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, 2018 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>
34 #include <sys/spa_impl.h>
36 #include <sys/dmu_impl.h>
38 #include <sys/sa_impl.h>
39 #include <sys/callb.h>
40 #include <sys/zfeature.h>
42 int32_t zfs_pd_bytes_max
= 50 * 1024 * 1024; /* 50MB */
43 int32_t send_holes_without_birth_time
= 1;
45 typedef struct prefetch_data
{
48 int32_t pd_bytes_fetched
;
52 zbookmark_phys_t pd_resume
;
55 typedef struct traverse_data
{
60 zbookmark_phys_t
*td_resume
;
62 prefetch_data_t
*td_pfd
;
64 uint64_t td_hole_birth_enabled_txg
;
67 boolean_t td_realloc_possible
;
70 static int traverse_dnode(traverse_data_t
*td
, const dnode_phys_t
*dnp
,
71 uint64_t objset
, uint64_t object
);
72 static void prefetch_dnode_metadata(traverse_data_t
*td
, const dnode_phys_t
*,
73 uint64_t objset
, uint64_t object
);
76 traverse_zil_block(zilog_t
*zilog
, blkptr_t
*bp
, void *arg
, uint64_t claim_txg
)
78 traverse_data_t
*td
= arg
;
84 if (claim_txg
== 0 && bp
->blk_birth
>= spa_min_claim_txg(td
->td_spa
))
87 SET_BOOKMARK(&zb
, td
->td_objset
, ZB_ZIL_OBJECT
, ZB_ZIL_LEVEL
,
88 bp
->blk_cksum
.zc_word
[ZIL_ZC_SEQ
]);
90 (void) td
->td_func(td
->td_spa
, zilog
, bp
, &zb
, NULL
, td
->td_arg
);
96 traverse_zil_record(zilog_t
*zilog
, lr_t
*lrc
, void *arg
, uint64_t claim_txg
)
98 traverse_data_t
*td
= arg
;
100 if (lrc
->lrc_txtype
== TX_WRITE
) {
101 lr_write_t
*lr
= (lr_write_t
*)lrc
;
102 blkptr_t
*bp
= &lr
->lr_blkptr
;
108 if (claim_txg
== 0 || bp
->blk_birth
< claim_txg
)
111 SET_BOOKMARK(&zb
, td
->td_objset
, lr
->lr_foid
,
112 ZB_ZIL_LEVEL
, lr
->lr_offset
/ BP_GET_LSIZE(bp
));
114 (void) td
->td_func(td
->td_spa
, zilog
, bp
, &zb
, NULL
,
121 traverse_zil(traverse_data_t
*td
, zil_header_t
*zh
)
123 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 blocks that are already stable in read-only mode.
129 if (claim_txg
== 0 && spa_writeable(td
->td_spa
))
132 zilog_t
*zilog
= zil_alloc(spa_get_dsl(td
->td_spa
)->dp_meta_objset
, zh
);
133 (void) zil_parse(zilog
, traverse_zil_block
, traverse_zil_record
, td
,
134 claim_txg
, !(td
->td_flags
& TRAVERSE_NO_DECRYPT
));
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
;
182 int zio_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
;
184 if (!(td
->td_flags
& TRAVERSE_PREFETCH_METADATA
))
187 * If we are in the process of resuming, don't prefetch, because
188 * some children will not be needed (and in fact may have already
191 if (td
->td_resume
!= NULL
&& !ZB_IS_ZERO(td
->td_resume
))
193 if (BP_IS_HOLE(bp
) || bp
->blk_birth
<= td
->td_min_txg
)
195 if (BP_GET_LEVEL(bp
) == 0 && BP_GET_TYPE(bp
) != DMU_OT_DNODE
)
198 if ((td
->td_flags
& TRAVERSE_NO_DECRYPT
) && BP_IS_PROTECTED(bp
))
199 zio_flags
|= ZIO_FLAG_RAW
;
201 (void) arc_read(NULL
, td
->td_spa
, bp
, NULL
, NULL
,
202 ZIO_PRIORITY_ASYNC_READ
, zio_flags
, &flags
, zb
);
206 prefetch_needed(prefetch_data_t
*pfd
, const blkptr_t
*bp
)
208 ASSERT(pfd
->pd_flags
& TRAVERSE_PREFETCH_DATA
);
209 if (BP_IS_HOLE(bp
) || BP_IS_EMBEDDED(bp
) ||
210 BP_GET_TYPE(bp
) == DMU_OT_INTENT_LOG
)
216 traverse_visitbp(traverse_data_t
*td
, const dnode_phys_t
*dnp
,
217 const blkptr_t
*bp
, const zbookmark_phys_t
*zb
)
220 arc_buf_t
*buf
= NULL
;
221 prefetch_data_t
*pd
= td
->td_pfd
;
223 switch (resume_skip_check(td
, dnp
, zb
)) {
224 case RESUME_SKIP_ALL
:
226 case RESUME_SKIP_CHILDREN
:
228 case RESUME_SKIP_NONE
:
234 if (bp
->blk_birth
== 0) {
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.
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.
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.
255 * Note that the meta-dnode cannot be reallocated.
257 if (!send_holes_without_birth_time
&&
258 (!td
->td_realloc_possible
||
259 zb
->zb_object
== DMU_META_DNODE_OBJECT
) &&
260 td
->td_hole_birth_enabled_txg
<= td
->td_min_txg
)
262 } else if (bp
->blk_birth
<= td
->td_min_txg
) {
266 if (pd
!= NULL
&& !pd
->pd_exited
&& prefetch_needed(pd
, bp
)) {
267 uint64_t size
= BP_GET_LSIZE(bp
);
268 mutex_enter(&pd
->pd_mtx
);
269 ASSERT(pd
->pd_bytes_fetched
>= 0);
270 while (pd
->pd_bytes_fetched
< size
&& !pd
->pd_exited
)
271 cv_wait_sig(&pd
->pd_cv
, &pd
->pd_mtx
);
272 pd
->pd_bytes_fetched
-= size
;
273 cv_broadcast(&pd
->pd_cv
);
274 mutex_exit(&pd
->pd_mtx
);
277 if (BP_IS_HOLE(bp
)) {
278 err
= td
->td_func(td
->td_spa
, NULL
, bp
, zb
, dnp
, td
->td_arg
);
284 if (td
->td_flags
& TRAVERSE_PRE
) {
285 err
= td
->td_func(td
->td_spa
, NULL
, bp
, zb
, dnp
,
287 if (err
== TRAVERSE_VISIT_NO_CHILDREN
)
293 if (BP_GET_LEVEL(bp
) > 0) {
294 uint32_t flags
= ARC_FLAG_WAIT
;
296 int32_t epb
= BP_GET_LSIZE(bp
) >> SPA_BLKPTRSHIFT
;
297 zbookmark_phys_t
*czb
;
299 ASSERT(!BP_IS_PROTECTED(bp
));
301 err
= arc_read(NULL
, td
->td_spa
, bp
, arc_getbuf_func
, &buf
,
302 ZIO_PRIORITY_ASYNC_READ
, ZIO_FLAG_CANFAIL
, &flags
, zb
);
306 czb
= kmem_alloc(sizeof (zbookmark_phys_t
), KM_SLEEP
);
308 for (i
= 0; i
< epb
; i
++) {
309 SET_BOOKMARK(czb
, zb
->zb_objset
, zb
->zb_object
,
311 zb
->zb_blkid
* epb
+ i
);
312 traverse_prefetch_metadata(td
,
313 &((blkptr_t
*)buf
->b_data
)[i
], czb
);
316 /* recursively visitbp() blocks below this */
317 for (i
= 0; i
< epb
; i
++) {
318 SET_BOOKMARK(czb
, zb
->zb_objset
, zb
->zb_object
,
320 zb
->zb_blkid
* epb
+ i
);
321 err
= traverse_visitbp(td
, dnp
,
322 &((blkptr_t
*)buf
->b_data
)[i
], czb
);
327 kmem_free(czb
, sizeof (zbookmark_phys_t
));
329 } else if (BP_GET_TYPE(bp
) == DMU_OT_DNODE
) {
330 uint32_t flags
= ARC_FLAG_WAIT
;
331 uint32_t zio_flags
= ZIO_FLAG_CANFAIL
;
333 int32_t epb
= BP_GET_LSIZE(bp
) >> DNODE_SHIFT
;
334 dnode_phys_t
*child_dnp
;
337 * dnode blocks might have their bonus buffers encrypted, so
338 * we must be careful to honor TRAVERSE_NO_DECRYPT
340 if ((td
->td_flags
& TRAVERSE_NO_DECRYPT
) && BP_IS_PROTECTED(bp
))
341 zio_flags
|= ZIO_FLAG_RAW
;
343 err
= arc_read(NULL
, td
->td_spa
, bp
, arc_getbuf_func
, &buf
,
344 ZIO_PRIORITY_ASYNC_READ
, zio_flags
, &flags
, zb
);
348 child_dnp
= buf
->b_data
;
350 for (i
= 0; i
< epb
; i
+= child_dnp
[i
].dn_extra_slots
+ 1) {
351 prefetch_dnode_metadata(td
, &child_dnp
[i
],
352 zb
->zb_objset
, zb
->zb_blkid
* epb
+ i
);
355 /* recursively visitbp() blocks below this */
356 for (i
= 0; i
< epb
; i
+= child_dnp
[i
].dn_extra_slots
+ 1) {
357 err
= traverse_dnode(td
, &child_dnp
[i
],
358 zb
->zb_objset
, zb
->zb_blkid
* epb
+ i
);
362 } else if (BP_GET_TYPE(bp
) == DMU_OT_OBJSET
) {
363 uint32_t zio_flags
= ZIO_FLAG_CANFAIL
;
364 arc_flags_t flags
= ARC_FLAG_WAIT
;
367 if ((td
->td_flags
& TRAVERSE_NO_DECRYPT
) && BP_IS_PROTECTED(bp
))
368 zio_flags
|= ZIO_FLAG_RAW
;
370 err
= arc_read(NULL
, td
->td_spa
, bp
, arc_getbuf_func
, &buf
,
371 ZIO_PRIORITY_ASYNC_READ
, zio_flags
, &flags
, zb
);
376 prefetch_dnode_metadata(td
, &osp
->os_meta_dnode
, zb
->zb_objset
,
377 DMU_META_DNODE_OBJECT
);
379 * See the block comment above for the goal of this variable.
380 * If the maxblkid of the meta-dnode is 0, then we know that
381 * we've never had more than DNODES_PER_BLOCK objects in the
382 * dataset, which means we can't have reused any object ids.
384 if (osp
->os_meta_dnode
.dn_maxblkid
== 0)
385 td
->td_realloc_possible
= B_FALSE
;
387 if (OBJSET_BUF_HAS_USERUSED(buf
)) {
388 if (OBJSET_BUF_HAS_PROJECTUSED(buf
))
389 prefetch_dnode_metadata(td
,
390 &osp
->os_projectused_dnode
,
391 zb
->zb_objset
, DMU_PROJECTUSED_OBJECT
);
392 prefetch_dnode_metadata(td
, &osp
->os_groupused_dnode
,
393 zb
->zb_objset
, DMU_GROUPUSED_OBJECT
);
394 prefetch_dnode_metadata(td
, &osp
->os_userused_dnode
,
395 zb
->zb_objset
, DMU_USERUSED_OBJECT
);
398 err
= traverse_dnode(td
, &osp
->os_meta_dnode
, zb
->zb_objset
,
399 DMU_META_DNODE_OBJECT
);
400 if (err
== 0 && OBJSET_BUF_HAS_USERUSED(buf
)) {
401 if (OBJSET_BUF_HAS_PROJECTUSED(buf
))
402 err
= traverse_dnode(td
,
403 &osp
->os_projectused_dnode
, zb
->zb_objset
,
404 DMU_PROJECTUSED_OBJECT
);
406 err
= traverse_dnode(td
,
407 &osp
->os_groupused_dnode
, zb
->zb_objset
,
408 DMU_GROUPUSED_OBJECT
);
410 err
= traverse_dnode(td
,
411 &osp
->os_userused_dnode
, zb
->zb_objset
,
412 DMU_USERUSED_OBJECT
);
417 arc_buf_destroy(buf
, &buf
);
420 if (err
== 0 && (td
->td_flags
& TRAVERSE_POST
))
421 err
= td
->td_func(td
->td_spa
, NULL
, bp
, zb
, dnp
, td
->td_arg
);
423 if ((td
->td_flags
& TRAVERSE_HARD
) && (err
== EIO
|| err
== ECKSUM
)) {
425 * Ignore this disk error as requested by the HARD flag,
426 * and continue traversal.
432 * If we are stopping here, set td_resume.
434 if (td
->td_resume
!= NULL
&& err
!= 0 && !td
->td_paused
) {
435 td
->td_resume
->zb_objset
= zb
->zb_objset
;
436 td
->td_resume
->zb_object
= zb
->zb_object
;
437 td
->td_resume
->zb_level
= 0;
439 * If we have stopped on an indirect block (e.g. due to
440 * i/o error), we have not visited anything below it.
441 * Set the bookmark to the first level-0 block that we need
442 * to visit. This way, the resuming code does not need to
443 * deal with resuming from indirect blocks.
445 * Note, if zb_level <= 0, dnp may be NULL, so we don't want
448 td
->td_resume
->zb_blkid
= zb
->zb_blkid
;
449 if (zb
->zb_level
> 0) {
450 td
->td_resume
->zb_blkid
<<= zb
->zb_level
*
451 (dnp
->dn_indblkshift
- SPA_BLKPTRSHIFT
);
453 td
->td_paused
= B_TRUE
;
460 prefetch_dnode_metadata(traverse_data_t
*td
, const dnode_phys_t
*dnp
,
461 uint64_t objset
, uint64_t object
)
464 zbookmark_phys_t czb
;
466 for (j
= 0; j
< dnp
->dn_nblkptr
; j
++) {
467 SET_BOOKMARK(&czb
, objset
, object
, dnp
->dn_nlevels
- 1, j
);
468 traverse_prefetch_metadata(td
, &dnp
->dn_blkptr
[j
], &czb
);
471 if (dnp
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
) {
472 SET_BOOKMARK(&czb
, objset
, object
, 0, DMU_SPILL_BLKID
);
473 traverse_prefetch_metadata(td
, DN_SPILL_BLKPTR(dnp
), &czb
);
478 traverse_dnode(traverse_data_t
*td
, const dnode_phys_t
*dnp
,
479 uint64_t objset
, uint64_t object
)
482 zbookmark_phys_t czb
;
484 if (object
!= DMU_META_DNODE_OBJECT
&& td
->td_resume
!= NULL
&&
485 object
< td
->td_resume
->zb_object
)
488 if (td
->td_flags
& TRAVERSE_PRE
) {
489 SET_BOOKMARK(&czb
, objset
, object
, ZB_DNODE_LEVEL
,
491 err
= td
->td_func(td
->td_spa
, NULL
, NULL
, &czb
, dnp
,
493 if (err
== TRAVERSE_VISIT_NO_CHILDREN
)
499 for (j
= 0; j
< dnp
->dn_nblkptr
; j
++) {
500 SET_BOOKMARK(&czb
, objset
, object
, dnp
->dn_nlevels
- 1, j
);
501 err
= traverse_visitbp(td
, dnp
, &dnp
->dn_blkptr
[j
], &czb
);
506 if (err
== 0 && (dnp
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
)) {
507 SET_BOOKMARK(&czb
, objset
, object
, 0, DMU_SPILL_BLKID
);
508 err
= traverse_visitbp(td
, dnp
, DN_SPILL_BLKPTR(dnp
), &czb
);
511 if (err
== 0 && (td
->td_flags
& TRAVERSE_POST
)) {
512 SET_BOOKMARK(&czb
, objset
, object
, ZB_DNODE_LEVEL
,
514 err
= td
->td_func(td
->td_spa
, NULL
, NULL
, &czb
, dnp
,
516 if (err
== TRAVERSE_VISIT_NO_CHILDREN
)
526 traverse_prefetcher(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
527 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
529 prefetch_data_t
*pfd
= arg
;
530 int zio_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
;
531 arc_flags_t aflags
= ARC_FLAG_NOWAIT
| ARC_FLAG_PREFETCH
|
532 ARC_FLAG_PRESCIENT_PREFETCH
;
534 ASSERT(pfd
->pd_bytes_fetched
>= 0);
538 return (SET_ERROR(EINTR
));
540 if (!prefetch_needed(pfd
, bp
))
543 mutex_enter(&pfd
->pd_mtx
);
544 while (!pfd
->pd_cancel
&& pfd
->pd_bytes_fetched
>= zfs_pd_bytes_max
)
545 cv_wait_sig(&pfd
->pd_cv
, &pfd
->pd_mtx
);
546 pfd
->pd_bytes_fetched
+= BP_GET_LSIZE(bp
);
547 cv_broadcast(&pfd
->pd_cv
);
548 mutex_exit(&pfd
->pd_mtx
);
550 if ((pfd
->pd_flags
& TRAVERSE_NO_DECRYPT
) && BP_IS_PROTECTED(bp
))
551 zio_flags
|= ZIO_FLAG_RAW
;
553 (void) arc_read(NULL
, spa
, bp
, NULL
, NULL
, ZIO_PRIORITY_ASYNC_READ
,
554 zio_flags
, &aflags
, zb
);
560 traverse_prefetch_thread(void *arg
)
562 traverse_data_t
*td_main
= arg
;
563 traverse_data_t td
= *td_main
;
564 zbookmark_phys_t czb
;
565 fstrans_cookie_t cookie
= spl_fstrans_mark();
567 td
.td_func
= traverse_prefetcher
;
568 td
.td_arg
= td_main
->td_pfd
;
570 td
.td_resume
= &td_main
->td_pfd
->pd_resume
;
572 SET_BOOKMARK(&czb
, td
.td_objset
,
573 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
574 (void) traverse_visitbp(&td
, NULL
, td
.td_rootbp
, &czb
);
576 mutex_enter(&td_main
->td_pfd
->pd_mtx
);
577 td_main
->td_pfd
->pd_exited
= B_TRUE
;
578 cv_broadcast(&td_main
->td_pfd
->pd_cv
);
579 mutex_exit(&td_main
->td_pfd
->pd_mtx
);
580 spl_fstrans_unmark(cookie
);
584 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
585 * in syncing context).
588 traverse_impl(spa_t
*spa
, dsl_dataset_t
*ds
, uint64_t objset
, blkptr_t
*rootbp
,
589 uint64_t txg_start
, zbookmark_phys_t
*resume
, int flags
,
590 blkptr_cb_t func
, void *arg
)
594 zbookmark_phys_t
*czb
;
597 ASSERT(ds
== NULL
|| objset
== ds
->ds_object
);
598 ASSERT(!(flags
& TRAVERSE_PRE
) || !(flags
& TRAVERSE_POST
));
600 td
= kmem_alloc(sizeof (traverse_data_t
), KM_SLEEP
);
601 pd
= kmem_zalloc(sizeof (prefetch_data_t
), KM_SLEEP
);
602 czb
= kmem_alloc(sizeof (zbookmark_phys_t
), KM_SLEEP
);
605 td
->td_objset
= objset
;
606 td
->td_rootbp
= rootbp
;
607 td
->td_min_txg
= txg_start
;
608 td
->td_resume
= resume
;
612 td
->td_flags
= flags
;
613 td
->td_paused
= B_FALSE
;
614 td
->td_realloc_possible
= (txg_start
== 0 ? B_FALSE
: B_TRUE
);
616 if (spa_feature_is_active(spa
, SPA_FEATURE_HOLE_BIRTH
)) {
617 VERIFY(spa_feature_enabled_txg(spa
,
618 SPA_FEATURE_HOLE_BIRTH
, &td
->td_hole_birth_enabled_txg
));
620 td
->td_hole_birth_enabled_txg
= UINT64_MAX
;
623 pd
->pd_flags
= flags
;
625 pd
->pd_resume
= *resume
;
626 mutex_init(&pd
->pd_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
627 cv_init(&pd
->pd_cv
, NULL
, CV_DEFAULT
, NULL
);
629 SET_BOOKMARK(czb
, td
->td_objset
,
630 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
632 /* See comment on ZIL traversal in dsl_scan_visitds. */
633 if (ds
!= NULL
&& !ds
->ds_is_snapshot
&& !BP_IS_HOLE(rootbp
)) {
634 enum zio_flag zio_flags
= ZIO_FLAG_CANFAIL
;
635 uint32_t flags
= ARC_FLAG_WAIT
;
639 if ((td
->td_flags
& TRAVERSE_NO_DECRYPT
) &&
640 BP_IS_PROTECTED(rootbp
))
641 zio_flags
|= ZIO_FLAG_RAW
;
643 err
= arc_read(NULL
, td
->td_spa
, rootbp
, arc_getbuf_func
,
644 &buf
, ZIO_PRIORITY_ASYNC_READ
, zio_flags
, &flags
, czb
);
647 * If both TRAVERSE_HARD and TRAVERSE_PRE are set,
648 * continue to visitbp so that td_func can be called
649 * in pre stage, and err will reset to zero.
651 if (!(td
->td_flags
& TRAVERSE_HARD
) ||
652 !(td
->td_flags
& TRAVERSE_PRE
))
656 traverse_zil(td
, &osp
->os_zil_header
);
657 arc_buf_destroy(buf
, &buf
);
661 if (!(flags
& TRAVERSE_PREFETCH_DATA
) ||
662 taskq_dispatch(spa
->spa_prefetch_taskq
, traverse_prefetch_thread
,
663 td
, TQ_NOQUEUE
) == TASKQID_INVALID
)
664 pd
->pd_exited
= B_TRUE
;
666 err
= traverse_visitbp(td
, NULL
, rootbp
, czb
);
668 mutex_enter(&pd
->pd_mtx
);
669 pd
->pd_cancel
= B_TRUE
;
670 cv_broadcast(&pd
->pd_cv
);
671 while (!pd
->pd_exited
)
672 cv_wait_sig(&pd
->pd_cv
, &pd
->pd_mtx
);
673 mutex_exit(&pd
->pd_mtx
);
675 mutex_destroy(&pd
->pd_mtx
);
676 cv_destroy(&pd
->pd_cv
);
678 kmem_free(czb
, sizeof (zbookmark_phys_t
));
679 kmem_free(pd
, sizeof (struct prefetch_data
));
680 kmem_free(td
, sizeof (struct traverse_data
));
686 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
687 * in syncing context).
690 traverse_dataset_resume(dsl_dataset_t
*ds
, uint64_t txg_start
,
691 zbookmark_phys_t
*resume
,
692 int flags
, blkptr_cb_t func
, void *arg
)
694 return (traverse_impl(ds
->ds_dir
->dd_pool
->dp_spa
, ds
, ds
->ds_object
,
695 &dsl_dataset_phys(ds
)->ds_bp
, txg_start
, resume
, flags
, func
, arg
));
699 traverse_dataset(dsl_dataset_t
*ds
, uint64_t txg_start
,
700 int flags
, blkptr_cb_t func
, void *arg
)
702 return (traverse_dataset_resume(ds
, txg_start
, NULL
, flags
, func
, arg
));
706 traverse_dataset_destroyed(spa_t
*spa
, blkptr_t
*blkptr
,
707 uint64_t txg_start
, zbookmark_phys_t
*resume
, int flags
,
708 blkptr_cb_t func
, void *arg
)
710 return (traverse_impl(spa
, NULL
, ZB_DESTROYED_OBJSET
,
711 blkptr
, txg_start
, resume
, flags
, func
, arg
));
715 * NB: pool must not be changing on-disk (eg, from zdb or sync context).
718 traverse_pool(spa_t
*spa
, uint64_t txg_start
, int flags
,
719 blkptr_cb_t func
, void *arg
)
722 dsl_pool_t
*dp
= spa_get_dsl(spa
);
723 objset_t
*mos
= dp
->dp_meta_objset
;
724 boolean_t hard
= (flags
& TRAVERSE_HARD
);
727 err
= traverse_impl(spa
, NULL
, 0, spa_get_rootblkptr(spa
),
728 txg_start
, NULL
, flags
, func
, arg
);
732 /* visit each dataset */
733 for (uint64_t obj
= 1; err
== 0;
734 err
= dmu_object_next(mos
, &obj
, B_FALSE
, txg_start
)) {
735 dmu_object_info_t doi
;
737 err
= dmu_object_info(mos
, obj
, &doi
);
744 if (doi
.doi_bonus_type
== DMU_OT_DSL_DATASET
) {
746 uint64_t txg
= txg_start
;
748 dsl_pool_config_enter(dp
, FTAG
);
749 err
= dsl_dataset_hold_obj(dp
, obj
, FTAG
, &ds
);
750 dsl_pool_config_exit(dp
, FTAG
);
756 if (dsl_dataset_phys(ds
)->ds_prev_snap_txg
> txg
)
757 txg
= dsl_dataset_phys(ds
)->ds_prev_snap_txg
;
758 err
= traverse_dataset(ds
, txg
, flags
, func
, arg
);
759 dsl_dataset_rele(ds
, FTAG
);
770 EXPORT_SYMBOL(traverse_dataset
);
771 EXPORT_SYMBOL(traverse_pool
);
773 module_param(zfs_pd_bytes_max
, int, 0644);
774 MODULE_PARM_DESC(zfs_pd_bytes_max
, "Max number of bytes to prefetch");
776 module_param_named(ignore_hole_birth
, send_holes_without_birth_time
, int, 0644);
777 MODULE_PARM_DESC(ignore_hole_birth
, "Alias for send_holes_without_birth_time");
779 module_param_named(send_holes_without_birth_time
,
780 send_holes_without_birth_time
, int, 0644);
781 MODULE_PARM_DESC(send_holes_without_birth_time
,
782 "Ignore hole_birth txg for zfs send");