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, Version 1.0 only 6 * (the "License"). You may not use this file except in compliance 7 * with the License. 8 * 9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 10 * or http://www.opensolaris.org/os/licensing. 11 * See the License for the specific language governing permissions 12 * and limitations under the License. 13 * 14 * When distributing Covered Code, include this CDDL HEADER in each 15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 16 * If applicable, add the following below this CDDL HEADER, with the 17 * fields enclosed by brackets "[]" replaced with your own identifying 18 * information: Portions Copyright [yyyy] [name of copyright owner] 19 * 20 * CDDL HEADER END 21 */ 22 /* 23 * Copyright 2005 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 27 #pragma ident "%Z%%M% %I% %E% SMI" 28 29 #include <sys/zfs_context.h> 30 #include <sys/spa.h> 31 #include <sys/dmu.h> 32 #include <sys/zap.h> 33 #include <sys/arc.h> 34 #include <sys/stat.h> 35 #include <sys/resource.h> 36 #include <sys/zil.h> 37 #include <sys/zil_impl.h> 38 #include <sys/dsl_dataset.h> 39 #include <sys/vdev.h> 40 41 /* 42 * The zfs intent log (ZIL) saves transaction records of system calls 43 * that change the file system in memory with enough information 44 * to be able to replay them. These are stored in memory until 45 * either the DMU transaction group (txg) commits them to the stable pool 46 * and they can be discarded, or they are flushed to the stable log 47 * (also in the pool) due to a fsync, O_DSYNC or other synchronous 48 * requirement. In the event of a panic or power fail then those log 49 * records (transactions) are replayed. 50 * 51 * There is one ZIL per file system. Its on-disk (pool) format consists 52 * of 3 parts: 53 * 54 * - ZIL header 55 * - ZIL blocks 56 * - ZIL records 57 * 58 * A log record holds a system call transaction. Log blocks can 59 * hold many log records and the blocks are chained together. 60 * Each ZIL block contains a block pointer (blkptr_t) to the next 61 * ZIL block in the chain. The ZIL header points to the first 62 * block in the chain. Note there is not a fixed place in the pool 63 * to hold blocks. They are dynamically allocated and freed as 64 * needed from the blocks available. Figure X shows the ZIL structure: 65 */ 66 67 /* 68 * These global ZIL switches affect all pools 69 */ 70 int zil_disable = 0; /* disable intent logging */ 71 int zil_always = 0; /* make every transaction synchronous */ 72 int zil_purge = 0; /* at pool open, just throw everything away */ 73 int zil_noflush = 0; /* don't flush write cache buffers on disks */ 74 75 static kmem_cache_t *zil_lwb_cache; 76 77 static int 78 zil_dva_compare(const void *x1, const void *x2) 79 { 80 const dva_t *dva1 = x1; 81 const dva_t *dva2 = x2; 82 83 if (DVA_GET_VDEV(dva1) < DVA_GET_VDEV(dva2)) 84 return (-1); 85 if (DVA_GET_VDEV(dva1) > DVA_GET_VDEV(dva2)) 86 return (1); 87 88 if (DVA_GET_OFFSET(dva1) < DVA_GET_OFFSET(dva2)) 89 return (-1); 90 if (DVA_GET_OFFSET(dva1) > DVA_GET_OFFSET(dva2)) 91 return (1); 92 93 return (0); 94 } 95 96 static void 97 zil_dva_tree_init(avl_tree_t *t) 98 { 99 avl_create(t, zil_dva_compare, sizeof (zil_dva_node_t), 100 offsetof(zil_dva_node_t, zn_node)); 101 } 102 103 static void 104 zil_dva_tree_fini(avl_tree_t *t) 105 { 106 zil_dva_node_t *zn; 107 void *cookie = NULL; 108 109 while ((zn = avl_destroy_nodes(t, &cookie)) != NULL) 110 kmem_free(zn, sizeof (zil_dva_node_t)); 111 112 avl_destroy(t); 113 } 114 115 static int 116 zil_dva_tree_add(avl_tree_t *t, dva_t *dva) 117 { 118 zil_dva_node_t *zn; 119 avl_index_t where; 120 121 if (avl_find(t, dva, &where) != NULL) 122 return (EEXIST); 123 124 zn = kmem_alloc(sizeof (zil_dva_node_t), KM_SLEEP); 125 zn->zn_dva = *dva; 126 avl_insert(t, zn, where); 127 128 return (0); 129 } 130 131 /* 132 * Read a log block, make sure it's valid, and byteswap it if necessary. 133 */ 134 static int 135 zil_read_log_block(zilog_t *zilog, blkptr_t *bp, char *buf) 136 { 137 uint64_t blksz = BP_GET_LSIZE(bp); 138 zil_trailer_t *ztp = (zil_trailer_t *)(buf + blksz) - 1; 139 zio_cksum_t cksum; 140 int error; 141 142 error = zio_wait(zio_read(NULL, zilog->zl_spa, bp, buf, blksz, 143 NULL, NULL, ZIO_PRIORITY_SYNC_READ, 144 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE)); 145 if (error) { 146 dprintf_bp(bp, "zilog %p bp %p read failed, error %d: ", 147 zilog, bp, error); 148 return (error); 149 } 150 151 if (BP_SHOULD_BYTESWAP(bp)) 152 byteswap_uint64_array(buf, blksz); 153 154 /* 155 * Sequence numbers should be... sequential. The checksum verifier for 156 * the next block should be: <logid[0], logid[1], objset id, seq + 1>. 157 */ 158 cksum = bp->blk_cksum; 159 cksum.zc_word[3]++; 160 if (bcmp(&cksum, &ztp->zit_next_blk.blk_cksum, sizeof (cksum)) != 0) { 161 dprintf_bp(bp, "zilog %p bp %p stale pointer: ", zilog, bp); 162 return (ESTALE); 163 } 164 165 if (BP_IS_HOLE(&ztp->zit_next_blk)) { 166 dprintf_bp(bp, "zilog %p bp %p hole: ", zilog, bp); 167 return (ENOENT); 168 } 169 170 if (ztp->zit_nused > (blksz - sizeof (zil_trailer_t))) { 171 dprintf("zilog %p bp %p nused exceeds blksz\n", zilog, bp); 172 return (EOVERFLOW); 173 } 174 175 dprintf_bp(bp, "zilog %p bp %p good block: ", zilog, bp); 176 177 return (0); 178 } 179 180 /* 181 * Parse the intent log, and call parse_func for each valid record within. 182 */ 183 void 184 zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func, 185 zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg) 186 { 187 blkptr_t blk; 188 char *lrbuf, *lrp; 189 zil_trailer_t *ztp; 190 int reclen, error; 191 192 blk = zilog->zl_header->zh_log; 193 if (BP_IS_HOLE(&blk)) 194 return; 195 196 /* 197 * Starting at the block pointed to by zh_log we read the log chain. 198 * For each block in the chain we strongly check that block to 199 * ensure its validity. We stop when an invalid block is found. 200 * For each block pointer in the chain we call parse_blk_func(). 201 * For each record in each valid block we call parse_lr_func(). 202 */ 203 zil_dva_tree_init(&zilog->zl_dva_tree); 204 lrbuf = zio_buf_alloc(SPA_MAXBLOCKSIZE); 205 for (;;) { 206 error = zil_read_log_block(zilog, &blk, lrbuf); 207 208 if (parse_blk_func != NULL) 209 parse_blk_func(zilog, &blk, arg, txg); 210 211 if (error) 212 break; 213 214 ztp = (zil_trailer_t *)(lrbuf + BP_GET_LSIZE(&blk)) - 1; 215 blk = ztp->zit_next_blk; 216 217 if (parse_lr_func == NULL) 218 continue; 219 220 for (lrp = lrbuf; lrp < lrbuf + ztp->zit_nused; lrp += reclen) { 221 lr_t *lr = (lr_t *)lrp; 222 reclen = lr->lrc_reclen; 223 ASSERT3U(reclen, >=, sizeof (lr_t)); 224 parse_lr_func(zilog, lr, arg, txg); 225 } 226 } 227 zio_buf_free(lrbuf, SPA_MAXBLOCKSIZE); 228 zil_dva_tree_fini(&zilog->zl_dva_tree); 229 } 230 231 /* ARGSUSED */ 232 static void 233 zil_claim_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t first_txg) 234 { 235 spa_t *spa = zilog->zl_spa; 236 int err; 237 238 dprintf_bp(bp, "first_txg %llu: ", first_txg); 239 240 /* 241 * Claim log block if not already committed and not already claimed. 242 */ 243 if (bp->blk_birth >= first_txg && 244 zil_dva_tree_add(&zilog->zl_dva_tree, BP_IDENTITY(bp)) == 0) { 245 err = zio_wait(zio_claim(NULL, spa, first_txg, bp, NULL, NULL)); 246 ASSERT(err == 0); 247 } 248 } 249 250 static void 251 zil_claim_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t first_txg) 252 { 253 if (lrc->lrc_txtype == TX_WRITE) { 254 lr_write_t *lr = (lr_write_t *)lrc; 255 zil_claim_log_block(zilog, &lr->lr_blkptr, tx, first_txg); 256 } 257 } 258 259 /* ARGSUSED */ 260 static void 261 zil_free_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t claim_txg) 262 { 263 zio_free_blk(zilog->zl_spa, bp, dmu_tx_get_txg(tx)); 264 } 265 266 static void 267 zil_free_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t claim_txg) 268 { 269 /* 270 * If we previously claimed it, we need to free it. 271 */ 272 if (claim_txg != 0 && lrc->lrc_txtype == TX_WRITE) { 273 lr_write_t *lr = (lr_write_t *)lrc; 274 blkptr_t *bp = &lr->lr_blkptr; 275 if (bp->blk_birth >= claim_txg && 276 !zil_dva_tree_add(&zilog->zl_dva_tree, BP_IDENTITY(bp))) { 277 (void) arc_free(NULL, zilog->zl_spa, 278 dmu_tx_get_txg(tx), bp, NULL, NULL, ARC_WAIT); 279 } 280 } 281 } 282 283 /* 284 * Create an on-disk intent log. 285 */ 286 static void 287 zil_create(zilog_t *zilog) 288 { 289 lwb_t *lwb; 290 uint64_t txg; 291 dmu_tx_t *tx; 292 blkptr_t blk; 293 int error; 294 295 ASSERT(zilog->zl_header->zh_claim_txg == 0); 296 ASSERT(zilog->zl_header->zh_replay_seq == 0); 297 298 /* 299 * Initialize the log header block. 300 */ 301 tx = dmu_tx_create(zilog->zl_os); 302 (void) dmu_tx_assign(tx, TXG_WAIT); 303 dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); 304 txg = dmu_tx_get_txg(tx); 305 306 /* 307 * Allocate the first log block and assign its checksum verifier. 308 */ 309 error = zio_alloc_blk(zilog->zl_spa, ZIO_CHECKSUM_ZILOG, 310 ZIL_MIN_BLKSZ, &blk, txg); 311 if (error == 0) { 312 ZIO_SET_CHECKSUM(&blk.blk_cksum, 313 spa_get_random(-1ULL), spa_get_random(-1ULL), 314 dmu_objset_id(zilog->zl_os), 1ULL); 315 316 /* 317 * Allocate a log write buffer (lwb) for the first log block. 318 */ 319 lwb = kmem_cache_alloc(zil_lwb_cache, KM_SLEEP); 320 lwb->lwb_zilog = zilog; 321 lwb->lwb_blk = blk; 322 lwb->lwb_nused = 0; 323 lwb->lwb_sz = BP_GET_LSIZE(&lwb->lwb_blk); 324 lwb->lwb_buf = zio_buf_alloc(lwb->lwb_sz); 325 lwb->lwb_max_txg = txg; 326 lwb->lwb_seq = 0; 327 lwb->lwb_state = UNWRITTEN; 328 mutex_enter(&zilog->zl_lock); 329 list_insert_tail(&zilog->zl_lwb_list, lwb); 330 mutex_exit(&zilog->zl_lock); 331 } 332 333 dmu_tx_commit(tx); 334 txg_wait_synced(zilog->zl_dmu_pool, txg); 335 } 336 337 /* 338 * In one tx, free all log blocks and clear the log header. 339 */ 340 void 341 zil_destroy(zilog_t *zilog) 342 { 343 dmu_tx_t *tx; 344 uint64_t txg; 345 346 mutex_enter(&zilog->zl_destroy_lock); 347 348 if (BP_IS_HOLE(&zilog->zl_header->zh_log)) { 349 mutex_exit(&zilog->zl_destroy_lock); 350 return; 351 } 352 353 tx = dmu_tx_create(zilog->zl_os); 354 (void) dmu_tx_assign(tx, TXG_WAIT); 355 dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); 356 txg = dmu_tx_get_txg(tx); 357 358 zil_parse(zilog, zil_free_log_block, zil_free_log_record, tx, 359 zilog->zl_header->zh_claim_txg); 360 zilog->zl_destroy_txg = txg; 361 362 dmu_tx_commit(tx); 363 txg_wait_synced(zilog->zl_dmu_pool, txg); 364 365 mutex_exit(&zilog->zl_destroy_lock); 366 } 367 368 void 369 zil_claim(char *osname, void *txarg) 370 { 371 dmu_tx_t *tx = txarg; 372 uint64_t first_txg = dmu_tx_get_txg(tx); 373 zilog_t *zilog; 374 zil_header_t *zh; 375 objset_t *os; 376 int error; 377 378 error = dmu_objset_open(osname, DMU_OST_ANY, DS_MODE_STANDARD, &os); 379 if (error) { 380 cmn_err(CE_WARN, "can't process intent log for %s", osname); 381 return; 382 } 383 384 zilog = dmu_objset_zil(os); 385 zh = zilog->zl_header; 386 387 /* 388 * Claim all log blocks if we haven't already done so. 389 */ 390 ASSERT3U(zh->zh_claim_txg, <=, first_txg); 391 if (zh->zh_claim_txg == 0 && !BP_IS_HOLE(&zh->zh_log)) { 392 zh->zh_claim_txg = first_txg; 393 zil_parse(zilog, zil_claim_log_block, zil_claim_log_record, 394 tx, first_txg); 395 dsl_dataset_dirty(dmu_objset_ds(os), tx); 396 } 397 ASSERT3U(first_txg, ==, (spa_last_synced_txg(zilog->zl_spa) + 1)); 398 dmu_objset_close(os); 399 } 400 401 void 402 zil_add_vdev(zilog_t *zilog, uint64_t vdev, uint64_t seq) 403 { 404 zil_vdev_t *zv; 405 406 if (zil_noflush) 407 return; 408 409 ASSERT(MUTEX_HELD(&zilog->zl_lock)); 410 zv = kmem_alloc(sizeof (zil_vdev_t), KM_SLEEP); 411 zv->vdev = vdev; 412 zv->seq = seq; 413 list_insert_tail(&zilog->zl_vdev_list, zv); 414 } 415 416 void 417 zil_flush_vdevs(zilog_t *zilog, uint64_t seq) 418 { 419 vdev_t *vd; 420 zil_vdev_t *zv, *zv2; 421 zio_t *zio; 422 spa_t *spa; 423 uint64_t vdev; 424 425 if (zil_noflush) 426 return; 427 428 ASSERT(MUTEX_HELD(&zilog->zl_lock)); 429 430 spa = zilog->zl_spa; 431 zio = NULL; 432 433 while ((zv = list_head(&zilog->zl_vdev_list)) != NULL && 434 zv->seq <= seq) { 435 vdev = zv->vdev; 436 list_remove(&zilog->zl_vdev_list, zv); 437 kmem_free(zv, sizeof (zil_vdev_t)); 438 439 /* 440 * remove all chained entries <= seq with same vdev 441 */ 442 zv = list_head(&zilog->zl_vdev_list); 443 while (zv && zv->seq <= seq) { 444 zv2 = list_next(&zilog->zl_vdev_list, zv); 445 if (zv->vdev == vdev) { 446 list_remove(&zilog->zl_vdev_list, zv); 447 kmem_free(zv, sizeof (zil_vdev_t)); 448 } 449 zv = zv2; 450 } 451 452 /* flush the write cache for this vdev */ 453 mutex_exit(&zilog->zl_lock); 454 if (zio == NULL) 455 zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL); 456 vd = vdev_lookup_top(spa, vdev); 457 ASSERT(vd); 458 (void) zio_nowait(zio_ioctl(zio, spa, vd, DKIOCFLUSHWRITECACHE, 459 NULL, NULL, ZIO_PRIORITY_NOW, 460 ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY)); 461 mutex_enter(&zilog->zl_lock); 462 } 463 464 /* 465 * Wait for all the flushes to complete. Not all devices actually 466 * support the DKIOCFLUSHWRITECACHE ioctl, so it's OK if it fails. 467 */ 468 if (zio != NULL) { 469 mutex_exit(&zilog->zl_lock); 470 (void) zio_wait(zio); 471 mutex_enter(&zilog->zl_lock); 472 } 473 } 474 475 /* 476 * Function called when a log block write completes 477 */ 478 static void 479 zil_lwb_write_done(zio_t *zio) 480 { 481 lwb_t *prev; 482 lwb_t *lwb = zio->io_private; 483 zilog_t *zilog = lwb->lwb_zilog; 484 uint64_t max_seq; 485 486 /* 487 * Now that we've written this log block, we have a stable pointer 488 * to the next block in the chain, so it's OK to let the txg in 489 * which we allocated the next block sync. 490 */ 491 txg_rele_to_sync(&lwb->lwb_txgh); 492 493 zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); 494 mutex_enter(&zilog->zl_lock); 495 lwb->lwb_buf = NULL; 496 if (zio->io_error) { 497 zilog->zl_log_error = B_TRUE; 498 mutex_exit(&zilog->zl_lock); 499 cv_broadcast(&zilog->zl_cv_seq); 500 return; 501 } 502 503 prev = list_prev(&zilog->zl_lwb_list, lwb); 504 if (prev && prev->lwb_state != SEQ_COMPLETE) { 505 /* There's an unwritten buffer in the chain before this one */ 506 lwb->lwb_state = SEQ_INCOMPLETE; 507 mutex_exit(&zilog->zl_lock); 508 return; 509 } 510 511 max_seq = lwb->lwb_seq; 512 lwb->lwb_state = SEQ_COMPLETE; 513 /* 514 * We must also follow up the chain for already written buffers 515 * to see if we can set zl_ss_seq even higher. 516 */ 517 while (lwb = list_next(&zilog->zl_lwb_list, lwb)) { 518 if (lwb->lwb_state != SEQ_INCOMPLETE) 519 break; 520 lwb->lwb_state = SEQ_COMPLETE; 521 /* lwb_seq will be zero if we've written an empty buffer */ 522 if (lwb->lwb_seq) { 523 ASSERT3U(max_seq, <, lwb->lwb_seq); 524 max_seq = lwb->lwb_seq; 525 } 526 } 527 zilog->zl_ss_seq = MAX(max_seq, zilog->zl_ss_seq); 528 mutex_exit(&zilog->zl_lock); 529 cv_broadcast(&zilog->zl_cv_seq); 530 } 531 532 /* 533 * Start a log block write and advance to the next log block. 534 * Calls are serialized. 535 */ 536 static lwb_t * 537 zil_lwb_write_start(zilog_t *zilog, lwb_t *lwb) 538 { 539 lwb_t *nlwb; 540 zil_trailer_t *ztp = (zil_trailer_t *)(lwb->lwb_buf + lwb->lwb_sz) - 1; 541 uint64_t txg; 542 uint64_t zil_blksz; 543 int error; 544 545 ASSERT(lwb->lwb_nused <= ZIL_BLK_DATA_SZ(lwb)); 546 547 /* 548 * Allocate the next block and save its address in this block 549 * before writing it in order to establish the log chain. 550 * Note that if the allocation of nlwb synced before we wrote 551 * the block that points at it (lwb), we'd leak it if we crashed. 552 * Therefore, we don't do txg_rele_to_sync() until zil_lwb_write_done(). 553 */ 554 txg = txg_hold_open(zilog->zl_dmu_pool, &lwb->lwb_txgh); 555 txg_rele_to_quiesce(&lwb->lwb_txgh); 556 557 /* 558 * Pick a ZIL blocksize. We request a size that is the 559 * maximum of the previous used size, the current used size and 560 * the amount waiting in the queue. 561 */ 562 zil_blksz = MAX(zilog->zl_cur_used, zilog->zl_prev_used); 563 zil_blksz = MAX(zil_blksz, zilog->zl_itx_list_sz + sizeof (*ztp)); 564 zil_blksz = P2ROUNDUP(zil_blksz, ZIL_MIN_BLKSZ); 565 if (zil_blksz > ZIL_MAX_BLKSZ) 566 zil_blksz = ZIL_MAX_BLKSZ; 567 568 error = zio_alloc_blk(zilog->zl_spa, ZIO_CHECKSUM_ZILOG, 569 zil_blksz, &ztp->zit_next_blk, txg); 570 if (error) { 571 txg_rele_to_sync(&lwb->lwb_txgh); 572 return (NULL); 573 } 574 575 ASSERT3U(ztp->zit_next_blk.blk_birth, ==, txg); 576 ztp->zit_nused = lwb->lwb_nused; 577 ztp->zit_bt.zbt_cksum = lwb->lwb_blk.blk_cksum; 578 ztp->zit_next_blk.blk_cksum = lwb->lwb_blk.blk_cksum; 579 ztp->zit_next_blk.blk_cksum.zc_word[3]++; 580 581 /* 582 * Allocate a new log write buffer (lwb). 583 */ 584 nlwb = kmem_cache_alloc(zil_lwb_cache, KM_SLEEP); 585 586 nlwb->lwb_zilog = zilog; 587 nlwb->lwb_blk = ztp->zit_next_blk; 588 nlwb->lwb_nused = 0; 589 nlwb->lwb_sz = BP_GET_LSIZE(&nlwb->lwb_blk); 590 nlwb->lwb_buf = zio_buf_alloc(nlwb->lwb_sz); 591 nlwb->lwb_max_txg = txg; 592 nlwb->lwb_seq = 0; 593 nlwb->lwb_state = UNWRITTEN; 594 595 /* 596 * Put new lwb at the end of the log chain, 597 * and record the vdev for later flushing 598 */ 599 mutex_enter(&zilog->zl_lock); 600 list_insert_tail(&zilog->zl_lwb_list, nlwb); 601 zil_add_vdev(zilog, DVA_GET_VDEV(BP_IDENTITY(&(lwb->lwb_blk))), 602 lwb->lwb_seq); 603 mutex_exit(&zilog->zl_lock); 604 605 /* 606 * write the old log block 607 */ 608 dprintf_bp(&lwb->lwb_blk, "lwb %p txg %llu: ", lwb, txg); 609 zio_nowait(zio_rewrite(NULL, zilog->zl_spa, ZIO_CHECKSUM_ZILOG, 0, 610 &lwb->lwb_blk, lwb->lwb_buf, lwb->lwb_sz, zil_lwb_write_done, lwb, 611 ZIO_PRIORITY_LOG_WRITE, ZIO_FLAG_MUSTSUCCEED)); 612 613 return (nlwb); 614 } 615 616 static lwb_t * 617 zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb) 618 { 619 lr_t *lrc = &itx->itx_lr; /* common log record */ 620 uint64_t seq = lrc->lrc_seq; 621 uint64_t txg = lrc->lrc_txg; 622 uint64_t reclen = lrc->lrc_reclen; 623 int error; 624 625 if (lwb == NULL) 626 return (NULL); 627 ASSERT(lwb->lwb_buf != NULL); 628 629 /* 630 * If it's a write, fetch the data or get its blkptr as appropriate. 631 */ 632 if (lrc->lrc_txtype == TX_WRITE) { 633 lr_write_t *lr = (lr_write_t *)lrc; 634 if (txg > spa_freeze_txg(zilog->zl_spa)) 635 txg_wait_synced(zilog->zl_dmu_pool, txg); 636 637 if (!itx->itx_data_copied && 638 (error = zilog->zl_get_data(itx->itx_private, lr)) != 0) { 639 if (error != ENOENT && error != EALREADY) { 640 txg_wait_synced(zilog->zl_dmu_pool, txg); 641 mutex_enter(&zilog->zl_lock); 642 zilog->zl_ss_seq = MAX(seq, zilog->zl_ss_seq); 643 zil_add_vdev(zilog, 644 DVA_GET_VDEV(BP_IDENTITY(&(lr->lr_blkptr))), 645 seq); 646 mutex_exit(&zilog->zl_lock); 647 return (lwb); 648 } 649 mutex_enter(&zilog->zl_lock); 650 zil_add_vdev(zilog, 651 DVA_GET_VDEV(BP_IDENTITY(&(lr->lr_blkptr))), seq); 652 mutex_exit(&zilog->zl_lock); 653 return (lwb); 654 } 655 } 656 657 zilog->zl_cur_used += reclen; 658 659 /* 660 * If this record won't fit in the current log block, start a new one. 661 */ 662 if (lwb->lwb_nused + reclen > ZIL_BLK_DATA_SZ(lwb)) { 663 lwb = zil_lwb_write_start(zilog, lwb); 664 if (lwb == NULL) 665 return (NULL); 666 if (lwb->lwb_nused + reclen > ZIL_BLK_DATA_SZ(lwb)) { 667 txg_wait_synced(zilog->zl_dmu_pool, txg); 668 mutex_enter(&zilog->zl_lock); 669 zilog->zl_ss_seq = MAX(seq, zilog->zl_ss_seq); 670 mutex_exit(&zilog->zl_lock); 671 return (lwb); 672 } 673 } 674 675 bcopy(lrc, lwb->lwb_buf + lwb->lwb_nused, reclen); 676 lwb->lwb_nused += reclen; 677 lwb->lwb_max_txg = MAX(lwb->lwb_max_txg, txg); 678 ASSERT3U(lwb->lwb_seq, <, seq); 679 lwb->lwb_seq = seq; 680 ASSERT3U(lwb->lwb_nused, <=, ZIL_BLK_DATA_SZ(lwb)); 681 ASSERT3U(P2PHASE(lwb->lwb_nused, sizeof (uint64_t)), ==, 0); 682 683 return (lwb); 684 } 685 686 itx_t * 687 zil_itx_create(int txtype, size_t lrsize) 688 { 689 itx_t *itx; 690 691 lrsize = P2ROUNDUP(lrsize, sizeof (uint64_t)); 692 693 itx = kmem_alloc(offsetof(itx_t, itx_lr) + lrsize, KM_SLEEP); 694 itx->itx_lr.lrc_txtype = txtype; 695 itx->itx_lr.lrc_reclen = lrsize; 696 itx->itx_lr.lrc_seq = 0; /* defensive */ 697 698 return (itx); 699 } 700 701 uint64_t 702 zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx) 703 { 704 uint64_t seq; 705 706 ASSERT(itx->itx_lr.lrc_seq == 0); 707 708 mutex_enter(&zilog->zl_lock); 709 list_insert_tail(&zilog->zl_itx_list, itx); 710 zilog->zl_itx_list_sz += itx->itx_lr.lrc_reclen; 711 itx->itx_lr.lrc_txg = dmu_tx_get_txg(tx); 712 itx->itx_lr.lrc_seq = seq = ++zilog->zl_itx_seq; 713 mutex_exit(&zilog->zl_lock); 714 715 return (seq); 716 } 717 718 /* 719 * Free up all in-memory intent log transactions that have now been synced. 720 */ 721 static void 722 zil_itx_clean(zilog_t *zilog) 723 { 724 uint64_t synced_txg = spa_last_synced_txg(zilog->zl_spa); 725 uint64_t freeze_txg = spa_freeze_txg(zilog->zl_spa); 726 uint64_t max_seq = 0; 727 itx_t *itx; 728 729 mutex_enter(&zilog->zl_lock); 730 while ((itx = list_head(&zilog->zl_itx_list)) != NULL && 731 itx->itx_lr.lrc_txg <= MIN(synced_txg, freeze_txg)) { 732 list_remove(&zilog->zl_itx_list, itx); 733 zilog->zl_itx_list_sz -= itx->itx_lr.lrc_reclen; 734 ASSERT3U(max_seq, <, itx->itx_lr.lrc_seq); 735 max_seq = itx->itx_lr.lrc_seq; 736 kmem_free(itx, offsetof(itx_t, itx_lr) 737 + itx->itx_lr.lrc_reclen); 738 } 739 if (max_seq > zilog->zl_ss_seq) { 740 zilog->zl_ss_seq = max_seq; 741 cv_broadcast(&zilog->zl_cv_seq); 742 } 743 mutex_exit(&zilog->zl_lock); 744 } 745 746 void 747 zil_clean(zilog_t *zilog) 748 { 749 /* 750 * Check for any log blocks that can be freed. 751 * Log blocks are only freed when the log block allocation and 752 * log records contained within are both known to be committed. 753 */ 754 mutex_enter(&zilog->zl_lock); 755 if (list_head(&zilog->zl_itx_list) != NULL) 756 (void) taskq_dispatch(zilog->zl_clean_taskq, 757 (void (*)(void *))zil_itx_clean, zilog, TQ_NOSLEEP); 758 mutex_exit(&zilog->zl_lock); 759 } 760 761 /* 762 * Push zfs transactions to stable storage up to the supplied sequence number. 763 */ 764 void 765 zil_commit(zilog_t *zilog, uint64_t seq, int ioflag) 766 { 767 uint64_t txg; 768 uint64_t max_seq; 769 uint64_t reclen; 770 itx_t *itx; 771 lwb_t *lwb; 772 spa_t *spa; 773 774 if (zilog == NULL || seq == 0 || 775 ((ioflag & (FSYNC | FDSYNC | FRSYNC)) == 0 && !zil_always)) 776 return; 777 778 spa = zilog->zl_spa; 779 mutex_enter(&zilog->zl_lock); 780 781 seq = MIN(seq, zilog->zl_itx_seq); /* cap seq at largest itx seq */ 782 783 for (;;) { 784 if (zilog->zl_ss_seq >= seq) { /* already on stable storage */ 785 cv_signal(&zilog->zl_cv_write); 786 mutex_exit(&zilog->zl_lock); 787 return; 788 } 789 790 if (zilog->zl_writer == B_FALSE) /* no one writing, do it */ 791 break; 792 793 cv_wait(&zilog->zl_cv_write, &zilog->zl_lock); 794 } 795 796 zilog->zl_writer = B_TRUE; 797 max_seq = 0; 798 799 if (zilog->zl_suspend) { 800 lwb = NULL; 801 } else { 802 lwb = list_tail(&zilog->zl_lwb_list); 803 if (lwb == NULL) { 804 mutex_exit(&zilog->zl_lock); 805 zil_create(zilog); 806 mutex_enter(&zilog->zl_lock); 807 lwb = list_tail(&zilog->zl_lwb_list); 808 } 809 } 810 811 /* 812 * Loop through in-memory log transactions filling log blocks, 813 * until we reach the given sequence number and there's no more 814 * room in the write buffer. 815 */ 816 for (;;) { 817 itx = list_head(&zilog->zl_itx_list); 818 if (itx == NULL) 819 break; 820 821 reclen = itx->itx_lr.lrc_reclen; 822 if ((itx->itx_lr.lrc_seq > seq) && 823 ((lwb == NULL) || (lwb->lwb_nused + reclen > 824 ZIL_BLK_DATA_SZ(lwb)))) 825 break; 826 827 list_remove(&zilog->zl_itx_list, itx); 828 txg = itx->itx_lr.lrc_txg; 829 ASSERT(txg); 830 831 mutex_exit(&zilog->zl_lock); 832 if (txg > spa_last_synced_txg(spa) || 833 txg > spa_freeze_txg(spa)) 834 lwb = zil_lwb_commit(zilog, itx, lwb); 835 else 836 max_seq = itx->itx_lr.lrc_seq; 837 kmem_free(itx, offsetof(itx_t, itx_lr) 838 + itx->itx_lr.lrc_reclen); 839 mutex_enter(&zilog->zl_lock); 840 zilog->zl_itx_list_sz -= reclen; 841 } 842 843 mutex_exit(&zilog->zl_lock); 844 845 /* write the last block out */ 846 if (lwb != NULL && lwb->lwb_nused != 0) 847 lwb = zil_lwb_write_start(zilog, lwb); 848 849 zilog->zl_prev_used = zilog->zl_cur_used; 850 zilog->zl_cur_used = 0; 851 852 mutex_enter(&zilog->zl_lock); 853 if (max_seq > zilog->zl_ss_seq) { 854 zilog->zl_ss_seq = max_seq; 855 cv_broadcast(&zilog->zl_cv_seq); 856 } 857 /* 858 * Wait if necessary for our seq to be committed. 859 */ 860 if (lwb) { 861 while (zilog->zl_ss_seq < seq && zilog->zl_log_error == 0) 862 cv_wait(&zilog->zl_cv_seq, &zilog->zl_lock); 863 zil_flush_vdevs(zilog, seq); 864 } 865 866 if (zilog->zl_log_error || lwb == NULL) { 867 zilog->zl_log_error = 0; 868 max_seq = zilog->zl_itx_seq; 869 mutex_exit(&zilog->zl_lock); 870 txg_wait_synced(zilog->zl_dmu_pool, 0); 871 mutex_enter(&zilog->zl_lock); 872 zilog->zl_ss_seq = MAX(max_seq, zilog->zl_ss_seq); 873 cv_broadcast(&zilog->zl_cv_seq); 874 } 875 /* wake up others waiting to start a write */ 876 zilog->zl_writer = B_FALSE; 877 mutex_exit(&zilog->zl_lock); 878 cv_signal(&zilog->zl_cv_write); 879 } 880 881 /* 882 * Called in syncing context to free committed log blocks and update log header. 883 */ 884 void 885 zil_sync(zilog_t *zilog, dmu_tx_t *tx) 886 { 887 uint64_t txg = dmu_tx_get_txg(tx); 888 spa_t *spa = zilog->zl_spa; 889 lwb_t *lwb; 890 891 ASSERT(zilog->zl_stop_sync == 0); 892 893 zilog->zl_header->zh_replay_seq = zilog->zl_replay_seq[txg & TXG_MASK]; 894 895 if (zilog->zl_destroy_txg == txg) { 896 bzero(zilog->zl_header, sizeof (zil_header_t)); 897 bzero(zilog->zl_replay_seq, sizeof (zilog->zl_replay_seq)); 898 zilog->zl_destroy_txg = 0; 899 } 900 901 mutex_enter(&zilog->zl_lock); 902 for (;;) { 903 lwb = list_head(&zilog->zl_lwb_list); 904 if (lwb == NULL) { 905 mutex_exit(&zilog->zl_lock); 906 return; 907 } 908 if (lwb->lwb_buf != NULL || lwb->lwb_max_txg > txg) 909 break; 910 list_remove(&zilog->zl_lwb_list, lwb); 911 zio_free_blk(spa, &lwb->lwb_blk, txg); 912 kmem_cache_free(zil_lwb_cache, lwb); 913 } 914 zilog->zl_header->zh_log = lwb->lwb_blk; 915 mutex_exit(&zilog->zl_lock); 916 } 917 918 void 919 zil_init(void) 920 { 921 zil_lwb_cache = kmem_cache_create("zil_lwb_cache", 922 sizeof (struct lwb), NULL, NULL, NULL, NULL, NULL, NULL, 0); 923 } 924 925 void 926 zil_fini(void) 927 { 928 kmem_cache_destroy(zil_lwb_cache); 929 } 930 931 zilog_t * 932 zil_alloc(objset_t *os, zil_header_t *zh_phys) 933 { 934 zilog_t *zilog; 935 936 zilog = kmem_zalloc(sizeof (zilog_t), KM_SLEEP); 937 938 zilog->zl_header = zh_phys; 939 zilog->zl_os = os; 940 zilog->zl_spa = dmu_objset_spa(os); 941 zilog->zl_dmu_pool = dmu_objset_pool(os); 942 943 list_create(&zilog->zl_itx_list, sizeof (itx_t), 944 offsetof(itx_t, itx_node)); 945 946 list_create(&zilog->zl_lwb_list, sizeof (lwb_t), 947 offsetof(lwb_t, lwb_node)); 948 949 list_create(&zilog->zl_vdev_list, sizeof (zil_vdev_t), 950 offsetof(zil_vdev_t, vdev_seq_node)); 951 952 return (zilog); 953 } 954 955 void 956 zil_free(zilog_t *zilog) 957 { 958 lwb_t *lwb; 959 zil_vdev_t *zv; 960 961 zilog->zl_stop_sync = 1; 962 963 while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) { 964 list_remove(&zilog->zl_lwb_list, lwb); 965 if (lwb->lwb_buf != NULL) 966 zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); 967 kmem_cache_free(zil_lwb_cache, lwb); 968 } 969 list_destroy(&zilog->zl_lwb_list); 970 971 while ((zv = list_head(&zilog->zl_vdev_list)) != NULL) { 972 list_remove(&zilog->zl_vdev_list, zv); 973 kmem_free(zv, sizeof (zil_vdev_t)); 974 } 975 list_destroy(&zilog->zl_vdev_list); 976 977 ASSERT(list_head(&zilog->zl_itx_list) == NULL); 978 list_destroy(&zilog->zl_itx_list); 979 980 kmem_free(zilog, sizeof (zilog_t)); 981 } 982 983 /* 984 * Open an intent log. 985 */ 986 zilog_t * 987 zil_open(objset_t *os, zil_get_data_t *get_data) 988 { 989 zilog_t *zilog = dmu_objset_zil(os); 990 991 zilog->zl_get_data = get_data; 992 zilog->zl_clean_taskq = taskq_create("zil_clean", 1, minclsyspri, 993 2, 2, TASKQ_PREPOPULATE); 994 995 return (zilog); 996 } 997 998 /* 999 * Close an intent log. 1000 */ 1001 void 1002 zil_close(zilog_t *zilog) 1003 { 1004 txg_wait_synced(zilog->zl_dmu_pool, 0); 1005 taskq_destroy(zilog->zl_clean_taskq); 1006 zilog->zl_clean_taskq = NULL; 1007 zilog->zl_get_data = NULL; 1008 1009 zil_itx_clean(zilog); 1010 ASSERT(list_head(&zilog->zl_itx_list) == NULL); 1011 } 1012 1013 /* 1014 * Suspend an intent log. While in suspended mode, we still honor 1015 * synchronous semantics, but we rely on txg_wait_synced() to do it. 1016 * We suspend the log briefly when taking a snapshot so that the snapshot 1017 * contains all the data it's supposed to, and has an empty intent log. 1018 */ 1019 int 1020 zil_suspend(zilog_t *zilog) 1021 { 1022 lwb_t *lwb; 1023 1024 mutex_enter(&zilog->zl_lock); 1025 if (zilog->zl_header->zh_claim_txg != 0) { /* unplayed log */ 1026 mutex_exit(&zilog->zl_lock); 1027 return (EBUSY); 1028 } 1029 zilog->zl_suspend++; 1030 mutex_exit(&zilog->zl_lock); 1031 1032 zil_commit(zilog, UINT64_MAX, FSYNC); 1033 1034 mutex_enter(&zilog->zl_lock); 1035 while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) { 1036 if (lwb->lwb_buf != NULL) { 1037 /* 1038 * Wait for the buffer if it's in the process of 1039 * being written. 1040 */ 1041 if ((lwb->lwb_seq != 0) && 1042 (lwb->lwb_state != SEQ_COMPLETE)) { 1043 cv_wait(&zilog->zl_cv_seq, &zilog->zl_lock); 1044 continue; 1045 } 1046 zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); 1047 } 1048 list_remove(&zilog->zl_lwb_list, lwb); 1049 kmem_cache_free(zil_lwb_cache, lwb); 1050 } 1051 mutex_exit(&zilog->zl_lock); 1052 1053 zil_destroy(zilog); 1054 1055 return (0); 1056 } 1057 1058 void 1059 zil_resume(zilog_t *zilog) 1060 { 1061 mutex_enter(&zilog->zl_lock); 1062 ASSERT(zilog->zl_suspend != 0); 1063 zilog->zl_suspend--; 1064 mutex_exit(&zilog->zl_lock); 1065 } 1066 1067 typedef struct zil_replay_arg { 1068 objset_t *zr_os; 1069 zil_replay_func_t **zr_replay; 1070 void *zr_arg; 1071 void (*zr_rm_sync)(void *arg); 1072 uint64_t *zr_txgp; 1073 boolean_t zr_byteswap; 1074 char *zr_lrbuf; 1075 } zil_replay_arg_t; 1076 1077 static void 1078 zil_replay_log_record(zilog_t *zilog, lr_t *lr, void *zra, uint64_t claim_txg) 1079 { 1080 zil_replay_arg_t *zr = zra; 1081 zil_header_t *zh = zilog->zl_header; 1082 uint64_t reclen = lr->lrc_reclen; 1083 uint64_t txtype = lr->lrc_txtype; 1084 int pass, error; 1085 1086 if (zilog->zl_stop_replay) 1087 return; 1088 1089 if (lr->lrc_txg < claim_txg) /* already committed */ 1090 return; 1091 1092 if (lr->lrc_seq <= zh->zh_replay_seq) /* already replayed */ 1093 return; 1094 1095 /* 1096 * Make a copy of the data so we can revise and extend it. 1097 */ 1098 bcopy(lr, zr->zr_lrbuf, reclen); 1099 1100 /* 1101 * The log block containing this lr may have been byteswapped 1102 * so that we can easily examine common fields like lrc_txtype. 1103 * However, the log is a mix of different data types, and only the 1104 * replay vectors know how to byteswap their records. Therefore, if 1105 * the lr was byteswapped, undo it before invoking the replay vector. 1106 */ 1107 if (zr->zr_byteswap) 1108 byteswap_uint64_array(zr->zr_lrbuf, reclen); 1109 1110 /* 1111 * If this is a TX_WRITE with a blkptr, suck in the data. 1112 */ 1113 if (txtype == TX_WRITE && reclen == sizeof (lr_write_t)) { 1114 lr_write_t *lrw = (lr_write_t *)lr; 1115 blkptr_t *wbp = &lrw->lr_blkptr; 1116 uint64_t wlen = lrw->lr_length; 1117 char *wbuf = zr->zr_lrbuf + reclen; 1118 1119 if (BP_IS_HOLE(wbp)) { /* compressed to a hole */ 1120 bzero(wbuf, wlen); 1121 } else { 1122 /* 1123 * A subsequent write may have overwritten this block, 1124 * in which case wbp may have been been freed and 1125 * reallocated, and our read of wbp may fail with a 1126 * checksum error. We can safely ignore this because 1127 * the later write will provide the correct data. 1128 */ 1129 (void) zio_wait(zio_read(NULL, zilog->zl_spa, 1130 wbp, wbuf, BP_GET_LSIZE(wbp), NULL, NULL, 1131 ZIO_PRIORITY_SYNC_READ, 1132 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE)); 1133 (void) memmove(wbuf, wbuf + lrw->lr_blkoff, wlen); 1134 } 1135 } 1136 1137 /* 1138 * We must now do two things atomically: replay this log record, 1139 * and update the log header to reflect the fact that we did so. 1140 * We use the DMU's ability to assign into a specific txg to do this. 1141 */ 1142 for (pass = 1; /* CONSTANTCONDITION */; pass++) { 1143 uint64_t replay_txg; 1144 dmu_tx_t *replay_tx; 1145 1146 replay_tx = dmu_tx_create(zr->zr_os); 1147 error = dmu_tx_assign(replay_tx, TXG_WAIT); 1148 if (error) { 1149 dmu_tx_abort(replay_tx); 1150 break; 1151 } 1152 1153 replay_txg = dmu_tx_get_txg(replay_tx); 1154 1155 if (txtype == 0 || txtype >= TX_MAX_TYPE) { 1156 error = EINVAL; 1157 } else { 1158 /* 1159 * On the first pass, arrange for the replay vector 1160 * to fail its dmu_tx_assign(). That's the only way 1161 * to ensure that those code paths remain well tested. 1162 */ 1163 *zr->zr_txgp = replay_txg - (pass == 1); 1164 error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lrbuf, 1165 zr->zr_byteswap); 1166 *zr->zr_txgp = TXG_NOWAIT; 1167 } 1168 1169 if (error == 0) { 1170 dsl_dataset_dirty(dmu_objset_ds(zr->zr_os), replay_tx); 1171 zilog->zl_replay_seq[replay_txg & TXG_MASK] = 1172 lr->lrc_seq; 1173 } 1174 1175 dmu_tx_commit(replay_tx); 1176 1177 if (error != ERESTART) 1178 break; 1179 1180 if (pass != 1) 1181 txg_wait_open(spa_get_dsl(zilog->zl_spa), 1182 replay_txg + 1); 1183 1184 dprintf("pass %d, retrying\n", pass); 1185 } 1186 1187 if (error) { 1188 char *name = kmem_alloc(MAXNAMELEN, KM_SLEEP); 1189 dmu_objset_name(zr->zr_os, name); 1190 cmn_err(CE_WARN, "ZFS replay transaction error %d, " 1191 "dataset %s, seq 0x%llx, txtype %llu\n", 1192 error, name, 1193 (u_longlong_t)lr->lrc_seq, (u_longlong_t)txtype); 1194 zilog->zl_stop_replay = 1; 1195 kmem_free(name, MAXNAMELEN); 1196 } 1197 1198 /* 1199 * The DMU's dnode layer doesn't see removes until the txg commits, 1200 * so a subsequent claim can spuriously fail with EEXIST. 1201 * To prevent this, if we might have removed an object, 1202 * wait for the delete thread to delete it, and then 1203 * wait for the transaction group to sync. 1204 */ 1205 if (txtype == TX_REMOVE || txtype == TX_RMDIR || txtype == TX_RENAME) { 1206 if (zr->zr_rm_sync != NULL) 1207 zr->zr_rm_sync(zr->zr_arg); 1208 txg_wait_synced(spa_get_dsl(zilog->zl_spa), 0); 1209 } 1210 } 1211 1212 /* 1213 * If this dataset has an intent log, replay it and destroy it. 1214 */ 1215 void 1216 zil_replay(objset_t *os, void *arg, uint64_t *txgp, 1217 zil_replay_func_t *replay_func[TX_MAX_TYPE], void (*rm_sync)(void *arg)) 1218 { 1219 zilog_t *zilog = dmu_objset_zil(os); 1220 zil_replay_arg_t zr; 1221 1222 zr.zr_os = os; 1223 zr.zr_replay = replay_func; 1224 zr.zr_arg = arg; 1225 zr.zr_rm_sync = rm_sync; 1226 zr.zr_txgp = txgp; 1227 zr.zr_byteswap = BP_SHOULD_BYTESWAP(&zilog->zl_header->zh_log); 1228 zr.zr_lrbuf = kmem_alloc(2 * SPA_MAXBLOCKSIZE, KM_SLEEP); 1229 1230 /* 1231 * Wait for in-progress removes to sync before starting replay. 1232 */ 1233 if (rm_sync != NULL) 1234 rm_sync(arg); 1235 txg_wait_synced(zilog->zl_dmu_pool, 0); 1236 1237 zilog->zl_stop_replay = 0; 1238 zil_parse(zilog, NULL, zil_replay_log_record, &zr, 1239 zilog->zl_header->zh_claim_txg); 1240 kmem_free(zr.zr_lrbuf, 2 * SPA_MAXBLOCKSIZE); 1241 1242 zil_destroy(zilog); 1243 } 1244