module/zfs/dmu_traverse.c

   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) 2013 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 int zfs_pd_blks_max = 100;
  42
  43 typedef struct prefetch_data {
  44         kmutex_t pd_mtx;
  45         kcondvar_t pd_cv;
  46         int pd_blks_max;
  47         int pd_blks_fetched;
  48         int pd_flags;
  49         boolean_t pd_cancel;
  50         boolean_t pd_exited;
  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_t *td_resume;
  59         int td_flags;
  60         prefetch_data_t *td_pfd;
  61         blkptr_cb_t *td_func;
  62         void *td_arg;
  63 } traverse_data_t;
  64
  65 #define TD_HARD(td)     (td->td_flags & TRAVERSE_HARD)
  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_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_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_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_is_before(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                 ASSERT3U(zb->zb_object, <=, td->td_resume->zb_object);
 169                 if (bcmp(zb, td->td_resume, sizeof (*zb)) == 0) {
 170                         bzero(td->td_resume, sizeof (*zb));
 171                         if (td->td_flags & TRAVERSE_POST)
 172                                 return (RESUME_SKIP_CHILDREN);
 173                 }
 174         }
 175         return (RESUME_SKIP_NONE);
 176 }
 177
 178 static void
 179 traverse_pause(traverse_data_t *td, const zbookmark_t *zb)
 180 {
 181         ASSERT(td->td_resume != NULL);
 182         ASSERT0(zb->zb_level);
 183         bcopy(zb, td->td_resume, sizeof (*td->td_resume));
 184 }
 185
 186 static void
 187 traverse_prefetch_metadata(traverse_data_t *td,
 188     const blkptr_t *bp, const zbookmark_t *zb)
 189 {
 190         uint32_t flags = ARC_NOWAIT | ARC_PREFETCH;
 191
 192         if (!(td->td_flags & TRAVERSE_PREFETCH_METADATA))
 193                 return;
 194         /*
 195          * If we are in the process of resuming, don't prefetch, because
 196          * some children will not be needed (and in fact may have already
 197          * been freed).
 198          */
 199         if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume))
 200                 return;
 201         if (BP_IS_HOLE(bp) || bp->blk_birth <= td->td_min_txg)
 202                 return;
 203         if (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_DNODE)
 204                 return;
 205
 206         (void) arc_read(NULL, td->td_spa, bp, NULL, NULL,
 207             ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
 208 }
 209
 210 static int
 211 traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp,
 212     const blkptr_t *bp, const zbookmark_t *zb)
 213 {
 214         int err = 0, lasterr = 0;
 215         arc_buf_t *buf = NULL;
 216         boolean_t pause = B_FALSE;
 217
 218         switch (resume_skip_check(td, dnp, zb)) {
 219         case RESUME_SKIP_ALL:
 220                 return (0);
 221         case RESUME_SKIP_CHILDREN:
 222                 goto post;
 223         case RESUME_SKIP_NONE:
 224                 break;
 225         default:
 226                 ASSERT(0);
 227         }
 228
 229         if (bp->blk_birth == 0) {
 230                 if (spa_feature_is_active(td->td_spa, SPA_FEATURE_HOLE_BIRTH)) {
 231                         /*
 232                          * Since this block has a birth time of 0 it must be a
 233                          * hole created before the SPA_FEATURE_HOLE_BIRTH
 234                          * feature was enabled.  If SPA_FEATURE_HOLE_BIRTH
 235                          * was enabled before the min_txg for this traveral we
 236                          * know the hole must have been created before the
 237                          * min_txg for this traveral, so we can skip it. If
 238                          * SPA_FEATURE_HOLE_BIRTH was enabled after the min_txg
 239                          * for this traveral we cannot tell if the hole was
 240                          * created before or after the min_txg for this
 241                          * traversal, so we cannot skip it.
 242                          */
 243                         uint64_t hole_birth_enabled_txg;
 244                         VERIFY(spa_feature_enabled_txg(td->td_spa,
 245                             SPA_FEATURE_HOLE_BIRTH, &hole_birth_enabled_txg));
 246                         if (hole_birth_enabled_txg < td->td_min_txg)
 247                                 return (0);
 248                 }
 249         } else if (bp->blk_birth <= td->td_min_txg) {
 250                 return (0);
 251         }
 252
 253         if (BP_IS_HOLE(bp)) {
 254                 err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg);
 255                 return (err);
 256         }
 257
 258         if (td->td_pfd && !td->td_pfd->pd_exited &&
 259             ((td->td_pfd->pd_flags & TRAVERSE_PREFETCH_DATA) ||
 260             BP_GET_TYPE(bp) == DMU_OT_DNODE || BP_GET_LEVEL(bp) > 0)) {
 261                 mutex_enter(&td->td_pfd->pd_mtx);
 262                 ASSERT(td->td_pfd->pd_blks_fetched >= 0);
 263                 while (td->td_pfd->pd_blks_fetched == 0 &&
 264                     !td->td_pfd->pd_exited)
 265                         cv_wait(&td->td_pfd->pd_cv, &td->td_pfd->pd_mtx);
 266                 td->td_pfd->pd_blks_fetched--;
 267                 cv_broadcast(&td->td_pfd->pd_cv);
 268                 mutex_exit(&td->td_pfd->pd_mtx);
 269         }
 270
 271         if (td->td_flags & TRAVERSE_PRE) {
 272                 err = td->td_func(td->td_spa, NULL, bp, zb, dnp,
 273                     td->td_arg);
 274                 if (err == TRAVERSE_VISIT_NO_CHILDREN)
 275                         return (0);
 276                 if (err == ERESTART)
 277                         pause = B_TRUE; /* handle pausing at a common point */
 278                 if (err != 0)
 279                         goto post;
 280         }
 281
 282         if (BP_GET_LEVEL(bp) > 0) {
 283                 uint32_t flags = ARC_WAIT;
 284                 int32_t i;
 285                 int32_t epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
 286                 zbookmark_t *czb;
 287
 288                 err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
 289                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
 290                 if (err != 0)
 291                         return (err);
 292
 293                 czb = kmem_alloc(sizeof (zbookmark_t), KM_PUSHPAGE);
 294
 295                 for (i = 0; i < epb; i++) {
 296                         SET_BOOKMARK(czb, zb->zb_objset, zb->zb_object,
 297                             zb->zb_level - 1,
 298                             zb->zb_blkid * epb + i);
 299                         traverse_prefetch_metadata(td,
 300                             &((blkptr_t *)buf->b_data)[i], czb);
 301                 }
 302
 303                 /* recursively visitbp() blocks below this */
 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                         err = traverse_visitbp(td, dnp,
 309                             &((blkptr_t *)buf->b_data)[i], czb);
 310                         if (err != 0) {
 311                                 if (!TD_HARD(td))
 312                                         break;
 313                                 lasterr = err;
 314                         }
 315                 }
 316
 317                 kmem_free(czb, sizeof (zbookmark_t));
 318
 319         } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) {
 320                 uint32_t flags = ARC_WAIT;
 321                 int32_t i;
 322                 int32_t epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT;
 323
 324                 err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
 325                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
 326                 if (err != 0)
 327                         return (err);
 328                 dnp = buf->b_data;
 329
 330                 for (i = 0; i < epb; i++) {
 331                         prefetch_dnode_metadata(td, &dnp[i], zb->zb_objset,
 332                             zb->zb_blkid * epb + i);
 333                 }
 334
 335                 /* recursively visitbp() blocks below this */
 336                 for (i = 0; i < epb; i++) {
 337                         err = traverse_dnode(td, &dnp[i], zb->zb_objset,
 338                             zb->zb_blkid * epb + i);
 339                         if (err != 0) {
 340                                 if (!TD_HARD(td))
 341                                         break;
 342                                 lasterr = err;
 343                         }
 344                 }
 345         } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) {
 346                 uint32_t flags = ARC_WAIT;
 347                 objset_phys_t *osp;
 348                 dnode_phys_t *dnp;
 349
 350                 err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
 351                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
 352                 if (err != 0)
 353                         return (err);
 354
 355                 osp = buf->b_data;
 356                 dnp = &osp->os_meta_dnode;
 357                 prefetch_dnode_metadata(td, dnp, zb->zb_objset,
 358                     DMU_META_DNODE_OBJECT);
 359                 if (arc_buf_size(buf) >= sizeof (objset_phys_t)) {
 360                         prefetch_dnode_metadata(td, &osp->os_groupused_dnode,
 361                             zb->zb_objset, DMU_GROUPUSED_OBJECT);
 362                         prefetch_dnode_metadata(td, &osp->os_userused_dnode,
 363                             zb->zb_objset, DMU_USERUSED_OBJECT);
 364                 }
 365
 366                 err = traverse_dnode(td, dnp, zb->zb_objset,
 367                     DMU_META_DNODE_OBJECT);
 368                 if (err && TD_HARD(td)) {
 369                         lasterr = err;
 370                         err = 0;
 371                 }
 372                 if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) {
 373                         dnp = &osp->os_groupused_dnode;
 374                         err = traverse_dnode(td, dnp, zb->zb_objset,
 375                             DMU_GROUPUSED_OBJECT);
 376                 }
 377                 if (err && TD_HARD(td)) {
 378                         lasterr = err;
 379                         err = 0;
 380                 }
 381                 if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) {
 382                         dnp = &osp->os_userused_dnode;
 383                         err = traverse_dnode(td, dnp, zb->zb_objset,
 384                             DMU_USERUSED_OBJECT);
 385                 }
 386         }
 387
 388         if (buf)
 389                 (void) arc_buf_remove_ref(buf, &buf);
 390
 391 post:
 392         if (err == 0 && (td->td_flags & TRAVERSE_POST)) {
 393                 err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg);
 394                 if (err == ERESTART)
 395                         pause = B_TRUE;
 396         }
 397
 398         if (pause && td->td_resume != NULL) {
 399                 ASSERT3U(err, ==, ERESTART);
 400                 ASSERT(!TD_HARD(td));
 401                 traverse_pause(td, zb);
 402         }
 403
 404         return (err != 0 ? err : lasterr);
 405 }
 406
 407 static void
 408 prefetch_dnode_metadata(traverse_data_t *td, const dnode_phys_t *dnp,
 409     uint64_t objset, uint64_t object)
 410 {
 411         int j;
 412         zbookmark_t czb;
 413
 414         for (j = 0; j < dnp->dn_nblkptr; j++) {
 415                 SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j);
 416                 traverse_prefetch_metadata(td, &dnp->dn_blkptr[j], &czb);
 417         }
 418
 419         if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
 420                 SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID);
 421                 traverse_prefetch_metadata(td, &dnp->dn_spill, &czb);
 422         }
 423 }
 424
 425 static int
 426 traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp,
 427     uint64_t objset, uint64_t object)
 428 {
 429         int j, err = 0, lasterr = 0;
 430         zbookmark_t czb;
 431
 432         for (j = 0; j < dnp->dn_nblkptr; j++) {
 433                 SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j);
 434                 err = traverse_visitbp(td, dnp, &dnp->dn_blkptr[j], &czb);
 435                 if (err != 0) {
 436                         if (!TD_HARD(td))
 437                                 break;
 438                         lasterr = err;
 439                 }
 440         }
 441
 442         if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
 443                 SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID);
 444                 err = traverse_visitbp(td, dnp, &dnp->dn_spill, &czb);
 445                 if (err != 0) {
 446                         if (!TD_HARD(td))
 447                                 return (err);
 448                         lasterr = err;
 449                 }
 450         }
 451         return (err != 0 ? err : lasterr);
 452 }
 453
 454 /* ARGSUSED */
 455 static int
 456 traverse_prefetcher(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 457     const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
 458 {
 459         prefetch_data_t *pfd = arg;
 460         uint32_t aflags = ARC_NOWAIT | ARC_PREFETCH;
 461
 462         ASSERT(pfd->pd_blks_fetched >= 0);
 463         if (pfd->pd_cancel)
 464                 return (SET_ERROR(EINTR));
 465
 466         if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp) ||
 467             !((pfd->pd_flags & TRAVERSE_PREFETCH_DATA) ||
 468             BP_GET_TYPE(bp) == DMU_OT_DNODE || BP_GET_LEVEL(bp) > 0) ||
 469             BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG)
 470                 return (0);
 471
 472         mutex_enter(&pfd->pd_mtx);
 473         while (!pfd->pd_cancel && pfd->pd_blks_fetched >= pfd->pd_blks_max)
 474                 cv_wait(&pfd->pd_cv, &pfd->pd_mtx);
 475         pfd->pd_blks_fetched++;
 476         cv_broadcast(&pfd->pd_cv);
 477         mutex_exit(&pfd->pd_mtx);
 478
 479         (void) arc_read(NULL, spa, bp, NULL, NULL, ZIO_PRIORITY_ASYNC_READ,
 480             ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &aflags, zb);
 481
 482         return (0);
 483 }
 484
 485 static void
 486 traverse_prefetch_thread(void *arg)
 487 {
 488         traverse_data_t *td_main = arg;
 489         traverse_data_t td = *td_main;
 490         zbookmark_t czb;
 491
 492         td.td_func = traverse_prefetcher;
 493         td.td_arg = td_main->td_pfd;
 494         td.td_pfd = NULL;
 495
 496         SET_BOOKMARK(&czb, td.td_objset,
 497             ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
 498         (void) traverse_visitbp(&td, NULL, td.td_rootbp, &czb);
 499
 500         mutex_enter(&td_main->td_pfd->pd_mtx);
 501         td_main->td_pfd->pd_exited = B_TRUE;
 502         cv_broadcast(&td_main->td_pfd->pd_cv);
 503         mutex_exit(&td_main->td_pfd->pd_mtx);
 504 }
 505
 506 /*
 507  * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
 508  * in syncing context).
 509  */
 510 static int
 511 traverse_impl(spa_t *spa, dsl_dataset_t *ds, uint64_t objset, blkptr_t *rootbp,
 512     uint64_t txg_start, zbookmark_t *resume, int flags,
 513     blkptr_cb_t func, void *arg)
 514 {
 515         traverse_data_t *td;
 516         prefetch_data_t *pd;
 517         zbookmark_t *czb;
 518         int err;
 519
 520         ASSERT(ds == NULL || objset == ds->ds_object);
 521         ASSERT(!(flags & TRAVERSE_PRE) || !(flags & TRAVERSE_POST));
 522
 523         /*
 524          * The data prefetching mechanism (the prefetch thread) is incompatible
 525          * with resuming from a bookmark.
 526          */
 527         ASSERT(resume == NULL || !(flags & TRAVERSE_PREFETCH_DATA));
 528
 529         td = kmem_alloc(sizeof (traverse_data_t), KM_PUSHPAGE);
 530         pd = kmem_zalloc(sizeof (prefetch_data_t), KM_PUSHPAGE);
 531         czb = kmem_alloc(sizeof (zbookmark_t), KM_PUSHPAGE);
 532
 533         td->td_spa = spa;
 534         td->td_objset = objset;
 535         td->td_rootbp = rootbp;
 536         td->td_min_txg = txg_start;
 537         td->td_resume = resume;
 538         td->td_func = func;
 539         td->td_arg = arg;
 540         td->td_pfd = pd;
 541         td->td_flags = flags;
 542
 543         pd->pd_blks_max = zfs_pd_blks_max;
 544         pd->pd_flags = flags;
 545         mutex_init(&pd->pd_mtx, NULL, MUTEX_DEFAULT, NULL);
 546         cv_init(&pd->pd_cv, NULL, CV_DEFAULT, NULL);
 547
 548         SET_BOOKMARK(czb, td->td_objset,
 549             ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
 550
 551         /* See comment on ZIL traversal in dsl_scan_visitds. */
 552         if (ds != NULL && !dsl_dataset_is_snapshot(ds) && !BP_IS_HOLE(rootbp)) {
 553                 uint32_t flags = ARC_WAIT;
 554                 objset_phys_t *osp;
 555                 arc_buf_t *buf;
 556
 557                 err = arc_read(NULL, td->td_spa, rootbp,
 558                     arc_getbuf_func, &buf,
 559                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, czb);
 560                 if (err != 0)
 561                         return (err);
 562
 563                 osp = buf->b_data;
 564                 traverse_zil(td, &osp->os_zil_header);
 565                 (void) arc_buf_remove_ref(buf, &buf);
 566         }
 567
 568         if (!(flags & TRAVERSE_PREFETCH_DATA) ||
 569             0 == taskq_dispatch(system_taskq, traverse_prefetch_thread,
 570             td, TQ_NOQUEUE))
 571                 pd->pd_exited = B_TRUE;
 572
 573         err = traverse_visitbp(td, NULL, rootbp, czb);
 574
 575         mutex_enter(&pd->pd_mtx);
 576         pd->pd_cancel = B_TRUE;
 577         cv_broadcast(&pd->pd_cv);
 578         while (!pd->pd_exited)
 579                 cv_wait(&pd->pd_cv, &pd->pd_mtx);
 580         mutex_exit(&pd->pd_mtx);
 581
 582         mutex_destroy(&pd->pd_mtx);
 583         cv_destroy(&pd->pd_cv);
 584
 585         kmem_free(czb, sizeof (zbookmark_t));
 586         kmem_free(pd, sizeof (struct prefetch_data));
 587         kmem_free(td, sizeof (struct traverse_data));
 588
 589         return (err);
 590 }
 591
 592 /*
 593  * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
 594  * in syncing context).
 595  */
 596 int
 597 traverse_dataset(dsl_dataset_t *ds, uint64_t txg_start, int flags,
 598     blkptr_cb_t func, void *arg)
 599 {
 600         return (traverse_impl(ds->ds_dir->dd_pool->dp_spa, ds, ds->ds_object,
 601             &ds->ds_phys->ds_bp, txg_start, NULL, flags, func, arg));
 602 }
 603
 604 int
 605 traverse_dataset_destroyed(spa_t *spa, blkptr_t *blkptr,
 606     uint64_t txg_start, zbookmark_t *resume, int flags,
 607     blkptr_cb_t func, void *arg)
 608 {
 609         return (traverse_impl(spa, NULL, ZB_DESTROYED_OBJSET,
 610             blkptr, txg_start, resume, flags, func, arg));
 611 }
 612
 613 /*
 614  * NB: pool must not be changing on-disk (eg, from zdb or sync context).
 615  */
 616 int
 617 traverse_pool(spa_t *spa, uint64_t txg_start, int flags,
 618     blkptr_cb_t func, void *arg)
 619 {
 620         int err, lasterr = 0;
 621         uint64_t obj;
 622         dsl_pool_t *dp = spa_get_dsl(spa);
 623         objset_t *mos = dp->dp_meta_objset;
 624         boolean_t hard = (flags & TRAVERSE_HARD);
 625
 626         /* visit the MOS */
 627         err = traverse_impl(spa, NULL, 0, spa_get_rootblkptr(spa),
 628             txg_start, NULL, flags, func, arg);
 629         if (err != 0)
 630                 return (err);
 631
 632         /* visit each dataset */
 633         for (obj = 1; err == 0 || (err != ESRCH && hard);
 634             err = dmu_object_next(mos, &obj, FALSE, txg_start)) {
 635                 dmu_object_info_t doi;
 636
 637                 err = dmu_object_info(mos, obj, &doi);
 638                 if (err != 0) {
 639                         if (!hard)
 640                                 return (err);
 641                         lasterr = err;
 642                         continue;
 643                 }
 644
 645                 if (doi.doi_bonus_type == DMU_OT_DSL_DATASET) {
 646                         dsl_dataset_t *ds;
 647                         uint64_t txg = txg_start;
 648
 649                         dsl_pool_config_enter(dp, FTAG);
 650                         err = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
 651                         dsl_pool_config_exit(dp, FTAG);
 652                         if (err != 0) {
 653                                 if (!hard)
 654                                         return (err);
 655                                 lasterr = err;
 656                                 continue;
 657                         }
 658                         if (ds->ds_phys->ds_prev_snap_txg > txg)
 659                                 txg = ds->ds_phys->ds_prev_snap_txg;
 660                         err = traverse_dataset(ds, txg, flags, func, arg);
 661                         dsl_dataset_rele(ds, FTAG);
 662                         if (err != 0) {
 663                                 if (!hard)
 664                                         return (err);
 665                                 lasterr = err;
 666                         }
 667                 }
 668         }
 669         if (err == ESRCH)
 670                 err = 0;
 671         return (err != 0 ? err : lasterr);
 672 }
 673
 674 #if defined(_KERNEL) && defined(HAVE_SPL)
 675 EXPORT_SYMBOL(traverse_dataset);
 676 EXPORT_SYMBOL(traverse_pool);
 677
 678 module_param(zfs_pd_blks_max, int, 0644);
 679 MODULE_PARM_DESC(zfs_pd_blks_max, "Max number of blocks to prefetch");
 680 #endif