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. 2381cd5c55SMatthew Ahrens * Copyright (c) 2011, 2015 by Delphix. All rights reserved. 245aeb9474SGarrett D'Amore * Copyright (c) 2011 Nexenta Systems, Inc. All rights reserved. 25fa9e4066Sahrens */ 26fa9e4066Sahrens 27de710d24SJosef 'Jeff' Sipek #include <sys/sysmacros.h> 28fa9e4066Sahrens #include <sys/zfs_context.h> 29ea8dc4b6Seschrock #include <sys/fm/fs/zfs.h> 30fa9e4066Sahrens #include <sys/spa.h> 31fa9e4066Sahrens #include <sys/txg.h> 32fa9e4066Sahrens #include <sys/spa_impl.h> 33fa9e4066Sahrens #include <sys/vdev_impl.h> 34fa9e4066Sahrens #include <sys/zio_impl.h> 35fa9e4066Sahrens #include <sys/zio_compress.h> 36fa9e4066Sahrens #include <sys/zio_checksum.h> 37b24ab676SJeff Bonwick #include <sys/dmu_objset.h> 38b24ab676SJeff Bonwick #include <sys/arc.h> 39b24ab676SJeff Bonwick #include <sys/ddt.h> 405d7b4d43SMatthew Ahrens #include <sys/blkptr.h> 4143466aaeSMax Grossman #include <sys/zfeature.h> 42fa9e4066Sahrens 43fa9e4066Sahrens /* 44fa9e4066Sahrens * ========================================================================== 45fa9e4066Sahrens * I/O type descriptions 46fa9e4066Sahrens * ========================================================================== 47fa9e4066Sahrens */ 4869962b56SMatthew Ahrens const char *zio_type_name[ZIO_TYPES] = { 4980eb36f2SGeorge Wilson "zio_null", "zio_read", "zio_write", "zio_free", "zio_claim", 5080eb36f2SGeorge Wilson "zio_ioctl" 5180eb36f2SGeorge Wilson }; 52fa9e4066Sahrens 53fa9e4066Sahrens /* 54fa9e4066Sahrens * ========================================================================== 55fa9e4066Sahrens * I/O kmem caches 56fa9e4066Sahrens * ========================================================================== 57fa9e4066Sahrens */ 58ccae0b50Seschrock kmem_cache_t *zio_cache; 59a3f829aeSBill Moore kmem_cache_t *zio_link_cache; 60fa9e4066Sahrens kmem_cache_t *zio_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT]; 61ad23a2dbSjohansen kmem_cache_t *zio_data_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT]; 62ad23a2dbSjohansen 63ad23a2dbSjohansen #ifdef _KERNEL 64ad23a2dbSjohansen extern vmem_t *zio_alloc_arena; 65ad23a2dbSjohansen #endif 66fa9e4066Sahrens 67738f37bcSGeorge Wilson #define ZIO_PIPELINE_CONTINUE 0x100 68738f37bcSGeorge Wilson #define ZIO_PIPELINE_STOP 0x101 69738f37bcSGeorge Wilson 70*a2cdcdd2SPaul Dagnelie #define BP_SPANB(indblkshift, level) \ 71*a2cdcdd2SPaul Dagnelie (((uint64_t)1) << ((level) * ((indblkshift) - SPA_BLKPTRSHIFT))) 72*a2cdcdd2SPaul Dagnelie #define COMPARE_META_LEVEL 0x80000000ul 7301f55e48SGeorge Wilson /* 7401f55e48SGeorge Wilson * The following actions directly effect the spa's sync-to-convergence logic. 7501f55e48SGeorge Wilson * The values below define the sync pass when we start performing the action. 7601f55e48SGeorge Wilson * Care should be taken when changing these values as they directly impact 7701f55e48SGeorge Wilson * spa_sync() performance. Tuning these values may introduce subtle performance 7801f55e48SGeorge Wilson * pathologies and should only be done in the context of performance analysis. 7901f55e48SGeorge Wilson * These tunables will eventually be removed and replaced with #defines once 8001f55e48SGeorge Wilson * enough analysis has been done to determine optimal values. 8101f55e48SGeorge Wilson * 8201f55e48SGeorge Wilson * The 'zfs_sync_pass_deferred_free' pass must be greater than 1 to ensure that 8301f55e48SGeorge Wilson * regular blocks are not deferred. 8401f55e48SGeorge Wilson */ 8501f55e48SGeorge Wilson int zfs_sync_pass_deferred_free = 2; /* defer frees starting in this pass */ 8601f55e48SGeorge Wilson int zfs_sync_pass_dont_compress = 5; /* don't compress starting in this pass */ 8701f55e48SGeorge Wilson int zfs_sync_pass_rewrite = 2; /* rewrite new bps starting in this pass */ 8801f55e48SGeorge Wilson 890a4e9518Sgw /* 90e14bb325SJeff Bonwick * An allocating zio is one that either currently has the DVA allocate 91e14bb325SJeff Bonwick * stage set or will have it later in its lifetime. 920a4e9518Sgw */ 93b24ab676SJeff Bonwick #define IO_IS_ALLOCATING(zio) ((zio)->io_orig_pipeline & ZIO_STAGE_DVA_ALLOCATE) 94b24ab676SJeff Bonwick 9535a5a358SJonathan Adams boolean_t zio_requeue_io_start_cut_in_line = B_TRUE; 9635a5a358SJonathan Adams 97b24ab676SJeff Bonwick #ifdef ZFS_DEBUG 98b24ab676SJeff Bonwick int zio_buf_debug_limit = 16384; 99b24ab676SJeff Bonwick #else 100b24ab676SJeff Bonwick int zio_buf_debug_limit = 0; 101b24ab676SJeff Bonwick #endif 1020a4e9518Sgw 103fa9e4066Sahrens void 104fa9e4066Sahrens zio_init(void) 105fa9e4066Sahrens { 106fa9e4066Sahrens size_t c; 107ad23a2dbSjohansen vmem_t *data_alloc_arena = NULL; 108ad23a2dbSjohansen 109ad23a2dbSjohansen #ifdef _KERNEL 110ad23a2dbSjohansen data_alloc_arena = zio_alloc_arena; 111ad23a2dbSjohansen #endif 112a3f829aeSBill Moore zio_cache = kmem_cache_create("zio_cache", 113a3f829aeSBill Moore sizeof (zio_t), 0, NULL, NULL, NULL, NULL, NULL, 0); 114a3f829aeSBill Moore zio_link_cache = kmem_cache_create("zio_link_cache", 115a3f829aeSBill Moore sizeof (zio_link_t), 0, NULL, NULL, NULL, NULL, NULL, 0); 116ccae0b50Seschrock 117fa9e4066Sahrens /* 118fa9e4066Sahrens * For small buffers, we want a cache for each multiple of 119b5152584SMatthew Ahrens * SPA_MINBLOCKSIZE. For larger buffers, we want a cache 120b5152584SMatthew Ahrens * for each quarter-power of 2. 121fa9e4066Sahrens */ 122fa9e4066Sahrens for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) { 123fa9e4066Sahrens size_t size = (c + 1) << SPA_MINBLOCKSHIFT; 124fa9e4066Sahrens size_t p2 = size; 125fa9e4066Sahrens size_t align = 0; 126e291592aSJonathan Adams size_t cflags = (size > zio_buf_debug_limit) ? KMC_NODEBUG : 0; 127fa9e4066Sahrens 128de710d24SJosef 'Jeff' Sipek while (!ISP2(p2)) 129fa9e4066Sahrens p2 &= p2 - 1; 130fa9e4066Sahrens 131cd1c8b85SMatthew Ahrens #ifndef _KERNEL 132cd1c8b85SMatthew Ahrens /* 133cd1c8b85SMatthew Ahrens * If we are using watchpoints, put each buffer on its own page, 134cd1c8b85SMatthew Ahrens * to eliminate the performance overhead of trapping to the 135cd1c8b85SMatthew Ahrens * kernel when modifying a non-watched buffer that shares the 136cd1c8b85SMatthew Ahrens * page with a watched buffer. 137cd1c8b85SMatthew Ahrens */ 138cd1c8b85SMatthew Ahrens if (arc_watch && !IS_P2ALIGNED(size, PAGESIZE)) 139cd1c8b85SMatthew Ahrens continue; 140cd1c8b85SMatthew Ahrens #endif 141fa9e4066Sahrens if (size <= 4 * SPA_MINBLOCKSIZE) { 142fa9e4066Sahrens align = SPA_MINBLOCKSIZE; 143cd1c8b85SMatthew Ahrens } else if (IS_P2ALIGNED(size, p2 >> 2)) { 144b5152584SMatthew Ahrens align = MIN(p2 >> 2, PAGESIZE); 145fa9e4066Sahrens } 146fa9e4066Sahrens 147fa9e4066Sahrens if (align != 0) { 148ad23a2dbSjohansen char name[36]; 1495ad82045Snd (void) sprintf(name, "zio_buf_%lu", (ulong_t)size); 150fa9e4066Sahrens zio_buf_cache[c] = kmem_cache_create(name, size, 151e291592aSJonathan Adams align, NULL, NULL, NULL, NULL, NULL, cflags); 152ad23a2dbSjohansen 153e291592aSJonathan Adams /* 154e291592aSJonathan Adams * Since zio_data bufs do not appear in crash dumps, we 155e291592aSJonathan Adams * pass KMC_NOTOUCH so that no allocator metadata is 156e291592aSJonathan Adams * stored with the buffers. 157e291592aSJonathan Adams */ 158ad23a2dbSjohansen (void) sprintf(name, "zio_data_buf_%lu", (ulong_t)size); 159ad23a2dbSjohansen zio_data_buf_cache[c] = kmem_cache_create(name, size, 160ad23a2dbSjohansen align, NULL, NULL, NULL, NULL, data_alloc_arena, 161e291592aSJonathan Adams cflags | KMC_NOTOUCH); 162fa9e4066Sahrens } 163fa9e4066Sahrens } 164fa9e4066Sahrens 165fa9e4066Sahrens while (--c != 0) { 166fa9e4066Sahrens ASSERT(zio_buf_cache[c] != NULL); 167fa9e4066Sahrens if (zio_buf_cache[c - 1] == NULL) 168fa9e4066Sahrens zio_buf_cache[c - 1] = zio_buf_cache[c]; 169ad23a2dbSjohansen 170ad23a2dbSjohansen ASSERT(zio_data_buf_cache[c] != NULL); 171ad23a2dbSjohansen if (zio_data_buf_cache[c - 1] == NULL) 172ad23a2dbSjohansen zio_data_buf_cache[c - 1] = zio_data_buf_cache[c]; 173fa9e4066Sahrens } 174ea8dc4b6Seschrock 175ea8dc4b6Seschrock zio_inject_init(); 176fa9e4066Sahrens } 177fa9e4066Sahrens 178fa9e4066Sahrens void 179fa9e4066Sahrens zio_fini(void) 180fa9e4066Sahrens { 181fa9e4066Sahrens size_t c; 182fa9e4066Sahrens kmem_cache_t *last_cache = NULL; 183ad23a2dbSjohansen kmem_cache_t *last_data_cache = NULL; 184fa9e4066Sahrens 185fa9e4066Sahrens for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) { 186fa9e4066Sahrens if (zio_buf_cache[c] != last_cache) { 187fa9e4066Sahrens last_cache = zio_buf_cache[c]; 188fa9e4066Sahrens kmem_cache_destroy(zio_buf_cache[c]); 189fa9e4066Sahrens } 190fa9e4066Sahrens zio_buf_cache[c] = NULL; 191ad23a2dbSjohansen 192ad23a2dbSjohansen if (zio_data_buf_cache[c] != last_data_cache) { 193ad23a2dbSjohansen last_data_cache = zio_data_buf_cache[c]; 194ad23a2dbSjohansen kmem_cache_destroy(zio_data_buf_cache[c]); 195ad23a2dbSjohansen } 196ad23a2dbSjohansen zio_data_buf_cache[c] = NULL; 197fa9e4066Sahrens } 198ea8dc4b6Seschrock 199a3f829aeSBill Moore kmem_cache_destroy(zio_link_cache); 200ccae0b50Seschrock kmem_cache_destroy(zio_cache); 201ccae0b50Seschrock 202ea8dc4b6Seschrock zio_inject_fini(); 203fa9e4066Sahrens } 204fa9e4066Sahrens 205fa9e4066Sahrens /* 206fa9e4066Sahrens * ========================================================================== 207fa9e4066Sahrens * Allocate and free I/O buffers 208fa9e4066Sahrens * ========================================================================== 209fa9e4066Sahrens */ 210ad23a2dbSjohansen 211ad23a2dbSjohansen /* 212ad23a2dbSjohansen * Use zio_buf_alloc to allocate ZFS metadata. This data will appear in a 213ad23a2dbSjohansen * crashdump if the kernel panics, so use it judiciously. Obviously, it's 214ad23a2dbSjohansen * useful to inspect ZFS metadata, but if possible, we should avoid keeping 215ad23a2dbSjohansen * excess / transient data in-core during a crashdump. 216ad23a2dbSjohansen */ 217fa9e4066Sahrens void * 218fa9e4066Sahrens zio_buf_alloc(size_t size) 219fa9e4066Sahrens { 220fa9e4066Sahrens size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; 221fa9e4066Sahrens 222f63ab3d5SMatthew Ahrens VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); 223fa9e4066Sahrens 2241ab7f2deSmaybee return (kmem_cache_alloc(zio_buf_cache[c], KM_PUSHPAGE)); 225fa9e4066Sahrens } 226fa9e4066Sahrens 227ad23a2dbSjohansen /* 228ad23a2dbSjohansen * Use zio_data_buf_alloc to allocate data. The data will not appear in a 229ad23a2dbSjohansen * crashdump if the kernel panics. This exists so that we will limit the amount 230ad23a2dbSjohansen * of ZFS data that shows up in a kernel crashdump. (Thus reducing the amount 231ad23a2dbSjohansen * of kernel heap dumped to disk when the kernel panics) 232ad23a2dbSjohansen */ 233ad23a2dbSjohansen void * 234ad23a2dbSjohansen zio_data_buf_alloc(size_t size) 235ad23a2dbSjohansen { 236ad23a2dbSjohansen size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; 237ad23a2dbSjohansen 238f63ab3d5SMatthew Ahrens VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); 239ad23a2dbSjohansen 2401ab7f2deSmaybee return (kmem_cache_alloc(zio_data_buf_cache[c], KM_PUSHPAGE)); 241ad23a2dbSjohansen } 242ad23a2dbSjohansen 243fa9e4066Sahrens void 244fa9e4066Sahrens zio_buf_free(void *buf, size_t size) 245fa9e4066Sahrens { 246fa9e4066Sahrens size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; 247fa9e4066Sahrens 248f63ab3d5SMatthew Ahrens VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); 249fa9e4066Sahrens 250fa9e4066Sahrens kmem_cache_free(zio_buf_cache[c], buf); 251fa9e4066Sahrens } 252fa9e4066Sahrens 253ad23a2dbSjohansen void 254ad23a2dbSjohansen zio_data_buf_free(void *buf, size_t size) 255ad23a2dbSjohansen { 256ad23a2dbSjohansen size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; 257ad23a2dbSjohansen 258f63ab3d5SMatthew Ahrens VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); 259ad23a2dbSjohansen 260ad23a2dbSjohansen kmem_cache_free(zio_data_buf_cache[c], buf); 261ad23a2dbSjohansen } 262b3995adbSahrens 263fa9e4066Sahrens /* 264fa9e4066Sahrens * ========================================================================== 265fa9e4066Sahrens * Push and pop I/O transform buffers 266fa9e4066Sahrens * ========================================================================== 267fa9e4066Sahrens */ 268fa9e4066Sahrens static void 269e14bb325SJeff Bonwick zio_push_transform(zio_t *zio, void *data, uint64_t size, uint64_t bufsize, 270e14bb325SJeff Bonwick zio_transform_func_t *transform) 271fa9e4066Sahrens { 272fa9e4066Sahrens zio_transform_t *zt = kmem_alloc(sizeof (zio_transform_t), KM_SLEEP); 273fa9e4066Sahrens 274e14bb325SJeff Bonwick zt->zt_orig_data = zio->io_data; 275e14bb325SJeff Bonwick zt->zt_orig_size = zio->io_size; 276fa9e4066Sahrens zt->zt_bufsize = bufsize; 277e14bb325SJeff Bonwick zt->zt_transform = transform; 278fa9e4066Sahrens 279fa9e4066Sahrens zt->zt_next = zio->io_transform_stack; 280fa9e4066Sahrens zio->io_transform_stack = zt; 281fa9e4066Sahrens 282fa9e4066Sahrens zio->io_data = data; 283fa9e4066Sahrens zio->io_size = size; 284fa9e4066Sahrens } 285fa9e4066Sahrens 286fa9e4066Sahrens static void 287e14bb325SJeff Bonwick zio_pop_transforms(zio_t *zio) 288fa9e4066Sahrens { 289e14bb325SJeff Bonwick zio_transform_t *zt; 290e14bb325SJeff Bonwick 291e14bb325SJeff Bonwick while ((zt = zio->io_transform_stack) != NULL) { 292e14bb325SJeff Bonwick if (zt->zt_transform != NULL) 293e14bb325SJeff Bonwick zt->zt_transform(zio, 294e14bb325SJeff Bonwick zt->zt_orig_data, zt->zt_orig_size); 295fa9e4066Sahrens 296b24ab676SJeff Bonwick if (zt->zt_bufsize != 0) 297b24ab676SJeff Bonwick zio_buf_free(zio->io_data, zt->zt_bufsize); 298fa9e4066Sahrens 299e14bb325SJeff Bonwick zio->io_data = zt->zt_orig_data; 300e14bb325SJeff Bonwick zio->io_size = zt->zt_orig_size; 301e14bb325SJeff Bonwick zio->io_transform_stack = zt->zt_next; 302fa9e4066Sahrens 303e14bb325SJeff Bonwick kmem_free(zt, sizeof (zio_transform_t)); 304fa9e4066Sahrens } 305fa9e4066Sahrens } 306fa9e4066Sahrens 307e14bb325SJeff Bonwick /* 308e14bb325SJeff Bonwick * ========================================================================== 309e14bb325SJeff Bonwick * I/O transform callbacks for subblocks and decompression 310e14bb325SJeff Bonwick * ========================================================================== 311e14bb325SJeff Bonwick */ 312e14bb325SJeff Bonwick static void 313e14bb325SJeff Bonwick zio_subblock(zio_t *zio, void *data, uint64_t size) 314e14bb325SJeff Bonwick { 315e14bb325SJeff Bonwick ASSERT(zio->io_size > size); 316e14bb325SJeff Bonwick 317e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) 318e14bb325SJeff Bonwick bcopy(zio->io_data, data, size); 319e14bb325SJeff Bonwick } 320e14bb325SJeff Bonwick 321e14bb325SJeff Bonwick static void 322e14bb325SJeff Bonwick zio_decompress(zio_t *zio, void *data, uint64_t size) 323e14bb325SJeff Bonwick { 324e14bb325SJeff Bonwick if (zio->io_error == 0 && 325e14bb325SJeff Bonwick zio_decompress_data(BP_GET_COMPRESS(zio->io_bp), 326b24ab676SJeff Bonwick zio->io_data, data, zio->io_size, size) != 0) 327be6fd75aSMatthew Ahrens zio->io_error = SET_ERROR(EIO); 328e14bb325SJeff Bonwick } 329e14bb325SJeff Bonwick 330e14bb325SJeff Bonwick /* 331e14bb325SJeff Bonwick * ========================================================================== 332e14bb325SJeff Bonwick * I/O parent/child relationships and pipeline interlocks 333e14bb325SJeff Bonwick * ========================================================================== 334e14bb325SJeff Bonwick */ 335a3f829aeSBill Moore /* 336a3f829aeSBill Moore * NOTE - Callers to zio_walk_parents() and zio_walk_children must 337a3f829aeSBill Moore * continue calling these functions until they return NULL. 338a3f829aeSBill Moore * Otherwise, the next caller will pick up the list walk in 339a3f829aeSBill Moore * some indeterminate state. (Otherwise every caller would 340a3f829aeSBill Moore * have to pass in a cookie to keep the state represented by 341a3f829aeSBill Moore * io_walk_link, which gets annoying.) 342a3f829aeSBill Moore */ 343a3f829aeSBill Moore zio_t * 344a3f829aeSBill Moore zio_walk_parents(zio_t *cio) 345a3f829aeSBill Moore { 346a3f829aeSBill Moore zio_link_t *zl = cio->io_walk_link; 347a3f829aeSBill Moore list_t *pl = &cio->io_parent_list; 348e14bb325SJeff Bonwick 349a3f829aeSBill Moore zl = (zl == NULL) ? list_head(pl) : list_next(pl, zl); 350a3f829aeSBill Moore cio->io_walk_link = zl; 351a3f829aeSBill Moore 352a3f829aeSBill Moore if (zl == NULL) 353a3f829aeSBill Moore return (NULL); 354a3f829aeSBill Moore 355a3f829aeSBill Moore ASSERT(zl->zl_child == cio); 356a3f829aeSBill Moore return (zl->zl_parent); 357a3f829aeSBill Moore } 358a3f829aeSBill Moore 359a3f829aeSBill Moore zio_t * 360a3f829aeSBill Moore zio_walk_children(zio_t *pio) 361a3f829aeSBill Moore { 362a3f829aeSBill Moore zio_link_t *zl = pio->io_walk_link; 363a3f829aeSBill Moore list_t *cl = &pio->io_child_list; 364a3f829aeSBill Moore 365a3f829aeSBill Moore zl = (zl == NULL) ? list_head(cl) : list_next(cl, zl); 366a3f829aeSBill Moore pio->io_walk_link = zl; 367a3f829aeSBill Moore 368a3f829aeSBill Moore if (zl == NULL) 369a3f829aeSBill Moore return (NULL); 370a3f829aeSBill Moore 371a3f829aeSBill Moore ASSERT(zl->zl_parent == pio); 372a3f829aeSBill Moore return (zl->zl_child); 373a3f829aeSBill Moore } 374a3f829aeSBill Moore 375a3f829aeSBill Moore zio_t * 376a3f829aeSBill Moore zio_unique_parent(zio_t *cio) 377a3f829aeSBill Moore { 378a3f829aeSBill Moore zio_t *pio = zio_walk_parents(cio); 379a3f829aeSBill Moore 380a3f829aeSBill Moore VERIFY(zio_walk_parents(cio) == NULL); 381a3f829aeSBill Moore return (pio); 382a3f829aeSBill Moore } 383a3f829aeSBill Moore 384a3f829aeSBill Moore void 385a3f829aeSBill Moore zio_add_child(zio_t *pio, zio_t *cio) 386e14bb325SJeff Bonwick { 387a3f829aeSBill Moore zio_link_t *zl = kmem_cache_alloc(zio_link_cache, KM_SLEEP); 388a3f829aeSBill Moore 389a3f829aeSBill Moore /* 390a3f829aeSBill Moore * Logical I/Os can have logical, gang, or vdev children. 391a3f829aeSBill Moore * Gang I/Os can have gang or vdev children. 392a3f829aeSBill Moore * Vdev I/Os can only have vdev children. 393a3f829aeSBill Moore * The following ASSERT captures all of these constraints. 394a3f829aeSBill Moore */ 395a3f829aeSBill Moore ASSERT(cio->io_child_type <= pio->io_child_type); 396a3f829aeSBill Moore 397a3f829aeSBill Moore zl->zl_parent = pio; 398a3f829aeSBill Moore zl->zl_child = cio; 399a3f829aeSBill Moore 400a3f829aeSBill Moore mutex_enter(&cio->io_lock); 401e14bb325SJeff Bonwick mutex_enter(&pio->io_lock); 402a3f829aeSBill Moore 403a3f829aeSBill Moore ASSERT(pio->io_state[ZIO_WAIT_DONE] == 0); 404a3f829aeSBill Moore 405a3f829aeSBill Moore for (int w = 0; w < ZIO_WAIT_TYPES; w++) 406a3f829aeSBill Moore pio->io_children[cio->io_child_type][w] += !cio->io_state[w]; 407a3f829aeSBill Moore 408a3f829aeSBill Moore list_insert_head(&pio->io_child_list, zl); 409a3f829aeSBill Moore list_insert_head(&cio->io_parent_list, zl); 410a3f829aeSBill Moore 411b24ab676SJeff Bonwick pio->io_child_count++; 412b24ab676SJeff Bonwick cio->io_parent_count++; 413b24ab676SJeff Bonwick 414e14bb325SJeff Bonwick mutex_exit(&pio->io_lock); 415a3f829aeSBill Moore mutex_exit(&cio->io_lock); 416e14bb325SJeff Bonwick } 417e14bb325SJeff Bonwick 418fa9e4066Sahrens static void 419a3f829aeSBill Moore zio_remove_child(zio_t *pio, zio_t *cio, zio_link_t *zl) 420e14bb325SJeff Bonwick { 421a3f829aeSBill Moore ASSERT(zl->zl_parent == pio); 422a3f829aeSBill Moore ASSERT(zl->zl_child == cio); 423e14bb325SJeff Bonwick 424a3f829aeSBill Moore mutex_enter(&cio->io_lock); 425e14bb325SJeff Bonwick mutex_enter(&pio->io_lock); 426a3f829aeSBill Moore 427a3f829aeSBill Moore list_remove(&pio->io_child_list, zl); 428a3f829aeSBill Moore list_remove(&cio->io_parent_list, zl); 429a3f829aeSBill Moore 430b24ab676SJeff Bonwick pio->io_child_count--; 431b24ab676SJeff Bonwick cio->io_parent_count--; 432b24ab676SJeff Bonwick 433e14bb325SJeff Bonwick mutex_exit(&pio->io_lock); 434a3f829aeSBill Moore mutex_exit(&cio->io_lock); 435a3f829aeSBill Moore 436a3f829aeSBill Moore kmem_cache_free(zio_link_cache, zl); 437e14bb325SJeff Bonwick } 438e14bb325SJeff Bonwick 439e14bb325SJeff Bonwick static boolean_t 440e14bb325SJeff Bonwick zio_wait_for_children(zio_t *zio, enum zio_child child, enum zio_wait_type wait) 441fa9e4066Sahrens { 442e14bb325SJeff Bonwick uint64_t *countp = &zio->io_children[child][wait]; 443e14bb325SJeff Bonwick boolean_t waiting = B_FALSE; 444e14bb325SJeff Bonwick 445e14bb325SJeff Bonwick mutex_enter(&zio->io_lock); 446e14bb325SJeff Bonwick ASSERT(zio->io_stall == NULL); 447e14bb325SJeff Bonwick if (*countp != 0) { 448b24ab676SJeff Bonwick zio->io_stage >>= 1; 449e14bb325SJeff Bonwick zio->io_stall = countp; 450e14bb325SJeff Bonwick waiting = B_TRUE; 451e14bb325SJeff Bonwick } 452e14bb325SJeff Bonwick mutex_exit(&zio->io_lock); 453e14bb325SJeff Bonwick 454e14bb325SJeff Bonwick return (waiting); 455e14bb325SJeff Bonwick } 456fa9e4066Sahrens 457e14bb325SJeff Bonwick static void 458e14bb325SJeff Bonwick zio_notify_parent(zio_t *pio, zio_t *zio, enum zio_wait_type wait) 459e14bb325SJeff Bonwick { 460e14bb325SJeff Bonwick uint64_t *countp = &pio->io_children[zio->io_child_type][wait]; 461e14bb325SJeff Bonwick int *errorp = &pio->io_child_error[zio->io_child_type]; 462fa9e4066Sahrens 463e14bb325SJeff Bonwick mutex_enter(&pio->io_lock); 464e14bb325SJeff Bonwick if (zio->io_error && !(zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 465e14bb325SJeff Bonwick *errorp = zio_worst_error(*errorp, zio->io_error); 466e14bb325SJeff Bonwick pio->io_reexecute |= zio->io_reexecute; 467e14bb325SJeff Bonwick ASSERT3U(*countp, >, 0); 46869962b56SMatthew Ahrens 46969962b56SMatthew Ahrens (*countp)--; 47069962b56SMatthew Ahrens 47169962b56SMatthew Ahrens if (*countp == 0 && pio->io_stall == countp) { 472e14bb325SJeff Bonwick pio->io_stall = NULL; 473e14bb325SJeff Bonwick mutex_exit(&pio->io_lock); 474e14bb325SJeff Bonwick zio_execute(pio); 475e14bb325SJeff Bonwick } else { 476e14bb325SJeff Bonwick mutex_exit(&pio->io_lock); 477fa9e4066Sahrens } 478fa9e4066Sahrens } 479fa9e4066Sahrens 480e14bb325SJeff Bonwick static void 481e14bb325SJeff Bonwick zio_inherit_child_errors(zio_t *zio, enum zio_child c) 482e14bb325SJeff Bonwick { 483e14bb325SJeff Bonwick if (zio->io_child_error[c] != 0 && zio->io_error == 0) 484e14bb325SJeff Bonwick zio->io_error = zio->io_child_error[c]; 485e14bb325SJeff Bonwick } 486e14bb325SJeff Bonwick 487fa9e4066Sahrens /* 488fa9e4066Sahrens * ========================================================================== 489e14bb325SJeff Bonwick * Create the various types of I/O (read, write, free, etc) 490fa9e4066Sahrens * ========================================================================== 491fa9e4066Sahrens */ 492fa9e4066Sahrens static zio_t * 493b24ab676SJeff Bonwick zio_create(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, 494fa9e4066Sahrens void *data, uint64_t size, zio_done_func_t *done, void *private, 49569962b56SMatthew Ahrens zio_type_t type, zio_priority_t priority, enum zio_flag flags, 4967802d7bfSMatthew Ahrens vdev_t *vd, uint64_t offset, const zbookmark_phys_t *zb, 497b24ab676SJeff Bonwick enum zio_stage stage, enum zio_stage pipeline) 498fa9e4066Sahrens { 499fa9e4066Sahrens zio_t *zio; 500fa9e4066Sahrens 501fa9e4066Sahrens ASSERT3U(size, <=, SPA_MAXBLOCKSIZE); 502fa9e4066Sahrens ASSERT(P2PHASE(size, SPA_MINBLOCKSIZE) == 0); 503e14bb325SJeff Bonwick ASSERT(P2PHASE(offset, SPA_MINBLOCKSIZE) == 0); 504fa9e4066Sahrens 505e14bb325SJeff Bonwick ASSERT(!vd || spa_config_held(spa, SCL_STATE_ALL, RW_READER)); 506e14bb325SJeff Bonwick ASSERT(!bp || !(flags & ZIO_FLAG_CONFIG_WRITER)); 507e14bb325SJeff Bonwick ASSERT(vd || stage == ZIO_STAGE_OPEN); 508088f3894Sahrens 509ccae0b50Seschrock zio = kmem_cache_alloc(zio_cache, KM_SLEEP); 510ccae0b50Seschrock bzero(zio, sizeof (zio_t)); 511e14bb325SJeff Bonwick 512e14bb325SJeff Bonwick mutex_init(&zio->io_lock, NULL, MUTEX_DEFAULT, NULL); 513e14bb325SJeff Bonwick cv_init(&zio->io_cv, NULL, CV_DEFAULT, NULL); 514e14bb325SJeff Bonwick 515a3f829aeSBill Moore list_create(&zio->io_parent_list, sizeof (zio_link_t), 516a3f829aeSBill Moore offsetof(zio_link_t, zl_parent_node)); 517a3f829aeSBill Moore list_create(&zio->io_child_list, sizeof (zio_link_t), 518a3f829aeSBill Moore offsetof(zio_link_t, zl_child_node)); 519a3f829aeSBill Moore 520e14bb325SJeff Bonwick if (vd != NULL) 521e14bb325SJeff Bonwick zio->io_child_type = ZIO_CHILD_VDEV; 522e14bb325SJeff Bonwick else if (flags & ZIO_FLAG_GANG_CHILD) 523e14bb325SJeff Bonwick zio->io_child_type = ZIO_CHILD_GANG; 524b24ab676SJeff Bonwick else if (flags & ZIO_FLAG_DDT_CHILD) 525b24ab676SJeff Bonwick zio->io_child_type = ZIO_CHILD_DDT; 526e14bb325SJeff Bonwick else 527e14bb325SJeff Bonwick zio->io_child_type = ZIO_CHILD_LOGICAL; 528e14bb325SJeff Bonwick 529fa9e4066Sahrens if (bp != NULL) { 530b24ab676SJeff Bonwick zio->io_bp = (blkptr_t *)bp; 531fa9e4066Sahrens zio->io_bp_copy = *bp; 532fa9e4066Sahrens zio->io_bp_orig = *bp; 533b24ab676SJeff Bonwick if (type != ZIO_TYPE_WRITE || 534b24ab676SJeff Bonwick zio->io_child_type == ZIO_CHILD_DDT) 535e14bb325SJeff Bonwick zio->io_bp = &zio->io_bp_copy; /* so caller can free */ 536f5383399SBill Moore if (zio->io_child_type == ZIO_CHILD_LOGICAL) 537e14bb325SJeff Bonwick zio->io_logical = zio; 538f5383399SBill Moore if (zio->io_child_type > ZIO_CHILD_GANG && BP_IS_GANG(bp)) 539f5383399SBill Moore pipeline |= ZIO_GANG_STAGES; 540fa9e4066Sahrens } 541e14bb325SJeff Bonwick 542e14bb325SJeff Bonwick zio->io_spa = spa; 543e14bb325SJeff Bonwick zio->io_txg = txg; 544fa9e4066Sahrens zio->io_done = done; 545fa9e4066Sahrens zio->io_private = private; 546fa9e4066Sahrens zio->io_type = type; 547fa9e4066Sahrens zio->io_priority = priority; 548e14bb325SJeff Bonwick zio->io_vd = vd; 549e14bb325SJeff Bonwick zio->io_offset = offset; 550b24ab676SJeff Bonwick zio->io_orig_data = zio->io_data = data; 551b24ab676SJeff Bonwick zio->io_orig_size = zio->io_size = size; 552e14bb325SJeff Bonwick zio->io_orig_flags = zio->io_flags = flags; 553e14bb325SJeff Bonwick zio->io_orig_stage = zio->io_stage = stage; 554e14bb325SJeff Bonwick zio->io_orig_pipeline = zio->io_pipeline = pipeline; 555fa9e4066Sahrens 556a3f829aeSBill Moore zio->io_state[ZIO_WAIT_READY] = (stage >= ZIO_STAGE_READY); 557a3f829aeSBill Moore zio->io_state[ZIO_WAIT_DONE] = (stage >= ZIO_STAGE_DONE); 558a3f829aeSBill Moore 559e14bb325SJeff Bonwick if (zb != NULL) 560e14bb325SJeff Bonwick zio->io_bookmark = *zb; 561e14bb325SJeff Bonwick 562e14bb325SJeff Bonwick if (pio != NULL) { 563e14bb325SJeff Bonwick if (zio->io_logical == NULL) 564ea8dc4b6Seschrock zio->io_logical = pio->io_logical; 565f5383399SBill Moore if (zio->io_child_type == ZIO_CHILD_GANG) 566f5383399SBill Moore zio->io_gang_leader = pio->io_gang_leader; 567e14bb325SJeff Bonwick zio_add_child(pio, zio); 568fa9e4066Sahrens } 569fa9e4066Sahrens 570fa9e4066Sahrens return (zio); 571fa9e4066Sahrens } 572fa9e4066Sahrens 5730a4e9518Sgw static void 574e14bb325SJeff Bonwick zio_destroy(zio_t *zio) 5750a4e9518Sgw { 576a3f829aeSBill Moore list_destroy(&zio->io_parent_list); 577a3f829aeSBill Moore list_destroy(&zio->io_child_list); 578e14bb325SJeff Bonwick mutex_destroy(&zio->io_lock); 579e14bb325SJeff Bonwick cv_destroy(&zio->io_cv); 580e14bb325SJeff Bonwick kmem_cache_free(zio_cache, zio); 5810a4e9518Sgw } 5820a4e9518Sgw 583fa9e4066Sahrens zio_t * 584a3f829aeSBill Moore zio_null(zio_t *pio, spa_t *spa, vdev_t *vd, zio_done_func_t *done, 585b24ab676SJeff Bonwick void *private, enum zio_flag flags) 586fa9e4066Sahrens { 587fa9e4066Sahrens zio_t *zio; 588fa9e4066Sahrens 589fa9e4066Sahrens zio = zio_create(pio, spa, 0, NULL, NULL, 0, done, private, 590a3f829aeSBill Moore ZIO_TYPE_NULL, ZIO_PRIORITY_NOW, flags, vd, 0, NULL, 591e14bb325SJeff Bonwick ZIO_STAGE_OPEN, ZIO_INTERLOCK_PIPELINE); 592fa9e4066Sahrens 593fa9e4066Sahrens return (zio); 594fa9e4066Sahrens } 595fa9e4066Sahrens 596fa9e4066Sahrens zio_t * 597b24ab676SJeff Bonwick zio_root(spa_t *spa, zio_done_func_t *done, void *private, enum zio_flag flags) 598fa9e4066Sahrens { 599a3f829aeSBill Moore return (zio_null(NULL, spa, NULL, done, private, flags)); 600fa9e4066Sahrens } 601fa9e4066Sahrens 602f63ab3d5SMatthew Ahrens void 603f63ab3d5SMatthew Ahrens zfs_blkptr_verify(spa_t *spa, const blkptr_t *bp) 604f63ab3d5SMatthew Ahrens { 605f63ab3d5SMatthew Ahrens if (!DMU_OT_IS_VALID(BP_GET_TYPE(bp))) { 606f63ab3d5SMatthew Ahrens zfs_panic_recover("blkptr at %p has invalid TYPE %llu", 607f63ab3d5SMatthew Ahrens bp, (longlong_t)BP_GET_TYPE(bp)); 608f63ab3d5SMatthew Ahrens } 609f63ab3d5SMatthew Ahrens if (BP_GET_CHECKSUM(bp) >= ZIO_CHECKSUM_FUNCTIONS || 610f63ab3d5SMatthew Ahrens BP_GET_CHECKSUM(bp) <= ZIO_CHECKSUM_ON) { 611f63ab3d5SMatthew Ahrens zfs_panic_recover("blkptr at %p has invalid CHECKSUM %llu", 612f63ab3d5SMatthew Ahrens bp, (longlong_t)BP_GET_CHECKSUM(bp)); 613f63ab3d5SMatthew Ahrens } 614f63ab3d5SMatthew Ahrens if (BP_GET_COMPRESS(bp) >= ZIO_COMPRESS_FUNCTIONS || 615f63ab3d5SMatthew Ahrens BP_GET_COMPRESS(bp) <= ZIO_COMPRESS_ON) { 616f63ab3d5SMatthew Ahrens zfs_panic_recover("blkptr at %p has invalid COMPRESS %llu", 617f63ab3d5SMatthew Ahrens bp, (longlong_t)BP_GET_COMPRESS(bp)); 618f63ab3d5SMatthew Ahrens } 619f63ab3d5SMatthew Ahrens if (BP_GET_LSIZE(bp) > SPA_MAXBLOCKSIZE) { 620f63ab3d5SMatthew Ahrens zfs_panic_recover("blkptr at %p has invalid LSIZE %llu", 621f63ab3d5SMatthew Ahrens bp, (longlong_t)BP_GET_LSIZE(bp)); 622f63ab3d5SMatthew Ahrens } 623f63ab3d5SMatthew Ahrens if (BP_GET_PSIZE(bp) > SPA_MAXBLOCKSIZE) { 624f63ab3d5SMatthew Ahrens zfs_panic_recover("blkptr at %p has invalid PSIZE %llu", 625f63ab3d5SMatthew Ahrens bp, (longlong_t)BP_GET_PSIZE(bp)); 626f63ab3d5SMatthew Ahrens } 627f63ab3d5SMatthew Ahrens 628f63ab3d5SMatthew Ahrens if (BP_IS_EMBEDDED(bp)) { 629f63ab3d5SMatthew Ahrens if (BPE_GET_ETYPE(bp) > NUM_BP_EMBEDDED_TYPES) { 630f63ab3d5SMatthew Ahrens zfs_panic_recover("blkptr at %p has invalid ETYPE %llu", 631f63ab3d5SMatthew Ahrens bp, (longlong_t)BPE_GET_ETYPE(bp)); 632f63ab3d5SMatthew Ahrens } 633f63ab3d5SMatthew Ahrens } 634f63ab3d5SMatthew Ahrens 635f63ab3d5SMatthew Ahrens /* 636f63ab3d5SMatthew Ahrens * Pool-specific checks. 637f63ab3d5SMatthew Ahrens * 638f63ab3d5SMatthew Ahrens * Note: it would be nice to verify that the blk_birth and 639f63ab3d5SMatthew Ahrens * BP_PHYSICAL_BIRTH() are not too large. However, spa_freeze() 640f63ab3d5SMatthew Ahrens * allows the birth time of log blocks (and dmu_sync()-ed blocks 641f63ab3d5SMatthew Ahrens * that are in the log) to be arbitrarily large. 642f63ab3d5SMatthew Ahrens */ 643f63ab3d5SMatthew Ahrens for (int i = 0; i < BP_GET_NDVAS(bp); i++) { 644f63ab3d5SMatthew Ahrens uint64_t vdevid = DVA_GET_VDEV(&bp->blk_dva[i]); 645f63ab3d5SMatthew Ahrens if (vdevid >= spa->spa_root_vdev->vdev_children) { 646f63ab3d5SMatthew Ahrens zfs_panic_recover("blkptr at %p DVA %u has invalid " 647f63ab3d5SMatthew Ahrens "VDEV %llu", 648f63ab3d5SMatthew Ahrens bp, i, (longlong_t)vdevid); 6495897eb49SJustin Gibbs continue; 650f63ab3d5SMatthew Ahrens } 651f63ab3d5SMatthew Ahrens vdev_t *vd = spa->spa_root_vdev->vdev_child[vdevid]; 652f63ab3d5SMatthew Ahrens if (vd == NULL) { 653f63ab3d5SMatthew Ahrens zfs_panic_recover("blkptr at %p DVA %u has invalid " 654f63ab3d5SMatthew Ahrens "VDEV %llu", 655f63ab3d5SMatthew Ahrens bp, i, (longlong_t)vdevid); 6565897eb49SJustin Gibbs continue; 657f63ab3d5SMatthew Ahrens } 658f63ab3d5SMatthew Ahrens if (vd->vdev_ops == &vdev_hole_ops) { 659f63ab3d5SMatthew Ahrens zfs_panic_recover("blkptr at %p DVA %u has hole " 660f63ab3d5SMatthew Ahrens "VDEV %llu", 661f63ab3d5SMatthew Ahrens bp, i, (longlong_t)vdevid); 6625897eb49SJustin Gibbs continue; 663f63ab3d5SMatthew Ahrens } 664f63ab3d5SMatthew Ahrens if (vd->vdev_ops == &vdev_missing_ops) { 665f63ab3d5SMatthew Ahrens /* 666f63ab3d5SMatthew Ahrens * "missing" vdevs are valid during import, but we 667f63ab3d5SMatthew Ahrens * don't have their detailed info (e.g. asize), so 668f63ab3d5SMatthew Ahrens * we can't perform any more checks on them. 669f63ab3d5SMatthew Ahrens */ 670f63ab3d5SMatthew Ahrens continue; 671f63ab3d5SMatthew Ahrens } 672f63ab3d5SMatthew Ahrens uint64_t offset = DVA_GET_OFFSET(&bp->blk_dva[i]); 673f63ab3d5SMatthew Ahrens uint64_t asize = DVA_GET_ASIZE(&bp->blk_dva[i]); 674f63ab3d5SMatthew Ahrens if (BP_IS_GANG(bp)) 675f63ab3d5SMatthew Ahrens asize = vdev_psize_to_asize(vd, SPA_GANGBLOCKSIZE); 676f63ab3d5SMatthew Ahrens if (offset + asize > vd->vdev_asize) { 677f63ab3d5SMatthew Ahrens zfs_panic_recover("blkptr at %p DVA %u has invalid " 678f63ab3d5SMatthew Ahrens "OFFSET %llu", 679f63ab3d5SMatthew Ahrens bp, i, (longlong_t)offset); 680f63ab3d5SMatthew Ahrens } 681f63ab3d5SMatthew Ahrens } 682f63ab3d5SMatthew Ahrens } 683f63ab3d5SMatthew Ahrens 684fa9e4066Sahrens zio_t * 685e14bb325SJeff Bonwick zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, 686e14bb325SJeff Bonwick void *data, uint64_t size, zio_done_func_t *done, void *private, 6877802d7bfSMatthew Ahrens zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb) 688fa9e4066Sahrens { 689fa9e4066Sahrens zio_t *zio; 690fa9e4066Sahrens 691f63ab3d5SMatthew Ahrens zfs_blkptr_verify(spa, bp); 692f63ab3d5SMatthew Ahrens 693b24ab676SJeff Bonwick zio = zio_create(pio, spa, BP_PHYSICAL_BIRTH(bp), bp, 694088f3894Sahrens data, size, done, private, 695e14bb325SJeff Bonwick ZIO_TYPE_READ, priority, flags, NULL, 0, zb, 696b24ab676SJeff Bonwick ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ? 697b24ab676SJeff Bonwick ZIO_DDT_CHILD_READ_PIPELINE : ZIO_READ_PIPELINE); 698fa9e4066Sahrens 699fa9e4066Sahrens return (zio); 700fa9e4066Sahrens } 701fa9e4066Sahrens 702fa9e4066Sahrens zio_t * 703e14bb325SJeff Bonwick zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, 704b24ab676SJeff Bonwick void *data, uint64_t size, const zio_prop_t *zp, 70569962b56SMatthew Ahrens zio_done_func_t *ready, zio_done_func_t *physdone, zio_done_func_t *done, 70669962b56SMatthew Ahrens void *private, 7077802d7bfSMatthew Ahrens zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb) 708fa9e4066Sahrens { 709fa9e4066Sahrens zio_t *zio; 710fa9e4066Sahrens 711e14bb325SJeff Bonwick ASSERT(zp->zp_checksum >= ZIO_CHECKSUM_OFF && 712e14bb325SJeff Bonwick zp->zp_checksum < ZIO_CHECKSUM_FUNCTIONS && 713e14bb325SJeff Bonwick zp->zp_compress >= ZIO_COMPRESS_OFF && 714e14bb325SJeff Bonwick zp->zp_compress < ZIO_COMPRESS_FUNCTIONS && 715ad135b5dSChristopher Siden DMU_OT_IS_VALID(zp->zp_type) && 716e14bb325SJeff Bonwick zp->zp_level < 32 && 717b24ab676SJeff Bonwick zp->zp_copies > 0 && 71880901aeaSGeorge Wilson zp->zp_copies <= spa_max_replication(spa)); 7190a4e9518Sgw 720fa9e4066Sahrens zio = zio_create(pio, spa, txg, bp, data, size, done, private, 721e14bb325SJeff Bonwick ZIO_TYPE_WRITE, priority, flags, NULL, 0, zb, 722b24ab676SJeff Bonwick ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ? 723b24ab676SJeff Bonwick ZIO_DDT_CHILD_WRITE_PIPELINE : ZIO_WRITE_PIPELINE); 724fa9e4066Sahrens 725c717a561Smaybee zio->io_ready = ready; 72669962b56SMatthew Ahrens zio->io_physdone = physdone; 727e14bb325SJeff Bonwick zio->io_prop = *zp; 728fa9e4066Sahrens 7295d7b4d43SMatthew Ahrens /* 7305d7b4d43SMatthew Ahrens * Data can be NULL if we are going to call zio_write_override() to 7315d7b4d43SMatthew Ahrens * provide the already-allocated BP. But we may need the data to 7325d7b4d43SMatthew Ahrens * verify a dedup hit (if requested). In this case, don't try to 7335d7b4d43SMatthew Ahrens * dedup (just take the already-allocated BP verbatim). 7345d7b4d43SMatthew Ahrens */ 7355d7b4d43SMatthew Ahrens if (data == NULL && zio->io_prop.zp_dedup_verify) { 7365d7b4d43SMatthew Ahrens zio->io_prop.zp_dedup = zio->io_prop.zp_dedup_verify = B_FALSE; 7375d7b4d43SMatthew Ahrens } 7385d7b4d43SMatthew Ahrens 739fa9e4066Sahrens return (zio); 740fa9e4066Sahrens } 741fa9e4066Sahrens 742fa9e4066Sahrens zio_t * 743e14bb325SJeff Bonwick zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, void *data, 74469962b56SMatthew Ahrens uint64_t size, zio_done_func_t *done, void *private, 7457802d7bfSMatthew Ahrens zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb) 746fa9e4066Sahrens { 747fa9e4066Sahrens zio_t *zio; 748fa9e4066Sahrens 749fa9e4066Sahrens zio = zio_create(pio, spa, txg, bp, data, size, done, private, 750e14bb325SJeff Bonwick ZIO_TYPE_WRITE, priority, flags, NULL, 0, zb, 751e14bb325SJeff Bonwick ZIO_STAGE_OPEN, ZIO_REWRITE_PIPELINE); 752fa9e4066Sahrens 753fa9e4066Sahrens return (zio); 754fa9e4066Sahrens } 755fa9e4066Sahrens 756b24ab676SJeff Bonwick void 75780901aeaSGeorge Wilson zio_write_override(zio_t *zio, blkptr_t *bp, int copies, boolean_t nopwrite) 758b24ab676SJeff Bonwick { 759b24ab676SJeff Bonwick ASSERT(zio->io_type == ZIO_TYPE_WRITE); 760b24ab676SJeff Bonwick ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 761b24ab676SJeff Bonwick ASSERT(zio->io_stage == ZIO_STAGE_OPEN); 762b24ab676SJeff Bonwick ASSERT(zio->io_txg == spa_syncing_txg(zio->io_spa)); 763b24ab676SJeff Bonwick 76480901aeaSGeorge Wilson /* 76580901aeaSGeorge Wilson * We must reset the io_prop to match the values that existed 76680901aeaSGeorge Wilson * when the bp was first written by dmu_sync() keeping in mind 76780901aeaSGeorge Wilson * that nopwrite and dedup are mutually exclusive. 76880901aeaSGeorge Wilson */ 76980901aeaSGeorge Wilson zio->io_prop.zp_dedup = nopwrite ? B_FALSE : zio->io_prop.zp_dedup; 77080901aeaSGeorge Wilson zio->io_prop.zp_nopwrite = nopwrite; 771b24ab676SJeff Bonwick zio->io_prop.zp_copies = copies; 772b24ab676SJeff Bonwick zio->io_bp_override = bp; 773b24ab676SJeff Bonwick } 774b24ab676SJeff Bonwick 775b24ab676SJeff Bonwick void 776b24ab676SJeff Bonwick zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp) 777b24ab676SJeff Bonwick { 7785d7b4d43SMatthew Ahrens 7795d7b4d43SMatthew Ahrens /* 7805d7b4d43SMatthew Ahrens * The check for EMBEDDED is a performance optimization. We 7815d7b4d43SMatthew Ahrens * process the free here (by ignoring it) rather than 7825d7b4d43SMatthew Ahrens * putting it on the list and then processing it in zio_free_sync(). 7835d7b4d43SMatthew Ahrens */ 7845d7b4d43SMatthew Ahrens if (BP_IS_EMBEDDED(bp)) 7855d7b4d43SMatthew Ahrens return; 7863b2aab18SMatthew Ahrens metaslab_check_free(spa, bp); 7879cb154a3SMatthew Ahrens 7889cb154a3SMatthew Ahrens /* 7899cb154a3SMatthew Ahrens * Frees that are for the currently-syncing txg, are not going to be 7909cb154a3SMatthew Ahrens * deferred, and which will not need to do a read (i.e. not GANG or 7919cb154a3SMatthew Ahrens * DEDUP), can be processed immediately. Otherwise, put them on the 7929cb154a3SMatthew Ahrens * in-memory list for later processing. 7939cb154a3SMatthew Ahrens */ 7949cb154a3SMatthew Ahrens if (BP_IS_GANG(bp) || BP_GET_DEDUP(bp) || 7959cb154a3SMatthew Ahrens txg != spa->spa_syncing_txg || 7969cb154a3SMatthew Ahrens spa_sync_pass(spa) >= zfs_sync_pass_deferred_free) { 7979cb154a3SMatthew Ahrens bplist_append(&spa->spa_free_bplist[txg & TXG_MASK], bp); 7989cb154a3SMatthew Ahrens } else { 7999cb154a3SMatthew Ahrens VERIFY0(zio_wait(zio_free_sync(NULL, spa, txg, bp, 0))); 8009cb154a3SMatthew Ahrens } 801b24ab676SJeff Bonwick } 802b24ab676SJeff Bonwick 803fa9e4066Sahrens zio_t * 804b24ab676SJeff Bonwick zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, 805b24ab676SJeff Bonwick enum zio_flag flags) 806fa9e4066Sahrens { 807fa9e4066Sahrens zio_t *zio; 8089cb154a3SMatthew Ahrens enum zio_stage stage = ZIO_FREE_PIPELINE; 809fa9e4066Sahrens 810fa9e4066Sahrens ASSERT(!BP_IS_HOLE(bp)); 811b24ab676SJeff Bonwick ASSERT(spa_syncing_txg(spa) == txg); 81201f55e48SGeorge Wilson ASSERT(spa_sync_pass(spa) < zfs_sync_pass_deferred_free); 813fa9e4066Sahrens 8145d7b4d43SMatthew Ahrens if (BP_IS_EMBEDDED(bp)) 8155d7b4d43SMatthew Ahrens return (zio_null(pio, spa, NULL, NULL, NULL, 0)); 8165d7b4d43SMatthew Ahrens 8173b2aab18SMatthew Ahrens metaslab_check_free(spa, bp); 8186e6d5868SMatthew Ahrens arc_freed(spa, bp); 8193b2aab18SMatthew Ahrens 8209cb154a3SMatthew Ahrens /* 8219cb154a3SMatthew Ahrens * GANG and DEDUP blocks can induce a read (for the gang block header, 8229cb154a3SMatthew Ahrens * or the DDT), so issue them asynchronously so that this thread is 8239cb154a3SMatthew Ahrens * not tied up. 8249cb154a3SMatthew Ahrens */ 8259cb154a3SMatthew Ahrens if (BP_IS_GANG(bp) || BP_GET_DEDUP(bp)) 8269cb154a3SMatthew Ahrens stage |= ZIO_STAGE_ISSUE_ASYNC; 8279cb154a3SMatthew Ahrens 828e14bb325SJeff Bonwick zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp), 82969962b56SMatthew Ahrens NULL, NULL, ZIO_TYPE_FREE, ZIO_PRIORITY_NOW, flags, 8309cb154a3SMatthew Ahrens NULL, 0, NULL, ZIO_STAGE_OPEN, stage); 8319cb154a3SMatthew Ahrens 832fa9e4066Sahrens return (zio); 833fa9e4066Sahrens } 834fa9e4066Sahrens 835fa9e4066Sahrens zio_t * 836b24ab676SJeff Bonwick zio_claim(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, 837b24ab676SJeff Bonwick zio_done_func_t *done, void *private, enum zio_flag flags) 838fa9e4066Sahrens { 839fa9e4066Sahrens zio_t *zio; 840fa9e4066Sahrens 8415d7b4d43SMatthew Ahrens dprintf_bp(bp, "claiming in txg %llu", txg); 8425d7b4d43SMatthew Ahrens 8435d7b4d43SMatthew Ahrens if (BP_IS_EMBEDDED(bp)) 8445d7b4d43SMatthew Ahrens return (zio_null(pio, spa, NULL, NULL, NULL, 0)); 8455d7b4d43SMatthew Ahrens 846fa9e4066Sahrens /* 847fa9e4066Sahrens * A claim is an allocation of a specific block. Claims are needed 848fa9e4066Sahrens * to support immediate writes in the intent log. The issue is that 849fa9e4066Sahrens * immediate writes contain committed data, but in a txg that was 850fa9e4066Sahrens * *not* committed. Upon opening the pool after an unclean shutdown, 851fa9e4066Sahrens * the intent log claims all blocks that contain immediate write data 852fa9e4066Sahrens * so that the SPA knows they're in use. 853fa9e4066Sahrens * 854fa9e4066Sahrens * All claims *must* be resolved in the first txg -- before the SPA 855fa9e4066Sahrens * starts allocating blocks -- so that nothing is allocated twice. 856b24ab676SJeff Bonwick * If txg == 0 we just verify that the block is claimable. 857fa9e4066Sahrens */ 858fa9e4066Sahrens ASSERT3U(spa->spa_uberblock.ub_rootbp.blk_birth, <, spa_first_txg(spa)); 859b24ab676SJeff Bonwick ASSERT(txg == spa_first_txg(spa) || txg == 0); 860b24ab676SJeff Bonwick ASSERT(!BP_GET_DEDUP(bp) || !spa_writeable(spa)); /* zdb(1M) */ 861fa9e4066Sahrens 862e14bb325SJeff Bonwick zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp), 863e14bb325SJeff Bonwick done, private, ZIO_TYPE_CLAIM, ZIO_PRIORITY_NOW, flags, 864e14bb325SJeff Bonwick NULL, 0, NULL, ZIO_STAGE_OPEN, ZIO_CLAIM_PIPELINE); 865fa9e4066Sahrens 866fa9e4066Sahrens return (zio); 867fa9e4066Sahrens } 868fa9e4066Sahrens 869fa9e4066Sahrens zio_t * 870fa9e4066Sahrens zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd, 87169962b56SMatthew Ahrens zio_done_func_t *done, void *private, enum zio_flag flags) 872fa9e4066Sahrens { 873fa9e4066Sahrens zio_t *zio; 874fa9e4066Sahrens int c; 875fa9e4066Sahrens 876fa9e4066Sahrens if (vd->vdev_children == 0) { 877fa9e4066Sahrens zio = zio_create(pio, spa, 0, NULL, NULL, 0, done, private, 87869962b56SMatthew Ahrens ZIO_TYPE_IOCTL, ZIO_PRIORITY_NOW, flags, vd, 0, NULL, 879fa9e4066Sahrens ZIO_STAGE_OPEN, ZIO_IOCTL_PIPELINE); 880fa9e4066Sahrens 881fa9e4066Sahrens zio->io_cmd = cmd; 882fa9e4066Sahrens } else { 883a3f829aeSBill Moore zio = zio_null(pio, spa, NULL, NULL, NULL, flags); 884fa9e4066Sahrens 885fa9e4066Sahrens for (c = 0; c < vd->vdev_children; c++) 886fa9e4066Sahrens zio_nowait(zio_ioctl(zio, spa, vd->vdev_child[c], cmd, 88769962b56SMatthew Ahrens done, private, flags)); 888fa9e4066Sahrens } 889fa9e4066Sahrens 890fa9e4066Sahrens return (zio); 891fa9e4066Sahrens } 892fa9e4066Sahrens 893fa9e4066Sahrens zio_t * 894fa9e4066Sahrens zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, 895fa9e4066Sahrens void *data, int checksum, zio_done_func_t *done, void *private, 89669962b56SMatthew Ahrens zio_priority_t priority, enum zio_flag flags, boolean_t labels) 897fa9e4066Sahrens { 898fa9e4066Sahrens zio_t *zio; 8990a4e9518Sgw 900e14bb325SJeff Bonwick ASSERT(vd->vdev_children == 0); 901e14bb325SJeff Bonwick ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE || 902e14bb325SJeff Bonwick offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE); 903e14bb325SJeff Bonwick ASSERT3U(offset + size, <=, vd->vdev_psize); 904fa9e4066Sahrens 905e14bb325SJeff Bonwick zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private, 9062a104a52SAlex Reece ZIO_TYPE_READ, priority, flags | ZIO_FLAG_PHYSICAL, vd, offset, 9072a104a52SAlex Reece NULL, ZIO_STAGE_OPEN, ZIO_READ_PHYS_PIPELINE); 908fa9e4066Sahrens 909e14bb325SJeff Bonwick zio->io_prop.zp_checksum = checksum; 910fa9e4066Sahrens 911fa9e4066Sahrens return (zio); 912fa9e4066Sahrens } 913fa9e4066Sahrens 914fa9e4066Sahrens zio_t * 915fa9e4066Sahrens zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, 916fa9e4066Sahrens void *data, int checksum, zio_done_func_t *done, void *private, 91769962b56SMatthew Ahrens zio_priority_t priority, enum zio_flag flags, boolean_t labels) 918fa9e4066Sahrens { 919fa9e4066Sahrens zio_t *zio; 9200a4e9518Sgw 921e14bb325SJeff Bonwick ASSERT(vd->vdev_children == 0); 922e14bb325SJeff Bonwick ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE || 923e14bb325SJeff Bonwick offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE); 924e14bb325SJeff Bonwick ASSERT3U(offset + size, <=, vd->vdev_psize); 925fa9e4066Sahrens 926e14bb325SJeff Bonwick zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private, 9272a104a52SAlex Reece ZIO_TYPE_WRITE, priority, flags | ZIO_FLAG_PHYSICAL, vd, offset, 9282a104a52SAlex Reece NULL, ZIO_STAGE_OPEN, ZIO_WRITE_PHYS_PIPELINE); 929fa9e4066Sahrens 930e14bb325SJeff Bonwick zio->io_prop.zp_checksum = checksum; 931fa9e4066Sahrens 9326e1f5caaSNeil Perrin if (zio_checksum_table[checksum].ci_eck) { 933fa9e4066Sahrens /* 9346e1f5caaSNeil Perrin * zec checksums are necessarily destructive -- they modify 935e14bb325SJeff Bonwick * the end of the write buffer to hold the verifier/checksum. 936fa9e4066Sahrens * Therefore, we must make a local copy in case the data is 937e14bb325SJeff Bonwick * being written to multiple places in parallel. 938fa9e4066Sahrens */ 939e14bb325SJeff Bonwick void *wbuf = zio_buf_alloc(size); 940fa9e4066Sahrens bcopy(data, wbuf, size); 941e14bb325SJeff Bonwick zio_push_transform(zio, wbuf, size, size, NULL); 942fa9e4066Sahrens } 943fa9e4066Sahrens 944fa9e4066Sahrens return (zio); 945fa9e4066Sahrens } 946fa9e4066Sahrens 947fa9e4066Sahrens /* 948e14bb325SJeff Bonwick * Create a child I/O to do some work for us. 949fa9e4066Sahrens */ 950fa9e4066Sahrens zio_t * 951e14bb325SJeff Bonwick zio_vdev_child_io(zio_t *pio, blkptr_t *bp, vdev_t *vd, uint64_t offset, 95269962b56SMatthew Ahrens void *data, uint64_t size, int type, zio_priority_t priority, 95369962b56SMatthew Ahrens enum zio_flag flags, zio_done_func_t *done, void *private) 954fa9e4066Sahrens { 955b24ab676SJeff Bonwick enum zio_stage pipeline = ZIO_VDEV_CHILD_PIPELINE; 956e14bb325SJeff Bonwick zio_t *zio; 957e14bb325SJeff Bonwick 958e14bb325SJeff Bonwick ASSERT(vd->vdev_parent == 959e14bb325SJeff Bonwick (pio->io_vd ? pio->io_vd : pio->io_spa->spa_root_vdev)); 960fa9e4066Sahrens 961fa9e4066Sahrens if (type == ZIO_TYPE_READ && bp != NULL) { 962fa9e4066Sahrens /* 963fa9e4066Sahrens * If we have the bp, then the child should perform the 964fa9e4066Sahrens * checksum and the parent need not. This pushes error 965fa9e4066Sahrens * detection as close to the leaves as possible and 966fa9e4066Sahrens * eliminates redundant checksums in the interior nodes. 967fa9e4066Sahrens */ 968b24ab676SJeff Bonwick pipeline |= ZIO_STAGE_CHECKSUM_VERIFY; 969b24ab676SJeff Bonwick pio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY; 970fa9e4066Sahrens } 971fa9e4066Sahrens 972e14bb325SJeff Bonwick if (vd->vdev_children == 0) 973e14bb325SJeff Bonwick offset += VDEV_LABEL_START_SIZE; 974e14bb325SJeff Bonwick 975b24ab676SJeff Bonwick flags |= ZIO_VDEV_CHILD_FLAGS(pio) | ZIO_FLAG_DONT_PROPAGATE; 976b24ab676SJeff Bonwick 977b24ab676SJeff Bonwick /* 978b24ab676SJeff Bonwick * If we've decided to do a repair, the write is not speculative -- 979b24ab676SJeff Bonwick * even if the original read was. 980b24ab676SJeff Bonwick */ 981b24ab676SJeff Bonwick if (flags & ZIO_FLAG_IO_REPAIR) 982b24ab676SJeff Bonwick flags &= ~ZIO_FLAG_SPECULATIVE; 983b24ab676SJeff Bonwick 984e14bb325SJeff Bonwick zio = zio_create(pio, pio->io_spa, pio->io_txg, bp, data, size, 985b24ab676SJeff Bonwick done, private, type, priority, flags, vd, offset, &pio->io_bookmark, 986b24ab676SJeff Bonwick ZIO_STAGE_VDEV_IO_START >> 1, pipeline); 987fa9e4066Sahrens 98869962b56SMatthew Ahrens zio->io_physdone = pio->io_physdone; 98969962b56SMatthew Ahrens if (vd->vdev_ops->vdev_op_leaf && zio->io_logical != NULL) 99069962b56SMatthew Ahrens zio->io_logical->io_phys_children++; 99169962b56SMatthew Ahrens 992e14bb325SJeff Bonwick return (zio); 99332b87932Sek } 99432b87932Sek 995e14bb325SJeff Bonwick zio_t * 996e14bb325SJeff Bonwick zio_vdev_delegated_io(vdev_t *vd, uint64_t offset, void *data, uint64_t size, 99769962b56SMatthew Ahrens int type, zio_priority_t priority, enum zio_flag flags, 998b24ab676SJeff Bonwick zio_done_func_t *done, void *private) 999fa9e4066Sahrens { 1000e14bb325SJeff Bonwick zio_t *zio; 1001fa9e4066Sahrens 1002e14bb325SJeff Bonwick ASSERT(vd->vdev_ops->vdev_op_leaf); 1003fa9e4066Sahrens 1004e14bb325SJeff Bonwick zio = zio_create(NULL, vd->vdev_spa, 0, NULL, 1005e14bb325SJeff Bonwick data, size, done, private, type, priority, 100669962b56SMatthew Ahrens flags | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY | ZIO_FLAG_DELEGATED, 1007e14bb325SJeff Bonwick vd, offset, NULL, 1008b24ab676SJeff Bonwick ZIO_STAGE_VDEV_IO_START >> 1, ZIO_VDEV_CHILD_PIPELINE); 1009fa9e4066Sahrens 1010e14bb325SJeff Bonwick return (zio); 1011e05725b1Sbonwick } 1012e05725b1Sbonwick 1013e05725b1Sbonwick void 1014e14bb325SJeff Bonwick zio_flush(zio_t *zio, vdev_t *vd) 1015e05725b1Sbonwick { 1016e14bb325SJeff Bonwick zio_nowait(zio_ioctl(zio, zio->io_spa, vd, DKIOCFLUSHWRITECACHE, 101769962b56SMatthew Ahrens NULL, NULL, 1018e14bb325SJeff Bonwick ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY)); 1019fa9e4066Sahrens } 1020fa9e4066Sahrens 10216e1f5caaSNeil Perrin void 10226e1f5caaSNeil Perrin zio_shrink(zio_t *zio, uint64_t size) 10236e1f5caaSNeil Perrin { 10246e1f5caaSNeil Perrin ASSERT(zio->io_executor == NULL); 10256e1f5caaSNeil Perrin ASSERT(zio->io_orig_size == zio->io_size); 10266e1f5caaSNeil Perrin ASSERT(size <= zio->io_size); 10276e1f5caaSNeil Perrin 10286e1f5caaSNeil Perrin /* 10296e1f5caaSNeil Perrin * We don't shrink for raidz because of problems with the 10306e1f5caaSNeil Perrin * reconstruction when reading back less than the block size. 10316e1f5caaSNeil Perrin * Note, BP_IS_RAIDZ() assumes no compression. 10326e1f5caaSNeil Perrin */ 10336e1f5caaSNeil Perrin ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF); 10346e1f5caaSNeil Perrin if (!BP_IS_RAIDZ(zio->io_bp)) 10356e1f5caaSNeil Perrin zio->io_orig_size = zio->io_size = size; 10366e1f5caaSNeil Perrin } 10376e1f5caaSNeil Perrin 1038fa9e4066Sahrens /* 1039fa9e4066Sahrens * ========================================================================== 1040e14bb325SJeff Bonwick * Prepare to read and write logical blocks 1041fa9e4066Sahrens * ========================================================================== 1042fa9e4066Sahrens */ 1043e14bb325SJeff Bonwick 1044e05725b1Sbonwick static int 1045e14bb325SJeff Bonwick zio_read_bp_init(zio_t *zio) 1046fa9e4066Sahrens { 1047e14bb325SJeff Bonwick blkptr_t *bp = zio->io_bp; 1048e05725b1Sbonwick 104903361682SJeff Bonwick if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_OFF && 1050f5383399SBill Moore zio->io_child_type == ZIO_CHILD_LOGICAL && 1051f5383399SBill Moore !(zio->io_flags & ZIO_FLAG_RAW)) { 10525d7b4d43SMatthew Ahrens uint64_t psize = 10535d7b4d43SMatthew Ahrens BP_IS_EMBEDDED(bp) ? BPE_GET_PSIZE(bp) : BP_GET_PSIZE(bp); 1054b24ab676SJeff Bonwick void *cbuf = zio_buf_alloc(psize); 1055e05725b1Sbonwick 1056b24ab676SJeff Bonwick zio_push_transform(zio, cbuf, psize, psize, zio_decompress); 1057e14bb325SJeff Bonwick } 1058fa9e4066Sahrens 10595d7b4d43SMatthew Ahrens if (BP_IS_EMBEDDED(bp) && BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA) { 10605d7b4d43SMatthew Ahrens zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; 10615d7b4d43SMatthew Ahrens decode_embedded_bp_compressed(bp, zio->io_data); 10625d7b4d43SMatthew Ahrens } else { 10635d7b4d43SMatthew Ahrens ASSERT(!BP_IS_EMBEDDED(bp)); 10645d7b4d43SMatthew Ahrens } 10655d7b4d43SMatthew Ahrens 1066ad135b5dSChristopher Siden if (!DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) && BP_GET_LEVEL(bp) == 0) 1067e14bb325SJeff Bonwick zio->io_flags |= ZIO_FLAG_DONT_CACHE; 1068fa9e4066Sahrens 1069bbfd46c4SJeff Bonwick if (BP_GET_TYPE(bp) == DMU_OT_DDT_ZAP) 1070bbfd46c4SJeff Bonwick zio->io_flags |= ZIO_FLAG_DONT_CACHE; 1071bbfd46c4SJeff Bonwick 1072b24ab676SJeff Bonwick if (BP_GET_DEDUP(bp) && zio->io_child_type == ZIO_CHILD_LOGICAL) 1073b24ab676SJeff Bonwick zio->io_pipeline = ZIO_DDT_READ_PIPELINE; 1074b24ab676SJeff Bonwick 1075e14bb325SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 1076fa9e4066Sahrens } 1077fa9e4066Sahrens 1078e05725b1Sbonwick static int 1079e14bb325SJeff Bonwick zio_write_bp_init(zio_t *zio) 10800a4e9518Sgw { 1081b24ab676SJeff Bonwick spa_t *spa = zio->io_spa; 1082e14bb325SJeff Bonwick zio_prop_t *zp = &zio->io_prop; 1083b24ab676SJeff Bonwick enum zio_compress compress = zp->zp_compress; 1084e05725b1Sbonwick blkptr_t *bp = zio->io_bp; 1085e14bb325SJeff Bonwick uint64_t lsize = zio->io_size; 1086b24ab676SJeff Bonwick uint64_t psize = lsize; 1087e14bb325SJeff Bonwick int pass = 1; 1088e05725b1Sbonwick 1089e14bb325SJeff Bonwick /* 1090e14bb325SJeff Bonwick * If our children haven't all reached the ready stage, 1091e14bb325SJeff Bonwick * wait for them and then repeat this pipeline stage. 1092e14bb325SJeff Bonwick */ 1093e14bb325SJeff Bonwick if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY) || 1094e14bb325SJeff Bonwick zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_READY)) 1095e14bb325SJeff Bonwick return (ZIO_PIPELINE_STOP); 10960a4e9518Sgw 1097e14bb325SJeff Bonwick if (!IO_IS_ALLOCATING(zio)) 1098e14bb325SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 10990a4e9518Sgw 1100b24ab676SJeff Bonwick ASSERT(zio->io_child_type != ZIO_CHILD_DDT); 1101b24ab676SJeff Bonwick 1102b24ab676SJeff Bonwick if (zio->io_bp_override) { 1103b24ab676SJeff Bonwick ASSERT(bp->blk_birth != zio->io_txg); 1104b24ab676SJeff Bonwick ASSERT(BP_GET_DEDUP(zio->io_bp_override) == 0); 1105b24ab676SJeff Bonwick 1106b24ab676SJeff Bonwick *bp = *zio->io_bp_override; 1107b24ab676SJeff Bonwick zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; 1108b24ab676SJeff Bonwick 11095d7b4d43SMatthew Ahrens if (BP_IS_EMBEDDED(bp)) 11105d7b4d43SMatthew Ahrens return (ZIO_PIPELINE_CONTINUE); 11115d7b4d43SMatthew Ahrens 111280901aeaSGeorge Wilson /* 111380901aeaSGeorge Wilson * If we've been overridden and nopwrite is set then 111480901aeaSGeorge Wilson * set the flag accordingly to indicate that a nopwrite 111580901aeaSGeorge Wilson * has already occurred. 111680901aeaSGeorge Wilson */ 111780901aeaSGeorge Wilson if (!BP_IS_HOLE(bp) && zp->zp_nopwrite) { 111880901aeaSGeorge Wilson ASSERT(!zp->zp_dedup); 111980901aeaSGeorge Wilson zio->io_flags |= ZIO_FLAG_NOPWRITE; 112080901aeaSGeorge Wilson return (ZIO_PIPELINE_CONTINUE); 112180901aeaSGeorge Wilson } 112280901aeaSGeorge Wilson 112380901aeaSGeorge Wilson ASSERT(!zp->zp_nopwrite); 112480901aeaSGeorge Wilson 1125b24ab676SJeff Bonwick if (BP_IS_HOLE(bp) || !zp->zp_dedup) 1126b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 1127b24ab676SJeff Bonwick 1128b24ab676SJeff Bonwick ASSERT(zio_checksum_table[zp->zp_checksum].ci_dedup || 1129b24ab676SJeff Bonwick zp->zp_dedup_verify); 1130b24ab676SJeff Bonwick 1131b24ab676SJeff Bonwick if (BP_GET_CHECKSUM(bp) == zp->zp_checksum) { 1132b24ab676SJeff Bonwick BP_SET_DEDUP(bp, 1); 1133b24ab676SJeff Bonwick zio->io_pipeline |= ZIO_STAGE_DDT_WRITE; 1134b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 1135b24ab676SJeff Bonwick } 1136b24ab676SJeff Bonwick } 11370a4e9518Sgw 113843466aaeSMax Grossman if (!BP_IS_HOLE(bp) && bp->blk_birth == zio->io_txg) { 1139e14bb325SJeff Bonwick /* 1140e14bb325SJeff Bonwick * We're rewriting an existing block, which means we're 1141e14bb325SJeff Bonwick * working on behalf of spa_sync(). For spa_sync() to 1142e14bb325SJeff Bonwick * converge, it must eventually be the case that we don't 1143e14bb325SJeff Bonwick * have to allocate new blocks. But compression changes 1144e14bb325SJeff Bonwick * the blocksize, which forces a reallocate, and makes 1145e14bb325SJeff Bonwick * convergence take longer. Therefore, after the first 1146e14bb325SJeff Bonwick * few passes, stop compressing to ensure convergence. 1147e14bb325SJeff Bonwick */ 1148b24ab676SJeff Bonwick pass = spa_sync_pass(spa); 1149b24ab676SJeff Bonwick 1150b24ab676SJeff Bonwick ASSERT(zio->io_txg == spa_syncing_txg(spa)); 1151b24ab676SJeff Bonwick ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 1152b24ab676SJeff Bonwick ASSERT(!BP_GET_DEDUP(bp)); 1153e05725b1Sbonwick 115401f55e48SGeorge Wilson if (pass >= zfs_sync_pass_dont_compress) 1155e14bb325SJeff Bonwick compress = ZIO_COMPRESS_OFF; 1156e05725b1Sbonwick 1157e14bb325SJeff Bonwick /* Make sure someone doesn't change their mind on overwrites */ 11585d7b4d43SMatthew Ahrens ASSERT(BP_IS_EMBEDDED(bp) || MIN(zp->zp_copies + BP_IS_GANG(bp), 1159b24ab676SJeff Bonwick spa_max_replication(spa)) == BP_GET_NDVAS(bp)); 1160e14bb325SJeff Bonwick } 1161fa9e4066Sahrens 1162e14bb325SJeff Bonwick if (compress != ZIO_COMPRESS_OFF) { 1163b24ab676SJeff Bonwick void *cbuf = zio_buf_alloc(lsize); 1164b24ab676SJeff Bonwick psize = zio_compress_data(compress, zio->io_data, cbuf, lsize); 1165b24ab676SJeff Bonwick if (psize == 0 || psize == lsize) { 1166e14bb325SJeff Bonwick compress = ZIO_COMPRESS_OFF; 1167b24ab676SJeff Bonwick zio_buf_free(cbuf, lsize); 11685d7b4d43SMatthew Ahrens } else if (!zp->zp_dedup && psize <= BPE_PAYLOAD_SIZE && 11695d7b4d43SMatthew Ahrens zp->zp_level == 0 && !DMU_OT_HAS_FILL(zp->zp_type) && 11705d7b4d43SMatthew Ahrens spa_feature_is_enabled(spa, SPA_FEATURE_EMBEDDED_DATA)) { 11715d7b4d43SMatthew Ahrens encode_embedded_bp_compressed(bp, 11725d7b4d43SMatthew Ahrens cbuf, compress, lsize, psize); 11735d7b4d43SMatthew Ahrens BPE_SET_ETYPE(bp, BP_EMBEDDED_TYPE_DATA); 11745d7b4d43SMatthew Ahrens BP_SET_TYPE(bp, zio->io_prop.zp_type); 11755d7b4d43SMatthew Ahrens BP_SET_LEVEL(bp, zio->io_prop.zp_level); 11765d7b4d43SMatthew Ahrens zio_buf_free(cbuf, lsize); 11775d7b4d43SMatthew Ahrens bp->blk_birth = zio->io_txg; 11785d7b4d43SMatthew Ahrens zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; 11795d7b4d43SMatthew Ahrens ASSERT(spa_feature_is_active(spa, 11805d7b4d43SMatthew Ahrens SPA_FEATURE_EMBEDDED_DATA)); 11815d7b4d43SMatthew Ahrens return (ZIO_PIPELINE_CONTINUE); 1182b24ab676SJeff Bonwick } else { 11835d7b4d43SMatthew Ahrens /* 118481cd5c55SMatthew Ahrens * Round up compressed size up to the ashift 118581cd5c55SMatthew Ahrens * of the smallest-ashift device, and zero the tail. 118681cd5c55SMatthew Ahrens * This ensures that the compressed size of the BP 118781cd5c55SMatthew Ahrens * (and thus compressratio property) are correct, 118881cd5c55SMatthew Ahrens * in that we charge for the padding used to fill out 118981cd5c55SMatthew Ahrens * the last sector. 11905d7b4d43SMatthew Ahrens */ 119181cd5c55SMatthew Ahrens ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); 119281cd5c55SMatthew Ahrens size_t rounded = (size_t)P2ROUNDUP(psize, 119381cd5c55SMatthew Ahrens 1ULL << spa->spa_min_ashift); 119481cd5c55SMatthew Ahrens if (rounded >= lsize) { 11955d7b4d43SMatthew Ahrens compress = ZIO_COMPRESS_OFF; 11965d7b4d43SMatthew Ahrens zio_buf_free(cbuf, lsize); 119781cd5c55SMatthew Ahrens psize = lsize; 11985d7b4d43SMatthew Ahrens } else { 119981cd5c55SMatthew Ahrens bzero((char *)cbuf + psize, rounded - psize); 120081cd5c55SMatthew Ahrens psize = rounded; 12015d7b4d43SMatthew Ahrens zio_push_transform(zio, cbuf, 12025d7b4d43SMatthew Ahrens psize, lsize, NULL); 12035d7b4d43SMatthew Ahrens } 1204e14bb325SJeff Bonwick } 1205e14bb325SJeff Bonwick } 1206c717a561Smaybee 1207e14bb325SJeff Bonwick /* 1208e14bb325SJeff Bonwick * The final pass of spa_sync() must be all rewrites, but the first 1209e14bb325SJeff Bonwick * few passes offer a trade-off: allocating blocks defers convergence, 1210e14bb325SJeff Bonwick * but newly allocated blocks are sequential, so they can be written 1211e14bb325SJeff Bonwick * to disk faster. Therefore, we allow the first few passes of 1212e14bb325SJeff Bonwick * spa_sync() to allocate new blocks, but force rewrites after that. 1213e14bb325SJeff Bonwick * There should only be a handful of blocks after pass 1 in any case. 1214e14bb325SJeff Bonwick */ 121543466aaeSMax Grossman if (!BP_IS_HOLE(bp) && bp->blk_birth == zio->io_txg && 121643466aaeSMax Grossman BP_GET_PSIZE(bp) == psize && 121701f55e48SGeorge Wilson pass >= zfs_sync_pass_rewrite) { 1218b24ab676SJeff Bonwick ASSERT(psize != 0); 1219b24ab676SJeff Bonwick enum zio_stage gang_stages = zio->io_pipeline & ZIO_GANG_STAGES; 1220e14bb325SJeff Bonwick zio->io_pipeline = ZIO_REWRITE_PIPELINE | gang_stages; 1221e14bb325SJeff Bonwick zio->io_flags |= ZIO_FLAG_IO_REWRITE; 1222e14bb325SJeff Bonwick } else { 1223e14bb325SJeff Bonwick BP_ZERO(bp); 1224e14bb325SJeff Bonwick zio->io_pipeline = ZIO_WRITE_PIPELINE; 1225e14bb325SJeff Bonwick } 1226fa9e4066Sahrens 1227b24ab676SJeff Bonwick if (psize == 0) { 122843466aaeSMax Grossman if (zio->io_bp_orig.blk_birth != 0 && 122943466aaeSMax Grossman spa_feature_is_active(spa, SPA_FEATURE_HOLE_BIRTH)) { 123043466aaeSMax Grossman BP_SET_LSIZE(bp, lsize); 123143466aaeSMax Grossman BP_SET_TYPE(bp, zp->zp_type); 123243466aaeSMax Grossman BP_SET_LEVEL(bp, zp->zp_level); 123343466aaeSMax Grossman BP_SET_BIRTH(bp, zio->io_txg, 0); 123443466aaeSMax Grossman } 1235e14bb325SJeff Bonwick zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; 1236e14bb325SJeff Bonwick } else { 1237e14bb325SJeff Bonwick ASSERT(zp->zp_checksum != ZIO_CHECKSUM_GANG_HEADER); 1238e14bb325SJeff Bonwick BP_SET_LSIZE(bp, lsize); 123943466aaeSMax Grossman BP_SET_TYPE(bp, zp->zp_type); 124043466aaeSMax Grossman BP_SET_LEVEL(bp, zp->zp_level); 1241b24ab676SJeff Bonwick BP_SET_PSIZE(bp, psize); 1242e14bb325SJeff Bonwick BP_SET_COMPRESS(bp, compress); 1243e14bb325SJeff Bonwick BP_SET_CHECKSUM(bp, zp->zp_checksum); 1244b24ab676SJeff Bonwick BP_SET_DEDUP(bp, zp->zp_dedup); 1245e14bb325SJeff Bonwick BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER); 1246b24ab676SJeff Bonwick if (zp->zp_dedup) { 1247b24ab676SJeff Bonwick ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 1248b24ab676SJeff Bonwick ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE)); 1249b24ab676SJeff Bonwick zio->io_pipeline = ZIO_DDT_WRITE_PIPELINE; 1250b24ab676SJeff Bonwick } 125180901aeaSGeorge Wilson if (zp->zp_nopwrite) { 125280901aeaSGeorge Wilson ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 125380901aeaSGeorge Wilson ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE)); 125480901aeaSGeorge Wilson zio->io_pipeline |= ZIO_STAGE_NOP_WRITE; 125580901aeaSGeorge Wilson } 1256b24ab676SJeff Bonwick } 1257b24ab676SJeff Bonwick 1258b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 1259b24ab676SJeff Bonwick } 1260b24ab676SJeff Bonwick 1261b24ab676SJeff Bonwick static int 1262b24ab676SJeff Bonwick zio_free_bp_init(zio_t *zio) 1263b24ab676SJeff Bonwick { 1264b24ab676SJeff Bonwick blkptr_t *bp = zio->io_bp; 1265b24ab676SJeff Bonwick 1266b24ab676SJeff Bonwick if (zio->io_child_type == ZIO_CHILD_LOGICAL) { 1267b24ab676SJeff Bonwick if (BP_GET_DEDUP(bp)) 1268b24ab676SJeff Bonwick zio->io_pipeline = ZIO_DDT_FREE_PIPELINE; 1269e14bb325SJeff Bonwick } 1270fa9e4066Sahrens 1271e05725b1Sbonwick return (ZIO_PIPELINE_CONTINUE); 1272fa9e4066Sahrens } 1273fa9e4066Sahrens 1274e14bb325SJeff Bonwick /* 1275e14bb325SJeff Bonwick * ========================================================================== 1276e14bb325SJeff Bonwick * Execute the I/O pipeline 1277e14bb325SJeff Bonwick * ========================================================================== 1278e14bb325SJeff Bonwick */ 1279e14bb325SJeff Bonwick 1280e14bb325SJeff Bonwick static void 1281ec94d322SAdam Leventhal zio_taskq_dispatch(zio_t *zio, zio_taskq_type_t q, boolean_t cutinline) 1282fa9e4066Sahrens { 128380eb36f2SGeorge Wilson spa_t *spa = zio->io_spa; 1284e14bb325SJeff Bonwick zio_type_t t = zio->io_type; 12855aeb9474SGarrett D'Amore int flags = (cutinline ? TQ_FRONT : 0); 12860a4e9518Sgw 12870a4e9518Sgw /* 1288bbe36defSGeorge Wilson * If we're a config writer or a probe, the normal issue and 1289bbe36defSGeorge Wilson * interrupt threads may all be blocked waiting for the config lock. 1290bbe36defSGeorge Wilson * In this case, select the otherwise-unused taskq for ZIO_TYPE_NULL. 12910a4e9518Sgw */ 1292bbe36defSGeorge Wilson if (zio->io_flags & (ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_PROBE)) 1293e14bb325SJeff Bonwick t = ZIO_TYPE_NULL; 12940a4e9518Sgw 12950a4e9518Sgw /* 1296e14bb325SJeff Bonwick * A similar issue exists for the L2ARC write thread until L2ARC 2.0. 12970a4e9518Sgw */ 1298e14bb325SJeff Bonwick if (t == ZIO_TYPE_WRITE && zio->io_vd && zio->io_vd->vdev_aux) 1299e14bb325SJeff Bonwick t = ZIO_TYPE_NULL; 13000a4e9518Sgw 130180eb36f2SGeorge Wilson /* 1302ec94d322SAdam Leventhal * If this is a high priority I/O, then use the high priority taskq if 1303ec94d322SAdam Leventhal * available. 130480eb36f2SGeorge Wilson */ 130580eb36f2SGeorge Wilson if (zio->io_priority == ZIO_PRIORITY_NOW && 1306ec94d322SAdam Leventhal spa->spa_zio_taskq[t][q + 1].stqs_count != 0) 130780eb36f2SGeorge Wilson q++; 130880eb36f2SGeorge Wilson 130980eb36f2SGeorge Wilson ASSERT3U(q, <, ZIO_TASKQ_TYPES); 13105aeb9474SGarrett D'Amore 13115aeb9474SGarrett D'Amore /* 13125aeb9474SGarrett D'Amore * NB: We are assuming that the zio can only be dispatched 13135aeb9474SGarrett D'Amore * to a single taskq at a time. It would be a grievous error 13145aeb9474SGarrett D'Amore * to dispatch the zio to another taskq at the same time. 13155aeb9474SGarrett D'Amore */ 13165aeb9474SGarrett D'Amore ASSERT(zio->io_tqent.tqent_next == NULL); 1317ec94d322SAdam Leventhal spa_taskq_dispatch_ent(spa, t, q, (task_func_t *)zio_execute, zio, 1318ec94d322SAdam Leventhal flags, &zio->io_tqent); 1319e14bb325SJeff Bonwick } 13200a4e9518Sgw 1321e14bb325SJeff Bonwick static boolean_t 1322ec94d322SAdam Leventhal zio_taskq_member(zio_t *zio, zio_taskq_type_t q) 1323e14bb325SJeff Bonwick { 1324e14bb325SJeff Bonwick kthread_t *executor = zio->io_executor; 1325e14bb325SJeff Bonwick spa_t *spa = zio->io_spa; 13260a4e9518Sgw 1327ec94d322SAdam Leventhal for (zio_type_t t = 0; t < ZIO_TYPES; t++) { 1328ec94d322SAdam Leventhal spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 1329ec94d322SAdam Leventhal uint_t i; 1330ec94d322SAdam Leventhal for (i = 0; i < tqs->stqs_count; i++) { 1331ec94d322SAdam Leventhal if (taskq_member(tqs->stqs_taskq[i], executor)) 1332ec94d322SAdam Leventhal return (B_TRUE); 1333ec94d322SAdam Leventhal } 1334ec94d322SAdam Leventhal } 13350a4e9518Sgw 1336e14bb325SJeff Bonwick return (B_FALSE); 1337e14bb325SJeff Bonwick } 1338e05725b1Sbonwick 1339e14bb325SJeff Bonwick static int 1340e14bb325SJeff Bonwick zio_issue_async(zio_t *zio) 1341e14bb325SJeff Bonwick { 134235a5a358SJonathan Adams zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE); 1343e14bb325SJeff Bonwick 1344e14bb325SJeff Bonwick return (ZIO_PIPELINE_STOP); 13450a4e9518Sgw } 13460a4e9518Sgw 1347e14bb325SJeff Bonwick void 1348e14bb325SJeff Bonwick zio_interrupt(zio_t *zio) 13490a4e9518Sgw { 135035a5a358SJonathan Adams zio_taskq_dispatch(zio, ZIO_TASKQ_INTERRUPT, B_FALSE); 1351e14bb325SJeff Bonwick } 13520a4e9518Sgw 1353e14bb325SJeff Bonwick /* 1354e14bb325SJeff Bonwick * Execute the I/O pipeline until one of the following occurs: 1355f7170741SWill Andrews * 1356f7170741SWill Andrews * (1) the I/O completes 1357f7170741SWill Andrews * (2) the pipeline stalls waiting for dependent child I/Os 1358f7170741SWill Andrews * (3) the I/O issues, so we're waiting for an I/O completion interrupt 1359f7170741SWill Andrews * (4) the I/O is delegated by vdev-level caching or aggregation 1360f7170741SWill Andrews * (5) the I/O is deferred due to vdev-level queueing 1361f7170741SWill Andrews * (6) the I/O is handed off to another thread. 1362f7170741SWill Andrews * 1363f7170741SWill Andrews * In all cases, the pipeline stops whenever there's no CPU work; it never 1364f7170741SWill Andrews * burns a thread in cv_wait(). 1365e14bb325SJeff Bonwick * 1366e14bb325SJeff Bonwick * There's no locking on io_stage because there's no legitimate way 1367e14bb325SJeff Bonwick * for multiple threads to be attempting to process the same I/O. 1368e14bb325SJeff Bonwick */ 1369b24ab676SJeff Bonwick static zio_pipe_stage_t *zio_pipeline[]; 13700a4e9518Sgw 1371e14bb325SJeff Bonwick void 1372e14bb325SJeff Bonwick zio_execute(zio_t *zio) 1373e14bb325SJeff Bonwick { 1374e14bb325SJeff Bonwick zio->io_executor = curthread; 13750a4e9518Sgw 1376e14bb325SJeff Bonwick while (zio->io_stage < ZIO_STAGE_DONE) { 1377b24ab676SJeff Bonwick enum zio_stage pipeline = zio->io_pipeline; 1378b24ab676SJeff Bonwick enum zio_stage stage = zio->io_stage; 1379e14bb325SJeff Bonwick int rv; 13800a4e9518Sgw 1381e14bb325SJeff Bonwick ASSERT(!MUTEX_HELD(&zio->io_lock)); 1382b24ab676SJeff Bonwick ASSERT(ISP2(stage)); 1383b24ab676SJeff Bonwick ASSERT(zio->io_stall == NULL); 13840a4e9518Sgw 1385b24ab676SJeff Bonwick do { 1386b24ab676SJeff Bonwick stage <<= 1; 1387b24ab676SJeff Bonwick } while ((stage & pipeline) == 0); 1388e14bb325SJeff Bonwick 1389e14bb325SJeff Bonwick ASSERT(stage <= ZIO_STAGE_DONE); 13900a4e9518Sgw 13910a4e9518Sgw /* 1392e14bb325SJeff Bonwick * If we are in interrupt context and this pipeline stage 1393e14bb325SJeff Bonwick * will grab a config lock that is held across I/O, 1394b24ab676SJeff Bonwick * or may wait for an I/O that needs an interrupt thread 1395b24ab676SJeff Bonwick * to complete, issue async to avoid deadlock. 139635a5a358SJonathan Adams * 139735a5a358SJonathan Adams * For VDEV_IO_START, we cut in line so that the io will 139835a5a358SJonathan Adams * be sent to disk promptly. 13990a4e9518Sgw */ 1400b24ab676SJeff Bonwick if ((stage & ZIO_BLOCKING_STAGES) && zio->io_vd == NULL && 1401e14bb325SJeff Bonwick zio_taskq_member(zio, ZIO_TASKQ_INTERRUPT)) { 140235a5a358SJonathan Adams boolean_t cut = (stage == ZIO_STAGE_VDEV_IO_START) ? 140335a5a358SJonathan Adams zio_requeue_io_start_cut_in_line : B_FALSE; 140435a5a358SJonathan Adams zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, cut); 1405e14bb325SJeff Bonwick return; 14060a4e9518Sgw } 14070a4e9518Sgw 1408e14bb325SJeff Bonwick zio->io_stage = stage; 1409bf16b11eSMatthew Ahrens rv = zio_pipeline[highbit64(stage) - 1](zio); 14100a4e9518Sgw 1411e14bb325SJeff Bonwick if (rv == ZIO_PIPELINE_STOP) 1412e14bb325SJeff Bonwick return; 14130a4e9518Sgw 1414e14bb325SJeff Bonwick ASSERT(rv == ZIO_PIPELINE_CONTINUE); 1415e14bb325SJeff Bonwick } 14160a4e9518Sgw } 14170a4e9518Sgw 1418e14bb325SJeff Bonwick /* 1419e14bb325SJeff Bonwick * ========================================================================== 1420e14bb325SJeff Bonwick * Initiate I/O, either sync or async 1421e14bb325SJeff Bonwick * ========================================================================== 1422e14bb325SJeff Bonwick */ 1423e14bb325SJeff Bonwick int 1424e14bb325SJeff Bonwick zio_wait(zio_t *zio) 14250a4e9518Sgw { 1426e14bb325SJeff Bonwick int error; 14270a4e9518Sgw 1428e14bb325SJeff Bonwick ASSERT(zio->io_stage == ZIO_STAGE_OPEN); 1429e14bb325SJeff Bonwick ASSERT(zio->io_executor == NULL); 14300a4e9518Sgw 1431e14bb325SJeff Bonwick zio->io_waiter = curthread; 1432e05725b1Sbonwick 1433e14bb325SJeff Bonwick zio_execute(zio); 14340a4e9518Sgw 1435e14bb325SJeff Bonwick mutex_enter(&zio->io_lock); 1436e14bb325SJeff Bonwick while (zio->io_executor != NULL) 1437e14bb325SJeff Bonwick cv_wait(&zio->io_cv, &zio->io_lock); 1438e14bb325SJeff Bonwick mutex_exit(&zio->io_lock); 143932b87932Sek 1440e14bb325SJeff Bonwick error = zio->io_error; 1441e14bb325SJeff Bonwick zio_destroy(zio); 144232b87932Sek 1443e14bb325SJeff Bonwick return (error); 144432b87932Sek } 144532b87932Sek 1446e14bb325SJeff Bonwick void 1447e14bb325SJeff Bonwick zio_nowait(zio_t *zio) 14480a4e9518Sgw { 1449e14bb325SJeff Bonwick ASSERT(zio->io_executor == NULL); 1450fa9e4066Sahrens 1451a3f829aeSBill Moore if (zio->io_child_type == ZIO_CHILD_LOGICAL && 1452a3f829aeSBill Moore zio_unique_parent(zio) == NULL) { 1453ea8dc4b6Seschrock /* 1454e14bb325SJeff Bonwick * This is a logical async I/O with no parent to wait for it. 145554d692b7SGeorge Wilson * We add it to the spa_async_root_zio "Godfather" I/O which 145654d692b7SGeorge Wilson * will ensure they complete prior to unloading the pool. 1457ea8dc4b6Seschrock */ 1458e14bb325SJeff Bonwick spa_t *spa = zio->io_spa; 145954d692b7SGeorge Wilson 14606f834bc1SMatthew Ahrens zio_add_child(spa->spa_async_zio_root[CPU_SEQID], zio); 1461e14bb325SJeff Bonwick } 1462ea8dc4b6Seschrock 1463e14bb325SJeff Bonwick zio_execute(zio); 1464e14bb325SJeff Bonwick } 1465ea8dc4b6Seschrock 1466e14bb325SJeff Bonwick /* 1467e14bb325SJeff Bonwick * ========================================================================== 1468e14bb325SJeff Bonwick * Reexecute or suspend/resume failed I/O 1469e14bb325SJeff Bonwick * ========================================================================== 1470e14bb325SJeff Bonwick */ 1471fa9e4066Sahrens 1472e14bb325SJeff Bonwick static void 1473e14bb325SJeff Bonwick zio_reexecute(zio_t *pio) 1474e14bb325SJeff Bonwick { 1475a3f829aeSBill Moore zio_t *cio, *cio_next; 1476a3f829aeSBill Moore 1477a3f829aeSBill Moore ASSERT(pio->io_child_type == ZIO_CHILD_LOGICAL); 1478a3f829aeSBill Moore ASSERT(pio->io_orig_stage == ZIO_STAGE_OPEN); 1479f5383399SBill Moore ASSERT(pio->io_gang_leader == NULL); 1480f5383399SBill Moore ASSERT(pio->io_gang_tree == NULL); 1481e05725b1Sbonwick 1482e14bb325SJeff Bonwick pio->io_flags = pio->io_orig_flags; 1483e14bb325SJeff Bonwick pio->io_stage = pio->io_orig_stage; 1484e14bb325SJeff Bonwick pio->io_pipeline = pio->io_orig_pipeline; 1485e14bb325SJeff Bonwick pio->io_reexecute = 0; 148680901aeaSGeorge Wilson pio->io_flags |= ZIO_FLAG_REEXECUTED; 1487e14bb325SJeff Bonwick pio->io_error = 0; 1488a3f829aeSBill Moore for (int w = 0; w < ZIO_WAIT_TYPES; w++) 1489a3f829aeSBill Moore pio->io_state[w] = 0; 1490e14bb325SJeff Bonwick for (int c = 0; c < ZIO_CHILD_TYPES; c++) 1491e14bb325SJeff Bonwick pio->io_child_error[c] = 0; 14920a4e9518Sgw 1493b24ab676SJeff Bonwick if (IO_IS_ALLOCATING(pio)) 1494b24ab676SJeff Bonwick BP_ZERO(pio->io_bp); 1495d58459f4Sek 1496e14bb325SJeff Bonwick /* 1497e14bb325SJeff Bonwick * As we reexecute pio's children, new children could be created. 1498a3f829aeSBill Moore * New children go to the head of pio's io_child_list, however, 1499e14bb325SJeff Bonwick * so we will (correctly) not reexecute them. The key is that 1500a3f829aeSBill Moore * the remainder of pio's io_child_list, from 'cio_next' onward, 1501a3f829aeSBill Moore * cannot be affected by any side effects of reexecuting 'cio'. 1502e14bb325SJeff Bonwick */ 1503a3f829aeSBill Moore for (cio = zio_walk_children(pio); cio != NULL; cio = cio_next) { 1504a3f829aeSBill Moore cio_next = zio_walk_children(pio); 1505e14bb325SJeff Bonwick mutex_enter(&pio->io_lock); 1506a3f829aeSBill Moore for (int w = 0; w < ZIO_WAIT_TYPES; w++) 1507a3f829aeSBill Moore pio->io_children[cio->io_child_type][w]++; 1508e14bb325SJeff Bonwick mutex_exit(&pio->io_lock); 1509a3f829aeSBill Moore zio_reexecute(cio); 1510fa9e4066Sahrens } 1511e05725b1Sbonwick 1512e14bb325SJeff Bonwick /* 1513e14bb325SJeff Bonwick * Now that all children have been reexecuted, execute the parent. 151454d692b7SGeorge Wilson * We don't reexecute "The Godfather" I/O here as it's the 151554d692b7SGeorge Wilson * responsibility of the caller to wait on him. 1516e14bb325SJeff Bonwick */ 151754d692b7SGeorge Wilson if (!(pio->io_flags & ZIO_FLAG_GODFATHER)) 151854d692b7SGeorge Wilson zio_execute(pio); 15190a4e9518Sgw } 15200a4e9518Sgw 1521e14bb325SJeff Bonwick void 1522e14bb325SJeff Bonwick zio_suspend(spa_t *spa, zio_t *zio) 15230a4e9518Sgw { 1524e14bb325SJeff Bonwick if (spa_get_failmode(spa) == ZIO_FAILURE_MODE_PANIC) 1525e14bb325SJeff Bonwick fm_panic("Pool '%s' has encountered an uncorrectable I/O " 1526e14bb325SJeff Bonwick "failure and the failure mode property for this pool " 1527e14bb325SJeff Bonwick "is set to panic.", spa_name(spa)); 15280a4e9518Sgw 1529e14bb325SJeff Bonwick zfs_ereport_post(FM_EREPORT_ZFS_IO_FAILURE, spa, NULL, NULL, 0, 0); 15300a4e9518Sgw 1531e14bb325SJeff Bonwick mutex_enter(&spa->spa_suspend_lock); 1532fa9e4066Sahrens 1533e14bb325SJeff Bonwick if (spa->spa_suspend_zio_root == NULL) 153454d692b7SGeorge Wilson spa->spa_suspend_zio_root = zio_root(spa, NULL, NULL, 153554d692b7SGeorge Wilson ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 153654d692b7SGeorge Wilson ZIO_FLAG_GODFATHER); 1537fa9e4066Sahrens 1538e14bb325SJeff Bonwick spa->spa_suspended = B_TRUE; 1539fa9e4066Sahrens 1540e14bb325SJeff Bonwick if (zio != NULL) { 154154d692b7SGeorge Wilson ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER)); 1542e14bb325SJeff Bonwick ASSERT(zio != spa->spa_suspend_zio_root); 1543e14bb325SJeff Bonwick ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 1544a3f829aeSBill Moore ASSERT(zio_unique_parent(zio) == NULL); 1545e14bb325SJeff Bonwick ASSERT(zio->io_stage == ZIO_STAGE_DONE); 1546e14bb325SJeff Bonwick zio_add_child(spa->spa_suspend_zio_root, zio); 1547e14bb325SJeff Bonwick } 1548fa9e4066Sahrens 1549e14bb325SJeff Bonwick mutex_exit(&spa->spa_suspend_lock); 1550e14bb325SJeff Bonwick } 1551fa9e4066Sahrens 155254d692b7SGeorge Wilson int 1553e14bb325SJeff Bonwick zio_resume(spa_t *spa) 1554e14bb325SJeff Bonwick { 155554d692b7SGeorge Wilson zio_t *pio; 1556fa9e4066Sahrens 1557b3995adbSahrens /* 1558e14bb325SJeff Bonwick * Reexecute all previously suspended i/o. 1559b3995adbSahrens */ 1560e14bb325SJeff Bonwick mutex_enter(&spa->spa_suspend_lock); 1561e14bb325SJeff Bonwick spa->spa_suspended = B_FALSE; 1562e14bb325SJeff Bonwick cv_broadcast(&spa->spa_suspend_cv); 1563e14bb325SJeff Bonwick pio = spa->spa_suspend_zio_root; 1564e14bb325SJeff Bonwick spa->spa_suspend_zio_root = NULL; 1565e14bb325SJeff Bonwick mutex_exit(&spa->spa_suspend_lock); 1566e14bb325SJeff Bonwick 1567e14bb325SJeff Bonwick if (pio == NULL) 156854d692b7SGeorge Wilson return (0); 1569e14bb325SJeff Bonwick 157054d692b7SGeorge Wilson zio_reexecute(pio); 157154d692b7SGeorge Wilson return (zio_wait(pio)); 1572e14bb325SJeff Bonwick } 1573e14bb325SJeff Bonwick 1574e14bb325SJeff Bonwick void 1575e14bb325SJeff Bonwick zio_resume_wait(spa_t *spa) 1576e14bb325SJeff Bonwick { 1577e14bb325SJeff Bonwick mutex_enter(&spa->spa_suspend_lock); 1578e14bb325SJeff Bonwick while (spa_suspended(spa)) 1579e14bb325SJeff Bonwick cv_wait(&spa->spa_suspend_cv, &spa->spa_suspend_lock); 1580e14bb325SJeff Bonwick mutex_exit(&spa->spa_suspend_lock); 1581fa9e4066Sahrens } 1582fa9e4066Sahrens 1583fa9e4066Sahrens /* 1584fa9e4066Sahrens * ========================================================================== 1585e14bb325SJeff Bonwick * Gang blocks. 1586e14bb325SJeff Bonwick * 1587e14bb325SJeff Bonwick * A gang block is a collection of small blocks that looks to the DMU 1588e14bb325SJeff Bonwick * like one large block. When zio_dva_allocate() cannot find a block 1589e14bb325SJeff Bonwick * of the requested size, due to either severe fragmentation or the pool 1590e14bb325SJeff Bonwick * being nearly full, it calls zio_write_gang_block() to construct the 1591e14bb325SJeff Bonwick * block from smaller fragments. 1592e14bb325SJeff Bonwick * 1593e14bb325SJeff Bonwick * A gang block consists of a gang header (zio_gbh_phys_t) and up to 1594e14bb325SJeff Bonwick * three (SPA_GBH_NBLKPTRS) gang members. The gang header is just like 1595e14bb325SJeff Bonwick * an indirect block: it's an array of block pointers. It consumes 1596e14bb325SJeff Bonwick * only one sector and hence is allocatable regardless of fragmentation. 1597e14bb325SJeff Bonwick * The gang header's bps point to its gang members, which hold the data. 1598e14bb325SJeff Bonwick * 1599e14bb325SJeff Bonwick * Gang blocks are self-checksumming, using the bp's <vdev, offset, txg> 1600e14bb325SJeff Bonwick * as the verifier to ensure uniqueness of the SHA256 checksum. 1601e14bb325SJeff Bonwick * Critically, the gang block bp's blk_cksum is the checksum of the data, 1602e14bb325SJeff Bonwick * not the gang header. This ensures that data block signatures (needed for 1603e14bb325SJeff Bonwick * deduplication) are independent of how the block is physically stored. 1604e14bb325SJeff Bonwick * 1605e14bb325SJeff Bonwick * Gang blocks can be nested: a gang member may itself be a gang block. 1606e14bb325SJeff Bonwick * Thus every gang block is a tree in which root and all interior nodes are 1607e14bb325SJeff Bonwick * gang headers, and the leaves are normal blocks that contain user data. 1608e14bb325SJeff Bonwick * The root of the gang tree is called the gang leader. 1609e14bb325SJeff Bonwick * 1610e14bb325SJeff Bonwick * To perform any operation (read, rewrite, free, claim) on a gang block, 1611e14bb325SJeff Bonwick * zio_gang_assemble() first assembles the gang tree (minus data leaves) 1612e14bb325SJeff Bonwick * in the io_gang_tree field of the original logical i/o by recursively 1613e14bb325SJeff Bonwick * reading the gang leader and all gang headers below it. This yields 1614e14bb325SJeff Bonwick * an in-core tree containing the contents of every gang header and the 1615e14bb325SJeff Bonwick * bps for every constituent of the gang block. 1616e14bb325SJeff Bonwick * 1617e14bb325SJeff Bonwick * With the gang tree now assembled, zio_gang_issue() just walks the gang tree 1618e14bb325SJeff Bonwick * and invokes a callback on each bp. To free a gang block, zio_gang_issue() 1619e14bb325SJeff Bonwick * calls zio_free_gang() -- a trivial wrapper around zio_free() -- for each bp. 1620e14bb325SJeff Bonwick * zio_claim_gang() provides a similarly trivial wrapper for zio_claim(). 1621e14bb325SJeff Bonwick * zio_read_gang() is a wrapper around zio_read() that omits reading gang 1622e14bb325SJeff Bonwick * headers, since we already have those in io_gang_tree. zio_rewrite_gang() 1623e14bb325SJeff Bonwick * performs a zio_rewrite() of the data or, for gang headers, a zio_rewrite() 1624e14bb325SJeff Bonwick * of the gang header plus zio_checksum_compute() of the data to update the 1625e14bb325SJeff Bonwick * gang header's blk_cksum as described above. 1626e14bb325SJeff Bonwick * 1627e14bb325SJeff Bonwick * The two-phase assemble/issue model solves the problem of partial failure -- 1628e14bb325SJeff Bonwick * what if you'd freed part of a gang block but then couldn't read the 1629e14bb325SJeff Bonwick * gang header for another part? Assembling the entire gang tree first 1630e14bb325SJeff Bonwick * ensures that all the necessary gang header I/O has succeeded before 1631e14bb325SJeff Bonwick * starting the actual work of free, claim, or write. Once the gang tree 1632e14bb325SJeff Bonwick * is assembled, free and claim are in-memory operations that cannot fail. 1633e14bb325SJeff Bonwick * 1634e14bb325SJeff Bonwick * In the event that a gang write fails, zio_dva_unallocate() walks the 1635e14bb325SJeff Bonwick * gang tree to immediately free (i.e. insert back into the space map) 1636e14bb325SJeff Bonwick * everything we've allocated. This ensures that we don't get ENOSPC 1637e14bb325SJeff Bonwick * errors during repeated suspend/resume cycles due to a flaky device. 1638e14bb325SJeff Bonwick * 1639e14bb325SJeff Bonwick * Gang rewrites only happen during sync-to-convergence. If we can't assemble 1640e14bb325SJeff Bonwick * the gang tree, we won't modify the block, so we can safely defer the free 1641e14bb325SJeff Bonwick * (knowing that the block is still intact). If we *can* assemble the gang 1642e14bb325SJeff Bonwick * tree, then even if some of the rewrites fail, zio_dva_unallocate() will free 1643e14bb325SJeff Bonwick * each constituent bp and we can allocate a new block on the next sync pass. 1644e14bb325SJeff Bonwick * 1645e14bb325SJeff Bonwick * In all cases, the gang tree allows complete recovery from partial failure. 1646fa9e4066Sahrens * ========================================================================== 1647fa9e4066Sahrens */ 1648e14bb325SJeff Bonwick 1649e14bb325SJeff Bonwick static zio_t * 1650e14bb325SJeff Bonwick zio_read_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) 1651fa9e4066Sahrens { 1652e14bb325SJeff Bonwick if (gn != NULL) 1653e14bb325SJeff Bonwick return (pio); 1654fa9e4066Sahrens 1655e14bb325SJeff Bonwick return (zio_read(pio, pio->io_spa, bp, data, BP_GET_PSIZE(bp), 1656e14bb325SJeff Bonwick NULL, NULL, pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), 1657e14bb325SJeff Bonwick &pio->io_bookmark)); 1658e14bb325SJeff Bonwick } 1659e14bb325SJeff Bonwick 1660e14bb325SJeff Bonwick zio_t * 1661e14bb325SJeff Bonwick zio_rewrite_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) 1662e14bb325SJeff Bonwick { 1663e14bb325SJeff Bonwick zio_t *zio; 1664e14bb325SJeff Bonwick 1665e14bb325SJeff Bonwick if (gn != NULL) { 1666e14bb325SJeff Bonwick zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp, 1667e14bb325SJeff Bonwick gn->gn_gbh, SPA_GANGBLOCKSIZE, NULL, NULL, pio->io_priority, 1668e14bb325SJeff Bonwick ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); 1669fa9e4066Sahrens /* 1670e14bb325SJeff Bonwick * As we rewrite each gang header, the pipeline will compute 1671e14bb325SJeff Bonwick * a new gang block header checksum for it; but no one will 1672e14bb325SJeff Bonwick * compute a new data checksum, so we do that here. The one 1673e14bb325SJeff Bonwick * exception is the gang leader: the pipeline already computed 1674e14bb325SJeff Bonwick * its data checksum because that stage precedes gang assembly. 1675e14bb325SJeff Bonwick * (Presently, nothing actually uses interior data checksums; 1676e14bb325SJeff Bonwick * this is just good hygiene.) 1677fa9e4066Sahrens */ 1678f5383399SBill Moore if (gn != pio->io_gang_leader->io_gang_tree) { 1679e14bb325SJeff Bonwick zio_checksum_compute(zio, BP_GET_CHECKSUM(bp), 1680e14bb325SJeff Bonwick data, BP_GET_PSIZE(bp)); 1681e14bb325SJeff Bonwick } 1682b24ab676SJeff Bonwick /* 1683b24ab676SJeff Bonwick * If we are here to damage data for testing purposes, 1684b24ab676SJeff Bonwick * leave the GBH alone so that we can detect the damage. 1685b24ab676SJeff Bonwick */ 1686b24ab676SJeff Bonwick if (pio->io_gang_leader->io_flags & ZIO_FLAG_INDUCE_DAMAGE) 1687b24ab676SJeff Bonwick zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES; 1688fa9e4066Sahrens } else { 1689e14bb325SJeff Bonwick zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp, 1690e14bb325SJeff Bonwick data, BP_GET_PSIZE(bp), NULL, NULL, pio->io_priority, 1691e14bb325SJeff Bonwick ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); 1692fa9e4066Sahrens } 1693fa9e4066Sahrens 1694e14bb325SJeff Bonwick return (zio); 1695e14bb325SJeff Bonwick } 1696fa9e4066Sahrens 1697e14bb325SJeff Bonwick /* ARGSUSED */ 1698e14bb325SJeff Bonwick zio_t * 1699e14bb325SJeff Bonwick zio_free_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) 1700e14bb325SJeff Bonwick { 1701b24ab676SJeff Bonwick return (zio_free_sync(pio, pio->io_spa, pio->io_txg, bp, 1702b24ab676SJeff Bonwick ZIO_GANG_CHILD_FLAGS(pio))); 1703fa9e4066Sahrens } 1704fa9e4066Sahrens 1705e14bb325SJeff Bonwick /* ARGSUSED */ 1706e14bb325SJeff Bonwick zio_t * 1707e14bb325SJeff Bonwick zio_claim_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) 1708fa9e4066Sahrens { 1709e14bb325SJeff Bonwick return (zio_claim(pio, pio->io_spa, pio->io_txg, bp, 1710e14bb325SJeff Bonwick NULL, NULL, ZIO_GANG_CHILD_FLAGS(pio))); 1711e14bb325SJeff Bonwick } 1712fa9e4066Sahrens 1713e14bb325SJeff Bonwick static zio_gang_issue_func_t *zio_gang_issue_func[ZIO_TYPES] = { 1714e14bb325SJeff Bonwick NULL, 1715e14bb325SJeff Bonwick zio_read_gang, 1716e14bb325SJeff Bonwick zio_rewrite_gang, 1717e14bb325SJeff Bonwick zio_free_gang, 1718e14bb325SJeff Bonwick zio_claim_gang, 1719e14bb325SJeff Bonwick NULL 1720e14bb325SJeff Bonwick }; 1721fa9e4066Sahrens 1722e14bb325SJeff Bonwick static void zio_gang_tree_assemble_done(zio_t *zio); 1723fa9e4066Sahrens 1724e14bb325SJeff Bonwick static zio_gang_node_t * 1725e14bb325SJeff Bonwick zio_gang_node_alloc(zio_gang_node_t **gnpp) 1726e14bb325SJeff Bonwick { 1727e14bb325SJeff Bonwick zio_gang_node_t *gn; 1728fa9e4066Sahrens 1729e14bb325SJeff Bonwick ASSERT(*gnpp == NULL); 1730fa9e4066Sahrens 1731e14bb325SJeff Bonwick gn = kmem_zalloc(sizeof (*gn), KM_SLEEP); 1732e14bb325SJeff Bonwick gn->gn_gbh = zio_buf_alloc(SPA_GANGBLOCKSIZE); 1733e14bb325SJeff Bonwick *gnpp = gn; 1734e14bb325SJeff Bonwick 1735e14bb325SJeff Bonwick return (gn); 1736fa9e4066Sahrens } 1737fa9e4066Sahrens 1738fa9e4066Sahrens static void 1739e14bb325SJeff Bonwick zio_gang_node_free(zio_gang_node_t **gnpp) 1740fa9e4066Sahrens { 1741e14bb325SJeff Bonwick zio_gang_node_t *gn = *gnpp; 1742fa9e4066Sahrens 1743e14bb325SJeff Bonwick for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) 1744e14bb325SJeff Bonwick ASSERT(gn->gn_child[g] == NULL); 1745e14bb325SJeff Bonwick 1746e14bb325SJeff Bonwick zio_buf_free(gn->gn_gbh, SPA_GANGBLOCKSIZE); 1747e14bb325SJeff Bonwick kmem_free(gn, sizeof (*gn)); 1748e14bb325SJeff Bonwick *gnpp = NULL; 1749fa9e4066Sahrens } 1750fa9e4066Sahrens 1751e14bb325SJeff Bonwick static void 1752e14bb325SJeff Bonwick zio_gang_tree_free(zio_gang_node_t **gnpp) 1753fa9e4066Sahrens { 1754e14bb325SJeff Bonwick zio_gang_node_t *gn = *gnpp; 1755fa9e4066Sahrens 1756e14bb325SJeff Bonwick if (gn == NULL) 1757e14bb325SJeff Bonwick return; 1758fa9e4066Sahrens 1759e14bb325SJeff Bonwick for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) 1760e14bb325SJeff Bonwick zio_gang_tree_free(&gn->gn_child[g]); 1761fa9e4066Sahrens 1762e14bb325SJeff Bonwick zio_gang_node_free(gnpp); 1763fa9e4066Sahrens } 1764fa9e4066Sahrens 1765e14bb325SJeff Bonwick static void 1766f5383399SBill Moore zio_gang_tree_assemble(zio_t *gio, blkptr_t *bp, zio_gang_node_t **gnpp) 1767fa9e4066Sahrens { 1768e14bb325SJeff Bonwick zio_gang_node_t *gn = zio_gang_node_alloc(gnpp); 1769e14bb325SJeff Bonwick 1770f5383399SBill Moore ASSERT(gio->io_gang_leader == gio); 1771e14bb325SJeff Bonwick ASSERT(BP_IS_GANG(bp)); 1772fa9e4066Sahrens 1773f5383399SBill Moore zio_nowait(zio_read(gio, gio->io_spa, bp, gn->gn_gbh, 1774e14bb325SJeff Bonwick SPA_GANGBLOCKSIZE, zio_gang_tree_assemble_done, gn, 1775f5383399SBill Moore gio->io_priority, ZIO_GANG_CHILD_FLAGS(gio), &gio->io_bookmark)); 1776e14bb325SJeff Bonwick } 1777fa9e4066Sahrens 1778e14bb325SJeff Bonwick static void 1779e14bb325SJeff Bonwick zio_gang_tree_assemble_done(zio_t *zio) 1780e14bb325SJeff Bonwick { 1781f5383399SBill Moore zio_t *gio = zio->io_gang_leader; 1782e14bb325SJeff Bonwick zio_gang_node_t *gn = zio->io_private; 1783e14bb325SJeff Bonwick blkptr_t *bp = zio->io_bp; 1784fa9e4066Sahrens 1785f5383399SBill Moore ASSERT(gio == zio_unique_parent(zio)); 1786b24ab676SJeff Bonwick ASSERT(zio->io_child_count == 0); 1787fa9e4066Sahrens 1788e14bb325SJeff Bonwick if (zio->io_error) 1789e14bb325SJeff Bonwick return; 1790fa9e4066Sahrens 1791e14bb325SJeff Bonwick if (BP_SHOULD_BYTESWAP(bp)) 1792e14bb325SJeff Bonwick byteswap_uint64_array(zio->io_data, zio->io_size); 1793fa9e4066Sahrens 1794e14bb325SJeff Bonwick ASSERT(zio->io_data == gn->gn_gbh); 1795e14bb325SJeff Bonwick ASSERT(zio->io_size == SPA_GANGBLOCKSIZE); 17966e1f5caaSNeil Perrin ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC); 1797e05725b1Sbonwick 1798e14bb325SJeff Bonwick for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) { 1799e14bb325SJeff Bonwick blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g]; 1800e14bb325SJeff Bonwick if (!BP_IS_GANG(gbp)) 1801e14bb325SJeff Bonwick continue; 1802f5383399SBill Moore zio_gang_tree_assemble(gio, gbp, &gn->gn_child[g]); 1803e14bb325SJeff Bonwick } 1804fa9e4066Sahrens } 1805fa9e4066Sahrens 1806e14bb325SJeff Bonwick static void 1807e14bb325SJeff Bonwick zio_gang_tree_issue(zio_t *pio, zio_gang_node_t *gn, blkptr_t *bp, void *data) 1808fa9e4066Sahrens { 1809f5383399SBill Moore zio_t *gio = pio->io_gang_leader; 1810e14bb325SJeff Bonwick zio_t *zio; 1811fa9e4066Sahrens 1812e14bb325SJeff Bonwick ASSERT(BP_IS_GANG(bp) == !!gn); 1813f5383399SBill Moore ASSERT(BP_GET_CHECKSUM(bp) == BP_GET_CHECKSUM(gio->io_bp)); 1814f5383399SBill Moore ASSERT(BP_GET_LSIZE(bp) == BP_GET_PSIZE(bp) || gn == gio->io_gang_tree); 1815fa9e4066Sahrens 1816e14bb325SJeff Bonwick /* 1817e14bb325SJeff Bonwick * If you're a gang header, your data is in gn->gn_gbh. 1818e14bb325SJeff Bonwick * If you're a gang member, your data is in 'data' and gn == NULL. 1819e14bb325SJeff Bonwick */ 1820f5383399SBill Moore zio = zio_gang_issue_func[gio->io_type](pio, bp, gn, data); 1821fa9e4066Sahrens 1822e14bb325SJeff Bonwick if (gn != NULL) { 18236e1f5caaSNeil Perrin ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC); 1824fa9e4066Sahrens 1825e14bb325SJeff Bonwick for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) { 1826e14bb325SJeff Bonwick blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g]; 1827e14bb325SJeff Bonwick if (BP_IS_HOLE(gbp)) 1828e14bb325SJeff Bonwick continue; 1829e14bb325SJeff Bonwick zio_gang_tree_issue(zio, gn->gn_child[g], gbp, data); 1830e14bb325SJeff Bonwick data = (char *)data + BP_GET_PSIZE(gbp); 1831e14bb325SJeff Bonwick } 1832fa9e4066Sahrens } 1833fa9e4066Sahrens 1834f5383399SBill Moore if (gn == gio->io_gang_tree) 1835f5383399SBill Moore ASSERT3P((char *)gio->io_data + gio->io_size, ==, data); 1836e05725b1Sbonwick 1837e14bb325SJeff Bonwick if (zio != pio) 1838e14bb325SJeff Bonwick zio_nowait(zio); 1839fa9e4066Sahrens } 1840fa9e4066Sahrens 1841e05725b1Sbonwick static int 1842e14bb325SJeff Bonwick zio_gang_assemble(zio_t *zio) 1843fa9e4066Sahrens { 1844e14bb325SJeff Bonwick blkptr_t *bp = zio->io_bp; 1845fa9e4066Sahrens 1846f5383399SBill Moore ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == NULL); 1847f5383399SBill Moore ASSERT(zio->io_child_type > ZIO_CHILD_GANG); 1848f5383399SBill Moore 1849f5383399SBill Moore zio->io_gang_leader = zio; 1850fa9e4066Sahrens 1851e14bb325SJeff Bonwick zio_gang_tree_assemble(zio, bp, &zio->io_gang_tree); 1852e05725b1Sbonwick 1853e05725b1Sbonwick return (ZIO_PIPELINE_CONTINUE); 1854fa9e4066Sahrens } 1855fa9e4066Sahrens 1856e05725b1Sbonwick static int 1857e14bb325SJeff Bonwick zio_gang_issue(zio_t *zio) 1858fa9e4066Sahrens { 1859e14bb325SJeff Bonwick blkptr_t *bp = zio->io_bp; 1860fa9e4066Sahrens 1861e14bb325SJeff Bonwick if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE)) 1862e14bb325SJeff Bonwick return (ZIO_PIPELINE_STOP); 1863fa9e4066Sahrens 1864f5383399SBill Moore ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == zio); 1865f5383399SBill Moore ASSERT(zio->io_child_type > ZIO_CHILD_GANG); 1866fa9e4066Sahrens 1867e14bb325SJeff Bonwick if (zio->io_child_error[ZIO_CHILD_GANG] == 0) 1868f5383399SBill Moore zio_gang_tree_issue(zio, zio->io_gang_tree, bp, zio->io_data); 1869e14bb325SJeff Bonwick else 1870f5383399SBill Moore zio_gang_tree_free(&zio->io_gang_tree); 1871fa9e4066Sahrens 1872e14bb325SJeff Bonwick zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; 1873e05725b1Sbonwick 1874e05725b1Sbonwick return (ZIO_PIPELINE_CONTINUE); 1875fa9e4066Sahrens } 1876fa9e4066Sahrens 1877fa9e4066Sahrens static void 1878e14bb325SJeff Bonwick zio_write_gang_member_ready(zio_t *zio) 1879fa9e4066Sahrens { 1880a3f829aeSBill Moore zio_t *pio = zio_unique_parent(zio); 1881f5383399SBill Moore zio_t *gio = zio->io_gang_leader; 188244cd46caSbillm dva_t *cdva = zio->io_bp->blk_dva; 188344cd46caSbillm dva_t *pdva = pio->io_bp->blk_dva; 1884fa9e4066Sahrens uint64_t asize; 1885fa9e4066Sahrens 1886e14bb325SJeff Bonwick if (BP_IS_HOLE(zio->io_bp)) 1887e14bb325SJeff Bonwick return; 1888e14bb325SJeff Bonwick 1889e14bb325SJeff Bonwick ASSERT(BP_IS_HOLE(&zio->io_bp_orig)); 1890e14bb325SJeff Bonwick 1891e14bb325SJeff Bonwick ASSERT(zio->io_child_type == ZIO_CHILD_GANG); 1892b24ab676SJeff Bonwick ASSERT3U(zio->io_prop.zp_copies, ==, gio->io_prop.zp_copies); 1893b24ab676SJeff Bonwick ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(zio->io_bp)); 1894b24ab676SJeff Bonwick ASSERT3U(pio->io_prop.zp_copies, <=, BP_GET_NDVAS(pio->io_bp)); 189544cd46caSbillm ASSERT3U(BP_GET_NDVAS(zio->io_bp), <=, BP_GET_NDVAS(pio->io_bp)); 1896fa9e4066Sahrens 1897fa9e4066Sahrens mutex_enter(&pio->io_lock); 1898e14bb325SJeff Bonwick for (int d = 0; d < BP_GET_NDVAS(zio->io_bp); d++) { 189944cd46caSbillm ASSERT(DVA_GET_GANG(&pdva[d])); 190044cd46caSbillm asize = DVA_GET_ASIZE(&pdva[d]); 190144cd46caSbillm asize += DVA_GET_ASIZE(&cdva[d]); 190244cd46caSbillm DVA_SET_ASIZE(&pdva[d], asize); 190344cd46caSbillm } 1904fa9e4066Sahrens mutex_exit(&pio->io_lock); 1905fa9e4066Sahrens } 1906fa9e4066Sahrens 19070a4e9518Sgw static int 1908e14bb325SJeff Bonwick zio_write_gang_block(zio_t *pio) 1909fa9e4066Sahrens { 1910e14bb325SJeff Bonwick spa_t *spa = pio->io_spa; 1911e14bb325SJeff Bonwick blkptr_t *bp = pio->io_bp; 1912f5383399SBill Moore zio_t *gio = pio->io_gang_leader; 1913e14bb325SJeff Bonwick zio_t *zio; 1914e14bb325SJeff Bonwick zio_gang_node_t *gn, **gnpp; 1915fa9e4066Sahrens zio_gbh_phys_t *gbh; 1916e14bb325SJeff Bonwick uint64_t txg = pio->io_txg; 1917e14bb325SJeff Bonwick uint64_t resid = pio->io_size; 1918e14bb325SJeff Bonwick uint64_t lsize; 1919b24ab676SJeff Bonwick int copies = gio->io_prop.zp_copies; 1920b24ab676SJeff Bonwick int gbh_copies = MIN(copies + 1, spa_max_replication(spa)); 1921e14bb325SJeff Bonwick zio_prop_t zp; 1922fa9e4066Sahrens int error; 1923fa9e4066Sahrens 1924b24ab676SJeff Bonwick error = metaslab_alloc(spa, spa_normal_class(spa), SPA_GANGBLOCKSIZE, 1925b24ab676SJeff Bonwick bp, gbh_copies, txg, pio == gio ? NULL : gio->io_bp, 1926e14bb325SJeff Bonwick METASLAB_HINTBP_FAVOR | METASLAB_GANG_HEADER); 1927e05725b1Sbonwick if (error) { 1928e14bb325SJeff Bonwick pio->io_error = error; 1929e05725b1Sbonwick return (ZIO_PIPELINE_CONTINUE); 1930e05725b1Sbonwick } 1931fa9e4066Sahrens 1932f5383399SBill Moore if (pio == gio) { 1933f5383399SBill Moore gnpp = &gio->io_gang_tree; 1934e14bb325SJeff Bonwick } else { 1935e14bb325SJeff Bonwick gnpp = pio->io_private; 1936e14bb325SJeff Bonwick ASSERT(pio->io_ready == zio_write_gang_member_ready); 1937fa9e4066Sahrens } 1938fa9e4066Sahrens 1939e14bb325SJeff Bonwick gn = zio_gang_node_alloc(gnpp); 1940e14bb325SJeff Bonwick gbh = gn->gn_gbh; 1941e14bb325SJeff Bonwick bzero(gbh, SPA_GANGBLOCKSIZE); 1942fa9e4066Sahrens 1943e14bb325SJeff Bonwick /* 1944e14bb325SJeff Bonwick * Create the gang header. 1945e14bb325SJeff Bonwick */ 1946e14bb325SJeff Bonwick zio = zio_rewrite(pio, spa, txg, bp, gbh, SPA_GANGBLOCKSIZE, NULL, NULL, 1947e14bb325SJeff Bonwick pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); 1948fa9e4066Sahrens 1949e14bb325SJeff Bonwick /* 1950e14bb325SJeff Bonwick * Create and nowait the gang children. 1951e14bb325SJeff Bonwick */ 1952e14bb325SJeff Bonwick for (int g = 0; resid != 0; resid -= lsize, g++) { 1953e14bb325SJeff Bonwick lsize = P2ROUNDUP(resid / (SPA_GBH_NBLKPTRS - g), 1954e14bb325SJeff Bonwick SPA_MINBLOCKSIZE); 1955e14bb325SJeff Bonwick ASSERT(lsize >= SPA_MINBLOCKSIZE && lsize <= resid); 1956e14bb325SJeff Bonwick 1957f5383399SBill Moore zp.zp_checksum = gio->io_prop.zp_checksum; 1958e14bb325SJeff Bonwick zp.zp_compress = ZIO_COMPRESS_OFF; 1959e14bb325SJeff Bonwick zp.zp_type = DMU_OT_NONE; 1960e14bb325SJeff Bonwick zp.zp_level = 0; 1961b24ab676SJeff Bonwick zp.zp_copies = gio->io_prop.zp_copies; 196280901aeaSGeorge Wilson zp.zp_dedup = B_FALSE; 196380901aeaSGeorge Wilson zp.zp_dedup_verify = B_FALSE; 196480901aeaSGeorge Wilson zp.zp_nopwrite = B_FALSE; 1965e14bb325SJeff Bonwick 1966e14bb325SJeff Bonwick zio_nowait(zio_write(zio, spa, txg, &gbh->zg_blkptr[g], 1967e14bb325SJeff Bonwick (char *)pio->io_data + (pio->io_size - resid), lsize, &zp, 196869962b56SMatthew Ahrens zio_write_gang_member_ready, NULL, NULL, &gn->gn_child[g], 1969e14bb325SJeff Bonwick pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), 1970e14bb325SJeff Bonwick &pio->io_bookmark)); 1971e14bb325SJeff Bonwick } 1972e05725b1Sbonwick 197344cd46caSbillm /* 1974e14bb325SJeff Bonwick * Set pio's pipeline to just wait for zio to finish. 197544cd46caSbillm */ 1976e14bb325SJeff Bonwick pio->io_pipeline = ZIO_INTERLOCK_PIPELINE; 1977e14bb325SJeff Bonwick 1978e14bb325SJeff Bonwick zio_nowait(zio); 1979e14bb325SJeff Bonwick 1980e14bb325SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 1981fa9e4066Sahrens } 1982fa9e4066Sahrens 198380901aeaSGeorge Wilson /* 198480901aeaSGeorge Wilson * The zio_nop_write stage in the pipeline determines if allocating 198580901aeaSGeorge Wilson * a new bp is necessary. By leveraging a cryptographically secure checksum, 198680901aeaSGeorge Wilson * such as SHA256, we can compare the checksums of the new data and the old 198780901aeaSGeorge Wilson * to determine if allocating a new block is required. The nopwrite 198880901aeaSGeorge Wilson * feature can handle writes in either syncing or open context (i.e. zil 198980901aeaSGeorge Wilson * writes) and as a result is mutually exclusive with dedup. 199080901aeaSGeorge Wilson */ 199180901aeaSGeorge Wilson static int 199280901aeaSGeorge Wilson zio_nop_write(zio_t *zio) 199380901aeaSGeorge Wilson { 199480901aeaSGeorge Wilson blkptr_t *bp = zio->io_bp; 199580901aeaSGeorge Wilson blkptr_t *bp_orig = &zio->io_bp_orig; 199680901aeaSGeorge Wilson zio_prop_t *zp = &zio->io_prop; 199780901aeaSGeorge Wilson 199880901aeaSGeorge Wilson ASSERT(BP_GET_LEVEL(bp) == 0); 199980901aeaSGeorge Wilson ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE)); 200080901aeaSGeorge Wilson ASSERT(zp->zp_nopwrite); 200180901aeaSGeorge Wilson ASSERT(!zp->zp_dedup); 200280901aeaSGeorge Wilson ASSERT(zio->io_bp_override == NULL); 200380901aeaSGeorge Wilson ASSERT(IO_IS_ALLOCATING(zio)); 200480901aeaSGeorge Wilson 200580901aeaSGeorge Wilson /* 200680901aeaSGeorge Wilson * Check to see if the original bp and the new bp have matching 200780901aeaSGeorge Wilson * characteristics (i.e. same checksum, compression algorithms, etc). 200880901aeaSGeorge Wilson * If they don't then just continue with the pipeline which will 200980901aeaSGeorge Wilson * allocate a new bp. 201080901aeaSGeorge Wilson */ 201180901aeaSGeorge Wilson if (BP_IS_HOLE(bp_orig) || 201280901aeaSGeorge Wilson !zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_dedup || 201380901aeaSGeorge Wilson BP_GET_CHECKSUM(bp) != BP_GET_CHECKSUM(bp_orig) || 201480901aeaSGeorge Wilson BP_GET_COMPRESS(bp) != BP_GET_COMPRESS(bp_orig) || 201580901aeaSGeorge Wilson BP_GET_DEDUP(bp) != BP_GET_DEDUP(bp_orig) || 201680901aeaSGeorge Wilson zp->zp_copies != BP_GET_NDVAS(bp_orig)) 201780901aeaSGeorge Wilson return (ZIO_PIPELINE_CONTINUE); 201880901aeaSGeorge Wilson 201980901aeaSGeorge Wilson /* 202080901aeaSGeorge Wilson * If the checksums match then reset the pipeline so that we 202180901aeaSGeorge Wilson * avoid allocating a new bp and issuing any I/O. 202280901aeaSGeorge Wilson */ 202380901aeaSGeorge Wilson if (ZIO_CHECKSUM_EQUAL(bp->blk_cksum, bp_orig->blk_cksum)) { 202480901aeaSGeorge Wilson ASSERT(zio_checksum_table[zp->zp_checksum].ci_dedup); 202580901aeaSGeorge Wilson ASSERT3U(BP_GET_PSIZE(bp), ==, BP_GET_PSIZE(bp_orig)); 202680901aeaSGeorge Wilson ASSERT3U(BP_GET_LSIZE(bp), ==, BP_GET_LSIZE(bp_orig)); 202780901aeaSGeorge Wilson ASSERT(zp->zp_compress != ZIO_COMPRESS_OFF); 202880901aeaSGeorge Wilson ASSERT(bcmp(&bp->blk_prop, &bp_orig->blk_prop, 202980901aeaSGeorge Wilson sizeof (uint64_t)) == 0); 203080901aeaSGeorge Wilson 203180901aeaSGeorge Wilson *bp = *bp_orig; 203280901aeaSGeorge Wilson zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; 203380901aeaSGeorge Wilson zio->io_flags |= ZIO_FLAG_NOPWRITE; 203480901aeaSGeorge Wilson } 203580901aeaSGeorge Wilson 203680901aeaSGeorge Wilson return (ZIO_PIPELINE_CONTINUE); 203780901aeaSGeorge Wilson } 203880901aeaSGeorge Wilson 2039fa9e4066Sahrens /* 2040fa9e4066Sahrens * ========================================================================== 2041b24ab676SJeff Bonwick * Dedup 2042fa9e4066Sahrens * ========================================================================== 2043fa9e4066Sahrens */ 2044b24ab676SJeff Bonwick static void 2045b24ab676SJeff Bonwick zio_ddt_child_read_done(zio_t *zio) 2046b24ab676SJeff Bonwick { 2047b24ab676SJeff Bonwick blkptr_t *bp = zio->io_bp; 2048b24ab676SJeff Bonwick ddt_entry_t *dde = zio->io_private; 2049b24ab676SJeff Bonwick ddt_phys_t *ddp; 2050b24ab676SJeff Bonwick zio_t *pio = zio_unique_parent(zio); 2051b24ab676SJeff Bonwick 2052b24ab676SJeff Bonwick mutex_enter(&pio->io_lock); 2053b24ab676SJeff Bonwick ddp = ddt_phys_select(dde, bp); 2054b24ab676SJeff Bonwick if (zio->io_error == 0) 2055b24ab676SJeff Bonwick ddt_phys_clear(ddp); /* this ddp doesn't need repair */ 2056b24ab676SJeff Bonwick if (zio->io_error == 0 && dde->dde_repair_data == NULL) 2057b24ab676SJeff Bonwick dde->dde_repair_data = zio->io_data; 2058b24ab676SJeff Bonwick else 2059b24ab676SJeff Bonwick zio_buf_free(zio->io_data, zio->io_size); 2060b24ab676SJeff Bonwick mutex_exit(&pio->io_lock); 2061b24ab676SJeff Bonwick } 2062b24ab676SJeff Bonwick 2063b24ab676SJeff Bonwick static int 2064b24ab676SJeff Bonwick zio_ddt_read_start(zio_t *zio) 2065b24ab676SJeff Bonwick { 2066b24ab676SJeff Bonwick blkptr_t *bp = zio->io_bp; 2067b24ab676SJeff Bonwick 2068b24ab676SJeff Bonwick ASSERT(BP_GET_DEDUP(bp)); 2069b24ab676SJeff Bonwick ASSERT(BP_GET_PSIZE(bp) == zio->io_size); 2070b24ab676SJeff Bonwick ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 2071b24ab676SJeff Bonwick 2072b24ab676SJeff Bonwick if (zio->io_child_error[ZIO_CHILD_DDT]) { 2073b24ab676SJeff Bonwick ddt_t *ddt = ddt_select(zio->io_spa, bp); 2074b24ab676SJeff Bonwick ddt_entry_t *dde = ddt_repair_start(ddt, bp); 2075b24ab676SJeff Bonwick ddt_phys_t *ddp = dde->dde_phys; 2076b24ab676SJeff Bonwick ddt_phys_t *ddp_self = ddt_phys_select(dde, bp); 2077b24ab676SJeff Bonwick blkptr_t blk; 2078b24ab676SJeff Bonwick 2079b24ab676SJeff Bonwick ASSERT(zio->io_vsd == NULL); 2080b24ab676SJeff Bonwick zio->io_vsd = dde; 2081b24ab676SJeff Bonwick 2082b24ab676SJeff Bonwick if (ddp_self == NULL) 2083b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 2084b24ab676SJeff Bonwick 2085b24ab676SJeff Bonwick for (int p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { 2086b24ab676SJeff Bonwick if (ddp->ddp_phys_birth == 0 || ddp == ddp_self) 2087b24ab676SJeff Bonwick continue; 2088bbfd46c4SJeff Bonwick ddt_bp_create(ddt->ddt_checksum, &dde->dde_key, ddp, 2089bbfd46c4SJeff Bonwick &blk); 2090b24ab676SJeff Bonwick zio_nowait(zio_read(zio, zio->io_spa, &blk, 2091b24ab676SJeff Bonwick zio_buf_alloc(zio->io_size), zio->io_size, 2092b24ab676SJeff Bonwick zio_ddt_child_read_done, dde, zio->io_priority, 2093b24ab676SJeff Bonwick ZIO_DDT_CHILD_FLAGS(zio) | ZIO_FLAG_DONT_PROPAGATE, 2094b24ab676SJeff Bonwick &zio->io_bookmark)); 2095b24ab676SJeff Bonwick } 2096b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 2097b24ab676SJeff Bonwick } 2098b24ab676SJeff Bonwick 2099b24ab676SJeff Bonwick zio_nowait(zio_read(zio, zio->io_spa, bp, 2100b24ab676SJeff Bonwick zio->io_data, zio->io_size, NULL, NULL, zio->io_priority, 2101b24ab676SJeff Bonwick ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark)); 2102b24ab676SJeff Bonwick 2103b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 2104b24ab676SJeff Bonwick } 2105e14bb325SJeff Bonwick 2106b24ab676SJeff Bonwick static int 2107b24ab676SJeff Bonwick zio_ddt_read_done(zio_t *zio) 2108b24ab676SJeff Bonwick { 2109b24ab676SJeff Bonwick blkptr_t *bp = zio->io_bp; 2110b24ab676SJeff Bonwick 2111b24ab676SJeff Bonwick if (zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_DONE)) 2112b24ab676SJeff Bonwick return (ZIO_PIPELINE_STOP); 2113b24ab676SJeff Bonwick 2114b24ab676SJeff Bonwick ASSERT(BP_GET_DEDUP(bp)); 2115b24ab676SJeff Bonwick ASSERT(BP_GET_PSIZE(bp) == zio->io_size); 2116b24ab676SJeff Bonwick ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 2117b24ab676SJeff Bonwick 2118b24ab676SJeff Bonwick if (zio->io_child_error[ZIO_CHILD_DDT]) { 2119b24ab676SJeff Bonwick ddt_t *ddt = ddt_select(zio->io_spa, bp); 2120b24ab676SJeff Bonwick ddt_entry_t *dde = zio->io_vsd; 2121b24ab676SJeff Bonwick if (ddt == NULL) { 2122b16da2e2SGeorge Wilson ASSERT(spa_load_state(zio->io_spa) != SPA_LOAD_NONE); 2123b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 2124b24ab676SJeff Bonwick } 2125b24ab676SJeff Bonwick if (dde == NULL) { 2126b24ab676SJeff Bonwick zio->io_stage = ZIO_STAGE_DDT_READ_START >> 1; 212735a5a358SJonathan Adams zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE); 2128b24ab676SJeff Bonwick return (ZIO_PIPELINE_STOP); 2129b24ab676SJeff Bonwick } 2130b24ab676SJeff Bonwick if (dde->dde_repair_data != NULL) { 2131b24ab676SJeff Bonwick bcopy(dde->dde_repair_data, zio->io_data, zio->io_size); 2132b24ab676SJeff Bonwick zio->io_child_error[ZIO_CHILD_DDT] = 0; 2133b24ab676SJeff Bonwick } 2134b24ab676SJeff Bonwick ddt_repair_done(ddt, dde); 2135b24ab676SJeff Bonwick zio->io_vsd = NULL; 2136b24ab676SJeff Bonwick } 2137b24ab676SJeff Bonwick 2138b24ab676SJeff Bonwick ASSERT(zio->io_vsd == NULL); 2139b24ab676SJeff Bonwick 2140b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 2141b24ab676SJeff Bonwick } 2142b24ab676SJeff Bonwick 2143b24ab676SJeff Bonwick static boolean_t 2144b24ab676SJeff Bonwick zio_ddt_collision(zio_t *zio, ddt_t *ddt, ddt_entry_t *dde) 2145b24ab676SJeff Bonwick { 2146b24ab676SJeff Bonwick spa_t *spa = zio->io_spa; 2147b24ab676SJeff Bonwick 2148b24ab676SJeff Bonwick /* 2149b24ab676SJeff Bonwick * Note: we compare the original data, not the transformed data, 2150b24ab676SJeff Bonwick * because when zio->io_bp is an override bp, we will not have 2151b24ab676SJeff Bonwick * pushed the I/O transforms. That's an important optimization 2152b24ab676SJeff Bonwick * because otherwise we'd compress/encrypt all dmu_sync() data twice. 2153b24ab676SJeff Bonwick */ 2154b24ab676SJeff Bonwick for (int p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { 2155b24ab676SJeff Bonwick zio_t *lio = dde->dde_lead_zio[p]; 2156b24ab676SJeff Bonwick 2157b24ab676SJeff Bonwick if (lio != NULL) { 2158b24ab676SJeff Bonwick return (lio->io_orig_size != zio->io_orig_size || 2159b24ab676SJeff Bonwick bcmp(zio->io_orig_data, lio->io_orig_data, 2160b24ab676SJeff Bonwick zio->io_orig_size) != 0); 2161b24ab676SJeff Bonwick } 2162b24ab676SJeff Bonwick } 2163b24ab676SJeff Bonwick 2164b24ab676SJeff Bonwick for (int p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { 2165b24ab676SJeff Bonwick ddt_phys_t *ddp = &dde->dde_phys[p]; 2166b24ab676SJeff Bonwick 2167b24ab676SJeff Bonwick if (ddp->ddp_phys_birth != 0) { 2168b24ab676SJeff Bonwick arc_buf_t *abuf = NULL; 21697adb730bSGeorge Wilson arc_flags_t aflags = ARC_FLAG_WAIT; 2170b24ab676SJeff Bonwick blkptr_t blk = *zio->io_bp; 2171b24ab676SJeff Bonwick int error; 2172b24ab676SJeff Bonwick 2173b24ab676SJeff Bonwick ddt_bp_fill(ddp, &blk, ddp->ddp_phys_birth); 2174b24ab676SJeff Bonwick 2175b24ab676SJeff Bonwick ddt_exit(ddt); 2176b24ab676SJeff Bonwick 21771b912ec7SGeorge Wilson error = arc_read(NULL, spa, &blk, 2178b24ab676SJeff Bonwick arc_getbuf_func, &abuf, ZIO_PRIORITY_SYNC_READ, 2179b24ab676SJeff Bonwick ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, 2180b24ab676SJeff Bonwick &aflags, &zio->io_bookmark); 2181b24ab676SJeff Bonwick 2182b24ab676SJeff Bonwick if (error == 0) { 2183b24ab676SJeff Bonwick if (arc_buf_size(abuf) != zio->io_orig_size || 2184b24ab676SJeff Bonwick bcmp(abuf->b_data, zio->io_orig_data, 2185b24ab676SJeff Bonwick zio->io_orig_size) != 0) 2186be6fd75aSMatthew Ahrens error = SET_ERROR(EEXIST); 21873b2aab18SMatthew Ahrens VERIFY(arc_buf_remove_ref(abuf, &abuf)); 2188b24ab676SJeff Bonwick } 2189b24ab676SJeff Bonwick 2190b24ab676SJeff Bonwick ddt_enter(ddt); 2191b24ab676SJeff Bonwick return (error != 0); 2192b24ab676SJeff Bonwick } 2193b24ab676SJeff Bonwick } 2194b24ab676SJeff Bonwick 2195b24ab676SJeff Bonwick return (B_FALSE); 2196b24ab676SJeff Bonwick } 2197b24ab676SJeff Bonwick 2198b24ab676SJeff Bonwick static void 2199b24ab676SJeff Bonwick zio_ddt_child_write_ready(zio_t *zio) 2200b24ab676SJeff Bonwick { 2201b24ab676SJeff Bonwick int p = zio->io_prop.zp_copies; 2202b24ab676SJeff Bonwick ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp); 2203b24ab676SJeff Bonwick ddt_entry_t *dde = zio->io_private; 2204b24ab676SJeff Bonwick ddt_phys_t *ddp = &dde->dde_phys[p]; 2205b24ab676SJeff Bonwick zio_t *pio; 2206b24ab676SJeff Bonwick 2207b24ab676SJeff Bonwick if (zio->io_error) 2208b24ab676SJeff Bonwick return; 2209b24ab676SJeff Bonwick 2210b24ab676SJeff Bonwick ddt_enter(ddt); 2211b24ab676SJeff Bonwick 2212b24ab676SJeff Bonwick ASSERT(dde->dde_lead_zio[p] == zio); 2213b24ab676SJeff Bonwick 2214b24ab676SJeff Bonwick ddt_phys_fill(ddp, zio->io_bp); 2215b24ab676SJeff Bonwick 2216b24ab676SJeff Bonwick while ((pio = zio_walk_parents(zio)) != NULL) 2217b24ab676SJeff Bonwick ddt_bp_fill(ddp, pio->io_bp, zio->io_txg); 2218b24ab676SJeff Bonwick 2219b24ab676SJeff Bonwick ddt_exit(ddt); 2220b24ab676SJeff Bonwick } 2221b24ab676SJeff Bonwick 2222b24ab676SJeff Bonwick static void 2223b24ab676SJeff Bonwick zio_ddt_child_write_done(zio_t *zio) 2224b24ab676SJeff Bonwick { 2225b24ab676SJeff Bonwick int p = zio->io_prop.zp_copies; 2226b24ab676SJeff Bonwick ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp); 2227b24ab676SJeff Bonwick ddt_entry_t *dde = zio->io_private; 2228b24ab676SJeff Bonwick ddt_phys_t *ddp = &dde->dde_phys[p]; 2229b24ab676SJeff Bonwick 2230b24ab676SJeff Bonwick ddt_enter(ddt); 2231b24ab676SJeff Bonwick 2232b24ab676SJeff Bonwick ASSERT(ddp->ddp_refcnt == 0); 2233b24ab676SJeff Bonwick ASSERT(dde->dde_lead_zio[p] == zio); 2234b24ab676SJeff Bonwick dde->dde_lead_zio[p] = NULL; 2235b24ab676SJeff Bonwick 2236b24ab676SJeff Bonwick if (zio->io_error == 0) { 2237b24ab676SJeff Bonwick while (zio_walk_parents(zio) != NULL) 2238b24ab676SJeff Bonwick ddt_phys_addref(ddp); 2239b24ab676SJeff Bonwick } else { 2240b24ab676SJeff Bonwick ddt_phys_clear(ddp); 2241b24ab676SJeff Bonwick } 2242b24ab676SJeff Bonwick 2243b24ab676SJeff Bonwick ddt_exit(ddt); 2244b24ab676SJeff Bonwick } 2245b24ab676SJeff Bonwick 2246b24ab676SJeff Bonwick static void 2247b24ab676SJeff Bonwick zio_ddt_ditto_write_done(zio_t *zio) 2248b24ab676SJeff Bonwick { 2249b24ab676SJeff Bonwick int p = DDT_PHYS_DITTO; 2250b24ab676SJeff Bonwick zio_prop_t *zp = &zio->io_prop; 2251b24ab676SJeff Bonwick blkptr_t *bp = zio->io_bp; 2252b24ab676SJeff Bonwick ddt_t *ddt = ddt_select(zio->io_spa, bp); 2253b24ab676SJeff Bonwick ddt_entry_t *dde = zio->io_private; 2254b24ab676SJeff Bonwick ddt_phys_t *ddp = &dde->dde_phys[p]; 2255b24ab676SJeff Bonwick ddt_key_t *ddk = &dde->dde_key; 2256b24ab676SJeff Bonwick 2257b24ab676SJeff Bonwick ddt_enter(ddt); 2258b24ab676SJeff Bonwick 2259b24ab676SJeff Bonwick ASSERT(ddp->ddp_refcnt == 0); 2260b24ab676SJeff Bonwick ASSERT(dde->dde_lead_zio[p] == zio); 2261b24ab676SJeff Bonwick dde->dde_lead_zio[p] = NULL; 2262b24ab676SJeff Bonwick 2263b24ab676SJeff Bonwick if (zio->io_error == 0) { 2264b24ab676SJeff Bonwick ASSERT(ZIO_CHECKSUM_EQUAL(bp->blk_cksum, ddk->ddk_cksum)); 2265b24ab676SJeff Bonwick ASSERT(zp->zp_copies < SPA_DVAS_PER_BP); 2266b24ab676SJeff Bonwick ASSERT(zp->zp_copies == BP_GET_NDVAS(bp) - BP_IS_GANG(bp)); 2267b24ab676SJeff Bonwick if (ddp->ddp_phys_birth != 0) 2268b24ab676SJeff Bonwick ddt_phys_free(ddt, ddk, ddp, zio->io_txg); 2269b24ab676SJeff Bonwick ddt_phys_fill(ddp, bp); 2270b24ab676SJeff Bonwick } 2271b24ab676SJeff Bonwick 2272b24ab676SJeff Bonwick ddt_exit(ddt); 2273b24ab676SJeff Bonwick } 2274b24ab676SJeff Bonwick 2275b24ab676SJeff Bonwick static int 2276b24ab676SJeff Bonwick zio_ddt_write(zio_t *zio) 2277b24ab676SJeff Bonwick { 2278b24ab676SJeff Bonwick spa_t *spa = zio->io_spa; 2279b24ab676SJeff Bonwick blkptr_t *bp = zio->io_bp; 2280b24ab676SJeff Bonwick uint64_t txg = zio->io_txg; 2281b24ab676SJeff Bonwick zio_prop_t *zp = &zio->io_prop; 2282b24ab676SJeff Bonwick int p = zp->zp_copies; 2283b24ab676SJeff Bonwick int ditto_copies; 2284b24ab676SJeff Bonwick zio_t *cio = NULL; 2285b24ab676SJeff Bonwick zio_t *dio = NULL; 2286b24ab676SJeff Bonwick ddt_t *ddt = ddt_select(spa, bp); 2287b24ab676SJeff Bonwick ddt_entry_t *dde; 2288b24ab676SJeff Bonwick ddt_phys_t *ddp; 2289b24ab676SJeff Bonwick 2290b24ab676SJeff Bonwick ASSERT(BP_GET_DEDUP(bp)); 2291b24ab676SJeff Bonwick ASSERT(BP_GET_CHECKSUM(bp) == zp->zp_checksum); 2292b24ab676SJeff Bonwick ASSERT(BP_IS_HOLE(bp) || zio->io_bp_override); 2293b24ab676SJeff Bonwick 2294b24ab676SJeff Bonwick ddt_enter(ddt); 2295b24ab676SJeff Bonwick dde = ddt_lookup(ddt, bp, B_TRUE); 2296b24ab676SJeff Bonwick ddp = &dde->dde_phys[p]; 2297b24ab676SJeff Bonwick 2298b24ab676SJeff Bonwick if (zp->zp_dedup_verify && zio_ddt_collision(zio, ddt, dde)) { 2299b24ab676SJeff Bonwick /* 2300b24ab676SJeff Bonwick * If we're using a weak checksum, upgrade to a strong checksum 2301b24ab676SJeff Bonwick * and try again. If we're already using a strong checksum, 2302b24ab676SJeff Bonwick * we can't resolve it, so just convert to an ordinary write. 2303b24ab676SJeff Bonwick * (And automatically e-mail a paper to Nature?) 2304b24ab676SJeff Bonwick */ 2305b24ab676SJeff Bonwick if (!zio_checksum_table[zp->zp_checksum].ci_dedup) { 2306b24ab676SJeff Bonwick zp->zp_checksum = spa_dedup_checksum(spa); 2307b24ab676SJeff Bonwick zio_pop_transforms(zio); 2308b24ab676SJeff Bonwick zio->io_stage = ZIO_STAGE_OPEN; 2309b24ab676SJeff Bonwick BP_ZERO(bp); 2310b24ab676SJeff Bonwick } else { 231180901aeaSGeorge Wilson zp->zp_dedup = B_FALSE; 2312b24ab676SJeff Bonwick } 2313b24ab676SJeff Bonwick zio->io_pipeline = ZIO_WRITE_PIPELINE; 2314b24ab676SJeff Bonwick ddt_exit(ddt); 2315b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 2316b24ab676SJeff Bonwick } 2317b24ab676SJeff Bonwick 2318b24ab676SJeff Bonwick ditto_copies = ddt_ditto_copies_needed(ddt, dde, ddp); 2319b24ab676SJeff Bonwick ASSERT(ditto_copies < SPA_DVAS_PER_BP); 2320b24ab676SJeff Bonwick 2321b24ab676SJeff Bonwick if (ditto_copies > ddt_ditto_copies_present(dde) && 2322b24ab676SJeff Bonwick dde->dde_lead_zio[DDT_PHYS_DITTO] == NULL) { 2323b24ab676SJeff Bonwick zio_prop_t czp = *zp; 2324b24ab676SJeff Bonwick 2325b24ab676SJeff Bonwick czp.zp_copies = ditto_copies; 2326b24ab676SJeff Bonwick 2327b24ab676SJeff Bonwick /* 2328b24ab676SJeff Bonwick * If we arrived here with an override bp, we won't have run 2329b24ab676SJeff Bonwick * the transform stack, so we won't have the data we need to 2330b24ab676SJeff Bonwick * generate a child i/o. So, toss the override bp and restart. 2331b24ab676SJeff Bonwick * This is safe, because using the override bp is just an 2332b24ab676SJeff Bonwick * optimization; and it's rare, so the cost doesn't matter. 2333b24ab676SJeff Bonwick */ 2334b24ab676SJeff Bonwick if (zio->io_bp_override) { 2335b24ab676SJeff Bonwick zio_pop_transforms(zio); 2336b24ab676SJeff Bonwick zio->io_stage = ZIO_STAGE_OPEN; 2337b24ab676SJeff Bonwick zio->io_pipeline = ZIO_WRITE_PIPELINE; 2338b24ab676SJeff Bonwick zio->io_bp_override = NULL; 2339b24ab676SJeff Bonwick BP_ZERO(bp); 2340b24ab676SJeff Bonwick ddt_exit(ddt); 2341b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 2342b24ab676SJeff Bonwick } 2343b24ab676SJeff Bonwick 2344b24ab676SJeff Bonwick dio = zio_write(zio, spa, txg, bp, zio->io_orig_data, 234569962b56SMatthew Ahrens zio->io_orig_size, &czp, NULL, NULL, 2346b24ab676SJeff Bonwick zio_ddt_ditto_write_done, dde, zio->io_priority, 2347b24ab676SJeff Bonwick ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark); 2348b24ab676SJeff Bonwick 2349b24ab676SJeff Bonwick zio_push_transform(dio, zio->io_data, zio->io_size, 0, NULL); 2350b24ab676SJeff Bonwick dde->dde_lead_zio[DDT_PHYS_DITTO] = dio; 2351b24ab676SJeff Bonwick } 2352b24ab676SJeff Bonwick 2353b24ab676SJeff Bonwick if (ddp->ddp_phys_birth != 0 || dde->dde_lead_zio[p] != NULL) { 2354b24ab676SJeff Bonwick if (ddp->ddp_phys_birth != 0) 2355b24ab676SJeff Bonwick ddt_bp_fill(ddp, bp, txg); 2356b24ab676SJeff Bonwick if (dde->dde_lead_zio[p] != NULL) 2357b24ab676SJeff Bonwick zio_add_child(zio, dde->dde_lead_zio[p]); 2358b24ab676SJeff Bonwick else 2359b24ab676SJeff Bonwick ddt_phys_addref(ddp); 2360b24ab676SJeff Bonwick } else if (zio->io_bp_override) { 2361b24ab676SJeff Bonwick ASSERT(bp->blk_birth == txg); 2362b24ab676SJeff Bonwick ASSERT(BP_EQUAL(bp, zio->io_bp_override)); 2363b24ab676SJeff Bonwick ddt_phys_fill(ddp, bp); 2364b24ab676SJeff Bonwick ddt_phys_addref(ddp); 2365b24ab676SJeff Bonwick } else { 2366b24ab676SJeff Bonwick cio = zio_write(zio, spa, txg, bp, zio->io_orig_data, 236769962b56SMatthew Ahrens zio->io_orig_size, zp, zio_ddt_child_write_ready, NULL, 2368b24ab676SJeff Bonwick zio_ddt_child_write_done, dde, zio->io_priority, 2369b24ab676SJeff Bonwick ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark); 2370b24ab676SJeff Bonwick 2371b24ab676SJeff Bonwick zio_push_transform(cio, zio->io_data, zio->io_size, 0, NULL); 2372b24ab676SJeff Bonwick dde->dde_lead_zio[p] = cio; 2373b24ab676SJeff Bonwick } 2374b24ab676SJeff Bonwick 2375b24ab676SJeff Bonwick ddt_exit(ddt); 2376b24ab676SJeff Bonwick 2377b24ab676SJeff Bonwick if (cio) 2378b24ab676SJeff Bonwick zio_nowait(cio); 2379b24ab676SJeff Bonwick if (dio) 2380b24ab676SJeff Bonwick zio_nowait(dio); 2381b24ab676SJeff Bonwick 2382b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 2383b24ab676SJeff Bonwick } 2384b24ab676SJeff Bonwick 23853f9d6ad7SLin Ling ddt_entry_t *freedde; /* for debugging */ 23863f9d6ad7SLin Ling 2387b24ab676SJeff Bonwick static int 2388b24ab676SJeff Bonwick zio_ddt_free(zio_t *zio) 2389b24ab676SJeff Bonwick { 2390b24ab676SJeff Bonwick spa_t *spa = zio->io_spa; 2391b24ab676SJeff Bonwick blkptr_t *bp = zio->io_bp; 2392b24ab676SJeff Bonwick ddt_t *ddt = ddt_select(spa, bp); 2393b24ab676SJeff Bonwick ddt_entry_t *dde; 2394b24ab676SJeff Bonwick ddt_phys_t *ddp; 2395b24ab676SJeff Bonwick 2396b24ab676SJeff Bonwick ASSERT(BP_GET_DEDUP(bp)); 2397b24ab676SJeff Bonwick ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 2398b24ab676SJeff Bonwick 2399b24ab676SJeff Bonwick ddt_enter(ddt); 24003f9d6ad7SLin Ling freedde = dde = ddt_lookup(ddt, bp, B_TRUE); 2401b24ab676SJeff Bonwick ddp = ddt_phys_select(dde, bp); 2402b24ab676SJeff Bonwick ddt_phys_decref(ddp); 2403b24ab676SJeff Bonwick ddt_exit(ddt); 2404b24ab676SJeff Bonwick 2405b24ab676SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 2406b24ab676SJeff Bonwick } 2407b24ab676SJeff Bonwick 2408b24ab676SJeff Bonwick /* 2409b24ab676SJeff Bonwick * ========================================================================== 2410b24ab676SJeff Bonwick * Allocate and free blocks 2411b24ab676SJeff Bonwick * ========================================================================== 2412b24ab676SJeff Bonwick */ 2413e05725b1Sbonwick static int 2414fa9e4066Sahrens zio_dva_allocate(zio_t *zio) 2415fa9e4066Sahrens { 24168654d025Sperrin spa_t *spa = zio->io_spa; 2417b24ab676SJeff Bonwick metaslab_class_t *mc = spa_normal_class(spa); 2418fa9e4066Sahrens blkptr_t *bp = zio->io_bp; 2419fa9e4066Sahrens int error; 242009c9d376SGeorge Wilson int flags = 0; 2421fa9e4066Sahrens 2422f5383399SBill Moore if (zio->io_gang_leader == NULL) { 2423f5383399SBill Moore ASSERT(zio->io_child_type > ZIO_CHILD_GANG); 2424f5383399SBill Moore zio->io_gang_leader = zio; 2425f5383399SBill Moore } 2426f5383399SBill Moore 2427fa9e4066Sahrens ASSERT(BP_IS_HOLE(bp)); 2428fb09f5aaSMadhav Suresh ASSERT0(BP_GET_NDVAS(bp)); 2429b24ab676SJeff Bonwick ASSERT3U(zio->io_prop.zp_copies, >, 0); 2430b24ab676SJeff Bonwick ASSERT3U(zio->io_prop.zp_copies, <=, spa_max_replication(spa)); 2431fa9e4066Sahrens ASSERT3U(zio->io_size, ==, BP_GET_PSIZE(bp)); 2432fa9e4066Sahrens 243309c9d376SGeorge Wilson /* 243409c9d376SGeorge Wilson * The dump device does not support gang blocks so allocation on 243509c9d376SGeorge Wilson * behalf of the dump device (i.e. ZIO_FLAG_NODATA) must avoid 243609c9d376SGeorge Wilson * the "fast" gang feature. 243709c9d376SGeorge Wilson */ 243809c9d376SGeorge Wilson flags |= (zio->io_flags & ZIO_FLAG_NODATA) ? METASLAB_GANG_AVOID : 0; 243909c9d376SGeorge Wilson flags |= (zio->io_flags & ZIO_FLAG_GANG_CHILD) ? 244009c9d376SGeorge Wilson METASLAB_GANG_CHILD : 0; 2441e14bb325SJeff Bonwick error = metaslab_alloc(spa, mc, zio->io_size, bp, 244209c9d376SGeorge Wilson zio->io_prop.zp_copies, zio->io_txg, NULL, flags); 2443fa9e4066Sahrens 2444e14bb325SJeff Bonwick if (error) { 244509c9d376SGeorge Wilson spa_dbgmsg(spa, "%s: metaslab allocation failure: zio %p, " 244609c9d376SGeorge Wilson "size %llu, error %d", spa_name(spa), zio, zio->io_size, 244709c9d376SGeorge Wilson error); 2448e14bb325SJeff Bonwick if (error == ENOSPC && zio->io_size > SPA_MINBLOCKSIZE) 2449e14bb325SJeff Bonwick return (zio_write_gang_block(zio)); 2450fa9e4066Sahrens zio->io_error = error; 2451fa9e4066Sahrens } 2452e05725b1Sbonwick 2453e05725b1Sbonwick return (ZIO_PIPELINE_CONTINUE); 2454fa9e4066Sahrens } 2455fa9e4066Sahrens 2456e05725b1Sbonwick static int 2457fa9e4066Sahrens zio_dva_free(zio_t *zio) 2458fa9e4066Sahrens { 2459e14bb325SJeff Bonwick metaslab_free(zio->io_spa, zio->io_bp, zio->io_txg, B_FALSE); 2460fa9e4066Sahrens 2461e05725b1Sbonwick return (ZIO_PIPELINE_CONTINUE); 2462fa9e4066Sahrens } 2463fa9e4066Sahrens 2464e05725b1Sbonwick static int 2465fa9e4066Sahrens zio_dva_claim(zio_t *zio) 2466fa9e4066Sahrens { 2467e14bb325SJeff Bonwick int error; 2468e14bb325SJeff Bonwick 2469e14bb325SJeff Bonwick error = metaslab_claim(zio->io_spa, zio->io_bp, zio->io_txg); 2470e14bb325SJeff Bonwick if (error) 2471e14bb325SJeff Bonwick zio->io_error = error; 2472fa9e4066Sahrens 2473e05725b1Sbonwick return (ZIO_PIPELINE_CONTINUE); 2474fa9e4066Sahrens } 2475fa9e4066Sahrens 2476e14bb325SJeff Bonwick /* 2477e14bb325SJeff Bonwick * Undo an allocation. This is used by zio_done() when an I/O fails 2478e14bb325SJeff Bonwick * and we want to give back the block we just allocated. 2479e14bb325SJeff Bonwick * This handles both normal blocks and gang blocks. 2480e14bb325SJeff Bonwick */ 2481e14bb325SJeff Bonwick static void 2482e14bb325SJeff Bonwick zio_dva_unallocate(zio_t *zio, zio_gang_node_t *gn, blkptr_t *bp) 2483e14bb325SJeff Bonwick { 2484e14bb325SJeff Bonwick ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp)); 2485b24ab676SJeff Bonwick ASSERT(zio->io_bp_override == NULL); 2486e14bb325SJeff Bonwick 2487e14bb325SJeff Bonwick if (!BP_IS_HOLE(bp)) 2488b24ab676SJeff Bonwick metaslab_free(zio->io_spa, bp, bp->blk_birth, B_TRUE); 2489e14bb325SJeff Bonwick 2490e14bb325SJeff Bonwick if (gn != NULL) { 2491e14bb325SJeff Bonwick for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) { 2492e14bb325SJeff Bonwick zio_dva_unallocate(zio, gn->gn_child[g], 2493e14bb325SJeff Bonwick &gn->gn_gbh->zg_blkptr[g]); 2494e14bb325SJeff Bonwick } 2495e14bb325SJeff Bonwick } 2496e14bb325SJeff Bonwick } 2497e14bb325SJeff Bonwick 2498e14bb325SJeff Bonwick /* 2499e14bb325SJeff Bonwick * Try to allocate an intent log block. Return 0 on success, errno on failure. 2500e14bb325SJeff Bonwick */ 2501e14bb325SJeff Bonwick int 2502b24ab676SJeff Bonwick zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp, blkptr_t *old_bp, 2503b24ab676SJeff Bonwick uint64_t size, boolean_t use_slog) 2504e14bb325SJeff Bonwick { 2505e09fa4daSNeil Perrin int error = 1; 2506e14bb325SJeff Bonwick 2507b24ab676SJeff Bonwick ASSERT(txg > spa_syncing_txg(spa)); 2508b24ab676SJeff Bonwick 2509840345f6SGeorge Wilson /* 2510840345f6SGeorge Wilson * ZIL blocks are always contiguous (i.e. not gang blocks) so we 2511840345f6SGeorge Wilson * set the METASLAB_GANG_AVOID flag so that they don't "fast gang" 2512840345f6SGeorge Wilson * when allocating them. 2513840345f6SGeorge Wilson */ 2514840345f6SGeorge Wilson if (use_slog) { 2515b24ab676SJeff Bonwick error = metaslab_alloc(spa, spa_log_class(spa), size, 2516840345f6SGeorge Wilson new_bp, 1, txg, old_bp, 2517840345f6SGeorge Wilson METASLAB_HINTBP_AVOID | METASLAB_GANG_AVOID); 2518840345f6SGeorge Wilson } 2519e14bb325SJeff Bonwick 2520840345f6SGeorge Wilson if (error) { 2521b24ab676SJeff Bonwick error = metaslab_alloc(spa, spa_normal_class(spa), size, 2522840345f6SGeorge Wilson new_bp, 1, txg, old_bp, 252322e30981SGeorge Wilson METASLAB_HINTBP_AVOID); 2524840345f6SGeorge Wilson } 2525e14bb325SJeff Bonwick 2526e14bb325SJeff Bonwick if (error == 0) { 2527e14bb325SJeff Bonwick BP_SET_LSIZE(new_bp, size); 2528e14bb325SJeff Bonwick BP_SET_PSIZE(new_bp, size); 2529e14bb325SJeff Bonwick BP_SET_COMPRESS(new_bp, ZIO_COMPRESS_OFF); 25306e1f5caaSNeil Perrin BP_SET_CHECKSUM(new_bp, 25316e1f5caaSNeil Perrin spa_version(spa) >= SPA_VERSION_SLIM_ZIL 25326e1f5caaSNeil Perrin ? ZIO_CHECKSUM_ZILOG2 : ZIO_CHECKSUM_ZILOG); 2533e14bb325SJeff Bonwick BP_SET_TYPE(new_bp, DMU_OT_INTENT_LOG); 2534e14bb325SJeff Bonwick BP_SET_LEVEL(new_bp, 0); 2535b24ab676SJeff Bonwick BP_SET_DEDUP(new_bp, 0); 2536e14bb325SJeff Bonwick BP_SET_BYTEORDER(new_bp, ZFS_HOST_BYTEORDER); 2537e14bb325SJeff Bonwick } 2538e14bb325SJeff Bonwick 2539e14bb325SJeff Bonwick return (error); 2540e14bb325SJeff Bonwick } 2541e14bb325SJeff Bonwick 2542e14bb325SJeff Bonwick /* 2543b24ab676SJeff Bonwick * Free an intent log block. 2544e14bb325SJeff Bonwick */ 2545e14bb325SJeff Bonwick void 2546b24ab676SJeff Bonwick zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp) 2547e14bb325SJeff Bonwick { 2548b24ab676SJeff Bonwick ASSERT(BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG); 2549e14bb325SJeff Bonwick ASSERT(!BP_IS_GANG(bp)); 2550e14bb325SJeff Bonwick 2551b24ab676SJeff Bonwick zio_free(spa, txg, bp); 2552e14bb325SJeff Bonwick } 2553e14bb325SJeff Bonwick 2554fa9e4066Sahrens /* 2555fa9e4066Sahrens * ========================================================================== 2556fa9e4066Sahrens * Read and write to physical devices 2557fa9e4066Sahrens * ========================================================================== 2558fa9e4066Sahrens */ 2559738f37bcSGeorge Wilson 2560738f37bcSGeorge Wilson 2561738f37bcSGeorge Wilson /* 2562738f37bcSGeorge Wilson * Issue an I/O to the underlying vdev. Typically the issue pipeline 2563738f37bcSGeorge Wilson * stops after this stage and will resume upon I/O completion. 2564738f37bcSGeorge Wilson * However, there are instances where the vdev layer may need to 2565738f37bcSGeorge Wilson * continue the pipeline when an I/O was not issued. Since the I/O 2566738f37bcSGeorge Wilson * that was sent to the vdev layer might be different than the one 2567738f37bcSGeorge Wilson * currently active in the pipeline (see vdev_queue_io()), we explicitly 2568738f37bcSGeorge Wilson * force the underlying vdev layers to call either zio_execute() or 2569738f37bcSGeorge Wilson * zio_interrupt() to ensure that the pipeline continues with the correct I/O. 2570738f37bcSGeorge Wilson */ 2571e05725b1Sbonwick static int 257244cd46caSbillm zio_vdev_io_start(zio_t *zio) 2573fa9e4066Sahrens { 2574fa9e4066Sahrens vdev_t *vd = zio->io_vd; 257544cd46caSbillm uint64_t align; 25760a4e9518Sgw spa_t *spa = zio->io_spa; 25770a4e9518Sgw 2578e14bb325SJeff Bonwick ASSERT(zio->io_error == 0); 2579e14bb325SJeff Bonwick ASSERT(zio->io_child_error[ZIO_CHILD_VDEV] == 0); 2580fa9e4066Sahrens 2581e14bb325SJeff Bonwick if (vd == NULL) { 2582e14bb325SJeff Bonwick if (!(zio->io_flags & ZIO_FLAG_CONFIG_WRITER)) 2583e14bb325SJeff Bonwick spa_config_enter(spa, SCL_ZIO, zio, RW_READER); 2584fa9e4066Sahrens 2585e14bb325SJeff Bonwick /* 2586e14bb325SJeff Bonwick * The mirror_ops handle multiple DVAs in a single BP. 2587e14bb325SJeff Bonwick */ 2588738f37bcSGeorge Wilson vdev_mirror_ops.vdev_op_io_start(zio); 2589738f37bcSGeorge Wilson return (ZIO_PIPELINE_STOP); 2590fa9e4066Sahrens } 2591fa9e4066Sahrens 259244ecc532SGeorge Wilson /* 259344ecc532SGeorge Wilson * We keep track of time-sensitive I/Os so that the scan thread 259444ecc532SGeorge Wilson * can quickly react to certain workloads. In particular, we care 259544ecc532SGeorge Wilson * about non-scrubbing, top-level reads and writes with the following 259644ecc532SGeorge Wilson * characteristics: 2597738f37bcSGeorge Wilson * - synchronous writes of user data to non-slog devices 259844ecc532SGeorge Wilson * - any reads of user data 259944ecc532SGeorge Wilson * When these conditions are met, adjust the timestamp of spa_last_io 260044ecc532SGeorge Wilson * which allows the scan thread to adjust its workload accordingly. 260144ecc532SGeorge Wilson */ 260244ecc532SGeorge Wilson if (!(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && zio->io_bp != NULL && 260344ecc532SGeorge Wilson vd == vd->vdev_top && !vd->vdev_islog && 260444ecc532SGeorge Wilson zio->io_bookmark.zb_objset != DMU_META_OBJSET && 260544ecc532SGeorge Wilson zio->io_txg != spa_syncing_txg(spa)) { 260644ecc532SGeorge Wilson uint64_t old = spa->spa_last_io; 260744ecc532SGeorge Wilson uint64_t new = ddi_get_lbolt64(); 260844ecc532SGeorge Wilson if (old != new) 260944ecc532SGeorge Wilson (void) atomic_cas_64(&spa->spa_last_io, old, new); 261044ecc532SGeorge Wilson } 261144ecc532SGeorge Wilson 2612e14bb325SJeff Bonwick align = 1ULL << vd->vdev_top->vdev_ashift; 2613e14bb325SJeff Bonwick 26142a104a52SAlex Reece if (!(zio->io_flags & ZIO_FLAG_PHYSICAL) && 26152a104a52SAlex Reece P2PHASE(zio->io_size, align) != 0) { 26162a104a52SAlex Reece /* Transform logical writes to be a full physical block size. */ 2617ecc2d604Sbonwick uint64_t asize = P2ROUNDUP(zio->io_size, align); 2618ecc2d604Sbonwick char *abuf = zio_buf_alloc(asize); 2619e14bb325SJeff Bonwick ASSERT(vd == vd->vdev_top); 2620ecc2d604Sbonwick if (zio->io_type == ZIO_TYPE_WRITE) { 2621ecc2d604Sbonwick bcopy(zio->io_data, abuf, zio->io_size); 2622ecc2d604Sbonwick bzero(abuf + zio->io_size, asize - zio->io_size); 2623ecc2d604Sbonwick } 2624e14bb325SJeff Bonwick zio_push_transform(zio, abuf, asize, asize, zio_subblock); 2625ecc2d604Sbonwick } 2626ecc2d604Sbonwick 26272a104a52SAlex Reece /* 26282a104a52SAlex Reece * If this is not a physical io, make sure that it is properly aligned 26292a104a52SAlex Reece * before proceeding. 26302a104a52SAlex Reece */ 26312a104a52SAlex Reece if (!(zio->io_flags & ZIO_FLAG_PHYSICAL)) { 26322a104a52SAlex Reece ASSERT0(P2PHASE(zio->io_offset, align)); 26332a104a52SAlex Reece ASSERT0(P2PHASE(zio->io_size, align)); 26342a104a52SAlex Reece } else { 26352a104a52SAlex Reece /* 26362a104a52SAlex Reece * For physical writes, we allow 512b aligned writes and assume 26372a104a52SAlex Reece * the device will perform a read-modify-write as necessary. 26382a104a52SAlex Reece */ 26392a104a52SAlex Reece ASSERT0(P2PHASE(zio->io_offset, SPA_MINBLOCKSIZE)); 26402a104a52SAlex Reece ASSERT0(P2PHASE(zio->io_size, SPA_MINBLOCKSIZE)); 26412a104a52SAlex Reece } 26422a104a52SAlex Reece 2643f9af39baSGeorge Wilson VERIFY(zio->io_type != ZIO_TYPE_WRITE || spa_writeable(spa)); 26448ad4d6ddSJeff Bonwick 26458ad4d6ddSJeff Bonwick /* 26468ad4d6ddSJeff Bonwick * If this is a repair I/O, and there's no self-healing involved -- 26478ad4d6ddSJeff Bonwick * that is, we're just resilvering what we expect to resilver -- 26488ad4d6ddSJeff Bonwick * then don't do the I/O unless zio's txg is actually in vd's DTL. 26498ad4d6ddSJeff Bonwick * This prevents spurious resilvering with nested replication. 26508ad4d6ddSJeff Bonwick * For example, given a mirror of mirrors, (A+B)+(C+D), if only 26518ad4d6ddSJeff Bonwick * A is out of date, we'll read from C+D, then use the data to 26528ad4d6ddSJeff Bonwick * resilver A+B -- but we don't actually want to resilver B, just A. 26538ad4d6ddSJeff Bonwick * The top-level mirror has no way to know this, so instead we just 26548ad4d6ddSJeff Bonwick * discard unnecessary repairs as we work our way down the vdev tree. 26558ad4d6ddSJeff Bonwick * The same logic applies to any form of nested replication: 26568ad4d6ddSJeff Bonwick * ditto + mirror, RAID-Z + replacing, etc. This covers them all. 26578ad4d6ddSJeff Bonwick */ 26588ad4d6ddSJeff Bonwick if ((zio->io_flags & ZIO_FLAG_IO_REPAIR) && 26598ad4d6ddSJeff Bonwick !(zio->io_flags & ZIO_FLAG_SELF_HEAL) && 26608ad4d6ddSJeff Bonwick zio->io_txg != 0 && /* not a delegated i/o */ 26618ad4d6ddSJeff Bonwick !vdev_dtl_contains(vd, DTL_PARTIAL, zio->io_txg, 1)) { 26628ad4d6ddSJeff Bonwick ASSERT(zio->io_type == ZIO_TYPE_WRITE); 26638ad4d6ddSJeff Bonwick zio_vdev_io_bypass(zio); 26648ad4d6ddSJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 26658ad4d6ddSJeff Bonwick } 2666fa9e4066Sahrens 2667e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && 2668e14bb325SJeff Bonwick (zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE)) { 2669e14bb325SJeff Bonwick 267043466aaeSMax Grossman if (zio->io_type == ZIO_TYPE_READ && vdev_cache_read(zio)) 2671a3f829aeSBill Moore return (ZIO_PIPELINE_CONTINUE); 2672e14bb325SJeff Bonwick 2673e14bb325SJeff Bonwick if ((zio = vdev_queue_io(zio)) == NULL) 2674e14bb325SJeff Bonwick return (ZIO_PIPELINE_STOP); 2675e14bb325SJeff Bonwick 2676e14bb325SJeff Bonwick if (!vdev_accessible(vd, zio)) { 2677be6fd75aSMatthew Ahrens zio->io_error = SET_ERROR(ENXIO); 2678e14bb325SJeff Bonwick zio_interrupt(zio); 2679e14bb325SJeff Bonwick return (ZIO_PIPELINE_STOP); 2680e14bb325SJeff Bonwick } 2681e14bb325SJeff Bonwick } 2682e14bb325SJeff Bonwick 2683738f37bcSGeorge Wilson vd->vdev_ops->vdev_op_io_start(zio); 2684738f37bcSGeorge Wilson return (ZIO_PIPELINE_STOP); 2685fa9e4066Sahrens } 2686fa9e4066Sahrens 2687e05725b1Sbonwick static int 2688fa9e4066Sahrens zio_vdev_io_done(zio_t *zio) 2689fa9e4066Sahrens { 2690e14bb325SJeff Bonwick vdev_t *vd = zio->io_vd; 2691e14bb325SJeff Bonwick vdev_ops_t *ops = vd ? vd->vdev_ops : &vdev_mirror_ops; 2692e14bb325SJeff Bonwick boolean_t unexpected_error = B_FALSE; 2693e05725b1Sbonwick 2694e14bb325SJeff Bonwick if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE)) 2695e14bb325SJeff Bonwick return (ZIO_PIPELINE_STOP); 2696fa9e4066Sahrens 2697e14bb325SJeff Bonwick ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE); 2698e14bb325SJeff Bonwick 2699e14bb325SJeff Bonwick if (vd != NULL && vd->vdev_ops->vdev_op_leaf) { 2700e14bb325SJeff Bonwick 2701e14bb325SJeff Bonwick vdev_queue_io_done(zio); 2702fa9e4066Sahrens 2703e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_WRITE) 2704e14bb325SJeff Bonwick vdev_cache_write(zio); 2705e14bb325SJeff Bonwick 2706e14bb325SJeff Bonwick if (zio_injection_enabled && zio->io_error == 0) 27078956713aSEric Schrock zio->io_error = zio_handle_device_injection(vd, 27088956713aSEric Schrock zio, EIO); 2709e14bb325SJeff Bonwick 2710e14bb325SJeff Bonwick if (zio_injection_enabled && zio->io_error == 0) 2711e14bb325SJeff Bonwick zio->io_error = zio_handle_label_injection(zio, EIO); 2712e14bb325SJeff Bonwick 2713e14bb325SJeff Bonwick if (zio->io_error) { 2714e14bb325SJeff Bonwick if (!vdev_accessible(vd, zio)) { 2715be6fd75aSMatthew Ahrens zio->io_error = SET_ERROR(ENXIO); 2716e14bb325SJeff Bonwick } else { 2717e14bb325SJeff Bonwick unexpected_error = B_TRUE; 2718e14bb325SJeff Bonwick } 2719e14bb325SJeff Bonwick } 272051ece835Seschrock } 2721fa9e4066Sahrens 2722e14bb325SJeff Bonwick ops->vdev_op_io_done(zio); 2723e14bb325SJeff Bonwick 2724e14bb325SJeff Bonwick if (unexpected_error) 2725a3f829aeSBill Moore VERIFY(vdev_probe(vd, zio) == NULL); 2726e14bb325SJeff Bonwick 2727e14bb325SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 2728fa9e4066Sahrens } 2729fa9e4066Sahrens 273022fe2c88SJonathan Adams /* 273122fe2c88SJonathan Adams * For non-raidz ZIOs, we can just copy aside the bad data read from the 273222fe2c88SJonathan Adams * disk, and use that to finish the checksum ereport later. 273322fe2c88SJonathan Adams */ 273422fe2c88SJonathan Adams static void 273522fe2c88SJonathan Adams zio_vsd_default_cksum_finish(zio_cksum_report_t *zcr, 273622fe2c88SJonathan Adams const void *good_buf) 273722fe2c88SJonathan Adams { 273822fe2c88SJonathan Adams /* no processing needed */ 273922fe2c88SJonathan Adams zfs_ereport_finish_checksum(zcr, good_buf, zcr->zcr_cbdata, B_FALSE); 274022fe2c88SJonathan Adams } 274122fe2c88SJonathan Adams 274222fe2c88SJonathan Adams /*ARGSUSED*/ 274322fe2c88SJonathan Adams void 274422fe2c88SJonathan Adams zio_vsd_default_cksum_report(zio_t *zio, zio_cksum_report_t *zcr, void *ignored) 274522fe2c88SJonathan Adams { 274622fe2c88SJonathan Adams void *buf = zio_buf_alloc(zio->io_size); 274722fe2c88SJonathan Adams 274822fe2c88SJonathan Adams bcopy(zio->io_data, buf, zio->io_size); 274922fe2c88SJonathan Adams 275022fe2c88SJonathan Adams zcr->zcr_cbinfo = zio->io_size; 275122fe2c88SJonathan Adams zcr->zcr_cbdata = buf; 275222fe2c88SJonathan Adams zcr->zcr_finish = zio_vsd_default_cksum_finish; 275322fe2c88SJonathan Adams zcr->zcr_free = zio_buf_free; 275422fe2c88SJonathan Adams } 275522fe2c88SJonathan Adams 2756e05725b1Sbonwick static int 2757fa9e4066Sahrens zio_vdev_io_assess(zio_t *zio) 2758fa9e4066Sahrens { 2759fa9e4066Sahrens vdev_t *vd = zio->io_vd; 2760e14bb325SJeff Bonwick 2761e14bb325SJeff Bonwick if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE)) 2762e14bb325SJeff Bonwick return (ZIO_PIPELINE_STOP); 2763e14bb325SJeff Bonwick 2764e14bb325SJeff Bonwick if (vd == NULL && !(zio->io_flags & ZIO_FLAG_CONFIG_WRITER)) 2765e14bb325SJeff Bonwick spa_config_exit(zio->io_spa, SCL_ZIO, zio); 2766e14bb325SJeff Bonwick 2767e14bb325SJeff Bonwick if (zio->io_vsd != NULL) { 276822fe2c88SJonathan Adams zio->io_vsd_ops->vsd_free(zio); 2769e14bb325SJeff Bonwick zio->io_vsd = NULL; 2770ecc2d604Sbonwick } 2771ecc2d604Sbonwick 2772e14bb325SJeff Bonwick if (zio_injection_enabled && zio->io_error == 0) 2773ea8dc4b6Seschrock zio->io_error = zio_handle_fault_injection(zio, EIO); 2774ea8dc4b6Seschrock 2775fa9e4066Sahrens /* 2776fa9e4066Sahrens * If the I/O failed, determine whether we should attempt to retry it. 277735a5a358SJonathan Adams * 277835a5a358SJonathan Adams * On retry, we cut in line in the issue queue, since we don't want 277935a5a358SJonathan Adams * compression/checksumming/etc. work to prevent our (cheap) IO reissue. 2780fa9e4066Sahrens */ 2781e14bb325SJeff Bonwick if (zio->io_error && vd == NULL && 2782e14bb325SJeff Bonwick !(zio->io_flags & (ZIO_FLAG_DONT_RETRY | ZIO_FLAG_IO_RETRY))) { 2783e14bb325SJeff Bonwick ASSERT(!(zio->io_flags & ZIO_FLAG_DONT_QUEUE)); /* not a leaf */ 2784e14bb325SJeff Bonwick ASSERT(!(zio->io_flags & ZIO_FLAG_IO_BYPASS)); /* not a leaf */ 2785fa9e4066Sahrens zio->io_error = 0; 2786e14bb325SJeff Bonwick zio->io_flags |= ZIO_FLAG_IO_RETRY | 2787e14bb325SJeff Bonwick ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE; 2788b24ab676SJeff Bonwick zio->io_stage = ZIO_STAGE_VDEV_IO_START >> 1; 278935a5a358SJonathan Adams zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, 279035a5a358SJonathan Adams zio_requeue_io_start_cut_in_line); 2791e14bb325SJeff Bonwick return (ZIO_PIPELINE_STOP); 2792ea8dc4b6Seschrock } 2793fa9e4066Sahrens 2794e14bb325SJeff Bonwick /* 2795e14bb325SJeff Bonwick * If we got an error on a leaf device, convert it to ENXIO 2796e14bb325SJeff Bonwick * if the device is not accessible at all. 2797e14bb325SJeff Bonwick */ 2798e14bb325SJeff Bonwick if (zio->io_error && vd != NULL && vd->vdev_ops->vdev_op_leaf && 2799e14bb325SJeff Bonwick !vdev_accessible(vd, zio)) 2800be6fd75aSMatthew Ahrens zio->io_error = SET_ERROR(ENXIO); 2801e14bb325SJeff Bonwick 2802e14bb325SJeff Bonwick /* 2803e14bb325SJeff Bonwick * If we can't write to an interior vdev (mirror or RAID-Z), 2804e14bb325SJeff Bonwick * set vdev_cant_write so that we stop trying to allocate from it. 2805e14bb325SJeff Bonwick */ 2806e14bb325SJeff Bonwick if (zio->io_error == ENXIO && zio->io_type == ZIO_TYPE_WRITE && 28073b2aab18SMatthew Ahrens vd != NULL && !vd->vdev_ops->vdev_op_leaf) { 2808e14bb325SJeff Bonwick vd->vdev_cant_write = B_TRUE; 28093b2aab18SMatthew Ahrens } 2810e14bb325SJeff Bonwick 2811e14bb325SJeff Bonwick if (zio->io_error) 2812e14bb325SJeff Bonwick zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; 2813e14bb325SJeff Bonwick 281469962b56SMatthew Ahrens if (vd != NULL && vd->vdev_ops->vdev_op_leaf && 281569962b56SMatthew Ahrens zio->io_physdone != NULL) { 281669962b56SMatthew Ahrens ASSERT(!(zio->io_flags & ZIO_FLAG_DELEGATED)); 281769962b56SMatthew Ahrens ASSERT(zio->io_child_type == ZIO_CHILD_VDEV); 281869962b56SMatthew Ahrens zio->io_physdone(zio->io_logical); 281969962b56SMatthew Ahrens } 282069962b56SMatthew Ahrens 2821e05725b1Sbonwick return (ZIO_PIPELINE_CONTINUE); 2822fa9e4066Sahrens } 2823fa9e4066Sahrens 2824fa9e4066Sahrens void 2825fa9e4066Sahrens zio_vdev_io_reissue(zio_t *zio) 2826fa9e4066Sahrens { 2827fa9e4066Sahrens ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START); 2828fa9e4066Sahrens ASSERT(zio->io_error == 0); 2829fa9e4066Sahrens 2830b24ab676SJeff Bonwick zio->io_stage >>= 1; 2831fa9e4066Sahrens } 2832fa9e4066Sahrens 2833fa9e4066Sahrens void 2834fa9e4066Sahrens zio_vdev_io_redone(zio_t *zio) 2835fa9e4066Sahrens { 2836fa9e4066Sahrens ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_DONE); 2837fa9e4066Sahrens 2838b24ab676SJeff Bonwick zio->io_stage >>= 1; 2839fa9e4066Sahrens } 2840fa9e4066Sahrens 2841fa9e4066Sahrens void 2842fa9e4066Sahrens zio_vdev_io_bypass(zio_t *zio) 2843fa9e4066Sahrens { 2844fa9e4066Sahrens ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START); 2845fa9e4066Sahrens ASSERT(zio->io_error == 0); 2846fa9e4066Sahrens 2847fa9e4066Sahrens zio->io_flags |= ZIO_FLAG_IO_BYPASS; 2848b24ab676SJeff Bonwick zio->io_stage = ZIO_STAGE_VDEV_IO_ASSESS >> 1; 2849fa9e4066Sahrens } 2850fa9e4066Sahrens 2851fa9e4066Sahrens /* 2852fa9e4066Sahrens * ========================================================================== 2853fa9e4066Sahrens * Generate and verify checksums 2854fa9e4066Sahrens * ========================================================================== 2855fa9e4066Sahrens */ 2856e05725b1Sbonwick static int 2857fa9e4066Sahrens zio_checksum_generate(zio_t *zio) 2858fa9e4066Sahrens { 2859fa9e4066Sahrens blkptr_t *bp = zio->io_bp; 2860e14bb325SJeff Bonwick enum zio_checksum checksum; 2861fa9e4066Sahrens 2862e14bb325SJeff Bonwick if (bp == NULL) { 2863e14bb325SJeff Bonwick /* 2864e14bb325SJeff Bonwick * This is zio_write_phys(). 2865e14bb325SJeff Bonwick * We're either generating a label checksum, or none at all. 2866e14bb325SJeff Bonwick */ 2867e14bb325SJeff Bonwick checksum = zio->io_prop.zp_checksum; 2868e14bb325SJeff Bonwick 2869e14bb325SJeff Bonwick if (checksum == ZIO_CHECKSUM_OFF) 2870e14bb325SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 2871fa9e4066Sahrens 2872e14bb325SJeff Bonwick ASSERT(checksum == ZIO_CHECKSUM_LABEL); 2873e14bb325SJeff Bonwick } else { 2874e14bb325SJeff Bonwick if (BP_IS_GANG(bp) && zio->io_child_type == ZIO_CHILD_GANG) { 2875e14bb325SJeff Bonwick ASSERT(!IO_IS_ALLOCATING(zio)); 2876e14bb325SJeff Bonwick checksum = ZIO_CHECKSUM_GANG_HEADER; 2877e14bb325SJeff Bonwick } else { 2878e14bb325SJeff Bonwick checksum = BP_GET_CHECKSUM(bp); 2879e14bb325SJeff Bonwick } 2880e14bb325SJeff Bonwick } 2881fa9e4066Sahrens 2882e14bb325SJeff Bonwick zio_checksum_compute(zio, checksum, zio->io_data, zio->io_size); 2883fa9e4066Sahrens 2884e05725b1Sbonwick return (ZIO_PIPELINE_CONTINUE); 2885fa9e4066Sahrens } 2886fa9e4066Sahrens 2887e05725b1Sbonwick static int 2888e14bb325SJeff Bonwick zio_checksum_verify(zio_t *zio) 2889fa9e4066Sahrens { 289022fe2c88SJonathan Adams zio_bad_cksum_t info; 2891e14bb325SJeff Bonwick blkptr_t *bp = zio->io_bp; 2892e14bb325SJeff Bonwick int error; 2893fa9e4066Sahrens 2894b24ab676SJeff Bonwick ASSERT(zio->io_vd != NULL); 2895b24ab676SJeff Bonwick 2896e14bb325SJeff Bonwick if (bp == NULL) { 2897e14bb325SJeff Bonwick /* 2898e14bb325SJeff Bonwick * This is zio_read_phys(). 2899e14bb325SJeff Bonwick * We're either verifying a label checksum, or nothing at all. 2900e14bb325SJeff Bonwick */ 2901e14bb325SJeff Bonwick if (zio->io_prop.zp_checksum == ZIO_CHECKSUM_OFF) 2902e14bb325SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 2903fa9e4066Sahrens 2904e14bb325SJeff Bonwick ASSERT(zio->io_prop.zp_checksum == ZIO_CHECKSUM_LABEL); 2905e14bb325SJeff Bonwick } 2906fa9e4066Sahrens 290722fe2c88SJonathan Adams if ((error = zio_checksum_error(zio, &info)) != 0) { 2908e14bb325SJeff Bonwick zio->io_error = error; 2909373dc1cfSMatthew Ahrens if (error == ECKSUM && 2910373dc1cfSMatthew Ahrens !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) { 291122fe2c88SJonathan Adams zfs_ereport_start_checksum(zio->io_spa, 291222fe2c88SJonathan Adams zio->io_vd, zio, zio->io_offset, 291322fe2c88SJonathan Adams zio->io_size, NULL, &info); 2914e14bb325SJeff Bonwick } 2915fa9e4066Sahrens } 2916fa9e4066Sahrens 2917e05725b1Sbonwick return (ZIO_PIPELINE_CONTINUE); 2918fa9e4066Sahrens } 2919fa9e4066Sahrens 2920fa9e4066Sahrens /* 2921fa9e4066Sahrens * Called by RAID-Z to ensure we don't compute the checksum twice. 2922fa9e4066Sahrens */ 2923fa9e4066Sahrens void 2924fa9e4066Sahrens zio_checksum_verified(zio_t *zio) 2925fa9e4066Sahrens { 2926b24ab676SJeff Bonwick zio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY; 2927fa9e4066Sahrens } 2928fa9e4066Sahrens 2929fa9e4066Sahrens /* 2930e14bb325SJeff Bonwick * ========================================================================== 2931e14bb325SJeff Bonwick * Error rank. Error are ranked in the order 0, ENXIO, ECKSUM, EIO, other. 29325d7b4d43SMatthew Ahrens * An error of 0 indicates success. ENXIO indicates whole-device failure, 2933e14bb325SJeff Bonwick * which may be transient (e.g. unplugged) or permament. ECKSUM and EIO 2934e14bb325SJeff Bonwick * indicate errors that are specific to one I/O, and most likely permanent. 2935e14bb325SJeff Bonwick * Any other error is presumed to be worse because we weren't expecting it. 2936e14bb325SJeff Bonwick * ========================================================================== 2937fa9e4066Sahrens */ 2938e14bb325SJeff Bonwick int 2939e14bb325SJeff Bonwick zio_worst_error(int e1, int e2) 2940fa9e4066Sahrens { 2941e14bb325SJeff Bonwick static int zio_error_rank[] = { 0, ENXIO, ECKSUM, EIO }; 2942e14bb325SJeff Bonwick int r1, r2; 2943e14bb325SJeff Bonwick 2944e14bb325SJeff Bonwick for (r1 = 0; r1 < sizeof (zio_error_rank) / sizeof (int); r1++) 2945e14bb325SJeff Bonwick if (e1 == zio_error_rank[r1]) 2946e14bb325SJeff Bonwick break; 2947e14bb325SJeff Bonwick 2948e14bb325SJeff Bonwick for (r2 = 0; r2 < sizeof (zio_error_rank) / sizeof (int); r2++) 2949e14bb325SJeff Bonwick if (e2 == zio_error_rank[r2]) 2950e14bb325SJeff Bonwick break; 295144cd46caSbillm 2952e14bb325SJeff Bonwick return (r1 > r2 ? e1 : e2); 2953fa9e4066Sahrens } 2954fa9e4066Sahrens 2955fa9e4066Sahrens /* 2956fa9e4066Sahrens * ========================================================================== 2957e14bb325SJeff Bonwick * I/O completion 2958fa9e4066Sahrens * ========================================================================== 2959fa9e4066Sahrens */ 2960e14bb325SJeff Bonwick static int 2961e14bb325SJeff Bonwick zio_ready(zio_t *zio) 2962fa9e4066Sahrens { 2963e14bb325SJeff Bonwick blkptr_t *bp = zio->io_bp; 2964a3f829aeSBill Moore zio_t *pio, *pio_next; 2965fa9e4066Sahrens 2966b24ab676SJeff Bonwick if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY) || 2967b24ab676SJeff Bonwick zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_READY)) 2968f5383399SBill Moore return (ZIO_PIPELINE_STOP); 2969fa9e4066Sahrens 2970f5383399SBill Moore if (zio->io_ready) { 2971e14bb325SJeff Bonwick ASSERT(IO_IS_ALLOCATING(zio)); 297280901aeaSGeorge Wilson ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp) || 297380901aeaSGeorge Wilson (zio->io_flags & ZIO_FLAG_NOPWRITE)); 2974e14bb325SJeff Bonwick ASSERT(zio->io_children[ZIO_CHILD_GANG][ZIO_WAIT_READY] == 0); 2975fa9e4066Sahrens 2976e14bb325SJeff Bonwick zio->io_ready(zio); 2977e14bb325SJeff Bonwick } 2978fa9e4066Sahrens 2979e14bb325SJeff Bonwick if (bp != NULL && bp != &zio->io_bp_copy) 2980e14bb325SJeff Bonwick zio->io_bp_copy = *bp; 2981fa9e4066Sahrens 2982e14bb325SJeff Bonwick if (zio->io_error) 2983e14bb325SJeff Bonwick zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; 2984fa9e4066Sahrens 2985a3f829aeSBill Moore mutex_enter(&zio->io_lock); 2986a3f829aeSBill Moore zio->io_state[ZIO_WAIT_READY] = 1; 2987a3f829aeSBill Moore pio = zio_walk_parents(zio); 2988a3f829aeSBill Moore mutex_exit(&zio->io_lock); 2989a3f829aeSBill Moore 2990a3f829aeSBill Moore /* 2991a3f829aeSBill Moore * As we notify zio's parents, new parents could be added. 2992a3f829aeSBill Moore * New parents go to the head of zio's io_parent_list, however, 2993a3f829aeSBill Moore * so we will (correctly) not notify them. The remainder of zio's 2994a3f829aeSBill Moore * io_parent_list, from 'pio_next' onward, cannot change because 2995a3f829aeSBill Moore * all parents must wait for us to be done before they can be done. 2996a3f829aeSBill Moore */ 2997a3f829aeSBill Moore for (; pio != NULL; pio = pio_next) { 2998a3f829aeSBill Moore pio_next = zio_walk_parents(zio); 2999e14bb325SJeff Bonwick zio_notify_parent(pio, zio, ZIO_WAIT_READY); 3000a3f829aeSBill Moore } 3001fa9e4066Sahrens 3002b24ab676SJeff Bonwick if (zio->io_flags & ZIO_FLAG_NODATA) { 3003b24ab676SJeff Bonwick if (BP_IS_GANG(bp)) { 3004b24ab676SJeff Bonwick zio->io_flags &= ~ZIO_FLAG_NODATA; 3005b24ab676SJeff Bonwick } else { 3006b24ab676SJeff Bonwick ASSERT((uintptr_t)zio->io_data < SPA_MAXBLOCKSIZE); 3007b24ab676SJeff Bonwick zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES; 3008b24ab676SJeff Bonwick } 3009b24ab676SJeff Bonwick } 3010b24ab676SJeff Bonwick 3011a33cae98STim Haley if (zio_injection_enabled && 3012a33cae98STim Haley zio->io_spa->spa_syncing_txg == zio->io_txg) 3013a33cae98STim Haley zio_handle_ignored_writes(zio); 3014a33cae98STim Haley 3015e14bb325SJeff Bonwick return (ZIO_PIPELINE_CONTINUE); 3016fa9e4066Sahrens } 3017fa9e4066Sahrens 3018e14bb325SJeff Bonwick static int 3019e14bb325SJeff Bonwick zio_done(zio_t *zio) 3020d63d470bSgw { 3021e14bb325SJeff Bonwick spa_t *spa = zio->io_spa; 3022e14bb325SJeff Bonwick zio_t *lio = zio->io_logical; 3023e14bb325SJeff Bonwick blkptr_t *bp = zio->io_bp; 3024e14bb325SJeff Bonwick vdev_t *vd = zio->io_vd; 3025e14bb325SJeff Bonwick uint64_t psize = zio->io_size; 3026a3f829aeSBill Moore zio_t *pio, *pio_next; 3027d63d470bSgw 3028e14bb325SJeff Bonwick /* 3029f5383399SBill Moore * If our children haven't all completed, 3030e14bb325SJeff Bonwick * wait for them and then repeat this pipeline stage. 3031e14bb325SJeff Bonwick */ 3032e14bb325SJeff Bonwick if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE) || 3033e14bb325SJeff Bonwick zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE) || 3034b24ab676SJeff Bonwick zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_DONE) || 3035e14bb325SJeff Bonwick zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_DONE)) 3036e14bb325SJeff Bonwick return (ZIO_PIPELINE_STOP); 3037d63d470bSgw 3038e14bb325SJeff Bonwick for (int c = 0; c < ZIO_CHILD_TYPES; c++) 3039e14bb325SJeff Bonwick for (int w = 0; w < ZIO_WAIT_TYPES; w++) 3040e14bb325SJeff Bonwick ASSERT(zio->io_children[c][w] == 0); 3041e14bb325SJeff Bonwick 30425d7b4d43SMatthew Ahrens if (bp != NULL && !BP_IS_EMBEDDED(bp)) { 3043e14bb325SJeff Bonwick ASSERT(bp->blk_pad[0] == 0); 3044e14bb325SJeff Bonwick ASSERT(bp->blk_pad[1] == 0); 3045e14bb325SJeff Bonwick ASSERT(bcmp(bp, &zio->io_bp_copy, sizeof (blkptr_t)) == 0 || 3046a3f829aeSBill Moore (bp == zio_unique_parent(zio)->io_bp)); 3047e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_WRITE && !BP_IS_HOLE(bp) && 3048b24ab676SJeff Bonwick zio->io_bp_override == NULL && 3049e14bb325SJeff Bonwick !(zio->io_flags & ZIO_FLAG_IO_REPAIR)) { 3050e14bb325SJeff Bonwick ASSERT(!BP_SHOULD_BYTESWAP(bp)); 3051b24ab676SJeff Bonwick ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(bp)); 3052e14bb325SJeff Bonwick ASSERT(BP_COUNT_GANG(bp) == 0 || 3053e14bb325SJeff Bonwick (BP_COUNT_GANG(bp) == BP_GET_NDVAS(bp))); 3054e14bb325SJeff Bonwick } 305580901aeaSGeorge Wilson if (zio->io_flags & ZIO_FLAG_NOPWRITE) 305680901aeaSGeorge Wilson VERIFY(BP_EQUAL(bp, &zio->io_bp_orig)); 3057e14bb325SJeff Bonwick } 3058fa9e4066Sahrens 3059e14bb325SJeff Bonwick /* 3060b24ab676SJeff Bonwick * If there were child vdev/gang/ddt errors, they apply to us now. 3061e14bb325SJeff Bonwick */ 3062e14bb325SJeff Bonwick zio_inherit_child_errors(zio, ZIO_CHILD_VDEV); 3063e14bb325SJeff Bonwick zio_inherit_child_errors(zio, ZIO_CHILD_GANG); 3064b24ab676SJeff Bonwick zio_inherit_child_errors(zio, ZIO_CHILD_DDT); 3065b24ab676SJeff Bonwick 3066b24ab676SJeff Bonwick /* 3067b24ab676SJeff Bonwick * If the I/O on the transformed data was successful, generate any 3068b24ab676SJeff Bonwick * checksum reports now while we still have the transformed data. 3069b24ab676SJeff Bonwick */ 3070b24ab676SJeff Bonwick if (zio->io_error == 0) { 3071b24ab676SJeff Bonwick while (zio->io_cksum_report != NULL) { 3072b24ab676SJeff Bonwick zio_cksum_report_t *zcr = zio->io_cksum_report; 3073b24ab676SJeff Bonwick uint64_t align = zcr->zcr_align; 3074b24ab676SJeff Bonwick uint64_t asize = P2ROUNDUP(psize, align); 3075b24ab676SJeff Bonwick char *abuf = zio->io_data; 3076b24ab676SJeff Bonwick 3077b24ab676SJeff Bonwick if (asize != psize) { 3078b24ab676SJeff Bonwick abuf = zio_buf_alloc(asize); 3079b24ab676SJeff Bonwick bcopy(zio->io_data, abuf, psize); 3080b24ab676SJeff Bonwick bzero(abuf + psize, asize - psize); 3081b24ab676SJeff Bonwick } 3082b24ab676SJeff Bonwick 3083b24ab676SJeff Bonwick zio->io_cksum_report = zcr->zcr_next; 3084b24ab676SJeff Bonwick zcr->zcr_next = NULL; 3085b24ab676SJeff Bonwick zcr->zcr_finish(zcr, abuf); 3086b24ab676SJeff Bonwick zfs_ereport_free_checksum(zcr); 3087b24ab676SJeff Bonwick 3088b24ab676SJeff Bonwick if (asize != psize) 3089b24ab676SJeff Bonwick zio_buf_free(abuf, asize); 3090b24ab676SJeff Bonwick } 3091b24ab676SJeff Bonwick } 3092e14bb325SJeff Bonwick 3093e14bb325SJeff Bonwick zio_pop_transforms(zio); /* note: may set zio->io_error */ 3094e14bb325SJeff Bonwick 3095e14bb325SJeff Bonwick vdev_stat_update(zio, psize); 3096e14bb325SJeff Bonwick 3097e14bb325SJeff Bonwick if (zio->io_error) { 3098e14bb325SJeff Bonwick /* 3099e14bb325SJeff Bonwick * If this I/O is attached to a particular vdev, 3100e14bb325SJeff Bonwick * generate an error message describing the I/O failure 3101e14bb325SJeff Bonwick * at the block level. We ignore these errors if the 3102e14bb325SJeff Bonwick * device is currently unavailable. 3103e14bb325SJeff Bonwick */ 3104e14bb325SJeff Bonwick if (zio->io_error != ECKSUM && vd != NULL && !vdev_is_dead(vd)) 3105e14bb325SJeff Bonwick zfs_ereport_post(FM_EREPORT_ZFS_IO, spa, vd, zio, 0, 0); 3106e14bb325SJeff Bonwick 31078f18d1faSGeorge Wilson if ((zio->io_error == EIO || !(zio->io_flags & 31088f18d1faSGeorge Wilson (ZIO_FLAG_SPECULATIVE | ZIO_FLAG_DONT_PROPAGATE))) && 31098f18d1faSGeorge Wilson zio == lio) { 3110e14bb325SJeff Bonwick /* 3111e14bb325SJeff Bonwick * For logical I/O requests, tell the SPA to log the 3112e14bb325SJeff Bonwick * error and generate a logical data ereport. 3113e14bb325SJeff Bonwick */ 3114e14bb325SJeff Bonwick spa_log_error(spa, zio); 3115e14bb325SJeff Bonwick zfs_ereport_post(FM_EREPORT_ZFS_DATA, spa, NULL, zio, 3116e14bb325SJeff Bonwick 0, 0); 3117e14bb325SJeff Bonwick } 3118e14bb325SJeff Bonwick } 3119fa9e4066Sahrens 3120e14bb325SJeff Bonwick if (zio->io_error && zio == lio) { 3121e14bb325SJeff Bonwick /* 3122e14bb325SJeff Bonwick * Determine whether zio should be reexecuted. This will 3123e14bb325SJeff Bonwick * propagate all the way to the root via zio_notify_parent(). 3124e14bb325SJeff Bonwick */ 3125e14bb325SJeff Bonwick ASSERT(vd == NULL && bp != NULL); 3126b24ab676SJeff Bonwick ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 3127e14bb325SJeff Bonwick 3128b24ab676SJeff Bonwick if (IO_IS_ALLOCATING(zio) && 3129b24ab676SJeff Bonwick !(zio->io_flags & ZIO_FLAG_CANFAIL)) { 3130e14bb325SJeff Bonwick if (zio->io_error != ENOSPC) 3131e14bb325SJeff Bonwick zio->io_reexecute |= ZIO_REEXECUTE_NOW; 3132e14bb325SJeff Bonwick else 3133e14bb325SJeff Bonwick zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; 3134b24ab676SJeff Bonwick } 3135e14bb325SJeff Bonwick 3136e14bb325SJeff Bonwick if ((zio->io_type == ZIO_TYPE_READ || 3137e14bb325SJeff Bonwick zio->io_type == ZIO_TYPE_FREE) && 313844ecc532SGeorge Wilson !(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && 3139e14bb325SJeff Bonwick zio->io_error == ENXIO && 3140b16da2e2SGeorge Wilson spa_load_state(spa) == SPA_LOAD_NONE && 3141e14bb325SJeff Bonwick spa_get_failmode(spa) != ZIO_FAILURE_MODE_CONTINUE) 3142e14bb325SJeff Bonwick zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; 3143e14bb325SJeff Bonwick 3144e14bb325SJeff Bonwick if (!(zio->io_flags & ZIO_FLAG_CANFAIL) && !zio->io_reexecute) 3145e14bb325SJeff Bonwick zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; 314622fe2c88SJonathan Adams 314722fe2c88SJonathan Adams /* 314822fe2c88SJonathan Adams * Here is a possibly good place to attempt to do 314922fe2c88SJonathan Adams * either combinatorial reconstruction or error correction 315022fe2c88SJonathan Adams * based on checksums. It also might be a good place 315122fe2c88SJonathan Adams * to send out preliminary ereports before we suspend 315222fe2c88SJonathan Adams * processing. 315322fe2c88SJonathan Adams */ 3154d63d470bSgw } 3155d63d470bSgw 315667bd71c6Sperrin /* 3157e14bb325SJeff Bonwick * If there were logical child errors, they apply to us now. 3158e14bb325SJeff Bonwick * We defer this until now to avoid conflating logical child 3159e14bb325SJeff Bonwick * errors with errors that happened to the zio itself when 3160e14bb325SJeff Bonwick * updating vdev stats and reporting FMA events above. 316167bd71c6Sperrin */ 3162e14bb325SJeff Bonwick zio_inherit_child_errors(zio, ZIO_CHILD_LOGICAL); 31638654d025Sperrin 3164b24ab676SJeff Bonwick if ((zio->io_error || zio->io_reexecute) && 3165b24ab676SJeff Bonwick IO_IS_ALLOCATING(zio) && zio->io_gang_leader == zio && 316680901aeaSGeorge Wilson !(zio->io_flags & (ZIO_FLAG_IO_REWRITE | ZIO_FLAG_NOPWRITE))) 3167f5383399SBill Moore zio_dva_unallocate(zio, zio->io_gang_tree, bp); 3168f5383399SBill Moore 3169f5383399SBill Moore zio_gang_tree_free(&zio->io_gang_tree); 3170f5383399SBill Moore 317133a372edSGeorge Wilson /* 317233a372edSGeorge Wilson * Godfather I/Os should never suspend. 317333a372edSGeorge Wilson */ 317433a372edSGeorge Wilson if ((zio->io_flags & ZIO_FLAG_GODFATHER) && 317533a372edSGeorge Wilson (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) 317633a372edSGeorge Wilson zio->io_reexecute = 0; 317733a372edSGeorge Wilson 317833a372edSGeorge Wilson if (zio->io_reexecute) { 3179e14bb325SJeff Bonwick /* 3180e14bb325SJeff Bonwick * This is a logical I/O that wants to reexecute. 3181e14bb325SJeff Bonwick * 3182e14bb325SJeff Bonwick * Reexecute is top-down. When an i/o fails, if it's not 3183e14bb325SJeff Bonwick * the root, it simply notifies its parent and sticks around. 3184e14bb325SJeff Bonwick * The parent, seeing that it still has children in zio_done(), 3185e14bb325SJeff Bonwick * does the same. This percolates all the way up to the root. 3186e14bb325SJeff Bonwick * The root i/o will reexecute or suspend the entire tree. 3187e14bb325SJeff Bonwick * 3188e14bb325SJeff Bonwick * This approach ensures that zio_reexecute() honors 3189e14bb325SJeff Bonwick * all the original i/o dependency relationships, e.g. 3190e14bb325SJeff Bonwick * parents not executing until children are ready. 3191e14bb325SJeff Bonwick */ 3192e14bb325SJeff Bonwick ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); 3193fa9e4066Sahrens 3194f5383399SBill Moore zio->io_gang_leader = NULL; 3195e14bb325SJeff Bonwick 3196a3f829aeSBill Moore mutex_enter(&zio->io_lock); 3197a3f829aeSBill Moore zio->io_state[ZIO_WAIT_DONE] = 1; 3198a3f829aeSBill Moore mutex_exit(&zio->io_lock); 3199a3f829aeSBill Moore 320054d692b7SGeorge Wilson /* 320154d692b7SGeorge Wilson * "The Godfather" I/O monitors its children but is 320254d692b7SGeorge Wilson * not a true parent to them. It will track them through 320354d692b7SGeorge Wilson * the pipeline but severs its ties whenever they get into 320454d692b7SGeorge Wilson * trouble (e.g. suspended). This allows "The Godfather" 320554d692b7SGeorge Wilson * I/O to return status without blocking. 320654d692b7SGeorge Wilson */ 320754d692b7SGeorge Wilson for (pio = zio_walk_parents(zio); pio != NULL; pio = pio_next) { 320854d692b7SGeorge Wilson zio_link_t *zl = zio->io_walk_link; 320954d692b7SGeorge Wilson pio_next = zio_walk_parents(zio); 321054d692b7SGeorge Wilson 321154d692b7SGeorge Wilson if ((pio->io_flags & ZIO_FLAG_GODFATHER) && 321254d692b7SGeorge Wilson (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) { 321354d692b7SGeorge Wilson zio_remove_child(pio, zio, zl); 321454d692b7SGeorge Wilson zio_notify_parent(pio, zio, ZIO_WAIT_DONE); 321554d692b7SGeorge Wilson } 321654d692b7SGeorge Wilson } 321754d692b7SGeorge Wilson 3218a3f829aeSBill Moore if ((pio = zio_unique_parent(zio)) != NULL) { 3219e14bb325SJeff Bonwick /* 3220e14bb325SJeff Bonwick * We're not a root i/o, so there's nothing to do 3221e14bb325SJeff Bonwick * but notify our parent. Don't propagate errors 3222e14bb325SJeff Bonwick * upward since we haven't permanently failed yet. 3223e14bb325SJeff Bonwick */ 322433a372edSGeorge Wilson ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER)); 3225e14bb325SJeff Bonwick zio->io_flags |= ZIO_FLAG_DONT_PROPAGATE; 3226e14bb325SJeff Bonwick zio_notify_parent(pio, zio, ZIO_WAIT_DONE); 3227e14bb325SJeff Bonwick } else if (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND) { 3228e14bb325SJeff Bonwick /* 3229e14bb325SJeff Bonwick * We'd fail again if we reexecuted now, so suspend 3230e14bb325SJeff Bonwick * until conditions improve (e.g. device comes online). 3231e14bb325SJeff Bonwick */ 3232e14bb325SJeff Bonwick zio_suspend(spa, zio); 3233e14bb325SJeff Bonwick } else { 3234e14bb325SJeff Bonwick /* 3235e14bb325SJeff Bonwick * Reexecution is potentially a huge amount of work. 3236e14bb325SJeff Bonwick * Hand it off to the otherwise-unused claim taskq. 3237e14bb325SJeff Bonwick */ 32385aeb9474SGarrett D'Amore ASSERT(zio->io_tqent.tqent_next == NULL); 3239ec94d322SAdam Leventhal spa_taskq_dispatch_ent(spa, ZIO_TYPE_CLAIM, 3240ec94d322SAdam Leventhal ZIO_TASKQ_ISSUE, (task_func_t *)zio_reexecute, zio, 3241ec94d322SAdam Leventhal 0, &zio->io_tqent); 3242e14bb325SJeff Bonwick } 3243e14bb325SJeff Bonwick return (ZIO_PIPELINE_STOP); 3244fa9e4066Sahrens } 3245fa9e4066Sahrens 3246b24ab676SJeff Bonwick ASSERT(zio->io_child_count == 0); 324733a372edSGeorge Wilson ASSERT(zio->io_reexecute == 0); 3248e14bb325SJeff Bonwick ASSERT(zio->io_error == 0 || (zio->io_flags & ZIO_FLAG_CANFAIL)); 3249fa9e4066Sahrens 3250b24ab676SJeff Bonwick /* 3251b24ab676SJeff Bonwick * Report any checksum errors, since the I/O is complete. 3252b24ab676SJeff Bonwick */ 325322fe2c88SJonathan Adams while (zio->io_cksum_report != NULL) { 3254b24ab676SJeff Bonwick zio_cksum_report_t *zcr = zio->io_cksum_report; 3255b24ab676SJeff Bonwick zio->io_cksum_report = zcr->zcr_next; 3256b24ab676SJeff Bonwick zcr->zcr_next = NULL; 3257b24ab676SJeff Bonwick zcr->zcr_finish(zcr, NULL); 3258b24ab676SJeff Bonwick zfs_ereport_free_checksum(zcr); 325922fe2c88SJonathan Adams } 326022fe2c88SJonathan Adams 3261a3f829aeSBill Moore /* 3262a3f829aeSBill Moore * It is the responsibility of the done callback to ensure that this 3263a3f829aeSBill Moore * particular zio is no longer discoverable for adoption, and as 3264a3f829aeSBill Moore * such, cannot acquire any new parents. 3265a3f829aeSBill Moore */ 3266e14bb325SJeff Bonwick if (zio->io_done) 3267e14bb325SJeff Bonwick zio->io_done(zio); 3268fa9e4066Sahrens 3269a3f829aeSBill Moore mutex_enter(&zio->io_lock); 3270a3f829aeSBill Moore zio->io_state[ZIO_WAIT_DONE] = 1; 3271a3f829aeSBill Moore mutex_exit(&zio->io_lock); 3272fa9e4066Sahrens 3273a3f829aeSBill Moore for (pio = zio_walk_parents(zio); pio != NULL; pio = pio_next) { 3274a3f829aeSBill Moore zio_link_t *zl = zio->io_walk_link; 3275a3f829aeSBill Moore pio_next = zio_walk_parents(zio); 3276a3f829aeSBill Moore zio_remove_child(pio, zio, zl); 3277e14bb325SJeff Bonwick zio_notify_parent(pio, zio, ZIO_WAIT_DONE); 3278e14bb325SJeff Bonwick } 3279fa9e4066Sahrens 3280e14bb325SJeff Bonwick if (zio->io_waiter != NULL) { 3281e14bb325SJeff Bonwick mutex_enter(&zio->io_lock); 3282e14bb325SJeff Bonwick zio->io_executor = NULL; 3283e14bb325SJeff Bonwick cv_broadcast(&zio->io_cv); 3284e14bb325SJeff Bonwick mutex_exit(&zio->io_lock); 3285e14bb325SJeff Bonwick } else { 3286e14bb325SJeff Bonwick zio_destroy(zio); 3287e14bb325SJeff Bonwick } 3288fa9e4066Sahrens 3289e14bb325SJeff Bonwick return (ZIO_PIPELINE_STOP); 3290fa9e4066Sahrens } 329146341222Sperrin 329246341222Sperrin /* 3293e14bb325SJeff Bonwick * ========================================================================== 3294e14bb325SJeff Bonwick * I/O pipeline definition 3295e14bb325SJeff Bonwick * ========================================================================== 329646341222Sperrin */ 3297b24ab676SJeff Bonwick static zio_pipe_stage_t *zio_pipeline[] = { 3298e14bb325SJeff Bonwick NULL, 3299e14bb325SJeff Bonwick zio_read_bp_init, 3300b24ab676SJeff Bonwick zio_free_bp_init, 3301b24ab676SJeff Bonwick zio_issue_async, 3302e14bb325SJeff Bonwick zio_write_bp_init, 3303e14bb325SJeff Bonwick zio_checksum_generate, 330480901aeaSGeorge Wilson zio_nop_write, 3305b24ab676SJeff Bonwick zio_ddt_read_start, 3306b24ab676SJeff Bonwick zio_ddt_read_done, 3307b24ab676SJeff Bonwick zio_ddt_write, 3308b24ab676SJeff Bonwick zio_ddt_free, 3309e14bb325SJeff Bonwick zio_gang_assemble, 3310e14bb325SJeff Bonwick zio_gang_issue, 3311e14bb325SJeff Bonwick zio_dva_allocate, 3312e14bb325SJeff Bonwick zio_dva_free, 3313e14bb325SJeff Bonwick zio_dva_claim, 3314e14bb325SJeff Bonwick zio_ready, 3315e14bb325SJeff Bonwick zio_vdev_io_start, 3316e14bb325SJeff Bonwick zio_vdev_io_done, 3317e14bb325SJeff Bonwick zio_vdev_io_assess, 3318e14bb325SJeff Bonwick zio_checksum_verify, 3319e14bb325SJeff Bonwick zio_done 3320e14bb325SJeff Bonwick }; 3321ad135b5dSChristopher Siden 3322ad135b5dSChristopher Siden 3323ad135b5dSChristopher Siden 3324ad135b5dSChristopher Siden 3325*a2cdcdd2SPaul Dagnelie /* 3326*a2cdcdd2SPaul Dagnelie * Compare two zbookmark_phys_t's to see which we would reach first in a 3327*a2cdcdd2SPaul Dagnelie * pre-order traversal of the object tree. 3328*a2cdcdd2SPaul Dagnelie * 3329*a2cdcdd2SPaul Dagnelie * This is simple in every case aside from the meta-dnode object. For all other 3330*a2cdcdd2SPaul Dagnelie * objects, we traverse them in order (object 1 before object 2, and so on). 3331*a2cdcdd2SPaul Dagnelie * However, all of these objects are traversed while traversing object 0, since 3332*a2cdcdd2SPaul Dagnelie * the data it points to is the list of objects. Thus, we need to convert to a 3333*a2cdcdd2SPaul Dagnelie * canonical representation so we can compare meta-dnode bookmarks to 3334*a2cdcdd2SPaul Dagnelie * non-meta-dnode bookmarks. 3335*a2cdcdd2SPaul Dagnelie * 3336*a2cdcdd2SPaul Dagnelie * We do this by calculating "equivalents" for each field of the zbookmark. 3337*a2cdcdd2SPaul Dagnelie * zbookmarks outside of the meta-dnode use their own object and level, and 3338*a2cdcdd2SPaul Dagnelie * calculate the level 0 equivalent (the first L0 blkid that is contained in the 3339*a2cdcdd2SPaul Dagnelie * blocks this bookmark refers to) by multiplying their blkid by their span 3340*a2cdcdd2SPaul Dagnelie * (the number of L0 blocks contained within one block at their level). 3341*a2cdcdd2SPaul Dagnelie * zbookmarks inside the meta-dnode calculate their object equivalent 3342*a2cdcdd2SPaul Dagnelie * (which is L0equiv * dnodes per data block), use 0 for their L0equiv, and use 3343*a2cdcdd2SPaul Dagnelie * level + 1<<31 (any value larger than a level could ever be) for their level. 3344*a2cdcdd2SPaul Dagnelie * This causes them to always compare before a bookmark in their object 3345*a2cdcdd2SPaul Dagnelie * equivalent, compare appropriately to bookmarks in other objects, and to 3346*a2cdcdd2SPaul Dagnelie * compare appropriately to other bookmarks in the meta-dnode. 3347*a2cdcdd2SPaul Dagnelie */ 3348*a2cdcdd2SPaul Dagnelie int 3349*a2cdcdd2SPaul Dagnelie zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2, uint8_t ibs2, 3350*a2cdcdd2SPaul Dagnelie const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2) 3351*a2cdcdd2SPaul Dagnelie { 3352*a2cdcdd2SPaul Dagnelie /* 3353*a2cdcdd2SPaul Dagnelie * These variables represent the "equivalent" values for the zbookmark, 3354*a2cdcdd2SPaul Dagnelie * after converting zbookmarks inside the meta dnode to their 3355*a2cdcdd2SPaul Dagnelie * normal-object equivalents. 3356*a2cdcdd2SPaul Dagnelie */ 3357*a2cdcdd2SPaul Dagnelie uint64_t zb1obj, zb2obj; 3358*a2cdcdd2SPaul Dagnelie uint64_t zb1L0, zb2L0; 3359*a2cdcdd2SPaul Dagnelie uint64_t zb1level, zb2level; 3360ad135b5dSChristopher Siden 3361*a2cdcdd2SPaul Dagnelie if (zb1->zb_object == zb2->zb_object && 3362*a2cdcdd2SPaul Dagnelie zb1->zb_level == zb2->zb_level && 3363*a2cdcdd2SPaul Dagnelie zb1->zb_blkid == zb2->zb_blkid) 3364*a2cdcdd2SPaul Dagnelie return (0); 3365*a2cdcdd2SPaul Dagnelie 3366*a2cdcdd2SPaul Dagnelie /* 3367*a2cdcdd2SPaul Dagnelie * BP_SPANB calculates the span in blocks. 3368*a2cdcdd2SPaul Dagnelie */ 3369*a2cdcdd2SPaul Dagnelie zb1L0 = (zb1->zb_blkid) * BP_SPANB(ibs1, zb1->zb_level); 3370*a2cdcdd2SPaul Dagnelie zb2L0 = (zb2->zb_blkid) * BP_SPANB(ibs2, zb2->zb_level); 3371ad135b5dSChristopher Siden 3372ad135b5dSChristopher Siden if (zb1->zb_object == DMU_META_DNODE_OBJECT) { 3373*a2cdcdd2SPaul Dagnelie zb1obj = zb1L0 * (dbss1 << (SPA_MINBLOCKSHIFT - DNODE_SHIFT)); 3374*a2cdcdd2SPaul Dagnelie zb1L0 = 0; 3375*a2cdcdd2SPaul Dagnelie zb1level = zb1->zb_level + COMPARE_META_LEVEL; 3376*a2cdcdd2SPaul Dagnelie } else { 3377*a2cdcdd2SPaul Dagnelie zb1obj = zb1->zb_object; 3378*a2cdcdd2SPaul Dagnelie zb1level = zb1->zb_level; 3379ad135b5dSChristopher Siden } 3380ad135b5dSChristopher Siden 3381*a2cdcdd2SPaul Dagnelie if (zb2->zb_object == DMU_META_DNODE_OBJECT) { 3382*a2cdcdd2SPaul Dagnelie zb2obj = zb2L0 * (dbss2 << (SPA_MINBLOCKSHIFT - DNODE_SHIFT)); 3383*a2cdcdd2SPaul Dagnelie zb2L0 = 0; 3384*a2cdcdd2SPaul Dagnelie zb2level = zb2->zb_level + COMPARE_META_LEVEL; 3385*a2cdcdd2SPaul Dagnelie } else { 3386*a2cdcdd2SPaul Dagnelie zb2obj = zb2->zb_object; 3387*a2cdcdd2SPaul Dagnelie zb2level = zb2->zb_level; 3388*a2cdcdd2SPaul Dagnelie } 3389*a2cdcdd2SPaul Dagnelie 3390*a2cdcdd2SPaul Dagnelie /* Now that we have a canonical representation, do the comparison. */ 3391*a2cdcdd2SPaul Dagnelie if (zb1obj != zb2obj) 3392*a2cdcdd2SPaul Dagnelie return (zb1obj < zb2obj ? -1 : 1); 3393*a2cdcdd2SPaul Dagnelie else if (zb1L0 != zb2L0) 3394*a2cdcdd2SPaul Dagnelie return (zb1L0 < zb2L0 ? -1 : 1); 3395*a2cdcdd2SPaul Dagnelie else if (zb1level != zb2level) 3396*a2cdcdd2SPaul Dagnelie return (zb1level > zb2level ? -1 : 1); 3397*a2cdcdd2SPaul Dagnelie /* 3398*a2cdcdd2SPaul Dagnelie * This can (theoretically) happen if the bookmarks have the same object 3399*a2cdcdd2SPaul Dagnelie * and level, but different blkids, if the block sizes are not the same. 3400*a2cdcdd2SPaul Dagnelie * There is presently no way to change the indirect block sizes 3401*a2cdcdd2SPaul Dagnelie */ 3402*a2cdcdd2SPaul Dagnelie return (0); 3403*a2cdcdd2SPaul Dagnelie } 3404*a2cdcdd2SPaul Dagnelie 3405*a2cdcdd2SPaul Dagnelie /* 3406*a2cdcdd2SPaul Dagnelie * This function checks the following: given that last_block is the place that 3407*a2cdcdd2SPaul Dagnelie * our traversal stopped last time, does that guarantee that we've visited 3408*a2cdcdd2SPaul Dagnelie * every node under subtree_root? Therefore, we can't just use the raw output 3409*a2cdcdd2SPaul Dagnelie * of zbookmark_compare. We have to pass in a modified version of 3410*a2cdcdd2SPaul Dagnelie * subtree_root; by incrementing the block id, and then checking whether 3411*a2cdcdd2SPaul Dagnelie * last_block is before or equal to that, we can tell whether or not having 3412*a2cdcdd2SPaul Dagnelie * visited last_block implies that all of subtree_root's children have been 3413*a2cdcdd2SPaul Dagnelie * visited. 3414*a2cdcdd2SPaul Dagnelie */ 3415*a2cdcdd2SPaul Dagnelie boolean_t 3416*a2cdcdd2SPaul Dagnelie zbookmark_subtree_completed(const dnode_phys_t *dnp, 3417*a2cdcdd2SPaul Dagnelie const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block) 3418*a2cdcdd2SPaul Dagnelie { 3419*a2cdcdd2SPaul Dagnelie zbookmark_phys_t mod_zb = *subtree_root; 3420*a2cdcdd2SPaul Dagnelie mod_zb.zb_blkid++; 3421*a2cdcdd2SPaul Dagnelie ASSERT(last_block->zb_level == 0); 3422*a2cdcdd2SPaul Dagnelie 3423*a2cdcdd2SPaul Dagnelie /* The objset_phys_t isn't before anything. */ 3424*a2cdcdd2SPaul Dagnelie if (dnp == NULL) 3425ad135b5dSChristopher Siden return (B_FALSE); 3426*a2cdcdd2SPaul Dagnelie 3427*a2cdcdd2SPaul Dagnelie /* 3428*a2cdcdd2SPaul Dagnelie * We pass in 1ULL << (DNODE_BLOCK_SHIFT - SPA_MINBLOCKSHIFT) for the 3429*a2cdcdd2SPaul Dagnelie * data block size in sectors, because that variable is only used if 3430*a2cdcdd2SPaul Dagnelie * the bookmark refers to a block in the meta-dnode. Since we don't 3431*a2cdcdd2SPaul Dagnelie * know without examining it what object it refers to, and there's no 3432*a2cdcdd2SPaul Dagnelie * harm in passing in this value in other cases, we always pass it in. 3433*a2cdcdd2SPaul Dagnelie * 3434*a2cdcdd2SPaul Dagnelie * We pass in 0 for the indirect block size shift because zb2 must be 3435*a2cdcdd2SPaul Dagnelie * level 0. The indirect block size is only used to calculate the span 3436*a2cdcdd2SPaul Dagnelie * of the bookmark, but since the bookmark must be level 0, the span is 3437*a2cdcdd2SPaul Dagnelie * always 1, so the math works out. 3438*a2cdcdd2SPaul Dagnelie * 3439*a2cdcdd2SPaul Dagnelie * If you make changes to how the zbookmark_compare code works, be sure 3440*a2cdcdd2SPaul Dagnelie * to make sure that this code still works afterwards. 3441*a2cdcdd2SPaul Dagnelie */ 3442*a2cdcdd2SPaul Dagnelie return (zbookmark_compare(dnp->dn_datablkszsec, dnp->dn_indblkshift, 3443*a2cdcdd2SPaul Dagnelie 1ULL << (DNODE_BLOCK_SHIFT - SPA_MINBLOCKSHIFT), 0, &mod_zb, 3444*a2cdcdd2SPaul Dagnelie last_block) <= 0); 3445ad135b5dSChristopher Siden } 3446