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
  40 int zfs_pd_blks_max = 100;
  41
  42 typedef struct prefetch_data {
  43         kmutex_t pd_mtx;
  44         kcondvar_t pd_cv;
  45         int pd_blks_max;
  46         int pd_blks_fetched;
  47         int pd_flags;
  48         boolean_t pd_cancel;
  49         boolean_t pd_exited;
  50 } prefetch_data_t;
  51
  52 typedef struct traverse_data {
  53         spa_t *td_spa;
  54         uint64_t td_objset;
  55         blkptr_t *td_rootbp;
  56         uint64_t td_min_txg;
  57         zbookmark_t *td_resume;
  58         int td_flags;
  59         prefetch_data_t *td_pfd;
  60         blkptr_cb_t *td_func;
  61         void *td_arg;
  62 } traverse_data_t;
  63
  64 #define TD_HARD(td)     (td->td_flags & TRAVERSE_HARD)
  65
  66 static int traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp,
  67     uint64_t objset, uint64_t object);
  68 static void prefetch_dnode_metadata(traverse_data_t *td, const dnode_phys_t *,
  69     uint64_t objset, uint64_t object);
  70
  71 static int
  72 traverse_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg)
  73 {
  74         traverse_data_t *td = arg;
  75         zbookmark_t zb;
  76
  77         if (bp->blk_birth == 0)
  78                 return (0);
  79
  80         if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(td->td_spa))
  81                 return (0);
  82
  83         SET_BOOKMARK(&zb, td->td_objset, ZB_ZIL_OBJECT, ZB_ZIL_LEVEL,
  84             bp->blk_cksum.zc_word[ZIL_ZC_SEQ]);
  85
  86         (void) td->td_func(td->td_spa, zilog, bp, &zb, NULL, td->td_arg);
  87
  88         return (0);
  89 }
  90
  91 static int
  92 traverse_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg)
  93 {
  94         traverse_data_t *td = arg;
  95
  96         if (lrc->lrc_txtype == TX_WRITE) {
  97                 lr_write_t *lr = (lr_write_t *)lrc;
  98                 blkptr_t *bp = &lr->lr_blkptr;
  99                 zbookmark_t zb;
 100
 101                 if (bp->blk_birth == 0)
 102                         return (0);
 103
 104                 if (claim_txg == 0 || bp->blk_birth < claim_txg)
 105                         return (0);
 106
 107                 SET_BOOKMARK(&zb, td->td_objset, lr->lr_foid,
 108                     ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp));
 109
 110                 (void) td->td_func(td->td_spa, zilog, bp, &zb, NULL,
 111                     td->td_arg);
 112         }
 113         return (0);
 114 }
 115
 116 static void
 117 traverse_zil(traverse_data_t *td, zil_header_t *zh)
 118 {
 119         uint64_t claim_txg = zh->zh_claim_txg;
 120         zilog_t *zilog;
 121
 122         /*
 123          * We only want to visit blocks that have been claimed but not yet
 124          * replayed; plus, in read-only mode, blocks that are already stable.
 125          */
 126         if (claim_txg == 0 && spa_writeable(td->td_spa))
 127                 return;
 128
 129         zilog = zil_alloc(spa_get_dsl(td->td_spa)->dp_meta_objset, zh);
 130
 131         (void) zil_parse(zilog, traverse_zil_block, traverse_zil_record, td,
 132             claim_txg);
 133
 134         zil_free(zilog);
 135 }
 136
 137 typedef enum resume_skip {
 138         RESUME_SKIP_ALL,
 139         RESUME_SKIP_NONE,
 140         RESUME_SKIP_CHILDREN
 141 } resume_skip_t;
 142
 143 /*
 144  * Returns RESUME_SKIP_ALL if td indicates that we are resuming a traversal and
 145  * the block indicated by zb does not need to be visited at all. Returns
 146  * RESUME_SKIP_CHILDREN if we are resuming a post traversal and we reach the
 147  * resume point. This indicates that this block should be visited but not its
 148  * children (since they must have been visited in a previous traversal).
 149  * Otherwise returns RESUME_SKIP_NONE.
 150  */
 151 static resume_skip_t
 152 resume_skip_check(traverse_data_t *td, const dnode_phys_t *dnp,
 153     const zbookmark_t *zb)
 154 {
 155         if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume)) {
 156                 /*
 157                  * If we already visited this bp & everything below,
 158                  * don't bother doing it again.
 159                  */
 160                 if (zbookmark_is_before(dnp, zb, td->td_resume))
 161                         return (RESUME_SKIP_ALL);
 162
 163                 /*
 164                  * If we found the block we're trying to resume from, zero
 165                  * the bookmark out to indicate that we have resumed.
 166                  */
 167                 ASSERT3U(zb->zb_object, <=, td->td_resume->zb_object);
 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_pause(traverse_data_t *td, const zbookmark_t *zb)
 179 {
 180         ASSERT(td->td_resume != NULL);
 181         ASSERT0(zb->zb_level);
 182         bcopy(zb, td->td_resume, sizeof (*td->td_resume));
 183 }
 184
 185 static void
 186 traverse_prefetch_metadata(traverse_data_t *td,
 187     const blkptr_t *bp, const zbookmark_t *zb)
 188 {
 189         uint32_t flags = ARC_NOWAIT | ARC_PREFETCH;
 190
 191         if (!(td->td_flags & TRAVERSE_PREFETCH_METADATA))
 192                 return;
 193         /*
 194          * If we are in the process of resuming, don't prefetch, because
 195          * some children will not be needed (and in fact may have already
 196          * been freed).
 197          */
 198         if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume))
 199                 return;
 200         if (BP_IS_HOLE(bp) || bp->blk_birth <= td->td_min_txg)
 201                 return;
 202         if (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_DNODE)
 203                 return;
 204
 205         (void) arc_read(NULL, td->td_spa, bp, NULL, NULL,
 206             ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
 207 }
 208
 209 static int
 210 traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp,
 211     const blkptr_t *bp, const zbookmark_t *zb)
 212 {
 213         int err = 0, lasterr = 0;
 214         arc_buf_t *buf = NULL;
 215         boolean_t pause = B_FALSE;
 216
 217         switch (resume_skip_check(td, dnp, zb)) {
 218         case RESUME_SKIP_ALL:
 219                 return (0);
 220         case RESUME_SKIP_CHILDREN:
 221                 goto post;
 222         case RESUME_SKIP_NONE:
 223                 break;
 224         default:
 225                 ASSERT(0);
 226         }
 227
 228         if (BP_IS_HOLE(bp)) {
 229                 err = td->td_func(td->td_spa, NULL, NULL, zb, dnp, td->td_arg);
 230                 return (err);
 231         }
 232
 233         if (bp->blk_birth <= td->td_min_txg)
 234                 return (0);
 235
 236         if (td->td_pfd && !td->td_pfd->pd_exited &&
 237             ((td->td_pfd->pd_flags & TRAVERSE_PREFETCH_DATA) ||
 238             BP_GET_TYPE(bp) == DMU_OT_DNODE || BP_GET_LEVEL(bp) > 0)) {
 239                 mutex_enter(&td->td_pfd->pd_mtx);
 240                 ASSERT(td->td_pfd->pd_blks_fetched >= 0);
 241                 while (td->td_pfd->pd_blks_fetched == 0 &&
 242                     !td->td_pfd->pd_exited)
 243                         cv_wait(&td->td_pfd->pd_cv, &td->td_pfd->pd_mtx);
 244                 td->td_pfd->pd_blks_fetched--;
 245                 cv_broadcast(&td->td_pfd->pd_cv);
 246                 mutex_exit(&td->td_pfd->pd_mtx);
 247         }
 248
 249         if (td->td_flags & TRAVERSE_PRE) {
 250                 err = td->td_func(td->td_spa, NULL, bp, zb, dnp,
 251                     td->td_arg);
 252                 if (err == TRAVERSE_VISIT_NO_CHILDREN)
 253                         return (0);
 254                 if (err == ERESTART)
 255                         pause = B_TRUE; /* handle pausing at a common point */
 256                 if (err != 0)
 257                         goto post;
 258         }
 259
 260         if (BP_GET_LEVEL(bp) > 0) {
 261                 uint32_t flags = ARC_WAIT;
 262                 int32_t i;
 263                 int32_t epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
 264                 zbookmark_t *czb;
 265
 266                 err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
 267                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
 268                 if (err != 0)
 269                         return (err);
 270
 271                 czb = kmem_alloc(sizeof (zbookmark_t), KM_PUSHPAGE);
 272
 273                 for (i = 0; i < epb; i++) {
 274                         SET_BOOKMARK(czb, zb->zb_objset, zb->zb_object,
 275                             zb->zb_level - 1,
 276                             zb->zb_blkid * epb + i);
 277                         traverse_prefetch_metadata(td,
 278                             &((blkptr_t *)buf->b_data)[i], czb);
 279                 }
 280
 281                 /* recursively visitbp() blocks below this */
 282                 for (i = 0; i < epb; i++) {
 283                         SET_BOOKMARK(czb, zb->zb_objset, zb->zb_object,
 284                             zb->zb_level - 1,
 285                             zb->zb_blkid * epb + i);
 286                         err = traverse_visitbp(td, dnp,
 287                             &((blkptr_t *)buf->b_data)[i], czb);
 288                         if (err != 0) {
 289                                 if (!TD_HARD(td))
 290                                         break;
 291                                 lasterr = err;
 292                         }
 293                 }
 294
 295                 kmem_free(czb, sizeof (zbookmark_t));
 296
 297         } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) {
 298                 uint32_t flags = ARC_WAIT;
 299                 int32_t i;
 300                 int32_t epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT;
 301
 302                 err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
 303                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
 304                 if (err != 0)
 305                         return (err);
 306                 dnp = buf->b_data;
 307
 308                 for (i = 0; i < epb; i++) {
 309                         prefetch_dnode_metadata(td, &dnp[i], zb->zb_objset,
 310                             zb->zb_blkid * epb + i);
 311                 }
 312
 313                 /* recursively visitbp() blocks below this */
 314                 for (i = 0; i < epb; i++) {
 315                         err = traverse_dnode(td, &dnp[i], zb->zb_objset,
 316                             zb->zb_blkid * epb + i);
 317                         if (err != 0) {
 318                                 if (!TD_HARD(td))
 319                                         break;
 320                                 lasterr = err;
 321                         }
 322                 }
 323         } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) {
 324                 uint32_t flags = ARC_WAIT;
 325                 objset_phys_t *osp;
 326                 dnode_phys_t *dnp;
 327
 328                 err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
 329                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
 330                 if (err != 0)
 331                         return (err);
 332
 333                 osp = buf->b_data;
 334                 dnp = &osp->os_meta_dnode;
 335                 prefetch_dnode_metadata(td, dnp, zb->zb_objset,
 336                     DMU_META_DNODE_OBJECT);
 337                 if (arc_buf_size(buf) >= sizeof (objset_phys_t)) {
 338                         prefetch_dnode_metadata(td, &osp->os_groupused_dnode,
 339                             zb->zb_objset, DMU_GROUPUSED_OBJECT);
 340                         prefetch_dnode_metadata(td, &osp->os_userused_dnode,
 341                             zb->zb_objset, DMU_USERUSED_OBJECT);
 342                 }
 343
 344                 err = traverse_dnode(td, dnp, zb->zb_objset,
 345                     DMU_META_DNODE_OBJECT);
 346                 if (err && TD_HARD(td)) {
 347                         lasterr = err;
 348                         err = 0;
 349                 }
 350                 if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) {
 351                         dnp = &osp->os_groupused_dnode;
 352                         err = traverse_dnode(td, dnp, zb->zb_objset,
 353                             DMU_GROUPUSED_OBJECT);
 354                 }
 355                 if (err && TD_HARD(td)) {
 356                         lasterr = err;
 357                         err = 0;
 358                 }
 359                 if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) {
 360                         dnp = &osp->os_userused_dnode;
 361                         err = traverse_dnode(td, dnp, zb->zb_objset,
 362                             DMU_USERUSED_OBJECT);
 363                 }
 364         }
 365
 366         if (buf)
 367                 (void) arc_buf_remove_ref(buf, &buf);
 368
 369 post:
 370         if (err == 0 && (td->td_flags & TRAVERSE_POST)) {
 371                 err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg);
 372                 if (err == ERESTART)
 373                         pause = B_TRUE;
 374         }
 375
 376         if (pause && td->td_resume != NULL) {
 377                 ASSERT3U(err, ==, ERESTART);
 378                 ASSERT(!TD_HARD(td));
 379                 traverse_pause(td, zb);
 380         }
 381
 382         return (err != 0 ? err : lasterr);
 383 }
 384
 385 static void
 386 prefetch_dnode_metadata(traverse_data_t *td, const dnode_phys_t *dnp,
 387     uint64_t objset, uint64_t object)
 388 {
 389         int j;
 390         zbookmark_t czb;
 391
 392         for (j = 0; j < dnp->dn_nblkptr; j++) {
 393                 SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j);
 394                 traverse_prefetch_metadata(td, &dnp->dn_blkptr[j], &czb);
 395         }
 396
 397         if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
 398                 SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID);
 399                 traverse_prefetch_metadata(td, &dnp->dn_spill, &czb);
 400         }
 401 }
 402
 403 static int
 404 traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp,
 405     uint64_t objset, uint64_t object)
 406 {
 407         int j, err = 0, lasterr = 0;
 408         zbookmark_t czb;
 409
 410         for (j = 0; j < dnp->dn_nblkptr; j++) {
 411                 SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j);
 412                 err = traverse_visitbp(td, dnp, &dnp->dn_blkptr[j], &czb);
 413                 if (err != 0) {
 414                         if (!TD_HARD(td))
 415                                 break;
 416                         lasterr = err;
 417                 }
 418         }
 419
 420         if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
 421                 SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID);
 422                 err = traverse_visitbp(td, dnp, &dnp->dn_spill, &czb);
 423                 if (err != 0) {
 424                         if (!TD_HARD(td))
 425                                 return (err);
 426                         lasterr = err;
 427                 }
 428         }
 429         return (err != 0 ? err : lasterr);
 430 }
 431
 432 /* ARGSUSED */
 433 static int
 434 traverse_prefetcher(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 435     const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
 436 {
 437         prefetch_data_t *pfd = arg;
 438         uint32_t aflags = ARC_NOWAIT | ARC_PREFETCH;
 439
 440         ASSERT(pfd->pd_blks_fetched >= 0);
 441         if (pfd->pd_cancel)
 442                 return (SET_ERROR(EINTR));
 443
 444         if (bp == NULL || !((pfd->pd_flags & TRAVERSE_PREFETCH_DATA) ||
 445             BP_GET_TYPE(bp) == DMU_OT_DNODE || BP_GET_LEVEL(bp) > 0) ||
 446             BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG)
 447                 return (0);
 448
 449         mutex_enter(&pfd->pd_mtx);
 450         while (!pfd->pd_cancel && pfd->pd_blks_fetched >= pfd->pd_blks_max)
 451                 cv_wait(&pfd->pd_cv, &pfd->pd_mtx);
 452         pfd->pd_blks_fetched++;
 453         cv_broadcast(&pfd->pd_cv);
 454         mutex_exit(&pfd->pd_mtx);
 455
 456         (void) arc_read(NULL, spa, bp, NULL, NULL, ZIO_PRIORITY_ASYNC_READ,
 457             ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &aflags, zb);
 458
 459         return (0);
 460 }
 461
 462 static void
 463 traverse_prefetch_thread(void *arg)
 464 {
 465         traverse_data_t *td_main = arg;
 466         traverse_data_t td = *td_main;
 467         zbookmark_t czb;
 468
 469         td.td_func = traverse_prefetcher;
 470         td.td_arg = td_main->td_pfd;
 471         td.td_pfd = NULL;
 472
 473         SET_BOOKMARK(&czb, td.td_objset,
 474             ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
 475         (void) traverse_visitbp(&td, NULL, td.td_rootbp, &czb);
 476
 477         mutex_enter(&td_main->td_pfd->pd_mtx);
 478         td_main->td_pfd->pd_exited = B_TRUE;
 479         cv_broadcast(&td_main->td_pfd->pd_cv);
 480         mutex_exit(&td_main->td_pfd->pd_mtx);
 481 }
 482
 483 /*
 484  * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
 485  * in syncing context).
 486  */
 487 static int
 488 traverse_impl(spa_t *spa, dsl_dataset_t *ds, uint64_t objset, blkptr_t *rootbp,
 489     uint64_t txg_start, zbookmark_t *resume, int flags,
 490     blkptr_cb_t func, void *arg)
 491 {
 492         traverse_data_t *td;
 493         prefetch_data_t *pd;
 494         zbookmark_t *czb;
 495         int err;
 496
 497         ASSERT(ds == NULL || objset == ds->ds_object);
 498         ASSERT(!(flags & TRAVERSE_PRE) || !(flags & TRAVERSE_POST));
 499
 500         /*
 501          * The data prefetching mechanism (the prefetch thread) is incompatible
 502          * with resuming from a bookmark.
 503          */
 504         ASSERT(resume == NULL || !(flags & TRAVERSE_PREFETCH_DATA));
 505
 506         td = kmem_alloc(sizeof (traverse_data_t), KM_PUSHPAGE);
 507         pd = kmem_zalloc(sizeof (prefetch_data_t), KM_PUSHPAGE);
 508         czb = kmem_alloc(sizeof (zbookmark_t), KM_PUSHPAGE);
 509
 510         td->td_spa = spa;
 511         td->td_objset = objset;
 512         td->td_rootbp = rootbp;
 513         td->td_min_txg = txg_start;
 514         td->td_resume = resume;
 515         td->td_func = func;
 516         td->td_arg = arg;
 517         td->td_pfd = pd;
 518         td->td_flags = flags;
 519
 520         pd->pd_blks_max = zfs_pd_blks_max;
 521         pd->pd_flags = flags;
 522         mutex_init(&pd->pd_mtx, NULL, MUTEX_DEFAULT, NULL);
 523         cv_init(&pd->pd_cv, NULL, CV_DEFAULT, NULL);
 524
 525         SET_BOOKMARK(czb, td->td_objset,
 526             ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
 527
 528         /* See comment on ZIL traversal in dsl_scan_visitds. */
 529         if (ds != NULL && !dsl_dataset_is_snapshot(ds) && !BP_IS_HOLE(rootbp)) {
 530                 uint32_t flags = ARC_WAIT;
 531                 objset_phys_t *osp;
 532                 arc_buf_t *buf;
 533
 534                 err = arc_read(NULL, td->td_spa, rootbp,
 535                     arc_getbuf_func, &buf,
 536                     ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, czb);
 537                 if (err != 0)
 538                         return (err);
 539
 540                 osp = buf->b_data;
 541                 traverse_zil(td, &osp->os_zil_header);
 542                 (void) arc_buf_remove_ref(buf, &buf);
 543         }
 544
 545         if (!(flags & TRAVERSE_PREFETCH_DATA) ||
 546             0 == taskq_dispatch(system_taskq, traverse_prefetch_thread,
 547             td, TQ_NOQUEUE))
 548                 pd->pd_exited = B_TRUE;
 549
 550         err = traverse_visitbp(td, NULL, rootbp, czb);
 551
 552         mutex_enter(&pd->pd_mtx);
 553         pd->pd_cancel = B_TRUE;
 554         cv_broadcast(&pd->pd_cv);
 555         while (!pd->pd_exited)
 556                 cv_wait(&pd->pd_cv, &pd->pd_mtx);
 557         mutex_exit(&pd->pd_mtx);
 558
 559         mutex_destroy(&pd->pd_mtx);
 560         cv_destroy(&pd->pd_cv);
 561
 562         kmem_free(czb, sizeof (zbookmark_t));
 563         kmem_free(pd, sizeof (struct prefetch_data));
 564         kmem_free(td, sizeof (struct traverse_data));
 565
 566         return (err);
 567 }
 568
 569 /*
 570  * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
 571  * in syncing context).
 572  */
 573 int
 574 traverse_dataset(dsl_dataset_t *ds, uint64_t txg_start, int flags,
 575     blkptr_cb_t func, void *arg)
 576 {
 577         return (traverse_impl(ds->ds_dir->dd_pool->dp_spa, ds, ds->ds_object,
 578             &ds->ds_phys->ds_bp, txg_start, NULL, flags, func, arg));
 579 }
 580
 581 int
 582 traverse_dataset_destroyed(spa_t *spa, blkptr_t *blkptr,
 583     uint64_t txg_start, zbookmark_t *resume, int flags,
 584     blkptr_cb_t func, void *arg)
 585 {
 586         return (traverse_impl(spa, NULL, ZB_DESTROYED_OBJSET,
 587             blkptr, txg_start, resume, flags, func, arg));
 588 }
 589
 590 /*
 591  * NB: pool must not be changing on-disk (eg, from zdb or sync context).
 592  */
 593 int
 594 traverse_pool(spa_t *spa, uint64_t txg_start, int flags,
 595     blkptr_cb_t func, void *arg)
 596 {
 597         int err, lasterr = 0;
 598         uint64_t obj;
 599         dsl_pool_t *dp = spa_get_dsl(spa);
 600         objset_t *mos = dp->dp_meta_objset;
 601         boolean_t hard = (flags & TRAVERSE_HARD);
 602
 603         /* visit the MOS */
 604         err = traverse_impl(spa, NULL, 0, spa_get_rootblkptr(spa),
 605             txg_start, NULL, flags, func, arg);
 606         if (err != 0)
 607                 return (err);
 608
 609         /* visit each dataset */
 610         for (obj = 1; err == 0 || (err != ESRCH && hard);
 611             err = dmu_object_next(mos, &obj, FALSE, txg_start)) {
 612                 dmu_object_info_t doi;
 613
 614                 err = dmu_object_info(mos, obj, &doi);
 615                 if (err != 0) {
 616                         if (!hard)
 617                                 return (err);
 618                         lasterr = err;
 619                         continue;
 620                 }
 621
 622                 if (doi.doi_type == DMU_OT_DSL_DATASET) {
 623                         dsl_dataset_t *ds;
 624                         uint64_t txg = txg_start;
 625
 626                         dsl_pool_config_enter(dp, FTAG);
 627                         err = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
 628                         dsl_pool_config_exit(dp, FTAG);
 629                         if (err != 0) {
 630                                 if (!hard)
 631                                         return (err);
 632                                 lasterr = err;
 633                                 continue;
 634                         }
 635                         if (ds->ds_phys->ds_prev_snap_txg > txg)
 636                                 txg = ds->ds_phys->ds_prev_snap_txg;
 637                         err = traverse_dataset(ds, txg, flags, func, arg);
 638                         dsl_dataset_rele(ds, FTAG);
 639                         if (err != 0) {
 640                                 if (!hard)
 641                                         return (err);
 642                                 lasterr = err;
 643                         }
 644                 }
 645         }
 646         if (err == ESRCH)
 647                 err = 0;
 648         return (err != 0 ? err : lasterr);
 649 }
 650
 651 #if defined(_KERNEL) && defined(HAVE_SPL)
 652 EXPORT_SYMBOL(traverse_dataset);
 653 EXPORT_SYMBOL(traverse_pool);
 654
 655 module_param(zfs_pd_blks_max, int, 0644);
 656 MODULE_PARM_DESC(zfs_pd_blks_max, "Max number of blocks to prefetch");
 657 #endif