1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 */ 24 25 #include <sys/zfs_context.h> 26 #include <sys/fm/fs/zfs.h> 27 #include <sys/spa.h> 28 #include <sys/txg.h> 29 #include <sys/spa_impl.h> 30 #include <sys/vdev_impl.h> 31 #include <sys/zio_impl.h> 32 #include <sys/zio_compress.h> 33 #include <sys/zio_checksum.h> 34 #include <sys/dmu_objset.h> 35 #include <sys/arc.h> 36 #include <sys/ddt.h> 37 38 /* 39 * ========================================================================== 40 * I/O priority table 41 * ========================================================================== 42 */ 43 uint8_t zio_priority_table[ZIO_PRIORITY_TABLE_SIZE] = { 44 0, /* ZIO_PRIORITY_NOW */ 45 0, /* ZIO_PRIORITY_SYNC_READ */ 46 0, /* ZIO_PRIORITY_SYNC_WRITE */ 47 0, /* ZIO_PRIORITY_LOG_WRITE */ 48 1, /* ZIO_PRIORITY_CACHE_FILL */ 49 1, /* ZIO_PRIORITY_AGG */ 50 4, /* ZIO_PRIORITY_FREE */ 51 4, /* ZIO_PRIORITY_ASYNC_WRITE */ 52 6, /* ZIO_PRIORITY_ASYNC_READ */ 53 10, /* ZIO_PRIORITY_RESILVER */ 54 20, /* ZIO_PRIORITY_SCRUB */ 55 2, /* ZIO_PRIORITY_DDT_PREFETCH */ 56 }; 57 58 /* 59 * ========================================================================== 60 * I/O type descriptions 61 * ========================================================================== 62 */ 63 char *zio_type_name[ZIO_TYPES] = { 64 "zio_null", "zio_read", "zio_write", "zio_free", "zio_claim", 65 "zio_ioctl" 66 }; 67 68 /* 69 * ========================================================================== 70 * I/O kmem caches 71 * ========================================================================== 72 */ 73 kmem_cache_t *zio_cache; 74 kmem_cache_t *zio_link_cache; 75 kmem_cache_t *zio_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT]; 76 kmem_cache_t *zio_data_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT]; 77 78 #ifdef _KERNEL 79 extern vmem_t *zio_alloc_arena; 80 #endif 81 82 /* 83 * An allocating zio is one that either currently has the DVA allocate 84 * stage set or will have it later in its lifetime. 85 */ 86 #define IO_IS_ALLOCATING(zio) ((zio)->io_orig_pipeline & ZIO_STAGE_DVA_ALLOCATE) 87 88 boolean_t zio_requeue_io_start_cut_in_line = B_TRUE; 89 90 #ifdef ZFS_DEBUG 91 int zio_buf_debug_limit = 16384; 92 #else 93 int zio_buf_debug_limit = 0; 94 #endif 95 96 void 97 zio_init(void) 98 { 99 size_t c; 100 vmem_t *data_alloc_arena = NULL; 101 102 #ifdef _KERNEL 103 data_alloc_arena = zio_alloc_arena; 104 #endif 105 zio_cache = kmem_cache_create("zio_cache", 106 sizeof (zio_t), 0, NULL, NULL, NULL, NULL, NULL, 0); 107 zio_link_cache = kmem_cache_create("zio_link_cache", 108 sizeof (zio_link_t), 0, NULL, NULL, NULL, NULL, NULL, 0); 109 110 /* 111 * For small buffers, we want a cache for each multiple of 112 * SPA_MINBLOCKSIZE. For medium-size buffers, we want a cache 113 * for each quarter-power of 2. For large buffers, we want 114 * a cache for each multiple of PAGESIZE. 115 */ 116 for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) { 117 size_t size = (c + 1) << SPA_MINBLOCKSHIFT; 118 size_t p2 = size; 119 size_t align = 0; 120 size_t cflags = (size > zio_buf_debug_limit) ? KMC_NODEBUG : 0; 121 122 while (p2 & (p2 - 1)) 123 p2 &= p2 - 1; 124 125 if (size <= 4 * SPA_MINBLOCKSIZE) { 126 align = SPA_MINBLOCKSIZE; 127 } else if (P2PHASE(size, PAGESIZE) == 0) { 128 align = PAGESIZE; 129 } else if (P2PHASE(size, p2 >> 2) == 0) { 130 align = p2 >> 2; 131 } 132 133 if (align != 0) { 134 char name[36]; 135 (void) sprintf(name, "zio_buf_%lu", (ulong_t)size); 136 zio_buf_cache[c] = kmem_cache_create(name, size, 137 align, NULL, NULL, NULL, NULL, NULL, cflags); 138 139 /* 140 * Since zio_data bufs do not appear in crash dumps, we 141 * pass KMC_NOTOUCH so that no allocator metadata is 142 * stored with the buffers. 143 */ 144 (void) sprintf(name, "zio_data_buf_%lu", (ulong_t)size); 145 zio_data_buf_cache[c] = kmem_cache_create(name, size, 146 align, NULL, NULL, NULL, NULL, data_alloc_arena, 147 cflags | KMC_NOTOUCH); 148 } 149 } 150 151 while (--c != 0) { 152 ASSERT(zio_buf_cache[c] != NULL); 153 if (zio_buf_cache[c - 1] == NULL) 154 zio_buf_cache[c - 1] = zio_buf_cache[c]; 155 156 ASSERT(zio_data_buf_cache[c] != NULL); 157 if (zio_data_buf_cache[c - 1] == NULL) 158 zio_data_buf_cache[c - 1] = zio_data_buf_cache[c]; 159 } 160 161 zio_inject_init(); 162 } 163 164 void 165 zio_fini(void) 166 { 167 size_t c; 168 kmem_cache_t *last_cache = NULL; 169 kmem_cache_t *last_data_cache = NULL; 170 171 for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) { 172 if (zio_buf_cache[c] != last_cache) { 173 last_cache = zio_buf_cache[c]; 174 kmem_cache_destroy(zio_buf_cache[c]); 175 } 176 zio_buf_cache[c] = NULL; 177 178 if (zio_data_buf_cache[c] != last_data_cache) { 179 last_data_cache = zio_data_buf_cache[c]; 180 kmem_cache_destroy(zio_data_buf_cache[c]); 181 } 182 zio_data_buf_cache[c] = NULL; 183 } 184 185 kmem_cache_destroy(zio_link_cache); 186 kmem_cache_destroy(zio_cache); 187 188 zio_inject_fini(); 189 } 190 191 /* 192 * ========================================================================== 193 * Allocate and free I/O buffers 194 * ========================================================================== 195 */ 196 197 /* 198 * Use zio_buf_alloc to allocate ZFS metadata. This data will appear in a 199 * crashdump if the kernel panics, so use it judiciously. Obviously, it's 200 * useful to inspect ZFS metadata, but if possible, we should avoid keeping 201 * excess / transient data in-core during a crashdump. 202 */ 203 void * 204 zio_buf_alloc(size_t size) 205 { 206 size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; 207 208 ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); 209 210 return (kmem_cache_alloc(zio_buf_cache[c], KM_PUSHPAGE)); 211 } 212 213 /* 214 * Use zio_data_buf_alloc to allocate data. The data will not appear in a 215 * crashdump if the kernel panics. This exists so that we will limit the amount 216 * of ZFS data that shows up in a kernel crashdump. (Thus reducing the amount 217 * of kernel heap dumped to disk when the kernel panics) 218 */ 219 void * 220 zio_data_buf_alloc(size_t size) 221 { 222 size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; 223 224 ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); 225 226 return (kmem_cache_alloc(zio_data_buf_cache[c], KM_PUSHPAGE)); 227 } 228 229 void 230 zio_buf_free(void *buf, size_t size) 231 { 232 size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; 233 234 ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); 235 236 kmem_cache_free(zio_buf_cache[c], buf); 237 } 238 239 void 240 zio_data_buf_free(void *buf, size_t size) 241 { 242 size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; 243 244 ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); 245 246 kmem_cache_free(zio_data_buf_cache[c], buf); 247 } 248 249 /* 250 * ========================================================================== 251 * Push and pop I/O transform buffers 252 * ========================================================================== 253 */ 254 static void 255 zio_push_transform(zio_t *zio, void *data, uint64_t size, uint64_t bufsize, 256 zio_transform_func_t *transform) 257 { 258 zio_transform_t *zt = kmem_alloc(sizeof (zio_transform_t), KM_SLEEP); 259 260 zt->zt_orig_data = zio->io_data; 261 zt->zt_orig_size = zio->io_size; 262 zt->zt_bufsize = bufsize; 263 zt->zt_transform = transform; 264 265 zt->zt_next = zio->io_transform_stack; 266 zio->io_transform_stack = zt; 267 268 zio->io_data = data; 269 zio->io_size = size; 270 } 271 272 static void 273 zio_pop_transforms(zio_t *zio) 274 { 275 zio_transform_t *zt; 276 277 while ((zt = zio->io_transform_stack) != NULL) { 278 if (zt->zt_transform != NULL) 279 zt->zt_transform(zio, 280 zt->zt_orig_data, zt->zt_orig_size); 281 282 if (zt->zt_bufsize != 0) 283 zio_buf_free(zio->io_data, zt->zt_bufsize); 284 285 zio->io_data = zt->zt_orig_data; 286 zio->io_size = zt->zt_orig_size; 287 zio->io_transform_stack = zt->zt_next; 288 289 kmem_free(zt, sizeof (zio_transform_t)); 290 } 291 } 292 293 /* 294 * ========================================================================== 295 * I/O transform callbacks for subblocks and decompression 296 * ========================================================================== 297 */ 298 static void 299 zio_subblock(zio_t *zio, void *data, uint64_t size) 300 { 301 ASSERT(zio->io_size > size); 302 303 if (zio->io_type == ZIO_TYPE_READ) 304 bcopy(zio->io_data, data, size); 305 } 306 307 static void 308 zio_decompress(zio_t *zio, void *data, uint64_t size) 309 { 310 if (zio->io_error == 0 && 311 zio_decompress_data(BP_GET_COMPRESS(zio->io_bp), 312 zio->io_data, data, zio->io_size, size) != 0) 313 zio->io_error = EIO; 314 } 315 316 /* 317 * ========================================================================== 318 * I/O parent/child relationships and pipeline interlocks 319 * ========================================================================== 320 */ 321 /* 322 * NOTE - Callers to zio_walk_parents() and zio_walk_children must 323 * continue calling these functions until they return NULL. 324 * Otherwise, the next caller will pick up the list walk in 325 * some indeterminate state. (Otherwise every caller would 326 * have to pass in a cookie to keep the state represented by 327 * io_walk_link, which gets annoying.) 328 */ 329 zio_t * 330 zio_walk_parents(zio_t *cio) 331 { 332 zio_link_t *zl = cio->io_walk_link; 333 list_t *pl = &cio->io_parent_list; 334 335 zl = (zl == NULL) ? list_head(pl) : list_next(pl, zl); 336 cio->io_walk_link = zl; 337 338 if (zl == NULL) 339 return (NULL); 340 341 ASSERT(zl->zl_child == cio); 342 return (zl->zl_parent); 343 } 344 345 zio_t * 346 zio_walk_children(zio_t *pio) 347 { 348 zio_link_t *zl = pio->io_walk_link; 349 list_t *cl = &pio->io_child_list; 350 351 zl = (zl == NULL) ? list_head(cl) : list_next(cl, zl); 352 pio->io_walk_link = zl; 353 354 if (zl == NULL) 355 return (NULL); 356 357 ASSERT(zl->zl_parent == pio); 358 return (zl->zl_child); 359 } 360 361 zio_t * 362 zio_unique_parent(zio_t *cio) 363 { 364 zio_t *pio = zio_walk_parents(cio); 365 366 VERIFY(zio_walk_parents(cio) == NULL); 367 return (pio); 368 } 369 370 void 371 zio_add_child(zio_t *pio, zio_t *cio) 372 { 373 zio_link_t *zl = kmem_cache_alloc(zio_link_cache, KM_SLEEP); 374 375 /* 376 * Logical I/Os can have logical, gang, or vdev children. 377 * Gang I/Os can have gang or vdev children. 378 * Vdev I/Os can only have vdev children. 379 * The following ASSERT captures all of these constraints. 380 */ 381 ASSERT(cio->io_child_type <= pio->io_child_type); 382 383 zl->zl_parent = pio; 384 zl->zl_child = cio; 385 386 mutex_enter(&cio->io_lock); 387 mutex_enter(&pio->io_lock); 388 389 ASSERT(pio->io_state[ZIO_WAIT_DONE] == 0); 390 391 for (int w = 0; w < ZIO_WAIT_TYPES; w++) 392 pio->io_children[cio->io_child_type][w] += !cio->io_state[w]; 393 394 list_insert_head(&pio->io_child_list, zl); 395 list_insert_head(&cio->io_parent_list, zl); 396 397 pio->io_child_count++; 398 cio->io_parent_count++; 399 400 mutex_exit(&pio->io_lock); 401 mutex_exit(&cio->io_lock); 402 } 403 404 static void 405 zio_remove_child(zio_t *pio, zio_t *cio, zio_link_t *zl) 406 { 407 ASSERT(zl->zl_parent == pio); 408 ASSERT(zl->zl_child == cio); 409 410 mutex_enter(&cio->io_lock); 411 mutex_enter(&pio->io_lock); 412 413 list_remove(&pio->io_child_list, zl); 414 list_remove(&cio->io_parent_list, zl); 415 416 pio->io_child_count--; 417 cio->io_parent_count--; 418 419 mutex_exit(&pio->io_lock); 420 mutex_exit(&cio->io_lock); 421 422 kmem_cache_free(zio_link_cache, zl); 423 } 424 425 static boolean_t 426 zio_wait_for_children(zio_t *zio, enum zio_child child, enum zio_wait_type wait) 427 { 428 uint64_t *countp = &zio->io_children[child][wait]; 429 boolean_t waiting = B_FALSE; 430 431 mutex_enter(&zio->io_lock); 432 ASSERT(zio->io_stall == NULL); 433 if (*countp != 0) { 434 zio->io_stage >>= 1; 435 zio->io_stall = countp; 436 waiting = B_TRUE; 437 } 438 mutex_exit(&zio->io_lock); 439 440 return (waiting); 441 } 442 443 static void 444 zio_notify_parent(zio_t *pio, zio_t *zio, enum zio_wait_type wait) 445 { 446 uint64_t *countp = &pio->io_children[zio->io_child_type][wait]; 447 int *errorp = &pio->io_child_error[zio->io_child_type]; 448 449 mutex_enter(&pio->io_lock); 450 if (zio->io_error && !(zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 451 *errorp = zio_worst_error(*errorp, zio->io_error); 452 pio->io_reexecute |= zio->io_reexecute; 453 ASSERT3U(*countp, >, 0); 454 if (--*countp == 0 && pio->io_stall == countp) { 455 pio->io_stall = NULL; 456 mutex_exit(&pio->io_lock); 457 zio_execute(pio); 458 } else { 459 mutex_exit(&pio->io_lock); 460 } 461 } 462 463 static void 464 zio_inherit_child_errors(zio_t *zio, enum zio_child c) 465 { 466 if (zio->io_child_error[c] != 0 && zio->io_error == 0) 467 zio->io_error = zio->io_child_error[c]; 468 } 469 470 /* 471 * ========================================================================== 472 * Create the various types of I/O (read, write, free, etc) 473 * ========================================================================== 474 */ 475 static zio_t * 476 zio_create(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, 477 void *data, uint64_t size, zio_done_func_t *done, void *private, 478 zio_type_t type, int priority, enum zio_flag flags, 479 vdev_t *vd, uint64_t offset, const zbookmark_t *zb, 480 enum zio_stage stage, enum zio_stage pipeline) 481 { 482 zio_t *zio; 483 484 ASSERT3U(size, <=, SPA_MAXBLOCKSIZE); 485 ASSERT(P2PHASE(size, SPA_MINBLOCKSIZE) == 0); 486 ASSERT(P2PHASE(offset, SPA_MINBLOCKSIZE) == 0); 487 488 ASSERT(!vd || spa_config_held(spa, SCL_STATE_ALL, RW_READER)); 489 ASSERT(!bp || !(flags & ZIO_FLAG_CONFIG_WRITER)); 490 ASSERT(vd || stage == ZIO_STAGE_OPEN); 491 492 zio = kmem_cache_alloc(zio_cache, KM_SLEEP); 493 bzero(zio, sizeof (zio_t)); 494 495 mutex_init(&zio->io_lock, NULL, MUTEX_DEFAULT, NULL); 496 cv_init(&zio->io_cv, NULL, CV_DEFAULT, NULL); 497 498 list_create(&zio->io_parent_list, sizeof (zio_link_t), 499 offsetof(zio_link_t, zl_parent_node)); 500 list_create(&zio->io_child_list, sizeof (zio_link_t), 501 offsetof(zio_link_t, zl_child_node)); 502 503 if (vd != NULL) 504 zio->io_child_type = ZIO_CHILD_VDEV; 505 else if (flags & ZIO_FLAG_GANG_CHILD) 506 zio->io_child_type = ZIO_CHILD_GANG; 507 else if (flags & ZIO_FLAG_DDT_CHILD) 508 zio->io_child_type = ZIO_CHILD_DDT; 509 else 510 zio->io_child_type = ZIO_CHILD_LOGICAL; 511 512 if (bp != NULL) { 513 zio->io_bp = (blkptr_t *)bp; 514 zio->io_bp_copy = *bp; 515 zio->io_bp_orig = *bp; 516 if (type != ZIO_TYPE_WRITE || 517 zio->io_child_type == ZIO_CHILD_DDT) 518 zio->io_bp = &zio->io_bp_copy; /* so caller can free */ 519 if (zio->io_child_type == ZIO_CHILD_LOGICAL) 520 zio->io_logical = zio; 521 if (zio->io_child_type > ZIO_CHILD_GANG && BP_IS_GANG(bp)) 522 pipeline |= ZIO_GANG_STAGES; 523 } 524 525 zio->io_spa = spa; 526 zio->io_txg = txg; 527 zio->io_done = done; 528 zio->io_private = private; 529 zio->io_type = type; 530 zio->io_priority = priority; 531 zio->io_vd = vd; 532 zio->io_offset = offset; 533 zio->io_orig_data = zio->io_data = data; 534 zio->io_orig_size = zio->io_size = size; 535 zio->io_orig_flags = zio->io_flags = flags; 536 zio->io_orig_stage = zio->io_stage = stage; 537 zio->io_orig_pipeline = zio->io_pipeline = pipeline; 538 539 zio->io_state[ZIO_WAIT_READY] = (stage >= ZIO_STAGE_READY); 540 zio->io_state[ZIO_WAIT_DONE] = (stage >= ZIO_STAGE_DONE); 541 542 if (zb != NULL) 543 zio->io_bookmark = *zb; 544 545 if (pio != NULL) { 546 if (zio->io_logical == NULL) 547 zio->io_logical = pio->io_logical; 548 if (zio->io_child_type == ZIO_CHILD_GANG) 549 zio->io_gang_leader = pio->io_gang_leader; 550 zio_add_child(pio, zio); 551 } 552 553 return (zio); 554 } 555 556 static void 557 zio_destroy(zio_t *zio) 558 { 559 list_destroy(&zio->io_parent_list); 560 list_destroy(&zio->io_child_list); 561 mutex_destroy(&zio->io_lock); 562 cv_destroy(&zio->io_cv); 563 kmem_cache_free(zio_cache, zio); 564 } 565 566 zio_t * 567 zio_null(zio_t *pio, spa_t *spa, vdev_t *vd, zio_done_func_t *done, 568 void *private, enum zio_flag flags) 569 { 570 zio_t *zio; 571 572 zio = zio_create(pio, spa, 0, NULL, NULL, 0, done, private, 573 ZIO_TYPE_NULL, ZIO_PRIORITY_NOW, flags, vd, 0, NULL, 574 ZIO_STAGE_OPEN, ZIO_INTERLOCK_PIPELINE); 575 576 return (zio); 577 } 578 579 zio_t * 580 zio_root(spa_t *spa, zio_done_func_t *done, void *private, enum zio_flag flags) 581 { 582 return (zio_null(NULL, spa, NULL, done, private, flags)); 583 } 584 585 zio_t * 586 zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, 587 void *data, uint64_t size, zio_done_func_t *done, void *private, 588 int priority, enum zio_flag flags, const zbookmark_t *zb) 589 { 590 zio_t *zio; 591 592 zio = zio_create(pio, spa, BP_PHYSICAL_BIRTH(bp), bp, 593 data, size, done, private, 594 ZIO_TYPE_READ, priority, flags, NULL, 0, zb, 595 ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ? 596 ZIO_DDT_CHILD_READ_PIPELINE : ZIO_READ_PIPELINE); 597 598 return (zio); 599 } 600 601 zio_t * 602 zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, 603 void *data, uint64_t size, const zio_prop_t *zp, 604 zio_done_func_t *ready, zio_done_func_t *done, void *private, 605 int priority, enum zio_flag flags, const zbookmark_t *zb) 606 { 607 zio_t *zio; 608 609 ASSERT(zp->zp_checksum >= ZIO_CHECKSUM_OFF && 610 zp->zp_checksum < ZIO_CHECKSUM_FUNCTIONS && 611 zp->zp_compress >= ZIO_COMPRESS_OFF && 612 zp->zp_compress < ZIO_COMPRESS_FUNCTIONS && 613 zp->zp_type < DMU_OT_NUMTYPES && 614 zp->zp_level < 32 && 615 zp->zp_copies > 0 && 616 zp->zp_copies <= spa_max_replication(spa) && 617 zp->zp_dedup <= 1 && 618 zp->zp_dedup_verify <= 1); 619 620 zio = zio_create(pio, spa, txg, bp, data, size, done, private, 621 ZIO_TYPE_WRITE, priority, flags, NULL, 0, zb, 622 ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ? 623 ZIO_DDT_CHILD_WRITE_PIPELINE : ZIO_WRITE_PIPELINE); 624 625 zio->io_ready = ready; 626 zio->io_prop = *zp; 627 628 return (zio); 629 } 630 631 zio_t * 632 zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, void *data, 633 uint64_t size, zio_done_func_t *done, void *private, int priority, 634 enum zio_flag flags, zbookmark_t *zb) 635 { 636 zio_t *zio; 637 638 zio = zio_create(pio, spa, txg, bp, data, size, done, private, 639 ZIO_TYPE_WRITE, priority, flags, NULL, 0, zb, 640 ZIO_STAGE_OPEN, ZIO_REWRITE_PIPELINE); 641 642 return (zio); 643 } 644 645 void 646 zio_write_override(zio_t *zio, blkptr_t *bp, int copies) 647 { 648 ASSERT(zio->io_type == ZIO_TYPE_WRITE); 649 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 650 ASSERT(zio->io_stage == ZIO_STAGE_OPEN); 651 ASSERT(zio->io_txg == spa_syncing_txg(zio->io_spa)); 652 653 zio->io_prop.zp_copies = copies; 654 zio->io_bp_override = bp; 655 } 656 657 void 658 zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp) 659 { 660 bplist_append(&spa->spa_free_bplist[txg & TXG_MASK], bp); 661 } 662 663 zio_t * 664 zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, 665 enum zio_flag flags) 666 { 667 zio_t *zio; 668 669 dprintf_bp(bp, "freeing in txg %llu, pass %u", 670 (longlong_t)txg, spa->spa_sync_pass); 671 672 ASSERT(!BP_IS_HOLE(bp)); 673 ASSERT(spa_syncing_txg(spa) == txg); 674 ASSERT(spa_sync_pass(spa) <= SYNC_PASS_DEFERRED_FREE); 675 676 zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp), 677 NULL, NULL, ZIO_TYPE_FREE, ZIO_PRIORITY_FREE, flags, 678 NULL, 0, NULL, ZIO_STAGE_OPEN, ZIO_FREE_PIPELINE); 679 680 return (zio); 681 } 682 683 zio_t * 684 zio_claim(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, 685 zio_done_func_t *done, void *private, enum zio_flag flags) 686 { 687 zio_t *zio; 688 689 /* 690 * A claim is an allocation of a specific block. Claims are needed 691 * to support immediate writes in the intent log. The issue is that 692 * immediate writes contain committed data, but in a txg that was 693 * *not* committed. Upon opening the pool after an unclean shutdown, 694 * the intent log claims all blocks that contain immediate write data 695 * so that the SPA knows they're in use. 696 * 697 * All claims *must* be resolved in the first txg -- before the SPA 698 * starts allocating blocks -- so that nothing is allocated twice. 699 * If txg == 0 we just verify that the block is claimable. 700 */ 701 ASSERT3U(spa->spa_uberblock.ub_rootbp.blk_birth, <, spa_first_txg(spa)); 702 ASSERT(txg == spa_first_txg(spa) || txg == 0); 703 ASSERT(!BP_GET_DEDUP(bp) || !spa_writeable(spa)); /* zdb(1M) */ 704 705 zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp), 706 done, private, ZIO_TYPE_CLAIM, ZIO_PRIORITY_NOW, flags, 707 NULL, 0, NULL, ZIO_STAGE_OPEN, ZIO_CLAIM_PIPELINE); 708 709 return (zio); 710 } 711 712 zio_t * 713 zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd, 714 zio_done_func_t *done, void *private, int priority, enum zio_flag flags) 715 { 716 zio_t *zio; 717 int c; 718 719 if (vd->vdev_children == 0) { 720 zio = zio_create(pio, spa, 0, NULL, NULL, 0, done, private, 721 ZIO_TYPE_IOCTL, priority, flags, vd, 0, NULL, 722 ZIO_STAGE_OPEN, ZIO_IOCTL_PIPELINE); 723 724 zio->io_cmd = cmd; 725 } else { 726 zio = zio_null(pio, spa, NULL, NULL, NULL, flags); 727 728 for (c = 0; c < vd->vdev_children; c++) 729 zio_nowait(zio_ioctl(zio, spa, vd->vdev_child[c], cmd, 730 done, private, priority, flags)); 731 } 732 733 return (zio); 734 } 735 736 zio_t * 737 zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, 738 void *data, int checksum, zio_done_func_t *done, void *private, 739 int priority, enum zio_flag flags, boolean_t labels) 740 { 741 zio_t *zio; 742 743 ASSERT(vd->vdev_children == 0); 744 ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE || 745 offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE); 746 ASSERT3U(offset + size, <=, vd->vdev_psize); 747 748 zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private, 749 ZIO_TYPE_READ, priority, flags, vd, offset, NULL, 750 ZIO_STAGE_OPEN, ZIO_READ_PHYS_PIPELINE); 751 752 zio->io_prop.zp_checksum = checksum; 753 754 return (zio); 755 } 756 757 zio_t * 758 zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, 759 void *data, int checksum, zio_done_func_t *done, void *private, 760 int priority, enum zio_flag flags, boolean_t labels) 761 { 762 zio_t *zio; 763 764 ASSERT(vd->vdev_children == 0); 765 ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE || 766 offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE); 767 ASSERT3U(offset + size, <=, vd->vdev_psize); 768 769 zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private, 770 ZIO_TYPE_WRITE, priority, flags, vd, offset, NULL, 771 ZIO_STAGE_OPEN, ZIO_WRITE_PHYS_PIPELINE); 772 773 zio->io_prop.zp_checksum = checksum; 774 775 if (zio_checksum_table[checksum].ci_eck) { 776 /* 777 * zec checksums are necessarily destructive -- they modify 778 * the end of the write buffer to hold the verifier/checksum. 779 * Therefore, we must make a local copy in case the data is 780 * being written to multiple places in parallel. 781 */ 782 void *wbuf = zio_buf_alloc(size); 783 bcopy(data, wbuf, size); 784 zio_push_transform(zio, wbuf, size, size, NULL); 785 } 786 787 return (zio); 788 } 789 790 /* 791 * Create a child I/O to do some work for us. 792 */ 793 zio_t * 794 zio_vdev_child_io(zio_t *pio, blkptr_t *bp, vdev_t *vd, uint64_t offset, 795 void *data, uint64_t size, int type, int priority, enum zio_flag flags, 796 zio_done_func_t *done, void *private) 797 { 798 enum zio_stage pipeline = ZIO_VDEV_CHILD_PIPELINE; 799 zio_t *zio; 800 801 ASSERT(vd->vdev_parent == 802 (pio->io_vd ? pio->io_vd : pio->io_spa->spa_root_vdev)); 803 804 if (type == ZIO_TYPE_READ && bp != NULL) { 805 /* 806 * If we have the bp, then the child should perform the 807 * checksum and the parent need not. This pushes error 808 * detection as close to the leaves as possible and 809 * eliminates redundant checksums in the interior nodes. 810 */ 811 pipeline |= ZIO_STAGE_CHECKSUM_VERIFY; 812 pio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY; 813 } 814 815 if (vd->vdev_children == 0) 816 offset += VDEV_LABEL_START_SIZE; 817 818 flags |= ZIO_VDEV_CHILD_FLAGS(pio) | ZIO_FLAG_DONT_PROPAGATE; 819 820 /* 821 * If we've decided to do a repair, the write is not speculative -- 822 * even if the original read was. 823 */ 824 if (flags & ZIO_FLAG_IO_REPAIR) 825 flags &= ~ZIO_FLAG_SPECULATIVE; 826 827 zio = zio_create(pio, pio->io_spa, pio->io_txg, bp, data, size, 828 done, private, type, priority, flags, vd, offset, &pio->io_bookmark, 829 ZIO_STAGE_VDEV_IO_START >> 1, pipeline); 830 831 return (zio); 832 } 833 834 zio_t * 835 zio_vdev_delegated_io(vdev_t *vd, uint64_t offset, void *data, uint64_t size, 836 int type, int priority, enum zio_flag flags, 837 zio_done_func_t *done, void *private) 838 { 839 zio_t *zio; 840 841 ASSERT(vd->vdev_ops->vdev_op_leaf); 842 843 zio = zio_create(NULL, vd->vdev_spa, 0, NULL, 844 data, size, done, private, type, priority, 845 flags | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY, 846 vd, offset, NULL, 847 ZIO_STAGE_VDEV_IO_START >> 1, ZIO_VDEV_CHILD_PIPELINE); 848 849 return (zio); 850 } 851 852 void 853 zio_flush(zio_t *zio, vdev_t *vd) 854 { 855 zio_nowait(zio_ioctl(zio, zio->io_spa, vd, DKIOCFLUSHWRITECACHE, 856 NULL, NULL, ZIO_PRIORITY_NOW, 857 ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY)); 858 } 859 860 void 861 zio_shrink(zio_t *zio, uint64_t size) 862 { 863 ASSERT(zio->io_executor == NULL); 864 ASSERT(zio->io_orig_size == zio->io_size); 865 ASSERT(size <= zio->io_size); 866 867 /* 868 * We don't shrink for raidz because of problems with the 869 * reconstruction when reading back less than the block size. 870 * Note, BP_IS_RAIDZ() assumes no compression. 871 */ 872 ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF); 873 if (!BP_IS_RAIDZ(zio->io_bp)) 874 zio->io_orig_size = zio->io_size = size; 875 } 876 877 /* 878 * ========================================================================== 879 * Prepare to read and write logical blocks 880 * ========================================================================== 881 */ 882 883 static int 884 zio_read_bp_init(zio_t *zio) 885 { 886 blkptr_t *bp = zio->io_bp; 887 888 if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_OFF && 889 zio->io_child_type == ZIO_CHILD_LOGICAL && 890 !(zio->io_flags & ZIO_FLAG_RAW)) { 891 uint64_t psize = BP_GET_PSIZE(bp); 892 void *cbuf = zio_buf_alloc(psize); 893 894 zio_push_transform(zio, cbuf, psize, psize, zio_decompress); 895 } 896 897 if (!dmu_ot[BP_GET_TYPE(bp)].ot_metadata && BP_GET_LEVEL(bp) == 0) 898 zio->io_flags |= ZIO_FLAG_DONT_CACHE; 899 900 if (BP_GET_TYPE(bp) == DMU_OT_DDT_ZAP) 901 zio->io_flags |= ZIO_FLAG_DONT_CACHE; 902 903 if (BP_GET_DEDUP(bp) && zio->io_child_type == ZIO_CHILD_LOGICAL) 904 zio->io_pipeline = ZIO_DDT_READ_PIPELINE; 905 906 return (ZIO_PIPELINE_CONTINUE); 907 } 908 909 static int 910 zio_write_bp_init(zio_t *zio) 911 { 912 spa_t *spa = zio->io_spa; 913 zio_prop_t *zp = &zio->io_prop; 914 enum zio_compress compress = zp->zp_compress; 915 blkptr_t *bp = zio->io_bp; 916 uint64_t lsize = zio->io_size; 917 uint64_t psize = lsize; 918 int pass = 1; 919 920 /* 921 * If our children haven't all reached the ready stage, 922 * wait for them and then repeat this pipeline stage. 923 */ 924 if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY) || 925 zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_READY)) 926 return (ZIO_PIPELINE_STOP); 927 928 if (!IO_IS_ALLOCATING(zio)) 929 return (ZIO_PIPELINE_CONTINUE); 930 931 ASSERT(zio->io_child_type != ZIO_CHILD_DDT); 932 933 if (zio->io_bp_override) { 934 ASSERT(bp->blk_birth != zio->io_txg); 935 ASSERT(BP_GET_DEDUP(zio->io_bp_override) == 0); 936 937 *bp = *zio->io_bp_override; 938 zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; 939 940 if (BP_IS_HOLE(bp) || !zp->zp_dedup) 941 return (ZIO_PIPELINE_CONTINUE); 942 943 ASSERT(zio_checksum_table[zp->zp_checksum].ci_dedup || 944 zp->zp_dedup_verify); 945 946 if (BP_GET_CHECKSUM(bp) == zp->zp_checksum) { 947 BP_SET_DEDUP(bp, 1); 948 zio->io_pipeline |= ZIO_STAGE_DDT_WRITE; 949 return (ZIO_PIPELINE_CONTINUE); 950 } 951 zio->io_bp_override = NULL; 952 BP_ZERO(bp); 953 } 954 955 if (bp->blk_birth == zio->io_txg) { 956 /* 957 * We're rewriting an existing block, which means we're 958 * working on behalf of spa_sync(). For spa_sync() to 959 * converge, it must eventually be the case that we don't 960 * have to allocate new blocks. But compression changes 961 * the blocksize, which forces a reallocate, and makes 962 * convergence take longer. Therefore, after the first 963 * few passes, stop compressing to ensure convergence. 964 */ 965 pass = spa_sync_pass(spa); 966 967 ASSERT(zio->io_txg == spa_syncing_txg(spa)); 968 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 969 ASSERT(!BP_GET_DEDUP(bp)); 970 971 if (pass > SYNC_PASS_DONT_COMPRESS) 972 compress = ZIO_COMPRESS_OFF; 973 974 /* Make sure someone doesn't change their mind on overwrites */ 975 ASSERT(MIN(zp->zp_copies + BP_IS_GANG(bp), 976 spa_max_replication(spa)) == BP_GET_NDVAS(bp)); 977 } 978 979 if (compress != ZIO_COMPRESS_OFF) { 980 void *cbuf = zio_buf_alloc(lsize); 981 psize = zio_compress_data(compress, zio->io_data, cbuf, lsize); 982 if (psize == 0 || psize == lsize) { 983 compress = ZIO_COMPRESS_OFF; 984 zio_buf_free(cbuf, lsize); 985 } else { 986 ASSERT(psize < lsize); 987 zio_push_transform(zio, cbuf, psize, lsize, NULL); 988 } 989 } 990 991 /* 992 * The final pass of spa_sync() must be all rewrites, but the first 993 * few passes offer a trade-off: allocating blocks defers convergence, 994 * but newly allocated blocks are sequential, so they can be written 995 * to disk faster. Therefore, we allow the first few passes of 996 * spa_sync() to allocate new blocks, but force rewrites after that. 997 * There should only be a handful of blocks after pass 1 in any case. 998 */ 999 if (bp->blk_birth == zio->io_txg && BP_GET_PSIZE(bp) == psize && 1000 pass > SYNC_PASS_REWRITE) { 1001 ASSERT(psize != 0); 1002 enum zio_stage gang_stages = zio->io_pipeline & ZIO_GANG_STAGES; 1003 zio->io_pipeline = ZIO_REWRITE_PIPELINE | gang_stages; 1004 zio->io_flags |= ZIO_FLAG_IO_REWRITE; 1005 } else { 1006 BP_ZERO(bp); 1007 zio->io_pipeline = ZIO_WRITE_PIPELINE; 1008 } 1009 1010 if (psize == 0) { 1011 zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; 1012 } else { 1013 ASSERT(zp->zp_checksum != ZIO_CHECKSUM_GANG_HEADER); 1014 BP_SET_LSIZE(bp, lsize); 1015 BP_SET_PSIZE(bp, psize); 1016 BP_SET_COMPRESS(bp, compress); 1017 BP_SET_CHECKSUM(bp, zp->zp_checksum); 1018 BP_SET_TYPE(bp, zp->zp_type); 1019 BP_SET_LEVEL(bp, zp->zp_level); 1020 BP_SET_DEDUP(bp, zp->zp_dedup); 1021 BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER); 1022 if (zp->zp_dedup) { 1023 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 1024 ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE)); 1025 zio->io_pipeline = ZIO_DDT_WRITE_PIPELINE; 1026 } 1027 } 1028 1029 return (ZIO_PIPELINE_CONTINUE); 1030 } 1031 1032 static int 1033 zio_free_bp_init(zio_t *zio) 1034 { 1035 blkptr_t *bp = zio->io_bp; 1036 1037 if (zio->io_child_type == ZIO_CHILD_LOGICAL) { 1038 if (BP_GET_DEDUP(bp)) 1039 zio->io_pipeline = ZIO_DDT_FREE_PIPELINE; 1040 } 1041 1042 return (ZIO_PIPELINE_CONTINUE); 1043 } 1044 1045 /* 1046 * ========================================================================== 1047 * Execute the I/O pipeline 1048 * ========================================================================== 1049 */ 1050 1051 static void 1052 zio_taskq_dispatch(zio_t *zio, enum zio_taskq_type q, boolean_t cutinline) 1053 { 1054 spa_t *spa = zio->io_spa; 1055 zio_type_t t = zio->io_type; 1056 int flags = TQ_SLEEP | (cutinline ? TQ_FRONT : 0); 1057 1058 /* 1059 * If we're a config writer or a probe, the normal issue and 1060 * interrupt threads may all be blocked waiting for the config lock. 1061 * In this case, select the otherwise-unused taskq for ZIO_TYPE_NULL. 1062 */ 1063 if (zio->io_flags & (ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_PROBE)) 1064 t = ZIO_TYPE_NULL; 1065 1066 /* 1067 * A similar issue exists for the L2ARC write thread until L2ARC 2.0. 1068 */ 1069 if (t == ZIO_TYPE_WRITE && zio->io_vd && zio->io_vd->vdev_aux) 1070 t = ZIO_TYPE_NULL; 1071 1072 /* 1073 * If this is a high priority I/O, then use the high priority taskq. 1074 */ 1075 if (zio->io_priority == ZIO_PRIORITY_NOW && 1076 spa->spa_zio_taskq[t][q + 1] != NULL) 1077 q++; 1078 1079 ASSERT3U(q, <, ZIO_TASKQ_TYPES); 1080 (void) taskq_dispatch(spa->spa_zio_taskq[t][q], 1081 (task_func_t *)zio_execute, zio, flags); 1082 } 1083 1084 static boolean_t 1085 zio_taskq_member(zio_t *zio, enum zio_taskq_type q) 1086 { 1087 kthread_t *executor = zio->io_executor; 1088 spa_t *spa = zio->io_spa; 1089 1090 for (zio_type_t t = 0; t < ZIO_TYPES; t++) 1091 if (taskq_member(spa->spa_zio_taskq[t][q], executor)) 1092 return (B_TRUE); 1093 1094 return (B_FALSE); 1095 } 1096 1097 static int 1098 zio_issue_async(zio_t *zio) 1099 { 1100 zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE); 1101 1102 return (ZIO_PIPELINE_STOP); 1103 } 1104 1105 void 1106 zio_interrupt(zio_t *zio) 1107 { 1108 zio_taskq_dispatch(zio, ZIO_TASKQ_INTERRUPT, B_FALSE); 1109 } 1110 1111 /* 1112 * Execute the I/O pipeline until one of the following occurs: 1113 * (1) the I/O completes; (2) the pipeline stalls waiting for 1114 * dependent child I/Os; (3) the I/O issues, so we're waiting 1115 * for an I/O completion interrupt; (4) the I/O is delegated by 1116 * vdev-level caching or aggregation; (5) the I/O is deferred 1117 * due to vdev-level queueing; (6) the I/O is handed off to 1118 * another thread. In all cases, the pipeline stops whenever 1119 * there's no CPU work; it never burns a thread in cv_wait(). 1120 * 1121 * There's no locking on io_stage because there's no legitimate way 1122 * for multiple threads to be attempting to process the same I/O. 1123 */ 1124 static zio_pipe_stage_t *zio_pipeline[]; 1125 1126 void 1127 zio_execute(zio_t *zio) 1128 { 1129 zio->io_executor = curthread; 1130 1131 while (zio->io_stage < ZIO_STAGE_DONE) { 1132 enum zio_stage pipeline = zio->io_pipeline; 1133 enum zio_stage stage = zio->io_stage; 1134 int rv; 1135 1136 ASSERT(!MUTEX_HELD(&zio->io_lock)); 1137 ASSERT(ISP2(stage)); 1138 ASSERT(zio->io_stall == NULL); 1139 1140 do { 1141 stage <<= 1; 1142 } while ((stage & pipeline) == 0); 1143 1144 ASSERT(stage <= ZIO_STAGE_DONE); 1145 1146 /* 1147 * If we are in interrupt context and this pipeline stage 1148 * will grab a config lock that is held across I/O, 1149 * or may wait for an I/O that needs an interrupt thread 1150 * to complete, issue async to avoid deadlock. 1151 * 1152 * For VDEV_IO_START, we cut in line so that the io will 1153 * be sent to disk promptly. 1154 */ 1155 if ((stage & ZIO_BLOCKING_STAGES) && zio->io_vd == NULL && 1156 zio_taskq_member(zio, ZIO_TASKQ_INTERRUPT)) { 1157 boolean_t cut = (stage == ZIO_STAGE_VDEV_IO_START) ? 1158 zio_requeue_io_start_cut_in_line : B_FALSE; 1159 zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, cut); 1160 return; 1161 } 1162 1163 zio->io_stage = stage; 1164 rv = zio_pipeline[highbit(stage) - 1](zio); 1165 1166 if (rv == ZIO_PIPELINE_STOP) 1167 return; 1168 1169 ASSERT(rv == ZIO_PIPELINE_CONTINUE); 1170 } 1171 } 1172 1173 /* 1174 * ========================================================================== 1175 * Initiate I/O, either sync or async 1176 * ========================================================================== 1177 */ 1178 int 1179 zio_wait(zio_t *zio) 1180 { 1181 int error; 1182 1183 ASSERT(zio->io_stage == ZIO_STAGE_OPEN); 1184 ASSERT(zio->io_executor == NULL); 1185 1186 zio->io_waiter = curthread; 1187 1188 zio_execute(zio); 1189 1190 mutex_enter(&zio->io_lock); 1191 while (zio->io_executor != NULL) 1192 cv_wait(&zio->io_cv, &zio->io_lock); 1193 mutex_exit(&zio->io_lock); 1194 1195 error = zio->io_error; 1196 zio_destroy(zio); 1197 1198 return (error); 1199 } 1200 1201 void 1202 zio_nowait(zio_t *zio) 1203 { 1204 ASSERT(zio->io_executor == NULL); 1205 1206 if (zio->io_child_type == ZIO_CHILD_LOGICAL && 1207 zio_unique_parent(zio) == NULL) { 1208 /* 1209 * This is a logical async I/O with no parent to wait for it. 1210 * We add it to the spa_async_root_zio "Godfather" I/O which 1211 * will ensure they complete prior to unloading the pool. 1212 */ 1213 spa_t *spa = zio->io_spa; 1214 1215 zio_add_child(spa->spa_async_zio_root, zio); 1216 } 1217 1218 zio_execute(zio); 1219 } 1220 1221 /* 1222 * ========================================================================== 1223 * Reexecute or suspend/resume failed I/O 1224 * ========================================================================== 1225 */ 1226 1227 static void 1228 zio_reexecute(zio_t *pio) 1229 { 1230 zio_t *cio, *cio_next; 1231 1232 ASSERT(pio->io_child_type == ZIO_CHILD_LOGICAL); 1233 ASSERT(pio->io_orig_stage == ZIO_STAGE_OPEN); 1234 ASSERT(pio->io_gang_leader == NULL); 1235 ASSERT(pio->io_gang_tree == NULL); 1236 1237 pio->io_flags = pio->io_orig_flags; 1238 pio->io_stage = pio->io_orig_stage; 1239 pio->io_pipeline = pio->io_orig_pipeline; 1240 pio->io_reexecute = 0; 1241 pio->io_error = 0; 1242 for (int w = 0; w < ZIO_WAIT_TYPES; w++) 1243 pio->io_state[w] = 0; 1244 for (int c = 0; c < ZIO_CHILD_TYPES; c++) 1245 pio->io_child_error[c] = 0; 1246 1247 if (IO_IS_ALLOCATING(pio)) 1248 BP_ZERO(pio->io_bp); 1249 1250 /* 1251 * As we reexecute pio's children, new children could be created. 1252 * New children go to the head of pio's io_child_list, however, 1253 * so we will (correctly) not reexecute them. The key is that 1254 * the remainder of pio's io_child_list, from 'cio_next' onward, 1255 * cannot be affected by any side effects of reexecuting 'cio'. 1256 */ 1257 for (cio = zio_walk_children(pio); cio != NULL; cio = cio_next) { 1258 cio_next = zio_walk_children(pio); 1259 mutex_enter(&pio->io_lock); 1260 for (int w = 0; w < ZIO_WAIT_TYPES; w++) 1261 pio->io_children[cio->io_child_type][w]++; 1262 mutex_exit(&pio->io_lock); 1263 zio_reexecute(cio); 1264 } 1265 1266 /* 1267 * Now that all children have been reexecuted, execute the parent. 1268 * We don't reexecute "The Godfather" I/O here as it's the 1269 * responsibility of the caller to wait on him. 1270 */ 1271 if (!(pio->io_flags & ZIO_FLAG_GODFATHER)) 1272 zio_execute(pio); 1273 } 1274 1275 void 1276 zio_suspend(spa_t *spa, zio_t *zio) 1277 { 1278 if (spa_get_failmode(spa) == ZIO_FAILURE_MODE_PANIC) 1279 fm_panic("Pool '%s' has encountered an uncorrectable I/O " 1280 "failure and the failure mode property for this pool " 1281 "is set to panic.", spa_name(spa)); 1282 1283 zfs_ereport_post(FM_EREPORT_ZFS_IO_FAILURE, spa, NULL, NULL, 0, 0); 1284 1285 mutex_enter(&spa->spa_suspend_lock); 1286 1287 if (spa->spa_suspend_zio_root == NULL) 1288 spa->spa_suspend_zio_root = zio_root(spa, NULL, NULL, 1289 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 1290 ZIO_FLAG_GODFATHER); 1291 1292 spa->spa_suspended = B_TRUE; 1293 1294 if (zio != NULL) { 1295 ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER)); 1296 ASSERT(zio != spa->spa_suspend_zio_root); 1297 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 1298 ASSERT(zio_unique_parent(zio) == NULL); 1299 ASSERT(zio->io_stage == ZIO_STAGE_DONE); 1300 zio_add_child(spa->spa_suspend_zio_root, zio); 1301 } 1302 1303 mutex_exit(&spa->spa_suspend_lock); 1304 } 1305 1306 int 1307 zio_resume(spa_t *spa) 1308 { 1309 zio_t *pio; 1310 1311 /* 1312 * Reexecute all previously suspended i/o. 1313 */ 1314 mutex_enter(&spa->spa_suspend_lock); 1315 spa->spa_suspended = B_FALSE; 1316 cv_broadcast(&spa->spa_suspend_cv); 1317 pio = spa->spa_suspend_zio_root; 1318 spa->spa_suspend_zio_root = NULL; 1319 mutex_exit(&spa->spa_suspend_lock); 1320 1321 if (pio == NULL) 1322 return (0); 1323 1324 zio_reexecute(pio); 1325 return (zio_wait(pio)); 1326 } 1327 1328 void 1329 zio_resume_wait(spa_t *spa) 1330 { 1331 mutex_enter(&spa->spa_suspend_lock); 1332 while (spa_suspended(spa)) 1333 cv_wait(&spa->spa_suspend_cv, &spa->spa_suspend_lock); 1334 mutex_exit(&spa->spa_suspend_lock); 1335 } 1336 1337 /* 1338 * ========================================================================== 1339 * Gang blocks. 1340 * 1341 * A gang block is a collection of small blocks that looks to the DMU 1342 * like one large block. When zio_dva_allocate() cannot find a block 1343 * of the requested size, due to either severe fragmentation or the pool 1344 * being nearly full, it calls zio_write_gang_block() to construct the 1345 * block from smaller fragments. 1346 * 1347 * A gang block consists of a gang header (zio_gbh_phys_t) and up to 1348 * three (SPA_GBH_NBLKPTRS) gang members. The gang header is just like 1349 * an indirect block: it's an array of block pointers. It consumes 1350 * only one sector and hence is allocatable regardless of fragmentation. 1351 * The gang header's bps point to its gang members, which hold the data. 1352 * 1353 * Gang blocks are self-checksumming, using the bp's <vdev, offset, txg> 1354 * as the verifier to ensure uniqueness of the SHA256 checksum. 1355 * Critically, the gang block bp's blk_cksum is the checksum of the data, 1356 * not the gang header. This ensures that data block signatures (needed for 1357 * deduplication) are independent of how the block is physically stored. 1358 * 1359 * Gang blocks can be nested: a gang member may itself be a gang block. 1360 * Thus every gang block is a tree in which root and all interior nodes are 1361 * gang headers, and the leaves are normal blocks that contain user data. 1362 * The root of the gang tree is called the gang leader. 1363 * 1364 * To perform any operation (read, rewrite, free, claim) on a gang block, 1365 * zio_gang_assemble() first assembles the gang tree (minus data leaves) 1366 * in the io_gang_tree field of the original logical i/o by recursively 1367 * reading the gang leader and all gang headers below it. This yields 1368 * an in-core tree containing the contents of every gang header and the 1369 * bps for every constituent of the gang block. 1370 * 1371 * With the gang tree now assembled, zio_gang_issue() just walks the gang tree 1372 * and invokes a callback on each bp. To free a gang block, zio_gang_issue() 1373 * calls zio_free_gang() -- a trivial wrapper around zio_free() -- for each bp. 1374 * zio_claim_gang() provides a similarly trivial wrapper for zio_claim(). 1375 * zio_read_gang() is a wrapper around zio_read() that omits reading gang 1376 * headers, since we already have those in io_gang_tree. zio_rewrite_gang() 1377 * performs a zio_rewrite() of the data or, for gang headers, a zio_rewrite() 1378 * of the gang header plus zio_checksum_compute() of the data to update the 1379 * gang header's blk_cksum as described above. 1380 * 1381 * The two-phase assemble/issue model solves the problem of partial failure -- 1382 * what if you'd freed part of a gang block but then couldn't read the 1383 * gang header for another part? Assembling the entire gang tree first 1384 * ensures that all the necessary gang header I/O has succeeded before 1385 * starting the actual work of free, claim, or write. Once the gang tree 1386 * is assembled, free and claim are in-memory operations that cannot fail. 1387 * 1388 * In the event that a gang write fails, zio_dva_unallocate() walks the 1389 * gang tree to immediately free (i.e. insert back into the space map) 1390 * everything we've allocated. This ensures that we don't get ENOSPC 1391 * errors during repeated suspend/resume cycles due to a flaky device. 1392 * 1393 * Gang rewrites only happen during sync-to-convergence. If we can't assemble 1394 * the gang tree, we won't modify the block, so we can safely defer the free 1395 * (knowing that the block is still intact). If we *can* assemble the gang 1396 * tree, then even if some of the rewrites fail, zio_dva_unallocate() will free 1397 * each constituent bp and we can allocate a new block on the next sync pass. 1398 * 1399 * In all cases, the gang tree allows complete recovery from partial failure. 1400 * ========================================================================== 1401 */ 1402 1403 static zio_t * 1404 zio_read_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) 1405 { 1406 if (gn != NULL) 1407 return (pio); 1408 1409 return (zio_read(pio, pio->io_spa, bp, data, BP_GET_PSIZE(bp), 1410 NULL, NULL, pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), 1411 &pio->io_bookmark)); 1412 } 1413 1414 zio_t * 1415 zio_rewrite_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) 1416 { 1417 zio_t *zio; 1418 1419 if (gn != NULL) { 1420 zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp, 1421 gn->gn_gbh, SPA_GANGBLOCKSIZE, NULL, NULL, pio->io_priority, 1422 ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); 1423 /* 1424 * As we rewrite each gang header, the pipeline will compute 1425 * a new gang block header checksum for it; but no one will 1426 * compute a new data checksum, so we do that here. The one 1427 * exception is the gang leader: the pipeline already computed 1428 * its data checksum because that stage precedes gang assembly. 1429 * (Presently, nothing actually uses interior data checksums; 1430 * this is just good hygiene.) 1431 */ 1432 if (gn != pio->io_gang_leader->io_gang_tree) { 1433 zio_checksum_compute(zio, BP_GET_CHECKSUM(bp), 1434 data, BP_GET_PSIZE(bp)); 1435 } 1436 /* 1437 * If we are here to damage data for testing purposes, 1438 * leave the GBH alone so that we can detect the damage. 1439 */ 1440 if (pio->io_gang_leader->io_flags & ZIO_FLAG_INDUCE_DAMAGE) 1441 zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES; 1442 } else { 1443 zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp, 1444 data, BP_GET_PSIZE(bp), NULL, NULL, pio->io_priority, 1445 ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); 1446 } 1447 1448 return (zio); 1449 } 1450 1451 /* ARGSUSED */ 1452 zio_t * 1453 zio_free_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) 1454 { 1455 return (zio_free_sync(pio, pio->io_spa, pio->io_txg, bp, 1456 ZIO_GANG_CHILD_FLAGS(pio))); 1457 } 1458 1459 /* ARGSUSED */ 1460 zio_t * 1461 zio_claim_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) 1462 { 1463 return (zio_claim(pio, pio->io_spa, pio->io_txg, bp, 1464 NULL, NULL, ZIO_GANG_CHILD_FLAGS(pio))); 1465 } 1466 1467 static zio_gang_issue_func_t *zio_gang_issue_func[ZIO_TYPES] = { 1468 NULL, 1469 zio_read_gang, 1470 zio_rewrite_gang, 1471 zio_free_gang, 1472 zio_claim_gang, 1473 NULL 1474 }; 1475 1476 static void zio_gang_tree_assemble_done(zio_t *zio); 1477 1478 static zio_gang_node_t * 1479 zio_gang_node_alloc(zio_gang_node_t **gnpp) 1480 { 1481 zio_gang_node_t *gn; 1482 1483 ASSERT(*gnpp == NULL); 1484 1485 gn = kmem_zalloc(sizeof (*gn), KM_SLEEP); 1486 gn->gn_gbh = zio_buf_alloc(SPA_GANGBLOCKSIZE); 1487 *gnpp = gn; 1488 1489 return (gn); 1490 } 1491 1492 static void 1493 zio_gang_node_free(zio_gang_node_t **gnpp) 1494 { 1495 zio_gang_node_t *gn = *gnpp; 1496 1497 for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) 1498 ASSERT(gn->gn_child[g] == NULL); 1499 1500 zio_buf_free(gn->gn_gbh, SPA_GANGBLOCKSIZE); 1501 kmem_free(gn, sizeof (*gn)); 1502 *gnpp = NULL; 1503 } 1504 1505 static void 1506 zio_gang_tree_free(zio_gang_node_t **gnpp) 1507 { 1508 zio_gang_node_t *gn = *gnpp; 1509 1510 if (gn == NULL) 1511 return; 1512 1513 for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) 1514 zio_gang_tree_free(&gn->gn_child[g]); 1515 1516 zio_gang_node_free(gnpp); 1517 } 1518 1519 static void 1520 zio_gang_tree_assemble(zio_t *gio, blkptr_t *bp, zio_gang_node_t **gnpp) 1521 { 1522 zio_gang_node_t *gn = zio_gang_node_alloc(gnpp); 1523 1524 ASSERT(gio->io_gang_leader == gio); 1525 ASSERT(BP_IS_GANG(bp)); 1526 1527 zio_nowait(zio_read(gio, gio->io_spa, bp, gn->gn_gbh, 1528 SPA_GANGBLOCKSIZE, zio_gang_tree_assemble_done, gn, 1529 gio->io_priority, ZIO_GANG_CHILD_FLAGS(gio), &gio->io_bookmark)); 1530 } 1531 1532 static void 1533 zio_gang_tree_assemble_done(zio_t *zio) 1534 { 1535 zio_t *gio = zio->io_gang_leader; 1536 zio_gang_node_t *gn = zio->io_private; 1537 blkptr_t *bp = zio->io_bp; 1538 1539 ASSERT(gio == zio_unique_parent(zio)); 1540 ASSERT(zio->io_child_count == 0); 1541 1542 if (zio->io_error) 1543 return; 1544 1545 if (BP_SHOULD_BYTESWAP(bp)) 1546 byteswap_uint64_array(zio->io_data, zio->io_size); 1547 1548 ASSERT(zio->io_data == gn->gn_gbh); 1549 ASSERT(zio->io_size == SPA_GANGBLOCKSIZE); 1550 ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC); 1551 1552 for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) { 1553 blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g]; 1554 if (!BP_IS_GANG(gbp)) 1555 continue; 1556 zio_gang_tree_assemble(gio, gbp, &gn->gn_child[g]); 1557 } 1558 } 1559 1560 static void 1561 zio_gang_tree_issue(zio_t *pio, zio_gang_node_t *gn, blkptr_t *bp, void *data) 1562 { 1563 zio_t *gio = pio->io_gang_leader; 1564 zio_t *zio; 1565 1566 ASSERT(BP_IS_GANG(bp) == !!gn); 1567 ASSERT(BP_GET_CHECKSUM(bp) == BP_GET_CHECKSUM(gio->io_bp)); 1568 ASSERT(BP_GET_LSIZE(bp) == BP_GET_PSIZE(bp) || gn == gio->io_gang_tree); 1569 1570 /* 1571 * If you're a gang header, your data is in gn->gn_gbh. 1572 * If you're a gang member, your data is in 'data' and gn == NULL. 1573 */ 1574 zio = zio_gang_issue_func[gio->io_type](pio, bp, gn, data); 1575 1576 if (gn != NULL) { 1577 ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC); 1578 1579 for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) { 1580 blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g]; 1581 if (BP_IS_HOLE(gbp)) 1582 continue; 1583 zio_gang_tree_issue(zio, gn->gn_child[g], gbp, data); 1584 data = (char *)data + BP_GET_PSIZE(gbp); 1585 } 1586 } 1587 1588 if (gn == gio->io_gang_tree) 1589 ASSERT3P((char *)gio->io_data + gio->io_size, ==, data); 1590 1591 if (zio != pio) 1592 zio_nowait(zio); 1593 } 1594 1595 static int 1596 zio_gang_assemble(zio_t *zio) 1597 { 1598 blkptr_t *bp = zio->io_bp; 1599 1600 ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == NULL); 1601 ASSERT(zio->io_child_type > ZIO_CHILD_GANG); 1602 1603 zio->io_gang_leader = zio; 1604 1605 zio_gang_tree_assemble(zio, bp, &zio->io_gang_tree); 1606 1607 return (ZIO_PIPELINE_CONTINUE); 1608 } 1609 1610 static int 1611 zio_gang_issue(zio_t *zio) 1612 { 1613 blkptr_t *bp = zio->io_bp; 1614 1615 if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE)) 1616 return (ZIO_PIPELINE_STOP); 1617 1618 ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == zio); 1619 ASSERT(zio->io_child_type > ZIO_CHILD_GANG); 1620 1621 if (zio->io_child_error[ZIO_CHILD_GANG] == 0) 1622 zio_gang_tree_issue(zio, zio->io_gang_tree, bp, zio->io_data); 1623 else 1624 zio_gang_tree_free(&zio->io_gang_tree); 1625 1626 zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; 1627 1628 return (ZIO_PIPELINE_CONTINUE); 1629 } 1630 1631 static void 1632 zio_write_gang_member_ready(zio_t *zio) 1633 { 1634 zio_t *pio = zio_unique_parent(zio); 1635 zio_t *gio = zio->io_gang_leader; 1636 dva_t *cdva = zio->io_bp->blk_dva; 1637 dva_t *pdva = pio->io_bp->blk_dva; 1638 uint64_t asize; 1639 1640 if (BP_IS_HOLE(zio->io_bp)) 1641 return; 1642 1643 ASSERT(BP_IS_HOLE(&zio->io_bp_orig)); 1644 1645 ASSERT(zio->io_child_type == ZIO_CHILD_GANG); 1646 ASSERT3U(zio->io_prop.zp_copies, ==, gio->io_prop.zp_copies); 1647 ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(zio->io_bp)); 1648 ASSERT3U(pio->io_prop.zp_copies, <=, BP_GET_NDVAS(pio->io_bp)); 1649 ASSERT3U(BP_GET_NDVAS(zio->io_bp), <=, BP_GET_NDVAS(pio->io_bp)); 1650 1651 mutex_enter(&pio->io_lock); 1652 for (int d = 0; d < BP_GET_NDVAS(zio->io_bp); d++) { 1653 ASSERT(DVA_GET_GANG(&pdva[d])); 1654 asize = DVA_GET_ASIZE(&pdva[d]); 1655 asize += DVA_GET_ASIZE(&cdva[d]); 1656 DVA_SET_ASIZE(&pdva[d], asize); 1657 } 1658 mutex_exit(&pio->io_lock); 1659 } 1660 1661 static int 1662 zio_write_gang_block(zio_t *pio) 1663 { 1664 spa_t *spa = pio->io_spa; 1665 blkptr_t *bp = pio->io_bp; 1666 zio_t *gio = pio->io_gang_leader; 1667 zio_t *zio; 1668 zio_gang_node_t *gn, **gnpp; 1669 zio_gbh_phys_t *gbh; 1670 uint64_t txg = pio->io_txg; 1671 uint64_t resid = pio->io_size; 1672 uint64_t lsize; 1673 int copies = gio->io_prop.zp_copies; 1674 int gbh_copies = MIN(copies + 1, spa_max_replication(spa)); 1675 zio_prop_t zp; 1676 int error; 1677 1678 error = metaslab_alloc(spa, spa_normal_class(spa), SPA_GANGBLOCKSIZE, 1679 bp, gbh_copies, txg, pio == gio ? NULL : gio->io_bp, 1680 METASLAB_HINTBP_FAVOR | METASLAB_GANG_HEADER); 1681 if (error) { 1682 pio->io_error = error; 1683 return (ZIO_PIPELINE_CONTINUE); 1684 } 1685 1686 if (pio == gio) { 1687 gnpp = &gio->io_gang_tree; 1688 } else { 1689 gnpp = pio->io_private; 1690 ASSERT(pio->io_ready == zio_write_gang_member_ready); 1691 } 1692 1693 gn = zio_gang_node_alloc(gnpp); 1694 gbh = gn->gn_gbh; 1695 bzero(gbh, SPA_GANGBLOCKSIZE); 1696 1697 /* 1698 * Create the gang header. 1699 */ 1700 zio = zio_rewrite(pio, spa, txg, bp, gbh, SPA_GANGBLOCKSIZE, NULL, NULL, 1701 pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); 1702 1703 /* 1704 * Create and nowait the gang children. 1705 */ 1706 for (int g = 0; resid != 0; resid -= lsize, g++) { 1707 lsize = P2ROUNDUP(resid / (SPA_GBH_NBLKPTRS - g), 1708 SPA_MINBLOCKSIZE); 1709 ASSERT(lsize >= SPA_MINBLOCKSIZE && lsize <= resid); 1710 1711 zp.zp_checksum = gio->io_prop.zp_checksum; 1712 zp.zp_compress = ZIO_COMPRESS_OFF; 1713 zp.zp_type = DMU_OT_NONE; 1714 zp.zp_level = 0; 1715 zp.zp_copies = gio->io_prop.zp_copies; 1716 zp.zp_dedup = 0; 1717 zp.zp_dedup_verify = 0; 1718 1719 zio_nowait(zio_write(zio, spa, txg, &gbh->zg_blkptr[g], 1720 (char *)pio->io_data + (pio->io_size - resid), lsize, &zp, 1721 zio_write_gang_member_ready, NULL, &gn->gn_child[g], 1722 pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), 1723 &pio->io_bookmark)); 1724 } 1725 1726 /* 1727 * Set pio's pipeline to just wait for zio to finish. 1728 */ 1729 pio->io_pipeline = ZIO_INTERLOCK_PIPELINE; 1730 1731 zio_nowait(zio); 1732 1733 return (ZIO_PIPELINE_CONTINUE); 1734 } 1735 1736 /* 1737 * ========================================================================== 1738 * Dedup 1739 * ========================================================================== 1740 */ 1741 static void 1742 zio_ddt_child_read_done(zio_t *zio) 1743 { 1744 blkptr_t *bp = zio->io_bp; 1745 ddt_entry_t *dde = zio->io_private; 1746 ddt_phys_t *ddp; 1747 zio_t *pio = zio_unique_parent(zio); 1748 1749 mutex_enter(&pio->io_lock); 1750 ddp = ddt_phys_select(dde, bp); 1751 if (zio->io_error == 0) 1752 ddt_phys_clear(ddp); /* this ddp doesn't need repair */ 1753 if (zio->io_error == 0 && dde->dde_repair_data == NULL) 1754 dde->dde_repair_data = zio->io_data; 1755 else 1756 zio_buf_free(zio->io_data, zio->io_size); 1757 mutex_exit(&pio->io_lock); 1758 } 1759 1760 static int 1761 zio_ddt_read_start(zio_t *zio) 1762 { 1763 blkptr_t *bp = zio->io_bp; 1764 1765 ASSERT(BP_GET_DEDUP(bp)); 1766 ASSERT(BP_GET_PSIZE(bp) == zio->io_size); 1767 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 1768 1769 if (zio->io_child_error[ZIO_CHILD_DDT]) { 1770 ddt_t *ddt = ddt_select(zio->io_spa, bp); 1771 ddt_entry_t *dde = ddt_repair_start(ddt, bp); 1772 ddt_phys_t *ddp = dde->dde_phys; 1773 ddt_phys_t *ddp_self = ddt_phys_select(dde, bp); 1774 blkptr_t blk; 1775 1776 ASSERT(zio->io_vsd == NULL); 1777 zio->io_vsd = dde; 1778 1779 if (ddp_self == NULL) 1780 return (ZIO_PIPELINE_CONTINUE); 1781 1782 for (int p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { 1783 if (ddp->ddp_phys_birth == 0 || ddp == ddp_self) 1784 continue; 1785 ddt_bp_create(ddt->ddt_checksum, &dde->dde_key, ddp, 1786 &blk); 1787 zio_nowait(zio_read(zio, zio->io_spa, &blk, 1788 zio_buf_alloc(zio->io_size), zio->io_size, 1789 zio_ddt_child_read_done, dde, zio->io_priority, 1790 ZIO_DDT_CHILD_FLAGS(zio) | ZIO_FLAG_DONT_PROPAGATE, 1791 &zio->io_bookmark)); 1792 } 1793 return (ZIO_PIPELINE_CONTINUE); 1794 } 1795 1796 zio_nowait(zio_read(zio, zio->io_spa, bp, 1797 zio->io_data, zio->io_size, NULL, NULL, zio->io_priority, 1798 ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark)); 1799 1800 return (ZIO_PIPELINE_CONTINUE); 1801 } 1802 1803 static int 1804 zio_ddt_read_done(zio_t *zio) 1805 { 1806 blkptr_t *bp = zio->io_bp; 1807 1808 if (zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_DONE)) 1809 return (ZIO_PIPELINE_STOP); 1810 1811 ASSERT(BP_GET_DEDUP(bp)); 1812 ASSERT(BP_GET_PSIZE(bp) == zio->io_size); 1813 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 1814 1815 if (zio->io_child_error[ZIO_CHILD_DDT]) { 1816 ddt_t *ddt = ddt_select(zio->io_spa, bp); 1817 ddt_entry_t *dde = zio->io_vsd; 1818 if (ddt == NULL) { 1819 ASSERT(spa_load_state(zio->io_spa) != SPA_LOAD_NONE); 1820 return (ZIO_PIPELINE_CONTINUE); 1821 } 1822 if (dde == NULL) { 1823 zio->io_stage = ZIO_STAGE_DDT_READ_START >> 1; 1824 zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE); 1825 return (ZIO_PIPELINE_STOP); 1826 } 1827 if (dde->dde_repair_data != NULL) { 1828 bcopy(dde->dde_repair_data, zio->io_data, zio->io_size); 1829 zio->io_child_error[ZIO_CHILD_DDT] = 0; 1830 } 1831 ddt_repair_done(ddt, dde); 1832 zio->io_vsd = NULL; 1833 } 1834 1835 ASSERT(zio->io_vsd == NULL); 1836 1837 return (ZIO_PIPELINE_CONTINUE); 1838 } 1839 1840 static boolean_t 1841 zio_ddt_collision(zio_t *zio, ddt_t *ddt, ddt_entry_t *dde) 1842 { 1843 spa_t *spa = zio->io_spa; 1844 1845 /* 1846 * Note: we compare the original data, not the transformed data, 1847 * because when zio->io_bp is an override bp, we will not have 1848 * pushed the I/O transforms. That's an important optimization 1849 * because otherwise we'd compress/encrypt all dmu_sync() data twice. 1850 */ 1851 for (int p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { 1852 zio_t *lio = dde->dde_lead_zio[p]; 1853 1854 if (lio != NULL) { 1855 return (lio->io_orig_size != zio->io_orig_size || 1856 bcmp(zio->io_orig_data, lio->io_orig_data, 1857 zio->io_orig_size) != 0); 1858 } 1859 } 1860 1861 for (int p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { 1862 ddt_phys_t *ddp = &dde->dde_phys[p]; 1863 1864 if (ddp->ddp_phys_birth != 0) { 1865 arc_buf_t *abuf = NULL; 1866 uint32_t aflags = ARC_WAIT; 1867 blkptr_t blk = *zio->io_bp; 1868 int error; 1869 1870 ddt_bp_fill(ddp, &blk, ddp->ddp_phys_birth); 1871 1872 ddt_exit(ddt); 1873 1874 error = arc_read_nolock(NULL, spa, &blk, 1875 arc_getbuf_func, &abuf, ZIO_PRIORITY_SYNC_READ, 1876 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, 1877 &aflags, &zio->io_bookmark); 1878 1879 if (error == 0) { 1880 if (arc_buf_size(abuf) != zio->io_orig_size || 1881 bcmp(abuf->b_data, zio->io_orig_data, 1882 zio->io_orig_size) != 0) 1883 error = EEXIST; 1884 VERIFY(arc_buf_remove_ref(abuf, &abuf) == 1); 1885 } 1886 1887 ddt_enter(ddt); 1888 return (error != 0); 1889 } 1890 } 1891 1892 return (B_FALSE); 1893 } 1894 1895 static void 1896 zio_ddt_child_write_ready(zio_t *zio) 1897 { 1898 int p = zio->io_prop.zp_copies; 1899 ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp); 1900 ddt_entry_t *dde = zio->io_private; 1901 ddt_phys_t *ddp = &dde->dde_phys[p]; 1902 zio_t *pio; 1903 1904 if (zio->io_error) 1905 return; 1906 1907 ddt_enter(ddt); 1908 1909 ASSERT(dde->dde_lead_zio[p] == zio); 1910 1911 ddt_phys_fill(ddp, zio->io_bp); 1912 1913 while ((pio = zio_walk_parents(zio)) != NULL) 1914 ddt_bp_fill(ddp, pio->io_bp, zio->io_txg); 1915 1916 ddt_exit(ddt); 1917 } 1918 1919 static void 1920 zio_ddt_child_write_done(zio_t *zio) 1921 { 1922 int p = zio->io_prop.zp_copies; 1923 ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp); 1924 ddt_entry_t *dde = zio->io_private; 1925 ddt_phys_t *ddp = &dde->dde_phys[p]; 1926 1927 ddt_enter(ddt); 1928 1929 ASSERT(ddp->ddp_refcnt == 0); 1930 ASSERT(dde->dde_lead_zio[p] == zio); 1931 dde->dde_lead_zio[p] = NULL; 1932 1933 if (zio->io_error == 0) { 1934 while (zio_walk_parents(zio) != NULL) 1935 ddt_phys_addref(ddp); 1936 } else { 1937 ddt_phys_clear(ddp); 1938 } 1939 1940 ddt_exit(ddt); 1941 } 1942 1943 static void 1944 zio_ddt_ditto_write_done(zio_t *zio) 1945 { 1946 int p = DDT_PHYS_DITTO; 1947 zio_prop_t *zp = &zio->io_prop; 1948 blkptr_t *bp = zio->io_bp; 1949 ddt_t *ddt = ddt_select(zio->io_spa, bp); 1950 ddt_entry_t *dde = zio->io_private; 1951 ddt_phys_t *ddp = &dde->dde_phys[p]; 1952 ddt_key_t *ddk = &dde->dde_key; 1953 1954 ddt_enter(ddt); 1955 1956 ASSERT(ddp->ddp_refcnt == 0); 1957 ASSERT(dde->dde_lead_zio[p] == zio); 1958 dde->dde_lead_zio[p] = NULL; 1959 1960 if (zio->io_error == 0) { 1961 ASSERT(ZIO_CHECKSUM_EQUAL(bp->blk_cksum, ddk->ddk_cksum)); 1962 ASSERT(zp->zp_copies < SPA_DVAS_PER_BP); 1963 ASSERT(zp->zp_copies == BP_GET_NDVAS(bp) - BP_IS_GANG(bp)); 1964 if (ddp->ddp_phys_birth != 0) 1965 ddt_phys_free(ddt, ddk, ddp, zio->io_txg); 1966 ddt_phys_fill(ddp, bp); 1967 } 1968 1969 ddt_exit(ddt); 1970 } 1971 1972 static int 1973 zio_ddt_write(zio_t *zio) 1974 { 1975 spa_t *spa = zio->io_spa; 1976 blkptr_t *bp = zio->io_bp; 1977 uint64_t txg = zio->io_txg; 1978 zio_prop_t *zp = &zio->io_prop; 1979 int p = zp->zp_copies; 1980 int ditto_copies; 1981 zio_t *cio = NULL; 1982 zio_t *dio = NULL; 1983 ddt_t *ddt = ddt_select(spa, bp); 1984 ddt_entry_t *dde; 1985 ddt_phys_t *ddp; 1986 1987 ASSERT(BP_GET_DEDUP(bp)); 1988 ASSERT(BP_GET_CHECKSUM(bp) == zp->zp_checksum); 1989 ASSERT(BP_IS_HOLE(bp) || zio->io_bp_override); 1990 1991 ddt_enter(ddt); 1992 dde = ddt_lookup(ddt, bp, B_TRUE); 1993 ddp = &dde->dde_phys[p]; 1994 1995 if (zp->zp_dedup_verify && zio_ddt_collision(zio, ddt, dde)) { 1996 /* 1997 * If we're using a weak checksum, upgrade to a strong checksum 1998 * and try again. If we're already using a strong checksum, 1999 * we can't resolve it, so just convert to an ordinary write. 2000 * (And automatically e-mail a paper to Nature?) 2001 */ 2002 if (!zio_checksum_table[zp->zp_checksum].ci_dedup) { 2003 zp->zp_checksum = spa_dedup_checksum(spa); 2004 zio_pop_transforms(zio); 2005 zio->io_stage = ZIO_STAGE_OPEN; 2006 BP_ZERO(bp); 2007 } else { 2008 zp->zp_dedup = 0; 2009 } 2010 zio->io_pipeline = ZIO_WRITE_PIPELINE; 2011 ddt_exit(ddt); 2012 return (ZIO_PIPELINE_CONTINUE); 2013 } 2014 2015 ditto_copies = ddt_ditto_copies_needed(ddt, dde, ddp); 2016 ASSERT(ditto_copies < SPA_DVAS_PER_BP); 2017 2018 if (ditto_copies > ddt_ditto_copies_present(dde) && 2019 dde->dde_lead_zio[DDT_PHYS_DITTO] == NULL) { 2020 zio_prop_t czp = *zp; 2021 2022 czp.zp_copies = ditto_copies; 2023 2024 /* 2025 * If we arrived here with an override bp, we won't have run 2026 * the transform stack, so we won't have the data we need to 2027 * generate a child i/o. So, toss the override bp and restart. 2028 * This is safe, because using the override bp is just an 2029 * optimization; and it's rare, so the cost doesn't matter. 2030 */ 2031 if (zio->io_bp_override) { 2032 zio_pop_transforms(zio); 2033 zio->io_stage = ZIO_STAGE_OPEN; 2034 zio->io_pipeline = ZIO_WRITE_PIPELINE; 2035 zio->io_bp_override = NULL; 2036 BP_ZERO(bp); 2037 ddt_exit(ddt); 2038 return (ZIO_PIPELINE_CONTINUE); 2039 } 2040 2041 dio = zio_write(zio, spa, txg, bp, zio->io_orig_data, 2042 zio->io_orig_size, &czp, NULL, 2043 zio_ddt_ditto_write_done, dde, zio->io_priority, 2044 ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark); 2045 2046 zio_push_transform(dio, zio->io_data, zio->io_size, 0, NULL); 2047 dde->dde_lead_zio[DDT_PHYS_DITTO] = dio; 2048 } 2049 2050 if (ddp->ddp_phys_birth != 0 || dde->dde_lead_zio[p] != NULL) { 2051 if (ddp->ddp_phys_birth != 0) 2052 ddt_bp_fill(ddp, bp, txg); 2053 if (dde->dde_lead_zio[p] != NULL) 2054 zio_add_child(zio, dde->dde_lead_zio[p]); 2055 else 2056 ddt_phys_addref(ddp); 2057 } else if (zio->io_bp_override) { 2058 ASSERT(bp->blk_birth == txg); 2059 ASSERT(BP_EQUAL(bp, zio->io_bp_override)); 2060 ddt_phys_fill(ddp, bp); 2061 ddt_phys_addref(ddp); 2062 } else { 2063 cio = zio_write(zio, spa, txg, bp, zio->io_orig_data, 2064 zio->io_orig_size, zp, zio_ddt_child_write_ready, 2065 zio_ddt_child_write_done, dde, zio->io_priority, 2066 ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark); 2067 2068 zio_push_transform(cio, zio->io_data, zio->io_size, 0, NULL); 2069 dde->dde_lead_zio[p] = cio; 2070 } 2071 2072 ddt_exit(ddt); 2073 2074 if (cio) 2075 zio_nowait(cio); 2076 if (dio) 2077 zio_nowait(dio); 2078 2079 return (ZIO_PIPELINE_CONTINUE); 2080 } 2081 2082 ddt_entry_t *freedde; /* for debugging */ 2083 2084 static int 2085 zio_ddt_free(zio_t *zio) 2086 { 2087 spa_t *spa = zio->io_spa; 2088 blkptr_t *bp = zio->io_bp; 2089 ddt_t *ddt = ddt_select(spa, bp); 2090 ddt_entry_t *dde; 2091 ddt_phys_t *ddp; 2092 2093 ASSERT(BP_GET_DEDUP(bp)); 2094 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 2095 2096 ddt_enter(ddt); 2097 freedde = dde = ddt_lookup(ddt, bp, B_TRUE); 2098 ddp = ddt_phys_select(dde, bp); 2099 ddt_phys_decref(ddp); 2100 ddt_exit(ddt); 2101 2102 return (ZIO_PIPELINE_CONTINUE); 2103 } 2104 2105 /* 2106 * ========================================================================== 2107 * Allocate and free blocks 2108 * ========================================================================== 2109 */ 2110 static int 2111 zio_dva_allocate(zio_t *zio) 2112 { 2113 spa_t *spa = zio->io_spa; 2114 metaslab_class_t *mc = spa_normal_class(spa); 2115 blkptr_t *bp = zio->io_bp; 2116 int error; 2117 2118 if (zio->io_gang_leader == NULL) { 2119 ASSERT(zio->io_child_type > ZIO_CHILD_GANG); 2120 zio->io_gang_leader = zio; 2121 } 2122 2123 ASSERT(BP_IS_HOLE(bp)); 2124 ASSERT3U(BP_GET_NDVAS(bp), ==, 0); 2125 ASSERT3U(zio->io_prop.zp_copies, >, 0); 2126 ASSERT3U(zio->io_prop.zp_copies, <=, spa_max_replication(spa)); 2127 ASSERT3U(zio->io_size, ==, BP_GET_PSIZE(bp)); 2128 2129 error = metaslab_alloc(spa, mc, zio->io_size, bp, 2130 zio->io_prop.zp_copies, zio->io_txg, NULL, 0); 2131 2132 if (error) { 2133 if (error == ENOSPC && zio->io_size > SPA_MINBLOCKSIZE) 2134 return (zio_write_gang_block(zio)); 2135 zio->io_error = error; 2136 } 2137 2138 return (ZIO_PIPELINE_CONTINUE); 2139 } 2140 2141 static int 2142 zio_dva_free(zio_t *zio) 2143 { 2144 metaslab_free(zio->io_spa, zio->io_bp, zio->io_txg, B_FALSE); 2145 2146 return (ZIO_PIPELINE_CONTINUE); 2147 } 2148 2149 static int 2150 zio_dva_claim(zio_t *zio) 2151 { 2152 int error; 2153 2154 error = metaslab_claim(zio->io_spa, zio->io_bp, zio->io_txg); 2155 if (error) 2156 zio->io_error = error; 2157 2158 return (ZIO_PIPELINE_CONTINUE); 2159 } 2160 2161 /* 2162 * Undo an allocation. This is used by zio_done() when an I/O fails 2163 * and we want to give back the block we just allocated. 2164 * This handles both normal blocks and gang blocks. 2165 */ 2166 static void 2167 zio_dva_unallocate(zio_t *zio, zio_gang_node_t *gn, blkptr_t *bp) 2168 { 2169 ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp)); 2170 ASSERT(zio->io_bp_override == NULL); 2171 2172 if (!BP_IS_HOLE(bp)) 2173 metaslab_free(zio->io_spa, bp, bp->blk_birth, B_TRUE); 2174 2175 if (gn != NULL) { 2176 for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) { 2177 zio_dva_unallocate(zio, gn->gn_child[g], 2178 &gn->gn_gbh->zg_blkptr[g]); 2179 } 2180 } 2181 } 2182 2183 /* 2184 * Try to allocate an intent log block. Return 0 on success, errno on failure. 2185 */ 2186 int 2187 zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp, blkptr_t *old_bp, 2188 uint64_t size, boolean_t use_slog) 2189 { 2190 int error = 1; 2191 2192 ASSERT(txg > spa_syncing_txg(spa)); 2193 2194 if (use_slog) 2195 error = metaslab_alloc(spa, spa_log_class(spa), size, 2196 new_bp, 1, txg, old_bp, METASLAB_HINTBP_AVOID); 2197 2198 if (error) 2199 error = metaslab_alloc(spa, spa_normal_class(spa), size, 2200 new_bp, 1, txg, old_bp, METASLAB_HINTBP_AVOID); 2201 2202 if (error == 0) { 2203 BP_SET_LSIZE(new_bp, size); 2204 BP_SET_PSIZE(new_bp, size); 2205 BP_SET_COMPRESS(new_bp, ZIO_COMPRESS_OFF); 2206 BP_SET_CHECKSUM(new_bp, 2207 spa_version(spa) >= SPA_VERSION_SLIM_ZIL 2208 ? ZIO_CHECKSUM_ZILOG2 : ZIO_CHECKSUM_ZILOG); 2209 BP_SET_TYPE(new_bp, DMU_OT_INTENT_LOG); 2210 BP_SET_LEVEL(new_bp, 0); 2211 BP_SET_DEDUP(new_bp, 0); 2212 BP_SET_BYTEORDER(new_bp, ZFS_HOST_BYTEORDER); 2213 } 2214 2215 return (error); 2216 } 2217 2218 /* 2219 * Free an intent log block. 2220 */ 2221 void 2222 zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp) 2223 { 2224 ASSERT(BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG); 2225 ASSERT(!BP_IS_GANG(bp)); 2226 2227 zio_free(spa, txg, bp); 2228 } 2229 2230 /* 2231 * ========================================================================== 2232 * Read and write to physical devices 2233 * ========================================================================== 2234 */ 2235 static int 2236 zio_vdev_io_start(zio_t *zio) 2237 { 2238 vdev_t *vd = zio->io_vd; 2239 uint64_t align; 2240 spa_t *spa = zio->io_spa; 2241 2242 ASSERT(zio->io_error == 0); 2243 ASSERT(zio->io_child_error[ZIO_CHILD_VDEV] == 0); 2244 2245 if (vd == NULL) { 2246 if (!(zio->io_flags & ZIO_FLAG_CONFIG_WRITER)) 2247 spa_config_enter(spa, SCL_ZIO, zio, RW_READER); 2248 2249 /* 2250 * The mirror_ops handle multiple DVAs in a single BP. 2251 */ 2252 return (vdev_mirror_ops.vdev_op_io_start(zio)); 2253 } 2254 2255 /* 2256 * We keep track of time-sensitive I/Os so that the scan thread 2257 * can quickly react to certain workloads. In particular, we care 2258 * about non-scrubbing, top-level reads and writes with the following 2259 * characteristics: 2260 * - synchronous writes of user data to non-slog devices 2261 * - any reads of user data 2262 * When these conditions are met, adjust the timestamp of spa_last_io 2263 * which allows the scan thread to adjust its workload accordingly. 2264 */ 2265 if (!(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && zio->io_bp != NULL && 2266 vd == vd->vdev_top && !vd->vdev_islog && 2267 zio->io_bookmark.zb_objset != DMU_META_OBJSET && 2268 zio->io_txg != spa_syncing_txg(spa)) { 2269 uint64_t old = spa->spa_last_io; 2270 uint64_t new = ddi_get_lbolt64(); 2271 if (old != new) 2272 (void) atomic_cas_64(&spa->spa_last_io, old, new); 2273 } 2274 2275 align = 1ULL << vd->vdev_top->vdev_ashift; 2276 2277 if (P2PHASE(zio->io_size, align) != 0) { 2278 uint64_t asize = P2ROUNDUP(zio->io_size, align); 2279 char *abuf = zio_buf_alloc(asize); 2280 ASSERT(vd == vd->vdev_top); 2281 if (zio->io_type == ZIO_TYPE_WRITE) { 2282 bcopy(zio->io_data, abuf, zio->io_size); 2283 bzero(abuf + zio->io_size, asize - zio->io_size); 2284 } 2285 zio_push_transform(zio, abuf, asize, asize, zio_subblock); 2286 } 2287 2288 ASSERT(P2PHASE(zio->io_offset, align) == 0); 2289 ASSERT(P2PHASE(zio->io_size, align) == 0); 2290 VERIFY(zio->io_type != ZIO_TYPE_WRITE || spa_writeable(spa)); 2291 2292 /* 2293 * If this is a repair I/O, and there's no self-healing involved -- 2294 * that is, we're just resilvering what we expect to resilver -- 2295 * then don't do the I/O unless zio's txg is actually in vd's DTL. 2296 * This prevents spurious resilvering with nested replication. 2297 * For example, given a mirror of mirrors, (A+B)+(C+D), if only 2298 * A is out of date, we'll read from C+D, then use the data to 2299 * resilver A+B -- but we don't actually want to resilver B, just A. 2300 * The top-level mirror has no way to know this, so instead we just 2301 * discard unnecessary repairs as we work our way down the vdev tree. 2302 * The same logic applies to any form of nested replication: 2303 * ditto + mirror, RAID-Z + replacing, etc. This covers them all. 2304 */ 2305 if ((zio->io_flags & ZIO_FLAG_IO_REPAIR) && 2306 !(zio->io_flags & ZIO_FLAG_SELF_HEAL) && 2307 zio->io_txg != 0 && /* not a delegated i/o */ 2308 !vdev_dtl_contains(vd, DTL_PARTIAL, zio->io_txg, 1)) { 2309 ASSERT(zio->io_type == ZIO_TYPE_WRITE); 2310 zio_vdev_io_bypass(zio); 2311 return (ZIO_PIPELINE_CONTINUE); 2312 } 2313 2314 if (vd->vdev_ops->vdev_op_leaf && 2315 (zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE)) { 2316 2317 if (zio->io_type == ZIO_TYPE_READ && vdev_cache_read(zio) == 0) 2318 return (ZIO_PIPELINE_CONTINUE); 2319 2320 if ((zio = vdev_queue_io(zio)) == NULL) 2321 return (ZIO_PIPELINE_STOP); 2322 2323 if (!vdev_accessible(vd, zio)) { 2324 zio->io_error = ENXIO; 2325 zio_interrupt(zio); 2326 return (ZIO_PIPELINE_STOP); 2327 } 2328 } 2329 2330 return (vd->vdev_ops->vdev_op_io_start(zio)); 2331 } 2332 2333 static int 2334 zio_vdev_io_done(zio_t *zio) 2335 { 2336 vdev_t *vd = zio->io_vd; 2337 vdev_ops_t *ops = vd ? vd->vdev_ops : &vdev_mirror_ops; 2338 boolean_t unexpected_error = B_FALSE; 2339 2340 if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE)) 2341 return (ZIO_PIPELINE_STOP); 2342 2343 ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE); 2344 2345 if (vd != NULL && vd->vdev_ops->vdev_op_leaf) { 2346 2347 vdev_queue_io_done(zio); 2348 2349 if (zio->io_type == ZIO_TYPE_WRITE) 2350 vdev_cache_write(zio); 2351 2352 if (zio_injection_enabled && zio->io_error == 0) 2353 zio->io_error = zio_handle_device_injection(vd, 2354 zio, EIO); 2355 2356 if (zio_injection_enabled && zio->io_error == 0) 2357 zio->io_error = zio_handle_label_injection(zio, EIO); 2358 2359 if (zio->io_error) { 2360 if (!vdev_accessible(vd, zio)) { 2361 zio->io_error = ENXIO; 2362 } else { 2363 unexpected_error = B_TRUE; 2364 } 2365 } 2366 } 2367 2368 ops->vdev_op_io_done(zio); 2369 2370 if (unexpected_error) 2371 VERIFY(vdev_probe(vd, zio) == NULL); 2372 2373 return (ZIO_PIPELINE_CONTINUE); 2374 } 2375 2376 /* 2377 * For non-raidz ZIOs, we can just copy aside the bad data read from the 2378 * disk, and use that to finish the checksum ereport later. 2379 */ 2380 static void 2381 zio_vsd_default_cksum_finish(zio_cksum_report_t *zcr, 2382 const void *good_buf) 2383 { 2384 /* no processing needed */ 2385 zfs_ereport_finish_checksum(zcr, good_buf, zcr->zcr_cbdata, B_FALSE); 2386 } 2387 2388 /*ARGSUSED*/ 2389 void 2390 zio_vsd_default_cksum_report(zio_t *zio, zio_cksum_report_t *zcr, void *ignored) 2391 { 2392 void *buf = zio_buf_alloc(zio->io_size); 2393 2394 bcopy(zio->io_data, buf, zio->io_size); 2395 2396 zcr->zcr_cbinfo = zio->io_size; 2397 zcr->zcr_cbdata = buf; 2398 zcr->zcr_finish = zio_vsd_default_cksum_finish; 2399 zcr->zcr_free = zio_buf_free; 2400 } 2401 2402 static int 2403 zio_vdev_io_assess(zio_t *zio) 2404 { 2405 vdev_t *vd = zio->io_vd; 2406 2407 if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE)) 2408 return (ZIO_PIPELINE_STOP); 2409 2410 if (vd == NULL && !(zio->io_flags & ZIO_FLAG_CONFIG_WRITER)) 2411 spa_config_exit(zio->io_spa, SCL_ZIO, zio); 2412 2413 if (zio->io_vsd != NULL) { 2414 zio->io_vsd_ops->vsd_free(zio); 2415 zio->io_vsd = NULL; 2416 } 2417 2418 if (zio_injection_enabled && zio->io_error == 0) 2419 zio->io_error = zio_handle_fault_injection(zio, EIO); 2420 2421 /* 2422 * If the I/O failed, determine whether we should attempt to retry it. 2423 * 2424 * On retry, we cut in line in the issue queue, since we don't want 2425 * compression/checksumming/etc. work to prevent our (cheap) IO reissue. 2426 */ 2427 if (zio->io_error && vd == NULL && 2428 !(zio->io_flags & (ZIO_FLAG_DONT_RETRY | ZIO_FLAG_IO_RETRY))) { 2429 ASSERT(!(zio->io_flags & ZIO_FLAG_DONT_QUEUE)); /* not a leaf */ 2430 ASSERT(!(zio->io_flags & ZIO_FLAG_IO_BYPASS)); /* not a leaf */ 2431 zio->io_error = 0; 2432 zio->io_flags |= ZIO_FLAG_IO_RETRY | 2433 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE; 2434 zio->io_stage = ZIO_STAGE_VDEV_IO_START >> 1; 2435 zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, 2436 zio_requeue_io_start_cut_in_line); 2437 return (ZIO_PIPELINE_STOP); 2438 } 2439 2440 /* 2441 * If we got an error on a leaf device, convert it to ENXIO 2442 * if the device is not accessible at all. 2443 */ 2444 if (zio->io_error && vd != NULL && vd->vdev_ops->vdev_op_leaf && 2445 !vdev_accessible(vd, zio)) 2446 zio->io_error = ENXIO; 2447 2448 /* 2449 * If we can't write to an interior vdev (mirror or RAID-Z), 2450 * set vdev_cant_write so that we stop trying to allocate from it. 2451 */ 2452 if (zio->io_error == ENXIO && zio->io_type == ZIO_TYPE_WRITE && 2453 vd != NULL && !vd->vdev_ops->vdev_op_leaf) 2454 vd->vdev_cant_write = B_TRUE; 2455 2456 if (zio->io_error) 2457 zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; 2458 2459 return (ZIO_PIPELINE_CONTINUE); 2460 } 2461 2462 void 2463 zio_vdev_io_reissue(zio_t *zio) 2464 { 2465 ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START); 2466 ASSERT(zio->io_error == 0); 2467 2468 zio->io_stage >>= 1; 2469 } 2470 2471 void 2472 zio_vdev_io_redone(zio_t *zio) 2473 { 2474 ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_DONE); 2475 2476 zio->io_stage >>= 1; 2477 } 2478 2479 void 2480 zio_vdev_io_bypass(zio_t *zio) 2481 { 2482 ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START); 2483 ASSERT(zio->io_error == 0); 2484 2485 zio->io_flags |= ZIO_FLAG_IO_BYPASS; 2486 zio->io_stage = ZIO_STAGE_VDEV_IO_ASSESS >> 1; 2487 } 2488 2489 /* 2490 * ========================================================================== 2491 * Generate and verify checksums 2492 * ========================================================================== 2493 */ 2494 static int 2495 zio_checksum_generate(zio_t *zio) 2496 { 2497 blkptr_t *bp = zio->io_bp; 2498 enum zio_checksum checksum; 2499 2500 if (bp == NULL) { 2501 /* 2502 * This is zio_write_phys(). 2503 * We're either generating a label checksum, or none at all. 2504 */ 2505 checksum = zio->io_prop.zp_checksum; 2506 2507 if (checksum == ZIO_CHECKSUM_OFF) 2508 return (ZIO_PIPELINE_CONTINUE); 2509 2510 ASSERT(checksum == ZIO_CHECKSUM_LABEL); 2511 } else { 2512 if (BP_IS_GANG(bp) && zio->io_child_type == ZIO_CHILD_GANG) { 2513 ASSERT(!IO_IS_ALLOCATING(zio)); 2514 checksum = ZIO_CHECKSUM_GANG_HEADER; 2515 } else { 2516 checksum = BP_GET_CHECKSUM(bp); 2517 } 2518 } 2519 2520 zio_checksum_compute(zio, checksum, zio->io_data, zio->io_size); 2521 2522 return (ZIO_PIPELINE_CONTINUE); 2523 } 2524 2525 static int 2526 zio_checksum_verify(zio_t *zio) 2527 { 2528 zio_bad_cksum_t info; 2529 blkptr_t *bp = zio->io_bp; 2530 int error; 2531 2532 ASSERT(zio->io_vd != NULL); 2533 2534 if (bp == NULL) { 2535 /* 2536 * This is zio_read_phys(). 2537 * We're either verifying a label checksum, or nothing at all. 2538 */ 2539 if (zio->io_prop.zp_checksum == ZIO_CHECKSUM_OFF) 2540 return (ZIO_PIPELINE_CONTINUE); 2541 2542 ASSERT(zio->io_prop.zp_checksum == ZIO_CHECKSUM_LABEL); 2543 } 2544 2545 if ((error = zio_checksum_error(zio, &info)) != 0) { 2546 zio->io_error = error; 2547 if (!(zio->io_flags & ZIO_FLAG_SPECULATIVE)) { 2548 zfs_ereport_start_checksum(zio->io_spa, 2549 zio->io_vd, zio, zio->io_offset, 2550 zio->io_size, NULL, &info); 2551 } 2552 } 2553 2554 return (ZIO_PIPELINE_CONTINUE); 2555 } 2556 2557 /* 2558 * Called by RAID-Z to ensure we don't compute the checksum twice. 2559 */ 2560 void 2561 zio_checksum_verified(zio_t *zio) 2562 { 2563 zio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY; 2564 } 2565 2566 /* 2567 * ========================================================================== 2568 * Error rank. Error are ranked in the order 0, ENXIO, ECKSUM, EIO, other. 2569 * An error of 0 indictes success. ENXIO indicates whole-device failure, 2570 * which may be transient (e.g. unplugged) or permament. ECKSUM and EIO 2571 * indicate errors that are specific to one I/O, and most likely permanent. 2572 * Any other error is presumed to be worse because we weren't expecting it. 2573 * ========================================================================== 2574 */ 2575 int 2576 zio_worst_error(int e1, int e2) 2577 { 2578 static int zio_error_rank[] = { 0, ENXIO, ECKSUM, EIO }; 2579 int r1, r2; 2580 2581 for (r1 = 0; r1 < sizeof (zio_error_rank) / sizeof (int); r1++) 2582 if (e1 == zio_error_rank[r1]) 2583 break; 2584 2585 for (r2 = 0; r2 < sizeof (zio_error_rank) / sizeof (int); r2++) 2586 if (e2 == zio_error_rank[r2]) 2587 break; 2588 2589 return (r1 > r2 ? e1 : e2); 2590 } 2591 2592 /* 2593 * ========================================================================== 2594 * I/O completion 2595 * ========================================================================== 2596 */ 2597 static int 2598 zio_ready(zio_t *zio) 2599 { 2600 blkptr_t *bp = zio->io_bp; 2601 zio_t *pio, *pio_next; 2602 2603 if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY) || 2604 zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_READY)) 2605 return (ZIO_PIPELINE_STOP); 2606 2607 if (zio->io_ready) { 2608 ASSERT(IO_IS_ALLOCATING(zio)); 2609 ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp)); 2610 ASSERT(zio->io_children[ZIO_CHILD_GANG][ZIO_WAIT_READY] == 0); 2611 2612 zio->io_ready(zio); 2613 } 2614 2615 if (bp != NULL && bp != &zio->io_bp_copy) 2616 zio->io_bp_copy = *bp; 2617 2618 if (zio->io_error) 2619 zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; 2620 2621 mutex_enter(&zio->io_lock); 2622 zio->io_state[ZIO_WAIT_READY] = 1; 2623 pio = zio_walk_parents(zio); 2624 mutex_exit(&zio->io_lock); 2625 2626 /* 2627 * As we notify zio's parents, new parents could be added. 2628 * New parents go to the head of zio's io_parent_list, however, 2629 * so we will (correctly) not notify them. The remainder of zio's 2630 * io_parent_list, from 'pio_next' onward, cannot change because 2631 * all parents must wait for us to be done before they can be done. 2632 */ 2633 for (; pio != NULL; pio = pio_next) { 2634 pio_next = zio_walk_parents(zio); 2635 zio_notify_parent(pio, zio, ZIO_WAIT_READY); 2636 } 2637 2638 if (zio->io_flags & ZIO_FLAG_NODATA) { 2639 if (BP_IS_GANG(bp)) { 2640 zio->io_flags &= ~ZIO_FLAG_NODATA; 2641 } else { 2642 ASSERT((uintptr_t)zio->io_data < SPA_MAXBLOCKSIZE); 2643 zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES; 2644 } 2645 } 2646 2647 if (zio_injection_enabled && 2648 zio->io_spa->spa_syncing_txg == zio->io_txg) 2649 zio_handle_ignored_writes(zio); 2650 2651 return (ZIO_PIPELINE_CONTINUE); 2652 } 2653 2654 static int 2655 zio_done(zio_t *zio) 2656 { 2657 spa_t *spa = zio->io_spa; 2658 zio_t *lio = zio->io_logical; 2659 blkptr_t *bp = zio->io_bp; 2660 vdev_t *vd = zio->io_vd; 2661 uint64_t psize = zio->io_size; 2662 zio_t *pio, *pio_next; 2663 2664 /* 2665 * If our children haven't all completed, 2666 * wait for them and then repeat this pipeline stage. 2667 */ 2668 if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE) || 2669 zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE) || 2670 zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_DONE) || 2671 zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_DONE)) 2672 return (ZIO_PIPELINE_STOP); 2673 2674 for (int c = 0; c < ZIO_CHILD_TYPES; c++) 2675 for (int w = 0; w < ZIO_WAIT_TYPES; w++) 2676 ASSERT(zio->io_children[c][w] == 0); 2677 2678 if (bp != NULL) { 2679 ASSERT(bp->blk_pad[0] == 0); 2680 ASSERT(bp->blk_pad[1] == 0); 2681 ASSERT(bcmp(bp, &zio->io_bp_copy, sizeof (blkptr_t)) == 0 || 2682 (bp == zio_unique_parent(zio)->io_bp)); 2683 if (zio->io_type == ZIO_TYPE_WRITE && !BP_IS_HOLE(bp) && 2684 zio->io_bp_override == NULL && 2685 !(zio->io_flags & ZIO_FLAG_IO_REPAIR)) { 2686 ASSERT(!BP_SHOULD_BYTESWAP(bp)); 2687 ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(bp)); 2688 ASSERT(BP_COUNT_GANG(bp) == 0 || 2689 (BP_COUNT_GANG(bp) == BP_GET_NDVAS(bp))); 2690 } 2691 } 2692 2693 /* 2694 * If there were child vdev/gang/ddt errors, they apply to us now. 2695 */ 2696 zio_inherit_child_errors(zio, ZIO_CHILD_VDEV); 2697 zio_inherit_child_errors(zio, ZIO_CHILD_GANG); 2698 zio_inherit_child_errors(zio, ZIO_CHILD_DDT); 2699 2700 /* 2701 * If the I/O on the transformed data was successful, generate any 2702 * checksum reports now while we still have the transformed data. 2703 */ 2704 if (zio->io_error == 0) { 2705 while (zio->io_cksum_report != NULL) { 2706 zio_cksum_report_t *zcr = zio->io_cksum_report; 2707 uint64_t align = zcr->zcr_align; 2708 uint64_t asize = P2ROUNDUP(psize, align); 2709 char *abuf = zio->io_data; 2710 2711 if (asize != psize) { 2712 abuf = zio_buf_alloc(asize); 2713 bcopy(zio->io_data, abuf, psize); 2714 bzero(abuf + psize, asize - psize); 2715 } 2716 2717 zio->io_cksum_report = zcr->zcr_next; 2718 zcr->zcr_next = NULL; 2719 zcr->zcr_finish(zcr, abuf); 2720 zfs_ereport_free_checksum(zcr); 2721 2722 if (asize != psize) 2723 zio_buf_free(abuf, asize); 2724 } 2725 } 2726 2727 zio_pop_transforms(zio); /* note: may set zio->io_error */ 2728 2729 vdev_stat_update(zio, psize); 2730 2731 if (zio->io_error) { 2732 /* 2733 * If this I/O is attached to a particular vdev, 2734 * generate an error message describing the I/O failure 2735 * at the block level. We ignore these errors if the 2736 * device is currently unavailable. 2737 */ 2738 if (zio->io_error != ECKSUM && vd != NULL && !vdev_is_dead(vd)) 2739 zfs_ereport_post(FM_EREPORT_ZFS_IO, spa, vd, zio, 0, 0); 2740 2741 if ((zio->io_error == EIO || !(zio->io_flags & 2742 (ZIO_FLAG_SPECULATIVE | ZIO_FLAG_DONT_PROPAGATE))) && 2743 zio == lio) { 2744 /* 2745 * For logical I/O requests, tell the SPA to log the 2746 * error and generate a logical data ereport. 2747 */ 2748 spa_log_error(spa, zio); 2749 zfs_ereport_post(FM_EREPORT_ZFS_DATA, spa, NULL, zio, 2750 0, 0); 2751 } 2752 } 2753 2754 if (zio->io_error && zio == lio) { 2755 /* 2756 * Determine whether zio should be reexecuted. This will 2757 * propagate all the way to the root via zio_notify_parent(). 2758 */ 2759 ASSERT(vd == NULL && bp != NULL); 2760 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 2761 2762 if (IO_IS_ALLOCATING(zio) && 2763 !(zio->io_flags & ZIO_FLAG_CANFAIL)) { 2764 if (zio->io_error != ENOSPC) 2765 zio->io_reexecute |= ZIO_REEXECUTE_NOW; 2766 else 2767 zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; 2768 } 2769 2770 if ((zio->io_type == ZIO_TYPE_READ || 2771 zio->io_type == ZIO_TYPE_FREE) && 2772 !(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && 2773 zio->io_error == ENXIO && 2774 spa_load_state(spa) == SPA_LOAD_NONE && 2775 spa_get_failmode(spa) != ZIO_FAILURE_MODE_CONTINUE) 2776 zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; 2777 2778 if (!(zio->io_flags & ZIO_FLAG_CANFAIL) && !zio->io_reexecute) 2779 zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; 2780 2781 /* 2782 * Here is a possibly good place to attempt to do 2783 * either combinatorial reconstruction or error correction 2784 * based on checksums. It also might be a good place 2785 * to send out preliminary ereports before we suspend 2786 * processing. 2787 */ 2788 } 2789 2790 /* 2791 * If there were logical child errors, they apply to us now. 2792 * We defer this until now to avoid conflating logical child 2793 * errors with errors that happened to the zio itself when 2794 * updating vdev stats and reporting FMA events above. 2795 */ 2796 zio_inherit_child_errors(zio, ZIO_CHILD_LOGICAL); 2797 2798 if ((zio->io_error || zio->io_reexecute) && 2799 IO_IS_ALLOCATING(zio) && zio->io_gang_leader == zio && 2800 !(zio->io_flags & ZIO_FLAG_IO_REWRITE)) 2801 zio_dva_unallocate(zio, zio->io_gang_tree, bp); 2802 2803 zio_gang_tree_free(&zio->io_gang_tree); 2804 2805 /* 2806 * Godfather I/Os should never suspend. 2807 */ 2808 if ((zio->io_flags & ZIO_FLAG_GODFATHER) && 2809 (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) 2810 zio->io_reexecute = 0; 2811 2812 if (zio->io_reexecute) { 2813 /* 2814 * This is a logical I/O that wants to reexecute. 2815 * 2816 * Reexecute is top-down. When an i/o fails, if it's not 2817 * the root, it simply notifies its parent and sticks around. 2818 * The parent, seeing that it still has children in zio_done(), 2819 * does the same. This percolates all the way up to the root. 2820 * The root i/o will reexecute or suspend the entire tree. 2821 * 2822 * This approach ensures that zio_reexecute() honors 2823 * all the original i/o dependency relationships, e.g. 2824 * parents not executing until children are ready. 2825 */ 2826 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 2827 2828 zio->io_gang_leader = NULL; 2829 2830 mutex_enter(&zio->io_lock); 2831 zio->io_state[ZIO_WAIT_DONE] = 1; 2832 mutex_exit(&zio->io_lock); 2833 2834 /* 2835 * "The Godfather" I/O monitors its children but is 2836 * not a true parent to them. It will track them through 2837 * the pipeline but severs its ties whenever they get into 2838 * trouble (e.g. suspended). This allows "The Godfather" 2839 * I/O to return status without blocking. 2840 */ 2841 for (pio = zio_walk_parents(zio); pio != NULL; pio = pio_next) { 2842 zio_link_t *zl = zio->io_walk_link; 2843 pio_next = zio_walk_parents(zio); 2844 2845 if ((pio->io_flags & ZIO_FLAG_GODFATHER) && 2846 (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) { 2847 zio_remove_child(pio, zio, zl); 2848 zio_notify_parent(pio, zio, ZIO_WAIT_DONE); 2849 } 2850 } 2851 2852 if ((pio = zio_unique_parent(zio)) != NULL) { 2853 /* 2854 * We're not a root i/o, so there's nothing to do 2855 * but notify our parent. Don't propagate errors 2856 * upward since we haven't permanently failed yet. 2857 */ 2858 ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER)); 2859 zio->io_flags |= ZIO_FLAG_DONT_PROPAGATE; 2860 zio_notify_parent(pio, zio, ZIO_WAIT_DONE); 2861 } else if (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND) { 2862 /* 2863 * We'd fail again if we reexecuted now, so suspend 2864 * until conditions improve (e.g. device comes online). 2865 */ 2866 zio_suspend(spa, zio); 2867 } else { 2868 /* 2869 * Reexecution is potentially a huge amount of work. 2870 * Hand it off to the otherwise-unused claim taskq. 2871 */ 2872 (void) taskq_dispatch( 2873 spa->spa_zio_taskq[ZIO_TYPE_CLAIM][ZIO_TASKQ_ISSUE], 2874 (task_func_t *)zio_reexecute, zio, TQ_SLEEP); 2875 } 2876 return (ZIO_PIPELINE_STOP); 2877 } 2878 2879 ASSERT(zio->io_child_count == 0); 2880 ASSERT(zio->io_reexecute == 0); 2881 ASSERT(zio->io_error == 0 || (zio->io_flags & ZIO_FLAG_CANFAIL)); 2882 2883 /* 2884 * Report any checksum errors, since the I/O is complete. 2885 */ 2886 while (zio->io_cksum_report != NULL) { 2887 zio_cksum_report_t *zcr = zio->io_cksum_report; 2888 zio->io_cksum_report = zcr->zcr_next; 2889 zcr->zcr_next = NULL; 2890 zcr->zcr_finish(zcr, NULL); 2891 zfs_ereport_free_checksum(zcr); 2892 } 2893 2894 /* 2895 * It is the responsibility of the done callback to ensure that this 2896 * particular zio is no longer discoverable for adoption, and as 2897 * such, cannot acquire any new parents. 2898 */ 2899 if (zio->io_done) 2900 zio->io_done(zio); 2901 2902 mutex_enter(&zio->io_lock); 2903 zio->io_state[ZIO_WAIT_DONE] = 1; 2904 mutex_exit(&zio->io_lock); 2905 2906 for (pio = zio_walk_parents(zio); pio != NULL; pio = pio_next) { 2907 zio_link_t *zl = zio->io_walk_link; 2908 pio_next = zio_walk_parents(zio); 2909 zio_remove_child(pio, zio, zl); 2910 zio_notify_parent(pio, zio, ZIO_WAIT_DONE); 2911 } 2912 2913 if (zio->io_waiter != NULL) { 2914 mutex_enter(&zio->io_lock); 2915 zio->io_executor = NULL; 2916 cv_broadcast(&zio->io_cv); 2917 mutex_exit(&zio->io_lock); 2918 } else { 2919 zio_destroy(zio); 2920 } 2921 2922 return (ZIO_PIPELINE_STOP); 2923 } 2924 2925 /* 2926 * ========================================================================== 2927 * I/O pipeline definition 2928 * ========================================================================== 2929 */ 2930 static zio_pipe_stage_t *zio_pipeline[] = { 2931 NULL, 2932 zio_read_bp_init, 2933 zio_free_bp_init, 2934 zio_issue_async, 2935 zio_write_bp_init, 2936 zio_checksum_generate, 2937 zio_ddt_read_start, 2938 zio_ddt_read_done, 2939 zio_ddt_write, 2940 zio_ddt_free, 2941 zio_gang_assemble, 2942 zio_gang_issue, 2943 zio_dva_allocate, 2944 zio_dva_free, 2945 zio_dva_claim, 2946 zio_ready, 2947 zio_vdev_io_start, 2948 zio_vdev_io_done, 2949 zio_vdev_io_assess, 2950 zio_checksum_verify, 2951 zio_done 2952 }; 2953