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.
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.
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]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2016 by Delphix. All rights reserved.
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>
35 #include <sys/dmu_impl.h>
37 #include <sys/sa_impl.h>
38 #include <sys/callb.h>
39 #include <sys/zfeature.h>
41 int32_t zfs_pd_bytes_max
= 50 * 1024 * 1024; /* 50MB */
43 typedef struct prefetch_data
{
46 int32_t pd_bytes_fetched
;
52 typedef struct traverse_data
{
57 zbookmark_phys_t
*td_resume
;
59 prefetch_data_t
*td_pfd
;
61 uint64_t td_hole_birth_enabled_txg
;
64 boolean_t td_realloc_possible
;
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
);
73 traverse_zil_block(zilog_t
*zilog
, blkptr_t
*bp
, void *arg
, uint64_t claim_txg
)
75 traverse_data_t
*td
= arg
;
81 if (claim_txg
== 0 && bp
->blk_birth
>= spa_first_txg(td
->td_spa
))
84 SET_BOOKMARK(&zb
, td
->td_objset
, ZB_ZIL_OBJECT
, ZB_ZIL_LEVEL
,
85 bp
->blk_cksum
.zc_word
[ZIL_ZC_SEQ
]);
87 (void) td
->td_func(td
->td_spa
, zilog
, bp
, &zb
, NULL
, td
->td_arg
);
93 traverse_zil_record(zilog_t
*zilog
, lr_t
*lrc
, void *arg
, uint64_t claim_txg
)
95 traverse_data_t
*td
= arg
;
97 if (lrc
->lrc_txtype
== TX_WRITE
) {
98 lr_write_t
*lr
= (lr_write_t
*)lrc
;
99 blkptr_t
*bp
= &lr
->lr_blkptr
;
105 if (claim_txg
== 0 || bp
->blk_birth
< claim_txg
)
108 SET_BOOKMARK(&zb
, td
->td_objset
, lr
->lr_foid
,
109 ZB_ZIL_LEVEL
, lr
->lr_offset
/ BP_GET_LSIZE(bp
));
111 (void) td
->td_func(td
->td_spa
, zilog
, bp
, &zb
, NULL
,
118 traverse_zil(traverse_data_t
*td
, zil_header_t
*zh
)
120 uint64_t claim_txg
= zh
->zh_claim_txg
;
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.
127 if (claim_txg
== 0 && spa_writeable(td
->td_spa
))
130 zilog
= zil_alloc(spa_get_dsl(td
->td_spa
)->dp_meta_objset
, zh
);
132 (void) zil_parse(zilog
, traverse_zil_block
, traverse_zil_record
, td
,
138 typedef enum resume_skip
{
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.
153 resume_skip_check(traverse_data_t
*td
, const dnode_phys_t
*dnp
,
154 const zbookmark_phys_t
*zb
)
156 if (td
->td_resume
!= NULL
&& !ZB_IS_ZERO(td
->td_resume
)) {
158 * If we already visited this bp & everything below,
159 * don't bother doing it again.
161 if (zbookmark_subtree_completed(dnp
, zb
, td
->td_resume
))
162 return (RESUME_SKIP_ALL
);
165 * If we found the block we're trying to resume from, zero
166 * the bookmark out to indicate that we have resumed.
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
);
174 return (RESUME_SKIP_NONE
);
178 traverse_prefetch_metadata(traverse_data_t
*td
,
179 const blkptr_t
*bp
, const zbookmark_phys_t
*zb
)
181 arc_flags_t flags
= ARC_FLAG_NOWAIT
| ARC_FLAG_PREFETCH
;
183 if (!(td
->td_flags
& TRAVERSE_PREFETCH_METADATA
))
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
190 if (td
->td_resume
!= NULL
&& !ZB_IS_ZERO(td
->td_resume
))
192 if (BP_IS_HOLE(bp
) || bp
->blk_birth
<= td
->td_min_txg
)
194 if (BP_GET_LEVEL(bp
) == 0 && BP_GET_TYPE(bp
) != DMU_OT_DNODE
)
197 (void) arc_read(NULL
, td
->td_spa
, bp
, NULL
, NULL
,
198 ZIO_PRIORITY_ASYNC_READ
, ZIO_FLAG_CANFAIL
, &flags
, zb
);
202 prefetch_needed(prefetch_data_t
*pfd
, const blkptr_t
*bp
)
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
)
212 traverse_visitbp(traverse_data_t
*td
, const dnode_phys_t
*dnp
,
213 const blkptr_t
*bp
, const zbookmark_phys_t
*zb
)
216 arc_buf_t
*buf
= NULL
;
217 prefetch_data_t
*pd
= td
->td_pfd
;
219 switch (resume_skip_check(td
, dnp
, zb
)) {
220 case RESUME_SKIP_ALL
:
222 case RESUME_SKIP_CHILDREN
:
224 case RESUME_SKIP_NONE
:
230 if (bp
->blk_birth
== 0) {
232 * Since this block has a birth time of 0 it must be one of
233 * two things: a hole created before the
234 * SPA_FEATURE_HOLE_BIRTH feature was enabled, or a hole
235 * which has always been a hole in an object.
237 * If a file is written sparsely, then the unwritten parts of
238 * the file were "always holes" -- that is, they have been
239 * holes since this object was allocated. However, we (and
240 * our callers) can not necessarily tell when an object was
241 * allocated. Therefore, if it's possible that this object
242 * was freed and then its object number reused, we need to
243 * visit all the holes with birth==0.
245 * If it isn't possible that the object number was reused,
246 * then if SPA_FEATURE_HOLE_BIRTH was enabled before we wrote
247 * all the blocks we will visit as part of this traversal,
248 * then this hole must have always existed, so we can skip
249 * it. We visit blocks born after (exclusive) td_min_txg.
251 * Note that the meta-dnode cannot be reallocated.
253 if ((!td
->td_realloc_possible
||
254 zb
->zb_object
== DMU_META_DNODE_OBJECT
) &&
255 td
->td_hole_birth_enabled_txg
<= td
->td_min_txg
)
257 } else if (bp
->blk_birth
<= td
->td_min_txg
) {
261 if (pd
!= NULL
&& !pd
->pd_exited
&& prefetch_needed(pd
, bp
)) {
262 uint64_t size
= BP_GET_LSIZE(bp
);
263 mutex_enter(&pd
->pd_mtx
);
264 ASSERT(pd
->pd_bytes_fetched
>= 0);
265 while (pd
->pd_bytes_fetched
< size
&& !pd
->pd_exited
)
266 cv_wait_sig(&pd
->pd_cv
, &pd
->pd_mtx
);
267 pd
->pd_bytes_fetched
-= size
;
268 cv_broadcast(&pd
->pd_cv
);
269 mutex_exit(&pd
->pd_mtx
);
272 if (BP_IS_HOLE(bp
)) {
273 err
= td
->td_func(td
->td_spa
, NULL
, bp
, zb
, dnp
, td
->td_arg
);
279 if (td
->td_flags
& TRAVERSE_PRE
) {
280 err
= td
->td_func(td
->td_spa
, NULL
, bp
, zb
, dnp
,
282 if (err
== TRAVERSE_VISIT_NO_CHILDREN
)
288 if (BP_GET_LEVEL(bp
) > 0) {
289 uint32_t flags
= ARC_FLAG_WAIT
;
291 int32_t epb
= BP_GET_LSIZE(bp
) >> SPA_BLKPTRSHIFT
;
292 zbookmark_phys_t
*czb
;
294 err
= arc_read(NULL
, td
->td_spa
, bp
, arc_getbuf_func
, &buf
,
295 ZIO_PRIORITY_ASYNC_READ
, ZIO_FLAG_CANFAIL
, &flags
, zb
);
299 czb
= kmem_alloc(sizeof (zbookmark_phys_t
), KM_SLEEP
);
301 for (i
= 0; i
< epb
; i
++) {
302 SET_BOOKMARK(czb
, zb
->zb_objset
, zb
->zb_object
,
304 zb
->zb_blkid
* epb
+ i
);
305 traverse_prefetch_metadata(td
,
306 &((blkptr_t
*)buf
->b_data
)[i
], czb
);
309 /* recursively visitbp() blocks below this */
310 for (i
= 0; i
< epb
; i
++) {
311 SET_BOOKMARK(czb
, zb
->zb_objset
, zb
->zb_object
,
313 zb
->zb_blkid
* epb
+ i
);
314 err
= traverse_visitbp(td
, dnp
,
315 &((blkptr_t
*)buf
->b_data
)[i
], czb
);
320 kmem_free(czb
, sizeof (zbookmark_phys_t
));
322 } else if (BP_GET_TYPE(bp
) == DMU_OT_DNODE
) {
323 uint32_t flags
= ARC_FLAG_WAIT
;
325 int32_t epb
= BP_GET_LSIZE(bp
) >> DNODE_SHIFT
;
328 err
= arc_read(NULL
, td
->td_spa
, bp
, arc_getbuf_func
, &buf
,
329 ZIO_PRIORITY_ASYNC_READ
, ZIO_FLAG_CANFAIL
, &flags
, zb
);
334 for (i
= 0; i
< epb
; i
++) {
335 prefetch_dnode_metadata(td
, &cdnp
[i
], zb
->zb_objset
,
336 zb
->zb_blkid
* epb
+ i
);
339 /* recursively visitbp() blocks below this */
340 for (i
= 0; i
< epb
; i
++) {
341 err
= traverse_dnode(td
, &cdnp
[i
], zb
->zb_objset
,
342 zb
->zb_blkid
* epb
+ i
);
346 } else if (BP_GET_TYPE(bp
) == DMU_OT_OBJSET
) {
347 arc_flags_t flags
= ARC_FLAG_WAIT
;
349 dnode_phys_t
*mdnp
, *gdnp
, *udnp
;
351 err
= arc_read(NULL
, td
->td_spa
, bp
, arc_getbuf_func
, &buf
,
352 ZIO_PRIORITY_ASYNC_READ
, ZIO_FLAG_CANFAIL
, &flags
, zb
);
357 mdnp
= &osp
->os_meta_dnode
;
358 gdnp
= &osp
->os_groupused_dnode
;
359 udnp
= &osp
->os_userused_dnode
;
361 prefetch_dnode_metadata(td
, mdnp
, zb
->zb_objset
,
362 DMU_META_DNODE_OBJECT
);
364 * See the block comment above for the goal of this variable.
365 * If the maxblkid of the meta-dnode is 0, then we know that
366 * we've never had more than DNODES_PER_BLOCK objects in the
367 * dataset, which means we can't have reused any object ids.
369 if (osp
->os_meta_dnode
.dn_maxblkid
== 0)
370 td
->td_realloc_possible
= B_FALSE
;
372 if (arc_buf_size(buf
) >= sizeof (objset_phys_t
)) {
373 prefetch_dnode_metadata(td
, gdnp
, zb
->zb_objset
,
374 DMU_GROUPUSED_OBJECT
);
375 prefetch_dnode_metadata(td
, udnp
, zb
->zb_objset
,
376 DMU_USERUSED_OBJECT
);
379 err
= traverse_dnode(td
, mdnp
, zb
->zb_objset
,
380 DMU_META_DNODE_OBJECT
);
381 if (err
== 0 && arc_buf_size(buf
) >= sizeof (objset_phys_t
)) {
382 err
= traverse_dnode(td
, gdnp
, zb
->zb_objset
,
383 DMU_GROUPUSED_OBJECT
);
385 if (err
== 0 && arc_buf_size(buf
) >= sizeof (objset_phys_t
)) {
386 err
= traverse_dnode(td
, udnp
, zb
->zb_objset
,
387 DMU_USERUSED_OBJECT
);
392 (void) arc_buf_remove_ref(buf
, &buf
);
395 if (err
== 0 && (td
->td_flags
& TRAVERSE_POST
))
396 err
= td
->td_func(td
->td_spa
, NULL
, bp
, zb
, dnp
, td
->td_arg
);
398 if ((td
->td_flags
& TRAVERSE_HARD
) && (err
== EIO
|| err
== ECKSUM
)) {
400 * Ignore this disk error as requested by the HARD flag,
401 * and continue traversal.
407 * If we are stopping here, set td_resume.
409 if (td
->td_resume
!= NULL
&& err
!= 0 && !td
->td_paused
) {
410 td
->td_resume
->zb_objset
= zb
->zb_objset
;
411 td
->td_resume
->zb_object
= zb
->zb_object
;
412 td
->td_resume
->zb_level
= 0;
414 * If we have stopped on an indirect block (e.g. due to
415 * i/o error), we have not visited anything below it.
416 * Set the bookmark to the first level-0 block that we need
417 * to visit. This way, the resuming code does not need to
418 * deal with resuming from indirect blocks.
420 td
->td_resume
->zb_blkid
= zb
->zb_blkid
<<
421 (zb
->zb_level
* (dnp
->dn_indblkshift
- SPA_BLKPTRSHIFT
));
422 td
->td_paused
= B_TRUE
;
429 prefetch_dnode_metadata(traverse_data_t
*td
, const dnode_phys_t
*dnp
,
430 uint64_t objset
, uint64_t object
)
433 zbookmark_phys_t czb
;
435 for (j
= 0; j
< dnp
->dn_nblkptr
; j
++) {
436 SET_BOOKMARK(&czb
, objset
, object
, dnp
->dn_nlevels
- 1, j
);
437 traverse_prefetch_metadata(td
, &dnp
->dn_blkptr
[j
], &czb
);
440 if (dnp
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
) {
441 SET_BOOKMARK(&czb
, objset
, object
, 0, DMU_SPILL_BLKID
);
442 traverse_prefetch_metadata(td
, &dnp
->dn_spill
, &czb
);
447 traverse_dnode(traverse_data_t
*td
, const dnode_phys_t
*dnp
,
448 uint64_t objset
, uint64_t object
)
451 zbookmark_phys_t czb
;
453 if (td
->td_flags
& TRAVERSE_PRE
) {
454 SET_BOOKMARK(&czb
, objset
, object
, ZB_DNODE_LEVEL
,
456 err
= td
->td_func(td
->td_spa
, NULL
, NULL
, &czb
, dnp
,
458 if (err
== TRAVERSE_VISIT_NO_CHILDREN
)
464 for (j
= 0; j
< dnp
->dn_nblkptr
; j
++) {
465 SET_BOOKMARK(&czb
, objset
, object
, dnp
->dn_nlevels
- 1, j
);
466 err
= traverse_visitbp(td
, dnp
, &dnp
->dn_blkptr
[j
], &czb
);
471 if (err
== 0 && (dnp
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
)) {
472 SET_BOOKMARK(&czb
, objset
, object
, 0, DMU_SPILL_BLKID
);
473 err
= traverse_visitbp(td
, dnp
, &dnp
->dn_spill
, &czb
);
476 if (err
== 0 && (td
->td_flags
& TRAVERSE_POST
)) {
477 SET_BOOKMARK(&czb
, objset
, object
, ZB_DNODE_LEVEL
,
479 err
= td
->td_func(td
->td_spa
, NULL
, NULL
, &czb
, dnp
,
481 if (err
== TRAVERSE_VISIT_NO_CHILDREN
)
491 traverse_prefetcher(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
492 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
494 prefetch_data_t
*pfd
= arg
;
495 arc_flags_t aflags
= ARC_FLAG_NOWAIT
| ARC_FLAG_PREFETCH
;
497 ASSERT(pfd
->pd_bytes_fetched
>= 0);
501 return (SET_ERROR(EINTR
));
503 if (!prefetch_needed(pfd
, bp
))
506 mutex_enter(&pfd
->pd_mtx
);
507 while (!pfd
->pd_cancel
&& pfd
->pd_bytes_fetched
>= zfs_pd_bytes_max
)
508 cv_wait_sig(&pfd
->pd_cv
, &pfd
->pd_mtx
);
509 pfd
->pd_bytes_fetched
+= BP_GET_LSIZE(bp
);
510 cv_broadcast(&pfd
->pd_cv
);
511 mutex_exit(&pfd
->pd_mtx
);
513 (void) arc_read(NULL
, spa
, bp
, NULL
, NULL
, ZIO_PRIORITY_ASYNC_READ
,
514 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
, &aflags
, zb
);
520 traverse_prefetch_thread(void *arg
)
522 traverse_data_t
*td_main
= arg
;
523 traverse_data_t td
= *td_main
;
524 zbookmark_phys_t czb
;
525 fstrans_cookie_t cookie
= spl_fstrans_mark();
527 td
.td_func
= traverse_prefetcher
;
528 td
.td_arg
= td_main
->td_pfd
;
531 SET_BOOKMARK(&czb
, td
.td_objset
,
532 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
533 (void) traverse_visitbp(&td
, NULL
, td
.td_rootbp
, &czb
);
535 mutex_enter(&td_main
->td_pfd
->pd_mtx
);
536 td_main
->td_pfd
->pd_exited
= B_TRUE
;
537 cv_broadcast(&td_main
->td_pfd
->pd_cv
);
538 mutex_exit(&td_main
->td_pfd
->pd_mtx
);
539 spl_fstrans_unmark(cookie
);
543 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
544 * in syncing context).
547 traverse_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
)
553 zbookmark_phys_t
*czb
;
556 ASSERT(ds
== NULL
|| objset
== ds
->ds_object
);
557 ASSERT(!(flags
& TRAVERSE_PRE
) || !(flags
& TRAVERSE_POST
));
560 * The data prefetching mechanism (the prefetch thread) is incompatible
561 * with resuming from a bookmark.
563 ASSERT(resume
== NULL
|| !(flags
& TRAVERSE_PREFETCH_DATA
));
565 td
= kmem_alloc(sizeof (traverse_data_t
), KM_SLEEP
);
566 pd
= kmem_zalloc(sizeof (prefetch_data_t
), KM_SLEEP
);
567 czb
= kmem_alloc(sizeof (zbookmark_phys_t
), KM_SLEEP
);
570 td
->td_objset
= objset
;
571 td
->td_rootbp
= rootbp
;
572 td
->td_min_txg
= txg_start
;
573 td
->td_resume
= resume
;
577 td
->td_flags
= flags
;
578 td
->td_paused
= B_FALSE
;
579 td
->td_realloc_possible
= (txg_start
== 0 ? B_FALSE
: B_TRUE
);
581 if (spa_feature_is_active(spa
, SPA_FEATURE_HOLE_BIRTH
)) {
582 VERIFY(spa_feature_enabled_txg(spa
,
583 SPA_FEATURE_HOLE_BIRTH
, &td
->td_hole_birth_enabled_txg
));
585 td
->td_hole_birth_enabled_txg
= UINT64_MAX
;
588 pd
->pd_flags
= flags
;
589 mutex_init(&pd
->pd_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
590 cv_init(&pd
->pd_cv
, NULL
, CV_DEFAULT
, NULL
);
592 SET_BOOKMARK(czb
, td
->td_objset
,
593 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
595 /* See comment on ZIL traversal in dsl_scan_visitds. */
596 if (ds
!= NULL
&& !ds
->ds_is_snapshot
&& !BP_IS_HOLE(rootbp
)) {
597 uint32_t flags
= ARC_FLAG_WAIT
;
601 err
= arc_read(NULL
, td
->td_spa
, rootbp
,
602 arc_getbuf_func
, &buf
,
603 ZIO_PRIORITY_ASYNC_READ
, ZIO_FLAG_CANFAIL
, &flags
, czb
);
608 traverse_zil(td
, &osp
->os_zil_header
);
609 (void) arc_buf_remove_ref(buf
, &buf
);
612 if (!(flags
& TRAVERSE_PREFETCH_DATA
) ||
613 0 == taskq_dispatch(system_taskq
, traverse_prefetch_thread
,
615 pd
->pd_exited
= B_TRUE
;
617 err
= traverse_visitbp(td
, NULL
, rootbp
, czb
);
619 mutex_enter(&pd
->pd_mtx
);
620 pd
->pd_cancel
= B_TRUE
;
621 cv_broadcast(&pd
->pd_cv
);
622 while (!pd
->pd_exited
)
623 cv_wait_sig(&pd
->pd_cv
, &pd
->pd_mtx
);
624 mutex_exit(&pd
->pd_mtx
);
626 mutex_destroy(&pd
->pd_mtx
);
627 cv_destroy(&pd
->pd_cv
);
629 kmem_free(czb
, sizeof (zbookmark_phys_t
));
630 kmem_free(pd
, sizeof (struct prefetch_data
));
631 kmem_free(td
, sizeof (struct traverse_data
));
637 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
638 * in syncing context).
641 traverse_dataset(dsl_dataset_t
*ds
, uint64_t txg_start
, int flags
,
642 blkptr_cb_t func
, void *arg
)
644 return (traverse_impl(ds
->ds_dir
->dd_pool
->dp_spa
, ds
, ds
->ds_object
,
645 &dsl_dataset_phys(ds
)->ds_bp
, txg_start
, NULL
, flags
, func
, arg
));
649 traverse_dataset_destroyed(spa_t
*spa
, blkptr_t
*blkptr
,
650 uint64_t txg_start
, zbookmark_phys_t
*resume
, int flags
,
651 blkptr_cb_t func
, void *arg
)
653 return (traverse_impl(spa
, NULL
, ZB_DESTROYED_OBJSET
,
654 blkptr
, txg_start
, resume
, flags
, func
, arg
));
658 * NB: pool must not be changing on-disk (eg, from zdb or sync context).
661 traverse_pool(spa_t
*spa
, uint64_t txg_start
, int flags
,
662 blkptr_cb_t func
, void *arg
)
666 dsl_pool_t
*dp
= spa_get_dsl(spa
);
667 objset_t
*mos
= dp
->dp_meta_objset
;
668 boolean_t hard
= (flags
& TRAVERSE_HARD
);
671 err
= traverse_impl(spa
, NULL
, 0, spa_get_rootblkptr(spa
),
672 txg_start
, NULL
, flags
, func
, arg
);
676 /* visit each dataset */
677 for (obj
= 1; err
== 0;
678 err
= dmu_object_next(mos
, &obj
, FALSE
, txg_start
)) {
679 dmu_object_info_t doi
;
681 err
= dmu_object_info(mos
, obj
, &doi
);
688 if (doi
.doi_bonus_type
== DMU_OT_DSL_DATASET
) {
690 uint64_t txg
= txg_start
;
692 dsl_pool_config_enter(dp
, FTAG
);
693 err
= dsl_dataset_hold_obj(dp
, obj
, FTAG
, &ds
);
694 dsl_pool_config_exit(dp
, FTAG
);
700 if (dsl_dataset_phys(ds
)->ds_prev_snap_txg
> txg
)
701 txg
= dsl_dataset_phys(ds
)->ds_prev_snap_txg
;
702 err
= traverse_dataset(ds
, txg
, flags
, func
, arg
);
703 dsl_dataset_rele(ds
, FTAG
);
713 #if defined(_KERNEL) && defined(HAVE_SPL)
714 EXPORT_SYMBOL(traverse_dataset
);
715 EXPORT_SYMBOL(traverse_pool
);
717 module_param(zfs_pd_bytes_max
, int, 0644);
718 MODULE_PARM_DESC(zfs_pd_bytes_max
, "Max number of bytes to prefetch");