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, 2016 by Delphix. All rights reserved.
24 * Copyright 2016 Gary Mills
27 #include <sys/dsl_scan.h>
28 #include <sys/dsl_pool.h>
29 #include <sys/dsl_dataset.h>
30 #include <sys/dsl_prop.h>
31 #include <sys/dsl_dir.h>
32 #include <sys/dsl_synctask.h>
33 #include <sys/dnode.h>
34 #include <sys/dmu_tx.h>
35 #include <sys/dmu_objset.h>
39 #include <sys/zfs_context.h>
40 #include <sys/fs/zfs.h>
41 #include <sys/zfs_znode.h>
42 #include <sys/spa_impl.h>
43 #include <sys/vdev_impl.h>
44 #include <sys/zil_impl.h>
45 #include <sys/zio_checksum.h>
48 #include <sys/sa_impl.h>
49 #include <sys/zfeature.h>
52 #include <sys/zfs_vfsops.h>
55 typedef int (scan_cb_t
)(dsl_pool_t
*, const blkptr_t
*,
56 const zbookmark_phys_t
*);
58 static scan_cb_t dsl_scan_scrub_cb
;
59 static void dsl_scan_cancel_sync(void *, dmu_tx_t
*);
60 static void dsl_scan_sync_state(dsl_scan_t
*, dmu_tx_t
*);
61 static boolean_t
dsl_scan_restarting(dsl_scan_t
*, dmu_tx_t
*);
63 int zfs_top_maxinflight
= 32; /* maximum I/Os per top-level */
64 int zfs_resilver_delay
= 2; /* number of ticks to delay resilver */
65 int zfs_scrub_delay
= 4; /* number of ticks to delay scrub */
66 int zfs_scan_idle
= 50; /* idle window in clock ticks */
68 int zfs_scan_min_time_ms
= 1000; /* min millisecs to scrub per txg */
69 int zfs_free_min_time_ms
= 1000; /* min millisecs to free per txg */
70 int zfs_resilver_min_time_ms
= 3000; /* min millisecs to resilver per txg */
71 int zfs_no_scrub_io
= B_FALSE
; /* set to disable scrub i/o */
72 int zfs_no_scrub_prefetch
= B_FALSE
; /* set to disable scrub prefetch */
73 enum ddt_class zfs_scrub_ddt_class_max
= DDT_CLASS_DUPLICATE
;
74 int dsl_scan_delay_completion
= B_FALSE
; /* set to delay scan completion */
75 /* max number of blocks to free in a single TXG */
76 ulong zfs_free_max_blocks
= 100000;
78 #define DSL_SCAN_IS_SCRUB_RESILVER(scn) \
79 ((scn)->scn_phys.scn_func == POOL_SCAN_SCRUB || \
80 (scn)->scn_phys.scn_func == POOL_SCAN_RESILVER)
83 * Enable/disable the processing of the free_bpobj object.
85 int zfs_free_bpobj_enabled
= 1;
87 /* the order has to match pool_scan_type */
88 static scan_cb_t
*scan_funcs
[POOL_SCAN_FUNCS
] = {
90 dsl_scan_scrub_cb
, /* POOL_SCAN_SCRUB */
91 dsl_scan_scrub_cb
, /* POOL_SCAN_RESILVER */
95 dsl_scan_init(dsl_pool_t
*dp
, uint64_t txg
)
99 spa_t
*spa
= dp
->dp_spa
;
102 scn
= dp
->dp_scan
= kmem_zalloc(sizeof (dsl_scan_t
), KM_SLEEP
);
106 * It's possible that we're resuming a scan after a reboot so
107 * make sure that the scan_async_destroying flag is initialized
110 ASSERT(!scn
->scn_async_destroying
);
111 scn
->scn_async_destroying
= spa_feature_is_active(dp
->dp_spa
,
112 SPA_FEATURE_ASYNC_DESTROY
);
114 err
= zap_lookup(dp
->dp_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
115 "scrub_func", sizeof (uint64_t), 1, &f
);
118 * There was an old-style scrub in progress. Restart a
119 * new-style scrub from the beginning.
121 scn
->scn_restart_txg
= txg
;
122 zfs_dbgmsg("old-style scrub was in progress; "
123 "restarting new-style scrub in txg %llu",
124 scn
->scn_restart_txg
);
127 * Load the queue obj from the old location so that it
128 * can be freed by dsl_scan_done().
130 (void) zap_lookup(dp
->dp_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
131 "scrub_queue", sizeof (uint64_t), 1,
132 &scn
->scn_phys
.scn_queue_obj
);
134 err
= zap_lookup(dp
->dp_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
135 DMU_POOL_SCAN
, sizeof (uint64_t), SCAN_PHYS_NUMINTS
,
138 * Detect if the pool contains the signature of #2094. If it
139 * does properly update the scn->scn_phys structure and notify
140 * the administrator by setting an errata for the pool.
142 if (err
== EOVERFLOW
) {
143 uint64_t zaptmp
[SCAN_PHYS_NUMINTS
+ 1];
144 VERIFY3S(SCAN_PHYS_NUMINTS
, ==, 24);
145 VERIFY3S(offsetof(dsl_scan_phys_t
, scn_flags
), ==,
146 (23 * sizeof (uint64_t)));
148 err
= zap_lookup(dp
->dp_meta_objset
,
149 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SCAN
,
150 sizeof (uint64_t), SCAN_PHYS_NUMINTS
+ 1, &zaptmp
);
152 uint64_t overflow
= zaptmp
[SCAN_PHYS_NUMINTS
];
154 if (overflow
& ~DSL_SCAN_FLAGS_MASK
||
155 scn
->scn_async_destroying
) {
157 ZPOOL_ERRATA_ZOL_2094_ASYNC_DESTROY
;
161 bcopy(zaptmp
, &scn
->scn_phys
,
162 SCAN_PHYS_NUMINTS
* sizeof (uint64_t));
163 scn
->scn_phys
.scn_flags
= overflow
;
165 /* Required scrub already in progress. */
166 if (scn
->scn_phys
.scn_state
== DSS_FINISHED
||
167 scn
->scn_phys
.scn_state
== DSS_CANCELED
)
169 ZPOOL_ERRATA_ZOL_2094_SCRUB
;
178 if (scn
->scn_phys
.scn_state
== DSS_SCANNING
&&
179 spa_prev_software_version(dp
->dp_spa
) < SPA_VERSION_SCAN
) {
181 * A new-type scrub was in progress on an old
182 * pool, and the pool was accessed by old
183 * software. Restart from the beginning, since
184 * the old software may have changed the pool in
187 scn
->scn_restart_txg
= txg
;
188 zfs_dbgmsg("new-style scrub was modified "
189 "by old software; restarting in txg %llu",
190 scn
->scn_restart_txg
);
194 spa_scan_stat_init(spa
);
199 dsl_scan_fini(dsl_pool_t
*dp
)
202 kmem_free(dp
->dp_scan
, sizeof (dsl_scan_t
));
209 dsl_scan_setup_check(void *arg
, dmu_tx_t
*tx
)
211 dsl_scan_t
*scn
= dmu_tx_pool(tx
)->dp_scan
;
213 if (scn
->scn_phys
.scn_state
== DSS_SCANNING
)
214 return (SET_ERROR(EBUSY
));
220 dsl_scan_setup_sync(void *arg
, dmu_tx_t
*tx
)
222 dsl_scan_t
*scn
= dmu_tx_pool(tx
)->dp_scan
;
223 pool_scan_func_t
*funcp
= arg
;
224 dmu_object_type_t ot
= 0;
225 dsl_pool_t
*dp
= scn
->scn_dp
;
226 spa_t
*spa
= dp
->dp_spa
;
228 ASSERT(scn
->scn_phys
.scn_state
!= DSS_SCANNING
);
229 ASSERT(*funcp
> POOL_SCAN_NONE
&& *funcp
< POOL_SCAN_FUNCS
);
230 bzero(&scn
->scn_phys
, sizeof (scn
->scn_phys
));
231 scn
->scn_phys
.scn_func
= *funcp
;
232 scn
->scn_phys
.scn_state
= DSS_SCANNING
;
233 scn
->scn_phys
.scn_min_txg
= 0;
234 scn
->scn_phys
.scn_max_txg
= tx
->tx_txg
;
235 scn
->scn_phys
.scn_ddt_class_max
= DDT_CLASSES
- 1; /* the entire DDT */
236 scn
->scn_phys
.scn_start_time
= gethrestime_sec();
237 scn
->scn_phys
.scn_errors
= 0;
238 scn
->scn_phys
.scn_to_examine
= spa
->spa_root_vdev
->vdev_stat
.vs_alloc
;
239 scn
->scn_restart_txg
= 0;
240 scn
->scn_done_txg
= 0;
241 spa_scan_stat_init(spa
);
243 if (DSL_SCAN_IS_SCRUB_RESILVER(scn
)) {
244 scn
->scn_phys
.scn_ddt_class_max
= zfs_scrub_ddt_class_max
;
246 /* rewrite all disk labels */
247 vdev_config_dirty(spa
->spa_root_vdev
);
249 if (vdev_resilver_needed(spa
->spa_root_vdev
,
250 &scn
->scn_phys
.scn_min_txg
, &scn
->scn_phys
.scn_max_txg
)) {
251 spa_event_notify(spa
, NULL
, ESC_ZFS_RESILVER_START
);
253 spa_event_notify(spa
, NULL
, ESC_ZFS_SCRUB_START
);
256 spa
->spa_scrub_started
= B_TRUE
;
258 * If this is an incremental scrub, limit the DDT scrub phase
259 * to just the auto-ditto class (for correctness); the rest
260 * of the scrub should go faster using top-down pruning.
262 if (scn
->scn_phys
.scn_min_txg
> TXG_INITIAL
)
263 scn
->scn_phys
.scn_ddt_class_max
= DDT_CLASS_DITTO
;
267 /* back to the generic stuff */
269 if (dp
->dp_blkstats
== NULL
) {
271 vmem_alloc(sizeof (zfs_all_blkstats_t
), KM_SLEEP
);
273 bzero(dp
->dp_blkstats
, sizeof (zfs_all_blkstats_t
));
275 if (spa_version(spa
) < SPA_VERSION_DSL_SCRUB
)
276 ot
= DMU_OT_ZAP_OTHER
;
278 scn
->scn_phys
.scn_queue_obj
= zap_create(dp
->dp_meta_objset
,
279 ot
? ot
: DMU_OT_SCAN_QUEUE
, DMU_OT_NONE
, 0, tx
);
281 dsl_scan_sync_state(scn
, tx
);
283 spa_history_log_internal(spa
, "scan setup", tx
,
284 "func=%u mintxg=%llu maxtxg=%llu",
285 *funcp
, scn
->scn_phys
.scn_min_txg
, scn
->scn_phys
.scn_max_txg
);
290 dsl_scan_done(dsl_scan_t
*scn
, boolean_t complete
, dmu_tx_t
*tx
)
292 static const char *old_names
[] = {
294 "scrub_ddt_bookmark",
295 "scrub_ddt_class_max",
304 dsl_pool_t
*dp
= scn
->scn_dp
;
305 spa_t
*spa
= dp
->dp_spa
;
308 /* Remove any remnants of an old-style scrub. */
309 for (i
= 0; old_names
[i
]; i
++) {
310 (void) zap_remove(dp
->dp_meta_objset
,
311 DMU_POOL_DIRECTORY_OBJECT
, old_names
[i
], tx
);
314 if (scn
->scn_phys
.scn_queue_obj
!= 0) {
315 VERIFY(0 == dmu_object_free(dp
->dp_meta_objset
,
316 scn
->scn_phys
.scn_queue_obj
, tx
));
317 scn
->scn_phys
.scn_queue_obj
= 0;
321 * If we were "restarted" from a stopped state, don't bother
322 * with anything else.
324 if (scn
->scn_phys
.scn_state
!= DSS_SCANNING
)
328 scn
->scn_phys
.scn_state
= DSS_FINISHED
;
330 scn
->scn_phys
.scn_state
= DSS_CANCELED
;
332 if (dsl_scan_restarting(scn
, tx
))
333 spa_history_log_internal(spa
, "scan aborted, restarting", tx
,
334 "errors=%llu", spa_get_errlog_size(spa
));
336 spa_history_log_internal(spa
, "scan cancelled", tx
,
337 "errors=%llu", spa_get_errlog_size(spa
));
339 spa_history_log_internal(spa
, "scan done", tx
,
340 "errors=%llu", spa_get_errlog_size(spa
));
342 if (DSL_SCAN_IS_SCRUB_RESILVER(scn
)) {
343 mutex_enter(&spa
->spa_scrub_lock
);
344 while (spa
->spa_scrub_inflight
> 0) {
345 cv_wait(&spa
->spa_scrub_io_cv
,
346 &spa
->spa_scrub_lock
);
348 mutex_exit(&spa
->spa_scrub_lock
);
349 spa
->spa_scrub_started
= B_FALSE
;
350 spa
->spa_scrub_active
= B_FALSE
;
353 * If the scrub/resilver completed, update all DTLs to
354 * reflect this. Whether it succeeded or not, vacate
355 * all temporary scrub DTLs.
357 vdev_dtl_reassess(spa
->spa_root_vdev
, tx
->tx_txg
,
358 complete
? scn
->scn_phys
.scn_max_txg
: 0, B_TRUE
);
360 spa_event_notify(spa
, NULL
, scn
->scn_phys
.scn_min_txg
?
361 ESC_ZFS_RESILVER_FINISH
: ESC_ZFS_SCRUB_FINISH
);
363 spa_errlog_rotate(spa
);
366 * We may have finished replacing a device.
367 * Let the async thread assess this and handle the detach.
369 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
372 scn
->scn_phys
.scn_end_time
= gethrestime_sec();
374 if (spa
->spa_errata
== ZPOOL_ERRATA_ZOL_2094_SCRUB
)
380 dsl_scan_cancel_check(void *arg
, dmu_tx_t
*tx
)
382 dsl_scan_t
*scn
= dmu_tx_pool(tx
)->dp_scan
;
384 if (scn
->scn_phys
.scn_state
!= DSS_SCANNING
)
385 return (SET_ERROR(ENOENT
));
391 dsl_scan_cancel_sync(void *arg
, dmu_tx_t
*tx
)
393 dsl_scan_t
*scn
= dmu_tx_pool(tx
)->dp_scan
;
395 dsl_scan_done(scn
, B_FALSE
, tx
);
396 dsl_scan_sync_state(scn
, tx
);
400 dsl_scan_cancel(dsl_pool_t
*dp
)
402 return (dsl_sync_task(spa_name(dp
->dp_spa
), dsl_scan_cancel_check
,
403 dsl_scan_cancel_sync
, NULL
, 3, ZFS_SPACE_CHECK_RESERVED
));
406 static void dsl_scan_visitbp(blkptr_t
*bp
, const zbookmark_phys_t
*zb
,
407 dnode_phys_t
*dnp
, dsl_dataset_t
*ds
, dsl_scan_t
*scn
,
408 dmu_objset_type_t ostype
, dmu_tx_t
*tx
);
409 inline __attribute__((always_inline
)) static void dsl_scan_visitdnode(
410 dsl_scan_t
*, dsl_dataset_t
*ds
, dmu_objset_type_t ostype
,
411 dnode_phys_t
*dnp
, uint64_t object
, dmu_tx_t
*tx
);
414 dsl_free(dsl_pool_t
*dp
, uint64_t txg
, const blkptr_t
*bp
)
416 zio_free(dp
->dp_spa
, txg
, bp
);
420 dsl_free_sync(zio_t
*pio
, dsl_pool_t
*dp
, uint64_t txg
, const blkptr_t
*bpp
)
422 ASSERT(dsl_pool_sync_context(dp
));
423 zio_nowait(zio_free_sync(pio
, dp
->dp_spa
, txg
, bpp
, pio
->io_flags
));
427 dsl_scan_ds_maxtxg(dsl_dataset_t
*ds
)
429 uint64_t smt
= ds
->ds_dir
->dd_pool
->dp_scan
->scn_phys
.scn_max_txg
;
430 if (ds
->ds_is_snapshot
)
431 return (MIN(smt
, dsl_dataset_phys(ds
)->ds_creation_txg
));
436 dsl_scan_sync_state(dsl_scan_t
*scn
, dmu_tx_t
*tx
)
438 VERIFY0(zap_update(scn
->scn_dp
->dp_meta_objset
,
439 DMU_POOL_DIRECTORY_OBJECT
,
440 DMU_POOL_SCAN
, sizeof (uint64_t), SCAN_PHYS_NUMINTS
,
441 &scn
->scn_phys
, tx
));
444 extern int zfs_vdev_async_write_active_min_dirty_percent
;
447 dsl_scan_check_pause(dsl_scan_t
*scn
, const zbookmark_phys_t
*zb
)
449 uint64_t elapsed_nanosecs
;
453 /* we never skip user/group accounting objects */
454 if (zb
&& (int64_t)zb
->zb_object
< 0)
457 if (scn
->scn_pausing
)
458 return (B_TRUE
); /* we're already pausing */
460 if (!ZB_IS_ZERO(&scn
->scn_phys
.scn_bookmark
))
461 return (B_FALSE
); /* we're resuming */
463 /* We only know how to resume from level-0 blocks. */
464 if (zb
&& zb
->zb_level
!= 0)
469 * - we have scanned for the maximum time: an entire txg
470 * timeout (default 5 sec)
472 * - we have scanned for at least the minimum time (default 1 sec
473 * for scrub, 3 sec for resilver), and either we have sufficient
474 * dirty data that we are starting to write more quickly
475 * (default 30%), or someone is explicitly waiting for this txg
478 * - the spa is shutting down because this pool is being exported
479 * or the machine is rebooting.
481 mintime
= (scn
->scn_phys
.scn_func
== POOL_SCAN_RESILVER
) ?
482 zfs_resilver_min_time_ms
: zfs_scan_min_time_ms
;
483 elapsed_nanosecs
= gethrtime() - scn
->scn_sync_start_time
;
484 dirty_pct
= scn
->scn_dp
->dp_dirty_total
* 100 / zfs_dirty_data_max
;
485 if (elapsed_nanosecs
/ NANOSEC
>= zfs_txg_timeout
||
486 (NSEC2MSEC(elapsed_nanosecs
) > mintime
&&
487 (txg_sync_waiting(scn
->scn_dp
) ||
488 dirty_pct
>= zfs_vdev_async_write_active_min_dirty_percent
)) ||
489 spa_shutting_down(scn
->scn_dp
->dp_spa
)) {
491 dprintf("pausing at bookmark %llx/%llx/%llx/%llx\n",
492 (longlong_t
)zb
->zb_objset
,
493 (longlong_t
)zb
->zb_object
,
494 (longlong_t
)zb
->zb_level
,
495 (longlong_t
)zb
->zb_blkid
);
496 scn
->scn_phys
.scn_bookmark
= *zb
;
498 dprintf("pausing at DDT bookmark %llx/%llx/%llx/%llx\n",
499 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_class
,
500 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_type
,
501 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_checksum
,
502 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_cursor
);
503 scn
->scn_pausing
= B_TRUE
;
509 typedef struct zil_scan_arg
{
511 zil_header_t
*zsa_zh
;
516 dsl_scan_zil_block(zilog_t
*zilog
, blkptr_t
*bp
, void *arg
, uint64_t claim_txg
)
518 zil_scan_arg_t
*zsa
= arg
;
519 dsl_pool_t
*dp
= zsa
->zsa_dp
;
520 dsl_scan_t
*scn
= dp
->dp_scan
;
521 zil_header_t
*zh
= zsa
->zsa_zh
;
524 if (BP_IS_HOLE(bp
) || bp
->blk_birth
<= scn
->scn_phys
.scn_cur_min_txg
)
528 * One block ("stubby") can be allocated a long time ago; we
529 * want to visit that one because it has been allocated
530 * (on-disk) even if it hasn't been claimed (even though for
531 * scrub there's nothing to do to it).
533 if (claim_txg
== 0 && bp
->blk_birth
>= spa_first_txg(dp
->dp_spa
))
536 SET_BOOKMARK(&zb
, zh
->zh_log
.blk_cksum
.zc_word
[ZIL_ZC_OBJSET
],
537 ZB_ZIL_OBJECT
, ZB_ZIL_LEVEL
, bp
->blk_cksum
.zc_word
[ZIL_ZC_SEQ
]);
539 VERIFY(0 == scan_funcs
[scn
->scn_phys
.scn_func
](dp
, bp
, &zb
));
545 dsl_scan_zil_record(zilog_t
*zilog
, lr_t
*lrc
, void *arg
, uint64_t claim_txg
)
547 if (lrc
->lrc_txtype
== TX_WRITE
) {
548 zil_scan_arg_t
*zsa
= arg
;
549 dsl_pool_t
*dp
= zsa
->zsa_dp
;
550 dsl_scan_t
*scn
= dp
->dp_scan
;
551 zil_header_t
*zh
= zsa
->zsa_zh
;
552 lr_write_t
*lr
= (lr_write_t
*)lrc
;
553 blkptr_t
*bp
= &lr
->lr_blkptr
;
556 if (BP_IS_HOLE(bp
) ||
557 bp
->blk_birth
<= scn
->scn_phys
.scn_cur_min_txg
)
561 * birth can be < claim_txg if this record's txg is
562 * already txg sync'ed (but this log block contains
563 * other records that are not synced)
565 if (claim_txg
== 0 || bp
->blk_birth
< claim_txg
)
568 SET_BOOKMARK(&zb
, zh
->zh_log
.blk_cksum
.zc_word
[ZIL_ZC_OBJSET
],
569 lr
->lr_foid
, ZB_ZIL_LEVEL
,
570 lr
->lr_offset
/ BP_GET_LSIZE(bp
));
572 VERIFY(0 == scan_funcs
[scn
->scn_phys
.scn_func
](dp
, bp
, &zb
));
578 dsl_scan_zil(dsl_pool_t
*dp
, zil_header_t
*zh
)
580 uint64_t claim_txg
= zh
->zh_claim_txg
;
581 zil_scan_arg_t zsa
= { dp
, zh
};
585 * We only want to visit blocks that have been claimed but not yet
586 * replayed (or, in read-only mode, blocks that *would* be claimed).
588 if (claim_txg
== 0 && spa_writeable(dp
->dp_spa
))
591 zilog
= zil_alloc(dp
->dp_meta_objset
, zh
);
593 (void) zil_parse(zilog
, dsl_scan_zil_block
, dsl_scan_zil_record
, &zsa
,
601 dsl_scan_prefetch(dsl_scan_t
*scn
, arc_buf_t
*buf
, blkptr_t
*bp
,
602 uint64_t objset
, uint64_t object
, uint64_t blkid
)
604 zbookmark_phys_t czb
;
605 arc_flags_t flags
= ARC_FLAG_NOWAIT
| ARC_FLAG_PREFETCH
;
607 if (zfs_no_scrub_prefetch
)
610 if (BP_IS_HOLE(bp
) || bp
->blk_birth
<= scn
->scn_phys
.scn_min_txg
||
611 (BP_GET_LEVEL(bp
) == 0 && BP_GET_TYPE(bp
) != DMU_OT_DNODE
))
614 SET_BOOKMARK(&czb
, objset
, object
, BP_GET_LEVEL(bp
), blkid
);
616 (void) arc_read(scn
->scn_zio_root
, scn
->scn_dp
->dp_spa
, bp
,
617 NULL
, NULL
, ZIO_PRIORITY_ASYNC_READ
,
618 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SCAN_THREAD
, &flags
, &czb
);
622 dsl_scan_check_resume(dsl_scan_t
*scn
, const dnode_phys_t
*dnp
,
623 const zbookmark_phys_t
*zb
)
626 * We never skip over user/group accounting objects (obj<0)
628 if (!ZB_IS_ZERO(&scn
->scn_phys
.scn_bookmark
) &&
629 (int64_t)zb
->zb_object
>= 0) {
631 * If we already visited this bp & everything below (in
632 * a prior txg sync), don't bother doing it again.
634 if (zbookmark_subtree_completed(dnp
, zb
,
635 &scn
->scn_phys
.scn_bookmark
))
639 * If we found the block we're trying to resume from, or
640 * we went past it to a different object, zero it out to
641 * indicate that it's OK to start checking for pausing
644 if (bcmp(zb
, &scn
->scn_phys
.scn_bookmark
, sizeof (*zb
)) == 0 ||
645 zb
->zb_object
> scn
->scn_phys
.scn_bookmark
.zb_object
) {
646 dprintf("resuming at %llx/%llx/%llx/%llx\n",
647 (longlong_t
)zb
->zb_objset
,
648 (longlong_t
)zb
->zb_object
,
649 (longlong_t
)zb
->zb_level
,
650 (longlong_t
)zb
->zb_blkid
);
651 bzero(&scn
->scn_phys
.scn_bookmark
, sizeof (*zb
));
658 * Return nonzero on i/o error.
659 * Return new buf to write out in *bufp.
661 inline __attribute__((always_inline
)) static int
662 dsl_scan_recurse(dsl_scan_t
*scn
, dsl_dataset_t
*ds
, dmu_objset_type_t ostype
,
663 dnode_phys_t
*dnp
, const blkptr_t
*bp
,
664 const zbookmark_phys_t
*zb
, dmu_tx_t
*tx
)
666 dsl_pool_t
*dp
= scn
->scn_dp
;
667 int zio_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_SCAN_THREAD
;
670 if (BP_GET_LEVEL(bp
) > 0) {
671 arc_flags_t flags
= ARC_FLAG_WAIT
;
674 int epb
= BP_GET_LSIZE(bp
) >> SPA_BLKPTRSHIFT
;
677 err
= arc_read(NULL
, dp
->dp_spa
, bp
, arc_getbuf_func
, &buf
,
678 ZIO_PRIORITY_ASYNC_READ
, zio_flags
, &flags
, zb
);
680 scn
->scn_phys
.scn_errors
++;
683 for (i
= 0, cbp
= buf
->b_data
; i
< epb
; i
++, cbp
++) {
684 dsl_scan_prefetch(scn
, buf
, cbp
, zb
->zb_objset
,
685 zb
->zb_object
, zb
->zb_blkid
* epb
+ i
);
687 for (i
= 0, cbp
= buf
->b_data
; i
< epb
; i
++, cbp
++) {
688 zbookmark_phys_t czb
;
690 SET_BOOKMARK(&czb
, zb
->zb_objset
, zb
->zb_object
,
692 zb
->zb_blkid
* epb
+ i
);
693 dsl_scan_visitbp(cbp
, &czb
, dnp
,
694 ds
, scn
, ostype
, tx
);
696 arc_buf_destroy(buf
, &buf
);
697 } else if (BP_GET_TYPE(bp
) == DMU_OT_DNODE
) {
698 arc_flags_t flags
= ARC_FLAG_WAIT
;
701 int epb
= BP_GET_LSIZE(bp
) >> DNODE_SHIFT
;
704 err
= arc_read(NULL
, dp
->dp_spa
, bp
, arc_getbuf_func
, &buf
,
705 ZIO_PRIORITY_ASYNC_READ
, zio_flags
, &flags
, zb
);
707 scn
->scn_phys
.scn_errors
++;
710 for (i
= 0, cdnp
= buf
->b_data
; i
< epb
;
711 i
+= cdnp
->dn_extra_slots
+ 1,
712 cdnp
+= cdnp
->dn_extra_slots
+ 1) {
713 for (j
= 0; j
< cdnp
->dn_nblkptr
; j
++) {
714 blkptr_t
*cbp
= &cdnp
->dn_blkptr
[j
];
715 dsl_scan_prefetch(scn
, buf
, cbp
,
716 zb
->zb_objset
, zb
->zb_blkid
* epb
+ i
, j
);
719 for (i
= 0, cdnp
= buf
->b_data
; i
< epb
;
720 i
+= cdnp
->dn_extra_slots
+ 1,
721 cdnp
+= cdnp
->dn_extra_slots
+ 1) {
722 dsl_scan_visitdnode(scn
, ds
, ostype
,
723 cdnp
, zb
->zb_blkid
* epb
+ i
, tx
);
726 arc_buf_destroy(buf
, &buf
);
727 } else if (BP_GET_TYPE(bp
) == DMU_OT_OBJSET
) {
728 arc_flags_t flags
= ARC_FLAG_WAIT
;
732 err
= arc_read(NULL
, dp
->dp_spa
, bp
, arc_getbuf_func
, &buf
,
733 ZIO_PRIORITY_ASYNC_READ
, zio_flags
, &flags
, zb
);
735 scn
->scn_phys
.scn_errors
++;
741 dsl_scan_visitdnode(scn
, ds
, osp
->os_type
,
742 &osp
->os_meta_dnode
, DMU_META_DNODE_OBJECT
, tx
);
744 if (OBJSET_BUF_HAS_USERUSED(buf
)) {
746 * We also always visit user/group accounting
747 * objects, and never skip them, even if we are
748 * pausing. This is necessary so that the space
749 * deltas from this txg get integrated.
751 dsl_scan_visitdnode(scn
, ds
, osp
->os_type
,
752 &osp
->os_groupused_dnode
,
753 DMU_GROUPUSED_OBJECT
, tx
);
754 dsl_scan_visitdnode(scn
, ds
, osp
->os_type
,
755 &osp
->os_userused_dnode
,
756 DMU_USERUSED_OBJECT
, tx
);
758 arc_buf_destroy(buf
, &buf
);
764 inline __attribute__((always_inline
)) static void
765 dsl_scan_visitdnode(dsl_scan_t
*scn
, dsl_dataset_t
*ds
,
766 dmu_objset_type_t ostype
, dnode_phys_t
*dnp
,
767 uint64_t object
, dmu_tx_t
*tx
)
771 for (j
= 0; j
< dnp
->dn_nblkptr
; j
++) {
772 zbookmark_phys_t czb
;
774 SET_BOOKMARK(&czb
, ds
? ds
->ds_object
: 0, object
,
775 dnp
->dn_nlevels
- 1, j
);
776 dsl_scan_visitbp(&dnp
->dn_blkptr
[j
],
777 &czb
, dnp
, ds
, scn
, ostype
, tx
);
780 if (dnp
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
) {
781 zbookmark_phys_t czb
;
782 SET_BOOKMARK(&czb
, ds
? ds
->ds_object
: 0, object
,
784 dsl_scan_visitbp(DN_SPILL_BLKPTR(dnp
),
785 &czb
, dnp
, ds
, scn
, ostype
, tx
);
790 * The arguments are in this order because mdb can only print the
791 * first 5; we want them to be useful.
794 dsl_scan_visitbp(blkptr_t
*bp
, const zbookmark_phys_t
*zb
,
795 dnode_phys_t
*dnp
, dsl_dataset_t
*ds
, dsl_scan_t
*scn
,
796 dmu_objset_type_t ostype
, dmu_tx_t
*tx
)
798 dsl_pool_t
*dp
= scn
->scn_dp
;
801 bp_toread
= kmem_alloc(sizeof (blkptr_t
), KM_SLEEP
);
804 /* ASSERT(pbuf == NULL || arc_released(pbuf)); */
806 if (dsl_scan_check_pause(scn
, zb
))
809 if (dsl_scan_check_resume(scn
, dnp
, zb
))
815 scn
->scn_visited_this_txg
++;
818 * This debugging is commented out to conserve stack space. This
819 * function is called recursively and the debugging addes several
820 * bytes to the stack for each call. It can be commented back in
821 * if required to debug an issue in dsl_scan_visitbp().
824 * "visiting ds=%p/%llu zb=%llx/%llx/%llx/%llx bp=%p",
825 * ds, ds ? ds->ds_object : 0,
826 * zb->zb_objset, zb->zb_object, zb->zb_level, zb->zb_blkid,
830 if (bp
->blk_birth
<= scn
->scn_phys
.scn_cur_min_txg
)
833 if (dsl_scan_recurse(scn
, ds
, ostype
, dnp
, bp_toread
, zb
, tx
) != 0)
837 * If dsl_scan_ddt() has aready visited this block, it will have
838 * already done any translations or scrubbing, so don't call the
841 if (ddt_class_contains(dp
->dp_spa
,
842 scn
->scn_phys
.scn_ddt_class_max
, bp
)) {
847 * If this block is from the future (after cur_max_txg), then we
848 * are doing this on behalf of a deleted snapshot, and we will
849 * revisit the future block on the next pass of this dataset.
850 * Don't scan it now unless we need to because something
851 * under it was modified.
853 if (BP_PHYSICAL_BIRTH(bp
) <= scn
->scn_phys
.scn_cur_max_txg
) {
854 scan_funcs
[scn
->scn_phys
.scn_func
](dp
, bp
, zb
);
857 kmem_free(bp_toread
, sizeof (blkptr_t
));
861 dsl_scan_visit_rootbp(dsl_scan_t
*scn
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
866 SET_BOOKMARK(&zb
, ds
? ds
->ds_object
: DMU_META_OBJSET
,
867 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
868 dsl_scan_visitbp(bp
, &zb
, NULL
,
869 ds
, scn
, DMU_OST_NONE
, tx
);
871 dprintf_ds(ds
, "finished scan%s", "");
875 dsl_scan_ds_destroyed(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
877 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
878 dsl_scan_t
*scn
= dp
->dp_scan
;
881 if (scn
->scn_phys
.scn_state
!= DSS_SCANNING
)
884 if (scn
->scn_phys
.scn_bookmark
.zb_objset
== ds
->ds_object
) {
885 if (ds
->ds_is_snapshot
) {
888 * - scn_cur_{min,max}_txg stays the same.
889 * - Setting the flag is not really necessary if
890 * scn_cur_max_txg == scn_max_txg, because there
891 * is nothing after this snapshot that we care
892 * about. However, we set it anyway and then
893 * ignore it when we retraverse it in
894 * dsl_scan_visitds().
896 scn
->scn_phys
.scn_bookmark
.zb_objset
=
897 dsl_dataset_phys(ds
)->ds_next_snap_obj
;
898 zfs_dbgmsg("destroying ds %llu; currently traversing; "
899 "reset zb_objset to %llu",
900 (u_longlong_t
)ds
->ds_object
,
901 (u_longlong_t
)dsl_dataset_phys(ds
)->
903 scn
->scn_phys
.scn_flags
|= DSF_VISIT_DS_AGAIN
;
905 SET_BOOKMARK(&scn
->scn_phys
.scn_bookmark
,
906 ZB_DESTROYED_OBJSET
, 0, 0, 0);
907 zfs_dbgmsg("destroying ds %llu; currently traversing; "
908 "reset bookmark to -1,0,0,0",
909 (u_longlong_t
)ds
->ds_object
);
911 } else if (zap_lookup_int_key(dp
->dp_meta_objset
,
912 scn
->scn_phys
.scn_queue_obj
, ds
->ds_object
, &mintxg
) == 0) {
913 ASSERT3U(dsl_dataset_phys(ds
)->ds_num_children
, <=, 1);
914 VERIFY3U(0, ==, zap_remove_int(dp
->dp_meta_objset
,
915 scn
->scn_phys
.scn_queue_obj
, ds
->ds_object
, tx
));
916 if (ds
->ds_is_snapshot
) {
918 * We keep the same mintxg; it could be >
919 * ds_creation_txg if the previous snapshot was
922 VERIFY(zap_add_int_key(dp
->dp_meta_objset
,
923 scn
->scn_phys
.scn_queue_obj
,
924 dsl_dataset_phys(ds
)->ds_next_snap_obj
,
926 zfs_dbgmsg("destroying ds %llu; in queue; "
927 "replacing with %llu",
928 (u_longlong_t
)ds
->ds_object
,
929 (u_longlong_t
)dsl_dataset_phys(ds
)->
932 zfs_dbgmsg("destroying ds %llu; in queue; removing",
933 (u_longlong_t
)ds
->ds_object
);
938 * dsl_scan_sync() should be called after this, and should sync
939 * out our changed state, but just to be safe, do it here.
941 dsl_scan_sync_state(scn
, tx
);
945 dsl_scan_ds_snapshotted(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
947 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
948 dsl_scan_t
*scn
= dp
->dp_scan
;
951 if (scn
->scn_phys
.scn_state
!= DSS_SCANNING
)
954 ASSERT(dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0);
956 if (scn
->scn_phys
.scn_bookmark
.zb_objset
== ds
->ds_object
) {
957 scn
->scn_phys
.scn_bookmark
.zb_objset
=
958 dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
959 zfs_dbgmsg("snapshotting ds %llu; currently traversing; "
960 "reset zb_objset to %llu",
961 (u_longlong_t
)ds
->ds_object
,
962 (u_longlong_t
)dsl_dataset_phys(ds
)->ds_prev_snap_obj
);
963 } else if (zap_lookup_int_key(dp
->dp_meta_objset
,
964 scn
->scn_phys
.scn_queue_obj
, ds
->ds_object
, &mintxg
) == 0) {
965 VERIFY3U(0, ==, zap_remove_int(dp
->dp_meta_objset
,
966 scn
->scn_phys
.scn_queue_obj
, ds
->ds_object
, tx
));
967 VERIFY(zap_add_int_key(dp
->dp_meta_objset
,
968 scn
->scn_phys
.scn_queue_obj
,
969 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, mintxg
, tx
) == 0);
970 zfs_dbgmsg("snapshotting ds %llu; in queue; "
971 "replacing with %llu",
972 (u_longlong_t
)ds
->ds_object
,
973 (u_longlong_t
)dsl_dataset_phys(ds
)->ds_prev_snap_obj
);
975 dsl_scan_sync_state(scn
, tx
);
979 dsl_scan_ds_clone_swapped(dsl_dataset_t
*ds1
, dsl_dataset_t
*ds2
, dmu_tx_t
*tx
)
981 dsl_pool_t
*dp
= ds1
->ds_dir
->dd_pool
;
982 dsl_scan_t
*scn
= dp
->dp_scan
;
985 if (scn
->scn_phys
.scn_state
!= DSS_SCANNING
)
988 if (scn
->scn_phys
.scn_bookmark
.zb_objset
== ds1
->ds_object
) {
989 scn
->scn_phys
.scn_bookmark
.zb_objset
= ds2
->ds_object
;
990 zfs_dbgmsg("clone_swap ds %llu; currently traversing; "
991 "reset zb_objset to %llu",
992 (u_longlong_t
)ds1
->ds_object
,
993 (u_longlong_t
)ds2
->ds_object
);
994 } else if (scn
->scn_phys
.scn_bookmark
.zb_objset
== ds2
->ds_object
) {
995 scn
->scn_phys
.scn_bookmark
.zb_objset
= ds1
->ds_object
;
996 zfs_dbgmsg("clone_swap ds %llu; currently traversing; "
997 "reset zb_objset to %llu",
998 (u_longlong_t
)ds2
->ds_object
,
999 (u_longlong_t
)ds1
->ds_object
);
1002 if (zap_lookup_int_key(dp
->dp_meta_objset
, scn
->scn_phys
.scn_queue_obj
,
1003 ds1
->ds_object
, &mintxg
) == 0) {
1006 ASSERT3U(mintxg
, ==, dsl_dataset_phys(ds1
)->ds_prev_snap_txg
);
1007 ASSERT3U(mintxg
, ==, dsl_dataset_phys(ds2
)->ds_prev_snap_txg
);
1008 VERIFY3U(0, ==, zap_remove_int(dp
->dp_meta_objset
,
1009 scn
->scn_phys
.scn_queue_obj
, ds1
->ds_object
, tx
));
1010 err
= zap_add_int_key(dp
->dp_meta_objset
,
1011 scn
->scn_phys
.scn_queue_obj
, ds2
->ds_object
, mintxg
, tx
);
1012 VERIFY(err
== 0 || err
== EEXIST
);
1013 if (err
== EEXIST
) {
1014 /* Both were there to begin with */
1015 VERIFY(0 == zap_add_int_key(dp
->dp_meta_objset
,
1016 scn
->scn_phys
.scn_queue_obj
,
1017 ds1
->ds_object
, mintxg
, tx
));
1019 zfs_dbgmsg("clone_swap ds %llu; in queue; "
1020 "replacing with %llu",
1021 (u_longlong_t
)ds1
->ds_object
,
1022 (u_longlong_t
)ds2
->ds_object
);
1023 } else if (zap_lookup_int_key(dp
->dp_meta_objset
,
1024 scn
->scn_phys
.scn_queue_obj
, ds2
->ds_object
, &mintxg
) == 0) {
1025 ASSERT3U(mintxg
, ==, dsl_dataset_phys(ds1
)->ds_prev_snap_txg
);
1026 ASSERT3U(mintxg
, ==, dsl_dataset_phys(ds2
)->ds_prev_snap_txg
);
1027 VERIFY3U(0, ==, zap_remove_int(dp
->dp_meta_objset
,
1028 scn
->scn_phys
.scn_queue_obj
, ds2
->ds_object
, tx
));
1029 VERIFY(0 == zap_add_int_key(dp
->dp_meta_objset
,
1030 scn
->scn_phys
.scn_queue_obj
, ds1
->ds_object
, mintxg
, tx
));
1031 zfs_dbgmsg("clone_swap ds %llu; in queue; "
1032 "replacing with %llu",
1033 (u_longlong_t
)ds2
->ds_object
,
1034 (u_longlong_t
)ds1
->ds_object
);
1037 dsl_scan_sync_state(scn
, tx
);
1040 struct enqueue_clones_arg
{
1047 enqueue_clones_cb(dsl_pool_t
*dp
, dsl_dataset_t
*hds
, void *arg
)
1049 struct enqueue_clones_arg
*eca
= arg
;
1052 dsl_scan_t
*scn
= dp
->dp_scan
;
1054 if (dsl_dir_phys(hds
->ds_dir
)->dd_origin_obj
!= eca
->originobj
)
1057 err
= dsl_dataset_hold_obj(dp
, hds
->ds_object
, FTAG
, &ds
);
1061 while (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= eca
->originobj
) {
1062 dsl_dataset_t
*prev
;
1063 err
= dsl_dataset_hold_obj(dp
,
1064 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, FTAG
, &prev
);
1066 dsl_dataset_rele(ds
, FTAG
);
1071 VERIFY(zap_add_int_key(dp
->dp_meta_objset
,
1072 scn
->scn_phys
.scn_queue_obj
, ds
->ds_object
,
1073 dsl_dataset_phys(ds
)->ds_prev_snap_txg
, eca
->tx
) == 0);
1074 dsl_dataset_rele(ds
, FTAG
);
1079 dsl_scan_visitds(dsl_scan_t
*scn
, uint64_t dsobj
, dmu_tx_t
*tx
)
1081 dsl_pool_t
*dp
= scn
->scn_dp
;
1086 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp
, dsobj
, FTAG
, &ds
));
1088 if (scn
->scn_phys
.scn_cur_min_txg
>=
1089 scn
->scn_phys
.scn_max_txg
) {
1091 * This can happen if this snapshot was created after the
1092 * scan started, and we already completed a previous snapshot
1093 * that was created after the scan started. This snapshot
1094 * only references blocks with:
1096 * birth < our ds_creation_txg
1097 * cur_min_txg is no less than ds_creation_txg.
1098 * We have already visited these blocks.
1100 * birth > scn_max_txg
1101 * The scan requested not to visit these blocks.
1103 * Subsequent snapshots (and clones) can reference our
1104 * blocks, or blocks with even higher birth times.
1105 * Therefore we do not need to visit them either,
1106 * so we do not add them to the work queue.
1108 * Note that checking for cur_min_txg >= cur_max_txg
1109 * is not sufficient, because in that case we may need to
1110 * visit subsequent snapshots. This happens when min_txg > 0,
1111 * which raises cur_min_txg. In this case we will visit
1112 * this dataset but skip all of its blocks, because the
1113 * rootbp's birth time is < cur_min_txg. Then we will
1114 * add the next snapshots/clones to the work queue.
1116 char *dsname
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
1117 dsl_dataset_name(ds
, dsname
);
1118 zfs_dbgmsg("scanning dataset %llu (%s) is unnecessary because "
1119 "cur_min_txg (%llu) >= max_txg (%llu)",
1121 scn
->scn_phys
.scn_cur_min_txg
,
1122 scn
->scn_phys
.scn_max_txg
);
1123 kmem_free(dsname
, MAXNAMELEN
);
1128 if (dmu_objset_from_ds(ds
, &os
))
1132 * Only the ZIL in the head (non-snapshot) is valid. Even though
1133 * snapshots can have ZIL block pointers (which may be the same
1134 * BP as in the head), they must be ignored. So we traverse the
1135 * ZIL here, rather than in scan_recurse(), because the regular
1136 * snapshot block-sharing rules don't apply to it.
1138 if (DSL_SCAN_IS_SCRUB_RESILVER(scn
) && !ds
->ds_is_snapshot
)
1139 dsl_scan_zil(dp
, &os
->os_zil_header
);
1142 * Iterate over the bps in this ds.
1144 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1145 dsl_scan_visit_rootbp(scn
, ds
, &dsl_dataset_phys(ds
)->ds_bp
, tx
);
1147 dsname
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
1148 dsl_dataset_name(ds
, dsname
);
1149 zfs_dbgmsg("scanned dataset %llu (%s) with min=%llu max=%llu; "
1151 (longlong_t
)dsobj
, dsname
,
1152 (longlong_t
)scn
->scn_phys
.scn_cur_min_txg
,
1153 (longlong_t
)scn
->scn_phys
.scn_cur_max_txg
,
1154 (int)scn
->scn_pausing
);
1155 kmem_free(dsname
, ZFS_MAX_DATASET_NAME_LEN
);
1157 if (scn
->scn_pausing
)
1161 * We've finished this pass over this dataset.
1165 * If we did not completely visit this dataset, do another pass.
1167 if (scn
->scn_phys
.scn_flags
& DSF_VISIT_DS_AGAIN
) {
1168 zfs_dbgmsg("incomplete pass; visiting again");
1169 scn
->scn_phys
.scn_flags
&= ~DSF_VISIT_DS_AGAIN
;
1170 VERIFY(zap_add_int_key(dp
->dp_meta_objset
,
1171 scn
->scn_phys
.scn_queue_obj
, ds
->ds_object
,
1172 scn
->scn_phys
.scn_cur_max_txg
, tx
) == 0);
1177 * Add descendent datasets to work queue.
1179 if (dsl_dataset_phys(ds
)->ds_next_snap_obj
!= 0) {
1180 VERIFY(zap_add_int_key(dp
->dp_meta_objset
,
1181 scn
->scn_phys
.scn_queue_obj
,
1182 dsl_dataset_phys(ds
)->ds_next_snap_obj
,
1183 dsl_dataset_phys(ds
)->ds_creation_txg
, tx
) == 0);
1185 if (dsl_dataset_phys(ds
)->ds_num_children
> 1) {
1186 boolean_t usenext
= B_FALSE
;
1187 if (dsl_dataset_phys(ds
)->ds_next_clones_obj
!= 0) {
1190 * A bug in a previous version of the code could
1191 * cause upgrade_clones_cb() to not set
1192 * ds_next_snap_obj when it should, leading to a
1193 * missing entry. Therefore we can only use the
1194 * next_clones_obj when its count is correct.
1196 int err
= zap_count(dp
->dp_meta_objset
,
1197 dsl_dataset_phys(ds
)->ds_next_clones_obj
, &count
);
1199 count
== dsl_dataset_phys(ds
)->ds_num_children
- 1)
1204 VERIFY0(zap_join_key(dp
->dp_meta_objset
,
1205 dsl_dataset_phys(ds
)->ds_next_clones_obj
,
1206 scn
->scn_phys
.scn_queue_obj
,
1207 dsl_dataset_phys(ds
)->ds_creation_txg
, tx
));
1209 struct enqueue_clones_arg eca
;
1211 eca
.originobj
= ds
->ds_object
;
1213 VERIFY0(dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
1214 enqueue_clones_cb
, &eca
, DS_FIND_CHILDREN
));
1219 dsl_dataset_rele(ds
, FTAG
);
1224 enqueue_cb(dsl_pool_t
*dp
, dsl_dataset_t
*hds
, void *arg
)
1229 dsl_scan_t
*scn
= dp
->dp_scan
;
1231 err
= dsl_dataset_hold_obj(dp
, hds
->ds_object
, FTAG
, &ds
);
1235 while (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0) {
1236 dsl_dataset_t
*prev
;
1237 err
= dsl_dataset_hold_obj(dp
,
1238 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, FTAG
, &prev
);
1240 dsl_dataset_rele(ds
, FTAG
);
1245 * If this is a clone, we don't need to worry about it for now.
1247 if (dsl_dataset_phys(prev
)->ds_next_snap_obj
!= ds
->ds_object
) {
1248 dsl_dataset_rele(ds
, FTAG
);
1249 dsl_dataset_rele(prev
, FTAG
);
1252 dsl_dataset_rele(ds
, FTAG
);
1256 VERIFY(zap_add_int_key(dp
->dp_meta_objset
, scn
->scn_phys
.scn_queue_obj
,
1257 ds
->ds_object
, dsl_dataset_phys(ds
)->ds_prev_snap_txg
, tx
) == 0);
1258 dsl_dataset_rele(ds
, FTAG
);
1263 * Scrub/dedup interaction.
1265 * If there are N references to a deduped block, we don't want to scrub it
1266 * N times -- ideally, we should scrub it exactly once.
1268 * We leverage the fact that the dde's replication class (enum ddt_class)
1269 * is ordered from highest replication class (DDT_CLASS_DITTO) to lowest
1270 * (DDT_CLASS_UNIQUE) so that we may walk the DDT in that order.
1272 * To prevent excess scrubbing, the scrub begins by walking the DDT
1273 * to find all blocks with refcnt > 1, and scrubs each of these once.
1274 * Since there are two replication classes which contain blocks with
1275 * refcnt > 1, we scrub the highest replication class (DDT_CLASS_DITTO) first.
1276 * Finally the top-down scrub begins, only visiting blocks with refcnt == 1.
1278 * There would be nothing more to say if a block's refcnt couldn't change
1279 * during a scrub, but of course it can so we must account for changes
1280 * in a block's replication class.
1282 * Here's an example of what can occur:
1284 * If a block has refcnt > 1 during the DDT scrub phase, but has refcnt == 1
1285 * when visited during the top-down scrub phase, it will be scrubbed twice.
1286 * This negates our scrub optimization, but is otherwise harmless.
1288 * If a block has refcnt == 1 during the DDT scrub phase, but has refcnt > 1
1289 * on each visit during the top-down scrub phase, it will never be scrubbed.
1290 * To catch this, ddt_sync_entry() notifies the scrub code whenever a block's
1291 * reference class transitions to a higher level (i.e DDT_CLASS_UNIQUE to
1292 * DDT_CLASS_DUPLICATE); if it transitions from refcnt == 1 to refcnt > 1
1293 * while a scrub is in progress, it scrubs the block right then.
1296 dsl_scan_ddt(dsl_scan_t
*scn
, dmu_tx_t
*tx
)
1298 ddt_bookmark_t
*ddb
= &scn
->scn_phys
.scn_ddt_bookmark
;
1303 bzero(&dde
, sizeof (ddt_entry_t
));
1305 while ((error
= ddt_walk(scn
->scn_dp
->dp_spa
, ddb
, &dde
)) == 0) {
1308 if (ddb
->ddb_class
> scn
->scn_phys
.scn_ddt_class_max
)
1310 dprintf("visiting ddb=%llu/%llu/%llu/%llx\n",
1311 (longlong_t
)ddb
->ddb_class
,
1312 (longlong_t
)ddb
->ddb_type
,
1313 (longlong_t
)ddb
->ddb_checksum
,
1314 (longlong_t
)ddb
->ddb_cursor
);
1316 /* There should be no pending changes to the dedup table */
1317 ddt
= scn
->scn_dp
->dp_spa
->spa_ddt
[ddb
->ddb_checksum
];
1318 ASSERT(avl_first(&ddt
->ddt_tree
) == NULL
);
1320 dsl_scan_ddt_entry(scn
, ddb
->ddb_checksum
, &dde
, tx
);
1323 if (dsl_scan_check_pause(scn
, NULL
))
1327 zfs_dbgmsg("scanned %llu ddt entries with class_max = %u; pausing=%u",
1328 (longlong_t
)n
, (int)scn
->scn_phys
.scn_ddt_class_max
,
1329 (int)scn
->scn_pausing
);
1331 ASSERT(error
== 0 || error
== ENOENT
);
1332 ASSERT(error
!= ENOENT
||
1333 ddb
->ddb_class
> scn
->scn_phys
.scn_ddt_class_max
);
1338 dsl_scan_ddt_entry(dsl_scan_t
*scn
, enum zio_checksum checksum
,
1339 ddt_entry_t
*dde
, dmu_tx_t
*tx
)
1341 const ddt_key_t
*ddk
= &dde
->dde_key
;
1342 ddt_phys_t
*ddp
= dde
->dde_phys
;
1344 zbookmark_phys_t zb
= { 0 };
1347 if (scn
->scn_phys
.scn_state
!= DSS_SCANNING
)
1350 for (p
= 0; p
< DDT_PHYS_TYPES
; p
++, ddp
++) {
1351 if (ddp
->ddp_phys_birth
== 0 ||
1352 ddp
->ddp_phys_birth
> scn
->scn_phys
.scn_max_txg
)
1354 ddt_bp_create(checksum
, ddk
, ddp
, &bp
);
1356 scn
->scn_visited_this_txg
++;
1357 scan_funcs
[scn
->scn_phys
.scn_func
](scn
->scn_dp
, &bp
, &zb
);
1362 dsl_scan_visit(dsl_scan_t
*scn
, dmu_tx_t
*tx
)
1364 dsl_pool_t
*dp
= scn
->scn_dp
;
1366 zap_attribute_t
*za
;
1368 if (scn
->scn_phys
.scn_ddt_bookmark
.ddb_class
<=
1369 scn
->scn_phys
.scn_ddt_class_max
) {
1370 scn
->scn_phys
.scn_cur_min_txg
= scn
->scn_phys
.scn_min_txg
;
1371 scn
->scn_phys
.scn_cur_max_txg
= scn
->scn_phys
.scn_max_txg
;
1372 dsl_scan_ddt(scn
, tx
);
1373 if (scn
->scn_pausing
)
1377 if (scn
->scn_phys
.scn_bookmark
.zb_objset
== DMU_META_OBJSET
) {
1378 /* First do the MOS & ORIGIN */
1380 scn
->scn_phys
.scn_cur_min_txg
= scn
->scn_phys
.scn_min_txg
;
1381 scn
->scn_phys
.scn_cur_max_txg
= scn
->scn_phys
.scn_max_txg
;
1382 dsl_scan_visit_rootbp(scn
, NULL
,
1383 &dp
->dp_meta_rootbp
, tx
);
1384 spa_set_rootblkptr(dp
->dp_spa
, &dp
->dp_meta_rootbp
);
1385 if (scn
->scn_pausing
)
1388 if (spa_version(dp
->dp_spa
) < SPA_VERSION_DSL_SCRUB
) {
1389 VERIFY0(dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
1390 enqueue_cb
, tx
, DS_FIND_CHILDREN
));
1392 dsl_scan_visitds(scn
,
1393 dp
->dp_origin_snap
->ds_object
, tx
);
1395 ASSERT(!scn
->scn_pausing
);
1396 } else if (scn
->scn_phys
.scn_bookmark
.zb_objset
!=
1397 ZB_DESTROYED_OBJSET
) {
1399 * If we were paused, continue from here. Note if the
1400 * ds we were paused on was deleted, the zb_objset may
1401 * be -1, so we will skip this and find a new objset
1404 dsl_scan_visitds(scn
, scn
->scn_phys
.scn_bookmark
.zb_objset
, tx
);
1405 if (scn
->scn_pausing
)
1410 * In case we were paused right at the end of the ds, zero the
1411 * bookmark so we don't think that we're still trying to resume.
1413 bzero(&scn
->scn_phys
.scn_bookmark
, sizeof (zbookmark_phys_t
));
1414 zc
= kmem_alloc(sizeof (zap_cursor_t
), KM_SLEEP
);
1415 za
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
1417 /* keep pulling things out of the zap-object-as-queue */
1418 while (zap_cursor_init(zc
, dp
->dp_meta_objset
,
1419 scn
->scn_phys
.scn_queue_obj
),
1420 zap_cursor_retrieve(zc
, za
) == 0) {
1424 dsobj
= strtonum(za
->za_name
, NULL
);
1425 VERIFY3U(0, ==, zap_remove_int(dp
->dp_meta_objset
,
1426 scn
->scn_phys
.scn_queue_obj
, dsobj
, tx
));
1428 /* Set up min/max txg */
1429 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp
, dsobj
, FTAG
, &ds
));
1430 if (za
->za_first_integer
!= 0) {
1431 scn
->scn_phys
.scn_cur_min_txg
=
1432 MAX(scn
->scn_phys
.scn_min_txg
,
1433 za
->za_first_integer
);
1435 scn
->scn_phys
.scn_cur_min_txg
=
1436 MAX(scn
->scn_phys
.scn_min_txg
,
1437 dsl_dataset_phys(ds
)->ds_prev_snap_txg
);
1439 scn
->scn_phys
.scn_cur_max_txg
= dsl_scan_ds_maxtxg(ds
);
1440 dsl_dataset_rele(ds
, FTAG
);
1442 dsl_scan_visitds(scn
, dsobj
, tx
);
1443 zap_cursor_fini(zc
);
1444 if (scn
->scn_pausing
)
1447 zap_cursor_fini(zc
);
1449 kmem_free(za
, sizeof (zap_attribute_t
));
1450 kmem_free(zc
, sizeof (zap_cursor_t
));
1454 dsl_scan_free_should_pause(dsl_scan_t
*scn
)
1456 uint64_t elapsed_nanosecs
;
1461 if (scn
->scn_visited_this_txg
>= zfs_free_max_blocks
)
1464 elapsed_nanosecs
= gethrtime() - scn
->scn_sync_start_time
;
1465 return (elapsed_nanosecs
/ NANOSEC
> zfs_txg_timeout
||
1466 (NSEC2MSEC(elapsed_nanosecs
) > zfs_free_min_time_ms
&&
1467 txg_sync_waiting(scn
->scn_dp
)) ||
1468 spa_shutting_down(scn
->scn_dp
->dp_spa
));
1472 dsl_scan_free_block_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
1474 dsl_scan_t
*scn
= arg
;
1476 if (!scn
->scn_is_bptree
||
1477 (BP_GET_LEVEL(bp
) == 0 && BP_GET_TYPE(bp
) != DMU_OT_OBJSET
)) {
1478 if (dsl_scan_free_should_pause(scn
))
1479 return (SET_ERROR(ERESTART
));
1482 zio_nowait(zio_free_sync(scn
->scn_zio_root
, scn
->scn_dp
->dp_spa
,
1483 dmu_tx_get_txg(tx
), bp
, 0));
1484 dsl_dir_diduse_space(tx
->tx_pool
->dp_free_dir
, DD_USED_HEAD
,
1485 -bp_get_dsize_sync(scn
->scn_dp
->dp_spa
, bp
),
1486 -BP_GET_PSIZE(bp
), -BP_GET_UCSIZE(bp
), tx
);
1487 scn
->scn_visited_this_txg
++;
1492 dsl_scan_active(dsl_scan_t
*scn
)
1494 spa_t
*spa
= scn
->scn_dp
->dp_spa
;
1495 uint64_t used
= 0, comp
, uncomp
;
1497 if (spa
->spa_load_state
!= SPA_LOAD_NONE
)
1499 if (spa_shutting_down(spa
))
1501 if (scn
->scn_phys
.scn_state
== DSS_SCANNING
||
1502 (scn
->scn_async_destroying
&& !scn
->scn_async_stalled
))
1505 if (spa_version(scn
->scn_dp
->dp_spa
) >= SPA_VERSION_DEADLISTS
) {
1506 (void) bpobj_space(&scn
->scn_dp
->dp_free_bpobj
,
1507 &used
, &comp
, &uncomp
);
1513 dsl_scan_sync(dsl_pool_t
*dp
, dmu_tx_t
*tx
)
1515 dsl_scan_t
*scn
= dp
->dp_scan
;
1516 spa_t
*spa
= dp
->dp_spa
;
1520 * Check for scn_restart_txg before checking spa_load_state, so
1521 * that we can restart an old-style scan while the pool is being
1522 * imported (see dsl_scan_init).
1524 if (dsl_scan_restarting(scn
, tx
)) {
1525 pool_scan_func_t func
= POOL_SCAN_SCRUB
;
1526 dsl_scan_done(scn
, B_FALSE
, tx
);
1527 if (vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
))
1528 func
= POOL_SCAN_RESILVER
;
1529 zfs_dbgmsg("restarting scan func=%u txg=%llu",
1531 dsl_scan_setup_sync(&func
, tx
);
1535 * Only process scans in sync pass 1.
1537 if (spa_sync_pass(dp
->dp_spa
) > 1)
1541 * If the spa is shutting down, then stop scanning. This will
1542 * ensure that the scan does not dirty any new data during the
1545 if (spa_shutting_down(spa
))
1549 * If the scan is inactive due to a stalled async destroy, try again.
1551 if (!scn
->scn_async_stalled
&& !dsl_scan_active(scn
))
1554 scn
->scn_visited_this_txg
= 0;
1555 scn
->scn_pausing
= B_FALSE
;
1556 scn
->scn_sync_start_time
= gethrtime();
1557 spa
->spa_scrub_active
= B_TRUE
;
1560 * First process the async destroys. If we pause, don't do
1561 * any scrubbing or resilvering. This ensures that there are no
1562 * async destroys while we are scanning, so the scan code doesn't
1563 * have to worry about traversing it. It is also faster to free the
1564 * blocks than to scrub them.
1566 if (zfs_free_bpobj_enabled
&&
1567 spa_version(dp
->dp_spa
) >= SPA_VERSION_DEADLISTS
) {
1568 scn
->scn_is_bptree
= B_FALSE
;
1569 scn
->scn_zio_root
= zio_root(dp
->dp_spa
, NULL
,
1570 NULL
, ZIO_FLAG_MUSTSUCCEED
);
1571 err
= bpobj_iterate(&dp
->dp_free_bpobj
,
1572 dsl_scan_free_block_cb
, scn
, tx
);
1573 VERIFY3U(0, ==, zio_wait(scn
->scn_zio_root
));
1575 if (err
!= 0 && err
!= ERESTART
)
1576 zfs_panic_recover("error %u from bpobj_iterate()", err
);
1579 if (err
== 0 && spa_feature_is_active(spa
, SPA_FEATURE_ASYNC_DESTROY
)) {
1580 ASSERT(scn
->scn_async_destroying
);
1581 scn
->scn_is_bptree
= B_TRUE
;
1582 scn
->scn_zio_root
= zio_root(dp
->dp_spa
, NULL
,
1583 NULL
, ZIO_FLAG_MUSTSUCCEED
);
1584 err
= bptree_iterate(dp
->dp_meta_objset
,
1585 dp
->dp_bptree_obj
, B_TRUE
, dsl_scan_free_block_cb
, scn
, tx
);
1586 VERIFY0(zio_wait(scn
->scn_zio_root
));
1588 if (err
== EIO
|| err
== ECKSUM
) {
1590 } else if (err
!= 0 && err
!= ERESTART
) {
1591 zfs_panic_recover("error %u from "
1592 "traverse_dataset_destroyed()", err
);
1595 if (bptree_is_empty(dp
->dp_meta_objset
, dp
->dp_bptree_obj
)) {
1596 /* finished; deactivate async destroy feature */
1597 spa_feature_decr(spa
, SPA_FEATURE_ASYNC_DESTROY
, tx
);
1598 ASSERT(!spa_feature_is_active(spa
,
1599 SPA_FEATURE_ASYNC_DESTROY
));
1600 VERIFY0(zap_remove(dp
->dp_meta_objset
,
1601 DMU_POOL_DIRECTORY_OBJECT
,
1602 DMU_POOL_BPTREE_OBJ
, tx
));
1603 VERIFY0(bptree_free(dp
->dp_meta_objset
,
1604 dp
->dp_bptree_obj
, tx
));
1605 dp
->dp_bptree_obj
= 0;
1606 scn
->scn_async_destroying
= B_FALSE
;
1607 scn
->scn_async_stalled
= B_FALSE
;
1610 * If we didn't make progress, mark the async
1611 * destroy as stalled, so that we will not initiate
1612 * a spa_sync() on its behalf. Note that we only
1613 * check this if we are not finished, because if the
1614 * bptree had no blocks for us to visit, we can
1615 * finish without "making progress".
1617 scn
->scn_async_stalled
=
1618 (scn
->scn_visited_this_txg
== 0);
1621 if (scn
->scn_visited_this_txg
) {
1622 zfs_dbgmsg("freed %llu blocks in %llums from "
1623 "free_bpobj/bptree txg %llu; err=%u",
1624 (longlong_t
)scn
->scn_visited_this_txg
,
1626 NSEC2MSEC(gethrtime() - scn
->scn_sync_start_time
),
1627 (longlong_t
)tx
->tx_txg
, err
);
1628 scn
->scn_visited_this_txg
= 0;
1631 * Write out changes to the DDT that may be required as a
1632 * result of the blocks freed. This ensures that the DDT
1633 * is clean when a scrub/resilver runs.
1635 ddt_sync(spa
, tx
->tx_txg
);
1639 if (dp
->dp_free_dir
!= NULL
&& !scn
->scn_async_destroying
&&
1640 zfs_free_leak_on_eio
&&
1641 (dsl_dir_phys(dp
->dp_free_dir
)->dd_used_bytes
!= 0 ||
1642 dsl_dir_phys(dp
->dp_free_dir
)->dd_compressed_bytes
!= 0 ||
1643 dsl_dir_phys(dp
->dp_free_dir
)->dd_uncompressed_bytes
!= 0)) {
1645 * We have finished background destroying, but there is still
1646 * some space left in the dp_free_dir. Transfer this leaked
1647 * space to the dp_leak_dir.
1649 if (dp
->dp_leak_dir
== NULL
) {
1650 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
1651 (void) dsl_dir_create_sync(dp
, dp
->dp_root_dir
,
1653 VERIFY0(dsl_pool_open_special_dir(dp
,
1654 LEAK_DIR_NAME
, &dp
->dp_leak_dir
));
1655 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
1657 dsl_dir_diduse_space(dp
->dp_leak_dir
, DD_USED_HEAD
,
1658 dsl_dir_phys(dp
->dp_free_dir
)->dd_used_bytes
,
1659 dsl_dir_phys(dp
->dp_free_dir
)->dd_compressed_bytes
,
1660 dsl_dir_phys(dp
->dp_free_dir
)->dd_uncompressed_bytes
, tx
);
1661 dsl_dir_diduse_space(dp
->dp_free_dir
, DD_USED_HEAD
,
1662 -dsl_dir_phys(dp
->dp_free_dir
)->dd_used_bytes
,
1663 -dsl_dir_phys(dp
->dp_free_dir
)->dd_compressed_bytes
,
1664 -dsl_dir_phys(dp
->dp_free_dir
)->dd_uncompressed_bytes
, tx
);
1666 if (dp
->dp_free_dir
!= NULL
&& !scn
->scn_async_destroying
) {
1667 /* finished; verify that space accounting went to zero */
1668 ASSERT0(dsl_dir_phys(dp
->dp_free_dir
)->dd_used_bytes
);
1669 ASSERT0(dsl_dir_phys(dp
->dp_free_dir
)->dd_compressed_bytes
);
1670 ASSERT0(dsl_dir_phys(dp
->dp_free_dir
)->dd_uncompressed_bytes
);
1673 if (scn
->scn_phys
.scn_state
!= DSS_SCANNING
)
1676 if (scn
->scn_done_txg
== tx
->tx_txg
) {
1677 ASSERT(!scn
->scn_pausing
);
1678 /* finished with scan. */
1679 zfs_dbgmsg("txg %llu scan complete", tx
->tx_txg
);
1680 dsl_scan_done(scn
, B_TRUE
, tx
);
1681 ASSERT3U(spa
->spa_scrub_inflight
, ==, 0);
1682 dsl_scan_sync_state(scn
, tx
);
1686 if (scn
->scn_phys
.scn_ddt_bookmark
.ddb_class
<=
1687 scn
->scn_phys
.scn_ddt_class_max
) {
1688 zfs_dbgmsg("doing scan sync txg %llu; "
1689 "ddt bm=%llu/%llu/%llu/%llx",
1690 (longlong_t
)tx
->tx_txg
,
1691 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_class
,
1692 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_type
,
1693 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_checksum
,
1694 (longlong_t
)scn
->scn_phys
.scn_ddt_bookmark
.ddb_cursor
);
1695 ASSERT(scn
->scn_phys
.scn_bookmark
.zb_objset
== 0);
1696 ASSERT(scn
->scn_phys
.scn_bookmark
.zb_object
== 0);
1697 ASSERT(scn
->scn_phys
.scn_bookmark
.zb_level
== 0);
1698 ASSERT(scn
->scn_phys
.scn_bookmark
.zb_blkid
== 0);
1700 zfs_dbgmsg("doing scan sync txg %llu; bm=%llu/%llu/%llu/%llu",
1701 (longlong_t
)tx
->tx_txg
,
1702 (longlong_t
)scn
->scn_phys
.scn_bookmark
.zb_objset
,
1703 (longlong_t
)scn
->scn_phys
.scn_bookmark
.zb_object
,
1704 (longlong_t
)scn
->scn_phys
.scn_bookmark
.zb_level
,
1705 (longlong_t
)scn
->scn_phys
.scn_bookmark
.zb_blkid
);
1708 scn
->scn_zio_root
= zio_root(dp
->dp_spa
, NULL
,
1709 NULL
, ZIO_FLAG_CANFAIL
);
1710 dsl_pool_config_enter(dp
, FTAG
);
1711 dsl_scan_visit(scn
, tx
);
1712 dsl_pool_config_exit(dp
, FTAG
);
1713 (void) zio_wait(scn
->scn_zio_root
);
1714 scn
->scn_zio_root
= NULL
;
1716 zfs_dbgmsg("visited %llu blocks in %llums",
1717 (longlong_t
)scn
->scn_visited_this_txg
,
1718 (longlong_t
)NSEC2MSEC(gethrtime() - scn
->scn_sync_start_time
));
1720 if (!scn
->scn_pausing
) {
1721 scn
->scn_done_txg
= tx
->tx_txg
+ 1;
1722 zfs_dbgmsg("txg %llu traversal complete, waiting till txg %llu",
1723 tx
->tx_txg
, scn
->scn_done_txg
);
1726 if (DSL_SCAN_IS_SCRUB_RESILVER(scn
)) {
1727 mutex_enter(&spa
->spa_scrub_lock
);
1728 while (spa
->spa_scrub_inflight
> 0) {
1729 cv_wait(&spa
->spa_scrub_io_cv
,
1730 &spa
->spa_scrub_lock
);
1732 mutex_exit(&spa
->spa_scrub_lock
);
1735 dsl_scan_sync_state(scn
, tx
);
1739 * This will start a new scan, or restart an existing one.
1742 dsl_resilver_restart(dsl_pool_t
*dp
, uint64_t txg
)
1746 tx
= dmu_tx_create_dd(dp
->dp_mos_dir
);
1747 VERIFY(0 == dmu_tx_assign(tx
, TXG_WAIT
));
1749 txg
= dmu_tx_get_txg(tx
);
1750 dp
->dp_scan
->scn_restart_txg
= txg
;
1753 dp
->dp_scan
->scn_restart_txg
= txg
;
1755 zfs_dbgmsg("restarting resilver txg=%llu", txg
);
1759 dsl_scan_resilvering(dsl_pool_t
*dp
)
1761 return (dp
->dp_scan
->scn_phys
.scn_state
== DSS_SCANNING
&&
1762 dp
->dp_scan
->scn_phys
.scn_func
== POOL_SCAN_RESILVER
);
1770 count_block(zfs_all_blkstats_t
*zab
, const blkptr_t
*bp
)
1775 * If we resume after a reboot, zab will be NULL; don't record
1776 * incomplete stats in that case.
1781 for (i
= 0; i
< 4; i
++) {
1782 int l
= (i
< 2) ? BP_GET_LEVEL(bp
) : DN_MAX_LEVELS
;
1783 int t
= (i
& 1) ? BP_GET_TYPE(bp
) : DMU_OT_TOTAL
;
1787 if (t
& DMU_OT_NEWTYPE
)
1790 zb
= &zab
->zab_type
[l
][t
];
1792 zb
->zb_asize
+= BP_GET_ASIZE(bp
);
1793 zb
->zb_lsize
+= BP_GET_LSIZE(bp
);
1794 zb
->zb_psize
+= BP_GET_PSIZE(bp
);
1795 zb
->zb_gangs
+= BP_COUNT_GANG(bp
);
1797 switch (BP_GET_NDVAS(bp
)) {
1799 if (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
1800 DVA_GET_VDEV(&bp
->blk_dva
[1]))
1801 zb
->zb_ditto_2_of_2_samevdev
++;
1804 equal
= (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
1805 DVA_GET_VDEV(&bp
->blk_dva
[1])) +
1806 (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
1807 DVA_GET_VDEV(&bp
->blk_dva
[2])) +
1808 (DVA_GET_VDEV(&bp
->blk_dva
[1]) ==
1809 DVA_GET_VDEV(&bp
->blk_dva
[2]));
1811 zb
->zb_ditto_2_of_3_samevdev
++;
1812 else if (equal
== 3)
1813 zb
->zb_ditto_3_of_3_samevdev
++;
1820 dsl_scan_scrub_done(zio_t
*zio
)
1822 spa_t
*spa
= zio
->io_spa
;
1824 abd_free(zio
->io_abd
);
1826 mutex_enter(&spa
->spa_scrub_lock
);
1827 spa
->spa_scrub_inflight
--;
1828 cv_broadcast(&spa
->spa_scrub_io_cv
);
1830 if (zio
->io_error
&& (zio
->io_error
!= ECKSUM
||
1831 !(zio
->io_flags
& ZIO_FLAG_SPECULATIVE
))) {
1832 spa
->spa_dsl_pool
->dp_scan
->scn_phys
.scn_errors
++;
1834 mutex_exit(&spa
->spa_scrub_lock
);
1838 dsl_scan_scrub_cb(dsl_pool_t
*dp
,
1839 const blkptr_t
*bp
, const zbookmark_phys_t
*zb
)
1841 dsl_scan_t
*scn
= dp
->dp_scan
;
1842 size_t size
= BP_GET_PSIZE(bp
);
1843 spa_t
*spa
= dp
->dp_spa
;
1844 uint64_t phys_birth
= BP_PHYSICAL_BIRTH(bp
);
1845 boolean_t needs_io
= B_FALSE
;
1846 int zio_flags
= ZIO_FLAG_SCAN_THREAD
| ZIO_FLAG_RAW
| ZIO_FLAG_CANFAIL
;
1850 if (phys_birth
<= scn
->scn_phys
.scn_min_txg
||
1851 phys_birth
>= scn
->scn_phys
.scn_max_txg
)
1854 count_block(dp
->dp_blkstats
, bp
);
1856 if (BP_IS_EMBEDDED(bp
))
1859 ASSERT(DSL_SCAN_IS_SCRUB_RESILVER(scn
));
1860 if (scn
->scn_phys
.scn_func
== POOL_SCAN_SCRUB
) {
1861 zio_flags
|= ZIO_FLAG_SCRUB
;
1863 scan_delay
= zfs_scrub_delay
;
1865 ASSERT3U(scn
->scn_phys
.scn_func
, ==, POOL_SCAN_RESILVER
);
1866 zio_flags
|= ZIO_FLAG_RESILVER
;
1868 scan_delay
= zfs_resilver_delay
;
1871 /* If it's an intent log block, failure is expected. */
1872 if (zb
->zb_level
== ZB_ZIL_LEVEL
)
1873 zio_flags
|= ZIO_FLAG_SPECULATIVE
;
1875 for (d
= 0; d
< BP_GET_NDVAS(bp
); d
++) {
1876 vdev_t
*vd
= vdev_lookup_top(spa
,
1877 DVA_GET_VDEV(&bp
->blk_dva
[d
]));
1880 * Keep track of how much data we've examined so that
1881 * zpool(1M) status can make useful progress reports.
1883 scn
->scn_phys
.scn_examined
+= DVA_GET_ASIZE(&bp
->blk_dva
[d
]);
1884 spa
->spa_scan_pass_exam
+= DVA_GET_ASIZE(&bp
->blk_dva
[d
]);
1886 /* if it's a resilver, this may not be in the target range */
1888 if (DVA_GET_GANG(&bp
->blk_dva
[d
])) {
1890 * Gang members may be spread across multiple
1891 * vdevs, so the best estimate we have is the
1892 * scrub range, which has already been checked.
1893 * XXX -- it would be better to change our
1894 * allocation policy to ensure that all
1895 * gang members reside on the same vdev.
1899 needs_io
= vdev_dtl_contains(vd
, DTL_PARTIAL
,
1905 if (needs_io
&& !zfs_no_scrub_io
) {
1906 vdev_t
*rvd
= spa
->spa_root_vdev
;
1907 uint64_t maxinflight
= rvd
->vdev_children
* zfs_top_maxinflight
;
1909 mutex_enter(&spa
->spa_scrub_lock
);
1910 while (spa
->spa_scrub_inflight
>= maxinflight
)
1911 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
1912 spa
->spa_scrub_inflight
++;
1913 mutex_exit(&spa
->spa_scrub_lock
);
1916 * If we're seeing recent (zfs_scan_idle) "important" I/Os
1917 * then throttle our workload to limit the impact of a scan.
1919 if (ddi_get_lbolt64() - spa
->spa_last_io
<= zfs_scan_idle
)
1922 zio_nowait(zio_read(NULL
, spa
, bp
,
1923 abd_alloc_for_io(size
, B_FALSE
), size
, dsl_scan_scrub_done
,
1924 NULL
, ZIO_PRIORITY_SCRUB
, zio_flags
, zb
));
1927 /* do not relocate this block */
1932 dsl_scan(dsl_pool_t
*dp
, pool_scan_func_t func
)
1934 spa_t
*spa
= dp
->dp_spa
;
1937 * Purge all vdev caches and probe all devices. We do this here
1938 * rather than in sync context because this requires a writer lock
1939 * on the spa_config lock, which we can't do from sync context. The
1940 * spa_scrub_reopen flag indicates that vdev_open() should not
1941 * attempt to start another scrub.
1943 spa_vdev_state_enter(spa
, SCL_NONE
);
1944 spa
->spa_scrub_reopen
= B_TRUE
;
1945 vdev_reopen(spa
->spa_root_vdev
);
1946 spa
->spa_scrub_reopen
= B_FALSE
;
1947 (void) spa_vdev_state_exit(spa
, NULL
, 0);
1949 return (dsl_sync_task(spa_name(spa
), dsl_scan_setup_check
,
1950 dsl_scan_setup_sync
, &func
, 0, ZFS_SPACE_CHECK_NONE
));
1954 dsl_scan_restarting(dsl_scan_t
*scn
, dmu_tx_t
*tx
)
1956 return (scn
->scn_restart_txg
!= 0 &&
1957 scn
->scn_restart_txg
<= tx
->tx_txg
);
1960 #if defined(_KERNEL) && defined(HAVE_SPL)
1961 module_param(zfs_top_maxinflight
, int, 0644);
1962 MODULE_PARM_DESC(zfs_top_maxinflight
, "Max I/Os per top-level");
1964 module_param(zfs_resilver_delay
, int, 0644);
1965 MODULE_PARM_DESC(zfs_resilver_delay
, "Number of ticks to delay resilver");
1967 module_param(zfs_scrub_delay
, int, 0644);
1968 MODULE_PARM_DESC(zfs_scrub_delay
, "Number of ticks to delay scrub");
1970 module_param(zfs_scan_idle
, int, 0644);
1971 MODULE_PARM_DESC(zfs_scan_idle
, "Idle window in clock ticks");
1973 module_param(zfs_scan_min_time_ms
, int, 0644);
1974 MODULE_PARM_DESC(zfs_scan_min_time_ms
, "Min millisecs to scrub per txg");
1976 module_param(zfs_free_min_time_ms
, int, 0644);
1977 MODULE_PARM_DESC(zfs_free_min_time_ms
, "Min millisecs to free per txg");
1979 module_param(zfs_resilver_min_time_ms
, int, 0644);
1980 MODULE_PARM_DESC(zfs_resilver_min_time_ms
, "Min millisecs to resilver per txg");
1982 module_param(zfs_no_scrub_io
, int, 0644);
1983 MODULE_PARM_DESC(zfs_no_scrub_io
, "Set to disable scrub I/O");
1985 module_param(zfs_no_scrub_prefetch
, int, 0644);
1986 MODULE_PARM_DESC(zfs_no_scrub_prefetch
, "Set to disable scrub prefetching");
1988 module_param(zfs_free_max_blocks
, ulong
, 0644);
1989 MODULE_PARM_DESC(zfs_free_max_blocks
, "Max number of blocks freed in one txg");
1991 module_param(zfs_free_bpobj_enabled
, int, 0644);
1992 MODULE_PARM_DESC(zfs_free_bpobj_enabled
, "Enable processing of the free_bpobj");