1fa9e4066Sahrens /* 2fa9e4066Sahrens * CDDL HEADER START 3fa9e4066Sahrens * 4fa9e4066Sahrens * The contents of this file are subject to the terms of the 5ea8dc4b6Seschrock * Common Development and Distribution License (the "License"). 6ea8dc4b6Seschrock * You may not use this file except in compliance with the License. 7fa9e4066Sahrens * 8fa9e4066Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9fa9e4066Sahrens * or http://www.opensolaris.org/os/licensing. 10fa9e4066Sahrens * See the License for the specific language governing permissions 11fa9e4066Sahrens * and limitations under the License. 12fa9e4066Sahrens * 13fa9e4066Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14fa9e4066Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15fa9e4066Sahrens * If applicable, add the following below this CDDL HEADER, with the 16fa9e4066Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17fa9e4066Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18fa9e4066Sahrens * 19fa9e4066Sahrens * CDDL HEADER END 20fa9e4066Sahrens */ 21fa9e4066Sahrens /* 223f9d6ad7SLin Ling * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23840345f6SGeorge Wilson * Copyright (c) 2012 by Delphix. All rights reserved. 245aeb9474SGarrett D'Amore * Copyright (c) 2011 Nexenta Systems, Inc. All rights reserved. 25fa9e4066Sahrens */ 26fa9e4066Sahrens 27fa9e4066Sahrens #include <sys/zfs_context.h> 28ea8dc4b6Seschrock #include <sys/fm/fs/zfs.h> 29fa9e4066Sahrens #include <sys/spa.h> 30fa9e4066Sahrens #include <sys/txg.h> 31fa9e4066Sahrens #include <sys/spa_impl.h> 32fa9e4066Sahrens #include <sys/vdev_impl.h> 33fa9e4066Sahrens #include <sys/zio_impl.h> 34fa9e4066Sahrens #include <sys/zio_compress.h> 35fa9e4066Sahrens #include <sys/zio_checksum.h> 36b24ab676SJeff Bonwick #include <sys/dmu_objset.h> 37b24ab676SJeff Bonwick #include <sys/arc.h> 38b24ab676SJeff Bonwick #include <sys/ddt.h> 39fa9e4066Sahrens 40fa9e4066Sahrens /* 41fa9e4066Sahrens * ========================================================================== 42fa9e4066Sahrens * I/O priority table 43fa9e4066Sahrens * ========================================================================== 44fa9e4066Sahrens */ 45fa9e4066Sahrens uint8_t zio_priority_table[ZIO_PRIORITY_TABLE_SIZE] = { 46fa9e4066Sahrens 0, /* ZIO_PRIORITY_NOW */ 47fa9e4066Sahrens 0, /* ZIO_PRIORITY_SYNC_READ */ 48fa9e4066Sahrens 0, /* ZIO_PRIORITY_SYNC_WRITE */ 49fa9e4066Sahrens 0, /* ZIO_PRIORITY_LOG_WRITE */ 5080eb36f2SGeorge Wilson 1, /* ZIO_PRIORITY_CACHE_FILL */ 5180eb36f2SGeorge Wilson 1, /* ZIO_PRIORITY_AGG */ 5280eb36f2SGeorge Wilson 4, /* ZIO_PRIORITY_FREE */ 5380eb36f2SGeorge Wilson 4, /* ZIO_PRIORITY_ASYNC_WRITE */ 5480eb36f2SGeorge Wilson 6, /* ZIO_PRIORITY_ASYNC_READ */ 55fa9e4066Sahrens 10, /* ZIO_PRIORITY_RESILVER */ 56fa9e4066Sahrens 20, /* ZIO_PRIORITY_SCRUB */ 57c7cd2421SGeorge Wilson 2, /* ZIO_PRIORITY_DDT_PREFETCH */ 58fa9e4066Sahrens }; 59fa9e4066Sahrens 60fa9e4066Sahrens /* 61fa9e4066Sahrens * ========================================================================== 62fa9e4066Sahrens * I/O type descriptions 63fa9e4066Sahrens * ========================================================================== 64fa9e4066Sahrens */ 65fa9e4066Sahrens char *zio_type_name[ZIO_TYPES] = { 6680eb36f2SGeorge Wilson "zio_null", "zio_read", "zio_write", "zio_free", "zio_claim", 6780eb36f2SGeorge Wilson "zio_ioctl" 6880eb36f2SGeorge Wilson }; 69fa9e4066Sahrens 70fa9e4066Sahrens /* 71fa9e4066Sahrens * ========================================================================== 72fa9e4066Sahrens * I/O kmem caches 73fa9e4066Sahrens * ========================================================================== 74fa9e4066Sahrens */ 75ccae0b50Seschrock kmem_cache_t *zio_cache; 76a3f829aeSBill Moore kmem_cache_t *zio_link_cache; 77fa9e4066Sahrens kmem_cache_t *zio_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT]; 78ad23a2dbSjohansen kmem_cache_t *zio_data_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT]; 79ad23a2dbSjohansen 80ad23a2dbSjohansen #ifdef _KERNEL 81ad23a2dbSjohansen extern vmem_t *zio_alloc_arena; 82ad23a2dbSjohansen #endif 8309c9d376SGeorge Wilson extern int zfs_mg_alloc_failures; 84fa9e4066Sahrens 850a4e9518Sgw /* 86e14bb325SJeff Bonwick * An allocating zio is one that either currently has the DVA allocate 87e14bb325SJeff Bonwick * stage set or will have it later in its lifetime. 880a4e9518Sgw */ 89b24ab676SJeff Bonwick #define IO_IS_ALLOCATING(zio) ((zio)->io_orig_pipeline & ZIO_STAGE_DVA_ALLOCATE) 90b24ab676SJeff Bonwick 9135a5a358SJonathan Adams boolean_t zio_requeue_io_start_cut_in_line = B_TRUE; 9235a5a358SJonathan Adams 93b24ab676SJeff Bonwick #ifdef ZFS_DEBUG 94b24ab676SJeff Bonwick int zio_buf_debug_limit = 16384; 95b24ab676SJeff Bonwick #else 96b24ab676SJeff Bonwick int zio_buf_debug_limit = 0; 97b24ab676SJeff Bonwick #endif 980a4e9518Sgw 99fa9e4066Sahrens void 100fa9e4066Sahrens zio_init(void) 101fa9e4066Sahrens { 102fa9e4066Sahrens size_t c; 103ad23a2dbSjohansen vmem_t *data_alloc_arena = NULL; 104ad23a2dbSjohansen 105ad23a2dbSjohansen #ifdef _KERNEL 106ad23a2dbSjohansen data_alloc_arena = zio_alloc_arena; 107ad23a2dbSjohansen #endif 108a3f829aeSBill Moore zio_cache = kmem_cache_create("zio_cache", 109a3f829aeSBill Moore sizeof (zio_t), 0, NULL, NULL, NULL, NULL, NULL, 0); 110a3f829aeSBill Moore zio_link_cache = kmem_cache_create("zio_link_cache", 111a3f829aeSBill Moore sizeof (zio_link_t), 0, NULL, NULL, NULL, NULL, NULL, 0); 112ccae0b50Seschrock 113fa9e4066Sahrens /* 114fa9e4066Sahrens * For small buffers, we want a cache for each multiple of 115fa9e4066Sahrens * SPA_MINBLOCKSIZE. For medium-size buffers, we want a cache 116fa9e4066Sahrens * for each quarter-power of 2. For large buffers, we want 117fa9e4066Sahrens * a cache for each multiple of PAGESIZE. 118fa9e4066Sahrens */ 119fa9e4066Sahrens for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) { 120fa9e4066Sahrens size_t size = (c + 1) << SPA_MINBLOCKSHIFT; 121fa9e4066Sahrens size_t p2 = size; 122fa9e4066Sahrens size_t align = 0; 123e291592aSJonathan Adams size_t cflags = (size > zio_buf_debug_limit) ? KMC_NODEBUG : 0; 124fa9e4066Sahrens 125fa9e4066Sahrens while (p2 & (p2 - 1)) 126fa9e4066Sahrens p2 &= p2 - 1; 127fa9e4066Sahrens 128*cd1c8b85SMatthew Ahrens #ifndef _KERNEL 129*cd1c8b85SMatthew Ahrens /* 130*cd1c8b85SMatthew Ahrens * If we are using watchpoints, put each buffer on its own page, 131*cd1c8b85SMatthew Ahrens * to eliminate the performance overhead of trapping to the 132*cd1c8b85SMatthew Ahrens * kernel when modifying a non-watched buffer that shares the 133*cd1c8b85SMatthew Ahrens * page with a watched buffer. 134*cd1c8b85SMatthew Ahrens */ 135*cd1c8b85SMatthew Ahrens if (arc_watch && !IS_P2ALIGNED(size, PAGESIZE)) 136*cd1c8b85SMatthew Ahrens continue; 137*cd1c8b85SMatthew Ahrens #endif 138fa9e4066Sahrens if (size <= 4 * SPA_MINBLOCKSIZE) { 139fa9e4066Sahrens align = SPA_MINBLOCKSIZE; 140*cd1c8b85SMatthew Ahrens } else if (IS_P2ALIGNED(size, PAGESIZE)) { 141fa9e4066Sahrens align = PAGESIZE; 142*cd1c8b85SMatthew Ahrens } else if (IS_P2ALIGNED(size, p2 >> 2)) { 143fa9e4066Sahrens align = p2 >> 2; 144fa9e4066Sahrens } 145fa9e4066Sahrens 146fa9e4066Sahrens if (align != 0) { 147ad23a2dbSjohansen char name[36]; 1485ad82045Snd (void) sprintf(name, "zio_buf_%lu", (ulong_t)size); 149fa9e4066Sahrens zio_buf_cache[c] = kmem_cache_create(name, size, 150e291592aSJonathan Adams align, NULL, NULL, NULL, NULL, NULL, cflags); 151ad23a2dbSjohansen 152e291592aSJonathan Adams /* 153e291592aSJonathan Adams * Since zio_data bufs do not appear in crash dumps, we 154e291592aSJonathan Adams * pass KMC_NOTOUCH so that no allocator metadata is 155e291592aSJonathan Adams * stored with the buffers. 156e291592aSJonathan Adams */ 157ad23a2dbSjohansen (void) sprintf(name, "zio_data_buf_%lu", (ulong_t)size); 158ad23a2dbSjohansen zio_data_buf_cache[c] = kmem_cache_create(name, size, 159ad23a2dbSjohansen align, NULL, NULL, NULL, NULL, data_alloc_arena, 160e291592aSJonathan Adams cflags | KMC_NOTOUCH); 161fa9e4066Sahrens } 162fa9e4066Sahrens } 163fa9e4066Sahrens 164fa9e4066Sahrens while (--c != 0) { 165fa9e4066Sahrens ASSERT(zio_buf_cache[c] != NULL); 166fa9e4066Sahrens if (zio_buf_cache[c - 1] == NULL) 167fa9e4066Sahrens zio_buf_cache[c - 1] = zio_buf_cache[c]; 168ad23a2dbSjohansen 169ad23a2dbSjohansen ASSERT(zio_data_buf_cache[c] != NULL); 170ad23a2dbSjohansen if (zio_data_buf_cache[c - 1] == NULL) 171ad23a2dbSjohansen zio_data_buf_cache[c - 1] = zio_data_buf_cache[c]; 172fa9e4066Sahrens } 173ea8dc4b6Seschrock 17409c9d376SGeorge Wilson /* 17509c9d376SGeorge Wilson * The zio write taskqs have 1 thread per cpu, allow 1/2 of the taskqs 17609c9d376SGeorge Wilson * to fail 3 times per txg or 8 failures, whichever is greater. 17709c9d376SGeorge Wilson */ 17809c9d376SGeorge Wilson zfs_mg_alloc_failures = MAX((3 * max_ncpus / 2), 8); 17909c9d376SGeorge Wilson 180ea8dc4b6Seschrock zio_inject_init(); 181fa9e4066Sahrens } 182fa9e4066Sahrens 183fa9e4066Sahrens void 184fa9e4066Sahrens zio_fini(void) 185fa9e4066Sahrens { 186fa9e4066Sahrens size_t c; 187fa9e4066Sahrens kmem_cache_t *last_cache = NULL; 188ad23a2dbSjohansen kmem_cache_t *last_data_cache = NULL; 189fa9e4066Sahrens 190fa9e4066Sahrens for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) { 191fa9e4066Sahrens if (zio_buf_cache[c] != last_cache) { 192fa9e4066Sahrens last_cache = zio_buf_cache[c]; 193fa9e4066Sahrens kmem_cache_destroy(zio_buf_cache[c]); 194fa9e4066Sahrens } 195fa9e4066Sahrens zio_buf_cache[c] = NULL; 196ad23a2dbSjohansen 197ad23a2dbSjohansen if (zio_data_buf_cache[c] != last_data_cache) { 198ad23a2dbSjohansen last_data_cache = zio_data_buf_cache[c]; 199ad23a2dbSjohansen kmem_cache_destroy(zio_data_buf_cache[c]); 200ad23a2dbSjohansen } 201ad23a2dbSjohansen zio_data_buf_cache[c] = NULL; 202fa9e4066Sahrens } 203ea8dc4b6Seschrock 204a3f829aeSBill Moore kmem_cache_destroy(zio_link_cache); 205ccae0b50Seschrock kmem_cache_destroy(zio_cache); 206ccae0b50Seschrock 207ea8dc4b6Seschrock zio_inject_fini(); 208fa9e4066Sahrens } 209fa9e4066Sahrens 210fa9e4066Sahrens /* 211fa9e4066Sahrens * ========================================================================== 212fa9e4066Sahrens * Allocate and free I/O buffers 213fa9e4066Sahrens * ========================================================================== 214fa9e4066Sahrens */ 215ad23a2dbSjohansen 216ad23a2dbSjohansen /* 217ad23a2dbSjohansen * Use zio_buf_alloc to allocate ZFS metadata. This data will appear in a 218ad23a2dbSjohansen * crashdump if the kernel panics, so use it judiciously. Obviously, it's 219ad23a2dbSjohansen * useful to inspect ZFS metadata, but if possible, we should avoid keeping 220ad23a2dbSjohansen * excess / transient data in-core during a crashdump. 221ad23a2dbSjohansen */ 222fa9e4066Sahrens void * 223fa9e4066Sahrens zio_buf_alloc(size_t size) 224fa9e4066Sahrens { 225fa9e4066Sahrens size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; 226fa9e4066Sahrens 227fa9e4066Sahrens ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); 228fa9e4066Sahrens 2291ab7f2deSmaybee return (kmem_cache_alloc(zio_buf_cache[c], KM_PUSHPAGE)); 230fa9e4066Sahrens } 231fa9e4066Sahrens 232ad23a2dbSjohansen /* 233ad23a2dbSjohansen * Use zio_data_buf_alloc to allocate data. The data will not appear in a 234ad23a2dbSjohansen * crashdump if the kernel panics. This exists so that we will limit the amount 235ad23a2dbSjohansen * of ZFS data that shows up in a kernel crashdump. (Thus reducing the amount 236ad23a2dbSjohansen * of kernel heap dumped to disk when the kernel panics) 237ad23a2dbSjohansen */ 238ad23a2dbSjohansen void * 239ad23a2dbSjohansen zio_data_buf_alloc(size_t size) 240ad23a2dbSjohansen { 241ad23a2dbSjohansen size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; 242ad23a2dbSjohansen 243ad23a2dbSjohansen ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); 244ad23a2dbSjohansen 2451ab7f2deSmaybee return (kmem_cache_alloc(zio_data_buf_cache[c], KM_PUSHPAGE)); 246ad23a2dbSjohansen } 247ad23a2dbSjohansen 248fa9e4066Sahrens void 249fa9e4066Sahrens zio_buf_free(void *buf, size_t size) 250fa9e4066Sahrens { 251fa9e4066Sahrens size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; 252fa9e4066Sahrens 253fa9e4066Sahrens ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); 254fa9e4066Sahrens 255fa9e4066Sahrens kmem_cache_free(zio_buf_cache[c], buf); 256fa9e4066Sahrens } 257fa9e4066Sahrens 258ad23a2dbSjohansen void 259ad23a2dbSjohansen zio_data_buf_free(void *buf, size_t size) 260ad23a2dbSjohansen { 261ad23a2dbSjohansen size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; 262ad23a2dbSjohansen 263ad23a2dbSjohansen ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); 264ad23a2dbSjohansen 265ad23a2dbSjohansen kmem_cache_free(zio_data_buf_cache[c], buf); 266ad23a2dbSjohansen } 267b3995adbSahrens 268fa9e4066Sahrens /* 269fa9e4066Sahrens * ========================================================================== 270fa9e4066Sahrens * Push and pop I/O transform buffers 271fa9e4066Sahrens * ========================================================================== 272fa9e4066Sahrens */ 273fa9e4066Sahrens static void 274e14bb325SJeff Bonwick zio_push_transform(zio_t *zio, void *data, uint64_t size, uint64_t bufsize, 275e14bb325SJeff Bonwick zio_transform_func_t *transform) 276fa9e4066Sahrens { 277fa9e4066Sahrens zio_transform_t *zt = kmem_alloc(sizeof (zio_transform_t), KM_SLEEP); 278fa9e4066Sahrens 279e14bb325SJeff Bonwick zt->zt_orig_data = zio->io_data; 280e14bb325SJeff Bonwick zt->zt_orig_size = zio->io_size; 281fa9e4066Sahrens zt->zt_bufsize = bufsize; 282e14bb325SJeff Bonwick zt->zt_transform = transform; 283fa9e4066Sahrens 284fa9e4066Sahrens zt->zt_next = zio->io_transform_stack; 285fa9e4066Sahrens zio->io_transform_stack = zt; 286fa9e4066Sahrens 287fa9e4066Sahrens zio->io_data = data; 288fa9e4066Sahrens zio->io_size = size; 289fa9e4066Sahrens } 290fa9e4066Sahrens 291fa9e4066Sahrens static void 292e14bb325SJeff Bonwick zio_pop_transforms(zio_t *zio) 293fa9e4066Sahrens { 294e14bb325SJeff Bonwick zio_transform_t *zt; 295e14bb325SJeff Bonwick 296e14bb325SJeff Bonwick while ((zt = zio->io_transform_stack) != NULL) { 297e14bb325SJeff Bonwick if (zt->zt_transform != NULL) 298e14bb325SJeff Bonwick zt->zt_transform(zio, 299e14bb325SJeff Bonwick zt->zt_orig_data, zt->zt_orig_size); 300fa9e4066Sahrens 301b24ab676SJeff Bonwick if (zt->zt_bufsize != 0) 302b24ab676SJeff Bonwick zio_buf_free(zio->io_data, zt->zt_bufsize); 303fa9e4066Sahrens 304e14bb325SJeff Bonwick zio->io_data = zt->zt_orig_data; 305e14bb325SJeff Bonwick zio->io_size = zt->zt_orig_size; 306e14bb325SJeff Bonwick zio->io_transform_stack = zt->zt_next; 307fa9e4066Sahrens 308e14bb325SJeff Bonwick kmem_free(zt, sizeof (zio_transform_t)); 309fa9e4066Sahrens } 310fa9e4066Sahrens } 311fa9e4066Sahrens 312e14bb325SJeff Bonwick /* 313e14bb325SJeff Bonwick * ========================================================================== 314e14bb325SJeff Bonwick * I/O transform callbacks for subblocks and decompression 315e14bb325SJeff Bonwick * ========================================================================== 316e14bb325SJeff Bonwick */ 317e14bb325SJeff Bonwick static void 318e14bb325SJeff Bonwick zio_subblock(zio_t *zio, void *data, uint64_t size) 319e14bb325SJeff Bonwick { 320e14bb325SJeff Bonwick ASSERT(zio->io_size > size); 321e14bb325SJeff Bonwick 322e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) 323e14bb325SJeff Bonwick bcopy(zio->io_data, data, size); 324e14bb325SJeff Bonwick } 325e14bb325SJeff Bonwick 326e14bb325SJeff Bonwick static void 327e14bb325SJeff Bonwick zio_decompress(zio_t *zio, void *data, uint64_t size) 328e14bb325SJeff Bonwick { 329e14bb325SJeff Bonwick if (zio->io_error == 0 && 330e14bb325SJeff Bonwick zio_decompress_data(BP_GET_COMPRESS(zio->io_bp), 331b24ab676SJeff Bonwick zio->io_data, data, zio->io_size, size) != 0) 332e14bb325SJeff Bonwick zio->io_error = EIO; 333e14bb325SJeff Bonwick } 334e14bb325SJeff Bonwick 335e14bb325SJeff Bonwick /* 336e14bb325SJeff Bonwick * ========================================================================== 337e14bb325SJeff Bonwick * I/O parent/child relationships and pipeline interlocks 338e14bb325SJeff Bonwick * ========================================================================== 339e14bb325SJeff Bonwick */ 340a3f829aeSBill Moore /* 341a3f829aeSBill Moore * NOTE - Callers to zio_walk_parents() and zio_walk_children must 342a3f829aeSBill Moore * continue calling these functions until they return NULL. 343a3f829aeSBill Moore * Otherwise, the next caller will pick up the list walk in 344a3f829aeSBill Moore * some indeterminate state. (Otherwise every caller would 345a3f829aeSBill Moore * have to pass in a cookie to keep the state represented by 346a3f829aeSBill Moore * io_walk_link, which gets annoying.) 347a3f829aeSBill Moore */ 348a3f829aeSBill Moore zio_t * 349a3f829aeSBill Moore zio_walk_parents(zio_t *cio) 350a3f829aeSBill Moore { 351a3f829aeSBill Moore zio_link_t *zl = cio->io_walk_link; 352a3f829aeSBill Moore list_t *pl = &cio->io_parent_list; 353e14bb325SJeff Bonwick 354a3f829aeSBill Moore zl = (zl == NULL) ? list_head(pl) : list_next(pl, zl); 355a3f829aeSBill Moore cio->io_walk_link = zl; 356a3f829aeSBill Moore 357a3f829aeSBill Moore if (zl == NULL) 358a3f829aeSBill Moore return (NULL); 359a3f829aeSBill Moore 360a3f829aeSBill Moore ASSERT(zl->zl_child == cio); 361a3f829aeSBill Moore return (zl->zl_parent); 362a3f829aeSBill Moore } 363a3f829aeSBill Moore 364a3f829aeSBill Moore zio_t * 365a3f829aeSBill Moore zio_walk_children(zio_t *pio) 366a3f829aeSBill Moore { 367a3f829aeSBill Moore zio_link_t *zl = pio->io_walk_link; 368a3f829aeSBill Moore list_t *cl = &pio->io_child_list; 369a3f829aeSBill Moore 370a3f829aeSBill Moore zl = (zl == NULL) ? list_head(cl) : list_next(cl, zl); 371a3f829aeSBill Moore pio->io_walk_link = zl; 372a3f829aeSBill Moore 373a3f829aeSBill Moore if (zl == NULL) 374a3f829aeSBill Moore return (NULL); 375a3f829aeSBill Moore 376a3f829aeSBill Moore ASSERT(zl->zl_parent == pio); 377a3f829aeSBill Moore return (zl->zl_child); 378a3f829aeSBill Moore } 379a3f829aeSBill Moore 380a3f829aeSBill Moore zio_t * 381a3f829aeSBill Moore zio_unique_parent(zio_t *cio) 382a3f829aeSBill Moore { 383a3f829aeSBill Moore zio_t *pio = zio_walk_parents(cio); 384a3f829aeSBill Moore 385a3f829aeSBill Moore VERIFY(zio_walk_parents(cio) == NULL); 386a3f829aeSBill Moore return (pio); 387a3f829aeSBill Moore } 388a3f829aeSBill Moore 389a3f829aeSBill Moore void 390a3f829aeSBill Moore zio_add_child(zio_t *pio, zio_t *cio) 391e14bb325SJeff Bonwick { 392a3f829aeSBill Moore zio_link_t *zl = kmem_cache_alloc(zio_link_cache, KM_SLEEP); 393a3f829aeSBill Moore 394a3f829aeSBill Moore /* 395a3f829aeSBill Moore * Logical I/Os can have logical, gang, or vdev children. 396a3f829aeSBill Moore * Gang I/Os can have gang or vdev children. 397a3f829aeSBill Moore * Vdev I/Os can only have vdev children. 398a3f829aeSBill Moore * The following ASSERT captures all of these constraints. 399a3f829aeSBill Moore */ 400a3f829aeSBill Moore ASSERT(cio->io_child_type <= pio->io_child_type); 401a3f829aeSBill Moore 402a3f829aeSBill Moore zl->zl_parent = pio; 403a3f829aeSBill Moore zl->zl_child = cio; 404a3f829aeSBill Moore 405a3f829aeSBill Moore mutex_enter(&cio->io_lock); 406e14bb325SJeff Bonwick mutex_enter(&pio->io_lock); 407a3f829aeSBill Moore 408a3f829aeSBill Moore ASSERT(pio->io_state[ZIO_WAIT_DONE] == 0); 409a3f829aeSBill Moore 410a3f829aeSBill Moore for (int w = 0; w < ZIO_WAIT_TYPES; w++) 411a3f829aeSBill Moore pio->io_children[cio->io_child_type][w] += !cio->io_state[w]; 412a3f829aeSBill Moore 413a3f829aeSBill Moore list_insert_head(&pio->io_child_list, zl); 414a3f829aeSBill Moore list_insert_head(&cio->io_parent_list, zl); 415a3f829aeSBill Moore 416b24ab676SJeff Bonwick pio->io_child_count++; 417b24ab676SJeff Bonwick cio->io_parent_count++; 418b24ab676SJeff Bonwick 419e14bb325SJeff Bonwick mutex_exit(&pio->io_lock); 420a3f829aeSBill Moore mutex_exit(&cio->io_lock); 421e14bb325SJeff Bonwick } 422e14bb325SJeff Bonwick 423fa9e4066Sahrens static void 424a3f829aeSBill Moore zio_remove_child(zio_t *pio, zio_t *cio, zio_link_t *zl) 425e14bb325SJeff Bonwick { 426a3f829aeSBill Moore ASSERT(zl->zl_parent == pio); 427a3f829aeSBill Moore ASSERT(zl->zl_child == cio); 428e14bb325SJeff Bonwick 429a3f829aeSBill Moore mutex_enter(&cio->io_lock); 430e14bb325SJeff Bonwick mutex_enter(&pio->io_lock); 431a3f829aeSBill Moore 432a3f829aeSBill Moore list_remove(&pio->io_child_list, zl); 433a3f829aeSBill Moore list_remove(&cio->io_parent_list, zl); 434a3f829aeSBill Moore 435b24ab676SJeff Bonwick pio->io_child_count--; 436b24ab676SJeff Bonwick cio->io_parent_count--; 437b24ab676SJeff Bonwick 438e14bb325SJeff Bonwick mutex_exit(&pio->io_lock); 439a3f829aeSBill Moore mutex_exit(&cio->io_lock); 440a3f829aeSBill Moore 441a3f829aeSBill Moore kmem_cache_free(zio_link_cache, zl); 442e14bb325SJeff Bonwick } 443e14bb325SJeff Bonwick 444e14bb325SJeff Bonwick static boolean_t 445e14bb325SJeff Bonwick zio_wait_for_children(zio_t *zio, enum zio_child child, enum zio_wait_type wait) 446fa9e4066Sahrens { 447e14bb325SJeff Bonwick uint64_t *countp = &zio->io_children[child][wait]; 448e14bb325SJeff Bonwick boolean_t waiting = B_FALSE; 449e14bb325SJeff Bonwick 450e14bb325SJeff Bonwick mutex_enter(&zio->io_lock); 451e14bb325SJeff Bonwick ASSERT(zio->io_stall == NULL); 452e14bb325SJeff Bonwick if (*countp != 0) { 453b24ab676SJeff Bonwick zio->io_stage >>= 1; 454e14bb325SJeff Bonwick zio->io_stall = countp; 455e14bb325SJeff Bonwick waiting = B_TRUE; 456e14bb325SJeff Bonwick } 457e14bb325SJeff Bonwick mutex_exit(&zio->io_lock); 458e14bb325SJeff Bonwick 459e14bb325SJeff Bonwick return (waiting); 460e14bb325SJeff Bonwick } 461fa9e4066Sahrens 462e14bb325SJeff Bonwick static void 463e14bb325SJeff Bonwick zio_notify_parent(zio_t *pio, zio_t *zio, enum zio_wait_type wait) 464e14bb325SJeff Bonwick { 465e14bb325SJeff Bonwick uint64_t *countp = &pio->io_children[zio->io_child_type][wait]; 466e14bb325SJeff Bonwick int *errorp = &pio->io_child_error[zio->io_child_type]; 467fa9e4066Sahrens 468e14bb325SJeff Bonwick mutex_enter(&pio->io_lock); 469e14bb325SJeff Bonwick if (zio->io_error && !(zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 470e14bb325SJeff Bonwick *errorp = zio_worst_error(*errorp, zio->io_error); 471e14bb325SJeff Bonwick pio->io_reexecute |= zio->io_reexecute; 472e14bb325SJeff Bonwick ASSERT3U(*countp, >, 0); 473e14bb325SJeff Bonwick if (--*countp == 0 && pio->io_stall == countp) { 474e14bb325SJeff Bonwick pio->io_stall = NULL; 475e14bb325SJeff Bonwick mutex_exit(&pio->io_lock); 476e14bb325SJeff Bonwick zio_execute(pio); 477e14bb325SJeff Bonwick } else { 478e14bb325SJeff Bonwick mutex_exit(&pio->io_lock); 479fa9e4066Sahrens } 480fa9e4066Sahrens } 481fa9e4066Sahrens 482e14bb325SJeff Bonwick static void 483e14bb325SJeff Bonwick zio_inherit_child_errors(zio_t *zio, enum zio_child c) 484e14bb325SJeff Bonwick { 485e14bb325SJeff Bonwick if (zio->io_child_error[c] != 0 && zio->io_error == 0) 486e14bb325SJeff Bonwick zio->io_error = zio->io_child_error[c]; 487e14bb325SJeff Bonwick } 488e14bb325SJeff Bonwick 489fa9e4066Sahrens /* 490fa9e4066Sahrens * ========================================================================== 491e14bb325SJeff Bonwick * Create the various types of I/O (read, write, free, etc) 492fa9e4066Sahrens * ========================================================================== 493fa9e4066Sahrens */ 494fa9e4066Sahrens static zio_t * 495b24ab676SJeff Bonwick zio_create(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, 496fa9e4066Sahrens void *data, uint64_t size, zio_done_func_t *done, void *private, 497b24ab676SJeff Bonwick zio_type_t type, int priority, enum zio_flag flags, 498b24ab676SJeff Bonwick vdev_t *vd, uint64_t offset, const zbookmark_t *zb, 499b24ab676SJeff Bonwick enum zio_stage stage, enum zio_stage pipeline) 500fa9e4066Sahrens { 501fa9e4066Sahrens zio_t *zio; 502fa9e4066Sahrens 503fa9e4066Sahrens ASSERT3U(size, <=, SPA_MAXBLOCKSIZE); 504fa9e4066Sahrens ASSERT(P2PHASE(size, SPA_MINBLOCKSIZE) == 0); 505e14bb325SJeff Bonwick ASSERT(P2PHASE(offset, SPA_MINBLOCKSIZE) == 0); 506fa9e4066Sahrens 507e14bb325SJeff Bonwick ASSERT(!vd || spa_config_held(spa, SCL_STATE_ALL, RW_READER)); 508e14bb325SJeff Bonwick ASSERT(!bp || !(flags & ZIO_FLAG_CONFIG_WRITER)); 509e14bb325SJeff Bonwick ASSERT(vd || stage == ZIO_STAGE_OPEN); 510088f3894Sahrens 511ccae0b50Seschrock zio = kmem_cache_alloc(zio_cache, KM_SLEEP); 512ccae0b50Seschrock bzero(zio, sizeof (zio_t)); 513e14bb325SJeff Bonwick 514e14bb325SJeff Bonwick mutex_init(&zio->io_lock, NULL, MUTEX_DEFAULT, NULL); 515e14bb325SJeff Bonwick cv_init(&zio->io_cv, NULL, CV_DEFAULT, NULL); 516e14bb325SJeff Bonwick 517a3f829aeSBill Moore list_create(&zio->io_parent_list, sizeof (zio_link_t), 518a3f829aeSBill Moore offsetof(zio_link_t, zl_parent_node)); 519a3f829aeSBill Moore list_create(&zio->io_child_list, sizeof (zio_link_t), 520a3f829aeSBill Moore offsetof(zio_link_t, zl_child_node)); 521a3f829aeSBill Moore 522e14bb325SJeff Bonwick if (vd != NULL) 523e14bb325SJeff Bonwick zio->io_child_type = ZIO_CHILD_VDEV; 524e14bb325SJeff Bonwick else if (flags & ZIO_FLAG_GANG_CHILD) 525e14bb325SJeff Bonwick zio->io_child_type = ZIO_CHILD_GANG; 526b24ab676SJeff Bonwick else if (flags & ZIO_FLAG_DDT_CHILD) 527b24ab676SJeff Bonwick zio->io_child_type = ZIO_CHILD_DDT; 528e14bb325SJeff Bonwick else 529e14bb325SJeff Bonwick zio->io_child_type = ZIO_CHILD_LOGICAL; 530e14bb325SJeff Bonwick 531fa9e4066Sahrens if (bp != NULL) { 532b24ab676SJeff Bonwick zio->io_bp = (blkptr_t *)bp; 533fa9e4066Sahrens zio->io_bp_copy = *bp; 534fa9e4066Sahrens zio->io_bp_orig = *bp; 535b24ab676SJeff Bonwick if (type != ZIO_TYPE_WRITE || 536b24ab676SJeff Bonwick zio->io_child_type == ZIO_CHILD_DDT) 537e14bb325SJeff Bonwick zio->io_bp = &zio->io_bp_copy; /* so caller can free */ 538f5383399SBill Moore if (zio->io_child_type == ZIO_CHILD_LOGICAL) 539e14bb325SJeff Bonwick zio->io_logical = zio; 540f5383399SBill Moore if (zio->io_child_type > ZIO_CHILD_GANG && BP_IS_GANG(bp)) 541f5383399SBill Moore pipeline |= ZIO_GANG_STAGES; 542fa9e4066Sahrens } 543e14bb325SJeff Bonwick 544e14bb325SJeff Bonwick zio->io_spa = spa; 545e14bb325SJeff Bonwick zio->io_txg = txg; 546fa9e4066Sahrens zio->io_done = done; 547fa9e4066Sahrens zio->io_private = private; 548fa9e4066Sahrens zio->io_type = type; 549fa9e4066Sahrens zio->io_priority = priority; 550e14bb325SJeff Bonwick zio->io_vd = vd; 551e14bb325SJeff Bonwick zio->io_offset = offset; 552b24ab676SJeff Bonwick zio->io_orig_data = zio->io_data = data; 553b24ab676SJeff Bonwick zio->io_orig_size = zio->io_size = size; 554e14bb325SJeff Bonwick zio->io_orig_flags = zio->io_flags = flags; 555e14bb325SJeff Bonwick zio->io_orig_stage = zio->io_stage = stage; 556e14bb325SJeff Bonwick zio->io_orig_pipeline = zio->io_pipeline = pipeline; 557fa9e4066Sahrens 558a3f829aeSBill Moore zio->io_state[ZIO_WAIT_READY] = (stage >= ZIO_STAGE_READY); 559a3f829aeSBill Moore zio->io_state[ZIO_WAIT_DONE] = (stage >= ZIO_STAGE_DONE); 560a3f829aeSBill Moore 561e14bb325SJeff Bonwick if (zb != NULL) 562e14bb325SJeff Bonwick zio->io_bookmark = *zb; 563e14bb325SJeff Bonwick 564e14bb325SJeff Bonwick if (pio != NULL) { 565e14bb325SJeff Bonwick if (zio->io_logical == NULL) 566ea8dc4b6Seschrock zio->io_logical = pio->io_logical; 567f5383399SBill Moore if (zio->io_child_type == ZIO_CHILD_GANG) 568f5383399SBill Moore zio->io_gang_leader = pio->io_gang_leader; 569e14bb325SJeff Bonwick zio_add_child(pio, zio); 570fa9e4066Sahrens } 571fa9e4066Sahrens 572fa9e4066Sahrens return (zio); 573fa9e4066Sahrens } 574fa9e4066Sahrens 5750a4e9518Sgw static void 576e14bb325SJeff Bonwick zio_destroy(zio_t *zio) 5770a4e9518Sgw { 578a3f829aeSBill Moore list_destroy(&zio->io_parent_list); 579a3f829aeSBill Moore list_destroy(&zio->io_child_list); 580e14bb325SJeff Bonwick mutex_destroy(&zio->io_lock); 581e14bb325SJeff Bonwick cv_destroy(&zio->io_cv); 582e14bb325SJeff Bonwick kmem_cache_free(zio_cache, zio); 5830a4e9518Sgw } 5840a4e9518Sgw 585fa9e4066Sahrens zio_t * 586a3f829aeSBill Moore zio_null(zio_t *pio, spa_t *spa, vdev_t *vd, zio_done_func_t *done, 587b24ab676SJeff Bonwick void *private, enum zio_flag flags) 588fa9e4066Sahrens { 589fa9e4066Sahrens zio_t *zio; 590fa9e4066Sahrens 591fa9e4066Sahrens zio = zio_create(pio, spa, 0, NULL, NULL, 0, done, private, 592a3f829aeSBill Moore ZIO_TYPE_NULL, ZIO_PRIORITY_NOW, flags, vd, 0, NULL, 593e14bb325SJeff Bonwick ZIO_STAGE_OPEN, ZIO_INTERLOCK_PIPELINE); 594fa9e4066Sahrens 595fa9e4066Sahrens return (zio); 596fa9e4066Sahrens } 597fa9e4066Sahrens 598fa9e4066Sahrens zio_t * 599b24ab676SJeff Bonwick zio_root(spa_t *spa, zio_done_func_t *done, void *private, enum zio_flag flags) 600fa9e4066Sahrens { 601a3f829aeSBill Moore return (zio_null(NULL, spa, NULL, done, private, flags)); 602fa9e4066Sahrens } 603fa9e4066Sahrens 604fa9e4066Sahrens zio_t * 605e14bb325SJeff Bonwick zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, 606e14bb325SJeff Bonwick void *data, uint64_t size, zio_done_func_t *done, void *private, 607b24ab676SJeff Bonwick int priority, enum zio_flag flags, const zbookmark_t *zb) 608fa9e4066Sahrens { 609fa9e4066Sahrens zio_t *zio; 610fa9e4066Sahrens 611b24ab676SJeff Bonwick zio = zio_create(pio, spa, BP_PHYSICAL_BIRTH(bp), bp, 612088f3894Sahrens data, size, done, private, 613e14bb325SJeff Bonwick ZIO_TYPE_READ, priority, flags, NULL, 0, zb, 614b24ab676SJeff Bonwick ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ? 615b24ab676SJeff Bonwick ZIO_DDT_CHILD_READ_PIPELINE : ZIO_READ_PIPELINE); 616fa9e4066Sahrens 617fa9e4066Sahrens return (zio); 618fa9e4066Sahrens } 619fa9e4066Sahrens 620fa9e4066Sahrens zio_t * 621e14bb325SJeff Bonwick zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, 622b24ab676SJeff Bonwick void *data, uint64_t size, const zio_prop_t *zp, 623e14bb325SJeff Bonwick zio_done_func_t *ready, zio_done_func_t *done, void *private, 624b24ab676SJeff Bonwick int priority, enum zio_flag flags, const zbookmark_t *zb) 625fa9e4066Sahrens { 626fa9e4066Sahrens zio_t *zio; 627fa9e4066Sahrens 628e14bb325SJeff Bonwick ASSERT(zp->zp_checksum >= ZIO_CHECKSUM_OFF && 629e14bb325SJeff Bonwick zp->zp_checksum < ZIO_CHECKSUM_FUNCTIONS && 630e14bb325SJeff Bonwick zp->zp_compress >= ZIO_COMPRESS_OFF && 631e14bb325SJeff Bonwick zp->zp_compress < ZIO_COMPRESS_FUNCTIONS && 632ad135b5dSChristopher Siden DMU_OT_IS_VALID(zp->zp_type) && 633e14bb325SJeff Bonwick zp->zp_level < 32 && 634b24ab676SJeff Bonwick zp->zp_copies > 0 && 635b24ab676SJeff Bonwick zp->zp_copies <= spa_max_replication(spa) && 636b24ab676SJeff Bonwick zp->zp_dedup <= 1 && 637b24ab676SJeff Bonwick zp->zp_dedup_verify <= 1); 6380a4e9518Sgw 639fa9e4066Sahrens zio = zio_create(pio, spa, txg, bp, data, size, done, private, 640e14bb325SJeff Bonwick ZIO_TYPE_WRITE, priority, flags, NULL, 0, zb, 641b24ab676SJeff Bonwick ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ? 642b24ab676SJeff Bonwick ZIO_DDT_CHILD_WRITE_PIPELINE : ZIO_WRITE_PIPELINE); 643fa9e4066Sahrens 644c717a561Smaybee zio->io_ready = ready; 645e14bb325SJeff Bonwick zio->io_prop = *zp; 646fa9e4066Sahrens 647fa9e4066Sahrens return (zio); 648fa9e4066Sahrens } 649fa9e4066Sahrens 650fa9e4066Sahrens zio_t * 651e14bb325SJeff Bonwick zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, void *data, 652e14bb325SJeff Bonwick uint64_t size, zio_done_func_t *done, void *private, int priority, 653b24ab676SJeff Bonwick enum zio_flag flags, zbookmark_t *zb) 654fa9e4066Sahrens { 655fa9e4066Sahrens zio_t *zio; 656fa9e4066Sahrens 657fa9e4066Sahrens zio = zio_create(pio, spa, txg, bp, data, size, done, private, 658e14bb325SJeff Bonwick ZIO_TYPE_WRITE, priority, flags, NULL, 0, zb, 659e14bb325SJeff Bonwick ZIO_STAGE_OPEN, ZIO_REWRITE_PIPELINE); 660fa9e4066Sahrens 661fa9e4066Sahrens return (zio); 662fa9e4066Sahrens } 663fa9e4066Sahrens 664b24ab676SJeff Bonwick void 665b24ab676SJeff Bonwick zio_write_override(zio_t *zio, blkptr_t *bp, int copies) 666b24ab676SJeff Bonwick { 667b24ab676SJeff Bonwick ASSERT(zio->io_type == ZIO_TYPE_WRITE); 668b24ab676SJeff Bonwick ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 669b24ab676SJeff Bonwick ASSERT(zio->io_stage == ZIO_STAGE_OPEN); 670b24ab676SJeff Bonwick ASSERT(zio->io_txg == spa_syncing_txg(zio->io_spa)); 671b24ab676SJeff Bonwick 672b24ab676SJeff Bonwick zio->io_prop.zp_copies = copies; 673b24ab676SJeff Bonwick zio->io_bp_override = bp; 674b24ab676SJeff Bonwick } 675b24ab676SJeff Bonwick 676b24ab676SJeff Bonwick void 677b24ab676SJeff Bonwick zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp) 678b24ab676SJeff Bonwick { 679cde58dbcSMatthew Ahrens bplist_append(&spa->spa_free_bplist[txg & TXG_MASK], bp); 680b24ab676SJeff Bonwick } 681b24ab676SJeff Bonwick 682fa9e4066Sahrens zio_t * 683b24ab676SJeff Bonwick zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, 684b24ab676SJeff Bonwick enum zio_flag flags) 685fa9e4066Sahrens { 686fa9e4066Sahrens zio_t *zio; 687fa9e4066Sahrens 6883f9d6ad7SLin Ling dprintf_bp(bp, "freeing in txg %llu, pass %u", 6893f9d6ad7SLin Ling (longlong_t)txg, spa->spa_sync_pass); 6903f9d6ad7SLin Ling 691fa9e4066Sahrens ASSERT(!BP_IS_HOLE(bp)); 692b24ab676SJeff Bonwick ASSERT(spa_syncing_txg(spa) == txg); 693b24ab676SJeff Bonwick ASSERT(spa_sync_pass(spa) <= SYNC_PASS_DEFERRED_FREE); 694fa9e4066Sahrens 695e14bb325SJeff Bonwick zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp), 696b24ab676SJeff Bonwick NULL, NULL, ZIO_TYPE_FREE, ZIO_PRIORITY_FREE, flags, 697e14bb325SJeff Bonwick NULL, 0, NULL, ZIO_STAGE_OPEN, ZIO_FREE_PIPELINE); 698fa9e4066Sahrens 699fa9e4066Sahrens return (zio); 700fa9e4066Sahrens } 701fa9e4066Sahrens 702fa9e4066Sahrens zio_t * 703b24ab676SJeff Bonwick zio_claim(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, 704b24ab676SJeff Bonwick zio_done_func_t *done, void *private, enum zio_flag flags) 705fa9e4066Sahrens { 706fa9e4066Sahrens zio_t *zio; 707fa9e4066Sahrens 708fa9e4066Sahrens /* 709fa9e4066Sahrens * A claim is an allocation of a specific block. Claims are needed 710fa9e4066Sahrens * to support immediate writes in the intent log. The issue is that 711fa9e4066Sahrens * immediate writes contain committed data, but in a txg that was 712fa9e4066Sahrens * *not* committed. Upon opening the pool after an unclean shutdown, 713fa9e4066Sahrens * the intent log claims all blocks that contain immediate write data 714fa9e4066Sahrens * so that the SPA knows they're in use. 715fa9e4066Sahrens * 716fa9e4066Sahrens * All claims *must* be resolved in the first txg -- before the SPA 717fa9e4066Sahrens * starts allocating blocks -- so that nothing is allocated twice. 718b24ab676SJeff Bonwick * If txg == 0 we just verify that the block is claimable. 719fa9e4066Sahrens */ 720fa9e4066Sahrens ASSERT3U(spa->spa_uberblock.ub_rootbp.blk_birth, <, spa_first_txg(spa)); 721b24ab676SJeff Bonwick ASSERT(txg == spa_first_txg(spa) || txg == 0); 722b24ab676SJeff Bonwick ASSERT(!BP_GET_DEDUP(bp) || !spa_writeable(spa)); /* zdb(1M) */ 723fa9e4066Sahrens 724e14bb325SJeff Bonwick zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp), 725e14bb325SJeff Bonwick done, private, ZIO_TYPE_CLAIM, ZIO_PRIORITY_NOW, flags, 726e14bb325SJeff Bonwick NULL, 0, NULL, ZIO_STAGE_OPEN, ZIO_CLAIM_PIPELINE); 727fa9e4066Sahrens 728fa9e4066Sahrens return (zio); 729fa9e4066Sahrens } 730fa9e4066Sahrens 731fa9e4066Sahrens zio_t * 732fa9e4066Sahrens zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd, 733b24ab676SJeff Bonwick zio_done_func_t *done, void *private, int priority, enum zio_flag flags) 734fa9e4066Sahrens { 735fa9e4066Sahrens zio_t *zio; 736fa9e4066Sahrens int c; 737fa9e4066Sahrens 738fa9e4066Sahrens if (vd->vdev_children == 0) { 739fa9e4066Sahrens zio = zio_create(pio, spa, 0, NULL, NULL, 0, done, private, 740e14bb325SJeff Bonwick ZIO_TYPE_IOCTL, priority, flags, vd, 0, NULL, 741fa9e4066Sahrens ZIO_STAGE_OPEN, ZIO_IOCTL_PIPELINE); 742fa9e4066Sahrens 743fa9e4066Sahrens zio->io_cmd = cmd; 744fa9e4066Sahrens } else { 745a3f829aeSBill Moore zio = zio_null(pio, spa, NULL, NULL, NULL, flags); 746fa9e4066Sahrens 747fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 748fa9e4066Sahrens zio_nowait(zio_ioctl(zio, spa, vd->vdev_child[c], cmd, 749fa9e4066Sahrens done, private, priority, flags)); 750fa9e4066Sahrens } 751fa9e4066Sahrens 752fa9e4066Sahrens return (zio); 753fa9e4066Sahrens } 754fa9e4066Sahrens 755fa9e4066Sahrens zio_t * 756fa9e4066Sahrens zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, 757fa9e4066Sahrens void *data, int checksum, zio_done_func_t *done, void *private, 758b24ab676SJeff Bonwick int priority, enum zio_flag flags, boolean_t labels) 759fa9e4066Sahrens { 760fa9e4066Sahrens zio_t *zio; 7610a4e9518Sgw 762e14bb325SJeff Bonwick ASSERT(vd->vdev_children == 0); 763e14bb325SJeff Bonwick ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE || 764e14bb325SJeff Bonwick offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE); 765e14bb325SJeff Bonwick ASSERT3U(offset + size, <=, vd->vdev_psize); 766fa9e4066Sahrens 767e14bb325SJeff Bonwick zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private, 768e14bb325SJeff Bonwick ZIO_TYPE_READ, priority, flags, vd, offset, NULL, 769fa9e4066Sahrens ZIO_STAGE_OPEN, ZIO_READ_PHYS_PIPELINE); 770fa9e4066Sahrens 771e14bb325SJeff Bonwick zio->io_prop.zp_checksum = checksum; 772fa9e4066Sahrens 773fa9e4066Sahrens return (zio); 774fa9e4066Sahrens } 775fa9e4066Sahrens 776fa9e4066Sahrens zio_t * 777fa9e4066Sahrens zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, 778fa9e4066Sahrens void *data, int checksum, zio_done_func_t *done, void *private, 779b24ab676SJeff Bonwick int priority, enum zio_flag flags, boolean_t labels) 780fa9e4066Sahrens { 781fa9e4066Sahrens zio_t *zio; 7820a4e9518Sgw 783e14bb325SJeff Bonwick ASSERT(vd->vdev_children == 0); 784e14bb325SJeff Bonwick ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE || 785e14bb325SJeff Bonwick offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE); 786e14bb325SJeff Bonwick ASSERT3U(offset + size, <=, vd->vdev_psize); 787fa9e4066Sahrens 788e14bb325SJeff Bonwick zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private, 789e14bb325SJeff Bonwick ZIO_TYPE_WRITE, priority, flags, vd, offset, NULL, 790fa9e4066Sahrens ZIO_STAGE_OPEN, ZIO_WRITE_PHYS_PIPELINE); 791fa9e4066Sahrens 792e14bb325SJeff Bonwick zio->io_prop.zp_checksum = checksum; 793fa9e4066Sahrens 7946e1f5caaSNeil Perrin if (zio_checksum_table[checksum].ci_eck) { 795fa9e4066Sahrens /* 7966e1f5caaSNeil Perrin * zec checksums are necessarily destructive -- they modify 797e14bb325SJeff Bonwick * the end of the write buffer to hold the verifier/checksum. 798fa9e4066Sahrens * Therefore, we must make a local copy in case the data is 799e14bb325SJeff Bonwick * being written to multiple places in parallel. 800fa9e4066Sahrens */ 801e14bb325SJeff Bonwick void *wbuf = zio_buf_alloc(size); 802fa9e4066Sahrens bcopy(data, wbuf, size); 803e14bb325SJeff Bonwick zio_push_transform(zio, wbuf, size, size, NULL); 804fa9e4066Sahrens } 805fa9e4066Sahrens 806fa9e4066Sahrens return (zio); 807fa9e4066Sahrens } 808fa9e4066Sahrens 809fa9e4066Sahrens /* 810e14bb325SJeff Bonwick * Create a child I/O to do some work for us. 811fa9e4066Sahrens */ 812fa9e4066Sahrens zio_t * 813e14bb325SJeff Bonwick zio_vdev_child_io(zio_t *pio, blkptr_t *bp, vdev_t *vd, uint64_t offset, 814b24ab676SJeff Bonwick void *data, uint64_t size, int type, int priority, enum zio_flag flags, 815fa9e4066Sahrens zio_done_func_t *done, void *private) 816fa9e4066Sahrens { 817b24ab676SJeff Bonwick enum zio_stage pipeline = ZIO_VDEV_CHILD_PIPELINE; 818e14bb325SJeff Bonwick zio_t *zio; 819e14bb325SJeff Bonwick 820e14bb325SJeff Bonwick ASSERT(vd->vdev_parent == 821e14bb325SJeff Bonwick (pio->io_vd ? pio->io_vd : pio->io_spa->spa_root_vdev)); 822fa9e4066Sahrens 823fa9e4066Sahrens if (type == ZIO_TYPE_READ && bp != NULL) { 824fa9e4066Sahrens /* 825fa9e4066Sahrens * If we have the bp, then the child should perform the 826fa9e4066Sahrens * checksum and the parent need not. This pushes error 827fa9e4066Sahrens * detection as close to the leaves as possible and 828fa9e4066Sahrens * eliminates redundant checksums in the interior nodes. 829fa9e4066Sahrens */ 830b24ab676SJeff Bonwick pipeline |= ZIO_STAGE_CHECKSUM_VERIFY; 831b24ab676SJeff Bonwick pio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY; 832fa9e4066Sahrens } 833fa9e4066Sahrens 834e14bb325SJeff Bonwick if (vd->vdev_children == 0) 835e14bb325SJeff Bonwick offset += VDEV_LABEL_START_SIZE; 836e14bb325SJeff Bonwick 837b24ab676SJeff Bonwick flags |= ZIO_VDEV_CHILD_FLAGS(pio) | ZIO_FLAG_DONT_PROPAGATE; 838b24ab676SJeff Bonwick 839b24ab676SJeff Bonwick /* 840b24ab676SJeff Bonwick * If we've decided to do a repair, the write is not speculative -- 841b24ab676SJeff Bonwick * even if the original read was. 842b24ab676SJeff Bonwick */ 843b24ab676SJeff Bonwick if (flags & ZIO_FLAG_IO_REPAIR) 844b24ab676SJeff Bonwick flags &= ~ZIO_FLAG_SPECULATIVE; 845b24ab676SJeff Bonwick 846e14bb325SJeff Bonwick zio = zio_create(pio, pio->io_spa, pio->io_txg, bp, data, size, 847b24ab676SJeff Bonwick done, private, type, priority, flags, vd, offset, &pio->io_bookmark, 848b24ab676SJeff Bonwick ZIO_STAGE_VDEV_IO_START >> 1, pipeline); 849fa9e4066Sahrens 850e14bb325SJeff Bonwick return (zio); 85132b87932Sek } 85232b87932Sek 853e14bb325SJeff Bonwick zio_t * 854e14bb325SJeff Bonwick zio_vdev_delegated_io(vdev_t *vd, uint64_t offset, void *data, uint64_t size, 855b24ab676SJeff Bonwick int type, int priority, enum zio_flag flags, 856b24ab676SJeff Bonwick zio_done_func_t *done, void *private) 857fa9e4066Sahrens { 858e14bb325SJeff Bonwick zio_t *zio; 859fa9e4066Sahrens 860e14bb325SJeff Bonwick ASSERT(vd->vdev_ops->vdev_op_leaf); 861fa9e4066Sahrens 862e14bb325SJeff Bonwick zio = zio_create(NULL, vd->vdev_spa, 0, NULL, 863e14bb325SJeff Bonwick data, size, done, private, type, priority, 864e14bb325SJeff Bonwick flags | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY, 865e14bb325SJeff Bonwick vd, offset, NULL, 866b24ab676SJeff Bonwick ZIO_STAGE_VDEV_IO_START >> 1, ZIO_VDEV_CHILD_PIPELINE); 867fa9e4066Sahrens 868e14bb325SJeff Bonwick return (zio); 869e05725b1Sbonwick } 870e05725b1Sbonwick 871e05725b1Sbonwick void 872e14bb325SJeff Bonwick zio_flush(zio_t *zio, vdev_t *vd) 873e05725b1Sbonwick { 874e14bb325SJeff Bonwick zio_nowait(zio_ioctl(zio, zio->io_spa, vd, DKIOCFLUSHWRITECACHE, 875e14bb325SJeff Bonwick NULL, NULL, ZIO_PRIORITY_NOW, 876e14bb325SJeff Bonwick ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY)); 877fa9e4066Sahrens } 878fa9e4066Sahrens 8796e1f5caaSNeil Perrin void 8806e1f5caaSNeil Perrin zio_shrink(zio_t *zio, uint64_t size) 8816e1f5caaSNeil Perrin { 8826e1f5caaSNeil Perrin ASSERT(zio->io_executor == NULL); 8836e1f5caaSNeil Perrin ASSERT(zio->io_orig_size == zio->io_size); 8846e1f5caaSNeil Perrin ASSERT(size <= zio->io_size); 8856e1f5caaSNeil Perrin 8866e1f5caaSNeil Perrin /* 8876e1f5caaSNeil Perrin * We don't shrink for raidz because of problems with the 8886e1f5caaSNeil Perrin * reconstruction when reading back less than the block size. 8896e1f5caaSNeil Perrin * Note, BP_IS_RAIDZ() assumes no compression. 8906e1f5caaSNeil Perrin */ 8916e1f5caaSNeil Perrin ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF); 8926e1f5caaSNeil Perrin if (!BP_IS_RAIDZ(zio->io_bp)) 8936e1f5caaSNeil Perrin zio->io_orig_size = zio->io_size = size; 8946e1f5caaSNeil Perrin } 8956e1f5caaSNeil Perrin 896fa9e4066Sahrens /* 897fa9e4066Sahrens * ========================================================================== 898e14bb325SJeff Bonwick * Prepare to read and write logical blocks 899fa9e4066Sahrens * ========================================================================== 900fa9e4066Sahrens */ 901e14bb325SJeff Bonwick 902e05725b1Sbonwick static int 903e14bb325SJeff Bonwick zio_read_bp_init(zio_t *zio) 904fa9e4066Sahrens { 905e14bb325SJeff Bonwick blkptr_t *bp = zio->io_bp; 906e05725b1Sbonwick 90703361682SJeff Bonwick if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_OFF && 908f5383399SBill Moore zio->io_child_type == ZIO_CHILD_LOGICAL && 909f5383399SBill Moore !(zio->io_flags & ZIO_FLAG_RAW)) { 910b24ab676SJeff Bonwick uint64_t psize = BP_GET_PSIZE(bp); 911b24ab676SJeff Bonwick void *cbuf = zio_buf_alloc(psize); 912e05725b1Sbonwick 913b24ab676SJeff Bonwick zio_push_transform(zio, cbuf, psize, psize, zio_decompress); 914e14bb325SJeff Bonwick } 915fa9e4066Sahrens 916ad135b5dSChristopher Siden if (!DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) && BP_GET_LEVEL(bp) == 0) 917e14bb325SJeff Bonwick zio->io_flags |= ZIO_FLAG_DONT_CACHE; 918fa9e4066Sahrens 919bbfd46c4SJeff Bonwick if (BP_GET_TYPE(bp) == DMU_OT_DDT_ZAP) 920bbfd46c4SJeff Bonwick zio->io_flags |= ZIO_FLAG_DONT_CACHE; 921bbfd46c4SJeff Bonwick 922b24ab676SJeff Bonwick if (BP_GET_DEDUP(bp) && zio->io_child_type == ZIO_CHILD_LOGICAL) 923b24ab676SJeff Bonwick zio->io_pipeline = ZIO_DDT_READ_PIPELINE; 924b24ab676SJeff Bonwick 925e14bb325SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 926fa9e4066Sahrens } 927fa9e4066Sahrens 928e05725b1Sbonwick static int 929e14bb325SJeff Bonwick zio_write_bp_init(zio_t *zio) 9300a4e9518Sgw { 931b24ab676SJeff Bonwick spa_t *spa = zio->io_spa; 932e14bb325SJeff Bonwick zio_prop_t *zp = &zio->io_prop; 933b24ab676SJeff Bonwick enum zio_compress compress = zp->zp_compress; 934e05725b1Sbonwick blkptr_t *bp = zio->io_bp; 935e14bb325SJeff Bonwick uint64_t lsize = zio->io_size; 936b24ab676SJeff Bonwick uint64_t psize = lsize; 937e14bb325SJeff Bonwick int pass = 1; 938e05725b1Sbonwick 939e14bb325SJeff Bonwick /* 940e14bb325SJeff Bonwick * If our children haven't all reached the ready stage, 941e14bb325SJeff Bonwick * wait for them and then repeat this pipeline stage. 942e14bb325SJeff Bonwick */ 943e14bb325SJeff Bonwick if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY) || 944e14bb325SJeff Bonwick zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_READY)) 945e14bb325SJeff Bonwick return (ZIO_PIPELINE_STOP); 9460a4e9518Sgw 947e14bb325SJeff Bonwick if (!IO_IS_ALLOCATING(zio)) 948e14bb325SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 9490a4e9518Sgw 950b24ab676SJeff Bonwick ASSERT(zio->io_child_type != ZIO_CHILD_DDT); 951b24ab676SJeff Bonwick 952b24ab676SJeff Bonwick if (zio->io_bp_override) { 953b24ab676SJeff Bonwick ASSERT(bp->blk_birth != zio->io_txg); 954b24ab676SJeff Bonwick ASSERT(BP_GET_DEDUP(zio->io_bp_override) == 0); 955b24ab676SJeff Bonwick 956b24ab676SJeff Bonwick *bp = *zio->io_bp_override; 957b24ab676SJeff Bonwick zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; 958b24ab676SJeff Bonwick 959b24ab676SJeff Bonwick if (BP_IS_HOLE(bp) || !zp->zp_dedup) 960b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 961b24ab676SJeff Bonwick 962b24ab676SJeff Bonwick ASSERT(zio_checksum_table[zp->zp_checksum].ci_dedup || 963b24ab676SJeff Bonwick zp->zp_dedup_verify); 964b24ab676SJeff Bonwick 965b24ab676SJeff Bonwick if (BP_GET_CHECKSUM(bp) == zp->zp_checksum) { 966b24ab676SJeff Bonwick BP_SET_DEDUP(bp, 1); 967b24ab676SJeff Bonwick zio->io_pipeline |= ZIO_STAGE_DDT_WRITE; 968b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 969b24ab676SJeff Bonwick } 970b24ab676SJeff Bonwick zio->io_bp_override = NULL; 971b24ab676SJeff Bonwick BP_ZERO(bp); 972b24ab676SJeff Bonwick } 9730a4e9518Sgw 974e14bb325SJeff Bonwick if (bp->blk_birth == zio->io_txg) { 975e14bb325SJeff Bonwick /* 976e14bb325SJeff Bonwick * We're rewriting an existing block, which means we're 977e14bb325SJeff Bonwick * working on behalf of spa_sync(). For spa_sync() to 978e14bb325SJeff Bonwick * converge, it must eventually be the case that we don't 979e14bb325SJeff Bonwick * have to allocate new blocks. But compression changes 980e14bb325SJeff Bonwick * the blocksize, which forces a reallocate, and makes 981e14bb325SJeff Bonwick * convergence take longer. Therefore, after the first 982e14bb325SJeff Bonwick * few passes, stop compressing to ensure convergence. 983e14bb325SJeff Bonwick */ 984b24ab676SJeff Bonwick pass = spa_sync_pass(spa); 985b24ab676SJeff Bonwick 986b24ab676SJeff Bonwick ASSERT(zio->io_txg == spa_syncing_txg(spa)); 987b24ab676SJeff Bonwick ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 988b24ab676SJeff Bonwick ASSERT(!BP_GET_DEDUP(bp)); 989e05725b1Sbonwick 990e14bb325SJeff Bonwick if (pass > SYNC_PASS_DONT_COMPRESS) 991e14bb325SJeff Bonwick compress = ZIO_COMPRESS_OFF; 992e05725b1Sbonwick 993e14bb325SJeff Bonwick /* Make sure someone doesn't change their mind on overwrites */ 994b24ab676SJeff Bonwick ASSERT(MIN(zp->zp_copies + BP_IS_GANG(bp), 995b24ab676SJeff Bonwick spa_max_replication(spa)) == BP_GET_NDVAS(bp)); 996e14bb325SJeff Bonwick } 997fa9e4066Sahrens 998e14bb325SJeff Bonwick if (compress != ZIO_COMPRESS_OFF) { 999b24ab676SJeff Bonwick void *cbuf = zio_buf_alloc(lsize); 1000b24ab676SJeff Bonwick psize = zio_compress_data(compress, zio->io_data, cbuf, lsize); 1001b24ab676SJeff Bonwick if (psize == 0 || psize == lsize) { 1002e14bb325SJeff Bonwick compress = ZIO_COMPRESS_OFF; 1003b24ab676SJeff Bonwick zio_buf_free(cbuf, lsize); 1004b24ab676SJeff Bonwick } else { 1005b24ab676SJeff Bonwick ASSERT(psize < lsize); 1006b24ab676SJeff Bonwick zio_push_transform(zio, cbuf, psize, lsize, NULL); 1007e14bb325SJeff Bonwick } 1008e14bb325SJeff Bonwick } 1009c717a561Smaybee 1010e14bb325SJeff Bonwick /* 1011e14bb325SJeff Bonwick * The final pass of spa_sync() must be all rewrites, but the first 1012e14bb325SJeff Bonwick * few passes offer a trade-off: allocating blocks defers convergence, 1013e14bb325SJeff Bonwick * but newly allocated blocks are sequential, so they can be written 1014e14bb325SJeff Bonwick * to disk faster. Therefore, we allow the first few passes of 1015e14bb325SJeff Bonwick * spa_sync() to allocate new blocks, but force rewrites after that. 1016e14bb325SJeff Bonwick * There should only be a handful of blocks after pass 1 in any case. 1017e14bb325SJeff Bonwick */ 1018b24ab676SJeff Bonwick if (bp->blk_birth == zio->io_txg && BP_GET_PSIZE(bp) == psize && 1019e14bb325SJeff Bonwick pass > SYNC_PASS_REWRITE) { 1020b24ab676SJeff Bonwick ASSERT(psize != 0); 1021b24ab676SJeff Bonwick enum zio_stage gang_stages = zio->io_pipeline & ZIO_GANG_STAGES; 1022e14bb325SJeff Bonwick zio->io_pipeline = ZIO_REWRITE_PIPELINE | gang_stages; 1023e14bb325SJeff Bonwick zio->io_flags |= ZIO_FLAG_IO_REWRITE; 1024e14bb325SJeff Bonwick } else { 1025e14bb325SJeff Bonwick BP_ZERO(bp); 1026e14bb325SJeff Bonwick zio->io_pipeline = ZIO_WRITE_PIPELINE; 1027e14bb325SJeff Bonwick } 1028fa9e4066Sahrens 1029b24ab676SJeff Bonwick if (psize == 0) { 1030e14bb325SJeff Bonwick zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; 1031e14bb325SJeff Bonwick } else { 1032e14bb325SJeff Bonwick ASSERT(zp->zp_checksum != ZIO_CHECKSUM_GANG_HEADER); 1033e14bb325SJeff Bonwick BP_SET_LSIZE(bp, lsize); 1034b24ab676SJeff Bonwick BP_SET_PSIZE(bp, psize); 1035e14bb325SJeff Bonwick BP_SET_COMPRESS(bp, compress); 1036e14bb325SJeff Bonwick BP_SET_CHECKSUM(bp, zp->zp_checksum); 1037e14bb325SJeff Bonwick BP_SET_TYPE(bp, zp->zp_type); 1038e14bb325SJeff Bonwick BP_SET_LEVEL(bp, zp->zp_level); 1039b24ab676SJeff Bonwick BP_SET_DEDUP(bp, zp->zp_dedup); 1040e14bb325SJeff Bonwick BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER); 1041b24ab676SJeff Bonwick if (zp->zp_dedup) { 1042b24ab676SJeff Bonwick ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 1043b24ab676SJeff Bonwick ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE)); 1044b24ab676SJeff Bonwick zio->io_pipeline = ZIO_DDT_WRITE_PIPELINE; 1045b24ab676SJeff Bonwick } 1046b24ab676SJeff Bonwick } 1047b24ab676SJeff Bonwick 1048b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 1049b24ab676SJeff Bonwick } 1050b24ab676SJeff Bonwick 1051b24ab676SJeff Bonwick static int 1052b24ab676SJeff Bonwick zio_free_bp_init(zio_t *zio) 1053b24ab676SJeff Bonwick { 1054b24ab676SJeff Bonwick blkptr_t *bp = zio->io_bp; 1055b24ab676SJeff Bonwick 1056b24ab676SJeff Bonwick if (zio->io_child_type == ZIO_CHILD_LOGICAL) { 1057b24ab676SJeff Bonwick if (BP_GET_DEDUP(bp)) 1058b24ab676SJeff Bonwick zio->io_pipeline = ZIO_DDT_FREE_PIPELINE; 1059e14bb325SJeff Bonwick } 1060fa9e4066Sahrens 1061e05725b1Sbonwick return (ZIO_PIPELINE_CONTINUE); 1062fa9e4066Sahrens } 1063fa9e4066Sahrens 1064e14bb325SJeff Bonwick /* 1065e14bb325SJeff Bonwick * ========================================================================== 1066e14bb325SJeff Bonwick * Execute the I/O pipeline 1067e14bb325SJeff Bonwick * ========================================================================== 1068e14bb325SJeff Bonwick */ 1069e14bb325SJeff Bonwick 1070e14bb325SJeff Bonwick static void 107135a5a358SJonathan Adams zio_taskq_dispatch(zio_t *zio, enum zio_taskq_type q, boolean_t cutinline) 1072fa9e4066Sahrens { 107380eb36f2SGeorge Wilson spa_t *spa = zio->io_spa; 1074e14bb325SJeff Bonwick zio_type_t t = zio->io_type; 10755aeb9474SGarrett D'Amore int flags = (cutinline ? TQ_FRONT : 0); 10760a4e9518Sgw 10770a4e9518Sgw /* 1078bbe36defSGeorge Wilson * If we're a config writer or a probe, the normal issue and 1079bbe36defSGeorge Wilson * interrupt threads may all be blocked waiting for the config lock. 1080bbe36defSGeorge Wilson * In this case, select the otherwise-unused taskq for ZIO_TYPE_NULL. 10810a4e9518Sgw */ 1082bbe36defSGeorge Wilson if (zio->io_flags & (ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_PROBE)) 1083e14bb325SJeff Bonwick t = ZIO_TYPE_NULL; 10840a4e9518Sgw 10850a4e9518Sgw /* 1086e14bb325SJeff Bonwick * A similar issue exists for the L2ARC write thread until L2ARC 2.0. 10870a4e9518Sgw */ 1088e14bb325SJeff Bonwick if (t == ZIO_TYPE_WRITE && zio->io_vd && zio->io_vd->vdev_aux) 1089e14bb325SJeff Bonwick t = ZIO_TYPE_NULL; 10900a4e9518Sgw 109180eb36f2SGeorge Wilson /* 109280eb36f2SGeorge Wilson * If this is a high priority I/O, then use the high priority taskq. 109380eb36f2SGeorge Wilson */ 109480eb36f2SGeorge Wilson if (zio->io_priority == ZIO_PRIORITY_NOW && 109580eb36f2SGeorge Wilson spa->spa_zio_taskq[t][q + 1] != NULL) 109680eb36f2SGeorge Wilson q++; 109780eb36f2SGeorge Wilson 109880eb36f2SGeorge Wilson ASSERT3U(q, <, ZIO_TASKQ_TYPES); 10995aeb9474SGarrett D'Amore 11005aeb9474SGarrett D'Amore /* 11015aeb9474SGarrett D'Amore * NB: We are assuming that the zio can only be dispatched 11025aeb9474SGarrett D'Amore * to a single taskq at a time. It would be a grievous error 11035aeb9474SGarrett D'Amore * to dispatch the zio to another taskq at the same time. 11045aeb9474SGarrett D'Amore */ 11055aeb9474SGarrett D'Amore ASSERT(zio->io_tqent.tqent_next == NULL); 11065aeb9474SGarrett D'Amore taskq_dispatch_ent(spa->spa_zio_taskq[t][q], 11075aeb9474SGarrett D'Amore (task_func_t *)zio_execute, zio, flags, &zio->io_tqent); 1108e14bb325SJeff Bonwick } 11090a4e9518Sgw 1110e14bb325SJeff Bonwick static boolean_t 1111e14bb325SJeff Bonwick zio_taskq_member(zio_t *zio, enum zio_taskq_type q) 1112e14bb325SJeff Bonwick { 1113e14bb325SJeff Bonwick kthread_t *executor = zio->io_executor; 1114e14bb325SJeff Bonwick spa_t *spa = zio->io_spa; 11150a4e9518Sgw 1116e14bb325SJeff Bonwick for (zio_type_t t = 0; t < ZIO_TYPES; t++) 1117e14bb325SJeff Bonwick if (taskq_member(spa->spa_zio_taskq[t][q], executor)) 1118e14bb325SJeff Bonwick return (B_TRUE); 11190a4e9518Sgw 1120e14bb325SJeff Bonwick return (B_FALSE); 1121e14bb325SJeff Bonwick } 1122e05725b1Sbonwick 1123e14bb325SJeff Bonwick static int 1124e14bb325SJeff Bonwick zio_issue_async(zio_t *zio) 1125e14bb325SJeff Bonwick { 112635a5a358SJonathan Adams zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE); 1127e14bb325SJeff Bonwick 1128e14bb325SJeff Bonwick return (ZIO_PIPELINE_STOP); 11290a4e9518Sgw } 11300a4e9518Sgw 1131e14bb325SJeff Bonwick void 1132e14bb325SJeff Bonwick zio_interrupt(zio_t *zio) 11330a4e9518Sgw { 113435a5a358SJonathan Adams zio_taskq_dispatch(zio, ZIO_TASKQ_INTERRUPT, B_FALSE); 1135e14bb325SJeff Bonwick } 11360a4e9518Sgw 1137e14bb325SJeff Bonwick /* 1138e14bb325SJeff Bonwick * Execute the I/O pipeline until one of the following occurs: 1139e14bb325SJeff Bonwick * (1) the I/O completes; (2) the pipeline stalls waiting for 1140e14bb325SJeff Bonwick * dependent child I/Os; (3) the I/O issues, so we're waiting 1141e14bb325SJeff Bonwick * for an I/O completion interrupt; (4) the I/O is delegated by 1142e14bb325SJeff Bonwick * vdev-level caching or aggregation; (5) the I/O is deferred 1143e14bb325SJeff Bonwick * due to vdev-level queueing; (6) the I/O is handed off to 1144e14bb325SJeff Bonwick * another thread. In all cases, the pipeline stops whenever 1145e14bb325SJeff Bonwick * there's no CPU work; it never burns a thread in cv_wait(). 1146e14bb325SJeff Bonwick * 1147e14bb325SJeff Bonwick * There's no locking on io_stage because there's no legitimate way 1148e14bb325SJeff Bonwick * for multiple threads to be attempting to process the same I/O. 1149e14bb325SJeff Bonwick */ 1150b24ab676SJeff Bonwick static zio_pipe_stage_t *zio_pipeline[]; 11510a4e9518Sgw 1152e14bb325SJeff Bonwick void 1153e14bb325SJeff Bonwick zio_execute(zio_t *zio) 1154e14bb325SJeff Bonwick { 1155e14bb325SJeff Bonwick zio->io_executor = curthread; 11560a4e9518Sgw 1157e14bb325SJeff Bonwick while (zio->io_stage < ZIO_STAGE_DONE) { 1158b24ab676SJeff Bonwick enum zio_stage pipeline = zio->io_pipeline; 1159b24ab676SJeff Bonwick enum zio_stage stage = zio->io_stage; 1160e14bb325SJeff Bonwick int rv; 11610a4e9518Sgw 1162e14bb325SJeff Bonwick ASSERT(!MUTEX_HELD(&zio->io_lock)); 1163b24ab676SJeff Bonwick ASSERT(ISP2(stage)); 1164b24ab676SJeff Bonwick ASSERT(zio->io_stall == NULL); 11650a4e9518Sgw 1166b24ab676SJeff Bonwick do { 1167b24ab676SJeff Bonwick stage <<= 1; 1168b24ab676SJeff Bonwick } while ((stage & pipeline) == 0); 1169e14bb325SJeff Bonwick 1170e14bb325SJeff Bonwick ASSERT(stage <= ZIO_STAGE_DONE); 11710a4e9518Sgw 11720a4e9518Sgw /* 1173e14bb325SJeff Bonwick * If we are in interrupt context and this pipeline stage 1174e14bb325SJeff Bonwick * will grab a config lock that is held across I/O, 1175b24ab676SJeff Bonwick * or may wait for an I/O that needs an interrupt thread 1176b24ab676SJeff Bonwick * to complete, issue async to avoid deadlock. 117735a5a358SJonathan Adams * 117835a5a358SJonathan Adams * For VDEV_IO_START, we cut in line so that the io will 117935a5a358SJonathan Adams * be sent to disk promptly. 11800a4e9518Sgw */ 1181b24ab676SJeff Bonwick if ((stage & ZIO_BLOCKING_STAGES) && zio->io_vd == NULL && 1182e14bb325SJeff Bonwick zio_taskq_member(zio, ZIO_TASKQ_INTERRUPT)) { 118335a5a358SJonathan Adams boolean_t cut = (stage == ZIO_STAGE_VDEV_IO_START) ? 118435a5a358SJonathan Adams zio_requeue_io_start_cut_in_line : B_FALSE; 118535a5a358SJonathan Adams zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, cut); 1186e14bb325SJeff Bonwick return; 11870a4e9518Sgw } 11880a4e9518Sgw 1189e14bb325SJeff Bonwick zio->io_stage = stage; 1190b24ab676SJeff Bonwick rv = zio_pipeline[highbit(stage) - 1](zio); 11910a4e9518Sgw 1192e14bb325SJeff Bonwick if (rv == ZIO_PIPELINE_STOP) 1193e14bb325SJeff Bonwick return; 11940a4e9518Sgw 1195e14bb325SJeff Bonwick ASSERT(rv == ZIO_PIPELINE_CONTINUE); 1196e14bb325SJeff Bonwick } 11970a4e9518Sgw } 11980a4e9518Sgw 1199e14bb325SJeff Bonwick /* 1200e14bb325SJeff Bonwick * ========================================================================== 1201e14bb325SJeff Bonwick * Initiate I/O, either sync or async 1202e14bb325SJeff Bonwick * ========================================================================== 1203e14bb325SJeff Bonwick */ 1204e14bb325SJeff Bonwick int 1205e14bb325SJeff Bonwick zio_wait(zio_t *zio) 12060a4e9518Sgw { 1207e14bb325SJeff Bonwick int error; 12080a4e9518Sgw 1209e14bb325SJeff Bonwick ASSERT(zio->io_stage == ZIO_STAGE_OPEN); 1210e14bb325SJeff Bonwick ASSERT(zio->io_executor == NULL); 12110a4e9518Sgw 1212e14bb325SJeff Bonwick zio->io_waiter = curthread; 1213e05725b1Sbonwick 1214e14bb325SJeff Bonwick zio_execute(zio); 12150a4e9518Sgw 1216e14bb325SJeff Bonwick mutex_enter(&zio->io_lock); 1217e14bb325SJeff Bonwick while (zio->io_executor != NULL) 1218e14bb325SJeff Bonwick cv_wait(&zio->io_cv, &zio->io_lock); 1219e14bb325SJeff Bonwick mutex_exit(&zio->io_lock); 122032b87932Sek 1221e14bb325SJeff Bonwick error = zio->io_error; 1222e14bb325SJeff Bonwick zio_destroy(zio); 122332b87932Sek 1224e14bb325SJeff Bonwick return (error); 122532b87932Sek } 122632b87932Sek 1227e14bb325SJeff Bonwick void 1228e14bb325SJeff Bonwick zio_nowait(zio_t *zio) 12290a4e9518Sgw { 1230e14bb325SJeff Bonwick ASSERT(zio->io_executor == NULL); 1231fa9e4066Sahrens 1232a3f829aeSBill Moore if (zio->io_child_type == ZIO_CHILD_LOGICAL && 1233a3f829aeSBill Moore zio_unique_parent(zio) == NULL) { 1234ea8dc4b6Seschrock /* 1235e14bb325SJeff Bonwick * This is a logical async I/O with no parent to wait for it. 123654d692b7SGeorge Wilson * We add it to the spa_async_root_zio "Godfather" I/O which 123754d692b7SGeorge Wilson * will ensure they complete prior to unloading the pool. 1238ea8dc4b6Seschrock */ 1239e14bb325SJeff Bonwick spa_t *spa = zio->io_spa; 124054d692b7SGeorge Wilson 124154d692b7SGeorge Wilson zio_add_child(spa->spa_async_zio_root, zio); 1242e14bb325SJeff Bonwick } 1243ea8dc4b6Seschrock 1244e14bb325SJeff Bonwick zio_execute(zio); 1245e14bb325SJeff Bonwick } 1246ea8dc4b6Seschrock 1247e14bb325SJeff Bonwick /* 1248e14bb325SJeff Bonwick * ========================================================================== 1249e14bb325SJeff Bonwick * Reexecute or suspend/resume failed I/O 1250e14bb325SJeff Bonwick * ========================================================================== 1251e14bb325SJeff Bonwick */ 1252fa9e4066Sahrens 1253e14bb325SJeff Bonwick static void 1254e14bb325SJeff Bonwick zio_reexecute(zio_t *pio) 1255e14bb325SJeff Bonwick { 1256a3f829aeSBill Moore zio_t *cio, *cio_next; 1257a3f829aeSBill Moore 1258a3f829aeSBill Moore ASSERT(pio->io_child_type == ZIO_CHILD_LOGICAL); 1259a3f829aeSBill Moore ASSERT(pio->io_orig_stage == ZIO_STAGE_OPEN); 1260f5383399SBill Moore ASSERT(pio->io_gang_leader == NULL); 1261f5383399SBill Moore ASSERT(pio->io_gang_tree == NULL); 1262e05725b1Sbonwick 1263e14bb325SJeff Bonwick pio->io_flags = pio->io_orig_flags; 1264e14bb325SJeff Bonwick pio->io_stage = pio->io_orig_stage; 1265e14bb325SJeff Bonwick pio->io_pipeline = pio->io_orig_pipeline; 1266e14bb325SJeff Bonwick pio->io_reexecute = 0; 1267e14bb325SJeff Bonwick pio->io_error = 0; 1268a3f829aeSBill Moore for (int w = 0; w < ZIO_WAIT_TYPES; w++) 1269a3f829aeSBill Moore pio->io_state[w] = 0; 1270e14bb325SJeff Bonwick for (int c = 0; c < ZIO_CHILD_TYPES; c++) 1271e14bb325SJeff Bonwick pio->io_child_error[c] = 0; 12720a4e9518Sgw 1273b24ab676SJeff Bonwick if (IO_IS_ALLOCATING(pio)) 1274b24ab676SJeff Bonwick BP_ZERO(pio->io_bp); 1275d58459f4Sek 1276e14bb325SJeff Bonwick /* 1277e14bb325SJeff Bonwick * As we reexecute pio's children, new children could be created. 1278a3f829aeSBill Moore * New children go to the head of pio's io_child_list, however, 1279e14bb325SJeff Bonwick * so we will (correctly) not reexecute them. The key is that 1280a3f829aeSBill Moore * the remainder of pio's io_child_list, from 'cio_next' onward, 1281a3f829aeSBill Moore * cannot be affected by any side effects of reexecuting 'cio'. 1282e14bb325SJeff Bonwick */ 1283a3f829aeSBill Moore for (cio = zio_walk_children(pio); cio != NULL; cio = cio_next) { 1284a3f829aeSBill Moore cio_next = zio_walk_children(pio); 1285e14bb325SJeff Bonwick mutex_enter(&pio->io_lock); 1286a3f829aeSBill Moore for (int w = 0; w < ZIO_WAIT_TYPES; w++) 1287a3f829aeSBill Moore pio->io_children[cio->io_child_type][w]++; 1288e14bb325SJeff Bonwick mutex_exit(&pio->io_lock); 1289a3f829aeSBill Moore zio_reexecute(cio); 1290fa9e4066Sahrens } 1291e05725b1Sbonwick 1292e14bb325SJeff Bonwick /* 1293e14bb325SJeff Bonwick * Now that all children have been reexecuted, execute the parent. 129454d692b7SGeorge Wilson * We don't reexecute "The Godfather" I/O here as it's the 129554d692b7SGeorge Wilson * responsibility of the caller to wait on him. 1296e14bb325SJeff Bonwick */ 129754d692b7SGeorge Wilson if (!(pio->io_flags & ZIO_FLAG_GODFATHER)) 129854d692b7SGeorge Wilson zio_execute(pio); 12990a4e9518Sgw } 13000a4e9518Sgw 1301e14bb325SJeff Bonwick void 1302e14bb325SJeff Bonwick zio_suspend(spa_t *spa, zio_t *zio) 13030a4e9518Sgw { 1304e14bb325SJeff Bonwick if (spa_get_failmode(spa) == ZIO_FAILURE_MODE_PANIC) 1305e14bb325SJeff Bonwick fm_panic("Pool '%s' has encountered an uncorrectable I/O " 1306e14bb325SJeff Bonwick "failure and the failure mode property for this pool " 1307e14bb325SJeff Bonwick "is set to panic.", spa_name(spa)); 13080a4e9518Sgw 1309e14bb325SJeff Bonwick zfs_ereport_post(FM_EREPORT_ZFS_IO_FAILURE, spa, NULL, NULL, 0, 0); 13100a4e9518Sgw 1311e14bb325SJeff Bonwick mutex_enter(&spa->spa_suspend_lock); 1312fa9e4066Sahrens 1313e14bb325SJeff Bonwick if (spa->spa_suspend_zio_root == NULL) 131454d692b7SGeorge Wilson spa->spa_suspend_zio_root = zio_root(spa, NULL, NULL, 131554d692b7SGeorge Wilson ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 131654d692b7SGeorge Wilson ZIO_FLAG_GODFATHER); 1317fa9e4066Sahrens 1318e14bb325SJeff Bonwick spa->spa_suspended = B_TRUE; 1319fa9e4066Sahrens 1320e14bb325SJeff Bonwick if (zio != NULL) { 132154d692b7SGeorge Wilson ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER)); 1322e14bb325SJeff Bonwick ASSERT(zio != spa->spa_suspend_zio_root); 1323e14bb325SJeff Bonwick ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 1324a3f829aeSBill Moore ASSERT(zio_unique_parent(zio) == NULL); 1325e14bb325SJeff Bonwick ASSERT(zio->io_stage == ZIO_STAGE_DONE); 1326e14bb325SJeff Bonwick zio_add_child(spa->spa_suspend_zio_root, zio); 1327e14bb325SJeff Bonwick } 1328fa9e4066Sahrens 1329e14bb325SJeff Bonwick mutex_exit(&spa->spa_suspend_lock); 1330e14bb325SJeff Bonwick } 1331fa9e4066Sahrens 133254d692b7SGeorge Wilson int 1333e14bb325SJeff Bonwick zio_resume(spa_t *spa) 1334e14bb325SJeff Bonwick { 133554d692b7SGeorge Wilson zio_t *pio; 1336fa9e4066Sahrens 1337b3995adbSahrens /* 1338e14bb325SJeff Bonwick * Reexecute all previously suspended i/o. 1339b3995adbSahrens */ 1340e14bb325SJeff Bonwick mutex_enter(&spa->spa_suspend_lock); 1341e14bb325SJeff Bonwick spa->spa_suspended = B_FALSE; 1342e14bb325SJeff Bonwick cv_broadcast(&spa->spa_suspend_cv); 1343e14bb325SJeff Bonwick pio = spa->spa_suspend_zio_root; 1344e14bb325SJeff Bonwick spa->spa_suspend_zio_root = NULL; 1345e14bb325SJeff Bonwick mutex_exit(&spa->spa_suspend_lock); 1346e14bb325SJeff Bonwick 1347e14bb325SJeff Bonwick if (pio == NULL) 134854d692b7SGeorge Wilson return (0); 1349e14bb325SJeff Bonwick 135054d692b7SGeorge Wilson zio_reexecute(pio); 135154d692b7SGeorge Wilson return (zio_wait(pio)); 1352e14bb325SJeff Bonwick } 1353e14bb325SJeff Bonwick 1354e14bb325SJeff Bonwick void 1355e14bb325SJeff Bonwick zio_resume_wait(spa_t *spa) 1356e14bb325SJeff Bonwick { 1357e14bb325SJeff Bonwick mutex_enter(&spa->spa_suspend_lock); 1358e14bb325SJeff Bonwick while (spa_suspended(spa)) 1359e14bb325SJeff Bonwick cv_wait(&spa->spa_suspend_cv, &spa->spa_suspend_lock); 1360e14bb325SJeff Bonwick mutex_exit(&spa->spa_suspend_lock); 1361fa9e4066Sahrens } 1362fa9e4066Sahrens 1363fa9e4066Sahrens /* 1364fa9e4066Sahrens * ========================================================================== 1365e14bb325SJeff Bonwick * Gang blocks. 1366e14bb325SJeff Bonwick * 1367e14bb325SJeff Bonwick * A gang block is a collection of small blocks that looks to the DMU 1368e14bb325SJeff Bonwick * like one large block. When zio_dva_allocate() cannot find a block 1369e14bb325SJeff Bonwick * of the requested size, due to either severe fragmentation or the pool 1370e14bb325SJeff Bonwick * being nearly full, it calls zio_write_gang_block() to construct the 1371e14bb325SJeff Bonwick * block from smaller fragments. 1372e14bb325SJeff Bonwick * 1373e14bb325SJeff Bonwick * A gang block consists of a gang header (zio_gbh_phys_t) and up to 1374e14bb325SJeff Bonwick * three (SPA_GBH_NBLKPTRS) gang members. The gang header is just like 1375e14bb325SJeff Bonwick * an indirect block: it's an array of block pointers. It consumes 1376e14bb325SJeff Bonwick * only one sector and hence is allocatable regardless of fragmentation. 1377e14bb325SJeff Bonwick * The gang header's bps point to its gang members, which hold the data. 1378e14bb325SJeff Bonwick * 1379e14bb325SJeff Bonwick * Gang blocks are self-checksumming, using the bp's <vdev, offset, txg> 1380e14bb325SJeff Bonwick * as the verifier to ensure uniqueness of the SHA256 checksum. 1381e14bb325SJeff Bonwick * Critically, the gang block bp's blk_cksum is the checksum of the data, 1382e14bb325SJeff Bonwick * not the gang header. This ensures that data block signatures (needed for 1383e14bb325SJeff Bonwick * deduplication) are independent of how the block is physically stored. 1384e14bb325SJeff Bonwick * 1385e14bb325SJeff Bonwick * Gang blocks can be nested: a gang member may itself be a gang block. 1386e14bb325SJeff Bonwick * Thus every gang block is a tree in which root and all interior nodes are 1387e14bb325SJeff Bonwick * gang headers, and the leaves are normal blocks that contain user data. 1388e14bb325SJeff Bonwick * The root of the gang tree is called the gang leader. 1389e14bb325SJeff Bonwick * 1390e14bb325SJeff Bonwick * To perform any operation (read, rewrite, free, claim) on a gang block, 1391e14bb325SJeff Bonwick * zio_gang_assemble() first assembles the gang tree (minus data leaves) 1392e14bb325SJeff Bonwick * in the io_gang_tree field of the original logical i/o by recursively 1393e14bb325SJeff Bonwick * reading the gang leader and all gang headers below it. This yields 1394e14bb325SJeff Bonwick * an in-core tree containing the contents of every gang header and the 1395e14bb325SJeff Bonwick * bps for every constituent of the gang block. 1396e14bb325SJeff Bonwick * 1397e14bb325SJeff Bonwick * With the gang tree now assembled, zio_gang_issue() just walks the gang tree 1398e14bb325SJeff Bonwick * and invokes a callback on each bp. To free a gang block, zio_gang_issue() 1399e14bb325SJeff Bonwick * calls zio_free_gang() -- a trivial wrapper around zio_free() -- for each bp. 1400e14bb325SJeff Bonwick * zio_claim_gang() provides a similarly trivial wrapper for zio_claim(). 1401e14bb325SJeff Bonwick * zio_read_gang() is a wrapper around zio_read() that omits reading gang 1402e14bb325SJeff Bonwick * headers, since we already have those in io_gang_tree. zio_rewrite_gang() 1403e14bb325SJeff Bonwick * performs a zio_rewrite() of the data or, for gang headers, a zio_rewrite() 1404e14bb325SJeff Bonwick * of the gang header plus zio_checksum_compute() of the data to update the 1405e14bb325SJeff Bonwick * gang header's blk_cksum as described above. 1406e14bb325SJeff Bonwick * 1407e14bb325SJeff Bonwick * The two-phase assemble/issue model solves the problem of partial failure -- 1408e14bb325SJeff Bonwick * what if you'd freed part of a gang block but then couldn't read the 1409e14bb325SJeff Bonwick * gang header for another part? Assembling the entire gang tree first 1410e14bb325SJeff Bonwick * ensures that all the necessary gang header I/O has succeeded before 1411e14bb325SJeff Bonwick * starting the actual work of free, claim, or write. Once the gang tree 1412e14bb325SJeff Bonwick * is assembled, free and claim are in-memory operations that cannot fail. 1413e14bb325SJeff Bonwick * 1414e14bb325SJeff Bonwick * In the event that a gang write fails, zio_dva_unallocate() walks the 1415e14bb325SJeff Bonwick * gang tree to immediately free (i.e. insert back into the space map) 1416e14bb325SJeff Bonwick * everything we've allocated. This ensures that we don't get ENOSPC 1417e14bb325SJeff Bonwick * errors during repeated suspend/resume cycles due to a flaky device. 1418e14bb325SJeff Bonwick * 1419e14bb325SJeff Bonwick * Gang rewrites only happen during sync-to-convergence. If we can't assemble 1420e14bb325SJeff Bonwick * the gang tree, we won't modify the block, so we can safely defer the free 1421e14bb325SJeff Bonwick * (knowing that the block is still intact). If we *can* assemble the gang 1422e14bb325SJeff Bonwick * tree, then even if some of the rewrites fail, zio_dva_unallocate() will free 1423e14bb325SJeff Bonwick * each constituent bp and we can allocate a new block on the next sync pass. 1424e14bb325SJeff Bonwick * 1425e14bb325SJeff Bonwick * In all cases, the gang tree allows complete recovery from partial failure. 1426fa9e4066Sahrens * ========================================================================== 1427fa9e4066Sahrens */ 1428e14bb325SJeff Bonwick 1429e14bb325SJeff Bonwick static zio_t * 1430e14bb325SJeff Bonwick zio_read_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) 1431fa9e4066Sahrens { 1432e14bb325SJeff Bonwick if (gn != NULL) 1433e14bb325SJeff Bonwick return (pio); 1434fa9e4066Sahrens 1435e14bb325SJeff Bonwick return (zio_read(pio, pio->io_spa, bp, data, BP_GET_PSIZE(bp), 1436e14bb325SJeff Bonwick NULL, NULL, pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), 1437e14bb325SJeff Bonwick &pio->io_bookmark)); 1438e14bb325SJeff Bonwick } 1439e14bb325SJeff Bonwick 1440e14bb325SJeff Bonwick zio_t * 1441e14bb325SJeff Bonwick zio_rewrite_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) 1442e14bb325SJeff Bonwick { 1443e14bb325SJeff Bonwick zio_t *zio; 1444e14bb325SJeff Bonwick 1445e14bb325SJeff Bonwick if (gn != NULL) { 1446e14bb325SJeff Bonwick zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp, 1447e14bb325SJeff Bonwick gn->gn_gbh, SPA_GANGBLOCKSIZE, NULL, NULL, pio->io_priority, 1448e14bb325SJeff Bonwick ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); 1449fa9e4066Sahrens /* 1450e14bb325SJeff Bonwick * As we rewrite each gang header, the pipeline will compute 1451e14bb325SJeff Bonwick * a new gang block header checksum for it; but no one will 1452e14bb325SJeff Bonwick * compute a new data checksum, so we do that here. The one 1453e14bb325SJeff Bonwick * exception is the gang leader: the pipeline already computed 1454e14bb325SJeff Bonwick * its data checksum because that stage precedes gang assembly. 1455e14bb325SJeff Bonwick * (Presently, nothing actually uses interior data checksums; 1456e14bb325SJeff Bonwick * this is just good hygiene.) 1457fa9e4066Sahrens */ 1458f5383399SBill Moore if (gn != pio->io_gang_leader->io_gang_tree) { 1459e14bb325SJeff Bonwick zio_checksum_compute(zio, BP_GET_CHECKSUM(bp), 1460e14bb325SJeff Bonwick data, BP_GET_PSIZE(bp)); 1461e14bb325SJeff Bonwick } 1462b24ab676SJeff Bonwick /* 1463b24ab676SJeff Bonwick * If we are here to damage data for testing purposes, 1464b24ab676SJeff Bonwick * leave the GBH alone so that we can detect the damage. 1465b24ab676SJeff Bonwick */ 1466b24ab676SJeff Bonwick if (pio->io_gang_leader->io_flags & ZIO_FLAG_INDUCE_DAMAGE) 1467b24ab676SJeff Bonwick zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES; 1468fa9e4066Sahrens } else { 1469e14bb325SJeff Bonwick zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp, 1470e14bb325SJeff Bonwick data, BP_GET_PSIZE(bp), NULL, NULL, pio->io_priority, 1471e14bb325SJeff Bonwick ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); 1472fa9e4066Sahrens } 1473fa9e4066Sahrens 1474e14bb325SJeff Bonwick return (zio); 1475e14bb325SJeff Bonwick } 1476fa9e4066Sahrens 1477e14bb325SJeff Bonwick /* ARGSUSED */ 1478e14bb325SJeff Bonwick zio_t * 1479e14bb325SJeff Bonwick zio_free_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) 1480e14bb325SJeff Bonwick { 1481b24ab676SJeff Bonwick return (zio_free_sync(pio, pio->io_spa, pio->io_txg, bp, 1482b24ab676SJeff Bonwick ZIO_GANG_CHILD_FLAGS(pio))); 1483fa9e4066Sahrens } 1484fa9e4066Sahrens 1485e14bb325SJeff Bonwick /* ARGSUSED */ 1486e14bb325SJeff Bonwick zio_t * 1487e14bb325SJeff Bonwick zio_claim_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) 1488fa9e4066Sahrens { 1489e14bb325SJeff Bonwick return (zio_claim(pio, pio->io_spa, pio->io_txg, bp, 1490e14bb325SJeff Bonwick NULL, NULL, ZIO_GANG_CHILD_FLAGS(pio))); 1491e14bb325SJeff Bonwick } 1492fa9e4066Sahrens 1493e14bb325SJeff Bonwick static zio_gang_issue_func_t *zio_gang_issue_func[ZIO_TYPES] = { 1494e14bb325SJeff Bonwick NULL, 1495e14bb325SJeff Bonwick zio_read_gang, 1496e14bb325SJeff Bonwick zio_rewrite_gang, 1497e14bb325SJeff Bonwick zio_free_gang, 1498e14bb325SJeff Bonwick zio_claim_gang, 1499e14bb325SJeff Bonwick NULL 1500e14bb325SJeff Bonwick }; 1501fa9e4066Sahrens 1502e14bb325SJeff Bonwick static void zio_gang_tree_assemble_done(zio_t *zio); 1503fa9e4066Sahrens 1504e14bb325SJeff Bonwick static zio_gang_node_t * 1505e14bb325SJeff Bonwick zio_gang_node_alloc(zio_gang_node_t **gnpp) 1506e14bb325SJeff Bonwick { 1507e14bb325SJeff Bonwick zio_gang_node_t *gn; 1508fa9e4066Sahrens 1509e14bb325SJeff Bonwick ASSERT(*gnpp == NULL); 1510fa9e4066Sahrens 1511e14bb325SJeff Bonwick gn = kmem_zalloc(sizeof (*gn), KM_SLEEP); 1512e14bb325SJeff Bonwick gn->gn_gbh = zio_buf_alloc(SPA_GANGBLOCKSIZE); 1513e14bb325SJeff Bonwick *gnpp = gn; 1514e14bb325SJeff Bonwick 1515e14bb325SJeff Bonwick return (gn); 1516fa9e4066Sahrens } 1517fa9e4066Sahrens 1518fa9e4066Sahrens static void 1519e14bb325SJeff Bonwick zio_gang_node_free(zio_gang_node_t **gnpp) 1520fa9e4066Sahrens { 1521e14bb325SJeff Bonwick zio_gang_node_t *gn = *gnpp; 1522fa9e4066Sahrens 1523e14bb325SJeff Bonwick for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) 1524e14bb325SJeff Bonwick ASSERT(gn->gn_child[g] == NULL); 1525e14bb325SJeff Bonwick 1526e14bb325SJeff Bonwick zio_buf_free(gn->gn_gbh, SPA_GANGBLOCKSIZE); 1527e14bb325SJeff Bonwick kmem_free(gn, sizeof (*gn)); 1528e14bb325SJeff Bonwick *gnpp = NULL; 1529fa9e4066Sahrens } 1530fa9e4066Sahrens 1531e14bb325SJeff Bonwick static void 1532e14bb325SJeff Bonwick zio_gang_tree_free(zio_gang_node_t **gnpp) 1533fa9e4066Sahrens { 1534e14bb325SJeff Bonwick zio_gang_node_t *gn = *gnpp; 1535fa9e4066Sahrens 1536e14bb325SJeff Bonwick if (gn == NULL) 1537e14bb325SJeff Bonwick return; 1538fa9e4066Sahrens 1539e14bb325SJeff Bonwick for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) 1540e14bb325SJeff Bonwick zio_gang_tree_free(&gn->gn_child[g]); 1541fa9e4066Sahrens 1542e14bb325SJeff Bonwick zio_gang_node_free(gnpp); 1543fa9e4066Sahrens } 1544fa9e4066Sahrens 1545e14bb325SJeff Bonwick static void 1546f5383399SBill Moore zio_gang_tree_assemble(zio_t *gio, blkptr_t *bp, zio_gang_node_t **gnpp) 1547fa9e4066Sahrens { 1548e14bb325SJeff Bonwick zio_gang_node_t *gn = zio_gang_node_alloc(gnpp); 1549e14bb325SJeff Bonwick 1550f5383399SBill Moore ASSERT(gio->io_gang_leader == gio); 1551e14bb325SJeff Bonwick ASSERT(BP_IS_GANG(bp)); 1552fa9e4066Sahrens 1553f5383399SBill Moore zio_nowait(zio_read(gio, gio->io_spa, bp, gn->gn_gbh, 1554e14bb325SJeff Bonwick SPA_GANGBLOCKSIZE, zio_gang_tree_assemble_done, gn, 1555f5383399SBill Moore gio->io_priority, ZIO_GANG_CHILD_FLAGS(gio), &gio->io_bookmark)); 1556e14bb325SJeff Bonwick } 1557fa9e4066Sahrens 1558e14bb325SJeff Bonwick static void 1559e14bb325SJeff Bonwick zio_gang_tree_assemble_done(zio_t *zio) 1560e14bb325SJeff Bonwick { 1561f5383399SBill Moore zio_t *gio = zio->io_gang_leader; 1562e14bb325SJeff Bonwick zio_gang_node_t *gn = zio->io_private; 1563e14bb325SJeff Bonwick blkptr_t *bp = zio->io_bp; 1564fa9e4066Sahrens 1565f5383399SBill Moore ASSERT(gio == zio_unique_parent(zio)); 1566b24ab676SJeff Bonwick ASSERT(zio->io_child_count == 0); 1567fa9e4066Sahrens 1568e14bb325SJeff Bonwick if (zio->io_error) 1569e14bb325SJeff Bonwick return; 1570fa9e4066Sahrens 1571e14bb325SJeff Bonwick if (BP_SHOULD_BYTESWAP(bp)) 1572e14bb325SJeff Bonwick byteswap_uint64_array(zio->io_data, zio->io_size); 1573fa9e4066Sahrens 1574e14bb325SJeff Bonwick ASSERT(zio->io_data == gn->gn_gbh); 1575e14bb325SJeff Bonwick ASSERT(zio->io_size == SPA_GANGBLOCKSIZE); 15766e1f5caaSNeil Perrin ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC); 1577e05725b1Sbonwick 1578e14bb325SJeff Bonwick for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) { 1579e14bb325SJeff Bonwick blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g]; 1580e14bb325SJeff Bonwick if (!BP_IS_GANG(gbp)) 1581e14bb325SJeff Bonwick continue; 1582f5383399SBill Moore zio_gang_tree_assemble(gio, gbp, &gn->gn_child[g]); 1583e14bb325SJeff Bonwick } 1584fa9e4066Sahrens } 1585fa9e4066Sahrens 1586e14bb325SJeff Bonwick static void 1587e14bb325SJeff Bonwick zio_gang_tree_issue(zio_t *pio, zio_gang_node_t *gn, blkptr_t *bp, void *data) 1588fa9e4066Sahrens { 1589f5383399SBill Moore zio_t *gio = pio->io_gang_leader; 1590e14bb325SJeff Bonwick zio_t *zio; 1591fa9e4066Sahrens 1592e14bb325SJeff Bonwick ASSERT(BP_IS_GANG(bp) == !!gn); 1593f5383399SBill Moore ASSERT(BP_GET_CHECKSUM(bp) == BP_GET_CHECKSUM(gio->io_bp)); 1594f5383399SBill Moore ASSERT(BP_GET_LSIZE(bp) == BP_GET_PSIZE(bp) || gn == gio->io_gang_tree); 1595fa9e4066Sahrens 1596e14bb325SJeff Bonwick /* 1597e14bb325SJeff Bonwick * If you're a gang header, your data is in gn->gn_gbh. 1598e14bb325SJeff Bonwick * If you're a gang member, your data is in 'data' and gn == NULL. 1599e14bb325SJeff Bonwick */ 1600f5383399SBill Moore zio = zio_gang_issue_func[gio->io_type](pio, bp, gn, data); 1601fa9e4066Sahrens 1602e14bb325SJeff Bonwick if (gn != NULL) { 16036e1f5caaSNeil Perrin ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC); 1604fa9e4066Sahrens 1605e14bb325SJeff Bonwick for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) { 1606e14bb325SJeff Bonwick blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g]; 1607e14bb325SJeff Bonwick if (BP_IS_HOLE(gbp)) 1608e14bb325SJeff Bonwick continue; 1609e14bb325SJeff Bonwick zio_gang_tree_issue(zio, gn->gn_child[g], gbp, data); 1610e14bb325SJeff Bonwick data = (char *)data + BP_GET_PSIZE(gbp); 1611e14bb325SJeff Bonwick } 1612fa9e4066Sahrens } 1613fa9e4066Sahrens 1614f5383399SBill Moore if (gn == gio->io_gang_tree) 1615f5383399SBill Moore ASSERT3P((char *)gio->io_data + gio->io_size, ==, data); 1616e05725b1Sbonwick 1617e14bb325SJeff Bonwick if (zio != pio) 1618e14bb325SJeff Bonwick zio_nowait(zio); 1619fa9e4066Sahrens } 1620fa9e4066Sahrens 1621e05725b1Sbonwick static int 1622e14bb325SJeff Bonwick zio_gang_assemble(zio_t *zio) 1623fa9e4066Sahrens { 1624e14bb325SJeff Bonwick blkptr_t *bp = zio->io_bp; 1625fa9e4066Sahrens 1626f5383399SBill Moore ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == NULL); 1627f5383399SBill Moore ASSERT(zio->io_child_type > ZIO_CHILD_GANG); 1628f5383399SBill Moore 1629f5383399SBill Moore zio->io_gang_leader = zio; 1630fa9e4066Sahrens 1631e14bb325SJeff Bonwick zio_gang_tree_assemble(zio, bp, &zio->io_gang_tree); 1632e05725b1Sbonwick 1633e05725b1Sbonwick return (ZIO_PIPELINE_CONTINUE); 1634fa9e4066Sahrens } 1635fa9e4066Sahrens 1636e05725b1Sbonwick static int 1637e14bb325SJeff Bonwick zio_gang_issue(zio_t *zio) 1638fa9e4066Sahrens { 1639e14bb325SJeff Bonwick blkptr_t *bp = zio->io_bp; 1640fa9e4066Sahrens 1641e14bb325SJeff Bonwick if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE)) 1642e14bb325SJeff Bonwick return (ZIO_PIPELINE_STOP); 1643fa9e4066Sahrens 1644f5383399SBill Moore ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == zio); 1645f5383399SBill Moore ASSERT(zio->io_child_type > ZIO_CHILD_GANG); 1646fa9e4066Sahrens 1647e14bb325SJeff Bonwick if (zio->io_child_error[ZIO_CHILD_GANG] == 0) 1648f5383399SBill Moore zio_gang_tree_issue(zio, zio->io_gang_tree, bp, zio->io_data); 1649e14bb325SJeff Bonwick else 1650f5383399SBill Moore zio_gang_tree_free(&zio->io_gang_tree); 1651fa9e4066Sahrens 1652e14bb325SJeff Bonwick zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; 1653e05725b1Sbonwick 1654e05725b1Sbonwick return (ZIO_PIPELINE_CONTINUE); 1655fa9e4066Sahrens } 1656fa9e4066Sahrens 1657fa9e4066Sahrens static void 1658e14bb325SJeff Bonwick zio_write_gang_member_ready(zio_t *zio) 1659fa9e4066Sahrens { 1660a3f829aeSBill Moore zio_t *pio = zio_unique_parent(zio); 1661f5383399SBill Moore zio_t *gio = zio->io_gang_leader; 166244cd46caSbillm dva_t *cdva = zio->io_bp->blk_dva; 166344cd46caSbillm dva_t *pdva = pio->io_bp->blk_dva; 1664fa9e4066Sahrens uint64_t asize; 1665fa9e4066Sahrens 1666e14bb325SJeff Bonwick if (BP_IS_HOLE(zio->io_bp)) 1667e14bb325SJeff Bonwick return; 1668e14bb325SJeff Bonwick 1669e14bb325SJeff Bonwick ASSERT(BP_IS_HOLE(&zio->io_bp_orig)); 1670e14bb325SJeff Bonwick 1671e14bb325SJeff Bonwick ASSERT(zio->io_child_type == ZIO_CHILD_GANG); 1672b24ab676SJeff Bonwick ASSERT3U(zio->io_prop.zp_copies, ==, gio->io_prop.zp_copies); 1673b24ab676SJeff Bonwick ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(zio->io_bp)); 1674b24ab676SJeff Bonwick ASSERT3U(pio->io_prop.zp_copies, <=, BP_GET_NDVAS(pio->io_bp)); 167544cd46caSbillm ASSERT3U(BP_GET_NDVAS(zio->io_bp), <=, BP_GET_NDVAS(pio->io_bp)); 1676fa9e4066Sahrens 1677fa9e4066Sahrens mutex_enter(&pio->io_lock); 1678e14bb325SJeff Bonwick for (int d = 0; d < BP_GET_NDVAS(zio->io_bp); d++) { 167944cd46caSbillm ASSERT(DVA_GET_GANG(&pdva[d])); 168044cd46caSbillm asize = DVA_GET_ASIZE(&pdva[d]); 168144cd46caSbillm asize += DVA_GET_ASIZE(&cdva[d]); 168244cd46caSbillm DVA_SET_ASIZE(&pdva[d], asize); 168344cd46caSbillm } 1684fa9e4066Sahrens mutex_exit(&pio->io_lock); 1685fa9e4066Sahrens } 1686fa9e4066Sahrens 16870a4e9518Sgw static int 1688e14bb325SJeff Bonwick zio_write_gang_block(zio_t *pio) 1689fa9e4066Sahrens { 1690e14bb325SJeff Bonwick spa_t *spa = pio->io_spa; 1691e14bb325SJeff Bonwick blkptr_t *bp = pio->io_bp; 1692f5383399SBill Moore zio_t *gio = pio->io_gang_leader; 1693e14bb325SJeff Bonwick zio_t *zio; 1694e14bb325SJeff Bonwick zio_gang_node_t *gn, **gnpp; 1695fa9e4066Sahrens zio_gbh_phys_t *gbh; 1696e14bb325SJeff Bonwick uint64_t txg = pio->io_txg; 1697e14bb325SJeff Bonwick uint64_t resid = pio->io_size; 1698e14bb325SJeff Bonwick uint64_t lsize; 1699b24ab676SJeff Bonwick int copies = gio->io_prop.zp_copies; 1700b24ab676SJeff Bonwick int gbh_copies = MIN(copies + 1, spa_max_replication(spa)); 1701e14bb325SJeff Bonwick zio_prop_t zp; 1702fa9e4066Sahrens int error; 1703fa9e4066Sahrens 1704b24ab676SJeff Bonwick error = metaslab_alloc(spa, spa_normal_class(spa), SPA_GANGBLOCKSIZE, 1705b24ab676SJeff Bonwick bp, gbh_copies, txg, pio == gio ? NULL : gio->io_bp, 1706e14bb325SJeff Bonwick METASLAB_HINTBP_FAVOR | METASLAB_GANG_HEADER); 1707e05725b1Sbonwick if (error) { 1708e14bb325SJeff Bonwick pio->io_error = error; 1709e05725b1Sbonwick return (ZIO_PIPELINE_CONTINUE); 1710e05725b1Sbonwick } 1711fa9e4066Sahrens 1712f5383399SBill Moore if (pio == gio) { 1713f5383399SBill Moore gnpp = &gio->io_gang_tree; 1714e14bb325SJeff Bonwick } else { 1715e14bb325SJeff Bonwick gnpp = pio->io_private; 1716e14bb325SJeff Bonwick ASSERT(pio->io_ready == zio_write_gang_member_ready); 1717fa9e4066Sahrens } 1718fa9e4066Sahrens 1719e14bb325SJeff Bonwick gn = zio_gang_node_alloc(gnpp); 1720e14bb325SJeff Bonwick gbh = gn->gn_gbh; 1721e14bb325SJeff Bonwick bzero(gbh, SPA_GANGBLOCKSIZE); 1722fa9e4066Sahrens 1723e14bb325SJeff Bonwick /* 1724e14bb325SJeff Bonwick * Create the gang header. 1725e14bb325SJeff Bonwick */ 1726e14bb325SJeff Bonwick zio = zio_rewrite(pio, spa, txg, bp, gbh, SPA_GANGBLOCKSIZE, NULL, NULL, 1727e14bb325SJeff Bonwick pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); 1728fa9e4066Sahrens 1729e14bb325SJeff Bonwick /* 1730e14bb325SJeff Bonwick * Create and nowait the gang children. 1731e14bb325SJeff Bonwick */ 1732e14bb325SJeff Bonwick for (int g = 0; resid != 0; resid -= lsize, g++) { 1733e14bb325SJeff Bonwick lsize = P2ROUNDUP(resid / (SPA_GBH_NBLKPTRS - g), 1734e14bb325SJeff Bonwick SPA_MINBLOCKSIZE); 1735e14bb325SJeff Bonwick ASSERT(lsize >= SPA_MINBLOCKSIZE && lsize <= resid); 1736e14bb325SJeff Bonwick 1737f5383399SBill Moore zp.zp_checksum = gio->io_prop.zp_checksum; 1738e14bb325SJeff Bonwick zp.zp_compress = ZIO_COMPRESS_OFF; 1739e14bb325SJeff Bonwick zp.zp_type = DMU_OT_NONE; 1740e14bb325SJeff Bonwick zp.zp_level = 0; 1741b24ab676SJeff Bonwick zp.zp_copies = gio->io_prop.zp_copies; 1742b24ab676SJeff Bonwick zp.zp_dedup = 0; 1743b24ab676SJeff Bonwick zp.zp_dedup_verify = 0; 1744e14bb325SJeff Bonwick 1745e14bb325SJeff Bonwick zio_nowait(zio_write(zio, spa, txg, &gbh->zg_blkptr[g], 1746e14bb325SJeff Bonwick (char *)pio->io_data + (pio->io_size - resid), lsize, &zp, 1747e14bb325SJeff Bonwick zio_write_gang_member_ready, NULL, &gn->gn_child[g], 1748e14bb325SJeff Bonwick pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), 1749e14bb325SJeff Bonwick &pio->io_bookmark)); 1750e14bb325SJeff Bonwick } 1751e05725b1Sbonwick 175244cd46caSbillm /* 1753e14bb325SJeff Bonwick * Set pio's pipeline to just wait for zio to finish. 175444cd46caSbillm */ 1755e14bb325SJeff Bonwick pio->io_pipeline = ZIO_INTERLOCK_PIPELINE; 1756e14bb325SJeff Bonwick 1757e14bb325SJeff Bonwick zio_nowait(zio); 1758e14bb325SJeff Bonwick 1759e14bb325SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 1760fa9e4066Sahrens } 1761fa9e4066Sahrens 1762fa9e4066Sahrens /* 1763fa9e4066Sahrens * ========================================================================== 1764b24ab676SJeff Bonwick * Dedup 1765fa9e4066Sahrens * ========================================================================== 1766fa9e4066Sahrens */ 1767b24ab676SJeff Bonwick static void 1768b24ab676SJeff Bonwick zio_ddt_child_read_done(zio_t *zio) 1769b24ab676SJeff Bonwick { 1770b24ab676SJeff Bonwick blkptr_t *bp = zio->io_bp; 1771b24ab676SJeff Bonwick ddt_entry_t *dde = zio->io_private; 1772b24ab676SJeff Bonwick ddt_phys_t *ddp; 1773b24ab676SJeff Bonwick zio_t *pio = zio_unique_parent(zio); 1774b24ab676SJeff Bonwick 1775b24ab676SJeff Bonwick mutex_enter(&pio->io_lock); 1776b24ab676SJeff Bonwick ddp = ddt_phys_select(dde, bp); 1777b24ab676SJeff Bonwick if (zio->io_error == 0) 1778b24ab676SJeff Bonwick ddt_phys_clear(ddp); /* this ddp doesn't need repair */ 1779b24ab676SJeff Bonwick if (zio->io_error == 0 && dde->dde_repair_data == NULL) 1780b24ab676SJeff Bonwick dde->dde_repair_data = zio->io_data; 1781b24ab676SJeff Bonwick else 1782b24ab676SJeff Bonwick zio_buf_free(zio->io_data, zio->io_size); 1783b24ab676SJeff Bonwick mutex_exit(&pio->io_lock); 1784b24ab676SJeff Bonwick } 1785b24ab676SJeff Bonwick 1786b24ab676SJeff Bonwick static int 1787b24ab676SJeff Bonwick zio_ddt_read_start(zio_t *zio) 1788b24ab676SJeff Bonwick { 1789b24ab676SJeff Bonwick blkptr_t *bp = zio->io_bp; 1790b24ab676SJeff Bonwick 1791b24ab676SJeff Bonwick ASSERT(BP_GET_DEDUP(bp)); 1792b24ab676SJeff Bonwick ASSERT(BP_GET_PSIZE(bp) == zio->io_size); 1793b24ab676SJeff Bonwick ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 1794b24ab676SJeff Bonwick 1795b24ab676SJeff Bonwick if (zio->io_child_error[ZIO_CHILD_DDT]) { 1796b24ab676SJeff Bonwick ddt_t *ddt = ddt_select(zio->io_spa, bp); 1797b24ab676SJeff Bonwick ddt_entry_t *dde = ddt_repair_start(ddt, bp); 1798b24ab676SJeff Bonwick ddt_phys_t *ddp = dde->dde_phys; 1799b24ab676SJeff Bonwick ddt_phys_t *ddp_self = ddt_phys_select(dde, bp); 1800b24ab676SJeff Bonwick blkptr_t blk; 1801b24ab676SJeff Bonwick 1802b24ab676SJeff Bonwick ASSERT(zio->io_vsd == NULL); 1803b24ab676SJeff Bonwick zio->io_vsd = dde; 1804b24ab676SJeff Bonwick 1805b24ab676SJeff Bonwick if (ddp_self == NULL) 1806b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 1807b24ab676SJeff Bonwick 1808b24ab676SJeff Bonwick for (int p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { 1809b24ab676SJeff Bonwick if (ddp->ddp_phys_birth == 0 || ddp == ddp_self) 1810b24ab676SJeff Bonwick continue; 1811bbfd46c4SJeff Bonwick ddt_bp_create(ddt->ddt_checksum, &dde->dde_key, ddp, 1812bbfd46c4SJeff Bonwick &blk); 1813b24ab676SJeff Bonwick zio_nowait(zio_read(zio, zio->io_spa, &blk, 1814b24ab676SJeff Bonwick zio_buf_alloc(zio->io_size), zio->io_size, 1815b24ab676SJeff Bonwick zio_ddt_child_read_done, dde, zio->io_priority, 1816b24ab676SJeff Bonwick ZIO_DDT_CHILD_FLAGS(zio) | ZIO_FLAG_DONT_PROPAGATE, 1817b24ab676SJeff Bonwick &zio->io_bookmark)); 1818b24ab676SJeff Bonwick } 1819b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 1820b24ab676SJeff Bonwick } 1821b24ab676SJeff Bonwick 1822b24ab676SJeff Bonwick zio_nowait(zio_read(zio, zio->io_spa, bp, 1823b24ab676SJeff Bonwick zio->io_data, zio->io_size, NULL, NULL, zio->io_priority, 1824b24ab676SJeff Bonwick ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark)); 1825b24ab676SJeff Bonwick 1826b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 1827b24ab676SJeff Bonwick } 1828e14bb325SJeff Bonwick 1829b24ab676SJeff Bonwick static int 1830b24ab676SJeff Bonwick zio_ddt_read_done(zio_t *zio) 1831b24ab676SJeff Bonwick { 1832b24ab676SJeff Bonwick blkptr_t *bp = zio->io_bp; 1833b24ab676SJeff Bonwick 1834b24ab676SJeff Bonwick if (zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_DONE)) 1835b24ab676SJeff Bonwick return (ZIO_PIPELINE_STOP); 1836b24ab676SJeff Bonwick 1837b24ab676SJeff Bonwick ASSERT(BP_GET_DEDUP(bp)); 1838b24ab676SJeff Bonwick ASSERT(BP_GET_PSIZE(bp) == zio->io_size); 1839b24ab676SJeff Bonwick ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 1840b24ab676SJeff Bonwick 1841b24ab676SJeff Bonwick if (zio->io_child_error[ZIO_CHILD_DDT]) { 1842b24ab676SJeff Bonwick ddt_t *ddt = ddt_select(zio->io_spa, bp); 1843b24ab676SJeff Bonwick ddt_entry_t *dde = zio->io_vsd; 1844b24ab676SJeff Bonwick if (ddt == NULL) { 1845b16da2e2SGeorge Wilson ASSERT(spa_load_state(zio->io_spa) != SPA_LOAD_NONE); 1846b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 1847b24ab676SJeff Bonwick } 1848b24ab676SJeff Bonwick if (dde == NULL) { 1849b24ab676SJeff Bonwick zio->io_stage = ZIO_STAGE_DDT_READ_START >> 1; 185035a5a358SJonathan Adams zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE); 1851b24ab676SJeff Bonwick return (ZIO_PIPELINE_STOP); 1852b24ab676SJeff Bonwick } 1853b24ab676SJeff Bonwick if (dde->dde_repair_data != NULL) { 1854b24ab676SJeff Bonwick bcopy(dde->dde_repair_data, zio->io_data, zio->io_size); 1855b24ab676SJeff Bonwick zio->io_child_error[ZIO_CHILD_DDT] = 0; 1856b24ab676SJeff Bonwick } 1857b24ab676SJeff Bonwick ddt_repair_done(ddt, dde); 1858b24ab676SJeff Bonwick zio->io_vsd = NULL; 1859b24ab676SJeff Bonwick } 1860b24ab676SJeff Bonwick 1861b24ab676SJeff Bonwick ASSERT(zio->io_vsd == NULL); 1862b24ab676SJeff Bonwick 1863b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 1864b24ab676SJeff Bonwick } 1865b24ab676SJeff Bonwick 1866b24ab676SJeff Bonwick static boolean_t 1867b24ab676SJeff Bonwick zio_ddt_collision(zio_t *zio, ddt_t *ddt, ddt_entry_t *dde) 1868b24ab676SJeff Bonwick { 1869b24ab676SJeff Bonwick spa_t *spa = zio->io_spa; 1870b24ab676SJeff Bonwick 1871b24ab676SJeff Bonwick /* 1872b24ab676SJeff Bonwick * Note: we compare the original data, not the transformed data, 1873b24ab676SJeff Bonwick * because when zio->io_bp is an override bp, we will not have 1874b24ab676SJeff Bonwick * pushed the I/O transforms. That's an important optimization 1875b24ab676SJeff Bonwick * because otherwise we'd compress/encrypt all dmu_sync() data twice. 1876b24ab676SJeff Bonwick */ 1877b24ab676SJeff Bonwick for (int p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { 1878b24ab676SJeff Bonwick zio_t *lio = dde->dde_lead_zio[p]; 1879b24ab676SJeff Bonwick 1880b24ab676SJeff Bonwick if (lio != NULL) { 1881b24ab676SJeff Bonwick return (lio->io_orig_size != zio->io_orig_size || 1882b24ab676SJeff Bonwick bcmp(zio->io_orig_data, lio->io_orig_data, 1883b24ab676SJeff Bonwick zio->io_orig_size) != 0); 1884b24ab676SJeff Bonwick } 1885b24ab676SJeff Bonwick } 1886b24ab676SJeff Bonwick 1887b24ab676SJeff Bonwick for (int p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { 1888b24ab676SJeff Bonwick ddt_phys_t *ddp = &dde->dde_phys[p]; 1889b24ab676SJeff Bonwick 1890b24ab676SJeff Bonwick if (ddp->ddp_phys_birth != 0) { 1891b24ab676SJeff Bonwick arc_buf_t *abuf = NULL; 1892b24ab676SJeff Bonwick uint32_t aflags = ARC_WAIT; 1893b24ab676SJeff Bonwick blkptr_t blk = *zio->io_bp; 1894b24ab676SJeff Bonwick int error; 1895b24ab676SJeff Bonwick 1896b24ab676SJeff Bonwick ddt_bp_fill(ddp, &blk, ddp->ddp_phys_birth); 1897b24ab676SJeff Bonwick 1898b24ab676SJeff Bonwick ddt_exit(ddt); 1899b24ab676SJeff Bonwick 1900b24ab676SJeff Bonwick error = arc_read_nolock(NULL, spa, &blk, 1901b24ab676SJeff Bonwick arc_getbuf_func, &abuf, ZIO_PRIORITY_SYNC_READ, 1902b24ab676SJeff Bonwick ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, 1903b24ab676SJeff Bonwick &aflags, &zio->io_bookmark); 1904b24ab676SJeff Bonwick 1905b24ab676SJeff Bonwick if (error == 0) { 1906b24ab676SJeff Bonwick if (arc_buf_size(abuf) != zio->io_orig_size || 1907b24ab676SJeff Bonwick bcmp(abuf->b_data, zio->io_orig_data, 1908b24ab676SJeff Bonwick zio->io_orig_size) != 0) 1909b24ab676SJeff Bonwick error = EEXIST; 1910b24ab676SJeff Bonwick VERIFY(arc_buf_remove_ref(abuf, &abuf) == 1); 1911b24ab676SJeff Bonwick } 1912b24ab676SJeff Bonwick 1913b24ab676SJeff Bonwick ddt_enter(ddt); 1914b24ab676SJeff Bonwick return (error != 0); 1915b24ab676SJeff Bonwick } 1916b24ab676SJeff Bonwick } 1917b24ab676SJeff Bonwick 1918b24ab676SJeff Bonwick return (B_FALSE); 1919b24ab676SJeff Bonwick } 1920b24ab676SJeff Bonwick 1921b24ab676SJeff Bonwick static void 1922b24ab676SJeff Bonwick zio_ddt_child_write_ready(zio_t *zio) 1923b24ab676SJeff Bonwick { 1924b24ab676SJeff Bonwick int p = zio->io_prop.zp_copies; 1925b24ab676SJeff Bonwick ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp); 1926b24ab676SJeff Bonwick ddt_entry_t *dde = zio->io_private; 1927b24ab676SJeff Bonwick ddt_phys_t *ddp = &dde->dde_phys[p]; 1928b24ab676SJeff Bonwick zio_t *pio; 1929b24ab676SJeff Bonwick 1930b24ab676SJeff Bonwick if (zio->io_error) 1931b24ab676SJeff Bonwick return; 1932b24ab676SJeff Bonwick 1933b24ab676SJeff Bonwick ddt_enter(ddt); 1934b24ab676SJeff Bonwick 1935b24ab676SJeff Bonwick ASSERT(dde->dde_lead_zio[p] == zio); 1936b24ab676SJeff Bonwick 1937b24ab676SJeff Bonwick ddt_phys_fill(ddp, zio->io_bp); 1938b24ab676SJeff Bonwick 1939b24ab676SJeff Bonwick while ((pio = zio_walk_parents(zio)) != NULL) 1940b24ab676SJeff Bonwick ddt_bp_fill(ddp, pio->io_bp, zio->io_txg); 1941b24ab676SJeff Bonwick 1942b24ab676SJeff Bonwick ddt_exit(ddt); 1943b24ab676SJeff Bonwick } 1944b24ab676SJeff Bonwick 1945b24ab676SJeff Bonwick static void 1946b24ab676SJeff Bonwick zio_ddt_child_write_done(zio_t *zio) 1947b24ab676SJeff Bonwick { 1948b24ab676SJeff Bonwick int p = zio->io_prop.zp_copies; 1949b24ab676SJeff Bonwick ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp); 1950b24ab676SJeff Bonwick ddt_entry_t *dde = zio->io_private; 1951b24ab676SJeff Bonwick ddt_phys_t *ddp = &dde->dde_phys[p]; 1952b24ab676SJeff Bonwick 1953b24ab676SJeff Bonwick ddt_enter(ddt); 1954b24ab676SJeff Bonwick 1955b24ab676SJeff Bonwick ASSERT(ddp->ddp_refcnt == 0); 1956b24ab676SJeff Bonwick ASSERT(dde->dde_lead_zio[p] == zio); 1957b24ab676SJeff Bonwick dde->dde_lead_zio[p] = NULL; 1958b24ab676SJeff Bonwick 1959b24ab676SJeff Bonwick if (zio->io_error == 0) { 1960b24ab676SJeff Bonwick while (zio_walk_parents(zio) != NULL) 1961b24ab676SJeff Bonwick ddt_phys_addref(ddp); 1962b24ab676SJeff Bonwick } else { 1963b24ab676SJeff Bonwick ddt_phys_clear(ddp); 1964b24ab676SJeff Bonwick } 1965b24ab676SJeff Bonwick 1966b24ab676SJeff Bonwick ddt_exit(ddt); 1967b24ab676SJeff Bonwick } 1968b24ab676SJeff Bonwick 1969b24ab676SJeff Bonwick static void 1970b24ab676SJeff Bonwick zio_ddt_ditto_write_done(zio_t *zio) 1971b24ab676SJeff Bonwick { 1972b24ab676SJeff Bonwick int p = DDT_PHYS_DITTO; 1973b24ab676SJeff Bonwick zio_prop_t *zp = &zio->io_prop; 1974b24ab676SJeff Bonwick blkptr_t *bp = zio->io_bp; 1975b24ab676SJeff Bonwick ddt_t *ddt = ddt_select(zio->io_spa, bp); 1976b24ab676SJeff Bonwick ddt_entry_t *dde = zio->io_private; 1977b24ab676SJeff Bonwick ddt_phys_t *ddp = &dde->dde_phys[p]; 1978b24ab676SJeff Bonwick ddt_key_t *ddk = &dde->dde_key; 1979b24ab676SJeff Bonwick 1980b24ab676SJeff Bonwick ddt_enter(ddt); 1981b24ab676SJeff Bonwick 1982b24ab676SJeff Bonwick ASSERT(ddp->ddp_refcnt == 0); 1983b24ab676SJeff Bonwick ASSERT(dde->dde_lead_zio[p] == zio); 1984b24ab676SJeff Bonwick dde->dde_lead_zio[p] = NULL; 1985b24ab676SJeff Bonwick 1986b24ab676SJeff Bonwick if (zio->io_error == 0) { 1987b24ab676SJeff Bonwick ASSERT(ZIO_CHECKSUM_EQUAL(bp->blk_cksum, ddk->ddk_cksum)); 1988b24ab676SJeff Bonwick ASSERT(zp->zp_copies < SPA_DVAS_PER_BP); 1989b24ab676SJeff Bonwick ASSERT(zp->zp_copies == BP_GET_NDVAS(bp) - BP_IS_GANG(bp)); 1990b24ab676SJeff Bonwick if (ddp->ddp_phys_birth != 0) 1991b24ab676SJeff Bonwick ddt_phys_free(ddt, ddk, ddp, zio->io_txg); 1992b24ab676SJeff Bonwick ddt_phys_fill(ddp, bp); 1993b24ab676SJeff Bonwick } 1994b24ab676SJeff Bonwick 1995b24ab676SJeff Bonwick ddt_exit(ddt); 1996b24ab676SJeff Bonwick } 1997b24ab676SJeff Bonwick 1998b24ab676SJeff Bonwick static int 1999b24ab676SJeff Bonwick zio_ddt_write(zio_t *zio) 2000b24ab676SJeff Bonwick { 2001b24ab676SJeff Bonwick spa_t *spa = zio->io_spa; 2002b24ab676SJeff Bonwick blkptr_t *bp = zio->io_bp; 2003b24ab676SJeff Bonwick uint64_t txg = zio->io_txg; 2004b24ab676SJeff Bonwick zio_prop_t *zp = &zio->io_prop; 2005b24ab676SJeff Bonwick int p = zp->zp_copies; 2006b24ab676SJeff Bonwick int ditto_copies; 2007b24ab676SJeff Bonwick zio_t *cio = NULL; 2008b24ab676SJeff Bonwick zio_t *dio = NULL; 2009b24ab676SJeff Bonwick ddt_t *ddt = ddt_select(spa, bp); 2010b24ab676SJeff Bonwick ddt_entry_t *dde; 2011b24ab676SJeff Bonwick ddt_phys_t *ddp; 2012b24ab676SJeff Bonwick 2013b24ab676SJeff Bonwick ASSERT(BP_GET_DEDUP(bp)); 2014b24ab676SJeff Bonwick ASSERT(BP_GET_CHECKSUM(bp) == zp->zp_checksum); 2015b24ab676SJeff Bonwick ASSERT(BP_IS_HOLE(bp) || zio->io_bp_override); 2016b24ab676SJeff Bonwick 2017b24ab676SJeff Bonwick ddt_enter(ddt); 2018b24ab676SJeff Bonwick dde = ddt_lookup(ddt, bp, B_TRUE); 2019b24ab676SJeff Bonwick ddp = &dde->dde_phys[p]; 2020b24ab676SJeff Bonwick 2021b24ab676SJeff Bonwick if (zp->zp_dedup_verify && zio_ddt_collision(zio, ddt, dde)) { 2022b24ab676SJeff Bonwick /* 2023b24ab676SJeff Bonwick * If we're using a weak checksum, upgrade to a strong checksum 2024b24ab676SJeff Bonwick * and try again. If we're already using a strong checksum, 2025b24ab676SJeff Bonwick * we can't resolve it, so just convert to an ordinary write. 2026b24ab676SJeff Bonwick * (And automatically e-mail a paper to Nature?) 2027b24ab676SJeff Bonwick */ 2028b24ab676SJeff Bonwick if (!zio_checksum_table[zp->zp_checksum].ci_dedup) { 2029b24ab676SJeff Bonwick zp->zp_checksum = spa_dedup_checksum(spa); 2030b24ab676SJeff Bonwick zio_pop_transforms(zio); 2031b24ab676SJeff Bonwick zio->io_stage = ZIO_STAGE_OPEN; 2032b24ab676SJeff Bonwick BP_ZERO(bp); 2033b24ab676SJeff Bonwick } else { 2034b24ab676SJeff Bonwick zp->zp_dedup = 0; 2035b24ab676SJeff Bonwick } 2036b24ab676SJeff Bonwick zio->io_pipeline = ZIO_WRITE_PIPELINE; 2037b24ab676SJeff Bonwick ddt_exit(ddt); 2038b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 2039b24ab676SJeff Bonwick } 2040b24ab676SJeff Bonwick 2041b24ab676SJeff Bonwick ditto_copies = ddt_ditto_copies_needed(ddt, dde, ddp); 2042b24ab676SJeff Bonwick ASSERT(ditto_copies < SPA_DVAS_PER_BP); 2043b24ab676SJeff Bonwick 2044b24ab676SJeff Bonwick if (ditto_copies > ddt_ditto_copies_present(dde) && 2045b24ab676SJeff Bonwick dde->dde_lead_zio[DDT_PHYS_DITTO] == NULL) { 2046b24ab676SJeff Bonwick zio_prop_t czp = *zp; 2047b24ab676SJeff Bonwick 2048b24ab676SJeff Bonwick czp.zp_copies = ditto_copies; 2049b24ab676SJeff Bonwick 2050b24ab676SJeff Bonwick /* 2051b24ab676SJeff Bonwick * If we arrived here with an override bp, we won't have run 2052b24ab676SJeff Bonwick * the transform stack, so we won't have the data we need to 2053b24ab676SJeff Bonwick * generate a child i/o. So, toss the override bp and restart. 2054b24ab676SJeff Bonwick * This is safe, because using the override bp is just an 2055b24ab676SJeff Bonwick * optimization; and it's rare, so the cost doesn't matter. 2056b24ab676SJeff Bonwick */ 2057b24ab676SJeff Bonwick if (zio->io_bp_override) { 2058b24ab676SJeff Bonwick zio_pop_transforms(zio); 2059b24ab676SJeff Bonwick zio->io_stage = ZIO_STAGE_OPEN; 2060b24ab676SJeff Bonwick zio->io_pipeline = ZIO_WRITE_PIPELINE; 2061b24ab676SJeff Bonwick zio->io_bp_override = NULL; 2062b24ab676SJeff Bonwick BP_ZERO(bp); 2063b24ab676SJeff Bonwick ddt_exit(ddt); 2064b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 2065b24ab676SJeff Bonwick } 2066b24ab676SJeff Bonwick 2067b24ab676SJeff Bonwick dio = zio_write(zio, spa, txg, bp, zio->io_orig_data, 2068b24ab676SJeff Bonwick zio->io_orig_size, &czp, NULL, 2069b24ab676SJeff Bonwick zio_ddt_ditto_write_done, dde, zio->io_priority, 2070b24ab676SJeff Bonwick ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark); 2071b24ab676SJeff Bonwick 2072b24ab676SJeff Bonwick zio_push_transform(dio, zio->io_data, zio->io_size, 0, NULL); 2073b24ab676SJeff Bonwick dde->dde_lead_zio[DDT_PHYS_DITTO] = dio; 2074b24ab676SJeff Bonwick } 2075b24ab676SJeff Bonwick 2076b24ab676SJeff Bonwick if (ddp->ddp_phys_birth != 0 || dde->dde_lead_zio[p] != NULL) { 2077b24ab676SJeff Bonwick if (ddp->ddp_phys_birth != 0) 2078b24ab676SJeff Bonwick ddt_bp_fill(ddp, bp, txg); 2079b24ab676SJeff Bonwick if (dde->dde_lead_zio[p] != NULL) 2080b24ab676SJeff Bonwick zio_add_child(zio, dde->dde_lead_zio[p]); 2081b24ab676SJeff Bonwick else 2082b24ab676SJeff Bonwick ddt_phys_addref(ddp); 2083b24ab676SJeff Bonwick } else if (zio->io_bp_override) { 2084b24ab676SJeff Bonwick ASSERT(bp->blk_birth == txg); 2085b24ab676SJeff Bonwick ASSERT(BP_EQUAL(bp, zio->io_bp_override)); 2086b24ab676SJeff Bonwick ddt_phys_fill(ddp, bp); 2087b24ab676SJeff Bonwick ddt_phys_addref(ddp); 2088b24ab676SJeff Bonwick } else { 2089b24ab676SJeff Bonwick cio = zio_write(zio, spa, txg, bp, zio->io_orig_data, 2090b24ab676SJeff Bonwick zio->io_orig_size, zp, zio_ddt_child_write_ready, 2091b24ab676SJeff Bonwick zio_ddt_child_write_done, dde, zio->io_priority, 2092b24ab676SJeff Bonwick ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark); 2093b24ab676SJeff Bonwick 2094b24ab676SJeff Bonwick zio_push_transform(cio, zio->io_data, zio->io_size, 0, NULL); 2095b24ab676SJeff Bonwick dde->dde_lead_zio[p] = cio; 2096b24ab676SJeff Bonwick } 2097b24ab676SJeff Bonwick 2098b24ab676SJeff Bonwick ddt_exit(ddt); 2099b24ab676SJeff Bonwick 2100b24ab676SJeff Bonwick if (cio) 2101b24ab676SJeff Bonwick zio_nowait(cio); 2102b24ab676SJeff Bonwick if (dio) 2103b24ab676SJeff Bonwick zio_nowait(dio); 2104b24ab676SJeff Bonwick 2105b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 2106b24ab676SJeff Bonwick } 2107b24ab676SJeff Bonwick 21083f9d6ad7SLin Ling ddt_entry_t *freedde; /* for debugging */ 21093f9d6ad7SLin Ling 2110b24ab676SJeff Bonwick static int 2111b24ab676SJeff Bonwick zio_ddt_free(zio_t *zio) 2112b24ab676SJeff Bonwick { 2113b24ab676SJeff Bonwick spa_t *spa = zio->io_spa; 2114b24ab676SJeff Bonwick blkptr_t *bp = zio->io_bp; 2115b24ab676SJeff Bonwick ddt_t *ddt = ddt_select(spa, bp); 2116b24ab676SJeff Bonwick ddt_entry_t *dde; 2117b24ab676SJeff Bonwick ddt_phys_t *ddp; 2118b24ab676SJeff Bonwick 2119b24ab676SJeff Bonwick ASSERT(BP_GET_DEDUP(bp)); 2120b24ab676SJeff Bonwick ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 2121b24ab676SJeff Bonwick 2122b24ab676SJeff Bonwick ddt_enter(ddt); 21233f9d6ad7SLin Ling freedde = dde = ddt_lookup(ddt, bp, B_TRUE); 2124b24ab676SJeff Bonwick ddp = ddt_phys_select(dde, bp); 2125b24ab676SJeff Bonwick ddt_phys_decref(ddp); 2126b24ab676SJeff Bonwick ddt_exit(ddt); 2127b24ab676SJeff Bonwick 2128b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 2129b24ab676SJeff Bonwick } 2130b24ab676SJeff Bonwick 2131b24ab676SJeff Bonwick /* 2132b24ab676SJeff Bonwick * ========================================================================== 2133b24ab676SJeff Bonwick * Allocate and free blocks 2134b24ab676SJeff Bonwick * ========================================================================== 2135b24ab676SJeff Bonwick */ 2136e05725b1Sbonwick static int 2137fa9e4066Sahrens zio_dva_allocate(zio_t *zio) 2138fa9e4066Sahrens { 21398654d025Sperrin spa_t *spa = zio->io_spa; 2140b24ab676SJeff Bonwick metaslab_class_t *mc = spa_normal_class(spa); 2141fa9e4066Sahrens blkptr_t *bp = zio->io_bp; 2142fa9e4066Sahrens int error; 214309c9d376SGeorge Wilson int flags = 0; 2144fa9e4066Sahrens 2145f5383399SBill Moore if (zio->io_gang_leader == NULL) { 2146f5383399SBill Moore ASSERT(zio->io_child_type > ZIO_CHILD_GANG); 2147f5383399SBill Moore zio->io_gang_leader = zio; 2148f5383399SBill Moore } 2149f5383399SBill Moore 2150fa9e4066Sahrens ASSERT(BP_IS_HOLE(bp)); 2151b420f3adSRichard Lowe ASSERT3U(BP_GET_NDVAS(bp), ==, 0); 2152b24ab676SJeff Bonwick ASSERT3U(zio->io_prop.zp_copies, >, 0); 2153b24ab676SJeff Bonwick ASSERT3U(zio->io_prop.zp_copies, <=, spa_max_replication(spa)); 2154fa9e4066Sahrens ASSERT3U(zio->io_size, ==, BP_GET_PSIZE(bp)); 2155fa9e4066Sahrens 215609c9d376SGeorge Wilson /* 215709c9d376SGeorge Wilson * The dump device does not support gang blocks so allocation on 215809c9d376SGeorge Wilson * behalf of the dump device (i.e. ZIO_FLAG_NODATA) must avoid 215909c9d376SGeorge Wilson * the "fast" gang feature. 216009c9d376SGeorge Wilson */ 216109c9d376SGeorge Wilson flags |= (zio->io_flags & ZIO_FLAG_NODATA) ? METASLAB_GANG_AVOID : 0; 216209c9d376SGeorge Wilson flags |= (zio->io_flags & ZIO_FLAG_GANG_CHILD) ? 216309c9d376SGeorge Wilson METASLAB_GANG_CHILD : 0; 2164e14bb325SJeff Bonwick error = metaslab_alloc(spa, mc, zio->io_size, bp, 216509c9d376SGeorge Wilson zio->io_prop.zp_copies, zio->io_txg, NULL, flags); 2166fa9e4066Sahrens 2167e14bb325SJeff Bonwick if (error) { 216809c9d376SGeorge Wilson spa_dbgmsg(spa, "%s: metaslab allocation failure: zio %p, " 216909c9d376SGeorge Wilson "size %llu, error %d", spa_name(spa), zio, zio->io_size, 217009c9d376SGeorge Wilson error); 2171e14bb325SJeff Bonwick if (error == ENOSPC && zio->io_size > SPA_MINBLOCKSIZE) 2172e14bb325SJeff Bonwick return (zio_write_gang_block(zio)); 2173fa9e4066Sahrens zio->io_error = error; 2174fa9e4066Sahrens } 2175e05725b1Sbonwick 2176e05725b1Sbonwick return (ZIO_PIPELINE_CONTINUE); 2177fa9e4066Sahrens } 2178fa9e4066Sahrens 2179e05725b1Sbonwick static int 2180fa9e4066Sahrens zio_dva_free(zio_t *zio) 2181fa9e4066Sahrens { 2182e14bb325SJeff Bonwick metaslab_free(zio->io_spa, zio->io_bp, zio->io_txg, B_FALSE); 2183fa9e4066Sahrens 2184e05725b1Sbonwick return (ZIO_PIPELINE_CONTINUE); 2185fa9e4066Sahrens } 2186fa9e4066Sahrens 2187e05725b1Sbonwick static int 2188fa9e4066Sahrens zio_dva_claim(zio_t *zio) 2189fa9e4066Sahrens { 2190e14bb325SJeff Bonwick int error; 2191e14bb325SJeff Bonwick 2192e14bb325SJeff Bonwick error = metaslab_claim(zio->io_spa, zio->io_bp, zio->io_txg); 2193e14bb325SJeff Bonwick if (error) 2194e14bb325SJeff Bonwick zio->io_error = error; 2195fa9e4066Sahrens 2196e05725b1Sbonwick return (ZIO_PIPELINE_CONTINUE); 2197fa9e4066Sahrens } 2198fa9e4066Sahrens 2199e14bb325SJeff Bonwick /* 2200e14bb325SJeff Bonwick * Undo an allocation. This is used by zio_done() when an I/O fails 2201e14bb325SJeff Bonwick * and we want to give back the block we just allocated. 2202e14bb325SJeff Bonwick * This handles both normal blocks and gang blocks. 2203e14bb325SJeff Bonwick */ 2204e14bb325SJeff Bonwick static void 2205e14bb325SJeff Bonwick zio_dva_unallocate(zio_t *zio, zio_gang_node_t *gn, blkptr_t *bp) 2206e14bb325SJeff Bonwick { 2207e14bb325SJeff Bonwick ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp)); 2208b24ab676SJeff Bonwick ASSERT(zio->io_bp_override == NULL); 2209e14bb325SJeff Bonwick 2210e14bb325SJeff Bonwick if (!BP_IS_HOLE(bp)) 2211b24ab676SJeff Bonwick metaslab_free(zio->io_spa, bp, bp->blk_birth, B_TRUE); 2212e14bb325SJeff Bonwick 2213e14bb325SJeff Bonwick if (gn != NULL) { 2214e14bb325SJeff Bonwick for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) { 2215e14bb325SJeff Bonwick zio_dva_unallocate(zio, gn->gn_child[g], 2216e14bb325SJeff Bonwick &gn->gn_gbh->zg_blkptr[g]); 2217e14bb325SJeff Bonwick } 2218e14bb325SJeff Bonwick } 2219e14bb325SJeff Bonwick } 2220e14bb325SJeff Bonwick 2221e14bb325SJeff Bonwick /* 2222e14bb325SJeff Bonwick * Try to allocate an intent log block. Return 0 on success, errno on failure. 2223e14bb325SJeff Bonwick */ 2224e14bb325SJeff Bonwick int 2225b24ab676SJeff Bonwick zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp, blkptr_t *old_bp, 2226b24ab676SJeff Bonwick uint64_t size, boolean_t use_slog) 2227e14bb325SJeff Bonwick { 2228e09fa4daSNeil Perrin int error = 1; 2229e14bb325SJeff Bonwick 2230b24ab676SJeff Bonwick ASSERT(txg > spa_syncing_txg(spa)); 2231b24ab676SJeff Bonwick 2232840345f6SGeorge Wilson /* 2233840345f6SGeorge Wilson * ZIL blocks are always contiguous (i.e. not gang blocks) so we 2234840345f6SGeorge Wilson * set the METASLAB_GANG_AVOID flag so that they don't "fast gang" 2235840345f6SGeorge Wilson * when allocating them. 2236840345f6SGeorge Wilson */ 2237840345f6SGeorge Wilson if (use_slog) { 2238b24ab676SJeff Bonwick error = metaslab_alloc(spa, spa_log_class(spa), size, 2239840345f6SGeorge Wilson new_bp, 1, txg, old_bp, 2240840345f6SGeorge Wilson METASLAB_HINTBP_AVOID | METASLAB_GANG_AVOID); 2241840345f6SGeorge Wilson } 2242e14bb325SJeff Bonwick 2243840345f6SGeorge Wilson if (error) { 2244b24ab676SJeff Bonwick error = metaslab_alloc(spa, spa_normal_class(spa), size, 2245840345f6SGeorge Wilson new_bp, 1, txg, old_bp, 2246840345f6SGeorge Wilson METASLAB_HINTBP_AVOID | METASLAB_GANG_AVOID); 2247840345f6SGeorge Wilson } 2248e14bb325SJeff Bonwick 2249e14bb325SJeff Bonwick if (error == 0) { 2250e14bb325SJeff Bonwick BP_SET_LSIZE(new_bp, size); 2251e14bb325SJeff Bonwick BP_SET_PSIZE(new_bp, size); 2252e14bb325SJeff Bonwick BP_SET_COMPRESS(new_bp, ZIO_COMPRESS_OFF); 22536e1f5caaSNeil Perrin BP_SET_CHECKSUM(new_bp, 22546e1f5caaSNeil Perrin spa_version(spa) >= SPA_VERSION_SLIM_ZIL 22556e1f5caaSNeil Perrin ? ZIO_CHECKSUM_ZILOG2 : ZIO_CHECKSUM_ZILOG); 2256e14bb325SJeff Bonwick BP_SET_TYPE(new_bp, DMU_OT_INTENT_LOG); 2257e14bb325SJeff Bonwick BP_SET_LEVEL(new_bp, 0); 2258b24ab676SJeff Bonwick BP_SET_DEDUP(new_bp, 0); 2259e14bb325SJeff Bonwick BP_SET_BYTEORDER(new_bp, ZFS_HOST_BYTEORDER); 2260e14bb325SJeff Bonwick } 2261e14bb325SJeff Bonwick 2262e14bb325SJeff Bonwick return (error); 2263e14bb325SJeff Bonwick } 2264e14bb325SJeff Bonwick 2265e14bb325SJeff Bonwick /* 2266b24ab676SJeff Bonwick * Free an intent log block. 2267e14bb325SJeff Bonwick */ 2268e14bb325SJeff Bonwick void 2269b24ab676SJeff Bonwick zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp) 2270e14bb325SJeff Bonwick { 2271b24ab676SJeff Bonwick ASSERT(BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG); 2272e14bb325SJeff Bonwick ASSERT(!BP_IS_GANG(bp)); 2273e14bb325SJeff Bonwick 2274b24ab676SJeff Bonwick zio_free(spa, txg, bp); 2275e14bb325SJeff Bonwick } 2276e14bb325SJeff Bonwick 2277fa9e4066Sahrens /* 2278fa9e4066Sahrens * ========================================================================== 2279fa9e4066Sahrens * Read and write to physical devices 2280fa9e4066Sahrens * ========================================================================== 2281fa9e4066Sahrens */ 2282e05725b1Sbonwick static int 228344cd46caSbillm zio_vdev_io_start(zio_t *zio) 2284fa9e4066Sahrens { 2285fa9e4066Sahrens vdev_t *vd = zio->io_vd; 228644cd46caSbillm uint64_t align; 22870a4e9518Sgw spa_t *spa = zio->io_spa; 22880a4e9518Sgw 2289e14bb325SJeff Bonwick ASSERT(zio->io_error == 0); 2290e14bb325SJeff Bonwick ASSERT(zio->io_child_error[ZIO_CHILD_VDEV] == 0); 2291fa9e4066Sahrens 2292e14bb325SJeff Bonwick if (vd == NULL) { 2293e14bb325SJeff Bonwick if (!(zio->io_flags & ZIO_FLAG_CONFIG_WRITER)) 2294e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ZIO, zio, RW_READER); 2295fa9e4066Sahrens 2296e14bb325SJeff Bonwick /* 2297e14bb325SJeff Bonwick * The mirror_ops handle multiple DVAs in a single BP. 2298e14bb325SJeff Bonwick */ 2299e14bb325SJeff Bonwick return (vdev_mirror_ops.vdev_op_io_start(zio)); 2300fa9e4066Sahrens } 2301fa9e4066Sahrens 230244ecc532SGeorge Wilson /* 230344ecc532SGeorge Wilson * We keep track of time-sensitive I/Os so that the scan thread 230444ecc532SGeorge Wilson * can quickly react to certain workloads. In particular, we care 230544ecc532SGeorge Wilson * about non-scrubbing, top-level reads and writes with the following 230644ecc532SGeorge Wilson * characteristics: 230744ecc532SGeorge Wilson * - synchronous writes of user data to non-slog devices 230844ecc532SGeorge Wilson * - any reads of user data 230944ecc532SGeorge Wilson * When these conditions are met, adjust the timestamp of spa_last_io 231044ecc532SGeorge Wilson * which allows the scan thread to adjust its workload accordingly. 231144ecc532SGeorge Wilson */ 231244ecc532SGeorge Wilson if (!(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && zio->io_bp != NULL && 231344ecc532SGeorge Wilson vd == vd->vdev_top && !vd->vdev_islog && 231444ecc532SGeorge Wilson zio->io_bookmark.zb_objset != DMU_META_OBJSET && 231544ecc532SGeorge Wilson zio->io_txg != spa_syncing_txg(spa)) { 231644ecc532SGeorge Wilson uint64_t old = spa->spa_last_io; 231744ecc532SGeorge Wilson uint64_t new = ddi_get_lbolt64(); 231844ecc532SGeorge Wilson if (old != new) 231944ecc532SGeorge Wilson (void) atomic_cas_64(&spa->spa_last_io, old, new); 232044ecc532SGeorge Wilson } 232144ecc532SGeorge Wilson 2322e14bb325SJeff Bonwick align = 1ULL << vd->vdev_top->vdev_ashift; 2323e14bb325SJeff Bonwick 2324ecc2d604Sbonwick if (P2PHASE(zio->io_size, align) != 0) { 2325ecc2d604Sbonwick uint64_t asize = P2ROUNDUP(zio->io_size, align); 2326ecc2d604Sbonwick char *abuf = zio_buf_alloc(asize); 2327e14bb325SJeff Bonwick ASSERT(vd == vd->vdev_top); 2328ecc2d604Sbonwick if (zio->io_type == ZIO_TYPE_WRITE) { 2329ecc2d604Sbonwick bcopy(zio->io_data, abuf, zio->io_size); 2330ecc2d604Sbonwick bzero(abuf + zio->io_size, asize - zio->io_size); 2331ecc2d604Sbonwick } 2332e14bb325SJeff Bonwick zio_push_transform(zio, abuf, asize, asize, zio_subblock); 2333ecc2d604Sbonwick } 2334ecc2d604Sbonwick 2335ecc2d604Sbonwick ASSERT(P2PHASE(zio->io_offset, align) == 0); 2336ecc2d604Sbonwick ASSERT(P2PHASE(zio->io_size, align) == 0); 2337f9af39baSGeorge Wilson VERIFY(zio->io_type != ZIO_TYPE_WRITE || spa_writeable(spa)); 23388ad4d6ddSJeff Bonwick 23398ad4d6ddSJeff Bonwick /* 23408ad4d6ddSJeff Bonwick * If this is a repair I/O, and there's no self-healing involved -- 23418ad4d6ddSJeff Bonwick * that is, we're just resilvering what we expect to resilver -- 23428ad4d6ddSJeff Bonwick * then don't do the I/O unless zio's txg is actually in vd's DTL. 23438ad4d6ddSJeff Bonwick * This prevents spurious resilvering with nested replication. 23448ad4d6ddSJeff Bonwick * For example, given a mirror of mirrors, (A+B)+(C+D), if only 23458ad4d6ddSJeff Bonwick * A is out of date, we'll read from C+D, then use the data to 23468ad4d6ddSJeff Bonwick * resilver A+B -- but we don't actually want to resilver B, just A. 23478ad4d6ddSJeff Bonwick * The top-level mirror has no way to know this, so instead we just 23488ad4d6ddSJeff Bonwick * discard unnecessary repairs as we work our way down the vdev tree. 23498ad4d6ddSJeff Bonwick * The same logic applies to any form of nested replication: 23508ad4d6ddSJeff Bonwick * ditto + mirror, RAID-Z + replacing, etc. This covers them all. 23518ad4d6ddSJeff Bonwick */ 23528ad4d6ddSJeff Bonwick if ((zio->io_flags & ZIO_FLAG_IO_REPAIR) && 23538ad4d6ddSJeff Bonwick !(zio->io_flags & ZIO_FLAG_SELF_HEAL) && 23548ad4d6ddSJeff Bonwick zio->io_txg != 0 && /* not a delegated i/o */ 23558ad4d6ddSJeff Bonwick !vdev_dtl_contains(vd, DTL_PARTIAL, zio->io_txg, 1)) { 23568ad4d6ddSJeff Bonwick ASSERT(zio->io_type == ZIO_TYPE_WRITE); 23578ad4d6ddSJeff Bonwick zio_vdev_io_bypass(zio); 23588ad4d6ddSJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 23598ad4d6ddSJeff Bonwick } 2360fa9e4066Sahrens 2361e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && 2362e14bb325SJeff Bonwick (zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE)) { 2363e14bb325SJeff Bonwick 2364e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ && vdev_cache_read(zio) == 0) 2365a3f829aeSBill Moore return (ZIO_PIPELINE_CONTINUE); 2366e14bb325SJeff Bonwick 2367e14bb325SJeff Bonwick if ((zio = vdev_queue_io(zio)) == NULL) 2368e14bb325SJeff Bonwick return (ZIO_PIPELINE_STOP); 2369e14bb325SJeff Bonwick 2370e14bb325SJeff Bonwick if (!vdev_accessible(vd, zio)) { 2371e14bb325SJeff Bonwick zio->io_error = ENXIO; 2372e14bb325SJeff Bonwick zio_interrupt(zio); 2373e14bb325SJeff Bonwick return (ZIO_PIPELINE_STOP); 2374e14bb325SJeff Bonwick } 2375e14bb325SJeff Bonwick } 2376e14bb325SJeff Bonwick 2377e05725b1Sbonwick return (vd->vdev_ops->vdev_op_io_start(zio)); 2378fa9e4066Sahrens } 2379fa9e4066Sahrens 2380e05725b1Sbonwick static int 2381fa9e4066Sahrens zio_vdev_io_done(zio_t *zio) 2382fa9e4066Sahrens { 2383e14bb325SJeff Bonwick vdev_t *vd = zio->io_vd; 2384e14bb325SJeff Bonwick vdev_ops_t *ops = vd ? vd->vdev_ops : &vdev_mirror_ops; 2385e14bb325SJeff Bonwick boolean_t unexpected_error = B_FALSE; 2386e05725b1Sbonwick 2387e14bb325SJeff Bonwick if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE)) 2388e14bb325SJeff Bonwick return (ZIO_PIPELINE_STOP); 2389fa9e4066Sahrens 2390e14bb325SJeff Bonwick ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE); 2391e14bb325SJeff Bonwick 2392e14bb325SJeff Bonwick if (vd != NULL && vd->vdev_ops->vdev_op_leaf) { 2393e14bb325SJeff Bonwick 2394e14bb325SJeff Bonwick vdev_queue_io_done(zio); 2395fa9e4066Sahrens 2396e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_WRITE) 2397e14bb325SJeff Bonwick vdev_cache_write(zio); 2398e14bb325SJeff Bonwick 2399e14bb325SJeff Bonwick if (zio_injection_enabled && zio->io_error == 0) 24008956713aSEric Schrock zio->io_error = zio_handle_device_injection(vd, 24018956713aSEric Schrock zio, EIO); 2402e14bb325SJeff Bonwick 2403e14bb325SJeff Bonwick if (zio_injection_enabled && zio->io_error == 0) 2404e14bb325SJeff Bonwick zio->io_error = zio_handle_label_injection(zio, EIO); 2405e14bb325SJeff Bonwick 2406e14bb325SJeff Bonwick if (zio->io_error) { 2407e14bb325SJeff Bonwick if (!vdev_accessible(vd, zio)) { 2408e14bb325SJeff Bonwick zio->io_error = ENXIO; 2409e14bb325SJeff Bonwick } else { 2410e14bb325SJeff Bonwick unexpected_error = B_TRUE; 2411e14bb325SJeff Bonwick } 2412e14bb325SJeff Bonwick } 241351ece835Seschrock } 2414fa9e4066Sahrens 2415e14bb325SJeff Bonwick ops->vdev_op_io_done(zio); 2416e14bb325SJeff Bonwick 2417e14bb325SJeff Bonwick if (unexpected_error) 2418a3f829aeSBill Moore VERIFY(vdev_probe(vd, zio) == NULL); 2419e14bb325SJeff Bonwick 2420e14bb325SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 2421fa9e4066Sahrens } 2422fa9e4066Sahrens 242322fe2c88SJonathan Adams /* 242422fe2c88SJonathan Adams * For non-raidz ZIOs, we can just copy aside the bad data read from the 242522fe2c88SJonathan Adams * disk, and use that to finish the checksum ereport later. 242622fe2c88SJonathan Adams */ 242722fe2c88SJonathan Adams static void 242822fe2c88SJonathan Adams zio_vsd_default_cksum_finish(zio_cksum_report_t *zcr, 242922fe2c88SJonathan Adams const void *good_buf) 243022fe2c88SJonathan Adams { 243122fe2c88SJonathan Adams /* no processing needed */ 243222fe2c88SJonathan Adams zfs_ereport_finish_checksum(zcr, good_buf, zcr->zcr_cbdata, B_FALSE); 243322fe2c88SJonathan Adams } 243422fe2c88SJonathan Adams 243522fe2c88SJonathan Adams /*ARGSUSED*/ 243622fe2c88SJonathan Adams void 243722fe2c88SJonathan Adams zio_vsd_default_cksum_report(zio_t *zio, zio_cksum_report_t *zcr, void *ignored) 243822fe2c88SJonathan Adams { 243922fe2c88SJonathan Adams void *buf = zio_buf_alloc(zio->io_size); 244022fe2c88SJonathan Adams 244122fe2c88SJonathan Adams bcopy(zio->io_data, buf, zio->io_size); 244222fe2c88SJonathan Adams 244322fe2c88SJonathan Adams zcr->zcr_cbinfo = zio->io_size; 244422fe2c88SJonathan Adams zcr->zcr_cbdata = buf; 244522fe2c88SJonathan Adams zcr->zcr_finish = zio_vsd_default_cksum_finish; 244622fe2c88SJonathan Adams zcr->zcr_free = zio_buf_free; 244722fe2c88SJonathan Adams } 244822fe2c88SJonathan Adams 2449e05725b1Sbonwick static int 2450fa9e4066Sahrens zio_vdev_io_assess(zio_t *zio) 2451fa9e4066Sahrens { 2452fa9e4066Sahrens vdev_t *vd = zio->io_vd; 2453e14bb325SJeff Bonwick 2454e14bb325SJeff Bonwick if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE)) 2455e14bb325SJeff Bonwick return (ZIO_PIPELINE_STOP); 2456e14bb325SJeff Bonwick 2457e14bb325SJeff Bonwick if (vd == NULL && !(zio->io_flags & ZIO_FLAG_CONFIG_WRITER)) 2458e14bb325SJeff Bonwick spa_config_exit(zio->io_spa, SCL_ZIO, zio); 2459e14bb325SJeff Bonwick 2460e14bb325SJeff Bonwick if (zio->io_vsd != NULL) { 246122fe2c88SJonathan Adams zio->io_vsd_ops->vsd_free(zio); 2462e14bb325SJeff Bonwick zio->io_vsd = NULL; 2463ecc2d604Sbonwick } 2464ecc2d604Sbonwick 2465e14bb325SJeff Bonwick if (zio_injection_enabled && zio->io_error == 0) 2466ea8dc4b6Seschrock zio->io_error = zio_handle_fault_injection(zio, EIO); 2467ea8dc4b6Seschrock 2468fa9e4066Sahrens /* 2469fa9e4066Sahrens * If the I/O failed, determine whether we should attempt to retry it. 247035a5a358SJonathan Adams * 247135a5a358SJonathan Adams * On retry, we cut in line in the issue queue, since we don't want 247235a5a358SJonathan Adams * compression/checksumming/etc. work to prevent our (cheap) IO reissue. 2473fa9e4066Sahrens */ 2474e14bb325SJeff Bonwick if (zio->io_error && vd == NULL && 2475e14bb325SJeff Bonwick !(zio->io_flags & (ZIO_FLAG_DONT_RETRY | ZIO_FLAG_IO_RETRY))) { 2476e14bb325SJeff Bonwick ASSERT(!(zio->io_flags & ZIO_FLAG_DONT_QUEUE)); /* not a leaf */ 2477e14bb325SJeff Bonwick ASSERT(!(zio->io_flags & ZIO_FLAG_IO_BYPASS)); /* not a leaf */ 2478fa9e4066Sahrens zio->io_error = 0; 2479e14bb325SJeff Bonwick zio->io_flags |= ZIO_FLAG_IO_RETRY | 2480e14bb325SJeff Bonwick ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE; 2481b24ab676SJeff Bonwick zio->io_stage = ZIO_STAGE_VDEV_IO_START >> 1; 248235a5a358SJonathan Adams zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, 248335a5a358SJonathan Adams zio_requeue_io_start_cut_in_line); 2484e14bb325SJeff Bonwick return (ZIO_PIPELINE_STOP); 2485ea8dc4b6Seschrock } 2486fa9e4066Sahrens 2487e14bb325SJeff Bonwick /* 2488e14bb325SJeff Bonwick * If we got an error on a leaf device, convert it to ENXIO 2489e14bb325SJeff Bonwick * if the device is not accessible at all. 2490e14bb325SJeff Bonwick */ 2491e14bb325SJeff Bonwick if (zio->io_error && vd != NULL && vd->vdev_ops->vdev_op_leaf && 2492e14bb325SJeff Bonwick !vdev_accessible(vd, zio)) 2493e14bb325SJeff Bonwick zio->io_error = ENXIO; 2494e14bb325SJeff Bonwick 2495e14bb325SJeff Bonwick /* 2496e14bb325SJeff Bonwick * If we can't write to an interior vdev (mirror or RAID-Z), 2497e14bb325SJeff Bonwick * set vdev_cant_write so that we stop trying to allocate from it. 2498e14bb325SJeff Bonwick */ 2499e14bb325SJeff Bonwick if (zio->io_error == ENXIO && zio->io_type == ZIO_TYPE_WRITE && 2500e14bb325SJeff Bonwick vd != NULL && !vd->vdev_ops->vdev_op_leaf) 2501e14bb325SJeff Bonwick vd->vdev_cant_write = B_TRUE; 2502e14bb325SJeff Bonwick 2503e14bb325SJeff Bonwick if (zio->io_error) 2504e14bb325SJeff Bonwick zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; 2505e14bb325SJeff Bonwick 2506e05725b1Sbonwick return (ZIO_PIPELINE_CONTINUE); 2507fa9e4066Sahrens } 2508fa9e4066Sahrens 2509fa9e4066Sahrens void 2510fa9e4066Sahrens zio_vdev_io_reissue(zio_t *zio) 2511fa9e4066Sahrens { 2512fa9e4066Sahrens ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START); 2513fa9e4066Sahrens ASSERT(zio->io_error == 0); 2514fa9e4066Sahrens 2515b24ab676SJeff Bonwick zio->io_stage >>= 1; 2516fa9e4066Sahrens } 2517fa9e4066Sahrens 2518fa9e4066Sahrens void 2519fa9e4066Sahrens zio_vdev_io_redone(zio_t *zio) 2520fa9e4066Sahrens { 2521fa9e4066Sahrens ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_DONE); 2522fa9e4066Sahrens 2523b24ab676SJeff Bonwick zio->io_stage >>= 1; 2524fa9e4066Sahrens } 2525fa9e4066Sahrens 2526fa9e4066Sahrens void 2527fa9e4066Sahrens zio_vdev_io_bypass(zio_t *zio) 2528fa9e4066Sahrens { 2529fa9e4066Sahrens ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START); 2530fa9e4066Sahrens ASSERT(zio->io_error == 0); 2531fa9e4066Sahrens 2532fa9e4066Sahrens zio->io_flags |= ZIO_FLAG_IO_BYPASS; 2533b24ab676SJeff Bonwick zio->io_stage = ZIO_STAGE_VDEV_IO_ASSESS >> 1; 2534fa9e4066Sahrens } 2535fa9e4066Sahrens 2536fa9e4066Sahrens /* 2537fa9e4066Sahrens * ========================================================================== 2538fa9e4066Sahrens * Generate and verify checksums 2539fa9e4066Sahrens * ========================================================================== 2540fa9e4066Sahrens */ 2541e05725b1Sbonwick static int 2542fa9e4066Sahrens zio_checksum_generate(zio_t *zio) 2543fa9e4066Sahrens { 2544fa9e4066Sahrens blkptr_t *bp = zio->io_bp; 2545e14bb325SJeff Bonwick enum zio_checksum checksum; 2546fa9e4066Sahrens 2547e14bb325SJeff Bonwick if (bp == NULL) { 2548e14bb325SJeff Bonwick /* 2549e14bb325SJeff Bonwick * This is zio_write_phys(). 2550e14bb325SJeff Bonwick * We're either generating a label checksum, or none at all. 2551e14bb325SJeff Bonwick */ 2552e14bb325SJeff Bonwick checksum = zio->io_prop.zp_checksum; 2553e14bb325SJeff Bonwick 2554e14bb325SJeff Bonwick if (checksum == ZIO_CHECKSUM_OFF) 2555e14bb325SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 2556fa9e4066Sahrens 2557e14bb325SJeff Bonwick ASSERT(checksum == ZIO_CHECKSUM_LABEL); 2558e14bb325SJeff Bonwick } else { 2559e14bb325SJeff Bonwick if (BP_IS_GANG(bp) && zio->io_child_type == ZIO_CHILD_GANG) { 2560e14bb325SJeff Bonwick ASSERT(!IO_IS_ALLOCATING(zio)); 2561e14bb325SJeff Bonwick checksum = ZIO_CHECKSUM_GANG_HEADER; 2562e14bb325SJeff Bonwick } else { 2563e14bb325SJeff Bonwick checksum = BP_GET_CHECKSUM(bp); 2564e14bb325SJeff Bonwick } 2565e14bb325SJeff Bonwick } 2566fa9e4066Sahrens 2567e14bb325SJeff Bonwick zio_checksum_compute(zio, checksum, zio->io_data, zio->io_size); 2568fa9e4066Sahrens 2569e05725b1Sbonwick return (ZIO_PIPELINE_CONTINUE); 2570fa9e4066Sahrens } 2571fa9e4066Sahrens 2572e05725b1Sbonwick static int 2573e14bb325SJeff Bonwick zio_checksum_verify(zio_t *zio) 2574fa9e4066Sahrens { 257522fe2c88SJonathan Adams zio_bad_cksum_t info; 2576e14bb325SJeff Bonwick blkptr_t *bp = zio->io_bp; 2577e14bb325SJeff Bonwick int error; 2578fa9e4066Sahrens 2579b24ab676SJeff Bonwick ASSERT(zio->io_vd != NULL); 2580b24ab676SJeff Bonwick 2581e14bb325SJeff Bonwick if (bp == NULL) { 2582e14bb325SJeff Bonwick /* 2583e14bb325SJeff Bonwick * This is zio_read_phys(). 2584e14bb325SJeff Bonwick * We're either verifying a label checksum, or nothing at all. 2585e14bb325SJeff Bonwick */ 2586e14bb325SJeff Bonwick if (zio->io_prop.zp_checksum == ZIO_CHECKSUM_OFF) 2587e14bb325SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 2588fa9e4066Sahrens 2589e14bb325SJeff Bonwick ASSERT(zio->io_prop.zp_checksum == ZIO_CHECKSUM_LABEL); 2590e14bb325SJeff Bonwick } 2591fa9e4066Sahrens 259222fe2c88SJonathan Adams if ((error = zio_checksum_error(zio, &info)) != 0) { 2593e14bb325SJeff Bonwick zio->io_error = error; 2594e14bb325SJeff Bonwick if (!(zio->io_flags & ZIO_FLAG_SPECULATIVE)) { 259522fe2c88SJonathan Adams zfs_ereport_start_checksum(zio->io_spa, 259622fe2c88SJonathan Adams zio->io_vd, zio, zio->io_offset, 259722fe2c88SJonathan Adams zio->io_size, NULL, &info); 2598e14bb325SJeff Bonwick } 2599fa9e4066Sahrens } 2600fa9e4066Sahrens 2601e05725b1Sbonwick return (ZIO_PIPELINE_CONTINUE); 2602fa9e4066Sahrens } 2603fa9e4066Sahrens 2604fa9e4066Sahrens /* 2605fa9e4066Sahrens * Called by RAID-Z to ensure we don't compute the checksum twice. 2606fa9e4066Sahrens */ 2607fa9e4066Sahrens void 2608fa9e4066Sahrens zio_checksum_verified(zio_t *zio) 2609fa9e4066Sahrens { 2610b24ab676SJeff Bonwick zio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY; 2611fa9e4066Sahrens } 2612fa9e4066Sahrens 2613fa9e4066Sahrens /* 2614e14bb325SJeff Bonwick * ========================================================================== 2615e14bb325SJeff Bonwick * Error rank. Error are ranked in the order 0, ENXIO, ECKSUM, EIO, other. 2616e14bb325SJeff Bonwick * An error of 0 indictes success. ENXIO indicates whole-device failure, 2617e14bb325SJeff Bonwick * which may be transient (e.g. unplugged) or permament. ECKSUM and EIO 2618e14bb325SJeff Bonwick * indicate errors that are specific to one I/O, and most likely permanent. 2619e14bb325SJeff Bonwick * Any other error is presumed to be worse because we weren't expecting it. 2620e14bb325SJeff Bonwick * ========================================================================== 2621fa9e4066Sahrens */ 2622e14bb325SJeff Bonwick int 2623e14bb325SJeff Bonwick zio_worst_error(int e1, int e2) 2624fa9e4066Sahrens { 2625e14bb325SJeff Bonwick static int zio_error_rank[] = { 0, ENXIO, ECKSUM, EIO }; 2626e14bb325SJeff Bonwick int r1, r2; 2627e14bb325SJeff Bonwick 2628e14bb325SJeff Bonwick for (r1 = 0; r1 < sizeof (zio_error_rank) / sizeof (int); r1++) 2629e14bb325SJeff Bonwick if (e1 == zio_error_rank[r1]) 2630e14bb325SJeff Bonwick break; 2631e14bb325SJeff Bonwick 2632e14bb325SJeff Bonwick for (r2 = 0; r2 < sizeof (zio_error_rank) / sizeof (int); r2++) 2633e14bb325SJeff Bonwick if (e2 == zio_error_rank[r2]) 2634e14bb325SJeff Bonwick break; 263544cd46caSbillm 2636e14bb325SJeff Bonwick return (r1 > r2 ? e1 : e2); 2637fa9e4066Sahrens } 2638fa9e4066Sahrens 2639fa9e4066Sahrens /* 2640fa9e4066Sahrens * ========================================================================== 2641e14bb325SJeff Bonwick * I/O completion 2642fa9e4066Sahrens * ========================================================================== 2643fa9e4066Sahrens */ 2644e14bb325SJeff Bonwick static int 2645e14bb325SJeff Bonwick zio_ready(zio_t *zio) 2646fa9e4066Sahrens { 2647e14bb325SJeff Bonwick blkptr_t *bp = zio->io_bp; 2648a3f829aeSBill Moore zio_t *pio, *pio_next; 2649fa9e4066Sahrens 2650b24ab676SJeff Bonwick if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY) || 2651b24ab676SJeff Bonwick zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_READY)) 2652f5383399SBill Moore return (ZIO_PIPELINE_STOP); 2653fa9e4066Sahrens 2654f5383399SBill Moore if (zio->io_ready) { 2655e14bb325SJeff Bonwick ASSERT(IO_IS_ALLOCATING(zio)); 2656e14bb325SJeff Bonwick ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp)); 2657e14bb325SJeff Bonwick ASSERT(zio->io_children[ZIO_CHILD_GANG][ZIO_WAIT_READY] == 0); 2658fa9e4066Sahrens 2659e14bb325SJeff Bonwick zio->io_ready(zio); 2660e14bb325SJeff Bonwick } 2661fa9e4066Sahrens 2662e14bb325SJeff Bonwick if (bp != NULL && bp != &zio->io_bp_copy) 2663e14bb325SJeff Bonwick zio->io_bp_copy = *bp; 2664fa9e4066Sahrens 2665e14bb325SJeff Bonwick if (zio->io_error) 2666e14bb325SJeff Bonwick zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; 2667fa9e4066Sahrens 2668a3f829aeSBill Moore mutex_enter(&zio->io_lock); 2669a3f829aeSBill Moore zio->io_state[ZIO_WAIT_READY] = 1; 2670a3f829aeSBill Moore pio = zio_walk_parents(zio); 2671a3f829aeSBill Moore mutex_exit(&zio->io_lock); 2672a3f829aeSBill Moore 2673a3f829aeSBill Moore /* 2674a3f829aeSBill Moore * As we notify zio's parents, new parents could be added. 2675a3f829aeSBill Moore * New parents go to the head of zio's io_parent_list, however, 2676a3f829aeSBill Moore * so we will (correctly) not notify them. The remainder of zio's 2677a3f829aeSBill Moore * io_parent_list, from 'pio_next' onward, cannot change because 2678a3f829aeSBill Moore * all parents must wait for us to be done before they can be done. 2679a3f829aeSBill Moore */ 2680a3f829aeSBill Moore for (; pio != NULL; pio = pio_next) { 2681a3f829aeSBill Moore pio_next = zio_walk_parents(zio); 2682e14bb325SJeff Bonwick zio_notify_parent(pio, zio, ZIO_WAIT_READY); 2683a3f829aeSBill Moore } 2684fa9e4066Sahrens 2685b24ab676SJeff Bonwick if (zio->io_flags & ZIO_FLAG_NODATA) { 2686b24ab676SJeff Bonwick if (BP_IS_GANG(bp)) { 2687b24ab676SJeff Bonwick zio->io_flags &= ~ZIO_FLAG_NODATA; 2688b24ab676SJeff Bonwick } else { 2689b24ab676SJeff Bonwick ASSERT((uintptr_t)zio->io_data < SPA_MAXBLOCKSIZE); 2690b24ab676SJeff Bonwick zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES; 2691b24ab676SJeff Bonwick } 2692b24ab676SJeff Bonwick } 2693b24ab676SJeff Bonwick 2694a33cae98STim Haley if (zio_injection_enabled && 2695a33cae98STim Haley zio->io_spa->spa_syncing_txg == zio->io_txg) 2696a33cae98STim Haley zio_handle_ignored_writes(zio); 2697a33cae98STim Haley 2698e14bb325SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 2699fa9e4066Sahrens } 2700fa9e4066Sahrens 2701e14bb325SJeff Bonwick static int 2702e14bb325SJeff Bonwick zio_done(zio_t *zio) 2703d63d470bSgw { 2704e14bb325SJeff Bonwick spa_t *spa = zio->io_spa; 2705e14bb325SJeff Bonwick zio_t *lio = zio->io_logical; 2706e14bb325SJeff Bonwick blkptr_t *bp = zio->io_bp; 2707e14bb325SJeff Bonwick vdev_t *vd = zio->io_vd; 2708e14bb325SJeff Bonwick uint64_t psize = zio->io_size; 2709a3f829aeSBill Moore zio_t *pio, *pio_next; 2710d63d470bSgw 2711e14bb325SJeff Bonwick /* 2712f5383399SBill Moore * If our children haven't all completed, 2713e14bb325SJeff Bonwick * wait for them and then repeat this pipeline stage. 2714e14bb325SJeff Bonwick */ 2715e14bb325SJeff Bonwick if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE) || 2716e14bb325SJeff Bonwick zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE) || 2717b24ab676SJeff Bonwick zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_DONE) || 2718e14bb325SJeff Bonwick zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_DONE)) 2719e14bb325SJeff Bonwick return (ZIO_PIPELINE_STOP); 2720d63d470bSgw 2721e14bb325SJeff Bonwick for (int c = 0; c < ZIO_CHILD_TYPES; c++) 2722e14bb325SJeff Bonwick for (int w = 0; w < ZIO_WAIT_TYPES; w++) 2723e14bb325SJeff Bonwick ASSERT(zio->io_children[c][w] == 0); 2724e14bb325SJeff Bonwick 2725e14bb325SJeff Bonwick if (bp != NULL) { 2726e14bb325SJeff Bonwick ASSERT(bp->blk_pad[0] == 0); 2727e14bb325SJeff Bonwick ASSERT(bp->blk_pad[1] == 0); 2728e14bb325SJeff Bonwick ASSERT(bcmp(bp, &zio->io_bp_copy, sizeof (blkptr_t)) == 0 || 2729a3f829aeSBill Moore (bp == zio_unique_parent(zio)->io_bp)); 2730e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_WRITE && !BP_IS_HOLE(bp) && 2731b24ab676SJeff Bonwick zio->io_bp_override == NULL && 2732e14bb325SJeff Bonwick !(zio->io_flags & ZIO_FLAG_IO_REPAIR)) { 2733e14bb325SJeff Bonwick ASSERT(!BP_SHOULD_BYTESWAP(bp)); 2734b24ab676SJeff Bonwick ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(bp)); 2735e14bb325SJeff Bonwick ASSERT(BP_COUNT_GANG(bp) == 0 || 2736e14bb325SJeff Bonwick (BP_COUNT_GANG(bp) == BP_GET_NDVAS(bp))); 2737e14bb325SJeff Bonwick } 2738e14bb325SJeff Bonwick } 2739fa9e4066Sahrens 2740e14bb325SJeff Bonwick /* 2741b24ab676SJeff Bonwick * If there were child vdev/gang/ddt errors, they apply to us now. 2742e14bb325SJeff Bonwick */ 2743e14bb325SJeff Bonwick zio_inherit_child_errors(zio, ZIO_CHILD_VDEV); 2744e14bb325SJeff Bonwick zio_inherit_child_errors(zio, ZIO_CHILD_GANG); 2745b24ab676SJeff Bonwick zio_inherit_child_errors(zio, ZIO_CHILD_DDT); 2746b24ab676SJeff Bonwick 2747b24ab676SJeff Bonwick /* 2748b24ab676SJeff Bonwick * If the I/O on the transformed data was successful, generate any 2749b24ab676SJeff Bonwick * checksum reports now while we still have the transformed data. 2750b24ab676SJeff Bonwick */ 2751b24ab676SJeff Bonwick if (zio->io_error == 0) { 2752b24ab676SJeff Bonwick while (zio->io_cksum_report != NULL) { 2753b24ab676SJeff Bonwick zio_cksum_report_t *zcr = zio->io_cksum_report; 2754b24ab676SJeff Bonwick uint64_t align = zcr->zcr_align; 2755b24ab676SJeff Bonwick uint64_t asize = P2ROUNDUP(psize, align); 2756b24ab676SJeff Bonwick char *abuf = zio->io_data; 2757b24ab676SJeff Bonwick 2758b24ab676SJeff Bonwick if (asize != psize) { 2759b24ab676SJeff Bonwick abuf = zio_buf_alloc(asize); 2760b24ab676SJeff Bonwick bcopy(zio->io_data, abuf, psize); 2761b24ab676SJeff Bonwick bzero(abuf + psize, asize - psize); 2762b24ab676SJeff Bonwick } 2763b24ab676SJeff Bonwick 2764b24ab676SJeff Bonwick zio->io_cksum_report = zcr->zcr_next; 2765b24ab676SJeff Bonwick zcr->zcr_next = NULL; 2766b24ab676SJeff Bonwick zcr->zcr_finish(zcr, abuf); 2767b24ab676SJeff Bonwick zfs_ereport_free_checksum(zcr); 2768b24ab676SJeff Bonwick 2769b24ab676SJeff Bonwick if (asize != psize) 2770b24ab676SJeff Bonwick zio_buf_free(abuf, asize); 2771b24ab676SJeff Bonwick } 2772b24ab676SJeff Bonwick } 2773e14bb325SJeff Bonwick 2774e14bb325SJeff Bonwick zio_pop_transforms(zio); /* note: may set zio->io_error */ 2775e14bb325SJeff Bonwick 2776e14bb325SJeff Bonwick vdev_stat_update(zio, psize); 2777e14bb325SJeff Bonwick 2778e14bb325SJeff Bonwick if (zio->io_error) { 2779e14bb325SJeff Bonwick /* 2780e14bb325SJeff Bonwick * If this I/O is attached to a particular vdev, 2781e14bb325SJeff Bonwick * generate an error message describing the I/O failure 2782e14bb325SJeff Bonwick * at the block level. We ignore these errors if the 2783e14bb325SJeff Bonwick * device is currently unavailable. 2784e14bb325SJeff Bonwick */ 2785e14bb325SJeff Bonwick if (zio->io_error != ECKSUM && vd != NULL && !vdev_is_dead(vd)) 2786e14bb325SJeff Bonwick zfs_ereport_post(FM_EREPORT_ZFS_IO, spa, vd, zio, 0, 0); 2787e14bb325SJeff Bonwick 27888f18d1faSGeorge Wilson if ((zio->io_error == EIO || !(zio->io_flags & 27898f18d1faSGeorge Wilson (ZIO_FLAG_SPECULATIVE | ZIO_FLAG_DONT_PROPAGATE))) && 27908f18d1faSGeorge Wilson zio == lio) { 2791e14bb325SJeff Bonwick /* 2792e14bb325SJeff Bonwick * For logical I/O requests, tell the SPA to log the 2793e14bb325SJeff Bonwick * error and generate a logical data ereport. 2794e14bb325SJeff Bonwick */ 2795e14bb325SJeff Bonwick spa_log_error(spa, zio); 2796e14bb325SJeff Bonwick zfs_ereport_post(FM_EREPORT_ZFS_DATA, spa, NULL, zio, 2797e14bb325SJeff Bonwick 0, 0); 2798e14bb325SJeff Bonwick } 2799e14bb325SJeff Bonwick } 2800fa9e4066Sahrens 2801e14bb325SJeff Bonwick if (zio->io_error && zio == lio) { 2802e14bb325SJeff Bonwick /* 2803e14bb325SJeff Bonwick * Determine whether zio should be reexecuted. This will 2804e14bb325SJeff Bonwick * propagate all the way to the root via zio_notify_parent(). 2805e14bb325SJeff Bonwick */ 2806e14bb325SJeff Bonwick ASSERT(vd == NULL && bp != NULL); 2807b24ab676SJeff Bonwick ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 2808e14bb325SJeff Bonwick 2809b24ab676SJeff Bonwick if (IO_IS_ALLOCATING(zio) && 2810b24ab676SJeff Bonwick !(zio->io_flags & ZIO_FLAG_CANFAIL)) { 2811e14bb325SJeff Bonwick if (zio->io_error != ENOSPC) 2812e14bb325SJeff Bonwick zio->io_reexecute |= ZIO_REEXECUTE_NOW; 2813e14bb325SJeff Bonwick else 2814e14bb325SJeff Bonwick zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; 2815b24ab676SJeff Bonwick } 2816e14bb325SJeff Bonwick 2817e14bb325SJeff Bonwick if ((zio->io_type == ZIO_TYPE_READ || 2818e14bb325SJeff Bonwick zio->io_type == ZIO_TYPE_FREE) && 281944ecc532SGeorge Wilson !(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && 2820e14bb325SJeff Bonwick zio->io_error == ENXIO && 2821b16da2e2SGeorge Wilson spa_load_state(spa) == SPA_LOAD_NONE && 2822e14bb325SJeff Bonwick spa_get_failmode(spa) != ZIO_FAILURE_MODE_CONTINUE) 2823e14bb325SJeff Bonwick zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; 2824e14bb325SJeff Bonwick 2825e14bb325SJeff Bonwick if (!(zio->io_flags & ZIO_FLAG_CANFAIL) && !zio->io_reexecute) 2826e14bb325SJeff Bonwick zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; 282722fe2c88SJonathan Adams 282822fe2c88SJonathan Adams /* 282922fe2c88SJonathan Adams * Here is a possibly good place to attempt to do 283022fe2c88SJonathan Adams * either combinatorial reconstruction or error correction 283122fe2c88SJonathan Adams * based on checksums. It also might be a good place 283222fe2c88SJonathan Adams * to send out preliminary ereports before we suspend 283322fe2c88SJonathan Adams * processing. 283422fe2c88SJonathan Adams */ 2835d63d470bSgw } 2836d63d470bSgw 283767bd71c6Sperrin /* 2838e14bb325SJeff Bonwick * If there were logical child errors, they apply to us now. 2839e14bb325SJeff Bonwick * We defer this until now to avoid conflating logical child 2840e14bb325SJeff Bonwick * errors with errors that happened to the zio itself when 2841e14bb325SJeff Bonwick * updating vdev stats and reporting FMA events above. 284267bd71c6Sperrin */ 2843e14bb325SJeff Bonwick zio_inherit_child_errors(zio, ZIO_CHILD_LOGICAL); 28448654d025Sperrin 2845b24ab676SJeff Bonwick if ((zio->io_error || zio->io_reexecute) && 2846b24ab676SJeff Bonwick IO_IS_ALLOCATING(zio) && zio->io_gang_leader == zio && 2847b24ab676SJeff Bonwick !(zio->io_flags & ZIO_FLAG_IO_REWRITE)) 2848f5383399SBill Moore zio_dva_unallocate(zio, zio->io_gang_tree, bp); 2849f5383399SBill Moore 2850f5383399SBill Moore zio_gang_tree_free(&zio->io_gang_tree); 2851f5383399SBill Moore 285233a372edSGeorge Wilson /* 285333a372edSGeorge Wilson * Godfather I/Os should never suspend. 285433a372edSGeorge Wilson */ 285533a372edSGeorge Wilson if ((zio->io_flags & ZIO_FLAG_GODFATHER) && 285633a372edSGeorge Wilson (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) 285733a372edSGeorge Wilson zio->io_reexecute = 0; 285833a372edSGeorge Wilson 285933a372edSGeorge Wilson if (zio->io_reexecute) { 2860e14bb325SJeff Bonwick /* 2861e14bb325SJeff Bonwick * This is a logical I/O that wants to reexecute. 2862e14bb325SJeff Bonwick * 2863e14bb325SJeff Bonwick * Reexecute is top-down. When an i/o fails, if it's not 2864e14bb325SJeff Bonwick * the root, it simply notifies its parent and sticks around. 2865e14bb325SJeff Bonwick * The parent, seeing that it still has children in zio_done(), 2866e14bb325SJeff Bonwick * does the same. This percolates all the way up to the root. 2867e14bb325SJeff Bonwick * The root i/o will reexecute or suspend the entire tree. 2868e14bb325SJeff Bonwick * 2869e14bb325SJeff Bonwick * This approach ensures that zio_reexecute() honors 2870e14bb325SJeff Bonwick * all the original i/o dependency relationships, e.g. 2871e14bb325SJeff Bonwick * parents not executing until children are ready. 2872e14bb325SJeff Bonwick */ 2873e14bb325SJeff Bonwick ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 2874fa9e4066Sahrens 2875f5383399SBill Moore zio->io_gang_leader = NULL; 2876e14bb325SJeff Bonwick 2877a3f829aeSBill Moore mutex_enter(&zio->io_lock); 2878a3f829aeSBill Moore zio->io_state[ZIO_WAIT_DONE] = 1; 2879a3f829aeSBill Moore mutex_exit(&zio->io_lock); 2880a3f829aeSBill Moore 288154d692b7SGeorge Wilson /* 288254d692b7SGeorge Wilson * "The Godfather" I/O monitors its children but is 288354d692b7SGeorge Wilson * not a true parent to them. It will track them through 288454d692b7SGeorge Wilson * the pipeline but severs its ties whenever they get into 288554d692b7SGeorge Wilson * trouble (e.g. suspended). This allows "The Godfather" 288654d692b7SGeorge Wilson * I/O to return status without blocking. 288754d692b7SGeorge Wilson */ 288854d692b7SGeorge Wilson for (pio = zio_walk_parents(zio); pio != NULL; pio = pio_next) { 288954d692b7SGeorge Wilson zio_link_t *zl = zio->io_walk_link; 289054d692b7SGeorge Wilson pio_next = zio_walk_parents(zio); 289154d692b7SGeorge Wilson 289254d692b7SGeorge Wilson if ((pio->io_flags & ZIO_FLAG_GODFATHER) && 289354d692b7SGeorge Wilson (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) { 289454d692b7SGeorge Wilson zio_remove_child(pio, zio, zl); 289554d692b7SGeorge Wilson zio_notify_parent(pio, zio, ZIO_WAIT_DONE); 289654d692b7SGeorge Wilson } 289754d692b7SGeorge Wilson } 289854d692b7SGeorge Wilson 2899a3f829aeSBill Moore if ((pio = zio_unique_parent(zio)) != NULL) { 2900e14bb325SJeff Bonwick /* 2901e14bb325SJeff Bonwick * We're not a root i/o, so there's nothing to do 2902e14bb325SJeff Bonwick * but notify our parent. Don't propagate errors 2903e14bb325SJeff Bonwick * upward since we haven't permanently failed yet. 2904e14bb325SJeff Bonwick */ 290533a372edSGeorge Wilson ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER)); 2906e14bb325SJeff Bonwick zio->io_flags |= ZIO_FLAG_DONT_PROPAGATE; 2907e14bb325SJeff Bonwick zio_notify_parent(pio, zio, ZIO_WAIT_DONE); 2908e14bb325SJeff Bonwick } else if (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND) { 2909e14bb325SJeff Bonwick /* 2910e14bb325SJeff Bonwick * We'd fail again if we reexecuted now, so suspend 2911e14bb325SJeff Bonwick * until conditions improve (e.g. device comes online). 2912e14bb325SJeff Bonwick */ 2913e14bb325SJeff Bonwick zio_suspend(spa, zio); 2914e14bb325SJeff Bonwick } else { 2915e14bb325SJeff Bonwick /* 2916e14bb325SJeff Bonwick * Reexecution is potentially a huge amount of work. 2917e14bb325SJeff Bonwick * Hand it off to the otherwise-unused claim taskq. 2918e14bb325SJeff Bonwick */ 29195aeb9474SGarrett D'Amore ASSERT(zio->io_tqent.tqent_next == NULL); 29205aeb9474SGarrett D'Amore (void) taskq_dispatch_ent( 2921e14bb325SJeff Bonwick spa->spa_zio_taskq[ZIO_TYPE_CLAIM][ZIO_TASKQ_ISSUE], 29225aeb9474SGarrett D'Amore (task_func_t *)zio_reexecute, zio, 0, 29235aeb9474SGarrett D'Amore &zio->io_tqent); 2924e14bb325SJeff Bonwick } 2925e14bb325SJeff Bonwick return (ZIO_PIPELINE_STOP); 2926fa9e4066Sahrens } 2927fa9e4066Sahrens 2928b24ab676SJeff Bonwick ASSERT(zio->io_child_count == 0); 292933a372edSGeorge Wilson ASSERT(zio->io_reexecute == 0); 2930e14bb325SJeff Bonwick ASSERT(zio->io_error == 0 || (zio->io_flags & ZIO_FLAG_CANFAIL)); 2931fa9e4066Sahrens 2932b24ab676SJeff Bonwick /* 2933b24ab676SJeff Bonwick * Report any checksum errors, since the I/O is complete. 2934b24ab676SJeff Bonwick */ 293522fe2c88SJonathan Adams while (zio->io_cksum_report != NULL) { 2936b24ab676SJeff Bonwick zio_cksum_report_t *zcr = zio->io_cksum_report; 2937b24ab676SJeff Bonwick zio->io_cksum_report = zcr->zcr_next; 2938b24ab676SJeff Bonwick zcr->zcr_next = NULL; 2939b24ab676SJeff Bonwick zcr->zcr_finish(zcr, NULL); 2940b24ab676SJeff Bonwick zfs_ereport_free_checksum(zcr); 294122fe2c88SJonathan Adams } 294222fe2c88SJonathan Adams 2943a3f829aeSBill Moore /* 2944a3f829aeSBill Moore * It is the responsibility of the done callback to ensure that this 2945a3f829aeSBill Moore * particular zio is no longer discoverable for adoption, and as 2946a3f829aeSBill Moore * such, cannot acquire any new parents. 2947a3f829aeSBill Moore */ 2948e14bb325SJeff Bonwick if (zio->io_done) 2949e14bb325SJeff Bonwick zio->io_done(zio); 2950fa9e4066Sahrens 2951a3f829aeSBill Moore mutex_enter(&zio->io_lock); 2952a3f829aeSBill Moore zio->io_state[ZIO_WAIT_DONE] = 1; 2953a3f829aeSBill Moore mutex_exit(&zio->io_lock); 2954fa9e4066Sahrens 2955a3f829aeSBill Moore for (pio = zio_walk_parents(zio); pio != NULL; pio = pio_next) { 2956a3f829aeSBill Moore zio_link_t *zl = zio->io_walk_link; 2957a3f829aeSBill Moore pio_next = zio_walk_parents(zio); 2958a3f829aeSBill Moore zio_remove_child(pio, zio, zl); 2959e14bb325SJeff Bonwick zio_notify_parent(pio, zio, ZIO_WAIT_DONE); 2960e14bb325SJeff Bonwick } 2961fa9e4066Sahrens 2962e14bb325SJeff Bonwick if (zio->io_waiter != NULL) { 2963e14bb325SJeff Bonwick mutex_enter(&zio->io_lock); 2964e14bb325SJeff Bonwick zio->io_executor = NULL; 2965e14bb325SJeff Bonwick cv_broadcast(&zio->io_cv); 2966e14bb325SJeff Bonwick mutex_exit(&zio->io_lock); 2967e14bb325SJeff Bonwick } else { 2968e14bb325SJeff Bonwick zio_destroy(zio); 2969e14bb325SJeff Bonwick } 2970fa9e4066Sahrens 2971e14bb325SJeff Bonwick return (ZIO_PIPELINE_STOP); 2972fa9e4066Sahrens } 297346341222Sperrin 297446341222Sperrin /* 2975e14bb325SJeff Bonwick * ========================================================================== 2976e14bb325SJeff Bonwick * I/O pipeline definition 2977e14bb325SJeff Bonwick * ========================================================================== 297846341222Sperrin */ 2979b24ab676SJeff Bonwick static zio_pipe_stage_t *zio_pipeline[] = { 2980e14bb325SJeff Bonwick NULL, 2981e14bb325SJeff Bonwick zio_read_bp_init, 2982b24ab676SJeff Bonwick zio_free_bp_init, 2983b24ab676SJeff Bonwick zio_issue_async, 2984e14bb325SJeff Bonwick zio_write_bp_init, 2985e14bb325SJeff Bonwick zio_checksum_generate, 2986b24ab676SJeff Bonwick zio_ddt_read_start, 2987b24ab676SJeff Bonwick zio_ddt_read_done, 2988b24ab676SJeff Bonwick zio_ddt_write, 2989b24ab676SJeff Bonwick zio_ddt_free, 2990e14bb325SJeff Bonwick zio_gang_assemble, 2991e14bb325SJeff Bonwick zio_gang_issue, 2992e14bb325SJeff Bonwick zio_dva_allocate, 2993e14bb325SJeff Bonwick zio_dva_free, 2994e14bb325SJeff Bonwick zio_dva_claim, 2995e14bb325SJeff Bonwick zio_ready, 2996e14bb325SJeff Bonwick zio_vdev_io_start, 2997e14bb325SJeff Bonwick zio_vdev_io_done, 2998e14bb325SJeff Bonwick zio_vdev_io_assess, 2999e14bb325SJeff Bonwick zio_checksum_verify, 3000e14bb325SJeff Bonwick zio_done 3001e14bb325SJeff Bonwick }; 3002ad135b5dSChristopher Siden 3003ad135b5dSChristopher Siden /* dnp is the dnode for zb1->zb_object */ 3004ad135b5dSChristopher Siden boolean_t 3005ad135b5dSChristopher Siden zbookmark_is_before(const dnode_phys_t *dnp, const zbookmark_t *zb1, 3006ad135b5dSChristopher Siden const zbookmark_t *zb2) 3007ad135b5dSChristopher Siden { 3008ad135b5dSChristopher Siden uint64_t zb1nextL0, zb2thisobj; 3009ad135b5dSChristopher Siden 3010ad135b5dSChristopher Siden ASSERT(zb1->zb_objset == zb2->zb_objset); 3011ad135b5dSChristopher Siden ASSERT(zb2->zb_level == 0); 3012ad135b5dSChristopher Siden 3013ad135b5dSChristopher Siden /* 3014ad135b5dSChristopher Siden * A bookmark in the deadlist is considered to be after 3015ad135b5dSChristopher Siden * everything else. 3016ad135b5dSChristopher Siden */ 3017ad135b5dSChristopher Siden if (zb2->zb_object == DMU_DEADLIST_OBJECT) 3018ad135b5dSChristopher Siden return (B_TRUE); 3019ad135b5dSChristopher Siden 3020ad135b5dSChristopher Siden /* The objset_phys_t isn't before anything. */ 3021ad135b5dSChristopher Siden if (dnp == NULL) 3022ad135b5dSChristopher Siden return (B_FALSE); 3023ad135b5dSChristopher Siden 3024ad135b5dSChristopher Siden zb1nextL0 = (zb1->zb_blkid + 1) << 3025ad135b5dSChristopher Siden ((zb1->zb_level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)); 3026ad135b5dSChristopher Siden 3027ad135b5dSChristopher Siden zb2thisobj = zb2->zb_object ? zb2->zb_object : 3028ad135b5dSChristopher Siden zb2->zb_blkid << (DNODE_BLOCK_SHIFT - DNODE_SHIFT); 3029ad135b5dSChristopher Siden 3030ad135b5dSChristopher Siden if (zb1->zb_object == DMU_META_DNODE_OBJECT) { 3031ad135b5dSChristopher Siden uint64_t nextobj = zb1nextL0 * 3032ad135b5dSChristopher Siden (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT) >> DNODE_SHIFT; 3033ad135b5dSChristopher Siden return (nextobj <= zb2thisobj); 3034ad135b5dSChristopher Siden } 3035ad135b5dSChristopher Siden 3036ad135b5dSChristopher Siden if (zb1->zb_object < zb2thisobj) 3037ad135b5dSChristopher Siden return (B_TRUE); 3038ad135b5dSChristopher Siden if (zb1->zb_object > zb2thisobj) 3039ad135b5dSChristopher Siden return (B_FALSE); 3040ad135b5dSChristopher Siden if (zb2->zb_object == DMU_META_DNODE_OBJECT) 3041ad135b5dSChristopher Siden return (B_FALSE); 3042ad135b5dSChristopher Siden return (zb1nextL0 <= zb2->zb_blkid); 3043ad135b5dSChristopher Siden } 3044