dmu_traverse.c revision 307266
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, 2016 by Delphix. All rights reserved. 24 * Copyright (c) 2015 Chunwei Chen. All rights reserved. 25 */ 26 27#include <sys/zfs_context.h> 28#include <sys/dmu_objset.h> 29#include <sys/dmu_traverse.h> 30#include <sys/dsl_dataset.h> 31#include <sys/dsl_dir.h> 32#include <sys/dsl_pool.h> 33#include <sys/dnode.h> 34#include <sys/spa.h> 35#include <sys/zio.h> 36#include <sys/dmu_impl.h> 37#include <sys/sa.h> 38#include <sys/sa_impl.h> 39#include <sys/callb.h> 40#include <sys/zfeature.h> 41 42int32_t zfs_pd_bytes_max = 50 * 1024 * 1024; /* 50MB */ 43 44typedef struct prefetch_data { 45 kmutex_t pd_mtx; 46 kcondvar_t pd_cv; 47 int32_t pd_bytes_fetched; 48 int pd_flags; 49 boolean_t pd_cancel; 50 boolean_t pd_exited; 51 zbookmark_phys_t pd_resume; 52} prefetch_data_t; 53 54typedef struct traverse_data { 55 spa_t *td_spa; 56 uint64_t td_objset; 57 blkptr_t *td_rootbp; 58 uint64_t td_min_txg; 59 zbookmark_phys_t *td_resume; 60 int td_flags; 61 prefetch_data_t *td_pfd; 62 boolean_t td_paused; 63 uint64_t td_hole_birth_enabled_txg; 64 blkptr_cb_t *td_func; 65 void *td_arg; 66 boolean_t td_realloc_possible; 67} traverse_data_t; 68 69static int traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp, 70 uint64_t objset, uint64_t object); 71static void prefetch_dnode_metadata(traverse_data_t *td, const dnode_phys_t *, 72 uint64_t objset, uint64_t object); 73 74static int 75traverse_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg) 76{ 77 traverse_data_t *td = arg; 78 zbookmark_phys_t zb; 79 80 if (BP_IS_HOLE(bp)) 81 return (0); 82 83 if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(td->td_spa)) 84 return (0); 85 86 SET_BOOKMARK(&zb, td->td_objset, ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, 87 bp->blk_cksum.zc_word[ZIL_ZC_SEQ]); 88 89 (void) td->td_func(td->td_spa, zilog, bp, &zb, NULL, td->td_arg); 90 91 return (0); 92} 93 94static int 95traverse_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg) 96{ 97 traverse_data_t *td = arg; 98 99 if (lrc->lrc_txtype == TX_WRITE) { 100 lr_write_t *lr = (lr_write_t *)lrc; 101 blkptr_t *bp = &lr->lr_blkptr; 102 zbookmark_phys_t zb; 103 104 if (BP_IS_HOLE(bp)) 105 return (0); 106 107 if (claim_txg == 0 || bp->blk_birth < claim_txg) 108 return (0); 109 110 SET_BOOKMARK(&zb, td->td_objset, lr->lr_foid, 111 ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp)); 112 113 (void) td->td_func(td->td_spa, zilog, bp, &zb, NULL, 114 td->td_arg); 115 } 116 return (0); 117} 118 119static void 120traverse_zil(traverse_data_t *td, zil_header_t *zh) 121{ 122 uint64_t claim_txg = zh->zh_claim_txg; 123 zilog_t *zilog; 124 125 /* 126 * We only want to visit blocks that have been claimed but not yet 127 * replayed; plus, in read-only mode, blocks that are already stable. 128 */ 129 if (claim_txg == 0 && spa_writeable(td->td_spa)) 130 return; 131 132 zilog = zil_alloc(spa_get_dsl(td->td_spa)->dp_meta_objset, zh); 133 134 (void) zil_parse(zilog, traverse_zil_block, traverse_zil_record, td, 135 claim_txg); 136 137 zil_free(zilog); 138} 139 140typedef enum resume_skip { 141 RESUME_SKIP_ALL, 142 RESUME_SKIP_NONE, 143 RESUME_SKIP_CHILDREN 144} resume_skip_t; 145 146/* 147 * Returns RESUME_SKIP_ALL if td indicates that we are resuming a traversal and 148 * the block indicated by zb does not need to be visited at all. Returns 149 * RESUME_SKIP_CHILDREN if we are resuming a post traversal and we reach the 150 * resume point. This indicates that this block should be visited but not its 151 * children (since they must have been visited in a previous traversal). 152 * Otherwise returns RESUME_SKIP_NONE. 153 */ 154static resume_skip_t 155resume_skip_check(traverse_data_t *td, const dnode_phys_t *dnp, 156 const zbookmark_phys_t *zb) 157{ 158 if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume)) { 159 /* 160 * If we already visited this bp & everything below, 161 * don't bother doing it again. 162 */ 163 if (zbookmark_subtree_completed(dnp, zb, td->td_resume)) 164 return (RESUME_SKIP_ALL); 165 166 /* 167 * If we found the block we're trying to resume from, zero 168 * the bookmark out to indicate that we have resumed. 169 */ 170 if (bcmp(zb, td->td_resume, sizeof (*zb)) == 0) { 171 bzero(td->td_resume, sizeof (*zb)); 172 if (td->td_flags & TRAVERSE_POST) 173 return (RESUME_SKIP_CHILDREN); 174 } 175 } 176 return (RESUME_SKIP_NONE); 177} 178 179static void 180traverse_prefetch_metadata(traverse_data_t *td, 181 const blkptr_t *bp, const zbookmark_phys_t *zb) 182{ 183 arc_flags_t flags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH; 184 185 if (!(td->td_flags & TRAVERSE_PREFETCH_METADATA)) 186 return; 187 /* 188 * If we are in the process of resuming, don't prefetch, because 189 * some children will not be needed (and in fact may have already 190 * been freed). 191 */ 192 if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume)) 193 return; 194 if (BP_IS_HOLE(bp) || bp->blk_birth <= td->td_min_txg) 195 return; 196 if (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_DNODE) 197 return; 198 199 (void) arc_read(NULL, td->td_spa, bp, NULL, NULL, 200 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); 201} 202 203static boolean_t 204prefetch_needed(prefetch_data_t *pfd, const blkptr_t *bp) 205{ 206 ASSERT(pfd->pd_flags & TRAVERSE_PREFETCH_DATA); 207 if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp) || 208 BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG) 209 return (B_FALSE); 210 return (B_TRUE); 211} 212 213static int 214traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp, 215 const blkptr_t *bp, const zbookmark_phys_t *zb) 216{ 217 zbookmark_phys_t czb; 218 int err = 0; 219 arc_buf_t *buf = NULL; 220 prefetch_data_t *pd = td->td_pfd; 221 boolean_t hard = td->td_flags & TRAVERSE_HARD; 222 223 switch (resume_skip_check(td, dnp, zb)) { 224 case RESUME_SKIP_ALL: 225 return (0); 226 case RESUME_SKIP_CHILDREN: 227 goto post; 228 case RESUME_SKIP_NONE: 229 break; 230 default: 231 ASSERT(0); 232 } 233 234 if (bp->blk_birth == 0) { 235 /* 236 * Since this block has a birth time of 0 it must be one of 237 * two things: a hole created before the 238 * SPA_FEATURE_HOLE_BIRTH feature was enabled, or a hole 239 * which has always been a hole in an object. 240 * 241 * If a file is written sparsely, then the unwritten parts of 242 * the file were "always holes" -- that is, they have been 243 * holes since this object was allocated. However, we (and 244 * our callers) can not necessarily tell when an object was 245 * allocated. Therefore, if it's possible that this object 246 * was freed and then its object number reused, we need to 247 * visit all the holes with birth==0. 248 * 249 * If it isn't possible that the object number was reused, 250 * then if SPA_FEATURE_HOLE_BIRTH was enabled before we wrote 251 * all the blocks we will visit as part of this traversal, 252 * then this hole must have always existed, so we can skip 253 * it. We visit blocks born after (exclusive) td_min_txg. 254 * 255 * Note that the meta-dnode cannot be reallocated. 256 */ 257 if ((!td->td_realloc_possible || 258 zb->zb_object == DMU_META_DNODE_OBJECT) && 259 td->td_hole_birth_enabled_txg <= td->td_min_txg) 260 return (0); 261 } else if (bp->blk_birth <= td->td_min_txg) { 262 return (0); 263 } 264 265 if (pd != NULL && !pd->pd_exited && prefetch_needed(pd, bp)) { 266 uint64_t size = BP_GET_LSIZE(bp); 267 mutex_enter(&pd->pd_mtx); 268 ASSERT(pd->pd_bytes_fetched >= 0); 269 while (pd->pd_bytes_fetched < size && !pd->pd_exited) 270 cv_wait(&pd->pd_cv, &pd->pd_mtx); 271 pd->pd_bytes_fetched -= size; 272 cv_broadcast(&pd->pd_cv); 273 mutex_exit(&pd->pd_mtx); 274 } 275 276 if (BP_IS_HOLE(bp)) { 277 err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg); 278 if (err != 0) 279 goto post; 280 return (0); 281 } 282 283 if (td->td_flags & TRAVERSE_PRE) { 284 err = td->td_func(td->td_spa, NULL, bp, zb, dnp, 285 td->td_arg); 286 if (err == TRAVERSE_VISIT_NO_CHILDREN) 287 return (0); 288 if (err != 0) 289 goto post; 290 } 291 292 if (BP_GET_LEVEL(bp) > 0) { 293 arc_flags_t flags = ARC_FLAG_WAIT; 294 int i; 295 blkptr_t *cbp; 296 int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT; 297 298 err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf, 299 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); 300 if (err != 0) 301 goto post; 302 cbp = buf->b_data; 303 304 for (i = 0; i < epb; i++) { 305 SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object, 306 zb->zb_level - 1, 307 zb->zb_blkid * epb + i); 308 traverse_prefetch_metadata(td, &cbp[i], &czb); 309 } 310 311 /* recursively visitbp() blocks below this */ 312 for (i = 0; i < epb; i++) { 313 SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object, 314 zb->zb_level - 1, 315 zb->zb_blkid * epb + i); 316 err = traverse_visitbp(td, dnp, &cbp[i], &czb); 317 if (err != 0) 318 break; 319 } 320 } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) { 321 arc_flags_t flags = ARC_FLAG_WAIT; 322 int i; 323 int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT; 324 325 err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf, 326 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); 327 if (err != 0) 328 goto post; 329 dnode_phys_t *child_dnp = buf->b_data; 330 331 for (i = 0; i < epb; i++) { 332 prefetch_dnode_metadata(td, &child_dnp[i], 333 zb->zb_objset, zb->zb_blkid * epb + i); 334 } 335 336 /* recursively visitbp() blocks below this */ 337 for (i = 0; i < epb; i++) { 338 err = traverse_dnode(td, &child_dnp[i], 339 zb->zb_objset, zb->zb_blkid * epb + i); 340 if (err != 0) 341 break; 342 } 343 } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) { 344 arc_flags_t flags = ARC_FLAG_WAIT; 345 346 err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf, 347 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); 348 if (err != 0) 349 goto post; 350 351 objset_phys_t *osp = buf->b_data; 352 prefetch_dnode_metadata(td, &osp->os_meta_dnode, zb->zb_objset, 353 DMU_META_DNODE_OBJECT); 354 /* 355 * See the block comment above for the goal of this variable. 356 * If the maxblkid of the meta-dnode is 0, then we know that 357 * we've never had more than DNODES_PER_BLOCK objects in the 358 * dataset, which means we can't have reused any object ids. 359 */ 360 if (osp->os_meta_dnode.dn_maxblkid == 0) 361 td->td_realloc_possible = B_FALSE; 362 363 if (arc_buf_size(buf) >= sizeof (objset_phys_t)) { 364 prefetch_dnode_metadata(td, &osp->os_groupused_dnode, 365 zb->zb_objset, DMU_GROUPUSED_OBJECT); 366 prefetch_dnode_metadata(td, &osp->os_userused_dnode, 367 zb->zb_objset, DMU_USERUSED_OBJECT); 368 } 369 370 err = traverse_dnode(td, &osp->os_meta_dnode, zb->zb_objset, 371 DMU_META_DNODE_OBJECT); 372 if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) { 373 err = traverse_dnode(td, &osp->os_groupused_dnode, 374 zb->zb_objset, DMU_GROUPUSED_OBJECT); 375 } 376 if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) { 377 err = traverse_dnode(td, &osp->os_userused_dnode, 378 zb->zb_objset, DMU_USERUSED_OBJECT); 379 } 380 } 381 382 if (buf) 383 arc_buf_destroy(buf, &buf); 384 385post: 386 if (err == 0 && (td->td_flags & TRAVERSE_POST)) 387 err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg); 388 389 if (hard && (err == EIO || err == ECKSUM)) { 390 /* 391 * Ignore this disk error as requested by the HARD flag, 392 * and continue traversal. 393 */ 394 err = 0; 395 } 396 397 /* 398 * If we are stopping here, set td_resume. 399 */ 400 if (td->td_resume != NULL && err != 0 && !td->td_paused) { 401 td->td_resume->zb_objset = zb->zb_objset; 402 td->td_resume->zb_object = zb->zb_object; 403 td->td_resume->zb_level = 0; 404 /* 405 * If we have stopped on an indirect block (e.g. due to 406 * i/o error), we have not visited anything below it. 407 * Set the bookmark to the first level-0 block that we need 408 * to visit. This way, the resuming code does not need to 409 * deal with resuming from indirect blocks. 410 * 411 * Note, if zb_level <= 0, dnp may be NULL, so we don't want 412 * to dereference it. 413 */ 414 td->td_resume->zb_blkid = zb->zb_blkid; 415 if (zb->zb_level > 0) { 416 td->td_resume->zb_blkid <<= zb->zb_level * 417 (dnp->dn_indblkshift - SPA_BLKPTRSHIFT); 418 } 419 td->td_paused = B_TRUE; 420 } 421 422 return (err); 423} 424 425static void 426prefetch_dnode_metadata(traverse_data_t *td, const dnode_phys_t *dnp, 427 uint64_t objset, uint64_t object) 428{ 429 int j; 430 zbookmark_phys_t czb; 431 432 for (j = 0; j < dnp->dn_nblkptr; j++) { 433 SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j); 434 traverse_prefetch_metadata(td, &dnp->dn_blkptr[j], &czb); 435 } 436 437 if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) { 438 SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID); 439 traverse_prefetch_metadata(td, &dnp->dn_spill, &czb); 440 } 441} 442 443static int 444traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp, 445 uint64_t objset, uint64_t object) 446{ 447 int j, err = 0; 448 zbookmark_phys_t czb; 449 450 if (object != DMU_META_DNODE_OBJECT && td->td_resume != NULL && 451 object < td->td_resume->zb_object) 452 return (0); 453 454 if (td->td_flags & TRAVERSE_PRE) { 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 } 464 465 for (j = 0; j < dnp->dn_nblkptr; j++) { 466 SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j); 467 err = traverse_visitbp(td, dnp, &dnp->dn_blkptr[j], &czb); 468 if (err != 0) 469 break; 470 } 471 472 if (err == 0 && (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) { 473 SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID); 474 err = traverse_visitbp(td, dnp, &dnp->dn_spill, &czb); 475 } 476 477 if (err == 0 && (td->td_flags & TRAVERSE_POST)) { 478 SET_BOOKMARK(&czb, objset, object, ZB_DNODE_LEVEL, 479 ZB_DNODE_BLKID); 480 err = td->td_func(td->td_spa, NULL, NULL, &czb, dnp, 481 td->td_arg); 482 if (err == TRAVERSE_VISIT_NO_CHILDREN) 483 return (0); 484 if (err != 0) 485 return (err); 486 } 487 return (err); 488} 489 490/* ARGSUSED */ 491static int 492traverse_prefetcher(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 493 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 494{ 495 prefetch_data_t *pfd = arg; 496 arc_flags_t aflags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH; 497 498 ASSERT(pfd->pd_bytes_fetched >= 0); 499 if (bp == NULL) 500 return (0); 501 if (pfd->pd_cancel) 502 return (SET_ERROR(EINTR)); 503 504 if (!prefetch_needed(pfd, bp)) 505 return (0); 506 507 mutex_enter(&pfd->pd_mtx); 508 while (!pfd->pd_cancel && pfd->pd_bytes_fetched >= zfs_pd_bytes_max) 509 cv_wait(&pfd->pd_cv, &pfd->pd_mtx); 510 pfd->pd_bytes_fetched += BP_GET_LSIZE(bp); 511 cv_broadcast(&pfd->pd_cv); 512 mutex_exit(&pfd->pd_mtx); 513 514 (void) arc_read(NULL, spa, bp, NULL, NULL, ZIO_PRIORITY_ASYNC_READ, 515 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &aflags, zb); 516 517 return (0); 518} 519 520static void 521traverse_prefetch_thread(void *arg) 522{ 523 traverse_data_t *td_main = arg; 524 traverse_data_t td = *td_main; 525 zbookmark_phys_t czb; 526 527 td.td_func = traverse_prefetcher; 528 td.td_arg = td_main->td_pfd; 529 td.td_pfd = NULL; 530 td.td_resume = &td_main->td_pfd->pd_resume; 531 532 SET_BOOKMARK(&czb, td.td_objset, 533 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 534 (void) traverse_visitbp(&td, NULL, td.td_rootbp, &czb); 535 536 mutex_enter(&td_main->td_pfd->pd_mtx); 537 td_main->td_pfd->pd_exited = B_TRUE; 538 cv_broadcast(&td_main->td_pfd->pd_cv); 539 mutex_exit(&td_main->td_pfd->pd_mtx); 540} 541 542/* 543 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are 544 * in syncing context). 545 */ 546static int 547traverse_impl(spa_t *spa, dsl_dataset_t *ds, uint64_t objset, blkptr_t *rootbp, 548 uint64_t txg_start, zbookmark_phys_t *resume, int flags, 549 blkptr_cb_t func, void *arg) 550{ 551 traverse_data_t td; 552 prefetch_data_t pd = { 0 }; 553 zbookmark_phys_t czb; 554 int err; 555 556 ASSERT(ds == NULL || objset == ds->ds_object); 557 ASSERT(!(flags & TRAVERSE_PRE) || !(flags & TRAVERSE_POST)); 558 559 td.td_spa = spa; 560 td.td_objset = objset; 561 td.td_rootbp = rootbp; 562 td.td_min_txg = txg_start; 563 td.td_resume = resume; 564 td.td_func = func; 565 td.td_arg = arg; 566 td.td_pfd = &pd; 567 td.td_flags = flags; 568 td.td_paused = B_FALSE; 569 td.td_realloc_possible = (txg_start == 0 ? B_FALSE : B_TRUE); 570 571 if (spa_feature_is_active(spa, SPA_FEATURE_HOLE_BIRTH)) { 572 VERIFY(spa_feature_enabled_txg(spa, 573 SPA_FEATURE_HOLE_BIRTH, &td.td_hole_birth_enabled_txg)); 574 } else { 575 td.td_hole_birth_enabled_txg = UINT64_MAX; 576 } 577 578 pd.pd_flags = flags; 579 if (resume != NULL) 580 pd.pd_resume = *resume; 581 mutex_init(&pd.pd_mtx, NULL, MUTEX_DEFAULT, NULL); 582 cv_init(&pd.pd_cv, NULL, CV_DEFAULT, NULL); 583 584 /* See comment on ZIL traversal in dsl_scan_visitds. */ 585 if (ds != NULL && !ds->ds_is_snapshot && !BP_IS_HOLE(rootbp)) { 586 arc_flags_t flags = ARC_FLAG_WAIT; 587 objset_phys_t *osp; 588 arc_buf_t *buf; 589 590 err = arc_read(NULL, td.td_spa, rootbp, 591 arc_getbuf_func, &buf, 592 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, NULL); 593 if (err != 0) 594 return (err); 595 596 osp = buf->b_data; 597 traverse_zil(&td, &osp->os_zil_header); 598 arc_buf_destroy(buf, &buf); 599 } 600 601 if (!(flags & TRAVERSE_PREFETCH_DATA) || 602 0 == taskq_dispatch(system_taskq, traverse_prefetch_thread, 603 &td, TQ_NOQUEUE)) 604 pd.pd_exited = B_TRUE; 605 606 SET_BOOKMARK(&czb, td.td_objset, 607 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 608 err = traverse_visitbp(&td, NULL, rootbp, &czb); 609 610 mutex_enter(&pd.pd_mtx); 611 pd.pd_cancel = B_TRUE; 612 cv_broadcast(&pd.pd_cv); 613 while (!pd.pd_exited) 614 cv_wait(&pd.pd_cv, &pd.pd_mtx); 615 mutex_exit(&pd.pd_mtx); 616 617 mutex_destroy(&pd.pd_mtx); 618 cv_destroy(&pd.pd_cv); 619 620 return (err); 621} 622 623/* 624 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are 625 * in syncing context). 626 */ 627int 628traverse_dataset_resume(dsl_dataset_t *ds, uint64_t txg_start, 629 zbookmark_phys_t *resume, 630 int flags, blkptr_cb_t func, void *arg) 631{ 632 return (traverse_impl(ds->ds_dir->dd_pool->dp_spa, ds, ds->ds_object, 633 &dsl_dataset_phys(ds)->ds_bp, txg_start, resume, flags, func, arg)); 634} 635 636int 637traverse_dataset(dsl_dataset_t *ds, uint64_t txg_start, 638 int flags, blkptr_cb_t func, void *arg) 639{ 640 return (traverse_dataset_resume(ds, txg_start, NULL, flags, func, arg)); 641} 642 643int 644traverse_dataset_destroyed(spa_t *spa, blkptr_t *blkptr, 645 uint64_t txg_start, zbookmark_phys_t *resume, int flags, 646 blkptr_cb_t func, void *arg) 647{ 648 return (traverse_impl(spa, NULL, ZB_DESTROYED_OBJSET, 649 blkptr, txg_start, resume, flags, func, arg)); 650} 651 652/* 653 * NB: pool must not be changing on-disk (eg, from zdb or sync context). 654 */ 655int 656traverse_pool(spa_t *spa, uint64_t txg_start, int flags, 657 blkptr_cb_t func, void *arg) 658{ 659 int err; 660 dsl_pool_t *dp = spa_get_dsl(spa); 661 objset_t *mos = dp->dp_meta_objset; 662 boolean_t hard = (flags & TRAVERSE_HARD); 663 664 /* visit the MOS */ 665 err = traverse_impl(spa, NULL, 0, spa_get_rootblkptr(spa), 666 txg_start, NULL, flags, func, arg); 667 if (err != 0) 668 return (err); 669 670 /* visit each dataset */ 671 for (uint64_t obj = 1; err == 0; 672 err = dmu_object_next(mos, &obj, B_FALSE, txg_start)) { 673 dmu_object_info_t doi; 674 675 err = dmu_object_info(mos, obj, &doi); 676 if (err != 0) { 677 if (hard) 678 continue; 679 break; 680 } 681 682 if (doi.doi_bonus_type == DMU_OT_DSL_DATASET) { 683 dsl_dataset_t *ds; 684 uint64_t txg = txg_start; 685 686 dsl_pool_config_enter(dp, FTAG); 687 err = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); 688 dsl_pool_config_exit(dp, FTAG); 689 if (err != 0) { 690 if (hard) 691 continue; 692 break; 693 } 694 if (dsl_dataset_phys(ds)->ds_prev_snap_txg > txg) 695 txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; 696 err = traverse_dataset(ds, txg, flags, func, arg); 697 dsl_dataset_rele(ds, FTAG); 698 if (err != 0) 699 break; 700 } 701 } 702 if (err == ESRCH) 703 err = 0; 704 return (err); 705} 706