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. 23286ef713SPaul Dagnelie * Copyright (c) 2012, 2016 by Delphix. All rights reserved. 24fa9e4066Sahrens */ 25fa9e4066Sahrens 26fa9e4066Sahrens #include <sys/zfs_context.h> 27fa9e4066Sahrens #include <sys/dmu_objset.h> 28fa9e4066Sahrens #include <sys/dmu_traverse.h> 29fa9e4066Sahrens #include <sys/dsl_dataset.h> 30fa9e4066Sahrens #include <sys/dsl_dir.h> 31fa9e4066Sahrens #include <sys/dsl_pool.h> 32fa9e4066Sahrens #include <sys/dnode.h> 33fa9e4066Sahrens #include <sys/spa.h> 34fa9e4066Sahrens #include <sys/zio.h> 35fa9e4066Sahrens #include <sys/dmu_impl.h> 360a586ceaSMark Shellenbaum #include <sys/sa.h> 370a586ceaSMark Shellenbaum #include <sys/sa_impl.h> 3888b7b0f2SMatthew Ahrens #include <sys/callb.h> 3943466aaeSMax Grossman #include <sys/zfeature.h> 40fa9e4066Sahrens 4134d7ce05SGeorge Wilson int32_t zfs_pd_bytes_max = 50 * 1024 * 1024; /* 50MB */ 4244f92b7dSChris Kirby 436e0cbcaaSMatthew Ahrens typedef struct prefetch_data { 4488b7b0f2SMatthew Ahrens kmutex_t pd_mtx; 4588b7b0f2SMatthew Ahrens kcondvar_t pd_cv; 4634d7ce05SGeorge Wilson int32_t pd_bytes_fetched; 4788b7b0f2SMatthew Ahrens int pd_flags; 4888b7b0f2SMatthew Ahrens boolean_t pd_cancel; 4988b7b0f2SMatthew Ahrens boolean_t pd_exited; 509c3fd121SMatthew Ahrens zbookmark_phys_t pd_resume; 516e0cbcaaSMatthew Ahrens } prefetch_data_t; 52fa9e4066Sahrens 536e0cbcaaSMatthew Ahrens typedef struct traverse_data { 5488b7b0f2SMatthew Ahrens spa_t *td_spa; 5588b7b0f2SMatthew Ahrens uint64_t td_objset; 5688b7b0f2SMatthew Ahrens blkptr_t *td_rootbp; 5788b7b0f2SMatthew Ahrens uint64_t td_min_txg; 587802d7bfSMatthew Ahrens zbookmark_phys_t *td_resume; 5988b7b0f2SMatthew Ahrens int td_flags; 606e0cbcaaSMatthew Ahrens prefetch_data_t *td_pfd; 617fd05ac4SMatthew Ahrens boolean_t td_paused; 62f7950bf1SMatthew Ahrens uint64_t td_hole_birth_enabled_txg; 6388b7b0f2SMatthew Ahrens blkptr_cb_t *td_func; 6488b7b0f2SMatthew Ahrens void *td_arg; 65286ef713SPaul Dagnelie boolean_t td_realloc_possible; 666e0cbcaaSMatthew Ahrens } traverse_data_t; 67fa9e4066Sahrens 686e0cbcaaSMatthew Ahrens static int traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp, 691b912ec7SGeorge Wilson uint64_t objset, uint64_t object); 70b4709335SMatthew Ahrens static void prefetch_dnode_metadata(traverse_data_t *td, const dnode_phys_t *, 711b912ec7SGeorge Wilson uint64_t objset, uint64_t object); 7214843421SMatthew Ahrens 73b24ab676SJeff Bonwick static int 745dabedeeSbonwick traverse_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg) 75ea8dc4b6Seschrock { 766e0cbcaaSMatthew Ahrens traverse_data_t *td = arg; 777802d7bfSMatthew Ahrens zbookmark_phys_t zb; 78ea8dc4b6Seschrock 7943466aaeSMax Grossman if (BP_IS_HOLE(bp)) 80b24ab676SJeff Bonwick return (0); 815dabedeeSbonwick 8288b7b0f2SMatthew Ahrens if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(td->td_spa)) 83b24ab676SJeff Bonwick return (0); 8488b7b0f2SMatthew Ahrens 85b24ab676SJeff Bonwick SET_BOOKMARK(&zb, td->td_objset, ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, 86b24ab676SJeff Bonwick bp->blk_cksum.zc_word[ZIL_ZC_SEQ]); 87b24ab676SJeff Bonwick 881b912ec7SGeorge Wilson (void) td->td_func(td->td_spa, zilog, bp, &zb, NULL, td->td_arg); 89b24ab676SJeff Bonwick 90b24ab676SJeff Bonwick return (0); 91ea8dc4b6Seschrock } 92ea8dc4b6Seschrock 93b24ab676SJeff Bonwick static int 945dabedeeSbonwick traverse_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg) 95ea8dc4b6Seschrock { 966e0cbcaaSMatthew Ahrens traverse_data_t *td = arg; 97ea8dc4b6Seschrock 98ea8dc4b6Seschrock if (lrc->lrc_txtype == TX_WRITE) { 99ea8dc4b6Seschrock lr_write_t *lr = (lr_write_t *)lrc; 100ea8dc4b6Seschrock blkptr_t *bp = &lr->lr_blkptr; 1017802d7bfSMatthew Ahrens zbookmark_phys_t zb; 102ea8dc4b6Seschrock 10343466aaeSMax Grossman if (BP_IS_HOLE(bp)) 104b24ab676SJeff Bonwick return (0); 1055dabedeeSbonwick 10688b7b0f2SMatthew Ahrens if (claim_txg == 0 || bp->blk_birth < claim_txg) 107b24ab676SJeff Bonwick return (0); 10888b7b0f2SMatthew Ahrens 1096e0cbcaaSMatthew Ahrens SET_BOOKMARK(&zb, td->td_objset, lr->lr_foid, 1106e0cbcaaSMatthew Ahrens ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp)); 111b24ab676SJeff Bonwick 1121b912ec7SGeorge Wilson (void) td->td_func(td->td_spa, zilog, bp, &zb, NULL, 113b24ab676SJeff Bonwick td->td_arg); 114ea8dc4b6Seschrock } 115b24ab676SJeff Bonwick return (0); 116ea8dc4b6Seschrock } 117ea8dc4b6Seschrock 118ea8dc4b6Seschrock static void 1196e0cbcaaSMatthew Ahrens traverse_zil(traverse_data_t *td, zil_header_t *zh) 120ea8dc4b6Seschrock { 1215dabedeeSbonwick uint64_t claim_txg = zh->zh_claim_txg; 122ea8dc4b6Seschrock zilog_t *zilog; 123ea8dc4b6Seschrock 1245dabedeeSbonwick /* 1255dabedeeSbonwick * We only want to visit blocks that have been claimed but not yet 126b24ab676SJeff Bonwick * replayed; plus, in read-only mode, blocks that are already stable. 1275dabedeeSbonwick */ 1288ad4d6ddSJeff Bonwick if (claim_txg == 0 && spa_writeable(td->td_spa)) 1295dabedeeSbonwick return; 1305dabedeeSbonwick 13188b7b0f2SMatthew Ahrens zilog = zil_alloc(spa_get_dsl(td->td_spa)->dp_meta_objset, zh); 132ea8dc4b6Seschrock 13388b7b0f2SMatthew Ahrens (void) zil_parse(zilog, traverse_zil_block, traverse_zil_record, td, 1345dabedeeSbonwick claim_txg); 135ea8dc4b6Seschrock 136ea8dc4b6Seschrock zil_free(zilog); 137ea8dc4b6Seschrock } 138ea8dc4b6Seschrock 139ad135b5dSChristopher Siden typedef enum resume_skip { 140ad135b5dSChristopher Siden RESUME_SKIP_ALL, 141ad135b5dSChristopher Siden RESUME_SKIP_NONE, 142ad135b5dSChristopher Siden RESUME_SKIP_CHILDREN 143ad135b5dSChristopher Siden } resume_skip_t; 144ad135b5dSChristopher Siden 145ad135b5dSChristopher Siden /* 146ad135b5dSChristopher Siden * Returns RESUME_SKIP_ALL if td indicates that we are resuming a traversal and 147ad135b5dSChristopher Siden * the block indicated by zb does not need to be visited at all. Returns 148ad135b5dSChristopher Siden * RESUME_SKIP_CHILDREN if we are resuming a post traversal and we reach the 149ad135b5dSChristopher Siden * resume point. This indicates that this block should be visited but not its 150ad135b5dSChristopher Siden * children (since they must have been visited in a previous traversal). 151ad135b5dSChristopher Siden * Otherwise returns RESUME_SKIP_NONE. 152ad135b5dSChristopher Siden */ 153ad135b5dSChristopher Siden static resume_skip_t 154ad135b5dSChristopher Siden resume_skip_check(traverse_data_t *td, const dnode_phys_t *dnp, 1557802d7bfSMatthew Ahrens const zbookmark_phys_t *zb) 156ad135b5dSChristopher Siden { 157ad135b5dSChristopher Siden if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume)) { 158ad135b5dSChristopher Siden /* 159ad135b5dSChristopher Siden * If we already visited this bp & everything below, 160ad135b5dSChristopher Siden * don't bother doing it again. 161ad135b5dSChristopher Siden */ 162a2cdcdd2SPaul Dagnelie if (zbookmark_subtree_completed(dnp, zb, td->td_resume)) 163ad135b5dSChristopher Siden return (RESUME_SKIP_ALL); 164ad135b5dSChristopher Siden 165ad135b5dSChristopher Siden /* 166ad135b5dSChristopher Siden * If we found the block we're trying to resume from, zero 167ad135b5dSChristopher Siden * the bookmark out to indicate that we have resumed. 168ad135b5dSChristopher Siden */ 169ad135b5dSChristopher Siden if (bcmp(zb, td->td_resume, sizeof (*zb)) == 0) { 170ad135b5dSChristopher Siden bzero(td->td_resume, sizeof (*zb)); 171ad135b5dSChristopher Siden if (td->td_flags & TRAVERSE_POST) 172ad135b5dSChristopher Siden return (RESUME_SKIP_CHILDREN); 173ad135b5dSChristopher Siden } 174ad135b5dSChristopher Siden } 175ad135b5dSChristopher Siden return (RESUME_SKIP_NONE); 176ad135b5dSChristopher Siden } 177ad135b5dSChristopher Siden 178ad135b5dSChristopher Siden static void 179b4709335SMatthew Ahrens traverse_prefetch_metadata(traverse_data_t *td, 1807802d7bfSMatthew Ahrens const blkptr_t *bp, const zbookmark_phys_t *zb) 181b4709335SMatthew Ahrens { 1827adb730bSGeorge Wilson arc_flags_t flags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH; 183b4709335SMatthew Ahrens 184b4709335SMatthew Ahrens if (!(td->td_flags & TRAVERSE_PREFETCH_METADATA)) 185b4709335SMatthew Ahrens return; 186b4709335SMatthew Ahrens /* 187b4709335SMatthew Ahrens * If we are in the process of resuming, don't prefetch, because 188b4709335SMatthew Ahrens * some children will not be needed (and in fact may have already 189b4709335SMatthew Ahrens * been freed). 190b4709335SMatthew Ahrens */ 191b4709335SMatthew Ahrens if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume)) 192b4709335SMatthew Ahrens return; 193b4709335SMatthew Ahrens if (BP_IS_HOLE(bp) || bp->blk_birth <= td->td_min_txg) 194b4709335SMatthew Ahrens return; 195b4709335SMatthew Ahrens if (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_DNODE) 196b4709335SMatthew Ahrens return; 197b4709335SMatthew Ahrens 1981b912ec7SGeorge Wilson (void) arc_read(NULL, td->td_spa, bp, NULL, NULL, 1991b912ec7SGeorge Wilson ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); 200b4709335SMatthew Ahrens } 201b4709335SMatthew Ahrens 20206315b79SMatthew Ahrens static boolean_t 20306315b79SMatthew Ahrens prefetch_needed(prefetch_data_t *pfd, const blkptr_t *bp) 20406315b79SMatthew Ahrens { 20506315b79SMatthew Ahrens ASSERT(pfd->pd_flags & TRAVERSE_PREFETCH_DATA); 20606315b79SMatthew Ahrens if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp) || 20706315b79SMatthew Ahrens BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG) 20806315b79SMatthew Ahrens return (B_FALSE); 20906315b79SMatthew Ahrens return (B_TRUE); 21006315b79SMatthew Ahrens } 21106315b79SMatthew Ahrens 212fa9e4066Sahrens static int 2136e0cbcaaSMatthew Ahrens traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp, 2147802d7bfSMatthew Ahrens const blkptr_t *bp, const zbookmark_phys_t *zb) 215fa9e4066Sahrens { 2167802d7bfSMatthew Ahrens zbookmark_phys_t czb; 2177fd05ac4SMatthew Ahrens int err = 0; 21888b7b0f2SMatthew Ahrens arc_buf_t *buf = NULL; 2196e0cbcaaSMatthew Ahrens prefetch_data_t *pd = td->td_pfd; 220cd088ea4SVictor Latushkin boolean_t hard = td->td_flags & TRAVERSE_HARD; 221fa9e4066Sahrens 222ad135b5dSChristopher Siden switch (resume_skip_check(td, dnp, zb)) { 223ad135b5dSChristopher Siden case RESUME_SKIP_ALL: 224ad135b5dSChristopher Siden return (0); 225ad135b5dSChristopher Siden case RESUME_SKIP_CHILDREN: 226ad135b5dSChristopher Siden goto post; 227ad135b5dSChristopher Siden case RESUME_SKIP_NONE: 228ad135b5dSChristopher Siden break; 229ad135b5dSChristopher Siden default: 230ad135b5dSChristopher Siden ASSERT(0); 231ad135b5dSChristopher Siden } 232ad135b5dSChristopher Siden 23343466aaeSMax Grossman if (bp->blk_birth == 0) { 23443466aaeSMax Grossman /* 235286ef713SPaul Dagnelie * Since this block has a birth time of 0 it must be one of 236286ef713SPaul Dagnelie * two things: a hole created before the 237286ef713SPaul Dagnelie * SPA_FEATURE_HOLE_BIRTH feature was enabled, or a hole 238286ef713SPaul Dagnelie * which has always been a hole in an object. 239286ef713SPaul Dagnelie * 240286ef713SPaul Dagnelie * If a file is written sparsely, then the unwritten parts of 241286ef713SPaul Dagnelie * the file were "always holes" -- that is, they have been 242286ef713SPaul Dagnelie * holes since this object was allocated. However, we (and 243286ef713SPaul Dagnelie * our callers) can not necessarily tell when an object was 244286ef713SPaul Dagnelie * allocated. Therefore, if it's possible that this object 245286ef713SPaul Dagnelie * was freed and then its object number reused, we need to 246286ef713SPaul Dagnelie * visit all the holes with birth==0. 247286ef713SPaul Dagnelie * 248286ef713SPaul Dagnelie * If it isn't possible that the object number was reused, 249286ef713SPaul Dagnelie * then if SPA_FEATURE_HOLE_BIRTH was enabled before we wrote 250286ef713SPaul Dagnelie * all the blocks we will visit as part of this traversal, 251286ef713SPaul Dagnelie * then this hole must have always existed, so we can skip 252286ef713SPaul Dagnelie * it. We visit blocks born after (exclusive) td_min_txg. 253286ef713SPaul Dagnelie * 254286ef713SPaul Dagnelie * Note that the meta-dnode cannot be reallocated. 25543466aaeSMax Grossman */ 256286ef713SPaul Dagnelie if ((!td->td_realloc_possible || 257286ef713SPaul Dagnelie zb->zb_object == DMU_META_DNODE_OBJECT) && 258286ef713SPaul Dagnelie td->td_hole_birth_enabled_txg <= td->td_min_txg) 25943466aaeSMax Grossman return (0); 26043466aaeSMax Grossman } else if (bp->blk_birth <= td->td_min_txg) { 26143466aaeSMax Grossman return (0); 262fa9e4066Sahrens } 263fa9e4066Sahrens 26406315b79SMatthew Ahrens if (pd != NULL && !pd->pd_exited && prefetch_needed(pd, bp)) { 26534d7ce05SGeorge Wilson uint64_t size = BP_GET_LSIZE(bp); 26688b7b0f2SMatthew Ahrens mutex_enter(&pd->pd_mtx); 26734d7ce05SGeorge Wilson ASSERT(pd->pd_bytes_fetched >= 0); 26834d7ce05SGeorge Wilson while (pd->pd_bytes_fetched < size && !pd->pd_exited) 26988b7b0f2SMatthew Ahrens cv_wait(&pd->pd_cv, &pd->pd_mtx); 27034d7ce05SGeorge Wilson pd->pd_bytes_fetched -= size; 27188b7b0f2SMatthew Ahrens cv_broadcast(&pd->pd_cv); 27288b7b0f2SMatthew Ahrens mutex_exit(&pd->pd_mtx); 273fa9e4066Sahrens } 274fa9e4066Sahrens 27506315b79SMatthew Ahrens if (BP_IS_HOLE(bp)) { 27606315b79SMatthew Ahrens err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg); 27706315b79SMatthew Ahrens if (err != 0) 27806315b79SMatthew Ahrens goto post; 27906315b79SMatthew Ahrens return (0); 28006315b79SMatthew Ahrens } 28106315b79SMatthew Ahrens 28288b7b0f2SMatthew Ahrens if (td->td_flags & TRAVERSE_PRE) { 2831b912ec7SGeorge Wilson err = td->td_func(td->td_spa, NULL, bp, zb, dnp, 2843f9d6ad7SLin Ling td->td_arg); 28599d5e173STim Haley if (err == TRAVERSE_VISIT_NO_CHILDREN) 28699d5e173STim Haley return (0); 287ad135b5dSChristopher Siden if (err != 0) 288ad135b5dSChristopher Siden goto post; 28988b7b0f2SMatthew Ahrens } 29088b7b0f2SMatthew Ahrens 29188b7b0f2SMatthew Ahrens if (BP_GET_LEVEL(bp) > 0) { 2927adb730bSGeorge Wilson arc_flags_t flags = ARC_FLAG_WAIT; 29388b7b0f2SMatthew Ahrens int i; 29488b7b0f2SMatthew Ahrens blkptr_t *cbp; 29588b7b0f2SMatthew Ahrens int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT; 29688b7b0f2SMatthew Ahrens 2971b912ec7SGeorge Wilson err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf, 29888b7b0f2SMatthew Ahrens ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); 2993b2aab18SMatthew Ahrens if (err != 0) 3007fd05ac4SMatthew Ahrens goto post; 30188b7b0f2SMatthew Ahrens cbp = buf->b_data; 302b4709335SMatthew Ahrens 303b4709335SMatthew Ahrens for (i = 0; i < epb; i++) { 30488b7b0f2SMatthew Ahrens SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object, 30588b7b0f2SMatthew Ahrens zb->zb_level - 1, 30688b7b0f2SMatthew Ahrens zb->zb_blkid * epb + i); 3071b912ec7SGeorge Wilson traverse_prefetch_metadata(td, &cbp[i], &czb); 308b4709335SMatthew Ahrens } 309b4709335SMatthew Ahrens 310b4709335SMatthew Ahrens /* recursively visitbp() blocks below this */ 311b4709335SMatthew Ahrens for (i = 0; i < epb; i++) { 312b4709335SMatthew Ahrens SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object, 313b4709335SMatthew Ahrens zb->zb_level - 1, 314b4709335SMatthew Ahrens zb->zb_blkid * epb + i); 3151b912ec7SGeorge Wilson err = traverse_visitbp(td, dnp, &cbp[i], &czb); 3167fd05ac4SMatthew Ahrens if (err != 0) 31788b7b0f2SMatthew Ahrens break; 31888b7b0f2SMatthew Ahrens } 31988b7b0f2SMatthew Ahrens } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) { 3207adb730bSGeorge Wilson arc_flags_t flags = ARC_FLAG_WAIT; 32114843421SMatthew Ahrens int i; 32288b7b0f2SMatthew Ahrens int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT; 32388b7b0f2SMatthew Ahrens 3241b912ec7SGeorge Wilson err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf, 32588b7b0f2SMatthew Ahrens ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); 3263b2aab18SMatthew Ahrens if (err != 0) 3277fd05ac4SMatthew Ahrens goto post; 3289c3fd121SMatthew Ahrens dnode_phys_t *child_dnp = buf->b_data; 329b4709335SMatthew Ahrens 330b4709335SMatthew Ahrens for (i = 0; i < epb; i++) { 3319c3fd121SMatthew Ahrens prefetch_dnode_metadata(td, &child_dnp[i], 3329c3fd121SMatthew Ahrens zb->zb_objset, zb->zb_blkid * epb + i); 333b4709335SMatthew Ahrens } 33488b7b0f2SMatthew Ahrens 33588b7b0f2SMatthew Ahrens /* recursively visitbp() blocks below this */ 336b4709335SMatthew Ahrens for (i = 0; i < epb; i++) { 3379c3fd121SMatthew Ahrens err = traverse_dnode(td, &child_dnp[i], 3389c3fd121SMatthew Ahrens zb->zb_objset, zb->zb_blkid * epb + i); 3397fd05ac4SMatthew Ahrens if (err != 0) 34088b7b0f2SMatthew Ahrens break; 34188b7b0f2SMatthew Ahrens } 34288b7b0f2SMatthew Ahrens } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) { 3437adb730bSGeorge Wilson arc_flags_t flags = ARC_FLAG_WAIT; 34488b7b0f2SMatthew Ahrens 3451b912ec7SGeorge Wilson err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf, 34688b7b0f2SMatthew Ahrens ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); 3473b2aab18SMatthew Ahrens if (err != 0) 3487fd05ac4SMatthew Ahrens goto post; 34988b7b0f2SMatthew Ahrens 3509c3fd121SMatthew Ahrens objset_phys_t *osp = buf->b_data; 3519c3fd121SMatthew Ahrens prefetch_dnode_metadata(td, &osp->os_meta_dnode, zb->zb_objset, 352b4709335SMatthew Ahrens DMU_META_DNODE_OBJECT); 353286ef713SPaul Dagnelie /* 354286ef713SPaul Dagnelie * See the block comment above for the goal of this variable. 355286ef713SPaul Dagnelie * If the maxblkid of the meta-dnode is 0, then we know that 356286ef713SPaul Dagnelie * we've never had more than DNODES_PER_BLOCK objects in the 357286ef713SPaul Dagnelie * dataset, which means we can't have reused any object ids. 358286ef713SPaul Dagnelie */ 359286ef713SPaul Dagnelie if (osp->os_meta_dnode.dn_maxblkid == 0) 360286ef713SPaul Dagnelie td->td_realloc_possible = B_FALSE; 361286ef713SPaul Dagnelie 362b4709335SMatthew Ahrens if (arc_buf_size(buf) >= sizeof (objset_phys_t)) { 363b4709335SMatthew Ahrens prefetch_dnode_metadata(td, &osp->os_groupused_dnode, 36448f1b90eSMatthew Ahrens zb->zb_objset, DMU_GROUPUSED_OBJECT); 36548f1b90eSMatthew Ahrens prefetch_dnode_metadata(td, &osp->os_userused_dnode, 3661b912ec7SGeorge Wilson zb->zb_objset, DMU_USERUSED_OBJECT); 367b4709335SMatthew Ahrens } 368b4709335SMatthew Ahrens 3699c3fd121SMatthew Ahrens err = traverse_dnode(td, &osp->os_meta_dnode, zb->zb_objset, 370b24ab676SJeff Bonwick DMU_META_DNODE_OBJECT); 37114843421SMatthew Ahrens if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) { 3729c3fd121SMatthew Ahrens err = traverse_dnode(td, &osp->os_groupused_dnode, 3739c3fd121SMatthew Ahrens zb->zb_objset, DMU_GROUPUSED_OBJECT); 37414843421SMatthew Ahrens } 37514843421SMatthew Ahrens if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) { 3769c3fd121SMatthew Ahrens err = traverse_dnode(td, &osp->os_userused_dnode, 3779c3fd121SMatthew Ahrens zb->zb_objset, DMU_USERUSED_OBJECT); 37888b7b0f2SMatthew Ahrens } 37988b7b0f2SMatthew Ahrens } 38088b7b0f2SMatthew Ahrens 38188b7b0f2SMatthew Ahrens if (buf) 382*dcbf3bd6SGeorge Wilson arc_buf_destroy(buf, &buf); 38388b7b0f2SMatthew Ahrens 384ad135b5dSChristopher Siden post: 3857fd05ac4SMatthew Ahrens if (err == 0 && (td->td_flags & TRAVERSE_POST)) 3861b912ec7SGeorge Wilson err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg); 3877fd05ac4SMatthew Ahrens 3887fd05ac4SMatthew Ahrens if (hard && (err == EIO || err == ECKSUM)) { 3897fd05ac4SMatthew Ahrens /* 3907fd05ac4SMatthew Ahrens * Ignore this disk error as requested by the HARD flag, 3917fd05ac4SMatthew Ahrens * and continue traversal. 3927fd05ac4SMatthew Ahrens */ 3937fd05ac4SMatthew Ahrens err = 0; 394ad135b5dSChristopher Siden } 395ad135b5dSChristopher Siden 3967fd05ac4SMatthew Ahrens /* 3977fd05ac4SMatthew Ahrens * If we are stopping here, set td_resume. 3987fd05ac4SMatthew Ahrens */ 3997fd05ac4SMatthew Ahrens if (td->td_resume != NULL && err != 0 && !td->td_paused) { 4007fd05ac4SMatthew Ahrens td->td_resume->zb_objset = zb->zb_objset; 4017fd05ac4SMatthew Ahrens td->td_resume->zb_object = zb->zb_object; 4027fd05ac4SMatthew Ahrens td->td_resume->zb_level = 0; 4037fd05ac4SMatthew Ahrens /* 4047fd05ac4SMatthew Ahrens * If we have stopped on an indirect block (e.g. due to 4057fd05ac4SMatthew Ahrens * i/o error), we have not visited anything below it. 4067fd05ac4SMatthew Ahrens * Set the bookmark to the first level-0 block that we need 4077fd05ac4SMatthew Ahrens * to visit. This way, the resuming code does not need to 4087fd05ac4SMatthew Ahrens * deal with resuming from indirect blocks. 4099c3fd121SMatthew Ahrens * 4109c3fd121SMatthew Ahrens * Note, if zb_level <= 0, dnp may be NULL, so we don't want 4119c3fd121SMatthew Ahrens * to dereference it. 4127fd05ac4SMatthew Ahrens */ 4139c3fd121SMatthew Ahrens td->td_resume->zb_blkid = zb->zb_blkid; 4149c3fd121SMatthew Ahrens if (zb->zb_level > 0) { 4159c3fd121SMatthew Ahrens td->td_resume->zb_blkid <<= zb->zb_level * 4169c3fd121SMatthew Ahrens (dnp->dn_indblkshift - SPA_BLKPTRSHIFT); 4179c3fd121SMatthew Ahrens } 4187fd05ac4SMatthew Ahrens td->td_paused = B_TRUE; 4193f9d6ad7SLin Ling } 42088b7b0f2SMatthew Ahrens 4217fd05ac4SMatthew Ahrens return (err); 42288b7b0f2SMatthew Ahrens } 42388b7b0f2SMatthew Ahrens 424b4709335SMatthew Ahrens static void 425b4709335SMatthew Ahrens prefetch_dnode_metadata(traverse_data_t *td, const dnode_phys_t *dnp, 4261b912ec7SGeorge Wilson uint64_t objset, uint64_t object) 427b4709335SMatthew Ahrens { 428b4709335SMatthew Ahrens int j; 4297802d7bfSMatthew Ahrens zbookmark_phys_t czb; 430b4709335SMatthew Ahrens 431b4709335SMatthew Ahrens for (j = 0; j < dnp->dn_nblkptr; j++) { 432b4709335SMatthew Ahrens SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j); 4331b912ec7SGeorge Wilson traverse_prefetch_metadata(td, &dnp->dn_blkptr[j], &czb); 434b4709335SMatthew Ahrens } 435b4709335SMatthew Ahrens 436b4709335SMatthew Ahrens if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) { 437b4709335SMatthew Ahrens SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID); 4381b912ec7SGeorge Wilson traverse_prefetch_metadata(td, &dnp->dn_spill, &czb); 439b4709335SMatthew Ahrens } 440b4709335SMatthew Ahrens } 441b4709335SMatthew Ahrens 44214843421SMatthew Ahrens static int 4436e0cbcaaSMatthew Ahrens traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp, 4441b912ec7SGeorge Wilson uint64_t objset, uint64_t object) 44514843421SMatthew Ahrens { 4467fd05ac4SMatthew Ahrens int j, err = 0; 4477802d7bfSMatthew Ahrens zbookmark_phys_t czb; 44814843421SMatthew Ahrens 4499c3fd121SMatthew Ahrens if (object != DMU_META_DNODE_OBJECT && td->td_resume != NULL && 4509c3fd121SMatthew Ahrens object < td->td_resume->zb_object) 4519c3fd121SMatthew Ahrens return (0); 4529c3fd121SMatthew Ahrens 453a2cdcdd2SPaul Dagnelie if (td->td_flags & TRAVERSE_PRE) { 454a2cdcdd2SPaul Dagnelie SET_BOOKMARK(&czb, objset, object, ZB_DNODE_LEVEL, 455a2cdcdd2SPaul Dagnelie ZB_DNODE_BLKID); 456a2cdcdd2SPaul Dagnelie err = td->td_func(td->td_spa, NULL, NULL, &czb, dnp, 457a2cdcdd2SPaul Dagnelie td->td_arg); 458a2cdcdd2SPaul Dagnelie if (err == TRAVERSE_VISIT_NO_CHILDREN) 459a2cdcdd2SPaul Dagnelie return (0); 460a2cdcdd2SPaul Dagnelie if (err != 0) 461a2cdcdd2SPaul Dagnelie return (err); 462a2cdcdd2SPaul Dagnelie } 463a2cdcdd2SPaul Dagnelie 46414843421SMatthew Ahrens for (j = 0; j < dnp->dn_nblkptr; j++) { 46514843421SMatthew Ahrens SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j); 4661b912ec7SGeorge Wilson err = traverse_visitbp(td, dnp, &dnp->dn_blkptr[j], &czb); 4677fd05ac4SMatthew Ahrens if (err != 0) 46814843421SMatthew Ahrens break; 4693f9d6ad7SLin Ling } 4703f9d6ad7SLin Ling 471a2cdcdd2SPaul Dagnelie if (err == 0 && (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) { 472b4709335SMatthew Ahrens SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID); 4731b912ec7SGeorge Wilson err = traverse_visitbp(td, dnp, &dnp->dn_spill, &czb); 4747fd05ac4SMatthew Ahrens } 475a2cdcdd2SPaul Dagnelie 476a2cdcdd2SPaul Dagnelie if (err == 0 && (td->td_flags & TRAVERSE_POST)) { 477a2cdcdd2SPaul Dagnelie SET_BOOKMARK(&czb, objset, object, ZB_DNODE_LEVEL, 478a2cdcdd2SPaul Dagnelie ZB_DNODE_BLKID); 479a2cdcdd2SPaul Dagnelie err = td->td_func(td->td_spa, NULL, NULL, &czb, dnp, 480a2cdcdd2SPaul Dagnelie td->td_arg); 481a2cdcdd2SPaul Dagnelie if (err == TRAVERSE_VISIT_NO_CHILDREN) 482a2cdcdd2SPaul Dagnelie return (0); 483a2cdcdd2SPaul Dagnelie if (err != 0) 484a2cdcdd2SPaul Dagnelie return (err); 485a2cdcdd2SPaul Dagnelie } 4863f9d6ad7SLin Ling return (err); 48714843421SMatthew Ahrens } 48814843421SMatthew Ahrens 48988b7b0f2SMatthew Ahrens /* ARGSUSED */ 49088b7b0f2SMatthew Ahrens static int 491b24ab676SJeff Bonwick traverse_prefetcher(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 4927802d7bfSMatthew Ahrens const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 49388b7b0f2SMatthew Ahrens { 4946e0cbcaaSMatthew Ahrens prefetch_data_t *pfd = arg; 4957adb730bSGeorge Wilson arc_flags_t aflags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH; 49688b7b0f2SMatthew Ahrens 49734d7ce05SGeorge Wilson ASSERT(pfd->pd_bytes_fetched >= 0); 498a2cdcdd2SPaul Dagnelie if (bp == NULL) 499a2cdcdd2SPaul Dagnelie return (0); 50088b7b0f2SMatthew Ahrens if (pfd->pd_cancel) 501be6fd75aSMatthew Ahrens return (SET_ERROR(EINTR)); 50288b7b0f2SMatthew Ahrens 50306315b79SMatthew Ahrens if (!prefetch_needed(pfd, bp)) 50488b7b0f2SMatthew Ahrens return (0); 50588b7b0f2SMatthew Ahrens 50688b7b0f2SMatthew Ahrens mutex_enter(&pfd->pd_mtx); 50734d7ce05SGeorge Wilson while (!pfd->pd_cancel && pfd->pd_bytes_fetched >= zfs_pd_bytes_max) 50888b7b0f2SMatthew Ahrens cv_wait(&pfd->pd_cv, &pfd->pd_mtx); 50934d7ce05SGeorge Wilson pfd->pd_bytes_fetched += BP_GET_LSIZE(bp); 51088b7b0f2SMatthew Ahrens cv_broadcast(&pfd->pd_cv); 51188b7b0f2SMatthew Ahrens mutex_exit(&pfd->pd_mtx); 51288b7b0f2SMatthew Ahrens 5131b912ec7SGeorge Wilson (void) arc_read(NULL, spa, bp, NULL, NULL, ZIO_PRIORITY_ASYNC_READ, 5141b912ec7SGeorge Wilson ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &aflags, zb); 51588b7b0f2SMatthew Ahrens 51688b7b0f2SMatthew Ahrens return (0); 51788b7b0f2SMatthew Ahrens } 51888b7b0f2SMatthew Ahrens 51988b7b0f2SMatthew Ahrens static void 52088b7b0f2SMatthew Ahrens traverse_prefetch_thread(void *arg) 52188b7b0f2SMatthew Ahrens { 5226e0cbcaaSMatthew Ahrens traverse_data_t *td_main = arg; 5236e0cbcaaSMatthew Ahrens traverse_data_t td = *td_main; 5247802d7bfSMatthew Ahrens zbookmark_phys_t czb; 52588b7b0f2SMatthew Ahrens 52688b7b0f2SMatthew Ahrens td.td_func = traverse_prefetcher; 52788b7b0f2SMatthew Ahrens td.td_arg = td_main->td_pfd; 52888b7b0f2SMatthew Ahrens td.td_pfd = NULL; 5299c3fd121SMatthew Ahrens td.td_resume = &td_main->td_pfd->pd_resume; 53088b7b0f2SMatthew Ahrens 531b24ab676SJeff Bonwick SET_BOOKMARK(&czb, td.td_objset, 532b24ab676SJeff Bonwick ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 5331b912ec7SGeorge Wilson (void) traverse_visitbp(&td, NULL, td.td_rootbp, &czb); 53488b7b0f2SMatthew Ahrens 53588b7b0f2SMatthew Ahrens mutex_enter(&td_main->td_pfd->pd_mtx); 53688b7b0f2SMatthew Ahrens td_main->td_pfd->pd_exited = B_TRUE; 53788b7b0f2SMatthew Ahrens cv_broadcast(&td_main->td_pfd->pd_cv); 53888b7b0f2SMatthew Ahrens mutex_exit(&td_main->td_pfd->pd_mtx); 539fa9e4066Sahrens } 540fa9e4066Sahrens 541fa9e4066Sahrens /* 54288b7b0f2SMatthew Ahrens * NB: dataset must not be changing on-disk (eg, is a snapshot or we are 54388b7b0f2SMatthew Ahrens * in syncing context). 544fa9e4066Sahrens */ 54588b7b0f2SMatthew Ahrens static int 546ad135b5dSChristopher Siden traverse_impl(spa_t *spa, dsl_dataset_t *ds, uint64_t objset, blkptr_t *rootbp, 5477802d7bfSMatthew Ahrens uint64_t txg_start, zbookmark_phys_t *resume, int flags, 548ad135b5dSChristopher Siden blkptr_cb_t func, void *arg) 549fa9e4066Sahrens { 5506e0cbcaaSMatthew Ahrens traverse_data_t td; 5516e0cbcaaSMatthew Ahrens prefetch_data_t pd = { 0 }; 5527802d7bfSMatthew Ahrens zbookmark_phys_t czb; 55388b7b0f2SMatthew Ahrens int err; 554fa9e4066Sahrens 555ad135b5dSChristopher Siden ASSERT(ds == NULL || objset == ds->ds_object); 556ad135b5dSChristopher Siden ASSERT(!(flags & TRAVERSE_PRE) || !(flags & TRAVERSE_POST)); 557ad135b5dSChristopher Siden 55888b7b0f2SMatthew Ahrens td.td_spa = spa; 559ad135b5dSChristopher Siden td.td_objset = objset; 56088b7b0f2SMatthew Ahrens td.td_rootbp = rootbp; 56188b7b0f2SMatthew Ahrens td.td_min_txg = txg_start; 562ad135b5dSChristopher Siden td.td_resume = resume; 56388b7b0f2SMatthew Ahrens td.td_func = func; 56488b7b0f2SMatthew Ahrens td.td_arg = arg; 56588b7b0f2SMatthew Ahrens td.td_pfd = &pd; 56688b7b0f2SMatthew Ahrens td.td_flags = flags; 5677fd05ac4SMatthew Ahrens td.td_paused = B_FALSE; 568286ef713SPaul Dagnelie td.td_realloc_possible = (txg_start == 0 ? B_FALSE : B_TRUE); 569fa9e4066Sahrens 570f7950bf1SMatthew Ahrens if (spa_feature_is_active(spa, SPA_FEATURE_HOLE_BIRTH)) { 571f7950bf1SMatthew Ahrens VERIFY(spa_feature_enabled_txg(spa, 572f7950bf1SMatthew Ahrens SPA_FEATURE_HOLE_BIRTH, &td.td_hole_birth_enabled_txg)); 573f7950bf1SMatthew Ahrens } else { 574286ef713SPaul Dagnelie td.td_hole_birth_enabled_txg = UINT64_MAX; 575f7950bf1SMatthew Ahrens } 576f7950bf1SMatthew Ahrens 57788b7b0f2SMatthew Ahrens pd.pd_flags = flags; 5789c3fd121SMatthew Ahrens if (resume != NULL) 5799c3fd121SMatthew Ahrens pd.pd_resume = *resume; 58088b7b0f2SMatthew Ahrens mutex_init(&pd.pd_mtx, NULL, MUTEX_DEFAULT, NULL); 58188b7b0f2SMatthew Ahrens cv_init(&pd.pd_cv, NULL, CV_DEFAULT, NULL); 582fa9e4066Sahrens 5836e0cbcaaSMatthew Ahrens /* See comment on ZIL traversal in dsl_scan_visitds. */ 584bc9014e6SJustin Gibbs if (ds != NULL && !ds->ds_is_snapshot && !BP_IS_HOLE(rootbp)) { 5857adb730bSGeorge Wilson arc_flags_t flags = ARC_FLAG_WAIT; 5863b2aab18SMatthew Ahrens objset_phys_t *osp; 5873b2aab18SMatthew Ahrens arc_buf_t *buf; 5886e0cbcaaSMatthew Ahrens 5893b2aab18SMatthew Ahrens err = arc_read(NULL, td.td_spa, rootbp, 5903b2aab18SMatthew Ahrens arc_getbuf_func, &buf, 5913b2aab18SMatthew Ahrens ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, NULL); 5923b2aab18SMatthew Ahrens if (err != 0) 5936e0cbcaaSMatthew Ahrens return (err); 5946e0cbcaaSMatthew Ahrens 5953b2aab18SMatthew Ahrens osp = buf->b_data; 5963b2aab18SMatthew Ahrens traverse_zil(&td, &osp->os_zil_header); 597*dcbf3bd6SGeorge Wilson arc_buf_destroy(buf, &buf); 5986e0cbcaaSMatthew Ahrens } 5996e0cbcaaSMatthew Ahrens 600b4709335SMatthew Ahrens if (!(flags & TRAVERSE_PREFETCH_DATA) || 60188b7b0f2SMatthew Ahrens 0 == taskq_dispatch(system_taskq, traverse_prefetch_thread, 60288b7b0f2SMatthew Ahrens &td, TQ_NOQUEUE)) 60388b7b0f2SMatthew Ahrens pd.pd_exited = B_TRUE; 604fa9e4066Sahrens 6056e0cbcaaSMatthew Ahrens SET_BOOKMARK(&czb, td.td_objset, 606b24ab676SJeff Bonwick ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 6071b912ec7SGeorge Wilson err = traverse_visitbp(&td, NULL, rootbp, &czb); 608fa9e4066Sahrens 60988b7b0f2SMatthew Ahrens mutex_enter(&pd.pd_mtx); 61088b7b0f2SMatthew Ahrens pd.pd_cancel = B_TRUE; 61188b7b0f2SMatthew Ahrens cv_broadcast(&pd.pd_cv); 61288b7b0f2SMatthew Ahrens while (!pd.pd_exited) 61388b7b0f2SMatthew Ahrens cv_wait(&pd.pd_cv, &pd.pd_mtx); 61488b7b0f2SMatthew Ahrens mutex_exit(&pd.pd_mtx); 61588b7b0f2SMatthew Ahrens 61688b7b0f2SMatthew Ahrens mutex_destroy(&pd.pd_mtx); 61788b7b0f2SMatthew Ahrens cv_destroy(&pd.pd_cv); 61888b7b0f2SMatthew Ahrens 61988b7b0f2SMatthew Ahrens return (err); 620fa9e4066Sahrens } 621fa9e4066Sahrens 62288b7b0f2SMatthew Ahrens /* 62388b7b0f2SMatthew Ahrens * NB: dataset must not be changing on-disk (eg, is a snapshot or we are 62488b7b0f2SMatthew Ahrens * in syncing context). 62588b7b0f2SMatthew Ahrens */ 62688b7b0f2SMatthew Ahrens int 6279c3fd121SMatthew Ahrens traverse_dataset_resume(dsl_dataset_t *ds, uint64_t txg_start, 6289c3fd121SMatthew Ahrens zbookmark_phys_t *resume, 6299c3fd121SMatthew Ahrens int flags, blkptr_cb_t func, void *arg) 630fa9e4066Sahrens { 631ad135b5dSChristopher Siden return (traverse_impl(ds->ds_dir->dd_pool->dp_spa, ds, ds->ds_object, 6329c3fd121SMatthew Ahrens &dsl_dataset_phys(ds)->ds_bp, txg_start, resume, flags, func, arg)); 6339c3fd121SMatthew Ahrens } 6349c3fd121SMatthew Ahrens 6359c3fd121SMatthew Ahrens int 6369c3fd121SMatthew Ahrens traverse_dataset(dsl_dataset_t *ds, uint64_t txg_start, 6379c3fd121SMatthew Ahrens int flags, blkptr_cb_t func, void *arg) 6389c3fd121SMatthew Ahrens { 6399c3fd121SMatthew Ahrens return (traverse_dataset_resume(ds, txg_start, NULL, flags, func, arg)); 640ad135b5dSChristopher Siden } 641ad135b5dSChristopher Siden 642ad135b5dSChristopher Siden int 643ad135b5dSChristopher Siden traverse_dataset_destroyed(spa_t *spa, blkptr_t *blkptr, 6447802d7bfSMatthew Ahrens uint64_t txg_start, zbookmark_phys_t *resume, int flags, 645ad135b5dSChristopher Siden blkptr_cb_t func, void *arg) 646ad135b5dSChristopher Siden { 647ad135b5dSChristopher Siden return (traverse_impl(spa, NULL, ZB_DESTROYED_OBJSET, 648ad135b5dSChristopher Siden blkptr, txg_start, resume, flags, func, arg)); 649fa9e4066Sahrens } 650fa9e4066Sahrens 65188b7b0f2SMatthew Ahrens /* 65288b7b0f2SMatthew Ahrens * NB: pool must not be changing on-disk (eg, from zdb or sync context). 65388b7b0f2SMatthew Ahrens */ 65488b7b0f2SMatthew Ahrens int 655bbfd46c4SJeff Bonwick traverse_pool(spa_t *spa, uint64_t txg_start, int flags, 656bbfd46c4SJeff Bonwick blkptr_cb_t func, void *arg) 657fa9e4066Sahrens { 6587fd05ac4SMatthew Ahrens int err; 65988b7b0f2SMatthew Ahrens dsl_pool_t *dp = spa_get_dsl(spa); 66088b7b0f2SMatthew Ahrens objset_t *mos = dp->dp_meta_objset; 661cd088ea4SVictor Latushkin boolean_t hard = (flags & TRAVERSE_HARD); 662fa9e4066Sahrens 66388b7b0f2SMatthew Ahrens /* visit the MOS */ 664ad135b5dSChristopher Siden err = traverse_impl(spa, NULL, 0, spa_get_rootblkptr(spa), 665ad135b5dSChristopher Siden txg_start, NULL, flags, func, arg); 6663b2aab18SMatthew Ahrens if (err != 0) 66788b7b0f2SMatthew Ahrens return (err); 668fa9e4066Sahrens 66988b7b0f2SMatthew Ahrens /* visit each dataset */ 6709c3fd121SMatthew Ahrens for (uint64_t obj = 1; err == 0; 6719c3fd121SMatthew Ahrens err = dmu_object_next(mos, &obj, B_FALSE, txg_start)) { 67288b7b0f2SMatthew Ahrens dmu_object_info_t doi; 673fa9e4066Sahrens 67488b7b0f2SMatthew Ahrens err = dmu_object_info(mos, obj, &doi); 6753b2aab18SMatthew Ahrens if (err != 0) { 6767fd05ac4SMatthew Ahrens if (hard) 677cd088ea4SVictor Latushkin continue; 6787fd05ac4SMatthew Ahrens break; 679cd088ea4SVictor Latushkin } 680fa9e4066Sahrens 6812acef22dSMatthew Ahrens if (doi.doi_bonus_type == DMU_OT_DSL_DATASET) { 68288b7b0f2SMatthew Ahrens dsl_dataset_t *ds; 683468c413aSTim Haley uint64_t txg = txg_start; 684468c413aSTim Haley 6853b2aab18SMatthew Ahrens dsl_pool_config_enter(dp, FTAG); 68688b7b0f2SMatthew Ahrens err = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); 6873b2aab18SMatthew Ahrens dsl_pool_config_exit(dp, FTAG); 6883b2aab18SMatthew Ahrens if (err != 0) { 6897fd05ac4SMatthew Ahrens if (hard) 690cd088ea4SVictor Latushkin continue; 6917fd05ac4SMatthew Ahrens break; 692cd088ea4SVictor Latushkin } 693c1379625SJustin T. Gibbs if (dsl_dataset_phys(ds)->ds_prev_snap_txg > txg) 694c1379625SJustin T. Gibbs txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; 695bbfd46c4SJeff Bonwick err = traverse_dataset(ds, txg, flags, func, arg); 69688b7b0f2SMatthew Ahrens dsl_dataset_rele(ds, FTAG); 6977fd05ac4SMatthew Ahrens if (err != 0) 6987fd05ac4SMatthew Ahrens break; 699fa9e4066Sahrens } 700fa9e4066Sahrens } 70188b7b0f2SMatthew Ahrens if (err == ESRCH) 70288b7b0f2SMatthew Ahrens err = 0; 7037fd05ac4SMatthew Ahrens return (err); 704fa9e4066Sahrens } 705