9e4066Sahrens } 1463fa9e4066Sahrens 1464fa9e4066Sahrens typedef struct zil_replay_arg { 1465fa9e4066Sahrens objset_t *zr_os; 1466fa9e4066Sahrens zil_replay_func_t **zr_replay; 1467fa9e4066Sahrens void *zr_arg; 1468fa9e4066Sahrens boolean_t zr_byteswap; 1469fa9e4066Sahrens char *zr_lrbuf; 1470fa9e4066Sahrens } zil_replay_arg_t; 1471fa9e4066Sahrens 1472fa9e4066Sahrens static void 1473fa9e4066Sahrens zil_replay_log_record(zilog_t *zilog, lr_t *lr, void *zra, uint64_t claim_txg) 1474fa9e4066Sahrens { 1475fa9e4066Sahrens zil_replay_arg_t *zr = zra; 1476d80c45e0Sbonwick const zil_header_t *zh = zilog->zl_header; 1477fa9e4066Sahrens uint64_t reclen = lr->lrc_reclen; 1478fa9e4066Sahrens uint64_t txtype = lr->lrc_txtype; 147967bd71c6Sperrin char *name; 14801209a471SNeil Perrin int pass, error; 1481fa9e4066Sahrens 14821209a471SNeil Perrin if (!zilog->zl_replay) /* giving up */ 1483fa9e4066Sahrens return; 1484fa9e4066Sahrens 1485fa9e4066Sahrens if (lr->lrc_txg < claim_txg) /* already committed */ 1486fa9e4066Sahrens return; 1487fa9e4066Sahrens 1488fa9e4066Sahrens if (lr->lrc_seq <= zh->zh_replay_seq) /* already replayed */ 1489fa9e4066Sahrens return; 1490fa9e4066Sahrens 1491da6c28aaSamw /* Strip case-insensitive bit, still present in log record */ 1492da6c28aaSamw txtype &= ~TX_CI; 1493da6c28aaSamw 14941209a471SNeil Perrin if (txtype == 0 || txtype >= TX_MAX_TYPE) { 14951209a471SNeil Perrin error = EINVAL; 14961209a471SNeil Perrin goto bad; 14971209a471SNeil Perrin } 14981209a471SNeil Perrin 1499fa9e4066Sahrens /* 1500fa9e4066Sahrens * Make a copy of the data so we can revise and extend it. 1501fa9e4066Sahrens */ 1502fa9e4066Sahrens bcopy(lr, zr->zr_lrbuf, reclen); 1503fa9e4066Sahrens 1504fa9e4066Sahrens /* 1505fa9e4066Sahrens * The log block containing this lr may have been byteswapped 1506fa9e4066Sahrens * so that we can easily examine common fields like lrc_txtype. 1507fa9e4066Sahrens * However, the log is a mix of different data types, and only the 1508fa9e4066Sahrens * replay vectors know how to byteswap their records. Therefore, if 1509fa9e4066Sahrens * the lr was byteswapped, undo it before invoking the replay vector. 1510fa9e4066Sahrens */ 1511fa9e4066Sahrens if (zr->zr_byteswap) 1512fa9e4066Sahrens byteswap_uint64_array(zr->zr_lrbuf, reclen); 1513fa9e4066Sahrens 1514fa9e4066Sahrens /* 1515fa9e4066Sahrens * If this is a TX_WRITE with a blkptr, suck in the data. 1516fa9e4066Sahrens */ 1517fa9e4066Sahrens if (txtype == TX_WRITE && reclen == sizeof (lr_write_t)) { 1518fa9e4066Sahrens lr_write_t *lrw = (lr_write_t *)lr; 1519fa9e4066Sahrens blkptr_t *wbp = &lrw->lr_blkptr; 1520fa9e4066Sahrens uint64_t wlen = lrw->lr_length; 1521fa9e4066Sahrens char *wbuf = zr->zr_lrbuf + reclen; 1522fa9e4066Sahrens 1523fa9e4066Sahrens if (BP_IS_HOLE(wbp)) { /* compressed to a hole */ 1524fa9e4066Sahrens bzero(wbuf, wlen); 1525fa9e4066Sahrens } else { 1526fa9e4066Sahrens /* 1527fa9e4066Sahrens * A subsequent write may have overwritten this block, 1528fa9e4066Sahrens * in which case wbp may have been been freed and 1529fa9e4066Sahrens * reallocated, and our read of wbp may fail with a 1530fa9e4066Sahrens * checksum error. We can safely ignore this because 1531fa9e4066Sahrens * the later write will provide the correct data. 1532fa9e4066Sahrens */ 1533ea8dc4b6Seschrock zbookmark_t zb; 1534ea8dc4b6Seschrock 1535ea8dc4b6Seschrock zb.zb_objset = dmu_objset_id(zilog->zl_os); 1536ea8dc4b6Seschrock zb.zb_object = lrw->lr_foid; 1537ea8dc4b6Seschrock zb.zb_level = -1; 1538ea8dc4b6Seschrock zb.zb_blkid = lrw->lr_offset / BP_GET_LSIZE(wbp); 1539ea8dc4b6Seschrock 1540fa9e4066Sahrens (void) zio_wait(zio_read(NULL, zilog->zl_spa, 1541fa9e4066Sahrens wbp, wbuf, BP_GET_LSIZE(wbp), NULL, NULL, 1542fa9e4066Sahrens ZIO_PRIORITY_SYNC_READ, 1543ea8dc4b6Seschrock ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &zb)); 1544fa9e4066Sahrens (void) memmove(wbuf, wbuf + lrw->lr_blkoff, wlen); 1545fa9e4066Sahrens } 1546fa9e4066Sahrens } 1547fa9e4066Sahrens 1548fa9e4066Sahrens /* 1549fa9e4066Sahrens * We must now do two things atomically: replay this log record, 15501209a471SNeil Perrin * and update the log header sequence number to reflect the fact that 15511209a471SNeil Perrin * we did so. At the end of each replay function the sequence number 15521209a471SNeil Perrin * is updated if we are in replay mode. 1553fa9e4066Sahrens */ 15541209a471SNeil Perrin for (pass = 1; pass <= 2; pass++) { 15551209a471SNeil Perrin zilog->zl_replaying_seq = lr->lrc_seq; 15561209a471SNeil Perrin /* Only byteswap (if needed) on the 1st pass. */ 15571209a471SNeil Perrin error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lrbuf, 15581209a471SNeil Perrin zr->zr_byteswap && pass == 1); 1559fa9e4066Sahrens 156067bd71c6Sperrin if (!error) 156167bd71c6Sperrin return; 156267bd71c6Sperrin 156367bd71c6Sperrin /* 156467bd71c6Sperrin * The DMU's dnode layer doesn't see removes until the txg 156567bd71c6Sperrin * commits, so a subsequent claim can spuriously fail with 15661209a471SNeil Perrin * EEXIST. So if we receive any error we try syncing out 15671209a471SNeil Perrin * any removes then retry the transaction. 156867bd71c6Sperrin */ 15691209a471SNeil Perrin if (pass == 1) 157067bd71c6Sperrin txg_wait_synced(spa_get_dsl(zilog->zl_spa), 0); 1571fa9e4066Sahrens } 1572fa9e4066Sahrens 1573d30e856aSNeil Perrin bad: 15741209a471SNeil Perrin ASSERT(error); 157567bd71c6Sperrin name = kmem_alloc(MAXNAMELEN, KM_SLEEP); 157667bd71c6Sperrin dmu_objset_name(zr->zr_os, name); 157767bd71c6Sperrin cmn_err(CE_WARN, "ZFS replay transaction error %d, " 1578da6c28aaSamw "dataset %s, seq 0x%llx, txtype %llu %s\n", 1579da6c28aaSamw error, name, (u_longlong_t)lr->lrc_seq, (u_longlong_t)txtype, 1580da6c28aaSamw (lr->lrc_txtype & TX_CI) ? "CI" : ""); 15811209a471SNeil Perrin zilog->zl_replay = B_FALSE; 158267bd71c6Sperrin kmem_free(name, MAXNAMELEN); 158367bd71c6Sperrin } 1584fa9e4066Sahrens 158567bd71c6Sperrin /* ARGSUSED */ 158667bd71c6Sperrin static void 158767bd71c6Sperrin zil_incr_blks(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg) 158867bd71c6Sperrin { 158967bd71c6Sperrin zilog->zl_replay_blks++; 1590fa9e4066Sahrens } 1591fa9e4066Sahrens 1592fa9e4066Sahrens /* 159313f5297eSperrin * If this dataset has a non-empty intent log, replay it and destroy it. 1594fa9e4066Sahrens */ 1595fa9e4066Sahrens void 15961209a471SNeil Perrin zil_replay(objset_t *os, void *arg, zil_replay_func_t *replay_func[TX_MAX_TYPE]) 1597fa9e4066Sahrens { 1598fa9e4066Sahrens zilog_t *zilog = dmu_objset_zil(os); 1599d80c45e0Sbonwick const zil_header_t *zh = zilog->zl_header; 1600d80c45e0Sbonwick zil_replay_arg_t zr; 160113f5297eSperrin 16023589c4f0SNeil Perrin if ((zh->zh_flags & ZIL_REPLAY_NEEDED) == 0) { 1603d80c45e0Sbonwick zil_destroy(zilog, B_TRUE); 160413f5297eSperrin return; 160513f5297eSperrin } 1606fa9e4066Sahrens 1607fa9e4066Sahrens zr.zr_os = os; 1608fa9e4066Sahrens zr.zr_replay = replay_func; 1609fa9e4066Sahrens zr.zr_arg = arg; 1610d80c45e0Sbonwick zr.zr_byteswap = BP_SHOULD_BYTESWAP(&zh->zh_log); 1611fa9e4066Sahrens zr.zr_lrbuf = kmem_alloc(2 * SPA_MAXBLOCKSIZE, KM_SLEEP); 1612fa9e4066Sahrens 1613fa9e4066Sahrens /* 1614fa9e4066Sahrens * Wait for in-progress removes to sync before starting replay. 1615fa9e4066Sahrens */ 1616fa9e4066Sahrens txg_wait_synced(zilog->zl_dmu_pool, 0); 1617fa9e4066Sahrens 16181209a471SNeil Perrin zilog->zl_replay = B_TRUE; 161967bd71c6Sperrin zilog->zl_replay_time = lbolt; 162067bd71c6Sperrin ASSERT(zilog->zl_replay_blks == 0); 162167bd71c6Sperrin (void) zil_parse(zilog, zil_incr_blks, zil_replay_log_record, &zr, 1622d80c45e0Sbonwick zh->zh_claim_txg); 1623fa9e4066Sahrens kmem_free(zr.zr_lrbuf, 2 * SPA_MAXBLOCKSIZE); 1624fa9e4066Sahrens 1625d80c45e0Sbonwick zil_destroy(zilog, B_FALSE); 1626a4611edeSahrens txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg); 16271209a471SNeil Perrin zilog->zl_replay = B_FALSE; 1628fa9e4066Sahrens } 1629436b2950Sperrin 1630436b2950Sperrin /* 1631436b2950Sperrin * Report whether all transactions are committed 1632436b2950Sperrin */ 1633436b2950Sperrin int 1634436b2950Sperrin zil_is_committed(zilog_t *zilog) 1635436b2950Sperrin { 1636436b2950Sperrin lwb_t *lwb; 1637b19a79ecSperrin int ret; 1638b19a79ecSperrin 1639b19a79ecSperrin mutex_enter(&zilog->zl_lock); 1640b19a79ecSperrin while (zilog->zl_writer) 1641b19a79ecSperrin cv_wait(&zilog->zl_cv_writer, &zilog->zl_lock); 1642b19a79ecSperrin 1643b19a79ecSperrin /* recent unpushed intent log transactions? */ 1644b19a79ecSperrin if (!list_is_empty(&zilog->zl_itx_list)) { 1645b19a79ecSperrin ret = B_FALSE; 1646b19a79ecSperrin goto out; 1647b19a79ecSperrin } 1648436b2950Sperrin 1649b19a79ecSperrin /* intent log never used? */ 1650b19a79ecSperrin lwb = list_head(&zilog->zl_lwb_list); 1651b19a79ecSperrin if (lwb == NULL) { 1652b19a79ecSperrin ret = B_TRUE; 1653b19a79ecSperrin goto out; 1654b19a79ecSperrin } 1655436b2950Sperrin 1656436b2950Sperrin /* 1657b19a79ecSperrin * more than 1 log buffer means zil_sync() hasn't yet freed 1658b19a79ecSperrin * entries after a txg has committed 1659436b2950Sperrin */ 1660b19a79ecSperrin if (list_next(&zilog->zl_lwb_list, lwb)) { 1661b19a79ecSperrin ret = B_FALSE; 1662b19a79ecSperrin goto out; 1663b19a79ecSperrin } 1664b19a79ecSperrin 1665436b2950Sperrin ASSERT(zil_empty(zilog)); 1666b19a79ecSperrin ret = B_TRUE; 1667b19a79ecSperrin out: 1668b19a79ecSperrin cv_broadcast(&zilog->zl_cv_writer); 1669b19a79ecSperrin mutex_exit(&zilog->zl_lock); 1670b19a79ecSperrin return (ret); 1671436b2950Sperrin } 1672