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) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
26 #include <sys/dsl_scan.h>
27 #include <sys/dsl_pool.h>
28 #include <sys/dsl_dataset.h>
29 #include <sys/dsl_prop.h>
30 #include <sys/dsl_dir.h>
31 #include <sys/dsl_synctask.h>
32 #include <sys/dnode.h>
33 #include <sys/dmu_tx.h>
34 #include <sys/dmu_objset.h>
38 #include <sys/zfs_context.h>
39 #include <sys/fs/zfs.h>
40 #include <sys/zfs_znode.h>
41 #include <sys/spa_impl.h>
42 #include <sys/vdev_impl.h>
43 #include <sys/zil_impl.h>
44 #include <sys/zio_checksum.h>
47 #include <sys/sa_impl.h>
48 #include <sys/zfeature.h>
50 #include <sys/zfs_vfsops.h>
53 typedef int (scan_cb_t
)(dsl_pool_t
*, const blkptr_t
*,
54 const zbookmark_phys_t
*);
56 static scan_cb_t dsl_scan_scrub_cb
;
57 static void dsl_scan_cancel_sync(void *, dmu_tx_t
*);
58 static void dsl_scan_sync_state(dsl_scan_t
*, dmu_tx_t
*tx
);
60 int zfs_top_maxinflight
= 32; /* maximum I/Os per top-level */
61 int zfs_resilver_delay
= 2; /* number of ticks to delay resilver */
62 int zfs_scrub_delay
= 4; /* number of ticks to delay scrub */
63 int zfs_scan_idle
= 50; /* idle window in clock ticks */
65 int zfs_scan_min_time_ms
= 1000; /* min millisecs to scrub per txg */
66 int zfs_free_min_time_ms
= 1000; /* min millisecs to free per txg */
67 int zfs_resilver_min_time_ms
= 3000; /* min millisecs to resilver per txg */
68 boolean_t zfs_no_scrub_io
= B_FALSE
; /* set to disable scrub i/o */
69 boolean_t zfs_no_scrub_prefetch
= B_FALSE
; /* set to disable scrub prefetch */
70 enum ddt_class zfs_scrub_ddt_class_max
= DDT_CLASS_DUPLICATE
;
71 int dsl_scan_delay_completion
= B_FALSE
; /* set to delay scan completion */
72 /* max number of blocks to free in a single TXG */
73 uint64_t zfs_free_max_blocks
= UINT64_MAX
;
75 #define DSL_SCAN_IS_SCRUB_RESILVER(scn) \
76 ((scn)->scn_phys.scn_func == POOL_SCAN_SCRUB || \
77 (scn)->scn_phys.scn_func == POOL_SCAN_RESILVER)
79 extern int zfs_txg_timeout
;
81 /* the order has to match pool_scan_type */
82 static scan_cb_t
*scan_funcs
[POOL_SCAN_FUNCS
] = {
84 dsl_scan_scrub_cb
, /* POOL_SCAN_SCRUB */
85 dsl_scan_scrub_cb
, /* POOL_SCAN_RESILVER */
89 dsl_scan_init(dsl_pool_t
*dp
, uint64_t txg
)
93 spa_t
*spa
= dp
->dp_spa
;
96 scn
= dp
->dp_scan
= kmem_zalloc(sizeof (dsl_scan_t
), KM_SLEEP
);
100 * It's possible that we're resuming a scan after a reboot so
101 * make sure that the scan_async_destroying flag is initialized
104 ASSERT(!scn
->scn_async_destroying
);
105 scn
->scn_async_destroying
= spa_feature_is_active(dp
->dp_spa
,
106 SPA_FEATURE_ASYNC_DESTROY
);
108 err
= zap_lookup(dp
->dp_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
109 "scrub_func", sizeof (uint64_t), 1, &f
);
112 * There was an old-style scrub in progress. Restart a
113 * new-style scrub from the beginning.
115 scn
->scn_restart_txg
= txg
;
116 zfs_dbgmsg("old-style scrub was in progress; "
117 "restarting new-style scrub in txg %llu",
118 scn
->scn_restart_txg
);
121 * Load the queue obj from the old location so that it
122 * can be freed by dsl_scan_done().
124 (void) zap_lookup(dp
->dp_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
125 "scrub_queue", sizeof (uint64_t), 1,
126 &scn
->scn_phys
.scn_queue_obj
);
128 err
= zap_lookup(dp
->dp_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
129 DMU_POOL_SCAN
, sizeof (uint64_t), SCAN_PHYS_NUMINTS
,
136 if (scn
->scn_phys
.scn_state
== DSS_SCANNING
&&
137 spa_prev_software_version(dp
->dp_spa
) < SPA_VERSION_SCAN
) {
139 * A new-type scrub was in progress on an old
140 * pool, and the pool was accessed by old
141 * software. Restart from the beginning, since
142 * the old software may have changed the pool in
145 scn
->scn_restart_txg
= txg
;
146 zfs_dbgmsg("new-style scrub was modified "
147 "by old software; restarting in txg %llu",
148 scn
->scn_restart_txg
);
152 spa_scan_stat_init(spa
);
157 dsl_scan_fini(dsl_pool_t
*dp
)
160 kmem_free(dp
->dp_scan
, sizeof (dsl_scan_t
));
167 dsl_scan_setup_check(void *arg
, dmu_tx_t
*tx
)
169 dsl_scan_t
*scn
= dmu_tx_pool(tx
)->dp_scan
;
171 if (scn
->scn_phys
.scn_state
== DSS_SCANNING
)
172 return (SET_ERROR(EBUSY
));
178 dsl_scan_setup_sync(void *arg
, dmu_tx_t
*tx
)
180 dsl_scan_t
*scn
= dmu_tx_pool(tx
)->dp_scan
;
181 pool_scan_func_t
*funcp
= arg
;
182 dmu_object_type_t ot
= 0;
183 dsl_pool_t
*dp
= scn
->scn_dp
;
184 spa_t
*spa
= dp
->dp_spa
;
186 ASSERT(scn
->scn_phys
.scn_state
!= DSS_SCANNING
);
187 ASSERT(*funcp
> POOL_SCAN_NONE
&& *funcp
< POOL_SCAN_FUNCS
);
188 bzero(&scn
->scn_phys
, sizeof (scn
->scn_phys
));
189 scn
->scn_phys
.scn_func
= *funcp
;
190 scn
->scn_phys
.scn_state
= DSS_SCANNING
;
191 scn
->scn_phys
.scn_min_txg
= 0;
192 scn
->scn_phys
.scn_max_txg
= tx
->tx_txg
;
193 scn
->scn_phys
.scn_ddt_class_max
= DDT_CLASSES
- 1; /* the entire DDT */
194 scn
->scn_phys
.scn_start_time
= gethrestime_sec();
195 scn
->scn_phys
.scn_errors
= 0;
196 scn
->scn_phys
.scn_to_examine
= spa
->spa_root_vdev
->vdev_stat
.vs_alloc
;
197 scn
->scn_restart_txg
= 0;
198 scn
->scn_done_txg
= 0;
199 spa_scan_stat_init(spa
);
201 if (DSL_SCAN_IS_SCRUB_RESILVER(scn
)) {
202 scn
->scn_phys
.scn_ddt_class_max
= zfs_scrub_ddt_class_max
;
204 /* rewrite all disk labels */
205 vdev_config_dirty(spa
->spa_root_vdev
);
207 if (vdev_resilver_needed(spa
->spa_root_vdev
,
208 &scn
->scn_phys
.scn_min_txg
, &scn
->scn_phys
.scn_max_txg
)) {
209 spa_event_notify(spa
, NULL
, ESC_ZFS_RESILVER_START
);
211 spa_event_notify(spa
, NULL
, ESC_ZFS_SCRUB_START
);
214 spa
->spa_scrub_started
= B_TRUE
;
216 * If this is an incremental scrub, limit the DDT scrub phase
217 * to just the auto-ditto class (for correctness); the rest
218 * of the scrub should go faster using top-down pruning.
220 if (scn
->scn_phys
.scn_min_txg
> TXG_INITIAL
)
221 scn
->scn_phys
.scn_ddt_class_max
= DDT_CLASS_DITTO
;
225 /* back to the generic stuff */
227 if (dp
->dp_blkstats
== NULL
) {
229 kmem_alloc(sizeof (zfs_all_blkstats_t
), KM_SLEEP
);
231 bzero(dp
->dp_blkstats
, sizeof (zfs_all_blkstats_t
));
233 if (spa_version(spa
) < SPA_VERSION_DSL_SCRUB
)
234 ot
= DMU_OT_ZAP_OTHER
;
236 scn
->scn_phys
.scn_queue_obj
= zap_create(dp
->dp_meta_objset
,
237 ot
? ot
: DMU_OT_SCAN_QUEUE
, DMU_OT_NONE
, 0, tx
);
239 dsl_scan_sync_state(scn
, tx
);
241 spa_history_log_internal(spa
, "scan setup", tx
,
242 "func=%u mintxg=%llu maxtxg=%llu",
243 *funcp
, scn
->scn_phys
.scn_min_txg
, scn
->scn_phys
.scn_max_txg
);
248 dsl_scan_done(dsl_scan_t
*scn
, boolean_t complete
, dmu_tx_t
*tx
)
250 static const char *old_names
[] = {
252 "scrub_ddt_bookmark",
253 "scrub_ddt_class_max",
262 dsl_pool_t
*dp
= scn
->scn_dp
;
263 spa_t
*spa
= dp
->dp_spa
;
266 /* Remove any remnants of an old-style scrub. */
267 for (i
= 0; old_names
[i
]; i
++) {
268 (void) zap_remove(dp
->dp_meta_objset
,
269 DMU_POOL_DIRECTORY_OBJECT
, old_names
[i
], tx
);
272 if (scn
->scn_phys
.scn_queue_obj
!= 0) {
273 VERIFY(0 == dmu_object_free(dp
->dp_meta_objset
,
274 scn
->scn_phys
.scn_queue_obj
, tx
));
275 scn
->scn_phys
.scn_queue_obj
= 0;
279 * If we were "restarted" from a stopped state, don't bother
280 * with anything else.
282 if (scn
->scn_phys
.scn_state
!= DSS_SCANNING
)
286 scn
->scn_phys
.scn_state
= DSS_FINISHED
;
288 scn
->scn_phys
.scn_state
= DSS_CANCELED
;
290 spa_history_log_internal(spa
, "scan done", tx
,
291 "complete=%u", complete
);
293 if (DSL_SCAN_IS_SCRUB_RESILVER(scn
)) {
294 mutex_enter(&spa
->spa_scrub_lock
);
295 while (spa
->spa_scrub_inflight
> 0) {
296 cv_wait(&spa
->spa_scrub_io_cv
,
297 &spa
->spa_scrub_lock
);
299 mutex_exit(&spa
->spa_scrub_lock
);
300 spa
->spa_scrub_started
= B_FALSE
;
301 spa
->spa_scrub_active
= B_FALSE
;
304 * If the scrub/resilver completed, update all DTLs to
305 * reflect this. Whether it succeeded or not, vacate
306 * all temporary scrub DTLs.
308 vdev_dtl_reassess(spa
->spa_root_vdev
, tx
->tx_txg
,
309 complete
? scn
->scn_phys
.scn_max_txg
: 0, B_TRUE
);
311 spa_event_notify(spa
, NULL
, scn
->scn_phys
.scn_min_txg
?
312 ESC_ZFS_RESILVER_FINISH
: ESC_ZFS_SCRUB_FINISH
);
314 spa_errlog_rotate(spa
);
317 * We may have finished replacing a device.
318 * Let the async thread assess this and handle the detach.
320 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
323 scn
->scn_phys
.scn_end_time
= gethrestime_sec();
328 dsl_scan_cancel_check(void *arg
, dmu_tx_t
*tx
)
330 dsl_scan_t
*scn
= dmu_tx_pool(tx
)->dp_scan
;
332 if (scn
->scn_phys
.scn_state
!= DSS_SCANNING
)
333 return (SET_ERROR(ENOENT
));
339 dsl_scan_cancel_sync(void *arg
, dmu_tx_t
*tx
)
341 dsl_scan_t
*scn
= dmu_tx_pool(tx
)->dp_scan
;
343 dsl_scan_done(scn
, B_FALSE
, tx
);
344 dsl_scan_sync_state(scn
, tx
);
348 dsl_scan_cancel(dsl_pool_t
*dp
)
350 return (dsl_sync_task(spa_name(dp
->dp_spa
), dsl_scan_cancel_check
,
351 dsl_scan_cancel_sync
, NULL
, 3, ZFS_SPACE_CHECK_RESERVED
));
354 static void dsl_scan_visitbp(blkptr_t
*bp
, const zbookmark_phys_t
*zb
,
355 dnode_phys_t
*dnp
, dsl_dataset_t
*ds
, dsl_scan_t
*scn
,
356 dmu_objset_type_t ostype
, dmu_tx_t
*tx
);
357 static void dsl_scan_visitdnode(dsl_scan_t
*, dsl_dataset_t
*ds
,
358 dmu_objset_type_t ostype
,
359 dnode_phys_t
*dnp
, uint64_t object
, dmu_tx_t
*tx
);
362 dsl_free(dsl_pool_t
*dp
, uint64_t txg
, const blkptr_t
*bp
)
364 zio_free(dp
->dp_spa
, txg
, bp
);
368 dsl_free_sync(zio_t
*pio
, dsl_pool_t
*dp
, uint64_t txg
, const blkptr_t
*bpp
)
370 ASSERT(dsl_pool_sync_context(dp
));
371 zio_nowait(zio_free_sync(pio
, dp
->dp_spa
, txg
, bpp
, pio
->io_flags
));
375 dsl_scan_ds_maxtxg(dsl_dataset_t
*ds
)
377 uint64_t smt
= ds
->ds_dir
->dd_pool
->dp_scan
->scn_phys
.scn_max_txg
;
378 if (ds
->ds_is_snapshot
)
379 return (MIN(smt
, dsl_dataset_phys(ds
)->ds_creation_txg
));
384 dsl_scan_sync_state(dsl_scan_t
*scn
, dmu_tx_t
*tx
)
386 VERIFY0(zap_update(scn
->scn_dp
->dp_meta_objset
,
387 DMU_POOL_DIRECTORY_OBJECT
,
388 DMU_POOL_SCAN
, sizeof (uint64_t), SCAN_PHYS_NUMINTS
,
389 &scn
->scn_phys
, tx
));
392 extern int zfs_vdev_async_write_active_min_dirty_percent
;
395 dsl_scan_check_pause(dsl_scan_t
*scn
, const zbookmark_phys_t
*zb
)
397 /* we never skip user/group accounting objects */
398 if (zb
&& (int64_t)zb
->zb_object
< 0)
401 if (scn
->scn_pausing
)
402 return (B_TRUE
); /* we're already pausing */
404 if (!ZB_IS_ZERO(&scn
->scn_phys
.scn_bookmark
))
405 return (B_FALSE
); /* we're resuming */
407 /* We only know how to resume from level-0 blocks. */
408 if (zb
&& zb
->zb_level
!= 0)
413 * - we have scanned for the maximum time: an entire txg
414 * timeout (default 5 sec)
416 * - we have scanned for at least the minimum time (default 1 sec
417 * for scrub, 3 sec for resilver), and either we have sufficient
418 * dirty data that we are starting to write more quickly
419 * (default 30%), or someone is explicitly waiting for this txg
422 * - the spa is shutting down because this pool is being exported
423 * or the machine is rebooting.
425 int mintime
= (scn
->scn_phys
.scn_func
== POOL_SCAN_RESILVER
) ?
426 zfs_resilver_min_time_ms
: zfs_scan_min_time_ms
;
427 uint64_t elapsed_nanosecs
= gethrtime() - scn
->scn_sync_start_time
;
428 int dirty_pct
= scn
->scn_dp
->dp_dirty_total
* 100 / zfs_dirty_data_max
;
429 if (elapsed_nanosecs
/ NANOSEC
>= zfs_txg_timeout
||
430 (NSEC2MSEC(elapsed_nanosecs
) > mintime
&&
431 (txg_sync_waiting(scn
->scn_dp
) ||
432 dirty_pct
>= zfs_vdev_async_write_active_min_dirty_percent
)) ||
433 spa_shutting_down(scn
->scn_dp
->dp_spa
)) {
435 dprintf("pausing at bookmark %llx/%llx/%llx/%llx\n",
436 (longlong_t
)zb
->zb_objset
,
437 (longlong_t
)zb
->zb_object
,
438 (longlong_t
)zb
->zb_level
,
439 (longlong_t
)zb
->zb_blkid
);
440 scn
->scn_phys
.scn_bookmark
= *zb
;
442 dprintf("pausing at DDT bookmark %llx/%llx/%llx/%llx\n",
443 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_class
,
444 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_type
,
445 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_checksum
,
446 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_cursor
);
447 scn
->scn_pausing
= B_TRUE
;
453 typedef struct zil_scan_arg
{
455 zil_header_t
*zsa_zh
;
460 dsl_scan_zil_block(zilog_t
*zilog
, blkptr_t
*bp
, void *arg
, uint64_t claim_txg
)
462 zil_scan_arg_t
*zsa
= arg
;
463 dsl_pool_t
*dp
= zsa
->zsa_dp
;
464 dsl_scan_t
*scn
= dp
->dp_scan
;
465 zil_header_t
*zh
= zsa
->zsa_zh
;
468 if (BP_IS_HOLE(bp
) || bp
->blk_birth
<= scn
->scn_phys
.scn_cur_min_txg
)
472 * One block ("stubby") can be allocated a long time ago; we
473 * want to visit that one because it has been allocated
474 * (on-disk) even if it hasn't been claimed (even though for
475 * scrub there's nothing to do to it).
477 if (claim_txg
== 0 && bp
->blk_birth
>= spa_first_txg(dp
->dp_spa
))
480 SET_BOOKMARK(&zb
, zh
->zh_log
.blk_cksum
.zc_word
[ZIL_ZC_OBJSET
],
481 ZB_ZIL_OBJECT
, ZB_ZIL_LEVEL
, bp
->blk_cksum
.zc_word
[ZIL_ZC_SEQ
]);
483 VERIFY(0 == scan_funcs
[scn
->scn_phys
.scn_func
](dp
, bp
, &zb
));
489 dsl_scan_zil_record(zilog_t
*zilog
, lr_t
*lrc
, void *arg
, uint64_t claim_txg
)
491 if (lrc
->lrc_txtype
== TX_WRITE
) {
492 zil_scan_arg_t
*zsa
= arg
;
493 dsl_pool_t
*dp
= zsa
->zsa_dp
;
494 dsl_scan_t
*scn
= dp
->dp_scan
;
495 zil_header_t
*zh
= zsa
->zsa_zh
;
496 lr_write_t
*lr
= (lr_write_t
*)lrc
;
497 blkptr_t
*bp
= &lr
->lr_blkptr
;
500 if (BP_IS_HOLE(bp
) ||
501 bp
->blk_birth
<= scn
->scn_phys
.scn_cur_min_txg
)
505 * birth can be < claim_txg if this record's txg is
506 * already txg sync'ed (but this log block contains
507 * other records that are not synced)
509 if (claim_txg
== 0 || bp
->blk_birth
< claim_txg
)
512 SET_BOOKMARK(&zb
, zh
->zh_log
.blk_cksum
.zc_word
[ZIL_ZC_OBJSET
],
513 lr
->lr_foid
, ZB_ZIL_LEVEL
,
514 lr
->lr_offset
/ BP_GET_LSIZE(bp
));
516 VERIFY(0 == scan_funcs
[scn
->scn_phys
.scn_func
](dp
, bp
, &zb
));
522 dsl_scan_zil(dsl_pool_t
*dp
, zil_header_t
*zh
)
524 uint64_t claim_txg
= zh
->zh_claim_txg
;
525 zil_scan_arg_t zsa
= { dp
, zh
};
529 * We only want to visit blocks that have been claimed but not yet
530 * replayed (or, in read-only mode, blocks that *would* be claimed).
532 if (claim_txg
== 0 && spa_writeable(dp
->dp_spa
))
535 zilog
= zil_alloc(dp
->dp_meta_objset
, zh
);
537 (void) zil_parse(zilog
, dsl_scan_zil_block
, dsl_scan_zil_record
, &zsa
,
545 dsl_scan_prefetch(dsl_scan_t
*scn
, arc_buf_t
*buf
, blkptr_t
*bp
,
546 uint64_t objset
, uint64_t object
, uint64_t blkid
)
548 zbookmark_phys_t czb
;
549 arc_flags_t flags
= ARC_FLAG_NOWAIT
| ARC_FLAG_PREFETCH
;
551 if (zfs_no_scrub_prefetch
)
554 if (BP_IS_HOLE(bp
) || bp
->blk_birth
<= scn
->scn_phys
.scn_min_txg
||
555 (BP_GET_LEVEL(bp
) == 0 && BP_GET_TYPE(bp
) != DMU_OT_DNODE
))
558 SET_BOOKMARK(&czb
, objset
, object
, BP_GET_LEVEL(bp
), blkid
);
560 (void) arc_read(scn
->scn_zio_root
, scn
->scn_dp
->dp_spa
, bp
,
561 NULL
, NULL
, ZIO_PRIORITY_ASYNC_READ
,
562 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SCAN_THREAD
, &flags
, &czb
);
566 dsl_scan_check_resume(dsl_scan_t
*scn
, const dnode_phys_t
*dnp
,
567 const zbookmark_phys_t
*zb
)
570 * We never skip over user/group accounting objects (obj<0)
572 if (!ZB_IS_ZERO(&scn
->scn_phys
.scn_bookmark
) &&
573 (int64_t)zb
->zb_object
>= 0) {
575 * If we already visited this bp & everything below (in
576 * a prior txg sync), don't bother doing it again.
578 if (zbookmark_subtree_completed(dnp
, zb
,
579 &scn
->scn_phys
.scn_bookmark
))
583 * If we found the block we're trying to resume from, or
584 * we went past it to a different object, zero it out to
585 * indicate that it's OK to start checking for pausing
588 if (bcmp(zb
, &scn
->scn_phys
.scn_bookmark
, sizeof (*zb
)) == 0 ||
589 zb
->zb_object
> scn
->scn_phys
.scn_bookmark
.zb_object
) {
590 dprintf("resuming at %llx/%llx/%llx/%llx\n",
591 (longlong_t
)zb
->zb_objset
,
592 (longlong_t
)zb
->zb_object
,
593 (longlong_t
)zb
->zb_level
,
594 (longlong_t
)zb
->zb_blkid
);
595 bzero(&scn
->scn_phys
.scn_bookmark
, sizeof (*zb
));
602 * Return nonzero on i/o error.
603 * Return new buf to write out in *bufp.
606 dsl_scan_recurse(dsl_scan_t
*scn
, dsl_dataset_t
*ds
, dmu_objset_type_t ostype
,
607 dnode_phys_t
*dnp
, const blkptr_t
*bp
,
608 const zbookmark_phys_t
*zb
, dmu_tx_t
*tx
)
610 dsl_pool_t
*dp
= scn
->scn_dp
;
611 int zio_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_SCAN_THREAD
;
614 if (BP_GET_LEVEL(bp
) > 0) {
615 arc_flags_t flags
= ARC_FLAG_WAIT
;
618 int epb
= BP_GET_LSIZE(bp
) >> SPA_BLKPTRSHIFT
;
621 err
= arc_read(NULL
, dp
->dp_spa
, bp
, arc_getbuf_func
, &buf
,
622 ZIO_PRIORITY_ASYNC_READ
, zio_flags
, &flags
, zb
);
624 scn
->scn_phys
.scn_errors
++;
627 for (i
= 0, cbp
= buf
->b_data
; i
< epb
; i
++, cbp
++) {
628 dsl_scan_prefetch(scn
, buf
, cbp
, zb
->zb_objset
,
629 zb
->zb_object
, zb
->zb_blkid
* epb
+ i
);
631 for (i
= 0, cbp
= buf
->b_data
; i
< epb
; i
++, cbp
++) {
632 zbookmark_phys_t czb
;
634 SET_BOOKMARK(&czb
, zb
->zb_objset
, zb
->zb_object
,
636 zb
->zb_blkid
* epb
+ i
);
637 dsl_scan_visitbp(cbp
, &czb
, dnp
,
638 ds
, scn
, ostype
, tx
);
640 (void) arc_buf_remove_ref(buf
, &buf
);
641 } else if (BP_GET_TYPE(bp
) == DMU_OT_DNODE
) {
642 arc_flags_t flags
= ARC_FLAG_WAIT
;
645 int epb
= BP_GET_LSIZE(bp
) >> DNODE_SHIFT
;
648 err
= arc_read(NULL
, dp
->dp_spa
, bp
, arc_getbuf_func
, &buf
,
649 ZIO_PRIORITY_ASYNC_READ
, zio_flags
, &flags
, zb
);
651 scn
->scn_phys
.scn_errors
++;
654 for (i
= 0, cdnp
= buf
->b_data
; i
< epb
; i
++, cdnp
++) {
655 for (j
= 0; j
< cdnp
->dn_nblkptr
; j
++) {
656 blkptr_t
*cbp
= &cdnp
->dn_blkptr
[j
];
657 dsl_scan_prefetch(scn
, buf
, cbp
,
658 zb
->zb_objset
, zb
->zb_blkid
* epb
+ i
, j
);
661 for (i
= 0, cdnp
= buf
->b_data
; i
< epb
; i
++, cdnp
++) {
662 dsl_scan_visitdnode(scn
, ds
, ostype
,
663 cdnp
, zb
->zb_blkid
* epb
+ i
, tx
);
666 (void) arc_buf_remove_ref(buf
, &buf
);
667 } else if (BP_GET_TYPE(bp
) == DMU_OT_OBJSET
) {
668 arc_flags_t flags
= ARC_FLAG_WAIT
;
672 err
= arc_read(NULL
, dp
->dp_spa
, bp
, arc_getbuf_func
, &buf
,
673 ZIO_PRIORITY_ASYNC_READ
, zio_flags
, &flags
, zb
);
675 scn
->scn_phys
.scn_errors
++;
681 dsl_scan_visitdnode(scn
, ds
, osp
->os_type
,
682 &osp
->os_meta_dnode
, DMU_META_DNODE_OBJECT
, tx
);
684 if (OBJSET_BUF_HAS_USERUSED(buf
)) {
686 * We also always visit user/group accounting
687 * objects, and never skip them, even if we are
688 * pausing. This is necessary so that the space
689 * deltas from this txg get integrated.
691 dsl_scan_visitdnode(scn
, ds
, osp
->os_type
,
692 &osp
->os_groupused_dnode
,
693 DMU_GROUPUSED_OBJECT
, tx
);
694 dsl_scan_visitdnode(scn
, ds
, osp
->os_type
,
695 &osp
->os_userused_dnode
,
696 DMU_USERUSED_OBJECT
, tx
);
698 (void) arc_buf_remove_ref(buf
, &buf
);
705 dsl_scan_visitdnode(dsl_scan_t
*scn
, dsl_dataset_t
*ds
,
706 dmu_objset_type_t ostype
, dnode_phys_t
*dnp
,
707 uint64_t object
, dmu_tx_t
*tx
)
711 for (j
= 0; j
< dnp
->dn_nblkptr
; j
++) {
712 zbookmark_phys_t czb
;
714 SET_BOOKMARK(&czb
, ds
? ds
->ds_object
: 0, object
,
715 dnp
->dn_nlevels
- 1, j
);
716 dsl_scan_visitbp(&dnp
->dn_blkptr
[j
],
717 &czb
, dnp
, ds
, scn
, ostype
, tx
);
720 if (dnp
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
) {
721 zbookmark_phys_t czb
;
722 SET_BOOKMARK(&czb
, ds
? ds
->ds_object
: 0, object
,
724 dsl_scan_visitbp(&dnp
->dn_spill
,
725 &czb
, dnp
, ds
, scn
, ostype
, tx
);
730 * The arguments are in this order because mdb can only print the
731 * first 5; we want them to be useful.
734 dsl_scan_visitbp(blkptr_t
*bp
, const zbookmark_phys_t
*zb
,
735 dnode_phys_t
*dnp
, dsl_dataset_t
*ds
, dsl_scan_t
*scn
,
736 dmu_objset_type_t ostype
, dmu_tx_t
*tx
)
738 dsl_pool_t
*dp
= scn
->scn_dp
;
739 arc_buf_t
*buf
= NULL
;
740 blkptr_t bp_toread
= *bp
;
742 /* ASSERT(pbuf == NULL || arc_released(pbuf)); */
744 if (dsl_scan_check_pause(scn
, zb
))
747 if (dsl_scan_check_resume(scn
, dnp
, zb
))
753 scn
->scn_visited_this_txg
++;
756 "visiting ds=%p/%llu zb=%llx/%llx/%llx/%llx bp=%p",
757 ds
, ds
? ds
->ds_object
: 0,
758 zb
->zb_objset
, zb
->zb_object
, zb
->zb_level
, zb
->zb_blkid
,
761 if (bp
->blk_birth
<= scn
->scn_phys
.scn_cur_min_txg
)
764 if (dsl_scan_recurse(scn
, ds
, ostype
, dnp
, &bp_toread
, zb
, tx
) != 0)
768 * If dsl_scan_ddt() has aready visited this block, it will have
769 * already done any translations or scrubbing, so don't call the
772 if (ddt_class_contains(dp
->dp_spa
,
773 scn
->scn_phys
.scn_ddt_class_max
, bp
)) {
779 * If this block is from the future (after cur_max_txg), then we
780 * are doing this on behalf of a deleted snapshot, and we will
781 * revisit the future block on the next pass of this dataset.
782 * Don't scan it now unless we need to because something
783 * under it was modified.
785 if (BP_PHYSICAL_BIRTH(bp
) <= scn
->scn_phys
.scn_cur_max_txg
) {
786 scan_funcs
[scn
->scn_phys
.scn_func
](dp
, bp
, zb
);
791 dsl_scan_visit_rootbp(dsl_scan_t
*scn
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
796 SET_BOOKMARK(&zb
, ds
? ds
->ds_object
: DMU_META_OBJSET
,
797 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
798 dsl_scan_visitbp(bp
, &zb
, NULL
,
799 ds
, scn
, DMU_OST_NONE
, tx
);
801 dprintf_ds(ds
, "finished scan%s", "");
805 dsl_scan_ds_destroyed(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
807 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
808 dsl_scan_t
*scn
= dp
->dp_scan
;
811 if (scn
->scn_phys
.scn_state
!= DSS_SCANNING
)
814 if (scn
->scn_phys
.scn_bookmark
.zb_objset
== ds
->ds_object
) {
815 if (ds
->ds_is_snapshot
) {
816 /* Note, scn_cur_{min,max}_txg stays the same. */
817 scn
->scn_phys
.scn_bookmark
.zb_objset
=
818 dsl_dataset_phys(ds
)->ds_next_snap_obj
;
819 zfs_dbgmsg("destroying ds %llu; currently traversing; "
820 "reset zb_objset to %llu",
821 (u_longlong_t
)ds
->ds_object
,
822 (u_longlong_t
)dsl_dataset_phys(ds
)->
824 scn
->scn_phys
.scn_flags
|= DSF_VISIT_DS_AGAIN
;
826 SET_BOOKMARK(&scn
->scn_phys
.scn_bookmark
,
827 ZB_DESTROYED_OBJSET
, 0, 0, 0);
828 zfs_dbgmsg("destroying ds %llu; currently traversing; "
829 "reset bookmark to -1,0,0,0",
830 (u_longlong_t
)ds
->ds_object
);
832 } else if (zap_lookup_int_key(dp
->dp_meta_objset
,
833 scn
->scn_phys
.scn_queue_obj
, ds
->ds_object
, &mintxg
) == 0) {
834 ASSERT3U(dsl_dataset_phys(ds
)->ds_num_children
, <=, 1);
835 VERIFY3U(0, ==, zap_remove_int(dp
->dp_meta_objset
,
836 scn
->scn_phys
.scn_queue_obj
, ds
->ds_object
, tx
));
837 if (ds
->ds_is_snapshot
) {
839 * We keep the same mintxg; it could be >
840 * ds_creation_txg if the previous snapshot was
843 VERIFY(zap_add_int_key(dp
->dp_meta_objset
,
844 scn
->scn_phys
.scn_queue_obj
,
845 dsl_dataset_phys(ds
)->ds_next_snap_obj
,
847 zfs_dbgmsg("destroying ds %llu; in queue; "
848 "replacing with %llu",
849 (u_longlong_t
)ds
->ds_object
,
850 (u_longlong_t
)dsl_dataset_phys(ds
)->
853 zfs_dbgmsg("destroying ds %llu; in queue; removing",
854 (u_longlong_t
)ds
->ds_object
);
857 zfs_dbgmsg("destroying ds %llu; ignoring",
858 (u_longlong_t
)ds
->ds_object
);
862 * dsl_scan_sync() should be called after this, and should sync
863 * out our changed state, but just to be safe, do it here.
865 dsl_scan_sync_state(scn
, tx
);
869 dsl_scan_ds_snapshotted(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
871 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
872 dsl_scan_t
*scn
= dp
->dp_scan
;
875 if (scn
->scn_phys
.scn_state
!= DSS_SCANNING
)
878 ASSERT(dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0);
880 if (scn
->scn_phys
.scn_bookmark
.zb_objset
== ds
->ds_object
) {
881 scn
->scn_phys
.scn_bookmark
.zb_objset
=
882 dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
883 zfs_dbgmsg("snapshotting ds %llu; currently traversing; "
884 "reset zb_objset to %llu",
885 (u_longlong_t
)ds
->ds_object
,
886 (u_longlong_t
)dsl_dataset_phys(ds
)->ds_prev_snap_obj
);
887 } else if (zap_lookup_int_key(dp
->dp_meta_objset
,
888 scn
->scn_phys
.scn_queue_obj
, ds
->ds_object
, &mintxg
) == 0) {
889 VERIFY3U(0, ==, zap_remove_int(dp
->dp_meta_objset
,
890 scn
->scn_phys
.scn_queue_obj
, ds
->ds_object
, tx
));
891 VERIFY(zap_add_int_key(dp
->dp_meta_objset
,
892 scn
->scn_phys
.scn_queue_obj
,
893 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, mintxg
, tx
) == 0);
894 zfs_dbgmsg("snapshotting ds %llu; in queue; "
895 "replacing with %llu",
896 (u_longlong_t
)ds
->ds_object
,
897 (u_longlong_t
)dsl_dataset_phys(ds
)->ds_prev_snap_obj
);
899 dsl_scan_sync_state(scn
, tx
);
903 dsl_scan_ds_clone_swapped(dsl_dataset_t
*ds1
, dsl_dataset_t
*ds2
, dmu_tx_t
*tx
)
905 dsl_pool_t
*dp
= ds1
->ds_dir
->dd_pool
;
906 dsl_scan_t
*scn
= dp
->dp_scan
;
909 if (scn
->scn_phys
.scn_state
!= DSS_SCANNING
)
912 if (scn
->scn_phys
.scn_bookmark
.zb_objset
== ds1
->ds_object
) {
913 scn
->scn_phys
.scn_bookmark
.zb_objset
= ds2
->ds_object
;
914 zfs_dbgmsg("clone_swap ds %llu; currently traversing; "
915 "reset zb_objset to %llu",
916 (u_longlong_t
)ds1
->ds_object
,
917 (u_longlong_t
)ds2
->ds_object
);
918 } else if (scn
->scn_phys
.scn_bookmark
.zb_objset
== ds2
->ds_object
) {
919 scn
->scn_phys
.scn_bookmark
.zb_objset
= ds1
->ds_object
;
920 zfs_dbgmsg("clone_swap ds %llu; currently traversing; "
921 "reset zb_objset to %llu",
922 (u_longlong_t
)ds2
->ds_object
,
923 (u_longlong_t
)ds1
->ds_object
);
926 if (zap_lookup_int_key(dp
->dp_meta_objset
, scn
->scn_phys
.scn_queue_obj
,
927 ds1
->ds_object
, &mintxg
) == 0) {
930 ASSERT3U(mintxg
, ==, dsl_dataset_phys(ds1
)->ds_prev_snap_txg
);
931 ASSERT3U(mintxg
, ==, dsl_dataset_phys(ds2
)->ds_prev_snap_txg
);
932 VERIFY3U(0, ==, zap_remove_int(dp
->dp_meta_objset
,
933 scn
->scn_phys
.scn_queue_obj
, ds1
->ds_object
, tx
));
934 err
= zap_add_int_key(dp
->dp_meta_objset
,
935 scn
->scn_phys
.scn_queue_obj
, ds2
->ds_object
, mintxg
, tx
);
936 VERIFY(err
== 0 || err
== EEXIST
);
938 /* Both were there to begin with */
939 VERIFY(0 == zap_add_int_key(dp
->dp_meta_objset
,
940 scn
->scn_phys
.scn_queue_obj
,
941 ds1
->ds_object
, mintxg
, tx
));
943 zfs_dbgmsg("clone_swap ds %llu; in queue; "
944 "replacing with %llu",
945 (u_longlong_t
)ds1
->ds_object
,
946 (u_longlong_t
)ds2
->ds_object
);
947 } else if (zap_lookup_int_key(dp
->dp_meta_objset
,
948 scn
->scn_phys
.scn_queue_obj
, ds2
->ds_object
, &mintxg
) == 0) {
949 ASSERT3U(mintxg
, ==, dsl_dataset_phys(ds1
)->ds_prev_snap_txg
);
950 ASSERT3U(mintxg
, ==, dsl_dataset_phys(ds2
)->ds_prev_snap_txg
);
951 VERIFY3U(0, ==, zap_remove_int(dp
->dp_meta_objset
,
952 scn
->scn_phys
.scn_queue_obj
, ds2
->ds_object
, tx
));
953 VERIFY(0 == zap_add_int_key(dp
->dp_meta_objset
,
954 scn
->scn_phys
.scn_queue_obj
, ds1
->ds_object
, mintxg
, tx
));
955 zfs_dbgmsg("clone_swap ds %llu; in queue; "
956 "replacing with %llu",
957 (u_longlong_t
)ds2
->ds_object
,
958 (u_longlong_t
)ds1
->ds_object
);
961 dsl_scan_sync_state(scn
, tx
);
964 struct enqueue_clones_arg
{
971 enqueue_clones_cb(dsl_pool_t
*dp
, dsl_dataset_t
*hds
, void *arg
)
973 struct enqueue_clones_arg
*eca
= arg
;
976 dsl_scan_t
*scn
= dp
->dp_scan
;
978 if (dsl_dir_phys(hds
->ds_dir
)->dd_origin_obj
!= eca
->originobj
)
981 err
= dsl_dataset_hold_obj(dp
, hds
->ds_object
, FTAG
, &ds
);
985 while (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= eca
->originobj
) {
987 err
= dsl_dataset_hold_obj(dp
,
988 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, FTAG
, &prev
);
990 dsl_dataset_rele(ds
, FTAG
);
995 VERIFY(zap_add_int_key(dp
->dp_meta_objset
,
996 scn
->scn_phys
.scn_queue_obj
, ds
->ds_object
,
997 dsl_dataset_phys(ds
)->ds_prev_snap_txg
, eca
->tx
) == 0);
998 dsl_dataset_rele(ds
, FTAG
);
1003 dsl_scan_visitds(dsl_scan_t
*scn
, uint64_t dsobj
, dmu_tx_t
*tx
)
1005 dsl_pool_t
*dp
= scn
->scn_dp
;
1009 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp
, dsobj
, FTAG
, &ds
));
1011 if (dmu_objset_from_ds(ds
, &os
))
1015 * Only the ZIL in the head (non-snapshot) is valid. Even though
1016 * snapshots can have ZIL block pointers (which may be the same
1017 * BP as in the head), they must be ignored. So we traverse the
1018 * ZIL here, rather than in scan_recurse(), because the regular
1019 * snapshot block-sharing rules don't apply to it.
1021 if (DSL_SCAN_IS_SCRUB_RESILVER(scn
) && !ds
->ds_is_snapshot
)
1022 dsl_scan_zil(dp
, &os
->os_zil_header
);
1025 * Iterate over the bps in this ds.
1027 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1028 dsl_scan_visit_rootbp(scn
, ds
, &dsl_dataset_phys(ds
)->ds_bp
, tx
);
1030 char *dsname
= kmem_alloc(ZFS_MAXNAMELEN
, KM_SLEEP
);
1031 dsl_dataset_name(ds
, dsname
);
1032 zfs_dbgmsg("scanned dataset %llu (%s) with min=%llu max=%llu; "
1034 (longlong_t
)dsobj
, dsname
,
1035 (longlong_t
)scn
->scn_phys
.scn_cur_min_txg
,
1036 (longlong_t
)scn
->scn_phys
.scn_cur_max_txg
,
1037 (int)scn
->scn_pausing
);
1038 kmem_free(dsname
, ZFS_MAXNAMELEN
);
1040 if (scn
->scn_pausing
)
1044 * We've finished this pass over this dataset.
1048 * If we did not completely visit this dataset, do another pass.
1050 if (scn
->scn_phys
.scn_flags
& DSF_VISIT_DS_AGAIN
) {
1051 zfs_dbgmsg("incomplete pass; visiting again");
1052 scn
->scn_phys
.scn_flags
&= ~DSF_VISIT_DS_AGAIN
;
1053 VERIFY(zap_add_int_key(dp
->dp_meta_objset
,
1054 scn
->scn_phys
.scn_queue_obj
, ds
->ds_object
,
1055 scn
->scn_phys
.scn_cur_max_txg
, tx
) == 0);
1060 * Add descendent datasets to work queue.
1062 if (dsl_dataset_phys(ds
)->ds_next_snap_obj
!= 0) {
1063 VERIFY(zap_add_int_key(dp
->dp_meta_objset
,
1064 scn
->scn_phys
.scn_queue_obj
,
1065 dsl_dataset_phys(ds
)->ds_next_snap_obj
,
1066 dsl_dataset_phys(ds
)->ds_creation_txg
, tx
) == 0);
1068 if (dsl_dataset_phys(ds
)->ds_num_children
> 1) {
1069 boolean_t usenext
= B_FALSE
;
1070 if (dsl_dataset_phys(ds
)->ds_next_clones_obj
!= 0) {
1073 * A bug in a previous version of the code could
1074 * cause upgrade_clones_cb() to not set
1075 * ds_next_snap_obj when it should, leading to a
1076 * missing entry. Therefore we can only use the
1077 * next_clones_obj when its count is correct.
1079 int err
= zap_count(dp
->dp_meta_objset
,
1080 dsl_dataset_phys(ds
)->ds_next_clones_obj
, &count
);
1082 count
== dsl_dataset_phys(ds
)->ds_num_children
- 1)
1087 VERIFY0(zap_join_key(dp
->dp_meta_objset
,
1088 dsl_dataset_phys(ds
)->ds_next_clones_obj
,
1089 scn
->scn_phys
.scn_queue_obj
,
1090 dsl_dataset_phys(ds
)->ds_creation_txg
, tx
));
1092 struct enqueue_clones_arg eca
;
1094 eca
.originobj
= ds
->ds_object
;
1096 VERIFY0(dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
1097 enqueue_clones_cb
, &eca
, DS_FIND_CHILDREN
));
1102 dsl_dataset_rele(ds
, FTAG
);
1107 enqueue_cb(dsl_pool_t
*dp
, dsl_dataset_t
*hds
, void *arg
)
1112 dsl_scan_t
*scn
= dp
->dp_scan
;
1114 err
= dsl_dataset_hold_obj(dp
, hds
->ds_object
, FTAG
, &ds
);
1118 while (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0) {
1119 dsl_dataset_t
*prev
;
1120 err
= dsl_dataset_hold_obj(dp
,
1121 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, FTAG
, &prev
);
1123 dsl_dataset_rele(ds
, FTAG
);
1128 * If this is a clone, we don't need to worry about it for now.
1130 if (dsl_dataset_phys(prev
)->ds_next_snap_obj
!= ds
->ds_object
) {
1131 dsl_dataset_rele(ds
, FTAG
);
1132 dsl_dataset_rele(prev
, FTAG
);
1135 dsl_dataset_rele(ds
, FTAG
);
1139 VERIFY(zap_add_int_key(dp
->dp_meta_objset
, scn
->scn_phys
.scn_queue_obj
,
1140 ds
->ds_object
, dsl_dataset_phys(ds
)->ds_prev_snap_txg
, tx
) == 0);
1141 dsl_dataset_rele(ds
, FTAG
);
1146 * Scrub/dedup interaction.
1148 * If there are N references to a deduped block, we don't want to scrub it
1149 * N times -- ideally, we should scrub it exactly once.
1151 * We leverage the fact that the dde's replication class (enum ddt_class)
1152 * is ordered from highest replication class (DDT_CLASS_DITTO) to lowest
1153 * (DDT_CLASS_UNIQUE) so that we may walk the DDT in that order.
1155 * To prevent excess scrubbing, the scrub begins by walking the DDT
1156 * to find all blocks with refcnt > 1, and scrubs each of these once.
1157 * Since there are two replication classes which contain blocks with
1158 * refcnt > 1, we scrub the highest replication class (DDT_CLASS_DITTO) first.
1159 * Finally the top-down scrub begins, only visiting blocks with refcnt == 1.
1161 * There would be nothing more to say if a block's refcnt couldn't change
1162 * during a scrub, but of course it can so we must account for changes
1163 * in a block's replication class.
1165 * Here's an example of what can occur:
1167 * If a block has refcnt > 1 during the DDT scrub phase, but has refcnt == 1
1168 * when visited during the top-down scrub phase, it will be scrubbed twice.
1169 * This negates our scrub optimization, but is otherwise harmless.
1171 * If a block has refcnt == 1 during the DDT scrub phase, but has refcnt > 1
1172 * on each visit during the top-down scrub phase, it will never be scrubbed.
1173 * To catch this, ddt_sync_entry() notifies the scrub code whenever a block's
1174 * reference class transitions to a higher level (i.e DDT_CLASS_UNIQUE to
1175 * DDT_CLASS_DUPLICATE); if it transitions from refcnt == 1 to refcnt > 1
1176 * while a scrub is in progress, it scrubs the block right then.
1179 dsl_scan_ddt(dsl_scan_t
*scn
, dmu_tx_t
*tx
)
1181 ddt_bookmark_t
*ddb
= &scn
->scn_phys
.scn_ddt_bookmark
;
1182 ddt_entry_t dde
= { 0 };
1186 while ((error
= ddt_walk(scn
->scn_dp
->dp_spa
, ddb
, &dde
)) == 0) {
1189 if (ddb
->ddb_class
> scn
->scn_phys
.scn_ddt_class_max
)
1191 dprintf("visiting ddb=%llu/%llu/%llu/%llx\n",
1192 (longlong_t
)ddb
->ddb_class
,
1193 (longlong_t
)ddb
->ddb_type
,
1194 (longlong_t
)ddb
->ddb_checksum
,
1195 (longlong_t
)ddb
->ddb_cursor
);
1197 /* There should be no pending changes to the dedup table */
1198 ddt
= scn
->scn_dp
->dp_spa
->spa_ddt
[ddb
->ddb_checksum
];
1199 ASSERT(avl_first(&ddt
->ddt_tree
) == NULL
);
1201 dsl_scan_ddt_entry(scn
, ddb
->ddb_checksum
, &dde
, tx
);
1204 if (dsl_scan_check_pause(scn
, NULL
))
1208 zfs_dbgmsg("scanned %llu ddt entries with class_max = %u; pausing=%u",
1209 (longlong_t
)n
, (int)scn
->scn_phys
.scn_ddt_class_max
,
1210 (int)scn
->scn_pausing
);
1212 ASSERT(error
== 0 || error
== ENOENT
);
1213 ASSERT(error
!= ENOENT
||
1214 ddb
->ddb_class
> scn
->scn_phys
.scn_ddt_class_max
);
1219 dsl_scan_ddt_entry(dsl_scan_t
*scn
, enum zio_checksum checksum
,
1220 ddt_entry_t
*dde
, dmu_tx_t
*tx
)
1222 const ddt_key_t
*ddk
= &dde
->dde_key
;
1223 ddt_phys_t
*ddp
= dde
->dde_phys
;
1225 zbookmark_phys_t zb
= { 0 };
1227 if (scn
->scn_phys
.scn_state
!= DSS_SCANNING
)
1230 for (int p
= 0; p
< DDT_PHYS_TYPES
; p
++, ddp
++) {
1231 if (ddp
->ddp_phys_birth
== 0 ||
1232 ddp
->ddp_phys_birth
> scn
->scn_phys
.scn_max_txg
)
1234 ddt_bp_create(checksum
, ddk
, ddp
, &bp
);
1236 scn
->scn_visited_this_txg
++;
1237 scan_funcs
[scn
->scn_phys
.scn_func
](scn
->scn_dp
, &bp
, &zb
);
1242 dsl_scan_visit(dsl_scan_t
*scn
, dmu_tx_t
*tx
)
1244 dsl_pool_t
*dp
= scn
->scn_dp
;
1248 if (scn
->scn_phys
.scn_ddt_bookmark
.ddb_class
<=
1249 scn
->scn_phys
.scn_ddt_class_max
) {
1250 scn
->scn_phys
.scn_cur_min_txg
= scn
->scn_phys
.scn_min_txg
;
1251 scn
->scn_phys
.scn_cur_max_txg
= scn
->scn_phys
.scn_max_txg
;
1252 dsl_scan_ddt(scn
, tx
);
1253 if (scn
->scn_pausing
)
1257 if (scn
->scn_phys
.scn_bookmark
.zb_objset
== DMU_META_OBJSET
) {
1258 /* First do the MOS & ORIGIN */
1260 scn
->scn_phys
.scn_cur_min_txg
= scn
->scn_phys
.scn_min_txg
;
1261 scn
->scn_phys
.scn_cur_max_txg
= scn
->scn_phys
.scn_max_txg
;
1262 dsl_scan_visit_rootbp(scn
, NULL
,
1263 &dp
->dp_meta_rootbp
, tx
);
1264 spa_set_rootblkptr(dp
->dp_spa
, &dp
->dp_meta_rootbp
);
1265 if (scn
->scn_pausing
)
1268 if (spa_version(dp
->dp_spa
) < SPA_VERSION_DSL_SCRUB
) {
1269 VERIFY0(dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
1270 enqueue_cb
, tx
, DS_FIND_CHILDREN
));
1272 dsl_scan_visitds(scn
,
1273 dp
->dp_origin_snap
->ds_object
, tx
);
1275 ASSERT(!scn
->scn_pausing
);
1276 } else if (scn
->scn_phys
.scn_bookmark
.zb_objset
!=
1277 ZB_DESTROYED_OBJSET
) {
1279 * If we were paused, continue from here. Note if the
1280 * ds we were paused on was deleted, the zb_objset may
1281 * be -1, so we will skip this and find a new objset
1284 dsl_scan_visitds(scn
, scn
->scn_phys
.scn_bookmark
.zb_objset
, tx
);
1285 if (scn
->scn_pausing
)
1290 * In case we were paused right at the end of the ds, zero the
1291 * bookmark so we don't think that we're still trying to resume.
1293 bzero(&scn
->scn_phys
.scn_bookmark
, sizeof (zbookmark_phys_t
));
1295 /* keep pulling things out of the zap-object-as-queue */
1296 while (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
1297 scn
->scn_phys
.scn_queue_obj
),
1298 zap_cursor_retrieve(&zc
, &za
) == 0) {
1302 dsobj
= strtonum(za
.za_name
, NULL
);
1303 VERIFY3U(0, ==, zap_remove_int(dp
->dp_meta_objset
,
1304 scn
->scn_phys
.scn_queue_obj
, dsobj
, tx
));
1306 /* Set up min/max txg */
1307 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp
, dsobj
, FTAG
, &ds
));
1308 if (za
.za_first_integer
!= 0) {
1309 scn
->scn_phys
.scn_cur_min_txg
=
1310 MAX(scn
->scn_phys
.scn_min_txg
,
1311 za
.za_first_integer
);
1313 scn
->scn_phys
.scn_cur_min_txg
=
1314 MAX(scn
->scn_phys
.scn_min_txg
,
1315 dsl_dataset_phys(ds
)->ds_prev_snap_txg
);
1317 scn
->scn_phys
.scn_cur_max_txg
= dsl_scan_ds_maxtxg(ds
);
1318 dsl_dataset_rele(ds
, FTAG
);
1320 dsl_scan_visitds(scn
, dsobj
, tx
);
1321 zap_cursor_fini(&zc
);
1322 if (scn
->scn_pausing
)
1325 zap_cursor_fini(&zc
);
1329 dsl_scan_free_should_pause(dsl_scan_t
*scn
)
1331 uint64_t elapsed_nanosecs
;
1336 if (scn
->scn_visited_this_txg
>= zfs_free_max_blocks
)
1339 elapsed_nanosecs
= gethrtime() - scn
->scn_sync_start_time
;
1340 return (elapsed_nanosecs
/ NANOSEC
> zfs_txg_timeout
||
1341 (NSEC2MSEC(elapsed_nanosecs
) > zfs_free_min_time_ms
&&
1342 txg_sync_waiting(scn
->scn_dp
)) ||
1343 spa_shutting_down(scn
->scn_dp
->dp_spa
));
1347 dsl_scan_free_block_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
1349 dsl_scan_t
*scn
= arg
;
1351 if (!scn
->scn_is_bptree
||
1352 (BP_GET_LEVEL(bp
) == 0 && BP_GET_TYPE(bp
) != DMU_OT_OBJSET
)) {
1353 if (dsl_scan_free_should_pause(scn
))
1354 return (SET_ERROR(ERESTART
));
1357 zio_nowait(zio_free_sync(scn
->scn_zio_root
, scn
->scn_dp
->dp_spa
,
1358 dmu_tx_get_txg(tx
), bp
, 0));
1359 dsl_dir_diduse_space(tx
->tx_pool
->dp_free_dir
, DD_USED_HEAD
,
1360 -bp_get_dsize_sync(scn
->scn_dp
->dp_spa
, bp
),
1361 -BP_GET_PSIZE(bp
), -BP_GET_UCSIZE(bp
), tx
);
1362 scn
->scn_visited_this_txg
++;
1367 dsl_scan_active(dsl_scan_t
*scn
)
1369 spa_t
*spa
= scn
->scn_dp
->dp_spa
;
1370 uint64_t used
= 0, comp
, uncomp
;
1372 if (spa
->spa_load_state
!= SPA_LOAD_NONE
)
1374 if (spa_shutting_down(spa
))
1376 if (scn
->scn_phys
.scn_state
== DSS_SCANNING
||
1377 (scn
->scn_async_destroying
&& !scn
->scn_async_stalled
))
1380 if (spa_version(scn
->scn_dp
->dp_spa
) >= SPA_VERSION_DEADLISTS
) {
1381 (void) bpobj_space(&scn
->scn_dp
->dp_free_bpobj
,
1382 &used
, &comp
, &uncomp
);
1388 dsl_scan_sync(dsl_pool_t
*dp
, dmu_tx_t
*tx
)
1390 dsl_scan_t
*scn
= dp
->dp_scan
;
1391 spa_t
*spa
= dp
->dp_spa
;
1395 * Check for scn_restart_txg before checking spa_load_state, so
1396 * that we can restart an old-style scan while the pool is being
1397 * imported (see dsl_scan_init).
1399 if (scn
->scn_restart_txg
!= 0 &&
1400 scn
->scn_restart_txg
<= tx
->tx_txg
) {
1401 pool_scan_func_t func
= POOL_SCAN_SCRUB
;
1402 dsl_scan_done(scn
, B_FALSE
, tx
);
1403 if (vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
))
1404 func
= POOL_SCAN_RESILVER
;
1405 zfs_dbgmsg("restarting scan func=%u txg=%llu",
1407 dsl_scan_setup_sync(&func
, tx
);
1411 * If the scan is inactive due to a stalled async destroy, try again.
1413 if ((!scn
->scn_async_stalled
&& !dsl_scan_active(scn
)) ||
1414 spa_sync_pass(dp
->dp_spa
) > 1)
1417 scn
->scn_visited_this_txg
= 0;
1418 scn
->scn_pausing
= B_FALSE
;
1419 scn
->scn_sync_start_time
= gethrtime();
1420 spa
->spa_scrub_active
= B_TRUE
;
1423 * First process the async destroys. If we pause, don't do
1424 * any scrubbing or resilvering. This ensures that there are no
1425 * async destroys while we are scanning, so the scan code doesn't
1426 * have to worry about traversing it. It is also faster to free the
1427 * blocks than to scrub them.
1429 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DEADLISTS
) {
1430 scn
->scn_is_bptree
= B_FALSE
;
1431 scn
->scn_zio_root
= zio_root(dp
->dp_spa
, NULL
,
1432 NULL
, ZIO_FLAG_MUSTSUCCEED
);
1433 err
= bpobj_iterate(&dp
->dp_free_bpobj
,
1434 dsl_scan_free_block_cb
, scn
, tx
);
1435 VERIFY3U(0, ==, zio_wait(scn
->scn_zio_root
));
1437 if (err
!= 0 && err
!= ERESTART
)
1438 zfs_panic_recover("error %u from bpobj_iterate()", err
);
1441 if (err
== 0 && spa_feature_is_active(spa
, SPA_FEATURE_ASYNC_DESTROY
)) {
1442 ASSERT(scn
->scn_async_destroying
);
1443 scn
->scn_is_bptree
= B_TRUE
;
1444 scn
->scn_zio_root
= zio_root(dp
->dp_spa
, NULL
,
1445 NULL
, ZIO_FLAG_MUSTSUCCEED
);
1446 err
= bptree_iterate(dp
->dp_meta_objset
,
1447 dp
->dp_bptree_obj
, B_TRUE
, dsl_scan_free_block_cb
, scn
, tx
);
1448 VERIFY0(zio_wait(scn
->scn_zio_root
));
1450 if (err
== EIO
|| err
== ECKSUM
) {
1452 } else if (err
!= 0 && err
!= ERESTART
) {
1453 zfs_panic_recover("error %u from "
1454 "traverse_dataset_destroyed()", err
);
1457 if (bptree_is_empty(dp
->dp_meta_objset
, dp
->dp_bptree_obj
)) {
1458 /* finished; deactivate async destroy feature */
1459 spa_feature_decr(spa
, SPA_FEATURE_ASYNC_DESTROY
, tx
);
1460 ASSERT(!spa_feature_is_active(spa
,
1461 SPA_FEATURE_ASYNC_DESTROY
));
1462 VERIFY0(zap_remove(dp
->dp_meta_objset
,
1463 DMU_POOL_DIRECTORY_OBJECT
,
1464 DMU_POOL_BPTREE_OBJ
, tx
));
1465 VERIFY0(bptree_free(dp
->dp_meta_objset
,
1466 dp
->dp_bptree_obj
, tx
));
1467 dp
->dp_bptree_obj
= 0;
1468 scn
->scn_async_destroying
= B_FALSE
;
1469 scn
->scn_async_stalled
= B_FALSE
;
1472 * If we didn't make progress, mark the async
1473 * destroy as stalled, so that we will not initiate
1474 * a spa_sync() on its behalf. Note that we only
1475 * check this if we are not finished, because if the
1476 * bptree had no blocks for us to visit, we can
1477 * finish without "making progress".
1479 scn
->scn_async_stalled
=
1480 (scn
->scn_visited_this_txg
== 0);
1483 if (scn
->scn_visited_this_txg
) {
1484 zfs_dbgmsg("freed %llu blocks in %llums from "
1485 "free_bpobj/bptree txg %llu; err=%u",
1486 (longlong_t
)scn
->scn_visited_this_txg
,
1488 NSEC2MSEC(gethrtime() - scn
->scn_sync_start_time
),
1489 (longlong_t
)tx
->tx_txg
, err
);
1490 scn
->scn_visited_this_txg
= 0;
1493 * Write out changes to the DDT that may be required as a
1494 * result of the blocks freed. This ensures that the DDT
1495 * is clean when a scrub/resilver runs.
1497 ddt_sync(spa
, tx
->tx_txg
);
1501 if (!scn
->scn_async_destroying
&& zfs_free_leak_on_eio
&&
1502 (dsl_dir_phys(dp
->dp_free_dir
)->dd_used_bytes
!= 0 ||
1503 dsl_dir_phys(dp
->dp_free_dir
)->dd_compressed_bytes
!= 0 ||
1504 dsl_dir_phys(dp
->dp_free_dir
)->dd_uncompressed_bytes
!= 0)) {
1506 * We have finished background destroying, but there is still
1507 * some space left in the dp_free_dir. Transfer this leaked
1508 * space to the dp_leak_dir.
1510 if (dp
->dp_leak_dir
== NULL
) {
1511 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
1512 (void) dsl_dir_create_sync(dp
, dp
->dp_root_dir
,
1514 VERIFY0(dsl_pool_open_special_dir(dp
,
1515 LEAK_DIR_NAME
, &dp
->dp_leak_dir
));
1516 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
1518 dsl_dir_diduse_space(dp
->dp_leak_dir
, DD_USED_HEAD
,
1519 dsl_dir_phys(dp
->dp_free_dir
)->dd_used_bytes
,
1520 dsl_dir_phys(dp
->dp_free_dir
)->dd_compressed_bytes
,
1521 dsl_dir_phys(dp
->dp_free_dir
)->dd_uncompressed_bytes
, tx
);
1522 dsl_dir_diduse_space(dp
->dp_free_dir
, DD_USED_HEAD
,
1523 -dsl_dir_phys(dp
->dp_free_dir
)->dd_used_bytes
,
1524 -dsl_dir_phys(dp
->dp_free_dir
)->dd_compressed_bytes
,
1525 -dsl_dir_phys(dp
->dp_free_dir
)->dd_uncompressed_bytes
, tx
);
1527 if (!scn
->scn_async_destroying
) {
1528 /* finished; verify that space accounting went to zero */
1529 ASSERT0(dsl_dir_phys(dp
->dp_free_dir
)->dd_used_bytes
);
1530 ASSERT0(dsl_dir_phys(dp
->dp_free_dir
)->dd_compressed_bytes
);
1531 ASSERT0(dsl_dir_phys(dp
->dp_free_dir
)->dd_uncompressed_bytes
);
1534 if (scn
->scn_phys
.scn_state
!= DSS_SCANNING
)
1537 if (scn
->scn_done_txg
== tx
->tx_txg
) {
1538 ASSERT(!scn
->scn_pausing
);
1539 /* finished with scan. */
1540 zfs_dbgmsg("txg %llu scan complete", tx
->tx_txg
);
1541 dsl_scan_done(scn
, B_TRUE
, tx
);
1542 ASSERT3U(spa
->spa_scrub_inflight
, ==, 0);
1543 dsl_scan_sync_state(scn
, tx
);
1547 if (scn
->scn_phys
.scn_ddt_bookmark
.ddb_class
<=
1548 scn
->scn_phys
.scn_ddt_class_max
) {
1549 zfs_dbgmsg("doing scan sync txg %llu; "
1550 "ddt bm=%llu/%llu/%llu/%llx",
1551 (longlong_t
)tx
->tx_txg
,
1552 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_class
,
1553 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_type
,
1554 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_checksum
,
1555 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_cursor
);
1556 ASSERT(scn
->scn_phys
.scn_bookmark
.zb_objset
== 0);
1557 ASSERT(scn
->scn_phys
.scn_bookmark
.zb_object
== 0);
1558 ASSERT(scn
->scn_phys
.scn_bookmark
.zb_level
== 0);
1559 ASSERT(scn
->scn_phys
.scn_bookmark
.zb_blkid
== 0);
1561 zfs_dbgmsg("doing scan sync txg %llu; bm=%llu/%llu/%llu/%llu",
1562 (longlong_t
)tx
->tx_txg
,
1563 (longlong_t
)scn
->scn_phys
.scn_bookmark
.zb_objset
,
1564 (longlong_t
)scn
->scn_phys
.scn_bookmark
.zb_object
,
1565 (longlong_t
)scn
->scn_phys
.scn_bookmark
.zb_level
,
1566 (longlong_t
)scn
->scn_phys
.scn_bookmark
.zb_blkid
);
1569 scn
->scn_zio_root
= zio_root(dp
->dp_spa
, NULL
,
1570 NULL
, ZIO_FLAG_CANFAIL
);
1571 dsl_pool_config_enter(dp
, FTAG
);
1572 dsl_scan_visit(scn
, tx
);
1573 dsl_pool_config_exit(dp
, FTAG
);
1574 (void) zio_wait(scn
->scn_zio_root
);
1575 scn
->scn_zio_root
= NULL
;
1577 zfs_dbgmsg("visited %llu blocks in %llums",
1578 (longlong_t
)scn
->scn_visited_this_txg
,
1579 (longlong_t
)NSEC2MSEC(gethrtime() - scn
->scn_sync_start_time
));
1581 if (!scn
->scn_pausing
) {
1582 scn
->scn_done_txg
= tx
->tx_txg
+ 1;
1583 zfs_dbgmsg("txg %llu traversal complete, waiting till txg %llu",
1584 tx
->tx_txg
, scn
->scn_done_txg
);
1587 if (DSL_SCAN_IS_SCRUB_RESILVER(scn
)) {
1588 mutex_enter(&spa
->spa_scrub_lock
);
1589 while (spa
->spa_scrub_inflight
> 0) {
1590 cv_wait(&spa
->spa_scrub_io_cv
,
1591 &spa
->spa_scrub_lock
);
1593 mutex_exit(&spa
->spa_scrub_lock
);
1596 dsl_scan_sync_state(scn
, tx
);
1600 * This will start a new scan, or restart an existing one.
1603 dsl_resilver_restart(dsl_pool_t
*dp
, uint64_t txg
)
1607 tx
= dmu_tx_create_dd(dp
->dp_mos_dir
);
1608 VERIFY(0 == dmu_tx_assign(tx
, TXG_WAIT
));
1610 txg
= dmu_tx_get_txg(tx
);
1611 dp
->dp_scan
->scn_restart_txg
= txg
;
1614 dp
->dp_scan
->scn_restart_txg
= txg
;
1616 zfs_dbgmsg("restarting resilver txg=%llu", txg
);
1620 dsl_scan_resilvering(dsl_pool_t
*dp
)
1622 return (dp
->dp_scan
->scn_phys
.scn_state
== DSS_SCANNING
&&
1623 dp
->dp_scan
->scn_phys
.scn_func
== POOL_SCAN_RESILVER
);
1631 count_block(zfs_all_blkstats_t
*zab
, const blkptr_t
*bp
)
1636 * If we resume after a reboot, zab will be NULL; don't record
1637 * incomplete stats in that case.
1642 for (i
= 0; i
< 4; i
++) {
1643 int l
= (i
< 2) ? BP_GET_LEVEL(bp
) : DN_MAX_LEVELS
;
1644 int t
= (i
& 1) ? BP_GET_TYPE(bp
) : DMU_OT_TOTAL
;
1645 if (t
& DMU_OT_NEWTYPE
)
1647 zfs_blkstat_t
*zb
= &zab
->zab_type
[l
][t
];
1651 zb
->zb_asize
+= BP_GET_ASIZE(bp
);
1652 zb
->zb_lsize
+= BP_GET_LSIZE(bp
);
1653 zb
->zb_psize
+= BP_GET_PSIZE(bp
);
1654 zb
->zb_gangs
+= BP_COUNT_GANG(bp
);
1656 switch (BP_GET_NDVAS(bp
)) {
1658 if (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
1659 DVA_GET_VDEV(&bp
->blk_dva
[1]))
1660 zb
->zb_ditto_2_of_2_samevdev
++;
1663 equal
= (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
1664 DVA_GET_VDEV(&bp
->blk_dva
[1])) +
1665 (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
1666 DVA_GET_VDEV(&bp
->blk_dva
[2])) +
1667 (DVA_GET_VDEV(&bp
->blk_dva
[1]) ==
1668 DVA_GET_VDEV(&bp
->blk_dva
[2]));
1670 zb
->zb_ditto_2_of_3_samevdev
++;
1671 else if (equal
== 3)
1672 zb
->zb_ditto_3_of_3_samevdev
++;
1679 dsl_scan_scrub_done(zio_t
*zio
)
1681 spa_t
*spa
= zio
->io_spa
;
1683 zio_data_buf_free(zio
->io_data
, zio
->io_size
);
1685 mutex_enter(&spa
->spa_scrub_lock
);
1686 spa
->spa_scrub_inflight
--;
1687 cv_broadcast(&spa
->spa_scrub_io_cv
);
1689 if (zio
->io_error
&& (zio
->io_error
!= ECKSUM
||
1690 !(zio
->io_flags
& ZIO_FLAG_SPECULATIVE
))) {
1691 spa
->spa_dsl_pool
->dp_scan
->scn_phys
.scn_errors
++;
1693 mutex_exit(&spa
->spa_scrub_lock
);
1697 dsl_scan_scrub_cb(dsl_pool_t
*dp
,
1698 const blkptr_t
*bp
, const zbookmark_phys_t
*zb
)
1700 dsl_scan_t
*scn
= dp
->dp_scan
;
1701 size_t size
= BP_GET_PSIZE(bp
);
1702 spa_t
*spa
= dp
->dp_spa
;
1703 uint64_t phys_birth
= BP_PHYSICAL_BIRTH(bp
);
1705 int zio_flags
= ZIO_FLAG_SCAN_THREAD
| ZIO_FLAG_RAW
| ZIO_FLAG_CANFAIL
;
1708 if (phys_birth
<= scn
->scn_phys
.scn_min_txg
||
1709 phys_birth
>= scn
->scn_phys
.scn_max_txg
)
1712 count_block(dp
->dp_blkstats
, bp
);
1714 if (BP_IS_EMBEDDED(bp
))
1717 ASSERT(DSL_SCAN_IS_SCRUB_RESILVER(scn
));
1718 if (scn
->scn_phys
.scn_func
== POOL_SCAN_SCRUB
) {
1719 zio_flags
|= ZIO_FLAG_SCRUB
;
1721 scan_delay
= zfs_scrub_delay
;
1723 ASSERT3U(scn
->scn_phys
.scn_func
, ==, POOL_SCAN_RESILVER
);
1724 zio_flags
|= ZIO_FLAG_RESILVER
;
1726 scan_delay
= zfs_resilver_delay
;
1729 /* If it's an intent log block, failure is expected. */
1730 if (zb
->zb_level
== ZB_ZIL_LEVEL
)
1731 zio_flags
|= ZIO_FLAG_SPECULATIVE
;
1733 for (int d
= 0; d
< BP_GET_NDVAS(bp
); d
++) {
1734 vdev_t
*vd
= vdev_lookup_top(spa
,
1735 DVA_GET_VDEV(&bp
->blk_dva
[d
]));
1738 * Keep track of how much data we've examined so that
1739 * zpool(1M) status can make useful progress reports.
1741 scn
->scn_phys
.scn_examined
+= DVA_GET_ASIZE(&bp
->blk_dva
[d
]);
1742 spa
->spa_scan_pass_exam
+= DVA_GET_ASIZE(&bp
->blk_dva
[d
]);
1744 /* if it's a resilver, this may not be in the target range */
1746 if (DVA_GET_GANG(&bp
->blk_dva
[d
])) {
1748 * Gang members may be spread across multiple
1749 * vdevs, so the best estimate we have is the
1750 * scrub range, which has already been checked.
1751 * XXX -- it would be better to change our
1752 * allocation policy to ensure that all
1753 * gang members reside on the same vdev.
1757 needs_io
= vdev_dtl_contains(vd
, DTL_PARTIAL
,
1763 if (needs_io
&& !zfs_no_scrub_io
) {
1764 vdev_t
*rvd
= spa
->spa_root_vdev
;
1765 uint64_t maxinflight
= rvd
->vdev_children
* zfs_top_maxinflight
;
1766 void *data
= zio_data_buf_alloc(size
);
1768 mutex_enter(&spa
->spa_scrub_lock
);
1769 while (spa
->spa_scrub_inflight
>= maxinflight
)
1770 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
1771 spa
->spa_scrub_inflight
++;
1772 mutex_exit(&spa
->spa_scrub_lock
);
1775 * If we're seeing recent (zfs_scan_idle) "important" I/Os
1776 * then throttle our workload to limit the impact of a scan.
1778 if (ddi_get_lbolt64() - spa
->spa_last_io
<= zfs_scan_idle
)
1781 zio_nowait(zio_read(NULL
, spa
, bp
, data
, size
,
1782 dsl_scan_scrub_done
, NULL
, ZIO_PRIORITY_SCRUB
,
1786 /* do not relocate this block */
1791 dsl_scan(dsl_pool_t
*dp
, pool_scan_func_t func
)
1793 spa_t
*spa
= dp
->dp_spa
;
1796 * Purge all vdev caches and probe all devices. We do this here
1797 * rather than in sync context because this requires a writer lock
1798 * on the spa_config lock, which we can't do from sync context. The
1799 * spa_scrub_reopen flag indicates that vdev_open() should not
1800 * attempt to start another scrub.
1802 spa_vdev_state_enter(spa
, SCL_NONE
);
1803 spa
->spa_scrub_reopen
= B_TRUE
;
1804 vdev_reopen(spa
->spa_root_vdev
);
1805 spa
->spa_scrub_reopen
= B_FALSE
;
1806 (void) spa_vdev_state_exit(spa
, NULL
, 0);
1808 return (dsl_sync_task(spa_name(spa
), dsl_scan_setup_check
,
1809 dsl_scan_setup_sync
, &func
, 0, ZFS_SPACE_CHECK_NONE
));