1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2012, 2014 by Delphix. All rights reserved. 24 */ 25 26 #include <sys/zfs_context.h> 27 #include <sys/dmu_objset.h> 28 #include <sys/dmu_traverse.h> 29 #include <sys/dsl_dataset.h> 30 #include <sys/dsl_dir.h> 31 #include <sys/dsl_pool.h> 32 #include <sys/dnode.h> 33 #include <sys/spa.h> 34 #include <sys/zio.h> 35 #include <sys/dmu_impl.h> 36 #include <sys/sa.h> 37 #include <sys/sa_impl.h> 38 #include <sys/callb.h> 39 #include <sys/zfeature.h> 40 41 int32_t zfs_pd_bytes_max = 50 * 1024 * 1024; /* 50MB */ 42 43 typedef struct prefetch_data { 44 kmutex_t pd_mtx; 45 kcondvar_t pd_cv; 46 int32_t pd_bytes_fetched; 47 int pd_flags; 48 boolean_t pd_cancel; 49 boolean_t pd_exited; 50 zbookmark_phys_t pd_resume; 51 } prefetch_data_t; 52 53 typedef struct traverse_data { 54 spa_t *td_spa; 55 uint64_t td_objset; 56 blkptr_t *td_rootbp; 57 uint64_t td_min_txg; 58 zbookmark_phys_t *td_resume; 59 int td_flags; 60 prefetch_data_t *td_pfd; 61 boolean_t td_paused; 62 uint64_t td_hole_birth_enabled_txg; 63 blkptr_cb_t *td_func; 64 void *td_arg; 65 } traverse_data_t; 66 67 static int traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp, 68 uint64_t objset, uint64_t object); 69 static void prefetch_dnode_metadata(traverse_data_t *td, const dnode_phys_t *, 70 uint64_t objset, uint64_t object); 71 72 static int 73 traverse_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg) 74 { 75 traverse_data_t *td = arg; 76 zbookmark_phys_t zb; 77 78 if (BP_IS_HOLE(bp)) 79 return (0); 80 81 if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(td->td_spa)) 82 return (0); 83 84 SET_BOOKMARK(&zb, td->td_objset, ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, 85 bp->blk_cksum.zc_word[ZIL_ZC_SEQ]); 86 87 (void) td->td_func(td->td_spa, zilog, bp, &zb, NULL, td->td_arg); 88 89 return (0); 90 } 91 92 static int 93 traverse_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg) 94 { 95 traverse_data_t *td = arg; 96 97 if (lrc->lrc_txtype == TX_WRITE) { 98 lr_write_t *lr = (lr_write_t *)lrc; 99 blkptr_t *bp = &lr->lr_blkptr; 100 zbookmark_phys_t zb; 101 102 if (BP_IS_HOLE(bp)) 103 return (0); 104 105 if (claim_txg == 0 || bp->blk_birth < claim_txg) 106 return (0); 107 108 SET_BOOKMARK(&zb, td->td_objset, lr->lr_foid, 109 ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp)); 110 111 (void) td->td_func(td->td_spa, zilog, bp, &zb, NULL, 112 td->td_arg); 113 } 114 return (0); 115 } 116 117 static void 118 traverse_zil(traverse_data_t *td, zil_header_t *zh) 119 { 120 uint64_t claim_txg = zh->zh_claim_txg; 121 zilog_t *zilog; 122 123 /* 124 * We only want to visit blocks that have been claimed but not yet 125 * replayed; plus, in read-only mode, blocks that are already stable. 126 */ 127 if (claim_txg == 0 && spa_writeable(td->td_spa)) 128 return; 129 130 zilog = zil_alloc(spa_get_dsl(td->td_spa)->dp_meta_objset, zh); 131 132 (void) zil_parse(zilog, traverse_zil_block, traverse_zil_record, td, 133 claim_txg); 134 135 zil_free(zilog); 136 } 137 138 typedef enum resume_skip { 139 RESUME_SKIP_ALL, 140 RESUME_SKIP_NONE, 141 RESUME_SKIP_CHILDREN 142 } resume_skip_t; 143 144 /* 145 * Returns RESUME_SKIP_ALL if td indicates that we are resuming a traversal and 146 * the block indicated by zb does not need to be visited at all. Returns 147 * RESUME_SKIP_CHILDREN if we are resuming a post traversal and we reach the 148 * resume point. This indicates that this block should be visited but not its 149 * children (since they must have been visited in a previous traversal). 150 * Otherwise returns RESUME_SKIP_NONE. 151 */ 152 static resume_skip_t 153 resume_skip_check(traverse_data_t *td, const dnode_phys_t *dnp, 154 const zbookmark_phys_t *zb) 155 { 156 if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume)) { 157 /* 158 * If we already visited this bp & everything below, 159 * don't bother doing it again. 160 */ 161 if (zbookmark_subtree_completed(dnp, zb, td->td_resume)) 162 return (RESUME_SKIP_ALL); 163 164 /* 165 * If we found the block we're trying to resume from, zero 166 * the bookmark out to indicate that we have resumed. 167 */ 168 if (bcmp(zb, td->td_resume, sizeof (*zb)) == 0) { 169 bzero(td->td_resume, sizeof (*zb)); 170 if (td->td_flags & TRAVERSE_POST) 171 return (RESUME_SKIP_CHILDREN); 172 } 173 } 174 return (RESUME_SKIP_NONE); 175 } 176 177 static void 178 traverse_prefetch_metadata(traverse_data_t *td, 179 const blkptr_t *bp, const zbookmark_phys_t *zb) 180 { 181 arc_flags_t flags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH; 182 183 if (!(td->td_flags & TRAVERSE_PREFETCH_METADATA)) 184 return; 185 /* 186 * If we are in the process of resuming, don't prefetch, because 187 * some children will not be needed (and in fact may have already 188 * been freed). 189 */ 190 if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume)) 191 return; 192 if (BP_IS_HOLE(bp) || bp->blk_birth <= td->td_min_txg) 193 return; 194 if (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_DNODE) 195 return; 196 197 (void) arc_read(NULL, td->td_spa, bp, NULL, NULL, 198 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); 199 } 200 201 static boolean_t 202 prefetch_needed(prefetch_data_t *pfd, const blkptr_t *bp) 203 { 204 ASSERT(pfd->pd_flags & TRAVERSE_PREFETCH_DATA); 205 if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp) || 206 BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG) 207 return (B_FALSE); 208 return (B_TRUE); 209 } 210 211 static int 212 traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp, 213 const blkptr_t *bp, const zbookmark_phys_t *zb) 214 { 215 zbookmark_phys_t czb; 216 int err = 0; 217 arc_buf_t *buf = NULL; 218 prefetch_data_t *pd = td->td_pfd; 219 boolean_t hard = td->td_flags & TRAVERSE_HARD; 220 221 switch (resume_skip_check(td, dnp, zb)) { 222 case RESUME_SKIP_ALL: 223 return (0); 224 case RESUME_SKIP_CHILDREN: 225 goto post; 226 case RESUME_SKIP_NONE: 227 break; 228 default: 229 ASSERT(0); 230 } 231 232 if (bp->blk_birth == 0) { 233 /* 234 * Since this block has a birth time of 0 it must be a 235 * hole created before the SPA_FEATURE_HOLE_BIRTH 236 * feature was enabled. If SPA_FEATURE_HOLE_BIRTH 237 * was enabled before the min_txg for this traveral we 238 * know the hole must have been created before the 239 * min_txg for this traveral, so we can skip it. If 240 * SPA_FEATURE_HOLE_BIRTH was enabled after the min_txg 241 * for this traveral we cannot tell if the hole was 242 * created before or after the min_txg for this 243 * traversal, so we cannot skip it. 244 */ 245 if (td->td_hole_birth_enabled_txg < td->td_min_txg) 246 return (0); 247 } else if (bp->blk_birth <= td->td_min_txg) { 248 return (0); 249 } 250 251 if (pd != NULL && !pd->pd_exited && prefetch_needed(pd, bp)) { 252 uint64_t size = BP_GET_LSIZE(bp); 253 mutex_enter(&pd->pd_mtx); 254 ASSERT(pd->pd_bytes_fetched >= 0); 255 while (pd->pd_bytes_fetched < size && !pd->pd_exited) 256 cv_wait(&pd->pd_cv, &pd->pd_mtx); 257 pd->pd_bytes_fetched -= size; 258 cv_broadcast(&pd->pd_cv); 259 mutex_exit(&pd->pd_mtx); 260 } 261 262 if (BP_IS_HOLE(bp)) { 263 err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg); 264 if (err != 0) 265 goto post; 266 return (0); 267 } 268 269 if (td->td_flags & TRAVERSE_PRE) { 270 err = td->td_func(td->td_spa, NULL, bp, zb, dnp, 271 td->td_arg); 272 if (err == TRAVERSE_VISIT_NO_CHILDREN) 273 return (0); 274 if (err != 0) 275 goto post; 276 } 277 278 if (BP_GET_LEVEL(bp) > 0) { 279 arc_flags_t flags = ARC_FLAG_WAIT; 280 int i; 281 blkptr_t *cbp; 282 int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT; 283 284 err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf, 285 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); 286 if (err != 0) 287 goto post; 288 cbp = buf->b_data; 289 290 for (i = 0; i < epb; i++) { 291 SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object, 292 zb->zb_level - 1, 293 zb->zb_blkid * epb + i); 294 traverse_prefetch_metadata(td, &cbp[i], &czb); 295 } 296 297 /* recursively visitbp() blocks below this */ 298 for (i = 0; i < epb; i++) { 299 SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object, 300 zb->zb_level - 1, 301 zb->zb_blkid * epb + i); 302 err = traverse_visitbp(td, dnp, &cbp[i], &czb); 303 if (err != 0) 304 break; 305 } 306 } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) { 307 arc_flags_t flags = ARC_FLAG_WAIT; 308 int i; 309 int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT; 310 311 err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf, 312 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); 313 if (err != 0) 314 goto post; 315 dnode_phys_t *child_dnp = buf->b_data; 316 317 for (i = 0; i < epb; i++) { 318 prefetch_dnode_metadata(td, &child_dnp[i], 319 zb->zb_objset, zb->zb_blkid * epb + i); 320 } 321 322 /* recursively visitbp() blocks below this */ 323 for (i = 0; i < epb; i++) { 324 err = traverse_dnode(td, &child_dnp[i], 325 zb->zb_objset, zb->zb_blkid * epb + i); 326 if (err != 0) 327 break; 328 } 329 } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) { 330 arc_flags_t flags = ARC_FLAG_WAIT; 331 332 err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf, 333 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); 334 if (err != 0) 335 goto post; 336 337 objset_phys_t *osp = buf->b_data; 338 prefetch_dnode_metadata(td, &osp->os_meta_dnode, zb->zb_objset, 339 DMU_META_DNODE_OBJECT); 340 if (arc_buf_size(buf) >= sizeof (objset_phys_t)) { 341 prefetch_dnode_metadata(td, &osp->os_groupused_dnode, 342 zb->zb_objset, DMU_GROUPUSED_OBJECT); 343 prefetch_dnode_metadata(td, &osp->os_userused_dnode, 344 zb->zb_objset, DMU_USERUSED_OBJECT); 345 } 346 347 err = traverse_dnode(td, &osp->os_meta_dnode, zb->zb_objset, 348 DMU_META_DNODE_OBJECT); 349 if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) { 350 err = traverse_dnode(td, &osp->os_groupused_dnode, 351 zb->zb_objset, DMU_GROUPUSED_OBJECT); 352 } 353 if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) { 354 err = traverse_dnode(td, &osp->os_userused_dnode, 355 zb->zb_objset, DMU_USERUSED_OBJECT); 356 } 357 } 358 359 if (buf) 360 (void) arc_buf_remove_ref(buf, &buf); 361 362 post: 363 if (err == 0 && (td->td_flags & TRAVERSE_POST)) 364 err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg); 365 366 if (hard && (err == EIO || err == ECKSUM)) { 367 /* 368 * Ignore this disk error as requested by the HARD flag, 369 * and continue traversal. 370 */ 371 err = 0; 372 } 373 374 /* 375 * If we are stopping here, set td_resume. 376 */ 377 if (td->td_resume != NULL && err != 0 && !td->td_paused) { 378 td->td_resume->zb_objset = zb->zb_objset; 379 td->td_resume->zb_object = zb->zb_object; 380 td->td_resume->zb_level = 0; 381 /* 382 * If we have stopped on an indirect block (e.g. due to 383 * i/o error), we have not visited anything below it. 384 * Set the bookmark to the first level-0 block that we need 385 * to visit. This way, the resuming code does not need to 386 * deal with resuming from indirect blocks. 387 * 388 * Note, if zb_level <= 0, dnp may be NULL, so we don't want 389 * to dereference it. 390 */ 391 td->td_resume->zb_blkid = zb->zb_blkid; 392 if (zb->zb_level > 0) { 393 td->td_resume->zb_blkid <<= zb->zb_level * 394 (dnp->dn_indblkshift - SPA_BLKPTRSHIFT); 395 } 396 td->td_paused = B_TRUE; 397 } 398 399 return (err); 400 } 401 402 static void 403 prefetch_dnode_metadata(traverse_data_t *td, const dnode_phys_t *dnp, 404 uint64_t objset, uint64_t object) 405 { 406 int j; 407 zbookmark_phys_t czb; 408 409 for (j = 0; j < dnp->dn_nblkptr; j++) { 410 SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j); 411 traverse_prefetch_metadata(td, &dnp->dn_blkptr[j], &czb); 412 } 413 414 if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) { 415 SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID); 416 traverse_prefetch_metadata(td, &dnp->dn_spill, &czb); 417 } 418 } 419 420 static int 421 traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp, 422 uint64_t objset, uint64_t object) 423 { 424 int j, err = 0; 425 zbookmark_phys_t czb; 426 427 if (object != DMU_META_DNODE_OBJECT && td->td_resume != NULL && 428 object < td->td_resume->zb_object) 429 return (0); 430 431 if (td->td_flags & TRAVERSE_PRE) { 432 SET_BOOKMARK(&czb, objset, object, ZB_DNODE_LEVEL, 433 ZB_DNODE_BLKID); 434 err = td->td_func(td->td_spa, NULL, NULL, &czb, dnp, 435 td->td_arg); 436 if (err == TRAVERSE_VISIT_NO_CHILDREN) 437 return (0); 438 if (err != 0) 439 return (err); 440 } 441 442 for (j = 0; j < dnp->dn_nblkptr; j++) { 443 SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j); 444 err = traverse_visitbp(td, dnp, &dnp->dn_blkptr[j], &czb); 445 if (err != 0) 446 break; 447 } 448 449 if (err == 0 && (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) { 450 SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID); 451 err = traverse_visitbp(td, dnp, &dnp->dn_spill, &czb); 452 } 453 454 if (err == 0 && (td->td_flags & TRAVERSE_POST)) { 455 SET_BOOKMARK(&czb, objset, object, ZB_DNODE_LEVEL, 456 ZB_DNODE_BLKID); 457 err = td->td_func(td->td_spa, NULL, NULL, &czb, dnp, 458 td->td_arg); 459 if (err == TRAVERSE_VISIT_NO_CHILDREN) 460 return (0); 461 if (err != 0) 462 return (err); 463 } 464 return (err); 465 } 466 467 /* ARGSUSED */ 468 static int 469 traverse_prefetcher(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 470 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 471 { 472 prefetch_data_t *pfd = arg; 473 arc_flags_t aflags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH; 474 475 ASSERT(pfd->pd_bytes_fetched >= 0); 476 if (bp == NULL) 477 return (0); 478 if (pfd->pd_cancel) 479 return (SET_ERROR(EINTR)); 480 481 if (!prefetch_needed(pfd, bp)) 482 return (0); 483 484 mutex_enter(&pfd->pd_mtx); 485 while (!pfd->pd_cancel && pfd->pd_bytes_fetched >= zfs_pd_bytes_max) 486 cv_wait(&pfd->pd_cv, &pfd->pd_mtx); 487 pfd->pd_bytes_fetched += BP_GET_LSIZE(bp); 488 cv_broadcast(&pfd->pd_cv); 489 mutex_exit(&pfd->pd_mtx); 490 491 (void) arc_read(NULL, spa, bp, NULL, NULL, ZIO_PRIORITY_ASYNC_READ, 492 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &aflags, zb); 493 494 return (0); 495 } 496 497 static void 498 traverse_prefetch_thread(void *arg) 499 { 500 traverse_data_t *td_main = arg; 501 traverse_data_t td = *td_main; 502 zbookmark_phys_t czb; 503 504 td.td_func = traverse_prefetcher; 505 td.td_arg = td_main->td_pfd; 506 td.td_pfd = NULL; 507 td.td_resume = &td_main->td_pfd->pd_resume; 508 509 SET_BOOKMARK(&czb, td.td_objset, 510 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 511 (void) traverse_visitbp(&td, NULL, td.td_rootbp, &czb); 512 513 mutex_enter(&td_main->td_pfd->pd_mtx); 514 td_main->td_pfd->pd_exited = B_TRUE; 515 cv_broadcast(&td_main->td_pfd->pd_cv); 516 mutex_exit(&td_main->td_pfd->pd_mtx); 517 } 518 519 /* 520 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are 521 * in syncing context). 522 */ 523 static int 524 traverse_impl(spa_t *spa, dsl_dataset_t *ds, uint64_t objset, blkptr_t *rootbp, 525 uint64_t txg_start, zbookmark_phys_t *resume, int flags, 526 blkptr_cb_t func, void *arg) 527 { 528 traverse_data_t td; 529 prefetch_data_t pd = { 0 }; 530 zbookmark_phys_t czb; 531 int err; 532 533 ASSERT(ds == NULL || objset == ds->ds_object); 534 ASSERT(!(flags & TRAVERSE_PRE) || !(flags & TRAVERSE_POST)); 535 536 td.td_spa = spa; 537 td.td_objset = objset; 538 td.td_rootbp = rootbp; 539 td.td_min_txg = txg_start; 540 td.td_resume = resume; 541 td.td_func = func; 542 td.td_arg = arg; 543 td.td_pfd = &pd; 544 td.td_flags = flags; 545 td.td_paused = B_FALSE; 546 547 if (spa_feature_is_active(spa, SPA_FEATURE_HOLE_BIRTH)) { 548 VERIFY(spa_feature_enabled_txg(spa, 549 SPA_FEATURE_HOLE_BIRTH, &td.td_hole_birth_enabled_txg)); 550 } else { 551 td.td_hole_birth_enabled_txg = 0; 552 } 553 554 pd.pd_flags = flags; 555 if (resume != NULL) 556 pd.pd_resume = *resume; 557 mutex_init(&pd.pd_mtx, NULL, MUTEX_DEFAULT, NULL); 558 cv_init(&pd.pd_cv, NULL, CV_DEFAULT, NULL); 559 560 /* See comment on ZIL traversal in dsl_scan_visitds. */ 561 if (ds != NULL && !ds->ds_is_snapshot && !BP_IS_HOLE(rootbp)) { 562 arc_flags_t flags = ARC_FLAG_WAIT; 563 objset_phys_t *osp; 564 arc_buf_t *buf; 565 566 err = arc_read(NULL, td.td_spa, rootbp, 567 arc_getbuf_func, &buf, 568 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, NULL); 569 if (err != 0) 570 return (err); 571 572 osp = buf->b_data; 573 traverse_zil(&td, &osp->os_zil_header); 574 (void) arc_buf_remove_ref(buf, &buf); 575 } 576 577 if (!(flags & TRAVERSE_PREFETCH_DATA) || 578 0 == taskq_dispatch(system_taskq, traverse_prefetch_thread, 579 &td, TQ_NOQUEUE)) 580 pd.pd_exited = B_TRUE; 581 582 SET_BOOKMARK(&czb, td.td_objset, 583 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 584 err = traverse_visitbp(&td, NULL, rootbp, &czb); 585 586 mutex_enter(&pd.pd_mtx); 587 pd.pd_cancel = B_TRUE; 588 cv_broadcast(&pd.pd_cv); 589 while (!pd.pd_exited) 590 cv_wait(&pd.pd_cv, &pd.pd_mtx); 591 mutex_exit(&pd.pd_mtx); 592 593 mutex_destroy(&pd.pd_mtx); 594 cv_destroy(&pd.pd_cv); 595 596 return (err); 597 } 598 599 /* 600 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are 601 * in syncing context). 602 */ 603 int 604 traverse_dataset_resume(dsl_dataset_t *ds, uint64_t txg_start, 605 zbookmark_phys_t *resume, 606 int flags, blkptr_cb_t func, void *arg) 607 { 608 return (traverse_impl(ds->ds_dir->dd_pool->dp_spa, ds, ds->ds_object, 609 &dsl_dataset_phys(ds)->ds_bp, txg_start, resume, flags, func, arg)); 610 } 611 612 int 613 traverse_dataset(dsl_dataset_t *ds, uint64_t txg_start, 614 int flags, blkptr_cb_t func, void *arg) 615 { 616 return (traverse_dataset_resume(ds, txg_start, NULL, flags, func, arg)); 617 } 618 619 int 620 traverse_dataset_destroyed(spa_t *spa, blkptr_t *blkptr, 621 uint64_t txg_start, zbookmark_phys_t *resume, int flags, 622 blkptr_cb_t func, void *arg) 623 { 624 return (traverse_impl(spa, NULL, ZB_DESTROYED_OBJSET, 625 blkptr, txg_start, resume, flags, func, arg)); 626 } 627 628 /* 629 * NB: pool must not be changing on-disk (eg, from zdb or sync context). 630 */ 631 int 632 traverse_pool(spa_t *spa, uint64_t txg_start, int flags, 633 blkptr_cb_t func, void *arg) 634 { 635 int err; 636 dsl_pool_t *dp = spa_get_dsl(spa); 637 objset_t *mos = dp->dp_meta_objset; 638 boolean_t hard = (flags & TRAVERSE_HARD); 639 640 /* visit the MOS */ 641 err = traverse_impl(spa, NULL, 0, spa_get_rootblkptr(spa), 642 txg_start, NULL, flags, func, arg); 643 if (err != 0) 644 return (err); 645 646 /* visit each dataset */ 647 for (uint64_t obj = 1; err == 0; 648 err = dmu_object_next(mos, &obj, B_FALSE, txg_start)) { 649 dmu_object_info_t doi; 650 651 err = dmu_object_info(mos, obj, &doi); 652 if (err != 0) { 653 if (hard) 654 continue; 655 break; 656 } 657 658 if (doi.doi_bonus_type == DMU_OT_DSL_DATASET) { 659 dsl_dataset_t *ds; 660 uint64_t txg = txg_start; 661 662 dsl_pool_config_enter(dp, FTAG); 663 err = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); 664 dsl_pool_config_exit(dp, FTAG); 665 if (err != 0) { 666 if (hard) 667 continue; 668 break; 669 } 670 if (dsl_dataset_phys(ds)->ds_prev_snap_txg > txg) 671 txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; 672 err = traverse_dataset(ds, txg, flags, func, arg); 673 dsl_dataset_rele(ds, FTAG); 674 if (err != 0) 675 break; 676 } 677 } 678 if (err == ESRCH) 679 err = 0; 680 return (err); 681 } 682