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 */
42 int32_t ignore_hole_birth
= 1;
44 typedef struct prefetch_data
{
47 int32_t pd_bytes_fetched
;
51 zbookmark_phys_t pd_resume
;
54 typedef struct traverse_data
{
59 zbookmark_phys_t
*td_resume
;
61 prefetch_data_t
*td_pfd
;
63 uint64_t td_hole_birth_enabled_txg
;
66 boolean_t td_realloc_possible
;
69 static int traverse_dnode(traverse_data_t
*td
, const dnode_phys_t
*dnp
,
70 uint64_t objset
, uint64_t object
);
71 static void prefetch_dnode_metadata(traverse_data_t
*td
, const dnode_phys_t
*,
72 uint64_t objset
, uint64_t object
);
75 traverse_zil_block(zilog_t
*zilog
, blkptr_t
*bp
, void *arg
, uint64_t claim_txg
)
77 traverse_data_t
*td
= arg
;
83 if (claim_txg
== 0 && bp
->blk_birth
>= spa_first_txg(td
->td_spa
))
86 SET_BOOKMARK(&zb
, td
->td_objset
, ZB_ZIL_OBJECT
, ZB_ZIL_LEVEL
,
87 bp
->blk_cksum
.zc_word
[ZIL_ZC_SEQ
]);
89 (void) td
->td_func(td
->td_spa
, zilog
, bp
, &zb
, NULL
, td
->td_arg
);
95 traverse_zil_record(zilog_t
*zilog
, lr_t
*lrc
, void *arg
, uint64_t claim_txg
)
97 traverse_data_t
*td
= arg
;
99 if (lrc
->lrc_txtype
== TX_WRITE
) {
100 lr_write_t
*lr
= (lr_write_t
*)lrc
;
101 blkptr_t
*bp
= &lr
->lr_blkptr
;
107 if (claim_txg
== 0 || bp
->blk_birth
< claim_txg
)
110 SET_BOOKMARK(&zb
, td
->td_objset
, lr
->lr_foid
,
111 ZB_ZIL_LEVEL
, lr
->lr_offset
/ BP_GET_LSIZE(bp
));
113 (void) td
->td_func(td
->td_spa
, zilog
, bp
, &zb
, NULL
,
120 traverse_zil(traverse_data_t
*td
, zil_header_t
*zh
)
122 uint64_t claim_txg
= zh
->zh_claim_txg
;
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.
129 if (claim_txg
== 0 && spa_writeable(td
->td_spa
))
132 zilog
= zil_alloc(spa_get_dsl(td
->td_spa
)->dp_meta_objset
, zh
);
134 (void) zil_parse(zilog
, traverse_zil_block
, traverse_zil_record
, td
,
140 typedef enum resume_skip
{
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.
155 resume_skip_check(traverse_data_t
*td
, const dnode_phys_t
*dnp
,
156 const zbookmark_phys_t
*zb
)
158 if (td
->td_resume
!= NULL
&& !ZB_IS_ZERO(td
->td_resume
)) {
160 * If we already visited this bp & everything below,
161 * don't bother doing it again.
163 if (zbookmark_subtree_completed(dnp
, zb
, td
->td_resume
))
164 return (RESUME_SKIP_ALL
);
167 * If we found the block we're trying to resume from, zero
168 * the bookmark out to indicate that we have resumed.
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
);
176 return (RESUME_SKIP_NONE
);
180 traverse_prefetch_metadata(traverse_data_t
*td
,
181 const blkptr_t
*bp
, const zbookmark_phys_t
*zb
)
183 arc_flags_t flags
= ARC_FLAG_NOWAIT
| ARC_FLAG_PREFETCH
;
185 if (!(td
->td_flags
& TRAVERSE_PREFETCH_METADATA
))
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
192 if (td
->td_resume
!= NULL
&& !ZB_IS_ZERO(td
->td_resume
))
194 if (BP_IS_HOLE(bp
) || bp
->blk_birth
<= td
->td_min_txg
)
196 if (BP_GET_LEVEL(bp
) == 0 && BP_GET_TYPE(bp
) != DMU_OT_DNODE
)
199 (void) arc_read(NULL
, td
->td_spa
, bp
, NULL
, NULL
,
200 ZIO_PRIORITY_ASYNC_READ
, ZIO_FLAG_CANFAIL
, &flags
, zb
);
204 prefetch_needed(prefetch_data_t
*pfd
, const blkptr_t
*bp
)
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
)
214 traverse_visitbp(traverse_data_t
*td
, const dnode_phys_t
*dnp
,
215 const blkptr_t
*bp
, const zbookmark_phys_t
*zb
)
218 arc_buf_t
*buf
= NULL
;
219 prefetch_data_t
*pd
= td
->td_pfd
;
221 switch (resume_skip_check(td
, dnp
, zb
)) {
222 case RESUME_SKIP_ALL
:
224 case RESUME_SKIP_CHILDREN
:
226 case RESUME_SKIP_NONE
:
232 if (bp
->blk_birth
== 0) {
234 * Since this block has a birth time of 0 it must be one of
235 * two things: a hole created before the
236 * SPA_FEATURE_HOLE_BIRTH feature was enabled, or a hole
237 * which has always been a hole in an object.
239 * If a file is written sparsely, then the unwritten parts of
240 * the file were "always holes" -- that is, they have been
241 * holes since this object was allocated. However, we (and
242 * our callers) can not necessarily tell when an object was
243 * allocated. Therefore, if it's possible that this object
244 * was freed and then its object number reused, we need to
245 * visit all the holes with birth==0.
247 * If it isn't possible that the object number was reused,
248 * then if SPA_FEATURE_HOLE_BIRTH was enabled before we wrote
249 * all the blocks we will visit as part of this traversal,
250 * then this hole must have always existed, so we can skip
251 * it. We visit blocks born after (exclusive) td_min_txg.
253 * Note that the meta-dnode cannot be reallocated.
255 if (!ignore_hole_birth
&& (!td
->td_realloc_possible
||
256 zb
->zb_object
== DMU_META_DNODE_OBJECT
) &&
257 td
->td_hole_birth_enabled_txg
<= td
->td_min_txg
)
259 } else if (bp
->blk_birth
<= td
->td_min_txg
) {
263 if (pd
!= NULL
&& !pd
->pd_exited
&& prefetch_needed(pd
, bp
)) {
264 uint64_t size
= BP_GET_LSIZE(bp
);
265 mutex_enter(&pd
->pd_mtx
);
266 ASSERT(pd
->pd_bytes_fetched
>= 0);
267 while (pd
->pd_bytes_fetched
< size
&& !pd
->pd_exited
)
268 cv_wait_sig(&pd
->pd_cv
, &pd
->pd_mtx
);
269 pd
->pd_bytes_fetched
-= size
;
270 cv_broadcast(&pd
->pd_cv
);
271 mutex_exit(&pd
->pd_mtx
);
274 if (BP_IS_HOLE(bp
)) {
275 err
= td
->td_func(td
->td_spa
, NULL
, bp
, zb
, dnp
, td
->td_arg
);
281 if (td
->td_flags
& TRAVERSE_PRE
) {
282 err
= td
->td_func(td
->td_spa
, NULL
, bp
, zb
, dnp
,
284 if (err
== TRAVERSE_VISIT_NO_CHILDREN
)
290 if (BP_GET_LEVEL(bp
) > 0) {
291 uint32_t flags
= ARC_FLAG_WAIT
;
293 int32_t epb
= BP_GET_LSIZE(bp
) >> SPA_BLKPTRSHIFT
;
294 zbookmark_phys_t
*czb
;
296 err
= arc_read(NULL
, td
->td_spa
, bp
, arc_getbuf_func
, &buf
,
297 ZIO_PRIORITY_ASYNC_READ
, ZIO_FLAG_CANFAIL
, &flags
, zb
);
301 czb
= kmem_alloc(sizeof (zbookmark_phys_t
), KM_SLEEP
);
303 for (i
= 0; i
< epb
; i
++) {
304 SET_BOOKMARK(czb
, zb
->zb_objset
, zb
->zb_object
,
306 zb
->zb_blkid
* epb
+ i
);
307 traverse_prefetch_metadata(td
,
308 &((blkptr_t
*)buf
->b_data
)[i
], czb
);
311 /* recursively visitbp() blocks below this */
312 for (i
= 0; i
< epb
; i
++) {
313 SET_BOOKMARK(czb
, zb
->zb_objset
, zb
->zb_object
,
315 zb
->zb_blkid
* epb
+ i
);
316 err
= traverse_visitbp(td
, dnp
,
317 &((blkptr_t
*)buf
->b_data
)[i
], czb
);
322 kmem_free(czb
, sizeof (zbookmark_phys_t
));
324 } else if (BP_GET_TYPE(bp
) == DMU_OT_DNODE
) {
325 uint32_t flags
= ARC_FLAG_WAIT
;
327 int32_t epb
= BP_GET_LSIZE(bp
) >> DNODE_SHIFT
;
328 dnode_phys_t
*child_dnp
;
330 err
= arc_read(NULL
, td
->td_spa
, bp
, arc_getbuf_func
, &buf
,
331 ZIO_PRIORITY_ASYNC_READ
, ZIO_FLAG_CANFAIL
, &flags
, zb
);
334 child_dnp
= buf
->b_data
;
336 for (i
= 0; i
< epb
; i
+= child_dnp
[i
].dn_extra_slots
+ 1) {
337 prefetch_dnode_metadata(td
, &child_dnp
[i
],
338 zb
->zb_objset
, zb
->zb_blkid
* epb
+ i
);
341 /* recursively visitbp() blocks below this */
342 for (i
= 0; i
< epb
; i
+= child_dnp
[i
].dn_extra_slots
+ 1) {
343 err
= traverse_dnode(td
, &child_dnp
[i
],
344 zb
->zb_objset
, zb
->zb_blkid
* epb
+ i
);
348 } else if (BP_GET_TYPE(bp
) == DMU_OT_OBJSET
) {
349 arc_flags_t flags
= ARC_FLAG_WAIT
;
352 err
= arc_read(NULL
, td
->td_spa
, bp
, arc_getbuf_func
, &buf
,
353 ZIO_PRIORITY_ASYNC_READ
, ZIO_FLAG_CANFAIL
, &flags
, zb
);
358 prefetch_dnode_metadata(td
, &osp
->os_meta_dnode
, zb
->zb_objset
,
359 DMU_META_DNODE_OBJECT
);
361 * See the block comment above for the goal of this variable.
362 * If the maxblkid of the meta-dnode is 0, then we know that
363 * we've never had more than DNODES_PER_BLOCK objects in the
364 * dataset, which means we can't have reused any object ids.
366 if (osp
->os_meta_dnode
.dn_maxblkid
== 0)
367 td
->td_realloc_possible
= B_FALSE
;
369 if (arc_buf_size(buf
) >= sizeof (objset_phys_t
)) {
370 prefetch_dnode_metadata(td
, &osp
->os_groupused_dnode
,
371 zb
->zb_objset
, DMU_GROUPUSED_OBJECT
);
372 prefetch_dnode_metadata(td
, &osp
->os_userused_dnode
,
373 zb
->zb_objset
, DMU_USERUSED_OBJECT
);
376 err
= traverse_dnode(td
, &osp
->os_meta_dnode
, zb
->zb_objset
,
377 DMU_META_DNODE_OBJECT
);
378 if (err
== 0 && arc_buf_size(buf
) >= sizeof (objset_phys_t
)) {
379 err
= traverse_dnode(td
, &osp
->os_groupused_dnode
,
380 zb
->zb_objset
, DMU_GROUPUSED_OBJECT
);
382 if (err
== 0 && arc_buf_size(buf
) >= sizeof (objset_phys_t
)) {
383 err
= traverse_dnode(td
, &osp
->os_userused_dnode
,
384 zb
->zb_objset
, DMU_USERUSED_OBJECT
);
389 arc_buf_destroy(buf
, &buf
);
392 if (err
== 0 && (td
->td_flags
& TRAVERSE_POST
))
393 err
= td
->td_func(td
->td_spa
, NULL
, bp
, zb
, dnp
, td
->td_arg
);
395 if ((td
->td_flags
& TRAVERSE_HARD
) && (err
== EIO
|| err
== ECKSUM
)) {
397 * Ignore this disk error as requested by the HARD flag,
398 * and continue traversal.
404 * If we are stopping here, set td_resume.
406 if (td
->td_resume
!= NULL
&& err
!= 0 && !td
->td_paused
) {
407 td
->td_resume
->zb_objset
= zb
->zb_objset
;
408 td
->td_resume
->zb_object
= zb
->zb_object
;
409 td
->td_resume
->zb_level
= 0;
411 * If we have stopped on an indirect block (e.g. due to
412 * i/o error), we have not visited anything below it.
413 * Set the bookmark to the first level-0 block that we need
414 * to visit. This way, the resuming code does not need to
415 * deal with resuming from indirect blocks.
417 * Note, if zb_level <= 0, dnp may be NULL, so we don't want
420 td
->td_resume
->zb_blkid
= zb
->zb_blkid
;
421 if (zb
->zb_level
> 0) {
422 td
->td_resume
->zb_blkid
<<= zb
->zb_level
*
423 (dnp
->dn_indblkshift
- SPA_BLKPTRSHIFT
);
425 td
->td_paused
= B_TRUE
;
432 prefetch_dnode_metadata(traverse_data_t
*td
, const dnode_phys_t
*dnp
,
433 uint64_t objset
, uint64_t object
)
436 zbookmark_phys_t czb
;
438 for (j
= 0; j
< dnp
->dn_nblkptr
; j
++) {
439 SET_BOOKMARK(&czb
, objset
, object
, dnp
->dn_nlevels
- 1, j
);
440 traverse_prefetch_metadata(td
, &dnp
->dn_blkptr
[j
], &czb
);
443 if (dnp
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
) {
444 SET_BOOKMARK(&czb
, objset
, object
, 0, DMU_SPILL_BLKID
);
445 traverse_prefetch_metadata(td
, DN_SPILL_BLKPTR(dnp
), &czb
);
450 traverse_dnode(traverse_data_t
*td
, const dnode_phys_t
*dnp
,
451 uint64_t objset
, uint64_t object
)
454 zbookmark_phys_t czb
;
456 if (object
!= DMU_META_DNODE_OBJECT
&& td
->td_resume
!= NULL
&&
457 object
< td
->td_resume
->zb_object
)
460 if (td
->td_flags
& TRAVERSE_PRE
) {
461 SET_BOOKMARK(&czb
, objset
, object
, ZB_DNODE_LEVEL
,
463 err
= td
->td_func(td
->td_spa
, NULL
, NULL
, &czb
, dnp
,
465 if (err
== TRAVERSE_VISIT_NO_CHILDREN
)
471 for (j
= 0; j
< dnp
->dn_nblkptr
; j
++) {
472 SET_BOOKMARK(&czb
, objset
, object
, dnp
->dn_nlevels
- 1, j
);
473 err
= traverse_visitbp(td
, dnp
, &dnp
->dn_blkptr
[j
], &czb
);
478 if (err
== 0 && (dnp
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
)) {
479 SET_BOOKMARK(&czb
, objset
, object
, 0, DMU_SPILL_BLKID
);
480 err
= traverse_visitbp(td
, dnp
, DN_SPILL_BLKPTR(dnp
), &czb
);
483 if (err
== 0 && (td
->td_flags
& TRAVERSE_POST
)) {
484 SET_BOOKMARK(&czb
, objset
, object
, ZB_DNODE_LEVEL
,
486 err
= td
->td_func(td
->td_spa
, NULL
, NULL
, &czb
, dnp
,
488 if (err
== TRAVERSE_VISIT_NO_CHILDREN
)
498 traverse_prefetcher(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
499 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
501 prefetch_data_t
*pfd
= arg
;
502 arc_flags_t aflags
= ARC_FLAG_NOWAIT
| ARC_FLAG_PREFETCH
;
504 ASSERT(pfd
->pd_bytes_fetched
>= 0);
508 return (SET_ERROR(EINTR
));
510 if (!prefetch_needed(pfd
, bp
))
513 mutex_enter(&pfd
->pd_mtx
);
514 while (!pfd
->pd_cancel
&& pfd
->pd_bytes_fetched
>= zfs_pd_bytes_max
)
515 cv_wait_sig(&pfd
->pd_cv
, &pfd
->pd_mtx
);
516 pfd
->pd_bytes_fetched
+= BP_GET_LSIZE(bp
);
517 cv_broadcast(&pfd
->pd_cv
);
518 mutex_exit(&pfd
->pd_mtx
);
520 (void) arc_read(NULL
, spa
, bp
, NULL
, NULL
, ZIO_PRIORITY_ASYNC_READ
,
521 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
, &aflags
, zb
);
527 traverse_prefetch_thread(void *arg
)
529 traverse_data_t
*td_main
= arg
;
530 traverse_data_t td
= *td_main
;
531 zbookmark_phys_t czb
;
532 fstrans_cookie_t cookie
= spl_fstrans_mark();
534 td
.td_func
= traverse_prefetcher
;
535 td
.td_arg
= td_main
->td_pfd
;
537 td
.td_resume
= &td_main
->td_pfd
->pd_resume
;
539 SET_BOOKMARK(&czb
, td
.td_objset
,
540 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
541 (void) traverse_visitbp(&td
, NULL
, td
.td_rootbp
, &czb
);
543 mutex_enter(&td_main
->td_pfd
->pd_mtx
);
544 td_main
->td_pfd
->pd_exited
= B_TRUE
;
545 cv_broadcast(&td_main
->td_pfd
->pd_cv
);
546 mutex_exit(&td_main
->td_pfd
->pd_mtx
);
547 spl_fstrans_unmark(cookie
);
551 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
552 * in syncing context).
555 traverse_impl(spa_t
*spa
, dsl_dataset_t
*ds
, uint64_t objset
, blkptr_t
*rootbp
,
556 uint64_t txg_start
, zbookmark_phys_t
*resume
, int flags
,
557 blkptr_cb_t func
, void *arg
)
561 zbookmark_phys_t
*czb
;
564 ASSERT(ds
== NULL
|| objset
== ds
->ds_object
);
565 ASSERT(!(flags
& TRAVERSE_PRE
) || !(flags
& TRAVERSE_POST
));
567 td
= kmem_alloc(sizeof (traverse_data_t
), KM_SLEEP
);
568 pd
= kmem_zalloc(sizeof (prefetch_data_t
), KM_SLEEP
);
569 czb
= kmem_alloc(sizeof (zbookmark_phys_t
), KM_SLEEP
);
572 td
->td_objset
= objset
;
573 td
->td_rootbp
= rootbp
;
574 td
->td_min_txg
= txg_start
;
575 td
->td_resume
= resume
;
579 td
->td_flags
= flags
;
580 td
->td_paused
= B_FALSE
;
581 td
->td_realloc_possible
= (txg_start
== 0 ? B_FALSE
: B_TRUE
);
583 if (spa_feature_is_active(spa
, SPA_FEATURE_HOLE_BIRTH
)) {
584 VERIFY(spa_feature_enabled_txg(spa
,
585 SPA_FEATURE_HOLE_BIRTH
, &td
->td_hole_birth_enabled_txg
));
587 td
->td_hole_birth_enabled_txg
= UINT64_MAX
;
590 pd
->pd_flags
= flags
;
592 pd
->pd_resume
= *resume
;
593 mutex_init(&pd
->pd_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
594 cv_init(&pd
->pd_cv
, NULL
, CV_DEFAULT
, NULL
);
596 SET_BOOKMARK(czb
, td
->td_objset
,
597 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
599 /* See comment on ZIL traversal in dsl_scan_visitds. */
600 if (ds
!= NULL
&& !ds
->ds_is_snapshot
&& !BP_IS_HOLE(rootbp
)) {
601 uint32_t flags
= ARC_FLAG_WAIT
;
605 err
= arc_read(NULL
, td
->td_spa
, rootbp
,
606 arc_getbuf_func
, &buf
,
607 ZIO_PRIORITY_ASYNC_READ
, ZIO_FLAG_CANFAIL
, &flags
, czb
);
612 traverse_zil(td
, &osp
->os_zil_header
);
613 arc_buf_destroy(buf
, &buf
);
616 if (!(flags
& TRAVERSE_PREFETCH_DATA
) ||
617 0 == taskq_dispatch(system_taskq
, traverse_prefetch_thread
,
619 pd
->pd_exited
= B_TRUE
;
621 err
= traverse_visitbp(td
, NULL
, rootbp
, czb
);
623 mutex_enter(&pd
->pd_mtx
);
624 pd
->pd_cancel
= B_TRUE
;
625 cv_broadcast(&pd
->pd_cv
);
626 while (!pd
->pd_exited
)
627 cv_wait_sig(&pd
->pd_cv
, &pd
->pd_mtx
);
628 mutex_exit(&pd
->pd_mtx
);
630 mutex_destroy(&pd
->pd_mtx
);
631 cv_destroy(&pd
->pd_cv
);
633 kmem_free(czb
, sizeof (zbookmark_phys_t
));
634 kmem_free(pd
, sizeof (struct prefetch_data
));
635 kmem_free(td
, sizeof (struct traverse_data
));
641 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
642 * in syncing context).
645 traverse_dataset_resume(dsl_dataset_t
*ds
, uint64_t txg_start
,
646 zbookmark_phys_t
*resume
,
647 int flags
, blkptr_cb_t func
, void *arg
)
649 return (traverse_impl(ds
->ds_dir
->dd_pool
->dp_spa
, ds
, ds
->ds_object
,
650 &dsl_dataset_phys(ds
)->ds_bp
, txg_start
, resume
, flags
, func
, arg
));
654 traverse_dataset(dsl_dataset_t
*ds
, uint64_t txg_start
,
655 int flags
, blkptr_cb_t func
, void *arg
)
657 return (traverse_dataset_resume(ds
, txg_start
, NULL
, flags
, func
, arg
));
661 traverse_dataset_destroyed(spa_t
*spa
, blkptr_t
*blkptr
,
662 uint64_t txg_start
, zbookmark_phys_t
*resume
, int flags
,
663 blkptr_cb_t func
, void *arg
)
665 return (traverse_impl(spa
, NULL
, ZB_DESTROYED_OBJSET
,
666 blkptr
, txg_start
, resume
, flags
, func
, arg
));
670 * NB: pool must not be changing on-disk (eg, from zdb or sync context).
673 traverse_pool(spa_t
*spa
, uint64_t txg_start
, int flags
,
674 blkptr_cb_t func
, void *arg
)
678 dsl_pool_t
*dp
= spa_get_dsl(spa
);
679 objset_t
*mos
= dp
->dp_meta_objset
;
680 boolean_t hard
= (flags
& TRAVERSE_HARD
);
683 err
= traverse_impl(spa
, NULL
, 0, spa_get_rootblkptr(spa
),
684 txg_start
, NULL
, flags
, func
, arg
);
688 /* visit each dataset */
689 for (obj
= 1; err
== 0;
690 err
= dmu_object_next(mos
, &obj
, B_FALSE
, txg_start
)) {
691 dmu_object_info_t doi
;
693 err
= dmu_object_info(mos
, obj
, &doi
);
700 if (doi
.doi_bonus_type
== DMU_OT_DSL_DATASET
) {
702 uint64_t txg
= txg_start
;
704 dsl_pool_config_enter(dp
, FTAG
);
705 err
= dsl_dataset_hold_obj(dp
, obj
, FTAG
, &ds
);
706 dsl_pool_config_exit(dp
, FTAG
);
712 if (dsl_dataset_phys(ds
)->ds_prev_snap_txg
> txg
)
713 txg
= dsl_dataset_phys(ds
)->ds_prev_snap_txg
;
714 err
= traverse_dataset(ds
, txg
, flags
, func
, arg
);
715 dsl_dataset_rele(ds
, FTAG
);
725 #if defined(_KERNEL) && defined(HAVE_SPL)
726 EXPORT_SYMBOL(traverse_dataset
);
727 EXPORT_SYMBOL(traverse_pool
);
729 module_param(zfs_pd_bytes_max
, int, 0644);
730 MODULE_PARM_DESC(zfs_pd_bytes_max
, "Max number of bytes to prefetch");
732 module_param(ignore_hole_birth
, int, 0644);
733 MODULE_PARM_DESC(ignore_hole_birth
, "Ignore hole_birth txg for send");