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) 2013 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
*, const zbookmark_t
*);
55 static scan_cb_t dsl_scan_scrub_cb
;
56 static void dsl_scan_cancel_sync(void *, dmu_tx_t
*);
57 static void dsl_scan_sync_state(dsl_scan_t
*, dmu_tx_t
*tx
);
59 int zfs_top_maxinflight
= 32; /* maximum I/Os per top-level */
60 int zfs_resilver_delay
= 2; /* number of ticks to delay resilver */
61 int zfs_scrub_delay
= 4; /* number of ticks to delay scrub */
62 int zfs_scan_idle
= 50; /* idle window in clock ticks */
64 int zfs_scan_min_time_ms
= 1000; /* min millisecs to scrub per txg */
65 int zfs_free_min_time_ms
= 1000; /* min millisecs to free per txg */
66 int zfs_resilver_min_time_ms
= 3000; /* min millisecs to resilver per txg */
67 int zfs_no_scrub_io
= B_FALSE
; /* set to disable scrub i/o */
68 int zfs_no_scrub_prefetch
= B_FALSE
; /* set to disable srub prefetching */
69 enum ddt_class zfs_scrub_ddt_class_max
= DDT_CLASS_DUPLICATE
;
70 int dsl_scan_delay_completion
= B_FALSE
; /* set to delay scan completion */
72 #define DSL_SCAN_IS_SCRUB_RESILVER(scn) \
73 ((scn)->scn_phys.scn_func == POOL_SCAN_SCRUB || \
74 (scn)->scn_phys.scn_func == POOL_SCAN_RESILVER)
76 /* the order has to match pool_scan_type */
77 static scan_cb_t
*scan_funcs
[POOL_SCAN_FUNCS
] = {
79 dsl_scan_scrub_cb
, /* POOL_SCAN_SCRUB */
80 dsl_scan_scrub_cb
, /* POOL_SCAN_RESILVER */
84 dsl_scan_init(dsl_pool_t
*dp
, uint64_t txg
)
88 spa_t
*spa
= dp
->dp_spa
;
91 scn
= dp
->dp_scan
= kmem_zalloc(sizeof (dsl_scan_t
), KM_SLEEP
);
95 * It's possible that we're resuming a scan after a reboot so
96 * make sure that the scan_async_destroying flag is initialized
99 ASSERT(!scn
->scn_async_destroying
);
100 scn
->scn_async_destroying
= spa_feature_is_active(dp
->dp_spa
,
101 SPA_FEATURE_ASYNC_DESTROY
);
103 err
= zap_lookup(dp
->dp_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
104 "scrub_func", sizeof (uint64_t), 1, &f
);
107 * There was an old-style scrub in progress. Restart a
108 * new-style scrub from the beginning.
110 scn
->scn_restart_txg
= txg
;
111 zfs_dbgmsg("old-style scrub was in progress; "
112 "restarting new-style scrub in txg %llu",
113 scn
->scn_restart_txg
);
116 * Load the queue obj from the old location so that it
117 * can be freed by dsl_scan_done().
119 (void) zap_lookup(dp
->dp_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
120 "scrub_queue", sizeof (uint64_t), 1,
121 &scn
->scn_phys
.scn_queue_obj
);
123 err
= zap_lookup(dp
->dp_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
124 DMU_POOL_SCAN
, sizeof (uint64_t), SCAN_PHYS_NUMINTS
,
127 * Detect if the pool contains the signature of #2094. If it
128 * does properly update the scn->scn_phys structure and notify
129 * the administrator by setting an errata for the pool.
131 if (err
== EOVERFLOW
) {
132 uint64_t zaptmp
[SCAN_PHYS_NUMINTS
+ 1];
133 VERIFY3S(SCAN_PHYS_NUMINTS
, ==, 24);
134 VERIFY3S(offsetof(dsl_scan_phys_t
, scn_flags
), ==,
135 (23 * sizeof (uint64_t)));
137 err
= zap_lookup(dp
->dp_meta_objset
,
138 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SCAN
,
139 sizeof (uint64_t), SCAN_PHYS_NUMINTS
+ 1, &zaptmp
);
141 uint64_t overflow
= zaptmp
[SCAN_PHYS_NUMINTS
];
143 if (overflow
& ~DSL_SCAN_FLAGS_MASK
||
144 scn
->scn_async_destroying
) {
146 ZPOOL_ERRATA_ZOL_2094_ASYNC_DESTROY
;
150 bcopy(zaptmp
, &scn
->scn_phys
,
151 SCAN_PHYS_NUMINTS
* sizeof (uint64_t));
152 scn
->scn_phys
.scn_flags
= overflow
;
154 /* Required scrub already in progress. */
155 if (scn
->scn_phys
.scn_state
== DSS_FINISHED
||
156 scn
->scn_phys
.scn_state
== DSS_CANCELED
)
158 ZPOOL_ERRATA_ZOL_2094_SCRUB
;
167 if (scn
->scn_phys
.scn_state
== DSS_SCANNING
&&
168 spa_prev_software_version(dp
->dp_spa
) < SPA_VERSION_SCAN
) {
170 * A new-type scrub was in progress on an old
171 * pool, and the pool was accessed by old
172 * software. Restart from the beginning, since
173 * the old software may have changed the pool in
176 scn
->scn_restart_txg
= txg
;
177 zfs_dbgmsg("new-style scrub was modified "
178 "by old software; restarting in txg %llu",
179 scn
->scn_restart_txg
);
183 spa_scan_stat_init(spa
);
188 dsl_scan_fini(dsl_pool_t
*dp
)
191 kmem_free(dp
->dp_scan
, sizeof (dsl_scan_t
));
198 dsl_scan_setup_check(void *arg
, dmu_tx_t
*tx
)
200 dsl_scan_t
*scn
= dmu_tx_pool(tx
)->dp_scan
;
202 if (scn
->scn_phys
.scn_state
== DSS_SCANNING
)
203 return (SET_ERROR(EBUSY
));
209 dsl_scan_setup_sync(void *arg
, dmu_tx_t
*tx
)
211 dsl_scan_t
*scn
= dmu_tx_pool(tx
)->dp_scan
;
212 pool_scan_func_t
*funcp
= arg
;
213 dmu_object_type_t ot
= 0;
214 dsl_pool_t
*dp
= scn
->scn_dp
;
215 spa_t
*spa
= dp
->dp_spa
;
217 ASSERT(scn
->scn_phys
.scn_state
!= DSS_SCANNING
);
218 ASSERT(*funcp
> POOL_SCAN_NONE
&& *funcp
< POOL_SCAN_FUNCS
);
219 bzero(&scn
->scn_phys
, sizeof (scn
->scn_phys
));
220 scn
->scn_phys
.scn_func
= *funcp
;
221 scn
->scn_phys
.scn_state
= DSS_SCANNING
;
222 scn
->scn_phys
.scn_min_txg
= 0;
223 scn
->scn_phys
.scn_max_txg
= tx
->tx_txg
;
224 scn
->scn_phys
.scn_ddt_class_max
= DDT_CLASSES
- 1; /* the entire DDT */
225 scn
->scn_phys
.scn_start_time
= gethrestime_sec();
226 scn
->scn_phys
.scn_errors
= 0;
227 scn
->scn_phys
.scn_to_examine
= spa
->spa_root_vdev
->vdev_stat
.vs_alloc
;
228 scn
->scn_restart_txg
= 0;
229 scn
->scn_done_txg
= 0;
230 spa_scan_stat_init(spa
);
232 if (DSL_SCAN_IS_SCRUB_RESILVER(scn
)) {
233 scn
->scn_phys
.scn_ddt_class_max
= zfs_scrub_ddt_class_max
;
235 /* rewrite all disk labels */
236 vdev_config_dirty(spa
->spa_root_vdev
);
238 if (vdev_resilver_needed(spa
->spa_root_vdev
,
239 &scn
->scn_phys
.scn_min_txg
, &scn
->scn_phys
.scn_max_txg
)) {
240 spa_event_notify(spa
, NULL
,
241 FM_EREPORT_ZFS_RESILVER_START
);
243 spa_event_notify(spa
, NULL
,
244 FM_EREPORT_ZFS_SCRUB_START
);
247 spa
->spa_scrub_started
= B_TRUE
;
249 * If this is an incremental scrub, limit the DDT scrub phase
250 * to just the auto-ditto class (for correctness); the rest
251 * of the scrub should go faster using top-down pruning.
253 if (scn
->scn_phys
.scn_min_txg
> TXG_INITIAL
)
254 scn
->scn_phys
.scn_ddt_class_max
= DDT_CLASS_DITTO
;
258 /* back to the generic stuff */
260 if (dp
->dp_blkstats
== NULL
) {
261 dp
->dp_blkstats
= kmem_alloc(sizeof (zfs_all_blkstats_t
),
262 KM_PUSHPAGE
| KM_NODEBUG
);
264 bzero(dp
->dp_blkstats
, sizeof (zfs_all_blkstats_t
));
266 if (spa_version(spa
) < SPA_VERSION_DSL_SCRUB
)
267 ot
= DMU_OT_ZAP_OTHER
;
269 scn
->scn_phys
.scn_queue_obj
= zap_create(dp
->dp_meta_objset
,
270 ot
? ot
: DMU_OT_SCAN_QUEUE
, DMU_OT_NONE
, 0, tx
);
272 dsl_scan_sync_state(scn
, tx
);
274 spa_history_log_internal(spa
, "scan setup", tx
,
275 "func=%u mintxg=%llu maxtxg=%llu",
276 *funcp
, scn
->scn_phys
.scn_min_txg
, scn
->scn_phys
.scn_max_txg
);
281 dsl_scan_done(dsl_scan_t
*scn
, boolean_t complete
, dmu_tx_t
*tx
)
283 static const char *old_names
[] = {
285 "scrub_ddt_bookmark",
286 "scrub_ddt_class_max",
295 dsl_pool_t
*dp
= scn
->scn_dp
;
296 spa_t
*spa
= dp
->dp_spa
;
299 /* Remove any remnants of an old-style scrub. */
300 for (i
= 0; old_names
[i
]; i
++) {
301 (void) zap_remove(dp
->dp_meta_objset
,
302 DMU_POOL_DIRECTORY_OBJECT
, old_names
[i
], tx
);
305 if (scn
->scn_phys
.scn_queue_obj
!= 0) {
306 VERIFY(0 == dmu_object_free(dp
->dp_meta_objset
,
307 scn
->scn_phys
.scn_queue_obj
, tx
));
308 scn
->scn_phys
.scn_queue_obj
= 0;
312 * If we were "restarted" from a stopped state, don't bother
313 * with anything else.
315 if (scn
->scn_phys
.scn_state
!= DSS_SCANNING
)
319 scn
->scn_phys
.scn_state
= DSS_FINISHED
;
321 scn
->scn_phys
.scn_state
= DSS_CANCELED
;
323 spa_history_log_internal(spa
, "scan done", tx
,
324 "complete=%u", complete
);
326 if (DSL_SCAN_IS_SCRUB_RESILVER(scn
)) {
327 mutex_enter(&spa
->spa_scrub_lock
);
328 while (spa
->spa_scrub_inflight
> 0) {
329 cv_wait(&spa
->spa_scrub_io_cv
,
330 &spa
->spa_scrub_lock
);
332 mutex_exit(&spa
->spa_scrub_lock
);
333 spa
->spa_scrub_started
= B_FALSE
;
334 spa
->spa_scrub_active
= B_FALSE
;
337 * If the scrub/resilver completed, update all DTLs to
338 * reflect this. Whether it succeeded or not, vacate
339 * all temporary scrub DTLs.
341 vdev_dtl_reassess(spa
->spa_root_vdev
, tx
->tx_txg
,
342 complete
? scn
->scn_phys
.scn_max_txg
: 0, B_TRUE
);
344 spa_event_notify(spa
, NULL
, scn
->scn_phys
.scn_min_txg
?
345 FM_EREPORT_ZFS_RESILVER_FINISH
:
346 FM_EREPORT_ZFS_SCRUB_FINISH
);
348 spa_errlog_rotate(spa
);
351 * We may have finished replacing a device.
352 * Let the async thread assess this and handle the detach.
354 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
357 scn
->scn_phys
.scn_end_time
= gethrestime_sec();
359 if (spa
->spa_errata
== ZPOOL_ERRATA_ZOL_2094_SCRUB
)
365 dsl_scan_cancel_check(void *arg
, dmu_tx_t
*tx
)
367 dsl_scan_t
*scn
= dmu_tx_pool(tx
)->dp_scan
;
369 if (scn
->scn_phys
.scn_state
!= DSS_SCANNING
)
370 return (SET_ERROR(ENOENT
));
376 dsl_scan_cancel_sync(void *arg
, dmu_tx_t
*tx
)
378 dsl_scan_t
*scn
= dmu_tx_pool(tx
)->dp_scan
;
380 dsl_scan_done(scn
, B_FALSE
, tx
);
381 dsl_scan_sync_state(scn
, tx
);
385 dsl_scan_cancel(dsl_pool_t
*dp
)
387 return (dsl_sync_task(spa_name(dp
->dp_spa
), dsl_scan_cancel_check
,
388 dsl_scan_cancel_sync
, NULL
, 3));
391 static void dsl_scan_visitbp(blkptr_t
*bp
,
392 const zbookmark_t
*zb
, dnode_phys_t
*dnp
, arc_buf_t
*pbuf
,
393 dsl_dataset_t
*ds
, dsl_scan_t
*scn
, dmu_objset_type_t ostype
,
395 inline __attribute__((always_inline
)) static void dsl_scan_visitdnode(
396 dsl_scan_t
*, dsl_dataset_t
*ds
, dmu_objset_type_t ostype
,
397 dnode_phys_t
*dnp
, arc_buf_t
*buf
, uint64_t object
, dmu_tx_t
*tx
);
400 dsl_free(dsl_pool_t
*dp
, uint64_t txg
, const blkptr_t
*bp
)
402 zio_free(dp
->dp_spa
, txg
, bp
);
406 dsl_free_sync(zio_t
*pio
, dsl_pool_t
*dp
, uint64_t txg
, const blkptr_t
*bpp
)
408 ASSERT(dsl_pool_sync_context(dp
));
409 zio_nowait(zio_free_sync(pio
, dp
->dp_spa
, txg
, bpp
, pio
->io_flags
));
413 dsl_scan_ds_maxtxg(dsl_dataset_t
*ds
)
415 uint64_t smt
= ds
->ds_dir
->dd_pool
->dp_scan
->scn_phys
.scn_max_txg
;
416 if (dsl_dataset_is_snapshot(ds
))
417 return (MIN(smt
, ds
->ds_phys
->ds_creation_txg
));
422 dsl_scan_sync_state(dsl_scan_t
*scn
, dmu_tx_t
*tx
)
424 VERIFY0(zap_update(scn
->scn_dp
->dp_meta_objset
,
425 DMU_POOL_DIRECTORY_OBJECT
,
426 DMU_POOL_SCAN
, sizeof (uint64_t), SCAN_PHYS_NUMINTS
,
427 &scn
->scn_phys
, tx
));
431 dsl_scan_check_pause(dsl_scan_t
*scn
, const zbookmark_t
*zb
)
433 uint64_t elapsed_nanosecs
;
436 /* we never skip user/group accounting objects */
437 if (zb
&& (int64_t)zb
->zb_object
< 0)
440 if (scn
->scn_pausing
)
441 return (B_TRUE
); /* we're already pausing */
443 if (!ZB_IS_ZERO(&scn
->scn_phys
.scn_bookmark
))
444 return (B_FALSE
); /* we're resuming */
446 /* We only know how to resume from level-0 blocks. */
447 if (zb
&& zb
->zb_level
!= 0)
450 mintime
= (scn
->scn_phys
.scn_func
== POOL_SCAN_RESILVER
) ?
451 zfs_resilver_min_time_ms
: zfs_scan_min_time_ms
;
452 elapsed_nanosecs
= gethrtime() - scn
->scn_sync_start_time
;
453 if (elapsed_nanosecs
/ NANOSEC
> zfs_txg_timeout
||
454 (NSEC2MSEC(elapsed_nanosecs
) > mintime
&&
455 txg_sync_waiting(scn
->scn_dp
)) ||
456 spa_shutting_down(scn
->scn_dp
->dp_spa
)) {
458 dprintf("pausing at bookmark %llx/%llx/%llx/%llx\n",
459 (longlong_t
)zb
->zb_objset
,
460 (longlong_t
)zb
->zb_object
,
461 (longlong_t
)zb
->zb_level
,
462 (longlong_t
)zb
->zb_blkid
);
463 scn
->scn_phys
.scn_bookmark
= *zb
;
465 dprintf("pausing at DDT bookmark %llx/%llx/%llx/%llx\n",
466 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_class
,
467 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_type
,
468 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_checksum
,
469 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_cursor
);
470 scn
->scn_pausing
= B_TRUE
;
476 typedef struct zil_scan_arg
{
478 zil_header_t
*zsa_zh
;
483 dsl_scan_zil_block(zilog_t
*zilog
, blkptr_t
*bp
, void *arg
, uint64_t claim_txg
)
485 zil_scan_arg_t
*zsa
= arg
;
486 dsl_pool_t
*dp
= zsa
->zsa_dp
;
487 dsl_scan_t
*scn
= dp
->dp_scan
;
488 zil_header_t
*zh
= zsa
->zsa_zh
;
491 if (BP_IS_HOLE(bp
) || bp
->blk_birth
<= scn
->scn_phys
.scn_cur_min_txg
)
495 * One block ("stubby") can be allocated a long time ago; we
496 * want to visit that one because it has been allocated
497 * (on-disk) even if it hasn't been claimed (even though for
498 * scrub there's nothing to do to it).
500 if (claim_txg
== 0 && bp
->blk_birth
>= spa_first_txg(dp
->dp_spa
))
503 SET_BOOKMARK(&zb
, zh
->zh_log
.blk_cksum
.zc_word
[ZIL_ZC_OBJSET
],
504 ZB_ZIL_OBJECT
, ZB_ZIL_LEVEL
, bp
->blk_cksum
.zc_word
[ZIL_ZC_SEQ
]);
506 VERIFY(0 == scan_funcs
[scn
->scn_phys
.scn_func
](dp
, bp
, &zb
));
512 dsl_scan_zil_record(zilog_t
*zilog
, lr_t
*lrc
, void *arg
, uint64_t claim_txg
)
514 if (lrc
->lrc_txtype
== TX_WRITE
) {
515 zil_scan_arg_t
*zsa
= arg
;
516 dsl_pool_t
*dp
= zsa
->zsa_dp
;
517 dsl_scan_t
*scn
= dp
->dp_scan
;
518 zil_header_t
*zh
= zsa
->zsa_zh
;
519 lr_write_t
*lr
= (lr_write_t
*)lrc
;
520 blkptr_t
*bp
= &lr
->lr_blkptr
;
523 if (BP_IS_HOLE(bp
) ||
524 bp
->blk_birth
<= scn
->scn_phys
.scn_cur_min_txg
)
528 * birth can be < claim_txg if this record's txg is
529 * already txg sync'ed (but this log block contains
530 * other records that are not synced)
532 if (claim_txg
== 0 || bp
->blk_birth
< claim_txg
)
535 SET_BOOKMARK(&zb
, zh
->zh_log
.blk_cksum
.zc_word
[ZIL_ZC_OBJSET
],
536 lr
->lr_foid
, ZB_ZIL_LEVEL
,
537 lr
->lr_offset
/ BP_GET_LSIZE(bp
));
539 VERIFY(0 == scan_funcs
[scn
->scn_phys
.scn_func
](dp
, bp
, &zb
));
545 dsl_scan_zil(dsl_pool_t
*dp
, zil_header_t
*zh
)
547 uint64_t claim_txg
= zh
->zh_claim_txg
;
548 zil_scan_arg_t zsa
= { dp
, zh
};
552 * We only want to visit blocks that have been claimed but not yet
553 * replayed (or, in read-only mode, blocks that *would* be claimed).
555 if (claim_txg
== 0 && spa_writeable(dp
->dp_spa
))
558 zilog
= zil_alloc(dp
->dp_meta_objset
, zh
);
560 (void) zil_parse(zilog
, dsl_scan_zil_block
, dsl_scan_zil_record
, &zsa
,
568 dsl_scan_prefetch(dsl_scan_t
*scn
, arc_buf_t
*buf
, blkptr_t
*bp
,
569 uint64_t objset
, uint64_t object
, uint64_t blkid
)
572 uint32_t flags
= ARC_NOWAIT
| ARC_PREFETCH
;
574 if (zfs_no_scrub_prefetch
)
577 if (BP_IS_HOLE(bp
) || bp
->blk_birth
<= scn
->scn_phys
.scn_min_txg
||
578 (BP_GET_LEVEL(bp
) == 0 && BP_GET_TYPE(bp
) != DMU_OT_DNODE
))
581 SET_BOOKMARK(&czb
, objset
, object
, BP_GET_LEVEL(bp
), blkid
);
583 (void) arc_read(scn
->scn_zio_root
, scn
->scn_dp
->dp_spa
, bp
,
584 NULL
, NULL
, ZIO_PRIORITY_ASYNC_READ
,
585 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SCAN_THREAD
, &flags
, &czb
);
589 dsl_scan_check_resume(dsl_scan_t
*scn
, const dnode_phys_t
*dnp
,
590 const zbookmark_t
*zb
)
593 * We never skip over user/group accounting objects (obj<0)
595 if (!ZB_IS_ZERO(&scn
->scn_phys
.scn_bookmark
) &&
596 (int64_t)zb
->zb_object
>= 0) {
598 * If we already visited this bp & everything below (in
599 * a prior txg sync), don't bother doing it again.
601 if (zbookmark_is_before(dnp
, zb
, &scn
->scn_phys
.scn_bookmark
))
605 * If we found the block we're trying to resume from, or
606 * we went past it to a different object, zero it out to
607 * indicate that it's OK to start checking for pausing
610 if (bcmp(zb
, &scn
->scn_phys
.scn_bookmark
, sizeof (*zb
)) == 0 ||
611 zb
->zb_object
> scn
->scn_phys
.scn_bookmark
.zb_object
) {
612 dprintf("resuming at %llx/%llx/%llx/%llx\n",
613 (longlong_t
)zb
->zb_objset
,
614 (longlong_t
)zb
->zb_object
,
615 (longlong_t
)zb
->zb_level
,
616 (longlong_t
)zb
->zb_blkid
);
617 bzero(&scn
->scn_phys
.scn_bookmark
, sizeof (*zb
));
624 * Return nonzero on i/o error.
625 * Return new buf to write out in *bufp.
627 inline __attribute__((always_inline
)) static int
628 dsl_scan_recurse(dsl_scan_t
*scn
, dsl_dataset_t
*ds
, dmu_objset_type_t ostype
,
629 dnode_phys_t
*dnp
, const blkptr_t
*bp
,
630 const zbookmark_t
*zb
, dmu_tx_t
*tx
, arc_buf_t
**bufp
)
632 dsl_pool_t
*dp
= scn
->scn_dp
;
633 int zio_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_SCAN_THREAD
;
636 if (BP_GET_LEVEL(bp
) > 0) {
637 uint32_t flags
= ARC_WAIT
;
640 int epb
= BP_GET_LSIZE(bp
) >> SPA_BLKPTRSHIFT
;
642 err
= arc_read(NULL
, dp
->dp_spa
, bp
, arc_getbuf_func
, bufp
,
643 ZIO_PRIORITY_ASYNC_READ
, zio_flags
, &flags
, zb
);
645 scn
->scn_phys
.scn_errors
++;
648 for (i
= 0, cbp
= (*bufp
)->b_data
; i
< epb
; i
++, cbp
++) {
649 dsl_scan_prefetch(scn
, *bufp
, cbp
, zb
->zb_objset
,
650 zb
->zb_object
, zb
->zb_blkid
* epb
+ i
);
652 for (i
= 0, cbp
= (*bufp
)->b_data
; i
< epb
; i
++, cbp
++) {
655 SET_BOOKMARK(&czb
, zb
->zb_objset
, zb
->zb_object
,
657 zb
->zb_blkid
* epb
+ i
);
658 dsl_scan_visitbp(cbp
, &czb
, dnp
,
659 *bufp
, ds
, scn
, ostype
, tx
);
661 } else if (BP_GET_TYPE(bp
) == DMU_OT_USERGROUP_USED
) {
662 uint32_t flags
= ARC_WAIT
;
664 err
= arc_read(NULL
, dp
->dp_spa
, bp
, arc_getbuf_func
, bufp
,
665 ZIO_PRIORITY_ASYNC_READ
, zio_flags
, &flags
, zb
);
667 scn
->scn_phys
.scn_errors
++;
670 } else if (BP_GET_TYPE(bp
) == DMU_OT_DNODE
) {
671 uint32_t flags
= ARC_WAIT
;
674 int epb
= BP_GET_LSIZE(bp
) >> DNODE_SHIFT
;
676 err
= arc_read(NULL
, dp
->dp_spa
, bp
, arc_getbuf_func
, bufp
,
677 ZIO_PRIORITY_ASYNC_READ
, zio_flags
, &flags
, zb
);
679 scn
->scn_phys
.scn_errors
++;
682 for (i
= 0, cdnp
= (*bufp
)->b_data
; i
< epb
; i
++, cdnp
++) {
683 for (j
= 0; j
< cdnp
->dn_nblkptr
; j
++) {
684 blkptr_t
*cbp
= &cdnp
->dn_blkptr
[j
];
685 dsl_scan_prefetch(scn
, *bufp
, cbp
,
686 zb
->zb_objset
, zb
->zb_blkid
* epb
+ i
, j
);
689 for (i
= 0, cdnp
= (*bufp
)->b_data
; i
< epb
; i
++, cdnp
++) {
690 dsl_scan_visitdnode(scn
, ds
, ostype
,
691 cdnp
, *bufp
, zb
->zb_blkid
* epb
+ i
, tx
);
694 } else if (BP_GET_TYPE(bp
) == DMU_OT_OBJSET
) {
695 uint32_t flags
= ARC_WAIT
;
698 err
= arc_read(NULL
, dp
->dp_spa
, bp
, arc_getbuf_func
, bufp
,
699 ZIO_PRIORITY_ASYNC_READ
, zio_flags
, &flags
, zb
);
701 scn
->scn_phys
.scn_errors
++;
705 osp
= (*bufp
)->b_data
;
707 dsl_scan_visitdnode(scn
, ds
, osp
->os_type
,
708 &osp
->os_meta_dnode
, *bufp
, DMU_META_DNODE_OBJECT
, tx
);
710 if (OBJSET_BUF_HAS_USERUSED(*bufp
)) {
712 * We also always visit user/group accounting
713 * objects, and never skip them, even if we are
714 * pausing. This is necessary so that the space
715 * deltas from this txg get integrated.
717 dsl_scan_visitdnode(scn
, ds
, osp
->os_type
,
718 &osp
->os_groupused_dnode
, *bufp
,
719 DMU_GROUPUSED_OBJECT
, tx
);
720 dsl_scan_visitdnode(scn
, ds
, osp
->os_type
,
721 &osp
->os_userused_dnode
, *bufp
,
722 DMU_USERUSED_OBJECT
, tx
);
729 inline __attribute__((always_inline
)) static void
730 dsl_scan_visitdnode(dsl_scan_t
*scn
, dsl_dataset_t
*ds
,
731 dmu_objset_type_t ostype
, dnode_phys_t
*dnp
, arc_buf_t
*buf
,
732 uint64_t object
, dmu_tx_t
*tx
)
736 for (j
= 0; j
< dnp
->dn_nblkptr
; j
++) {
739 SET_BOOKMARK(&czb
, ds
? ds
->ds_object
: 0, object
,
740 dnp
->dn_nlevels
- 1, j
);
741 dsl_scan_visitbp(&dnp
->dn_blkptr
[j
],
742 &czb
, dnp
, buf
, ds
, scn
, ostype
, tx
);
745 if (dnp
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
) {
747 SET_BOOKMARK(&czb
, ds
? ds
->ds_object
: 0, object
,
749 dsl_scan_visitbp(&dnp
->dn_spill
,
750 &czb
, dnp
, buf
, ds
, scn
, ostype
, tx
);
755 * The arguments are in this order because mdb can only print the
756 * first 5; we want them to be useful.
759 dsl_scan_visitbp(blkptr_t
*bp
, const zbookmark_t
*zb
,
760 dnode_phys_t
*dnp
, arc_buf_t
*pbuf
,
761 dsl_dataset_t
*ds
, dsl_scan_t
*scn
, dmu_objset_type_t ostype
,
764 dsl_pool_t
*dp
= scn
->scn_dp
;
765 arc_buf_t
*buf
= NULL
;
768 bp_toread
= kmem_alloc(sizeof (blkptr_t
), KM_PUSHPAGE
);
771 /* ASSERT(pbuf == NULL || arc_released(pbuf)); */
773 if (dsl_scan_check_pause(scn
, zb
))
776 if (dsl_scan_check_resume(scn
, dnp
, zb
))
782 scn
->scn_visited_this_txg
++;
785 * This debugging is commented out to conserve stack space. This
786 * function is called recursively and the debugging addes several
787 * bytes to the stack for each call. It can be commented back in
788 * if required to debug an issue in dsl_scan_visitbp().
791 * "visiting ds=%p/%llu zb=%llx/%llx/%llx/%llx buf=%p bp=%p",
792 * ds, ds ? ds->ds_object : 0,
793 * zb->zb_objset, zb->zb_object, zb->zb_level, zb->zb_blkid,
797 if (bp
->blk_birth
<= scn
->scn_phys
.scn_cur_min_txg
)
800 if (dsl_scan_recurse(scn
, ds
, ostype
, dnp
, bp_toread
, zb
, tx
,
805 * If dsl_scan_ddt() has aready visited this block, it will have
806 * already done any translations or scrubbing, so don't call the
809 if (ddt_class_contains(dp
->dp_spa
,
810 scn
->scn_phys
.scn_ddt_class_max
, bp
)) {
816 * If this block is from the future (after cur_max_txg), then we
817 * are doing this on behalf of a deleted snapshot, and we will
818 * revisit the future block on the next pass of this dataset.
819 * Don't scan it now unless we need to because something
820 * under it was modified.
822 if (BP_PHYSICAL_BIRTH(bp
) <= scn
->scn_phys
.scn_cur_max_txg
) {
823 scan_funcs
[scn
->scn_phys
.scn_func
](dp
, bp
, zb
);
826 (void) arc_buf_remove_ref(buf
, &buf
);
828 kmem_free(bp_toread
, sizeof (blkptr_t
));
832 dsl_scan_visit_rootbp(dsl_scan_t
*scn
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
837 SET_BOOKMARK(&zb
, ds
? ds
->ds_object
: DMU_META_OBJSET
,
838 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
839 dsl_scan_visitbp(bp
, &zb
, NULL
, NULL
,
840 ds
, scn
, DMU_OST_NONE
, tx
);
842 dprintf_ds(ds
, "finished scan%s", "");
846 dsl_scan_ds_destroyed(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
848 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
849 dsl_scan_t
*scn
= dp
->dp_scan
;
852 if (scn
->scn_phys
.scn_state
!= DSS_SCANNING
)
855 if (scn
->scn_phys
.scn_bookmark
.zb_objset
== ds
->ds_object
) {
856 if (dsl_dataset_is_snapshot(ds
)) {
857 /* Note, scn_cur_{min,max}_txg stays the same. */
858 scn
->scn_phys
.scn_bookmark
.zb_objset
=
859 ds
->ds_phys
->ds_next_snap_obj
;
860 zfs_dbgmsg("destroying ds %llu; currently traversing; "
861 "reset zb_objset to %llu",
862 (u_longlong_t
)ds
->ds_object
,
863 (u_longlong_t
)ds
->ds_phys
->ds_next_snap_obj
);
864 scn
->scn_phys
.scn_flags
|= DSF_VISIT_DS_AGAIN
;
866 SET_BOOKMARK(&scn
->scn_phys
.scn_bookmark
,
867 ZB_DESTROYED_OBJSET
, 0, 0, 0);
868 zfs_dbgmsg("destroying ds %llu; currently traversing; "
869 "reset bookmark to -1,0,0,0",
870 (u_longlong_t
)ds
->ds_object
);
872 } else if (zap_lookup_int_key(dp
->dp_meta_objset
,
873 scn
->scn_phys
.scn_queue_obj
, ds
->ds_object
, &mintxg
) == 0) {
874 ASSERT3U(ds
->ds_phys
->ds_num_children
, <=, 1);
875 VERIFY3U(0, ==, zap_remove_int(dp
->dp_meta_objset
,
876 scn
->scn_phys
.scn_queue_obj
, ds
->ds_object
, tx
));
877 if (dsl_dataset_is_snapshot(ds
)) {
879 * We keep the same mintxg; it could be >
880 * ds_creation_txg if the previous snapshot was
883 VERIFY(zap_add_int_key(dp
->dp_meta_objset
,
884 scn
->scn_phys
.scn_queue_obj
,
885 ds
->ds_phys
->ds_next_snap_obj
, mintxg
, tx
) == 0);
886 zfs_dbgmsg("destroying ds %llu; in queue; "
887 "replacing with %llu",
888 (u_longlong_t
)ds
->ds_object
,
889 (u_longlong_t
)ds
->ds_phys
->ds_next_snap_obj
);
891 zfs_dbgmsg("destroying ds %llu; in queue; removing",
892 (u_longlong_t
)ds
->ds_object
);
895 zfs_dbgmsg("destroying ds %llu; ignoring",
896 (u_longlong_t
)ds
->ds_object
);
900 * dsl_scan_sync() should be called after this, and should sync
901 * out our changed state, but just to be safe, do it here.
903 dsl_scan_sync_state(scn
, tx
);
907 dsl_scan_ds_snapshotted(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
909 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
910 dsl_scan_t
*scn
= dp
->dp_scan
;
913 if (scn
->scn_phys
.scn_state
!= DSS_SCANNING
)
916 ASSERT(ds
->ds_phys
->ds_prev_snap_obj
!= 0);
918 if (scn
->scn_phys
.scn_bookmark
.zb_objset
== ds
->ds_object
) {
919 scn
->scn_phys
.scn_bookmark
.zb_objset
=
920 ds
->ds_phys
->ds_prev_snap_obj
;
921 zfs_dbgmsg("snapshotting ds %llu; currently traversing; "
922 "reset zb_objset to %llu",
923 (u_longlong_t
)ds
->ds_object
,
924 (u_longlong_t
)ds
->ds_phys
->ds_prev_snap_obj
);
925 } else if (zap_lookup_int_key(dp
->dp_meta_objset
,
926 scn
->scn_phys
.scn_queue_obj
, ds
->ds_object
, &mintxg
) == 0) {
927 VERIFY3U(0, ==, zap_remove_int(dp
->dp_meta_objset
,
928 scn
->scn_phys
.scn_queue_obj
, ds
->ds_object
, tx
));
929 VERIFY(zap_add_int_key(dp
->dp_meta_objset
,
930 scn
->scn_phys
.scn_queue_obj
,
931 ds
->ds_phys
->ds_prev_snap_obj
, mintxg
, tx
) == 0);
932 zfs_dbgmsg("snapshotting ds %llu; in queue; "
933 "replacing with %llu",
934 (u_longlong_t
)ds
->ds_object
,
935 (u_longlong_t
)ds
->ds_phys
->ds_prev_snap_obj
);
937 dsl_scan_sync_state(scn
, tx
);
941 dsl_scan_ds_clone_swapped(dsl_dataset_t
*ds1
, dsl_dataset_t
*ds2
, dmu_tx_t
*tx
)
943 dsl_pool_t
*dp
= ds1
->ds_dir
->dd_pool
;
944 dsl_scan_t
*scn
= dp
->dp_scan
;
947 if (scn
->scn_phys
.scn_state
!= DSS_SCANNING
)
950 if (scn
->scn_phys
.scn_bookmark
.zb_objset
== ds1
->ds_object
) {
951 scn
->scn_phys
.scn_bookmark
.zb_objset
= ds2
->ds_object
;
952 zfs_dbgmsg("clone_swap ds %llu; currently traversing; "
953 "reset zb_objset to %llu",
954 (u_longlong_t
)ds1
->ds_object
,
955 (u_longlong_t
)ds2
->ds_object
);
956 } else if (scn
->scn_phys
.scn_bookmark
.zb_objset
== ds2
->ds_object
) {
957 scn
->scn_phys
.scn_bookmark
.zb_objset
= ds1
->ds_object
;
958 zfs_dbgmsg("clone_swap ds %llu; currently traversing; "
959 "reset zb_objset to %llu",
960 (u_longlong_t
)ds2
->ds_object
,
961 (u_longlong_t
)ds1
->ds_object
);
964 if (zap_lookup_int_key(dp
->dp_meta_objset
, scn
->scn_phys
.scn_queue_obj
,
965 ds1
->ds_object
, &mintxg
) == 0) {
968 ASSERT3U(mintxg
, ==, ds1
->ds_phys
->ds_prev_snap_txg
);
969 ASSERT3U(mintxg
, ==, ds2
->ds_phys
->ds_prev_snap_txg
);
970 VERIFY3U(0, ==, zap_remove_int(dp
->dp_meta_objset
,
971 scn
->scn_phys
.scn_queue_obj
, ds1
->ds_object
, tx
));
972 err
= zap_add_int_key(dp
->dp_meta_objset
,
973 scn
->scn_phys
.scn_queue_obj
, ds2
->ds_object
, mintxg
, tx
);
974 VERIFY(err
== 0 || err
== EEXIST
);
976 /* Both were there to begin with */
977 VERIFY(0 == zap_add_int_key(dp
->dp_meta_objset
,
978 scn
->scn_phys
.scn_queue_obj
,
979 ds1
->ds_object
, mintxg
, tx
));
981 zfs_dbgmsg("clone_swap ds %llu; in queue; "
982 "replacing with %llu",
983 (u_longlong_t
)ds1
->ds_object
,
984 (u_longlong_t
)ds2
->ds_object
);
985 } else if (zap_lookup_int_key(dp
->dp_meta_objset
,
986 scn
->scn_phys
.scn_queue_obj
, ds2
->ds_object
, &mintxg
) == 0) {
987 ASSERT3U(mintxg
, ==, ds1
->ds_phys
->ds_prev_snap_txg
);
988 ASSERT3U(mintxg
, ==, ds2
->ds_phys
->ds_prev_snap_txg
);
989 VERIFY3U(0, ==, zap_remove_int(dp
->dp_meta_objset
,
990 scn
->scn_phys
.scn_queue_obj
, ds2
->ds_object
, tx
));
991 VERIFY(0 == zap_add_int_key(dp
->dp_meta_objset
,
992 scn
->scn_phys
.scn_queue_obj
, ds1
->ds_object
, mintxg
, tx
));
993 zfs_dbgmsg("clone_swap ds %llu; in queue; "
994 "replacing with %llu",
995 (u_longlong_t
)ds2
->ds_object
,
996 (u_longlong_t
)ds1
->ds_object
);
999 dsl_scan_sync_state(scn
, tx
);
1002 struct enqueue_clones_arg
{
1009 enqueue_clones_cb(dsl_pool_t
*dp
, dsl_dataset_t
*hds
, void *arg
)
1011 struct enqueue_clones_arg
*eca
= arg
;
1014 dsl_scan_t
*scn
= dp
->dp_scan
;
1016 if (hds
->ds_dir
->dd_phys
->dd_origin_obj
!= eca
->originobj
)
1019 err
= dsl_dataset_hold_obj(dp
, hds
->ds_object
, FTAG
, &ds
);
1023 while (ds
->ds_phys
->ds_prev_snap_obj
!= eca
->originobj
) {
1024 dsl_dataset_t
*prev
;
1025 err
= dsl_dataset_hold_obj(dp
,
1026 ds
->ds_phys
->ds_prev_snap_obj
, FTAG
, &prev
);
1028 dsl_dataset_rele(ds
, FTAG
);
1033 VERIFY(zap_add_int_key(dp
->dp_meta_objset
,
1034 scn
->scn_phys
.scn_queue_obj
, ds
->ds_object
,
1035 ds
->ds_phys
->ds_prev_snap_txg
, eca
->tx
) == 0);
1036 dsl_dataset_rele(ds
, FTAG
);
1041 dsl_scan_visitds(dsl_scan_t
*scn
, uint64_t dsobj
, dmu_tx_t
*tx
)
1043 dsl_pool_t
*dp
= scn
->scn_dp
;
1048 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp
, dsobj
, FTAG
, &ds
));
1050 if (dmu_objset_from_ds(ds
, &os
))
1054 * Only the ZIL in the head (non-snapshot) is valid. Even though
1055 * snapshots can have ZIL block pointers (which may be the same
1056 * BP as in the head), they must be ignored. So we traverse the
1057 * ZIL here, rather than in scan_recurse(), because the regular
1058 * snapshot block-sharing rules don't apply to it.
1060 if (DSL_SCAN_IS_SCRUB_RESILVER(scn
) && !dsl_dataset_is_snapshot(ds
))
1061 dsl_scan_zil(dp
, &os
->os_zil_header
);
1064 * Iterate over the bps in this ds.
1066 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1067 dsl_scan_visit_rootbp(scn
, ds
, &ds
->ds_phys
->ds_bp
, tx
);
1069 dsname
= kmem_alloc(ZFS_MAXNAMELEN
, KM_PUSHPAGE
);
1070 dsl_dataset_name(ds
, dsname
);
1071 zfs_dbgmsg("scanned dataset %llu (%s) with min=%llu max=%llu; "
1073 (longlong_t
)dsobj
, dsname
,
1074 (longlong_t
)scn
->scn_phys
.scn_cur_min_txg
,
1075 (longlong_t
)scn
->scn_phys
.scn_cur_max_txg
,
1076 (int)scn
->scn_pausing
);
1077 kmem_free(dsname
, ZFS_MAXNAMELEN
);
1079 if (scn
->scn_pausing
)
1083 * We've finished this pass over this dataset.
1087 * If we did not completely visit this dataset, do another pass.
1089 if (scn
->scn_phys
.scn_flags
& DSF_VISIT_DS_AGAIN
) {
1090 zfs_dbgmsg("incomplete pass; visiting again");
1091 scn
->scn_phys
.scn_flags
&= ~DSF_VISIT_DS_AGAIN
;
1092 VERIFY(zap_add_int_key(dp
->dp_meta_objset
,
1093 scn
->scn_phys
.scn_queue_obj
, ds
->ds_object
,
1094 scn
->scn_phys
.scn_cur_max_txg
, tx
) == 0);
1099 * Add descendent datasets to work queue.
1101 if (ds
->ds_phys
->ds_next_snap_obj
!= 0) {
1102 VERIFY(zap_add_int_key(dp
->dp_meta_objset
,
1103 scn
->scn_phys
.scn_queue_obj
, ds
->ds_phys
->ds_next_snap_obj
,
1104 ds
->ds_phys
->ds_creation_txg
, tx
) == 0);
1106 if (ds
->ds_phys
->ds_num_children
> 1) {
1107 boolean_t usenext
= B_FALSE
;
1108 if (ds
->ds_phys
->ds_next_clones_obj
!= 0) {
1111 * A bug in a previous version of the code could
1112 * cause upgrade_clones_cb() to not set
1113 * ds_next_snap_obj when it should, leading to a
1114 * missing entry. Therefore we can only use the
1115 * next_clones_obj when its count is correct.
1117 int err
= zap_count(dp
->dp_meta_objset
,
1118 ds
->ds_phys
->ds_next_clones_obj
, &count
);
1120 count
== ds
->ds_phys
->ds_num_children
- 1)
1125 VERIFY0(zap_join_key(dp
->dp_meta_objset
,
1126 ds
->ds_phys
->ds_next_clones_obj
,
1127 scn
->scn_phys
.scn_queue_obj
,
1128 ds
->ds_phys
->ds_creation_txg
, tx
));
1130 struct enqueue_clones_arg eca
;
1132 eca
.originobj
= ds
->ds_object
;
1134 VERIFY0(dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
1135 enqueue_clones_cb
, &eca
, DS_FIND_CHILDREN
));
1140 dsl_dataset_rele(ds
, FTAG
);
1145 enqueue_cb(dsl_pool_t
*dp
, dsl_dataset_t
*hds
, void *arg
)
1150 dsl_scan_t
*scn
= dp
->dp_scan
;
1152 err
= dsl_dataset_hold_obj(dp
, hds
->ds_object
, FTAG
, &ds
);
1156 while (ds
->ds_phys
->ds_prev_snap_obj
!= 0) {
1157 dsl_dataset_t
*prev
;
1158 err
= dsl_dataset_hold_obj(dp
, ds
->ds_phys
->ds_prev_snap_obj
,
1161 dsl_dataset_rele(ds
, FTAG
);
1166 * If this is a clone, we don't need to worry about it for now.
1168 if (prev
->ds_phys
->ds_next_snap_obj
!= ds
->ds_object
) {
1169 dsl_dataset_rele(ds
, FTAG
);
1170 dsl_dataset_rele(prev
, FTAG
);
1173 dsl_dataset_rele(ds
, FTAG
);
1177 VERIFY(zap_add_int_key(dp
->dp_meta_objset
, scn
->scn_phys
.scn_queue_obj
,
1178 ds
->ds_object
, ds
->ds_phys
->ds_prev_snap_txg
, tx
) == 0);
1179 dsl_dataset_rele(ds
, FTAG
);
1184 * Scrub/dedup interaction.
1186 * If there are N references to a deduped block, we don't want to scrub it
1187 * N times -- ideally, we should scrub it exactly once.
1189 * We leverage the fact that the dde's replication class (enum ddt_class)
1190 * is ordered from highest replication class (DDT_CLASS_DITTO) to lowest
1191 * (DDT_CLASS_UNIQUE) so that we may walk the DDT in that order.
1193 * To prevent excess scrubbing, the scrub begins by walking the DDT
1194 * to find all blocks with refcnt > 1, and scrubs each of these once.
1195 * Since there are two replication classes which contain blocks with
1196 * refcnt > 1, we scrub the highest replication class (DDT_CLASS_DITTO) first.
1197 * Finally the top-down scrub begins, only visiting blocks with refcnt == 1.
1199 * There would be nothing more to say if a block's refcnt couldn't change
1200 * during a scrub, but of course it can so we must account for changes
1201 * in a block's replication class.
1203 * Here's an example of what can occur:
1205 * If a block has refcnt > 1 during the DDT scrub phase, but has refcnt == 1
1206 * when visited during the top-down scrub phase, it will be scrubbed twice.
1207 * This negates our scrub optimization, but is otherwise harmless.
1209 * If a block has refcnt == 1 during the DDT scrub phase, but has refcnt > 1
1210 * on each visit during the top-down scrub phase, it will never be scrubbed.
1211 * To catch this, ddt_sync_entry() notifies the scrub code whenever a block's
1212 * reference class transitions to a higher level (i.e DDT_CLASS_UNIQUE to
1213 * DDT_CLASS_DUPLICATE); if it transitions from refcnt == 1 to refcnt > 1
1214 * while a scrub is in progress, it scrubs the block right then.
1217 dsl_scan_ddt(dsl_scan_t
*scn
, dmu_tx_t
*tx
)
1219 ddt_bookmark_t
*ddb
= &scn
->scn_phys
.scn_ddt_bookmark
;
1224 bzero(&dde
, sizeof (ddt_entry_t
));
1226 while ((error
= ddt_walk(scn
->scn_dp
->dp_spa
, ddb
, &dde
)) == 0) {
1229 if (ddb
->ddb_class
> scn
->scn_phys
.scn_ddt_class_max
)
1231 dprintf("visiting ddb=%llu/%llu/%llu/%llx\n",
1232 (longlong_t
)ddb
->ddb_class
,
1233 (longlong_t
)ddb
->ddb_type
,
1234 (longlong_t
)ddb
->ddb_checksum
,
1235 (longlong_t
)ddb
->ddb_cursor
);
1237 /* There should be no pending changes to the dedup table */
1238 ddt
= scn
->scn_dp
->dp_spa
->spa_ddt
[ddb
->ddb_checksum
];
1239 ASSERT(avl_first(&ddt
->ddt_tree
) == NULL
);
1241 dsl_scan_ddt_entry(scn
, ddb
->ddb_checksum
, &dde
, tx
);
1244 if (dsl_scan_check_pause(scn
, NULL
))
1248 zfs_dbgmsg("scanned %llu ddt entries with class_max = %u; pausing=%u",
1249 (longlong_t
)n
, (int)scn
->scn_phys
.scn_ddt_class_max
,
1250 (int)scn
->scn_pausing
);
1252 ASSERT(error
== 0 || error
== ENOENT
);
1253 ASSERT(error
!= ENOENT
||
1254 ddb
->ddb_class
> scn
->scn_phys
.scn_ddt_class_max
);
1259 dsl_scan_ddt_entry(dsl_scan_t
*scn
, enum zio_checksum checksum
,
1260 ddt_entry_t
*dde
, dmu_tx_t
*tx
)
1262 const ddt_key_t
*ddk
= &dde
->dde_key
;
1263 ddt_phys_t
*ddp
= dde
->dde_phys
;
1265 zbookmark_t zb
= { 0 };
1268 if (scn
->scn_phys
.scn_state
!= DSS_SCANNING
)
1271 for (p
= 0; p
< DDT_PHYS_TYPES
; p
++, ddp
++) {
1272 if (ddp
->ddp_phys_birth
== 0 ||
1273 ddp
->ddp_phys_birth
> scn
->scn_phys
.scn_max_txg
)
1275 ddt_bp_create(checksum
, ddk
, ddp
, &bp
);
1277 scn
->scn_visited_this_txg
++;
1278 scan_funcs
[scn
->scn_phys
.scn_func
](scn
->scn_dp
, &bp
, &zb
);
1283 dsl_scan_visit(dsl_scan_t
*scn
, dmu_tx_t
*tx
)
1285 dsl_pool_t
*dp
= scn
->scn_dp
;
1287 zap_attribute_t
*za
;
1289 if (scn
->scn_phys
.scn_ddt_bookmark
.ddb_class
<=
1290 scn
->scn_phys
.scn_ddt_class_max
) {
1291 scn
->scn_phys
.scn_cur_min_txg
= scn
->scn_phys
.scn_min_txg
;
1292 scn
->scn_phys
.scn_cur_max_txg
= scn
->scn_phys
.scn_max_txg
;
1293 dsl_scan_ddt(scn
, tx
);
1294 if (scn
->scn_pausing
)
1298 if (scn
->scn_phys
.scn_bookmark
.zb_objset
== DMU_META_OBJSET
) {
1299 /* First do the MOS & ORIGIN */
1301 scn
->scn_phys
.scn_cur_min_txg
= scn
->scn_phys
.scn_min_txg
;
1302 scn
->scn_phys
.scn_cur_max_txg
= scn
->scn_phys
.scn_max_txg
;
1303 dsl_scan_visit_rootbp(scn
, NULL
,
1304 &dp
->dp_meta_rootbp
, tx
);
1305 spa_set_rootblkptr(dp
->dp_spa
, &dp
->dp_meta_rootbp
);
1306 if (scn
->scn_pausing
)
1309 if (spa_version(dp
->dp_spa
) < SPA_VERSION_DSL_SCRUB
) {
1310 VERIFY0(dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
1311 enqueue_cb
, tx
, DS_FIND_CHILDREN
));
1313 dsl_scan_visitds(scn
,
1314 dp
->dp_origin_snap
->ds_object
, tx
);
1316 ASSERT(!scn
->scn_pausing
);
1317 } else if (scn
->scn_phys
.scn_bookmark
.zb_objset
!=
1318 ZB_DESTROYED_OBJSET
) {
1320 * If we were paused, continue from here. Note if the
1321 * ds we were paused on was deleted, the zb_objset may
1322 * be -1, so we will skip this and find a new objset
1325 dsl_scan_visitds(scn
, scn
->scn_phys
.scn_bookmark
.zb_objset
, tx
);
1326 if (scn
->scn_pausing
)
1331 * In case we were paused right at the end of the ds, zero the
1332 * bookmark so we don't think that we're still trying to resume.
1334 bzero(&scn
->scn_phys
.scn_bookmark
, sizeof (zbookmark_t
));
1335 zc
= kmem_alloc(sizeof (zap_cursor_t
), KM_PUSHPAGE
);
1336 za
= kmem_alloc(sizeof (zap_attribute_t
), KM_PUSHPAGE
);
1338 /* keep pulling things out of the zap-object-as-queue */
1339 while (zap_cursor_init(zc
, dp
->dp_meta_objset
,
1340 scn
->scn_phys
.scn_queue_obj
),
1341 zap_cursor_retrieve(zc
, za
) == 0) {
1345 dsobj
= strtonum(za
->za_name
, NULL
);
1346 VERIFY3U(0, ==, zap_remove_int(dp
->dp_meta_objset
,
1347 scn
->scn_phys
.scn_queue_obj
, dsobj
, tx
));
1349 /* Set up min/max txg */
1350 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp
, dsobj
, FTAG
, &ds
));
1351 if (za
->za_first_integer
!= 0) {
1352 scn
->scn_phys
.scn_cur_min_txg
=
1353 MAX(scn
->scn_phys
.scn_min_txg
,
1354 za
->za_first_integer
);
1356 scn
->scn_phys
.scn_cur_min_txg
=
1357 MAX(scn
->scn_phys
.scn_min_txg
,
1358 ds
->ds_phys
->ds_prev_snap_txg
);
1360 scn
->scn_phys
.scn_cur_max_txg
= dsl_scan_ds_maxtxg(ds
);
1361 dsl_dataset_rele(ds
, FTAG
);
1363 dsl_scan_visitds(scn
, dsobj
, tx
);
1364 zap_cursor_fini(zc
);
1365 if (scn
->scn_pausing
)
1368 zap_cursor_fini(zc
);
1370 kmem_free(za
, sizeof (zap_attribute_t
));
1371 kmem_free(zc
, sizeof (zap_cursor_t
));
1375 dsl_scan_free_should_pause(dsl_scan_t
*scn
)
1377 uint64_t elapsed_nanosecs
;
1382 elapsed_nanosecs
= gethrtime() - scn
->scn_sync_start_time
;
1383 return (elapsed_nanosecs
/ NANOSEC
> zfs_txg_timeout
||
1384 (NSEC2MSEC(elapsed_nanosecs
) > zfs_free_min_time_ms
&&
1385 txg_sync_waiting(scn
->scn_dp
)) ||
1386 spa_shutting_down(scn
->scn_dp
->dp_spa
));
1390 dsl_scan_free_block_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
1392 dsl_scan_t
*scn
= arg
;
1394 if (!scn
->scn_is_bptree
||
1395 (BP_GET_LEVEL(bp
) == 0 && BP_GET_TYPE(bp
) != DMU_OT_OBJSET
)) {
1396 if (dsl_scan_free_should_pause(scn
))
1397 return (SET_ERROR(ERESTART
));
1400 zio_nowait(zio_free_sync(scn
->scn_zio_root
, scn
->scn_dp
->dp_spa
,
1401 dmu_tx_get_txg(tx
), bp
, 0));
1402 dsl_dir_diduse_space(tx
->tx_pool
->dp_free_dir
, DD_USED_HEAD
,
1403 -bp_get_dsize_sync(scn
->scn_dp
->dp_spa
, bp
),
1404 -BP_GET_PSIZE(bp
), -BP_GET_UCSIZE(bp
), tx
);
1405 scn
->scn_visited_this_txg
++;
1410 dsl_scan_active(dsl_scan_t
*scn
)
1412 spa_t
*spa
= scn
->scn_dp
->dp_spa
;
1413 uint64_t used
= 0, comp
, uncomp
;
1415 if (spa
->spa_load_state
!= SPA_LOAD_NONE
)
1417 if (spa_shutting_down(spa
))
1419 if (scn
->scn_phys
.scn_state
== DSS_SCANNING
||
1420 scn
->scn_async_destroying
)
1423 if (spa_version(scn
->scn_dp
->dp_spa
) >= SPA_VERSION_DEADLISTS
) {
1424 (void) bpobj_space(&scn
->scn_dp
->dp_free_bpobj
,
1425 &used
, &comp
, &uncomp
);
1431 dsl_scan_sync(dsl_pool_t
*dp
, dmu_tx_t
*tx
)
1433 dsl_scan_t
*scn
= dp
->dp_scan
;
1434 spa_t
*spa
= dp
->dp_spa
;
1438 * Check for scn_restart_txg before checking spa_load_state, so
1439 * that we can restart an old-style scan while the pool is being
1440 * imported (see dsl_scan_init).
1442 if (scn
->scn_restart_txg
!= 0 &&
1443 scn
->scn_restart_txg
<= tx
->tx_txg
) {
1444 pool_scan_func_t func
= POOL_SCAN_SCRUB
;
1445 dsl_scan_done(scn
, B_FALSE
, tx
);
1446 if (vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
))
1447 func
= POOL_SCAN_RESILVER
;
1448 zfs_dbgmsg("restarting scan func=%u txg=%llu",
1450 dsl_scan_setup_sync(&func
, tx
);
1453 if (!dsl_scan_active(scn
) ||
1454 spa_sync_pass(dp
->dp_spa
) > 1)
1457 scn
->scn_visited_this_txg
= 0;
1458 scn
->scn_pausing
= B_FALSE
;
1459 scn
->scn_sync_start_time
= gethrtime();
1460 spa
->spa_scrub_active
= B_TRUE
;
1463 * First process the free list. If we pause the free, don't do
1464 * any scanning. This ensures that there is no free list when
1465 * we are scanning, so the scan code doesn't have to worry about
1468 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DEADLISTS
) {
1469 scn
->scn_is_bptree
= B_FALSE
;
1470 scn
->scn_zio_root
= zio_root(dp
->dp_spa
, NULL
,
1471 NULL
, ZIO_FLAG_MUSTSUCCEED
);
1472 err
= bpobj_iterate(&dp
->dp_free_bpobj
,
1473 dsl_scan_free_block_cb
, scn
, tx
);
1474 VERIFY3U(0, ==, zio_wait(scn
->scn_zio_root
));
1476 if (err
== 0 && spa_feature_is_active(spa
,
1477 SPA_FEATURE_ASYNC_DESTROY
)) {
1478 ASSERT(scn
->scn_async_destroying
);
1479 scn
->scn_is_bptree
= B_TRUE
;
1480 scn
->scn_zio_root
= zio_root(dp
->dp_spa
, NULL
,
1481 NULL
, ZIO_FLAG_MUSTSUCCEED
);
1482 err
= bptree_iterate(dp
->dp_meta_objset
,
1483 dp
->dp_bptree_obj
, B_TRUE
, dsl_scan_free_block_cb
,
1485 VERIFY0(zio_wait(scn
->scn_zio_root
));
1488 /* finished; deactivate async destroy feature */
1489 spa_feature_decr(spa
, SPA_FEATURE_ASYNC_DESTROY
,
1491 ASSERT(!spa_feature_is_active(spa
,
1492 SPA_FEATURE_ASYNC_DESTROY
));
1493 VERIFY0(zap_remove(dp
->dp_meta_objset
,
1494 DMU_POOL_DIRECTORY_OBJECT
,
1495 DMU_POOL_BPTREE_OBJ
, tx
));
1496 VERIFY0(bptree_free(dp
->dp_meta_objset
,
1497 dp
->dp_bptree_obj
, tx
));
1498 dp
->dp_bptree_obj
= 0;
1499 scn
->scn_async_destroying
= B_FALSE
;
1502 if (scn
->scn_visited_this_txg
) {
1503 zfs_dbgmsg("freed %llu blocks in %llums from "
1504 "free_bpobj/bptree txg %llu",
1505 (longlong_t
)scn
->scn_visited_this_txg
,
1507 NSEC2MSEC(gethrtime() - scn
->scn_sync_start_time
),
1508 (longlong_t
)tx
->tx_txg
);
1509 scn
->scn_visited_this_txg
= 0;
1511 * Re-sync the ddt so that we can further modify
1512 * it when doing bprewrite.
1514 ddt_sync(spa
, tx
->tx_txg
);
1516 if (err
== ERESTART
)
1520 if (scn
->scn_phys
.scn_state
!= DSS_SCANNING
)
1523 if (scn
->scn_done_txg
== tx
->tx_txg
) {
1524 ASSERT(!scn
->scn_pausing
);
1525 /* finished with scan. */
1526 zfs_dbgmsg("txg %llu scan complete", tx
->tx_txg
);
1527 dsl_scan_done(scn
, B_TRUE
, tx
);
1528 ASSERT3U(spa
->spa_scrub_inflight
, ==, 0);
1529 dsl_scan_sync_state(scn
, tx
);
1533 if (scn
->scn_phys
.scn_ddt_bookmark
.ddb_class
<=
1534 scn
->scn_phys
.scn_ddt_class_max
) {
1535 zfs_dbgmsg("doing scan sync txg %llu; "
1536 "ddt bm=%llu/%llu/%llu/%llx",
1537 (longlong_t
)tx
->tx_txg
,
1538 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_class
,
1539 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_type
,
1540 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_checksum
,
1541 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_cursor
);
1542 ASSERT(scn
->scn_phys
.scn_bookmark
.zb_objset
== 0);
1543 ASSERT(scn
->scn_phys
.scn_bookmark
.zb_object
== 0);
1544 ASSERT(scn
->scn_phys
.scn_bookmark
.zb_level
== 0);
1545 ASSERT(scn
->scn_phys
.scn_bookmark
.zb_blkid
== 0);
1547 zfs_dbgmsg("doing scan sync txg %llu; bm=%llu/%llu/%llu/%llu",
1548 (longlong_t
)tx
->tx_txg
,
1549 (longlong_t
)scn
->scn_phys
.scn_bookmark
.zb_objset
,
1550 (longlong_t
)scn
->scn_phys
.scn_bookmark
.zb_object
,
1551 (longlong_t
)scn
->scn_phys
.scn_bookmark
.zb_level
,
1552 (longlong_t
)scn
->scn_phys
.scn_bookmark
.zb_blkid
);
1555 scn
->scn_zio_root
= zio_root(dp
->dp_spa
, NULL
,
1556 NULL
, ZIO_FLAG_CANFAIL
);
1557 dsl_pool_config_enter(dp
, FTAG
);
1558 dsl_scan_visit(scn
, tx
);
1559 dsl_pool_config_exit(dp
, FTAG
);
1560 (void) zio_wait(scn
->scn_zio_root
);
1561 scn
->scn_zio_root
= NULL
;
1563 zfs_dbgmsg("visited %llu blocks in %llums",
1564 (longlong_t
)scn
->scn_visited_this_txg
,
1565 (longlong_t
)NSEC2MSEC(gethrtime() - scn
->scn_sync_start_time
));
1567 if (!scn
->scn_pausing
) {
1568 scn
->scn_done_txg
= tx
->tx_txg
+ 1;
1569 zfs_dbgmsg("txg %llu traversal complete, waiting till txg %llu",
1570 tx
->tx_txg
, scn
->scn_done_txg
);
1573 if (DSL_SCAN_IS_SCRUB_RESILVER(scn
)) {
1574 mutex_enter(&spa
->spa_scrub_lock
);
1575 while (spa
->spa_scrub_inflight
> 0) {
1576 cv_wait(&spa
->spa_scrub_io_cv
,
1577 &spa
->spa_scrub_lock
);
1579 mutex_exit(&spa
->spa_scrub_lock
);
1582 dsl_scan_sync_state(scn
, tx
);
1586 * This will start a new scan, or restart an existing one.
1589 dsl_resilver_restart(dsl_pool_t
*dp
, uint64_t txg
)
1593 tx
= dmu_tx_create_dd(dp
->dp_mos_dir
);
1594 VERIFY(0 == dmu_tx_assign(tx
, TXG_WAIT
));
1596 txg
= dmu_tx_get_txg(tx
);
1597 dp
->dp_scan
->scn_restart_txg
= txg
;
1600 dp
->dp_scan
->scn_restart_txg
= txg
;
1602 zfs_dbgmsg("restarting resilver txg=%llu", txg
);
1606 dsl_scan_resilvering(dsl_pool_t
*dp
)
1608 return (dp
->dp_scan
->scn_phys
.scn_state
== DSS_SCANNING
&&
1609 dp
->dp_scan
->scn_phys
.scn_func
== POOL_SCAN_RESILVER
);
1617 count_block(zfs_all_blkstats_t
*zab
, const blkptr_t
*bp
)
1622 * If we resume after a reboot, zab will be NULL; don't record
1623 * incomplete stats in that case.
1628 for (i
= 0; i
< 4; i
++) {
1629 int l
= (i
< 2) ? BP_GET_LEVEL(bp
) : DN_MAX_LEVELS
;
1630 int t
= (i
& 1) ? BP_GET_TYPE(bp
) : DMU_OT_TOTAL
;
1634 if (t
& DMU_OT_NEWTYPE
)
1637 zb
= &zab
->zab_type
[l
][t
];
1639 zb
->zb_asize
+= BP_GET_ASIZE(bp
);
1640 zb
->zb_lsize
+= BP_GET_LSIZE(bp
);
1641 zb
->zb_psize
+= BP_GET_PSIZE(bp
);
1642 zb
->zb_gangs
+= BP_COUNT_GANG(bp
);
1644 switch (BP_GET_NDVAS(bp
)) {
1646 if (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
1647 DVA_GET_VDEV(&bp
->blk_dva
[1]))
1648 zb
->zb_ditto_2_of_2_samevdev
++;
1651 equal
= (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
1652 DVA_GET_VDEV(&bp
->blk_dva
[1])) +
1653 (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
1654 DVA_GET_VDEV(&bp
->blk_dva
[2])) +
1655 (DVA_GET_VDEV(&bp
->blk_dva
[1]) ==
1656 DVA_GET_VDEV(&bp
->blk_dva
[2]));
1658 zb
->zb_ditto_2_of_3_samevdev
++;
1659 else if (equal
== 3)
1660 zb
->zb_ditto_3_of_3_samevdev
++;
1667 dsl_scan_scrub_done(zio_t
*zio
)
1669 spa_t
*spa
= zio
->io_spa
;
1671 zio_data_buf_free(zio
->io_data
, zio
->io_size
);
1673 mutex_enter(&spa
->spa_scrub_lock
);
1674 spa
->spa_scrub_inflight
--;
1675 cv_broadcast(&spa
->spa_scrub_io_cv
);
1677 if (zio
->io_error
&& (zio
->io_error
!= ECKSUM
||
1678 !(zio
->io_flags
& ZIO_FLAG_SPECULATIVE
))) {
1679 spa
->spa_dsl_pool
->dp_scan
->scn_phys
.scn_errors
++;
1681 mutex_exit(&spa
->spa_scrub_lock
);
1685 dsl_scan_scrub_cb(dsl_pool_t
*dp
,
1686 const blkptr_t
*bp
, const zbookmark_t
*zb
)
1688 dsl_scan_t
*scn
= dp
->dp_scan
;
1689 size_t size
= BP_GET_PSIZE(bp
);
1690 spa_t
*spa
= dp
->dp_spa
;
1691 uint64_t phys_birth
= BP_PHYSICAL_BIRTH(bp
);
1692 boolean_t needs_io
= B_FALSE
;
1693 int zio_flags
= ZIO_FLAG_SCAN_THREAD
| ZIO_FLAG_RAW
| ZIO_FLAG_CANFAIL
;
1697 if (phys_birth
<= scn
->scn_phys
.scn_min_txg
||
1698 phys_birth
>= scn
->scn_phys
.scn_max_txg
)
1701 count_block(dp
->dp_blkstats
, bp
);
1703 ASSERT(DSL_SCAN_IS_SCRUB_RESILVER(scn
));
1704 if (scn
->scn_phys
.scn_func
== POOL_SCAN_SCRUB
) {
1705 zio_flags
|= ZIO_FLAG_SCRUB
;
1707 scan_delay
= zfs_scrub_delay
;
1709 ASSERT3U(scn
->scn_phys
.scn_func
, ==, POOL_SCAN_RESILVER
);
1710 zio_flags
|= ZIO_FLAG_RESILVER
;
1712 scan_delay
= zfs_resilver_delay
;
1715 /* If it's an intent log block, failure is expected. */
1716 if (zb
->zb_level
== ZB_ZIL_LEVEL
)
1717 zio_flags
|= ZIO_FLAG_SPECULATIVE
;
1719 for (d
= 0; d
< BP_GET_NDVAS(bp
); d
++) {
1720 vdev_t
*vd
= vdev_lookup_top(spa
,
1721 DVA_GET_VDEV(&bp
->blk_dva
[d
]));
1724 * Keep track of how much data we've examined so that
1725 * zpool(1M) status can make useful progress reports.
1727 scn
->scn_phys
.scn_examined
+= DVA_GET_ASIZE(&bp
->blk_dva
[d
]);
1728 spa
->spa_scan_pass_exam
+= DVA_GET_ASIZE(&bp
->blk_dva
[d
]);
1730 /* if it's a resilver, this may not be in the target range */
1732 if (DVA_GET_GANG(&bp
->blk_dva
[d
])) {
1734 * Gang members may be spread across multiple
1735 * vdevs, so the best estimate we have is the
1736 * scrub range, which has already been checked.
1737 * XXX -- it would be better to change our
1738 * allocation policy to ensure that all
1739 * gang members reside on the same vdev.
1743 needs_io
= vdev_dtl_contains(vd
, DTL_PARTIAL
,
1749 if (needs_io
&& !zfs_no_scrub_io
) {
1750 vdev_t
*rvd
= spa
->spa_root_vdev
;
1751 uint64_t maxinflight
= rvd
->vdev_children
* zfs_top_maxinflight
;
1752 void *data
= zio_data_buf_alloc(size
);
1754 mutex_enter(&spa
->spa_scrub_lock
);
1755 while (spa
->spa_scrub_inflight
>= maxinflight
)
1756 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
1757 spa
->spa_scrub_inflight
++;
1758 mutex_exit(&spa
->spa_scrub_lock
);
1761 * If we're seeing recent (zfs_scan_idle) "important" I/Os
1762 * then throttle our workload to limit the impact of a scan.
1764 if (ddi_get_lbolt64() - spa
->spa_last_io
<= zfs_scan_idle
)
1767 zio_nowait(zio_read(NULL
, spa
, bp
, data
, size
,
1768 dsl_scan_scrub_done
, NULL
, ZIO_PRIORITY_SCRUB
,
1772 /* do not relocate this block */
1777 dsl_scan(dsl_pool_t
*dp
, pool_scan_func_t func
)
1779 spa_t
*spa
= dp
->dp_spa
;
1782 * Purge all vdev caches and probe all devices. We do this here
1783 * rather than in sync context because this requires a writer lock
1784 * on the spa_config lock, which we can't do from sync context. The
1785 * spa_scrub_reopen flag indicates that vdev_open() should not
1786 * attempt to start another scrub.
1788 spa_vdev_state_enter(spa
, SCL_NONE
);
1789 spa
->spa_scrub_reopen
= B_TRUE
;
1790 vdev_reopen(spa
->spa_root_vdev
);
1791 spa
->spa_scrub_reopen
= B_FALSE
;
1792 (void) spa_vdev_state_exit(spa
, NULL
, 0);
1794 return (dsl_sync_task(spa_name(spa
), dsl_scan_setup_check
,
1795 dsl_scan_setup_sync
, &func
, 0));
1798 #if defined(_KERNEL) && defined(HAVE_SPL)
1799 module_param(zfs_top_maxinflight
, int, 0644);
1800 MODULE_PARM_DESC(zfs_top_maxinflight
, "Max I/Os per top-level");
1802 module_param(zfs_resilver_delay
, int, 0644);
1803 MODULE_PARM_DESC(zfs_resilver_delay
, "Number of ticks to delay resilver");
1805 module_param(zfs_scrub_delay
, int, 0644);
1806 MODULE_PARM_DESC(zfs_scrub_delay
, "Number of ticks to delay scrub");
1808 module_param(zfs_scan_idle
, int, 0644);
1809 MODULE_PARM_DESC(zfs_scan_idle
, "Idle window in clock ticks");
1811 module_param(zfs_scan_min_time_ms
, int, 0644);
1812 MODULE_PARM_DESC(zfs_scan_min_time_ms
, "Min millisecs to scrub per txg");
1814 module_param(zfs_free_min_time_ms
, int, 0644);
1815 MODULE_PARM_DESC(zfs_free_min_time_ms
, "Min millisecs to free per txg");
1817 module_param(zfs_resilver_min_time_ms
, int, 0644);
1818 MODULE_PARM_DESC(zfs_resilver_min_time_ms
, "Min millisecs to resilver per txg");
1820 module_param(zfs_no_scrub_io
, int, 0644);
1821 MODULE_PARM_DESC(zfs_no_scrub_io
, "Set to disable scrub I/O");
1823 module_param(zfs_no_scrub_prefetch
, int, 0644);
1824 MODULE_PARM_DESC(zfs_no_scrub_prefetch
, "Set to disable scrub prefetching");